Radc TR 85 91 - May1985

Downloaded from http://www.everyspec.
com
AD-A158 843
RADC-TR-85-91
.Final Technical Report
Mel 196S
IMPACT OF NONOPERATING PERIODS ON
EOUIPMENT RELIABILITY
Ir Rtesearch Institute
DarEd W. Colt end Mary 0. Prior
-M FON P9KIC
U i" M DIRN1U ULMUED
:-:: • SEP 0 4 0
LL.
:: ROME AIR DEVELOPMENT CENTER

Air Force Systems Command
:* Griffiss Air Force Base, NY 13441-5700
Reproduced From
r\• Best Available Copy
"""
"
6
Downloaded from http://www.everyspec.com
-7177
This report has been reviewed by the RADC
isa releasable to the National Technical public Affairt 'Office (PA) and
Information Service' (NTIS)-. At NTIS
it will-be releasable to the general public,
including- foreigna nations.
RADC-TR-85-91 has been reviewed and is
apProved for publication.
APPROVED:
~. -
PRESTON R. MACDIARNID
- -: -.
Project Engineer
APPROVED: 400
'7
U. S. TUTHILL, Colonel, US"?

Chief, Reliability &Compatibility Disof
FOR THE CO*O(1MER: 0I
Plan& Office * -
If your address has changed or if You.

wish to be"removed'.
list, or if the addressee is-no longer foth. aD m in
employed by your, 0 9mn at~on, : please
notizy RAWC (RBE-z) Griffjisg APB NY :
13441-5700. This il, assis81tus In-
taining, a current mailing list. main- .
Do not return copies 'ofthis report

unless* contractual 6bligatiotw or-
on a-specific document requie noticees
ta It be returned..:
UNCTASSTVTrn
SECURITY CLASSIFICATION OF THIS PAGE
REPORT DOCUMENTATION PAGE

1,. REPORT SICURIT't CLASSIFICATION 10I. RE6STRIOCTIVE MARKINGS
UNCLASSIFIED - N/A
2&. SECURITY CLASSIFICATION AUTHORITY 3. OISTRIOUTIONIAVASLAIILITY OP REPORT
UN/A Approved for public release; distribution
[20 OIECLASSIPSICATION/IOOWGRtAOING SCHEDULE unlimited.
4 ~~N/A__________________ __
A4. PtAPORMI'NG ORGANIZATIUN REPORT NUMBER($) 4. MONITORING ORGANIZATION REPORT NUMSESRISI
N ,/A RLADC-TR-85-91
Os&NAME OF PERFORMING ORGANIZATION 0.OFFICE SYMBOL 7s. NAME OP MONITORING ORGANIZATION
* ITT Research Institute IRome Air Development Center (RBE-2)
G&.
ACORESS (City. Stae and ZIP Code) 70. ADDREaSS ICity. SUNaned ZIP Code)
Turin Road, North Grif fiss APB NY 13441-5700
P.O. Box 180
R~RNY 114411_______
90. NAME OF PUNOING/SPONSOMING O. FFPICESYMUO. 9. PROCUREMENT INSTRUMENT IOENTIPICATION NUMBER
Rome Air Develoement Center RBE-2 F30602-83-C-0056

Be. AGORESS (City. State and ZIP Caft -10. Sounce OP FUNDING NOS$.
PROGAAM IPROJE1CT TASK WOK UNIT
Grif flag AFB NY 13441-5700 E LEMONT No. No. No. N.1
11I. TI TLELflneivad. Sfmuhty CIlaIfleat.io 20F 23 29

IMPACT OF N0NOP)PRATING PERIODS ON EQUIPMENT RELIABILITY
12. PERSONAL AUTHORIISI
David W. Coit, Mary G. Priore
13&. TYPE OF REPORT 131L TIME COVERED IA. OA40TS
OP REPORT (Y.. Joe.. D'i S.PAGE COUNT
Final PROM Feb 83 T _S 84May 1985 446
141.SUPPLEMENTARY NOTATION
N/A
I?. COSATI COCES Its. SU JECT TEENS IceatIfnue a^ maa it Neessay and identify by bSleft Number)
FIED GROUP I sun. oR. IRlability Dormancy MIL-HDBK-217
IELD Failure Rate Storage
I nINonoperating
l Power On-Off Cycling
IS. ABSTRACT fCeaUntiu on .weeeleAt mpeoemmr end ideftif, by blork numbrtI
e objective of this study was to develop a procedure to viredict the quantitative effects
of nonoperating periods on electronic equipmnent reliability. A series of nonoperating
failure rate prediction models were developed at the component level. The models are
capable of evaluating component nonoperating failure rate for any anticipated environment
with the exception of a satellite environment.
* The proposed nonoperating failure rate prediction methodology is Intended to provide the
ability to predict the component nonoperating failure rate and reliability as a function
of the characteristics of the devices, technology employed in producing the device, and
I' external factors such as environmental stresses which have a significant effect on device
nonoperating reliability. The prediction methodology is presented ina form compatible
withh J!-L-HDBK-2l7 as an Appendix to the technical report. /
-Ai
~~~~~1
~4 446--2-Li
200 OISTRIE1UTION,A ILAEI BTV OP AESTRA49 6 3 1. ASTRAACT SECURIITY CLASSIFIC-ATION4
uN~CLAaso~sIFEUN.imITvoE3 SAME AS RQPT.(3 OTIC USERS E3 UNCLASSIFIED

220, NAME OF RESPONEIOLE INDIVIDUAL 22b, TELEP 22c. OFFICE SYMBOL
-.---. -~~ t o.
ooudcadet
Preston R 'NacDiarmtt- (315) 330-7095 RADC (RBE-2)
/WFORM 1473, 83 APR EDITION OF I JAN 73 IS OBSOL.ETE. UNCLASSIFIED
0 SECURITY CLASSIFICATION OP THIS PA ;a
/h ~-~&&
fb ~ * 44 C .
UNCWiSIFIED
qCUNITY CLASSIPICA1ION OP THIS PACE
17. Cosati Codes (Continued)
Field Group
14 04
UNISSFE
"SCOT LI.CTO 0 omPa

PREFACE
This Final Report was prepared by lIT Research Institute, Rome, New
York, for the Rome Air Development Center, Griffiss AFB, New York, under
Contract F30602-83-C-0056. The RADC laboratory contract manager for this
* program was Mr. Preston MacDiarmid (RBE-2). The study originator and
initial RADC laboratory contract manager was Mr. Lester Gubbins. This
report covers the work performed from February 1983 to September 1984.
The principalinvestigator for this project was David W. Coit.

Donald
Valuable assistance was provided by Mary Priore, Donald Fulton,
Rymer, James Carey, Kieron Dey, Thomas Ballou, Bernard Radigan, Harold
Lauffenburger, William Denson, Michael Rossi, Donald Rash and David Dylis.
Report production was coordinated by Pamela Coe and Virginia Dwyer.
i AccesssiOfi vo
NITIS GRA&I
DTIC TAB
Unannounced 03
justificatio
By-
D smt r i bu t i on / __ _ . - -
AvailabilitY CodeS
Av...ii. and/or
Dist Special
Volp
INSPCCTED
I:
MANAGEMENT SUMMARY
The objective of this study was to develop a procedure to predict the

quantitative effects of nonoperating periods on electronic equipment
reliability. A series of nonoperating failure rate prediction models were

developed at the component level. The models are capable of evaluating
component nonoperating failure rate for any anticipated environment with
the exception of a satellite environment.
•" The proposed nonoperating failure rate prediction methodology is

intended to provide the ability to predict the component nonoperating
failure rate and reliability as a function of the characteristics of the
devices, technology employed in producing the device, and external factors
such as environmental stresses which have a significant effect on device
nonoperating reliability. The prediction methodology is presented in a
form compatible with MIL-HDBK-217 as an Appendix to the technical report.
Observed nonoperating failure rate data were collected from a variety

of sources. The following criteria were established for an acceptable
source of data:
o Data available to the part level

o Primary failures can be separated from total maintenance actions
o Nonoperating failures can be separated from operating failures
o Sufficient detail can be identified for components
o Sufficient equipment nonoperating hours to expect failures
A summary of the collected data is presented in Table MS-1.
* The model development approach was based primarily on empirical data

analysis. Thus, the proposed models include variables which can be shown
to significantly affect nonoperating failure rate. A model development
matrix is presented in Table MS-2. The model development matrix indicates
the part class, data sources, empirical model factors, assumed/theoretical
model factors and also provides a discussion.
,•" :" "'• "• .""3'_• i •" ." • r •i

A clear and concise procedure was

also developed to apply the
"component nonoperating failure rate prediction
models to assess the impact
of nonoperating periods at the equipment level.
comprehensive reliability prediction Additionally, a
methodology was presented which
describes the use of the proposed nonoperating
failure rate prediction
models together with the documented
operating failure rate assessment
techniques (i.e. MIL-HDBK-217D).
It is recommended that the proposed

nonoperating failure rate
prediction models developed during this
study be incorporated into NIL-
"HDBK-217. Additionally, it is recommended
that the nonoperating failure
rate prediction models be updated periodically
to reflect changes in
techisclogy or other factors which
temporarily result in an inaccurate
or
missing model.
"* •
iv
--- I
; -- i.--
/ F
TABLE MS-i: NONOPERATING FAILURE RATE DATA SUM4ARY
Model Development Model Evaluation
Data Failures Part Data Failures Part

Part Class Hrs.
Records Hrs. Records
(x161) (x2332
IMonolithic Micro- 50 1947

circuits 644 353 25862 11
Hybrid Microcir-
27 2082 50050 0 0 0
cubts
106 121 21983 16 23 8422
Transistors
139 57 22181 28 77 25468
Diodes
73 87 49833 0 0 0
Inductive Devices
413 34 113119 21 17 16390
•Resistors
324 48 41841 22 14 5004

Capacitors
47 364 795 0 0 0
/Tubes
Rotating Mechanisms 8 20 149 0 0 0
Relays 13 36 1360 0 0 0
"\Switches 16 35 408 0 0 0
Connectors 6 1 82444 0 0 0
Interconnection 3 3 2977 0 0 0
'-:I-:Assemblies
Misc. Parts 8 5 129 0 0 0
,-,I. . ,-, \V.( .' . •
S• • -" / ' • • " \ " t - - '" "- - • - - '

W43
0) d41r
0in
CW to-U~ 39 1
a,- G)ul(
4A .C4J(1,)'0
to "4 0EWa=0 ) .)-- C W ,.,m0 .,,,.
W In>l 0 WO . M=W4-
- 4E
OWLV C.0
O - ,- O"- *1 e-
*4-0
c U..- U (A.0 *.- 0.U W) Uto I.. - to . ,W
In> >IW~ WW U >t~o C*-U4J
n 0. - " - S- .)" ,-sO >w .. fo-- s-
0 to 3 - - W•to
) 4-.
o 1O 'a 41 . 4.-• O WM-.-- . 5 W
S(L) • .S. OJ W"- >jm 0=,
O Cn= r= 4•go ,, =e-W
41
L 1 0.• =-.0
.0. C S S-E~ * L 004CM4 4 6 3 C. S-.CM
4to =-.O
Wo 4-4-' W.--- OWO0 CO
o)L.41 S- U*.= 4J S-.9# . EN
LS..n 4J(DL.0 >
""
M 0-WW5.4MM W- 0 iuC0-C-
C3
L o) faWE j0 0 • 4)"--._0
I--
0 (
" S- 4J 4 w-C
u 41
,, 0 .L. 00 '0 4-" .0 .0 06 (d '0 0
( 41
3)>E4, 3 U4 06L0
- "1r 4""M0 )- %
0.ý
uLJ n
= 4
"k "bW J4 WG-
I C.J udo x 06"- O3r5- U

C #
I-...r(A
*"In
f ý0 (A to c
*.-( C)0 %e-U3tIn-W4J'
-1 0. WL0 C 0
431
0 40
s-
,0
• 4A VV
0 ZO
0 CR
u I0% kI
CL..uo u L Co
vIn
> 4-
03t W 0(
c~ ~ ~ 5
4J055-CA
0. )4) .003
w A0 c3c0 41 (U t
C0 4 If 4.-0 6 4J f" 4A ) fa
Ii I--I=i=hi
.9 W34JU L- = RC W4
$-to 30
CAl M
4EL W W-CU 9- U)>W(
t
W)
Um) ) 0 a 40. 40W.06 S.
3 30w(
Uý C S..f45C *u)
M- C
0-
03 4)
0 4)0 Atoý 9 - to %-S 40)-4
.CO 4- J2 4- . -f C 4 '0
0 W
W~~9 S- 4*4J4J 4
cc 03 0= C
0) 0)0
> - > - >, -
_ ~
9(
U4
r
W U 0)
, U
C-)
4
) >
=I- M.0 03-4 34 034)
uI. fa>, >03 Ct u
0) 0 0 IV 0
Wi 4) 4) 4)
W) 44) 41
4W- ~ 4-J %C0-

%- 4 ic) ZO uZO4U
IV I4h 0.0 4' 03 -4)19 w
-c. .tU g...E 0.
CL.0
4J M 0h -v-
0 4) 4-345-
4) CL C)05
CL. cc
r to
S
u- . (
01-
0
0 r' i. 4 ., 0 *0 ,v-. 5•a• '..L_ 0 0"
• , .Wr-01-.
0.1 e- 0
" '-4 ,. 42
,0..-
- m" 0 1-03
4-4-. i wOEE"
W UL>
0
00W5a-ý
,. OW43.'-. 0 3.-
O .- -n
S- W,.
4-3
Qj' 1-1-al
> - 40
e.-4 ... 43,L
(A 0 S- 0 - 4J CL-
W S 01 0)W c(A 0 0"
W
r_ r= c u u0) 410 C1
0 tp
Wto 43 -C o _Q
C._.4-- M 4Sa in 9 o 6j4
WL0
w
W - 4J W.
4-b04a
4 ,fd.wf-, 4- 0 • L
S- 5, u 4-
U 0r r-C
"- 0 0 to(v X>%-CiiS '

IA1-J C I c(
CD
LLIX
0 0J u .
%-.4 43 .
431 0)1 oin w W1

J2 -u .u -SUný1
0 0
w. () med)3
¶g jl
0 .4,3
c•
0 w~
P -, ,
43 ) Ul C"." ,
owr.U csO ()• l~.U)C%•4•
•"1 S-
4W 001.
-1 a
*c.
.
S- 4
4- CW.- CW
00
4-
W - .-
U) 4..0 0 43r- €
C• 4C
0 S- .-.
34) 9.e' ,.- u-W
CEU:OJ# 4..--
tM
c45 - c W * -r-C(0(A
C.UL~E
W'o-a t; 4- 0IU 0 0 M43 E4C)%44
o -" (P
4 - -.- 4 aU
o go,- r 0 =~
u U . d) w4j aw>%
to • 4Jr
W•43I-•SUW-~ .O U>w l0
,.U)(
Ci
)CC
4J W
4.L ( ,
9- w>4
w 0 oJx C J3 C 4J(U0 S-.3 w W4J 4-

06 =))~ I 0.31 0U= . 0=4-= OI.-~
.to
-4.,)
o CA
CU)
u) S
fa C
MI
_i #A
04J 4)a
Ci C
_EU I.... .E Cr'i r
U) C
coJ
uix #A S
CiC
S- %A04J4
to 00 04
ixi
" ~~I-,in"
'd
51 cL.1 4- in 4-
S.-.- Q,-
We
l.-" cC .-
'""-- 0 - I 0 W-o• WC3.
00 L.4- 40 0 r-. ,4. . >41 S,- 4- >1s- o
.- 0 -0
::0',- "tM"¢
0 1:0 %- 4M 4-S4-.L= W U)
C.L 3c r-"
CL3.0W uS-
C-" 0 W A -. 4) a) 4- CL. c . WLin
CO'r- 0)W."" d
e'-~~~~~ .u. ' M 4,J Ue a,Mto4).-
3 (A 41. :3
0 M 4J-
IA 4-J C" ()
o (A cmý 4n--) u n)
U >C-10 ; 0 U •- .0 Q,
in S'- 4.._ .- 0 >,r- a 4J MO0 4J tO
"" -I-4J () =EdSO 4. ( -. .) = ) C--4- 4-.
.- r - oO >% u 4-J 0-4- E'-- E (-
i.
n 4- 0 -- (DW t >% uLC• I- ' ,- 4.,) - 4J
. ,. ) 0
u u- 0J S.f.- S-=
,4.3 uJ 4) (DJ (A (M -- f,- ,--
W W)(.
Q ".- C C,...0 "- '-
U (V- , M OS-, W- CD
"0 CO >.04)J 00 D f,.-- t.,.-d,
McU
D 0-- EWM LUi 0 .L
4- 4) 4.2 (M c "4.-: c W >1 0,.- >)0",-
S.W•. VS- - (L 'Wa1) > -, r-. gO
V. 0
~cW3*-E
M Wn4-E
.C0.C
4J-4- > I.-- O t7.,-
co>
D: r to
0o>
ge
0
-41.
o in >v' ,*
._j 4-) 4. i4
S4.).
to oC
%..4n . 4c r- C-
W WL- W
(A WUW 4 0W
S. • -C 0-c
u wc
WW
S0C _r= 0
0 4-34
CM
S- 4) 4
E u- a* #
0...
oj tot
4) 04 C 4J0) 0
tin
Ed IA i
to W 4) inX
xC
u W-
S.. to U - toC jU
Ed- (A3
'CC
3:x
~ ,.
to 0 4 01to
1.1.01 CA 4)
+-
a,3 "0 +3 r- 0
(A
0.C~~~ 03Oi - 4
'a 0 -J r-04- 1 3 0 0
ý 4- C~ =~ 4 r
CL
to >0 (A *.4 0 0 A
0J-- 4JliC4- S W>% 01. >j

Cl -&- *(-
In 0.'0 O
uJ4u'a).1
-0
to ( ) in) to~ r- U
to1 c L.V f
013C
C 4-) .
*d) 4 41 S- u' 0
4- -
04-.
U U 0 4-- 0
0 r 4 .1 0
4- ~ 0
CC 0
S- 04J
Mi r54- 01)
o1 #A
4,J 3:0
x n LL. 0. 01 ) C- J
in0 40
400 LA-
4)~0 C 1. 00
04j
0L
CD .4 S- '
'0
0 0 1 0
.4J 4)1
In S 06 4 a)4-
>) ) i.
41-
4)i 41
U- CA-vi
C ~~(>1.4.3 . ~ 4
CLto UV> 4)- 4
E~ LL 4J X
0L U29. ua4)1
Lii 4)01
'V..d 00 ..
)u 41 M.-
0
CL S4> to011
- u 0 4J
4) 03 0(A Ca -
4
uJ1 4-)- - w
4O 1.0
0
4 Or- w U-)
0C-U r=1 -
~,. 054-4-
S-0
4A-
4
0 *1
- 4-)4
4) C IAU.
1.E inC" ..
0 000
xi
TABLE OF CONTENTS
Page
1.0 INTRODUCTION ................................................. 1-1

1.1 Objective ............................................. 1-1
1.2 Background ............................................ 1-1
1.3 Definitions ........................................... 1-4
2.0 DATA/INFORMATION COLLECTION .................................. 2-1
2.1 Literature Search ...................................... 2-1
2.2 Data Collection Approach .............................. 2-4
2.3 Data Summary .......................................... 2-
3.0 DATA ANALYSIS OVERVIEW ....................................... 3-1
3.1 Data Deficiencies ..................................... 3-1
3.2 Statistical Methods ................................... 3-3
3.3 Zero Failure Data ..................................... 3-8
3.4 Effect of Nonoperating Failures on Operating Models... 3-11
4.0 NONOPERATING FAILURE RATE MODELING CONCEPTS .................. 4-1
4.1 Failure Rate Modeling Approach ........................ 4-1
4.2 Theoretical Model Development ......................... 4-6
4.3 Equipment Power Cycling Effects ....................... 4-7
4.3.1 Dormanc) and Power On-Off Cycling Effects on. 4-8
Electronic Equipment and Part Reliability
(RADC-lTR-73-248)
4.3.2 Planning Research Corporation Studies ........ 4-9
4.3.3 ARINC Study .................................. 4-10
4.3.4 Hughes Presentations ......................... 4-11
4.3.5 Equipment Power Cycling Conclusions .......... 4-11
4.3.6 Eouipment Power Cycling Analysis ............. 4-12
4.4 Temperature Effects ............. ..................... 4-16
4.5 Environmental Factor Analysis ......................... 4-22
4.6 Screening Effectiveness ............................... 4-44
5.0 NONOPERATING FAILURE RATE MODEL DEVELOPMENT .................. 5-1
5.1 Microcircuits ...... .................................. 5-4
xii
TABLE OF CONTENTS (CO)NT'D)
Page
5.1.1 Monolithic Microclrcumt Nonoperating Failure. 5-4
Rate Predictior Models
5.1.2 Digital Microcircuit Model Development ....... 5-9
5.1.3 Linear/Interface Microcircuits ............... 5-38
5.1.4 Memory Device Model Development .............. 5-50
5.1.5 Monolithic Microcircuit Model Validation ..... 5-52
5.1.6 Hybrid Microcircuit Nonoperating Failure ..... .5-57
Rate Model,
5.1.7 Hybrid Model Development ....................... 5-58
5.1.8 Proposed, Magnetic Bubble Memory Nonoperating.. 5-74
Failure Rate Prediction Model
5.1.9 Magnetic Bubble Memory Model Development ...... 5-76
5.2 Discrete Semiconductors ................................ 5-79
5.2.1 Discrete Semiconductor Nonoperating Failure... 5-79
Rate Prediction Models
5.2.2 Transistor Model Development .................. 5-86
5.2.3 Diode Model Development ....................... 5-101
5.2.4 Opto-electronic Semiconductor Model .......... 5-116
Development
5.2.5 Model Validation ......................... 5-120
5.3.1 Proposed Resistor Nonoperating Failure Rate... 5-125

Prediction Model
5.3.2 Model Development .......................... 5-127
5.3.3 Model Validation .............................. 5-148
5.4 Capacitors... .................. ............. 5-152
5.4.1 Proposed Capacitor Nonoperating Failure Rate.. 5-152
Prediction Model
5.4.2 Model Development ............................. 5-153
5.4.3 Model Validation ......................... 5-173
xiii
t/
TABLE OF CONTENTS (CONT'D)
Page
5.5 Inductive Devices ............ .................... 5-176
5.5.1 Proposed Inductor Nonoperating Failure Rate... 5-176
Prediction Model
5.5.2 Inductor Model Development .................... 5-178
5.6 Lasers ................................................. 5-191
5.6.1 Proposed Laser Nonoperating Failure Rate ...... 5-191
Prediction Models
5.6.2 Laser Model Development ....................... 5-194
5.7 Tubes .................................................. 5-199
5.7.1 Proposed Tube Nonoperating Failure Rate ....... 5-196
Prediction Model
5.7.2 Tube Model Development ........................ 5-197
5.8 Mechanical/Electromechanical Devices .................. 5-208
5.8.1 Proposed Mechanical/Electromechanical Device.. 5-208
Nonoperating Failure Rate Prediction Models
5.8.2 Rotating Mechanisms ........................... 5-212
5.8.3 Contact Device Model Development ............... 5-215
5.8.4 Connectors .................................... 5-227
5.9 Interconnection Assemblies ............................. 5-232
5.9.1 Proposed Interconnection Assembly Nonoper-.... 5-232
ating Failure Rate Prediction Model
5.9.2 Interconnection Assembly Model Development .... 5-233
5.10 Connections .... ........................................ 5-240
5.10.1 Proposed Connections Nonoperating Failure ..... 5-240
Rate Prediction Model
5.10.2 Model Development ............................. 5-241
5.11 Miscellaneous Parts .................................... 5-244
5.11.1 Proposed Miscellaneous Parts Nonoperating ..... 5-244
Failure Rates
5.11.2 Model Development ........................ 5-245
xiv
TABLE OF CONTENTS (CONT-D)
Page
6.0 APPLICATION OF NONOPERATING RELIABILITY MODELS ............... 6-1
.6.1 Comprehensive Reliability Models ...................... 6-1
6.2 Proposed Comprehensive Reliability Prediction Method.. 6-3
7.0 COMPARISON OF OPERATING AND NONOPERATING FAILURE RATES ........ 7-1
4b8.0 CONCLUSIONS .............................................. 8-1
9.0 RECOMM4ENDATIONS ........................ 9-1
REFERENCES..................................................... R-1
Appendix A: Proposed Nonoperating Failure Rate Prediction Models. A-i
xv
LIST OF FIGURES
Page
FIGURE 3.3-1: CONCEPTUAL FAILURE RATE DISTRIBUTION ................ 3-10
FIGURE 4.1-1: MODEL DEVELOPMENT FLOW CHART ........................ 4-2
FIGURE 4.4-1: LINEAR MICROCIRCUIT STORAGE FAILURE RATE VS ......... 4-19
TEMPERATURE
FIGURE 4.5-1: FAILURE MECHANISM DISTRIBUTION, ACCELERATED BY ...... 4-36
ENVIRONMENTAL STRESS (CASE I)
FIGURE 4.5-2: FAILURE MECHANISM DISTRIBUTION, ACCELERATED BY ...... 4-36
OPERATIONAL STRESS (CASE II)
FIGURE 4.5-3: FAILURE MECHANISM DISTRIBUlION, ACCELERATED BY..... 4-37
COMBINED OPERATIONAL/ENVIRONMENTAL STRESS (CASE III)
FIGURE 4.6-1: CONCEPTUAL SCREENING EFFECTIVENESS COMPARISON ....... 4-46
FIGURE 5.1.2-1: FAILURE RATE VS. TEMPERATURE ........................ 5-24
xvi .
LIST OF TABLES
Page
TABLE 2.1-1: LITERATURE SEARCH RESOURCES ........................ 2-3
TABLE 2.2-1: NONOPERATING FAILURE RATE DATA SOURCES ............... 2-7
TABLE 2.3-1: SUMMARIZED NONC hERATING FAILURE RATE DATA ............ 2-8
TABLE 3.2-1: EXAMPLE OF QUALITATIVE REGRESSION ANALYSIS ........... 3-5
TABLE 4.5-1: ENVIRONMENT CATEGORIES ............................. 4-25
TABLE 4.5-2: MIL-HDBK-217D ENVIRONMENTAL FACTOR CLASSIFICATION .... 4-32
TABLE 4.5-3: ENVIRONMENTAL FACTOR ANALYSIS (CASE II) -............ 4-41
SIMULATANEOUS EQUATIONS
TABLE 5.0-1: PART CLASS CATEGORIZATION .......................... 5-1
TABLE 5.0-2: MODEL DEVELOPMENT OVERVIEW ................ ....... 5-2
TABLE 5.1.1-1: DIGITAL MICROCIRCUIT NONOPERATING TEMPERATURE ........ 5-6
FACTOR CONSTANTS
TABLE 5.1.1-2: MICROCIRCUIT NONOPERATING QUALITY FACTORS ............ 5-6
* TABLE 5.1.1-3: MICROCIRCUIT NONOPERATING ENVIRONMENTAL FACTORS ...... 5-7
(wNE)
TABLE 5.1.2-1: DIGITAL MICROCIRCUIT CHARACTERIZATION VARIABLES ...... 5-11
TABLE 5.1.2-2: DIGITAL MICROCIRCUIT NONOPERATING FAILURE RATE DATA.. 5-12
TABLE 5.1.2-3: DIGITAL MICROCIRCUIT TEMPERATURE MATRIX .............. 5416
TABLE 5.1.2-4: DIGITAL MICROCIRCUIT QUALITY VARIABLE MATRIX ......... 5-18
TABLE 5.1.2-5: DIGITAL MICROCIRCUIT REGRESSION RESULTS .............. 5-19
TABLE 5.1.2-6: DIGITAL MICROCIRCUIT NONOPERATING QUALITY FACTOR..... 5-21
TABLE 5.1.2-7: MICROCIRCUIT NONOPERATING TEMPERATURE FACTOR ......... 5-23
"COEFFICIENTS
TABLE 5.1.2-8: MOS DIGITAL SSI/MSI FAILURE MECHANISM DISTRIBUTION... 5-32
TABLE 5.1.2-9: BIPOLAR DIGITAL SSI/MSI FAILURE MECHANISM ............ 5-33
DISTRIBUTION
TABLE 5.1.2-10: RANDOM LOGIC LSI FAILURE MECHANISM DISTRIBUTION ...... 5-34
TABLE 5.1.2-11: K3 and K4 TEMPERATURE FACTOR CONSTANTS FOR ........... 5-37
DIGITAL MICROCIRCUITS
TABLE 5.1.3-1: LINEAR/INTERFACE MICROCIRCUIT CHARACTERIZATION ....... 5-39
"VARIABLES
xvii . .~
-
-
.
.
W1ý VW
~ 4'4
VW~~
N
LIST OF TABLES
Page
TABLE 5.1.3-2: LINEAR/INTERFACE MICROCIRCUIT NONOPERATING FAILURE... 5-40
RATE DATA
TABLE 5.1.3-3: LINEAR/INTERFACE INITIAL REGRESSION RESULTS .......... 5-43
TABLE 5.1.3-4: LINEAR/INTERFACE FINAL REGRESSION RESULTS ............ 5-45
TABLE 5.1.4-1: MEMORY DEVICE NONOPERATING FAILURE RATE DATA......... 5-49
TABLE 5.1.5-1: MONOLITHIC MICROCIRCUIT MODEL VALIDATION DATA ........ 5-56
TABLE 5.1.6-1: HYBRID NONOPERATING ENVRIONMENT FACTORS (rNE) ......... 5-59
TABLE 5.1.7-1: HYBRID PART CHARACTERIZATION ....................... 5-60
TABLE 5.1.7-2: SUMMARIZED HYBRID NONOPERATING RELIABILITY DATA ....... 5-62
TABLE 5.1.7-3: HYBRID VARIABLE CORRELATION COEFFICIENT MATRIX ....... 5-64
TABLE 5.1.7-4: HYBRID REGRESSION RESULTS (0)........................ 5-68
TABLE 5.1.7-5: HYBRID REGRESSION RESULTS (II) ...................... 5-68
TABLE 5.1.7-6: HYBRID COEFFICIENT CONFIDENCE INTERVALS .............. 5-71
TABLE 5.1.7-7: HYBRID NONOPERATING QUALITY FACTORS .................. 5-72,
r TABLE 5.2.1-1: DISCRETE SEMICONDUCTOR NONOPERATING TEMPERATURE ...... 5-80
FACTOR PARAMETERS
TABLE 5.2.1-2: TRANSISTOR NONOPERATING ENVRIONMENAL FACTORS (ITNE)... 5-82
TABLE 5.2.1-3: DIODE NONOPERATING ENVRIONMENTAL FACTORS (ONE) ....... 5-84
TABLE 5.2.1-4: OPTO-ELECTRONIC SEMICONDUCTOR NONOPERATING ........... 5-85
ENVIRONMENTAL FACTORS (wNE)
TABLE 5.2.2-1: TRANSISTOR PART CHARACTERIZATION ..................... 5-87
TABLE 5.2.2-2: TRANSISTOR NONOPERATING FAILURE RATE DATA ............ 5-89
TABLE 5.2.2-3: TRANSISTOR STYLE VARIABLE MATRIX ..................... 5-90
TABLE 5.2.2-4: TRANSISTOR REGRESSION RESULTS II ..................... 5-92
TABLE 5.2.2-5: SIGNAL AND POWER TRANSISTOR FAILURE MECHANISM ...... .. 5-95•
DISTRIBUTION
TABLE 5.2.2-6: FET FAILURE MECHANISM DISTRIBUTION ................... 5-96
TABLE 5.2.2-7: HIGH TEMPERATURE STORAGE LIFE TEST DATA .............. 5-99
TAELE 5.2.3-1: DIODE CHARACTERIZATION VARIABLES ..................... 5-103
TABLE 5.2.3-2: DIODE NONOPERATING FAILURE RATE DATA................ 5-104
TABLE 5.2.3-3: DIODE STYLE VARIABLE MATRIX .......................... 5-106
TABLE 5.2.3-4: DIODE INITIAL REGRESSION RESUI.TS ..................... 5-107
xviiii
V
LIST OF TABLES
Page
TABLE 5.2.3-5: DIODE FINAL REGRESSION RESULTS ..................... 5-110
TABLE 5.2.3-6: DIODE NONOPERATING QUALITY FACTORS ................... 5-111

TABLE 5.2.3-7: ZENER DIODE FAILURE MODE/MECHANISM DISTRIBUTIONS ..... 5-112
TA13LE 5.2.3-8: SMALL SIGNAL DIODE FAILURE MODE/MECHANISM ............ 5-113
DISTRIBUTIONS
TABLE 5.2.3-9: DIODE NONOPERATING BASE FAILURE RATES ................ 5-115
TABLE 5.2.4-1: OPTO-ELECTRONIC DEVICE CHARACTERIZATION VARIABLES .... 5-117
TABLE 5.2.4-2: OPTO-ELECTRONIC NONOPERATING BASE FAILURE RATES ...... 5-120
TABLE 5.2.5-2: TRANSISTOR MODEL VALIDATION DATA ..................... 5-122
TABLE 5.2.5-2: DIODE MODEL VALIDATION DATA ......................... 5-123
TABLE 5.3.1-1: RESISTOR NONOPERATING ENVIRONMENTAL FACTORS ('TNE),,, 5-126

TABLE 5.3.2-1: RESISTOR PART CHARACTERIZATION ...................... 5-128
TABLE 5.3.2-2: RESISTOR NONOPERATING FAILURE RATE DATA ............. 5-130
TABLE 5.3.2-3: RESISTOR STYLE VARIABLE MATRIX ...................... 5-134
TABLE 5.3.2-4: RESISTOR INITIAL REGRESSION RESULTS ................. 5-136
TABLE 5.3.2-5: RESISTOR NONOPERATING QUALITY FACTORS ............... 5-140
TABLE 5.3.2-6: COMPOSITION RESISTOR FAILURE MECHANISM DISTRIBUTION. 5-142
TABLE 5.3.2-7: FILM RESISTOR FAILURE MECHANISM DISTRIBUTION ........ 5-142
TABLE 5.3.2-8: WIREWOUND RESISTOR FAILURE MECHANISM DISTRIBUTION,,, 5-143
TABLE 5.3.2-9: VARIABLE WIREWOUND RESISTOR FAILURE MECHANISM ....... 5-143
DISTRIBUTION
TABLE 5.3.2-10: VARIABLE COMPOSITION RESISTOR FAILURE MECHANISM ..... 5-144
DISTRIBUTION
TABLE 5.3.2-11: THERMISTOR FAILURE MECHANISM DISTRIBUTION........... 5-144
TABLE 5.3.2-12: RESISTOR NONOPERATING BASE FAILURE RATES ............ 5-146
TABLE 5.3.3-1: RESISTOR MODEL VALIDATION DATA...................... 5-151
I TABLE 5.4.1-1: CAPACITOR NONOPERATING ENVIRONMENTAL FACTORS ........ 5-154
TABLE 5.4.2-1: CAPACITOR PART CHARACTERIZATION ..................... 5-155
TABLE 5.4.2-2: CAPACITOR NONOPERATING FAILURE RATE DATA TABLE ...... 5-157
TABLE 5.4.2-3: CAPACITOR DEVICE STYLE VARIABLE MATRIX .............. 5-159
TABLE 5.4.2-4: CAPACITOR INITIAL REGRESSION RESULTS ................ 5-162
TABLE 5.4.2-5: CAPACITOR NONOPERATING QUALITY FACTORS .............. 5-167
xix
LIST OF TABLES
Page
TABLE 5.4.2-6: PAPER AND PLASTIC CAPACITOR FAILURE MECHANISM ....... 5-168
DISTRIBUTION
TABLE 5.4.2-7: CERAMIC CAPACITOR FAILURE MECHANISM DISTRIBUTION .... 5-168
TABLE 5.4.2-8: MICA CAPACITOR FAILURE MECHANISM DISTRIBUTION ....... 5-169
TABLE 5.4.2-9: GLASS CAPACITOR FAILURE MECHANISM DISTRIBUTION ...... 5-169
TABLE 5.4.2-10: WET FOIL CAPACITOR FAILURE MECHANISM DISTRIBUTION... 5-170
TABLE 5.4.2-11: SOLID TANTULUM CAPACITOR FAILURE MECHANISM .......... 5-170
DISTRIBUTION
TABLE 5.4.2-12: ALUMINUM ELECTROLYTIC CAPACITOR FAILURE MECHANISM.. 5-171
jol DISTRIBUTION
TABLE 5.4.2-13: VARIABLE CAPACITOR FAILURE MECHANISM DISTRIBUTION... 5-171
TABLE 5.4.2-14: CAPACITOR NONOPERATING BASE FAILURE RATES ........... 5-172
TABLE 5.4.3-1: CAPACITOR MODEL VALIDATION DATA ..................... 5-174
TABLE 5.5.1-1: INDUCTOR NONOPERATING ENVIRONMENTAL FACTORS ......... 5-177
TABLE 5.5.2-1: INDUCTOR PART CHARACTERIZATION ...................... 5-179
TABLE 5.5.2-2: INDUCTOR NONOPERATING FAILURE RATE DATA ............. 5-180
TABLE 5.5.2-3: DEVICE QUALITY VARIABLE MATRIX ...................... 5-183
TABLE 5.5.2-4: INDUCTOR INITIAL REGRESSION RESULTS ................. 5-184
TABLE 5.5.2-5: INDUCTOR NONOPERATING QUALITY FACTORS ............... 5-188
TABLE 5.5.2-6: RF COIL FAILURE MECHANISM DISTRIBUTION .............. 5-190
TABLE 5.5.2-7: TRANSFORMER FAILURE MECHANISM DISTRIBUTION .......... 5-190
TABLE 5.6.1-1: LASER NONOPERATING ENVIRONMENTAL FACTORS ............ 5-193I
TABLE 5.7.1-1: TUBE NONOPERATING ENVIRONMENTAL FACTORS ............. 5-197
TABLE 5.7.2-1: TUBE PART CLASSIFICATION 5-198
TABLE 5.7.2-2: TUBE NONOPERATING FAILURE RATE DATA .................. 5-201
TABLE 5.7.2-3: DEVICE STYLE VARIABLE MATRIX ........................ 5-202
TABLE 5.7.2-4: TUBE REGRESSION RESULTS ........................... 5-203
TABLE 5.7.2-5: TUBE OPERATING AND NONOPERATING FAILURE RATE ........ 5-207
COMPARISON
TABLE 5.8.1-1: ROTATING DEVICE AVERAGE NONOPERATING FAILURE RATES.. 5-209
TABLE 5.8.1-2: RELAY NONOPERATING ENVIRONMENTAL FACTORS ............ 5-210
TABLE 5.8.1-3: SWITCH NONOPERATiNG ENVIROIIMENTAL FACTORS ........... 5-211
xx
LIST OF TABLES
Page
TABLE 5.8.1-4: CONNECTOR NONOPERATING ENVIRONMENTAL FACTORS ........ 5-212
TABLE 5.8.2-1: ROTATING MECHANISM NONOPERATING FAILURE RATE DATA... 5-214
TABLE 5.8.3-1: RELAY PART CHARACTERIZATION ......................... 5-217
TABLE 5.8.3-2: SWITCH PART CHARACTERIZATION ........................ 5-218

TABLE 5.8.3-3: RELAY NONOPERATING FAILURE RATE DATA............... 5-220
TABLE 5.8.3-4: RELAY NONOPERATING QUALITY FACTORS .................. 5-222
TABLE 5.8.3-5: ARMATURE RELAY FAILURE MECHANISM DISTRIBUTION ....... 5-223
TABLE 5.8.3-6: TOGGLE SWITCH MECHANISM COMPARISONS ................. 5-225
TABLE 5.8.3-7: SWITCH NONOPERATING FAILURE RATE DATA............ 5-226
TABLE 5.8.4-1: CONNECTOR NONOPERATING FAILURE RATE DATA ............ 5-229
TABLE 5.9.1-1: INTERCONNECTION ASSEMBLY NONOPERATING ENVIRONMENTAL. 5-233
FACTORS
TABLE 5.9.2-1: INTERCONNECTION ASSEMBLY CHARACTERIZATION........... 5-235
TABLE 5.9.2-2: INTERCONNECTION ASSEMBLY NONOPERATING FAILURE RATE. 5-236
j DATA
TABLE 5.10.1-1: CONNECTIONS NONOPERATING ENVIRONMENTAL FACTORS...... 5-241
TABLE 5.11.1-1: NONOPERATING FAILURE RATES FOR MISCELLANEOUS PARTS.. 5-244
TABLE 5.11.2-1: MISCELLANEOUS PART NONOPERATING FAILURE RATE DATA... 5-245
STABLE 6.2-1: COMPREHENSIVE RELIABILITY PREDICTION TASKS .......... 6-3
TABLE 7.0-1: SAMPLE CALCULATIONS COMPARING NONOPERATING AND ...... 7-2
MIL-HDBK-217D MODELS
TABLE 7.0-2: RATIOS OF MIL-HDBK-217D TO NONOPERATING FAILURE ..... 7-6
RATES
TABLE 7.0-3: DISCRETE SEMICONDUCTOR OPERATING TO NONOPERATING .... 7-8
4
MFAILURE RATE RATIOS
rro
xxiI
1.0 INTRODUCTION
1.1 Objective
The objective of this study was to develop a procedure to predict the

quantitative effects of nonoperating periods on electronic equipment
I reliability. A series of nonoperating failure rate prediction models were
developed at the component level. The models are capable of evaluating
component nonoperating failure rate for any anticipated environment with
the exception of a satellite environment. A clear and concise procedure
was also developed to apply the component nonoperating failure rate
prediction models to assess the impact of nonoperating periods at the y
equipment level. Additionally, a comprehensive reliability prediction
methodology was presented which describes the use of the proposed
nonoperating fiilu-e rate prediction models together with the documented
operating failure rate assessment techniques (i.e. MIL-HDBK-217D).
Th? proposed nonoperating failure rate prediction methodology was

Sintended to provide the ability to predict the component nonoperating
failure rate and reliability as a function of the characteristics of the
p~i devices, technology employed in producing the device, and external factors
such as environmental stresses which have a significant effect on device
nonoperating reliability. An analytical approach using observed data was

taken for model development where possible. Thus, the proposed models
only include variables which can be shown to significantly affect
nonoperating failure rate. The prediction methodolugy is presented in a
form compatible with MIL-HDBK-217 in Appendix A.
.Z
1.2 Background
Failure rate and reliability prediction capabilities are essential

tools in the development and maintenance of reliable electronic
equipments. Predictions
performed during the design phase yield early
estimates of the anticipated equipment reliability and provide a
quantitative basis for performing proposal evaluations, design trade-off
"i 1 .-i
analyses, reliability growth monitoring and life-cycle cost studies. This

is particularly important when an electronic equipment is in a
U
nonoperating state for a prolonged period of time. Equipment nonoperating
reliability cannot be determined until power is applied. If no test
schedule has been established, an equipment in a nonoperating state has an
unknown reliability until the point in time when it is required.
Unexpected poor performance can have disastrous results. Therefore, it is
essential that nonoperating reliability assessment techniques are
"I implemented on the same scale as operating reliability assessment.
Manufacturers and government customers for missiles have long

recognized the need for nonoperating reliability assessment. Conversely,
U.S. Air Force airborne electronic equipments have often been assumed to
possess a nonoperating failure rate of zero. This has been an oversight,
particularly when it is considered that a typical fighter aircraft is
exposed to over twenty times more nonoperating time than operating time.
Even if the nonoperating failure rate is a small fraction of the
corresponding operating failure rate, the total number of nonoperating
failures may be equivalent to the number of operating failures. Two
studies sponsored by RADC (References 1 and 2) identified this oversight
as a possible reason that operating failure rate predictions differ from
observed field data (where it is generally assumed that all failures are
operating failures).
Previous attempts (References 3, 4, 5 and 6) to predict nonoperating

reliability have generally been in one of two categories. Either (1) an
equipment level multiplicative "K" factor has been applied to an equipment S,
operating failure rate, or (2) operating failure rate relationships have
been extrapolated to zero stress. The first method has merit under
controlled circumstances. The second method hias little or no merit. As a
general nonoperating failure rate prediction methodology which can be hF7
applied for any equipment in any environment, neither of these previous

methods are acceptable.
1-2
Use of a multiplicative "K" factor has merit under certain

circumstances. The "K" factor can be accurately used to predict
nonoperating failure rate if it was based on equipment level data from the
same contractor on a similar equipment type with similar derating and
I.• screening.
approximate method
assume that
In any other circumstances,
operating
at best.
and
Additionally,
nonoperating
the use of a "K" factor is a very
it
failure
is intuitively wrong to
rates are directly
proportional. Many application and design variables would be anticipated
"to have a pronounced effect on operating failure rate, yet negligible
effect on nonoperating failure rate. Derating is one example. It has
been observed that derating results in a significant decrease in operating
failure rate, but a similar decrease would not be expected with no power
applied. Additionally, the stresses on parts are different in the
nonoperating state, and therefore, there is no reason to believe that the
I operating factors for temperature, environment, quality and application
would also be applicable for nonoperating reliability prediction purposes.
"An invalid approach for nonoperating failure rate assessment has been
to extrapolate operating failure rate relationships to zero electrical
stress. All factors in MIL-HDBK-217D, whether for electrical stress,
temperature or another factor, represent empirical relationships (as
opposed to theoretical relationships). An empirical relationship is based
on observed data, and proposed because of the supposedly good fit to the
data. However, empirical relationships may not be valid beyond the range
"of parameters found in the data and this range does not include zero
* electrical stress for MIL-HDBK-217D operating reliability relationships.
Extrapolation of empirical relationships beyond the range found in the
data can be particularly dangerous when the variable is part of an
exponential relationship. A relatively small error in the exponent can
correspond to a large error in the resultant predicted failure rate.
Additionally, there are many intuitive or qualitative reasons why small
amounts of applied power can be preferable to pure storage conditions.
For nonhermetic microcircuits, the effect of humidity is the primary
failure accelerating stress. A small current will result in a temperature
rise, burning off moisture, and probably decreasing device failure rate.
1r-
Also, the detrimental effects of equipment power on-off cycling would be

* expected to be less for any electronic part when a small current is
applied to the part. Another example where nonoperating failure rates
* could be expected to be higher than low stress operating failure rates is
for storage degradation components such as electrolytic capacitors, motors
* and electromechanical devices. These part types benefit from periodic
operation.
One of the assumptions which dominated these two approaches was also
necessary in this study. No empirical nonoperating failure rate data were
available for other than ground based environments. Nonoperating
environmental factor values were, therefore, determined based on an in-
depth study and comparison of operating and nonoperating failure
mechanisms and failure causing stresses. However, for the most part, the
factors and models presented in this report represent empirical
nonoperating relationships determined from observed nonoperating failure
rate data.
1.3 Definitions
Many terms have been used during previous studies of nonoperating

reliability. Unfortunately, the definitions given in the literature are
hardly standardized, and in many cases conflicting. It was not considered
desirable to develop unique definitions for this study. This would only
further confuse the issue. Therefore, the appropriate definitions, which
were used for this study, were based on a review of the available
literature, and were adopted from Reference 7.
Operating - Operating is defined as the state of a subsystem, assembly

or component when it is activated (as designed) by electrical or
mechanical means at any level of stress. At the subsystem level, if any
portion of the subsystem is operating, then the entire subsystem is
considered to be operating.
1-4
Nonoperating - A subsystem, assembly, or component is considered to be

nonoperating when it is experiencing none of the electrical or mechanical
stresses inherent in the (designed) activation of that subsystem,
assembly, orscomponent. It may however be experiencing stress caused by
the environm~ent, transportation and handling, such as captive carry G-
forces, etc.
Storage - Storage is defined as the state in which a system,

subsystem, assembly, or component is zero percent activated, and is in its
normal configuration in a storage area.
Dormancy - Dormancy is defined as those states wherein an equipment is

in its normal operation configuration and is not operating, or is
maintained in operationally ready storage. Dormancy includes the
nonoperating portions of alert, captive carry, transportation and
handling, and launcher carriage. rThe operationally ready storage mode is
predominant in that this state is where "long periods" of dormancy occur.
The ability of a system to withstand these "long periods" may be
21 influenced by relatively short periods of test time or the stress inherent
in other states such as transportation, captive carry, or launcher
carriage.
Shelf *Life -According to AFR 136-1, "Nonnuclear Munitions Product

Assurance Program," September 1979, shelf life is defined as: the length
* Of time an item may remain in storage under prescribed packaging and
storage conditions, and operate satisfactorily when removed from storage.
Equipment Power On-Off Cycle An equipment power on-off cycle is

defined as that state during which. an electronic system goes from the zero
electrical activation level to its normal design system activation level
plus that state during which it returns to zero.
Equipment Power On-Off Cycling Frequency - Power cycling frequency is

defined as the number of equipment power on-off cycles for a given time
* interval (1000 nonoperating hours was used for this study).
1-5
S
It should be noted that definition of dormancy can vary
the
considerably. According to Reference 5, dormancy is equal to any
condition where the electrical activation level is less than or equal to
10% of the normal design level. The dormancy definition adopted for this
study was consistent with information defined in the RADC statement of
work explicitly defining nonoperating.
1-6
2.0 DATAINFORMATION COLLECTION
The proposed nonoperating failure rate prediction model development

approach involved the analysis of large amounts of empirical data.
Therefore, the data collection task was a prerequisite for numerical
analyses and an integral part of the overall process. Additionally, the
proposed models were scrutinized with a theoretical evaluation. Thus, a
thorough literature search was required to locate pertinent information.
This section describes the literature search, the data collection approach'
and the available data.
4 2.1 Literature Search
The literature search was an essential part of the study approach.

Information obtained through the literature search was used to develop
theoretical models, evaluate the proposed models and complement the
analyses for part families without sufficient data. To insure an
efficient and successful'literature search, a concise methodology was
developed to locate applicable nonoperating reliability information.
Two major information sources were used. The first source consisted
of libraries and other data resources covering open technical literature.
3 The second source consisted of contacts established in past and present
electronic systems reliability projects. This included contacts whose
work has not yet been published in the open literature. Use of these
diverse sources ensured that a comprehensive review of the field was
achieved. Four technical areas which were extensively researched are:
o Actual equipment level environmental stresses (environmental

profiles).
o Component susceptibility to environmental stresses.

o Equipment level operating scenarios (percentage of on/off time per
mission and distribution of mission types).
o Theoretical nonoperating reliability relationships for
temperature, environment, and equipment power on-off cycling.
The information gathered for the first two areas listed was used in
the development of nonoperating environmental factors. The information
gathered for the third area was sought for all equipment environments.
The information was occasionally required for the derivation of the number
of nonoperating hours. This information was necessary to develop a
nonoperating failure rate from data collected on equipments which did not
contain elapsed time meters or where the time numeric recorded was flight
hours. The information in the fourth area was used to aid in the
development of theoretical models and to complement the data analysis
* task.
In order to efficiently conduct a comprehensive literature search it

was necessary to define and implement a methodology, whose critical
elements included the following:
o Problem/goal definition
In this step, the key concepts of the search were defined and also
any related areas that could potentially yield any information
were identified. Other factors that were considered at this stage
included the time span of the search, and its general scope.
0 Identification of information resources
The main activity in this area consisted of identifying the
relevant abstracts, indexes, reference works and technical
journals for the problems defined in the previous step.
o Search strategy formulation
A search strategy was devised which identified those information
resources which could most effectively yield the information
defined in the first step of the process.
o Literature Survey
Using the search strategy as a roadmap, the information specialist
surveyed the information resources for potentially relevant
information. Both manual and automated search methods were used.
o Evaluation
This was the most important step. The results of the search were
reviewed and search strategy was redefined as required.
2-2
o Literature Search
An in-depth search was then conducted. Close contact was
maintained between the project engineer and information specialist
to insure the goals of the search were met.
A variety of information resources were used. Table 2.1-1 presents a

brief description the more useful resources.
TABLE 2.1-1: LITERATURE SEARCH RESOURCES

Resource Description
Defense Technical Information DTIC maintains a large computer-

Center (DTIC) ized database of technical
documents produced by government
sponsored efforts.
Reliability Analysis Center RAC is a DoD information analysis
(RAC) center primarily concerned with
J, electronics component and system
reliability. The center has an
automated library and data base
with numerous hardware reliabil-
ity references.
IITRI Computer Search Center This service located at the IITRI

(CSC) Chicago office is staffed by pro-
fessi-nal information specialists
who are experts in searching com-
puterized bibliographiic data
bases. The Center has access to
approximately 150 individual data
bases.
Government Industry Data The GIDEP data base contains four
Exchange Program (GIDEP) separate data banks. Of these
the Engineering Data Bank, the
Reliability-Maintainability Data
Bank, and the Failure Experience
Data Bank were most relevant to
this study.
The literature search task was very successful. The most relevant
documents and technical articles are listed in the References section of
this report.
2-3
2.2 Data Collection Approach
The development of nonoperating failure rate prediction methodologies

should be derived from field failure rate data representing a large range
of application and construction variables. A general data collection
approach was developed to efficiently collect a large data base.
The Reliability Analysis Center (RAC) operated by the lIT Research

Institute at Griffiss Air Force Base was solicited to aid in the data
collection process. Numerous contacts in both government and industry
have been established at RAC as part of the regular data collection
process. Additionally, high temperature storage life test data for
microcircuits were available as part of the RAC microcircuit data base.
Two key conditions caused data collection to be particularly difficult

for this study. First, the objective of the study was to develop
nonoperating failure rate prediction models for all part classes in MiL-
HDBK-217D. The range and variety of the part types included in MIL-HDEXK-
217D is extremely large. Data collection could not be concentrated on one
generic part family at the expense of other part types. The other
condition which made data collection difficult was the inherently low
failure rate which many part types exhibit in storage or dormant
applications, and the further fact that many of the part types fall in the
category of low population parts. Many potential data sources could not
be used simply because insufficient nonope,-ating time had accumulated to
expect any failures.
To insure an efficient and effective data collection process, five

criteria were established for an acceptable data source. Each potential
data source was evaluated with these criteria before proceeding with data
summarization. These five criteria were:
1. Data available to the part level.
2. Primary failures can be separated from total maintenance actions.
2-4
!•.
• •2-•.• 5, • F••t•
r'W•%• •?•, • ~ er -• • •¢••¢• •;u: •-v• • ¢.-•--w--t.-- < I
3. Nonoperating failures can be separated from operating failures.
4. Sufficient detail can be identified for components.
5. Sufficient equipment nonoperating hours to expect failures.
These five attributes were used as a guide

to determine suitable
candidate equipments. Information collection trips were made to (1) U.S.
Army MICOM (Missile Command), Redstone Arsenal, (2) U.S. Army ARRADCOM
(Armament Research and Development Command), (3) U.S. Navy Sea Systems
Command, (4) U.S. Department of Commerce, and (4) Tobyhanna Army Depot to
evaluate potential sources of data. These organizations were chosen based
on information provided in initial telephone contacts. In person contact
was required to emphasize the importance of this study, get acquainted
with additional reliability data bases, retrieve raw data and/or inspect
raw data more closely. Many sources of information could be accessed by
telephone or written requests and did not require on-site visits.
Telephone contact was made with numerous other industrial and government
organizations including the Product Performance Agreement Center at
Wright-Patterson Air Force Base. The conclusion was reached after
evaluation of the potential sources to concentrate data collection efforts
on large, preferably automated data bases which had already been
summarized (i.e. nonoperating failures identified, part characterization
performed, part hours computed) and pertained only to nonoperating
reliability. The separation of nonoperating and operating failures proved
to be difficult and prevented the inclusion of many possible data sources.
The Product Performance Agreement Center at Wright-Patterson Air Force

Base provided information regarding U.S. Air Force equipments purchased
under Reliability Improvement Warranty (RIW) contracts. In general, RIW
failure reporting is preferable to that of large automated military
U reliability data bases. The reporting allows for decisions to be made
regarding primary versus secondary failures, and is generally more
complete. As with other data sources, the issue of separating
nonoperating failures from total failures was difficult. The F-16 heads-
2-5
up-display (HUD) was selected from the list of RIW contracts to summarize
for nonoperating reliability data. The F-16 HUD met each of five
requirements for an acceptable data source. Additionally, there were a
relatively wide range of component styles included in the design.
Decisions for operating versus nonoperating failures were made using the
on-equipment maintenance action "'when discovered" code.
The Storage Reliability of Missile Materiel Program maintained by U.S.

Army MICOM, Redstone Arsenal provided the best source of nonoperating
failure rate data. Nonoperating failure rate data were available for a
wide range of part types for a number of missile programs. A summary of
this data is presented in Reference 3. Time and budget constraints would
have prevented independent summarization of a data base as large as the
MICOM data base. MICOM has periodically issued a set of documents
presenting the data in various formats, describing data analyses, and
presenting nonoperating failure rate prediction models for missile
electronics. A 1982 revision of these documents was issued and then
) recalled. This action was not indicative of poor or inaccurate data, but
was necessary because of errors in presentation of the data and prediction
models, and clerical errors. In fact, the same sources of data were used
in the 1982 documents as in the previous revisions (which did not include
errors). Therefore, the use of this data and the problems encountered by
MICOM do not adversely affect in any manner the validity or accuracy of
the analyses or proposed nonoperating failure rate prediction models
presented in this technical report. The successful completion of this
study relied heavily on the use of large summarized data from sources such
as MICOM.
U
Another organization which contributed data was the French group
Association Francaise pour le Controle Industriel et la Qualite (AFCIQ).
This group produced a document (Reference 8) which includes a sumlary of
the MICOM data, as well as additional data from European sources. This
data were used for model validation for microcircuits, discrete
2-6
I
II
semiconductors, resistors and capacitors, and for model development for
the remaining part styles with data.
Many other data sources were identified and evaluated as part of the
data collection approach. Table 2.2-1 presents a list of the
organizations and/or equipments which supplied data in sufficient quality
and quantity for numerical analysis.
TABLE 2.2-1: NONOPERATING FAILURE RATE DATA SOURCES
Source Equipment Data Type

MICOM Missile electronics field/test
F-16 HUD/RIW HUD field (RIW)
RAC S~testing N/A high temperature
Sandia National Labs nuclear weapon field
electronics
U.S. Army ERADCOM N/A high humidity
testing
U.S. Army TSARCOM generator sets field
Planning Research Corp. satellite field
Martin Marietta various field
AFCIQ various field
2.3 Lata Summary
A summary of the collected nonoperating failure rate data by device

style is presented in Table 2.3-1. Table 2.3-1 presents the sum total of
observed failures and part hours. A more detailed list of the data,
including data sources, is included in the individual model development
sections.
2-7
?• 2-7
TABLE 2.3-1: SUMMARIZED NONOPERATING FAILURE RATE DATA

A.\
Part Class Failures Part Hours (XlO6 )
Random Logic Microcircuits 155 21441.1
Linear/Int'erface Microcircuits 76 3605.8

Memory Microcircuits 122 814.6
Hybrid Microcircuits 2082 50049.9
Transistors 121 21983.1
Diodes 57 22180.6 _
Resistors 34 113118.9
Capacitors :
IA
48 41840.6
Inductive Devices 87 49833.0
Tubes 364 794.8 ••
Relays 36 1360.4
Switches 35
j_
408.3 •'i
Connectors 1 82444.0
2-8
3.0 DATA ANALYSIS OVERVIEW
The nonoperating failure rate prediction models presented in this

study are primarily based on numerical analysis of observed data. This
section presents a brief overview of the applicable data analysis
techniques. Additionally, inherent problems with the available data are
discussed, and the treatment of zero failure data records is described.
3.1 Data Deficiencies
Analytical evaluation of the nonoperating failure rate for electronic

or electromechanical components requires a large data base. Development
of nonoperating failure rate data bases which have sufficient quantity and[I
detail is difficult, if not impossible. This section presents a brief
overview of inherent problems with the available data and data quality
control measures implemented to minimize error. A more detailed -,l
description of data deficiencies and data quality control is presented in
Reference 9.
Available sources of nonoperating failure rate data were generally

either high temperature storage life test data or equipment level field
experience data. Each type of datahas several inherent difficulties.
Life test data generally are of a high statistical quality because

there is very little uncertainty with regard to recorded failures, number
of parts on test, test time and environmental conditions. The majo•
deficiencies with life test data are (1) the data often consists of many
parts cn test for a relatively short time each, and (2) the test
conditions are usually not representative of the actual usage environment.-
Field experience data are the more desirable type of data. This type
of nonoperating failure rate data represents what actually occurred in the
field, which is what the proposed model attempts to predict. The inherent
difficulties with field experience data are related to the accuracy with
which a failure can be defined, the precision with which the number of
Io
4
3-1
4
nonoperating part hours can be measured, and the ability to determine the
environmental stresses applied to a part.
Several classes of parts degrade during nonoperating periods and would

be anticipated to have a time dependent failure rate. A problemn
associated with available sources of field experience data was ti.at
individual nonoperating times to failure could not be determined. Data
were only available in the form of X failures observed in Y part hours.
The nonoperating part hours are a cumulative count of nonoperating 'Ours
from individual components. The result of this data deficiency is that
the exponential reliability function (i.e., constant failure rate) mu;L 'Ce
assumed for all part types. For most electronic parts, it was rot
believed that this assumption introduced significant error. However, 'or
electromechanical parts, electrolytic capacitors, tubes and other part
types where degradation failure mechanisms are significant, the constant
failure rate calculated by dividing the observed failures by the
nonoperating part hours may not be meaningful. This is particularly true
when the data consists of many parts on test (or fielded) for a relatively
short time each. To minimize error, data sources were sought with long
storage times per equipment. In these instances, the calculated constant
nonoperating failure rate represents an average failure rate value for the
data collection interval. The average failure rate is equal to the
failure rate due to random stresses plus an average failure rate
contribution due to wearout or degradation stresses.
Field nonoperating failure data samples for most electronic and

electromechanical parts are
necessarily restricted because the average
L
mean time to failure (approximately 106 to 1010 part hours) is longer than
the technology has been available. However good the failure rate data, it
can only cover the first few percentiles of the probability density
function. One result of these relatively high mean time to failures for A!
most part types during nonoperating periods is data records with zero
observed failures. The presence of zero failure data records weakens the
analysis. There exists no accurate method for determining a nonoperating
failure rate without observEd failures. However, these entries could not
3-2
--
,~-,-,.,.-,.*-. -- -- 4 -- -- -- -- -- -0-
-glected. In somte instances, a large percentage of the collected data

ifailure records. In other instances, the zero failure data
indicate that the predicted failure rate should be lower than what
the data records containing failures indicates. The treatment of zero
failure data records is discussed in Section 3.3.
Another data deficiency was related to the data collection approach.

Data were collected from any and all sources. This was a necessary
approach because of the anticipated lack of data. However, the resulting
data bases often contained variables, (i.e. part types, screen level,
application environment, etc.) which were correlated. Statistical
analysis can not be correctly applied if the variables are not reasonably
independent. Error was minimized in these instances by carefully choosing
independent variables, and by complementing the data analysis task with
theoretical or empirical relationships located during the literature
search.
3.2 Statistical Methods
This section presents a very brief discussion of statistical methods

*used in this study. References are provided which include much greater
detail.
Stepwise Multiple Linear Regression Analysis. Regression analysis is

an important statistical tool and was used to develop the nonoperating
failure rate prediction models for the majority of the devices. A more
thorough discussion of stepwise multiple linear regression analysis is
given in Reference 10. A brief description follows.
The stepwise multiple linear regression analysis technique assumes a

preliminary model of the form
Y b+
o b1 X1 +b 2 X2 + + biXi
3-3
where Y is the resultant dependent variable, X1 , X2 ,...,XI are the

independent variables which are thought to influence the value of Y, and
bo, bl, b2 ,...,bi are the coefficients which are to be found by the
regression.
To perform a regression, a number of data points, each consisting of a

known Y and its corresponding X variables are required. A proper
regression requires that the X variables are independent and that there
are many more data points than X variables. Estimates of the bi
coefficients are then made by minimizing the sum of the squared residuals.
The significance of independent variables can be analyzed by comparing

the F-ratio to the critical F numeric for a specified confidence level.
P In this manner, different regression solutions can be found depending on
the specified confidence level. Only independent variables with a F-ratio
greater than the critical F are included in the solution.
Nonoperating failure rate prediction models are rarely assumed to be

in the additive form of the regression solution. However, by using
transformations, many possible model forms can assume the additive form.
An example can best illustrate this point. The equivalent Arrhenius
relationship was determined to be applicable to the nonoperating failure
rate of microcircuits, and takes the following form,
X = A exp(-B/T)
where T is the independent variable, X is the dependent variable and A and

B are constants. By taking the natural logarithm of each side, the
equation becomes,
i InX = lnA - TB
which can be solved by regression analysis with 1/T the independent

variable and lnX the dependent variable. Other transformations a-e
3-4
available such that stepwise multiple linear regression can be used to

quantify a variety of failure rate model forms.
The previous paragraphs have discussed how regression analysis can be

useful in developing nonoperating failure rate prediction models in which
failure rate is a function of quantitative variables such as temperature
or number of gates.. However, there are often si~gnificant variables which
can not be measured on a continuous quantitative scale. Application
environment and quality level are examples of variables which are
* qualitative. To determine numerical quantities for qualitative factors in
a regression analysis, a matrix of "dumm~y variables" (0 or 1) is used as
,Athe independent variables. The regression solution by least squares
:computes
~z- numerical values of the coefficients which can be used to
determine numerical factors.
An example can best illustrate qualitative regression. Consider a

* part type which is represented by a multiplicative model and has four
clearly defined quality levels based on the amount of screening. The four
quality levels are signified as L1, L2 , L3 and L.4. The following matrix
given in Table' 3.2-1 presents quality level as a function of three
qualitative I'dunmny variables" (Q1, Q2, Q3).
TABLE 3.2-1: EXAMPLE OF QUALITATIVE REGRESSION ANALYSIS
Quality Level Q1 Q2 Q
Ll 0 0 0
L21 0 0
L0 1 0
L4 0 0 1
*Nonoperating failure rate data would be introduced into the analysis

for Ll parts by setting (Q1, Q2, Q3) = (0,0, 0). Data would be similarly
entered into the analysis for L2 , L3 and L4 parts. Determination of
coefficients for Q1, Q2 and Q3 allows for computation of quality factor
3-5
I ROW,-.v*-.-'
values by use of the following equations. These equations assume that the
dependent variable was the natural logarithm of nonoperating failure rate.
wNQ = exp(blQ1 + b2Q2 + b 3 Q3 )

•NQ1 = exp(O + 0 0O)=
+ 1
wNQ2 = exp(bl + 0 + 0) = exp(b1)
1TNQ3 = exp(O + b2 + 0).= exp(b2)
NNQ4 = exp(O + 0 + b 3 ) = exp(b3)
This example was constructed such that a level L1 nonoperating quality

factor was equal to one. Any of the other nonoperating quality levels
could have been set equal to one without changing the overall results.
The relative differences caused by changing which factor was set equal to
one would be compensated for by a change in the base failure rate.
It often happens in reliability analyses that some of the data records

used in a regression analysis are less precise than others. For example,
a data entry with 100 observed failures would be expected to be relatively
more precise than a similar data entry with only one failure. To
accommodate this difference, it is sometimes advantageous to "weight" the
data records. A suitable weighting factor is the reciprocal of the
variance. However, this information was not available for empirical
nonoperating failure rate data. As an approximate weighting factor, the
number of observed failures could be used. Several weighted regressions
were performed in this study to assess the effects of equipment power on-
off cycling.
F-Ratio and Critical F. The F-ratio and critical F are parameters

which are used in conjunction with regression analysis to determine
significance of independent variables. The critical F value corresponds
to the degrees of freedom of the model (equal to the number of data points
minus the number of bi coefficients minus one) and a specified confidence
limit. This number may be used to test the significance of each variable
as it is considered for addition to or deletion from the model. The F-
ratio value for a regression is the quotient of the mean square due to
3-6
-A7.1,CRUW N
regression and the mean square due to residual variation. If the F-ratio
value for any independent variable is greater than the critical F value,
then it was considered a significant factor influencing nonoperating
4 failure rate and was included in the regression solution.
R-Sguared Coefficient. The R2 coefficient or multiple coefficient of

determination is equal to the ratio of the sum of squares of the variance
explained by the regression to the sum of the squares of the variance of
the observed data. The R2 value is often used as a means to determine the
accuracy of a regression model. The coefficient ranges from 0 to 1.0. A
coefficient vallue of 1.0 indicates a perfect fit between the model and
observed data.
The Correlation Coefficient. The correlation coefficient is a measure

of the relation between, any two variables. It varies between -1 and 1
(from perfect negative to perfect positive correlation). A correlation
coefficient of zero indicates that two variables have no correlation.
Standard Error of Estimate. The standard error of estimate allows for

computation of a confidence interval for an individual bi regression
coefficient. The standard error is equal to the square root of the
residual mean square (the estimate of the variance about the regression).
I Upper and lower confidence limits of t.he regression coefficients can be
determined from the standard error tor a predetermined confidence (ai) by,
bi +tn..2(S.E.)
where
bi regression coefficient
tn-.2 1 -apercentage point of a t -distribution with n-2 degrees
of freedom
n =number of observations
S.E. standard error of estimate
3-7
When the assumed failure rate model form is a multiplicative model, the
upper and lower confidence limit values are not exact, but are approximate
due to the transformation. Values for the t - distribution are given in
Reference 10.
Chi-squared Confidence Intervals. The chi-squared statistic is used

to compute a confidence interval around the nonoperating failure rate
point estimate for an exponentially distributed reliability function. It
is possible to define a 90% confidence interval such that 90% of all
possible intervals (of which ours is just one) will contain the true
failure rate point est-imate. Assumptions concerning data censoring are
made to calculate the confidence interval values. These values are
calculated as follows:
Upper Confidence Limit = X(2(r + 1), a/2)
2T
Point estimate = r
Lower Confidence Limit = X2 (2r, 1 - a/2)

2T
where
r = number of observed failures

T = part hours
X (a,b) = chi-squared statistic with "a" degrees of freedom at the bth
2
percentile
1-a = confidence (100 x (1-a) is the confidence expressed as a
percentage)
3.3 Zero Failure Data
For "zero failure" data records, the standard method of dividing the
number of observed failures by the part hours results in a failure rate
value of zero. This value was considered to be intuitively
3-8
unsatisfactory. Zero observed f'tilures can be a result of a very low

inherent nonoperating part failure rate, but it can also be a result of
insufficient collected part hours. Any potential data record will exhibit
zero failures if the data collection time period is short enough. In many
instances, the nonoperating failure rates studied were extremely low, and
*there were many zeru failure data records as a result. This section
presents a discussion of zero failure data and how it was treated in this
study.
Zero failure data could not be ignored or discarded for several

reasons, including,
o For some part types, zero failure data records represented a

substantial percentage of all data.
o A data sample with only failure data would include failure rates
higher than the mean more often than corresponding lower failure
rates.
*The second item warranted further discussion. Given an infinite

sample of data, some of the calculated failure rates would be higher than
the "true" mean failure rate, and the rest would be less. However, given
a time-truncated data sample for inherently reliable components (as was
the case in this study), the failed items will more likely come from the
sub-population with failure rates above the "true" mean. Indeed, high
failure rates are characterized by observed failures. For the
corresponding low failure rates, no failures would be expected because of
the number of available part hours. This is especially true when the
nonoperating part hours fall into a certain range due to practical
constraints. Extremely high part hours are unattainable due to physical
constraints, and sources with very low part hours are avoided during data
collection. A conceptual illustration of this behavior is presented in
Figure 3.3-1. The distribution estimated from the failure data has a
A higher estimated mean failure rate than the true, unknown distribution.
Given large sample sizes, the true and the estimated means would be
expected to align themselves. Howlever, this may not necessarily be true
3-9
for small sample sizes. This was considered to be one reason that zero
failure data could not be ignored.
TRUE DISTRIBUTION FAILURE DATA
FREQUENCY
AS FAILURE RATE xt
FIGURE 3.3-1: CONCEPTUAL FAILURE RATE DISTRIBUTION
Zero failure data hypothetically could be segregated into two

categories; (1) data records indicative of a very low failure rate, and
(2) data records indicative of insufficient part hours. Practically, it
was impossible to divide the zero failure data into one category or
another. However, a largely intuitive method was developed to ascertain
which data had sufficient part hours to estimate a failure rate without
observed failures.
A preliminary analysis was performed using only failure data to

determine which zero failure data records had sufficient part hours to
estimate a failure rate. This method was used for data records on similar
part types in similar applications. If the number of part hours for each
of the zero failure data records were larger than the mean-time-to-failureR
for the data with failures, then it was assumed that the data entry had
3-10
4) Downloaded from http://www.everyspec.com
sufficient part hours. An approximate nonoperating failure rate was

determined by calculating an upper, single-sided 60% confidence limit.
This value represents an upper bound on failure rate. If an individual
zero failure data record had less part hours than the mean-time-to-
N failure, then it was assumed that there were insufficient part hours to
expect a failure, and the data record was discarded. This method was
intuitively satisfactory for two reasons. First, a zero failure data
record with more part hours than the mean-time-to-failure could be used as
an argument that the failure rate (computed from failure data) was too
high, and thus, required additional data. The second reason was that by
including the failure rates computed without failures, the estimated
failure rate mean (depicted in Figure 3.3-1) would move closer to the
true, unknown value.
It was concluded during this study that zero failure data could
* neither be arbitrarily discarded nor arbitrarily used. An intuitive
V approach was developed and subsequently implemented to decide which zero
failure data records had sufficient part hours to estimate a failure rate.
3.4 Effect of Nonoperating Failures on Operating Models
The failure rate models which appear in the current version of MIL-
HDBK-217 may, or may not, include a contribution due to nonoperating
failures. For example; if the failure rate model is based primarily upon
• life test, physics or failure data or verified operating failures, then
the model would exclude nonoperating failures. In contrast, if the

failure rate model were based primarily upon field experience utilizing
military aircraft where all failures were assumed to be operating
failures, then the model may include nonoperating failures mixed within
ý6 the operating failures.
9
3-11
V4
L
OR_ __ _ _
The difference between the MIL-HDBK-217D failure rate and the actual
operating failure rate is dependent on,
o the extent which nonoperating failures were separated from

operating failures
o the ratio of operating to nonoperating fdilure rate
o the ratio of operating to nonoperating time
If no attempt was made to eliminate nonoperating failures from the

total number of failures, then the following set of equations represent
the relationship of MIL-HDBK-217 failure rate, inherent operating failure
£ rate and nonoperating failure rate.
1 fo + fn
fo fn
fn
+fo Tn
TO oTn
=Xo + Xn
where
A2 1 7 = MIL-HDBK-217 failure rate (if nonoperating failures were not

separated)
fo = operating failures
fn = nonoperating failures
To = operating hours
Tn = nonoperating hours
X0 = operating failure rate
An = nonoperating failure rate
3-12
S"
An applicable multiplicative correction factor would then be equal to

the ratio of )o to X217, or,
correction factor = Xo/X217
Xn Tn
X217 To
To fully investigate the influence of nonoperating failures on MIL-

HDBK-217D models, the ratio of nonoperating time to operating time would
have to be determined for-all data sources used in the model development
process. The RADC technical reports describing each model development
effort were located to determine this information. Unfortunately, the
extent to which nonoperating failures were considered and the duty cycles
for each data source were generally not presented in the technical
reports. Additionally, the data sources often were not identified,
thereby preventing further investigations. As a result, correction
factors could not be determined to eliminate the effect of nonoperating
failures on the MIL-HDBK-217D models. However, based on the information
which was available, it was concluded that much of the data used to
determine the MIL-HDBK-217D models were either from life testing or from
equipments which experience significant operating times. Therefore, the
correction factor would generally be close to one.
3-13
Y,~
4.0 NONOPERATING FAILURE RATE MODELING CONCEPTS
This section presents an overview of the general nonoperating failure

rate modeling approach developed for this study. In addition, the effects
of equipment power on-off cycling, temperature, environment and screening
are discussed in detail.
4.1 Failure Rate Modeling Approach
This study involved the development of a large number of nonoperating

failure rate prediction models. A general failure rate modeling approach
was established and applied to all generic part classes. Use of a similar
modeling approach for all part classes resulted in nonoperating failure
rate prediction models which were consistent and complementary.
The basic nonoperating failure rate modeling approach is presented in

Figure 4.1-1. This approach was used for all generic part categories.
Deviation from the general approach was occasionally required because of
imbalanced data sets, inconsistent results or other reasons. Detailed
descriptions of the model development process for each part category are
included in the appropriate subsection of Section 5 of this report. The
following paragraphs provide a discussion of the techniques referred to in
Figure 4.1-1.
The initial phase of the nonoperating failure rate modeling process

was to identify potential variables. For each generic part category, of
part construction and application variables were identified to properly
characterize the subject part in a nonoperating environment. Emphasis
during the variable identification task was directed towards
identification of variables which would be accessible to the prediction
model users. These variables represent possible prediction model input
parameters. In addition, the identification of potential variables was a
prerequisite for both the data collection and theoretical model
development tasks.
4-1
MODEL
QCQ
FIGUE41-1: MOD LYS

DEVELOPMEN FLWC ART ALE
4-2
FIT' .IS
GOOD FI
FAL TRME-
Development of a theoretical nonoperating failure rate prediction

model was an integral part of the 'Overall model development process. It
was recognized early in this study that limited data resources might
prevent the identification of all significant variables using purely
statistical techniques. Unbalanced data sets or correlated variables
could result in inconsistent or incorrect conclusions. Additionally,
decisions regarding the optimal model form (i.e. multiplicative, additive,
etc.). were enhanced by a fundamentally sound theoretical model. Also, the
theoretical model proved to be useful for interpreting or explaining
observed analysis results.
An extensive data collection effort was performed concurrent with the

theoretical model development. The data collection approach and a summary
of the collected data is described in great detail in Section 2.0 of this
report.
Following the dat& collection activities, a data quality control task

was performed. All data items received during the data collection efforts
were reviewed for completeness and examined for any inherent biases. Any
data submitted which displayed obv~ious bias was not considered in
subsequent analyses. Those data recovds lacking sufficient detail were
not considered~ until the necessary additional information was acquired.
Additionally, the number of observed nonoperating failures and the
corresponding failure definition were scrutinized.
The objective of the correlation coefficient analysis was to identify

highly correlated variables. Definition of correlation coefficient and a
brief explanation is presented in Section 3.2. During the analysis,
correlation coefficients were computed for each pair of independent
variables. Regression analysis requires that all independent variables v
are uncorrelated. Therefore, the effects of correlated variables could
not be simultaneously quantified. If the variables were correlated
inherently (e.g. number of pins and number of gates for SSI/MSI devices),
then a decision was mrade to include only the most significant variable in
the regression analysis. If the variables werE correlated due to chance
4-3
(e.g. quality vs. temperature), then several options were considered. If

a valid theoretical or empirical relationship was found for one of the
correlated variables, then the effect of that variable was removed from
the data by assuming the relationship to be correct. If this assumption
was correct, then the effect of the remaining correlated variable could be
accurately assessed by data analysis.
4 The next step in the model development process was to apply stepwise
multiple regression analysis. Regression analysis is defined and briefly
described in Section 3.2 of this report. This technique was used to
compute the coefficients of an assumed model form in a least squares fit
to the data. Regression solutions were found for decreasing confidence
limits beginning with 90%. In addition, F-ratios and standard error
statistics were computed for each significant variable to obtain an
indication of the degree of significance and the accuracy of coefficient
estimates. Additionally, upper and lower 90% confidence interval values
were determined for each coefficient. In general, variables were not
* -included in the proposed model if they did not significantly effect
nonoperating failure rate with at least 70% confidence. However, if a
variable such as device screening was known to influence nonoperating
failure rate, then coefficients were computed with less than 70%
confidence and a corresponding factor proposed. In these instances, the
- resultant factor was considered approximate. This was necessary only
i ioccasionally, and no factors were proposed with less than 50% confidence.
- The goodness of fit of the regression solution was then tested using
the R-squared statistic. No absolute acceptable limit was Oefined to
* determine what constituted a "good fit" because of the relative
variability between part classes and because of different sample sizes.
For example, the acceptable R-squared value computed for hybrid
microcircuits would have been unacceptable for monolithic microcircuits.
The inherent variability of hybrid nonoperating failure rate and the large
number of potential variables (vs. a smaller number of data records)
- prevented a highly accurate model for hybrid dvices. Nevertheless, an R-
squared value was computed and the proposed model evaluated for each part
4-4
i:,
•!IN
•,- •
class. If the proposed model was determined to be unacceptable, then the

model development process returned to the variable identification task to
identify missing variables.
The next phase of the general model development process was to perform
an extreme case analysis. Predictions were made using the proposed model
for parameters beyond the ranges found in the data. The intent of the
extreme case analysis was two-fold. The first objective was to identify
any set of conditions which cause the proposed model to numerically "blow
up". The other objective of the extreme case analysis was to identify any
set of conditions which predict a nonoperating failure rate which is
intuitively incorrect. For instance, a model that predicted an unscreened
device with a lower failure rate than a similar screened device, or
predicted a negative failure rate would be examples of an intuitively
incorrect model. Reasons for failing the extreme case analysis primarily
involve an incorrect choice of model form. If the extreme case analysis
indicated that the proposed model was unacceptable, then the entire model
"development process was begun again.
The final phase of the general model development process was a model
"validation task. Data which had been withheld from the model development
process were used to evaluate the accuracy of the proposed models. Data
obtained from the European organization AFCIQ were used for this purpose
for microcircuits, diodes, transistors, resistors and capacitors. In each
*•, case, the AFCIQ data indicated that the proposed models pro ided accurate
predictions. If this had not been the case, then the model development
process would have beer started agairk with the AFCIQ data as part of the
data base. For other pf.'t types, scarcity of data required that all data
were i,:cluded in the mod6l develioment process.
Establishing i general ncnoperating failure rate model development

* process all:cd fr; .:Mfiient data analyses, and consistent proposed
models. T.-a fc•'oeI. sections include more detailed discussions of the
theoreticý.i ,•ccel ,o?•:flopment
process, and the effects on nonoperating
4 -5
;77 .'*~ J~',vs ~

failure rate of equipment power on-off cycling, temperature, application

environment and screening.
4.2 Theoretical Model Development
A series of theoretical nonoperating failure rate prediction models

was hypothesized to provide direction to the model development process and
to provide the resultant models with a theoretical foundation.
4 Information obtained through the literature search was evaluated and
reviewed to aid in development of the theoretical model. Both theoretical
and intuitive nonuperating reliability relationships located in the
literature proved to be extremely relevant in the later stages of model
development.
In general, the theoretical model development consisted of evaluation

of the hypothetical effect on nonoperating failure rate of the following
factors.
o Function
o Technology
- Fabrication Techniques
- Fabrication Process Maturity
- Failure Mode/Mechanism Experience
o Complexity
o Packaging Techniques
o Effectiveness of Process Controls
o Effectiveness of Screening and Test Techniques
o Nonoperating Environment and Temperatk're
o Frequency of Equipment Power Cycles
In addition, different model forms were investigated and an optimal model
form hypothesized. For example, assumed relationships between

nonoperating failure, rate and equipment power on-off cycling had been
referenced in many sources (Refences 5, 11 - 17). The proposed equipment
power oa,-off cycling factors were based on the general agreement between
4-6
several of these references. It was doubtful that a similar factor would

have been developed using only analytical methods.
Specifically, three major advantages were gained by the theoretical

"model development process. (1) Selection of significant variables was not
always possible by purely statistical techniques. For example,
application environment was believed to be a significant variable, but
V could not be evaluated empirically because all data were from a ground
based environment. The theoretical model complemented the data analysis
in these instances. (2) Another benefit from the theoretical model was
the selection of an appropriate model form. Many of the model forms
- presented in this study were based on conclusions made in the theoretical
model development. The form of proposed factors for temperature, quality
and equipment power on-off cycling were assumed based- on the theoretical
model, and then quantified with the available data. (3) A fir, advantage
of the theoretical model development was to identif) existing
* relationships in the literature. These relationships were often valuable
for the analysis of correlated variables.
4.3 Equipment Power Cycling Effects
Equipments in long term storage are often energized periodically in an

attempt to increase the reliability and availability of the inventory. It
has generally been assumed that the majority of observed failures were
storage related and thus, the testing was justified as a means to increase
availability. However, if the majority of failures were related to the
test procedure itself, then the testing may be unnecessary and excessively
. costly. One of the major objectives of this study was to investigate the
effects of equipment power on-off cycling on electronic and
electromechanical components. For the purpose of this study, an equipment
*• power on-off cycle was defined as the state during which an electronic
equipment goes from zero electrical activation level to the normal design
activation level and then returns to zero.
4-7
A proper investigation of nonoperating reliability must also include a

study of the effects of equipment power on-off cycling. Storage related
failures and failures induced by the power on-off cycle are the result of
quite different catisal factors. However, both types of failures can
generally only be detected when power is applied to the equipment and are
subsequently grouped together in the available data. As a result,
failures due to storage can not be distinguished from failures due to the
power on-off cycling. A two phase approach was taken to properly assess
the effects of power on-off cycling. The first phase was to investigate
previous work done in this area and to reach conclusions based on the
available information. The second phase of the power cycling
investigation consisted of treating equipment power on-off cycling
frequency as a quantitative variable in a nonlinear regression analysis.
Nonoperating failure rate data were collected for a wide range of cycling
frequencies. If cycling frequency was not identified as a significant
variable, then it could be concluded that the observed failures were
dominated by storage related failures. Conversely, if the regression
solution indicated that nonoperating failure rate (in units of failures
per 106 nonoperating hours) was approximately proportional to cycling
frequency, then it could be concluded that the observed failures were
primarily induced by the power cycling.
4.3.1 Dormancy and Power On-Off Cycling Effects on Electronic

Equipment and Part Reliability (RADC-TR-73-248)
The issue of equipment power on-off cycling was addressed in RADC-TR-

73-248, "Dormancy and Power On-off Cycling Effects on Electronic Equipment
and Part Reliability" (Reference 5) performed by Martin Marietta
Aerospace. Included in this study was a mathematical analysis of failure
rate versus equipment power cycling. Equation 1.1.2-9 in Reference 5
presents a mathematical relationship for service life failure rate for
systems which undergo long term storage and dormancy. The service life
failure rate was presented as a function of dormant failure rate and
average equipment power zy:ling rate. The mathematical expression was
4-8
developed based on intuitive reliability relationships. This expression

is given by the following equation.
XSL = (1 + KC/D(NC)) XD
where
XSL = service life failure rate

KC/D = ratio of cyclic failure rate to dormancy failure rate (in
hours of dormancy per cycle)
NC = average cycling rate expected during the service life of an
electronic system (in cycles per total unit time of service
life)
XD =.dormant failure rate
This equation was then extended to component nonoperating failure rates

and estimates of the KC/D term were made. It was noted that the KC/D term
ideally should vary with part type, part quality, cyclic rate, temperature
effects, transient suppression protection and environmental application.
However, the KC/D estimates were only made based on part types because of
practical restraints.
4.3.2 Planning Research Corporation Studies
The effects of equipment power on-off cycling have been investigated

by Planning Research Corporation (PRC) during several studies performed
for the Navy Space Systems Activity, and Goddard Space Flight Center
(References 11, 12, 13 and 14). Equipment and component level observed
"data were collected and compared for continuous operation, standby
operation, intermittent operation, cyclical operation, dormancy and
storage. The subject equipments were spacecraft electronic systems. No
significant difference was identified during the studies between cyclical
operation and steady state operation. This was due primarily to data
limitations rather than a definitive conclusion concerning cyclical vs.
steady state failure rate. Nevertheless, it was stated in Reference 13,
4- 9
"It is rather clear, however, that cycled components in general do not

have order of magnitude worse failure rates than their non-cycled
counterpart!."
4.3.3 ARINC Study
Cited in the PRC reports was a ARINC study (Reference 15) concerned
with the effect of equipment power cycling on shipboard electronic
equipment. A summary of the conclusions from this study are as follows.
o The expected number of equipment malfunctions per hour of

operation, bt, is given by,
bt = bo(l + 8N)
where
bo = expected number of malfunctions per hour of continuous
operation
N = the number of cycles per hour of continuous operation
o More severe environments would be expected to increase the cycling

factor and more benign environments to reduce it.
o The study revealed a substantially higher level of equipment
malfunctions during a short period of time after each power turn-
on than during any other operating time interval. It could not be
determined which of these failures occurred in dormancy and which
were due to turn-on transients.
o On/off cycling was not observed to introduce any specific type of
failure.
The cycling equation constant of 8 was empirically determined from a

broad class of shipboard electronic equipme ts, primarily of the vacuum -
tube variety. Similarities in form were noted between this equation and
the previously presented mathematical expression for service life failure
rate from the Martin Marietta report. In both instances, the equipment
power cycling frequency was multiplied by a constant term and then added
to one.
4-10
I rdAS"-
4.3.4 Hughes Presentations
Missile storage test data compiled by Hughes Aircraft Co. (Reference

16 and 17) indicates that missile testing is the dominant variable
affecting the performance of missile electronics in storage. Equipment
level test data were collected for the Maverick and TOW missiles. The
data indicates that a similar percentage of missiles fail regardless of
the storage interval. It was concluded by Hughes that this observation
was because the testing process had induced a large majority of the
observed failures. Extended to the component level, this observation
would result in nonoperating failure rate prediction models (measured in
units of failures per 106 nonoperating hours) directly proportional to the
testing frequency. The data appears to strongly support the conclusion,
although no trend analysis or statistical tests were described in the
references to help validate the conclusion.
It should also be noted that failures caused by operator or

maintenance error would tend to be concentrated during the test interval
as opposed to being evenly distributed throughout the storage period.
Therefore, if the data were not accurately recorded, screened and
characterized, then other factors could be partially responsible for the
relatively constant percentage of "failed" missiles observed by Hughes.
4.3.5 Equipment Power Cycling Conclusions
Two major conclusions were made based on the literature review. The
first conclusion was that equipment power cycling must be considered as
part of any effort to investigate or predict nonoperating failure rates.
'~ 'The !'acond conclusion was that the effects of equipment power on-off
cycling can be predicted by a multiplicative model of the following form.
Trcyc 1 + Ki(Nc)
4-11
where
wcyc = equipment power on-off cycling factor

K1 = constant
Nc = number of power on-off cycles per 103 nonoperating hours
The form of this cycling factor was based on the agreement between the
Martin Marietta and ARINC results. In addition, a factor of this form is
intuitively appealing. At extremely low cycling rates, the predicted
nonoperating failure rate would become independent of cycling rate and
equal to the dormant/storage failure rate. At high cycling rates, a
predicted nonoperating failure rate would be proportional to the cycling
rate.
A cycling rate factor of this form would not be applicable when the
power cycles interfere with one another (i.e. the equipment has not yet
cooled down from the previous cycle when a new one is initiated).
However, this restriction was not believed to limit the utility of the
results of this study.
The units for equipment power cycling rate were chosern to be number of
equipment power on-off cycles per 103 nonoperating hours. This decision
was made for convenience. Storage intervals as long as ten yeari (.011
cycles/lO3 hrs.) and as short as one day (41.67 cycles/10 3 hrs.) could be
expressed easily.
4.3.5 Equipment PowerCycling Analysis
The second phase of the equipment power cycling analysis was to

quantify the effects of power cycli% by application of appropriate
statistical techniques. As a prerequisite tn data analysis, it was
required that the cycling rate must he identified for each data source
described in Section 2.3. This tasi, proved to be more difficult than
anticipated. For some cases, the test schedule information was
proprietary. In other instances, týe testing frequency was unknown or
4-12
variable. average cycling rates were obtained or calculated

Nevertheless,
for the majority of data sources. For data sources where the testing
frequency was variable (e.g. Hawk, Maverick and TOW missiles) an average
cycling rate was computed for the data collection interval.
A multiplicative model with the assumeJ equipmeint power cycling factor

was nonlinear. Thus, linear regression analysis could not be directly
applied. An iterativs approach was taken to solve the nonlinear equation.
Specifically, the iterative process began by defining three distinct
cycling rate categories. Then, the effects of all significant variables
were quantiFied including computation of unique regression coefficients
for each of the three cycling rate categories. In effect, equipment power
cycling was initially treated as a qualitative variable. Conceptually,
this first step can be represented by the following equations.
The first equation depicts an example nonlinear nonoperating failure

rate prediction model. The model is nonlinear due to the effect of power
cycling, and is also a function of device style, quality and temperature.
p= XnbVNQ7rNT(1 + KI(Nc))
where
Xp = part nonoperating failure rate

Xnb = base failure rate, based on device style
StNQ nonoperating quality factor
wNT = temperature factor
K1 = cycling factor constant
Nc = number of equipment power on-off cycles per 103 nonoperating
hours
Three distinct cycling rate categories were defined to enable the

r computation of unique regression coefficients for each category. The
resultant model then becomes linear (with a logarithm transformation) and
4-13
,..
regression analysis can be, applied to quantify each variable. The model
then takes the following form.
= nb 7YIQ '7TAi
where
Aj = multiplicative cycling rate constant
Al = constant for N1 2>Nc >N2, cycling rate category oneI

A2 = constant for N2 >Nc >N3 , cycling rate category two
A3 = constant for N3 .2,Nc > N4, cycling rate category three
The preliminary cycling rate factor could then be computed by

performing a two-dimensional regression (Ai vs. average cycling rate) to
obtain estimates for the cycling factor constant (K1). The following
equation depicts this relationship.
Ai = K2(1 + lR)
where Ri is the mean cycling rate for each cycling rate category, and K2
is a normalization constant.
The iterative process was required unless Ai correlated exactly with

Ni (correlation coefficient = 1). Practically, it is doubtful that a
correlation coefficient of exactly one will be observed, and the iterative
process was required in all cases for this study.
As the iterative process continues, the preliminary equipment power

cycling factor would then be assumed exact, and the coefficients for all
6other variables recalculated. Generally, the observed change in
coefficient values was small. Then, those coefficients would be assumed
to be exact and the eq~uipment cycling factor recalculated. The iterative
process continued uintil observed changes In coefficient estimates were
negligible. In practice, it was found that less than three iterations
were generally required. For example, the coefficienis stabilized after
4-14
Y'
one iteration for resistors and three iterations for capacitors. However,I
six iterations were required for linear/interface microcircuits.
K1 constant values were computed for each generic part category where
it was determined that equipment power on-off cycling was a significant
variable. Ideally, unique K1 values should be determined for every
conceivable combination of part class, quality, temperature and
II
-
environment. However, an extremely large amount of nonoperating failure

rate data would have been needed to quantify an equipment power cycling
factor as a function of each of those variables. Since the available data
was limited, the equipment power cycling factors determined inthis study
represent the best possible values. When additional nonoperating
reliability data becomes available, these factors should be investigated
to determine their validity. The equipment power cycling factors for
specific part classes are presented in the respective model development
se ction of this report.
Whether or not equipment power cycling significantly increases the

reliability and availability of an inventory of stored equipments, it must
be recognized that the testing process accelerates the frequency of
electronic component failure. Conversely, the opposite can be assumed toj
be true for certain electromechanical components. Long periods of storage
for switches, relays, electric motors, servomechanisms and connectors can
result in severe degradation. Effects of the degradation process can
often be minimized by periodic actuation of contact devices or periodic
turn on for motors and servomrechanisms. Similarly, certain styles of
capacitors are known to degrade in storage and require reforming of the
dielectric at certain time intervals.
Switches and relays designed to control low voltage digital circuits

are particularly susceptible to storage degradation. A resistive surface
film will form on normally open contacts depending on package type,
contact material and environment. Periodic actuation of the contacts can

break thru the resistive surface film to provide electrical contact. If
4-15
becomes relatively thick and thereby, contact.

results in poor
Unfortunately there were insufficient data to quantify the effects of
equipment power on-off cycling on contact devices.
For motors and servomechanisms, periodic turn-on

maintain the integrity of the lubrication. As was the case with switches
and relays, there were insufficient data to accurately ascertain the
is essential to
I
effects of equipment power on-off cycling for these rotating mechanisms.
It is recommended that future studies addressing the issue of equipment z
power cycling investigate the effects of cycling on these mechanical and
electromechanical part types.
4.4 Temperature Effects
An investigation into the effects of temperature on nonoperating

failure rate was a crucial part of this study effort. Electronic
assemblies in a ground storage environment can be exposed to a relatively
wide range of temperatures depending on the applicable mission profile,
geographic location, and availability of environment controlled storage
facilities. To investigate temperature effects, high temperature storage
life test data were collected for microcircuits and discrete
semiconductors. Data were available for test temperatures ranging up to
3500C. As a result, unique nunoperating temperature factors were
determined for microcircuits, transistors and diodes.
The impact of storage temperature was believed to be one of the most

significant variables effecting microcircuit no,'operating failure rate.
Most microcircuit failure mechanisms involve one or more physical or 41
chemical processes which occur at a rate which is dependent on

temperature. It was assumed that the Arrhenius model applies to the
reaction rate of microcircuit storage failure mechanisms. The Arrhenius
model was based on empirical data and predicts that the rate of a given
physical or chemical reaction will be exponential with the inverse of
temperature. Conceptually, the Arrhenius model is given by the following
equation.
4-16
SN
Reaction Rate exp(-cea/KT)
where
Lea = activation energy (eV)

K = Boltzman's constant
= 8.63 x 10-5 (eV/OK)
T = temperature (OK)
Every chemical and physical reaction has a unique activation energy

associated with it. During the storage life of an electronic component
there are several such reactions proceeding simultaneously, each capable
of causing a part failure. Individual consideration of each of these
different reactions would result in very complex nonoperating failure rate
prediction models which are not in accordance with the simple form of the
Arrhenius model. Consideration of each physical and chemical failure
mechanism separately (and assuming each viechanism is independent) would
result in a nonoperating failure rate predictior model similar to,
n
where i•1(-eai/KT)
S~where
P i 1
i Xp = nonoperating failure rate -
n = number of failure mechanisms

Eeai
I_.
= activation energy of the ith failure mechanism
This relationship was determined to be much too complex t' be

quantified with the available data. Therefore, alternate mathematical
expressions for device nonoperating failure rate vs. temperatut.ve were
explored. The activation energy Arrhenius relationship concept has been
applied to microcircuit failure rates (instead of failure mechanism
reaction rates) often enough (References 18 and 19) to warrant further
investigation. It has been found for general classes of components with
similar failure mechanism distributions, the cumulative effects of the
4-17
iz
various reactions
can be approximated by an Arrhenius model for a
specified temperature range. This relationship was designated as the
"equivalent Arrhenius relationship." Because of the documented accuracy
of this approximation, it was decided to investigate the effects of
temperature using the equivalent Arrhenius relationship. It should be
emphasized that at extreme high and low temperatures, this relationship
will no longer be applicable. Use of the Arrhenius model to predict
failure rate can be a very useful and accurate tool. However, the
limitations of this assumption must be fully understood.
The storage failure rate of linear microcircuits has been observed to

approximate the equivalent Arrhenius model over a wide range of storage
temperatures. Figure 4.4-1, taken from Reference 20, presents the
logarithm of observed storage failure rates for linear devices as a
function of the inverse of temperature (which is equivalent to the
Arrhenius relationship). The information depicted in Figure 4.4-1 is for
a range of temperatures of 150 0C to 3500C. The relationship appears
linear at least until 3000C. In addition, the microcircuit nonoperating
data collected in support of this study included application temperatures
from 180C for field data to 350 0C for high temperature storage life teAt
data.
A preliminary multipl':ative temperature factor for microcircuits was

determined to be the following equation.
Tt,p = exp(-An(I/T))
where
wtp= preliminary temperature factor

T temperature (OK)
An =constant (equivalent activation energy)/(Boltzman's constant)
It was decided to add a reference temperature term to improve the

* utility of the proposed nonoperating microcircuit models. A reference
4-18
_......... .
F-6. few
D10
.4
17
!!!i .....
7VS..
iU'
303
RT VS TEMPERATUR
4-1
0C (Linear I/KO Scale)
:: ?,.3,Teiiperature
,,, ."FIGURE 4.4-I: LINEAR MICROCIRCUIT .. TORAGE FAILURE

i "- ",•;'RATE VS. TEMPERATURE
•. •-,,;4-19
temperature of 298 0 K (i.e. 250C) was selected to be consistent with MIL-

HDBK-2170. The preliminary temperature factor with a reference
temperature term is therefore, given by,
wt = exp(-An(4. T-r
where
'Nt,p = temperature factor

An = constant
Tr = reference temperature (OK) = 2980 K
Mathematically, adding the reference temperature term to the proposed

model has no effect on the resultant nonoperating failure rate prediction.
Relative differences caused by selection of the reference temperature are
compensated by corresponding changes in the proposed base failure rate.
The following series of mathematical expressions depict this relationship.
"1 =X2exp(-An())
=X 2 exp(-An(T - + Tr))
=X 2 exp(-An(Tr))exp(-An(T -r))
4 =A 3 exp(-An( - ))
where
X 1 = nonoperating failure rate

A 2 = preliminary base failure rate
A 3 = base failure rate =X 2 exp(-An( 1 ))
4-20
* .. -- *- .
Initially, introduction of the reference temperature term seemed to

needlessly complicate the proposed model. However, it was decided to
include the reference temperature for two reasons:
1) A proposed model with the reference temperature term provides more

information. The base failure rate would be equal to the device
failure rate at the reference temperature. Thus, inspection of
the base failure rate value provides meaningful information for
quick analyses. Without the reference temperature term, the base
failure rate would correspond to when (l/T) approaches zero, and
would thus be meaningless by itself.
2) A proposed model with the reference temperature term would

minimize the need for exponential numbers (e.g. 7 x 1034) and
would therefore result in models which are easier to use. The
temperature factor would be equal to one when the ambient
temperature equals the reference temperature, and would generally
be below 100 for even the highest possible ambient temperatures
found in nonoperating applications.
A modification was then made to the preliminary temperature factor so

the predicted nonoperating failure rates would approach constant values at
low temperatures and become asymptotic to the operating failure rate at
high temperatures. The need for this modification was identified during
the microcircuit model validation stage. The proposed microcircuit
nonoperating temperature factor was, therefore, given by the following
expression.
St K3 + K4 exp(-An( 1
where
K3 , K4 constants
It was concluded after evaluating all the available information that

microcircuit nonoperating failure rates could be predicted by use of this
temperature relationship for the range of nonoperating temperatures found
during normal usage (-55 0 C to 1250C, maximum). Unique Arrhenius
4-21
relationship constants were determined as a function of logic type for

digital, linear and memory devices.
High temperature storage life test data was also collected for
discrete semiconductors with temperatures ranging from 1500C to 200 0 C. It
could not be verified whether the equivalent Arrhenius relationship was
linear for the total range of temperature values found in the data; 180C
to 2000C. Therefore, an additional term was added to the general form of
the temperature factor expression to allow for a nonlinear relationship
for the logarithm of failure rate versus the inverse of temperature. The
multiplicative temperature factor expression for discrete semiconductors
was assumed to be the described by following equation.
ITr exp(-An(~1 1. + (T)P)
" ~where
Tm, P = shaping parameters
If the equivalent Arrhenius relationship was determined to be applicable

for the entire range of temperature values, then the second term would
become negligible.
The derivation of appropriate temperature factors for microcircuits

and discrete semiconductors was an essential part of the methodology
developed to assess the effects of nonoperating periods on equipment
reliability. Appropriate temperature factors were determined empirically
and are included in the respective nonoperating failure rate prediction
models. Temperature factors for less temperature dependent part types
could not be developed due to data limitations.
4.5 Environmental Factor Analysis
The primary objective of this study was to develop a nonoperating

failure rate prediction methodology which can be applied to any
4-22
conceivable mission profile with nonoperating periods. Therefore, it was

essential that any proposed prediction models account for the apparent
differences in nonoperating failure rate caused by environmental stresses
such as temperature cycling, vibration, humidity, mechanical shock,
electromagnetic interference and many other stresses. There is a wide
variety of environmental stresses that electronic equipments can be
exposed to while in th• nonoperating condition. For example, at one
extreme, missile guidance systems in a captive-carry application are
exposed to the extreme vibration and temperature cycling stresses
associated with an airborne uninhabited environment. Conversely, missile
electronics stored in a sealed container are subjected to significantly
lower levels of environmental stress. In the development of theoretical
nonoperating failure rate prediction models, it was determined that
environment would indeed have a significant effect on nonoperating lure
rate for each major part category (i.e., microcircuits, f1screte
semiconductors, resistors, etc.)
The early stages of this study effort included plans to derive

nonoperating failure rate prediction models as a function of specific
environmental stress values (i.e. relative humidity, g-force, etc.). A
model of this form would provide maximum prediction accuracy and proper
discrimination against known failure mechanism accelerating factors.
However, after careful consideration, this approach was rejected for the
following reasons.
o 'The nonoperating reliability data available for analysis for this

study generally did not include specific values for all
environmental stresses. In fact, the ambient temperature was
usually the only environment variable which could be precisely
determined. Therefore, assumptions would have to be made to
assess the quantitative effect on device nonoperating failure
rate. This would clearly diminish the significance of analysis
results.
o One of the more important objectives of this study was to develop

failure rate prediction models which are relatively easy to use.
In the design phase of an equipment, specific values for all
environment related variables may not be available. Therefore,
the benefits afforded by including specific environmental stress
'4-23
N
--
variables would be negated by a decreased ability to use the

models.
o It would be difficult for the government costumer to verify or

document the environmental stress input parameters. This would
introduce additional uncertainties regarding equipment reliability
predictions. /
As an alternative to the analysis of specific environmental stresses,

the approach was taken to consider well defined environment categories
which experience similar levels of environmental stress. This is
consistent with the methods used in MIL-HDBK-2170. Benefits offered by
this approach are that the models would be easy to use while still
discriminating against variables known to increase failure rate. The only
variable analyzed apart from the generic environment categories was
ambient temperature. This was for three reasons. The first, ambient
temperature is the most significant of individual environmental stresses
for many part types. The second reason was that the ambient temperature
for electronic assemblies in a ground storage environment can vary
substantially depending on the geographic location and the availability of
environment controlled storage facilities. The last reason was that
ambient temperature was generally available from the data sources used for
this study. Therefore, statistical techniques including regression
"analysis were used to determine the effect of failure rate versus ambient
temperature for microcircuits, transistors and diodes.
The number of application environments included in MIL-HDBK-217D,

Notice 1 has been expanded to 26 as a result of the work presented in
References 4 and 21. The environment categories are presented in Table
4.5-1 with a brief description and the appropriate abbreviation.
Electronic equipments can hypothetically experience nonoperating periods
in any of these environments. Therefore, these same environment
categories, with the exception of the space flight environment, were
considered for nonoperating failure rate prediction purposes. In addition
to these categories it was anticipated that reliability differences
between storage, dormant and captive-carry missile applications would be
considered as part of the environmental factor.
4-24
TABLE 4.5-1: ENVIRONMENT CATEGORIES
ENVIRONMENT SYMBOL DESCRIPTION
Ground, Benign GB Nonmobile, laboratory environment readily

accessible to maintenance; includes
laboratory instruments and test equipment,
medical electronic equipment, business and
scientific computer complexes.
Ground, Fixed GF Conditions less than ideal such as
installation in permanent racks with adequate
cooling air and possible installation in
unheated buildings; includes permanent
installation of air traffic control, radar
and communications facilities.
Ground, Mobile GM Equipment installed on wheeled or tracked

vehicles; includes tactical missile ground
support equipment, mobile communication
equipment, and tactical fire direction
systems.
Space, Flight SF Earth orbital. Approaches benign ground
conditions. Vehicle neither under powered
flight nor in atmospheric reentry; includes
satellites and shuttles.
Manpack Mp Portable electronic equipment being manually
transported while in operation; includes
portable field communications equipment and
laser designations and range finders.
Naval, Sheltered NS Sheltered or below deck conditions, protected

from weather; includes surface ships
communication, cumputer, and sonar equipment.
Naval, Unsheltered NU Nonprotected surface shipborne equipment
exposed to weather conditions; includes most
mounted equipments and missile/projectile
fire control equipment.
Naval, Undersea, NUU Equipment immersed in salt water; includes

Unsheltered sonar sensors and special purp se anti-
submarine warfare equipment.
Naval, Submarine NSB Equipment installed in submarines; includes
navigation and launch control systems.
>3 4-25
WI
TABLE 4.5-1: ENVIRONMENT CATEGORIES (CONT'D)
Naval, Hydrofoil NH Equipment installed in a hydrofoil vessel.

Airborne, AIC Typical conditions in cargo compartments
Inhabited, Cargo occupied by aircrew without environment
extremes of pressure, temperature, shock and
vibration and installed on long mission cargo
V aircraft.
Airborne, AIT Same as AIC but installed on high performance
Inhabited, Trainer aircraft such as trainer aircraft.
Airborne, AIB Typical conditions in bomber compartments
Inhabited, Bomber occupied by aircrew without' environment
extremes of pressure, temperature, shock and
vibration and installed on long mission
bomber aircraft.
Airborne, AIA Same as AIC but installed on high performance
Inhabited, Attack aircraft such as used for ground support.
I.Airborne, AIF Same as AIC but installed on high performance
Inhabited, Fighter aircraft such as fighters and intercepters.
Airborne, AUC Bomb bay, equipment bay, tail or where ex-
Uninhabited, Cargo treme pressure, vibration and temperature
cycling may be aggravated by contamination
from oil, hydraulic fluid and engine exhaust.
Installed on long mission transport aircraft.
Airborne, AIJT Same as AUC but installed on high performance
Uninhabited, aircraft such as used for trainer aircraft.
Trainer
Airborne, AUB Bomb bay, equipment bay, tail or where ex-
Uninhabited, treme pressure, vibration and temperature
Bomber cycling may be aggravated by contamination
from oil, hydraulic fluid and engine exhaust.
Installed on long mission bomber aircraft.
Airborne, AUA Same as AUC. but installed on high performance
Uninhabited Attack aircraft such as used for ground support.
Airborne, AUF Same as AUC but installed on high performance
Uninhabited Fighter aircraft such as fighters and intercepters.
4-26
TABLE 4.5-1: ENVIRONMENT CATEGORIES (CONT'D)
Airborne, Rotary ARW Equipment installed on helicopters, includes

Winged laser designators and fire control systems.
Missile, Launch f Severe conditions related to missile launch
(air and ground), and space vehicle boost
into orbit, vehicle re-entry and landing by
parachute. Conditions may also apply to
rocket propulsion powered flight.
Cannon, Launch CL Extremely severe conditions related to cannon
launching of 155 mm and 5 inch guided
projectiles. Conditions apply from launch to
target impact.
Undersea, Launch UsL Conditions related to undersea torpedo
mission and missile launch.
Missile, Free MFF Missiles in non-powered free flight.
Flight
Airbreathing MFA Conditions related to powered flight of air
Missile, Flight breathing missile; includes cruise missiles.
4-27
The optimal approach to determine nonoperating environmental factors

would have been to analyze observed nonoperating field data for all ?6
environment categories presented in Table 4.5-1. However, this method was
not feasible due to severe data restrictions. In fact, all collected data
were for ground based environments despite considerable efforts expended
to collect data for other environments. There were several major reasons
why nonoperating data were not available for the more stressfu)
environments. Several of these reasons are presented in the following
paragraphs.
One of the primary reasons that nonoperating data were not available
for airborne environments was that aircraft mission profiles include a
composite of environments. Many airborne equipments (i.e. heads-up-
display, radar) are powered up prior to take-off and remain on throughout
the flight, and therefore do not experience any nonoperating airborne

stresses. Data of this nature was available, and subsequently summarized
for the F-16 heads-up-display. However, the data is not representative of
an airborne nonoperating environment. Equipments which are nonoperating
while airborne, such as laser target designators, captive-carry missiles
or countermeasures sets, experience a variety of nonoperating environments
and corresponding stresses. The equipments are exposed to a nonoperating
ground fixed environnent between flights, a nonoperating ground mobile
environment during ground taxi time, and a nonoperating airborne
environment during flights. Unfortunately, nonoperating failures can only
be detected when the equipment is powered up.
ground mobile or airborne.

Therefore, observed
nonoperating failures could not be accurately designated as ground fixed,
Additional problems, including depot data
I
recording practices and airborne "when discovered" code definitions, also
prevented summarization of nonoperating data for composite environment
applications.
4-28
* .. ..- - ____ -. ,_______,_.__-__,•,.+ . • - - +. % .- '• • • • ' ,-•

Another reason that data was unavailable for stressful nonoperating
environments was the common misconception in the electronics industry that

all failures are attributable to operating stresses. With the exception
of missile storage applications, no attempt generally has been made to
identify nonoperating failures. Therefore, no centralized database has
been established which includes nonoperating reliability for airborne
environments. In fact, the Missile Materiel Storage Reliability Program

managed by U.S. Army MICOM, Redstone Arsenal, AL represents the only
centralized reliability database which includes nonoperating data,
Another major reason that nonoperating data for airborne and other
non-ground environments were not general lack of
available was the
accurate piece part data for any operating or nonoperating application.
The additional requirement that the data include only primary failures,
correspond to a nonoperating state, and correspond to a non-ground
environment further compounds an. already difficult problem. Data from .
large automated data bases like the U.S. Air Force D056, U.S. Navy 3M and
U.S. Army Sample Data Collection Program could not meet the requirements
of this study. Specifically, the D056 and 3M data bases are incomplete at
the piece part level (i.e.. not all depot level repair actions are reported
into the system). In addition, the task of separating primary (or
inherent) part failures from secondary (or induced) failures is very
difficult. These observations are documented for the U.S. Air Force data
collection system in Reference 22. The Sample Data Collection program is
operated by the Cobro Corporation and managed by U.S. Army TSARCOM, St.
Louis, MO. This program offered several distinct advantages in regard to
- separating operating from nonoperating failures, and primary from
secondary failures. However, the Sample Data Collection program was also
inadequate for tracking failures to the piece part level.
The lack of traceability between on-equipment maintenance actions and_

•- ~depot part replacements was another reason that collection of nonoperating
data for airborne environments was unsuccessful. On-equipment maintenance
actions can often be designated as operating or nonoperating failures by
IL use of a "when discovered" code. It can be assumed that failures observed
4-29
,, .- ,. ,
when the equipment was energized are due to the effects of rionoperating
period or'the result of the power on-off cycle. However, this information
is of no use unless depot level part replacements can be accurately traced
back to the on-equipment action. Unfortunately, traceability is usually
poor due to time and location differences between on-equipment and depot
maintenance, and incomplete data recording practices.
An alternate nonoperating environmental factor development method was

determined to compensate for the lack of available data. The alternate
method was based on the hypothesis that a series of nonoperating
environmental factors could be developed from the documented operating
environmental factors. Two recent studies (References 4 and 21) evaluated
and proposed operating environmental factors for every environment
category presented in Table 4.5-1. It was assumed that these numerical
values properly characterize the effects of environmental stress for the
operating applications. It was also assumed that nonoperating
environmental factors could be generated from the operating environmental
factors by comparing differences in failure mechanism accelerating factors
and average application temperatures.
/
Specifically, the nonoperating environmental factor development
process took place in four phases. The first. phase consisted of an
analysis of the operating environmental factor in the context of the MIL-
HDBK-217D failure rate prediction mndel. Several factors, including the
presence of a separate temperature factor, influence the definition of the
environmental factor. The second phase of the nonoperating environmental
factor development process was to compare typical temperature differences
between the operating and nonoperating state. If both operating and
proposed nonoperating models do not include separate temperature factors, i
then an adjustment was possibly required to the operating environmental
factors. The third phase of this method was to compare the differences
between operating and nonoperating failure mechanism accelerating factors
(i.e. environmental or operational stresses which accelerate the
occurrence of failure mechanisms). Decisions were made regarding whether
the magnitude of nonoperating environmental factors would be greater than,
4-30
less than, or approximately the same as the MIL-HDBK-2170 operating

environmental factors. The final phase of the environmental factor
development process was to determine numerically the magnitude of
environmental factor differences, if any. Theoretical methods and failure
simulations based on assumed distributioais were utilized to aid in the
process. The following paragraphs present a general discussion of the
environmental factor determination method. Results for specific part
types are included in the respective model development sections of this
report.
The relationship of the operating environmental factor to the MIL-

HDBK-217D failure rate prediction model was identified in the first phase
of the nonoperating environmental factor development process. It was
found that the MIL-HDBK-217D failure rate prediction models incorporated
the environmental factor four separate ways, which are,
1) multiplicative environmental factor without a separate temperature

dependent factor.
2) multiplicative environmental factor with a separate temperature
dependent factor.
3) environmental factor part of the non-linear microcircuit model.I
4) no environmental factor.
Table 4.5-2 presents the distribution of part types in MIL-HDBK-217D

into these categories. The distinction between multiplicative
I
environmental factors without a separate temperature dependent factor
* (category 1) and factors with a separate temperature dependent factor
(category 2) was very important. If a model does not inriude a separate
temperature dependent factor, then the environmental fact-or accounts for
gall environmental stresses including temperature. If the model does
include a separate temperature dependent factor, then the environmental
factor accounts for all environmental stresses except changes, in
temperature. The temperature factor (or base failure rate as a function
of temperature) predicts the effect of changes in temperature due to both
ambient temperature and internal heat generation. In. effect, the
I
4-31
environmental factor alone, for category 2 models, predicts the failure

rate difference due to environmental stress given that the temperature is
constant.
TABLE 4.5-2: MIL-HDBK-217D ENVIRONMENTAL FACTOR CLASSIFICATION
Category 1 (1) Category 2 (2) Category 3 (3) Category 4 (4)
Tubes Transistors Microcircuits Motors

Lasers Diodes Hybrids Misc. Parts
Thermistors Opto-electronics Bubble Memories
Switches Resistors
P.W. Assemblies Capacitors
Connection Transformers
Crystals (5) Coils
Fuses (5) Synchros
Inc. Lamps (5) Resolvers
Meters (5) Relays
Circuit Breakers (5) Connectors
Notes: 1) Type 1 environmental factors predict the effects of all

environmental stresses.
2) Type 2 environmental factors predict the effects of all
environmental stresses except temperature
3) Type 3 environmental factors are part of a non-linear
microcircuit model.
4) Type 4 part type models have no environmental factors.
5) An updated model for, these part types is presented in

Reference 43.
The second phase of the nonoperating environmental factor development

process was to determine numerical adjustment factors to account for
differences in operating and nonoperating temperatures. If both operating
and proposed nonoperating failure rate prediction models include separate
temperature and environmental factors, then no temperature adjustment was
required. Failure rate differences due to internal heat generation are
predicted by the respective temperature factors, of the operating and
proposed nonoperating models. This was the case for transistors, diodes
and microcircuits.
4-32
If neither the operating or proposed nonoperating model included a

temperature factor, then (as a minimum) the operating factors would have
to be adjusted to account for the relatively lower temperatures found in
the nonoperating condition.. The difference in operating and nonoperating
temperature is greater for the more stressful environments. There are
several reasons for this observation including better ventilation and
environmental control available for ground based environments.
Determination of the anticipated temperature difference between operating
and nonoperating environments was difficult. The range of possible
missions, applications and locations for electronic equipments is so great
.J that determination of an average temperature difference based on a sample
of mission profiles was not deemed feasible. In addition, the available
literature concerning ambient temperatures for different application
environments was inconclusive at best. However, many technical documents
were reviewed including the documented temperature relationships in MIL-
HDBK-217D. Table 5.2-17 "Ambient Temperature for all Parts" and Table
5.1.2.5-4 Note 2 "Microcircuit Case Temperatures" from MIL-HDBK-217D,
Notice 1 were utilized to estimate the relative difference between
operating and nonoperating applications. Those tables indicated that the
temperature difference between operating and nonoperating is approximately
0 C more for airborne
* i20 uninhabited environments than for the ground benign
environment. This value was used to compute environmental factor
temperature adjustment factors for tubes, lasers, thermistors, switches,
printed wiring assemblies, connections, crystals, fuses, incandescent
lamps, meters and circuit breakers.
A' third case was identified where the operating failure rate
prediction model included a separate temperature factor and the proposed
nonoperating failure rate prediction model did not. No nonoperating
* temperature factors were determined in this case because of data
deficiencies. For these part types, the increase in temperature due to
the applied power is accounted for by the operating temperature dependent
factor and not the environmental factor. Therefore no large temperature
adjustment was required as in the previous case. A study of environmental
profiles (Reference 4 and 24) revealed that temperature 1'n,7 and
4-33
temperature extremes varied radically

for more stressful nonoperating
environments such as airborne uninhabited fighter. However, the typical
change in average temperature was much less for the nonoperating
"" environment than the operating environment. It was determined that
appropriate temperature adjustment factors would be small in magnitude.
'The relatively smaller temperature adjustment for these part types would
be less than the inherent variability involved with failure rate
-• estimation, and thus were not warranted. Therefore, no temperature
adjustment was made for opto-electronic semiconductors, resistors,
- capacitors, transformers, coils, synchros, resolvers, relays and
connectors.
The third phase of the environmental factor analysis task was to

. compare operating and nonoperating failure mechanism accelerating factors.
Decisions were made as to whether the magnitude of proposed nonoperating
. environmental factors would be greater than, less than, or approximately
the same as operating environmental factors. A literature search was
. performed to identify documented sources of operating and nonoperating
I failure mechanism distributions. In addition, the microcircuit failure
4 mode/mechanism data base at the RAC was available for analysis. After
completion of the literature search and preliminary analyses, it was
concluded that there was severe absence of substantial, documented failure
mechanism distribution information. Much of the available information was
incomplete and/or conflicting. This deficiency was particularly acute for
nonoperating applications. However, the available information provided a
good boundary for the failure mechanism distributions and average failure
mechanism and failure mechanism accelerating factor distributions were
estimated for each major part class after evaluating all available
information. These failure mechanism distributions are presented in the
appropriate subsections of Section 5.0 of this technical report.
Each failure mechanism for the specific part types was categorized as
either (1) primarily accelerated by environmental stresses, (2) primarily,
accelerated by operational stresses, (e.g., applied voltage or current,
mechanical actuations, etc.) or (3) accelerated by simultaneous exposure
4-34
to environmental and operational stresses. Conceptually, the implications

of these three types of failure mechanisms are presented in Figures 4.5-1,
4.5-2 and 4.5-3. Each figure presents a hypothetical distribution of
failure mechanism accelerating factors for ground and airborne
environments, and for operating and nonoperating applications. The
failure rate due only to operational stress (e.g. electrical current) is
shown to be constant with regard to environment. The failure rates due to
environmental stress and combined environmental/operational stress are
shown to increase with the amount of environmental stress. Additionally,
all nonoperating failure rates are shown to be accelerated only by
environmental stress. The relative increase in failures dlue to
6 environmental
nonoperating.
stress is
These conceptual
shown to be the same for operating and
examples assume
differences have been numerically compensated for.
that temperature
*. Figure 4.5-4 represents the failure behavior of a specific part type

which has operating failure mechanisms primarily accelerated by
environmental stresses. The difference in operating and nonoperating
failure rate is relatively small, and the ratio of airborne to ground
failure rate is slightly greater for nonoperating than operating
"applications. Proposed nonoperating environmental factors would be
slighter greater than operating factors for part types with failure
* behavior similar to that depicted in Figure 4.5-1. The contribution of
"* the constant operational stress term causes this phenomenon. There is no
operational stress contribution for the nonoperating case and therefore
the increase in environmental stress failure rate is a larger percentage
. of the total failure rate. Multiplicative environmental factors predict
the relative increase in total failure rate. Thus, the nonoperating
environmental factors would be slightly larger. It must be emphasized
that this does not mean that the nonoperating failure rate would be
slightly larger, but that the rate of change of failure rate is greater in
the nonoperating mode.
Figure 4.5-2 represents the failure behavior of i specific part type

which has operating failure mechanisms primarily accelerated by
4-35
Yy. , ?
KEY:
0 Failures due to Operational Stress
E Failures due to Environmental Stress
Failures due to combined Operational/
Environmental Stress
OP.
U.'
NON-OP.
OP. E
0
RON-OP.
E E
0 1
GROUND AIRBORNE
FAILURE MECHANISMS FAILURE MECHANISMS
FIGURE 4.5-1: FAILURE MECHANISM DISTRIBUTION,

ACCELERATED BY ENVIRONMENTAL STRESS (CASE I)
OP.
S OP. E
- E
" 0 0
"~ 0NON-OP.
NOW-OP
GROUND AIPRORNE
FAILURE MECHANISMS FAILURE MECHANISMS

ACCELERATED BY OPERATIONAL STRESS (CASE II)
4-36
V4 N
KEY:
0 Failures due to Operational Stress
E Failures due to Environmental Stress
Failures due to combined Operational/
L Environmental Stress
OP.
E
LU
L-J
Lj-
OP.
LU
NON-OP.
- ~ NON-OPE
GROUND AIRBORNE
7AILURE M4ECHANISMS FAILURE MECHANISMS

ACCELERATED BY COMBIN ED OPERAT Ia@AL/NV IRONMENTAL STRESS
(CASE III)
4-37
y,*..
,7
Pi Downloaded from http://www.everyspec.com
"operational" stresses. The difference in operating

and nonoperating
failure rate is relatively large, and the ratio of airborne to ground
failure rate is significantly larger for nonoperating than operating
applications. As the environment changes from ground to airborne for this
class, the number of failures would be expected to increase for both
operating and nonoperating applications. This expected increase would be
relatively less for the nonoperating condition because of the smaller
probability of failures due to combined environmental/operational stress.
However, the expected failure rate increase for the nonoperating
environment would be a much greater percentage of the overall failure
rate. This is because the nonoperating failure rate is relatively low in
the ground based environments. Therefore, the magnitude of proposed
nonoperating environmental factors would need to be greater than the
multiplicative operating MIL-HDBK-217D environmental factors. This can be
"observed in Figure 4.5-2.
Figure 4.5-3 represents the failure behavior of a specific part type

in which operating failure mechanisms are primarily accelerated by the
simultaneous exposure to environmental and operational stresses. This
case represents perhaps the most common and also the most complex. The
example depicted in Figure 4.5-3 shows the ratio of airborne to ground
failure rate to be greater for operating than nonoperating applications.
The example was constructed with the rate of increase for the combined
stress failure mechanisms greater than the rate of increase for the
environmental stress failure mechanisms. This sevmed logical because the
expected number of failures was greater for mechanisms accelerated by the
combined stresses. However, this was not necessarily the case, and
additionally, the relative difference between the rates of increase had
important implications for nonoperating environmental factor development.
To further investigate this situation, numerical simulations were
I performed assuming a variety of failure mechanism distributions and rates
of increase. The rate of increase for the sum total of all failure
*. mechanisms was set equal to MIL-HDBK-217D operating environmental factor.
Results of the simulation process (for part types dominated by failure
mechanisms accelerated by combined stress) were that, (1) on average, the
I! 4-38
/"
.,.4
._°
magnitude of multiplicative nonoperating environmental factors were

slightly less than the operating factors, and (2) as the percentage of
failure mechanisms due to only operational, or only environmental stresses
increase, the magnitude of the nonoperating environmental factors approach
and possibly surpass the operational environmental factors. It must again
"be emphasized that this does not infer that the nonoperating failure rates
are higher. The operating and nonoperating environmental factors are in
separate ranges. Therefore the nonoperating environmental factors can be
higher, but the resultant nonoperating failure rates lower.
The failure mechanism distributions estimated previously for specific

part types were further investigated to determine which of the examples
depicted in Figures 4.5-1, 4.5-2, and 4.5-3 more closely approximated the
respective failure mechanism distributions. It was found for most part
types that the operating failure mechanisms were accelerated prirArily by
"the combination of operational/environmental stress, or by environmental
stress alone. In either case, the expected difference between operating
and nonooerating environmental factors was anticipated to be small.
Therefore, for these cases, it was proposed that the operating
environmental factors could be applied for nonoperating failure rate
prediction purposes. The inexact nature of the nonoperating environmental
"factor determination process and the imprecision involved with estimating
failure mechanism distributions precluded the use of adjustment factors
which were much smaller than the inherent noise included in part level
reliability data or the variability involved in failure rate prediction.
The failure mechanisms for several part types were primarily

accelerated by operational stress, and therefore the operating
"environmental factors could not be applied to nonoperating failure rate
prediction. For these part types, the range of nonoperating environmental
factors were anticipated to be significantly larger than the range of
temperature adjusted (if required) operating environmental factors.
Incandescent lamps, la3ers and microwave tubes were included in this
category.
"4-39
• -:" .', -- •'-••-=•-• •z: .• .- : , , ... •-<- ,--•-.- ... ... .4...
; • • :•.- ;' •••
•.'•: ]•.'.•
i-"t]•
A general nonoperating environment factor expression was derived for

these parts by solving simultaneous equations to determine nonoperating
environmental factor as a function of operating envirunmental factor, R1,
. R2 and R3. R1 was defined as the fraction of operating failures due to
only environmental stress. R2 was defined as the fraction of operating
failures due to simultaneous exposure to environmental and operational
stress. R3 was defined as the fraction of operating failures due to only
operational related stress. The equations which were solved
"simultaneously are presented in Table 4.5-3. There were six unknowns
(i.e., YNE, Xei, Xef, Xeoi, Xeof, X0 ) and six indeoendent equations.
Equation 5 in Table 4.5-3 is not independent. 'TOE (the MIL-HDBK-217
* environmental factor) is a known quantity. This method presupposes that
estimates of R1 , R2 and R3 could be found from the available literature.
* The nonoperating environmental factor expression was determined to be the
following expression.
IrOE R3
*'NE R1 +R 2 R + R2
where
nNE = nonoperating environmental factor
OE = operating environmental factor

R1 = fraction of operating failures due to only environmental stress
(0 < R1 < 1)
R2 = fraction of operating failures due to simultaneous exposure to
environmental and operational stress (0 < K2 < 1)
R3 = fraction of operating failures due to only operational related
stress (0 < R3 - 1)
It was noted that the expression only applies when (R1 + R2 ) is greater
than 0.20. Also, it should be noted that R1 + R2 + R3 = 1. Estimates of
R1 , R2 and R3 valves were determined from the available literature. Rj,
R2 and R3 values used for lamps, lasers, and microwave tubes are presented
in the respective model development section.
4-40
Sae
TABLE 4.5-3: ENVIRONMENTAL FACTOR ANALYSIS (CASE

II) - SIMULTANEOUS EQUATIONS
NO. EQUATION BASIS
I rNE )-ei/Xef Definition
OE ef + Xeoi + ko Definition
3 RI )tef Definition
';..Kef + keof +- Xk
• X eof
4 R2 =eof
Xef + Xeof + Xo Definition
5 R3 X Definition
Xef + ef+A
6 R1 + R2 + R3 = 1 Equations 3, 4 and 5
7 Xei Xef Assumption

Xeoi Xý
NOTES: fNE = nonoperating environmental factor

TOE = operating environmental factor
"Xei = failure rate due to environmental stress in the ith
environment
Xef failure rate due to environmental stress in a fixed
environment
Xeoi = failure rate due to combined operational/environmen-
tal stress in the ith environment
Xeof = failure rate due to combined operational/environmen-
tal stress in a fixed environment
Xo = failure rate due to only operational stress
(independent of environment)
U•4-41
Vo
/; .. ,". :Y
The fourth phase of the nonoperating environmental factor development

process was to consolidate the findings from the first three phases and
propose numerical Eleven generic part categories
adjustment factors.
. required environmental factor adjustments. For the remainlng part types,
it was determined that the effects of environmental stress wre
sufficiently similar to propose the sane numerical values for nonoperating
environmental factors as the documented MIL-IIISK-217D operating
environmental factors. For thermistors, switches, printed wiring
assembiies, connections, crystals, fuses, meters and circuit breakers, the
, nonoperating environmental factor was determined to be given by the
following equation.
INE = ArOE
where
;fNE = nonoperating environmental factor

AT = temperature adjustment factor
= f(environment, part type)
WOE = MIL-HDBK-217D operating environmental factor
For tubes, lasers and incandescent lamps, the nonoperating environmental

factor is given by the following equation.
S:"•N
.
ATMOE
=R1 + R2
ITNER+R
R3
R1 + R2 jI
where all variables have been previously defined.
An additional modification was required for microcircuit nonoperating

environmental factor development for two reasons. First, the operating
environmental factor is part of a complex, nonlinear model form.
I
Conversely, the proposed nonoperating failure rate prediction model is a
multiplicative model with one nonlinear term. The second reason was that
the nonoperating failure mzchanism distribution for hermetic and
nonhermetic devices varies considerably. Specifically, nonhermetic
"4-42
S.-- .,.,• . "9- .' . .--- '.'' . ,',"," -,."'"."- . ,,:•., '•:..' Z , .% -.- '''''''...? ' .. ' .•r:
devices were determined to be significantly more sensitive to

environmental stresses, particularly humidity and contaminated
environments. The microcircuit environmental factor derivation was
determined to be.
WNE = AiwOE + Bi
where
Al and BI = normalization constants for hermetic devices

A2 and B2 = normalization constants for nonhermetic devices
Numerical values for Al and B1 were estimated to be 0.66 and 0.75

respectively. These values were based on the typical model parameters
presented in MIL-HDBK-217D, Notice 1, Table 5.2-18, "Model Parameters for
Random Logic and Memory Microelectronic Semiconductor Devices".
Additionally, the equation was normalized so that the nonoperating
environmental factor for a ground benign environment was equal to one.
Estimates for A2 and B2 for nonhermetic devices were determined to be 1.4
and 0.46. The higher environmental factor values for nonhermetic devices
were hypothesized using the results ie the Panama nonhermetic microcircuit
life test data (described in Section 2.3) as a foundation.
Determination of nonoperating environmental factors allows for

analysis of the reliability for any proposed mission profile. Many
specific mission profiles consist of a composite of environments. For
example, missile mission profiles may consist of storage, dormancy,
transportation and captive carry environments. Definition of the
appropriate mission profile is required including specific time periods in

individual environments and a detailed description of the test plan. This
information can be used with the methods described in Section 6.0,
"Equipment Nonoperating ReliaThlity Prediction P-ocedure" to assess the
reliability of any conceivable mission profile. For example, a comparison
of storage and dormant applications can be performed by defining the
respective environmental profiles and test plans, and then applying the
I -4-43
_, /j ~ 1. *"
"4 ' /
Iy
"methods developed in this study. This approach is preferable to previous
attempts to compare storage failure rates to dormant failure rates
directly. For example, missile electronics storage applications can range
from very benign environment controlled facilities to more stressful
uncontrolled storage. Dormant applications can also vary but generally
are in a ground fixed environment. Therefore, a comparison of storage
versus dormant missile applications is virtually meaningless without

proper environmental characterization.
In conclusion, it would have been preferable to analyze observed

nonoperating failure rate data in all 26 environment categories. However,
"inherent difficulties of collecting nonoperating reliability data
prevented the collection of data for any non-ground environments. A
viable approach was developed to compare temperatures and failure
mechanisms, and to generate nonoperating environmental factors from the
documented MIL-HDBK-217D operating environmental factors. This method was
used to derive nonoperating environmental factors for eleven part types.
For the remaining part types, it was determined that no adjustment was
required and that the operating environmental factors also approximate the
effects of environmental stress on nonoperating failure rate.
4.6 Screening Effectiveness
During the nonoperating failure rate model development process, it was

observed that the magnitude of the nonoperating quality factors were
generally small in comparison to the corresponding MIL410BK-217C operating
quality factors. Initially, this observation seemed unusual because the
devices are screened before entering into either an operating or
nonoperating state, and it was unclear why screening would effect the
nonoperating failure rate any differently than the operating failure rate.
The observation would probably have been attributed to variab'klity in the
data if not for the consistency of the results. The same trend was
noticed for microcircuits, transistors, diodes, resistors, capacitors and
inductors. In each case, the empirically determined nonoperating quality
factors were small in comparison with the corresponding operating values.
4-44
•,* - .<.,. •-
S•A..
To evaluate this apparent trend, a separate study was initiated concerning

k• the effectiveness of screening on nonoperating failure rates. The
argument presented in the following paragraphs is offered not as a
rationale to propose assumed nonoperating quality factors but as a r
possible explanation for the observed trend.
The intention of device screening is to lower the field failure rate

of a component lot by eliminating inherently weak devices. The fact that
the failure rate decreases as screening level increases is evidence that
the instantaneous failure rate is not constant with time, but is
decreasing. Since the number of weak devices is independent of
operational mode, the rate at which the hazard rate decreases following
the screen is the important parameter for quality factor comparisons.
Whether a sample of parts is intended for long term storage or

immediate operation, the percentage of inherently weak devices surviving
an identical screen is relatively constant. However, the rate at which
the surviving weak components fail is what indicates the relative effect
on operating or nonoperating failure rate. Weak devices will clearly fail
more quickly in an operating mode. This is why devices are "burned-in"
with the power applied.
The operatingquality factors have generally been determined from

empirical data and represent the ratio of the failure rate for unscreened,
and various levels of screened parts. The MIL-HDBK-217D failure rates are
constant with .time and therefore represent some average failure rate over
an unknown data collection interval. Since the weak devices surviving the
screen fail more quickly during operation, the computed operating quality
factors appear to be more sensitive to screening. However, given infinite
time, the same number of inherently weak devices for a similar screen
would fail in either operating or nonoperating applications.
An example is provided to illustrate that the observed nonoperating

failure rate screening effectiveness was probably also the expected
result. Figure 4.6-1 presents two graphs of failure rate versus time for
4-45
-- X
S.. . . •. • " , . . ; . (
a, I
a JJ
., •. I-,,'. 7 •'I-. ,'" w.'," • • -,+ ,.r,.•.-•• I•-,••- -e-'••• . .--. • ..
• -,
", -•,--v•
. 4." ; • • -. .- '. "• " 4. :. .

. . . % . ., m ,% I.,. .I , . . .,
4-46
06/15/2020 13:17 3153301991 PUBLIC AFFAIRS PAGE 02
similar samples of parts screened to similar specifications. In the upper

graph, the devices are operating. The operating failure rate is shown to
be decreasing and asymptotic to some constant failure rate. In the lower
graph, the devices are nonoperating and the failure rate is asymptotic to
some lower constant failure rate. In each case, the number of weak
devices surviving the screening process is constant and represented by the
area between the failure rate curve and the dotted line. The time at
which virtually all the devices have failed is much shorter for the
operating case. Applying the concepts of geometry, it can be concluded
that the differential between the instantaneous failure rate and the
constant failure rate must be smaller in the nonoperating state if the
number of weak devices (i.e. area between the solid and dotted lines) is
the same.
Additionally, both operating and proposed nonoperating quality factors

correspond to some unknown data collection interval near the beginning of
the devices life. As the data collection interval becomes shorter and
closer to time zero, the expected difference between operating and
nonoperating quality factors would be maximized. As the data collection
interval approaches infinity, the ratio of operating to nonoperating
quality factors would theoretically become unity. Therefore, the
numerical quality factor values, both operating and nonoperating, are an
indication of screening effectiveness, data collection interval length and
operational mode.
To further investigate the issue, a literature search was performed to

identify any sources which address the relative effects of screening on
operating and nonoperating failure rate. One document was located
(Reference 25) which concluded on the basis of a theoretical analysis that
dormant and operating failure rates for electronic parts tend to equality
as part quality improves. This conclusion was equivalent to stating that
as part quality worsens, the difference between dormant and operating
failure rates would increase. Thus, MIL-HDBK-217D style quality factors
would have to be numerically larger in the operating state to reflect the
increasing difference in failure rate.
4-47
JUN-15-1900 12:57 3153301991 P.002

This discussion represents a brief comparison of operating and

nonoperating instantaneous failure rates and screening effectiveness. The
concepts presented here were not used to develop or assume any of the
nonoperating quality factors presented in this report. The nonoperating
quality factors were all determined empirically where data ailowed and
extrapolated for quality levels without data. This discussion was
presented simply as a possible explanation for an observed trend which
consistently occurred during the electronic component nonoperating failure
rate modeling efforts performed in this study.
i.
p,.
4-48
•''- ."-".•'-.
.".,',,,a''•,'• ."', .. ' ."-"".-.,".,"j " ".'-> :• ..: . •,..
*. .-..
. ,o.j
A,- • , .~ •_L ,,2 -. .- ..- -"-"-"- -"
\ 5.0 NONOPERATING FAILURE RATE MODEL DEVELOPMENT
Nonoperating failure rate prediction models were developed for each

part type in MIL-HDBK-217D. The integrity of the proposed models is
related to the diversity and quantity of the available nonoperating
"failure rate data. Table 5.0-1 preseaits an overview of the amount of data
available for analysis and the basis for model development. Each part
class is divided into a category of either "sufficient data", "some data"
or "assumptions" depending on the basis for model development.
Categorization decisions were subjective. Consideration was given to the
number of data records, number of failures, part hours and number of
potential variables.
TABLE 5.0-1: PART CLASS CATEGORIZATION
SUFFICIENT DATA SOME DATA ASSUMPTIONS

Digital ICs (1) Memory ICs Bubble Memories
Linear/Interface ICs Hybrid ICs Opto-Electronics
Transistors Tubes Lasers
Diodes Rotating Mechanisms Connections
"Resistors Relays
. Capacitors Switches
- Inductors Connectors
"P.W. Assemblies
Miscellaneous Parts
NOTES (1): Digital ICs refers to digital SSI/MSI, random logic LSI and
microprocessor devices.
The general model development approach described in Section 4.1 was

"followed to the extent possible. For part classes without empirical data,
the proposed models were based on theoretical considerations and
hypothetical reliability relationships. Required assumptions are clearly
* stated in each model development section. Table 5.0-2 presents an
overview of the model development process.
5-1
TABLE 5.0-2: MODEL DEVELOPMENT OVERVIEW

PART CLASS DATA SOURCES EMPIRICAL
FACTORS ASSUMED/THEORETICAL
FACTORS
Monolithic ICs MICOM complexity environment

Martin Marietta temperature
F-16 HUD/RIW logic
Sandia screening
PRC enclosure type
RAC
N
ERADCOM power cycling
Hybrid ICs MICOM # ICs environment

F-16 HUD/RIW # transistors quality
# diodes
Bubble
Memories none none # gates
# loops
temperature
environment
logic
Transistors MICOM device style environment
Martin Marietta temperature
F-16 HUD/RIW quality
AFCIQ power cycling
PRC k~
Diodes
ERADCOM
MICOM
Martin Marietta
F-16 HUD/RIW
device style
quality
power cycling
environment
temperature
H
NO
PRC
Opto-elec- none none device style
tronics quality
environment .'
Resistors MICOM device style environment
Martin Marietta quality
F-16 HUD/RIW power cycling
PRC
L_4
r•L.
5-2
TABLE 5.0-2: MODEL DEVELOPMENT OVERVIEW (CONT'D)

EMPIRICAL ASSUMED/THEORETICAL
PART CLASS DATA SOURCES FACTORS FACTORS
Capacitors MICOM device style environment

Martin Marietta quality -
F-16 HUD/RIW power cyclina
Inductive MICOM device style environment

Devices Martin Marietta qu3lity
F-16 HUD/RIW power cycling
AFCIQ
Southern Tech.
Lasers none none environment

style
device
Tubes MICOM device style environment
Rotating MICOM device style none

Mechanisms Martin Marietta
Relays MICOM quality enclosure type

Martin Marietta contact rating
Hughes environment
AFCIQ
Switches MICOM none enclosure type

Martin Marietta contact rating
environment
quality
Connectors MICOM none device style

Martin Marietta environment
P.W. Assemblies MICOM technology # PTHs

AFCIQ # circuit
planes environment
i
Connections none none type
Misc. Parts MICOM device type none K

Martin-Marietta
I
5-3
'i mp'- , -. ,
5.1 Microcircuits
The generic category of microcircuits includes an overwhelming number

of devices with diverse characteristics. Failure rate model development
could not be attempted at the "microcircuit level" without making gross
generalities and oversimplifications. Therefore the nonoperating failure
rate model development process (and the collected data) needed to be
divided into a workable number of categories which would provide
sufficient data per category, while properly categorizing the microcircuit
family. It was decided to divide the microcircuit model development
process into five distinct sections. The five sections were (1) digital
SSI/MSI, random logic, and microprocessors, (2) linear/interface, (3)
memory, (4) hybrid, and (5) bubble memories. Data was segregated into
these categories, and the reliability attributes were studied
individually. Ideally it would have been advantageous to divide the model
development process to even finer subcategories. However, practical
restraints, most notably data limitations, prevented any finer division
beyond these categories.
5.1.1 Monolithic Microcircuit Nonoperating Failure Rate Prediction

Models
This section presents the proposed nonoperating failure rate

prediction models for monolithic microcircuits. Separate models were
developed for (1) digital SSI/MSI, random logic and microprocessor
devices, (2) linear/interface devices, and (3) memory devices. The models
are presented in Appendix A in a form compatible with MIL-HDBK-217D.
Digital SSI/MSI, Random Logic Devices and Microprocessors
The proposed model for digital SSI/MSI, random logic and

microprocessors is represented by the following equation.
=nb rNT rNQ -aNE ffcyc

Xp
5-4 .. ,,.
where
Xp = digital SSI/MSI, random logic LSI and microprocessor

nonoperating failure rate
6
Xnb = nonoperating base failure rate (failures/10 hours)
= .00029(Ng)"', Ng <j3100 gates
where
= .014, Ng > 3100 gates
9I
Ng = number of gates
nNT = nonoperating temperature factor
1 1
= K3 + K(4 exp(-An(T 1-8)
where
T = nonoperating temperature (OK)

K3, K4, An = temperature coefficients (See Table 5.1.1-1) •
•NQ = nonoperating quality factor (see Table 5.1.1-2)

•NE = nonoperating environmental factor (see Table 5.1.1-3)
•cyc = equipment power on-off cycling factor
1 + n2(Nc)
where

hours
5-5
TABLE 5.1.1-1: DIGITAL MICROCIRCUIT NONOPERATING TEMPERATURE

FACTOR CONSTANTS
Technology K3 K4 An
TTL,HTTL,DTL,ECL .91 .09 4813
LTTL,STTL .90 .10 5261
LSTTL .89 .11 57111
IIL .86 .14 6607
MNOS .61 .39 6607
PMOS- .68 .32 5711
NMOS, CCD .65 .35 6159
CMOS, CMOS/SOS ,58 .42 7059
TABLE 5.1.1-2: MICROCIRCUIT NONOPERATING QUALITY FACTORS
Quality Level wNQ
S 0.53
B 1.0
B-1 1.4
B-2 2.0
C 2.3
C-1 2.4
D 2.5
D-1 8.7 '
5-6
TABLE 5.1.1-3: MICROCIRCUIT NONOPERATING ENVIRONMENTAL FACTORS (1tNE)
Hermetic Nonhennetic
Environment Devices Devices
GB11I
GF 2.4 4.0
GM3.5 6.5
MP 3.2 5.9
NSB 3.4 6.2I

NS 3.4 6.2
NU 4.5 8.6
NH 4.6 8.9
-NUU 4.9 9.5
ARW 6.3 13
AIC 2..4 4.0
AIT 2.7 4.7
AIS 4.0 7.6
AlA 3.4 6.2
AIF 4.7 9.0
AUC 2.7 2.7
AUT 3.4 3.4
AUB 5.7 11
AUA 4.7 9.0.
AUF 6.7 13 14,
MFF 3.3 6.0

MFA 4.3 8.2
USL 8.0 16
ML 9.3 19LI'
CL 150 310
Notes: (1) Space flight environment was not addressed in this study.
5-7
Linear and Interface Devices
The proposed model for linear and interface devices is the following
equation.
Xp= Xnb "NT VNQ "NE lTcyc
where
Xp = linear/interface device nonoperating failure rate.•-'•,,
V
6
Xnb = nonoperating base failure rate (failures/10 nonoperating
hours)
= O.O0024(Nt)" 87
where
Nt = number of transistors
"wNT = nonoperating temperature factor
= exp(-4748(T - ))
wNQ = nonoperating quality factor (see Table 5.1.1-2)
1!NE = nonoperating environmental factor (See Table 5.1.1-3)
wcyc = equipment power on-off cycling factor
= 1 + .031(Nc)
V
.
where

hours "
Memory Devices
The proposed nonoperating failure rate prediction model is for RAM,

ROM, PROM and CCD memory devices, and is given by the following equation..
Xp =Xnb iiNT 1iNQ "nNE ircyc
5-8
¶;. •,,
where
Ip = memory device nonoperating failure rate

Xnb = nonoperating base failure rate (failures/10 6 nonoperating
hours ) /.,"
= 0.0034, bipolar memory devices

= 0.0017, MOS memory devices
INT = nonoperating temperature factor
= K3 + K4 exp(-An(
ep-nT 29-8
where VP
K3 , K4, An = temperature coefficients (See Table 5.1.1-1)

•NQ = nonoperating quality factor (see Table 5.1.1-2)
INE = nonoperating environmental factor (see Table 5.1.1-3)
kcyc = equipment power on-off cycling factor
H
= 1 + .O2(0)
where

hours
The proposed models for monolithic microcircuit are also presented in

Appendix A. The following sections describe the model development V
processes.
5.1.2 Digital Microcircuit Model Development
The failure rate modeling approach described in Section 4.1 was

successfully implemented to develop a model for digital SSI/MSI, random
logic devices dnd microprocessors. Model development for memory devices . ,
is presented in Section 5.1.4. A theoretical nonoperating failure rate
model was hypothesized based on known failure mechanisms and physics of
failure information (References 3, 26 and 27). The theoretical model was
5-9
evaluated and quantified with the available data. Failure rate prediction ,
model parameters for complexity, scrcening and temperature were determined

empirically from the data. Modifying factors for equipment power cycling
and application environment were determined by alternate methods.
Application and construction variables which characterize digital

microcircuits in a nonoperating environment are presented in Table 5.1.2-
1. These factors were determined whenever possible for all summarized
data. The variables presented in Table 5.1.2-1 represent possible failure
rate model parameters.
Development of a theoretical model for digital microcircuits was

conducted in two phases. The initial phase consisted of identifying
variables which hypothetically have the greatest impact on reliability.
It was determined that nonoperating ambient temperature theoretically has
the most significant influence on nonoperating reliability for hermetic
digital microcircuits. Other factors which have been identified in the
literature (References 3, 26, 27, 23 and 29) are humidity, complexity,
enuipment power cycling and screening. The second phase of the
theoretical model development process consisted of a study of various
model forms to determine which would be applicable to microcircuit
nonoperating failure rate prediction. Model forms which were studied
include multiplicative riodels, additive models and nonlinear model forms.
The MIL-HDBK-217D operating failure rate prediction model for

microcircuits is a combination of an additive and multiplicative model
form (Reference 19). The application stresses during nonoperating periods
are sufficiently different to preclude an assumption, without additional
justification, that, the nonoperating model form is the same or similar to
the operating model form. Nevertheless, a model form similar to the MIL-
HDBK-2170 microcircuit model was considered for nonoperating failure rate
model development. The mathematically attractive purely additive and
purely multiplicative models were also considered. After evaluating the
advantages and disadvantages of the various model forms, the decision was
5-10
TABLE 5.1.2-1: DIGITAL MICROCIRCUIT CHARACTERIZATION VARIABLES
I. Technology
A. DTL E. ECL I. IIL M. CCD
B. TTL F. LTTL J. MNOS N. CMOS
C. HTTL G. STTL K. PMOS 0. CMOS/SOS
D. DTL H. LSTTL L. NMOS P. HMOS
II. Number of Gates
III. Construction
A. Dip E. Chip Carrier
B. Can F. Quad In-Line (staggered leads)
C. Flatpack G. In-Line
D. Square
IV. Enclosure
A. Hermetic
B. Non-Hermetic
V. Package Material
A. Metal E. Glass I. Metal/Epoxy
B. Ceramic F. Plastic/Ceramic J. Silicon
C. Metal/Ceramic G. Epoxy K. Phenolic
D. Metal/Glass H. Ceramic/Plastic/Window
VI. Number of Pins
VII. Number of Interconnects
VIII. Die Bond

A. Eutectic
B. Epoxy
C. Glass
IX. Quality Level

A. S C. B-I E. C G. 0
B. B D. B-2 F. C-i H. 0-1
X. Application Environment
XI. Temperature
A. Rated
73 B. Actual
3
XII. Number of Power On/Off Cycles per 10 Nonoperating Hours
5-11
made to initially proceed with the general theoretical

model form
developed for this study. This model form is a multiplicative
model with
the exception of the equipment power cycling contribution.
The summarized digital microcircuit data collected in support of this

study effort is presented in Table 5.1.2-2. The nonoperating
failure rate
data collectively consists of 155 failures, 21441.1 x 106
part hours and
399 individual data records. Data was available for parts screened to S,
< B, C, D and D-1 quality levels as defined in MIL-HDBK-217D.
TABLE 5.1.2-2: CIGITAL MICROCIRCUIT NONOPERATING

FAILURE RATE DATA
Data Part
* No. Equipment/Source Records Hours6
Complexity Tech. Failures (x10 )
1 Hawk/MICOM 6 SSI bipolar 1 768.9
2 Maverick/MICOM 3 SSI bipolar
3 Maverick/MICOM 1 432.4
2 MSI bipolar 1 308.9
4 F-16 HUD/RIW 36 SSI bipolar 10 4'20.7
5 F-16 HUD/RIW 50 MSI bipolar 7
6 F-16 HUD/RIW 3398.2
2 LSI bipolar 0 129.8
7 F-16 HUD/RIW 2 LSI MOS 0
* 8 26.0
Martin Marietta 3 SSI/MSI bipolar 18 9550.8
9 Sandia I LSI MOS 1
10 Sandia 46.7
1 SSI bipolar 0 2432.1
11 PRC 2 SSI/MSI bipolar 0
12 RAC (1) 3.8
120 SSI bipolar 6.9
13 RAC (1) 14 SSI M51 43
14 RAC (1) 130 0.4
MSI bipolar 20 4.3
15 RAC (1) 2 LSI bipolar 0
16 RAC (1) 1 VLSI (2) 0.1
bipolar 1 0.1
17 ERADCOM (3) 17 SSI bipolar 18
18 ERADCOM (3) 7.7
7 SSI MOS 17 3.3
TOTALS 399 155 21441.1
1) High temperature storage life test data.
1 Storage temperatures
range from 150 0 C to 350 0C.
I 2) VLSI parts defined as having greater than 3000 gates.
3) Panama life test data for nonhermetic devices.

temperature 300C, relative humidity = 90%. Ambient
5-12
-V•.
N ~Early in the model development process, the issue, of device complexity

was addressed. The terms "complexity" and "complexity factor" are
convenient, yet ambiguous. Complexity is a relative term. A device may
be relatively complex in a number of ways; process, function, size, etc.
It was considered desirable to investigate the effects of nonoperating
failure rate versus a quantitative measure of complexity. Possible
measures of complexity were evaluated for digital microcircuits. Several
of the possible complexity measures which were given consideration are the
following:
o Number of gates
o Number of transistors
o Number of package pins
o Number of package 1/0 functicnis
o Die area
In addition, the possibilities. were considered that more than one

complexity measure should be included in the optimal model form for
digital microcircuits, or that perhaps complexity had only a negligible
effect on nonoperating failure rate.
After evaluating the different measures of complexity, it was

determined that the number of gates and the number of package pins, were
the two best complexity measures. These two comiplexity measures would be
further investigated by analysis of the available data. Both parameters
represent information which is readily availahie, and theoretically have
an effect on failure rate. For VLSI digital devices, the gate count may
not be available. This potential problem is,analyzed at a later stage of
the model development process. Die area was initially thought to have
merit as a measure of complexity. Ho~wever, it was found that die area
information is often vendor proprietary, and thus would not be suitable
for failure rate prediction purposes.
The digital microcircuit data were initially analyzed without merging

data records. Point estimate failure rates were calculated for data
5-13
entries with observed failures. Thc- methods described in Section 3.3 were
used to determine which zero faiiurp ýAt. ,n'ries had sufficient part
hours to estimate a failure rate without failures. Correlation
coefficients were computed for every pair of 4 ndependent variables. In
addition, the available data was evalupted to identify weaknesses in the
data set. Results from these initial analyses are given below.
1) All collected data were from a ground based environment or from

storage life testing. Therefore there was a insufficient range of
environmental stresses represented in the data to develop
environmental factors based on data analysis.
2) All data were for digital microcircuits with one of the following
logic types; DTL, TTL, LSTTL, ECL, CMOS, or CMOS/SOS.
3) 88% of all MOS data records were for high temperature or high
humidity applications.
4) 86% of all field data records were for 14 or 16 pin devices.

5) The correlation coefficient for number of gates vs. number of pins
was 0.52.
6) 100% of all data on nonhermetic devices were from high temperature
or high humidity testing.
It had been desired to analyze the effect that both number of pins and
number of gates had on nonoperating failure rate. However,
characteristics of the data base and conclusions from the preliminary data
analysis made it very difficult to properly address the effect of package
pins on failure rate. Proper application of regression analysis required
that all independent variables were uncorrelated (i.e. correlation
coefficient not significantly different from zero) and that there was a
large range of variable values represented in the data.
The observation that gate count and pin count were correlated was
further investigated. A recent study (Reference 30) regarding VLSI device
reliability indicated that there was little or no correlation between gate
count and pin count. After further studying the issue, it was determined
that there is theoretically an intermediate amount of correlation between
5-14
II
4i
pin count and gate count for SSI and MSI devices. Then, as device
complexity enters into the LSI and VLSI ranges, the correlation decreases.
Field data was not available for the most current state-of-the-art LSI and
VLSI devices. The long nonoperating times required to observe failures
iD,•.precludes the existence of such data. Nsa
available for analysis were concentrated in the SSI and MSI ranges. This
explains the observed correlation. The implications of the observed
correlation are two-fold. First, regression analysis could not be
correctly applied to quantify the effects of both variables. Second, a
regression analysis for failure rate versus gate count implicitly includes
the effects of pin count for SSI and MSI devices. That is, a regression
analysis for failure rate against either one of the two variables includes
the effects of both because the number of pins and the number of gates
increase together for the lower complexities. Therefore it was decided to
include only gate count as a measure of complexity. The natural logarithm
of the complexity was designated as C1.
It was determined in the theoretical model development process that

ambient temperature was the most significant variable influencing
nonoperating failure rate for digital microcircuits. It was also assumed
that the effects of temperature would be different depending on the logic
type of the device. In addition, the equivalent Arhennius relationship
(described in Section 4.4) was determined to be applicable to nonoperating
failure rate prediction. The preliminary form of the nonoperating
temperature factor was determined to be:
lNT exp(-el(1
where
IC" 7NT = nonoperating temperature factor

eV = equivalent activation energy
K = Bnltzman's constant = 8.63 x 10- 5 eV/OK
STr = reference temperature = 298 OK
5-15
- W C , . r. W - rr
The addition of the reference temperature term was for convenience. The
temperature factor was defined to be equal to unity when the ambient
temperature is equal to the reference temperature. The value of 298 0 K was
selected to be consistent with the reference temperature term used for
failure rate prediction in MIL-HDBK-217D.
It was assumed that the equivalent activation energy would vary with
logic type. Data was available for DTL, TTL, LSTTL, ECL, CMOS and
CMOS/SOS. Introducing temperature (or inverse of temperature) into a
regression by itself would result in only one temperature coefficient.
Therefore, a temperature matrix with two temperature variables was defined
to derive a temperature factor of the desired form. Introducing the
temperature matrix as independent variables in a regression analysis
allowed for computation of unique activation energies. The logic types
were grouped into the categories of (1) bipolar or (2) MOS technology.
The temperature matrix is presented in Table 5.1.2-3.
TABLE 5.1.2-3: DIGITAL MICROCIRCUIT TEMPERATURE MATRIX
Technology 1/T1 1/T2
bipolar 1/T 0
MOS 0 1/T
The following equations depict the relationship of nonoperating

failure rate versus I/T1 and 1/T2. These equations are conceptual in
nature, and temporarily ignore all variables except temperature.
ln(X) = bo + bl(1/T1) + b 2 (1/T2) + -
X= f(1/T1, I/T2) exp(bo)exp(bl(lfI1) + b 2 (1/T2))
Al = f(1/T,O) = exp(bo)exp(b 1 (1/T1) + b2 (0)), bipolar

= exp(bo)exp(b 1 (1/T1)), bipolar
5-16
I-I
X2 f(O,1/T) = exp(bo)evp(bl(O) + b2 (1/T2)), MOS

= exp(bo)exp(b2(1/T2)), MOS
bo, bI and b2 are regression coefficients and e is the residual. The

equivalent activation energy for the logic types in the bipolar category
was equal to the bI coefficient multiplied by Boltzman's constant.
Similarly, the equivalent activation energy for the MOS logic types was
equal to the b 2 coefficient multiplied by Boltzman's constant.
The theoretical model for digital microcircuits was determined to

include an equipment power cycling factor. The factor was assumed to be
of the following form.
Cycling Factor = 1 + K(cycles/10 3 hr)
where K is a constant. The other factors in the theoretical model were

multiplicative. Development of an equipment power cycling factor of this
form resulted in a proposed model which is neither linear nor can be
transformed into linear form. Therefore, application of regression
analysis was more difficult. This problem and a proposed solution is
described in-depth in Section 4.3. As recommended in Section 4.3,
equipment cycling rate was temporarily treated as a qualitative variable
and three "dummy variables" were defined (cycl, cyc2, and cyc3).
A matrix of "dummy variables" (0 or 1) was defined to accommodate

quality level id quantitative analyses. The quality level matrix is
presented in Table 5.1.2-4. In addition, a technology qualitative
variable was designated TECH. The TECH variable was assigned a value of
zero for bipolar devices and one for MOS devices.
5-17
!Z.M
TABLE 5.1.2-4: DIGITAL MICROCIRCUIT QUALITY VARIABLE MATRIX
Quality Level QI Q2 Q3 Q4
S 1 0 0 0
B 0 0 0 0
C 0 1 0 0
D 0 0 1 0
D-1 0 0 0 1
The data were then merged according to equipment, technology,

temperature, quality level and gate count. The data were merged by
summing failures and part hours.
part hours per data record, data within
intervals were merged.
To maximize the number of failures and
complexity categories of 10 gate

For example, all data for devices which were
similar in regard to equipment, t~chnology, temperature and quality level,
and which have between 21 and 30 gates were merged. The average number of
I3
gates for the category was used as the measure of complexity for each
merged data record. There were a total of 33 merged data records
available for further analyses.
At this stage in the model development process, the data were in a

format suitablefor application of regression analysis. All of the
available data were utilized except the Panama life test data. This
testing was performed for noihermetic devices found in the Panama jungle.
The observed failure rates from the Panama testing were not representative A
of nonhermetic microcircuits in a ground fixed environment. Additionally,

there was neither an accurate method to extrapolate the observed failure
rates to typical ground fixed conditions nor sufficient detail in all data t
sources to develop a unique humidity factor. Therefore, this data were
not included in the regression analysis.
Results of the regression analysis are given in Table 5.1.2-5. The

dependent variable was ln(failure rate) and the independent variables were
TECH, 1/T1, I/T2, QI, Q2, Q3, Q4, C1 , cycl, cyc2, and cyc3. The R-squared
value computed for the analysis was 0.98. However, this value was
misleading. The absence of temperature values between 200C and 150oC
5-18
7e
together valuethethen
R-squared with dominant
would effect
have of
beentemperature resulted
observed if in a much higher
the temperatures were
evenly distributed.
TABLE 5.1.2-5: DIGITAL MICROCIRCUIT REGRESSION RESULTS

standard confidence
variable coefficient error f-ratio limit
TECH 7.524 1.368 30.26 0.90 V'PA
1/TI -4812.803 205.468 548.67 0.90

1/T2 -7056.722 550.861 164.10 0.90
Q1 -0.637 0.698 0.83 0.60 U
Q2 0.836 0.698 1.44 0.70
Q3 0.882 0.279 9.99 0.90 ¾,
Q4 2.158 0.520 17.20 0.90
C1 0.477 0.086 30.88 0.90
bo 9.368 --.. " --
The coefficients given in Table 5.1.2-5 were the results of a

regression with the dependent variable equal to the natural logarithm of
failure rate. Transforming the regression solution into an equation where
failure rate (as opposed to log of failure rate) is the dependent variable Y.
results in the following preliminary multiplicative model.
Xd Ab(# gates)"7 exp(-An/T) wNQ JA
where l
Xd = predicted digital microcircuit nonoperating failure rate

(failures/106 hours)
Ab = preliminary base failure rate constant
= exp(7.524 + 9.368(TECH))
= 1.852 X 103 , bipolar
= 2.168 x 107, MOS
5-19
SLIL,
An = tPmperature coefficient
= 4813, bipolar
=7057, MOS
rNQ = nonoperating quality factor
= exp(-0.637(Qi) + 0.836(Q2) + 0.882(Q3) + 2.158(Q4))
=0.53, S
=1.0, B
- 2.3, C
= 2.4, 0
= 8.7, D-1
The variables TECH, Q1, Q2, Q3, and Q4 have been previously defined. It
is emphasized that the above equation represents a preliminary model,
which is only applicable to ground based environment.
Conclusions
results.
and observations were made based
These conclusions and observations are given in the following
on the regression
i1
paragraphs.
The nonoperating failure rate of D and D-1

significantly different, with 90% confidence, from devices screened to the
B quality level. The nonoperating failure rate difference between devices
quality devices was El
screened to C and B levels was significant at a 70% confidence, and the
difference between S and B quality level parts was significant at a 60%
confidence. It was considered very encouraging that the ranking of
quality factor values was precisely as expected based on the screening
specifications and screening effectiveness characteristics. Generally,
separate factors were not proposed in this study when the confidence limit
was lower than 0.70. However there are strong theoretical reasons why the
failure rate for S quality devices is lower than B quality devices. It
was assumed that the failure rate difference between
quality microcircuits (less than two to one)
magnitude as the statistical noise inherent
explain the low confidence.
S quality
was approximately the same
in the data. This would
and B
>1
......
5-20
wrp" a
The standard error statistic allows for the computation of confidence

intervals around the nonoperating quality factor estimates. Table 5.1.2-6
presents point estimate quality factor values, as well as lower and upper
90% confidence interval values. Confidence limit values could
not be
computed for B quality because all factors are in comparison to the B
level. In addition, Table 5.1.2-6 presents .interpolated quality factor
values for quality levels where no data was available.
TABLE 5.1.2-6: DIGITAL MICROCIRCUIT NONOPERATING QUALITY FACTOR
Quality Level WNQ, .05 1TNQI( 1 ) 1TNQ, .95 ItNQ2(2)
S 0.16 0.53 1.7 0.53

B 1.0 -- 1.0
B-I --... 1.4
B-2 ...... 2.0
C 0.70 2.3 7.6 2.3
C-1 -- -- 2.4
D 1.5 2.4 3.9 2.4
D-1 3.6 8.7 21.1 8.7
Notes: 1) TNQI is the point estimate quality factor.

2) nNQ2 is the point estimate quality factor supplemented by
interpolated values for B-i, B-2 and C-i quality levels. ,
It was noticed that the range of the nonoperating quality factors

(0.53, 8.7) was much less than the range of microcircuit operating quality
factors (0.50, 35) found in Table 5.1.2.5-1 of MIL-HDBK-217D. This was
believed to be indicative of a significantly slower rate for "infant
mortality" failure occurrence during periods of nonoperation. Therefore,
the effects of screening are less evident. This same trend was noticed
for practically all electronic component types, and is discussed in
Section 4.6 of this report. Conversely, package type appeared to have a
more severe effect on nonoperating failure rate. The best indication of V
the failure rate difference between hermetic and nonhermetic devices is
the ratio of the D-I quality factor to the D quality factor. The ratio is
2:1 for operating failure rate based on the information provided in MIL-
HDBK-217D. The ratio is 3.6:1 for nonoperating failure rate based on the
5-21
where
wNT,p = preliminary nonoperating temperature factor

An 2 temperature coefficient (given in Table 5.1.2-7)
T - nonoperating temperature (OK)
TABLE 5.1.2-7: MICROCIRCUIT NONOPERATING TEMPERATURE FACTOR COEFFICIENTS

Technology An
TTL, HTTL, DTL, ECL 4,813
LTTL, STTL 5,261
LSTTL 5,711
IIL, MNOS 6,607
PMOS 5,711
NMOS, CCD 6,159
CMOS, CMOS/SOS 7,057
No distinction was made during the nonoperating temperature factor

development process between hermetic and nonhermetic digital devices.
This was done for several reasons. All available data for nonhermetic
devices was life test data with a test temperature of 150 0 C. Therefore
separate temperature coefficients for nonhermetic devices could not be
determined empirically. A wide range of temperatures would have been
required to determine an empirical relationship. A second reason was that
the nonoperating failure rate difference between hermetic and nonhermetic
devices was accounted for by both the nonoperating quality factor and
environmental factor. It was determined that failure rate effects caused
by humidity and moisture intrusion are the major differences between
hermetic and nonhermetic devices. These factors were better predicted by
use of Yhe environmental factors. A third reason was that an assumption
of a more temperature dependent relationship for nonhermetic devices could
result in optimistic failure rate predictions (i.e. predicts too low a
failure rate) for temperatures less than the 150 0C test temperature. An
example which illustrates this relationship is presented in the following
paragraph.
5-23
The relationship of failure rate versus temperature is shown for two

different equivalent activation energies in Figure 5.1.2-1. If it is
assumed that nonhermetic devices exhibit a more temperature dependent
failure rate, then the example depicted in Figure 5.1.2-1 is analogous to
the study of activation energies for nonhermetic devices. The activation
energy (Eal) for the more temperature dependent failure rate is greater
than the activation energy (Ea2) for the less temperature dependent
failure rate, as would be expected. However, the failure rate is lower
for the more temperature dependent relationship for all points to the
right of the intersection point. It should also be noted that as
temperature decreases, the (1/KT) term increases. Therefore in the
example depicted in Figure 5.1.2-1, the failure rate is less for the Eal
activation energy relationship for all temperatures less than To (i.e. to
the right of the intersection point). The Ea2 activation energy is .
equivalent to the observed values for hermetic devices. The Eal
activation energy would be equivalent to an assumed higher An value for
ncnhermetic devices.
1nX
I V-".
k7--
FIGURE 5.1.2-1: FAILURE RATE VS. TEMPERATURE
The temperature intersection point was 150 0C for nonhermetic digital

devices. Any proposed model will predict the same failure rate at 150 0C
regardless of the activation energy because all nonhermetic data was for a
150oC test temperature. Therefore, assumption of a lower activation
energy (i.e. the regression solution values for hermetic devices) resulted
in a conservative failure rate estimate for all nonoperating temperatures
less than 1500C.
5-24
k
The number of gates was determined to be a significant variable with

90% confidence. The preliminary model assumed that digital microcircuit
nonoperating failure rate was proportional to the number of the gates
raised to a constant value. The regression solution for the constant was
0.477. Upper and lower 90% confidence interval values were 0.624 and
"0.330 respectively. The range of gate counts found in the available data
was from I to 3100 gates. The observed relationship between nonoperating
failure rate and the number of gates was a empirical relationship (as
opposed to a theoretical relationship). Therefore, extrapolation of the
observed relationship beyond 3100 gates could not be justified.
It was decided to include the gate count variable into an equation for
base failure rate. The base failure rate for digital microcircuits was
therefore given by the following equation:
47 7
.Xnb = Ab(# gates)"
"where
"Xnb= nonoperating base failure rate, (# gates < 3100)

Ab = constant
The constant term (Ab) was defined as a function of device technology and
the assumed reference temperature of 298 0 K (or 250C) due to normalization
of the nonoperating temperature factor. The temperature factor was
defined to be equal to a value of one for a temperature of 250C.
Therefore, the Ab constant must also correspond to 250C. In addition, the
technology dummy variable (TECH) coefficient was included in the equation
for base failure rate. Determination of appropriate values for the Ab
constant was computed by the following equation.
Ab = exp(9.368 + 7.524(TECH) - An(i/298))
where all variables have been previously defined. The 9.368 and 7.524
values were the bo and TECH coefficients previously presented in Table
5-25
5.1.2-5. Only the observed An values of 4813 for TTL,

DTL and ECL, and
7057 for CMOS and CMOS/SOS were used for the Ab determination process.
The extrapolated An values for other logic types were to determine the
relative (as opposed to absolute) effect of temperature. The respective
base failure rate constants for bipolar and MOS were therefore determined
to be,
bipolar: Ab = exp(9.368 - 4813(1/298)) =,.001134 failures/10 6 hrs.

MOS: Ab = exp(9.368 + 7.524 - 7057(1/298)) = .001126 failures/10 6 hrs.
It was noted that the two base failure rate constant values were the
same up to three significant digits. The fact that these two numbers are
"extremely close must in part be attributed to coincidence because failure
rate estimation techniques are not accsirate to three significant digits.
However, it was concluded from this analysis that the nonoperating failure
rates of bipolar and MOS devices were indistinguishable at 250C, if all
other parameters are equal.
The base failure rate expression was determined to be applicable from

1 to 3100 gates. This range of values includes all SSI/MSI devices and
enters into the LSI range of devices. The gate count range 'or LST
devices begins at 100 gates. The cut-off between LSI and VLSI is less
defined. Nonoperating failure rate estimation of digital devices with
greater than 3100 gates was further investigated to determine whether
extrapolated or assumed relationships beyond 3100 gates could be justified
or were warrented.
It was considered essential to determine methods to predict the

nonoperating failure rate for state-of-the-art devices. Continued
advancements in integrated circuit fabrication and processing techniques
have resulted in an electronics industry which is in a constant state of
trdnsition. Therefore, numerical analyses which require large quantities

of observed field data are necessarily outdated by the time the data is
collected. This problem was particularly acute for nonoperating failure
rate prediction model development because of relatively long mean-time-to-
5-26
failures in the nonoperating state. As a result, prediction models for

state-of-the-art devices could not be determined empirically.
Nevertheless, one of the primary objectives of this study effort was to
develop methods which can be applied to any conceivable equipment type and
mission profile. Therefore, alternate methods to assess high gate count
LSI and VLSI devices were considered.
One option which was considered to predict the nonoperating failure

rate for high gate devices was to assume that the observed relationship
for devices with less than 3100 gates (i.e. failure rate = (#gates). 4 77 )
would continue to be accurate for higher gate counts. This option was
rejected, however, based on the information provided in Reference 30. The
results of this recent study (documented in Reference 30) concluded that
"no numerical correlation between operating failure rate and number of
gates could be found for devices in the VLSI range with greater than 3000
gates. A possible reason for this lack of correlation was that the
majority of VLSI failure mechanisms are directly related to particular
fabrication process steps (such as mask registration). These failure
mechanisms effect all similar components on a die uniformally, and not
f random individual die components as would be indicated by a complexity
factor. Another factor confounding the results was the fact that as
"device complexity increases the associated fabrication processes are
improving, thus masking any true effect complexity may have. It was
assumed that this observed pattern would also be true for nonoperating
failure rate.
It was determined that an approximate base failure rate value for

random logic VLSI devices would be constant with respect to gate count
within the VLSI range, and could be found by solving the previously
determined expression for SSI/MSI/LSI devices for number of gates equal to
3100. Therefore, the approximate random logic VLSI nonoperating base
[1
failure rate was computed with the following expression.
Xnb,VLSI .00113(3100).7 477 .0523 failures/10 6 hrs
L- 5-27
. This value is differeit from the base failure rate presented in Section
.4 5.1.1 because the model had not been normalized for environment at this
stage of the model development process. It is emphasized that this
average value is only approximate. However, the proposed value is
""-'tuitively appealing for two reasons. First, the nonoperating failure
rate is independent of gate count for random logic VLSI devices. Second,
"the predicted nonoperating failure rates provide a smooth transition from
SSI/MSI to the LSI and VLSI ranges of device complexities.
None of the three equipment power cycling variables were significant
at the 60% confidence level. This observation was contrary to the
assumption made in the theoretical model development process that

equipment power cycling had an effect on observed nonoperating failure
rate. Therefore, this observation was further studied. Given the size of
the merged data set (i.e. 33 records), it was concluded that only the most
significant variables could be identified through data analysis. In this
case, temperature, technology, quality and complexity were identified as
significant variables. Thus, results of the regression analysis were not
j necessarily that equipment power cycling had no effect on nonoperating
-•. failure rate, but that the effect was less evident than that of
temperature, technology, quality and complexity over the range of values
.•. found in the data base.
The most obvious treatment of equipment power cycling was to exclude
A. the variable from digital microcircuit nonoperating failure rate model

development because of the apparent lack of correlation with failure rate.
This probably would have been the proposed approach if not for the
U observation that equipment power cycling was determined to be a
significant model parameter for linear/interface microcircuits (discussed
in Section 5.1.3). The observed nonoperating failure rate difference
resulting from power cycling was found to be 2.2 to 1 for linear devices.
o > This difference was relatively small, and therefore it was not surprising
that the effect could not be estimated for digital devices. A comparison 1-
was then made between the anticipated effects of power cycling for digital
"vs. linear devices.
5-28
'V.
One possible reason for the apparent difference between equipment

power cycling effects of digital and linear devices was that a major _
failure mode resulting from power cycling is wire bond fatigue and
failure. The rate at which wire bond fatigue occurs is related to the
power dissipation and corresponding thermal/mechanical stresses which
occur during the power on-off cycle. It was hypothesized that, on
average, linear devices exhibit slightly higher levels of power
dissipation than the primarily SSI/MSI digital devices included in the
data base. This would possibly explain the iQpparently higher degree of
dependence on equipment power cycling for linear devices. However, this
can not be generalized for all linear and digital devices. Many linear
, N ~ devices have power dissipation equal to or lower than those typically I
encountered in digital devices.
It was determined from this comparison of power cycling effects forI

digital and linear devices that a proposed digital microcircuit model
required an equipment power cycling factor. This determination was made
because (1) the comparison did not reveal substantial reasons why power
cycling would not effect digital nonoperating failure rate given that
power cycling was observed to effect linear device failure rate, and *(2)
the series of microcircuit models should be consistent with regard to
model parameters.
The assumed equipment power cycling factor was determined to be,
ircyc z1 + .02(power cycles/103 hours)
The (.02) constant term was based on two premises. First, the power
cycling constant term would be less than or
equal to the corresponding i
Ptermfound vinoted availabe mircruthan
data The oserved prelaios hipasforth
power cycl-ng factor would have less effect (over the range of values -ij
temperature, quality and complexity. The (.02) term represents the

largest numerical value which satisfies the two constraints. It is noted
5-29 Q
'2N %;
that this proposed power cycling factor will assume a value of one for
essentially all missile storage test schedules.
The next phase of the model development process was to investigate the
effects of the environment. The theoretical model developed for digital
microcircuits' included a nonoperating environmental factor. As previously
stated, there were insufficient data to develop a complete series of
nonoperating environmental factors empirically. Nevertheless, the
objective of this study effort was to develop a methodology which can be
employed for nonoperating reliability evaluation for any potential
application. Therefore, it was imperative that appropriate nonoperating
environmental factors were determined.
The methods presented in Section 4.5 were applied to develop

appropriate nonoperating environmental factors. This method assumes that
a series of nonoperating environmental factors can be generated from the
MIL-HDBK-217D operating environmental factors based on a comparison of
operating and nonoperating failure mechanism accelerating factors.
Additionally, the differences between average temperature for operating
and nonoperating states were investigated. Operating temperatures are
higher because of the internal heat generation associated with applying
power to the equipment. For microcircuits, the effect of this temperature
difference was predicted by the respective operating and proposed
nonoperating temperature factors. Thus, no temperature adjustment was
required for environmental factor development purposes..I
The proposed digital microcircuit nonoperating environmental factors
were presented in Table 5.1.1-1 in the previous section. Separate

nonoperating environmental factors were proposed for hermetic and
nonhermetic devices because nonhermetic devices have been observed to be
more sensitive to environmental stress, primarily humidity and
contaminated environments. An environmental factor conversion was
required because the operating environmental factor is part of a complex
nonlinear model form as opposed to the proposed multiplicative model for
nonoperating failure rate prediction. The conversion factor expression
5-30
for microcircuits was presented in Section 4.5. The conversion factor.
constants used to compensate for the difference in model forms were based
on average MIL-HDBK-217D microcircuit values. For nonhermetic devices,
the conversion factor constants were based on average factor values
complemented by the results of the Panama nonhermetic microcircuit storage
life testing (data entries 17 and 18 in Table 5.1.2-2). In addition, the
nonoperating environmental factors were normalized to a ground benign
value equal to one. This was done so that the proposed microcircuit
nonoperating environmental factors would be consistent with the other
proposed models, and to provide the proposed model with increased utility.
Conversely, the MIL-HDBK-217D, Notice 1 operating microcircuit
environmental factors are not normalized to any environment (i.e. noI
factor is equal to one), and therefore the numerical values for the base
failure rate constants have no physical meaning by themselves.
Tables 5.1.2-8 through 5.1.2-10 present failure mechanism and failure

acceleration factor distributions for MOS SSI/MSI, bipolar SSI/MSI and ij
random logic LSI digital devices. References 31 through 40 included
microcircuit failure mode or failure mechanism distribution information.
Accurate quantitative failure mechanism information was difficult to
obtain for nonoperating conditions and often c~nflicting for operating
conditions. Therefore, the information presented in Tables 5.1.2-8
through 5.1.2-10 represent typical failure mechanism distributions which
may not be similar for all specific device styles and applications.
The issue of device maturity was also addressed in this study effort.
All digital microcircuits which were represented in the collected data
were mature devices. Thus, the effect of device maturity could not be
evaluated by data analysis. Additionally, it is strongly recommended that
any equipment subjected to prolonged periods of storage be designed with _-_
only mature part types. Therefore, the proposed digital microcircuit

nonoperating failure rate prediction model does not apply to any of the
following conditions.
1) New device in initial production.
5-31
CL
oco
Cie)
cz
Z- 4
I-I
-4 C
-4
03
CO 0 O~nU CJ Icm
(J
CL 4J 4
cz 0141)..
-Jý
-4 0;T4->
M4 (0 4J4)O oa
0 5- = 0
65. 0 0 0.
to4J )40.
UU
0) 4- 0 C ~ 4-r_ l>
4
-0
0 c
'D4 -M4J t a)-9
01 0) C44)4
E 03 .- C CO
5-32
J(%J.4~fl %IC-J
0 4J Il l
O Ll
atlL IAl~l
C3 Aft
co
04)
uiii
U-4
tm C
- ~~4J 10L
tn4~ E
4) a- * C. '-I
C3 (V + 4J
= -.--
03 c s4- +3 3 4 j
030 ~u0 30
O iE
>3 >O 4 V3
0J(A
LUL
0 ) E00
+1.
*-* 5-33
~~V~W'~~'( 0) 4J 4.J
r-;v~v 4JWWW.~ .
u. ~
4J~ 00 L Ln
00000- ~o r LinO 004DL
0 -
CO LA OLA
C LA OLn
V) 4J 4-i
(Ln
m
I1
lA
1
~rC'.. II .- 4 lA lA
m ma
L.0 0- CDo.J- )0 n
CLS
L.S
-4J
>-4
m > 00
m- 4-3
< m . .4..)
(c) 0V S-41 441
U) CM= 1c
0 E CC0
CIQ > U .C 4-)( (A>
U~U
O~- 4.14) 0 00
UM- 0 . - .---
(V r C-ý
4~) * * 4J I
4- r-J0- ft oo
(D 0-L N~
- 4-S- .4C to )L#IU

-a 3 )0 b
W )LA)xu ) 4 F
U-.miC
5-4 '
, -
2) Where major changes in design or process have occurred. Is!. **,
3) Where there has been an extended interruption in production or a

change in line personnel (radical expansion).
The final phase of the model development process was to normalize the
base failure rate constant to correspond to a ground benign environment.

All observed field data were for a ground fixed environment. Therefore,
the base failure rate constant of 0.00113 was divided by the ground fixed
hermetic environmental factor of 2.4 to obtain a normalized base failure
;.*
rate. Additionally, a relatively small numerical adjustment was required

because of the assumed factor for equipment power on-off cycling. The L
base failure rate constant needed to correspond to a cycling rate of zero.
Therefcre, the base failure rate was multiplied by an adjustment factor of
0.80. This term was equal to the inverse of the average equipment power
cycling factor for the cycling rate values found in the data base. The
nonoperating base failure rate was therefore equal to .00038 after
normalizing for environment and equipment power cycling. A corresponding
change was also made for the approximate, constant base failure rate value
for random logic microcircuits with greater than 3,100 gates.
A modification to the preliminary temperature factor was then required

to prevent situations where the predicted nonoperating failure rate
exceeds the MIL-HDBK-217D operating failure rate. The first phase of the
model validation task (described in Section 5.1.5 for monolithic
microcircuits) was to compare the proposed nonoperating models with MIL-
HDBK-217D models. It was observed that in a very small percentage (<2%)
of the failure rate comparisons, the preliminary nonoperating failure rate f.
exceeded the corresponding operating failure rate prediction. . -
In each of the instances where the nonoperating failure rate was

higher than the MIL-HDBK-217D failure rate, the part type in question had'
relatively low power dissipation and junction-to-case thermal resistence,
and was exposed to a high ambient temperature. In other words, the
difference between operating and nonoperating junction temperature was
small, and temperature was the dominant failure causal stress. In these
5-35
4A.
cases, the "true" operating failure rate will be slightly higher than the
"true" nonoperating failure rate, and given an infinite data supply, this
would have been the result of the operating to nonoperating failure rate
model comparisons. However, both the MIL-HDBK-217D models and the
proposed nonoperating failure rate prediction models were based on data
samples obtainzd from different data sources. The predicted failure rate
is therefore a random variable distributed about an unknown "true" mean LAW
failure rate. As the difference between the "true" operating and
nonoperating failure rates approach zero, the probability of the predicted
nonoperating failure rate exceeding the predicted operating failure rate
approaches 50%. Thus, the small percentage of instances where the
preliminary nonoperating predicted failure rate exceeds the operating
failure rate is a natural phenomenom, and is in no manner indicative of an
incorrect analysis or invalid data. It must always be remembered that
predicted failure rates are average failure rate values computed from a
data sample of parts with similar characteristics. Predicted failure
rates are not an inherent property of the part such as capacitance, device
dimensions or number of gates.
Despite the conclusions of the previous paragraph, it is important

that the proposed nonoperating failure rate prediction models are
completely consistent with the operating failure rate prediction models.
Therefore, a modified nonoperating temperature relationship was proposed
and is given by the following equation.
"nNT = K3 + K4 exp(-An(T - 19
where
'NT nonoperating temperature factor
K3 , K4 constants (given in Table 5.1.2-11)
An temperature coefficient (given in Table 5.1.2-7) * I,
5-36 . .
L-4-
LV'
The K3 and K4 constants were determined by (1) assuming that the

operating and nonoperating failure rates are asymptotic as temperature
increases, and (2) defining the temperature factor to be equal to one at
250C. The temperature factor modification process resulted in a factor of
1.29 difference between the regression solution failure rate and the
modified temperature factor solution at 250C. This difference was because
, ., ..
the average temperature for the ground storage data was 200C, and thus, L.:
both models predict an identical failure rate at 200C. The difference
between the two failure rate models increases with temperature and is .
equal to 1.29 at 250. The previously determined nonoperating base failure
rates then had to be divided by 1.29 to be compatible with the modified
temperature factor. The modified nonoperating base failure rate constant
was equal to .00029 and the VLSI nonoperating base failure rate was equal
to .014 failures/10 6 hours.'
TABLE 5.1.2-11: K3 and K4 TEMPERATURE FACTOR CONSTANTS FOR DIGITAL

MICROCIRCUITS
Technology K3 K4
TTL, HTTL, DTL, ECL .91 .09
LTTL, STTL .90 .10 ,
LSTTL .89 .11
IIL .86 .14
MNOS .61 .39
PMOS .68 .32
NMOS, CCD .65 .35
CMOS, CMOS/SOS .58 .42
Normalization of the base failure rate and modification of the

temperature concluded the model development process for digital
factor
microcircuits. The proposed nonoperating failure rate prediction model
was found to be a function of gate count, technology, temperature,
screening, hermeticity, equipment power on-off cycling and application
environment. The model is presented in Section 5.1.1 and in Appendix A in
a form compatible with MIL-HDBK-217D. Monolithic microcircuit model
5-37
validation was performed using data from AFCIQ and is described in Section A.
5.1.5.
5.1.3 Linear/Interface Microcircuits
A nonoperating failure rate modeling approach was also successfully

ipplied for linear/interface microcircuits. The approach taken was
simiar to the digital microcircuit model development. A theoretical
nonoperating failure rate prediction model was hypothesized, and then ,
evaluated and quantified with the available data. The resultant

nonoperating failure rate prediction model was found to be a function of
complexity, quality, temperature, environment and equipment power on-off
cycling frequency. The model is presented in Section 5.1.1 and in
Appendix A in a format compatible with MIL-HDBK-21?D.
The initial step in the model development process was to identify

application and construction variables which characterize linear and
interface microcircuits in a nonoperating environment. These variables
represent possible failure rate model paramneters and are presented in
Table 5.1.3-1.
The theoretical model for linear/interface devices was essentially the L

same as the digital microcircuit theoretical model. It was assumed that
the model would be multiplicative, but nonlinear due to the equipment
power cycling factor. Additionally, the theoretical model included
V
complexity, technology, hermeticity, screening, temperature, environment U-
and equipment power cycling as model parameters.'
The summarized linear/interface microcircuit nonoperating reliability

data is presented in Table 5.1.3-2. The collected data consists of 76 A .
observed failures, 3605.8 x 106 nonoperating part hours and 124 individual
data records. Data was available for parts screened to S, B, B-i, B-2, D
and D-1 quality levels.
5-38
""E 5.1.3-1: LINEAR/INTERFACE MICROCIRCUIT CHARACTERILATION VARIABLES
iear
interface
II. Technology
A. DTL E. ECL I. IIL M. CCD
B. TTL F. LTTL J. MNOS N. CMOS
C. HTTL G. STTL K. PMOS 0. CMOS/SOS
D. DTL H. LSTTL L. NMOS P. HMOS
III. Number of Transistors
IV. Construction
A. Dip E. Chip Carrier
B. Can F. Quad In-Line (staggered leads)
C. Flatpack G. In-Line
D. Square
V. Enclosure
A. Hermetic
B. Non-Hermetic
VI. Package Material
A. Metal E. Glass I. Metal/Epoxy
B. Ceramic F. Plastic/Ceramic 3. Silicon
C. Metal/Ceramic G. Epoxy K. Phenolic
D. Metal/Glass H. Ceramic/Plastic/Window
VII. Number of Pins
VIII. Number of Interconnects
IX. Die Bond
A. Eutectic
B. Epoxy
C. Glass
X. Quality Level
A. S C. B-1 E. C G. D
B. B 0. B-2 F. C-1 H. D-1
XI. Application Environment
XII. Temperature
A. Rated
B. Actual
XIII. Number of Power On/Off Cycles per 103 Nonoperating Hours
5-39
TABLE 5.1.3-2: LINEAR/INTERFACE MICROCIRCUIT NONOPERATING

FAILURE RATE DATA
Data Complexity Part Hours

No. Equipment/Source Records (# transistors) Failures (x 106)
1 Hawk/MICOM 3 1-30 3 1,123.0
2 Maverick/MICOM 4 1-30 6 885.4
3 F-16 HUD/RIW 8 1-30 5 285.5
4 F-16 HUD/RIW 11 31-100 15 519.0
5 F-16 HUD/RIW 1 >100 2 13.0
6 Martin Marietta 2 1-30 3 771.0
7 PRC 3 1-30 0 0.6
8 RAC (1) 66 1-30 6 4.7
9 RAC (1) 20 31-100 32 1.6
10 RAC (1) 2 >100 2 0.1
11 RAC (1) 4 (2) 2 1.9
Totals 124 76 3,605.8
Notes 1) High temperature storage life test data. Storage temperatures

range from 150oC to 200 0C
2) Unknown
The collected nonoperating failure rate data for linear and. interface
devices was initially subjected to a preliminary analysis. Weaknesses in
the data base were identified, and correlation coefficients were
computed
for each pair of independent variables. Additionally, zero failure data
were analyzed to determire which data entries had sufficient part
hours to
estimate an upper bound on nonoperating failure rate. Results from
these
initial analyses are the following:
1) All collected data were from a ground based environment or from

storage life testing.
2) All data were for bipolar linear/interface devices.
5-40
Z.:
3) 73% of the high temperature data records and 98% of the high
temperature storage failures were for 0 or D-1 quality level
parts.
4) 100% of the ground field data were for S or B quality level parts
5) The correlation coefficient for number of transistors vs. number
of pins was equal to 0.70.
The distribution of quality levels presented a problem. The higher

quality devices were concentrated in the lower temperature applications.
Similarly, the lower quality devices were concentrated in the high
temperature applications. Therefore, the effects of temperature and
quality could not both be determined empirically. After consideration of
several options, it was decided to assume that the nonoperating quality
factcrs developed for digital microcircuits would also apply for
linear/interface microcircuits. This assumption was based on the premise
that screening impacts all monolithic microcircuits similarly. Thus, the
nonoperating quality factors presented in Table 5.1.2-6 of the previous
V,' section were used for nonoperating failure rate prediction of
linear/interface devices. To remove the effect of quality level from the
data, the number of part hours were multiplied by the respective
nonoperating quality factor to normalize the data. After normalizing the
data for quality level, the effect of temperature could then be evaluated.
It had been desired to analyze the effect that both number of pins and
number of transistors had on nonoperating failure rate. However, the
apparent correlation between these two variables prevented evaluation of
the effect of both variables by use of regression analysis. This observed
"correlation between number of transistors and number of pins was hardly
unexpected. As a result of the high degree of correlation, it was
concluded that only one measure of device complexity would be included in
the model. The natural logarithm of number of transistors was chosen to
be included in the regression and was designated C1.
It should also be noted that

if two variables are correlated
inherently (as opposed to coincidentially), then it would not be desirable
5-41
A.X2-7-
H7
to determine separate factors. A model with separate factors for
transistor count and pin count would be correct physically, but probably t
less accurate. At least one of the factors would have to be based on an
assumption. Assumptions always introduce some degree of error.
Determination of the other factor would be adversely effected by an
inaccurate assumption, and thus, both factors would be incorrect.
Conversely, an empirical relationship versus either one of the inherently
correlated variables represents the actual failure experience of the
device, and would be preferable.
The theoretical model for linear/interface devices included a factor

for equipment power on-off cycling. The multiplicative factor was assumed
to be of the following form.
= 1 + K(cycles/10 3 hours) SFactor
where K is a constant. A model with a factor of this form is not linear

and regression analysis could not be directly applied. This potential
problem and a proposed solution is described in Section 4.3. The proposed
solution involved performing iterative regression analyses. As
recommended in Section 4.3, three qualitative "dummy" variables were
defined (cycl, cyc2 and cyc3) to represent four distinct equipment power
cycling frequency categories, which are,
o cycling r3tes less than one power cycle every two years (cycl,
cyc2, cyc3 = 0,0,0). .
o cycling rates between one cycle per year and one cycle every two
years (cycl, cyc2, cyc3 = 1,0,0).
o cycling, rates greater than one cycle per year (cycl, cyc2, cyc3 =
0,1,0).
o unknown cycling rates (cycl, cyc2, cyc3 = 0,0,1).
The linear/interface nonoperating failure rate data were then merged

according to equipment, temperature, quality level and transistor count.
To maximize the number of failures and part hours per data record, data
5-42
'4
within 10 transistor intervals were merged. For example, the observed

failures and part hours were summed for all B quality linear devices in
the Maverick missile with between 21 and 30 transistors. The average
number of transistors for the category was used as the measure of
complexity in the regression analysis. There were a total of 23 merged
data records for linear/interface devices.
Regression analysis was then applied to the data. Results of the

regression analysis are given in Table 5.1.3-3. The dependent variable
was the natural logarithm of nonoperating failure rate. The number of
transistors (C1 ) and the inverse of temperature (designated I/T) were
determined to significantly effect nonoperating failure rate with 90%
confidence. The equipment power cycling variable for cycling rates
greater than one cycle per year (cyc2) was significant with 70%
confidence.
TABLE 5.1.3-3: LINEAR/INTERFACE INITIAL REGRESSION RESULTS
Standard Confidence
Variable Coefficient Error F-Ratio Limit
1/T -4027.505 555.208 52.62 0.90

C1 0.841 0.340 6.11 0.90
cyc2 0.983 0.802 1.50 0.70
bo 6.065 -- -- --

regression analysis with the dependent variable equal to the natural
logarithm of nonoperating failure rate normalized to remove the effects of
quality level. Transforming the regression solution into an equation
where nonoperating failure rate (as opposed to log of normalized
nonoperating failure rate) is the dependent variable results in the
following preliminary multiplicative model for linear/interface devices.
XI/i = 0.00058(# transistors)* 841 4028(i -I-)) rNQicycc
5-43
I -________________
where
XI/i = predicted linear/interface microcircuit nonoperating failure

rate (failures/10 6 hours)
INQ = assumed nonoperating quality factors
1rcyc
= equipment power cycling factor (preliminary)
= exp(O.983(cyc2))
= 1.0, cycling rate < .057 cycles/10 3 hour
= 1.0, .057 . cycling rate < .114 cycles/10 3 hour
= 2.67, cycling rate-.- .114 cycles/lO3 hour
C residual
The reference temperature term of 298 0 K was added to the previous

equation for convenience. Therefore, the multiplicative constant of
0.000581 corresponds to a temperature equal to the reference temperature.
It was hypothesized that the reason that the cycl and cyc3 variables
were not determined to be significant was that storage related failures
(as opposed to failures induced by the power on-off cycle) dominate the
total failure rate for cycling rates less than 0.114 cycles/lO3 hours.
Therefore, the anticipated difference in failure rate would be less than L
the inherent variability in the data and could not be detected. It was
considered encouraging that the unknown cycling frequency data entries
were not significantly different than the data in the two lower cycling
frequency categories.
these data sources, it
Although the cycling frequency was
was known that the equipments were
unknown for
energized
Li
relatively infrequently.
The next step in the model development process was to perform

iterative regression analyses to accommodate the assumed nonlinear model
form. The process began by performing a two-dimensional regression with ,ri
the cycling rate coefficients from the initial regression as the dependent
variable, and the mean cycling rate for each category as the independent
variable. The result was an expression for equipment power cycling factor
5-44
of the desired form (i.e. cycling factor = mx + b). As the iterative

process continued, the equipment power cycling expression was assumed
exact and the coefficients for all other variables were recalculated.
Then, those coefficients were assumed exact and the equipment power
cycling factor was recalculated. The iterative process continued until
the observed changes in coefficient values were negligible. For
linear/interface microcircuits, six iterations were required before the
optimal nonlinear regression solution was found. On the final iteration,
the temperature coefficient changed by 1/10th of 1% and the complexity
coefficient changed by 4/10th of 1%. The coefficients for l/T, C1 and bo
after the sixth iteration are given in Table 5.1.3-4. The optimal form of
the equipment power cycling factor expression was determined to be,
Wcyc = 1 + .030 9 (Nc)
where

hours
TABLE 5.1.3-4: LINEAR/INTERFACE FINAL REGRESSION RESULTS V
Standard Confidence 4
Variable Coefficient Error F-Ratio Limit
1/T -4748.303 448.970 111.85 0.90
C1 0.887 0.297 9.24 0.90
b 6.065 ......
As had been expected, temperature was determined to be the dominant

variable effecting failure rate. The linear/interface temperature
coefficient of -4748 is equivalent to an activation energy of 0.41eV. The
standard error statistic allowed for computation of a 90% confidence
interval around the point estimate temperature coefficient value. Upper
and lower 90% confidence interval values for linear/interface devices were
determined to be -3974 and -5523 respectively. These values are only
5-45
approximate because of the log transformation. It was decided to include

the effect of temperature into a nonoperating temperature factor. For
linear/interface devices, the temperature factor was determined to be the
following equation. As previously stated, the 298 term is a reference
temperature and was added for convenience.
'TNT = exp(-4748(• -
where
7TNT = nonoperating temperature factor
A base failure rate for linear/interface devices was defined as a

function of the number of transistors. Upper and lower 90% confidence
interval values for the base failure rate complexity coefficient of 0.887
were computed and found to be 1.40 and 0.375 respectively. As this stage
of the model development process, the base failure must be considered
preliminary because the value has yet to be normalized for environment. A
preliminary base failure rate expression was determined to be,
n 0.000576(# transistors) 8 8 7
~nb
where
Xnb preliminary base failure rate
The next phase of the model development process for linear and
interface microcircuits was to investigate the effects of the environment.

Data were only available for ground based environments, and thus, a series
of nonoperating environmental factors could not be determined empirically.
However, it was essential that an appropriate series of factors were t -S-
determined so the proposed model could be used for any potential
application with linear/interface devices.
5-46
41
Ix
Several possibilities were evaluated to determine appropriate

nonoperating environmental factors for linear and interface microcircuits.
Consideration was given to development of a unique series of factors for •
linear/interface devices using the methods described in Section 4.5 of 2,
this report. However, it was eventually decided to apply the same factors.--
developed for nonoperating failure rate prediction of digital
microcircuits (presented in Table 5.1.1-3 in Section 5.1.1). This factor

includes unique nonoperating environmental factor values for hermetic and . .
nonhermetic microcircuits. This decision was made for two reasons. •J

First, the use of one series of environmental factors for various
microcircuit types has a precedent. The operating failure rate prediction
procedure for microcircuits presented in •!IL-HDBK-217D includes a single ~
series of factors for all monolithic microcircuit3. The second reason for
using the digital nonoperating environmental factors was that the
nonoperating environmental factor development process was only
approximate. Therefore, the relatively smaller differences between ?•"'2
specific microcircuit styles could not be justified. ••
The final step in the linear/interface model development process was

to normalize the base failure rate eq'iation to correspond to a ground,--,•!:
benign environment. All observed field data were for a ground fixed•',••,
environment. Therefore, the preliminary base failure rate constant of ••
0.000576 was divided by the ground fixed hermetic environmental factor of ••.
2.4 to obtain a normalized base failure rate constant equal to 0.000240.••iii
The first phase of the model validation task (described in Section

4.1.5) resulted in the observation that the preliminary nonoperating
r
• !
failure rate model exceeded the MIL-HDBK-217D failure rate prediction a ,•••
small percentage of the time. To prevent these occurrances, a ,.:
modification was made to the nonoperating temperature factor. The
modified form of the temperature factor is given by the following •
equ at ion.
5-47 .
The 0.50 constants were based on a vigorous exercising of both operating

and nonoperating failure rate models. A small adjustment was also
required to the nonoperating base failure rate because the average
temperature for the field data was 200 C compared to the reference
temperature of 250C (i.e. 2980K). The adjustment factor of 1.16 resulted
in a nonoperating base failure rate constant equal to 0.000208
fai lures/10 6 hours.
Definition and determination of a nonoperating temperature factor,

base failure rate and nonoperating environmental factor, and normalization
of the base failure rate constant concluded the model development process
for linear/interface devices. The proposed model for linear/interface
devices is thereby given by the following equation. The residual term was
removed from the expression. -
XI/i Xnb WNT TNQ INE ltcyc
where
XI/i predicted linear/interface microcircuit nonoperating failure

rate ..
Xnb = base failure rate (failure/10 6 nonoperating hours) I..
88 7
= O.00021(Nt)"
where
= 0.50 + 0.50 exp(-4748(l - 1
T 29
INQ = nonoperating quality factor (presented in Table 5.1.1-2
in Section 5.1.1) rn
7NE = nonoperating environmental factor (presented in Table 5.1.1-3
in Section 5.1.1) -
t
'cyc = equipment power on-off cycling factor
= 1 + .031(Nc)
"5-48
determined microcircuit nonoperating

failure rate factors for quality,
environment, temperature and equipment power on-off cycling frequency.
The results of the first regression indicated that these factors could
also be applied for memory nonoperating failure rate prediction purposes. -2
These factors were then assumed to be correct and the data were subjected
to a second regression. Introduced into the second regression were
variables for memory type, technology and the number of gates. Only the
technology variable was significant with a 70% confidence. Interpretation
of the regression results were that bipolar memory devices have a
nonoperating failure rate which is 2.1 times higher than MOS memory
devices at the reference temperature of 250C.
TABLE 5.1.4-1: MEMORY DEVICE NONOPERATING FAILURE RATE DATA
Part
Data Complexity Hours6
No. Equipment/Source Records (# bits) Failures (x10 )
1 RAC (1) 7 < 256 n1

0.4
2 RAC (1) 24 256 13 1.0
3 RAC (1) 4 320 3 1.1
4 RAC (1) 2 512 0 0.2
5 RAC (1) 1 567 0 0.1
6 RAC (1) 48 1024 23 4.0
7 RAC (1) 3 2048 2 0.2
8 RAC (1) 12 4096 8 1.5 "
9 RAC (1) 11 8192 67 0.9
10 RAC (1) 4 16384 1 0.6
11 F-16 HUD/RIW 1 64 0 77.9
12 F-16 HUD/RIW 2 1024 1 337.4
13 F-16 HUD/RIW 1 4096 3 181.7
14 F-16 HUD/RIW 1 16384 0 207.6
Totals 121 122 814.6

NOTES: (1) High temperature storage life test data.
Additional regression solutions were found to further investigate the

effects of number of bits and memory type. The difference in nonoperating
failure rate between RAM and ROM memory devices was not significant with
50% confidence. When forced into the regression solution, the ratio of
RAM to ROM nonoperating failure rate was found to be 0.67. However,
5-49
where
Nc =number of equipment power on-off cycles per 103 nonoperating -R,

hours
The proposed linear/interface nonoperating failure rate prediction

model is presented in Section 5.1.1 and in Appendix A in a form compatible
with MIL-HDBK-217D,. Monolithic microcircuit model validation including
linear/interface devices is described in Section 5.1.4. L
5.1.4 Memory Device Model Development
A nonoperating failure rate prediction model was also successfully

developed for memory devices. The model development approach was similar
to the digItal and. linear device model development processes. The
proposed model was determined to be a function of device style, quality,
temperature, environment, hermeticity and screening level. The model is
presented in Section 5.1.1 and in Appendix A in a format compatible with
MIL-HDBK-217D. This ,,ection describes the model development process for
memory devices.
The theoretical model for memory devices was essentially the same as
the models for other microcircuit types. The theoretical model was
assumed to be a function of complexity, technology, hermeticity,
screening, temperature, environment and equipment power on-off cycling.
The summarized memory device nonoperating failure rate data are

presented in Table 5.1.4-1. The data collectively consists of 122
observed failures, 814.6 x 106 nonoperating part hours and 121 individual
data records. All data were either from the F-16 HUD, or from high
temperature storage life testing. The inherent data base weaknesses
identified for other microcircuit types also applied for memory ICs.
Two regression analyses were applied to the data to quantify the

theoretical model. A first regression was used to evaluate the previously
5-50
numerical values computed with low confidence are not necessarily k

meaningful. Similarly, the number of bits was not significant with 50%
confidence. When forced into regression solution, the data indicated that
nonoperating failure rate was negatively correlated with the number of .-
bits. Intuitively incorrect solutions, such as this, are not unusual when
variables are forced into a regression solution with very low confidence.
Neither memory type or number of bits was included in the proposed model
because of the low observed correlation with nonoperating failure rate.
It was considered unusual that complexity was not a significant

variable for memory devices given that the number of gates and number of
transistors were significant variables for random logic and linear
microcircuits respectively. One possible explanation for this observation
is that as complexity has historically increased, corresponding advances
have been made to optimize manufacturing technology. Thus, the expected
increase in failure rate may be negated by corresponding increases in
production technology, quality control and screening. Another possible
explanation for the observed lack of correlation was that the statistical
noise in the data prevented all but the most significant variables from
being identified. However, neither of these explanations indicate why
complexity would be significant for random logic and linear devices, but'
not memories. It was decided not to assume a complexity relationship for
several reasons. Including an assumed complexity factor would needlessly
complicate the proposed model when no observed correlation can be
detected, for whatever the reason.
The proposed model for memory devices is therefore given by the

following equation. The nonoperating base failure rates are the numerical
values determined from the data. The other factors were previously
determined for random logic microcircuits.
Xnb 7rNT fNQ TNE Icyc -5
5-51
- W.,7Pr rnr
)
where
X
Xnb
memory device nonoperating failure rate
= nonoperating base failure rate
= .0034, bipolar memory devices
V
= .0%17, MOS memory devices
9NT = nonoperating temperature factor
= exp(-An(T- 298))
where L
T = nonoperating temperature ()K)"
An = temperature coefficient (given in Table 5.1.2-ý in Section
5.1.2)
7TNQ= nonoperating quality factor (given in Table 5.1.1-1 in Section
5.1.1)
"WNE = nonoperating environmental factor (given in Table 5.1.1-2 in
Section 5.1.1)
ffcyc = equipment power on-off cycling frequency
= 1 + .02(Nc)
where

hours y "
Determination of the memory device nonoperating base failure rates was

the final phase of the model development process. The proposed model is
presented in Appendix A in a format compatible with MIL-HDBK-217D.
5.1.5 Monolithic Microcircuit Model Validation
The proposed microcircuit nonoperating failure rate prediction models

were next subjected to a thorough model validation process. The first
phase of the model validation process consisted of an in-depth comparison
5-52
of nonoperating failure rate prediction models to the M:L-HDBK-217D

models. This phase served to identify cases where the predicted
nonoperating failure rate exceeded the predicted operating failure rate.
"The second phase of the model validation task consisted of an extreme case
"analysis. Predictions were made using the proposed models for parameters
beyond the ranges found in the data. The intent of the extreme case
analysis was to identify any set of conditions which cause the predicted
nonoperating failure rates to approach infinity or predict an intuitively
incorrect failure rate. This exercise did not actually serve to validatc
"the proposed model, but was an important step in the overall model
development process and was useful to insure that the proposed models were
physically correct. The third phase of the model validation task was to
compare the predicted nonoperating failure rates with observed field data
which had been withheld from the model development process. Nonoperating
microcircuit data were collected from AFCIQ but not used because of part
characterization problems.
The first phase of the model validation task consisted of a vigorous

exercising of the models to identify cases where the predicted
nonoperating failure rate exceeds the MIL-HDBK-217D operating failure
"rate. The probability of this seemingly unlikely event approaches 50% (if
the failure rate estimates are distributed normally) as the inherent
nonoperating failure rate becomes close to the inherent operating failure
rate. This stems from the fact that failure rate is a random variable and
the predicted failure rates represent averages computed from a limited
data sample. The averages computed from different data samples would be
distributed about some unknown inherent mean failure rate. Some of the
average failure rates will be higher than the inherent failure rate and
some will lower. Therefore, as the inherent operating and nonoperating
failure rates become close, there will be close to a 50% probability that
the average failure rate computed from a sample of nonoperating failure
data will exceed the average failure rate computed from a sample of
operating failure data. An example where the inherent nonoperating
failure rate would be anticipated to be close to the inherent operating
failure rate is for highly screened devices exposed to a high ambient
5-53
..............
N..- ' .. . . . . . . ...
temperature and with a low power dissipation and IC junction-to-case

thermal resistance. In other words, there would be little difference
between operating and nonoperating junction temperature, and temperature
would be the dominant failure causal stress.
Despite the conclusions of the previous paragraph, it was believed

that nonoperating failure rate predictions which exceed the corresponding
operating failure rate
prediction (even if this occurs only a small
percentage of the time) could seriously detract from the credibility of
MIL-HDBK-217D. Many handbook users incorrectly treat MIL-HDBK-217D
failure rate predictions as an inherent property of the device. It was
determined that a modification to the model would have little effect on
model accuracy yet would increase the credibility of the models.
To identify instances where the nonoperating failure rate prediction

exceeded the MIL-HDBK-217D failure rate, a sample size of 40 microcircuits
were selected. The selected microcircuits rep-esented the full range of
available quality levels, technologies, and complexities. These same 40
microcircuits are presented in Table 7.0-1 in Section 7.0, "Comparison of
Operating and Nonoperating Failure Rates." Operating and nonoperating
failure rate predictions were made using MIL-HDBK-217D and the preliminary
nonoperating failure rate prediction model for three different ambient
temperatures (200C, 40oC, and 600C) and two environments (ground, fixed
and airborne, uninhabited fighter). Comparisons were then made for
operating to nonoperating ground fixed environments (e.g., most ground
based equipments), operating to nonoperating airborne uninhabited fighter
environment (e.g., captive carry missiles), and operating airborne
uninhabited fighter to nonoperat•,,g ground fixed environments (e.g., most
avionic equipments). A total -f 480 failure rate predictions and 360
failure rate corparisons we-e cmiputed. The nonoperating failure rate
prediction exce-d-- 2' MI
P DBK-217D ;:iaure rate prediction in seven |
instances or in 1.9% of th(- comparisois. In each of these instances, the
ambient temperature wa, V',...
64--
wy-:4g2
After this phase of the model evaluation task, a modification was made
to the nonoperating temperature factor as part cf the model development
task. The modified models have the same equivalent activation energy for
high temperatures. The only difference is that the modified temperature
factor model approaches a constant value as temperature becomes very low.
The second phase of the model validation task was an extreme case
analysis. This analysis did not indicate any deficiencies with the
proposed models. Quantitative factors for temperature, complexity and
equipment power on-off cycling were tested by this method. At extremely
high levels of equipment power cycling, the nonoperating failure rate for
each miicrocircuit model becomes directly proportional to the equipment
power cycling frequency. At low levels of equipment power cycling, the
proposed microcircuit nonoperating failure rates approach a constant
value. These extreme case results confirm intuitive reliability
relationships. At extremely high equipment power cycling rates, the
nonoperating failure rate would be expected to be dominated by failures
induced by the power on-off cycle. Therefore, it would be anticipated
that the relationship would be proportional. At extremely low equipment
power cycling rates, the failure rate would be expected to be equal to the
inherent storage/dormant failure rate, and independent of power cycling
frequency.
The microcircuit nonoperating failure rate data collected from AFCIO

were not used in the model development process because device complexity,
equipment power cycling frequency and screening level could not be
determined. However, typical values for these parameters were assumed and
the observed nonoperating failure rates were compared to the corresponding
predicted values. Table 5.1.5-1 presents the rmodel v;*lidaticn data.
Additionally, the predicted nonoperating failure rate is included in Table
S5.1.5-1. Equivalent quality levels were provided by AFCVj. The
complexity levels were known to be less than 100 gates for digital devices
and 100 transistors for linear devices. Fifty gates and 50 transistors
were assumed to compute the predicted 'failure rates. One equipment power
* cycle per 103 nonoperating periods was also assumed. Due to the number of
5-55
assumptions, the model validation with AFCIQ data could only serve to
identify large errors.
TABLE 5.1.5-1: MONOLITHIC MICROCIRCUIT MODEL VALIDATION DATA
Part
Hourý Temp. Equiv.
No. Fail. (x10 0 ) Description (OC) Quality X.0 5 Xo X.95 Xpre
1 0 22 digital, bipolar (1) D . .-- .1361 .0115

2 0 240 digital, bipolar 25 D -- .0042 .0125 .0115
3(2) 1 2 digital, bipolar 150 D .0307 .5988 2.84 .0605
4 22 1330 digital, bipolar 5-30 D .0111 .0165 .0235 .0115
5 2 59 digital, bipolar 5-30 B .0060 .0339 .1067 .0046
6(3) 20 25 digital, bipolar 85-135 B .5302 .8000 1.16 .0579
7 2 159 linear, bipolar (1) D .0022 .0126 .0396 .0406
8 0 24 linear, bipolar 25 D -- -- .1248 .0406
9(2) 0 <1 linear, bipolar 150 D -- -- 7.30 1.08
10 3 80 linear, bipolar 5-30 D .0133 .0375 .0969 .0406
11 0 6 linear, bipolar 5-30 -.. .4992 ,S3
MV 1947
NOTES: (1) Temperature was unknown for these data entries. 250C was
assumed.
(2) Data entry was from high temperature storage life testing.
(3) Data entry was from temperature cycling test (850C to 1350C)
Environment assumed to be airborne, uninhabited fighter.
The model evaluation process indicated that the proposed models were
accurate. Although no substantial conclusion could be reached because of
the assumptions. The geometric mean of the ratio of observed to predicted
failure rate was equal to 2.1 for those data entries with observed
failures or sufficient part hours to estimate a failure rate without
observed failures (i.e. eati entries 2, 3, 4, 5, 6, 7, 10). Strict
interpretation would therefore indicate that the proposed model predicted
a failure rate which is lower than actual. However, it should also be
noted that the geometric mean of the ratio of observed to predicted
failure rate is equal to 0.97 for only the field data entries. High
temperature life test data is inherently more variable because it
necessarily reflects only the beginning of the device lifetime.
5-56
°#r
Additionally, test data generally consists of many parts on test for a

relatively short time each.
The proposed nonoperating failure rate prediction models for

monolithic microcircuits are probably the most important models presented
in this report. The proposed models provide proper discrimination against
application variables which influence failure rate. Additionally, they
are easy to apply and have been shown to be relatively accurate. The use
of the proposed monolithic microcircuit nonoperating failure rate
prediction models, as well as the other models presented in this report,
will greatly enhance nonoperating failure rate assessment capabilities.
5.1.6 Hybrid Microcircuit Nonoperating Failure Rate Model

prediction model for hybrid microcircuits. The proposed model is
in Appendix A in a form compatible with IIL-HDBK-217D. The S
proposed model is given by the following relationship.
Xp Xnb 1INQ ITNE
~~where
Xp = predicted hybrid microcircuit nonoperating failure rate

nb= nonoperating base failure rate (includes an average failure
rate contribution for capacitors, packaged resistors, substrate
resistors and the substrate) (failures/1O6 hours)
= A1exp(.45(ND) + .15(NT) + .81(NIC)), ND + NT + 1.8NIC < 12.2
= A2 exp(.033(ND) + .033(NT) + .059(NIC)), ND + NT + 1.8NIC > 12.2
7. where
ND = number of diodes
NT -= number of transistors
NIC = number of integrated circuits
5-57
A1 a .0000817
A2 = .013
iNQ = nonoperating quality factor
= 0.53, S
=1.0, B
- 8.6, D
nNE = nonoperating environmental factors (see Table 5.1.6-1)
The next section describes the nonoperating hybrid model development

process.
5.1.7 Hybrid Model Development

implemented for hybrid devices. Two assumed theoretical models were
investigated by analysis of the available hybrid nonoperating reliability
data. Both resulting failure rate prediction models were thoroughly
analyzed and an optimum model determined. The first theoretical model was
a function of the number of diodes, transistors, integrated circuits
(ICs), resistors and capacitors in the hybrid. The second theoretical
model was similar to the operating hybrid failure rate model developed by
IITRI (Reference 30) which was determined to be a function of the number
of interconnects. In addition, the effects of device screening, seal
perimeter, environment and equipment power on-off cycling frequency were
investigated as part of each theoretical model.
The first step in the model development process was to identify

application and construction variables which properly characterize hybrids
in a nonoperating environment. Table 5.1.7-1 presents a list of the part
characterization variables for hybrids. These variables are possible
failure rate model input parameters which were analyzed using statistical
methods when possible.
5-58
- Al
TABLE 5.1.6-1: HYBRID NONOPERATING ENVIRONMENT FACTORS (INE)
Environment ITNE
GB 1
GF 2.4
GM 3.5
MP 3.2
NS8 3.4
NS 3.4
NU 4.5
NH 4.6
NUU 4.9
ARW 6.3
AIC 2.4
AIT 2.7
A18 4.0
AIA 3.4
AIF 4.7
AUC 2.7
Aiyr 3.4
AUB 5.7
AUA 4.7
AUF 6.7
SF 1.3
MFF 3.3
MFA 4.3
USL 8.0
ML 9.~3
CL 150
5-59
TABLE 5.1.7-1: HYBRID PART CHARACTERIZATION
I. Functional Group
A. Digital
B. Linear
II. Number of Components
A. Packaged Resistors
B. Capacitors (packaged/chip)
C. Diodes (packaged/die)
D. Transistors (packaqed/die)
E. Microcircuits (packaged/die)
III. Number of Chip and Substrate Resistors
IV. Number of Active Interconnections -

A. Bimetal Bonds
B. Single Metal Bonds
V. Seal Perimeter (inches)
VI. Substrate Area (Square Inches)

VII. Quality Level J,
A. S
B. B Li'
C. 0
VIII. Application Environment
IX. Temperature
A. Rated
B. Actual
X. Number of Equipment Power On/Off Cycles per 103 Nonoperating Hours
5-60
>-
The first theoretical model for hybrids was determined, to be the

following equation (f denotes a function).
Ahl = f(ND, NT, NIC, NR1, NR2, Nc, S) INQ "NE Icyc
where
A1= predicted hybrid nonoperating failure rate (failures/10 6 hours)

NT = number of transistors
NR1 = number of packaged resistors
NR2 = number of chip or substrate resistors
Nc = number of capacitors
S = seal perimeter (inches)
wNQ = nonoperating quality factor, based on device screening
INE = nonoperating environmental factor
'rcyc = equipment power on-off cycling factor
= 1 + K(Nc)
where
K = constant
Nc = equipment power on-off cycling rate (cycles/lO3 nonop. hours)
The second theoretical model was based on the premise that the total
number of interconnections sufficient y characterizes hybrid device
complexity. The number of "nterconnections is typically equal to one for
each diode and external lead, equal to two tor each transistor, capacitor
and chip resistor and equal to the number of chip bonding pads for
integrated circuits. The second theoretical model was the following
equation.
2= f(NI,S) INQ 'NE 71cyc
5-61
where
NI = number of interconnects
S = seal perimeter (inches)
Presentation of these theoretical models does not imply that each of

the variables is significant. Statistical methods were applied as a basis
for establishing the significance of variables. The theoretical model was
intended to define the relationship of the independent variables to each
other.
The summarized hybrid nonoperating reliability data collected in

support of this study is presented in Tables 5.1.7-2. The data consists
of 27 individual data records, 2082 observed failures with 50,049.89 X 106
part hours. ,7-
TABLE 5.1.7-2: SUMMtARIZED HYBRID NONOPERATING RELIABILITY DATA
Equipment/Source
Maverick/MICOM
Hawk/MICOM
# Data
Records
17
7
# Failures
1,969
109
Part Hours
(X 106)
22,529.73
27,416.36
F ,
F-16 HUD/RIW 3 4 103.80

Totals 27 2,082 50,049.89
Point estimate failure rates were computed for data records with
observed failures. The methods described in Section 3.3 were applied to
determine which zero failure data records had sufficient part hours to be
included in the analysis. All twenty seven data records had either
observed failures or sufficient part hours to estimate a failure rate
without observed failures.
The number of variables which could be analyzed empirically was

limited by the nature of the available data. The following attributes of
5-62
Z. .w
the collected hybrid data were identified during a preliminary data

analysis:
1) There was an insufficient range of environments represented in the

data to develop a series of environmental factors based only on
the data.
2) There was an insufficient range of quality levels to develop a.
quality factor series based upon the data.
3) There was an insufficient range of ambient temperature to study
the effects of temperature analytically.
4) All data were for digital hybrids.
•) There was a high degree of correlation between many of the
"independent" variables.
One major problem identified in the available data was a high level of
correlation between a large number of the potential model variables. The
following pairs of variables had a correlation coefficient greater than
0.40:
0 interconnects and number of rsitod s

o interconnects and number of risors
0 interconnects and number of ICsL
o interconnects and cycling rate
number of resistors and number of diodes
o number of resistors and cycling rate
o number of capacitors and number of transistors
o number of diodes and seal perimeter
o number of transistors number of ICs
O number of transistors and seal perimeter
o number of transistors and cycling rate r
o number of ICs and cycling rate
This many correlated variables made proper application of regression

analysis impossible. A correlation coefficient matrix is presented in
Table 5.1.7-3. A discussion of correlation coefficients is presented in
Section 3.2.
5-63
Yrt
TABLE 5.1.7-3: HYBRID VARIABLE CORRELATION COEFFICIENT MATRIX y
Variables
IN R C D T IC S CYC
IN 1.0
R .67 1.0
C -.21 -.15 1.0
D .41 .75 -.25 1.0
T -.33 -.33 .45 .04 1.0
IC .69 .33 -.10 -.18 -.64 1.0
S .15 .21 .06 .42 .41 -.17 1.0
CYC .80 .57 -.26 .13 -.75 .78 -.25 1.0
NOTE (1): IN =# interconnects, R =# resistors, C # capacitors, D

diodes, T = # transistors, IC =# integrated circuits, S =seal
perimeter (inches), CYC = equipment power cycling rate
(cycles/b 3 hrs.)
The correlation coefficients for hybrids were much larger than those
for any other part type in this study. Probable reasons for the large
coefficients were the small number of data sources available with hybrid
nonoperating failure rate data, and the small number of hybrids typically
used in an equipment. Also, many of the correlations were expected
because as the hybrid becomes more complex' the size and the number of
components must increase. As a result of the high correlation, multiple
linear regression analysis could not be correctly applied to the data toh
simultaneously investigate the effects of all variables. This observation
was one of the reasons to attempt two different modeling approaches. The
high degree of correlation between interconnects and resistors, diodes,
and ICs prevented a numerical analysis of both the number of interconnects L
and the number of components. However, if the number of interconnects

sufficiently characterizes hybrid complexity by itself, then a two-
dimensional regression (failure rate vs. number of interconnects) could
be correctly applied to develop a hybrid nonoperating failure rate
prediction model. This is not to say that only interconnects fail in a
hybrid device. If the number of interconnects increase simultaneously
5-64
with the number of components, then a regression against either one of

these parameters will include the effects of both.
should be noted that the observed correlation between number of

It
interconnects and number of active components was hardly unexpected. in
fact, these two parameters must be correlated because of the nature of a
hybrid design. Therefore, it would be undesirable to include both number L.
of interconnects and number of components in a proposed prediction model, I-
even thourh both factors intuitively have an effect on nonoperating ,o.•j
failure rate. Quantifying two correlated variables separately must always

rely on assumptions, thus introducing error. L.
The large correlation between equipment power on-off cycling frequency

and interconnects, resistors, transistors, and ICs must be attributed to
coincidence and the result of a relatively small data base. Therefore,
the effects of equipment power on-off cycling could not be quantified with
the avaiiable data. This was unfortunate because equipment power on-off
cycling potentially has a significant effect on hybrid nonoperating
failure rate. There was a wide range of power cycling values represented
in the collected data. Equipment power cycling frequency values ranged
from 0.038 cycles/l 3 nonoperating hours (i.e. approximately one power
cycle every three years) for the Hawk missile to 40.12 cycles/103
nonoperating hours (i.e. approximately one power cycle every 25 hours) for
the F-16 HUD. Therefore, the resultant hybrid microcircuit nonoperating
failure rate prediction model corresponds to some unknown, average cycling
rate.
Although no temperature relationship could be derived from the data,

it was determined that temperature dependence hypothetically should be a
factor in the failure rate prediction model. However, no temperature
factor was proposed due to (1) the limited range of temperatures found in
the data, (2) the fact that hybrids are made of a composite of devices
each with different temperature characteristics, and (3) the large variety
of hybrid designs precludes the assumption of a temperature factor.
5-65
Therefore, the proposed model is based on a constant average nonoperating

temperature.
The preliminary model refinement process resulted in the conclusion

that both theoretical models had merit. The variables analyzed for the
first model were number of capacitors, number of resistors, number of
diodes, number of transistors, number of ICs, seal perimeter, quality
level, and environment. The variables evalLated versus nonoperating
failure rate for the second model were number of interconnects, quality
level, and environment. These variables, with the exception of quality
and environment, were analyzed using statistical techniques. Both
theoretical models were explored by application of regression analysis.
The negative effect of correlated variables was minimized by carefully
choosing independent variables.
For the first model, the dependent variables wera number of resistors
(NR2), number of capacitors (NC), number of diodes (N0 ), number of
transistors (NT),
number of microcircuits (NIC), and package seal
perimeter (S). The results of this regression indicated that nonoperating ,"
failure rate was negatively correlated to NC and S, as well as positively,.
correlated with NT, ND and NIC. These intuitively incorrect results were
mcst likely the result of the scarcity of data, the intercorrelation
between variables, and/or the large variability observed in the data.
Next, the regression was run with these variables omitted. The results of
this regression are presented in Table 5.1.7-4. NT, ND and NIC were
significant at a 70% confidence limit. A properly applied regression
analysis requires that there is no correlation between independent
variables. Therefore, the regression results must be considered
approximate. The R-squared value for this regression was .44. This value
was considered to be relatively low, but not unexpected due to the
variability in the data.
In the second model, the dependent variable was the number of

interconnects (1). The results of the second regression analysis are
given in Table 5.1.7-5. Number of interconnects was significant at a 90%
5-66
confidence level. The R-squared value for this regression was .31. In
other words, only 31% of the variability in the data could be explained by
the regression solution. This relatively low value can be explained by
the fact that there are many variables effecting hybrid nonoperating
reliability, including design and processing variables which can not be.
quantified. An extremely accurate model was not considered feasible
because of the problems associated with hybrid nonoperating failure
modeling.
Interpretation of these two initial regression solutions are:
1) A nonoperating hybrid failure rate prediction model as a function

of the number of interconnects can be supported with the data.
2) The number of transistors, diodes and microcircuits were
significant at a 70% confidence limit for the first model.
3) The high level of intercorrelation between variables, the large
variability in observed data, the scarcity of data, and the many
variables involved make a highly accurate model unfeasible.
4) Although number of resistors, number of capacitors and seal
perimeter were not significant factors in either analysis, this
"does not mean they do not have an effect on hybrid microcircuit
nonoperating failure rates. Rather, the effect cannot be detected
with the available data over the range of values found in the data
base. In addition, since these devices have inherently lower
failure rates than discrete semiconductors and microcircuits, the
results of the regression were encouraging.
The first model does not include capacitors, packaged resistors,

substrate resistors and the substrate implicitly in the proposed model.
However, the model was based on observed data for hybrids designed with
these components, and therefore, the base failure rate constant must
"include an average failure rate contribution for capacitors, packaged
M. resistors, substrate resistors and the substrate itself. Thus, the effect
of these factors was not ignored although they are not included
specifically as part of the model.
5-67
-N
'The fiirs t ode% 4r-`1fd*S a count of integrated circuits

without
considering tht ýoraplexity (f individual IC chips.
S
A model factor based
on specific C chip complexties would have been impossible
to quantify
from the avoilable data.
TABLE 5.1.7-4: HYBRID REGRESSION RESULTS (1)
Standard Confidence
Variable Coefficient Error F-ratio Limit
NO .6157 .1717 12.86 .70

NT .2843 .172" 2.73 .70
NIC .8115 .2696 9.06 .70
bo -8.5364 -- --.
TABLE 5.1.7-5: HYBRID REGRESSION RESULTS (II)
Standard Confidence
Variable Coefficient Error F-ratio Limit
NI 2.1114 .6269 11.34 .90

bo-11.9330 -- -- --
The regression analyses were performed with the dependent

variable
Sequal to the natural logarithm of nonoperating failure rate.
Transforming
the regression solutions into equations where the dependent
variable is
simply
models: the nonoperating failure rate results in the following preliminary
(1) )hl = Alexp(al(ND) + t2(NT) + a3(NIC))
5-68
where
Xhl = predicted hybrid nonoperating fdilure rate (failure/lO6 hours)

A1 = constant (includes average failure rate contributiuns for
capacitors, packaged resistors, substrate resistors and
substrate)
= .000196
ot1 = constant
= .62
= constant
= .28
a3 = constant
= .81
(II) Xh2 = A2 (NI)a4
where
Xh2 = predicted hybrid nonoperating failure rate (failures/10 6 hours)

A2 = constant
= .00000657
NI = number of interconnections
az4 = constant
= 2.11
These models are preliminary and only applicable to a ground fixed

environment.
Both preliminary models had merit and were easy to apply. However,
neither preliminary model was particularly accurate. This was not a
result of an incorrect or oversimplified analysis, but was due to the
5-69
I
large variability associated with hybrid device failure behavior. Hybrid
devices tend to be manufactured in lower quantities, and the design and
processing controls are less uniform than with monolithic microcircuits.
This observation and ot1`-r factors result in the high variability. The
decision was made to proceed with the first nonoperating failure rate
• prediction model, as a function of number of microcircuits, number of
transistors and number of diodes. This decision was made because of the
better fit to the data of the first regression solution.
The standard error statistic allows for computation of confidence

intervals around regression solution coefficients. A brief discussion
concerning confidence intervals is included in Section 3.2. Table 5.1.7-6
presents upper and lower 90% confidence interval values of the
coefficients for ND, NT and NIC. The cr~fficient for NIC was considered
to be intuitively correct. Interpretation of the coefficient was that the
number of integrated circuits would have a greater influence on
Snonoperating failure rate than either the number of transistors or
diodes. Conversely, the coefficients for diodes (MD) and transistors (NT)
IA
seemed c3nflicting. Intuitively, the number transistors should have a
greater effect than the number of diodes.
regression solution was erroneous.
Initially, it was felt that the
After further investigation of the
respective coefficients and the confidence intervals (presented in Table
5.1.7-6), it was concluded that the results were not in error but that the
relatively smaller difference between the coefficient values could not be
detected with the available data. However, it wa; strongly believed that
the proposed models should be physically correct. Therefore, an identical
average value was included in the proposed model for both diodes and
transistors.
5-70
. . . . . • • - . . . , ,
V ".
TABLE 5.1.7-6: HYBRID COEFFICIENT CONFIDENCE INTERVALS
Lower Upper
Variable 90% Limit Coefficient 90% Limit
ND 0.321 0.617 0.910
NT -0.011 0.284 0.580
NIC 0.349 0.812 1.274
The next task in the model development was to determine a series of

hybrid nonoperating quality factors for S, B and 0 quality levels. No
quality factor was determined 0-1 quality hybrids. This is consistent
with MIL-HDBK-217D. The proposed series of nonoperating quality factors
was developed based upon the operating hybrid quality factors and the
relationship of the digital IC nonoperating quality factors (developed
empirically in this study) to the monolithic microelectronic device
operating values from MIL-HDBK-217D, Table 5.1.2.5-1. The derivation
process was based on two assumptions. First, it was assumed that the
ratio of S to B nonoperating quality factors was the same for hybrid
microcircuits as monolithic microcircuits, or,
(1TNQS/TNQB)monolithic ICs = !NQS/iNQB)hybrids = .53
where
'rNQS is the nonoperating S level quality factor

-rNQB is the nonoperating B level quality factor
Thus making 'rNQB = 1.0 and irNQS = 0.53 for hybrids
The second assumption was that the relative difference between the
I proposed nonoperating quality factors and the operating quality factors
were the same for hybrid microcircuits as monolithic ICs. In effect, the
previously determined factors for monolithic ICs were used as a scaling
5-71
1.
factor to determine the nonoperating quality factor for 0 level hybrids.

This relationship is given by,
(TNqD/1TNQB NQD/ NQB)

M(
wh
QD/ wQB monolithic ICs TQD/TQB hybrids
where
1TNQB is the nooioperating B-level quality factor (= 1.0 for monolithic

and hybrid)
WNQD is the nonoperating D-level quality factor (= 2.5 for monolithic)
TQB is the operating B-level quality factor (= 1.0 for monolithic
and hybrid)
wQD is the operating D-level quality factor (= 17.5 for monolithic
and 60 for hybrid)
Solving for the hybrid wNQD gives: 1NQD 8.57
The resulting series of hybrid nonoperating quality factors are

presented in Table 5.1.7-7. These numerical values are based on intuitive
relationships, and should be evaluated when more hybrid nonoperating
failure rate data becomes available. The proposed nonoperating quality
factors are based on the screening of the hybrid and not on the screening
H
of the components in the hybrid. It was assumed that a high level of
hybrid screening was indicative of an overall quality part. Conversely, a
low level of hybrid screening was believed to be indicative of an overall
lower level of quality.
Pt
-AX
TABLE 5.1.7-7: HYBRID NONOPERATING QUALITY FACTORS ri

Quality Level rNQ
s 0.53
B 1.0
D 8.6
5-72
The next step in the model development process was to determine a

series of nonoperating environmental factors for hybrid microcircuits.
After consideration of several possibilities, it was assumed that the IC
chip typically dominates the failure rate of the hybrid, and thus the
nonoperating environmental factors for digital iCs could also be applied
to hybrids. Table 5.1.6-1 in Section 5.1.6 presents a listing of the
nonoperating environmental factors for devices with both hermetic and non-
hermetic packaging.
Finally, the base failure rate constant for hybrids was normalized to
correspond to a ground benign environment. The base failure rates
determined from the data corrcspond to a ground fixed environment. The
observed nonoperating base failure rate was divided by the ground fixed
environment factor (ItNE,GF = 2.4 for hermetic devices). The normalized
hybrid nonoperating base failure rate constant was therefore determined to
be .0000817.
The proposed hybrid nonoperating failure rate prediction model was

next subjected to an extreme case analysis. Predictions were made using
the proposed model for parameter values beyond those found in the data
base. It was found that the proposed model predicted extremely high
nonoperating failure rates for quantities of active components beyond that
found in the data base. This was not necessarily indicative of an
incorrect analysis. Empirical relationships are only strictly applicable
to the range of values found in the data. Nevertheless, it was essential
that the proposed hybrid model be physically correct and not limited to
lower complexity hybrids. Therefore a second base failure rate equation
was hypothesized for complexities beyond that found in the data.
Additionally a more definitive complexity numeric was defined to be equal
to the number of diodes plus the number of transistors plus 1.8 times the
5-73
number of integrated circuits. The second base failure rate is given by

the following equation.
)tnb = .013exp(.033(ND) + .033(NT) + .059(NIC))
and is applicable when ND + NT + 1.8NIC >'12.2
The 12.2 cut-off value was equal to the highest complexity found in
the data. The second base failure rate is intuitively correct for several
reasons. First, it provides continuity with the low complexity base
failure rate. Additionally, it would be anticipated that an increase in
hybrid components for very complex hybrids would have less overall effect
on the hybrid nonoperating failure rate as a similar increase on a simple
device. Additionally, the second base failure rate iquatior still
increases with the number of diodes, number of transistors and number of
integrated circuits.
Determination of the second base failure rate equation concluded the

model development for hybrid microcircuits. Despite several difficulties,
a viable approach was developed for hybrid model development and a
proposed model was determined as a function of the number of integrated
circuits, number of transistors, number of diodes, quality level and
environment.
5.1.8 Proposed Magnetic Bubble Memory Nonoperating Failure Rate

Prediction Model
The proposed model for magnetic bubble memory devices is represented

by the following equation. The proposed model is presented in Appendix A
in a form compatible with MIL-HDBK-217D.
p= (nbl lTNT1 + Xnb2 7TNT2) iNE
5-74
/ W
/LM
where
Xp = magnetic bubble memory predicted nonoperating failure rate

XnbI = control structure nonoperating base failure rate (failures/
106 nonoperating hours)
477
= .0015(Ng)"
where
Ng = number of gates
= number of transfer gates plus number of dissipative control
gates plus number of major loops
INT1 = control structure nonoperating temperature factor
exp(-6159( 1
=exp-uJ.J~T 28)
F-
where
T = temperature (OK) = 273 + T (OC)

Xnb2 = magnetic memory structure nonoperating base failure rate
(failures/10 6 ncnoperating hours)
= .0089(NL)
where
NL = number of loops
= number of major loops plus number of functional minor loops
ITNT2 = magnetic memory structure nonoperating temperature factor

exp(.An(• 296 -
where
An = temperature coefficient (see Table 5.1.2-7 in Section 5.1.2) •:

T = temperature (OK) = 273 + T (OC)
'NE = nonoperating environmental factors (see Table 5.1.1-2 in
Section 5.1.1)
5-75
5.1.9 Magnetic Bubble Memory Model Development
The magnetic bubble memory device is a hybrid assembly of two major

structural elements. The first segment is a basic memory and control
structure consisting of thin-film elements on a crystalline substrate.
The second major structural segment is a magnetic structure to provide a
controlled magnetic field consisting of a magnet, magnetic coils and a
housing. These two major structural segments of the hybrid are
interconnected by a mechanical substrate and lead frame. The interconnect
substrate is normally a printed circuit board in present technology.
No quantitative nonoperating failure rate data were available for

magnetic bubble memory devices. Therefore, a hypothetical model was
developed based on the monolithic and hybrid microcircuit nonoperating
failure rate prediction model development, and the MIL-HDBK-2170 operating
failure rate prediction model for magnetic bubble memories. The model was
simplified because the anticipated precision of the proposed model did not
warrant a ve.ry complex equation. The hypothetical model is represented by
the following expression.
Xbm (Abl(Ng)"7 NT1 + Ab2(NL) TNT2) iTNE -
where
'bm =magnetic bubble memory predicted nonoperating failure rate

(failures/10 6 nonoperating hours)
Abl control structure nonoperating base failure rate constant
Ng number of gates
number of transfer gates plus number of dissipative control
gates plus number of major loops
•NT1 control structure nonoperating temperature factor __
exp(-6159(i - 1 -
5-76
4...
.A
where
T = ambient nonoperating temperature (OK)

Ab2 = magnetic memory structure nonoperating base failure rate
constant
NL = number of loops
= number of major loops plus number of functional minor loops
wNT2 = magnetic memory structure nonoperating temperature factor
= exp(-An(T - M-))
where
An = temperature coefficient
T = ambient nonoperating temperature (OK)
wNE = nonoperating environmental factor
The first term in the nonoperating failure rate equation represents

the contribution of the control structure. The temperature coefficient
was the p'reviously determined value for NMOS technology. The complexity
exponent was the previously determined value for random logic digital
microcircuits. The second term in the equation represents the
nonoperating failure rate contribution of the magnetic memory structure.
It was assumed that the temperature coefficients and nonoperating Pr
environmental factors for monolithic microcircuits were applicable for
magnetic bubble memory devices. This assumption was necessary because of
the lack of nonoperating failure rate data.
The remaining factors to be quantified were the control structure K%'

nonoperating base failure rate constant and the magnetic memory structure
nonoperating base failure rate. Estimates for these values were based on
the ratio of operating to nonoperating failure rate for monolithic
microcircuits, and the MIL-HDBK-217D magnetic bubble memory operating
5-77
failure rate prediction model. Estimates for Abi and Ab2 are the
following.
Abl = control structure nonoperating base failure rate constant

- 0.00153
Ab2 = magnetic memory structure nonoperating base failure rate K.
= 0.00885
Determination of the two nonoperating base failure rate constants

concluded the model development process for magnetic bubble memories. The
proposed model was based entirely on assumptions and must be considered
approximate. It was essential that a nonoperating failure rate model for
magnetic bubble memories devices was proposed because of the requirement
for a comprehensive nonoperating failure rate prediction methodology.
V8
5-78
5.2 Discrete Semiconductors
5.2.1 Discrete Semiconductor Nonoperating Failure Rate Prediction

Models

prediction models for discrete semiconductors. Separate models were
developed for transistors, diodes, and opto-electronic semiconductor
devices. These models are presented in Appendix A in a form compatible
with MIL-HDBK-217D.
Transistors
The proposed model for transistors is:
Xp = Xnb wNT 'INQ iTNE ircyc ,
where
Xp predicted transistor nonoperating failure rate

Xnb = nonoperating base failure rate (failure/10 6 hours)
= .00027, Si,NPN (Group I)
= .00027, Si,PNP (Group I)
= .00040, Ge,PNP (Group I)
=.00040, Ge,NPN (Group I)
= .00039, FET (Group II)
= .0013, Unijunction (Group III)
o
.041, Microwave (Group IX)
= exp(-An( 1 + ( )P)
where
T temperature (OK) V
An = temperature coefficient (see Table 5.2.1-1)
5-79 •..
TABLE 5.2.1-1: DISCRETE SEMICONDUCTOR. NONOPERATING TEMPERATURE

FACTOR PARAMETERS
Part Group Style ATm

Transistors I Si,NPN 3356 448 10.5
Si,PNP 3541 448 14.2
Ge,PNP 4403 373 20.8
Ge,NPN 4482 373 19
II FET 3423 448 13.8
III Uni junction 4040 448 13.8
Diodes IV Si, Gen. Purpose 4399 448 17.7
Ge, Gen. Purpose 5829 373 22.5
V Zener/Avalanche 3061 448 14
VI Thyristors 4311 448 9.6
VII Microwave 2738 421, 16.6
VIII Impatt., Gunn, 3423 448 13.8
Varactor, Pin, Step
Recovery & Tunnel
Transistors IX Microwave 5700 623 20
5-80
ji V ~ - - . . - - - - -~
Tm,P - shaping parameters (see Table 5.2.1-1)

wNQ - nonoperating quality factors
= 0.57, JANTXV
= 1.0, JANTX
= 3.6, JAN
= 13, lower, hermetic
= 23, plastic
iNE a nonoperating environmental factor (see TablE 5.2.1-2)
IWcyc = equipment power on-off cycling factor
= 1 + .050(Nc)
where

hours
Diodes
The proposed model for diodes is represented by the following

equation.
xXp =Xnb "NT "NQ ONE fcyc
where
x)p = predicted diode nonoperating failure rate

,nb = nonoperating base failure rate (failure/lO6 hours)
= .00017, Si, General Purpose (Group IV)
= .0004?, Ge, General Purpose (Group IV)
= .00040, Zener/Avalanche (Group V)
= .0027, Microwave (Group VII)
= .00063, Thyristor (Group VI)
.0027, Impatt, Gunn, Varactor, Pin, Step Recovery and Tunnel
(Group VIII)
5-81
TABLE 5.2.1-2: TRANSISTOR NONOPERATING ENVIRONMENTAL FACTORS (iNE)
Env. Group I Group II Group III Group IX

GB 1 1 1 1
5.8 4.0 4.0 2.0
GM 18 18 18 7.8
Mp 12 12 12 7.4
NSB 9.8 6.0 9.3 3.6
NS9.8 8.6 9.3 4.7
NU 21 21 21 11
NH 19 19 19 11
NUU 20 20 20 12
ARW 27 27 27 16
AIC 9.5 7.5 9.5 2.5
AIT 15 9 15 3.5
AIB 35 35 35 6.0
AIA 20 30 20 3.5
AIF 40 40 40 6.0
AUC 15 10 15 5.0
AUT 25 15 ?5 7.0
AUB 60 55 60 10
AUA 35 50 35 7.0
AUF 65 65 65 10
SF (1) (1) (1) (1)
MFF 12 12 12 7.5
FFA 17 17 17 11
USL 36 36 36 22
ML 41 41 41 25
CL 690 690 690 250
NOTES: (1) This study did not address Space Flight environment
5-82
INT - nonoperating temperature factor

- exp(-An( 1-1 ) + ( T)P)
where
T - temperature (OK)
An a temperature coefficient (see Table 5.2.1-1)
Tm,P a shaping parameters (see Table 5.2.1-1)
INQ a nonoperating quality factors
- 0.57, JANTXV
- 1.0, JANTX
- 3.6, JAN
- 13, lower, hermetic
a 23, plastic
RNE a nonoperating environmental factor (see Table 5.2.1-3)
,cyc o equipment power on-off cycling factor
- 1 + .-8 3(Nc)
where
Nc a number of equipment power on-off cycles per 103 nonoperating

hours
Opto-electronic Devices
The proposed model for opto-electronic semiconductor devices is the

following equation.
"p= Xnb fNQ 'NE
where
x = predicted opto-electronic semiconductor nonoperating failure

rate
5-83
.......
...
TABLE 5.2.1-3: DIODE NONOPERATING ENVIRONMENTAL FACTORS (WNE)
Env. Group IV Group V Group VI Group VII Group VIII

GB 1 1 1 1 1
GF 3.9 3.9 3.9 6.4 3.9
GM 18 18 18 31 18
Mp 12 12 12 35 12
NSB 4.8 5.8 5.8 8.0 5.8
NS 4.8 8.7 8.7 11 8.7
NU 21 21 21 33 21
NH 19 19 19 54 19
NUU 20 20 20 58 20
ARW 27 27 27 78 27
AIC 15 4.5 9.5 30 4.5
AIT 20 6.5 15 40 6.5

AIB 30 45 35 65 45
AIA 25 25 20 50 25
AIF 35 45 40 70 45
AUC 25 7.5 15 50 7.5
AUT 30 10 25 60 10
AUB 50 70 60 105 70
AUA 40 40 35 ý80 40
AUF 50 70 65 110 70
SF (1) (1) (1) (1) (1)
MFF 12 12 12 36 12
FFA 17 17 17 50 17
USL 36 36 36 110 36
ML 41 41 41 120 41
CL 690 690 690 2000 690
r69 60 20 6
Xnb = nonoperating base failure rate (failures/106 nonoperating hours)

= .00016, LED
= .00070, Single Isolator
= .00089, Dual Isolator
- .00038, Phototransi stor
= .00028, Photodiode
= .00025, Alpha-Numeric Displays

friQ = nonoperating quality factors
- 0.57, JANTXV
n 1.0, JANTX
= 3.6, JAN
= 13, lower, hermetic
= 23, plastic
•NE = nonoperating environmental factor (see Table 5.2.1-4)
TABLE 5.2.1-4: OPTO-ELECTRONIC SEMICONDUCTOR NONOPERATING

ENVIRONMENTAL FACTORS (CNE)
Env. 7NE Env. WNE

G8 1 AIA 3.5
GF 2.4 AIF 8.0
GM 7.8 AUC 3.0
Mp 7'.7 AUT 5.5
NSB 3.7 AUB 8.0
NS 5.7 AUA 5.5
NU 11 AUF 10
NH 12 SF (1)
NUU 13 MFF 7.8
ARW 17 MFA 11
SAIC 2.5 USL 23
AIT 3.5 ML 26
AIB 5.5 CL 450
"5-85
5.2.2 Transistor Model Development
The general nonoperating failure rate model development approach

described in Section 4.1 was successfully implemented for transistors. A
theoretical model was hypothesized based on information located during the
literature search. Selection and quantification of the independent
variables was accomplished by analyzing the available data. The proposed
nonoperating failure rate prediction model for transistors was found to be
a function of device style, screening, temperature, application
environment and equipment power on-off cycling frequency.
The application and construction variables which were identified for

transistors are presented in Table 5.2.2-1. These variables represent
possible nonoperating failure rate model parameters. Parameter values for
these factors were identified whenever possible as part of the data
collection effort. If sufficient detail could not be determined for a
specific data entry, it was deleted from the analysis.
The general nonoperating reliability concepts presented in Section 4

for temperature, screening, environment and equipment power on-off cycling
were studied for development of a transistor theoretical model. These
concepts were complemented by an investigation of transistor nonoperating
failure mechar,isms to determine the theoretical model. For transistors,
the theoretical model was multiplicative but nonlinear due to the assumed
effects of equipment power on-off cycling and temperature. The
theoretical transistor failure rate prediction model was initially
determined to be a function of device style, application, complexity,
rated power, quality, environment, temperature, and equipment power on-off
cycling frequency. After further qualitative analyses, it was determined
that application and rated power were important variables for operating
failure rate prediction, but were not nearly as significant for
nonoperating failure rate prediction. Therefore, these two variables were
deleted from the theoretical model.
5-86
,..vi
TABLE 5.2.2-1: TRANSISTOR PART CHARACTERIZATION
I. Device Style
A. Group I
1. SI,NPN
2. Si,PNP
3. Ge,NPN
4. Ge,PNP
B. Group II
1. Si,FET
2. GaAs,FET
C. Group III (Unijunction)
D. Group IX (Microwave)
I1. Application
A. Linear E. Driver
B. Switch F. Pulse Amplifier
C. Low Noise G. Continuous Wave
D. High Frequency H. Oscillator
III. Complexity
* A. Single Device E. Dual Emitter
B. Dual (matched) F. Multiple Emitter
C. Dual (unmatched) G. Complementary Pair
D. Darlington Pair H. Tetrode
- IV. Rated Power (watts)
V. Quality Level
A. JANTXV
B. JANTX
C. JAN
D. Lower, Hermetic
E. Plastic
VI. Application Environment
VII. Temperature
A. Rated
B. Actual
VIII. Number of Power On/Off Cycles per 103 Nonoperating Hours
5-87
}S
The collected transistor nonoperating reliability data is summarized

in Table 5.2.2-2. The table presents the number of data records, observed
failures and summed part hours for each transistor style and equipment
represented in the collected data. The data were initially analyzed
separately without merging data records. The data were subjected to a
thorough preliminary analysis to determine weaknesses of the data set.
The following attributes of the available data were identified.
" 1) All collected data were from a ground based environment.

Therefore, appropriate nonoperating environmental factors could
not be determined empirically.
2) Only one source had data for JAN level transistors.

3) All high temperature data were from AFCIQ, and were not screened
to identical specifications.
4) High temperature storage data were only available for Si,PNP and 70
Si,NPN device styles.
Application environment was hypothesized to be a significant variable

in the theoretical model development. However, nonoperating environmental
factors could not be determined empirically because all collected data was
for ground based environments. The methods described in Section 4.5 were
used to determine appropriate nonoperating environmental factors after
statistically analyzing the other variables. 4
Matrices of "dummy variables" (0 or 1) were defined for device style ' '
and quality level to accommodate quantitative analyses. The matrix for

transistor device style is presented in Table 5.2.2-3. The matrix was
defined to represent (1) Si,NPN, (2) Si,PNP, (3) FET, and (4) microwave
transistors. Data in sufficient quantity were not available for other
transistor styles. The matrix for quality level was simply a "0" for
JANTX parts and a "1" for JAN parts. This variable was designated as QI.
Data were not available for quality levels other than JANTX and JAN, aId
therefore, the quality matrix was not defined to represent other quality
levels.
5-88
.ýQ 0,7-.
TABLE 5.2.2-2: TRANSISTOR NON(JPERATING FAILURE RATE DATA
Part
Data Hours
Equipment/Source Records Quality Style Failures (X16
Hawk/MICOM 9 JANTX Si,NPN 6 5189.5
Hawk/MICOM 5 JANTX Si,PNP 2 2484.1
Hawk/MICOM 5 JANTX FET 3 2501.2
Hawk/MICOM 1 JANTX Microwave 1 17.0
Maverick/MICOM 6 JANTX Si,NPN 4 1400.1
Maverick/MICOM 3 JANTX Si,PNP 4 1482.5
Maverick/MICOM 1 JANTX FET 0 41.2
Sparrow/MICOM 1 JAN Si,PNP 0 25.5
Sparrow/MICOM 9 JAN Si,NPN 1 408.3
Sparrow/MICOM 3 JAN FET 0 25.5
Sprint/MICOM 8 JANTX Si,PNP 0 489.8
Sprint/MICOM 3 JANTX Si,NPN 2 1446.2
TOW/MICOM 3 JANTX Ge,PNP 0 13.1
TOW/MICOM 4 JANTX Si,NPN 0 107.7
TOW/MICOM 3 JANIX Si,PNP 0 39.4
Lance/MICOM 4 JANTX Si,PNP 0 11.2
Lance/MICOM 4 JANTX Si,NPN 0 15.7
Martin Marietta 3 JANTX Si,PNP 1 1326.4
Martin Marietta. 3 JANTX Si,NPN 6 4076.1
Martin Marietta 1 JANTX Ge,NPN 0 20.8
Martin Marietta 1 JANTX Ge,PNP 0 44.8
Martin Marietta 1 JANTX FET 0 71.7
Martin Marietta 1 ANX Uni junction 0 1.0
F-16 HUD/RIW 8 JANTX S1,NPN 4 350.0
F-16 HUD/RIW 5 JANTX Si,PNP 0 363.3
AFCIQ (1) 4 (2) Si,PNP 3 8.7 ..
AFCIQ (1) 4 (2) Si,NPN 12 20.8

PRC 1 JANTX FET 0 0.3
ERADCOM (3) 1 Plastic Si,NPN 33 0.9H
ERAOCOM (3) 1 Plastic Si,PNP 39 0.3
Totals 106 121 21983.1
NOTES: 1) High temperature storage life test data. Storage

temperatures range from 1500 C to 2000C.
2) These devices not screened to identical specifications.
3) Panama life test data for nonhermetic devices. Ambient
temperature =300C, relative humidity =90%.
L
5-89 .
TABLE 5.2.2-3: TRANSISTOR STYLE VARIABLE MATRIX
Device Style S1 S2 S3
Si,NPN 0 0 0
SI,PNP 1 0 0
FET 0 1 0
There were several problems associated with the determination of an

appropriate nonoperating temperature factor for transistors. One problem
with the high temperature storage life test data from AFCIQ was that the
transistors were not screened to similar specifications as the other data.
Another problem was that data were only available for Si,PNP and Si,NPN
transistor styles. A third problem was that no empirical or theoretical
relationship for nonoperating failure rate vs. temperature could be
located in the literature. Additionally, the assumed transistor
temperature relationship presented in Section 4.4 could not be transformed
into a linear form. As a solution to this dilemma, approximate quality
levels were determined and various nonlinear relationships were fit to the
data after statistical analysis of the other variables and assumption of
an applicable series of nonoperating environmental factors.
Three qualitative "dummy variables" were then defined to represent

four distinct equipment power cycling rate categories. The three
variables were designated cycl, cyc2 and cyc3. The definitions of the
three categories and the corresponding cycl, cyc2 and cyc3 values are the
following.
o cycling rates less than one power cycle every two years (cycl,
cyc2, cyc3 = 0,0, 0).
years (cycl, cyc2, cyc3 = 1, 0, 0).
o cycling rates greater than one cycle per year (cycl, cyc2, cyc3 -
0, 0, 1).
o unknown cycling frequencies. L
5-90
. • •I
Definition of the cycling rate variables was necessary because of the .

nonlinear form of the proposed model. Equipment power on-off cy~..ing was
temporarily treated as a qualitative variable and an approximate cycling
factor relationship was determined. Then, an iterative approach was taken
for nonlinear regression. This procedure is described in Section 4.3.
The assumed form of the equipment cycling factor is expressed by the
following equation.
Cycling Factor = 1 + Kl(cycles/10 3 hours)
where
Ki1 constant
The data were then merged and regress analysis applied. The data
were merged according to equipment, device style, complexity and quality
level by si,nming the number of failures and part hours. For example, the
number of failures and part hours were summed for all JANTX, FET devices
in the Hawk missile. Initially, two regression analyses were performed.
In each case the dependent variable was equal to the natural logarithm of
nonoperating failure rate. One regression was performed to evaluate the
validity of the diode nonoperating quality factor (described in Section
5.2.3) for transistor nonoperating failure rate prediction. This approach
was taken because of the belief that screening effects transistors and
diodes similarly, and because only one data source was available for JA,
transistors.
The initial regression indicated that the ratio of JAN to JANTX device
nonoperating failure rate was equal to 2.24. This was compared to the
corresponding value of 3.63 for diodes. A 90% confidence interval around
the point estimate ratio for transistors was computed to be 0.83 and 6.03.
This relatively large interval was due in part to the limited amourt of
JAN device data. The data did not disprove that the observed ratio for
diodes could also be apo'ied for transistor nonoperating failure rate r
prediction. Therefore, the ratio of 3.63 determined for diodes was
5-91
- * .. . .
S.J.,
appliled to transistors. The complete series of nonoperating quality

factors are presented in Table 5.2.3-6 in the following section. A second
regression analysis was then performed assuming that the diode
nonoperating quality factors were accurate. Results of the second
regression analysis are presented in Table 5.2.2-4. The variable
coefficient, the standard error and the F-ratio are presented for each ":
significant variable.
TABLE 5.2.2-4: TRANSISTOR REGRESSION RESULTS II
Standard Confidence
Variable Coefficient Error F-Ratio Limit -:

S3 4.054 0.501 65.43 0.90
cycl 1.000 0.376 7.09 0.90
cyc2 1.651 0.376 19.29 0.90
bo -6.887 ....- --
Interpretation of the regression results are the following.
1) The nonoperating failure rate for microwave transistors was

significantly different from the othev transistor styles with 90%
confidence.
2) The difference in failure rate between Si,NPN, Si,PNP, and FET
devices was not significant at a 70% confidence limit.
3) There was no observed difference between single and dual I.
complexity transistors.
4) The cycl and cyc2 equipment power cycling variables were
significant at 90%. The resultant factors were in descending
order from the highest to lowest cycling rates, as expected.
There was no significant difference, with 70% confidence, between
the lowest cycling rate category and the unknown cycling rate data
entries. K.
The next step in the model developrnent task for transistors was to
perform iterative regression analyses to solve the nonlinear model form.
The nonlinear form resulted from the assumed equipment power on-off
cycling relationship. This process was described in Section 4.3. For °.-
5-92
L
transistors, three iterations were required. The regression solution

after the third iteration is given by the following equation. The
dependent variable was transformed to nonoperating failure rate.
Previously, the dependent variable was the natural logarithm of
nonoperating failure rate adjusted to compensate for quality. The
preliminary model at this stage of the model development process is:
XtXn-b lNQ lTcyc
where
Xt = predicted transistor nonoperating failure rate

Xnb = preliminary nonoperating base failure -ate (failure/106
nonoperating hours)
= .00130, Si,PNP
= .00130, SiNPN
= .00130, FET
= .0588, Microwave
•NQ = assumed nonoperating quality factors
7rcyc = equipment power on-off cycling factor .
= 1 + .0 6 8 9 (Nc)
where .

hours
It must be emphasized that the above equation is preliminary, and only,

applies to a ground fixed environment with an average nonoperating ambient
temperature of 200C. The nonoperating base failure rate in the proposed
model (Section 5.2.1) corresponds to a reference temperature of 250C.
Therefore, determination of the proposed nonoperating base failure rate
was dependent on the determination of transistor nonoperating temperature
factors.
5-93
>-.;" r.T.7"-;,.w•:,:,
•,C••=• ,, ;-:,:
.'w-ei.•'
:-,"T•;"x•.',..,~'v•<.-,• -•-,• •¢•,w'-ww'~-,w:., €, •v 9 i
The next phase of the nonoperating failure rate model development was
to estimate appropriate nonoperating environmental factors. The methods
presented in Section 4.5 were applied to determine applicable factors.
This method assumes that a series of nonoperating environmental factors
can be generated from the MIL-HDBK-217 operating environmental factors
based on (1) a comparison of application temperature differences, and (2)
a comparison of operating and nonoperating failure mechanism accelerating A
factors. Operating temperatures are typically higher because of the
internal heat generation. However, the effects of this temperature •N "
difference are predicted by the respective operating and proposed
nonoperating temperature factors. Therefore, no temperature adjustment L
was required for transistors. A comparison was made between operating and
nonoperating failure, mechanisms and failure accelerating factors. A
surrntry of the failure mechanism comparisons are presented in Tables
5.2.2-5 and 5.2.2-6 for bipolar single and power transistors, and FETs . /
respectively. References 42 and 43 included transistor failure V-
mode/mechanism information. Quantitative failure mechanism distributions
were essentially impossible to locate for nonoperating applications, and
often conflicting for operating applications. Nevertheless, it was
determined that the operating transistor environmental factors approximate
the effect of environmental stress for nonoperating failure rate. These
factors were presented in Table 5.2.1-2 in Section 5.2.1.
The next stage in the model development process was to determine

nonoperating temperature factors. Data were available from AFCIQ with
temperatures ranging from 150oC to 2000C for Si,PNP and Si,NPN
transistors. However, this range of values was insufficient to determine
whether the equivalent Arhennius equation was applicable for transistor
nonoperating failure rate. It was believed that the Arhennius
relationship was not applicable because of the relationship for transistor
failure rate vs. temperature in MIL-HDBK-217D. Therefore, a modified
Arhennius equation was assumed. The modified temperature factor allowed
for a nonlinear relationship between the natural logarithm of nonoperating
failure rate and the inverse of temperature. Additionally, a reference .,
0
temperature term of 298 K was added to increase the utility of the
5-94
.
M... .
T..... .... .. ........
..
oCCI
LA CD 0 n L
0 4J
zm C
W ~ 4 +3-4 " -4C' 'J %C3~
"4
V- V-
9-4 L -4 .9-4U n
4d 4
*9-*-)4)1 4)
- a, 4)4) CC * ** 4
be -V SI-- to OE
C (A
OUo0U
4J 4) c V
to 0 4) W V
S- S-) a, L0
>U 4a, Ua, 4 ) A
C40 C -
0 G3e 40 ,
4-*- 9- V 4V
Ki e-f4 ) 'u4
cc' VC W'
a a4 )LN 0C OC
S..R 4 )J 4U L
a, 0
'a . u C 1.
* LL. 0 V 4)
4-) 0
cococ-t OLf
C.O Ln m
LnU 12?
S-41 -4 r4 -4 -4v-4C' (% . I
C..0 -44-O OIn UIn -4ý
o 4.
ccC.
*9-C*9- J(DU
mLn qc OLC)
C>O
o- V-c4q-I 1-4 "4-4 , , -

r-4 -4 Ln U)
d) 9-ý
0.L
0)CD4-217
C-.)
u~.9
!mNeS
Aj 4jjWf
Im
W
goU
U.
-
00
041
= S-
4)
-
S
C .
-
M..
0 0)
S-
II
Q>
4J .J
*l 4-3 af~lS. E 0.- a0
to 0 (UF >
S.- - 4141 a, 4-3 4
*ý()
- 4-) r_) 7).9 . - Ct
LA C- d) U toE -le- 00
S. . .9 4) 4.). S- 44) 4)
-L. E, S-~ S- E *.
-- .- =0 =a 0 a) V d) >S
> >4. > 4-3 .Cf4-
.4.
0 0o
0 4-C-00
.- U) 4- E to.
4-a.--o
u , .-&-
0 a) E 0 0
C 4-)U N 0 C
Eu 0
uaIM 00 E~
a, 4)V - *9- CL) 0~ 4- 0
U4-~ a, 04-i
a S- 4a,-) -4 .)* (1)
- a) 0
ao -.- 0
LL- 0 (A) A-
5-96
proposed model. The temperature factor for transistors was assumed to be

the following equation.
1TNT " exp(-An(T - )+ (1m)
where
iNT = nonoperating temperature factor

An = temperature coefficient
Tr = reference temperature = 298 0K
Tm, P = shaping parameters
The form of the assumed nonoperating temperature factor was based on the
relationship of operating failure rate versus temperature in MIL-HDBK-
217D. In approximate terms, the Tm shaping parameter is the temperature
(expressed in degrees Kelvin) were the equivalent Arhennius relationship
is no longer applicable, and the P shaping parameter provides an
indication of how fast the assumed relationship deviates from the
equivalent Arhennius relationship.
Several decisions were required before the assumed temperature factor

could be quantified. First, an equivalent quality level of lower,
hermetic was determined for the high temperature storage data based on
information provided by AFCIQ. Second, there were insufficient data to
estimate all three unknowns (i.e., An, Tm, P) in the assumed temperature
relationship. It was decided that the Tm and P parameters which are used
for operating failure rate prediction would be applicable for nonoperating
failure rate prediction purposes. Unique An values were then determined
empirically. Ideally, it would have been preferable to solve for all
three parameters. However, data limitations prevented this.
Five data entries were available for transistor high temperature

storage life testing. To compute the An temperature coefficients, the
regression solution was assumed to be exact for 200C, and it was then
5-97
assumed that the relatively higher failure rates from the high temperature
data were due only to the increased temperatures. The effects of the
other variables were eliminated by using the previously determined factors
for quality, equipment power cycling and environment. Specifically, two
simultaneous equations were solved for each high temperature data entry.
Unknown parameters were the nonoperating base failure rate and the An
temperature coefficient. The nonoperating base failure rate was 'Unknown
because it was dependent on the reference temperature, and thus, also

dependent on the temperature coefficient.
The two equations used for the temperature analysis, and solved
simultaneously were the following. The first equation was the regression
solution found earlier for an average field storage temperature of 200C
(or 293 0 K). The second equation was the assumed model for the high to
temperature storage data. In the first equation, the preliminary base
6
failure rate (.00130 f/10 hours) corresponded to JANTX quality, a ground
fixed environment, a temperature of 200C, and no equipment power on-off
cycling. Therefore, the previously determined values for those parameters
were inserted in the first equation given below. In the second equation,
the observed high temperature storage life test failure rate corresponded
to an equivalent quality of lower, hermetic, ground benign environment,
temperatures ranging from 150oC to 2000C and approximately 3 power cycles PAZ
per 103 hours. The corresponding factors for these parameters were
inserted in the second equation.
1) Xt Xnb 7NQ 7NE •cyc(exp(-An(2293)P

I 298 ) + (293) -Tm
.00130 Xnb(1)(5.8)(1 + .069(1))exp(-An( - 1 + (298
2) Xt = Xnb rNQ TINE

1cyc(exp(-An(T - + (T-)P
Xo0 Xnb(3.63)(1)(I + .069(3))exp(-An( T- 1 + (T-)
5-98
where
Tm = 448, Table 5.1.3-21, MIL-HDBK-217D

P = 10.5, Si,NPN, Table 5.1.3-2, MIL-HDBK-217D
= 14.2, Si,PNP, Table 5.1.3-2, MIL-HDBK-217D
Ts = storage temperature (OK) •.__
,o = observed nonoperating failure rate (failures/10 6 hours)
Unique solutions for An were found for each of the five high
temperature storage life test data entries. These data entries and the
solutions for An are presented in Table 5.2.2-7. The proposed temperature
coefficient of 3356 for Si,NPN transistors was the average An value for
entries I and 2 in Table 5.2.2-7. Similarly, the proposed temperature
coefficient of 3541 for Si,PNP transistors was the average An value for
entries 3, 4 and 5. It was noted that the proposed values were higher
than the corresponding operating temperature coefficients (NT in Table
5.1.3-2, MIL-HDBK-217D). One possible explanation for this observation
was that there are fewer stresses acting on a part during nonoperation,
and therefore. the failure rate is more sensitive to each individual
stress. It was also noted that the absolute difference between the
proposed An values for Si,PNP and Si,NPN was very similar to the
corresponding difference between operating temperature coefficients.
TABLE 5.2.2-7: HIGH TEMPERATURE STORAGE LIFE TEST DATA
Part
Entry Hour Equivalent Temperature
No. Failures (X1Ob) Xo(1) Style Quality (oC) An(2)
1 3 7.35 0.41 Si,NPN lower 150 4021
2 9 13.41 0.67 Si,NPN lower 200 ?690 .0ý
3 3 0.88 3.41 Si,PNP lower 175 4979
4 0 6.53 <0.15 SI,PNP lower 150 <3159
5 0 1.30 <0.77 Si,PNP lower 200 <2486
NOTES: (1) Observed failure rate. One failure was assumed for data
entries without failures
(2) An transistor temperature coefficient
5-99
Estimates of the temperature coefficients for Ge,PNP, Ge,NPN, FET, and

Unijunction were calculated by (1) assuming that the nonoperating
temperature coefficient rankings would be the same as the operating
temperature coefficient rankings, and (2) the temperature coefficient,- -
difference between operating and nonoperating ý.ould be the same for the
other device styles as the observed difference for Si,PNP and Si,NPN. The
MIL-HDBK-217D failure rate prediction model for microwave transistors
assumes the equivalent Arhennius relationship applied for the entire
temperature range found during normal usage. The complete set of
temperature coefficients and shaping parameters is presented in Table
5.2.1-1 of Section 5.2.1.
A by-product of the temperature coefficient determination process was

that the nonoperating base failure rate was also found for Si,NPN, and
Si,PNP transistors. The value was computed to be 0.00027 failures/10 6
nonoperating hours as part of the simultaneous equation solution. The
same value could have been determined by dividing the preliminary base
failure (0.00130 failures/10 6 nonoperating hours) by both the ground fixed
nonoperating environmental factor of 5.8 and the nonoperating temperature
factor of 0.83 for 200C. This would have been necessary to normalize the
base failure rate to correspond to the desired parameters. A comolete
series of nonoperating base failure rates were then determined by
normalizing the preliminary base' failure rates to the desired parameters.
Additionally, nonoperating base failure rates were determined by
extrapolation for transistor types without data.
Determination of the nonoperating

base failure rates concluded the
model development process for trarsistors. The proposed model for
transistor nonoperating failure rate is therefore given by, I
Xt =nb 'NT T7NQ 7NE '-cyc
4 i,
5400 •=.•
where
Xt - transistor nonoperating failure rate

Xnb = nonoperating base failure rate (failures/lO6 nonoperating
hours)
= .00027, Si,NPN
= .00027, Si,PNP
= .00040, Ge,PNP
= .00040, Ge,NPN
= .00039, FET
= .0013, Unijunction
.041,Microwave
;rNT nonoperating temperature factor
= K
fNQ = assumed nonoperating quality factor
NNE = nonoperating environmental factor
1 + .069(Nc)
where

hours
The proposed transistor nonoperating failure rate prediction model is

presented in Section 5.2.1 and in Appendix A in a form compatible with
MIL-HDBK-217D. Transistor model validation is described in Section 5.2.5.
5.2.3 Diode Model Development
A nonoperating failure rate prediction model for diodes was developed

by hypothesizing a theoretical model, and then by analyzing empirical data
to quantify the model parameters. The proposed model was determined to be
a function of device style, environment, temperature, quality level and
equipment power on-off cycling. The model factors for device style,
quality level and equipment power on-off cycling were determined
5-101
• '.- .- ',,,•, •.•", n " """-, "., ,•: - o.w-, .y '. • ",-, '"" ' :,:.. ... :)
analytically. The factors for temperature and application environment

were assumed. The proposed model is presented in Section 5.2.1 and in
Appendix A in a form compatible with MIL-HDBK-2170.
The initial phase of the model development process was to identify

application and construction variables which characterize diodes in a
nonoperating envir,-nment. These variables, presented in Table 5.2.3-1, 1._..
represent possibl'e nonoperating failure rate moeel parameters.
Additionally, identification of the application antd construction variables
was a prerequisite to both the theoretical model development and data r:.' -,I
collection tasks.
A theoretical model was developel for diode nonoperating failure rate

by studying both documented and intuitive reliability relationships. The
theoretical model was nonlinear due to the assumed effects of temperature
and equipment power on-off cycling frequency. The theoretical model was
also assumed to be a function of device style, application, rated power, ,. .o
quality level and application environment.
Further refinement of the theoretical model for diodes resulted in the ,

deletion of part application and rated power from the model. It was
believed that these parameters had negligible effect on diode nonoperating V.%
failure rate. The refined theoretical model formed the basis for model 0 1W
development and subsequent analyses.
An extensive data collection effort was performed in support of this

study. The collected diode nonoperating failure rate data is presented in ___
a summarized form in Table 5.2.3-2. Data were collected in sufficient

quantity for general purpose, zener, microwave and thyristor diodes.
5-102
, .S( WI
N%
is
TABLE 5.2.3-1: DIODE CHARACTERIZATION VARIABLES
I. Device Style .J
A. Group IV Purpose
1. Si, General Purpose
2. Ge, General Purpose
B. Group V (Zener & Avalanche)
C. Group VI (Thyristors)
D. Group VII (Microwave Diodes)'
1. Microwave Detector
2. Microwave Mixer
3. Schottky Detector "/.
E. Group VIli
1. Varactor
2. PIN
3. IMPATT
4. Step recovery
5. Tunnel
6. Gunn
II. Application
A. Analog Circuits
B. Switching
C. Power Rectifier
D. Voltage Regulator
E. Voltrage Reference
III. Rated Power (watts)
IV. Quality Level

A. JANTXV
B. JANTX
C. JAN
D. Lower, Hermetic
E. Plastic
V. Temperature
A. Rated
B. Actual
VII. Number of Equipment Power On/Off Cycles per 103 Nonoperating Hours
5-103
V•
TABLE 5.2.3-2: DIODE NONOPERATING FAILURE RATE DATA
Equipment/Source Data Records Quality Style Failures Part Hours

(X 106)
Hawk/MICOM 6 JANTX Si,Gen.Purpose 0 4355.8
Hawk/MICOM 13 JANTX Zener 1 351.3
Hawk/MICOM 1 JANTX Microwave 3 204.2 w
Maverick/MI COM 9 JANTX Si,Gen.Purpose 2 4176.9
Maverick/MICOM
.Sparrow/MI COM
Sparrow/MICOM
Sprint/MI11OM
Sprint/MICOM
9
28
14
2
7
JANTX
JAN
JAN
JANTX
JANTX
Zener.
Si,Gen.Purpose
Zener
Si,Gen.Purpose
Zener
1
2
0
0
0
411.8
1084.6
118.6
11.6
81.3
I
Sprint/MICOM 3 JANTX SCR 1 509.2
TOW/MICOM 7 JANTX Si,Gen.Purpose 0 162.9
TOW/MI COM 2 JANIX Zener 0 5.3
Lance/MICOM 7 JANTX Si,Gen.Purpose 0 67.1
Lance/MICOM 5 JANTIX Zener 0 10.1
Martin Marietta 2 JAN Si,Gen.Purpose 41 6262.0
Martin Marietta 1 JAN Zener 0 607.0
Martin Marietta 1 JANIX Zener 1 898.0
Martin Marietta 1 JANTX Tunnel 0 1.9 0
Martin Marietta 1 JANTX Varactor 0 1.9
F-16 HUD/RIW 19 JANTX Si,Gen.Purpose 5 2192.8
PRC 1 JAN TX Zener 0 0.3
Totals 139 57 22180.6
5-104
Z;ý. ~ '.J j
A preliminary analysis of the collected data was then performed to

identify relative strengths and weaknesses. The primary weakness was that
all data were from a ground based environment. Therefore, there were an
insufficient range of environmental stresses represented in the data to
develop a series of nonoperating environmental factors empirically. This
deficiency was common to all generic part types considered in this study.
An alternate nonoperating environmental factor development procedure was I -4
developed (described in Section 4.5) to compensate for this deficiency.
This method was applied after application of regression analysis to
evaluate the effects of device style and quality. Another major
deficiency with the summarized data set was that there was an insufficient
range of temperatures to study the effects of temperature analytically.
It was considered to be essential that a proposed diode nonoperating
failure rate prediction model include the effects of temperature. -""i
Therefore, an applicable nonoperating temperature factor was assuaned based
on the MIL-HDBK-217D diode failure rate prediction model, and the proposed
temperature factor for transistors.
"Dummy variable" matrices were defined for device style and quality.
Statistical techniques such as regression analysis and correlation
coefficient analysis require that variables are quantitative. Conversely,
device style and quality level are qualitative variables. Therefore,
definition of the matrices was necessary to accommodate the techniques
required for nonoperating failure rate model development. The matrix for
device style is presented in Table 5.2.3-3. Three variables were required
to represent four distinct diode styles. The three variables were
designated S1, S2, and S3 . The matrix for quality level was simply a "0"11
for JANTX diodes and a "I" for JAN diodes. This variable was designated
"Q10
.- ,. • . . . . . •5 . . . .
TABLE 5.2.3-3: DIODE STYLE VARIABLE MATRIX
Diode Style S1 S2 S3
Si, General Purpose 0 0 0 •.
Zener 1 0 0
Thyristor 0 1 0
Microwave 0 0 r
The correlation coefficient analysis indicated no large correlation

between independent variables. The data were then merged according to
equipment, device style and quality level by summing the observed failures
and the part hours. For example, all JANTX zener diode data were merged
for the Maverick missile by summing the observed failures and part hours.
Similar to the discussion for transistor model development, three

equipment power on-off cycling variables (cycl, cyc2
were and cyc3)
defined to represent four distinct equipment cycling rate categories. An
approach of this type was necessary because the assumed relationship
between nonoperating failure rate and equipment power on-off cycling -
resulted in a nonlinear model form. An iterative approach was taken to
solve the nonlinear model. The four cycling rate categories were
designated as (1) low, (2) intermediate, (3) high, and (4) unknown cycling V .
rates, dnd were defined as,
cyc2, cyc3 = 0,0,0).
0 cycling rates between one cycle per year and one Lycle every two
years (cycl, cyc2, cyc3:= 1,0,0).
o cycling rates greater than one cycle per year (cycl, cyc2, cyc3
1,0,0).
o unknown cycl ing, rates (cvcl, cyc2, cyc3 0,0,1).
At this stage of the model development process, regression analysis

w.s applied to the data. Results of the initial regression analysis are
presented in Table 5.2.3-4. The dependent variable was the natural
logarithm of nonoperating failure rate and the independent variables y .
5-106 "
introduced into the analysis were S1, S2, S3 , Q1, cycl, cyc2 and cyc3.
The S3 variable was significant with 90% confidence, the Q1 variable was
significant with 70% and S1, S2 and cyc2 were significant with 60%
confidence.
TABLE 5.2.3-4: DIODE INITIAL REGRESSION RESULTS
Variable Coefficient Standard Error F-ratio Confidence Limit
Si 0.916 0.716 1.64 0.60

S2 1.309 1.155 1.29 0.60
S3 3.322 1.155 8.28 0.90
Q1 1.291 0.740 3.05 0.70
cyc2 1.459 1.155 1.60 0.60
bo -7.542 -- -- --
The regression results were next evaluated. Interpretation of the

regression solution was:
1) Microwave diode nonoperating failure rate was significantly

different from Si, general purpose diode failure rate with 90%
confidence.
2) Zener diode and thyristor nonoperating failure rates were
significantly different from Si, general purpose diode failure
rate with 60% confidence.
"3) It was determined with 70% confidence that quality significantly
"affects diode nonoperating failure rate.
4) The nonoperating failure rate for high equipment power cycling
rates was significantly larger, with 60% confidence, from devices
with low, intermediate or unknown cycling rates. There was no
significant difference, at the 60% confidence limit, between low,
intermediate and unknown cycling rates.

nonoperating failure rate. Transformning the regression solution into an
"5-107
SdZ Z
equation where failure rate (as opposed to log of failure rate) is the
dependent variable results in the following multiplicative model,.
Ad = Anb lTNQ •cyc e f
where
Ad
Anb
=
=
predicted diode nonoperating failure rate
preliminary nonoperating base failure rate (failures/106
exp(-7.542 + 0.916($ 1 ) + 1o309(S2) + 3.322($ 3 ))

I
F
=0.000530, Si, General Purpose•?

= 0.00133, Zener i,
= 0.00196, Thyristor
= 0.0147, Microwave
ITNQ = nonoperating quality factor
= exp(1.291(Q1))
= 1.0, JANTX
= 3.63, JAN
= exp(1.459(cyc2))
= 1.0, cycling rate < .057 cycles/lO3 hr.
m1.0', .057 .5 cycling rate < .114 cycles/lO3 hr.
3
= 4.30, cycling rate• ,j114cycles/10) hr.
S= residual
The variables S1, $2, $3, Q1 and cyc2 have been previously defined. It
must be emphasized that this equation represents a preliminary model, )
which is only applicable to a ground based environment with an average •!
nonoperating temperature of 20oC. r
The next step in the model development process was to perform Z
iterative regression analyses to accommnodate the assumed nonlinear model \,

form. For diodes, only one iteration was required. The regression after
respective confidence limits for S1 , S2 and Q1 increased after one

iteration. It was believed that this was evidence that the nonlinear
nonoperating
model form was indeed the optimal form for predicting diode
failure rate. The preliminary model at this stage of the model
development process is given by the following equation. The residual term
was dropped from the equation at this point. However, the fact that a
residual term should be included as part of a predictive model is implied.
X.d Xnb TNQlcyc
where
6
Xnb = preliminary nonoperating base failure rate (failures/10
nonoperating hours)
= 0.000528, Si, general purpose
= 0.00132, Zener
= 0.00196, Thyristor
= 0.0146, Microwave
jNQ = nonoperating quality factor
= 1.0, JANTX
= 3.64, JAN
itcyc = equipment power on-off cycling factor
I,= 1 + .083(Nc)
where

hours
:• ~5-109
TABLE 5.2.3-5: DIODE FINAL REGRESSION RESULTS
Variable Coefficient Standard Error F-ratio Confidence Limit
S1 0.917 0.607 2.28 0.70

$2 1.310 1.011 1.68 0.70
S3 3.323 1.011 10.80 0.90
Q1 1.291 0.640 4.08 0.80
bo -7.547 ......
The magnitude of the ratio of the JAN to JANTX nonoperating quality

factors was small in comparison to the corresponding ratio from MIL-HDBK-
217D. It was hypothesized that observation was due to a significantly
slower rate of failure for inherently weak devices in the nonoperating
state. This trend is described in Section 4.6 of this report.
Data were only available for JANTX and JAN diode quality levels.
However, it was essential that a proposed diode nonoperating failure rate
prediction model include applicable nonoperating quality factors for each
quality level presented previously in Table 5.2.3-1. Therefore,
appropriate nonoperating quality factor values were extrapolated by
assuming (1) that the nonoperating values would have a similar numerical
ranking as the corresponding operating values, and (2) that the
nonoperating values follow an increasing geometric progression. Table
5.2.3-6 presents the regression solution quality factor values, the
extrapolated values for JANTXV, lower and plastic, and a 90% confidence
interval computed around the point estimate value for JAN quality. The
proposed values are normalized to a JANTX factor equal to one by
definition. Conversely, the MIL-HDBK-217D diode quality factors do not
appear to be normalized to any quality level.
5-110
j. A
TABLE 5.2.3-6: DIODE NONOPERATING QUALITY FACTORS
•NQ2( 2)
Quality Level irNQ,.05 •NQ1() 7rNQ,.95
JANTXV ..-....- 0.57

JANTX -- 1.0 -- 1.0
JAN 1.00 3.64 13.2 3.6
Lower, Hermetic -- -- -- 13
Plastic ...... 23
NOTES: (1) 1TNQ1 are the observed point estimate nonoperating quality
factors.
(2) TNQ2 are the point estimate values complemented by

extrapolated values for JANTXV, lower and plastic.
The next phase of the model development process was to apply the
methods described in Section 4.5 to dezermine nonoperdting environmental
factors. This method assumed that appropriate nonoperating environmental
factors could be generated from the corresponding MIL-HDBK-217D operating
environmental factors by (1) comparing the difference in temperature, and
(2) comparing the operating and nonoperating failure mechanism
accelerating factors. No temperature adjustment was required for diode
nonoperating environmental factors because the operating base failure rate
(which is a function of temperature) and the proposed nonoperating
temperature factor account for relative failure rate differences caused by
internally generated heat. The comparison of failure mechanisms and
failure mechanism accelerating factors indicated that the effects of
environmental stress on nonoperating failure rate could be predicted by
use of the MIL-HDBK-217D environmental factors. The complete series of
diode nonoperating environmental factors is presented in Table 5.2.1-3 in
Section 5.2.1. Tables 5.2.3-7 and 5.2.3-8 on the following pages present
failure mode/mechanism and failure acceleration factor distributions for
diodes in both the operating and nonoperating state. Accurate
quantitative nonoperating failure mechanism information was extremely
difficult to locate. Therefore, much of the information provided in
Tables 5.2.3-7 and 5.2.3-8 was hypothesized. References 41, 42 and 43
proved to be useful for developing these distributions.
5-111 'Y
49- C ~ 0L
L4.)
= 0 I- I4 -4 V-I1
GI.0 Vw.4 -0 . 0
0 - C-C~JC
o 4.)
I-A
M C
CO-44 MI
4-.J4.) I I z-'I 01'C 00u
II I I Ontfl
I I
go. = D-4 - 4L r
9-a
0)-
Cie
cc9
U- 4-3
0 0 0
P- 0.0 00
m*
"9 4)4. 4J 4 4)E
u 00 E L 00 00E
0-
U C-
4) E u~ r--0 .
CV 4) 0
0 4 (A - 4OJ4- #
(D4- 41C 0
0 (D 00 4J t 0) 4-3
C .- 4J' .- 0-
m )4 c
(AU 4- -
UD 0 43' L)O 4J 4)
'o (L01-). CU Ej r'

~~*39-~~ 4) 0U - O
(D 0Vb J. t _4
00 M..
U.. 0 O
5N1-W
410 Ln LflW IfLO ) Ln
a-)= n _4-4C a In -4 ii
(A 0. *-;.
Ol 04-1
ZCO0 Otf)
0L ul)) U~LnLn Ln
=~~~ 4)4-4l -4 i
Ms1-
U.S
-. J4
la 0 0 0
CDS 4)
41)14
40 4-3
4.)4
-~4 4J4 4J 4 +bE *
tn . . c0 .0.0 4.0
be.be . E) d Ne- .*9 U9
v- v )4)S_ S_.)u
U 00o1s 1-00 02
co u .0 .- = .C C .C
I fll4)U
W VU u AfA.
-1 41
4o
+3 W, 4-
I.- 0 -' 10
4)3 V MO '04-'
C 4) -
0 m 4J. .4)
2A
S @ L 4-
4) ~ W
~ to.
M=
0.
u )
C'
4. 4)
C
0.
(
C 4- U *'C S_1
u o a)- 4-J 0 (A '4tM 4-
U). 41" o~0 1-CD
-
v fd 31"0 43 '0~
w to 1-MC 1- 9
V9 d) 0 D5
S 0. &_ 44S. -
iLi_ 0 an' , -
5-113
'7U-7
The effect of nonoperating temperature was the next parameter

evaluated. There was an insufficient range of temperature values
represented in the data to determine a nonoperating temperature factor by
the desired analytical techniques. However, the objective of this study
was to develop nonoperating failure rate prediction methodologies which
could be used for any conceivable mission profile (with the exception of
satellites). Therefore, alternate methods were explored to determine an
appropriate temperature factor. In the general discussion of temperature
effects described in Section 4.4, a nonoperating temperature factor was
presented for transistors and diodes. The factor was represented by the
following equation.
1TT=1 1 T P
'NT :exp(-An(y- 2) + (m) )
where
lYNT : nonoperating temperature factor

T = temperature (OK) V
An = temperature coefficient W.
Tm,P = shaping parameters
High temperature storage life test data for transistors were used to
estimate An terms during the transistor nonoperating failure rate model
development. The Tm and P shaping parameter terms were assumed because of
data limitations. Nevertheless, the transistor temperature factor was
based on empirical data, and, therefore, is accurate in the range of 1500C
to 2000C. For other temperatures, accuracy of the proposed transistor
temperature factor is dependent on the accuracy of the assumptions. An
values for transistor styles without data were extrapolated using the MIL-
HDBK-217D relationship for failure rate vs. temperature. It was decided
to extend the extrapolation process to also include diodes as a method to
determine appropriate diode nonoperating temperature factors. This ;--
assumption was justified because of documented similarities (i.e. MIL- -':
HDBK-217D) between the temperature relationships of transistors and
5-114
diodes. The extrapolated series of An values and the assumed values for
Tm and P are presented in Table 5.2.1-1 of Section 5.2.1.
The final phase of the diode nonoperating failure rate model

development process was to normalize the preliminary base failure rates to
correspond to a ground benign environment and a nonoperating temperature
of 250C (i.e. the reference temperature). Additionally, base failure
rate values were extrapolated for diode types without data. Table 5.2.3-9
presents the normalized nonoperating base failure rates, the extrapolated
base failure rate values and 90% confidence limits for zener, thyristor
and microwave diodes. Use of the standard error statistic and the t-
statistic allows for computation of the 90% confidence interval values. V
Because of log transformation, the values are only approximate.
TABLE 5.2.3-9: DIODE NONOPERATING BASE FAILURE RATES
Device Style Xb,.05 Xbl(1) Xb,.95 Xb2(2)
Si, General Purpose -- .000174 -- .000174 , •.

Ge, General Purpose ...--. .000420
Zener/Avalanche .000118 .000402 .00136 .000402
Microwave .000349 .00266 .0205 .00266
Thyristor .000083 .000632 .00485 .000632
Group VIII (3) ...... .00266
NOTES: (1)Xb1 is the observed point estimate base failure rates

(2) Xbj is the observed values complemented with extrapolated -*
va ues.
(3) Group VIII diodes include varactor, PIN, IMPATT, step

recovery, tunnel and gunn devices
5L1
5-115
Normalization of the nonoperating base failure rates was the final

phase of the diode model development process. The proposed model is,
Xp= Xnb wNT -NQ '•NE ,'cyc
where all factors have been previously defined.
The proposed model is presented in Section 5.2.1 and in Appendix A in a

form compatible with MIL-HDBK-217D. Model validation for discrete
semiconductors is presented in Section 5.2.5.
5.2.4 Opto-electronic Semiconductor Model Development
The desired modeling approach utilizing analysis of a large data base

was not feasible for opto-electronic semiconductor devices. No
nonoperating failure rate data were collected for light emitting diodes
(LEDs), isolators, phototransistors, photodiodes or alpha-numerlc
displays. Nevertheless, the objective of this study was to develop
nonoperating failure rate prediction models which could be used for any
proposed electronic equipment in any conceivable mission profile.
Therefore, an alternate modeling approach was taken based on the ratio of
operating to nonoperating failure rate for other semiconductor devices.
Initially, it was believed that sufficient nonoperating reliability

data would be collected for opto-electronic semiconductors, and the
general failure rate modeling approach (described in Section 4.1) would
also be implemented for these devices. As a prerequisite for the planned
data collection and theoretical model development, a list of application
and construction variables were determined for opto-electronic devices.
These variables are presented in Table 5.2.4-1, and represent possible
nonoperating failure rate model parameters. 7- 77,
5-116
S.~
74ýý/ / /M
I -/
TABLE 5.2.4-1: OPTO-ELECTRONIC DEVICE CHARACTERIZATION VARIABLES
I. Device Style
A. Light Emitting Diode (LED)
B. Isolator
1. Photodiode Detector
2. Phototransistor Detector
3. Light Sensitive Resistor ,
C. Phototransistor
D. Photodiode
E. Alpha-numeric Display
II. Complexity t
A. Single Isolator
B. Dual Isolator
III. Quality Level
A. JANTXV
B. JANTX
C. JAN
D. Lower, Hermetic
E. Plastic
IV. Number of Pins
V. Temperature
VII. Number of Equipment Power On/Off Cycles per 103 Nonoperating Hours
5-117
a\
A theoretical model was developed for opto-electronic devices based

primarily on intuitive reliability relationships. Very little documented
information concerning the storage/dormant reliability of these devices
was available. The theoretical model was determined to be a function of
device style, complexity, quality level, number of pins and equipment
power on-off cycling frequency. In the absence of data, the theoretical
model development process was the basis of the proposed nonoperating k. -
failure rate prediction model.
Despite extensive data collection efforts, no data were collected for

opto-electronic semiconductors. Therefore, the anticipated failure
behavior of opto-electronic devices was compared with other semiconductor
devices. Two conclusions were made based on this comparison. First, the
relative effect of device screening was anticipated to be similar for
opto-electronic devices as the effect on transistors and diodes. As a
result of this conclusion, the proposed nonoperating quality factors for
diodes (also applied for transistors) were used to assess the effect of
quality on opto-electronic semiconductors. These proposed nonoperating L
quality factors were presented in Table 5.2.3-6 in the previous section.
The second conclusion was based on the observation that the MIL-HDBK-217D
operating environmental factors were also assumed to accurately predict
the effects of environmental stress during nonoperating periods. A
similar assumption was made for opto-electronic devices. The MIL-HDBK-
217D cpto-electronic environmental factors were also applied for
nonoperating failure rate prediction purposes, and were presented in Table
5.2.1-4 in Section 5.2.1.
The theoretical model for opto-electronic semiconductor devices was -

also assumed to be a function of device style, complexity, temperature and
equipment power on-off cycling. It would have been extremely presumptuous
to assume relationships for each of these variables without additional
justification (i.e. observed data). A proposed nonoperating failure rate
prediction model based entirely on assumptions must be considered %
approximate at best. Therefore, the only additional factor included in
the proposed model was a nonoperating base failure rate as a function of
5-118
device style. Additional factors would only serve to complicate the •..-.
proposed model without resulting in any significant improvement in model '•"
accuracy. The proposed model is therefore represented by the following• -,i'ii
equati on. •'•..
Xoe = Xnb 'TNQ WINE
where •}
X'oe =opto-electronic semiconductor predicted nonoperating failure • ;LI
=f(device style)
lINQ -- assumned nonoperating quality factor •'
•NE = nonoperating environmental factor .• •
Determination of numier~cal for the opto-electronic base

quantities ... ,•
f allure rate was conducted in two phases. The first phase was the • ... ,
selection of device style categories. It was decided to estimate separate L ::
base failure rate values for (1) I.EDs, (2) single isolators, (3) dual )i;!:•
isolators, (4).phototransistors, (5) photodiodes, and (6) alpha-numierlc •...
displays. There are obviously other variables within each of these ••, -
categories which effect nonoperating failure rate. However, no finer ;':':
division was justified because of the lack of quantitative information. ••;ii
The second phase of the nonoperating base failure rate determination
proces was to caclt naeaenonoperatin to oeaigfiuert
ratio based on the proposed nonoperating failure rate prediction models
f or transistors and diodes, and the corresponding sections of MIL-HDBK- '". !i ilil
2170. It was found the average ratio for other semiconductor types was .<
equal to 1:8. Thus, this adjustment factor was applied to the operating *••:
failure rate prediction model values. The resultant nonoperating base :
failure rates are presented in Table 5.2.4-2. L
5-119.
TABLE 5.2.4-2: OPTO-ELECTRONIC NONOPERATING BASE FAILURE RATES
Device Style Anb (Failures/10 6

nonoperating hours)
Single LED .000163
Single Isolator .000699
Dual Isolator .000885
Phototransi stor .000375
Photodiode .000275
Alpha-numeric Display .000254
It is emphasized that the proposed nonoperating failure rate

prediction model for opto-electronic semiconductors is based solely
on
assumptions, and therefore must be considered approximate. However, a
viable approach was taken to determine numerical values for the proposed
model. It is strongly recommended that the accuracy of the proposed model
is evaluated when nonoperating failure rate data becomes
available for
opto-electronic semiconductors.
5.2.5 Model Validation

r
The proposed discrete semiconductor nonoperating failure rate -
prediction models were next subjected to an extreme case analysis
and a
model evaluation process with observed data. The extreme case analysis
consisted of computing nonoperating failure rate predictions with
the
proposed models for parameters beyond the ranges found in the data.
The
intent of the extreme case analysis was to identify any set of conditions
which cause the proposed models to approach infinity, or predict
intuitively inconsistent results. The proposed models were then tested
with observed data which had been withheld from the model development
process. Nonoperating failure rate data for discrete semiconductors had
been collected from AFCIQ. However, only the high temperature storage
life test data from AFCIQ were used for model development purposes.
The extreme case analyses did not indicate any deficiencies with the
proposed models. At extremely high levels of equipment power cyclino,
the
5-120
"WINWN-•'
nonoperating failure rate become proportional to the equipment power

cycling frequency. This was considered to ý,e an intuitively satisfactory
"result because, at high cycling frequencies, true storage related failures
would be negligible in comparison to the number of failures induced by
power on-off cycling. At extremely low cycling frequencies, the proposed
models would predict nonoperating failure rates independent of equipment
power cycling frequency. This was also considered to be an intuitively
appealing result.
The field experience nonoperating failure rate data from AFCIQ were
not used in the model development process because vital information such
as equipment power cycling frequency and part screen class could not be
precisely determined. However, these parameters were estimated and the
data were used for model validation purposes. Tables 5.2.5-1 and 5.2.5-2
presents the model validation data for transistors and diodes
respectively.
On average, the model validation process indicated that the proposed

models were extremely accurate. However, individual data entries varied
substantially from the predicted values. The geometric mean of the ratio
of observed to predicted nonoperating failure rate was equal to 1.03 for
those data entries with either observed failures or sufficient part hours
to estimate a failure rate without failures. When the two most divergent
data entries (numbers 4 and 13 in Table 5.2.5-2) were deleted, the
"geometric mean of the ratio became equal to 1.16. For transistors alone,
the ratio of observed to predicted nonoperating failure rate was equal to
0.81. The geometric mean of the diode data, except entries 4 and 13, was
- equal to 1.35. These ratios were very close to unity, and this
observation was considered to be indicative of accurate nonoperating
failure rate prediction models.
'1
5-121
Z . .- ..
TABLE 5.2.5-1: TRANSISTOR MODEL VALIDATION DATA
Part
Hours Temp. Equiv.
No. Failures (xj0 6 ) Style (oC) Quality X.0 5 )-.95 ýpre(1)
"1 6 1900 Si,PNP 20-30 JAN .00138 .00316 .0062 .00567

2 0 18 Si,PNP 25 JAN -- (3) .166 .00567
3 0 38 Si,PNP 25 JAN -- (3) .0788 .00567
4 0 5 Si,PNP 25 JAN -- (3) .599 .00567
5 3 1100 Si,PNP 20-30 JAN .00075 .00273 .0071 .00567
*1 6 0 72 Si,NPN 25-30 JAN -- (3) .0416 .00567
S7 0 42 Si,NPN 25-30 JAN -- (3) .0713 .00567
8 1 30 Si,NPN 25 JAN .00171 .03333 .158 .00567
9 4 2800 Si,NPN 20-30 JAN .00049 .00143 .0033 .00567
10 1 1253 Si,NPN (2) JANTX .00004 .00080 .0038 .00157,
2 11 8 1000 FET 20-30 JAN .00398 .00800 .0144 .00565
* 12 0 28 FET 25 JAN -- (3) .107 .00565
13 0 6 FET 25 JAN -- (3) .499 .00565
"14 0 6 FET 25 JAN -- (3) .499 .00565 F
15 0 96 FET 5-30 JAN -- (3) .0312 .00565
16 0 5 Unijun. 25 JAN -- (3) .599 .00565
STotals 23 8422
Notes: 1) X predicted nonoperating failure rate. A cycling rate
0.114 cy.les/10 3 nonoperating hours was assumed.
2) Temperature was unknown for this data entry. Temperature

i • assumed to be 250C.,
3) There were insufficient part hours for these data entries to
estimate a failure rate without observed failures.
5IL
5-122 ""
~ ~ A' / -
TABLE 5.2.5-2: DIODE MODEL VALIDATION DATA Xk
Part
Hours Temp. Equiv.
No. Failures (xj0 6 ) Style (oC) Quality X.05 Xo X.95 Xpre(1)
1 20 779 Si,GP (2) Plastic .01702 .02567 .0371 .01539

2 5 675 Si,GP (2) Plastic .00292 .00741 .0156 .01539
3 2 136 Si,GP 20-30 JAN .00261 .01471 .0463 .00241
4 2 16000 Si,GP 20-30 JAN <.00001 .00001 .0004 .00241
5 4 906 Si,GP (2) JAN .00151 .00442 .0101 .00241
6 0 41 Si,GP 25 JAN -- (5) .0730 .00241
7 1 247 Si,GP 25 JAN .00021 .00405 .0192 .00241
8 0 36 Si,GP 25 JAN -- (5) .0832 .00241
9 1 400 Si,GP 5-30 JAN .00013 .00250 .0119 .00241
10 1 1200 Si,GP 0-30 JAN .00004 .00083 .0040 .00241
11(3) 0 3422 Si,GP (2) JANTX -- .00029 .0009 .00067
12 5 47 Zener (2) Plastic .04191 .1064 .2237 .03622
13 2 14 Zener 20-30 JAN .02539 .1429 .4496 .00567
14 1 130 Zener (2) JAN .00039 .00769 .0365 .00567
15 0 12 Zener 25 JAN -- (5) .2496 .00567
16 0 24 Zener 25 JAN -- (5) .1249 .00567
17 0 150 Zener 5-30 JAN -- (5) .0200 .00567
18 0 624 Zener 0-30 JAN -- .00160 .0048 .00567
19(3) 1 175 Zener (2) JANTX .00029 .00571 .0271 .00157
20 15 104 Si,GP (2) Plastic .0889 .1442 .2205 .01539
21 0 30 Si,GP 25 JAN -- (5) .0998 .00241
22 0 67 Si,GP 5-30 JAN -- (5) .0447 .00241
23 0 72 Si,GP (2) JANTX -- (5) .0416 .00067
24(4) 16 10 Thyr. (2) Plastic (4) (4) (4) .0570
25 0 2 Thyr. (2) Plastic -- (5) 1.498 .0570
26 0 19 Thyr. 20-30 JAN -- (5) .0365 .00893
27 1 130 Thyr. 20-30 JAN .00039 .00769 .0365 .00893
28 0 16 Thyr. 0-30 JAN -- (5) .872 .00893
Totals 77 25468
NOTES: 1) Xr predicted nonoperating failure rate. A cycling rate

of 0.0114 cycles/l03 hours was assumed.
2) Temperature was unknown for these data entries. Temperature

assumed to be 250C.
3) The application environment was unknown for these data
entries. An environment of ground fixed was assumed.
4) This part was over-stressed.
5) There were insufficient part hours for these data entries to

estimate a failure rate without observed failures.
5-123
Inspection of the diode model validation data (Table 5.2.5-2) gives a

S)good
indication of the inherent variability which exists for nonoperating
failure rates. It is not unusual for there to be order of magnitude
differences for similar devices in similar applications. For this reason,
incorrect conclusions can be made if insufficient data resources were
available for analysis. This was one reason such a large emphasis was
placed on was
validation the an-integral
data collection
part of effort in this
the overall model study, and why
development model
process.
zK
5-124
-Xi
5.3 Resistors
5.3.1 Proposed Resistor Nonoperating Failure Rate Prediction Model
prediction model for resistors. The proposed model is presented in
Appendix A in a form compatible with MIL-HDBK-217D. The proposed model

is:
Xp = Xnb INQ INE Icyc
where .
Xp = predicted resistor nonoperating failure rate

6
= .000063, fixed composition (RC, RCR)
- .00010, fixed film (RN, RD, RL, RLR, RNR)
- .00043, film network (RZ) •,. •
- .00057, fixed wire wound (RW, RB, RE, RBR, RWR, RER)
= .00099, variable wirewound trimmer (RT, RTR)
= .0052, precision, semiprecision or power variable wirewound
(RR, RA, RK, RP)
- .0052, variable non-wirewound (RV, RJ, RVC, RQ, RJR)
- .0027, thermistor (RTH)
tNQ nonoperating quality factor
0.15, S0.2,'
=0.28, R _!
--0.52, P
: 1.0, M
- 2.4, MIL-SPEC
-4.4, lower J
7NE nonoperating environmental factor (see Table 5.3.1-1)
5-125
TABLE 5.3.1-1: RESISTOR NONOPERATING ENVIRONMENTAL FACTORS (liNE)
Env. Comp. Film Network WW Therm. Var. WW Var. non-WW
GB 1 1 1 1 11 1
GF 2.9 2.4 2.4 2.1 4.8 2.5 2.5
GM 8.3 8.3 7.8 8.8 23 13 13
Mp 8.5 9.9 8.8 11 17 .18 19
NSB 4 4.7 4.2 5 7.9 7 7
NS 5.2 4.9 4.7 5 14 7 7
NU 12 15 14 15 17 14(3) 17
NH 13 16 14 17 25 29 29
NUU 14 17 15 18 27 31 31 yjý
ARW 19 22 19 24 33 41 41
I
AIC 3 3 2.5 4.3 4.3 5.5 12
AIT 3.5 4.5 3 7.3 7.7 6.6 16
AIB 5 6.8 6.5 12 19 9.5 24
AIA 3.5 5.8 6 9.7 15 8.6 22
AIF 6.5 9.5 9 13 38 14 33
AUC 5 7.5 6 10 4.6 6.5(3) 19
AUT 7 11 6.5 13 8.6 9 (3) 25
AUB 10 18 15 23 21 18 (3) 37
AUA 7 13 15 18 17 13 (3) 32
AUF 15 23 20 28 42 20 (3) 52
SF (2) (2) (2) (2) (2) (2) (2) R •!
MFF 8.6 10 8.9 11 15 20 (3) 19

MFA 13 14 12 16 21 28 (3) 26
USL 25 30 26 33 49 58 (3) 56
ML 29 35 30 38 51 66 (3) 64 'S
CL 490 590 510 610 950 1100 (3) 1100
NOTES: 1) WW Wirewound
2) Space flight environment was not considered in this study
3) These factors are not applicable for semiprecision wirewound

or high power wirewound variable resistors. ,,
5-126
"'.-MT
ixA.,z, It
icyc = equipment power on-off cycling factor

= 1 + .063(Nc)
where

hours
5.3.2 Model Development

successfully implemented for resistors. An assumed theoretical model was
supplemented by analysis of the available resistor nonoperating
reliability data to derive the failure rate prediction model. The model
was determined to be a function of 'vice style, quality level,
environment and equipment power on-off cycling frequency. The proposed
nonoperating failure rate prediction model fr- resietors is presented in
Appendix A in a form compatible with MTH '%XK-217D. This section
describes the model development process.
The first step in the model development process was to identify

application and construction variables which properly characterize
resistors in a nonoperating environment. Table 5.3.2-1 presents a list of
the part characterization variables for resistors. These variables
represent possible failure rate model input parameters which were analyzed
in greater depth using statistical methods. The part characterization
variables were determined whenever possible for all collected data. If

sufficient detail could not be determined for an individual data entry,
then it was removed from the model development process.
The general nonoperating reliability concepts presented in Section 4.0

were complemented by a study of resistor nonoperating failure mechanisms
and physics of failure information to determine a theoretical model for
resistors., The theoretical model assumes a nonlinear failure rate
equation which is constant with respect to time. The theoretical model
5-127
'tZ:3V
/
TABLE 5.3.2-1: RESISTOR PART CHARACTERIZATION
I. Device Style
A. Fixed
1. Composition (RC, RCR)

2. Film (RN, RD, RL, kLR, RNR)
3. Network, Film (RZ)
4. Wirewound (RW, RB, RE, RBR, RWR, RER)
F
B. Variable
1. Non-wirewound (RV, RJ, RVC, RQ, RJR)
2. Wirewound (RA, RP, RR, RT, RK, RTR)
C. Thermistor (RTH)
1. Bead
2.. Disk
3. Rod
II. Resistance (ohms)
III. Rated Power (watts)V
IV. Quality Level
A. S
B. P
C. R
D.-
E.
F.
M
Mi1-Spec
lower
r
V. Number of Potentiometer Taps
VI. Construction Class
A. Enclosed 3,
B. Unenclosed
VII. Number of Pins (style RZ only)
VIII. Number of Resistors (style RZ only) L
IX. Temperature
A. Rated
B. Actual
XI. Number of Power On/Off Cycles per 103 Nonoperating Hours
5-128
F-~:.
developed for resistors is described by the following equation (f denotes • :

a function). *tt
"Xr= f(device style, temp., R, P,)TNQITNEITcyc
where
Xr = predicted resistor nonoperating failure rate (failures/j06 hrs)

R = resistance (ohms)
P = rated power (watts)
'TNQ = nonoperating quality factor, based on device screening
-•NE = nonoperating environmental factor
Ircyc = equipment power on-off cycling factor 1 + KI(Nc)
K1 = constant
Nc = equipment power on-off cycling rate (cycles/lO3 nonop. hrs.)
Presentation of the theoretical model does not imply that each of

these variables was significant by anything other than theoretical
grounds. Statistical methods, where appropriate, were used as a basis for
decisions regarding significance of variables. The theoretical model was
intended to define the relationship of the independent variables to each
other.
The summarized resistor nonoperating reliability data collected in

support of this study is presented in Table 5.3.2-2. The data
collectively consists of 413 individual data records, 34 observed failures
and 113,119 x 106 part hours.
Initially the data was evaluated with each data record separate (i.e.
no data records were merged). Point estimate failure rates were computed
for data records with observed failures. The methods described in Section
3.3 were applied to determine which zero failure data records had
sufficient part hours to be included in the analysis, and to estimate
failure rate for those records. It was found that 43 data records either
5-129
* %
TABLE 5.3.2-2: RESISTOR NONOPERATING FAILURE RATE DATA
Data Parts
Equipment/Source Records Style Fai'lures Hours (x106 )
Sprint/MICOM 3 Wirewound 0 31.0
Sprint/MICOM 4 Composition 0 4300.0
Sprint/MICOM 1 Film 0 147.1
Sparrow/MICOM 116 Composition 2 4478.9 -
Sparrow/MICOM 173 Film 0 5346.6

Sparrow/MICOM 13 Wirewound 1 268.0
Sparrow/MiCOM 6 Var., Non-WW 1 102.1
Sparrow/MICOM 1 Thermistor 0 13.0
TOW/MICOM 6 Film 0 609.7
TOW/MICOM 7 Composition 0 141.9
TOW/MICOM 1 Wirewound 0 42.1
Lance/MICOM 4 Film 0 122.0
Lance/MICOM 4 Composition 0 239.3
Lance/MICOM 2 Wirewound 0 23.4
Lance/MICOM 1 Var., WW 0 2.2 0
Lance/MICOM 1 Thermistor 0 1.1
Lance/MICOM 1 Film Ntwk. 0 1.1
Hawk/MICOM 8 Film 0 38576.5
Hawk/MICOM 4 Wirewound 11225.1
Hawk/MICQM 1 Thermistor 0 51.0L
Hawk/MICOM 3 Var.,1 Non-WW 0 102.1
Hawk/MICOM 1 Var., WW 0 17.0
Maverick/MICOM 8 Film 0 5806.4
Maverick/MICOM 4 Wirewound 0 782.4
Maverick/MICOM 7 Composition 0 3486.5
Maverick/MICOM 3 Var., Non-WW 0 370.6
Maverick/MICOM 2 Var., WW 1 514.8
Martin Marietta 2 Composition 0 11548.1
Martin Marietta 4 Film 1 15937.5
Martin Marietta 3 Thermistor 3 102.0
Martin Marietta 6 Wirewound 2 4339.0
Martin Marietta 2 Var., Non-WW 1 24.1
Martin Marietta 1 Var., WW 0 1.5
PRC 1 Composition 0 159.0
PRC 1 Film 0 192.0
F-16 HUD/RIW 3 Film 18 10328.2
F-16 HUD/RIW 2 Wirewound 2 570.9
F-16 HUD/RIW 2 Composition 0 2893.5
F-16 HUD/RIW I Var., WW4 1 220.6 0"
Totals 413 34 113,118.94
5-130
had observed failures or had sufficient part hours to estimate a failure

rate without observed failures.

-- restricted because of the nature of the available data The following "-
..
attributes of the collected nonoperating resistor data were identified

from the preliminary data analysis.
1) All collected data were from a ground based environment.

Therefore, there was not a sufficient range of environments to
develop a series of environmental factors based solely on the
data. k
2) There was an insufficient range of ambient temperatures to study
3) The inherent failure rate of resistors in storage/dormant

applications is very low. Therefore, the number of data records
with observed failures and the number of failures per data record
were naturally limited. Precise estimates of field failure rates
and quantification of failure rate model parameters can not be
made without observed failures. Thus, the number of variables
which could be analyzed empirically was also limited.
The theoretical model was further scrutinized to determine the

relative effect of the independent variables so that decisions could be
made regarding the selection of variables. Failure mechanism acceleration
factors were studied to aid in the variable evaluation process.

that resistor device style, quality level, application environment and
equipment power on-off cycling frequency are the dominant variables V
effecting nonoperating failure rate. These variables, with the exception

of application environment, were further analyzed using Statistical
techniques. It was also determined that resistance level, and rated power
were significant variables for assessing operating resistor failure rate,
but tieoretically had little effect on nonoperating failure rate. These
two variables were not included in subsequent analyses.
5-131 ,'
+\
It was assumed that application environment was a very important

parameter effecting resistor nonoperating failure rate. Therefore it was
essential that a proposed resistor nonoperating failure rate prediction
model include an appropriate environmental factor to properly discriminate
against known failure accelerating f-:tors. All data collected in support
of this study was for ground based environments, and therefore
environmental factors could not be developed empirically. This was a
major obstacle to the successful completion of this study and is discussed
in Section 4.5 of this report. Section 4.5 also proposes a method to
develop appropriate environmental factors based on a comparison of
operating and nonoperating failure mechanisms, and a study of the r
documented MIL-HDBK-217D operating environmental factors. This method was
applied to resistor nonoperating failure rate modeling after statistically
analyzing the other variables.
The fact that temperature could not be empirically analyzed was al;o a
problem. However, after careful consideration of many factors, it was
decided not to include temperature in the failure rate prediction model.
This decision was based on three reasons or observations. First., a
temperature relationship could not be determined from the data.
Therefore, any proposed temperature factor would be based on assumptions,
which introduce inaccuracies. The second reason was that no theoretical
relationship for resistor nonoperating failure rate vs. temperature could
be located in the literature. In addition, there was no basis for
assumptions that the nonoperating failure rate behaves similarly to the
operating rate with respect to temperature. The stresses on the part are
necessarily different due to the absence of an applied current. Therefore
an approach of that nature was determined to be invalid. The third reason
was that the temperature dependence of resistor nonoperating failure rate
was believed to be significantly less than the temperature dependence for
microcircuits and discrete semiconductors. The proposed nonoperating
failure rate for these devices (described in Sections 5.1 and 5.2 of this
report) does include a temperature factor. Therefore, the absence of a
temperature factor for low failure rate devices, such as resistors, would
rot siMnificantly effect equipment-level failure rate predictions. It also
5-132
should be noted that the very high temperatures during operation are not
observed during nonoperating periods because of the absence of internal
heat generation and the associated cooling problems. In conclusion, it is
emphasized that direct numerical analysis of the relationship of resistor
nonoperating failure rate and temperature would have been desirable.
However, it is also emphasized that deletion of ambient temperature from
the model development process was believed to have little effect on the
accuracy of the proposed resistor nonoperating failure rate prediction
model.
Matrices of "dummy variables" (0 or 1) were defined for device style PEP`

and quality level so that correlation coefficient analysis .and regression i,'
analysis could be applied to the data. These analyses require
quantitative variables, whereas device style and quality level are
qualitative variables. The matrix for device style is given in Table
5.3.2-3. No distinction was made between the different styles of
variable resistors. The specific styles of variable resistors with data
were wirewound precision, non-wirewound trimmer, wirewound trimmer, and
non-wirewound film. This was done because of the limited amount of data
(i.e. only 4 variable resistor observed failures) and because the failure
rate difference between specific variable resistor types was assumed to
much less than the difference between variable and fixed resistors. The
difference in failure rate between the specific styles of variable
resistors were analyzed at a later stage of the model development process.
The matrix for quality level was simply a "0" for high-reliability parts
and a "1" for mil-spec quality parts. This variable was designated as QI.
The quality matrix was not defined to accommodate lower quality resistors
because no data was available for lower quality resistors.
The data were then merged according to equipment, device style, and
quality level. The data were merged by summing observed failures and
summing part hours. For example, all data for B screen class, fixed
composition resistors in the Hawk missile were merged. The correlation
coefficient analysis indicated no large correlation between independent
variables for the merged data set. K
5-133
-p ~pp -p~
¶:~-Vpj~.'-~ * p ~*
. **• .j "*'J*
TABLE 5.3.2-3: RESISTOR STYLE VARIABLE MATRIX
Device Style S1 S2 S3 S4
Variable 0 0 0 0
Fixed, Wirewound 1 0 0 0
Fixed, Film 0 1 0 0
Fixed, Composition 0 0 i 0
Thermistor 0 0 0 1
The theoretical resistor model was nonlinear due to the assumed

relationship for equipment power cycling frequency. Thus, linear
regression analysis could not be directly applied. An iterative approach
to nonlinear regression was described in Section 4.3. As suggested in
Section 4.3, equipment power cycling frequency was temporarily treated as
a qualitative variable. Constant multiplicative values were computed by
regression for three distinct cycling rate categories. Then a two-
dimensional regression was performed with the cycling rate coefficients
from the initial regression as the dependent variable, and the mean
cycling rate value for each category as the independent variable. The
result would be an expression for equipment cycling factor of the desired
form (i.e. cycling factor = mx + b). As the iterative process continued,
the equipment power cycling factor expression would be assumed exact, and
the coefficients for all other variables recalculated. Then, those
coefficients would be assumed to he exact and the equipment cycling factor
recalculated. The iterative process would continued untii observed r
changes were negligible.
Three qualitative "dummy" variables were defined (cycl, cyc2, and

cyc3) to represent four distinct equipment power cycling rate categories,
which arp,
o cycling rates less than one power cycle every two years (cyci,
cyc2, cyc3 = 0,0,0),
o cycling rates between one cycle Der year and one cycle every two
years (cycl, cyc2, cyc3 = 1,0,0),
o cycling rates greater than one cycle per year (cycl, cyc2, cyc3
0,1,0), and
5-134
o unknown cycling rates (cycl, cyc2, cyc3 0,0,1).
The first three categories were designated as (1) low, (2) intermediate,
and (3) high cycling frequencies, respectively.
Results of the initial regression analysis are given in Table 5.3.2-4.

The dependent variable was ln(failure rate). The independent variables
were S1, S2, S3, S4 , Q1, cycl, cyc2, and cyc3, which were previously
defined. Device style variables SI, S2, S3, quality variable Q1 and
equipment power cycling variable cyc2 were significant at a 90% confidence
limit. No other variables were significant at a 70% confidence limit.
The R-squared value for the regression was 0.85. That is, 85% of the
variability in the observed data can be explained by the regression
solution. Interpretation of the regression results are:
1) The nonoperating failure rates for fixed composition, fixed film,

and fixed wirewound resistors were significantly different, with a
90%' confidence, from the variable resistor and thermistor failure
rate, as would be expected.
2) The difference in failure rate between variable resistors and

thermistors was not significant at a 70% confidence limit.
3) As expected, it was determined with 90% confidence that resistor
screening impacts the field nonoperating failure rate.
Interpretation of the regression results was that the failure rate
of Mil-spec quality resistors is 4.8 times higher than resistors
screened to the established reliability specification.
4) The cyc2 equipment power cycling variable was significant with 90%
confidence. This substantiated the assumption that equipment
power cycling is an important variable for predicting nonoperating
failure rate. The significance of cyc2 also served as a rationale
to proceed with the cycling factor analysis. The failure rate
difference between low, intermediate and unknown cycling
frequencies was not significant even at a 70% confidence.
Regression solutions for confidence limits lower than 70% were

to further investigate and
comouted
variable resistor failure rate. Atthe50%difference
confidence,between thermistor
the difference became
significant (i.e. the f-ratio for the S4 variable exceeded the critical-f
5-135
j.. ..
A.7
value). Interpretation of the results were that the ratio of thermistor

to variable resistor failure rate was equal to 2.0. However, after
careful consideration, it was decided that there was insufficient
confidence to propose unique values for variable resistors and
thermistors.
TABLE 5.3.2-4: RESISTOR INITIAL REGRESSION RESULTS
standard confidence
variable coefficient error f-ratio limit
S1 -2.386 0.637 14.04 .90
S2 -3.957 0.584 45.91 .90

S3 -4,257 0.622 46.80 .90
Q1 1.560 0.512 9.28 .90
cyc2 1.294 0.622 4.33 .90
"bo
-5.014 ......
'•- :The coefficients given in Table 5.3.2-4 were the results of a

failure rate. Transforming the regression results into an equation where
failure rate (as opposed to log of failure rate) is the dependent variable
results in the following preliminary multiplicative model:
Xr nb"NQ cyc
where
/Xr =predicted resistor nonoperating failure rate (failures/10 6

hours)
Xnb = base failure rate, preliminary Li
Sexp(-5.014 - 2.386(S$) - 3.957(S 2 ) - 4.257(S 3 ))
= .00611, fixed wirewound resistors
= .000127, fixed film resistors •
= .0000941, fixed composition resistors
= .00664, variable resistors
5-136
= .00664, thermistors
'rNQ = quality factor

= exp(1.560(Q1))
= 1.0, Hi-rel
"=4.76, Mil-spec
kcyc = equipment power cycling factor (preliminary)
= exp(1.294(cyc2))
= 1.0, cycling rate < .057 cycles/lO3 hr.
= 1.0, .057 < cycling rate < .114 cycles/10 3 hr.
= 3.65, cycling rate 2 .114 cycles/lO3 hr.
C = residual
The variables S1, S2 , S3, Q1 and cyc2 have been previously defined. It is
emphasized that the above equation represents a preliminary model, which
is only applicable to ground based environments.
Two hypothetical reasons were determined as to why the intermediate

and low equipment power cycling frequency data were statistically
indistinguishable. The first reason was that "true" storage related
failures (as opposed to failures induced by the power-on cycle) dominate
"the total failure rate for cycling rates less than .114 cycles/lO3 hr
(i.e. powered-up less than once a year). Therefore, the anticipated
1 difference in failure rate would potentially be less than the inherent
"~ variability in the data, and therefore would not be detected. The second
hypothetical reason was that the merged data set included only one data
entry in the intermediate range of cycling frequencies. Therefore, if the
data from that one source was biased, then the results would be effected.
It is also noted that the unknown cycling frequency data entries were
statistically indistinguishable from the data entries with low and
intermediate power cycling frequencies. This was considered to be an
dencouraging sign. Although the cycling frequency was unknown for these
data entries, the data was from missile storage applications where it was
known that the cycling was relatively infrequent.
' 5-137 n
The next step in the model development process was to perform the
iterative regression analyses to accommodate the assumed non-linear model
form. It has been documented (Reference 42) that the effect of equipment
power on-off cycling is best represented by an equation of the following
f orm:
'rcyc 1 + K1 (cycles/10 3 hr.)
As previously described, the iterative process began by performing a two-

dimensional regression analysis for "equipment power cycling factor"
versus average cycling rate for the three previously defined cycling rate
categories. The average cycling rate for resistors exposed to 'low
equipment cycling rates was .042 cycles/lO3 hrs. The data in this
category was primarily from the Sparrow and Hawk missile program. A
regression coefficient was also computed for the intermediate cycling rate
category, so that the method could be properly applied. The average
cycling rate for the intermediate category was .079 cycles/10 3 hr. This
* data was primarily from the TOW missile program. The average cycling rate
for resistors exposed to high equipment cycling frequencies was 40,12
cycles/10 3 hr. This data was from the F-16 HUD RIW data. The data from
sources with unknown cycling frequencies were not included in this
regression. The data entries were weighted by the number of observations
per category. An explanation of weighted regression is included in
Section 3.2. The first iteration for equipment power cycling factor
resulted in the following equation.
K21Tcyc = 0.977 + .0614(cycles/10 3 hr.)

'Tcyc = 1 + .0628(cycles/10 3 hr.)
where
K2 = normalization constant
= 1/0.977 = 1.023
5-138
44.
For resistors, only one iteration was required. The preliminary model
at this stage of the model development process is given by the following
equation:
Xr Xnb 1TNQ 'Tcyc
where
Xr = predicted resistor nonoperating failure-rate (failures/10 6 hr.)

Xnb = base failure rate, preliminary
= .000617, fixed wirewound resistors
= .000128, fixed film resistors
= .0000953, fixed composition resistors
= .00671, variable resistors
= .00671, thermistors
= quality S•NQ
factor
= 1.0, Hi-rel
= 4.69, Mil-spec
S•cyc
= equipment power cycling factor
= 1 + .0628(cycles/10 3 hr.)
The magnitude of the nonoperating quality factor for Mil-spec

resistors was small in comparison to the ratio of Mil-spec to established
reliability (i.e. S, R, P and M) operating quality factors from MIL-HDBK-
2170. In other words; screening apparently has less effect on
nonoperating failure rate than operating failure rate. It was
hypothesized that this trend was due to a significantly slower rate of
failure for "infant mortality" failures (i.e. weak devices remaining and
subsequently failing after the screen) during nonoperating periods. This
trend was noted for every electronic component generic family, and is
discussed in Section 4.6 of this report.
It was decided to derive unique nonoperating quality factor values for

the established reliability (ER) quality levels defined in the ER
specifications. A complete series of quality factors were required to
5439 j.
'/'.÷
provide the mndels with maximum utility and to properly discriminate

against known quality related factors. The available nonoperating failure
rate for ER (or Hi-rel) resistors did not specify the particular
established reliability level. In general, the average quality level was
R or P. It was assumed that the average ER level for the collected data
was P to determine a complete series of nonoperating quality factors. The
assumption was made that nonoperating quality factors would follow a
geometric progression. This was based on the observation that the MIL-
HDBK-217D operating quality factors follow a similar progression. Table
5.3.2-5 presents point estimate nonoperating quality factors, upper and
lower 90% confidence interval values, extrapolated nonoperating quality
factors for S, R, M and lower qualities, and a series of quality factors
normalized to an M factor equal to one.
TABLE 5.3.2-5: RESISTOR NONOPERATING QUALITY FACTORS
Quality Level 'NQ,.0 5 TNQ1(1) 'NQ,.95 'NQ2(2) 'NQ3(3)

S ...... .28 0.15
R -- .. .55 0.28
P -- 1.0 -- 1.0 0.52
M -- -- -- 1.9 1.0
Mil-spec 2.2 4.7 10.1 4.7 2.4
lower -- -- -- 8.6 4.4
NOTES: (1) INQ1 Is the observed point estimate quality factor.
(2) 'NQ2 is the pointvalues

by-extrapolated estimate
for quality
S, R, M factors
and lowersupplemented
qua*'ity
levels.
(3) fNQ3 is the quality factor normalized to M quality

factor equal to one.
The next task in the model development process was to apply the ..
methods presented in Section 4.5 to determine nonoperating environmental
factors. This method assumes that a series of nonoperating environmental
factors can be derived from the MIL-HDBK-217D operating environmental
factors. A conversion algorithm was determined based on a comparison of
5-140
failure mechanism accelerating factors, and differences in average

temperature. In general, if the distribution of failure mechanism
accelerating factors were equivalent for operating and nonoperating
conditions, then the one series of environmental factors would apply.
Conversely, if the failure mechanism distributions were significantly
different, then methods were determined to estimate the magnitude of the
difference, if any. Since every MIL-HDBK-217D resistor model has a unique
temperature factor, no numerical adjustment was required to account for
the difference in *average temperature between operating and nonoperating
applications. An adjustment was required, however, for the thermistor
nonoperating environmental factor.
The proposed resistor nonoperating environmental factors were

presented in Table 5.3.1-1 in the previous section. It was determined
that a minimum of seven different series of environmental factors were
required to properly assess the effects of environmental stress on
nonoperating failure rate. The results of the nonoperating environmental
factor development were that the thermistor environmental factors required
an adjustment to account for temperature differences. The appropriate
temperature adjustment factor was based on anticipated differences between
operating and nonoperating applications of up to 240 C for airborne
uninhabited environments. The other MIL-HOBK-217D resistor environmental
factors approximate the effects of environmental stress on resistor
nonoperating failure rate. MIL-HDBK-217D includes 15 separate series of
environmental factors for resistors. The decision was made to group and
average the environmental factors for similar resistor styles. This was
done for several reasons. The primary reason was that the nonoperating
environmental factor development process was inexact. Therefore, the
relatively small differences in environmental factors between similar
styles were not justified for nonoperating failure rate prediction
purposes.
Tables 5.3.2-6 through mode/mechanism and

5.3.2-11 show failure
failure acceleration factor distributions for resistors. Accurate failure
mode/mechanism information was difficult to obtain for operating
5-141
S....
t ••. • ,.-.• . , .> . . . I,--,-,•7
.. - . ...... . . • ..-- ,., . . •. • - . ,.. . .. • ... - • ... .. ,, •- :
TABLE 5.3.2-6: COMPOSITION RESISTOR FAILURE MECHANISM DISTRIBUTION
Failure Failure Operating Nonoperating

Accelerating Distribution Distribution
Mode Mechanism Factors()()
resistance moisture moisture, temp. 35-55 45-65

drift intrusion
non-uniform voltage/current, 5-25
comp. material 5-20
temp.,
contaminants voltage/current, 5-25 5-20
temp.
open lead defects moisture, temp., 15-35 15-30
voltage/current
ii' TABLE 5.3.2-7: FILM RESISTOR FAILURE MECHANISM DISTRIBUTION
Failure Failure Operating Nonoperating

Accelerating Distribution Distribution
Mode Mechanism Factors() )
resistance
drift
moisture
ingression
moisture,temp.,
contamination.
20-40 30-50 [
substrate moisture, temp., 15-35
defects 5-25
voltage/current
film moisture, temp., 15-35
imperfections 5-25
voltage/rýurrent
open lead termina- shocK, vibration, 1-20 1-20
tion temp., Voltage!
current
film material moisture, temp., 1-20
damage 1-20
voltage/current
TABLE 5.3.2-8: WIREWOUND RESISTOR FAILURE MECHANISM DISTRIBUTION
Operating Nonoperating
Failure Failure Accelerating Distribution Distribution
Mode Mechanism Factors (%) (%)
resistance wire im- voltage/ 10-35 1-10
drift perfection current, temp.
wire insula- voltage/ 15-40 1-15
tion flow current, temp.
corrosion temp., humidity 10-35 35-55
open lead defect shock, vibration, 1-20 1-20
voltage/current
wire im- voltage/current 1-20 1-10
perfection
corrosion temp., humidity 1-20 15-40
short intrawinding temp., voltage/ 1-10 1-10
insulation current
breakdown
TABLE 5.3.2-9: VARIABLE WIREWOUND RESISTOR FAILURE MECHANISM DISTRIBUTION
Mode Mechanism Factor (%) (%)
resistance contamination temp., conta- 15-35 20-45
drift mination
noise moisture, temp. 1-18 5-40
short insulation
breakdown
moisture, temp.,
voltage/current
5-25 5-15 .
contamination contamination, 1-12 5-20
bridging moisture, temp.
open wiper arm mechanical 1-20 0-5
wear actuations
seal defects contamination, 1-20 5-30
moisture, temp.
mechanical jamming, mechanical 7-27 0-5
stripping actuations
5-143
TABLE 5.3.2-10: VARIABLE COMPOSITION RESISTOR FAILURE MECHANISM DISTRIBUTION
resistance corrosion temp., humidity 40-60 45-65
drift
moisture moisture, temp. 20-40 25-45 '•

intrusion -
wiper move- shock, vibration 1-20 5-2-

ment
mechanical binding, mechanical actu- 1-15 1-5

failure jamming ation, corrosion
open terminal voltage/current, 1-13 1-5

defect temp.
burnout of voltage/current 1-12 0
resistive
element
TABLE 5.3.2-11: THERMISTOR FAILURE MECHANISM DISTRIBUTION
resistance moisture in- moisture, temp. 25-45 35-55

drift trusion
body
anomalies
temp., voltage/
current
20-40 14-35
p
open lead termina- vibration, 10-30 5-25 , :
tion defect temp., voltage/
current j
non-uniform temp., voltage 1-20 1-20
resistance
material
short various temp., contamina- 0-15 0-15
tion voltage/
currelt
5-144
-/
applications and essentially impossible to locate for nonoperating

applications. Therefore, much of information presented in Tables 5.3.2-6
through 5.3.2-11 was theorized. References 41, 42, and 43 provided
important resistor failure mode information. Relationships identified
L
during the failure mechanism analysis and the failure mechanism
comparisons are,
o Operating and nonoperating failure mechanisms for composition

resistors, thermistors and all types of variable resistors are
primarily accelerated by environmental stresses.
o Operating failure mechanisms for fixed film and wirewound
resistors are accelerated by a combination of environmental and
operational stresses.
o Carbon composition resistors are affected by moisture but usually

keep themselves dry during operation because of self-generated
"heat. However, if the equipment must stand for long periods under
humid conditions without power applied, then the resistance value
will change.
o Contaminants, especially when combined with moisture, can result

in resistance shifts for composition resistors. The probability
of experiencing contamination-related resistance drifts increases
with storage time.
The next step in the model development process was to extrapolate base
failure rate values for resistor styles which had no data. In addition,
the base failure rate values were normalized to correspond to a ground
benign environment and a M quality level. It was considered desirable to
normalize the base failure rate values to be consistent with MIL-HDBK-
217D. The base failure rates determined from the data correspond to a
ground fixed environment and a P screen class. The observed base failure
rates were divided by the ground fixed environmental factor and the P
quality factor (ITNQ3 = 0.52 in Table 5.3.2-5) to obtain the normalized
base failure rates. The nonoperating environmental factor and
nonoperating quality factor had previously been normalized to the desired
levels. The normalization process had no effect whatsoever on the
resultant resistor nonoperating failure rate prediction. In fact, the
base failure rate, nonoperatinq environmental factor, and nonoperating
-AA
IrI
quality factor could have been

normalized to any combination of X,
environment and quality without effecting the resultant prediction. The
increase in one factor would be compensated by a decrease in another.
Table 5.3.2-12 presents the normalized base failure rates, the ' t
extrapolated base failure rate values, and upper and lower confidence
interval values for resistor styles with data. The upper and lower 90%
confidence interval values were also normalized to correspond to a ground
benign environment and M quality level. ri.
TABLE 5.3.2-12: RESISTOR NONOPERATING BASE FAILURE RATES
Type Style Xnb,.05 Xnb Xnb,.095
Fixed Composition .000022 .000063 .00018

Fixed Film .000039 .000103 .00027
Fixed Film Network .000433
Fixed Wirewound .000195 .000565 .00164
Variable Wirewound -- .00516 --
Variable Non-wi rewound -- .00516 --
Thermistor Any -- .00267 --
The final, phase of the resistor model development process was to

investigate the failure rate differences between the specific styles of
variable resistors. The regression analysis indicated that there was no
significant difference between thermistors and variable resistor failure
rate in the ground fixed environment. However, the specific styles of
variable resistors were grouped because of data limitations and failure
rate differences could not be eetected. Point estimate failure rates,
predicted failure rates and t0% chi-squared confidence interval values
were computed for each variable resistor data record. No data record had ,,,,
greater than one failure and thus the observed failure rate estimates were
imprecise. A description of chi-squared confidence intervals is included
in Section 3.2 of this report. If the regression solution (or predicted) "
failure rate was contained in the 90% chi-squared confidence interval,
5-146
'VW. 1-7(.
-6ýr ~
then the variable resistor point estimate base failure rate previously
presented in Table 5.3.2-12, was applied to the respective variable
resistor style. This was the case for non-wirewound trimmer, wirewound
precision, and non-wirewound film variable resistors. For wirewound
trimmer variable resistors, the upper 90% confidence interval value was
less than the predicted failure rate. Therefore, it was concluded that
the nonoperating failure rate of wirewound trimmer variable resistors was
divergent from the remaining variabie resistor styles and required a
unique base failure rate value. A multiplicative base failure rate
adjustment factor for wirewound trimmer variable resistors was derived by
computing the ratio of the observed point estimate failure rate to the
predicted failure rate (Adjustment Factor = .00453/.0237 - .191). The
variable resistor base failure rate was then adjusted accordingly. For
variable resistor styles where no data were available, the operating
failure rate rankings found from MIL-HDBK-217 were inspected to aid in the
analysis. It was found that the operating failure rates for the variable
resistor styles without data (i.e. wirewound semiprecision, wirewound
power, non-wirewound composition, and non-wirewound precision) were ranked
between non-wirewound trimmer and wirewound precision. It was assumed
that the nonoperating failure rates would have a similar ranking and
therefore the regression solution base failure rate would also be
applicable to these resistor styles. It should also be noted that the
operating MIL-HDBK-217D failure rate prediction for wirewound trimmer
variable resistnrs was also substantially lower than other variable
resistor styles.
Adjustment of the base failure rate value for wirewound trimmer

variable resistors concluded the resistor model development process. The
proposed nonoperating failure rate prediction model for resistors is given
by the following equation. Two significant digits were used for all .
factors.
Xr nb 7NQ 7NE 7cyc
5-147
K I
where
Ar = predicted resistor nonoperating failure rate

Anb = base failure rate (failures/t06 hours)
= .000063, fixed composition

- .00010, fixed film
- .00043, film network
= .00057, fixed wirewound
"- .00099, variable wirewound, trimmer
= .0052, precision, semiprecision or power variable wirewound
= .0052, vari:ble non-wirewound
0027, thermistor
-.
fNQ = nonoperating quality factor (INQ3 in Table 5.3.2-5)

NE= nonoperating quality factor (given in Table 5.3.1-1)
"1cyc = equipment power on-off cycling factor
= 1 + .063(Nc)
where

hours
The proposed resistor nonoperating failure rate prediction model is

presented in a format similar to MIL-HDBK-217D in Appendix A.
The proposed resistor nonoperating failure rate prediction model was

next subjected to a thorough model validation process. The model
validation process took place in two phases. The first phase consisted of
an extreme case analysis. Predictions were made using the proposed model
for parameters beyond the ranges found in the data. The intent of the
extreme case analysis was to identify any set of conditions which cause
the proposed model to numerically "blow up", and also to identify any set
of conditions which predict a failure rate which is theoretically
5-148
inconsistent (e.g. a negative failure rate) or intuitively wrong. The

second phase of the model validation process was to compare the proposed
model with observed field data which had been withheld from the model
development process. Nonoperating resistor failure rate data had been
collected from the French Organization AFCIQ (Reference 8) but not used
"because of part characterization problems.
The extreme case analysis did not indicate any deficiencies with the
proposed model. At extremely high levels of equipment power cycling, the
nonoperating failure rate becomes directly proportional to the equipment
power cycling frequency. At low levels of equipment power cycling, the
l) proposed resistor nonoperating failure rate approaches a constant value.
These extreme case results confirm initiative reliability relationships.
At extremely high equipment power cycling rates, the nonoperating failure
"rate would be expected to be dominated by failures induced by the power
on-off cycle. Thus, the proportional relationship between the two
variables was anticipated. At extremely low equipment power cycling
rates, the failure rate would be expected to be equal to the inherent
storage/dormant failure rate, and independent of power cycling frequency.
The resistor nonoperating failure rate data collected from AFCIQ were
not used in the model development process because vital information such
as equipment power cycling frequency and part screen class could not be
precisely determined. However, the AFCIQ data was extremely useful for
model evaluation purposes. Table 5.3.3-1 presents the AFCIQ resistor
nonoperating failure rate data and the corresponding failure rate
prediction from the proposed model. The part types were not screened
according to identical specifications. However, equivalent quality levels
were determined. Additionally, one equipment power cycle per year was
assumed to compute the failure rate prediction. All AFCIQ data records
were for a ground fixed environment except record number 13, which
corresponds to a laboratory environment.
The model evaluation process substantiated the accuracy of the

proposed resistor model. The geometric mean of the ratio of observed to
5-149
•N-6
.~ %•
predicted failure rate was equal to 1.23 for those data entries with
without observed failures (i.e. record numbers 5, 9, 10, 14, 15, 16, 17 in
Table 5.3.3-1). It is also noted that data records 5, 9 and 17 were for
commercial grade resistors. The proposed nonoperating quality factor for
commercial quality resistors was an extrapolated value. Therefore, it was
considered very encouraging that the geometric mean of the ratio of
observed to predicted failure rate for these three data entries was equal
to 0.91. It should also be noted that the predicted failure rate was not
within the 90% chi-squared confidence interval for two of these three
commercial quality data records. Field failure rates for lower quality
parts always exhibit a higher degree of vdriability, and therefore this A
observation was not surprising.
The proposed resistor nonoperating failure rate prediction model is an

integral part of the methodology, presented in this report, which provides
capabilities to assess the effects of nonoperating periods on equipment
reliability. The proposed resistor nonoperating failure rate prediction
model provides proper discrimination against application variables which
influence failure rate. Additionally, the proposed model is relatively
easy to use, and has been shown to be accurate. The application of the
nonoperating failure rate prediction methods developed in this study will
grectly enhance nonoperating failure rate assessment capabilities.
5-150
4..
/ j
Ii
TABLE 5.3.3-1: RESISTOR MODEL VALIDATION DATA
Part
Hours Equiv.
No. Failures (X106) Style Quality ).05 Xo(1) X.95 Xpre(2)
' 1 0 20 Comp. MIL-SPEC -- (3) .150 .00044

2 0 40 Comp. MIL-SPEC -- (3) .0749 .00044
3 0 147 Comp. MIL-SPEC -- (3) .0204 .00044
4 0 420 Comp. MIL-SPEC -- (3) .0071 .00044
5 1 143 WW lower .00036 .00699 .0332 .00530
6 0 13 WW MIL-SPEC -- (3) .230 .00289
7 0 36 WW MIL-SPEC -- (3) .0832 .00289
8 0 10 WW MIL-SPEC -- (3) .300 .00289
9 9 3,422 film lower .00137 .00263 .0046 .00106
10 0 2,383 film MIL-SPEC -- .00042 .0013 .00058
11 0 155 film MIL-SPEC -- (3) .0193 .00058
12 0 84 film MIL-SPEC -- (3) .0357 .00058
13(4) 0 55 film MIL-SPEC -- (3) .0546 (4)
14
16
1
2
0
299
4,448
4,273
film
film
film
MIL-SPEC
MIL-SPEC
HI-REL
.00017
.00008
--
.00334
.00045
.00023
S15
.0159
.00142
.00070
.00058
.00058
.00013
17 4 300 var. (5) lower .00455 .01333 .0305 .0576
18 0 78 var. (5) MIL-SPEC -- (3) .0384 .0314
19 0 11 var. (5) HI-REL -- (3) .272 .00681
20 0 28 var., ww MIL-SPEC -- (3) .107 .0314
21 0 25 v., trimm. MIL-SPEC -- (3) .120 .00598
Totals 17 16,390
NOTES: 1) Xo is the point estimate failure rate
2) Apre is the predicted failure rate. A cycling rate of 0.114

cycles/lO3 nonoperating hours was assumed.
3) There were insufficient part hours to estimate a failure
rate without observed failures.
4) This data record was from life testing performed at 155 0 C.
5) Specific variable resistor style unknown.
5-151
S•.
Capacitors S~5.4 L
S5.4.1 Nonoperating Failure Rate Prediction Model

proposed Capacitor

prediction model for capacitors. The proposed model is presented in -
Appendix A in a form compatible with MIL-HDBK-217D. The proposed model

is:
X'p =X)nb WNQ ~'NE 1Tcyc
where
,p = predicted capacitor nonoperating failure rate

./ Xnb = nonoperating base failure rate (failure/lO6 hours)
=.0011, fixed paper/plastic film (CP, CZ, CA, CPV, CH, CHR, CQ,
CQR, CFR, CRH)
=.00039, fixed ceramic (CC, CCR, CK, CKR)
=.00075, fixed mica (CB, CM, CMR)
--.00045, fixed glass (CY, CYR)
=.00018, fixed tantalum, solid (CSR)
=.0064, fixed tantalum, non-solid (CL, CLR)
=.0064, fixed electrolytic, aluminum (CE, CU)
=.015, variable air trimmier (CT)
=.012, variable ceramic (CV) Y
=.0038, variable piston (PC)
=.046, var'iable vacuum (C(,)
XNQ =nonoperating quality factor
= .05, T (mica capacitors only)
=
=.10,
.23,
S
R
L
= .46, P -
= 1.0, M •
= 1.7, L"•i
= 2.5, MIL-SPEC
5-152.
= 5.3, lower
wNE = nonoperating environmental factor (See Table 5.4.1-1)
ncyc = equipment power on-off cycling factor
= 1 + .16(Nc)
whe i

hours
An accurate and useful nonoperating failure rate prediction model for

capacitors was developed by applying the modeling approach described in
Section 4.1. Analysis of the available capacitor nonoperating reliability
data and development of an assumed theoretical model were used to derive
the failure rate prediction model. The model was determined to be a
function of device style, quality level, environment and equipment power
on-off cycling frequency.
Initially, application and construction variables were identified

which characterize capacitors in a nonoperating environment. Table 5.4.2-
1 presents a list of the part characterization variables for capacitors.
These variables were analyzed in greater depth to determine their effect
on capacitor nonoperating reliability. Statistical methods were used when
appropriate. The part characterization variables were determined whenever
possible as an integral part of the data collection task. If sufficient
detail could not be determined for an individual data entry, then it was
removed fruum the model development process.
A theoretical nonoperating failure rate prediction model for

capacitors was developed by studying documented and intuitive nonoperating
reliability relationships. The theoretical a nonlinear
model assumes
failure rate equation because of the assumed effect of equipment power on-
off cycling. In addition, the theoretical model is constant with respect
5-153
JI
TABLE 5.4.1-1: CAPACITOR NONOPERATING ENVIRONMENTAL FACTORS (lINE)
Tant. Tant. Elec.

Env. Film Mica/Glass Cer. Solid Nonsolid A]. Variable
GB 1 1 1 1 1 1 1
GF 2.2 2.1 2.0 2.4 1.4 2 3.3
GM 8.3 8.8 8.3 7.8 10 12 9.6
Mp 9.9 11 11 9.2 11 12 17
NSB 4.7 5.0 5.0 4.4 5.0 5.8 7.7
NS 6.3 5.9 5.2 4.9 6.7 6.7 8.2
NU 14 15 15 13 15 13 18
NH 15 16 16 14 16 19 25
NUU 16 17 18 15 17 20 27
ARW 21 23 24 20 23 27 36
AIC 3.2 3.5 2.7 2.5 2.5 9.5 5
AIT 4.3, 4.0 3.3 2.5 4.0 10 5.3
AIB 7.0 8.0 6.2 7 6.5 10 7.7
AIA 4.9 4.0 5.0 3 6 15 13
AIF 9.2 10 8.0 7.5 10 15 13
AUC 7.6 15 6.0 4.5 8.5 28
AUT 13 20
15 12 6.0 15 30 38
AUB 23 35 15 25 20 30 57
AUA 17 15 17 10 20 30 50
AUF 33 40 30 30 40 40 85
MFF 9.9 11 11 9.3 11 12 16(1)
MFA 13 15 15 13 15 17 22
USL 23 31 32 27 31 36 47(1)
ML 33
CL
36 36 31 36 41 54(1) £ 4
560 610 610 510 610 690 930'
Notes: (1) This environment not applicable for vacuum or gas, fixed and I
variable (CG) type capacitors.
(2) Cer. - Ceramic, Tant. Tantalum, Elec. Al. Aluminum
Electrolytic
5-154
*A 6
TABLE 5.4.2-1: CAPACITOR PART CHARACTERIZATION
I. Device Style
A. Fixed
1. Paper/Plastic Film (CP,CZ,CA,CPV,CH,CHR,CQ,CQR,CFR,CRH)
2. Mica (CR,CM,CMR)
3. Glass (CY,CYR)
4. Ceramic (CC,CCR,CR,CRR)
5. Electrolytic (CE,CL,CLR,CSR,CU)
6. Fixed Vacuum or Gas (CG)
B. Variable (CV,CTPC,CG)
II. Capacitance (microfarads)
III. Quality level

A. T
B. S
C. R
D. P
E. M
F. L
G. Mil-Spec
H. Lower
IV. Construction (,CLR type only)
A. Slug
B. Foil
V. Seal
A.
B. Hermetic
Nonhermet i c• P i'
VI. Body Dimensions
VII. Temperature
A. Rated
B. Actual
IX. Number of Power On/Off Cycles per 103 Nonoperating Hours.
"-.• -7ý
5-155
%'~
Sto time. The theoretical model for capacitors is given by the following
f :equation (f denotes a function).
Xc = f(device style, temperature, C)1TNQ 'TNE Ircyc
where
xc = predicted capacitor nonoperating failure rate (failures/10 6

hours)
C = capacitance (microfarads)
'NQ nonoperating quality factor, based on device screening
wNE = nonoperating environmental factor
7cyc = equipment power on-off cycling factor = I + KI(Nc)

K1 = constant
Nc = equipment power on-off cycling rate (cycles/10 3 nonop. hrs.)
The summarized capacitor nonoperating reliability data collected in
support of this study is presented in Table 5.4.2-2. The data

collectively consists of 324 individual data records, 48 observed failures
and 41,840.6 X 106 part hours.
A preliminary analysis was then performed to identify strengths and

weaknesses of the collected data set. Initially the data was evaluated
without merging data records. The methods described in Section 3.3 were
applied to determine which zero failure data records had sufficient part
hours to estimate a failure rate without observed failures. Ninety eight
data records either had observed failures or had sufficient part hours to
estimate a failure rate without observed failures. Limitations imposed by
the nature of the available data restricted the n'imber of variables which
could be statistically evaluated. Four deficiencies of the collected
Li
nonoperating capacitor data base were identified from the preliminary data
analysis. These deficiencies are,
1) All collected data was from a ground based environment.

Therefore, there was not a sufficient range of environments to
5-156
', C
TABLE 5.4.2-2: CAPACITOR NONOPERATING FAILURE RATE DATA TABLE
Data Part
Equipment/Source Records Style Failures Hours (xj06)
Hawk/MICOM 2 Film 0 1271.9
Hawk/MICOM 10 Ceramic 2 7299.3
Hawk/MICOM 3 Mica 0 1071.9
Hawk/MICOM 3 Electrolytic 0 3896.4
Hawk/MICOM 2 Variable 0 85.1
Maverick/MICOM 4 Film 1 453.0
Maverick/MICOM 1 Ceramic 0 1997.2
Maverick/MICOM 2 Mica 0 1235.4
Maverick/MICOM 2 Electrolytic 0 4838.7
Sparrow/MICOM 50 Film 6 1620.6
Sparrow/MICOM 77 Ceramic 0 650.8
Sparrow/MICOM 14 Mica 0 51.0
Sparrow/MICOM 42 Electrolytic 0 663.6
Sparrow/MICOM 34 Glass 0 1225.0
Sprint/MICOM 6 Mica 1 230.4
Sprint/MICOM 1 Glass 0 71.6
Sprint/MICOM 5 Ceramic 0 909.9
Sprint/MICOM 4 Electrolytic 0 104.6
Sprint/MICOM 1 Variable 1 9.7
TOW/MICOM 3 Film 1 63.1
TOW/MICOM 1 Mica 0 65.7
TOW/MICOM 1 Ceramic 0 131.4
TOW/MICOM 2 Electrolytic 0 63.1
Lance/MICOM 6 Film 0 32.4
Lance/MICOM 3 Mica 0 4.4
Lance/MICOM 6 Ceramic 0 11.2
Lance/MICOM 4 Electrolytic 2 17.9
Martin Marietta 7 Film 4 413.3
Martin Marietta 4 Mica 1 660.0
Martin Marietta 2 Glass 0 299.4
Martin Marietta 2 Ceramic 5 3832.4
Martin Marietta 5 Electrolytic 7 2613.2
Martin Marietta 4 Variable 1 133.4
F-16 HUD/RIW 7 Ceramic 14 4748.9 r
F-16 HUD/RIW 2 Electrolytic 2 1012.1
F-16 HUD/RIW 1 Glass 0 25.6 p
PRC 1 Electrolytic 0 27.0 K
Totals 324 48 41,840.6
5-157 :1
develop a series of environmental faztors based solely on the

data. ,
3) The inherent failure rate of capacitors in storage/dormant

applications is very low. Therefore, the number of data records
with observed failures and the number of failures per data record
were naturally limited. Precise estimates of field failure rates
and quantification of failure rate model parameters can not be
made without observed failures. Thus, the number of variables
which could be analyzed empirically was limited.
4) No data was available for Aluminum Oxide electrolytic capacitors

or for any variable capacitor style except variable, air trimmer.
K
The theoretical capacitor nonoperating failure rate prediction model
was further studied to determine the relative effect of the independent
variables. It was concluded that capacitor device style, quality level,
application environment and equipment power on-off cycling frequency were
the dominant variables effecting nonoperating failure rate. Capacitance
level was not anticipated to have a significant effect on nonoperating
failure rate. This variable was therefore not included in subsequent
analyses due to data limitations.
A matrix of "dummy variables" (0 or 1) was defined for capacitor

device style so that correlation coefficient analysis and regression
analysis could be applied to the data. Device style is a qualitative
variable, whereas regression and correlation analyses require quantitative
variables. The "dummy variable" matrix allows for application of these
w
numerical methods. The matrix for device style is given in Table 5.4.2-3.
Of particular interest was the specific device styles within the
electrolytic capacitors category. Aluminum electrolytic capacitors are a
"shelf life" item which is known to significantly degrade during storage.
Non-solid tantalum capacitors are also believed to be sensitive to
prolonged storage. The non-solid dielectric is particularly sensitive to
a loss of seal. Unfortunately no data were available for aluminum
electrolytic capacitors. However, data were available for both solid
tantalum and non-solid tantalum electrolytic capacitors, and the device fr-
5-158
style matrix was defined to allow for a comparison of observed

nonoperating failure rate.
A qualitative matrix was also defined for quality level. The matrix
was simply a "0" for high-reliability parts and a "1" for mil-spec quality
parts. This variable was designated as QI. The quality matrix was not
defined to accommodate lower quality capacitors because no data of this
type was available.
TABLE 5.4.2-3: CAPACITOR DEVICE STYLE VARIABLE MATRIX
Device Style S S2
S S3 S4 S5
Paper/plastic film 0 0 0 0 0
Ceramic 1 0 0 0 0
Mica 0 1 0 0 0
Solid tantalum 0 0 1 0 0
Non-solid tantalum 0 0 0 1 0
Variable, air trimmer 0 0 0 0 1
The correlation coefficient analysis indicated no large correlation

between independent variables for the merged data set. The data were
merged according to equipment, device style, and quality level. For
example, all data for B screen class, fixed mica cdpacitors in the Hawk
missile were merged.
Since the theoretical capacitor model was nonlinear due to the assumed
relationship for equipment power cycling frequency, the approach to
nonlinear regression described in Section 4.3 was implemented to quantify
the nonlinear model. Equipment power cycling frequency was temporarily
treated as a qualitative variable. Constant multiplicative values were
computed by regression for three distinct cycling rate categories. Then a
two-dimensional regression was performed with the cycling rate .
coefficients from the initial regression as the dependent variable, and
the mean cycling rate value for each category as the ihdependent variable.
The result would be an expression for equipment cycling factor of the
desired form (i.e. cycling factor = mx + b). An iterative process was
5-159
• '• .:
then required. First, the equipment power cycling factor expression was
assumed exact, and the coefficients for all other variables recalculated.
Then, those coefficients were assumed to be exact and the equipment
cycling factor recalculated. The iterative process would continue until
observed changes were negligible. For capacitors, this process required
three iterations.
Three qualitative "dummy" variables were defined to represent four

distinct equipment power cycling rate categories. The three variables
were designated as cycl, cyc2 and cyc3. The categories were defined as,
cyc2, cyc3 = 0,0,0).
years (cycl, cyc2, cyc3 = 1,0,0), (3) cycling rates greater than
one cycle per year (cycl, cyc2, cyc3 = 0,1,0).
o cycling rates greater than one cycle per year (cycl, cyc2, cyc3 =
1,0,0).
o unknown cycling rates (cyci, cyc2, cyc3 =0,0,1).
The first three categories designated as (1) low,

were (2)
intermediate, and (3) high cycling frequencies, respectively.
Application environment was assumed to be a significant variable

effecting capacitor nonoperating failure rate. It was essential that a
proposed capacitor nonoperating failure rate prediction model include an
appropriate environmental factor to properly discriminate against known
failure accelerating environmental stresses. Unfortunately, 'all collected
data was for ground based environments. Therefore, environmental factors
could not be developed empirically. This data deficiency existed for the
model development of each part type, and is discussed in-depth in Section
4.5 of this report. Section 4.5 also proposes a method to develop
appropriate environmental factors based on a comparison of operating and
nonoperating failure mechanisms. This method was used to determine
5-160
MA/
capacitor nonoperating environmental factors after application of

regression analysis.
The fact that temperature could not be empirically analyzed was also a
problem. However, after careful consideration of many factors, it was
decided not to include temperature in the nonoperating failure rate
prediction model. Similar to the discussion for resistor nonoperating
failure rate model development, there were three reasons determined to
support this decision. The first was that no temperature relationship
could be derived from the data due to the limited range of temperatures
available. The second reason was that no theoretical relationship was
identified in the literature which corresponded to nonoperating
conditions. The third reason was that the effect of temperature on
nonoperating failure was believed to be relatively low in comparison to K
microcircuits and discrete semiconductors. Therefore, an ;pproximate
assumed temperature relation for capacitors was not warranted.
At this stage of the model development process, the data were in the 4.
format required for application of regression analysis. Results of the
initial regression analysis are presented in Table 5.4.2-4. The dependent
variable was In(failure rate). The independent variables were SI, S2, S3 , v
S4 , S5 , QI, cycl, cyc2 and cyc3.
Device style variables S3 , S4, and S5 were significant at a 90%

confidence limit. Device style variable S1 , device quality variable Q1
and equipment cycling frequency variables cycl, cyc2, and cyc3 were
significant at a 80% confidence limit. Device style variable S2 was
significant at a 65% confidence limit. The R-squared value for the
regression was .85. That is 85% of the variability in the observed data
can be explained by the' regression solution. Interpretation of the
regression results are:
1) The variable, non-solid and solid tantalum capacitor styles were

significantly differcnt, with a 90% confidence limit from the
other device styles.
5-161 .. ..
%%%
/ \
/
/ Downloaded from http://www.everyspec.com
2) All other styles were significantly different from paper/plastic

film at a 60% confidence limit.
3) The difference in predicted failure rate between mica and paper
plastic film capacitors was not significant at a 80% confidence
limit.
4) As expected, it was determined with 80% confidence that capacitor
screening effects the field nonoperating failure rate.
S~~Speci
Interpretation of the initial regression solution was that the
failure rate of mil-spec quality resistors is 3.9 times higher
than fi capacitors
cation. screened to the established reliability IL!
5) The cycl, cyc2 and cyc3 equipment power cycling variables were
significant with 80% confidence. This substantiated the
assumption that equipment power cycling is aro important variable
for predicting nonoperating failure rate.
TABLE 5.4.2-4: CAPACITOR INITIAL REGRESSION RESULTS
Variable Coefficient Error f-ratio Limit
Si -1.578 .528 8.94 .80

"S2 -0.601 .622 0.93 .65
S3 -2.180 .627 12.08 .90
S4 0.875 .662 1.75 .90
S5 2.283 .817 7.80 .90
QI 1.189 .505 5.55 .80
cycl 0.891 .541 2.71 .80
cyc2 2.C24 .736 12.70 .80
cyc3 1.500 .440 11.63 .80
I
bo -6.775 -- -- --

failure rate. Transtorming the regression results into an equation where
failure rate (as opposed to loq of failure rate) is the dependent variable
results in the followiog rultýnlicative model:
Xc =nb 'NQ kcyc,
5-162
! ,
.~' ~ /
where
6
Xc = predicted capacitor nonoperating failure rate (failures/10
hours)
Xnb = base failure rate
= exp(-7.089 - 1.306(S1).- 1.879(S 3 ) + 1.120(S4) + 2.650(S 5 ))
= .000834, paper/plastic film
= .000834, mica
= .000226, ceramic
= .000127. solid tantalum
= .00256, nonsolid tantalum
"=.0118, variable
S•NQ = quality factor
= exp(1.361(Q1))
= 1.0, hi-rel
= 3.9, mil-spec
ircyc = equipment power cycling factor
= exp(.970(cycl) + 2.651(cyc2) + 1.446(cyc3))
3
= 1.0, cycling rate <.057 cycles/lO hr.
3
= 2.64, .057 < cycling rate < .114 cycles/10 hr.
3
= 14.17, cycling rate > .114 cycles/10 hr.
= 4.25, unknown cycling rate
= residual
The variables S1, S2, S3, S4, S5, Q1, cycl, cyc2 and cyc3 *have been
previously defined. It is emphasized that the above equation represents a
preliminary model, which is only applicable to ground based environments.
The observation that equipment power cycling variables cycl and cyc2
were identified as significant variables supports the assumption that
The
equipment power on-off cycling effects nonoperating failure rate.
be
cyc3 variable for unknown power cycling was also observed to
significantly different, with 80% confidence, from the low cycling
frequency values. This was attributed to several possible reasons. This
observation may be because the average cycling frequency for the unknown
5-163
.;.•.-•
I •" . "••:5 .".". '" ,-"-" .' "•',t >" , •," .•->•; " •" :;- ,'' :"'".•'.:'-'--'.-:-'-..-% ''• • .•'-.N'•
data entries was greater than 0.114 cycles/10 3 nonoperating hours.

Another possible reason was that the data in the unknown category may have
been biased and that the observed failure rates are not representative of
capacitors in a nonoperating environment. Neither contention can be
proved. However, the magnitude of the ratio of unknown to low equipment
cycling was not large enough to cause any concern.
The next step in the model development process was to perform the
iterative regression analyses to accommodate the assumed non-linear model
form. It
has been documented (Reference 5) that the effect of equipment
power on-off cycling is best represented by an equation of the following
form:
ffcyc = 1 + Kl(cycles/10 3 hr.)
As previously described, the iterative process began by performing a two-

dimensional regression analysis for "equipment power cycling factor"
versus average cyclin9 rate for the three previously defined cycling rate
categories. The average cycling rate for capacitors exposed to low
equipment cycling rates was .042 cycles/lO3 hrs. The data in this
category was primarily from the Sparrow and Hawk missile programs. The
3
average cycling rate for the intermediate category was .079 cycles/lO hr.
This data was primarily from the Maverick and TOW missile programs. The
average cycling rate for capacitors exposed to high equipment cycling
frequencies was 40.12 cycles/10 3 hr. This data was from the F-16 HUD RIW
data. The data from sources with unknown cycling frequencies were not
included in this regression. The data entries were weighted by the number
of observations per category. An explanation of weighted regression is
included in Section 3.2. The third iteration for equipment power cycling
factor resulted in the following equation.
K2fcyc = .60 + .0962(cycles/10 3 hr.)

3 hr.)
cyc = 1 + .16(cycles/10
5-164
!.
where
K2 = normalization constant
= 1/.60 = 1.66
The preliminary model at this stage of the model development process

is given by the following equation. The residual was dropped from the
equation at this stage of model development.
Xc = Xnb INQ Icyc
where
Xnb = preliminary base failure rate

= .00231 paper plastic/film
= .000782 ceramic
= .00158 mica
= .000434 solid tantalum
= .00896 non-solid tantalum
= .05001 variable
wNQ = quality factor
= 1.0, Hi-rel
= 2.47, mil-spec
vcyc = equipment power cycling factor
= 1 + .16(cycles/10 3 hr.)
It was noticed that the ratio of non-solid tantalum to solid tantalum

preliminary base failure rates was greater than twenty to one. It had
been anticipated that the non-solid tantalum nonoperating failure rate
would be relatively higher because of the sensitivity of the non-solid
dielectric to even minimal loss of seal properties. The observation that
the observed nonoperating failure rate was twenty times as high was
considered to be additional evidence that non-solid tantalum capacitors
are relatively more sensitive to nonoperating periods. The corresponding
operating failure rate ratio (from MIL-HDBK-217D) is approximately 1.2 to
"5-165
1. This difference between operating and nonoperating ratios was another

reason why operating and nonoperating failure rate prediction must be
treated separately.
The magnitude of the nonoperating quality factor for mil-spec

capacitors was small in comparison to the ratio of mil-spec to established
reliability (i.e. T, S. R. P, M and L) operating quality factors from MIL-
HDBK-217D. It was hypothesized that this trend was due to a significantly
slower rate of failure for "infant mortality" failures (i.e. weak devices
remaining after the screen and subsequently failing) during nonoperating
periods. This trend is discussed in Section 4.6 of this report.
It was decided to derive unique nonoperating quality factor values for

the established reliability (ER) quality levels defined in the ER
specifications. The available nonoperating failure rate for ER (or Hi-
rel) capacitors did not specify the particular established reliability
level. In general, the average capacitor quality level was R, P or M. It
Swas assumed that the average ER level for the collected data was M. A P
quality level had been assumed for resistors. However the observed effect
of quality was less for capacitors and it was decided to make a "worst
case" assumption. Inspection of the MIL-HDBK-217D capacitor quality
factors revealed that the numerical values increase geometrically. The
assumption was made that nonoperating quality factors would also follow a
geometric progression. Ideally, a unique nonoperating quality factor
series may have been required for each capacitor style. However, due to
the amount of available data, a single series of nonoperating quality
factors was proposed for all capacitor styles. Numerical values for
quality levels without data were extrapolated by assuming a similar

ranking as the MIL-HDBK-217D operating quality factors. Table 5.4.2-5
presents point estimate nonoperating quality factors, upper and lower 90%
confidence interval values, and extrapolated nonoperating quality factors
for T, S, R, P, L and lower qualities.
5-166
TABLE 5.4.2-5: CAPACITOR NONOPERATING QUALITY FACTORS
Quality Level nNQ,.05 WNQ1(1) NNQ,.95 ' tNQ2(2)

T ...... .05
S ...... .10
R ...... .23
P ...--. .46
ML-- 1.0 1.00
L-- -- -- 1.70
MIL-SPEC .810 2.47 7.46 2.47
Lower -- -- -- 5.25
NOTES: (1) lNQ1 is the observed point estimate nonoperating quality

factor.
(2) WNQ2 is the point estimate quality factors supplemented by

extrapolated values for T, S, R, P, L and Lower quality
levels.
The next task in the capacitor model development process was to apply
the methods presented in Section 4.5 to determine nonoperating
environme-tal factors. The proposed capacitor nonoperating environmental
factors are presented in Table 5.4.1-1 in Section 5.4.1. Eight series of

environmental factors were required to properly assess the effect of
environmental stress on capacitor nonoperating failure rate. The 19
different environmental factor series in MIL-HDBK-217D were grouped and
averaged for similar capacitor styles. Since the nonoperating
environmental factor development process was inexact, the small
environmental factor differences between similar styles for the operating
condition were not justified for nonoperating failure rate prediction
purposes.
Table 5.4.2-6 through 5.4.2-13 present failure mode/mechanism and

failure acceleration factor distributions for capacitors in both the
operating and nonoperating conditions. Accurate distribution information
for operating applications was difficult to find and essentially
impossible to find for nonoperating applications. Therefore, much of the
5-167
S.... 1V.
a~ c I 0)
41 41 10 0 4J4.
4%O I l LALALnL m
0 ý4- c-4NN M =- "-4 -4 LAnU
LAn
Lin. OL 0) kn
04~~ Nn 04D4-.-.- Lc4
L A0O V-4 00 kN
ce-4
(A
4M
0 .- Nn b. Cn
I L AADL
0m
V)- 4J- 14 .0b I

~
-=-4 *O -
NCJ -
4 nALn I.~
IV LA I n 0L
G)4Jb 4
9-94- - -4 (041) LnAýLALA-4 V- V4-4ý .J
u) C
9.- 4J)
> U.
39 >1 0 0. >1
C 0)
w) 0)4
C 0)
(D I-.o
0~ -0 9 S. 4J CS
0. L fol o ~
> 41 uS 4 > -j4
t. . 4- I) u J.
fo.. -) to ...
au. E.0 0. .0 a) U- O*.
0..00 C
0.M
0. (U E
EM
40
4. 4141). 4J 4J4- 4-) 4J 4J 4- 4-) .. .
LLii
4-) 4-)
a)) C .-
CC cC4 4.)
s 0 NL
C9- c j9 . C L '
0 0o
0) Ca C #0 ;l,
wJ LA; 11" 0 C
*L 00 a0
0 to I" 4j c 0
to OCS- c. cU
9
404) 4J S- Q.C 4. -0.
0)o' E- -- 00
- I" U u L 0)a
(UCA S L. 0- w s.C-) C- >0 d
S- 4 4 4 o4 4.) 0000 1.. '--C
cLuu u r. CA=' a. S
M U(D U W r, a LCU4 CO
. 0-)
ilo
E
4~)4.)4
"a-
0
S-f
S-
C M
u
m
4.)
C
,T
- WU
~ ) C M-W u
1*-
4J
(
.1.&
UC 0)0)00 W 0) (UL (20+) 0
c0 4-
0- w V C- '-~~
IL 000~
S) to ~ CL
o 0L A
T~~am 40%o)-
4. I. Ln LA) C) C)4 .- 4 n %
'at 0. 00
o0 f0 C%J 1-4.'-4 0 4-J 0 I-. LA

C CA
o0
I- S 0 0 1-s- n
4) 4J - )43
m 1~~~.4 ). - C m
a CA CL
0 Q
cm.. -
01C Ln
LL. IA3 CL4;L.(
m, 0 0 >0
S- 41 >3 1.1*4.
W -- - U - >-
4)~~- 4J-&J4J4 d
4J4- 41 S-.
= =
-~~0 =~ 0
0 4J. 4 0 l
43 3 ' to
3t r-43 9 D C
LLJ.
X, 4*w- 0 jb 43
Wm a0 u 4- 2L 4-1 4.(
S.- 4- S.. 4-
c. S-0 -)g 0 C c$- 0
LA. to 0o 'Vu 0o 'OAD
'I. -- o) U. - W
S- w 43 %- 4 41 4
C-4 -4- -4 go :3 -4 %NJ Ln '.
I.~ I IIL O tII

041 -'00 - ' 4 4 V- - 0 041 QO-4 (%JOC r-O ,4-4
cik
-l 0I
(M. 0 00 0
C-4-) CD
000 (y) m mv Ln Lc) C 4- L Lc)Ln Ln 0D LAOC
V) C%jCI\ .- 4 -4 r-4 ",4 C%J
i J%
41 411 Ik 41a
04-)
d) 04 004 1-. L
CL4 04- -- 0) 41 C4J. -4 Ln.4 -4 Ul V-4
z CD
-S.. 41 4) U
> > C tIm0

0A r . A.- 0. *
.4J 5-. 0)00
CL (M3 .41 . 41)-4J
. 4
to 0 C EC to t;0 t o) M0 go 0 0a 0)0M
4- c) &- C I" -ý
r_ r_ 4J C-
L)~~I
W 0)
~ ~ =0 r-) fa 0- C~
a-c 0 0u u0
to ai
. -c 04 a 00
t 00.M
- ~ ~ 04-)0
O 0>O00 0.06 .4L
-Wdo
4
-
~0
C.~ 06.0
CoE
41> 41 CA 4) 4J 414+'.C > t
4--
Lii 0
CC
0 V 0J4 0 +
C m4t) 4
-o 0 eo .- 40 )
0) 0
co ~ ~ .w -j
x
(a
to >1
M) > 0 /4
0u -4 to C 4-( S 4 o C
0)Q 04 - V C0
) 00410 U- 0a 4
0)(f . ) C- 4J f0j d) wA fa 4 M
- 'I) CU 4)' (- ) kAC

"O~ LA
- 0
0-=
0 -
-
u 0-
s-C )
to 0>) 4X) 0 W0- 0 )
00 -.- 4- 44-- 000.4A ~0
004
U0go 0)LA ~ '4 - IA00
4)X~4 4)4)1)
0_ I- .C .c 4-V
0)f., 0- IV0 0). 0' V 0 .
50 7
4.) *1*LI n n
m o 0 J4J I f00 cr CD
CM41 In COj I~
CL4L. Ln 0~ ) CL4 CD 0 CD 0
O*41 Oat .n-4 C.J CNI O41 r--4 "%C%.4
~
t*
C~j
zo
V)
.- c.-C n A
c 1 C d *,D L ) c4
0 L.o
a~ C
UU
C 2l 4 > >~ > > CL4

4Ja 0-8) C1 83 1+- 4JaS-83 S-
o tm to 0 0t 0 C; C
4J to0 -. .. u3 C
41IU
* L O > .IV OU u u GiC
LALL) U- > . cc
*t . ' -
C> -'&U . *-U U 4.J
to cc to
483 6 4 06 *.- 8
>
1.. 4-3- S. 4 >8
0~~~~ . 00 C .C 0
LL~~
In ~~
0 U 8 L.
information in these tables was theorized.

capacitor failure mode information. The
identified during the failure mechanism analysis:
References 33 and 41 provided
following relationships were
I:
o For aluminum electrolytic capacitors, the aluminum cases corrode
in salt air environment.
o Both glass atid mica capacitors are adequate for most military
environments which include exposure to humidity and high levels of
shock and vibration.
o In the nonoperating mode, shorts and opens in ceramic capacitors
can be accelerated by temperature cycling and to a lesser extent,
mechanical stress. Low insulation resistance may occur as a
result of surface conditioning during manufacture or because of
S~moisture
penetration through a defective epoxy encapsulation.
Monolithic ceramics when properly cured and fixed are impervious
to moisture as the chip itself is concerned, but the shunt path
formed by contaminants surface moisture may result in circuit
failure.
The next step in the model development process was to extrapolate base
failure rate values for capacitor styles for which there was no data.- In
addition, the base failure rate values were normalized to a ground benign
environment. The observed data corresponded to a ground fixed
environment. Table 5.4.2-14 presents the normalized base failure rates,
the extrapolated base failure rates, and upper and lower confidence
interval values for capacitor styles for which there were observed data.
TABLE 5.4-14: CAPACITOR NONOPERATING BASE FAILURE RATES
Type Style Xiib'* 05 )Lib Xnb,. 95

Fixed Paper/plas. film -- .00105 --
Fixed Ceramic .000119 .000391 .00128
Fixed Mica .000182 .000752 .00314
A
Fixed Glass -- .000450 --
Fixed Ta, Solid .0000436 .000181 .000750
Fixed Ta, Nonsolid .00143 .00640 .0287
Fixed Al Oxide -- 00640 --
Variable Air, Trimmer .00265 .0152 .0867
Variable Ceramic -- .0122 --
Variable Piston -- .00376 --
Variable Vacuum -- .0456 --
5-172
- -
Normalization of the base failure rates concluded the model

development prceess. The capacitor nonoperating failure rate prediction
model was found to be a function of device style, quality, environment
power on-off cycling frequency. The proposed model is presented in
Section 5.4.1 and in Appendix A in a format compatible with MIL-HDBK-217D.
The proposed capacitor nonoperating failure rate prediction model was

subjected to a thorough model validation process. First, the model was
subjected to an extreme case analyses. That is, predictions were made
using the proposed model for parameters beyond the ranges in the data.
The purpose of this analysis was to test the model for any set of
conditions for which this model would "blow up," or predict a failure rate
which is theoretically inconsistent. No deficiencies with the proposed
model were found. Secondly, the prediction model was compared with field
data (AFCIQ) which was purposely withheld from the model development
process. Table 5.4.3-1 presents this nonoperating field data and the
corresponding failure rate prediction from the proposed model. Although
these parts were not screened according to the identical specifications
used for the model development data, equivalent quality levels were able
to be determined.
Except where otherwise inldicated, all data was from a ground, fixed p.
environment. Generally, the Hi-rel quality level was equivalent to a
level of M.
5-173
%9
TABLE 5.4.3-1: CAPACITOR MODEL VALIDATION DATA
Part
Hours Equiv.
No. Failures (x10 6 ) Style Quality X. 0 5 Xo(1) X*.95 Xpre(2)
1 1 25 Ceramic MIL-SPEC .00205 .0400 .190 .00197

2 5 1637 Ceramic MIL-SPEC .00120 .00305 .0064 .00197
3 0 84 Ceramic MIL-SPEC -- (4) .036 .00197
4 0 835 Ceramic MIL-SPEC -- .00120 .0036 .00197
5(3) 0 82 Ceramic HI-REL -- (4) .037 (3)
6(3) 0 518 Ceramic HI-REL -- (4) .0058 (3)
7 1 42 Film MIL-SPEC .00122 .0238 .113 .00581
8 0 37 Film MIL-SPEC -- (4) .081 .00581
9 0 18 Film MIL-SPEC -- (4) .170 .00581
10(3) 0 31 Film HI-REL -- (4) .097 (3)
11 0 30 Mica MIL-SPEC -- (4) .100 .00397
12 0 87 Mica MIL-SPEC -- (4) .034 .00397
13 0 2 Paper MIL-SPEC -- (4) 1.497 .00581
14 0 36 Paper MIL-SPEC -- (4) .083 .00581
15 0 4 ta, solid Lower -- (4) .749 .00231
16 1 42 ta, solid MIL-SPEC .00122 .0238 .113 .00109
17 2 855 ta, solid MIL-SPEC .00042 .00234 .0074 .00109
18 0 46 ta, solid MIL-SPEC -- (4) .065 .00109
S19 1 27 ta, solid MIL-SPEC .00190 .0370 .176 .00109
20(3) 0 5 ta, solid HI-REL -- (4) .599 (3)
21 3 560 ta, non-s MIL-SPEC .00146 .00536 .0138 .0225
22 0 1 Glass MIL-SPEC -- (4) 2.995 .00238
Totals 14 5004
(1) Xo is the point estimate failure rate. L

(2),r is the predicted failure rate. A cycling rate of .114
cycles/lO nonoperating hours was assumed.
(3) Environment was unknown, so prediction with this data entry was
not possible.
(4) Insufficient part hours to estimate a failure rate without
observed failures.
5-174
Vi'. .-, . . -- - , - " - ','" •+ ' J, +, '. ++ w " " ""•w+•+ •"" + " •+'-"v 2•<
" •+."w+° t'• '% •++" +-%+
The model validation process indicated that the proposed nonoperating • -

failure rate prediction model for capacitors was relatively accurate.
Three data entries (i.e. 1, 16 and 19) indicated that the proposed model
was optimistic, and one data entry (i.e. 21) indicated that the proposed
model was nessimistic. The predicted failure rate for four data entries
(i.e. 2, 4, 7 and 17) was contained within the 90% chi-squared confidence
interval and were considered to be representative of an accurate model.
Further investigation of the model validation data proved to be
enlightening. Data entry number one was for ceramic capacitors and
suggested that the proposed model was optimistic. However, data entry
number 2 was also for ceramic capacitors and had a greater number of
failures and part hours. The predicted and observed failure rates were
close for this data entry. Similarly, data entries 16 and 19 were for
solid tantalum capacitors and indicated that the proposed model was
optimistic. However, data entry number 17 was also for solid tantalum
capacitors and had more observed failures and part hours than either of
the other two data entries. As before, this data entry indicated that the
proposed model was accurate. The apparent variability of some' of the
observed failure rat,,\s ,.,th a small number of failures and part hours was
not unusual. It was concluded that observed nonoperating failure behavior
requires some minimum time interval to stabilize. It was considered very
encouraging that the three model validation data entries with the largest
number of part hours (e.g. 2, 4 and 17) each indicated that the proposed
model was accurate. N
The proposed capacitor nonoperating failure rate prediction model

provides proper discrimination against application variables which
influence nonoperating failure rate, and is an integral part of the
methodology presented in this report. It was designed to be easy to use,
and has been proven to be relatively accurate.
5-175
, /. t ' '.' ,'
/ .',
• ','
5.5 Inductive Devices
5.5.1 Proposed Inductor Nonoperating Failure Rate Prediction Model

prediction model for inductive devices. The proposed model is presented
in Appendix A in a format compatible with MIL-HDBK-2170. The proposed
model is:
C41
p NQ iNE I1cyc
nb Wb 1.
where F
•'p = predicted inductor nonoperating failure rate
Xnb = nonoperating base failure rate
= .000055, low power pulse and audio transformers
= .00028, high power pulse, power and RF transformers
= .00015, RF coils
fNQ = nonoperating quality factor
= .06, S
.15, R
.38, P ".
=1.0, M
3.1, Mil-spec
lower
11,
wNE = nonoperating environmental factor (see Tabie 5.5.1-1)
= 1 + .75(Nc), transformers
= 1 + . 38(Nc), coils
where
Nc number of equipment power on-off cycles per 10 nonooeratingV
hours
5-176
/
TABLE 5.5.1-1: INDUCTOR NONOPERATING ENV TRONMENTAL FACTORS
Environment Transformers Coils
GB1 1
GF 5.7 3.6
GM 12 12
MP 11 11
NSB 5.1 5.1
NS 5.7 5.7
NU 14 14
NH 16 16
NUU 18 18
ARW 24 24
AIC 4.5 4
4IT 6 4.5
AIB 6 5.5
AIA 6 4.5
AIF 9 9
AUC 6.5 5
AUT 6.5 6.5
AUB 7.5 7.5
AUA 7.5 6.5
AUF 10 10
SF(1) (1)
MFF 11 11
MFA 15 15
USL 32 32
ML 36 36
CL 310 610
NOTES: (1) Space flight environment was not considered in this study.
5.-177
5.5.2 Inductor Model Development
The failu-e rate modeling approach described in section 4.1 was

implemiented for inductors. Analysis of the available nonoperating
inductor failure rate data was complemented by an assuxned theoretical
model to derive the failure rate prediction model. The model was
determined to be a function of inductor style, quality level, environment,
and power on-off cycling frequency. For the purposes of this study,
inductive devices were considered to be pulse, audio, power and RF
transformers and RF coils. The proposed nonoperating failure rate
prediction model for inductors is included in Appendix A in a format
similar to MIL-HDBK-217D models.
Identification of application and construction variables which

properly characterize inductors in a nonoperating environment was the
first step of the preliminary model development. Table 5.5.2-1 presents a
list of these variables for inductors. The part characterization
variables represent possible failure rate model input parameters, analyzed
in greater depth using statistical methods. As with all other models, the
part characterization variables were determined whenever possible for all
inductor data. If sufficient detail could not be achieved for an
individual data entry, it was then removed from the model development
process.
A theoretical model for inductors was determined based upon the

general nonoperating reliability concepts presented in section 4.0 and a
study of inductor nonoperating failure mechanisms and physics of failure
information. The theoretical model was assumed to be a nonlinear failure
rate equation Lzcause of the contribution of equipment power on-off
I• cycling.
5-178
r.- r. j%.
TABLE 5.5.2-1: INDUCTOR PART CHAKACTERIZATION
I. Device Style
A. Coils
1. Construction
a. Fixed, RF
b. Variable, RF
c. IF
2. Insulation Class
a. A
b. B
c. C
d. F
B. Transformers
1. Type
a. Audio (MIL-T-27)
b. Power (MIL-T-27)
c. High Power Pulse (MIL-T-27)
d. Low Power Pulse (MIL-T-21038)
e. IF (tIL-T-55631)
f. RF (MIL-T-55631)
g. Discriminator (MIL-T-55631)
II. Device Construction
A. Coils
1. Numbers of Coils per Device
2. Core Material
3. Potting Material
B. Transformers
1. Number of Primary Windings
2. Number of Secondary Windings
3. Core Material
4. Potting Material
III. Inductance (coils only)
IV. Quality Level
A. S D. M
B. R E. MIL-SPEC
C. P F. Lower
V. Temperature
A. Rated
B. Actual
VI. Application Environ~ment
VII. Number of Power On/Off Cycles per 103 Nonoperating Hours
5-179
The summarized inductornonoperating reliability data collected in

support of this study is presented in Table 5.5.2-2. The data consists of
73 individual data records, with 87 failures and 49,833 x 106 part hours.
TABLE 5.5.2-2: INDUCTOR NONOPERATING FAILURE RATE DATA
Data Part Hours

" Equipment/Source Rccords Style Failures (X 106)
Hawk/MICOM 3 RF Coil 2 935.8

Hawk/MICOM 1 Discriminator 0 17.0
Hawk/MICOM 3 RF Transformer 0 204.2
Hawk/MICOM 1 Power Transformer 0 17.0
Maverick/MICOM 1 RF Coil 0 329.4
Maverick/MICOM 1 Transformer 0 41.2
Sparrow/MICOM 3 RF Coil 0 599.7
Sparrow/MICOM 4 Transformer 0 140.4
Sprint/MICOM 6 RF Coil 0 118.1
Sprint/MICOM 5 Power Transformer 0 17.4
Sprint/MICOM 5 RF Transformer 0 65.8
Sprint/MICOM 2 AF Transformer 0 9.7
Sprint/MICOM 3 Pulse Transformer 0 21.3
TOW/MICOM 2 Transformer 0 3.7
Martin Marietta 2 RF Coil 0 291.2
Martin Marietta 2 Transformer 12 3437.3
F-16 HUD/RIW 5 RF Coil 5 1831.8
AFCIQ 4 RF Coil 1 243.4
AFCIQ 9 Transformer 18 429.0
Southern Tech. 3 RF Coil 16 32235.9
Southern Tech. 8 Transformer 33 8843.7
Totals 73 87 49,833.0
The data was evaluated initially with each data record separate (no
data records merged). Point estimate failure rates were computed for data
records with observed failures. The methods described in Section 3.3 were
applied to determine which zero failure data records had sufficient part
J hours to be included in the analysis, and to estimate failure rates for
those records. Twenty data records in total were found to have either
without observed failures.
)L
C.T"
5-180

restricted because of the nature of the available data. The following __
Sattributes
of the nonoperating inductor data were observed from the
preliminary data analysis:
1) There was not a sufficient range of environments in the data to

develop a series of environmental factors based solely on the
data.
2) There was an insufficient range of ambient temperature to study

3) Data were not available in sufficient quantity to evaluate

specific device styles within the generic categories of coils and
S~transformers. 1i
4 The previously derived theoretical model was further examined to
determine the relative effect of the independent variables, leading to the

selection of variables. Failure mechanism acceleration factors were
studied to aid in the variable evaluation process.

that inductor device style, quality level, apr'cation environment and
equipment power on-off cycling frequency were the dominant variables
effecting inductor nonoperating f~ilure rate. These variables, with the
exception of application enviornment, were further analyzed using
statistical methods. Other application and construction variables were
* considered to have minimal effect on inductor nonoperating failure rate.
- It was assumed that application environment was an important factor

effecting inductor nonoperating failure rate. Therefore, it was essential
that a proposed inductor nonoperating failure rate prediction model
included an appropriate environment factor to properly discriminate
against known failure accelerating factors. Section 4.5 proposes a method
to develop appropriate environmental factors based on a comparison of
operating and nonoperating failure mechanisms, and a study of the
documented MIL-HDBK--2170 operating environmental factors. This method was
5-181
4No
~ I%
applied to the inductor nonoperating failure rate mo",- after

statistically analyzing the other variables.
The fact that the effect of temperature could not be emp, ic~lly
analyzed was also a problem. However, after careful consideration, ".was Al
decided not to include temperature in the inductor failure rate predicv n
model. This decision was based on the following observations:
o A temperature relationship could not be determined from the data.

Therefore, any proposed temperature factor would be based on
assumptions, which introduce inaccuracies.
o No theoretical relationship for inductor nonoperating failure rate
vs. temperature could be located in the literature.
o There is no basis for the assumption that the nonoperating failure

rate behaves similarly to the operating rate with respect to
temperature. The stresses on the part are necessarily different
due to the absence of an applied current.
o The average effect of ambient temperature will be included in the
environmental factor since average ambient temperature changes
with application environment.
The temperature dependence of inductor nonoperating failure rate
was believed to be much less than the temperature dependence of
microcircuits and discrete semiconductors, which do have a
temperature factor in their proposed model.
The absence of a temperature factor for inductors was therefore not

expected to have a significant effect on equipment level failure rate
predictions.
To allow correlation coefficient analysis and regression analysis to

be applied to the data, matrices of "dummy variables" (0 or 1) were
defined for device style and quality level. These analyses require o!
quantitative variables and device style and quality are qualitative
variables. Table 5.5.2-3 illustrates the matrix for device quality level. •. ,
The matrix for device style was simply a "0" for coils and a "1" for •.
transformers. This varialle was designated as S1 . The device style
5-182
TV 11 .1. It~~p
matrix was not defined to accommodate a further breakdown of the styles

since sufficient detail in the data were not available.
TABLE 5.5.2-3: DEVICE QUALITY VARIABLE MATRIX
Device Quality Q1 Q2
Hi-Rel 0 0
MIL-SPEC 1 0
Lower 0 1
Data were then merged according to equipment, device style and quality
level by summing failures and part hours. For example, all data for Hi-
Rel coils from the Hawk missile were merged. The correlation coefficient
analysis indicated no large correlation between independent variables for
the merged data set. Quality levels for inductors in the F-16 data were
not available. was therefore assumed that these parts were Hi-rel
It
because of the requirements of the military application (Air Inhabited).
Regression results later indicated that this assumption was reasonable.
The theoretical inductor model was nonlinear due to the assumed

relationship for equipment power cycling frequency. An iterative approach
was used to determine the nonlinear regression solution. Initially, two
Squalitative "dummy" variables were defined (cycl and cyc2) to represent
three distinct equipment power cycling rate categories within the inductor
data. For other part types, four categories where used. However there
was an insufficient spread of equipment cycling frequency values to
propose four categories for inductors. The three categories are:
o Cycling rate less than one power cycle every two years (cycl, cyc2
0,o0)
o Cycling rate greater than one cycle per year (cycl, cyc2 = 1, 0)
o Unknown cycling rates (cycl, cyc2 = 0, 1)
5-183
• I
frequencies......
These categories were designated as low, high, and unknown cycling

./•-" frequencies.
The data were then

in a form suitable for regression analysis.
Results of the initial regression analysis are given in Table 5.5.2-4. .
The dependent variable was ln(failure rate). The independent variables
introduced into the analysis were S1, Q1, Q2, cycl, cyc2, which were
previously defined.
TABLE 5.5.2-4: INDUCTOR INITIAL REGRESSION RESULTS
Standard Confidence
Variable Coefficient Error F-Ratio Limit '. .
Si 1.689 1.103 2.34 .80

Q2 2.477 1.208 4.20 .80
cycl 3.414 1.636 4.36 .75
bo -7.504 .. 4-
Device style (S1 ) and the lower device quality variable (Q2) were
significant at a 80% confidence limit. The cycling rate variable (cycl)
was significant at a 75% confidence limit. No other variables were
significant with 70% confidence. The available data did not indicate a
significant difference between nonoperating failure rate for Hi-Rel and
MIL-SPEC coils. This was unusual because screening was expected to have a
significant effect. The R-squared value for the regression was .69. That
is, 69% of the variability in the observed data can be explained by the
regression solution. Interpretation of the regression results are:
1. The nonoperating failure rates for coils and transformers are •. i

significantly different from each other with a 80% confidence •>
limit.
2. There were insufficient data to determine the difference between .

Hi-Rel and MIL-SPEC coils.
3. The lower device quality level was significantly different from
Hi-Rel and MIL-SPEC at 80% confidence limit. ..
5-184 ~'
* MT;;
4. The cycl equipment power cycling variable was significant at a 75%

confidence limit.
5. The low cycling rate variable and the unknown cycling rate
variable were statistically indistinguishable.
Although cycling frequency was unknown for several data entries, it

was not surprising that the unknown cycling data entries would be
statistically indistinguishable from the low cycling frequency data
entries. It was known that this data were from missile storage
applications where cycling were relatively infrequent.
The coefficients ir Table 5.5.2-4 were the result of a regression with

the dependent variable equal to the natural logarithm of failure rate.
Transforming the regression solution into an equation where failure rate
is the dependent variable results in the following multiplicative model
Xi Xnb TNQ wcyc E k.

where
6
Xi = predicted inductor nonoperating failure rate (failure/lO
nonoperating hours) L i
Xnb = base failure rate, preliminary
= exp(-7.509 + 1.689(S 1 ))
.000551, Coils
.002982 Transformers
'TNQ = quality factor
= exp(2.477(QI))
1.0, Hi-Rel and MIL-SPEC
: 11.91, Lower
itcyc = equipmeat power c'/,fling factor (preliminary)
= exp(3.414(cycI))
= 1, low cycling rate (= .039 cycles/lO3 hours)
= 30.386, high cycling rate (= 40.12 cycles/.0 3 hours)
C = residual, _
5-185
The variables S1 , Q1, and cycl have been previously defined. The
above equation represents a preliminary model, applicable only to ground
based environments.
The next step in the model development was, to dpply iterative

regression analyses to quantify the assumed nonlinear model form. As
previously discussed, the effect of equipment power on-off cycling was
represented by an equation of the following form:
Wcyc = 1 + Kl(cycles/1O3 hours)
The data in the first category, low cycling rate, was primarily from
the Hawk missile program, with an average cycling rate of .038 cycles/103
hour. The data in the second category, high equipment cycling, was from
the F-16 HUD RIW data with a cycling rate of 40.12 cycles/lO3 hour. The
following equation for a power cycling factor resulted:
K2wcyc = .972 + .730(cycles/10 3 hour)

Icyc = 1 + .751(cycles/10 3 hour)
S....
.where" •
K2 normalization constant
: 1/.972 1.029
Only one iteration was required for changes in the regression solution
to be negligible. The preliminary model at this stage of the development
process is given by:
)k nb wNQ 1cyc ):':
where1 N
'nb base failure rate

.000555, coils
Y/k-
= .00328, transformers
"1NQ = nonoperating quality factor
- 1.0, Hi-Rel and Mil-spec
= 11.36, lower
wcyc = inductor equipment power cycling factor
= 1 + .751(cycles/10 3 hour)
Transformers were anticipated to be one of the more sensitive part

types to equipment power on-off cycling. The nonoperating failure rate of
coils was also anticipated to be sensitive to equipment power on-off
cycling. However, the relative effect was thought to be less for coils. LL
Unfortunately, there were insufficient data to determine unique factors
for bcth part classes. It was strongly believed that the proposed models
should be physically correct. Therefore a separate, hypothesized power
cycling constant was proposed for coils. The hypothesized value was one
half of the observed constant for the inductor family.
The observation that coil nonoperating failure rate was statistically

indistinguishable for HI-Rel and Mil-Spec devices was further studied.
The fact that the lower quality variable (Q2) was significant at a 75%
confidence limit was considered to be sufficient rationale to assume tnat
quality provisions result in a lower observed nonoperating failure rat-.
Theoretically, screening effectiveness properties supported the premise
that screening improves failure rate. Also, a complete series of
nonoperating quality factors were required to provide the models with
maximum utility and to properly discriminate against known quality related
factors.
To determine a complete series of coil nonoperating quality factor

values, the point estimate quality factors for Hi-Rel and Mil-Spec quality
levels were assumed to correspond to an average quality level of M
(according to applicable ER specifications). Only, Nil-spec and lower
quality levels are applicable for transformers. Coil nonoperating quality
factors for S, R, P and Mil-spec were determined by extrapolation by
5-187
/ /
assuming (1) the observed nonoperating quality factors are accurate for M
and lower quality levels, (2) nonoperating quality factors would follow a
similar ranking to the MIL-HUBK-217D coil operating quality factors, and
(3) the nonoperating quality factors are distributed geometrically. Table
5.5.2-5 presents point estimate nonoperating quality factors, upper and
lower 90% confidence interval values, and extrapolated nonoperating
quality factors for S, R, P, and Mil-spec. Transformer data were only for
Mil-spec and lower categories. Therefore, the preliminary transformer
base failure rate was divided by the Mil-spec nonoperating quality factor
of 3.08. In this maner, one nonoperating quality factor can be used for
all inductive devices.
TABLE 5.5.2-5: INDUCTOR NONOPERATING QUALITY FACTORS

Quality Level wNQ, .05 TNQ1(1) TNQ,.95 WNQ2(I)
S .... .06
R .. .15
P --- -. 38
N - 1.0 -- 1.00
MIL-SPEC ...... 3.08
Lower 1.37 11.36 94.51 11.36
NOTES: (1) WNQ1 are the observed nonoperating quality factors. wNQ2
are the observed factors supplemented by extrapolated values
for S, R, P and MIL-SPEC qualities.
The next task in the inductor model development process was to

determine nonoperating environmental factors by the methods described in
Section 4.5. This method assumes that a series of nonoperating -l'
environmental factors can be generated from the MIL-HDBK-2170 operating
environmental factors. The proposed indicator nonoperating fa.tors were
presented in Table 5.5.1-1 in the previous section. It was determined
that separate environmental factors ?ould be applied for coils and
transformers.
5-188 ,..
.3 . '.. .%
"Tables 5.5.2-6 and 5.5.2-7 show failure mode/mechanism and failure ...
acceleration factor distributions for inductors. Accurate failure
mode/mechanism information was difficult to obtain for operating
applications and essentially impossible for nonoperating applications.
Much of the information in these tables was therefore theorized, although
References 33, 41 and 42 provided important failure mode information.
The final phase of the inductive device model development process was
to normalize the preliminary base failure rates to correspond to a ground
benign environment. The preliminary base failure rates were normalized by
dividing with the respective greund fixed environmental factors. .
"Additionally a unique base failure rate value for RF transformers, power
transformers and high power pulse transformers was extrapolated by
assuming the base failure rate was the average value for all transformer
styles and using the operating failure rate relationships from MIL-HDBK-
217D. The nonoperating base failure rates were therefore determined tn
be:

.000055, lower power pulse and audio transformers
= .00028, high power pulse, power and RF transformers
.00015, RF coils
*
Normalization of the nonoperating base failure rates concluded the

model development for inductive devices. The prposed ,model is presented
in its entirety in Section 5.5.1.
5-189
\4
TABLE 5.5.2-6: RF COIL FAILURE MECHANISM DISTRIBUTION

., ., 1£
Operating Nonoperat I ng
- open wire over- voltage, current 25 - 45 0 - 5

stress
faulty lead vibration, shock 5 25 20 - 40
short insulation voltage, humidity, 5 - 20 5 - 20

breakdown temperature
unstable, insulation temp., humidity 20 - 40 40 - 65

drift deterioration
TABLE 5.5.2-7: TRANSFORMER FAILURE MECHANISM DISTRIBUT!ON
j Operating Nonoperating
open wire over- voltage, current 15 -35 0 -5
stresst
faulty leads vibration, shock 0 - 15 0 - 15
short corroded humidity, temp. 15 - 35 20 - 45
windings
insulation voltage, humidity, 15 - 35 20 - 40
breakdown temp.
insulation humidity, temp. 10 - 30 20 " 40
deter ioration
5-190
/
5.6 Lasers
5.6.1 Proposed Laser Nonoperating Failure Rate Prediction Models
This section presents nonoperating failure rate

the proposed
prediction models for lasers. The proposed models are presented in
Appendix A in a format compatible with MIL-HDBK-217D. The proposed
nonoperating failure rate prediction models are approximate, and based on
the observed operating to nonoperating ratios for vacuum tubes.
Helium Neon Lasers

A
The proposed model for helium/neon lasers is:
Xp : .1 'NE
"where
6
X p = helium/neon laser nonoperating failure rate (failures/10
nonoperating hours)
Argon Ion Lasers
The proposed model for argon ion lasers is:

,_- p .61 1NE
•.where
,X, = argon ion laser nonoperating failure rate (failure/10 6

nonoperating hours)
".NE nonoperating environmental factor (see Table 5.6.1-1)
5-191
Carbon Dioxide, Sealed Lasers
The proposed model for CO2 sealed lasers is:
Xp = (.65 + .013(Nop)) wNE
where
Xp = CO2 sealed laser nonoperating failure rate (failures/10 6

nonoperating hours)
Nop = number of active optical surfaces
TNE = nonoperating environmental factor (see Table 5.6.1-1)
Carbon Dioxide, Flowing Lasers
The proposed model for CO2 flowing lasers is:
p= .039(Nop) ffNE
where
XP= CO2 flowing laser nonoperating failure rate (failures/10 6

nonoperating hours)
Nop = number of active optical surfaces
lINE = nonoperating environmental factor (see Table 5.6.1-1)
Solid State Lasers (Nd:YAG or Ruby Rod)
The proposed model for solid state lasers is:
p= (0.062 + .021(Nop)) 7NE
5-192
-C 9 Ila - W -C
or - - -.. -V W t tSVW t . td r
where
Xp = solid state laser nonoperating failure rate (failures/10 6

nonoperating hours)
Nop= number of active optical surfaces
wNE = nonoperating environmental factor (see Table 5.6.1-1)
The proposed model for solid state lasers is for either Nd:YAG or ruby
rod lasers, and for either xenon or krypton flashlamps.
TABLE 5.6.1-1: LASER NONOPERATING ENVIRONMENTAL FACTORS
Environment fNE Environment vNE

U GB 1 AIA 44
GF 9.0 AIF 62
GM 44 AUC 35
Mp 22 AUT 42
NSB 10 AUB 71
NS 49 AUA 57
V NU 49 AUF 71
'p NH 35 SF (1)
NUU 38 MFF 21
ARW 46 MFA 29
AIC 26 USL 69
"AIT 35 ML 71
AIB 58 CL
"NOTES: (1) Space Flight environment was not addressed in this study.
F. *5-193
--
---
---
-
5.6.2 Laser Model DevelopmentL

Li:
A series of simple laser nonoperating failure rate prediction models
were determined based on assumptions. The nonoperating failure rate model
development process consisted of five separate steps, which are: lk
1. Compute average MIL-HDBK-2170 operating failure rates.
2. Determine an appropriate operating to nonoperating failure rate

ratio.
3. Compute approximate nonoperating failure rates.
4. Assume a nonoperating environmental factor.

5. Normalize the failure rates.
The first step in the process was simple enough. Average operating
failure rates were computed for helium/neon, argon ion, CO2 , and solid
state lasers by using the methods in MIL-HDBK-217D. It was preferable to
proceed with the average failure rate values rather than assuming the
operating factors would also apply during nonoperating periods. Many of A
laser operating factors would clearly not apply during nonoperating
periods. For example, factors for pulse repetition, cooling, tube current
and pulse energy would intuitively not apply to nonoperating periods.
In general, the dominant nonoperating failure mechanisms for the

generic laser famiily are either related to the tube (i.e. laser tube for a
He/Ne laser, flashlamp for a Nd:YAG laser) or degraded optical surfaces.
The operating to nonoperating failure rate ratio for vacuum tubes was
selected because of similar tube failure mechanisms. For the electronic
vacuum tubes with data, the average ratio was 84.8 to 1. This ratio was
then applied to the average operating failure rates to determine
approximate nonoperating failure rates.
The next phase of the model development process was to assume a series
of nonoperating environmental factors. It was determined that the
operating environmental factor could not accurately be applied for
5-194
~-.-.-...~-..----..--.---V
-------------------.
,-.-.--...~ ..
nonoperating failure rate prediction purposes. Therefore, a series of

nonoperating environmental factors were generated using the methods
presented in Section 4.5. The proposed laser nonoperating environmental
factors were presented in Table 5.6.1-1 of the previous section.
The final step for laser model development was to normalize the
approximate nonoperating failure rates to correspond to a ground benign
environment. This was done to provide consistency with the other proposed
models. The normalization process consisted of dividing the failure rates
lili
by the ground fixed environmental factor of nine.
The proposed laser nonoperating failure rate prediction models must be

considered approximate because they were based exclusively on assumptions.
However, the objective of this study was not to analyze data but to
develop models which can be used to predict the nonoperating failure rate
of any equipment type. Additionally, laser target designators and range
finders with Nd:YAG solid state lasers generally are exposed to
significant periods of nonoperation. Therefore, a laser nonoperating
failure rate prediction methodology is essential.
5-195 '-
5.7 Tubes
5.7.1 Proposed Tube Nonoperating Failure Rate Prediction Model
This section presents the proposed tube nonoperating failure rate

model. The model is also presented in Appendix A in a form compatible
with MIL-HDBK-217D. The proposed model for electronic vacuum, microwave
and other tube types is given by the following equation.
p= Xnb ItNE
where
Xp = predicted tube nonoperating failure rate

Xnb = nonoperating base failure rate (failure/lO6 hrs.)
= .0040, receiver; triode, tetrode or pentode
=.0090, receiver; power rectifier {-•'
= .049, vidicon
= .013, CRT
.32, thyratron
= 1.29, crossed field amplifier
ri
= 1.03, pulsed gridded
= .56, transmitting; triode, tetrode or pentode 3!r•
= 1.61, transmitting, any style with peak power _<200 kW,

frequency < 200 MHz, or average power < 2 kW I, i
= 2.60, twystron Af
= 1.02, magnetron..
= 1.20, CW klystron
= 1.15, pulsed klystron
= .69, TWT 3
fNE = nonoperating environmental factor (see Table 5.7.1-1)
5-196
TABLE 5.7.1-1: TUBE NONOPERATING ENVIRONMENTAL FACTORS
Environment WNE Environment iTNE

GB 1 AIA 23
GF 3.0 AIF 35
GM 31 AUC 8.2
Mp 31 AUT 23
NSB 15 AUB 33
NS 29 AUA 27
NU 47 AUF 43
NH 110 SF (1)
NUU 120 MFF 63
ARW 140 MFA 91
AIC 6.2 USL 210
AIT 19 ML 220
AIB 25 CL 3600
NOTES: (1) Space Flight environment was not addressed in this study.
5.7.2 Tube Model Development
The modeling approach discussed in Section 4.1 was implemented for

Analysis of the available tube nonoperating reliability data and
H
tubes.
development of a theoretical model were used to determine the nonoperating
failure rate prediction model. The proposed model was determined to be a
function of tube style and environment.
Application and construction variables were identified which

characterize tubes in a nonoperating environment. Table 5.7.2-1 presents
a list of the device classification variables. These variables were
possible nonoperating failure rate model input parameters which were
analyzed using statistical methods where appropriate. These variables
were determined whenever possible for all collected data. If sufficient
information could not be determined for any specific data entry, then it
was removed from the model development process.
5-197
TABLE 5.7.2-1: TUBE PART CLASSIFICATION
I. Type
A. Receiver
1. Triode
2. Tetrode
3. Pentode >*
4. Power Rectifier
B. CRT
C. Vidicon
D. Thyratron
E. Crossed Field Amplifier
F. Pulsed Gridded
G. Transmitting
1. Triode
2. Tetrode
3. Pentode
H. TWT
I. Twystron
J Magnetron
1. Conventional
2. Coaxial
K. Klystron
II. Pulsed vs. CW
III. Temperature
A. Rated
V
B. Actual
IV. Rated Power
V. Frequency

VII. Number of Power On/Off Cycles per 103 Nonoperating Hours
5-198 5-,1
The nonoperating reliability concepts presented in Section 4.0 and a

study of tube nonoperating failure mechanisms were combined to determine a
theoretical model for tubes. The theoretical model assumes a nonlinear
failure rate equation which is time dependent. The theoretical
nonoperating failure rate model for tubes is represented by the following
equation (f denotes a function).
Xt f(device style, P, F, time, temp.) "NE 1cyc *
where
Xt = predicted tube nonoperating failure rate (failures/10 6 hrs.)

P = rated power
F = operating frequency
MNE = nonoperating environmental factor
rcyc equipment power on-off cycling factor 1 + KI(Nc)
where
KI = constant
Nc = equipment power on-off cycling rate (cycles/1O3 nonop. hrs.)
It was hypothesized that the theoretical tube model would be time

dependent. This assumption was based on two observations. First, many of
the anticipated storage failure mechanisms are degradation mechanisms.
For example, the dominate electronic vacuum tube failure mechanism
according to Reference I is a loss of vacuum during prolonged storage. A
loss of vacuum integrity would tend to cause failures later in a tube
nonoperating life (as opposed to early in the tube life). Even though a
microscopic leak in the tube envelope may be present early, the tube would
not actually fail as a result of the leak until a later point in time.
This mechanism would be indicative of a time increasing nonoperating
failure rate. Another tube storage failure mechanism is corrosion and
embrittlement of the filament structure. Again, the frequency of
occurrence for these failures would not be anticipated to be random with
5-199
I",
time, but would increase with time. The second observation which led to
the time dependent tube nonoperating failure rate conclusion was the
"shelf life" phenomenon. Even if nothing were known about tube storage
failure mechanisms, the fact that some tube types are believed to have
some inherent "shelf life" was considered as evidence of a time dependent
failure rate. In general cerms, it can be assumed that the nonoperating
failure rate is relatively low until the specified "shelf life" and then
the instantaneous failure rate would be expected to increase dramatically.
It should also be noted that many high quality military tube types
including TWTs and magnetrons are no longer believed to have a "shelf
life" and thus, the nonoperating failure rate could be accurately assumed
to be constant with time. However, this is certainly not true for all
tube styles considered in this study.
Unfortunately, there was little which could have done to quantify a

time dependent nonoperating failure rate model. All available data were
in the form: X observed failures in Y nonoperating part hours.
Individual nonoperating times-to-failure would have been required to
quantify a time dependent model. It is doubtful that accurate times-to- ',-
failure can ever be detected during nonoperating applications because a 7
failure generally can not be detected until power is applied to the tube.
If power were applied to the tube frequently enough to determine times-to-
failures, then the majority of failures would probably have been induced
by the frequent power on-off cycling. As a result of this lack of time to
failure information, the model development process continued by computing
average nonoperating failures. It should be cautioned that instantaneous
failure rates can differ significantly from the average for short time
intervals.
The summarized tube nonoperating reliability data collected in support L

of thi; study is presented in Table 5.7.2-2. The data consists of 47
individual data records, 364 observed failures, and 794.8 x 106 part
hours. The source for each data record was the MICOM storage data base.
Forty-seven data records were found to either have observed failures or
5-200
had sufficient part hours to estimate a failure rate without observed

fai Iures.
TABLE 5.7.2-2: TUBE NONOPERATING FAILURE RATE DATA r . .
Part
Data Records Style Failures (X1O6)
11 Magnetron 163 156.5 •,

2 TWT 11 3.6
2 Twystron 8 1.6
11 Klystron, pulsed 98 31.3
14 Klystron, CW 32 11.0
4 Pulsed Gridded 38 8.7
1 Vidicon 3 20.6
1 Receiver, Triode 1 191.4
1 Receiver, Pentode 10 370.1
47
The number of variables which could
364
be analyzed
794.8
empirically was
F
limited by the nature of the available data. The following attributes of
the collected nonoperating tube data were identified from the preliminary
data analysis.
1) There was not a sufficient range of environments to develop a

series of environmental factors based solely on the data.

3) There was insufficient cycling rate information to analyze the
effects of power on-off cycling on the nonoperating failure rate
of tubes.
It was unfortunate that equipment power on-off cycling could not be

analyzed empirically. The change in temperature resulting from turning
.the equipment power on and off would hypothetically have an adverse effect
on nonoperating failure rate. However, no relationship could be
determined empirically and no theoretical relationship could be located in
the literature. Thus, the proposed model corresponds to some average,
unknown equipment power on-off cycling frequency.
5-201

that tube style and application environment were the dominant variables _.
effecting nonoperating failure rate. Variables such as rated power and
frequency were believed to have little effect on nonoperating failure
rate. The effect of style was further analyzed using statistical
techniques. It was assumed that application environment was an important ___
parameter effecting nonoperating failure rate. Althuugh a series of

environmental factors could not be derived from the data, it was essential i-
to include applicable factors in the proposed tube nonoperating failure
rate prediction model. This was essential to properly discriminate
against known failure accelerating environmental factors. Section 4.5
proposes a method to develop appropriate nonoperating environmental
factors. This method was applied to tube nonoperating failure rate model
after evaluating the effect of tube style.
A matrix of "dummy variables" (0 or 1) was defined for tube style so

that correlation coefficient analysis and regression analysis could be , 44
applied to the data. These aralyses require quantitative variables and
device style is a qualitative variable. The matrix for tube style is
il'ustrated in Table 5.7.2-3.
TABLE 5.7.2-3: DEVICE SfYLE VARIABLE MATRIX
Device Style 51 S2 S3 S4 S5 S6 S7 S8
Klystron, Pulsed 0 0 0 0 0 0 0 0
Twystron 1 0 0 0 0 0 0 0
Klystron, CW 0 1 0 0 0 0 0 0
Magnietron 0 0 1 0 0 0 0 0
TWT 0 0 0 1 0 0 0 0 t-
Receiver, Triode 0 0 0 0 1 0 0 0
Vidicon 0 0 0 0 0 1 0 0
Receiver, Pentode 0 0 0 0 0 0 1 0
Pulsed Gridded 0 0 0 0 0 0 0 1
V
5-202
r~~~~"
- *-4i *.
As the model development process continued, regression analysis was

then applied to the data. Results of the regression analysis are given in
Table 5.7.2-4. The dependent variable was the natural logarithm of
nonoperating failure rate. The independent variables were S1 through S8.
Device style variables S5 , S6, and S7 were significant at a 90% confidence
limit. S1 was significant with 70% confidence, and S4 with 50%
confidence. The R-squared value for the regression was .65, so 65% of the
variability in the observed data can be explained by the regression
solution. There was no significant difference, with 50% confidence,
between the nonoperating failure rate for pulsed klystrons, CW klystrons,
magnetrons and pulsed gridded tubes.
TABLE 5.7.2-4: TUBE REGRESSION RESULTS
Standard Confidence
Variable Coefficient Error f-ratio Limit
SI .819 .770 1.13 0.70

S2 .147 .404 .01 <0.50
S3 -.115 .427 .07 <0.50
S4 -. 512 .770 .44 0.50
S5 -6.491 1.047 38.46 0.90
S6 -3.163 1.047 9.13 0.90
S7 -4.848 1.047 21.45 0.90
"S8
-. 106 .585 .03 <0.50
bo 1.237 -- -- --
Regression solutions below 50% were then computed to further evaluate

the nonoperating failure rate of CW klystrons, magnetrons, and pulsed
gridded tubes. As a general practice in this study, unique failure rate
modifying factors were not proposed for variables which did not
"significantly effect nonoperating failure rate with greater than 70%
confidence. However, an exception was made to this general rule for
tubes. These various tube styles belong to the same generic part family,
but are physically very different. Therefore it was decided to propose
the best estimates available for nonoperating failure rate regardless of
the significance level.
5-203'

the nonoperating failure rate. Transforming the regression results into
an equation where nonoperating failure rate (as opposed to log of
nonoperating failure rate) was the dependent variable resulted in the
following preliminary multiplicative model.
Xt exp(1.237 + .819(S 1 ) + .147(S 2) - .115(S 3) - .512(S 4 ) - 6.491(S 5 )

- 3.163(S 6 ) - 4.848(S 7 ) - .106(S 8 ))
= 3.44, klystron, pulsed

= 7.81, twystron
= 3.61, klystron, CW
3.07, magnetron
=2.06, TWT
= .00523, Receiver, triode

= .146, Vidicon
.0271, Receiver, pentode
= 3.10, Pulsed gridded
The next phase of the model development process for tubes was to
develop appropriate nonoperating environmental factors using the methods
presented in Section 4.5. Tubes presented a unique situation for
determination of nonoperating environmental factors. The anticipated
failure mechanisms in the operating state are dominated by failures which
are accelerated by the internal heat rise due to either the current flow
from one element of the tube to another element, the power used to raise
an electron-emitting cathode to operating temperature, or another source
of internal heat rise. Excessive heat can result in several general
failure mechanisms including deterioration of the seal, wearout of
electron emission surfaces, evolution of gas, and contaminated or damaged
emission surfaces resulting in increased electronic emission. It was
concluded from the failure mechanism comparisons that the MIL-HDBK-217D
operating environmental factors clearly do not represent the effects of
environmental stress in a nonoperating application.
5-204
iii
Part types which have operating failure rates which are dominated by
factors unique to the operating state were categorized in Section 4.5.
Tubes, as well as incandescent lamps and laser tubes, fit into this
category. For these part types, the difference between operating and
nonoperating failure rate was believed to be relatively large. However,
the rate of increase for failure rate versus environmental stress would be
greater in the nonoperating state. Failures due to purely environmental
stresses, such as vibration and mechanical shock, would not occur in
greater numbers in the nonoperating state, but these failures would be a
much larger percentage of the total failures. Thus, a similar increase in
the number of failures due to a more stressful environment would have a
much more apparent effect for nonoperating applications. Proper
determination of applicable nonoperating environmental factors was not
possible with the available information. However, an approximate formula
was developed (and described in Section 4.5) to estimate nonoperating
. environmental factors for these part types.
The nonoperating environmental factor conversion was based on the MIL-

HDBK-217D operating environmental factors, the fraction of failures due to
purely operational stress, the fraction of
failures due to purely
environmental stress, and the fraction of
total failures due to
simultaneous exposure to operational/environmental stresses. To compute
the environmental factors, it was estimated that 50% of operating tube
failures could be attributed to entirely operational stress. The
remaining percentage was attributed to combined operational/environmental
stress or other factors such as environment related temperature rises.
The resultant factors were presented in Table 5.7.1-1 of the previous
section. It is emphasized that these values are approximate. However,
the described approach did not ignore the observation that the operating
and nonoperating environmental factors would not be expected to be the
same for tubes. Therefore, this approach was preferable to the erroneous
assumption that the environmental factors would be the same.
Additionally, the tube nonoperating environmental factors were normalized
to a ground benign value equal to one to be consistent with the other
"models presented in this report.
5-205
The next phase of the tube .•V'ie d-'-elupment process was to

extrapolate nonoperating failure rates for those ,.ube styles without data.
Table 5.7.2-5 presents the observed rnonope-r•ting failure rate, the
corresponding MIL-HDBK-217D operating f•;lure rate for a ground fixed
environment, and the ratio between the two. The results of this exercise
were very interesting. First, the correlation coefficient for operating
and nonoperating failure rate was found to be 0.73. This was considered
to be extremely high in consideration of the inherent failure rate
variability. The high correlation coefficient provides justification for
using the operating failure rates to extrapolate nonoperating failure
rates for tube types without data. Another important observation was the
ratios fell into two categories. There was a high ratio (geometric mean =
374) for receiving tubes and vidicons, and a somewhat lower ratio
(geometric mean = 51.7) for the remaining tube styles. No explanation for
this apparent difference was determined. However, it was decided to group
the tube styles without data into one of the two groups based on device
construction. CRTs was grouped with vidicons and receiving tubes.
Thyratrons, crossed field amplifiers, and transmitting tubes were grouped
with the remaining tube types. The appropriate ratios were then used to
determine ground fixed nonoperating failure rates.
The final phase of nonoperating failure rate model development was to

normalize the ground fixed failure rates to correspond to a ground, benign
environment. It was desired that all models presented in this report be
normalized to a ground benign environment to provide consistency. The
I
normalization was done by dividing the previously presented failure rates.
by the ground fixed nonoperating envi'ronmental factor of three.
Normalizing the tube nonoperating failure rate concluded the model
development process. The proposed models were based on data in ground

based environments and accurately predicts the nonoperating failure rate
in those environments. A series of nonoperating environmental factors
were assumed based largely on intuitive reliability relationships These
factors should be evaluated when additional tube nonoperating failure rate
data becomes available for airborne and other stressful environments.
5-206
r'1
TABLE 5.7.2-5: TUBE OPERATING AND NONOPERATING FAILURE RATE COMPARISON
Failure Failure
Rgte Rage Ratio
Tube Type (f/lO hrs.) (f/10 hrs.) (Xop/xnonop)
Receiver
Triode, Tetrode, Pentode 5 .012(2) 417
Power Rectifier 10 ..--
Vidicon 49(1) .146 336
CRT 15 --..
Thyratron 50(2) ....
200(2) -- --
Crossed Field Amplifier
Pulsed Gridded 200(2) 3.10 64.5
Transmitting --
Triode, Tetrode, Pentode 87(2)
250 -- --
Other (3)
Twystron 460(2) 7.81 58.9
Magnetron 435(2) 3.07 141
CW Klystron 68(2) 3.61 18.8
Pulsed Klystron 84(2) 3.44 24.4
160(2) 2.06 77.7
TWTs
NOTES: (1) Vidicon failure rates from Reference 45.

(2) Geometric mean is given.
(3) Any transmitting tube with peak power < kW, frequency • 200
MHz, or average power < 2 kW.
k
"i
5-207
5.8 Mechanical/Electromechanical Devices
There are a number of families of electromechanical devices that are

susceptible to degradation damage when exposed to many of the military
environments for long periods of time in a nonoperating state. In some
cases this degradation may be more severe than the operating effects in
the same environment. The device families in question include switches,
relays, electric motors, servomechanisms, and connectors.
The following sections present the proposed nonoperating failure rate

prediction models for mechanical and electromechanical devices.
Additionally, discussions are presented which address the nature of the
degradation effects and actions which may ameliorate the impact of long
term exposure in a nonoperating state. The elements of the environment
which cause degradation include vibration, shock, ambient temperature,
humidity, and airborne contamination such as salt, sulfur and products of
petroleum combustion.
Very little
nonoperating failure rate data were collected for
Smechanical/electromechanical devices. Therefore, model development for
these devices was based on intuitive nonoperating reliability
relationships supplemented by whatever data were available. It was
decided to propose simple models for these part types because of the
general lack of quantitative information. More complex assumed models
would not necessarily be more accurate but would delude the model users''
into thinking that they were.
.5.8.1 Proposed Mechanical/Electromechanical Device Nonoperating

Failure Rate Prediction Models

prediction models for mechanical and electromechanical devices. The
devices studied are those mechanical and electromechanical devices
included in MIL-HDBK-217D. Separate models were developed for relays,
switches and connectors. A table of average nonoperating failure rates
5-208
was determined for rotating mechanisms. The proposed models are presented
in Appendix A in a format compatible with MIL-HDBK-217D.
Rotating Mechanisms ,
Table 5.8.1-1 presents average nonoperating failure rates for motors,

synchros, resolvers, and elapsed time meters. The average nonoperating
failure rate for motors pertains to motors with power ratings below one
horsepower.
TABLE 5.8.1-1: ROTATING DEVICE AVERAGE NONOPERATING FAILURE RATES
Nonoperating Failure Rate

Device (failures/lO6 nonoperating hours)
Motors (ac or dc) .045

Synchros .14
Resolvers .14
Elapsed Time Meters 1.2
Relays
The proposed nonoperating failure rate prediction model for relays is:
Xp =nb lTNQ 'TNE
where
X• relay nonoperating failure rate

* Xnb = nonoperating base failure rate (failures/lO6 hours)
= .010, nonhermetic, contact voltage < 50 millivolts when operated
= .002, nonhermetic, contact voltage > 50 millivolts when operated
- .0004, hermetic packaged relays
TNQ nonoperating quality factor
- 0.46, established reliability
5-209
= 1.0, Mil-spec
= 4.2, lower EA__
7rNE = nonoperating environmental factor (see Table 5.8.1-2)
TABLE 5.8.1-2: RELAY NONOPERATING ENVIRONMENTAL FACTORS ~3
Environment iTNE Environment irNE

GF 2.3. AIF 10
GM 8.2 AUG 8.0

MP 21 AUT 9.0 A
NSB 8.0 AUB 15
NS 8.0 AUA 10
NU 14 AUF 15
NH 32 SF (1.)
NUU 34 MFF 21
ARW 46 MFA 29
AIC 5.5 USL 62
AIT 6 ML 71
AIB 10 CL N/A
NOTES: (1)Space Flight environment was not addressed in this study.
Swi tches
The proposed nonoperating failure rate prediction model for switches

is:
Xp Xnb wNQ WrNE
where
xp= switch nonoperating failure rate

=.030, contact voltage <50 milliamps when operated
5-210
/N
S.006, contact voltage > 50 milliamps when operated

1rNQ = nonoperating quality factor
= 0.46, established reliability
= 1.0, Mil-spec
= 4.2, lower
TABLE 5.8.1-3: SWITCH NONOPERATING ENVIRONMENTAL FACTORS
Environment ITNE Environment 9NE

GB 1 AIA 14
GF 2.9 AIF 18
GM 13 AUC 9
Mp 21 AUT 9
NSB 7.9 AUB 18
NS 7.9 AUA 18
NU 18 AUF 23
NH 32 SF (1)
NUU 34 MFF 19
ARW
AIC
AIT
41
7.2
7.2
MFA
USL
ML
26
63
64
I
AIB 14 CL 1200
Connectors
The proposed nonoperating failure rate prediction model for connectors

is: Li
p= nb ITNE
5-211
* * ~ ~ l ,
where
connector nonoperating failure rate

-
=p
Xnb = connector nonoperating base failure rate (failures/10 6 hours)

.00044, circular, coaxial and power connectors
= .0029, rack and panel, and printed wiring board connectors
WNE = nonoperating environmental factor (see Table 5.8.1-4)
TABLE 5.8.1-4: CONNECTOR NONOPERATING ENVIRONMENTAL FACTORS k
Environment irNE Environment •NE

GB 1 AIA 5.3
GF 2.3 AIF 11
GM 8.3 AUC 4.3
Mp 8.5 AUT 15
NSB 4.1 AUB 9.8
NS 5.5 AUA 8.0
NU 13 AUF 15
NH 13 SF (1)
NUU 14 MFF 8.5
ARW 19 MFA 12
AIC 2.8 USL 25
AIT 4.8 ML 29
AIB 7.0 CL 490
5.8.2 Rotating Mechanisms
The rotating mechanisms included in MIL-HDBK-217D are motors, m

synchros, resolvers and elapsed time meters. Average nonoperating failure I
rates were determined for each of these device types. Additionally, a
qualitative assessment of rotating mechanisms was performed, and is'1
described in this section.
5-212
21
The following discussion is limited to low torque electric motors and

servomechanisms, their nonoperating degradation mechanisms, and the
associated environmental causal stresses.
Ball bearings are normally used to support the rotating member in

those devices using grease as a lubricant. Prior to assembly of the
bearings into the device, they are uniformly packed with grease. In the
first few rotations of the device, the balls form a channel in the grease.
This channel, if uniform without voids or discontinuities, serves to
concentrate the oils released from the grease on the wear path insuring
good lubrication and long bearing life. The grease must be uniformly
packed initially or a continuous channel will not be formed and short life
will result. Prolonged nonoperating periods can effect the uniformity of
the packed grease. In the nonoperating state, temperatures above 700,
vibration, or both will cause the grease to sag resulting in a nonuniform

channel when the device is again operated.
It is recommended that these devices be run up to speed every six

months when exposed to 70OF or higher temperatures. Where these rotating-i
mechanisms are simultaneously exposed to temperatures in excess of 70OF
and vibration, the time between run ups should not exceed three months. ,
If the time between run ups is longer than these recorinmended intervals, it
would be anticipated that the nonoperating failure rate would be higher.
Commutator devices such as DC motors and certain servomechanisms may

develop resistive coatings on the commutator much as contact devices
(discussed in Section 5.8.3). Prolonged storage results in the formation
of resistive surface films. When less than six volts potential exist,
running these devices up to speed every three months will also preclude
erratic operations.
It was hypothesized that device style would be the most significant

variable effecting the nonoperating failure rate of motors. Commutator
motors have two major nonoperating failure mechanisms; contamination of
5-213
commutative surfaces and loss of a uniform lubricant. Conversely, the L

loss of lubricant is the dominant mechanism acting on non-commutator
motors. In general, AC motors are non-commutator devices. DC motors may
be either commutator or non-commutator.
Nonoperating failure rate data for rotating mechanisms were available

from MICOM. This data are presented in Table 5.8.2-1. The data
collectively consists of 20 observed failures in 149.0 x 106 nonoperating
part hours. There were insufficient data to statistically analyze the
effects of independent variables with regression analysis.
TABLE 5.8.2-1: ROTATING MECHANISM NONOPERATING FAILURE RATE DATA
Part Hours
Equipment/Source Device Style Failures (X10 6 )
Martin Marietta AC motor 1 2.16

Martin Marietta torque motor 0 4.16
Martin Marietta resolver 2 14.20
MICOM DC motor 3 6.34
MICOM AC motor 0 0.15
MICOM DC motor 0 55.25
Maverick/MICOM torque motor 14 41.18
Sparrow/MICOM DC motor 0 25.52
Totals 20 148.96
The nonoperating failure rate for the merged AC motor data was over 12
r
times as high as the nonoperating failure rate for the merged DC motor
data. This observation was contrary to the earlier hypothesib that
commutator motors (if some of the DC motors were commutators) should have
a higher nonoperating failure rata. It was assumed that there were
insufficient data to accurately distinguish the difference between
coffutator and non-commutator motor types. Therefore, an average
nonoperating failure rate was computed by dividing the sum of the failures
by the sum of the part hours for both AC and DC motor styles. The average
nonoperating failure rate was therefore found to be 0.0447 failures per
106 nonoperating hours.
5-214
x,,
77t
Similar to the discussion for motors, there were insufficient data for
synchros and resolvers to develop nonoperating failure rate prediction
models analytically. Therefore, an average nonoperating failure rate of
0.14 was determined from the data. It was then assumed that this value
was applicable for all styles of synchros and resolvers.
No nonoperating failure rate data were available for elapsed time

meters. Therefore, an average nonoperating failure rate was determined by
(1) calculating an average operating failure rate using MIL-HDBK-217D, (2)
determine an operating to nonoperating ratio for rotating devices using
the data for other devices, and (3) computing an average, approximate
failure rate. The applicable ratio was determined to be 54:1 and the
average nonoperating failure rate for elapsed time meters was computed to
be 1.2 failures/10 6 nonoperating hours.
Failure rates based on small sample sizes are potentially biased.

However, the average nonoperating failure rates presented in this section
for motors, synchros, resolvers and elapsed time meters represent the best
possible values given the data constraints. It is recommended that these
average failure rates be evaluated when additional data resources become
available.
5.8.3 Contact Device Model Development
For the purpose of this study, contact devices were considered to be

switches and relays. The desired analytical model development approach
could not be applied for contact devices because of a general lack of
quantitative nonoperating failure rate data. Therefore, intuitive
nonoperating reliability relationships formed the basis of the model
development process. The limited data supply was used to complement the
intuitive relationships. The proposed nonoperating failure rate
prediction model for relays was determined to be a function of package.
type, contact voltage, quality level and application environment. The
switch nonoperating failure rate prediction model was found to be a
function of contact voltage, quality level and application environment.
5-215
1K`;-'z
A list of construction and application variables were determined for

both relays and switches. These variables are presented in Table 5.8.3-1
and 5.8.3-2 respectively. In each case, these variables represent
possible nonoperating failure rate prediction model parameters.
Switches arnd nonhermetic relays utilized to control low voltage and

current (typical of digital circuits) are of concern regarding
nonoperating reliability. This type of operation is called low level •
circuit operation where the voltage and current are less than 50
millivolts and 50 microamps respectively. rhe contacts are particularly a
reliability concern when contacts are normally open. The voltage in
digital circuits is too low to produce an arc on contact closure which
would break thru resistive surface films which form during nonoperating
periods. The transfer is by a tunneling mechanism which can not take
place thru a resistive surface film. Therefore, the contact
voltage/current and the number of normally open contacts are important
variables for predictingthe nonoperating failure rate for nonhermetic .
c3ntact devices. Conversely, contact voltage/current woild not be
anticipated to be a significant variable for hermetic relays because the
hermetic seal would preclude the formation of substantial resistive
surface films.
The rate of build up and nature of the surface resistive film is a

function of the environment to which the contacts are exposed, as well as
the contact material. The ability of t•e contact to remove a surface film
depends on the design, contact material, surface roughness, and contact
pressure.
From the previous paragraph, it can be said that reliable operation -, -
depends on maintaining clean contact surfaces and that long termn. ---
nonoperating conditions may defeat this objective. The alternative
approach to prevent nonoperating contact failures is periodic actuation
when long periods of nonoperation are the rule. The literature contains
no quantitative information in this regard. A survey of leading U-'--
manufacturers was more helpful in a qualitative sense, but not
5-2 16
.-. . .- . . . ~ * *. law-
TABLE 5.8.3-1: RELAY PART CHARACTERIZATION
I. Application Type
A. Low Level Circuit G. Time Delay

B. General Purpose H. Latching
C. Sensitive (0-100 mw) I. High Power
D. Polarized J. Medium Power
E. Vibrating Reed K. Contactors
F. High Speed
II. Construction
A. Armature E. Magnetic Latching
B. Dry Reed F. Vacuum
C. Mercury Wetted G. Mechanical Latching
D. Solenoid
III. Configuration
A. SPST E. 4PST
B. DPST F. DPDT
C. SPDT G. 4PDT
D. 3PST H. 6PDT
IV. Contact Rating (Amps)
V. Operating Load Type

A. Resistive
B. Inductive
C. Lamp
VI. Quality Level

A. R D. L
"B. P E. Mil-Spec
C. M F. Lower
VII. Temperature
A. Rated
B. Actual
IX. Number of Power On/Off Cycles per 103 Nonoperating Hours
5-217
TABLE 5.8.3-2: SWITCH PART CHARACTERIZATION
I. Device Style
A. Toggle
B. Push Bottom
C. Rotary
D. Basic Sensitive
E. Heavy Duty Contactor
F. Inertial
II. Actuation Style

A. Snap Action
B. Non-Snap Action
III. Configuration
A. SPST E. 4PST
"B. DPST F. DPDT
C. SPOT G. 4PDT
D. 3PST H. 6PDT
IV. Operating Load Type
A. Resistive
B. Inductive
C. Lamp
,: V. Contact Rating (Amps)
VI. Number of Switch Positions
VII. Actuation Differential
VIII. Temperature
A. Rated
B. Actual
IX. Application Environment
X. Number of Power On/Off Cycles Per Unit Time,
XI. Quality
A. Mil-Spec
B. Lower
XII. Wafer Material

A. Ceramic
B. Glass
5-218
quantitatively. The manufacturers all agreed to the wisdom of periodic

actuation. However, due to varying experience, the manufacturers
contacted had different views on the optimal frequency of actuation.
Frequencies were expressed ranging from daily for a dirty industrial
environment to every six months in a computer room like environment. None
of the manufacturers contacted took exception to monthly actuation as
being adequate for all but worst case environments.
The philosophy of periodic actuations for contact devices is the

reverse of the observation that equipment power on-off cycling degrades
electronic component device reliability. For contact devices, it would be
A anticipated that as equipment power on-off cycling increases, the observed
.4 nonoperating failure rate due to the formation of resistive films would
decrease. This trend would be expected for all equipment power on-off
cycling frequencies encountered during equipment storage. Fo~r equipments
which are power cycled frequently (e.g. every day), the observed failure
rate would be anticipated to begin to increase due to contact wear.
However, it was hypothesized that the effect of equipment power on-off
cycling was not a significant variable over the range of equipment power
cycling frequencies typically found during normal usage. The beneficial
effects of periodic actuation would be approximately canceled by the
$1 effects of contact wear. Therefore, the proposed models for contact

devices did not include an equipment power on-off cycling factor.
As a result of the qualitative assessment of contact devices, it.was

determined that the, dominant variables effecting relay nonoperating
failure rate were package type (i.e. hermetic vs. nonhermetic), contact
voltage/current, quality level, environment and the number of normally
open contacts. It was believed that these variables were more significant
than the respective differences between specific relay styles. It was
7_7 anticipated that there would not be a large nonoperating failure rate
difference between specific relay styles because the dominant nonoperating
failure mechanisms act on the contacts (which are relatively similar), and
do not act on the actuation mechanisms (which vary considerably).
5-219
The nonoperating failure rate data collected for relays is presented

in Table 5.8.3-3. The data were merged according to data source,
construction and quality level. Merged data records with zero failures
were evaluated by the methods presented in Section 3.3 to determine
whether there were sufficient part hours to estimate a nonoperating
failure rate. There were a total of seven merged data entries available
for analysis. This was considered to be insufficient data to quantify the
effects of the independent variables by use of regression analysis.
TABLE 5.8.3-3: RELAY NONOPERATING FAILURE RATE DATA
Part
Hours
Equipment/Source Construction Quality Failures (X106)
"Maverick/MICOM Armature Mil-spec 2 61.7
Maverick/MICOM Other Mil-spec 0 20.6
Sparrow/MICOM Other Mil-spec 2 382.8
"Lance/MICOM Armature Mil-spec 0 4.5
MICOM Other Lower 1 3.6
Hughes Armature Mil-spec 0 22.1
Hughes Other Mil-spec 20 620.0
Martin Marietta Other Mil-spec 0 153.5
Martin Marietta Other Mil-spec 0 12.3
AFCIQ Armature Lower 11 57.0
AFCIQ Other Lower 0 1.5
Totals 36 1360.4
The proposed model form for relays was then hypothesized to be the
following equation.
[bXr : Xnb vNQ 7NE
where
Xr = relay nonoperating failure rate

Xnb = nonoperating base failure rate (failures/10 6 nonoperating hrs.)
= f(package type, contact voltage/current)
F~,• 5-220
1%Q = nonoperating quality factor

'NE = nonoperating environmental factor
The number of normally open contacts was not included in the model
because it could not be determined for the collected data. Therefore the
proposed model corresponds to an average, unknown quantity of normally
open contacts.
A nonoperating base failure rate for nonhermetic relays was determined

from the available data for armature relays. It was assumed that the data
typically were for low level circuit applications. Nonoperating base
failure rates for hermetic relays and nonhermetic relays with greater than
50 millivolts across the contacts were hypothesized. The data where the
construction and application were unknown provided a lower limit on
hypothesized nonoperating failure rate. The preliminary base failure
*rrates were determined to be:
Xnb,pre preliminary nonoperating base failure rate

= .023 nonhermetic, contact voltage < 50 millivolts when
operated
.005 nonhermetic, contact voltage > 50 millivolts when
operated
= .001 hermetic relays
The base failure rates are preliminary because they correspond to a

ground fixed environment. Conversely, the proposed model corresponds to a
$ground benign environment, Only one base failure rate was required for
hermetic relays because the effect of contact voltage is negligible
without the build-up of surface films.
A ratio between lower quality relays and mil-spec relays was computed
from the available data to determine a nonoperating quality factor for
commercial grade relays. Both the armature and unknown relay style data
were 'tilized to calculate a commercial grade nonoperating quality factor
equal to 4.2. An average established reliability (ER) quality factor was
5-221
obtained from RADC-73-248 (Reference 5). This document presented generic

relay nonoperating data for established reliability and Nil-spec devices.
The ratio of the observed failure rates was equal to 0.46. It was assumed
that this ratio was applicable for all relay types. The relay
nonoperating quality factors are presented in Table 5.8.3-4. The table
presents the proposed factors for established reliability, Mil-spec and
commercial grade relays.
TABLE 5.8.3-4: RELAY NONOPERATING QUALITY FACTORS
Quality Level WNQ
Established Reliability 0.46

Mil-spec 1.0
Lower 4.2
The next phase of the model development process was to assume

appropriate nonoperating environmental factors. Based on a comparison of
failure mechanism accelerating factors, it was assumed that the operating
relay environmental factors would be applicable for nonoperating failure
rate prediction purposes. These factors were given in Table 5.8.1-2 in
Section 5.8.1. Table 5.8.3-5, on the following page, presents the
summarized results of the failure mechanism comparison for relays.
References 41, 42 and 46 provided failure mode or mechanism information
for relays.
The final phase of the nonoperating failure rate model development for
relays was to normalize the base failure rates by dividing by the ground
fixed environmental factor of 2.3. The proposed model is presented in
Section 5.8.1 and in Appendix A in a format compatible with MIL-HDBK-2170.
A similar intuitive approach to model development was taken for

switches. Similar to the discussion for relays, it was hypothesized that
the nonoperating failure rate of the switch actuation mechanism is small
in comparison to the nonoperating failure rate of the contacts.
5-222
.0
'vD
-44 I
0.41 0) -4 -4
V4 CLO "-4 .C
04) 0
CL 4-3 -4r4.-4
.O7t 4-4 clt 4- ~ .4 -
CL(-4 - 4r4U)"
LiJ
>%
M r- 4J 03
LLJ CL0-0 0
> U
w .00 a
41 - >.- 41
S-
4C - C C
00 o0(U 0M 0 )% 0
S-4-) $ 4 4-) 41.4- 41 .4
IAU4-)4-J C -E C4-3 41
U;
-J
41
00(
S 41Cr r
.4- C.4- 0 01
C 0 (D0
u 4)1411S- *00
a4 411.
o MU'-0 - 4J 0
UO4J U'0 t41 u- Oil
0- 0( 4-. S..
"ao 410o41 '- 41
uO
04)4
41u -
O 000 .0
o 0 CL0~
C C co 0u
41L41C4 41. 4: 0
V ~ a,0 .- U
Therefore, the dominant variables were believed to be contact

voltage/current and the number of normally open contacts. Additionally,
it was assumed that application environment and quality were significant
variables.
All. collected nonoperating switch data were for Mil-spec quality

devices. Therefore, the nonoperating quality factors developed for relays
were also applied for switches. This assumption was justified because the
contacts are physically similar for both generic part families.
A series of nonoperating environmental factors for switches was then

determined by applying the methods presented in Section 4.5. This method
assumed that nonoperating environmental factors could be generated from
the corresponding operating factors. The conversion process was conducted
in two phases. First, typical temperature differences between operating f'
and nonoperating applications were compared. A temperature adjustment was
required for switches because the MIL-HDBK-217D switch models do not have
a unique temperature factor to* account for the temperature difference.
Temperature adjustment factors were multiplied with the operating
environmental factors. The temperature adjustment factors were based on 7
documented relationships (i.e. MIL-HDBK-217D) of failure rate vs.
temperature for other component *ypes. The second phase of the
nonoperating environmental factor determination process was to compare
typical operating and nonoperating failure mechanism accelerating factors. .
Table 5.8.3-6 presents the results of the failure mechanism comparison for
toggle switches. References 41, 42 'and 46 contained valuable switch
failure mode and mechanism information. It was concluded, based on the
comparison, that the temperature adjusted switch environmental factors
also represented the effect of environmental stress on nonoperating.
failure rate. The proposed switch nonoperating environmental factors were
presented in Table 5.8.1-3 in Section 5.8.1.
The next phase of the switch model development process was to

determine an average base failure rate from the available switch
nonoperating failure rate data. Table 5.8.3-7 presents the summarized
5-224
M~c
4
0C. - en U) r-4 U, (')
LM .- 4 L) O --40 U, -lt
r- 431
(A 4-3 .0 0f
go C"C)O ' UjU,) 4c U,
6-A 0L (A m
~~0 -4-4.-4 .- 4 -4 U C
(CA
'-41
* C
c C
0
4-
V)
-J ~4 4Ji,
(A 0 PA
t
4.3 S- c4
m4 0 0
&-- 4-)
C
a) 1 -432C4J 4
4)S-
4) U- - )- "- 4
co .0 S4-3 S-3 4J
4) u =0
fU 3: C
U4-)E
aU 0
-C
a 4-Cr
-0
o) (V to 4 *-
a003 30cu4-
CC 4- 4J 0 (A
t~.- 4- 4-) C . 4-4 .
R ro aC 40 4-3 M
42u C: cC - -an
S- *00 00CL 600
Euu 4J Ua (A.UU
v 0. .C W
U.. 0 LAn
5-2254
..... .
switch data. The average nonoperating base failure rate was determined by
dividing the sum of the failures by the sum of the part hours, and then
dividing by the ground fixed environmental factor. The average
nonoperating base failure rate was determined to be 0.030 failures per 106
nonoperating hours.
4m
TABLE 5.8.3-7: SWITCH NONOPERATING FAILURE RATE DATA
Part Hours
Equipment/Source Device Style Failures (XPO6)
Martin Marietta Toggle/Pushbutton 0 1.61

Martin Marietta Pressure 4 48.30
Martin Marietta Sensitive 0 1.64
Martin Marietta Stepping 2 5.00
Martin Marietta Inertial 9 137.10
Hawk/MICOM Toggle/Pushbutton 1 17.00
Maverick/MICOM Thermostatic 0 20.59
Maverick/MICOM Solenoid 9 82.36
Sparrow/MICOM Toggle/Pushbutton 0 12.76
Sparrow/MICOM Pressure 0 25.52
Sparrow/MICOM Inertial 0 12.76
TOW/MICOM Toggle/Pushbutton 0 5.26
MICOM Toggle/Pushbutton 0 1.88
MICOM Pressure 10 31.00
MICOM Thermostatic 0 3.88
MICOM Sensitive 0 1.64
Totals 35 408.30
The contact voltage/current was unknown for the available data.

However, this was considered to be the most important variable for
assessing switch nonoperating failure rate. Resistive films on the
contact surfaces are much more harmful for low voltage, low current
switching applications. Higher. voltage applications produce an arc on
contact closure which would break thru resistive surface films which form
during storage. It was assumed that the available data corresponded to
low level circuit applications to compute separate base failure rates.
Additionally, it was assumed that the relative difference between low
level circuit and other applications was the same as for relays. Two
5
5-226
~i
unique nonoperating base failure rates were determined by applying these

assumptions and are given by,
Xnb = nonoperating base failure rate (failures/ll 6 hours)

= 0.006, contact voltage > 50 millivolts when operated ,:,'.
= .030, contact voltage < 50 millivolts when operated
Determination of the nonoperating base failure rates concluded the

model development for contact devices.L Both the proposed model for
switches and relays are presented in Section 5.8.1 and in Appendix A.
Both models are based primarily on intuitive rather than empirical
relationships. The dominant model parameters were determined by carefully
evaluating device construction and the anticipated failure mechanisms.
When more data becomes available, the proposed models can be evaluated.
It is hypothesized that the number of normally open contacts is the most
significant variable not included in the proposed model.
5.8.4 Connectors
The model development for connectors is described in this section.

Additionally, a qualitative assessment of connector nonoperating failure
rate is provided. The proposed model is presented in Section 5.8.1 and in
Appendix A in a format compatible with MIL-HDBK-217D.
Printed wiring board, and rack and panel connectors are the most
simply constructed of all connector types. This does not mean they are
the most reliable since they are not designed to seal against moisture
assimilation or protect the contacts from harmful agents in the
atmosphere. How this effects performance in a nonoperating state requires
an understanding of the nature of the contact and certain chemical effects
on the contact.
Achieving good electrical contact in a connector is a function of

surface films (oxides and sulphides), surface roughness, contact area,
plastic deformation of the contacting materials and load applied.
5-227
X:K
Since even the best machined, polished and coated surfaces look rough
and uneven when -, ewed microscopically, the common concept of a flat,
smooth contact is grossly oversimplified. In reality, the connector
interface is basically an insulating barrier with a few widely scattered
points of microscopic contact. The performance of the connector is
dependent upoi the chemical, thermal and mechanical behavior at these
contact points.
Current flow between mating metals is constricted at the interface to

the small points on the surfaces which are in electrical contact. This
flow pattern causes differences of potential to exist along the contact
interface, and causes current bunching at points of lower resistance. As
a result, contact resistance and capacitance are introduced into the
circuit, and certain chemical effects evolve.
Most nonoperating failures of connectors are induced by the growth of

films at points of contact. These films can cause increased contact
resistance or an open circuit when power is applied. Contact resistance
causes higher temperatures during equipment turn on at the point of
contact and thus, increasing chemical activity.
Normally, the purity of metals in contact is not considered a

reliability problem. However, ions in impurities or contamination in the
surface pores will migrate to the points of highest potential, which are
frequently the localized hotspots. Ions interfacing with electrons and i•
other constituents at the points of high chemical activity generate films,

usually non-conducting. There is also a continuous supply of material for
the growth of insulating films from environments where there are corrosive
elements such as hydrogen sulfide, water vapor, oxygen, ozone,
hydrocarbons and various dusts.
The connector plugged to its mate during much of its life is

characterized by a typical catastrophic nonoperating failure rate based on
the factors described. It would appear that periodic demating and mating
whether operational or nonoperational would be beneficial. However, such
5-228
%XW
.%N. W..
action exposes, by repeated opening of the mating pin surfaces, the

contacts to a fresh supply of local corrosive contaminants. There is also
the problem of physical wear on the connecting interfaces. Surface
contact points become worn which cause unsymmetrical contacts. The result :';
is increased interface resistance and degradation of the connection.
It was hypothesized that the dominant variables influencing connector

nonoperating failure rate were the connector style, the number of pins and
the application environment. The device style was considered to be an
extremely important variable for connectors because the amount of
protection to the environment can vary considerable depending on the
specific connector style. In general, all connector styles can be grouped
into one of two categories. Circular, coaxial, and power connectors are
characterized by good protection from the environment which minimizes the
build-up of resistive surface films during storage. Conversely, rack and
panel, and printed wiring board (both one-piece and two-piece) connectors
offer less protection from the environment and would be anticipated to
have a higher nonoperating failure rate.
The collected nonoperating reliability data for connectors is

presented in Table 5.8.4-1. All available connector data were from MICOM.
Unfortunately, no part characterization information was available.
TABLE 5.8.4-1: CONNECTOR NONOPERATING FAILURE RATE DATA
6 )
No. Equipment/Source Failures Part Hours (X10
1 Martin Marietta 0 163.0

2 Hawk/MICOM 0 2024.7
3 Sparrow/MICOM 1 344.5
4 Lance/MICOM 0 3.4
5 MICOM 0 47.4 01
6 MICOM 0 79861.0
Totals 1 82444.0
Analysis of the data revealed large variability. Data entry number
six in Table 5.8.4-1 indicated such a low nonoperating failure rate that
it was considered suspect. There were insufficient data to statistically
5-229
4,
test for outliers. However, data entry number six was not further
considered in the analysis because of the belief that the extremely low
failure rate was not indicative of connectors in nonoperating
applications. Of the remaining data, only data entries 2 and 3 had
failures or sufficient part hours to estimate nonoperating failure rate.
As stated before, no part characterization information was available.
However, it was assumed that the two calculated nonoperating failure rates
provided a typical lower and upper bound on connector nonoperating failure
rate, and were set equal to preliminary base failure rates for ;..
circular, coaxial and power connectors and (2) rack and panel, and printeI
wiring board connectors, respectively. These values were preliminary
because a series of nonoperating environmental factors were late.-
determined and the ba.e failure rates were normalized to a ground benign
value equal to one. The preliminary base failure rates are,
,nb,pre = preliminary nonoperating base failure rate

= .00044, circular, coaxial, and power connectors
= .0029, rack and panel, and printed wiriiig board connectors
The methods presented in Section 4.5 were then applied to determine

approximate nonoperating environmental factors. A comparison of failure
mechanism accelerating factors was used as a rationale to assume that the
operating environmental factors would also predict the effect of
environmental stress on nonoperating failure rate. The proposed
nonoperating environmental factors were presented in Table 5.8.1-4 in ..
Section 5.8.1. These numerical values are average values of printed
wiring board (Table 5.1.12.2-4, MIL-HDBK-217D) and other connector styles
(Table 5.1.12.1-6, MIL-HDBK-217D).
The final stage of the model development process was to normalize the
nonoperating base failure rates by dividing by the ground fixed
nonoperating environmental factor of 2.3. The proposed nonoperating base,•
failure rates were presented in Section 5.8.1.
5-230
•'t be emphasized that the objective of this proposed study was to

dev -operating failure rate prediction models which can be applied
for any equipment type. Therefore, it was absolutely essential that the
proposed methodology include a nonoperating failure rate prediction model
, for connectors. The proposed connector model was based on practically no
quantitative information and must be considered approximate. However, the
proposed model is physically correct with proposed factors for device
* style and environment.
J ,
5-231
5.9 Interconnection Assemblies
5.9.1 Proposed Interconnection Assembly Nonoperating Failure Rate

Prediction Model
This section presents the proposed interconnection assembly

nonoperating failure rate prediction model. The interconnection assembly
-,
-. model predicts the nonoperating failure rate of the printed wiring board
and the solder connections. The model is also presented in Appendix A in
a format compatible with MIL-HDBK-217D. The proposed model is:
Xp Xnb x Npth X INE
where
Xp = interconnection assembly nonoperating failure rate (includes

both the interconnection board and the solder connections)
Xnb = nonoperating base failure rate (failures/10 6 nonoperating
"hours)
= .0000014, double-sided soldered printed wiring boards
= .0000028, multilayer soldered printed wiring boards
= .0000089, discrete wiring w/electroless deposited PTH
Npth = number of functional plated-through holes (includes nonsoldered
functional via holes)
iTNE = nonoperating environmental factor (see Table 5.9.1-1)
This model pertains to all interconnection assemblies using plated through

holes.
5-232
TABLE 5.9.1-1: INTERCONNECTION ASSEMBLY NONOPERATING ENVIRONMENTAL FACTORS
Environment NNE Environment •NE

GB 1 AIA 5.0
GF 2.3 AIF 9.0
GM 6.9 AUC 5.4
Mp 6.9 AUT 11
NSB 4.1 AUB 18
NS 5.3 AUA 14
NU 11 AUF 25
NH 13 SF (1)
NUU 14 MFF 7.8
ARW 17 MFA 11
AIC 2.3 USL 25
AIT 4.1 ML 26
AIB 7.2 CL 500
5.9.2 Interconnection Assembly Model Development
¶ An interconnection assembly failure rate prediction model

simultaneously predicts the failure rate of the printed wiring board (or
discrete wiring board) and the solder connections. The failure rate
sd modeling approach described in Section 4.1 was implemented for
interconnection assemblies. A theoretical model was developed and then
quantified with the available data. Selection of significant variables
was accomplished as part of the theoretical model development because a
relatively small nonoperating failure rate data base prevented the
selection of variables empirically. The proposed model was determined to
be a function of interconnection technology, the number of circuit planes,
the number of functional plated through holes, and the application
environment. The proposed model is presented in Appendix A in a format
compatible with MIL-HDBK-217D.
5-233
Ir T
Identification of application and construction variables which

properly characterize interconnection assemblies was the first step of the
model development process. Table 5.9.2-1 presents a list of the variables
identified for interconnection assemblies. These variables represent
possible nonoperating failure rate prediction model parameters which were
further studied by both qualitative and quantitative techniques.
A theoretical model for interconnection assemblies was developed based

yon information located during the literature search. The theoretical
model assumed a multiplicative model as a function of the number of
circuit planes, the number of plated through holes, solder application
technique, application environment and equipment power on-off cycling
frequency. Additionally, the theoretical nonoperating failure rate
prediction model was hypothesized to be directly proportional to the
number of plated through holes. This hypothesis was based on two
observations. First, the dominate areas of failure for interconnection
assemblies are (1) the plated through hole and substrate interface, (2)
the plated through hole barrel, (3) the plated through hole and solder
connection interface, (4) the solder connection, and (5) the solder
connection and component lead interface. In each instance, the number of
potential failure areas is approximately equal to the number of plated
through holes. Specific examples may exist where the number of plated
through holes and the number of solder connection are not correlated.
However, on the average, the number of solder connections can be
accurately approximated by the number of plated through holes. The second
observation was that each potential failure area is physically isolated
from each other. Therefore, it follows that the assembly failure rate
would be' equal to the sum of the individual plated through hole, solder
connection and interface failure rates. Thus, the assembly nonoperating
failure rate would be proportional to the number of plated through holes.
The sunmmarized printed wiring board and solder connection nonoperating

failure rate data is presented in Table 5.9.2-2. The data collectively
consists of 11 data records, 3 observed failures, 2976.6 x 106 assembly
,hours and 2,065,262.2 x 106 connection hours. Two data entries were
5-234
Y . .
TABLE 5.9.2-1: INTERCONNECTION ASSEMBLY CHARACTERIZATION
I. Interconnection Technology
A. Printed Wiring
B. Discrete Wiring
II. Number of Circuit Planes
III. Number of Functional Plated Through Holes
IV. Board Dimensions
V. Solder Application Technique
A. Wave Solder
B. Hand Solder
C. Reflow Solder
VI. Percentage of Solder Connections Requiring Rework
VII. Conformal Coating
VIII. Quality Level
"A. MIL-SPEC
B. Lower
IX. Temperature
A. Rated
B. Actual
XI. Number of Equipment Power On/off Cycles per 103 Nonoperating Hours
52.3
•..• 5-235
I
available for double sided printed wiring boards and six data entries were
available for multilayer printed wiring boards. Efforts to determine
specific circuit plane quantities for multilayer boards proved to be
futile. Additionally, it was assumed that the average number of plated
through holes per board was equal to 600 to determine the number of
connection hours. The 600 value was based on a sample from the Maverick
missile.
TABLE 5.9.2-2: INTERCONNECTION ASSEMBLY NONOPERATING FAILURE RATE DATA
Assembly Connection
Data Hours6 Hours6
Equipment/Source Records Description(I) Failures (X10 ) (X1O )
Hawk/MICOM(2) 1 Multilayer PWB 0 2,569.2 1541520.0

Maverick/MICOM(2) 1 Multilayer PWB 0 308.9 185310.0
Sprint/MICOM(2) 1 Multilayer PWB 1 67.8 40655.4
TOW/MICOM(2) 1 Multilayer PWB 1 26.3 15768.0
* Lance/MICOM(2) 1 Multilayer PWB 0 4.5 2683.8
AFCIQ 1 Multilayer PWB 1 (3) 405000.0
AFCIQ 2 Double Sided PWB 0 (3) 5940.0
Totals 8 3 >2,976.7 2196877.2
NOTES: (1) PWB = printed wiring board
(2) 600 plated through holes per PWB assumed for these data
records.
(3) Unknown.
The number of observed variables which could be evaluated empirically

was limited because of the nature of the available data. Environment,
temperature and equipment power on-off cycling could not be properly
evaluated because there were an insufficient range of these parameters
represented in the data. The number of circuit planes, solder application
technique and solder rework percentage could not be determined for the
majority of data sources, and therefore these variables also could not be
evaluated. As a result of the data restrictions, it was decided that a
5-236
I
relatively simple model was warranted for interconnection assembly
Selection of significant variables was then done based primarily on

intuitive nonoperating reliability relationships. The assumed form of the
interconnection nonoperating failure rate prediction model was determined
to be the following equation.
Xia Xnb x NpthX tNE
where
Xia = interconnection assembly nonoperating failure rate

Xnb = nonoperating base failure rate (failures/10 6 nonoperating hrs.)
= f(interconnection technology)
Npth = number of plated through holes
*NE = nonoperating environmental factor
The next phase of the model development process was to determine

appropriate nonoperating environmental factors by applying the methods
described in Section 4.5. This method assumes that a series of
nonoperating environmental factors can be generated from the MIL-HDBK-217D
environmental factors by (1) comparing anticipated differences in
application temperature, and (2) comparing operating and nonoperating
failure mechanism accelerating factors. Operating temperatures are
necessarily higher because of the heat generation associated with power
being applied. It was required that the interconnection assembly MIL-
HDBK-217D environmental factors be adjusted to compensate for the higher
temperature differences in the more stressful environments. The
environmental factor temperature conversion was computed by the following
equation.
FNE =AT 5T3OE
5-237
Vý?
V
where
ME = nonoperating environmental factor

AT = temperature adjustment factor
VOE = operating environmental factor
The AT factor values ranged from a value of 1.0 for all ground based
environments to a value of 0.72 for all airborne, uninhabited
environments. The temperature adjustment factors were determined based on
the average, documented temperature dependence for other part types. It
was determined that the resultant MNE factors sufficiently characterized
the effects of nonoperating environmental stress on interconnection
assembly nonoperating failure rate. These factors were presented in Table
5.9.1-1 in the previous section.
Determination of a nonoperating base failure rate for multilayer

printed wiring assemblies was the next step in the model development
process. A preliminary base failure rate was computed by taking the
geometric average of the connection nonoperating failure rate (i.e.
failures divided by connection hours) for those data entries with observed
failures or sufficient connection hours to estimate failure rate without
observed failures. The preliminary base failure for multilayer printed
wiring assemblies was determined to be 0.00000642 failures per 106 hours.
This numerical value is equivalent to the failure rate of one plated
through hole/solder joint connection in a ground fixed environment.
final base failure rate was computed by dividing the preliminary base
The
r
failure rate by the ground fixed nonoperating environmental factor of 2.3.
This was required because the nonoperating environmental factors were
normalized to a ground benign value equal to one. Thus, the multilayer 4-.
printed wiring assembly base failure rate was equal to 0.00000279.
There were insufficient connection hours to estimate a base failure

rate constant for doubled sided printed wiring assemblies. Therefore, an
extrapolated value was determined from the corresponding MIL-HDBK-217D,
Notice 1 operating interconnection assembly model (Reference 47).
5-238
MIR.-
Additionally, a base failure rate value was extrapolated for discrete

wiring with electroless deposited plated through holes. The final
nonoperating base failure rates were therefore given by,

= .00000140, double sided printed wiring assemblies
= .00000279, multilayer printed wiring assemblies
= .00000885, discrete wiring w/electroless deposited PTH
Extrapolation of the base failure rates concluded the model

development process. The proposed model is largely intuitive due to
deficiencies with the collected data. It is recommended that the proposed
model be evaluated when additional interconnection assembly nonoperating
failure rate data becomes available.
5-239
5.10 Connections
5.10.1 Proposed Connections Nonoperating Failure Rate Prediction

Model
The proposed model for connections is given by the following equation.
Xp = 1NEi 1 (NiXnbi)
where
Xp = connections nonoperating failure rate

NNE = nonoperating environmental factor (see Table 5.10.1-1)
Ni = number of connections of the ith type
Xnbi = nonoperating base failure rate of the ith type connection
= .000089, hand solder
= .000013, crimp
= .0000017, weld
.00000012, solderless wrap
.0000048, wrapped and soldered
= .0000041, clip termination
.0000024, reflow solder
This nonoperating failure rate prediction model applies to connections

used on all interconnection assemblies except those with plated through
holes. The nonoperating failure rate of the structure which supports the
connections and components should be considered to have a negligible
5-240
TABLE 5.10.1-1: CONNECTIONS NONOPERATING ENVIRONMENTAL FACTORS
Environment TNE Environment iTNE

GB 1 AIA 4.5
GF 2.1 AIF 6.8
GM 6.6 AUC 2.7
Mp 7.3 AUT .5.4
NSB 3.5 AUB 6.8
NS 4.4 AUA 6.3
NU 8.9 AUF 8.6
NH 11 SF (1)
NUU 12 MFF 6.6
ARW 14 MFA 9.0
AIC 2.3 USL 22
AIT 4.1 ML 23
AIB 5.0 CL 420
The nonoperating failure rate model development for connections could

not be performed by the desired, analytical approach because of a complete
lack of quantitative information. However, the objective of this study
effort was to determine a nonoperating failure rate prediction methodology
which could be applied for any equipment in any conceivable mission
profile (with the exception of satellite applications). Therefore, an
alternative nonoperating failure rate model development approach was
determined. The alternate approach was based solely on assumptions and ,
therefore must be considered approximate.
The alternate nonoperating failure rate development approach for

connections was conducted in three phases. The first phase was to
determine appropriate nonoperating environmental factors using the methods
described in Section 4.5. The second phase of the alternate model
development approach was to determine an applicable operating to
5-241
nonoperating failure rate ratio. Computation of approximate connection

nonoperating base failure rates using this ratio was the final phase of
the alternate model development process. It would have been preferable to
analyze observed nonoperating failure rate data. However, the alternate
approach represents the best possible method to determine connections
nonoperating failure rate prediction models given the data constraints.
Connections nonoperating environmental factors were determined using

the methods presented in Section 4.5. This method assumes that
appropriate nonoperating environmental factors can be generated from the
corresponding MIL-HDBK-217D environmental fdctors. Specifically, the
environmental factor conversion process required two quantitative
comparisons. First, the difference between typical operating and
nonoperating temperatures were investigated. The operating temperature is
generally higher because of the internal heat generation. For
connections, a temperature adjustment was required because there was no
separate temperature dependent factor to account for the temperature
difference. The second part of the nonoperating environmental factor
determination process was to compare operating and nonoperating failure
mechanism accelerating factors. The MIL-HDBK-217D factors were not
applicable for part types with operating failure mechanisms which are r
primarily accelerated by "operational stress" (e.g. current, voltage,
actuations). However, this was not the case for connections. Most
connection failure mechanisms are accelerated by environmental stress in
both operating and nonoperating states. Therefore, the temperature
adjusted environmental factors approximate the effect
stress on connections nonoperating failure rate.
of environmental
The connections
k
nonoperating environmental factors were presented in Table 5.10.1-1 in the
previous section.
The second phase of the alternate model development approach was to

assume an appropriate operating to nonoperating failure rate ratio. It
was determined that the most comparable part category was interconnection
assemblies. The interconnection assembly failure rate predicts the
failure rate of both the printed wiring board and the solder connections.
5-242
The average operating to nonoperating ratio for interconnection assemblies

was equal to 29:1.
The final phase of the connections model development approach was to

compute approximate base failure rate values. The proposed values were
computed by dividing the operating base failure rates (Table 5.1.14-1,
MIL-HDBK-217D, Notice 1) by the assumed ratio. In this manner,
nonoperating base failure rates were determined for hand solder, crimp,
weld, solderless wrap, wrapped and soldered, clip termination and reflow
solder. One average base failure rate value was proposed for crimp
connections. Without additional justification, the crimp connection
factors for tool type and quality could not be assumed to apply for
nonoperating failure rate prediction purposes.
The proposed model for connections was based entirely on assumptions.

In a project the magnitude of this study, it was inevitable that
sufficient data would not be collected for all parts. However, the
proposed model represents the best possible nonoperating failure rate
values and are based on sound assumptions. When additional information
becomes available, the proposed model should be evaluated, and updated if
required.
"5-243 -- . :..
5.11 Miscellaneous Parts
5.11.1 Proposed Miscellaneous Parts Nonoperating Failure Rates
Table 5.11.1-1 presents average nonoperating failure rates for

miscellaneous parts. the part types considered in this section include
all parts in Table 5.1.15-1, MIL-HDBK-217D.
TABLE 5.11.1-1: NONOPERATING FAILURE RATES FOR MISCELLANEOUS PARTS
Part Type Nonoperating Failure Rate

(failures/i06 hours)
Vi brators 3.3 ,• e*
Quartz Crystals .039
Fuses .0014
Lamps
Neon .029
Incandescent .11
Fiber Optic Cables .014 (per fiber kin)
(single fiber types only)
Single Fiber Optic Connectors .014
Meters 1.4
Circuit Breakers .29
Microwave Elements (coaxial and waveguide)
Fixed Elements
couplers, (directional
fixed stubs and cavities) Negligible :'
Variable Elements (tuned
stubs and tuned cavities)
Microwave Ferrite Devices
.014
.043
V
Dummy Loads .011 A
Terminations (thin or thick film .010

loads used in stripline and thin
film circuits)
NOTES: (1) The nonoperating failure rate of attenuators should be

calculated the same as for Style RD resistors.
5-244 ,
1%41
The miscellaneous part types considered in this study are the same
- miscellaneous parts which are included in MIL-HDBK-217D. Nonoperating
failure rate data were available for quartz crystals, fuses, neon lamps
"' and incandescent lamps. The operating to nonoperating failure rate ratio
for crystals and incandescent lamps were then applied for the other
miscellaneous parts. There were insufficient data to estimate
nonoperating failure rate for fuses and neon lamps.
4 There was no physical reason for proportional failure rate assumption

to be true. However, both incandescent lamps and crystals are part styles
which degrade in storage. Therefore, the assumption represents a worst
case scenario. Additionally, the geometric mean of the numerical values
for vibrators, microwave ferrite devices and dummy loads was used. The
p average ratio for operating to nonoperating failure rate was determined to

be 7:1. The average fuse nonoperating failure rate was divided by an
additional factor of ten because of the nature of fuse operating and
nonoperating failure mechanisms.
The available nonoperating failure rate data for miscellaneous parts

is presented in Table 5.11.2-1. All available data were from MICOM or
Martin Marietta. The merged nonoperating failure rate for crystals was
computed to be .039 failures per 106 nonoperating hours, and the merged
nonoperating failure rate was equal to 0.11 for incandescent lamps.
TABLE 5.11.2-1: MISCELLANEOUS PART NONOPERATING FAILURE RATE DATA

Part Hours
Equipment/Source Part Type Failures (X 106)
Hawk/MICOM Crystals 4 51.0
Maverick/MICOM Crystals 0 21.0
Sprint/MICOM Crystals 0 9.7
Martin Marietta Crystals 0 20.1
Martin Marietta Fuses 0 2.1
Martin Marietta Inc. Lamps 1 9.5
Sparrow/MICOM Neon Lamps 0 12.8
TOW/MICOM Fuses 0 2.6
Totals 5 128.8
5-245
The miscellaneous part nonoperating failure rates are approximate for

all devices except crystals or incandescent lamps. However, the part
types are all low population parts and therefore inaccurate nonoperating
failure rates would have little effect on the equipment level nonoperating
failure rate prediction.
5-;•46
6.0 APPLICATION OF NONOPERATING RELIABILITY MODELS
6.1 Comprehensive Reliability Models
With the inclusion of nonoperating failure rate models as a part of

this document, comprehensive reliability models are made feasible. A
,. comprehensive reliability model is one which considers all possible states
and the time in each state which an equipment must survive. While it is
true that most contractual reliability requirements are given in
operational equipment terms (i.e. MTBF or probability of survival for a
given mission time), the performance of an electronic equipment depends on
the effects of other events such as transportation, testing, equipment
power on-off cycling and nonoperation.
A simple comprehensive model which considers only one operational and

one nonoperational state takes the following form.
Rxp-•(oto tN
R = exp(-UAo(T+ tN) + N(to+ tN)) T)
where
R = reliability for a service life time of T

Xo= equipment operating failure rate
.N = equipment nonoperating failure rate
to = one mission operating time
tN = nonoperating time between missions
T = to + tN
* The (to/(to + tN)) term is simply the operating duty cycle, and the
* (tN/(to + tN)) term is the nonoperating duty cycle. In cases where there
is no definitive service life time interval (T), estimates can often be
* made of the respective duty cycles. Reliability would then be a function
' of one unknown instead of two.
6-1
This form also reduces to the following simplier equation.
R = exp(-(Xot 0 + XNtN))
However, the fractional time terms represent duty cycle, and is best
viewed as initially presented in many instances.
This is a simple model which could be applied to a ground fixed

electronic system since the two equipment states are exposed to the same
environment.
The model for an avionics system becomes more complex. A possible

scenario of events would consist of preflight power up testing, the
airborne mission comprised of operating and nonoperating periods,
postflight power up testing, and a nonoperating period. The mo, for
this scenario takes the following form.
R=exp(-(XoG to) + XoA (t) + XNA (t)+ XNG (t) T)
where
XoG = ground operating equipment failure rate

XoA = airborne operating equipment-failure rate
XNA = airborne nonoperating equipment failure rate
XNG = ground nonoperating equipment failure rate
toG = ground power on test time
toA = airborne power on time
tNA = airborne nonoperating time
tNG = ground nonoperating time
T = total time covered by the scenario
This model quantifies reliability in terms of the major events that

effect, in varying degree, the reliability of a system. It can be seen
that such a model can be developed for virtually'any system exposed to a
6-2
.9 ~*S'V~'
single or combination of environments. The model has the virtue of

focusing attention on the impact on reliability of both the operating and
nonoperating states, as well as the environments in which these states
occur.
6.2 Proposed Comprehensive Reliability Prediction Method
An eight part procedure was defined to use the proposed nonoperating

failure rate prediction models together with existing MIL-HDBK-217D
operating reliability assessment techniques. Table 6.2-1 presents the
eight specific tasks.
TABLE 6.2-1: COMPREHENSIVE RELIABILITY PREDICTION TASKS
Task No. Description

1 Define Mission Profile
2 Define Use Scenario
3 Determine Component Operating Failure Rates
4 Determine Equipment Operating Failure Rate
5 Determine Component Nonoperating Failure Rates
6 Determine Equipment Nonoperating Failure Rate
7 Compute Average Service Life Failure Rate
8 Compute Reliability
Definition of a mission profile (task 1) must be the initial phase of

any reliability assessment process. Specific time intervals and stresses
in every applicable operating environment must be defined as part of the
mission profile definition. For this study, a use scenario was considered
"to be equal to the mission profile plus the nonoperating time between
missions. In addition to the information provided from the mission
profile, spe'ific time intervals and stresses encountered durinq the
nonoperating phase of the use scenario must be defined as part of task 2.
Tasks 3 and 4 are well established and will not be further discussed.
Tasks 5 and 6 represent the corresponding reliability assessment tasks for
nonoperating periods. An equipment nonoperating failure rate is
determined based on the use scenario. The product of Tasks 3 through 6 is
6-3
•46ý
the determination of two unique parameters characterizing equipment

reliability; the inherent equipment operating failure rate and the
inherent nonoperating failure rate. At this point in the comprehensive
reliability prediction process, the two parameters are seemingly
independent and do not complement one another.
The next task is to compute an average service life failure rate. A

service life failure rate was defined as the number of failures per unit
.: time regardless of operational mode. The service life failure rate is
therefore dependent on operating failure rate, nonoperating failure rate,
operating duty cycle and nonoperating duty cycle, and is given by the
following equation for one operational state and one nonoperational state.
•L •Oto too
+ tN +•to tN
+ tN)
where
SSL = service life failure rate
to
to + tN= operating duty cycle
tN
to + tN = nonoperating duty cycle
The equation becomes more complex when there are more than two states.
However, the equation would take the same general form with more
Soperationalstates. The service life failure rate can
or nonoperational
be the most informative comprehensive reliability parameter for some
eouipment types. The service life failure rate is constant with time, and
is therefore useful for comparing or evaluating the total reliability for
equipments with an indeterminate mission duration. The service life
parameter is generally a good indication of the reliability for equipments
which are exposed to intermediate amounts of both operating and
nonoperating periods. For equipments with a use scenario dominated by
L. 6-4
N
dormancy, the service life failure rate approximates the dormant equipment
failure rate. Note in the equation that as to approaches zero, XSL
approaches XN. Regardless of the relative merit of this parameter, the
service life failure rate should be computed for every electronic
equipment exposed to nonoperating periods. The service life failure rate
can be computed with specific values for to and tN, or with appropriate
duty cycle values.
Computation of the equipment reliability is the final phase of the

comprehensive reliability prediction method. Given the specific time
intervals defined as part of the use scenario, the equipment reliability
is given as
R = exp(-(Z Xoitoi + E XNi tNi))
"where
R = reliability
Xoi = operating failure rate in the ith operating state
toi = time in the ith operating state
XNi = nonoperating failure rate in the ith nonoperating state
tNi = time in the ith nonoperating state
Reliability is the more important parameter for equipments exposed to

prolonged periods of nonoperation. Failures can not be detected when the
equipment is not operating, and therefore the probability of surviving a
specified nonoperating time interval without failure is essential
information. However, regardless of the relative merit of this
reliability parameter, the reliability should be computed for every
I electronic equipment for the time interval equal to the mission duration
plus the time between missions.
6-5
I' __ _____
These sections described two key comprehensive reliability parameters

which should be evaluated as part every reliability program for equipments
exposed to nonoperating periods. The equations to compute service life
failure rate and comprehensive reliability were also described.
kN
6-6 L
7.0 COMPARISON OF OPERATING AND NONOPERATING FAILURE RATES
Operating and nonoperating failure rates were compared using MIL-HDBK-

2170 and the proposed nonoperating failure rate prediction models. Table
7.0-1. presents the calculated
failure rates for a wide range of
microcircuits. Table 7.0-2 presents the ratio of operating to nonoperating
failure rate for three possible mission profiles; (1) ground fixed
operating to nonoperating, (2) airborne uninhabited fighter operating toI
nonoperating, and (3) airborne uninhabited fighter operating to ground

fixed nonoperating, and three different ambient temperatures. Table 7.0-3
presents operating to nonoperating failure rate ratios for discrete
semiconductors.
It was noted that in Table 7.0-2 the ratio for microcircuits was
greater for lower quality parts. This was the anticipated result. As
the device screening increases, the operating and nonoperating failure
rate difference decreases. This trend is because inherently weak devices
fail more quickly in the operating state, and was discussed in Section
4.6. Additionally, it was noted that the ratios for nonhermetic
unscreened microcircuits was lower than the ratios for hermetic unscreened
microcircuits. This would mean that package type is a relatively more
significant variable in the nonoperating state than in the operating
state.
Ratios for airborne avionic equipments are necessarily different from

ground based equipments. Airborne equipments typically operate in ground
fixed, ground mobile and airborne environments. In the nonoperating
state, the same equipment is primarily exposed to a ground fixed
environment. The predicted operating failure rate is higher (than the
ground fixed case) because of the numerically higher environmental factors
and increased temperatures. The predicted nonoperating failure rate would
also increase due to a higher level of equipment power on-off cycling.
The operating to nonoperating ratio is approximately 2-5 times higher for
r airborne equipments.
7-11
, 2t
TABLE 7.0-1: SAMPLE CALCULATIONS COMPARING NONOPERATING

AND MIL-HDBK-217D MODELS
Pw I vaT
O DESCUIPTI0N CMIPIBXIM TEDidaLo6YvW.LIY TYPEU-20C W-40W I-68C -2K AF-4K AUN-C
100102DC FSC MGTE 6 Ea 0 0 .5807 .5079 .5237 1.791 1.79" 1.114
1001020C FSC GTE 66 ECL 0 N .0040 .0045 .0057 .0112 .0126 .1161
10470 FSC 41 PM 40968 ECL D 0 1.618 3.238 6.441 2.796 4.409 7.633
10470 FSC 4( WI 40968 Ec. D N .0200 .0227 .0287 .0159 .0433 .W003
1103 SISNETICS IK WI 10248 mIGS C 0 .4364 .3502 1.79" AM943 1.316 2.247
1103 SISNETICS IK FM 10240 RINS C N .1017 .0123 .0182 .0243 .0337 .0107
140156 NOT SHIFT REGISTIER 664 0109 0-I 0 .6805 .7911 2.075 2.135. 2.318 3.42
140150 MOT SHIFT REGISTR 4"8 I 0-1 N .0A3N .1399 .5442 .2072 .4544 1.360
1821 RCA IK WI 10243 DIGS S 0 .0132 .0271 .0764 .0370 .0509 .091 -
1821 RCA 15IKI 15248 0W0S S N .0019 10141 .1122 .052 .0114 .0346
2110M INTEL 4KMI 40968 R40S 3-1 0 .16n .3=30 .86W,5 .3777 .5668 1.081
2114A INTEL 4K WI 40968 NOS 8'-1 N .051 .A91 .0213 .0142 .0254 .0595
2513 SIGNETICS RON 2560MID RS D 0 .5349 .8172 1.526 1.582 1.862 2.571
2513 SI]NETICS RON 25608 RIGS 0 N .A093 .0152 .0315 .0253 .0425 .M000
2580 SIS'ETICS SK501 31929 RIGS 0-1 0 2.576 7.5N 25.16 5.392 11.40 27.91
2580 SIGNETICS W RON 81929 RIS D-1 N .0537 M03 .1829 .1745 .2868 .5943
25104 MoI SHIFT RE61STER 2630 LTT". D 0 .6052 .8426 1.359 1.793 2.031 2.547
251.04 MI SHIFT REGISTER 2636 LTTL D N .0245 .03 .0384 .0671 .0739 .1073
251•,9299AI SHIFT RE61"RER 1046 LSt. s 0 .8131 .0161 .0235 .0439 .0468 .01542
251.929 MID SHIFT REGISTER 1346 LSITL S N .0033 .0039 .0059 .0092 .0111 .0164
2981A MID mICROPROCESSOR

5429 LSTTL B-1 0 .1515 .2329 .3329 .4811 .5324 .A42
2901A om MICROPRMESSOR 5426 LSTTL 8-1 N .0191 .0230 .0341 .0532 .0641 .0953
3601 INTEL 1KP"Om 10248 LTTiL. D-I 0 .6812 .8684 1.420 2.199 2.406 2.933
3601 INTEL I1 PRMI 10248 LTTL 0-I N .1147 .1077 .214 .3729 .3499 .69M2
7-2
".A.w

AND MIL-HDBK-217D MODELS (CONT'D)
4144444
4444444 44444444444444444*41414414441444* 4444444444HIH
• iH•HD]HHH 4444 H414 H HHH# 44e4
4144144444
mrT I VEN0OR OESCRIPTION COMPLEITY TECHNOLOY ORLITY TYPE OF-20C V-40C 6SF-60C AF-20C AF-40C AUF-61C
4004 INTEL 4 ICROPROCESSOR 759

T817 RIOS 0-1 0 .8208 1.597 4.674 2.455 3.231 6.310
R0IS D-1 N .2163 .3555 .7368 .7030 1.155 2.395 t w

4614 INTEL 4 BIT MICROPRICESSOR 7598
"4619 NSC SUFFER 66 OIOS 0-1 0 .5311 .5741 .8784 1.8 1.942 2.246
40094 NSC BUFFER 66 06OS 6-1 N .0263 .0557 .1135 .0659 .1810 .3687
4020A RCA BINARY COWNTER 1326 0109 0 0 .2556 .3290 .6162 .8723 .9458 1.233
4020A RCA BINAY COWNTER 1326 0m08 0 N .0153 .0336 .1005 .0427 .8938 .2964
45191 RCA LATCH

DECODER 1026 090S C 0 .2183 .2530 .3866 .7627 .7974 .9310
45151 RCA LATCH

DECODER 1026 010S C N .0125 .1273 .0918 .0348 .0763 .2282
54C221 NSC FLIP-FLOP 206 090S 9-1 0 .0497 .0551 .0766 .1755 .1909 .2125
54C221 NSC FLIP-FLOP 206 010S 6-2 N .0035 .0077 .0229 .0098 .0214 .U646
34LS197 TI B1INAY. COUWTER 346 LSTTL D 0 .2576 .2853 .3626 .8981 .9258 1.063
541.S197 TI BINARY COWNTER 346 LSTTL D N .0084 .0109 .0163 .0253 .0305 .1454
6116 1108 16K PA 163848 0OS 0 0 1.398 3.337 10.42 3.637 5.576 t2.66
6116 ZILO 16K WI 263849 010 0 N .0088 .0192 .0574 .0244 .0536 .1613
732 FSC DeCOOULATOR 39T BIPOLAR 0-1 0 7.238 17.14 65.88 21.34 31.26 08.01
732 FSC DEM0OULATOR 39T BIPOLAR ý-! N .1664 .2997 .5993 .5407 .9708 1.948
741072 TI JKFLIP-FLOP 8a HTTL D-! 0 .4608 .4801 .5279 1.634 1.653 1.711
741172 TI AJFLIP-FLOP 8a H11rL 0-1 N .0265 .0361 .0382 .0863 .0978 .1246E
74S206 NSC 256 BIT RIA 2568 STTL C-I 0 .3922 .6840 1.339 .9656 1.257 1.913
74S206 NSC 256 BIT RAN 2568 S1T. C-1 N .0190 .0223 .0303 .0530 .0623 .0847
"9"90A DM0 MICROPROCESSOR 11006 R40S 9-2 0 .5324 .7978 1.473 1.582 1.848 2.523
"9080A AND MICROPROCESSOR 11006 was 8-2 N .0349 .0625 .1466 .0975 .1744 .4694
9218 *0 16K ROI 163848 R4OS 9-1 0 .1911 .3853 '.973 .4660 .6•62 1.182
928 16K RO 163848 ,0S 8-1 N .0051 .0091 .0621 .8142 .0254 .M
7-3 --
".:"
," """ "•" " V'r"
x" "" • -" • -" - " -" - •" .' " ,' " ,• -" " -.. .. "... . .. . . . . .. .. . . .

MR yew9
4''41141414 14414•
DESCRIPTION
41414H41414HI'4
"==-'-''''HHH
COMU1.W1Y TECHNOLOGY
44 41144
H 4
QUALITY TYPE6F-20C GF-40C OF-66C AOF-20C AUF-46C MLF-68C1.
HII HlI 11111H4 4H4111*11HI4 111411 4'414 44
JlIH1141111,, 1111411
HHI tI
9425 FC IIKWIM 10248 TTL C-I 0' .5063 .8882 .686 1.144 1.546 2.344 K
9X25 FEC IK w 10248 T. C-I N .0192 .0218 .0276 .0536 .0608 .077.1
""4 FSC New 40968 111. 0-I 0 3.715 18.42 88.93 6.005 20.71 91.22
93461 FEC MEMORY 4968 IlL 0-I N .1127 .1502 .2M33 .362 .4880 .8915
M415 FSC IK WM12024 LTTL 8 0 .0602 .12, .271 .1026 .1719 .3214
?X415 FSC 2KWIN 10248 LITi. 8 N .0079 .8009 .0126 .0221 .0260 .5353
9402 FSC GENERTOR 1148 111 0- 0 .66M6 1.619 6.175 1.903 2.859 7.415 J
?461 FEC 686FATOR 1146 IlL 0"- N .0920 .1226 .224 .2990 .3M .72•0
"m TI MICROPROCESSOR 31006 IIL 0-1 0 22M1 6300 15871 2297 6307 15878
991A TI MICROPROCESSOR
31006 i11. D-I N .4449 .5929 1.083 1.446 1.927 3.520
C2416 INTEL 16K CCD

SERIAL 116. 168 CCO D 0 2.296 5.795 14.98 4.098 7.596 16.78 .-..
C2416 INTEL 16K CCD

SERIAL 1191. 163848 CCO 0 N .0091 .0162 .0381 .0253 .0453 .1063
M"425iK RCA GATE 36 010 B-I 0 .0309 .0336 .0443 .1096 .1122 .1229
C0482S RCA GTE 36 CHOS 8-I N .0014 .0031 .0092 .0039 .0086 .0258 .
OK74L87 NSC COMPAPTOR 338 LTTL D-I 0 .6179 .6673 .811692.174 2.223 2.362
3I74L8B4 NSC CLOMPARATOR 336 LIIL 0-1 N .0517 .0608 .0827 .1681 .1976 .2W8
L1112 NEC VOLTAGE

COMPAATOR 231 BIPOLAR S 0 .0100 .0170 .0426 .0304 .0375 .0630
1111 NEC VOLTAGE
C01PMTOR 23T BIPOLARS N .0038 .0068 .0137 .0106 .0191 .0383
LM741 NSC DUAL

OPAMP 237 BIPOLAR 0 0 .2815 .3869 .7845 .9469 1.052 1.450
LM741 NSC OtwAL

OPAMP 231 BIPOLAR 0 N .0180 .0322 .0647 .0501 .0900 .1806
MC1355P-2 MOT FREQU4NCY

AM 22T BIPOLAR 0-I 0 1.763 6.939 33.28 3.879 9.055 35.39
'C1355P-2 MOT FRE•UB4Y AMP 221 BIPOLAR 0-I N .1001 .1798 .3607 .3254 .5842 1.172
MC140BL-8 NOT D/A C'4•ERTER 601 BIPOLAR0 0 .6463 .9056 1.963 2.131 2.396 3.453 L
MC140BL-8 MOT D/A CONERTER 60I 8IPOLAR 0 N .0420 .0735 .1514 .1173 .2107 .4227 .
7-4 4..
njo-

MET I vow08 DESCRIPIONI

CMIPEIfY TECHNOLOGYV
OWITY TYPE OF-20C SF-40C SF-60C AUF-20C AOF-40C AUW-60C
203636 INTEL. 16KPROM 163841 Tn. a 0 1.014 1.639 2.883 2.457 3.082 4.326
MD3636 INTEL 16KPR3I 16384 Tn. 0 N .0200 .022 .0287 .0559 .0633 .0803
"II205D NMI 2K RUN 20438 ITTI. 0 0 .3318 .4268 64"4 1.059 1.153 1,372
MOD205 191 2KIII 20488 ITTI. D N .01 " .0233 .0464 .0552 .0649 .0882
WA12784 291 16KKIM 163848 LTTL D-I 0 2.568 2.5U6 5.019 4.237 5.255 7.6N
88ý
MUM278 191 16KROW 163848 LTTL. 0-1 N .1147 .1349 .1833 .3729 .4383 .5958
TLC271CP TI op 99 4T LIN M10 0-1 0 1.162 7.209 31.79 2.594 8.187 32.77
TLC271CP TI w i 4T LIN 010 0-1 N .0221 .0396 .0795 .0717 .1288 .2584
7-5
>..,
F -
TABLE 7,0-2: RATIOS OF MIL-HDBK-217D

TO NONOPERATING FAILURE RATES
M I v" DESCRIPTION CWLtEXITY TECHNOLOGY

GAMTY PES PIN GF/GFN AtFO/AtWN AIF/FN".
100102CC FSC WTE 66 ECL 0 HOIP 24 125.17 142.73 449.53
10470 FSC 41KWN 40968 ECL 0 HDIP 18 80.88 69.64 220.43
1203 SIMErICS IK IN 10248 MlOS C HOIP 10 50.16 39.05 151.27
14215 mOT SHIFT REGISTER 668 OIOS D-I NoIP 16 9.53 5.09 36.32 •,..
1821 RCA IK ?A 10243 O10s S HOIP 16 6.95 4.46 26.77
2114A INTEL 4K WI 40968 IMOS 8-I HOIP 18 32.12 13.33 111.15
2513 SIVETICS ROM 25608 PHOS 0 HOWP19 57.73 43.82 200.25
2580
251.04
25LS2"9
2901A
3601
SI
AVi
iD
Mon
INTEL
R0M
8K(ICS
SHIFT RESISTER
SHIFT RESISTER
ICRIPWROCESSOR
IK PROM
5426
8199
2636
1046
10248
P'S
LTTL
LSTTL
LSTTL
LTTL
0-1
8-1
0-I
NOIP 24
HOIP 21
HOIP 20
HOIP 4d
NOIP 16
47.97
24.70
3.98
7.94
5.93
36.27
."5.73
4.21
8.30
6.87
193.72
82.87
14.17
27.93
20.98
r
4004 INTEL 4 8IT MICROPROCESSOR 7596 IMGS 0-I NOIP 16 3.79 2.79 14.93
409N NSC BUFFER 66 OIOS

M-I NOIP 15 2.61 10.72 95.66
4120A RCA BItN1 COiNTER 1329 0105 D HFPK16 16.70 10.08 61.81
45158 IRA DECODER

LATCH 1026 dIOS C HFPK 24 17.46 10.45 63.79 ýo-
54C221 NSC FLIP-FLOP 206 0M05 9-1 HOOP16 14.19 8.45 5.16
54LS197 TI BI11RY COV4TER 346 LSTTL 0 H0IP 14 30.66 30.34 110.22
6116 ZILOS 16KRM 163848 010S 0 HDIP 24 15S.81 104.02 633.61 L ,
732 FSC MODULATOR 39T BIPOLAR 0-1 N0IP 14 43.49 32.20 187.9
74H72 TI JK FLIP-FLOP 86 HTTL 0-I NOIP 13 17.38 i6.90 62.38
74S206 NSC 25P 1I7T

WI 2568 STTL C-I HOIP 16 20.64 20.la 66.17
9080A Am MIICROPROCESSOR11006 NMOS 8-2 HoIP 40 15.25 10.59 52.94
9218 WID 1U ROM 163848 NOS B-1 HOIP 24 37.47 25.99 129,44
7-6
''¼'-1 '. '

TABLE 7.0-2: RATIOS OF MIL-HDBK-217D

TO NONOPERATING FAILURE RATES (CONT'D)
•
PART YO CipLEXITY TEC4OL06Y ML.ITY PK6
DESQCRIPTION PIN PF0/GFN AOAIF/M AU 6/iFN
"93425 FSC IK RWI 1024B TTL C-I MIP 16 26.37 25.42 0.51
93481 FSC II6IORY '46963 I 0-1 NDIP 16 32.9' 42.42 133.71
91415 FSC 1KRIA 10248 LTTn 6 HFPIC16 7.62 6.41 21.75
9401 FSC GBEMTNRT 1146 I1. 0-1 NIP 12 7.21 7.22 31.07
990
MA TI MICROPR•OESSOR 31006 IlL 0-I NSIP 59 5146.1 3273.5 14137.6
C2416 INTEL 3611 SERIAL Mm.

CCO 16334 CODSIP 18 25.34 167.68 334.72
C04025
D RCA MTE 36 010S 3-I IUPIC14 22.07 13.67 U.17
0l74LBS2 NSC CD WTODR 336 LTTL 0-1 NSIP 16 11.95 11.24 42.99
S"3LMIII NBC CU'PM

VOLTAGE 23'T BIPLAR S HCAN4
8 2.63 1.96 9.86
L.34741 NBC DUL PFif 23T BIPOLAR 0 NCW 7 35.69 11.60 58.44
14CI335.-2 MOT FREOEICY AV 221 BIPOLAR 0-I SIP 14 17.61 35.53 96.46
" MCIOL-B NOT 0/A CM E 61T IIPOLA D SDIP 16 15.39 11.37 57.65
"MD36 INTEL 11 PROM 136 TTL D SIP 24 6.73 43.69 154.12
1911
MM5265D 2K104 2840 LTTL D SDIP 16 36.07 17.77 53.62
M91
MM19627M 16K RON 36334 L~lL 0-3 SDIP 24 13.67 11.93 45.31
TLC27ICP TI o iMP 4T LIOIOS 0"- NSIP 3 73.12 63.56 373.5
NOTE: The ground fixed failure rates were computed with an ambient
temperature of 200 C. The airborne uninhabited fighter failure
rates were computed with an ambient temperature of 400 C.
7-7
8.0 CONCLUSIONS
Nonoperating failure rate prediction models were developed at the

component level for the entire range of device types included in MIL-HDBK-
2170. The models were primarily based on empirical data analysis and
represent accurate, easy to use and essential nonoperating reliability
assessment tools. Additionally, a comprehensive reliability prediction
method was presented to allow for use of the proposed nonoperating failure
rate prediction models together with the documented MIL-HDBK-217D methods.
The proposed methods will greatly improve upon current failure rate
prediction capabilities.
Previously, no single source of nonoperating failure rates were

available. A large number of intuitive and, for the most part, invalid
methods were generally used to estimate nonoperating reliability. The
inclusion of the proposed methods as part of a reliability program fills
a large void and will allow for more consistent evaluations of
nonoperating reliability predictions, reliability trade-offs and life
cycle cost analyses.
-ufficient nonoperating failure rate data were available for most

major part classes. However, for the following part classes, the proposed
models were largely intuitive and complemented with whatever data were
available.
o Magnetic Bubble Memories

o Opto-electronic Semiconductors
o Relays
o Switches
o Connectors
o Motors
"None of the nonoperating failure rate prediction models developed

during this study are as sophisticated as was originally intended. This
8-1
was entirely due to a lack of sufficient data, or a lack of detail in the

data. Several of the reasons for these data deficiences are:
o The inherent nonoperating failure rate for some parts is extremely

low. Thus, observed failure data is very limited.
o Many part types in MIL-HDBK-217 are low population part types.

o Data contributors are generally reluctant to incur any expenditure
to further refine data and information they provide without
charge.
o It is often difficult to segregate nonoperating failures from
operating failures or total maintenance actions.
o The wide scope of the study (i.e. all MIL-HDBK-217 parts) made it
impossible to concentrate efforts on one part type.
o Potential data contributors are hesitant to allow visitors access

to their proprietary data bases.
Consequently, many of the factors initially considered could not be

properly evaluated as part of the data analysis phase. Nevertheless,
sufficient data were collected and models developed for all major part
classes.
Additionally, it was concluded that the present MIL-HDBK-217D failure

rate prediction models may be influenced by nonoperating failures. The
extent to which the MIL-HDBK-217D models are affected by nonoperating
failures is dependent on,
o whether nonoperating failures were segregated from the total

number of failures during model development
o the ratio of operating to nonoperating failure rate
o the ratio of operating to nonoperating time
It was impossible to accurately estimate the difference between the MIL-

HDBK-217D models and the inherent mean operating failure rate. However,
it was concluded that the difference was small based on the model
development information which was studied.
8-2
9.0 RECOMMENDATIONS
It is recommended that the proposed nonoperating failure rate

prediction models developed during this study be incorporated into MIL-
"HDBK-217. Additionally, it is recommended that nonoperating failure rate
prediction models be updated periodically to reflect changes in technology
or other factors which temporarily result in an inaccurate or missing
model.
Although the proposed nonoperating failure rate prediction procedures

greatly improve prediction capabilities, several recommendations are
necessary in light of all available information. Due to the nature of
Snonoperating failure rates, plentiful amounts of data were not always
"available for analysis and consequently, many of the proposed models were
"based on limited data resources. The nonoperating failure rate of many of
these devices can constitute a large percentage of the total nonoperating
equipment reliability. Therefore, the respective nonoperating failure
rate prediction models should be further investigated to enhance their
accuracy and sensitivity. Included among these part types are:
o Hybrid Microcircuits
o Bubble Memories
, o Relays
o Switches
0 Rotating Mechanisms
o Connectors
- o Opto-electronic Devices
So Lasers
o Tubes
It is also recommended that a study be initiated to investigate the

"feasibility of predicting nonoperating failure rate based on accelerated
life testing to simulate more stressful environments. This would allow
for expeditious analyses of emerging technologies and would possibly allow
"for the development of unique nonoperating environmental factors.
9-1
-,,w-:
Additionally, it is recommended that future studies be initiated to

address the issue of equipment power cycling effects for mechanical and
electromechanical part types. For this family of parts, periodic
equipment power cycling can be a useful tool to decrease the probability
of failure during extended nonoperating periods. Unfortunately, there
were insufficient data for this class of part types to properly
investigate this effect. A study should therefore be established which
specifically analyzes both qualitatively and quantitatively the effects of
equipment power on-off cycling on mechanical and electromechanical part
types, and to determine optimal testing/operational schedules.
The availability of empirical data is essential for any valid

reliability assessment. When inherent difficulties prevent the
availability of data, an exhaustive data collection effort must be
undertaken, thereby, consuming valuable funds and manpower. Therefore, it
is recommended that the government should investigate methods of
identifying when reliability data are generated, of analyzing the merits
of available data, of providing a method of purchasing the data (if
required) and of storing the data in a central repository which is
available to all government contractors.
9-2
S ," T
I Downloaded from http://www.everyspec.com
REFERENCES
1. Kern, G. A. and T. M. Drnas, Operational Influences on Reliability,

RADC-TR-76-366, Hughes Aircraft Co., December 1976.
2. Kern, G. A., I. Quart, S. Tung and K. Wong, Nonoperating Failure

Rates for Avionics Study, Hughes Aircraft Co., RADC-TR-80-136, April
1980.
3. Malik, D., Storage Reliability Analysis Summary Report, LC-78-2,

Raytheon, January 1978.
4. Kremp, B. F. and E. W. Kimball, Revision of Environmental Factors for

MIL-HDBK-217B, RADC-TR-80-299, Martin Marietta, September 1980.
5. Bauer, J. A., D. F. Cottrell, T. R. Gagnier and E. W. Kimball, et.

al., Dormancy and Power On-Off Cycling Effects on Electronic Equipment
and Part Reliability, RADC-TR-73-248, Martin Marietta, August 1973.
6. Neuner, G. E. and E. H. Barnett, An Investigation of the Ratio Between

Stand-by and Operating Part Failure Rates, GIDEP Report E151-1405, TRW
Systems Group, November 1971.
7. Graber, R., et. al., An Approach In Assessing Missile System Dormant

Reliability, BDM/A-81-016-TR, January 1981.
8. Association Francais Pour le Control Industrial et la Qualite, Donnes

de Faubilite en Stockage des Composants Electroniques, Second Edition,
July 1983.
9. Coit, D. W., K. A. Dey, and W. E. Turkowski, "Operation & Analysis of

A Reliability Database," Proceedings of the 8th Advances in
Reliability Technology, University of Bradford, April 1984.
10. Draper, N. R. and H. Smith, Applied Regression Analysis, Wiley, 1966.
R-1
*i
REFERENCES (CONT'D)
11. Bean, E. E. and C. E. Bloomquist, An Investigation of On/Off Effects

on Equipment Operating in Space, PRC-R-1293, Planning Research
Corporation, October 1969.
12. Bean, E. E. and C. E. Bloomquist, The Effects of Ground Storage, Space

Dormancy, Standby Operation, and On/Off Cycling on Satellite
Electronics, PRC-R-1435, Planning Research Corp., May 1970.
13. Bean, E. E. and C. E. Bloomquist, Reliability Data from In-flight

Spacecraft, 1958-1970, PRC-R-1453, Planning Research Corporation,
November 1971.
14. Bloomquist, C., and W. Grahm, Analysis of Spacecraft On-Orbit
Anomalies and Lifetimes, PRC-R-3579, Planning Research Corporation,

February 1983.
15. Boteilko, R. J., "Effects of On/Off Cycling on Equipment Reliability",

Proceedings of the Seventh National Symposium on Reliability and
Quality Control in Electronics, January 1961.
16. Stanbery, R. L., "Military Electronics Field Effectiveness,"

Washington, D.C. Presentations, March 1983.
17. Stanbery, R. L., "TOW Failure Rates in Test and Flight," Washington,
D.C. Presentations, March 1983.
18. MIL-HDBK-217D, Military Handbook, Reliability Prediction of Electronic

Equipment, January 1982.
19. Rickers, H. C., LSI/Microprocessor Reliability Prediction Model

Development, RADC-TR-79-97.
R-2
-Mý,
7T
R/
REFERENCES (CONT'D)
20. Hommey, G. A., and T. E. Paquette, Linear/Interface Data, RAC

Publication MDR-6, Autumn 1977.
21. Flint, S., J. Steinkirchner and E. Edwards, Avionic Environmental

Factors for MIL-HDBK-217, RADC-TR-81-374, lIT Research Institute,
January 1982.
22. Fiorentino, Eugene, The Use of Air Force Field Maintenance Data for
R&M Assessments of Ground Electronic Systems, RADC-TR-79-103, April
1979.
23. Reed, A. C., Failure Rates of Non-Homogeneous Parts Populations,

Report No. TOR-0172 (2133)-l, September 1971.
24. Livesay, B. R. and E. J. Scheibner, Reliability Factors for Electronic

Components in a Storage Environment, DD-14-23 Georgia Institute of
Technology, September 1977.
25. Frank, R., L. McTigue and R. Provence, "Storage Reliability of Chip

and Bond Wire Electronic Devices", Proceedings 1976 Reliability and
Maintainability Symposium, January 1976.
26. Hakim, E. and H. Shaver, "Panama Field Test Results of Plastic

Encapsulated Devices," Proceedings of the Symposium on Plastic
Encapsulated/Polymer Sealed Semiconductor Devices for Army Equipment,
U.S. Army ERADCOM, Miy 1978.
m 27. Merren, G. T., "Dormant Storage Reliability Assessments - Data Based,"

IEEE Transactions on Components, Hybrids, and Manufacturing
Technology, Vol. CHMT-4, No. 4, Sandia National Laboratories, December
1981.
CL R-3
REFERENCES (CONT'D)
28. Coit, D., et. al., VLSI Device Reliability Models, RADC-TR-84-182, lIT
Research Institute, 1984.
29. Manno, P., Failure Rate Prediction Methodology (Today and Tomorrow),
Rome Air Development Center.
30. Peck, D. S., "New Concerns About Integrated Circuit Reliability,"

International Reliability Physics Symposium, 1978.
31. Electronic Reliability Design Handbook, lIT Research Institute, MIL-

HDBK-338 (Proposed).
32. Cook, D., S. Rosenburg, and B. Euzert, "IHMOS Reliability,"

International Reliability Physics Symposium, 1978.
33. Edwards, J. R., et. al., "VMOS Reliability," International Reliability

Physics Symposium Proceedings, 1978.
34. Branner, J. B., et. al., IEEE Transactions on Reliability, R-24, 1975 XS
p. 238.
35. Reynolds, F. H., and J. W. Stevins, "Semiconductor Components

Reliability in an Equipment Operating in Electromechanical Telephone
Exchanges," International Reliability Physics Symposium Proceedings,
1978.
36. Bramhilla, D. F., F. Funtini and A. Mattona, "Reliability Problems

with VLSI," Microelectronics Reliability, Vol. 24, No. 2, 1984.
37. Dey, K. A., and W. E. Turkowski, Memory/Digital LSI Data, RAC

Publication MDR-18, Winter 81/82.
77
U
R-4
-5i'
REFERENCES (CONT'D)
38. Ballou, T. A., Digital SSI/MSI Data, RAC Publication MDR-19, Spring
1984.
39. European Space Research and Technology Center, Specification QRA-4,

Issue 3, February 1976.
40. Information for Reliability Prediction, G.E. Technical Memorandum,

Report No. ASD-R-05-64-1, General Electric Co., May 1964.
41. Transistor/Diode Data, RAC Publication DRS-3, Winter 1979/80.
42. Tyler, D. and J. Wilbur, FMECA R&M Standard, Naval Avionics Center
Standard No. NAC-83-RM-914-OOC-1, May 1984.
43. Coit, D., and J. Steinkirchner, Reliability Modeling of Critical

Electronic Devices, RADC-TR-83-108, IIT Research Institute, May 1983.
44. Arno, R., Nonelectronic Parts Reliability Data, RAC Publication NPRD-
2, Summer 1981.
45. Coit, D. W., Printed Wiring Assembly and Interconnection Reliability,

RADC-TR-81-318, IIT Research Institute, November 1981.
R-5
? 11
APPENDIX A
Proposed Revision Pages
For
MIL-HDBK-217
A-i K
~7
MIL-HDBK-217
5.2 Nonoperating Reliability Prediction
5.2.1 Discussion of Nonoperating Failure Rate Prediction
5.2.1.1 Applicability
The nonoperating failure rate prediction models are applicable when

most of the design is completed and a detailed parts list is available.
This section contains nonoperating failure rate models for a wide variety
of parts used in electronic equipment. The models are to be used when a
subsystem, or assembly is experiencing none of the electrical or
mechanical stresses inherent in the designed activation of that subsystem
or system. It may, however, be experiencing stress caused by the
environment, transportation and handling, captive carry G-forces, etc.
5.2.1.2 General Model Factors
The following factors are common for the majority of the nonoperating
failure rate prediction models.
S'INO Nonoperating Quality Factor, accounts for effects of

different quality levels.
0 wNE - Nonoperating Environmental Factor, accounts for the
influence of all environmental stresses. For microelectronic
devices and discrete ser,;iconductors, the factor accounts for all
environmental stress except temperature.
o RNT - Nonoperating Temperature Factor, accounts for the effects of
ronoperating ambient temperature.
o lc c - Equipment Power On-Off Cycling Factor, accounts for the

effects of transients due to equipment power cycling.
A-2
S, •,:K
MIL-HDBK-217
5.2.1.3 Model Application
The nonoperating failure rate prediction models can be used separately

to predict nonoperating failure rate and reliability, or can be used to
complement the operating failure rate prediction models in Section 5.1.
The following three equations illustrate the methods for predicting
nonoperating reliability, service life failure rate and combined
operating/nonoperating reliability.
(1) Rn = exp(-E Xnitni)
(2) 'SL = Z DoiXoi + E DniXni
(3) Ro/n = exp(-(E. Xnitni + Z Xoitoi))
where
Rn = nonoperating reliability
Xni = nonoperating failure rate in the ith nonoperating environment
tni = nonoperating time in the ith nonoperating environment
XSL = service life failure rate, equal to the number of failures per
ur,it time regardless of operational mode
Doi = duty cycle in the ith operating environment, equal to the time
in the ith operating environment divided by total operating
time plus total nonoperating time
XIi operating failure rate in the ith operating envrionment
A-3
d. ~ ~ v~
T'~ -. :kV.
MIL-HDBK-217
U DnDi = duty cycle in the ith nonoperating environment, equal to the

time in the ith nonoperating environment divided by total
. operating time plus total nonoperating time
Roin - Reliability for the mission duration plus nonoperating time

between missions
5.2.1.4 Cautions
The following cautions are offered to prevent the misuse of the

nonoperating failure rate models.
o Temperature in the models for discrete semiconductors and

microelectronic devices is the ambient nonoperating temperature,
"not operating case or junction temperatures.
o Nonoperating environment is the actual environment to which the
component is exposed. For example, an airborne radar between
missions is most likely exposed to a ground fixed environment.
o Equipment power on-off cycling is determined at the equipment

level. The parameter does not refer to actuations of switches or
relays, nor specific circuit applications within the operating
state.
A-4
V^I*L" f ^L-VA.
MIL-HDBK-217
MICROELECTRONIC DEVICES
MONOLITHIC
5.2.2 Microelectronic Devices
5.2.2.1 Monolithic (Bipolar and MOS) Digital SSI/MSI, Random Logic

LSI. and Microprocessor Devices
This section represents the nonoperating failure rate prediction model

for monolithic (bipolar and MOS) digital SSI/MSI, random logic LSI and
microprocessor devices. The prediction model is as follows:
Xp = Xnb 1NT NNQ 'NE Ncyc failures/10 6 nonoperating hours
where
"Xp = predicted monolithic microelectronic device nonoperating

failure rate
.)nb = nonoperating base failure rate, based on gate count (Ng)
=,.00029(Ng)"477 for Ng < 3100 gates (See Table 5.2.2.4-1)
= .014 for Ng > 3100 gates (See Table 5.2.2.4-1)
TNT = nonoperating temperature factor, based on technology (See

Table 5.2.2.4-3)
,"NQ = nonoperating quality factor (See Table 5.2.2.4-5)
wNE = nonoperating environmental factor (See Table 5.2.2.4-8)
, equipment power on-off cycling factor (See Table 5.2.2.4-6)

=cyc
-. * A-5
ýW-
No*
MIL-HDBK-217
i MONOLITHIC
5.2.2.2 Monolithic (Bipolar and MOS) Linear/Interface Devices

for monolithic (bipolar and MOS) linear/interface devices. The prediction
model is as follows:
"1p=nb TNT 'NQ wNE 'cyc failures/10 6 nonoperating hours
•. where
XP = linear/interface device predicted nonoperating failure rate
Xnb - nonoperating base failure rate, based on transistor count (Nt)
Xnb - O.O0021(Nt)"8 8 7 where Nt = number of transistors (See

Table 5.2.2.4-2)
TNT - nonoperating temperature factor, based on technology (See

Table 5.2.2.4-3)
'NQ nonoperating quality factor (See Table 5.2.2.4-5)
'NE - nonoperating environmental factor (See Table 5.2.2.4-8)
wcyc = equipment power on-off cycling factor (See Table 5.2.2.4-7)
A-6
MIL-HDBK-217
MONOLITHIC
5.2.2.3 Monolithic RAM, CCD. ROM, and PROM Memory Devices

for monolith c (bi lar and MOS) memory devices (RAMs, CCDs, ROMs and
PROMs). The ppred tion model is as follows:
Ap = nb wNT wNQ itNE cyc failures/1O6 nonoperating hours
where
xp - memory device predicted nonoperating failure rate
)Xnb = nonoperating base failure rate, based on technology
z 0.0034 for Bipolar Memory Devices
= 0.0017 for MOS Memory Devices
irNT = nonoperating temperature factor, based on technology (See

Table 5.2.2.4-3)
INQ = nonoperating quality factor (See Table 5.2.2.4-5)
l-NE = nonoperating envirornental factor (See Table 5.2.2.4-8)
-cyc - equipment power on-off cycling factor (See Table 5.2.2.4-6)
A-7
*'' - . . Ao
- ,..~k-.'
MIL-HDBK-217
MONOLITHIC
5.2.2.4 Monolithic Microcircuit Nonoperating Failure Rate Tables
TABLE 5.2.2.4-1: DIGITAL SSI/MSI, RANDOM LOGIC LSI AND MICROPROCESSOR

NONOPERATING BASE FAILURE RATE (Xnb)
No. Xgb No. ( ,n

Gates (f/10 hrs) Gates 1f/t0 hrs)
1 .00029 80 .00234
2 .00040 90 .00248
4 .00056 100 .00261
6 .00068 150 .00317
8 .00078 200 .00363
"10 .00087 250 .00404
12 .00095 300 .00440
14 .00102 350 .00474
16 .00109 400 .00505
18 .00115 450 .00534
20 .00121 500 .00562
22 .00127 550 .00588
24 .00132 600 .00613
26 .00137 650 .00637
28 .00142 700 .00660
30 .00147 750 .00682
32 .00151 800 .00704
34 .00157 850 .00724
36 .00160 900 .00744
38 .00164 950 .00763
40 .00169 1000 .00786
42 .00172 1200 .00855
44 .00176 1400 .00916
"46 .00180 1600 .00977
"48 .00184 1800 .0104
50 .00188 2000 .0109
55 .00196 2400 .0119
60 .00205 2800 .0128
65 .00212 3100 .0134
70 .00220 >3100 .014
"nb =, .00029(Ng)' 4 77
where
"Ng = number of gates
A-8
MIL-HDBK-217
MONOLITHIC
TABLE 5.2.2.4-2: LINEAR/INTERFACE MICROCIRCUIT NONOPERATING

BASE FAILURE RATE (Xnb)
No. Xnb No. I Xnb

Transistors (f/10 6 hrs.) Transistors (f/10 6 hrs.)
4 .00072 148 .0177

8 .00133 156 .0186
12 .00190 164 .0193
16 .00246 172 .0202
"20 .00299 180 .0210
24 .00352 188 .0219
28 .00403 196 .0227
32 .00454 204 .0235
36 .00504 220 .0251
40 .00746 236 .0268
"44 .00603 252 .0284
48 .00651 268 .0299
52 .00699 284 .0315
56 .00746 300 .0331
60 .00794 350 .0379
64 .00840 400 .0427
68 .00884 450 .0474
72 .00936 500 .0521
76 .00978 550 .0566
80 .0102 600 .0612
84 .0107 650 .0656
88 .0111 700 .0701
92 .0116 750 .0746
96 .0121 800 .0789
100 .0125 850 .0833
108 .0134 900 .0875
116 .0143 950 .0919
124 .0151 1000 .0963
132 .0160
140 .0168
88 7
Xnb = .00021(Nt).
where
A-9
II
MIL-HDBK-217
MONOLITHIC
TABLE 5.2.2.4-3: MONOLITHIC DEVICE NONOPERATING TEMPERATURE FACTOR
(OtNT)
T TTLHTTL, LTTL, LSTTL IIL MNOS PMOS NMOS, CMOS, Linear

(OC) DTL,ECL STTL CCD CMOS/SOS Devices
0 0.93 0.92 0.91 0.88 0.66 0.73 0.70 0.63 0.62
5 0.94 0.93 0.91 0.89 0.69 0.76 0.73 0.66 0.66
10 0.95 0.94 0.93 0.90 0.73 0.79 0.76 0.70 0.72
20 0.98 0.97 0.97 0.95 0.88 0.91 0.89 0.86 0.88
25 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
30 1.03 1.03 1.04 1.06 1.17 1.12 1.14 1.20 1.15
35 1.06 1.08 1.10 1.15 1.41 1.28 1.34 1.49 1.34
40 1.11 1.14 1.17 1.27 1.74 1.49 1.59 1.88 1.58
45 1.16 1.21 1.27 1.43 2.18 1.75 1.94 2.44 1.86
50 1.23 1.30 1.39 1.65 2.77 2.10 2.39 3.20 2.22
55 1.31 1.41 1.54 1.94 3.57 2.54 2.97 4.23 2.65
60 1.41 1.55 1.74 2.32 4.62 3.09 3.73 5.63 4.62
65 1.54 1.72 1.99 2.82 5.98 3.79 4.69 7.49 3.80
70 1.68 1.93 2.30 3.46 7.73 4.67 5.95 9.97 4.55
75 1.86 2.18 2.68 4.29 10.0 5.i3 7.49 13.2 5.44
80 2.06 2.50 3.14 5.34 12.9 7.05 9.42 1T.3 6.50
85 2.30 2.87 3.71 6.69 16.6 8.66 11.8 22.7 7.70
90 2.58 3.31 4.41 8.38 21.2 10.6 14.9 29.7 9.15
95 2.92 3.83 5.25 10.5 27.0 13.0 18.6 38.5 10.9
00 3.30 4.45 6.27 13.1 34.2 15.9 23.0 49.6 12.8
05 3.74 5.18 7.46 16.3 43.0 19.3 28.5 63.4 15.1
10 4.28 6.03 8.90 20.3 53.9 23.3 35.1 81.0 17.7
15 4.85 7.02 10.6 25.1 67.1 28.1 43.1 102 20.7
20 5.52 8.18 12.6 31.1 83.5 33.8 52.6 129 24.1
25 6.29 9.51 14.9 38.2 103 40.3 63.8 161 27.9
30 7.16 11.0 17.6 46.7 126 48.0 77.2 201 32.3
35 8.14 12.8 20.8 56.8 154 57.0 93.0 249 37.2
40 9.26 14.8 24.6 69.0 187 67.7 112 307 42.8
45 10.5 17.1 28.8 83.4 227 79.6 133 376 49.0
50 11.9 19.7 33.7 100 273 93.4 158 457 56.0
55 13.5 22.6 39.4 120 328 109 188 558 63.5
60 15.2 26.0 45.7 143 390
wNT = K1 + K2 exp(-An(T r+273 -

127
1
))
T = ambient nonoperating temperature (oC)
221
1
667 72.5
. -
K1, K2 , An = temperature coefficients, based on technology (See

Table 5.2.2.4-4)
A- 10
//
MIL-HDBK-217
MONOLITHIC
TABLE 5.2.2.4-4: NONOPERATING TEMPERATURE COEFFICIENT (An)

STechnology An K1 K2
TTL, HTTL, DTL, ECL 4813 0.91 0.09

LTTL, STTL 5261 0.90 0.10
LSTTL 5711 0.89 0.11

IIL 6607 0.86 0.14
MNOS 6607 0.61 0.39
PMOS 5711 0.68 0.32
NMOS, CCD 6159 0.b5 0.35
CMOS, CMOS/SOS 7057 0.58 0.42
Linear 4748 0.50 0.50
"f"T = K1 + K2 expk 1 1
-n(T + 273 2)
TABLE 5.2.2.4-5: MONOLITHIC DEVICE NONOPERATING QUALITY

FACTOR (•NQ)
Quality Level* wNQ

S 0.53
B 1.0
B-1 1.4
B-2 2.0
VC 2..3
C-I 2.4
, •D 2.5
" •-2:D-1 8.7
•ii *Quality level definitions are provided in Table 5.1.2.5-1 in Section
S~5.1.2.5-1
.e N
A-11
MIL-HOBK-217
"MICROELECTRONIC DEVICES
MONOLITHIC
TABLE 5.2.2.4-6: DIGITAL/MEMORY EQUIPMENT POWER ON-OFF

CYCLING FACTOR (wcyc)
Cycling Rate* (Nc) Mean-Time-Between

"(Power Cycles/10 3 hrs.) Power Cycles (Hours) Ircyc
<1 >1000 1
1 1000 1.02
2 500 1.04
3 333 1.06
4 250 1.08
5 200 1.10
10 100 1.20
20 50 1.40
50 20 2.00
Wcyc = 1 + .02(Nc)
Nc Number of equipment power on-off cycles per 1000 nonoperating hours
* An equipment power on-off cycle is defined as the state during which

an electronic equipment goes from zero electrical activation level to
the normal design activation level plus the state during which it
returns to zero.
A-12
S ..... I 4,1
MIL-HDBK-217
MONOLITHIC
TABLE 5.2.2.4-7: LINEAR/INTERFACE EQUIPMENT POWER ON-OFF

CYCLING FACTOR (lrcyc)
Cycling Rate* (Nc) Mean-Time-Between

(Power Cycles/10 3 hrs.) Power Cycles (Hours) ircyc
<1 >1000 1
1 1000 1.03
2 500 1.06
3 333 1.09
4 250 1.12
5 200 1.16
10 100 1.31
20 50 1.62
50 20 2.55
•fcyc = 1 + .031(Nc)
Nc = Number of equipment power on-off cycles per 1000 nonoperating hours
• An equipment power on-off cycle is defined as the state during which

the normal design activation level plus the state during which it
returns to zero.
- -
_ Z
-. ~~ . .4
MIL-HDBK-217
MONOLITHIC
TABLE 5.2.2.4-8: MONOLITHIC DEVICE NONOPERATING

"ENVIRONMENTAL FACTORS (1NE)
Hermetic Nonhermetic
Environment Devices Devices
GB 1 1
GF 2.4 4.0
GM 3.5 6.5
Mp 3.2 5.9
NSB 3.4 6.2
NS 3.4 6.2
NU 4.5 8.6
"NH 4.6 8.9
"NUU 4.9 9.5
"ARW 6.3 13
AIC 2.4 4.0
AIT 2.7 4.7
AIB 4.0 7.6
AIA 3.4 6.2
AIF 4.7 9.0
AUC 2.7 2.7
AUT 3.4 3.4
AUB 5.7 11
AUA 4.7 9.0
AUF 6.7 13
SF 1.0 1.0
MFF 3.3 6.0
MFA 4.3 8.2
USL 8.0 16
"ML 9.3 19
CL ISO 310 •:;
A-14
; • - :< - -
MIL-HDBK-217
MONOLITHIC
5.2.2.5 Example Nonoperating Failure Rate Calculations
Example One
Given: A hermetically sealed monolithic linear device, M38510/11402P is

being used in a greund fixed environment at an ambient
nonoperating temperature of 200C. The quality level is class S
and the equipment on-off cycling rate is 1 per 1,000 hours.
Step 1: The nonoperating failure rate model is shown in Section 5.2.2.2

to be Xp = Xnb ffNT wNQ TNE ircyc
Step 2: Examine Table 5.1.2.5-28 "Microelectronic Parameters" in Section
5.1.2.5 and determine the number of transistors for an
M385010/11402P. Thirty-two transistors is the correct number.
Step 3: Refer to Table 5.2.2.4-2 and find the correct nonoperating base
failure rate(),nb) of 0.00454 failures per million hours for the kW
32 transistor device.
Step 4: Refer to Table 5.2.2.4-3 and find the nonoperating temperature
factor rNT of 0.88 for a linear device at 200C ambient
nonoperating temperature.
Step 5: Next refer to Table 5.2.2.4-5 and find correct nonoperating
quality factor (INQ) of 0.53 for a quality level S device.
Step 6: Refer to Table 5.2.2.4-7 and find correct ircyc factor of 1.03 for
a cycling rate of 1 per 1000 hours.
Step 7: Refer to Table 5.2.2.4-8 and find correct nonoperating
environmental factor (nNE) of 2.4 for a hermetically sealed
linear device.
Step 8: Determine the predicted linear monolithic device failure rate as
follows:
Xp = Xnb wNT lTNQ TNE wcyc

p= .00454 x 0.88 x 0.53 x 2.4 x 1.03
XP= .00523 failures for million hours
A-15
/ -A
MIL-HDBK-217
MI7ROELECTRONIC DEVICES
MONOLITHIC
Example Two
Given: A hermetically sealed monolithic bipolar memory device TTL,

M38510/20202-F is being used in a ground mobile environment at an
ambient nonoperating temperature of 400C. The quality level is
class B and the equipment on-off cycling rate is 3 every 1000
hours.
Step 1: The nonoperating failure rate model is shown in section 5.2.2.3

"tobe Xp = Xnb WNT rNQ ItNE Ncyc
Step 2: Refer to nonoperating base failure rate constant Xnb of 0.0034
for a bipolar memory device from Section 5.2.2.3.
factor (TNT) of 1.11 for a TTL memory device at 400C ambient
quality factor (TNQ) of 1.0 for a quality level-B device.
Step 5: Refer to Table 5.2.2.4-6 and find correct IrcYc factor of 1.06 for
a cycling rate of 3 per 1000 hours (refer to digital/memory
device listing).
environmental factor WNE of 3.5 for ground mobile (GM),
hermetically sealed TTL memory device.
Step 7: Determine the predicted TTL monolithic memory device failure rate
as follows:
Xp = Xnb wNT NNQ INE wcyc
Xp= 0.0034 x 1.11 x 1.0 x 3.5 x 1.06

p= .0140 failures per million hours
A- 16 "- "':•
?.
MIL-HDBK-217
MONOLITHIC
Example Three
Given: A hermetically sealed monolithic TTL, M38510/00903-A, device is

being used in the ground mobile environment at an ambient
nonoperating temperature of 600C. The quality level is Class B
and the equipment on-off cycling rate is 1 per 500 hours.
Step 1: The nonoperating failure rate model is shown in Section 5.2.2.1 K
to be Xp = Xnb ANT rNQ icyc ANE
Step 2: Examine Table 5.1.2.5-28, "Microelectronic Parameters" in Section
5.1.2.5 and determine the number of gates for a TTL type,
M38510/00903-A device. Thirty-six gates is the correct number. •.,;
Step 3: Refer to Table 5.2.2.4-1 and find the correct nonoperating base
failure rate (Xnb) of 0.00160 failures per million hours for the
36 gate device.

factor (irNT) of 1.41 for a TTL device at 600C ambient
quality factor (nNQ) of 1.0 for a quality level B device.
Step 6: Refer to Table 5.2.2.4-6 and find correct ncyc factor of 1.04 for
a mean-t 4 me-between power cycles of 500 hours.
environmental factor (6NE) of 3.5 for a ground mobile (GM),
hermetically sealed TTL device.
Step 8: Determine the predicted TTL monolithic device nonoperating

failure rate as follows:
Xp =Xnb ANT "NQ 1rNE cyc

p= .00160 x 1.41 x 1.0 x 3.5 x 1.04
Xp 0.00821 failures per million hours
A-17
... '
MIL-HDBK-217
HYBRID
5.2.2.6 Hybrid Microcircuits
This section includes the nonoperating failure rate prediction model

for hybrids.
The general model for hybrid devices is as follows:
1'.W XP= Xnb INQ •NE failures/10 6 nonoperating hours
where
Xp predicted hybrid microelectronic device nonoperating failure

rate
.nb = nonoperating base failure rate, based on component count (See
Table 5.2.2.6-1)
fNQ = nonoperating quality factor (see Table 5.2.2.6-2)
T NE = nonoperating environmental factor (see Table 5.2.2.6-3)
5.2.2.6.1 Example Failure Rate Calculation
Given: A hermetically sealed hybrid procured to MIL-STD-883, Method

5004, 5005 and MIL-M-38510 contains 18 diodes, 15 transistors,
and 3 integrated circuits, and is installed in a ground fixed
environment.
Step 1: The nonoperating failure rate model is shown in Section 5.2.2.6
to be Xp = Xnb 'NQ TNE
Step 2: The next step is to determine which hybrid nonoperating base
failure rate equation to use. Refer to Table 5.2.2.6-1. Adding
the number of diodes (18) plus the number of transistors (15)
plus 1.8 times the number of integrated circuits (3) yields 38.4.
Therefore, since 38.4 is greater than 12.2, the second base
failure rate equation is selected to determine the hybrid
nonoperating base failure rate. Do not count capaciLors,
packaged resistors, substrate resistors, substrate and
interconnections as these parameters are all included in the base
failure rate constant.
A-18
• / Downloaded from http://www.everyspec.com
MIL-HDBK-217
HYBRID
Step 3: The correct nonoperating base failure rate model (Equation 2) is

shown in Section 5.2.2.6 to be
03 3 (ND + NT) + .059NIC)
Xnb = 0.013 (e.
= 0.046
Step 4: Refer to Table 5.2.2.6-2 and select the nonoperating quality

factor of 1.0 for a quality level B hybrids.
Step 5: Refer to Table 5.2.2.6-3 and select the ground fixed (GF)
nonoperating environmental factor of 2.4 for the hermetically
sealed hybrid.
Step 6: Determined the predicted hybrid nonoperating failure rate as
follows:
XP = Xnb 7TNQ WNE
= .046 x 1.0 x 2.4
= .110 failures/lO6 nonoperating hours
!m
-A-I9
MIL-HDBK-217
HYBRID
TABLE 5.2.2.6-1: HYBRID NONOPERATING BASE FAILURE RATE (Xnb)
Nonoperating Base Failure Rate Model*
Xnb = A exp(blND + b2NT + b3NIC)
where
Case Complexity A bl b2 b3
I ND + NT + 1.8 NIC < 12.2 .000817 .45 .45 .81
II ND + NT + 1.8 NIC > 12.2 .013 .033 .033 .059
* Hybrid base failure rate includes the contribution of an

average number of capacitors, packaged resistors, and
substrate resistors.
A-20
* ~*..-
~'**. N!
~*~-. .,.T . t7
MIL-HDBK-217
HYBRID
TABLE 5.2.2.6-2: HYBRID NONOPERATING QUALITY FACTORS (7TNQ)
QUALITY LEVEL DESCRIPTION 7Q
S The test procedures for this quality level 0.53

are currently being developed. Until such
time that they are included in MIL-STD-883
and MIL-M-38510, the procuring activity will
provide the necessary testing requirements.
B Procured to the Class B requirements of:
MIL-STD-883, Method 5008 and Appendix G of
MIL-M-38510
or 1.0
MIL-STD-883, Methods 5004 and 5005 and
MIL-M-38510
D Commercial Part, hermetically sealed, with 8.6
no screening beyond manufacturer's normal
quality assurance practices.
,-,- -2
MIL-HDBK-217
HYBRID
TABLE 5.2.2.6-3: HYBRID NONOPERATING ENVIRONMENT FACTORS (WNE)
Environment WNE
GB 1
GF 2.4
GM 3.5
Mp 3.2
NSB 3.4
NS 3.4
NU 4.5
NH 4.6
NUU 4.9
ARW 6.3
AIC 2.4
AIT 2.7
AIB 4.0
AIA 3.4
AIF 4.7
AUC 2.7
AUT 3.4
AUB 5.7
AUA 4.7
AUF 6.7
SF 1.3
MFF 3.3
MFA 4.3
USL 8.0
ML 9.3
CL 150
A-22
XM
U
MIL-HDBK-217
BUBBLE MEMORIES
5.2.2.7 Magnetic Bubble Memories
The magnetic bubble memory device in its present form is a hybrid

assembly of two major structural segments:
I
a. A basic memory and control structure consisting of thin-film
elements on a crystalline substrate, and
b. A magnetic structure to provide a controlled magnetic field
consisting of a magnet, magnetic coils and a housing.
These two major structural segments of the hybrid are interconnected

by a mechanical substrate and lead frame. The interconnect substrate in
present technology is normally a printed circuit board. The general form
of the operating failure rate model is:
Xp =nl+ n2
where
xp = nonoperating failure rate in failures/10 6 hrs.

Xnl = nonoperating failure rate of the control structure
Xn2 = nonoperating failure rate of the magnetic memory structure.
Nonoperating Failure Rate of the Control Structure (Xnl)
INE failures/10 6 nonoperating hours

SXnl
Xnbl INT1
where
Xnbl = control structure nonoperating base failure rate, based on gate

count (Ng)
A-23
14
MIL-HDBK-217
BUBBLE MEMORIES
= .0015(Ng)- 4 7 7 , where Ng is the number of transfer gates plus

the number of dissipative control gates plus the number of
major loops.
.fNT1 = control structure nonoperating temperature factor, based upon

NMOS technology for digital microcircuits (See Table 5.2.2.4-3
for NMOS)
1i
=exp(-6159( - T 8)), where T is temperature (OK)
wNE = nonoperating environmental factor (See Table 5.2.2.4-7 for

hnonhermetic devices)
Nonoperating Failure Rate of the Magnetic Memory Structure (En?)
• Xn2 = .0089 (NL) 'TNT "NE
S~where
NL = number of loops, equal to the number of major loops plus the

number of functional minor loops
'NT = magnetic memory structure nonoperating temperature factor, based

on technology (See Table 5.2.2.4-3)
,rNE = nonoperating environmental factor (See Table 5.2.2.4-8 for

nonhermetic devices)
A- 24
BUBBLE MEMORIES
5.2.2.7-1 Example Nonoperating Failure Rate Calculation-
Given: A 92K bit magnetic bubble memory with 10 connected pins, 1 major
loop, 3 dissipative control elements (generate, replicate and
detector bridge), 144 transfer gates, 640 bits per major loop,
and 157 memory minor loops with 144 functional is installed in a
ground benign nonoperating environment with a nonoperating
ambient temperature of 250C.
Step 1: From the example statement, N was equal to 148, found by adding
the number of major loops (1) plus the number of dissipative
control elements (3) plus the number of transfer gates (144).
Step 2: From Section 5.2.2.7, the control structure nonoperating base

failure rate can be found by, 1
Xnbl = .0015 (Ng)47
= .0015(148). 477
= .0163
Step 3: From Table 5.2.2.4-3 for NMOS technology, a nonoperating

temperature factor of 1 was found for 25 0 C.
Step 4: From Table 5.2.2.4-8, an appropriate nonoperating environmental

factor of 1.0 was selected.
Step 5: The control structure nonoperating failure rate is
Xn1 = Xnbl 7NT TNE

= .0163 x 1.0 x 1.0
= .0163
Step 6: From the example statement, the number of loops was found to be
145 equal to the number of major loops (1) plus the number of
functional minor loops (144).
A-25
.* .-. '.-'.
MIL-HDBK-217
BUBBLE MEMORIES
Step 7: From Table 5.2.2.4-3, the magnetic memory structure nonoperating

temperature factor was determined to be 1.0 for 250C.
Step 8: Since the nonoperating environmental factor was already
determined, the magnetic memory structure can be found by the
following expression
p= .0089(NL) 'rT 7NE
= .0089(145) x (1.0) x (1.0)

= 1.29
Step 9: From Section 5.2.2.7, the magnetic bubble memory nonoperating
failure rate is equal the control structure nonoperating failure
rate plus the magnetic memory structure
Xp = Xnl + Xn2
= .0163 + 1.29
= 1.31 failures/10 6 nonoperating hours
JA2
MIL-HDBK-217
DISCRETE SEMICONDUCTORS
5.2.3 Discrete Semiconductors
This section includes the nonoperating failure rate prediction models

for discrete semiconductors.
5.2.3.1 Transistor and Diode Semiconductor Devices
The general nonoperating failure rate prediction model for transistors

and diodes is as follows:
p Xnb RNT WNE 7NQ wcyc failures/10 6 nonoperating hours
where
Xp = predicted transistor or diode nonoperating failure rate

Xnb = nonoperating base failure rate (See Table 5.2.3-1) %..
wNT = nonoperating temperature factor, based on device style (see
Table 5.2.3-2 for transistors, Table 5.2.3-3 for diodes and
Table 5.2.3-4 for temperature factor parameter) ,2I
rNE = nonoperating environmental factor (See Table 5.2.3-5)
7NQ = nonoperating quality factor (See Table 5.2.3-6)
ncyc equipment power on-off cycling factor (See Table 5.2.3-7 for
transistors and Table 5.2.3-8 for diodes)
5.2.3.1.1 Example Nonoperating Failure Rate Calculation
Given: A Silicon, NPN general purpose JAN grade transistor is in a

fixed ground installation at 30 degrees C ambient nonoperating
temperature. The power is cycled "on" once every 1,000 1
nonoperating hours.
Steep1: The nonoperating failure rate model shown in Section 5.2.3.1
above is XpP Anb 7NT TNE TNQ Itcyc
"A-27
MIL-HDBK-217
Step 2: Refer to Table 5.2.3-1 and find the appropriate Xnb of .00027
failures per 100 hrs for a Group I Silicon, NPN type
transistor.
Step 3: From Table 5.2.3-1 note that transistors, Silicon, NPN fall
into the Group I category of discrete semiconductors.
Step 4: Next refer to Table 5.2.3-2 and find appropriate noncperating

temperature factor of 1.22 for a Group I Silicon, NPN
transistor at 300C ambient nonoperating temperature.
Step 5: Refer to Table 5.2.3-5 and find the correct nonoperating
environmental factor (wNE) of 5.8 for a Group I transistor in
a ground fixed environment (GF).
Step 6: Refer to Table 5.2.3-6 and find correct nonoperating quality
factor INQ of 3.6 for a JAN type transistor quality level.
Step 7: Next use Table 5.2.3-7 and find appropriate power on/off
equipment cycling factor (rcyc) of 1.05 for a transistor which
has an equipment cycling rate of 1 per 1000 nonoperating
hours.
Step 8: Determine the predicted transistor nonoperat.ng failure rate
(Xp) in failures per million hours.
= Xnb INT wNE wNQ Ircyc j* -
XP= .00027 x 1.22 x 5.8 x 3.6 x 1.05

Xp = 0.00722 failures per million hours
5.2.3.2 Opto-Electronic Semiconductor Devices
The general nonoperating failure rate prediction model for opto-

electronic semiconductor devices is as follows:
XP =nb wNE wNQ failures/10 6 nonoperating hours
where
Ap predicted nonoperating opto-electronic device failure rate
A-28
MIL-HDBK-217
Xnb - nonoperating base failure rate (See Table 5.2.3-1)

lNE -,nonoperating environmental factor (See Table 5.2.3-5)
liNQ = nonoperating quality factor (See Table 3.2.3-6)
5.2.3.2.1 Example Nonoperating Failure Rate Calculation
Given: A commercial quality plastic-encapsulated single opto-isolatnr

is being stored in a ground, benign application
Step 1: The nonoperating failure rate model shown in 5.2.3.2 above for
opto-electronic devices is Xp - Xnb lNE 1INQ
Step 2: From the example statement, the device is a single opto-

isolator. Refer to Table 5.2.3-1 and find appropriate Group
X, single isolator, nonoperating base failure rate of 0.00070 • '
failures per million hours for this particular opto-electroric
device.
Step 3: Next use Table 5.2.3-5 and select Ground Benign (GB) factor of
1 for this group X device. _
Step Refer to Table 5.2.3-6 and select "plastic" quality level

factor of 23 for this device,
3tep 5: Determine the predicted single isolator nonoperating failure
rate (Xp) in failures per million hours.
Xp = Xnb wNE NQ L
Xp - 0.00070 x 1 x 23
p= 0.016 failures per million hours
A-29
MIL-HDBK-217
DISCREIE SEMICONDUCTORS
"TABLE 5.2.3-1: DISCRETE SEMICONDUCTOR NONOPERATING BASE FAILURE

RATE (Xnb)
SXnb (F~ilures
Part Class Group Part Type per 100 hrs)
A. Transistors I Si, NPN .00027

Si, PNP .00027
Ge, PNP .00040
"Ge, NPN .00040
II FET .00039
III Unijunction .0013

B. Diodes IV Si, Gen. Purpose .00017
and Gi, Gen. Purpose .00042
Rectifiers
V Zener/Avalanche .00040
VI Thyristors .00063
C. Microwave VII Detectors .0027
Semiconductors Mixers
and Special
Devices VIII Varactors .0027
Step Recovery
IX Microwave .041
Transistors
D. Opto-Electronic X LED .00016
Devices Single Isolator .00070
Dual Isolator .00089
Phototransistor .00038
Photo Diode .00029
Alpha-Numeric .00025
Displays
A-30
. ....
MIL-HDBK-217
DISCRETE SEMI CONDUCTORS
TABLE 5.2.3-2: TRANSISTOR NONOPERATING TEMPERATURE

FACTORS (Groups I, II, III, IX) (TNT)
N
GROUP
I I II III IX
""'T
TGUnijunc- Si Be
Micro-
oc NPN PNP PNP NPN FET tion wave
0 0.36 0.34 0.26 0.25 0.35 0.29 0.17

10 0.55 0.53 0.46 0.45 0.54 0.49 0.36
20 0.83 0.82 0.78 0.78 0.82 0.80 0.72
25 1.00 1.00 1.00 1.00 1.00 1.00 1.00
30 1.22 1.22 1.29 1.31 1.21 1.26 1.37
40 1.76 1.78 2.08 2.13 1.75 1.93 2.50
"50 2.47 2.53 3.30 3.42 2.46 2.89 4.39
"55 2.91 3.00 4.14 4.32 2.90 3.50 5.75
60 3.41 3.54 5.19 5.47 3.40 4.23 7.47
65 3.99 4.16 6.54 6.93 3.97 5.08 9.62
70 4.66 4.86 8.28 8.83 4.63 6.07 12.3
75 5.41 5.67 10.6 11.4 5.37 7.23 15.6
80 6.27 6.59 13.7 14.8 6.22 8.58 19.7
85 7.26 7.64 18.2 19.7 7.17 10.2 24.7
90 8.38 8.83 24.9 26.9 8.26 12.0 30.7
95 9.67 10.2 9.50 14.1 38.0
100 11.1 11.7 10.9 16.5 46.8
105 12.8 13.5 12.5 19.4 57.3
110 14.8 15.6 14.4 22.7 69.8
115 17.0 17.9 16.5 26.6 84.5
"120 19.6 20.7 18.9 31.2 102
125 22.6 23.9 21.8 36.7 122
130 26.1 27.6 25.2 43.1 146
135 30.3 32.1 50.9 174
140 35.2 37.5 60.4 206
145 41.1 44.0 72.0 243
150 48.2 52.1 86.4 285
155 56.8 62.3 105 334
160 67.4 75.3 128 389
T : ambient nonoperating temperature (oc)

Temperature factor equation and parameter values are given in Table 5.2.3-4
A-31
MIL-HDBK-217
TABLE 5.2.3-3: DIODE NONOPERATING TEMPERATURE

FACTORS (Groups IV, V, VI, VII, VIII) (TNT)
GROUP
IV V VI VII VIII
T Gen Purpose Zener/ Varactor,
Av. Thyristor Microwave etc.
oc Si Ge
0 0.26 0.17 0.39 0.27 0.43 0.35
10 0.46 0.36 0.58 0.47 0.62 0.54
20 0.78 0.72 0.84 0.79 0.86 0.82
25 1.00 1.00 1.00 1.00 1.00 1.00
30 1.28 1.39 1.19 1.30 1.17 1.21
40 2.03 2.60 1.65 2.07 1.56 1.75
50 3.14 4.73 2.24 3.20 2.06 2.46
55 3.87 6.32 2.59 3.95 2.35 2.90
60 4.74 8.45 2.99 4.85 2.68 3.40
65 5.78 11.3 3.44 5.92 3.04 3.97
70 7.00 15.2 3.94 7.21 3.44 4.63
75 8.43 20.5 4.50 8.73 3.89 5.37
80 10.1 28.1 5.13 10.5 4.40 6.22
85 12.1 39.5 5.84 12.7 4.97 7.17
90 14.4 57.1 6.63 15.2 5.61 8.26
95 17.1 7.52 18.2 6.34 9.50
100 20.2 8.52 21.8 7.18 10.9
105 23.9 9.65 26.0 8.16 12.5
110 28.2 10.9 31.0 9.31 14.4
115 33.2 12.4 36.9 10.7 16.5
120 39.1 14.1 43.9 12.4 18.9
125 46.2 16.0 52.2 14.5 21.8
130 54.6 18.2 62.3 17.1 25.2
135 64.8 20.9 74.3 20.6 29.1
140 77.3 24.1 88.8 33.9
145 92.9 27.9 106 39.7
150 113 32.6 128 46.9
155 138 38.4 154 55.8
160 172 45.8 187 67.1
T = ambient nonoperating temperature (OC)

Temperature factor equation and parameter values are given in Table 5.2.3-4
A-32
MIL-HDBK-217
TABLE 5.2.3-4: DISCRETE SEMICONDUCTOR NONOPERATING

TEMPERATURE FACTOR PARAMETERS
Temp. Constants
Group Part Type At TM P
Transistors
Si, NPN 3356 448 10.5
Si, PNP 3541 448 14.2
'Ge, PNP 4403 373 20.8
Ge, NPN 4482 373 19
II FET 3423 448 13.8
III Unijunction 4040 448 13.8

4,,
"Diodes
4399 448 17.7
IV Si, Gen. Purpose
Ge, Gen. Purpose 5829 373 22.5
V Zener/Avalanche 3061 448 14
VI Thryistors 4311 448 9.6
VII Microwave 2738 423 16.6
IMPATT, Gunn,
-:j Varactor, PIN, 3423 448 13.8
VIII Step Recovery &
Tunnel
Transistors Microwave 5700 623 20

IX
7•NT
=exp(-At(T - M98)+
"where
"T= temperature ( 0K) T(*C) + 273
A*-33
MIL-HDBK-217
ENVIRONMENTAL FACTOR (1NE)
GROUP
Env. I II III IV V Vi VII VIII IX X
Gg 1 1 1 1 1 1 1 1 1 1
GF 5.8 4.0 4.0 3.9 3.9 3.9 6.4 3.9 2.0 2.4
GM 18 18 18 18 18 18 31 18 7.8 7.8
Mp 12 12 12 12 12 12 35 12 7.4 7.7
NSB 9.8 6.0 9.3 4.8 5.8 5.8 8.0 5.8 3.6 3.7
NS 9.8 8.6 9.3 4.8 8.7 8.7 11 8.7 4.7 5.7

NU 21 21 21 21 21 21 33 21 11 11
NH 19 19 19 19 19 19 54 19 11 12
NUU 20 20 20 20 20 20 58 20 12 13
ARW 27 27 27 27 27 27 78 27 16 17
AIC 9.5 7.5 9.5 15 4.5 9.5 30 4.5 2.5 2.5
AIT 15 9 15 20 6.5 15 40 6.5 3.5 3.5
AIB 35 35 35 30 45 35 65 45 6.0 5.5
AIA 20 30 20 25 25 20 50 25 3.5 3.5
AIF 40 40 40 35 45 40 70 45 6.0 5.5
AUC 15 10 15 25 7.5 15 50 7.5 5.0 3.0
AUT 25 15 25 30 10 25 60 10 7.0 5.5
V AUB 60 55 60 50 70 60 105 70 10 8.0
AUA 35 50 35 40 40 35 80 40 7.0 5.5
AUF 65 65 65 50 70 65 110 70 10 10
SF 1 1 1 1 1 1 1 1 1 1
MFF 12 12 12 12 12 12 36 12 7.5 7.8
mMFA 17 17 17 17 17 i7 50 17 11 11
USL 36 36 36 36 36 36 110 36 22 23
z ML 41 41 41 41 41 41 120 41 25 26
CL 690 690 690 690 690 690 2000 690 250 450
A-34
i3
MIL-HDBK-217
QUALITY FACTOR (•NQ)
Quality Level ffNQ
JANTXV 0.57
JANTX 1.0
JAN 3.6
Lower, Hermetic* 13
Plastic** 23
applies to all hermetic packaged discrete semiconductors devices and

to Non-JAN hermetic packaged devices.
** applies to all discrete semiconductor devices encapsulated with

organic material
TABLE 5.2.3-7: TRANSISTOR EQUIPMENT POWER ON-OFF

CYCLING FACTOR (rrcyc) (Groups I, II, III, IX)
Cycling Rate***(Nc)
(Powqr Cycles/ Mean-Time-Between
10J hrs.) Power Cycles lfcyc
<1 >1000 1.00

1 1000 1.05
2 500 1.10
3 333 1.15
4 250 1.20
5 200 1.25
10 100 1.50
20 50 2.00
50 20 3.50
ncyc= 1 + .050(Nc)
Nc number of equipment power on-off cycles per 1000 nonoperating hours
An equipment power on-off cycle is defined as the state

S*** duiring which
an electronic e quiipment goes from zero electrical activation level to
the normal design activation level plus the state during which is
returns to zero.
A-35
6 [-
MIL-HDBK-217
TABLE 5.2.3-8: DIODE EQUIPMENT POWER ON-OFF CYCLING

g FACTOR (Wcyc) (Groups IV,V,VI, VII, VIII)
Cycling Rate***(Nc)
(Power Cycles/ Mean-Time-Between
103 hrs.) Power Cycles cyc
<0.6 >1667 1.00

1 1000 1.08
2 500 1.17
34 333
250
1.25
1.33 "'
5 200 1.42•
10 i00 1.83
20 50 2.66
50 20 5.15•
•cyc = 1 +,.083(Nc) •~
Nc = number of equipment power on-off cycles per 1000 nonoperating hours
An equipment power on-off cycle is defined as the state during which

returns to zero.
A-36
* MIL-HDBK-217
•_jTUBES S...
I
5.2.4 Tubes

for electronic vacuum, microwave and other tube types.
The general model for tubes is as follows:
Xp = •NE
7nb failures/10 6 nonoperating hours
where
Xp = predicted tube nonoperating failure rate

Xnb = nonoperating base failure rate (See Table 5.2.4-1)
TNE = nonoperating environmental factor (See Table 5.2.4-2)
5.2.4.1 Example Nonoperating Failure Rate Calculation
Given: A pentode receiver type tube is being used in a Ground Mobile

environment.
Step 1: The nonoperating failure rate model is shown in section 5.2.4 to

be Xp = Xnb tNE.
Step 2: Refer to Table 5.2.4-1 and select correct nonoperating base

failure rate of 0.0040 for a pentode receiver tube
Step 3: Next select from Table 5.2.4-2 the nonoperating environmental
factor of 31 for a ground mobile (GM) environment.
Step 4: Determine the predicted tube nonoperating failure rate as
follows:
p Xnb iTNE
X' = 0.0040 x 31
= 0.124 failures/10 6 hours

-
'1 A-37
7 73,,ý
t4c
MIL-HDBK-217
TUBES
TABLE 5.2.4-1: TUBE NONOPERATING BASE FAILURE RATE (Xflb)
TYPE TYPE Xnb(failures/10 6 hrs)
RECEIVER
Triode, Tetrode, Pentode 0.0040
Power Rectifier 0.0090
CRT 0.013
I
THYRATRON 0.32
VIDICON 0.049
CROSSED FIELD AMPLIFIER
PULSED GRIDDEDa
TRANSMITTING
Triode, Tetrode, Pentode 0.56
Any Style with Peak Pwr. < 200 kW, Freq. 1.61
< 200MHz, or Average Pwr. < 2KW
(All types) 2.60
2 MAGNETRON
(All types), 1.02U
KLYSTRON
Continuous Wave 1.20
Low Power 0.19
Pulsed 1.15
TWT 0.69
A-38
-SLI
MIL-HDBK-217
TUBES
TABLE 5.2.4-2: TUBE NONOPERATING ENVIRONMENTAL FACTORS (iTNE)
Environment iTfNE
GB 1
GF 3.0
GM 31
Mp
NSB
31
15 F
NU 47
NH 110H
NUU 120
ARW 140
AIC 6.2
AIT 19
AIB 25
AIA 23
AIF 35
AUC 8.2
AUT 23
AUB 33
AUA 27
AUF 43
SF 1
MFF 63
MFA 91
USL 210
ML 220
CL 3600 '
A-39
all.
MIL-HDBK-217
LASERS
5.2.5 Lasers
This section presents nonoperating failure rate models for laser

peculiar items used in the following five major classes of laser
equipment: i!W
o 'Helium/Neon
o Argon Ion
o CO2 Sealed
0
o
CO
2 Flowing
Solid State
II °;,
The models and failure rates presented in t.his section apply to the
laser peculiar items only, i.e., those items wherein the lasing action is
generated and controlled. In addition to the laser peculiar items, there
are other assemblies used with lasers that contain electronic parts and
mechanical devices (pumps, valves, hoses, etc.). The failure rates for
these parts should be determined with the same procedures as used for
M
other electronic and mechanical devices in the equipment or system of
which the laser is a part. The electronic device failure rates are in k 4M
other parts of this Handbook and the mechanical device failure rates are
in Bibliography Item 47.
The laser failure rate models have been developed at the "functional,"
rather than "piece part," level because the available data were not
sufficient for "piece part" model development. Nevertheless, the laser
functional models are included in this Handbook in the interest of
completeness. These laser models will be revised to include piece part
models and other laser types when the data become available.
Because each laser family can be designed using a variety of

approaches, the failure rate models have been structured on three basic
A-40
V -V
MIL-HDBK-217
LASERS
laser functions which are common to most laser families, but may differ in
the hardware implementation of a given function. These functions are the
lasing media, the laser pumping mechanism (or pump), and the coupling
method.
Helium/Neon Lasers
the genera! nonoperating failure rate prediction model for helium/neon

lasers is as follows:
XHe/Ne = .11 "NE failures/10 6 nonoperating hours
where
XHe/Ne = helium/neon laser nonoperating failure rate (includes the

nonoperating failure rate contribution for the lasing media,
the laser pumping mechanism and the coupling method)
"•NE = nonoperating environmental factor (See Table 5.2.5-1)
Argon Ion Lasers
The general nonoperating failure rate prediction model for Argon Ion
lasers is as follows: I!
ýAI .61 "NE failures/10 6 nonoperating hours
where
-AI argon ion laser nonoperating failure rate (includes the

nonoperating failure rate contribution for the lasing media, the
"laser pumping mechanism and the coupling method)
hNE nonoperating environmental factor (See Table 5.2.5-1)
'ft A-4 1
tii
MIL-HDBK-217 *'.
LASERS
Carbon Dioxide. Sealed Lasers
The general nonoperating failure rate prediction model for carbon

dioxide sealed lasers is as follows:
XCO2 SEALED= (.65 + .O13(Nop)) rNE failures/lO6 nonoperating hours
where
CO2 SEALED = carbon dioxide sealed nonoperating failure rate

(includes the nonoperating failure rate contribution
for the lasing media, the laser pumping mechanism and
the coupling method)
Nop= number of active optical surfaces (determine from Figure
5.1.5.7-4 in Section 5.1.5.7)
"NE = nonoperating environmental factor (See Table 5.2.5-1)
Carbon Dioxide, Flowing Lasers rj

The general nonoperating failure rate prediction model for carbon
dioxide flowing lasers is as follows:
•0O
2 FLOWING= .039(Nop) 7NE
where
X FLOWING = carbon dioxide sealed laser nonoperating failure rate

(includes the nonoperating failure rate contribution
for the lasing media, the laser pumping mechanism and
the coupling method).
Nop= number of active optical surfaces (determine from
Figure 5.1.5.7-4 in Section 5.1.5.7) .
•NE nonoperating environmental factor (See Table 5.2.5-1)
A-42
MIL-HDBK-217
LASERS .
Solid State Lasers (Nd:YAG or Ruby Rod)
The general nonoperating failure rate pr-ediction model for solid state
lasers (either pumped by xenon or Krypton flashlamp) is as follows:
)ss (.062 + .021(Nop)) NNE
where
Xss state laser nonoperating failure rate (includes the

=solid
nonoperating failure rate contribution for the lasing media, the

laser pumping mechanism and the coupling method)
No number of active optical surfaces (determined from Figure
5.1.5.7-4 in Section 5.1.5.7)
NTNE =nonoperating environmental factor (See Table 5.2.5-1) :
TABLE 5.2.5-1: LASER NONOPERATING ENVIRONMENTAL FACTOR (wNE)
Environment WNE Envirotment lINE
18 AIA 44
GF 9.0 AIF 62
GM 44 AUC 35ý
MP 22 AUT 42
NSB 10 AUB '71 2
NS 49 AUA 57
NU 49 AUF 71
NH 35 SF 1
NUU 38 MFF 21
ARW 46 MFA 29
AIC 26 USL 69
AT35 ML 71
A18 58 CL
A-43
MIL-HDBK-217
LASERS
Given: Nd:YAG laser designator using a xenon flashlamp is being used in

the AUF nonoperating environment. The laser includes one totally
reflective (TR) mirror, one prismatic "Q" switch, one partially
reflective (PR) mirror and one exit lens.
Step 1: The nonoperating failure rate model for solid state lasers shown
in Section 5.2.5 is Xss = (.062 + .021(Nop))TNE
Step 2: From Figure 5.1.5.7-1 in Section 5.1.5.7, determine the number of

optical surfaces (Nop) by,
(1) One TR mirror = 1 optical surfaces
(2) One "Q" switch = 2 optical surfaces
(3) One PR mirror = 2 optical surfaces
(4) One exit 'ens = 2 optical surfaces
Total active optical surfaces = 7
Step 3: Refer to Table 5.2.5-1 and find the correct nonoperating

environmental factor (•NE) of 71 for an AUF environment.
Step 4: Determine the predicted nonoperating Nd:YAG nonoperating failure
rate in failures per 106 nonoperating hours.
ss (.062 + .021(7)) x 71 = 14.8 failures/10 6 nonoperating
hours
A-44
.~)~-~' ,' Z
MIL-HDBK-217
RESISTORS
5.2.6 Resistors

for resistors.
The general model for resistor devices is as follows:
P= Xnb INE •NQ Icyc failures/10 6 nonoperating hours
where
Xp = predicted resistor nonoperating failure rate

"Xnb = nonoperating base failure rate (See Table 5.2.6-1)
7NE = nonoperating environmental factor (See Table 5.2.6-2)
wNQ = nonoperating quality factor (See Table 5.2.6-3)
'Tcyc = equipment power on-off cycling factor (See Table 5.2.6-4)

. .
Given: Type RCR fixed composition 12,000 ohm resistor per MIL-R-39008,
level M rated at 0.5 watts is being used in a trainer aircraft
cockpit equipment. The power is cycled "on" twenty times every
1,000 nonoperating hours.
Step 1: The nonoperating failure rate model is shown in Section 5.2.6

"•tobe p Xnb 'TNE 7NQ !Tcyc.
Step 2: Refer to Table 5.2.6-1 and find RCR style and appropriate Xnb of
.00063 failures per million hours and find appropriate resistor
type (Fixed Composition) for the specific resistor RCR style.
A-45
~
.**-*~ .*.-,**
,-. *~-* , *~-~
Y*-
MIL-HDBK-217
RESISTORS
TABLE 5.2.6-1: RESISTOR NONOPERATING BASE FAILURE RATE (Xnb)
Anb (Fgilures
Specification Style per 10 hrs)
Composition, Fixed
MIL-R-11 Resistors, Fixed, Composition (Insulated) RC .000063
MIL-R-39008 Resistors, Fixed, Composition (Insulated) RCR .000063
Established Reliability
Film, Fixed
MIL-R-l0509 Resistors, Fixed, Film (High Stability) RN .00010
MIL-R-11804 Resistors, Fixed, Film (Power Type) RD .00010
MIL-R-22684 Resistors, Fixed, Film, Insulated RL .00010
MIL-R-39017 Resistors, Fixed, Film, Insulated, RLR .00010
MIL-R-55182 Resistors, Fixed, Film, Established Reliability RN(R, C or N) .00010
Network, Film, Fixed

MIL-R-83401 Resistor Network, Fixed, Film RZ .00043
Wirewound, Fixed
MIL-R-26 Resistors, Fixed, Wirewound (Power Type) RW .00057
MIL-R-93 Resistors, Fixed, Wirewound (Accurate) RB .00057
MIL-R-18546 Resistors, Fixed, Wirewound (Power Type, RE .00057
Chassis Mounted
MIL-R-19005 Resistors, Fixed, Wirewound (Accurate), R8R .00057
MIL-R-39007 Resistors, Fixed, Wirewound (Power Type) RWR .00057
MIL-R-3900g Resistors, Fixed, Wirewound (Power Type RER .00057
Chassis Mounted) Established Reliability
Thermistor
MIL-T-23648 Thermistor (Thermally Sensitive Resistor) RTH .0027
Insulated
Non-wirewound, Variable
MIL-R-94 Resistors, Variable, Composition RV .0052
MIL-R-22097 Resistors, Variable, Non-wirewound RJ .0052
(Lead Screw Actuated)
MIL-R-23285 Resistors, Variable, Film RVC .0052
MIL-R-39023 Resistors, Variable, Non-wirewound, Precision RQ .0052
MIL-R-39035 Resistors, Variable, Cermet, or Carbon Film RJR .0052
(Lead Screw Actuated) Established Reliability
Wirewound, Variable
MIL-R-19 Resistors, Variable, Wirewound (Low Operating RA .0052
Temperature)
MIL-R-22 Resistors, Variable, Wirewound (Power Type) RP .0052
MIL-R-12934 Resistors, Variable, Wirewound, Precision RR .0052
MIL-R-27208 Resistors, Variable, Wirewound (Lead Screw RT .00099
". Actuated)
MIL-R-39002 Resistors, Variable, Wirewound, Semi-Precision RK .0052
MIL-R-39015 Resistors, Variable, Wirewound (Lead Screw RTR .00099
Actuated), Established Reliability
A-46
- W
MIL-HDBK-217
RESISTORS
TABLE 5.2.6-2: RESISTOR NONOPERATING ENVIRONMENTAL FACTORS (TNE)

Fixed Fixed Film Fixed
Env. Comp. Film Network WW Therm. Var. non-WW Var. WW
GB 1 1 1 1 1 1 1
GF 2.9 2.4 2.4 2.1 4.8 2.5 2.5
GM 8.3 8.3 7.8 8.8 23 13 13
Mp 8.5 9.9 8.8 11 17 19 18
NSB 4 4.7 4.2 5 7.9 7 7
NS 5.2 4.9 4.7 5 14 7 7
NU 12 15 14 15 17 17 14(2)
NH 13 16 14 17 25 29 29
NUU 14 17 15 18 27 31 31
ARW 19 22 19 24 33 41 41
AIC 3 3 2.5 4.3 4.3 12 5.5
AIT 3.5 4.5 3 7.3 7.7 16 6.6
AIB 5 6.8 6.5 12 19 24 9.5.
AIA 3.5 5.8 6 9.7 15 22 8.6
AIF 6.5 9.5 9 13 38 33 14
5 7.5 6 10 4.6 19 SAUC 6.5(2)
AUT 7 11 6.5 13 8.6 25 9(2)
AUB 10 18 15 23 21 37 18(2)
AUA 7 13 15 18 17 32 13(2)
AUF 15 23 20 28 42 52 20(2)
SSF (1) (1) (1) (1) (1) 1 1
MFF 8.6 10 8.9 11 15 19 20(2)
MFA 13 14 12 16 21 26 28(2)
.USL 25 30 26 33 49 56 58(2)
* ML 29 35 30 38 51 64 66(2)
CL 490 590 510 610 950 1100 1100(2)
NOTES: 1) Env. : Environment, Comp, = Composition, WW Wirewound,
Therm. = Thermistor, Var. = Variable
2) Semiprecision wirewound or high power wirewound variable

resistors shall not be used in these environments.
A-47
Ira
U
MIL-HDBK-217
RESISTORS
TABLE 5.2.6-3: RESISTOR NONOPERATING QUALITY FACTOR
Quality Level wNQ
S 0.15
R 0.28
P 0.52
M 1.0
MIL-SPEC 2.4
Lower 4.4
TABLE 5.2.6-4: RESISTOR EQUIPMENT POWER ON-OFF CYCLING FACTOR
Cycling Rate* Mean-Time-Between

(Power Cycles/lO3 hrs.) Power Cycles (Hours) rcyc
'S< 0.8 >1250 1

1 1000 1.06
2 500 1.13
"3 333 1.19
4 250 1.25
5 200 1.32
.. - 10 100 1.63
20 50 2.26
50 20 4.15
ircyc 1 + .0 6 3(Nc)
Nc = Number of power on-off cycles per 1000 nonoperating hours
*An equipment power on-off cycle is defined as the state during which an
electronic equipment goes from zero electrical activation level to the
normal design activation level plus the state during which it returns to
zero.
I A-48
SMIL-HDBK-217
RESISTORS
Step 3: Refer to Table 5.2.6-2 and select appropriate iNE factor of 3.5
for Fixed Composition type resistor and in an AIT environment.
Step 4: Refer to Table 5.2.6-3 and select appropriate INQ factor of 1.0
for a level M quality part.
Step 5: Refer to Table 5.2.6-4 and select appropriate wcyc factor of

2.26 for a cycling rate of 20 per 1000 nonoperating hours.
Step 6: Xp =Xnb x TNE 1x NQ 1x cyc

= .000063 x (3.5) x (1) x (2.26)
Xp = .000498 failures/10 6 hours
A-49
MIL-HDBK-217
CAPACITORS
5.2.7 Capacitors

for capacitors.
The general model for capacitor devices is as follows:
Xp = Xnb 1tNE wNQ wcyc
where
),p = predicted capacitor nonoperating failure rate

Xnb = nonoperating base failure rate (..e Table 5.2.7-1)
nNE = nonoperating environmental factor (See Table 5.2.7-2)
wNQ = nonoperating quality factor (See TaL•i 5.?.7-3)
ircyc = equipment power on-off cycling factjr ýSee Table 5.2.7-4)
y Given: Style CSR solid tantalum capacitor MIL-C-39003, level R, rated

at 40 Vdc is being used in a ground fixed environment. The
power is cycled "on" twice every 1,000 nonoperating hours.
Step 1: The nonoperating failure rate model is shown in Section 5.2.7
to be Xp = ýnb wNE wNQ wcyc.
Step 2: Refer to Table 5.2.7-1 and find CSR style and appropriate Xnb of
.00018 failures per million hours.
Step 3: Refer to Table 5.2.7-1 and find appropriate capacitor type
(Electrolytic-Tantulum Solid) for the specific capacitor CSR
style. 0-
A-50
W w
MIL-HDBK-217
CAPACITORS
TABLE 5.2.7-1: CAPACITOR NONOPERATING BASE FAILURE RATE (Xnb)
Specification Style
S'pernb (Fgilures
106 hrs)
Paper/Plastic Film
MIL-C-25 Capacitors, Fixed, Paner CP .0011
MIL-C-11693 Capacitors, Fixed, Paper, Metallized Paper, CZ .0011
Metallized Plastic, RFI Feed-Thru, Established
Reliability and Non-Established Reliability
SMIL-C-12889 Capacitors, Fixed, Paper, RFI Bypass CA .0011
MIL-C-14157 Capacitors, Fixed, Paper-Plastic, Established CPV .0011
• " Reliability
MIL-C-18312 Capacitors, Metallized Paper, Paper-Plastic, CH .0011
Plastic
MIL-C-19978 Capacitors, Fixed, Plastic (or Paper-Plastic), CQ/CQR .0011
AIN Established and Non-Established Reliability
MIL-C-39022 Capacitors, Fixed, Metallized, Paper-Plastic Film CHR .0011
or Plastic Film Dielectric, Established Reliability
MIL-C-55514 Capacitors, Plastic, Metallized Plastic, CFR .0011
* Established Reliability
MIL-C-83421 Capacitors, Super-Metallized Plastic, Established CRH .0011
Mica -Reliability
Mica '
MIL-C-5 Capacitors, Fixed, Mica CM .00075
MIL-C-10950 Capacitors, Fixed, Mica, Button Sytle CB .00075
MIL-C-39001 Capacitors, Fixed, Mica, Established Reliability CMR .00075
Glass
MIL-C-11272 Capacitors, Glass CY .00045
MIL-C-23269 Capacitors, Fixed, Glass, Established Reliability CYR .00045
Ceramic
MIL-C-20 Capacitors, Fixed, Ceramic (Temperature CC/CCR .00039
Compensating)
MJL-C-11015 Capacitors, Fixed, Ceramic (General Purpose) CK .00039
MIL-C-39014 Capacitors, Fixed, Ceramic (General Purpose), CKR .00039
Electrolytic
MIL-C-62 Capacitors, Fixed, Electrolytic (DC, Aluminum, CE .0064
Dry Electrolyte, Polarized)
MIL-C-3965 Capacitors, Fixed, Electrolytic (Non-solid CL .0064
Electrolyte), Tantalum
MIL-C-39003 Capacitors, Fixed, Electrolytic, Tantalum, Solid CSR .00018
Electrolyte, Established Reliability
MIL-C-39006 Capacitors, Fixed, Electrolytic, Tantalum, Non- CLR .0064
solid Electrolyte, Established Reliability
MIL-C-39018 Capacitors, Fixed, Electrolytic, Aluminum Oxide CU .0064
Variable Capacitors
MIL-C-81 Capacitors, Variable, Ceramic CV .012
"MIL-C-92 Capacitors, Air, Trimmer CT .015
"MIL-C-14409 Capacitors, Variable, Piston Type, Tubular Trimmer PC .0038
MIL-C-23183 Capacitors, Vacuum or Gas, Fixed and Variable CG .046
A-51
**,'~ -*,-*- V;*~.., * '..-

MIL-HDBK-217
CAPACITORS
TABLE 5.2.7-2: CAPACITOR NONOPERATING ENVIRONMENTAL FACTORS (WNE)
:_-___Electrolytic
Paper/ Mica/ Tant. Tant.
Env. Plas. Film Glass Cer. Solid Non-Solid Al. Var.
GB 1.0 1.0 1.0 1.0 1.0 1.0 1.0
GF 2.2 2.1 2.0 2.4 1.4 2.0 3.3
GM 8.3 8.8 8.3 7.8 10 12 9.6
Mp 9.9 11 11 9.2 11 12 17
NSB 4.7 5.0 5.0 4.4 5.0 5.8 7.7
NS 6.3 5.9 5.2 4.9 6.7 6.7 8.2
- NU 14 15 15 13 15 13 18
NH 15 16 16 14 16 19 25
NUU 16 17 18 15 17 20 27
ARW 21 23 24 20 23 27 36
AIC 3.2 3.5 2.7 2.5 2.5 9.5 5.0
AIT 4.3 4.0 3.3 2.5 4.0 10 5.3
AIB 7.0 8.0 6.2 7.0 6.5 10 7.8
AIA 4.9 4.0 5.0 3.0 6.0 10 7.7
AIF 9.8 10 8.0 7.5 10 15 13
AUC 7.6 15 6.0 4.5 8.5 28 20
AUT 13 15 12 6.0 15 30 38
AUB 23 35 15 25 20 30 57
AUA 17 15 17 10 20 30 50
AUF 33 40 30 30 40 40 85
SF 1.0 1.0 1.0 1.0 1.0 1.0 1.0
MFF 9.9 11 11 9.3 11 12 16(1)
MFA 13 15 15 13 15 17 22(1)
USL 23 31 32 27 31 36 47(1)
ML 33 36 36 31 36 41 54(1)
CL 560 610 610 510 610 690 930
NOTES: 1) Vacuum or Gas, fixed and variable (CG) style capacitors '
shall not be used in these environments.
2) Plas.
s = Plastic,
Env ironment. Tant. = Tantalum, Al. Aluminum, Env. ,• '
A-52
i,,.
MIL-HDBK-217
CAPACITORS
7 TABLE 5.2.7-3: CAPACITOR NONOPERATING QUALITY FACTOR
Quality Level INQ
T 0.05
S 0.10
R
P
0.23
0.46
SM 1.0
L 1.7
MIL-SPEC 2.5
Lower 5.3
TABLE 5.2.7-4: CAPACITOR EQUIPMENT POWER ON-OFF CYCLING FACTOR
Cycling Rate* Mean-Time-Between

(Power Cycles/t0 3 hrs.) Power Cycles (Hours) 'cyc
< 0.3 >3333 1

0.5 2000 1.08 1,4
1 1000 1.16
2 500 1.32
3 333 1.48
4 250 1.64
5 200 1.80
10 100 2.60
20 50 4.20
50 20 9.00
1Tcyc : 1 + O.1 6 (Nc)
Nc = Number of equipment power on-off cycles per 1000

nonoperating hours
*An equipment power on-off cycle is defined as the state during which an
electronic eqq_!ipment goes from zero electrical activation level to the
* normal design activation level plus the state during which it returns to
zero.
A-53
VJ
MIL-HDBK-217
CAPACITORS
Step 4: Refer to Table 5.2.7-2 and select appropriate TNE factor of 2.4
for Electrolytic-Tantalum Solid type resistor in a GF
environment.
Step 5: Refer to Table 5.2.7-3 and select appropriate 'INQ factor of 0.23
for a level R quality part.
Step_6: Refer to Table 5.2.7-4 and select appropirate ir•yc factor of
1.32 for a cycling rate of 2 per 1000 nonoperating hours.
Ste_ 7: )p Xnb x INE x 1NQ x 7cyc

.00018 x (2.4) x (0.23) x (1.32)
.0001312 failures/106 hours
FNl
A-54
'777~
MIL-HDBK-217
INDUCTIVE DEVICES
5.2.8 Inductive Devices
This section presents t• nonoperating failure rate prediction model

for inductive devices. The inductive devices included in this section are
transformers and coils.
The general model for inductive devices is as follows:
Xp= nb TNQ 7NE Icyc failures/10 6 nonoperating hours
where
Xp =
predicted transformer or coil nonoperating failure rate
Xnb =
nonoperating base failure rate (See Table 5.2.8-1)
wNQ =
nonoperating quality factor (See Table 5.2.8-2)
ffNE =
nonoperating environmental factor (See Table 5.2.8-3)
7rcyc = equipment power on-off cycling factor (See Table 5.2.8-4 for
transformers and 5.2.8-5 for coils)
Given: Power transformer with type designation TF5SXO3GA203 procured per

the requirements of MIL-T-27. The power is cycled "on" ten times
every 1000 nonoperating hours, and is in a ground fixed
environment.
Step 1: The nonoperating failure rate model is shown in Section 5.2.8 to
be P = Xnb 1TNQ fNE ncyc
Step 2: Refer to Table 5.2.8-1 and find MIL-T-27 power transformer part
class, and appropriate Xnb of .00028.
Step 3: Refer to Table 5.2.8-2 and select appropriate ?TNQ of 3.1 for a
MIL-SPEC quality part.
A-55
4 .1
MIL-HDBK-217
INDUCTIVE DEVICES
Stp4 Refer to Table 5.2.8-3 and select appropriate XNE of 5.1 for a
ground fixed environment.
Step 5: Refer to Table 5.2.8-4 and select appropriate itc of 8.5 for an
equipment power cycling rate of 10 cycles per 1000 hours.
Step 6: Xpfb~NXYEwy
- (.&OJO28) x (3.1) x (5.7) x (8.5)
- .042 failureif10 6 nonoperating hours
"A
A-56
MIL-HDBK-217
INDUCTIVE DEVICES
TABLE 5.2.8-1: INDUCTIVE DEVICE NONOPERATING BASE FAILURE RATE (Xnb)
Part Class Xnb
Transformers
MIL-T-27 Transformers and Inductors, Audio .000055
MIL-T-27 Transformers and Inductors, Power .00028
MIL-T-27 Transformers and Inductors, High Power Pulse .00028
MIL-T-21038 Transformers, Low Power Pulse .000055
MIL-T-55631 Transformers, IF, RF, and Discriminator .00028
Coils
MIL-C-15305 Coils, Fixed and Variable, RF .00015
MIL-C-39010 Coils, Molded, RF, ER .00015
TABLE 5.2.8-2: INDUCTIVE DEVICE NONOPERATING QUALITY FACTOR (1!NQ)
Quality Level W 1TNQ
S .06
R .15
P .38
M 1.0
MIL-SPEC 3.1
Lower 1
*S,R,P and M levels refer to coils only.
A-57
MIL-HDBK-217
INDUCTIVE DEVICES
TABLE 5.2.8-3: INDUCTIVE DEVICE NONOPERATING ENVIRONMENTAL

FACTORS (WNE)
Environment TrnfresCoils
GB 1 1
GF 5.7 3.6
GM 12 12
MP11 11
NSB 5.1 5.1
NS 5.7 5.7
NU14 14
NH 16 16
NUU 18 18
ARW 24 24
AIC 4.5 4
AIT 6 4.5
AIB 6 5.5
AIA 6 4.5
AIF 99
AUC 6.5 5
AUT 6.5 6.5
AUB 7.5 7.5
AUA 7.5 6.5
AUF 10 10
SF 1 1
MFF 11* 11
MFA 15 15
USL 32 32
ML 36 36
CL 310 610
A-58
MIL-HDBK-217
INDUCTIVE DEVICES
TABLE 5.2.8-4: TRANSFORMER EQUIPMENT POWER ON-OFF
Cycling Rate*
(Power Cycles/ Mean-Time-Between
103 hrs.) Power Cycles 7cyc
.05 >20,000 1.0

1. 10,000 1.08
.2 5,000 1.15
.5 2,000 1.38
1 1,000 1.75
2 500 2.50
5 200 4.75
10 100 8.50
20 50 16.0
50 20 38.5
'cyc - 1 + .75(Nc)
Nc = number of equipment power on-off cycles per 1000 nonoperating hours
TABLE 5.2.8-6: COIL EQUIPMENT POWER ON-OFF

Cycling Rate*
"(Power Cycles/ Mean-Time-Between
10J hrs.) Power Cycles i tcyc
• 1 .>10,000 1.00
.2 5,000 1.08
.5 2,000 1.19
1 1,000 1.38
2 500 1.76
5 200 2.90
10 100 4.80
20 50 8.60
50 20 20.0
-,cyc :1 + .38(Nc)
Nc number of equipment power on-off cycles per 1000 nonoperating hours

* iLAn equipmnwt power on-off cycle is defined as the state during which
an electronic equIpment goes from zero electrical activation level to
returns to zero.
A-59
1'
MIL-HDBK-217
ROTATING DEVICES
5.2.9 Rotating Devices
This section presents the method to be used for estimating the

nonoperating failure rate for motors with power ratings below one
"horsepower, synchros and resolvers, and elapsed time meters.
Average nonoperating failure rates for rotating devices are presented

in Table 5.2.9-1.
TABLE 5.2.9-1: NONOPERATING FAILURE RATES FOR ROTATING DEVICES

Part Type Nonoperating Failure Rate

Motors (ac or dc) .045

Synchros .14
Resolvers .14
Elapsed Time Meters 1.2
/
SI
*, .,
.Ii
A- go60,
tkiI MIL-HDBK-217
2 .: RELAYS
5.2.10 Relays

for relays:
The general model for relays is as follows:
= Xnb WNQ •NE failures/lO6 nonoperating hours
where
xrp = predicted relay nonoperating failure rate

-nb = nonoperating base failure rate (See Table 5.2.10-1)
nNQ = nonoperating quality factor (See Table 5.2.10-2)
1tNE = nonoperating environmental factor (See Table 5.2.10-3)
TABLE 5.2.10-1: RELAY NONOPERATING BASE FAILURE RATE (Xnb)
Contact Voltage Nonoperating Base Failure Rate

Package Type (when operated) (failures/106 nonoperating hrs)
Nonhermetic <50 millivolts .010

Nonhermetic 750 millivolts .002
Hermetic any .0004
TABLE 5.2.10-2: RELAY NONOPERATING QUALITY FACTOR (wNQ)
Quality Level 1rNQ

MIL-SPEC 1.0
Lower 4.2
A-61
AWI6o.
'" . ',• •/ '/,••I•• • - .J " • • •', ' '-'- ,•x ,-••- •'• •,,.'' o• - -- •-•..... .... . .. .. .•.. . . - . ... .
MIL-HDBK-217
RELAYS
TABLE 5.2.5-1: RELAY NONOPERATING ENVIRONMENTAL FACTOR (rNE)
Environment "NE Environment "NE
GB 1 AIA 7.5
GF 2.3 AIF 10
GM 8.2 AUC 8.0
Mp 21 AUT 9.0
NSB 8.0 AUB 15
NS 8.0 AUA 10
NU 14 AUF 15
NH 32 SF 1
NUU 34 MFF 21
ARW 46 MFA 29
AIC 5.5 USL 62
IAIT 6 ML 71
AIB 10 CL N/A

Given: A MIL-SPEC double-pole, double-throw armature relay is being used
in a nonoperating ground fixed environment. The design operating
contact voltage is 0.5 volts.
be Xp = Xnb "NQ "NE
Step 2: Since armature relays are nonhermetic, refer to Table 5.2.10-1
and find appropriate Xnb of .002 for nonhermetic relays with a
contact voltage greater than 50 milliviIts when operated.
Step 3: Refer to Table 5.2.10-2 and find appropriate "NQ o• 1.0 for MIL-
SPEC parts.
Step 4: Refer to Table 5.2.10-3 and select "NE of 2.3 for a ground fixed
environment.
Step 5: X = XNE NQ
:nb
= (.002) x (1.0) x (2.3)
= .0046 failures/10 6 nonoperating hours
A-62
½ v
MIL-HDBK .217
SWITCHES
* 5.2.11 Switches

for switches.
The general model for switches is as follows:
X Xnb INQ fiNE failures/10 6 nonoperating hours
where
Xp = predicted switch nonoperating failure rate

7-NQ = nonoperating quality factor (See Table 5.2.11-2)
•NE = nonoperating environmental factor (See Table 5.2.11-3
TABLE 5.2.11-1: SWITCH NONOPERATING BASE FAILURE RATE (Xnb)
- Contact Voltage Nonoperating Base Failure Rate

(when operated)
<50 millivolts 0.030

>50 millivolts 0.006
TABLE 5.2.11-2: SWITCH NONOPERATING QUALITY FACTOR (INQ)
Quality Level 1TNQ

MIL-SPEC 1.0
Lower 4.2
A-63
MIL-HDBK-217
SWITCHES
TABLE 5.2.11-3: SWITCH NONOPERATING ENVIRONMENTAL FACTOR (WNE)
Environment "NE Environment iTNE
GB 1 AIA 14
GF 2.9 AIF 18
GM 13 AUC 9
Mp 21 AUT 9
NSB 7.9 AUB 18
NS 7.9 AUA 18
NU 18 AUF 23
NH 32 SF 1
NUU 34 MFF 19
ARW 41 MFA 26
AIC 7.2 USL 63
AIT 7.2 ML 64
AIB 14 CL 1200

Given: A MIL-SPEC rotary switch is installed in an airborne inhabited,
cargo nonoperating environment. The switch, when operated, is
used to transfer digital circuits with less than 50 millivolts.
be Xp = Xnb •NQ ffNE.
Step 2: Refer to Table 5.2.11-1 and select nonoperating base failure rate
of 0.030 for contact voltage less than 50 millivolts
Step 3: Refer to Table 5.2.11-2 and select appropriate nonoperating
quality factor of 1.0 for MIL-SPEC devices.
Step 4: Refer to Table 5.2.11-3 and select iTNE of 7.2 for airborne,
inhabited environment.
Ste•_9___: nb 1tNQ ITNE
X~p =: X~,'N N K
Step 5:
- (.030) x (1.0) x (7.2)
A-64
-7-ý--11'-'17'Lý
MIL-HDBK-217
CONNECTORS
5.2.12 Connectors

for connectors:
The general model for connectors is as follows:
- - Xp = Xnb •NE failures/10 6 nonoperating hours
where
Xp = predicted connector nonoperating failure rate

7 " NE = nonoperating environmental factor (See Table 5.2.12-2)
TABLE 5.2.12-1: CONNECTOR NONOPERATING BASE FAILURE

RATE (Xnb)
Connector Type Nonoperating Base Failure Rate

(failures/106 nonoperating hrs)
J
Circular .00044
Coaxial .00044
Power .00044
Rack and Panel .0029
Printed Wiring Board .0029
A-65w
'.4
!I
y
MIL-HDBK-217
CONNECTORS
TABLE 5.2.12-2: CONNECTOR NONOPERATING ENVIRONMENTAL FACTOR (WNE)
Environment wNE Environment WNE
1GB AIA 5.3

GF 2.3 AIF 11
GM 8.3 AUC 4.3
Mp 8.5 AUT 15
NSB 4.1 AUB 9.8
NS 5.5 AUA 8.0
NU 13 AUF 15
NH 13 SF 1
NUU 14 MFF 8.5
ARW 19 MFA 12
AIC 2.8 USL 25
AIT 4.8 ML 29
AIB 7.0 CL 490
Given: A MIL-SPEC circular connector with 20 pins is installed in a

ground fixed environment.
be xp = Xnb ME
Step 2: Refer to Table 5.2.12-1 and select appropriate )nb of .00044. L
Step 3: Refer to Table 5.2.12-2 and select nNE of 2.3 for a ground fixed
environment.
Step 4: Xp = Xnb ME
= (.00044) x (2.3)
.00101 failures/10 6 nonoperating hours F,7
A-66
/ 'e.<
' !:,
MIL-HDBK-217__
INTERCONNECTION ASSEMBLIES __
5.2.13 Interconnection Assemblies with Plated Through Holes (PTHs)

for interconnection assemblies with PTHs. The interconnettion assembly
model predicts the nonoperating failure rate for both the interconnection
board and the solder connections used to connect the compoients to the
interconnection board. For interconnection assemblies withoU, PTHs, use
Section 5.2.14, Connections. f
The general model for interconnection assemblies is as follows:
xp = Xnb x Npth x "NE failures /106 hours
where
xp = predicted interconnection assembly nonoperating failure !-ate

Npth = number of functional plated through holes (includes nonsoldered
functional via holes)
"NE nonoperating environmental factor (See Table 5.2.13-2)
TABLE 5.2.13-1: INTERCONNECTION ASSEMBLY NONOPERATING BASE FAILURE

RATE (Xnb)
Intercornnection Technology Xnb
Double-Sided Soldered Printed Wiring .0000014

Multilayer Soldered Printed Wiring .0000028
Discrete Wiring w/Electroless Deposited PTH .0000089
A-67
/
MIL-HDBK-217
INTERCONNECTION ASSEMBLIES
TABLE 5.2.13-2: INTERCONNECTION NONOPERATING ENVIRONMENTAL
FACTOR ("NE)
Environment iNE Environment wNE
GB 1 AIA 5.0
GF 2.3 AIF 9.0
GM 6.9 AUC 5.4
Mp 6.9 AUT 11
NSB 4.1 AUB 18

NS 5.3 AUA 14
NU 11 AUF 25
NH 13 SF 1
"NUU 14 MFF 7.8
ARW 17 MFA 11
AIC 2.3 USL 25
AIT 4.1 ML 26
AIB 7.2 CL 500
Given: A plated through hole printed wiring assembly having 6 circuit

planes and 700 PTHs is installed in an airborne, uninhabited
fighter nonoperating environment.

be Ap 2 Anbx Npth X "NE
Step 2: Refer to Table 5.2.13-1 and select Xnb of .0000028 for multilayer -
soldered printed wiring assemblies.
Step 3: From the example statement, Npth = 700
Step 4: Refer to Table 5.2.13-2 and select "NE of 25 for airborne, L :.

uninhabited fighter.
Step 5. X= nb x Npth x "NE
.0000028 x 700 x 25
A-68
MIL-HDBK-217
CONNECTIONS
5.2.14 Connections
This section presents the nonoperating failure rate model for

connections including interconnection assemblies without plated through
holes. The nonoperating failure rate of the structure which supports the
connections and parts should be considered zero.
The general model for connections is as follows:
xp =T (Ni Xnbi)
where
xp = predicted connections nonoperating failure rate

SNE = nonoperating environmental factor (See Table 5.2.14-2)
Ni = number of connections of the ith type
/I' Xnbi = nonoperating base failure rate of the ith type connection (See
Table 5.2.14-1)
TABLE 5.2.14-1: CONNECTIONS NONOPERATING BASE FAILURE,

RATE (Xnb)
Connection Type Xnbi _
Hand Solder .000089

Crimp .000013
Weld .0000017
Solderless Wrap .00000012
Wrapped and Soldered .0000048
Clip Termination .0000041
Reflow Solder .0000024
A-69
IA
' ~/
7 MIL-HDBK-217
CONNECTIONS
TABLE 5.2.14-2: CONNECTIONS NONOPERATING ENVIRONMENTAL

FACTOR (ONE)
Environment "NE Environment INE

GB 1 AIA 4.5 N
GF 2.1 AIF 6.8
GM 6.6 AUC 2.7
Mp 7.3 AUJT 5.4
NSB 3.5 AUB 6.8
NS 4.4 AUA 6.3
NU 8.9 AUF 8.6
NH 11 SF 1
NUU 12 MFF 6.6
ARW 14 MFA 9.0
AIC 2.3 USL 22
AIT 4.1 ML 23
AIB 5.0 CL 420
i
Given: A srnldr~rless wrap discrete wiring assembly is to be used in an

inhabited cargo airborne nonoperating envlronmert. The assembly
consists of 1560 wraps and 156 of the posts are connected to
either the ground or voltage planes with reflow solder.
Step 1: The expanded nonoperating failure rate expression from Section
5.2.14 is:
p= INE (N1)nbl + N24,b2)
Step 2: From the example statement, N1 = 1560 wraps and N2 156 reflow
solder connections.
A-70
44
)~
MIL-HDBK-217
CONNECTIONS
Step 3: Refer to Table 5.2.14-1 and select a )tnbl value of .00000012 for
solder wrap and .0000024 for reflow solder.

Step 4: Refer to Table 5.2.14-2 and select a wNE value of 2.3 for AIC t
Step 5: ,p = wNE (NIknbl + N2nb2)

- 2.3((1560 X .00000012) + (156 X .0000024))
= .00129 failures/l0 6 nonoperating hours •
-V
A-71
MIL-HDBK-217
MISCELLANEOUS PARTS
5.2.15 Miscellaneous Parts
This section includes average nonoperating failure rates for

miscellaneous parts.
Table 5.2.15 includes average nonoperating failure rates for

miscellaneous parts.
TABLE 5.2.15-1: NONOPERATING FAILURE RATES FOR MISCELLANEOUS PARTS

Part Type Specification Nonoperati ng

Failure Rate
Vibrators MIL-V-95 3.3

Quartz Crystals MIL-C-3098 .039
Fuses .0014
Lamps
Neon .029
Incandescent .11
Fiber Optic Cables per fiber km .014
(single fiber types only)
Single Fiber Optic Connectors .014
Meters MIL-M-10304 1.4
Circuit Breakders .29
Microwave Elements (coaxial
and waveguide)
Attenuators
I Fixed Elements (direc-
tional couplers, fixed
studs and cavities) negligible
Variable Elements (tuned
stubs and tuned cavities) .014
Microwave Ferrite Devices .043
SDummy Loads .011
Terminations (thin or thick
film loads used in strip-
line and thin film circuits) .010
)P
* The nonoperating failure rate of attenuators should be calculated
the same as for Style RD resistors in Section 5.2.6.
A-72
• i .
* v'.-~r.~ z r'Of
Rowx~ir~vel~v~w- ine
PAVC ptin6 and executez 48i eaAch, de'L ~~tes- an.
hetec~tad acqiqZt:a on n
iastm a Cobt,?ma4 Cd rr
o6cv
Conmwtictic.on6 and int;b (&i
1~zc uies't Tt<-- .
and en@.Ue.cttq~vtt
rgt& ax't&.. ng1.:~tcZ
zcpci1
S iz~Lp'vtided to 621)l P-toq':,- Oý.;'LCC
c (P0,i) a;M atzz7,t EST,
etzmentA. The p'utnc44e. teckŽ&tcaZnt~$c4 -s A~te
4-s uet~~c.Utianc o4d

goIwta4on r'f~: f-
4 x A~~pflOLL z arc .eteitýon zc O
phrp~eas ext cit nt cA'&$t IVtttNc
ctd~ttn
i~limfat =..

Radc TR 85 91 - May1985

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Radc TR 85 91 - May1985

Uploaded by

Copyright:

Available Formats

Downloaded from http://www.everyspec.

IMPACT OF NONOPERATING PERIODS ON

DarEd W. Colt end Mary 0. Prior

:: ROME AIR DEVELOPMENT CENTER

U. S. TUTHILL, Colonel, US"?

FOR THE CO*O(1MER: 0I

If your address has changed or if You.

Do not return copies 'ofthis report

REPORT DOCUMENTATION PAGE

Rome Air Develoement Center RBE-2 F30602-83-C-0056

11I. TI TLELflneivad. Sfmuhty CIlaIfleat.io 20F 23 29

uN~CLAaso~sIFEUN.imITvoE3 SAME AS RQPT.(3 OTIC USERS E3 UNCLASSIFIED

17. Cosati Codes (Continued)

"SCOT LI.CTO 0 omPa

The principalinvestigator for this project was David W. Coit.

The objective of this study was to develop a procedure to predict the

reliability. A series of nonoperating failure rate prediction models were

•" The proposed nonoperating failure rate prediction methodology is

Observed nonoperating failure rate data were collected from a variety

o Data available to the part level

A summary of the collected data is presented in Table MS-1.

* The model development approach was based primarily on empirical data

,•" :" "'• "• .""3'_• i •" ." • r •i

A clear and concise procedure was

It is recommended that the proposed

TABLE MS-i: NONOPERATING FAILURE RATE DATA SUM4ARY

Model Development Model Evaluation

Data Failures Part Data Failures Part

IMonolithic Micro- 50 1947

324 48 41841 22 14 5004

Misc. Parts 8 5 129 0 0 0

,-,I. . ,-, \V.( .' . •

S• • -" / ' • • " \ " t - - '" "- - • - - '

I C.J udo x 06"- O3r5- U

> - > - >, -

=I- M.0 03-4 34 034)

uI. fa>, >03 Ct u

4W- ~ 4-J %C0-

C._.4-- M 4Sa in 9 o 6j4

"- 0 0 to(v X>%-CiiS '

431 0)1 oin w W1

owr.U csO ()• l~.U)C%•4•

w 0 oJx C J3 C 4J(U0 S-.3 w W4J 4-

0J-- 4JliC4- S W>% 01. >j

1.0 INTRODUCTION ................................................. 1-1

TABLE OF CONTENTS (CO)NT'D)

5.3.1 Proposed Resistor Nonoperating Failure Rate... 5-125

TABLE OF CONTENTS (CONT'D)

TABLE OF CONTENTS (CONT-D)

TABLE 5.2.3-6: DIODE NONOPERATING QUALITY FACTORS ................... 5-111

TABLE 5.3.1-1: RESISTOR NONOPERATING ENVIRONMENTAL FACTORS ('TNE),,, 5-126

TABLE 5.8.3-2: SWITCH PART CHARACTERIZATION ........................ 5-218

The objective of this study was to develop a procedure to predict the

Th? proposed nonoperating failure rate prediction methodology was

nonoperating reliability. An analytical approach using observed data was

Failure rate and reliability prediction capabilities are essential

analyses, reliability growth monitoring and life-cycle cost studies. This

Manufacturers and government customers for missiles have long

Previous attempts (References 3, 4, 5 and 6) to predict nonoperating

applied for any equipment in any environment, neither of these previous

Use of a multiplicative "K" factor has merit under certain

Also, the detrimental effects of equipment power on-off cycling would be

Many terms have been used during previous studies of nonoperating