Photonic Integrated Circuits for Passive Optical Networks: Outlook and Case Study of Integrated Quasi-Coherent Receiver

Abstract

Photonic Integrated Circuits (PICs) are taking a major role in the telecommunications and datacenter markets. The increased complexity of coexisting and fast evolving standards for Passive Optical Networks (PONs) suggests the introduction of PICs will be the next step in PON related optoelectronics. The PICs ecosystem has greatly matured in the past years, becoming a solution that can cope with the requirements of industry and academia, and presenting the flexibility of combining multiple platforms available towards viable commercial solutions. In this review, the evolution of PONs and PICs is presented, with a focus on the optoelectronic integration of PICs for PONs and coherent PONs. To demonstrate the potential of PICs and their combination with electronics, a quasi-coherent receiver based on co-hosted PIC and Application Specific Integrated Circuit (ASIC) is presented and characterized.

1. Introduction

An ever-increasing demand for bandwidth has led to the evolution of fiber optic networks, with Passive Optical Networks (PONs) being the ultimate representation of a complex structure that connects the whole world. This evolution is not only in terms of bandwidth but also in latency and power consumption of the networks, two pillars that the standardization bodies take into consideration when developing the next generation of networks. To keep up with the evolution in a sustained and sustainable manner, electro-optical hardware needs to move into a stage of higher integration.

With this vision, the authors present this paper to provide an overview of PON, and their electro-optical hardware evolution, with an integrated case study. In Section 2, a first sub-section with a thorough description of the PONs’ evolution roadmap is presented, followed by a sub-section with the options and feasibility of electro-optical hardware implementation for PONs, including the most recent advances for 50 Gbps PONs and different possible visions of implementation, such as coherent PONs. In Section 3, the genesis of photonic integration up to the present day’s generic availability is presented, followed by a sub-section dedicated to a review of the photonic integration possibilities for PONs. Section 4 presents a case study of an integrated quasi-coherent receiver, co-packaged with an application specific integrated circuit (ASIC), including the description of its assembly and the preliminary characterizations performed on it to demonstrate its applicability to PONs. Section 5 presents the take-away conclusions of the paper.

2. Passive Optical Networks: Roadmap and Electro-Optical Implementations

2.1. Passive Optical Networks Roadmap Review

PONs were developed to bring, in a cost-effective manner, high speed and reliable communications to a massive number of users. Since the 1990s, PONs have been deployed for various applications stemming from Fiber to the Home (FTTH) to mobile (Fiber to the Cell-FTTC) technologies being used as the backbone for most of the current day to day communications [1]. PONs comprise a point to multi point (P2MP) topology connecting to a central office (CO), where the optical line terminal (OLT) connects multiple optical network units/terminals (ONUs/ONTs), also known as customer premise equipment (CPEs), which can be installed at various points of the network. To guarantee P2MP, access to the medium is via Time Division Multiplexing (TDM), i.e., in the Downstream (DS) direction (from the OLT to the multiple ONUs) there is a constant broadcast and in the reverse, Upstream (US) communication only occurs in bursts, in defined time intervals. Figure 1 depicts the standard PON topology; on the left-hand side is the OLT, that is connected to multiples ONUs through the Optical Distribution Network (ODN), which is composed only of passive elements (coexistence element (CEx), fiber, splitters), establishing the reach of the PON [2]. The CEx multiplexes the multiple technologies deployed over the same PON, which is then routed to multiple case scenarios (residential, business, or wireless back haul/mid haul).

To deploy a worldwide technology that runs regardless of the various operators sharing the fiber infrastructure, standardization is key. Three industry organizations lead the efforts on PON standardization: the Full Service Access Network (FSAN) Group, the ITU Telecommunication Standardization Sector (ITU-T) Question 2/Study Group 15 (Q2/15), and the IEEE 802.3 Ethernet Working Group [1,3]. In 2004, through FSAN and Q2/15, Gigabit PON (GPON) was standardized [4]. Almost two decades after being standardized, GPON has a global footprint and meets most of the current market’s demand. Nevertheless, as the process of standardization and first field trials required to qualify a new technology can take up to 10 years, the different standardization bodies kept on defining new technologies, so that the increasing bandwidth demand coming from the continued evolution of the internet, and also its ever-increasing number of applications, can be met. The standardization bodies worked to, in 2009, release 10G-EPON (IEEE, 8022.3av Clause 75–77) and in 2010 XG-PON (ITU Q2/15, G987.x), paving the way for 10 Gbps PON [1]. With GPON widely deployed, operators were educated about the need to increase the bandwidth in order to offer better and more services, and to keep at the forefront of the technology in the technological race, however, these upgrades need to be cost-effective, and the technologies around 2010 were not mature enough to support low-cost 10 Gbps transmitters and receivers. While efforts were made to push 10 Gbps hardware technologies to a mature place, ITU standardized XGS-PON (G9807.x) in 2016, and NG-PON2 (G989.x) in 2015 [1]. The latter comprised the biggest technological leap, offering not only TDM but also Wavelength Division Multiplexing (WDM), and was the first PON TWDM standardized, offering up to 8 wavelengths with 10 Gbps per wavelength, allowing an aggregate 80 Gbps in each direction.

With the first deployments of 10G PONs in the Asian markets now in place, PON standards are now committed to identify the next trend demanded by the increase of traffic and use of devices, where by 2022 it was expected that data traffic will exceed 4 exabytes/month, with a predicted 28.5 billion networked devices and the number of Global Internet Users predicted to reach 5.3 billion by 2023 [5,6]. In parallel, 10 Gbps per wavelength networks, with a reach of 50 km and up to 16 wavelengths, are under standardization in both bodies, with the name of “Super-PON”. The efforts to standardize this Super-PON are being led by Google fiber. The latest effort of ITU to standardize PONs with line rates above 10 Gbps is called Higher-Speed PON (G.9804.x), that was consented to in November 2019 [7]. This recommendation works as a baseline for standardization beyond 10 Gbps PON, ensuring coexistence with previous technologies and providing a common transmission convergence layer. The first Physical Media Dependent (PMD) layer to be consented to was 50G PON, in April 2021 [8,9], whose first deployments are expected to occur during the end of the 2020s and peak at high volume by the mid-2030s [10,11].

Figure 2 summarizes, the changing PON standards over time, and the evolution of task forces with respect to the maximum aggregated DS bandwidth, depicting the different generations correspondent to the different technological steps between the different family standards.

2.2. Passive Optical Networks and Their Electro-Optical Hardware

Given the TDM characteristic of PONs, ONUs are equipped with a burst mode transmitter and a continuous mode receiver, whereas on the OLT a continuous transmitter and a burst mode receiver operate. Burst mode poses a challenge in any PON deployment, on the one hand the ONU transmitter needs to be fast and accurately controlled so that no time overlap with other ONUs occurs, on the other hand, the OLT’s dynamic range of the receiver poses a challenge as two consecutive messages can come from different distances, thus with different attenuation and power levels. A typical electro-optic transceiver for a PON is composed of a bidirectional optical sub-assembly (BOSA) and its driving electronics, packaged into a standard form factor in accordance with bitrate and power consumption targets. Inside the BOSA, US and DS band splitters are used to separate/combine the signals; on the transmitter side a laser is packaged, while on the receiver a photodiode with a transimpedance amplifier (TIA) is the usual assembly configuration. Concerning the typical driving circuitry, the key building blocks are a laser driver, commercially available as continuous or burst mode, a limiting amplifier (LIM), which can include Clock Digital Recovery (CDR), and a microprocessor to control the stated peripherals and provide monitor parameters to the host equipment. Following this baseline, enhancements and an increase in complexity may need to occur according to the PON technology under implementation, as will be discussed in the subsequent paragraphs.

GPON offers asymmetric bitrates of 2.5/1.25 Gbps in the DS/US direction. Where the upstream burst mode 1.25 Gbps communication is at the coarse (+/− 30 nm) 1310 nm O-band, corresponding to the minimum chromatic dispersion of standard single mode fiber (SSMF), allowing for low-cost deployment [12]. The 2.5 Gbps laser is at the 1490 nm C-Band and the power budget was extended to be as high as 39 dB with the definition of the C+ class in 2008 [4]. Today, the GPON ecosystem is fully established and, through deployment, prices have fallen to mass-production prices of below $5 for ONU BOSA and below $30 for C+ OLT optics [13].

Concerning 10G PON technologies, at the physical level, 10GEPON and XGS-PON have minimum differences. The implementation that poses higher technological challenges is the symmetric 10 Gbps/10 Gbps. While the US transmission is at the coarse (+/− 10 nm) 1270 nm O-Band, DS is at C-Band 1577 nm (+/− 3 nm) [1]. On the receiver side, no major technological challenges are related to the implementation of these technologies, the photodiodes in use are Avalanche Photodiodes (APDs), that evolved from the ones used in GPON, and the requirements of the burst mode receiver on the OLT are for an improved TIA with ns-scale settling time to synchronize 10 Gbps data. Similar to GPON, 10G EPON and XGS-PON take advantage of reduced chromatic dispersion in the US direction at O-Band. If this were not the case, chromatic dispersion would be exacerbated by the increase of bitrate [12]. The low dispersion region associated with +/− 10 nm wandering, allowed the implementations to re-use GPON laser structures with enhanced bandwidth, i.e., low-cost direct modulated lasers (DML), typically implemented through Distributed Feedback Lasers (DFBs). For the downstream, extra complexity is added at the transmitter level: high output power (+4 to +7 dBm for N2 class) associated with 10 Gbps, transmission at C-Band, and a reach of 20 km poses a roadblock to the use of DMLs. The use of an external modulator such as Electro Absorption Modulator (EAM), mitigates the chirp characteristics at the expense of extra hardware, extra packaging steps, and reduced output power. However, the latter can be overcome with a decrease in the laser yield or the introduction of an integrated optical amplifier. The control of the laser with its external modulator, the +/− 3 nm stabilization window associated with the need to maintain the operation stable for a temperature range (industry standards demand minimum 0 to 70 °C operation), leads to the use of thermoelectric cooler (TEC) elements to package the transmitter. One can then observe a clear complexity increase in the OLT transmitter: 10 Gbps C-Band communication leads to the use of EAM on a cooled package, which also leads to the correspondent need for extra electronic TEC control. Fortunately, this is on the OLT side where there is a higher cost-resilience. XGSPON is under rapid deployment worldwide and there are multiple vendors offering commercial solutions that are combinations of what was described above, e.g., SFP+ transceiver for N2 class OLT, based on cooled EML and high sensitivity APD, or BOSA for ONU with uncooled DML and high sensitivity APD.

Adding tunability to a PON standard means adding a full new set of capabilities that the operators can explore to optimize their capital expenditures on the network. NG-PON2, through its multi-wavelength, multi-rate capabilities allows bandwidth flexibility assignment; “pay as you grow” model, i.e., the network can start its deployment with one wavelength and, through time as the customer numbers increase, the operator can then add other wavelengths without having to change OLT cards or ONU equipment; resilient network maintenance, where the clients that sit on a wavelength can be seamlessly migrated to another one while the former goes into maintenance, and channel bonding allowing several channels to be directed to a single CPE, increasing the bandwidth up to 40 or 80 Gbps [14,15,16]. Coexistence with other PON standards is maintained, meaning that the tunability cannot be traded-off by the power budget or time division multiplexing characteristics. It is then possible to conclude that NG-PON2 electro-optical hardware represents an increased step in complexity. Another concern that arises with tunability is to guarantee that at each moment the transmitter and receiver are tuned to the right channel, and that no crosstalk between channels exists, meaning that from the +/− 30 nm wavelength coarse in GPON that moved to +/− 10 nm in XGS-PON and 10GEPON, NG-PON2 specifies a 100 GHz, i.e., ~0.8 nm channel spacing, where the transmitter needs to ensure a maximum spectral excursion of +/− 20 GHz, i.e., approximately +/− 0.16 nm [17].

For NG-PON2 OLT electro-optical hardware, the major challenge, when compared to XGS-PON or 10GEPON technologies, is to ensure stable wavelength operation of the C-Band laser through time, as the wandering window is shorter. The rationale behind the hardware choice follows the same line of thought as XGS-PON and 10G-EPON, with typical implementations using EAM as the external modulator and TEC to ensure wavelength stability. The receiver side focuses on the burst mode 10 Gbps receiver with fast (ns-scale) settling time TIAs.

For NG-PON2 ONUs, contrary to the other technologies presented previously, where the wavelength plan established was C-band wavelengths, also it specifies an output power varying from +2~+7 dBm for Type B and +4~+9 dBm for Type A ONU with operators looking only to Type A ONU to be used, meaning tunable transmitters with a high output power and reduced or managed chirp. Beyond these characteristics, NG-PON2 specifies power spectral densities Out-of-Channel (OOC), Out-of-Band (OOB), and When-not-Enabled (WNE), to ensure that in a populated PON with multiple technologies no crosstalk occurs at the OLT receivers. Tunability is defined in classes according to the maximum tuning time between any channel communication and the main classes establish a maximum of 1 s (Class 3) or a maximum of 25 ms (Class 2). The receiver also needs to be tunable, while maintaining a high sensitivity performance to ensure a PON typical link budget (−28 dBm for 10E-3 Bit Error Rate (BER) reference level). For NG-PON2 to succeed and supersede other 10 Gbps technologies, the new technological challenges cannot mean a proportional increase in the ONU cost, implying that low-cost tunable electro-optical hardware need to become available. From the transmitter side, the first challenge concerns successful transmission on 20 km SSMF of 10 Gbps signals, which at C-Band is almost immediately translated into an external modulated laser (EML) requirement due to the chromatic dispersion of SSMF on that region, associated with the typical chirp of C-Band 10 Gbps lasers. The wavelength stability and tunability pose the other major challenges of NG-PON2. A solution would be to package multiple lasers with different wavelengths and electronically control their switching, however that poses a problem due to the footprint and cost increase that ONU PON equipment cannot cope with. The simplest alternative for tunability is thermal tuning, where the laser cavity temperature is controlled, and thus its wavelength of operation. For NG-PON2 requirements previously presented this means a high precision control, to ensure that the laser is stable within the +/− 20 GHz window. Added to this, there is the challenge of burst mode transmission where, as the laser is turned on and off, its average temperature, and thus its wavelength, changes which, with the tight spectral characteristics of NG-PON2, becomes a problem that needs to be controlled. In summary, for an NG-PON2 transmitter, and for a successful deployment on the transmitter side, 10 Gbps C Band lasers, preferably directly modulated, that operate in burst mode in the tight window of +/− 20 GHz and with a considerable high output power, need to be packaged (with its thermo-electric circuitry) and developed at the right price point. Several players have been engaged in this endeavor: Finisar have presented a Distributed Bragg Reflector (DBR) laser that could be directly modulated due to an innovative technique of chirp suppression [18]; Nokia have developed a multi-electrode laser that allows the laser to operate in burst mode within tight spectral conditions [19]; and PICadvanced have developed algorithms for the driving and temperature control for burst mode operation of a 10 Gbps C-Band DFB laser. On the receiver side, the straightforward way to achieve tunability is to add a temperature-controlled filter, such as a thin-film-filter (TFF), which has a cyclic spectral response. The filter design, typically a thin-film-filter, poses a challenge as it needs to have a low insertion loss, a flat spectral response within the passband, fast roll-off between channels, and a small foot-print. On top of the tight filter design conditions, its packaging needs to be optimized so that there is proper thermal efficiency with the TEC for an efficient tuning and minimum free space propagation insertion losses in the path to the high-sensitivity photodiode. NG-PON2 presents major advantages from a network perspective brought by tunability coming at the expense of tight requirements on the electro-optic hardware, which is highly sensitive to cost (e.g., 10GEPON and XGS-PON ONU have no need for thermal control, whereas on the NG-PON2 two thermal controllers are needed). At a commercial level, one can find in PICadvanced portfolio compliant NG-PON2 solutions for Type A ONU Class 3 and Class 2 tuning times that are based on a cooled transmitter based in DML and a cooled receiver based on APD.

The efforts of PON standardization have been focused on the delivery of bandwidth and quality of signal to new applications and case studies that are under development for the decades to come, while maintaining coexistence with previous technologies and looking for the best trade-offs between the target performance and electro-optical implementation. This is the train of thought being implemented in High Speed PON (HSP) standardization, and in particular in its first, PMD standardized, 50G-PON. Taking advantage of the reduced dispersion window of fiber, both DS and US are at O-Band. For the DS the only line rate that was standardized was 50 Gbps, with two power classes, N1 and C+, whose minimum mean launch power are +5.5 dBm and +8.5 dBm, respectively. The link budget, i.e., the balance between the power launched into the fiber and the sensitivity at the ONU side, is defined as a function of Transmitter and Dispersion Eye Closure (TDEC), a figure of merit defined for this standard that is computed using histograms of the eye diagram for the worst case scenario of transmission, allowing flexibility of electro-optical hardware implementation while maintaining the target link budget. While for DS the wavelength of operation is 1340–1344 nm, for US there are two possible windows of implementation: 1260–1280 nm or 1290–1310 nm, each of them overlapping with legacy technologies, the former with XGS-PON and the latter with GPON. This choice of spectra forces the operators to decide if they will go directly from GPON to 50G-PON, skipping XGS-PON, or removing GPON (or XGS-PON but less likely) from the network in case they already made the deployment of both legacy technologies. This conflict is being addressed as a possible amendment to the standard, allowing full coexistence and placing the window of operation on 1284–1288 nm. For the upstream line rates, 12.5 Gbps and 25 Gbps are already standardized with the bit error reference level for both at 1E-2 and required sensitivity of −26 and −24 dBm at the OLT side, respectively. The mean launch power is defined between +5 and +9 dBm for both line rates, with the extinction ratio defined at 6 dB for the ONU transmitters and 7 dB for the OLT. To aid the reader,

Table 1. PON technology optoelectronic key performance indicators.

Standard		Upstream					Downstream
		Wavelength (nm)	Max. Bit Rate (Gbps)	Max. Output Power (dBm)	Sensitivity (dBm)	Commercial Availability	Wavelength (nm)	Max. Bit Rate ( Gbps)	Max. Output Power (dBm)	Sensitivity (dBm)	Commercial Availability
GPON	Regular	1290–1330	1.244	4 C+	−28 BER 1E-10	Mass	1480–1500	2.488	7 C+	−32 BER 1E-10	Mass
	Narrow	1300–1320
XGSPON		1260–1280	9.953	9 N2	−28 BER 10E-3	Mass	1575–1580	9.953	9 N2	−28 BER 10E-3	Mass
NGPON2		1532.68	9.953	9 Type A	−28 BER 10E-3	Ramp up	1596.34	9.953	7 N2	−28 BER 10E-3	Ramp up
		1533.47					1597.19	9.953
		1534.25					1598.04	9.953
		1535.04					1598.89	9.953
50G	UW1–Wide	1290–1310	49.764	9 N1	−24 BER 1E-2	Lab trials	1340–1344	24.883	5.5 N1	−24.5 BER 1E-2	Lab trials
	UW2–Wide	1260–1280

summarizes the key performance indicators of the different standardized PON technologies.

The evolution of electro-optical hardware is dependent on the evolution of PON standards and its related technologies. The authors of [20] identify two paradigms on PON hardware development, one which is called “old paradigm” where the hardware was being pushed by long-haul and metro applications and the “new paradigm” where for 25 Gbps and 50 Gbps per channel the hardware comes from mature Data Center applications.

As of 2022 we are in the turning point between the old and new paradigm except for Super-PON where the 10 Gbps trend is maintained with the technological enhancement of extended fiber reach and higher number of WDM channels under use, demanding proper and careful driving whose electro-optic possible options are presented in [21]. For Higher Speed PON, the new paradigm is applied and adaption of already existing 25 Gbps and 50 Gbps can be done, being the major challenges on differentiation from the Data Center applications the burst mode requisites demanded by the TDM characteristics of PON. While the major guidelines for optical requirements for the new standards were being defined, and to aid with standard development, electro-optical hardware has been developed and proposed for the next PON generations by pursuing different technological approaches, one example proposes the use of low bandwidth optics (10 Gbps class optics) to implement 25 Gbps and 50 Gbps per channel communications through the use of multi-amplitude modulation format and equalization techniques, such as feed-forward or decision-forward equalization [22]. From Nokia, in order to solve the burst mode problem at 25 Gbps, the multi-electrode laser presented to operate as DML for NG-PON2, was also tested as EML for 25 Gbps demonstrating tight spectral conditions at the burst transients and during burst windows [19]. On the receiver side, for burst mode receiver electronics and techniques, IBM and Ghent University presented a burst mode adaptative equalization for 25 Gbps operation [23]. A thorough literature analysis for next generation PON hardware was provided in the tutorial conference in the 2020 Optical Fiber Communication Conference (OFC) [24]. In the session, the author addressed the next generation PON implementation challenges from a network and component perspective, including, but not limited to, burst mode challenges and possible implementations for transmitters and receivers at 25 Gbps, demonstrating the ecosystem preparation for the next PON generations.

While the standard was taking its final shape, multiple group studies worked on the feasibility of implementing 25 Gbps upstream and 50 Gbps downstream using bandwidth limited hardware (i.e., 25 Gbps capable hardware) to do so; the introduction of Digital Signal Processing (DSP) would be mandatory for the standard. These studies also considered multiple amplitude level in comparison to the NRZ standard option [25,26]. In an effort to demonstrate the trade-offs of having 25 Gbps over the C-Band, where the attenuation is lower but dispersion is higher, compared to the O-Band, as well as to test burst mode transmitters and receivers to emulate upstream links at 25 Gbps, simulations and experimental demonstrations were performed to help guide the standard [25,27]. With the publication of the standard defining 50 Gbps on the DS, it became clear for the opto-electronic researcher that the major effort would be on the 50 Gbps transmitter of the OLT, whereas on the receiver side of the ONU, 25 Gbps APD were already commercially available at the time of the standard’s publication, and the implementation of DSP would overcome the limit of bandwidth of the referred devices. Nevertheless, efforts to develop advanced optimized structures for 50 Gbps are being pursued, for example with Ge/Si APD compatibles with existing CMOS processes to ensure the required high volume of this application [28]. On the 50 Gbps transmitter side, to achieve an NRZ modulation, bandwidth in excess of 37.5 GHz is required which, given the commercially available devices and fabrication processes, requires EML based on EAM with an O-Band DFB. The trade-off of PON budget, power consumption, and bandwidth requires the development of state-of-the-art structures and techniques to bring 50 Gbps communications into PON. A thorough review of the options for EML implementation for multiple telecommunication applications is provided in [29]. In this review, the authors address means to achieve high power 10 Gbps EML for XGS-PON OLT and to evolve next generation lasers to cope with the next generation of standards, from the fabrication techniques to the packaging of the laser through flip-chip and self-alignment. The ultimate structure for HSP is an EML structure composed of a DFB, and an EAM and an SOA, which, by using the latter in a saturation region, can deliver the required high output power and help to compensate for chirp effects, as was presented in [30]. There are research groups that are focused on taking the lead on optoelectronic development of 50 Gbps PON, by the end of 2020 the first demonstration of a 50 Gbps with 35.6 dB of link budget over 20 km was performed by using 25G class EML co-fabricated with an SOA and 25G APD. To achieve 50 Gbps performance, DSP was used for both transmitter and receiver [31]. Subsequently, one year later, also using 25 Gbps hardware aided by DSP, a carrier lab trial was performed demonstrating 40 Gbps capacity with 80µs latency over a 10 km SSMF PON link [32]. Through a simulation of network worst case scenarios, the researchers were able to optimize their demonstration setups and optoelectronic hardware, achieving an E2 class link with 35.6 dB real time link budget over 20 km of SSMF [33]. Later, through further optimization, a laser structure comprising a DFB + EAM + SOA was presented with a bandwidth of 37 GHz, allowing 50 Gbps without a DSP on the transmitter side, demonstrating a link up to 40 km with emitting output power of +13 dBm and 6 dB ER [34]. By using the same devices, in a PON with a common SOA that acts as pre-amplifier at OLT, a 30 dB budget link with 50 km SSMF with burst mode was shown [35].

In another stream of thought, and to achieve a high link budget, hence the desired sensitivity at the receiver side, a possible solution presented to achieve extended performance in future PONs is to use a coherent receiver, also known as a coherent PON, which refers to the PON networks that use coherent receivers, i.e., architecture of receiver that not only is able to recover amplitude information but also phase with the aid of a reference local oscillator. The implementation of optoelectronic circuitry based on coherent mechanisms brings major advantages to the network such as high-sensitivity bringing increased power budget to the PON, possible higher bit rates (through the use of phase and amplitude modulation schemes), increased spectral efficiency (as the tuning is done with a reference local oscillator), and opening of the C-Band for communication, as dispersion can be mitigated through DSP techniques. All these main advantages come at the cost of extra hardware complexity, as to implement coherent reception implies the use of an extra laser source in the receiver as well as increases in the number of photodiodes needed when compared with a single-photodiode for the typical On-Of-Keying (OOK) reception. The topic of coherent PONs has been revised in terms of architectures and hardware implementation through time, even recently with 50 G-PON simulation and experimental models have been presented to achieve the desired data rate and link budget using coherent receiver techniques [36,37]. The concept of ultra-dense WDM-PON (udWDM-PON) arose as an alternative/complementary solution for TDM-PON, as with coherent reception, channel selectivity is intrinsic and a dedicated channel could be attributed to a single ONU or set of ONUs. This type of PON was widely reviewed in [38], where the authors present the possible topologies for udWDM-PON. and implement an 8 × 2.5 Gbps QPSK system over 80 km of SSMF with off-the-shelf components, showing the commercial availability for this type of networks. In [39] the authors provided a full set of possible coherent PON architectures and looked at lowering the cost of implementation of coherent access with the use of DFB (low cost, higher linewidth) lasers as local oscillators. Albeit there is proof that coherent PON can be built with off-the-shelf components, cost and hardware complexity should be of the same order of magnitude when compared with existing PONs under deployment, such as NG-PON2, where the tunability could be implemented through a coherent receiver rather than with the use of thermally tuned filters as addressed before. In order to maintain the right level of cost and complexity, the coherent receiver should use a low-cost laser as the local oscillator, implement the minimum number of photodiodes to achieve phase diversity, should not use DSP (due to power and cost consumption), and should guarantee polarization insensitive operation–coherent receivers are polarization sensitive, which in many applications is used to increase the bitrate, nevertheless in PON the systems are single polarization.

The FP-7 project COCONUT [40] explored and developed low cost implementation of udWDM-PONs showing different topologies for PON implementation and a polarization independent receiver for OOK receivers [41,42,43]. The results from the project were further exploited to develop low complexity/low-cost reduced digital transceivers for udWDM-PON enabling the possibility to introduce, in a cost-effective manner, DSP into PON [44]. Besides udWDM-PON which takes better advantage of the available spectrum, there are other topologies and their electro-optical hardware implementation under study to enable coherent PONs, for example analog and digital implementations for simplified OOK coherent hardware and suggestions of a network topology to use Dual-LO coherent receivers are presented by Lavery et al. in [45], with further detail on low-complexity coherent PON ONU implementation in [46]. A full experimental demonstration of 10 Gbps NG-PON2, with an extended power budget of 35 dB with 40 km, was conducted based on coherent reception through the development of analog integrated circuitry to process OOK signals, exhibiting the advantages of coherent reception on current PON technologies [47].

3. Photonic Integrated Circuits (PICs) Evolution and PON Applications

3.1. PIC from Concept to Generic Availability

The concept of photonic integration was introduced in the late 1960s, when different types of waveguides, simple lasers, modulators, and other basic building blocks manipulating light were designed to be fabricated with photolithographic processes that were already available from the integrated electronics industry [48]. The major goal would be to start composing photonic integrated circuits, envisioned as integrated interconnections between optical building blocks put together to implement defined functions–the opposite of discrete packaging where the different building blocks (lasers, couplers, photodiodes) were packaged separately through free space optics. The advantages of integration were already clearly observable at the time. Despite the breakthrough in the late 1960s, it took four decades for the first complex PIC to become commercially available, when Infinera launched, in 2005, a PIC based transceiver capable of 100 Gbps transmission and reception with 10 channels of 10 Gbps in a WDM scheme [49]. Infinera kept on introducing PIC based products to the market, and in parallel Data Center technologies under implementation from major companies such as Intel, Cisco, or Neophotonics were also based on photonic integration, allowing them to reach 400 Gbps [50,51,52].

The time gap between photonic integration and the first commercial introduction was mainly due to a lack of coordination of the investments required to promote and evolve photonic integration as an open platform, posing the risk of development to big companies without synergies developed among them leading to the development of different technologies for similar applications, making process transfer more difficult [53]. The way to mitigate this gap was through the introduction of a generic photonic integration path, paving the way to a common photonic development infrastructure and opening the access to the technology to more users, making it fabrication independent to the designers with the creation of process design kits (PDKs) for multiple available foundries [53,54]. The initial work on this path was to identify the two technology branches possible for photonic integration: InP-based, that supports the highest level of integration, offering the possibility to jointly integrate passive devices (filters, waveguides) with active elements (lasers, amplifiers); and Silicon Photonics, which comes at a lower entry cost as it offers the advantages of a mature CMOS process, but is so far not able to monolithically integrate lasers and amplifiers. These two main platforms were then joined by a third one, TriPlex, offering dielectric waveguide technology, which increased the performance of the waveguides and offered thermo-optic functions [53]. For each of the technologies identified, platform organizations were created opening the access to an increased number of users: JePPIX from TU Eindhoven for the InP based photonic integration [55]; ePIXfab for SOI-technology of IMEC [56]; and Lionix, a Dutch company, opening for TriPlex technology [57]. All platform integration started organizing Multi Project Wafer (MPW) runs, a concept migrated from microelectronics, allowing external users to the different fabs to share a part of a wafer with a dedicated design, reducing the R&D cost, technological risks, and helping the different fabs to improve their processes. These platforms also offered brokerage services allowing the creation of a Photonic Integration ecosystem in Europe. While these steps were taken in Europe in the late-2000s to early-2010s, the United States launched their own generic platform integration later on, with a major focus on Silicon Photonics development but also offering MPW through different fabs that until then were closed to open access. The project consortium, called AIM Photonics [58], has been evolving over the past years offering MPWs, education services, and brokerage services fostering further development on generic photonic integration.

Generic photonic integration allows the creation of a multitude of architectures through four basic categories of building blocks: Passive, Phase, Amplitude, and Polarisation, whose possible functionalities are depicted in Figure 3. Falling into these categories, a multitude of building blocks offered in the PDKs of the foundries can be offered, which are in constant evolution. As an example, in the technological roadmap of JePPIX for 2020, electro absorption modulators exceeding 56 Gbd transmission rates, and polarization splitters with more than 20 dB polarization extinction ratio, were predicted [59]. The same roadmap depicts the access of the users to the MPWs offered, showing an increased number of overall users with an increased trend of industry-based tape-outs, demonstrating that the generic photonic integration eases the way to start-ups and SMEs for prototyping and testing PIC-based devices, accelerating the time to market. The constant evolution in maturity of the different platforms, and the increasing number of users, is increasing the number of PICs fabricated and the density of the circuits, closing the gap between photonics and electronics, and approaching Moore’s law to photonics, as predicted in 2012 by Meint Smit [60].

The goal of generic photonic integration is to develop a platform that can jointly bring the gain offered by InP with the compactness and maturity of fabrication processes of Silicon Photonics. Over the past decades many steps were taken to reach this goal and the authors in [61] provided a review of the evolution of the generic photonic integration platform and defined the trends for its future. They focused on InP and Silicon Photonic co-integration with a highlight of the IMOS (InP membrane on Silicon) under development by TU Eindhoven, which already demonstrated wafer bonding level integration between an InP membrane and a BiCMOS wafer, which is done by transferring the InP membrane to a CMOS wafer by adhesive bonding with a BCB (benzocyclobutene, a robust polymer) [61,62].

The availability of a generic photonic integration platform is key for PIC development, but the packaging of the fabricated PICs plays a major role in complexity and cost. According to the JePPIX roadmap [59], assembly testing and packaging represent more than 80% of the cost of PIC-based products. Figure 4 compares the cost structure for the discrete and integrated implementation of a PIC-based transceiver with tunability capabilities. Packaging plays a major role on the success of PIC-based devices. This led to the creation at the European level of packaging pilot lines for the different PIC areas: PIX4life focus on CMOS processing compatible SiN photonics IC, and the establishment of a supply chain for visible range targeting biophotonic applications; MIRPHAB is dedicated to prototyping and production of sources and sensors for the mid-IR range and PIXAPP focus purely on establishing a supply chain for assembly and packaging from early stage to low-to-medium volumes [63,64,65,66].

PIC packaging comprises a multitude of challenges, from that for the optical interconnection, which is required to achieve minimum losses and high packaging throughput, to that for the electric, which requires robust packages, and without compromising the target bandwidth and thermal challenges to ensure thermal stability of the PIC and its devices. The problems are applicable across the different PIC platforms, and in [67] the authors assess the state of the art solutions for the different packaging challenges existing in Si Photonics. For example, practical studies on a laser-to-pic integration are provided, demonstrating the effort among different packaging groups on solving packaging challenges. A focus on thermal management is provided, as most of the components in a PIC have their performance tightly related to temperature, and a temperature stabilization for the different components is needed for the final module to be able to operate according to the specifications. The thermal challenges for ONU PIC packaging arise in different manners and are assessed through the development of a heat propagation theoretical model which was confirmed via simulation and experiments as reported in [68]. Following this train of thought, an electro-thermal circuit for generic PIC packaging with active cooling through a TEC, its theoretical model and analysis, are provided in [69]. Here the author makes conclusions about the relative role of the thickness of the heat spreader and the thermal resistance chain between the PIC and the ambient temperature. The same author later presents solutions for active cooling of the PIC based on microfluidics [70]. A solution where a holder platform is used for PIC packaging (electrical, optical, and thermal) is thermally analyzed concerning the holder material, thermal cross talk of the PIC active elements, and the epoxies used in the packaging bonding between the PIC and the holder [71].

3.2. An Outlook of PICs for PONs

The increase in complexity of PONs’ requirements increased the appetite for the PIC community to contribute on an academic and commercial level to the electro-optical interfaces, mainly due to the proliferation of WDM PON and the increase of PON bandwidth. In order to maintain the cost at the right point, photonic integration will be a key solution enabling the reduction of space floor occupied (planar chip vs. packaged components with free space optics in a BOSA format), reducing the number of optic/optic and optic/electric interfaces (i.e., reducing the number of packaging steps), and enabling mass-production fabrication with all-wafer processing, i.e., bringing the advantages of integration (either electronic or photonic) into the PON market.

Key lessons from the previous sections indicate that challenges related to electro-optical hardware for the current and next generation PONs comprise passive building blocks that are able to separate and combine upstream and downstream bands and wavelengths of the different chosen channels of a given PON technology, fast and tunable/multi wavelength transmitters. and receivers’ architectures that can operate in the integrated platform with a wide temperature range and tight wavelength control that may also need to be burst mode enabled. Through the past decade, many contributions from the PIC ecosystem to solve these problems arose, by focusing on a FlexPON approach, i.e., a PON architecture that is WDM and “on-demand” TDM with wavelength agnostic ONUs, an architecture of a multi-wavelength transmitter capable of supporting 8 wavelengths was proposed, fabricated, and tested, achieving 12.5 Gbps operation [72]. The laser architecture in this implementation was based on a DBR with a dedicated Mach Zehnder Modulator (MZM) per channel and the wavelength multiplexing was achieved through an Arrayed Waveguide Grating (AWG). In the same work, a WDM receiver architecture was demonstrated with wavelength de-multiplexing achieved with AWG and a dedicated photodiode per channel; both transmitter and receiver were fabricated through JePPIX MPW. Another example is the 16-channel transmitter PIC based on an array of DFB lasers combined with an AWG to attain wavelength multiplexing and amplification with a Semiconductor Optical Amplifier (SOA) that was designed and fabricated with a compact size of 4.25 × 4.5 mm; the design was oriented to WDM PON applications like NG-PON2 or Super-PON [73]. Fraunhofer HHI, one of the foundries available in the MPW of JePPIX, developed a polymer-based integration platform called Polyboard that allows passive optical functions and efficient coupling with InP active elements; through this platform several building blocks and PICs were developed and presented in [74]. From the contributions presented in [74], the author of this work would like to showcase the WDM-PON devices developed: a 40 channel 100 GHz spacing receiver based photodiode arrays combined with an AWG, and the 10 Gbps tunable C-Band ONU with a Dual Gain chip on InP and the polymer Bragg grating and band mux-demux achieved on the polymer platform. Moreover, a GPON OLT on chip, for extended link budget and supporting multiple channels on the transceiver, was presented in [75], where the authors developed passive low-loss polarization insensitive building blocks for band splitting, and on-chip power monitoring, demonstrating an economically viable path to introduce photonic integration into the widely deployed market of GPON through the implementation of a quad-transceiver in a single port of the OLT. Alternative WDM networks have also been presented. An example is the use of wavelength locking techniques that can simplify the ONU tunable hardware and improve the overall network capacity. In [76] the authors present PIC based ONU and OLT designs for the implementation of a full C-Band tunable WDM-PON with a fully integrated transmitter array at the OLT side, and a tunable DBR based transmitter at the ONU. Efficient on-chip optic amplification is key for PIC based PON transceiver amplification to succeed, to ensure low power consumption and the adequate PON link budget, a thorough study on the integration of SOA on SOI PIC based transmitters has been conducted in [77]. Polarization insensitive receiver architectures were presented for an ONU receiver, including the considerations needed for flip-chip integration of the SOA chip into the SOI PIC, on the transmitter side a WDM array of 32 transmitters working at 25 Gb/s per channel is presented. On the trend of monolithic integration, novel low polarization dependent over C- and L-Band SOA were fabricated and integrated with passive waveguides [78]. As a final example, a multi-wavelength transmitter and receiver for ONU and OLT of WDM-PON, where the transmitter is based on a DBR laser and a Mach Zehnder modulator and wavelength multiplexing achieved through AWG, was designed fabricated and demonstrated at 10 Gbps with a 1 mW coupled output power [79].

With NG-PON2 being the standardized TWDM technology, focus on research and investigation has occurred to develop PIC in order to bring NG-PON2 to a massive deployment with the use of photonic integration. Almost simultaneously with the standard’s approval, a proposal of a single PIC for ONU and OLT was presented, with the novelty of ONU and OLT interchangeability achieved through AWG match of Free Spectral Range and optical switching through SOA [80]. OKI, a Japanese system and component manufacturer, recently proposed a Silicon Photonics based PIC for NG-PON2 [81]. The transmitter is a co-packaged InP laser array whose wavelengths are multiplexed through an AWG and a single Mach Zehnder modulator is used. As the device is bidirectional, band-splitting is achieved through a polarization rotation Bragg grating and the receiver path comprises a polarization beam splitter followed on one path by a polarization rotator and on the other just an AWG. Outputs of both AWGs are then input to a photodiode array grown on the substrate, achieving in this manner the required polarization independency. The chip was fabricated and packaged, being 10 Gbps operation of transmitter and receiver demonstrated for the various channels. In [82], an array of four lasers with SOA for NG-PON2 was presented showing 80 mW output power for temperature operation from 15 °C to 45 °C, through a proprietary discrete mode technology. A thorough analysis of the methodologies and implementations to bring PICs into NG-PON2 was presented in the tutorial session of [83], where the full methodologies and ecosystem were revised and novel architectures for transmitter and receivers for NG-PON2 PIC were presented. An assessment of the implementation of ONU for NG-PON2 is provided in [84], where the trade-offs for monolithic implementation of NGPON2 ONU PIC based transceivers are assessed and compared against the different technological barriers to meet, and the standard requirements both at a network and component level. The authors present the comparison of their released NG-PON2 ONU based on discrete optics and their progress with PIC, where the receiver is based on a quasi-coherent architecture, which is shown in Figure 5.

Table 2. Summary of key parameters of PIC based NG-PON2 implementations.

Reference No.	Transceiver Type	Integration	Technology	Target Equipment
80	Tx	Monolithic	DBR array + SOA + AWG	ONU and OLT (interchangeable)
	Rx		AWG + array of PIN
81	Tx	Hybrid (InP + Si)	Laser array + AWG + single MZM	ONU
	Rx		AWG + array of PD per polarization	ONU
82	Tx	Monolithic	Laser array (80 mW output power)	ONU or OLT
84	Rx	Monolithic	Quasi-coherent	ONU

presents an overview of the NG-PON2 PIC based implementations featured in this section.

Coherent technologies for PON have been a focus of study as presented at the end of Section 2.2. Developments of PIC for coherent were focused so far on high-capacity applications like data centers, however with the change of paradigm to the “New paradigm” where the data centers’ electro-optic solutions are being transferred to PONs, and with a high motivation to develop PIC for PON, the development of low-cost integrated coherent architectures is arising.

A dual polarization phase diversity coherent receiver is composed of a polarization beam splitter (PBS) that receives the input signal and splits both polarizations that go through similar paths composed by a hybrid (many implementations are possible for a 90° hybrid, a review is provided in [85], and for 120° hybrid in [86]) that on the one hand inputs the signal and splits it in amplitude and phase at its output, and on the other hand inputs a reference local oscillator (LO) whose polarization is aligned with the one from the signal. At the hybrid outputs, a combination of the LO with the signal that presents a phase shift among them, allowing after photodetection, and with the right photocurrent combination, to recover both the phase and amplitude of the signal. An example of a dual polarization coherent receiver based on a 120° hybrid is depicted in Figure 6. This exemplary architecture requires six photodiodes instead of the eight photodiodes required for a 90º hybrid. While the 120° hybrid, considering the implementation of single ended photodiodes, can be considered easier to integrate on to a PIC based solution, the 90° hybrid is simpler in the case of use of balanced photodiodes. Additionally, the 120° hybrid is a more compact integrated device when compared to a 90° hybrid. For PON, with the current standardized technologies and standardization occurring, coherent is useful for WDM PONs where the LO can be used to tune the signal, it can also be useful to achieve a high sensitivity with PIN photodiodes. For the latter case, it is important to state that the integrated platforms do not provide APDs, whilst coherent can reach better sensitivities than generic APD based receivers due to the gain provided by the power of the LO. For PON based on OOK, simplified architectures should be developed providing polarization insensitive operation and only amplitude recovery.

Infinera have put a large amount of tine and effort into the development of coherent monolithically integrated PICs for datacenter applications, delivering state of the art, and the best industry rated, transceivers based on monolithic PICs. That same company recently thoroughly reviewed its production data and statistics, with the best practices for system on chip PIC based development over the past years, providing an example for academia and industries working in the field of photonic integration fabrication [87]. For their coherent transceiver development, in 2017 they reported a full C-Band tunable multi-channel PIC transmitter and coherent receiver capable of reaching up to 44 Gbaud 16-QAM dual polarization with an aggregated capacity higher than 1.2 Tbps [88,89]. Evolving from there, in 2019 the company presented the platform to reach a transmitter and coherent receiver of 100 Gbaud [90], and in 2020, using this platform, they demonstrated a 1.6 Tbps 2-channel transceiver based on a monolithic Tx/Rx InP PIC, co-packaged with a SiGe ASIC [91]. Other academic and industry players are working towards PIC based coherent receivers: Ericsson presented a 120° hybrid on a SOI PIC [92]; Sumitomo developed a very compact coherent receiver based on a 90° hybrid capable of reaching 224 Gbps with dual polarization 16-QAM [93]; and Fraunhofer HHI are actively working to improve their balanced photoreceiver, that reduces to two the number of photodiodes needed on a 90° hybrid, having demonstrated 100 Gbaud operation [94] and a best of class waveguide integrated balanced photodiode with a 3 dB bandwidth of 115 GHz [95]. Simplified architectures for coherent PONs were demonstrated at 10 Gbps and 25 Gbps in [96], where a 2 × 1 coupler, a PBS, and two photodiodes were used in conjunction with a dedicated ASIC per polarization, that worked as an envelope detector. The authors demonstrated up to −28 dBm sensitivity at 25 Gbps with a commercial TIA and the dedicated envelope detector [97]. Subsequently a TIA was designed and co-packaged with the envelope detector, showing −30.5 dBm sensitivity at 10 Gbps [98]. Finally a customized TIA has been developed and co-packed with the ASIC for envelope detection, this achieved a −28 dBm sensitivity at 10 Gbps [99]. In [84] a coherent receiver was implemented in a PIC platform and it was stated to be polarization independent.

4. Quasi-Coherent Receiver Based on Co-Hosted InP PIC and SiGe ASIC

With the aim of developing PIC based transceivers for PON applications, work has been carried out to develop transmitter and receiver architectures that can be implemented in an integrated manner and have a performance comparable to, or better than, current BOSA based implementations. With the focus on the receiver side, to achieve the desired sensitivity and the tunability required for WDM applications, coherent receivers are chosen. The authors present an electro-optical receiver that co-hosts an InP PIC and a Silicon Germanium (SiGe) ASIC, designed to match performance and implement coherent detection for IM, polarization independent applications [100]. By using a multi-pocket host of Silicon (Si), both ASIC and PIC are co-packaged, and the device characterization is performed. The setup and results are presented in the next subsections with an extended set of results.

4.1. Co-Hosted Device and Experimental Setup

Figure 7a shows a detailed image of the co-hosted InP PIC + SiGe ASIC mounted on a specially designed Printed Circuit Board (PCB). The bottom layer of the device is a TEC that is externally controlled using the Thorlabs TED200C(Thorlabs Inc., Newton, NJ, USA), to define the base temperature, either 20 or 35 °C. The TEC is attached to the PCB with thermal paste, where the PCB includes vias for thermal dissipation. Following that, a proprietary Si multi-pocket host is thermally and mechanically connected to the PCB using thermal epoxy H20E from EPO-TEK. The Si host contains cavities for the ASIC and PIC. It is geometrically designed to match the PIC and ASIC input/output interfaces and dimensions, while minimizing the distance between their high-speed interfaces. The host also contains electrical traces and pads for the various interconnections required between the integrated circuits and the PCB as well as V-grooves for fiber alignment. Wire-bonds between the PIC and the ASIC connect the high-speed PIN receiver output to the TIA input stage of the ASIC. The ASIC’s output is wire-bonded to the Si host RF lines, which provide a bandwidth of more than 30 GHz with an insertion loss smaller than 1 dB. The host RF lines are then wire-bonded to the PCB RF lines. The remaining wire-bonded DC electrical connections are performed from the PIC/ASIC to the Si host. Both PIC and ASIC are attached to the Si host via the aforementioned thermal epoxy.

The InP PIC was designed using the generic integration processes of Fraunhofer Heinrich-Hertz Institute (HHI) [101], featuring a simplified coherent architecture, capable of detecting the signal amplitude [102]. An LO, used in heterodyne regime, is mixed with the input signal and then separated in two PIN receivers with equal amplitude and orthogonal components to achieve polarization independence. The electrical output signals from the coherent receiver feed the ASIC, whose input stage is a high-speed linear TIA, followed by an envelope detection for down-conversion of the signals to baseband, the last stage that comprises the electrical combination of the two components.

Figure 7b shows the setup used to characterize the device. The green box containing the quasi-coherent receiver with two photodiodes, represents the PIC while the blue box contains the ASIC, enclosing the two TIAs and envelope detectors, whose outputs are combined and wire-bonded to the PCB’s differential output traces. The transmitter consists of an XFP form factor evaluation board where the XFP transceiver can be changed according to the laser modulation technique (DML or EML) and wavelength band. A modulated signal coming from the Bit Error Rate Tester (BERT) ExoBERT2904 with 2⁷-1 Pseudo Random Sequences (PRBS) at 5 or 10 Gbps, drives the XFP. The output signal from the transmitter goes through a polarization controller (PC) Thorlabs PM100S, that is used for the polarization dependence characterization scenario, and feeds a 99%/1% 1 × 2 coupler to enable a monitoring photodiode (PD) on the 1% output. The signal, coming from the 99% output of the previously described 1 × 2 coupler, then traverses a 2 × 2 coupler, where the remaining input is fed with an external tunable laser APEX AP3350A/3352A with wavelengths from 1526 to 1608 nm and a ~300 kHz linewidth, which is used in the wideband operation characterization as an LO, since the tunability of the integrated LO would not suffice for the test scenarios and the PIC was designed to allow the use of either an internal or external LO. At the 1% output of the 2 × 2 coupler, an Optical Spectrum Analyzer Advantest EQ8384 (OSA) is used for monitoring the incoming optical signal to the PIC, which corresponds to the 99% output of the 2 × 2 coupler. The optical interfacing is performed through a cleaved SSMF and a Spot Size Converter (SSC) at the PIC side. The differential electrical output of the recovered signal is then connected to the ExoBERT 2904 for real-time BER measurement, or a digital scope TEKTRONIX DSA72004 with a 20 GHz Bandwidth and 50 GS/s.

4.2. Experimental Results

The first set of results, depicted in Figure 8, were extracted to verify the band of operation, electrical bandwidth, and resilience to the characteristics of different commercial transmitters. The wavelength of the external tunable local oscillator for a given transmitter was swept around the optimum point of sensitivity for the BER 1E-3 to determine the penalty stemming from detuning, hence the bandwidth of the device was attained. This was performed for both 5 and 10 Gbps. In Figure 8a, the transmitter is a C-Band DML operating at 1532 nm, and in Figure 8b the transmitter is an L-band EML operating at 1596 nm. Figure 8a,b conclude that the intrinsic penalty between 5 and 10 Gbps arises, and that the behavior around the LO detuning is similar for both transmitter types. The measured bandwidth, shown by arrows, is broader when the transmitter is an EML, which is explained by the different spectral characteristics. Although the DML features a broader spectral width when compared with the EML, both setups are limited by the bandwidth of the ASIC (30 GHz). On both graphs, eye diagrams are shown for the optimal point of 10 Gbps operation.

In order to verify polarization independence, the internal LO of the InP PIC was used to recover the signal coming from the DML transmitter. The wavelength of the LO was controlled via a closed loop temperature stabilization method embedded in the control circuit of the PCB. By controlling the PIC, the Best-Case-Condition (BCC) and Worst-Case-Condition (WCC) of State of Polarization (SoP) were obtained for the measurements, using the BER at 1E-3 as the point of reference. After defining the BCC and WCC, the BER curves were obtained and the results are depicted in Figure 9, where the 0 of the normalized Received Optical Power (ROP) corresponds to the BCC SOP at 1E-3. It was observed that all SOP could be recovered, and that the BER curve slopes for the BCC and WCC are similar. The penalty between both cases varies between 3.5 and 4.5 dB. Intrinsic to the architecture employed, a 3 dB penalty was theoretically expected, the remaining experimental penalty is explained by the polarization dependence of the PIC components, e.g., the responsivity of the PD.

To assess the dependency of the power of the LO, an external tunable laser was used. By sweeping the signal optical power at the input of the PIC, BER levels were taken for the LO powers of +5 dBm, +7 dBm, and +10 dBm. Figure 10 depicts the results, which are normalized in terms of a penalty with respect to the 10E-3 BER level when the LO is +10 dBm. With the increase of the LO’s power, an enhancement of the performance is observed due to the gain effect of the quasi-coherent mechanism. An approximate 2 dB improvement can be noted between the curves of +5 and +7 dBm. However, when the LO power reaches +10 dBm, the improvement does not continue, which may be related to the saturation of the receiver. The recovered eye diagram for the case of LO power at +10 dBm is also shown as an inset in Figure 10.

To exploit the tunability range, the power penalty was characterized for a BER of 1E-3 using 4 WDM channels separated by 100 GHz for both the C- and L-bands, referring to the upstream and downstream channels of NG-PON2. In the case of the C-band, two different LOs were used so that the impact of the laser linewidth could also be assessed. One of the LO sources was an external tunable laser and the other was a DFB whose linewidth is of the order of a few hundred MHz, which is comparable to that of the LO monolithically integrated into the PIC. The XFP used as a transmitter signal is a commercial NG-PON2 ONU based on a DML. In the L-band experiment, the external tunable laser was used as LO and the transmitter signal was generated by a commercial XFP OLT based on external modulation. The results can be observed in Figure 11, where it is possible to conclude that the penalty within each band is less than 1–1.5 dB for the C- and L-bands, respectively. This demonstrates the wideband tunability of the implemented architecture and methodology. As seen in the C-band, the DFB laser used as an LO leads to a higher penalty.

5. Conclusions

PONs have been evolving over the past decades following consumers’ demand for higher bitrates. With this evolution complexity arises, being defined the current next trends as WDM PON with 25 or 50 Gbps per channel. For this increase in capacity per channel, Data Center technologies are being adapted to be able to meet burst mode transmission (ONU) and reception (OLT), and an increased link budget, with electro-optical hardware that has already had feasibility demonstrations performed.

Although photonic integration was introduced in the late 60s, the steps to harmonize PIC development only occurred four decades after that, with the creation of generic photonic integration fabrication and packaging lines. The availability of photonic generic processes allowed the increase of accessibility to PIC development. For PONs, the need is to develop passive building blocks specific to the wavelength band and channel separation, as well as transmitter and receiver integrated architectures that can meet the tight standard requirements, associated with the development of integrated drivers for the dedicated PIC architectures and assembly steps at the electrical optical and thermal levels that can meet on the one hand, the electro-optical performance required, and on the other hand, the power consumption required by the target multi-source agreement form factors. Extensive work in this field was demonstrated in this paper, with a focus on receiver integrated architectures, where a way to achieve the required tunability and high sensitivity is to use coherent architectures, a path already explored by several authors. Work has also focused on the simplifications that a coherent receiver can have when applied to a PON. A case study of an integrated receiver, where a PIC and ASIC are co-packaged, was presented, demonstrating its core characteristics of polarization robustness, bandwidth, wideband, and LO power dependency. All these developments pave the way to bring PIC-based transceivers to PONs in a rapid and cost-accessible manner, and to be able to efficiently deliver the performance required of the electro-optic equipment of PONs. Challenges stem from the high volume manufacturability capabilities of PIC based solutions with the proper yields for the application, passing by new design platforms and building blocks on the different bands of communication (in particular O-band), and finalizing novel packaging technologies that allow them to efficiently achieve passive or quasi-passive optical alignment between different PICs and their optical input/outputs while ensuring proper electrical and thermal performances.