PHOTONIC INTEGRATION

This is a continuation from the previous tutorial - Real-time implementation of high-speed digital coherent transceivers

 

 

1.  INTRODUCTION

Advanced modulation formats, together with wavelength-division multiplexing \(\text{(WDM)}\) and polarization-division multiplexing \(\text{(PDM)}\), have become key enablers to increase the spectral efficiency and the overall capacity per fiber. Coherent optical transmission has been employed in long-haul communications with channel data rates at \(\text{100G}\) and beyond and with transmission distance typically beyond \(\text{2000km}\).

PDM quadrature phase-shift keying \(\text{(QPSK)}\) is the format of choice in the current deployed \(\text{100G}\) networks. Next-generation \(\text{400G}\) networks may utilize \(\text{PDM}\) 16-ary quadrature-amplitude modulation \(\text{(16QAM)}\). Coherent transmission technology is also well-suited for metro and short-reach optical networks, in which small transceiver footprint along with reduced power consumption is strongly required. 

A coherent transponder mainly consists of three parts, namely optical components, modulator drivers, and transimpedance amplifiers \(\text{(TIAs)}\), and application-specific integrated circuits \(\text{(ASICs)}\), as shown in Figure 1. for a dual-polarization \(\text{QPSK}\) or \(\text{16QAM}\) transceiver.

The optical front-end consists of a number of high-performance optical components, such as narrow-linewidth lasers, high-speed modulators, high-dynamic-range photodetectors \(\text{(PDs)}\), polarization beam splitters and combiners \(\text{(PBS/Cs)}\), polarization rotators \(\text{(PRs)}\), and \(90^\circ\) optical hybrids.

The \(\text{ASICs}\) include electronic processing elements such as analog-to-digital converters \(\text{(ADCs)}\) and digital-to-analog converters \(\text{(DACs)}\) to capture and generate optical signals, respectively, digital signal processing \(\text{(DSP)}\) core to process and condition the

 

 

signals, and multiplexers and demultiplexers to aggregate and disaggregate the information payload. 

In current optical transceivers, discrete optical components with optimized performance are used for meeting the very challenging transmission requirements. Multisource-agreement \(\text{(MSA)}\) modules based on optical internetworking forum \(\text{(OIF)}\) recommendations have emerged, with the transceiver module size of \(5\times7\) square inches, power consumption of less than \(\text{80W}\), and the modulation format of \(\text{PDM}\)-\(\text{QPSK}\).

Nevertheless, the size and power consumption of this \(\text{MSA}\) transceivers for long-haul applications are not suitable for metro networks where port densities on both line side and client side are important. Smaller-form-factor \(\text{100G}\) coherent transceivers are under development by various subsystem vendors, equipment manufacturers, and module suppliers.

There are a few options of form factors under consideration, which all require high level of photonic integration. One choice is a nonpluggable \(4\times5\)-square inch MSA module with less than \(\text{40W}\) of power. Other choices include the compact form-factor pluggable \(\text{(CFP)}\), \(\text{CFP2}\), and even smaller \(\text{CFP4}\), as shown in Figure 2. For \(\text{CFP}\)/\(\text{CFP2}\), the size and power are limited to \(82\times145\text{mm}^2\)/\(41.5\times107\text{mm}^2\) and \(\text{32 W}\)/\(\text{12W}\), respectively.

Implementing a \(\text{100-G}\) coherent transceiver in such small modules is very challenging since considerably less space is available and low heat dissipation is required. Therefore, the \(\text{CFP}\) implementation requires a high degree of optical integration, low-power driver/\(\text{TIAs}\), and power-optimized \(\text{ASICs}\).

The power consumption for a coherent transceiver mainly includes the power consumed by the \(\text{ASICs}\) \(\text{(DACs}\), \(\text{ADCs}\), \(\text{DSPs)}\), lasers, modulator drivers, and \(\text{TIAs}\).

In order to pack all the components in a smaller form factor, the power consumption need to be reduced. The power consumption of the \(\text{ASICs}\) is a significant part of the overall power consumption.

 

 

This power can be reduced if implemented with smaller transistor feature size. Today, 40-nm complementary metal–oxide–semiconductor \(\text{(CMOS)}\) represents the state-of-the-art for mixed-signal \(\text{ASICs}\).

However, more advanced \(\text{CMOS}\) processes such as 28-nm will be available soon and may become the preferred feature size if low-power \(\text{ASICs}\) are required. In addition, compared with long-haul application, the metro link reach may be less than a few hundred kilometers.

The shorter transmission reach could require less link dispersion equalization, namely fewer filter taps and less processing are needed, leading to lower power consumption. The rest of the power consumption related to optical components become more critical once the \(\text{ASIC}\) power drops. Energy-efficient lasers and low-voltage electro-optic modulators would then be preferable. 

Photonic integration appears to be a must for \(\text{CFPx}\) modules. A typical length of a \(\text{LiNbO}_3\) modulator is about 100mm which makes it difficult to fit into even the largest \(\text{CFP}\) modules.

Alternative modulators based on indium phosphide \(\text{(InP)}\) and silicon photonic integrated circuits \(\text{(PICs)}\) are emerging as the most promising modulator candidates. To save space, it is also becoming critical to employ optical packaging without the use or with only limited use of micro-optic components, such as lenses, power splitters, polarizers, \(\text{PBS}/\text{Cs}\), and \(\text{PRs}\).

In summary, photonic integration is required to achieve compact coherent optical transceivers for next-generation pluggable modules to be used in metro and even shorter-distance coherent optical systems. In Section 2., we review three photonic integration technologies that may fulfill the stringent requirements for this application.

In Section 3., we review the progress in integrated transmitters, while in Section 4. we review the progress in integrated receivers. We finally make concluding remarks in Section 5.

 

 

2. OVERVIEW OF PHOTONIC INTEGRATION TECHNOLOGIES

There are mainly three promising optical technologies currently in development aiming for compact and low-power coherent transponders. Planar lightwave circuit \(\text{(PLC)}\) technology is used to realize passive optical components which may consist of power splitters/combiners and \(\text{PBSs}\)/\(\text{PBCs}\). By packaging high-speed modulators/detectors with \(\text{PLCs}\), “hybrid” coherent transmitters/receivers can be realized. \(\text{InP}\) and silicon monolithic \(\text{PICs}\) are the second and the third technologies, respectively.

Most of the demonstrated coherent transmitters and receivers do not have monolithic integration of the laser. Integration of the laser with the modulators or detectors may have limited advantages. This integration significantly increases the integration complexity for \(\text{InP}\) and silicon \(\text{PICs}\).

Moreover, a laser generally requires cooling, which consumes significant power. In order to save power, it would be better not to cool other devices such as modulators and detectors. Furthermore, a single laser can be shared in the transmitter and receiver to reduce the cost and power, which makes the laser integration less attractive.

For \(\text{PLC}\) hybrid technology, silica or polymer \(\text{PLCs}\) have been co-packaged with \(\text{LiNbO}_3\) modulators for transmitters, or have been assembled with \(\text{III}\)–\(\text{V}\) photodetectors for receivers. The overall sizes for optical front-end and the packaging cost could be concerns.

Higher-degree photonic integration can be achieved by \(\text{III}\)–\(\text{V}\) (mainly \(\text{InP)}\) and silicon \(\text{PICs}\). The PICs could significantly reduce the device sizes and simplify the packaging procedures and hence the cost. Both InP and silicon \(\text{PICs}\) are very promising, each with their own merits and limitations. \(\text{InP}\) \(\text{PICs}\) can provide integrated lasers, but it is challenging to integrate lasers, modulators, photodetectors and polarization elements all on a single chip.

Moreover, its yield and cost could be concerns. Silicon \(\text{PICs}\) take the advantage of \(\text{CMOS}\) foundries with large wafer size, high yield, and low cost. Silicon waveguides also have the flexibility to implement polarization combiners and rotators, which offers advantages in polarization diversity circuits. The emerging hybrid wafer-scale integration technology to bond \(\text{InP}\) on silicon offers another possibility to support integrated amplifiers and lasers.

\(\text{InP}\)-based \(\text{PICs}\) have been realized to integrate high-speed modulators or photodetectors, wavelength multiplexing filters, and even integrated lasers on the same chip for coherent applications.

However, polarization diversity is typically achieved by off-chip micro-optic polarization elements.

Silicon \(\text{PICs}\) also show very promising progress for coherent transceivers. Compact silicon microring modulators as well as silicon \(\text{Mach}\)–\(\text{Zehnder}\) modulators \(\text{(MZMs)}\)  have been demonstrated for in-phase/quadrature \(\text{(I/Q)}\) modulation.

Dual-polarization \(\text{I/Q}\) modulators have been implemented on a single chip. Silicon–organic \(\text{I/Q}\) modulators with low-power consumption at 40 Gbaud have been reported.

In addition, monolithic silicon \(\text{PICs}\) were also demonstrated for polarization-diversity coherent receivers and a single-chip transceiver was reported in. These demonstrations confirm the potential of silicon photonics for high-level integration capability for dual-polarization coherent transceivers.

Table 1. illustrates the comparisons of the three technologies discussed earlier. As each technology was evolving very rapidly and dramatically during the last few years, and is expected to continue to do so in the near future, this comparison serves as a general guideline rather than an accurate prediction.

If one does not consider monolithic integration of the laser, the integration level is highest for silicon \(\text{PICs}\) as polarization elements can be monolithically integrated on-chip, while it is lowest for \(\text{PLC}\) hybrids, as both high-speed optical devices and polarization elements need to

 

TABLE 1.  High-level comparison of different photonic integration technologies for coherent optical transceivers. As each technology evolves, this table may only serve as a guideline

 

be assembled together. Considering the integration level point of view, the packaging cost for silicon \(\text{PICs}\) maybe the lowest, with compact footprints as well. However, the optical performance of \(\text{PLC}\) hybrids may be the best as \(\text{PLCs}\) provide both low insertion loss and high-performance passive components, together with individually optimized modulators and \(\text{PDs}\) made from other materials. 

 

 

3. TRANSMITTERS

In this section, we review \(\text{I/Q}\) modulator technologies. This section is organized as follows. Section Dual-Polarization Transmitter Circuits describes the transmitter circuits for the realization of dual polarization\(\text{ I/Q}\) modulators.

Section High-Speed Modulators. examines the critical component in an \(\text{I/Q}\) modulator, namely the \(\text{MZM}\). Sections 3-5 summarize previously reported \(\text{I/Q}\) modulators based on \(\text{PLC}\) hybrids, \(\text{InP}\) \(\text{PICs}\), and silicon \(\text{PICs}\), respectively. 

 

Dual-Polarization Transmitter Circuits

The most widely used modulator for full E-field encoding is the nested \(\text{MZM}\), also called an \(\text{I/Q}\) modulator or a vector modulator. It consists of two \(\text{MZMs}\) in parallel, one for encoding the in-phase part of light, and the other for encoding the quadrate-phase part of light.

A power combiner combines the outputs of two \(\text{MZMs}\) with a \(\pi/2\) phase difference. Since a single-mode fiber can support two orthogonal polarization states, a dual-polarization \(\text{I/Q}\) modulator meets the requirement for maximizing information encoding.

The building blocks for such a dual-polarization \(\text{I/Q}\) modulator are a polarization-maintaining power splitter to split the continuous-wave \(\text{(CW)}\) laser power into equal powers to be launched into each \(\text{I/Q}\) modulator. At the output, one needs to combine one signal with the other signal in the orthogonal polarization using a \(\text{PR}\) followed by a \(\text{PBC}\).

Figure 3. shows a circuit diagram for a dual-polarization \(\text{I/Q}\) modulator. A multilevel \(\text{QAM}\) signal can be synthesized by using parallel \(\text{I/Q}\) modulators with binary electrical driving signals, or by a single \(\text{I/Q}\) modulator with multilevel electrical driving signals. 

Four \(\text{RF}\) driving signals and at least six tunable phase-control elements are required. The four RF electrical driving signals are usually termed as \(\text{IX}\), \(\text{QX}\), \(\text{IY}\), and \(\text{QY}\), where \(\text{I/Q}\) represents in-phase/quadrature components, and \(\text{X/Y}\) refers to the two polarization states.

 

 

For \(\text{QPSK}\) generation, the \(\text{RF}\) driving signals typically have two levels, while for \(\text{16QAM}\) generation, the RF driving signals typically have four levels. The \(\text{MZMs}\) inside each \(\text{I/Q}\) modulator are usually operated under push–pull condition, which makes them more tolerant to the variations of drive voltage and at the same time achieve zero-chirp operation. Each \(\text{MZM}\) (also sometimes referred to as “daughter” \(\text{MZM)}\) needs to be biased at its minimum transmission point. Four-phase elements \((\boldsymbol{\phi_1}-\boldsymbol{\phi_4})\) in the daughter \(\text{MZMs}\) are, therefore, present in order to tune the phase difference between the two arms of \(\text{MZMs}\) to be at \(\pi\).

In commercial \(\text{I/Q}\) modulators, active locking to \(\pi\)-phase differences are required by monitoring the light from \(\text{MZMs}\). For \(\pi/2\)-phase difference between \(I\) and \(Q\) branches, two additional phase elements \((\boldsymbol{\phi_5}\;\text{and}\;\boldsymbol{\phi_6})\) after the daughter \(\text{MZMs}\) are needed. 

Different technologies have different monolithic integration levels. For \(\text{PLC}\) hybrids, the passive power coupler and phase elements, sometimes with variable optical attenuators \(\text{(VOAs)}\), are on \(\text{PLCs}\), while \(\text{MZMs}\), \(\text{PRs}\), and \(\text{PBCs}\) are off chip.

For \(\text{InP}\) \(\text{PICs}\), \(\text{PR}\) and \(\text{PBC}\) are typically external, and the rest can be monolithically integrated. For silicon \(\text{PICs}\), all components (except the laser) can be on a single chip.

 

High-Speed Modulators

Modulators are one of the most crucial and challenging optical devices in coherent transceivers. The performance of an \(\text{MZM}\) can be characterized by three most important parameters, that is, the \(\text{3-dB}\) electro-optic bandwidth, the half-wave voltage swing \(V_\boldsymbol{\pi}\), and the optical insertion loss.

For coherent transceivers, it is highly desirable to have both higher bandwidth and lower \(V_\boldsymbol{\pi}\). This is driven by the need for low power consumption, high data rate operation, as well as difficulties in developing high-voltage broad-band linear electronic amplifiers.

However, reducing \(V_\boldsymbol{\pi}\) usually results in reduced bandwidth and increased insertion loss because of the need of longer phase shifter in the \(\text{MZM}\). 

There are two common types of \(\text{MZMs}\) operating in push-pull configuration, that is, single drive and dual drive. For a single-drive push–pull \(\text{MZM}\), the two arms of \(\text{MZM}\) are electrically connected and a single \(\text{RF}\) signal drives the two arms simultaneously.

With a proper electrode arrangement, the push–pull operation can be obtained, which is important for zero-chirp \(\text{QPSK}\) and \(\text{16QAM}\) modulation formats. For dual-drive push–pull, the two electrodes that drive two arms are independent. If an \(\text{RF}\) signal and its complementary are applied on the two electrodes simultaneously, push–pull operation is achieved.

Typically, \(V_\boldsymbol{\pi}\) is defined as a voltage swing for a phase change of \(\boldsymbol{\pi}\) between two arms of the \(\text{MZMs}\), where the voltage is applied on one arm. Therefore, for single-drive \(\text{InP}\) and silicon \(\text{MZMs}\), a full \(V_\boldsymbol{\pi}\) is needed to achieve a phase change of \(\boldsymbol{\pi}\) between the two arms, while for dual-drive, only \(V_\boldsymbol{\pi}/2\) is required on each arm.

Depending on the availability of modulator drivers and driver output, either single drive or dual drive may be preferable.

For the device design, Single-drive \(\text{InP}\) and silicon \(\text{MZMs}\) are better for broad bandwidth, but may have higher insertion loss. The bandwidth benefit comes from the fact that the two arms connect to the \(\text{RF}\) transmission line are in series, resulting in the loaded capacitance being reduced to half.

However, in order to use same drive voltage amplitude, the phase shifter in single-drive \(\text{MZMs}\) needs to be twice as long as the dual-drive \(\text{MZMs}\), which can make the insertion loss higher, and also degrades the bandwidth.

The \(\text{InP}\) material system (which also includes ternary materials such as InGaAs, and quaternary materials such as \(\text{InGaAsP}\) and \(\text{InGaAsAl)}\) can provide both electro-absorption \(\text{(EAM)}\) and electro-refraction modulations.

Although \(\text{InP}\) \(\text{EAMs}\) have been employed to achieve \(\text{QPSK}\) and \(\text{16QAM}\), \(\text{I/Q}\) modulators based on electro-refractive or phase modulation are more popular. The high-speed modulation in \(\text{InP}\) waveguides can be achieved by applying a reverse voltage across a p–i–n or n–i–n junction, as shown in Figure 4.

There are two main electro-optic effects involved. The first one is the Pockels effect, which is similar to that in \(\text{LiNbO}_3\). The Pockels effect occurs in crystals that lack inversion symmetry. The refractive index change is proportional to the strength of the electrical field and the sign of the index change depends on the orientation of the electrical field with respect to the crystal axis.

This effect is an ultra-fast electro-optic effect; however, it is typically a weak effect, and may require centimeter-long device to achieve the desired \(\boldsymbol{\pi}\)-phase shift for a low \(V_\boldsymbol{\pi}\). The second effect is a shifting of the band edge with an applied voltage, leading to a change in the absorption.

It is called the Franz–Keldysh effect in bulk material and the quantum-confined Stark effect \(\text{(QCSE)}\) in multiple quantum wells \(\text{(MQWs)}\). The applied electrical field shifts the band edge toward longer wavelengths, which causes changes to both absorption and refractive index for a

 

 

signal whose wavelength is located a few tens to about one hundred nanometers larger than the band edge wavelength. Depending on the signal wavelength, \(\text{EAMs}\) can be realized if the shifted band edge reaches or is close to the signal wavelength, and therefore the absorption effect results in a modulation of light. If the signal wavelength is far away from the band edge, \(\text{MZMs}\) can be realized by exploiting the phase change that occurs with the applied voltage. 

In recent years, silicon photonic \(\text{MZMs}\) have been undergoing significant developments. Silicon itself does not have strong electro-optic effects such as the Kerr effect and the Franz–Keldysh effect as the signal wavelength at 1550 or 1330 nm is too far away from the silicon band edge.

Furthermore, silicon does not exhibit the Pockels effect due to inversion symmetry. Researchers have reported electro-optic modulators by employing other hybrid materials deposited on silicon, such as germanium, polymer, and \(\text{III}\)–\(V\) semiconductors. Nevertheless, having an all-silicon modulator would potentially yield a very reliable structure. High-speed modulation in silicon itself can be realized by free-carrier-induced refractive index change. The carrier-density modulation in a silicon waveguide can be obtained with carrier injection in a forward-biased p–i–n diode structure, carrier accumulation in an \(\text{MOS}\) capacitor structure, or carrier depletion in a reverse-biased p–n diode structure (see Figure 5).

Carrier injection is the most efficient modulation mechanism, but it is difficult to achieve high speed since it suffers from high junction capacitance and slow free-carrier recombination (Figure 5a). Complicated driving signals with pre-emphasis can be used to speed up the free-carrier injection/recovery process and thus increase the speed of the modulator. Carrier accumulation in \(\text{MOS}\) capacitors (Figure 5c) has better modulation efficiency than carrier-depletion in a reverse-biased p–n junction (Figure 5b), but the speed of \(\text{MOS}\)-type modulators may be limited by the high capacitances from the thin oxide layer. Carrier depletion has the worst modulation efficiency, yet comes with the best high-speed performance, as the junction capacitance can be reduced by optimizing doping profiles.

Table 2. summarizes the performance of \(\text{LiNbO}_3\), \(\text{InP}\), and silicon \(\text{MZMs}\) reported in the literature. For both \(\text{LiNbO}_3\) and \(\text{InP}\) modulators, a \(V_\boldsymbol{\pi}\) of 1–2.5\(V\) can be achieved with a bandwidth about 30–40 \(\text{GHz}\), while for silicon \(\text{MZMs}\), the best reported \(V_\boldsymbol{\pi}\) is 3.0–4.0\(V\) with a bandwidth 10–20 \(\text{GHz}\) for reverse-biased p–n junction modulators.

In addition, silicon \(\text{MZMs}\) tend to have higher insertion loss as p–n junctions in the waveguide center result in significant free-carrier-induced propagation loss. For \(\text{MOS}\)-type silicon \(\text{MZMs}\), the \(V_\boldsymbol{\pi}\) is

 

 

 

TABLE 2. Performance comparisons of \(\text{MZMs}\) with different technologies 

 

comparable to the p–n junction modulator but the insertion loss may be even higher. In silicon–organic modulators, where polymers are deposited on silicon slot waveguides or photonic crystals, the \(V_\boldsymbol{\pi}\) can be comparable with \(\text{LiNbO}_3\) and \(\text{InP}\) modulators, but the insertion loss is higher and also the reliability is a concern.

It is to be noted that although the bandwidth-\(V_\boldsymbol{\pi}\) performance of InP modulators are comparable to those of \(\text{LiNbO}_3\), the fiber coupling to \(\text{LiNbO}_3\) \(\text{MZMs}\) is much easier due to better mode matching.

In addition, \(\text{InP}\) modulators typically have phase shifters with lengths of a few millimeters, similar to that of silicon \(\text{MZMs}\), while for \(\text{LiNbO}_3\), the length can be a few centimeters. 

 

\(\text{PLC}\) Hybrid \(\text{I/Q}\) Modulator 

It was recognized that several functional modulators could be realized using multiple I/Q modulators by appropriately combining these modulators with passive PLCs both at the input and at the output.

As seen in Figure 6, devices such as a \(\text{QPSK}\) modulator, a \(\text{DP}\)-\(\text{QPSK}\), a 64-\(\text{QAM}\) modulator, and a dual-carrier orthogonal frequency-division multiplexing \(\text{(OFDM)}\) transmitter have been demonstrated using the versatile hybrid integration platform.

The \(\text{PLC}\) part incorporated passive elements such as splitters, combiners, \(\text{VOAs}\), phase shifters, and polarization beam rotators/combiners, whereas the \(\text{LiNbO}_3\) part incorporated several \(\text{MZMs}\) that could be combined to form multiple \(\text{I/Q}\) modulators.

Nevertheless, with the widespread adoption of \(\text{DACs}\) at the transmitter to synthesize multilevel modulation formats using a single dual-polarization \(\text{I/Q}\) modulator, these functional modulators have been largely used for research demonstrations and have not made it into commercial products. 

 

\(\text{InP}\) Monolithic \(\text{I/Q}\) Modulator

In 3., we list some examples of monolithically integrated \(\text{InP}\) \(\text{I/Q}\) modulators as well as transmitters that include a laser. The examples clearly show that \(\text{InP}\) modulators have made excellent progress in the last few years, with the latest \(\text{InP}\) modulators with a \(V_\boldsymbol{\pi}\) of 1.5\(\text{V}\) and bandwidth \(>40\;\text{GHz}\). 

 

 

 

TABLE 3.  Some reported \(\text{I/Q}\) modulators on \(\text{InP}\) \(\text{PICs}\) 

 

There may exist some challenges to implementing and operating \(\text{InP}\)-based \(\text{I/Q}\) modulator. Unlike \(\text{LiNbO}_3\) modulators, which purely rely on the electro-optic \(\text{(EO)}\) interaction via the Pockels effect, \(\text{InP}\) \(\text{MZMs}\) exhibit a more complex behavior due to the interplay of both the electro-optic effect as well as the electro-absorption effect, as discussed in Section High-Speed Modulators The phase modulation of an \(\text{InP}\) \(\text{MZM}\) via voltage modulation exhibits some nonlinearity, with increasing modulation efficiency at deeper bias.

In addition, there is the voltage-dependent optical absorption associated with the shift and broadening of the exciton peak \(\text{QCSE}\). The two effects make operation of 

 

 

the \(\text{InP}\) \(\text{MZM}\) in a “chirp-free” manner challenging and need calibration and look-up tables across the wavelength range. In addition, the epitaxial layer structure for modulation is quite different from that needed for a tunable laser.

Nevertheless, there has been progress in fully integrated transmitter-modulator assemblies, as shown in Table 3. 

As an example, Chandrasekhar et al. presented a two-chip integrated transmitter, which was directly driven by \(\text{CMOS}\) \(\text{DACs}\) (differential output of 1 \(V\)) to generate 32-Gbaud \(\text{QPSK}\) and \(\text{16QAM}\) formats without the use of modulator drivers.

The schematic of the integrated device is shown in Figure 7. The laser was a digital supermode \(\text{(DS)}\) distributed Bragg reflector \(\text{(DBR)}\) laser, and it was optically coupled with micro-optics to a second InP chip that consisted of a pulse carver followed by a single polarization vector modulator.

The laser was tunable over the entire \(C\)-band and the analog bandwidth of the MZMs that comprised the vector modulator had a bandwidth of about 12 \(\text{GHz}\), limited by the package.

Nevertheless, using \(\text{DSP}\) at the transmitter to equalize the bandwidth roll-off, operation up to 32 Gbaud was achieved and transmission over 8000 km for the 100-Gb/s \(\text{PDM}\)-\(\text{QPSK}\) format and 900 km for the 200-\(\text{Gb/s}\) \(\text{PDM}\)-\(\text{16QAM}\) format was successfully demonstrated. 

 

Silicon Monolithic \(\text{I/Q}\) Modulator 

In 2012, an \(\text{I/Q}\) modulator based on nested single-drive push–pull silicon \(\text{MZMs}\) was reported in, where a 50-\(\text{Gb/s}\) \(\text{QPSK}\) signal was generated with only 2.7-\(\text{dB}\) optical signal-to-noise ratio \(\text{(OSNR)}\) penalty from the theoretical limit at a bit error ratio (BER) of \(10^{-3}\).

Compared with commercial \(\text{LiNbO}_3\) \(\text{I/Q}\) modulators, there is only \(\sim1\;\text{dB}\) \(\text{OSNR}\) penalty.

This is the first successful demonstration of advanced modulation formats using silicon \(\text{MZMs}\).

By further integrating two \(\text{I/Q}\) modulators and an on-chip polarization rotator and \(\text{PBC}\), a monolithic single-chip dual-polarization coherent modulator was implemented to generate a 112-\(\text{Gb/s}\) \(\text{PDM}\)-\(\text{QPSK}\) and a 224-\(\text{Gb/s}\) \(\text{PDM}\)-\(\text{16QAM}\) signal, as shown in Figure 8(a) and (b).

It was shown that the integration of on-chip polarization elements introduces an additional 0.9-\(\text{dB}\) penalty due to polarization-dependent loss \(\text{(PDL)}\). The \(\text{PR}\) is 

 

 

realized by a \(\text{SiN}\)-assisted taper structure. This \(\text{PIC}\) is the first monolithic single-chip dual-polarization \(\text{I/Q}\) modulator, with highest photonic integration in this particular application.

Dual-polarization \(\text{I/Q}\) modulators were further demonstrated, where a novel polarization rotator based on silicon-only structure was employed, as shown in Figure 8(c)., 128-\(\text{Gb/s}\) \(\text{PDM}\)-\(\text{QPSK}\) was generated, with \(\text{BER}\) performance shown in Figure 8(d). 

Single-polarization silicon \(\text{I/Q}\) modulators were also reported using \(\text{MOS}\)-capacitor type silicon modulators with performance comparable to \(\text{LiNbO}_3\)-based \(\text{I/Q}\) modulators.

Here, a compact \(\text{I/Q}\) modulator with a size of less than \(1\text{mm}\times25\mu m\) per \(\text{MZM}\) was used, and the \(\text{MZM}\) was driven by a low-power \(\text{CMOS}\) driver with less than 200 mW per channel.

The 56-\(\text{Gb}\)/s \(\text{QPSK}\) signals were generated, externally polarization multiplexed to 112-Gb/s, and transmitted through 2427-km standard single mode fiber.

The performance showed negligible \(\text{OSNR}\) penalties were associated with the silicon modulator compared to \(\text{LiNbO}_3\) modulators for back-to-back operation, and a small penalty was found after transmission due to chromatic dispersion. 

Single-polarization \(\text{I/Q}\) modulators were also realized using silicon–organic modulators where an electro-optic polymer is spin-coated on silicon slot waveguides. High modulation efficiency could be realized through both the high \(\text{EO}\) coefficient of the polymer as well as the high E-field concentration in the slot. Up to 40-Gbaud \(\text{QAM}\) signals were generated.

 

 

4. RECEIVERS 

In this section, we review coherent optical receiver technologies. This section is organized as follows. Section Polarization Diversity Receiver Circuits describes the principle of coherent detection and optical circuits for polarization diversity receivers. Sections \(\text{PLC}\) Hybrid Receivers, \(\text{InP}\) Monolithic Receivers and Silicon Monolithic Receivers summarize previously reported coherent receivers based on \(\text{PLC}\) hybrids, \(\text{InP}\) \(\text{PICs}\), and silicon \(\text{PICs}\), respectively. Section Coherent Receiver with \(120^\circ\) Optical Hybrids presents coherent receivers based on \(120^\circ\) optical hybrids. 

 

Polarization Diversity Receiver Circuits 

The fundamental principle of coherent detection is to measure the resulting electrical field following the beating of a signal and a \(\text{CW}\) local oscillator \(\text{(LO)}\) laser. Generation of the beat signal requires mixing the signal and \(\text{LO}\) in an optical mixer.

For example, mixing the signal and \(\text{LO}\) using a \(2\times2\) coupler will result in a beat signal. Balanced detection can be employed at the two outputs to suppress direct current \(\text{(DC)}\) components.

However, only single-quadrature information of the signal can be measured with such \(2\times2\) coupler. In order to detect the full electric field in the complex plane, one typically uses an optical \(90^\circ\) hybrid, which sends the mixed portions of the beat signal to four outputs.

The four outputs represent a beat signal that is \(0^\circ\), \(90^\circ\), \(180^\circ\), and \(270^\circ\) shifted in phase between the signal and \(\text{LO}\). If the signal and \(\text{LO}\) are launched into a \(90^\circ\) optical hybrid, the four output fields become 

\[\tag{1}E_i=\frac{1}{2}\left(E_s+e^{j(i-1)\frac{\pi}{2}}E_L\right)\]

where \(\text{E}_L\) and \(\text{E}_S\) are the electrical fields of the \(\text{LO}\) and signal, respectively; \(i=1-4\) for four outputs. Output photocurrents from balanced photodetections are then given as

\[\tag{2}\left\{\begin{array}\text{I_I=I_1-I_3=|E_S||E_L|\cos(\phi)}\\I_Q=I_2-I_4=|E_S||E_L|\sin(\phi)\end{array}\right.\]

Here, \(\phi\) represents the phase difference between the signal and \(\text{LO}\). The above-mentioned photocurrents produce the \(\text{I/Q}\) components of the signal without \(\text{DC}\) components, fulfilling the requirements to measure both the magnitude and phase of the signal. 

For a dual-polarization signal such as \(\text{PDM}\)-\(\text{QPSK}\), a polarization-diversity receiver is required. Figure 9. shows three possible optical circuits for polarization diversity coherent receiver. In the first circuit shown in Figure 9(a), the input polarization of the LO is \(45^\circ\).

Two \(\text{PBSs}\) are used to split the signal and \(\text{LO}\) into transverse-electric \(\text{(TE)}\) and transverse-magnetic \(\text{(TM)}\) components, respectively. The \(\text{TE}\) (or \(\text{TM)}\) components of signal and \(\text{LO}\) are mixed with a \(90^\circ\) optical hybrid designed for \(\text{TE}\) (or \(\text{TM)}\).

In this circuit, a \(\text{PR}\) is not needed, but the optical hybrids and the \(\text{PDs}\) need to be designed for two polarizations, which is not trivial for \(\text{InP}\) and silicon \(\text{PICs}\).

In the second circuit shown in Figure 9(b), a \(\text{PBS}\) is used for the signal port only to split the signal into \(\text{TE}\) and \(\text{TM}\) components. For the \(\text{TM}\) component, a \(\text{PR}\) rotates the \(\text{TM}\) to \(\text{TE}\) polarization.

In the \(\text{LO}\) side, only \(\text{TE}\) is coupled in and then a power splitter is used. Both the two \(90^\circ\) hybrids and the following \(\text{PDs}\) are designed for \(\text{TE}\) polarization only.

The advantages of this circuit include \(\text{(i)}\) the optical hybrids and \(\text{PDs}\) are designed only for one polarization and \(\text{(ii)}\) the input \(\text{LO}\) is polarized only to \(\text{TE}\), which matches the typical output polarization from a semiconductor laser. In this scheme, an additional component of

 

 

 

\(\text{PR}\) is required, which may further cause unwanted \(\text{PDL}\). A more symmetric circuit shown in Figure 9(c) uses two \(\text{PBSs}\) and two \(\text{PRs}\) for both the signal and \(\text{LO}\) ports. Furthermore, the input polarization of \(\text{LO}\) can be fine-tuned to compensate the \(\text{PDL}\). 

 

PLC Hybrid Receivers 

Silica \(\text{PLCs}\) offer high-performance passive devices such as optical power splitters/ combiners, \(90^\circ\) hybrids, and arrayed waveguide gratings \(\text{(AWGs)}\), but not high-speed \(\text{PDs}\). To implement coherent receivers, \(\text{PD}\) arrays can be bonded or packaged with \(\text{PLCs}\). Such \(\text{PLC}\)-based hybrid coherent receivers have achieved excellent device and system performance in a compact footprint.

To illustrate the versatility of the platform to incorporate various optical elements, we cite the work by Kurata et al.

In this tutorial, heterogeneously integrated eight high-speed \(\text{PDs}\) on a silica-based PLC platform with a PBS, \(90^\circ\) optical hybrids, and a \(\text{VOA}\).

The use of a 2.5% index contrast waveguide reduced the receiver \(\text{PLC}\) size to \(11\times11\;\text{mm}^2\). The schematic of the chip layout is shown in Figure 10(a), and the optical circuit employed is the one shown in Figure 9(b). 

The \(\text{VOA}\) driven by the thermo-optic effect is composed of a Mach–Zehnder interferometer \(\text{(MZI)}\) and thin-film heaters. The \(\text{PBS}\) is also composed of an \(\text{MZI}\) and two polyimide quarter waveplates. The waveplates were tilted at \(0^\circ\) and \(90^\circ\), respectively, and inserted in the two arms of the \(\text{MZI}\). A half-wavelength waveplate is inserted in

 

 

 

the pathway of one polarization after the signal passes through the \(\text{PBS}\) to realize the polarization rotation. In the \(90^\circ\) hybrids, a \(2\times2\) multimode interference \(\text{(MMI)}\) coupler was designed for a \(90^\circ\) phase shifter instead of a delay line, because an MMI is more tolerant to fabrication errors such as refractive index and waveguide errors.

The interference signals from the hybrids are reflected into the \(\text{PDs}\) by the micromirror fabricated at the chip edge. The fabrication sequences for such heterogeneous integration are shown schematically in Figure 10(b).

The \(\text{PLC}\) is first fabricated and is bonded to an InP chip. Next, the \(\text{InP}\) substrate is removed and only the top epitaxial layers are left for the following conventional \(\text{PD}\) fabrication. Finally, the micromirror is fabricated by dry tilt etching and aluminum deposition. 

The integrated coherent receiver exhibited a total insertion loss of about 11 dB, a polarization extinction ratio \(\text{(PER)}\) of more than 20 dB, optical hybrid phase error of \(\pm2^\circ\), and photodetector external responsivities of about 0.015 A/W.

These values are quite comparable to the state-of-the-art coherent receivers. The integrated receivers were subsequently packaged with \(\text{TIA}\) arrays into a very compact module with a volume of only \(1.3\;\text{cm}^3\).

 

\(\text{InP}\) Monolithic Receivers

The second approach to achieve a small form factor coherent receiver is to use \(\text{InP}\) \(\text{PIC}\) to implement the \(90^\circ\) hybrid and the balanced \(\text{PDs}\). In some cases, the \(\text{LO}\) laser has also been integrated.

The challenges are to realize monolithic \(\text{PBSs}\) and \(\text{PRs}\) and also to achieve uniform performance across the full C-band wavelength win-dow for such devices to become useful in commercial applications.

Some examples of monolithically integrated all-\(\text{InP}\) coherent receivers that have been demonstrated are listed in Table 4. Micro-optic \(\text{PRs}\) and \(\text{PBSs}\) are typically employed by packaging methods, a four-channel dual-polarization coherent receiver was realized on a monolithic \(\text{InP}\) \(\text{PIC}\) by the use of a novel interleave-chirped \(\text{AWG}\) that acts simultaneously as a wavelength demultiplexer, \(90^\circ\) hybrids, and also polarization splitter.  a 10-channel receiver chip monolithically integrated with 10 \(\text{LO}\) lasers and also an \(\text{InP}\) \(\text{AWG}\) was reported. The receiver chip can detect 100-\(\text{Gb}\)/\(\text{s}\) \(\text{PDM}\)-\(\text{QPSK}\) signals but external \(\text{PBSs}\) and \(\text{PRs}\) were used.

 

Silicon Monolithic Receivers

Monolithic dual-polarization coherent receivers have been demonstrated based on silicon \(\text{PICs}\). Doerr et al. demonstrated a grating-assisted coherent receiver \(\text{PIC}\), where two-dimensional \(\text{(2D)}\) gratings are used for fiber coupling (shown in Figure 11a). A \(\text{2D}\) grating coupler is a photonic crystal that couples a wave traveling normal to a substrate to a wave guided parallel to the substrate.

Light incident on the grating from the vertical direction is phase-matched to waveguide-guiding modes in a certain direction at the designed polarization state. With these gratings, the \(\text{TE}\) and \(\text{TM}\) components of signal (and \(\text{LO})\) can be coupled and separated into different silicon waveguides. Once coupled in, all of the light on the chip is \(\text{TE}\) polarized. Therefore, the \(\text{2D}\) gratings can also realize polarization splitting, polarization 

 

TABLE  4. Some previously reported coherent receivers based on InP PICs 

 

 

rotation, and \(50/50\) power splitting. The coupled signal and \(\text{LO}\) pass through two \(90^\circ\) hybrids based on \(2\times2\) \(\text{MMIs}\) and eight germanium photodetectors. The optical circuit for polarization-diversity coherent detection is essentially the same as in Figure 9(b).

Using this \(\text{PIC}\) packaged with in-house \(\text{TIAs}\), a 112-\(\text{Gb}\)/\(\text{s}\) \(\text{PDM}\)-\(\text{QPSK}\) signal was successfully detected with \(\text{BER}\) performance comparable to a commercial coherent receiver. 

While the \(\text{2D}\) gratings are elegant optical elements for a polarization-diversity receiver, the coupling loss is typically high and also the 1-\(\text{dB}\) bandwidth is limited to \(\sim40\)nm.

To solve these problems, edge coupling can be used, but on-chip \(\text{PBSs}\) and \(\text{PRs}\) are required. Dong et al. implemented a silicon coherent receiver by integrating on-chip polarization rotators and splitters, with the optical circuit shown in Figure 9(c). The optical signal enters the \(\text{PIC}\) from one facet with two polarizations, as shown in Figure 11(b).

The optical \(\text{LO}\) is coupled into the silicon \(\text{PIC}\) from the other facet. Once coupled in, the signal and \(\text{LO}\) are divided into \(\text{TE}\) and \(\text{TM}\) polarizations by two \(\text{PBSs}\). The \(\text{TE}\)-polarized light proceeds to the \(4\times4\) \(\text{MMI}\)-based \(90^\circ\) hybrids, whose four outputs are detected by four germanium photodetectors on the left side of the \(\text{PIC}\).

The \(\text{TM}\) components from the output of the \(\text{PBSs}\) are converted to TE polarization by two polarization rotators. The converted \(\text{TE}\) light enters the right-side \(4\times4\) \(\text{MMI}\)-based \(90^\circ\) hybrid, which is identical to that for the \(\text{TE}\) mode.

Using this \(\text{PIC}\) co-packaged with \(\text{TIAs}\), a 112-\(\text{Gb}\)/\(\text{s}\) \(\text{PDM}\)-\(\text{QPSK}\) signal and a 224-\(\text{Gb}\)/\(\text{s}\) \(\text{PDM}\)-\(\text{16QAM}\) signal were successfully detected.

The monolithic silicon coherent receivers have been used in the transmission experiment where both the transmitter and receiver front-ends are silicon \(\text{PICs}\) fabricated from the same wafers.

A 2560-\(\text{km}\) fiber transmission of a 112-\(\text{Gb}\)/\(\text{s}\) \(\text{PDM}\)-\(\text{QPSK}\) signal has been obtained at a \(\text{BER}\) of \(3.8\times10^{−3}\) (the typical threshold of a 7%-overhead hard decision forward error correction), as shown in Figure 12. This experiment validates the readiness of silicon \(\text{PICs}\) in optical coherent links.

Doerr et al. demonstrated a monolithic silicon PIC that contains all the optical front-end for a 100-\(\text{Gb}\)/\(\text{s}\) coherent transceiver except the laser (Figure 13). The silicon PIC was co-packaged with linear drivers and \(\text{TIAs}\) in a very compact gold-box module with a size of \(27\times35.5\text{mm}^2\) and power consumption \(<4.5\)\(\text{W}\). This \(\text{PIC}\) further verifies the high integration level of silicon photonic for this particular application, mainly by taking advantage of polarization elements on-chip. 

 

 

 

 

 

Coherent Receiver with \(120^\circ\) Optical Hybrids 

Instead of \(90^\circ\) hybrids, coherent detection can also be achieved by mixing a signal with an \(\text{LO}\) in \(120^\circ\) hybrids with three single-ended detectors. In general, balanced detection after \(90^\circ\) hybrids is superior in suppressing various noise sources.

This is preferred for long-haul transmission driven by high-performance requirements. Single-ended detection, however, may be lower in cost because of simpler \(\text{TIAs}\) and less \(\text{RF}\) connections between optical and electrical interfaces. Compared with \(\text{MMI}\)-based \(90^\circ\) hybrids, \(120^\circ\) hybrids may have a broader optical bandwidth, larger fabrication tolerance, and the ability to mitigate hardware-induced imperfections from \(\text{ADCs}\) and digital signal processors

If a signal and an \(\text{LO}\) are launched into a symmetric \(3\times3\) \(\text{MMI}\) with \(1:1:1\) power splitting, its three output photocurrents are given by Xie et al.

\[\tag{3}\left(\begin{align}I_1\\I_2\\I_3\end{align}\right)=\frac{1}{3}\left(\begin{array}&|E_L|^2+|E_S|^2\\|E_L|^2+|E_S|^2\\|E_L|^2+|E_S|^2\end{array}\right)+\frac{2}{3}\left(\begin{array}&|E_L||E_S|\cos(\phi+2/3\pi)\\|E_L||E_S|\cos(\phi)\\|E_L||E_S|\cos(\phi-2/3\pi)\end{array}\right)\]

where \(\text{E}_L\) and \(\text{E}_s\) are the electrical fields of the \(\text{LO}\) and signal, respectively, and \(\phi\) represents the phase difference between them. The first term in Equation 3. is the direct-detection term and the second term is the beat term, with a \(120^\circ\) difference among its three components.

A proper rearrangement of the photocurrents can yield the currents for \(\text{I}\)/\(\text{Q}\) components with the suppression of the direct-detection term:

\[\tag{4}\left\{\begin{align}&I_I=I_2-0.5I_1-0.5I_3=|E_L||E_S|\cos(\phi)\\&I_Q=\sqrt 3/2(I_3-I_1)=|E_L||E_S|\sin(\phi)\end{align}\right.\]

 

 

A monolithic polarization diversity coherent receiver by employing such \(120^\circ\) optical hybrids on a silicon \(\text{PIC}\). This \(\text{PIC}\) monolithically integrates silicon inverse tapers for fiber coupling, silicon polarization splitters, germanium high-speed photodetectors, and \(120^\circ\) optical hybrids based on \(3\times3\) \(\text{MMIs}\), as shown in Figure 14.

The chip size is only \(2.3\times1.5\text{mm}^2\). The 112-\(\text{Gb}\)/\(\text{s}\) \(\text{PDM}\)-\(\text{QPSK}\) signals were detected in the wavelength range of 1530–1580 nm. The \(\text{BERs}\) as a function of \(\text{OSNR}\), presented in Figure 14(d), demonstrate comparable performance to commercial receivers. The broadband operation indicates that this type of receiver may find applications in low-cost transceivers. 

 

 

5. CONCLUSIONS

This tutorial reviewed recent progress on photonic integration for coherent optical transceivers. In order to achieve compact and low-power pluggable modules for metro or short-distance applications, photonic integration is perceived as critical and necessary in order to address the need to have several optical elements in a small footprint.

In order to reduce optical packaging cost, it is also highly preferable for high-level monolithic photonic integration. Three main integration technologies have been reviewed in this chapter, that is, \(\text{PLC}\) hybrids, \(\text{InP}\), and silicon \(\text{PICs}\).

While \(\text{PLC}\) hybrids have the least monolithic integration of optical components, the performance tends to be the best. \(\text{InP}\) \(\text{PICs}\) have the advantages of monolithic integration of lasers with \(\text{MZMs}\) and \(\text{PDs}\). Silicon \(\text{PICs}\) excel in monolithic integration of polarization elements such as \(\text{PBS}\)/\(\text{Cs}\) and \(\text{PRs}\). While all the three technologies have demonstrated promising progresses in recent years, the cost, power, and size of the final modules may decide which technology will likely dominate. It is also possible that all these three technologies will be used in the future.