High-speed transceivers for short-reach (few cm to few tens of km) and long-haul optical communication links require high-performance modulators in their transmitter optical subassembly.^1–3 The key performance parameters for an efficient modulator are (a) high modulation efficiency^4–9 in amplitude or phase by a drive signal that is compliant with the complementary metal-oxide-semiconductor (CMOS) circuitry^5,6 [for phase modulators, the modulation efficiency is defined as the product of voltage Vπ applied to the phase shifter to achieve π phase shift and the length L of the phase shifter—a smaller Vπ·L represents higher modulation efficiency; for amplitude modulation, the ratio of extinction ratio (ER) to an applied voltage for a given length of the modulator represents the modulation efficiency and large values are desired]; (b) low insertion loss (IL),^10–13 both low loss in the active and passive parts of the modulator; (c) tens of gigabits per second (Gb/s)^14–17 modulation speed to support the future capacity demands of the optical communication networks; (d) energy consumption to be as low as a few tens of fJ/bit;^18–20 (e) large ER—typically ∼4dB for ∼2km transmission links and ∼8dB for tens of km long transmission links—to ensure high optical signal-to-noise ratio for maintaining a low bit error rate;^21–25 (f) low chirp^26–28 to realize higher-order modulation schemes^27–29 and to enhance dispersion tolerance of transmission links;^26,30 and (g) compact size^9,12,31–33 to enable high-integration density.^{1,2,21,34–36}

Generally, the modulator figure of merit (FoM) captures the performance of a modulator. In the literature, a variety of definitions for modulator FoM exist.^1,2,21,37,38 In its simplest form, Vπ·L represents the FoM for a phase modulator. Vπ·L·α is a phase modulator FoM that takes the phase shifter loss α into account. A phase modulator FoM given as fc·[CL·(Vπ·L)2·α/8]−1, where fc represents the phase shifter cutoff frequency and CL represents the phase shifter capacitance, relates the FoM in terms of modulator bandwidth, loss, and energy per bit.^2,4,21 The ratio ER/IL gives the FoM for amplitude modulators.^20,38 Dynamic optical modulation amplitude is a dynamic FoM of a modulator, as it takes modulator parameters and the modulation format into account.³⁷

High-index-contrast silicon photonics (SiPh) provides dense integration of complex photonic functionalities, such as high-speed modulators, using technology and process toolsets from CMOS electronics.³⁹ Other photonic integrated circuit (PIC) platforms such as indium phosphide,^40–42 which typically relies on 100 mm wafers, and lithium niobate (LiNbO3),⁴³ which commonly relies on 150 mm wafers, are fabricated in customized fabs. SiPh integrated circuits are typically fabricated on 200 or 300 mm wafers. The large wafer size allows for a large number of compact dies per wafer at low-cost scaling to large commercial volumes in existing CMOS commercial fabs and R&D pilot lines.^39,44–48 SiPh modulators promise to provide the best price-performance ratio and possible routes for monolithic integration of electronics and photonics.^{2,21,22,39,49} The commercial launch of SiPh-enabled transceivers is a testament to the promise of SiPh as a photonic integration technology to deliver efficient high-speed modulators.^{3,44–48,50–52}

The push for performant high-speed modulators stems from the ever booming growth of internet traffic. This growth has driven the modern data centers (DCs) into the so-called “Zettabyte Era,” where the annual global traffic is expected to exceed 2.2 zettabytes/year. This value corresponds to a threefold increase in just five years. The biggest contribution to this huge traffic originates from data moving inside the DCs over inter-DC connections.⁵³ Nonetheless, a remarkable increase in data exchanged between different DCs is also reported within the last few years (inter-DC connections). Future massive-scale DCs are expected to spread widely across the globe in order to meet the enormous bandwidth demands. An increase in baud rate imposes itself, yet at an even lower power consumption per bit. Consequently, crucial optimization of the intra- and inter-data center interconnects (DCIs)—where high-speed modulators are a key building block—is needed. More specifically:

For both intra- and inter-DCIs, SiPh appears to be a promising technology platform,^59,60 mainly because of the dense integration of photonics^59,60 while minimizing the overall cost-power efficiency budget.^18,19,39 Currently, the plasma dispersion effect, which is based on the movement (injection,^7,37,61–67 accumulation,^12,68–70 or depletion^{9,13,14,16,22,32,33,71–87}) of carriers to induce a refractive index change in a silicon waveguide, is the most widely deployed phenomenon to implement high-speed modulators in SiPh.^{5,7–19,22,32,33,37,61–69,69–71,71–121} Plasma dispersion modulators in SiPh have successfully demonstrated operations of 90 (Ref. 87) and 100Gb/s^122,123 for the on–off keying (OOK) modulation scheme. Furthermore, various plasma dispersion modulators implementations for advanced modulation schemes such as QPSK,^55,56,124 QAM,^125,126 PAM-4,^29,52,127 and PAM-8¹¹⁵ also exist.

The plasma dispersion effect is a weak electro-refractive effect.¹²⁸ A 1017cm−3 change in carrier concentration introduces an index change of Δn∼10−4. Furthermore, the phase modulation provided by the plasma dispersion effect is always associated with spurious amplitude modulation.^21,26 Unlike Pockels modulators, plasma dispersion modulators do not have a linear dependence of voltage and refractive index change Δn (thus, they show chirp), which makes them a less promising candidate for cases that require pure phase modulation.^{26,55,56,124,125,129} Nevertheless, R&D efforts have resulted in the performance of SiPh plasma dispersion modulators that is comparable with other photonic integration technologies.²⁷

Typically, plasma dispersion modulators rely on interferometric structures [such as the Mach–Zehnder interferometer (MZI)^{13,19,78,84,108,118} or optical cavities^{9,12,32,33,100,115}] for the translation of phase modulation into amplitude modulation. The carrier depletion-based interferometric plasma dispersion modulators, such as MZMs, are large in size (typically a phase shifter length of a few mm).^21,22 Injection or accumulation of carriers offers higher efficiency plasma dispersion effect phase shifters, enabling compact interferometric modulators.^37,65 The implementations of plasma dispersion high-speed modulators involving optical cavities [such as ring resonators (RRs)] result in compact devices (typically a few hundreds of μm2).^9,12,32,33 However, these RR-based devices suffer from high temperature and fabrication sensitivity.^130,131 Literature reports solving these problems to a certain extent through smart design strategies and technological advancements.¹³² Nonetheless, the limited operational optical bandwidth of the ring modulator (RM) is intrinsic to them.

It is a daunting task to deliver the future demands for single lane 100 Gbaud and beyond data rates in an energy-efficient manner. Recently, there have been a growing number of efforts to integrate electro-optic materials with a strong Pockels effect on SiPh platforms. The integration of such electro-optic materials promises to provide additional routes to upscale the performance of SiPh modulators to deliver the future demands for single lane 100 Gbaud and beyond data rates in an energy-efficient manner. Table 1 shows a rich landscape of SiPh modulators, where SiPh is incorporating a variety of other materials for implementing performant high-speed modulators. Modulation of amplitude by the Franz–Keldysh (FK) effect,^{20,38,128,133–140} quantum-confined Stark (QCS) effect,^141–152 and electrical gating, ^153–160 or the modulation of phase by employing the plasma dispersion effect,^161–166 Pockels effect,^{167–192,192–200} and inter-band transitions,^201–208 is feasible either directly in silicon^{20,31,38,133–139,143} or by integrating a variety of efficient electro-optic materials with SiPh platforms.^{161,172,182,184,202} The integration of new materials into SiPh has to counter various challenges. Invariably, any new material integration requires new technology and process development that allows for low-loss integration with the SiPh platform. The integration process must ensure high material quality. The introduction of defects in the integration process downgrades the yield of the PIC manufacturing process. Consequently, the integration of new materials always adds to manufacturing costs. Materials allowing wafer-scale integration and the ones for which the fab can leverage from the existing learning curve of introduction into other platforms reduce the financial impact and technological complexities, and they are thus preferred.

The electrical driving of the modulator comprises an electrode configuration and the driving scheme (such as single-ended drive with dual-arm push-pull, differential drive with dual-arm push-pull, differential drive with dual-arm push-pull with shared drive, or dual-differential drive with dual-arm push-pull) to feed the electrodes.²⁵ Generally, the length of the phase shifter determines the electrode configuration.²⁰⁹ Modulators with a short phase shifter length (phase shifter length <λRF/10)^2,25 use lumped electrodes.^209,210 Lumped driving is applicable to modulators comprising interferometric structures,^65,210,211 as well as cavity-based structures^{9,12,62,70,115} and slow-light structures.^112,113 With lumped electrodes, the applied voltage is approximately identical across the entire length of the phase shifter. Lumped driving does not necessarily require a matched termination of the driving electrodes. With unterminated lumped driving, the phase shifter’s capacitive load determines the energy consumption per bit (Ebit) for a modulator. Typically, the bandwidth of lumped electrode phase shifter driving is limited by its impedance (intrinsic, parasitic, and the one contributed by the driving circuitry) and the time taken by the photons to transit through the phase shifter. The latter effect is more pronounced for the slow-light modulators.^113,114

The traveling wave (TW) is the most common driving scheme for carrier depletion plasma dispersion MZMs.^{2,4,25,210,212} In this scheme, a single driver drives the entire electrode of the phase shifter.^25,212 The TW electrode typically comprises a coplanar waveguide transmission line. It terminates with a resistive impedance matched to the wave impedance of the electrode. The RF power applied to the TW electrode is consumed by the RF losses in the TW, capacitive loading by the phase shifter, and the matched termination. The matched resistance prevents electrical reflections and consequent interference with the electrical signal stream. But, it also increases the power consumption of the TW electrode as a fraction of the total input RF power is always consumed by the matched termination. Generally, TW driving schemes are six to eight times more energy-hungry than the lumped driving.^2,4 The detrimental factors impacting the bandwidth of the modulators with TW driving are: (a) RC time constant of the phase shifter; (b) walk-off (phase mismatch due to mismatch between the propagation speed of the electrical signal in the transmission line and the optical signal in the waveguide) between the electrical and optical signals; and (c) RF-losses in the TW electrode.

The segmented topology with distributed driving comprises a series of short phase shifters, which are individually driven by lumped electrodes.^6,22,66 Segmentation of the phase shifter with distributed driving avoids the RF attenuation due to a long transmission line while enabling the same voltage swing across the whole length of the phase shifter.²¹² As compared to the TW architecture, the segmented electrodes require smaller-size voltage drivers.²⁵ The precise delivery of the electrical signal to a phase shifter segment requires time synchronization with the transit time of light in that segment, which requires timing control circuitry for precise delivery of the electrical signal to each segment. This additional overhead undermines the energy efficiency of the segmented drivers and adds to the driver complexity. The number of phase shifter segments and their length determine the modulator bandwidth and its power efficiency. Typically, segmented electrodes with distributed drivers require tighter connectivity between the electronic drivers and the phase shifter segments to preserve the signal integrity and prevent parasitic effects. Therefore, this driving scheme demands either monolithic integration of electronics and photonics or flip-chip bonding photonic and electronic chips through an array of microbumps. Electrical driving plays a crucial role in defining the performance (primarily the power consumption and bandwidth) of high-speed phase modulators.^2,122,210 In the early years, most of the high-speed modulator demonstrations focused on the optimization of modulator performance in isolation from the driver design.¹²² Recent demonstrations of high-speed modulators have established that concurrent co-design and co-optimization of electrical driving and photonic phase shifter design^25,52,122 are a key to extracting the best performance for all modulator performance attributes.^52,213 This holds for modulators involving monolithic or hybrid integration with electronics.^25,122,213

This paper aims to provide a comprehensive review of the new technologies that provide additional, viable routes to enhance the performance of SiPh modulators. These new technologies involve the integration of materials with high electro-optic coefficients with SiPh platforms. As a result, they promise to deliver modulator performance that is required to meet the blistering surge in transmission rates for the optical links. This paper also compares these new technologies against the current mainstream SiPh modulator implementations. A comprehensive coverage in full detail is very challenging and difficult to be performed in a paper with a limited page number. Therefore, we will limit the discussion in this paper to some of the most important and relevant implementations of high-speed modulators in SiPh. It is important to note that the SiPh modulators have been used to modulate signals by means of advanced modulation schemes such as QPSK,^55,56,124 QAM,^125,126 PAM-4,^29,52,127 and PAM-8.¹¹⁵ However, in this paper, the operating speed is quoted only for OOK. Furthermore, the discussion in this paper is limited to C-band and O-band implementations of SiPh modulators for telecommunication and data communication applications; though SiPh modulators are also reported for mid-infrared (mid-IR) wavelengths^214,215 and for non-telecom/datacom applications.²¹⁵

Research to develop SiPh modulators dates back to the era when it was still a niche choice for photonic integration.^{61,64,88–90,216} Unstrained silicon has a fundamentally zero linear electro-optic effect (the Pockels effect, used in traditional LiNbO3 modulators) due to its crystallographic structure (i.e., centrosymmetric crystal lattice), albeit strained silicon is investigated for high-speed all-silicon modulation.^217–219 This intrinsic property of Si directed the researchers to investigate other optical phenomenons, such as the plasma dispersion effect to realize high-performance modulators in SiPh. This section provides a summary of the plasma dispersion modulators to implement SiPh high-speed modulators.

The refractive index change by the plasma dispersion effect in silicon is larger than the change by the Kerr^61,220 and FK effects.^31,63,141 The plasma dispersion effect is broadband going all the way from telecom to mid-IR wavelengths²²¹ and is relatively temperature independent.^222–225 Furthermore, its implementation is inherently simple, using standard CMOS processing technology.⁸¹ All of these factors made the plasma dispersion effect one of the current mainstream mechanisms to implement efficient high-speed modulators in SiPh.^2,21,35,49 Years of development, led by both industry and academic players, have resulted in the demonstration of plasma dispersion effect-based phase modulators providing losses of <10dB/cm,^{71,80,116,118,121} phase modulation efficiencies Vπ·L<0.3V·cm,^5,69,104 energy consumption as low as tens of fJ/bit,^{18,99,110,111,115} and data rates reaching 100Gb/s^{14,87,122,123} for OOK modulation with minimal influence of temperature on the performance of the plasma dispersion modulators (III–V modulators rely on the QCS effect and are influenced by temperature). It is important to note that all of these specs are not provided simultaneously by a single plasma dispersion modulation implementation.²¹

Plasma dispersion is an electro-refractive effect. The change of free carrier concentration by the movement of charge carriers into or out of a waveguide results in phase modulation of the optical signal.^21,49,61 Phase modulation is translated into intensity modulation by embedding the phase modulator into MZIs,^{15–17,77,79,108,109} RRs,^{18,32,33,62,110,111,119} Bragg reflectors,⁸⁸ Michelson interferometers,^8,121 photonic crystal cavities,^98,112–114 and Fabry–Perot cavities.¹⁰⁰ The three prominent schemes to introduce change in the free carrier concentration are (a) the injection of minority carriers^7,37,61–67 by forward biasing a PIN junction; (b) the accumulation of majority carriers^{12,68,69,69,70} of opposing polarity across an insulating section in a waveguide; (c) the depletion of majority carriers^{9,13,14,16,22,32,33,71,71–87} from a PN junction by reversely biasing it.

Carrier injection is the most efficient plasma dispersion-based modulation scheme.^7,8,37,98 Most of the early demonstrations of high-speed modulators relied on carrier injection of minority carriers by forward biasing a PIN junction built along a waveguide.^{63,89,90,93,216,226} The carrier injection-based modulators provide a high modulation efficiency due to their large diffusion capacitance (typically ∼10pF) when implemented on a waveguide with a small cross section with stronger confinement.^{7,37,65–67,91–94} Carrier injection-based modulators are reported using lateral PIN^63,65,67,92 and vertical PIN configurations (see Fig. 1).^{63,89,93,216,226} Owing to the feasibility for a short phase shifter (∼ a few hundreds of μm), carrier injection modulators typically use lumped driving schemes.^7,65,67 The use of pre-emphasized electrical signaling schemes, in which a fraction of the electrical driver signal has a voltage ≫Vπ of the modulator, boosts the limited speeds of carrier injection-based modulators^67,95–97 with the obvious ramification of increased power consumption. Recent years have seen the development of passive equalization techniques to boost the high-speed operation of the carrier injection modulators.^65,67 The studies have shown that the factors that limit the speed of the carrier injection-based modulation are (a) recombination time of the injected electron–hole pairs and (b) the sum of the electrical driver output resistance and the bulk resistance of the p- and n-doped regions.^7,65,67 The slower of these two effects determines the speed limit of the carrier injection modulators.⁶⁷ Practically, it is possible to improve the bandwidth of carrier injection modulators by embedding an RECE passive equalizer such that RECE=RFCF and CF≫CE, where RE,F and CE,F are the equalizer and the forward resistance and the capacitance, respectively.^7,65,91 Current state-of-the-art carrier injection modulators use this passive equalization to demonstrate >20Gb/s operation.^7,65 Recently, 70 Gbaud operation of a carrier injection MZM was enabled by passive RC equalization.⁷ The modulator has a 37-GHz 3-dB bandwidth using a 0.25-mm long-phase shifter having ∼28dB/cm loss and a Vπ·L of 2V·cm.⁷ One of the best modulation efficiencies of 0.274V·cm has been reported in Ref. 67, where a side-wall corrugated MZM with a bandwidth of 12.5 GHz is operated at 25Gb/s using a finite impulse response filter for frequency compensation necessary for the broadband operation.⁶⁷ The demonstrated modulator had a 250-μm-long phase shifter that exhibited a propagation loss of 5.3dB/mm.

The carrier accumulation plasma dispersion phase shifter requires forward biasing of a metal-oxide-semiconductor (MOS) capacitor.^2,68,101,102 A typical implementation of an MOS (also known as silicon–insulator–silicon capacitor—SISCAP) modulator comprises a vertical slot waveguide with an overlapping stack of a sub-μm-sized thick p-type poly-silicon layer on top of a sub-μm-sized thick n-type c-silicon layer with a few nm (typically <20nm) thick insulating gate oxide layer in the middle (i.e., p-Si/insulator/c-Si vertical slot waveguide)—see Fig. 1(a).^69,101,102 Under accumulation conditions, in which a positive bias is applied to the p-type layer, the vertically confined optical mode overlaps strongly with the highly capacitive gate section.^101,102 This enables <1V·cm modulation efficiency for SISCAP modulators³⁷ with state-of-the-art values as low as 0.2 and 0.16V·cm.^5,68,69 Owing to this higher modulation efficiency, typically ≤0.5-mm-long phase shifters with lumped electrical driving are implemented in SISCAP modulators.^{65,67,69,69,103} The higher capacitance, which leads to higher modulation efficiency, caps the modulation speed of SISCAP modulators.^2,104 The modulation efficiency and speed can be traded-off in SISCAP modulators by adopting the capacitance to a value that can provide either an efficient modulator (i.e., high capacitance) with low Vπ·L or high-speed modulator (i.e., low capacitance) without any degradation of the loss parameter for the modulator (the increase or decrease in capacitance is independent of the doping level and is controlled by the thickness of gate oxide and the relative permitivity of the material used as the gate oxide).^{2,70,101,102,104} Typically, a vertical capacitor configuration is used in SISCAP modulators due to the ease of fabrication.^{5,68,69,69,70,104} It is straightforward to fabricate by the deposition of poly-Si.^2,101 Techniques such as epitaxial lateral growth^101,102 and laser-induced epitaxial growth crystallize the poly-silicon layer to overcome the higher scattering loss due to defects in the lattice.^2,227 Recently, SISCAP modulators using a c-Si/insulator/c-Si in a vertical slot configuration⁷⁰ and in a horizontal slot configuration¹⁰³ without requiring any additional processing steps have been reported [see Fig. 1(b)]. One of the first 40Gb/s Mach–Zehnder SISCAP modulators relied on a vertical capacitor configuration, and it delivered an ER of 9 dB.⁶⁸ The design consisted of a 400-μm-long MOS capacitor with a 6.5-dB/mm loss delivering a Vπ·L of 0.2 cm for the O-band. The same MZM is driven by a CMOS driver to demonstrate its low-power operation for NRZ-OOK and for advanced modulation formats such as PAM-4 using a segmented driver.⁵ A recent demonstration reports high modulation efficiency of 0.16V·cm for 25Gb/s operation using a 60-μm-long phase shifter with a loss of 3.5dB/mm.⁶⁹ The MZM provided one of the best FoMs (α·Vπ·L) of <7dB·V. A microdisk-based SISCAP modulator demonstrated 15Gb/s operation on a small footprint of ∼30μm2 consuming only 55fJ/bit.⁷⁰ In order to deliver >50Gb/s operation using SISCAP modulators, a parametric study has been performed validating the trade-off between high-speed operation, modulation efficiency, and loss.¹⁰⁴

Depletion of electrons and holes by reversely biasing a PN junction built in or around it is the most widely adopted scheme to implement the plasma dispersion-based phase modulator.^{8–11,13–19,22,32,33,71,71–87,106–121} Depletion-based modulators were the first all-silicon modulator to reach speeds of 40¹⁰⁶ and 50Gb/s.¹⁶ The current highest reported data rates of 100Gb/s using OOK modulation were also demonstrated in carrier depletion modulators.^122,123 Carrier depletion phase shifters exhibit a relatively limited capacitance of 0.2 to 0.8pF/mm^21,91 and, hence, a limited modulation efficiency. In order to improve the modulation efficiency, the capacitance can be increased by either shrinking the mode size (determined by the waveguide geometry) or by reducing the width of the depletion region (higher transition capacitance).²¹ The latter requires higher doping concentrations, which leads to higher loss due to free carrier absorption. Over the years, a wide variety of PN junction designs are reported to solve this challenge of optimizing modulator speed, efficiency, and loss at the same time.^{13,15,17,18,22,71–76,78,107–109,116,117,119} Distinct categories of PN junction implementation are as follows:

One of the first 40Gb/s modulators relied on a vertical PIN junction and delivered a modulation efficiency of 4V·cm and phase shifter loss of 18dB/cm.¹⁰⁶ Using a vertical P(I)N junction design, an athermal microdisk modulator with a high modulation efficiency of 250pm/V consuming 0.9fJ/bit while operating at 25Gb/s using NRZ-OOK signaling has been reported.¹⁸ At 25Gb/s, the modulator had an IL of ∼1dB and an ER of 6 dB. One of the highest operating speeds of 60Gb/s with 3.8 dB ER comprised a symmetric PN junction.¹⁷ In this case, the modulator operated in a series of push–pull modes by connecting back-to-back the phase shifter PN junction embedded on both arms of the MZI. The series connection of the two phase shifters halves the net capacitance, and the push–pull scheme limits the chirp of the modulator. The modulator shows 50 GHz bandwidth at 4 V reverse bias and an average modulation efficiency of 3.2V·cm for a 4-mm-long PN junction.¹⁷ In another implementation, a symmetric carrier depletion MZM with sub-1 V drive signal operated at 20Gb/s and delivered a power consumption as low as 200fJ/bit.¹¹⁸ Yet another variant of the symmetric lateral PN junction used the doping compensating technique¹²⁰ to demonstrate a 50-Gb/s MZM comprising a 3-mm-long phase shifter with a modulation efficiency of 1.85V·cm and a loss of 3.9 dB. An RM relying on a symmetric phase shifter configuration delivered a modulation efficiency of 25pm/V at 0 V¹¹⁹ to demonstrate 36fJ/bit energy consumption for 40Gb/s operation on a compact footprint of 0.5mm2. A recent implementation demonstrates a 64-Gb/s MZM, in which the carrier depletion-based phase shifter is implemented through a 200-μm-long photonic crystal slow light structure.¹¹⁴ The modulator used a meandered RF electrode terminated with a 20Ω resistor to provide a bandwidth of 38 GHz by preventing the velocity mismatch of the electrical and optical signals. The modulator required a voltage swing of 5.5 V to provide an energy consumption of 21pJ/bit and provided 4.8 dB ER. The literature also reports various other demonstrations of asymmetric (where there is an offset of the PN junction from the waveguide center) carrier depletion-based phase shifters.^{8,16,72,74,76} For example, a Michelson interferometer with a partially shifted PN-junction provided one of the highest modulation efficiencies of 0.72V·cm at 1 V reverse bias and 4.7 dB loss for a 500-μm-long PN junction with a lumped driving scheme to operate at 40Gb/s through a bandwidth extension scheme involving a resistor parallel to the PN junction.⁸ The asymmetric PN diodes are also exploited by RRs^72,76 to demonstrate low-power (sub-100fJ/bit) and compact depletion modulators. The literature also reports fully shifted PN junctions, where a self-aligning process allows for the alignment of the n-doped region to the edge of the waveguide.^16,74 In one implementation of a fully shifted PN junction, a 50-Gb/s MZM modulator provided a modulation efficiency of 2.8V·cm and an IL of 4 dB for a 1-mm-long phase shifter.¹⁶ By further optimizing the implementation of self-aligned PN junctions,⁷⁴ a demonstration provided an ∼1dB/mm loss without compromising the speed of the modulator. The self-aligning process is further exploited to implement a PIPIN phase shifter with a modulation efficiency of 3.5V·cm for a 0.95-mm-long phase shifter introducing 4.5 dB optical loss when operated at 40Gb/s.⁷⁹ Depletion-based modulators relying on an interleaved phase shifter design have also delivered high-speed operation. A modulator comprising a 3-mm-long interleaved phase shifter reported a high modulation efficiency of 0.62V·cm with a static IL of 2.8 dB while delivering 40Gb/s modulation speed.⁸¹ An RM comprising a ring resonator with a 100-μm radius, and an MZM with a 0.95-mm phase shifter length operated at 40Gb/s speed using an interleaved phase shifter delivering a modulation efficiency and loss of 2.4V·cm and 2.1dB/mm, respectively.³² Apart from standalone implementations of high-speed phase shifters through these three baseline junction configurations and their adaptations, there are reports on junction implementations that rely on combining these three schemes in a single PN junction design.^{9,33,79,85,86} An example is the substrate-removed “zig-zag” phase shifter for an MZM delivering 1.6V·cm modulation efficiency for a 2-mm-long phase shifter with 4.4 dB loss.¹⁴ The modulator provided 55 GHz bandwidth and modulation speed of 90Gb/s.^14,87 Another example includes a phase shifter for 40Gb/s operation of a racetrack RM by combining horizontal, vertical, and interleaved approaches. This implementation delivered a modulation efficiency of 0.76V·cm at −1V reverse bias and a loss of 3.5dB/mm.³³ A recent result demonstrated an RM with a record low modulation efficiency of 0.52V·cm by employing a phase shifter comprising a combination of horizontal and vertical junctions by wrapping the n-doped region over the intrinsic and the p-doped region.⁹ The modulator showed 50 GHz bandwidth and delivered 70fJ/bit energy consumption for 64Gb/s operation.

In the discussion presented above, we have provided a summary of the latest developments made for the improvement of plasma dispersion effect-based modulators. Figure 1(a) represents the three baseline plasma dispersion phase shifter architectures and their respective biasing configuration. Figure 1(b) gives an overview of a variety of other phase shifter configurations for carrier injection, carrier depletion, and carrier accumulation phase shifters.

Table 2 summarizes the typical performance for the injection, accumulation, and depletion-type plasma dispersion modulators. Table 2 also presents the best-reported results for the modulation efficiencies, losses, data rates, and energy consumption. There is an interplay of a vast pool of variables, such as junction type and its actuation mechanism, doping concentrations (typically ranging from ∼1017 to ∼1018cm−3) and their positions, the thickness of the slab waveguide (typically ranging from 1/3 to 1/4 of the guiding silicon thickness), the thickness of the guiding silicon layer (typically <1μm), choice of the passive structure for the translation of phase modulation to amplitude modulation, and the design of the driving electrode (lumped, TW, and segmented), which can influence the performance of plasma dispersion modulators. In many cases, one performance parameter has to be delicately traded-off for another specification. This makes the design of a plasma dispersion modulator, which can meet all of the performance requirements concurrently, far from trivial. Operation of plasma dispersion modulators at speeds >70Gb/s^7,14,122 consuming tens of fJ/bit with high modulation efficiency enabling micrometer-sized footprints has been accomplished. Recent developments have resulted in a co-designed SiPh modulator-driver offering 100Gb/s while preserving a good modulation efficiency¹²² and without any customized fabrication process development. The plasma dispersion modulators deliver sufficient linearity and ER to handle advanced coherent modulation schemes,^{3,55,56,124–126} such as QAM, and amplitude modulation schemes,^{9,17,29,52,66,115,123} such as PAM. Furthermore, the plasma dispersion modulators have shown modest chirp-induced penalty for short- and long-haul links.^3,26,30 The co-design and co-optimization of the driver–modulator design is the key to taking the performance of plasma dispersion modulators to a new level.¹²²

Recent years have seen growing efforts to integrate materials with strong electro-optic effects on SiPh platforms.^{153–192,192–208,228} The integration of new materials provides routes for optically broadband and energy-efficient high-speed modulation with little or no current flow.^{176,181,184,188,202} Furthermore, in some cases, these new materials provide an excellent decoupling between amplitude and phase modulation, which is difficult to achieve with contemporary all-silicon plasma dispersion modulation schemes.^174,178,184

Through the wafer-scale integration of these new material systems with SiPh platforms, one can achieve impressive modulator performance, while at the same time preserving the benefits provided by the mature SiPh technology.¹⁸¹ These new materials can be broadly categorized into the following:

Early results of high-speed modulators by integrating electro-optic materials with the SiPh platforms have shown a lot of promise. Many research teams are attempting to take their maturity and performance beyond the performance of plasma dispersion modulators. The sections below discuss the key modulator implementations by integrating these materials into SiPh platforms.

The FK effect and the QCS effect provide possible routes to implement compact, low-power, and high-speed electro-absorption modulators (EAMs) in SiPh.^{31,143,144,146,149} Under the influence of an applied electric field, the band edge tilts to enhance the absorption coefficient for GeSi bulk material (FK effect) or Ge/SiGe quantum well (QW) structures (QCS effect).^{31,38,145,146} These effects are intrinsically fast through their sub-ps response time.^34,229 The QCS effect is stronger than the FK effect^134,230 due to the strong excitonic effect and discretized density of states. The integration of bulk GeSi or Ge/SiGe QW requires epitaxial growth on SiPh platforms. The FK effect is considered more attractive than the QCS effect in the context of integration feasibility. The latter requires a more complex epitaxial growth process for band alignment and strain balancing.^152,230 A large change in the absorption coefficient (typically ranging from 100 to 1000cm−1 in the C-band¹⁴⁶) enables the implementation of a compact and energy-efficient EAM using the FK effect and QCS effect.^31,230 Both effects rely on band engineering. Therefore, they have limited optical bandwidth (∼30nm for the FK effect and ∼20nm for the QCS effect).^134,149 In one of the first SiPh EAMs, the L-band absorption spectrum is shifted to the C-band by a tensiled 0.8% GeSi alloy.³¹ On a small footprint of only 30μm, the EAM actuated by a vertical PIN diode provided an energy consumption of only 50fJ/bit. This EAM could operate over a 14-nm bandwidth from 1539 to 1553 nm and delivered an efficiency (defined as Δα/αon, where Δα is the absorption difference between ON-state and OFF-state, and αon is the absorption for the ON-sate, in which no electric field is applied) of 2. Despite the limited bandwidth of 1 GHz, this promising demonstration triggered other demonstrations of EAMs.^{20,38,128,133–139,231,232} An evanescently coupled 55μm2 GeSi EAM has been reported on a thick SOI platform (silicon guiding layer thickness of 3μm), actuated by a horizontal PIN junction.¹³⁵ This EAM is butt-coupled to the thick SOI waveguide through an appropriately designed taper.¹³⁵ The modulator has an efficiency of ∼1. This EAM operated in the C-band and provided 28Gb/s operation with 60fJ/bit energy consumption.¹³⁵ Literature also reports an L-band Ge EAM and a GeSi C-band EAM with an electrical 3-dB bandwidth of ∼50GHz for an optical bandwidth of 30 nm.¹³⁶ In this case, the EAM is butt-coupled to a silicon waveguide through a tapered waveguide section. The Ge or GeSi alloy, which is grown on a silicon recess, has a lateral PIN diode. A highly doped contact on the silicon layer biases the EAM. The modulator consumed 12.8fJ/bit energy at modulation speeds of 56Gb/s and provided an FoM of <1. Using this GeSi EAM, a digital to analog converter-less and digital signal processing (DSP)-free 100Gb/s transmission over a 500-m-long single-mode fiber using NRZ-OOK signaling has been reported.^20,137

There are various reports on the implementation of the Ge/SiGe QCS effect-based modulators.^141–152 In one case, a Ge/SiGe QW modulator is butt-coupled to a thin SOI rib waveguide in such a way that the QWs and the rib waveguide are vertically aligned with each other to have maximum overlap.¹⁴⁷ The active section of the modulator, which comprised 15 pairs of Ge/Si0.15Ge0.85 QWs, is driven by a vertical PIN diode. The 8-μm2 modulator showed 3.5 GHz bandwidth by using a 1-V voltage swing (5.5 to 6.5 V) to provide ∼3dB of ER. In another implementation, Ge QWs are epitaxially grown at temperatures below 450°C on the silicon-rich Si1−xGex low-loss waveguide platform.¹⁴¹ The x concentration is selected such that it simultaneously ensures low indirect bandgap absorption and a sufficiently small lattice mismatch with the QWs. The 400-μm2 modulator with 6.3 GHz bandwidth and 0 to 3 V swing provided ∼4dB ER and 3 dB IL.

The FK effect or QCS effect electro-absorption amplitude modulators allow for the implementation of coherent I/Q^231,232 and advanced amplitude modulation schemes such as PAM-4.²³³ These modulation schemes are a promising solution for short- and long-reach links because they are spectrally efficient while being tolerant of optical fiber transmission impairments.^54,55,231 The implementation of an EAM-based I/Q or PAM-4 modulators relies on balanced MZIs with an EAM and a fixed phase-shift in the arms of the MZI.^231–233 Typically, an unequal splitting of the optical carrier is required by the splitter of the MZI to achieve the appropriate eye-opening for PAM-4 or constellation for the I/Q signal.²³³ Literature reports an EAM-based PAM-4 modulator by integrating an EAM in each arm of a two-port balanced MZI.^231–233 In this modulator, a fixed phase shift is provided by embedding a thermal phase shifter in one of the two arms of the MZI. The device modulated a 128-Gb/s PAM-4 signal without requiring any DSP. Furthermore, there are reports on the demonstrations of modulating QPSK and 16-QAM signals using EAM-based coherent modulators.^231,232 The energy consumption of these modulators takes a hit due to the power-hungry thermo-optic effect required to provide a fixed phase shift in the arms of the MZIs.

Figure 2 represents the typical cross sections of FK EAMs and QCS effect-based EAMs in SiPh. This figure also summarizes the integration routes of these materials with SiPh. Table 3 summarizes the typical and the best-reported performance for the (silicon-)germanium modulators. This table covers the performance of the FK effect-based EAMs as well as QCS effect-based EAMs. The FoM for both FK and QCS effect-based EAMs is typically <2, although QCS effect-based modulators can potentially deliver better FoM due to their stronger excitonic absorption peaks.^144,149,152 In most cases, FK-effect-based EAMs operate either in the L-band or C-band.^150,152 It is possible to tune the operating wavelength of the QCS effect modulators by a bias voltage.^150,152 Therefore, the QCS effect modulators can operate in the O-band as well. In terms of high-speed operation, FK effect-based EAMs have reached line rates of up to 100Gb/s for OOK modulation schemes,¹³⁷ whereas the QCS effect-based modulators have not yet demonstrated such a high speed of operation.

Ferroelectrics are materials possessing spontaneous polarization that can be altered by the application of a sufficiently strong electric field.²³⁴ As ferroelectric materials are non-centrosymmetric, they exhibit a Pockels effect.²³⁴ Many ferroelectrics have large Pockels coefficients compared to non-ferroelectric inorganic materials. The dominant contribution to the Pockels coefficients in these ferroelectrics tends to come from lattice vibrations rather than electron movements.^234,235 Because of their strong Pockels effect, the integration of ferroelectric materials such as LiNbO3,^167–174 barium titanate (BaTiO3),^175–182 and lead zirconate titanate (PZT)^183–185 with SiPh provides a promising route for the implementation of efficient, high-speed modulators.^2,175,228 A variety of integration techniques have been explored, such as molecular beam epitaxy (MBE),^176,178 pulsed laser deposition,^236,237 chemical solution deposition (CSD),²³⁸ RF magnetron sputtering,²³⁹ metal organic chemical vapor deposition,²⁴⁰ and atomic layer deposition.²⁴¹ The two key requirements for integrating ferroelectric materials with SiPh are to ensure a high electro-optic coefficient and low propagation loss.¹⁸² The following sections will review the latest developments to implement high-speed modulators by integrating ferroelectric materials with SiPh.

LiNbO3 is a widely used material for modulators and provides high-speed devices with a bandwidth of >35GHz.^27,242,243LiNbO3 modulators have been used for a long time in fiber-optic networks.^242,244 These devices are based on low-index-contrast waveguides in bulk LiNbO3,²⁴² in which the index contrast between the core and cladding of a LiNbO3 waveguide is provided by the diffusion of Ti or Li ions.²⁴² The weak optical guiding resulting from the low-index-contrast requires the electrical contacts to be placed far away from the waveguides, resulting in typical Vπ·L of 10 to 20V·cm.²⁴² Recent demonstrations of high-index-contrast LiNbO3-on-insulator (LNOI) modulators have shown energy-efficient operation (a few tens of fJ/bit energy consumption) with 100-GHz bandwidth, ∼2V·cm modulation efficiency, and <0.5dB loss for a device that is a few millimeters long (typically 2 to 20 mm long).^{243,245–247} These thin-film LNOI modulators rely on LiNbO3 layers of a few hundred nanometers thickness, which are bonded on top of SiO2.²⁴³ Several approaches have been reported in the last two decades to integrate LiNbO3 with silicon at room temperature^245,248,249 and to demonstrate high-speed modulation by bonding LiNbO3 with silicon^{167–169,250} or silicon nitride.^170–172 The Pockels coefficient of integrated LiNbO3 is comparable to that of bulk LiNbO3,²⁴⁹ but most of these integration attempts have resulted in modulator bandwidths that are smaller than the all-silicon modulator implementations.¹⁷³ Recently, a tremendous boost in the speeds (not yet the overall performance) of such modulators has been reported.^173,174 For example, for a 5-mm long TW-MZM that was realized by direct oxide bonding of thin-film LiNbO3 on top of a silicon waveguide, a bandwidth of 100 GHz and a loss of 7.6 dB have been reported.¹⁷³ The modulator had a limited efficiency of 6.7V·cm due to <100% vertical coupling of light into the unpatterned LiNbO3 membrane on top of the silicon waveguide. In another implementation, the complete coupling of light from a silicon waveguide to a patterned LiNbO3 rib waveguide through a vertical adiabatic coupler resulted in a modulation efficiency of ∼2.5V·cm.¹⁷⁴ With a 2.5-dB loss, the 5-mm-long LiNbO3 phase shifter embedded in an MZI and driven by a TW electrode in the push–pull configuration provided 70 GHz bandwidth and enabled single-lane 100Gb/s NRZ-OOK transmission. Unlike the other reported implementation,¹⁷³ the integration of LiNbO3 with the SiPh PIC was carried out using BCB-bonding. Monolithic integration of LiNbO3 with SiPh PICs has yet to be reported.

In recent years, there has been a lot of interest in the integration of BaTiO3 with SiPh.^175–182BaTiO3 is a promising material for integration with SiPh because of (a) its high Pockels coefficient [in bulk form, LiNbO3 has a Pockels coefficient of ∼30pm/V,^27,175,249 whereas BTO offers one of the highest Pockels coefficients (>1000pm/V) in bulk form^176,251]; (b) feasibility of integration with a silicon substrate;²⁵² (c) capability to offer high-speed modulation;²⁵³ (d) high chemical and thermal stability of BTO; and (e) feasibility of low-loss hybrid BTO-Si passive circuitry.^178,179 The integration of BTO with SiPh relies on epitaxial growth enabling integration compatible with 200 and 300 mm silicon wafers.¹⁸¹ The large lattice mismatch (∼27%) between BTO and Si makes the integration of the two materials far from trivial.²⁵⁴ A SrTiO3 (STO) buffer layer of a couple of nanometers thickness reduces this lattice mismatch to <2%.²²⁸ Among the various possible routes to grow BTO on SiPh platforms, MBE is known to provide the highest (>900pm/V) Pockels coefficient in BTO thin films.¹⁷⁶ In one approach, the BTO growth is done directly on an SOI wafer.¹⁷⁷ Afterward, an amorphous silicon layer with a thickness equal to the thickness of the c-Si layer of the SOI wafer is grown. The resulting heterostructure is patterned lithographically to form a horizontal slot waveguide. Stronger guiding of the optical signal in the low-index BTO layer (nBTO∼2.38) enhances the overlap of the optical and electrical field in the BTO layer. This approach provided a Pockels coefficient of 213±49pm/V.¹⁷⁷ A ring-based modulator with a modulation efficiency of 1.5V·cm and a bandwidth of 4.9 GHz has been reported, where the limited bandwidth is associated with the parasitic electrical effects.¹⁷⁷ A key challenge faced by the integration approach mentioned in Ref. 177 and other integration approaches^176,178 is the high propagation loss (>40dB/cm).^{176–178,255} The origin of this high loss is attributed to the absorption of hydrogen by the STO seed layer during the fabrication of the slot waveguide.¹⁷⁸ Furthermore, it has been shown that a loss of 6dB/cm in a strip-loaded horizontal slot waveguide can be achieved by low-temperature annealing.¹⁷⁸ Apart from the aforementioned approach, where BTO is grown directly on an SOI wafer,¹⁷⁷ another possible approach entails the formation of a BTO/Si heterostructure by bonding an SOI wafer with a BTO layer grown on a host SOI wafer.^176,179 For such a bonded BTO/Si heterostructure, a Pockels coefficient of 923pm/V has been reported (30 times larger than that of a LiNbO3 structure).¹⁷⁹ In one implementation, a plasmonic MZM with a 10-μm-long BTO/Si phase shifter with 25 dB loss delivered 72Gb/s and required a Vπ of 14.5 V (modulation efficiency of 0.0145V·cm).¹⁸⁰ The modulator operated successfully for temperatures up to 130°C. Furthermore, a wafer-level bonding of Si/BTO heterostructures on the back-end-of-line (BEOL) dielectric layer of an advanced SiPh platform has been reported.¹⁸¹ This integration led to the demonstration of an MZM with a limited bandwidth of 2 GHz (mostly limited by the electrical bandwidth of the electrodes).¹⁸¹ The modulator had a modulation efficiency of 0.2V·cm and loss of <6dB for a 1-mm-long phase shifter.^181,182

PZT is an orthorhombic lead-containing ceramic ferroelectric material.^256,257 Though BTO has a large Pockels coefficient, its use in SiPh requires complex and expensive integration technology.^{178,181,182,255} As compared to BTO, PZT has a smaller Pockels coefficient, but it can be deposited by a low-cost and simple CSD method.^185,257 Ultra-thin lanthanide-based intermediate seed layers enable the deposition of high-quality thin (up to 200 nm) PZT layers on silicon,²⁵⁷ with Pockels coefficients up to 150pm/V.²²⁸ PZT thin films have been integrated with silicon nitride photonic circuits. An RM with a bandwidth of 33 GHz and an MZM with a bandwidth of 27 GHz have been demonstrated,¹⁸⁴ with a modulation efficiency of 3.2V·cm and a loss of 1dB/cm. The PZT layer was deposited on a planarized silicon nitride PIC (i.e., the PIC consisted of silicon nitride strip waveguides with oxide bottom and side cladding, while the top cladding was air). The PZT thin film was poled by applying a voltage of 60 to 80 V for 1 h.¹⁸⁵

Figure 3 represents the typical cross sections of ferroelectric modulators in SiPh. This figure also summarizes the integration routes of the respective materials with SiPh. Table 4 presents a summary of the typical and the best-reported modulation efficiencies, losses, data rates, and energy consumption for the ferroelectric-based modulators in SiPh. For all three types (i.e., LiNbO3, BTO, and PZT), modulation speeds ≥40Gb/s have been demonstrated. LiNbO3-based modulators are outstanding in terms of low-loss operation and have delivered speeds of ∼100Gb/s. In terms of modulation efficiency, BTO-based modulators have shown a lot of promise with typical reported values of <0.5V·cm.

Organic electro-optic materials are typically based on π-conjugated molecules with strong electron donor and acceptor groups at the ends of the π-conjugated structure.^188,258 The delocalized electrons associated with this π-conjugated structure are confined by a strongly asymmetric potential well, giving rise to large Pockels coefficients.^188,258 In contrast to the ferroelectrics discussed above, the Pockels effect in these organic electro-optic materials is mostly electronic in nature and can thus provide an ultra-fast response.^{188,190,228,258} Organic electro-optic materials can be processed at room temperature using inexpensive solution-based techniques, whereas the growth of inorganic Pockels materials typically requires high temperatures and/or the use of vacuum equipment.^{177,182,188,255,258}

A combination of organic materials with SiPh devices has resulted in the so-called silicon-organic hybrid (SOH) class of modulators.^{186–188,197} The underlying organic materials in SOH modulators are designed to provide high electro-optic activity with experimentally demonstrated values of n3r>2000pm/V.¹⁹⁰ This results in modulators with a low-loss,¹⁹⁴ a low-chirp parameter of α=−0.02,¹⁹³ and a high modulation efficiency (Vπ·L≈0.03V·cm),¹⁹⁰ along with a high modulation bandwidth (>100GHz)^191,259 and low-energy consumption of a few fJ/bit.^192,228 With π-voltages below 1 V, SOH modulators may be directly driven by standard CMOS circuits without the need for any additional drive amplifiers,²⁶⁰ thereby opening a path toward greatly reduced power dissipation in highly integrated transceivers. For the fabrication of SOH devices, the organic electro-optic material is deposited onto fully processed SiPh chips in a postprocessing step, which requires removal of the BEOL layer stack of a standard SiPh process flow.²²⁸ SOH modulators rely on slot waveguides,^188,191 which guarantees enhanced interaction between the highly confined optical mode in the slot waveguide and the organic electro-optic material filling the slot. To pole the electro-optic material after deposition, the devices are heated, and a DC voltage is applied across the slot waveguides. This non-volatile poling transforms the random orientation of the electro-optic chromophores in the slot into an acentric orientation, thereby leading to a non-zero Pockels coefficient. After cooling down the device to ambient temperature, the voltage can be removed, and the poled electro-optic material preserves its orientation. SOH modulators have shown exemplary performance over the years. Some key results are reported in Refs. 189–191, 193, 196, and 261. The main skepticism faced by the SOH modulators comes from their long-term photochemical and photothermal stability. Recent research results reported in Refs. 195 and 262 have shown long-term stability of SOH devices following Telecordia standards. These early experiments have shown that in storing the SOH modulators at elevated temperatures of 85°C, the degradation of the electro-optic activity converges to a stable level where the Vπ·L-product increases by <15% after 2400 h.

The limitations of poled organic materials can be overcome by substituting them with crystalline organic materials, which are inherently resilient to higher temperatures with good photochemical stability and do not require any poling.¹⁹⁹ The electro-optic coefficients of the organic crystals are smaller (typically n3r<500pm/V)²⁶³ than those of the poled organic polymers.¹⁹⁰ The integration of organic crystals requires growth to functionalize the silicon waveguides.^198,199 An N-benzyl-2-methyl-4-nitroaniline crystal has been deposited on an SOI waveguide to demonstrate an MZM that could operate at 12.5Gb/s with a modulation efficiency of 1.2V·cm and a loss of 16 dB in the phase modulator with a 1.5 mm length.¹⁹⁹

Typically, the analogue bandwidth of current SOH modulators is limited to ∼25GHz by the finite conductivity of the doped silicon slabs that provide an electrical contact with the rails of the slot waveguide.¹⁹⁰ These bandwidths may be improved to more than 100 GHz by exploiting optimized waveguide designs that rely on multi-step doping profiles.²⁵⁹ To that end, doped silicon slabs have also been replaced by BTO to feed the modulating RF signal to the EO material in the slot waveguide.²⁰⁰ This so-called capacitively coupled SOH MZM delivered a bandwidth of 65 GHz, which was sufficient to produce a 100-Gb/s OOK signal.²⁰⁰ The modulator had a modulation efficiency of 0.13V·cm; the loss from the phase shifter section is not reported. Yet another variant of the SOH modulator is the plasmonic-organic hybrid (POH) modulator,^{258,264–269} where the silicon slot waveguide is replaced by a metal plasmonic slot waveguide to provide even higher optical confinement.^258,265 The deposition of an organic electro-optic material on the metal slot waveguide leads to compact high-speed devices with bandwidths of hundreds of gigahertz.^266,270 POH modulators are compact with typical lengths of a few microns and offer a low-energy consumption of a few tens of fJ/bit along with ultra-low voltage-length products of typically ∼0.005V·cm²⁵⁸—nearly an order of magnitude lower than those of classical SOH modulators. The obvious limitation of the POH modulators is the high propagation loss in the plasmonic slot-waveguide section with typical values of 200dB/mm.^200,258 Despite the high efficiencies, this leads to substantial π-voltage-loss-products of the order of 10V·dB.¹⁸⁸

Table 4 provides a summary of typical modulation efficiencies, losses, speeds, and energy consumption for organic electro-optic modulators in SiPh. When compared with the other Pockels modulators, the organic electro-optic modulators provide the best modulation efficiency with a reported value of 0.032V·cm.¹⁹⁰ Organic electro-optic modulators have shown up to 100Gb/s¹⁹⁶ operation and sub-fJ/bit energy consumption.¹⁹²Figure 3(d) represents the typical cross section of organics modulators in SiPh.

The integration of III–V semiconductors with SiPh platforms is a well-established research field, as it provides a viable route for the integration of laser sources with SiPh PICs.^271–275 Direct bonding,²⁷² BCB-aided adhesive bonding,²⁷¹ III–V/Si flip-chip bonding,²⁷⁵ and transfer printing²⁷⁴ are some of the most established bonding mechanisms for III–V/Si integration. Demonstrations of III–V integration with advanced SiPh platforms have been reported.^276,277 Demonstrations of epitaxial growth of III–V on Si have also been reported despite the large lattice mismatch between Si and III–V (for example, InP has 8% lattice mismatch with Si).^278,279 The integration of III–V on silicon also provides a route for the implementation of efficient high-speed modulators because they provide (a) a large electron-induced refractive-index change, (b) a high electron mobility, and (c) a low carrier-plasma absorption.¹⁶⁶

The reported implementations of III–V/Si modulators typically use a MOS (SISCAP) modulator architecture.^161–163 In these modulators, n-doped III–V material, for example, InGaAsP, is bonded to a p-doped Si layer with an insulating layer separating the two.^161–163 This results in a SISCAP architecture, where the insulating layer acts as a capacitor. A change in the refractive index takes place due to the charge accumulation due to the biasing of the doped silicon and III–V layer. According to Drude’s model, the plasma dispersion induced refractive index change −Δn is inversely proportional to the effective masses of the electrons and holes.^61,89 As compared to the electron mass of Si, the lighter electron mass of n-doped III–V results in a 17 times stronger electron-induced refractive index change.¹⁶¹ Furthermore, an n-doped III–V provides a larger ratio of −Δn to Δk, where Δk is the change in the absorption coefficient,¹⁶¹ resulting in a phase modulator with reduced spurious intensity modulation (“pure-phase” modulation). Similarly, p-doped silicon provides larger −Δn/Δk than p-doped III–Vs for carrier concentrations below 1019cm−3.^161–163 Therefore, a hybrid III–V/Si MOS modulator brings the best of III–V and Si together. Consequently, this approach offers modulators with a 3 to 5 times better modulation efficiency compared to all-silicon MOS modulators and a phase shifter loss that is significantly lower than in all-silicon MOS modulators.^161,163

In one report, an InGaAsP/Al2O3/Si MOS structure provided a modulation efficiency of 0.047V·cm and a phase shifter loss of ∼19.4dB/cm.¹⁶¹ In another report, an InGaAsP/SiO2/Si MOS structure provided¹⁶³ a modulation efficiency of 0.09V·cm with an absorption loss of 26dB/cm in the phase shifter. Both demonstrations show an impressive modulation efficiency. But, the electro-optic cut-off frequency is rather limited due to the large oxide capacitance in the carrier accumulation mode of operation.^161,163

One possible approach to break the trade-off between the high-speed operation and high modulation efficiency banks on operating the III–V on Si MOS phase shifter in reverse bias.^165,280 The III–V on Si MOS modulator that is operated in reverse bias is architecturally similar to the forward biased III–V on Si modulator.^{161,163,165,280} For a given thickness of the oxide layer, the reverse biased III–V on Si phase shifter has a reduced depletion capacitance as compared to the accumulation capacitance.^165,280 The combined plasma dispersion effect and FK effect in the reverse biased III–V on Si MOS phase shifter ensure a modulation efficiency that is comparable to that in the accumulation mode while ensuring a low-loss operation due to the limited hole density in the n-doped InGaAsP layer of the modulator.^165,280 A III–V on Si MOS modulator with a 250-μm-long phase shifter and a 9-nm-thick oxide layer has been reported with a modulation efficiency of 0.11V·cm.¹⁶⁵ As compared to the carrier accumulation MOS, the reverse biased MOS provides a 3.3× lower depletion capacitance, hence paving the way for a 3.3× improvement in modulation bandwidth and energy consumption per bit. The implementation of III–V on Si MOS modulators is not limited to MZIs only.¹⁶⁴ By implementing a III–V on Si MOS modulator in a racetrack configuration (implementing III–V on Si MOS modulators in a ring is challenging due to the long taper length needed for the III–V on Si hybrid waveguide), the energy consumption of a 40-GHz (3-dB bandwidth) modulator has been improved by a factor of 3.7 as compared to the III–V on Si MZMs, while delivering a modulation efficiency of 0.064V·cm.¹⁶⁴ The 100-μm-long phase shifter section of the modulators comprised a thin aluminum oxide gate layer separating the n-type III–V and the p-doped Si layer. Recently, a proof-of-concept, taper-less integration of a few tens of nanometers thick III–V layer with Si has been demonstrated to reduce the footprint and loss of the device²⁸¹ without compromising the modulation efficiency of the III–V on Si MOS capacitor.

Apart from phase modulators, EAMs have been reported by the bonding of III–V QWs with silicon.^140,282 These EAMs use a PIN architecture, where the application of a bias voltage changes the optical absorption of the intrinsic multiple QWs due to the QCS effect.²⁸² A segmented electrode, comprising a low-impedance active modulation section cascaded by a high-impedance passive section, provides high-bandwidth operation while ensuring low reflection and good modulation efficiency.¹⁴⁰ A >67GHz bandwidth hybrid III–V on silicon EAM has been demonstrated using this scheme.¹⁴⁰ In the O-band, this modulator has an IL of 4.9 dB, and it shows >9dB of dynamic ER for 50Gb/s operation over a 16-km-long optical fiber link.

Figure 4 represents the typical cross sections of III–V on Si phase modulators and EAMs. This figure also summarizes the routes to integrate III–V on Si. Table 5 presents a summary of the typical and best-reported values for the modulation efficiencies, losses, data rates, and energy consumption for forward and reverse biased III–V on Si phase modulators as well as for III–V on Si EAMs. III–V on Si modulators have shown an impressive modulation efficiency and low-loss operation, but so far experimental demonstrations of high-speed operation (∼100Gb/s) remain elusive.

Atomically thick 2D materials, the most prominent graphene, have received considerable interest in the last decade to implement efficient modulators by integration with Si/SiN PICs.^{1,153–160,201–208} The implementation of graphene-based modulators in SiPh involves the transfer of the graphene layer on a Si or SiN waveguide.¹ Mature and wafer-scale chemical vapor deposition is the widely adopted scheme to deposit graphene on a variety of metallic and dielectric substrates.^283–285 The transfer of the graphene layer on a Si or SiN waveguide is performed after delaminating graphene from the substrate while adding a polymer carrier to handle the graphene film.^154,159 Recently, microtransfer printing technology has shown promising results to transfer graphene to the target substrate in an automatic fashion.¹⁵⁹ A single graphene layer, which is only a fraction of a nanometer thick, can absorb 2.3% of light²⁸⁶ with a wavelength ranging from the visible to THz wavelengths (graphene is a gapless material). Furthermore, graphene has a very high electron mobility of >100,000cm2·V−1·s−1 (silicon has a carrier mobility of 1400cm2·V−1·s−1).^228,287,288 These two properties enable the implementation of high-speed electro-absorptive¹⁵³ or electro-refractive²⁰¹ modulators through the integration of graphene with SiPh ICs.

The two prominent types of graphene modulators in SiPh comprise a SiPh rib waveguide, which is either loaded with a single layer or a double layer of graphene^153,156 (see Fig. 5). The operating principle of these modulators relies on the charging and discharging of a capacitor. A single-layer graphene modulator comprises a graphene loaded Si rib waveguide with a doped slab region.¹⁵³ A few nm thick dielectric spacer, such as SiO2, Al2O3, or hBN, separates the waveguide and the graphene layer to form a capacitor.¹ The first reported graphene modulator¹⁵³ had a 4-dB ER and relied on a single-layer approach. The 25-μm2 EAM provided 1 GHz operation, limited by the RC time constant, and an electroabsorption modulation of 0.1dB/μm.¹⁵³ The absorption in graphene is dependent on the bias voltage.¹ This voltage tunes the Fermi level of graphene to enable or inhibit the interband transitions by excitation of electrons through the incident photons.^201,289 This change in absorption due to carrier accumulation in the graphene sheet results in modulation of light propagating in the waveguide. The doped Si waveguide in the single-layer graphene modulator is a source of loss, and it requires an implantation step.^153,156 Double-layer graphene modulators remedy these limitations.¹⁵⁶ They are separated by a dielectric layer that resides on top of a completely passive waveguide. (This architecture does not require any doping of the waveguide slab.)¹⁵⁶ The first double-layer graphene modulator reported 1 GHz bandwidth and an electroabsorption modulation of 0.16dB/μm.¹⁵⁶ The 40-μm-long modulator provided a 6-dB ER and required a 5-V driving voltage. One of the first multi-Gb/s graphene-silicon modulators¹⁵⁷ relied on a single-layer graphene architecture that provided a 5.2-dB ER for a 50-μm-long EAM operating at ∼2.5V. At 10Gb/s, the modulator provided a modulation efficiency of 1.5dB/V. The ON state loss is 3.8 dB with a power consumption of about 350fJ/bit. Another result reports a double-layer graphene RM¹⁵⁸ using a critical coupling effect in a 40-μm SiN ring. In this implementation, a 50-nm thick Al2O3 layer separates the two graphene layers. By biasing, the change in the transparency of the graphene capacitor led to the modulation of the optical signal. The reduced capacitance resulting from the thick spacer between the graphene layers led to a modulator showing 30 GHz bandwidth with a modulation efficiency of 1.5dB/V but required a −30-V DC bias and 7.5Vp−p.¹⁵⁸ The modulator could transmit 22Gb/s while consuming 800fJ/bit. A positive chirp of a graphene EAM is an interesting property.¹⁶⁰ This feature is useful in compensating for the chromatic dispersion in an optical transmission link for the transmission of 10-Gb/s without any optical signal to noise ratio penalty.¹⁶⁰

Apart from amplitude modulation, it is also feasible to use graphene for phase modulation.^201–208 Phase modulation is more exciting for the encoding of complex modulation formats such as QPSK. A single-layer graphene architecture, in which the graphene layer is separated from the silicon waveguide by a thin dielectric spacer layer, acts as a phase modulator by shifting the Fermi level of graphene.^{201,205,207,290,291} A single-layer graphene phase modulator of 300μm length is integrated on a Si MZI and showed a modulation efficiency of 0.28V·cm. The modulator delivered an ER of 35 dB, an EO bandwidth of 5 GHz, and a loss of 236dB/cm.²⁰¹ With a 2-Vp−p drive, the transmission of 10Gb/s over a 50-km fiber span is demonstrated. The higher loss in the modulator is associated with the suboptimal transfer of graphene.²⁰¹ The quality of graphene, which is typically determined by the scattering time and is related to the carrier mobility, plays a detrimental role in the performance of graphene modulators.¹ Typically, scattering time of >100fs is required to implement energy-efficient graphene modulators with a high ER.¹ Achieving 10,000cm2·V−1·s−1 graphene carrier mobility, which roughly corresponds to 300 fs scattering time, upon the complete modulator integration process would be ideal. Accomplishing this would lead to extremely low IL.²⁸⁹ In perspective, the combination of low IL in a dual-layer graphene structure, the high graphene electro-absorption, or electro-refractive efficiency would lead to a modulator with a good FoM.¹ It is important to mention that graphene can be used for detection,¹ and this makes graphene suitable for a telecommunication platform on a passive integrated circuit operating in several telecommunications bands.

Inspired by the promising results shown by graphene, other 2D materials such as tungsten disulfide (WS2) and molybdenum disulfide (MoS2) have also been integrated with SiPh. Results show phase change relative to the change in absorption for these materials at telecom wavelengths when integrated with SiN.²⁰⁴

Figure 5 represents typical cross sections of a single layer graphene modulator (a) and a double layer graphene modulator (b) in SiPh. Table 6 provides the current state of high-speed phase and amplitude modulators in SiPh by integrating 2D materials. The currently reported performances of 2D modulators, in particular graphene modulators, are mostly limited by technological immaturity and not by conceptual limits. Graphene requires more development to reach the best performances. More specifically, graphene needs to preserve the high carrier mobility after the integration in the photonic circuit. This preservation is not the case yet. By achieving a high level of technological maturity, graphene has the potential to provide devices that can work over a wide range of wavelengths due to its gapless band structure, miniaturized devices due to its large EO efficiency, feasibility for BEOL integration with the SiPh fabrication process, and a possibility to offer a complete platform for communication applications, as graphene possesses properties for efficient modulation and detection.

Years of academic and industrial R&D have enabled a tremendous boost in the performance of plasma dispersion phase modulators. Operation at speeds of 100Gb/s with high modulation efficiency is now feasible. Plasma dispersion modulators can operate with a sub-1-V drive signal consuming a few tens of fJ/bit in a μm-sized footprint. Demonstrations of advanced modulation schemes such as PAM-4, PAM-8, QPSK, 16-QAM, and 32-QAM have been reported in plasma dispersion modulators. The modulators have an insignificant chirp-induced penalty for long- and short-haul links. A closer synergy through co-design and co-optimization of electronic–photonic modulator design is the key to take plasma dispersion modulators beyond 100Gb/s operation in an efficient manner. The last five years have seen a new paradigm in achieving higher speed and lower loss operations of SiPh modulators by integrating materials such as ferroelectrics, organics, SiGe alloys, III–Vs, and 2D materials. Unlike plasma dispersion modulators, which are commercially deployed and have a high manufacturing readiness level (MRL) of >8, these new modulators are still in the early R&D stage (MRL <7) and are currently not included in any commercial product.

Electro-absorption modulation by (silicon-)germanium FK or QCS devices allows for micrometer-sized footprints. FK-EA modulation provides 100Gb/s operation, but these modulators have a limited ER as well as limited operating optical bandwidth. QCS-EA modulation has integration challenges with SiPh. Moreover, their operating speed is <10Gb/s. LiNbO3, BaTiO3, PZT, graphene, and organic thin films on Si PICs offer high Pockels coefficients, paving the way for linear and chirp-free high-performance SiPh modulators. Early results reported in the literature indicate that promising performance is achievable by integrating thin-film LiNbO3, BaTiO3, or PZT materials on Si. The integration of organic materials on Si has also shown 100Gb/s operation and impressive modulation efficiency. Progress has been reported to address the lifetime issues, but more development is necessary before the commercial launch of high-speed modulators relying on organic electro-optic materials. Bringing together III–V and Si brings the best out of III–V and Si to implement MOS-type phase modulators and EAMs. The product of Vπ and phase shifter loss is impressive for the III–V on Si phase modulators, but high-speed operation remains elusive. (Theoretically, they have shown impressive results.) Integrating 2D materials, in particular graphene, with SiPh has the potential to provide a complete solution for modulation and detection at high speed. Multi-Gb/s operation has been reported for the graphene phase and amplitude modulators. A technique to preserve the high carrier mobility after integration in the photonic circuit is essential to unleash the true potential of graphene modulators in SiPh.

To conclude, modulator schemes that rely on the integration of efficient electro-optic materials with SiPh are gaining rapid technological maturity. They provide an additional route to fulfill the compelling needs of bandwidth, loss, and linearity at low-power consumption and cost for future high-speed optical links. It is important to note that the modulators based on the integration of efficient electro-optic materials with SiPh will not necessarily cannibalize or replace the plasma dispersion modulators. It is a way to overcome bandwidth and loss limits that stem from the physics of the plasma dispersion effect. However they will come at the cost of a more complex fabrication process. For this reason, the optimized solution will depend on the application.