The development of photonic integrated circuits has historically been driven by communications. However, emerging markets and opportunities require platforms that can operate over a broader wavelength range, offer additional capabilities, or improve performance. This paper provides a perspective on these emerging opportunities and the most promising approaches for addressing them, with a focus on platforms that support on-chip light generation and amplification.
INTRODUCTION
Photonic integrated circuits (PICs) have been developed for more than 30 years,1 primarily driven by communications to enable dramatic data rates increase while enabling continuous price (per bit) reduction. Much of the development was targeted to address the needs around 1550 nm, primarily driven by telecom industry, and to address the needs around 1310 nm, primarily driven by datacom industry. Both wavelengths were addressed by indium-phosphide (InP) platform.1 There has been research in gallium-arsenide (GaAs) integrated platform,2 but commercial deployment has been limited. A significant part of optical connectivity is still provided by the GaAs material system, typically around 850 nm wavelengths, but, here, the level of integration is low, and most systems leverage VCSELs (vertical-cavity surface-emitting lasers) and assembled components.3
In the recent decade, we have seen tremendous activity in silicon photonics4,5 with the key driver being the advantage of lower loss waveguides (compared to InP) and excellent uniformity, primarily resulting from the high quality and mature processing of the silicon CMOS ecosystem. Silicon PICs enable high-performance, complex systems for communications and, more recently, longer wavelength LIDAR systems but are limited in wavelength of operation (with silicon bandgap of 1.12 eV or ∼1100 nm), power handling limitation (two photon absorption), the lack of on-chip sources (as silicon is an indirect bandgap material), and fiber coupling efficiency (due to high confinement and small mode area of standard silicon platform). The first three limitations are inherent to the silicon material platform, while the fourth is a result of both the material properties and typical geometrical properties (and lithography limitations) of the platform.
There has been a lot of activity to address third limitation of lacking the on-chip sources,6 either by advanced packaging or co-packaging of multiple photonic chips, e.g., silicon chip and InP chip, or by integration of multiple materials such as in a heterogeneous process where InP is integrated directly on silicon wafer before co-processing,7–9 or developing direct growth on silicon.6 The next generation of connectivity systems is moving from pluggables to on-chip optics, and the race to develop and commercialize them is on. These systems will need to support more wavelengths (including multi-wavelength sources), higher bandwidths, and higher modulation formats to push from 112 to 224 Gbps and beyond, with the main requirements being density and energy per bit.10
BEYOND COMMUNICATIONS
While communications are a well-established market with clear roadmaps for increasing capacity and density (even if technology path to meet future requirements is not known yet), there are many emerging applications that can benefit from PICs due to their potential for SWaP (size, weight, and power) improvement, cost reduction, scalability, and performance improvement. These applications include augmented/mixed/virtual reality (AR/MR/VR), quantum computing, sensing and timing,11 positioning, navigation and timing, radio-frequency systems, computer vision systems, sensing, healthcare, and others. Each of these applications has specific requirements, often significantly different than the datacom/telecom products/markets, which necessitate the development of different photonic platforms and material systems. Figure 1 illustrates the wavelengths of interest, while Table I summarizes some of the general requirements for each application. It is important to note that the details in the table are just illustrative and may vary depending on the specific system implementation.
Wavelength ranges that are accessible with waveguide materials (bottom rows), wavelength ranges that are accessible with gain materials (center), and wavelength ranges that are commonly used in some applications/market verticals as also described in the text and in Table I. Three vertical yellow shaded areas show wavelength ranges developed for communications industry (where the 850 nm wavelength range generally does not have any photonic integration). The wavelength ranges for applications are illustrative, with the key message being that PICs from ultraviolet (UV) to IR are needed for future systems and applications.
Wavelength ranges that are accessible with waveguide materials (bottom rows), wavelength ranges that are accessible with gain materials (center), and wavelength ranges that are commonly used in some applications/market verticals as also described in the text and in Table I. Three vertical yellow shaded areas show wavelength ranges developed for communications industry (where the 850 nm wavelength range generally does not have any photonic integration). The wavelength ranges for applications are illustrative, with the key message being that PICs from ultraviolet (UV) to IR are needed for future systems and applications.
Summary of some application drivers and specifications for some emerging market verticals.
Market . | Application . | Wavelength of operation . | Key requirements . | Driver . |
---|---|---|---|---|
AR/MR/VR | Displays | RGB (460–630 nm) | Very low on-chip loss/scattering, high-performance switching, high wall-plug efficiency | Integrated laser display engine for SWaP-C improvement |
AR/MR/VR | Sensing | 850–980 nm | Small size, high wall-plug efficiency, steering/reconfigurability of pattern | Pairing of photonic illuminators with CMOS camera for cost-efficient sensing |
Quantum computing | Trapped atoms/ions, generation of entangled photon pairs | ∼250 to ∼1700 nm depending on configuration | Generally, narrow to very narrow linewidth, low noise, high extinction ratio of on/off gating, very high signal to total spontaneous emission ratio, high quality filters, high density | SWaP-C improvement and enabling scaling of systems |
Quantum sensing | Magnetic field, electric field, temperature, gravity, force measurements, etc. | Various, often driven by atomic transitions | Narrow to very narrow linewidth, low noise, modulation for laser locking | SWaP-C and performance improvement compared to current systems |
Positioning, navigation, and timing (PNT) | Gyroscopes and optical clocks | Driven by atomic transitions, on-chip losses or fiber performance | Very narrow-linewidth (atomic, resonant), broadband (interferometric), very low on-chip propagation loss, very high level of polarization control, low coupling losses | SWaP-C and performance improvement compared to current systems |
Radio-frequency systems | Stable microwave oscillators, widely tunable microwave oscillators | N/A (output is RF), generally best performance where on-chip losses are lowest | Ultra-narrow linewidth (Hz-level of below), high-performance photodetectors (high-power, high-bandwidth), combs for downconversion | State-of-the-art phase noise performance, high-frequency performance, continuous tunability |
Spectroscopic sensors | Varies, depends on the specimen to be detected | Essentially all possible wavelengths, depending on application | Varies, but generally broadband operation is a requirement, comb-based sources are a plus (including dual-comb spectroscopy) | SWaP-C improvement compared to current systems |
Healthcare | Varies, depends on biomarker to be identified | Current systems utilize from ∼500 to 1700+ nm, broader operation beneficial | Varies, but generally broadband operation is a requirement | Continuous monitoring or home point of care, SWaP-C improvement |
Highly robust comm. systems | Very high temperature, uncooled communication systems (e.g., 125 °C) | ∼9xx nm wavelength range seems to be superior in terms of thermal performance | Very high temp lasing, medium to high bandwidths, WDM is a plus | Improved robustness communication systems for uncooled operation at extreme environments |
Market . | Application . | Wavelength of operation . | Key requirements . | Driver . |
---|---|---|---|---|
AR/MR/VR | Displays | RGB (460–630 nm) | Very low on-chip loss/scattering, high-performance switching, high wall-plug efficiency | Integrated laser display engine for SWaP-C improvement |
AR/MR/VR | Sensing | 850–980 nm | Small size, high wall-plug efficiency, steering/reconfigurability of pattern | Pairing of photonic illuminators with CMOS camera for cost-efficient sensing |
Quantum computing | Trapped atoms/ions, generation of entangled photon pairs | ∼250 to ∼1700 nm depending on configuration | Generally, narrow to very narrow linewidth, low noise, high extinction ratio of on/off gating, very high signal to total spontaneous emission ratio, high quality filters, high density | SWaP-C improvement and enabling scaling of systems |
Quantum sensing | Magnetic field, electric field, temperature, gravity, force measurements, etc. | Various, often driven by atomic transitions | Narrow to very narrow linewidth, low noise, modulation for laser locking | SWaP-C and performance improvement compared to current systems |
Positioning, navigation, and timing (PNT) | Gyroscopes and optical clocks | Driven by atomic transitions, on-chip losses or fiber performance | Very narrow-linewidth (atomic, resonant), broadband (interferometric), very low on-chip propagation loss, very high level of polarization control, low coupling losses | SWaP-C and performance improvement compared to current systems |
Radio-frequency systems | Stable microwave oscillators, widely tunable microwave oscillators | N/A (output is RF), generally best performance where on-chip losses are lowest | Ultra-narrow linewidth (Hz-level of below), high-performance photodetectors (high-power, high-bandwidth), combs for downconversion | State-of-the-art phase noise performance, high-frequency performance, continuous tunability |
Spectroscopic sensors | Varies, depends on the specimen to be detected | Essentially all possible wavelengths, depending on application | Varies, but generally broadband operation is a requirement, comb-based sources are a plus (including dual-comb spectroscopy) | SWaP-C improvement compared to current systems |
Healthcare | Varies, depends on biomarker to be identified | Current systems utilize from ∼500 to 1700+ nm, broader operation beneficial | Varies, but generally broadband operation is a requirement | Continuous monitoring or home point of care, SWaP-C improvement |
Highly robust comm. systems | Very high temperature, uncooled communication systems (e.g., 125 °C) | ∼9xx nm wavelength range seems to be superior in terms of thermal performance | Very high temp lasing, medium to high bandwidths, WDM is a plus | Improved robustness communication systems for uncooled operation at extreme environments |
AR/MR/VR (xR) is a potential frontier for our connectivity that could initially be paired with our smartphones but eventually have full functionality integrated into our glasses and replace smartphones altogether. AR/MR/VR, to meet the stringent requirements needed to replace cellphones as our main connectivity device, relies heavily on the size, weight, power, and cost (SWaP-C) advantages of photonic integration and semiconductors. Two main xR sub-systems need photonics: (1) Displays operating in visible wavelength range, typically by providing red–green–blue (RGB) emission,12 and (2) near-infrared (NIR) wavelength PICs/sensors that are typically paired with cost-effective CMOS sensors at around 850–980 nm wavelength.13 Both sub-systems have demanding requirements. They must be lightweight and highly efficient in terms of wall-plug efficiency to enable long battery operation. These requirements strongly favor the use of photonic integrated circuit (PIC)-based approaches.
Quantum computing is being developed using a variety of approaches. Here, we will focus on optical/photonic-based approaches such as the use of entangled photons,14 trapped ions,15 or cold atoms.16 Entangled photons can technically work at any wavelength, but trapped ions and cold atoms have specific wavelength requirements that depend on the type of atom or ion used. In most cases, the wavelengths used for ion/cold atom systems are below 1.2 μm, which requires sources and PICs that go beyond traditional datacom/telecom applications. Entangled-photon approaches, on the other hand, generally require nonlinearities and very low losses. To create useful quantum computers, it is necessary to significantly scale the number of qubits—which is a requirement that will require high-level of integration.
Quantum sensors, providing capabilities such as gravimeters,17 inertial sensing,18 electro-magnetic sensing, magnetometers,19 are a very broad field, but in many cases, similarly to ion/atom-based computing, they require specific wavelengths to manipulate atoms. On top of the specific wavelength requirements, they generally require very high-performance lasers with low phase and amplitude noise. Today, a large number of sensors are demonstrated in laboratory environments, and the ones that have left laboratory environment are characterized by large sizes (often exceeding 1 U) and high costs with the optical system often being the dominant cost/size driver. Photonic integration is expected to enable significant cost reduction, while also enabling size reduction and improved robustness for field deployment.
Positioning, navigation, and timing (PNT) encompasses systems such as LIDAR, gyroscopes, and (optical) clocks. LIDAR is outside of scope of this paper, with heavy research and commercialization in the past 10–15 years in part focused on high-level of integration to drive down the cost. A subset of optical gyroscopes includes interferometric gyroscopes20 and resonant gyroscopes.21 While both require low propagation losses and high-performance modulators, the interferometric gyroscopes typically use broadband optical sources, while resonant ones utilize narrow-linewidth lasers. Wider deployment of (optical) gyroscopes will require significant cost reduction to compete with MEMS based alternatives currently used in consumer electronics, and drastic cost reduction is only possible with increased levels of photonic integration. Optical clocks,22,23 similarly, require extremely narrow linewidths of lasers operating at the particular transitions of interest. Wavelengths of interest span broadly inside visible and NIR parts of the spectrum, depending on the choice of the atom with typical choices being Rb, Cs, and less commonly Li, Sr, and Yb.
Radio-frequency (RF) systems are transitioning to photonic approaches for the cases requiring very high-performance and/or leveraging the low-cost, large bandwidth and small-size of the optical fibers (when compared to high-speed RF cables). Most stable microwave oscillators today are realized by optical downconversion,24 in which extremely stable references at hundreds of THz are down-converted to GHz level ranges, with division reducing the phase noise in dB by 20 × log(division factor) and resulting with huge improvements of up to 80+ dB if we can divide fully from the optical carrier frequencies to RF frequencies. Such systems today, although state-of-the-art (SOTA), are characterized by very high cost and large size. Photonic integration that would combine extremely narrow-linewidth lasers (ideally Hz-level or lower) with supporting components needed to provide division to GHz-ranges would enable much broader deployments. Progress is already being made with first chip scale demonstrations.25
(Spectroscopic) sensing and healthcare have similar requirements, with the main difference being the specimen being sensed. In healthcare, we typically sense markers related to human health, while in general spectroscopic sensing, we detect the presence of atoms, molecules, and chemicals in the environment. In both cases, the wavelength of operation is defined by the application, and either narrow-linewidth tunable lasers to scan across a particular marker/resonance and/or broadband operation is generally required for optimal performance. The use of combs, and more specifically dual-comb spectroscopy, can further improve the quality of sensing.26 Combs can be generated by electro-optic modulation or by nonlinear processes. In all cases, significant improvements in SWaP-C (size, weight, power, and cost) are necessary for the widespread deployment of these technologies, leading again to photonic integration as necessary development.
Robust communication (R. Comms) systems would be a subset of communication systems in which operation is needed in, especially, challenging environment conditions. Examples could be various systems needing to meet e.g. military specifications including extended temperature range and increased exposure to shock and vibration requiring significant ruggedization. 9xx nm wavelength range seems to be especially suitable for this application as semiconductor lasers operating around 9xx nm generally provide highest wall-plug efficiencies and highest lasting temperatures that can exceed 180 °C.27
OVERVIEW OF THE PLATFORMS
Indium-phosphide (InP) is, historically, the only “single-material” photonic integrated circuit platform that enabled commercialized, advanced PICs (with tens or more components) with on-chip sources.1 Gallium-arsenide (GaAs) technically supports similar functionalities but to-date has not been developed to a nearly comparable degree. Both InP and GaAs have limitations related to scaling functionality. Combining multiple active components and/or passives typically requires multiple regrowths, and (ultra) low-loss waveguides are usually not achievable in pure III/V platforms. Furthermore, additional limitations in terms of scalability and uniformity arise due to using inferior quality substrates (with larger defect densities than silicon), and older processing tools and packaging techniques.
This was one of key drivers to develop silicon photonics and use superior quality of wafers and advanced processing capabilities developed and funded by the microelectronics industry in the field of photonics.4,5 Silicon offers many advantages as an optical material but lacks one key element for complete wafer-scale integration: an efficient electrically pumped laser. An elegant, wafer-scale solution to problem was developed by the University of California, Santa Barbara with the development of heterogeneous integration28 and commercialized by Intel at multimillion per annum scale providing unprecedented uniformity and scale.9 Heterogeneous silicon photonics provide a state-of-the-art photonic platform with 300 mm wafers and unmatched lithography but is limited in wavelength of operation to >1.2 μm due to silicon bandgap.
CHOICE OF WAVEGUIDE
Extending the wavelength range of operation requires the use of different materials, especially to cover the shorter wavelength range in which high-bandgap dielectric materials generally have to be used to provide low on-chip propagation loss. In most cases, the cladding can be SiO2, which has excellent performance due to the very high-bandgap. The choice of the waveguide core material varies between platforms, but some of the most commonly reported materials include silicon nitride (SiN), tantalum pentoxide (Ta2O5), and lithium niobate (LiNbO3), with aluminum nitride (AlN) and aluminum oxide (Al2O3) also being researched.
SiN offers transparency over an extremely wide wavelength range from ∼400 nm to beyond 2000 nm and has demonstrated record low on-chip losses.29 Losses lower than 0.1 dB/m are routinely achieved around 1550 nm30 and are typically at least order of magnitude lower than other material systems. Losses tend to increase toward blue wavelength range due to increased material losses and higher scattering loss (for same roughness). Material optimizations such as using PECVD films31 or oxide rich films can be used to reduce losses and achieve <1 dB/cm all the way down to 400 nm.32 The low propagation loss in silicon nitride (SiN) enables the realization of chip-scale lasers with hertz-level linewidths.30 Efficient nonlinear operation has been demonstrated in both thick and thin waveguide geometries. In thick geometries, dispersion is carefully managed to achieve phase matching for efficient nonlinear generation.33 In thin geometries, the high field intensities leveraging high quality factors and high finesses of the resonators for power buildup are utilized to achieve efficient nonlinear effects.30 SiN is a standard CMOS material, so it can be readily processed at all wafer sizes up to 300 mm (see Fig. 2). This allows the use of the most advanced lithography techniques for small feature sizes, high throughput, and exceptional yield. Finally, SiN arguably has the lowest cost of all the platforms as it uses silicon wafers that are either oxidized or SiO2 cladding is deposited before depositing SiN films, all of which are low-cost, scalable processes with very good control and uniformity.
SiN arguably provides the best on-chip propagation performance of all platforms with (1) ∼0.1 dB/m losses around 1550 nm and (2) <1 dB/m losses around 950 nm fabricated in scalable fashion using 200 mm (3) or 300 mm (4) wafers. Further improvements in propagation loss are possible but might be more challenging to be integrated with other (active) functionalities.
SiN arguably provides the best on-chip propagation performance of all platforms with (1) ∼0.1 dB/m losses around 1550 nm and (2) <1 dB/m losses around 950 nm fabricated in scalable fashion using 200 mm (3) or 300 mm (4) wafers. Further improvements in propagation loss are possible but might be more challenging to be integrated with other (active) functionalities.
Ta2O5 has garnered attention as yet another CMOS compatible material platform mainly due to its lower thermo-optic coefficient (∼8.8 × 10−6/K) compared to Si (1.8 × 10−4/K), and other high-index material platforms suitable for compact PICs that could potentially help reducing the thermo-refractive noise floor of lasers that utilize Ta2O5 waveguides as a part of their cavity. Note that SiN thermo-optic coefficient (2.5 × 10−5/K) is larger than thermo-optic coefficient of Ta2O5, while Ta2O5 has higher mechanical loss at the small offset frequencies.34 Other driving forces for the use of Ta2O5 as a PIC platform are ultraviolet (UV) operation with bandgap transparency theoretically enabling operation down to ∼300 nm wavelength, lower residual stress of the film (enabling thicker films), and higher Kerr coefficient compared to, e.g., SiN platform,35 which makes it a promising candidate for nonlinear on-chip processes. There have been demonstrations of high-Q Ta2O5 micro-resonators in C-band35 and near-infrared wavelength of 780 nm,36 and nonlinear processes including frequency comb generation and optical-parametric oscillation,35 but propagation loss performance is still inferior to SiN.37 The low temperature deposition of sputtered Ta2O5 is another factor, making it uniquely suitable for integration in PIC platforms with restricted thermal budget (e.g., heterogeneous lasers and/or PICs integrated with microelectronics), but this can also be limitation in achieving best on-chip losses that, in many cases, require high-temperature anneals, especially when operating around 1520 nm wavelength.35,37 Low loss amorphous Ta2O5 experiences a phase transition to a more stable polycrystalline state upon getting exposed to high energy irradiation like x-ray or annealing at relatively high temperatures (around 600 °C with exact temperature depending on the stress imposed on the film). This may limit the annealing temperature of the Ta2O5 PICs or limit the use in fabrication involving x-ray irradiation (e.g., metallization using e-beam evaporation).
LiNbO3 has emerged as a promising material platform for PICs, mainly due to its unique combination of optical and mechanical properties. High electro-optic coefficient in LiNbO3 combined with its large transparency window has made it a promising PIC platform for operation through a large spectrum, ranging from ultraviolet to mid-infrared38 while providing high-performance modulators. The challenges of using this platform include its non-CMOS compatibility, fabrication difficulties, sensitivity to higher optical intensities (especially at shorter wavelengths), and temperature sensitivity. Nonetheless, LiNbO3 PICs have witnessed impressive progress with low-loss, and high-Q micro-rings demonstrated across the spectrum from visible to the near-infrared wavelengths,39 paving the way for efficient nonlinear processes on a chip, for both classical and quantum applications. The unique combination of second and third order nonlinearities in LiNbO3 along with improvements of the fabrication processes and material properties has led to efficient nonlinear processes on a chip.40,41 Material properties can be improved by, e.g., doping with Mg, Zn, or by using lithium tantalate (LiTaO3). However, the challenges mentioned earlier combined with significantly higher cost of thin-film wafers (when compared to silicon wafers) and inferior uniformity will likely limit scaling of PICs based on LiNbO3 for large volume, consumer applications when compared to SiN.
AlN42 and Al2O343 are two additional large bandgap materials explored to push the PIC operation to shorter wavelengths. They are especially promising for operation in ultraviolet (UV), with AlN transparency extending to around 300 nm and Al2O3 transparency almost reaching 200 nm wavelengths. AlN, furthermore, supports electro-optic modulation capability extending the active functionality to modulators and tuners.
We expect the research will continue with all the materials, and most will have, at least limited, commercial deployment, but we strongly believe that SiN will be the material of choice for standard, high volume products owing to its performance, cost, and robustness of processing.
CHOICE OF LIGHT GENERATION
Electrically pumped, direct-generation is generally the preferred way to make lasers at a particular wavelength due to the higher wall-plug efficiencies, higher powers, smaller size, and simpler operation, but there is a part of the visible spectrum where direct emission cannot provide lasing spanning from ∼535 to ∼630 nm wavelengths (so-called green–yellow gap). Materials for direct emission include gallium-nitride (GaN) based actives covering the range from ∼375 to ∼535 nm, and GaAs and InP based actives covering the range from ∼630 nm to ∼1.7 μm and beyond as also summarized in Fig. 1. The green–yellow gap is an area of active research and progress with demonstrations of LEDs (spontaneous emission) in full wavelength range,44 and optically pumped sources in part of the green–yellow gap range45 have recently been made. The goal of the development is to use advanced growth techniques to fully close the gap.
Frequency conversion is another approach to generate light in visible and NIR, and it can utilize frequency doubling,46 frequency tripling,47,48 and optical-parametric oscillation (OPO).49 High efficiencies of conversion have been demonstrated in several waveguide systems (including SiN, LiNbO3, and Ta2O5), but in all cases, they are sensitive to fabrication tolerances, as high efficiency requires good phase matching. The fabrication tolerances can limit the ability to scale such systems and impact yield. Nevertheless, today, this is the only way to access green–yellow gap with PICs.
GAIN INTEGRATION
Active components (gain, modulation, and detection) integration is necessary to provide system-on-a-chip (SOC) level functionality, and laser is arguably the greatest challenge. Although some early demonstrations have moved from benchtop laser-driven systems to hybrid integrated lasers/gain (typically using butt-coupling), we believe that the rationale for gain integration in emerging markets is the same as it was for datacom/telecom markets. In those markets, wafer-scale integration was pursued with the promise of enabling wafer-scale testing and singulation, which would reduce cost and size while improving robustness.
Some early systems might continue utilizing hybrid integration in which two (or more) pre-processed dies are mechanically aligned during packaging to provide optical coupling. While hybrid integration can achieve good performance at lower volumes,50 it comes with a high price tag and uniformity issues. These issues are likely to be even more pronounced in emerging markets, which typically operate at shorter wavelengths with tighter tolerances. Additionally, the robustness of hybrid systems is typically reduced, especially for operation over a wide temperature range, due to differences in the coefficient of thermal expansion (CTE) between the materials used. Finally, integrating more than one element significantly increases the challenges of this approach.
Heterogeneous III/V on SiN integration with four types of gain materials including GaN, GaAs (two types), and InP integration.
Heterogeneous III/V on SiN integration with four types of gain materials including GaN, GaAs (two types), and InP integration.
Photonic wire bonding is another approach that promises to enable precise alignment between two (or more) pre-processed dies.51 A waveguide and/or micro-optical component is written using additive 3D nanofabrication techniques to couple between the chips after they are placed on a common substrate. However, research in photonic wire bonding is still ongoing, and no large-scale commercial systems have been deployed. Most of the current research is focused on longer wavelengths (1550 nm),52 and going to shorter wavelengths will increase challenges related to material loss and scattering loss. Additionally, there are open questions about scalability (if a larger number of active functionalities are needed), cost, and robustness at scale for rugged, low-cost PICs.
Transfer printing is another approach that has been intensively researched in the past 20 years.53,54 However, to date, it has not been commercialized at scale with in-plane optics, even at longer wavelengths. We believe that the challenges of this approach only become harder at shorter wavelengths, as tolerances are tighter. Furthermore, to realize a SoC, multiple functionalities need to be integrated. This complicates the placement accuracy, as a pre-populated carrier wafer will likely be required to account for two or more active functionalities, resulting in at least two alignment steps, each of which brings tolerance concerns.
Heterogeneous integration is the only approach that has been commercialized at scale for datacom using InP on Si integration9 and has provided the highest reliability transceivers to date.10 Moving the technology to shorter wavelengths requires developments to account for different materials (GaAs and GaN), different etch chemistries, different metallization stacks, different coefficients of thermal expansion (CTEs), etc. This has been a focus at Nexus Photonics, and we have developed high-bandgap GaAs,55 GaAs,27 and InP (see Fig. 3) actives integration in a wafer-scale process, with GaN still in development but with already demonstrated electrically pumped heterogeneous sources (see Fig. 3). We strongly believe heterogeneous integration is the most promising approach due to already demonstrated ability to scale, exceptional reliability (as demonstrated in challenging datacom markets) and robustness. Exceptional robustness to thermal variations has been demonstrated with heterogeneously tunable narrow-linewidth lasers in which linewidth of the laser was essentially unchanged from room temperature to 150 °C, showing that the coupling efficiency between gain and passive cavity (corresponding to cavity loss) has essentially not changed in such large temperature range.27 Finally, the integration of 1000 s of active components with lithographically defined alignment is supported at wafer-scale enabling unprecedented complexity and density, as was demonstrated internally at Nexus with density exceeding two active devices per mm2 across the wafer.
Typical steps in heterogeneous process include passive processing (e.g., definition of SiN, LiNbO3, Ta2O5 waveguides, and components), III/V bonding, and III/V process + back end of line (BEOL). Bonding uses unpatterned III–V thin film dies with a coarse alignment, and then devices and their alignment to passive components are lithographically defined on the full wafer scale using common alignment marks on the wafer. The unpatterned III–V dies can be prescreened for defects to improve yield.
Typical steps in heterogeneous process include passive processing (e.g., definition of SiN, LiNbO3, Ta2O5 waveguides, and components), III/V bonding, and III/V process + back end of line (BEOL). Bonding uses unpatterned III–V thin film dies with a coarse alignment, and then devices and their alignment to passive components are lithographically defined on the full wafer scale using common alignment marks on the wafer. The unpatterned III–V dies can be prescreened for defects to improve yield.
Heterogeneous integration is flexible, supporting integration of actives on various passive waveguides, including SiN, Ta2O5, LiNbO3, and others and can also accommodate not only III–V materials for gain with broad spectral coverage but also a variety of other exotic materials to provide magnetic properties, phase-change materials, materials with higher nonlinearities, etc. A common process for heterogeneous integration is shown in Fig. 4.8,27
Scalability of heterogeneous integration using commercial foundries. In this proof-of-concept runs, SiN is processed at 200 mm scale using a CMOS foundry with extremely high uniformity between multiple wafers/dies, while bonding and III/V process are performed at 100 mm using a commercial foundry with excellent process control. The commercial foundry process enables very high uniformity and yields as demonstrated by screening results (inset) showing monitor laser thresholds and monitor photodetector dark currents of 500 devices distributed across the wafer.
Scalability of heterogeneous integration using commercial foundries. In this proof-of-concept runs, SiN is processed at 200 mm scale using a CMOS foundry with extremely high uniformity between multiple wafers/dies, while bonding and III/V process are performed at 100 mm using a commercial foundry with excellent process control. The commercial foundry process enables very high uniformity and yields as demonstrated by screening results (inset) showing monitor laser thresholds and monitor photodetector dark currents of 500 devices distributed across the wafer.
CHALLENGES AND FUTURE
There has been tremendous progress in expanding the wavelength range of integrated photonics, but most of the work has been in research and early prototypes. Moving the technology to products will require further developments, and one of the key challenges is fractured markets with complex and different requirements. Datacom and telecom both had strongly defined requirements, primarily driven by the use of optical fibers, which clearly defined the wavelengths of operation. Most of the work in these markets was focused on power, wall-plug efficiency, noise, and, eventually, tunability. At the same time, both markets were large, and significant investments were poured into maturing and qualifying platforms and products over more than 30 years.
To accelerate the development of broadband integrated photonics, market consolidation (e.g., determining a subset of best wavelengths for, e.g., quantum applications that are notorious by number of “requested” wavelengths) and identifying key volume applications/drivers (e.g., xR, healthcare, others) would be preferred. Ideally, multiple market verticals could share the same photonic platform, which would allow ramping up the fabrication scale. Today, such processes can be run at 100 mm with very high yield (see Fig. 5), and this can serve specific markets, but ideally, we have to look for opportunities that are at least comparable to datacom market, and preferably 10× larger. Healthcare with, e.g., glucose sensing could be one of such markets, or xR and/or drone vision systems with advanced photonic illuminators.
Progression of integrated photonic platforms to address emerging applications starting from (1) co-packaged chips on a carrier to (2) heterogeneous integration, which is currently being commercialized for sub-bands covered with single III/V gain material (see Figs. 1 and 4), and then (3) co-packaging of multiple heterogeneous chiplets toward (4) full heterogeneous integration for single-chip solutions. Increasing complexity requires increased volumes to justify the development costs.
Progression of integrated photonic platforms to address emerging applications starting from (1) co-packaged chips on a carrier to (2) heterogeneous integration, which is currently being commercialized for sub-bands covered with single III/V gain material (see Figs. 1 and 4), and then (3) co-packaging of multiple heterogeneous chiplets toward (4) full heterogeneous integration for single-chip solutions. Increasing complexity requires increased volumes to justify the development costs.
Such volume scaling would also justify moving to larger wafers, 200 mm and preferably 300 mm in which low-cost, optimized packaging could be leveraged and could reduce the costs of single (packaged) PIC to few USD. Such price points are only achievable if there is a wafer-scale process throughout the full supply chain until final packaging.
In terms of functionality, addressing the green–yellow gap is most important. We believe that frequency conversion will be utilized for some applications, but it inherently has challenges in scaling complexity due to unavoidable process variations impacting yield. Material developments have already covered the gap with LEDs grown on GaN, and this continues to be an area of very active research. The promise of on-chip gain and lasing, at these wavelengths, would enable much higher densities and efficiencies of PICs.
ROADMAP AND CONCLUSIONS
On-chip gain, and more specifically laser, is the key challenge in scaling photonic systems. Close engagement between material development groups, photonic integrated circuit groups, sub-system integrators, and system integrators is needed to coordinate and accelerate the development of PICs and whole ecosystem for commercial scaling. Identifying large volume market drivers can accelerate developments, which are currently primarily driven by government and R&D funding. A big advantage for the upcoming technologies is the vast accumulated knowledge from scaling datacom/telecom markets, which can significantly accelerate developments. We already see this in scaling (SiN) heterogeneous photonics where, despite additional introduced materials, the pace is very aggressive, and very broadband operational range was developed in only a few years including transfer to commercial foundries.
Once gain (from single substrate materials such as GaN, GaAs, or InP) integration is scaled and commercialized, additional functionalities can be introduced (by adding more materials) to address further requirements, and, eventually, we could envision complete state-of-the-art (SOTA) broadband functionality integrated on a single chip, or potentially several optical chiplets if co-processing of some materials is especially challenging. In the chiplet approach, it is important to highlight that somewhat higher losses between chiplets might be acceptable if these losses do not reside inside the laser cavity where they have largest impact on performance (as in hybrid approach). Eventually, we believe integration of multiple III/V materials will be done on the same wafer as soon as there is commercial drive to do so (see Fig. 6).
ACKNOWLEDGMENTS
This work was supported in part by the Defense Advanced Research Projects Agency (DARPA) under Contract Nos. HR0011-20-C-0135 and HR0011-22-C-0017. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the DARPA. Approved for Public Release, Distribution Unlimited.
AUTHOR DECLARATIONS
Conflict of Interest
All authors are employees or own parts of Nexus Photonics.
Author Contributions
Chong Zhang: Conceptualization (equal); Data curation (equal); Investigation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Frank Levinson: Supervision (equal). John E. Bowers: Supervision (equal); Writing – review & editing (equal). Tin Komljenovic: Funding acquisition (lead); Supervision (lead); Writing – original draft (equal). Minh Anh Tran: Conceptualization (equal); Data curation (equal); Investigation (equal); Methodology (equal); Writing – original draft (equal). Zeyu Zhang: Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Visualization (equal); Writing – review & editing (equal). Ali E. Dorche: Data curation (equal); Formal analysis (equal); Investigation (equal); Visualization (equal); Writing – original draft (equal). Yang Shen: Data curation (equal); Formal analysis (equal); Investigation (equal); Visualization (equal); Writing – review & editing (equal). Boqiang Shen: Data curation (equal); Software (equal); Visualization (equal); Writing – review & editing (equal). Kaustubh Asawa: Data curation (equal); Investigation (equal); Methodology (equal). Glenn Kim: Data curation (equal); Investigation (equal). Nathan Kim: Data curation (equal); Investigation (equal).
DATA AVAILABILITY
Data sharing is not applicable to this article as no new data were created or analyzed in this study.