Hybrid integrated chip-scale laser systems

A photonic integrated circuit (PIC) is a microsystem that integrates multiple photonic functions such as light generation, detection, modulation, filtering, routing, and nonlinear processing on a small chip where the waveguide dimension can be accurately controlled and the chip can be mass produced. Like electronic integrated circuits, PICs can significantly reduce the system’s size, weight, operation power, and cost (SWaP-c). PIC technology has transformed many optical systems, such as optical communication systems, nonlinear optical systems, and quantum optical systems, which have traditionally relied on tabletop systems and bulky components, into chip-scale platforms. This device and system miniaturization has successfully led to a wide range of practical applications in computing, sensing, spectroscopy, and communication.

Unlike the electronics industry, where silicon is the dominant material platform, multiple material systems have successfully been used to create devices on the PIC platform, including optical transceivers, tunable lasers, optical modulators, and wavelength selective detectors. This is mainly because different material platforms offer various advantages and disadvantages and, therefore, there is not one dominant material platform in the field of integrated photonics. Commercially available photonic integrated circuits are generally fabricated using III–V semiconductors (GaAs and InP) and Silicon/SiO₂/Si₃N₄ material platforms.¹ III–V semiconductors are optimal for light generation and detection, but the integration of both active and passive devices requires complex epitaxy regrowth. The silicon-based platform has desirable properties for passive components (SiO₂ and Si₃N₄) and modulators (Si), but it is very difficult to realize electrically driven laser sources due to its indirect bandgap material property. Silicon on insulator (SOI) waveguides can provide tight optical confinement and large third order optical nonlinearity, but two photon absorption and free carrier absorption induced by two photon absorption limit the maximum light intensity. Therefore, SOI waveguides are not suitable for nonlinear processes that require high power continuous wave (CW) light input, such as micro-resonator-based frequency comb generation.² 

In recent years, hybrid/heterogeneous integration of active and passive components in different material systems has attracted intense interest in creating next generation multifunctional photonic integrated circuits.³ Some examples of such can be seen in Fig. 1. We envision that future hybrid/heterogeneous integrated laser systems will have multiple gain sections and provide a number of advanced functions such as power scaling, linewidth narrowing, wavelength tuning, and nonlinear conversion. In addition, many other optical components such as modulators, isolators, amplifiers, and detectors will be co-integrated on the same chip for a wide range of communication, computing, sensing, and signal processing applications. The main motivation is to realize each desired photonic function by using the best suitable material and creating the whole system based on hybrid/heterogeneous integration. Though they have the same goal, hybrid and heterogeneous integration offer different methods to achieve multi-functional PICs with integrated laser sources. In the hybrid integration approach, active and passive components are separately fabricated on different sub-chips and then integrated through some method of optical coupling.⁴ This integration method often requires accurate sub-chip dimension control and/or mechanical alignment/assembly. Comparatively, the heterogeneous approach directly integrates III–V gain materials on silicon through wafer bonding or heteroepitaxy growth to create electrically pumped lasers and amplifiers for silicon photonics components.³ In hybrid integration, individual components can be tested prior to integration to increase device reliability, whereas this is not possible in heterogeneous integration. It can also be used on a wide variety of material platforms. Heterogeneous integration is currently more optimal for higher throughput applications, but due to the greater ease of integration and a wider range of usability, we will be focusing on hybrid integration approaches for chip-scale laser systems in this paper. Specifically, we will be looking at hybrid integration using a silicon substrate, as it is the most studied.

Hybrid/heterogeneous integration is accomplished by utilizing a variety of coupling methods, each with its own set of parameters that allow for optimal performance in key areas. These coupling methods are shown below in Fig. 2. As with any scientific method, there are tradeoffs to be considered when selecting a coupling method for a specific application. These will be briefly discussed below.

This coupling method is usually considered in heterogeneous integration applications, as it involves growing one material system on top of another. This method unlocks a plethora of capabilities, as “active silicon” devices can be created. The largest setback within hetero-epitaxy coupling is lattice and thermal mismatch between materials. This mismatch causes many dislocations to be formed on the material surface and can inhibit output power and overall performance. Research is currently being pursued to minimize the effects of lattice and thermal mismatch through avenues of semiconductor nanostructures like quantum dots, but much ground still needs to be gained before this method can be reliably realized.^5–7 

Wafer bonding integration methods help alleviate the issue of lattice/thermal mismatch associated with hetero-epitaxy coupling. This is performed using one of two bonding methods: direct bonding or adhesive bonding. Direct bonding is advantageous due to its ability to bond two materials without any additional polymers that may impact light propagation or thermal characteristics. However, to achieve covalent direct bonding, high temperature annealing is required, which is lengthy and requires ultra-flat and ultra-clean surface quality. Adhesive bonding minimizes material quality requirements and uses polymers such as Benzocyclobutene (BCB) and SU-8 to bond materials in a soft-bake process. This makes it an optimal candidate for mass produced devices, with the trade-off of poor thermal conductivity. The light in the active region is coupled using a spot size converter (SSC) that is adiabatically tapered from a wide tip to a narrow tip to push the optical mode to the passive waveguide bonded below the active device. In general, heterogeneous integration needs a close index match to facilitate the mode transition between the active and passive devices; however, there are also recent research developments that allow heterogeneous integration between Si₃N₄ and III–V, which utilizes either an intermediate layer or more complex processing.^8,9

Off-chip coupling methods are emerging as the main option for hybrid integration. Vertical coupling makes use of mirrors and gratings to couple light from an external gain chip into a passive chip. The biggest setback in off-chip coupling methods lies within alignment tolerance.^10–12 Vertical coupling is highly dependent on the angle of the incident light. Total Internal Reflection (TIR) mirrors are placed at 45° at the gain output to allow light to be coupled vertically into passive cavities. Micro-optical devices such as prisms are also used in conjunction with grating couplers to minimize the impact of incident angle error and maximize coupling efficiency.¹³ Even still, the alignment of the devices is critical to their performance.

Like vertical coupling, lensed coupling utilizes two separate devices and uses a medium to couple light from one to the other. Lensed coupling does not rely on incidence angle nearly as much as vertical coupling but rather relies on the XYZ positional alignment of both devices. Micro-optical devices such as ball lenses and prisms are integral to this coupling method. Light from the gain chip enters a micro-optical device, which directs the light into the passive chip. The presence of a micro-optical device as a medium between the two chips maximizes the alignment tolerance of the two chips, allowing for the scalability of the hybrid solution.^14,15

This is a very specific application of lensed coupling, where instead of using a typical micro-optic lens, a polymer material is used to create a wire that connects the active and passive chips. This is optimal due to the mechanical flexibility of a polymer waveguide as well as the ability to use 3D printing methods to create these types of devices. Typically, the photonic wire is printed in the shape of an SSC to convert the optical mode from the active waveguide to the passive waveguide. Mechanical structures can be used to achieve an accurate alignment of the polymer waveguide and maximize the alignment tolerance of the device. 3D printing of photonic wires could allow for the cost minimization of hybrid devices.¹⁶ 

Edge coupling is the simplest form of hybrid integration, as it involves optimizing the position of the two chips, removing the need for any medium in between the chips. As one could imagine, the alignment accuracy of edge coupling is critical. However, this method's simplistic manufacturing process makes it optimal for research and other applications that otherwise require longer processing times. In addition, progress in passive alignment technology can allow this technology to be brought to scale.^17–20 

One of the greatest advantages of edge coupling is its robust nature across many applications. Primarily, it can be used within a wide variety of passive and active material systems, unlike other coupling schemes, which either depend on refractive index contrast or require additional optical components. The ease of integration also makes edge coupling a highly optimal solution for hybrid integration. For these reasons, we will be focusing on hybrid edge coupled laser systems in this paper.

Edge coupling is made possible by converting the large mode size of an active chip to the smaller mode size of a single-mode passive waveguide. This is important to minimize the coupling loss between the active and passive devices. The difference in mode size is apparent in Figs. 3(a) and 3(b). The mode size for the active device in Fig. 3(a) is much larger than the mode size of the waveguide in Fig. 3(b), which causes a significant mode mismatch. The larger mode in the active region must be converted to a smaller mode in the passive region to achieve optimal confinement and yield low coupling loss. This is performed using an SSC on the passive chip, which in its simplest form is a taper from a large tip width to accommodate the large mode to a smaller tip width to accommodate the smaller mode. The SSC is sufficiently long to achieve adiabatic, or lossless, mode conversion.²¹ The larger active mode couples into the larger tip of the SSC, and the mode field diameter (MFD) then decreases as the light is squeezed down to the single-mode waveguide width, as shown in Fig. 4(a). SSCs can also use an inverse design and taper from an ultra-small tip width at the facet, depending on the application. An ultra-small tip width allows the mode to propagate in the cladding until it evanescently couples into the passive waveguide, as shown in Fig. 4(b). This allows for a much lower reflection at the facet and a much lower mode mismatch with the active device, which can result in a reduced coupling loss. It is important to note that these are not state-of-the-art taper designs, but they help to explain the fundamental principles of SSC design.

The edge-coupled architecture is reliant on facet quality at the edges of both passive and active devices. Perfect mirror facets will remove light scattering, leading to lower coupling loss. Mirror facets are created by either a cleaving or dicing/polishing process. The cleaving process consists of scribing a small portion of the chip that will not be used as a coupling facet, then applying slight pressure to the scribe until the chip splits along the scribe line, following the crystal lattice structure.²² This is most optimal in III–V devices due to the III–V material properties. Passive chips are typically diced using a diamond blade and then polished to smooth the facet in order to reduce light scattering. An anti-reflective (AR) coating can then be applied to minimize reflections at the facet. An angled facet design can also be used to further reduce the reflection for some applications.

The difference in mode size between the active and passive devices shows the clear need for an SSC and is indicative of the importance of accurate alignment between the two devices. Coupling loss can be induced by both lateral, vertical, and angular misalignments. The alignment tolerance is highly dependent on the architecture of the SSC. The alignment tolerance of the inverse taper is typically much lower than that of the regular taper. The coupling efficiency of such a device is shown while the alignment of the active device is shifted in Fig. 5. In this instance, a lateral misalignment on the order of 0.6 μm resulted in a coupling loss of 1 dB.^23, Figure 5 also shows the dependence of the coupling efficiency on the gap between the active and passive devices. Figure 6 shows that vertical misalignment also has a large impact on coupling efficiency, even more so than lateral misalignment. Misalignment tolerance can be increased by altering the SSC geometry by creating “fork” structures, where multiple tapers combine into one waveguide, as exemplified in Fig. 7(a). This device structure allows the maximum amount of light to be captured at the coupling interface, allowing for a much wider alignment tolerance.²³ This device greatly widened the misalignment tolerance to near 2.2 μm, as shown in Fig. 7(b).

Alignment can either be performed using active or passive methods. The active method employs a power monitor to measure the power output of the passive device as the alignment is performed. The alignment is tweaked until the maximum output is achieved. This is the most simple and reliable method. On the other hand, passive alignment utilizes alignment markers (fiducials) that indicate the position of several cameras integrated with computer vision software to help with the placement of optical components. This is well-suited for high throughput applications where a longer initial lead time is acceptable in order to yield higher throughput via automatic processes. It is important to note that the active alignment process can also be automated through the use of a feedback loop in order to create optimally aligned devices in a high throughput setting.

A plethora of applications can be realized using hybrid integration, specifically the edge coupling method. Narrow linewidth devices, beam combinations, frequency combs, and Pockels lasers primarily serve as major current applications of chip-scale laser systems based on hybrid integration through edge coupling. These devices are made possible by high quality active lasing devices and low-loss passive devices. This allows for minimal phase noise in the case of narrow linewidth applications, making it suitable for spectroscopy and LiDAR applications.²⁴ Beam combination is made possible due to the low loss in the passive device that allows multiple beams to be coupled together to maximize output power from a single semiconductor laser. Frequency combs and Pockels lasers need a high quality laser pump, which can be optimized separately from the low-loss passive device, allowing for complex laser systems to be created.

The physical path-length of ECLs can be very large on small, passive chips using spiral device geometry, while the effective optical path-length can also be largely increased by taking advantage of micro-ring resonators (MRR) in a Vernier configuration. This scheme entails two MRRs of slightly different radii (and thus different free spectral ranges) connected in parallel with low coupling coefficients, thus increasing the round-trip distance of photons that enter the ring resonator and greatly increasing the effective optical path length. The MRRs also act as a frequency filter to isolate the lasing mode of the device. Thermo-optic or electro-optic phase shifting is used to tune the output to a resonant wavelength (so that the peak output of both ring resonators overlaps at the resonant frequency). This unlocks the ability to tune the resonant wavelength within a certain range, enabling tunable lasers to be realized. All of this is only possible on a low loss, passive platform that allows photons to travel large distances without losing energy, thereby increasing the photon’s lifetime.

The development of hybrid integrated ECLs has enabled breakthrough technology in narrow linewidth and wavelength tunability applications. In recent research, hybrid integrated ECLs have enabled the Schawlow-Townes linewidth to be reduced from on the order of MHz (as in distributed Bragg reflectors and distributed feedback monolithic lasers) to the order of a few Hz. In recent research, wavelength tunability has also been largely increased from spanning just the C-band for communication applications to spanning as large as 172 nm. External cavity devices have also been used to realize narrow linewidth and wavelength tunability at a variety of wavelengths, extending both into the 1 μm range and the 2 μm range. Table I shows the advancements made from initial devices to recent research.

Some of the initial work performed in Ref. 27 was aimed at minimizing spectral linewidth to below 100 kHz in order to use the devices in multilevel modulation schemes, such as 16 quadrature amplitude modulation (QAM). This was performed using a passive silicon platform integrated with two high quality-factor (Q-factor) MRRs integrated in a vernier configuration. The device layout is shown in Fig. 8(a). A low MRR coupling coefficient of 0.15 was selected so that the optical path length was 3.0 mm. The importance of an extended optical path length is shown in Fig. 8(b), which shows the relationship between optical path length, quality factor, and spectral linewidth. This work yielded a minimum linewidth of 46 kHz at the resonant wavelength of 1584.5 nm. The group also achieved a linewidth of less than 70 kHz for the entire L-band.

Though a narrow linewidth was found to be achievable, the feasibility of such a device in the commercial market is limited due to the high input power required to achieve optimal device performance. This is mainly due to the low wall plug efficiency (WPE) of the hybrid laser design. An increased WPE would drastically improve the success of narrow linewidth devices in the commercial market. To achieve ultra-low power Si photonics (SiPh), a WPE greater than 10% would have to be achieved.²⁹ WPE is discussed in depth in Ref. 46.

In 2019, a group set out to inch closer to this goal by making a highly efficient tunable laser while also minimizing unwanted optical feedback. The device was created using an RSOA edge coupled to a SOI device that contained two vernier MRRs integrated with thermo-optic heaters. The device layout is shown in Fig. 9(a). The tuning range of the device was found to be 150 nm, which was achieved by optimizing the coupling coefficients and ring geometry of the device. The maximum WPE was found to be 18% while providing an output power of 20 mW and maintaining laser stability up to −19 dB of external optical feedback. Designing for high WPE comes with the trade-off of a broader linewidth than less efficient designs. However, in this instance, a linewidth between 30 and 300 kHz was achieved, depending on the selected operating coupling coefficient. The relationship between the coupling coefficient, bias current, WPE, and optical path length for the described device is shown in Fig. 9(b).³⁷ 

Current research is primarily focused on narrowing linewidth into the Hz range, drastically increasing the wavelength tunability range, and minimizing input power. There are also many researchers trying to unlock tunability in higher frequency bands, such as the 2 μm band, as in Ref. 39. The widest tunability range to date of 172 nm was achieved in 2022 by butt coupling an RSOA with a passive chip containing two MRRs and a tunable Sagnac loop (TSL) integrated with an MZI.⁴³ The device structure is shown in Fig. 10(a). The power coupling coefficient of the MRRs is tuned between 0.06 and 0.15 thermo-optically to drastically increase and tune the optical path length. This is what allowed for the wide tuning range of 172 nm between 1487 and 1659 nm wavelengths. The output lasing spectra of the ECL are shown in Fig. 10(b).

The research discussed earlier focuses on hybrid integrated narrow-linewidth ECLs using RSOAs as the gain chip. There exists an alternative approach for achieving narrow-linewidth hybrid integrated lasers known as self-injection-locking lasers. In contrast to ECLs, this category of lasers achieves linewidth narrowing by locking a semiconductor laser [such as a Fabry–Pérot (FP) or distributed-feedback laser (DFB)] to a high-quality-factor microresonator based on the self-injection locking effect. Using lasers instead of RSOAs offers the advantage of increased robustness to coupling loss and fluctuations in reflections.⁴⁷ Jin et al. demonstrated a self-injection locking laser in 2022 that utilized a DFB laser butt-coupled to ultra-high-Q microresonators with a Q-factor over 260 × 10⁶.⁴¹  Figure 11(a) illustrates the configuration of their device and the measurement apparatus employed in their study. The combination of the self-injection locking effect and the utilization of the ultra-high-Q resonator resulted in significantly suppressed frequency noise by five orders of magnitude, yielding a white-noise floor of 0.2 Hz² Hz⁻¹, corresponding to a short-term linewidth of 1.2 Hz, as shown in Fig. 11(b).

Self-injection locking narrow-linewidth hybrid lasers also allow for broadband/coarse wavelength tuning, but with a slightly more complex procedure compared to ECLs. To reach the locked regime with stabilized single-longitudinal lasing in self-injection locking lasers, the resonance frequency of the external microresonator needs to be aligned with the mode of the semiconductor laser. Additionally, the external phase delay needs to be adjusted to ensure that the feedback field constructively interferes with the light field inside the laser cavity. Once reaching the locked regime, one can achieve frequency pulling or mode-hop-free frequency fine tuning by varying the resonance frequency of the microresonator. This distinctive characteristic is exclusive to self-injection-locking lasers, setting them apart from ECLs. Recent studies have demonstrated self-injection locking lasers that offer mode-hop-free fine-tuning capabilities of up to tens of GHz as well as broadband coarse tuning capabilities of up to tens of nanometers.^42,48 The items marked with asterisks (*) in Table I highlight a selection of recent examples of self-injection locking lasers.

Hybrid integration can also be used for beam combination applications. Beam combining is a highly sought after solution to achieve high power, high brightness laser systems. These types of laser systems are prevalent in laser material processing, remote sensing, fundamental science, and directed-energy system applications.^49–52 High brightness systems require spatial coherence over the emitting aperture to generate a diffraction limited, single-lobe far field. In individual laser systems, high power operation results in thermo-optic and optical nonlinear effects that destroy the desired spatial coherence and degrade beam quality.^53–55 Beam combining offers a solution to obtain high optical power and brightness by combining multiple lasers.

Conventional beam combining systems based on bulky free space or fiber optical components can be divided into two categories: coherent beam combining (CBC) and wavelength beam combining (WBC).⁵⁵ WBC systems rely on diffractive components such as dispersive or volume gratings to spatially overlap beams from different channels that operate at different wavelengths.^56–61 WBC has the advantage of being a phase insensitive, incoherent combining method, meaning it does not require phase control and can be scaled to many array elements. However, WBC is not suitable for some applications that require a narrow linewidth, and its scalability is ultimately limited by the gain bandwidth. On the contrary, CBC is a phase sensitive, coherent combining technique that requires accurate modal and/or phase control. In the CBC system, all lasers are operated at the same wavelength and phase-locked. CBC systems can be implemented either by a filled-aperture approach, where the overall emitting aperture is always equal to a single array element,⁵⁵ or a tiled-aperture approach, where the overall emitting aperture scales with the number of the array elements. Depending on whether feedback is applied to control the phase of individual array elements, CBC systems can be divided into two categories: active combining and passive combining.

Traditional beam combining systems, both WBC and CBC, are based on free space optical components, so they are usually bulky, complex, and not robust against vibration. Hybrid integrated PIC-based beam combining systems allow for component miniaturization and integration, which can significantly reduce the optical system SWaP.

Hybrid integration can be used to improve both active and passive CBC systems. Here we focus on passive CBC systems due to their simplicity. To implement a hybrid integrated PIC-based CBC system, the Fourier lens external cavity, which provides the coherent coupling in the conventional CBC system, is replaced with a cascading, tree-like coupler array in a passive silicon chip,⁶² as shown in Fig. 12. There is no fill factor problem as in the free space system since the light is confined in optical waveguides.

The whole CBC system can be realized on a single chip by utilizing hybrid integration.^62,63 The diode laser amplifier array is directly coupled to the passive coupler array through direct facet-to-facet coupling. For the coherent coupler array, only one output port has nonzero reflection to provide feedback, as shown in Fig. 12. The DBR is used to increase the reflectivity and select the operation wavelength. This configuration forces all the lasers to constructively interfere at the nonzero feedback port and destructively interfere at the zero feedback port. Therefore, the CBC of multiple laser amplifiers is obtained.

To realize chip-scale WBC systems, the diffraction grating in free space WBC systems is replaced with an arrayed waveguide grating (AWG).⁶² The AWG-based WBC system simultaneously realizes beam combining and wavelength control of the laser array elements. In addition, it can increase the spectrum utilization efficiency of the WBC system to improve scalability. As shown in Fig. 13, the functions of the transform lens, diffraction grating, and output coupler in a free space WBC system are all integrated into the passive PIC chip, and the diode laser amplifiers are directly coupled to the input waveguides of the AWG through mode matching. The AWG provides the same function as the diffraction grating, the input waveguide array provides the same function as the transform lens, and the output waveguide array with coatings or DBRs provides the same function as the output coupler. Through hybrid integration, the laser can only obtain feedback from the output waveguide of the AWG, either through a DBR reflector or a straight cleaved facet, to reach the threshold. Because of the spectral filtering property of the AWG, different laser array elements operate at different wavelengths, corresponding to different passbands of the AWG. At the output waveguide, all the optical power from different laser array elements will be combined, just like in the free space WBC system, where the output coupler forces the coaxial propagation of different beams. Since there is no cross-coupling between the laser array elements, every wavelength channel can be operated (turned on/off) independently, which provides spectral control flexibility for the integrated WBC system as a multi-wavelength source. Due to the routing function of the waveguide, the specific location of each laser array element decouples from its operation wavelength, leading to large array scalability and high spectrum utilization efficiency.

Hybrid integrated WBC has been demonstrated in a chip-scale package by using the AWG architecture integrated on the III–V/Si₃N₄ platform, as in Ref. 64. In this realization, the AWG central wavelength was designed near 1 μm, with an FSR of 9.6 nm and 8 channels. The chip-to-chip coupling configuration is shown in Fig. 14. The experimental results of this setup are shown in Fig. 15. The red line (CH1, 3 + CH5, 7) in Fig. 15(a) is the output power sum of the single chip based hybrid laser (channel 1, 3 + channel 5, 7) when each individual group SOA was turned on. The blue line (CH1, 3, 5, 7 WBC) is the combined wavelength of LI when all four SOAs (two chips) were turned on simultaneously. The combined result shows good agreement with the summed result in Fig. 15(a), which indicates that the total WBC on-chip laser power is over 1 W. The spectral results are shown in Fig. 16. The black dashed line (WBC) in Fig. 16(b) refers to the spectrum from the AWG output when all the four SOAs were turned on simultaneously, which perfectly overlaps with the blue line (channel 1,3) and purple line (channel 5,7). This result proves that the AWG-based hybrid integration successfully combined the outputs of different lasers in different wavelength bands. For the combined wavelength LI and spectrum results from channels 2, 4/6, and 8, a similar conclusion can be drawn from Figs. 15(b) and 16(d).

Hybrid integrated chip-scale CBC has been demonstrated on III–V/Si₃N₄ using four SOAs and a 1 × 4 DC array, as in Ref. 64. Multiple 1 × 2 DCs are configured to realize a 1 × 4 DC array and are butt-coupled with two identical SOAs, as represented in Fig. 17. The LI characteristics of the setup are shown in Fig. 18(a), and the output spectrum for an injection current of 620 mA is shown in Fig. 18(b). Each SOA group had a threshold current of 260 mA and a maximum current injection level of 800 mA. The output power of the single chip-based hybrid laser and the coherently combined lasers is 70 and 430 mW, respectively. This system was optimized using water cooling and could be further expanded to include more SOA devices.⁶⁵ 

On-chip optical frequency combs are another application of chip-scale devices that are improved using hybrid integration. These devices allow for mutually coherent, uniformly spaced spectral lines that are precise in both frequency and timing.⁶⁶ This is realized through Kerr nonlinearity in microresonators.⁶⁶ So-called micro-combs suit applications that require strict frequency and timing stability, such as spectroscopy, optical clocks, distance ranging, coherent LiDAR, and telecommunications.^66,67 Hybrid integration allows Kerr microresonators to be integrated into highly compact, scalable, and power efficient devices.⁶⁶ 

Aside from the microresonator approach, frequency combs have traditionally been generated using external solid state or fiber lasers to pump mode-locked laser systems to create frequency combs with spectroscopic coverage from 400 nm to 4 μm.⁶⁸ These solutions, while offering high performance, are bulky and not ideal for applications that require lower power and device footprints. Microresonator-based frequency combs provide a suitable solution for such applications by taking advantage of high quality semiconductor laser sources and utilizing resonant enhancement.⁶⁸ 

Microcombs are produced via the cascaded four-wave mixing process inside a high-Q dispersion-engineered microresonator. A continuous-wave pump wave launched into the microresonator produces a set of equally spaced narrow-linewidth comb lines with a mode spacing determined by the free-spectral range of the microresonator. Appropriate excitation of a micro-comb can produce a phase-locked comb, namely a Kerr soliton microcomb, in which all comb lines share a common phase, which manifests as a periodic soliton pulse train in the time domain. The spectral bandwidth of a soliton microcomb can be controlled by engineering the group-velocity dispersion of the microresonator. With superior coherence, such a phase-locked soliton microcomb exhibits great potential for broad applications ranging from data communication, optical frequency synthesis, coherent microwave generation, spectroscopy, to sensing applications.^69,70

Hybrid microcombs are typically constructed using a Si₃N₄ platform, taking advantage of its wide transparency window and high power compatibility. They can be integrated with either DFB lasers or RSOAs, depending on the tunability requirements.^66,71 Self-injection locking solutions for DFB active sources can also be explored for improved linewidth in the pump laser. The ring resonators on such passive devices are able to achieve ultra-high quality factors with minimal loss.⁶⁶ This property of minimal loss is key for low power operation, which is one of the main advantages of chip-scale microcombs. Hybrid integration allows the RSOA to be coupled into the passive cavity, passing through Vernier resonators to achieve a highly tunable, narrow linewidth, single mode source that is necessary to achieve a high quality frequency comb.⁶⁹ This hybrid approach is advantageous over traditional tabletop frequency comb systems that utilize high-power solid state and fiber lasers as their pump source due to the smaller footprint and lower power requirement of the hybrid system.⁶⁶ 

An example of a hybrid integrated Kerr microresonator is shown in Fig. 19. This work utilized a high quality nonlinear MRR to create a frequency comb through parametric FWM, as previously described. The microresonator was hybrid integrated with a III–V RSOA that was configured as an external cavity diode laser with a double ring resonator to form a tunable, narrow linewidth pump laser. This device had a 9.5 mW output power with a Kerr comb spanning more than 8 THz. The device spectrum is shown in Fig. 20.⁶⁶ More importantly, this device was able to be fully integrated into a battery-operated package due to the low-power nature of the design and its small form factor. A single AAA battery provided 98 mW of electrical pump power and was able to generate 1.3 mW of optical output power matching the single-soliton profile. This output spectrum is shown in Fig. 21.

Another group was able to demonstrate a sub-100 GHz line spacing Kerr comb that operates using less than one watt of electrical power.⁷² This device is shown in Fig. 22. The pump laser is self-injecting and locked to the resonator. It was centered at 1530 nm and was able to produce a maximum optical output of 100 mW with a Gaussian linewidth of 60 MHz. The spectrum of this device is shown in Fig. 23.

As hybrid integration is explored within a plethora of material systems, many functions may be realized that were not previously possible on an integrated platform. This is the case with hybrid integrated Lithium Niobate (LN) devices, which allow for many non-linear optical effects to be observed. One large bottleneck of Si and Si₃N₄ external cavity devices is their limited ability to quickly and efficiently tune frequency. In these devices, frequency is usually tuned by utilizing the thermo-optic effect, as previously described. However, this method is not power efficient and is quite slow (on the order of kHz). However, the Pockels effect allows the frequency to be tuned electro-optically. Electro-optic (EO) tuning is much more power efficient and considerably faster.⁷³ Lithium niobate is the leading material for utilization in EO tuning applications. Therefore, hybrid integration with LN opens many new doors for applications that require fast switching, such as LiDAR, microwave photonics, atomic physics, and AR/VR sensing.⁷³ In addition, it is possible to integrate both high performance lasers and high speed modulators into the hybrid LN platform,⁷⁴ which opens doors for a wide range of applications for integrated laser transmitters.

One example of a Pockels laser is shown in Fig. 24(a). It was built on a lithium niobate on insulator architecture and was butt coupled to a III–V RSOA. This device incorporated two high-Q racetrack resonators in a Vernier configuration to achieve a narrow linewidth, fast switching device by utilizing both the thermo-optic and electro-optic effects. One ring was coarsely tuned using thermo-optic heaters to achieve a tuning range of 20 nm from 1576 to 1596 nm. The electro-optic electrode on the other resonator allowed for modulation frequencies ranging between 0.1 and 50 MHz within the tuning range. Additionally, a record frequency tuning speed of 2.0 EHz/s (2 × 10¹⁸ Hz/s) was achieved at a linewidth of 11.3 kHz.⁷³ The modulation speed measurement is shown in Fig. 24(b).

Another group managed to achieve the first high power integrated transmitter through flip-chip bonding. This was performed by integrating a TFLN electro-optical modulator with a high-power DFB laser using butt coupling through flip-chip bonding and passive alignment. This device featured high linearity, low driving voltage, and low propagation loss. The DFB was capable of outputting more than 200 mW of optical power, and the on-chip optical power was measured to be 19.8 mW at a pump current of 1 A. Therefore, the insertion loss between the DFB and TFLN was conservatively estimated to be 10 dB. The EO modulator is shown below in Fig. 25(a). The modulator featured a bandwidth greater than 50 GHz. The EO phase index (S₂₁) value throughout the entirety of the bandwidth is shown in Fig. 25(b). The V_π was found to be 4.3 V, which demonstrated that the DFB emission could be efficiently modulated in on-chip architecture. This hybrid integrated solution achieves low propagation loss, low drive voltage, and high optical output power in a single flip-chip bonded device.⁷⁴ 

Edge coupled hybrid solutions will continue to develop toward a wide range of practical applications. There is still much ground to cover, primarily in the arena of alignment and fabrication processes. The alignment tolerance of devices must be increased as much as possible, and alignment methods must maintain greater accuracy throughout the integration process. Both passive and active alignment methods discussed earlier are being developed to allow for higher positional accuracy in edge coupling applications, and both are beginning to be integrated into commercial platforms. Some commercial processes employ a rough passive alignment with an accuracy of 5 μm before operating the active device at full power to perform an active alignment accurate to 200 nm.⁷⁵ The active alignment is performed by instantly measuring and feeding back near/far field and beam pointing, divergence, and quality. This helps to solve the issue of scalability in hybrid devices, as many hybrid devices could be manufactured and tested automatically. Once the devices are properly aligned, an epoxy sealing process can be performed to ensure that the alignment is not altered throughout the rest of the fabrication process. This epoxy seal allows multiple devices to be rigidly mounted on a common sub-mount for further packaging and testing. It is also worth noting that automation and epoxy sealing can also be used in smaller settings to greatly improve alignment accuracy and minimize coupling loss.

Coupling loss is being minimized by increasing the quality of the mirror facets of the passive device. Silicon is a hard material, making the dicing process very slow and defect ridden. However, many solutions have been proposed and are currently under research to discover an improved methodology for creating ideal mirror facets. Laser ablation has been proposed as a method to minimize random stress from the standard dicing process using a diamond saw blade to cut and scribe thin silicon wafers. This process would allow a laser to focus on a specific layer of silicon and melt or sublimate the material in the interaction zone. Different laser sources will provide varying fracture strengths, thus allowing the process to be optimized on a case-by-case basis.⁷⁶ It has also been proposed to place the passive device on a non-silicon substrate to increase scribe quality. This would increase the strength of the thin, passive device at the expense of cost and manufacturing ease. For instance, a substrate made of a III–V material such as GaAs would result in a much higher quality mirror facet (as is the case with active devices) due to its ideal cleaving properties.

These strides toward minimizing coupling loss through alignment and fabrication methods are integral to the commercial success of hybrid devices. Lower coupling losses will allow the application base for edge coupled hybrid devices in particular to grow exponentially, reaching into the quantum technology, communication, and spectroscopy markets and greatly advancing the capabilities of chip-level photonics.

To address some emerging applications, chip-scale hybrid laser systems can be further integrated with other optical components such as optical phase arrays, on-chip optical sensors, optical computing units, etc. For example, in Ref. 35, chip-scale, tunable narrow-linewidth hybrid integrated diode lasers are demonstrated through edge-coupling to a silicon nitride PIC at 1, 1.3, and 1.55 μm bands at the same time. These hybrid lasers are further integrated with waveguide surface gratings to create a fully integrated beam steerer. By tuning the hybrid laser wavelength at different bands, wide-angle beam steering is obtained. These types of component integration with hybrid lasers are the key to creating next generation multifunctional PICs that address many novel applications in communication, computing, and sensing.

Hybrid integration of different components/systems, in particular chip-scale laser systems, will be indispensable for practical applications of quantum technology in the real world beyond laboratory demonstration. It is well known that all quantum systems require tremendously supportive classical equipment/components to operate properly, which include lasers, detectors, electro-optic controlling elements, electronic controlling/monitoring units, etc. Practical implementation of quantum systems in the real world would require co-packaging the quantum functioning units with these supporting units. A typical example is quantum photonic integrated circuits (QPICs),^77–79 which are actively developed around the world. These quantum PIC chips currently rely on off-chip lasers for proper operations in a well-controlled laboratory. Such separate operation of the laser source and the quantum chip(s) is extremely susceptible to environmental perturbations. Since QPICs are being developed on a wide range of material platforms, direct chip-to-chip edge coupling provides an excellent solution to realize the hybrid integration of chip-scale classical lasers with quantum chips.

In summary, the ability of hybrid integration to bridge the gap between material platforms makes it a promising method for realizing novel on-chip laser devices and systems. Specifically, edge coupling allows for the realization of many emerging applications and can be scaled into practical industrial applications. Applications like narrow linewidth lasers, beam combination, frequency combs, and Pockels lasers are already being demonstrated via hybrid integration and will only be improved as more research goes into edge coupled hybrid integrated devices. As hybrid technology is improved, more applications will be realized that were previously only possible on large optical tables in lab environments. Hybrid integration has greatly improved the usability of PICs and will only continue to improve their usability going forward.

The authors acknowledge the helpful discussions with Professor Qiang Lin and the use of the Gatech Nanotechnology Research Center Facility and associated support services in the completion of this work.

Army Research Office (Grant No. W911NF-18-1-0176); Office of Naval Research (Grant No. N00014-17-1-2556); National Science Foundation (Grant No. NSF 2137776).

The authors have no conflicts to disclose.

C. Porter: Investigation (equal); Writing – original draft (equal); Writing – review & editing (equal). S. Zeng: Investigation (equal). X. Zhao: Investigation (equal). L. Zhu: Conceptualization (equal); Supervision (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.