An advanced III-V-on-silicon photonic integration platform

Yingtao Hu, Di Liang* and Raymond G. Beausoleil

Keywords: Si photonics; III-V-on-Si laser; photonic integration; epitaxy regrowth

Introduction

With global data traffic growing exponentially， the electrical interconnects have hit a brick wall to satisfy the huge demand of high speed， energy-efficient and cost-effective data transmission in next-generation data centers，high-performance computers， and many emerging applications. By leveraging the mature complementary metal oxide semiconductor （CMOS） manufacturing processes， silicon （Si） photonics has experienced blooming developments over the past 20 years. It now holds the great promise to overcome the data transmission bottleneck problem by providing fast， reliable， and low-cost optical links. However， due to the intrinsic nature of indirect bandgap of Si material， a laser source on Si is the key challenge for Si photonics. Approaches by engineering group IV materials or their alloys have resulted in Ge and Ge-alloy （GeSi and GeSn） lasers on Si1-3， but the laser performances are still far behind the requirements for practical applications. A number of approaches by integrating group III-V materials or devices onto a Si substrate have been widely developed. The hybrid integration which directly assembling finished III-V laser die onto Si photonic chip with flip-chip4technique is the current mainstream commercial solution. This approach has the advantages of independent III-V and Si optimization and qualification and high light-emitting efficiency.But it requires precise assembly to couple light between the two which leads to high packaging cost and it's not suitable for low-cost mass manufacturing and dense integration. A more favorable approach， the heterogeneous integration， is to transfer III-V optical gain material on patterned silicon-on-insulator （SOI） wafer by wafer bonding and then process the large-scale wafer with CMOS compatible fabrication5. This approach allows low-loss evanescent optical coupling from III-V active medium to Si photonic circuits which would reduce packaging costs. It also has the advantage to bond different epitaxial materials onto a single Si substrate6and is able to realize dense integration. A third category is the direct epitaxial growth of III-V layers on Si or SOI substrate to realize monolithic integration. There are two main approaches among this category. One is the selective area growth of a relative small volume of III-V materials on pre-patterned7，8or oxide-hollow-template-assisted9-11silicon substrate. Excellent epitaxial material have been achieved and room temperature distributed feedback （DFB） lasers7， comb lasers8or microcavity lasers have been successfully demonstrated with good performance. However， it still requires extensive R&D to enable electrically pumped lasers and reliability for the selective growth approach. The other approach is planar growth on full wafers by using intermediate buffer layers to minimize dislocations propagating into the active region. As it's potentially to be the ultimate solution for very high density integration， it has regained huge interest12-17. As a result of newly developed novel substrate patterns， intermediate buffer layers， and the use of defect-tolerant active regions， e.g.， quantum dot （QD）， significant progress has been achieved with the demonstrations of lasers with high efficiency13， low threshold14， direct modulation16and good lifetime18. However， several other critical issues associated with this approach currently include potentially more challenges in heteroepitaxy on a SOI substrate over bulk Si ones； difficulty in achieving efficient light coupling from the III/V active region to the Si waveguide due to inevitable several μmthick buffer layers； and extra optical loss when light propagates in the dense dislocation zones. The fourth approach is transfer printing which is transferring III-V gain materials or devices on Si photonic chips by using a soft elastomeric PDMS stamp. While the proof-ofconcept of demonstrations of wideband-emission LEDs and high-speed photodetectors show great potential to be a cost-effective III-V on Si integration technology， it still requires extensive research to address challenges include transfer yield， degraded bonding quality for large area transfer printing and alignment issue. Also large cost to partially or completely fabricate the III-V components on its 50 or 75 mm native substrate is still inevitable.

More recently， an advanced heterogeneous integration approach by combining the advantages of direct monolithic epitaxy and wafer bonding approaches has gained increasing research interest. By using wafer bonding to bring InP-based active material onto a Si substrate and then epitaxially growing InP onto the bonded III-V material to bury the pre-defined mesa， researchers from NTT have developed a lateral p-i-n for lateral-currentinjection heterostructure membrane laser19. Very impressive demonstrations of high performance lasers show great potential of this scheme for compact and high efficiency lasers on Si20-23. However， it requires a relatively complex and unconventional process to make p- and ndoping regions and to form a lateral p-i-n diode structure through an extra regrowth step. This unique process is not readily available in most commercial epitaxial growers. Coincidentally， the advanced heterogeneous integration of epitaxial regrowth of III-V onto III-V-on-Si bonded substrate has been reported from several other research groups while they mainly focus on conventional vertical p-i-n-type of diode lasers on silicon. A research group form Sophia University has demonstrated a double heterostructure （DH） laser23-25and a multiquantum well （MQW） laser26，27on Si substrate with this combined approach. We have studied the quality of regrowth material extensively and demonstrated Fabry-Perot （FP） lasers on SOI substrate with light easily coupled to Si waveguides28-30. Researchers from III-V Lab have reported their results by using epitaxial growth on InP-SiO2/Si bonding substrate as well31，32. These proof-of-concept demonstrations of lasers on Si with this advanced heterogeneous integration approach show its great potential. In this paper， we review the aforementioned research work on this advanced heterogeneous integration platform， particularly formation of conventional vertical p-i-n diode laser structure via thick III-V epitaxial growth. An outlook of this young area is discussed at the end of this paper. This paper is organized as follows. First， the developments of the III-V-on-Si bonding templates for epitaxial regrowth are described. Next， the epitaxy regrowth and materials characterizations is provided. Additionally， the performance characterizations of lasers are detailed. Finally， a discussion and a summary of this advanced heterogeneous integration method are given.

Development of III-V-on-Si bonding templates

The development of III-V-on-Si bonding templates for regrowth is basically a III-V-to-Si wafer bonding process.As the epitaxy temperature of III-V material is normally as high as 650 °C， only a direct wafer bonding technology may be suitable for such regrowth template preparation while the adhesive bonding， e.g.， DVS-BCB bonding， is not as the decomposition temperature of most of the suitable polymers， which is much lower than the epitaxy temperature. Demonstrations of bonding templates from different research groups are all based on direct wafer bonding with some slight differences between each other. Figure 1 is a schematic drawing to summarize the major process steps to form such a vertical p-i-n diode laser structure on Si. We highlight the differences in bonding techniques， epitaxy and post-epi fabrications from different research groups.

The Si substrate could be pure Si or SOI wafer while the later provides the possibility to create Si waveguides or other silicon photonic structures under the III-V template layer. To realize photonic structures in Si underneath the III-V template layer， pre-patterning on the SOI wafer is necessary before the wafer bonding. In our work，we also patterned vertical outgassing channels （VOC） in Si to avoid bonding defect formation， i.e.， gas voids， due to gas product as a result of polymerization reactions at bonding interface. III-V Lab used a 200 nm thermal oxide on top of Si wafer as a hydrogen reservoir to allow diffusion of gas product and subsequently improve bonding quality. While there is no patterns or oxide layer in Sophia University's bonding method but both Si substrate and template wafer underwent a hydrophilic treatment by H2SO4cleaning. The III-V template epitaxy wafer basically includes a bonding layer and a etch stop layer on a native substrate. For the InP based templates， the bonding layer is InP and is normally ndoped to act as a contact layer of an active device. The InGaAs etch stop layer is used to selectively etch off InP substrate in the HCl-based solution， and then result in an ideal surface for regrowth once it is selectively removed.After both top surfaces of Si/SOI and III-V template wafers are cleaned and activated with O2plasma， bonding dielectric deposition or H2SO4solution treatment，the III-V wafer is bonded to Si/SOI wafers. By applying pressure and annealing at 300 - 400 °C， strong covalent bonds are formed. The next is to remove the III-V substrate and etch stop layer. Upon selectively removing the etch stop layer to expose the n-InP template layer， the bonding template is ready for epitaxial regrowth. The pictures from left to right in Fig. 2 show the fabricated InP-on-Si bonded wafers from our work， III-V Lab and Sophia University's work， respectively. We carried out the bonding with a quarter of 2 inch InP wafer to a SOI wafer， while III-V Lab and Sophia University realized the growth templates by bonding a 4 inch full InP wafer on a 6 inch SiO2/Si wafer， and a 2 inch full InP wafer on a 2 inch Si wafer， separately. All bonded growth templates were verified to be robust enough for elevated epitaxial growth temperature.

Fig. 1 | A schematic drawing of the process flow for the advanced heterogeneous integration by using regrowth on III-V-on-Si bonding template.

After the development of bonded growth templates，the following is epitaxial regrowth of lattice-matched IIIV device layer stack with metalorganic vapor phase epitaxy （MOVPE）. Finally， the wafer experienced subsequent processing of III-V laser devices fabrication. We note that the type of Si-based substrate （SOI or Si） and the specific design of III-V layers determine the optical mode profiles. By using pre-patterned SOI and thin III-V template layer， a hybrid mode can be realized which enables evanescent optical coupling from III-V active layers to Si. When a pure Si substrate is used， thick n-layer is necessary to allow the optical mode mainly confined in III-V without leaking into the Si substrate. The bonded growth templates developed from these three research groups are suitable for regrowing laser structures and making vertical injection lasers. Considering significant difference between lateral p-i-n diode laser structure in NTT's solution and vertical ones from the rest of 3 groups， we focus on reviewing progress to demonstrate vertical p-i-n lasers on this platform in this paper.

Epitaxial regrowth and material characterizations

Epitaxial regrowth

In our work， we conducted the epitaxial regrowth of a laser structure on the bonding template with a MOVPE system in a commercial vendor. The growth temperature is 600 °C. The epitaxy structure， as shown in Fig. 3（a），is similar to previously demonstrated heterogeneous laser structures33with low-loss evanescent coupling to the Si photonic circuits. The InGaAsP-based MQW is designed for light emission at 1.31 μm. The total epitaxial thickness is ～2 μm. In III-V Lab's work， an AlGaInAsbased MQW laser with total thickness of 3 μm （Fig. 3（b））was grown at 610 °C in a MOVPE system. In Sophia University's work， a InGaAsP-based MQW laser structure was grown in a low pressure MOVPE system at an epitaxy temperature of 650 °C. The epitaxy structure is shown in Fig. 3（c）， which shows the total regrowth thickness over 2 μm. Though the laser structure designs are different， the thicknesses are all above 2 μm for conventional vertical injection lasers. The introduction of the bonded template eliminated lattice and polarity mismatches between the Si and III/V material， while the thermal mismatch still exists. The impact of this remaining mismatch can be studied by comparing the epitaxy quality on the bonded substrate and that on a native substrate. All experiments have included the same growth on native InP substrates for comparisons of material quality or device performance.

Fig. 2 | Pictures of the fabricated InP-on-Si bonded wafers from HPE, III-V Lab and Sophia University. Figure repoduced with permisson from: (a) ref.29, under a Creative Commons Attribution 4.0 International License; (b) ref.31, (c) ref.25, John Wiley and Sons.

Fig. 3 | Epitaxial regrowth laser structures on bonded templates from HPE, III-V Lab and Sophia University. Figure repoduced with permisson from: (a) ref.29, under a Creative Commons Attribution 4.0 International License; (b) ref.32, (c) ref.26, IEEE.

Surface characterizations

Surface roughnesses of the regrowth epitaxy were characterized with atomic force microscopy （AFM）， we obtained a root mean square （RMS） value of 0.2 nm while that of III-V Lab's growth is 0.7 nm. Figure 4（a） and 4（b）show the AFM images of the epitaxy regrowth of our work and III-V Lab's work. Particularly， we obtained an identical RMS value of 0.2 nm on the substance epitaxially grown on the reference InP substrate. This indicated a comparable material quality of subtance epitaxially grown on a bonded substrate versus on a native substrate. Surface quality of the epitaxy material was estimated with Nomarski imaging， as shown in Fig. 4（c），which shows that the surface is smooth while with 2.5%of bonding caused voids occupancy in Sophia University's work. More study27shows that the density of bonding caused voids which intuitively represents epitaxy quality affects the lasing threshold current density accordingly.

Dislocation characterizations

Another routine of characterization is to study material cross-section with transmission electron microscopy（TEM） to investigate possible defects. Figure 5（a） and 5（b） show the cross-sectional TEM images of the MQW epitaxy on the bonded substrate from our work， while Fig. 5（c） is the cross-sectional STEM and Fig. 5（d） is the cross-sectional TEM images from III-V Lab's work. No defect was observed in both TEM specimens （Fig. 5（a）and 5（c）） and MQW layers with good contrast are clearly exhibited in the high-magnification TEM images （Fig.5（b） and 5（d）） from both work. Particularly in our work，since we were unable to find any threading dislocation（TD） in the epitaxy from cross-sectional TEM imaging on TEM specimens whose plan-view area is around 0.75 μm × 15 μm， we then performed plan-view TEM imaging which provides a relatively large area （30 μm × 12 μm） observations. Nevertheless， no TDs but some misfit dislocations were observed with the plan-view TEM.Those misfits were observed at the interface between the upper SCH and InP cladding layer and within the InP cladding layer close to the interface by tilting cross-sectional TEM specimens. The existence of the misfits is explained as the result of thermal strain in the InP bonding template and the following epitaxy due to the difference in their thermal expansion coefficients and implies that the TDs are far away across the observed plan-view TEM area.

Fig. 4 | AFM images and Nomarski microscope image of the epitaxial regrowth on bonded substrate from HPE, III-V Lab and Sophia University. Figure repoduced with permisson from: (a) ref.29,under a Creative Commons Attribution 4.0 International License; (b) ref.32, (c) ref.26,IEEE.

Fig. 5 | Cross-sectional TEM (or STEM) images of the MQW or bulk epitaxy on the bonded substrate from HPE and III-V Lab.Figure repoduced with permisson from: (a, b) ref.29, under a Creative Commons Attribution 4.0 International License; (c) ref.32, IEEE; (d)ref.31, John Wiley and Sons.

Fig. 6 | (a) The electron channeling patterns corresponding to the three-beam (400) and (220) imaging conditions that were used in ECCI characterization. (b) A representative ECCI image with only one TD. Figure repoduced with permisson from ref.29, under a Creative Commons Attribution 4.0 International License.

Furthermore， we used electron-channeling contrast imaging （ECCI） to quantify the dislocation density in the plan view. Figure 6（a） shows the electron channeling patterns corresponding to the three-beam （400） and （220）imaging conditions that were used. Figure 6（b） shows a representative image with only one TD. A total of 20 TDs were counted in 100 images with a total mapping area of 100 × 14.5 × 14.5 μm2. This led to a dislocation density of 9.5 × 104cm-2， only one order of magnitude higher than that of native InP substrates and two orders of magnitude lower than state-of-the-art values of conventional monolithic growth with a thick buffer layer34. Additional ECCI investigations observed small areas with more concentrated TDs， which were likely caused by bonding voids or dirt particles. We applied the same ECCI process to an InP witness sample for comparison but could not see any TD due to a very low dislocation density for epitaxy on the native substrate.

Photoluminescence (PL) measurements

Figure 7 shows the PL measurements at room temperature （RT） for the epitaxy on both InP and the bonded substrate from the three groups. We found that the measured PL intensity of the MQW from the bonded substrate sample was 2.53 times higher than that from the InP substrate sample， very similar to the 2.7 times higher PL intensity reported in reference21. This is mainly because strong reflections from the Si and buried-oxide layers couple with reflection from the top III/V surface to form a resonance cavity that enhances the PL pump efficiency in the Si substrate sample， as shown in the schematic drawing in the inset of Fig. 7（a）. III-V Lab's PL measurements （Fig. 7（b）） show the similar phenomenon that the intensity from epitaxy on bonded InP/SiO2/Si substrate is 5 times higher than that from the InP substrate. As there is no material interface with strong refractive index contrast to cause reflections in Sophia University's case， they reported that the PL intensity from epitaxy on bonded substrate is equivalent to that on the reference InP substrate. The observations of no significant difference on the peak wavelengths， the full widths at half maximum （FWHMs） and PL profiles from all work indicated that the epitaxy quality on the bonded substrate is comparable with that on the InP substrate.

Fig. 7 | The PL measurements at room temperature for the epitaxy on both InP and the bonded substrate from HPE, III-V Lab and Sophia University. Figure repoduced with permisson from: (a) ref.29, under a Creative Commons Attribution 4.0 International License; (b) ref.32,(c) ref.26, IEEE.

X-ray diffraction (XRD) measurements

XRD measurement on the epitaxy sample is a method to further investigate the epitaxy quality. Three plots in Fig. 8 are the measured ω-2θ curves both for the epitaxy on bonded and on native substrates for comparison from our work （Fig. 8（a））， III-V Lab （Fig. 8（b）） and Sophia University （Fig. 8（c））. All the three comparisons show that the MQW satellite peaks on both curves exhibit almost identical characteristic signatures， thus indicating good MQW structural similarity and integrity. This means that the epitaxy on bonded substrates has similar high crystal quality to that on native substrates despite total epitaxial thickness is over the critical thickness. Detailed XRD rocking curve analysis and high-resolution omega scans in Fig. 8（a） provide more information of the epitaxy including the strain condition， dislocation density estimation which is a further evidence to our ECCI observations of one order of magnitude higher threading dislocation density （TDD） in the epitaxy on the bonded substrate than that on the native InP substrate29.

Laser demonstrations

III-V-on-SOI FP laser with hybrid waveguide facets

In our work， we treated the regrowth wafer as a conventional heterogeneous wafer and applied the same fabrication procedure to fabricate two types of Fabry-Perot （FP）lasers. One type is FP laser with III-V-on-Si hybrid waveguide facets as front and back mirrors and the other is FP laser with Si waveguide laser facet mirrors. Figure 9（a-c）show a microscope image of a FP laser with hybrid facets， schematic of the device cross-section， and SEM of the hybrid facet， respectively. Figure 9（d） and 9（e） show the light-current-voltage （LIV） curves at RT （20 °C） and LI curves up to a stage temperature of 40 °C， both of which are under the pulsed injection mode （0.5 μs， 0.25%duty cycle）. Large diode voltage in Fig. 9（d） indicates a large series resistance due to fabrication imperfection.The simulated fundamental TE mode profile is shown in the inset of Fig. 9（e）. The 1.9-mm-long device starts lasing at 61.8 mA and emits 4.2 mW from a single facet under a 120-mA current injection， corresponding to a reasonable threshold current density of 813 A/cm2and an overall slope efficiency of 0.14 W/A. The observation of lasing at approximately 1313 nm under the pulsed mode at RT （Fig. 9（f）） matches our MQW design well. Figure 9（g） shows continuous-wave （cw） LI curves up to a stage temperature of 20 °C， with increased thresholds due to device joule heating.

III-V-on-SOI FP laser with Si waveguide facets

To prove convenient integration with other Si photonic circuits， the FP laser with Si waveguide facets by using two 50-μm-long III/V-to-Si tapers to evanescently couple hybrid lasing mode to Si waveguide before reaching polished Si mirrors were demonstrated in our work. Figure 10 （a） shows a microscope image of a FP laser with Si waveguide facets and a SEM image of a III/V-to-Si taper.Figure 10 （b-c） show the LIV at RT and the LI curves up to a stage temperature of 35 °C under the same pulsed mode. The threshold current density of a 2.1-mm-long device with a 2.0-mm-long active region was calculated to be 1125 A/cm2. The simulated fundamental output mode profile at the Si facet is shown in the inset of Fig. 10（c）. We note that multiple fabrication issues including large series resistance and quite high internal loss due to unexpected high p-doping in p-InP cladding， etc.，significantly limiting the device performance. However，the demonstration of this laser exhibits the great advantage of readily light coupling from III-V active layer to Si waveguide over other heterogeneous or monolithic works which use pure Si substrate or thick III-V buffer layer where excellent passive optical waveguides in Si is not employed.

Fig. 8 | XRD measurements on the epitaxy samples that from the three different research groups. Figure repoduced with permisson from:(a) ref.29, under a Creative Commons Attribution 4.0 International License; (b) ref.31, John Wiley and Sons; (c) ref.27, Elsevier.

Fig. 9 | (a) A microscope image of a FP laser with hybrid facets. (b) Schematic drawing of the device cross-section and (c) SEM of the hybrid facet. (d) RT pulsed LIV. (e) Pulsed LI up to 40 °C (inset: mode profile at facets). (f) Device spetrum. (g) cw LI up to 25 °C. Figure repoduced with permisson from ref.29, under a Creative Commons Attribution 4.0 International License.

FP laser on Si with III-V facets

III-V Lab demonstrated broad-area FP lasers on both the bonded growth template （Fig. 3 （center）） and native substrate. Figure 11（a） shows a SEM image of the fabricated laser on bonded substrate. The mesa width is more than 50 μm and FP cavity length is 660 μm. The pulse mode current density-light （J-L） characteristics at 20 °C of broad-area lasers on bonded and native substrate are shown in Fig. 11（b）. The threshold current densities（TCD） at 20 °C of the laser on bonded substrate and on native substrate are obtained as 400 A/cm2and 700 A/cm2， respectively. The TCD differences could be caused by the difference on the Zn-doping profiles or cleaved facets variations. The measured slope efficiency are 0.092 and 0.095 W/A for the laser on bonded substrate and on native substrate， respectively. The facts that the two TCD values are very close and quantum efficiencies are similar indicate that the epitaxy quality grown on a bonded substrate is comparable to that on a native substrate. Figure 11（c） shows the pulse mode J-L characteristics measured at different temperatures up to 50 °C for the laser on bonded substrate. Figure 11（d） shows the TCD evolution against temperature for the laser on bonded （blue） and on native substrate （red）. The laser characteristic temperature T0is extracted to be 53 °C for both lasers. This result confirms that material quality is preserved for several μm-thick epitaxial growth on a bonded substrate.

Fig. 10 | (a) A microscope image of a FP laser with Si waveguide facets and a SEM image of a III/V-to-Si taper. (b) RT pulsed LIV (inset: microscope image of the device), (c) pulsed LI up to 40 °C (inset: mode profile at facets). Figure repoduced with permisson from ref.29, under a Creative Commons Attribution 4.0 International License.

Fig. 11 | (a) SEM image of the fabricated FP laser on InP-on-Si substrate. (b) J-L characteristics in pulse injection mode at 20 °C: laser on bonded substrate (solid line) and the laser on InP (dash line). (c) J-L characteristics in pulse injection mode measured at different temperatures for the laser on bonded substrate. (d) Threshold current density evolution against temperature for the laser on bonded substrate (blue) and on InP(red). Figure repoduced with permisson from ref.32, IEEE.

Fig. 12 | (a) Measured lasing spectra of five FP lasers under C-W operation for a driving current of 100 mA at 20 °C. (b-d) L-I characteristics under C-W operation for different temperatures for the lasers emitting at 1515 nm, 1580 nm and 1635 nm. Figure repoduced with permisson from ref.35, IEEE.

III-V-on-Si FP laser array

III-V Lab also demonstrated FP laser array covering 155 nm spectra band by selective area growth （SAG） on the bonded InP-on-Si substrate35. By defining the widths of etching-exposed SiO2mask that on both sides of fixed width of InP template strips， the thicknesses of the regrown MQW can be tuned accordingly which leads to variations on the MQW photoluminescence emission.Figure 12（a） shows the measured lasing spectra of five FP lasers based on five different selectively grown MQW active regions. The spectra were measured under cw operation at a driving current of 100 mA at 20 °C for the 500 μm-long lasers. It can be seen that a 155 nm spectra band has been achieved by the single SAG on adjusting the widths of template openings. Figure 12（b-d） shows the cw operation LI curves at different temperatures from 20 °C to 70 °C of lasers lasing at 1515 nm， 1580 nm and 1635 nm. Decent performance of a maximum of 20 mW for a 200 mA driving current and up to 70°C for the lasers emitting from 1515 nm to 1610 nm were obtained on these SAG lasers. In addition， III-V Lab is adapting their platform to use SOI instead of Si substrate for light coupling into Si waveguides.

FP laser on Si by using Si substrate as one contact

Sophia University has demonstrated InGaAsP-InP bulk DH lasers36，37and MQW lasers26，27on the regrowth on bonded substrates platform. In their laser structure， current was injected through the bonding interface between InP and Si by using Si substrate as the n contact. A typical laser structure is shown in Fig. 13（a）. The laser cavities were formed by the manual cleavage of both facets without additional coating on them. The lowest threshold current density for DH lasers at RT and pulse mode is 1800 A/cm2lasing at a wavelength of 1.5 μm24while the threshold current density is 2850 A/cm2for the MQW lasers. Figure 13（b） shows the typical LI characteristics of the MQW laser at various temperatures. The laser was measured at pulse current mode with pulse width of 0.5 μs and duty ratio of 0.05%. Figure 13（c）shows a lasing spectrum for the MQW laser at an injection current density of J=6.83 kA/cm2， where the lasing wavelength was 1383 nm. They compared lasing characteristics among four different type of lasers： DH bulk laser on bonded substrates and on native substrates，MQW laser on bonded and on native substrates. Figure 13（d） summarized the temperature dependence of the threshold current density for the four types of lasers. We can see that the threshold current density is the same for DH lasers on the two different substrates while it's quite different for the MQW lasers on the two substrates. The increasing trend in the threshold current density with the temperature is almost the same for the DH lasers on bonded and native substrates while it's different for the MQW lasers on the bonded and native substrates. As the density of the void in directly bonded InP/Si substrate caused by the growth of the SCH-MQW LD was observed slightly higher than that in the InP/Si substrate with the bulk LD， the difference on the laser performance could be explained as the bonding quality influencing the laser performance significantly as the current has to inject through the bonding interface. Another possible reason might be that MQW-based active region is much more dependent on material quality than that of bulk material. While the similarity of the DH lasers on the two substrates indicates that epitaxy quality on a bonded substrate that with good bonding quality could be comparable to that on a native substrate.

Fig. 13 | (a) A typical laser structure. (b) The typical I-L characteristics of the MQW laser on the bonded substrate at various temperatures. (c) A lasing spectrum for the MQW laser at an input current of J=6.83 kA/cm2. (d) The temperature dependence of the threshold current density for the DH bulk lasers and MQW lasers on bonded substrates and on native substraes. Figure repoduced with permisson from: (a) ref.26, IEEE; (b, c, d)ref.27, Elsevier.

As a comparison， we summarized the main features including bonding methods， epitaxy characterizations and laser performance of the demonstrations from the reviewed three research groups in Table. 1. It can be seen that III-V lab has achieved the best laser performance while our work demonstrated the ability of light coupling from active region to Si. We believe that more achievements would be inspired by these proof-ofconcept laser demonstrations in the future.

Discussion and summary

Research progress from different research groups reviewed above has indicated growing interest and huge potential in this novel concept of regrowth on III-V-on-Si bonded substrate. Bonded growth templates with variations in Si side have been successfully developed， and high quality of III-V MQW epitaxy has been confirmed by different characterization methods. Proof-of-concept laser demonstrations further concluded that the electrical and optical properties are well preserved to be comparable to the growth on a native substrate. According to our detailed discussions in our previous work29， the regrowth on bonded template method is of huge interest and high value for further R&D for the following main reasons.

1） High-quality epitaxy with low dislocation density on Si. The approach of epitaxy on bonded template eliminates two of the three major root causes for dislocations：lattice and polarity mismatches between the substrate（e.g.， Si） and the function material （e.g.， III/V） from epitaxial growth. The thermal mismatch between the substrate and template material would still cause defects in the regrown materials， but it is measured to be at a significantly low level. According to the aging tests of InAs QD lasers on Si near room temperature18， a reduction in the TDD from 108cm-2to 106cm-2can extend the laser lifetime from a few months to over 100 years. It is reasonable to expect that lasers from the regrowth on the bonded template with even lower dislocation density would eliminate the defect-induced lifetime concerns for all practical applications.

2） It's a generic method for many other heterogeneous material combinations. As the epitaxy quality is high and thickness is up to several μm， this growth on bonding template approach could be a generic method for many other heterogeneous material combinations.The substrate could be semiconductors （e.g.， Si）， dielectrics （e.g.， Si3N4）， metals， etc.， and the top grown material could be bulk materials， QWs， QDs or other nanostructures. Sequential multiple growths on the same template can be a routine procedure to enable advanced，large wafer-scale， dense photonic integration. A good example in silicon photonics is the integration of light sources， amplifiers， modulators and detectors on a single chip with close proximity and low coupling loss by implementing multiple selective regrowth on a single bonding template instead of bonding three or four types of epitaxial structures on each chip38. We believe that various functional materials grown on a single bonded substrate could be emerging as large-scale process.

3） It's cost competitive over the other existing III-Von-Si integration approaches. Table 2 is a qualitative comparison of the production and operation costs of the same diode laser， built using different III/V-on-silicon integration approaches. All costs， including the substrate material （Si and InP substrate）， III/V epitaxy， device fabrication （bonding and device fab）， chip packaging and operation， are compared separately among the four integration approaches. The marks （x， xx， xxx） indicate the relative cost level among the four integration approaches within each column but do not indicate the cost differences over columns. The epitaxy plus bonding approach is enabling material growth on much larger and cheaper substrate than its native substrate. The natural feasibility of multiple regrowth on the same growth template can lead to very high-density integration， thus minimizing the chip size and packaging effort. Therefore， thebonding plus epitaxy approach can be very cost competitive overall.

Table 1 | A comparison of the main features in the demonstrations from the three research groups.

Table 2 | Qualitative comparison of production and operation costs of the same diode laser built on different III/V-on-silicon integration approaches.

As direct wafer bonding is critical to this platform based on regrowth on bonded substrate， efficient， reliable and high-quality large area wafer bonding technology is needed for this approach to be adapted in standard CMOS foundries towards commercialization for practical applications. It still needs extensive research and development to conduct multiple and selective regrowth on the template to fully explore the capability and potential of this platform. In addition， investigating options to reuse the III-V substrate for template preparation41is another interesting subject to study for maximizing cost saving nature of this platform. In summary， epitaxy regrowth on bonded substrate is emerging as a novel integration method for III-V-on-Si photonic integration. It is a general approach for combining different materials onto various substrates and particularly suitable for on-chip light source and other functional devices for Si photonics. Current high-quality epitaxy material regrowth and proof-of-concept pulsed and cw lasers demonstrations will inspire more R&D efforts on this integration platform. It is worth looking forward to in a number of all-in-one photonic integration applications for its low cost， high scalability， and high integration density features.