GaAs-on-Insulator Platform Overview

• GaAs-on-insulator is a high-index-contrast platform that places a thin GaAs layer on an insulating oxide to achieve strong optical confinement and enhanced nonlinear interactions.
• It supports efficient frequency conversion, Kerr comb generation, and integration of active quantum emitters through heterogeneous bonding.
• Precise nanofabrication and surface chemistry control are key to reducing losses and ensuring uniform device performance across various architectures.

Searching arXiv for relevant GaAs-on-insulator papers to ground the article. GaAs-on-insulator denotes a class of integrated semiconductor and photonic platforms in which a thin gallium-arsenide or GaAs-family device layer is placed on a low-index insulating undercladding, most commonly $\mathrm{SiO_2}$, so that the high refractive index, direct bandgap, electro-optic response, and large optical nonlinearities of III–V arsenides can be exploited in a high-contrast planar geometry. In the literature, the term spans several distinct realizations: fully oxide-clad heterogeneously bonded GaAs waveguides for nonlinear optics (Chang et al., 2018), AlGaAs-on-insulator nanophotonic circuits for Kerr comb generation (Pu et al., 2015, Xie et al., 2020), electrically contacted GaAs membranes with embedded quantum dots bonded onto silica-based substrates and coupled to SiN circuits (Salamon et al., 6 Aug 2025), and suspended GaAs membranes that function as a practical surrogate for true GaAs-on-insulator high-contrast photonics (Jiang et al., 2019). Some works are only indirectly relevant: a [111]-grown GaAs/Ge/GaAs quantum well can be regarded as a GaAs-compatible insulating heterostructure platform in the quantum-well sense rather than a conventional oxide-isolated wafer (Zhang et al., 2013), while GaAs-containing superlattice-on-insulator proposals extend the concept to atomically engineered electro-optic cores on oxide (Leonardis et al., 2022). Across these variants, the common idea is that moving GaAs or related III–V arsenides from low-contrast native-substrate geometries into an insulator-supported thin-film format converts attractive material properties into compact, dispersion-engineerable, low-power integrated devices.

1. Platform concept and scope

The defining architectural move in a GaAs-on-insulator platform is the replacement of a native high-index substrate or weakly confining semiconductor undercladding with an insulating environment, typically $\mathrm{SiO_2}$, so that the optical mode is concentrated in a submicron GaAs-family core. In the heterogeneously integrated nonlinear platform of (Chang et al., 2018), a thin GaAs guiding layer is bonded onto an oxidized silicon wafer and fully surrounded by $\mathrm{SiO_2}$. In the AlGaAs-on-insulator implementations of (Pu et al., 2015) and (Xie et al., 2020), an epitaxial AlGaAs film is transferred onto oxide and then patterned into nanowaveguides and microresonators. In the quantum-photonic platform of (Salamon et al., 6 Aug 2025), a $170~\text{nm}$ GaAs membrane containing self-assembled InAs quantum dots and a p-i-n junction is adhesively bonded via an approximately $260~\text{nm}$ epoxy layer onto a silica/silicon or silica/SiN/silicon host.

This high-contrast geometry is used because conventional GaAs/AlGaAs waveguides on native substrates have weak vertical confinement. One paper quotes the prior vertical refractive-index contrast as $\Delta n \approx 0.2$, which limited intensity, dispersion engineering, and effective nonlinearity (Chang et al., 2018). By contrast, when GaAs or AlGaAs is embedded in low-index dielectric surroundings, optical modes become much smaller, waveguide geometry exerts stronger control over dispersion and birefringence, and compact bends, couplers, and resonators become feasible (Chang et al., 2018, Xie et al., 2020).

The literature also uses related but nonidentical architectures. Suspended GaAs membranes above an air gap, released from an AlGaAs sacrificial layer and later encapsulated with oxide, are not permanent GaAs-on-insulator stacks, yet they reproduce many of the same high-index-contrast design idioms and therefore function as a suspended analogue of GaAs-on-insulator high-contrast photonics (Jiang et al., 2019). A different extension appears in short-period superlattice-on-insulator proposals, where the active core is not homogeneous GaAs but a GaAs-containing superlattice such as $(\mathrm{GaAs})_N/(\mathrm{Ge}_2)_M$ or $(\mathrm{ZnSe})_N/(\mathrm{GaAs})_M$, bonded onto oxide and used for electro-optic devices (Leonardis et al., 2022). By contrast, the [111]-grown GaAs/Ge/GaAs quantum-well topological-insulator proposal is not a buried-oxide or bonded GaAs-on-insulator wafer; its relevance is that GaAs acts as a confining wide-gap barrier around an electronically active Ge well, producing a quantum-well insulator with topological edge conduction rather than an oxide-isolated PIC substrate (Zhang et al., 2013).

A useful editorial distinction is therefore between direct GaAs-on-insulator, GaAs-family-on-insulator, and GaAs-compatible high-contrast analogues. The first includes oxide-clad bonded GaAs thin films (Chang et al., 2018, Stanton et al., 2019, Salamon et al., 6 Aug 2025). The second includes AlGaAsOI, whose platform logic is largely shared with GaAsOI (Pu et al., 2015, Xie et al., 2020, Placke et al., 2020). The third includes suspended GaAs membranes and GaAs-based quantum wells whose insulating functionality is architectural rather than oxide-based [(Jiang et al., 2019); (Zhang et al., 2013)].

2. Material physics and why the insulator matters

The central attraction of GaAs and related III–V arsenides is the coexistence of high refractive index, direct-bandgap optoelectronic functionality, non-centrosymmetric $\chi^{(2)}$, and large $\chi^{(3)}$. For GaAs or AlGaAs, one nonlinear photonics comparison lists $\mathrm{SiO_2}$0 pm/V and $\mathrm{SiO_2}$1, alongside a compact optical mode size of approximately $\mathrm{SiO_2}$2 (Chang et al., 2018). For $\mathrm{SiO_2}$3, the reported refractive index is $\mathrm{SiO_2}$4, the bandgap is $\mathrm{SiO_2}$5–$\mathrm{SiO_2}$6, and the derived material nonlinear index is $\mathrm{SiO_2}$7 (Pu et al., 2015). For $\mathrm{SiO_2}$8, another platform paper reports $\mathrm{SiO_2}$9 at $\mathrm{SiO_2}$0 and a bandgap of approximately $\mathrm{SiO_2}$1 (Xie et al., 2020). A quantum-emitter paper states that GaAs has $\mathrm{SiO_2}$2 while $\mathrm{SiO_2}$3 has $\mathrm{SiO_2}$4, which explains why a silica-supported GaAs membrane retains strong confinement (Salamon et al., 6 Aug 2025).

The insulator matters because these material properties only translate into exceptionally strong device-level interactions when the effective mode area becomes small. The standard nonlinear waveguide parameter is written as

$\mathrm{SiO_2}$5

In AlGaAsOI, the small $\mathrm{SiO_2}$6 produced by the insulator geometry yields $\mathrm{SiO_2}$7 for a $\mathrm{SiO_2}$8 waveguide, with $\mathrm{SiO_2}$9 (Pu et al., 2015). A simulation-driven study makes the same point more generally: the semiconductor’s material nonlinearity is already large, but the on-insulator geometry increases effective nonlinearity by reducing $170~\text{nm}$0, strengthening mode overlap and making waveguide dispersion dominate material dispersion (Placke et al., 2020). For a representative $170~\text{nm}$1 AlGaAsOI waveguide, the mode area is $170~\text{nm}$2, and the resulting nonlinear coefficient is stated to be more than 400 times higher than in SiN waveguides (Xie et al., 2020).

This same confinement logic underlies $170~\text{nm}$3 devices. In thin-film GaAs-on-insulator waveguides for second-harmonic generation, modal birefringence and geometry-dependent effective index are used to satisfy phase matching between a TE pump near $170~\text{nm}$4 and a TM second harmonic near $170~\text{nm}$5 (Stanton et al., 2019). In thin-film GaAs bonded to oxide for $170~\text{nm}$6 SHG, the phase-matching condition is expressed through

$170~\text{nm}$7

with perfect phase matching when $170~\text{nm}$8 (Stanton et al., 2019). In an earlier GaAsOI SHG platform, the same modal-matching logic is expressed as $170~\text{nm}$9, or equivalently $260~\text{nm}$0 (Chang et al., 2018).

For electro-optic and quantum functionality, the insulating support contributes more than optical confinement. A bonded GaAs membrane on silica is mechanically more robust than a fully suspended beam and remains compatible with CMOS-style processing and with buried waveguide platforms such as SiN (Salamon et al., 6 Aug 2025). A review of GaAs quantum photonics explicitly argues that future high-density integration will require replacing conventional GaAs/AlGaAs waveguides by a GaAs/$260~\text{nm}$1 structure, with GaAs membranes bonded onto $260~\text{nm}$2-coated Si (Dietrich et al., 2016). This suggests that the insulator is not merely a passive cladding but an integration substrate that connects GaAs optoelectronics to larger heterogeneous photonic systems.

3. Realizations of GaAs-family-on-insulator

The major experimentally demonstrated GaAs-family-on-insulator variants differ in bonding method, cladding symmetry, and targeted function.

Realization	Core architecture	Primary function
GaAsOI nonlinear waveguides	Thin GaAs film bonded to oxidized Si and fully or partially oxide-clad	$260~\text{nm}$3 frequency conversion (Chang et al., 2018, Stanton et al., 2019)
AlGaAsOI nanophotonics	Thin AlGaAs layer transferred to oxide and patterned into nanowaveguides and resonators	$260~\text{nm}$4 combs, OPO, low-loss resonators (Pu et al., 2015, Xie et al., 2020)
Bonded GaAs quantum photonics	$260~\text{nm}$5 GaAs membrane with QDs and p-i-n junction on silica-based host	Coherent single-photon generation and hybrid GaAs/SiN routing (Salamon et al., 6 Aug 2025)
Suspended GaAs high-contrast PICs	Released 220 nm GaAs membrane above air gap with oxide above after encapsulation	Passive PIC components in a GOI-like surrogate (Jiang et al., 2019)
GaAs-containing SLOI	Short-period GaAs-based superlattice bonded onto oxide	Pockels modulation and switching proposals (Leonardis et al., 2022)

The heterogeneously integrated GaAs nonlinear platform of (Chang et al., 2018) begins with a 150 nm GaAs film on a 500 nm $260~\text{nm}$6 sacrificial/etch-stop layer atop a 500 $260~\text{nm}$7 GaAs substrate. After deposition of a 5 nm sputtered SiN layer to improve bond strength, the III–V die is bonded to a silicon wafer carrying $260~\text{nm}$8 thermal $260~\text{nm}$9, with $\Delta n \approx 0.2$0 vertical channels at $\Delta n \approx 0.2$1 spacing to release gas during bonding. Substrate removal by polishing and wet etching leaves a thin-film GaAs layer that is later oxide overclad, producing a fully oxide-clad GaAs nanowire on oxidized Si (Chang et al., 2018).

The high-yield wafer-scale GaAsOI SHG platform of (Stanton et al., 2019) uses MBE growth of a 158 nm GaAs layer on 150 nm $\Delta n \approx 0.2$2, direct bonding to a Si wafer with $\Delta n \approx 0.2$3 thermal $\Delta n \approx 0.2$4, and substrate removal to create partially etched GaAs ridges on oxide with air top cladding. The transferred film shows approximately $\Delta n \approx 0.2$5 nm RMS roughness over $\Delta n \approx 0.2$6, thickness uniformity of $\Delta n \approx 0.2$7 or $\Delta n \approx 0.2$8 nm within a 20 mm radius, and nearly perfect transferred-film yield over a 76 mm wafer (Stanton et al., 2019).

In AlGaAsOI nonlinear photonics, one realization uses a 320 nm $\Delta n \approx 0.2$9 layer bonded via 90 nm benzocyclobutene onto a second substrate with 10 nm silica, with 3 $(\mathrm{GaAs})_N/(\mathrm{Ge}_2)_M$0m PECVD $(\mathrm{GaAs})_N/(\mathrm{Ge}_2)_M$1 and final $(\mathrm{GaAs})_N/(\mathrm{Ge}_2)_M$2 overcladding (Pu et al., 2015). A later low-loss version uses a 400 nm $(\mathrm{GaAs})_N/(\mathrm{Ge}_2)_M$3 film transferred by heterogeneous wafer bonding onto silicon with $(\mathrm{GaAs})_N/(\mathrm{Ge}_2)_M$4 thermal $(\mathrm{GaAs})_N/(\mathrm{Ge}_2)_M$5, plus thin $(\mathrm{GaAs})_N/(\mathrm{Ge}_2)_M$6 passivation and $(\mathrm{GaAs})_N/(\mathrm{Ge}_2)_M$7 PECVD $(\mathrm{GaAs})_N/(\mathrm{Ge}_2)_M$8 overcladding (Xie et al., 2020). These two papers collectively establish AlGaAsOI as a GaAs-family-on-insulator platform with both extreme effective nonlinearity and ultrahigh $(\mathrm{GaAs})_N/(\mathrm{Ge}_2)_M$9 capability.

The bonded quantum-photonic platform of (Salamon et al., 6 Aug 2025) differs in that the transferred film is not a passive optical layer but an electrically functional membrane containing epitaxial quantum dots and a p-i-n junction. The bonding target can be a plain silica-on-silicon chip or a foundry-fabricated SiN PIC with buried low-loss waveguides; after polishing, the remaining oxide above SiN is about 200–220 nm, so together with the $(\mathrm{ZnSe})_N/(\mathrm{GaAs})_M$0 nm adhesive the GaAs membrane sits approximately 480 nm above the buried SiN guide (Salamon et al., 6 Aug 2025). This is a distinctly heterogeneous GaAs-on-insulator platform in which GaAs provides the active quantum-emitter layer and SiN provides low-loss passive transport.

By contrast, the suspended passive GaAs platform of (Jiang et al., 2019) begins from a 220 nm GaAs film above a 1.5 $(\mathrm{ZnSe})_N/(\mathrm{GaAs})_M$1m AlGaAs sacrificial layer, then uses a three-etch silicon-photonics-like process and HF release to create a membrane with air below and oxide above. It is not a bonded GOI stack, yet it validates the transferability of grating couplers, rib waveguides, MMIs, and microrings to high-index-contrast GaAs (Jiang et al., 2019). This suggests that many passive PDK-like elements do not depend on the literal presence of buried oxide so long as the undercladding index is sufficiently low.

4. Device classes and representative performance

The GaAs-on-insulator literature is notable for spanning nonlinear optics, passive photonics, and quantum photonics rather than converging on a single device type.

Nonlinear waveguides and second-harmonic generation

A fully oxide-clad GaAsOI waveguide of width $(\mathrm{ZnSe})_N/(\mathrm{GaAs})_M$2, thickness 150 nm, and length 1.4 mm was designed to phase match the fundamental TE mode at $(\mathrm{ZnSe})_N/(\mathrm{GaAs})_M$3 to the fundamental TM mode at $(\mathrm{ZnSe})_N/(\mathrm{GaAs})_M$4 and achieved a maximum single-pass conversion efficiency of $(\mathrm{ZnSe})_N/(\mathrm{GaAs})_M$5, corresponding to a normalized efficiency of $(\mathrm{ZnSe})_N/(\mathrm{GaAs})_M$6 (Chang et al., 2018). Pump-wavelength loss near $(\mathrm{ZnSe})_N/(\mathrm{GaAs})_M$7 was approximately $(\mathrm{ZnSe})_N/(\mathrm{GaAs})_M$8–$(\mathrm{ZnSe})_N/(\mathrm{GaAs})_M$9 (Chang et al., 2018). The measured SHG bandwidth was 0.93 nm, close to the 0.90 nm theoretical prediction, and the efficiency spectrum followed the expected sinc behavior associated with phase-matched propagation (Chang et al., 2018).

A related wafer-scale GaAsOI platform achieved a peak measured single-pass SHG efficiency of $\chi^{(2)}$0, reported as 40 $\chi^{(2)}$1, in a 2.9 mm partially etched GaAs waveguide converting a $\chi^{(2)}$2 TE pump to a $\chi^{(2)}$3 TM second harmonic (Stanton et al., 2019). The nonlinear efficiency was defined as

$\chi^{(2)}$4

with $\chi^{(2)}$5 incorporating phase mismatch and loss mismatch, and $\chi^{(2)}$6 capturing the mode-overlap integral (Stanton et al., 2019). This device had pump propagation loss $\chi^{(2)}$7, SH propagation loss $\chi^{(2)}$8, a 148 GHz bandwidth, and thermal tuning over 45 $\chi^{(2)}$9C with slope $\chi^{(3)}$0 (Stanton et al., 2019).

These two SHG results are not numerically interchangeable: one is reported in the conventional single-pass GaAsOI normalization of $\chi^{(3)}$1 with explicit area-length normalization and the other emphasizes record normalized efficiency in $\chi^{(3)}$2 (Chang et al., 2018, Stanton et al., 2019). Taken together, however, they show that GaAsOI can support both high absolute efficiency and broad-enough bandwidth for system applications such as integrated $\chi^{(3)}$3-to-$\chi^{(3)}$4 self-referencing (Stanton et al., 2019).

Kerr nonlinear photonics and microcombs

AlGaAsOI extends the GaAs-family-on-insulator concept to telecom-band Kerr devices. In a $\chi^{(3)}$5 waveguide, $\chi^{(3)}$6 and propagation loss of approximately $\chi^{(3)}$7 were extracted from continuous-wave four-wave mixing (Pu et al., 2015). Bus-coupled microresonators with loaded $\chi^{(3)}$8 around $\chi^{(3)}$9 to $\mathrm{SiO_2}$00 produced optical parametric oscillation with a measured threshold of $\mathrm{SiO_2}$01, described as a record low threshold power of 3 mW, and an 810-$\mathrm{SiO_2}$02m resonator generated a 98 GHz comb spanning about 350 nm under 72 mW coupled pump (Pu et al., 2015).

A later ultrahigh-$\mathrm{SiO_2}$03 AlGaAsOI platform pushed the propagation loss to $\mathrm{SiO_2}$04 and the intrinsic microring quality factor to $\mathrm{SiO_2}$05, with finesse as high as $\mathrm{SiO_2}$06 (Xie et al., 2020). In anomalous-dispersion rings of width $\mathrm{SiO_2}$07, comb generation thresholds were $\mathrm{SiO_2}$08 for a 1 THz FSR ring and $\mathrm{SiO_2}$09 for a 90 GHz FSR ring; at $\mathrm{SiO_2}$10 and $\mathrm{SiO_2}$11, those resonators produced comb spans of 150 nm and 200 nm, respectively (Xie et al., 2020). This strongly suggests that once sidewall and surface scattering are sufficiently suppressed, GaAs-family-on-insulator platforms can reach the low-loss regime previously associated mainly with SiN while retaining far larger material nonlinearity.

Passive photonic integrated circuits

Suspended GaAs, used as a practical surrogate for GOI high-contrast photonics, has demonstrated a complete passive device set: grating couplers, waveguides, 2×2 MMIs, and ring resonators (Jiang et al., 2019). The device layer is 220 nm GaAs above an initially $\mathrm{SiO_2}$12 AlGaAs sacrificial layer. Reported metrics include grating-coupler period $\mathrm{SiO_2}$13 nm, duty cycle $\mathrm{SiO_2}$14, etch depth $\mathrm{SiO_2}$15 nm, a measured fiber-to-fiber peak transmission of about $\mathrm{SiO_2}$16 dB near 1550 nm corresponding to roughly 4 dB per coupler, ring radius 25 $\mathrm{SiO_2}$17m, bus-ring gap 325 nm, optical $\mathrm{SiO_2}$18, and propagation loss estimated from the ring as $\mathrm{SiO_2}$19, characterized as an upper bound because the ring is likely overcoupled (Jiang et al., 2019).

Although these devices are suspended rather than bonded, the design lesson is explicit: once the undercladding index is made sufficiently low, many silicon-photonics-like process flows and component geometries can be ported to GaAs with only limited redesign (Jiang et al., 2019). A plausible implication is that true GOI could remove membrane-specific complications such as release holes, tethers, and sagging while retaining similar passive topologies.

Quantum photonics and coherent single-photon generation

The bonded GaAs quantum-emitter platform of (Salamon et al., 6 Aug 2025) demonstrates that electrically controlled self-assembled quantum dots survive transfer to an insulating host. The GaAs membrane remains single-mode and the simulated $\mathrm{SiO_2}$20-factor decreases only from $\mathrm{SiO_2}$21 in suspended GaAs waveguides to $\mathrm{SiO_2}$22 with underlying silica (Salamon et al., 6 Aug 2025). The embedded p-i-n diode shows low leakage $\mathrm{SiO_2}$23 from 1 to 1.5 V at 1.6 K, corresponding to power dissipation $\mathrm{SiO_2}$24, and enables Stark tuning with $\mathrm{SiO_2}$25 (Salamon et al., 6 Aug 2025).

Optically, a representative resonance-fluorescence transition at 930.6 nm shows

$\mathrm{SiO_2}$26

with linewidths across the sample in the $\mathrm{SiO_2}$27–$\mathrm{SiO_2}$28 range (Salamon et al., 6 Aug 2025). The time-resolved fluorescence decay rate is $\mathrm{SiO_2}$29, implying a natural linewidth $\mathrm{SiO_2}$30, so the measured line is broadened by less than a factor of two over the Fourier limit (Salamon et al., 6 Aug 2025). Under pulsed resonant driving, the single-photon purity is $\mathrm{SiO_2}$31 (Salamon et al., 6 Aug 2025). The same platform supports vertical coupling into buried SiN via $\mathrm{SiO_2}$32-long GaAs tapers, with measured two-taper insertion loss $\mathrm{SiO_2}$33, corresponding to $\mathrm{SiO_2}$34 efficiency per taper (Salamon et al., 6 Aug 2025).

This result is significant because it shifts GaAsOI from purely passive or nonlinear optics into electrically controlled heterogeneous quantum photonics. It also clarifies that GOI can host active epitaxial semiconductor stacks rather than merely passive transferred slabs.

5. Fabrication routes and engineering constraints

The dominant fabrication paradigm is heterogeneous bonding followed by donor-substrate removal. The nonlinear GaAsOI platform of (Chang et al., 2018) uses plasma-activated bonding of a III–V die onto thermal oxide, followed by annealing at $\mathrm{SiO_2}$35 for 12 hours under pressure, mechanical polishing of the original GaAs substrate to $\mathrm{SiO_2}$36, selective wet etching, and buffered HF removal of the AlGaAs layer (Chang et al., 2018). The wafer-scale GaAsOI SHG platform of (Stanton et al., 2019) uses direct bonding between native-oxide GaAs and $\mathrm{SiO_2}$37, with room-temperature bond initiation, a $\mathrm{SiO_2}$38 anneal for 1 hour, and a measured bond energy of $\mathrm{SiO_2}$39 (Stanton et al., 2019). AlGaAsOI fabrication employs either BCB-assisted bonding (Pu et al., 2015) or direct heterogeneous wafer bonding with $\mathrm{SiO_2}$40 interlayers, surface activation, and outgassing trenches in thermal oxide (Xie et al., 2020).

Several recurring process constraints emerge.

First, surface and interface chemistry matter strongly. In GaAsOI SHG, preserving the native oxide on the bonding surface and using a post-fabrication HCl clean before regrowth of a cleaner native oxide reduce As–As bond-related absorption near the SH wavelength (Stanton et al., 2019). In AlGaAsOI microresonators, replacing InGaP etch stops with $\mathrm{SiO_2}$41, smoothing resist by thermal reflow, and using MBE material instead of MOCVD significantly reduce sidewall and surface roughness (Xie et al., 2020).

Second, high contrast makes roughness unusually costly. In AlGaAsOI, sidewall roughness estimated from SEM line-edge analysis decreased from $\mathrm{SiO_2}$42 to $\mathrm{SiO_2}$43 after photoresist reflow, and this alone improved $\mathrm{SiO_2}$44 from $\mathrm{SiO_2}$45 to $\mathrm{SiO_2}$46 in a 30 $\mathrm{SiO_2}$47 radius ring (Xie et al., 2020). In the optimized MBE platform, top and bottom surface roughness fell to 0.15 nm and 0.17 nm, enabling $\mathrm{SiO_2}$48 (Xie et al., 2020).

Third, GaAs-based thin films can be highly sensitive to dimensional nonuniformity. In the GaAsOI SHG platform, thickness variation over a device can reach $\mathrm{SiO_2}$49, while the relevant thickness tolerance is only $\mathrm{SiO_2}$50, causing longer waveguides to underperform despite the nominal $\mathrm{SiO_2}$51 scaling of SHG (Stanton et al., 2019). In the earlier bonded GaAsOI platform, a 1 nm thickness change shifts the phase-matching wavelength by 11 nm, and a 10 nm width change shifts phase matching by 1 nm (Chang et al., 2018). This suggests that thin-film GaAsOI, especially for phase-matched $\mathrm{SiO_2}$52 devices, is often limited by nanometer and sub-nanometer uniformity rather than by the underlying nonlinear coefficients.

Fourth, electrical integration on transferred GaAs requires adapted packaging. In (Salamon et al., 6 Aug 2025), bond pads are routed onto the silica surface rather than placed directly on the GaAs membrane because the adhesive bond is not strong enough for direct ball-bonding onto the membrane without risking delamination. This indicates that GOI electrical design must co-optimize contact routing, membrane adhesion, and optical layout.

A different but related materials route appears in thin-buffer GaAs-on-Si donor structures intended to improve III–V films before any potential transfer. One paper shows that high-quality GaAs can be grown on on-axis Si(001) using a 5 nm AlSb nucleation layer and a likely 1045 nm GaSb buffer, followed by 200 nm low-temperature GaAs and 300 nm high-temperature GaAs, with total thickness 1.55 $\mathrm{SiO_2}$53m (Cheng et al., 2023). The purpose there is direct epitaxy on Si rather than GOI, but the broader lesson is that engineered interfacial misfit-dislocation arrays can localize strain relief in relatively thin stacks (Cheng et al., 2023). A plausible implication is that such donor-wafer strategies could feed later transfer-to-insulator workflows, though the paper itself does not perform bonding.

6. Opportunities, limitations, and contested interpretations

The literature agrees on the platform’s functional promise but is careful about its present boundaries. Several misconceptions are explicitly corrected by the underlying papers.

A first misconception is to equate every GaAs-based high-contrast structure with conventional GOI. The suspended passive platform with air below and oxide above is mechanically and thermally distinct from a permanently oxide-bonded device layer (Jiang et al., 2019). The [111]-grown GaAs/Ge/GaAs topological-insulator quantum well is not a classic GaAs-on-insulator wafer at all; it is a GaAs-compatible insulating heterostructure whose “insulating” behavior comes from quantum-well confinement and bulk bandgap rather than a buried oxide (Zhang et al., 2013). Conversely, AlGaAsOI papers are directly relevant to GaAsOI in platform logic even though the core is an alloy rather than pure GaAs (Pu et al., 2015, Xie et al., 2020, Placke et al., 2020).

A second misconception is that GaAs-family-on-insulator is already a mature foundry platform. Several papers explicitly stop short of that claim. The suspended passive work argues scalability mainly at the level of design methodology and process modularity, not industrial manufacturability (Jiang et al., 2019). The bonded quantum-emitter platform demonstrates coherence preservation and hybrid coupling but not optimized brightness, yield, or large-scale packaging (Salamon et al., 6 Aug 2025). Even the wafer-scale GaAsOI SHG result, despite 76 mm transferred-film uniformity, remains constrained by coupling loss and second-harmonic propagation loss (Stanton et al., 2019).

A third misconception is that oxide support automatically solves all platform problems. In fact, new bottlenecks emerge. For GaAsOI SHG, surface-state absorption from As–As bonds creates a strong absorption feature centered near 950 nm and contributes to the measured SH loss of $\mathrm{SiO_2}$54 (Stanton et al., 2019). In AlGaAsOI, high confinement magnifies sensitivity to sidewall roughness, and residual scattering remains the dominant loss mechanism even after top and bottom surfaces become nearly atomically smooth (Xie et al., 2020). In quantum photonics, the bonded membrane preserves narrow linewidths, but blinking at the $\mathrm{SiO_2}$55 level remains in supplementary data, indicating that environmental charge noise is reduced, not eliminated (Salamon et al., 6 Aug 2025).

There is also a materials-design tradeoff specific to GaAs-family alloys. In AlGaAsOI telecom Kerr devices, the aluminum fraction must be chosen so that operation lies below half the bandgap and two-photon absorption is strongly reduced. A comparison between 15% and 17% Al shows that $\mathrm{SiO_2}$56 with bandgap 1.61 eV did not sufficiently suppress TPA and showed no OPO, whereas $\mathrm{SiO_2}$57 with bandgap 1.64 eV reduced TPA to a negligible level and enabled OPO (Pu et al., 2015). Pure GaAs offers higher refractive index and strong $\mathrm{SiO_2}$58, but its smaller bandgap restricts the practical SHG window to wavelengths above roughly 900 nm and makes surface absorption near 1 $\mathrm{SiO_2}$59m a recurring issue (Stanton et al., 2019).

From a system perspective, the strongest present opportunities lie in domains where GaAs brings functionalities absent or weaker in Si and SiN. These include coherent epitaxial single-photon sources with electrical tuning (Salamon et al., 6 Aug 2025), strong $\mathrm{SiO_2}$60 frequency conversion without periodic poling (Chang et al., 2018, Stanton et al., 2019), combined $\mathrm{SiO_2}$61 and $\mathrm{SiO_2}$62 nonlinear photonics in the same material family (Pu et al., 2015, Xie et al., 2020), and prospective electro-optic superlattice-on-insulator modulators with predicted $\mathrm{SiO_2}$63 up to $\mathrm{SiO_2}$64 and $\mathrm{SiO_2}$65 down to $\mathrm{SiO_2}$66 in $\mathrm{SiO_2}$67 structures (Leonardis et al., 2022). Because the latter is theoretical and based on non-relaxed heterointerfaces and approximate ionic contributions, it should be read as a design proposal rather than a demonstrated GOI technology (Leonardis et al., 2022).

The overall trajectory is therefore coherent but not uniform. GaAs-on-insulator is best understood not as a single standardized wafer format, but as a family of high-index-contrast III–V thin-film platforms whose principal engineering objective is to retain GaAs-family active, electro-optic, and nonlinear advantages while obtaining the confinement, compactness, and heterogeneous-integration benefits traditionally associated with oxide photonics. The strongest established results are low-loss AlGaAsOI resonators with $\mathrm{SiO_2}$68 and comb thresholds down to $\mathrm{SiO_2}$69 (Xie et al., 2020), GaAsOI SHG with 40 $\mathrm{SiO_2}$70 single-pass efficiency on a 76 mm wafer (Stanton et al., 2019), record normalized SHG efficiency of $\mathrm{SiO_2}$71 in a 1.4 mm GaAsOI waveguide (Chang et al., 2018), and coherent electrically tunable quantum-dot emission from a bonded GaAs membrane with linewidths below $\mathrm{SiO_2}$72 (Salamon et al., 6 Aug 2025). These results collectively support the view that GaAs-on-insulator has moved beyond a purely conceptual extension of silicon photonics, but has not yet converged into a single mature manufacturing platform.