On-chip stimulated Brillouin scattering via surface acoustic waves

Surface acoustic waves (SAWs)—mechanical waves that travel along the surface of a material—are harnessed in many signal processing and filtering applications for their high fidelity and compact footprint,^1,2 as well as a multitude of sensing applications—mechanical, structural, chemical, and biological—due to their sensitivity to the topography and surrounding of the surface.^3–5 Typical approaches to generating and detecting SAWs utilize piezoelectric materials via interdigital transducers (IDTs).^1,2 Exciting and detecting SAWs all-optically would enable new opportunities to achieve coherent coupling for quantum applications,^6–8 more dynamic and tunable signal processing without bandwidth and frequency limitations set by the IDTs, and highly sensitive optical readout techniques for sensing applications. Furthermore, it has the potential to extend SAW technologies to non-piezoelectric materials, allowing for integration with photonic circuits and providing new functionalities to photonic chips.

Stimulated Brillouin scattering (SBS) is a nonlinear optical effect that couples acoustic and optical waves.⁹ Predicted more than 100 years ago,¹⁰ SBS has found many applications in fiber optical systems,¹¹ and more recently, it was successfully demonstrated in chip-scale waveguides.¹² Inducing SBS on-chip is challenging due to the short interaction length and the requirement of mode overlap between the optical and acoustic modes. The need for simultaneous guidance of the optical and acoustic modes limits the Brillouin gain in most complementary metal–oxide semiconductor (CMOS) materials such as SiN^13,14 or requires under-etching in the case of silicon-on-insulator waveguides.^15,16 Soft chalcogenide glasses, on the contrary, provide simultaneous optical and acoustic guidance,^17,18 resulting in demonstrations of as high as 50 dB on-chip Brillouin gain.¹⁹ 

The large SBS gain enabled by the unique properties of chalcogenide together with the narrow linewidth of the SBS response makes it ideal for signal processing applications in a small footprint, such as compact narrowband filters.²⁰ SBS is furthermore an exquisite sensor, as the SBS response changes with waveguide geometry, temperature, and/or strain.^21–25 Here, the high SBS gain in chalcogenide waveguides enabled record spatial resolution distributed Brillouin sensing of below one mm;²⁶ however, only geometrical variations to the waveguide core could be sensed,²⁷ as the interacting modes were confined to the waveguide core and, therefore, isolated from the surrounding environment.

Inducing acoustic waves that are traveling at the surface via Brillouin interactions—referred to as surface acoustic wave stimulated Brillouin scattering (SAW-SBS)—opens new opportunities for high-resolution sensing applications via coherent optical coupling between SAWs and optical modes in compact photonic integrated circuits that are sensitive to the surface topography and the surrounding of the waveguide.

Furthermore, guiding the acoustic wave at the surface of the waveguide rather than in the waveguide core can provide a pathway to induce narrowband SBS responses in photonic circuits that otherwise lack acoustic confinement, opening a path toward novel tunable, high-resolution signal processing functionalities. In addition, the lower frequency of SAWs relative to their bulk counterparts might lead to even higher frequency resolution.

Recent theoretical studies have shown that, for example, lithium niobate waveguides—attractive for their low loss and large electro-optic coefficient—do not show strong SBS gain from optoacoustic interactions in the waveguide core but could support strong SBS from SAWs.²⁸ Indeed, SAWs excited via IDTs in silicon-on-insulator (SOI) waveguides experimentally demonstrated ultra-narrow bandwidth filter functions without the requirement of under-etching the silicon waveguide to provide guidance for the acoustic modes.²⁹ Spontaneous Brillouin light backscattering by SAWs in a waveguide without the use of IDTs has been studied numerically and experimentally in subwavelength optical fiber tapers and photonic crystal fibers,^30–32 and forward Brillouin interaction has been observed at the surface of microspheres.³³ However, an experimental demonstration of optically excited and detected coherent SAWs via stimulated Brillouin scattering in a waveguide remains elusive.

In this work, we present the first experimental observation of on-chip excitation of surface acoustic waves via stimulated Brillouin scattering. This novel regime of Brillouin scattering was demonstrated in highly nonlinear Ge_11.5As₂₄Se_64.5 chalcogenide waveguides that were carefully designed and engineered to optimize the overlap with the guided optical modes and the acoustic wave propagating along the surface. We performed systematic numerical modeling of various waveguide geometries to enhance the overlap between optical and surface acoustic waves, which matches our experimental observation. Tailored waveguides were fabricated, and we measured 203 W⁻¹ m⁻¹ of SBS gain from surface acoustic waves in Ge_11.5As₂₄Se_64.5 waveguides with a cross section of 2600 nm by 116 nm and an on-chip circuit length of 8.5 cm. In addition, we studied the linewidth of the surface acoustic wave resonance and measured linewidth narrowing from 34 to 20 MHz. Our first demonstration of surface acoustic wave stimulated Brillouin scattering marks a new regime for SBS and opens the door for a plethora of on-chip SAW-based applications.

Exciting their shear transverse counterparts, or more specifically coupling to surface acoustic waves [Fig. 1(a), right waveguide], involves a smaller Ω_B [Fig. 1(b)], as the velocity of surface waves is lower than the velocity of longitudinal acoustic waves, typically between 0.87 and 0.95 times the shear velocity of the material.³⁴ The lower velocity translates to a lower Ω_B of the scattered Stokes light [Eq. (1)], illustrated in Fig. 1(b) by a smaller frequency difference between the optical pump and the Stokes wave.

In the context of on-chip SBS, waveguides with geometries providing large overlap between the optical and the longitudinal acoustic field are used to enhance the Brillouin gain (Ref. 18). The longitudinal acoustic waves can be localized in the core of the waveguide when the acoustic velocity in the waveguide core is smaller than in the surrounding material [Fig. 1(c), left waveguide]. On the contrary, the displacement field of surface acoustic waves, as the name suggests, is localized at the surface of the waveguide, with its polarization dominantly in the transverse direction [Fig. 1(c), right waveguide]. Achieving significant Brillouin gain, SAW scattering requires optimization of the waveguide geometry to improve the overlap between the optical and the surface acoustic fields. Highly nonlinear materials, such as chalcogenide glasses, have a large intrinsic SBS gain coefficient¹⁷ and, hence, are a promising candidate for demonstrating SAW-SBS on a chip.

Multiple chalcogenide compositions have been explored for on-chip SBS waveguides, such as As₂S₃, which offers low propagation loss and large Brillouin gain Refs. 17 and 19, and more recently GeSbS³⁵ and AsSe.³⁶ To push the acoustic wave to the surface while simultaneously achieving good overlap with the optical modes, new waveguide designs, materials, and fabrication methods are required. We identified an alternative chalcogenide glass composition, Ge_11.5As₂₄Se_64.5, that has the potential to support on-chip SAW-SBS; Ge_11.5As₂₄Se_64.5 has a relative high refractive index of 2.634 at infrared wavelengths, similar elastic properties as As₂S₃, and does not require any protective layer on top of the waveguide that inhibits or dampens the SBS process at the surface.³⁷ 

We numerically model the Brillouin gain of Ge_11.5As₂₄Se_64.5 waveguides with different geometries to find the optimum dimensions and waveguide structures that can support SAW-SBS. The numerical modeling of the Brillouin response is carried out in the open-source Numerical Brillouin Analysis Tool (NumBAT), a software tool developed for modeling of Brillouin interactions in waveguides of arbitrary geometries³⁸ based on Finite Element Method (FEM) mode solvers. As the first step, we use NumBAT to determine the solutions for the optical and elastic modes of a given waveguide structure and geometry. Next, we evaluate the overlap integrals of the (specified) optical and elastic modal fields, which describe the interaction between the fields, to provide a spectrum of the SBS gain response of the waveguide.

We simulate a rectangular waveguide structure comprising a 2600 nm wide Ge_11.5As₂₄Se_64.5 core of different thicknesses on top of a SiO₂ substrate and, importantly, no over-cladding [Fig. 2(a)]. All the relevant material parameters, refractive index (n = 2.634), density (ρ = 4495 kg/m³), and stiffness tensor components (c₁₁ = 23.837 GPa, c₁₂ = 9.736 GPa, and c₄₄ = 7.050 GPa), have been taken from Ref. 37. For thicker waveguide structures [Fig. 2(b) left], the fundamental optical mode is well confined in the core, whereas for thin structures [Fig. 2(b) right], the optical mode has a strong evanescent field component. This potentially leads to a large overlap with surface acoustic modes that are localized at the surface of the waveguide. Hence, a sweep of the waveguide core thickness is performed to find a waveguide structure that has a large mode overlap between the optical mode and any surface acoustic modes. The modeling provides the dimensions for the fabrication of the waveguides that support SAW-SBS. A scanning electron microscope (SEM) image of a typical thin Ge_11.5As₂₄Se_64.5 waveguide is shown in Fig. 2(c). We note that the SEM image shows some sidewall roughness, which might be caused by the GeAsSe being slightly dissolved in the alkaline developer solution during fabrication (see Sec. V). A wider waveguide design was chosen to reduce the overlap of the optical mode with the sidewall and minimize the impact of sidewall roughness-induced scattering loss.

To experimentally characterize and measure the SBS response of different waveguide structures, we use a pump–probe setup [Fig. 2(d)]. The pump–probe method, first developed for characterizing SBS in optical fiber,³⁹ is now the accepted standard for characterizing the Brillouin gain in integrated circuits.⁴⁰ In essence, it is a measurement of the beating between an optical carrier and a sweeping probe signal using a fast photodetector and an electrical spectrum or network analyzer. A schematic of a typical pump–probe setup and the principle of the measurement technique in the frequency domain are shown in Figs. 2(d) and 2(e), respectively. Laser light (1550 nm) that serves as a carrier is split into two arms—a pump and a probe. The light in both arms is modulated using electro-optic modulators (EOMs), filtered, and amplified by an erbium-doped fiber amplifier (EDFA). On one side of the photonic chip, the strong optical pump wave is coupled to the waveguide via lensed tip fibers in order to generate the Brillouin response of the material. From the opposite side of the chip, a probe signal is coupled into the chip. A vector network analyzer (VNA) is used to sweep the frequency of the probe sideband via an electro-optic modulator relative to the pump frequency, such that the difference between the pump and the probe is 1–10 GHz. Using electrical spectrum/network analyzers allows for high-resolution measurement of the Brillouin spectrum. After being transmitted through the chip, the probe passes through a circulator, and the beating between the probe carrier and the sweeping probe sideband is measured via a photodetector (PD). An optical filter is used before the PD to remove any reflections from the optical pump wave from the chip. Changes in the amplitude of the probe signal due to Brillouin amplification are then measured by the VNA, providing a high-resolution mapping of the Brillouin spectrum of the waveguide from the optical to the electrical domain. We note that the pump–probe technique measures the on–off Brillouin gain in dB, which will be shown in all following measurement results.

Numerical modeling of the stimulated Brillouin gain as a function of frequency for different Ge_11.5As₂₄Se_64.5 waveguide core thicknesses (50–900 nm, fixed, 2600 nm width) provides a map of the different resonances supported by the different waveguide geometries [Fig. 3(a)]. The resonance map results show multiple Brillouin resonances appearing and disappearing as the waveguide thickness changes and the appearance of distinct lower frequency modes for thinner waveguide dimensions. We select a waveguide thickness of 116 nm (the actual thickness of the fabricated waveguide) that supports two lower frequency modes with strong calculated Brillouin gain, and the calculated Brillouin spectrum is presented in Fig. 3(b). We note that the Brillouin gain in Fig. 3(b) is normalized as only very few studies of Ge_11.5As₂₄Se_64.5 for Brillouin interactions have been reported at present,⁴¹ and the required components of the photo-elastic tensor and acoustic loss tensor were not available in the literature. We would like to emphasize that the photo-elastic and acoustic loss tensor components solely determine the magnitude of the Brillouin gain and linewidth of the Brillouin response and not the nature of the modes or frequency (for further details regarding simulation parameters, see Sec. V). The modeled spectrum [Fig. 3(b)] shows two strong resonances at 3.98 and 4.89 GHz and a weak resonance at 7.41 GHz.

Analyzing the characteristics reveals that the lowest frequency mode at 3.98 GHz has all the characteristics of a SAW. The acoustic mode profile is localized at the surface of the structure; the polarization is dominantly in the transverse fraction f_T with a small longitudinal polarization fraction f_L component (f_T = 0.919, f_L = 0.081), and the acoustic velocity v_p is about 0.9 times the shear velocity of the material (v_p = 1125 m/s). All three characteristics verify that the resonance is due to scattering from surface waves. On the contrary, the acoustic mode profile of the second peak (4.89 GHz) has a reduced transverse polarization fraction f_T = 0.143 and, accordingly, an increased longitudinal polarization fraction f_L = 0.857, suggesting that this resonance involves a hybrid of shear transverse and longitudinal acoustic waves. Brillouin scattering mediated by these types of resonances has been referred to as hybrid acoustic wave (HAW) Brillouin scattering and has previously been demonstrated in fiber tapers³⁰ and As₂S₃ waveguides.⁴² The small peak at 7.41 GHz can be assigned to a mainly longitudinal acoustic mode, typically observed in chalcogenide waveguides with larger cross section.¹⁷ 

Based on the results from our numerical modeling, we fabricated a photonic chip comprising 116 nm thick Ge_11.5As₂₄Se_64.5 waveguides. A thin 116 nm Ge_11.5As₂₄Se_64.5 layer has been deposited on top of an SiO₂ substrate and etched to form a 2600 nm wide waveguide (for more details on the fabrication of the samples, see Sec. V). Cut-back measurements for different waveguide lengths of the fabricated waveguides showed a 4.24 dB coupling loss per facet and an estimated propagation loss of around 1.11 dB/cm.

High-resolution pump–probe measurements of the SBS spectrum for an 11.7-cm-long circuit with a coupled pump and probe power of 150 and 11 mW, respectively, show two strong resonances around 3.81 and 4.64 GHz and a weak signal at 7.04 GHz [Fig. 3(c)]. Comparing the experimental findings with our modeling results, we find that our model agrees well with the measured gain spectrum. The consistent offset of about 0.2 GHz between the experimentally measured resonance frequencies and the numerically modeled spectrum likely finds its origin in the uncertainty of the composition of the GeAsSe glass and the actual dimensions of the fabricated waveguide. However, the overall good agreement between the numerical modeling and the experimental measurements allows us to draw conclusions about the nature of the measured resonances. Based on our modeling, where the resonance at 3.98 GHz is due to SAW-SBS, we identify the lower frequency resonance at 3.81 GHz as being caused by the same mechanism. This marks the first experimental observation of on-chip excitation of surface acoustic waves by stimulated Brillouin scattering. In addition, the same reasoning leads to the conclusion that the resonance at 4.64 GHz originates from Brillouin scattering from hybrid shear longitudinal acoustic waves.

We further investigate the characteristics of the observed SAW-SBS (4). In particular, the value of the Brillouin gain coefficient and the linewidth of the SAW resonance. To that end, we measure the Brillouin spectrum for different pump powers using the pump–probe setup [Fig. 2(d)]. The measured high-resolution Brillouin spectra [Fig. 4(a)] of an 8.5-cm-long waveguide circuit show how the amplitude of the Brillouin-SAW resonance at 3.81 GHz increases with increasing coupled pump power.

The zoomed-in high-resolution spectrum measurements reveal a second feature: a shoulder on the higher frequency side of the resonance. Zooming in on the modeled resonance map reveals that for 2600 nm wide and 116 nm thick waveguides, a second surface acoustic wave appears close to the fundamental SAW [Fig. 4(b)]. The second SAW can be identified as a higher-order SAW, with the two gain peaks overlapping in the spectrum presented in Fig. 4(a).

To extract the linewidth and gain of the two overlapping SAW resonances, we fit two Lorentzian curves, which both add up to a curve matching the measured data [Fig. 4(c)]. From the Lorentzian fits, we obtain the peak gain values of the two resonances and plot them as a function of coupled pump power [Fig. 4(d)]. We find that both resonances grow exponentially with pump power, verifying the nonlinear behavior of the SBS interaction. We extract the slope of the measured increase in peak gain via a linear fit function. From the slope, we extract a Brillouin gain coefficient of G_SBS = 203 W⁻¹ m⁻¹ for the fundamental SAW mode and a gain coefficient of G_SBS = 166 W⁻¹ m⁻¹ for the higher-order SAW mode. We note that the measured gain coefficient for SAW-SBS is about 25 times higher than the recorded value of Brillouin light scattering via SAWs in silica fiber tapers.³⁰ In addition, the value of the gain coefficient is in the same order of magnitude as with SBS from LAWs in chalcogenide waveguides, which ranges from 300 to 500 W⁻¹ m⁻¹.¹² From the Lorentzian fits to the gain spectra, we extract an estimated linewidth of the fundamental and higher-order Brillouin SAW resonances as a function of coupled pump power [Fig. 4(e)]. The linewidth of the fundamental SAW narrows from around 34–20 MHz, which is a typical characteristic of stimulated Brillouin scattering.^43,44 The higher-order SAW shows no narrowing in the measured power range and warrants further investigation.

To summarize, in this work, we have demonstrated a first observation of on-chip stimulated Brillouin scattering mediated by surface acoustic waves. Thin waveguides made of nonlinear Ge_11.5As₂₄Se_64.5 chalcogenide glass enabled strong overlap between optical and acoustic waves propagating at the waveguide surface. We used finite element method mode solvers to comprehensively model the Brillouin response of said waveguides and numerically find Brillouin SAW and HAW resonances. Experimental measurements of tailor-made Ge_11.5As₂₄Se_64.5 agree well with the theoretical predictions. We observed surface acoustic wave stimulated Brillouin scattering with a resonance frequency of 3.81 GHz in 116 nm thick chalcogenide waveguides. The SBS gain was measured for a range of pump powers, and a gain coefficient of G_SBS = 203 W⁻¹ m⁻¹ and a linewidth of 20 MHz for SAW-SBS were extracted.

This proof-of-principle demonstration of SAW-SBS in integrated circuits opens the door for novel on-chip sensing applications. As the acoustic wave propagates along the surface, it could enable sensing of the waveguide surrounding and topography, as well as opportunities to further functionalize the surface. In addition, being able to excite multiple acoustic modes in a single waveguide with different ratios of longitudinal and transverse components might open the door to distinguishing effects from different strains and temperatures on the Brillouin response.

Accessing acoustic modes with a lower frequency shift compared to the frequency shift from Brillouin scattering via longitudinal acoustic modes guided in the waveguide core might provide a pathway to high-resolution signal processing as the lifetime of acoustic waves typically increases as the frequency reduces. The longer lifetime could enable longer delays in light storage schemes⁴⁵ but also translate into a narrower linewidth, relevant for signal processing.¹² 

Our demonstration shows a pathway toward optically driven SAW applications in non-piezoelectric materials, as opposed to their electronic counterparts excited via IDTs, which opens the door toward further photonic integration. We demonstrated previously that chalcogenide waveguides can be seamlessly integrated onto versatile CMOS-compatible silicon platforms and low-loss Ge:SiO₂ platforms using horizontal or vertical taper technology.^36,40,46,47 Heterogeneous integration of SAW-SBS waveguides with active and passive photonic components on a single chip paths the way toward fully integrated, small-footprint optical SAW devices.

The following provides an overview of all the material parameters that are used in the modeling of SBS using NumBAT. Table I provides values of the refractive indices n, densities ρ, and stiffness tensor components (c₁₁, c₁₂, c₄₄). Note that chalcogenide is an amorphous material, which means only c₁₁ and c₁₂ are independent and c₄₄ can be calculated from the former two parameters. All values have been taken from existing material parameters provided by NumBAT and from Ref. 37.

As shown in Table I, the refractive index of the core material is much larger than the refractive index of the silica substrate. The large index contrast allows for the optical modes to be confined to the GeAsSe, preventing the optical modes from leaking into the surroundings. The densities and the stiffness tensor components are used to calculate the acoustic wave velocity in the materials. Consequently, the acoustic wave velocities in the waveguide are much lower (v_GeAsSe = 2300 m/s) than the acoustic wave velocity in the cladding material (v_SiO = 6000 m/s). This large difference allows for guidance of the acoustic modes. The combination of both strong optical and acoustic confinement allows for large optoacoustic overlap, which is essential for large Brillouin gain.

The values of the photoelastic tensor and the acoustic loss tensor used for the modeling are provided in Table II. As the values for GeAsSe are not reported in the literature, we rely on a combination of experimentally measured values (viscosity) and reported values for As₂S₃ (photoelasticity), which has similar acoustic properties. Again, we note that the viscosity components only influence the linewidth and the photoelasticity components only the overall gain, but not the nature and frequency of the modeled acoustic waves. Hence, the modeling shows only the normalized gain for the different modes.

Although As₂S₃ waveguides have shown record high SBS gain (Ref. 19), new waveguide fabrication methods, designs, and materials are required for SAW-SBS. The reason is that the As₂S₃ platform requires a protective polymer layer to prevent chemical attacks from the alkaline developer used in the photolithography process,^48,49 which prevents As₂S₃ waveguides from being fabricated without any over-cladding, which will suppress scattering from SAWs. GeAsSe is a waveguide platform that does not require the protective over-cladding.

The fabrication of the GeAsSe waveguides was done by the Australian National Fabrication Facility (ANFF) OptoFab ACT Node. We used a chip layout that comprises waveguides of four different lengths: 2.3, 8.5, 11.7, and 23.7 cm. Initially, a 116 nm chalcogenide film was deposited on a 5 μm thick thermal oxide substrate at room temperature through the thermal evaporation technique. Subsequently, resist patterning was executed via a UV lithography process, followed by inductively coupled plasma-reactive ion etching (ICP-RIE) with an argon-trifluoromethane gas mixture to etch the exposed Ge_11.5As₂₄Se_64.5. At the given parameters (an RF bias power of 30 W, an induction power of 300 W, a CHF3 flow rate of 20 sccm, a gas pressure of 10 mTorr, and a substrate temperature of 20 °C), the etching process yielded an etch rate of 200 nm/min.

By carrying out transmission measurements of all four different waveguide lengths and plotting the propagation losses as a function of waveguide length, we extract a propagation loss of 1.11 dB/cm and a fiber-to-chip coupling loss of 4.24 dB per facet. Using the propagation loss, we estimate the values of the effective lengths to be 1.74, 3.47, and 3.71 cm for waveguides with physical lengths of 2.3, 8.5, and 11.7 cm, respectively.

We use a pump–probe setup to characterize the Brillouin response of the waveguides and precisely measure the Brillouin frequency shift, linewidth, and gain coefficient. A schematic of the setup is shown in Fig. 5.

A continuous narrow linewidth laser (TeraXion distributed feedback (DFB) laser, 45 kHz linewidth), operated at λ_c = 1550.12 nm, is coupled to the pump and probe arm using a 3 dB-coupler. An electro-optic modulator (Fujitsu FTM7938EZ, denoted by EOM), driven by a signal generator (Keysight E8257D), is used to amplitude modulate the pump signal at a frequency of ω₀ = 15 GHz. Using an optical bandpass filter (Alnair BVF-300CL, denoted by BPF), one sideband is selected. An erbium-doped fiber amplifier (Amonic AEDFA-PS-33-B-FA, denoted by EDFA) is used to amplify the pump signal, followed by a Gaussian tunable filter (Santec OTF-320, denoted by GTF) to filter out the amplified spontaneous emission. After the filter, the amplified pump is sent through a circulator and polarization controller (PC) before it is injected into the photonic chip. The polarization of the pump and probe light is aligned to couple into the TE mode of the thin waveguide, which only supports a single optical mode. The light is coupled into the chip using lensed tip fibers that are precisely aligned using I3005 three-axis positioners from Luminos Photonics. After passing the chip, the light is blocked from further propagation by an optical isolator. The probe signal in the other arm is modulated by a second EOM (Thorlabs LNLVL-IM-Z), which is driven by a vector network analyzer (Agilent PNA N5224A, denoted by VNA) at frequency ω_RF. The probe carrier and sidebands then pass an EDFA, GTF, and PC before being coupled into the chip from the opposite side of the pump. After propagation through the chip, the probe and the back-reflected pump pass through the circulator. A high-speed photodetector (Finisar BPDV2120R, denoted by PD) is used to convert the probe signal to the RF domain so it can be measured by the VNA. A narrow bandpass filter (Alnair BVF-300CL) ensures that only the probe carrier and sidebands are measured by the PD. The beating between the probe carrier and probe sidebands is then measured by the VNA.

We acknowledge the support of the Australian National Fabrication Facility (ANFF) OptoFab ACT Node in carrying out this research. The current release of NumBAT can be found at https://github.com/michaeljsteel/NumBAT. During the preparation of this manuscript, we became aware of the following two arXiv papers on stimulated Brillouin scattering via SAWs in LiNbO₃ waveguides.^50,51 This work has been supported through the European Research Council Consolidator (Grant No. 101043229 TRIFFIC). We also acknowledge the support of the Australian Research Council (Grant Nos. DP220101431 and DP200101893), the Australian Research Council Center of Excellence in Optical Microcombs for Breakthrough Science (Project No. CE230100006), and the Office of Naval Research (Grant Nos. N00014-24-1-2009 and N00014-23-1-2597).

The authors have no conflicts to disclose.

Govert Neijts: Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Software (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Choon Kong Lai: Conceptualization (supporting); Investigation (equal); Methodology (equal); Project administration (equal); Software (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Maren Kramer Riseng: Investigation (equal); Methodology (equal); Software (equal). Duk-Yong Choi: Investigation (equal); Methodology (equal); Resources (equal); Writing – review & editing (equal). Kunlun Yan: Investigation (equal); Methodology (equal). David Marpaung: Funding acquisition (equal); Project administration (equal); Resources (equal); Supervision (equal). Stephen J. Madden: Funding acquisition (equal); Project administration (equal); Resources (equal). Benjamin J. Eggleton: Conceptualization (supporting); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Supervision (equal); Writing – original draft (equal); Writing – review & editing (equal). Moritz Merklein: Conceptualization (lead); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Supervision (equal); Writing – original draft (equal); Writing – review & editing (equal).

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