Advances in nanofabrication and photonic engineering have propelled cavity optomechanical systems to a pivotal role for probing fundamental optical-force-mediated mechanical phenomena and enabling precision control over light-matter interactions [1]. These optomechanical systems have evolved rapidly with improved coupling efficiencies and tunable responses, promising innovations in vast fields including metrology [2–4], quantum information [5–7], and advanced sensing [8,9]. A key figure of merit for these optomechanical devices is the product of the mechanical frequency (fm) and quality factor (Qm), as a large fm·Qm is critical to realizing exceptional measurement sensitivity, enabling quantum-level control, and minimizing thermal decoherence [10]. Despite the significant progress achieved with conventional materials such as silicon (Si) [11–14], silicon dioxide (SiO2) [15–17], and silicon nitride (Si3N4) [18,19], the quest for better performance has driven attention toward other candidates such as aluminum nitride (AlN) [20,21], diamond [22,23], gallium phosphide (GaP) [24,25], and lithium niobate (LN) [26–28] for their broadband optical transparency, strong optical confinement, and excellent mechanical properties that enable high-frequency mechanical modes with minimal dissipation. Among them, silicon carbide (SiC) stands out due to its exceptional thermal stability, mechanical robustness, and compatibility with CMOS processing [29–32]. For example, the fm·Qm product of SiC is predicted to be 6.4×1014Hz in the Akhiezer regime, which is one order of magnitude higher than that of Si (3.9×1013Hz) and diamond (3.7×1013Hz) [33–35]. However, the very characteristics that make SiC attractive—its high mechanical rigidity and chemical inertness—also present fabrication challenges that constrain the realization of high-performance SiC optomechanical resonators. To date, only a few experiments have successfully demonstrated optomechanical resonators in the polytype of 3C-SiC [36]. In 4H-SiC, the reported SiC mechanical resonators are all bulky and only support mechanical modes with relatively low frequencies (<10MHz) [31,32]. In addition, the lack of integrated optical access necessitates additional excitation mechanisms, further complicating device design and scaling potential.

In this work, we report the first demonstration of a high-performance, integrated 4H-SiC optomechanical resonator that simultaneously functions as a high-finesse optical cavity. As illustrated in Fig. 1, our design employs an ultracompact (with radius around 4 μm) microdisk resonator suspended on a low-loss 4H-SiC-on-insulator (4H-SiCOI) platform, which supports whispering-gallery modes at telecom wavelengths with intrinsic optical quality factors up to ∼1×106. The strong optomechanical coupling resulting from highly co-localized optical and mechanical modes enables sensitive readout of the Brownian motion through optical photodetection, achieving a displacement sensitivity of 0.144fm/Hz. With this method, we characterize the fundamental mechanical mode to exhibit a mechanical frequency of 950 MHz and a mechanical quality factor near 19,200. The corresponding fm·Qm product is estimated to be 1.82×1013Hz, which is among the highest reported metrics of optomechanical cavities tested in ambient environment at room temperature. Notably, this value surpasses the threshold of fm·Qm>6.1×1012Hz that is typically required in observation of quantum sideband asymmetry or feedback cooling a mechanical oscillator to its ground state [10], providing a straightforward solution for applications where low dissipation and high coherence are essential without needing dissipation engineering. Finally, we also achieve radiation-pressure-driven regenerative (or self-sustained) optomechanical oscillations and harmonic generation with an on-chip optical power well below 1 mW.

An optomechanical resonator requires strong coupling between the optical and mechanical modes. As illustrated in Fig. 1, the optical whispering-gallery mode supported by the SiC microdisk generates radiation pressure along the radial direction, efficiently driving the fundamental mechanical radial-breathing mode with dominant radial displacement [see Fig. 1(a)]. For such mechanical modes, numerical simulations based on the finite element method (FEM) point to a mechanical frequency above the very-high-frequency (VHF) band (i.e., >300MHz) for SiC microdisks with radii below 13 μm [see Fig. 1(c)]. In addition, the optomechanical coupling coefficient resulting from the moving-boundary effect scales as gom≈ωo/R, where ωo is the angular optical resonance frequency and R is the radius. While a small R benefits a high mechanical frequency and strong optomechanical coupling, the optical quality factor of the SiC microdisk could drop significantly if the radius is close to the optical wavelength (1.55 μm in this experiment). As such, we choose to work with disks with radii near 4 μm to balance the mechanical and optical properties. A detailed simulation shows that a 4-μm-radius SiC disk exhibits a mechanical frequency fm≈1GHz, an effective mass of meff≈71pg, and gom/2π≈44GHz/nm for the fundamental transverse-electric (TE00) mode. These numbers translate to a vacuum optomechanical coupling rate g0=gom/ℏ/(4πmefffm) ≈15kHz (ℏ≡h/2π with h being the Planck’s constant) for the interaction between optical and mechanical modes.

Proceeding to nanofabrication, a 4H-SiCOI chip consisting of a 700-nm-thick 4H-SiC layer on top of 2-μm-thick silicon dioxide (NGK Insulators) is employed for SiC disks with radii varied from 4.3 to 4.7 μm [37]. The device fabrication begins with defining photonic structures using a negative-tone resist (flowable oxide 16) in e-beam lithography (EBL), which is followed by fluorine-based plasma dry etching to target 600 nm removal of SiC. Next, a positive-tone e-beam resist (PMMA) is employed to pattern circular release windows surrounding the microdisk resonators, within which the remaining 100 nm SiC layer is etched away for the subsequent undercutting process [see Fig. 1(b)]. Note that during this step, the top surface of the SiC microdisks is no longer protected by resist, resulting in an approximate 100 nm reduction in the SiC thickness (i.e., the final microdisk thickness is around 600 nm instead of the original 700 nm). After the two-step EBL and dry etching process, the SiC chip is finally dipped in buffered oxide etch solution to isotropically remove the SiO₂ layer underneath the SiC microdisk. The wet etch time is carefully controlled to maximize the undercutting ratio while avoiding structural collapse in microdisk resonators. In practice, we manage to attain an undercut ratio (undercut width normalized by the disk radius) close to 80% for 4.3-μm-radius SiC microdisks. For larger disks with radii including 4.5 and 4.7 μm, the undercut width is similar but the undercut ratio becomes smaller.

The experimental setup for the optomechanical characterization of the SiC chip is shown in Fig. 2, which is carried out at room temperature and in atmosphere. Briefly, we first perform linear transmission scans at low optical powers to identify the resonant modes of SiC microdisks, focusing on the TE00 mode family, which is expected to exhibit the highest optical quality factors in the telecom band. Next, we fix the laser detuning relative to the optical resonance and characterize the mechanical properties and optomechanical interactions by sending the detected signal to an electrical spectrum analyzer (ESA). For these purposes, a narrow-linewidth tunable laser (linewidth <100kHz with a tuning range of 1510–1640 nm) is employed, whose output power is fixed at 5 mW and attenuated externally using an in-loop variable optical attenuator (0–60 dB). In addition, the polarization state of light is manually adjusted with the help of a fiber polarization controller. To achieve efficient coupling between the fiber and the SiC chip, a pair of grating couplers is designed for each microdisk in the 1550 nm band, attaining a total insertion loss of 10–13 dB [37]. After transmission, the collected signal is split into two paths using a 10:90 fiber coupler: the 10% portion is connected to a low-speed (MHz) photodetector (Thorlabs PDB450C) with large electronic gains for the resonance scan, while the 90% portion is sent to a high-speed photodetector (Newport AD-40 with 12 GHz bandwidth) followed by a real-time ESA (Tektronix RSA5106 with 6.2 GHz bandwidth).

Systematic linear optical characterization is performed for SiC microdisks with radii ranging from 4.3 to 4.7 μm. The example provided in Fig. 3 corresponds to a waveguide-coupled 4.5-μm-radius suspended SiC microdisk. As shown by the scanning electron micrograph in Fig. 3(a), the access waveguide in the coupling region is tapered down to 600 nm in width to increase the field overlap with the confined resonant modes. The linear transmission scan displayed in Fig. 3(b) confirms that the TE00 mode family is efficiently excited, which has an estimated free spectral range of 4.2 THz. The zoom-in plots of two TE00 resonances highlighted in Fig. 3(b) reveal mode splitting resulting from the coupling between the clockwise and counterclockwise modes induced by the sidewall roughness [38]. Furthermore, numerical fitting based on a doublet model points to an intrinsic optical quality factor up to 1.2 million. It is worth noting that the intrinsic quality factors of the TE00 mode family exhibit variations among different azimuthal orders and devices, with the majority falling in the range of 0.3–0.8 million. Such variations are consistent with under-coupled microresonators with their loss dominated by the scattering from the roughness introduced to sidewalls in nanofabrication [39].

Next, we proceed to the characterization of the mechanical properties of SiC microdisks. In this regard, the undercut ratio, defined as the undercut width normalized by the radius of the microdisk, is a critical parameter determining the intrinsic mechanical loss. Given that the undercut width is similar among microdisk resonators with different radii (varied from 4.3 to 4.7 μm), the 4.3-μm-radius SiC microdisk is expected to exhibit the highest mechanical Q due to its largest undercut ratio. Figure 4(a) shows the optical resonance measurement for a 4.3-μm-radius SiC microdisk, where the TE00 mode is found to possess an intrinsic optical quality factor of 3.4×105 at 1592 nm, along with a higher-order TE10 mode with an intrinsic quality factor near 1.8×105 at 1610 nm.

The large optical quality of the observed WGMs combined with strong optomechanical coupling enables efficient optical transduction of mechanical motion. To measure the mechanical resonance, we fix the laser frequency to the blue side of the optical resonance while recording the optical transmission using a high-speed photodetector and an electrical spectrum analyzer (ESA). The optical power dropped into the microdisk (Pd) from the coupling waveguide is defined as Pd=Pin(1−To), with Pin and To representing the input optical power in the waveguide and the normalized optical transmission, respectively. The optical dropped power is initially maintained at a low level (≈2.5μW) to minimize dynamic optical back-action, so that the thermomechanical (Brownian) motion of the microdisk dominates in the detected signal. As shown in Fig. 4(b), the fundamental RBM at 950.1 MHz is clearly observed in the ESA spectrum, which agrees with the simulated value of 945.93 MHz in Fig. 1(c) reasonably well. The difference is mainly attributed to the uncertainty in estimating the undercut width and disk radius. Moreover, we can extract the mechanical quality factor (Qm) by fitting the measured RF spectrum to a damped simple harmonic resonator model [40], revealing a Qm of 1.92×104 [see Fig. 4(c)]. Note that this value does not vary with the optical dropped power provided that it is small enough (Pd<10μW), suggesting that the mechanical resonator is mainly limited by damping effects with negligible contribution from the optical back-action. Consequently, the fundamental RBM achieves an fm·Qm product of 1.82×1013Hz, which is on par with the highest values achieved among the reported WGM-type optomechanical microresonators (a more detailed comparison is provided in Table 1) [20,22].Table 1.

Survey of Reported Optomechanical and Electromechanical Microresonators across Different Material Platforms^a

The displacement sensitivity of our SiC microdisk resonator can be estimated by relating the amplitude of the thermomechanical motion to the RF power spectrum. For example, the spectral density of the thermomechanical motion at the resonance frequency is given by [46] Sx,th1/2(ωm)=4kBTQmωm3meff,(1)where kB is the Boltzmann constant, T is the temperature in Kelvin (300 K in our case), ωm=2πfm, and meff is the effective mass (meff≈71pg for 4.3-μm-radius SiC microdisk). Using this formula, the data shown in Fig. 4(b) translates to an exceptionally high transduction responsivity of ∼143nV/fm and displacement sensitivity of 0.144fm/Hz1/2. For convenience, we also plot the spectral density in the displacement domain on the right y-axis of Fig. 4(c) [47].

In addition to the specific example (4.3-μm-radius SiC microdisk) discussed in Fig. 4, an array of SiC microdisks co-fabricated on the same chip is also characterized with the results summarized in Fig. 5. Notably, we observe a consistent degradation of the mechanical quality factor with increased radius. This trend indicates that the mechanical quality factor is likely to be limited by pedestal-induced anchor loss, as the larger-sized microdisks are compromised by reduced undercutting ratios. This observation motivates future FEM simulations of energy dissipation pathways, specifically probing anchor-dependent losses and modal strain redistribution mechanisms amplified by the diminishing pedestal-to-radius ratio.

So far our mechanical characterization has been performed at low enough optical power levels to suppress the optical back-action, a necessary condition for the extraction of mechanical frequencies and quality factors in the linear regime. When operated at high enough optical power levels, however, the strong optomechanical back-action can surpass intrinsic mechanical damping and hence induce self-sustained regenerative optomechanical oscillations. We showcase such dynamic optomechanical interplay using the 4.3-μm-radius SiC microdisk discussed in Fig. 4 as an example. In Fig. 6(a), the measured RF spectrum of the fundamental RBM as a function of the optical powers is plotted. A close-up view of the mechanical spectrum at an approximate dropped power of 32 μW, shown in Fig. 6(b), reveals a significant reduction in the mechanical linewidth from the linear case of ≈50kHz to ≈1.0kHz, achieving an effective mechanical quality factor as high as 9.3×105. In addition, a full range of RF scan displays harmonic generation up to the fifth order for the fundamental RBM [Fig. 6(c)]. In the same spectrum, we also observe a secondary mechanical mode near the frequency of 3.8 GHz. A zoomed-in display of this ultrahigh-frequency mode reveals a mechanical quality factor of 227 as shown in the inset of Fig. 6(c). While FEM simulations predict second- and third-order RBMs at ∼2.6GHz and ∼4.1GHz, respectively, both modes are absent in the measured RF spectrum. Instead, the observed 3.8 GHz resonance is situated between these expected frequencies, suggesting a possible cause due to modal mixing between these high-order modes. Further investigation into the origin of this secondary mechanical mode is ongoing.

Finally, the performance of our optomechanical oscillator can be more effectively quantified by plotting the mechanical energy as a function of the optical dropped power, as shown in Fig. 6(d). Here, the mechanical energy is obtained by integrating the transduced spectral density over the relevant frequency range, which is then normalized by the thermomechanical energy at room temperature in the absence of dynamic back-action. The resulting plot in Fig. 6(d) exhibits a clear phonon-lasing behavior, featuring low mechanical energy at power levels below the threshold (≈14μW) and a dramatic increase with a large slope above the threshold. This threshold power is close to a theoretically predicted value of 19 μW, though we note the model used in earlier experiments has assumed a Lorentzian resonance, which may require revision for optomechanical resonators showing mode splitting [15,22,43]. In addition, it is likely important to include other contributions such as the photoelastic effect, which could play an important role in determining the overall optomechanical coupling [48]. A more comprehensive model incorporating these factors is planned for future work. In Fig. 6(d), when the optical power is large enough, the mechanical energy begins to saturate due to the optomechanically induced cavity frequency shifts approaching or exceeding the cavity linewidth.

To benchmark the performance of our SiC optomechanical microdisk resonator, in particular in terms of the attained fm·Qm product, we compare our results to existing optomechanical and electromechanical resonators across common material platforms in Table 1. Note that this survey excludes designs based on the dissipation dilution technique, which allows mechanical resonators to achieve an fm·Qm product above the intrinsic material (Akhiezer) limit [10,49]. As can be seen, the superior optical and mechanical properties of 4H-SiC have already elevated our fm·Qm product to be among the best reported numbers, with only a few experiments surpassing our result. Moreover, both the AlN [20] and LN [27] experiments that reported higher fm·Qm products are for thin-film thickness modes that require non-optical excitation mechanisms. On the other hand, within the 4H-SiC realm, although a higher fm·Qm product has been reported in an earlier work [45] using electrical transduction, the reported mechanical resonator is substantially larger in dimensions, requires external actuation, and can only maintain low dissipation in a vacuum environment. All these limiting factors can greatly restrict device miniaturization, integration, and scaling. In contrast, our devices are specifically designed for seamless integration with other photonic integrated circuit components and feature a notably compact footprint that is comparable to the mechanical wavelength.

In summary, we have demonstrated the first optomechanical resonators in the 4H-SiCOI platform based on ultracompact suspended 4H-SiC microdisks. These resonators, with radii varied from 4.3 to 4.7 μm, exhibit exceptional optical quality factors up to ∼1×106, with their fundamental radial breathing mode featuring mechanical frequency around 950 MHz and mechanical quality factors up to 1.92×104. The corresponding fm·Qm product can be as high as 1.82×1013Hz, which is on par with the highest value reported among all whispering-gallery-mode optomechanical resonators measured in ambient on different material platforms. This advancement offers considerable promise for applications requiring high-precision sensing and metrology. In addition, the inherent material advantages of SiC render these devices well-suited for operations in extreme environments. Such unique combinations of high-performance sensing capabilities and robust material characteristics position SiC optomechanical systems as an ideal platform for addressing challenging applications where conventional technologies face significant limitations.

Acknowledgment. The authors would like to acknowledge helpful discussions with Prof. Gianluca Piazza, as well as contributions from Dr. Ruixuan Wang and Sam Gou in the early stages of the experiment. The CMU team was supported by NSF. Y. Liu and P. Feng are thankful to the partial support from NSF IUCRC MIST Center, NSF Central Florida Semiconductor Innovation Engine, and DARPA DSO OpTIm Program.