Integrated optomechanical ultrasonic sensors with nano-Pascal-level sensitivity

Principle

As depicted in Fig. 1a, the proposed integrated ultrasonic sensor comprises a suspended SiO₂ membrane and a high-Q Si₃N₄ microring resonator. The SiO₂ membrane is clamped at its periphery to the silicon substrate and suspended at its center to maximize mechanical displacement under ultrasound excitation. The Si₃N₄ microring is embedded within the SiO₂ membrane, enabling optomechanical readout by transducing the mechanical displacement of the membrane into optical resonance shifts of the ring resonator. To maximize the readout signal, we systematically optimize the geometries of the sensor, yielding final radii of 450 μm for the membrane and 235 μm for the microring resonator (see Materials and methods “Device design” Section). As shown in the right inset of Fig. 1a, by locking the frequency of an input continuous wave (CW) laser on the blue-detuned side of the optical resonance, the ultrasound-induced displacement is converted into the laser intensity modulation. The higher the optical Q factor, the steeper the slope of the transmission spectrum, and the greater the readout sensitivity.

Mechanical resonance further amplifies the displacement of the SiO₂ membrane, significantly enhancing the sensitivity to acoustic signals. The bottom left inset of Fig. 1a compares the displacement response of a mechanical oscillator at different frequencies, illustrating the amplification of the mechanical resonance. Figure 1b presents a schematic comparison of integrated ultrasonic sensors with three configurations: (i) conventional non-suspended structure, suspended structure operating at (ii) off-mechanical and (iii) on-mechanical resonance frequencies, highlighting the mechanical resonance-enhanced sensor response. It is worth noting that the sensor’s response to ultrasonic waves is dominated by optical mode shifts caused by variations in the microring radius. Specifically, the contributions from the changes in the coupling gap or refractive index variations induced by photoelastic effect are negligible, as detailed in Supplementary Information Section IB.

Sensor design and fabrication

We fabricate the devices at the wafer scale, with the process flow outlined in Fig. 2a. The fabrication begins with electron beam lithography (EBL) and reactive ion etching (RIE) to create the Si₃N₄ microring resonator⁴⁷, with both processes optimized to achieve near-vertical sidewalls. A SiO₂ cladding layer is deposited on the Si₃N₄ layer to preserve resonator optical performance, with Fig. 2b showing a photograph of a full 4-inch wafer after the deposition. We then flip the wafer upside down and perform photolithography and deep reactive ion etching (DRIE) processes to selectively etch the silicon underneath Si₃N₄ microring resonators. Figure 2c presents a photograph of a 3 × 3 cm² chip after releasing the suspended SiO₂ membranes. For full integration and enhanced robustness, two mode conversion fibers are used to couple light into the microring resonator and are co-packaged with the device (Fig. 2d). This results in a compact and portable ultrasonic sensor with reliable operation in diverse environments.

Device characterization

Figure 2e shows an optical microscopy image of the sensor, with the optical field distribution shown in the inset. The microring resonator exhibits an optical Q factor of 1.33 × 10⁶ in water (Fig. 2f), which is similar to that of 1.35 × 10⁶ in air (Fig. 7e), confirming that the packaging robustly preserves the microring optical performance in various operational environments. The high optical Q factor also ensures the sensor has sufficiently high readout sensitivity, enabling thermal noise to dominate around mechanical resonance frequencies.

The experimental set-up for sensitivity characterization is depicted in Materials and methods “Ultrasound sensitivity characterization” Section. Specifically, a 1550 nm laser is coupled into the microring resonator and detuned to the optimal slope of the optical resonance. The probe laser is locked to the blue-detuned side of the optical mode using a proportional-integral-derivative (PID) controller. The stability of the device is discussed in Supplementary Information Section IIC. A pre-calibrated ultrasound transducer, driven by a vector network analyzer (VNA) and positioned above the sensor, serves as the ultrasound source. The ultrasound-induced optical resonance shift modulates the transmitted light intensity and is detected by a photodetector. The transmission spectrum of the optical mode, noise power spectral density (PSD), and the ultrasound response of the sensor are measured by an oscilloscope, electronic spectrum analyzer (ESA), and VNA, respectively.

Noise PSD across the first-order flapping mode ν_{(0, 0)} is measured in both air and water, as shown in the red and blue dots in Fig. 3a, b, respectively. The orange dash-dotted, green dashed, and black solid curves represent the calculated thermal noise, thermorefractive noise (TRN), and total noise, respectively. Notably, ν_{(0, 0)} mode (with its mechanical profile shown in the insets of Fig. 3a, b) exhibits the greatest spatial overlap with incident ultrasound from above the sensor and therefore is expected to have better sensitivity. Near the mechanical resonance frequency, thermal noise–arising from both surrounding molecule collisions and intrinsic damping of the mechanical resonator⁴⁴–dominates as Lorentzian peaks (orange dashed curves). Resonance frequencies of the ν_{(0, 0)} mode in air and water are 289 kHz and 52 kHz with linewidths of 10 kHz and 2 kHz, respectively. The resonance frequency shifts downward in water compared to air due to the increased effective mass and viscosity⁴⁸. Away from the mechanical resonance frequency, TRN dominates (green dashed curves), originating from temperature fluctuations induced refractive index variation via the thermo-optic effect⁴⁹. TRN can be mitigated through using cavity materials with a smaller thermo-optic coefficient, increasing the cavity mode volume, or improving the thermal stability of the device, with details provided in Supplementary Information Section IA. As investigated in previous works, achieving thermal noise-limited performance is essential for enhanced sensitivity⁴⁴ (Supplementary Information Section IA).

The sensor’s single-frequency linear dynamic range (LDR) at mechanical resonance is approximately 52 dB in both air and water (Supplementary Information Section ID), and all subsequent ultrasound pressures are maintained within this linear range. The ultrasound response of the sensor is measured in air and water by sweeping the applied frequency to the transducer using the VNA, with the results shown in red and blue dots in Fig. 3c, d, respectively. The observed spectral peak matches the mechanical mode identified in Fig. 3a and b, confirming that the mechanical mode amplifies the membrane’s displacement response. As illustrated in the inset of Fig. 3c, the on-resonance case achieves a signal-to-noise ratio (SNR) of 42 dB at 30 Hz resolution bandwidth (RBW), a 7 dB improvement over the off-resonance case (35 dB). Deviations of responses in the off-resonance frequency regions (red and blue dots in Fig. 3c, d) from the theoretical mechanical response curves using mechanical susceptibility functions (black curves in Fig. 3c, d) result from interferences with other mechanical modes near the ν_{(0, 0)} mode (see Supplementary Information Section IC for details). Additionally, the observed response undulations are caused by sound waves reflections between the ultrasound transducer and the sensor chip (Supplementary Information Section IIB).

The NEPs of the sensor at different frequencies are derived from the noise spectra, single-frequency response, and response spectra, as shown in Fig. 3e (air) and 3f (water). The NEP spectra exhibit minima at mechanical resonance frequencies, consistent with theoretical predictions (Supplementary Information Sections IA and IC). Remarkably, our sensor achieves optimal NEPs of 218 nPa Hz^−1/2 at 289 kHz in air and 9.6 nPa Hz^−1/2 at 52 kHz in water, representing a record in microcavity-based ultrasonic sensors. The superior sensitivity in water compared to that in air likely arises from improved acoustic impedance match at the water-sensor interface. We also measured the NEPs in a larger frequency range from 20 kHz to 1 MHz, displayed in the insets of Fig. 3e, f, where the shaded areas correspond to the ν_{(0, 0)} mode, and the pentagrams mark the best sensitivities. Device reproducibility is confirmed across nine sensors from the same batch, showing 1% variation in mechanical resonance frequencies and 38% deviations in sensitivities (Supplementary Information Section IIA). Furthermore, the ν_{(0, 0)} mode demonstrates excellent directional uniformity, with a 3 dB angular bandwidth exceeding 100^∘ (Supplementary Information Section IIB). Crucially, suspended-membrane sensors outperform non-suspended (without DRIE process) devices by two orders of magnitude in NEP (Fig. 9), highlighting the critical role of membrane suspension.

Photoacoustic gas spectroscopy

Highly sensitive air-coupled ultrasonic sensors are indispensable for gas detection in applications ranging from breath analysis and environmental monitoring to hazardous gas detection⁵⁰. Among spectroscopic techniques, photoacoustic spectroscopy (PAS) has emerged as a powerful approach^51,52, leveraging the photoacoustic effect where gas molecules convert intensity-modulated light into acoustic waves through periodic thermal expansion. PAS offers unique advantages, including background-free detection, high sensitivity, and broad spectral coverage⁵³, making it particularly suitable for trace gas analysis.

We demonstrate the application of our highly sensitive integrated ultrasonic sensor in PAS of C₂H₂ gas, with the experimental setup shown in Fig. 4a (details in Materials and methods “Experimental details for photoacoustic gas spectroscopy” Section). Specifically, a pump laser with its intensity modulated by an electro-optic modulator (EOM) is injected into the gas cell containing a mixture of 1% C₂H₂ and 99% N₂, while the integrated ultrasonic sensor within the gas cell detects the generated acoustic waves. PA signals are measured at two modulation frequencies: 288 kHz (on mechanical resonance) and 268 kHz (off resonance), at different pump powers. As the PA signal scales linearly with pump power⁵⁴, the noise-equivalent concentration (NEC) is inversely proportional to the pump power (Fig. 4b). At 350 mW pump power and 1 s integration time, NECs of 2.9 ppm (on resonance) and 13.1 ppm (off resonance) are achieved, demonstrating significant sensitivity enhancement through mechanical resonance. The inset of Fig. 4b quantifies this improvement, showing SNRs of 61 dB at 288 kHz (on resonance) versus 48 dB at 268 kHz (off resonance). The background-free nature of PAS is confirmed by comparing PA signals from C₂H₂/N₂ mixture and pure N₂ (Supplementary Information Section IIIA), with PA responses exclusively emerging at C₂H₂ absorption wavelengths. The PA spectrum of our sensor is obtained by scanning the pump laser wavelength with a step of 0.01 nm, as shown in the blue curve of Fig. 4c, which matches well with the high-resolution transmission molecular absorption database (HITRAN) simulations (red curve). The relative residuals between the two are below 5% (black curve), validating the system’s spectroscopic accuracy.

Underwater ultrasound imaging

Ultrasound imaging technology is a powerful non-destructive technique widely used in medical diagnostics, industrial inspection, and marine acoustics. We demonstrate these capabilities through underwater ultrasound imaging using our highly sensitive sensors. As depicted in Fig. 4d, the experimental setup includes a water tank containing a 500 kHz focused ultrasound transducer, our sensor, and an acrylic sample featuring an “F” shaped groove. When immersed in water, the groove traps air, creating a localized acoustic impedance mismatch. The sample is raster-scanned using a two-dimensional horizontal translational stage. When the ultrasound beam generated by the ultrasound transducer is focused on the “F” region, the air-filled groove reflects the ultrasound waves, resulting in significant attenuation of the transmitted ultrasound signal. This generates a strong contrast between the groove and the surrounding homogeneous regions, enabling high-resolution imaging. For optimal performance, we operate the sensor at the ν_{(3, 0)} mechanical mode of 517 kHz (mode profile shown in the inset of Fig. 10b), achieving higher spatial resolution than the ν_{(1, 0)} due to its shorter wavelength.

The sample is scanned with a 1 mm step size, and the sensor’s response is recorded at each position. Figure 4e(i) and (ii) represent the imaging results obtained with a driving ultrasound pressure of 0.3 mPa and driving frequencies of 517 kHz (on-resonance) and 550 kHz (off-resonance), respectively, revealing significantly improved contrast and clarity under resonant conditions, a direct consequence of mechanical resonance-enhanced sensitivity. Further quantitative analysis (Supplementary Information Section IIIB) confirms a spatial resolution of 1.89 mm, demonstrating the potential of our sensor for high-resolution imaging and long-range target detection in underwater applications. This spatial resolution is primarily determined by the focusing geometry of the external ultrasound transducer, instead of by the sensor itself.

To provide a comprehensive performance benchmark, we include a comparison with a commercial hydrophone. Remarkably, our sensor outperforms a commercial hydrophone even at ultrasound pressures three orders of magnitude lower (0.3 mPa versus 0.7 Pa), as shown in Fig. 4e(iii). This demonstrates the sensor’s imaging capability under extremely weak ultrasound signals. Notably, conventional hydrophones are engineered for fundamentally different operational regimes, typically prioritizing wide bandwidth and high-pressure detection capability. This comparison underscores their complementary nature, with our sensor excelling in scenarios requiring ultra-high sensitivity within specific frequency bands while conventional hydrophones remain preferred for wide bandwidth applications at higher ultrasound pressure levels.

While only transmissive imaging is demonstrated here, the compact size of the sensor facilitates easy integration with an ultrasound transmitter to achieve reflective measurement. Furthermore, adding an impedance-matching layer (such as polymers) on the membrane can further enhance the device’s robustness for contact detection.