The experiment was conducted in air at room temperature and atmospheric pressure, and the result presents a device-level balance among sensitivity, response speed and integration rather than the lowest noise value reported for every thermal sensor. The platform combines a suspended microring optical cavity with an ultrathin central membrane made from thin-film lithium niobate. The material offers low optical loss, strong piezoelectricity, a large thermal expansion coefficient of 16.7 parts per million per kelvin and low thermal conductivity of 4.6 watts per metre-kelvin.

After the final suspension step, the measured intrinsic optical quality factor was approximately 1 million, allowing mechanical vibration to be detected through phase modulation. The researchers used a tunable continuous-wave laser, Pound–Drever–Hall locking and balanced homodyne detection to suppress laser-intensity and cavity-detuning noise. The optical output could also be suitable for remote fiber-based interrogation, although the reported measurements used a laboratory setup.

At the center of the design is a bowl-shaped resonator with outer and inner radii of 20 and 17 micrometres. Its outer rim supports a circulating optical mode, while four narrow anchors, each 400 nanometres wide and 20 micrometres long, connect the structure to the slab layer and substrate. A two-step fabrication process first patterns the microring and bus waveguides, leaving an 80-nanometre-thick slab, and then defines the anchors and air gaps before release.

The central membrane was selectively trimmed from 300 to 80 nanometres to reduce thermal mass. Because thermal conductance and thermal capacitance ordinarily scale proportionally when a membrane is thinned, the design was intended to reduce the heat capacity while retaining the thermally isolating anchor structure. Finite-element simulations estimated an intrinsic thermal conductance of 108 microwatts per kelvin and a thermal capacitance of 540 picojoules per kelvin for the optimized device.

These values correspond to an intrinsic time constant of 5 microseconds and a thermal bandwidth of 31.8 kilohertz. Simulated heating applied across the resonator produced a 5-microsecond response, while heating the surrounding silicon substrate produced a 6.8-microsecond response because heat had to reach the device through an additional path. The experiment used square-wave-modulated visible light absorbed by the silicon substrate as a stand-in for a dedicated on-chip absorber.

Fitting the resulting optical waveform gave a loaded response time of 7.2 microseconds, consistent with the simulation. The fundamental radial mode shifted downward in frequency as temperature increased. Between 289 and 306 kelvin, the researchers measured a temperature coefficient of frequency of minus 124 parts per million per kelvin for the 83.7-megahertz mode.

Three measured modes showed coefficients ranging from minus 106 to minus 263 parts per million per kelvin. The reported fundamental-mode value exceeds the magnitudes below 50 parts per million per kelvin cited for other integrated platforms in the study. The resonator also showed a thermo-optic coefficient of 30 picometres per kelvin without spectral enhancement.

Its thermomechanical output places the temperature-dependent frequency shift on an electronically accessible megahertz carrier, allowing continuous tracking without repeated acquisition of an optical spectrum. Optical pumping increased the mechanical response once the device crossed a self-oscillation threshold. The measured threshold was 200 microwatts at the experimental detuning, although the researchers estimated that optimizing the laser–cavity detuning could reduce the threshold to 110 microwatts.

Above threshold, the peak mechanical spectral intensity increased by 80 decibels and the linewidth narrowed to 1.6 kilohertz, corresponding to an effective mechanical quality factor of 5.1 times 10 to the fourth power. The experiment also yielded a single-photon optomechanical coupling rate of 4.2 kilohertz, consistent with simulation. This coupling combined moving-boundary, photoelastic and electro-optic contributions because mechanical displacement in lithium niobate can create an internal electric field through its piezoelectric response and then modulate the optical field through the electro-optic effect.

Allan-deviation measurements were used to convert frequency fluctuations into an estimate of detectable optical power. After oscillation began, the corresponding value fell by two orders of magnitude to 6.2 nanowatts per square-root hertz. At longer averaging times, however, frequency flicker noise and random-walk noise became more prominent.

The authors note that high-power laser instability may contribute to these effects and could be reduced through improved detection or locking. Operation under vacuum was not tested; the study projects that reducing ambient pressure to 1 millibar could lower thermal conductance by three orders of magnitude and yield a projected noise-equivalent power as low as 6.2 picowatts per square-root hertz. The researchers also demonstrated optical-power detection by illuminating the suspended region with 780-nanometre light while applying 20 microwatts of telecom-wavelength power to the resonator itself.

Below 1 microwatt of incident visible power, the mechanical-frequency response was approximately linear, with a responsivity of 30 kilohertz per microwatt; at higher power, the response gradually saturated, defining the demonstrated linear operating range. The present device is wavelength-selective and produced no measurable response to telecom or mid-infrared input. The authors propose integrating a patterned absorber directly onto the suspended membrane to increase the fraction of incident power converted into a measurable response and extend the operating wavelength range.

That absorber was not part of the reported experiment. Several system-level limitations remain. The noise measurements came from a single sensing element rather than a completed infrared array, and array-level operation was not demonstrated.

The uncooled measurements were conducted in air at room temperature and atmospheric pressure, where gas damping contributes to mechanical loss and thermal conductance. Long-term frequency stability became less favorable as flicker and random-walk noise emerged after self-oscillation, and the optical-power test used visible-light absorption rather than a purpose-built infrared layer. The study nevertheless shows that co-designing the optical cavity, mechanical mode and heat-transfer path can produce a compact sensor with a microsecond-scale response and integrated optical readout.

Better absorbers, optimized anchors, piezoelectric feedback, improved frequency locking and reduced-pressure operation remain proposed development routes rather than demonstrated outcomes of this study.

Subject of Research: Experimental thin-film lithium niobate optomechanical sensing of temperature and optical power at room temperature

Article Title: Uncooled low-noise optomechanical sensors on thin-film lithium niobate

Article References: Yu, Y., Yin, R., Anderson, I., Wang, Y., Kramer, J., Christen, I., Lee, C.-H., Ren, X., Chen, Z., Povinelli, M. L., Wasserman, D., Lu, R., & Yu, M. (2026). Uncooled low-noise optomechanical sensors on thin-film lithium niobate. Light: Science & Applications, 15(1), Article 389. https://doi.org/10.1038/s41377-026-02506-z

Image Credits: AI Generated

DOI: 10.1038/s41377-026-02506-z

Keywords: thin-film lithium niobate, optomechanical sensor, thermal sensing, optical power detection, microring resonator, low-noise measurement, self-oscillation, room-temperature operation, thermal response, infrared detection, Uncooled, low-noise