Frequency-stable nanophotonic microcavities via integrated thermometry

TL;DR

The paper tackles the challenge of environmental and crosstalk-induced thermal perturbations that limit the long-term frequency stability of on-chip high-Q microresonators and Kerr combs. It introduces a fully integrated thermometry scheme using a thin-film Pt resistor above the microcavity, enabling a one-time calibration that maps resonance wavelength to thermometer resistance and allows absolute resonance tuning by thermometry alone. The approach delivers RMSE below 0.8 pm over days, enables a DFB laser to be locked with ~48× suppression of drift to within ±0.5 pm over 50 hours, and stabilizes a Kerr comb without photodetection, all in an unpackaged, chip-scale platform. This method provides a scalable, hardware-efficient path to robust photonic frequency references and chip-based Kerr comb devices for field-deployed classical and quantum applications.

Abstract

Field-deployable integrated photonic devices co-packaged with electronics will enable important applications such as optical interconnects, quantum information processing, precision measurements, spectroscopy, and microwave generation. Significant progress has been made over the past two decades on increasing the functional complexity of photonic chips. However, a critical challenge that remains is the lack of scalable techniques to overcome thermal perturbations arising from the environment and co-packaged electronics. Here, we demonstrate a fully integrated scheme to monitor and stabilize the temperature of a high-Q microresonator on a Si-based chip, which can serve as a photonic frequency reference. Our approach relies on a thin-film metallic resistor placed directly above the microcavity, acting as an integrated resistance thermometer, enabling unique mapping of the cavity's absolute resonance wavelength to the thermometer's electrical resistance. Following a one-time calibration, the microresonator can be accurately and repeatably tuned to any desired absolute resonance wavelength using thermometry alone with a root-mean squared wavelength error of <0.8 pm over a timespan of days. We frequency-lock a distributed feedback (DFB) laser to the microresonator and demonstrate a 48x reduction in its frequency drift, resulting in its center wavelength staying within +-0.5 pm of the mean over the duration of 50 hours in the presence of significant ambient fluctuations, outperforming many commercial DFB and wavelength-locker-based laser systems. Finally, we stabilize a soliton mode-locked Kerr comb without the need for photodetection, paving the way for Kerr-comb-based photonic devices that can potentially operate in the desired mode-locked state indefinitely.

Figures (13)

• Figure 1: Integrated thermometry.a, Conceptual illustration of integrated thermometry applied to stabilize a high-Q monolithic microresonator against thermal fluctuations arising from ambient heat sources and crosstalk from other thermally-tuned devices on the same chip. A thin-film metallic (Platinum) resistor, typically used as a microheater, exhibits a temperature-dependent resistance due to the metal’s natural temperature-dependent resistivity. As a result of the low heat capacity of the thin-film resistor, small fluctuations in the heat flow in its vicinity lead to large changes in its temperature, enabling the real-time monitoring of on-chip temperature by simply measuring its electrical resistance. A second identical resistor is used as a heater to perform active stabilization purely using the resistance thermometer, thereby eliminating the need for optical probing for stabilization. Image for data center sourced from Adobe Stock, Ref. dataCent. Image courtesy for picture of co-packaged device: Kaylx Jang. b, Current-voltage-resistance (I-V-R) characteristics of the resistance thermometer fabricated in this work. The I-V measurements (pale blue dots) deviate from the linear trend (dark blue line), and the resistance (yellow dots) exhibits a quadratic dependence (purple line) on the thermometer voltage. c, The measured thermometer resistance (lavender dots) also exhibits a quadratic dependence (green line) on the voltage applied across the (second) heater, indicating a linear dependence on power dissipated across the heater. d, Resonance frequency shift (green circles) of a high-Q microcavity of free-spectral range (FSR) 76 GHz, measured using a calibrated piezo-tuned probe laser, and measured change in thermometer resistance (gold line), for a sinusoidal perturbation applied via the heater. The cavity temperature and resonance frequency exhibit a strong anti-correlation. e, Measured resonance frequency shift plotted versus temperature measured by the thermometer, same data as in d. The linear fit provides an in-situ measurement of the thermo-optic coefficient of the fundamental spatial mode of the cavity waveguide.
• Figure 1: Thermal design for integrated thermometry.a, Numerically simulated steady-state temperature profile for the scheme described in this work. The heater power is set to $100$ mW for the purposes of illustration. The initial temperature of the chip is $293$ K. b, Heater-thermometer spacing design trade-off, showing the core-thermometer temperature differential (left y-axis, in blue), and the heater power required for core heating of $0.1$ K (right y-axis, in dark red), versus heater-thermometer spacing.
• Figure 2: Long-term stabilization of the absolute resonance frequency of a high-Q SiN microresonator using integrated thermometry. a, Simplified experimental schematic, depicting a tunable probe laser that is piezo-scanned across a microcavity resonance. Absolute frequency calibration of the scan is performed by producing a heterodyne note between the tunable laser and a stable reference laser, and simultaneously tracking the reference laser’s wavelength drift using a benchtop wavelength meter. The heterodyne beat note as well as the cavity resonance line-shape are measured synchronously with the piezo sweep using a real-time oscilloscope connected to a computer interface for periodic data acquisition every 10 seconds. For $\xi$=1, the pseudorandom perturbation applied via the heater yields a root mean-squared (RMS) of the induced cavity resonance frequency fluctuation that is equal to the cavity resonance full-width-at-half-maximum (FWHM) linewidth, which is 75 MHz for the microcavity fabricated in this work. b, Drift in the microcavity’s calibrated resonance frequency over the duration of 24 hours, for free-running ($\xi$=0), open-loop ($\xi$=7.55), and stabilized ($\xi$=7.55) operation. The free-running cavity exhibits a slow but strong drift in its resonance frequency due to ambient temperature drift in the lab, whereas the open-loop cavity additionally exhibits strong "fast" fluctuations due to the perturbation introduced. The stabilized cavity’s resonance frequency remains highly stable in the presence of very strong ambient and crosstalk-induced perturbations. c, Allan Deviation (ADEV) versus averaging time, computed using the data displayed in b. The stabilized cavity exhibits an ADEV that is well below the free-running and open-loop modes of operation. The shape of the ADEV curve of the stabilized cavity matches that of $1/f$ noise, indicating that performance is only limited by control electronics. d, Histogram plots depicting the fraction of time over 24 hours (y-axis) that the microcavity’s absolute resonance frequency drift lies within a given frequency bin with a width of 5 MHz.
• Figure 2: Measurement of temperature coefficient of resistance of the integrated Platinum thermometer.a, Experimental schematic depicting a test chip, containing a platinum resistor (referred to here as device under test (DUT)) fabricated on top of a SiN microresonator, placed on a temperature-stabilized hot plate chuck inside a vacuum chamber. Its electrical resistance is probed as a function of temperature by performing current-voltage characterization using a Keithley $2400$ series SourceMeter®. b, Measured electrical resistance of DUT versus temperature (dark-blue filled circles) plotted alongside a linear fit (magenta solid line).
• Figure 3: Demonstration of thermometric absolute-resonance-wavelength lookup table.a., Simplified experimental schematic, depicting a tunable probe laser that is piezo-scanned across a microcavity resonance. The absolute resonance wavelength is measured by producing a heterodyne beat between the tunable laser and a stable reference laser, and simultaneously tracking the reference laser’s wavelength using a highly-stable benchtop wavelength meter. The current produced by the low-noise current source is monitored by measuring the voltage across a series $50.0$$\Omega$ bulk resistor using a precision 6.5-digit digital multimeter (DMM). The PID loop is implemented using a commercial FPGA, and the set-point voltage is input via a computer interface. b. Calibration measurement of the absolute resonance wavelength vs the thermometer resistance. The linear fit extracted from this one-time calibration is utilized in subsequent measurements to compute the desired thermometer resistance to achieve any desired resonance wavelength within the calibration range. c. Demonstration of arbitrary absolute-wavelength tuning of the microresonator via thermometry. A string of 10,000 pseudorandom "target" wavelengths (red hollow circles) are provided as inputs to the thermometric control system illustrated in a., which computes the desired temperature for each target wavelength and stabilizes the microresonator to the computed temperature. The resonance wavelength is measured (blue-magenta filled circles) and compared with the target wavelengths. A zoomed-in portion of 100 consecutive measurements out of 10,000 is shown. d. Histograms showing the deviation between the target and measured absolute resonance wavelengths following a one-time calibration. The histogram bin size is 0.1 pm. The first set of 10,000 measurements, which concluded 12 hours after the calibration run, exhibits a root-mean squared absolute-wavelength error of 0.11 pm. The second set of measurements three days later shows a similar narrow distribution of absolute-wavelength error, but exhibits a systematic error, resulting in a 0.77 pm.