🔬 applied physics

On-Chip Frequency Noise Cancellation in Nanomechanical Resonators using Cavity Optomechanics

This paper demonstrates an ultracoherent, on-chip nanomechanical resonator system that utilizes correlated frequency fluctuations between two perimeter modes and nonlinear optomechanical transduction to generate a "difference signal" with vastly reduced intrinsic noise, enabling high-precision frequency tracking for sensing applications.

Original authors: Bhavesh Kharbanda, Amirali Arabmoheghi, Letizia Catalini, Mohammad Bereyhi, Geena Benga, Alessio Zicoschi, Christian L. Degen, Tobias J. Kippenberg, Alexander Eichler, Nils J. Engelsen

Published 2026-03-31

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Bhavesh Kharbanda, Amirali Arabmoheghi, Letizia Catalini, Mohammad Bereyhi, Geena Benga, Alessio Zicoschi, Christian L. Degen, Tobias J. Kippenberg, Alexander Eichler, Nils J. Engelsen

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Problem: The "Jittery" Watch

Imagine you have a very expensive, high-tech watch that is supposed to tick perfectly. This watch is actually a tiny, vibrating machine (a nanomechanical resonator) used for super-precise sensing—like weighing a single virus or detecting a single atom.

The problem is that this watch is jittery. Even in a quiet room, its ticking speed wobbles up and down.

Why? It's like trying to listen to a whisper in a windstorm. The "wind" comes from two places:
1. The Room: The temperature changes slightly, making the metal expand and contract, changing the pitch.
2. The Watch Itself: Tiny defects inside the material act like little "ghosts" that randomly push the watch, causing it to speed up or slow down. This is called flicker noise.

Scientists have been trying to build better watches for years. They made the materials purer and the designs stronger to reduce the "wind." But the "ghosts" inside the material are still there, and they are hard to get rid of.

The New Idea: The "Twin Watch" Trick

The researchers at ETH Zurich and EPFL came up with a clever solution. Instead of trying to stop the ghosts, they decided to cancel them out using a trick similar to noise-canceling headphones.

They built a device with two tiny vibrating watches (mechanical modes) right next to each other on the same chip:

Watch A (The Out-of-Plane mode).
Watch B (The In-Plane mode).

The Secret: Even though these two watches vibrate differently, they are made of the same material and sit in the same environment. So, when the room gets slightly warmer, or when a "ghost" pushes the material, both watches wobble at the exact same time and in the exact same way.

The Magic: Mixing the Signals

Here is where the science gets cool. The researchers used a laser (light) to "listen" to both watches simultaneously. Because of the way light interacts with these tiny machines (a field called cavity optomechanics), the laser doesn't just hear Watch A and Watch B separately.

It acts like a mixer in a DJ booth. When two signals are mixed, they create new "beats."

If Watch A vibrates at 1,070,000 Hz and Watch B at 1,092,000 Hz, the laser creates a new signal at the difference: 22,000 Hz.

The Analogy: Imagine two singers singing the same note, but one is slightly sharp and the other is slightly flat. If you subtract one voice from the other, the "wobble" (the sharpness/flatness) cancels out, leaving you with a perfectly steady tone.

In this experiment, the "wobble" (the frequency noise) is correlated. Because both watches are wobbly for the same reasons, when you subtract one from the other, the wobble disappears!

The Result: A Super-Stable Signal

The researchers tested this by trying to detect a tiny, artificial change in the frequency of Watch A (simulating a particle landing on it).

Without the trick: The signal was buried in the noise. It was like trying to hear a pin drop while a truck was driving by.
With the trick: They looked at the "difference signal." The noise vanished. The tiny change became crystal clear.

They found that this new "difference signal" was 20 times more stable than looking at either watch alone. It was so stable that they could track the frequency with incredible precision, even over long periods.

Why This Matters

This isn't just about making a better watch. It's about sensing the invisible.

Mass Spectrometry: Weighing single molecules.
Medical Imaging: Detecting tiny magnetic signals from atoms inside our bodies.
Quantum Computing: Keeping delicate quantum states stable.

The Takeaway

The researchers didn't try to build a perfect, noise-free machine (which is nearly impossible). Instead, they built a smart system that uses two imperfect machines to cancel out their own imperfections.

It's like having two people who are both slightly unsteady on their feet. If they hold hands and walk together, their individual wobbles cancel out, and they can walk in a perfectly straight line. This "on-chip" trick allows scientists to see the world with a clarity they never thought possible.

1. Problem Statement

Precise tracking of mechanical eigenfrequencies is critical for applications such as atomic clocks, mass spectrometry, scanning force microscopy, and magnetic resonance imaging. However, measurement precision is fundamentally limited by three sources of frequency noise:

Thermomechanical noise: Creates a flat frequency noise power spectral density (PSD) at all timescales.
Readout (detection) noise: Creates a frequency noise PSD scaling as $\Omega^2$ , dominating at short timescales.
Intrinsic frequency fluctuations: Includes flicker noise (jitter) and thermal drifts, creating a noise PSD scaling as $1/\Omega^\alpha$ (where $0.5 < \alpha < 2$ ). This dominates at long timescales and is difficult to mitigate via simple time averaging.

While strategies exist to minimize thermomechanical and readout noise (e.g., high-Q designs and cavity optomechanics), intrinsic frequency fluctuations remain a major challenge. Previous methods to cancel these fluctuations relied on post-processing subtraction of correlated modes, which is susceptible to uncorrelated detector noise and cannot be applied in real-time. There was no existing method for on-chip, calibration-free cancellation of these fluctuations.

2. Methodology

The authors developed an on-chip optomechanical platform to address this challenge:

Device Architecture: They co-integrated an ultralow-dissipation perimeter-mode nanomechanical resonator (fabricated from 250 nm stoichiometric silicon nitride, partially thinned to 100 nm) with a photonic crystal optical microcavity (Fabry-Pérot cavity).
- The resonator features a "soft-clamping" effect, yielding extremely high quality factors ( $Q$ ).
- Two specific modes were utilized: an out-of-plane (OOP) mode ( $\omega_1 \approx 1.070$ MHz, $Q_1 \approx 4 \times 10^7$ ) and an in-plane (IP) mode ( $\omega_2 \approx 1.092$ MHz, $Q_2 \approx 3 \times 10^5$ ).
Optical Detection: A continuous-wave laser (1552.6 nm) is red-detuned from the cavity resonance by $\sim 6$ GHz. The mechanical motion of the resonator modulates the cavity detuning, which is read out via the reflected light intensity.
Nonlinear Transduction Strategy:
- Both mechanical modes ( $\omega_1$ and $\omega_2$ ) are driven simultaneously.
- Due to the nonlinear optomechanical transduction of the cavity, the interaction generates mixing products, specifically signals at the sum ( $\omega_1 + \omega_2$ ) and difference ( $\omega_d = \omega_2 - \omega_1$ ) frequencies.
- Crucially, the difference signal $\omega_d$ is generated directly on the chip within the optical cavity before detection.
Hypothesis: Since the intrinsic frequency fluctuations of $\omega_1$ and $\omega_2$ are strongly correlated (stemming from shared environmental temperature changes and intrinsic material defects), their difference ( $\omega_d$ ) should cancel out these common-mode fluctuations, leaving a signal with vastly reduced noise.

3. Key Contributions

Discovery of Strong Correlation: The authors experimentally demonstrated that the intrinsic frequency fluctuations (specifically the $1/\Omega$ flicker noise and thermal drifts) of two distinct mechanical modes on the same chip are strongly correlated.
On-Chip Noise Cancellation: They successfully generated a "difference signal" ( $\omega_d$ ) via nonlinear optomechanical mixing that inherently cancels these correlated fluctuations without requiring external post-processing or calibration.
Superiority over Post-Processing: They proved that this on-chip cancellation outperforms traditional post-processing subtraction ( $\omega_{cal} = \omega_2 - \omega_1$ ) because the on-chip method samples the detection noise only once, whereas post-processing samples it twice (uncorrelated noise adds up).
Proof-of-Concept Sensing: They demonstrated the utility of this stable signal in a frequency tracking experiment mimicking mass sensing, showing significantly improved signal-to-noise ratios (SNR).

4. Key Results

Frequency Stability:
- Individual modes ( $\omega_1, \omega_2$ ) exhibited frequency fluctuations of $>30$ Hz over several hours.
- The difference signal ( $\omega_d$ ) showed a maximum change of only 0.4 Hz, a reduction of roughly two orders of magnitude.
Noise PSD Analysis:
- The $1/\Omega$ noise branch for $\omega_d$ was reduced by an order of magnitude compared to the individual modes at sideband frequencies below 1 Hz.
- The noise floor of $\omega_d$ is limited by the thermomechanical noise of the lower- $Q$ mode ( $\omega_2$ ), estimated at $\approx 1.4 \times 10^{-4}$ Hz $^2$ /Hz.
Allan Deviation:
- At long integration times ( $\tau = 25$ s), the Allan deviation ( $\sigma_\omega$ ) for the difference signal reached 7.54 mHz, which is 20 times lower than that of the individual modes (26.15 mHz).
Sensing Performance (SNR):
- In a frequency jump experiment (simulating mass detection), the SNR for the on-chip difference signal was 4.1, compared to 2.4 for the raw OOP mode and 1.3 for the post-processed difference.
- With a simple moving average filter, the SNR for $\omega_d$ increased to 6.2, consistently outperforming the post-processed signal (4.5).
Reproducibility: The effects were verified across multiple devices and different mode combinations, confirming the robustness of the mechanism.

5. Significance and Outlook

This work presents a breakthrough in nanomechanical sensing by solving the long-standing problem of intrinsic frequency noise cancellation in real-time.

Impact on Sensing: The method enables ultra-precise frequency tracking essential for mass spectroscopy (single-molecule detection) and photothermal particle detection.
Quantum Applications: The combination of high force sensitivity and low frequency noise is crucial for nuclear spin detection and nanoscale magnetic resonance imaging (MRI).
Future Potential: The authors suggest that using two high- $Q$ modes (rather than one high- $Q$ and one lower- $Q$ ) could further reduce the thermomechanical noise floor, potentially achieving frequency noise PSDs below $3 \times 10^{-5}$ Hz $^2$ /Hz. This would create a precise time reference stable across both short and long timescales.

In summary, the paper demonstrates a practical, scalable, and calibration-free method to suppress intrinsic frequency noise in nanomechanical systems, leveraging the nonlinear properties of cavity optomechanics to generate a "clean" difference signal directly on the chip.