🔬 applied physics

1/f frequency-noise reduction in nanomechanical resonators by dissipation dilution

This paper identifies the origin of the $1/f$ frequency-noise floor in nanomechanical resonators as thermal tension fluctuations governed by structural damping, demonstrating that dissipation dilution via high tensile stress effectively suppresses this noise and extends the sensor's useful averaging time.

Original authors: Kostas Kanellopulos, Silvan Schmid

Published 2026-10-15

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Kostas Kanellopulos, Silvan Schmid

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

Imagine a tiny, invisible drumhead, no wider than a human hair, stretched tight across a microscopic frame. This is a nanomechanical resonator, a device so sensitive it can feel the weight of a single protein or the whisper of a single atom. Scientists use these microscopic drums to measure the world with incredible precision, but they face a frustrating ceiling. No matter how quiet the room or how perfect the equipment, the drum's pitch never stays perfectly steady. It wavers in a specific, stubborn way known as "flicker noise," a jitter that gets worse the longer you listen. For decades, this noise floor has been the ultimate barrier, preventing these sensors from reaching their full potential, and no one knew exactly what was causing it.

A team of researchers at the Institute of Sensor and Actuator Systems in Vienna has finally identified the culprit. They discovered that this persistent noise is not a mystery of complex defects or hidden quantum glitches, but simply the thermal noise of the tension itself. Just as a guitar string vibrates because of the heat energy in its atoms, the tension holding this microscopic drumhead also jitters. Because the drumhead is fixed at both ends, it cannot stretch to relieve this jitter; instead, the fluctuating tension causes the pitch of the drum to wobble. The researchers showed that this wobble is a direct consequence of the material's internal friction, a property known as the loss angle, which dictates how much energy the material absorbs as it deforms.

The team tested this idea using silicon nitride strings and membranes, ranging from 30 to 340 nanometers in thickness. They measured how the noise changed when they altered the tension, the size, and the temperature of the devices. They found that the noise floor behaves exactly as their theory predicted: if you increase the tension, the noise drops dramatically, following a precise mathematical rule where the noise level is inversely proportional to the square of the tension. This means that by simply pulling the string tighter, you can silence the noise. They also found that the noise depends on the volume of the material and its intrinsic quality, confirming that the source is the material itself, not the way the device is mounted or the air around it.

Crucially, the researchers ruled out the idea that the noise comes from the surface of the material alone. By comparing devices of different thicknesses, they showed that the noise comes from defects located both inside the volume of the material and on its surfaces. For thicker films, the volume dominates the noise, while for very thin films, the surface plays a larger role. This distinction allowed them to pinpoint the exact density of the microscopic defects causing the trouble. They also examined how the noise behaves at different temperatures. At room temperature and down to about 78 Kelvin, the noise follows the predicted pattern, linking the low-frequency jitter to the high-frequency vibrations of the material. However, below 78 Kelvin, the connection breaks down; the noise stops following the material's measured loss, suggesting that the microscopic defects behave differently when the temperature drops too low, perhaps switching from a thermal activation process to a quantum tunneling one.

This discovery changes how scientists think about building better sensors. Previously, the focus was on making the material as perfect as possible to remove all external sources of damping. The new understanding shows that the limit is not just about removing external noise, but about managing the internal tension. By increasing the prestress on the resonator, engineers can dilute the effect of the material's internal friction, pushing the noise floor lower without needing to find a new, perfect material. This provides a clear, actionable path forward: to build a sensor that can listen to the faintest whispers of the universe, one must simply pull the string tighter. The longest useful time a sensor can measure is now understood to be set by its tension, its volume, its temperature, and the inherent loss of the material, offering a complete explanation for a problem that has puzzled the field for years.

Technical Summary: 1/f Frequency-Noise Reduction in Nanomechanical Resonators by Dissipation Dilution

Problem Statement
The frequency stability of nanomechanical resonators is critical for applications ranging from single-atom mass detection to ultra-precise frequency references. While the short-term stability is well-understood as being limited by thermomechanical noise and detection noise, the long-term stability is constrained by a "flicker" or $1/f$ frequency-noise floor. This floor sets a limit on the useful averaging time (Allan deviation), beyond which further integration yields no improvement. Despite extensive study, the physical origin of this floor has remained unknown. Previous work established that the product of the flicker coefficient ( $h^{-1}$ ) and resonator volume ( $V$ ) is invariant across materials and geometries, suggesting a universal density of material defects (fluctuators), but could not distinguish between volume and surface origins, nor determine the dependence on prestress.

Methodology and Theoretical Framework
The authors propose that the $1/f$ frequency-noise floor is the thermal noise of the resonator's tension, derived directly from the Fluctuation-Dissipation Theorem (FDT) without invoking a specific microscopic model.

Theoretical Derivation: For a resonator with fixed ends, the elongation is preset, allowing the tension to fluctuate. The FDT relates the spectral density of stress fluctuations to the material's loss angle ( $\phi$ ). Assuming structural damping (a frequency-independent loss angle characteristic of amorphous solids like silicon nitride), the fractional frequency noise spectrum follows $S_y(f) = h^{-1}/f$ .
Key Relation: The derived flicker coefficient is given by $h^{-1} = \frac{E k_B T}{2\pi V \sigma_0^2 Q_{el}}$ , where $E$ is Young's modulus, $k_B$ is Boltzmann's constant, $T$ is temperature, $\sigma_0$ is the prestress, and $Q_{el}$ is the intrinsic quality factor under uniform elongation.
Dissipation Dilution Connection: The paper posits that the same mechanism of dissipation dilution that enhances the mechanical quality factor ( $Q$ ) by storing energy in tension rather than bending also dilutes the frequency noise. The noise floor is inversely proportional to the square of the prestress ( $\sigma_0^{-2}$ ) and the volume ( $V^{-1}$ ).
Experimental Approach: The authors tested these predictions using silicon nitride (SiN) strings and membranes with thicknesses ranging from 30 to 340 nm and prestresses from 100 MPa to 1 GPa. They utilized:
- Stress-Engineered Devices: Strings with tapered clamping points to vary prestress within a single film (eliminating composition variables).
- Batch Variations: Membranes of varying lateral sizes to test volume scaling.
- Composition Variations: Stoichiometric vs. silicon-rich films to decouple stress from material composition.
- Cryogenic Data: Re-analysis of published data (Fong et al.) spanning 78 K to 296 K to test temperature dependence and the structural-damping assumption.

Key Results

Volume Scaling: The flicker coefficient $h^{-1}$ scales inversely with volume ( $V^{-1}$ ), confirming the invariance of $h^{-1}V$ across different devices.
Stress Dilution: The data confirms a $\sigma_0^{-2}$ dependence for the flicker floor. This was demonstrated across different devices and, crucially, within a single film using stress-engineered strings. Increasing the prestress by 23% reduced the flicker noise by a factor consistent with the predicted 1.52 reduction.
Thickness Dependence: The inferred loss angle from the noise floor ( $\phi = 1/Q_{el}$ ) exhibits a thickness dependence distinct from the measured bending loss ( $1/Q_{int}$ ). While bending loss is dominated by surface effects for thin films, the noise-active loss angle rises as $1/h$ only below a crossover thickness of $\approx 100$ nm (compared to $\approx 320$ nm for bending loss). This indicates that noise-active two-level systems (TLS) reside in both the volume and the surfaces, with the surface contribution being weighted differently for elongation (noise) versus bending (dissipation).
Temperature Dependence: Between 78 K and 296 K, the noise floor follows the predicted linear temperature dependence scaled by the measured loss angle. Below 78 K, the noise floor decouples from the loss measured at the resonance frequency; the noise remains high while the resonant loss drops, marking the limit of the structural-damping assumption where the TLS distributions probed by the noise and the resonance diverge.

Significance and Claims
The paper claims to resolve the origin of the $1/f$ frequency-noise floor in tensioned nanomechanical resonators, identifying it as the thermal noise of the tension governed by the material's structural damping loss angle.

No Free Parameters: The theoretical model predicts the absolute level of the noise floor using only independently measured material parameters (Young's modulus, density, intrinsic loss) and device geometry, with no fitting parameters.
Mechanism Clarification: The work establishes that the gap between the thermomechanical limit and the observed flicker floor is not due to a new noise source but is a consequence of the tension's thermal fluctuations, which are diluted by prestress.
Design Implications: The findings suggest that the ultimate stability of a resonant sensor is set by its prestress, volume, temperature, and intrinsic loss angle. Improving stability requires increasing prestress or reducing the intrinsic loss angle, rather than merely removing external damping channels that affect the measured $Q$ but not the tension noise.
Universality: The relation holds for amorphous materials (SiN) and explains discrepancies in previous surveys involving crystalline materials, where the lack of stress-induced dilution and different loss mechanisms result in different noise floors.

The authors conclude that the $1/f$ floor is a fundamental thermal limit for tensioned resonators, determined by the interplay between the Fluctuation-Dissipation Theorem and dissipation dilution.

More like this