betaFeedback

WorksPhysics and Astronomy

Enhanced bandwidth in radiation sensors operating at the fundamental temperature fluctuation noise limit

Abstract

Temperature-based radiation detectors are an essential tool for long optical wavelengths detection even if they often suffer from important bandwidth limitations. Their responsivity, and hence their noise equivalent power (NEP), typically degrade at frequencies exceeding the cutoff set by their characteristic thermal response time (τthτ_\text{th}), i.e., at ω>τth−1ω> τ_\text{th}^{-1}. Here we show that this bandwidth limitation can be broken when a radiation sensor operates at its fundamental temperature fluctuation noise limit. The key enabler of this demonstration is a nanomechanical sensor in which frequency stability is limited by fundamental temperature fluctuations over an unprecedentedly large bandwidth of 54 Hz\text{Hz}. In this range, the sensor performance remains within a factor 3 from its peak detectivity (DT∗=7.4×109 cm⋅Hz1/2W−1D_T^* = 7.4 \times 10^9~\mathrm{cm \cdot Hz^{1/2} W^{-1}}) even though the thermal cutoff frequency is 30 times lower (i.e., 1/2πτth=1.8 Hz1/2\mathrmπ τ_\text{th} = 1.8~\text{Hz}). We also derive and validate experimentally closed-form expression predicting maximum bandwidth enhancement in the context of nanomechanical resonators interfaced with a closed-loop frequency tracking scheme.

Intro

Temperature-sensitive nanomechanical resonators (NMRs) [4, 5, 6, 7, 8] [1, 2, 3] have attracted significant interest as an alternative to traditional thermal sensing approaches that rely on electrical sensing (e.g., resistive bolometers [9, 10], pyroelectric detectors [4, 5] [3], and thermopiles [11, 12]. In principle, mechanical-based temperature detection in NMRs can provide immunity to electrical noise (e.g., Johnson–Nyquist noise) that limits performance in traditional thermal-based detectors. This immunity creates a path to reaching the fundamental detectivity limit of thermal photon fluctuation at room temperature DT,photon∗≈1010 cm⋅Hz1/2W−1D_{\mathrm{T,photon}}^{*} \approx10^{10}\ \mathrm{cm} \cdot\mathrm{Hz}^{1/2}\mathrm{W}^{-1} [13].

Reaching this fundamental detectivity limit with NMR-based radiation sensors requires eliminating all non-fundamental sources of frequency fluctuations. In recent years, these non-fundamental fluctuations have been shown to originate in large part from thermomechanical resonator fluctuations, and from noise in the readout used to capture resonator vibration. Comprehensive studies of these noise sources can be found in [14, 15, 16, 17], which also outlines the central importance of including the frequency tracking schemes—such as phase-locked loop (PLL) or self-sustaining oscillators—when predicting frequency noise.

Despite these recent progress in noise modeling, demonstration of NMRs in which these noise sources are minimized below fundamental temperature fluctuation remain scarce. Previous work [18] explored the modeling of thermal fluctuation noise in a PLL frequency tracking scheme with preliminary experimental validation within a limited frequency range. Other recent comparative analysis of silicon nitride nanomechanical resonators have approached the temperature fluctuation noise limit [19] but without reaching it clearly over a broad frequency bandwidth.

Here, we demonstrate NMRs with a frequency stability dominated by temperature fluctuation noise over an unprecedentedly large bandwidth of 54 Hz. We also provide updated modeling for temperature fluctuation noise within closed-loop frequency tracking scheme, thus greatly improving predictability of noise figure in this regime.

Building on this result, we unveil an important benefit of reaching the fundamental temperature fluctuation noise limit over such a large bandwidth. In this regime, we show that detectivity and noise equivalent power remain undegraded at frequencies greatly exceeding the resonator thermal responsivity cutoff (i.e., at ω>1/τth\omega> 1/\tau_{\mathrm{th}}, where τth\tau_{\mathrm{th}} is the sensor characteristic thermal response time).

Theory

When using NMRs to perform thermal radiation sensing, noise equivalent power (NEP) in W⋅Hz−1/2\mathrm{W} \cdot\mathrm{Hz}^{-1/2} can be defined as:

NEP=SyR,(1)\mathrm{NEP} = \frac{\sqrt{S_y}}{R}, \tag*{(1)}

where SyS_y quantifies unwanted resonator frequency fluctuations, and RR quantifies the sensor responsivity, i.e., its sensitivity to absorbed radiation. We use unitless fractional frequency (y=δω/ωry = \delta\omega/\omega_r) throughout this work such that SyS_y is a noise spectral density (one-sided) expressed in units of Hz−1\mathrm{Hz}^{-1}. Similarly, RR is in units of W−1\mathrm{W}^{-1} and is given by

R(ω)=γαG∣Hth(ω)∣,(2)R(\omega) = \frac{\gamma\alpha}{G}\left|H_{\mathrm{th}}(\omega)\right|, \tag*{(2)}

where GG is the total thermal conductance between the NMR and its environment, in W/K, γ\gamma is the optical absorption coefficient at target detection wavelength, and α\alpha represents the temperature coefficient of fractional frequency shift, in K−1^{-1}. This coefficient α\alpha is mostly dictated by the NMR’s material properties and is discussed below (see Eq. (17)) for the specific resonator used in this work. Hth(ω)H_{\mathrm{th}}(\omega) accounts for NMR thermal response time τth\tau_{\mathrm{th}} under uniform illumination, and is defined as:

Hth(ω)=11+jωτth.(3)H_{\mathrm{th}}(\omega) = \frac{1}{1+\mathrm{j}\omega\tau_{\mathrm{th}}}. \tag*{(3)}

For the general case of an arbitrary white noise SyS_y, we see from Eq. (1) that NEP degrades rapidly at frequencies exceeding the resonator thermal response time (i.e., for ω>τth−1\omega> \tau_{\mathrm{th}}^{-1}) because of its inverse dependence on the thermal responsivity filter (Eq. (3)). However, this is not always true if we examine more closely the different possible contributions to SyS_y.

Here, we consider the three most common noise sources for the overall noise profile of NMRs, namely thermomechanical noise Sy,tmechS_{y,\mathrm{tmech}}, instrument readout noise Sy,readS_{y,\mathrm{read}}, and fundamental temperature fluctuation noise Sy,TS_{y,T}:

Sy=Sy,tmech+Sy,read+Sy,T.(4)S_y = S_{y,\mathrm{tmech}} + S_{y,\mathrm{read}} + S_{y,T}. \tag*{(4)}

Among these three noise sources, STS_T is the universal limiting noise for thermal-based radiation sensors of all types [13] (e.g., bolometer, thermopile, pyroelectric detector), including NMRs. For NMRs, Sy,TS_{y,T} can be expressed as:

Sy,T(ω)=4kBT2α2G∣Hth,eff(ω)∣2,(5)S_{y,T}(\omega) = \frac{4k_B T^2\alpha^2}{G}\left|H_{\mathrm{th,eff}}(\omega)\right|^2, \tag*{(5)}

where kBk_B is the Boltzmann constant and TT is the surrounding temperature. Hth,eff(ω)H_{\mathrm{th,eff}}(\omega) denotes the effective thermal response of the resonator, averaging all local thermal energy fluctuations over the sensor volume. As shown in [19], the value of Hth,eff(ω)H_{\mathrm{th,eff}}(\omega) is close to Hth(ω)H_{\mathrm{th}}(\omega), especially when the resonator is coupled to its environment primarily by radiation. We therefore assume Hth,eff(ω)≈Hth(ω)H_{\mathrm{th,eff}}(\omega) \approx H_{\mathrm{th}}(\omega) in the following.

In an ideal case where all non-fundamental noise sources (i.e., Sy,tmech+Sy,readS_{y,\mathrm{tmech}} + S_{y,\mathrm{read}}) are minimized compared to the fundamental temperature fluctuation noise Sy,TS_{y,T} (i.e., Sy≈Sy,TS_y \approx S_{y,T}), NEP reduces to

NEPT=4kBT2Gγ.(6)\mathrm{NEP}_T = \frac{\sqrt{4k_B T^2G}}{\gamma}. \tag*{(6)}

Furthermore, Eq. (6) can be normalized by the sensor area to provide the maximum possible sensor detectivity (D∗=A1/2/NEPD^* = A^{1/2}/\mathrm{NEP}, where AA is the sensor area):

DT∗=γA4kBT2G.(7)D_T^* = \gamma\sqrt{\frac{A}{4k_B T^2G}}. \tag*{(7)}

Another limiting case occurs when the sensor is thermally coupled to its surroundings primarily by radiation on both its faces: G≈2⋅4σSBεAT3G \approx2 \cdot4\sigma_{\mathrm{SB}}\varepsilon A T^3, where σSB\sigma_{\mathrm{SB}} is the Stefan–Boltzmann constant, ε\varepsilon is the total hemispherical emissivity, and a perfect emitter/absorber is assumed (ε=γ=1\varepsilon= \gamma= 1). In this case, D∗D^* converges to the well-known fundamental limit of thermal photon fluctuation (DT,photon∗=1.3 cm⋅Hz1/2W−1D^*_{T,\mathrm{photon}} = 1.3\ \mathrm{cm}\cdot\mathrm{Hz}^{1/2}\mathrm{W}^{-1} for a two-side coupled sensor).

Of central importance, we see in Eq. (6)–(7) that a thermal sensor limited purely by temperature fluctuation noise Sy,TS_{y,T} is unaffected by the thermal filter Hth(ω)H_{\mathrm{th}}(\omega) and can therefore, in principle, have infinitely large sensing bandwidth. More specifically, a sensor can be expected to operate at its fundamental NEP and D∗D^* performance limits (Eq. (6)–(7)) at any frequency where its overall noise Sy(ω)S_y(\omega) is limited by fundamental temperature fluctuations Sy,T(ω)S_{y,T}(\omega). In this case, the sensor bandwidth is not defined anymore by the thermal corner frequency ωth=τth−1\omega_{\mathrm{th}} = \tau_{\mathrm{th}}^{-1}, but instead by the limit frequencies (ωc,read\omega_{c,\mathrm{read}} and ωc,tmech\omega_{c,\mathrm{tmech}}) above which other non-fundamental noise sources (respectively Sy,readS_{y,\mathrm{read}} and Sy,tmechS_{y,\mathrm{tmech}} whose expressions can be found in Supplementary Section S1) begin to dominate over Sy,TS_{y,T}.

To quantify the range of frequencies over which fundamental-level performances are possible, we aim to provide expressions for critical intersection frequencies (ωc\omega_c) above which either Sy,tmechS_{y,\mathrm{tmech}} or Sy,readS_{y,\mathrm{read}} start dominating over Sy,TS_{y,T}. To do so, we must however consider not only the intrinsic noise contributions to SyS_y, but their values after filtering by the frequency tracking apparatus; an optical fiber interferometer combined with a phase-locked loop (PLL) in the present study (see Figure 1a). In this setup, an optical fiber tip is positioned in close proximity to the surface of the NMR, creating a low finesse interferometer between the tip of the optical fiber and the NMR. The reflected light signal is collected by a photodetector and subsequently processed by a lock-in amplifier with built-in PLL frequency tracking.

Figure 1. Frequency tracking setup schematics. (a) Schematic of a frequency tracking apparatus for nanomechanical resonators (NMRs), showing a high-level representation of an optical interferometer and a lock-in amplifier equipped with a phase-locked loop (PLL) function. (b) Block diagram illustrating the signal chain for the filtering process of various fractional frequency noise spectral densities (in units of Hz−1/2^{-1/2}). Here, the noise inputs are unfiltered temperature fluctuation noise S‾y,T\overline{S}_{y,T}, thermomechanical noise S‾y,tmech\overline{S}_{y,tmech}, and instrument readout noise S‾y,read\overline{S}_{y,read}. The overline donation denotes the white noise magnitude before filtering, i.e., S‾y=Sy(0)\overline{S}_y=S_y(0). (c) Image of the NMR used in this work.

The block diagram in Figure 1(b) outlines the noise filtering process when using a PLL to track the resonance frequency. This block diagram configuration is largely inspired from Demir et al. [14], with the addition of temperature fluctuation noise Sy,TS_{y,T}. Under the PLL frequency tracking scheme, the filtered thermomechanical noise can be expressed as

Sy,tmechPLL(ω)=2kBTmeffωr3QArss2∣HmechPLL(ω)∣2,(8)S^{\mathrm{PLL}}_{y,\mathrm{tmech}}(\omega) = \frac{2k_B T}{m_{\mathrm{eff}}\omega_r^3 Q A_{\mathrm{rss}}^2}\left|H^{\mathrm{PLL}}_{\mathrm{mech}}(\omega)\right|^2, \tag*{(8)}

where QQ is the mechanical quality factor, ωr\omega_r is the resonance frequency at a given eigenmode (in rad/s), meffm_{\mathrm{eff}} is the effective mass, ArssA_{\mathrm{rss}} is the resonator driven vibration amplitude (in m), and HmechPLL(ω)H^{\mathrm{PLL}}_{\mathrm{mech}}(\omega) accounts for the PLL filtering effect on thermomechanical noise, as given in [14, 20]:

HmechPLL(ω)=(jωKp+Ki)Hdemod(ω)(jω)2+jωτmech+(jωKp+Ki)Hdemod(ω)(9)H^{PLL}_{mech}(\omega)=\frac{(j\omega K_p+K_i)H_{demod}(\omega)}{(j\omega)^2+\frac{j\omega}{\tau_{mech}}+(j\omega K_p+K_i)H_{demod}(\omega)} \tag*{(9)}

where KiK_i and KpK_p are the corresponding control loop parameters of the PI controller used in the PLL, τmech=2Q/ωr\tau_{mech}=2Q/\omega_r is the mechanical time constant of the NMR, and Hdemod(ω)H_{demod}(\omega) is the lock-in amplifier demodulation filter. The filtered instrument readout noise Sy,readS_{y,read} is given by [21]:

Sy,readPLL(ω)=Sx2Q2Arss2∣HmechPLL(ω)Hmech(ω)∣2(10)S^{PLL}_{y,read}(\omega)=\frac{S_x}{2Q^2A_{rss}^2}\left|\frac{H^{PLL}_{mech}(\omega)}{H_{mech}(\omega)}\right|^2 \tag*{(10)}

where SxS_x is the displacement noise floor of the readout instrument in m2^2/Hz, and Hmech(ω)=(1+jωτmech)−1H_{mech}(\omega)=(1+j\omega\tau_{mech})^{-1} is the intrinsic mechanical filter of the resonator.

We model the fundamental temperature fluctuation Sy,TS_{y,T} as an external noise whereas the intrinsic thermal filter Hth(ω)H_{th}(\omega) of the NMR is outside of the signal loop, as in the block diagram in Figure 1(b). This is analogous to a step frequency change in the resonator described in [17] and is therefore included in the loop in the same way. In this configuration, the filtered Sy,TS_{y,T} becomes:

Sy,TPLL(ω)=4kBT2α2G∣HmechPLL(ω)⋅Hth(ω)∣2(11)S^{PLL}_{y,T}(\omega)=\frac{4k_BT^2\alpha^2}{G}\left|H^{PLL}_{mech}(\omega)\cdot H_{th}(\omega)\right|^2 \tag*{(11)}

Conversely, frequency changes resulting from radiation impinging the resonator follow the same signal chain, such that the intrinsic sensor responsivity RR is filtered by the PLL, resulting in:

RPLL(ω)=γαG∣HmechPLL(ω)⋅Hth(ω)∣(12)R^{PLL}(\omega)=\frac{\gamma\alpha}{G}\left|H^{PLL}_{mech}(\omega)\cdot H_{th}(\omega)\right| \tag*{(12)}

Note that Eq. (11)–(12) corrects our previous model for temperature fluctuation noise in [18]. The conclusions of [18] remain unchanged, but the model proposed herein greatly improves predictability over a broad range of PLL parameters.

We find the limit frequencies (ωc,read\omega_{c,read} and ωc,tmech\omega_{c,tmech}) by solving for the intersection between Sy,TPLLS^{PLL}_{y,T} and, respectively, Sy,readPLLS^{PLL}_{y,read} or Sy,tmechPLLS^{PLL}_{y,tmech}, as illustrated in Figure 2. The critical frequency above which thermomechanical noise exceeds temperature fluctuation noise is well approximated by:

Figure 2. Critical intersection frequency. Example theoretical frequency fluctuation SyS_y outlining the critical intersection frequency ωc\omega_c for low and high drive amplitude ArssA_{rss}. Resonator parameters are the same as presented in the results section, with a phase-locked loop PLL frequency tracking set at 8 Hz bandwidth, with at demodulation BW set at 5 kHz.

ωc,tmech≃2Tα2meffωr3QArss2G⋅ωth(13)\omega_{c,tmech}\simeq\sqrt{\frac{2T\alpha^2m_{\mathrm{eff}}\omega_r^3Q A_{rss}^2}{G}}\cdot\omega_{th} \tag*{(13)}

when the term under the square root is greater than 1. Here, ωth=τth−1\omega_{th}=\tau_{th}^{-1} is the thermal cut-off frequency of the NMR. Likewise, the frequency above which readout noise exceeds temperature fluctuation noise is approximated as:

ωc,read≃(2kBT2ωr2α2Arss2SxG)1/4⋅ωth(14)\omega_{c,read}\simeq\left(\frac{2k_BT^2\omega_r^2\alpha^2A_{rss}^2}{S_xG}\right)^{1/4}\cdot\sqrt{\omega_{th}} \tag*{(14)}

Again, assuming the term under the fourth root is greater than 1.

Interestingly, the analytical expressions for ωc,tmech\omega_{c,tmech} and ωc,read\omega_{c,read} indicate that the limit frequencies are independent of the PLL frequency tracking parameters, but rather on parameters intrinsic to the NMR itself (i.e., α\alpha, meffm_{eff}, QQ, ArssA_{rss}, ωr\omega_r, GG) and the instrument readout noise (SxS_x). In other words, there is no fundamental benefit to adjusting the demodulation bandwidth or PLL control parameter settings to achieve better thermal sensing performances.

Finally, by solving for the intersection between Sy,TPLLS^{PLL}_{y,T} and Sy,readPLL+Sy,tmechPLLS^{PLL}_{y,read} + S^{PLL}_{y,tmech}, one can obtain the overall limit frequency ωc\omega_c above which all combined non-fundamental noises exceed the fundamental temperature fluctuation. This is not straightforward analytically, and we instead provide a phenomenological approximation using the analytically obtained ωc,tmech\omega_{c,tmech} and ωc,read\omega_{c,read}:

ωc−N≈ωc,read−N+ωc,tmech−N,(15)\omega_c^{-N} \approx\omega_{c,read}^{-N} + \omega_{c,tmech}^{-N}, \tag*{(15)}

which yields good results in the current work for N≈3N \approx3, as illustrated in Figure 2.

It is worth noting that the overall intersection frequency ωc\omega_c (Eq. (15)) can transition from being dominated by either ωc,read\omega_{c,read} or ωc,tmech\omega_{c,tmech}, as we vary the drive amplitude. Even though Sy,tmechPLLS^{PLL}_{y,tmech} and Sy,readPLLS^{PLL}_{y,read} both scale with Arss−2A_{rss}^{-2} (see Eq. (8) and Eq. (10)), the intersection equations (Eq. (13)–Eq. (14)) do not, due to the different H(ω)H(\omega) filters. As shown in Figure 2, for the typical resonator parameters and readout noise used in this work, increasing ArssA_{rss} causes the limit frequency ωc\omega_c to shift from being limited by thermomechanical noise (e.g., for Arss=6A_{rss} = 6 nm) to being limited by readout noise (for Arss=120A_{rss} = 120 nm).

Obviously, driving NMR beyond its mechanical linearity would eventually introduce unwanted frequency fluctuation, originating from Duffing nonlinearity. In this work, we consider the amplitude limit given by Manzaneque et al. [22]:

Acrit=234Fsωrβ,(16)A_{crit} = \frac{2}{\sqrt[4]{3}}\sqrt{\frac{F_s}{\omega_r\beta}}, \tag*{(16)}

where β\beta is the Duffing coefficient (in m−2^{-2}) and FsF_s is the sampling frequency of the frequency fluctuations signal. Here, Eq. (16) accounts for the fast-sensing regime in which the intended sampling frequency is much higher than the mechanical cut-off frequency of the NMR (i.e., ω>τmech−1\omega> \tau_{mech}^{-1}). Note that Eq. (16) predicts considerably higher AcritA_{crit} than the more commonly accepted limit [23] (i.e., Acrit=0.56L(σ/QE)1/2A_{crit} = 0.56L(\sigma/QE)^{1/2}, where LL is the size of the NMR, σ\sigma is the tensile stress and EE is the Young’s modulus). For example, setting Fs≈50F_s \approx50 Hz and considering our typical resonator parameters (next section), we obtain Acrit≈250A_{crit} \approx250 nm using Eq. (16).

Methods & Results

In this work, we use a low-stress (70 MPa), 90-nm-thick silicon nitride (SiN) NMR (see Figure 1c) that has a surface

area A≈4.2×10−7A \approx4.2 \times10^{-7} m2^2. The structure is fabricated via laser ablation [24, 25] of a plain square SiN membrane. Detailed fabrication process and characterization of multiple mechanical eigenmodes are subject of a separate article in preparation. We mount this NMR on a steel plate using three pairs of disk magnets and mechanically excite it with a piezoelectric actuator. All components are placed in a custom-built vacuum chamber operating at pressure ∼10−6\sim10^{-6} Torr to sufficiently suppress air damping, therefore maximizing the NMR QQ-factor. We measure NMR displacement using a custom-assembled optical fiber interferometer with the configuration shown in Figure 1(a). It consists of a 1564 nm Orion™ distributed feedback (DFB) laser, a 90:10 coupler, a 5 dB optical attenuator, an optical isolator, and a Thorlabs PDA20CS2 photodetector. A Zurich Instrument Ltd. MFLI is used for both exciting and tracking the NMR resonance frequency.

We first perform a set of measurements on some of the NMR fundamental thermal and mechanical properties. We pick a low-order mechanical mode at 60.2 kHz that has both a high QQ-factor (2.4 × 10^6, see Figure 3a) and that shows a strong fluctuation response, i.e., that is well aligned with the tip of our optical fiber readout (see Figure 3c). Using finite element simulation (see Supplementary Section S3), we find the effective mass meffm_{eff} for this particular mode is roughly a third of the actual NMR mass m0m_0 (i.e., meff=m0/2.9m_{eff}=m_0/2.9, where m0=hAρm_0=hA\rho, hh is the NMR’s thickness and ρ\rho is SiN material density). Using this mode, we vary the actuation amplitude to measure the Duffing nonlinearity. By fitting the frequency–amplitude data shown in Figure 3(b), we extract an experimental Duffing coefficient of β=3.2×1010 m−2\beta=3.2\times10^{10}\ \mathrm{m}^{-2}. We also measure the instrument displacement noise floor (Sx1/2=0.04 pm⋅Hz−1/2S_x^{1/2}=0.04\ \mathrm{pm}\cdot\mathrm{Hz}^{-1/2}) as pictured in Figure 3(c). This calibration is done in a separate experiment by moving the optical fiber tip towards the membrane by a known distance to calibrate the interferometer sensitivity (in V/m), allowing transformation of raw voltage noise to displacement noise. We measure the NMR thermal response time (τth\tau_{th}) by intermittently exposing the NMR to a broadband IR source modulated using an optical chopper at 4 Hz and 10% duty cycle. From the fitted transient response (Figure 3d), we measure τth=90\tau_{th}=90 ms. In this measurement, we set the PLL bandwidth to 40 Hz, which is significantly higher than the thermal cut-off frequency, ωth/2π=1.8\omega_{th}/2\pi=1.8 Hz.

Figure 3. General characterization of the resonator mechanical and thermal response. (a) Amplitude-time trace of a mechanical ringdown for extracting the NMR mechanical Q-factor. (b) Frequency-amplitude plot used for determining the Duffing nonlinearity coefficient of the NMR. (c) Snapshot of the readout instrument displacement noise floor Sx1/2S_x^{1/2}, with thermally excited mechanical modes. (d) NMR’s thermal time constant τth\tau_{th} measurement by intermittently exposing the NMR to broadband IR source modulated using an optical chopper at 4 Hz and 10% duty cycle.

The total thermal conductance GG between the NMR and its environment, as well as the temperature coefficient of fractional frequency shift α\alpha are evaluated by a combination of analytical models and finite-element analysis. We calculate radiative contribution to the thermal conductance using Stefan-Boltzmann law Grad=2⋅4σSBεAT3=4.8×10−7G_{rad}=2\cdot4\sigma_{SB}\varepsilon AT^3=4.8\times10^{-7} W/K, considering a total emissivity ε=0.1\varepsilon=0.1 [26]. The factor of 2 accounts for radiative thermal exchange on both front and back NMR surface. Using finite element simulation (see Supplementary Section S2) and considering a thermal conductivity of 2.72.7 W m−1^{-1}K−1^{-1} [27] for SiN, we find that xrad=65%x_{rad}=65\% of the total thermal exchanges with the surrounding environment occurs through radiation (i.e., 35% via solid-state conduction). The total thermal conductance is then given by G=Grad/xrad=7.4×10−7G=G_{rad}/x_{rad}=7.4\times10^{-7} W/K. Again, using finite element simulation (see Supplementary Section S3), we obtain a temperature coefficient of fractional frequency shift α=6.5×10−3\alpha=6.5\times10^{-3} K−1^{-1}. Noteworthy, this value is very close (within 1%) from the value calculated using the analytical expression derived for drum resonators [26]:

α≃EαT2σ(1−ν)(17)\alpha\simeq\frac{E\alpha_T}{2\sigma(1-\nu)} \tag*{(17)}

where E=300E=300 GPa is the Young’s modulus, αT=2.2×10−6\alpha_T=2.2\times10^{-6} K−1^{-1} is the thermal expansion coefficient of SiN, σ=70\sigma=70 MPa is the tensile stress and ν=0.27\nu=0.27 is the Poisson ratio of SiN.

When measuring fractional frequency noise of the NMR using PLL, we set the demodulation bandwidth of the lock-in amplifier to a high value (5 kHz), ensuring that noise originating from the readout instrument Sy,readPLLS_{y,read}^{PLL} remains largely unfiltered. We then set data acquisition rate significantly higher than the demodulation bandwidth (53 kHz) to avoid signal aliasing. We begin by recording frequency-time traces for multiple levels of drive amplitude ArssA_{rss}, at a fixed PLL bandwidth of 8 Hz, and thereafter convert those to noise spectral density Sy1/2S_y^{1/2} and Allan deviation σA\sigma_A (Figure 4a–Figure 4b). The experimental data, in colored lines, closely match our theoretical predictions (Sy,tmechPLL+Sy,readPLL+Sy,TPLLS_{y,tmech}^{PLL}+S_{y,read}^{PLL}+S_{y,T}^{PLL}) in grey dashed lines, without the need for fitting parameters.

Figure 4. Frequency fluctuation measurements. (a) Comparison between the experimental fractional frequency spectral density Sy1/2S_y^{1/2} (in colorful solid lines) and their corresponding theoretical predictions (in grey dashed lines), at various vibration amplitude ArssA_{\mathrm{rss}} with fixed demodulation bandwidth set to 5 kHz and a phase-locked loop PLL bandwidth of 8 Hz. Theoretical limit frequency ωc/2π\omega_c/2\pi at each ArssA_{\mathrm{rss}} is given as a visual guide. (b) Allan deviation σA\sigma_A at various ArssA_{\mathrm{rss}} using the same data as in (a). (c) Comparison between the experimental Sy1/2S_y^{1/2} (in colorful solid lines) and their corresponding theoretical prediction (in grey dashed lines) at varying PLL bandwidth, with fixed Arss=30A_{\mathrm{rss}} = 30 nm, and a demodulation bandwidth of 5 kHz. (d) Allan deviation σA\sigma_A at various ArssA_{\mathrm{rss}} using the same data as in (c).

As we increase the drive amplitude (ArssA_{rss}), traces increasingly overlap with the predicted Sy,TPLLS_{y,T}^{PLL} (red dashed line in Figure 4a–Figure 4b), indicating that they are dominated by fundamental temperature fluctuation noise over an increasingly wide bandwidth. To make the measurement bandwidth extension effect visually clearer, in Figure 4(a), we include the theoretical overall limit frequency ωc\omega_c for each trace, indicating the point where the noise profile is predicted to transition from the fundamental temperature fluctuation dominated regime to non-fundamental noise dominated regime. In specific, we see that at maximum Arss=120A_{rss}=120 nm, the NMR can operate at the Sy,TPLLS_{y,T}^{PLL} dominated regime up to slightly above 54 Hz—roughly two orders of magnitude greater than the NMR’s intrinsic thermal cutoff frequency ωth/2π=1.8\omega_{th}/2\pi=1.8 Hz. In contrast, this measurement bandwidth enhancement is not seen in Figure 4(c–Figure 4d) where we solely vary the PLL bandwidth while using a fixed ArssA_{rss} of 30 nm. This aligns with our theoretical prediction, as Eqs. (13)–(14) are independent of the PLL bandwidth.

In the noise traces of Figure 4, we observe some discrepancy between the experimental and the theoretical prediction. At low sampling frequency ω≤ωth\omega\leq\omega_{th}, we observe the typically unavoidable additional noise due to systematic drift. At higher sampling frequencies (ω>ωth\omega>\omega_{th}), we also observe some discrepancy between the theoretical and experimental roll-off in Sy,TPLLS_{y,T}^{PLL}. This is especially visible for higher ArssA_{rss} traces. We hypothesize that this discrepancy arises from our assumption of Hth,eff(ω)≈Hth(ω)H_{th,eff}(\omega)\approx H_{th}(\omega) in our theoretical calculation. As hinted in [19], averaging local energy fluctuations into an effective Hth,eff(ω)H_{th,eff}(\omega) thermal filter may result in a greater noise bandwidth than the thermal response measured under uniform heating (i.e., Hth,eff(ω)>Hth(ω)H_{th,eff}(\omega)>H_{th}(\omega)). This phenomenon warrants quantitative experimental investigation in future work.

Readout laser back-action [16, 19] and Duffing nonlinearity [22] are also known to cause unwanted frequency fluctuations, but those are ruled out in the present case. We verify that laser power fluctuation plays a negligible role in this experiment by repeating measurements with a variable optical attenuator (see Supplementary Section S4). We also observe that the frequency fluctuation originating from Duffing nonlinearity is negligible. We note in Figure 4(a) that higher ArssA_{rss} never increases noise. High ArssA_{rss} either reduces noise at very high frequencies (where additive noise dominate) or has no effect on the noise (at frequencies where temperature fluctuations noise Sy,TS_{y,T} dominate). Negligible nonlinear noise is consistent with the critical amplitude predicted by Eq. (16). We are mostly interested in ω/2π\omega/2\pi with sampling frequencies FsF_s between 4 to 54 Hz. By using our measured β\beta value into Eq. (16), we estimate AcritA_{\mathrm{crit}} increases from 70 to 255 nm for this range of FsF_s, which is generally above our experimental ArssA_{\mathrm{rss}}.

Another important aspect to confirm the validity of our model is to confirm the predicted effect of PLL filtering on the NMR’s thermal responsivity, i.e., RPLLR_{\mathrm{PLL}}. In Figure 5, we record traces of the NMR’s responsivity at largely different PLL bandwidths (4 and 40 Hz). For each trace, we modulate the thermal source from 0.5 Hz to 100 Hz. The experimental data closely matches theoretical prediction from Eq. (12) (red dashed lines) in both cases. In contrast, the black dashed line in Figure 5 includes only the NMR’s intrinsic thermal filter Hth=(1+jωτth)−1H_{\mathrm{th}} = (1 + j\omega\tau_{\mathrm{th}})^{-1} which, as expected, does not align with our measurement.

Figure 5. Thermal responsivity measurement. Circular dots represent experimental fractional frequency shift Δω/ωr\Delta\omega/\omega_r induced by impinging thermal radiation at various modulation frequencies under closed-loop frequency tracking. Blue dots represent data with phase-locked loop PLL set to 4 Hz and a demodulation bandwidth of 5 kHz. Orange dots represent with PLL set to 40 Hz and a demodulation bandwidth of 5 kHz. Red dashed lines represent theoretical prediction (Eq. (12)) corresponding to each experimental trace. Black dashed line represents the modulated theoretical thermal responsivity curve without closed-loop frequency tracking (i.e., without PLL).

Lastly, we obtain our sensor NEP and D∗D^* by combining our experimentally measured noise SyS_y (i.e., Figure 4) with our experimentally validated responsivity model (Figure 5). We assume γ=0.4\gamma= 0.4, which corresponds to the peak material absorption of 90-nm-thick SiN membrane at a wavelength λ=11\lambda= 11 µm [26]. Of course, in a practical setting,

this material absorption peak would be too narrow, and a broadband optical absorber would be required. We expect this step to be straightforward since previous work by our group and others [5, 6] have demonstrated that including such radiation absorber does not degrade the sensor mechanical performances.

We present both NEP and D∗D^* in Figure 6 for various ArssA_{\mathrm{rss}} and PLL bandwidth. In all cases, our calculated ωc\omega_c values (using Eq. (15)) accurately predict the corner frequency at which experimental D∗D^* declines. For the highest ArssA_{\mathrm{rss}} drive, this decline occurs at 54 Hz, i.e., at a 30 times higher frequency than the sensor intrinsic thermal bandwidth (ωth/2π=1.8\omega_{\mathrm{th}}/2\pi= 1.8 Hz). As a visual guide, in Figure 6(b), we also include both theoretical D∗D^* and theoretical overall limit frequency ωc\omega_c for each ArssA_{\mathrm{rss}}. We observe strong agreement between the theoretical D∗D^* traces. The minor discrepancies are due to the same reasons discussed in the context of Figure 4.

Figure 6. Enhanced NEP and D∗D^* bandwidth. Experimental noise equivalent power NEP (a) and detectivity (b) of the nanomechanical resonator at various vibration amplitudes ArssA_{\mathrm{rss}} and phase-locked loop PLL bandwidths. As predicted, the critical frequency (ωc\omega_c) at which sensor performance decline significantly is greater than the thermal cutoff frequency (ωth\omega_{\mathrm{th}}), by up to a factor 30. The curves also confirm that ωc\omega_c depends strongly on the drive amplitude (ArssA_{\mathrm{rss}}), but not on the frequency tracking PLL bandwidth (the curves at different PLL completely overlap and appear green). Experimental specific detectivity D∗D^* traces are obtained from the NEP values in (a) using A/NEP\sqrt{A}/\mathrm{NEP}, where A=300 μm\sqrt{A}=300\ \mu\mathrm{m} is the characteristic size of the center pad of the NMR. Theoretical predictions for D∗D^* are shown in dashed lines with matching colors. These predictions assume γ=0.4\gamma=0.4 as the absorption, which corresponds to the material absorption of 100 nm thick SiN at λ=11 μm\lambda=11\ \mu\mathrm{m} wavelength.

As expected also from our theoretical model, PLL bandwidth have negligible influence on the NEP and D∗D^*. At a fixed Arss=40A_{\mathrm{rss}} = 40 nm drive, we observe that varying the PLL bandwidth from 2 to 25 Hz leads to no difference. in sensing performance as the three traces are perfectly superimposed. PLL filtering affects both the responsivity and the noise in the same manner and therefore cancels out in the final NEP and D∗D^*.

Conclusion

In conclusion, we have demonstrated an NMR radiation sensor limited by temperature fluctuation noise over an unprecedented measurement bandwidth (i.e., 30×ωth30 \times\omega_{\mathrm{th}}). This demonstration opens avenues for developing fast NMR-based thermal sensors operating at the fundamental detectivity D∗D^* limit at room temperature, which could serve as next-generation long-wavelength infrared (LWIR) thermal sensors. An obvious next step for this work is to demonstrate the same signal bandwidth enhancement in a sensor comprising a broadband optical absorber, rather than relying solely on the intrinsic material absorption of SiN. This should be straightforward since previous work has demonstrated that including such a broadband radiation absorber does not degrade the sensor mechanical performances.

S1. Expressions for intrinsic thermomechanical and instrument readout noise

Here, we provide expressions for intrinsic thermomechanical noise (Sy,tmechS_{y,\mathrm{tmech}}) and instrument readout noise (Sy,readS_{y,\mathrm{read}}) (i.e., without the loop dynamics shown in Figure 1). The thermomechanical noise in the open-loop case can be expressed as:

Sy,tmech(ω)=2kBTmeffωr3QArss2∣Hmech(ω)∣2,(18)S_{y,\mathrm{tmech}}(\omega) = \frac{2k_{\mathrm{B}}T}{m_{\mathrm{eff}}\omega_{r}^{3}QA_{\mathrm{rss}}^{2}} \lvert H_{\mathrm{mech}}(\omega)\rvert^{2}, \tag*{(18)}

where Hmech(ω)H_{\mathrm{mech}}(\omega) is the intrinsic mechanical filter of the NMR, calculated as 1/(1+jωτmech)1/(1+j\omega\tau_{\mathrm{mech}}) where τmech\tau_{\mathrm{mech}} is the mechanical time constant. The instrument readout noise can be expressed as:

Sy,read(ω)=Sx2Q2Arss2∣Hdemod(ω)∣2,(19)S_{y,\mathrm{read}}(\omega) = \frac{S_{x}}{2Q^{2}A_{\mathrm{rss}}^{2}}\lvert H_{\mathrm{demod}}(\omega)\rvert^{2}, \tag*{(19)}

where Hdemod(ω)H_{\mathrm{demod}}(\omega) accounts for the readout instrument filter (e.g., lock-in amplifier demodulation filter), calculated as 1/(1+jωτdemod)1/(1+j\omega\tau_{\mathrm{demod}}) where τdemod\tau_{\mathrm{demod}} represents the demodulation speed.

S2. Heat transfer simulation

We perform a heat transfer simulation via COMSOL, considering both solid-state conduction and radiative heat transfer. We apply an uniform heat flux of Q=1 μWQ = 1\ \mu\mathrm{W} being absorbed by the SiN nanomechanical resonator (NMR) surface and set the boundary condition at constant room temperature Troom=294.15 KT_{\mathrm{room}} = 294.15\ \mathrm{K}. Figure S1 shows the 2-D temperature profile in steady state during constant heating.

2-D temperature profile of the SiN nanomechanical resonator

Figure S1. 2-D temperature profile of the SiN nanomechanical resonator considering solid-state conduction and radiative heat transfer during 1 μ\muW constant heating.

In COMSOL, we obtain heat flux leaving the NMR via solid-state conduction for a single tether to be Qcond,1 tether=43.425 nWQ_{\mathrm{cond},1\ \mathrm{tether} = 43.425\ \mathrm{nW}}. The total heat flux leaving the NMR via conduction can be calculated as Qcond=8×43.425=347.4 nWQ_{\mathrm{cond}} = 8 \times43.425 = 347.4\ \mathrm{nW} in which the factor of 8 accounts for the total number of tethers supporting the NMR. Due to only two heat transfer modes (e.g., conduction and radiation) included in this simulation, we can calculate the percentage of heat flux leaving the NMR via radiation as:

xrad=Q−QcondQ=1×10−6−347.4×10−91×10−6≈65%.(20)x_{\mathrm{rad}} = \frac{Q-Q_{\mathrm{cond}}}{Q} = \frac{1 \times10^{-6} - 347.4 \times10^{-9}}{1 \times10^{-6}} \approx65\%. \tag*{(20)}

S3. Effective mass and temperature coefficient of fractional frequency shift simulation

Using SiN material density ρ=2900 kg/m3\rho= 2900\ \mathrm{kg}/\mathrm{m}^{3}, and the dimension of the fabricated NMR, we perform dynamic vibration analysis in COMSOL which solves for eigen modes and their corresponding eigen frequencies frf_{r}. According to this simulation, at room temperature T0=294.15 KT_{0} = 294.15\ \mathrm{K}, frf_{r} of mechanical mode (1,2) is 60.8 kHz (see Figure S2) when the tensile stress of the SiN membrane is set to σ=70 MPa\sigma= 70\ \mathrm{MPa}. This closely aligns with the chosen mode in our experiment (fr=60.2 kHzf_{r} = 60.2\ \mathrm{kHz}, see main text). From this simulation, we also obtain an effective mass, meff≈36.5 ngm_{\mathrm{eff}} \approx36.5\ \mathrm{ng} for mode (1,2). This value is roughly a third of the actual mass, m0=hA⋅ρm_{0} = hA \cdot\rho of the NMR (i.e., meff=m0/2.9m_{\mathrm{eff}} = m_{0}/2.9).

Mechanical mode (1,2) of the SiN nanomechanical resonator

Figure S2. Mechanical mode (1,2) of the SiN nanomechanical resonator in COMSOL simulation.

By applying uniform heating on the surface of the NMR in COMSOL until its average temperature increases by ≈1 K\approx1\ \mathrm{K}, we observe frf_{r} of the mechanical mode (1,2) decreases by 393 Hz. From this, we extract the temperature coefficient of fractional frequency α≈6.45×10−3 K−1\alpha\approx6.45 \times10^{-3}\ \mathrm{K}^{-1}. Intriguingly, this value matches closely (i.e., <1%< 1\% discrepancy) with the commonly used equation for estimating α\alpha of the fundamental mechanical mode of a drum resonator under tensile stress, listed as follows:

α=EαT2σ(1−ν).(21)\alpha= \frac{E\alpha_{T}}{2\sigma(1-\nu)}. \tag*{(21)}

We use Young’s modulus E=300 GPaE = 300\ \mathrm{GPa}, Poisson’s ratio: ν=0.27\nu= 0.27, thermal expansion coefficient αT=2.2×10−6 K−1\alpha_{T} = 2.2 \times10^{-6}\ \mathrm{K}^{-1} and tensile stress σ=70 MPa\sigma= 70\ \mathrm{MPa} for this work.

Frequency fluctuation under varying readout laser power

The power intensity fluctuation of the probing laser is known to induce unwanted frequency noise to the temperature-sensitive NMR. This effect arises from the probing laser acting as a continuous, point heat source on the NMR. When the laser power fluctuates, it induces frequency fluctuation whose magnitude scales with both the laser power and the NMR’s thermal responsivity RR (see Eq. (2) in main text), on the timescale set by NMR’s thermal time constant τth\tau_{\mathrm{th}}.

To investigate this, we use a variable optical attenuator (Thorlabs V1550A) to vary the incident laser power on the NMR for multiple traces of frequency noise measurement. In Figure S3, we present frequency noise measurements at different laser power attenuation levels, ranging from 0.25 to 10 dB. We see no increase of the NMR noise when reducing attenuation below the value used in the main text (5 dB). In turn, a low attenuation (i.e., high laser power) helps improve the displacement sensitivity SxS_x of the readout instrument, resulting in a reduction in readout noise Sy,readPLLS^{\mathrm{PLL}}_{y,\mathrm{read}} at higher sampling frequency (f≫1/2πτthf \gg1/2\pi\tau_{\mathrm{th}}).

Frequency fluctuation measurement at varying laser power

Figure S3. Frequency fluctuation measurement at fixed vibration amplitude ArssA_{\mathrm{rss}} and phase-locked loop settings (i.e., PLL and demodulation bandwidth) with varying laser power. 5dB attenuation is used in the main text.

References

  1. [4]Y. Hui, J. S. Gomez-Diaz, Z. Qian, A. Alu, and M. Rinaldi, Plasmonic piezoelectric nanomechanical resonator for spectrally selective infrared sensing, Nature communications 7, 11249 (2016).DOI
  2. [5]M. Piller, J. Hiesberger, E. Wistrela, P. Martini, N. Luhmann, and S. Schmid, Thermal ir detection with nanoelectromechanical silicon nitride trampoline resonators, IEEE Sensors Journal 23, 1066 (2022).arxiv.org/abs/2105.03999
  3. [6]C. Zhang, E. K. Yalavarthi, M. Giroux, W. Cui, M. Stephan, A. Maleki, A. Weck, J.-M. Ménard, and R. St-Gelais, High detectivity terahertz radiation sensing using frequency-noise-optimized nanomechanical resonators, APL Photonics 9 (2024).
  4. [7]Y. Zhang, Y. Watanabe, S. Hosono, N. Nagai, and K. Hirakawa, Room temperature, very sensitive thermometer using a doubly clamped microelectromechanical beam resonator for bolometer applications, Applied Physics Letters 108 (2016).DOI
  5. [8]L. Vicarelli, A. Treducci, and A. Pitanti, Micromechanical bolometers for subterahertz detection at room temperature, ACS Photonics 9, 360 (2022).DOI
  6. [9]B. Wang, J. Lai, E. Zhao, H. Hu, Q. Liu, and S. Chen, Vanadium oxide microbolometer with gold black absorbing layer, Optical Engineering 51, 074003 (2012).DOI
  7. [10]P. Renoux, S. Æ. Jónsson, L. J. Klein, H. F. Hamann, and S. Ingvarsson, Sub-wavelength bolometers: Uncooled platinum wires as infrared sensors, Optics express 19, 8721 (2011).DOI
  8. [11]A. L. Hsu, P. K. Herring, N. M. Gabor, S. Ha, Y. C. Shin, Y. Song, M. Chin, M. Dubey, A. P. Chandrakasan, J. Kong, et al., Graphene-based thermopile for thermal imaging applications, Nano letters 15, 7211 (2015).DOI
  9. [12]B. C. St-Antoine, D. Ménard, and R. Martel, Single-walled carbon nanotube thermopile for broadband light detection, Nano letters 11, 609 (2011).
  10. [13]A. Rogalski, Infrared detectors (CRC press, 2000).
  11. [14]A. Demir and M. S. Hanay, Fundamental sensitivity limitations of nanomechanical resonant sensors due to thermomechanical noise, IEEE Sensors Journal 20, 1947 (2019).DOI
  12. [15]A. Demir, Understanding fundamental trade-offs in nanomechanical resonant sensors, Journal of Applied Physics 129 (2021).DOI
  13. [16]P. Sadeghi, A. Demir, L. G. Villanueva, H. Kähler, and S. Schmid, Frequency fluctuations in nanomechanical silicon nitride string resonators, Physical Review B 102, 214106 (2020).DOI
  14. [17]H. Bešić, A. Demir, J. Steurer, N. Luhmann, and S. Schmid, Schemes for tracking resonance frequency for micro-and nanomechanical resonators, Physical Review Applied 20, 024023 (2023).
  15. [18]C. Zhang and R. St-Gelais, Demonstration of frequency stability limited by thermal fluctuation noise in silicon nitride nanomechanical resonators, Applied Physics Letters 122 (2023).DOI
  16. [19]K. Kanellopulos, F. Ladinig, S. Emminger, P. Martini, R. G. West, and S. Schmid, Comparative analysis of nanomechanical resonators: sensitivity, response time, and practical considerations in photothermal sensing, Microsystems & Nanoengineering 11, 28 (2025).DOI
  17. [20]A. Demir, Understanding fundamental trade-offs in nanomechanical resonant sensors, J. Appl. Phys. 129, 044503 (2021).DOI
  18. [21]S. Schmid, L. G. Villanueva, and M. L. Roukes, Fundamentals of nanomechanical resonators, Vol. 49 (Springer, 2016).
  19. [22]T. Manzaneque, M. K. Ghatkesar, F. Alijani, M. Xu, R. A. Norte, and P. G. Steeneken, Resolution limits of resonant sensors, Physical Review Applied 19, 054074 (2023).DOI
  20. [23]H. Postma, I. Kozinsky, A. Husain, and M. Roukes, Dynamic range of nanotube-and nanowire-based electromechanical systems, Applied Physics Letters 86 (2005).DOI
  21. [24]X. Xie, R. Nikbakht, M. Couillard, R. St-Gelais, and A. Weck, Laser machining of free-standing silicon nitride membranes, Journal of Materials Processing Technology 318, 118001 (2023).DOI
  22. [25]R. Nikbakht, X. Xie, A. Weck, and R. St-Gelais, High quality factor silicon nitride nanomechanical resonators fabricated by maskless femtosecond laser micromachining, Journal of Vacuum Science & Technology B 41 (2023).DOI
  23. [26]C. Zhang, M. Giroux, T. A. Nour, and R. St-Gelais, Radiative heat transfer in freestanding silicon nitride membranes, Physical Review Applied 14, 024072 (2020).DOI
  24. [27]N. Snell, C. Zhang, G. Mu, A. Bouchard, and R. St-Gelais, Heat transport in silicon nitride drum resonators and its influence on thermal fluctuation-induced frequency noise, Physical Review Applied 17, 044019 (2022).

Paper details

vlnc:W_6k7d7nw5
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents