An aerosol deposition based MEMS piezoelectric accelerometer for low noise measurement

Vibration measurement is crucial in applications such as structural health monitoring (SHM), machine condition-based monitoring, inertial navigation, earthquake early warning, and patient health monitoring. Low-noise accelerometers are generally required. Traditional piezoelectric accelerometers offer superior low-noise performance but there is demand for lighter, cheaper devices. MEMS technology enables miniaturization and cost reduction, leading to low-noise capacitive MEMS accelerometers with in-band noise densities from sub-µg/√Hz to tens of µg/√Hz. However, capacitive MEMS devices often require vacuum packaging to achieve high Q, potentially increasing circuit noise, cost, and fabrication complexity, and they can suffer thermal drift with temperature changes. Piezoelectric accelerometers inherently have high quality factors, reducing the need for vacuum encapsulation, with good temperature durability and linearity, making them reliable for industrial applications. Utilizing microfabrication to develop piezoelectric MEMS accelerometers is thus a promising path to provide low-noise, small form-factor devices. Thick PZT films are desirable to increase sensitivity without excessively thinning mechanical beams, but conventional sol-gel approaches face cracking risks for thicker films. Aerosol deposition (AD) enables room-temperature formation of dense ceramic layers tens of micrometers thick, with easy patterning, low process temperature, and high deposition rate—attributes advantageous for MEMS. In this work, a cantilever-beam, da31-mode piezoelectric MEMS accelerometer using aerosol-deposited PZT is designed, simulated, fabricated, and tested to evaluate its suitability for low-noise SHM applications.

Cantilever beam and symmetric suspension structures are commonly used for piezoelectric MEMS accelerometers due to higher sensitivities at small scale. Among sensing materials (AlN, PZT, ZnO), PZT offers much higher piezoelectric constants, while AlN has lower dielectric constant and better CMOS compatibility. Prior PZT films for accelerometers have typically been ~2 µm thick via sol-gel or magnetron sputtering. Wang et al. used sol-gel PZT with annular diaphragm structures, achieving sensitivities 0.77–7.6 pC/g and resonant frequencies 35.3–3.7 kHz. Hewa-Kasakarage et al. developed PZT cantilever accelerometers via sol-gel with 3.4–50 pC/g sensitivities and 60–1.5 kHz resonance. Saayujya et al. demonstrated a ZnO cantilever with 1.69 mV/g voltage sensitivity and 2.19 kHz natural frequency. Using AlN, Gesing et al. built a four-symmetric suspension beam accelerometer with 510 µg/√Hz noise density, 0.0981 pC/g sensitivity, and 19.1 kHz natural frequency. Trivedi et al. implemented a four-symmetric structure with 1 µm PZT, yielding 5800 µg/√Hz noise density, 8.12 mV/g sensitivity, and 9.62 kHz natural frequency. Despite the desirability of thick PZT films (tens of µm) to boost sensitivity under mechanical strength constraints, prior micro-accelerometer studies have not applied thick-film deposition due to limitations of sol-gel processes (cracking risk from shrinkage and CTE mismatch during repeated thermal treatments). Aerosol deposition overcomes these issues by forming dense, uniform, hard ceramic layers at room temperature, with straightforward patterning and rapid deposition, making it attractive for MEMS.

Design: A 1-axis cantilever accelerometer with a proof mass at the free end and a composite beam (bottom electrode/supporting layer + PZT + top electrode) operating in d31 mode. Layers/materials: top electrode Pt/Ti (~0.22 µm), piezoelectric layer PZT (thick film via aerosol deposition), bottom electrode/supporting layer stainless steel, proof mass tungsten. The PZT and bottom electrode form the composite beam; the PZT sensing area is slightly smaller than the bottom electrode to protect it during wet etching; the top electrode is smaller to prevent shorting. Target specifications for SHM: working bandwidth >100 Hz and noise density <30 µg/√Hz. Simulation (COMSOL Multiphysics): 3D model excludes the very thin top electrode thickness. Material properties: PZT strain-charge form with compliance matrix (units 10^-12 1/Pa), permittivity matrix (relative permittivity times 8.854×10^-12 F/m), and piezoelectric constants (d31=17, d33=73, d15=125.5 in 10^-12 C/N; coupling matrix representation provided). Stainless steel: density 7800 kg/m^3, Young’s modulus 185 GPa, Poisson’s ratio 0.27. Tungsten: density 17800 kg/m^3, Young’s modulus 360 GPa, Poisson’s ratio 0.27. Mechanical damping: Rayleigh damping with damping ratio ζ=0.006, derived from ζ=1/(2Q) using the measured Q. Boundary conditions: left-side surfaces clamped; inertial body load applied as uniform harmonic force density ρ×1 g over the proof mass and adjacent areas. Electrostatics: bottom surface of PZT grounded (0 V); top PZT surface under the top electrode defined as charge output terminal. Noise modeling: Thermal-mechanical and thermal-electrical acceleration noise densities computed as a_nm = sqrt(4 k_B T ω0 /(m Q)) and a_ne = sqrt(4 k_B T η C /(ω0 Q)), with total noise a_na from root-sum-square and expressed as a_na = sqrt(4 k_B T/(ω0 C0 + m ω0^2 Q_r^2)) using parameters: T=297 K, f0=867.4 Hz, m≈0.15 g (from tungsten density × proof mass volume), Q=84, dissipation factor η=0.03 @ 40 Hz, capacitance C=4.58 nF, charge sensitivity Q_r=22.74 pC/g. Experimental measurements: Frequency response measured from 10 to 1300 Hz at multiple excitation levels (0.05–2 g), deriving transmissibility (output charge per acceleration). Working bandwidth characterized by repeating frequency response 14 times at different accelerations over 10–200 Hz and assessing deviation relative to 95 Hz transmissibility. Linearity assessed by output charge versus acceleration from 0.05 to 5 g at multiple frequencies (e.g., around 95 Hz). Noise equivalent acceleration measured from 10 to 300 Hz. Comparative tests: vibration of a fan measured with the fabricated sensor and a commercial piezoelectric accelerometer; shaker-based comparison of noise floor versus a low-noise capacitive MEMS accelerometer (ADXL1001).

- Simulated first natural frequency: 898.9 Hz; measured first resonance frequency: 867.4 Hz (at 0.1 g). - Working bandwidth: 10–200 Hz within ±5% deviation (flat response at low frequencies across 0.05–0.2 g). Four transmissibility curves coincide within 95% confidence; relative error at four resonance peaks ~6%. - Sensitivity: measured charge sensitivity 22.74 pC/g (used in noise modeling); simulated short-circuit sensitivity 36.8 pC/g at 95 Hz (linear output charge vs acceleration). - Noise: simulated thermal-mechanical noise a_nm ≈ 8.6 ng/√Hz (negligible vs electrical); simulated total noise equivalent acceleration a_na ≈ 8.3, 5.9, and 2.7 µg/√Hz at 10, 20, and 95 Hz, respectively. Measured noise equivalent acceleration: 5.6 µg/√Hz at 20 Hz; measured across 10–300 Hz. - Application demonstrations: fan vibration measurements with the fabricated sensor matched well with a commercial piezoelectric accelerometer; shaker tests showed the fabricated sensor has a much lower noise level than ADXL1001 (low-noise capacitive MEMS accelerometer). - Design targets for SHM (bandwidth >100 Hz and noise density <30 µg/√Hz) were met.

The cantilever d31-mode accelerometer with an aerosol-deposited thick PZT layer achieved high sensitivity and low noise suitable for SHM, where small vibrations and bandwidths from tens to hundreds of hertz are typical. The thicker PZT increases the active piezoelectric volume without compromising beam mechanical integrity, improving charge output for a given device footprint. Simulation-guided design predicted sufficient bandwidth and low noise, with measurements confirming a flat 10–200 Hz response and micro-g/√Hz-level noise density, meeting or surpassing SHM requirements. The close agreement between the fabricated sensor and a commercial piezoelectric accelerometer in fan vibration measurements validates practical applicability. Compared to a state-of-the-art low-noise capacitive MEMS device (ADXL1001), the fabricated sensor exhibits a lower noise floor, while avoiding vacuum packaging and associated complexity and thermal drift issues that often challenge capacitive designs. The resonance near 867 Hz ensures a broad sub-resonant measurement band for low-frequency vibration monitoring. Overall, the results substantiate that aerosol-deposited thick-film PZT enables competitive or superior low-noise performance in a compact MEMS form factor, with favorable robustness and process simplicity.

This work demonstrates a MEMS piezoelectric accelerometer fabricated using aerosol deposition of a thick PZT film on a cantilever-beam structure. The device delivers a measured charge sensitivity of 22.74 pC/g, a first resonance frequency of 867.4 Hz, a flat working bandwidth of 10–200 Hz (±5%), and a noise equivalent acceleration as low as 5.6 µg/√Hz at 20 Hz. Simulations corroborate the measured bandwidth and noise levels, indicating electrical-thermal noise dominance and supporting the design choices. Comparative experiments confirm good agreement with a commercial piezoelectric accelerometer and a lower noise floor than a low-noise capacitive MEMS accelerometer (ADXL1001). The aerosol deposition process provides a practical route to thick, high-quality PZT films at low temperature with straightforward patterning, enabling high sensitivity without vacuum packaging. Future work could optimize film properties (e.g., dissipation factor), electrode configurations, and beam geometry to further reduce electrical noise, expand bandwidth, and integrate on-chip readout electronics for system-level miniaturization.