Neuromorphic Acoustic Sensors Using Piezoelectric Mems Resonators with Epitaxially Grown BiFeO3 Films

The neuromorphic acoustic sensor was developed by the piezoelectric MEMS resonator array and physical reservoir computing (PRC). By using the resonators as neurons in PRC, high sensitivity and edge AI processing can be achieved simultaneously. Mn-substituted <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{BiFeO}_{3}$</tex> epitaxial thin film was employed as the piezoelectric film because it has moderate relative permittivity (~ 100) and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$e_{31,f}$</tex> piezoelectric constant (~−6 C/m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>), which allow the resonator to be miniaturized without losing sensitivity due to its low susceptibility to parasitic capacitance. The sensor showed a sensitivity of 7.7 mV/Pa at a resonance frequency of 433.3 Hz. For information processing capability, good prediction with a normalized mean squared error of 0.013 was demonstrated in second-order time series prediction of a nonlinear autoregressive moving average task.