Machine Learning-Assisted Nonlinearity Decouoling for MEMS Resonators Within Closed-Loop Configuration

This paper presents a machine learning-based approach to decouple nonlinear effects in MEMS resonators operating in a closed-loop configuration. The resonator features capacitive-drive and piezoresistive-detection transduction, exhibiting mechanical nonlinear characteristics in both open-loop and closed-loop configurations. A multilayer perception (MLP) algorithm is trained to analyze the amplitude-frequency (A-f) response of the resonator in the nonlinear regime in open-loop configuration and estimate the key parameters of the nonlinear system. These parameters are then incorporated in a decoupling model to mitigate the nonlinear frequency shifts observed in closed-loop configuration. The experimental results show this approach enables the resonator to increase critical amplitudes in the nonlinear regime by 8 dB compared to closed-loop operation without decoupling, demonstrating its effectiveness in mitigating nonlinear effects.