Deep Learning for Nonlinear Characterization of Electrostatic Vibrating Beam MEMS

In this paper, we integrate deep learning techniques with the motion-induced current method to analyze the nonlinear response of electrostatic MEMS resonators consisting of vibrating beams under electrostatic actuation. The motion-induced current method relies on a transduction mechanism that converts the motion of the resonator to a current signal. The third harmonic of the induced current captures the motion characteristics of the MEMS resonator. We conduct electrical measurements on a MEMS device comprising a microcantilever beam subject to electrostatic actuation using a side electrode. The electrical measurements are verified against their optical counterparts to confirm the suitability of the motion-induced current method to analyze the motion of the MEMS resonator. Next, we develop a model by combining deep learning methods with experimental data aiming to detect the nonlinear dynamics associated with the motion of the resonator when subjected to large actuation voltages. The results demonstrate high prediction accuracy of the data-driven model in terms of capturing the peak resonance, the onset of bifurcation, the occurrence hysteresis and its bandwidth.