Reservoir Computing with a MEMS Nonlinear Resonator for In-Sensor Computing

Reservoir Computing (RC) is a promising neuromorphic paradigm for future applications on the Internet of Things(IoT) and is characterized by low training costs and hardware compatibility. This study introduces a MEMS resonator as a nonlinear node for non-delayed RC(ND-RC), specifically tailored for processing time-varying data. To investigate the nonlinear dynamics of a clamped-clamped silicon beam subjected to electrostatic driving, this study employs both experimental and numerical analyses. By utilizing the critical Duffing nonlinear parameters of a single MEMS resonator, ND-RC is implemented, reducing processing delays and enhancing the system dimensionality. Parametric optimization of the RC operation was conducted through numerical analysis, demonstrating its feasibility. The resonator exhibited high-performance capabilities, as evidenced by its success in a handwritten digit recognition benchmark, surpassing the reported works.