Improved Learning Performance in Physical Reservoir Computing Using Coupled Triple MEMS Nonlinear Resonators
Kosuke Shima, Hiroki Takemura, Masaki Shimofuri, Amit Banerjee, Jun Hirotani, Toshiyuki Tsuchiya · 2025
This paper reports on physical reservoir computing (PRC) using electrostatically coupled triple MEMS nonlinear resonators and demonstrates for the first time that it is effective in improving learning performance to increase the number of resonators to obtain virtual nodes from one to three. The resonator of doubly supported beam is made of single crystal silicon using a SOI wafer. We first numerically evaluated the learning performance of this novel computing system, followed by experimental evaluation by the actual device. Numerical simulation and experiments show similar results, indicating that using all resonator motion states for machine learning increases memory capacity in a nonlinear task.