Parallel reservoir computing exploiting a single MEMS device via blue sideband excitation

Abstract

Physical reservoir computing (PRC) offers a promising pathway for energy-efficient edge intelligence; however, existing micro-electromechanical systems (MEMS) implementations struggle to balance computational dimensionality with hardware complexity. Here, we introduce a parallel reservoir architecture realized within a single MEMS resonator via Blue Sideband Excitation (BSE). Unlike traditional approaches that rely on physical arrays or time-multiplexing, which often compromise structural complexity or processing speed—we utilize a unique combined-frequency driving scheme to simultaneously activate two distinct modes into their nonlinear regimes. This strategy enables two distinct but coupled modal responses to be simultaneously accessed as parallel reservoir readout channels within a single device footprint, doubling the state-space dimensionality without increasing fabrication overhead. Built upon such a mechanism, it is demonstrated that fusing features from these co-existing modes can significantly enhance the reservoir computing ability of the proposed system for complex classification tasks. Therefore, the results suggest that exploiting the inherent multi-modal dynamics of continuous mechanical structures suggests a compact route toward multimodal MEMS reservoir computing within a single-device footprint in next-generation computing hardware.