Ultrafast Reservoir Computing based on Nonlinear Nanomechanical Resonators at Ambient Conditions

Abstract

Reservoir computing offers an energy-efficient alternative to deep neural networks (DNNs) by replacing complex hidden layers with a fixed nonlinear system and training only the final layer. This work investigates nanoelectromechanical system (NEMS) resonators for reservoir computing, utilizing inherent nonlinearities and the fading memory effect from NEMS's transient response. This approach transforms input data into a higher-dimensional space for effective classification. The smaller size and higher operating frequencies of the NEMS resonators enable faster processing rates than recent demonstrations with micromechanical systems, while their compact footprint and ability to operate under ambient conditions simplify integration into practical applications. Through an MNIST handwritten digit recognition test, this system achieved 90% accuracy with a 3.3-microsecond processing time per pixel, highlighting the potential for various applications that require efficient and fast information processing in resource-constrained environments.