Physical Reservoir Computing Using Microelectromechanical Systems

The growing interest in artificial intelligence (AI) and edge computing has motivated the need for alternative computational frameworks such as neuromorphic computing, which can efficiently handle data‐intensive tasks. Reservoir computing (RC) is a neuro‐inspired approach for processing complex temporal data using a fixed network (the reservoir) and a simple trainable readout layer. To reduce the complexity of computation and energy consumption, physical reservoir computing (PRC) extends this concept to dynamical physical systems by exploiting their intrinsic nonlinearities and memory effects. Among the different substrates that have been proposed for PRC, microelectromechanical systems (MEMS) have shown promise due to their nonlinear mechanical behavior and direct interaction with physical stimuli. Also, being compact and scalable, MEMS devices are widely used as sensors and transducers in commercial applications, making them a suitable choice for low‐power in‐sensor computing and edge intelligence applications. This review discusses recent progress in MEMS‐based PRC with a focus on its underlying principles, device architectures, and the potential of MEMS in neuromorphic computing applications.