Device-level nonlinearity and temporal memory in optoelectronic reservoir computing

Abstract Reservoir computing (RC) has emerged as a promising computational paradigm for processing temporally correlated and nonlinear data with low training cost. Among various physical implementations, optoelectronic devices provide a unique opportunity to directly interface light with nonlinear dynamical systems, enriching the reservoir state space through device-intrinsic responses. Light can encode information in wavelength, intensity, and pulse duration, and stimulate multiple nodes in parallel with minimal delay or added power. Recent advances in photodiodes, optically modulated memristors, and phototransistors have revealed device-level pathways to enhance nonlinearity, temporal memory, and node diversity, moving beyond purely electrical control toward hybrid optical–electrical tuning. This review revisits these developments from a device physics perspective, highlighting mechanisms for multi-state generation, bidirectional synaptic weight modulation, and temporal response tailoring. We compare diverse excitation schemes, ranging from wavelength- and intensity-selective photocarrier modulation to con optical-assisted filament control and gate–light co-modulation. We also discuss their impact on reservoir performance in pattern recognition, time-series prediction, and dynamic signal processing. We connect material design, device architecture, and reservoir dynamics to outline emerging strategies for scaling optoelectronic RC. This review provides timely insights for researchers working at the intersection of device engineering and neuromorphic computing. Graphical Abstract