Memristive Physical Reservoir Computing

Reservoir computing (RC) has emerged as an efficient neuromorphic framework for temporal information processing, offering low training complexity and hardware-friendly implementation. Memristors' nonlinear dynamics and input-dependent memory effects make them ideal candidates for high-performance physical RC. Based on their conductance modulation, memristors can be classified as electronic or optoelectronic types. However, no systematic review has compared electrically and optically controlled memristive RC. This review fills that gap by comparing them from the device to the system level. We first summarize the resistive switching mechanisms of electronic and optoelectronic memristors, highlighting their distinct roles in RC encoding and processing temporal signals. We then review recent advances in electronic memristive RC, emphasizing architecture innovations and performance improvements in pattern recognition and sequence prediction. Subsequently, we focus on optoelectronic memristive RC, where the high parallelism of optical inputs are harnessed for color vision processing, dynamic gesture recognition, and multi-signals fusion. Notably, we provide a systematic comparison between single-modal and multi-modal RC implementations, demonstrating how hybrid electro-optical stimulation enhances feature diversity and task accuracy. Finally, we outline key challenges and future research directions, including the development of fully hardware-integrated RC systems, system-level multi-modal RC architectures, and novel encoding paradigms.