From optoelectronic synapses to physical reservoir computing: materials, progress and application

Optoelectronic neuromorphic computing systems have emerged to overcome the Von Neumann bottleneck by mimicking the structure and function of the human brain. These systems integrate sensing, memory, and computing within a single device, enabling low-power operation and fast processing speeds. Their operation is intertwined with synaptic plasticity and learning behavior, which are governed by intrinsic material properties and underpinned by distinctive light–matter interactions. In this study, we highlight emerging material platforms such as two-dimensional materials, perovskites, nanostructures, and conventional materials. Their superior optoelectronic performance and simple fabrication have facilitated the development of optical neuromorphic devices for both conventional neural networks and physical reservoir computing. In particular, physical reservoir computing exploits nonlinearity and fading-memory behavior, which distinguish it from conventional neural networks. Unlike traditional neural network that rely on static weight updates, reservoir computing uses system dynamics to compute and store temporal information. Recent demonstrations have exhibited real-time, low-power, and high-accuracy performance in diverse classification tasks, such as fingerprint recognition and human behavior analysis.