Electro-optical memristors: from the origin to device implementation

With the rapid growth of artificial intelligence, conventional von Neumann architectures face increasing limitations arising from memory-processor separation, high data-transfer latency, and power consumption. Memristive devices offer a promising route towards in-memory and neuromorphic computing by combining information storage and processing within compact two-terminal structures. Among emerging memristive technologies, electro-optical memristors are particularly important because they integrate electrical programmability with optical modulation, enabling high-bandwidth, parallel, and energy-efficient signal processing. Unlike purely electrical memristors, which are often limited by interconnect bottlenecks, switching variability, and bandwidth constraints, electro-optical memristors provide additional degrees of freedom through light intensity, wavelength, pulse duration, and timing. Compared with purely optical memory elements, they offer nonvolatile or tunable electrical conductance states that can be directly interfaced with electronic circuits. This review critically summarizes the evolution from conventional resistive-switching memristors to optical and electro-optical memristive systems, with emphasis on material platforms, switching mechanisms, device architectures, synaptic and neuronal functionalities, and system-level integration. Particular attention is given to electro-optical memristors for optoelectronic synapses, in-sensor computing, photonic neuromorphic processing, and brain-inspired hardware. Finally, the key challenges, including device variability, endurance, optical/electrical integration, scalability, and commercialization, are discussed together with future perspectives for next-generation electro-optical neuromorphic computing.