Emerging light-driven neuromorphic hardware for artificial intelligence

Optoelectronic neuromorphic systems have emerged as a promising hardware paradigm for next-generation artificial intelligence, combining the high bandwidth, parallelism, and wavelength selectivity of photonics with the adaptive plasticity of electronic materials. This mini-review provides a focused and critical overview of recent advances in material platforms, device architectures, and system-level implementations enabling light-driven neuromorphic computation. We comparatively analyze key photoresponsive materials-including halide perovskites, low-dimensional semiconductors, phase-change and oxide systems, and organic-inorganic hybrids-highlighting their underlying physical mechanisms such as photocarrier generation, charge trapping, ion migration, and excitonic effects. Particular emphasis is placed on device concepts, including optoelectronic synapses, neurons, and crossbar arrays, as well as their integration into in-sensor and hybrid photonic-electronic architectures for machine vision and real-time perception. A benchmarking analysis is presented to evaluate trade-offs in speed, energy consumption, retention, scalability, and stability across different material systems. Finally, we discuss key technological bottlenecks-including device variability, lack of standardized metrics, and integration challenges-and outline future research directions toward scalable, energy-efficient, and application-specific optoelectronic neuromorphic processors.