Vision associative memory based on single ferroelectric semiconductor α-In₂Se₃

Ferroelectric-based optoelectronic devices can integrate light sensing, data storage, and in-memory computing, providing a compact and energy-efficient paradigm for artificial vision hardware. Here, we demonstrate an image reconstruction application based on vision associative memory in the single ferroelectric semiconductor α-In₂Se₃, enabling direct hardware-level emulation of advanced neuromorphic functionality without external computation or peripheral circuitry. This device utilizes light-induced nonvolatile ferroelectric polarization switching to tune interfacial band bending and carrier injection, thereby generating multilevel resistance states that function as synaptic weights and enable synaptic-like weight updates. Notably, its intrinsic ferroelectric polarization relaxation effect naturally mimics biological adaptive forgetting process. Several key visual synapse functions have been realized in the α-In₂Se₃ device, including the transition from short-term memory to long-term memory, paired-pulse facilitation, learning–forgetting–relearning behavior, long-term potentiation, long-term depression, and Pavlov’s classical conditioning. Importantly, we demonstrate a new image reconstruction strategy based on the memory decay curves of light-sensing images. By associating and comparing memory fragments of reference images, the original information can be restored. This work establishes a compact neuromorphic-vision platform that co-integrates photodetection, synaptic plasticity, nonvolatile memory, and image reconstruction based on associative memory within a single ferroelectric device. The resulting architecture provides a practical route toward energy-efficient artificial retinas and brain-inspired visual systems.