Bifacial Interface Engineering of Perovskite Memristors for High-Performance Optoelectronic Synapses and Neuromorphic Vision Systems

Abstract

Conventional complementary metal-oxide-semiconductor (CMOS)-based intelligent visual systems suffer from inherent bottlenecks of high energy consumption and latency, while perovskite memristors are constrained by critical issues including defect-induced instability, unimodal modulation, and random ion migration, hindering their practical deployment. Herein, a universal bifacial engineering strategy is proposed to address these challenges, featuring CF3–PEAI modification at both the prenucleation and postgrowth interfaces of the perovskite layer. The resulting memristor devices exhibit excellent cycling stability, superior uniformity, multiple distinguishable resistance states, and high linearity under both electrical and optical stimulation, enabling reliable emulation of key synaptic plasticity behaviors such as excitatory postsynaptic current (EPSC), short-term memory (STM) to long-term memory (LTM) transition, paired-pulse facilitation (PPF), and learning-forgetting-relearning processes. Furthermore, the device enables high precision dynamic path correction in a feedback controlled trajectory tracking system. The practical feasibility of the device in real world object detection is further verified via digital conductance mapping to YOLOv8 on the BDD100 K data set. This work establishes an effective bifacial engineering approach for high performance perovskite optoelectronic synapses, paving the way for their wide application in high-efficiency neuromorphic visual systems such as autonomous driving, robotic navigation, and smart surveillance.