Interface Engineered Perovskite-Oxide Heterojunction All-Photonic Synapses for Multibit Memory, Optical Logic, and Wearable Neuromorphic Vision.

Abstract

Bio-inspired photonic synapses integrate optical sensing, memory, and processing in a single platform, overcoming the bottlenecks of traditional CMOS-based vision sensors. Most existing neuromorphic vision systems either rely on electrical inputs for bidirectional conductance modulation, limiting their operating speed and bandwidth, or lack nonvolatility, leading to nonlinear weight updates and poor efficiency in image recognition tasks. Here, we report a two-terminal, fully light-controlled synaptic memristor based on CsPbBr3/ZnO nanorod heterojunction that demonstrates 4-bit memory storage and optical logic operations within a single architecture. The device emulates essential functions of both excitatory and inhibitory synapses, utilizing positive photoconductivity under UV illumination (λ = 375 nm) and anomalous negative photoconductivity under visible light (λ = 450 nm). By controlling defect concentrations at the CsPbBr3/ZnO nanorod interface, the device exhibits nonvolatile multibit memory with near-linear, symmetric conductance modulation, achieving 92.4% image recognition accuracy with a convolutional neural network. The bidirectional photoresponse enables reconfigurable optical logic operations, demonstrating integrated logic-in-memory. Additionally, when integrated on a flexible platform, the device demonstrates stable synaptic performance under repeated mechanical bending. These results highlight the potential of CsPbBr3/ZnO nanorod heterojunction-based all-photonic synapses as building blocks for multibit storage, optical information processing, and wearable neuromorphic vision systems.