Energy-Level Regulated Organic Memristive Synapses for High-Accuracy Visual Recognition

Abstract

Organic semiconductors have emerged as compelling candidates for synaptic devices due to their biocompatibility, mechanical flexibility, tunable optoelectronic properties, and scalable production. Here, we report an energy-level-regulated organic synaptic device based on a photovoltaic material, in which interfacial engineering and bulk heterojunction (BHJ) blending are jointly employed to modulate charge injection, transport, and accumulation. By introducing a MoO3 interfacial layer and constructing BHJ active layers with ZY-4Cl and P3HT, the device exhibits stable, reversible conductance modulation at a low operating voltage and reduced leakage current, highlighting the effectiveness of energy-level alignment and interfacial control. The resulting devices exhibit a range of memristive and synapse-relevant electrical characteristics, as well as dual-mode modulation under both electrical and optical stimuli. In particular, photoresponses enable effective optical erasing and writing, providing an additional degree of freedom for device operation. When integrated into a convolutional neural network for MNIST handwritten digit classification, it achieves 94% recognition accuracy. This work highlights the importance of energy-level engineering and interfacial design in organic memristive devices. It provides a materials-oriented strategy for developing low-power, multifunctional memristive elements compatible with neuromorphic and in-memory computing architectures.