Novel Optoelectronic Reconfigurable Transistors Based on Graphene/VO2 Heterojunction for Efficient Neuromorphic Perception, Computation, and Storage.

Abstract

Optoelectronic artificial neuromorphic devices, inspired by biological vision systems, have overcome bottlenecks of the von Neumann architecture. The innovation and integration of neuromorphic hardware systems represent a pivotal challenge for advancing the iteration of artificial intelligence. Accordingly, a novel optoelectronic reconfigurable neuromorphic transistor (ORNT) is designed to integrate three functions, enabling the perception, computation, and storage of optical information in a manner analogous to visual nervous systems. Based on the electrode-inserted graphene/VO₂ nanoparticles heterostructure and photovoltaic effect, the ORNT demonstrates broadband self-powered responsiveness from the ultraviolet to near-infrared (365-940 nm). Leveraging the photogating effect and the photoinduced phase transition in VO₂, the differentiated electrode design enables wide-electrode ORNTs to exhibit synaptic behavior under bias voltages, whereas narrow-electrode ORNTs demonstrate data storage capability and multistage photomodulation. Furthermore, an integrated optical communication and processing-in-memory system is developed, achieving a full-process demonstration from optical perception to computation and storage. Overall, the ORNTs introduced in this work provide an innovative strategy for optimizing the hardware resource allocation of chips and enhancing the adaptability and scalability of systems.