Robust optoelectronic dual-mode memristor enabled by ZnO/MoS<inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline" id="TIMEQ1"><mml:msub><mml:mrow/><mml:mn>2</mml:mn></mml:msub></mml:math></inline-formula> heterojunction for synaptic bionics and in-memory computing

The current von Neumann computing architecture based on traditional complementary metal oxide semiconductor (CMOS) transistors, which separates memory from process units, faces severe challenges of insufficient computing capability and high power consumption in artificial intelligence. To address these key bottlenecks, memristor-enabled neuromorphic computing has emerged as a promising solution. Herein, an optoelectronic dual-mode memristor based on a ZnO/MoS$_2$-heterojunction featuring tunable characteristics through interface engineering has been proposed. By applying different types of electrical and optical signals, the device successfully emulates synaptic plasticity including paired-pulse facilitation, long-term potentiation and long-term depression. Besides, the MoS$_2$ insertion layer significantly enhances the stability and uniformity of the memristor, achieving 580-cycle resistive switching with over 10$^4$ s retention time and low coefficient of variation of 18.30% for the high resistance state and 7.26% for the low resistance state, indicating excellent operational reliability. Furthermore, we employed SPICE simulations to implement memristor-based AND, OR, and NOT logic circuits. At the same time, a convolutional neural network for handwritten digit recognition is constructed, reaching 99% recognition accuracy. In addition, a biomimetic artificial visual system is implemented using a 3$\times$3 ZnO/MoS$_2$ memristor array. Therefore, the optoelectronic memristor developed in this work provides a viable solution to promote the development of next-generation in-memory and neuromorphic computing.