Self-Resetting Synaptic Device toward Low-Power Neuromorphic Vision Systems

Abstract

Integrating photodetection and optoelectronic synaptic functions into a single device is critical for neuromorphic vision systems. However, such integration typically relies on continuous external biasing and electrical reset pulses for weight regulation, incurring significant power overhead. Here, we present a dual-interface modulation MoOx/MoS2/intrinsic-Si heterojunction device. At the MoOx/MoS2 interface, the device forms a self-resetting optoelectronic synapse, where the high-resistance intrinsic-Si layer and ultrathin MoOx enable slow growth and spontaneous rupture of oxygen-vacancy filaments without electrical reset pulses. Furthermore, the MoS2/Si interface functions as a self-powered photodetector achieving μs-scale response times, enabled by a single-sided depletion region that drives efficient carrier separation. Leveraging this dual-mode capability, the device demonstrates key synaptic behaviors in synaptic mode, while exhibiting logic gate functionality and imaging capabilities in photodetection mode. When integrated with an artificial neural network, the device enables handwritten digit classification and vehicle tracking under both visible and near-infrared wavelengths. This work provides an innovative solution for developing next-generation low-power neuromorphic vision systems.