Direct-Write Defects for Bias-Free Two-Dimensional Optical Synapses Towards In-Sensor Motion Detection

Abstract Optoelectronic synaptic devices, particularly those based on two-dimensional (2D) transition metal dichalcogenides (TMDs), emulate biological synaptic functions, promising in neuromorphic computing. However, most TMD-based synapses rely on external gate or bias control, leading to complicated device architecture and high-power consumption. Here, we demonstrate a gate-free and bias-free optical synapses, based on monolayer molybdenum disulfide (MoS2) homojunction with its counterpart made of spatially controlled sulfur (S) vacancies, obtained through laser patterning. By precisely controlling the defect density from 0.15 nm–2 to 0.78 nm–2, we effectively modify the synaptic memory from a short retention time (∼125 s) to a long retention time (>450 s). We further demonstrate neuromorphic computing from image denoising and recognition to motion trajectory prediction for autonomous driving, highlighting the device’s applicability in on-chip visual preprocessing. This work shows great promise for neuromorphic computing by offering a scalable platform for energy-efficient optoelectronic neuromorphic technology.