Light-modulated synaptic response of planar WS2–PVA hybrid memristors for artificial synapses and quantization-aware neural networks

Two-dimensional-material-based artificial synapses that combine stable electrical switching with optical tunability are promising for neuromorphic systems. Here, a planar Cr/WS 2 –PVA/Cr memristive device was fabricated using liquid-phase-exfoliated WS 2 sheets embedded in a PVA matrix. The device exhibited stable analog bipolar resistive switching with self-rectifying behavior over 300 cycles. Its switching characteristics were associated with defect-assisted charge trapping, Poole–Frenkel-type emission, and space-charge-assisted transport within the WS 2 –PVA hybrid active layer. Sulfur-vacancy-related states and polymer-assisted trapping centers contributed to the gradual modulation of conductance. Under 425 nm illumination, extending the exposure time from 30 to 120 s increased the conductance from 1.7 to 5.4 μS, facilitating light-assisted learning and relaxation-based forgetting behavior after light removal. Hardware-aware ResNet-20 simulations using the measured conductance states achieved a CIFAR-10 accuracy of 81.91 ± 0.29% after 120 s of illumination, averaged over five independent seeds. The device-aware models maintained classification accuracies above 81% under both dark and illuminated conditions, demonstrating the feasibility of integrating experimentally obtained WS 2 –PVA conductance states into neuromorphic inference. These findings highlight the potential of WS 2 –PVA hybrid memristors for electrically and optically tunable artificial synapses and neuromorphic inference.