WS 2 nanosheet-based memristors for neuromorphic pattern recognition

Abstract Although two-dimensional transition-metal dichalcogenides offer a promising platform for energy-efficient neuromorphic computing, achieving robust synaptic fidelity remains challenging. In this study, we report a vertical Al/WS 2 nanosheet/ITO memristor that exhibits reliable analog resistive switching. Conductance modulation is governed by the dynamic evolution of conductive filaments, which are physically mediated by the field-driven migration of sulfur vacancies. The device demonstrated highly stable and linear long-term potentiation and long-term depression, which are essential for effective weight updates in neural networks. By leveraging these experimental synaptic characteristics, system-level simulations showed that a conventional artificial neural network achieved 97.14% accuracy on the modified MNIST handwritten dataset, whereas a convolutional neural network attained 93.27% accuracy on the German Traffic Sign Recognition Benchmark. These results underscore the potential of defect-engineered WS 2 memristors as high-performance artificial synapses for next-generation pattern recognition hardware.