Modeling Defect Dynamics in MoS₂ Monolayers for Resistive Switching Devices Using Machine Learning Potentials and Kinetic Monte Carlo Simulations

Defective MoS₂ monolayers are promising building blocks for next-generation low-power, nonvolatile memory devices due to their enhanced electrostatic control and unique electronic properties. Site-specific irradiation can create nanometer-scale defect-rich regions in MoS₂ monolayers, where the reversible drift of sulfur vacancies under electric fields modulates resistance, enabling memristive functionality. Understanding defect dynamics at the atomic scale is crucial for optimizing device performance. However, the limited scalability of ab initio methods confines modeling to small scales, rendering them inadequate for simulating large-scale phenomena such as interacting defect migration. To overcome this limitation, we developed an equivariant neural network potential to drive kinetic Monte Carlo simulations, enabling large-scale modeling of memristive processes with DFT-like accuracy at reduced computational cost. Our simulations model vacancy migration, agglomeration, and dissolution in MoS₂ monolayers with varying sulfur defect concentrations and distributions under external electric fields. We evaluate their impact on the device's low and high resistance states. Moreover, we examine defect relocations during cycling, yielding insights into the reliability and endurance of lateral defect-based MoS₂ memristors within neuromorphic architectures.