Atomistic Insights into Cycle-to-Cycle Variability in Au/MoS2/Au Atomristors through Multiscale Simulations

Two-dimensional (2D) transition metal dichalcogenide atomristors are promising candidates for next-generation nonvolatile memory and neuromorphic computing, but inherent cycle-to-cycle (C2C) variability poses a challenge to their practical deployment. Although many experimental and theoretical studies have explored these devices, the atomistic mechanisms responsible for this variability are still not well understood. Here, we employ a multiscale computational framework integrating reactive molecular dynamics (RMD), electro-thermal finite-element modeling (FEM), and quantum transport (QT) simulations to investigate switching in Au/MoS2/Au memristors and uncover the atomistic origin of variability. We show that SET is driven by electric-field-induced migration of partially charged Au atoms, forming a conductive filament, while RESET involves reverse-field-driven dissolution with incomplete MoS2 structural recovery. Electro-thermal coupling reveals that Joule heating lowers migration barriers, reducing switching fields and accelerating dynamics. Over repeated cycles, residual Au clusters, partially refilled voids, and lattice distortions persist, continuously modifying the energy landscape and lowering activation barriers for subsequent switching. As a result, C2C variability emerges from deterministic, history-dependent structural evolution rather than purely stochastic effects. After a few cycles, the resulting small energy separation between metastable states further enables thermally assisted and potentially volatile switching regimes. By linking atomistic structure with electrical conductivity and temperature, we establish a direct structure–energy–property relationship governing switching behavior. These findings provide a unified understanding of variability in 2D memristors and offer design guidelines for improving device stability.