A Variability-Aware Behavioral Model of Monolayer MoS2 RRAM for Tunable Stochastic Sources

Abstract

Stochastic switching in resistive random-access memories (RRAMs), while presenting challenges in digital memory applications, can be leveraged beyond von Neumann's stochastic computing and hardware security applications. In this regard, it is crucial to identify and model RRAMs where microscopic stochastic events can enable sizeable and tunable variability in macroscopic device characteristics. In this regard, chalcogen vacancy-mediated multifilamentary switching consisting of a multitude of hotspots in monolayer transition metal dichalcogenide (TMDCs) RRAMs can be promising candidates for high-quality, tunable stochastic sources. In this work, an efficient physics-based model is developed to capture the behavior of stochastic switching in monolayer MoS₂ RRAMs. The microscopic origin of stochasticity, arising from clusters of sulfur vacancies transforming into metallic hotspots, is modeled using the kinetic Monte Carlo method. The rate equations designed to capture the physics of abrupt SET and gradual RESET processes provide an excellent fit to experimental data, allowing to extract key material parameters. The calibrated macroscopic model is then employed to explore multiple non-volatile resistance states in the gradual RESET process, area scalability trends and cycle-to-cycle C2C variability over 100k cycles. Furthermore, the statistical distribution of HRS and LRS variability is modeled and large tunability of the distribution is demonstrated using stop voltage in RESET. Finally, it is demonstrated that these devices are excellent candidates as bit stream generators for stochastic computing applications with accuracy values comparable to an ideal source. It is envisioned that the work will induce significant interest in the deployment of 2D materials-based RRAMs for high-quality tunable stochastic sources.