Unraveling Origin of Stochasticity in Multi-Filamentary Memristor

Abstract

Recent advances in computing including security applications, Monte Carlo simulations, and probabilistic computing, have increased the demand for robust probabilistic elements. Ion-motion-mediated volatile memristors with threshold switching (TS) characteristics have emerged as promising physical entropy sources because of their stochastic conductive filament (CF) formation and rupture. However, optimizing a memristor as an entropy source requires a material system that actively promotes ion motion and the associated CF formation/rupture, along with a quantitative understanding of their coupled electrothermal behavior. In this study, by integrating a porous nanorods (NRs)-based oxide layer that enhances ion-motion pathways, we achieved rapid, device-centric digital and analog random outputs without the need for post-processing. Moreover, we directly visualized the stochastic dynamics of multiple CFs using scanning thermal microscopy (SThM) and verified our findings through electrothermal simulations, confirming the device's inherent randomness. Finally, a bimodal (digital and analog) true random number generator (TRNG) and a probabilistic computing platform demonstrated the versatility of TS memristors as tunable and robust sources of randomness for probability-oriented applications.