The computing paradigm is shifting from traditional deterministic models to probabilistic models inspired by the brain. This shift aims to address complex problems in fields such as artificial intelligence, optimization, and security with greater energy efficiency. However, achieving efficient sources of randomness at the hardware level remains a significant challenge. Memristors exhibit intrinsic nanoscale resistive switching randomness, offering a physical foundation for probabilistic computing and hardware security. Traditionally, this variability has been viewed as a reliability issue in memory applications; however, the emerging concept of “stochasticity-by-design” is transforming it into an exploitable functional property. This review provides a comprehensive, materials-focused perspective on this rapidly evolving field. We analyze the physical origins of randomness, namely, “entropy sources”, in different memristive material systems, covering conductive bridges (ECMs), valence state changes (VCMs), Mott transitions, and proton transport mechanisms. Building on this foundation, we systematically outline a toolbox of “control knobs” for precisely tuning these random properties, spanning material, device, and operational levels. We then establish clear performance benchmarks for two key applications: probabilistic computing and hardware security, directly linking the device’s physical properties to application requirements. Finally, we explore the challenges of moving from single devices to large-scale system integration, discuss the trade-off between randomness and determinism, and highlight recent advances in multifunctional devices with tunable randomness. Here, we provide researchers with a unified framework from fundamental physics to system applications, offering a roadmap for optimizing random memristive materials for next-generation security hardware.
Jiang et al. (Thu,) studied this question.