Spin–Orbit Torque MRAM: Fundamentals, Device Engineering, and Applications in Probabilistic Computing

ABSTRACT Spin–orbit torque magnetoresistive random‐access memory (SOT‐MRAM) has emerged as one of the most promising next‐generation non‐volatile memories, overcoming the endurance and energy limitations of conventional memory technologies. By harnessing the spin–orbit torque effect, SOT‐MRAM offers fast switching, high endurance, and separation of read and write paths, making it suitable for both memory and computing applications. This review summarizes the fundamental writing mechanisms of SOT‐MRAM, the progress in materials and device engineering, and recent advances toward wafer‐scale integration. Beyond its role as embedded memory or cache, SOT‐MRAM provides unique opportunities in unconventional computing. In particular, SOT‐MTJs can serve as probabilistic bits (p‐bits), enabling hardware implementations of probabilistic inference, Boltzmann machines, and Ising models. We highlight both the opportunities and challenges of leveraging the intrinsic stochasticity of SOT‐MRAM, which originates from thermal fluctuations rather than uncontrolled device variability, for large‐scale probabilistic computing. We also discuss future research directions for bridging device‐level innovations with system‐level applications.