Reinforcement learning to reduce failures in SOT-MRAM switching

Abstract

A reinforcement learning strategy is applied to find a reliable switching scheme for deterministic switching of a perpendicularly magnetized spin-orbit torque magnetoresistive memory cell. Current pulses sent along orthogonal metal wires allow the field-free reversal of the magnetization. The current pulses are optimized such that reliable switching can be achieved over a wide range of material parameters. Micromagnetic simulations confirm the reliability of the presented approach.