Reinforcement Learning to Reduce Failures in SOT-MRAM Switching

Abstract

We demonstrate the use of reinforcement learning for achieving efficient switching schemes for a field-free operation of spin-orbit torque magnetoresistive random access memory cells. A cell is switched purely electrically by applying two orthogonal current pulses. It is shown that using a reinforcement learning approach, a neural network model can be trained on a fixed material parameter set for finding optimal switching pulse sequences. This model is not only suitable to switch a memory cell in the presence of thermal fluctuations, but also for varied cell material parameters.