Voltage-Gated Domain Wall Magnetic Tunnel Junction for Neuromorphic Computing Applications

We propose a novel spin-orbit torque (SOT) driven and voltage-gated domain wall motion (DWM)-based MTJ device and its application in neuromorphic computing. Using the voltage-controlled gating effect on the DWM, the sneak path current problem in the synaptic crossbar implementations is mitigated. The proposed device scheme is more robust against synapse writing errors, and device behavior shows good linear characteristics. The voltage-controlled DW gating performance of the device in the crossbar array is quantified in-depth. The device linearity dependence on material parameters such as chiral Dzyaloshinskii Moriya interaction (DMI) is evaluated at different temperatures for estimation of harsh environment performance; furthermore, the proposed device is used as the synapse in a spiking neural network (SNN). The DW-MTJ conductance was mapped to the weights of the SNN. When trained and tested on the CIFAR-10 dataset, the architecture based on the DW-MTJ synapse achieved an accuracy of around 85%.