Impact of Free Layer Thickness and Damping Factor Variation on the Performance of Spin Orbit Torque Neuron-based of Spiking Neural Networks

Performance of spiking neural network (SNN) with spin-orbit torque based in-plane magnetic tunnel junctions as neurons have been evaluated on the MNIST dataset in light of the variations of the free layer (FL) thickness and damping factor. The variations, modeled through stochastic Landau–Lifshitz–Gilbert equations, reveal that for 10% variation of FL thickness and damping, network accuracy drops by 16% and 10% respectively. The loss of network accuracy is traced back to the variations of the output layer neurons of the SNN.