Perpendicular-spin-transfer-torque magnetic-tunnel-junction neuron for spiking neural networks depending on the nanoscale grain size of the MgO tunnelling barrier

Abstract

Unlike conventional neuromorphic chips fabricated with C-MOSFETs and capacitors, those utilizing p-STT MTJ neuron devices can achieve fast switching (on the order of several tens of nanoseconds) and extremely low power consumption (<0.2 pJ per spike). A p-STT MTJ neuron with a sensing circuit, which is composed of one p-STT MTJ neuron device, seven n-MOSFETs, three p-MOSFETs, and one reference resistor, was constructed in this study and presented integrate-and-fire characteristics for use in spiking neural networks. In particular, the difference in resistance between the no-spiking input and after the implementation of integration-and-fire was found to be principally determined by the average nanoscale grain size (i.e., 0.418 to 1.141 nm) and face-centered-cubic crystallinity of the MgO tunnelling barrier of the p-STT MTJ neuron devices. Therefore, a larger grain size and better crystallinity led to a larger resistance difference in these devices. MNIST pattern recognition tests (achieving a testing accuracy of 90.34%) using the p-STT MTJ neurons were conducted for demonstrating a spiking neural network. This journal is