Radio-frequency-modulated artificial synapses based on magnetic tunnel junctions with perpendicular magnetic anisotropy

Abstract

Magnetic-tunnel-junction- (MTJ) based spintronic devices have demonstrated significant potential in neuromorphic computing. Here, we report an artificial synapse, which can be modulated by rf signals directly based on the nanoscale MTJs with perpendicular magnetic anisotropy (PMA). To utilize multiple rf signals in parallel, we take an approach to change the resonance frequencies of MTJs by changing the PMA between the $\mathrm{Co}\text{−}\mathrm{Fe}\text{−}\mathrm{B}$ free layer and $\mathrm{Mg}\mathrm{O}$ barrier, which can expand the application range of rf signal processing. Moreover, we experimentally demonstrate that MTJs with PMA can serve as an rf synapse with adjustable positive and negative weights. We have achieved effective classification of rf signals with an accuracy exceeding 96% through experimental results as synaptic weights, comparable to that of equivalent software-based neural networks. This work may pave the way for the development of rf-oriented hardware artificial neural networks.