Stochastic artificial synapses based on nanoscale magnetic tunnel junction for neuromorphic applications

Abstract

Bio-inspired neuromorphic computing has aroused great interest due to its potential to realize on-chip learning with bio-plausibility and energy efficiency. Realizing spike-timing-dependent plasticity (STDP) in synaptic electronics is critical toward bio-inspired neuromorphic computing systems. Here, we report on stochastic artificial synapses based on nanoscale magnetic tunnel junctions that can implement STDP harnessing stochastic magnetization switching. We further demonstrate that both the magnitude and the temporal requirements for STDP can be modulated via engineering the pre- and post-synaptic voltage pulses. Moreover, based on arrays of binary magnetic synapses, unsupervised learning can be realized for neuromorphic computing tasks such as pattern recognition with great computing accuracy and efficiency. Our study suggests a potential route toward on-chip neuromorphic computing systems.