Dynamic $(\mathrm{Bi}_{\mathrm{x}}\mathrm{Sb}_{1-\mathrm{x}})_{2}\mathrm{Te}_{\mathrm{3}}$ Synaptic Devices with Programmable Spatio-Temporal Responses

Neuromorphic computing has emerged as a new computing paradigm to tackle the von Neumann bottleneck and enable a more energy-efficient processing of today's large-scale datasets, especially for data-intensive applications such as image and pattern recognition, speech recognition, and autonomous machines ¹ . Despite the staggering success of deep neural networks (DNNs), their need for labelled data and the energy intensive training process are limiting their widespread usage. Recently, spiking neural networks (SNNs), which resembles more closely to biological neural networks via their coupled spatio-temporal dynamics, have attracted intense research interests ^1,2 in hope of mimicking the superb energy efficiency of the human brain, which exceeds the petaflop mark in performance with a power consumption as low as 20 W.