High-performance artificial synapse based on oxidized Fe3GeTe2 with ultra-low energy consumption

Abstract

It holds significant importance to modulate electronic characteristics and enable neuromorphic computing by environmental oxidation. In this study, various synaptic plasticity and Bienenstock-Cooper-Munro (BCM) learning rules can be imitated in few layer oxidized Fe₃GeTe₂ (FGT) device. Moreover, this oxidized FGT synaptic device exhibits ultra low power consumption (2 fJ), near ideal linearity, and stability during 1000 wt updates, providing a feasible solution for improving training accuracy in hardware-based neural networks. Finally, based on the short-term plasticity and history-dependent plasticity of the FGT device, achieving lower training costs and faster reservoir computing. It serves as an ideal component for constructing reservoir computing systems. These demonstrate that oxidized FGT synaptic devices are attractive for improving neuromorphic computing.