Artificial synapses based on boron ions-modulated transistors for neuromorphic applications

Abstract

The development of large-scale manufacturing and cost-saving artificial synaptic devices is of great significance for the realization of hardware neural networks. In this work, the In2O3 synaptic transistors with boron ion-doped ZrOx as the dielectric layer were fabricated by the solution process. The large counterclockwise hysteresis indicates that the nonvolatile memory capacity of the synaptic transistors was improved by boron ions doping. Typical synaptic functions including excitatory postsynaptic current, paired pulse facilitation, and high-pass filtering properties were simulated by applying gate bias. The multilevel channel conductance can be modulated by successive positive and negative pulses. Additionally, small nonlinearities were shown in the potentiation and depression processes, and high accuracies of 92.0% (small digits) and 93.8% (large digits) were obtained in digital recognition training. This study indicates that oxide-gated synaptic transistors based on the boron ions doping strategy show great potential for neuromorphic applications.