Low-Voltage Solution-Processed Artificial Optoelectronic Hybrid-Integrated Neuron Based on 2d Mxene for Multi-Task Spiking Neural Network

Abstract

Incorporating optoelectronic integrated capability into artificial neurons can offer critical benefits of tunable device properties, diverse functions, and efficient computing capacity for artificial intelligent system. However, current reports are mostly focused on artificial neurons using an electric-driving mono-mode, while a facile and efficient approach to integrate electrical and optical signals is still lacking. Herein, a multifunctional optoelectronic hybrid-integrated neuron based on Ag nanoparticles-decorated MXene is proposed to achieve optoelectronic spatiotemporal information integration with low operating voltage of 0.93 V and high on/off ratio of 103, which are superior to those of majority of artificial neurons. An integrated visual perception system is developed by integrating artificial synapses, artificial optoelectronic neuron and robotic hand to emulate human conditional response. By integrating the optical sensory signals and electrical training signals, the response time of the system is significantly reduced. Finally, benefiting from the ability of spatiotemporal information integration, a multi-task pattern recognition in the spiking neural network composed of artificial synapses and neurons is completed, which can simultaneously recognize the digit patterns and rotation angles. Hence, this work exhibits the superiority in sensory and recognition tasks, which can pave the way for future application in neuromorphic circuits.

Keywords: artificial neuron, integrate-and-fire, spatiotemporal integration, multi-task, spiking neural network