The Neuron Circuit with Property of Sensitization or Habituation Based on Threshold Switch Devices

摘要

Brain-inspired computing is an efficient method to implement artificial neural network, which is the foundation of artificial intelligence. In the brain-inspired computing, neuron circuits play an important role in information processing and computing. But in recent researches, most efforts have been paid to improve the calculating performance with the cost of adding more functions. Nonetheless, the biological characteristics of neuron have attracted less attention. This paper aims to emulate two important biological characteristics (habituation and sensitization) of neuron employed CMOS circuits based on threshold switch devices. In specific, we fabricate and test an Ag/SiTe/TiN threshold switch device with more than 100 times switching ratio. Based on this device, we design a habitual neuron circuit and a sensitive neuron circuit. These two circuit can achieve nonassociative learning. It can be used in image processing and improve efficiency and accuracy in feature extraction and image recognition.