Approaching the Area of Neuromorphic Computing Circuit and System Design

The traditional von Neumann architecture has met limitations in both computation and energy efficiency. Researchers' attention has been diverted to neuromorphic computing with the progression of neuroscience. With the inspiration of mammal neural systems, neuromorphic chips are designed and fabricated. This paper will introduce the basic concept and elements of neuromorphic computing circuit design, such as spiking neurons and encoders. Spiking encoders convert analog signals to spikes and lead to high power efficiency while maintaining low hardware implementation costs. Spiking neural networks that utilize the delay-feedback property have been designed and fabricated. One of them is the delay-feedback reservoir (DFR) network that is more computational efficient than the conventional recurrent neural network (RNN). The others are hybrid neural networks (HNN) that combine DFR with other neural networks like multilayer perceptron (MLP) and convolutional neural network (CNN). Finally, the measurement performance for different applications of these neural networks (NNs) will also be demonstrated.