A hardware spiking neural network system

Abstract

The invention discloses a hardware pulse neural network system, comprising: the input node layer and the unsupervised learning layer are connected through a synaptic connection unit through a neuron full connection mode, and the unsupervised learning layer and the supervised learning layer are connected by a neuron full connection The method is connected through another synaptic connection unit; the input node layer realizes the information input under different encoding methods, the unsupervised learning layer adopts the unsupervised learning method, and the supervised learning layer adopts the supervised learning method; the synaptic connection unit is composed of electronic The synaptic device is implemented so that the synaptic connection unit has pulse timing-dependent plasticity STDP, and the synaptic array unit receives the stimulus signal carrying information from the previous layer of neurons as a presynaptic pulse, combined with the actions excited by the subsequent layer of neurons The potential pulse acts as a post-synaptic pulse, and the time difference between the pre-synaptic pulse and the post-synaptic pulse determines the amount of synaptic weight adjustment of the synaptic connection unit. The neural network system provided by the invention has wide application value.

Description

technical field [0001] The present invention relates to the technical field of artificial neural network, and more specifically, relates to a hardware impulse neural network system. Background technique [0002] In the context of the era of big data, the traditional data processing method in which memory and processor are separated has brought about the von Neumann bottleneck problem, and this processing method has gradually been unable to meet people's growing data processing needs. However, the information activities of the human brain nervous system have the characteristics of large-scale parallelism, distributed storage and processing, self-organization, self-adaptation and self-learning, and there is no obvious boundary between data storage and processing. Theories, methods, technologies and application systems that expand and expand human intelligence, as well as the simulation of the information process of human consciousness and thinking, have great potential in the ...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the defects of the prior art, the purpose of the present invention is to combine the supervised and unsupervised learning methods, as well as the brain-like mechanism of side inhibition and pulse synchronous distribution, to solve the problem of robustness in the existing network training methods of the spiking neural network. The performance is poor, the recognition rate is low, and the STDP characteristics of various types of electronic synaptic devices cannot be satisfied at the same time. The technical problems of hardware application

Images