Research On Routing Arbitration Policy And Mapping Algorithm Of Spiking Neural Network Hardware System

In order to research and study biological treatment information and learning mechanisms in depth,Artificial Neural Networks(ANNs)was proposed.Spiking Neural Networks(SNNs),as a third-generation ANN,have an electric potential spiking trigger mechanism similar to that of biological neurons.This mechanism makes SNN compared with the traditional ANN has stronger computing power and is suitable for processing complex time-space information.Given its good biological properties and strong computational power,a research goal is to establish brain-like systems that mimic the key information processing mechanisms of the mammalian brain.However,the simulation is performed by using a conventional software calculation,the execution speed is slow,and the system has poor scalability.However,common parallel computing methods,such as computing clusters,have disadvantages such as high power consumption.Therefore,in order to overcome the problems existing in the implementation of SNNs by existing software or hardware methods,the new full-custom hardware architecture needs to be studied.Based on the current research on hardware implementation of SNNs,The research and innovation of this paper mainly focuses on the implementation of SNN hardware systems by mainly uses the Networks-on-Chip(NoC)architecture and intelligent optimization algorithms,including Particle Swarm Optimization(PSO),Genetic Algorithm(GA)and Immune Algorithm(IA).In order to achieve an efficient SNN hardware interconnection system,the main works are:1.This paper first introduces the research background of the topic,including the origin of the topic and the main research content,and then gives an innovation point of this paper.The basic knowledge of SNN is then described,and the existing large-scale SNN implementation scheme is summarized and analyzed.At the same time,this paper briefly introduces the background knowledge involved in the implementation of hardware systems,namely,the NoC,intelligent optimization algorithms.2.This paper considers the characteristics of SNN,that is,SNN has different transmission frequencies between different neurons.Based on the structural system of NoC,a dynamic priority arbitration policy based on the frequency of spiking transmission is designed.This policy can reduce the average delay of the high-frequency path and the risk of system packet loss,improve the system's working stability,and solve the traffic load imbalance problem of SNN spiking transmission.At the same time,the program uses a two-dimensional mesh NoC system to realize the interconnection of neurons.3.In order to reduce the energy and delay of the SNN hardware system in the process of implementing the SNN hardware system by using NoC architecture,this paper designs the SNN mapping algorithm based on the intelligent optimization algorithms such as PSO,GA,and IA.Experimental results show that the SNN mapping algorithm designed in this paper can effectively reduce the energy and delay of the SNN hardware system.