Research On The Interconnection Network Of Spiking Neuron Networks Simulation Multi-Core Computer

As a third-generation new artificial network model,spiking neural network has not only developed rapidly in the fields of pattern recognition and data processing,but also promoted the development of brain-like computer-related projects based on spiking neural network.With the maturity of chip integration technology and the decline of processor cost,it is no longer a dream to realize a million-core system architecture for brain-like platforms.The network communication has a big problem:how to support massive spiking packets.Efficient transmission,This topic is based on the spiking neural network to simulate the interconnection network of multi-core computers.The main research contents and conclusions are as follows:Firstly,the theory and technology of multi-core computer are simulated by spiking neural network.This paper defines the format of three kinds of spiking data packets and the corresponding coding methods,which are point-to-point spiking data packets,multicast spiking data packets,cluster-based multicast data packets.A complete spiking data packet coding scheme is designed for cluster multicast data packets to realize flexible and variable spiking data packet coding,which greatly reduces the number of routing entries in the routing table.Secondly,this paper focuses on the design of the underlying communication network model of spiking neural network simulation multi-core computer,focusing on the transmission of spiking data packets.According to the mapping relationship of neurons in spiking neural network,based on the idea of heuristic AC algorithm and GA algorithm,an improved GA routing algorithm and AC routing algorithm are proposed.The shortest path is solved by the two algorithms.And recurring,innovative classification of ant colonies,and adaptive parameter design to ensure that the algorithm converges to the optimal solution.The experimental results show that compared with the traditional AC algorithm and GA algorithm,the improved routing algorithm not only shortens the path length of route addressing,but also shortens the execution time of route addressing.Finally,based on the spiking neural network to simulate the network topology of multi-core computers,a new routing addressing architecture is proposed,and the basic format of the dedicated routing table is designed.According to the 6-bit output port coding of the routing node,the process of a complete set of routing address of the spiking data packet is completed,and the corresponding communication fault tolerance mechanism is proposed.The static and dynamic methods are used to solve the problem of the spiking data packet when the link is damaged.Obstacle transmission.The design of this dedicated communication network model not only supports peer-to-peer data communication,but also supports efficient multicast communication,and successfully solves the problem of massive data communication in spiking neural network simulating multi-core computers.