Research On The Dynamic Functional Neuron Network NoC Architecture

Computational neuroscience is one of the important methods of neuroscience research,which studies the characteristics and structure of neuron networks by establishing biophysical models.In order to meet the needs of large-scale neuron network simulation,hardware simulation with field programmable gate array(FPGA)has become one of the main directions of computational neuroscience research.Due to the complexity of the characteristics and structure of the neuron network,the traditional on-chip simulation model of the neuron network cannot achieve its dynamic characteristics and functions.This paper proposes to introduce the on-chip network(NoC)architecture into the research of neuron network,which can realize the complex communication function of the neuron network by means of route addressing,and realize the dynamic simulation effect of the neuron network through the internal state machine control of FPGA.This paper proposes a Network-on-Chip(NoC)architecture to realize the dynamic functional neuron network with the controlling of the state machine based on FPGA.The most suitable NoC topology and routing algorithm are chosen according to the internal structure and physiological function of neuron network,which can transform the neuron information transmission to the routing addressing process.In the architecture,single neurons are set as the IP cores,and the routers are added in NoC to find the target path.Thus the architecture can realize the dynamic functions of the neuron network with the help of the state machine.In this paper,an improved CORDIC algorithm is proposed to solve the nonlinear operations in neuron networks.Improved CORDIC algorithm extends the convergence domain of the traditional CORDIC algorithm by interval compression and approaches the nonlinear operations by 20 times pipeline iteration structure,which gives the CORDIC algorithm a larger convergence domain and a higher accuracy to implement the nonlinear mathematical model in neuron network.Based on NoC architecture,two kinds of functional neuron network models are implemented in this paper.First,a dynamically extensible feed-forward network based on time is realized by the NoC architecture.The system takes XBFT as the network topology to transform the static single-layer network structure into the dynamic threelayer feedforward network structure.Second,a retinal neuron network model based on illumination control is realized by the NoC architecture.The system combines NoC architecture with peripheral devices driven by FPGA to build a dynamic neuron simulation system that can communicate with external signals in real time.Taking two kinds of neuron network models as examples,this paper deeply studies the NoC architecture of the neuron network,which verifies the dynamic characteristics and portability of the NoC architecture and provides a new idea for the hardware simulation of the neuron network.