Design Of Spiking Neuron Networks Based On FPGA

With the continuous development of brain science,cognitive science and neuroscience,it has become possible to observe the activity of brain neuron networks under different cognitive tasks.In recent years,the working mechanism of the brain has been gradually excavated,inspired by this,developing brain-liked intelligence and neuromorphic computing has become a research hotspot in the field of computational science.The manifestations of neuromorphic computing is a chip or a computer,it aims to achieve a smarter computing system by imitating the structure and working principle of the biological neural system,then bring a revolutionary breakthrough and development in artificial intelligence,computing and storage and other fields.How to implement biological neural networks on hardware is one of the most challenging technical problems in the field of computational science.Spiking Neuron Networks encode the information into pulse time and can get more information,which are suitable for dealing with a variety of neural signals problems.Based on the information transfer mechanism of neuron,with Field Programmable Gate Array(FPGA)as the hardware platform,we completed the design of Spiking Neuron Network circuit model by DSP Builder and the simulation analysis of neuronal action potential.The main work of this paper is as follows:Firstly,based on the Hodgkin-Huxley(H-H)neuron model,we completed the discretization of H-H neuron mathematical model.We built a H-H neuron circuit model that based on discrete time by DSP Builder and a H-H neuron circuit model that based on continuous time by Simulink.Then we simulated the action potential waveform of the two circuit models in the same current stimulation,and verified the feasibility of the modeling method.Secondly,according to the characteristics of different types of synapses in neuronal synapse theory,we choose chemical synapses to realize the coupling between neurons,and then combining with the mathematical model of chemical synapses we built a chemical synapse circuit model by DSP Builder.Finally,combining the topological structure of neural networks we completed the design of an annular Spiking Neuron Network circuit model which contains eight H-H neurons and the design of a Feed-Forward Spiking Neuron Network circuit model which contains six H-H neurons.Then we simulated the two kinds of Spiking Neuron Network model circuit model,obtained the action potential waveform of each neuron in the neural network,and verified the feasibility of hardware implementation of Spiking Neuron Networks.In this paper,the design and simulation of neural network circuit model provides an idea for the simulation and hardware implementation of large-scale neural networks.