FPGA-Based Design And Implementation Of Large-Scale STN-GPE Network

Parkinson’s disease(PD)is a kind of common neurodegenerative disease,which pathogenesis and treatment mechanism are still unclear.Abnormal oscillatory activity in the beta band(13-35 Hz)of the local field potential in the subthalamic nucleus(STN)-external globus pallidus(GPe)network is associated with parkinsonian motor symptoms.However,deep brain stimulation for STN,a commonly used target of stimulation,can effectively reduce the oscillatory activity of STN-GPe network and relieve movement disorders related to PD,but the underlying mechanism is unclear.Therefore,the high-speed numerical simulation of STNGPe network is of great significance for the exploration of the underlying mechanism of PD.Based on Field Programmable Gate Array(FPGA),this paper implements the real-time simulation of a large-scale STN-GPe network containing 512 single-compartment HodgkinHuxley type neurons,and analyzes the network physiological activities.Firstly,the model and parameter values of STN-GPe network are introduced.This paper systematically analyzes the FPGA simulation platform from the aspects of hardware system,data communication and design framework,etc.,and selects the hardware expansion scheme for the realization of large-scale STN-GPe network,and determines the overall development process of FPGA.Secondly,the FPGA design of single neuron is realized.In order to save resources on the chip,multiplier substitution,fixed-point operation,nonlinear function approximation and function re-integration are adopted,so that the design makes full use of various hardware resources on the FPGA(such as memory resources and embedded DSP blocks)to indirectly reduce the use of the total logic elements.The design idea of finite state machine is proposed,and the detailed digital structure FPGA implementation results of single neuron model are compared with MATLAB software simulation results to verify the accuracy of single neuron model FPGA implementation.Finally,the network topology is expanded,and the STN-GPe network with a size of 512 neurons is implemented using module reuse that increases the network size by 64 times at the expense of simulation time.According to FPGA simulation results,the network can accurately simulate firing activities in normal and PD states.The correlation coefficient between the neuron firing waveform of the FPGA platform and the software simulation waveform is 0.9756.Under the same physiological time,the FPGA platform simulation speed is 75 times that of the Intel Core i7-8700 K 3.70 GHz CPU 32 GB RAM computer simulation speed,which meets the requirements of real-time simulation.In addition,the established network platform was successfully applied to the analysis of the effects of simulated temporal pattern of deep brain stimulation waveforms on network firing activities.This paper provides an effective method for FPGA implementation of large-scale spiking neural network,and provides a more physiological simulation platform for the study of PD pathogenesis and DBS.