Research On Construction Of Simulation System For Neural Network In FPGA

Nervous system is a high-dimensional, strongly coupled system, with the characteristics of complex non-linear dynamics, and traditional computer softwares are difficult to reach the speed performance when large-scale neural network is simulated. The paper is committed to to achieve the modeling and analysis of neural network by the field programmable gate array (FPGA) technology.First, the hardware system is built. The system contains model library, control unit, parameter storage module, FPGA computing arrays and host computer and it provides an experimental platform for analysis of large-scale neural network, identification of parameter and control of neuron.Next, we model the Morris-Lecar neuron and network in DSP Builder, build the hardware system to analysis the effects of parameters on discharge and synchronization. It is found by comparison with the MATLAB simulation that the FPGA neuron and network meet the requirements of computational neuroscience. At the same time there is a certain speed advantage, and the larger the network, the more obvious advantages.Then, using the same modeling approach, the more complex hippocampal CA3 pyramidal neuron is modeled and analyzed by the hardware system. The results show that the hippocampal FPGA neuron meets the requirements of computational neuroscience.Finally, based on FPGA technology and adaptive synchronization theory, the parameter identification of normal neuron and the control of non-normal neuron are realized. It is shown that the hardware system can accurately identify the parameters of neuron, and the non-normal neuron can reach the same output as normal neuron eventually when the control is applied.The results show that the hardware system can achieve the simulation of neural networks, identification of parameters and control, providing an accurate, rapid experiment platform.