Neural Network Analysis And Multi-FPGA Implementation

The nervous system is constituted a complex network of a large number of neurons coupled. Network information transmission between neurons through synaptic currents and changes in membrane voltage implemented. Computational neuroscience research and simulate the nervous system in the computer simulation and mathematical analysis levels. Researchers can explore the nervous system from a computational point of view, thus creating the nervous system. But the traditional software simulation method has lower computational performance limitations of the study efficiency. Specific integrated circuit chip is a hardware category, with parallel computing capabilities, with its built circuit model neuron system can significantly computational efficiency pricey neural network. Currently, there are many domestic and foreign researchers of Applied Physiology characteristics of the neural circuit model of signal transduction in the nervous system. In this paper, we use multi-chip field-programmable gate array(FPGA) chip to build a model of neural information transmission circuit, and through the neurons and neural network analysis to prove the validity of the model.Firstly, we propose to build a multi-chip FPGA simulation platform neurons. We study the topology of the connection between the multi-chip-chip data transfer clock synchronization, configuration write model code and other issues; We explore the physiological characteristics of data communication problems neuronal conduction circuit model generated, digital to analog conversion proposed Ethernet packet capture, USB connection real-time communication LABVIEW built PC. We achieve a common graduate staff by 4 FPGA chip built-oriented and easy to operate, feature-rich integrated hardware and software platform for FPGA simulation.Then, we study single neurons and ML implementation framework FHN model based on small-world neural network on the set up of the FPGA hardware simulation platform. We studied the neuron model, the general FPGA hardware design techniques and design criteria synaptic model and network connection mode. This method ensures physiological energy in real time scale, to complete the simulation and analysis of large-scale neural network.Finally, Design 3-layer feedforward networks based on small world network and three different stimulation signals on multiple FPGA hardware simulation platform. By analyzing the characteristics of excitatory synaptic coupling connection, verify the feedforward signal transmission network synchronization process. At the same time, we calculated the root mean square error(RMSE) of word, which evaluate the performance of an FPGA-based hardware to build neural networks. We studied how to take full advantage of multi-chip FPGA to implement more effective neurological Network, which make a preliminary inquiry for future ultra-large-scale neural nuclei.