多FPGA神经元网络仿真平台设计  

Design and analysis of multi-FPGA based simulation platform for neural network

作　　者：孙凡[1] 李会艳[1] 赵爱清[2] 

机构地区：[1]天津职业技术师范大学天津市信息传感与智能控制重点实验室,天津300222 [2]天津现代职业技术学院,天津300350

出　　处：《天津职业技术师范大学学报》2017年第2期8-12,共5页Journal of Tianjin University of Technology and Education

基　　金：国家自然科学基金资助项目(61374182)

摘　　要：为解决多片FPGA之间的多时钟同步问题,提出了多时钟硬件同步的思想。对多片FPGA芯片之间的时钟同步方法进行研究,实现了由4片FPGA芯片搭建的多FPGA神经元网络仿真平台。利用3层的FHN(Fitz Hugh-Nagumo)模型对多FPGA神经元网络仿真平台进行验证。结果表明:相对于市场现有的FPGA仿真平台,多FPGA神经元网络仿真平台具有计算速度快、可靠性强、配置灵活、易于扩展、适用于大规模神经网络仿真等优点,能够应用于各种神经元网络的仿真与分析。In order to solve the problem of multi-clock synchronization between multiple FPGA, this study proposes the idea of muhi-clock hardware synchronization, analyses the method of clock synchronization between multi-FPGA, and designs a simulation platform of muhi-FPGA based on neural network, which is set up by 4 FPGA chips. The simulation plat- form of muhi-FPGA neural network is verified by the three layer FitzHugh-Nagumo （FHN） model. The verification results show that. compared with the existing FPGA platfrom, the muhi-FPGA neural network simulation platform has the advantages of high computational efficiency, high reliability, flexible configuration, easy to expand, suitable fur large-scale neural network simulation and so on. The proposed design can be widely applied in fields such as the artificial intelligence or the dynamical characteristics investigation of the neural networks.

关 键 词：FPGA 时钟硬件同步 FHN 神经网络 

分 类 号：TN791[电子电信—电路与系统]