面向高性能计算平台的脑科学仿真技术研究

Brain network simulation technology on high-performance computing platform

近年来,脑科学研究成为了热点研究领域,欧盟"人脑计划"、美国"尖端创新神经技术脑研究计划"、日本脑计划的相继提出标志着脑科学研究进入了研究高峰期.高性能计算机经过数十年的发展,计算性能的提升日新月异,应用的范围越来越广.本文结合当下脑科学研究热点生物神经网络以及高性能计算平台TH-1A(天河一号)的计算优势,在TH-1A上部署了神经网络模拟软件NEST,对用NEST构建的神经网络进行了内存占用和可扩展性分析,测试了单计算节点最大仿真规模,发现了NEST软件自身的一些瓶颈,并提出了改进的想法,进而实现了对大规模生物神经网络面向高性能计算平台的仿真研究工作.

Brain science is becoming an ever more active research area in recent years.The launch of human brain project（HBP）of European Union,brain research through advancing innovative Neurotechnologies（BRAIN）of United States,and the brain project of Japan indicate that the research of brain science is reaching apeak.On the other hand,after decades of developments,computational power of super-computer has been improved rapidly,and its application fields are ever-increasing.In this work,we implemented a platform for simulating large-size biological neural networks on the high-performance computer Tianhe No.1（TH-1A）.We employed the popular simulation software NEST on TH-1A,analyzed its memory usage and scalability,identified computation bottlenecks,and proposed solutions for future improvements.