面向“达尔文”芯片的脉冲神经网平台设计与实现

发布时间：2018-02-14 23:33

本文关键词： 类脑计算脉冲神经网络 “达尔文”芯片仿真平台　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：基于冯诺依曼结构的计算机体系结构由于功耗与散热的问题面临着挑战,类脑计算因此成为研究的热点。类脑计算的基本方法是应用生物神经网络的概念来设计计算机系统,并针对特定的应用来降低功耗和提高性能。脉冲神经网络是通过模拟生物神经网络发展而来的,因此成为类脑计算的重点研究方向。目前类脑芯片在硬件设计上有了突破性的进展,"达尔文"类脑计算芯片便是浙江大学与杭州电子科技大学联合研发的一款基于脉冲神经网络的协处理器芯片。“达尔文”芯片需要具体的应用系统来进行验证和评估,而构建"达尔文"芯片的应用系统需要适合的脉冲神经网络实现。基于以上背景,本文设计并实现了面向达尔文芯片的脉冲神经网络平台,为类脑计算芯片的应用设计提供了一个可定制的、直观的仿真平台,并以此为基础为达尔文芯片实现实时演示系统。本文主要有以下贡献:1)设计并实现基于"达尔文"芯片的脉冲神经网络,实现了人工神经网络转化成脉.冲神经网络的算法。2)构建并实现了一个脉冲神经网络仿真平台,该平台能够通过可视化操作界面配置网络结构和参数,以实现自定义结构的脉冲神经网络,并能够对脉冲神经网络仿真的结果和过程进行可视化。3)以脉冲神经网络仿真平台为基础,设计并实现了一个基于"达尔文"芯片的意念控制虚拟篮球运动实时演示系统。
[Abstract]:The computer architecture based on von Neumann structure is facing a challenge due to the problem of power consumption and heat dissipation, so brain-like computing has become a hot topic. The basic method of brain-like computing is to design computer system by using the concept of biological neural network. Pulse neural networks are developed by simulating biological neural networks. At present, there has been a breakthrough in the hardware design of brain-like chips. "Darwin" is the basis of the joint research and development of Zhejiang University and Hangzhou University of Electronic Science and Technology. The "Darwin" chip needs a specific application system to verify and evaluate. The application system of "Darwin" chip needs to be realized by pulse neural network. Based on the above background, this paper designs and implements a pulse neural network platform for Darwin chip. It provides a customizable and intuitionistic simulation platform for the application design of brain-like computing chip. In this paper, the following contributions are made to design and implement the pulse neural network based on "Darwin" chip. A pulse neural network simulation platform is constructed and implemented. The platform can configure the network structure and parameters through visual operation interface. Based on the pulse neural network simulation platform, we can visualize the simulation results and process of the pulse neural network by realizing the self-defined structure of the pulse neural network, and can visualize the simulation results and the process of the pulse neural network simulation on the basis of the pulse neural network simulation platform. A virtual basketball demonstration system based on Darwin chip is designed and implemented.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;TP391.9

【相似文献】