面向EDA课程设计的数字神经元电路实现

Implementation of digital neuron circuit oriented to EDA curriculum design

为提高电子类专业学生FPGA的应用能力,面向EDA课程设计,提出了一种利用FPGA实现电子神经元的通用架构。该架构包括PL端的模型函数模块、龙格库塔参量模块、变量更新模块、主控制模块、流输出模块以及PS端的接口模块。PL端与数值计算有关的模块均采用单精度浮点数计算,以保证中间结果的计算精度。PS端程序运行于FPGA的嵌入式ARM处理器上,主要负责数字神经元的对外接口,实现数字神经元初始值和模型参数的在线重置。利用该架构,用户只需编写每种神经元模型特有的模型函数模块,便可方便快速地在FPGA上构建其数字神经元电路。

In order to improve the FPGA application capability of students majored in the electronics class,a universal framework is proposed to realize electronic neurons oriented to EDA curriculum design.The framework is composed of modules in PL end such as model function block,Runge-Kutta parameter block,variable updating block,main controller,stream output block,and interface module in PS end.The modules concerning the numerical calculation in the PL end utilize single-precision floating-point computation to guarantee the computational accuracy of intermediate results.As an external interface,the embedded program in the PS end running on the ARM processor realizes online resetting of initial values and model parameters.Using the proposed framework,users only need to concentrate on programming the specific module function of each kind of neuron,such that the digital neuron circuit can be implemented conveniently and rapidly on FPGA.