摘要
在基于FPGA的神经网络设计中,提出一种采用直接坐标旋转数字计算机(CORDIC)算法计算神经元激励函数ex的方法,依靠移位和求和能够实现快速、精确的指数函数计算,较查表法和间接CORDIC算法既节省了大量片内资源,又提高了计算速度和精度.利用Xilinx公司ISE开发工具进行仿真实验,结果表明直接CORDIC算法的计算速度是间接CORDIC算法的14倍,证明了该算法计算指数函数的快速性与精确性.
In the design process of neural networks based on the FPGA, a new direct coordinate rotation digital computer (CORDIC) is proposed to calculate the activation function of neuron. It calculates fast and accurately rely on shift-add operations. Compared with the Look-table approach and indirect CORDIC approach, the direct CORDIC not only save large numbers of logic gates, but also improve the rate and precision of calculation. This paper uses ISE software of Xilinx company to simulate the CORDIC algorithm, and the results show the fast and accurate character of the proposed algorithm.
出处
《华中科技大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2009年第6期36-39,共4页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
湖北省自然科学基金资助项目(2007ABA281)
国家自然科学基金资助项目(60874047)
关键词
神经网络
神经元
坐标旋转数字计算机
激励函数
可编程逻辑门阵列
neural network
neruon
coordinate rotation digital computer
activation function
field programable gate array (FPGA)