一种基于逻辑最小项化简的非线性函数实现方法  

An Implementation Method for Nonlinear Function Based on Logic Minimization

作　　者：李昂[1] 王沁[1] 吴巍[1] 钱艺[1] 

机构地区：[1]北京科技大学信息工程学院,北京100083

出　　处：《小型微型计算机系统》2008年第3期563-566,共4页Journal of Chinese Computer Systems

摘　　要：神经网络的数字VLSI实现中,非线性激活函数是计算部件设计中的难点.区别于其他模拟近似方法,本文提出的LMN方法使用逻辑最小项化简,对查找表(即函数真值表)进行有效压缩,提炼出函数最简逻辑表达式后,可直接生成结构规整的门级电路,且除线延时外,电路只有数个门级延时.该方法使用程序实现,简化了设计过程.结果表明当定点位数小于12时,该方法在速度及误差方面具有更好的性能.Nonlinear function＇ s implementation is a difficult issue in VLSI design of neural networks. Differ from other approximation methods, the LMN method proposed in this paper uses logic minimization to compress the function＇s look-up-table （true table） effectively, and produces the simplest logic expressions. These expressions can be translated straightforward into gates level circuits with canonical structure and very little gate delays. Program was developed to simply the design process. Results show when fixed point numbers are less than 12 bits, this method has a better performance in speed and error.

关 键 词：神经网络 VLSI 非线性函数 逻辑化简 

分 类 号：TP331[自动化与计算机技术—计算机系统结构]