论文部分内容阅读
基于神经网络与模糊逻辑的手写体数字识别的算法，设计出一种结构紧凑灵活的可编程可配置电流型集成电路神经网络／模糊逻辑识别器用来进行手写体数字的识别。这种识别器既可实现单层或多层感知机的神经网络算法，也可实现模糊逻辑的识别算法。该识别器利用了电流信号的可操作性强的特点来实现神经网络与模糊逻辑中的乘加运算和非线性变换，并采用开关电流（ＳＩ）技术进行累加运算。ＨＳＰＩＣＥ模拟表明该识别器性能良好，并且其所需工艺完全与标准数字ＣＭＯＳ工艺兼容。
Based on the algorithm of neural network and fuzzy logic handwritten numeral recognition, a compact and flexible program-configurable current-type integrated circuit neural network / fuzzy logic recognizer is designed to recognize handwritten digits. This recognizer can not only realize the neural network algorithm of single-layer or multi-layer perceptron, but also realize the recognition algorithm of fuzzy logic. The recognizer utilizes the strong operability of the current signal to realize the multiply-add operation and the non-linear transform in the neural network and fuzzy logic, and uses the switch current (SI) technique to perform the accumulative operation. HSPICE simulations show that the recognizer performs well and that its required process is fully compatible with standard digital CMOS processes.