摘要
提出了一种新型的 sigm oid函数发生器 .它不仅简单、快速 ,与理想 sigm oid函数的拟合程度好 ,而且可实现阈值和增益因子的编程 ,因而有很大的应用范围和良好的应用前景 .设计了神经元以及 Gilbert乘法器、数字存储器、 D/ A转换器等神经网络的基本单元 .说明了遗传算法 (GA)作为人工神经网络 (ANN)学习算法的有利因素 .利用上述电路 ,采用 GA,设计了可重构 ANN.对各单元电路和整个 ANN都用标准 1.2 μm CMOS工艺的第 47级模型进行了 HSPICE模拟 .结果表明它们的功能正确、性能优良 .
A novel sigmoid function generator is proposed.It is not only simple,fast,fit quite well with the ideal sigmoid function,but also programmable for the threshold and the gain factor.Thus,the neuron has wide application area and good prospect.Basic cells of the neural networks including the neuron,Gilbert multiplier,digital memory,D/A converter are designed.The benefit of Genetic Algorithm (GA) as an algorithm of the ANN is explained.A reconfigurable ANN composed of the above cells is designed,using GA as the training algorithm.The cells and the whole ANN are simulated with HSPICE,using level 47 transistor models for a standard 12μm CMOS process.Simulation results show that they all operate correctly and excellently.
基金
国家自然科学基金资助项目! (批准号 :6963 60 3 0 )&&
关键词
人工神经网络
模拟集成电路
遗传算法
电路设计
Artificial Neural Network (ANN)
analog integrated circuits
reconfigurable
programmable