论文部分内容阅读
MCM是集成电路中的一种新技术.划分是MCM设计中极其重要的一个环节.本文应用Kohonen自组织神经网络求解以面积和时延为约束的、以芯片之间的连线代价和系统时钟周期为优化目标的MCM系统划分问题.算法用单元之间的联接度和组合逻辑单元的内部时延表示直接相联单元间的相似性,并应用模糊相似性变换建立间接相联单元间的相似性.算法将各单元映射到二维平面上,对应一个或者多个神经元.学习过程是通过单元之间有协作的移动,使相似性大的单元能够逐渐移到一起来完成的.
MCM is a new technology in integrated circuits.Classification is an extremely important part of MCM design.This paper applies Kohonen self-organizing neural network to solve the problem of area and delay constrained by the connection cost between chips and system clock Period is the objective of MCM system partitioning problem.The algorithm uses the degree of connection between units and the internal delay of combinatorial logic unit to represent the similarity between direct connected units and the fuzzy similarity transformation to establish the similarity between indirect units The algorithm maps each unit to a two-dimensional plane, corresponding to one or more neurons.The learning process is achieved by the coordinated movement between units, so that units with similarities can be gradually moved together.