自组织人工神经网络在集成电路布局中的应用

APPLICATION OF SELF-ORGANIZATION NEURAL NETWORKIN VLSI PLACEMENT

本文把T.Kohonen的自组织特征映射人工神经网络用于集成电路的布局,改进了已有的适用于两端线网的自组织网络布局算法.针对多端线网的布局提出了两种不同的处理思想.面向单元的布局和面向线网的布局,给出了它们用自组织网络实现的算法.实验结果表明作为对多端线网更有效的处理工具,面向线网的布局算法能在较短的时间内得到比面向单元算法好的布局,且布局结果明显优于传统的最小割布局.

In this paper, Kohonen's self-organization neural network is applied to VLSI cell placement. The existing algorithm suitable for two-terminal nets is modified. In respect to multi-terminal net problems, two different placement strategies, namely gate-as-point and ner-as-point respectively, are proposed. Algorithms based on each of them are presented. Experiments show that algorithm based on net-as-point strategy can get better results in relatively shorter computational time compared with the pate-as-point based one. The placements got by the former algorithm are superior to those generated by conventional Min-Cut algorithm in terms of total wire length.