用神经网络求解时延、功耗和连线三重驱动的布局问题

Neural Network Approach for Timing,Power Dissipation and Wire Connection Driven Placement

本文应用Ｋｏｈｏｎｅｎ 自组织神经网络求解时延、功耗和连线三重驱动的门阵列布局问题．算法用自组织学习算法和分配算法确定关键单元的位置，用迭代改善的方法确定非关键单元的位置，从而获得关键线网最短、散热大的单元离得尽可能远并且单元连线总长尽可能短的布局．本文还介绍了面向线网和功耗的样本矢量的概念，与面向单元的样本矢量相比，面向线网和功耗的样本矢量不仅可以直接处理多端线网，而且能够描述时延信息和热信息．

The Kohonen self\|organizing neural network approach is applied to the timing,power dissipation,and wire connection driven placement of gate array.In the algorithm,the critical nets are minimized,the distance between heat source cells is maximized,and the total length of nets is minimized.The self\|organizing learning approach and the assignment algorithm are used to fit the critical cells,and traditional iterative improvement method is used to fit the other cells.In addition,the concept of net and power dissipation oriented similarity vector is introduced.Compared with the cell oriented similarity vector,net and power dissipation oriented similarity vector can not only deal with the multi\|terminal nets directly,but also describe the information about timing and heat.The experimental result shows that it is an effective method.