纳米工艺下大规模线网的快速随机行走电容提取技术  

Efficient Techniques for Handling Large Nets in Capacitance Extraction Based on Floating Random Walk

作　　者：宋明烨 喻文健[1,2] Song Mingye;Yu Wenjian(Department of Computer Science and Technology,Tsinghua University,Beijing 100084;Beijing National Research Center for Information Science and Technology,Tsinghua University,Beijing 100084)

机构地区：[1]清华大学计算机科学与技术系,北京100084 [2]清华大学北京信息科学与技术国家研究中心,北京100084

出　　处：《计算机辅助设计与图形学学报》2022年第4期491-498,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(62090025);清华大学自主科研计划(2021Z11GHX006)。

摘　　要：为了解决悬浮随机行走算法在处理超过十万级别的大规模线网电容提取任务时耗时过长的问题,提出一种适用于悬浮随机行走电容提取算法中的快速虚拟高斯面构造技术.首先对已有的虚拟高斯面采样方法进行时间复杂度分析,并在其基础上发现影响效率的计算瓶颈——导体块之间的距离计算和高斯面相交关系计算;然后提出采用空间管理代替多线程并行加速的方法,利用网格的空间管理优化大线网的虚拟高斯面构造过程,从而显著减少构造的时间,极大地加快了大规模线网的高斯面构造.在一台32核服务器上的数值实验结果表明,在一个包含约100万块导体块的大规模测例上的高斯面构造过程中,该方法获得高达372×的加速比,远远优于多线程并行的效果.An approach for fast generation of virtual Gaussian surface in floating random walk algorithm is proposed,which resolves the difficulty of performing capacitance extraction on large wire nets including over one hundred thousand conductor blocks.Firstly,computational complexity of the existing virtual Gaussian surface sampling technique is analyzed.Based on it,techniques are proposed to improve the distance calculation among conductor blocks and the intersection calculation of block Gaussian surfaces.The techniques are based on space management and reduce the computational time for generating the Gaussian surfaces,achieving better performance than the approach based on parallel computing.The efficiency and correctness of the proposed approach is verified with the experiments carried out on a server with 32 cores.The experimental results show that,for a large structure including about one million conductor blocks the proposed approach accelerates generating the Gaussian surface for 372 times,which is much better than the acceleration brought by paralleling computing.

关 键 词：电容提取 悬浮随机行走 空间管理 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]