加权主元分析在统计互连寄生参数提取中的应用

Weighted Principal Factor Analysis in Statistical Interconnect Parasitic Extraction

针对随机工艺变动引起的统计寄生参数提取问题,采用Hermite多项式配置法提出加权主元分析技术来对随机变量进行消减,以减少独立变量数目,提高计算效率.在此基础上,利用并行计算技术,进一步减少统计寄生参数提取的运行时间.数值实验结果表明,相对于普通的主元分析,采用文中的加权主元分析能在同等精度情况下使寄生参数提取速度提高几倍至几十倍,而在含8个CPU计算机上的并行计算也取得了6.7倍的加速比.

To cope with the problems of statistical parasitic extraction induced by random process variations,the technique of weighted principal factor analysis（wPFA）based on Hermite Polynomial Collocation method is proposed to reduce the number of random variables and improve the computational efficiency.The parallel computing technique is also applied to further reduce the computational time.Numerical results show that,the wPFA is able to accelerate the statistical extraction using a normal principal factor analysis by several or several tens times.While,the parallel computing experiment on a machine with 8 CPU achieves a speedup of 6.7.