并行空间自适应多重网格集成电路热分析方法
Parallel Space Adaptive Multigrid Method for IC Thermal Analysis
-
摘要: 针对集成电路设计阶段的重要步骤之一——热分析中的稳态分析问题,将多线程并行计算技术和目前最先进的空间自适应多重网格方法相结合,提出一种基于多核并行空间自适应多重网格的热分析方法.根据CPU的核数对芯片进行区域划分,并根据划分情况构造适用于块Jacobi迭代的分块热导矩阵,每一个划分区域对应一个矩阵子块.对全局的稳态热分析问题采用块Jacobi迭代法,每一次块Jacobi迭代需要并行地计算每个子块方程组,每个子块方程组的计算采用空间自适应多重网格方法;用八叉森林数据结构记录每个子块的网格层次信息,为多重网格计算提供不同网格层次下的热导矩阵.实验结果表明,文中方法在保证稳态热分析精度的前提下,在32核CPU上可以达到近似线性的理想加速比,能够有效地应用于集成电路设计流程,提高热分析效率.Abstract: With the increasing scale of IC and the development of three-dimensional integration technology,thermal analysis becomes an important part in IC design flow.In iterative processes of IC synthesis in circuit design,IC thermal tool will be invoked frequently.Therefore,IC thermal analysis should provide adequate accuracy and efficiency to support IC synthesis.With the development of multicore CPU technology,utilizing multicore parallel technology to improve computational efficiency becomes a hot topic in IC design automation.Aim at steady-state thermal analysis in thermal analysis,parallel space adaptive multigrid method based on multicore is proposed.By partitioning thermal analysis problem by regions and parallelizing it using block Jacobi iterative method and space adaptive multigrid method,the proposed method improves the efficiency of thermal analysis remarkably.Experimental results show that parallel space adaptive multigrid method reached to a perfect speedup,which is close to theoretic linear speedup.This method can be applied to IC design flow effectively to accelerate the thermal analysis.