中文摘要：

受限于计算能力,在现有的电热分析研究中,无法考虑电压变化对电热分析的影响,从而降低了分析的精度.基于已有的研究成果,文中分析了芯片的功耗/电压/温度分布向量之间的相互关系,指出了在电热综合分析中考虑电压/温度变化的必要性,进而提出了一种迭代式的并行电热综合分析方法ETA_VT,该方法基于功耗与电压/温度之间的递归关系,进行迭代计算,最后将收敛后的功耗/电压/温度分布向量作为求解结果同时输出.基于多核CPU＋众核GPU异构计算机系统所提供的并行计算资源,为了提高电热综合分析的运行效率,文中不仅设计了一个具有主辅双进程的ETA_VT算法流程,而且还分别采用CPU多线程并行计算、GPU并行计算、CPU＋GPU协同并行计算技术对ETA_VT算法进行加速研究.实验数据表明：（1）考虑电压/温度变化的电热综合分析不仅可以获得较为精确的分析结果,而且可以同时计算出芯片的功耗/电压/温度分布;（2）采用并行计算技术、并合理分配计算资源,不仅可以解决电热综合分析中存在的功耗/电压/温度多参量相互影响的问题,而且还可以有效地提高电热综合分析的速度,获得多达44倍的加速效果.文中工作是将高性能计算引入电子设计自动化（EDA）算法研究的一次有益尝试,表明高性能计算技术不仅可以提高EDA算法的执行效率,而且可以促进芯片设计中存在的多参量相互影响综合分析问题的研究和解决.

英文摘要：

Owing to limited computing capacity,present electro-thermal（ET） analysis methods are unable to take the influence of voltage change into consideration,which in turn lowers ET analysis accuracy inevitably.Based on our ET researches,this paper analyzes the relationship among P/V/T,IC distribution vectors of power/voltage/temperature,and then points out the necessity to consider influences of V T changes in ET analysis,and further proposes an iterative ET co-analysis method named ETA_VT.ETA_VT uses iterative computing to conquer P/V/T＇s interaction and simultaneously outputs the converged P/V/T vectors at last.With parallel computing resources provided by a hetero-architecture computing system consisting of one multi-core CPU and one many-core GPU,in order to increase the efficiency of ET co-analysis,this work constructs a dual-process algorithm for ETA_VT and then proposes several speedup techniques including CPU＇s multi-thread parallel computing,GPU＇s many-thread parallel computing,and CPU＋GPU＇s cooperated parallel computing.Experimental data draw the following two conclusions.（1） As a single co-analysis platform,ETA_VT not only provides accurate P/V/T vectors but also can output them simultaneously;（2） The parallel computing technology can be used to solve the P/V/T interaction in the ET co-analysis and to increase the ET co-analysis efficiency up to the utmost of 44X speedup through rationally distributing computing resources.Thus this work is a helpful trial in introducing high performance computing（HPC） into EDA algorithm studies,which means that HPC not only increases the efficiency of EDA algorithm but also solves complicated co-analyses of multi-parameter interactions in IC design.