基于GPU集群层次式并行计算的3D芯片电热综合分析与综合优化

项目名称： 基于GPU集群层次式并行计算的3D芯片电热综合分析与综合优化

项目编号： No.61274033

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 无线电电子学、电信技术

项目作者： 骆祖莹

作者单位： 北京师范大学

项目金额： 80万元

中文摘要： 高温与供电困难是3D集成技术所面临的重要挑战，由于功耗/电压/温度/性能参数之间存在明显的相互影响，所以必须对3D芯片进行高效而精确的电热综合分析与综合优化。为了考虑功耗/电压/温度之间的相互影响，我们首次提出了电热综合分析方法，对功耗/电压/温度进行迭代更新计算以获得三者的精确解，此外我们还采用了并行计算技术对电热综合分析进行了并行加速。本项目计划开展基于GPU集群层次式并行计算的3D芯片电热综合分析与综合优化的算法研究，内容包括：(1)考虑功耗/温度/电压/性能诸参数之间相互影响的电热综合分析与综合优化的建模，(2)面向3D芯片电热综合分析与综合优化的并行计算方法研究，(3)利用GPU集群的层次式并行计算资源、对3D芯片电热综合分析与综合优化进行并行加速，(4)集成一套EDA工具软件原型系统。 本项目是已结题项目的滚动研究，以出高水平研究成果为目标，预计发表16-20 篇高质量论文。

中文关键词： 3D芯片；电热分析；电热优化；并行计算；GPU 集群

英文摘要： High temperature and power delivery difficulty are great challenges to the three-dimension (3D) integration technology. Since there are visible influences among power/ voltage/temperature/performance, it is necessary to do efficient and accurate electro-thermal (ET) co-analysis and co-optimization for 3D chips. In order to consider influences among power/ voltage/temperature, for the first time, we have proposed the ET co-analysis method, in which we use iterative computing to refresh three parameters and obtain the accurate solution at last. Furthermore we use the parallel computing to speed up the ET co-analysis. In this project, we plan to study the following algorithms on ET co-analysis and co-optimization for 3D chips based on hierarchical parallel computing resources provided by a GPU cluster. (1) Model the ET co-analysis and co-optimization with considering influences among power/voltage/temperature/ performance. (2) Parallel computing studies on ET co-analysis and co-optimization for 3D chips. (3) With hierarchical parallel computing resources provided by a GPU cluster, we plan to utilize hierarchical parallel computing techniques to algorithm studies on ET co-analysis and co-optimization for 3D chips. (4) Integrate a show-how EDA prototype system for ET co-analysis and co-optimization of 3D chips. Thi

英文关键词： 3D integration；Electro-thermal analysis；Electro-thermal optimization；Parallel computing；GPU cluster