Thermal-Aware Floorplanning Considering Empty Space Effect Based on Genetic Algorithms  

作　　者：LIU Yongpan YANG Huazhong LUO Rong WANG Hui 

机构地区：[1]Department of Electronics Engineering, Tsinghua University, Beijing 100084, China

出　　处：《Chinese Journal of Electronics》2007年第3期429-434,共6页电子学报（英文版）

基　　金：This work is supported by the National Natural Science Foundation of China （No.90207001, No.60506010） and National 863 Project of China （No.2004AA1Z1050, No.2005AA1Z1230）.

摘　　要：As power density increases steadily, temperature has become an important design concern. One of the most effective ways to reduce chip peak temperature is thermal-aware fioorplanning, in which thermal analysis has to be called iteratively. Thus, a computing efficient thermal model is quite important. Compact thermal models are promising, since they maintain reasonable accuracy and provide magnitudes speedup compared with numerical models. However, they are sensitive to empty spaces. In this paper, we analyze Empty space effects （ESE） in compact thermal models and reveal that filtering empty spaces under traditional area threshold is unsafe. A novel adaptive threshold procedure is proposed to acquire better accuracy with less than 1.79% maximum and 0.29% average errors. Furthermore, by integrating compact thermal model and empty space processing procedure, we implement a non-slicing thermal-aware fioorplanner based on multiobjective genetic algorithms for the first time. It relies on a novel two-stage fitness function and peak temperatures in different benchmarks are reduced from 4.1 to 57.4° under similar compact ratios. We show that thermalaware fioorplanning considering ESE always obtains better results than the one without considering ESE.

关 键 词：Thermal-aware floorplan Empty space effect Compact resistive thermal model Genetic algorithms 

分 类 号：TN0[电子电信—物理电子学]