A low energy cache design for multimedia applications exploiting set access locality

被引:6
作者
Yang, J [1 ]
Yu, J
Zhang, YT
机构
[1] Univ Calif Riverside, Comp Sci & Engn Dept, Riverside, CA 92521 USA
[2] Univ Texas, Dept Comp Sci, Richardson, TX 75083 USA
基金
美国国家科学基金会;
关键词
low power design; cache; set buffer; multimedia;
D O I
10.1016/j.sysarc.2005.02.003
中图分类号
TP3 [计算技术、计算机技术];
学科分类号
0812 ;
摘要
An architectural technique is proposed to reduce power dissipation in conventional caches. Our technique is based on the observation of cache access locality: current access is likely to touch the same cache set including the tags as the last access. We show that considerable amount of power driving the cache tag and data banks can be saved if this cache access locality is fully exploited. This is achieved through buffering and accessing the last accessed cache set instead of driving the tag and data banks. Unlikely previous designs, our technique does not incur performance degradation. Experimental results carried out on 8 KB/16 KB/32 KB data and instruction caches have respectively shown 31%/35%/36% and 51%/58%/66% power savings. (c) 2005 Elsevier B.V. All rights reserved.
引用
收藏
页码:653 / 664
页数:12
相关论文
共 22 条
[1]  
Albonesi D.H., 2000, J INSTRUCTION LEVEL, V2
[2]  
ALLARM M, 2000, ACM IEEE INT S LOW P, P155
[3]  
[Anonymous], P INT S LOW POW DES
[4]  
Burger D, 1997, 1342 U WISC MAD COMP
[5]   Predictive sequential associative cache [J].
Calder, B ;
Grunwald, D ;
Emer, J .
SECOND INTERNATIONAL SYMPOSIUM ON HIGH-PERFORMANCE COMPUTER ARCHITECTURE, PROCEEDINGS, 1996, :244-253
[6]  
Canal R, 2000, INT SYMP MICROARCH, P181, DOI 10.1109/MICRO.2000.898069
[7]  
GHOSE K, 1999, ACM IEEE INT S LOW P, P70
[8]   Dynamic fine-grain leakage reduction using leakage-biased bitlines [J].
Heo, S ;
Barr, K ;
Hampton, M ;
Asanovic, K .
29TH ANNUAL INTERNATIONAL SYMPOSIUM ON COMPUTER ARCHITECTURE, PROCEEDINGS, 2002, :137-147
[9]  
HUANG M, 2001, ACM IEEE INT S LOW P, P10
[10]  
INOUE K, 2002, ACM IEEE INT S LOW P, P148