Mitigating Soft Errors in Highly Associative Cache with CAM-based Tag

  • Authors:
  • Luong D. Hung;Masahiro Goshima;Shuichi Sakai

  • Affiliations:
  • Graduate School of Information Science and Technology, The University of Tokyo, Tokyo, Bunkyo, Hongo, Japan;Graduate School of Information Science and Technology, The University of Tokyo, Tokyo, Bunkyo, Hongo, Japan;Graduate School of Information Science and Technology, The University of Tokyo, Tokyo, Bunkyo, Hongo, Japan

  • Venue:
  • ICCD '05 Proceedings of the 2005 International Conference on Computer Design
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

Content Addressable Memories (CAM) are widely used for the tag portions in highly associative caches. Since data are not explicitly read out of tag array in CAM search, the detection of false misses caused by soft errors for such caches, is difficult. This paper presents a technique to detect the false miss in highly associative cache with CAMbased tag. The technique involves subdividing the tags and providing backup checking for cases the tags are partially matched. An original tag encoding scheme is proposed to reduce the frequency of back-up checking. Modifications to support the technique do not increase the cache access latency. The performance degradation incurred by additional cycles for false miss checking is very low.