Characterizing the Effects of Transient Faults on a High-Performance Processor Pipeline

  • Authors:
  • Nicholas J. Wang;Justin Quek;Todd M. Rafacz;Sanjay J. patel

  • Affiliations:
  • University of Illinois at Urbana-Champaign;University of Illinois at Urbana-Champaign;University of Illinois at Urbana-Champaign;University of Illinois at Urbana-Champaign

  • Venue:
  • DSN '04 Proceedings of the 2004 International Conference on Dependable Systems and Networks
  • Year:
  • 2004

Quantified Score

Hi-index 0.02

Visualization

Abstract

The progression of implementation technologies into thesub-100 nanometer lithographies renew the importance ofunderstanding and protecting against single-event upsets indigital systems.In this work, the effects of transient faults onhigh performance microprocessors is explored.To perform athorough exploration, a highly detailed register transfer levelmodel of a deeply pipelined, out-of-order microprocessor wascreated.Using fault injection, we determined that fewer than15% of single bit corruptions in processor state result in softwarevisible errors.These failures were analyzed to identifythe most vulnerable portions of the processor, which werethen protected using simple low-overhead techniques.Thisresulted in a 75% reduction in failures.Building upon thefailure modes seen in the microarchitecture, fault injectionsinto software were performed to investigate the level of maskingthat the software layer provides.Together, the baselinemicroarchitectural substrate and software mask more than 9out of 10 transient faults from affecting correct program execution.