News

Non-Uniform Fault Tolerance [abstract] (PDF)
Jonathan Chang, George A. Reis, and David I. August
Proceedings of the 2nd Workshop on Architectural Reliability (WAR), December 2006.

As devices become more susceptible to transient faults that can affect program correctness, processor designers will increasingly compensate by adding hardware or software redundancy. Proposed redundancy techniques and those currently in use are generally applied uniformly to a structure despite non-uniformity in the way errors within the structure manifest themselves in programs. This uniform protection leads to inefficiency in terms of performance, power, and area. Using case studies involving the register file, this paper motivates an alternative \emph{Non-Uniform Fault Tolerance} approach which improves reliability over uniform approaches by spending the redundancy budget on those areas most susceptible.