SIMD-based soft error detection

Zhi Chen, Alexandru Nicolau, Alexander V. Veidenbaum
2016 Proceedings of the ACM International Conference on Computing Frontiers - CF '16  
Soft error rates in processors have been increasing with decreasing feature size and larger chips. Both hardware or software based solutions have been proposed to address this problem. However, this leads to significant overheads in chip area, performance, and/or energy. This paper proposes a novel, software-only, solution to the problem. It uses instruction duplication to detect and eventually correct transient faults with lower overhead than prior solutions. This is achieved by exploiting
more » ... ndancy within SIMD instructions. The solution is implemented in the LLVM compiler. Execution of a set of compiled benchmarks shows that SIMD based instruction duplication introduces 12% and 9% performance and energy overheads, respectively, over the baseline. The same overheads become 21% and 14% when error checking and branching code is added.
doi:10.1145/2903150.2903170 dblp:conf/cd/ChenNV16 fatcat:k7z4l2sngnbknlzxofyq234ef4