Using software invariants for dynamic detection of transient errors

Abstract

Software based error detection techniques usually imply modification of the algorithms to be hardened, and almost certainly also demand a huge memory footprint and/or execution time overhead. In the software engineering field, program invariants have been proposed as a means to check program correctness during the development cycle. In this work we discuss the use of software invariants verification as a low cost alternative to detect soft errors after the execution of a given algorithm. A clear advantage is that this approach does not require any change in the algorithm to be hardened, and in case its computational cost and memory overhead are proven to be much smaller than duplication for a given algorithm, it may become a feasible option for hardening that algorithm against soft errors. The results of fault injection experiments performed with different algorithms are analyzed and some guidelines for future research concerning this technique are proposed.