Source code transformation for software-based on-line error detection

Abstract

With the widespread deployment of ICs into safety-critical systems, ensuring in-field reliability for the expected IC life span becomes a serious challenge. One solution to improving reliability is software-based on-line error detection; the idea is to empower the application program to detect run-time errors which may be permanent, intermittent, or transient. In this paper, we propose a set of high-level source code transformation rules to facilitate software-based on-line error detection. Compared to previous works, the proposed source code transformation rules improve the coverage on flow control constructs and can handle structures as well as structure pointers. Experiments on singlebit error injection show that in average only 1.08% of faults that modify the program behavior are undetected.