Data flow transformations to detect results which are corrupted by hardware faults

Abstract

Design diversity, which is generally used to detect software faults, can be used to detect hardware faults without any additional measures. Since design of diverse programs may use hardware parts in the same way, the hardware fault coverage obtained is insufficient. To improve hardware fault coverage, a method is presented that systematically transforms every instruction of a given program into a modified instruction (sequence), keeping the algorithm fixed. This transformation is based on a diverse data representation and accompanying modified instruction sequences, that calculate the original results in the diverse data representation. If original and systematically modified variants of a program are executed sequentially, the results can be compared online to detect hardware faults. For this method, different diverse data representation have been examined. For the most suitable representation, the accompanying modified instruction sequences have been generated at assembler level and at high language level. The theoretically estimated improvement of the fault coverage of design diversity by additionally using systematically generated diversity have been confirmed by practical examinations.