A Study on Selective Implementation Approaches for Soft Error Detection Using S-SWIFT-R

This article analyzes diverse criteria for effectively implementing selective hardening against soft errors through software-based strategies. The goal is to obtain maximum fault coverage with the least amount of overhead for each specific application. To achieve this objective, the analysis is conducted based on two important phases, pre-selection and selective hardening of registers. In the pre-selection phase, the impact of the two most used selection metrics has been examined: (1) selecting registers based on their memory interaction vs. (2) selecting registers depending on fault injection vulnerability. Toward the selective hardening phase, the impact of gradually increasing the number of registers to protect is examined. Experiments have been conducted on 8 academic case studies and 1 industrial case study. Faults have been injected into the case studies using our in-house fault injector. The results indicate that selecting registers based on the fault injected into the system has an overall 10% better performance in comparison to selecting registers based on the memory interaction in 6 out of 8 academic case studies and also the industrial case study. Additionally, there is a significant improvement in reliability when increasing the number of registers to protect at the expense of rising overhead. In this work, these comparisons and analyses are presented.