The exponentially increasing occurrence of soft errors makes the optimization of reliability, performance, hardware area, and power consumption one of the main concerns in modern embedded processors. Since the design cost of hardware techniques aimed at improving the reliability of microprocessors is quite expensive for resource-constrained embedded systems, software-level fault tolerance mechanisms have been proposed as an attractive solution for soft error threats. However, many software-level redundancy-based schemes are accompanied by considerable performance overhead, which is not acceptable for many embedded applications. In this work, we have introduced an ultra-low-cost soft error protection scheme for embedded applications, which works based on source-code analysis and identifying critical variables. After identification, these vital variables are adequately protected by placing runtime checks at critical points of execution. Our experimental results based on several applications demonstrate that the proposed scheme can mitigate the failure rate by 47% with negligible performance degradation.
Cost-effective soft-error protection for SRAM-based structures in GPGPUs