A Configurable Approach to Tolerate Soft Errors via Partial Software Protection

Compared with hardware-based methods, software-based methods which need not additional hardware costs are regarded as efficient methods to tolerate soft errors. Software-based methods which are implemented by software protection have performance sacrifice. This paper proposes a new configurable approach whose purpose is to balance system reliability and performance, to tolerate soft errors via partial software protection. Those unprotected software regions which are motivated by soft error mask on software level are related to statically dead codes, those codes whose probabilities to be executed are low and some partially dead codes. For those protected codes, we copy every data and operate every operation twice to ensure those data stored into memory are right. Additionally, we ensure every branch instruction can jump to the right address by checking condition and destination address. Finally, our approach is implemented by modification of compiler. System reliability and performance are evaluated with different configurations. Experimental results demonstrate our purpose to balance system reliability and performance.