Improving Code Generation for Software-based Error Detection

With modern hardware heading for smaller feature sizes and lower operating voltages, failure rates are expected to increase. Instead of adding hardware to protect against failures, which is expensive and inflexible, one can implement error detection in software. The focus of the present work is the AN encoding scheme. Improvements to code generation are introduced that enable a high level of fault coverage, namely over 98%, while reducing the performance overhead due to AN encoding by up to 45%. This raises the general question of whether code generation strategies can be designed that enable fault detection with a minimal impact on performance.