An approach to analyze effects of soft errors on program level

Computer hardware components are susceptible to soft errors which could result in incorrect program execution with improvement of semiconductor technology. To measure the reliability of hardware components, it is important to know the effect of soft error on program outcome. This paper analyzes the effects of soft errors on the level of the program. We choose an approach which combines a high level with a low level to model the program. According to the behaviors of soft errors on hardware components during the instruction pipeline, we build a soft error model. Based on our program model and soft error model, the effect of soft errors on the whole computing program is explored from the granularity of the function. Furthermore, data flow errors and control flow errors which are two main implicit program errors caused by soft errors are analyzed respectively. Based on our analyses, we give a method to estimate the Architectural Vulnerability Factor (AVF) of hardware components statically. Our experiments illustrate this method and indicate our analysis results that there are different program behaviors under the effects of soft errors, especially the ability for the program to mask program errors which are caused by soft errors.