Pruning single event upset faults with petri nets

Dependability of embedded systems is becoming a serious concern even for mass-market systems. Usually, designs are verified by means of fault injection campaigns, but the length of a thorough test often collides with the severe requirements about design cycle times. The number of fault injection experiments is thus usually reduced by performing random fault injections, or by focusing on selected fault models, or on components that depend on specific architectures and workloads. This forces to begin the validation campaign only when the system is fully designed, since specific details about the implementation or the workload are required. In this work, we propose to perform early fault pruning analysis on a formal model of the system, in order to identify the most critical components and computation cycles as soon as possible.