Towards Making Fault Injection on Abstract Models a More Accurate Tool for Predicting RT-Level Effects

Abstract

Fault injection and fault simulation are a typical approach to analyze the effect of a fault on a hardware/software system. Often fault injection is done on abstract models of the system either to retrieve early results when no implementation is available, yet, or to speed-up the runtime intensive fault simulation on detailed models. The simulation results from the abstract model are typically inaccurate because details of the concrete hardware are missing. Here, we propose an approach to relate faults from an abstract untimed algorithmic model to their counterparts in the concrete register transfer models. This allows to understand which faults are covered on the concrete model and to speed up the fault simulation process. We use a mapping between both models' variables and mapped timing states for fault injection to corresponding variables on both models. After fault simulations the results are compared to check, whether a given fault produces the same behavior on both models. The results show that an injected fault to corresponding variables leads to the same behavior of both models for a large share of faults.