Fault Model Qualification By Assertion Mining

The process of measuring the quality of a fault model is a key ingredient for implementing effective verification/testing phases based on fault injection. Most of the existing approaches for the qualification of a fault model base their evaluation on the comparison of the achieved fault coverage against other code coverage metrics, or against the fault coverage achieved by different fault models, sometimes at the varying of the abstraction level. However, these approaches do not explicitly provide a measure of the accuracy of the fault injection with respect to the actual functional behaviours implemented in the design under verification/testing (DUV/T). Thus, the achievement of 100% fault coverage does not necessarily imply that all the design's behaviours have been accurately perturbed by the selected fault model. To provide a more accurate evaluation of fault models, this paper proposes a methodology based on assertion mining, i.e., automatic extraction of temporal assertions from the simulation of the DUV/T. Mined assertions are then used to highlight behaviours of the DUV/T that are not accurately perturbed by the selected fault model.