About on-line functionally untestable fault identification in microprocessor cores for safety-critical applications

When microprocessor cores are used in safety-critical applications, in-field test must be performed to reach the target reliability figures. In turns, the in-field test must be organized so that it achieves a sufficient fault coverage. The fault list to be considered for computing the fault coverage should only include testable faults, i.e., faults which may cause a failure in the operating conditions. Hence, single permanent faults that in the operating conditions cannot be excited, or do not propagate to any output, or both, should be removed from the list. These faults, called on-line functionally untestable faults, require a significant effort to be identified. The contribution of this paper is twofold. From one side, it reports experiments, showing that in typical embedded safety-critical systems their number is often far from being negligible, and depends on many parameters, including the application code run by the processor. Secondly, it provides a semi-automated approach to their identification. Experimental results on a representative microprocessor are reported.