Observability solutions for in-field functional test of processor-based systems

The growing usage of electronic systems in safety-critical applications requires effective solutions to early identify possible faults affecting the hardware while it is in the operational phase. A possible approach leverages functional programs to be run by the CPU typically existing in such systems. These programs must exercise the different parts of the system, and produce a behavior different than the normal one in case of faults. However, their effectiveness depends on the adopted observation mechanism, which is deeply affected by the constraints imposed by the in-field application environment. This paper first describes different mechanisms for supporting the observation of possible fault effects; then, it reports and discusses the results of an experimental analysis performed on a multicore system, based on a representative pipelined processor. The gathered results allow to quantitatively evaluate the drop in fault coverage coming from the adoption of the different observation solutions with respect to the ideal case, and thus to better evaluate the advantages/disadvantages they provide.