Early Assessment of the Fault Tolerance of Complex Digital Components From Virtual Platform Based Profiling

The broad spectrum of functionalities offered by complex digital components built around CPU cores makes them essential for onboard computers (OBCs) or instrument controller units (ICUs). In order to obtain a quick but accurate first assessment of their fault tolerance in radiative environments, both development and test teams must use a set of representative programs, before the actual application software is available. In this context, we propose a methodology based on high level software profiling metrics only, without any details on the actual CPU microarchitecture, to early assess fault tolerance of complex digital components. A derating factor equation only based on profiling metrics is calibrated with available fault tolerance data for a set of programs. This derating factor is then applied to the static cross section of the target device. The approach has been applied to a set of benchmarks running on a RISC-V softcore. Early predictions obtained from this methodology are compared to real measures collected during neutron irradiation tests. Results show that all the evaluations are within the margins of error of the experiment. The impact on the methodology accuracy of combining several profiling metrics or using several regression types when calculating the derating factor is also discussed.