For maximum benefit, users of fault-tolerant (FT) systems must be assured that all components are fault free at the beginning of a mission. This is a very difficult task since, by design, faults are masked and tolerated.
Fidalgo et al. demonstrate that the controllability and observability features provided by on-chip debug (OCD) infrastructures can be easily enhanced to inject single-bit flip faults--simulating single-event upset (SEU) faults--into fault-tolerant systems to validate the reliability of the hardware at any instant in time. The paper contains a comparative analysis of various fault-injection methods, including offline and real-time methods using basic OCD structures, extended OCD structures, and customer-designed fault-injection modules. Experiments are performed on both standard and FT systems. The results suggest that designers of systems with OCD capabilities should consider adding fault-injection capacity, especially if the design involves FT features.
This paper is well written, technically sound, and readable by nonexperts.