An effective strategy for selective hardening of software

This paper presents an effective design space exploration strategy for the development of dependable systems using selective hardening techniques based on software. Instead of design space exploration approaches based on brute-force or time-consuming fault injection experiments, this strategy is grounded in an early estimation of the register file criticality in microprocessor-based systems. This estimation is possible thanks to an application-based criticality metric which relies on different criteria based on common features of executed applications. Experiments show a significant improvement with respect to approaches based on brute-force or the traditional AVF criticality metric.