Some Faults are Worse Than Others⎯And How That is Useful for Low- Cost Hardening

Traditionally, a fault handling strategy is considered effective if it covers large classes of faults, e.g., all single faults. This conventional wisdom has recently been challenged by identifying subsets of faults which are acceptable at system level. One example is a fault in an imaging microchip which does not result in a deterioration of the calculated image to the extent that a human viewer would notice the difference. In context of microand nanoelectronic circuits, dropping the restrictive requirement that all faults be covered enables cost-effective selective hardening solutions, where only parts of the circuit are equipped with fault protection mechanisms. In this way, the need for traditional massive redundancy schemes such as triplemodular redundancy, which are associated with massive area and energy consumption overheads, is eliminated. Handling most critical faults could be associated with overheads as low as 10%, which is practical even for cost-sensitive embedded systems with a limited energy budget.