Introduction

Modern digital systems are subject to a variety of potential faults that can corrupt their output and degrade their performance [Johnson, 1989; Pradhan, 1996; Siewiorek and Swarz, 1998]. In this context, a fault is a deviation of a given system from its required or expected behavior. The more complex a computational system is or the longer an algorithm runs for, the higher is the risk of a hardware malfunction that renders the overall functionality of the system useless. Depending on the duration of faults, two broad classes are defined [Johnson, 1989]:(i) Permanent faults manifest themselves in a consistent manner and include design or software errors, manufacturing defects, or irreversible physical damage. (ii) Transient faults do not appear on a consistent basis and only manifest themselves in a certain portion of system invocations; transient faults could be due to noise, such as absorption of alpha particles and electromagnetic interference, or environmental factors, such as overheating.