Checkers' No-Harm Alarms and Design Approaches to Tolerate Them

In this paper we analyze the probability that transient faults, multiple or single, affecting a checker of a self-checking circuit (with particular reference to the case of circuits using the two-rail code, the parity code, the Berger code and the Bose–Lin code) give rise to no-harm alarms, here defined as indications of errors neither denoting the presence of an incorrect word at the output of the functional block, nor denoting the presence of checker internal faults possibly compromising its ability to discriminate input codewords from input non-codewords. Differently from all other error indications, no-harm alarms could be conveniently ignored (or tolerated) by the system, with no need to adopt any recovery strategy upon their reception, otherwise for instance leading to exclude the self checking circuit from the whole system, or to degrade system’s performance. A new property (No-Harm Alarm Robustness) is defined for checkers, allowing to discriminate “true” error indications from no-harm alarms. A possible approach to design checkers featuring such a property is proposed. The behavior of the derived checkers has been verified by means of electrical level simulations, and their costs are discussed