Controllable signature check pointing scheme for transient error detection

A program can fail during execution owing to software, hardware, or extraneous faults. There are three types of errors caused by extraneous faults: transient, intermittent, and permanent. This work deals with transient faults. The effect of transient faults during a program execution is often unpredictable and temporary in nature. As a result, traditional approaches to error detection are not suitable for detecting transient errors. The authors present a flexible and controllable transient error detection scheme called the controllable signature check pointing scheme. The new signature monitoring differs from existing schemes in three major aspects: (1) the intervals in the program to be monitored are decided at a higher level of abstraction (high-level-language level); (2) the interval size is chosen not arbitrarily but by applying control-flow and bulk complexity measures; (3) the interval size can be controlled to suit the application. The location of the signature check points and the frequency of occurrence of the check points can be controlled by varying the interval control-flow complexity factor (INCCF). Thus the error coverage and latency, memory overhead, and performance degradation can also be tuned to suit the application. Future refinements and extensions include signature check pointing of sequential code and reducing signature embedding overhead.