Detection of soft errors through checksums in redundant execution systems

As technology trends keep pushing transistor dimensions to smaller geometries and higher densities, the gains in reliability over the past decades have started to reverse trends due to the gradual increase in soft error rates. Consumer applications that cannot afford hardware redundancy solutions to reduce the effects of soft errors, can find a cost effective alternative using time redundancy. The memory requirements needed for a redundant execution system can be minimized without compromising soft error detection by using hardware execution signatures. The proposed technique calculates checksum signatures that summarize the past and present state of execution of a microprocessor. In this system, the soft error detection logic generates partial checksum signatures during the first execution of a microprocessor and compares them to the signatures generated during the second execution. Soft errors are detected when there is a signature mismatch between the two program executions. To minimize time overhead, the microprocessor was optimized to automatically run redundant executions without the need of duplicating software. The system based on redundant execution using state execution signatures achieved high soft error detection while simultaneously maintaining reduced memory requirements.