2014 · 15 citations · 14 references
EngineeringComputer EngineeringComputer ScienceFault-tolerant ComputingFault-tolerant SystemsReal-time SystemsFault DetectionHardware Fault ToleranceReal-time Systems ArchitectureRedundancy TechniquesReliable System DesignTransient Fault HandlingFault Detection RateFault Tolerance MechanismsFault Tolerant ArchitecturesFault-tolerant ArchitectureFault-tolerant Processor DesignNoise MarginsTransient Fault MitigationTransient Faults
Performance improvements in transistors have also brought about lower threshold voltages and tighter noise margins. Therefore, microprocessors are becoming more vulnerable to transient faults, and Lockstep is a technology that has been developed to provide a good method to address such transient faults. In Lockstep, two processors perform and compare the same computation during every cycle. Consequently, transient faults can be detected with a high fault-detection rate. However, this technique only gives a warning of wrong information or executes a reset simply by comparing replications of the data in the cores. Thus, they are not appropriate for real-time systems used in the automotive industry. To this end, we propose a method to provide hardware redundancy in a system. This system uses an extra core and a voting algorithm when a fault occurs, and it is appropriate for real-time systems.
14
IBM's S/390 G5 microprocessor design
T. J. Slegel, R. M. Averill, M. A. Check et al. · IEEE Micro · 1999 · 349 citations