Abstract

Recently, many architectural level mechanisms such as dual or triple modular redundancies (DMR or TMR) have been included in high-end microprocessors to tolerate the continuously increasing electronic error rates along the process technology advancing direction. However, hardware in these space redundancy based systems is usually not well-balanced or it has been over-designed. In particular, it requires twice the hardware to achieve a reliable execution in a single issue processor, hence resulting in an area/performance efficiency decreased by half. This efficiency will be even worse when the redundancy is extended to multi-issue processors, which have already exchanged some area for performance enhancement. In this research, we propose a reliable 1.5-way superscalar processor by including time redundancy into the original space redundancy systems. This is usually not preferred when performance and permanent failure are considered. However, by this fusion of space and time redundancies, it is possible to perform dual executions on a triple-pipeline processor. Special working schemes have also been defined to distribute every dual-execution onto different units for both transient and permanent failure toleration. This architecture achieves better performance than a single issue DMR processor while balancing more area/performance than a dual-issue DMR processor. Our performance simulation and area synthesis results show that under equal reliability, the proposed scheme can have 22.6% better performance/area efficiency than a traditional DMR 2-way superscalar processor.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call