Description

This thesis addresses one of the fundamental challenges emerging in microprocessor design, namely hardware reliability and resilience. Since inception in the 70's, microprocessors have primarily benefited from technological advancements in semiconductors fabrication allowing for an exponential increase in computing capability of chips by shrinking transistors sizes. Unfortunately, forecasts indicate that further shrinking in size will be accompanied by variability in transistor performance and reliability. This thesis proposes novel designs and enhancements to provide hardware reliability for parallel workloads. In particular, it is provided noteworthy improvements in Redundant Multi Threading (RMT) fault-tolerant approaches, as well as novel Expected Miss Ratio (EMR) model to determine the impact of hard faults on cache memories.