Using tag-match comparators for detecting soft errors

Abstract

Soft errors caused by high energy particle strikes are becoming an increasingly important problem in microprocessor design. With increasing transistor density and die sizes, soft errors are expected to be a larger problem in the near future. Recovering from these unexpected faults may be possible by reexecuting some part of the program only if the error can be detected. Therefore it is important to come up with new techniques to detect soft errors and increase the number of errors that are detected. Modern microprocessors employ out-of-order execution and dynamic scheduling logic. Comparator circuits, which are used to keep track of data dependencies, are usually idle. In this paper, we propose various schemes to exploit on-chip comparators to detect transient faults. Our results show that around 50% of the errors on the wakeup logic can be detected with minimal hardware overhead by using the proposed techniques.