Impact of Microarchitectural Differences of RISC-V Processor Cores on Soft Error Effects

Abstract

In this paper, we compare how radiation-induced soft errors affect the execution results of user-level applications on different processor cores that implement the same instruction set architecture (ISA). We target two processor cores that support the same RISC-V ISA but have significant differences in their microarchitectural implementations (in-order versus out-of-order). The observed results from fault injection experiments show very strong correlations between the resulting effects from those two processor cores. This strong correlation property is not observed between the processor cores that have different ISAs. Based on this observation, we discuss how the resulting effects of soft errors on a target processor core can be predicted using a reduced set of fault injection experiments with a small number of benchmark applications. Using our heuristic method of selecting the applications to be used for the fault injection experiments on the target processor core, we achieved a high prediction accuracy with prediction errors of less than 7%. Our approach can be used for rapid error resilience evaluation of system designs that have the same ISA.