微处理器容软错误设计量化评估指标及评估方法

摘要

Aiming at highly reliable microprocessor soft error tolerance design,a new metric,eMWTF (enhanced mean work to failure),was proposed to capture the trade-off among reliability,performance,area and power.A quantitative approach for evaluating eMWTF was also presented.Two control flow checking techniques were quantitatively evaluated in reliability.The experimental results indicate that the control flow checking by compiler signatures and hardware checking achieves better trade-off among reliability,performance,area and power.Because the eMWTF metric takes into consideration performance,area and power overheads,the quantitative reliability evaluation can be more accurate by using this metric and corresponding methodology.Finally,the evaluation results can effectively guild the design exploring and optimization.%针对高可靠微处理器软容错设计,提出了一种新的可靠性度量标准,增强的平均无失效工作量,以解决现有度量标准没有综合考虑性能、面积、功耗开销带来的可靠性降低的缺点;提出了一种评估方法对增强的平均无失效工作量以及两种控制流检测技术进行定量评估.评估结果表明,软硬件结合的控制流检测技术较好地折中了可靠性、性能、面积和功耗.量化评估指标全面考虑了多种开销对微处理器可靠性的影响,采用相应的评估方法可以更加准确地对微处理器可靠性加固手段进行定量评估,以指导设计探索和设计优化.