基于错误传播分析的SDC脆弱指令识别方法

发布时间：2018-02-11 20:15

本文关键词： 单粒子翻转单粒子软错误 SDC脆弱指令错误注入错误传播　出处：《计算机研究与发展》2016年09期 　论文类型：期刊论文

【摘要】：单粒子软错误是高辐照空间环境下影响计算可靠性的主要因素.随着芯片晶体管数的快速增长,单粒子软错误的威胁日益严重.结果错误(silent data corruption,SDC)是单粒子软错误造成的一种故障类型.由于SDC是隐蔽传播的,SDC的检测是单粒子软错误防护的难点.寻找SDC脆弱指令是目前检测SDC的重要途径.现有方法需要进行巨量的错误注入,时间代价巨大.首先根据数据关联图建立了指令的数据依赖关系,研究了函数间和函数内部错误传播过程;进而推导出判定SDC脆弱指令的充分条件,提出了SDC脆弱指令识别方法,该方法在错误注入中依据充分条件推测潜在的SDC脆弱指令.实验表明,在保证较高准确率和覆盖率的前提下,时间代价显著减少.
[Abstract]:Single particle soft error is the main factor that affects the reliability of calculation in high irradiation space. With the rapid increase of the number of transistors in the chip, The threat of single particle soft error is becoming more and more serious. Result silent data fault is a kind of fault caused by single particle soft error. Detection of SDC is a difficult point of single particle soft error prevention. Order is an important way to detect SDC at present. Existing methods require a huge amount of error injection. The time cost is huge. Firstly, the data dependence of instructions is established according to the data association graph, and the error propagation process between and within functions is studied, and then the sufficient conditions for judging SDC fragile instructions are derived. In this paper, a method of SDC fragile instruction recognition is proposed, which infer the potential SDC fragile instruction based on sufficient conditions in error injection. Experiments show that the time cost is significantly reduced on the premise of high accuracy and coverage.
【作者单位】：东南大学计算机科学与工程学院;计算机网络和信息集成教育部重点实验室(东南大学);中国空间技术研究院总体部;
【分类号】：TP302.8

【相似文献】