利用AdaBoost的SDC错误检测方法  

SDC Error Detection Method Based on AdaBoost

作　　者：刘阳 庄毅[1] LIU Yang;ZHUANG Yi(Collaborative Innovation Center of Novel Software Technology and Industrialization,College of Computer Science and Technology,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China)

机构地区：[1]南京航空航天大学计算机科学与技术学院,软件新技术与产业化协同创新中心,南京211106

出　　处：《小型微型计算机系统》2022年第4期882-888,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金面上项目(61572253)资助;航空科学基金项目(2016ZC52030)资助。

摘　　要：单粒子翻转引起的静默数据损坏(SDC)错误是影响嵌入式系统可靠性的关键因素.针对SDC错误难以检测的问题,本文提出了一种基于AdaBoost的SDC错误检测方法.方法首先将物理空间中的单粒子多位翻转映射到单一指令中,并建立故障模型;然后根据故障模型对目标程序进行故障注入以获取故障注入仿真实验结果,使用AdaBoost算法训练指令SDC脆弱性模型;最后根据模型预测结果对目标程序进行粒度可配置的指令冗余以达到检测SDC错误的目的.实验结果表明,与现有方法相比,本文提出的利用AdaBoost的SDC错误检测方法有着较高的检测率和较低的时空开销.Silent data corruption(SDC)error caused by single event upset(SEU)is a key factor affecting the reliability of embedded systems.To solve the problem of SDC error detection,this paper proposes a SDC error detection method based on AdaBoost.The method first maps the single-particle multi-bit flip in physical space to a single instruction and builds a fault model;then performs fault injection to the target programs according to the fault model to obtain relevant data and uses the AdaBoost algorithm to train the instruction SDC vulnerability model;and finally performs granularity-adjustable instruction redundancy to the target program according to the model prediction results to detect SDC errors.Experimental results show that compared with the existing methods,the proposed SDC error detection method based on AdaBoost has higher detection rate and lower space-time overhead.

关 键 词：静默数据损坏 错误检测 ADABOOST 多位翻转 指令 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]