抗内存位翻转的软硬件协同检错纠错方法  

A Software/Hardware Cooperation EDAC Method against Bit-Flips of Memory

作　　者：黄江洪[1,2] 向阳霞[3] 章立生[1] 谢应科[1] 韩承德[1] 

机构地区：[1]中国科学院计算技术研究所,北京100190 [2]中国科学院研究生院,北京100049 [3]装甲兵工程学院信息工程系,北京100072

出　　处：《武汉大学学报（理学版）》2009年第1期35-40,共6页Journal of Wuhan University:Natural Science Edition

基　　金：国家自然科学基金资助项目(60303017)

摘　　要：针对硬件实现的纠1位错检2位错(SEC-DED)的编码用于内存的保护时纠错能力有限的问题,提出了一种软硬件协同的检错纠错(EDAC)编码方法.在硬件对水平方向的每个字进行SEC-DED编码的基础上,该方法在软件上对垂直方向的每个位片进行奇偶编码,并通过软硬件协同的方式来发现并纠正内存位翻转错误.相对于纯硬件的EDAC方法而言,该方法能纠正更多的内存位翻转错误,并且无需增加硬件资源开销.理论分析和实验结果表明,内存在软硬件协同EDAC保护下的可靠性比在纯硬件EDAC保护下的可靠性具有一定程度的提高.When hardware-implemented SEC-DED（single error correcting,double error detecting） code is used for protecting memory,its capability of correction is limited.In order to improve the capability of correction,this paper proposes an EDAC（error detection and correction） coding method with software/hardware cooperation.Based on hardware-implemented SEC-DED coding for each horizontal word,this method makes software-implemented parity coding for the vertical bit-slice,and corrects bit-flip errors of memory through software/hardware cooperation.In comparison with the hardware-only EDAC method,the proposed method can correct more bit-flip errors without increasing hardware cost.Theoretic analyses and experimental results show that,the reliability of the memory protected by the software/hardware cooperation method is higher than the reliability of memory protected by the hardware-only method to some extent.

关 键 词：检错纠错 软硬件协同 位翻转 可靠性 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]