Reassessing March Test Effectiveness in System Memory Error Detection through Inductive Fault Analysis

초록

At the system level, extensive memory testing is performed throughout all stages of the system lifecycle, including during manufacturing. Most memory testing is primarily based on March algorithms. However, as memory undergoes thorough testing during component manufacturing, repeating similar tests at the system level is not expected to be efficient. This work examined the effectiveness of March tests at the system level and experimentally demonstrated their limitations. Cisco Systems’ C480 M5 server was used as the target system, with memory errors collected from a memory module. A total of 23 March algorithms, covering most of the commonly used ones, were performed. To explore the correlation between system memory errors and the fault models (FMs) targeted by these algorithms, inductive fault analysis (IFA) was employed. A newly proposed metric, the FM correlation probability, was introduced to quantify this correlation. This metric was used to evaluate the correlation of 8 faulty system addresses (FSAs) against the 62 fault primitives (FPs) across the 9 FMs. No FSA exhibited a FM correlation probability greater than 30%, and none achieved 100% correlation with all 62 FPs. Additionally, March tests were also performed in different socket positions to see whether the observed trends are similar in a different system environment. © 2013 IEEE.

키워드