B-SEFI:A Binary Level Soft Error Fault Injection Tool

Ying Wang,Jian Dong,Sen Zhang,Decheng Zuo
DOI: https://doi.org/10.1109/qrs-c.2019.00053
2019-01-01
Abstract:Soft errors are becoming more prominent in modern computing systems due to the increasing integration of chips. These faults pose a major challenge for memories and logic circuits in high performance microprocessors. In this paper, a tool based on Pin was designed to simulate soft error, which can implement hardware fault injection at the machine code level. Our tool is based on binary instrumentation that supports an accurate and low-cost fault injection. We apply the tool to five classic machine learning programs and analyze the programs' weakness to soft errors by simulating bit flips. We investigate the soft fault effect on performance and accuracy according to experimental results. Our study reveals that these programs are highly sensitive to soft errors, therefore, the tool provides a new effective approach for future resiliency research.
What problem does this paper attempt to address?