Untestable faults identification in GPGPUs for safety-critical applications
Josie E.Rodriguez Condia (Politecnico di Torino)
F. Augusto da Silva (TU Delft - Computer Engineering, Cadence Design Systems)
S. Hamdioui (TU Delft - Quantum & Computer Engineering)
Christian Sauer (Cadence Design Systems)
Matteo Sonza Reorda (Politecnico di Torino)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Nowadays, General Purpose Graphics Processing Units (GPGPUs) devices are considered as promising solutions for high-performance safety-critical applications, such as those in the automotive field. However, their adoption requires solutions to effectively detect faults arising in the device during the operative life. Hence, effective in-field test solutions are required to guarantee high-reliability levels. In this paper, we leverage the results of Software-Based Self-Test (SBST) based approaches for GPGPUs by deploying new techniques for automating the identification of untestable faults (UF). Our methodology has achieved fault coverage of 82.8% when applied to an open-source implementation of the NVIDIA G80 GPU architecture. The proposed approach combining SBSTs and UFs identification appears as an effective solution for the reliability analysis of GPGPUs.