Using STLs for Effective In-field Test of GPUs

Reliability and functional safety are of primary importance for Graphics Processing Units (GPUs) widely used in safety-critical fields such as automotive and robotics. Moreover, cutting-edge technologies used in modern GPUs make these even more susceptible to in-field permanent faults due to aging and degradation; hence, the rising need for effective in-field test solutions to ensure the deployment of appropriate countermeasures. This work demonstrates that Self-Test Libraries (STLs) can be written and effectively applied for the in-field detection of permanent faults in a GPU. Simulations using an open-source GPU model (FlexGripPlus) confirm the validity and effectiveness of STLs. The results show that a fault coverage of 92.6% can be achieved, allowing at least ASIL B level for functional safety inside a GPU core.