Annachiara Ruospo
Reliability challenges and solutions in the new age of artificial intelligence-powered devices
In the technological landscape, AI models are evolving rapidly, reaching unexpected levels of accuracy and human-like capabilities. They are even more attractive to safety-critical systems, particularly in the automotive and aerospace sectors, where the imperative for dependability is paramount. With the shrinking of technology nodes, today’s hardware devices are more prone to hardware-induced failures, and the literature shows that their effects cannot be neglected, as even one of them can cause the application to fail. This, combined with the complexity of current software AI models, is making the assessment of their reliability and fault tolerance a real challenge. Therefore, it is necessary to propose innovative solutions for reliability assessment and improvement. This talk will present the latest proposed solutions for assessing the resilience of AI-powered devices by exploiting statistical methods and accurate error modeling. Then, the presentation will describe novel in-field testing techniques to detect the occurrence of transient and permanent faults during the operational phase of the device, focusing on software test libraries (STLs) and the recently introduced image test libraries (ITLs). Overall, the talk will summarize the most recent contributions in this field by the CAD & Reliability Group within the Dept. of Control and Computer Engineering of Politecnico di Torino.
back to overview
