AI-Assisted Verification and Reliability Enhancement in Complex VLSI Systems: A Data-Driven Framework

Abstract

The increasing complexity of the current VLSI systems has revealed the basic constraints
of traditional reliability analysis methodologies as well as verification. Simulation-based
verification has a hard time getting sufficient coverage within realistic time frames whereas
the static workload reliability models do not model dynamic interaction between the
behaviour of workloads, process variability, and environmental stress. The presented paper
suggests a data-based AI-supported framework of verification and reliability improvement in
complicated VLSI systems. The framework uses machine learning and reinforcement learning
to exploit the low-coverage and high-risk design states efficiently, which facilitates the
detection of corner-cases. Simultaneously, a time-dependent failure prediction model (based
on an AI) is trained to predict failure rates that significantly depend on verify data along with
thermal and aging-induced stress. Experimental comparison between representative system
on-chip and AI accelerator giving shows better results of functional coverage, efficiency of
simulation, and predicted reliability than traditional and control learning-assisted techniques.
This is because the results reveal that AI-controlled verification flows could be used to design
next-generation VLSI systems in establish able and reliably aware design automation.