Vmin Shift Prediction Using Machine Learning-Based Methodology for Automotive Products

Predicting aging behavior is essential for product development to guarantee in-field lifetime. Conventionally, aging margin is determined by identifying the maximum shift value of minimum operating voltage (Vmin) through a series of high-temperature operation lifetime (HTOL) tests. In this paper, we propose a novel approach that leverages the machine learning (ML) techniques to predict Vmin shifts before conducting the HTOL test. Compared to the conventional fixed aging margin, this ML-based methodology offers the adaptive aging margins on voltage groups, resulting in significant power savings. The reduction in the aging margin is estimated to be > 20%. In addition, this proposed methodology enables the use of more sensitive monitors for detecting reliability degradation compared to the on-chip NAND and NOR based RO. In our experiment, the ML-derived monitor demonstrated the 3x sensitivity to negative bias temperature instability (NBTI) than NOR.