Beyond Limits: Machine Learning Driven Reliability Forecasting for Nanoscale ULSI Circuits

This paper presents a machine learning-based predictive model for negative bias temperature instability (NBTI) degradation in nanoscale PMOS transistors. Through extensive HSPICE Synopsys simulations over a decade, encompassing diverse process and bias conditions, the study elucidates the influence of Gate Oxide, Threshold Voltage, Temperature, and Supply Voltage on NBTI. Multiple machine learning algorithms including Linear Regression, Support Vector Regression, Tree Regression, Ensembles of Trees Regression, Gaussian Process Regression, and Neural Networks are trained and validated. Results showcase more than 99% accuracy in forecasting NBTI degradation, facilitating robust design strategies for VLSI digital circuits in industrial 45nm technology nodes.