Machine Learning Approaches for Variability Analysis in Integrated Circuits

This paper discusses the performance of various regression-based machine learning approaches for the variability analysis of integrated circuits. The considered approaches include Support Vector Machine, Least Square-Support Vector Machine, Random Forest and Gaussian process regression. The efficacies of these approaches are demonstrated using two circuit examples- a CMOS LC oscillator and a low-noise amplifier. Here, the performance of all the approaches in the presence and absence of numerical noise is compared to the state-of-the-art methods such as sparse Polynomial Chaos expansion and Monte Carlo analysis to provide a complete overview. This study guides the readers to choose an appropriate learning model for similar applications of variability analysis of the integrated circuits.