Power Estimation in VLSI Circuits using Supervised Machine Learning

Power estimation plays a pivotal role in the design and optimization of VLSI circuits to meet the ever-growing demands for energy efficiency. This research explores pre-synthesis and post-synthesis power estimation using advanced machine learning models, with a particular focus on supervised learning techniques. A comprehensive analysis was conducted on multiple machine learning models, including CatBoost, LightGBM, XGBoost, Random Forest, and others, utilizing benchmark EPFL circuits. The results indicate that CatBoost outperforms all other models, achieving the lowest Mean Squared Error (MSE) and a near-perfect R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> score, demonstrating its exceptional capability for accurate power estimation. LightGBM follows closely, offering competitive performance. This study emphasizes the importance of integrating machine learning approaches for efficient and accurate power estimation, paving the way for future advancements in low-power VLSI design. The presented methodology and findings establish a robust framework for incorporating intelligent models in the circuit design flow, highlighting the potential of CatBoost as a reliable tool for power estimation.