COMPARATIVE ANALYSIS FOR ESTIMATING POWER IN VLSI CIRCUITS USING MACHINE LEARNING MODELS

Abstract

This study explores power estimation in CMOS VLSI circuits through a passive ML- based approach, utilizing various circuit attributes. Based on recent advancements, machine learning (ML) algorithms have become integral to engineering applications for modeling complex systems using historical data. By employing a supervised learning method, the approach ensures rapid and precise power estimation without compromising accuracy. Notably, the XGBoost algorithm emerges as the superior method for power estimation. Experimental outcomes reveal that XGBoost achieves the lowest Mean Squared Error (MSE) and highest R2 score compared to Random Forest and BPNN models. Cross-validation confirms XGBoost's robustness, highlighting its potential as the optimal choice for CMOS VLSI power estimation tasks for ISCAS’89 benchmark circuits.