Offline Power Estimation in CMOS VLSI Circuits Using Machine Learning Model

Accurate power estimation in CMOS VLSI circuits is vital for energy efficiency, performance optimization, and thermal management. This paper introduces a novel machine learning-based method for power estimation using a Random Forest model, with activity factor $(\alpha)$ as a crucial parameter. We propose a method for predicting power consumption by analysing circuit parameters and power consumption reports. The prediction accuracy is validated by comparing simulated and predicted values. For internal and total power, an activity factor $(\alpha)$ value of 0.1 yields better overall prediction accuracy compared to an $\alpha$ value of 0.4. Nevertheless, using the default $\alpha$ value, the predicted values closely approximate the simulated results. Early power estimation is essential for design engineers to make informed decisions during the design phase, ensuring efficient power management and reliable circuit performance. Our ML-based approach achieves accuracy comparable to traditional techniques and enhances computational efficiency, providing a valuable tool for the design process.