Ensemble Model for Power Estimation of CMOS Circuits

Abstract—Accurate power estimation in complementary metaloxide-semiconductor (CMOS) very largescale integration (VLSI) design is vital for optimizing performance and reducing energy consumption. Traditional methods like SPICE simulations are computationally expensive and time-intensive, necessitating more scalable solutions. To overcome these limitations, an ensemble approach using machine learning (ML) and deep learning (DL) models to enhance power prediction accuracy has been proposed in this work. Different model combinations, such as random forest (RF) with extreme gradient boosting (XGB) and support vector regression (SVR) with convolutional neural network (CNN), were developed and evaluated. Among these, the SVR and CNN ensemble model achieved the highest performance, with an accuracy of 97.12%, a mean squared error (MSE) of 0.3705, and an R² score of 0.9634. The CNN component was employed for automatic feature extraction, while SVR handled precise regression, with weight allocation optimized using R²- based optimization. The results emphasize the effectiveness of combining DL with regression models for power estimation, illustrating the promising role of ensemble framework in advancing CMOS VLSI design methodologies.