Power Estimation of Digital VLSI Circuits Using Machine Learning

For any VLSI design, accurate power estimation is significant to make it optimized. Conventional approaches for power estimation are based on some analytical models and involve simulation, which is typically computationally expensive and time-consuming depending on the level of abstraction where it is performed. As an alternative machine learning-based approaches have gained popularity due to their ability to predict power consumption with higher accuracy and speed. In this paper, we presented a comparative analysis of four best performing machine learning-based approaches for power estimation of digital circuits. The dataset of various digital circuits was prepared based on the features collected from the synthesis reports obtained by Quartus tool. The Root Mean Squared Error (RMSE) and R-squared values are used to compare the performance of the models. In our results the best values of R-squared and RMSE obtained are 0.99 and 0.29581 respectively.