Power Estimation of Synchronous Sequential VLSI Circuits Using Boosting Techniques

Abstract

Power has a notable influence on the functionality and reliability of the Very Large-Scale Integration (VLSI) circuits. Thus, estimation of consumed power at an initial phase is extremely necessary. This paper describes the comparative study of supervised ensemble based boosting machine learning techniques to predict synchronous sequential VLSI circuits. We implemented three supervised ensemble based boosting learning algorithms for power estimation: Adaptive Boosting (AdaBoost), Gradient Boosting (GB) and Extreme Gradient Boosting (XgBoost). Ensemble boosting techniques are tuned by using Grid Search and Random Search hyper-parameter optimization techniques. The ensemble based boosting techniques are applied on IEEE ISCAS’89 benchmark circuits. The coefficient of determination (R) and Root Mean Squared Error (RMSE) are the statistical parameters. These statistical parameters are calculated for each boosting algorithm. The experimental results show that gradient boosting with grid search hyper-parameter optimization approach is a strong preference for predicting the power of synchronous sequential VLSI circuits. Since it has remarkable coefficient of determination of 0.99746 and lower RMSE of 3.143e-5.