Enhancing VLSI Design with Learning Algorithms: A Neural Network Perspective

Abstract

Semiconductor technology is rapidly growing, from small gadgets to big electrical products. Designing an optimal circuit that consumes less power and does optimal work is challenging in this case. To overcome these challenges, we propose a neural network model that can predict the total power consumption of different designs. For this, we have created a new dataset consisting of 8 features, like number gates, input and output ports, etc. Our model consistently predicts power consumption with a mean absolute rate of 1.431788 and an R2 error of 0.995. Moreover, we also made an inference of each feature to find and analyze which feature affects VLSI designs, including cost and power.