3D-IC Power Distribution Network (PDN) Impedance Prediction using Deep Learning (DL)

In this paper, we propose a 3D-IC PDN impedance prediction method using deep learning (DL). Unlike conventional application process (AP) and printed circuit board (PCB) design, our method proposes an effective solution to the problem of PDN design of 3D-IC, which requires accurate and fast impedance evaluation. For prediction, we construct regression neural network model based on VGG-16 with additional layers. Additionally, boundary element method (BEM) and Voronoi diagram are used to generate training data. We verified the proposed method in terms of accuracy and time. The impedance can be predicted nearly 4 times faster than ground truth which calculated by BEM with a low error.