A Novel CNN-Based BEOL Contour Modeling Solution for Parasitic R/C Extraction

In this article, we propose a machine-learning (ML) based wafer contour approach for parasitic R/C modeling, leveraging a convolutional neural network (CNN) architecture. We demonstrate that our NN critical dimension (CD) model outperforms traditional rule-based approaches in predicting metal contours. With this capability, our NN RC-model exhibits excellent accuracy in predicting parasitic R/C. Furthermore, our NN RC-model is more efficient than the conventional 3D field solver tools. This demonstration manifests the possibility of an electrical-aware pattern optimization, enabling the cooptimization from layout design to final electrical performance through the CNN-based CD/RC algorithm.