DNN-based Predictive Digital Twin for FHE Manufacturing

This work presents a novel predictive model and a virtual metrology for Flexible and Printable Electronic (FHE) device fabrication using deep learning techniques. Our method leverages the Deep Neural Network (DNN) based model Pix2Pix to predict outcomes of 3D inkjet printing from layout images. We prepared an experimental FHE dataset that captures printed electronics’s intricate patterns and printing process variability. By training the Pix2Pix model on this dataset, we successfully demonstrated its ability to learn and accurately predict the outcome of inkjet printing of structures using a conductive ink. The best mean absolute errors (MAE) for width and gap prediction are 5.07 microns and 7.99 microns. This approach enables improvement of efficiency and accuracy of printed electronics, paving the way for advancements in FHE manufacturing technology and workflows.