Double U-Net based Virtual Metrology on Plasma-Etch CD-SEM Images : AM: Advanced Metrology

Optimizing lithography and plasma etching is the key for manufacturing nanoelectronics and Micro-Electromechanical Systems (MEMS). We propose a Deep Learning (DL) based virtual metrology algorithm for CD-SEM images that addresses the background noise, the low contrast edges, and the discrepancies between design and wafer that occur in such images and achieves automatic and accurate segmentation that is robust to changes in structure and process recipe. Accurate segmentation of CD-SEM images of the nanostructures provides the flexibility to make a very diverse set of measurements. The results show that the DL applied to CD-SEM images significantly improves the accuracy and robustness of virtual metrology.