The Adoption of Machine Learning in the Measurement of Copper Contact on the Main Chip in Advanced 3D NAND Technology Nodes

Abstract

Metrology measurements of the top copper contact in the main chip area is critical to predict the subsequent electrical performance in advanced 3D NAND technology nodes. Conventional CDSEM is used in the determination of top copper CD while the accurate measurement of copper depth remains challenging. In this paper, we propose a new approach that successfully explored the use of machine learning to combine the advantages of optical Critical Dimension (OCD) and picosecond ultrasonic technology (PULSE™) for high volume manufacturing (HVM) measurements in the main chip area. Results demonstrate that by using machine learning, we were able to combine the PULSE reference with cross-section microscopy results to successfully train the OCD data set. OCD measurements are rapid at <1s/site and meets the HVM need for extensive sampling and allows for in-line process control of the top copper contact height. This approach also opens the possibilities for application of machine learning for in-line 3D NAND monitoring process control by combining multiple methods and exploiting the full potential of each of these technologies.