Liu, Ming-JieMing-JieLiuHuang, Yu-TangYu-TangHuangLIANG-CHIA CHENSendelbach, Matthew J.Schuch, Nivea G.2025-07-072025-07-072025-02-24https://www.scopus.com/record/display.uri?eid=2-s2.0-105007140723&origin=resultslisthttps://scholars.lib.ntu.edu.tw/handle/123456789/730606Optical image metrology is vital in semiconductor manufacturing, offering non-contact, high-resolution, high-speed measurements. This AI-driven approach integrates CNN with a physics-based 3D point spread function (PSF) for enhanced critical dimension (CD) measurement. A focused ion beam (FIB) artifact captures the PSF, while imaging noise is minimized using a parameterized 3D PSF from the Gibson-Lanni model with Zernike polynomials. A super-resolution CNN with Wiener deconvolution refines resolution. Preliminary experimental results show this method doubles the depth of field (DOF) and enables nanometer-accurate CD measurement, improving semiconductor manufacturing and automated optical inspection (AOI).Advanced semiconductor packagingAI deep learningAutomated Optical Inspection (AOI)Critical Dimension (CD)Optical MetrologySuper resolutionconference paper10.1117/12.3051689