Synthetic mask layout generation for machine learning-based OPC modeling in NA 0.33 EUV lithography

Abstract

As NA 0.33 EUV lithography operates increasingly close to fundamental resolution limits, accurate and robust optical proximity correction (OPC) models become critical for advanced logic technology development. In early technology phases, OPC calibration is challenged by limited availability of real product layouts. To address these challenges, this work integrates Layout Schematic Generation with machine learning (ML) based OPC modeling to enhance calibration pattern diversity for 28nm metal pitch in single‑patterning NA 0.33 EUV lithography. Synthetic layouts are generated from a limited set of manufacturable ground rules. The generated patterns are fabricated using a bright‑field low‑n attenuated phase‑shift mask and metal‑organic resist with dry development. ML OPC model experiments demonstrate that incorporating LSG‑generated layouts leads to systematic improvements in prediction accuracy across multiple pattern categories, with root‑mean‑square (RMS) error reductions ranging from approximately 28% to 15% compared with conventional calibration strategies. The proposed LSG‑driven calibration framework provides an efficient and scalable solution to improve OPC model robustness for NA 0.33 EUV lithography and is directly applicable to future high‑NA EUV technology nodes.