Decomposition-learning-based thick-mask model for partially coherent lithography system.

Area of Science:

• Semiconductor Manufacturing
• Computational Electromagnetics
• Optical Lithography

Background:

• Accurate aerial image calculation in immersion lithography requires simulating thick-mask diffraction near-fields (DNF).
• Partially coherent illumination (PCI) is crucial for enhancing pattern fidelity in practical lithography tools.
• Existing simulation methods for DNFs under PCI can be computationally intensive.

Purpose of the Study:

• To extend a previously developed learning-based thick-mask model to handle partially coherent illumination (PCI).
• To enable precise and efficient simulation of DNFs under PCI for immersion lithography applications.
• To validate the model's accuracy and computational performance for various critical dimensions (CD).

Main Methods:

• Established a training library of DNFs under oblique illumination using a rigorous electromagnetic field (EMF) simulator.
• Extended a learning-based thick-mask model from coherent to partially coherent illumination conditions.
• Analyzed simulation accuracy against mask patterns with varying critical dimensions (CD).

Main Results:

• The proposed learning-based model achieves high-precision DNF simulations under PCI.
• The model demonstrates suitability for technology nodes of 14nm and larger.
• Computational efficiency is improved by up to two orders of magnitude compared to EMF simulators.

Conclusions:

• The extended learning-based thick-mask model provides accurate and efficient DNF simulations for PCI in immersion lithography.
• This advancement supports the development and optimization of semiconductor manufacturing processes for 14nm nodes and beyond.
• The model offers a significant speedup over traditional rigorous electromagnetic field simulations.