Fast lithography aerial image calculation method based on machine learning

Area of Science:

• Optics
• Computational Physics
• Semiconductor Manufacturing

Background:

• Aerial image calculation for thick masks is crucial for lithography simulations but is computationally intensive.
• Existing methods face challenges in speed and accuracy, particularly for partially coherent systems.

Purpose of the Study:

• To develop a rapid and accurate aerial image calculation method for thick masks in partially coherent lithography.
• To leverage machine learning to accelerate a critical step in lithography simulation.

Main Methods:

• Utilized sparse sampling of the source plane to represent partially coherent illumination.
• Developed training libraries of thick-mask diffraction near-fields using representative mask features.
• Employed nonparametric kernel regression and a post-processing technique for aerial image calculation and error compensation.

Main Results:

• The proposed machine learning method significantly reduces computation time for thick-mask aerial image calculations.
• The method demonstrates high accuracy, validated through simulations at 45 nm and 14 nm technology nodes.
• Analysis provided insights into key factors affecting the accuracy and efficiency of the developed technique.

Conclusions:

• The developed fast thick-mask aerial image calculation method based on machine learning offers a viable solution for accelerating lithography simulations.
• This approach enhances computational efficiency without compromising accuracy, crucial for advanced semiconductor process development.
• The study validates the effectiveness of machine learning in addressing computational bottlenecks in photolithography simulation.