Fast aerial image model for EUV lithography using the adjoint fully convolutional network.

Area of Science:

• Semiconductor Manufacturing
• Optical Lithography
• Deep Learning Applications

Background:

• Extreme ultraviolet (EUV) lithography is crucial for advanced semiconductor manufacturing.
• Calculating aerial images in EUV systems is challenging due to thick-mask and oblique incidence effects.
• Accurate aerial image modeling is essential for process control and yield optimization.

Purpose of the Study:

• To develop a fast and accurate aerial image model for EUV lithography.
• To address the computational challenges posed by thick-mask and oblique incidence effects.
• To leverage deep learning for enhanced lithography process modeling.

Main Methods:

• Development of a novel Adjoint Fully Convolutional Network (AFCN) deep learning framework.
• Utilizing two adjoint data paths to recover real and imaginary parts of the complex mask diffraction-near-field (DNF).
• Employing a feature-swapping technique for information exchange between data paths and training on a rigorous thick-mask DNF dataset.

Main Results:

• The AFCN model significantly reduces calculation errors by over 80% compared to traditional methods.
• Achieved a 25-fold improvement in computational efficiency for aerial image calculation.
• Demonstrated the effectiveness of the deep learning approach in handling complex EUV lithography conditions.

Conclusions:

• The proposed AFCN model offers a highly efficient and accurate solution for EUV lithography aerial image calculation.
• This deep learning framework effectively overcomes the limitations of traditional methods in complex lithography scenarios.
• The study paves the way for faster and more precise process development in advanced semiconductor manufacturing.