Fast source mask optimization adopting mask prediction and feedback method with similarity penalty

Area of Science:

• Computational lithography
• Semiconductor manufacturing
• Image processing

Background:

• Source Mask Optimization (SMO) is vital in computational lithography.
• Enhancing SMO efficiency is critical for advanced semiconductor nodes.
• Gradient descent is a common but improvable SMO technique.

Purpose of the Study:

• To develop a more efficient SMO method for advanced nodes.
• To improve optimization speed without compromising lithographic quality.
• To leverage mask and gradient information more effectively.

Main Methods:

• Introduced a novel SMO method incorporating a similarity penalty term in the loss function.
• Employed a mask prediction feedback (MPF) mechanism.
• Integrated MPF with the similarity penalty term for synergistic benefits.

Main Results:

• The proposed method reduced running time by at least 30% compared to the Adam optimizer.
• Ablation studies confirmed the synergistic effect, outperforming individual components (MPF or similarity penalty).
• Optimization conditions were maintained while achieving significant speedup.

Conclusions:

• The developed SMO method offers a substantial improvement in computational lithography efficiency.
• Combining mask predictive feedback with a similarity penalty term is a highly effective strategy.
• This approach is particularly beneficial for optimizing at advanced semiconductor nodes.