High-precision lithography thick-mask model based on a decomposition machine learning method.

Area of Science:

• Optical lithography
• Computational electromagnetics
• Semiconductor manufacturing

Background:

• Traditional thick-mask models in optical lithography face challenges in accurately simulating 3D photomask diffraction behavior.
• Edge interference effects in the diffraction near-field (DNF) are critical but often simplified in existing models.

Purpose of the Study:

• To propose an improved thick-mask model with high precision for simulating 3D photomask diffraction.
• To enhance the accuracy of optical lithography simulations by incorporating detailed DNF analysis.

Main Methods:

• Introduction of a diffraction transfer matrix (DTM) to map layout patterns to DNF.
• DTM learning from a training library of rigorous DNF for representative mask segments.
• Decomposition of thick-mask patterns into segments, local DNF calculation using DTM, and synthesis of local DNF segments.

Main Results:

• The proposed thick-mask model significantly improves simulation accuracy compared to traditional filter-based methods.
• The method achieves high computation speed, making it efficient for complex simulations.
• Accurate simulation of the entire thick-mask DNF is achieved through segment synthesis.

Conclusions:

• The improved thick-mask model provides a more precise and computationally efficient approach for optical lithography simulations.
• Accurate modeling of DNF, particularly edge interference effects, is crucial for advanced semiconductor manufacturing.
• This DTM-based method offers a promising solution for simulating complex 3D photomask diffraction patterns.