Modeling an extreme ultraviolet lithography mask by a deep fully convolutional network

Unmet Need

Extreme ultraviolet (EUV) lithography is a patterning method that is at the forefront of semiconductor fabrication technology. EUV lithography uses wavelengths of light that are as small as 13 nm to produce incredibly high-resolution features. However, a key challenge in achieving high quality lithographic patterning is accurately modeling the electromagnetic scattering of EUV light as it passes through the lithography mask. There is a need for a sophisticated model that can accurately determine the scattered fields for any given mask without requiring intensive computation or simulation.

Technology

Duke inventors have developed a machine learning method that can accurately predict the scattering of light in EUV lithography three orders of magnitude faster than similar simulation tools. This is intended to be used during semiconductor manufacturing by mask designers to ensure precise patterning of nanometer-scale features through the mask while accounting for unintentional light scattering. Specifically, a deep fully convolutional network has been trained on scattering data determined from known mask unit cells to determine the relationship between light scattering and mask patterns. This method has been demonstrated to accurately predict the scattering fields for complex mask patterns as small as 128 nm by 128 nm and as large as 4000 nm by 4000 nm ensuring a wide range of applicability to semiconductor manufacturing.

Advantages

• Low computational requirements for analysis
• Can accurately predict the patterning of electromagnetic fields for EUV lithography
• Offers fast and precise simulation that outperforms other computational methods