Lin Jiaxin, Dong Lisong, Fan Taian, Ma Xu, Chen Rui, Wei Yayi
Appl Opt. 2020 Mar 20;59(9):2829-2838. doi: 10.1364/AO.384407.
Near-field calculation for a three-dimensional (3D) mask is a fundamental task in extreme ultraviolet (EUV) lithography simulations. This paper develops a fast 3D mask near-field calculation method based on machine learning for EUV lithography. First, the training libraries of rigorous mask near fields are built based on a set of representative mask samples and reference source points. In the testing stage, the mask under consideration is first segmented into a set of non-overlapped patches. Then the local near field of each patch is calculated based on the non-parametric regression and data fusion techniques. Finally, the entire mask near field is synthesized based on the image stitching and data fitting methods. The proposed method is shown to achieve higher accuracy compared to the traditional domain decomposition method. In addition, the computational efficiency is improved up to an order of magnitude compared to the rigorous electromagnetic field simulator.
三维（3D）掩膜的近场计算是极紫外（EUV）光刻模拟中的一项基础任务。本文针对EUV光刻开发了一种基于机器学习的快速3D掩膜近场计算方法。首先，基于一组具有代表性的掩膜样本和参考源点构建严格掩膜近场的训练库。在测试阶段，将所考虑的掩膜首先分割成一组不重叠的小块。然后基于非参数回归和数据融合技术计算每个小块的局部近场。最后，基于图像拼接和数据拟合方法合成整个掩膜近场。结果表明，与传统的区域分解方法相比，所提方法具有更高的精度。此外，与严格的电磁场模拟器相比，计算效率提高了一个数量级。
