NiCE-sim：一個使用Navier-Stokes和濃度方程的非均勻旋轉微影模擬器

Abstract

2023年GTC 2023 Keynote Nivida展示了新技術culitho，這項科技可以幫助IC製造公司在光罩製作以及圖案投影，在這項技術還沒有被實現時，根據Nivida公布執行這項模擬會消耗百億個cpu小時，資料中心需要全年無休來做光罩產生。由此可知，隨著製程3奈米節點後繼續縮小，IC製造公司越是需要需要更精確且多種物理及化學模擬，預測IC製造時可能會發生的問題，才有辦法在現實中做一些補償，以降低生產成本。由於模擬實際現象需要多種物理方程式，本論文只針對曝光烘乾後的流程，也就是微影溶解。在我們的研究中，利用finite difference method 離散化Navier-stokes equation 以及concentration equation，並利用迭代法加速求解稀疏線性系統，可以看到我們的結果在輸入資料量很大時，記憶體使用量以及運算時間有顯著的效果。

In the 2023 GTC Keynote, Nvidia presented a new technology called culitho, which can assist IC manufacturing companies in mask fabrication and pattern projection. Before the implementation of this technology, according to Nvidia's disclosure, performing simulations for these purposes would consume billions of CPU hours, requiring data centers to operate continuously throughout the year for mask generation. This indicates that as the process node continues to shrink beyond the 3-nanometer threshold, IC manufacturing companies increasingly require more precise and diverse physical and chemical simulations to predict potential issues during IC fabrication. These simulations enable them to make compensations in the real world, ultimately reducing production costs. Given that simulating real-world phenomena involves multiple physical equations, this paper focuses solely on the post-exposure baking process, specifically lithography dissolution. In our research, we discretize the Navier-Stokes equation and concentration equation using the finite difference method. We employ an iterative method to accelerate the solution of the sparse linear system. Our results demonstrate significant improvements in memory usage and computation time when dealing with large input datasets.