透過您的圖書館登入
IP:44.220.131.93
  • 學位論文

透過強化視覺問答模型進行微影製程熱點偵測之佈局圖案擷取

Layout Pattern Extraction through an Enhanced Visual Question Answering Model for Lithographic Hotspot Detection

指導教授 : 江蕙如
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


隨著半導體製程技術快速進步,積體電路佈局圖案受製程變異影響益發嚴重。即使存在許多解析度增強技術,探索有效的微影製程熱點佈局圖案並盡早修正對於確保晶片的良率和可製造性來說極為重要。因此,微影熱點偵測已成為半導體製程實體驗證的重要步驟。熱點圖案經過微影模擬後能進一步確知並分類出造成不同製造錯誤的熱點型態。當代先進製程設計佈局中的微影熱點不僅與單一特定佈局層相關,更是多佈局層的總和效應的結果。然而,現有微影熱點偵測的方法僅處理單層(至多包含相鄰層)電路佈局,並且無法將偵測之佈局圖案與相關熱點型態連結。在本篇論文中,我們著眼於偵測微影熱點關鍵佈局圖案並確認所屬熱點型態來改善實體驗證覆蓋範圍。我們首先提出將多層微影熱點關鍵佈局圖案擷取建模成視覺問答 (Visual Question Answering) 的問題,將多層佈局圖案作為視覺資料,以及微影熱點型態作為問題,我們提出的模型可以回答此多層佈局圖案對於詢問的微影熱點型態是否關鍵。我們更進一步提出根據微影熱點型態對佈局層注意力機制。實驗結果證明,我們的模型具有相當高的準確性和問答能力,並且有能力從超過三十層的先進製程佈局資料庫中進一步識別出從未考慮過的潛在的微影熱點佈局圖案。

並列摘要


Along with the rapid advance of semiconductor process technologies, layout patterns become highly susceptible to lithography process fluctuations. Even equipped with a variety of resolution enhancement techniques, exploring effective lithographic hotspot patterns and correcting them as early as possible is extremely crucial to guarantee chip yield and manufacturability. Lithographic hotspot detection is thus a key step of physical verification of semiconductor process technology. Through lithography simulation, lithographic hotspot patterns are confirmed and classified into different hotspot types based on induced manufacturing failures. Hotspots in modern layouts are related to not only one specific layer but also the accumulated influence over many layers. Existing hotspot detection methods, however, handle only single-layer (or with at most two adjacent layers) layout patterns, and they cannot recognize the corresponding hotspot type for a hotspot pattern. In this thesis, we focus on a new lithographic hotspot detection challenge; our goal is to extract critical hotspot patterns from modern design database and identify corresponding hotspot types for improving the coverage of physical verification. We first propose to model the many-layer critical hotspot pattern extraction task as a visual question answering (VQA) problem: Considering a many-layer layout pattern as an image and a hotspot type as a question, we devise an enhanced VQA model to answer whether the pattern is critical to the queried hotspot type. We further design a layer attention mechanism to identify the importance and relevance of each layer. Experimental results show that the proposed model has superior accuracy and question-answering ability, and it can further identify unseen hotspot patterns from sub-14nm layouts with more than thirty layout layers.

參考文獻


[1] P. Anderson, X. He, C. Buehler, D. Teney, M. Johnson, S. Gould, and L. Zhang, "Bottom-up and top-down attention for image captioning and visual question answering," in Proceedings of Conference on Computer Vision and Pattern Recognition (CVPR), 2018.
[2] S. Antol, A. Agrawal, J. Lu, M. Mitchell, D. Batra, C. L. Zitnick, and D. Parikh, "VQA: Visual question answering," in Proceedings of International Conference on Computer Vision (ICCV), 2015.
[3] W.-C. Chang, I. H.-R. Jiang, Y.-T. Yu, and W.-F. Liu, "iClaire: A fast and general layout pattern classication algorithm," in Proceedings of 54th Design Automation Conference (DAC), 2017.
[4] X. Glorot and Y. Bengio, "Understanding the difficulty of training deep feed-forward neural networks," in Proceedings of 13th International Conference on Arti cial Intelligence and Statistics (AISTATS), 2010.
[5] S. Hochreiter and J. Schmidhuber, "Long short-term memory," Neural Computation, vol. 9, no. 8, pp. 1735-1780, 1997.

延伸閱讀