在半導體製造中，光罩(Mask)上的線寬及間距不斷縮小，當逼近所使用之深 紫外光波(KrF、ArF氣體雷射)的成像解析度極限時，由於光刻(Photolithography)中 光線干涉及繞射的行為更加顯著，加上蝕刻中化學反應過程內含的不確定因素， 使得這些微影製程後的電路產品與原本的設計圖樣間􏰀在失真。這些從設計圖樣 到微影成像電路中的製程變異(Process variation)，目前乃是製程技術進步的一大 瓶頸，若能有效掌握此種形變的預估或模擬，則能預判出圖樣形變後相應可能產 生的問題，對於電晶體品質的控制將有相當大的提升。
對於此形變偏差，現今常見的應對做法，是利用對形變的預估而在設計圖 上進行預先補強，例如光學接近修正(Optical Proximity Correction)即是建立光學模 型進行輪廓模擬(Contour simulation)，並迭代地修正設計樣式使曝光後的電路盡 量接近要求。然而，因微影成像過程中􏰀在各式變因，要全面地考量各種光學 及化學的因素所帶來的影響實屬困難，且有限的計算資源大幅限縮了所建光學 模型的複雜度，在建模的資料收集也僅能針對簡易的圖樣，這些限制都降低了 光學接近修正系統中對變形模擬的實用性，使其在時間成本的考量下不敷使用。
在本篇論文中，我們將提出一套以卷積神經網路模型作為工具的微影成像模 擬方法，透過數據驅動化(Data-Driven)的類神經網路學習，在效率及泛用性上能 有所提昇，並在拍攝電子掃描式顯微鏡影像的成本高昂之前提下，分析如何達 到最有效的主動式學習(Active learning)。我們嘗試以不同演算法將電路設計圖進 行分群(Clustering)，主動找出其中對網路訓練最具意義的電路樣式作為訓練資料 集，即分析如何用最少的輸入(電路設計圖)—輸出(電路之電子掃描式顯微鏡拍 攝影像)訓練資料對，達成最好的變形預估成效。最後，我們也提供了神經網路 的視覺化分析，希望能讓使用者對這個黑盒子般的工具有更多的了解，增加人 工智慧用於半導體產業研究的適用分析。

In semiconductor manufacturing, with the minimization of width or space on our design pattern, the light interference and diffraction play more significant role while performing photolithography, which can lead to a deformation between the exposed pattern and the original mask, further cause open or short situation on integral circuit. These process variations had become the primary bottleneck of manufacturing, hence well-prediction of these deformation is crucial to the improvement of quality control.
Dealing with the distortion, Optical Proximity Correction (OPC) has been developed to correct the layout in advance, such that the exposed result would differ from the pattern we want as less as possible. However, a rule-based mathematical model has a limitation of efficiency, iteratively simulation on a new layout pattern can be very time-consuming. Furthermore, considering every effect of lithographic and chemical is almost impossible for a manually design model.
In this paper, we present a Convolutional Neural Networks (CNNs) based model to perform the simulation of pattern distortion during photolithography, which can be more efficient and robust compared to conventional simulators. Furthermore, under the premise of limited ground truth labeling resource, which in our case, taking Scanning Electron Microscopy (SEM) photos of the etched circuit, we analysis the low-shot learning of our simulator, which will perform by clustering the layout pattern to actively choose the most meaningful data pairs for training. Last but not least, we provide the network visualization analysis, hope to gain more understanding about this mysterious black box.

[1] B. Settles, “Active learning literature survey,” University of Wisconsin– Madison, Computer Sciences Technical Report 1648, 2009.
[2] D. D. Lewis, “A sequential algorithm for training text classifiers: Corrigendum and additional data,” SIGIR Forum, vol. 29, no. 2, pp. 13–19, 1995.
[3] S. Tong and D. Koller, “Support vector machine active learning with applications to text classification,” Journal of Machine Learning Research, vol. 2, pp. 45–66, 2001.
[4] A. J. Joshi, F. Porikli, and N. Papanikolopoulos, “Multi-class active
learning for image classification,” in CVPR, 2009.
[5] A. J. Joshi, F. Porikli, and N. P. Papanikolopoulos, “Scalable active
learning for multiclass image classification,” IEEE Trans. Pattern Anal.
Mach. Intell., vol. 34, no. 11, pp. 2259–2273, 2012.
[6] A. Kapoor, G. Hua, A. Akbarzadeh, and S. Baker, “Which faces to tag:
Adding prior constraints into active learning,” in ICCV, 2009.
[7] K. Wang, D. Zhang, Y. Li, R. Zhang, and L. Lin, “Cost-effective active
learning for deep image classification,” TCSVT, 2016.
[8] Trigeorgis, G., Bousmalis, K., Zafeiriou, S., and Schuller, B. A deep semi-nmf model for learning hidden representations. In Proceedings of the International Conference on Machine Learning (ICML), pages 1692–1700, 2014.
[9] E. Aljalbout, V. Golkov, Y. Siddiqui, and D. Cremers. Clustering with Deep Learning: Taxonomy and New Methods. arXiv:1801.07648, 2018.
[10] Xie, J., Girshick, R., and Farhadi, A. Unsupervised deep embedding for clustering analysis. In International Conference on Machine Learning (ICML), pages 478–487, 2016.
[11] Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. Generative adversarial nets. In Advances in neural information processing systems, pages 2672–2680, 2014.
[12] Ronneberger, O., Fischer, P., Brox, T.:Unet: Convolutional networks for biomedical image segmentation. In: MICCAI. pp. 234–241. Springer, 2015.
[13] P. Ji, T. Zhang, H. Li, M. Salzmann, and I. Reid. Deep subspace clustering networks. In Neural Information Processing Systems (NIPS), 2017.
[14] J. Xie, R. Girshick, and A. Farhadi. Unsupervised deep embedding for clustering analysis. In ICML, 2016.
[15] Mathilde Caron, Piotr Bojanowski, Armand Joulin, and Matthijs Douze. Deep clustering for unsupervised learning of visual features. In European Conference on Computer Vision (ECCV), 2018.
[16] E. Elhamifar and R. Vidal. Sparse subspace clustering. In Proceedings of IEEE International Conference on Computer Vision and Pattern Recognition, pages 2790–2797, 2009.
[17] E. Elhamifar and R. Vidal. Sparse subspace clustering: Algorithm, theory, and applications. TPAMI, 35(11):2765–2781, 2013.
[18] Lin, F., Cohen, W.W.: Power iteration clustering. In: ICML. (2010)
[19] Zeiler, M.D., Fergus, R.: Visualizing and understanding convolutional networks. In: ECCV. (2014)