在近幾年以來，在半導體業界為了迎合更輕薄短小、熱性能的處理、更好的電性需求，以及腳位輸出輸入端的增加，電子封裝技術不斷的更新開發。電子封裝技術也由傳統封裝到覆晶(Flip Chip)、晶片尺寸封裝(Chip Scale Packaging)、晶圓級封裝(Wafer level Packaging)，和面板級封裝(Panel level Packaging)等先進封裝。
其中的扇出型面板級封裝(Fan-Out Panel Level Packaging, FO-PLP)，有別於其他封裝體的地方是先進行封裝後再切割成獨立IC的製程技術。在製程規格上，面板級封裝的面積可到達500×500〖mm〗^2的矩形面積，相較於12吋晶圓之圓形面積大上數倍。此外矩形面板的設計使材料使用率上亦高於其他封裝體，兼具了大面積和高材料使用率等優勢，使得此封裝體成為討論的話題。
在面板級封裝製程階段中，會經歷封膠過程(Molding Process)和後熟化(Post-Molding Curing)加熱使來促使膠體完全熟化，封裝體和面板會加熱到熟化溫度，再降溫至室溫，此階段的溫度變化容易造成翹曲現象發生。翹曲量過大會影響後續的製程，是至今仍需解決的問題。所以量測在不同尺度封裝體的翹曲量成為重要的工作之一，故本研究將致力於建立三維模型，並測得其翹曲量。
在建立三維模型，本研究採用業界廣泛使用的有限元素法(Finite Element Method, FEM)來進行模擬，並且引用由實驗驗證過的等效熱膨脹係數(Equivalent CTE)，對封膠材料環氧模壓樹脂的固化反應進行簡化，建立扇出型面板級封裝模型。
不論模擬或實驗上要得到翹曲數據，皆需要大量的時間或是成本，且常因人為疏失導致結果不同。本研究為了降低上述情況，藉由人工智慧(Artificial Intelligence, AI)進行不同封裝尺度下的翹曲量預估。運用有限單元法建立出不同尺度之扇出型面板級封裝模型的訓練數據庫，藉由人工智慧得出可靠的已訓練預估模型。而後在面臨不同尺度的封裝體，可以直接藉由此預估模型在短時間內得出翹曲結果。
本研究在演算法為梯度卷積神經網路(Modified Convolution Neural Network)，為卷積神經網路後續發展的演算法。運用了傳統卷積神經網路之卷積層(Convolution Layer)和過濾器(Filter)之概念，擷取訓練資料中重要的資料點，以此優化神經網路之預估精度，和減少計算時間。
使用神經網路目標在於求出最佳的預估數值，在神經網路中的梯度下降法(Gradient Decent)更新時，容易因為神經網路結構設定的不同，和隨機參數的不確定性，陷入神經網路中的區域最低值(Local Minimum)，求出錯誤的預估數值。為了避免上述情況發生，本研究引進粒子群最佳化法(Particle Swarm Optimization, PSO)，利用群體智慧的力量改善上述情況，並提升整體的預估精度。
粒子群最佳化法運用散佈的粒子群在不同維度中作求解，根據粒子群的回饋數值，作為更新前進的方向。以個體最佳解(Personal Best)和群體最佳解(Group Best)，代表在群體中個體和群體的經驗，並找尋最適合的解。此演算法因為精度高、收斂速度快，和更新方式簡易，常被運於求解許多工程領域上的問題，和優化參數。本研究引進此演算法尋找神經網路之初始權重，代替原先神經網路的隨機權重，避免落入區域最低值的情況發生。

In recent years, electronic packaging technology has been continuously updated and developed for the sake of lighter, thinner, shorter, thermal performance, better electrical requirements and increase in pin output and input in the semiconductor industry. Electronic packaging technology has also changed from traditional packaging to advanced packaging such as flip chip, chip scale packaging, wafer level packaging, and panel level packaging.
Among the advanced packaging, Fan-Out Panel Level Packaging (FO-PLP) is different from other packages in the process technology of packaging and then cutting into individual ICs. In terms of process specifications, the rectangular area of the panel level packaging which reaches 500×500〖mm〗^2 is several times larger than the circular area of a 12-inch wafer. Moreover, the design of the rectangular panel makes the material usage rate higher than other packages. The combination of large area and high material usage makes this package a topic of discussion.
In the panel-level packaging process stage, it will go through the molding process and post-molding curing heating to promote the complete curing of the molding compound. The package will be heated to the curing temperature and then cooled down to the room temperature. Temperature changes at this stage are likely to cause warpage. Excessive warpage affects the subsequent manufacturing process. Warpage is still a problem that still needs to be solved. Measuring the warpage of packages at different scales has become one of the important tasks. Therefore, this research will focus on establishing a three-dimensional model and measuring the amount of warpage.
For establishing the three-dimensional model, the Finite Element Method has been introduced in this study, which is widely used to simulate in the industry. Here also introduced the equivalent coefficient of thermal expansion which is verified by experiments as the CTE of package material, epoxy molding compound. The equivalent CTE is used to simplify the simulation of the curing reaction, and establish the fan-out panel-level package model.
Whether to obtain warpage data in simulation or experiment, a lot of time or cost is required, and the results are often different due to human error. In order to reduce the situation, this study uses Artificial Intelligence (AI) to estimate the amount of warpage under different package sizes. The finite element method is used to establish a training database for fan-out panel-level packaging models of different scales. Through artificial intelligence, a reliable trained prediction model is obtained. When facing packages of different scales, we can get the prediction of warpage directly from the trained model in a short time.
The algorithm in this research is Modified Convolution Neural Network (Modified Convolution Neural Network), which is the development of convolutional neural networks. The concepts of Convolution Layer and Filter of the traditional convolutional neural network are used to capture important data points in the training data to optimize the prediction accuracy of the neural network and reduce the calculation time.
The goal of using a neural network is to find the best prediction. When the gradient descent method in the neural network is updating, it falls into Local Minimum of neural network easily because of the different neural structure and the uncertainty of random parameters. In order to avoid the above situation, this research introduces Particle Swarm Optimization (PSO) to use the power of swarm intelligence to improve the overall accuracy of prediction.
The particle swarm optimization uses scattered particle swarms to solve in different dimensions. According to the feedback value of the particle swarm, it is used as the direction of the update. The Personal Best and the Group Best represent the experience of individuals and groups in the particle swarms. The algorithm will use these two parameter in particles to find the best solution. This algorithm is often used to solve problems in many engineering fields and optimize parameters because of its high accuracy, fast convergence speed, and simple updating. This research introduces this algorithm to find the initial weight of the neural network, instead of the original random weight of the neural network, to avoid the situation of falling into the lowest value of the loss function.

摘要 I
Abstract III
目錄 VI
圖目錄 X
表目錄 XIII
第一章 緒論 1
1.1 簡介 1
1.2 研究動機與目標 2
1.3 文獻回顧 3
1.4 扇出行晶圓級與面板級封裝製程 16
第二章 基礎理論 18
2.1 有限元素法理論 18
2.1.1 有限元素法線彈性理論 18
2.2 有限元素接觸理論 22
2.2.1 懲罰函數法 23
2.2.2 拉格朗日乘子法 24
2.2.3 增廣拉格朗日乘子法 24
2.3 翹曲現象 25
2.3.1 熱膨脹係數不匹配 25
2.3.2 固化反應 26
2.4 P-V-T-C方程式 27
2.5 等效熱膨脹係數法 28
2.6 機器學習 31
2.6.1 感知器 33
2.6.2 人工神經網路 34
2.6.3 梯度下降法 35
2.6.4 學習速率常數 36
2.6.5 反向誤差傳播法 40
2.6.6 激活函數 41
2.6.7 卷積神經網路 44
2.6.8 圖像補零法 46
2.7 機器學習優化和評估 47
2.7.1 資料前處理 47
2.7.2 交叉驗證 49
2.7.3 擬和度問題 50
2.7.4 正規化 51
2.8 粒子群最佳化法 53
2.8.1 個體最佳解和群體最佳解 53
2.8.2 粒子更新機制 55
2.8.3 應用於神經網路 57
第三章 有限元素模型建立 58
3.1 扇出型面板級封裝模型 58
3.1.1 材料參數和元素尺寸設定 59
3.1.2 邊界條件與負載設定 61
3.2 翹曲量分析 62
3.2.1 載板脫離製成分析 64
第四章 機器學習 67
4.1 機器學習資料庫 68
4.1.1 數據庫建立 69
4.1.2 封裝體尺寸對翹曲量影響分析 72
4.2 演算法資料點選取 78
4.2.1 閥值設定 79
4.2.2 過濾器選擇 81
4.2.3 資料點選取 83
4.3 粒子群最佳化產生初始權重 87
4.3.1 速度權重選擇 87
4.3.2 參數設置探討 91
4.3.3 粒子數目選擇 95
第五章 結果與討論 99
5.1 神經網路最佳化參數 100
5.2 修正型卷積神經網路之Feature Map參數探討 103
5.3 粒子群最佳化法應用於修正型卷積神經網路 110
第六章 結論與建議未來工作 119
參考文獻 121

[1] H. J. Borchardt and F. Daniels, “The application of differential thermal analysis to the study of reaction kinetics,” Journal of the American Chemical Society, vol. 79, no. 1, pp. 41-46, 1957.
[2] K. E. J. Barrett, “Determination of rates of thermal decomposition of polymerization initiators with a differential scanning calorimeter,” Journal of Applied Polymer Science, vol. 11, no. 9, pp. 1617-1626, 1967.
[3] K. Horie, H. Hiura, M. Sawada, I. Mita, and H. Kambe, “Calorimetric investigation of polymerization reactions. III. Curing reaction of epoxides with amines,” Journal of Polymer Science Part A‐1: Polymer Chemistry, vol. 8, no. 6, pp. 1357-1372, 1970.
[4] M. Kamal and S. Sourour, “Kinetics and thermal characterization of thermoset cure,” Polymer Engineering & Science, vol. 13, no. 1, pp. 59-64, 1973.
[5] M. R. Kamal and M. Ryan, “The behavior of thermosetting compounds in injection molding cavities,” Polymer Engineering & Science, vol. 20, no. 13, pp. 859-867, 1980.
[6] S. J. Hwang and Y. S. Chang, “PVTC equation for epoxy molding compound,” IEEE Transactions on components and Packaging Technologies, vol. 29, no. 1, pp. 112-117, 2006.
[7] S. S. Deng, S. J. Hwang, H. H. Lee, D. Y. Huang, Y. R. Chen, and G. S. Shen, “Simulation and experiments of fan-out wafer level package during encapsulation,” in 2009 4th International Microsystems, Packaging, Assembly and Circuits Technology Conference(IMPACT), Taipei, Taiwan, pp. 48-51, October 21-23, 2009.
[8] K. Oota and M. Saka, “Cure shrinkage analysis of epoxy molding compound,” Polymer Engineering & Science, vol. 41, no. 8, pp. 1373-1379, 2001.
[9] J. W. Kong, J. K. Kim, and M. M. Yuen, “Warpage in plastic packages: effects of process conditions, geometry and materials,” IEEE transactions on electronics packaging manufacturing, vol. 26, no. 3, pp. 245-252, 2003.
[10] T. Lin et al., “Warpage simulation and experiment for panel level fan-out package,” in 2016 IEEE CPMT Symposium Japan (ICSJ), Kyoto, Japan, pp. 129-131, November 7-9, 2016.
[11] F. Hou et al., “Experimental verification and optimization analysis of warpage for panel-level fan-out package,” IEEE Transactions on Components, Packaging and Manufacturing Technology, vol. 7, no. 10, pp. 1721-1728, 2017.
[12] M. Su et al., “Warpage simulation and experimental verification for 320 mm× 320 mm panel level fan-out packaging based on die-first process,” Microelectronics Reliability, vol. 83, pp. 29-38, 2018.
[13] R. A. Fisher, “The Use of Multiple Measurements in Taxonomic Problems”, Annals of Eugenics, Vol.7, Issue 2, pp. 179–188, 1936.
[14] D. R. Cox, “The regression analysis of binary sequences (with discussion)”, Journal of the Royal Statistical Society, Vol 20, pp. 215-242, 1958.
[15] F. Rosenblatt "The Perceptron: A Probabilistic Model For Information Storage And Organization In The Brain", Psychological Review, Vol 65, Issue 6, pp. 386–408, 1958.
[16] T. Cover and P. Hart, “Nearest neighbor pattern classification,” IEEE transactions on information theory, vol. 13, no. 1, pp. 21-27, 1967
[17] D. E. Rumelhart, Geoffrey E. Hinton, and Ronald J. Williams, “Learning internal representations by back-propagating errors”, Nature, Vol. 323, pp.533-536, 1986.
[18] J. R. Quinlan, “Induction of Decision Trees”, Machine Learning, Vol.1, Issue 1, pp. 81–106, 1986.
[19] T. K. Ho, “Random Decision Forest”, Proceedings of 3rd International Conference on Document Analysis and Recognition, Montreal, Quebec, Canada, Vol 1, pp. 278-282, 1995.
[20] W. S. McCulloch and W. Pitts, “A logical calculus of the ideas immanent in nervous activity”, The bulletin of mathematical biophysics, Vol.5, Issue 4, pp 115–133, 1943.
[21] Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner, ” Gradient-based learning applied to document recognition” IEEE, Vol 86, Issue 11, pp 2278-2324, 1998.
[22] C. Cortes and V. Vapnik, “Support-Vector Networks”, Machine Learning, Vol. 20, Issue 3, pp. 273–297, 1995.
[23] T. Braun et al., “Large area compression molding for Fan-out Panel Level Packing,” in 2015 IEEE 65th Electronic Components and Technology Conference (ECTC), San Diego, Californa, USA, May 26-29, 2015.
[24] 蔡佳翰,“使用等校熱膨脹係數於扇出型面板及封裝之翹曲研究”, 國立清華大學動力機械工程學系,碩士論文, 2018
[25] K. N. Chiang, H. C. Cheng, and W. H. Chen. “Large-Scaled 3-D Area Array Electronic Packaging Analysis”, Journal of Computer Modeling and Simulation in Engineering, Vol. 4, No. 1, pp. 4-11, 1999.
[26] S. S. Shai and B. D. Shai. “Understanding Machine Learning From Throry to Algorithms” Cambridge University Press, 2014.
[27] S. Aich, K. Younga, K. L. Hui, A. A. Al-Absi and M. Sain, “A nonlinear decision tree based classification approach to predict the Parkinson's disease using different feature sets of voice data”, in 2018 20th International Conference on Advanced Communication Technology (ICACT), Korea (South), Feb. 11-14, 2018.
[28] D. Wu, C. Jennings , J. Terpenny, R. X. Gao, and S. Kumara, “A Comparative Study on Machine Learning Algorithms for Smart Manufacturing: Tool Wear Prediction Using Random Forests”, Journal of Manufacturing Science and Engineering, Vol.139, Issue 7, 071018, 2017.
[29] Tomáš Mikolov, Martin Karafiát, Lukáš Burget, Jan Černocký, and Sanjeev Khudanpur, “Recurrent Neural Network Based Language Model” in 2010 37th International Symposium on Computer Architecture (ISCA), Makuhari, Chiba, Japan, Sep. 26-30, 2010.
[30] D. H. Hubel and T. N. Wiesil “Receptive Fields, Binocular Interaction and Functional Architecture in the Cat’s Visual Cortex” J. Physiology, Vol. 160, Issue 1 pp. 106-154, Jan 1962.
[31] K. Fukushima and S. Miyake, “Neocognitron: A self-organizing neural network model for a mechanism of visual pattern recognition,” in Competition and cooperation in neural nets: Springer, pp. 267-285, 1982.
[32] Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner, ” Gradient-based learning applied to document recognition” IEEE, Vol 86, Issue 11, pp 2278-2324, 1998.
[33] A. Krizhevsky, I. Sutskever, and G. K. Hinton “ImageNet classification with deep convolutional neural networks” Advances in Neural Information Processing Systems, Vol. 1, pp 1097-1105,2012.
[34] J. Deng, W. Dong, and R. Socher, L. J. Li, K. Li, and L. Fei Fei, “Imagenet: A large-scale hierarchical image database,” in 2009 IEEE conference on computer vision and pattern recognition, 2009: Ieee, pp. 248-255.
[35] 周佩勳,“以人工神經網路回歸模型評估晶圓級封裝之可靠度”, 國立清華大學動力機械工程學系,碩士論文, 2019
[36] 沈奕廷“以支援向量回歸模型評估晶圓級封裝之可靠度”, 國立清華大學動力機械工程學系,碩士論文, 2019
[37] T. Wang, Y. Chen, M. Qiao, and H. Snoussi, “A fast and robust convolutional neural network-based defect detection model in product quality control,” The International Journal of Advanced Manufacturing Technology, vol. 94, no. 9-12, pp. 3465-3471, 2018.
[38] F. Yin, H. Mao, L. Hua, W. Guo, and M. Shu, “Back Propagation neural network modeling for warpage prediction and optimization of plastic products during injection molding” Materials & Design, 2011.
[39] S. Taghizadeh, A. Ozdemir, and O. Uluer, “Warpage prediction in plastic injection molded part using artificial neural network,” 2013.
[40] C. Selvanayagam, P. L. T. Duong, R. Mandal, and N. Raghavan, “Machine learning approach to improve accuracy of warpage simulations,” in 2019 IEEE 69th Electronic Components and Technology Conference (ECTC), Las Vegas, Nevada, USA, pp. 834-841, May 28-31,2019.
[41] J. Kennedy and R. Eberhart, “Particle swarm optimization,” in Proceedings of the IEEE International Conference on Neural Networks, vol. 4, pp. 1941-1948, December 1995.
[42] W. C. Yeh, "New parameter-free simplified swarm optimization for artificial neural network training and its application in the prediction of time series," IEEE Trans Neural Netw Learn Syst, vol. 24, no. 4, pp. 661-5, April 2013
[43] S. W. Fei, M. J. Wang, J. Tu, and C. L. Liu, “Particle swarm optimization-based support vector machine for forecasting dissolved gases content in power transformer oil,” Energy Convers. Manage, vol. 50, no. 6, pp. 1604-1609, 2009.
[44] 陳珉宇,“以修正型卷積神經網路法預估扇出型面板級封裝之翹曲研究”, 國立清華大學動力機械工程學系,碩士論文, 2020
[45] S. Raschka and V. Mirjalili, Python machine learning. Packt Publishing Ltd, 2017
[46] K. J. Bathe, Finite element procedures. Klaus-Jurgen Bathe, 2006.