AI人工智慧目前廣為學術單位及企業使用。AI人工智慧在使用過程，常須搭配大數據以進行建模預測。實務上，企業欲解決的問題不一定都有大數據可供使用。例如，為因應消費型電子之市場需求，半導體企業常需開發領先市場之新產品。此種領先市場之新產品開發過程，經常沒有相關之歷史數據可供建模預測。若為了使用大數據來建立模型以改善產品品質，則企業需進行大量實驗來收集數據。大量實驗對企業造成額外成本支出，也不一定可以找出關鍵控制因子，進而導致模型之預測能力不佳。
半導體企業對於產品品質之要求非常嚴謹，若因品質問題影響產品之出貨時間，則會嚴重影響產品之市占率。所以要如何快速的提升產品品質為一重要之研究課題。許多企業在產品改善過程中，因為沒有太多的數據可供參考分析，所以常利用試誤法進行品質改善。但試誤法常找不到最佳的因子設定條件。更重要的是花費了大量的實驗成本，也拉長了改善時間。
本文提出一個整合實驗設計(部分因子實驗、反應曲面法、田口方法…)與人工智慧(類神經網路、基因演算法…)之改善方法。此方法協助個案公司快速且有效地收集產品數據、篩選關鍵屬性與建立一個預測模型以改善DRAM產品之W結構厚度偏差及DRAM產品之CoSi2阻值變異。在應用此方法之後，DRAM產品之W結構之厚度變異從45.0 Å減少到了12.9 Å而CoSi2阻值變異也從1.440 ohms減少到0.302 ohms。

Artificial intelligence (AI) is widely used by academic institutions and enterprises. In using artificial intelligence, it is necessary to use big data for modeling and predicting. In practice, big data is not always available for the problems that enterprises want to improve. For example, in response to the market request for consumer electronics, semiconductor enterprises often develop new advanced products in the market. In the market-leading new product development process, there is often no relevant historical data for modeling and predicting. If enterprises want to use big data to construct models to improve product quality, construct need to perform a lot of experiments to collect data. A large number of experiments have caused additional costs for the enterprises, and it may not find the key control factors, which leads to the poor prediction of the model.
Semiconductor companies have very strict requirements for product quality. If quality problems affect the shipping schedule, it will seriously affect the product's market share. Therefore, how to quickly improve the quality of products is an important study topic. During the product improvement, many companies use trial and error to improve product quality because there is not big data for reference. But the trial and error method often fails to find the best setting of the control factors. More importantly, a lot of experimental costs were spent, which also delayed the improvement schedule.
This thesis proposed an improvement approach that integrates experimental design (e.g. fractional factorial design, response surface method, Taguchi’s method…) and artificial intelligence (e.g. neural networks, genetic algorithms...). The proposed approach helps case companies to collect data, select important control factors, and construct a prediction model for reducing the W-shaped structure thickness deviation and the CoSi2 resistance variation of DRAM products. After applying the proposed approach, the average thickness variation of the W-shaped structure is reduced from 45.0 Å to 12.9 Å, and the CoSi2 resistance variation of DRAM products is reduced from 1.440 ohms to 0.302 ohms.

[1] D.C. Montgomery, Design and Analysis of Experiments, 6th ed., New York, NY, USA: John Wiley & Sons, 2005.
[2] G. E. Box and J. S. Hunter, “The 2k-p fractional factorial designs,” Technometrics, vol. 3, no. 3, pp. 311-351, 1961, doi: 10.1080/00401706.1961.10489951.
[3] S. C. Dias and J. A. Dias, “Effects of the dealumination methodology on the FER zeolite acidity: A study with fractional factorial design,” Molecular Catalysis, vol. 458, pp. 139-144, 2018, doi: 10.1016/j.mcat.2017.12.028.
[4] S. P. Harborne, D. Wotherspoon, J. Michie, A. McComb, T. Kotila, S. Gilmour and A. Goldman, “Revolutionising the design and analysis of protein engineering experiments using fractional factorial design,” bioRxiv, 298273, 2018, doi: 10.1101/298273.
[5] G. E. Box and K. B. Wilson, “On the experimental attainment of optimum conditions,” Journal of the Royal Statistical Society: Series B (Methodological), vol. 13, no. 1, pp. 1-45, 1951, doi: 10.1111/j.2517-6161.1951.tb00067.x.
[6] H. Sharifi, S. M. Zabihzadeh and M. Ghorbani, “The application of response surface methodology on the synthesis of conductive polyaniline/cellulosic fiber nanocomposites,” Carbohydrate polymers, vol. 194, pp. 384-394, 2018, doi: 10.1016/j.carbpol.2018.04.083.
[7] M. Tuzen, A. Sarı and T. A. Saleh, “Response surface optimization, kinetic and thermodynamic studies for effective removal of rhodamine B by magnetic AC/CeO2 nanocomposite,” Journal of environmental management, vol. 206, pp. 170-177, 2018, doi: 10.1016/j.jenvman.2017.10.016.
[8] R. H. Myers, D. C. Montgomery, and C. M. Anderson-Cook, Response Surface Methodology: Process and Product Optimization Using Designed Experiments, 3rd ed., New Jersey, USA: John Wiley & Sons, Inc., 2009.
[9] A. I. Khuri, Response Surface Methodology and Related Topics. London, UK: World Scientific, 2006, doi: 10.1142/5915.
[10] C. T. Su, Quality Engineering: Off-Line Methods and Applications, 1st ed., Boca Raton, FL, USA: CRC Press, 2013.
[11] A. N. Özakın and F. Kaya, “Experimental thermodynamic analysis of air-based PVT system using fins in different materials: Optimization of control parameters by Taguchi method and ANOVA,” Solar Energy, vol. 197, pp. 199-211, Feb. 2020, doi: 10.1016/j.solener.2019.12.077.
[12] L. Jiang, Y. Li, Y. Huang, J. Yu, X. Qiao, Y. Wang, C. Huang, and Y. Cao, “Optimization of multi-stage constant current charging pattern based on Taguchi method for Li-Ion battery,” Applied Energy, vol. 259, no. 1, Feb. 2020, 114148, doi: 10.1016/j.apenergy.2019.114148.
[13] P. V. Dandale and S. S. Pathak, “Parameters Optimization on Compressive Strength of Polypropylene Fiber Concrete by Taguchi Method,” In Techno-Societal 2018. Cham, Switzerland: Springer, 2020, pp. 789-797, Springer, Cham, doi: 10.1007/978-3-030-16848-3_71.
[14] N. Berkane, S. Meziane, and S. Aziri, “Optimization of Congo red removal from aqueous solution using Taguchi experimental design,” Separation Science and Technology, vol. 55, no. 2, pp. 278-288, 2020, doi: 10.1080/01496395.2019.1577442.
[15] P. J. Ross, Taguchi Techniques for Quality Engineering: Loss Function, Orthogonal Experiments, Parameter and Tolerance Design, 2nd ed., New York, NY, USA: McGraw-Hill, 1996.
[16] F. Rosenblatt, Principles of Neurodynamics: Perceptrons and the Theory of Brain Mechanisms, New York, NY, USA: Spartan Books, 1962.
[17] H. S. Stern, “Neural networks in applied statistics,” Technometrics, vol. 38, no. 3, pp. 205-220, 1996, doi: 10.1080/00401706.1996.10484497.
[18] S. Kang, E. Kim, J. Shim, S. Cho, W. Chang, and J. Kim, “Mining the relationship between production and customer service data for failure analysis of industrial products,” Computers & Industrial Engineering, vol. 106, pp. 137-146, 2017, doi: 10.1016/j.cie.2017.01.028.
[19] T. Kessler, T. Schwartz, H.-W. Wong, and J. H. Mack, “Screening Compounds for Fast Pyrolysis and Catalytic Biofuel Upgrading Using Artificial Neural Networks,” In ASME 2019 Internal Combustion Engine Division Fall Technical Conference, American Society of Mechanical Engineers Digital Collection, 2019, doi: 10.1115/ICEF2019-7170.
[20] A. Ansari, M. Heras, J. Nones, M. Mohammadpoor, and F. Torabi, “Predicting the performance of steam assisted gravity drainage (SAGD) method utilizing artificial neural network (ANN),” Petroleum, 2019, doi: 10.1016/j.petlm.2019.04.001.
[21] M. Sheikholeslami, M. B. Gerdroodbary, R. Moradi, A. Shafee, and Z. Li, “Application of Neural Network for estimation of heat transfer treatment of Al2O3-H2O nanofluid through a channel,” Computer Methods in Applied Mechanics and Engineering, vol. 344, pp. 1-12, 2019, doi: 10.1016/j.cma.2018.09.025.
[22] I. Kaastra and M. S. Boyd, “Forecasting futures trading volume using neural networks,” J. of Futures Markets, vol. 15, no. 8, pp. 953-970, 1995, doi: 10.1002/fut.3990150806.
[23] W.-C. Chiang, T. L. Urban, and G. W. Baldridge, “A neural network approach to mutual fund net asset value forecasting,” Omega, vol. 24, no. 2, pp. 205-215, 1996, doi: 10.1016/0305-0483(95)00059-3.
[24] D. E. Goldberg, Genetic Algorithm in Search, Optimization and Machine Learning, 1st ed., New York, NY, USA: Addison-Wesley, 1989.
[25] T.-H. Nguyen, T.-T. Do, D.-N. Nguyen, D. N. Lu, and H. N. Nguyen, “A Hybrid Method Based on Genetic Algorithm and Ant Colony System for Traffic Routing Optimization,” VNU Journal of Science: Comp. Science & Com. Eng., 2019, [Online]. Available: http://eprints.uet.vnu.edu.vn/eprints/id/eprint/ 3609.
[26] M. S. H. Kalathingal, S. Basak, and J. Mitra, “Artificial neural network modeling and genetic algorithm optimization of process parameters in fluidized bed drying of green tea leaves,” Journal of Food Process Engineering, e13128. 2020, doi: 10.1111/jfpe.13128.
[27] R. Alipour-Sarabi, Z. Nasiri-Gheidari, F. Tootoonchian, and H. Oraee, “Improved winding proposal for wound rotor resolver using genetic algorithm and winding function approach,” IEEE Transactions on Industrial Electronics, vol. 66, no. 2, pp. 1325-1334, 2019, doi: 10.1109/TIE.2018.2821091.
[28] M. Gen and R. Cheng, Genetic Algorithms and Engineering Design, 1st ed., New York, NY, USA: John Wiley & Sons, 1997, doi:10.1002/9780470172254.
[29] K. Linliu, DRAM (Dynamic Random Access Memory) Process Flow, Independently Published, 2020.
[30] H. Xiao, 3D IC Devices, Technologies, and Manufacturing, Washington, WA, USA: SPIE Press Monographs, 2016.
[31] B. Prince, Semiconductor Memories: A Handbook of Design, Manufacture and Application, 2nd ed., New York, NY, USA: John Wiley & Sons, 1996.
[32] G. Lee, S. Kim, S. Seok, J. Lee, J. Lee, S. Choi, T. Yoon, J.Park and H. S. Ko, "A study on film thickness control of vertical flow showerhead reactor for RF plasma deposition," 2017 28th Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC), Saratoga Springs, NY, USA, pp. 217-221, 2017, doi: 10.1109/ASMC.2017.7969233.
[33] V. Bhosle, N. Bateman, D. Raj, and T. Miller, "Boron Deposition Control for p-type Plasma Doping," 2018 22nd International Conference on Ion Implantation Technology (IIT). IEEE, pp. 338-341, 2018, doi: 10.1109/IIT.2018.8807913.
[34] J. M. Lee, D. S. Park, S. C. Yew, S. H. Shin, J. Y. Noh, H. S. Kim, and B. D. Choi, "Novel Approach for the Reduction of Leakage Current Characteristics of 20 nm DRAM Capacitors With ZrO2–Based High-k Dielectrics," IEEE Electron Device Letters, 38(11), pp. 1524-1527, 2017, doi: 10.1109/LED.2017.2755050.