近年來，如何有效降低生產成本與提升良率，已成為半導體製程產業致力改進的方向。影響半導體製造成本因素有線寬的縮小、晶圓尺寸的增大、良率的提高及設備的整體效益提升。線寬縮小的貢獻將維持在12%~14%，晶圓尺寸的增大的貢獻只有佔2%的貢獻，而良率的提高所衍生的成本降低更只有1%的效益，剩下的成本降低空間則有賴設備整體效益的提升。為了改善或提升上述的關鍵指標，晶圓廠在引進製程設備後，無不投入大量資源，致力於掌握設備使用技巧，提升晶圓生產線的可靠度與彈性，以獲得最大的產出良率與數量，可減少壞片，而降低生產成本。對減少不良品或廢品的發生機率，在線上進行調整製程配方，以確保產品品質不因機台特性飄移而受影響，並減少量測造成的監視片，將有助於設備的整體效益。過去二十幾年來半導體產業，技術及產能大致能遵循著莫爾定律的預測不停的提升進步，但隨著製程邁入ULSI紀元後，晶圓尺寸加大，線徑縮小、製程日漸複雜，製程的控制也日益困難。
本文係使用晶圓廠線上實際測試的數據，配合田口實驗設計，得到最佳化之控制因子組合及控制因子對薄膜目標厚度及製程變異之影響性，採用田口實驗設計法使得實驗參數之間無交互作用，建立最佳化之製程線性迴歸(Linear Regression)數學模型，並用二階段製程線性迴歸模型指數加權平均(Exponentially Weighted Moving Average ; EWMA)控制器來控制擴散(Diffusion)製程的薄膜厚度，第一階段使用機台保養線性迴歸模型指數加權平均控制器，使製程收斂快速，第二階段使用穩定化線性迴歸模型指數加權平均控制器，使得後續批次製程變異穩定，以達成精確製程控制之最佳化方法。

In recent years, how to reduce the costs and enhance the qualified yield has become the direction of improvement for the semiconductor industry. The factors related to cost saving include the reduction of the wired width, increase of wafer size, enhancement of the qualified yield and the promotion of the overall efficiency of the equipment. Among those contributions, 12% -14% results from the reduction of wired width, only 2% results from the size of the wafer and only 1 % goes to the enhancement of the qualified yield. The rest of the contributions to cost reduction are attributed to the enhancement of the equipment efficiency. In order to improve above key indicators, the manufacturers always invest massive resources after installing their production equipment to increase the techniques of equipment operation, enhance the reliability and elasticity of the production line, and obtain the maximum qualified yield so as to reduce the costs. To reduce the rate of wafer scrap and the rate of non-qualified yield, the recipe on the production line is adjusted to make sure the quality of the products not to be impacted by the process drift of the equipment and reduce the consumption of the monitor wafers so as to result in the overall efficiency of the equipment. Over the past two decades, the technology and capacity of the semiconductor industry could be improved smoothly based on the forecast of the Mole Law. Thus while entering the ULSI era, the process of production is more hardly controlled due to the increase of wafer size, reduction of wired width and the complexity of technology day after day.

This study applies the Taguchi method with the data on the production line to obtain the optimal combination of the control factors as well as the effects of the control factors on the thickness of the film and the deviation of the process. The Taguchi method is adopted to establish the optimal linear regression model since there in no interaction among the experiment parameters. Furthermore, the two-stage Linear Regression model of EWMA controller is applied to control the thickness of the film in the diffusion process. In the first stage, the periodical maintenance Linear Regression model of EWMA controller is applied to make the process converge speedily. In the second stage, the stable Linear Regression model of EWMA controller is applied to stabilize the process so as to obtain the optimal method for accurately controlling the process.

1.Jansen, F.,“AVS Short Course: PECVD,＂ American Vacuum Society,(1990).
2. 莊達人,“VLSI 製造技術,＂ 高立圖書有限公司, (2000).
3. Shewhart, W. A., Economic Control of Quality of Manufactured Product,Princeton, NJ: Van Nostrand Reinhold (1931).
4. Juran, J. M., Juran on Quality by Design, Free Press, (1992).
5. Cai, D. Q.,Xie, M. and Goh, T. N., "SPC in an Automated Manufacturing Environment," International Journal of Computer Integrated Manufacturing, Vol. 14,No. 2, pp. 206-211 (2001).
6. Montgomery, D. C., and Mastrangelo , C. M., "Some Statistical Process Control Methodsfor Autocorrelated Data," Journal of Quality Technology, Vol. 23, No. 3, pp. 179-193(1991).
7 Wang, X. A. and Mahajan, R. L., “Artificial neural network model-based run-to-run process controller,” IEEE Transactions on Components, Packaging,and Manufacturing Technology-Part C, 9, 19-26 (1996).

8. Holmes, D.,Aluise, T.,Burt, K. and Ellsworth, S., “Combing SPC with APC to 60 improve quality,” Quality Engineering, 10, 575-578 (1998).
9. Shao, Y. E., “Integrated application of the cumulative score control engineering process control,” Statistical Sinica, 8, 239-252 (1998).
10. MacGregor, J. F., “On-line statistical process control,” Chemical Engineering Process, 21-31 (1988).
11. 楊宗儒，「SPC 與EPC 整合架程研究所碩士論文，(2000)。
12.Box, G., and Kramer, T., “Statistical process monitoring and feedback adjustment-a discussion,” Technometrics, 34, 3, 251-267 (1992).
13. Wiel, S. A. V., Tucker, W. T., Faltin, F. W., and Doganaksoy, N., “Algorithmic statistical process control: concepts and an application,” Technometrics, 34, 3,286-297 (1992).
14. Sachs, E., Hu, A., and Ingolfsson, A., “Run-to-run process control: combining SPC and feedback control,” IEEE Transactions on Semiconductor Manufacturing, 8, 1, 26-43 (1995).
15. Boning, D., Moyne, W.,Smith, T., et. al., “Run by run control of Chemical-Mechanical Polishing,” IEEE/CPMT Int’l Electronics Manufacturing Technology Symposium, pp. 81 –87, (2001).
16. 李輝煌,＂田口方法-品質設計的原理與實務＂,高立圖書有限公
司,(2000)
17. Roberts, S., “Control chart tests based on geometric moving averages,” Technometrics, 1, 239–250(1959).
18. Hotelling, H., “Analysis of a Complex of Statistical Variables into Principal Components”, Journal of Education Psychology, Vol. 24, pp. 498-520, (1933).
19. 張建邦(1997)，多變量分析，台北：三民，(2001)。
20. Tseng, S. T., Chou, R. J. and Lee, S. P., “A study of multivariate EWMA controller,” IIE Transactions, Vol. 34, pp. 541-549, (2002).