銅化學機械研磨之製程參數最佳化

Process Recipe Optimization in Copper CMP

張益瑞(I-Jui Chang)

指導教授 ： 張耀仁

中原大學/工學院/機械工程研究所/碩士(2005年)

https://doi.org/10.6840/cycu200500482

摘要

摘 要
為了因應市場趨勢，國際大廠必須設計積集度高而且多工之積體電路，以符合消費性電子產品需求，在下游的晶圓代工業中，製程技術必須不斷的改變更新以適應新產品，才能縮短開發時程以應付客戶短交期的壓力。在今日，快速而有效的尋求最佳製程參數已成為半導體業關注的焦點。
近年來直交表及基因演算法在工業上被廣泛使用在尋找新導入製程的最佳參數，智慧型基因演算法是一種最佳化演算法，其主要的觀念是將基因演算法與直交表兩種截然不同的最佳化機制加以緊密的結合在一起，改良傳統基因演算法中的交配運算(crossover operation)。
本研究利用晶圓銅膜移除量與可控因子直交表訓練類神經網路建構銅化學機械研磨(Copper CMP)之製程模型並結合智慧型基因演算法(IGA)蒐尋出全域最佳解，以達成最佳化參數設計，並驗証此一方法的有效性。
本研究以CMP機台之可控因子：(1)研磨壓力，(2)研磨墊轉速，(3)晶圓轉速，(4)研磨時間作為智慧型基因演算法搜尋對象。當晶圓銅膜移除量目標值設為5500Å，在演化30代後即快速收斂，第96代時出現更高適應值之最佳化參數。將此最佳化參數輸入CMP機台後得到之實際輸出平均值為5567 Å，標準差為24 Å。此結果顯示本研究所使用的方法在求最佳化的問題上相較於產業界之田口方法，具有更佳的準確度(accuracy)與精密度(precision)。

關鍵字

製程參數 ； 銅化學機械研磨 ； 最佳化

並列摘要

ABSTRACT
In order to accommodate the market trend and satisfy the demand of consumable electronic product, the major design houses tend to design a chip with high density and multiple working capabilities. Relatively, to reduce the development time and resolve the delivery pressure coming from the customer, the foundries need to continuously promote their process technology to adopt these new products. Therefore, the foundries currently focus on the subject of finding the optimal process parameters quickly and efficiently.

Recently, the orthogonal arrays and simple genetic algorithm (SGA) have been widely used to find the optimal parameters for a new developed process. Intelligent genetic algorithm (IGA) is an optimized algorithm, which closely combines the two totally different optimization mechanisms of orthogonal experimental design and SGA, and it also improves the “crossover operation” in the traditional SGA.

In this research, we use orthogonal arrays (OAs) to train a neural network (NN), and then use this NN to create a Copper CMP process model. Finally, we combine IGA and this process model to find the global optimum, achieving the goal of optimization of parameter design, and also verifying the effectiveness of this method.

Some controllable factors, such as polish pressure, polish speed, and polish time, are used to be the key optimal parameters of IGA. In the case of Cu film removal rate target 5500 Å, the data converge at the 30th generation, and finally a set of optimal parameters appears at the 96th generation. After applying those parameters to the real CMP tool, we get an average removal rate of 5567 Å, and a standard deviation of 24 Å.

The result has approved that, comparing to the traditional Taguchi method, the approach in this study is able to get a set of optimized parameters with better accuracy and precision.

並列關鍵字

optimization ； process recipe ； Copper CMP

參考文獻

[5]江孟龍，“多響應系統之製程配方最佳化”，中原大學機械研究所碩士論文，(2002)。

[1] Mirra CMP Process Concepts & Measurement, Applied Materials, INC , (2003).

[10] J. Torres, “Cu dual damascene for advanced metallisation(0.18/spl mu/m and beyond)”, Proceedings of the IEEE Interconnect Technology, International Conference, pp.253-255, (2000).

[11]F. B. Kaufman, D. B. Thompson, R. E. Broadie, M. A. Jaso, W. L. Guthrie, D. J. Pearsons, and M. B. Small, J. Electrochem. Soc., 138, 3460, (1991).

[20] G. Derringer and R. Suich, “Simultaneous Optimization of Several Response Variables”, Journal of Quality Technology, Vol.12,pp.214-219, (1980).