應用類神經網路與田口基因演算法於表面聲波器黃光製程最佳化-以表面聲波濾波器為例

Applying Neural Network and Taguchi Genetic Algorithm to Optimization of Photolithography Process of SAW Device-A Case Study on SAW Filter

陳俊傑(Jun-Jei Chen)

指導教授 ： 張耀仁

中原大學/工學院/機械工程研究所/碩士(2009年)

https://doi.org/10.6840/cycu200901341

摘要

摘要

表面聲波元件(Surface Acoustic Wave Device，簡稱SAW Device)主要為在一壓電性材料上，經由黃光製程控制指叉電極(Inter-Digital Transducer，IDT)粗細 ，另外在鍍膜製程控制金屬薄膜的均勻性與厚度，經由此兩者來得到適當的頻率。在半導體製程中，表面聲波元件在製程的複雜性算是較為不繁雜的，但頻率的控制與集中性卻是頻率元件發展最難控制的一環，如何控制製程中的各項參數以取得最佳配方，使頻率的標準差(Sigma)值最小化，但如何在數道製程與不同機台中找到最佳配方，減少製程工程師繁複的實驗與次數，並精確地獲得最佳配方，改善產品良率，是我們所努力與嚮往的目標。本研究使用田口法(Taguchi Method)設計實驗，可控制因子為(1)光阻塗佈轉速(2)光阻烘烤溫度(3)光阻烘烤時間(4)曝光時間(5)曝光後烘烤溫度(6)曝光後烘烤時間(7)顯影時間，輸出為頻率的Sigma。使用田口法設計的實驗當作樣本，用來訓練類神經網路的模型，逐漸獲得輸入和輸出(非隨機樣式)彼此之對應關係。並將此模型用於表面聲波器製程，然後結合田口基因演算法取得最佳配方。過去使用類神經研究最佳配方不在少數但橫跨多道製程與不同機台的探討最佳化本論文的重點。
本研究使用徑向基底類神經網路（Radial Basis Function Neural Network，RBFN）作模型建構，並搭配田口基因演算法﹙Taguchi Genetic Algorithms，TGA﹚求取跨機台的最佳配方不但快速且有效，在本研究產品良率有約20%顯著改善，未來不僅可用於表面聲波器其他產品製程的最佳化取得，更可應用此方法於其他製程研究。

關鍵字

類神經網路與田口基因演算法

並列摘要

Abstract
The fabrication of inter-digital transducer (IDT) on piezoelectric material is the key of the surface acoustic wave (SAW) devices, which includes the photolithography process to control the width of IDT and the sputtering process to dominate the uniformity and thickness of deposited metal. The aim is to obtain the designed frequency. Even though the fabrication process of SAW devices is simpler than that of other semiconductor devices, the control of fabricated SAW frequencies and concentration level is difficult. To achieve a fabrication with low cost and high production yield, process optimization is the solution. In this study, an experimental design based on Taguchi method is used. Controllable parameters in this experiment are spinning speed, soft bake temperature and time, exposure time, hard bake temperature and time, and development time. Neural network is used to establish the process model. Then, the optimal recipe of SAW fabrication process is obtained by Taguchi genetic algorithm. In this study, we obtain the optimal recipe for a serial of process tools in the photolithography process.
By using radial basis function neural network (RBFN) and Taguchi genetic algorithm, an improvement in the fabrication of SAW devices was observed. The yield was raised up to 20%. This method can also be used for different processes.

並列關鍵字

Applying Neural Network and Taguchi Genetic Algorithm

參考文獻

[4] 林祖強,“微影製程之疊對控制﹐”中原大學碩士論文,2005

[7] 張益瑞,“銅化學機械研磨之製程參數最佳化﹐”中原大學碩士論文,2005.

[5] D. J. H. Wilson and G. W. Irwin, “RBF Principal Manifolds for Process Monitoring,”IEEE Transaction on neural Networks, Vol. 10

No. 6, November, 1999.

[14] J.-T. Tsai and J.-H. Chou, “Tuning the Structure and Parameters of a Neural Network by Using Hybrid Taguchi-Genetic Algorithm﹐”IEEE Transactions on Neural Networks, Vol. 17, No. 1, Januray 2006.