The Use of Mahalanobis-Taguchi System in Solving the Multidimensional Quality Characteristics Problems

• Main Authors: Yuan-TingCheng, 鄭元婷
• Other Authors: Taho Yang
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/75786284854781186055

博士 === 國立成功大學 === 製造資訊與系統研究所碩博士班 === 98 === Quality inspection is one of important issues in the optical electrical and semiconductor industries which are two of high-tech industries in Taiwan. These two industries need huge investments. Therefore, how to increase the inspection efficiency of processing quality to raise the production capacity has become an important issue. A good quality inspection method must have a good diagnosis and forecasting capability, especially in the high-tech industries, which reveals very limited numbers of nonconforming items. A robust quality inspection method is necessary and the method must be not influenced by data distribution. MTS is a new diagnosis and forecasting technique for multivariate data. MTS establishes a classification model by constructing a continuous measurement scale rather than learning from training data set. Therefore, MTS is not influenced by data distribution. In many applications, Mahalanobis-Taguchi System (MTS) verified that it has a more robust classification and enhanced features to allow improved selection. In this study, MTS is used to solve the sputtering process thin-film thickness uniformity quality problem in the optical electrical industry and to improve the bumping process inspection efficiency in the semiconductor industry by using its classification and ability to selection important features. To solve the sputtering process thin-film thickness uniformity quality problem, the mean, the standard deviation and the range of the thickness are adopted as the quality characteristics, as opposed to the general approach that only considers the thickness mean in the measurements of thin-film thickness uniformity. The present study proposes Mahalanobis distance (MD) as a method to detect the nonconforming items in the sputtering process. The empirical results demonstrate that the MD method is more accurate and efficient in solving a multivariate, class imbalance problem. Moreover, two stages quality inspection is proceeded in order to improve the bumping process inspection efficiency and reduce cost by selecting suitable inspection positions. In the first stage, the quality inspection system of the single quality is constructed and the suitable inspection positions are selected; in the second stage, the quality inspection system of the multidimensional quality is constructed and the suitable inspection positions are selected. In the experiment with first stage, the concept of control chart is used to generate a suitable MS and the bisection algorithm is used to determine the threshold value. The empirical result demonstrates that the numbers of bump height inspection features are significantly reduced from 10 to 6 without losing classification accuracy; and inspection time can be reduced by 40% in the single quality inspection system. In the experiment with second stage, the concept of MD is used to convert multidimensional inspection quality characteristics into a single-dimensional inspection index and subsequently MTS is used to reduce costly inspection time and maintain the quality of the bumping process. The empirical result demonstrates that the numbers of bump height inspection features are significantly reduced from 10 to 7 without losing classification accuracy; and inspection time can be reduced by 30% in the multidimensional quality inspection system. Furthermore, by virtue of reducing the bump height inspection position, the time of bump inspection was reduced. The inspection staff can select the suitable inspection position in sequence according to the significance of features selected by the MTS method. Meanwhile, they can reduce the number of inspection positions when checking acceptable quality of bump height. On the other hand, SPC has been used in this study, to select a suitable normal example for the full MS measurement; while it can also monitor the process. If the process detects defects that are distinct from preceding shift, the MS will be re-calculated in order to maintain the quality of the process.