Improvement For Semiconductor Porcess Yield By Big Data Analysis Technology With A Machine Learning And Statistical Method.

• Main Authors: Shou-Cheng Cheng, 鄭守成
• Other Authors: Chung-Hong Lee
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/89841155167867728899

碩士 === 國立高雄應用科技大學 === 電機工程系博碩士班 === 103 === It is a well-known fact that Taiwan is of superior manufacturing technology on semiconductor in the world, and one of the main characteristics is that the related manufacturers can produce quality products in a very short time. In the production of semiconductor encapsulation, the production yield is one of critical issues concerned by various foundries. It is because that yield rate can directly affects the quality of the final product and the profitability. When abnormal problems of production quality occurred, engineers should carry out the subsequent revision action immediately and make trouble-shooting analysis to improve production line according to the relevant production records. As a result, it would be useful for finding out possible factors that may affect the product quality in advance, by using the large-volume of related data collected in the manufacturing process, and analytical techniques of such big data. Also, such methods are beneficial in detecting abnormality, in order to make a significant improvement for production yield and reducing of production cost and shorten the time of manual inspection and classification. In this work we took the analysis of defect product samples during encapsulation manufacturing process as an example for case study, and as such we applied machine learning techniques to classify the defect products and verified the resulting accuracy to insure the feasibility of the experimental result, in order for establishing a classification system to reduce the cost of manual inspection. This work started with the utilization of Support Vector Machines (SVM) method to carry out the task of classification, and then compared the accuracy with the Back-propagation net (BPN) model. After that, a statistical analysis technique, Pearson product-moment correlation coefficient method, was utilized to formulate the influential factors for production quality and explain the related concept and effectiveness of big data analysis.