Harnessing Decision Tree to Analyze the Abnormality in the Semiconductor Manufacturing Industry-A Case Study from S Company

• Main Authors: CHEN CHAO-HSIEN, 陳昭賢
• Other Authors: 黃正魁
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/uh3gsy

碩士 === 國立中正大學 === 企業管理學系碩士在職專班 === 105 === The semiconductor industry has been treated as one of benchmark developments in Taiwan. Taiwan is the largest chip-manufacturing base in the world. The total sales of semiconductor industry is over one fifth of the global semiconductor industry. The wafer production is the core component of various devices, such as computers, smart phones, cameras, and other technology equipments. This industry has bringing huge economic values and employment opportunities to Taiwan. However, to meet the progressively matured technologies, many international competitors are bogged down the price competition, resulting in profit squeeze for all manufacturers. Hence, the semiconductor firms are facing a difficult environment so that they have to maintain a strict cost control for keeping their competitive advantage. For performing the strategy of cost control, an abnormality detection in the production process is one of approaches to reduece the cost of production. Howeve, the abnormality, sometimes, is not contributing to only one machine in a work station but is from a covariance factor among different machines in different work stations. The sophisticated managers or enginners could not easily figure out the problem at first with their traditional approach, i.e. statistical method. The reason is that they need to address all possible cases from different machines of different work stations and then verifiy all of them for the abnormality detection. Unfortunately, the detection would be taking exponential time to find the outcome as the number of machines or work stations is increasing. In this study, we utilize a data mining technique, decision tree, to address the problem above. Its advantages are (1) it can discover the correction among the different machines of different work stations and does not need to combine all possible cases and (2) as collecting the data from the manufacturing process, managers or engeineers can systematically and efficiently get a result without their past experience. In the experiment of this study, we employ a real-world datasest to test the effectiveness of this technique. Keywords: Semiconductor, Abnormality analysis, Data mining, Decision tree