The research of apply Data Mining to Machine assembly in semiconductor and to WAT mode

• Main Authors: Yi-Ying Yang, 楊怡穎
• Other Authors: 王建智
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/ye99cc

碩士 === 明志科技大學 === 工程管理研究所 === 94 === In the semiconductor manufacturing process, the engineers always cannot find out the abnormal process stations and machines quickly, based on their know-how and experience, from the huge database, because the manufacturing process is very precise and complicated. Moreover, there are many reasons causing finished products functional failed. However, past research regarded machine assembling as random factor and seldom studied the influence on yield rate by machine assembling. And most of the researches were more focusing on facility planning and production line balancing and rarely mentioned the correlation between semiconductor machine assembly and the yield rate. This paper is taking one semiconductor manufacturing DRAM for example, setting up an analytical model, doing data mining from huge database, probing into whether machine assembly affects Vt parameter of Wafer Acceptance Test or not, and finding out the particular station or machine that may have an effect on the manufacturing process. First of all, this research is grouping the data with I-MR control chart and K -times standard deviation grouping method, then establishing the rules of machine assembling using the association rule. It analyzes and verifies thirty-seven hundred and twenty data from the actual numbers provided by the merchant. Concerning the support percentage for judging standard of association rule, support> 80%, 75 %< support <80%, and 70% <support <75% are set up, confidence percentage is defined as >85%. From the analyzing result, it is suggested that if the support percentage is between 70% and 80% and confidence percentage equals 85%, via the association rule, thirteen stations affecting Vt parameter more are picked up quickly out of thirty-nine stations. Those abnormal machines gather at station 046000(EH308), 049200(DW005), 049050(DI303), and 060000(DC393) in the later half manufacturing process of that factory. After verified by engineers, the result coheres with the current situation of that factory. It is said that the analyzing method in this research is correct. The association rule defining abnormal machines by this research is not only used to detect machine during manufacturing process of semiconductor product, but also used to provide references for engineers to analyze and schedule the production.