Using Uniformity to Predict Wafer Yield - A Feasibility Study of a DRAM Foundry

• Main Authors: Jen-hsu Lin, 林仁旭
• Other Authors: Cheng-Kiang Farn
• Format: Others
• Language: zh-TW
• Published: 2008
• Online Access: http://ndltd.ncl.edu.tw/handle/15354275312078395297

碩士 === 國立中央大學 === 資訊管理學系碩士在職專班 === 96 === Abstract DRAM plays a central role in the electronics industry due to its wide spread applications in various products, like PC, mobile phone, optical disk drive, digital camera, printer, digital TV, etc. However, DRAM is a standardized product, and is treated as a commodity in the market. Thus, a DRAM manufacturer cannot keep its customer with unique product specifications like other semiconductor ICs. The most important issue in improving profitability is to reduce its production costs. In general, there are three methods to reduce its production costs: reducing manufacturing cycle time, introducing advanced technology and improving yield. Yield improvement is very important for semiconductor industry. The company in this study produces about 1,000 million chips per year. If the yield rate can be improved by 1%, it translates to an additional 10 million finished chips per year without further investment. Assuming the selling price of a chip to be 2 USD, it can result in additional 20 million USD per year in profit for the company. Thus, every DRAM foundry tries their best to find any possible means to improve its yield rate. Among the various plans that the company in this study attempts, is to improve its production uniformity, which is a statistics collected for daily quality management purpose, hoping that it would improve yield. It sounds logical to use uniformity as an indicator for predicting yield, but whether it is an effective indicator in practice is not known. Thus, this study attempts to investigate the feasibility by statistically analyzing the data collected previously, to establish any relationship between yield and uniformity. Results of the in depth analysis show that uniformity alone, as collected for quality management purpose, is not an effective predictor of yield. Therefore, we do not recommend using the uniformity statistics as a predictor of yield.