應用CART決策樹與資料視覺技術於低良率晶圓成因探討

Causing factor analysis of low-yield wafer using CART decision tree and data visualization

蔡智政

指導教授 ： 孫天龍

元智大學/工程學院/工業工程與管理學系/碩士(2002年)

摘要

由於半導體晶圓良率高低影響產品的成本以及競爭力，因此良率的改善成為各家廠商競爭的重要課題之一。在半導體製造過程中，都會蒐集晶圓通過機台時自動紀錄的參數資料來進行產品的監控或是製程分析，從這些資料中，工程師往往根據經驗選擇可能影響低良率的因子，利用實驗設計方法及實驗批來驗證，但是在半導體製程往往包含了數百個，甚至上千個互相影響的製程參數，要從中選擇出影響良率的因子非常困難，因此本研究應用決策樹演算法來輔助工程師找出影響製程變異的因子。半導體製程資料種類很多，其中晶圓批量製程管制(Lot in-process control, LPC)資料為晶圓在製程中經過每一道製程加工後記錄量測的資料，可以用來直接找出製程中造成低良率的成因，本研究使用具有處理大量連續型資料屬性之CART(classification and regression tree)決策樹演算法來分析LPC資料，除了Gini 指標外，我們也設計了兩種新的指標做為決策樹選擇分割條件的依據，利用此三種指標導出之LPC資料的決策樹，提供了製程工程師在一些參數條件組合下所產生的晶圓良率，可輔助製程工程師選擇影響製程良率的因子。

關鍵字

良率分析 ； IC製造 ； CART決策樹 ； 資料探勘 ； 資料視覺

並列摘要

Everyday, there are many production data in the corporation. It is difficult that get any information by analyzing production data rapidly in the situation, which fill up question, variable and competition everywhere. The people who manages the production can’t read and analysis the data which is large amount and variable immediately.
Yield analysis is critical for IC manufacturing since yield is directly related to production cast and competitively in the market. To monitor the manufacturing process and product quality, manufacturing data are automatically collected during wafer fabrication. Engineers use these data to select possible causing factors of low-yield wafer and employ statistical techniques (e.g. design of experiment) to verity the hypothesis. This approach, however, is difficult due to the large amount of parameters (from hundreds to thousands) and the complicated interactions among them. This research employs CART decision tree to help engineers select the possible causing factors if low-yield wafers. The input data is the lot in-process control (LPC) data, which are recorded by metrology machine during wafer fabrication. In addition to the Gini index, two indexes are developed for tree node splitting. Decision trees generated using these tree node splitting criteria provides engineers useful rules to analysis the causing factors of low-yield wafers.

並列關鍵字

yield analysis ； IC manufacturing ； CART decision tree ； data mining ； data visualization

參考文獻

Brada, D., and A. Shmilovici, “Data mining for improving a cleaning process in the semiconductor industry,” IEEE Transactions on Semiconductor Manufacturing, Vol. 15, No. 1, pp. 91-101, 2002.

Cunningham, S. P., C. J. Spanos, and K. Voros, “Semiconductor yield improvement : Results and best practices ,” IEEE Transactions Semiconductor Manufacturing, Vol. 8, pp. 103-109, 1995.

Inselberg, A., and B. Dimsdale, “Parallel coordinates: A tool for visualizing multi-dimensional geometry,” Proc. Visualization’90, San Francisco, CA, pp. 361-370, 1990.

Jaynes, E. T., Probability theory — the logic of science, Physics, St. Louis, 1994.

Keim, D. A., H. P. Kriegel, and M. Ankerst, “Recursive Pattern: A Technique for Visualizing Very Large Amounts of Data,” Proc. Visualization’95, Atlanta, GA, pp. 279-286, 1995.