Paper Kerdprasop
Paper Kerdprasop
Paper Kerdprasop
Actual class
Class= 1 58 1
Class= -1 98 311
Class= 1 59 0
Class= 1 44 15
Class= 1 59 0
Class= -1 66 343
TABLE I
FAULT DETECTION MODEL ASSESSMENT
Fig. 7 TP rate versus FP rate comparison of each fault detection
model k-Nearest Logistic Naïve Decision
Neighbor Regression Bayes Tree
V. CONCLUSION [12] S.J. Qin, G. Cherry, R. Good, J. Wang, and C.A. Harrison,
“Semiconductor manufacturing process control and monitoring: a fab-
In semiconductor manufacturing process control and wide framework,” Journal of Process Control, vol.16, pp.179-191,
monitoring, hundreds of metrology data are available for 2006.
[13] E. Tafazzoli and M. Saif, “Application of combined support vector
process engineers to analyze for the purpose of maintaining machines in process fault diagnosis,” in Proc. American Control
efficient operations and getting optimum yield of high Conf., St. Louis, USA, 10-12 June 2009, pp.3429-3433.
quality products. For such a large volume of measurement [14] G. Verdier and A. Ferreira, “Fault detection with an adaptive distance
data, automatic fault detection technique is essential. We for the k-nearest neighbor rule,” in Proc. Int. Conf. Computer &
Industrial Engineering, Troyes, France, 6-9 July 2009, pp.1273-1278.
thus investigate the application of data mining techniques [15] G. Verdier and A. Ferreira. (2010). Adaptive Mahalanobis distance
such as decision tree induction, naïve Bayes analysis, and k-nearest neighbor rule for fault detection in semiconductor
logistic regression, and k-nearest neighbor classification for manufacturing. IEEE Trans. Semiconductor Manufacturing.
Available: doi:10.1109/TSM.2010.2065531
creating an accurate model for fault case detection in the [16] WEKA. (2010, October 23). Available: http://www.cs.waikato.ac.nz/
wafer fabrication process of semiconductor industries. ml/weka/
From a series of experimentation, we found that naïve
Bayes model built from a subset of features selected by a
gain ration criteria can detect the fault cases at the very high
rate of 90%. But the false alarm rate, or false positive, is
also as high as 80%. The decision tree method built from
our MeanDiff feature selection method generates a more
comprehensible form of fault detection model with false
alarm rate at only 4.5%. But the precision and true positive
rate, or recall, of the tree model are still low at 20.5% and
16%, respectively.
We thus devise a boosting technique to improve the
precision of tree-based model for fault detection by
pumping the number of rare cases, or fault test, to the equal
number of majority cases, or pass test. The outcome is
surprising that the true positive rate of the tree-based model
can increase up to 100%, whereas the false alarm rate is still
low at the 16%. We plan to investigate this boosting
technique to other domains that show imbalance among data
classes in our future research.
REFERENCES
[1] E. Epistola. (2010, December 4). Semiconductor manufacturing.
Available: http://www.siliconfareast.com/manufacturing.htm
[2] Z. Ge and Z. Song, “Semiconductor manufacturing process
monitoring based on adaptive substatistical PCA,” IEEE Trans.
Semiconductor Manufacturing, vol.23, no.1, pp.99-108, Feb. 2010.
[3] B.E. Goodlin, D.S. Boning, H.H. Sawin, and B.M. Wise,
“Simultaneous fault detection and classification for semiconductor
manufacturing tools,” Journal of The Electrochemical Society,
vol.150, no.12, pp.G778-G784, 2003.
[4] I. Guyon and A. Elisseeff, “An introduction to variable and feature
selection,” Journal of Machine Learning Research, vol.3, pp.1157-
1182, 2003.
[5] Q.P. He and J. Wang, “Fault detection using the k-nearest neighbor
rule for semiconductor manufacturing processes,” IEEE Trans.
Semiconductor Manufacturing, vol.20, no.4, pp.345-354, Nov. 2007.
[6] A.M. Ison, W. Li, and C.J. Spanos, “Fault diagnosis of plasma etch
equipment,” in Proc. IEEE Int. Symp. Semiconductor Manufacturing,
San Francisco, 6-8 October 1997, pp.B-49-B-52.
[7] A.M. Ison and C. Spanos, “Robust fault detection and fault
classification of semiconductor manufacturing equipment,” in Proc.
Int. Symp. Semiconductor Manufacturing, Tokyo, 2-4 October 1996.
[8] G.S. May and C.J. Spanos, Fundamentals of Semiconductor
Manufacturing and Process Control. John Wiley & Sons, 2006.
[9] M. McCann, Y. Li, L. Maguire, and A. Johnston, “Causality
challenge: benchmarking relevant signal components for effective
monitoring and process control,” in Proc. JMLR Workshop, Canada,
12 December 2008, pp.277-288.
[10] SEmi COnductor Manufacturing. (2010, November 25). Available:
http://www.causality.inf.ethz.ch/repository.php
[11] G. Spitzlsperger, C. Schmidt, G. Ernst, H. Strasser, and M. Speil,
“Fault detection for a via etch process using adaptive multivariate
methods,” IEEE Trans. Semiconductor Manufacturing, vol.18, no.4,
pp.528-533, Nov. 2005.