TY - JOUR
T1 - A look-ahead back propagation network to predict wafer lot output time
AU - Chen, Tin-Chih
PY - 2006/5/1
Y1 - 2006/5/1
N2 - Output time prediction is a critical task to a wafer fab (fabrication plant). However, traditional wafer lot output time prediction methods are based on the historical data of the fab. The importance of the (future) release plan has been neglected. In addition, a lot that will be released in the future might appear in front of another lot that currently exists in the fab. For these reasons, to further improve the accuracy of wafer lot output time prediction, the future release plan of the fab has to be considered and the traditional BPN can be restructured by incorporating the plan. A look-ahead BPN called the BPNf is then constructed. For evaluating the effectiveness and efficiency of the BPNf and to make some comparisons with two traditional approaches - BPN and CBR, PS is applied in this study to generate test data. Then all the three methods are applied to five cases elicited from the test data. According to experimental results, the prediction accuracy represented with the RMSE of the BPNf was significantly better than those of the other approaches in all cases by achieving a 5%-24% (and an average of 12%) reduction in the RMSE over the comparison basis - the BPN. The advantage over the CBR was 1%-12%.
AB - Output time prediction is a critical task to a wafer fab (fabrication plant). However, traditional wafer lot output time prediction methods are based on the historical data of the fab. The importance of the (future) release plan has been neglected. In addition, a lot that will be released in the future might appear in front of another lot that currently exists in the fab. For these reasons, to further improve the accuracy of wafer lot output time prediction, the future release plan of the fab has to be considered and the traditional BPN can be restructured by incorporating the plan. A look-ahead BPN called the BPNf is then constructed. For evaluating the effectiveness and efficiency of the BPNf and to make some comparisons with two traditional approaches - BPN and CBR, PS is applied in this study to generate test data. Then all the three methods are applied to five cases elicited from the test data. According to experimental results, the prediction accuracy represented with the RMSE of the BPNf was significantly better than those of the other approaches in all cases by achieving a 5%-24% (and an average of 12%) reduction in the RMSE over the comparison basis - the BPN. The advantage over the CBR was 1%-12%.
KW - Back propagation network
KW - Future release plan
KW - Fuzzy set application
KW - Output time prediction
UR - http://www.scopus.com/inward/record.url?scp=33744547519&partnerID=8YFLogxK
M3 - Article
AN - SCOPUS:33744547519
VL - 5
SP - 910
EP - 915
JO - WSEAS Transactions on Computers
JF - WSEAS Transactions on Computers
SN - 1109-2750
IS - 5
ER -