TY - GEN
T1 - Similarity-based Local Feature Extraction for Wafer Bin Map Pattern Recognition
AU - Kim, Jieun
AU - Baek, Jun Geol
N1 - Funding Information:
This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (NRF-2019R1A2C2005949). Also, this work was supported by Brain Korea 21 FOUR and Samsung Electronics Co., Ltd(IO201210-07929-01).
Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - A wafer bin map consists of a local chip containing key information and a global chip present in all patterns. The defect pattern shows a specific pattern shape on the wafer bin map and is defined based on the existing area information. Global information is not differentiated from local information in classification problems and is recognized as a major characteristic, so it affects the identification of the characteristics of defective patterns. In preparation for this, a method of extracting key local information has been proposed. In this paper, we propose a Skip Connections Denoising Autoencoder-based methodology to extract regional information of defect patterns. Randomly distributed chips are recognized as noise by defining anomaly scores based on the probability of each chip appearing in the wafer bin map. We propose a data transformation and reconstruction methodology for extracting local information based on the anomaly score, which is an uncertainty score index. Through the proposed methodology, it was confirmed that the main information that could not be extracted from the convolutional neural network (CNN) was extracted, and it was confirmed that the method proposed in this paper for WM-811K data is superior to the existing method.
AB - A wafer bin map consists of a local chip containing key information and a global chip present in all patterns. The defect pattern shows a specific pattern shape on the wafer bin map and is defined based on the existing area information. Global information is not differentiated from local information in classification problems and is recognized as a major characteristic, so it affects the identification of the characteristics of defective patterns. In preparation for this, a method of extracting key local information has been proposed. In this paper, we propose a Skip Connections Denoising Autoencoder-based methodology to extract regional information of defect patterns. Randomly distributed chips are recognized as noise by defining anomaly scores based on the probability of each chip appearing in the wafer bin map. We propose a data transformation and reconstruction methodology for extracting local information based on the anomaly score, which is an uncertainty score index. Through the proposed methodology, it was confirmed that the main information that could not be extracted from the convolutional neural network (CNN) was extracted, and it was confirmed that the method proposed in this paper for WM-811K data is superior to the existing method.
KW - Anomaly localization
KW - Data augmentation
KW - Defect pattern recognition
KW - Semiconductor manufacturing process
KW - Skip connections denoising autoencoder
UR - http://www.scopus.com/inward/record.url?scp=85127675112&partnerID=8YFLogxK
U2 - 10.1109/ICAIIC54071.2022.9722665
DO - 10.1109/ICAIIC54071.2022.9722665
M3 - Conference contribution
AN - SCOPUS:85127675112
T3 - 4th International Conference on Artificial Intelligence in Information and Communication, ICAIIC 2022 - Proceedings
SP - 56
EP - 59
BT - 4th International Conference on Artificial Intelligence in Information and Communication, ICAIIC 2022 - Proceedings
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 4th International Conference on Artificial Intelligence in Information and Communication, ICAIIC 2022
Y2 - 21 February 2022 through 24 February 2022
ER -