2021
DOI: 10.1016/j.procs.2021.08.090
|View full text |Cite
|
Sign up to set email alerts
|

Anomaly detection for sensor data of semiconductor manufacturing equipment using a GAN

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
2
1

Citation Types

0
5
0

Year Published

2022
2022
2024
2024

Publication Types

Select...
5
1

Relationship

0
6

Authors

Journals

citations
Cited by 11 publications
(5 citation statements)
references
References 3 publications
0
5
0
Order By: Relevance
“…It is easy to understand that the left side has no anomaly s with little variation, while the right side has lots of anomaly situation with larger variation. Between the two, it is reasonable that display panels which are manufactured by the abnormal state equipment are more likely to have more defects, while panels produced by stable equipment are qualified [8].…”
Section: Figure 2 Data Elimination By Under-sampling (C) Over Samplingmentioning
confidence: 99%
“…It is easy to understand that the left side has no anomaly s with little variation, while the right side has lots of anomaly situation with larger variation. Between the two, it is reasonable that display panels which are manufactured by the abnormal state equipment are more likely to have more defects, while panels produced by stable equipment are qualified [8].…”
Section: Figure 2 Data Elimination By Under-sampling (C) Over Samplingmentioning
confidence: 99%
“…RAIM (recurrent attentive and intensive model; Xu et al, 2018 ) was a model including an attention mechanism, which used multi-channel attention to improve the prediction of the model. Hashimoto et al (2021) introduced RAIM into GAN to detect time series anomalies generated by semiconductor sensors. To this end, we consider a multi-channel attention mechanism, and an attention module is connected before both the encoder and the decoder.…”
Section: Proposed Approachmentioning
confidence: 99%
“…The input data is a series of time series data of length , where the data at time is an -dimensional vector. Let the -th subsequence be , when X is split by a sliding window ( Hashimoto et al, 2021 ). Then the importance of the time dimension is calculated by the following equation:…”
Section: Proposed Approachmentioning
confidence: 99%
“…Lee et al explored GAN‐based anomaly detection on temporal data by proposing a new evaluation framework, using simulated data and water treatment system data 8 . Hashimoto et al analyzed multivariate timing data of semiconductor manufacturing sensors with GAN and successfully detected anomalous defects of components that could not be directly observed 9 . Li et al proposed MAD‐GAN, an unsupervised multivariate anomaly detection method based on GAN, to perform fault detection on timing data from the cyber–physical system 10 …”
Section: Introductionmentioning
confidence: 99%
“…8 Hashimoto et al analyzed multivariate timing data of semiconductor manufacturing sensors with GAN and successfully detected anomalous defects of components that could not be directly observed. 9 Li et al proposed MAD-GAN, an unsupervised multivariate anomaly detection method based on GAN, to perform fault detection on timing data from the cyber-physical system. 10 To avoid blind training of traditional GAN, many studies have proposed GAN variants, among which the auxiliary classifier GAN 11 (ACGAN) based on CGAN has shown superior performance.…”
Section: Introductionmentioning
confidence: 99%