Anomaly Detection for Semiconductor Tools Using Stacked Autoencoder Learning

Liao, Da-Yin; Chen, Chieh-Yu; Tsai, Wen-Pao; Chen, Hsuan-Tseng; Wu, Y.-W.; Chang, Shuo‐Hung

doi:10.1109/issm.2018.8651179

Cited by 14 publications

(13 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In semiconductor manufacturing, most of ESD time series when plotted over time axis approximately show piecewise linear value curves [ 31 ]. For example, the sample sequence of the cycle depicted by Figure 1 starts at value 250, rises to a range between 325 and 375 after 30 s, and drops back to 250 at 180 s [ 27 , 28 ].…”

Section: Anomaly Detection Problems and Challengesmentioning

confidence: 99%

“…In semiconductor equipment AD, there have been only few spectral-analysis related approaches. Liao et al [ 27 ] considered the spectral transformation of ESD cycles and preliminarily investigated the value of spectral analysis for semiconductor AD. Chen et al [ 28 ] further indicated how spectral analysis helps detect a drift in ESD anomaly at very low frequencies.…”

Section: Introductionmentioning

confidence: 99%

“…Liao et al [ 27 ] assessed the feasibility of using a Stacked AutoEncoder (SAE) for unsupervised learning of normal features for AD from semiconductor tool ESD. An SAE contains multiple hidden layers of nodes in addition to the input and output layers and can be viewed as the “stacking” of multiple general autoencoders [ 29 ].…”

Section: Introductionmentioning

confidence: 99%

“…In this paper, on top of the preliminary design and insights achieved by Liao et al [ 27 ] and Chen, Chang and Liao [ 28 ], we shall further define four design problems of semiconductor equipment AD exploiting ESD. To address the four problems and resolve their challenges, we shall propose a novel framework of spectral and time autoencoder learning for anomaly detection (STALAD).…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Equipment Anomaly Detection for Semiconductor Manufacturing by Exploiting Unsupervised Learning from Sensory Data

Chen

Chang

Liao³

2020

Sensors

Self Cite

View full text Add to dashboard Cite

In-line anomaly detection (AD) not only identifies the needs for semiconductor equipment maintenance but also indicates potential line yield problems. Prompt AD based on available equipment sensory data (ESD) facilitates proactive yield and operations management. However, ESD items are highly diversified and drastically scale up along with the increased use of sensors. Even veteran engineers lack knowledge about ESD items for automated AD. This paper presents a novel Spectral and Time Autoencoder Learning for Anomaly Detection (STALAD) framework. The design consists of four innovations: (1) identification of cycle series and spectral transformation (CSST) from ESD, (2) unsupervised learning from CSST of ESD by exploiting Stacked AutoEncoders, (3) hypothesis test for AD based on the difference between the learned normal data and the tested sample data, (4) dynamic procedure control enabling periodic and parallel learning and testing. Applications to ESD of an HDP-CVD tool demonstrate that STALAD learns normality without engineers’ prior knowledge, is tolerant to some abnormal data in training input, performs correct AD, and is efficient and adaptive for fab applications. Complementary to the current practice of using control wafer monitoring for AD, STALAD may facilitate early detection of equipment anomaly and assessment of impacts to process quality.

show abstract

Section: Anomaly Detection Problems and Challengesmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Equipment Anomaly Detection for Semiconductor Manufacturing by Exploiting Unsupervised Learning from Sensory Data

Chen

Chang

Liao³

2020

Sensors

Self Cite

View full text Add to dashboard Cite

show abstract

“…Barbee (2007)). Since 2010, there has been a positive trend in general, such as with studies from Meidan et al (2011) and Moyne et al (2014)) that turned into a significant increase from 2017 to 2018 (e.g., with studies from Chiu et al (2017) and Liao et al (2018)). This increase correlates with the data-driven trends in manufacturing that have been identified and discussed in 2.3.…”

Section: Overviewmentioning

confidence: 99%

Analysis and Evaluation of the Impacts of Predictive Analytics on Production System Performances in the Semiconductor Industry.

Preis¹

View full text Add to dashboard Cite

Problem Statement: Predictive Analytics (PA) may effectively support semiconductor industry (SI) companies in order to manage the special challenges in SI value chains. To discover the implications of PA, the realistic benefits as well as its limitations of its application to semiconductor manufacturing, it is necessary to assess in which ways the application of PA affects the production system (PS) performances. However, based on the literature survey, the influences of PA on the various performance characteristics of an SI PS are not as clear as expected for the efficiently operative application. Besides, the existing performance models are not effective to predict the impacts of PA on the SI PS performances. Therefore, the overall aim of this thesis is to analyse and evaluate the impacts of PA on the SI PS performances and to identify under which conditions a PA application would generate the most significant performance improvements. The focus of this thesis is predictive maintenance (PdM). Research Methodology: Based on a post-positivist philosophy, the thesis applies a deductive research approach using mixed-methods for data collection. The research design has the following stages: (1) theory, (2) hypothesis, (3) state of research, (4) case study and (5) verification. Main Achievements: (1) The systematic literature review is carried out to identify the gaps of the existing research and based on these findings, a conceptual framework is proposed and developed. (2) The existing performance models are analysed and evaluated against their applicability to this study. (3) A causal loop model for SI PS is generated based on the assessment of experts with industrial engineering and equipment maintenance expertise. (4) An expert system is developed and evaluated in order to investigate transitive and contradictory effects of PdM on SI PS performances. (5) A simulation model is developed and validated for investigating the strengths and limitations of PdM regarding SI PS performances under different circumstances. Results: The results of the logical inference study show that PdM has 34 positive effects as well as 4 contradictory effects on SI PS performance characteristics. Based on the various simulation experiments, it has been found that (1) ’Mean Time to Repair’ decreases only if PdM supports proportionate reduction of failures and repair times. (2) Logistics performance improves only if the underlying workcenter is limited in capacity or the four partners are nonsynchronous. (3) PdM supports optimal cost decreases for workcenters where the degree of exhausting wear limits can be most effectively improved and (4) the degree of yield improvement gained by PdM is dependent on the operation scrap rate. However, (5) if a workcenter has overcapacity, PdM will potentially worsen PS performances, even if the particular workcenter performance can be improved. These new insights advance existing knowledge in production managements when adopting predictive technologies at SI PS in order to improve PS performances. The findings above enable SI practitioners to justify a PdM investment and to select suitable workcenters in order to improve SI PS performances by applying the proposed PdM. Contributions: The main contributions of this PhD project can be divided into practical application and theoretical work. The contributions from the theoretical perspective are: 1) The critical review and evaluation of the state of the research for PA in the context of semiconductor manufacturing and the models for predicting and evaluating SI PS performances. 2) A new framework for investigating the implications of PA on the challenges such as gaining high utilizations and controlling the variability in production processes in SI value chains. 3) The new knowledge about transitive and contradictory effects of PdM on SI PS performances, which indicates that PdM can be used to improve PS performances beyond a single machine. 4) The new knowledge about strengths and limitations of PdM in order to improve SI PS performances under particular circumstances. The contributions from the practical application perspective are: 1) A practical method for identifying workcenters where PdM delivers the most significant benefits for SI PS performances. 2) An expert system that provides a comprehensive knowledge base about causes and effects within SI PS in order to justify a PdM investment. 3) A concise review of important PA applications, their capabilities for the wafer fabrication and the most suited PA methods. These findings can be adopted by SI practitioners.

show abstract

The anomaly detection mechanism using deep learning in a limited amount of data for fog networking

Horng

Liu

Hsu

2021

Computer Communications

View full text Add to dashboard Cite

Anomaly Detection for Semiconductor Tools Using Stacked Autoencoder Learning

Cited by 14 publications

References 4 publications

Equipment Anomaly Detection for Semiconductor Manufacturing by Exploiting Unsupervised Learning from Sensory Data

Equipment Anomaly Detection for Semiconductor Manufacturing by Exploiting Unsupervised Learning from Sensory Data

Analysis and Evaluation of the Impacts of Predictive Analytics on Production System Performances in the Semiconductor Industry.

The anomaly detection mechanism using deep learning in a limited amount of data for fog networking

Contact Info

Product

Resources

About