Publications

Detailed Information

Effective Data Selection for Robust Training of Generative Adversarial Network for Lithography Pattern Alignment : 리소그래피 패턴 정렬용 생성적 적대 신경망의 강건한 훈련을 위한 효과적인 데이터 선택

DC Field Value Language
dc.contributor.advisor김도년-
dc.contributor.author곽노홍-
dc.date.accessioned2023-06-29T01:50:24Z-
dc.date.available2023-06-29T01:50:24Z-
dc.date.issued2023-
dc.identifier.other000000174406-
dc.identifier.urihttps://hdl.handle.net/10371/193077-
dc.identifier.urihttps://dcollection.snu.ac.kr/common/orgView/000000174406ko_KR
dc.description학위논문(석사) -- 서울대학교대학원 : 공과대학 기계공학부, 2023. 2. 김도년.-
dc.description.abstractCritical Dimension SEM (CD-SEM) is a dedicated system for measuring the shape, size and roughness of patterns formed on semiconductor wafers. As designs shrink and product development challenges increase, the ability to quickly measure large amounts of samples for accurate Optical Proximity Correction (OPC) is required. Design Based Metrology (DBM) technology allowed the rapid creation of large volumes of recipes using design images and reduced measurement time. However, there were still many problems in the alignment between the design image and the SEM image, and to solve this problem, a new pattern alignment method using Generative Adversarial Network (GAN) technology was developed. In this paper, training patterns are classified according to polygon types of design patterns and the alignment effect according to each type is confirmed. We also studied how to effectively select a training set for model training through the relationship between training set and alignment accuracy.-
dc.description.abstractCritical Dimension SEM(CD-SEM)은 반도체 웨이퍼에 형성된 패턴의 모양, 크기 및 거칠기를 측정하는 전용 시스템이다. 설계가 축소되고 제품 개발 과제가 증가함에 따라 정확한 Optical Proximity Correction (OPC)를 위해 대량의 샘플을 신속하게 측정할 수 있는 기능이 필요하다. Design Based Metrology (DBM) 기술을 통해 설계 이미지를 사용하여 대량의 레시피를 빠르게 생성하고 측정 시간을 단축할 수 있었다. 그러나 디자인 이미지와 SEM 이미지 간의 정렬에는 여전히 많은 문제가 있었고, 이를 해결하기 위해 Generative Adversarial Network (GAN) 기술을 사용한 새로운 패턴 정렬 방법이 개발되었다. 본 논문에서는 디자인 패턴의 폴리곤 유형에 따라 학습 패턴을 분류하고 각 유형에 따른 정렬 효과를 확인하였다. 또한 훈련 세트와 정렬 정확도의 관계를 통해 모델 훈련을 위한 훈련 세트를 효과적으로 선택하는 방법을 연구하였다.-
dc.description.tableofcontentsChapter 1. Introduction 1
1-1. Semiconductor design and optical proximity correction 2
1-2. SEM inspection and design-based metrology 5
1-3. A new method of lithography pattern alignment 9
Chapter 2. Motivation 16
2-1. Relationship between accuracy and sample set 17
Chapter 3. Experiments 24
3-1. Relationship between train pattern size and alignment result 26
3-1-1. The result of inferring an image twice the size of the pattern learned by the model 28
3-1-2. The result of inferring an image three times the size of the pattern learned by the model 29
3-1-3. The result of inferring an image four times the size of the pattern learned by the model 30
3-2. Dense-to-iso or iso-to-dense pattern relationship 32
3-2-1. Dense to iso pattern relationship 33
3-2-2. Iso to dense pattern relationship 34
3-3. The Pattern shifted and direction 35
Chapter 4. Methodology and Result 39
4-1. How to select the minimum sample that can cover the entire pattern 42
4-2. Proposed method and improvement 45
4-2-1. Proposed method 46
4-2-2. Improvement using effective distance based pattern extraction 49
Chapter 5. Conclusion 56
Bibliography 59
Abstract in Korean 66
-
dc.format.extentvii, 66-
dc.language.isoeng-
dc.publisher서울대학교 대학원-
dc.subjectScanning Electron Microscopy-
dc.subjectDesign-based metrology-
dc.subjectGenerative Adversarial Network-
dc.subjectSupervised learning-
dc.subject.ddc621-
dc.titleEffective Data Selection for Robust Training of Generative Adversarial Network for Lithography Pattern Alignment-
dc.title.alternative리소그래피 패턴 정렬용 생성적 적대 신경망의 강건한 훈련을 위한 효과적인 데이터 선택-
dc.typeThesis-
dc.typeDissertation-
dc.contributor.AlternativeAuthorNohong Kwak-
dc.contributor.department공과대학 기계공학부-
dc.description.degree석사-
dc.date.awarded2023-02-
dc.contributor.major역학-
dc.identifier.uciI804:11032-000000174406-
dc.identifier.holdings000000000049▲000000000056▲000000174406▲-
Appears in Collections:
Files in This Item:

Altmetrics

Item View & Download Count

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share