5월 24일 (수) 세미나 안내

세미나 주제: AI 기술을 활용한 반도체 simulation

세미나 강사 : 선행 simulation 그룹 허인

(Bio)
– ’22 SEC Annual Awards 수상
– ’17 삼성전자 DS 부문 반도체 연구소 입사
– ’17 서강대 석사
– ’15 서강대 학사

(Title) Structure-preserving representation learning: its application to semiconductor design analysis

(Abstract) In recent years, there has been a significant surge of interest in learning the underlying geometric, topological and symmetry structures of data by using machine learning models. In parallel, an emerging set of findings in the deep learning community suggests that preserving the geometric structures of data may be a fundamental principle of achieving a good representation of data.
In this talk, I will review some important literatures on the topic of such a structure-preserving representation learning. It will cover deep generative frameworks combined with geometric concepts including the group invariance and Riemannian geometry. Then, I will present my recent work on the deep representation model that introduces the concept of the quotient manifold and local isometry to the variational auto-encoders. Instead of discussing theoretical properties of the developed model, I will emphasize the application of it from the perspective of the semiconductor engineering field, including the use-cases in integrated circuit layout analysis and wafer map clustering. I will conclude my talk with briefly outlining the future research direction with some challenging real-world problems in the semiconductor manufacturing process.