Detecting gaps in deep learning models used for mask process modeling

Ketan Sethi; Parikshit Kulkarni; Sabrina Aliyeva; Alex Zepka

doi:10.1117/12.2573032

13 October 2020 Detecting gaps in deep learning models used for mask process modeling

Ketan Sethi, Parikshit Kulkarni, Sabrina Aliyeva, Alex Zepka

Author Affiliations +

Proceedings Volume 11518, Photomask Technology 2020; 115180I (2020) https://doi.org/10.1117/12.2573032
Event: SPIE Photomask Technology + EUV Lithography, 2020, Online Only

Abstract

Deep neural networks (DNN) have shown excellent performance in classification and regression problems in multiple fields. Recent work has demonstrated the use of deep learning techniques for modeling mask processes (MILAN) where it greatly reduces the time and effort to build an accurate model [1]. Models used in semiconductor fabrication have rigorous requirements about generalization and users need to be aware of potential gaps in the model. It is therefore imperative that we can detect when a deep learning process model is not able to generalize on unseen data. We discuss an approach to solve this problem using variational autoencoders [2] in this paper. Variational autoencoders (VAE) are generative models that compute a probability distribution for describing an observation in the latent space of the model [3]. Therefore, any input to the model can be described in terms of probability distributions of its latent variables. We leverage this property of VAEs to augment MILAN models to determine their robustness. We also present results of the latent space distribution on unseen data when the deep learning model fails to generalize. The results show that our approach provides us with an ability to indicate vulnerabilities in deep neural network models used for mask manufacture proposed.

Conference Presentation

Citation Download Citation

Ketan Sethi, Parikshit Kulkarni, Sabrina Aliyeva, and Alex Zepka "Detecting gaps in deep learning models used for mask process modeling", Proc. SPIE 11518, Photomask Technology 2020, 115180I (13 October 2020); https://doi.org/10.1117/12.2573032

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

;

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE