Communications and Signal Processing Seminar

Machine Learning challenges in Metrology in Semiconductor Device Industry

Min-Yeong MoonLead Algorithm EngineerKLA
WHERE:
1690 Beyster BuildingMap
SHARE:

Abstract:  Metrology is critical for process and device performance control and its variation. Metrology is expected to increase when more complicated architecture is introduced in a new generation of devices. To automate process control in metrology system, machine learning (ML) and deep learning (DL) is widely adopted in various industry. The focus of the talk is to overview machine learning challenges in metrology problems to predict critical dimensions of critical layers in the semiconductor industry. Typical ML/DL models’ confidence is guaranteed given enough numbered referenced (labeled with ground truth) data. However, in practical industry, there often time exists an extremely small number of real ground truth data available due to the manufacturing cost. Moreover, ML/DL models are sensitive to complex process variations and varying tool states; and thus, the trained ML recipe may not be able to survive on the inline monitoring system. These variations may be gradual over time or abrupt such as an induced process change or erroneous flier wafers. The talk presents insight into an overview of how ML/DL is used in the semiconductor industry, what are ML challenges, and what could be potential solutions.

Bio:  MinYeong Moon is Lead Algorithm Engineer in the Advanced Algorithm Group at KLA Corporation. Her group develops advanced machine learning/deep learning algorithms to solve challenges in metrology. She obtained her Ph.D. in Mechanical Engineering at the University of Iowa in 2017. After that, she was a post-doctoral research scholar focusing on confidence-based reliability design optimization/reliability assessment/uncertainty quantification given the lack of testing data. Her primary focus currently broadly includes machine learning, deep learning, and probabilistic modeling, especially robustness improvement given the lack of limited real-world data and uncertainty quantification of machine learning model prediction.

***Event will take place in a hybrid format. The location for in-person attendance will be room 1690 Beyster Building.   Attendance will also be available via Zoom.

Join Zoom Meeting:  https://umich.zoom.us/j/91414297851

Meeting ID: 914 1429 7851

Passcode: XXXXXX (Will be sent via e-mail to attendees)

Zoom Passcode information is also available upon request to Michele Feldkamp ([email protected]) or Sher Nickrand([email protected]).

See full seminar by Professor Moon

Faculty Host

Qing QuAssistant Professor, Electrical Engineering and Computer ScienceUniversity of Michigan