Bulletin of the American Physical Society
73rd Annual Gaseous Electronics Virtual Conference
Volume 65, Number 10
Monday–Friday, October 5–9, 2020; Time Zone: Central Daylight Time, USA.
Session BM3: Workshop I: Artificial Intelligence & Machine Learning in Plasma Science and BeyondLive
|
Hide Abstracts |
Chair: Satoshi Hamaguchi, Osaka University, Japan |
Monday, October 5, 2020 8:30AM - 8:45AM Live |
BM3.00001: Artificial Intelligence {\&} Machine Learning in Plasma Science and Beyond; Introduction Invited Speaker: Satoshi Hamaguchi Artificial intelligence (AI) and machine learning (ML) can play important roles in plasma science and its application to a wide range of technologies in various ways. For example, they can be used to extract useful information from a large amount of data produced in experiments and numerical simulations. As the technologies for plasma diagnostics, supercomputing, and data management continue to advance and their costs continue to decrease, data produced in this field is expected to grow exponentially and must be analyzed nearly automatically. For physical phenomena that are too complex to be understood by deduction from the first principles, AI and ML may help one to find correlation or even causation among seemingly unrelated physical quantities or physical events, inferring possible underlying physical mechanisms or even phenomenological predictive models. In this Workshop, speakers specializing in different branches of plasma science and technologies will present their latest research work on the utilization of data in their own specialties. The Workshop is intended to provide an exemplary insight into this exciting development of “new means” of analyses in plasma science and technologies. [Preview Abstract] |
Monday, October 5, 2020 8:45AM - 9:15AM Live |
BM3.00002: Optimizing Etching Conditions by Reinforcement Learning and Data Compensation Method Invited Speaker: Hyakka Nakada To achieve nanoscale semiconductor-processing, the number of control parameters in etchers has increased. However, it has been difficult to optimize a recipe, that is an etching condition, with many parameters. Therefore, we proposed to explore optimal recipes for a target etching profile by AI. We present two exploring methods: one using reinforcement learning (RL) and the other using data compensation technique. They were developed to optimize recipes with multi-steps and recipes for fine patterns, respectively. We developed the method for multi-step etching by utilizing an in-situ optical monitor [1]. Applying RL to the monitor data, recipes can be optimized step by step. A vertical trench with a width of 750 nm and an aspect ratio (AR) of 4 was etched with predicted 5-step recipes. Next, we developed the method for fine pattern etching [2]. Nanoscale patterns are prone to collapse because of excessive side etching. To avoid the learning data shortage due to the collapse, missing data were compensated according to process knowledge. Applying supervised learning to the compensated data, a vertical trench with a width of 12.5 nm and an AR of 10 was etched with predicted recipes. [1] H. Nakada, et al., GEC2018, [2] H. Nakada, et al., DPS2019 [Preview Abstract] |
Monday, October 5, 2020 9:15AM - 9:45AM Live |
BM3.00003: Neural network surrogate modelling of tokamak plasma turbulence Invited Speaker: Jonathan Citrin Tokamak plasma confinement is constrained by turbulence. Reactor optimization demands accurate and tractable turbulence calculations, infeasible with high-fidelity DNS. Neural Network (NN) surrogate models circumvent the conflicting constraints of accuracy and tractability. A key enabling step is the development of reduced order models, validated by nonlinear simulations. Reduced model calculation time is sufficient for constructing extensive databases of model input-output mappings using HPC resources. These are used as training sets for NN regression. A key aspect is the physics-informed customization of regression variables and optimization cost functions, to capture known features of the system. The resultant NN transport models accurately reproduce the original reduced turbulence model with significant speedup, providing near-realtime capability, 1 trillion times faster than the anchoring nonlinear simulations. We summarize the state-of-the-art in tokamak transport NN surrogate development based on the QuaLiKiz model, ranging from grid-based input-space approaches, sampling input space based on pre-selected experimental databases including clustering and data-reduction approaches, and modification of the NN topologies to better capture the structure of the input-output mapping. [Preview Abstract] |
Monday, October 5, 2020 9:45AM - 10:15AM Live |
BM3.00004: Pairing plasma diagnostics, theory and simulations with machine learning to overcome the limitations of big data analytics in complex plasma processes Invited Speaker: Jun Shinagawa Big data analytics is the methodology around examining large data sets to uncover correlations among/within them and other insights of practical value. The application of big data analytics in industry is surging; the semiconductor industry is no exception. Typical examples involve advanced machine learning techniques such as PCA and pattern recognition. Rare are examples integrating the underlying physics for the purpose of extracting target metrics such as CD and etch depth. Plasma diagnostics paired with appropriate sensor technologies can reduce the advanced data processing load by enabling the direct extraction of variables that should correlate with target metrics via theory or a model. We call this “augmented big data analytics”. Here we present the successful real-world process example of pairing of plasma diagnostics with a for a critical device parameter. As a judge of success, we consider multiple factors and important ones are model robustness and need for retraining preventative maintenance (PM) cycles. We conclude with a few words on the importance of sensor-to-sensor and installation-to- installation variability. [Preview Abstract] |
Monday, October 5, 2020 10:15AM - 10:30AM |
BM3.00005: Break (10:15am - 10:30am) |
Monday, October 5, 2020 10:30AM - 11:00AM Live |
BM3.00006: Learning-based Control for Non-equilibrium Plasmas Invited Speaker: Ali Mesbah Learning-based control is a form of adaptive control, whereby controller and/or process model parameters are modified based on system measurements. Learning-based control can create unprecedented opportunities for process control of non-equilibrium plasmas (NEPs), which are increasingly used for treatment of heat and pressure sensitive (bio)materials in surface etching/functionalization, environmental, and biomedical applications. Some of the main challenges in process control of NEP applications arise from their inherent complexity and variability. Firstly, the dynamics of NEPs are highly nonlinear and spatio-temporally distributed, which are both expensive and also difficult to model due to their mechanistic complexity. Secondly, the NEP effects on complex surfaces are generally poorly understood. And thirdly, NEPs exhibit run-to-run variations and time-varying dynamics, whereby the same experiment may be carried out under similar conditions, but yield different results. In this talk, we will demonstrate the usefulness of learning-based predictive control approaches for NEP treatment of complex surfaces. We will discuss how machine learning approaches such as Gaussian process regression and Bayesian neural networks can be leveraged to learn the complex plasma and surface dynamics in real-time, toward safe and high-performance NEP treatment of complex surfaces. [Preview Abstract] |
Monday, October 5, 2020 11:00AM - 11:30AM Live |
BM3.00007: Data reprocessing in omics-driven approaches in plasma medicine Invited Speaker: Kristian Wende Omics-approaches like metabolomics, proteomics, and lipidomics generate massive data sets$.$ In plasma medicine, researcher seeks to unveil the impact of the treatment on cells or tissues. It is assumed that reactive species lead to signal processes that ultimately yield to physiological consequences. Two strategies are applied in our lab: the analyzing the direct impact of plasma-generated species on biomolecules (proteins, lipids, amino acids) and the analysis of expression changes in complex targets (skin, cancer models). The subsequent data analysis is the bottleneck, needing improvement. Especially in lipidomics, bottom-up approaches are hard to analyze. We tackle this situation by a number of bioinformatics workflows, engineering new analysis tools to identify phospholipid oxidation products from positive mode Orbitrap data. Additionally, the benefit of public data repositories to disseminate data and the comparison between CAP derived changes in the lipidome/proteome with other physical or chemical entities, e.g. radiation or pulsed electric fields is tested. [Preview Abstract] |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700