Hyungmin An
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Hyungmin An has a total of 20 co-authors. Here are the top co-authors by number of shared publications:
- Seungwu Han
Seoul National University
6 shared publications - Sangmin Oh
Seoul National University
3 shared publications - Changho Hong
Department of Materials Science and Engineering and Research Institute of Advanced Materials, Seoul National University, Seoul 08826, Korea
2 shared publications - Dongyean Oh
SK Hynix Inc, Icheon-si, Gyeonggi-do 17336, Republic of Korea
2 shared publications - Gijin Kim
Department of Materials Science and Engineering, Seoul National University, Seoul 08826, Korea
2 shared publications Jae-Hyeon Ko
2 shared publications
Latest Publications
- 20 July 2026
- Published by American Chemical Society (ACS) in Journal of Chemical Information and Modeling
- Vol. 66 (15) , 8875-8892
A Lightweight Universal Machine-Learning Interatomic Potential via Knowledge Distillation for Scalable Atomistic Simulations
We introduce a lightweight universal machine-learning interatomic potential (uMLIP), SevenNet-Nano, based on the graph neural network architecture SevenNet and enabled by a knowledge-distillation framework. The model inherits the broad generalization capability of a large multitask foundation model, SevenNet-Omni, trained on diverse materials data sets across chemical, configurational, and computational spaces. By learning chemical representations from high-quality inference data generated by the teacher model within a unified computational framework, SevenNet-Nano achieves high accuracy and strong transferability despite its compact architecture. The model also accurately captures a wide range of interatomic interactions, enabling reliable simulations under both equilibrium and extreme conditions, including plasma etching of SiO2. Comprehensive benchmarks on static and dynamical properties─such as Li-ion diffusion and liquid densities─demonstrate its broad applicability across diverse domains, with optional fine-tuning to further enhance performance if required. Importantly, SevenNet-Nano significantly reduces computational cost, achieving over an order-of-magnitude speedup and enabling large-scale atomistic simulations involving thousands of atoms.
- 20 July 2026
- Published by American Chemical Society (ACS) in Journal of Chemical Information and Modeling
- Vol. 66 (15) , 8875-8892
A Lightweight Universal Machine-Learning Interatomic Potential via Knowledge Distillation for Scalable Atomistic Simulations
We introduce a lightweight universal machine-learning interatomic potential (uMLIP), SevenNet-Nano, based on the graph neural network architecture SevenNet and enabled by a knowledge-distillation framework. The model inherits the broad generalization capability of a large multitask foundation model, SevenNet-Omni, trained on diverse materials data sets across chemical, configurational, and computational spaces. By learning chemical representations from high-quality inference data generated by the teacher model within a unified computational framework, SevenNet-Nano achieves high accuracy and strong transferability despite its compact architecture. The model also accurately captures a wide range of interatomic interactions, enabling reliable simulations under both equilibrium and extreme conditions, including plasma etching of SiO2. Comprehensive benchmarks on static and dynamical properties─such as Li-ion diffusion and liquid densities─demonstrate its broad applicability across diverse domains, with optional fine-tuning to further enhance performance if required. Importantly, SevenNet-Nano significantly reduces computational cost, achieving over an order-of-magnitude speedup and enabling large-scale atomistic simulations involving thousands of atoms.
- 22 September 2025
- Published by Elsevier in Applied Surface Science
- Vol. 716
Etching-to-deposition transition in SiO2/Si 3 N 4 using CH x F y ion-based plasma etching: An atomistic study with neural network potentials
- 22 September 2025
- Published by Elsevier in Applied Surface Science
- Vol. 716
Etching-to-deposition transition in SiO2/Si 3 N 4 using CH x F y ion-based plasma etching: An atomistic study with neural network potentials
- 29 May 2025
- Published by IOP Publishing in Modelling and Simulation in Materials Science and Engineering
- Vol. 33 (4) , 045015
Atomistic simulation of HF diffusion on ammonium fluorosilicate surface using neural network potential
Plasma etching is an essential technology in semiconductor fabrication, enabling precise nanoscale patterning. For high-aspect-ratio channel hole etching in 3D NAND flash memory, cryogenic etching using hydrogen fluoride (HF) gas shows great potential. This process often leads to the deposition of ammonium hexafluorosilicate (AFS) on sidewalls, which critically impacts surface diffusion. Understanding such phenomena requires accurate atomistic modeling, and while density functional theory (DFT) provides reliable and accurate results, its significant computational cost makes it challenging to apply to large-scale or dynamic simulations. As a promising alternative, neural network potentials (NNPs) provide DFT-level accuracy at a fraction of the computational cost. In this study, we develop a fine-tuned NNP based on the pretrained SevenNet-0 model to simulate HF diffusion on AFS surfaces. Although SevenNet-0 is trained on a broad chemical space and exhibits great generalization capabilities, it requires further refinement to accurately capture the complex energy landscape occurring during cryogenic etching, particularly for configurations far from equilibrium. To address this, we fine-tune the SevenNet-0 model using a minimal DFT dataset. The resulting fine-tuned NNP demonstrates superior accuracy and stability in molecular dynamics simulations compared to both the NNP trained from scratch and the SevenNet-0 model. Our analysis reveals that the additive gas IF5 enhances HF diffusivity by reducing chain formation and lowering the diffusion barrier. This work underscores the potential of fine-tuned NNPs for simulating complex etching processes, offering valuable insights for advancing semiconductor manufacturing.
- 29 May 2025
- Published by IOP Publishing in Modelling and Simulation in Materials Science and Engineering
- Vol. 33 (4) , 045015
Atomistic simulation of HF diffusion on ammonium fluorosilicate surface using neural network potential
Plasma etching is an essential technology in semiconductor fabrication, enabling precise nanoscale patterning. For high-aspect-ratio channel hole etching in 3D NAND flash memory, cryogenic etching using hydrogen fluoride (HF) gas shows great potential. This process often leads to the deposition of ammonium hexafluorosilicate (AFS) on sidewalls, which critically impacts surface diffusion. Understanding such phenomena requires accurate atomistic modeling, and while density functional theory (DFT) provides reliable and accurate results, its significant computational cost makes it challenging to apply to large-scale or dynamic simulations. As a promising alternative, neural network potentials (NNPs) provide DFT-level accuracy at a fraction of the computational cost. In this study, we develop a fine-tuned NNP based on the pretrained SevenNet-0 model to simulate HF diffusion on AFS surfaces. Although SevenNet-0 is trained on a broad chemical space and exhibits great generalization capabilities, it requires further refinement to accurately capture the complex energy landscape occurring during cryogenic etching, particularly for configurations far from equilibrium. To address this, we fine-tune the SevenNet-0 model using a minimal DFT dataset. The resulting fine-tuned NNP demonstrates superior accuracy and stability in molecular dynamics simulations compared to both the NNP trained from scratch and the SevenNet-0 model. Our analysis reveals that the additive gas IF5 enhances HF diffusivity by reducing chain formation and lowering the diffusion barrier. This work underscores the potential of fine-tuned NNPs for simulating complex etching processes, offering valuable insights for advancing semiconductor manufacturing.
- Top Cited PapersOpen Access
- 7 May 2025
- Published by Royal Society of Chemistry (RSC) in Digital Discovery
- Vol. 4 (6) , 1544-1559
Application of pretrained universal machine-learning interatomic potential for physicochemical simulation of liquid electrolytes in Li-ion batteries
Achieving higher operational voltages, faster charging, and broader temperature ranges for Li-ion batteries necessitates advancements in electrolyte engineering. However, the complexity of optimizing combinations of solvents, salts, and additives has limited the effectiveness of both experimental and computational screening methods for liquid electrolytes. Recently, pretrained universal machine-learning interatomic potentials (MLIPs) have emerged as promising tools for computational exploration of complex chemical spaces with high accuracy and efficiency. In this study, we evaluated the performance of the state-of-the-art equivariant pretrained MLIP, SevenNet-0, in predicting key properties of liquid electrolytes, including solvation behavior, density, and ion transport. To assess its suitability for extensive material screening, we considered a dataset comprising 20 solvents. Although SevenNet-0 was predominantly trained on inorganic compounds, its predictions for the properties of liquid electrolytes showed good agreement with experimental and ab initio data. However, systematic errors were identified, particularly in the predicted density of liquid electrolytes. To address this limitation, we fine-tuned SevenNet-0, achieving improved accuracy at a significantly reduced computational cost compared to developing bespoke models. Analysis of the training set suggested that the model achieved its accuracy by generalizing across the chemical space rather than memorizing trained configurations. This work highlights the potential of SevenNet-0 as a powerful tool for future engineering of liquid electrolyte systems.
- 7 May 2025
- Published by Royal Society of Chemistry (RSC) in Digital Discovery
- Vol. 4 (6) , 1544-1559
Top Cited PapersOpen AccessApplication of pretrained universal machine-learning interatomic potential for physicochemical simulation of liquid electrolytes in Li-ion batteries
Achieving higher operational voltages, faster charging, and broader temperature ranges for Li-ion batteries necessitates advancements in electrolyte engineering. However, the complexity of optimizing combinations of solvents, salts, and additives has limited the effectiveness of both experimental and computational screening methods for liquid electrolytes. Recently, pretrained universal machine-learning interatomic potentials (MLIPs) have emerged as promising tools for computational exploration of complex chemical spaces with high accuracy and efficiency. In this study, we evaluated the performance of the state-of-the-art equivariant pretrained MLIP, SevenNet-0, in predicting key properties of liquid electrolytes, including solvation behavior, density, and ion transport. To assess its suitability for extensive material screening, we considered a dataset comprising 20 solvents. Although SevenNet-0 was predominantly trained on inorganic compounds, its predictions for the properties of liquid electrolytes showed good agreement with experimental and ab initio data. However, systematic errors were identified, particularly in the predicted density of liquid electrolytes. To address this limitation, we fine-tuned SevenNet-0, achieving improved accuracy at a significantly reduced computational cost compared to developing bespoke models. Analysis of the training set suggested that the model achieved its accuracy by generalizing across the chemical space rather than memorizing trained configurations. This work highlights the potential of SevenNet-0 as a powerful tool for future engineering of liquid electrolyte systems.
- 6 September 2024
- Published by American Chemical Society (ACS) in Journal of Chemical Theory and Computation
- Vol. 20 (18) , 8024-8034
Modified Activation-Relaxation Technique (ARTn) Method Tuned for Efficient Identification of Transition States in Surface Reactions
Exploring potential energy surfaces (PES) is essential for unraveling the underlying mechanisms of chemical reactions and material properties. While the activation-relaxation technique (ARTn) is a state-of-the-art method for identifying saddle points on PES, it often faces challenges in complex energy landscapes, especially on surfaces. In this study, we introduce iso-ARTn, an enhanced ARTn method that incorporates constraints on an orthogonal hyperplane and employs an adaptive active volume. By leveraging a neural network potential (NNP) to conduct an exhaustive saddle point search on the Pt(111) surface with 0.3 monolayers of surface oxygen coverage, iso-ARTn achieves a success rate that is 8.2% higher than the original ARTn, with 40% fewer force calls. Moreover, this method effectively finds various saddle points without compromising the success rate. Combined with kinetic Monte Carlo simulations for event table construction, iso-ARTn with NNP demonstrates the capability to reveal structures consistent with experimental observations. This work signifies a substantial advancement in the investigation of PES, enhancing both the efficiency and breadth of saddle point searches.
- 6 September 2024
- Published by American Chemical Society (ACS) in Journal of Chemical Theory and Computation
- Vol. 20 (18) , 8024-8034
Modified Activation-Relaxation Technique (ARTn) Method Tuned for Efficient Identification of Transition States in Surface Reactions
Exploring potential energy surfaces (PES) is essential for unraveling the underlying mechanisms of chemical reactions and material properties. While the activation-relaxation technique (ARTn) is a state-of-the-art method for identifying saddle points on PES, it often faces challenges in complex energy landscapes, especially on surfaces. In this study, we introduce iso-ARTn, an enhanced ARTn method that incorporates constraints on an orthogonal hyperplane and employs an adaptive active volume. By leveraging a neural network potential (NNP) to conduct an exhaustive saddle point search on the Pt(111) surface with 0.3 monolayers of surface oxygen coverage, iso-ARTn achieves a success rate that is 8.2% higher than the original ARTn, with 40% fewer force calls. Moreover, this method effectively finds various saddle points without compromising the success rate. Combined with kinetic Monte Carlo simulations for event table construction, iso-ARTn with NNP demonstrates the capability to reveal structures consistent with experimental observations. This work signifies a substantial advancement in the investigation of PES, enhancing both the efficiency and breadth of saddle point searches.
- 28 August 2024
- Published by American Chemical Society (ACS) in ACS Applied Materials & Interfaces
- Vol. 16 (36) , 48457-48469
Atomistic Simulation of HF Etching Process of Amorphous Si3N4 Using Machine Learning Potential
- 28 August 2024
- Published by American Chemical Society (ACS) in ACS Applied Materials & Interfaces
- Vol. 16 (36) , 48457-48469
Atomistic Simulation of HF Etching Process of Amorphous Si3N4 Using Machine Learning Potential