Eunae Cho

Sang Ho Jeon

Bora Lee

Youngho Kang

Kye Yeop Kim

Kanghoon Yim

Joohee Lee

Yong Youn

Hochul Song

Gijae Kang

Kyuhyun Lee

Dongsun Yoo

Dongheon Lee

Miso Lee

Wonseok Jeong

Sungwoo Kang

Sangtae Kim

Kyeongpung Lee

Jisu Jung

Changho Hong

Suyeon Ju

Jeong Min Choi

Deokgi Hong

Tae-Eun Kim

Sohee Park

Doyeon Kim

Choongki Lee

Hyo-Shin Ahn

Won-joon Son

The lab focuses on computational materials science, data-driven materials discovery, and the application of machine learning techniques to materials research. Specific research areas include:Atomistic simulation of materials,Development of machine learning interatomic potentials (MLIPs),Data-driven analysis of materials properties,Simulation of cryogenic etching processes,Simulation of liquid electrolytes in Li-ion batteries

계산 재료 과학, 데이터 기반 재료 발견 및 재료 연구에 대한 기계 학습 기술 적용에 중점을 둡니다. 특정 연구 분야는 재료의 원자 시뮬레이션, 기계 학습 원자 간 전위(MLIP) 개발, 재료 특성의 데이터 기반 분석, 극저온 에칭 공정 시뮬레이션, Li-ion 배터리의 액체 전해질 시뮬레이션입니다.

This research elucidates the atomistic mechanisms governing the alkaline wet etching of GaN,a crucial process in the fabrication of nitride semiconductor nanostructures, employing largescale molecular dynamics (MD) simulations driven by a developed neural network potential(NNP). The developed Behler-Parrinello type NNP overcomes the computational limitationsof density functional theory in simulating long timescale, large-scale chemical dynamics,marking a breakthrough in the study of complex material interfaces. By applying temperatureaccelerated dynamics, the NNP-MD simulations reproduce the surface-specific morphologicalchanges of GaN nanorods that are consistent with experimental observations. The analysis alsoreproduces surface-specific etching behaviors of surface morphologies. Crucially, enhancedsampling simulations were used to quantify the free energy surfaces of the Ga dissolutionpathway under ambient conditions, explaining the etching anisotropy observed oncrystallographic surfaces by correlating macroscopic etching features with microscopicchemical reaction barriers. Mechanistically, this work offers a quantitative, atomisticexplanation for the observed anisotropic etching of GaN. Furthermore, the study identified the formation and stability of Ga-O-Ga bridges on the etched surfaces, suggesting they couldpotentially act as detrimental charge carrier traps that impact device performance. Ultimately,the findings establish atomic-level guidelines to suppress the formation of efficiency-degradingsurface defects, paving the way for the fabrication of high-performance GaN electronic andoptoelectronic devices. This work was published in npj Computational Materials 11 , 311 (2025).

Neural network-driven molecular insights into alkaline wet etching of GaN: toward atomistic precision in nanostructure fabrication

Amorphous GeSe is a promising material for ovonic threshold switching (OTS) selectors, but the microscopic origins of its defect states have remained unclear. In this study, large-scale molecular dynamics simulations accelerated by machine learning interatomic potentials (MLIPs) were employed to investigate defect motifs in amorphous GeSe. Benchmarking across Behler–Parrinello neural networks (BPNN), moment tensor potentials (MTP), and SevenNet models revealed that capturing at least four body interactions and sufficient effective receptive field is essential to avoid spurious defects. Using an optimized SevenNet model, two dominant defect motifs were identified in 20 independent 960-atom cells: aligned Ge chains giving rise to conduction band defect states, and overcoordinated Ge chains producing valence band defect states. Importantly, correlations were established between electronic defect levels and structural features such as bond-angle alignment in chains and Peierls distortion around overcoordinated Ge atoms. These results provide a structural–electronic framework to interpret experimental observations from DLTS and IPES, offering new insights into defect-driven OTS behavior in amorphous chalcogenides. This study was published in ACS Applied Materials & Interfaces 17 , 57260 (2025).

Unveiling defect motifs in amorphous GeSe using machine learning interatomic potentials

Cryogenic etching has emerged as a promising solution for creating the intricate, high-aspect-ratio structures required by modern semiconductors. This low-temperature process is crucial for manufacturing devices like 3D NAND flash memory. However, understanding the process at an atomic level is challenging. Traditional simulation methods, such as density functional theory (DFT), are often too slow and computationally expensive to model the complex, dynamic systems involved, limiting our ability to fully optimize the process.
To overcome this limitation, we developed an innovative approach using a fine-tuned neural network potential (NNP). Our work focuses on a critical phenomenon in cryogenic etching: the diffusion of hydrogen fluoride (HF) on the surface of ammonium hexafluorosilicate (AFS), a byproduct that deposits on sidewalls during the process. We started with a pre-trained universal model, SevenNet-0, and fine-tuned it using a small, specialized dataset from DFT calculations. This significantly improved the model's accuracy, allowing us to achieve DFT-level performance with a fraction of the computational cost.
Using our refined model, we conducted simulations and discovered that an additive gas, such as iodine pentafluoride (IF 5 ), drastically enhances HF diffusivity. Our analysis showed that IF 5 works by inhibiting the formation of HF molecular chains and lowering the diffusion barrier on the AFS surface. These findings provide valuable insights into the role of additive gases in optimizing cryogenic etching. Our research demonstrates that fine-tuned NNPs are a powerful and efficient tool for modeling complex semiconductor processes, paving the way for further advancements in manufacturing technology. This study was published in Modelling and Simulation in Materials Science and Engineering 2025, 33 , 045015.

Accelerating atomistic simulation for cryogenic etching with fine-tuned neural network potentials

Improving Li-ion batteries to support higher voltages, faster charging, and wider temperature ranges requires better electrolyte materials, but exploring the vast combinations of solvents, salts, and additives is challenging using traditional experimental and computational methods. Here, we evaluated a state-of-the-art pretrained universal machine-learning interatomic potential (MLIP), SevenNet-0, for its ability to predict key properties of liquid electrolytes, including solvation structure, density, and ion transport. Although primarily trained on inorganic systems, SevenNet-0 performed well on organic electrolytes, showing good agreement with experimental and ab initio data, though we identified systematic errors, particularly in density predictions, and addressed them by fine-tuning the model. This approach improved accuracy significantly while keeping computational costs low, and analysis of the training set indicated that the model achieved its performance by generalizing across chemical space rather than memorizing specific configurations. This study highlights the potential of MLIPs like SevenNet-0 to accelerate the discovery and design of advanced electrolyte materials, and this work was published in Digital Discovery 2025, 4 , 1544.

Application of pretrained universal machine-learning interatomic potential for physicochemical simulation of liquid electrolytes in Li-ion batteries

Machine learning potentials (MLPs) offer an efficient way to replicate potential energy surfaces(PES) derived from first-principles calculations, significantly lowering computational costs.However, generating high-quality training datasets remains expensive, particularly forapplications requiring high-level theoretical accuracy. Here, we introduce a cost-awaremultifidelity training strategy for MLPs that allows simultaneous utilization of low- and highfidelity datasets. Using an equivariant graph neural network as the underlying MLP architecture,we validate this method across organic molecules, inorganic solids, and pretrained universalMLPs. Our findings reveal that abundant, correlated data from low-fidelity sources caneffectively complement sparse high-fidelity datasets, bridging gaps in configurational andcompositional coverage. By strategically subsampling low-fidelity configurations, highfidelity PES can be reproduced accurately with minimal computational effort. Furthermore,this approach addresses limitations associated with conventional methods for high-fidelityMLPs, such as Δ-learning and transfer learning. This study was published in Journal of theAmerican Chemical Society 2025, 147 , 1042.

Multifidelity machine learning potential for reaching high-level ab initio calculations

Are diffusion models ready for materials discovery in unexplored chemical space?

A computational study for screening high-selectivity inhibitors in area-selective atomic layer deposition on amorphous surfaces

Optimizing cross-domain transfer for universal machine learning interatomic potentials

Discovery of oxide Li-conducting electrolytes in uncharted chemical space via topology-constrained crystal structure prediction

Reactive machine learning interatomic potentials for chemistry and materials science

Latent Indium alloy defect formation inherently limiting InGaN light-emitting diode efficiency

An efficient forgetting-aware fine-tuning framework for pretrained universal machine-learning interatomic potentials

Atomistic insights into hydrogen migration in IGZO from machine-learning interatomic potential: linking atomic diffusion to device performance

Etching-to-deposition transition in SiO2/Si3N4 using CHxFy ion-based plasma etching: an atomistic study with neural network potentials

Atomistic insights into Cu/amorphous-TaxN interfacial adhesion via machine learning interatomic potentials: effects of stoichiometry and interface construction

BaGe2P2 and BaGe2As2 pnictides as promising ferroelecric semiconductors for thin film solar cell

Electronic structures of crystalline and amorphous GeSe and GeSbTe compounds using machine learning empirical pseudopotentials

Atomistic simulation of HF diffusion on ammonium fluorosilicate surface using neural network potential

Flash TP: fused, sparsity-aware tensor product for machine learning interatomic potentials

Nonmelting disordering facilitated by electron delocalization

Roadmap for the development of machine learning-based interatomic potentials

Data-efficient multifidelity training for high-fidelity machine learning interatomic potentials

Chemisorption manipulation by adjusting the carrier concentration of the adsorbent and its application to adsorbate identification

Modified activation-relaxation technique (ARTn) method tuned for efficient identification of transition states in surface reactions

Atomistic simulation of HF etching process of amorphous Si3N4 using machine learning potential

Disorder-dependent Li diffusion in Li6PS5Cl investigated by machine-learning potential

Scalable parallel algorithm for graph neural network interatomic potentials in molecular dynamics simulations

Bonding structure and dry etching characteristics in amorphous B-C-N films for hardmask applications

Predicting melting temperature of inorganic crystals via crystal graph neural network enhanced by transfer learning

Electrochemical degradation of Pt3Co nanoparticles investigated by off-lattice kinetic Monte Carlo simulations with machine-learned potentials

Atomic structure of conducting nanofilaments in TiO 2 resistive switching memory

Development of new interatomic potentials appropriate for crystalline and liquid iron

Development of an interatomic potential for phosphorus impurities in α-iron

Magnetic ordering at the edges of graphitic fragments: Magnetic tail interactions between the edge-localized states

Electrical manipulation of nanofilaments in transition-metal oxides for resistance-based memory

Best practices in machine learning for chemistry

First-principles study on doping and phase stability of

Organolead Halide Perovskites for Low Operating Voltage Multilevel Resistive Switching.

Hydrated Manganese(II) Phosphate (Mn 3 (PO 4 ) 2 ·3H 2 O) as a Water Oxidation Catalyst

Al‐Doped TiO 2 Films with Ultralow Leakage Currents for Next Generation DRAM Capacitors

Oxygen Vacancy Clustering and Electron Localization in Oxygen-Deficient : <?format ?> Study

Novel high-κ dielectrics for next-generation electronic devices screened by automated ab initio calculations

Capacitors with an equivalent oxide thickness of< 0.5 nm for nanoscale electronic semiconductor memory

First-principles study of point defects in rutile

A detailed understanding of the electronic bipolar resistance switching behavior in Pt/TiO2/Pt structure

Wafer-scale transferable molybdenum disulfide thin-film catalysts for photoelectrochemical hydrogen production

Anion control as a strategy to achieve high-mobility and high-stability oxide thin-film transistors

SIMPLE-NN: An efficient package for training and executing neural-network interatomic potentials

Materials Data & Informatics Laboratory

교수님 소개

한승우

연구실 소개

연구실 키워드

알럼나이가 간 회사

연구실 지원 방법

대우 조건

연도별 논문 인용수

저널 & 학회 논문 출판 수

연구실 논문 목록

Latent Indium alloy defect formation inherently limiting InGaN light-emitting diode efficiency

An efficient forgetting-aware fine-tuning framework for pretrained universal machine-learning interatomic potentials

Atomistic insights into hydrogen migration in IGZO from machine-learning interatomic potential: linking atomic diffusion to device performance

연구실 보유 특허

대학별 입시끈

우리 연구실 모집

교수님 소개

한승우

연구실 키워드

알럼나이가 간 회사

연구실 지원 방법

대우 조건

연도별 논문 인용수

저널 & 학회 논문 출판 수

연구실 논문 목록

Latent Indium alloy defect formation inherently limiting InGaN light-emitting diode efficiency

An efficient forgetting-aware fine-tuning framework for pretrained universal machine-learning interatomic potentials

Atomistic insights into hydrogen migration in IGZO from machine-learning interatomic potential: linking atomic diffusion to device performance

연구실 보유 특허

최근 3년간 연구실적 높은 연구실

대학별 입시끈

우리 연구실 모집