Korean Research Team Develops Intuitive Model to Predict Complex Material Structures Without Supercomputers

The Ministry of Science and ICT (Minister Yu Sang-wook) announced that a research team led by Professor Wooyoung Shim at Yonsei University has developed a new model capable of predicting the structures of complex materials without the need for intensive computer simulations. This “phenomenological model” utilizes intuitive atomic properties—such as electronegativity, ionic radius, and charge—to forecast material structures with surprising accuracy.

A Simpler Alternative to Complex Simulations

Until now, predicting the crystal structure of newly discovered compounds required high-precision simulations and massive computational resources. While accurate, these methods were time-consuming and accessible only to experts, limiting their practical use.

Earlier models based on simple atomic traits—such as Pauling’s rules—could only handle binary compounds (those made from two elements) and were inadequate for more complex materials. However, today’s materials innovation often involves ternary or more complex compounds (three or more elements), demanding a new, simpler predictive tool.

From Prediction to Real-World Synthesis

In response to this need, the research team developed a new phenomenological model specifically tailored for ternary A–III–V compounds composed of alkali metals, group III, and group V elements. The model converts atomic properties into quantitative indicators of chemical bonding tendencies, enabling the prediction of whether a material will form a layered structure.

Using this approach, the team accurately classified 35 known layered materials and identified 9 previously unknown candidates. Among these, two—K₂In₂P₃ and Na₂In₂As₃—were successfully synthesized in the lab. Structural analysis using X-ray diffraction (XRD) and electron microscopy confirmed that the actual structures matched the predictions. This marks a significant advance in materials discovery, offering a low-cost and efficient alternative to conventional simulation-based methods.

Toward Next-Generation Electronics

The model’s potential goes beyond structure prediction. The team used Na₂In₂As₃ to fabricate a 2D indium arsenide (InAs) layer, which was successfully applied in a next-generation memory device (memtransistor). The fact that the model not only guided material synthesis but also enabled real device fabrication highlights its practical importance.

Dr. Jongbeom Won and Dr. Taeyoung Kim, who led much of the work, emphasized that "like Pauling’s rules, our model is simple and intuitive—something any researcher can use." Professor Shim added, “We’re the first in the world to propose both a new material and a predictive model for it. Korea has world-class strength in materials science, and we should lead the discovery of future materials ourselves.”

Globally Recognized Impact

This groundbreaking research was supported by the Ministry of Science and ICT and the National Research Foundation of Korea. Results were published in two leading international journals—Advanced Materials (May) and Nature Communications (July)—underscoring the model’s credibility and global relevance.

Phenomenological prediction and experimental validation of layered cation-eutaxy structures

in ternary A–III–V compounds

[Reference] Won J. et al., (2025) “Mapping cation-eutaxy ternary with a phenomenological model” Nature Communications

Kim T. et al., (2025) “Memristive InAs-based semiconductors with anisotropic ion transport” Advanced Materials

[Main Author] Jongbum Won (Yonsei University), Taeyoung Kim (Yonsei university), Wooyoung Shim (Yonsei University)

* Contact email : Professor Wooyoung Shim (wshim@yonsei.ac.kr)

KAIST (President Lee Kwang-hyung) announced that Professor Jeung Ku Kang’s team from the Department of Materials Science and Engineering designed a machine learning algorithm to predict the performance of multiprincipal element alloy catalysts and developed water-splitting catalysts with performance surpassing existing precious metal-based catalysts.

Currently, electrocatalysts for water electrolysis mainly contain large amounts of precious metals such as platinum and iridium, facing problems in terms of economic feasibility and sustainability from the catalyst production stage. Additionally, the existing experimental catalyst development method based on trial-and-error has the problem of requiring astronomical amounts of time and cost to explore the vast compositional space of multi-element alloys.

In response, Professor Kang’s research team developed a machine learning-based catalyst design method that enables rapid exploration of compositional space and discovery of optimal compositions. Particularly noteworthy is that they applied differential uncertainty to compositional regions based on the insight that the amount of information obtained through sampling in high-entropy compositional regions is greater than in low-entropy regions. Through this, the research team explained that they applied thermodynamic knowledge from the design stage of the machine learning model, enabling the model to perform chemical inductive reasoning. The machine learning framework developed by the research team applied Bayesian optimization techniques combining Gaussian Process Regression models and catalyst entropy. The optimal alloy with composition Fe₁₂Co₂₈Ni₃₃Mo₁₇Pd₅Pt₅ showed low overpotentials of 24mV for hydrogen evolution reaction and 204mV for oxygen evolution reaction, which significantly surpasses the performance of platinum/iridium oxide catalysts composed of combinations of precious metals optimal for each reaction. Additionally, significant progress was made in terms of mass activity, which represents the utilization efficiency of precious metals.

The core of this technology lies in efficient exploration of compositional space through design of experiments and design of acquisition functions including thermodynamic information. The model simultaneously provides predicted values (mean values) and uncertainty (standard deviation), and they built an AI model that can independently verify the contribution of each element to overpotential in multi-element compositional space. They also used compositional entropy to find catalysts with optimal compositions without falling into local minima.

The research results of Professor Kang’s team (KAIST), jointly participated by Professor William Goddard’s team from California Institute of Technology (Caltech), were published in the prestigious international academic journal ‘Proceedings of the National Academy of Sciences (PNAS)’. (Paper title: AI-driven design of multiprincipal element alloys for optimal water splitting).

Machine learning framework for the composition optimization of multiprincipal element alloys.

[Reference] AI-driven design of multiprincipal element alloys for optimal water splitting, Proceedings of the National Academy of Sciences (PNAS).

[Main Author] Jeung Ku Kang (Professor, Department of Materials Science and Engineering, KAIST), William A. Goddard (Professor, Chemistry & Chemical Engineering department, Caltech), Jihoon Kim (Graduate student, Department of Materials Science and Engineering, KAIST)

Stereoisomerism, arising from the distinctive spatial arrangements of atoms despite identical molecular formulae, often displays different chemical reactivities. Herein, we demonstrate how geometric isomerism of multifunctional electrolyte additives affects aqueous zinc metal batteries. Inspired by natural bacteria, we compared trans-butenedioic acid (fumaric acid) and its cis-isomer (maleic acid), revealing different hydrogen bonding dynamics and solvation environments, as confirmed by femtosecond transient absorption spectroscopy and computational simulations. The trans-isomer promotes the formation of favorable interfacial structures and ion pathways, improving Zndeposition reversibility and cycling stability. As a result, Zn symmetric cells showed stable plating/strippingfor over 6150h at 1 mA cm-2 and 1 mAh cm-2 and 1500h at 5 mA cm-2 and 5 mAh cm-2. The Zn-predeposited Cu||MnVO full cells exhibited a capacity retention exceeding 70% after 1000 cycles at 2 A g-1, ultimately achieving over 270 cycles for initially anodeless Cu||zincated MnVO cells at a high current density of 30 mA cm-2. The application of isomerism concept on the design of new electrolyte materials and associated solvated and interphasial chemistries offers new pathway to the next generation batteries.

Concept and interaction chemistry by Stereoisomers

[Reference] Huang S. et al., (2025) “Stereoisomerism of Multi-Functional Electrolyte Additives for Initially Anodeless Aqueous Zinc Metal Batteries.” Nature Communications

[Main Author] Shengyang Huang, Hao Fu, Sae Byeok Jo, Ho Seok Park 

* Contact : phs0727@skku.edu

The formation of ice can serve as a remarkable and previously overlooked pathway for mineral genesis. The National Research Foundation of Korea announced that a research team led by Professor Giehyeon Lee at Yonsei University has uncovered that minerals with distinctive properties form within ice during the freezing of natural waters. The study was published in the Proceedings of the National Academy of Sciences (PNAS) in May.

For decades, chemical reactions in ice were largely neglected, primarily due to the perception that reactions proceed too slowly at low temperatures and are restricted in solid phases. However, growing evidence now suggests that certain chemical reactions can be accelerated in freezing solutions, driven by the freeze concentration effect. As water freezes, dissolved ions and compounds are excluded from the growing ice lattice and become concentrated in thin, unfrozen channels at ice grain boundaries, known as liquid-like layers. These layers act as confined microenvironments where reactions can proceed under conditions vastly different from those in the original liquid solution. 

Building on this insight, the researchers investigated how the freeze concentration effect might operate in nature, specifically in mineral formation. Minerals are fundamental to Earth’s biogeochemical cycles, playing essential roles in the transport, storage, and transformation of elements. To simulate natural conditions, the team froze solutions containing manganese and carbonate ions, key players in photosynthesis and carbon cycling, respectively. 

To their surprise, they found that rhodochrosite (MnCO3), a manganese carbonate mineral, formed in freezing solutions even under theoretically impossible conditions, which was achieved when the saturation level (the relative concentrations of dissolved manganese and carbonate ions compared to those required for rhodochrosite formation) was increased approximately 30,000 times at room temperature. Furthermore, the rhodochrosite crystals formed in ice appeared not as typical cubic or rhombohedral crystals but as spherical aggregates of numerous nanocrystallites, indicating a distinctly different crystallization pathway. 

“This result is due to the dramatically increased saturation level and the low temperature within the so-called liquid-like layers at ice grain boundaries,” said Professor Lee. “It demonstrates that ice can create a unique chemical environment for mineral formation, unlike any other environments observed in natural settings.” He added, “Our findings have broad implications, not only for Earth science and planetary research, but also for practical applications in areas such as freeze-assisted syntheses of functional materials, cryopreservation, and frozen food stability.”

Processes of mineral formation in aqueous (25 ℃) and freezing (-20 ℃) solutions

[Reference] Won Y. et al. (2025) "Freeze-induced crystallization: An overlooked pathway for mineral genesis in natural waters." Proceedings of the National Academy of Sciences. DOI: https://doi.org/10.1073/pnas.2421822122

[Main Author] Giehyeon Lee (Yonsei University), Younghoon Won (Yonsei University)

* Contact : ghlee@yonsei.ac.kr