Kaist

KAIST Characterizes Barium Titanate’s Optical Prop..

< (From left) Ph.D. candidate Jiwon Kang from KAIST, Dr. Nina Hong from J.A. Woollam Co., Professor Harry A. Atwater from Caltech, and Professor Min Seok Jang from KAIST > Faster, more efficient optical communications require precise control of light and minimal optical loss. A KAIST-led research team has characterized how light propagates through and is absorbed by barium titanate (BTO), a promising material for next-generation optical communication devices. By clarifying optical properties that had been difficult to determine because of the material’s complex internal structure, the study provides a foundation for designing these devices. < Intrinsic optical properties of barium titanate revealed by Mueller matrix ellipsometry. > KAIST (President Choongsik Bae) announced on November 8 that a research team led by Professor Min Seok Jang of its School of Electrical Engineering had precisely characterized BTO’s refractive index and absorption properties across wavelengths ranging from the ultraviolet to the far-infrared. The study was conducted in collaboration with Professor Harry A. Atwater’s team at the California Institute of Technology (Caltech), the Korea Research Institute of Standards and Science (KRISS), and U.S.-based J.A. Woollam Co. BTO’s refractive index changes when an electric field is applied. Its strong electro-optic response, known as the Pockels effect, makes it a promising material for optical modulators, which encode information from electrical signals onto light. < Figure 2. Structural characteristics of single-crystal barium titanate and a schematic of the optical constant measurements. > Designing these devices requires accurate values for the refractive index and extinction coefficient, which describe how light propagates through a material and how strongly it is absorbed, respectively. In a waveguide—a structure that guides light—these values help designers predict how light will travel and how much of the signal will be lost to absorption. However, BTO can contain small regions, known as domains, with different crystallographic orientations. These regions can respond differently to light, making it difficult to determine the material’s intrinsic optical properties from their combined response. Reported values have varied widely for decades. The research team used Mueller-matrix spectroscopic ellipsometry, a technique that measures changes in the polarization of light reflected from or transmitted through a material. By analyzing two types of single-crystal samples with relatively uniform domain orientations, the researchers reduced the effects of mixed responses from differently oriented domains and determined the material’s intrinsic optical properties. The analysis covered wavelengths from 190 nanometers (nm) to 40 micrometers (μm), spanning the ultraviolet to the far-infrared. Because light interacts with a material differently depending on its wavelength, the researchers divided this range into three spectral regions and used an analytical approach suited to each region. This allowed them to determine how the refractive index and absorption properties vary with wavelength and crystal orientation. < Figure 3. Intrinsic complex refractive index of tetragonal BTO determined over a broad spectral range. > The researchers also quantified very weak absorption in the wavelength range where BTO is nearly transparent. By combining ellipsometry with separate transmittance measurements, they determined extinction coefficients as low as approximately 10⁻⁵ in the 2–10 μm wavelength range. These data can help predict absorption losses in components that transmit or confine light. The team also presented a method for calculating the overall optical response of a multidomain crystal from the orientations and relative proportions of its domains. These optical data provide a basis for designing BTO-based photonic devices. Before fabrication, designers can use the data to more accurately predict how light will propagate and how much will be absorbed in devices of different sizes and shapes. This could reduce the time and cost associated with repeated fabrication and redesign. Potential applications include optical modulators for communications, devices that generate light at new wavelengths, and circuits that process quantum information using light. “This study reveals the optical response of barium titanate, which had long remained unclear, and provides a reliable reference for how the material interacts with light,” said Professor Min Seok Jang. He added that the findings are expected to contribute to the development of optical communication devices that reduce light loss and enable faster data transmission, as well as quantum photonic devices that require precise control of light. Jiwon Kang, a Ph.D. candidate in KAIST’s School of Electrical Engineering, and Dr. Nina Hong of J.A. Woollam Co. are co-first authors of the paper. Professors Min Seok Jang of KAIST and Harry A. Atwater of Caltech are the corresponding authors. The findings were published online on October 20, 2026, in the optics journal ACS Photonics. Paper title: Broadband Optical Characterization of Tetragonal Barium Titanate by Mueller-Matrix Spectroscopic Ellipsometry DOI: 10.1021/acsphotonics.6c01514 This research was supported by the Culture, Sports and Tourism R&D Program through the Korea Creative Content Agency (KOCCA), with funding from the Ministry of Culture, Sports and Tourism; the Basic Science Research Program through the National Research Foundation of Korea (NRF), with funding from the Ministry of Education; the Ministry of Science and ICT; and the U.S. Air Force Office of Scientific Research (AFOSR).

Devilray Exposes Blind Spots in FBS Detection​

DAEJEON, South Korea — Governments use them to track phones. Criminals drive them through city streets and send texts that look like they came from your bank. For 20 years, researchers have been building tools to catch these fake cell towers. They have tested those tools mostly against fake towers they built themselves. That loop has left gaps. Researchers at KAIST built a platform called Devilray that reproduces 2,592 distinct fake-base-station behaviors. They ran it against seven existing detection systems, and every one of them missed at least 192. The finding points to a larger problem with how the field tests itself: detectors have been evaluated against the wrong adversary. An echo chamber Fake base stations go by several names: IMSI catchers, cell-site simulators, StingRays. They pose as legitimate cell towers and get nearby phones to connect. Once a phone is connected, the fake tower can make it give up the identifier that ties it to its owner or reveal where it is. It can also cut the phone off, push it down to an older and weaker network, or send it messages that never passed through a carrier. Police and intelligence agencies have used these devices for surveillance for years. Criminals now use them for illegal surveillance and for SMS-blaster scams, in which a device carried in a car sends phishing texts to every phone nearby. The commercial machines are expensive, and their sale and operation are tightly restricted. Universities generally cannot buy one. Carriers and investigative agencies are the ones responsible for finding fake towers in the field, but even they cannot get every model, let alone every new variant. So researchers do the reasonable thing. They build a fake base station, study the traces it leaves, design a detector that recognizes those traces and report that it works. The next paper improves on the last one, and detection rates climb. But the attacker in those experiments is one the defender already knows. The attack used to design a detector and the attack used to test it rest on the same assumptions. A detector that scores well may simply be recognizing its own reflection. What the field has lacked, the researchers argue, is adversarial thinking. The question should not be whether a detector catches a known attack. It should be what an attacker would do instead. "Detection has been built around the attacks we already know," said Yongdae Kim, the KAIST professor of electrical engineering who led the work. "Real attackers don't stay there. Until now, the playing field has tilted toward them." A standard full of open doors Attackers have that much room because of how LTE works. When an LTE phone first connects, it does not check whether the base station is genuine. It acts on some messages before any authentication takes place, and it can be pushed down to older generations of network. These are not implementation bugs that a patch can fix. They are written into the standard, so an attacker can choose from many messages and stay fully standard-compliant. Take the phone's permanent identifier, the main prize for a surveillance device. A fake tower can simply ask for it. That request is the move that Rayhunter, the Electronic Frontier Foundation's open-source detector, watches for. The tower can also send a rejection message that makes the phone reconnect and hand over the identifier without being asked. The standard allows both. A detector watching one door does not see the other. "LTE is fundamentally broken," Professor Kim said. "It is full of standard-compliant messages that can be used for bad purposes." Nobody is going to fix that. LTE's security design was settled long ago, and the standards community has moved on to 5G. Yet LTE remains the backbone of mobile service in most of the world. Many 5G networks still lean on it, and 5G phones drop back to LTE whenever 5G is unavailable, or whenever an attacker forces them to. These weaknesses will stay in the field as long as LTE does. As a result, defending against fake base stations looks less like fixing a software bug and more like an ongoing contest: a new attack appears, a detector learns to catch it, the attacker changes. That contest only works if defenders can see the moves attackers are able to make. Thinking like the attacker Devilray reverses the usual order. It maps the attacker first and looks at the detectors only afterward. The team began with a commercial fake base station, a unit sold to law enforcement and intelligence agencies. It is the first such device to be analyzed in an academic study. The device could do several things that academic prototypes rarely model: • copy the identity of a legitimate tower, • hop between frequencies, • run two networks at once, • dial its power down to cover only a few meters. The researchers then looked past that single device. They went through prior research and the 3GPP standards that govern cellular networks for other routes to the same ends. They broke a fake-base-station operation into four stages: surveying the local network, launching a fake cell, luring phones to connect, and carrying out the attack. Across those stages they identified seven dimensions an attacker can vary. Twenty-one dependency rules remove combinations that could not occur together in a real exchange. What remained were 2,592 configurations. Every configuration was then run against real smartphones to confirm that the full procedure completed. That check matters in both directions. An attack space derived only from the specifications could include thousands of combinations no phone would actually accept. A baseline copied from a single commercial device would inherit that device's limits. Devilray is anchored in the real device and extends outward into behaviors the standard permits. The researchers stress that 2,592 is not a count of attacks seen in the wild, and not a list of 2,592 ways to beat every detector. It is something the field has not had before: a realistic adversary, grounded in a real device and the standard itself, against which detectors can finally be tested. Seven detectors, seven sets of blind spots The seven detectors the team tested are publicly available and work on different principles: signaling messages, base station information, radio signals, statistical patterns and machine learning. Every one had blind spots, and most traced back to an assumption built into its design. Some detectors key on one telltale message, so they missed attacks that reached the same goal by another route. Others flag the unusually strong signal of a cheap fake tower; a fake tower that turns its power down, as the commercial unit can, stays under their thresholds. One statistical detector watches how often phones in an area are asked to reveal their identifiers. A fake tower that collects identifiers patiently, from a fraction of the phones nearby instead of all of them, produced a pattern that looked like an ordinary bus terminal. The harder question The deepest blind spot is the one where the attack starts to look like a normal network. The researchers tested this directly. They retrained a machine-learning detector on the Devilray attacks it had missed. On the radio link layer (RRC) model, its detection rate rose from 51 percent to 96 percent, while its false-alarm rate on ordinary carrier traffic nearly quadrupled, from 1.4 percent to 5.3 percent. On the core network layer (NAS) model, the detection rate rose from 27 percent to 63 percent, and its false-alarm rate increased from 4.7 percent to 8.8 percent. When trained on the hardest case, that false-alarm rate reached 15 percent, particularly for the NAS layer model. A detector that catches everything but constantly flags ordinary networks is of little use, and neither is a quiet one that misses carefully chosen attacks. The problem is not simply to maximize detection. It is to find the boundary between malicious and legitimate behavior when the standard lets an attacker move right up against it. "The real question is whether we can reduce misses without raising false alarms," Professor Kim said. The researchers argue that future detectors should report both numbers together. Claiming better coverage without testing against ordinary traffic gives an incomplete picture of whether a detector can actually be deployed. An attacker for defenders A platform that can imitate thousands of attack behaviors could also be misused, and the researchers acknowledge the risk. They share Devilray only with vetted researchers and organizations, under controlled access. What defenders have lacked is the same range of behavior to test against. With Devilray, carriers, investigators and detector developers can do that without having to buy every device on the market. The name is a jab at the StingRay, the commercial device whose name became shorthand for the whole category. A devil ray is a much bigger relative. The work is already reaching defenders. The team gave EFF the attack cases Rayhunter misses, and the foundation is now working on ways to detect them. The GSMA, the global association of mobile operators, invited the researchers to present Devilray to its Fraud and Security Group. There, standards and industry representatives discussed where fake-base-station detection should go next. The study was led by KAIST in collaboration with South Korea's National Security Research Institute, Kyung Hee University, the University of Florida and Cape, a U.S. mobile network company. It will be presented in December at ACSAC 2026 in Los Angeles. Professor Kim's group is now building a detection system of its own. No single vantage point may see enough, so it combines what phones observe, what carriers can see inside their networks and what independent radio sniffers capture. The aim is to detect fake base stations and also to locate them. "Devilray is not meant to defeat existing detectors," Professor Kim said. "It gives defenders the same choices attackers already have, so that attack and defense can finally advance together." Paper: "Devilray: A Reconfigurable Adversarial Baseline for Revealing Blind Spots in Fake Base Station Detection," IEEE ACSAC 2026, Los Angeles, Dec. 7–11, 2026. arXiv: https://arxiv.org/abs/2605.19232 · Project page and demo videos: https://sites.google.com/view/devilray Access: Devilray is available to vetted researchers and organizations under controlled access. Requests: https://sites.google.com/view/devilray Authors: Led by KAIST — Taekkyung Oh and Duckwoo Kim (co-first authors), Beomseok Oh and Yongdae Kim (KAIST); Hansung Bae and Byeongdo Hong (National Security Research Institute); CheolJun Park (Kyung Hee University); Tyler Tucker, Nathaniel Bennett and Patrick Traynor (University of Florida); Sangwook Bae (Cape). Funding: Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government(MSIT) (No. RS-2024-00441762, Global Advanced Cybersecurity Human Resources Development). Contact: Professor Yongdae Kim, School of Electrical Engineering, KAIST · yongdaek＠kaist.ac.kr

KAIST Makes Virtual Roughness Visible to Improve V..

< The awards ceremony on October 7. From left: Professor Dooyoung Kim, Ph.D. candidate Minju Baeck, and Professor Woontack Woo of KAIST, and Professor Bruce Thomas (award committee member) > What if you could see how rough an object feels, even without touching it? A KAIST research team has developed a technique that makes the roughness of virtual objects visible through changes in the outline of a virtual hand. By allowing users without haptic devices to perceive roughness information visually, the technique enables more effective collaboration in virtual reality (VR) between people using different equipment. KAIST (President Choongsik Bae) announced on October 29 that researchers from the Ubiquitous Virtual Reality Lab (UVR Lab), led by Professor Woontack Woo of the Graduate School of Metaverse, received the Best Paper Award at the IEEE International Symposium on Mixed and Augmented Reality (ISMAR 2026), a leading international conference on extended reality (XR). This marks the team’s third consecutive year winning the Best Paper Award, following wins in 2024 and 2025. In collaborative VR environments, users may have access to different types of sensory information depending on their equipment. For example, one person may feel an object’s roughness through a haptic device that delivers vibrations to the fingertips, while a partner without such a device must rely on visual information alone. This imbalance can make it difficult for users to compare, select, or sort virtual objects together. < Figure 1. Sensory asymmetry between collaborators in virtual reality > To address this challenge, the research team developed a technique that encodes an object's roughness as visual cues using the outline of a virtual hand.. By modifying the outline's shape and movement, the technique represents three levels of roughness: low, medium, and high. Rather than reproducing the tactile sensation on the skin, this approach enables users without a haptic device to distinguish between roughness levels visually. In two preliminary experiments, the team confirmed that users could distinguish between the visual cues and that the cues were consistent with roughness perceived through haptic feedback. They then conducted a collaborative VR experiment involving 48 participants organized into 24 pairs, with one participant using a haptic device and the other working without one.. Each pair collaborated to sort virtual objects according to their roughness. < Figure 2. Hand-outline roughness visualization and the VR collaboration experiment > When roughness cues was displayed on both participants’ virtual hands, the pairs completed the sorting task approximately 23％ faster than when no visual roughness cues were provided, while maintaining high sorting accuracy. Participants without haptic devices also reported greater confidence in their judgments and a stronger sense of contribution when the cues appeared on their own virtual hands. These findings suggest that visualizing tactile information can help users make shared decisions even when only one partner has access to haptic feedback. The team also found that collaboration was influenced not only by how roughness was represented, but also by whose virtual hand displayed the information. The findings could inform the design of VR education and training programs, as well as remote collaboration environments involving users with different equipment. The team previously received the Best Paper Award in 2024 for research on mixed reality (MR) remote collaboration. In 2025, two of its papers—one on spatial sharing and another on visual and tactile experiences—received Best Paper Awards. This year’s award continues a record of achievement in XR research spanning successive cohorts of students and researchers in the lab. < The Best Paper Award certificate and plaque > “Receiving Best Paper Awards for three consecutive years is a meaningful achievement made possible by government support for talent development and research in virtual convergence, KAIST’s research infrastructure, and the interdisciplinary efforts of our young researchers,” said Professor Woo. He added that “We will continue advancing virtual convergence research that integrates XR and AI to enable people to share knowledge and experiences beyond time and space, address real-world challenges, and expand human capabilities.” This research was supported by the Virtual Convergence Graduate School Support Program, funded by the Ministry of Science and ICT (MSIT) and administered by the Institute for Information & Communications Technology Planning & Evaluation (IITP). KAIST’s Graduate School of Metaverse is dedicated to cultivating interdisciplinary talent in extended reality (XR), including virtual and augmented reality, as well as artificial intelligence (AI) and spatial computing.

KAIST Cuts AI Fine-Tuning from 10 Hours to 3… Fast..

< From left: Ph.D. student Hyunjin Kim and Professor Jae-Gil Lee (KAIST School of Computing). Inset photos at top right, from left: Youngeun Nam, postdoctoral researcher (University of Toronto); Jaemin Han, master’s student (KAIST Graduate School of Data Science); and Wonhyeok Choi, master’s student (KAIST Graduate School of Data Science). > Can AI fine-tuning, that once took 10 hours, be completed in just 3 hours? KAIST researchers have developed a technology that selects the most useful training data and distributes it evenly across multiple GPUs. By reducing repetitive learning of similar data, the method shortened training time by up to 69.6％ while maintaining nearly the same AI performance. The technology is expected to help develop AI tailored to specific tasks, such as corporate customer service or assisting medical professionals with information retrieval, more quickly and at lower cost. KAIST (President Choongsik Bae) announced on November 2 that a research team led by Professor Jae-Gil Lee from the School of Computing developed “CluSTER,” a large language model fine-tuning technology that groups and selects data with similar effects on learning and distributes them evenly across multiple GPUs. To use a large language model in a specific field such as coding or medicine, it must be additionally trained on relevant data. This process is called “fine-tuning.” It is similar to teaching a broadly knowledgeable AI coding problems and solutions or medical questions and answers so that it can perform tasks in that particular field more effectively. This process requires substantial computation, so multiple GPUs divide the data and process it in parallel. A GPU is a semiconductor device capable of rapidly performing the calculations required for AI training. Just as several people can divide up a large set of problems and solve them together, multiple GPUs split the training data, process it, and combine the results to improve a single AI model. However, if the data are distributed randomly, multiple GPUs may repeatedly process data that have similar learning effects. This is like a student repeatedly solving problems that are very similar to ones they have already mastered, resulting in relatively little additional learning compared with the amount of computation required. The research team therefore focused not only on “how much data to train on,” but also on “which data to select and how to distribute them.” The goal was to reduce repetitive learning while ensuring that the AI did not miss diverse information it needed to learn. CluSTER first analyzes how each piece of data affects AI learning and groups together data with similar effects. The grouping is not based simply on whether the data cover the same topic or use similar words. Even problems that look different on the surface are classified into the same group if they teach the AI similar problem-solving abilities. For example, when teaching a student addition, a problem about combining two apples with three apples and a problem about combining two pencils with three pencils tell different stories, but the calculation being learned is the same. In this way, the research team identified and grouped data that had similar effects on learning even when they were expressed differently. The system then selects data from each group and distributes them so that multiple GPUs process a diverse range of learning information evenly. It is similar to reducing repetitive problems in a workbook while making sure that all important types of problems remain and are divided among several people to solve. The key feature is that the method does not merely reduce the amount of data, but also enables multiple GPUs to use the selected data efficiently. The research team fine-tuned large language models with approximately 7 billion and 13 billion parameters using three training datasets from the coding and medical fields. Parameters are internal numerical values adjusted as an AI model learns and are used to reflect acquired knowledge in its responses. The team evaluated the effectiveness of the technology through a total of eight performance benchmarks. In the experiments, the models maintained nearly the same performance even when trained on only about 31％ of the full dataset, while training time was reduced by up to 69.6％. Applying the maximum reduction rate, training that previously took 10 hours could be completed in about 3 hours. The research team primarily conducted experiments in an environment using four GPUs and also confirmed better performance than conventional methods in an eight-GPU environment. This study shows that AI training efficiency can be improved without adding more GPUs if the training data are selected and distributed effectively. In the future, the technology is expected to help companies and research institutions reduce the time and GPU costs required for additional training when developing AI for specific purposes such as coding assistance or medical question answering. < Figure 1. Conceptual diagram of the clustering-based balanced data sampling and allocation method developed by the research team. > Professor Jae-Gil Lee said, “To improve the efficiency of AI training, it is important not only to use more GPUs, but also to intelligently select which data should be used for learning,” adding, “This study is significant because it shows that existing GPU resources can be used more efficiently to reduce the training burden of large-scale AI models.” Hyunjin Kim, a Ph.D. student in the KAIST School of Computing, participated in the study as the first author, with Professor Jae-Gil Lee as the corresponding author. Youngeun Nam, a postdoctoral researcher at the University of Toronto, and Jaemin Han and Wonhyeok Choi, master’s students at the KAIST Graduate School of Data Science, participated as co-authors. The research findings were presented in October 2026 at the Conference on Empirical Methods in Natural Language Processing (EMNLP 2026), a major international conference in natural language processing. Paper title: Clustering-Based Balanced Sampling and Allocation with Data Parallelism for High-Performance Fine-Tuning, https://arxiv.org/pdf/2609.12584 Author information: Hyunjin Kim (Ph.D. student, KAIST School of Computing, first author), Youngeun Nam (postdoctoral researcher, University of Toronto; KAIST alumnus, co-author), Jaemin Han (master’s student, KAIST Graduate School of Data Science, co-author), Wonhyeok Choi (master’s student, KAIST Graduate School of Data Science, co-author), Jae-Gil Lee (Professor, KAIST School of Computing, corresponding author) This research was supported by the Institute of Information & Communications Technology Planning & Evaluation (IITP), funded by the Ministry of Science and ICT, through the SW Computing Industry Core Technology Development Program’s SW Star Lab project. The project title is “DB4DL: High-Usability and Performance In-Memory Distributed DBMS for Deep Learning” (RS-2020-II200862).

KAIST Develops Ruthenium Catalyst that Matches Pla..

< Research team and schematic illustration of the atom-layer neighbor structure. Top row, from left: Research Professor Seongbeen Kim, Dr. Hyunwoo Jun, and Dr. Jihye Park of KAIST; and Dr. Sungjun Kim of KRICT. Bottom row, from left: Dr. Joonhee Moon of KBSI; Professor Jang Yong Lee of Konkuk University; and Professors Hyungjun Kim and Jinwoo Lee of KAIST. > Lowering the cost of green hydrogen production requires catalysts that produce hydrogen efficiently while using smaller amounts of expensive precious metals. A Korean research team has developed a catalyst that brings three or four ruthenium (Ru) atoms together so they can work cooperatively. The catalyst achieved hydrogen production performance comparable to that of a commercial platinum catalyst while using only one-quarter as much precious metal, potentially helping to lower the cost of green hydrogen production. KAIST (President Choongsik Bae) announced on October 28 that a research team led by Professor Jinwoo Lee from the Department of Chemical and Biomolecular Engineering has developed a high-performance water electrolysis catalyst based on an “atom-layer neighbor” structure. The work was carried out in collaboration with Professor Hyungjun Kim from the Department of Chemistry, Professor Jang Yong Lee from Konkuk University, Dr. Joonhee Moon from the Korea Basic Science Institute (KBSI), and Dr. Sungjun Kim from the Korea Research Institute of Chemical Technology (KRICT). Water electrolysis uses electricity to split water into hydrogen and oxygen. When powered by renewable energy sources such as solar and wind, the process produces “green hydrogen” with low carbon emissions. However, catalysts for the hydrogen-producing reaction rely on expensive precious metals, creating a need for technologies that maintain performance while reducing precious metal use. In conventional precious metal nanoparticle catalysts, many atoms are clustered together, leaving those buried inside the particles largely unable to participate in the reaction. As a result, only a fraction of the precious metal is actively used. To address this problem, researchers have explored “single-atom catalysts,” in which individual metal atoms are dispersed across a support. However, their isolated arrangement can limit performance in reactions that require several atoms to work together. The research team sought to maximize the use of each atom while enabling neighboring atoms to cooperate. The team created “atom-layer neighbors,” small groups of ruthenium atoms arranged in a layer just one atom thick. This structure keeps the atoms close enough to work together in the reaction without forming large particles. Maintaining this arrangement requires a support—a base material that holds the ruthenium atoms in place. During the initial stage of catalyst synthesis, the researchers used molecular interactions to prevent the atoms from aggregating into large particles. During subsequent heat treatment, strong metal–support interactions anchored the atoms to a tungsten carbide (WCₓ) support. The team also varied the number of atoms in each group to identify the most effective structure for hydrogen production. The researchers also optimized the local environment surrounding the active atoms. Hydroxyl species (OH) on the support surface strengthened the hydrogen-bond network among water molecules, facilitating faster proton (hydrogen ion) transfer. By controlling both the arrangement of the catalytically active atoms and the interactions among surrounding water molecules, the team enabled the reaction to proceed efficiently. When tested in an anion-exchange membrane water electrolyzer (AEMWE) cell, the catalyst achieved a current density of 9.82 amperes per square centimeter (A/cm²) at 2.0 volts (V) and 80°C. Current density measures the current flowing per unit of electrode area and indicates the rate of hydrogen production over that area. The catalyst delivered performance comparable to that of a commercial platinum catalyst while using 75％ less precious metal. Its catalytic activity per unit mass of precious metal was approximately four times that of the platinum catalyst. The team also evaluated the catalyst’s durability during extended operation. In a 500-hour continuous operation test, its performance loss was about half that of the platinum catalyst used for comparison. The results demonstrate that the catalyst can deliver high performance with a small amount of precious metal while maintaining stability over extended operation times. The findings could help reduce precious metal use in hydrogen production equipment and lower catalyst replacement and maintenance costs. If the performance is validated at a larger scale and over longer operating periods, the catalyst could help lower the cost of green hydrogen production. “In this study, we brought small groups of precious metal atoms together and optimized the surrounding environment for water reactions, enabling efficient and sustained hydrogen production with a small amount of precious metal,” said Research Professor Seongbeen Kim, a co-first author of the study. He added that the team hopes the findings will help reduce precious metal use in hydrogen production equipment and guide efforts to improve other catalysts. Professor Kim and researchers Hyunwoo Jun and Jihye Park of KAIST, together with Dr. Sungjun Kim of KRICT, contributed as co-first authors. The findings were published online in Nature Communications on September 22 (Vol. 17, Article No. 11021). Paper title: Stabilizing ruthenium atom layer neighbors through molecular and metal support interactions for water electrolyzers DOI: https://doi.org/10.1038/s41467-026-78037-3 This research was supported by National Research Foundation of Korea (NRF) grants funded by the Ministry of Science and ICT (MSIT) (RS-2023-00235596, RS-2026-25474505, RS-2026-25477690, and RS-2024-00466554) and by the Basic Science Research Program of the Korea Basic Science Institute (KBSI) (C526111).

KAIST Uses Wireless Neural Chip to Track Brain Cha..

< The research team. From left: Professor Kyung-In Jang, Ph.D. candidate Jieun Son, Postdoctoral Researcher Saehyuck Oh, and Postdoctoral Researcher Jeongdae Ha > What happens in the brain after an Alzheimer’s drug is administered? A joint research team led by KAIST has used a small, implantable neural chip to wirelessly monitor real-time changes in a neurotransmitter involved in memory and learning. In miniature pigs (minipigs), the team observed an increase in the neurotransmitter signal following drug administration, followed by a gradual return to baseline. The technology could help researchers assess how drugs act in the brain and develop stimulation methods that adapt to the brain’s response. KAIST (President Choongsik Bae) announced on October 27 that a research team led by Professor Kyung-In Jang from the Department of Mechanical Engineering, in collaboration with researchers from the Daegu Gyeongbuk Institute of Science and Technology (DGIST), the Korea Brain Research Institute (KBRI), and the Korea Research Institute of Bioscience and Biotechnology (KRIBB), has developed an implantable neural chip that integrates a sensor for detecting the neurotransmitter acetylcholine, a light-stimulation component, and miniaturized wireless measurement technology < Figure 1. Structure and operating principle of the implantable wireless neural chip that measures acetylcholine changes in the brain > Acetylcholine transmits signals between nerve cells and plays an important role in memory and learning. Donepezil, a drug used to treat Alzheimer’s disease, inhibits acetylcholinesterase, the enzyme that breaks down acetylcholine, allowing the neurotransmitter to remain active for longer periods of time. Tracking changes in acetylcholine in specific brain regions after drug administration can help researchers better understand the drug’s action. The research team implanted the neural chip in the prefrontal cortex of minipigs, administered donepezil, and wirelessly measured acetylcholine signals. The prefrontal cortex is involved in cognitive functions such as judgment, attention, and memory. Following drug administration, the acetylcholine signal increased and then gradually returned to baseline. The findings demonstrate that chemical responses in a specific brain region can be continuously monitored using an external device. At the heart of the sensing technology is a layer that converts changes in neurotransmitter levels into electrical signals. The researchers combined porous chitosan with ultrathin silver–platinum (Ag＠Pt) nanowires. The porous structure provides a large surface area for attaching the enzymes used to detect acetylcholine. The nanowires convert hydrogen peroxide produced by the enzymatic reaction into electrical signals. The team also designed the chip to minimize interference from other molecules. The researchers placed an enzyme-coated electrode on one side of the thin probe inserted into the brain and an electrode without enzymes on the other. By comparing the two signals, the system reduces interference detected by both electrodes and isolates changes associated with acetylcholine. The measurements are processed by a miniaturized circuit and transmitted to an external device via Bluetooth. In addition to sensing changes in the brain, the neural chip can stimulate nerve cells with light. In a separate experiment involving transgenic mice, the team confirmed that acetylcholine signals increased in response to light stimulation delivered by a micro-light-emitting diode (μLED) integrated into the chip. The animal experiments drew on the expertise of the partner institutions. KRIBB provided the minipigs and support for surgical procedures, while KBRI supported mouse surgeries and experiments. The technology could help researchers investigate how drugs for brain diseases act in the brain. In particular, it could enable comparisons of neurotransmitter responses across different drug doses and dosing intervals. The study also lays the groundwork for research into closed-loop neural interfaces, which adjust stimulation based on ongoing measurements of the brain’s response. In such systems, a sensor detects changes following stimulation, and the intensity or timing of subsequent stimulation is adjusted accordingly. By integrating sensing and light stimulation into a single chip, the study provides a foundation for developing these systems. However, the study measured neurotransmitter changes in animal brains and did not demonstrate therapeutic effects on Alzheimer’s disease or improvements in memory. Further validation is needed before the technology can be used in humans, including assessments of its long-term safety and the stability of its measurement performance after implantation. < Figure 2. Implantable neural probe placed on a fingertip > “This study lays the groundwork for technologies that monitor changes in the brain and adjust stimulation based on its response,” said Professor Jang. He added that the technology could help researchers investigate how drugs act in the brain and identify stimulation methods tailored to the brain’s condition. The research team plans to commercialize the technology through N-Side Inc. (CEO Kyung-In Jang), a faculty-founded startup. Jieun Son, Saehyuck Oh, and Jeongdae Ha from DGIST contributed as co-first authors, and Professor Jang served as the corresponding author. The findings were published on October 14 in Device, a Cell Press journal covering devices and applied technologies. Paper title: Implantable neural chip with electrochemical sensing and optogenetic stimulation for monitoring of cholinergic pharmacological responses, DOI: 10.1016/j.device.2026.101329

KAIST Predicts the Lifetime of Blue OLEDs by Measu..

< A magneto-electroluminescence measurement setup and a blue phosphorescent OLED sample, shown with Professor Seunghyup Yoo (left) and Dr. Jeoungmin Ji (right). > Could a magnetic field be used to identify in advance which OLEDs will keep smartphone and TV screens bright the longest? A KAIST research team has developed a technique that predicts OLED lifetime by analyzing small changes in brightness caused by a magnetic field. The technique reduces the need for long operating tests and is expected to support the development of longer-lasting displays. KAIST (President Choongsik Bae) announced on October 22 that a research team led by Professor Seunghyup Yoo of the School of Electrical Engineering has developed a method that predicts the lifetime of blue phosphorescent OLEDs from changes in their brightness under a magnetic field. The study was conducted jointly with Professor Jang Hyuk Kwon’s team at Kyung Hee University, Professor Jae-Hyun Lee’s team at Hanbat National University, and researchers at ASTA Inc. OLEDs are devices that emit light on their own when an electric current flows through them, and they are used in smartphone and TV screens. A screen produces a wide range of colors by combining red, green, and blue light. The blue phosphorescent OLED examined in this study uses electrical energy efficiently to emit blue light. Achieving a long lifetime while maintaining high efficiency is a key challenge in its development. Conventionally, determining how long an OLED will last has required operating it for tens to hundreds of hours while monitoring the decrease in brightness. This process takes a long time, and the device’s performance also degrades during testing. The research team focused on the phenomenon in which a magnetic field slightly changes the brightness of an OLED. Inside an OLED, light is produced when electrically charged particles meet and recombine, and a magnetic field can affect this process. The researchers operated OLEDs at a constant current and precisely measured changes in brightness while varying the magnetic field strength with an electromagnet. In this experiment, the OLEDs became slightly brighter under the magnetic field. This change contains several different responses. The team separated the component most closely related to lifetime degradation and converted it into a single evaluation metric. < Figure 3. Measurement of changes in OLED light emission under a magnetic field, and the principle of lifetime prediction. The left panel is a schematic of the measurement, in which a magnetic field is applied to an OLED operated at a constant current and the resulting change in light emission is measured. The right panel shows the relationship between the magnetic-field-based lifetime evaluation metric and the operational lifetime. The analysis was performed separately at the points where the luminance had decreased to 70％ (LT₇₀) and 50％ (LT₅₀) of its initial value, and both showed a strong linear correlation between the metric and the logarithm of the lifetime. The inset shows the metric for each device measured in the initial state (LT₁₀₀), before long-term operation. > The team applied this metric to four types of blue phosphorescent OLEDs that were made with the same light-emitting material but had different internal structures. Devices that showed a smaller brightness response to the magnetic field from the start maintained their initial luminance longer in long-term operating tests. This indicates that the brightness response to a magnetic field can serve as an indicator of OLED lifetime. The magnetic field was not used to extend OLED lifetime. Instead, the researchers measured how OLEDs respond to a magnetic field to identify in advance which devices will last longer. By measuring and analyzing brightness changes too small to see with the naked eye, the team obtained a basis for estimating how long a device will operate. The team also investigated why this response occurs. Over long periods of use, bonds in the organic molecules inside an OLED can break and form defects that interfere with light emission. Through molecular simulations and analysis, the researchers explained that these defects can affect the light-emitting process and that this effect can be reflected in the brightness response to a magnetic field. Display companies and research institutes can use this technique when developing new OLEDs. For example, when several prototypes with different internal structures are made from the same light-emitting material, early measurements can identify the structures most likely to last longer, so that further development can focus on them. Because not every prototype needs to be operated for a long time before results are available, the method is expected to help reduce development time and cost. However, further validation is needed before this relationship can be applied to a wider range of OLEDs. When the materials or internal structure change, the main factors that respond to a magnetic field can also change, so each type of device requires its own analysis. < concept image > “This result can be used to quickly select materials and device structures with long lifetimes in the early stages of OLED development,” said Professor Seunghyup Yoo. He added that, with further validation under various conditions, the method is expected to support the development of high-efficiency, long-lifetime blue OLEDs and to be used as a non-destructive evaluation technique. Dr. Jeoungmin Ji of KAIST led the study as first author, and Professor Seunghyup Yoo served as corresponding author. The results were published online in Nature Communications on September 30. Paper title: Magnetic Electroluminescent response as a predictive indicator of operational lifetime in phosphorescent blue OLEDs DOI: 10.1038/s41467-026-78180-x This research was supported by the Technology Innovation Program (20020408, Development of direct analytic methods for exciton dynamics in OLEDs and lifetime-enhancing technologies based thereon) funded by the Ministry of Trade, Industry and Resource (Korea).

KAIST Develops Gravity-Independent Wearable Eyedro..

< The research team. From left: Jaehun Jeon, doctoral student (Department of Bio and Brain Engineering, KAIST); Ki-Hun Jeong, professor (Department of Bio and Brain Engineering, KAIST); Se Joon Woo, professor (Seoul National University Bundang Hospital) > Applying eyedrops is difficult in space, where there is almost no gravity. A KAIST research team has developed a wearable patch that delivers eyedrops to the eye without relying on gravity, for use in space. < Figure 1. Wearable eyedrop delivery patch for use in microgravity environments (left), its structure (center), and a photograph (right). > KAIST (President Choongsik Bae) announced on October 21 that a research team led by Professor Ki-Hun Jeong from the Department of Bio and Brain Engineering, in collaboration with a research team led by Professor Se Joon Woo from Seoul National University Bundang Hospital, has developed a wearable eyedrop delivery patch designed for the conditions of space. During long-duration space exploration, technologies that allow astronauts to manage their health without direct help from medical staff are essential. In particular, microgravity causes body fluids to shift toward the head, which can change the structure of the eyes and visual function. However, the standard method of applying eyedrops on Earth depends on drops falling by gravity, so it cannot be used in space without modification. The research team combined microchannels that carry liquid with a chemical reaction chamber that generates pressure, all in a single flexible patch. When the user wears the patch and presses the actuation area, the pressure of the gas generated inside delivers the eyedrop to the eye. The patch works through a reaction between vitamin C and baking soda. When ascorbic acid (vitamin C) and sodium bicarbonate (the main component of baking soda) come into contact with water, carbon dioxide gas is produced. The resulting pressure pushes the eyedrop along the microchannel. Instead of relying on gravity to move the eyedrop, the patch uses pressure generated inside it. A filter placed between the reaction chamber and the eyedrop chamber allows gas to pass but blocks liquid. This keeps the reactants and the eyedrop separate while still transmitting the gas pressure. Because the patch needs no external pump or electrical drive, it is expected to be suitable for settings where equipment is limited. The research team confirmed that the patch released eyedrops consistently even when the direction of gravity was changed. In an on-body evaluation with human participants, the patch also delivered the drug at a level similar to that of the conventional eyedrop method. These results demonstrate the potential of an eyedrop delivery method that does not depend on the direction of gravity. < Figure 2. Results of the on-body evaluation of the patch’s drug delivery performance. > The study presents a personal drug delivery technology that allows people to apply eyedrops by themselves in space. Note, however, that these results come from experiments with different gravity directions and an on-body evaluation conducted on Earth. Use in an actual space environment still requires further verification. The technology could also be used on Earth. It is expected to help people who have difficulty aiming an eyedrop container at the eye because of hand tremors or other physical limitations, such as patients with Parkinson’s disease. “The flexible eyedrop patch developed in this study is a wearable platform that can deliver medication without complex additional equipment, and we expect it to expand to space medicine and other settings,” said Professor Jeong. Jaehun Jeon, a doctoral student in the Department of Bio and Brain Engineering at KAIST, was the first author of the study. The results were published online on October 7 in the international journal Sensors and Actuators B: Chemical. Paper title: Wearable self-instillation microfluidic eyedrop patch for space medicine, DOI: https://doi.org/10.1016/j.snb.2026.140885 This research was supported by the Korean ARPA-H Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare (RS-2024-00512384); the National Research Foundation of Korea (NRF), funded by the Ministry of Science and ICT (RS-2025-00523089); and the Technology Innovation Program funded by the Ministry of Trade, Industry and Resource (RS-2024-00432381)

KAIST Develops AI That Reconstructs Cloud-Covered ..

< The research team. From left: Professor Munchurl Kim (School of Electrical Engineering) and Dr. Jeonghyeok Do (Information & Electronics Research Institute), KAIST > Satellite images of the Earth are not always easy to use. Clouds can block the view of the ground, and images captured with certain imaging methods look different from ordinary photographs, which makes them difficult to interpret. A KAIST research team has developed an artificial intelligence (AI) model that learns the appearance of diverse terrain and cities. The model can restore cloud-covered images and convert radar images into images that resemble ordinary photographs. It applies what it learned while generating satellite images to a range of image-processing tasks, so it is expected to expand the use of satellite data in disaster response and environmental monitoring. KAIST (President Choongsik Bae) announced on October 20 that a research team led by Professor Munchurl Kim from the School of Electrical Engineering has developed GeoCore-9B, an AI model that can be used to generate, restore, and convert satellite images. During training, the model’s core image-generating component learns image features by adjusting about 9 billion internal values, known as parameters. GeoCore-9B is pretrained on a large volume of satellite data and can then be adapted to a variety of tasks. For this reason, it is classified as a ‘foundation model,’ a type of AI model that serves as the basis for a wide range of AI technologies. < Conceptual illustration of the research applications of GeoCore-9B. > Satellite images are essential for monitoring changes in forests and cities and for responding to disasters such as floods and wildfires. However, optical images, which record sunlight reflected from the Earth’s surface, cannot clearly capture the ground when clouds are thick or at night. Synthetic aperture radar (SAR) observes the ground by transmitting radio waves and measuring the signals reflected back. It can acquire images of the Earth’s surface at night and in cloudy weather. However, SAR images differ from ordinary photographs in brightness and patterns, which makes them difficult to interpret. To create a common basis for processing these different types of satellite data, the research team trained the core image-generating model from scratch on Earth observation data. Instead of adapting the core component of a model built to generate ordinary photographs, the team trained a new model on overhead satellite views of terrain, forests, ports, and cities. The model was trained on 10 million pairs of satellite images from around the world and sentences describing each image. For example, an image of a port was paired with a description stating that ships and warehouses are located there. An image of a rural area was paired with a description stating that a road runs between fields. By learning from this large collection of described images, the AI learned the features of diverse terrain and urban areas. < From pretraining to task-specific applications. > A key feature of this study is that the AI learned not only what an image contains, but also which location on Earth the image shows and how large an area it covers. In satellite images, objects with similar shapes may be small buildings or large industrial facilities, depending on their actual size. It is therefore important to take into account both where an image was taken and its spatial scale. The team trained the model using latitude, longitude, and ground sample distance (GSD) together with the descriptive sentences. Latitude and longitude indicate a position on Earth. GSD indicates the size of the ground area represented by a single pixel, the smallest unit of an image. For example, at a GSD of 1 meter, one pixel represents an area of about 1 meter by 1 meter on the ground. Using this information, the AI learned to generate images that match a given location and spatial scale. The team also applied a training technique called ‘Geospatial Semantic Alignment,’ which helps the model represent the structure of terrain and cities more accurately. In this technique, the generative model learns by referring to features extracted by an AI specialized in satellite imagery. These features serve as a training reference that helps the model learn structural elements of an image, such as how roads connect and where farmland boundaries lie. The specialized AI is used only during training, so it does not add computation when the model generates images. The researchers then fine-tuned GeoCore-9B separately for cloud removal and for radar image conversion. They used LoRA (Low-Rank Adaptation), a method that keeps the core model fixed and trains only a small number of additional parameters. This approach reduces the burden of retraining an entire large AI model from scratch for each task. In evaluations using publicly available satellite datasets, GeoCore-9B outperformed existing task-specific models at restoring cloud-covered images. It scored higher on the key metrics that measure brightness, color, and structural similarity to the reference images. In converting radar images into optical images, it also achieved the best results on some of the metrics that assess the quality and fine detail of the generated images. < Restoration results for cloud-covered observation images. > < Changes in generated satellite images with ground sample distance (GSD). > These results demonstrate that knowledge gained from generating satellite images can also be applied to restoring and converting actual observation images. The model is expected to make greater use of data whose value has been limited by cloud cover or imaging methods. It is also expected to support the assessment of conditions in disaster areas and the monitoring of changes in forests, farmland, and cities. In addition, adapting a single common model to multiple tasks can reduce the cost and time required to develop a separate AI for each task. The team plans to validate the model’s performance across a wider range of sensors and regions. To apply the model to real-world disaster response, the team also plans to expand its research to accurately predicting changes on the ground, such as floods, building damage, wildfires, and urban expansion. The team will also explore defense applications such as surveillance and reconnaissance. “By training the AI on where each satellite image was taken and the actual size of the terrain it shows, we made it possible to use a single model for multiple tasks such as cloud removal and image conversion,” said Professor Kim. He added that he expects the work to make satellite data limited by weather or imaging methods more widely usable, which will help with assessing disaster situations and monitoring changes in forests and cities. Dr. Jeonghyeok Do of KAIST’s Information & Electronics Research Institute led the study as first author, and Professor Munchurl Kim of the School of Electrical Engineering participated as corresponding author. The findings have been accepted for presentation at NeurIPS 2026, a major international conference in artificial intelligence. The paper was posted on the preprint server arXiv on August 3, 2026. Paper title: GeoCore-9B: Towards Geo-Aware Generative Foundation Models in Earth Observation DOI: https://doi.org/10.48550/arXiv.2608.01896 Authors: Jeonghyeok Do, Munchurl Kim This research was carried out using high-performance computing resources provided through the ‘Advanced GPU Utilization Support Program’ of the Ministry of Science and ICT and the National IT Industry Promotion Agency. GPUs are semiconductor chips that process a large number of calculations at the same time, and they are used to train large-scale AI models.

Same “Depressed” Body Movements, Opposite Brain Re..

< The research team. From left: Professor Young-Gyun Park (Department of Bio and Brain Engineering); Ratna Aditya Apsari, master’s graduate (Program of Brain and Cognitive Engineering); and Somi Lee, master’s graduate (Department of Bio and Brain Engineering). > Can changing your posture change your mood? Previous studies have suggested that body posture and movement can influence emotions. KAIST researchers have extended this line of research by showing that, when people perform the same “depressed” body movements, their brain responses differ depending on the level of their depressive symptoms. The findings are expected to provide a basis for research on which movements or exercises help which people, and on mental health care methods suited to each individual’s condition. KAIST (President Choongsik Bae) announced on October 19 that a research team led by Professor Young-Gyun Park from the Department of Bio and Brain Engineering used a wireless brain imaging device to measure brain activity while people moved their bodies and identified differences in brain responses according to the level of depressive symptoms. When people feel sad or exhausted, they may hug their knees, sigh deeply, or let their shoulders droop. These movements express emotions. However, body posture can also influence emotions. One study has reported that people with depressive symptoms may experience more positive emotions and less fatigue when they adopt an upright posture. Building on previous studies of the relationship between posture and emotion, this study asked whether the brain responds differently to the same movements depending on the level of depressive symptoms. The researchers started from the question of whether brain responses can differ even when outward behavior is the same. < Figure 1. Emotion-expressing bodily movements and experimental setup used in this study. > Answering this question requires measuring brain activity while the body is moving. However, functional magnetic resonance imaging (fMRI), a widely used brain imaging method, requires participants to lie still inside a narrow scanner. This makes it difficult to examine brain responses during movements such as hugging the knees or skipping. The team instead used a wireless functional near-infrared spectroscopy (fNIRS) device worn on the forehead. The device uses light to measure changes in the oxygenation of blood in the brain. Because blood flow and oxygenation change when the brain is active, these measurements provide an indirect indicator of brain activity. Participants can move while wearing the sensor, which allows brain responses to be measured during actual movements. < Figure 2. Wireless fNIRS device and measurement channel layout. > Fifty-two adults aged 20 to 35 took part in the experiment. Using a depressive symptom questionnaire (PHQ-9) and a mental health screening interview, the researchers divided the participants into a group of 27 with minimal depressive symptoms and a group of 25 with elevated depressive symptoms. The questionnaire asks how often a person has experienced nine symptoms, such as loss of interest and sleep problems, over the past two weeks. The groups were defined by questionnaire-measured symptom levels rather than by a clinical diagnosis of depression. Participants in each group were randomly assigned to either a negative or a positive emotional movement task. The negative emotional movements were hugging the knees, sighing loudly, and drooping the shoulders. The positive emotional movements were a victory pose, skipping, and laughing while clapping. Participants performed each movement for 30 seconds followed by 60 seconds of rest, and repeated this four times. < Figure 3. Changes in prefrontal activity during negative emotional movements. > The researchers also took into account that, during movement, the sensor might shift and changes in scalp blood flow might mix into the measured signal. For example, a change caused by sensor movement could appear to be a change in brain activity. To address this, the team applied correction procedures that reduce these effects before analyzing brain responses. The results showed that the two groups had opposite brain responses when they hugged their knees, sighed, or drooped their shoulders. The difference appeared in the left anterior prefrontal cortex, the front part of the brain region located just behind the forehead. Activity in this region increased in the group with minimal depressive symptoms but decreased in the group with elevated depressive symptoms. < Figure 4. Differences in prefrontal activity between the two groups for each movement. > This difference in the same region appeared repeatedly across all three distinct movements: hugging the knees, sighing, and drooping the shoulders. In contrast, no common brain activation pattern was observed across the three positive emotional movements. These results show that the brain may process the same movements differently depending on the level of depressive symptoms. The researchers proposed the left anterior prefrontal cortex, where the difference appeared repeatedly, as an important candidate region for understanding the relationship between body movement and emotion. The findings suggest that future research on movement- or exercise-based mental health care should take each individual’s condition into account. For example, examining both brain responses and actual mood changes after the same movement could lead to studies that test which movements help people at different levels of depressive symptoms. The approach can also be used in studies that repeatedly measure responses to specific movements with portable brain imaging devices. It may further be extended to neurofeedback research, in which people practice movements while viewing their own brain responses on a screen. However, this study did not confirm that any specific movement has a therapeutic effect on depression. An increase in brain activity does not mean that a person’s mood improved, and a decrease does not mean that the person became more depressed. Further research is needed to determine why the two groups responded differently and how these differences relate to actual changes in emotion. “This study is significant in that it confirmed, through measurements taken during actual movements, that the brain responds differently to the same movements depending on the level of depressive symptoms,” said Professor Park. Professor Park added that the findings could provide a basis for understanding the relationship between body movement and emotion, and for future research on movement-based mental health care tailored to each individual’s condition. Ratna Aditya Apsari, who earned a master’s degree in the KAIST Program of Brain and Cognitive Engineering, and Somi Lee, who earned a master’s degree in the Department of Bio and Brain Engineering, served as co-first authors, with Professor Park as the corresponding author. The findings were published online on July 31 in Brain Research Bulletin, an international journal in the field of neuroscience. Paper title: Prefrontal responses to negative emotional movements in healthy and symptomatic individuals, DOI: 10.1016/j.brainresbull.2026.112064 This research was supported by National Research Foundation of Korea (NRF) grants funded by the Korean government (MSIT) (Grant Nos. RS-2021-NR061634 and RS-2024-00439379).

KAIST Develops Transistor-Based Encryption That Co..

< The research team. Top row, from left: Ph.D. candidates Taewook Go and Jingyao Yu. Bottom row, from left: Professor Kyung Min Kim and Ph.D. candidates Huisu Noh and Myeongchan Ko (Department of Materials Science and Engineering, KAIST) > Some information, such as exam questions or bid prices, must remain unreadable until an agreed time, even if it has already been delivered. KAIST researchers have used changes in the electric current inside a semiconductor to develop encryption in which information is revealed only after a set amount of time has passed, or can be read correctly only at a specific time. In addition to the key needed to restore the original information, the technology adds a second condition: when the information is read. KAIST (President Choongsik Bae) announced on October 14 that a research team led by Professor Kyung Min Kim from the Department of Materials Science and Engineering has developed a dual-state programmable oxide transistor (DUPOT), a single device in which long-term and short-term memory can be controlled independently. The team then used the transistor to implement time-based cryptography. < Overview of the dual-state programmable oxide transistor (DUPOT) and concept of time-based cryptography > A transistor is a small semiconductor component that controls the flow of electric current. The DUPOT is made from oxide materials, and its two electrical states can be independently set by applying appropriate voltages. The term “programmable” refers to this ability to set the device to a desired state. < Long-term and short-term memory characteristics of the dual-state programmable oxide transistor > Encryption methods that use time as a condition for accessing information have been studied before. What sets this study apart is that the time at which information can be read is determined by physical changes within the semiconductor itself, rather than by external clocks or software alone. To achieve this, the researchers implemented two types of memory in a single transistor: long-term memory, which is retained for a long period, and short-term memory, which changes over time. In this context, memory refers to a device’s ability to retain a state that has been changed by an applied voltage. The team used the long-lasting state to set the reference for the information and the time-varying state to control when the target information appears. Conventional memory transistors store information mainly by changing the threshold voltage, the voltage at which current begins to flow. The DUPOT, by contrast, can independently control both the threshold voltage and the saturation current, the level of current maintained under fixed operating conditions. As a result, the voltage at which current begins to flow and the amount of current that flows can be set separately. The two states are produced by different mechanisms. When a positive voltage is applied to the gate, the terminal that controls the current, electrons become trapped in hafnium oxide (HfO₂), an insulating layer that does not easily conduct electricity. This shifts the threshold voltage. Because this state is retained for a long time, it serves as long-term memory. When a negative voltage is applied to the gate, oxygen ions (electrically charged oxygen particles) move to the interface between the insulating layer and the channel through which current flows, which reduces the saturation current. Once the voltage is removed, the oxygen ions gradually redistribute and the current recovers. The team used this change as short-term memory. Using several transistors, the researchers configured how the current changed over time and set a readout criterion for each device to demonstrate two encryption methods. In time-release encryption, information appears only after a certain amount of time has passed. In the experiment, the difference in current between the target pattern and the background grew over time, and the letter “T” became readable after 8.4 seconds. In time-bound encryption, the correct information can be read only within a specified time window. In the experiment, a decoy pattern, the letter “K,” appeared at 1.5 seconds, while the target letter “T” appeared at the designated time of 37 seconds. The same array therefore displays different information depending on when it is read. These time points were chosen for the experiment. The team also proposed an encryption method for cloud environments, in which data are stored on remote servers accessed over the internet. In this method, different time conditions are applied to the ciphertext stored in the cloud and to the decryption key held by the user Ciphertext is data that has been converted into a form in which the original information cannot be recognized, and a decryption key is the information needed to convert it back. In a demonstration using the device array, the researchers restored the original information by combining the ciphertext and the key that matched each time condition. The study shows that a semiconductor can do more than store information: it can also determine when that information is read. The technology could be used to protect information that must not be disclosed before a certain time or that must be readable only under specific time conditions. However, the technology cannot revoke access to copies of information that have already been obtained. “The significance of this work is that long-lasting memory and time-varying memory can each be controlled independently within a single transistor and used for time-based encryption,” said Professor Kyung Min Kim. Kim added that the team plans to build on the time-dependent changes of the semiconductor itself to develop the technology further for cloud and smart-device security, as well as for a range of information-processing technologies. The current study has demonstrated the feasibility of time-based encryption at the array level. The team plans to lower the operating voltage and improve device-to-device uniformity and consistency over repeated operating cycles, so that the devices can work stably in a variety of environments. The researchers then aim to extend the technology to time-based security and to neuromorphic devices, which mimic how neurons in the brain work and connect. The study, with Ph.D. candidate Huisu Noh of the Department of Materials Science and Engineering as first author and Professor Kyung Min Kim as corresponding author, was published online in Advanced Materials on September 10, 2026. Paper title: Dual-State Programmable Oxide Transistor for Time-Based Cryptography, DOI: 10.1002/adma.74951 This research was supported by the Next-Generation Intelligent Semiconductor Technology Development Program of the National Research Foundation of Korea (NRF) (RS-2025-02433006).

KAIST Controls Where 2D Semiconductor Crystals Beg..

< The research team. Top row, from left: Jeongwon Park, Ph.D. (KAIST); Jiyun Kim, master’s student (KAIST); Sumin Kang, M.S. (KAIST); Jieun Oh, integrated M.S.–Ph.D. student (Hanyang University); and Youngmin Sunwoo, integrated M.S.–Ph.D. student (UNIST). Bottom row, from left: Kibum Kang, professor (KAIST); Woo-Hee Kim, professor (Hanyang University); and Byungjo Kim, professor (UNIST). > Two-dimensional (2D) semiconductors, only a few atomic layers thick, are a promising material for making AI chips faster and more energy efficient by stacking additional semiconductor layers on top of existing computing circuits. However, when crystals form at multiple random locations during growth, it is difficult to produce high-quality semiconductors at the desired positions. A KAIST joint research team has developed a technology that sets in advance where a crystal begins to grow and allows only one crystal to grow at that position. By controlling the starting point of semiconductor growth, the work addresses a key challenge in realizing next-generation stacked semiconductors. KAIST announced that a research team led by Professor Kibum Kang from the Department of Materials Science and Engineering, in collaboration with UNIST, Hanyang University, and TDS Innovation (Co-CEOs Kibum Kang and Sanghyun Mo), has developed a new growth technique that precisely controls the location of nucleation, the stage at which 2D semiconductor crystals first form, using chemical species released from an oxide barrier. < Figure 2. Schematic of the principle by which oxygen acts as an etching flux that suppresses crystal formation, so that a 2D semiconductor begins to grow at only one designated point > When AI generates responses or processes large amounts of data, time and power are consumed not only by computation but also by moving data within the chip. Stacking the devices used for computation and data storage vertically, so that they are placed close together, shortens the distance that data must travel and can reduce processing delays and power consumption. This requires a technology that forms high-quality semiconductors directly at the required positions on top of circuits that have already been fabricated. Deposited semiconductors, in which devices are made by forming a thin film of semiconductor material directly on the required surface, are a promising technology for building stacked chips. 2D semiconductors are particularly well suited for stacking thin device layers because they retain the electrical properties of a semiconductor while being only a few atomic layers thick. The challenge is that, as a thin semiconductor film grows, the positions where crystals begin to form are inherently random. A crystal is a structure in which the atoms that make up the semiconductor are arranged in a regular, repeating pattern. When crystals grow separately from several locations and meet, grain boundaries form between them. These boundaries hinder the movement of electrons and can degrade the quality of the semiconductor. Area-selective growth techniques, which define in advance the regions where a semiconductor will grow, already existed. However, it has been difficult to control exactly where crystals first form within those regions and how many of them form. The research team achieved deterministic nucleation, in which nucleation that previously occurred at random takes place one at a time at defined positions. In addition to defining the region where the semiconductor will grow, the team also determined the point within that region where growth begins. The researchers named the technique etching-flux-mediated single-centered nucleation (EF-SCN). The key element is oxygen released from an oxide barrier, such as hafnium oxide (HfO₂), that surrounds the growth region. This oxygen acts as an etching flux that suppresses the formation of crystal nuclei of molybdenum disulfide (MoS₂), a representative 2D semiconductor. Etching refers to chemically removing material, and flux refers to the flow of a specific chemical species that is transported and supplied in a way that affects nearby chemical reactions. The suppressing effect is strong at the edges near the oxide barrier and becomes weaker toward the center, which is farthest from the barrier. Using this difference, the team caused a single crystal to begin growing at only one point in the center. While the crystal that started at the center continued to grow, the formation of new crystals around it was also suppressed, so that the single crystal could grow larger. In experiments, the team grew one MoS₂ single crystal at a designated central position in more than 99％ of 400 different patterns. A single crystal has one continuous atomic arrangement, and because it reduces the boundaries that form where multiple crystals meet, it is advantageous for producing high-quality semiconductors. Using MoS₂–MoSe₂ lateral heterostructures grown with this method, the team also deposited a thin film of ruthenium (Ru) metal selectively on the molybdenum diselenide (MoSe₂) regions only. The researchers further analyzed the principle behind this position control through growth experiments and simulations. < Figure 3. Comparison of conventional area-selective growth and EF-SCN, with representative results of single-centered nucleation > The significance of this work is that it turns the starting point of semiconductor thin-film growth, which had been determined at random, into a process parameter that can be designed. Instead of fabricating 2D semiconductors elsewhere and then transferring them, this approach could lead to a manufacturing technology that grows high-quality semiconductors directly at the required positions on a chip. Deposited semiconductors are expected to be adopted in the 2030s to improve the performance and lower the power consumption of AI chips. If applied to stacked AI chips, the technology could reduce the time and energy spent on moving data, which would shorten the time users wait for AI responses and reduce the power demands of data centers. “In thin-film growth, where a crystal forms is inherently a matter of probability, and the key significance of this study is that it allows that starting point to be determined at a desired position,” said Professor Kibum Kang. Professor Kang added that the technology will serve as the basis for stacked-channel processes, in which semiconductor devices are stacked vertically, by growing high-quality 2D semiconductors directly where they are needed. “In 2D semiconductors, the position where growth begins can change in response to even small chemical changes in the surroundings,” said Dr. Jeongwon Park, a co-first author. Park explained that the significance of the study lies in using this property to control where the semiconductor begins to grow. Dr. Park and master’s student Jiyun Kim from the KAIST Department of Materials Science and Engineering, Sumin Kang, who received a master’s degree from the same department, integrated M.S.–Ph.D. student Jieun Oh from the Department of Materials Science and Chemical Engineering at Hanyang University, and integrated M.S.–Ph.D. student Youngmin Sunwoo from the Graduate School of Semiconductor Materials and Devices Engineering at UNIST participated as co-first authors. The findings were published in Nature on October 8. Paper title: Spatially deterministic nucleation of 2D semiconductors by etching flux DOI: 10.1038/s41586-026-11095-1 This research was supported by the Nano and Material Technology Development Program and the Basic Science Research Program of the National Research Foundation of Korea (NRF), funded by the Ministry of Science and ICT (MSIT). Additional support was provided by the International Cooperation Program for Source Technology Development, the Development of vdW Materials and Process Technologies for Ultra-High-Density Semiconductors Program, the Next-Generation Intelligent Semiconductor Technology Development Program, the Ministry of Trade, Industry and Energy (MOTIE), the Korea Institute for Advancement of Technology (KIAT), the Commercialization Promotion Agency for R&D Outcomes (COMPA), the Innopolis Foundation, the Ministry of SMEs and Startups (MSS), and NVIDIA.