Kaist

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.

KAIST Develops a Beetle-Inspired Soft Robot That O..

  < Figure 1. Schematic overview of the structure and operation of BeetleBot. >   Researchers have developed a technology that lets a user move a small robot by bending and straightening a finger. The user can also pick up and move objects while checking the images and contact information that the robot sends back. The system combines artificial muscles, a camera, and tactile sensors in a small soft robot modeled on a beetle. It is expected to be used in remote tasks, such as inspecting narrow spaces that are difficult for people to reach and retrieving small objects. KAIST (President Choongsik Bae) announced on October 13 that a research team led by Professor Sang Ouk Kim from the Department of Materials Science and Engineering has developed a small soft robot called “BeetleBot.” The team worked with the research groups of Professor Wei Gao at the California Institute of Technology (Caltech) and Professor Hirotaka Sato at NTU. < Figure 2. Photograph of the miniaturized BeetleBot and the finger patch for wireless gesture control system. >   When people grasp an object, they use their eyes to find where it is and the sensation in their fingertips to check the contact. A robot that adjusts its work to its surroundings also needs its movement and sensing functions to be connected. Soft robots are built from soft, deformable materials. In these robots, however, actuation, sensing, and control are often built as separate parts or depend on complex devices. This has made it difficult to integrate several functions into a small robot. The researchers designed the robot to walk and grip objects using artificial muscles that contract in a similar way to biological muscles. They linked these muscles with a camera, a pressure sensor, and a stretchable sensor patch worn on the finger. With this setup, the user can control the robot while checking its surroundings and its contact with objects. < Figure 3. Real-time synchronization of gesture input, robot behavior, and visual feedback from the robot’s onboard camera. >   The team first analyzed how the leg muscles of a beetle relate to the movement of its joints. They used the results to decide where to place BeetleBot’s artificial muscles and how to design its legs. The artificial muscles were made by 3D printing a liquid crystal elastomer (LCE), a polymer that contracts when heated. They were combined with an actuation module that produces heat from electricity. The muscles in each part of the robot can be operated independently, and the order in which they operate can be adjusted. This allows the robot to walk forward and turn left or right. The researchers also built a soft gripper based on the mandibles that beetles use to hold their prey. The gripper combines artificial muscles with a capacitive pressure sensor, so it can grip an object and detect the contact. When the artificial muscles contract, the gripper closes around the object. The pressure sensor then detects the pressure applied during contact. When the object presses on the sensor, the distance between its internal electrodes changes, and so does its capacitance. The sensor uses this change to determine the contact state. To control the robot, the team used a stretchable sensor patch whose electrical resistance changes as the finger moves. When the finger bends and stretches the patch, the conductive paths formed by carbon nanotubes (CNTs) inside it change, which changes the resistance signal. The researchers sorted these signals into four commands: move forward, turn left, turn right, and grip an object. The commands are sent to BeetleBot over Bluetooth Low Energy, and the robot carries out the matching action. The user views the robot’s surroundings through a small onboard camera and checks contact with objects using data from the pressure sensor in the gripper. Based on this information, the user decides on the next action and sends a command with a finger movement. In this way, the system provides bidirectional feedback. Commands go from the user to the robot, and the information the robot detects goes back to the user. Using this control method, the researchers showed that BeetleBot could avoid obstacles, approach a target object, and then grip and carry it. By connecting artificial muscle-based walking and gripping, visual and tactile sensing, and user control in a single system, the study shows that small soft robots can carry out practical tasks. < Figure 4. Demonstration of Beetlebot task execution through closed-loop human–robot interaction using remote control and visual and tactile feedback. >   The results are expected to support the development of soft robots for remote tasks, such as checking conditions and retrieving small objects in spaces that are hard for people to access. Before the robot can be used more widely, for example in environmental exploration or industrial equipment inspection, the actuation speed, energy efficiency, and power supply of the artificial muscles will need to be improved. Its performance will also need to be tested in real working environments. “This study expands the potential uses of soft robots by connecting the movement of artificial muscles, sensing, and user control in a single system,” said Professor Sang Ouk Kim. He added that he expects the joint research with Caltech and other partners to contribute to soft robots that can work in spaces that are difficult for people to access. “A key feature of this work is the bidirectional feedback, which links the information detected by the robot to the user’s next action,” said Professor Inho Kim of Sungkyunkwan University, the first author. He added that he plans to increase the actuation speed and energy efficiency of the artificial muscles and improve control methods that use sensory information, so that the robot can be developed into a soft robot that works in a variety of environments. Professor Inho Kim earned his Ph.D. from the KAIST Department of Materials Science and Engineering. He is now an assistant professor in the Department of Chemistry at Sungkyunkwan University, and he took part in the study as first author while working as a postdoctoral researcher at Caltech. Professor Sang Ouk Kim of KAIST and Professor Wei Gao of Caltech are co-corresponding authors. The results were published in Science Advances on August 19, 2026. Paper title: BeetleBot: An integrated bioinspired soft robot for multimodal sensing and adaptive interaction, DOI: 10.1126/sciadv.aeg8849 This research was supported by the Heritage Medical Research Institute, the Meta Caltech Research Award, the National Research Foundation of Korea (RS-2024-00411904, RS-2025-00521316, RS-2025-25442577), and the Singapore Ministry of Education (RG140/20, RG82/24).

KAIST to Chart a Platform-FirstCourse for “Physica..

  < KAIST Humanoid Robot >   KAIST hosted “Physical AI KOREA 2026” at its Daejeon campus on October 13. At the event, KAIST presented physical AI platform technologies that can be applied not only to manufacturing but also to healthcare, defense, logistics, and everyday life, as well as strategies for research, education, and entrepreneurship that can be utilized by a wide range of companies. Humanoid robots are beginning to surpass human records in robotic competitions modeled after Olympic events such as track and field, and pilot applications have also begun at some manufacturing sites. Expectations for physical AI are growing, but KAIST’s vision extends beyond stadiums and selected factories. KAIST will announce a vision to expand physical AI into technologies used across diverse industries and everyday life and develop it into a future industry for Korea. President Choongsik Bae emphasized that “the key competition in physical AI is not about how closely robots resemble humans, but about how widely we can deploy AI that understands and moves within the physical world across diverse environments,” adding, “True to its reputation for strong fundamentals, KAIST will build foundational technologies for sensing, movement, and cooperation among multiple machines so that they can be used across a wide range of robots and industries. We will also foster the talent and companies that will lead this field.” On the day of the event, President Bae appeared on stage in a wearable robot developed by Professor Kyoungchul Kong’s research team and demonstrated how the movement data measured by the robot could be analyzed. This served as an example of how physical AI can also be used to understand human movement and health. < President Choongsik Bae in the WalkON Suit >   The wearable robot developed by Professor Kong's research team has primarily been studied as a technology for rehabilitation therapy and for helping people with disabilities walk. When combined with physical AI technology, the robot could be used not only as a walking-assistance device but also as a sensing device that monitors a person's movement. President Bae's demonstration will show KAIST's vision that physical AI can extend beyond manufacturing and industrial robots into people's health and daily lives. The event opened with a video tracing the development of robotics research at KAIST, followed by an overview of global trends in physical AI technology. The program featured presentations on platform technologies and demonstrations of robots, including humanoid robots and a robot pilot. Speakers then addressed the challenges of industrializing physical AI and the role KAIST can play, and President Bae presented a vision for building a physical AI industrial ecosystem in Korea. The event  also included a joint Q&A session with the presenters and media interviews. The event went beyond the question of “what kind of robot should we build” and presented KAIST’s answer to “what technologies are commonly needed across different robots and industries.” KAIST plans to combine its accumulated research capabilities in robotics, artificial intelligence, sensors, actuators, and data to present a technological framework for physical AI and identify areas of future focus. At the center of this framework are perceptual intelligence, which enables machines to “sense” the physical world; motor intelligence, which allows them to “move” appropriately according to the situation; and collective intelligence, which enables multiple robots and systems to “make decisions and act together.” Humanoids are one important form through which these capabilities can be realized, but KAIST does not limit the scope of physical AI to humanoids. Perceptual intelligence is technology that measures physical signals such as force, movement, and position, allowing AI to understand its surroundings and the condition of people. Motor intelligence uses this information to generate movements appropriate to the situation, such as walking or grasping objects. Collective intelligence allows multiple robots and sensors to share information and make decisions and take actions together. Rather than concentrating its capabilities on perfecting a single specific robot, KAIST plans to secure platform technologies that can be commonly used across multiple robots and industries. Key platform technologies include actuators that move robots with precision; technologies that measure and process physical data from the real world; data center technologies that manage and operate large-scale data; motor intelligence that autonomously generates movement; and sensory intelligence that understands surrounding environments and people. The event also showcased a variety of implementation examples. Professor Hae-Won Park's research team in the Department of Mechanical Engineering introduced a humanoid robot capable of carrying objects weighing more than 15 kg. Professor Hyunchul Shim's research team in the School of Electrical Engineering demonstrated PIBOT, a humanoid robot pilot that can operate an aircraft, showing the potential for applications in specialized work environments. < Prof. Haewon Park from the Department of Mechanical Engineering >   < Prof. Hyunchul Shim from the School of Electrical Engineering >   Professor Hyun Myung's research team, also in the School of Electrical Engineering, introduced autonomous navigation technology that uses cameras and LiDAR to recognize unfamiliar spaces and determine the robot's own position and path. Professor Hyeon-Min Bae's research team in the same school presented technology that combines motors, sensors, and control systems to move robots with precision. Professor Kyoungchul Kong's research team in the Department of Mechanical Engineering introduced wearable robot technology that identifies the user's intended movements and assists walking. Together, the work of these five research teams shows how the technologies robots need to perceive their surroundings, generate force and movement, and carry out real tasks connect with one another. < Prof. Hyun Myung from the School of Electrical Engineering >   < Prof. Hyeon-Min Bae from the School of Electrical Engineering >   < Prof. Kyoungchul Kong from the Department of Mechanical Engineering >   Connecting technologies developed in the laboratory to industry is another major pillar of the vision. KAIST faculty startups including Angel Robotics and Raion Robotics are working to develop related technologies into platforms and apply them in real-world environments. President Bae outlined a strategy that connects the university’s foundational technology research with product development by faculty startups so that physical AI platform technologies can be utilized across a wide range of settings. KAIST is also continuing collaborative research on physical AI that understands human movement. Professors Jung Kim and Kyoungchul Kong serve as co-directors of the Human Physical AI Research Center, which is pursuing related research through collaboration with the NVIDIA AI Technology Center (NVAITC). Alongside technology development, KAIST also plans to build an ecosystem linking talent development, entrepreneurship, and industrial expansion. It will support education and laboratory-based startups for students and young researchers while expanding retraining programs for personnel working in industry. The goal is to ensure that research outcomes do not remain limited to individual experiments but lead to corporate technology development and real-world applications. KAIST plans to continue hosting “Physical AI KOREA” to share progress in technology development and remaining challenges and to develop the event into a platform for collaboration between researchers and companies. The event is open to all members of the public who are interested. Those who wish to attend should register in advance through the online registration page ( https://docs.google.com/forms/d/e/1FAIpQLSdwirZfcNDBxEZogWqpDVgfBKslVRwHuQ-s-2fWtCBZYOTuUQ/viewform). Registration is open until 12 p.m. (KST) on October 12. Because seating is limited and will be allocated on a first-come, first-served basis, registration may close earlier.

KAIST Reveals How Information Is Written in Ferroe..

  < The research team. From left: Dr. Batzorig Buyantogtokh, Professor Seungbum Hong, and Ph.D. candidate Sheung Hun Kim (KAIST). >   What happens at the nanoscale when information is written to memory? Researchers at KAIST have shown how tiny regions with a new polarization direction form while previously formed regions continue to expand in a promising ferroelectric material. By linking these nanoscale changes to electrical measurements, the team developed a model that captures both processes, offering a basis for designing faster and more stable memory. KAIST (President Choongsik Bae) announced on October 8 that a team led by Professor Seungbum Hong from the Department of Materials Science and Engineering had identified how information is recorded in hafnium zirconium oxide (HZO), a promising material for next-generation memory. The study was conducted in collaboration with Professor Byung Jin Cho’s team at KAIST’s School of Electrical Engineering and researchers at NaMLab/TU Dresden in Germany. The research team focused on ferroelectrics. Ferroelectric materials can retain their electrical polarization after an applied voltage is removed. Reversing this orientation allows the material to store information as 0s and 1s. The electrical state in which positive and negative charges inside a material are aligned in a particular direction is called polarization. HZO combines hafnium oxide, a material widely used in semiconductor manufacturing, with zirconium oxide. Its compatibility with existing semiconductor manufacturing processes makes it attractive for nonvolatile memory, which retains data without power. When a voltage is applied to a ferroelectric, tiny regions called domains, each with a uniform polarization direction, begin to change. As new domains form and existing domains expand into the surrounding area, the overall polarization direction reverses and information is recorded. Researchers have traditionally described this switching behavior using models that emphasize either nucleation, the formation of new domains, or growth, the expansion of existing domains. In hafnia-based thin films such as HZO, however, switching occurs across many nanometer-sized crystal grains. Nucleation and growth interact with grain boundaries and other structural features, making it difficult for models that emphasize either process alone to fully capture the observed behavior. Using high-resolution piezoresponse force microscopy (PFM), the researchers directly observed how nanometer-scale domains (one nanometer is one billionth of a meter) change as the applied voltage increases. They then compared these observations with the electrical switching behavior of the full devices. < Ferroelectric domains spread differently depending on the bottom electrode. >   The combined results showed that new domains continued to form while existing domains expanded. In other words, information is recorded through two simultaneous processes: new changes begin, and changes already under way spread. To describe this behavior, the team introduced a simultaneous nucleation and growth (SNG) model. The model accounts for both the continuing formation of new domains and the cumulative expansion of those already present, bringing the two processes together in a single framework. The team also linked the nanoscale changes seen under the microscope directly to the electrical behavior of real memory devices. This makes it possible to explain not only how quickly the memory switches, but also where information writing begins and how it spreads. The findings can also be used to improve the performance of next-generation memory. Once researchers understand how domains form and spread, they can adjust a material's structure or manufacturing process so that information is written more quickly and uniformly. This understanding could help researchers improve memory performance by adjusting material structures and fabrication conditions to make switching faster and more uniform. The findings could also guide the development of more reliable, energy-efficient nonvolatile memory and neuromorphic devices for AI. Neuromorphic devices mimic aspects of the brain’s information processing and storage, offering potential applications in low-power AI hardware. < Pulsed switching measurements show that the switching speed and uniformity of the ferroelectric depend on the electrode material. >   “When information is written in a ferroelectric material, which retains its electrical state even without power, small changes begin at multiple sites while those already under way spread into the surrounding regions,” said Professor Hong. “This study reveals how these two processes work together to write information, providing a new basis for designing faster and more reliable next-generation memory.” Batzorig Buyantogtokh, a postdoctoral researcher in KAIST’s Department of Materials Science and Engineering, is the study’s first author. The paper was published in Nano Letters on June 15, 2026, and was selected for a journal cover. Paper: “Nucleation-to-Propagation Switching Modes in Ferroelectric Hf₀.₅Zr₀.₅O₂ Capacitors” DOI: https://doi.org/10.1021/acs.nanolett.6c01580 This research was supported by the National Research Foundation of Korea (NRF), with funding from the Ministry of Science and ICT (MSIT), under grant numbers RS-2026-25487626, RS-2026-25468150, and RS-2023-00272257. < The cover of Nano Letters (Volume 26, Issue 33, August 26, 2026) featuring the research. >

KAIST Develops Vector Database Technology for Reli..

  < The research team. From left: Professor Min-Soo Kim and Darae Lee, a master >   When new information is continually added and old information is removed, an AI system may fail to find the correct material even when that material exists. KAIST researchers have developed a vector database technology that preserves the search paths leading to the needed information as data changes. This allows AI to find up-to-date information more accurately and quickly. KAIST (President Choongsik Bae) announced on October 5 that a research team led by Professor Min-Soo Kim from the School of Computing has developed CONDA (Connectivity-Aware Dynamic Index). CONDA is a dynamic vector database indexing technology that maintains high search accuracy even as data is continuously added and deleted. Generative AI models do not inherently have access to information that emerges after training. Retrieval-augmented generation (RAG) has been widely adopted to address this limitation. RAG searches sources such as recent news or internal company documents and uses the results in its answers. For RAG to work properly, the system needs a search technology that can quickly and accurately find the information relevant to a question among vast amounts of data. Many AI services today convert the meaning of documents or images into vectors, which are numerical representations that computers can process. They then link pieces of information with similar meanings and follow these links to find the material they need. The challenge is that data in real companies and on the internet changes constantly. New documents, news, and product information keep arriving, while older information is modified or deleted. As these changes repeat, the links built between data points can gradually break, a problem the researchers call "connectivity collapse." The needed information is still in the database, but the AI can no longer reach it, and search accuracy drops. In other words, the information itself has not been lost. The search path that leads to it has been broken. CONDA considers both distances between data points and whether the search paths needed to reach the information remain connected. When new data is added or existing data is deleted, CONDA preserves the links that are important for search. This prevents specific pieces of information from becoming isolated in the search graph. < Figure 1. The data insertion and deletion processes in CONDA. >   In experiments where data was constantly added and deleted, CONDA improved search accuracy (recall) by up to 24.5％ compared with state-of-the-art methods. It also increased data update throughput by up to 1.90 times. The team also ran searches and data updates at the same time for six hours on a large-scale dataset containing 100 million vectors. Throughout the test, CONDA recorded the lowest search latency and the highest search accuracy among the compared methods. This result suggests that the technology can be applied to large-scale AI services. The technology can be used in knowledge search systems where information such as internal company regulations and work documents changes frequently. It can also be used in a range of AI services where data changes continuously, including news search, product search, and recommendation systems. In particular, it is expected to serve as a foundational technology for RAG, in which generative AI searches external information and uses it in its answers. CONDA allows this up-to-date information to be retrieved reliably over extended periods. The research results are also being applied to a commercial product. CONDA is being integrated into AkasicDB, the database product of GraphAI, an AI data infrastructure company founded by Professor Kim. It is scheduled for commercial release in the fourth quarter of this year. "The key value of retrieval-augmented generation (RAG) is that it allows large language models (LLMs) to find and use the latest information they were not trained on at the moment it is needed," said Professor Kim. He added that the study is significant because it presents a data infrastructure that lets AI use up-to-date knowledge accurately over long periods, even in real-world environments where data changes constantly. Darae Lee, who completed a master's degree at the KAIST School of Computing, is the first author of the study, and Professor Kim is the corresponding author. The findings were presented on September 2 at VLDB 2026 (the International Conference on Very Large Data Bases), a leading international conference in the database field. Paper title: CONDA: A Connectivity-Aware Dynamic Index for Approximate Nearest Neighbor Search over Evolving Data DOI: 10.14778/3836663.3836694 This research was supported by the Mid-Career Researcher Program of the National Research Foundation of Korea (NRF) and the SW Star Lab program of the Institute of Information & Communications Technology Planning & Evaluation (IITP), both funded by the Ministry of Science and ICT (MSIT).  

KAIST Develops a Venus Flytrap-Inspired ＇Robot Han..

  < The research team. From left: Ph.D. candidates Hyeon Woo Yang, Gyeong Rok Lee, and Jin Han Kwon, and Professor Hong Chul Moon (corresponding author), all of the KAIST Department of Chemical and Biomolecular Engineering. Top, from left: Professor Yong Min Kim of Jeonbuk National University and Dr. Sungryong Kim of the KAIST Department of Chemical and Biomolecular Engineering. >   A Venus flytrap closes its leaves to trap prey when it senses a stimulus. Taking inspiration from this plant, KAIST researchers have combined the roles of a skin that senses approaching objects and a muscle that grasps them in a single soft material. When an electrically charged object comes near, ions inside the material move and produce an electrical signal. When ultraviolet (UV) light is applied, the material bends and grasps the object. Because it holds its deformed shape for more than 10 minutes after the light is turned off, the material shows potential for soft robotic hands that require fewer components and less wiring. KAIST (President Choongsik Bae) announced on September 30 that a research team led by Professor Hong Chul Moon from the Department of Chemical and Biomolecular Engineering has developed the ‘Ionograsper,’ an ionic soft robot that senses nearby objects and moves in response to light. An ionic material contains ions, small particles that carry a positive or negative electric charge, which allow the material to respond to electrical changes in its surroundings. < Figure 1. Design and operating principle of the low-power, bioinspired Ionograsper. >   Humans notice objects with their eyes or skin and grasp them by moving their muscles. Robots likewise need sensors to detect their surroundings and actuators to generate movement. However, when these devices are attached separately to a soft robot, which bends flexibly, the number of components and the amount of wiring increase, and the structure becomes more complex. The research team combined the roles of a robot’s skin and muscle in a single soft polymer material. Polymers are substances made of small molecules linked into long chains; plastics and rubber are common examples. The team combined azobenzene, a compound that changes shape when exposed to light, with a hygroscopic polymer that absorbs moisture from the air. This produced a network structure in which ions can move. When a charged object approaches, its electric field causes the positive and negative ions inside the material to redistribute. As a result, changes occur inside the material even without direct contact. The researchers used the resulting electrical signal to detect the approach and movement of objects without applying a separate sensing voltage. To make the material move, the researchers irradiate it with UV light. Azobenzene changes its molecular shape when it absorbs light, and moisture escapes from the irradiated side of the material. This causes the material to bend toward the light, which can be used to grasp an object. When the light is turned off, moisture re-enters the material through nanoscale pores that formed on its surface during irradiation. The material then bends in the opposite direction and holds its deformed shape for more than 10 minutes, until the polymer structure slowly relaxes. This means the light does not need to stay on for the material to maintain its grasping position. < Figure 2. Sensing performance and operation of the bioinspired Ionograsper. >   A defining feature of the study is that three functions, sensing an object, moving, and retaining a changed shape for a period of time, are realized in a single material. The researchers also analyzed how light-induced changes in the molecules and in moisture content lead to bidirectional bending and shape retention. This material design approach could help reduce the components and wiring required when sensors and actuators are mounted separately on soft robots. The work, however, remains at the research stage: the material currently detects charged objects and is actuated by UV light. The team plans to increase the sensing distance, actuation speed, durability under repeated use, and load-bearing capacity, and to make the material responsive to visible or near-infrared light. < Figure 3. How the Ionograsper, based on the Venus flytrap, works and how it could be used in the future. >   “Rather than attaching a robot’s skin and muscles separately, we created a single material that senses objects and moves when it receives light, and a key feature is that it maintains its deformed shape even after the light is turned off,” said Professor Hong Chul Moon. He added that the team plans to combine the material with AI control technologies and develop it into a sensing and actuation material for physical AI robots. Professor Yong Min Kim of the School of Chemical Engineering at Jeonbuk National University served as the first author, and Professor Hong Chul Moon of KAIST’s Department of Chemical and Biomolecular Engineering served as the corresponding author. The study was published online in the international journal Advanced Materials on September 14, 2026, and was selected as the front cover article of Issue 62. Paper title: A Multifunctional Ionograsper Enabling Ion-Redistribution Proximity Sensing and Structural-Reconfiguration-Driven Bidirectional Actuation, DOI: 10.1002/adma.74981 Authors: Yong Min Kim (Jeonbuk National University, first author); Jin Han Kwon, Hyeon Woo Yang, Sungryong Kim, and Gyeong Rok Lee (KAIST, co-authors); Hong Chul Moon (KAIST, corresponding author) Related Video: https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.74981 This research was supported by a National Research Foundation of Korea (NRF) grant funded by the Korean government (Ministry of Science and ICT, MSIT) (RS-2025-00554260) and by the Nano & Material Technology Development Program through the NRF, funded by the MSIT (RS-2025-02221332).

KAIST’s World-First Full-Course Robot Marathon Com..

  < RAIBO2 and the research team at the finish line of the marathon >   A quadruped robot ran 42.195 km, a distance that is not easy even for humans to complete, without replacing its battery. KAIST researchers have developed this world-first achievement into design principles for long-distance operation, published the results in Nature, and are now moving toward commercializing robots that can be used in real industrial environments. KAIST (President Choongsik Bae) announced on the September 24 that the long-distance running technology of RAIBO2, a quadruped robot developed by a research team led by Professor Jemin Hwangbo from the Department of Mechanical Engineering, and the results of its actual marathon completion were published in the international journal Nature. This is the first time that robotics research conducted in Korea has been published in the main journal of Nature. Even globally, it is extremely rare for robotics research to be published in Nature, making this achievement an important demonstration of both the technological capability and academic level of Korean robotics research. RAIBO2 completed the full 42.195 km course of the Sangju Dried Persimmon Marathon held in November 2024 in 4 hours, 19 minutes, and 52 seconds without charging or replacing its battery. It was the world’s first case of a quadruped robot completing the full course of an actual marathon event. The core significance of this paper does not lie simply in the record of a robot running 42.195 km. In the global robotics industry, competition has recently intensified over the running speeds and obstacle-negotiation capabilities of quadruped robots and humanoids. However, to operate robots for long periods in real industrial environments, it is necessary to secure not only momentary locomotion performance but also energy efficiency, thermal management, battery performance, and actuator stability. The research team approached long-distance operation not as a performance issue involving individual components such as batteries or motors, but as a “whole-robot system problem” in which mechanical structure, electrical systems, and locomotion control are closely interconnected. Quadruped robots lose energy in multiple areas – including actuators, electrical systems, and mechanical components – as they repeatedly move their legs. Improving the efficiency of only one component makes it difficult to achieve both fast and stable locomotion performance and long travel distances at the same time. The research team systematically analyzed energy losses throughout the entire robot and jointly designed the mechanical, electrical, actuation, and control systems so that energy consumption could be reduced while maintaining high locomotion performance. Through this approach, the researchers presented specific design principles for which factors must be considered together to enable quadruped robots to move stably for extended periods. < Figure 1. Energy efficiency comparison graphs for the quadruped robot RAIBO2 (left: comparison with animals and other locomotion platforms; right: comparison across quadruped robots) >   < Figure 2. Energy transfer path from RAIBO2 >   In particular, the research team went beyond short-distance laboratory running tests or maximum travel distances estimated through calculations and validated the technology on an actual full marathon course. During the marathon, the team recorded the robot’s internal voltage, current, temperature, and battery status in real time and synchronized the data with position, speed, and altitude information measured by GPS. RAIBO2 was officially registered as a participant, allowing formal split times to be recorded, while the researchers ran alongside the robot and filmed the entire journey, lasting more than four hours from the starting line to the finish. While the 42.195 km completion record demonstrated RAIBO2’s performance, the data accumulated throughout the process provided scientific evidence showing how that performance was made possible and how it was maintained in a real-world environment. The research team developed a single impressive demonstration into research results based on system design principles and actual measurement data that can be analyzed and revalidated. < Figure 3. Learning network for RAIBO2 >   The research team’s next challenge is to transfer the performance achieved with a single research robot into a product that can be used repeatedly in industrial environments. RAIBO2 is a research robot that was designed and built by hand by the researchers themselves. To develop it into a product that can be deployed in real industrial environments, the same performance must be achieved reliably across multiple robots while also reducing production costs. The product must also meet industrial reliability requirements, including water and dust resistance, operation across a wide temperature range, durability against shock and vibration, and electromagnetic interference and compatibility requirements (EMI/EMC). Raion Robotics, a KAIST faculty startup that originated from Professor Hwang’s laboratory, is responsible for connecting the research results to actual products. During the development of RAIBO2, Raion Robotics’ manufacturing technology also contributed to maintaining the high energy efficiency and locomotion performance achieved in the laboratory for more than four hours during the actual marathon. The research team and Raion Robotics developed not only the mechanical and electrical systems, software, and artificial intelligence but also the motor drivers that precisely control the robot’s motors, allowing them to optimize the entire system. Currently, Raion Robotics is using this systems-design capability to develop commercialization and mass-production technologies for quadruped robots that can operate for extended periods in real environments, going beyond research prototypes. The goal is to reproduce the world-class performance demonstrated by a single laboratory robot across multiple products and enable stable operation in a wide variety of environments. Professor Jemin Hwang said, “This study goes beyond demonstrating how far a quadruped robot can run. Using actual marathon data, we demonstrated how the entire robot must be designed to achieve both high locomotion performance and energy efficiency,” adding, “We will connect the world-class performance achieved in the laboratory to products that anyone can use reliably in industrial environments.” Dr. Choongin Lee and students Donghoon Youm and Jeongsoo Park of the KAIST Department of Mechanical Engineering participated in the study as co-first authors, and the research results were published in the international journal Nature on September 23 (local time). Paper title: A quadruped robot designed to complete a marathon on a single battery charge, DOI: 10.1038/s41586-026-11102-5 Related photos and videos: Comprehensive photo/video collection on Google Drive: https://drive.google.com/drive/folders/1ye82tTr2dCK2Bt7Ih0hSgWqkgqlVXk6X?usp=sharing Full video of the marathon completion on YouTube: https://www.youtube.com/watch?v=-HMFoa3g9nA The entire process of RAIBO2 completing the marathon can be viewed in the YouTube video.

KAIST Finds a Way to Redirect Cell Fate with Just ..

  < Research Team. From left, Ferio Brahmana (circled), an MS student in the KAIST Department of Bio and Brain Engineering (co-first author); Woojeong Lee, a PhD student (co-author); Professor Kwang-Hyun Cho; and Corbin Hopper, a PhD student (co-first author). >   Which genes need to be controlled to guide a stem cell toward a desired cell type, or to modulate an immune cell's inflammatory response? A Korean research team has developed a general-purpose computational framework that identifies the key targets needed to steer a cell toward a desired state with a single temporary intervention, without permanently altering its genes. KAIST (President Choongsik Bae) announced on September 23 that a research team led by Professor Kwang-Hyun Cho from the Department of Bio and Brain Engineering has developed NUDGE, a foundational technology that redirects cell fate in a desired direction through a minimal, temporary intervention by harnessing the gene regulatory dynamics already present within cells. Technologies that redirect cell fate have drawn attention as key tools for treating intractable diseases and advancing regenerative medicine, since they can be used to differentiate stem cells into cells with specific functions or to return diseased or aged cells to a near-normal state. Differentiation, for example, refers to the process by which a stem cell becomes a specific cell type with a distinct function, such as a cardiomyocyte or a neuron. < Figure 1. Temporary vs. permanent control frameworks for cell fate regulation >   However, existing methods for controlling cell fate have relied mainly on permanent intervention like keeping specific genes locked in an "on" or "off" state. While this approach can produce the desired cell state, it has a limitation. It can reduce plasticity, the flexibility that allows cells to adapt to future changes in their environment, and it can also give rise to abnormal cell states that do not exist in nature. Rather than forcing a cell into a fixed state, the research team devised a new strategy that lets the cell reach the desired state on its own through a single temporary stimulus. The approach nudges the gene regulatory dynamics already present inside the cell toward the target direction, much like briefly clearing a blocked channel and letting the water find its own course. NUDGE (natural dynamics control of gene regulatory networks), the technology the team developed, uses a computational logic model to examine how key genes inside a cell influence one another and to determine which state the cell ultimately reaches as a result. It then mathematically decomposes the logical functions that describe these gene interactions to identify the minimal combination of control factors needed to convert a cell from its current state to the desired one. < Figure 2. Stages of applying NUDGE >   For example, if the goal is to differentiate a stem cell into a cardiomyocyte, NUDGE functions as a "blueprint for minimal intervention" that identifies which molecular targets need to be temporarily activated or inhibited. A key strength of the framework is that its exact method is mathematically guaranteed to identify all minimal control combinations that lead to the desired phenotype within the model. The team also presented an efficient approximation method that extends the framework to large-scale molecular networks. NUDGE does more than identify which factors to stimulate. It can also analyze which pathways a stimulated cell follows to reach the target state and how stably that state is maintained once reached, making it possible to anticipate whether unexpected states might emerge during the transition. The researchers assessed permanent control in 63 published biological network models where fixing a single node was sufficient to achieve the target phenotype; 55 showed reduced plasticity or attractors absent from the uncontrolled network. Separately, NUDGE's approximation achieved an average intervention error below 0.01 in more than 90 percent of 552 control problems across 69 large biological networks. Error measured the fraction of simulated converged states with the undesired phenotype. Of the minimum-sized, error-free interventions pooled from the three methods for each problem, NUDGE found an average of 83 percent, compared with 59 percent for IBMFA and 53 percent for LDOI. To assess how well NUDGE explains real biological phenomena, the researchers applied it to three distinct processes: cardiomyocyte differentiation, fate determination in mast cells involved in allergic responses, and the conversion of macrophages, a type of immune cell, into an anti-inflammatory state. In cardiomyocyte differentiation, the framework reproduced the known pattern in which MESP1, a gene important for heart development and regeneration, is briefly active early in differentiation but inactive in mature cardiomyocytes. The model shows how a single temporary intervention involving MESP1 and another regulatory target can guide differentiation without keeping MESP1 continuously active. The team also identified GATA2 and GATA1 as key regulators of cell fate during mast cell formation. In the macrophage model, the team derived optimal strategies for a single temporary intervention, including one using IL-4, to induce anti-inflammatory M2 states while preserving the capacity to transition between inflammatory and anti-inflammatory phenotypes and among M2 subtypes. The significance of this technology lies in providing a general-purpose framework that identifies where a single temporary intervention can guide a cell to a desired state, provided that a stable state with the desired phenotype exists in the network's uncontrolled dynamics. It is expected to find broad application, from guiding normal stem cells to differentiate into specific cell types such as cardiomyocytes, to normalizing diseased or aged cells, regenerating damaged tissue, modulating immune responses, and designing cancer reversion therapies that return cancer cells to a near-normal state. < Figure 3. Performance comparison of NUDGE >   Professor Cho said, "NUDGE is not a technology limited to returning a specific cell to a healthy state, but a general-purpose cell-control technology that guides a cell from its current state to a desired state through a single minimal stimulus." He added that the technology is expected to serve as a foundational tool for designing new control strategies across stem cell differentiation, aging, autoimmune disease, regenerative medicine, and cancer reversion therapy. The study was conducted by Ferio Brahmana, who holds a master's degree, and Corbin Hopper, a PhD student, both in KAIST's Department of Bio and Brain Engineering, who contributed equally as co-first authors, with PhD student Woojeong Lee as a co-author. The findings were published online on September 14 in the Proceedings of the National Academy of Sciences of the United States of America (PNAS). Paper Title: Uncovering minimal control of cell fate by natural dynamics, DOI: 10.1073/pnas.2604777123 This work was supported by National Research Foundation of Korea grants funded by the Korea Government, the Ministry of Science and Information and Communication Technology (ICT) (RS-2023-NR077224, RS-2024-00405360, RS-2025-17652976, and 2021M3A9I4024447 [Bio & Medical Technology Development Program]). It was also supported by a grant of the Korea Dementia Research Project through the Korea Dementia Research Center, funded by the Ministry of Health & Welfare and the Ministry of Science and ICT, Republic of Korea (RS-2025-02264017) and by the Korea Advanced Institute of Science and Technology (KAIST) Quantum＋X Convergence R&D Project and KAIST Grand Challenge 30 Project.

KAIST Extends Lifespan of Graphite-Free Anode-Free..

  < (top row, from left) Eunji Kim, PhD student (KAIST); Hyunju Jung, PhD student (KAIST); Jinuk Kim, PhD student (KAIST). (bottom row, from left) Jinwoo Lee, Professor (KAIST); Hee-Tae Jung, Professor (KAIST); Yonghee Lee, Professor (Kyungpook National University). >   Korean researchers have developed a technology that could make electric-vehicle (EV) batteries smaller and lighter while helping EVs travel farther on a single charge. A KAIST research team has proposed a new solution to a persistent problem in anode-free batteries, which reduce weight and volume by eliminating anode materials such as graphite that normally store lithium during charging. In these batteries, lithium can grow into sharp, spike-like structures that shorten battery life. The key lies in using nanofabrication techniques employed in semiconductor manufacturing to create uniform sites for lithium deposition and promote the formation of a robust protective layer. KAIST (President Choongsik Bae) announced on September 22 that a joint research team led by Professors Jinwoo Lee and Hee-Tae Jung from the Department of Chemical and Biomolecular Engineering, working with researchers from Kyungpook National University (President Heo Young-woo) and the National NanoFab Center (President Park Heung Soo), has developed a new technology that extends the lifespan of anode-free batteries by applying the ultrafine fabrication processes used in semiconductor device manufacturing to batteries. Conventional lithium-ion batteries store lithium in an anode material such as graphite during charging. Anode-free batteries, by contrast, deposit lithium directly onto a thin copper foil without a separate anode material. Because they eliminate the weight and volume that the anode material would otherwise occupy, they are drawing attention as a next-generation technology capable of packing more energy into a battery of the same size. However, repeated charging and discharging cause lithium to deposit unevenly on the copper foil, growing into branch- or spike-like structures known as dendrites. The protective layer over the lithium also becomes unstable, causing the battery’s performance and lifespan to deteriorate rapidly. This is one of the most significant obstacles standing in the way of commercializing anode-free batteries. Some previous approaches to extending the lifespan of anode-free batteries involved adding excess lithium or applying a thick protective layer to the surface. These methods provide extra lithium to compensate for losses during use or coat the surface with a protective film. However, adding more material makes the battery heavier and thicker, undercutting the very advantage of a small, lightweight anode-free battery. To minimize added weight and volume, the research team modified the copper foil on which lithium is deposited. Much like marking out closely spaced parking bays in a flat lot, the team created an array of microscopic tube-shaped structures on the copper surface so that lithium spreads out and deposits evenly rather than concentrating in one spot. To do this, the team used a process called secondary sputtering lithography (SSL), used for the precise micromachining of semiconductors. Fabricating microscopic tubes about 300 nanometers in diameter and 150 nanometers in height on the copper foil increased the surface area available for lithium deposition to about four times that of flat copper foil, helping to reduce localized, spike-like lithium growth. The team then coated the copper foil with a roughly 10-nanometer-thick layer of MXene, a two-dimensional material, forming an ultrathin film only a few thousandths the thickness of a human hair. Rather than serving as a finished protective layer itself, the MXene acts like a primer that draws together the components needed to form one. In the LiPF6-based electrolyte used in the study, the MXene surface promotes the formation of a robust protective layer rich in lithium fluoride (LiF) as the battery operates. This protective layer reduces unwanted reactions between the lithium and the electrolyte and also suppresses the spike-like growth of lithium. By examining electrode surfaces and cross sections while minimizing air exposure, the research team confirmed that a nanoscale protective layer forms uniformly along the MXene and identified the mechanism behind its formation. The analysis employed X-ray photoelectron spectroscopy (XPS), time-of-flight secondary ion mass spectrometry (ToF-SIMS), and transmission electron microscopy (TEM), among other techniques. < Concept illustration (AI-generated) >   Professor Jinwoo Lee said, “This study shows how ultrafine fabrication techniques used in semiconductor manufacturing can create both uniform sites for lithium deposition and a stable protective layer without changing the bulk electrolyte formulation or adding excess lithium.” He added that the technology is expected to serve as a foundation for accelerating the commercialization of high-energy anode-free batteries by extending battery life while minimizing increases in weight and volume. KAIST researchers Eunji Kim, Hyunju Jung, and Jinuk Kim participated as co-first authors on the study. The findings were published online in the international journal Advanced Functional Materials on September 1. Paper title: Anion-Interactive Anode Interfaces for Stable Anode-Free Lithium Metal Batteries, DOI 10.1002/adfm.77931 Author information: Eunji Kim (KAIST/National NanoFab Center, co-first author), Hyunju Jung (KAIST, co-first author), Jinuk Kim (KAIST, co-first author), Hee-Tae Jung (KAIST, corresponding author), Yonghee Lee (Kyungpook National University, corresponding author), Jinwoo Lee (KAIST, corresponding author) This research was supported by the Nano&Material Technology Development Program through the National Research Foundation of Korea (NRF) funded by Ministry of Science and ICT (RS-2026-25542167), as well as the Global Semiconductor Advanced Fab Utilization Project through the National Nano Fab Center (NNFC). This work was also supported by Semiconductor-Secondary Battery Interfacing Platform Technology Development Project of NNFC.

KAIST Develops Smartphone AI That Uses Past Soluti..

  < The research team. From left: Dr. Youngjun Lee, Postdoctoral Researcher (KAIST Institute of Information Electronics); Professor Jae-Gil Lee (School of Computing). Inset photos, from left: Dr. Doyoung Kim (Amazon); Dr. Junhyeok Kang (LG AI Research); Professor Hwanjun Song (Department of Industrial and Systems Engineering, KAIST). >   A new AI technology has been developed that retains and reuses knowledge, much like writing down a solution in a notebook and applying it to similar problems rather than asking an expert for help each time. Researchers at KAIST have developed a way for small AI models running on smartphones to store and reuse knowledge from a large server model. The approach reduced server calls by an average of 55.61％ compared with a non-cumulative approach while maintaining high accuracy, suggesting that mobile AI could make faster decisions with less server support as it encounters similar problems. KAIST (President Choongsik Bae) announced on September 21 that a research team led by Professor Jae-Gil Lee from the School of Computing has developed CURE (Cumulative Knowledge Reuse), a technology that enables a small AI model on a device to work efficiently with a large model on a server. Smartphones have limited processing power and memory, so they typically use small, lightweight AI models. These models can handle simple tasks quickly but may be less accurate when identifying complex or unfamiliar images. Sending every input to a powerful server model can improve accuracy, but transferring data and waiting for a response takes time. It also adds to network traffic and the server’s computational workload, increasing delays for users and operating costs for service providers. Researchers have explored a collaborative approach in which the on-device model assesses each input first and sends only difficult cases to the server. Without a way to retain the server’s knowledge, however, the device uses each answer once and may need the same help when a similar input appears. This is much like a student who fails to write down the solution to a difficult problem and has to ask for help again when faced with another of the same kind. The team set out to turn these one-time answers into knowledge the device could continue to use. CURE first checks whether the on-device model can handle an input reliably. If it cannot, the system consults knowledge previously obtained from the server and stored on the device. It contacts the server only if that knowledge is also insufficient. The process follows three steps, moving from solving a problem independently to consulting previous lessons and, when necessary, asking an expert. For example, an on-device model that cannot identify a car model in a photo may turn to the server for help. CURE uses the server’s prediction and the image’s features to update the device’s knowledge store. When the device later receives a photo of a similar car model, it can use that knowledge to identify it locally. < Figure 1. Comparison between a non-cumulative hybrid approach and CURE. >   CURE does not keep a collection of the original photos processed by the server. Instead, it stores a summary of their shared features and differences. This is like noting a car’s distinguishing features, such as its body shape or headlight design, rather than memorizing the entire photo. The stored knowledge can therefore help the system recognize not only images it has already seen but also similar images it encounters for the first time. The team tested CURE using vision-language models, which connect images with text to understand visual information. These models link what they see to language, much as people do. The researchers used MobileCLIP2 on the device and EVA-CLIP, which has 18 billion parameters, on the server. Parameters are numerical values within a model that encode information learned during training. They help the model distinguish objects and identify relationships between them. In tests on a range of image classification datasets, CURE achieved accuracy close to that of an approach that sends every input to the large server model. It also made an average of 55.61 percent fewer server calls than a device-server collaboration baseline that does not retain knowledge from previous server responses. In end-to-end tests that included communication time, CURE ran up to 2.80 times as fast as the non-cumulative device-server collaboration baseline and up to 3.67 times as fast as the approach that sends every input to the server. Making fewer server calls reduced the time spent transferring data and waiting for responses. CURE requires no retraining of either the on-device model or the server model. It leaves both models unchanged and adds a separate store for knowledge obtained from the server, making it adaptable to a range of AI models and services. The technology could reduce repeated data transfers and server computation image recognition. This could mean shorter waits for users and lower operating costs for service providers. Potential applications also include robots and wearables that need to recognize their surroundings with limited computing resources. Robots operating over slow or unreliable networks, for example, could use previously acquired knowledge to make more decisions locally without waiting for a server response. < Schematic illustration of the study (AI-generated) >   The benefits in practice will depend on the device’s processing power and storage capacity, network conditions, and the characteristics of the input data. Further testing across different devices and environments is therefore needed. “CURE allows a small on-device AI model to remember and reuse knowledge it has already obtained, rather than repeatedly asking the server the same question,” said Professor Jae-Gil Lee. “By maintaining high accuracy while reducing communication demands and response times, we expect it to help smartphones, robots, and wearables use powerful AI models more efficiently.” Dr. Youngjun Lee, a postdoctoral researcher from the KAIST Institute of Information Electronics, was the study’s first author, and Professor Jae-Gil Lee from the School of Computing was the corresponding author. Co-authors were Doyoung Kim from Amazon, Junhyeok Kang from LG AI Research, and Professor Hwanjun Song from the KAIST Department of Industrial and Systems Engineering. The findings were presented on September 10 at the European Conference on Computer Vision (ECCV 2026), a leading international computer vision conference held in Malmö, Sweden, from September 8 to 12. Paper title “CURE: Cumulative Knowledge Reuse for Efficient Device-Server Hybrid Inference in Vision-Language Models”, DOI: https://doi.org/10.1007/978-3-032-37627-5_21 Author information — Dr. Youngjun Lee (postdoctoral researcher, KAIST Institute of Information Electronics, first author), Doyoung Kim (KAIST graduate, now at Amazon, co-author), Junhyeok Kang (KAIST graduate, now at LG AI Research, co-author), Professor Hwanjun Song (KAIST Department of Industrial and Systems Engineering, co-author), Professor Jae-Gil Lee (KAIST School of Computing, corresponding author) This work was supported by Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No. RS-2020-II200862, DB4DL: High-Usability and Performance In-Memory Distributed DBMS for Deep Learning, 50％ and No. RS-2022-II220157, Robust, Fair, Extensible Data-Centric Continual Learning, 40％), the InnoCORE pro-gram of the Ministry of Science and ICT (AI Meta-Scientist, N10260110, 10％), and Samsung Electronics Co., Ltd.(IO251216-14634-01).  

KAIST Technology Behind Toothbrushes and Sportswea..

  < The research team. From left: Professor Sang Ouk Kim, PhD candidate Jin Hyo Kim, and PhD candidate Sujin Cha (KAIST Department of Materials Science and Engineering) >   KAIST’s discovery of graphene oxide liquid crystals in 2011 has laid the foundation for 15 years of research worldwide, leading to technologies for producing graphene fibers with high strength and thermal conductivity. The fundamental materials research that helped bring antibacterial toothbrushes and functional sportswear to market is now advancing into materials for thermal management. < Figure 1. Principle of viscoelasticity-based superdrawing of graphene oxide liquid crystals dispersed in glycerol, and the resulting formation of highly aligned graphene fibers >   KAIST (President Choongsik Bae) announced on September 18 that Professor Sang Ouk Kim’s team in the Department of Materials Science and Engineering has published a commentary in the News & Views section of Nature Materials, highlighting recent advances in high-performance graphene fibers based on graphene oxide liquid crystals and their scientific significance. Graphene oxide consists of graphene, a single layer of carbon atoms arranged in a honeycomb structure, with oxygen functional groups attached. Graphene is strong and conducts heat and electricity well, but its poor dispersibility in common solvent makes it difficult to process in liquid form. By contrast, the oxygen functional groups allow graphene oxide to disperse readily in water and other solvents, making it easier to formulate into inks, apply as coatings, or spin into fibers. In 2011, Professor Kim’s team was the first in the world to report that graphene oxide dispersed in water above a certain concentration spontaneously forms a liquid-crystalline state in which its thin sheets align in a common direction. Above critical concentration conditions, graphene oxide sheets randomly dispersed in water align like a scattered deck of cards arranged to face the same direction. This alignment allows graphene oxide to be drawn into long fibers while maintaining the sheets’ orientation along the fiber axis. Following this discovery, researchers worldwide have developed methods for producing graphene fibers from graphene oxide liquid crystals. Conventional methods, however, have faced problems such as breaks in the liquid filament during drawing and insufficiently aligned or loosely packed graphene sheets, which leave voids and defects within the fibers. These problems have limited efforts to improve strength and thermal conductivity simultaneously. In their commentary, Professor Kim’s team discussed recent research addressing these limitations and analyzed how research on graphene oxide liquid crystals could lead to technologies for manufacturing high-performance fibers. Researchers at Zhejiang University in China recently dispersed graphene oxide in highly viscous glycerol, giving the dispersion viscoelastic properties similar to those of a polymer solution. This enabled “ultrahigh-ratio drawing” during wet spinning, allowing the graphene oxide dispersion to be stretched much further than before. During this process, the graphene oxide sheets become more closely aligned along the fiber axis, reducing internal voids and defects. Subsequent heat treatment at high temperatures promotes the growth of large, aligned graphitic crystallites, producing lightweight, strong graphene fibers with high thermal and electrical conductivity. < Figure 2. Professor Sang Ouk Kim holding a graphene fiber >   The graphene fibers produced in this recent study were reported to have a tensile strength of up to 5.9 gigapascals (GPa) and a thermal conductivity of up to 1,720 watts per meter per kelvin (W/(m·K)). In simple terms, the fibers resist breaking when pulled and transfer heat quickly. These results show how denser packing and better alignment of graphene oxide sheets can improve both strength and thermal conductivity. In the commentary, Professor Kim’s team highlighted how research on graphene oxide liquid crystals has progressed from observing a fundamental phenomenon to processing a two-dimensional material through flow and stretching, much like a polymer solution. < Figure 3. Commercialized graphene toothbrush >   KAIST's core technology related to graphene oxide liquid crystals has already moved beyond the laboratory into consumer products. Antibacterial graphene toothbrushes incorporating technology from Materials CreationCo., Ltd., a faculty startup founded by Professor Kim, have sold more than 14 million units since their 2023 launch. Functional fibers made with GrapheneTex material were used in the uniforms of the Taekwondo demonstration team at the 2024 Paris Olympics. More recently, the material has been applied to functional golf apparel, other sportswear, and bedding, as the company pursues expansion into global markets. “Since its discovery in 2011, graphene oxide liquid crystal has provided a foundation for research over worldwide that has now led to strong, high-performance graphene fibers that conduct heat efficiently,” said Professor Kim. “What makes this significant is that a single line of basic research has led to applications ranging from consumer products such as toothbrushes and functional clothing to advanced materials for electronics, mobility, and aerospace.” < Figure4. Research concept concept illustration (AI-generated) >   The commentary was written by PhD candidates Jin Hyo Kim and Sujin Cha and Professor Sang Ouk Kim, all of KAIST’s Department of Materials Science and Engineering, and was published in the News & Views section of Nature Materials on September 8. (END) Commentary title: Processing graphene fibres by viscous flow DOI: 10.1038/s41563-026-02735-y