%45%6c%65%63%74%72%69%63%61%6c%20%45%6e%67%69%6e%65%65%72%69%6e%67
-
KAIST Develops ‘Chameleon AI Semiconductor’ with Programmable Response Speeds
AI semiconductors are becoming more programmable. KAIST researchers have developed a device whose response characteristics can be programmed to process data changing at different speeds. The technology reduced prediction errors for time-varying data by up to 40-fold and is expected to enhance real-time AI performance in autonomous vehicles, robots, and wearable devices.
KAIST (President Choongsik Bae) announced on August 7 that a research team led by Chair Professor Shinhyun Choi from the School of Electrical Engineering and the Graduate School of Semiconductor Technology has developed a programmable dynamic memtransistor (PDM), a semiconductor device whose time-response characteristics can be adjusted to multiple states and retained, as well as an integrated array based on the device.
A memtransistor is a next-generation semiconductor device that combines the information-storage function of memory with the computing function of a transistor. In the developed PDM, the ability to process data while retaining previous information allows its response characteristics to be adjusted and retained for incoming data.
Today’s computers and smartphones require complex software processing to analyze data that changes over time, resulting in large computational loads and high power consumption. To address this, researchers have been studying technologies that allow semiconductor hardware itself to process data directly. However, conventional devices have had fixed response speeds that cannot be changed once the device is fabricated.
The research team overcame this limitation by introducing a dual-layer structure inside the transistor, combining a charge storage layer that accumulates and processes data with an electron trapping layer that controls the response speed in a nonvolatile manner.
In the PDM developed by the research team, incoming data is processed in the charge storage layer, while the electron trapping layer controls, across multiple levels, the recovery speed at which the semiconductor returns to its original state. In experiments, the team succeeded in tuning the current recovery time over an approximately 5-fold range and the characteristic frequency over a range of more than 10-fold.
In particular, in experiments involving the prediction of data in which fast and slow changes are intricately mixed, the PDM reduced prediction errors by as much as 40 times compared with conventional fixed-response semiconductor devices. The PDM enables accurate information processing even when handwriting or object-movement speeds vary, by using response characteristics configured to match different input timescales. Once the response characteristics are set, the device remembers them without requiring a continuous external power supply, and it does not require complex preprocessing of input data. Because it is fully compatible with materials used in widely adopted commercial semiconductor processes, it is also highly advantageous for mass production and commercialization.
The research team fabricated a PDM array and used it to predict complex data, confirming that it achieved accuracy comparable to conventional software-based systems while consuming far less energy.
“This study demonstrates an AI semiconductor whose response characteristics can be programmed to efficiently process data changing at different speeds,” said Chair Professor Choi. “We expect it to become a core technology that improves the performance of AI devices such as autonomous vehicles, robots, and wearables while reducing their power consumption.”
This research was led by KAIST Graduate School of Semiconductor Technology Ph.D. candidate Dae-won Kim as the first author, with Yoonho Cho, Seokho Seo, Yujin Kim, See-On Park, Taehwan Jang, and Chaebin Park participating as co-authors. Young Taek Oh and Fellow Jae-Duk Lee of Samsung Electronics’ Semiconductor R&D Center also participated as co-authors, and Chair Professor Shinhyun Choi served as the corresponding author. The research was published in July in the internationally renowned journal Nature Communications on July 4.
Paper title: Programmable memtransistor array with temporal dynamics modulation for efficient time-series data processing,
DOI: https://doi.org/10.1038/s41467-026-75211-5
This research was supported by the National R&D Program through the National Research Foundation of Korea funded by the Ministry of Science and ICT, the ETRI R&D Support Program of the Institute of Information & Communications Technology Planning & Evaluation, the HRD Program for Industrial Innovation of the Korea Institute for Advancement of Technology funded by the Ministry of Trade, Industry and Energy, Samsung Electronics, and others.
2026.08.07 View 914 -
KAIST Develops AI That Avoids Hallucinating Even at Night or in Smoke
Multimodal large language models (MLLMs), which process multiple types of sensory information such as text, images, and audio at the same time, are rapidly expanding the range of applications for artificial intelligence (AI). However, in real-world environments, these models can misinterpret the physical characteristics of sensors, mistakenly identify objects, or claim to hear sounds that are not actually present simply because a certain object appears in a video. These errors are known as hallucinations. A KAIST research team has developed a new technology that corrects such information confusion and physical misperceptions in AI.
KAIST (President Choongsik Bae) announced on the 31st of July that a research team led by Professor Yong Man Ro from the School of Electrical Engineering has developed two core technologies that overcome the tendency of existing large language models to rely too heavily on ordinary camera (RGB) images and enable AI to suppress cross-modal hallucinations that occur when different sensory inputs become mixed.
The first technology developed by the research team is the Diverse Negative Attributes (DNA) optimization method, which helps AI accurately understand the physical characteristics of special camera sensors such as thermal, depth, and X-ray sensors. Existing AI models often failed to understand the physical meaning of such images, for example by mistaking bright areas in thermal images for simple light reflection.
The research team built VS-TDX, the first comprehensive benchmark for evaluating diverse vision sensors, and used the types of wrong answers that AI frequently produces as learning signals to help the model internalize the characteristics of each sensor. As a result, the AI gained a “new eye” that allows it to accurately infer the state of objects even in darkness or smoke.
The second technology is Modality-Adaptive Decoding (MAD), a control method that blocks hallucinations caused by confusion between visual and auditory information at the source. This technology prevents AI from mistakenly claiming that it hears a sound that does not actually exist simply because a certain object appears in a video.
MAD works by having the AI self-assess whether vision or audio is more important for a given task, and then increasing the weight of the more relevant modality in real time. A key advantage of this technology is that it can immediately suppress hallucination errors without costly model retraining, as it is training-free.
Instead of retraining AI models at large scale with massive computing resources, the research team maximized cost efficiency by introducing the DNA method, which enables fine adjustment with only a small amount of data, and the
MAD plug-in approach, which requires no additional training at all.
These technologies can be applied to autonomous vehicles operating at night or in bad weather, robots performing missions in smoke-filled environments, and unmanned aerial vehicles using thermal cameras. They are also expected to be useful in fields that process multiple types of sensor information together, such as airport X-ray security screening and medical image analysis.
Professor Yong Man Ro said, “This research is significant because it reduces AI’s sensory bias and misperceptions without large-scale retraining,” adding, “It will serve as a foundation for building multimodal AI that can be trusted in real-life and industrial settings.”
This achievement was notable for its continuity, with Sangyun Chung, a doctoral student in KAIST’s School of Electrical Engineering, participating as first author in both studies. Dr. Youngjun Yoo also participated as co-first author in the DNA study.
Among the related papers, the MAD study was presented in June at the Conference on Computer Vision and Pattern Recognition (CVPR), the world’s leading international conference in AI and computer vision. The DNA study was published in IEEE Transactions on Image Processing, a leading international journal in the field of image processing.
Paper title: Enhanced Vision-Language Models for Diverse Sensor Understanding: Cost-Efficient Optimization and Benchmarking,
DOI: 10.48550/arXiv.2412.20750 Author information: Sangyun Chung (KAIST, co-first author), Youngjun Yoo (KAIST, co-first author), Se Yeon Kim (KAIST, third author), Youngchae Chee (KAIST, fourth author), Yong Man Ro (KAIST, corresponding author)
Paper title: MAD: Modality-Adaptive Decoding for Mitigating Cross-Modal Hallucinations in Multimodal Large Language Models,
DOI: 10.48550/arXiv.2601.21181
Author information: Sangyun Chung (KAIST, first author), Se Yeon Kim (KAIST, second author), Youngchae Chee (KAIST, third author), Yong Man Ro (KAIST, corresponding author)
Related demo video: https://youtu.be/VuP9i6Vfk8o
This research was supported by the Institute of Information & Communications Technology Planning & Evaluation’s (IITP’s) Human-Centered AI Core Technology Development Program and by a Center for Applied Research in Artificial Intelligence (CARAI) grant funded by the Defense Acquisition Program Administration (DAPA) and the Agency for Defense Development (ADD).
2026.08.03 View 747 -
KAIST Develops AI That Finds Its Own Hidden Weaknesses, Paving the Way for Safer Generative AI Models
KAIST researchers have developed a safety verification technology that uncovers roughly seven times more hidden vulnerabilities in AI than existing methods. The technology is expected to serve as a foundation for developing safer, more trustworthy AI.
KAIST (President Choongsik Bae) announced on the 30th of July that a research team led by Professor Junmo Kim from the School of Electrical Engineering has developed a new framework called Stable-GFlowNet (S-GFN), which overcomes the limitations of red-teaming—a safety verification process that deliberately attacks large language models (LLMs) to expose hidden weaknesses.
Red-teaming for generative AI is the process of crafting attack prompts designed to probe an AI's vulnerabilities and induce the AI to produce harmful or dangerous responses before the program is deployed. Since discovering a wider variety of attack methods allows more vulnerabilities to be addressed in advance, both the success rate and diversity of attacks are critical.
Previous approaches primarily relied on reinforcement learning—an AI technique trained to maximize reward—to generate attack prompts. However, these methods frequently suffered from mode collapse—a phenomenon where the model repeatedly converges on a narrow set of high-reward attack prompts rather than generating diverse outputs, thereby limiting its ability to uncover various vulnerabilities.
Generative Flow Networks (GFlowNets)—an AI generation technique trained to produce diverse outputs in proportion to their reward—were proposed as a solution. Yet GFlowNet training is computationally complex and unstable, and noisy reward signals can assign high rewards even to meaningless sentences, often causing training to collapse.
To address these issues, the research team developed three core techniques that help the model learn effective attacks more reliably while filtering out flawed ones.
First, much like comparing several paths to choose the best one, the team introduced Contrastive Trajectory Balance (CTB), which reduces computational complexity and stabilizes training by directly comparing pairs of generated attack trajectories.
Second, akin to filtering out background noise to focus on a single voice, the team applied Noise Gradient Pruning (NGP) to eliminate minor reward fluctuations and ensure the model learns exclusively from meaningful signals.
Third, the team applied the Min-K Fluency Stabilizer (MKS), which guides the model to generate attack prompts resembling text that a real user would write—just as a human reader naturally prefers coherent sentences to gibberish.
As a result, Stable-GFlowNet discovered 134 unique attack types—about seven times more than the 17 unique attack types found by the existing GFlowNet-based technique—while maintaining a high attack success rate of 92%.
Defense models trained using attacks generated by Stable-GFlowNet also demonstrated strong generalization, effectively defending against a wide range of attacks in cross-attack tests, which evaluate performance using attack techniques different from those used during training.
The team further demonstrated that CTB and NGP achieve faster and more stable performance than existing methods—not only in AI safety verification, but also in other distribution-matching tasks such as molecular generation for drug discovery.
Professor Kim said, "This technology is significant in that it can reliably uncover a wide range of AI vulnerabilities even in realistic conditions with limited data and high noise." He added, "Because it allows a broader range of risk factors to be identified and defended against before generative AI is deployed in real-world services, we expect it to become a core foundational technology for developing safer, more trustworthy AI."
The study was led by first author Minchan Kwon, a Ph.D. candidate from the School of Electrical Engineering, and was selected as a Spotlight paper—placing it in the top 2.2% of submissions—at the International Conference on Machine Learning (ICML) 2026, one of the world's most prestigious AI conferences.
※ Paper title: Stable-GFlowNet: Toward Diverse and Robust LLM Red-Teaming via Contrastive Trajectory Balance
arXiv: https://arxiv.org/abs/2605.00553
This research was supported by the Institute of Information & Communications Technology Planning & Evaluation’s (IITP) SW Star Lab program, funded by the Ministry of Science and ICT.
2026.07.30 View 1015 -
KAIST Opens a New Era of Webtoons: From “Viewing” to “Experiencing”
Webtoons are coming to life in the physical world, ushering in a new era in which comics are not merely viewed, but experienced.
A KAIST research team has developed the world’s first next-generation extended reality (XR) comics platform that enables a wide range of readers to enjoy immersive, three-dimensional comics in physical space. By expanding webtoons beyond the screen and into the real world, the team has opened up new possibilities for the future of comics.
KAIST (President Choongsik Bae) announced on the 21st of July that a research team led by Professor Ian Oakley from the School of Electrical Engineering has proposed core design principles and future directions for next-generation extended reality (XR) comics through a systematic user study involving 15 participants, including human-computer interaction (HCI) experts, professional webtoon creators, and readers.
The research team developed ComiXR, a new platform that enables users to both read and create comics in XR environments. Participants used the platform to transform a conventional print comic into an XR comic and explored how different visual, auditory, haptic, and interactive features could be combined.
Comics, which originated in printed books and newspapers, have evolved dramatically with the rise of smartphones. The vertical-scrolling format of webtoons has become particularly successful by adapting comics to the interaction methods of mobile devices.
The research team viewed XR devices as a potential next stage in this evolution. To explore how spatial depth, three-dimensional rendering, spatial audio, eye tracking, and facial expression tracking could be incorporated into comics, the team built ComiXR using a Meta Quest Pro headset.
While wearing the headset, participants freely positioned 3D characters, speech bubbles, sound effects, and other comic elements throughout a physical room. They were able to construct comic environments that they found comfortable, engaging, and immersive.
The results showed that readers strongly preferred designs that actively used the depth of physical space over simply displaying flat comic pages in a virtual environment. Immersion increased significantly when characters were positioned at a different depth from the background and speech bubbles were separated into distinct layers. In particular, an eye-tracking feature that revealed the next line of dialogue only when the reader looked at a specific character proved effective in preventing spoilers.
The platform also demonstrated new sensory experiences that are not possible in conventional comics. Special effects could be triggered in response to readers’ facial expressions, while haptic feedback could convey sensations such as a character’s heartbeat or the impact represented by an onomatopoeic effect.
Based on the study, the research team also proposed four key design concepts for XR comics. The first, “The Panel Gallery,” transforms the walls of a room into a gallery for displaying comic panels. The second, “The Pop-Up,” presents comics like pop-up books on desks or walls. The third, “Around Comic,” places 3D characters and other comic elements in outdoor spaces. The fourth, “Inclusive ComiX,” improves accessibility for a wide range of readers.
The research team expects XR comics to complement, rather than replace, existing smartphone-based webtoons. They could be used for special exhibitions and educational content that allow audiences to experience fictional worlds more vividly, as well as platforms that improve access to cultural content for a wider range of users.
Ammar Al-Taie, a postdoctoral researcher at the KAIST Information and Electronics Research Institute, participated as the first author, while Hyunyoung Han, a doctoral student in the School of Electrical Engineering, participated as a co-author.
The research was presented at the ACM Designing Interactive Systems Conference 2026, or ACM DIS 2026, one of the leading international conferences in human-computer interaction and design. The ComiXR platform has also been released as open-source software for public use.
Paper title: ComiXR: Exploring Comic Layouts in eXtended Reality
DOI: https://doi.org/10.1145/3800645.3812857
Related Video: https://drive.google.com/drive/folders/1D9Efp3T0biDbUm1K5Gu6HLSSy89Uaq8R?usp=sharing
Open-source platform: https://github.com/ammarjamal/ComiXR
The research was supported by the KAIST Jang Young Sil Fel¬lowship Program (Excellence Track). The authors acknowledge support from the IITP (Institute of Information & Communications Technology Planning & Evaluation)-ITRC (Information Technology Research Center) grant funded by the Korean government (Ministry of Science and ICT) (IITP-2026-RS-2024-00436398).
2026.07.21 View 1438 -
KAIST Develops Key Technology to Make Personalized AI Safer
“Create an AI assistant trained only on our company’s documents.”
The era of building “personalized AI” by training AI models on individual or corporate documents and data is beginning. However, while such customization can improve task performance, it can also weaken the model’s existing safety safeguards. KAIST researchers have developed a core AI technology that preserves customized performance while further strengthening safety.
KAIST (President Choongsik Bae) announced on the 15th of July that a research team led by Professor Changick Kim from its School of Electrical Engineering has developed “Buffer-and-Reinforce,” a training framework for safe fine-tuning that prevents safety degradation when large language models (LLMs), such as
ChatGPT, are retrained on data from individuals or companies to better suit their needs.
Until now, one of the biggest challenges in the era of personalized AI has been that fine-tuning improves a model’s ability to perform new tasks, but can also weaken its existing safety rules. The research team focused on prior findings showing that, counterintuitively, fine-tuning an AI model while it is in a temporarily jailbroken state — a state in which it may respond even to dangerous requests it would normally refuse — does not significantly compromise its safety.
The team then devised a new approach in which this jailbroken state is not used in actual services, but is applied only temporarily during the fine-tuning process through a buffering module called “BufferLoRA,” which is removed after training.
The research team was the first to clarify why this phenomenon occurs. They found that, in the temporarily jailbroken state, the AI model becomes less easily influenced by harmful information, while still effectively learning the new task abilities desired by the user. In other words, the model can continue learning useful knowledge without additionally absorbing harmful behaviors.
Based on this insight, the team developed a two-stage learning method consisting of “buffering” and “safety reinforcement.”
First, the temporary buffering module, BufferLoRA, is applied to the AI model during user fine-tuning, where it acts as a protective layer that prevents harmful data from directly affecting the base model. Once fine-tuning is complete, this module is removed.
Next, a safety reinforcement module called “ReinforceLoRA” is applied to restore and strengthen the model’s safety. In this process, the team used QR decomposition, a mathematical technique that separates different types of information and selectively reflects only the necessary components. This allowed the model to retain the new functions learned from user data while selectively reinforcing safety.
Simply put, the researchers first placed a temporary protective layer, BufferLoRA, over the AI model so that harmful data could not directly affect it, while allowing the model to learn the necessary task. They then removed the protective layer and applied ReinforceLoRA to strengthen the model’s safety safeguards. As a result, the model maintained its customized performance while achieving even stronger safety.
In experiments, the AI model maintained high safety even in an extreme setting where all user data consisted of harmful questions and answers. After fine-tuning, the rate at which the AI generated harmful responses was about 8%, lower than the roughly 18% observed in the original model that had not been fine-tuned at all. The framework also achieved strong customized performance and state-of-the-art safety without requiring additional safety data during user fine-tuning or significantly increasing computational cost, suggesting its practical applicability to real-world personalized AI services.
Professor Changick Kim stated, “This research provides a key foundational technology that allows anyone to build customized AI with their own data while using it more safely,” adding, “We expect it to contribute significantly to building a trustworthy AI service environment in the era of personalized AI and AI agents.”
This research was led by Seokil Ham, a doctoral student in KAIST’s School of Electrical Engineering, as first author. The paper was selected as a Spotlight presentation at the International Conference on Machine Learning (ICML) 2026, one of the world’s most prestigious conferences in artificial intelligence, an honor given to only about the top 2.2% of all submitted papers, drawing international attention.
※ Paper title: Jailbreak to Protect: Buffering and Reinforcing via Temporary Jailbreaking for Safe Fine-Tuning in Large Language Models
DOI: 10.48550/arXiv.2605.24550
※ Author information: Seokil Ham (KAIST, first author), Jaehyuk Jang (KAIST, second author), Wonjun Lee (KAIST, third author), Changick Kim (KAIST, corresponding author)
※ Related video: https://drive.google.com/file/d/1gfok06dE8699qtiUR7gVsRoVmBGADaWQ/view?usp=sharing
This work was supported by Institute of Information & Communication Technology Planning & Evaluation (IITP) grant funded by Ministry of Science and ICT(MSIT) (No. RS-2025-02215344, Development of AI Technology with Robust and Flexible Resilience Against Risk Factors).
2026.07.15 View 1119 -
KAIST Automates the Search for “Dream Semiconductor” 2D Semiconductors
The era of researchers manually searching for two-dimensional semiconductors, which are drawing attention as next-generation AI semiconductors, is coming to an end. KAIST researchers have automated semiconductor screening and device fabrication, analyzed thousands of devices, and revealed the relationship between thickness and performance that had long been difficult to identify. This achievement is expected to shift next-generation semiconductor research toward a data-driven approach and accelerate the commercialization of AI semiconductors and ultra-low-power semiconductors.
KAIST (President Choongsik Bae) announced on the 9th that a research team led by Professor Jimin Kwon of the School of Electrical Engineering and the Department of AI System has developed a technology that automatically identifies two-dimensional semiconductors from optical microscope images alone and connects the process to transistor fabrication, through joint research with UNIST, Hanbat National University, Hanyang University, and Washington University in St. Louis in the United States.
Two-dimensional semiconductors are ultrathin semiconductors only a few atomic layers thick. They are called “dream semiconductors” because they can enable smaller semiconductors that consume less electricity than conventional silicon semiconductors. Today’s silicon semiconductors are approaching physical limits, as continued miniaturization of circuits leads to greater power loss and heat generation. Two-dimensional semiconductors, which are attracting attention as next-generation materials to overcome these limits, are expected to be used in a wide range of future technologies, including AI semiconductors, smartphones, data centers, wearable devices, foldable or stretchable electronics, and ultra-small medical sensors.
However, in two-dimensional semiconductors made through solution processing, the position, size, and thickness of each small semiconductor flake all differ, requiring researchers to find the desired samples one by one under a microscope. They then had to manually design electrodes according to the identified positions, requiring substantial time and effort, and making it practically difficult to analyze thousands or more devices at once.
The research team used molybdenum disulfide (MoS₂), a representative two-dimensional semiconductor material. By using the fact that the RGB red, green, and blue brightness values seen under a microscope change depending on thickness, the team enabled a computer to automatically identify the desired semiconductor and automatically design the electrodes. Verification using atomic force microscopy (AFM) confirmed that even subtle thickness differences of three to eight layers could be accurately distinguished.
Through this approach, the team successfully selected suitable samples automatically from more than 120,000 semiconductor flakes and fabricated and analyzed 1,615 transistors.
The large-scale analysis also produced meaningful results. The team statistically clarified for the first time that as the semiconductor becomes thicker, current flows more easily, but the ability to switch electricity on and off actually decreases. This characteristic had been difficult to confirm previously because only a small number of samples could be analyzed, but the team revealed it through large-scale data.
The greatest significance of this study is that it did not simply automate the fabrication process, but transformed two-dimensional semiconductor research, which had relied on human experience, into data-driven research. Going forward, the technology is expected to enable researchers to fabricate and analyze more semiconductors more quickly, identify high-performance materials, and ultimately expand into research in which AI designs new semiconductors.
This study was conducted with Professor Jimin Kwon, Dr. Haksoon Jung, and Dr. Yongwoo Lee of KAIST as co-corresponding authors, and Sanghyun Lee of UNIST as the first author. The research results were published on April 3 in Advanced Functional Materials, a leading international journal in materials science, and were also selected as an Inside Back Cover article in the field of 2D Materials & Electronics.
※ Paper title: Statistically Resolving Thickness-Dependent Electrical Characteristics in Multilayer-MoS₂ Transistors, DOI: 10.1002/adfm.202532204
※ Author information: Professor Jimin Kwon (KAIST, corresponding author), Dr. Haksoon Jung (KAIST, corresponding author), Dr. Yongwoo Lee (KAIST, corresponding author), Sanghyun Lee (UNIST, first author), and participating researchers from partner institutions: Sumin Hong (UNIST), Minho Park (UNIST), Professor Seongju Kim (Hanbat National University), Professor Sang-Hoon Baek (Hanyang University), Professor Joonki Suh (KAIST), Seonguk Yang (KAIST), Professor Sang-Hoon Bae (Washington University in St. Louis), and Dr. Chang-Soo Lee (TDS)
This research was supported by the Individual Basic Research Program of the National Research Foundation of Korea (NRF), funded by the Ministry of Science and ICT (MSIT), and by the Advanced Strategic Industry Super-Gap Technology Development Program of the Korea Planning & Evaluation Institute of Industrial Technology (KEIT), funded by the Ministry of Trade, Industry and Energy (MOTIE).
2026.07.09 View 1388 -
KAIST Develops Core Display Technology That Prevents Image Distortion Even When Stretched
Beyond bendable and foldable displays, the era of stretchable displays, whose screens can expand freely like rubber, is now emerging. KAIST researchers have developed a core technology that allows text, images, and other on-screen information to retain their original shape even when the screen is stretched by up to 15%. The achievement is expected to help solve the problem of image distortion and accelerate the commercialization of next-generation high-quality stretchable displays.
KAIST (President Choongsik Bae) announced on the July 8 that a research team led by Professor Seunghyup Yoo of the School of Electrical Engineering, in collaboration with Professor Hanul Moon’s team at Dong-A University (President Hae Woo Lee), has successfully implemented an auxetic-based stretchable display platform. Auxetic structures expand in both width and length when pulled, allowing the display to stretch uniformly at the same ratio in all directions without distorting the image on the screen.
Conventional stretchable displays are generally made by forming light-emitting devices on a stretchable substrate, which serves as the base layer of the display. However, when such a substrate is stretched in one direction, it tends to shrink in the opposite direction, causing letters and images on the screen to become flattened or distorted. Auxetic structures have been used to address this problem, but most previous approaches were limited to maintaining the overall horizontal-to-vertical ratio of the screen, while the letters and images within the screen still remained vulnerable to distortion.
Instead of bonding the auxetic structure and the stretchable substrate across the entire surface, as in conventional methods, the research team proposed a new design approach that uses computational analysis to selectively connect only the necessary points that ensure isotropic expansion throughout the substrate.
In the conventional approach, the twisting deformation that occurs as the auxetic structure stretches is directly transferred to the substrate, distorting the image inside the screen. In contrast, the platform developed by the research team was designed so that each region moves evenly outward from its original position. This allows not only the entire screen but also small areas such as letters and images to expand together while maintaining their original shapes.
The research team verified the platform’s performance by repeatedly stretching a substrate patterned with letters and images in both the horizontal and vertical directions. In the conventional method, the patterns underwent local deformation, whereas in the new platform, the shapes of the letters and images remained intact. This demonstrates that not only the whole screen but also fine images on-screen can expand uniformly without distortion.
The team also integrated an LED array, a structure in which multiple LEDs are arranged at regular intervals, onto the platform to verify its performance as an working stretchable display. Even when stretched by up to 15% in both the horizontal and vertical directions, stable electrical operation and the screen brightness were maintained. After repeated stretching to 15%, the decrease in brightness remained below 2%, confirming the platform’s potential for practical display applications.
This technology is expected to serve as a core platform for next-generation electronics with freely changeable shapes, including wearable electronic devices, electronic skin, or e-skin, which refers to electronic devices that stretch like skin while sensing and displaying information, medical biosensors, soft robots, and curved displays for automobiles and aircraft.
Professor Seunghyup Yoo of KAIST said, “For stretchable displays to be used as actual information display devices, they must not only stretch well, but also preserve on-screen information accurately during stretching,” adding, “This platform enables uniform expansion from small areas of the screen to the entire display, and will serve as a key foundational technology for accelerating the commercialization of high-quality stretchable displays.”
This study was led by KAIST Dr. Su-Bon Kim and Dr. Junho Kim as co-first authors, with Professor Hanul Moon of Dong-A University and Professor Seunghyup Yoo of KAIST as co-corresponding authors. The research was published in the international journal Nature Communications on June 10.
※ Paper title: Hybrid auxetic metamaterial platforms enabling multiscale isotropic expansion for distortion-free stretchable displays, DOI: 10.1038/s41467-026-74141-6
This research was supported by the National Research Foundation of Korea (NRF) Mid-Career Researcher Program, the Future Display Strategic Research Laboratory Program, the Korea Planning & Evaluation Institute of Industrial Technology (KEIT), and the Korea Institute for Advancement of Technology (KIAT) HRD Program.
2026.07.08 View 1417 -
KAIST Identifies the “Hidden Energy Cost” of AI Agents for the First Time
As the era of AI agents—systems that can reason and act autonomously—begins, the power consumption of data centers is emerging as a critical challenge. A KAIST research team has, for the first time, analyzed the computational cost and energy consumption of AI agents, finding that they can consume up to 136.5 times energy per query than conventional generative AI. The study shows that competitiveness in the AI era is expanding beyond model performance to include the efficiency of data centers and power infrastructure.
KAIST announced that a research team led by Professor Minsoo Rhu of the School of Electrical Engineering has systematically analyzed, for the first time, how much computational resources and power AI agents require in real-world service environments.
Large language model (LLMs) powered applications such as ChatGPT have rapidly evolved beyond simply answering questions. They are now developing into AI agents: next-generation AI systems that can plan, use external tools such as web search, calculators, and code execution environments, and solve complex tasks by coordinating multiple steps on their own.
Although AI agents are increasingly being adopted in areas such as software development, research, and workplace automation, little has been known about the amount of electricity and operational cost required to run them in practice.
The research team defined AI agents not merely as software programs, but as a new type of workload that must be continuously processed by data-center servers and graphics processing units, or GPUs—high-performance chips used for large-scale AI computation. The team then analyzed the computational load and energy consumption incurred during actual AI agent execution.
The analysis found that AI agents perform, far higher volumes of LLM invocations than conventional chain-of-thought reasoning. Chain-of-thought, or CoT, refers to a method in which an AI model breaks down its reasoning process step by step to reach an answer, while an LLM invocation refers to each computational request made to a language model to generate a new judgment or response.
Because AI agents repeatedly call language models during execution, their response latency also increases significantly. The team found that response time can increase by up to 153.7 times, while GPUs remain idle for as much as 54.5 percent of the total execution time as external tools perform their tasks. In other words, as AI systems take on more complex tasks, a new form of inefficiency emerges in which expensive GPUs cannot be fully utilized.
The research team also analyzed the power consumption of AI agents at data-center scale. An AI agent using a 70-billion-parameter LLM—a scale comparable to current commercial AI services—consumed an average of 348.41 watt-hours per query. This is 136.5 times higher than the energy consumed by a conventional generative AI system performing simple question answering.
In addition, the team projected a future scenario in which 13.7 billion AI agent requests are generated per day — a volume equivalent to current Google search traffic. Under this scenario, data-center power demand would reach approximately 198.9 gigawatts, a level far exceeding the scale of AI data centers currently under development (which are in the range of a few gigawatts) and equivalent to roughly half of the average power consumption of the United States.
This study demonstrates that the focus of competition in the AI era is shifting from “smarter AI” to “optimally efficient AI.” Going forward, it will be essential not only to advance AI models, but also to jointly optimize AI semiconductors, data centers, and power infrastructure through co-design. Such an approach is expected to become a key strategy for reducing the operating cost of AI services and building sustainable AI infrastructure.
“This study is the first to quantitatively show not only how AI is becoming more intelligent, but also how much electricity and cost are required to implement and sustain that intelligence,” said Professor Rhu. “As AI agents become widespread, it will become increasingly important to take an integrated co-design approach that optimizes not only AI data-center infrastructure, but also AI agent models and power infrastructure.” He added, “Research and investment in this direction will be essential to dramatically reduce the cost for end users to access AI services while building sustainable AI infrastructure.”
The study was conducted with Jiin Kim, a Ph.D. student in the KAIST School of Electrical Engineering, as the first author. The paper was presented in February at the 32nd IEEE International Symposium on High-Performance Computer Architecture, or HPCA, one of the most prestigious international conferences in computer system design. The research team has also released the AI agent implementations and benchmarks used in the paper as open source to support follow-up studies by researchers worldwide.
Paper title: “The Cost of Dynamic Reasoning: Demystifying AI Agents and Test-Time Scaling from an AI Infrastructure Perspective”
Open-source repository: 10.1109/HPCA68181.2026.11408569
This research was supported by the Institute of Information & Communications Technology Planning & Evaluation (IITP) through the SW Starlab program, the K-Cloud Technology Development Program using AI semiconductors, and the Leading Technology Development Program for Advancing AI-Semiconductor-Based Data Centers, as well as by the Samsung Electronics Future Technology Incubation Center.
2026.07.07 View 1645 -
KAIST Teams Win Both International Challenges at ICRA 2026 and CVPR 2026
Two research teams from KAIST have claimed first place in international challenge competitions held at the world’s premier robotics and computer vision conferences.
KAIST (President Kwang-Hyung Lee) announced that the ACDC-K Team and the Curaytor Team, both from the laboratory of Prof. Hyun Myung in the School of Electrical Engineering, won first place in international challenge competitions held in conjunction with the IEEE International Conference on Robotics and Automation (ICRA 2026) and the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2026), respectively.
The achievement highlights the global competitiveness of KAIST’s robotic perception and spatial intelligence technologies, with two teams from the same laboratory securing victories in leading international competitions across distinct research fields.
The ACDC-K Team won first place among more than 60 participating teams in the SLAM (Simultaneous Localization And Mapping) category of the Hilti×Trimble SLAM Challenge 2026, held during the Open Challenges in Robotics for Asset Inspection (OCRAIM) Workshop at ICRA 2026 in Vienna, Austria, from June 1 to 5.
Jointly organized by Hilti, Trimble, and the University of Oxford, the challenge evaluates robotic localization and mapping performance using sensor data collected from real construction sites. Participants were required to address practical challenges frequently encountered in construction environments, including non-overlapping front and rear fisheye camera configurations, low-texture indoor scenes, and rapid camera motion.
To tackle these challenges, the ACDC-K Team developed a robust visual-inertial SLAM system that fuses front and rear fisheye camera data with inertial measurements. By integrating feature-point and feature-line observations with adaptive constraints and correction mechanisms, the team achieved highly reliable localization and mapping performance in complex construction environments.
Meanwhile, the Curaytor Team won first place among eight participating teams in the Nothing Stands Still (NSS) Challenge 2026, held during the Computer Vision for the Built World (CV4AEC) Workshop at CVPR 2026 in Denver, Colorado, from June 3 to 7.
Jointly organized by Stanford University, ETH Zurich, and Oregon State University, the NSS Challenge evaluates 3D point cloud registration technologies for construction and industrial environments that evolve over time.
The Curaytor Team developed a novel multi-registration framework capable of aligning multiple LiDAR scans collected across different times and locations. The framework integrates feature extraction, correspondence estimation, robust global registration, registration confidence assessment, and change-aware refinement techniques. As a result, the team achieved highly accurate registration performance even in environments containing structural changes and dynamic objects.
“This achievement demonstrates the robustness of our visual-inertial SLAM and 3D LiDAR registration technologies in complex and constantly changing real-world environments,” said Prof. Hyun Myung. “It is particularly meaningful that our students secured first-place finishes in highly competitive international challenges hosted at two of the world’s most prestigious conferences in robotics and computer vision.”
Prof. Hyun Myung’s laboratory has consistently demonstrated excellence in spatial intelligence research. The laboratory previously won first place in the LiDAR track and ranked first among academic teams in the vision track of the Hilti SLAM Challenge in 2023. In addition, the Curaytor Team successfully defended its title in the NSS Challenge, securing back-to-back championships in 2025 and 2026.
2026.06.19 View 3208 -
Breakthrough in Data Processing via Light Control... Enhancing AI Accelerators and Quantum Communication
< (From left) Undergraduate researcher Taewon Kim and Professor Sangsik Kim >
A new technology has been developed that allows light to be "designed" into desired forms, potentially making Artificial Intelligence (AI) and communication technologies faster and more accurate. A KAIST research team has developed an "integrated photonic resonator"—a core component of next-generation optical integrated circuits that process data using light. The research is particularly significant as it was led by an undergraduate student. This technology is expected to serve as a key foundation for next-generation security technologies such as high-speed data processing and quantum communication.
KAIST announced on the 15th that a research team led by Professor Sangsik Kim from the School of Electrical Engineering, in collaboration with Professor Jae Woong Yoon’s team from the Department of Physics at Hanyang University (President Kigeong Lee), has developed a new integrated photonic resonator structure capable of freely controlling optical signals by utilizing light interference (the phenomenon where two light waves meet and influence each other).
Photonic Integrated Circuits (PICs) process data at ultra-high speeds and with low power consumption using light. They are garnering significant attention as a fundamental platform technology for next-generation fields such as AI, data centers, and quantum information processing.
The core of this technology lies in the precision with which light can be controlled. Specifically, the ability to freely adjust the spectrum (color or wavelength distribution) and phase response (timing or wave position) of optical signals is essential for implementing high-performance optical communication and computing. However, conventional methods have faced fundamental limitations.
The integrated photonic resonator (optical resonator) focused on by the research team is a key optical device that traps light in a specific space to amplify it or select specific colors (wavelengths), similar to how the body of a musical instrument amplifies sound. However, existing single-bus resonators have had limitations in precisely adjusting the phase and spectrum of optical signals.
To overcome these challenges, the research team introduced a "dual-bus" structure. This design allows light that has passed through the resonator to recombine with light that has not, enabling precise control over interference. This allows for the free design of optical signals into desired forms, making it possible to control various types of light signals that were previously difficult to implement.
By applying this technology, the research team secured new characteristics for more precise control of wavelength properties and presented new possibilities for non-linear frequency conversion research (changing the color of light). Utilizing this technology enables faster and more accurate data processing, which is expected to provide the groundwork for performance enhancements in future high-speed data centers, AI accelerators, and quantum communication systems.
This research is especially meaningful as it was led by an undergraduate student. Taewon Kim, an undergraduate student who conducted the study through the KAIST Undergraduate Research Program (URP), stated, "I was able to develop the resonator principles I learned in the Introduction to Integrated Optics class into actual device designs and a published paper."
< Research Image of the Dual-bus Resonator >
Professor Sangsik Kim remarked, "This study goes beyond proposing a new device; it demonstrates that by precisely analyzing previously overlooked optical characteristics, physical limitations can be overcome. We expect this to contribute broadly to the development of optics-based AI accelerators and optical communication technologies."
KAIST undergraduate student Taewon Kim participated as the lead author of this study, and the results were published on March 6th in the international optics journal, Laser & Photonics Reviews.
Paper Title: Dual-bus resonator for multi-port spectral engineering DOI: 10.1002/lpor.202502935 Authors: Taewon Kim, Mehedi Hasan, Yu Sung Choi, Jae Woong Yoon, and Sangsik Kim
This research was supported by the KAIST URP Program, the Institute of Information & Communications Technology Planning & Evaluation (IITP), the U.S. Asian Office of Aerospace Research and Development (AOARD), and the National Research Foundation of Korea (NRF).
2026.04.16 View 2188 -
KAIST Enables Ultra-High-Resolution AR·VR Without Battery Burden
<(From Left) Professor Young Min Song, Ph.D candidate Hyo Eun Jeong, (Upper Left) Professor Hyeon-Ho Jeong, Dr. Joo Hwan Ko>
A new display technology has emerged that significantly increases resolution while consuming almost no power. A Korean research team has developed a “monopixel” structure in which a single pixel can independently change colors while consuming minimal energy to maintain them. This breakthrough opens the possibility of creating sharper AR/VR displays without heavy battery demands.
KAIST (President Kwang Hyung Lee) announced on the 29th of March that a research team led by Professor Young Min Song of the School of Electrical Engineering, in collaboration with Professor Hyeon-Ho Jeong’s team at Gwangju Institute of Science and Technology (GIST, President Ki-Cheol Lim), has developed a new low-power monopixel technology called a “reconfigurable Gires–Tournois resonator (r-GT).” This system uses electrochromic materials—substances that change color when electricity is applied—to produce colors with very low power consumption.
Displays have been making pixels increasingly smaller to achieve higher resolution. However, as pixels shrink, power consumption rises and brightness decreases. This is especially challenging for AR/VR devices, which must achieve both extremely small pixels and low power consumption due to their proximity to the human eye.
The r-GT pixel developed by the research team changes color when voltage is applied, and once changed, the color is maintained for a certain period even after the power is turned off. In other words, power is only required when changing colors, while maintaining color requires almost no energy.
The core of this technology lies in two elements. First is a conductive polymer, polyaniline (PANI), whose properties change when voltage is applied. This material responds even at voltages below 1 volt (V), altering its refractive index and thereby changing color. The refractive index refers to how much light bends when passing through a material, and changes in this value lead to changes in perceived color.
Second, the system incorporates a resonator structure that reflects light multiple times to amplify specific colors. This structure enhances even small changes, enabling vivid color expression with minimal power.
As a result, the system achieved a wide color variation exceeding 220° using ultra-low power (90 μW cm⁻²). In simple terms, it can express more than half of the full color wheel (360°) using only about 0.00009 watts per square centimeter.
Another key feature is the “monopixel” structure. Unlike conventional displays that divide a single pixel into red (R), green (G), and blue (B) subpixels, the monopixel approach allows one pixel to independently produce various colors. This enables more pixels within the same area, resulting in higher resolution and reduced light loss, leading to clearer images.
Additionally, PANI retains its color state even after the applied voltage is removed. This confirms the feasibility of a “memory-in-pixel” display, where energy is used only when changing colors, not when maintaining them.
<Reflective display AI image>
The research team demonstrated that this technology can achieve a wide color range (220.6°) and reduce pixel size to as small as 1.5 micrometers (μm), corresponding to an ultra-high resolution of up to approximately 16,900 PPI—beyond the level where individual pixels can be distinguished by the human eye.
Moreover, even with a single-pixel structure, the system can represent about 48.1% of the standard sRGB color gamut, and up to 69.9% with varied material combinations, enabling richer color expression.
The team fabricated a 5×5 monopixel array to verify performance. The energy required to change colors was extremely low (2.31 mJ), demonstrating up to 5.8 times lower power consumption compared to conventional LEDs. As a reflective display, it also becomes more visible under brighter ambient lighting, since it uses external light rather than emitting its own.
<Structure and Representative Results of an Electrically Tunable Single Reflective Resonant Device Using Conductive Polymers>
This study demonstrates that combining electrochemical materials with optical resonator structures enables full-color implementation at ultra-low power. It is expected to be applied in various fields requiring energy efficiency, including ultra-high-resolution near-eye displays for AR/VR, wearable devices, outdoor displays, and electronic paper. It also suggests the potential for sustainable and energy-efficient display technologies by minimizing power consumption during color retention.
Professor Young Min Song stated, “This technology allows a wide range of color changes using very little electricity,” adding, “When combined with future display driving methods, it could enable not only clearer and more energy-efficient ultra-high-resolution displays but also a variety of optical applications.”
This research was conducted with Hyo Eun Jeong, an integrated M.S./Ph.D. student at KAIST, as co-first author, and Professor Young Min Song as the corresponding author. The results were published online on February 28 in Light: Science & Applications, a leading international journal in optics.
※ Paper title: “Sub-1-volt, reconfigurable Gires-Tournois resonators for full-coloured monopixel array,” DOI: https://www.nature.com/articles/s41377-026-02228-2
This research was supported by multiple programs funded by the Ministry of Science and ICT, the National Research Foundation of Korea (NRF), the InnoCORE-GIST program, nanomaterials and technology development initiatives, future medical innovation programs, international collaboration hubs, and the Ministry of Trade, Industry and Energy (MOTIE).
2026.03.30 View 959 -
KAIST Expands Storage Capacity with Smart Gate Semiconductor Technology
<(From Left) Ph. D candidate Dae Hyun Kang, Professor Byung Jin Cho>
From smartphones to large-scale AI servers, most digital information in modern society is stored in NAND flash memory*. KAIST researchers have developed an innovative technology that can overcome the limitations of next-generation semiconductors, where more data must be stored in smaller spaces. This advancement is expected to serve as a key enabling technology for realizing ultra-high-capacity memory.*NAND flash memory: a non-volatile semiconductor memory used in storage devices such as smartphones and SSDs, where data such as photos, videos, and apps are retained even when power is turned off.
KAIST (President Kwang Hyung Lee) announced on the 20th of March that a research team led by Professor Byung Jin Cho of the School of Electrical Engineering has overcome the scaling limitations of 3D V-NAND memory* by implementing a “smart gate” structure that selectively controls electron movement depending on conditions, using a new material applied to an ultra-thin semiconductor layer thinner than a human hair.*3D V-NAND: a memory technology that stacks memory cells vertically, unlike conventional planar (2D) arrangements, enabling higher data storage density.
This research is particularly significant in that it addresses the longstanding issues of speed degradation and reliability during data write and erase operations by utilizing a novel material called boron oxynitride (BON).
In semiconductor memory, the tunneling layer—a thin insulating layer that acts as a pathway for electrons to move in and out of the memory cell—has historically faced a trade-off between performance and reliability.
With conventional materials, it has been difficult to achieve both simultaneously. For example, the widely used silicon oxynitride (SiON) increases data leakage when the tunneling path is widened to improve erase speed, while narrowing the path to prevent leakage significantly slows down data erasure. This trade-off has been a major obstacle to implementing next-generation penta-level cell (PLC) technology.
PLC technology stores 5 bits of data per memory cell by distinguishing 32 different voltage states, allowing much higher data density within the same physical size.
To overcome this limitation, the research team introduced BON, a completely new material beyond conventional silicon-based systems, into the tunneling layer. This material exhibits a unique physical property in which the energy barrier height differs depending on the type of charge carrier.
Leveraging this property, the team designed an asymmetric energy barrier structure that allows holes (positive charge carriers)—needed for data erase—to pass through easily, while blocking electrons, which represent stored data, from leaking out.
An asymmetric energy barrier refers to a structure in which the energy required for charge carriers to move varies depending on the type of charge. This enables efficient charge transport during erase operations while effectively preventing data loss. The concept is analogous to a “smart gate” that opens easily for entry but firmly blocks exit, implemented at the semiconductor level.
Experimental results showed that devices using the BON tunneling layer achieved up to a 23-fold improvement in erase speed compared to conventional technologies and demonstrated excellent durability with minimal performance degradation even after tens of thousands of operation cycles.
Notably, even under the highly demanding PLC operation—where 32 distinct voltage levels must be precisely controlled—the researchers achieved more than threefold improvement in controlling data distribution across devices.
< Schematic diagram of the asymmetric energy barrier structure and operating principle of the BON tunneling layer >
This achievement is considered by both academia and industry to be beyond a purely experimental result, reaching a level immediately applicable to real semiconductor manufacturing processes.
Professor Byung Jin Cho stated, “This research presents a novel technology that can be directly applied to the production of next-generation ultra-high-capacity memory,” adding, “It will significantly contribute to maintaining Korea’s technological leadership in the semiconductor industry.”
This study was implemented by Dae Hyun Kang, an integrated master’s–PhD student in Electrical Engineering, as the first author. The research was presented at the IEEE International Electron Devices Meeting (IEDM) on December 9, one of the most prestigious conferences in the semiconductor field, attracting global attention.
The work also received the Grand Prize (first place overall in the university category) at the 32nd Samsung Human Tech Paper Awards, marking a notable achievement as a traditional semiconductor device study in a competition typically dominated by AI-related research.
※ Paper title: “Bandgap-Engineered Boron Oxynitride Tunneling Layer for Reliable PLC Operation of 3D V-NAND Flash Memory Devices,” DOI: https://doi.org/10.1109/IEDM50572.2025.11353681
This research was supported by the National Semiconductor Research Lab Core Technology Development Program funded by the Ministry of Science and ICT.
2026.03.23 View 1395