A research team from the Korea Advanced Institute of Science and Technology has developed an ultra-small computer chip based on next-generation neuromorphic semiconductors that can learn and correct errors on their own. The research was published in the international scientific journal Nature Electronics .
Computer systems are inefficient at handling complex data, such as in artificial intelligence, because the data storage and processing components are separate. A team of researchers at KAIST has developed an integrated memristor-based system that simulates the way the brain processes information. The system could be used in a range of devices, such as smart security cameras, to instantly identify suspicious activity without relying on remote cloud servers, or in medical equipment to analyze health data in real time.
This research was carried out by a collaborative research team consisting of Professor Choi Shin-hyeon and Professor Yoon Young-kyu from the Department of Electrical Engineering at KAIST (Chairman Lee Kwang-hyeon).
The computing chip is unique in that it can learn and correct errors caused by non-ideal features that have posed challenges for previous neuromorphic devices. For example, when processing a video stream, the chip learns to automatically distinguish between moving objects and the background, improving over time.
Real-time image processing demonstrated this self-learning capability by achieving accuracy on par with the best computer simulations.Besides creating a brain-like component, the team’s main achievement was to create a reliable and useful system.
The team has developed the first ever integrated memristor-based system that can instantly adapt to changes in its environment. The team also came up with innovative solutions that overcome the limitations of current technology.
At the heart of this innovation is a next-generation semiconductor device called a memristor, whose tunable resistive properties could replace synapses in neural networks, allowing them to simultaneously store data and perform computations, just like human brain cells.
The team developed an extremely reliable memristor that can precisely adapt to resistance changes and an efficient self-learning system that eliminates complex compensation processes. This work is notable because it experimentally validates the commercial viability of next-generation neuromorphic semiconductor-based integrated systems that enable real-time learning and inference.
This technology will completely change how we use artificial intelligence in everyday devices by processing AI tasks locally, rather than relying on far-away cloud servers, making them faster, more energy efficient, and with better privacy.
” The system is like a smart workspace where everything is at your fingertips, without you having to go back and forth between your desk and file cabinets. “It’s similar to the way our brains process information, where all the information is processed efficiently at once ,” explained KAIST researchers Jeong Hak-cheon and Han Seung-jae, who led the development of the technology.
Both Cheong and Han were enrolled in an integrated master’s and doctoral program.
This research was funded by the National Research Foundation of Korea’s Outstanding New Researcher Project, AI PIM Semiconductor Core Technology Development Project, Next-Generation Smart Semiconductor Technology Development Project, and the Korea Electronics and Communications Research Institute’s R&D Support Project.
Journal References:
John, H., et al . (2025) Self-calibrated self-monitored video processing on an analog computing platform based on selectorless memristor arrays. Nature Electronics . doi.org/10.1038/s41928-024-01318-6
sauce:
Korea Advanced Institute of Science and Technology (KAIST)
