Scientists have created a computing chip capable of learning, correcting mistakes, and handling AI tasks.
Conventional computer systems often use separate components for data processing and storage, which can make them less efficient when dealing with complex information like artificial intelligence. A research team from KAIST has introduced a memristor-based integrated system that mimics the brain’s information processing. This advancement is suitable for use in various devices, such as smart security cameras that can promptly detect suspicious behaviors without the need for cloud servers, as well as medical devices for real-time health data analysis.
On January 17, KAIST (under the leadership of President Kwang Hyung Lee) announced that a collaborative research team, featuring Professors Shinhyun Choi and Young-Gyu Yoon from the School of Electrical Engineering, has developed an ultra-compact, neuromorphic semiconductor chip capable of independent learning and error correction.
This chip stands out because it can learn and rectify errors linked to non-optimal traits that previously troubled neuromorphic devices. For instance, while processing a video feed, the chip not only learns to distinguish moving objects from their backgrounds but gets progressively better at this task over time.
Its self-learning ability has been demonstrated by achieving real-time image processing accuracy comparable to that of ideal computer simulations. The significant achievement of the research team is the creation of a reliable and practical system that goes beyond just developing brain-like components.
The team has designed the first memristor-based integrated system globally that can adjust swiftly to changing environments, introducing a groundbreaking solution that addresses the constraints of current technology.
At the core of this innovation lies a next-generation semiconductor device known as a memristor*. This device’s variable resistance functions like synapses in neural networks, enabling simultaneous data storage and computation, akin to the functions of our brain cells.
*Memristor: A term combining “memory” and “resistor,” describing a future-generation electrical device where the resistance depends on the charge flow history between its terminals.
The research team has successfully crafted a highly dependable memristor that accurately manages resistance changes, developing an effective system that eliminates the need for complex compensatory methods by utilizing self-learning. This research is vital as it experimentally confirmed the feasibility of commercializing a next-gen neuromorphic semiconductor-based integrated system that enables real-time learning and inference.
This innovation is poised to transform how artificial intelligence is implemented in everyday devices, allowing processing to occur locally—without dependence on remote cloud servers—resulting in quicker, more secure, and energy-efficient operations.
“This system operates like a smart workspace, where everything is easily accessible without the need to shuffle between desks and file cabinets,” explained KAIST researchers Hakcheon Jeong and Seungjae Han, who spearheaded this technology’s development. “It’s reminiscent of the way our brain efficiently processes information all at once in one location.”
The team included Hakcheon Jeong and Seungjae Han, students in the Integrated Master’s and Doctoral Program at KAIST’s School of Electrical Engineering, and the findings were published online on January 8, 2025, in the international journal, Nature Electronics.
This research received support from the Next-Generation Intelligent Semiconductor Technology Development Project, the Excellent New Researcher Project, and the PIM AI Semiconductor Core Technology Development Project from the National Research Foundation of Korea, as well as the Electronics and Telecommunications Research Institute’s Research and Development Support Project from the Institute of Information & Communications Technology Planning & Evaluation.
