Researchers in South Korea have developed a miniature computing chip that can self-learn and correct errors like a brain. This technology could be used for smart security cameras to immediately recognize suspicious activities or for medical devices to analyze health data in real time. In particular, it is expected to allow AI tasks to be processed on the device itself rather than in the cloud, resulting in faster performance, enhanced privacy protection, and improved energy efficiency.
On the 17th, a joint research team led by Professors Jeon Jae-hyeon and Yoon Young-kyu from the Department of Electrical and Electronic Engineering at the Korea Advanced Institute of Science and Technology (KAIST) announced that they have developed a next-generation neuromorphic semiconductor-based miniature computing chip that can self-learn and correct errors. The research results were published online in the international journal "Nature Electronics" on the 8th.
The computing chip developed by the researchers can learn and correct errors arising from atypical characteristics that were difficult to resolve in existing neuromorphic devices. For instance, when processing video streams, the chip learns how to automatically separate moving objects from the background, improving its performance over time. This self-learning capability has achieved accuracy comparable to an ideal computer simulation in real-time video processing.
At the core of this technology is a next-generation semiconductor device called a memristor. A memristor, a combination of memory and resistor, is a next-generation electronic device whose resistance value is determined by the amount and direction of past charge flow. The variable resistance characteristics of this device can replace the synapses in neural networks, enabling simultaneous data storage and computation akin to our brain cells. The researchers developed the world’s first memristor-based integrated system capable of adapting to immediate environmental changes. Additionally, they created an efficient system that eliminates complex correction processes through self-learning.
The researchers noted, "This study is significant in that it experimentally verified the commercial viability of next-generation neuromorphic semiconductor-based integrated systems that support real-time learning and inference," adding, "As everyday devices can handle AI task processing locally without relying on remote cloud servers, we can expect faster performance, enhanced privacy protection, and increased energy efficiency."
Research Institute members Jeong Hak-cheon and Han Seung-jae from KAIST explained, "This system is like a smart workspace where everything is within reach instead of moving between a desk and a filing cabinet," adding, "This approach is very similar to how our brain processes information efficiently in one place."
Reference materials
Nature Electronics (2025), DOI: https://doi.org/10.1038/s41928-024-01318-6