A domestic research team has developed a next-generation image sensor that can automatically adapt to extreme brightness changes without separate image processing technology. This technology is expected to be applicable in various fields such as autonomous vehicles, smart robots, and security systems.
The National Research Foundation of Korea (NRF) announced on the 18th that a joint research team led by Professor Song Young-min of the Korea Advanced Institute of Science and Technology (KAIST) and Professor Kang Dong-ho of the Gwangju Institute of Science and Technology (GIST) has developed a ferroelectric-based optical device inspired by the neural structure of the brain, which can implement light detection, recording, and processing within the device itself, leading to the development of a next-generation image sensor. The results of this research were published in the international journal 'Advanced Materials' on the 28th.
As demand for 'seeing artificial intelligence (AI)' increases, developing high-performance visual sensors that can operate stably in various environments has emerged as an urgent task. However, existing image sensors that process each pixel's signals individually are prone to excessive exposure or information loss due to low brightness in environments where lighting changes drastically. In particular, collected data had to be separately corrected or post-processed externally.
To address this issue, the research team designed a ferroelectric-based image sensor capable of adapting to extreme environmental changes, inspired by biological neural structures and learning methods. Ferroelectric materials refer to substances that possess spontaneous electrical polarization even in the absence of an external electric field.
The research team revealed that by adjusting the polarization state of the ferroelectric material, they could maintain detected light information for extended periods and selectively amplify or suppress it, achieving functions such as contrast enhancement, brightness correction, and noise suppression without complex image post-processing. They proved stable face recognition that does not distinguish between day and night or indoors and outdoors, solely through in-sensor processing without reconstructing learning data or additional training. Additionally, the developed device showed high compatibility with existing AI learning algorithms.
Professor Song Young-min noted, 'This research is significant in that it expands the ferroelectric devices, which have mainly been used as electrical memory devices, into the fields of neuromorphic vision and in-sensor computing.' He added, 'In the future, we plan to develop a next-generation vision system that can precisely detect and process changes in light wavelength, polarization, and phase.'
References
Advanced Materials (2025), DOI: https://doi.org/10.1002/adma.202503475