[ECE Department] Professor Jaejoon Kim named May’s Scientist of the Month for contributions to lightweight AI models through integrated hardware–software co-design
Professor Jaejoon Kim, Department of Electrical and Computer Engineering, Seoul National University[Provided by the Ministry of Science and ICT. Redistribution and database storage prohibited.]
(Seoul = Yonhap News) Reporter Cho Seung-han – The Ministry of Science and ICT and the National Research Foundation of Korea announced that Professor Jaejoon Kim of the Department of Electrical and Computer Engineering at Seoul National University has been selected as the May recipient of the Scientist of the Month Award. Professor Kim was recognized for his contributions in developing lightweight AI models and designing semiconductor accelerators that efficiently compute these models, thereby proposing AI-based technologies that can operate with low power consumption across various environments.
Efficiency optimization of AI models—gaining attention recently due to power consumption concerns—can be divided into two approaches: software-centric model compression and hardware-centric semiconductor accelerator technologies. However, since these approaches have typically been developed independently, applying compressed models in practice has often resulted in reduced computational speed.
Professor Kim conducted integrated research to develop lightweight AI techniques that take hardware characteristics into account, while simultaneously designing a dedicated semiconductor chip to efficiently support them. This technology supports AI lightweight models with “variable bit precision,” which adjusts the number of bits used in computation according to the required accuracy level or hardware resource constraints. The system uses a single accelerator circuit to handle these models, enabling efficient support for variable precision with a simpler computational unit structure than before. In addition, Professor Kim challenged the conventional belief that achieving high computational accuracy requires floating-point units—which are large in chip area and consume significant power—and demonstrated that integer arithmetic units can achieve the same level of accuracy. These research results were published in the IEEE Journal of Solid-State Circuits in June 2022 and presented at the International Conference on Learning Representations (ICLR) in May 2023. Professor Kim stated, “I will continue to focus on designing low-power integrated circuits that minimize energy consumption to contribute to the sustainable development of AI,” and added, “I expect this research to serve as a core foundational technology enabling AI computation in mobile environments where power efficiency is critical.”
Professor Jaejoon Kim’s research team
[Provided by the Ministry of Science and ICT. Redistribution and database storage prohibited.]
Source: https://www.yna.co.kr/view/AKR20250430062200017?input=1195m
Translated by: Dohyung Kim, English Editor of the Department of Electrical and Computer Engineering, kimdohyung@snu.ac.kr
