loaction
공대뉴스광장
Research team of SNU Professor Jaejin Lee Developed FPGA Chip Core Software ... "Proprietary Technology Beyond Intel, XILINX"
-
작성자
관리자
-
등록일
2020.03.26.
-
조회수
1193
Research team of SNU Professor Jaejin Lee
Developed FPGA Chip Core Software ...
"Proprietary Technology Beyond Intel, XILINX"
- Development of proprietary programming technology that is easy, has high performance, and requires low power
- RPGA chip that changes circuit structure... Overcoming of limitations in application such as reconfiguration time
▲ From left, Professor Jaejin Lee (SNU), Dr. Gangwon Jo, Researchers Heehoon Kim and Jeesoo Lee
Amid growing demand for low-power, high-performance semiconductor chip systems for processing AI and big data, local researchers have developed core system software technologies for FPGA chips that outperform Intel's technologies.
SNU College of Engineering (Dean Kookheon Char) announced March 20 that a team led by Professor Jaejin Lee of the Department of Computer Science and Technology has developed a proprietary programming technology that is easy, shows high-performance but requires low energy when executed in software written in OpenCL language, an international standard.
Unlike other semiconductor chips, FPGA chips can change their circuit structure through programming whenever necessary, which gives them higher performance and power efficiency than conventional universal CPUs. However, limitations include the fact that it takes a long time to reconstruct the circuit structure of the FPGA by having to program it each time and that only a small number of professionals can perform this task.
Professor Jaejin Lee's team overcame this limitation by developing a technology that automatically analyzes the source code of software and creates an optimal FPGA circuit structure. Just as the CPU programming environment automatically creates a mechanical code to run the software, the FPGA programming environment automatically creates a circuit structure. As a result, the user can program and use the FPGA as easily as the universal CPU.
The research team used SPEC benchmarks to test the technology and found that the OpenCL programming environment developed by the team solved all the errors that the commercial OpenCL programming environment of FPGA world leaders such as Intel and Xilinx had. Furthermore, they showed to have better performance and power efficiency in comparison to Intel and Xilinx products.
"It was difficult to respond quickly to new artificial intelligence algorithms as semiconductor chips like NVIDIA GPU or Google's TPU, which were previously used to deal with artificial intelligence, were fixed in circuitry," said Gangwon Jo, a SNU researcher who was a high contributor to the study. "Through this research, we achieved high performance and low power requirements by quickly accepting new artificial intelligence algorithms whenever they come up through the FPGA,' he added.
"This research significance lies in the development of the world's first OpenCL programming environment for FPGA, which requires easy use of FPGA for high-performance calculations," said SNU Professor Jaejin Lee. "While the software technology for system semiconductors developed by U.S. companies is now commercialized, our research is an important event that shows that we can lead the next generation of system semiconductor software technologies."
The globally recognized research results have been selectively adopted as a paper at the International Symposium on Computer Architecture, an international academic conference in the field of computer architecture, and is scheduled to be presented in June this year. Meanwhile, the research was conducted with the support of the Korean National Research Foundation's ultra-high performance computing research group (Director Hwan-soo Han), which is part of the next generation information and computing technology development project.
담당부서대외협력실
전화번호880-9148