Content Shortcut menu Shortcut



Completion of a supercomputer to solve the challenges of commercializing fusion


Date of registration 2020-11-24

 - Research on nuclear fusion simulation leveraging 1PF high-performance supercomputer




Kraken, NFRI’s first supercomputer, which retired in January this year after 10 years of challenges and achievements

The first project chosen by Aurora, the first exaflops-class supercomputer in the U.S., which can calculate 1 quintillion floating point operations per second in 2018, was the Princeton Plasma Physics Research Institute (PPPL)'s artificial intelligence (AI) project for nuclear fusion research. The fact that the world’s top-notch supercomputer is planned for the nuclear fusion study indicates how significant and challenging nuclear fusion research is.

Though supercomputers have been globally mobilized, nuclear fusion research is yet to be realized on Earth for commercial purpose. This is because predicting the movement of plasma particles of more than 100 quintillion per unit volume in tokamaks and finding optimal operation conditions with proper control is not an easy task even with supercomputers.

This is not the first supercomputer that has been introduced in KFE. NFRI(KFE’s previous name) introduced the 60 teraflop-supercomputer "Kraken" in 2011 to understand and solve challenging issues such as plasma transport and turbulence. Although there have been research achievements in the fusion theory and modeling, including code development, which can interpret and analyze KSTAR plasma experiments, there have been limitations in accommodating the increasing computing resource demand.

The fusion plasma, of which temperature exceeds 100 million °C, can be modelled accurately by a five to six-dimensional kinetic model rather than a general hydrodynamic model due to its high temperature and torus-shaped magnetic field geometry. In addition, existing supercomputers are not powerful enough to prepare a 'Virtual DEMO' in which a virtual fusion device may be designed and verified for the development of fusion energy. This is where "KAIROS" would play a significant role in the future.

KFE is ready to search for a "best operating condition" with KAIROS, whose calculation capability per second is 25 times larger than the retiring supercomputer Kraken. In August 2020, 'KAIROS', a 1PF high-performance supercomputer, began a full-scale service for Korean fusion research.

KAIROS, the name given through a public contest, is an Ancient Greek word meaning the right, critical moment. While chronos, another greek word for time, means natural and continuous passage of time, kairos implies an "opportune moment or time for action". Since 1995, the KSTAR has been developed and upgraded and the research program has progressed continuously. KFE now seeks to take a critical step toward realizing nuclear fusion, by enhancing its soft power with KAIROS.

The theoretical performance (Rpeak) of KAIROS is 1.56 petaflops and the real performance (Rmax) measured by HPL benchmark is 1.01 petaflops. It is the largest in Korea among the supercomputers dedicated to a specific research field. KFE plans to use KAIROS to develop and extend simulation codes to fully analyze and predict results of the ITER experiments that will begin operation in 2025. Furthermore, it will be used to develop virtual fusion devices that are geared for efficient and reliable design and verification of K-DEMO.

KAIROS aims to strengthen the soft power of the nation's nuclear fusion technologies and will be opened up to domestic fusion research communities including universities and industries, thereby help boost up the domestic nuclear fusion research capabilities.

Authorities concerned