Research on spintronics, which overcomes the limitations of aggregation of silicon semiconductors currently used, is being actively done nowadays. It is to combine electronics ‘spin’ and ‘electronics’ together to develop ultra-low-power and high-performance next-generation semiconductors. In order to develop spintronics devices such as magnetic memory (MRAM), magnetic materials such as temperature stability are used. Speed to respond to changes must be accurately identified before they can be used for device development.
Spintronics is a new paradigm electronics that uses spin properties, which are quantum mechanical properties of electrons, to replace conventional silicon semiconductors.
To this end, South Korean researchers have developed artificial intelligence that analyzes the properties of magnetic materials, which are materials of spintronics, in an instant. The Korea Advanced Institute of Science and Technology (KIST) developed a technology to estimate magnetic properties from the magnetic characteristics of the electrons of magnetic bodies by using artificial intelligence technology through a joint study by Kwon Hee-young and Choi Joon-woo, a team of researchers at Kyung Hee University. This artificial intelligence is a deep learning technology that can solve material analysis that used to take up to several hours.
Magnetometers have magnetic domains which are the areas where spins, the micro-unit magnets that make up the material, are aligned in the same direction. As these magnetic domains form and change, various magnetic phenomena are known to occur. To date, scientists measured the properties of magnetic domains directly through a variety of experiments, and they had to spend a lot of time and resources on this.
KIST-Kyunghee University’s joint research team overcame the above limitations by utilizing deep learning technology. They applied machine learning algorithms to artificial intelligence to learn existing magnetic domain images. When it looks at new magnetic domain images, it estimates the magnetic properties of the material. This made the estimation of magnetic properties of a magnet in real time by entering an electron microscope image of a magnet. Compared to the actual observed data with the estimated values of artificial intelligence, the estimation accuracy was very high, with the error of around 1%.
Dr. Kwon Hee-young of KIST said, “We have presented a new path to how artificial intelligence technologies can be used to analyze the characteristics of magnetic domains.” “We hope that new research methods to analyze magnetic systems will strengthen the connection between experiments and theories, and furthermore, the convergence of artificial intelligence technology and pure science.”
KIST. (n.d.-b). Retrieved December 27, 2020, from https://www.kist.re.kr/kist_web/?sub_num=2935&state=view&idx=3591&ord=0
Kwon, H. Y., et al. “Magnetic Hamiltonian parameter estimation using deep learning techniques.” Science advances 6.39 (2020): eabb0872.