Machine learning boosts thermal material discovery : Page 2 of 2

February 12, 2020 //By Julien Happich
machine learning
Researchers in Japan have developed an approach that can better predict the properties of materials by combining high throughput experimental and calculation data together with machine learning.

The team first conducted a simple high throughput technique on the sample called combinatorial X-ray diffraction. The resulting X-ray diffraction curves provide detailed information about the crystallographic structure, chemical composition, and physical properties of the sample.

The team then used machine learning to break down this data into individual X-ray diffraction curves for every combination of the three elements. High throughput calculations helped define the magnetic properties of each combination. Finally, calculations were performed to reduce the difference between the experimental and calculation data.
Their approach allowed them to successfully map the ‘Kerr rotation’ of the iron, cobalt, and nickel composition spread, representing the changes that happen to light as it is reflected from its magnetized surface. This property is important for a variety of applications in photonics and semiconductor devices.



Image caption: (a) Kerr rotation mapping of an iron, cobalt, nickel composite spread using the more accurate high throughput experimentation method, (b) only high throughput calculation, and (c) the Iwasaki et al. combined approach. The combined approach provides a much more accurate prediction of the composite spread’s Kerr rotation compared to high throughput calculation on its own.

The researchers say their approach could still be improved but that, as it stands, it enables mapping the magnetic moments of composition spreads without the need to resort to more difficult and expensive high throughput experiments.

NEC Corporation – www.nec.com
 

Related article:


Vous êtes certain ?

Si vous désactivez les cookies, vous ne pouvez plus naviguer sur le site.

Vous allez être rediriger vers Google.