Published in the Science and Technology of Advanced Materials Journal under the title “Predicting material properties by integrating high-throughput experiments, high-throughput ab-initio calculations, and machine learning”, the new approach could speed up the development of new materials with particular electronic or magnetic properties.
Today, most scientists use high throughput experimentation, involving large numbers of parallel experiments, to quickly map the relationships between the compositions, structures, and properties of materials made from varying quantities of the same elements. This helps accelerate new material development, but usually requires expensive equipment. High throughput calculation, on the other hand, uses computational models to determine a material’s properties based on its electron density, a measure of the probability of an electron occupying an extremely small amount of space. It is faster and cheaper than the physical experiments but much less accurate.
Materials informatics expert Yuma Iwasaki of the Central Research Laboratories of NEC Corporation, together with colleagues in Japan, combined the two high-throughput methods, taking the best of both worlds, and paired them with machine learning to streamline the process. They tested their approach using a 100 nanometre-thin film made of iron, cobalt and nickel spread on a sapphire substrate. Various possible combinations of the three elements were distributed along the film. These ‘composition spread samples’ are used to test many similar materials in a single sample.