(Invited, Digital Presentation) The Challenge of Automated High-Throughput Experiments on a Variety of Powder Libraries

Abstract

In recent years, "Materials Genome Initiative (MGI)", "Materials Informatics (MI)", and "Process Informatics (PI)" research have attracted much attention as keywords for high-throughput materials discovery. Much of the MI and PI research that has been published so far emphasizes data mining, statistics, machine learning, and predictive analytics. The MGI proposed by the US government in 2011 indicated that material innovation requires "experimental tools" in addition to "computational tools" and "digital data. "Experimental tools" means the creation of big data using automated high-throughput synthesis, evaluation and analysis systems, but there are not many examples of this development. In order to promote highly accurate materials discovery, a three-way relationship is important. And in order to improve the overall speed of exploration, all processes need to be accelerated. In the production of ceramic powder, which is also the synthesis process of dielectrics, the crystalline phase and grain size vary depending on the raw material manufacturer, purity, sintering atmosphere, sintering temperature, and other conditions. These have a great impact on the physical properties. It is important to create a data set that includes not only composition, phase identification results, lattice parameters, and physical property data, but also detailed experimental process conditions. If the data can be collected in a unified high-throughput experimental environment, it will greatly contribute to data interpolation in MI and PI research. We expect that combining the data obtained from high-throughput automated experiments with the data-driven research found in MI and PI will accelerate more accurate research. To achieve this goal, we are continuing to work on the development of combinatorial technology, a high-throughput synthesis, evaluation, and analysis technique for multi-component ceramic powders. For the preparation of powder samples consisting of a vast number of combinations, we have developed two methods based on wet processes.1)2) These devices allow us to synthesize 100 samples per day, reducing the amount of raw material used per sample to about 1/1000. This has the advantage of not only increasing the high-throughput but also reducing the excessive use of raw materials. After the library is created, the crystal structure needs to be evaluated. In recent years, the high-speed detectors of experimental X-ray diffractometers have made it possible to obtain X-ray diffraction patterns in a short time. In 2014, Gregoire et al. developed a technique to simultaneously collect synchrotron XRD and XRF data for a group of multi-component thin film samples. They developed a technique to collect the data and showed that they could collect at least 5000 patterns per day in less than 18 seconds per sample3). Furthermore, Kusune et al. matched the synchrotron thin-film XRD patterns with the Inorganic Crystal Structure Database (ICSD) using an on-the-fly method, and greatly improved the classification accuracy of structural phases overlapping in the three-component reaction diagram.4) The use of XRD patterns for informatics is often discussed between phase information, lattice constants and physical properties, but crystallographic information such as fractional coordinates should be added to the data set if genomic machine learning is to be promoted. In order to obtain the diffraction intensities for high-throughput synthesized samples, which are necessary for structural refinement, collaboration with synchrotron radiation is required. We propose a high-throughput crystallographic information collection process combined with efficient measurement tools and an automated structure refinement program.5,6) 1) K. Fujimoto et al., Measurement Science and Technology 16 41-45 (2005).2) K. Fujimoto et al., Solid State Ionics 177 2639-2642 (2006).3) J. M. Gregoire et al., Journal of Synchrotron Radiation, 21(6), 1262-1268 (2014).4) Aaron Gilad Kusne et al., Scientific Report, 4: 6367 (2014).5) A. Aimi et al., ACS Combinatorial Science 22(1) 35-41 (2020).6) K. Fujimoto et al., ACS Combinatorial Science 22(12) 734-737 (2020).