This work allows scientists to accelerate the discovery of materials that prove metal-insulator transitions.
An interdisciplinary team of scientists from Northwestern Engineering and the Massachusetts Institute of Technology has used AI techniques to create new, free, and easy-to-use tools that allow scientists to increase the rate of discovery and investigation of the materials they are exhibiting a metallic insulator transition (MIT) as well as the identification of new properties that can describe this class of material.
One of the keys to making microelectronic devices faster & more energy efficient, In addition to the design of new computer architectures, it is the discovery of new materials with tunable electronic properties. The specific electrical resistance of MITs can exhibit metallic or insulating electronic behavior depending on the environmental properties.
Although some of the materials shown by MIT have already been implemented in electronic devices, fewer than 70 are known to exhibit this property, and even fewer show the performance required for integration into new electronic devices. In addition, these materials change electrically due to a variety of mechanisms, which makes a general understanding of this class of materials difficult.
“By providing a database, an online classifier, and new functionality, our work is opening new avenues to understand & discover this class of materials,” said James Rondinelli, Morris E. Fine Professor of Materials and Manufacturing at the McCormick School of Engineering & corresponding primary investigator of the study. “In addition, this work can be used by other scientists and applied to other classes of materials to accelerate the discovery and understanding of other classes of quantum materials.
“One of the key elements of our tools and models is that they are accessible to a wide audience. Scientists and engineers don’t need to understand machine learning to use it, just like you don’t need a deep understanding of search algorithms to surf the Internet, ”said Alexandru Georgescu, postdoc re-searcher in the Rondinelli laboratory who is the first co-author of the study.
The team presented their research in the article “Database, Characteristics and Machine Learning Model to Identify Thermally Driven Metal-Insulator Transitions Compounds” published July 6, 2021 in the journal Chemistry of Materials.
Daniel Apley, Professor of Industrial Engineering & Management Science at Northwestern Engineering, was Co-primary Investigator Elsa A. Olivetti, Esther, & Harold E. Edgerton Associate Professor in Materials Science & Engineering at the Massachusetts Institute of Technology, and was Co-primary Investigators.
Using their existing knowledge of MIT materials combined with Natural Language Processing (NLP), the researchers examined the existing literature to identify the 60 known MIT compounds as well as 300 materials that have similar chemical compositions but don’t show MIT . The team provided the received materials and related functions to scientists in the form of a publicly accessible database.
Using machine learning tools, the scientists then identified key properties for characterizing these materials and confirmed the importance of certain properties, such as the distances between the transition metal ions or the electrostatic repulsion between some of them, as well as the precision of the model. They also identified new, previously underestimated properties, such as how the size of different atoms from each other, or measures of how ionic or covalent the interatomic bonds are. These features have proven critical in developing a reliable machine learning model for MIT materials that has been packaged in an open access format.
“This free tool allows anyone to quickly get probabilistic estimates as to whether the material they are investigating a metal, an insulator, or a metal-insulator transition-compound,” Apley said.
The findings were published on ACS Publications.
