This new machine learning model can design materials of different elements and structures based on human-defined properties, offering a range of potential industrial benefits and use cases.
A joint study by scientists from the Low Energy Electronic Systems Interdisciplinary Research Group at Singapore-MIT Alliance for Research and Technology (SMART), Massachusetts Institute of Technology (MIT), National University of Singapore, and Nanyang Technological University have found a new way to conduct “general inverse design” with sufficiently high accuracy. This innovation is expected to further advance a rapidly growing field that could potentially allow for the use of machine learning to accurately pinpoint materials according to a specified set of desired properties.
One of the greatest obstacles in materials science and research is finding a way to create a material or compound with a specific set of characteristics and properties that is needed for a particular application or use case. To overcome this challenge, scientists have attempted to screen materials by using materials-property databases, which have since led to the discovery of compounds with user-defined functional properties. However, despite technological advancements that have enabled the use of high-performance computing, there are limitations in its practical applications, namely, the high computational cost required to perform the necessary calculations. This has prevented scientists from extensively exploring the theoretical materials space, cueing the search for a more comprehensive and efficient method to perform “materials prospecting.”
In recent years, researchers have begun to adopt “inverse design” techniques to aid their “materials prospecting” process. Unlike conventional design processes, where intuition usually guides the process, inverse design involves “reverse-engineering” new materials and compounds by inputting a set of desired properties and characteristics, after which an optimisation algorithm is used to produce a predicted solution. In the field of photonics, inverse design has attracted significant attention as an alternate solution to overcome the challenge of designing increasingly small but more powerful devices. By adopting inverse design techniques, developers can fabricate devices with the most optimal or effective properties including but not limited to chemical composition, shape, and structure.
Going one step further, SMART researchers and their collaborators have now created a viable method of “general” inverse design that is not limited to specific elements or crystal structures, but can access a wide range of elements and structures. In particular, the scientists developed a framework for the general composition- and structure-varying inverse design of inorganic crystals called Fourier-Transform Crystal Properties (FTCP).
Trained using more than 50,000 compounds in a materials database, the new algorithm learns and generalises the complex relationships between chemistry, properties, and structures to predict novel compounds or materials that possess user-targeted characteristics. Through sampling, decoding, and post-processing, the technology enabled the scientists to inverse design crystals based on user-specific properties.
“The aim of finding more effective and efficient ways to create materials or compounds with user-defined properties has long been the focus of materials science researchers. Our work demonstrates a viable solution that goes beyond specialised inverse design, allowing researchers to explore potential materials of varying composition and structure and thus enabling the creation of a much wider range of compounds. This is a pioneering example of successful general inverse design, and we hope to build on this success in further research efforts,” said Zekun Ren, lead author and Postdoctoral Associate at Low Energy Electronic Systems.
In three design cases, the scientists showed how the framework successfully generated 142 new crystals with user-defined formation energies, bandgap, and thermoelectric power factors. They then validated these predictions with simulations through density functional theory, proving its reasonably high accuracy. Additionally, the researchers demonstrated that FTCP can design new crystalline materials that are unlike known structures, proving to be an important step in advancing this nascent technology with potentially revolutionary implications for materials sciences and industrial applications.
“This is an incredibly exciting development for the field of materials research. Materials science researchers now have an effective and comprehensive tool that allows them to discover and create new compounds and materials by simply inputting the desired characteristics,” said Tonio Buonassisi, Principal Investigator at Low Energy Electronic Systems and Professor of Mechanical Engineering at MIT.
“In the next step of this journey, an important milestone will be to refine the algorithm to be able to better predict stability and manufacturability. These are exciting challenges that the SMART team is currently working on with collaborators in Singapore and globally,” added S. Isaac P. Tian, National University of Singapore graduate student and co-first author on the study. [APBN]
Source: Ren et al. (2022). An invertible crystallographic representation for general inverse design of inorganic crystals with targeted properties. Matter, 5(1), 314-335.