Artificial intelligence could help control material properties
Implementing strain to a semiconductor or crystalline material could distort the precise pattern of atoms in the structure, causing its properties to change, which could affect the material’s electrical, thermal and optical properties. By using artificial intelligence (AI), a research team found it is possible to predict and control this process.
The team – made up of researchers from MIT, USA, Skolkovo Institute of Science and Technology, Russia, and Nanyang Technological University (NTU), Singapore – found that using AI to control material properties changes could initiate research into developing advanced materials for high-tech devices.
According to the researchers, this advance could enable researchers to create materials adapted specifically for electronic, optoelectronic and photonic devices, which could be used in devices for communication, information processing and energy applications.
Strain engineering set itself aside from other industry methods, such as chemical doping, as it allows for temporary manipulation, while in comparison, chemical doping produces a permanent change.
However, the development of strain-engineered materials has been slow as there is such a vast amount of possibilities as strain can be applied in any of six ways, in three dimensions, each of these can then produce strain in and out, or sideways. As there is such a range of options and possibilities, it would quickly grow to 100 million calculations if the whole map of the elastic strain space would be to lay out, according to MIT Professor of Nuclear Science and Engineering, and Professor of Materials Science and Engineering, Dr Ju Li.
The solution is the team’s application of machine learning methods. The machine is able to give an organised way of search through the possibilities and find the right amount and direction of strain to get the desired properties for a specific purpose.
NTU President and Vannevar Bush Professor Emeritus, Subra Suresh said, ‘This work is an illustration of how recent advances in seemingly distant fields such as material physics, artificial intelligence, computing, and machine learning can be brought together to advance scientific knowledge that has strong implications for industry application.’
To study the effects of strain, the team used the bandgap – a central electronic property of semiconductors – in both silicon and diamond. By using a neural network algorithm, the researchers highly accurately predicted how different amounts and orientations of strain would impact the bandgap.
The research paper was published in Proceedings of the National Academy of Sciences.