Technologists at Nanyang Technological University, Singapore have teamed up with scientists at the Massachusetts Institute of Technology and the Skolkovo Institute of Science and Technology to build a machine learning system that is capable of predicting changes to the properties of materials from straining the material.
It is hoped that the new approach will lead to the engineering of new materials with tailored properties. Potential applications include communications, information processing and energy.
For this, as a proof-of-concept study, the researchers have shown how their use of artificial intelligence can identify optimal energy-efficient strain pathways capable of transforming diamonds into more effective semiconductors.
The reason this is significant is because, as a semiconductor material is strained, the atoms within its structure are perturbed. This has the effect of changing the material properties, including how the material conducts electricity or heat or how the transmission of light occurs. This manipulation of the material is referred to as ‘strain engineering’.
This process has previously been applied to silicon processor chips. Here, as a little as a one percent strain is sufficient to enable electrons to travel faster, leading to a 50 percent increase in computer processing speeds.
The researchers had earlier demonstrated how diamond nanoneedles can be bent and stretched by up to nine percent. This was initially surprising, since diamond is the hardest natural material on Earth. The deformation of the diamond achieved was close to the theoretical limit for a diamond. This opens up the potential for applications in microelectronics and drug delivery.
The technology used to examine material properties combined machine learning and ab initio calculations to identify specific pathways of interest (Ab Initio software is a Business Intelligence platform containing six data processing products). The method invokes artificial neural networks to predict, with a high degree of accuracy, the material properties as functions of the various input strain combinations, based on only a limited amount of data.
The new approach has been described in a paper published in the Proceedings of the National Academy of Sciences. The research paper is titled “Deep elastic strain engineering of bandgap through machine learning.”
