Google DeepMind researchers have a total of 2.2 million such crystal structures It was discovered Using an artificial intelligence tool called GNoME, which was previously unknown to scientists. The newly discovered collection of theoretically stable but experimentally unverified combinations of materials is 45 times larger than anything previously discovered in the history of science.
The researchers plan to make 380,000 of the most promising structures available for other researchers to build and test their viability in a variety of fields, from solar cells to superconductors to computing power. in nature Published study It exemplifies how the time of scientific experiments can be dramatically shortened to lead to better products and processes.
“For me, material science basically means where abstract thinking meets the physical universe. It is difficult to imagine a technology that would not be improved if it contained better materials,” said Ekin Dogus Koppuk, one of the authors of the study. The researchers’ goal was to discover new crystal structures in addition to 48,000 known crystal structures – and the range of materials already discovered ranges from those known for thousands of years (bronze, iron) to more complex materials identified in recent years.
photo_camera
The DeepMind algorithm found several crystal structures that had already been discovered by human research groups – such as the diamond-like alkaline earth metal Li4MgGe2S7, or the potential superconductor Mo5GeB2.
Source: Google DeepMind
During the research, they first generated potential crystal candidates using machine learning, and then evaluated their potential stability. According to DeepMind, the number of materials found this way is consistent with nearly 800 years of experimental results. “From microchips to batteries to solar cells, the discovery of inorganic crystals has been hampered by expensive experimental processes. Our work expands the range of stable materials known to humanity by an order of magnitude,” the researchers wrote.
According to Coppock, the discovered crystal structures have excellent uses in two areas: new compounds created from them could help in the invention of multilayer materials (for example, in increasing the efficiency of renewable energy sources), and they could also be used in the field of neural computing, To make slides that reflect the work of the brain.
Another, also published in the journal Nature Recent study According to the University of California (Berkeley) and Lawrence Berkeley National Laboratory, researchers have already used the results in their experiments to create new materials: 41 new compounds were generated from 58 selected structures, representing a success rate of more than 70 percent. – and this, according to DeepMind researchers, can be improved further.
