A Google DeepMind AI has imagined 2.2 million new materials, including 700 synthesized in the laboratory

⇧ (VIDEO) You may also like this partner content

DeepMind’s GNoME AI is revolutionizing materials discovery by predicting nearly 2.2 million new structures in record time. This technological advance, which links machine learning to materials chemistry, promises faster advances in materials design – which could be quickly used to develop new types of batteries, semiconductors and solar technologies.

Thanks to advances in supercomputing and simulation, researchers can now effectively explore new materials necessary for emerging technologies, thus avoiding conjectural and inefficient approaches to creating from scratch.

In this context, theartificial intelligence in DeepMind, a subsidiary of Google, with a new tool called “Graphical Networks for Materials Exploration” (GNoME), recently made a major breakthrough. It produced 2.2 million crystal structures, 380,000 of which were stable, making them potentially useful for future technologies.

This advancement, combining cutting-edge technology and scientific research, opens up new perspectives in the areas ofrenewable energy and electronics. This could potentially lead to better solar cells, batteries, computer chips and more. Reports of these experiments were published in the journal Nature.

Deep learning is shaking up the search for new materials

DeepMind’s GNoME tool is a major innovation in the field of artificial intelligence, making it possible to predict the structure of millions of materials in record time, something unprecedented in the history of materials science. This huge forecast of 2.2 million new materials was made possible thanks to the use of advanced graphics networks, a branch ofdeep learning specializing in the analysis and modeling of complex data.

Of these materials, more than 700 have already been synthesized in the laboratory, providing concrete validation of GNoME’s accuracy and efficiency. They are currently undergoing rigorous testing to assess their properties and potential applicability in various fields. This validation step is crucial because it allows us to move from theory to practice, transforming computational predictions into tangible and useful applications.

Six examples. ranging from an alkaline earth diamond-like optical material, the first of its kind (Li4MgGe2S7) to a potential superconductor (Mo5GeB2). © Google/Deepmind

On the other hand, GNoME radically changes the way materials are discovered. Traditionally, the discovery of new materials involves expensive and time-consuming laboratory experiments, often based on trial (and a lot of error).

GNoME, on the other hand, uses a combination of two deep learning models. The first generates more than a billion structures by making modifications to existing material elements. The second ignores existing structures and predicts the stability of new materials based solely on chemical formulas. The combination of these two models allows for a much wider range of possibilities.

discovery principle

Workflow used to discover new materials. © Google/Deepmind

Once candidate structures are generated, they are filtered by DeepMind’s GNoME models. The models predict the decomposition energy of a given structure, which is an important indicator of the material’s stability. GNoME selects the most promising candidates, which undergo further evaluation based on known theoretical frameworks.

This approach not only accelerated the discovery process but also increased the accuracy of predictions. For some materials, stability prediction accuracy exceeded 80%, a remarkably high rate that demonstrates the effectiveness of AI in this area.

material forecast

Barium (blue), niobium (white) and oxygen (green) form a new material here. © Materials Design/Berkeley Lab

GNoME can be described as an AlphaFold for materials discovery, according to Ju Li, professor of materials science and engineering at Massachusetts Institute of Technology (MIT). AlphaFold, a DeepMind AI system announced in 2020, predicts protein structures with high accuracy and has since advanced biological research and drug discovery. Thanks to GNoME, the number of known stable materials has increased almost tenfold, reaching 421 thousand.

Towards AI-powered technological innovations

The materials predicted by GNoME have considerable potential for various technology industries. In particular, the 528 lithium-ion conductors identified by GNoME could be harnessed to improve batteries, including the efficiency and durability of lithium-ion batteries – essential components of electric vehicles and mobile devices.

material testing

AI-guided robots have created more than 40 new materials envisioned by the Materials Project. The GNoME data was used as an additional check on whether these predicted materials would be stable. © Marilyn Sargent/Berkeley Lab

Furthermore, the implications of these discoveries extend to semiconductors and solar cells, where improved materials could lead to increased energy efficiency and reduced costs. This advancement is particularly relevant as the global demand for cleaner, more efficient energy solutions continues to grow.

Lawrence Berkeley National Laboratory played a crucial role in putting these discoveries into practice. By integrating the materials envisioned by GNoME into its Materials Project, the laboratory has taken an important step towards making these innovations a reality.

Its autonomous laboratory, A-Lab, is a perfect example of the application of artificial intelligence and robotics in the development of new materials. A-Lab has demonstrated an impressive ability to rapidly synthesize new compounds, with 41 materials created based on 58 proposed in just 17 days. This remarkable efficiency shows how automation and AI can accelerate research and development processes, paving the way for rapid and significant advances in many areas.

Source: Nature

Leave a Comment