Millions of new materials discovered with deep learning

Gnome Tool Ai finds 2.2 million new crystals, including 380,000 stable materials that can operate future technologies

Modern technologies of computer chips and batteries to solar panels depend on inorganic crystals. To enable new technologies, the crystals must be stable, otherwise they can degrade, and every new and stable crystal can be months of painstaking experiences.

Today, in a paper published in natureWe share the discovery of 2.2 million new crystals – equivalent to nearly 800 years of knowledge. We offer GNOME graphic networks, which is a new deep learning tool that greatly increases the speed and efficiency of discovery by predicting the stability of new materials.

With Genoom, we have struck the number of technically known material for mankind. Of the 2.2 million predictions, 380,000 are the most stable, making it promising candidates for an experimental synthesis. Among these candidates are materials that have the ability to develop future transformational technologies ranging from super connections, operating giant computers, and the next generation batteries to enhance electric cars efficiency.

Genoom explains the potential for the use of artificial intelligence to discover and develop new materials on a large scale. External researchers in laboratories all over the world have created 736 of these new structures experimentally in simultaneous work. In partnership with Google DeepMind, a team of researchers at Lawrence Berkeley National Laboratory also published a second paper in nature This explains how our artificial intelligence predictions can be used to synthesize independent materials.

We have made Gnome predictions available to the research community. We will contribute to 380,000 materials we expect to be stable for the material project, which is now processing vehicles and adding them to the online database. We hope that these resources will drive forward to the front crystals, and to cancel the promise of machine learning tools as evidence of experimentation

In the past, scientists have searched for new crystals by changing known crystals or experimenting with new groups of elements-an expensive process, experimental and support that may take months to provide limited results. Over the past decade, the mathematical methods led by the project of materials and other groups have helped to discover 28,000 new materials. But so far, new methods of AI-are essentially struck in their ability to accurately predict materials that could be viable in terms of experimental. The discovery of Gnome of 2.2 million materials will be the equivalent of about 800 years and shows an unprecedented measure and a level of accuracy in predictions.

For example, 52,000 new vehicles resemble graphics that have the ability to revolutionize electronics with the development of high conductors. Previously, about 1000 of these materials were identified. We also found 528 potential Li -Ion connectors, more than 25 times from a previous study, which can be used to improve the performance of rechargeable batteries.

We release the expected structures for 380,000 materials with the highest opportunity to make success in the laboratory and use them in viable applications. In order for the material to be considered stable, it should not be degraded to similar literature of less energy. For example, carbon in a stable graphic structure compared to carbon in diamonds. Mathematically, these materials are on a convex body. This project discovered 2.2 million new crystals in accordance with the current scientific standards and reduces the convex body of previous discoveries. Among them, 380,000 is the most stable, and requires the “final” convex body – the new standard that we have set for the stability of materials.

Genoom: Harmony of graphic networks to explore materials

Genoum is the modern nerve network model (GNN). GNNS input data takes a graphic form that can be similar to communications between atoms, making GNNS particularly suitable to discover new crystalline materials.

Genoom was originally trained with data on crystals and their stability, publicly available through the material project. We used Genoom to generate new candidate crystals, as well as to predict their stability. To assess the predictive strength of our model during the gradual training courses, we have repeatedly monitored its performance using known mathematical techniques known as the DFT job theory (DFT), used in physics, chemistry and materials to understand atomic structures, which is important to assess the stability of crystals.

We used a training process called “active learning”, which greatly strengthened the performance of Genome. Genoom will generate predictions of new stable crystals structures, which were then tested using DFT. Then high -quality training data was fed to our model training.

Our research has strengthened the discovery of the stability of materials from about 50 %, to 80 %-based on Matbench Discovery, an external standard set by previous modern models. We have also been able to increase the efficiency of our model by improving the discovery rate from less than 10 % to more than 80 % – this efficiency increases may have a significant impact on the amount of the account required for each discovery.

Amnesty International “Recipes” for new materials

The Genoom project aims to reduce the cost of discovering new materials. External researchers created 736 new GNOME materials in the laboratory, indicating that our model predictions of stable crystals reflect reality accurately. We have released our database from the newly discovered crystals to the research community. By giving scientists a full catalog to the promising “recipes” of new candidate materials, we hope this will help them to test and may make the best.

The development of new technologies based on these crystals depends on the ability to manufacture them. In a paper led by our collaborators at Berkeley Laboratory, the researchers showed that the automated laboratory can quickly create new materials with automated synthesis techniques. Using materials from the materials and visions project on stability from Genoom, the independent laboratory has created new recipes for crystal structures and the synthesis of more than 41 new materials, which opens new capabilities to synthesize the materials that AI drives.

New materials for new technologies

To build a more sustainable future, we need new materials. Gnome 380,000 stable crystals have discovered the ability to develop more green technologies – from better batteries for electric cars, to the superior connectors of the most efficient computing.

Our research – and our collaborators in Berkeley, Google Research Laboratory, and the teams around the world – show the possibility of using artificial intelligence to direct the discovery of materials, experimentation and synthesis. We hope Gnome with other AI tools will help revolutionize the discovery of materials today and form the future of this field.