AI leads the revolution in materials science! Google DeepMind's latest research published in Nature successfully predicted 2.2 million new materials

Using only one AI, we have acquired the knowledge that it took humans nearly 800 years to develop!

This is a material discovery tool newly researched by Google DeepMind. The paper has been published on Nature.

With this AI tool alone, they discovered 2.2 million theoretically stable new crystal materials, which will not only predict the accuracy of material stability It has increased from 50% to 80%, and 380,000 types have been put into testing.

Google DeepMind said that given that 28,000 stable materials have only been discovered in the past 10 years, this research is equivalent to nearly 800 years of knowledge accumulation.

Industry experts are really eye-opening at the rapid progress

According to the Financial Times, MIT professor Bilge Yildiz commented on this research:

This vast database of inorganic crystals should be filled with gems waiting to be discovered to advance solutions to clean energy and environmental challenges.

Currently, this topic has become a hot topic on Zhihu:

So what kind of AI tool is this?

What does the new tool GNoME look like

This article proposes a new tool called GNoME (Graph Networks for Materials Exploration).

The architecture of GNoME is a graph neural network (GNN), in which nodes are used to represent atoms in the crystal structure, and edges are used to represent bonding relationships# in the crystal structure. ##.

Subsequently, GNoME used a series of known stable material data sets for training, including Materials Project, Open Quantum Materials Database (OQMD), etc.

This tool discovers new material through

active learning.

First, candidate structures are generated based on known stable materials; then, GNoME will screen these candidate structures

Of course, GNoME initially screens out The structure cannot be used directly, but the structural stability needs to be verified based on density functional theory (DFT).

Subsequently, these verified structures will be fed to GNoME again as new training data to improve its prediction capabilities.

GNoME eventually discovered more than 2.2 million new stable crystal structures, which is the result of this approach

At the same time, it also showed With certain generalization ability, it can even accurately predict structures containing more than 5 unique elements.

So, what does this newly discovered 2.2 million stable crystal materials do?

What are 2.2 million kinds of crystals used for?

The most intuitive point of view is that there is hope for progress in the fields of new energy batteries (such as solar cells), superconductors, and chips.

Although GNoME can currently only calculate theoretically stable crystalline materials, once the experimental synthesis is successful, their properties can be evaluated

These newly discovered stable crystalline materials will undergo superconducting , ferroelectrics, optoelectronics and other properties, they can be used in fields such as energy, information communications and sensing.

According to reports, researchers have synthesized 736 materials in the laboratory to prove that GNoME calculated Crystals can be synthesized.

In addition, the synthesized materials may also be used as guidance for the design of new materials, or as new data sets to train and optimize other AI models.

For example, the University of California, Berkeley, and Lawrence Berkeley National Laboratory have used these discovered materials as part of their experimental work, and the paper was also published in Nature.

The team built an A-Lab and successfully synthesized 41 compounds from 58 calculated materials, with a success rate of more than 70%.

Regarding this research, some netizens are already imagining the prospect of materials taking off, such as the progress of pharmacy:

Some netizens also cue a wave of enthusiasm LK-99 gradually calmed down: Materials Science is back.

Some people hope that these discovered materials can be beneficial to the entire human race

Do you think these AI prediction materials are still good? In what fields can it be applied?

The above is the detailed content of AI leads the revolution in materials science! Google DeepMind's latest research published in Nature successfully predicted 2.2 million new materials. For more information, please follow other related articles on the PHP Chinese website!