GNoME AI unleashes millennia of materials

In an earth-shattering revelation, researchers at Google DeepMind announced a materials science breakthrough that gives mankind an unprecedented number of new pieces to the puzzle that is the universe we live in and unlocks a transformative course for all of technology as we know it.

Utilizing artificial intelligence, the DeepMind team employed a new tool that uses deep learning to dramatically speed up the process of discovering new materials. Called Graphical Networks for Material Exploration or GNoME, this technology has already been used to predict the structures for more than two million possible new materials – or, as the team at DeepMind said, "[the] findings are the equivalent of nearly 800 years' worth of knowledge.

This discovery has basically increased the number of known stable materials by a factor of 10.

Although the materials will still need to be synthesized and tested to determine their credibility, a process that may take months or even years, the latest development is expected to accelerate the discovery of new materials that can be applied to various technologies.

Energy and tech implications

Considering the current height of the energy transition, little needs to be said of the effect this may have on renewable energy and high-tech applications like energy storage, solar cells, and superconductor chips.

"While materials play a very critical role in almost any technology, we as humanity know only about a few tens of thousands of stable materials," said Ekin Dogus Cubuk, a staff research scientist at Google Brain who worked on GNoME.

That number gets even smaller when considering which materials are suitable for specific technologies, Cubuk told journalists at a briefing on Nov. 28.

"Let's say you want to find a new solid electrolyte for better batteries," he said. "These electrolytes have to be ionically good conductors but electronically bad conductors, and they should not be toxic, they should not be radioactive. Once you apply all these filters, it turns out we only have a few options that we can go with, which end up not really revolutionizing our batteries."

Typically, new stable materials are discovered through trial and error by making incremental changes to known materials or by mixing elements together in line with principles derived from the field of solid-state chemistry.

This process is often expensive and can take months or years, as human experimentation has yielded the structures of roughly 20,000 stable materials in total.

Efforts to utilize advanced computational power to predict new materials have obviously been made before, with the most significant being the Materials Project, a multinational research effort at Lawrence Berkeley National Laboratory.

This work has yielded slightly more than the total ever discovered manually, about 28,000 stable materials.

The power of GNoME

Trained using the data on material structures and their stability from the Materials Project, researchers had GNoME suggest new structures that its model determined would likely be stable. From there, established computational techniques were used to tune the materials generated by GNoME more accurately.

With this new, higher-quality data fed back into the algorithm, GNoME produced an increase of over 10,000% in the number of potentially stable materials.

Despite this absurd number, the research team took only 381,000 materials that they considered most likely to be stable from the 2.2 million and added them to the ISCD – the Inorganic Crystal Structures Database, the world's largest database of identified materials.

Right away, third-party researchers have begun synthesizing the materials to determine if the data is accurate, and so far, 736 of them have been successful.

Among these newly added materials, 528, in particular, exist as potential lithium-ion conductors that could potentially be used in new battery chemistries, while another 52,000 have a similar layered compound structure to graphene, opening up the possibility that some of these could be the basis for – well, pretty much anything graphene has been a miracle material for – from superconductors to anode material.

"We believe that some of these will be made in the lab, which will hopefully lead to very exciting applications," said Cubuk.

In addition to accurately predicting whether a material will be stable, GNoME can also predict whether it will behave as an efficient ionic conductor – an important property for batteries.

The Google DeepMind researchers are optimistic that future AI tools will be able to predict other useful properties.

"Machine learning models, when trained on a lot of data, really learn interesting aspects of quantum mechanics, and are able to generalize and make predictions about things that they were never trained on," Cubuk said. "Which makes us very excited about our next challenges, such as predicting synthesizability."

AI synthesis

On the other side of the aisle, while material generation has obviously leapt forward by nearly a millennium, it is hard to say how long it will take to go through all those materials and determine the maximum capabilities of each material compared to others for every aspect of technology.

One could say that we have suddenly become immensely backlogged with innovation, and just parsing through it may hamper innovation itself.

To help speed up the issue of verifying the materials, researchers at Lawrence Berkeley National Lab have been developing A-Lab, an automated materials synthesis system that can work 24 hours a day, seven days a week, and for as long as needed to catch up to the current list.

In a typical human-led lab, it takes much longer to make materials. "If you're unlucky, it can take months or even years," said Kristin Persson, founder of the Materials Project, at a press briefing. Most students give up after a few weeks, she said. "But the A-Lab doesn't mind failing. It keeps trying and trying."

So far, A-Lab has performed 355 experiments over 17 days and successfully synthesized 41 out of 58 proposed compounds.

DeepMind and Berkeley Lab researchers say these new AI tools can help accelerate hardware innovation in energy, computing, and many other sectors.

"Hardware, especially when it comes to clean energy, needs innovation if we are going to solve the climate crisis," said Persson. "This is one aspect of accelerating that innovation."

"This is the future-to design materials autonomously using computers, but also then to make them autonomously using these robotic labs and learn from the process," said Kristin Persson, founder of the Materials Project.