In search of the super battery with AI
The search for suitable materials for new types of batteries is a lengthy one. Researchers have now used artificial intelligence to analyse 32 million substances - and have found what they were looking for.
It is one of the many promises of artificial intelligence: the technology should massively accelerate the search for new materials and molecules and thus help to solve some of the most pressing problems of our time. Experts are hoping for chemical blueprints for better catalytic converters, more powerful batteries and other novel substances. Now a team from Microsoft in collaboration with the Pacific Northwest National Laboratory (PNNL), a research facility of the US Department of Energy, has announced that it has reached an important milestone on the way to realising this vision. With the help of AI, they filtered out a previously unknown material from 32 million possible substances and then synthesised it in the laboratory. According to the researchers, the substance has great potential as a resource-efficient energy storage medium. The results have not yet been independently verified.
Normally, research into new chemical substances is a complex, expensive and lengthy process. It usually takes years, if not decades, to find, produce and test new compounds. For example, the development of today's widely used lithium-ion battery took around two decades. Now, the entire process - from the search for suitable material candidates to the selection, testing and production of a battery prototype - has been shortened to nine months. "We are at the dawn of a new era of scientific discovery," said Jason Zander, Vice President of Strategic Operations and Technology at Microsoft, in a press release. "Our success in finding a new battery material using AI is just one of the many examples of how our innovative approach to materials research can improve our daily lives in the future."
Microsoft's Azure Quantum Elements platform draws on and combines various AI systems, cloud computing, high-performance computers and - in the more distant future - a quantum computer. The team first trained an AI to determine useful combinations of the various chemical elements for this specific application. The algorithm then suggested 32.6 million candidates. The scientists then used another AI system to determine all materials that form a stable configuration under natural conditions. A third AI tool filtered out the molecules that could be considered as battery materials based on their reactivity and ionic conductivity. This left around 800 substances. All AI models used for this selection process are based on a graphical neural network. Such networks can process data that can be represented in graphs.
The recipe is already being tested
AI may be fast, but it lacks the accuracy required in materials research. This is why the researchers next used conventional high-performance computers to simulate the molecular dynamics of the remaining materials, for example. This reduced the list to 150 candidates. Finally, the scientists assessed the availability, costs and numerous other parameters, resulting in 23 materials, five of which were already known. All of this is said to have taken just 80 hours - apart from the training and programming effort. The final selection of the material that had the greatest chance of success in all the required points was made by the experts at PNNL.
The formulation for the solid-state electrolyte, which is now being tested at the Ministry of Energy's research facility, contains both lithium and sodium as well as several other elements. This could significantly reduce the lithium content in batteries - possibly by up to 70 per cent. It was previously thought that sodium and lithium did not harmonise particularly well with each other. As lithium is already relatively scarce and expensive and mining it has numerous negative consequences for the environment, materials researchers around the world have been searching for alternatives for some time. In addition, solid-state electrolytes are considered safer than conventional liquid or gel-like variants and offer a higher energy density. The market-ready development of such an energy storage system would therefore have enormous ecological, safety and economic benefits.
It's less about this particular battery material and more about the speed with which it could be identified.
Brian Abrahamson, head of the digital department at PNNL, discusses, however, that the material is still at an early stage of research. The exact chemical composition still needs to be optimised and it could still prove unsuitable when tested on a larger scale. "It's less about this particular battery material and more about the speed at which it could be identified," he said. This is the crucial bottleneck that is slowing down research. After all, the number of material compositions that need to be researched and analysed in order to find innovative solutions may exceed the number of atoms in the known universe.
Spectrum of Science
We are partners of Spektrum der Wissenschaft and want to make well-founded information more accessible to you. Follow Spektrum der Wissenschaft if you like the articles.
[[small:]]
Header image: Shutterstock / IM Imagery
Experts from science and research report on the latest findings in their fields – competent, authentic and comprehensible.