In June 2023, Microsoft launched Azure Quantum Elements to harness artificial intelligence and cloud computing to accelerate scientific research. A first result has already been announced.
Scientific research is accelerating thanks toartificial intelligence (AI) peak. Combined with the next generation Cloud, scientists expect even faster acceleration. In this context, Microsoft announced in June 2023 Azure Quantum Elements, a solution combining the power of artificial intelligence and cloud computing, designed specifically for research in chemistry and materials science. It is based on Microsoft’s investments in AI, high-performance computing and future quantum technologies. “ Our goal is to compress the next 250 years of chemistry and materials science advancements into the next 25 », announced Satya Nadella, CEO of Microsoft during this launch.
Microsoft then partnered with the Pacific Northwest National Laboratory (PNNL), a US Department of Energy laboratory that conducts AI research in several areas, including chemistry and materials science. The two organizations are combining forces to demonstrate how chemistry and materials science can benefit from this acceleration.
A first battery material obtained by AI
And a first result is there! PNNL researchers are currently testing a new battery material to act as a solid-state battery electrolyte. The AI-derived material is a solid state electrolyte which uses lithium, sodium and other elements. It could reduce lithium content by up to 70%. Exact chemistry is subject to optimization, but might not work on a larger scale, warns Brian Abrahamson, director of digital technologies at PNNL, in a press release from Microsoft.
« CThe work paves the way for a new way to accelerate the resolution of pressing problems related to sustainability, pharmaceuticals and others, while providing insight into the advances that will be made possible by the advent of quantum computing », assure Microsoft.
Learn from successes, but also from mistakes!
Usually, researchers begin their work with a review of the scientific literature on the subject. This allows them to make hypotheses about how to achieve new results and improve what has been done in the past. But since researchers rarely publish their failures, it is difficult to rely on them to move forward. Thus, several false hypotheses can be tested by different researchers, which constitutes a waste of time.
In this new approach, AI and machine learning algorithms allow all potential materials to be taken into account and, by elimination, to arrive at a new material. This virtual laboratory digitally screened 32 million potential non-organic materials, to select 18 promising candidates likely to be used in the development of batteries, in just 80 hours. Then the researchers selected the configuration which presents the best properties to play the role of electrolyte in the envisaged battery. “ These AI models hint at a future where designing a new molecule could be as easy as asking Bing Image Creator to paint a picture “, said Matthias Troyer, researcher at Microsoft and vice president of Quantum at the launch last June.
The cloud, as a solution to endless waits
Most of the sorting was done via AI, which represents around 90% of the calculation time. “ The HPC part represented 10% of the calculation time, on a set of molecules already targeted », Shares Microsoft. If it is possible to save a lot of time in high-performance computing, it is thanks to cloud computing, the cloud having the advantage of always being accessible. Usually, researchers must reserve their turns to use the supercomputers of universities or research institutes, and can wait several weeks before having a slot for calculations. “ We believe the cloud is a great resource for improving the accessibility of research communities “, explains Brian Abrahamson.
The project is currently at the experimental stage. The material was successfully synthesized and transformed into battery prototypes which will undergo laboratory testing. Brian Abrahamson assures: “ We are on the cusp of the maturity of artificial intelligence models, the computing power needed to train them and make them useful, and the ability to train them in specific scientific fields with specific intelligence. We are entering a period of unprecedented acceleration. »