Microsoft uses AI and HPC to reduce the use of lithium in batteries

A partnership between Microsoft and the US Department of Energy’s Pacific Northwest National Laboratory (PNLL), which conducts research into innovations in sustainable energy sources, will exploit Artificial Intelligence (AI) and high-performance computing (HPC) in the cloud to accelerate discoveries in computational chemistry and materials science and thus help find a new battery compound capable of reducing the use of lithium by up to 70%.

“We believe that the intersection of AI, cloud and high-performance computing, together with the expertise of scientists, is the key to accelerating significant scientific results,” says Tony Peurrung, Deputy Director of Science and Technology at PNNL. “Our cooperation with Microsoft aims to make AI accessible to scientists. We recognize the potential of AI to reveal unexpected or unconventional material or approaches that are worth investigating. This is the first step in what promises to be an exciting journey to increase the pace of scientific discovery.”

What will be explored initially is what AI does best: synthesize data and present results quickly. To do this, the work will use Microsoft’s Azure Quantum Elements platform, which uses AI models developed specifically to aid scientific discoveries. PNNL researchers will seek to identify promising new materials and chemicals for applications in sustainable on-demand energy supply systems.

According to scientists, traditionally the first step in researching the synthesis of materials is to read all the published studies on other materials and formulate hypotheses. The next step is to test the hypotheses, usually an iterative and lengthy process. Just to cite an example, one of PNNL’s previous projects, involving a vanadium battery, took several years to develop a new material, according to Vijay Murugesan, leader of PNNL’s materials science group.

Now, PNNL scientists are testing a new battery material that was found in a matter of days, not years. With the help of Microsoft, different AI systems were trained to evaluate 32 million potential inorganic materials and suggest combinations. The AI system then found all the stable materials. Another AI tool filtered out candidate molecules based on reactivity and others based on their potential to conduct energy. In short, the 32 million candidates were reduced to 500,000, mostly new stable materials, and then to 800.

In another stage of the research, high-precision HPC resources were used with a smaller set of candidate materials. The first check made calculated the energy of each material in relation to all the other states it could be in. Next came molecular dynamics simulations that combined HPC and AI to analyze the movements of atoms and molecules in each material.

This process reduced the list to 150 candidates. Finally, the Microsoft researchers used HPC to evaluate the practicality of each material – availability, cost and other elements. This brought the list down to 23, of which five materials were already known.

So how long did this whole process take? Counting the AI-HPC combination, it took just 80 hours to discover the most promising materials for application in batteries. In previous scenarios, the bottleneck for this type of research would have been high-performance computing resources. Even in universities and research institutions, supercomputers are not usually adapted to a specific domain, and have to be shared between several scientific initiatives. In the research carried out by partners Microsoft and PNNL, HPC was responsible for only 10% of the computing power time, since a set of molecules to be tested had already been reduced in a previous phase. The other 90% of computing time was spent on the AI system that did most of the candidate selection.

The research is already at a practical stage, in which the materials have been successfully synthesized and transformed into functional battery prototypes that will undergo laboratory tests. Microsoft is also already working on digital tools to speed up the other parts of the scientific process.

The whole process, from simulating the candidates to producing a working battery, took less than nine months, a blink of an eye compared to traditional methods, according to the PNNL researchers.

Today, the majority of batteries used in various devices are made from lithium ions. This material is already relatively scarce and expensive, and demand is expected to grow five to 10 times by 2030, according to the US Department of Energy. What’s more, this type of battery has safety problems and can cause explosions or fires. To complicate matters, mining this material, and others in general, is an activity that traditionally suffers from environmental and geopolitical problems. Hence the relevance of research such as that carried out by Microsoft and PNNL.

“The results of the new batteries are just one example – proof, so to speak,” says Brian Abrahamson, PNNL’s chief digital officer. “We recognized from the start that the magic here lies in the speed of AI in helping to identify promising materials and our ability to immediately put those ideas into action in the lab. We are excited to move forward with the partnership between Microsoft and PNNL. The idea is to push the boundaries of what is possible through the fusion of cutting-edge technology and scientific knowledge.”

As Microsoft’s AI tools are trained for the field of chemistry, and not just for battery systems, they can be used for any type of materials research, such as those employed in the pharmaceutical field. In addition, the research could pave the way for other advances that will be possible with the help of quantum computing.