Researchers at Microsoft’s Azure Quantum Elements (AQE) team and the Department of Energy’s Pacific Northwest National Lab (PNNL) have collaborated on a project to speed up the development of experimental batteries using a combination of artificial intelligence and high-performance computing.
The traditional approach to battery development involves testing hypothesis after hypothesis until the ideal candidate is identified. However, this process can be time-consuming and inefficient. The AQE and PNNL teams took a different approach by using AI models to evaluate different materials and suggest promising combinations. Through multiple rounds of machine learning and simulation, they narrowed down the field of possibilities to 18 previously unknown compounds.
What’s unique about this project is the distribution of compute resources. Contrary to conventional wisdom, 90% of the compute resources were allocated to machine learning tasks aimed at narrowing down the options, while only 10% were used for high-precision simulation workloads. This highlights the importance of a smarter application of compute resources in solving complex problems.
One of the most successful findings from the project was a solid-state electrolyte consisting of 70% sodium and 30% lithium. Combining these elements resulted in a battery with promising energy density while utilizing a more sustainable and abundant resource. The process of synthesizing and transforming the compound into a battery took about ten hours, much faster than the compute stage.
While the identification of a new battery chemistry is a significant achievement, the real success lies in the speed at which the teams were able to accomplish it. Traditionally, battery research of this magnitude would take years, but the AQE-PNNL collaboration achieved it in a matter of weeks.
Moving forward, Microsoft and PNNL are exploring the concept of a digital twin for chemistry and material sciences, which could further accelerate the testing and development process. By creating a virtual replica to test production changes digitally, the teams hope to reduce the time required for physical prototyping and testing.
While further testing and prototyping are necessary to determine the viability of the newly identified battery chemistry, this collaboration between AQE and PNNL showcases the potential of combining AI, machine learning, and high-performance computing to speed up scientific discovery and innovation.