AI put to work in push for rapid battery development

JeanLucIchard / Shutterstock.com

ARTIFICIAL INTELLIGENCE (AI) is helping slash the time it takes to develop batteries, with Umicore and a US state laboratory both making strides through separate partnerships with Microsoft.

Umicore has signed a partnership to use Microsoft’s Azure OpenAI service in a bid to cut development times. The materials technology firm says its use of AI and machine learning has already enabled it to file AI-enabled patents for battery materials.

Umicore makes battery materials used in cars made by Volkswagen and BMW among others. Company CEO Mathias Miedreich told the Financial Times he expects AI will shave up to four years from the six year-cycle it can currently take to research new battery materials.

In order to beat the competition and protect Umicore’s intellectual property (IP), the partnership will create an isolated AI environment in which it will analyse decades of historical company data and feed in external information from simulations, experiments and images.

“With the support of Microsoft, Umicore will be the frontrunner in applying AI as a tool for our battery scientists to win time, efficiency and scale in our innovations while safeguarding our IP in this significant R&D area,” Miedreich said.

The deal follows hot on the heels of the US Department of Energy’s Pacific Northwest National Laboratory (PNNL) announcing it has worked with Microsoft to rapidly shortlist potential battery materials and identify a candidate that could reduce lithium use by 70%. This could bolster energy security for countries competing to import lithium and reduce demand for a material whose production is energy- and water-intensive.

20 years in a week

Microsoft trained a series of AI systems to identify potential battery materials and whittle them down to the most promising candidates. An algorithm first proposed 32m candidates, then subsequent AI systems filtered the candidates based on stability, reactivity, and conductivity.

By this stage the list of potential candidates had been reduced to 800 materials. These were filtered using high-performance computing (HPC) to calculate the energy of each material relative to all the other states it could be in. AI and HPC were then used to simulate the movements of atoms inside each material and then a final polish based on availability and cost arrived at 18 suggested candidates. The team estimates that they compressed what would have been 20 years of trial-and-error discovery into just 80 hours.

The Microsoft product used in this research is called Azure Quantum Elements. Other companies who have signed on to use it include Johnson Matthey, AkzoNobel, and BASF. Launching the product in June last year, Microsoft CEO Satya Nadella said “Our goal is to compress the next 250 years of chemistry and materials science progress into the next 25.”

Brian Abrahamson, PNNL’s chief digital officer, said: “When you think about the convergence of artificial intelligence models coupled with high-performance computing in the cloud, running alongside of co-pilots that are trained in specific scientific disciplines, and having all of that come to bear at the disposal of the individual researcher scientist, that's a paradigm shift.”

The team now faces the lengthier and unavoidable “human” stage of the development process that involves synthesising the material and testing its performance over many cycles to ensure it’s effective and safe. Vijay Murugesan, material sciences group lead at PNNL, said he hopes that one day digital twins could enable researchers to reduce these processes too through simulation.

AI for new drug discovery

Microsoft is not the only big tech firm looking to help companies increase the pace of their R&D by selling them AI services.

In January, Google’s Isomorphic Labs signed separate deals with pharma majors Eli Lilly and Novartis to use its AI technology to help them discover new drugs. The partnership involves using the revolutionary AlphaFold technology that can almost instantly predict the shape of a protein – a process that used to take drug researchers years to complete. The next generation of the technology will now be used to discover smaller molecule therapies including nucleic acids.

“Cutting-edge AI technologies such as AlphaFold hold the potential to transform how we discover new drugs and accelerate our ability to deliver life-changing medicines for patients,” said Fiona Marshall, president of biomedical research at Novartis.