AI electrifies chemistry and battery researchers
3/28/2024 Energy and raw material base in transition Article

AI electrifies chemistry and battery researchers

Finding new battery materials is a time-consuming endeavour: Years often pass before promising candidates are found. A combination of artificial intelligence and high-performance computers has now succeeded in massively speeding up the process - and could also revolutionise chemical and pharmaceutical research.

Battery for electric cars digital background 3D rendering The use of AI and high-performance computing can radically accelerate the development of battery materials.

It took Alwin Mittasch four years and 20,000 experiments to find a suitable catalyst for the synthesis of ammonia. Computers were not available to the BASF researcher, who started his project back in 1909 - yet Mittasch's system was enormously effective for his time - and the result still characterises the world today: ammonia synthesis using the Haber-Bosch process is still the basis for feeding the world today.

Similar to the search for a suitable catalyst, the development of new battery materials is also a complex and time-consuming process. Traditional approaches require years of research and experimentation to find promising material candidates. However, although experiments will probably remain indispensable in the future, American researchers have recently shown that the search for suitable materials can be radically accelerated through a combination of artificial intelligence (AI) and high-performance computing (HPC) - and have joined forces with tech giant Microsoft to do so.

Millions of material combinations tested in just a few hours

AI and HPC enable scientists to find new materials faster and more efficiently. By analysing millions of material combinations and simulating their properties, AI models can identify promising candidates that can then be tested experimentally.

Microsoft and the Pacific Northwest National Laboratory (PNNL) successfully used AI and HPC to find a new battery material from 32 possible candidates in just 80 hours. The material, which uses both lithium and sodium, has the potential to improve the performance and lifespan of batteries.

Microsoft first trained various AI systems to evaluate suitable elements and suggest combinations. The algorithm identified 32 million candidates. The AI system then selected all materials that were stable. Another AI tool filtered out the candidate molecules based on their reactivity and another on their potential to conduct energy. The number of possible combinations of substances was gradually narrowed down to 800 materials. Only then was the computationally intensive high-performance computing (HPC) applied.

In the first HPC verification, density functional theory was used to calculate the energy of each material in relation to all the other states it could occupy. Molecular dynamics simulations were then used, combining AI and HPC to analyse the motions of the atoms and molecules in each material. This allowed the list to be narrowed down initially to 150 and - after a further HPC calculation of the practicality of the materials - to 23. Five of these were already known. It is only with these promising candidates that the hard work in the laboratory finally begins: the production of the materials, the manufacture of batteries and their lengthy testing. But the end result could be much more powerful batteries.

Enormous potential for chemistry

The potential of the combination of AI and HPC is enormous: it is not only battery materials that can be calculated using the process, as computer technology was generally developed and trained for chemistry and materials research. It could also revolutionise the tedious catalyst research that Alwin Mittasch and many other chemists have been conducting for over 100 years. And there is also a lot of potential in the findings from earlier times: lithium was also discovered as a battery component at the end of the 19th century - and it took almost 100 years for it to become the Li-ion battery we know today.

These are the advantages of AI-supported materials research

  • Speed: AI can shorten the search for new materials by years.
  • Efficiency: AI can analyse millions of material combinations simultaneously and thus reduce the number of experimental tests.
  • Lower costs: AI-supported materials research can reduce the costs of developing new materials.
  • New possibilities: AI can discover new material properties and combinations that would not be possible with traditional approaches.

However, AI-supported materials research is still at an early stage of development. The biggest challenges currently include the accuracy of AI models and the availability of HPC resources.

With Azure Quantum Elements, Microsoft offers a cloud computing system that was developed for chemistry and materials science research with a view to future quantum computing. The company is already working on such models, tools and workflows. These models will be improved for future quantum computers, but they are already proving useful for advancing scientific discoveries with conventional computers.


Armin Scheuermann

Armin Scheuermann

Chemical engineer and freelance specialised journalist