At TU/e and DIFFER, researchers are developing intelligent materials labs and autonomous experimentation to accelerate breakthroughs in energy, semiconductors, healthcare, and sustainable technologies.
Combine all of the available theoretical and experimental knowledge on materials, and use AI to come up with novel combinations of chemical elements and structures that can unlock new functionalities. That is the ultimate dream Shuxia Tao, Professor of Intelligent Materials Theory at TU/e’s Department of Applied Physics and Science Education, shares with Süleyman Er, Head of the Chemical Energy Department at DIFFER and member of the EIRES management team.
‘My vision is to build a unified foundation model of materials, that can predict the properties of new materials and enable their reverse design.’ This is how Shuxia Tao succinctly sums up her lifelong ambition. This idea strongly resonates with Süleyman Er: ‘If we could combine all of the available data and knowledge we have acquired over time about any kind of material, we can aim for an AI to come up with new hypotheses to surpass human intuition.’
Both researchers have an impressive track record in advancing materials research. Where Tao focuses on physics-informed AI for materials with quantum functionality, Er is a recognized leader in energy materials research, specializing in physics-based computational methods. They are united in their belief that with the rapid rise of machine learning and AI, materials science is entering a new realm, and energy technology should profit from it.
Speeding up search
Er’s group is working on novel materials for energy storage and conversion, ranging from batteries and catalysts to structural materials for nuclear reactors. ‘AI can tremendously speed up this search, and open up an entirely new realm of candidate materials. Take catalysts, for example.’ In current electrolysers to produce green hydrogen, platinum is still the most used catalyst. But that is an expensive and scarce material that is not scalable to the amounts we need to meet future demands. Alternatives for platinum can consist of the wildest combinations of different elements, in all imaginable ratios, which constitutes an enormous chemical space.
‘What’s more, catalysts are all about surface chemistry, which depends on how the material is cut. So, besides the infinite amount of possible combinations of elements, you also have millions of different surfaces to consider. With AI, we can optimize multiple requirements at one go to find and rank and the most promising candidate materials that not only have the right properties, but can also be synthesized, are affordable, and are based on abundantly available elements.’
AI for materials
With her research group Intelligent Materials Theory, Tao is also working on AI and machine learning for materials. ‘In the energy domain, I use machine learning to study quantum functionalities in materials – like interactions between electrons, ions and light in solar cells – and processes relevant to energy storage technologies, including iron-power systems.’ As part of the Advanced Materials Flagship recently launched at TU/e, the physicist is a strong advocate for setting up a virtual Intelligent Materials Lab: a digital infrastructure connecting all knowledge, data and models available on campus about any type of material under investigation by any group. She estimates that at least one third of all of the researchers on campus is working on materials in one way or another, albeit with different methods and focusing on different aspects. ‘On campus, there is a lot of data available on materials, ranging from atomic-scale characteristics and opto-electronic properties to data driven morphology and materials’ responses in different environments.
Topics span from condensed matter and semiconductors, to complex supramolecular materials and soft matter. And research ranges from the scale of single atoms to complete systems, and is aimed at applications for health, for semiconductors, and for energy. We can benefit greatly from sharing knowledge. There are clear commonalities in the required hardware, in data analysis, and in optimizing synthesis. At TU/e, we have all of the required ingredients for such a unifying lab. We now need to develop a digital infrastructure to bring all of this expertise together and benefit from the synergies.’
Tao sees great potential in AI for future energy materials research. ‘AI can be used to accelerate quantum mechanical simulations to better understand and make more powerful predictions of the complex behavior of materials at real-life scales, while remaining grounded in physics and chemistry at the atomic level. Besides for theory, AI can also be used to accelerate experiments, especially through self-driving labs.’
Autonomous experimentation
DIFFER is already taking the first steps in this direction with its ambitious Self-Driving Lab program. By merging machine learning with high-throughput computational screening for identifying molecules and materials, the institute is setting up an automated, fully autonomous lab for energy materials discovery that combines design, synthesis, characterization and chemical testing. Er: ‘We want to close this loop, and feed lessons learned from each step into the next with the help of AI. Our first efforts in this direction are aimed at electrochemistry. After having developed and demonstrated the infrastructure for this purpose, we will expand toward other materials and domains like batteries or ammonia production. We are also appointing new group leaders to fill our knowledge gaps in this chain.’
Educating AI-native materials scientists
Attracting the right talent is another priority for Tao. ‘In Eindhoven, we have plenty of experts in experimentation, but the theory side is largely missing. We need machine learning and data science experts who are trained as materials scientists, but there aren’t many of these people around.’ Er has a rather radical suggestion to fill this need. ‘In the current day and age, we should rethink our education and turn AI into a basic pillar of any engineering curriculum. Instead of educating physicists or chemists who can later decide to specialize in AI, we should flip the educational system, and start teaching every student a basic understanding of AI and how to trust and use it, before they start to specialize into a basic science domain.’
Both researchers argue that now is the time to reap what has been sown in the past. ‘Materials are everywhere. If we are able to ride the wave of AI for Materials, we can not only accelerate the energy transition, but also radically reduce the development cycles for medicines, plastics, optoelectronics, semiconductors, … while ensuring the sustainable use of scarce resources. The Advanced Materials Flagship at TU/e seamlessly connects to the ambitions DIFFER has set out in its recent strategy. EIRES acts as a connector here, bringing together different departments and parties from outside TU/e, like DIFFER. Together, we turn Eindhoven into the hotspot for future materials exploration.’
Text: TU/e, EIRES
Go to the News page.