How AI can be used in materials development

Today Josua, Expert in materials development at ExoMatter, shares insights about digital materials development in accelerating R&D, and optimizing product performance. With expertise in materials science and a deep understanding of the potential of digital technologies, Josua is at the forefront of driving innovation in research and development (R&D) processes across various industries

How can digital materials development contribute to the research and development (R&D) process in various industries?

Any company using materials can benefit from digital materials development. I don't just mean the chemical industry - any company that produces a physical good can change which materials they use to make it. Today, most companies don't bother with it, because it is way too complicated and expensive to do it. But this will change soon, as companies such as ours democratize access to digital materials development.

We have customers in the aerospace, automotive, chemical, and academic research sectors, and this is just the beginning.

Can you discuss the role of simulation and modeling in digital materials development and its impact on R&D outcomes?

Most of the time, if a new material is needed, companies rely on their employee's expertise and existing knowledge from the literature. And then, someone tries a couple of materials and they see which one works. This is all fine, but it takes a long time to gather all this knowledge and try everything out. Simulation and modelling does not replace the experts and/or the experiments. But by simulating materials, the experts get results much faster and waste a lot less resources.

How can digital materials development enhance collaboration and knowledge sharing among materials scientists working on complex projects?

A lot of this has to do with managing digital data that is already available. For example, if a new material is tested, where does the data usually end up? In most companies, it is just stored on a shared drive or saved in an excel sheet. And over time, this knowledge gets lost. Nobody knows where to find it. This is why we at ExoMatter are building a platform to manage materials data. Data from simulations and actual experiments has the most value when it is all stored in a common place and easy to find.

What are the limitations or challenges associated with digital materials development in R&D, such as the accuracy of digital models compared to real-world materials?

The quantum-chemical methods used to describe materials are always using some kind of model to describe reality. For instance, density functional theory (DFT) simulates the size and position of the atoms and the electrons around it until a minimum in energy is reached, which is then returned as a solution. But there are uncertainties in these models, depending on the models that are used. 

The great thing is, however, that simulations are good enough to narrow down the choice of materials a lot. In the end, one still has to try a couple of candidates and see which ones work in the real world. But nobody has to try hundreds or thousands of materials anymore to get lucky.

How can digital materials development contribute to reducing the cost and time associated with traditional experimental approaches in R&D?

The cost of finding a new material is driven by the amount of experiments one has to run before finding it. Running most of those experiments in the computer instead of the lab reduces the time to market significantly and gives companies a competitive edge over those that don't use such technology.

Developing a new material using simulation methods is 6-12 months faster than traditionally, and the likelihood of overlooking anything is much lower.

Are there any ethical considerations that researchers and developers need to address when utilizing digital materials development in their R&D processes?

The ideal material found in simulations may not be able to be ethically sourced, or it may be too toxic to be used. This is why we actually take these considerations into account when screening materials, and customers can opt to exclude certain chemical elements. 

It is also possible to estimate the toxicity of a new material to some extent. But this always needs to be verified in practice because the model has its limitations.

What are some of the emerging technologies or techniques that are revolutionizing digital materials development in R&D?

One aspect are methods that combine first-principles calculations with Machine Learning. One technology called ML-DFT could revolutionize materials development by reducing the computational cost of finding new materials.

Another exciting field is generative AI in the form as large language models (LLMs), which the famous Chat-GPT engine is based on. There have been attempts of treating the composition of molecules kind of like a word, and LLMs were able to predict which kind of atoms have to follow one another to get a desired result. If all of these new technologies converge into one intelligent materials engine, we have something very powerful on our hands.

Do you have some advice for material researchers on how to start with digital materials development? 

Think about your most important challenge at the moment. Do you need to find better materials? Do you have issues with cost? Sustainability? Think about why the materials you currently use might not be the perfect solution, and then prioritize one or two topics. Don’t start with something that is nice to have, but something that really has an impact on your business. Companies like ours then have the perfect starting point to take care of all the rest.