ChatGPT for metamaterials in the making

Designing metamaterials that are both light and strong? That is how we will design materials in the future, according to researchers at TU Delft and ETH Zurich. In new research published in Nature Machine Intelligence, they introduce DiffuMeta: an AI model that works similarly to ChatGPT. But instead of generating text, it designs 3D materials. By representing shapes as mathematical sentences, the model can create completely new metamaterials that meet specific mechanical goals, such as how they bend, compress or absorb energy.

Designing metamaterials

Designing a new material is one of the most challenging problems in engineering. This is especially true for metamaterials, where geometry determines mechanical behaviour. Designing metamaterials with certain properties often requires exploring a huge number of possible shapes. Just a decade ago, discovering a new design could take an entire PhD track. Artificial intelligence is now starting to change that.

Desired properties as a starting point

Earlier, associate professor Sid Kumar and his colleague, professor Dennis Kochmann (ETH Zurich), showed that AI can be very effective in inverse design. Instead of starting with a structure and calculating its behaviour, inverse design starts from the desired properties and looks for structures that realise these properties.

New model

Now they are going a step further. With DiffuMeta, they are introducing a model inspired by large language models such as ChatGPT. Based on specific mechanical properties, it can generate entirely new metamaterials.

Translating shapes into algebra

One of the biggest challenges in applying AI to material design is that neural networks do not naturally have a good understanding of geometry. "That's why we turned geometries into mathematical equations," Kumar explains. "These elements act as our words, and using the rules of mathematics as grammar, we can form sentences that are understood by our AI model."

Diffusion process

For the desired stress-strain response, the DiffuMeta AI model generates a new algebraic equation, which determines the shape of metamaterials.
The model uses a technique known as a diffusion process, similar to the technology behind AI image generators. When you ask an image model to create an image, it starts with random noise and gradually refines it into an image. DiffuMeta works in a similar way, but with mathematical expressions instead of pixels. The end result is an equation that represents a 3D geometry. Importantly, the model does not produce just one solution: it generates several different designs that all meet the same criteria, allowing engineers to choose from several options.

3D printing of metamaterials

To test whether these designs actually work, the researchers went beyond simulations. They 3D-printed several AI-generated metamaterials and tested them in the lab. The results showed that the physical samples behaved as predicted, and in this case matched the intended stress-strain properties.

Towards a universal AI design tool for materials

DiffuMeta represents an important step towards a new way of designing materials, where engineers specify what they want and AI explores the vast design space for them. The researchers now want to extend the model to other types of materials, such as polymers and piezoelectric systems. Another important goal is to reduce the amount of training data needed, making the method more efficient and more widely applicable.

Ambitious long-term vision

"The long-term vision is ambitious. We want a single AI model that can design many different types of materials, across multiple properties and applications. If successful, this approach could fundamentally change the way materials are developed," Kumar said.

Opening photo: 3D-printed metamaterial designed with DiffuMeta AI (photo: TU Delft)

Also read: High-tech sports shoes thanks to programmable mono-material