AI spinout Polaron to accelerate design of advanced materials


Group photo of the Polaron founders

Polaron's founders

A new Imperial spinout company will use generative AI to help companies design higher performing materials for applications such as batteries.

Polaron has been launched by a team from Imperial’s Dyson School of Design Engineering to develop generative machine learning algorithms based on Imperial research that could improve the performance of products such as batteries and wind turbines.

The spinout’s technology promises to enable manufacturers of the advanced materials used in these products to accelerate their design by modelling the complex relationships between processing parameters, the microstructures of the materials produced, and the performance of the resulting products. 

AI and the materials ‘holy trinity’

Complex manufacturing processes are currently beyond the capability of today’s best computer simulations. But our generative AI approach can ‘learn’ the process-structure-performance relationships directly from microstructural image data. Dr Steve Kench CTO, Polaron

“This ‘process-structure-performance’ conundrum is the Holy Trinity of the advanced materials world,” said Dr Sam Cooper, one of the company’s co-founders. “In the case of battery electrodes, hundreds of parameters need to be carefully tuned, including the ratio of materials in the mix, the coating thickness, and the drying temperature, to name but a few.”

Because these parameters interact with each other in highly complex ways, they cannot be optimised in isolation. Each combination results in a distinct microstructural arrangement of particles, and ultimately different performance characteristics of a product, such as an electric car’s range and charging time. 

Manufacturers of advanced materials typically rely on engineers with decades of experience to carry out a costly process of trial and error to tweak the manufacturing parameters for each material. The Polaron team says that there is a strong market need for off-the-shelf generative AI design tools like theirs that can accelerate the design process based on the customer’s data.

“These complex manufacturing processes are currently beyond the capability of today’s best computer simulations,” said Dr Steve Kench, Polaron’s CTO and co-founder. “This is where our generative AI approach comes in: rather than trying to model everything, we can ‘learn’ the process-structure-performance relationships directly from microstructural image data. Amazingly, unlike other famous generative AI models such as DALL-E or Midjourney, our approach can be trained very quickly and cheaply, requiring only a small amount of training data.”

Polaron’s models are designed to be run on a manufacturer’s own hardware or cloud platform, allowing them to keep their training data and results private.

The company has strong connections to the Faraday Institution’s Multi-scale Modelling project, led by Professor Greg Offer of Imperial College London, which Dr Cooper and Dr Kench have been part of. Polaron’s technology is underpinned by Dr Kench’s PhD research, ‘Microstructural fingerprint: Machine learning for the advanced characterisation of battery materials’.

Microstructure images
Polaron’s AI platform is built on image-based generative AI algorithms, with microstructural image data at its core. Image: Polaron

Commercial opportunity

“The scale of the advanced manufacturing industry is huge, and we already see that AI is starting to have a big impact across many aspects of it,” said Isaac Squires, Polaron’s CEO and co-founder, who is also a doctoral student in the Dyson School of Design Engineering. “Generative AI has been identified by manufacturers as the top emerging AI technology for integrating into their workflows, but a tech skills shortage means they need off-the-shelf solutions.”

The scale of the advanced manufacturing industry is huge, and we already see that AI is starting to have a big impact across many aspects of it. Isaac Squires CEO, Polaron

The company is currently raising funding and is in conversation with major manufacturers to begin applying the techniques to large-scale problems. While its initial focus is on battery manufacturers, the team plans to market the platform for a variety of advanced materials applications. “Early customers are benefitting from having a hand in shaping our product – we’re directly solving their problems first,” says Dr Cooper. “We want to get these powerful AI tools into the hands of engineers so that we can help accelerate the green transition.”

Imperial has a pipeline of AI companies supporting complementary areas of research and development in advanced manufacturing. These include Monolith AI, which offers technology to help optimise the design and manufacture of components such as car doors, TOffeeAM, which optimises designs for additive manufacturing, and Quaisr, which offers a platform to help engineers and AI experts integrate their workflows. Industry collaboration and applied AI are supported by College initiatives such as I-X and the AI Network.

Dr Simon Hepworth, Director of Enterprise at Imperial, said: “Pioneering spinout companies like Polaron offer industry partners and investors a fantastic opportunity to benefit from the advanced research taking place in Imperial’s labs and to collaborate with our teams to drive forward the new possibilities presented by AI, backed by the resources we offer to make our spinout companies into commercial successes. I congratulate the Polaron team on their formation and look forward to seeing how they grow and develop.”

Image at top show the founders of Polaron, Dr Steve Kench, Isaac Squires and Dr Sam Cooper. Photo: Polaron


David Silverman

David Silverman

Click to expand or contract

Contact details

Tel: +44 (0)20 7594 8104

Show all stories by this author


Engineering-Transition-to-zero-pollution-economy, Comms-strategy-Entrepreneurial-ecosystem, Materials, Engineering-Design-Eng, Artificial-intelligence, Entrepreneurship, Enterprise
See more tags

Leave a comment

Your comment may be published, displaying your name as you provide it, unless you request otherwise. Your contact details will never be published.