As the world turns to AI to generate text, images, and videos, a startup led by a former DeepMind senior researcher is developing GenAI technology to support the production of new physical materials.
Founded by Jonathan Godwin, previously involved in DeepMind's materials research efforts, Orbital Materials is creating an AI-powered platform that can be used to discover materials ranging from batteries to carbon capture cells.
Godwin was inspired to found Orbital Materials after seeing how the technology behind AI systems like AlphaFold, DeepMind's AI that can predict the 3D structure of proteins from amino acid sequences, could be applied to materials science. states.
“Traditional methods of discovering new materials have long relied on a lengthy trial-and-error process in the laboratory, often requiring years of experimentation before success,” Godwin said. told TechCrunch in an email interview. “We felt that to take materials out of the computer and into the real world, we needed a new type of organization that included not only materials scientists but also his AI experts.”
Creating new materials, with or without AI assistance, is usually not a very intuitive process.
To achieve certain properties (e.g. lightness or stiffness), one must not only identify the corresponding physical and chemical structure, but also understand the processes (dissolution, evaporation, etc.) that reliably create that structure. is needed. Once a material is conceived, it must be stress tested under different conditions, such as extreme temperatures, depending on its intended use.
AI cannot solve all challenges inherent in materials design. (There is no substitute for real-world experimentation.) But you can save time and money by relying on calculations to plan which properties and processes will yield which types of materials.
“Technical decision makers at chemical and materials companies are struggling to develop new products because traditional methods of discovering new advanced materials are too slow and expensive to meet this demand. ,” Godwin said. “[Yet] As our economy continues to electrify and decarbonize, the demand for new advanced materials is increasing significantly. ”
Orbital Materials is not the first company to apply AI to materials research and development.
Led by former Googlers and backed by Y Combinator, Osmium AI enables industrial customers to predict the physical properties of new materials and leverage AI to improve and optimize those new materials. Masu. Several academic papers over the past decade have proposed ways to combine AI with vast databases of molecules to speed up material design workflows. DeepMind itself has been investigating his AI-derived materials, announcing last year that it had devised an algorithm to discover millions of crystals that could someday power commercial technology.
But what sets Orbital Materials apart, Godwin argues, is its unique AI model for materials science.
“We have taken a lot of inspiration from large-scale language models and the success of AlphaFold when building our dataset,” Godwin says. “What's really important with these models is that you get a lot of different types of data. Models like ChatGPT are trained on code, news articles, scientific documents, encyclopedias. This diversity It’s one of the things that gives models such amazing abilities.”
Orbital's model, called Linus, serves as the backbone of the startup's research lab in New Jersey, driving research and development in materials and chemistry. Linus was trained using simulations and large data sets of materials, from batteries and semiconductors to catalysts and organic molecules, Godwin said.
Scientists using Linus can input natural language instructions, such as “a substance that absorbs carbon dioxide well,” and the system generates a 3D molecular structure that meets the criteria. Starting with a random cloud of atoms, Linus iteratively refines the structure until he arrives at the one that best fulfills his instructions.
“[We’re] We are developing our materials pipeline in-house using a full-stack AI approach,” Godwin continued.
Like all GenAIs, Linus isn't perfect. It may produce materials that cannot be physically manufactured.But Godwin insists that have We have succeeded in developing at least one inexpensive and reliable filter for capturing carbon dioxide from the air. Orbital plans to announce further details this year.
Orbital, which is based in London and has a team of 13 people, has no plans to manufacture the filters themselves, or any other materials for that matter. Rather, the goal is to bring the material to the proof-of-concept or pilot-demonstration stage and then seek external manufacturers as partners.
To achieve that goal, Orbital recently raised $16 million in a Series A round led by Radical Ventures with participation from Toyota Ventures. The startup's total raised is approximately $21 million, and Godwin said the new capital will be used to expand Orbital's data science and wet lab teams.
“Just as AlphaFold is enabling new drugs to be discovered and brought to market faster, Orbital Materials' technology now enables us to design and commercialize new advanced materials at unprecedented speed.” Godwin said.