Discovered Materials offers $1M for AI to solve materials puzzles

The five problems

With an advisory group of academics from leading universities, the company defined the following challenges:

• A theory explaining cuprate superconductivity, where certain materials show zero electrical resistance at ultra-low temperatures.
• A numerical method to predict the structure of interfaces between silicon and silicon dioxide, a pairing found in almost all semiconductors.
• A numerical method linking grain structure to thermal properties, to help design better heat-dissipating materials.
• A theory predicting why shining light on materials can change their behavior, particularly around superconductivity.
• A numerical method to improve density functional theory, a core technique for modeling atoms and molecules.

Before launching, Discovered Materials let frontier lab models work on the problems autonomously for several days each, with internet access and other resources. They made little headway. Still, the company acknowledges that more compute or a newer model could change that.

Goalposts, not a race

Sridhar says it doesn't matter whether the company, a professor or a frontier lab finds the answers. The aim is to create goalposts and track whether models are climbing in this domain. Stanford professor Felipe Jornada, who helped develop the challenges, said explaining these phenomena would improve understanding of physics and shape how electronics and energy storage are built.

Sridhar describes the approach as a new way of doing science: define the verification and the questions first, then let anyone supply the ideas.

The startup's main focus remains new semiconductor materials, through research partnerships with chip giants and its own funded work. When it announced its seed round in August, Sridhar hoped for a useful material within a year. He now says timelines have accelerated and he is more confident of finding industrially useful materials in less than that.