Physics-trained AI accelerates material development

Researchers at Chalmers University of Technology have developed a new method to speed up the design of nanophotonic optical components. By training a neural network model with a fundamental understanding of physics, they have reduced computationally intensive simulations from months to days.

The research team focuses on nanophotonics, manipulating light at the nanoscale to create advanced optical materials. These can lead to thinner and more effective lenses for cameras and eyeglasses. The work also holds potential for quantum computing, particularly in optical data transmission between quantum processors.

Previously, training AI models for these simulations required extensive data generation, with each datapoint taking up to an hour and thousands of simulations needed. This process could consume an entire month per training cycle.

The new approach, which embeds the laws of physics into the AI's learning process, significantly cuts down on the data needed. Professor Philippe Tassin notes that their calculations now take one-tenth of the time. This accelerates the design cycle for optical components, enabling faster technological advancements.