Researchers at Chalmers University of Technology have developed a physics-trained neural network to accelerate the design of optical components. This digital system integrates the fundamental laws of nature to reduce calculation times for nanophotonics simulations. Consequently, the time required to design components for quantum computing and optical lenses has decreased to one-tenth of previous requirements.
Specifically, the research team focused on nanophotonics, a field where light is controlled on scales smaller than a single wavelength. To bypass the physical limitations of natural optical materials, the researchers design artificial materials using computer simulations. These artificial structures can make camera lenses and eyeglass lenses thinner, lighter, and more efficient. Meanwhile, this technology also has direct applications in the development of quantum communication networks.
“When we fed the super-brain information about the laws of physics, it immediately got much smarter. Our calculations now take one tenth of the time previously required,”
Philippe Tassin, Professor at the Department of Physics and Astronomy
Previously, training a standard neural network for these simulations was highly time-consuming. Generating a single data point took up to an hour, and the system required up to 40,000 simulations. As a result, researchers spent up to a month generating training data. However, the new method bypasses this limitation by incorporating electromagnetism equations prior to training, saving significant computational energy.
How the Physics-Trained Neural Network Operates
By teaching the system the laws of physics directly, the researchers eliminated the need for the network to deduce these laws from raw data. Consequently, the data generation process that previously took 30 days now takes only three days. Furthermore, once trained, the physics-trained neural network can analyze any structure and predict its optical properties within a millisecond.
In addition to lens design, this method supports the development of quantum computing infrastructure. Researchers at Chalmers are currently building Sweden’s first large-scale quantum computer. They plan to use mechanically compliant photonic crystals to transmit optical frequencies between quantum systems over long distances. Consequently, this research could accelerate the timeline for practical quantum networking.
The study, titled “A General Framework for Knowledge Integration in Machine Learning for Electromagnetic Scattering Using Quasinormal Modes,” was published in Laser & Photonics Reviews. Notably, the Swedish Research Council and the Knut and Alice Wallenberg Foundation funded the project. The team utilized computing resources from the Swedish National Infrastructure for Computing to train the physics-trained neural network.
