Scientists in Sweden have built a physics-trained neural network that slashes simulation time for optical materials by 90%. The breakthrough could accelerate development of everything from thinner camera lenses to quantum computer components.
What happens when you give an artificial intelligence a crash course in the laws of physics before it even starts learning? According to researchers at Chalmers University of Technology in Sweden, the AI becomes dramatically more efficient — completing in three days work that used to take an entire month.
A team led by Philippe Tassin, a professor in the Department of Physics and Astronomy, published findings in Laser & Photonics Reviews describing a neural network that has been pre-loaded with fundamental principles of electromagnetism. The result is a so-called “super-brain” capable of simulating the optical properties of advanced nanomaterials at a fraction of the computational cost previously required.
Why Training AI on Physics Is a Game Changer
Neural networks typically learn patterns by processing enormous volumes of data. In the field of nanophotonics — where light is manipulated at scales smaller than a single wavelength — generating that data is painfully slow. Each simulation data point can take anywhere from 10 minutes to an hour to produce, and a full training run might require up to 40,000 individual simulations.
Lead author Viktor Lilja, a doctoral student in the Department of Physics and Astronomy, described the bottleneck bluntly.
“It might take us a whole month to generate enough data to train the neural network. Then if you realise that you need to add more things, it can take another month,” Lilja said in a news release.
The team’s solution was to stop making the neural network figure out basic physics on its own. Instead of letting it rediscover the rules of electromagnetism from raw data, the researchers embedded those rules directly into the network’s architecture before training began. The network arrived at its first lesson already knowing how light must behave — and that head start made all the difference.
“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,” added Tassin.
The speed improvement is striking: a simulation workflow that previously demanded 30 days of computation can now be completed in roughly three days.
Smarter Predictions, Fewer Obvious Errors
Beyond raw speed, the physics-informed network also produces more reliable results. Because it operates within the constraints of known physical laws, it is less prone to generating predictions that violate basic principles — a common failure mode for data-only neural networks tackling complex physical systems.
Tassin noted that even deep expertise in electromagnetism has its limits when it comes to understanding novel materials intuitively.
“I know electromagnetism’s equations inside out and I teach them, but I still can’t draw all the conclusions that the neural network can. The physics is so complex that I don’t understand the properties of a material just by looking at it — but the computer does,” he said.
Once trained, the network can analyze any given material structure and return its optical properties almost instantly. As Lilja put it, “Once we’d trained the network, we could ask it to examine any structure at all and get the optical properties in a millisecond. With these new networks, we get better estimates and avoid obvious errors.”
From Eyeglass Lenses to Quantum Computers
The practical applications of faster nanophotonics simulation span a surprisingly wide range of consumer and cutting-edge technology. On the everyday end, the research could help engineers design camera and eyeglass lenses that are lighter, thinner and optically superior to current models by using engineered artificial materials rather than conventional glass.
On the frontier end, the Chalmers team is collaborating with colleagues in the university’s Department of Microtechnology and Nanoscience — where Sweden’s first large-scale quantum computer is under development — to explore whether specially designed nanostructured materials could guide light between quantum processors. The concept relies on mechanically compliant photonic crystals, tiny man-made structures with an exceptionally high capacity to reflect light, potentially allowing quantum computers to communicate using optical frequencies over both short and long distances.
Tassin summarized the immediate payoff simply: “Now that we can work so much faster, we can speed up design development for optical components.”
Why It Matters for Students and Researchers
For students studying physics, computer science, or engineering, this research offers a compelling illustration of how domain knowledge and machine learning can reinforce each other. Rather than treating AI as a black box that learns everything from scratch, the Chalmers approach shows that embedding known scientific principles into a model’s structure — a technique sometimes called physics-informed machine learning — can yield outsized gains in both speed and accuracy.
The study also underscores a broader trend in scientific computing: the most powerful AI tools of the near future may not be the ones trained on the most data, but the ones that arrive pre-equipped with the right conceptual framework. For students entering fields where simulation is central — materials science, optics, climate modeling, drug discovery — that is a principle worth understanding early.