Machine Learning Accelerates Metamaterial Design

Modeling technique calculates material design in 23 hours instead of 2000 years

Many white cylinders on a shiny white surface with a beam of red coming from the top.
Image courtesy of Willie Padilla, Duke University
Infrared light shining on a metamaterial whose geometric parameters were selected with machine learning. The designed metamaterials effectively absorb low energy light, providing a route to new devices that turn heat into electricity.

The Science

Metamaterials are materials made by people that have certain patterns that change how light and matter interact. As a result, they have properties not found in nature.  Researchers are increasingly using computer models to predict how light will interact with metamaterials. However, it’s hard to predict which metamaterial will produce a desired property. Scientists used machine learning techniques to analyze databases of information. The computer program predicted the ideal metamaterial design for absorbing low-energy light.  Conventional models would have taken more than two thousand years to find the best metamaterial. In contrast, the machine learning program calculated the solution in twenty-three hours.

The Impact

Machine learning is emerging as an important tool for advancing fundamental science.  In this research, scientists used a type of machine learning known as a deep neural network to model how artificial materials (metamaterials) would absorb infrared light.  The scientists designed a metamaterial that absorbs many wavelengths of infrared light but emits very few. This approach may enable scientists to develop practical thermophotovoltaic devices. These devices create electricity from heat instead of light.


Computer models of infrared optical absorption and emission from a metamaterial were used to fabricate novel metamaterials. The final design was a square array of cylindrical silicon pillars with varying radii and heights. To choose the geometric details, scientists trained a deep neural network with 18,000 individual simulations of potential designs. The network then modeled a much larger set of metamaterial designs.  Because this approach was nearly one million times faster than the modeling software that was used to train the neural network, the authors were able to calculate the properties of every relevant metamaterial design in only twenty-three hours.  After calculating the optical properties of all possible designs, a search algorithm selected the best design.  These new techniques significantly increase the viability of more complex all-dielectric metasurface designs and provide opportunities for the future of tailored light-matter interactions.


Willie J. Padilla
Department of Electrical and Computer Engineering, Duke University


Research was supported by the U.S. Department of Energy, Office of Science, Basic Energy Sciences. Individual support was provided by the Alfred P. Sloan Foundation through the Duke University Energy Data Analytics Fellowship (BH) and by the Duke University Energy Initiative (JM).


Nadell, Christian C., Bohao Huang, Jordan M. Malof, and Willie J. Padilla. "Deep learning for accelerated all-dielectric metasurface design." Optics Express 27, 27523 (2019). [DOI: 10.1364/OE.27.027523]

Related Links

Machine Learning Finds New Metamaterial Designs for Energy Harvesting, Duke University press release

Highlight Categories

Program: BES , MSE

Performer: University