A computational neural network (a form of artificial intelligence) was able to learn how a multilayered nanoparticle scatters light by learning the relationship between the nanoparticle’s structure and its behavior, based on thousands of training examples. After the relationship was learned, the program could essentially be run backward to design a particle with the desired set of light-scattering properties. This inverse design technique could lead to a way to custom-design nanoparticles for use in displays, cloaking systems, or biomedical devices. It could also provide a means to predict the physical properties of nanoengineered materials without requiring intensive, time-consuming computation. This cloaking grenade, used for hiding troop operations from view on the battlefield, is an example of nanoparticles that reflect a particular color of light based on their exact size and composition. Work by MIT researchers provides a way to predict the light-scattering properties of layered nanoparticles — or to design particles to match a desired type of light-scattering behavior. Courtesy of Wikipedia. Researchers from Massachusetts Institute of Technology (MIT) tested their neural network on a nanophotonics system composed of layers of nanoparticles. Each layer was made of a different material and was of a different thickness. The sizes of the nanoparticles were comparable to, or smaller than, the wavelengths of visible light. The team found that the particles scattered light differently, depending on the details of the layers and the wavelength of the incoming beam. Rather than computationally calculating these effects for multilayered nanoparticles, researchers decided to see if the neural network could predict the way a particle would scatter by determining an underlying pattern that would allow it to extrapolate. “The simulations are very exact. . . . But they are numerically quite intensive, so it takes quite some time. What we want to see here is, if we show a bunch of examples of these particles, many, many different particles, to a neural network, whether the neural network can develop ‘intuition’ for it,” said researcher John Peurifoy. The neural network was able to predict reasonably well the exact pattern of a graph of light scattering versus wavelength, and in much less time than the computer simulations. Researchers found that the network needed to be trained on only a small sampling of the data to approximate the simulation to high precision. The team’s next step was to run the program in reverse, using a set of desired scattering properties as a starting point, to see if the neural network could work out the exact combination of nanoparticle layers needed to achieve the desired output. Researchers found that when the trained neural network was run backward, it performed well, compared to other standard inverse design methods. “It will actually do it much quicker than a traditional inverse design,” said professor Marin Soljacic. Once the neural network is trained, it can simulate such optical processes orders of magnitude faster than conventional simulations. Further, the trained neural network can be used to solve nanophotonic inverse design problems by using back propagation, where the gradient is analytical, not numerical. “It’s difficult to have apples-to-apples exact comparisons, but you can effectively say that you have gains on the order of hundreds of times. So the gain is very substantial — in some cases, it goes from days down to minutes,” said Peurifoy. The research was published in Science Advances (doi:10.1126/sciadv.aar4206).