Neural network generates custom-designed nanoparticles
Image credit: Dreamstime
Researchers based at Massachusetts Institute of Technology (MIT) have developed a technique for customising nanoparticles, which could pave the way for cloaking systems, as well as new biomedical devices and displays.
The team arrived at this technique by applying machine learning techniques to nanophysics.
“[We wanted to see] whether we can use some of those techniques in order to help us in our physics research,” said Professor Marin Soljacic, who is an author of the study. “So basically: are computers ‘intelligent’ enough so that they can do some more intelligent tasks in helping us understand and work with some physical systems?”
Soljacic and his colleagues built an artificial neural network (ANN) – a program that learns through processes which mirror those that occur in biological brains – capable of working out how the structure of a nanoparticle changes its behaviour. In this project, they worked with systems of nanoparticles of different materials and thicknesses “layered like an onion” inside each other.
Calculating how these structures interact with incoming light and causes it to scatter typically requires lengthy, computationally intensive simulations, with the number of days required to complete the complex calculations increasing as the number of layers grows.
The researchers hoped that their ANN could predict how light scatters from these structures by identifying an underlying pattern from which it could extrapolate.
“The simulations are very exact, so when you compare these with experiments they all reproduce each other point by point,” said John Peurifoy, an MIT student involved with the study. “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 – a neural network, whether the neural network can develop ‘intuition’ for it.”
In this case, they used thousands of examples to teach the ANN how particles’ structures change the scattering of light. The MIT team found that it did a fairly good job of predicting how light would be scattered from the structures, and took considerably less time than simulations to provide the researchers with a set of predictions.
Next, the researchers ran the program in reverse to design nanoparticles which could scatter light the ways they wanted: this process is known as inverse design. Slightly to the team’s surprise, the ANN was capable of generating custom designs with little difficulty.
“We didn’t do any special preparation for this. We said: “okay, let’s try to run it backward.” And amazingly enough, when we compare it with some other more standard inverse design methods, this is one of the best ones,” said Soljacic. “It will actually do it much quicker than a traditional inverse design.”
According to the researchers, as well as being a tool for other researchers, the ANN could be used to customise nanoparticles with particular properties; these could include particles for use in ‘cloaking’ devices, or in new displays and biomedical devices.