Making complex structures out of nanoparticles or polymers, be they for photonic computing or solar cells, typically involves a lot of expensive and time-consuming trial and error in the lab. Theorists hope to simplify the process by developing computer models that will generate recipes that always come out right, but so far, the ones that they’ve made have been too complex to realize in the lab. Now, in the hope of making these algorithms useful to chemists, computer scientists at Microsoft have simplified a model that creates recipes for self-assembling materials.
The new Microsoft models, described this week in the Proceedings of the National Academy of Sciences, are intended to speed the design of new self-assembled structures. Using trial and error, materials scientists have employed nanoparticles to make structures on what’s called the mesoscale. These ordered arrangements of nanoscale particles can have remarkable optical, electrical, and other properties but are difficult to create. “Theory there is sorely lacking,” says Mila Boncheva, a senior scientist at Firmenich, in Geneva, who played an important role in early research on this kind of self-assembly at Harvard University. “What people are currently doing in design is mostly trial and error based on common sense.” The theoretical model is aimed at helping materials scientists figure out much more quickly what the right materials and conditions are for self-assembly of a given structure.
“If you have in mind a form or shape, the model will tell you how to get it,” says Henry Cohn, principal researcher at Microsoft Research New England, who led the work with MIT assistant professor of mathematics Abhinav Kumar. The properties of materials structured on the mesoscale are to a great extent determined by how the individual components, be they polymers or nanoparticles, are arranged relative to each other. For example, silver nanoparticles floating in solution reflect light differently depending on how closely they’re packed–a principle that’s being used to design devices for photonic computing.
Whether or not particles will assemble into a given structure is determined by the forces between them. Electrical charges, for example, play a particularly important role in determining whether two particles will attract or repel each other. The Microsoft model generates a map of how strong these forces need to be. That is, given a desired structure, what should the potential energy between each of the particles and its neighbors be? These models are called potential functions.
“It is easy to design potential functions [on a computer], and really, really hard to generate them in reality,” says George Whitesides, a professor of chemistry at Harvard University and pioneer in self-assembly. Generating these forces requires figuring out what modifications to the particles–say, adding more positively charged groups to polymers–will generate the appropriate forces between individual particles and lead to the assembly of the desired structure.
Cohn says that the aim of his work is to bridge this gap between theory and reality. Previous versions of these algorithms have generated very complex instructions for putting together these structures, stipulating that a very large number of parameters need to be met in order to get a structure to assemble. “If you’re allowed to make elaborate potential functions, you can do elaborate things” and make wonderful materials inside the computer, he says. Now the question for theorists, Cohn says, is “Can we achieve more using simpler interactions?”
The Microsoft and MIT researchers have taken an important step toward this simplification, says Salvatore Torquato, a professor of chemistry at the Princeton Institute for the Science and Technology of Materials. Their models require a much smaller number of these potential-energy relationships than did previous ones. “That takes it from very hypothetical to something more realistic to produce in the laboratory,” says Torquato. The sophistication of the Microsoft model comes in part from introducing ideas from information theory.
The next step is to work with chemists to create one of these predicted structures in the lab. “I believe the materials science of the future will be done this way,” Torquato says of computer modeling. Whitesides believes that the theorists are still far from realizing that future because it’s still unclear whether the types of functions being developed by Cohn can be used to make self-assembling structures at all, or whether some other theoretical approach will turn out to be more useful. But work on these types of algorithms, says Whitesides, “is worth pursuing, since the resulting shouting match will help define what needs to be done” to make them useful.
This new data poisoning tool lets artists fight back against generative AI
The tool, called Nightshade, messes up training data in ways that could cause serious damage to image-generating AI models.
Rogue superintelligence and merging with machines: Inside the mind of OpenAI’s chief scientist
An exclusive conversation with Ilya Sutskever on his fears for the future of AI and why they’ve made him change the focus of his life’s work.
Data analytics reveal real business value
Sophisticated analytics tools mine insights from data, optimizing operational processes across the enterprise.
Driving companywide efficiencies with AI
Advanced AI and ML capabilities revolutionize how administrative and operations tasks are done.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.