image: A high magnification transmission electron micrograph showing a snapshot of time during the nucleation and growth of carbon nanotubes from catalytically active iron nanoparticles supported on aluminum oxide.
Credit: Mostafa Bedewy
As phones and other electronics grow smaller and more powerful, designers must find new ways to efficiently keep them cool and connected. Increasingly, they are turning to nanomaterials: imperceptible particles that behave differently because of their size.
While essential for many emerging devices, manufacturing things in the nanoscale can be unpredictable and difficult to simulate and predict. Researchers Mostafa Bedewy, at the University of Pittsburgh Swanson School of Engineering, and Ahmed Aziz Ezzat, at Rutgers University, are seeking to advance understanding of these particles to improve nanomanufacturing. The researchers have received a $549,947 collaborative National Science Foundation (NSF) grant to study new ways of controlling the formation of alumina-supported iron nanoparticles by using machine learning (ML) to efficiently model, characterize, simulate, and predict their growth.
“By using a special type of microscope called the environmental transmission electron microscope, we can watch the process of creating nanocatalysts that act like seeds for growing ‘turfs’ of tiny structures called carbon nanotubes that are a million times smaller than grass blades,” said Bedewy, principal investigator and associate professor and Graduate Program Coordinator for materials science.
Bedewy likens the process to growing grass, or a tiny forest. “We’re putting 100 billion nanoparticles on a plot that’s one centimeter by one centimeter,” he said. “Our previous work indicates that not all of those will grow nanotubes, and our new project aims at revealing this mystery of which ones act as seeds—and why.”
Carbon nanotubes—web-like tube structures—have impressive properties. They can be stronger than steel and more conducting that copper. Also, they can dissipate heat in small devices packed with components, making them excellent for interfaces in three-dimensional electronics.
Creating ideal high-density nanotube structures, however, can be difficult.
“It’s sort of like cooking,” Bedewy said. “Controlling the chemical vapor deposition process we use to grow carbon nanotubes requires finding the right ingredients, temperatures, and conditions. But we’re talking about highly coupled physical and chemical processes at the atomic scale.”
Because these particles are so small, researchers must use in-situ environmental transmission electron microscopy (E-TEM) to observe their work. Even with advanced electron microscopes, which use beams of electrons to capture images of nanomaterials, obtaining and processing the data has traditionally been inefficient.
“Ten years ago, we used to process images manually, but it took so long to analyze a few images,” said Bedewy. “Today, with machine learning, we can collect hundreds of images per second, automate the processing, and, importantly, predict the behavior of nanoparticles at unprecedented resolutions.”
Bedewy will collaborate with Aziz Ezzat, assistant professor of industrial and systems engineering at Rutgers University and an expert in ML and spatio-temporal data science. Aziz Ezzat will develop a machine-learning-based system to automate the processing of E-TEM images and further predict the complex nanoparticle dynamics during the fabrication of nanotubes.
“Revealing the spatio-temporal dynamics of nanoparticles from large E-TEM data is a complex challenge which requires a rigorous data science treatment,” said Aziz Ezzat. “This project aims to extract scientific insights from E-TEM data through a data science lens, and to develop a powerful predictive simulation tool. It resonates well with the growing momentum to harness AI and data sciences for advancing materials research and accelerating scientific discovery.”