System Bits: Feb. 11

Modeling computer vision on human vision
University of Michigan scientists used digital foveation technology to render images that are more comprehensible to machine vision systems, while also reducing energy consumption by 80%. The effect is achieved by manipulating a camera’s firmware.

“It’ll make new things and things that were infeasible before, practical,” Professor Robert Dick says. “Instead of having to change a battery once a week, for example, it’ll work for five weeks.”

The technique could be useful in faster facial recognition and reading license plates, according to the research team. Human retinas have a small area called the fovea, which supports high-resolution vision. In digital foveation, a machine vision camera focuses on the important elements in an image, such as the number of people involved, while blurring background elements that are not crucial to machine vision tasks.

Machine vision is currently limited due to its existing architectures. Computers capture a high-resolution, uniform image with a camera, then transfer the data to an application processor. The processor runs image classification algorithms. Transferring and processing the image data takes up energy and time.

The Michigan team modified a computer camera to capture high-resolution information within a certain bounding box. The narrower data set is sent to an application processor, which assesses the information with a machine learning algorithm. If the computer requires more information, it can direct the camera to gather more detail.

Professor Dick and Ekdeep Singh Lubana concentrated on license plate recognition to demonstrate the technique, presenting their research at the 2018 International Conference on Hardware/Software Codesign and System Synthesis.

Making perovskite solar cells last longer
A team of researchers at the Georgia Institute of Technology, University of California-San Diego, and the Massachusetts Institute of Technology has come up with new knowledge about perovskite-based solar cells and how to make them operate longer than a couple of months.

“Perovskite solar cells offer a lot of potential advantages because they are extremely lightweight and can be made with flexible plastic substrates,” said Juan-Pablo Correa-Baena, an assistant professor in the Georgia Tech School of Materials Science and Engineering. “To be able to compete in the marketplace with silicon-based solar cells, however, they need to be more efficient.”

Photo credit: Rob Felt, Georgia Tech

The team published their results on Feb. 8 in the journal Science. The study was supported by the U.S. Department of Energy, the National Science Foundation, the Hellman Foundation, and the California Energy Commission.

“Perovskites could really change the game in solar,” said David Fenning, a professor of nanoengineering at UC-San Diego. “They have the potential to reduce costs without giving up performance. But there’s still a lot to learn fundamentally about these materials.”

The team used high-intensity X-ray mapping to show how adding small amounts of cesium and rubidium salt to improve the performance of lead-halide perovskites, which promise to be a useful material in solar cells.

“By looking at the composition within the perovskite material, we can see how each individual element plays a role in improving the performance of the device,” said Yanqi (Grace) Luo, a nanoengineering Ph.D. student in Fenning’s Solar Energy Innovation Lab and co-first author of the study.

“We found that uniformity in the chemistry and structure is what helps a perovskite solar cell operate at its fullest potential. Any heterogeneity in that backbone is like a weak link in the chain,” said Fenning.

Tuning graphene to induce superconductivity
Columbia University led a team in investigating methods to finely tune adjacent layers of graphene with the purpose of inducing superconductivity. Nearly a year ago, MIT researchers made the discovery that two graphene layers could conduct electricity without resistance when the twist angle between them is 1.1 degrees, known as the “magic angle.”

“Our work demonstrates new ways to induce superconductivity in twisted bilayer graphene, in particular, achieved by applying pressure,” said Cory Dean, assistant professor of physics at Columbia and the study’s principal investigator. “It also provides critical first confirmation of last year’s MIT results—that bilayer graphene can exhibit electronic properties when twisted at an angle—and furthers our understanding of the system, which is extremely important for this new field of research.”

He added, “The layers must be twisted to within roughly a tenth of a degree around 1.1, which is experimentally challenging. We found that very small errors in alignment could give entirely different results.”

The research team, which included scientists from the National Institute for Materials Science and the University of California-Santa Barbara, tested whether magic-angle conditions could be achieved at bigger rotations.

“Rather than trying to precisely control the angle, we asked whether we could instead vary the spacing between the layers,” said Matthew Yankowitz, a postdoctoral research scientist in Columbia’s physics department and first author on the study. “In this way any twist angle could, in principle, be turned into a magic angle.”

The study was supported by the U.S. Army Research Office, the Energy Department, the David and Lucile Packard Foundation, and the NSF.