An impatient researcher is using NVIDIA GPUs to speed up his ability to model invisible materials.
These so-called “metamaterials” exist for now only on blackboards, as pioneered by theoretical physicist Sir John Pendry, of the Imperial College of London.
But the notion has Attique Dawood, an instructor at the National University of Computer and Emerging Sciences, in Islamabad, buzzing. And others are equally excited.
Letting You See Things That Are Small, Or Not At All
Such materials could, in theory, be used to bend light around an object, making possible the ability to vanish in plain sight – a trick you may have seen in books and movies such as “Harry Potter,” “Star Trek” and “Lord of the Rings.”
They could also be used to build lenses that show objects smaller than a single wavelength of light, letting researchers see things too small to be observed directly.
“Metamaterials offer a whole new world of possibilities not limited to invisibility,” Dawood says. “Efficient and smaller antenna designs, super lenses and structures that can do amazing things.”
These objects already exist, but only in simulations. Pendry, the researcher who kicked off the field, has already modeled invisible materials. A team of researchers in China have taken that idea even further, modeling materials that allow an observer to see out through a material without being visible.
The problem: simulating these objects soaks up a huge amount of computing power. Even a simple model can take hours to simulate using common scientific software tools such as MatLab, Dawood says.
So Dawood – an avid videogamer who has volunteered as a developer for an open-source gaming project – turned to CUDA and his NVIDIA GPU to speed things up. He chose to model two different problems with the help of GPUs. The first is a slab that soaks up electromagnetic radiation and doesn’t reflect light.
The second is even more interesting, a cylindrical cloak that bends light around it to render a person or object “wearing” the cloak invisible. This is what was first proposed and tested by Pendry.
Using GPUs Dawood found he could simulate the way light flows over materials 10-15 times faster than with CPUs alone. Better still, he found that the larger the problem size, the better the performance gain.
A Need for Speed
“It was frustrating to wait for hours,” he says. “Each simulation took about 1 to 10 hours depending on simulation size – so my decision to learn CUDA was borne out of need more than anything else.”
Next up, Dawood – who hopes to start a Ph.D. next year in metamaterials – wants to begin building clusters of GPUs and CPUs to model metamaterials even more efficiently.
Such simulations are critical to getting to the next step: actually building the stuff. “It’s the most time-consuming step and you must be absolutely sure before sending your design for manufacturing,” Dawood says.
After all, if you’re really building something invisible you’ll want to make sure you know exactly how it will behave before you wind up losing it on the factory floor.