Tania Roy Wants to Engineer the Hardware for the Next AI Revolution
By Andrew Tie
10/28/25Pratt School of Engineering
At the Duke Langford Lecture, Roy outlined her vision for a smarter, greener solution to AI demands.
SHARE
Tania Roy Wants to Engineer the Hardware for the Next AI Revolution
When Tania Roy began her Thomas Langford Lecture on October 22, speaking to faculty from across Duke’s campus, she shared something quickly recognizable across the internet: a cat video.
The eight-second clip showed an eerily realistic AI-generated scene of an anthropomorphic feline cooking soup over a stove, complete with sizzling sounds.
Making the short video required as much energy as charging a smartphone several times over, she said. Now scale that by the billions of videos streamed, social media posts generated and AI queries processed every day. The artificial intelligence revolution happening right now isn’t just computationally powerful, it’s energy hungry.
“The question we have to ask,” Roy asked rhetorically, “is which part of this story is actually sustainable?”
For Roy, an associate professor of electrical and computer engineering at Duke Engineering, the solution to meeting the resource demands of data centers isn’t about decreasing AI usage or creating new software, it’s about changing the way AI computing operates at the hardware level.
Her lab is developing hardware capable of performing AI tasks close to where data is captured.
We have the ideas, we have the talent and we’re building the hardware that will carry us to the next great leap.
Tania RoyAssociate Professor of Electrical and Computer Engineering
From the Cloud to the Edge
Today’s devices, from smart assistants to augmented-reality glasses, rely on the cloud, even with a simple question like, “What time is it?”
The sound is captured at the device, sent to a remote server farm for processing and returned as an answer. The process, relatively speaking, consumes enormous energy and introduces lag, which limits real-time applications like autonomous navigation.
Roy’s vision is to shift that computing power from data centers to the devices themselves. The key, she explained, lies in closing the gap in where data is stored and where it’s processed. In conventional systems, information must travel between the device’s memory and a separate processing unit, often in a cloud location, wasting both time and power. By bringing computation and memory close by, her lab aims to make AI faster, more efficient and less dependent on the cloud.
AI at Duke Engineering
Duke Engineering researchers are developing and deploying the power of computing to design autonomous systems, improve communications, glean useful insights from masses of data, detect disease and improve health, and enhance security in our world and cyberspace.
The potential of hardware-level edge computing is vast: Drones that fly autonomously deep into wildfire zones, smart glasses that translate language on the spot or robotic dogs that help visually impaired people navigate sidewalks—all with far less energy than today’s AI systems require.
Her group has designed pixel-level components for digital photography that not only capture light but also store and compute from it, a feat that could allow cameras to identify objects, detect motion or sharpen images before any data ever leaves the device.
Of course, reaching that potential will require significant advancements in engineering in the coming years. For Roy, that starts with the building blocks of electronics: the materials themselves. Her lab is assessing new materials to complement all the current wonders and strengths of silicon.
To explain her idea to engineering and humanities professors alike, Roy described new semiconductors like the layers of a sandwich. As the dependable foundation of modern electronics, silicon is still needed as the bread. The meat, cheese and other toppings are emerging materials with different functions stacked on top of one another to create smarter chips. These might include two-dimensional materials only atoms thick, amorphous oxide semiconductors and gallium nitride.
Roy said new challenges would arise with more materials, particularly in engineering the chips to perform for long periods in harsh conditions. It’s intricate work that connects disciplines across materials science, physics and ECE.
The Next Great Leap
To close her talk, Roy referenced one of North Carolina’s defining stories. The Wright brothers’ first flight at Kitty Hawk in 1903 sparked an innovation chain that saw astronauts land on the moon in just 66 years. The invention of the transistor in 1947 launched a similar wave, giving rise to integrated circuits, the internet and now artificial intelligence.
“Making edge AI hardware is the Manhattan Project of our generation,” Roy said. “And just like the eras that came before, it will depend on talented engineers who are ready to build.
“It’s the right time to be in semiconductors. We have the ideas, we have the talent and we’re building the hardware that will carry us to the next great leap.”
