University of Dayton researcher Tarek Taha hopes his third National Science Foundation award is a charm in his quest to develop a brain-inspired computer chip that can learn by itself and be more secure, efficient and compact than current chips.
“Anything small you need to be smart and powerful. The key is teaching the chip to learn and then apply it. One thing that differentiates us is we’re looking at learning on the chip,” said Taha, who is using a three-year, $440,000 National Science Foundation award to work toward that goal. “We want to make these systems more autonomous, or independent of outside systems.”
Modern intelligent systems such as self-driving cars typically ship data gathered throughout the day to servers at the manufacturer’s facility for processing. This process will likely move into smaller items as artificial intelligence further pervades our everyday lives, Taha explained.
“But chips like these can be more expensive in terms of energy and time, especially energy,” he said. “Plus, if you cannot connect to the Internet or share data, or need enhanced security or do not want to share data, a chip like this is important.”
"We want to make these systems more autonomous, or independent of outside systems."
Robots could also benefit from a learning chip.
“Big batteries in robots are heavy. Batteries and computing components take up most of the space in robots,” Taha said. “This can shrink the size of robots.”
Deep learning, an artificial intelligence approach that has caught fire, is at the root of making this work, according to Taha.
“Deep learning has created a mini-revolution in the industry by replacing decades-old approaches,” he said. “Deep learning involves mimicking what we think the human brain may do, teaching a system. And once you teach it, it works on its own. This latest project is to do the actual teaching.”
The U.S. Postal Service has been using deep learning to recognize handwritten digits, Taha said, adding that the multibillion-dollar deep learning industry hopes to ramp up to large-scale networks for applications like Google voice translation and others.
Taha has been a rising researcher in the artificial intelligence field for a decade.
In 2007, the National Science Foundation awarded Taha, then at Clemson University, a $400,000 CAREER award that supports junior faculty who exemplify the role of teacher-scholar through outstanding research. A few years ago, he received another NSF award to examine ways to make computers smarter by mimicking human brains.
Taha’s research uses a type of nanoscale device, known as a memristor, which retains memory without power. His group has applied for four patents for their work.
Taha’s group is in the process of designing a new computer chip that can provide the equivalent performance of an entire supercomputer, while consuming nearly 1 million times less energy.