December 21, 2016
Hai “Helen” Li will join the faculty of the Department of Electrical and Computer Engineering at Duke University in January 2017. With significant experience at both academic and corporate institutions, Li will forge a new research group in “evolutionary intelligence” dedicated to developing the next generation of computer hardware based on the fastest, most efficient example we know of—the human brain.
Since 1965, the progress of computer technology has generally followed Moore’s Law, doubling in processing power while shrinking in size every few years. That trend is about to be bucked, however, as the number of transistors that can be packed together is fast approaching fundamental physical limits.
The only way to continue the evolution of computing is to seek new methods of storing and processing information that lie outside the realm of classical transistors, memory caches and data pathways. Many researchers, including Li, are looking to the human brain for inspiration. It’s a goal that was dubbed a “Grand Challenge” by the White House in 2015; Li was among 15 scientists nationwide invited to participate in a Department of Energy roundtable discussion to determine the best paths forward.
“People have been trying to mimic the brain using conventional transistors for decades with very little progress,” said Li, who joins Duke as an associate professor after four years on the faculty of the University of Pittsburgh. “We are instead trying to use emerging technology that can emulate the behaviors of synapses to realize the high complexity and volume of neural connections.”
One example of this work is dubbed a “memristor.” As the name suggests, the evolving electrical component merges properties of memory and resistors. By taking advantage of nanoscale properties, these devices use magnetism as well as electrical currents to store as well as process information, making them much more analogous to a synapse.
With these sorts of advances in hand, Li is working to create “neuromorphic” chips that attempt to model the incredibly large and complex parallel processing power of the human brain. If the individual components can change how they connect with each other in response to stimuli, then the computer chip can actually learn from its experiences as an actual neural network does.
“We actually have some prototype chips that have demonstrated this capability recently,” said Li, who works with colleagues in HP Labs, the Air Force Office of Scientific Research and the Department of Defense on the project. “Of course they can’t think like a human being yet because they are still very small, but our technology demonstrated the intelligence and complex reasoning needed to recognize the patterns of problems.”
Li will find a strong network of collaborators already in place when she arrives at Duke. She mentions strengths in electronic systems, neuroscience, biomolecular materials, machine learning, computer architecture and technology transfer as reasons for her excitement for her move. And then there’s the proven track record between Pratt and the Duke School of Medicine in developing devices for clinical use.
“I get the impression that Duke is very ambitious and willing to take risks with new ideas that could result in significant technological advances,” said Li. “We expect to integrate the local strengths and programs so we’re able to not only deliver the basic hardware itself, but entire environments. The ultimate goal of our research is to create general systems for many diverse, specific applications that users are eager and able to access without needing specialized training.”
Li’s eye toward fast integration into actual devices at least partially comes from her experience in industry. Before joining the faculty of the Polytechnic Institute of New York University in 2009, Li held positions at Qualcomm, Intel and Seagate.
“I really benefit from my industry experience so much,” said Li, who entered the private job market after earning her PhD in electrical and computer engineering from Purdue University. “I’m always thinking about what industry wants and what is realistic to deliver. And on the other side, I’m always learning about what direction their research is going so that I can tailor my pursuits to their future goals.”
One example slightly outside of the realm of brain-inspired computing is Li’s work on emerging memory. Her research into low-power static random-access memory has been used in Intel processors. She also led a circuit design team on spintronic memory and resistive memory.
“I also introduce a lot of my students to potential exchanges and experiences in industry,” said Li. “I always bring students into meetings with industry companies to talk about their research. Not only do students get to learn about what industry is working on, they learn how companies operate, what people really think and how to keep a project moving. This helps them be more prepared for industry jobs when they graduate.”