PITTSBURGH (Dec. 6, 2019) —In science fiction stories from
“I, Robot” to “Star Trek,” an android’s “positronic brain” enables it to
function like a human, but with tremendously more processing power and speed.
In reality, the opposite is true: a human brain - which today is still more
proficient than CPUs at cognitive tasks like pattern recognition - needs only 20 watts of power to
complete a task, while a supercomputer requires more than 50,000 times
that amount of energy.
For that reason, researchers are turning to neuromorphic
computer and artificial neural networks that work more like the human brain.
However, with current technology, it is both challenging and expensive to replicate
the spatio-temporal processes native to the brain, like short-term and
long-term memory, in artificial spiking neural networks (SNN).
Feng
Xiong, PhD, assistant professor of electrical and computer engineering at
the University of Pittsburgh’s Swanson School of Engineering, received a
$500,000 CAREER Award from the National Science Foundation (NSF) for his
work developing the missing element, a dynamic synapse, that will dramatically improve energy efficiency, bandwidth
and cognitive capabilities of SNNs.
“When the human brain sees rain and then feels wetness, or
sees fire and feels heat, the brain’s synapses link the two ideas, so in the
future, it will associate rain with wetness and fire with warmth. The two ideas
are strongly linked in the brain,” explains Xiong. “Computers, on the other
hand, need to be fed massive datasets to do the same task. Our dynamic synapse
would mimic the brain’s ability to create neuronal connections as a function of
the timing differences between stimulations, significantly improving the energy
efficiency required to perform a task.”
Current non-volatile memory devices that have been studied
for use as artificial synapses in SNNs haven’t measured up: they are designed
to retain data permanently and aren’t suited for the spatio-temporal dynamics
and high precision that the human brain is capable of. In the brain, it’s not only
the information that matters but also the timing of the information—for
example, in some situations, the closer two pieces of information are in time,
the stronger the synaptic strand between them.
By programming the conductor to conduct more electricity for
a stronger neural connection, it can function more like the synapses of the
human brain, giving more weight to items that are more closely linked as it
learns.
“The resulted change in the electrical conductance
(representing the synaptic weight or the synaptic connection strength) in the
dynamic synapse will have both a short-term and a long-term component,
mimicking the short-term and long-term memory/learning in the human brain,”
says Xiong.
Though researchers have demonstrated this kind of technology
before in the lab, this project is the first time it will be applied to an SNN.
The application could lead to the wide use of AI and revolutionary advances in
cognitive computing, self-driving vehicles, and autonomous manufacturing.
In addition to the research component of the project, Xiong
will use the opportunity to engage future engineers in his research. He plans
to develop an after-school outreach program, host nanotech workshops with the
Pennsylvania Junior Academy of Science, and welcome undergraduate engineering
majors at Pitt to engage with the research.
The project is titled “Scalable
Ionic Gated 2D Synapse (IG-2DS) with Programmable Spatio-Temporal Dynamics for
Spiking Neural Networks” and will begin on March 1, 2020.
Maggie Pavlick, 12/6/2019
Contact: Maggie Pavlick