For generations, the human brain has served as a source of inspiration for artists and scientists alike. Composed of neurons, blood vessels, and glial cells, the brain governs all the functions of a human body. Millions of individual pieces come together to make a person who they are, all in a relatively small package using a minimal amount of energy. Unsurprisingly, the brain has become a model for a relatively new area of computational science.
Neuromorphic computing is an approach to computation based on the model of the human brain with widespread potential applications, from medicine to autonomous vehicle development. In order to have such far-reaching effects, neuromorphic computing has to be both efficient and scalable. Mark Dean, interim dean of the Tickle College of Engineering and Fisher Distinguished Professor of Electrical Engineering and Computer Science, hopes to address these requirements with his JDRD work.
From the outside, neuromorphic computing systems look like any other computer, according to Dean. They might even have chips like traditional computers. The content of those chips, however, makes new computational skills possible.
“On the chip you might see artificial neurons and synapses built from traditional digital logic, but you might also see new forms of devices,” he said. “This means of storing information allows them to be used like synaptic elements.”
Dean suggests this way of storing and transferring information could affect the very functionality of such computing systems, allowing them to learn and improve over time.
“Right now, computers are pretty static. You program them to do something and that’s all they do,” he said. “Neuromorphic computing would be more flexible than that. It would be able to deal with variations in information and come up with a set of insights that traditional computers just couldn’t do.”
His team is currently working to develop low-power interconnects for neuromorphic elements to support the work of his LDRD partner Raphael Pooser, Quantum Sensing Team lead at ORNL. Dean’s goal is to show that neuromorphic elements can be connected in a way that will maximize efficiency without losing functionality.
“Our expectation is that we will demonstrate how neuromorphic components can be connected together in an efficient way that minimizes power consumption and optimizes scale,” he said. “We’re hoping to show that it can be done and done well.”