Portland Business Journal: PSU joins Princeton, Penn researchers in $26M future of computing project
Author: Malia Spencer, Portland Business Journal
Posted: February 21, 2018

Read the original story in the Portland Business Journal.

Two Portland State University researchers have received a slice of a $26 million project that addresses next-generation computing.

The project is funded by a larger $200 million program of the Semiconductor Research Corp. and the Defense Advanced Research Projects Agency.

Specifically, the Portland research teams will help develop chips and systems with technology that's based on the way the human brain works. The university is receiving $1.4 million for local work led by Christof Teuscher and Dan Hammerstrom, professors in the Maseeh College of Engineering and Computer Science.

The pair will work alongside teams from Arizona State University, Georgia Institute of Technology, Pennsylvania State University, Princeton University, University of Pennsylvania and University of Southern California. The project, called The Center for Brain-Inspired Computing Enabling Autonomous Intelligence (or C-BRIC), is headed by Purdue University researchers.

Teuscher’s 16-student lab wants to reinvent computer architectures, or the set of rules that describe the larger computer system's function, organization and implementation.

“It’s not just the hardware but the way we solve problems because I think everybody realizes by now the way we have been building computers and the way we have been using them is not sustainable,” said Teuscher. These limitations include the physical challenges of maintaining Moore’s Law — the observation made by Intel co-founder Gordon Moore that the number of transistors that can fit on a chip would double about every two years — and the power consumption of traditional systems.

“Data centers are sort of a bigger scale (for power consumption) but if you go to the other scale in terms of embedded systems, biomedical devices, stuff that you could potentially swallow ... small devices also have very stringent power requirements and constraints. You basically have to feed off your body’s energy somehow,” Teuscher said. “So it goes in both extremes.”

The C-BRIC project goes beyond universities chasing academic knowledge. The Semiconductor Research Corp., the program’s main funder, is a consortium that includes universities, government and major technology companies such as Intel Corp., Samsung Electronics Corp. and Taiwan Semiconductor Manufacturing Co.

The concept of computers running similar to the brain is called neuromorphic computing. This type of computing is considered potentially more effective for complex workloads such deep learning, image recognition and artificial intelligence. Such systems are based on spiking neurons, which would take the place of the traditional logic gate in current computers. These neurons spike whenever electrical pulses are sent through.

The technical publication IEEE Spectrum noted that as these spikes occur within certain timeframes, the neurons send along one or more new spikes based on the system programming. These spikes can have weighted values instead of the traditional 1 or 0 of current computing.

Neuromorphic computing is one research avenue that Intel is exploring. The company made headlines earlier this year with the availability of a neuromorphic research chip called Loihi.

“We are interested in properties that we understand in the brain that we think are useful, take these and use them in a machine in a similar way,” Teuscher said. “But we are not married to making that a faithful copy of natural processes. ... The brain is sort of like a proof-of-concept that we know works. It’s super efficient, especially energy efficient.”

Teuscher’s portion of the project explores the entire technology stack, including the hardware, the architecture and the algorithms it would run.

“These days you simply can’t propose a new device or new chip without having things figured out at the upper levels,” he said. “It’s great to have a new device, but you can’t use it unless you have an architecture and you can run algorithms on it.”

Since chips can’t be manufactured at PSU, the team will use Teuscher’s computing cluster to run simulations. From there, information will be sent to the partners who make chips that are then tested.

The program's funding will pay for the work of five PhD students over five years. Teuscher expects the first findings to arrive after the second year of the project. The project launched in January.