Computer errors discussed

By Eric Chima

Dozens of computer science experts from industry leaders and renowned academic institutions gathered at the University on Tuesday and Wednesday for the second annual workshop to deal with the effects of logic errors on computer systems.

The event, the System Effects of Logic Soft Errors workshop, better known as SELSE, tackled a growing problem in computer architecture. As transistors get smaller and closer together on a computer chip, they will occasionally suffer “soft errors” and lose the values they have stored, with wildly unpredictable effects. A single bit changing from a one to a zero could do nothing at all or dump millions of dollars from an account.

SELSE’s strength was its incorporation of academics and industry into one event, Program Chair Wendy Bartlett said. Such a combination is rare because computer corporations rarely want to share their research or reveal their errors, she said.

“You’ve got industry living with the problems, and the academics that are trying to solve them more abstractly, and Ph.D. researchers somewhere in the middle,” said Bartlett, who works for Hewlett Packard. “It’s a really good mix of industry and academia.”

SELSE’s presentation topics ranged from techniques for mitigating soft errors to the effects of bombarding computer chips with radiation. The keynote presentations were Intel Representative Shekhar Borkar’s speech on expanding the complexity of circuits and two moderated panels on error reduction featuring experts from around the computer industry, said Sarita Adve, University professor and local arrangements chair.

Get The Daily Illini in your inbox!

  • Catch the latest on University of Illinois news, sports, and more. Delivered every weekday.
  • Stay up to date on all things Illini sports. Delivered every Monday.
Thank you for subscribing!

The event was deliberately kept informal, Bartlett said, to emphasize communication between the presenters and guests. Speeches were frequently interrupted with questions, and “poster sessions” allowed guests to meet one-on-one with researchers that had their work summarized on large signs.

“We’re trying to make this really interactive, not just boring talks,” Bartlett said.

In all, the second annual SELSE event drew 89 registered guests from throughout the United States, Europe and Japan, an improvement over last year’s event. The workshop was held at the University both years because of the faculty’s enthusiasm for the project and the school’s history in computer science and engineering, said Conference Chair Subhasish Mitra, visiting from Stanford University.

Brian Gold, a Ph.D. student in computer architecture at Carnegie Mellon University, said SELSE did a good job of bringing the best researchers from different organizations together.

“Being in computer architecture, I go to computer architecture conferences,” Gold said. “I don’t go to conferences with all these other academic fields and different industry people. It’s nice to go to a place with a combination and get new perspectives.”

Gold said he was surprised that the industry representatives shared real data with the audience, because it involved admitting that errors occurred in their systems. Most speakers reported their findings in “arbitrary units,” a vague term that let them report on trends in errors without revealing exactly how many errors occurred.

Despite the arbitrary units, Gold said, it was clear that the number of soft errors was growing and would have to be dealt with.

“Even if you don’t know the exact numbers, if they’re growing exponentially, they’re eventually going to get you,” Gold said.