Princeton-led center focuses on the future of computer system design
Posted April 28, 2010; 08:00 a.m.
Princeton engineers are leading a research center that brings together scientists and engineers from multiple universities to determine how to improve the speed, reliability and energy-efficiency of tomorrow's computer systems while reducing their cost.
The Gigascale Systems Research Center, a consortium of dozens of researchers across 15 universities, will focus on solving a range of technical hurdles that are emerging as computing demands strain the capabilities of current processor design technology. The scientists will explore new designs, programming techniques and applications of computing.
"We are asking what computer processors will look like in 10 years and beyond, as well as how will they be programmed and what applications will they be used for," said Sharad Malik, the George Van Ness Lothrop Professor in Engineering at Princeton, who was named director of the center last fall. "This gives us the opportunity to come up with a fresh agenda, to try out creative ideas and take more risks."
Princeton researchers are leading the Gigascale Systems Research Center, a consortium of 37 researchers across 15 universities that will search for ways to make computers more powerful, energy-efficient and reliable. (Photo: Frank Wojciechowski)
The center is part of the Focus Center Research Program, a collection of six research centers that collectively receive about $40 million in funding from the federal government's Defense Advance Research Project Agency (DARPA) and a coalition of U.S. semiconductor, computer systems and defense companies, including IBM, Intel and Texas Instruments. The Princeton proposal was successful in winning one of the six centers in the most recent phase of the program, which started last fall.
Each of the centers addresses different aspects of the design and manufacture of computer systems, ranging from a center based at the University of California-Los Angeles that focuses on materials and manufacturing processes at the nanoscale level, to a new center at the University of California-Berkeley that focuses on complex integrated systems, such as the data centers at the heart of Internet companies like Google.
The Princeton-led center will concentrate on advancing both the hardware and software of future computer systems across a range of applications.
A major challenge in advancing the computation capabilities of computers is limits on their power consumption. This holds for large data-center computers due to energy costs, as well as for portable devices such as cell phones and laptops due to battery limitations.
Moving to "multi-core" designs, where a single processor is replaced by a collection of cooperating processors, helps with this problem. But there are significant challenges associated with expanding this basic idea to the thousands of cores needed in the future. The challenges are both in the design of these systems as well as the ability to program them effectively.
The center also will seek to make computing systems more reliable. With programs running simultaneously on multiple core processors, the failure of even one processor can crash an entire application. "We want to build systems that are reliable even though their parts may not be reliable," Malik said.
Other Princeton faculty taking part in the research include Margaret Martonosi and Naveen Verma from the Department of Electrical Engineering, and David August and Kai Li from the Department of Computer Science.
The other universities participating in the center are Carnegie-Mellon University, Columbia University, the Georgia Institute of Technology, Harvard University, the Massachusetts Institute of Technology, Stanford University, the University of California-San Diego, the University of California-Santa Barbara, the University of Illinois-Urbana/Champaign, the University of Massachusetts-Amherst, the University of Michigan and the University of Pennsylvania.
Additional information regarding the center can be found on its website.