BERKELEY, Calif. Tomorrow's computers need to be designed to meet the energy efficiency needs of growing data centers, and the power grid needs to evolve to be more like the Internet. That's the view of Randy Katz, a professor of computer science speaking at an annual gathering of researchers from the University of California at Berkeley.
Katz showed that computers consume far too much energy when not in use. In addition, he described the trend toward massive data centers that will consist of warehouses packed with hundreds of shipping containers, each one loaded with computers, communications, power and cooling systems.
"It used to be you designed computers at the chip level, then we moved to the board level and then racks," said Katz, in a 40-minute keynote address at the event. "Now something like a 20-foot container is the basic building block of the aggregated systems we are building, and there is a completely new set of knowledge needed to build it.
"This is the emergence of the industrialization of the information technology industry," he said describing tomorrow's million-server data centers. "This is the aluminum smelting plant of the 21st Century, and we need to think about what knowledge we need to build these things if we want to be the IT builders of the 21st Century, aware of the energy issues," he said.
Katz also described a new initiative at Berkeley called LoCal exploring the technologies needed to upgrade today's electric distribution grid. He called for an evolution toward a grid that could generate, store and switch energy flows on demand in ways similar to today's Internet.
"This will be a switching network that has energy and data flows integrated into it," Katz said. "That is not the way the current transmission architecture works."
Such an electric network could ultimately support an energy exchange where users can buy and sell power based on shifting demand. The LoCal group hopes to create some of the underlying algorithms to enable those transactions.
In a similar move, Jan Rabaey of the Berkeley Wireless Research Center said he has submitted a proposal to the National Science Foundation to research a spectrum exchange. The proposal includes participation from business and regulatory thinkers as well as engineers.
"I believe we have to go away from spectrum as an assigned resource to something that is dynamically traded," Rabaey said. "We are already experimenting with a spectrum brokerage in a limited fashion," he added.
In another research report, Kris Pister, director of the Berkeley Sensor and Actuator Center, described his group's work on printable nanowires. The group has demonstrated a technique to print high quality transistors up to ten layers thick with enough complexity to make FETs, gas sensors and diodes. The team also used the process to make crystalline nanowires for photovoltaics that showed a six-percent efficiency.
"That's darned good for a first effort," said Pister. "With this process you can print electronics on paper or metal or non-flat surfaces," he added.
In another talk, Kameshwar Poolla of Berkeley's Impact center talked about his team's work on next-generation chip manufacturing technologies. The group is trying to bridge the increasing gap between the models chip designers use and the physical realities of devices built in the fab.
"We see a breakdown in the design and manufacturing interface," he said. "It's getting just too complicated."
For example, actual circuits do not have the rectangular gates used in the models of chip designers. Instead they show significant rounding and other imperfections that can significantly impact leakage estimates. One Berkeley team was able to find a way to use data about the physical imperfections to scale devices down by 20 percent without a loss of performance, he said.