Panel assesses 3G pitfalls
By Brian Fuller
EE Times
(10/03/01, 12:25 p.m. EST)
SAN JOSE, Calif. — The market for third-generation (3G) mobile phones will emerge only if engineers solve thorny cost-performance-integration problems, developers find killers apps and service providers fix their business models, a panel said Tuesday (Oct. 2).
The remarks, from five panelists assembled here at the first Communications Design Conference, seem obvious enough but belie a roiling state of systems engineering in the wireless sector, which has been hard hit by a slump in consumer demand and technology misfires for next-generation offerings.
Tom Starnes, a market analyst with Gartner Dataquest (San Jose, Calif.) who moderated the panel, noted that the economic slump has probably pushed all his forecasts out a year. In an earlier revised forecast he had estimated that 700 million wireless handsets would be sold in 2005, down from a projected 1 billion. Of that total perhaps 100 million will be 3G-enabled, he said. But the Sept. 11 terrorist attacks and the weakening U.S. economy have changed forecasts.
"Unfortunately this is a moving target," Starnes said. "3G is pretty well out there before it gets accepted."
Even so, Starnes suggested and panelists largely agreed that standardization and integration needs to take place before that ramp takes off.
Chuck Brown, director of marketing for Intel Corp.'s Cellular Communications Division, noted that a contemporary 2G handset requires about 100,000-to-150,000 logic gates excluding a DSP and memory. With 3G designs using wideband CDMA, that gate count soars to more than 1 million, Brown said. More memory will be required to handle complicated protocol stacks and DSPs will be running at three times the clock rates they do now, he said.
Starnes noted that the biggest battle may be over dispersing processing power or integrating it into a system-on-chip. Some current thinking has separate applications processors, media processors and baseband processors all handling different tasks. That approach can be good at containing power consumption but often slows system performance.
Wanda Gass, a fellow with Texas Instruments Inc. (Houston), argued that the way to 3G is paved with integration — an applications platform that's programmable, a communications platform that's extensible and reference designs that offer OEM customers hardware and software solutions.
"Semiconductor manufacturers often provide pieces to the solution but the challenge is to provide all of it," she said.
Phones must be transitioned to more open architectures because the value of third-generation handsets will be based on their number of applications, content and ISVs supporting it, she added. She touted TI's current approach to wireless architecture which revolves around a TI C55X DSP and an ARM935 MCU using a traffic controller to talk between themselves and to communications and applications peripherals.
Intel's Brown said a return on investment for 3G spectrum licenses will be key but remains unfinished business at this point.
"The investment far outstrips the revenue they're getting for voice services," he said. "And there is no killer app. We know that. If you want five good applications you have to write 1,000 of them" for five to emerge as popular, he said.
While noting the importance of integration, Brown said that communications and computing functions must be handled by separate processors for security's sake. "If not, you'll have malicious apps taking down the network," he said. Apps should be written to a general-purpose processor — StrongARM in Intel's case — to create a common platform that could be exploited by many ISVs.
In addition, "for 3G to be realized, we must provide a parallel development environment" in which hardware and software are developed simultaneously, he added.
Thomas Brooks, director of marketing for StarCore's Technology Center (Atlanta), said higher integration depended on efficient code compilation. With an obvious demand for small-form-factor pc-boards to fit inside small phones, he said the compiler holds a key role in the tool chain.
"Cycles and code size are not truly independent," he said. Formulas can help cram code into the tightest piece of silicon real estate, then allow a system designer more flexibility to push hardware features into software, he added.
Paul Masters, vice president of technology at fabless semiconductor house QuickSilver Technology Inc., said even with shrinking process geometries "you're going to wind up with 50 DSPs in an array," which takes up board space and is extremely difficult to program. On the other hand, dropping a monolithic GHz-level system-on-chip into a handset design will suck down battery life.
"It all boils down to efficiency. You need to look at the transistors that are actually doing the work," Masters said. Typically 2 percent to 25 percent of transistors are doing most of the work at any one time, he said.
Without offering details, he said QuickSilver's approach is to adapt the silicon to processing needs in real-time to improve the so-called "activity rates" of transistors.
And the challenges are not limited to technology. The $150 billion that European service providers have spent for spectrum licenses has no rational return on investment strategy associated with it. And there are no killer apps in the wings, panelists agreed.
One audience member — a former construction worker — noted that one possible killer app for the construction sector would be mobile phones that allowed workers to beam video to local building inspectors, who are notoriously slow at visiting sites to approve work.
"How would they know what building you were working on?" Starnes asked.
"GPS," many audience members shouted in return.
eet.com |
