Squaw Valley, Calif. For seven years a small band of investors and technophiles has gathered in the late-summer shadows of this ski resort to divine the future of the communications industry. This year's assessment: The market recovery looks real, but a generational change in chip design looms ahead.
That was the rough conclusion drawn from executives, armchair investors and pundits at Telecosm 2003 here last week. "If we don't have any 9/11-type disasters, the economy is poised for a very significant recovery," said Steve Forbes, president and chief executive officer of Forbes Inc., the conference organizer. The crowd of some 150 tech-sector attendees gave the former presidential candidate a standing ovation after he pronounced, "This recovery is real."
Despite the enthusiasm, conference goers recalled better days. "Three years ago, this conference drove stock prices," said one investor. "People would leave sessions to go out and buy stocks on their cell phones in the lobby. The mood here is the opposite of that today."
The recovery has already begun, Brian Halla, CEO of National Semiconductor Corp. (Santa Clara, Calif.), told Telecosm attendees. He touted his company's emerging portfolio of intelligent sensors, dubbed proximity signal processors, as being in the vanguard of analog componentry for the next up cycle.
Separately, cellular and 802.11 advocates both pointed to signs of growth and agreed their technologies generally will complement, not compete, with each other (see story below).
Paul Jacobs, executive vice president of Qualcomm Inc. (San Diego), said South Korea's SK Telecom has already deployed CDMA 1x EV-DO technology, the CDMA follow-up capable of delivering hundreds of kilobits of data per second, to 10 percent of its subscribers. Japan's KDDI Corp. will deploy the technology by the end of the year, he added. In the United States, carriers Sprint and Verizon also use CDMA 1x EV-DO.
"The most aggressive people are saying 25 percent of carrier revenue will be from data in 2005," said Jacobs.
Geoff Barrall, chief technology officer for BlueArc Corp. (San Jose, Calif.), said he is seeing signs of an uptick in corporate IT spending. To date, the startup has sold more than 230 of its high-end file servers for storage-area networks, he said.
Others pointed to recent quarterly reports from Intel Corp. and Network Appliance Inc. that suggest those companies are seeing a significant pickup in spending on laptop computers and enterprise networks, respectively. And one technical manager from Analog Devices Inc. said his company has logged several quarters of single-digit growth, driven in part by power management components.
Optical networks do not appear to be participating in the recovery yet, according to one presenter who noted that current use of transatlantic optical cable amounts to a fraction of one provider's lit capacity. Conference host and author George Gilder said too much government regulation and too little support for deploying broadband are holding back the U.S. market.
"South Korea has 20 times the bandwidth per capita of the U.S. and five times the bandwidth to homes and businesses," Gilder said.
As many as 73 percent of Korean homes have broadband links, compared with about 20 percent in the states, Gilder said. Nevertheless, "as broadband is rolled out in the U.S., it will occasion another nonlinear surge in traffic that will take most experts by surprise and generate another spurt of panic equipment buying," he said.
As the economy slowly rebounds from a three-year downturn, the technical community faces a generational shift, according to some Telecosm attendees. The 30-year reign of the microprocessor and the treadmill of ever-finer CMOS process technology on which it has been based is coming to an end, said some industry observers. In their place, a new generation of programmable-logic devices and "good enough" CMOS processes will serve a widening circle of applications, they said.
"It's not going to be just PLDs and FPGAs," said Carver Mead, a pioneer of VLSI design and a computer science professor at the California Institute of Technology. "There will be as many types of programmable-logic devices as there are styles of microprocessors today."
A new crop of companies will define reconfigurable interconnects, processing elements and a host of other options, Mead said. One startup, NetHarmonix Inc. (Burlington, Vt.), is already working with Sprint on a kind of software architecture that is reprogrammable on the fly, said NetHarmonix's Dan Coffey, a former IBM PLD researcher.
"It's a very fertile ground for research. The trick is finding the demanding apps that will pay the bills and get the paradigm going," said Mead.
Today's work in programmable logic represents a new wave in thinking after years of brute-force advances in CMOS and microprocessors, Mead said. "The picture we drew in 1971 [of a 150-nanometer transistor] is pretty much the picture of what's being built today," Mead said in closing remarks at the conference. "There's been a lot of problem solving, but no new ideas. Not only have we not had new ideas in the structure of how we build semiconductors, we have not seen new ideas in the architecture of devices either."
Startup BlueArc, whose file server uses a whopping 13 high-end Altera Corp. FPGAs, served as a poster child for future systems architectures that will rely on programmable logic. "I've seen a number of startups moving to this technology because they couldn't do it with ASICs or microprocessors," said BlueArc's Barrall.
"I think we are in for a sea change that's as important as the microprocessor was," said analyst Nick Tredennick, who has written on the trend and consults for QuickSilver Technology Inc., a San Jose startup developing chips for handsets and mobile equipment.
The shift comes as many companies are experiencing diminishing returns for the latest CMOS processes, said Tredennick, who noted that semiconductor foun-
dries are seeing a slowing rate of adoption for new technologies such as 130-nm processes. Tredennick said designers will turn increasingly to "good enough" trailing-edge processes, adding that 250-nm technology offers the best bang for the buck in 2003.
Indeed, Jordan Plofsky, senior vice president of Altera (San Jose), said ASIC technology is hitting a wall and provided a detailed analysis of the prohibitive costs of designing a 90-nm chip. Such designs will cost $30 million and will target a diminishing number of multibillion-dollar systems markets, he said.
"Only a few major microprocessors will migrate to 90 nm, and that will leave a lot of applications looking for a host," said Plofsky.