Cheap Chips for Next Wireless Frontier
Silicon in 60-GHz band promises speedy downloads
Head north on the electromagnetic spectrum from the 5.7-gigahertz Wi-Fi band, go past dozens of dedicated satellite bands, and you’ll find 7 GHz of unlicensed bandwidth just sitting there. There’s enough bandwidth around 60GHz for a 2-gigabit-per-second communications link—fast enough to wirelessly join a high-definition DVD player and a high-definition TV, or to beam a movie to an iPod in a flash.
What’s needed for such applications, however, are radio transceiver chips cheap enough for consumer electronics. That means silicon, not the gallium arsenide used today at these frequencies.
Finally, researchers are coaxing silicon to operate in this band. Last winter, IBM engineers unveiled the first experimental 60-GHz transmitter and receiver chips. This month, researchers at the University of California, Los Angeles, are presenting three key transceiver components built in a widely available and inexpensive silicon process technology.
Millimeter-wave radios, those operating at 30 GHz and above, have been available for many years. But because of the high frequencies involved, they were built from costly, difficult-to-integrate gallium arsenide. Packaging the chips and connecting them to an antenna without losing most of the signal added to their cost.
Only recently has a millimeter-wave silicon radio in a cheap package even seemed possible. ”When we first talked about doing this in silicon, people laughed at us,” says Brian Gaucher, who headed IBM’s millimeter-wave radio effort at the Thomas J. Watson Research Center, in Yorktown Heights, N.Y. Actually, IBM does not use silicon alone. Rather, it uses a high-speed alloy of silicon and germanium. IBM’s latest silicon-germanium technology makes transistors that can switch at a rate as fast as 200 GHz. Gaucher and his colleagues built separate transmitter and receiver chips with antennas incorporated right into the plastic package, eliminating the need for signal- sapping interconnects and economizing on packaging. The chips communicated at 630megabits per second over a distance of 10 meters.
UCLA electrical engineering professor Behzad Razavi is taking a different approach from IBM’s. He’s making key parts of his transmitters and receivers using 130-nanometer and 90-nm silicon CMOS manufacturing technology—mature chip-making processes used today to make microprocessors [see photo, ” ”]. There could be two advantages to this. First, the process technology is now so common and widespread that the chips that result will no doubt be cheap. Second, as the many millions of transistors on a microprocessor attest, CMOS lets you integrate a lot of devices on the same chip. ”If I can put one antenna on a chip, I can put on four,” says Razavi.
And CMOS transmitters at 60 GHz will need all the antennas they can get. As individual transistors get smaller, they will be less able to handle the power required for RF transmission. So a likely solution will be to put multiple transmitter circuits on a chip in parallel.
”In general, for circuit design in deep submicron, it’s easier to design several low-power transmitters than one high-power one,” says Razavi. He plans to integrate a number of transceivers on the same chip and focus the resulting radio waves electronically into a beam powerful enough to reach the 10 meters needed for a personal area network.
Razavi concedes that his group is somewhat behind IBM’s, but he feels that history is on his side. In the past, radio chips, and 5-GHz Wi-Fi radios in particular, started out in nonstandard transistor technologies, he says. But, eventually, engineers found a way to get ordinary silicon CMOS to do the same job, and for less money.
In the past, the problem with switching to CMOS was that the transistors lacked raw speed. At 60 GHz things are far worse than they were at 5 GHz, and it’s not just slow transistors. Instead, the problem is an inability to predict how transistors and everything else on a chip will work. CMOS circuit models today are accurate at frequencies up to about 5 GHz; at 60 GHz everything behaves differently. Razavi notes, for example, that any interconnect on a 60-GHz chip longer than 20micrometers—that’s 20millionths of a meter—requires complex calculations to be accurately depicted.
IBM expected the same problem, but it turned out that its silicon-germanium circuit models worked surprisingly well at 60 GHz, says Gaucher. If they work as well at higher frequencies in the millimeter-wave regime, IBM could go after other big applications. Gaucher notes that Mercedes-Benz and other carmakers offer adaptive cruise control based on a 77-GHz transceiver made of gallium arsenide. Millimeter-wave imagers are also under consideration for airport security, because they can see through clothing.
However such specific applications may fare in the real world, nobody is laughing at the idea of high-speed silicon anymore. An IEEE standards group, 802.15.3c, is hard at work defining specifications for such chips in a 2-Gb/s short-range, personal area network. More than 20 companies say they intend to participate in writing the standard, including such heavy hitters as Fujitsu, Freescale, Hewlett-Packard, Intel, Philips, and Samsung.
Proposals for the standard are to be submitted by October, according to Reed E. Fisher, a senior scientist at Tokyo-based Oki Electric Industry Co. and head of the 60-GHz standard working group. Then, in several rounds of voting, the proposals will be pared down until only one is left.