'dozer, chuzz.
thanks for your replies.
some trade journal tidbits that might pique your keen interest ...
-chris.
-----
Electronic Engineering Times
July 26, 1999, Issue: 1071
Section: News
NewPort brings CMOS to Sonet with 2.5-Gbit/s chip
Craig Matsumoto
IRVINE, CALIF. -- While gallium arsenide continues to be the material of choice for high-speed networking chips, companies on the CMOS side find they're beginning to reach the same speeds with standard silicon chips, a development that could lead to cheaper, more-widely manufactured parts.
NewPort Communications Inc. last week announced its first product: the first straight-CMOS synchronous optical network (Sonet) transceiver for OC-48 (2.5-Gbit/second) speeds, a move that the company maintains renders gallium arsenide and silicon germanium obsolete at that speed grade. The device is shipping in sample quantities, company officials said.
But it hardly spells the end for GaAs vendors. Vitesse Semiconductor Corp. (Camarillo, Calif.), for example, admits it's investigating CMOS itself, but the company expects GaAs to be used for extremely fast parts for years to come. NewPort's is a good part, but "the concept of obsoleting other technologies is a little bit of ego," said Chris Gardner, vice president and general manager of Vitesse's telecommunications products.
Ironically, Vitesse pressed NewPort into action. NewPort wanted to keep its OC-48 parts under wraps a little longer, but Vitesse announced a backplane interconnect transceiver -- though without the jitter specifications required for Sonet compliance -- at a trade conference July 14. To calm investors who feared their pony had been beaten to the stretch, NewPort rushed its own press release to prove it was in the OC-48 game, said Anil Bedi, president and chief executive.
Next speed grade
Issues like that matter, because NewPort's chosen market is in constant flux. Its original entry point was to be OC-12 (622 Mbits/s), but as prices fell, NewPort's eye had to turn to the next speed grade. "The '12 was an old part of the market," Bedi said. But OC-48 "is the sweet spot, and it's a challenge even for the other technologies."
NewPort's transceiver, the NP2510, combines the physical-layer and framing portions of a Sonet transceiver onto one chip. The key to the technology lies in the analog expertise of founders Armond Hairapetian and Lorenzo Longo, who left Rockwell Semiconductor in 1996 to start their own company.
Those analog design tricks, including special phase detectors and frequency detectors, are what it takes to get CMOS performance to Sonet speeds. NewPort uses its own version of current-mode logic, which it dubs "Current-Controlled CMOS," or C3MOS. That kind of design continually steers current through its circuits, meaning careful attention has to be paid at the transistor level.
"You have to do everything custom. You have to do designs down to the device level," Hairapetian said.
The ability to produce Sonet parts in CMOS is important because CMOS parts operate at lower power than GaAs. In addition, CMOS and GaAs can't be integrated, so the prospect of producing the multiple pieces of a Sonet transceiver in CMOS opens the possibility of single-chip transceivers, a path NewPort followed.
Finally, CMOS' dominance across most chip markets means manufacturing is readily available -- NewPort is using Taiwan Semiconductor Manufacturing Corp. as its foundry.
NewPort may run into other challenges however, one being that its transceiver runs at 2.5 V. "The feedback we get says that's not particularly useful, because they've got 3.3 V. on their boards," Gardner said. That would mean NewPort parts would need an accompanying regulator, diminishing the parts' integration advantages.
Vitesse's commitment to GaAs remains, but the company is exploring the possibilities of CMOS. "At this point we're not real technology snobs," Gardner said.
Even if CMOS has caught up completely, GaAs "still offers advantages in terms of overall raw performance and jitter specs," Gardner said.
"Will CMOS some day do 10 Gig? Probably. But it'll be three to five years down the road. By then, GaAs will do something faster," Gardner said. "Physics still counts." |
