MURRAY HILL, N.J. Lucent Technologies is ramping up new process technologies in CMOS and silicon germanium, as part of a continuing focus on single-chip design methods that it plans to apply across a range of communications devices. Offering a look ahead at the road map, Lucent executives last week disclosed the company is in the early stages of production with its first chips to be built in a low-cost SiGe process. The company also tipped news of a number of integrated devices planned for the next year.
One conclusion from product plans in both access and broadband switching is that Lucent is driving down into silicon support for network-service provisioning that carriers can implement in telco gear. That means the kind of features currently provided in subscriber-management software from vendors like Shasta Networks Inc. and Redback Networks Inc. can in theory move to embedded firmware and silicon.
For example, this week, Lucent will roll out an ATM port controller that brings its Atlanta family to a new watermark in aggregating ATM traffic flows while handling in silicon key quality-of-service functions increasingly used to monitor and provision network services. "This is arguably the most sophisticated scheduling device we have ever cast in silicon," said Greg Waters, vice president of network and communication products at Lucent Microelectronics (Allentown, Pa.).
The so-called APC device handles ATM service types such as continuous bit-rate and available bit-rate service, as well as Bell Labs' algorithms for dynamic thresholding. "What we're supporting is very complementary to Multi-Protocol Label Switching, but we think specific support in silicon should stop around Layer 3," Waters said, referring to the Open Systems Interconnection layered-protocol stack. "We think it's wrong to mess around with Layer 4 and above."
Waters sketched out plans to deliver, further down the road, an architecture for edge-access aggregation that will spin off devices for ATM-specific access with built-in segmentation/reassembly and inverse mux functions. Also on the drawing board is a "Super Mapper" T1/E1 aggregator that Waters calls a digital access cross-connect system-on-a-chip, and an any-service, any-port "Ultra Mapper" capable of handling hundreds of ATM and HDLC channels.
Versions of mapping devices are even being developed for wireless third-generation (3G) access devices, where architectures will be optimized for the mix of voice and data encountered in next-generation wireless-phone switching systems.
Lucent's ability to ride the rising communications wave with integrated silicon has helped buoy up the company in a year when the semiconductor business slumped overall by about 8 percent. Lucent's chip revenue rose 15.9 percent in 1998 to about $3 billion, making it the 14th largest semiconductor vendor in the world, according to Dataquest Inc.
Lucent's chip revenue grew 24 percent in the first half of its fiscal year 1999 the fastest in the industry, according to John T. Dickson, president of the company's Microelectronics Group. "We want to play in communications, and we think that's the best market there is for the next five to 10 years," Dickson said.
To bolster its position further, Lucent is ramping a low-cost 0.25-micron silicon-germanium process in its Orlando, Fla., 8-inch fab that will require just one mask step beyond its mainstream CMOS recipe. The process will be used in a broad range of OC-192 and OC-768 wire-line applications, as well as RF applications for multi-gigahertz wireless networks, some of which will ship within the next 12 months.
One of the first SiGe commercial devices likely will be a 10-Gbit OC-192 transceiver. "That sort of product used to be one of the longest holdouts of gallium arsenide," said Waters.
"SiGe will be a very powerful tool for us as we move into broadband communications at OC-192 and OC-768," said Dickson. "We're seeing a very aggressive convergence between ICs and optoelectronics" the arena the SiGe investment aims, in part, to leverage.
The fact that the process can be employed as a single-mask module added to Lucent's standard CMOS process means Lucent expects to charge as little as a 5 percent premium for SiGe. "We think we have a significant cost advantage over IBM," said Dickson.
Mark Pinto, chief technology officer and vice president of platform technology at Lucent, said SiGe is one of several pay-as-you-go modules along with BiCMOS and precision linear that Lucent is adding to its standard CMOS. Separately, the company plans to have volume capability for a 0.16-micron process in the first half of next year. That so-called COM-2 process is 0.16-micron drawn, 0.13-micron L-effective technology with a "fast gate" option requiring three additional mask steps. It also produces some transistors with 0.12-micron drawn geometries (0.09-micron L-effective).
COM-2 will provide six metal layers at voltages ranging from 3.3 to 1 V and have densities of up to 100,000 routed gates per square millimeter. With the new process, Lucent expects to significantly expand its nascent use of embedded flash developed with Taiwan's Winbond Electronics. So far, the embedded-flash technology has been limited to a cordless-phone design.
In the area of ASIC support logistics, Lucent this week will start a new program that gives its customers the ability to make daily changes on the mix of products they run in Lucent's fabs. "This allows us to give customers access to a virtual factory if that's what they want," said Dickson. "They can manage fabs and load what they want, when they want, with the portion of capacity they have purchased."
Optoelectronics is another burgeoning area for the company. Lucent already has integrated 10-Gbit transceiver functions to single-module size, and will look for opportunities to merge optoelectronic components, Sonet framing components and future SiGe devices for fiber physical-layer functions. The company says it saw 80 percent growth in its opto business in the first half of its fiscal year, which Dickson attributed to "the fiberization of the world. We are totally capacity limited in this area."
Dickson said the company still must pursue one area in laser-diode functions to enhance component support for wave-division multiplexing: the manufacture of vertical-cavity surface-emitting lasers. Though Lucent has a VCSEL array development program in place, it is not yet in a position to manufacture such devices in volume.
There are some markets the company is frustrated to have missed. Since Lucent researchers originated many concepts in quadrature amplitude modulation, market developers made a mistake by not plunging full force into the QAM-centric cable-modem market, Dickson said.
"We looked at cable modems three or four years ago and said we didn't need that. Now things are changing and we need to address that," he said, without disclosing just when Lucent will launch such parts.
Meanwhile, Lucent is pushing full bore on its WildWire products for asymmetric digital subscriber lines. The company expects 1.9 million DSL connections to be in service worldwide this year, down from 5 million originally projected.
With its forthcoming Home Wire product line, Lucent is targeting nascent, cost-sensitive arenas like home networking, said Tony Grewe, Lucent's director for communications strategy and business development.
Lucent will develop 10-Mbit/second interface parts for HomePNA (the home phone-line network association) through its partnership with Epigram. That technology partnership will remain in place, all parties insist, even though Epigram has been acquired by Broadcom Corp., a Lucent competitor.
Bell Labs researchers now believe the technology will eventually scale to 100 Mbits/s with a new physical-layer chip, Grewe said. Lucent's Advanced Technology Labs in Homdel, N.J., meanwhile, has developed a network interface that transmits data over power lines, said Grewe. But so far, high costs and noise susceptibility are gating its commercial deployment.
Rick Merritt contributed to this report