Boom in Opto Chips. Networking for high speed data transmission is causing demand for optoelectronic devices to explode
From Page One of Electronic News: April 27, 1998 Issue
sumnet.com
By Gale Morrison
New York--Underlying the "networked world" hype of the later 1990s is one salient fact: Getting
data to the people at points around the network at an acceptable rate necessitates fiber optic
transmission. And, more importantly, for component makers, it necessitates the optoelectronics
that generate the pulses of light-encoding data, or turn that light back into the digital domain's 1s
and 0s. And these, of course, have to be properly packaged.
Business has never been so good for optoelectronics, semiconductor laser producers Lucent
Technologies, Uniphase, Ciena, Hewlett-Packard, Siemens, and also for the ASIC suppliers to
the localized infrastructure (yes, Gigabit Ethernet!), and the conditions are set for demand to
continue exploding (see story, page 56).
Just last week among the younger optoelectronics concerns, San Jose's Uniphase bought the
Netherlands-based Philips optoelectronics unit for an undisclosed sum. The deal was Uniphase's
third such acquisition of telecom laser scientists from a corporate parent looking to streamline;
1996's was a group from United Technologies in Connecticut and 1997's was a Zurich group of
IBM's. Two days later telco equipment provider Ciena (Linthicum, Md.) reported its acquisition of
start-up Terabit Technology to gain capability in optical signal receivers.
In the same streamlining vein as Uniphase has found, Motorola shut down development in Tempe,
Ariz., of its own telecom-targeted semiconductor laser unit in December 1997, and there is little
doubt that these scientists and that work will continue, somewhere. Motorola "has a lot of IP
there," according to Bob Steele, a director with optoelectronics research house Strategies
Unlimited, "I wouldn't be surprised to hear" an announcement soon.
Motorola's official statement is that the OptoBus technology program, for modules built around
vertical-cavity surface-emitting lasers (VCSELs), has been discontinued but "internal efforts at the
research level are set to continue."
Old Is New Again
The wonder of fiber optic communications is not new. No one blinks anymore at the statement
that just one latest-generation fiber of a hair's width carries about 26,000 phone calls at the same
time. Coupled with the pervasiveness of the Internet, this leads to the perception that components
in the fiber network would have already hit sales highs, or have already hit the point where sales on
a chart look like a "hockey-stick," as the idiom goes.
"We thought the hockey stick would be fiber to the home," Bill Diamond of market leader Lucent
Technologies admits with humor. "We were waiting for years. . . . Nobody really saw this
coming," he said, "the hockey stick is right now."
Mr. Diamond is director of marketing for Lucent Technologies' Optoelectronics Center in
Breinigsville, Pa., about a half-hour from Allentown, Pa. He is a Ph.D. and alumnus of Lasertron, a
Massachusetts pioneer in telecom lasers.
He points out that the high price tags for these semiconductors, usually quaternary (four-element)
compounds fabricated by known processing means, is deceiving. "The high-speed lasers sell in the
low, tens of thousands (of units) and sell for thousands of dollars apiece," he said, "but DSPs
shipped in the hundreds of thousands last year and sell for five dollars a piece," he said.
Mr. Diamond and the rest of the laser diode community aren't very forthcoming on pricing and
total market size, and they don't have to be, because there aren't that many companies with the
expertise and there also aren't that many customers, who in the case of telecom usually have deep
pockets.
Strategies Unlimited, a market research firm in Mountain View, Calif., tracks optoelectronics.
According to director Bob Steele, (semiconductor) diode lasers for telecom was a $1.15 billion
segment worldwide in 1997. In 1998, he forecasts sales to be $1.35 billion, up 17.4 percent with
unit growth up 46 percent.
The larger market for optoelectronic components for communications was $2 billion in 1997,
according to William Magill, an analyst with NationsBanc Montgomery Securities, and is expected
to grow to $8 billion in just four years, in 2001. This segment of the industry has existed in
relatively quiet and profitable peace from that of modem chipsets or DRAMs, for example, though
there is total consensus that prices have been eroding at about 15 percent a year.
Different Parts For Different Jobs
The laser sources are diodes, so a current is needed to generate the pulse of light from the device.
The current is manipulated via a "driver" circuit of gallium arsenide (GaAs) or bipolar silicon;
Siemens Microelectronics fabricates its drivers in the latter substrate. These drivers take data to be
sent in the "ECL" format typically, and induce the laser to output light pulses encoding that data.
But for the long haul (a wide, 2.5 Gbps signal over long distances), evolution has dictated that the
lasers are turned on, so known as continuous wave (CW), and then externally modulated, or made
to pulse, with a lithium niobate device. The lithium niobate device, Mr. Diamond explains, acts like
a high-speed shutter so the blocking of the constant light is driven by what the ECL data dictates.
But in the grand industry tradition, integration is the key. A modulator of gallium arsenide indium
phosphide can be layered over an indium phosphide substrate to get an "integrated optoelectronic
module." This integrated part is known as an electro-absorptive modulated laser (EML). Mr.
Diamond said Lucent now can produce these EMLs for 10 Gbps transmission.
In another way, the laser signal is modulated, and this presents the opto-electronic component
market that spurred last week's Uniphase: fiber Bragg grating. For instance, the 1550nm-output
lasers have several peaks of emission around 1550nm, and that is the basis for the term
"multi-mode" fiber from a so-called Fabryr-Perot interoferometer. Multimode lasers are employed
often for fiber transmission short-haul, typically inside one building.
But having several modes (wavelengths) is not precise enough for 2.5 Gbps transmission past 250
kilometers. "You need a laser that suppresses all but one mode," Lucent's Mr. Diamond said.
Through holographic lithography, he said, "you etch grating over the active layer (the quaternary
semiconductor device) to filter out the other modes." The holograms or x-ray diffusion can be used
for the grating, which is the nature of the work at Philips Opto, as well as the EML capability.
Ciena pointed out in its Terabit press statement that it has its own Bragg grating capability.
Lucent has prototypes of lasers capable of carrying 20 Gbps, and 40 Gbps lasers in the lab, Mr.
Diamond said. These are at the higher end where semiconductor, GaAs or BiCMOS, device
drivers cannot be employed. "Electronics needs to catch up," Mr. Diamond jokes.
The 1,000 kilometer-capable 2.5 Gbps source laser Lucent discussed last fall will sell for
somewhere between $3,000 and $7,000, Mr. Diamond said. In this upper echelon of telecom
lasers, Lucent does not have to be very specific. Lucent is in "volume production" of 360 and 640
kilometer-capable optoelectronics.
980nm-output indium gallium arsenide (InGaAs) diode lasers fabricated on GaAs substrates are
the "pump sources" in erbium-doped fiber optic line amplifiers. In layman's terms, one can think of
these as loops along the fiber optic line to add a boost of energy for the signal to make a long haul.
To catch the light pulses in systems at the receiving end, the suppliers build GaAsInP
"photodetectors." These are less expensive and easier to make than laser diode sources. The
component that combines the transmitter, which for a short haul is often a VCSEL--which HP
produces on a large scale in San Jose and Motorola stopped developing--with a receiver, or
photodiode, is the "transceiver" component. These transceivers may or may not have clock
recovery ICs.
Terry Unter, VP of Global Optoelectronics for component giant AMP notes that VCSELs are
"highly manufacturable." AMP formed a partnership at the end of 1997 with Somerset, N.J.-based
Emcore, which at the time had just purchased MicroOptical Devices (MODE) of Alberquerque,
N.M. for $30 million. The new Emcore subsidiary has the capability to make Gbps transmitting
VCSELs, which are marketed as "Gigalase" (EN, Dec. 15, 1997).
Core Market
The erbium-doped amplifiers for boosting signals down a long haul, before the light disperses and
lose the signal, is a core market for Uniphase to date, though the Philips acquisition, which includes
leasing a new 60,000 square-foot facility in Eindhoven, broadens their position.
Kevin Kalkoven, the energetic Australian chairman and CEO of Uniphase, could not say how
much the Philips buy would mean to Uniphase's revenue tally. "They'd kill me," he joked. But,
Uniphase's "run rate" is $200 million a year, up drastically from last year's $100 million, so the deal
is "substantially additive."
Mr. Kalkoven contends that Lucent's optoelectronics business is wholly captive, and he is alone in
that contention. Uniphase and SDL (formerly Spectradyne), he said, are the only merchant market
suppliers. Clearly, Lucent and Uniphase are headed for a showdown.
Philips is taking an equity stake in Uniphase via last week's agreement. When Uniphase bought
IBM's laser group in Switzerland for a reported $30 million, analysts marvelled at the billion-dollar
IP Uniphase procured.
As one would expect, the Japanese vertically integrated information age companies have similar
expertise. NEC, Fujitsu, and increasingly Mitsubishi, have capabilities in semiconductor lasers and
fiber optic components, and the supplier relationships they have fall along geographic lines.
Volume OEM Supply
In the end, fiber is approaching the desktop, by virtue of Gigabit Ethernet. Ever practical
Hewlett-Packard and also Siemens Microelectronics do not strive for thousand-kilometer
transmission and the sockets in undersea systems. The two, and the data communications
components of Lucent, have joined forces with suppliers like AMP of Harrisburg, Pa., to bring to
market the various parts that make up a fiber optic line to, say, a office's router and around to
various computers.
Gigabit Ethernet necessitates fiber optic lines and therefore components. Many of the highest-end
workstations have Gigabit Ethernet cards, so the lines do come right to the desktop. Increasingly
businesses are not using copper wire when they have to wire, or re-wire, a building for
communications, because of the scalability inherent in fiber. Many have seen that the entire
network computing vision would fulfil its promise with gigabit data rates.
As Albert Comparini, business unit manager for Siemens Fiber Optics in Cupertino, Calif., puts it:
"It doesn't make sense to use copper for Gigabit Ethernet, because of the performance limitation.
Using copper up at those data rates, it takes special shielding and special (precautions) to use
copper.
"There's performance issues, and pricing issues. It doesn't make sense to even consider it," Mr.
Comparini said, adding, "in my opinion."
The de facto standard "small form factor" is the call of the day here. Lucent says that its new
manufacturing operation in Breinigsville is the first in the world to fully automate production of
these components. Siecor, the Siemens and Corning fiber joint venture, participates in these efforts
as well.
The transceiver package does take a thermo electric cooler (TEC) which is employed to keep the
laser at 25 degrees Celsius, so this represents still another opportunity. Suppliers of these TECs
currently are companies most people "have never heard of," Mr. Diamond said.
One such buyer of these transceivers for integration is Intel. The company's latest routers and hubs
have ports to plug in fiber lines. The new eight-fiber port Ethernet Express 550F lists for $7,995.
$16,000 Passive Components
Mr. Steele said that the Optical Fiber Conference in San Jose in February had an amazing number
of entrants seeking the seemingly unthinkable: the "muxer/demuxer" passive components that can
sell for as much as $16,000. These parts skillfully congregrate, or disperse, fiber optic signals for
transmission, or after transmission. They enable wave division multiplexing (WDM), and at
16-channels and above, "dense" WDM.
"If you don't know someone in the wave division multiplexing (WDM) business, you will soon,"
Mr. Steele said. "Everybody had some kind of WDM component."
Lucent is readying an 80-channel multiplexer, which of course means there are 80, say, $4,000
semiconductor lasers feeding into the multiplexed line. Ciena, the Maryland supplier, has a
40-channel system and Italy's Pirelli, somewhat ironically the same company that makes race car
tires, has a 32-channel WDM system.
One can see how the price tag on machines of the new era of telco equipment can reach quickly
into the millions of dollars.
Other Top Stories
List of Issues
Electronic News Home Page |
