Unprecedented demand for pump modules in optical-amplifier market

Th 12spr08 1

The 980-nm and 1480-nm pump module market could approach $4.5 billion by 2004, according to Wall Street estimates.

Wavelength-division multiplexing (WDM) systems engineers continue to introduce systems with higher channel counts and increased transmission speeds. At the same time, the push is on to develop technologies utilizing more of optical fiber's inherent bandwidth. Systems designers are creating WDM architectures that use the space outside the traditional C-band, incorporating both the L- (long) and S- (short) bands to transmit signals. All these factors are driving the demand for more powerful optical amplifiers. To achieve this, erbium-doped fiber amplifier (EDFA) engineers have begun to design amplifiers with higher-power pump lasers and with additional pumping stages.

Notably, while the EDFA fostered widespread deployment of WDM systems, it was the appearance of highly reliable pump lasers that enabled development of the optical amplifier. The 1480-nm pump lasers (manufactured by Furukawa, Sumitomo, Alcatel, JDS Uniphase, Anritsu, and NEC among others) were the first to be designed in--as their lower power allowed for heightened reliability--a characteristic critical to telecommunication network acceptance. Not long thereafter, higher power 980-nm pumps begin to be used as companies like JDS Uniphase and SDL introduced lasers with life test data demonstrating 100-plus-years meantime-to-failure (MTTF) chips.Th 12spr08 1

Fig. 1. A typical high-power EDFA today could have three or more stages. The initial stage is pumped using 980-nm lasers, which allow the most signal gain without introducing significant amounts of noise. The latter stages typically use 1480-nm lasers, which offer more-efficient pumping (lower cost per mW output), significantly boost signal power relative to noise, and offer better distributed (more-uniform) gain.

A typical high-power EDFA today could have three or more stages versus the single stage common when the EDFA was first introduced. The initial stage is pumped using 980-nm lasers, which allow the most signal gain without introducing significant amounts of noise. The latter stages typically use 1480-nm lasers; these devices currently offer more-efficient pumping (lower cost per milliwatt of output), significantly boost signal power relative to noise, and offer better distributed (more uniform) gain (see Fig. 1).

Multistage, high-powered EDFAs use very large spools of erbium-doped fiber--the longer the spools, the greater the pumping power needed to excite the erbium and amplify the signal. The 1480-nm lasers can pump these long spools much harder than the 980-nm models without introducing nonlinearities and other bandwidth-crippling effects into the system. As more stages using 1480-nm lasers are added to EDFA design, we think the current 980-nm/1480-nm pump volume ratio--tilted in favor of the 980-nm laser--could reverse.

Higher-channel-count architectures utilizing the L- and S-bands will use more 1480-nm lasers because the outer bands must be pumped at lower power (~100mW) to avoid introducing added noise to the system. These low-cost, low-power 1480-nm pump modules most likely will come from Japanese volume producers like Furukawa and Sumitomo. We believe average prices for 1480-nm modules could fall from roughly $1400 today to around $675 by 2004. In contrast, pricing for 980-nm modules should decrease much more slowly as SDL and JDS Uniphase offset price reductions by offering higher power for the same price.

Today, dense wavelength-division multiplexing (DWDM) is the major force influencing EDFA design and the performance of the diode lasers that pump them. Increased channel count necessitates proportionately higher total pump-laser power. For example, if 64 wavelengths are used in place of 16, EDFAs must have four times higher output to continue launching the several milliwatts per channel required to reach the next EDFA. We believe that this trend of increasing WDM channel counts could boost the average number of pump lasers per high-power optical amplifier to nearly eight by 2004 from around three today.

In 980-nm pump lasers, we expect powers to grow 30% to 40% annually through 2004. Currently, SDL and JDS Uniphase dominate the market for high-powered 980-nm pump lasers, each producing 300mW chips. SDL expects a 500mW chip (250mW module) by the first quarter with commercial volumes beginning in the second quarter. The company will initially qualify its newly designed module at lower powers, but expects to go above 300mW output power sometime in 2000. Keeping pace, JDS Uniphase announced its plans to boost the power of its 300-mW GO4 chip 33% over roughly the same time frame when it introduces its GO5 chip. JDS Uniphase's GO6 chip is expected to debut later next year, approaching 500mW. We believe JDS Uniphase can achieve 800mW in the lab but the build-up of heat causes the mode of the signal to change (hop), making the pump ineffectual. SDL expects to introduce a 1W 980-nm chip in early 2001. Chips are placed on a heat sink to dissipate heat and prevent mode-hopping. We believe this fast-ramping power curve will continue to support very high barriers to entry (highest) and ensure relatively firm pricing through 2004. High chip powers don't mean a thing if the companies cannot manufacture them with yields sufficient to support commercial volumes.

We expect future high-power 980-nm pump laser chip designs to include two options for boosting power, both of which raise reliability issues:

  • Junction side down. Flipping the laser upside down improves thermal conductivity. This means better cooling and helps prevent mode-hopping (see Fig. 2a)
  • A tapered laser. Also called the trumpet approach, this technique funnels a broad-beam (higher power) laser into a focused, narrow area (see Fig. 2b).
Th 12spr08 2aTh 12spr08 2b
Fig. 2. Future high-power 980-nm pump laser chip designs most likely will boost power in one of two ways. Junction side down (a), which flips the laser upside down to improve thermal conductivity; or through the use of a tapered laser (b) also known as the trumpet approach, which funnels the broad beam into a narrow area.

EDFAs coherently amplify 1550-nm signals through the conversion of either 980-nm or 1480-nm pump laser light. The 980-nm lasers are manufactured almost exclusively in the United States and Europe; Japanese companies dominate 1480-nm production. Japanese EDFA designs tend to have the highest 1480-nm content; the rest of the world employs more 980-nm pumps. Fluctuations in the exchange rate between the U.S. dollar and the Japanese yen also influence wavelength design choices.

As DWDM system technology advances and more channels are squeezed down a single fiber, it is increasingly important to reduce amplification noise. One way to reduce this noise is to design 980-nm pump lasers into the EDFA. We estimate the market for 980-nm pump lasers at more than twice that for 1480-nm pump lasers today.

JDS Uniphase and SDL are the dominant manufacturers of 980-nm lasers with approximately 60% and 30% of the market, respectively. Both companies offer extremely reliable lasers (100-years-plus MTTF), but only SDL can package its own laser into a grating-stabilized package--a key competency as high-channel-count WDM systems designers require very precise stable output.

Currently, JDS Uniphase sells the bulk of its chips to Oak Industries' Lasertron division, Lucent Technologies, Nortel Networks, and Pirelli; all these companies package their own modules. JDS Uniphase is moving to add packaging capability by the end of this year. This is a critical move: A typical terrestrial high-powered (300mW) 980-nm chip may sell for around $500; a packaged terrestrial module can garner in excess of $2200. Failure to add packaging by year-end could limit upside margin potential for the JDS Uniphase unit in 2000.

In response to increased demand for 980-nm pump lasers, JDS Uniphase and SDL recently expanded capacity. JDS Uniphase opened a laser-fabrication plant in Zurich to produce its new generation 4 (GO4), 300mW 980-nm diode. Yields at this new plant are improving and the company expects to exit 1999 at a 250,000-chip run-rate.

SDL also greatly increased chip capacity this year, doubling production to 200,000 chips. The introduction of a wafer-fabrication plant this year could ramp chip capacity beyond 300,000 units by mid-2000. Because chip production has many benefits similar to semiconductor production and leverages fixed assets, we believe both companies stand to gain economies of scale and subsequent margin expansion.

Both JDS Uniphase and SDL remain capacity constrained in the production of 980-nm pumps. SDL's bottleneck is in packaging modules, JDS Uniphase's remains at the chip level. SDL's chip production is likely half of its capacity. SDL opened a 40,000-sq.-ft. packaging facility in Victoria, British Columbia, to help alleviate this bottleneck. We estimate SDL's current production capacity in Victoria at 10,000-plus units a quarter and expect the new facility to boost this number by two times or more.

The barriers to entry in 980-nm pump laser manufacturing are very high-higher than for any other optical component. Perhaps the most significant hurdle is the thousands of hours of both laboratory and field-test data both SDL and JDS Uniphase have logged proving their lasers meet network operators' stringent terrestrial and undersea requirements. The companies' failure-in-time scores, near 100, show their lasers will not fail in 100-plus years of continuous use. A second key barrier to entry is power requirements. As WDM system OEMs move to higher channel-count systems, they continuously demand higher-powered pump lasers. The fact that power and reliability are inversely related makes for a very steep learning curve, leaving Oak Industries' Lasertron division, Furukawa, ADC's Spectracom, and other competitors continually playing catch-up.

As pump lasers are pushed to their physical limits to provide maximum amplification power, high reliability is critical. The technologies behind reliable pumps involve molecular beam epitaxy (MBE, used by JDS Uniphase) and metal organic vapor phase epitaxy (MOVPE, used by SDL).

A key part of any reliable 980-nm pump laser is the mirror passivation. The area between the mirrors offers the lasing cavity needed for laser performance. Higher power can dramatically intensify heat buildup within the laser. If the cleaved mirror is not properly protected, its degradation from oxidation eventually leads to failure of the laser, or catastrophic optical mirror damage (COMD). JDS Uniphase's patented E2 mirror passivation completely suppresses COMD. The E2 process is an important proprietary IBM technology licensed to JDS Uniphase. Notably, older versions of the E2 process were licensed to Lasertron, Nortel, Pirelli, and Hewlett-Packard Co.

The general perception has been that 1480-nm lasers are more reliable than 980-nm devices. Indium gallium arsenide phosphide (InGaAsP) lasers (of which 1480s are made) tend not to fail suddenly but to degrade slowly and predictably over time.

Reliability issues have long been associated with 980-nm pump lasers. The crux of the problem is that when a 980-nm pump laser in a single-pumped EDFA fails, the entire system goes down. To ensure signal delivery, engineers designed the "dual-pumped" EDFAs, which utilize two 980-nm pump lasers, essentially making the amplifier fail-safe.

JDS Uniphase, recognizing the need for an improved 980-nm laser, introduced the E2 process in 1990 to resolve the 980-nm lasers' reliability problems. E2 allowed JDS Uniphase to differentiate its 980-nm product by making it extremely reliable. Prior to the E2 process, 980-nm lasers failed frequently, suddenly, and unpredictably due to mirror degradation. Wafer growth and the development of mirror passivation procedures like E2 have eliminated the major sudden-failure mechanisms, however.

Meanwhile, enough field data has been accumulated with JDS Uniphase's and SDL's 980-nm pump lasers to develop a convincing reliability picture. Extended life testing has allowed JDS Uniphase to raise its MTTF estimate for 980-nm lasers above 2-million hours; in 1998, it could claim only 1-million hours. Results suggest that as 980-nm pump lasers continue to accumulate a track record versus 1480-nm technology, the reliability gap will narrow further, or disappear altogether.

The 980-nm pumping band also has important implications. The absorption spectrum versus pump wavelength is strongly peaked around 980 nm. As pump powers were increased over the past year, in conjunction with higher channel (40 to 80) WDM systems, the pumps were required to be wavelength selective. As a result, 980-nm pumps are increasingly wavelength stabilized using fiber Bragg gratings (FBGs) to eliminate gain variations and improve spectral quality.

New 80-plus lambda systems place channels closer together than ever before. Additionally, in most long-haul and undersea DWDM systems, many EDFAs are cascaded in sequence. After each amplification, if EDFA gain flatness is not tightly controlled across the entire band (all 40 or 80 channels), the power of each channel can eventually vary widely introducing nonlinear effects in higher-power channels or signal degradation in lower-powered channels.

This combination of higher channel count and cascaded amplification makes FBG technology increasingly important-and it was this grating process that propelled SDL into the market in 1998. SDL manufactured an inherently unstable 980-nm chip that needed an FBG to work properly. Comparably, JDS Uniphase's chip was quite stable and did not need a stabilization feature--until the recent move to higher channel counts for WDM systems. When the WDM market went to higher channel counts sooner than expected, vendors packaging the JDS Uniphase chip were caught by surprise and could not come to the market with the grating feature in a qualified package soon enough to trump SDL. But Oak Industries' Lasertron division recently qualified its FBG modules at several OEMs.

The introduction of gratings didn't fix all the problems associated with advancing multiplexer technology, however. Different wavelengths tend to propagate at slightly different speeds in the optical fiber--an effect called dispersion. As WDM networks increasingly migrate to 10-Gbit/sec modulation rates, power levels must be managed effectively. An important concept is that both gain flattening and dispersion compensation are needed immediately after signals are amplified. Ironically, gratings introduce additional power loss, adding to the total pump-power requirements.

As a result of increased channel counts and the transition to OC-192 (10 Gbits/sec), the roughly 100-km amplifier spacing typical in long-haul DWDM networks is predicted to decrease over the next few years. Added EDFA functionality for DWDM introduces large internal losses that must be compensated for by higher power pumps and additional pump stages. More amplifiers per kilometer and more pumps per EDFA created a dramatic surge in pump laser-unit demand in 1998. In response, both 980-nm and 1480-nm pump manufacturers are redesigning their chips for higher power and rapidly adding capacity.

Because fabrication of both 980-nm and 1480-nm pumps require extraordinary skills, only a handful of companies have developed a Bellcore-qualified 980-nm pump laser manufacturing process, and even fewer can realize >200-mW output power with high reliability. We believe only JDS Uniphase and SDL have reached this level of reliability, power, and manufacturing capacity.

Bellcore-qualified 1480-nm chips are more widely available, with more than 10 suppliers actively vying for market share. But high-power 1480-nm pumps appear to be a different story. As with 980-nm pumps, high-power 1480-nm production is dominated by a few large players. We believe Furukawa currently holds about 40% of the market for 1480-nm pumps, Sumitomo about 20%, and JDS Uniphase and Alcatel about 15% each. Anritsu, Fujitsu, and others round out the space. This group offers 1480-nm modules with output powers exceeding 150mW or higher, although even 140mW and 150mW packages can be tough to purchase in volume.

A good guide to 1480-nm pump pricing in the most common, 100mW to 140mW range is $10/mW. Still-higher powers are available; Anritsu offers a 200-mW package and Furukawa claims to have a 250mW module. Prices for these packages can run well in excess of $5000 each and the units are not currently available in volume. The drivers for these high-powered 1480-nm pumps mirror those of 980-nm pump demand, leading to capacity issues in this market as well.

Its larger effective pumping power makes 1480-nm the preferred wavelength in the booster (second) stage of an amplifier, since noise generation can be accommodated by providing an adequate signal-power input. In general, the choice of pump wavelength in the booster is determined by power output (measured in dBm) per dollar; this metric currently favors 1480-nm lasers.

Two major factors have contributed to the 1480-nm pump's continued popularity. As channel counts for DWDM systems increased from 8 to 80, EDFAs required more output power. This demand was largely met with 140mW 1480-nm lasers because the equivalent 200mW 980-nm pumps were not yet commercially available. This is important because a 200mW 1480-nm pump module gives the same output power from an EDFA as a 300mW 980-nm pump module. The 980-nm photon has 50% more energy than a 1480-nm photon owing to its shorter wavelength. Although the 980-nm photon wastes 50% of its energy, it gives a much better signal-to-noise ratio than 1480-nm pumping, where one 1480-nm photon equals one 1550-nm photon.

Secondly, the decline of the yen gave Japanese producers of 1480-nm pumps a pricing advantage. JDS Uniphase believes the available power will favor 1480-nm through this year, but the price-performance gap will narrow somewhat as 200mW-plus 980-nm pump modules become available.

This year also saw the debut of Bellcore-qualified multimode aluminum gallium indium arsenide (AlGaInAs) pump lasers for next-generation fiber amplifiers capable of >30 dBm output power. Demand for high-power (>1W) multimode pump modules is driven by the need for fewer pump modules in high-power EDFAs. Typically, 70% to 80% of pump lasers' cost is in the butterfly package, so considerable cost can be spared using fewer modules, especially considering the additional cost of wavelength combiners for pump laser multiplexing. Many manufacturers are readying ytterbium/erbium cladding-pumped fiber amplifier and Raman-shifted amplifier technologies for commercial production in 2000. With reliable, high-power multimode pump lasers now available, these new amplifier technologies will enable even further DWDM capacity expansion at ever lower cost.

The introduction of 980-nm pump laser technology into submarine networks was another major market shift this year. Submarine networks are long-haul in nature, necessitating that many EDFAs be cascaded between landfalls. In the past, eight-channel, 2.5-Gbit/sec systems could tolerate the noise generation of 1480-nm pumps. With submarine designers moving to 16 OC-192 channels (in Alcatel's case, 32 channels), however, 980-nm pump lasers' improved noise performance is now indispensable. Submarine systems also require extremely high reliability, another reason undersea network designers chose 1480-nm pumps exclusively.

The 980-nm pumps' higher power should drive a rapid transition in the submarine market from the exclusive use of 1480 nm to amplifier designs utilizing both 1480-nm and 980-nm pumps, much like what happened in terrestrial systems. We believe the undersea opportunity could soon rival that of the terrestrial market in dollars, as these components can garner an average selling price three times that of their dry-land counterparts. We expect JDS Uniphase and SDL will account for virtually 100% of the undersea 980-nm chip market going forward given their extremely reliable, high-powered products. At this time, we give a slight advantage to SDL in terms of projected 980-nm undersea market share.

In sum, more amplifiers per kilometer, more pumps per EDFA, and more market applications are creating a dramatic surge in pump laser unit demand--such that it is outpacing current capacity. In response, both 980-nm and 1480-nm pump laser manufacturers are redesigning their chips for higher power and doubling or tripling capacity almost annually.

As strong demand persists, pricing for amplifiers and high-end pump lasers should stabilize relative to that for many other optical components. Couple this with robust undersea network growth, where components can garner three times the price of their terrestrial counterparts, and we believe pricing should stabilize. Th 12spr08 3

Fig. 3. The total markets for 980-nm and 1480-nm pump modules-including the lucrative undersea arena-could grow from about $272 million and $109 million, respectively, in 1999, to exceed $2.7 billion and $1.7 billion in 2004.

In dollar terms, we believe the respective markets for 980-nm and 1480-nm pumps modules could grow from about $272 million and $109 million this year to nearly $4.5 billion combined in 2004. The nascent market for Raman source pumps, probably operating around 1450-nm, could grow from $700,000 to $138 million over this same time frame, yielding a total pump lasers market in 2004 of $4.6 billion (see Fig. 3).

James Jungjohann and Rick Schafer are equity research analysts at CIBC World Markets Corp. in New York. James can be reached at e-mail: James.Jungjohann@us.cibc.com or at (212) 667-7013. Rick can be reached at e-mail: Richard.Schafer@us.cibc.com or at (212) 667-7905.

More in Transport