|Optical integration moves back onto the agenda|
FibreSystems Europe - May 8, 2006
Photonic integration was one of the hot new ideas that emerged during the telecoms boom years, but was subsequently shelved by cash-strapped component vendors. Now, as Roy Rubenstein reports, the technology could be set to make a big revival.
Considering that the market for 40 Gbit/s transmission is yet to take off, Infinera's recent announcement is quite extraordinary. The optical networking equipment maker has detailed a lab demonstration of an optical chip comprising 40 channels, each one operating at 40 Gbit/s. "The reasoning [for such a device] is purely economic," explained Jagdeep Singh, chief executive officer of the US systems vendor. "It is the lowest-cost way to deliver bandwidth."
Even though this 1.6 Tbit/s device will not be deployed for several years, it highlights the performance hike that can be achieved with a photonic integrated circuit (PIC) made entirely from indium phosphide (InP). More significantly, Infinera's PIC strategy could signal a forked path in the optical networking industry.
Indeed, Infinera's PICs are the optical industry's sole example to date of Moore's Law-style progress. The company's existing long-haul data-transmission network (DTN) equipment already exploits a 10 × 10 Gbit/s PIC that was introduced in 2004. However, views are mixed about how much progress has been made in integrated components since venture-capitalist funding kick-started scores of start-ups in 1999 and 2000.
"In the transceiver market, 10 leading companies now have 90% of the market," explained Vladimir Kozlov, the founder of Lightcounting, a transceiver market-research firm. "They have been shedding technologies that haven't been making money." Such technologies include optical integration.
One example is integrated 1.3 ?m vertical-cavity surface-emitting laser (VCSEL) arrays. "Edge-emitting lasers have been driven down in cost so much that it's simply not worth building the manufacturing infrastructure for a whole new technology [1.3 ?m VCSELs] to achieve a potential cost reduction of 20%," said Kozlov.
However, Andreas Umbach, the chief executive officer at u2t Photonics, a German manufacturer of high-speed photodetectors, believes that progress has been made in recent years. "It depends on the level of integration," he explained. Monolithic integration comprising parallel channels that each perform disparate optical functions is clearly not commonplace, but photodetectors and tunable lasers have advanced their level of integration, he says.
Indium phosphide, a member of the III-V semiconductor family, is already the material of choice for active optical devices, now accounting for more than 60% of the global telecoms market for discrete optical components, according to market analysts Ovum-RHK. Its dominance stems from its energy bandgap - which spans the full range of optical-networking wavelengths (920-1650 nm) - and its high refractive index, which yields more compact devices.
Infinera's PICs exploit monolithic integration, in which a single material system is used to implement all of the component's optical functions. InP is ideal for monolithic integration because it can support all of the key optical functions, including lasers, gain blocks, detectors and passive waveguides. "A PIC is anything monolithically integrated, not just indium phosphide," said Catherine Lacoursiere, an analyst at BCC Research. "But it [InP] is the most practical and technologically feasible."
The alternative approach is hybrid integration, in which several different materials are used, each one best suited to a particular optical function. The challenge lies in developing the assembly techniques and a platform to host the different materials.
"Integration is a good thing where component packaging and test costs dominate," said Ian Lealman, vice-president for device development at the Centre for Integrated Photonics (CIP) in the UK. However, overall yields start to suffer as more functions are implemented monolithically. "Unless you have a very high start yield, it becomes prohibitively expensive," explained Lealman (see "InP manufacturing yields").
Companies pursuing hybrid integration emphasize the advantages of using multiple materials. For example, Jeffrey Rittichier, chief executive officer at Xponent Photonics, US, says that the InP actives in the company's hybrid components for passive optical networking (PON) are sealed with silica - which avoids the need for hermetic packaging - while all of the optical functions are interconnected on a planar lightwave circuit (PLC) made from pure silicon. Xponent can thus use merchant silicon-fabrication facilities to reduce manufacturing costs.
Enablence of Canada and Neophotonics of the US are also using hybrid techniques to make PON diplexers and triplexers. Today's PON triplexers are typically implemented using discrete components, and include two-receiver photodetectors and a laser - all in TO-cans - and thin-film filters to separate the PON transmit and receive wavelengths. However, PLC-based hybrid designs are now emerging. "It [a hybrid design] eliminates 20-30 piece parts, the alignment steps and the thin-film filters, which are fairly expensive," said Ferris Lipscomb, vice-president for marketing at Neophotonics.
With a hybrid design, two photodetectors and a laser (as well as an associated monitoring detector) can be mounted onto the PLC platform, which acts like a circuit board. A transimpedance amplifier is added using flip-chip technology, while the filtering is embedded within the waveguide to eliminate the requirement for thin-film filters.
Enablence, for example, exploits two filtering stages within its waveguides. The first divides the upstream and downstream PON wavelengths, which share a common fibre. The second filtering stage separates the two closely spaced downstream wavelengths - the 1550 ?m analogue video signal and the 1490 nm data signal. Meanwhile, Xponent's triplexer design comprises seven parts and uses a so-called "surface-mount converter" to expand the laser-fibre tolerances to ±3 ?m, thus enabling passive alignment.
However, although these companies are exploiting the benefits of hybrid integration, they do not yet want to take the final step of producing a monolithically integrated triplexer. "Any time you implement a laser, detector and waveguide on a single [InP] wafer you inevitably sacrifice some performance," explained Matt Pearson, Enablence's chief technology officer and director for product development. "We like hybrid because you can use very-high-performance lasers and detectors, and make low-loss waveguides, and each part is optimized."
A PLC-based design can also be modified easily. For example, building a triplexer for a more demanding PON specification simply involves replacing the actives on the PLC platform with higher performance versions. In contrast, a monolithic implementation requires a new design with an associated InP process qualification.
However, some people believe that PON triplexers will inevitably migrate towards fully integrated PIC technology - especially as PON triplexers could be sold in their millions and will face fierce cost targets, around $50 each typically. "Monolithic integration is needed to get to the [target] price points. It has to [be used]," said Daryl Inniss, vice-president for communications components at Ovum-RHK. "It is too early - the market is still emerging - but one dominant player will win volume by pricing the hell out of the market and by taking a big risk."
For the moment, however, organizations such as CIP do not see PON beyond the current battle between discrete designs and hybrid integration. "Until there are large volumes, there is not a lot of point to go beyond discrete," said Sean Amos, CIP's vice-president for business development.
CIP's philosophy is to use as high a level of monolithic integration as makes sense, before completing the optical design with hybrid techniques. For example, its 40 Gbit/s all-optical regenerator uses monolithic integration to create dual- or quad-array InP semiconductor optical amplifiers, which then form a building block in a more complex hybrid PLC design (see "Regenerators get ready for 40G")
"The debate centres around what makes sense," said Neophotonics' Lipscomb. And that is dependent on several factors: the unit volumes, how stable a product is, the performance achieved and the target price points. "Anything you do as an integrated product has to be a drop-in replacement," he continued. "It may be significantly cheaper but it has got to be as good."
Another market segment that is particularly suited to hybrid integration is reconfigurable optical add-drop multiplexers (ROADMs). These can have channel counts of up to 40 and the required optical functions - waveguides, filtering, variable optical attenuators, photodetectors and switching - can be combined on one platform.
However, a ROADM design is a lot more complex than a triplexer and, although ROADM is another market that is likely to benefit from strong growth in the next five years, unit sales will be measured in thousands rather than millions. Both of these factors will restrict the degree of integration that makes sense in these devices.
High speed fuels PICs
While PON and perhaps ROADMs represent longer-term opportunities for InP PICs, the current development of monolithically integrated circuits is really being driven by high-speed transmission, in particular the move towards 40 Gbit/s and new 100 Gbit/s standards. Indeed, the most widely deployed InP PIC to date is an electro-absorption modulated laser (EML), which combines a laser and modulator in InP. "This has been around for some 10 years," explained Lightcounting's Kozlov.
Tunable lasers also integrate several functions in a single InP device. "Tunable lasers are selling in reasonable volumes - several tens of thousands a year - while EMLs are in the hundreds of thousands," said Kozlov.
Apogee Photonics, US, is one company concentrating on the 10 Gbit/s laser market. It has its origins in two InP start-ups: ASIP and ThreeFive Photonics, the latter having brought to market such monolithic devices as a multiwavelength receiver and an optical performance monitor.
Now Apogee's focus is to exploit its selective-area growth and asymmetric twin-waveguide technologies to optimize the optical functions in its range of laser products. "We connect functions using tapers and, with precise control of the composition and structure, we can determine the performance of the laser and modulator," said Milind Gokhale, Apogee's chief technology officer.
One example is Apogee's uncooled EML, which can operate over a wide temperature range. "It is fair to say that we are using one of the best integration platforms to make more down-to-earth products that the market needs in large quantities," explained Erik Pennings, Apogee's director for product marketing.
Cyoptics, US, is another InP specialist making EMLs and it also produces tunable lasers under contract for other firms. Like Apogee, its staff has made some exotic PICs in the past. "We made a distributed Bragg grating with phase control, a power detector, and a modulator for a tunable laser over the C-band," explained Robert Hartman, Cyoptics' vice-president for device design and development. "We even had a version with an SOA that was finished in development." However, the market wasn't ready for such devices, he says, and was not prepared to pay any more than a 10% premium for single-wavelength (untuned) devices.
Now, however, Cyoptics is noticing a renewed interest in PICs. "Regarding photonic integration, some of the meetings I had [at OFC/NFOEC 2006] were the best in five years," said Stefan Rochus, director of marketing and business development. Particular interest is being shown in the emerging 100 Gbit/s Ethernet standard.
U2t's Umbach notes that two schemes are being considered for 100 Gbit/s Ethernet: four channels at 25 Gbit/s or one 50 Gbit/s channel with modulation. If a phase-modulation scheme is chosen, integration will be needed to realize the complex detector and waveguide structures that will be required.
Meanwhile, at OFC/NFOEC, Cyoptics provided details of a high-speed interface that it is developing in conjunction with Cray as part of the US Government's High Productivity Computing Systems programme. "The supercomputer market needs terabits-per-second transmissions," said Hartman. Cyoptics demonstrated a prototype 80 Gbit/s transceiver that is made from a transmitter PIC and a hybrid receiver. The transmitter PIC comprises 8 × 10 Gbit/s channels transmitted over a CWDM system, which makes the design simpler. The PIC incorporates lasers and modulators, as well as a waveguide multiplexer and photodetectors for phase control.
The receiver comprises a demultiplexer - in silica for now, but eventually in InP - and an array of eight photodetectors. Hartman says that the real benefit of using a PIC in this application is the cost, and the biggest return on cost can be achieved by implementing the transmitter in a PIC (rather than a hybrid), although the cost difference is still significant for the receiver.
For the time being, though, carriers - and by implication system vendors - want to maintain flexibility by using only a fraction of a system's capacity. More channels are lit, or a channel is upgraded from 2.5 to 10 Gbit/s, only when the need arises. "For any WDM system, our philosophy is pay-as-you-grow," explained Emmanuel Desurvire, senior director for photonic technologies in Alcatel's photonic-network product group. This approach does not favour multichannel PIC designs - such as "32 lambdas at once" - for optical networking. "Also, in case of a single-channel failure, the corresponding line card can be replaced without hitting the full (e.g. 320 Gbit/s) traffic," said Desurvire.
When designing a new system, vendors such as Nortel Networks look at the platform specification in terms of the interfaces - and the density of interfaces wanted by the carrier - and the reach. "We look at what new technologies can be brought to bear," said Maurice O'Sullivan, manager of physical layer and optical design at Nortel. Depending on the maturity of the integration process, Nortel will consider the technology if it delivers power, size and, in some cases, improvements in system performance.
Alcatel stresses that adequate trade-offs between different performance criteria need to be made when considering integrated products. Adopting an integrated device will probably mean that the line card will need to be redesigned, and that adds cost. "The technology also has to be mature and proven - we don't take chances in the field," said Desurvire. "A highly integrated device is not necessarily the best solution. Throwing away what you have can have an impact across the [network] architecture."
In the system
With Nortel and Alcatel selling their optical-component divisions, it has also become more complicated to explore how integration can benefit system design. "It is difficult to predict what optical component players are doing and for them to second guess systems performance," said Michel Belanger, senior technical advisor in Nortel's next-generation optical network group.
Indeed, Infinera's decision to manufacture systems allows the company to put its PIC technology at the centre of its design. "If you don't have control of the optical technology, you can't have a differentiated product," said Dave Welch, Infinera's chief strategy officer.
Infinera's DTN platform addresses the cost issue of optical-electrical-optical conversions by integrating discrete transponder functionality into its transmit and receive PICs. Infinera will not detail the resulting cost savings, but one carrier suggests that its DTN platform is 30% cheaper than other DWDM systems. Since every wavelength is converted into the electrical domain, whether it is dropped or passed, several other key benefits result. The signal's condition can be monitored, switching and grooming can be performed, and the signal can be properly regenerated - unlike in the optical domain, where both signal and noise are only amplified.
OnFiber, a North American metro service provider, has been using Infinera's DTN platform since late 2004. Flag has also adopted the system for its WDM metro rings, which are linked to its submarine network. More recently, US service provider XO Communications selected Infinera's DTN platform as part of a major network upgrade.
Welch claims that Infinera's InP manufacturing process is robust enough to achieve high yield. "We make DFB [distributed-feedback] lasers in a similar fashion to everyone else, as we do our modulators," he said. What is different is that Infinera gets its engineers to design around the process. "We design what the process can manufacture."
Infinera's design also trades off optics and electronics. Forward-error correction and electronic dispersion-compensation techniques are used to relax the optical specifications, placing more of the link-budget burden on the electronics. Nortel makes a similar point about the importance of integrated components - not just optical, but also analogue and digital silicon chips.
"Infinera basically can do what others thought impossible because they expanded the domain over which they did their design trade-offs to span systems to processing," said Karen Liu, research director for components at Ovum-RHK.
While the industry is focused on a pay-as-you-grow approach, Infinera's strategy is to deliver wavelength blocks - in chunks of 10 × 10 Gbit/s - cheaply enough, whether or not they are all needed. Welch argues that a line card costs roughly the same, regardless of what is on it and the company's PIC shifts the industry from 10 to 100 Gbit/s on a line card. The company's prototype 40 × 40 Gbit/s PIC will deliver a further 10-fold hike in three to four years' time.
However, RHK-Ovum's Inniss is not as optimistic. "A tier-one carrier won't wholly replace its network with Infinera equipment in the next 5 or 10 years," he said. "It's a good idea, a radical idea, but the path they offer is way ahead of incumbent systems."
As to whether the pace of monolithic integration will hasten in the next five years, most believe not. But that doesn't mean that there won't be exciting developments. "Infinera's 10 × 10 Gbit/s is all-in-one-go, but there will be some baby steps by the industry towards its solution," said Inniss.
Perhaps the most exciting step is the placing of a tunable laser within an XFP package that supports line-side transmission distances of 80 km and greater. Agility (recently bought by JDSU) has an InP monolithically integrated tunable laser that is sufficiently small to fit within an XFP package.
This development indicates that a variety of transceiver types will converge to one form factor, with one laser and one receiver, explains Inniss. Unit volumes will increase, while the price of such a tunable, pluggable DWDM interface will dip below $1000.
Meanwhile, Infinera believes that it is just a matter of time before someone breaks away from the pack to make a PIC triplexer. At the same time, advances in hybrid technology continue to reduce packaging costs. "We can make optics cheaper than copper," said Xponent's Rittichier, hinting at a future direction for the firm. And that would shift the balance in backplane and interconnect technology firmly towards optical solutions.