For years the telecoms industry has yearned to duplicate the phenomenal success of the electronic integrated circuit in the optical components arena. Indeed, the concept of the photonic integrated circuit (PIC) was first outlined back in 1969 by Stewart E Miller from AT&T Bell Labs, who described how multiple miniature optical components could be interconnected via optical waveguides fabricated from thin-film dielectric materials.
Writing in the Bell System Technical Journal, his paper concluded: "There is a conviction that the new miniaturized optical circuitry will prove useful…We must wait a while longer to find out how useful this new technology will become."
Little did Miller know that it would be more than 25 years before the first PICs containing "passive" optical functions like filtering and multiplexing reached the market. And it wasn't until 2004 that PICs based around "active" elements such as lasers and detectors found commercial success inside the long-haul DWDM systems from US vendor Infinera.
Ready for prime time
Now many people in the industry believe that it's finally time for photonic integration to go mainstream. "The reason for greater optimism [about PICs] today is that the killer applications that proved elusive in photonics for years have now arrived," claimed Sterling Perrin, senior analyst at US market-research firm Heavy Reading.
The spiralling consumption of bandwidth combined with falling prices has pushed integration to the top of the agenda. Photonic integration is the optical industry's best hope for scaling network capacity while also reducing transport costs per bit; that's the key conclusion of Perrin's recent report on the subject.
David Smith, chief technology officer for the UK's Centre for Integrated Photonics (CIP), the company created from BT's former fibre-optic research activities, concurs with Perrin's view. "Current technology will be physically too large and energy hungry to deliver the levels of bandwidth growth demanded by users. A step change in technology is needed that can not only deliver this bandwidth demand at economic cost, but also significantly reduce the amount of energy required to power and cool it."
Of course, this is likely to be easier said than done. One factor that's had a considerable effect on the industry in recent years is the changing nature of the telecoms supply chain. Network equipment makers now depend on external suppliers, having sold their in-house components divisions. Such an arrangement works well when the suppliers are technically ahead, but not when technology lags behind market demand, as it does now, according to Perrin.
This could threaten optical innovation because all the equipment vendors work with a same select few integration specialists. "If you, as a system vendor, have the same technology [as your competitor], how do you differentiate? How do you compete?" he asked.
Meanwhile, components companies have slashed head counts, tightened R&D budgets and moved their manufacturing to low-cost countries, or outsourced it altogether. Venture capital has been tough to find and many optical start-ups have been acquired for bargain-basement prices or simply vanished. The end result has been an "innovation gap", especially in high-risk, capital-intensive areas like photonic integration.
The lack of investment is understandable, says Oliver Jahreis, head of product management, WDM, at Nokia Siemens Networks. Component firms' slim margins make investing in R&D hard to justify unless a quick return on investment of 12 months or less is guaranteed. "Many component companies are holding back until the demand for PICs is clearer," he said.
US components vendor Opnext is a case in point. Matt Traverso, senior manager, technical marketing at Opnext is in no doubt that photonic integration is the clear long-term winner, but for now he remains cautious. "For us [Opnext] doing a [line side] PIC in a big way without standards is risky," said Traverso. "This is why there is a lot of interest in monolithic PICs in the IEEE [Task Force's] standards work."
Standards work is also taking place in the Optical Internetworking Forum (OIF), which recently identified photonic integration as a key enabler of components for 100 Gbit/s long-haul transmission. The project's ultimate objective is to create multisource agreements for 100 Gbit/s transmitter and receiver modules.
Photonic pioneer
In Perrin's view there has been surprisingly little progress in photonic integration in recent years except for Infinera, whose equipment uses monolithically integrated indium phosphide PICs that deliver 10 channels at 10 Gbit/s per channel. (In monolithic integration all the components are created from the same bulk material.)
The vendor can boast some notable achievements. It has gone from being a start-up during the telecoms meltdown to number four in the long-haul DWDM market, according to Heavy Reading's research. Infinera also announced its first incumbent operator customer in June, Deutsche Telekom, which plans to deploy the vendor's gear across its pan-European network.
Infinera bucked the trend by being well funded in a down market – the start-up raised more than $300 m (€200 m) since it was founded in 2000, before managing a successful initial public offering in 2007. It also went against the grain by being vertically integrated at a time when other systems houses were shedding their components arms, which gives the company greater leeway with regard to PIC costs by selling higher value optical systems.
"There is a screaming need for much lower cost systems: operators will use what works and is available," said Perrin. "What Infinera has proved is that operators will go with the radical."
Infinera stresses the performance benefits of its photonic integration strategy. First, transmission economics – the cost per bit per km – are improved by increasing the density of the system. "For the same rack, you have 10 times the bandwidth [using the 10 x 10 Gbit/s PIC]," said David Welch, a founder and chief strategy officer at Infinera. Second, a PIC improves heat dissipation while improving reliability. "We have shown with the 10 x 10 Gbit/s PIC that the reliability is the same – or better – than a single 10 Gbit/s laser," Welch claimed.
Not surprisingly, component vendors disagree with Perrin, and point to progress in integration across a range of products to refute his claim of inactivity.
"We've been doing photonic integration since the late 1990s and our investment has been at the right level," said Craig Iwata, JDSU's senior director of marketing and business operations. "Science projects are no longer funded," he said, which is true across the industry as a whole, not just at JDSU. "Any photonic integration [work] is done to solve real problems."
JDSU has developed an integrated laser Mach-Zehnder modulator (ILMZ) chip, which will enable it to put a widely tunable laser into the XFP form factor (currently they are only available in larger 300-pin modules). The ILMZ, which is a direct result of the acquisition of Agility Communications in 2005, integrates a tunable laser, semiconductor optical amplifier and Mach-Zehnder modulator onto a single indium phosphide chip just a few millimetres long. Further work is needed on the driver electronics inside the module before the final product is announced, probably later this year.
Other products that will benefit from integration are transmitters and receivers for 40 and 100 Gbit/s transmission, where integration can be used to attack cost and fibre connection issues. JDSU is developing components for 100 Gbit/s based on dual-polarization quadrature phase shift keying (DP-QPSK) modulation. The receiver demands high-integrity signals in terms of timing and skew, which is difficult to achieve using fibre-based connections. "Integration is required to get the performance," said Ed Murphy, chief technology officer, integrated photonics, at JDSU.
Driving down costs
One integrated optics start-up that managed to survive the lean years is US vendor CyOptics, and it points to cost and power consumption as the main drivers for integration. "Further cost reduction is limited using discrete [optical component] technology, while the transition from XFP to SFP+ [transceivers] doesn't reduce the overall system cost," said Stefan Rochus, CyOptics' vice-president of marketing and business development. "Further cost reduction can be achieved by moving from single lasers to [integrated] array technology, where common elements such as a thermo-electric cooler can be shared."
Canadian firm Enablence Technologies highlights how integration can benefit functionality within an assembly, citing its variable optical attenuator (VOA) multiplexer as an example. Instead of adding 40 VOAs, or several VOA arrays, to an array waveguide (AWG) multiplexer and fusing fibre to each channel to sense its power, a single structure is used. "You collapse everything onto a chip-based solution with the power taps – like the AWG – all waveguide based," said Matt Pearson, vice-president of technology at Enablence. Integration using waveguides may not necessarily reduce power consumption of the active components, he says, but with planar lightwave circuits (PLCs) it's possible to make things athermally, eliminating the need for a thermo-electric cooler.
Talk to anyone about advanced component integration and one theme always crops up: yield. Failure of a single component renders the entire PIC unserviceable. Simply put, the more components on a chip the more chance there is that one has a fault. This is an area where systems vendor Infinera appears to have an advantage over the pure components players, because it can trade off its PIC design with system electronics to maximize the yield of the manufacturing process.
Earlier this year Infinera demonstrated transmit and receive PICs based on differential quadrature phase-shift keying (DQPSK) modulation, each containing 10 channels at 40 Gbit/s. This chip takes complexity to a new level – the 10 x 40 Gbit/s PIC has 230 optical elements, compared with 50 for the 10 x 10 Gbit/s PIC. Nevertheless, Infinera is confident that achieving sensible yields will not be an issue. The firm is claiming a 60-fold improvement in chip yield since making its first 10 x 10 Gbit/s PIC in 2004. "We expect the yield of the next-generation PIC to show similar efficiencies," said Welch.
Another DQPSK developer is Bookham, but it does not wish to make trade-offs between performance and yield. Currently, the company's indium-phosphide-based 40 Gbit/s DQPSK modulator works alongside its tunable laser chip. Although the vendor does not rule out integrating the components monolithically in future, it says there are performance advantages in keeping them separate: "Every dB is important here, plus a hybrid approach enables faster time to market," said Andy Carter, Bookham's vice-president of technology.
Monolithic vs hybrid
CyOptics, originally a proponent of monolithic integration using indium phosphide, has added silica-on-silicon PLC expertise through acquisition. Having both technologies enables the company to combine monolithic and hybrid integration techniques to maximize yields.
"The laser array yield has the most devastating impact on [the yield of] a photonically integrated solution," said Rochus. For this reason the company favours a four- or five-laser array whose yield approaches that of a single laser, it claims. "For an eight or 10 channel design, we believe a PLC hybrid is the better approach – using [two] arrays integrated on the PLC."
"You use the highest monolithic integration that makes commercial sense in terms of yield," said Graeme Maxwell, CIP's vice-president of hybrid integration, noting that what makes commercial sense is a continually moving target. There is also the issue of likely volumes. Since the volumes are rarely high (more than 100 000), getting the process mature enough to improve yield is a challenge. A hybrid approach tackles these problems by enabling active devices to be optimized and shared across several products lines to boost volumes. As an example, CIP is reusing actives for sensor networks as well as for telecoms.
CIP has developed a hybrid platform – HyBoard – that it compares to a printed circuit board onto which "daughter" cards holding, for example, actives can be mounted. Using this approach it can turn around different designs very rapidly, Maxwell claims. HyBoard will be used in an EU project to develop integrated solutions for 40 and 100 Gbit/s transmission (see "Governments prioritize photonics R&D" in Further information).
Optical access is a rare example where volumes are high enough – in the millions – to really benefit from integration. For its passive optical network (PON) triplexer, Enablence has a PLC-based platform onto which a laser and two photodetectors are coupled. Volumes are boosted further by using the same platform for BPON, EPON and GPON designs. All that changes are the actives, depending on whether it is a 10 km reach EPON design or a 20 km GPON transceiver.
Time to invest
Is having integration expertise confined to just a few companies handicapping innovation in system vendors? And, as Heavy Reading's Perrin speculates, will system vendors take component companies back in-house, reversing the trend of the last decade? Equipment vendors dismiss both possibilities.
"We are working closely with our suppliers and have deep discussions on capabilities and cost," said José Mir, director of product strategy for the WDM product group, Alcatel-Lucent's optics business. Working with several specialists is beneficial because the company doesn't want to be tied to a single source – an issue that arises with an in-house component arm. "We can't afford any more to go single source," he said.
Nevertheless, equipment vendors do agree that the level of investment in photonic integration has been less than desired. "Although the photonic integrated circuit is not a panacea for the industry, it does help to solve specific problems, so it would be a shame if investment did not pick up," said Nokia Siemens' Jahreis. The lack of investment is not impeding the industry right now, but it could do if developments don't start soon, he concludes.
Further information
Governments Prioritize Photonics R&D
A significant portion of R&D investment in photonic integration in recent years has come from government, in Europe through the European Union's 7th Framework Programme (FP7) as well as national governments, and through the National Institute of Standards and Technology (NIST) in the US. Here's a selection of the projects that are currently underway:
• German VCSEL specialist VertiLas is part of the POLYPLANAR project backed by the German Federal Ministry of Education and Research (BMBF) to apply hybrid integration in applications such as fibre-to-the-x (FTTx) and parallel optics. VertiLas is working with polymer waveguide experts at the Heinrich-Hertz-Institut (HHI), Fibre Optical Components, which is embedding thin-film filter technology within waveguides, and Aifotec and MergeOptics, which bring packaging and transceiver expertise, respectively.
• VertiLas is also part of the EU-funded GigaWAM project to develop next-generation WDM-PON components using hybrid integration. Leading the project is optical components specialist Ignis Photonyx from Denmark, which makes athermal arrayed-waveguide gratings. Also involved is Swedish systems vendor Ericsson, and Germany's FiconTEC, which makes automated photonics manufacturing equipment. The project has received €3 m in EU funding and will run until March 2011.
• UK firm CIP is involved in an EU project dubbed APACHE (for agile photonic integrated systems-on-chip) to enable WDM terabit networks. The project will use CIP's HyBoard platform (see below) for high-speed WDM network devices such as building blocks for 100 Gbit/s phased-based modulation schemes, and monolithically integrated transmit and receive arrays. The goal is to develop common elements such as laser arrays or modulators that can be dropped into a hybrid platform and used across different designs.
• CyOptics is working with silicon photonics firm Kotura on a $6 m (€4 m) three-year project to develop photonic integrated circuits (PICs) capable of terabit-per-second data rates, as part of NIST's Advanced Technology Programme. The company is developing a CWDM-based design on a hybrid platform. The transmitter uses two arrays of five distributed-feedback lasers coupled to a silicon-on-insulator multiplexer, while the receiver comprises a demultiplexer and discrete detectors. "By October 2008 we will have a 10 x 10 Gbit/s silicon-on-insulator device," said CyOptics' Stefan Rochus. The device will support links up to 10 km, while being cost competitive with shorter reach 300 m VCSEL-based parallel links, he claims. The plan is to develop a 20 x 25 Gbit/s device in the project's second year, and a 25-channel PIC, each at 40 Gbit/s, in the third.
• Find out more about these projects at fibresystems.org.