Fabless model delivers high-speed tunable transmitter chip

Tunable lasers hold the key to improving the flexibility and agility of optical networks. By being able to select a particular wavelength within a communication band, users can carry fewer spares, while manufacturers can streamline operations by eliminating the need to fabricate, stock and maintain multiple devices. This will eliminate the need for 96 different lasers or line cards to cover the C-band (1525–1565 nm) using a 50 GHz spacing.

Wavelength-agile networks are also simplified with tunable lasers. Reconfigurable optical add–drop multiplexers (ROADMs) and wavelength-based routing will enable service providers to offer differentiated services, meet the ever-increasing demand for bandwidth and deliver all-optical networking.

Despite these advantages, tunable lasers have not dominated the market since their launch around 10 years ago, as early sales were hampered by the relatively poor performance of the device. However, the outlook is now brighter, and recently there has been a surge in shipments, according to market analysts such as RHK and Communications Industry Researchers. Last year saw shipments of 60,000 tunable lasers, and 50% of all newly deployed long-haul transmitters are using this format. This figure is expected to increase to 90% by the end of the decade. Tunable lasers are also starting to penetrate the metropolitan-area network segment and are expected to eventually account for over a third of all metro transmitters. With growth in both markets, shipments of this type of laser are expected to be between 100,000 and 120,000 for 2007, and increase at 64% year-on-year through 2010.

However, tunable lasers are a work in progress. While network operators recognize that these devices have reached an acceptable level of performance and reliability, costs and guarantee of supply are still a cause for concern. Many of today s designs rely on moving parts and multiple components, which require specialized, proprietary manufacturing techniques with costly alignment and low packaging yields. Such manufacturing constraints, combined with the relatively large designs, are hindering deployment. In particular, bulky tunable laser platforms are incompatible with 10 Gbit/s XFPs, the industry s most compact standardized transceiver platform.

In short, the market is still looking for tunable lasers that cost no more than their fixed-wavelength cousins, but deliver the same performance, reliability, form factor and power consumption. At Syntune, which is based in Kista, Sweden, we have designed a new-generation laser that is a significant step in that direction – a practical, affordable tunable laser.

There has been, and still is, a wide variety of approaches to building tunable lasers. Each has its merits and drawbacks, but most are based on several discrete optical components, such as gain chips, tuning elements, lenses and mirrors, which are mounted and aligned together.

Intel, Paxera, which has recently been acquired by NeoPhotonics, and Pirelli all produce tunable lasers with an external cavity that requires the assembly of multiple discrete elements. Tight tolerances are implicit within these designs, as components have to be positioned at locations insensitive to mechanical vibration and flexing of the laser package. They also suffer from a larger "footprint" within the transmitter module than monolithically integrated chips.

Alternative approaches, such as that used by Santur, feature laser arrays, and use temperature tuning. However, this form of tuning is relatively slow and consumes a substantial portion of the power-dissipation budget of the module. It also places greater demands on the laser s reliability since the chip has to operate continuously at high temperatures to serve particular wavelength division multiplexing (WDM) channels.

All these forms of discrete assembly not only increase the "part count" relative to monolithic designs, but can also require additional active optical-alignment steps during module assembly. Active alignments increase manufacturing costs, as they demand more skilled-operator time and more sophisticated robotics.

Full monolithic integration is clearly a more attractive approach, as it can minimize part count, lower power consumption and cut packaging costs by decreasing the number of time-consuming optical-alignment steps. Integrated tunable chips are also highly compact, which simplifies integration within transmitters and enables deployment in the smallest form-factor packages, namely XFP transceivers.

Unfortunately, most early attempts at monolithic InP integration were hampered by design compromises and manufacturing difficulties that produced low yields. Also, most approaches used proprietary processes and had to rely on supply from a unique fab.

For success, these proprietary processes must be replaced with standard ones so that the photonics industry can mature and replicate the silicon model. Design and manufacturing can then be decoupled, giving rise to fabless companies and outsourced manufacturing. This provides several well-known advantages, including more efficient use of the capital-intensive fab capacity, which ultimately lowers manufacturing costs, and the opportunity to move production from one fab to another, which simplifies second-sourcing and licensing of technology.

Syntune has implemented this model for its monolithic laser by using standard InP design methodologies and fabrication techniques. The result is a practical, affordable tunable laser chip that addresses the needs of network operators.

We are not the first company to have developed a monolithically integrated tunable laser chip, and we share the common approach, which features a combination of "comb" filters that have a series of high reflection spikes spaced approximately evenly in wavelength. Electrical injection current tuning selects a particular wavelength by bringing the reflection peak of one filter into alignment with that of another filter elsewhere in the chip. All other peaks are misaligned, and lasing occurs at a single frequency, a condition that is reinforced by the round-trip optical distance between the points of reflection equalling a whole number of wavelengths.

The conventional approach, which is used by JDSU, involves placing one filter at the back of the laser and another at the front (output) end. This has a downside: the variations in current needed to tune the front filter for the particular WDM channel alter the optical absorption. This produces variations in output power with wavelength, which has to be compensated with a semiconductor optical amplifier (SOA) that consumes part of the electrical power dissipation budget of the package.

This drawback can be offset by replacing the front comb filter with a more complex multi-electrode chirped front reflector, as has been demonstrated by Bookham. However, this has the downsides of more pins and more complicated tuning schemes. It also lengthens the time taken to identify the WDM channels for each chip.

Syntune has addressed these weaknesses by eliminating a tunable filter at the light-output end of the laser, and by using a modulated-grating, Y-branched architecture (see figure 1). A simple static reflector is positioned at the front of the device and the two tunable filter sections at the back. The result is a tunable laser that delivers industry-leading power flatness over the tuning range.

We manufacture our device with conventional processes used for manufacturing standard fixed-wavelength electroabsorption modulated lasers (EMLs), which are also referred to as distributed feedback lasers with electroabsorption modulators (DFB-EAs). Hundreds of thousands of these fixed-wavelength lasers are manufactured every year.

Structurally, our tunable laser is similar to a normal distributed Bragg reflector (DBR) laser. However, it replaces the single grating reflector of the DBR with a parallel coupling of two reflectors that are combined using a multimode interference (MMI) coupler.

Simultaneous adjustment of both reflecting sections provides tuning of the lasing wavelength over the entire C-band at 50 GHz spacing, while maintaining a side-mode suppression exceeding 40 dB. Tuning is also fast, as switching from one frequency to another takes less than 50 ns, and it only requires relatively small currents that reduce any device heating.

Additional structures can be added to our design, such as SOAs, which can be placed in front of the gain section to amplify and further equalize power over the tuning range. In this case, the SOA does not have to provide extra gain to boost any low-power channels, so the chip can provide better operational efficiency than other approaches.

Output exceeds 40 mW at all wavelengths for this laser, which is packaged in a miniaturized DFB-style butterfly package with an internal wavelength locker. The output power coupled into the fiber is 13 dBm at all wavelengths (see figures 2 and 3).

We have added an on-chip Mach–Zehnder (MZ) modulator to our laser to produce the first commercially viable, completely monolithic 10 Gbit/s tunable transmitter (see figure 4). This can serve ultra-long-haul dispersion-managed fiber links of typically 1000 km, where it is essential to have "zero-chirp" (little or no transient optical frequency shift at the data-pulse rising and falling edges). A follow-on version of the transmitter will be built with negative chirp – a red shift on the data pulse s rising edge and a blue shift on its falling edge. This is designed for metro applications, which require transmission over distances exceeding 80 km without dispersion compensation.

The advantages of our combined modulator and tunable transmitter on a single chip are obvious: reduced costs, a smaller size, fewer parts, lower power consumption and no alignment issues related to the assembly of discrete components. This will enable the deployment of tunable lasers in even the smallest form factors, such as XFPs, which cannot be addressed with other types of tunable laser that have a larger footprint.

The consistency of this transmitter s performance from channel to channel is demonstrated by the similarities of the eye diagrams shown in figure 4. Output power averages 5 dBm, with only a 2.9 V drive required for each modulator arm. The high extinction ratio and low dispersion penalties needed for propagating long distances in optical fibers are demonstrated in figure 5.

Our tunable lasers can be manufactured with high yields because they feature a few robust standard elements built with standard processes. Unlike silicon ICs, yield is only weakly dependent on chip size because only a very small portion of the chip is active. Instead, yield depends on processing quality and structural issues. Adding the modulator does not significantly reduce yield, as the DFB-EA processes used for the DBR structure also deliver a high-quality modulator, due to the similarity between the structures.

Our tunable laser has already been fabricated successfully in several different facilities, which demonstrates our ability to operate as a "fabless" company. Going forward, we are set to benefit from the freedom of designs and manufacturing techniques that do not require proprietary processes or unique fabrication equipment. This will allow us to build on the successful launch of the first 10 Gbit/s single-chip tunable transmitter, as it positions us to meet the rapidly growing customer demand for affordable, tunable lasers and transmitters.