Leveraging long-wavelength VCSEL arrays
The advent of reliable manufacturing techniques brings the advantage of VCSELs to applications at 1300 nm.
MICK WILCOX, Cielo Communications
Over the past decade, optical communications technologies have migrated out of the long-haul backbones and steadily moved into the network edge, invading MAN and campus-level LAN environments. One of the most important consequences of this migration has been the need to develop more efficient and lower-cost optical interconnect solutions. These solutions and their components must be able to alleviate bandwidth bottlenecks by providing increased port densities to provide more links at higher speeds despite the decreasing availability of space to house optical-networking equipment.
One key technology that perhaps offers the best solution to this access problem is parallel optics. The challenges in designing robust parallel arrays revolve mainly around the source lasers, due to a relatively limited range of laser choices for designing links in higher-density environments. Optical links meeting the telecommunications industry's standards have traditionally required the use of edge-emitting lasers such as Fabry-Perot and distributed feedback (DFB), which are inherently expensive, hard to produce in high volumes, and ill-suited for creating high-density laser arrays.
The development and commercialization of arrays based on 850-nm vertical-cavity surface-emitting lasers (VCSELs) has opened the door to dramatic improvements in density for very-short-reach (VSR) optical interconnects. Unfortunately, similar developments for short-reach (SR) and intermediate-reach (IR) optical arrays have not been realized-until now. Recent development and refinement of production-viable techniques for manufacturing 1310-nm arrays using VCSEL technology have created a new low-cost, low-power, and highly reliable alternative that is dramatically shifting the economics of deploying high-density SR and IR links. In fact, this new development enables many real-world opportunities to redefine the optimum architecture for next-generation metro and access networks.
As the next wave of optical networking is deployed, it is becoming clear that the lion's share of the bandwidth must be carried via relatively short- to intermediate-length links. While large amounts of aggregated and groomed traffic will continue to be transported over long-haul links, the real make-or-break factor will be in the ability to economically interconnect diverse regional centers and network access points. In support of this assertion, industry estimates indicate that more than 90% of all metro- and local-area connections, such as those between central offices, points of presence, and access points, require links of less than 15 km.
These distance and bandwidth parameters are becoming increasingly important as the convergence of voice, data, and video onto shared networks drives up the amount of traffic, the number of origination points, and the heterogeneity of the traffic mix. Emerging optical-network infrastructures must accommodate a widening number and range of "feeder links," while simultaneously building-in the dynamic configurability to grow and adapt to constantly evolving topology and protocol-driven requirements.
Today's telecom infrastructures almost exclusively use singlemode fiber in the 1310-nm transmission window for implementing both SR and IR links. The inherent characteristics of singlemode fiber allow for optimal operation in wavelength windows centered at 1310 nm and 1550 nm, while other areas of the spectrum in the installed fiber base exhibit too much attenuation for practical transmission results. Therefore, the industry has standardized on 1310 nm for short-, intermediate-, and some long-range links, while generally using more expensive 1550-nm technologies only to implement longer-haul requirements.
Another arena in which 1310 nm is becoming increasingly important is for implementing extended backplanes and intrasystem interconnects within complex network operations centers, central offices, and edge/core aggregation points. The need for scaling these environments toward terabit levels has led to multichassis architectures spread out over entire campuses connected via fiber-optic links.
To accommodate growing networks, 1310-nm-based interconnects and the deployment of singlemode fiber for backplane applications are prerequisites for obtaining optimal performance, scalability, and deployment flexibility. But given the high port densities and tight power constraints inherent with these environments, the only viable solution is to implement these interchassis and intrasystem interconnects using efficient next-generation 1310-nm VCSEL arrays.
As the name implies, VCSELs emit light perpendicular from the surface of the device rather than from the edge, as is the case with Fabry-Perot and DFB lasers. A VCSEL is created by vertically stacking the key laser components-mirrors and an active region-on top of each other through precise epitaxial deposition in a single unified process. As many as 60 individual semiconductor layers may be stacked within a VCSEL structure that is only 10 mi-crons thick.
After growth, conventional semiconductor processing techniques are used to define the laser and complete the device. Typically, VCSELs are fabricated with a circular aperture of between 5 to 25 microns in diameter to enable single- or multiple-mode operation.
One of the major advantages of VCSEL technology is the ability to efficiently process large volumes of devices to completion and to fully test all VCSELs while still in a wafer form. In contrast, an edge-emitter requires a complex multistep crystal cleaving and coating process to form the laser mirror facets. In addition, individual edge-emitters must be die-cut, separated, and partially packaged before they can be functionally tested, thereby greatly in-creasing the cost of yielded laser die.
Another advantage is that VCSELs emit a collimated beam from the surface of the device that enables very efficient fiber coupling without the need, cost, or complexity of external lens assemblies. The VCSEL's collimated beam facilitates reliable coupling efficiencies as high as 80% with simple butt-coupled designs. On the other hand, most standard edge-emitters have elliptical, asymmetric, and highly divergent light-launching characteristics, which typically require the addition of complex external lenses to shape the beams for optimal coupling to the fiber.
Other advantages include the fact that VCSELs can operate at ultra-low threshold currents of 3 mA or less because of the inherent efficiency of the device. They exhibit much more consistent performance as a result of the relative immunity to external temperature variations of their wavelengths and operating thresholds. The VCSEL's completely embedded active region provides excellent reliability over ex-tended deployment lifetimes. The active region in edge-emitting lasers is exposed during the fabrication process. This geometry reduces lifetime and commonly requires hermetic sealing, which adds significant cost.
Over the past few years, 850-nm VCSEL technology has been widely deployed for VSR multimode-fiber links and has been shown to deliver consistent performance and robust operation at a much lower cost than can be achieved with edge-emitting devices. In fact, within two years of their introduction in the data-communications market, VCSELs completely displaced edge-emitting-based optical components, at speeds of 1 Gbit/sec, due to their higher reliability and lower manufacturing costs.
Researchers tried to duplicate this revolution in the 1310-nm arena for years with the hopes of applying the benefits of VCSEL technology to longer links. The development of new long-wavelength production-grade VCSEL technology using indium gallium arsenide nitride has resulted in bringing all of the VCSEL's intrinsic advantages to the creation of low-cost, low-power, high-volume 1310-nm laser devices.
Optical array modules are manufactured by taking an array of lasers directly from a wafer and incorporating them into an extremely dense transmitter module. A similar module with an array of receivers completes the optical link. In creating these modules, maximum density is achieved so that one transmit and receive array pair replaces as many as 12 alternative modules, while using a fraction of the board real estate, drawing less power and generating less heat inside the equipment (see Figure 1).
Array technologies also have been proven using 850-nm VCSELs in dense ultra-short-haul LAN environments over multimode fiber. But as mentioned previously, the benefits of using arrays were unrealized at the 1310-nm window because of the limits of the incumbent laser technologies. Clearly, the availability of 1310-nm VCSEL devices now opens the door for designing arrays to meet the growing requirements for higher-density, longer-distance, singlemode-fiber applications.
But creating 1310-nm arrays for singlemode fiber involves significant challenges that go well beyond those faced by 850-nm multimode arrays. A primary concern is the need for tighter control over laser-to-fiber alignment and light-launch characteristics. Be-cause 850-nm arrays are launching into a much larger-diameter multimode-fiber core (50-62.5 microns), a variation such as 2-3 microns can be well within acceptable manufacturing limits. With singlemode core diameters of only 9 microns, a shift of 2-3 microns can constitute as much as 30% of the available opening, thus causing much of the light to be lost. Therefore, 1310-nm array modules require uniform laser characteristics and tightly controlled manufacturing processes, using precision techniques such as laser welding management that ensure singlemode alignment tolerances are achieved in a volume manufacturing environment.
VCSELs' inherently circular light-launch characteristics represent a key starting point for the creation of reliable, high-performance arrays. While the need to add complex lenses to correct and shape edge-emitter beams was accepted as a fact of life for creating discrete devices, it has led to a virtually insurmountable barrier to creating viable edge-emitting arrays. Besides the cost and alignment challenges of as-sembling and tuning individual lens structures for each device in the array, manufacturers must also cope with optical crosstalk, a problem caused by spillover light from an adjacent laser that inadvertently launches into the wrong fiber.
In contrast, every laser in a VCSEL array can provide optimal coupling directly into its corresponding fiber, with no need for additional lenses, tuning, or tweaking. That is achieved by using the nearly collimated VCSEL beam combined with the ability to fabricate VCSELs with submicron-tolerance inter-laser pitch dimensions. These properties make it a straightforward design exercise to create complete arrays on a single wafer with consistent, highly accurate alignment with the fiber array. Thus, VCSEL technology enables extremely high coupling efficiencies with little or no other me-chanical intervention (see Figure 2).
VCSELs allow manufacturers to identify and select arrays directly at wafer-level testing, resulting in lower costs. The growth and fabrication of the entire array on a single wafer leads to consistent performance characteristics among all VCSELs in the array. Obviously, from a manufacturing standpoint, that is a much more desirable starting point than trying to assemble and align many separate edge-emitters into a functioning array.
To successfully commercialize arrays, the electronics behind the VCSELs must also support these new array module architectures. IC manufacturers have migrated from single serialization/deserialization devices to quad and octal devices. These changes have greatly simplified board designs and permitted more dense solutions on the electrical side of the module.
Arrays of laser drivers have also pushed the state of the art forward to further reduce part count, improve density, and reduce power consumption. By improving both the ICs and the optics hand-in-hand, new modules can be commercialized that fundamentally change the way optical-network equipment is designed (see photo on page 169).
On the fiber management side, VCSEL arrays also greatly simplify interface challenges by enabling an entire array to connect directly to fiber ribbon using high-density MTP connectors. Instead of using individual LC or SC connectors for every interface, an entire eight-port or 12-port array can be connected via a single MTP connector.
For connections between arrays in backplane or chassis-to-chassis applications, the most efficient configurations involve a direct ribbon cable link with MTP connectors on both ends. In addition, a wide range of pre-designed patch cords is available to efficiently connect arrays into existing fiber infrastructures, such as cables that fan out from a single MTP interface into 12 individual LC connectors at a patch panel.
When compared to traditional 1310-nm designs using separate discrete edge-emitting lasers, new-generation dense VCSEL-based 1310-nm arrays completely transform the design issues facing equipment designers. Current discrete 1310-nm modules have reached an impasse for creating denser solutions. In fact, these modules are limited by the size of the connectors that they use. The only way network-element manufacturers can provide a denser solution is to use array technology. VCSEL-based arrays and 1310-nm VCSELs both independently represent major disruptive technologies, which, when combined, can yield tremendous benefits to optical networks in terms of cost, size, and power.
By using VCSEL-based OC-48 arrays with 12 channels per array, equipment manufacturers can essentially fit 30 Gbits/sec of capacity onto a similar amount of board space that would be needed for four 2.5-Gbit/sec small-form-factor devices, thus improving the density three-fold over the discrete approach. The bottom line from a cost standpoint is a much more attractive gigabit-per-second-per-square-inch ra-tio as well as a significantly more flexible and supportable product design.
The ability of VCSEL-based 1310-nm arrays to deliver unprecedented advantages in cost, power, and density makes new optical backplane and intrasystem interconnect architectures that were previously unthinkable now practical. For example, highly scalable terabit-router designs can be built using modular backplane designs built entirely of dense OC-48 VCSEL arrays interconnected exclusively via ribbonized fiber terminated with MTP connectors.
In addition to reducing the cost of the laser modules themselves, VCSEL arrays can also yield exponentially increasing cost advantages at successively higher levels on the "design food chain." This simplification of board designs with higher port densities allows for more powerful systems with fewer boards and increased functionality per rack unit. Lower overall power dissipation allows for the use of smaller power supplies, and lower heat generation both simplifies airflow parameters and reduces the size of cooling fans.
From the carriers' perspective, smaller systems consume less floor space and power, thereby enabling telecommunications companies to minimize lease expenses for collocation space. Shrinking system footprints also can allow carriers to migrate out further toward users at the network edge by deploying compact, energy-efficient systems into mini central offices, remotely located pedestals, lit buildings, or other nontraditional environments.
The availability of new 1310-nm VCSEL array technology opens up an expanded realm of optical interconnect design possibilities. From a macro perspective, the densities and efficiencies achievable with VCSEL arrays will ultimately have impacts throughout virtually every arena of optical communications. These 1310-nm VCSELs will spur the rapid evolution of cost-effective optical networks, while simultaneously maintaining a standards-based bridge to leverage the telecom industry's existing fiber infrastructure investments.
Mick Wilcox is product engineer at Cielo Communications (Broomfield, CO).