Integrating active components improves system design

Kevin Affolter

To take advantage of additional bandwidth in dense wavelength-division multiplexing (DWDM), optical networking system vendors must find ways to integrate numerous components such as continuous-wave (CW) lasers, modulators, attenuators, and wavelength lockers in network architectures. The output of these components is then channeled through multiplexers and amplifiers to deliver additional bandwidth to customers. There are significant limitations to this approach, including requirements of space, money, and advanced skill sets to manage and maintain each component in the optical path.

High-capacity performance must be cost-effective. In a typical system, the CW laser provides the source for a specific wavelength of light, and the modulator, typically lithium niobate (LiNbO3), is used to apply the data signal to the light (see Fig. 1) In this example, the variable optical attenuator (VOA) provides pre-emphasis to the individual optical signals, thereby optimizing end-to-end system performance on a channel-by-channel basis. The wavelength locker keeps the laser wavelength locked to the International Telecommunications Union (ITU) grid. The multiplexer combines the individual wavelengths of light onto a single fiber to which gain is applied with an erbium-doped fiber amplifier (EDFA) before the signals are launched onto the transmission fiber.

This approach requires polarization-maintaining fiber (PMF) handling skills, consumes a large amount of space on a circuit board, and uses expensive individual components. Architectures like this have been widely used in DWDM long-haul networks for several years, and were broadly accepted, until now, largely due to the performance benefits achieved by selecting the best technology for each function. However, system vendors are under great pressure to improve the density (measured in Gbit/s per square foot) and cost (measured in dollars per Gbit/s per km) of their solutions. System density can be improved by increasing the line rate, to 40 Gbit/s for example, or by keeping the same line rate and squeezing more circuits into the identical footprint.

PASSIVE COMPONENT INTEGRATION
It is now possible to consider monolithic integration of a variable attenuator array with a DWDM multiplexer. Many component vendors already have solutions such as this available or on their roadmaps. The most promising platforms for this approach are planar waveguides, normally implemented either in silicon or silica. Typically, an arrayed waveguide grating (AWG) provides the multiplexing function and a thermo-optic Mach-Zehnder interferometer the attenuator function (see Fig. 2).

• Silica and silicon devices are highly temperature sensitive so must be temperature controlled or must provide temperature compensation.
• The most economic way to implement planar waveguide devices is to cram a high channel count into the product, typically 40 to 64, but potentially much more. The end customer, therefore, has to pay for all 40 or 64 channels up front, thereby making the initial cost high compared to more scalable technologies like thin-film dielectric filters or fiber gratings.
• Each channel still has to be individually accessed, using 40/64 full-size or compact connectors. So although the planar waveguide device itself is compact, by the time the channels are fanned out to something a customer can use, the density improvements over other technologies are less obvious.
• It is highly challenging to overcome polarization sensitivity routinely witnessed in these types of planar waveguide devices, especially the variable attenuator; this problem is compounded with parallel integration and will show up as low fabrication yields.
• Integration of a large port count AWG multiplexer with multiple VOAs provides a mammoth yield challenge—the need for large wafer size (potentially greater than 6 in.)—and massive wafer fabrication capital expenditures.

ACTIVE COMPONENT INTEGRATION
Integration of functions in active components is somewhat more mature than in passive devices. Attenuators can be monolithically integrated with high-speed modulators, in indium phosphide (InP), gallium arsenide (GaAs), or LiNbO3. Electroabsorption modulators (EAM) can be monolithically integrated with lasers to produce electroabsorption modulator lasers (EMLs). These approaches are widely used in long-haul networks and the benefits to systems vendors are significant. Benefits include greatly improved optical efficiency attained by keeping the light in the waveguides rather than coupling in and out of fiber, which results in many dBs of loss; and the solutions become more compact as the component count is reduced and fiber routing is eased. These approaches have limitations, however, especially in DWDM networks where the majority of these devices are used.

Tunable lasers, in particular, would benefit from integration. The need for tunable lasers has been a discussion point for several years, yet products meeting the specification demands of the optical networking industry have only started to become available in the last 10 months. The applications address the ease of planning (one time provisioning), inventory reduction of spares, reconfigurable optical add/drop multiplexers (OADMs), and dynamic wavelength routing. The first two issues are broadly understood and are most pertinent to this article; the latter two require development of new component technologies such as true optical crossconnects and tunable filter technology, both of which are major focuses for several companies.

For a DWDM network, planning on a wavelength-by-wavelength basis is complex and inflexible to changing demand patterns. It is broadly recognized in the optical networking industry that it is almost impossible to forecast demand with any accuracy. This fact has lead to component and system vendors carrying huge inventory burdens to provide some degree of flexibility. From a sparing perspective in the network, high costs are incurred to make sure that if a circuit pack containing a DWDM laser fails, a replacement can be quickly and seamlessly identified and deployed. When the circuit pack contains an individual DWDM laser and Mach-Zehnder modulator, the cost can be tens of millions of dollars.

Simplified planning, reduced costs of inventory, and sparing—through the introduction of tunable lasers throughout the network—is the solution that is becoming increasingly accepted among network providers and systems vendors.

The amplifier provides three functions: it increases output power, flattens power output across all wavelengths, and provides transmitter pre-emphasis functionality. Three of the functions discussed previously in Figure 1 can now be monolithically realized on one chip. The fourth function from Figure 1 is the wavelength locker, which can also be integrated at the hermetic package level using a combination of beamsplitters, etalons, and photodiodes. No additional space is consumed using this method; however, no practical monolithic solution exists today. The same basic product can also be targeted toward the long-haul and ultralong-haul markets by removing the EAM section. Recent results reported at OFC 2001 indicated output power in excess of +10 dBm over 90 wavelengths using this architecture, while maintaining high-side mode suppression ratio (>40dB) and low channel-to-channel power variation (<1.5dB). In this case only the modulator is external to the laser package so significant circuit board real estate is saved.1

Both of these monolithically integrated solutions are packaged using industry-standard, highly automated, packaging technology. The solutions are entirely solid state in nature so that device reliability can be proven through accelerated aging tests that are generally accepted throughout the industry and are well understood.

Other ways to implement tunable laser functionality using either vertical-cavity surface-emitting laser (VCSEL), external cavity laser (ECL), or distributed Bragg reflector (DBR) structures also exist. These solutions do not offer the same integration capability and therefore will have difficulty providing a low-cost solution. Either optically inefficient combinations of discrete chips and/or moving parts are required to perform the wavelength tuning and achieve the optical performance specifications.

These mechanically tuned solutions can be subject to wavelength variations as a result of vibration caused by equipment cooling fans or bump and shock testing. Great lengths must be undertaken to prove laser stability under these conditions if these solutions are to be successful. It is difficult to see how such discrete solutions can economically scale to high volumes and how telecom reliability can be proven.

REFERENCE

1. Greg Fish, Monolithic widely tunable DBR laser, OFC (March 2001).

Kevin Affolter is the director of marketing for Agility Communications, 421 Pine Ave., Santa Barbara, CA 93117. He can be reached at 805-879-9279 or [email protected].