Like it or not, component manufacturers will have to start thinking about higher levels of integration.
IGOR E. TROFIMOV,
Qusion Technologies Inc.
Optics is well entrenched in long-haul telecommunications networks, and the crusade continues as the industry tries to bring fiber to the home. This ambitious plan is facing tremendous challenges due to supply constraints in optical components, however.
In a perfect world, service providers would cut the umbilical cord of SONET and build all-optical networks of highly meshed DWDM that offer instantaneous traffic reconfiguration and fault recovery, without the need for electrical restoration. In reality, components enabling this type of architecture aren't available-yet.
As a result, carriers deploy more DWDM systems and SONET hardware. While this approach undoubtedly increases network capacity, it also calls for ever-increasing component production.
Three trends will drive developments related to component manufacturing:
- Transmission rates have quadrupled every two years for the last decade. Widespread deployment of 40-Gbit/sec systems is expected next year. Since the bandwidth of existing silicon-based electronics already has been exhausted, this transition is more challenging than previous upgrades since it will involve the development of electronics based on new materials.
- Channel density in WDM systems has been gradually increasing in the past decade. This trend demands higher tolerances in manufacturing of both optical and electronic components, and eventually will require new production technologies to replace "bolt-together" discrete components.
- New wavelength bands to accommodate more DWDM channels are actively being pursued. It is expected that within the next few years, the transmission of about 1,000 channels across a 400-nm-wide band will be possible. Deployment of such systems calls for efficient high-volume manufacturing.
No matter how difficult these challenges are, the biggest obstacle by far is the packaging of photonic components. Packaging accounts for roughly 60% to 90% of the cost in photonic-component manufacturing. Packaging also accounts for as much as 50% of the reduction in production-in a manufacturing arena known for notoriously low yield.
Many developers see a panacea in the integration of multiple components under the same hood. This approach not only decreases packaging costs, it relieves system manufacturers from the burden of low-level design. Companies will realize savings from cost reductions in development, technician training, and service in the field. Customers will recognize the advantages of working with multifunctional modules rather than individual components.
The issue is complicated by the wide range of optical materials and technologies, many of which pose integration challenges, resulting in costly and labor-intensive manufacturing processes. Ramping up existing manufacturing techniques-the bolt-it-together approaches-is not a solution. DWDM and high-bit-rate modules (above 10 Gbits/sec) may reach a tolerance threshold beyond which this method will cease to be productive.
One way or another, the industry needs more components to enable higher communication speeds at lower prices. Integration is clearly the key, but to what degree and in what form?
A step beyond the bolt-it-together approach is silicon optical-bench (SiOB) technology. The idea behind SiOB is similar to the use of the printed circuit board in the electronics industry. A network of silica-based optical waveguides is fabricated using precision photolithography and planar patterning techniques on the silicon substrate. Then, precision V-grooves-used for the passive aligning of fiber to waveguides-are added by means of an isotropic wet etching. The same technique, or reactive ion etching, is used for defining pits and standoffs for precision aligning of other components in the assembly. Finally, flip-chip bonding technology, either in self-aligning or active-aligning mode, allows not only the positioning of active components on the board, but also the establishment of electrical contacts (see Figure 1).
Silica-waveguide losses are very low (<0.01 dB/cm), which makes this technology highly attractive for the large-scale planar lightwave circuits used in waveguide routers (WGRs), switching matrixes, and star couplers. Silica-based technology is also benefiting from the advancements of the electronics industry, from the sophisticated equipment available at every step of manufacturing to the large-sized, reasonably priced wafers-a prerequisite for low-cost, scalable production. Many manufacturers, including Nortel Networks, Fujitsu, NTT, NEC, and Lucent Technologies, apply this technology for manufacturing transceiver modules; each company has its own modifications.
Bookham Technology went a step further, integrating passive and active electro-optical components on the same substrate using proprietary ASOC silicon-based manufacturing technology. The company's product line includes transceivers, variable optical attenuators, and WGR multiplexers and demultiplexers.
No matter how attractive SiOB is as an integration platform, it is already close to its limits for monolithic integration. Silica does not possess attractive material properties for manufacturing active components such as lasers, modulators, switches, and detectors. Recent progress in erbium-doped planar-waveguide amplifiers could become a big hit in SiOB integrated modules, adding compatibility for a monolithic integration of light sources and introducing a whole realm of loss-less passive components.
An interesting alternative to the silica/silicon platform is lithium niobate (LiNbO3)-based integrated optics. Low-loss waveguides are made by titanium diffusion into LiNbO3. Rare-earth ion doping of these waveguides creates a media suitable for optically pumped amplifiers and lasers. Moreover, unlike silica, LiNbO3 is an excellent electro-optical material, the same property that made LiNbO3 modulators a big commercial success. A number of LiNbO3 distributed Bragg reflector (DBR) lasers, tunable lasers, and optical amplifiers have already been demonstrated in labs. Successful monolithic integration of a DBR waveguide laser and Mach-Zehnder (MZ) intensity modulator as a part of transmitter module also has been reported.
All the benefits of LiNbO3 come at a price. Waveguide losses are higher (<0.1 dB/cm) than silica-based components, and the wafers are smaller and more expensive. Most importantly, this material lacks the manufacturing expertise developed for silica/silicon in the semiconductor industry. So far, despite all research and development efforts, commercial products have not made it past a standalone modulator.
Until recently, polymeric materials were primarily considered a cheaper replacement for glass in optical fibers. The discovery of electro-optical and light-emitting polymers immediately made these materials a serious contender in the field of low-cost photonic integrated circuits. Theoretically, polymers could provide the functionality of silicon/silica but with a substantial advantage in the cost structure. Polymers have two basic methods of manufacturing: precision injection molding and spin-on. Polymer manufacture also permits the use of cheaper raw materials and the integration with polymer fibers. These factors indicate that polymers have the potential to surpass SiOB as an integration platform.
Vertical-cavity surface-emitting lasers (VCSELs) have recently attracted a lot of attention as a low-cost technology for telecommunications. VCSELs can also be used for optical board-to-board or rack-to-rack interconnects for data communications. VCSELs emit light perpendicular to the wafer plane, which makes them simpler to couple with arrays of fibers and to integrate with driver electronics. Fabrication of high-performance VCSELs involves relatively cheap, high-yield methods, including a single epitaxial growth, wet-thermal oxidation technique, and relaxed tolerance coupling to multimode fiber. The combination produces a difficult-to-beat cost structure.
Monolithic integration using indium phosphide (InP)-based compound semiconductors may be the "ultimate solution" to the problem. This method not only promises to put active and passive optical components interconnected by a network of waveguides in the most compact assemblies, but also allows electronic and micromechanical components to be manufactured on the same chip (see Figure 2). InP provides an excellent substrate for epitaxial growth of indium gallium aluminum arsenide (InGaAlAs) and indium gallium arsenide phosphide (InGaAsP), the two major materials for telecom lasers and detectors and the most promising candidates for ultra-high-speed electronics. InP has good thermal conductivity and high optical damage threshold and eliminates material compatibility problems typical of silicon-based integration. Overall, InGaAlAsP/InP monolithic integration has huge potential for delivering devices with novel functionality for all-optical networks.
This approach is highly scalable, as demonstrated by its tremendous success in the microelectronics industry. Once the design is done, many identical chips can be manufactured on the wafer at the same time. Growth in the wafer's diameter results in the increase of the number of chips manufactured, lowering the price of components.
Most of the methods of monolithic integration are the same as those used in the manufacturing of individual components and hybrid modules-photolithography, wet and dry etching, and bonding, among others. The epicenter of activity here is in growth. The flexibility of III-V semiconductor materials allows the manufacturing of all kinds of components; however, the same flexibility is a nightmare for everyone involved in the manufacturing. Two methods of epitaxial growth compete as the best solution: metalloid organic chemical-vapor deposition and gas source molecular-beam epitaxy. Both methods provide very accurate material flux control, a prerequisite for the highly uniform composition and layer thickness employed in state-of-the-art photonic and electronic devices.
The key technological challenge for monolithic integration is how to achieve high spatial resolution when changing the material properties on the wafer. The transition region between different components should not only be sharp, but also should provide excellent coupling efficiency and eventually no backreflections. Several techniques have been developed to achieve this goal, none of which are perfect:
- Regrowth is by far the most flexible and widely used method. It allows the growth of a sequence of epitaxial layers for one component (usually active) on the chip, then locally removes regions on the wafer for the next component by wet or dry etching. It then grows another sequence of epitaxial layers in a separate step. This procedure can be repeated as many times as needed. The butt-coupling efficiency between components of better than 90% is achieved by this technique. Since every component is manufactured independently, the material properties for each component, such as thickness and composition of the layers, doping profile, and concentration, can be designed to optimize performance without constraints imposed by the next component. This method is complex and cumbersome, requires a fair amount of finesse, and worst of all, is expensive.
- Selective-area epitaxy (SAE) is based on a discovery made in the early 1980s that the growth rate and composition of III-V compounds on a patterned surface can differ drastically from those on the flat wafer. In most common interpretations of this method, very narrow openings in SiO2 masks photolithographically manufactured on the surface of an InP wafer are used to locally inhibit film growth. Because optical properties of multiple quantum well (MQW) depend strongly on the thickness of constituting layers, it is possible to achieve a bandgap difference of more then 40 nm between the masked and unmasked regions on the wafer. The obtained variations in material properties are big enough to construct a wide spectrum of devices, from lasers to detectors to low-loss waveguides. Certainly, this method lacks the flexibility of regrowth, and requires tradeoffs in design of components. The biggest problem involved in SAE is photolithographic processing of the wafer before epitaxy. Meticulous surface treatment is necessary to ensure subsequent growth of high quality semiconductor materials.
- Disordering method is quite similar to SAE, involving local modification of structural parameters of MQW. Here, after the initial growth of the wafer with MQW optimized for the active components is done, local structural disordering of MQW is performed by means of heat, local diffusion, or ion implantation. The bandgap of material in the areas subjected to treatment increases, allowing low-loss waveguide structures to be manufactured.
- Dual waveguide is another approach to reduce the number of regrowth steps by the design of coupled waveguides containing an active layer on top of the passive layer. Following the growth of such structure on the whole wafer, the active layer is removed by etching in the areas of passive components. Care should be taken to establish smooth mode conversion between dual- and single-waveguide regions. Substantial restrictions on active device design are imposed by dual waveguide, which can compromise performance.
Since its first demonstration in the early 1980s, InP-based monolithic integration has reached the production stage, and several modules are now commercially available. A receiver based on a pin diode with a transimpedance amplifier was the first successful manufacturing attempt of telecom modules using monolithic integration on InP. It still remains the only example of integration of photonic and electronic components on the same chip.
Historically, hybrid receiver modules outperformed modules based on InP monolithic integration. Both the cost and technical performance were hampered by immaturity of III-V materials growth and processing technologies in comparison with silicon. As bit rates approached 10 Gbits/sec, however, this handicap was practically eliminated. The balance may shift in favor of monolithic integration with 40-Gbit/ sec systems. The reason is not only the ever-improving technology of III-V semiconductors, but also that the material properties of silicon do not allow its use in electronics at such high frequencies. SixGe1-x, while an adequate relay at 10 Gbits/sec, may hit some roadblocks at 40 Gbits/sec, particularly in applications requiring high breakdown voltage, such as modulator drivers; it will almost inevitably give in beyond 40 Gbits/sec. The first high-performance monolithically integrated receivers will become available this year from u2t, which manufactures 40-Gbit/sec receivers using pin photodiodes with high electron mobility transfer (HEMT)-based amplifiers.
The other commercially available product that employs monolithic integration is a transmitter. In its simplest form, the transmitter consists of a laser and modulator connected via a waveguide. A transmitter developed by Lucent Technologies features a DFB laser monolithically integrated with electro-absorption MQW modulators. The device operates at data rates as high as 10 Gbits/sec and requires a lower driving voltage than a LiNbO3 modulator. Transmitter modules consisting of DFB lasers monolithically integrated with MZ modulators are available from Nortel.
At first glance, the slow progress of InP-integration in commercial applications does not justify its pretension. New devices under development may appear soon, however. These integrated components are necessary if the transition to an all-optical network of highly meshed DWDM without electrical regeneration-and with instantaneous traffic reconfiguration and fault recovery-is to become a reality. These devices include 3R (receive, reframe, retransmit) optical regenerators based on semiconductor optical amplifiers (SOAs) integrated with an MZ interferometer; wavelength converters based on nonlinear effects in SOAs; optical add/drop multiplexers based on integration of WGRs with 2x2 switches; all-optical switches; and optical time-division multiplexers.
Despite the attractiveness of InP monolithic integration in performance and manufacturing costs, the technology is taking its first steps in large-scale industrial applications. These materials are lossy (>0.25 dB/cm) and brittle. Their optical properties are strongly nonlinear and temperature-dependent, which implies the use of thermoelectric coolers. The InP wafers are expensive and small; optical density does not exceed 3 inches. Strong mode confinement, which can be achieved using InP-based materials, is beneficial for very compact devices, because the bend radii of the waveguides can be as sharp as 100 microns. At the same time, it calls for extremely tight fabrication tolerances.
The big difference in mode diameter between InP devices and silica fibers causes huge coupling losses. This problem is often cited as a major hurdle for integration. Indeed, coupling light in and out of waveguiding components would require expensive focusing optics, precision mechanics, and active-alignment with submicron tolerances.
Fortunately, the solution has been found in spot-size converters, which are integrated waveguide devices that smoothly transform the spot-size without incurring many losses. A point often overlooked is that monolithically integrated modules do not need to deal with multiple couplings of components with dissimilar mode parameters. Each component is made of InP-based materials, so the mode variations are reduced to a minimum. And while coupling to the fiber is still required for communications with the outside world, it is done only once at the input and once at the output of the module. All the components inside the module are automatically aligned by design.
The same can be said with regard to the polarization sensitivity of devices. While most InP components are inherently polarization-sensitive, this problem can be mitigated by proper design of the integrated module, instead of making every component polarization-insensitive.
The biggest problem with InP monolithic integration today is an absence of manufacturing equipment. Despite the tremendous achievements of the semiconductor industry in processing, growth of materials, and packaging of components, the photonic industry cannot take full advantage of these advances because there is no specialized equipment of the same class for optoelectronics. The photonics industry remains a foster child in the eyes of state-of-the-art capital equipment manufacturers. These companies do not see enough market in 2-inch InP wafer production lines. The introduction of large-scale integrated modules for DWDM, which require a lot of wafer real estate, high-volume manufacturing of photonic integrated circuits, and ultra-high speed electronics, will hopefully change this perception.
Integration is not a catchy word in the lexicon of telecommunications executives, but it is a bare-bones necessity in the optical-telecom industry. The specific requirements for integrated devices depend on their application and the network architecture. The strongest driving force for monolithic integration from a market perspective is the high-volume, low-cost requirements of access networks.
On the other hand, high-end ultra-fast systems call for new materials and manufacturing techniques. InP monolithic integration is also a possible answer for this challenge. We may even see a merger of ultra-high-speed electronics and photonics in the not-so-remote future.
Igor E. Trofimov is executive vice president of Qusion Technologies Inc. (Princeton, NJ).