As soon as laser diodes came into being, their disadvantages compared to other types of lasers were as obvious as their advantages. The biggest challenge proved to be controlling the spectral output. The wavelengths of lasers made from the same wafer typically varied by +10 nm and changed with the temperature of the laser by about 0.3 nm/°C. For several important applications, and in particular for optical communications, the spectral output was essentially uncontrolled.
Under control
Variations in wavelength of diode batches were brought under control by improvements to epitaxial-growth processes. Methods for precisely maintaining the diode temperature were standardized early on. Researchers could then focus their attention on obtaining stable single-mode laser output for applications including barcode scanning, laser printing, and optical read/write systems. The explosive growth of wavelength-division multiplexing (WDM) in the late 1990s, though, led to a multibillion-dollar market for frequency-controlled laser diodes. Frequency-stabilized lasers formed the basis, in turn, for high-frequency modulation and tunable laser diodes.
The first generation of optical transmission systems used the laser diodes available at the time, that is, aluminum gallium arsenide (AlGaAs) devices operating around 850 nm. These systems had the high values of loss and dispersion characteristic of silica fibers around this wavelength. The development of indium gallium arsenide phosphide (InGaAsP) lasers allowed the next generation of systems to take advantage of the low dispersion at 1.3 µm (theoretically equal to zero) for silica fibers. Bit transfer rates, however, were still limited to less than 100 Mbit/s by modal dispersion of the multimode laser output within the multimode fibers.
The third generation of optical-communication systems came to rely on single-frequency laser diodes as transmission sources. Gain-guided and other broad-stripe laser-diode designs oscillate in several transverse modes. The invention of index-guided designs in the 1970s, and the more sophisticated designs that followed, constrained oscillation to the lowest-order transverse mode.
Laser diodes with Fabry-Perot cavities—in which laser oscillation is determined by the distance between the cleaved reflecting facets at the device's ends—typically oscillate in at least a few longitudinal modes. Although the cavity is very short, typically less than a millimeter, and the longitudinal modes are thus widely spaced, the gain in the semiconductor is so large that that it easily supports a wide spectrum of oscillation.
In practice, the operation of the laser diode tends to lead to self-selection of a single mode of oscillation. Small changes in temperature, current, or even the smallest amount of retroreflected light will cause the laser to hop between modes or oscillate multimode. For this reason, while the instantaneous bandwidth may be much less than 0.1 nm, 2 to 4 nm is a more realistic bandwidth.
In a groove
Historically, there are four significant designs for stabilizing laser-diode output. Distributed-feedback (DFB) lasers, distributed-Bragg-reflection (DBR) lasers, and external-cavity lasers all rely on diffraction gratings. The gratings in each of these designs provide a high degree of loss for undesired frequency modes to select and narrow the spectrum of oscillation. Cleaved-coupled-cavity (C3) lasers lock a main active section to a separate section that provides a reference frequency; each section has its own independent current.
For DBR lasers, the grating replaces one of the reflecting facets to provide feedback to the cavity. While the device is still integrated, placing the grating outside of the active region simplifies the epitaxy of the laser itself. The DBRs still require that a layer transparent to the laser wavelength be grown over the grating, which causes some difficulties.
External-cavity designs potentially have the narrowest linewidth yet allow the laser to be tuned continuously across the entire gain spectrum. They suffer a distinct disadvantage compared to the other designs in that they are larger, more expensive, and, worst of all for telecommunication, less reliable. It should be noted, however, that some of the latest research developments for miniature surface-emitting arrays involve microscopic external-cavity mechanisms.
The C3 lasers are predecessors of more complex designs now in production and under development. The primary advantage of the C3 design is that the current for the main laser body can be kept constant and separate from the current adjusted to stabilize (or also tune) the laser.
Power is not a concern for any of these designs, because nonlinear effects in fibers limit sources to outputs in the milliwatt range. The simplicity of DFB lasers has made them the overwhelming choice for optical networks, but designs based on the other schemes remain important because of their potential advantages for tunability, a requirement for next-generation WDM devices.
Single-frequency Jell-O
Proposals and theory for frequency control ran ahead of fabrication capability by several years. Distributed-feedback oscillation was studied intensively by Kogelnik and Yariv, among others, in the early 1970s. The first DFB laser was made of gelatin doped with a laser dye on a glass substrate. The Scifres group at Xerox PARC (Palo Alto, CA) made a AlGaAs device using a single-heterojunction DFB design. The first continuous-wave room-temperature DFB device was achieved in 1975 by researchers at Bell Labs (Murray Hill, NJ).
To form a DFB laser, a grating with a pitch of a few hundred nanometers (equal to, say, one-third of the laser wavelength) and a depth of only a few nanometers is formed next to the active layer. To avoid damaging the active layer while making the grating, a layer for separate photon confinement is interposed.
For DFB lasers, unlike conventional Fabry-Perot cavities, there has been concern that the single-mode oscillation might degrade over time. Extensive experience has since confirmed that the small height of the grating corrugations combined with the placement of the optical confinement layer serve to prevent deterioration.
Despite the success of DFB designs in achieving single-mode operation, optical-communication specifications demand still greater stability. A source might be required to maintain 3 GHz of accuracy over a broad range of environmental conditions. The stability of DFB lasers is typically about 10 times less than this, at best. To maintain the required accuracy, a fraction of the laser output is sent to an etalon and a corrective feedback signal generated if the wavelength strays.
Modulation in all things
It is ironic that one of the first advantages of laser diodes over other types was the ability to directly modulate the laser. Modulation for high-speed communications systems is now provided largely by external means. The chirp induced in the source pulse by direct modulation is too great for even the small dispersion in high-speed networks. The use of electroabsorptive modulated lasers (EMLs) and external modulation using lithium niobate (LiNbO3) are now standard practice for these systems.
Speed is the biggest consideration in specifying a modulation design, and LiNbO3 is preferred for the highest-speed applications. However, because of the inherent advantages of smaller size and lower cost of EMLs, research is driving these truly integrated optoelectronic devices to catch up to lithium niobate.
A recent major development effort in spectral control has been to produce tunable laser diodes. Tunable lasers are essential for a nontechnical but still vital WDM concern—inventory control. It is not uncommon to have 80 or more separate channels or frequencies in a WDM system. Providing backup lasers for each individual channel, or maintaining an inventory of replacements at service centers, is a nontrivial task much simplified by the availability of one or a few modules than can be tuned in situ to the required frequency.
Tunable lasers
Tunable lasers also will enhance the add/drop capability of WDM systems, allowing the remote addition of new channels to the system (or removal of old ones) with no changes to the system hardware. In addition, recent experimental work has demonstrated that tunable lasers can direct information packets to different destinations on the same network, based entirely on the frequency of the signal, using only passive optical components.
The three basic approaches to making a tunable laser diode are use of an external cavity, temperature tuning, and current tuning. The external-cavity approach, mentioned earlier, is currently too large and unreliable to be practical. Temperature tuning is an inelegant method, which, while slow and of limited range, has the virtue that it can be used with an installed base of DFB lasers. Modern designs rely on the C3 concept combined with a DBR module, thereby providing separate light-generation and wavelength-selection sections and providing both broad wavelength tuning and low-chirp operation.
These designs can have two, three, or four sections in a single device. If only a gain section and a grating section are used, it is necessary to match the phase of the cavity mode to the phase in the grating section by temperature tuning the cavity. Adding a third current-tuned, phase-matching section eliminates temperature tuning.
One example uses an index-guided module to emit light from the active layer through a waveguide to separate phase- and wavelength-controlling sections. These sections do not contain active layers, but experience a change in their refractive indices when their current is varied. The selected wavelength is reflected back to the amplifying section and emitted though the output facet.
Even more precise
Early in the development of optical communications, scientists envisioned that the technology would be similar to radio in its use of coherent detection. The success of WDM in boosting and simplifying network capacity has offset the attraction of coherent communication, but to make full use of the terahertz potential of fiberoptic networks, it is likely that some form of coherent communications will be needed. Such systems likely will require diode sources that are three orders of magnitude more stable than current designs.
ACKNOWLEDGMENTS
We would like to thank the following individuals and organizations for providing images used in the Semiconductor Lasers 2000 series timeline: Michael W. Davidson, Florida State University; GE Research and Development Center; Zhores I. Alferov; Nick Holonyak, University of Illinois; Dan Botez, University of Wisconsin-Madison; Lucent Technologies, Bell Labs; Connie Chang-Hasnain, University of California Berkeley; Nichia Chemical Corporation; Jack Jewell, Picolight Corp. -Ed.
Next month the series recounts the growth of optical communications.