Sources, Modulators, and Detectors For Fiber-Optic Communication Systems Part 6

Distributed Feedback (DFB) Lasers

When the corrugation is put directly on the active region or its cladding, this is called distributed feedback (DFB). One typical example is shown in Fig. 19. As before, the grating spacing is chosen such that, for a desired wavelength nearwhere now ng is the effective group refractive index of the laser mode inside its waveguiding active region, and m is any integer. A laser operating under the action of this grating has feedback that is distributed throughout the laser gain medium. In this case, Eq. (49) is generalized to allow for the gain:where gL is the laser gain andEquations (49) to (54) remain valid, understanding that now 8 is complex.

The laser oscillation condition requires that after a round-trip inside the laser cavity, a wave must have the same phase that it started out with, so that successive reflections add in phase. Thus, the phase of the product of the complex reflection coefficients (which now include gain) must be an integral number ofThis forces r2 to be a positive real number. So, laser oscillation requires that:

On resonanceso that So is pure real for simple corrugations (k real).

Since the denominator in Eq. (49) is now pure imaginary, r2 is negative and the round-trip condition of Eq. (56) cannot be met. Thus, there is no on-resonance solution to a simple DFB laser with a corrugated waveguide and/or a periodic refractive index.

DFB Threshold. We look for an off-resonance solution to the DFB laser with a corrugated waveguide in the active region (k real). A laser requires sufficient gain that the reflection coefficient becomes infinite. That is,

FIGURE 19 Geometry for a DFB laser, showing a buried grating waveguide that forms the separate confinement heterostructure laser, which was grown on top of a grating-etched substrate. The cross-hatched region contains the MQW active layer. A stripe mesa is etched and regrown to provide a buried het-erostructure laser.

where

By simple algebraic manipulation, Eq. (57) can be written as:

Multiplying and dividing by gives:

The denominator is k2, which, for pure corrugations, is K2. For large gain,so that Eq.

(58) givesInserting this in the numerator, Eq. (60) becomes30:

This is a complex eigenvalue equation that has both a real and an imaginary part, which give both the detuning 8o and the required gain gL. Equating the phases gives:

There is a series of solutions, depending on the value of m. For the largest possible gains,

There are two solutions,These are two modes equally spaced around the Bragg resonance. Converting to wavelength units, the mode detuning becomeswhereis the deviation from the Bragg wavelength.

Consideringthis corresponds to nm. The mode spacing is twice this, or 0.7 nm.

The required laser gain is found from the magnitude of Eq. (61) through

For detuningthe gain can be found by plotting Eq. (64) as a function of gain gL, which gives K(gL), which can be inverted to give gL(K).

These results show that there is a symmetry around 8o = 0, so that there will tend to be two modes, equally spaced around Xo. Such a multimode laser is not useful for communication systems, so something must be done about this. The first reality is that there are usually cleaved facets, at least at the output end of the DFB laser. This changes the analysis from that given here, requiring additional Fresnel reflection to be added to the analysis. The additional reflection will usually favor one mode over the other, and the DFB will end up as a single mode. However, there is very little control over the exact positioning of these additional cleaved facets with respect to the grating, and this has not proven to be a reliable way to achieve single-mode operation. The most common solution to this multimode problem is to use a quarter-wavelength-shifted grating, as shown in Fig. 20. Midway along the grating, the phase changes by n/2 and the two-mode degeneracy is lifted. This is the way that DFB lasers are made today.

FIGURE 20 Side view of a quarter-wavelength-shifted grating, etched into a separate confinement waveguide above the active laser region. Light with wavelength in the medium Xg sees a n/4 phase shift, resulting in a single-mode DFB laser operating on line-center.

Quarter-Wavelength-Shifted Grating. Introducing an additional phase shift of n to the round-trip optical wave enables an on-resonance DFB laser. Thus, light traveling in each direction must pass through an additional phase shift ofThis is done by interjecting an additional phase region of lengthas shown in Fig. 20. This provides an additional phase in Eq. (63), so that the high-gain oscillation condition becomes:

Now there is a unique solution at m = 0, given by Eq. (64) with

Given a value for KL, the gain can be calculated. Alternatively, the gain can be varied, and the coupling coefficient used with that gain can be calculated. It can be seen that if there are internal losses a the laser must have sufficient gain to overcome them as well: gL + a;.

Quarter-wavelength-shifted DFB lasers are commonly used in telecommunications applications. There are a variety of ways in which the DFB corrugations are placed with respect to the active layer. Most common is to place the corrugations laterally on either side of the active region, where the evanescent wave of the guided mode experiences sufficient distributed feedback for threshold to be achieved. Alternative methods place the corrugations on a thin cladding above the active layer. Because the process of corrugation may introduce defects, it is traditional to avoid corrugating the active layer directly. Once a DFB laser has been properly designed, it will be single mode at essentially all power levels and under all modulation conditions. Then the single-mode laser characteristics described in the early part of this topic will be well satisfied. However, it is crucial to avoid reflections from fibers back into the laser, because instabilities may arise, and the output may cease to be single mode.

A different technique that is sometimes used is to spatially modulate the gain. This renders k complex and enables an on-resonance solution for the DFB laser, since S will then be complex on resonance. Corrugation directly on the active region makes this possible, but care must be taken to avoid introducing centers for nonradiative recombination.

There have been more than 35 years of research and development in semiconductor lasers for telecommunications. Today it appears that the optimal sources for telecommunications applications are strained quantum well distributed feedback lasers at 1.3 or 1.55 |im.

LIGHT-EMITTING DIODES (LEDS)

Sources for low-cost fiber communication systems, such as are used for communicating data, are typically light-emitting diodes (LEDs). These may be edge-emitting LEDs (E-LEDs), which resemble laser diodes, or, more commonly, surface-emitting LEDs (S-LEDs), which emit light from the surface of the diode and can be butt-coupled to multimode fibers.

When a PN junction is forward biased, electrons are injected from the N region and holes are injected from the P region into the active region. When free electrons and free holes coexist with comparable momentum, they will combine and may emit photons of energy near that of the bandgap, resulting in an LED. The process is called injection (or electro-) luminescence, since injected carriers recombine and emit light by spontaneous emission. A semiconductor laser diode below threshold acts as an LED. Indeed, a semiconductor laser without mirrors is an LED. Because LEDs have no threshold, they usually are not as critical to operate and are usually much less expensive. Also, they do not need the optical feedback of lasers (in the form of cleaved facets or distributed feedback). Because the LED operates by spontaneous emission, it is an incoherent light source, typically emitted from a larger aperture (out the top surface) with a wider far-field angle and a much wider wavelength range (30 to 50 nm). In addition, LEDs are slower to modulate than laser diodes. Nonetheless, they can be excellent sources for inexpensive multimode fiber communication systems. Also, LEDs have the advantages of simpler fabrication procedures, lower cost, and simpler drive circuitry. They are longer lived, exhibit more linear input-output characteristics, are less temperature sensitive, and are essentially noise-free electrical-to-optical converters. The disadvantages are lower power output, smaller modulation bandwidths, and distortion in fiber systems because of the wide wavelength band emitted.

In fiber communication systems, LEDs are used for low-cost, high-reliability sources typically operating with graded index multimode fibers (core diameters approximately 62 |m) at data rates up to 622 Mb/s. The emission wavelength will be at the bandgap of the active region in the LED; different alloys and materials have different bandgaps. For medium-range distances up to -10 km (limited by modal dispersion), LEDs of InGaAsP grown on InP and operating at X = 1.3 |m offer low-cost, high-reliability transmitters. For short-distance systems, up to 2 km, GaAs-based LEDs operating near 850 nm wavelength are used, because they have the lowest cost, both to fabricate and to operate, and the least temperature dependence. The link length is limited to -2 km because of chromatic dispersion in the fiber and the finite linewidth of the LED. For lower data rates (a few megabits per second) and short distances (a few tens of meters), very inexpensive systems consisting of red-emitting LEDs with GaAlAs or GaInP active regions emitting at 650 nm can be used with plastic fibers and standard silicon detectors. The 650-nm wavelength is a window in the absorption in acrylic plastic fiber, where the loss is -0.3 dB/m.

A typical GaAs LED heterostructure is shown in Fig. 21. The forward-biased pn junction injects electrons and holes into the GaAs active region. The AlGaAs cladding layers confine the carriers in the active region. High-speed operation requires high levels of injection (and/or doping) so that the recombination rate of electrons and holes is very high. This means that the active region should be very thin. However, nonradiative recombination increases at high carrier concentrations, so there is a trade-off between internal quantum efficiency and speed. Under some conditions, LED performance is improved by using quantum wells or strained layers. The improvement is not as marked as with lasers, however.

Spontaneous emission causes light to be emitted in all directions inside the active layer, with an internal quantum efficiency that may approach 100 percent in these direct band semiconductors. However, only the light that gets out of the LED and into the fiber is useful in a communication system, as illustrated in Fig. 21a. The challenge, then, is to collect as much light as possible into the fiber end. The simplest approach is to butt-couple a multimode fiber to the LED surface as shown in Fig. 21a (although more light is collected by lensing the fiber tip or attaching a high-index lens directly on the LED surface). The alternative is to cleave the LED, as in a laser (Fig. 1), and collect the waveguided light that is emitted out the edge. Thus, there are two generic geometries for LEDs: surface-emitting and edge-emitting. The edge-emitting geometry is similar to that of a laser, while the surface-emitting geometry allows light to come out the top (or bottom). Its inexpensive fabrication and integration process makes the surface-emitting LED the most common type for inexpensive data communication; it will be discussed first. The edge-emitting LEDs have a niche in their ability to couple with reasonable efficiency into single-mode fibers. Both LED types can be modulated at bit rates up to 622 Mb/s, an ATM standard, but many commercial LEDs have considerably smaller band-widths.

FIGURE 21 Cross-section of a typical GaAs light-emitting diode (LED) structure: (a) surface-emitting LED aligned to a multimode fiber, indicating the small fraction of spontaneous emission that can be captured by the fiber; (b) energy of the conduction band Ec and valence band Ev as a function of depth through the LED under forward bias V, as well as the Fermi energies that indicate the potential drop that the average electron sees.

Surface-Emitting LEDs

The geometry of a surface-emitting LED butt-coupled to a multimode graded index fiber is shown Fig. 21a. The coupling efficiency is typically small, unless methods are employed to optimize it. Because light is spontaneously emitted in all internal directions, only half of it is emitted toward the top surface, so that often a mirror is provided to reflect back the downward-traveling light. In addition, light emitted at too great an angle to the surface normal is totally internally reflected back down and is lost. The critical angle for total internal reflection between the semiconductor of refractive index ns and the output medium (air or plastic encapsulant) of refractive index no is given by sin 9c = no/ns. Because the refractive index of GaAs is ns ~ 3.3, the internal critical angle with air is 9c ~ 18°. Even with encapsulation, the angle is only 27°. A butt-coupled fiber can accept only spontaneous emission at those external angles that are smaller than its numerical aperture. For a typical fiber NA ~ 0.25, this corresponds to an external angle (in air) of 14°, which corresponds to 4.4° inside the GaAs. This means that the cone of spontaneous emission that can be accepted by the fiber is only ~0.2 percent of the entire spontaneous emission. Fresnel reflection losses makes this number even smaller. Even including all angles, less than 2 percent of the total internal spontaneous emission will come out the top surface of a planar LED.

The LED source is incoherent, a Lambertian emitter, and follows the law of imaging optics: a lens can be used to reduce the angle of divergence of LED light, but will enlarge the apparent source. The use of a collimating lens means that the LED source diameter must be proportionally smaller than the fiber into which it is to be coupled. Unlike a laser, the LED has no modal interference, and the output of a well-designed LED has a smooth Lambertian intensity distribution that lends itself to imaging.

The coupling efficiency can be increased in a variety of ways, as shown in Fig. 22. The LED can be encapsulated in materials such as plastic or epoxy, with direct attachment to a focusing lens (Fig. 22a). Then the output cone angle will depend on the design of this encapsulating lens; the finite size of the emitting aperture and resulting aberrations will be the limiting consideration. In general, the user must know both the area of the emitting aperture and the angular divergence in order to optimize coupling efficiency into a fiber. Typical commercially available LEDs at 850 nm for fiber-optic applications have external half-angles of ~25° without a lens and ~10° with a lens, suitable for butt-coupling to multimode fiber.

Additional improvement can be achieved by lensing the pigtailed fiber to increase its acceptance angle (Fig. 22b). An alternative is to place a microlens between the LED and the fiber (Fig. 22c). Perhaps the most effective geometry for capturing light is the integrated domed surface fabricated directly on the back side of an InP LED, as shown in Fig. 22d. Because the refractive index of encapsulating plastic is <1.5, compared to 3.3 of the semi-conductor, only a semiconductor dome can entirely eliminate total internal reflection. Integrated semiconductor domes require advanced semiconductor fabrication technology, but have been proven effective. In GaAs diodes the substrate is absorptive, but etching a well and inserting a fiber can serve to collect backside emission. For any of these geometries, improvement in efficiency of as much as a factor of two can be obtained if a mirror is provided to reflect backward-emitted light forward. This mirror can be either metal or a dielectric stack at the air-semiconductor interface, or it can be a DBR mirror grown within the semiconductor structure.

Current must be confined to the surface area of emission, which is typically 25 to 75 |im in diameter. This is done by constricting the flow of injection current by mesa etching or by using an oxide-defined (reflective) electrode. Regrowth using npn blocking layers or semi-insulating material in the surrounding areas (as in lasers) has the advantage of reducing thermal heating. Surface-emitting LEDs require that light be emitted out of the surface in a gaussian-like pattern; it must not be obscured by the contacting electrode. Typically, a highly conductive cap layer brings the current in from a ring electrode; alternatively, when light is collected out of the substrate side rather than the top side, electrical contact may be made to the substrate.

FIGURE 22 Typical geometries for coupling from LEDs into fibers: (a) hemispherical lens attached with encapsulating plastic; (b) lensed fiber tip; (c) microlens aligned through use of an etched well; and (d) spherical semiconductor surface formed on the substrate side of the LED.

Typical operating specifications for a surface-emitting LED at 1.3 |m pigtailed to a 62-|m core graded index fiber might be 15 |W at 100 mA input current, for -0.02 percent efficiency,31 with a modulation capability of 622 Mb/s. A factor of 2.5 times improvement in power can be achieved with a comparable reduction in speed. The LEDs are typically placed in lensed TO-18 cans, and a lens micromachined on the back of the InP die is used to achieve this output coupling efficiency. At 1.55 |m, the specifications are for 7 times less power and 3 times less speed.

Recently, improved S-LED performance has been obtained by using resonant cavities to reduce the linewidth and increase the bandwidth that can be transmitted through fibers. These devices have integral mirrors grown above and below the active region that serve to resonate the spontaneous emission. As such, they look very much like VCSELs below threshold (Sec. 4.9).

Edge-Emitting LEDs

Edge-emitting LEDs (E-LEDs or EELEDs) have a geometry that is similar to that of a conventional laser diode (Fig. 1), but without a feedback cavity. That is, light travels back and forth in the plane of the active region of an E-LED and it is emitted out one anti-reflection coated cleaved end. As in a laser, the active layer is 0.1 to 0.2 |m thick. Because the light in an E-LED is waveguided in the out-of-plane dimension and is lambertian in-plane, the output radiation pattern will be elliptical, with the largest divergence in-plane with a full width at half-maximum (FWHM) angle of 120°. The out-of-plane guided direction typically radiates with a 30° half-angle. An elliptical collimating lens will be needed to optimally couple light into a fiber. The efficiency can be doubled by providing a reflector on the back facet of the E-LED, just as in the case of a laser.

Edge-emitting LEDs can be coupled into fibers with greater efficiency because their source area is smaller than that of S-LEDs. However, the alignment and packaging is more cumbersome than with S-LEDs. Typically, E-LEDs can be obtained already pigtailed to fibers. Edge-emitting diodes can be coupled into single-mode fiber with modest efficiency. A single-mode fiber pigtailed to an E-LED can typically transmit 30 |W at 150 mA drive at 1 V, for an overall efficiency of 0.04 percent. This efficiency is comparable to the emission of surface-emitting lasers into multimode fiber with 50 times the area. Because of their wide emission wavelength bandwidth, E-LEDs are typically used as low-coherence sources for fiber sensor applications, rather than in communications applications.