Sources, Modulators, and Detectors For Fiber-Optic Communication Systems Part 7

Operating Characteristics of LEDs

In an LED, the output optical power Popt is linearly proportional to the drive current; the relation defines the output efficiency n:

tmp8-586_thumb[2][2][2][2][2]_thumb

This efficiency is strongly affected by the geometry of the LED. The power coupled into a fiber is further reduced by the coupling efficiency between the LED emitter and the fiber, which depends on the location, size, and numerical aperture of the fiber as well as on the spatial distribution of the LED output light and the optics of any intervening lens. The internal quantum efficiency (ratio of emitted photons to incident electrons) is usually close to 100 percent.

Figure 23 shows a typical result for power coupled into a graded index multimode fiber as a function of current for various temperatures. The nonlinearity in the light out versus current, which is much less than in a laser diode, nevertheless causes some nonlinearity in the modulation of LEDs. This LED nonlinearity arises both from material properties and device configuration; it may be made worse by ohmic heating at high drive currents. The residual nonlinearity is an important characteristic of any LED used in communication systems. Edge emitters are typically less linear because they operate nearer the amplified spontaneous limit.


There is ~10 percent reduction in output power for a 25°C increase in temperature (compared to ~50 percent reduction for a typical laser). Unlike a laser, there is no temperature-dependent threshold. Also, the geometric factors that determine the fraction of light emitted from the LED are not temperature dependent. Nonetheless, the InP-based LEDs have a stronger temperature dependence than GaAs-based LEDs, because of the larger presence of nonradiative recombination, particularly at the high injection levels required by high-speed LEDs.

The spectrum of the incoherent light emitted from an LED is roughly gaussian with a FWHM around 40 nm in the case of a typical GaAs/AlGaAs LED operating around 0.8 |im. This bandwidth, along with chromatic dispersion in graded index fibers, limits the distance over which these LEDs can be used in fiber systems. InGaAsP/InP LEDs have wider linewidths (due to alloy scattering, heavy doping, and temperature fluctuations), which depend on the details of their design. As temperature increases, the peak of the spectrum shifts to longer wavelength and the spectrum widens. The variation of the central wavelength with temperature is ~5 meV/°C. However, at 1.3 |im, graded index fibers have negligible chromatic dispersion, so this usually is not a problem; if it is, heat sinking and/or cooling can be provided. Resonant cavity LEDs can provide narrower linewidths, but are more difficult to fabricate.

LEDs do not suffer from the catastrophic optical damage that lasers do, because of their lower optical power densities. However, they do degrade with time. Lifetimes of 106 to 109 hours can be expected. Because degradation processes have an exponential dependence on temperature, LED life can be shortened by operating at excessive temperatures. Using concepts of thermally accelerated life testing, the power out P varies with time t as:

tmp8-587_thumb[2][2][2][2][2][2]

wheretmp8-588_thumb[2][2][2][2][2][2], with Wa as the activation energy, kB as Boltzman’s constant, and T as temperature. In GaAs LEDs, Wa is 0.6 to 1 eV. Of course, this assumes that the LEDs are placed in a proper electrical circuit.

Optical power coupled from an InGaAsP S- LED into graded index fiber at 1.3 |im wavelength as a function of drive current, for several temperatures.

FIGURE 23 Optical power coupled from an InGaAsP S- LED into graded index fiber at 1.3 |im wavelength as a function of drive current, for several temperatures.

LED light is typically unpolarized, since there is no preferred polarization for spontaneous emission.

Transient Response

Most LEDs respond in times faster than 1 |s; with optimization, they can reach the nanosecond response times needed for optical communication systems. To achieve the 125 Mb/s rate of the fiber distributed data interface (FDDI) standard requires a maximum rise time and fall time of 3.5 ns; to achieve the 622 Mb/s rate of the asynchronous transfer mode (ATM) standard, the necessary times drop to 0.7 ns.

The speed of an LED is limited by the lifetime of injected carriers; it does not have the turn-on delay of lasers, nor any relaxation oscillations, but it also does not have the fast decay of stimulated emission. The LED intrinsic frequency response (defined as the ratio of the AC components of the emitted light to the current) is33:

tmp8-591_thumb[2][2][2][2][2][2]

where t is the minority carrier lifetime in the injected region. It can be seen that high-speed LEDs require small minority carrier lifetimes. The square-root dependence comes out of solving the rate equations.

When the active region is doped more highly than the density of injected carriers (the low-injection regime), the lifetime tl is determined by the background doping density No:

tmp8-592_thumb[2][2][2][2][2][2]

The lifetime decreases as the doping increases. The challenge is to provide high levels of doping without increasing the fraction of nonradiative recombination. The fastest speeds that are usually obtained are -1 ns, although doping with beryllium (or carbon) at levels as high as 7 x 1019 cm-3 has allowed speeds to increase to as much as 0.1 ns, resulting in a cutoff frequency of 1.7 GHz (at the sacrifice of some efficiency).34

When operating in the high-injection regime, the injected carrier density N can be much larger than the doping density, andtmp8-593_thumb[2][2][2][2][2][2]But N is created by a current density J such that tmp8-594_thumb[2][2][2][2][2][2]Combining these two equations:

tmp8-597_thumb[2][2][2][2][2][2]

The recombination time may be reduced by thinning the active region and by increasing the drive current. However, too much injection may lead to thermal problems, which in turn may cause modulation nonlinearity. LEDs with thin active layers operated in the high-injection regime will have the fastest response. Bandwidths in excess of 1 GHz have been achieved in practical LEDs.

Because LEDs have such wide wavelength spectra, frequency chirping is negligible. That is, LEDs cannot be modulated fast enough for their wavelengths to be affected by the modulation. Because LEDs do not have optical cavities, as do lasers, they will not have modal interference and noise. Also, there will not be strong feedback effects coming from external fiber facets, such as the coherence collapse. Because of their inherent light-current linearity, the modulation response of LEDs should be a direct measure of their frequency response. They add no noise to the circuit, and they add distortion only at the highest drive levels.

Drive Circuitry and Packaging

The LED is operated under sufficient forward bias to flatten the bands of the pn junction. This voltage depends on the bandgap and doping and is typically between 1 and 2 V. The current will be converted directly to light; typically, ~100 mA is required to produce a few milliwatts of output, with a series resistor used to limit the current.

The LED is modulated by varying the drive current. A typical circuit might apply the signal to the base circuit of a transistor connected in series with the LED and a current-limiting resistor. The variation in current flowing through the LED (and therefore in the light out) is proportional to the input voltage in the base circuit. LEDs are typically mounted on standard headers such as TO-18 or TO-46 cans; SMA and ST connectors are also used. The header is covered by a metal cap with a clear glass top through which light can pass.

VERTICAL CAVITY SURFACE-EMITTING LASERS (VCSELS)

The vertical cavity surface-emitting laser (VCSEL) has advantages for low-cost data transmission. The use of a laser means that multigigahertz modulation is possible, and the stimulated emission is directional, rather than the isotropic spontaneous emission of LEDs. Because the light is emitted directly from the surface, single or multimode fiber can be directly butt-coupled with an inexpensive mounting technology, and the coupling efficiency can be very high. The VCSELs can also be fabricated in linear arrays that can be coupled inexpensively to linear arrays of fibers for parallel fiber interconnects with aggregate bit rates of several gigabits per second, amortizing the alignment cost over the number of elements in the array. VCSELs lend themselves to two-dimensional arrays as well, which makes them attractive to use with smart pixels. The planar fabrication of VCSELs allows for wafer-scale testing, another cost savings.

The VCSEL requires mirrors on the top and bottom of the active layer, forming a vertical cavity, as shown in Fig. 24. These lasers utilize the fact that a DBR (multilayer quarter-wavelength dielectric stack) can make a very high reflectance mirror. Thus, the very short path length through a few quantum wells (at normal incidence to the plane) is sufficient to reach threshold.

One example of a vertical cavity surface emitting laser (VCSEL) geometry. This is a passive antiguide region (PAR) VCSEL.35 Light is reflected up and down through the active region by the two DBR mirrors. After the laser post is etched, regrowth in the region outside the mesa provides a high-refractive-index AlGaAsnipi region to stop current flow and to provide excess loss to higher-order modes.

FIGURE 24 One example of a vertical cavity surface emitting laser (VCSEL) geometry. This is a passive antiguide region (PAR) VCSEL.35 Light is reflected up and down through the active region by the two DBR mirrors. After the laser post is etched, regrowth in the region outside the mesa provides a high-refractive-index AlGaAsnipi region to stop current flow and to provide excess loss to higher-order modes.

In the 1990s, the only commercial VCSELs were based on GaAs: either GaAs active regions that emit at 850 nm, or strained InGaAs active regions that emit at 980 nm. The former are of greater interest in low-cost communication systems because they are compatible with inexpensive silicon detectors. This section describes the design of VCSELs and some of their key characteristics.

Number of Quantum Wells

A single quantum well of GaAs requires -100 A/cm2 to achieve transparency; N wells require N times this current. To keep the threshold current less than 1 kA/cm2, then, means less than 10 QWs. The VCSEL provides an optical standing wave which, in GaAs, has a period of -120 nm. The gain region should be confined to the quarter-wavelength region at the peak of the optical standing wave, a region of about 60 nm. Thus, a typical active region might consist of 3 QWs of 10 nm thickness, each separated by -10 nm. The lowest threshold VCSELs are single quantum wells of InGaAs grown on GaAs, sacrificing power for threshold.

Mirror Reflectivity

When the mirror reflectivity R in a laser is very high, such thattmp8-599_thumb[2][2][2][2][2][2]a simple expression for the threshold gain-length product GLL is

tmp8-601_thumb[2][2][2][2][2][2]

Typical GaAs lasers have gains GL – 1000 cm-1. For a quantum well thickness of 10 nm, the gain per quantum well is 10-3 and reflectivities of -98 percent for each mirror should be sufficient to achieve threshold for 3 QW. Very often, however, in order to lower the threshold much higher reflectivities are used, particularly on the back mirror.

The on-resonance Bragg mirror reflectivity is the square of the reflection coefficient r, given by:

tmp8-602_thumb[2][2][2][2][2][2]

where there are N pairs of quarter-wavelength layers that alternate high-index and low-index (nh and nl, respectively), and nf and ni are the refractive index of the final and initial media, respectively.36

For high-reflectance Bragg mirrors, the second term in the numerator and denominator is small, and the reflectivity can be simplified to:

tmp8-603_thumb[2][2][2][2][2][2]

Higher reflectivity (smaller e) is provided by either more layer pairs or a larger refractive index difference between the two compositions in the layer pairs. Also, Eq. (74) shows that internal mirrors (nf = n) will have a smaller reflectivity than external mirrors (nf = 1) for the same number of layer pairs. If the layer pair is GaAs (n – 3.6) and AlAs (n – 3.0), a mirror consisting of 15 layer pairs will have an internal reflectivity R = 98 percent and external reflectivity R = 99.5 percent. Thirty layer pairs will increase the internal mirror reflectivity to 99.96 percent. Bragg mirrors with a smaller fraction of AlAs in the low-index layers will require more layer pairs to achieve the same reflectivity.

Some advanced technologies reduce the number of required layer pairs by selectively oxidizing the AlAs layers to lower their refractive index to n – 1.5. Using such techniques, reflectivities as high as 99.95 and 99.97 percent can be achieved from mirrors grown with only 7 interior pairs and 5 outside pairs, respectively; these mirrors can be used in VSCELs, but do not easily conduct current.

Electrical Injection

There is difficulty in injecting carriers from the top electrode down through the Bragg reflector, even if it is n-doped, because the GaAs layers provide potential wells that trap carriers. Furthermore, n-doping increases the optical loss in the mirrors. Possible solutions include reducing the AlAs concentration to < 60 percent; using graded compositions rather than abrupt layer pairs; using lateral carrier injection (which increases the operating voltage); using a separately deposited dielectric mirror on top of a transparent electrode; or accepting the high resistivity of the Bragg mirror and operating the laser at relatively high voltage.

The major issue for VCSELs, then, is to inject carriers efficiently, without resistive loss and without carrier leakage. Because resistance in n-doped mirrors is less than in p-doped mirrors, typically the top mirror is doped n-type and carrier injection comes from a top electrode. Light is emitted through a window hole in this top electrode. Carrier injection into the active region often requires rather high voltages because it may be difficult to drive carriers across the Bragg mirrors. Transverse current injection typically requires even higher voltages, although this method has been proven useful when highly conductive layers are grown just above and below the active region.

Some VCSELs use GRINSCH structures (similar to the composition used in edge emitters) to reduce the resistivity in the active region. Typical thresholds for VCSELs are about 3.5 V. Because the drive is limited by resistance, thresholds are typically given as voltages, rather than currents.

Planar VCSELs of fairly large diameter (>10 |im) are straightforward to make, and are useful when a low threshold is not required and multispatial mode is acceptable. Ion implantation outside the VCSEL controls the current in this region; the light experiences gain guiding and perhaps thermal lensing. Smaller diameters (3 to 10 |im) require etching mesas vertically through the Bragg mirror in order to contain the laser light that tends to diffract away.

Higher injection efficiency is obtained by defining the active region through an oxide window just above the active layer. This uses a selective lateral oxidation process that can double the maximum conversion efficiency to almost 60 percent. A high-aluminum fraction AlGaAs layer (~98 percent) is grown. A mesa is etched to below that layer. Then a long, soaking, wet-oxidization process selectively creates a ring of native oxide that stops vertical carrier transport. The chemical reaction moves in from the side of an etched pillar and is stopped when the desired diameter is achieved. Such a current aperture confines current only where needed. Threshold voltages of <6 V are common in diameters ~12 |im. This geometry is shown in Fig. 25. This oxide-defined current channel increases the efficiency, but tends to cause multiple transverse modes due to relatively strong oxide-induced index guiding. Single-mode requirements force the diameter to be very small (below 4 to 5 |im).

Spatial Characteristics of Emitted Light

Single transverse mode remains a challenge for VCSELs, particularly at the larger diameters. When VCSELs are modulated, lateral spatial instabilities tend to set in, and spatial hole burning causes transverse modes to jump. This can introduce considerable modal noise in coupling VCSEL light into fibers. Techniques for mode selection include incorporating a spatial filter, using an antiguide structure where the losses are much higher for higher order modes, or using sidewall scattering losses that are higher for higher-order modes. The requirement is that the mode selective losses must be large enough to overcome the effects of spatial hole burning.

Cross-sectional view of an oxidized GaAs VCSEL. An AlGaAs layer with high aluminum content grown just above the active region is chemically oxidized into AlxOy by a process that moves in from the edge of the etched mesa with time. Controlling the oxidization rate and time results in a suitable current aperture to obtain high conversion efficiency.

FIGURE 25 Cross-sectional view of an oxidized GaAs VCSEL. An AlGaAs layer with high aluminum content grown just above the active region is chemically oxidized into AlxOy by a process that moves in from the edge of the etched mesa with time. Controlling the oxidization rate and time results in a suitable current aperture to obtain high conversion efficiency.

One approach to achieving single transverse mode output is to include a passive antiguide region (PAR), the geometry shown in Fig. 24.35 The surrounding region has been etched and the sides backfilled with material of higher refractive index. This provides an antiguide for the laser, which has low loss only for the lowest order transverse mode. A single mode with a FWHM mode size of 7.4 |im (which matches single-mode fibers) can be achieved at 2.4 times threshold with VCSEL diameters of 15 |im. Current blocking outside the active area can be achieved by regrowing an nipi-doped antiguide. Typical thresholds for such lasers are 2 V (at 3 mA). A single-mode output of 1.7 mW with an input of 6.6 mA was reported, with more than 20 dB higher-order spatial mode suppression. Fixed polarization along one of the crystal orientations was observed during single-mode operation and attributed to asymmetry introduced in the etching and regrowth process. These structures have slightly higher thresholds than other geometries, but offer single-mode operation.

Other low-cost means of confining current are either proton implantation or etching mesas and then planarizing with polyimide. In both these cases, the regions surrounding the mesa will have a lower refractive index, which will cause the VCSEL to be a real index guide, which will tend toward multimode operation. This may introduce modal noise into fiber communication systems.

When the QWs are composed of InGaAs, the VCSELs will emit at 980 nm, and they can be designed to be bottom emitting, since the substrate is transparent. However, inexpensive silicon detectors can no longer be used at this wavelength, so these VCSELs offer fewer advantages in optical communication systems.

Light Out versus Current In

The VCSEL will, in general, have similar L-I performance to edge-emitting laser diodes, with some small differences. Because the acceptance angle for the mode is higher than in edge-emitting diodes, there will be more spontaneous emission, which will show up as a more graceful turn-on of light out versus voltage in. As previously mentioned, the operating voltage is 2 to 3 times that of edge-emitting lasers. Thus, Eq. (8) must be modified to take into account the operating voltage drop across the resistance R of the device. The operating power efficiency is:

tmp8-605_thumb[2][2][2][2][2][2]

Single-mode VCSELs of small diameter would typically have a 5 |im radius, a carrier injection efficiency of 80 to 90 percent, an internal optical absorption loss aiL of 0.003, an optical scattering loss of 0.001, and a net transmission through the front mirror of 0.005 to 0.0095. Carrier losses reducing the quantum efficiency are typically due to spontaneous emission in the wells, spontaneous emission in the barriers, Auger recombination, and carrier leakage.

Typical VCSELs designed for a compatibility with single-mode fiber incorporate an 8-|m proton implantation window and 10-|im-diameter window in the top contact. Such diodes may have threshold voltages of ~3 V and threshold currents of a few milliamps. These lasers may emit up to ~2 mW maximum output power. Devices will operate in zero-order transverse spatial mode with gaussian near-field profile when operated with DC drive current less than about twice the threshold. Output optical powers in single mode as high as 4.4 mW have been reported.

When there is emission in more than one spatial mode, or with both polarizations, there will usually be kinks in the L-I curve, as with multimode edge-emitting lasers.

Spectral Characteristics

Since the laser cavity is short, the longitudinal modes are much farther apart in wavelength, typically 8X ~ 50 nm, so only one longitudinal mode will appear, and there is longitudinal mode purity. The problem is with spatial modes, since at higher power levels the laser does not operate in a single spatial mode. Each spatial mode will have slightly different wavelengths, perhaps 0.01 to 0.02 nm apart. There is nothing in a typical VCSEL that selects a given polarization state. Thus, the VCSEL tends to oscillate in both polarization states, also with slightly different wavelengths.

When modulated, lateral spatial instabilities may set in, and spatial hole burning may cause transverse modes to jump. This can cause spectral broadening. In addition, external reflections can cause instabilities and increased relative intensity noise, just as in edge-emitting lasers.38 For very short cavities, such as between the VCSEL and a butt-coupled fiber (with ~4 percent reflectivity), instabilities do not set in, but the output power can be affected by the additional mirror, which forms a Fabry-Perot cavity with the output mirror and can reduce or increase its effective reflectivity, depending on the round-trip phase difference. When the external reflection comes from ~1 cm away, bifurcations and chaos can be introduced with a feedback parametertmp8-606_thumb[2][2][2][2][2][2]as defined in the discussion surrounding Eq. (45). Fortmp8-607_thumb[2][2][2][2][2][2]the feedback parametertmp8-608_thumb[2][2][2][2][2][2] and instabilities can be observed if one is not careful about back-reflections.

Polarization

Most VCSELs exhibit linear but random polarization states, which may wander with time (and temperature) and may have slightly different emission wavelengths. These unstable polarization characteristics are due to the in-plane crystalline symmetry of the quantum wells grown on (100) oriented substrates. Polarization-preserving VCSELs require breaking the symmetry by introducing anisotropy in the optical gain or loss. Some polarization selection may arise from an elliptical current aperture. The strongest polarization selectivity has come from growth on (311) GaAs substrates, which causes anisotropic gain.

VCSELs at Other Wavelengths

Long-wavelength VCSELs at 1.3 and 1.55 |im have been limited by their poor high-temperature characteristics and by the low reflectivity of InP/InGaAsP Bragg mirrors due to

tmp8-609_thumb[2][2][2][2][2][2]tmp8-610_thumb[2][2][2][2][2][2]tmp8-611_thumb[2][2][2][2][2][2] low index contrast between lattice-matched layers grown on InP. These problems have been overcome by using the same InGaAsP/InP active layers as in edge-emitting lasers, but providing mirrors another way: dielectric mirrors, wafer fusion, or metamorphic Bragg reflectors. Dielectric mirrors have limited thermal dissipation and require lateral injection, although carrier injection through a tunnel junction has shown promise. More success has been achieved by wafer-fusing GaAs/AlGaAs Bragg mirrors (grown lattice-matched onto GaAs) to the InP lasers. Wafer fusion occurs when pressing the two wafers together (after removing oxide off their surfaces) at 15 atm and heating to 630°C under hydrogen for 20 min. Typically one side will have an integrally grown InP/InGaAsP lattice-matched DBR (GaAlAsSb/AlAsSb mirrors also work). Mirrors can be wafer-fused on both sides of the VCSEL by etching away the InP substrate and one of the GaAs substrates. An integrated fabrication technology involves growing metamorphic GaAs/AlGaAs Bragg reflectors directly onto the InP structure. These high-reflectivity mirrors, grown by molecular beam epitaxy, have a large lattice mismatch and a high dislocation density. Nonetheless, because current injection is based on majority carriers, these mirrors can still be conductive, with high enough reflectivity to enable promising long-wavelength VCSELs.39

Next post:

Previous post: