Sources, Modulators, and Detectors For Fiber-Optic Communication Systems Part 5

QUANTUM WELL AND STRAINED LASERS

Quantum Well Lasers

We have seen that the optimum design for low-threshold LDs uses the thinnest possible active region to confine free carriers, as long as the laser light is wave guided. When the active layer has a thickness less than a few tens of nanometers (hundreds of angstroms), it becomes a quantum well (QW).

FIGURE 13 Regimes of stable and unstable operation for two laser diodes (O and •) when subject to external feedback at varying distances and of varying amounts.

That is, the layer is so thin that the confined carriers have energies that are quantized in the growth direction z, as described in Vol. 1.This changes the density of states and the gain (and absorption) spectrum. While bulk semiconductors have an absorption spectrum near the band edge that increases with photon energy above the bandgap energyquantum wells have an absorption spectrum that is steplike in photon energy at each of the allowed quantum states. Riding on this steplike absorption is a series of exciton resonances at the absorption steps that occur because of the Coulomb interaction between free electrons and holes, which can be seen in the spectra of Fig. 14. These abrupt absorption features result in much higher gain for quantum well lasers than for bulk semiconductor lasers. The multiple spectra in Fig. 14 record the reduction in absorption as the QW states are filled with carriers. When the absorption goes to zero, transparency is reached. Figure 14 also shows that narrower wells push the bandgap to higher energies, a result of quantum confinement. The QW thickness is another design parameter in optimizing lasers for telecommunications.

Because a single quantum well (SQW) is so thin, its optical confinement factor is small. It is necessary either to use multiple QWs (separated by heterostructure barriers that contain the electronic wave functions within individual wells) or to use a guided wave structure that focuses the light into a SQW. The latter is usually a GRIN structure, as shown in Fig. 2d. Band diagrams as a function of distance in the growth direction for typical quantum well separate confinement heterostructures are shown in Fig. 15. The challenge is to properly confine carriers and light using materials that can be reliably grown and processed by common crystal growth methods.

FIGURE 14 Absorption spectrum for multiple quantum wells of three different well sizes, for varying levels of optically induced carrier density, showing the decrease in absorption toward transparency. Note the stronger excitonic resonances and increased bandgap with smaller well size.25

Quantum wells have provided significant improvement over bulk active regions, as originally observed in GaAs lasers. In InP lasers, Auger recombination and other losses come into play at the high carrier densities that occur in quantum confined structures, which tends to degrade laser performance. However, it has been found that providing strain in the active region can improve the performance of QW InGaAsP lasers to a level comparable with GaAs lasers. Strained QW lasers are described in the next section.

The LD characteristics described in Secs. 5.2 to 5.5 hold for QW lasers as well as for bulk lasers. The primary difference is that the gain is larger and the optical confinement factor will be much smaller, because the light is not well confined in a single thin QW active region. The optical confinement factor in a typical QW of thickness d is dominated by the second term in the denominator of Eq. (2). When multiple quantum wells (MQWs) are used, dg can be the thickness of the entire region containing the MQWs and their barriers, but r must now be multiplied by the filling factor r of the quantum wells \ within the MQW region—that is, if there are Nw wells, each of thicknessWhen a GRINSCH structure is used, the optical confinement factor depends on the curvature of its refractive gradient near the center of the guide.

FIGURE 15 Typical band diagrams (energy of conduction band Ec and valence band Ev versus growth direction) for quantum wells in separate confinement laser heterostructures: (a) single quantum well; (b) multiple quantum wells; and (c) graded index separate confinement heterostructure (GRINSCH) and multiple quantum wells.

There are subtle differences in performance between different geometries, depending on how many QWs are used and the extent to which a GRINSCH structure is dominant. The lowest threshold current densities have been reported for the highest Q cavities (longest lengths or highest reflectivities) using single QWs. However, for lower Q cavities the lowest threshold current densities are achieved with MQWs, even though they require higher carrier densities to achieve threshold. This is presumably because Auger recombination depends on the cube of the carrier density, so that SQW lasers will have excess losses with their higher carrier densities. In general, MQWs are a better choice in long-wavelength lasers, while SQWs have the advantage in GaAs lasers. However, with MQW lasers it is important to realize that the transport of carriers moving from one well to the next during high-speed modulation must be taken into account. In addition, improvements through the use of strained layer QWs make single QW devices more attractive.

Strained Layer Quantum Well Lasers

Active layers containing strained quantum wells have proven to be an extremely valuable advance in high-performance long-wavelength InP lasers. They have lower thresholds, enhanced differential quantum efficiency nc, larger characteristic temperature To, reduced linewidth enhancement factor Pc (less chirp), and enhanced high-speed characteristics (larger relaxation oscillation frequency QR) compared to unstrained QW and bulk devices. This results from the effect of strain on the energy-versus-momentum band diagram. Bulk semiconductors have two valence bands that are degenerate at the potential well minimum, as shown in Fig. 16. They are called heavy-hole and light-hole bands, since the smaller curvature means a heavier effective mass. Quantum wells lift this degeneracy, and interaction between the two bands near momentum k = 0 causes a local distortion in the formerly parabolic bands, also shown in Fig. 16. As a result, the heavy hole effective mass becomes smaller, more nearly approaching that of the conduction band. This allows population inversion to become more efficient, increasing the differential gain; this is one factor in the reduced threshold of QW lasers.26

FIGURE 16 The effect of strain on the band diagram (energy E versus in-plane momentum kx) of III-V semiconductors: (a) no strain, showing the degeneracy of the heavy holes HH and light holes LH at kx = 0; (b) quantum wells, showing the separately quantized conduction bands (C1 and C2) and removal of the valence band degeneracy, with the lowest energy heavy holes HH1 no longer having the same energy as the lowest energy light holes LH1 at k = 0; (c) compressive strain, with enhanced separation between the light-hole and the lowest heavy-hole band; and (d) tensile strain, with light holes having the lowest energy.

Strain additionally alters this structure in a way that can improve performance even more. Compressive strain in the QW moves the heavy-hole and light-hole valence bands further apart and further reduces the hole effective mass. Strain also decreases the heavy-hole effective mass by a factor of two or more, further increasing the differential gain and reducing the threshold carrier density. Higher differential gain also results in a smaller linewidth enhancement factor. Tensile strain moves the heavy-hole and light-hole valence bands closer together. In fact, at one particular tensile strain value these bands become degenerate at k = 0. Further tensile strain results in the light hole having the lowest energy at k = 0. These lasers will be polarized TM, because of the angular momentum properties of the light-hole band. This polarization has a larger optical matrix element, which can enhance the gain over some wavelength regions.

In addition to the heavy- and light-hole bands, there is an additional, higher-energy valence band (called the split-off band) which participates in Auger recombination and intervalence band absorption, both of which reduce quantum efficiency. In unstrained material there is a near-resonance between the bandgap energy and the difference in energy between the heavy-hole and split-off valence bands, which enhances these mechanisms for nonradiative recombination. Strain removes this near-degeneracy and reduces those losses that are caused by Auger recombination and intervalence band absorption. This means that incorporating strain is essential in long-wavelength laser diodes intended to be operated at high carrier densities. The reliability of strained layer QW lasers is excellent, when properly designed. However, strain does increase the intraband relaxation time, making the gain compression factor worse, so strained lasers tend to be more difficult to modulate at high speed.

Specific performance parameters are strongly dependent on the specific material, amount of strain, size and number of QWs, and device geometry, as well as the quality of crystal growth. Calculations show that compressive strain provides the lowest transparency current density, but tensile strain provides the largest gain (at sufficiently high carrier densities), as shown in Fig. 17. The lowest threshold lasers, then, will typically be compressively strained. Nonetheless, calculations show that, far enough above the band edge, the differential gain is 4 times higher in tensile compared to compressive strain. This results in a smaller linewidth enhancement factor, even if the refractive index changes per carrier density are larger. It has also been found that tensile strain in the active region reduces the Auger recombination, decreasing the losses introduced at higher temperatures. This means that To can increase with strain, particularly tensile strain. Performance at 1.55 |im comparable with that of GaAs lasers has been demonstrated using strained layer QWs. Deciding between compressively and tensilely strained QWs will be a matter of desired performance for specific applications.

Threshold current densities under 200 A/cm2 have been reported at 1.55 |im; To values on the order of 140 K have been reported, 3 times better than bulk lasers. Strained QW lasers have improved modulation properties compared with bulk DH lasers. Because the gain coefficient can be almost double, the relaxation oscillation frequency is expected to be almost 50 percent higher, enhancing the modulation bandwidth and decreasing the relative intensity noise for the same output power. Even the frequency chirp under modulation will be less, because the linewidth enhancement factor is less. The typical laser geometry, operating characteristics, transient response, noise, frequency chirping, and the effects of external optical feedback are all similar in the strained QW lasers to what has been described previously for bulk lasers. Only the experimentally derived numerical parameters will be somewhat different; strained long-wavelength semiconductor lasers have performance parameters comparable to those of GaAs lasers. One difference is that the polarization of the light emitted from strained lasers may differ from that emitted from bulk lasers.

FIGURE 17 Modal gain at 1.55 |m in InGaAs QW lasers calculated as a function of the carrier density per unit area contained in the quantum well. Well widths were determined by specifying wavelength.

As explained in Sec. 3.3, the gain in bulk semiconductors is independent of polarization, but lasers tend to be polarized in-plane because of higher facet reflectivity for that polarization. The use of quantum wells causes the gain for the TE polarization to be slightly (-10 percent) higher than for the TM polarization, so lattice-matched QW lasers operate with in-plane polarization. Compressive strain causes the TE polarization to have significantly more gain than the TM polarization (typically 50 to 100 percent more), so these lasers are also polarized in-plane. However, tensile strain severely depresses the TE gain, and these lasers have the potential to operate in TM polarization.

Typical 1.3- and 1.5-|m InP lasers today use from 5 to 15 wells that are grown with internal strain. By providing strain-compensating compressive barriers, there is no net buildup of strain. Typical threshold current densities today are -1000 A/cm2, threshold currents -10 mA, To – 50 to 70 K, maximum powers -40 mW, differential efficiencies -0.3 W/A, and maximum operating temperatures -70°C before the maximum power drops by 50 percent. There are trade-offs on all these parameters; some can be made better at the expense of some of the others.

DISTRIBUTED FEEDBACK (DFB) AND DISTRIBUTED BRAGG REFLECTOR (DBR) LASERS

Rather than cleaved facets for feedback, some lasers use distributed reflection from corrugated waveguide surfaces. Each groove provides some slight reflectivity, which adds up coherently along the waveguide at the wavelength given by the corrugation. This has two advantages. First, it defines the wavelength (by choice of grating spacing) and can be used to fabricate single-mode lasers. Second, it is an in-plane technology (no cleaves) and is therefore compatible with monolithic integration with modulators and/or other devices.

Distributed Bragg Reflector (DBR) Lasers

The distributed Bragg reflector (DBR) laser replaces one or both laser facet reflectors with a waveguide diffraction grating located outside the active region, as shown in Fig. 18. The reflectivity of a Bragg mirror is the square of the reflection coefficient (given here for the assumption of lossless mirrors)29:

where k is the coupling coefficient due to the corrugation (which is real for corrugations that modify the effective refractive index in the waveguide, but would be imaginary for periodic modulations in the gain and could, indeed, be complex).

FIGURE 18 Schematic for DBR laser configuration in a geometry that includes a phase portion for phase tuning and a tunable DBR grating. Fixed-wavelength DBR lasers do not require this tuning region. Designed for 1.55-|im output, light is waveguided in the transparent layer below the MQW that has a bandgap at a wavelength of 1.3 |im. The guided wave reflects from the rear grating, sees gain in the MQW active region, and is partially emitted and partially reflected from the cleaved front facet. Fully planar integration is possible if the front cleave is replaced by another DBR grating.

Also, 8 is a detuning parameter that measures the offset of the optical wavelength X from that defined by the grating periodicity A. When the grating is used in the mth order,

where ng is the effective group refractive index of the waveguide mode, and m is any integer. Also, S is given by:

The Bragg mirror has its maximum reflectivity on resonance when 8 ^ 0 and the wavelength Xm is determined by the mth order of the grating spacing A:

The reflection coefficient on resonance is

and the Bragg reflectivity is:

where K is the coupling per unit length,and is larger for deeper corrugations or when the refractive index difference between the waveguide and the cladding is larger. The reflectivity falls off as the wavelength moves away from resonance and the detuning increases. When off resonance far enough thatit is more practical to define:

and the reflectivity has the form:

Note that when. For moderate values of the grating couplingKL, this value of the reflectivity is not very different from that given by Eq. (53). Thus,over most of the detuning range.

The half-width of the resonance can be found by noting that the reflectivity goes to zero whenwhere the cotangent goes to infinity. This occurs at a cutoff detuninggiven byThis fact allows us to define a reflection resonance half-width as 8c/2 and the full width as 8o. The width of the resonance is constantbut broadens for large KL. Typical numbers are, so it is reasonable to take

The detuning is related to the wavelength bandwidth of the mirror by differentiating Eq. (50):Then the wavelength bandwidth fori and the width of the resonance is 0.5 nm (when). This narrow resonance, fixable by choosing the grating spacing and variable by varying the refractive index (with, for example, carrier injection) makes the DBR laser very favorable for use in optical communication systems.

The characteristics of Fabry-Perot lasers previously described still hold for DBR lasers, except that the narrow resonance can ensure that these lasers are single mode, even at high excitation levels.