Sources, Modulators, and Detectors For Fiber-Optic Communication Systems Part 9

ELECTROABSORPTION MODULATORS FOR FIBER-OPTIC SYSTEMS

When modulators are composed of III-V semiconductors, they can be integrated directly on the same chip as the laser, or placed external to the laser chip. External modulators may be butt-coupled to the laser, coupled by means of a microlens, or coupled by means of a fiber pigtail.

Electroabsorption. Semiconductor modulators typically use electroabsorption, the electric field dependence of the absorption near the band edge of a semiconductor. Electroabsorption is particularly strong in quantum wells (QWs), where it is called the quantum-confined Stark effect (QCSE). An example of the frequency dependence of the QCSE is shown in Fig. 28. The absorption spectrum of QWs exhibits a peak at the exciton resonance. When a field is applied, the exciton resonance moves to longer wavelengths, becomes weaker, and broadens. This means that the absorption increases with field on the long-wavelength side, as the exci-ton resonance moves to longer wavelengths. At wavelengths closer to the exciton resonance, the absorption will first increase with field, then plateau, and finally decrease, as the field continues to grow. At wavelengths shorter than the zero-field exciton resonance, the absorption will decrease with increasing field, as the resonance moves to longer wavelengths.

While electroabsorption in QWs is much larger than in bulk, due to the sharpness of the excitonic-enhanced absorption edge, the useful absorption change must be multiplied by the filling factor of the QW in the waveguide, which reduces its effective magnitude. Under some conditions, electroabsorption near the band edge in bulk semiconductors (typically called the Franz-Keldysh effect) may also be useful in electroabsorption modulators.

FIGURE 28 Spectrum of quantum-confined Stark effect (QCSE) in InAsP/InP strained MQWs. The absorption changes with applied field.

Waveguide Modulators. When light traverses a length of QW material, the transmission will be a function of applied voltage. An electroabsorption modulator consists of a length of waveguide containing QWs. The waveguide is necessary to confine the light to the QW region so that it does not diffract away. Thus, low-refractive-index layers must surround the layer containing the QWs. Discrete electroabsorption modulators are typically made by using geometries very similar to those of edge-emitting lasers (Fig. 1). They are cleaved, antireflec-tion coated and then butt-coupled to the laser chip. They are operated by a reverse bias, rather than the forward bias of a laser. Alternatively, the modulator is integrated on the laser chip, with the electroabsorption modulator region following a DFB or DBR laser in the optical train, as shown in Fig. 29. This figure shows the simplest electroabsorption modulator, with the same MQW composition as the DFB laser. This ridge waveguide device has been demonstrated with a 3-dB bandwidth of 30 GHz. The on-off contrast ratio is 12.5 dB for a 3-V drive voltage in a 90-^m-long modulator.50 The use of the same QW is possible by setting the grating that determines the laser wavelength to well below the exciton resonance. Because of the inherently wide gain spectrum exhibited by strained layer MQWs, this detuning is possible for the laser and allows it to operate in the optimal wavelength region for the electroabsorption modulator.

Other integrated electroabsorption modulators use a QW composition in the electro-absorption region that is different from that of the laser medium. Techniques for integration are discussed later.

FIGURE 29 (a) Geometry for a channel electroabsorption modulator (foreground) integrated on the same chip with a DFB laser (background, under the Bragg mirror). (b) Side view, showing how the same MQW active layer can be used under forward bias with a grating to provide a DFB laser, and in a separate region under reverse bias for modulation, with the two regions electrically separated by proton implantation.

Intensity Modulation by Electroabsorption

In an electroabsorption waveguide modulator of length L, where the absorption is a function of applied field E, the transmission is a function of field:where a is the absorption per unit length, averaging the QW absorption over the entire waveguide. (That is, a is the QW absorption multiplied by the filling factor of the QW in the waveguide.) Performance is usually characterized by two quantities: insertion loss (throughput at high transmission) and contrast ratio (ratio of high transmission to low transmission). Assume that the loss in the QW, initially at low value a_, increases by 8a. The contrast ratio is given by:

The insertion loss is given by

A long path length L means a high contrast ratio but also a large insertion loss and large capacitance, which results in a slower speed. Choosing the most practical length for any given application requires trading off the contrast ratio against insertion loss and speed.

To keep a moderate insertion loss, waveguide lengths should be chosen so that This sets the contrast ratio as

The contrast ratio depends on the ratio of the change in absorption to the absorption in the low-loss state; this fact is used to design the QW composition and dimensions relative to the wavelength of operation. In general, the contrast ratio improves farther from the band edge, but the maximum absorption is smaller there, so the modulator must be longer, which increases its capacitance, decreases its speed, and increases its loss. Contrast ratios may reach 10/1 or more with <2 V applied for optimized electroabsorption modulators. The contrast ratio does not depend on the filling factor of the QW in the waveguide, but the required length L does. Since high-speed modulators require small capacitance and small length, the filling factor should be as high as possible.

Waveguide modulators are used at wavelengths where the absorption is not too large, well below the band edge. In this wavelength region, electroabsorption at a fixed wavelength can be modeled by a pure quadratic dependence on field. Thus:

where a2 will typically depend on the wavelength, the QW and barrier dimensions and composition, and the waveguide filling factor. Intimately connected with this change in absorption is a change in refractive index with a similar field dependence:

where n2 is also strongly dependent on wavelength. Both electroabsorption and electro-refraction are about an order of magnitude larger in QWs than in bulk material. Specific numerical values depend on the detailed design, but typical values are on the order of . for 2 V applied across an i region 2.5 |im thick, for a field of ~10 kV/cm. This meansAlso,

Applying a Field in a Semiconductor

The electric field is usually applied by reverse biasing a pin junction. The electric field is supported by the semiconductor depletion region that exists within a pin junction, or at a metal-semiconductor junction (Schottky barrier). Charge carrier depletion in the n and p regions may play a role in determining the electric field across thin intrinsic regions. Taking this into account while assuming an undoped i region, the electric field across the i region of an ideally abrupt pin junction is given by51:

where Nd is the (donor) doping density in the n region, Na is the (acceptor) doping density in the p region, e is the elementary charge, £ is the dielectric constant, di is the thickness of the intrinsic region, and Vtot is the sum of the applied and built-in field (defined positive for reverse bias). When the n and p regions are highly doped and the i region is undoped, most of the voltage is dropped across the i region. When di is sufficiently large, the square root can be approximated, the di terms in the numerator cancel, and Eq. (88) becomes:

which, to lowest order, is just the field across a capacitor of thickness di. For a typical applied voltage ofHow much absorption and refractive index change this results in depends on wavelength and, of course, material design.

Integrating the Modulator

Stripe-geometry modulators can be cleaved from a wafer, antireflection coated, and butt-coupled to either a laser or a fiber pigtail. Typical insertion losses may be -10 dB. Or, the modulator may be monolithically integrated with the laser. A portion of the same epitaxial layer grown for the laser active region can be used as an electroabsorption modulator by providing a separate contact and applying a reverse bias. When such a modulator is placed inside the laser cavity, a multielement laser results that can have interesting switching properties, including wavelength tunability. When the electroabsorption modulator is placed outside the laser cavity, it is necessary to operate an electroabsorption modulator at wavelengths well below the band edge. Then, the modulator region must have a higher energy bandgap than the laser medium. Otherwise, the incident light will be absorbed, creating free electron-hole pairs that will move to screen the applied field and ruin the modulator.

The integration of an electroabsorption modulator, therefore, usually requires that the light traverse some portion of the sample that has a different bandgap from that of the laser region. Four techniques have been developed: etching and regrowth, vertical coupling between layers, selective area epitaxy, and postgrowth well and barrier intermixing.

Etching and Regrowth. Typically, a first set of epitaxial layers is grown everywhere, which includes the laser structure up through the QW layer. Then the QW layer is etched away from the regions where it is not needed. The structure is then overgrown everywhere with the same upper cladding layers. This typically results in a bulk electroabsorption modulator, consisting of laser cladding material. A more complex fabrication process might mask the laser region during the regrowth process and grow a different QW composition that would provide an integrated butt-coupled modulator for the DFB (or DBR) QW laser.

Vertical Coupling Between Layers. This approach makes it possible to use a QW modulator as well as a QW laser, with a different QW composition in each. Two sets of QWs can be grown one on top of the other and the structures can be designed so that light couples vertically from one layer to the other, using, for example, grating assisted coupling. This may involve photolithographically defining a grating followed by a regrowth of cladding layers, depending on the design.

Selective Area Epitaxy. Growth on a patterned substrate allows the width of the QWs to be varied across the wafer during a single growth. The substrate is usually coated with a SiO2 mask in which slots are opened. Under a precise set of growth conditions no growth takes place on top of the dielectric, but surface migration of the group III species (indium) can take place for some distance across the mask to the nearest opening. The growth rate in the opened area depends on the width of the opening and the patterning on the mask. Another approach is epitaxial growth on faceted mesas, making use of the different surface diffusion lengths of deposited atomic species on different crystal facets.

Well and Barrier Intermixing. The bandgap of a QW structure can be modified after growth by intermixing the well and barrier materials to form an alloy. This causes a rounding of the initially square QW bandgap profile and, in general, results in an increase of the bandgap energy. This provides a way to fabricate lasers and bandgap-shifted QCSE modulators using only one epitaxial step. Intermixing is greatly enhanced by the presence of impurities or defects in the vicinity of the QW interfaces. Then the bandgap is modified using impurity induced disordering, laser beam induced disordering, impurity-free vacancy diffusion, or ion implantation enhanced interdiffusion. The challenge is to ensure that the electrical quality of the pin junction remains after interdiffusion; sometimes regrowth of a top p layer helps.52

Operating Characteristics

In addition to contrast ratio, insertion loss, and required voltage, the performance of elec-troabsorption modulators depends on speed, chirp, polarization dependence, optical power-handling capabilities, and linearity. These factors all depend on the wavelength of operation, the materials, the presence of strain, the QW and waveguide geometry, and the device design. There will be extensive trade-offs that must be considered to achieve the best possible operation for a given application. Modulators will differ, depending on the laser and the proposed applications.

Chirp. Because a change in refractive index is simultaneous with any absorption change, electroabsorption modulators, in general, exhibit chirp (frequency broadening due to the time-varying refractive index, also observed in modulated lasers), which can seriously limit their usefulness. As with semiconductor lasers, the figure of merit is:

Unlike with lasers, however, there are particular wavelengths of sizable absorption change at which 8n = 0. Studies have shown that these nulls in index change can be positioned where 8a is large by using coupled quantum wells (CQWs).53 These structures provide two, three, or more wells so closely spaced that the electron wave functions overlap between them. If desired, several sets of these CQWs may be used in a single waveguide, if they are separated by large enough barriers that they do not interact. Chirp-free design is an important aspect of electroabsorption modulators.

On the other hand, since the chirp can be controlled in electroabsorption modulators, there are conditions under which it is advantageous to provide a negative chirp to cancel out the positive chirp introduced by fibers. This allows 1.55-^m laser pulses to travel down normal dispensive fiber (zero material dispersion at 1.3 |im wavelength) without the pulses unduly spreading.

Polarization Dependence. In general, the quantum-confined Stark effect is strongly polarization dependent, although there may be specific wavelengths at which TE and TM polarized light experience the same values of electroabsorption (and/or electrorefraction). It turns out that polarization-independent modulation is more readily achieved by using strained QWs. In addition, the contrast ratio of electroabsorption change at long wavelengths can be improved by using strained QWs.

Optical Power Dependence. During the process of electroabsorption, the modulators can absorb some of the incident light. This will create electron-hole pairs. If these electron-hole pairs remain in the QWs, at high optical powers they will introduce a free carrier plasma field that can screen the exciton resonance. This broadens the absorption spectrum and reduces the contrast ratio. In some cases, electroabsorption modulators operating at the band edge of bulk semiconductors (the Franz-Keldysh effect) may be able to operate with higher laser power. A common approach is to use shallow QWs, so that the electrons and holes may escape easily.

Even when the electron-hole pairs created by absorption escape the QWs, they will move across the junction to screen the applied fields. This will tend to reduce the applied field, and the performance will depend on the magnitude of absorbed light. Photogenerated carriers must also be removed, or they will slow down the modulator’s response time. Carriers may be removed by leakage currents in the electrodes or by recombination.

Built-in Bias. Because pin junctions have built-in fields, even at zero applied voltage, electro-absorption modulators have a prebias. Some applications use a small forward bias to achieve even larger modulation depths. However, the large forward current resulting from the forward bias limits the usefulness of this approach. There are, at present, some research approaches to remove the internal fields using an internal strain-induced piezoelectric effect to offset the pn junction intrinsic field.

Advanced Concepts for Electroabsorption in QWs

Coupled QWs offer the possibility of chirp control, and strained QWs offer the possibility of polarization independence, as previously explained. Adding these degrees of freedom to elec-troabsorption modulator design has been crucial in obtaining the highest performance devices.

High-performance Discrete Electroabsorption Modulators. A discrete modulator at 1.3 |im uses compressively strained InAsP wells grown on InP with InGaP barriers that are under tensile stress for strain compensation. High-speed operation with 3-dB bandwidth of >10 GHz and operating voltage of <2 V has been reported with a 20-dB on-off ratio.54 The electroabsorptive layer contained five QWs, each 11 nm thick. The waveguide was an etched high-mesa structure 3 |im wide and 200 |im long. A modulator had a 10-dB insertion loss, and the measured electroabsorptive figure of merit was 8a/a- = 10/1.

A discrete modulator at 1.55 |im planned for polarization insensitivity and capable of handling high optical powers was designed with two strongly coupled tensilely strained QWs.55 An 8-dB extinction ratio at 1.5-V drive voltage with a 3-dB bandwidth of 20 GHz was reported for average optical powers as high as 20 mW. Two InGaAs wells 5 nm thick with a 0.5-nm InGaAlAs barrier between them were grown in pairs, with 9-nm barriers separating each pair. A total of 13 pairs of wells were grown and etched to form ridge waveguides 3 |im wide.

High-Performance Integrated Electroabsorption Modulators. When an electroabsorption modulator is integrated with a DFB laser, strain is not required because polarization insensi-tivity is not needed. Selective epitaxy has been used to grow a 200-^m-long modulator region consisting of lattice-matched quaternary wells ~5 nm in width. The reported extinction ratio was >13 dB at 1.5 V, with output powers in the on state of >4 mW, at a current of 100 mA.56

A two-step growth procedure provided a butt joint between a modulator and a DFB laser. Compressively strained wells were used to reduce the potential well that the hole sees, speeding the device.57 Providing 3 V to a 200-^m-long modulator reduced the DFB laser output from 25 mW to 1 mW, for a 25/1 extinction ratio. The 3-dB bandwidth was 15 GHz. While there was a condition of zero chirp at -2 V, biasing to a regime of negative chirp allowed cancellation of the chromatic dispersion of fiber at the 1.55-^m laser wavelength. As a result, 10 Gb/s non-return-to-zero (NRZ) transmission was demonstrated over 60 km of standard fiber.

Wannier Stark Localization. A variation on the quantum-confined Stark effect uses an array of closely coupled quantum wells, which exhibit Wannier Stark localization (WSL). Because of the close spacing of the QWs, in the absence of a field, the electron wave function is free to travel across all wells, creating a miniband. When a field is applied, the wells decouple, and the electrons localize within individual wells. This removes the miniband, sharpening the absorption spectrum and creating a decrease in absorption below the band edge.58

Electron Transfer Modulators. A large absorption change can be created by filling the states near the edge of the semiconductor bands with electrons (or holes). This filling requires free carriers to be injected into the optical modulator. Quantum wells enhance the magnitude of this absorption change. Using applied voltage to transfer electrons from a reservoir across a barrier into a QW produces an effective long-wavelength modulator, termed a barrier, reservoir, and quantum well electron transfer (BRAQWET) modulator.59 By changing the bias across the device, the bound states of the QW are moved above and below the Fermi level fixed by the electron reservoir. These states are then emptied or filled by a transfer of electrons to or from the reservoir region. Optical modulation is achieved due to state filling and by carrier screening of the coulombic interaction between the electrons and holes in the QW. The combined effects reduce the absorption as the QW fills with electrons. Since electron transfer across the spacer is a very fast process, these modulators can have high modulation speeds, demonstrated at almost 6 GHz.

ELECTRO-OPTIC AND ELECTROREFRACTIVE SEMICONDUCTOR MODULATORS

Some semiconductor modulators are based on phase modulation that is converted to amplitude modulation by using a Mach-Zehnder interferometer, in the same manner as discussed in Sec. 4.10. Such modulators can be integrated on the same substrate as the laser, but do not have the chirp issues that electroabsorption modulators exhibit.

Electro-Optic Effect in Semiconductors

The III-V semiconductors are electro-optic. Although they are not initially anisotropic, they become so when an electric field is applied, and so they can be used as phase modulators. Referring to the discussion of the electro-optic effect in Sec. 4.10 for definitions, the GaAs electro-optic coefficients have only one nonzero term:Crystals are typically grown on the (001) face, with the Z axis normal to the surface. This means that the field is usually applied along Z. The only electro-optically induced index change will be Inserting this into the equation for the index ellipsoid, the electric field causes a rotation of the index ellipsoid around Z. Performing the diagonalization shows that the new values of the index ellipsoid are:These axes are at 45° to the crystal axes.

Performing the differential gives the refractive index changes at 45° to crystal axes:

The direction of these new optic axes (45° to the crystal axes) turns out to be in the direction that the zincblende material cleaves. Thus, TE-polarized light traveling down a waveguide normal to a cleave experiences the index change shown here. Light polarized along Z will not see any index change. Depending on whether light is polarized along X’ or Y’, the index will increase or decrease.60

With an electro-optic coefficient ofin a field of 10 kV/cm (2 V across 2 |im), and since no = 3.3, the index change for the TE polarization in GaAs will be 2.5 x 10-5. The index change in InP-based materials is comparable. The phase shift in a sample of length L is .At 1-|im wavelength, this will require a sample of length 1 cm to achieve a n phase shift, so that the voltage-length product for electro-optic GaAs (or other semiconductor) will be -20 Vmm. Practical devices require larger refractive index changes, which can be achieved by using quantum wells and choosing the exciton resonance at a shorter wavelength than that of the light to be modulated. These wells have an electrorefractive effect.