Nano-engineering approaches to self-assembled InAs quantum dot laser medium

Journal of ELECTRONIC MATERIALS, Vol. 35, No. 5, 2006

Special Issue Paper

Nano-Engineering Approaches to Self-Assembled InAs Quantum Dot Laser Medium S. OKTYABRSKY,1,2 V. TOKRANOV,1 G. AGNELLO,1 J. VAN EISDEN,1 and M. YAKIMOV1 1.—College of Nanoscale Science and Engineering, University at Albany–SUNY, Albany, NY 12203. 2.—E-mail: [email protected]

A number of nano-engineering methods are proposed and tested to improve optical properties of a laser gain medium using the self-assembled InAs quantum dot (QD) ensemble. The laser characteristics of concern include higher gain, larger modulation bandwidth, higher efficiency at elevated temperatures, higher thermal stability, and enhanced reliability. The focus of this paper is on the management of QD properties through design and molecular beam epitaxial growth and modification of QD heterostructures. This includes digital alloys as high-quality wide-bandgap barrier; under- and overlayers with various compositions to control the dynamics of QD formation and evolution on the surface; shape engineering of QDs to improve electron-hole overlap and reduce inhomogeneous broadening; band engineering of QD heterostructures to enhance the carrier localization by reduction of thermal escape from dots; as well as tunnel injection from quantum wells (QWs) to accelerate carrier transfer to the lasing state. Beneficial properties of the developed QD media are demonstrated at room temperature in laser diodes with unsurpassed thermal stability with a characteristic temperature of 380 K, high waveguide modal gain .50 cm#1, unsurpassed defect tolerance over two orders of magnitude higher than that of QWs typically used in lasers, and efficient emission from a two-dimensional (2-D) photonic crystal nanocavity. Key words: InAs, quantum dots (QDs), molecular beam epitaxy (MBE)

INTRODUCTION In recent years, numerous technologies to manipulate materials at nanometer scale have been developed. One of the most productive and, in fact, the most mature areas of nanoscale science and technology involves quantum-confined semiconductor heteroepitaxial structures, which have given rise to numerous applications. Over two decades of research in this area have resulted in significant advances in various electronic devices including microwave and power transistors, laser diodes, and photodetectors, just to name a few. An ultimate three-dimensional (3-D) quantum confinement is achieved in quantum dots (QDs), where the corresponding energy states are discrete, giving rise to fundamentally different electronic properties that make them desirable for these types of applications. The ensemble of semi(Received October 12, 2005; accepted November 11, 2006)

822

conductor epitaxial QDs (such as InAs QDs in AlGaAs matrix) is even more attractive than quantum wells (QWs) for electronic and, in particular, for photonic devices. A good example is the projected performance of QD-based laser diodes. According to theoretical predictions, heterojunction lasers with QD active media will have superior characteristics as compared to the conventional QW lasers.1,2 The major reason for this superiority is the discrete atom-like electronic spectrum of QDs without thermal spreading of carriers. These characteristics include larger modulation bandwidth, higher efficiency at elevated temperatures, higher thermal stability, and enhanced reliability. However, these exciting benefits of the QD active media for lasers have not evolved into devices so far due to inhomogeneous broadening of the QD electronic spectrum resulting from the size dispersion of QDs, relatively slow relaxation of carriers to the ground level in the QDs, relatively low gain of the QD medium,

Nano-Engineering Approaches to Self-Assembled InAs Quantum Dot Laser Medium

existence of excited states in the dots, and evaporation of carriers from the dots to the wetting layer (WL) and barrier. These problems can potentially be solved using various nano-engineering approaches. Significant improvements in physics and technology of self-assembled QDs have recently allowed for demonstration of unparalleled performance characteristics in devices. A good example is the progress in laser diodes achieved in earlier work. An ensemble of shape-engineered InAs QDs embedded in high-quality wide bandgap QWs has been used in edge-emitting laser diodes that demonstrated unsurpassed thermal stability with a maximum lasing temperature of 219oC and an extremely high characteristic temperature of 380 K at roomtemperature operation.3 This type of QDs was also shown to have unsurpassed defect tolerance at room temperature, which exceeds by more than two orders of magnitude that of QW structures due to the suppressed escape of carriers from QDs.4 A tunnel QW-QD structure was employed by Bhattacharya et al.5 to improve modulation characteristics of QDs. The direct modulation at 23 GHz was demonstrated and intrinsic modulation bandwidth was as high as 43 GHz. The interconnect ‘‘bottleneck’’ in Si integrated circuits and a strong tendency of convergence of optical interconnects from long-haul, metro, and LANs to a board and chip level serve as a strong motivation for fast development of QD laser sources. In fact, high-frequency, directly modulated, thermally stable vertical cavity surface emitting lasers (VCSELs) and dense VCSEL arrays are expected to change the entire paradigm of short-range interconnections.6 In fact, the only way to provide the optical emitters, that can be directly integrated with Si electronics in a simple, inexpensive, and effective manner, is through development of a novel active medium to overcome the fundamental limitations of QWs. The QDs can potentially provide the required set of properties. In this study, we present the nano-engineering approaches and results on molecular beam epitaxy (MBE) growth of self-assembled InAs QDs. The focus of this paper is on the management of QD properties by variation of thermodynamic parameters, chemistry of under- and overlayer, QD overgrowth procedures, and band engineering of QD heterostructures. EXPERIMENTAL DETAILS Quantum dot heterostructures were grown on nGaAs(001) substrates using an EPI GEN II MBE system equipped with metal SUMO cell sources and As valved cracker, delivering an As2 molecular beam. Typical QD growth conditions include a growth temperature of 470°C, an InAs growth rate of 0.05 ML/s (ML stands for monolayer), a flux ratio As2/In 5 10, and a total thickness of InAs coverage of 2.4 ML. Details on the growth procedure and layout of heterostructures have been described

823

elsewhere.7 The QD growth was monitored using in-situ RHEED to determine the kinetics of QD formation.8 Auger measurements were carried out using PHI 10–155A spectrometer with 0.6% energy resolution, attached to the MBE system buffer chamber. For Auger measurements, the substrate was cooled after the growth under As2 flux at a rate of about 50°C/min., and transferred to the measurement position within 5 min., without breaking the UHV conditions. Plane-view and cross-sectional transmission electron microscopy (TEM) was performed on the samples using a 200 keV JEOL-2010FEG microscope (Japan Electron Optics Ltd., Tokyo). The (200) dark-field (DF) conditions were used primarily to image the structures. These imaging conditions provide a high chemical contrast in zinc-blende crystal structures because the (200) kinematic beam amplitude is proportional to the difference in atomic scattering amplitudes of group III and V atoms, therefore providing different contrast for, e.g., GaAs, InAs, and AlAs. On the other hand, the strain contrast is significantly suppressed. For these reasons, it is used extensively for TEM studies of QDs embedded in multilayered structures. The optical properties of the self-assembled QDs were characterized mainly at room temperature using an Ar1 laser (514 nm) with ;10 W/cm2 excitation intensity. The photoluminescence (PL) was dispersed by a 0.8-m monochromator and detected by a liquid-nitrogen-cooled Ge detector with lock-in amplifier. A gain of the QD medium was measured using gain-guided lasers with contact widths ranging from 50 mm to 200 mm, waveguide thickness of 0.8 mm, and cavity lengths between 0.1 mm and 8.0 mm. No coatings were applied to the cleaved facets for basic characterization. Stripe-up laser crystals were mounted on a heat sink. RESULTS AND DISCUSSION To approach ‘‘the perfect QD ensemble,’’ various methods and technologies can be employed. Though the most controllable methods for synthesis of semiconductor QDs involve various types of nanopatterning techniques, so far, they generate too many defects that introduce traps and recombination centers in the active medium of the device. This results in low-quality material as compared to the self-assembling method. Therefore, we are using the most common technology for QD formation, which is based on the phenomenon of self-organization in epitaxial heterostructures with high lattice mismatch. The most explored heterosystem is the In(Ga)As/(Al)GaAs structure with up to 7% lattice mismatch. The growth of InAs QDs on the GaAs surface is usually described rather simplistically in terms of a classic Stranski–Krastanov growth mechanism, where the structure attains the lowest total free energy through a trade-off between strain energy in the

824

Oktyabrsky, Tokranov, Agnello, Van Eisden, and Yakimov

growing film and the surface free energy of the islands. According to this mechanism, after formation of a thin coherent WL, growth spontaneously becomes 3-D through assemblage of the islands. In reality, the mechanism of self-assembly is further complicated by the involvement of kinetics in addition to thermodynamics. Wish List for QD Ensemble for Laser Gain Medium Table I specifies essential parameters for a QD ensemble to be used as a laser gain medium. Unfortunately, self-assembly limits the window for variation of the QD properties. In fact, usually, the improvement of one particular parameter (e.g., narrow size distribution) results in degradation of other properties (e.g., emission efficiency). Therefore, a technology using a self-assembly process is a set of trade-offs and optimizations. Self-assembly results in a quite wide distribution of QD sizes giving rise to a relatively wide inhomogeneously broadened emission and gain spectra. A single QD energy level is extremely narrow (;10 meV), but the ensemble of QDs exhibits a 25– 35 meV spectral width at best. Creating the QD medium with more uniform size (more specifically volume) distribution, and therefore, higher saturated gain and differential gain, is a significant challenge because the size distribution (or, more accurately, volume distribution) of the islands is determined by kinetics of fluctuations in phase transformation. If the island size fluctuations are purely statistical, and are of the order of N1/2, where N is the number of atoms in the island, the QD with ,N. 5 3000 atoms would have fluctuations of the order of 50 atoms, corresponding to an exciton recombination energy broadening for InAs QDs of about 3 meV.9 Practically, the full-width at half-maximum (FWHM) of the QD luminescence band is usually 30–50 meV and is far beyond the simple statistics of the number of condensed atoms. As we have mentioned already, the benefits of QDs are mainly related to strong localization of carriers. In order to keep localization at higher temperatures, the separation between ground and excited levels (Dexc) in QDs and the barrier height should be made much larger than the thermal energy. This

separation is determined by the barrier height and shape of the QDs. The barrier height also decreases with the size of the QDs due to increased quantum confinement energy. Therefore, small dots with high density will provide lower localization energy than large dots with low density. On the other hand, high QD density is essential to achieve higher gain in the laser medium. It is critical for miniature lasers such as VCSELs and short edge-emitting diodes. HighQD-density medium (.1011 cm#2) suffers from poorer localization and lower radiative efficiency, the latter likely due to higher defect density in the heterostructures with increased InAs/GaAs interface area. Other optimization trade-offs indicated in Table I will be discussed in the following sections. Management of optical properties of QDs relies on understanding of the basic technology-structure-property relationships controlling the QD ensemble. A set of issues essential for understanding these relationships are summarized in Table II, and will be also discussed in the following sections. Formation of QDs The growth of InAs QDs uses the large (;7%) lattice mismatch between InAs and GaAs to promote the formation of strained 3-D InAs islands through the Stranski–Krastanow growth mode. At first, InAs grows pseudomorphically on top of GaAs, forming a strained film (WL) with a thickness of ;1.5 ML.8 At this critical thickness, it becomes energetically favorable for InAs to form islands, or QDs, which allows for the effective release of strain. In order to increase the bandgap of the matrix for QDs, we have employed an AlAs/GaAs short period superlattice (SPSL) as a barrier material instead of AlGaAs alloy. The SPSL or ‘‘digital alloy’’ is a highquality wide bandgap material that can be grown at the low temperature (,500°C) required for QD growth. In addition, SPSL provides another variable to control QD growth since both GaAs and AlAs surfaces can be used as a starting surface for QD deposition. Figure 1 shows a cross-sectional TEM micrograph of a typical InAs QD embedded in a (2 ML-AlAs)/(8 ML-GaAs) SPSL. The image reveals all major components of the structure under discussion. Figure 2 shows the dependence of QD density on the growth temperature measured by plane-view

Table I. Desirable Parameters for a QD Ensemble for Laser Gain Medium and Associated Optimization Trade-Offs QD Parameter

Demonstrated Optimum

Trade-Offs

Narrow size distribution (small inhomogeneous broadening) Strong localization Large ground-excited states separation High density High efficiency (defects) High efficiency (e-h overlap) Fast carrier transport to GS

25–35 meV at hv 5 1.1 eV

High efficiency of radiative recombination

400–500 meV at RT ;100 meV ;(8–12 3 1010 cm#2 per layer Close to 100% at 100 A/cm2 Symmetrical shape ;5–10 ps at RT

High density Fast relaxation to GS Strong localization, high efficiency High density High density Thermal stability

Nano-Engineering Approaches to Self-Assembled InAs Quantum Dot Laser Medium

825

Table II. Issues of Management of QD Properties Problem Surface properties of In Evolution of QDs Overgrowth of QDs Management methods Optical properties

TEM. Our results are limited to a standard process window of 450–500°C (solid symbols). To make an assumption of the QD density behavior at low growth temperatures, we employed results obtained by other groups for growth on GaAs10 and AlAs.11 Though the growth conditions are different in these cases resulting in different absolute values of QD densities, the tendency is quite clear. The Arrhenius behavior with ;2.0 eV activation energy is evident up to QD densities as high as 1012 cm#2. The same activation energy has been recently obtained from analysis of results on InAs/GaAs reported in the literature.12 The density of dots is consistently higher by approximately a factor of 2, if formed on the AlAs surface, which is attributed to the lower surface diffusion rates of In adatoms on AlAs. The atomic force microscopy (AFM) imaging of the QDs allows for the assessment of QD size and heights (Fig. 3). As expected from the TEM studies, the dots grown on AlAs surface are smaller and have higher densities. When the QDs are grown at 475°C, the aspect ratio is about 6 for the dots on GaAs surface, and slightly higher on AlAs. In both cases, total thickness of deposited InAs was kept identical: 2.4 ML. Hence, there is no noticeable difference in the total volume of the QDs grown on AlAs or GaAs. Since AFM provides information on surface QDs, and further overgrowth as well as cooling of the surface dots may change their size and shape, the real QD heterostructures may contain QDs with different characteristics. In fact, we will show in the next

Fig. 1. (200) DF TEM image of InAs QDs embedded in a SPSL. Structural components of the image are labeled accordingly. The frame ‘‘A’’ indicates the portion of the image used for simulations.

Issues Thermodynamics of 2D-3D phase transition, adatom diffusion Tendency toward larger dots with time, evaporation and intermixing of In Segregation of In, lateral redistribution of In Band engineering, shape engineering, stress management (InGaAs QW) Electronic spectrum, defects, carrier dynamics

section, that the (volume 3 density) product can be controlled by QD overgrowth procedure. Using dynamic in-situ RHEED, we have monitored the formation of QDs after deposition of the WL.8 The time of QD formation, which is characterized by the increase of the RHEED spot intensity associated with 3-D InAs features on the surface (gradual growth of the dots), is plotted in Fig. 4. This time gradually decreases with temperature when the QD layer is grown on the AlAs surface and remains almost constant when it is grown on GaAs in the temperature range of interest (400– 500°C). Considering that the formation of QDs is determined by the diffusion properties of the In adatoms on the surface, we can estimate the diffusion coefficient of In on the surface as D 5 x2 =2t where x is an average distance between dots calculated from Fig. 2, and t is QD formation time. The results of these estimations are also presented in Fig. 3. The solid lines correspond to the Arrhenius behavior of the diffusion coefficients with activation energy 2.2 eV for AlAs and 2.0 eV for GaAs. Obviously, a small change in QD formation time in the

Fig. 2. Arrhenius plot of surface density of InAs QDs: solid symbols— grown on GaAs and 2 ML AlAs measured by plane-view TEM in the current study at growth temperatures between 450°C and 500°C; and open symbols—from Ref. 27 (circles) and Ref. 11 (squares) grown on GaAs and thick AlAs, respectively, and measured by AFM. Straight lines correspond to activation energy of 2.0 eV.

826

Oktyabrsky, Tokranov, Agnello, Van Eisden, and Yakimov

Fig. 3. AFM Images of QDs grown on AlAs and GaAs surfaces at 475°C. The QD ensembles have almost identical average lateral size: 31 nm on GaAs and 27 nm on AlAs, different aspect ratio (d/h): ;6 on GaAs and ;11 on AlAs, lower QD density on GaAs ;5.2 3 1010 cm#2, and ;9.8 3 1010 cm#2 on AlAs.

case of QD formation on the GaAs surface results in the same value of activation energy as for the QD density (Fig. 2). Recently, the activation energy of 1.6 eV for the In diffusion on the GaAs surface was determined from the analysis of a kinetic model applied to the results on QD density.12 Our results allow obtainment of the diffusion values directly from the dynamic measurements and are based on straightforward assumptions. The difference in QD formation behavior on GaAs and AlAs surfaces indicates that though the QDs are actually formed on top of the InAs WL, the process of QD formation still depends on the properties of the underlying layer. Manipulation of underlayer chemistry can be used for fine tuning the QD sizes, and is accompanied by corresponding changes in the optical properties of QD structures. Figure 5 shows the dependences of the normalized (to the intensity of an 8-nm-thick In0.2Ga0.8As QW) photoluminescence (PL) intensity and linewidth on the ground state

transition energy of QDs, Egr. All of the QD structures were grown at 475°C and were subjected to the shape-engineering procedure described below. The average volume of QDs, and, hence, the ground state transition energy, Egr, was controlled by the underlayer chemistry. The lowest Egr (the largest QD volume) corresponds to QDs grown on the GaAs surface, though the largest Egr (the smallest QD volume) was observed in the QDs grown on 2 ML of AlAs of (2MLAlAs)/(8ML-GaAs) SPSL. Noticeable degradation of the PL intensity and broader size distribution were measured for QDs grown on the AlAs surface. Intermediate properties were obtained for QDs grown on 1 ML of GaAs on SPSL terminated by AlAs. These QDs with AlAs capping exhibit reduced volume with Egr 5 1.11 eV, narrow size distribution (FWHM 5 34 meV), large ground-excited state separation energy Dexc 5 87 meV, and acceptable PL intensity. This QD ensemble might be considered as a first pass optimized design of a laser medium.

Fig. 4. Temperature dependence of QD formation time and estimated diffusion coefficients of In adatoms on GaAs and AlAs surfaces. The solid lines correspond to Arrhenius behavior of the diffusion coefficients with activation energy 2.2 eV for AlAs and 2.0 eV for GaAs.

Fig. 5. Normalized room-temperature PL intensity and linewidth (corresponding to the size distribution); FWHM/Egr as a function of ground state PL energy, Egr, of QDs on various underlayers.

Nano-Engineering Approaches to Self-Assembled InAs Quantum Dot Laser Medium

Evolution of QDs on a Surface and QD Overgrowth As mentioned above, the QD characteristics can change when the dots formed on a surface are overgrown. Through AFM measurements, we have been able to investigate the effect of both capping layer material (2 ML of GaAs or AlAs) and growth rate (0.06 or 0.11 ML/s) on the resulting structural property QDs (Table III). These results allow for the following major conclusions: (1) Evolution of QDs on the surface toward larger QDs with less density; (2) AlAs is more effective in preserving the density and shape of QDs; and (3) stronger redistribution of In from QDs occurs when capped by GaAs. (4) The first conclusion is quite obvious and comes from the observation that the uncapped dots that were left on the bare GaAs surface for the longest time exhibit the largest volume. It is well understood that if the QD growth process is allowed to progress, surface redistribution of In adatoms will continue and the formed islands will become larger and larger. This has been shown both experimentally13 and through theoretical modeling.14 This conclusion is very important for understanding the QD growth because it claims that there is no thermodynamic equilibrium size of the QDs on the surface, and the QDs will grow on the surface as long as they are not overgrown. Of course, some other requirements should be met for this process to continue, such as maintaining the As equilibrium overpressure, no escape (evaporation or solution) of In atoms from the surface, and no plastic strain relaxation. The initial size and density of QDs is determined by characteristics of critical nuclei at the 2-D–3-D phase transformation, and after this transformation takes place, within a few seconds, the QDs start growing slowly by dissolution of the smaller dots. It is, therefore, important to cap the QDs immediately after formation to preserve the high density QD ensemble. (5) To prevent QD evolution on the surface, a thin (2 ML) capping layer was deposited in an attempt to ‘‘freeze’’ the growth surface. This step is intended to halt QD evolution/decomposition and produce a high QD density. From this

827

standpoint, a faster capping layer growth rate (0.11 ML/s) should be more effective in the prevention of continued QD evolution as compared to a slower one (0.06 ML/s). It is observed that fast AlAs capping preserves the highest QD surface density, and keeps the dots from growing on the surface. Slow AlAs capping is also effective in ‘‘freezing’’ the QD density, but can be responsible for loss of material from the top of the QDs making them shorter (Table III). Slow (0.06 ML/s) GaAs capping results in the QD ensemble similar in size and density to the uncapped one and, therefore, is not effective for our purposes. Even fast GaAs cannot preserve high QD density, though the QD average size is noticeably smaller than in the uncapped case. From these observations, we can conclude that an AlAs capping layer is more effective in preserving the density and shape of QDs. (6) If we consider now that a QD ensemble capped with a fast growth rate AlAs layer is the one preserved at the earlier stages of evolution, the reduction in QD sizes with other capping procedures (except slow GaAs capping, which is similar to the uncapped ensemble) becomes obvious. This size reduction is likely due to a segregation or lateral redistribution of In from the dots to the surrounding matrix. In fact, this mechanism of QD size reduction works in a direction opposite that of the QD growth/evolution on the surface. To prove the existence of QD size reduction during capping by an independent method, we have grown two samples with QDs embedded in a SPSL double heterostructure. One sample incorporated a 2 ML AlAs capping layer grown at a fast rate of 0.11 ML/s, and another was capped by GaAs at the same high rate. Figure 6 shows the room-temperature PL spectra of these two structures. Ground state transition energy Egr is red shifted by 13 meV in the AlAs-capped QD sample. On the other hand, through our AFM measurements, we were able to obtain average QD lateral size as well as height values with which the average QD volume (atoms/dot) can be calculated under both AlAs and GaAs capping conditions. The calculated difference between the two was determined to be ;7,000 (In 1 As) atoms (45,000 atoms for GaAs capping and 52,000 atoms for AlAs capping). According to k"p calculations,9

Table III. Characteristics of QD Ensembles Derived from Statistical Analysis of AFM Images; All QD Layers Were Deposited on the GaAs Surface at 470°C Capping/Rate Uncapped AlAs/slow GaAs/slow AlAs/fast GaAs/fast

Average Height (nm) 6.7 3.3 5.4 4.2 3.6

6 6 6 6 6

0.8 0.5 0.7 0.8 0.5

Average Lateral Size (nm) 37 27 35 30 28

6 6 6 6 6

8 6 7 7 8

QD Areal Density (31010/cm2) 3.9 4.6 3.6 4.8 3.7

828

Oktyabrsky, Tokranov, Agnello, Van Eisden, and Yakimov

Fig. 6. PL of QD ensembles capped with fast growth rate GaAs and AlAs layers. The right-most peaks correspond to ground state emission Egr. The ground state emission Egr red shift of 13 meV in an AlAs-capped QD sample corresponds to an approximately 20% increase in QD volume.9

this difference corresponds to DEgr of ;10 meV, which is very close to the measured PL shift. We can, therefore, conclude that slow AlAs and fast GaAs capping results in material loss from QDs. The mechanism of this loss is discussed in the following section.

The observed increased intensity of GaAs can be attributed to double diffraction of the {111} beams as the specimen becomes thicker, which adds to the ‘‘observed’’ intensity of the (200) beam. The incorporation of Al atoms (alloy formation) into the WL also has a significant impact on TEM contrast, as evidenced by the simulation in Fig. 7c, with In occupancy in the WL reduced to 50%. The GaAs intensity increases again, bringing it closer to the experimental image. The nonuniform intensity of the GaAs QW was modeled considering vertical In distribution in the region above the wetting/capping layers (Fig. 7d). These occupancy variations result in image simulations that closely resemble experimental observations, where a dark band protrudes vertically from the capping layer, eventually ‘‘blending in’’ with the surrounding matrix. The fact is In segregation itself cannot account for the difference in QD capping by GaAs or AlAs layers. Therefore, to study the InAs surface segregation, we have measured in-situ Auger spectra on a 1-ML-thick wetting InAs layer overgrown by either GaAs or AlAs. Since the escape depth of Auger electrons is low, the In(405 eV) Auger signal is

Interestingly, the spectra of the AlAs and GaAs capped QDs show a difference between first excited and ground state PL bands [Dexc 5 83 meV and 72 meV, respectively (Fig. 6)]. Other than dot volume, the aspect ratio can also affect the electronic spectrum of the QDs. It has been reported previously through pseudo-potential calculations15 that QDs with a larger aspect ratio (higher dots) exhibit a larger spacing between ground and first excited states. This is qualitatively consistent with our results (Table III), though the difference in Dexc (11 meV) looks too large for direct comparison to theory. Indium Segregation and Redistribution A strong indication for In segregation can be found from the analysis of TEM images. Figure 7 shows a magnified (002) DF image of frame ‘‘A’’ from Fig. 1 containing electron scattering information from different components of the structure away from the QDs. To study the image contrast variations, we have constructed a supercell (Fig. 7f) consisting of two periods of (8 ML GaAs)/(2 ML AlAs) SPSL followed by the 1 ML InAs WL, 2 ML AlAs capping layer, and 20 ML GaAs QW. Figure 7b shows the result for a (200) DF image calculation using multislice image simulation software (MacTempas), where the intensity distribution agrees with the prediction based on structure factors (InAs and AlAs should have bright contrast and GaAs should be dark in (200) DF) but disagrees, entirely, with the experiment.

Fig. 7. (a) Magnified (002) DF image of the frame ‘‘A’’ in Fig. 1 including two periods of (2 ML-AlAs)/(8 ML-GaAs) SPSL, 1 ML InAs WL, 2 ML AlAs capping layer, and 20 ML GaAs QW. (b)–(d) Multislice image simulations of the image in (a) using supercell depicted in (f): (b) simulation in pure (002) imaging conditions; (c) simulation in a (002) imaging conditions allowing double diffraction of the {111} beams contributing to (002) DF image with In0.5Al0.8As alloyed WL but without vertical indium segregation; and (d) same simulation where indium segregation into the GaAs has been modeled. (f) Supercell used in the image simulation viewed along the [110] direction.

Nano-Engineering Approaches to Self-Assembled InAs Quantum Dot Laser Medium

mainly generated by In segregated on the surface.8 Reduction of the Auger In (MNN) signal during overgrowth is plotted in Fig. 8, and appears to be almost identical for both GaAs and AlAs overgrowth. These profiles can be fitted using the Muraki model16 (Fig. 8), describing the surface segregation phenomenon. The best fit of experimental data are obtained using an In segregation efficiency parameter of R 5 0.86, approximately the same for both GaAs and AlAs overgrowth at 480°C. This result is to be compared with the published values of R 5 0.8217 for GaAs and R 5 0.77 for AlAs17,18 at T 5 530°C. As a result, we obtain an InAs redistribution thickness of about 6 ML, which is qualitatively consistent with the results obtained from TEM studies. Since the In surface segregation is found to be almost identical, the difference in QD capping by GaAs or AlAs layers is likely to be due to lateral redistribution of In when the QDs are being overgrown. After the beginning of the capping process, the open surface InAs areas (primarily the tops of QDs) will start losing indium. This process is opposite to the QD expansion on the InAs surface prior to capping. The mobility of In atoms is especially high in the GaAs/InAs system, which directly leads to faster redistribution of In when the GaAs capping layer is growing. On the other hand, AlAs capping freezes the surface redistribution of In, thus preventing In loss from the QDs during overgrowth. Shape Engineering of QDs All-epitaxial InAs QDs assembled on the GaAs or AlAs surface always show pyramidal or close to pyramidal shapes, which are determined by surface energy of the growing InAs islands. These shapes are responsible for quite poor interaction between electrons and holes. More specifically, the electron and hole ground state wave functions have a small overlap integral due to the stress-induced piezoelectric field separating electrons and holes, resulting in

Fig. 8. Auger In (MNN) signal (405 eV) from the surface during the overgrowth. Solid line is a fit using the Muraki model16 with the In segregation efficiency parameter R 5 0.86.

829

small optical gain in the QD systems.19 The QDs with more symmetric shapes, such as disks or spheres, would be preferable to increase the electron-hole interaction. Though the saturated material gain in the QD medium is much higher than in the QW structures, the small total volume of QDs reduces their modal gain to quite low values. Figure 9 demonstrates the basic approach involved in the shape management of the QD ensemble. As described above, to prevent coarsening of these 3-D islands nucleated on the surface as well as redistribution of In on the growing surface, we have employed capping of the just formed QDs by a 2-ML-thick AlAs layer. The most critical step is the shape engineering of the QD ensemble.7 The QDs covered with 2 ML AlAs are overgrown with a GaAs layer with a thickness approximately corresponding to the QD heights (;6 nm). Due to the high mobility of Ga adatoms, the GaAs layer does not cover the QDs entirely but tends to cover the area with low stress between the QDs (Fig. 9a). Then, the sample temperature is rapidly increased by 100oC, resulting in indium redistribution mainly from the top of QDs, changing the shape and volume of QDs. Finally, the structure is overgrown by a thin layer of GaAs to create a QW for efficient collection of carriers and, further, by a SPSL (Fig. 9b). The described shape management procedure results in QDs with a shape of truncated pyramids with flat tops entirely covered by the AlAs layer, even though the partially overgrown dots were subjected to a heating step (Fig. 10a and b). The truncated QD ensemble has demonstrated better volume uniformity since larger QDs lose more In. This results in a measured inhomogeneous broadening of the emission spectrum of as low as 29 meV.7 The beneficial (more symmetrical) shape of the QDs has been shown to increase the electron-hole

Fig. 9. Schematics of QD structure before (a) and after (b) shapeengineering (truncation) procedure. The shape of QDs changes during the heating step.

830

coupling, which allows demonstration of higher emission efficiency and saturated ground state gain presented below. AlAs capping affects mostly the energy separation between the ground and first excited states; the optimized structures demonstrate a separation of 90–100 meV, considerably higher than the thermal energy. Apart from the increase of electron-hole coupling by shape engineering, the more ordinary way to increase the maximum gain is through an increase of total density of QDs by adding more layers. Noticeable advantages of shape-engineered QD ensembles for multilayer QDs are due to the reduced roughness of the entire structure with very smooth top SPSL interfaces (as contrasted with the waveshape top SPSL interfaces in nontruncated QDs in Fig. 10c and d). It also allows for reducing the spacing between the QD sheets in order to fit up to nine layers into a quarter-wavelength-thick region with high optical field in the VCSEL cavity. We have recently demonstrated a medium with seven QD layers separated by 20-nm-interlayer barrier distances (total thickness of ;125 nm)20 with maximum in-plane gain of 31 cm#1. Band Engineering and Defect Tolerance of QDs As noted above, the strong localization of carriers in QDs is responsible for the major advantages of the QD laser medium. Thermal escape of carriers from the QDs reduces gain in the laser medium. The same process is responsible for thermal quenching of QD luminescence due to recombination of carriers outside the dots. Obviously, strong localization requires a large barrier to escaping of the carriers, large energy separation between the ground state and excited states levels, and lack of any intermediate levels with high density of states such as those in a WL. In general, this set of requirements is good enough to ensure localization as long as the relaxation process of carriers to the QD ground state is fast enough.4 It should be emphasized that the wide

Fig. 10. TEM images ((200) DF) of triple-layer QDs: (a) shapeengineered QDs in SPSL with top AlAs and (b) magnified image; (c) nontruncated QDs in SPSL with top AlAs and (d) magnified image.

Oktyabrsky, Tokranov, Agnello, Van Eisden, and Yakimov

bandgap barrier by itself is not enough to guarantee strong localization in the dots. The band diagram of nano-engineered QDs in a well shown in Fig. 10b is sketched in Fig. 11. Within the capture/escape events, carriers are usually considered to behave as excitons or correlated electronhole pairs, rather than independent electrons and holes. The validity of this approach is usually justified by the fact that the activation energy for the escape process is typically equal to the sum of the barrier heights for electrons and holes. This also appears to be true in our experiments. It is important to note that Le Ru et al.21 have recently argued that the more realistic independent capture/escape of electrons and holes will also give a barrier height equal to the sum of the barrier heights for electrons and holes in a strong thermal quenching regime. Comparison of thermal PL quenching of two QD structures fabricated using nano-engineering approaches described above is shown in Fig. 12. One of the structures contains shape-engineered (truncated) QDs, as shown in Fig. 10a and b; and the other sample contains nontruncated QDs as in Fig. 10c and d. In both cases, QDs are embedded in identical GaAs QWs. The nontruncated QDs have somewhat larger volume, corresponding to a slight (;30 meV7) increase in barrier height. Therefore, nontruncated dots show a similar (within the accuracy of experiment) slope in the strong thermal quenching regime. However, at intermediate temperatures (close to room temperature), we observe a twofold improvement of PL intensity in the shapeengineered sample versus the nontruncated one. This improvement can be attributed to the suppression of carrier escape through some intermediate states, e.g., WL, but more data are needed to make a definite conclusion. It is important to emphasize that the key improvements in the nano-engineered QD medium result from the observed large barrier for escape

Fig. 11. Schematic of the electron-hole pair energy levels in the QD structure. QD energy levels correspond to the ground state and first excited state PL peak energies, GaAs QW energy level is calculated using its width, barrier corresponds to bulk Al0.2Ga0.8As. Energies are given at room temperature.

Nano-Engineering Approaches to Self-Assembled InAs Quantum Dot Laser Medium

from QDs (450 meV). This energy simply corresponds to the transitions between QD and GaAs QW ground states. This fact leads to the assumptions that (1) no significant excitation transport through states in the InAs WL is involved in the dynamics of the system; and (2) the excited states in QDs are weakly populated in the entire temperature range studied and therefore also do not contribute to the dynamics. Considering the reasons for assumption (1), we can speculate that the states in the WL are either pushed above those in the QW due to intermixing with the capping AlAs layer or these states are strongly localized (due to AlAs capping) and do not contribute to carrier transport. Assumption (2) corresponds to a large energy separation between the ground and first excited states (;100 meV) in the shapeengineered QDs demonstrated elsewhere.7 The large escape barrier and a large energy separation between the ground and excited states are also responsible for the unsurpassed thermal stability of the QD laser threshold current described in the Introduction.3 The reliability of optical emitters is a critical parameter and might be a real show stopper for implementation of optical interconnects at elevated temperatures that are typical for Si chips. In general, due to stronger localization of carriers, the QD medium is expected to demonstrate superior defect tolerance and, hence, reliability than other types of media with weaker localization. A direct way to test radiation or defect tolerance is to introduce nonradiative defects with ion implantation and to test the reduction of luminescence efficiency. As with thermal stability, a superior defect tolerance of QDs over QWs was demonstrated at temperatures below 77 K.22,23 At higher temperatures, the carriers start to spend

Fig. 12. Temperature dependence of the integrated PL intensity in an Arrhenius plot of shape-engineered and regular QD structures. All intensities are normalized to PL at 77 K. Symbols correspond to the experimental data and the line is a fit with the dynamic model,4 which neglects the carrier escape through intermediate levels. Activation energy deduced from the region of strong quenching is about 450 meV.

831

significant time outside the QDs and eventually recombine outside the dots. Because of strong localization at high temperatures, the nano-engineered QDs exhibit unsurpassed, over two orders of magnitude higher, defect tolerance than QWs at room temperature, as shown in Fig. 13. We emphasize the two major properties of the nano-engineered QD ensemble that give rise to its extremely high room-temperature defect tolerance and high PL efficiency at room and elevated temperatures. First, a relatively large barrier energy for carrier escape from the QDs (D 5 450 meV) results in a high-temperature offset (250 K) for carrier escape from the dots. Second, a low nonradiative rate in the barrier material (which is a 7.5-nm-thick GaAs QW in the sample studied) extends the high probability of carrier recombination in the QDs at temperatures as high as 320 K,4 and consequently shifts the offset of PL thermal quenching to higher temperatures. Resonant Tunneling QD-QW Structures As we have already mentioned, the maximum (saturated) gain of the QD medium is usually significantly less than that of QWs. This value can be increased by nano-engineering approaches described above: reduction of inhomogeneous broadening in the QD ensemble, enhancement of overlap between electrons and holes, and reduction of escape of carriers from the lasing (ground state) QD levels. Another effective way to improve the gain is to increase capture to the lasing level using tunnel injection of carriers into QDs from a QW.5,24 The QW is working as a reservoir for electrons and holes with high density of states and, therefore, with large capture cross section. Hence, carrier capture in QWs is more efficient than in QDs, and may provide a path for fast carrier transfer into the dots

Fig. 13. Quenching of the integrated PL intensity versus proton irradiation dose of nano-engineered QD and QW structures at 77 and 300 K.

832

by resonant tunneling. As a result, tunnel-coupled QW-QDs media exhibited extremely high maximum saturated gain24 and improved 23 GHz modulation bandwidth.5 We have evaluated QDs-on-QW tunnel-coupled structures to determine their optical properties. It was found that PL spectra (total efficiency and FWHMs) of QDs and QW were not affected at GaAs barrier thicknesses from 30 nm to 2.5 nm. We have chosen a GaAs barrier thickness of 2.5 nm (9 ML) that is definitely suitable for carrier tunneling but did not affect the efficiency of radiative recombination. A TEM image of this structure is shown in Fig. 14. Samples with different QW-QDs ground state energy separations were grown and studied by PL at 77 K and room temperature (Fig. 15). All samples exhibited good and comparable levels of PL efficiency at room temperature and 1 W/cm2 excitation intensity, with easily identifiable peaks corresponding to QD and QW ground states. The dependence of the QW/QDs PL ratio on ground state (GS) energy separation at 300 K was in good agreement with the thermodynamic equilibrium of carriers with an Arrhenius slope of 26 meV, corresponding to room temperature. The dependence of the QW/QDs PL ratio on GS energy separation at 77 K exhibited non-Arrhenius behavior, indicating nonthermodynamic carrier distribution for larger energy separations of 60– 120 meV. Increased intensity of QW excitons, with respect to the thermal activation behavior at low temperatures, takes place when the rate of electron transfer from QW to QDs becomes slower than the recombination rate. We can conclude, therefore, that the resonant tunneling in QDs is a dominant carrier loss channel from a QW when the difference in QW and QD ground state energies is less than ;50 meV at least for the 2.5-nm GaAs barrier. In fact, laser diodes with the tunnel QD-QW pairs has demonstrated three- to fourfold increase in the maximum optical gain, as described in the next section.

Oktyabrsky, Tokranov, Agnello, Van Eisden, and Yakimov

Fig. 14. TEM (200) DF micrographs of a 4x(QDs-on-QW) structure. Shape-engineered InAs QDs are separated from a 6-nm-thick InGaAs QW by a 2.5-nm-thick GaAs barrier.

A significant improvement of gain is achieved in tunnel QW-QD structures.25 The 4x(QDs-on-QW) structure shown in Fig. 14 and the recently developed 3x(QW-on-QDs), with a reverse order of growth of QDs and QW and possibly faster tunneling, have shown extremely high saturated modal gain. Despite the higher threshold current density, a 3x(QW-on-QDs) structure appears to be more promising for high-frequency modulation and use in VCSELs due to the observed highest modal gain of .50 cm#1. Another important emerging application of QDs is an active medium for photonic crystal nanocavities. Recently, we have demonstrated strong

Properties of Nano-Engineered QDs Tested in Optical Structures and Devices Table IV summarizes the parameters of different QD laser media fabricated using various nano-engineering techniques. The 3xQD medium indicates the nano-engineered triple layer QD structures, as shown in Fig. 10a and b. These structures demonstrated an unsurpassed characteristic temperature of 380 K at room-temperature operation,3 as described in the Introduction, as well as the unparalleled defect tolerance summarized in the previous section. To increase the gain, we have tested a seven-layer QD structure (7xQD) consisting of similar nanoengineered QDs in the well. The maximum gain is increased by a factor of 2, but the gain per layer is slightly reduced and is likely due to lower injection efficiency.

Fig. 15. QW/QDs PL intensity ratio of single pair tunnel QDs-on-QW samples versus GS energy separation at 300 K and 77 K. Arrhenius activation behaviors at two temperatures are shown by two straight solid lines.

Nano-Engineering Approaches to Self-Assembled InAs Quantum Dot Laser Medium

833

Table IV. Comparison of Tunnel Injection Lasers with Multilayer QD Lasers; Jth is a Laser Threshold Current and Lmin is a Minimum Cavity Length for Lasing MQD Lasers Lasing wavelength, mm Minimum Jth, A/cm2 Lmin, mm Maximum modal gain (per layer), cm#1

3xQD

7xQD

4x(QDs on QW)

3x(QW on QDs)

1.22 56 0.87 16 (5.3)

1.12 155 0.41 31 (4.4)

1.16 95 0.49 26 (6.6)

1.13–1.11 255 0.25 .50 (.17)

room-temperature QD emission into cavity modes of a 2-D waveguide photonic crystal at CW excitation intensity of as low as 500 W/cm2.26 The luminescence efficiency is high despite the high surface recombination at the air holes in GaAs-based waveguide structures. SUMMARY Semiconductor QD structures with discrete atomic-like electronic spectra have significant advantages over QWs that are widely used in optoelectronic devices. However, it is essential to employ various nano-engineering technologies to achieve these benefits. We studied the formation and properties of InAs QDs self-assembled on surfaces of different compositions, QD capping, and shape-engineering within a MBE reactor. These approaches were employed to control size, density, shape, and ultimately both homogeneous and inhomogeneous electronic spectra of QDs. The analysis of QD capping results can be explained by two processes on the surface. The first one is an evolution of QDs on the InAs surface toward larger sizes with time. The second effect is the surface redistribution of In out of the QDs on top of the growing capping layer. The effectiveness of AlAs as compared with GaAs as a capping layer for QDs is attributed to ‘‘freezing’’ of the growth surface during capping and prevention of continued QD evolution or In redistribution. Band engineering of QD heterostructures using digital alloys as high-quality wide bandgap barriers to enhance carrier localization by reduction in thermal escape of carriers from the dots, as well as tunnel injection from QWs to accelerate carrier transfer to the lasing state, was used to improve optical gain and efficiency of the QD medium. Beneficial properties of the developed QD media are demonstrated at room temperature in laser diodes with unsurpassed thermal stability with a characteristic temperature of 380 K, high waveguide modal gain .50 cm#1, unsurpassed defect tolerance over two orders of magnitude higher than that of QWs typically used in lasers, and emission from a 2-D photonic crystal nanocavity. ACKNOWLEDGEMENTS The authors highly appreciate the support by MARCO and DARPA through the Focus Center for Hyperintegration and the National Science Foundation (Contract No. ECS0334994).

REFERENCES 1. Y. Arakawa and H. Sakaki, Appl. Phys. Lett. 40, 939 (1982). 2. D. Bimberg et al., Thin Solid Films 367, 235 (2000). 3. V. Tokranov, M. Yakimov, A. Katsnelson, M. Lamberti, and S. Oktyabrsky, Appl. Phys. Lett. 83, 833 (2003). 4. S. Oktyabrsky, M. Lamberti, V. Tokranov, G. Agnello, and M. Yakimov, J. Appl. Phys. 98, 53512 (2005). 5. P. Bhattacharya, S. Ghosh, S. Pradhan, J. Singh, Z.-K. Wu, J. Urayama, K. Kim, and T.B. Norris, IEEE J. Quantum Electron. 39, 952 (2003). 6. S. Oktyabrsky, J. Castracane, and A. Kaloyeros, Proc. SPIE 4652, 213 (2002). 7. V. Tokranov, M. Yakimov, A. Katsnelson, M. Lamberti, and S. Oktyabrsky, Proc. SPIE 4999, 79 (2003). 8. M. Yakimov, V. Tokranov, G. Agnello, J. van Eisden, and S. Oktyabrsky, J. Vac. Sci. Technol. B 23, 1221 (2005). 9. C. Pryor, Phys. Rev. B: Condens. Matter Mater. Phys. 60, 2869 (1999). 10. O. Suekane, T. Okui, M. Takata, S. Hasegawa, and H. Nakashima, 13th Int. Conf. Indium Phosphide and Related Materials, Proc. (Piscataway, NJ: IEEE, 2001), p. 288. 11. P. Ballet, J.B. Smathers, H. Yang, C.L. Workman, and G.J. Salamo, J. Appl. Phys. 90, 481 (2001). 12. K. Shiramine, T. Itoh, and S. Muto, J. Vac. Sci. Technol. B 22, 642 (2004). 13. J.M. Moison, F. Houzay, F. Barthe, L. Leprince, E. Andre, and O. Vatel, Appl. Phys. Lett. 64, 196 (1994). 14. C. Heyn and C. Dumat, J. Cryst. Growth 227–228, 990 (2001). 15. J. Kim, L.-W. Wang, and A. Zunger, Phys. Rev. B: Condens. Matter Mater. Phys. 57, R9408 (1998). 16. K. Muraki, S. Fukatsu, Y. Shiraki, and R. Ito, Appl. Phys. Lett. 61, 557 (1992). 17. A. Rosenauer, D. Gerthsen, D. Van Dyck, M. Arzberger, G. Bohm, and G. Abstreiter, Phys. Rev. B64, 245334/1 (2001). 18. M. Schowalter, A. Rosenauer, D. Gerthsen, M. Arzberger, M. Bichler, and G. Abstreiter, Appl. Phys. Lett. 79, 4426 (2001). 19. L.V. Asryan, M. Grundmann, N.N. Ledentsov, O. Stier, R.A. Suris, and D. Bimberg, J. Appl. Phys. 90, 1666 (2001). 20. V.E. Tokranov, M. Yakimov, A. Katsnelson, M. Lamberti, G. Agnello, and S. Oktyabrsky, Proc. SPIE 5365, 72 (2004). 21. E.C. Le Ru, J. Fack, and R. Murray, Phys. Rev. B: Condens. Matter Mater. Phys. 67, 245318 (2003). 22. P.G. Piva, R.D. Goldberg, I.V. Mitchell, D. Labrie, R. Leon, S. Charbonneau, Z.R. Wasilewski, and S. Fafard, Appl. Phys. Lett. 77, 624 (2000). 23. N.A. Sobolev, A. Cavaco, M.C. Carmo, M. Grundmann, F. Heinrichsdorff, and D. Bimberg, Phys. Status Solidi B 224, 93 (2001). 24. G. Walter, T. Chung, and J.N. Holonyak, Appl. Phys. Lett. 80, 1126 (2002). 25. V. Tokranov, M. Yakimov, G. Agnello, J. Van Eisden, and S. Oktyabrsky, Proc. SPIE 5734, 65 (2005). 26. Y. Tang, A.M. Mintairov, J. Merz, V. Tokranov, and S. Oktyabrsky, 5th IEEE Conf. on Nanotechnology (Piscataway, NJ: IEEE, 2005), p. 189.