GaNAs and GaAsBi: Structural and electronic properties of two resonant state semiconductor alloys

Transcription

1 GaNAs and GaAsBi: Structural and electronic properties of two resonant state semiconductor alloys by Erin Christina Young B.Eng., McMaster University, M.A.Sc., The University of British Columbia, A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in The Faculty of Graduate Studies (Metals and Materials Engineering) THE UNIVERSITY OF BRITISH COLUMBIA December 13, 2006 c Erin Christina Young, 2006

2 Abstract ii Abstract Semiconductor alloys that are lattice matched to GaAs but have a smaller energy band gap are of interest for numerous applications, including infrared lasers for telecommunications, high efficiency solar cells, and high electron mobility transistors. For high optoelectronic efficiency, these materials must be highly perfect single crystals with low defect densities. In this thesis, two substitutional GaAs-based alloy families, nitrides and bismides, are investigated experimentally. In the first alloy, GaNAs, the addition of N results in a large band gap reduction, though the small size of the N relative to As introduces tensile strain into the lattice, and the high electronegativity of N attracts electrons. The second alloy, GaAsBi, also has a smaller band gap and is formed by the addition of the very large Bi atom to GaAs, which introduces compressive strain and tends to attract holes. The experimental investigations of these alloys focused on elucidating the relationships between the growth process, atomic structure, and electronic properties. Films were grown by molecular beam epitaxy (MBE) with in-situ process monitoring and subject to post-growth structural and electronic characterization. For GaNAs and a related alloy, InGaNAs, degradation in luminescence efficiency, mobility and structural integrity were observed as the nitrogen content of the alloy was increased. A comprehensive study of strain relaxation in compressively strained InGaNAs and InGaAs quantum wells revealed that the nitrogen alloying did not have an effect on the critical thickness for dislocation formation, or the dislocation density in relaxed films. At large lattice mismatch, InGaNAs quantum wells were observed to relax by means of unusually oriented pure edge-type misfit dislocations aligned with 100 directions, likely due to the high stress associated with the large misfit. Use of bismuth as a non-incorporating surfactant during growth was successfully applied to improve the material quality of the nitrides. The Bi surface layer during growth was investigated using in-situ electron diffraction intensity measurements, and was found to improve both the smoothness of nitride films, by promoting a layer-bylayer growth mode, as well as increasing the photoluminescence (PL) intensity by a factor of 2.4. The improvement in PL is attributed to a reduction in nitrogen

3 Abstract iii clusters with the surfactant. In addition, an increase in nitrogen content of up to 50% was observed in films grown with Bi over films grown without the surfactant. The increase in the nitrogen incorporation scales with the Bi flux, and saturates at full Bi coverage. A modified Langmuir model was applied to describe the behaviour of Bi on the surface, as well as the increase in nitrogen incorporation. The bismide alloy family requires atypical MBE growth conditions such as low substrate temperature and low As overpressure in order to achieve Bi incorporation. The conditions are close to those where Ga and Bi droplets form. However, insitu light scattering was used to identify and avoid growth with droplets, resulting in films with high structural quality as determined by x-ray diffraction, and strong photoluminescence. 1% Bi alloying in GaAs was also found to result in a minimal 15% decrease in electron mobility, as compared with a 94% reduction for 1% N alloying. Finally, a preliminary investigation was made into the concept of co-doping GaAs with both N and Bi. GaNAsBi films showed room temperature PL at long wavelengths commensurate with contributions to band gap reduction from both N and Bi. Since lattice matching to GaAs can be achieved with a Bi/N ratio of 1.7, GaNAsBi and GaAsBi are promising new alloys for the applications described.

4 Contents iv Contents Abstract Contents List of Tables List of Figures Acknowledgements ii iv vii viii xiv 1 Introduction Experimental Methods Crystal growth by molecular beam epitaxy Nitrogen plasma source In-situ monitoring techniques RHEED Mass spectrometry Elastic light scattering Ex-situ characterization High resolution x-ray diffraction Photoluminescence Atomic force microscopy Transmission electron microscopy Growth and properties of dilute nitrides Electronic defects and dilute nitrides Structural defects and dilute nitrides: the strain problem Relaxation by misfit dislocations Relaxation by surface roughening Relaxation by crack formation

5 Contents v Relaxation by phase separation Strain relaxation study of quantum wells Discussion Summary Conclusion Bismuth surfactant growth of the dilute nitrides Surfactant-assisted epitaxy Experimental details for Bi surfactant assisted growth Effects of Bi Surfactant Morphology Photoluminescence Nitrogen incorporation Understanding the surfactant effect Measuring the Bi surface coverage with RHEED A model for the surface coverage Modelling the increase in nitrogen incorporation In-situ mass spectrometry of the surface Discussion Summary Conclusion Growth and properties of dilute bismides History of GaAsBi and related alloys Exploring the growth conditions for GaAsBi In-situ light scattering and surface morphology Bismuth incorporation Optical and electronic properties Photoluminescence Electron mobility from THz spectroscopy Dilute nitride-bismides Summary Conclusion Conclusion Future Work

6 Contents vi Bibliography

7 List of Tables vii List of Tables 1.1 Properties of elements of interest from groups III and V; electronic structure, atomic mass, covalent radius and Pauling s electronegativity [GS01] Electron mobilities from Drude fits to the complex conductivity in GaAsBi, GaAsN and GaNAsBi films N and Bi concentrations and peak photoluminescence emission energy for four as-grown GaN x As 1 x y Bi y samples, as well as band gap of GaAs for reference. Bi concentrations determined by RBS and N concentrations determined by x-ray diffraction

8 List of Figures viii List of Figures 1.1 Band gap as a function of lattice parameter for elemental and compound semiconductors. Bowing due to N and Bi alloying with GaAs is indicated schematically by the dashed lines. The lattice parameter for the metal GaBi is also indicated Schematic illustration of surfactant-assisted semiconductor growth. The surfactant species segregates to the growth front, changing the structure and energy of the surface Schematic illustration of band structure for GaNAs and GaAsBi alloys, showing resonant states due to Bi, N and a nitrogen defect state, NN 2. Bands shown are the conduction band (CB), heavy hole band (HH), light hole band (LH) and split-off band (SO) due to spin-orbit coupling. Bi primarily affects the valence band, while N affects the conduction band Nitrogen content normalized by growth rate for GaN x As 1 x films grown at 450 C as a function of Pirani voltage in the plasma discharge tube. Corresponding growth chamber pressure shown on top axis Schematic of in-situ elastic light scattering experiment, courtesy of M.B. Whitwick. UV light is incident on the growing film, and detected at a non-specular angle with a photomultiplier tube. A rougher growth surface increases the intensity of scattered light Bandgap dependence on nitrogen content of GaN x As 1 x films showing good agreement between photoluminescence data from samples grown in this study, a fit to other experimental data [TFS02], and LCAO model [NAB + 04]

9 List of Figures ix 3.2 Room temperature photoluminescence of InGaNAs quantum wells with increasing nitrogen and indium concentrations. Legend lists composition and thickness estimates corresponding to PL of quantum wells from left to right (increasing emission wavelength). All samples were grown at temperatures between 450 and 475 C Possible misfit dislocation geometries and associated slip systems in GaAs a) typical 110 dislocation with b oriented 60 to the interface and line direction and b) unusual 100 dislocation with b oriented 45 to the interface and 90 to the line direction High-resolution x-ray diffraction patterns from GaAs/QW/GaAs heterostructures, lattice mismatch (1.7 %) where the QW is (a) InGaNAs (b) InGaNAs grown with a Bi surfactant, and (c) InGaAs. In the dilute nitride QWs the interface coherence begins to degrade between a 7.5 and 9 nm QW thickness. The use of a Bi surfactant during growth (1 x 10 7 Torr) had no apparent effect on the critical thickness. The stronger fringes in the case of InGaAs indicate that the strain is more uniform compared to the dilute nitride samples, but the strain relaxation begins between 8 and 12 nm, similar to the other series. Data are offset for clarity Photoluminescence from the series of GaAs/InGaAsN/GaAs QWs (In- 28% N-1%) at room temperature as a function of QW thickness. At the onset of relaxation between a QW thickness of 7.5 and 9 nm the PL peak intensity decreases by more than 2 orders of magnitude. The 10 nm QW has almost no detectable photoluminescence, consistent with the presence of numerous defects Plan-view TEM images of GaAs/QW/GaAs (001) samples obtained with a g = (220) diffraction condition in the direction indicated by the arrow. Figure a) is a strong g, bright field (BF) image of a InGaAsN QW (9 nm) showing the beginnings of dislocation loop formation; b) strong g, BF and weak-beam dark-field (WBDF) image pair from an InGaAsN QW (10 nm), c) BF and WBDF pair from a InGaAs QW (12 nm) and d) WBDF of a InGaAsN QW (12 nm). All dislocations are in contrast for (220) diffraction but depending on the deviation from perfect Bragg diffraction of the sample, the loop interiors are in contrast in b) WBDF and c) BF and out of contrast in all other images. Data courtesy of K. Kavanagh [YKT + 05]

10 List of Figures x 3.7 Elastic light scattering signal recorded during growth of three, capped InGaNAs QWs at 450 C with thicknesses as indicated. The QW growth begins at Time = 0 for a duration of less than a minute followed by a 240 nm GaAs capping layer Elastic light scattering signal recorded during growth at 450 C of three films with 1.7% mismatch past their critical thickness. The top signal corresponds to growth of a 175 nm InGaNAs layer, the middle signal corresponds to growth of a 220 nm InGaNAs layer with a Bi flux of 1 x 10 6 Torr and the bottom layer corresponds to growth of a 220 nm InGaAs layer. The Bi surfactant does not affect the critical thickness but reduces the initial rate of relaxation in the InGaNAs layer. Roughening rates at the onset of relaxation are indicated for each growth RHEED oscillations during growth at 440 C of (a) GaAs, without Bi (top) and with a Bi flux of 1 x 10 7 Torr (bottom) and (b) GaN x As 1 x. The top set of oscillations correspond to a film with x = 1.4% N grown with a Bi flux, the bottom set correspond to a film with x = 0.01% grown with a Bi flux, and the middle two sets of oscillations correspond to films with x = 0.45%, one grown with Bi flux and the other without, as indicated. All samples grown with Bi had a flux of 1 x 10 7 Torr Room temperature PL from as-grown and annealed 6 nm InGaNAs quantum wells grown at 450 C with and without Bi surfactant (Bi flux of 10 7 Torr). Max intensity increased 2.4x for the sample grown with Bi for both conditions. Dashed lines correspond to samples annealed for 60 s at Temperature dependent PL spectra for 250 nm thick GaN x As 1 x films grown under identical conditions at 500 C, except that (a) r1349 was grown with no Bi and (b) r1350 was exposed to a Bi flux of 4x10 6 Torr. A low energy tail attributed to N cluster states is seen in the spectrum of r1349 and is greatly reduced for the sample grown with the surfactant. PL data courtesy of D. Beaton [Bea03]

11 List of Figures xi 4.4 High resolution XRD data for two GaN x As 1 x films grown at 430 C with (r lower diffraction pattern) and without (r upper diffraction pattern) a Bi flux of 4 x 10 6 Torr. Dashed lines represent simulations from dynamical theory indicating that N content increases by 36% for the film grown with Bi. Data are offset for clarity RHEED patterns of (3x1) Bi-stabilized reconstruction along 110 azimuths during growth at 370 C Change in RHEED specular intensity from GaAs substrate at different substrate temperatures when exposed to a Bi flux of 1.4x10 5 Torr, and subsequent decrease in intensity when flux is removed. Inset shows desorption rate vs. inverse temperature (a) Bi coverage vs. growth temperature at constant Bi flux and (b) Bi coverage vs. Bi flux at constant substrate temperature inferred from RHEED data Nitrogen concentration vs. Bi flux for GaNAs samples grown at 400 and 460 C Line-of-sight RGA partial pressure of mass number 89 (AsN) under two different growth conditions at 450 C, with arrows correspond to opening and closing of Ga and Bi shutters as indicated. a) Bi flux is introduced during plasma operation, resulting in increase in AsN pressure. b) Bi flux is introduced during growth of GaNAs (plasma on, Ga shutter open), again resulting in an increase in AsN pressure AFM images of two GaAsBi surfaces showing large metal droplets: a) Film r1710 grown at 350 C with high Bi flux. Image shows a 5x5 µm scan with 300 nm vertical scale. b) Film r1688 grown at 375 C with droplet formation when the As flux was dropped below a critical value. Growth of GaAs was continued after the occurrence of droplets. Image shows a 10x10 µm scan with 250 nm vertical scale In-situ diffuse UV light scattering intensity and As 2 pressure as a function of time during dilute bismide growth of r1688 by molecular beam epitaxy at 365 C. The arrow indicates the opening of the shutter on the Ga and Bi sources. The chamber pressure is a measure of the As 2 flux on the growth surface which is controlled by a mechanical valve on the arsenic cracker

12 List of Figures xii 5.3 In-situ diffuse light scattering intensity and As 2 pressure during growth of r1713 GaAs 1 x Bi x with x=0.8% at 385 C. The initial surface roughening and subsequent rapid smoothing is caused by the thermal oxide desorption followed by GaAs buffer layer growth. The roughening at long growth times is due to the reduction in As 2 and is associated with the formation of the rippled surface morphology illustrated in Figure x2 µm AFM images of three films grown at 385 C showing increasing scale of roughness with increasing Bi content a) r1712, GaAs, no Bi detected by x-ray. Vertical scale is 10 nm and rms roughness is 1.1 nm b) r1711, GaAs 1 x Bi x, x=0.33%. Vertical scale is 25 nm and rms roughness is 3.8 nm c) r1713, GaAs 1 x Bi x, x=0.8%. Vertical scale is 50 nm and rms roughness is 7.7 nm Atomic force microscope image of a dilute bismide film (r1698) grown at 390 C. The image area is 2x2 µm, the vertical scale is 3 nm and the rms roughness is 0.2 nm High resolution x-ray diffraction θ/2θ scans for three GaAs 1 x Bi x films. The top scan corresponds to a 210 nm film with x=0.5% (r1696) grown at 330 C, the middle scan to a 125 nm film with x=1.9% (r1698) grown at 390 C, and the bottom scan to a 30 nm QW with x=3.0% (r1459) grown at 340 C, capped with 290 nm of GaAs. Data are offset for clarity. The dashed lines show diffraction scans simulated by dynamical theory. The presence of pendellösung fringes in the experimental data indicates the high structural quality of the films Plot of Bi concentration vs. inverse temperature. The Bi concentration data is from Huang et al.[hofy05], and is plotted with an Arrhenius function with an activation energy of 1.34 ev and saturation value of 4.75% Bi. The predicted Bi concentration is based on Equation Room temperature photoluminescence spectra for three GaAs 1 x Bi x films with x = 1.7, 1.9 and 3.0%. Growth temperatures were 370, 390, and 340 C respectively. PL spectrum for a 10 level InGaAs multiquantum well grown at 450 C is also shown for reference (dashed line)

13 List of Figures xiii 5.9 Extracted complex conductivity for (a) GaAs buffer layer, (b) GaAsBi (0.84% Bi), (c) GaNAs (0.84%N), and (d) GaNAsBi (0.85%N, 1.4% Bi),10 ps after 400 nm excitation at a fluence of 3.7 µj/cm 2 and carrier concentrations of 2-3 x cm 3. The solid and dashed lines are Drude fits to the real and imaginary parts of the conductivity (data taken by D.G. Cooke at University of Alberta) Properties of GaNAsBi alloys a) Room temperature photoluminescence and electroreflectance (ER) spectra showing the optical bandgap for r1316 GaN x As 1 x y Bi y with x=0.85%, y=1.4% (top sample in (b)). The dotted line is a fit to the ER data. b) Experimental and simulated high resolution x-ray diffraction patterns for two GaNAsBi compositions as shown (data offset for clarity)

14 Acknowledgements xiv Acknowledgements Without hesitation, the first acknowledgement here must go to my supervisor Tom Tiedje. Thank you Tom, for the countless opportunities you made possible to me during my PhD; from conferences to collaborations I have been a very privileged graduate student. Thanks for always having faith in my work, and for always being available to discuss some aspect of physics or my MBE problems, or give advice (when solicited) with characteristic logic. Your breadth of knowledge and passion for science are truly inspiring. Thanks to Warren Poole, my co-supervisor, for once again acting in this role for me in a project outside your direct area of research! Your guidance and support throughout the project in many diverse ways is much appreciated, and this interdisciplinary project could not have been realized without you. Karen Kavanagh has also been a important mentor, collaborator and a committee member over the course of this work. It s been a pleasure to work with you. There have been many other professors at UBC who have been helpful to me along the way; Chad Sinclair, Matthias Militzer and Elizabeth Croft in particular. Thanks for your advice and support. Some aspects of this work were performed in collaboration with individuals at UBC and other institutions. I would like to thank my collaborators for their important contributions to this work, and for making it so fun and interesting to design and produce new samples. Thanks to Karen Kavanagh and Alexey Koveshnikov at Simon Fraser University for all the transmission electron microscopy work (and for teaching me everything I know about TEM), to Dan Beaton at UBC who obtained such beautiful low temperature photoluminescence measurements on my samples, and David Cooke and Frank Hegmann at University of Alberta for performing the THz spectroscopy to determine electron mobilities. Angelo Mascarenhas, Sebastien Francoeur, and Brian Fluegel at National Renewable Energy Lab also deserve a special mention for initiating the bismide project several years ago, and their continued interest and support of this work. And of course I have to thank my fellow MBE growers. First and foremost, thanks

15 Acknowledgements xv are due to Sebastien Tixier, for teaching me how to do MBE in the first place, your collaboration and assistance, and for being such an all-around great human being and friend. Also to Mike Whitwick, for keeping the MBE up and running with me during the home stretch, help with the light scattering experiments, and lending your brute strength when necessary! To Nikolaj Zangenberg for commiserating with me during the worst of MBE downtimes, and Jens Schmid who gave me the good advice to not get so stressed about it. Extra special thanks to Jim Mackenzie, who is not a grower but knows the MBE better than anyone and can fix it too... Life inevitably undergoes many ups and downs during the period of time it takes to complete a PhD, and through it all the lab remained a stable and comfortable constant in my life that I really valued. I will miss our many discussions over coffee (even though they all too often turned to cows), group lunches at the arsenic table, and Friday afternoon liquid nitrogen beers. I might even miss the lab hikes (ok, just hearing about them when everyone finally manages to make it back). Thanks to all my labmates for your support, technical and otherwise over the years; Anders Ballestad, Martin Adamcyk, Dan Beaton, Scott Webster and Eric Nodwell (both particularly for all your help with computer, Latex and Linux issues), Richard Mar, Jens Schmid, Shawn Penson, Raveen Kumaran, Mike Whitwick, Ryan Lewis, Johanna Hansen, Christopher Robin, Nikolaj Zangenberg, Sebastien Tixier, Kevin Mitchell (actually, special thanks to Kevin for having such great taste in music). And more thanks to Miryam Elouneg for your friendship and unique perspectives on life and physics! My time in graduate school was greatly enriched by my work with a number of committees and projects, often around the themes of science outreach and women in science/engineering. Thanks to all the inspiring people I worked with at SCWIST, IWIS, DAWEG, NEW@UBC, Dec. 6 committee, YWCA mentorship, the JADE Project and UBC Tri-mentoring, for your vision and leadership. Thanks especially to Donna Dykeman, and Elana Brief for sharing my passion for these causes, and becoming such good friends. Most importantly, thanks to all my friends and family for your love, support, and for providing me with perspective from outside the academic bubble. And finally, many thanks to Pat, whose support through these last stages has been invaluable.

16 Chapter 1. Introduction 1 Chapter 1 Introduction It is difficult to overstate the role that compound semiconductors have played in the technological developments of the past 30 years. From CD and DVD players to fiber optic communication and medical imaging, from cell phones and Blackberries to sensors and solid-state lighting, there has constantly been a new materials system to explore, a promising solution to a long-perplexing problem, or a novel application made possible for the first time. There has been a close synergy between these technological successes and new fundamental materials science knowledge that has been achieved through semiconductor research. The sheer diversity of compounds and properties available is a function of researchers having the combinatoric power of 4 full periodic table columns at their disposal, as well as an ever-improving ability to manipulate material at atomic scales with unprecedented purity and control. Non-equilibrium growth techniques such as molecular beam epitaxy (MBE) make it possible to produce metastable compounds not found in nature. In addition, the fact that there are generally only 2 crystallographic systems available to group IV, III-V, and II-VI compounds simplifies the alloying process and allows for heteroepitaxy within the constraints of lattice parameter. These conditions set the stage for a fascinating and dynamic field of research that may be as fundamental or applied as one wishes to practice. At least until the present day, it can be argued that GaAs has been the most technologically important compound semiconductor. GaAs exhibits a number of key properties that allow it and its alloys to find diverse application in electronics and optoelectronics. It is a direct bandgap semiconductor, allowing for more efficient generation of light than elemental semiconductors such as Si and Ge and is therefore well suited to light emitting devices including lasers. Carrier mobilities are also high in GaAs alloys, allowing the production of high speed transistors. Finally, one of the most significant advantages of GaAs is that it can easily be alloyed with other group III (Al, In) and group V (P, Sb) elements to achieve well-controlled bandgap engineering, since E g is a function of the composition of the alloy. The majority of the III-V semiconductors share the cubic zincblende crystal structure, and while alloying

17 Chapter 1. Introduction 2 may change E g significantly, the difference in lattice parameter is frequently small enough that different GaAs-based alloys can be grown epitaxially upon each other, enabling the high quality heterostructures of varying optical and electronic properties that are essential for optoelectronic devices. In recent years a number of previously untested group V elements have been introduced into the GaAs alloy system with interesting results. In the 1990s, small amounts of nitrogen were added to GaAs to form GaN x As 1 x, where x is less than 4% [KUHM94]. Since the compound GaN has a wide band gap of 3.2 ev, and the properties of ternary semiconductors can usually be well approximated by an interpolation of the properties of the two binary constituents, it was quite a surprising result that the dilute nitride GaNAs has a smaller band gap than the 1.42 ev bandgap of GaAs, i.e. GaNAs exhibits giant bandgap bowing. It has been shown that the introduction of only 1% nitrogen (x = 0.01) can reduce the bandgap by 0.2 ev [TFS02], as illustrated schematically in the graph of band gap as a function of composition for several III-V materials in Figure 1.1. This band gap reduction is anomalously large and is greater than the bowing associated with alloying any other group V element into GaAs. Since this discovery, dilute GaNAs and InGaNAs alloys have received considerable attention as new candidates for long wavelength (i.e. small E g ) light emitting devices in the technologically important near infrared range and a number of lasers have already been produced [FGR + 03]. These lasers are useful for telecommunications, as light signals at wavelengths of 1.3 or 1.55 µm have the lowest losses during transmission along fiber optic cables. There is also interest in InGaNAs for superluminescent diodes as a broadband near-infrared light source for a medical imaging technique known as optical coherence tomography [Web04]. In addition, high efficiency solar cells are a possible application for this material [BCT02]. However, while the addition of N into GaAs alloys is successful from a bandgap engineering point of view, it has been less successful from a material quality perspective. Work done at both UBC and elsewhere indicates that a number of key properties deteriorate with increasing nitrogen including photoluminescence [YHJW99], an indicator of the optical efficiency of a material, and electron mobility [Str01, MVD + 03]. This degradation may be due to the tendency of nitrogen to cluster to relieve its substitutional strain in the lattice, or be an intrinsic effect due to N clusters produced by random distribution of N [FNJ + 99]. Depending on the process conditions, the incorporation of nitrogen during the growth of GaNAs by the principal production technique of MBE can create a morphological growth instability, resulting in undesirably rough surfaces and interfaces [ATR + 01]. Finally, the small size of the N atoms

18 Chapter 1. Introduction 3 Figure 1.1: Band gap as a function of lattice parameter for elemental and compound semiconductors. Bowing due to N and Bi alloying with GaAs is indicated schematically by the dashed lines. The lattice parameter for the metal GaBi is also indicated. relative to the As atoms they are replacing introduces strain into the lattice, and thus when GaNAs is grown on GaAs the lattice mismatch can result in elastic or plastic relaxation processes, the artifacts of which (surface roughness, dislocations, cracks) severely compromise the functionality of the material [WBGB05, LWN05]. In the case of InGaNAs on GaAs, the small atomic size of the nitrogen helps to compensate for the larger atomic size of the In, but highly strained structures are still necessary to achieve the required infrared bandgap. A complete understanding of strain relaxation processes in semiconductors, and particularly nitrogen containing semiconductors is lacking, yet its avoidance is critical for synthesis of optoelectronic materials. One approach to improving the quality of the nitrides of interest (GaNAs and InGaNAs) that is discussed in this thesis is to introduce surfactants during growth of the material by MBE - surfactants in the form of a heavy group V element. As these elements are isoelectronic with As they are not expected to be electronically active in the material, even if they incorporate. Researchers at Stanford have reported success with antimony [YBW + 05] and at UBC, bismuth [TAY + 02]. For both cases, improvements in the luminescence efficiency of the material are observed for surfactant-assisted growth, as well as smoother surfaces. Due to their position in the periodic table, both of these elements have large atomic radii, in the case of Bi so large that atoms tend not to incorporate into the growing film under typical growth

19 Chapter 1. Introduction 4 Figure 1.2: Schematic illustration of surfactant-assisted semiconductor growth. The surfactant species segregates to the growth front, changing the structure and energy of the surface. conditions, but rather segregate to the surface, as illustrated schematically in Figure 1.2. The use of Bi as a surfactant progressed to successful recent attempts to incorporate Bi into GaAs. This was achieved by dramatic deviation from typical GaAs growth conditions in MBE [TAT + 03]. Similar to the effect of N in GaNAs, GaAsBi was also found to have a large band gap bowing with increasing Bi content (0.088 ev/%bi), and to exhibit luminescence [FSM + 03]. The other end member of the GaAsBi compound, GaBi (lattice parameter indicated on Figure 1.1), is predicted to be a metal [JWZ02], so a band gap reduction is perhaps more intuitive than for the case of N. However, similar to the case with N, the confirmation that Bi reduces the band gap of GaAs means that the GaAsBi is a potential candidate for the same types of optical applications as GaNAs, and may even be more suitable if the material quality is superior to GaNAs. The discovery also supports predictions that co-doping of Bi and N could provide a material lattice matched to GaAs with emission further in the infrared region than any other III-V compound [MZVS01]. One further point of interest is that GaAsBi was recently discovered to have giant bowing as a function of Bi content in the spin-orbit splitting energy, o, [FFM + 06]. This describes the strength of coupling between the orbital angular momentum and the spin angular momentum, and is expected to be greater for heavier atoms. This result may be relevant to the new field of spintronics. Comparison of the two alloys, GaNAs and GaAsBi, provides an interesting study. GaNAs has been studied extensively over the last decade and the literature is well developed, while GaAsBi is quite new. However a number of remarkable similarities have been made apparent in a short time. Both materials are nonconventional alloys,

20 Chapter 1. Introduction 5 Table 1.1: Properties of elements of interest from groups III and V; electronic structure, atomic mass, covalent radius and Pauling s electronegativity [GS01] Element Electronic Structure Atomic Mass Covalent Radius Electronegativity Ga (Ar) 3d 10 4s 2 4p nm 1.81 In (Kr) 4d 10 5s 2 5p nm 1.78 N (He) 2s 2 2p nm 3.04 As (Ar) 3d 10 4s 2 4p nm 2.18 Sb (Kr) 4d 10 5s 2 5p nm 2.05 Bi (Xe) 4f 14 5d 10 6s 2 6p nm 2.02 in the sense that the alloying element (N or Bi) introduces bound states in addition to energy bands, and behaves more like an isoelectronic acceptor (N) or donor (Bi) than a true alloying element [ZMW05]. The large reduction in band gap due to N is thought to be due to a resonant interaction between the nitrogen 2s state and the bottom of the conduction band [SWA + 99]. A similar resonant interaction is expected between the Bi 6p state and the valence band maximum [ZMW05]. A schematic drawing of these band structures for GaNAs and GaAsBi is shown in Figure 1.3. This difference in behaviour can be understood by examining the substitution process required to form the alloy. Usually, substitution of an isovalent impurity results in a weak perturbation to the host band structure. For the case of GaNAs or GaAsBi, N or Bi replaces some of the As anions in the lattice. The N atom has a smaller atomic radius than As, and introduces tensile strain, while the Bi is larger and introduces compressive strain. A large difference in the atomic potential can produce a localized center attractive to holes or electrons depending on its position relative to the conduction band or valence band. In addition to its small size, N is highly electronegative (a measure of its core potential), and is attractive to electrons, while Bi is large and tends to attracts holes. Atomic properties of group III and V elements of interest are shown in Table 1.1, illustrating the large differences in electronegativity and atomic size between As, and Bi and N which replace it in the lattice. The growing demand for new technologies coupled with the unusual properties of these materials provides the motivation for the current study. In this thesis we investigate the growth and properties of dilute (i.e. 0-3%) N and Bi containing alloys of GaAs, with particular attention to the relationships between growth, atomic

21 Chapter 1. Introduction 6 Figure 1.3: Schematic illustration of band structure for GaNAs and GaAsBi alloys, showing resonant states due to Bi, N and a nitrogen defect state, NN 2. Bands shown are the conduction band (CB), heavy hole band (HH), light hole band (LH) and split-off band (SO) due to spin-orbit coupling. Bi primarily affects the valence band, while N affects the conduction band. structure, and optical and electronic properties. A general introduction to the experimental methods used to grow and characterize semiconductor thin films is given in Chapter 2, with emphasis on the in-situ monitoring techniques that provide insight into the atomic processes occurring at the surface during growth. In Chapter 3, further introduction to the electronic and structural properties of dilute nitrides is given, and a study of strain relaxation in dilute nitride quantum wells is presented. In Chapter 4 we review the concept of surfactant-assisted crystal growth, and present results on the effects of the Bi surfactant on the properties of the dilute nitrides. Work done to characterize and model the Bi surface layer is also included. In Chapter 5, we move to an investigation of the growth conditions necessary to incorporate Bi into GaAs, as well as a study of the optical and electronic properties of the resultant GaAsBi films. Each of the principal body chapters (3,4,and 5) begins with a review of relevant literature and summarizes experimental issues unique to the topics discussed therein. Important results for each chapter are summarized at its conclusion.

22 Chapter 2 Chapter 2. Experimental Methods 7 Experimental Methods The growth of single crystal semiconductors with the purity (ppb) and defect density (5 x 10 3 /cm 2 or less) required for optoelectronic applications requires extreme measures. The technique of molecular beam epitaxy (MBE) is performed under ultrahigh vacuum (UHV), and is accomplished with thermally evaporated beams of atoms and molecules directed at a heated substrate. The mean free path of a gas molecule at UHV pressures (10 10 Torr) is on the order of several hundred kilometres. This degree of vacuum lends two main advantages to the MBE method of crystal growth, the first clearly being that the probability of impurity incorporation, particularly of oxygen and/or water, is exceedingly low. The second advantage is the high degree of control over the phase transformation from vapour to solid state. In particular, MBE is well suited to the growth of GaAs and its substitutional alloys, as condensation of the vapour into Ga 0.5 As 0.5 is thermodynamically preferred for a large window of MBE-accessible temperatures and pressures. This MBE growth window is bounded by a two phase region consisting of solid GaAs and a Ga-rich liquid. Provided that the system is As-rich, achieved in the MBE by applying an overpressure of As 2, GaAs is stable, and high quality crystals can be grown. 2.1 Crystal growth by molecular beam epitaxy Samples discussed in this thesis were grown in a VG-V80H molecular beam epitaxy deposition system with standard Knudson effusion cells generating the flux of Ga, In, Al, and Bi and a two zone thermal valved cracker source to produce As 2 dimers as the primary group V flux. Electromagnetic or pneumatic shutters on each cell are opened and closed manually or by computer control to regulate the combination of elements incorporating at a given time. UHV on the order of Torr is required preceding growth, while the fluxes, or beam equivalent pressures (BEP) of the constituent elements during growth ranged from Torr depending on the target composition. The BEPs for each source element were measured using a retractable ion gauge over a range of cell temperatures. As is typical in MBE, the

23 Chapter 2. Experimental Methods 8 group V/III flux ratio was kept high at values between 1 and 8, meaning that there was an overpressure of group V elements while the arrival of the group III elements at the crystal surface controlled the growth rate. Growth rates of the individual group III elements were determined by high resolution x-ray diffraction (see section 2.3.1), and combined with flux calibrations to predict compositions and thicknesses over a range of source temperatures. Typical growth rates are on the order of 1 µm/hr. All films were grown on 350 µm thick, polished, (001)-oriented GaAs substrates with a maximum axial tilt of ± 0.5. Prior to growth the substrates were degassed for 1 hour at 480 C in the MBE preparation chamber. They were then loaded into the growth chamber and heated to 615 C under an arsenic overpressure for 15 minutes to remove the native surface oxide. The thermal desorption process tends to leave large pits in the growth surface, typically separated by a distance of approximately 1 µm with a maximum depth of around 40 nm [Bal05]. Following desorption, the substrate temperature was dropped to 580 C to smooth the pitted surface and grow a GaAs buffer of nm before the desired films were grown at a specified temperature. Temperatures between 400 and 500 C provide the best compromise between optimum nitrogen incorporation and material quality [Ada02], so most nitride films were grown in this range. Films containing indium were also typically grown in this temperature range in order to avoid the surface roughening due to strain that InGaAs films are prone to at higher growth temperatures [CPE96], and since In concentration drops quickly at temperatures above 500 C. Films incorporating bismuth require much lower growth temperatures, and this is discussed in detail in Chapter 5. Several types of structures were grown in this work. Structures studied include bulk films (up to 1000 nm thick, though typically 250 nm), quantum wells (QWs), which are thin layers (between 5 and 30 nm thick) of small bandgap materials, bracketed on both sides by a higher bandgap material, i.e. GaAs, for electron confinement, and combination structures with both QW and bulk layers. In some instances QWs were also grown to have thin AlGaAs cladding layers on either side, or bulk layers were grown on top of thick AlGaAs or AlAs layers to provide a barrier between the film of interest and the substrate. Samples were rotated at moderate speeds ( 0.5 Hz) during growth to ensure uniform deposition. Temperature during growth was monitored using an optical bandgap thermometer with an absolute accuracy of approximately ± 2.5 C [JT97, JLT93]. Depending on the experiment, some samples were subject to a post-growth annealing treatment, either in the MBE growth chamber or in an external rapid thermal annealer (RTA). The RTA uses heat lamps to achieve heating in a quartz

24 Chapter 2. Experimental Methods 9 chamber under nitrogen atmosphere. Semiconductor films in this study were annealed at temperatures between 600 and 800 C for up to 60 s Nitrogen plasma source Due to the high chemical stability of the triply bonded N 2 molecule, a plasma source is necessary to provide reactive nitrogen species for incorporation. This is accomplished with a helical radio frequency (RF) resonator plasma source located in an effusion cell port on the MBE. The plasma source was previously built and designed at UBC, and consists of a ten-turn gold plated copper helix inside a coaxial tube that is coupled to an oscillator and power amplifier through a coupling loop and a coaxial cable. In the absence of a plasma discharge there are two unloaded resonances at about 63 MHz and 180 MHz. Since the source is a resonant device, high fields are present in the unloaded resonator, so there is spontaneous ignition of the plasma under a range of operating pressures, an advantage not common to commercially available plasma sources. Ignition is clearly indicated by the observation of a violet glow though a small window at the back of the plasma source. The plasma source is operated with high purity ( %) N 2 gas that is passed through an in-line purifier and then introduced to the active region of the source via a leak valve. The active region is a pyrolytic boron nitride (PBN) tube located inside the copper helix, and gas pressure inside the tube is measured with a Pirani gauge. A two stage PBN baffle at the exit of the tube acts to reduce the number of energetic ions incident on the substrate by 2-4 orders of magnitude [ZTN + 05]. During plasma operation, the forward and reflected RF power, P F and P R, are monitored with a directional wattmeter and the difference is the net power to the load, P L = P F P R. From measurements of the frequency and power required to start and maintain the plasma [ZTN + 05], the helix resonance is believed to be at 149 MHz with the plasma on, though another maximum in the net power to the load P L (typically around 85 W) occurs at 187 MHz. Once ignited the plasma discharge remains ignited throughout a wide frequency range, so some tuning is performed to maximize the input power. Most nitride films discussed in this thesis are grown in this second high power mode of operation, which we refer to as off-resonance. An increase photoluminescence intensity provides some evidence that the electronic quality of material grown under this mode of operation is superior to material grown in the on-resonance condition. It is unclear which of the species generated by the plasma source play a dom-

25 Chapter 2. Experimental Methods 10 inant role in the growth of dilute nitride semiconductors. Both atomic N as well as various metastable species are generated by the source, as determined by measuring the optical emission spectrum of the plasma [ZTN + 05]. However, under the off-resonance growth conditions typically used in this study, there appears to be an abundance of one particular metastable species (denoted N 2), which is 6 ev above the N 2 ground state and is also believed to be the main active species in the growth of GaN [PMM + 99] For a given growth rate, the nitrogen content of the film is controlled by varying the pressure inside the discharge tube, as shown in Figure 2.1. Once calibrated, the nitrogen content of the films as a function of gas pressure inside the tube remains reproducible over several months. However, a gradual decrease in the efficiency of nitrogen incorporation at similar plasma operation conditions was observed over a period of several years of operation of the source, particularly during times when the MBE was vented, and subsequently baked, more frequently. Eventually, it was decided that an investigation into the condition of the source was warranted. When the source was disassembled, a gray deposit was observed in the discharge tube and in the baffle. This deposit was assumed to be arsenic metal. After cleaning the discharge tube and baffle in aqua regia (HNO 3 :HCl 1:3), the efficiency of N incorporation increased by a factor of approximately 3. It is possible that the As coating on the discharge tube changes the relative ratios of the atoms and excited state molecules, though the plasma emission spectrum was similar before and after the tube cleaning [ZTN + 05]. As a result it may be more likely that the metal on the tube walls acts as a catalyst for recombination of atomic N, or deactivates the N 2 species. 2.2 In-situ monitoring techniques RHEED The in-situ monitoring technique of reflection high-energy electron diffraction (RHEED) is indispensable for high quality MBE film growth. RHEED is a surface sensitive diffraction technique that works on the principle that if electrons are incident on a crystal surface in grazing incidence, their penetration distance will be very small due to the small component of incident electron momentum normal to the surface. The resultant diffraction pattern will be representative of the surface atomic arrangement because the penetration distance is so short. Due to the very small wavelength of electrons at typical RHEED energies, the radius of the Ewald sphere is extremely

26 Chapter 2. Experimental Methods 11 N Concentration (%) x Growth Rate (nm/min) Chamber Pressure (Torr) Normalized [N] content Growth Temp 450 C Pirani Voltage (V) Figure 2.1: Nitrogen content normalized by growth rate for GaN x As 1 x films grown at 450 C as a function of Pirani voltage (a measure of the gas pressure in the plasma discharge tube). Corresponding growth chamber pressure shown on top axis. large compared to the reciprocal lattice spacings in the crystal, and thus the circumference of the Ewald sphere is coincident with the (00) rods in the 2D reciprocal lattice. This results in a diffraction pattern of elongated streaks, the spacing of which are related to the lattice parameter along a particular azimuth, taking into account the diffraction geometry. The surface unit cell measured with RHEED is often different than that which would be measured in a bulk crystal. This is due to the relaxation process known as surface reconstruction, in which surface atoms reposition themselves to new equilibrium spacings in order to lower the energy associated with dangling bonds. The standard convention for surface reconstruction is to describe it in two dimensions, in terms of the lattice parameter of the bulk. By this convention, if no surface reconstruction occurs, the structure is called (1x1). In (001)-oriented GaAs, the stable reconstruction at high temperature (600 C) under As overpressure is (2x4), which means that the wavelength of the surface periodicity has doubled in the [1 10] direction, and quadrupled in the [110] direction, resulting in a surface atomic density that is greatly reduced over that of the bulk. Because RHEED monitors changes in the crystal structure of the sample surface, it is sensitive to many processes critical to film growth, including oxide desorption, strain relaxation, phase transformations, etc. For example, the large mounds and pits

27 Chapter 2. Experimental Methods 12 left behind by the thermal oxide desorption process are clearly visible with RHEED. The three-dimensional pitted surface results in a typically spotty RHEED pattern, as the incident RHEED beam is diffracted by the 3-D material within the pits. As the growth proceeds, diffusional processes tend to smooth out the pits, and the streaky RHEED pattern of reciprocal lattice rods appears. RHEED is also well suited to study film growth dynamics, since the intensity of the diffracted beams oscillates when growth follows a two-dimensional layer-bylayer growth mode, and in fact the period of the oscillations corresponds to the monolayer growth rate. This means that a RHEED camera and appropriate software can be used to track RHEED intensity and thus to calibrate growth rates. Though there has been some controversy in the literature on whether or not the maximum intensity in the oscillations actually corresponds directly to the completion of a single monolayer, in a model proposed by Joyce and co-workers, [ZNDJ87], the oscillations are explained by an interference effect due to multiple electron scattering events between the top and the bottom of the reconstructed adlayers. In this work we perform RHEED experiments with a Staib electron gun operating at energies of kev, a high sensitivity CCD camera and EE2000 software developed by SPECS Scientific Instruments Inc Mass spectrometry Experiments performed with an in-situ quadrupole mass spectrometer are also discussed in this thesis. These experiments were performed with a Stanford Research Systems Residual Gas Analyzer (RGA) 300, capable of detecting gas species with atomic mass up to 300 amu with less than 1 amu resolution. The RGA is typically installed on the MBE growth chamber to analyze the composition of residual gases before growth and aid in leak detection. For the in-situ experiments, the RGA was placed in a source port on the MBE, with line-of-sight to the sample surface during growth. The RGA has a channel electron multiplier that allows detection of partial pressures down to the Torr range Elastic light scattering The growth of some samples, particularly those discussed in Chapter 5, were monitored using in-situ elastic light scattering (LS) with a 488 nm Ar ion laser as a light source. The Ar laser was typically frequency doubled to achieve UV incident light at a wavelength of 244 nm, with incident intensities in the mw range. This

28 Chapter 2. Experimental Methods 13 technique allows real time measurements of the surface morphology evolution, as the scattered light intensity is directly proportional to the power spectrum of the surface topography, i.e. the surface roughness. Most of the light incident on the mirror-like semiconductor surface is reflected specularly, but as the sample develops roughness a larger fraction of the light is scattered in non-specular directions, allowing us to examine surface evolution as a function of growth conditions. More details on this technique as applied to monitoring MBE growth may be found in the paper by Pinnington et al. [PLMT99]. In our set-up, light was incident through a standard optical viewport at an angle of either 25 or 55 with respect to the sample normal, depending on the period of growth. Scattered light was detected by a photomultiplier tube located at another viewport 73 out of the plane of incidence, and a lock-in amplifier was used to reduce ambient noise. The set-up is shown schematically in Figure 2.2. The measured spatial frequency q that the configuration is sensitive to depends on the incident and scattered angles, θ i and θ s, as well as the wavelength of light λ, according to q = 2π λ (sin θ i sin θ s ) (2.1) In particular, the UV light and backscattering geometry are chosen to maximize the sensitivity to small-scale surface features (large q). For the two geometries used in this study, the spatial frequencies are 17.5 µm 1 and 32 µm 1, corresponding to roughness on length scales of approximately 360 nm and 196 nm, respectively. 2.3 Ex-situ characterization High resolution x-ray diffraction High resolution X-Ray Diffraction (XRD) is a particularly useful structural characterization tool for thin epitaxial films as it can provide information about composition, thickness, and strain. Unless explicitly described otherwise, composition of films discussed here were determined by XRD. XRD was performed at UBC on a double axis Bede diffractometer with a Rigaku rotating anode x-ray source. The x-rays are produced using the Kα 1 transition of copper (λ= nm), and are monochromated with a Ge 220 channel cut crystal. The beam has a footprint of several mm 2 on the sample surface. Measurements were taken over a range of ± several thousand arcseconds after centering and optimizing the 004 reflection (the 004 peak has the highest intensity for the (001)-oriented zincblende lattice structure). Counting time varied

29 Chapter 2. Experimental Methods 14 Figure 2.2: Schematic of in-situ elastic light scattering experiment, courtesy of M.B. Whitwick. UV light is incident on the growing film, and detected at a non-specular angle with a photomultiplier tube. A rougher growth surface increases the intensity of scattered light. from 5-25 seconds/point depending on the required scan quality, and a large slit of 5 mm was used on the detector side. XRD simulations were performed using Bede RADS 4.0 software, which relies on the dynamical theory for x-ray diffraction. For a detailed description of dynamical theory, see, for example Warren s book [War90]. Analysis and modelling The composition of (001)-oriented, pseudomorphic, ternary films such as InGaAs, GaNAs, and GaAsBi can be determined very accurately by a single x-ray reflection from the (004) planes. Since the lattice parameter, and thus interplanar spacing, d, in the growth direction for the epitaxial layer is different than for the substrate, an x-ray rocking curve of intensity as a function of angle will show two peaks, one corresponding to the satisfied Bragg condition of the substrate and the other to the Bragg condition of the film. If the film is coherently strained, the relationship between the peak-splitting, θ, and difference in the interplanar spacing, d can be found by differentiating Bragg s law [Few93]. Bragg s law for constructive interference can be written nλ = 2d sin θ B (2.2)

30 Chapter 2. Experimental Methods 15 where n is an integer, λ is the wavelength of the incident photons and θ B is the Bragg angle of the substrate. Differentiating, we get nλ d = θ (2.3) 2 tan θ B sin θ B And finally, a combination of equations 2.2 and 2.3 gives an expression that is valid for small values of d; θ = d d tanθ B (2.4) where d can be defined as the misfit strain in the epilayer, ǫ. Once the misfit strain d is known, then Vegard s law [Veg21] gives the composition explicitly. Vegard s law proposes a linear variation of the lattice parameter with composition between the lattice parameter values corresponding to the end compounds. For GaNAs films, a peak splitting of 526 arcsec corresponds to a nitrogen concentration of 1% of the total group V composition [Ada02], while the addition of 1% Bi into GaAs is indicated by a peak splitting of 300 arcsec [TAT + 03]. To determine the composition of a pseudomorphic quaternary film, more information from another source is required. In addition to compositional information, it is also possible to determine layer thickness from high resolution x-ray diffraction, provided that the film has no structural defects such as dislocations and there is no lateral compositional gradient or interface roughness present. If these conditions are met, and the epilayer is of finite thickness, then it is typical to observe a series of decaying oscillations in intensity on either side of the Bragg peak (see, for example, Figure 4.4). These oscillations have been coined pendellösung fringes from the German word for pendulum solution, and arise from an effect explained by dynamical theory whereby interference occurs due to an exchange in intensity between the forward travelling (incident) x-ray beam and the diffracted beam, up to a depth of 2πξ g in the crystal. ξ g is the extinction distance of the beam in the sample. If the thickness t of the epilayer is less than the attenuation length of x-rays in the crystal, fringes should be visible. After measuring the interference peak separation, θ p, the thickness can be determined through the following relation [BT98]: t = λ 2 θ p cosθ B (2.5)

31 Chapter 2. Experimental Methods Photoluminescence Luminescence in a semiconductor material results from a relaxation process in which electron-hole pairs are first created by incident light, carrier injection in pn junctions, or electron bombardment [GS01]. In this study, photoluminescence (PL) is studied, using laser light to excite electron-hole pairs. Ultimately these pairs recombine, emitting a photon, with the primary emission wavelength at room temperature typically, though not always, representing the bandgap transition. Structural defects such as impurities and dislocations provide sites for non-radiative recombination; thus PL intensity is an important indicator of material quality. Photoluminescence of quantum wells and bulk layers is measured with a SpectraPro 300i spectrograph with a liquid-nitrogen cooled InGaAs array detector. Samples are optically pumped at room temperature with a frequency doubled, diode-pumped YLF solid-state laser emitting green light at 523 nm. The pulse length is 20 ns Atomic force microscopy Surface morphology of grown films is probed using atomic force microscopy for realspace measurements. Samples are scanned as soon as possible after removal from UHV using a Digital Instruments Multimode Scanning Probe Microscope in tapping mode with a Si tip of 30 nm radius. Scan sizes may range from 1 x 1 µm 2 to 100 x 100 µm 2 with atomic resolution on the z-axis and lateral resolution of approximately 30 nm Transmission electron microscopy Transmission microscopy for this study was carried out in collaboration with the Kavanagh group at Simon Fraser University using a Hitachi-8000 TEM with a LaB 6 filament with emission at 200 kev. Thin, plan-view samples were prepared by mechanically polishing the substrates to µm thickness, followed by etching the thinned substrate surface in bromine-methanol solutions ( %) to produce holes. The thin, wedge-shaped region of material near the hole in each sample was then examined in the TEM under a range of diffraction conditions including bright field (BF), dark field (DF), and weak beam dark field (WBDF), described in more detail in Chapter 3.

32 Chapter 3. Growth and properties of dilute nitrides 17 Chapter 3 Growth and properties of dilute nitrides The anomalous properties of dilute nitride semiconductors have precipitated a debate on whether nitrogen is better described as an alloying element or impurity in GaAs. With this frame of reference, a discussion of the electronic and structural properties of dilute nitride alloys is given in this chapter. In particular, detailed results from a study of strain relaxation and associated defects in these alloys are presented. In many ways, dilute nitride semiconductors are not dissimilar to other III-V alloys, where Groups III and V elements combine in equiatomic proportions to form compound semiconductors that are both isovalent and structurally similar with Group IV elemental semiconductors such as Si and Ge. As with the Group IV semiconductors, III-V bonding is predominantly covalent, although the metallic character of the bond increases for elements further down the columns of the periodic table. Tetrahedral bond symmetry is a consequence of sp 3 hybridization and as a result III-V compounds are prone to the sphalerite, or zincblende lattice structure, similar to the diamond cubic structure of Si and Ge. This crystal structure may be regarded as face-centered cubic (fcc) with two atoms occupying each lattice site: the Group III atoms at (0,0,0) positions and the Group V atoms displaced by ( 1, 1, 1 ). Although the equilibrium structure of the compound GaN is hexagonal wurtzite, the dilute nitride alloy GaN x As 1 x up to at least x = 0.1 maintains the zincblende structure with nitrogen replacing arsenic substitutionally in the lattice. Many properties of ternary and quaternary III-V semiconductors can be approximated quite well by applying Vegard s law [Veg21], which proposes a linear variation of the property of interest with composition between the values for the end compounds. Lattice parameter, elastic constants and energy bandgap are examples where this law may be successfully applied. For example, in the case of the lattice parameter of the nitrogen-containing compounds GaN x As 1 x and In x Ga 1 x N y As 1 y, the great majority of papers both theoretical and experimental assume that the lattice constant of the alloy is a linear combination involving the lattice constanst of

33 Chapter 3. Growth and properties of dilute nitrides 18 GaAs and cubic GaN [SDMC06]. Thus the lattice parameter a InGaNAs would be a InGaNAs = (1 x)[(1 y)a GaAs + ya GaN ] + x[(1 y)a InAs + ya InN ] (3.1) This assumption has been shown to be valid for GaN x As 1 x films grown by organometallic vapour phase epitaxy (OMVPE) for nitrogen concentrations up to x 3% [BMD + 04], a range encompassing all of the samples discussed in this thesis. Outside this composition range, negative deviation from Vegard s law has been attributed to a strongly anisotropic N-N interaction leading to preferential orientation of substitutional N along certain crystallographic directions, while positive deviation is explained by the presence of N interstitials [SDMC06]. The assumption of Vegard s law also breaks down for other properties of the dilute nitrides. The most significant finding on the first synthesis of GaNAs by MBE in 1992 [WSA92] was the highly non-linear redshift of the band gap with increasing nitrogen content, as shown in Figure 1.1. This was a surprising result considering that the band gap of pure GaN is actually greater than that of GaAs. The negative band gap bowing for GaNAs means that a linear interpolation is not sufficient to predict E g as a function of nitrogen content x. For the dilute nitrides, E g is better described as E GaNAs g = xe GaN g + (1 + x)e GaAs g bx(1 x) (3.2) where b is the bowing parameter as measured experimentally. Various estimates of b as a function of composition exist, typically taken from experimental data [TFS02]. One explanation for the origin of the GaNAs bandgap bowing is a two level anti-crossing model for the conduction band of GaNAs, though there is a considerable amount of controversy on this topic in the literature [MSY + 03]. The band anti-crossing model (BAC) [SWA + 99] is based on the assumption that the conduction band states of the alloy are formed due to an anticrossing interaction between the states of the GaAs bulk band edge and a nearly localized nitrogen 2s state at slightly higher energy created by substitutional nitrogen atoms on As sites. There have been other approaches to the band structure of the dilute nitrides, for example the tight binding approach taken by Nodwell et al. using Linear Combination of Atomic Orbitals (LCAO) to approximate the electronic wavefunctions in the solid [NAB + 04]. In Figure 3.1, the optical bandgap of a series of GaNAs films grown during the course of the current work is compared with the tight binding model from Nodwell et al., as well as to a fit to other experimental data compiled by Tisch,

34 Chapter 3. Growth and properties of dilute nitrides Tisch et al. fit to experiment Nodwell et al. LCAO model PL data from this study Bandgap (ev) Nitrogen Content, x (atom fraction) Figure 3.1: Bandgap dependence on nitrogen content of GaN x As 1 x films showing good agreement between photoluminescence data from samples grown in this study, a fit to other experimental data [TFS02], and LCAO model [NAB + 04]. [TFS02], showing good agreement. Predicting the bandgap of quaternary nitrogen containing alloys as a function of composition is more complicated, though Webster proposed a simple one-parameter fitting procedure that involves a summation of the contributions to bandgap reduction from the individual alloying elements [TWY + 05]. This method works extremely well for InGaNAs alloys, and also demonstrates that the effect on the bandgap of co-alloying In and N is less than the sum of effects of the two elements separately. 3.1 Electronic defects and dilute nitrides As was briefly introduced in Chapter 1, the presence of nitrogen in GaAs seems to introduce intrinsic defects that reduce photoluminescence (PL) efficiency. The reduction in PL due to nitrogen is a concern for the application of nitrogen containing alloys. InGaNAs quantum wells (QWs) are of interest to achieve light emission at specific wavelengths (1.3 and 1.55 µm) that correspond to optimum light transmission in fiber optic cables. In order to achieve these wavelengths, the maximum In and N possible must be incorporated into the QWs without strain relaxation, which will be discussed in the following section. Figure 3.2 shows room temperature photoluminescence for a series of InGaAs and InGaNAs QWs of varying thickness grown during

35 Chapter 3. Growth and properties of dilute nitrides r nm InGaAs QW 25% In r nm InGaNAs QW 28% In, 1% N r nm InGaNAs QW 28% In, 1% N r nm InGaNAs QW 30% In, 1.8% N r nm InGaNAs:Bi QW 30% In, 1.8% N r nm InGaNAs:Bi QW 32% In, 1.8%N RT PL intensity (au) Wavelength (nm) Figure 3.2: Room temperature photoluminescence of InGaNAs quantum wells with increasing nitrogen and indium concentrations. Legend lists composition and thickness estimates corresponding to PL of quantum wells from left to right (increasing emission wavelength). All samples were grown at temperatures between 450 and 475 C.

36 Chapter 3. Growth and properties of dilute nitrides 21 the course of this thesis project. All samples were grown at temperatures between 450 and 475 C. The red shift in emission wavelength with increasing N and In is evident; however the red shift is accompanied by a decrease in peak PL intensity by an order of magnitude. The PL intensity can be improved by annealing [DCN + 05, GAR + 01], but does not recover to the same intensities associated with N-free InGaAs. One theory for the degradation in dilute nitride alloys is that of cluster formation. Several researchers [FNJ + 99, KZ99] have observed mid band gap emission states that correspond to theoretical predictions for emission from so-called nitrogen clusters, and excitons bound to individual nitrogen pairs have been observed with high resolution photoluminescence spectroscopy [KMA + 06]. It is thought that N tends to cluster to minimize the strain energy due to its substitution for the much larger As atom, the same reason that GaNAs alloys tend to phase separate [SMP + 99], though random distributions of N may be sufficient to explain experimental observations. A nitrogen cluster is defined as a minimum of two interacting nitrogen atoms in next nearest neighbour configurations. The N clusters are labelled NN i, where i is proportional to the separation of the N atoms, d i = i/2a. At low N concentrations, the NN i clusters form bound states in the forbidden gap just below the conduction band, due to the strong electronic perturbation of the highly electronegative N. This was shown schematically for the NN 2 state in Figure 1.3. A larger nitrogen cluster with more atoms would be expected to introduce a much deeper mid-bandgap level. As the conduction band moves downwards with higher N concentration, it eventually overtakes some of the more weakly bound cluster states, so they overlap with the conduction band. This results in a transition from N doping in the dilute limit to a nitride alloy. Cluster states are thought to result in reduced efficiency for radiative recombination due to carrier recombination through the mid-bandgap states [Bea03]. The individual substitutional N atoms also strongly perturb the bottom of the conduction band and may be responsible for the decline in electronic mobility that is observed for increasing N content [Str01, MVD + 03]. 3.2 Structural defects and dilute nitrides: the strain problem The epitaxial growth process, first predicted in 1949 by Frank and van der Merwe [FvdM49], is one in which the structural properties of a single crystal substrate (lattice parameter, orientation, crystallography) are imposed on a film that is deposited on

37 Chapter 3. Growth and properties of dilute nitrides 22 the substrate with a high degree of crystalline perfection. It follows that unless the epilayer has the same composition as the substrate, it will be strained in-plane according to the mismatch between its natural lattice parameter and that of the substrate. The lattice mismatch f is equal to the misfit strain ǫ (also known as the parallel strain ǫ ). The misfit strain is isotropic for cubic systems and is defined as f = ǫ = a f a s a s (3.3) where a f is the lattice parameter of the film and a s is the lattice parameter of the substrate. For a f < a s the film is under tensile stress, and for a f > a s the stress is compressive. While elasticity in cubic systems is not isotropic, in practice it is common to use engineering constants such as the Young s modulus, E, the shear modulus, G, and Poisson s ratio ν for non-rigorous treatments 1. Elasticity theory (Hooke s Law) gives the biaxial stress,σ, generated by the misfit strain as the following expression: 1 + ν σ = 2Gǫ (3.4) 1 ν Because of the presence of a free surface, the lattice may expand out of plane without constraint, i.e. the stress in this direction is zero. This tetragonal distortion roughly preserves the unit cell volume, though this is not strictly necessary in elasticity theory. Due to the Poisson s ratio there is an out-of plane strain, ǫ, which is related to the misfit, ǫ, by ǫ = 2ν 1 ν ǫ (3.5) The strain energy stored in a mismatched layer is proportional to its volume. Thus as the strained layer grows, the stored energy may become so large that processes such as forming misfit dislocations, roughening the surface, or initiating cracks are energetically favoured over maintaining coherency Relaxation by misfit dislocations The formation of misfit dislocations seems a natural solution to the problem of coherency strain. Such dislocations have Burgers vectors, b, with components in the 1 The parameters used to characterize the elastic properties of a solid for a given orientation are simply related to the elastic constants C ij of the corresponding single crystal are as follows: E = (c11 c12)(c11+2c12) c 11+c 12, G = c11 c12 2, and ν = c12 c 11+c 12.

38 Chapter 3. Growth and properties of dilute nitrides 23 strain direction that accommodate the lattice mismatch at the interface between substrate and film. The amount of strain δu relieved by a single misfit dislocation of Burgers vector b is the projection of b on the interfacial plane normal to the dislocation line: δu = b cosφ (3.6) where φ is the angle between b and the direction of strain relaxation. The total strain energy that would be released per unit length, l of dislocation is E s = σ δu h (3.7) l where σ is the stress defined above and h is the layer thickness. However, the strain energy released is balanced by the energy necessary to create a dislocation in a perfect crystal. The concept of energy minimization is applied to this problem by Matthews and Blakeslee in their classic paper [MB74]. The energy associated with a unit length of dislocation in an elastically isotropic medium is E d l = Gb2 4π [ 1 ν cos 2 θ 1 ν ] ln αr b (3.8) where θ is the angle between b and the dislocation line direction (θ = 0 for screw dislocations and 90 for edge dislocations), α is a value representing the contributions of core energy (α = 4 for most semiconductors), and R is the range of elastic field of a dislocation, equal to the shorter of half the distance to the nearest dislocation or the distance to the nearest free surface. Different values of α tend to make little difference in the numerical results unless the ratio R/b is small. Then the strain energy/plastic relaxation balance for thermodynamic equilibrium is as follows; E s = N E d (3.9) where N = 1 for an uncapped layer and 2 for an embedded layer, since dislocations must then form at two interfaces [Hwa95]. Equating the two and simplifying for an uncapped layer, the critical thickness for plastic relaxation is: ( ) ( ) ( b 1 ν cos 2 θ h c = ln αh ) c 8πf cosφ 1 + ν b (3.10) The balance between dislocation formation energy and strain energy that results in the Matthews equation has been derived numerous times over the last 30 years

39 Chapter 3. Growth and properties of dilute nitrides 24 resulting in many slight variations; however the basic tennets remain the same. Theoretically, anytime this critical thickness is reached or surpassed, misfit dislocations should form since there would be a reduction in the overall energy of the system. In practice however, it is unusual for a material to fully undergo plastic relaxation at the Matthews-Blakeslee limit due to difficulties associated with nucleation of dislocations and glide to the appropriate interfacial position. These energetic barriers may typically be much larger than the thermal energy (kt) and thus the metastable range of pseudomorphic strain accommodation decreases with increasing mismatch and growth temperature and decreasing growth rate. Any relaxation that does occur at the equilibrium critical thickness typically involves the motion of already present threading dislocations from the substrate, though current substrates are typically of such high quality that there are insufficient threading dislocations ( 10 4 /cm 2 [Dun97]) for substantial relaxation. Indeed, many authors speak of a second critical thickness where nucleation of new dislocations occurs [DT87]. Dislocations in GaAs In general, dislocations in zincblende semiconductors have Burgers vectors parallel to close-packed 110 directions and confined to atomically dense {111} glide planes, in accordance with conventional dislocation theory [HL82]. As in other fcc systems, the activation energy for glide on the 110 {111} slip system is minimized due to the wide {111} spacing, as the Peierl s stress varies inversely with interplanar spacing. Typical dislocations may be of screw, edge or mixed character with Burgers vectors of the type b = a/ The common mixed screw-edge dislocation has a Burgers vector 60 to the 110 line direction as shown in Figure 3.3 a). These dislocations tend to dissociate into Shockley partial dislocations separated by a stacking fault [Jon00]. The dissociation reaction is a 2 [1 10] a 6 [1 21] + a 6 [2 1 1] (3.11) The actual situation is even more complex due to the two different fcc sublattices that combine to form the zincblende structure, leading to two possible types of partial dislocations; those with Ga cores (also known as β dislocations), and those with As cores (α dislocations). Ga-core dislocations lie parallel to the [110] direction while As-core dislocations lie parallel to [1 10]. Screw dislocations may dissociate into two partials, one of Ga (β) type and the other As (α). The existence of two dislocation core chemistries in GaAs is a complicating fac-

40 Chapter 3. Growth and properties of dilute nitrides 25 [100] [100] interface interface (111) [001] (011) [001] [010] b [010] b Growth direction [001] Dislocation line along <110> Slip on {111} planes b = a/2 <110> 60 to the interface Dislocation line along <100> Slip on {110} planes b = a/2 <110> 45 to the interface Figure 3.3: Possible misfit dislocation geometries and associated slip systems in GaAs a) typical 110 dislocation with b oriented 60 to the interface and line direction and b) unusual 100 dislocation with b oriented 45 to the interface and 90 to the line direction. The [001] growth direction and the (001) interfacial plane are also indicated.

41 Chapter 3. Growth and properties of dilute nitrides 26 tor for the case of strain relaxation since the two defect types will exhibit different properties. Rates of nucleation and resistance to glide are different for Ga-core and As-core dislocations, which leads to anisotropy in strain relaxation in the typical 110 in-plane directions [KCH + 88]. Experimentally it has been found that α dislocations have a higher glide rate than β dislocations and are often reported to form first during strain relaxation of InGaAs/GaAs interfaces, although some authors report higher densities of β dislocations [GKW + 98]. Although the vast majority of misfit dislocations in III-V semiconductors are reported to lie along the 110 direction with slip on {111} planes, there are a small number of reports of an unusual secondary slip system [BHWM92, ASHB93, ZCCS94, ZJF + 96, Fri01] usually in cases with large lattice mismatch, for example In x Ga 1 x As/GaAs with x > In these cases the misfits are aligned along 100 line directions, with {110} slip planes, and Burgers vectors b = a/2 110 which are 45 to the [001] growth normal, as shown in Figure 3.3 b). These misfits are pure edge dislocations, as determined by the standard TEM invisibility conditions i.e. g b= 0 and g (b x u)= 0, and actually relieve more strain than 110 oriented dislocations by a factor of 2. (The unit vector u is the dislocation line direction, and g is the diffraction vector.) Obviously this is energetically favourable from a strain relief point of view, and in fact the Schmid factor that resolves the applied stress onto specific slip systems is also higher for the 110 {110} slip system than the 110 {111} slip system. However the Peierl s stress, or activation energy for glide along the {110} planes is prohibitively high due to the relatively smaller interplanar spacing. It is suggested that this slip system may be activated in response to the large stresses in the highly mismatched systems, and glide is thought to have been observed in-situ in the TEM [BHWM92] although there is also evidence that the 001 dislocations can form by climb at lower mismatch if there is a sufficient vacancy population [ZCCS94]. The 001 oriented dislocations have been observed both alone and in addition to arrays of 110 oriented dislocations in the same samples [ZJF + 96]. It is an open question why 001 oriented dislocations form in the case of some highly strained systems and not in others Relaxation by surface roughening Surface roughening represents another mechanism by which a system under strain from lattice mismatch may relax, in some cases completely separated from the formation of misfit dislocations, in other cases complementary [GN99]. Because of the

42 Chapter 3. Growth and properties of dilute nitrides 27 high temperature growth environment necessary for most types of thin film growth, and the small dimensions involved, diffusive mass transport is very fast and allows the surface morphology to change continuously in response to changes in the dominant thermodynamic forces in the system, namely elastic strain energy and surface energy of the film. The problem of surface roughening can be examined by comparing the stability of a flat film to a film with a wavy perturbation described by a cosine function. Studies [Sro89] indicate that there exists a critical wavelength, λ c, given by 1 ν λ c = πγ s (3.12) E(1 + ν)ǫ 2 where γ s is the surface energy and E is Young s Modulus. If λ < λ c, the chemical potential is higher at a surface peak than a surface trough, so diffusive processes will tend to smooth the rough surface, whereas if λ > λ c the opposite is true and atoms will be transported from troughs to peaks. The intrinsic length scale of the roughening is given by the ratio of the surface energy density, γ s, to the strain energy density, Eǫ 2, and the tendency to roughen will be dominated by this ratio. Several semiconductor systems have been observed to relax by this method, including Si 1 x Ge x /Si alloys, as well as InGaAs/GaAs grown at high temperature (580 C) [CPE96]. GaNAs/GaAs films also exhibit a morphological growth instability [ATR + 01] that involves a {111} faceting transition favoured at high temperatures and nitrogen concentrations. The process of surface roughening often results in surface morphologies with cusped valleys and rounded peaks, and a thorough explanation of the energetics of this surface may be found in Gao and Nix [GN99]. These authors point out that a significant stress concentration exists at the cusped valleys that lower the activation energy for defect formation. Diffusing atoms are driven away from the highly stressed regions, and at a critical point a dislocation loop can be nucleated from vacancies with little or no activation barrier, then expand to create misfit dislocations at the substrate/film interface. This mechanism has been observed in both InGaAs/GaAs and SiGe/Si, systems that tend to surface roughen as mentioned above, and is thought to result from the formation of a sessile 90 Lomer-Cottrell dislocation with b = a/2 110 due to collapse of a double ledge at the surface followed by diffusion of atoms to the site, trapping the dislocation into place[gn99]. For the case of the GaNAs/GaAs system, it has recently been observed that there is an interplay between the surface morphology and the mechanisms of strain relaxation [LWN05, WBGB05]. Wu et al. [WBGB05] observe that relaxation in GaNAs films occurs by morphological changes involving surface cusps, stacking faults

43 Chapter 3. Growth and properties of dilute nitrides 28 and twins, and that 90 dislocations nucleate at surface cusps. Li et al. [LWN05] also find a correlation between the rough surfaces resulting from the surface morphological instablity at high N concentrations and the presence of dislocations and twins, while for films with smoother surfaces and lower N concentrations, the preferred mechanism of relaxation is crack formation (see section 3.2.3) Relaxation by crack formation The observation of cracks in semiconductor films in response to strain was first reported in 1972 by Matthews and Klokholm [MK72], who obtained the critical thickness for their formation as h c = γ s(1 ν) 2 πg(1 + ν)f (3.13) where γ s is the surface energy. This critical thickness was formulated using the Griffith criterion, which is to say that a balance must be struck between the rate of decrease in potential energy, in this case due to stored elastic energy, and the resultant rate of increase in surface energy due to the crack [Her96]. This condition can be written G = 2γ s. Misfit strain-related cracks only appear where the film is under tensile stress. While there have been few in-depth studies of cracking in semiconductors, there are reports of cracks in tensile systems such as InGaAs/InP [WW99], GaNP/GaP [LWT + 96], AlGaN/GaN [HHL + 00], and GaNAs/GaAs [ATR + 01, LWN05]. One particularly interesting result is that for the case of GaNAs/GaAs and InGaAs/InP the interfacial cracks observed actually change direction after penetration into the substrate from {110} plane to the {111} planes Relaxation by phase separation While phase separation of the dilute nitrides has not been investigated explicitly in this study, it is a significant concern for the GaNAs and InGaNAs alloy systems due to the miscibility gap between GaN and GaAs [SMP + 99]. The high strain energy associated with the small size of the nitrogen atom increases the internal energy of the alloy, providing a driving force for decomposition. Asomoza et al. [AEPS02] calculate that InGaNAs alloys are deeply inside the spinodal decomposition range at their growth temperatures, and several groups observe In-rich and In-poor inhomogeneous regions laterally along quantum wells [GAR + 01, DCN + 05].

44 Chapter 3. Growth and properties of dilute nitrides Strain relaxation study of quantum wells In previous work by Adamcyk et al. [AST + 02], it was shown using in-situ stress monitoring during growth that the rate of interfacial strain relaxation in single epilayers of low-mismatch, In x Ga 1 x As 0.99 N 0.01 /GaAs(001) heterointerfaces is slower compared to InGaAs/GaAs, given an equivalent mismatch. In other words, the degree of strain relaxation via the usual nucleation and glide of 60 misfit dislocations was 20% less for In x Ga 1 x As 0.99 N 0.01 compared to InGaAs, for an equivalent lattice mismatch of 0.5% (x = 0.08 (InGaAs), compared to x = 0.12 (InGaAsN)). The misfit dislocation line densities were smaller consistent with a lower strain relaxation rate, but a higher number of dislocation threads was observed. It was concluded that the N was acting to reduce the dislocation glide velocity, but this was partially compensated by a higher nucleation rate. In this work, we investigated the effect of N on the relaxation of more highly strained InGaAsN single quantum wells. Higher mismatch (i.e. higher In content) films are of greater technological interest as they have emission wavelengths closer to 1.3 and 1.55 µm as needed for fiber optic communication. InGaAs and InGaAsN QWs with equivalent lattice mismatch were compared. This is accomplished using the observation from Equation 3.1 that when x = 3y, In x Ga 1 x N y As 1 y is lattice matched to GaAs. Therefore 1% N alloying can be compensated for in terms of strain by an additional 3% In alloying. It was expected that N would both act to reduce the rate of relaxation via decreased dislocation nucleation and higher resistance to glide. In this study, several series of quantum well samples varying in thickness up to 24 nm were grown by MBE according to procedures described in section 2.1 at a growth temperature of 450 C. Each quantum well was capped with a layer of GaAs nominally 250 nm thick. Concentrations were calibrated by measuring N concentrations in ternary GaNAs films from peak splitting in the high-resolution x-ray diffraction spectra (see Chapter 2.3.1) as were the In concentrations in InGaAs films. In In- GaNAs films the nitrogen and indium concentrations were assumed to be the same as their relative concentrations in ternary InGaAs and GaNAs films grown at identical growth conditions. Using this method, the two types of QWs in this study were determined to have the following compositions at the same lattice mismatch (1.7%): In 0.28 Ga 0.72 As 0.99 N 0.01 or In 0.25 Ga 0.75 As. These are the compositions of QWs referred to in the remainder of this chapter. A Bi flux of 1 x 10 7 was applied during growth of some of the quantum wells. Due to its large atomic radius, Bi tends to surface segregate and does not incorporate under the growth conditions described here, even

45 Chapter 3. Growth and properties of dilute nitrides 30 at the ppm level [TAY + 02]. The Bi surfactant tends to affect surface morphology during the growth, as will be discussed in detail in Chapter 4, and thus might be expected to play a role in strain relaxation. X-ray diffraction results As expected, x-ray diffraction studies showed a clear dependence of the material strain state on the QW thickness. X-ray diffraction spectra obtained from a series of samples of increasing QW thickness, 7.5, 9.0, 10 or 12 nm, grown with and without N, and with and without a Bi surfactant, are shown in Figure 3.4. This data indicates that the crystalline perfection of the dilute nitride QW structures begins to degrade for QW thicknesses between 7.5 and 9 nm independent of the presence of Bi. By 10 nm the pendellösung fringes have completely disappeared, indicating that the uniformity of the capping layer lattice constant has become sufficiently distorted to annihilate constructive interference between the x-rays scattered from the sample surface and the top QW interface. Kinematical x-ray simulations predict the extinction of the pendellösung fringes when the amplitude of the surface roughness of the cap layer is 100 nm. From AFM measurements we know that the cap layer is much smoother than this; therefore, the disappearance of the fringes is presumed to be due to the presence of dislocations. The InGaAs QW samples show stronger fringes, but the fringes still begin to degrade between 8 and 12 nm and have completely disappeared by 12 nm. The critical thickness detected here is in good agreement with an estimate based on the Matthews-Blakeslee theory presented in section Analysis predicts that relaxation should occur at 7.5 nm for films at this mismatch (assuming 60 misfits and elastic constants determined using Vegard s law [GAR + 01]. Photoluminescence Room temperature PL spectra from the three InGaNAs QWs with x-ray diffraction patterns as shown in Figure 3.4 a), are shown in Figure 3.5. Consistent with the x-ray data, the PL peak intensity decreases by orders of magnitude for QW thickness between 7.5 and 9 nm, indicating the formation of defects that are sources for nonradiative recombination. These defects are almost certainly dangling bonds due to misfit dislocations. The redshift in emission wavelength can be attributed to the increasing well thickness.

46 Chapter 3. Growth and properties of dilute nitrides (004) Diffracted Intensity (a.u.) nm 9 nm 7.5 nm 10 nm 9 nm 7.5 nm 12 nm 8 nm 7 nm θ ( arcsec ) θ ( arcsec ) θ (arcsec) Figure 3.4: High-resolution x-ray diffraction patterns from GaAs/QW/GaAs heterostructures, lattice mismatch (1.7 %) where the QW is (a) InGaNAs (b) InGaNAs grown with a Bi surfactant, and (c) InGaAs. In the dilute nitride QWs the interface coherence begins to degrade between a 7.5 and 9 nm QW thickness. The use of a Bi surfactant (1 x 10 7 Torr) had no apparent effect on the critical thickness. The stronger fringes in the case of InGaAs indicate that the strain is more uniform compared to the dilute nitride samples, but the strain relaxation begins between 8 and 12 nm, similar to the other series. Data are offset for clarity.

47 Chapter 3. Growth and properties of dilute nitrides 32 RT PL intensity (arb. units) nm QW 9 nm QW (x5) 10 nm QW (x5) Wavelength (nm) Figure 3.5: Photoluminescence from the series of GaAs/InGaAsN/GaAs QWs (In-28% N-1%) at room temperature as a function of QW thickness. At the onset of relaxation between a QW thickness of 7.5 and 9 nm the PL peak intensity decreases by more than 2 orders of magnitude. The 10 nm QW has almost no detectable photoluminescence, consistent with the presence of numerous defects.

48 Chapter 3. Growth and properties of dilute nitrides 33 Transmission electron microscopy Planview TEM images of studied samples reveal details of the strain relaxation process in the QWs. TEM images were obtained by Koveshnikov and Kavanagh, with the participation of the author [YKT + 05]. Bright field (BF) and weak beam dark field (WBDF) TEM images were taken from representative InGaAs and InGaAsN QW samples for QW thicknesses of 9, 10, and 12 nm. Images were obtained with the samples tilted close to g=(220) diffraction conditions, where g is the diffraction vector in the plane of the sample (g = k, the difference in incident and reflected wavevectors). In the BF technique, an aperture is inserted around the main beam so that only the transmitted electrons form the image, i.e. contrast is primarily due to differences in the intensity of diffraction in the sample. Dynamical scattering effects can also influence the image when the sample is tilted to an exact Bragg angle (also called a strong beam condition). For a DF image, the electron beam is tilted such that a particular diffracted beam (hkl) is parallel to the optic axis. With the same aperture, bright regions in a DF image now indicate where strong diffraction has occurred, giving information that is complementary to the BF image. In WBDF, the beam tilt is the same magnitude but in the opposite direction to that of DF. This produces an image based on differences in the intensity of diffraction into a third order spot. This technique is useful for enhancing resolution for detecting strain variations in the sample, e.g. those due to dislocation cores. The relaxation process in both InGaAs and InGaAsN occurs via the formation of dislocation loops that are assumed to nucleate in the QW layer, as shown in Figure 3.6. These are first visible as coffee-bean contrast in the 9 nm InGaAsN samples from small loops with a density of 150 µm 2, as seen in Figure 3.6 a). When the QW is thicker and strain relaxation has progressed further, the loops seem to have elongated along both top and bottom QW interfaces forming extended loops aligned with both 100 directions. Figure 3.6 b) is a BF and WBDF image pair of a 10 nm InGaAsN, respectively. The BF images in a) and b) were obtained under strong g (220) conditions highlighting the strain variations generated by the dislocations and interfacial roughness. The WBDF image b) highlights diffraction specifically from dislocation lines. The BF and WBDF image pair in Figure 3.6 c) are also from a 12 nm InGaAs QW obtained away from a strong g = (220) condition. In the 12 nm sample it is clear that the majority of visible dislocations exist as pairs aligned along in-plane 100 directions. The short segments at the ends of each loop thread through the QW connecting the long segments located at the top and bottom QW

49 Chapter 3. Growth and properties of dilute nitrides 34 Figure 3.6: Plan-view TEM images of GaAs/QW/GaAs (001) samples obtained with a g = (220) diffraction condition in the direction indicated by the arrow. Figure a) is a strong g, bright field (BF) image of a InGaAsN QW (9 nm) showing the beginnings of dislocation loop formation; b) strong g, BF and weak-beam dark-field (WBDF) image pair from an InGaAsN QW (10 nm), c) BF and WBDF pair from a InGaAs QW (12 nm) and d) WBDF of a InGaAsN QW (12 nm). All dislocations are in contrast for (220) diffraction but depending on the deviation from perfect Bragg diffraction of the sample, the loop interiors are in contrast in b) WBDF and c) BF and out of contrast in all other images. Data courtesy of K. Kavanagh [YKT + 05].

50 Chapter 3. Growth and properties of dilute nitrides 35 interfaces. The average loop spacing d = 80 nm, or linear density D = 12 µm 1, are essentially the same in both 100 directions. Figure 3.6 d) is a WBDF image from a 12 nm InGaAsN QW sample indistinguishable from the InGaAs QW of Figure 3.6 c). Comparisons of large areas of the 12 nm InGaAsN and InGaAs QWs found an average linear density of misfit dislocations of 22 ± 3 and 24 ± 4 dislocations/mm, respectively. Surprisingly, analysis of images obtained under perpendicular g = (400) conditions confirm that the the segments of the loops that lie along the interface are pure edge dislocations, with their slip vectors, b = a/2 110, directed at a 45 angle out of the plane of the interface, 90 to the line direction. As discussed in section 3.2.1, these unusual dislocations have been observed previously in systems with large lattice mismatch, and they relieve more strain than the typical 60 dislocations. In this case the dislocation loop densities can be used to determine the degree of strain relaxation in each 100 direction and can be calculated from ǫ = b /d with b = a/2, the magnitude of the in-plane, perpendicular component of the dislocation Burgers vector. For an average loop spacing of 80 nm, ǫ = or approximately 25% of the lattice mismatch (0.017) has relaxed by a thickness of 12 nm. Dark contrast is also visible in the images in Figure 3.6 in the interior of some of the dislocation loops. This is due to diffraction contrast from the expansion associated with the relaxed QW material inside the loop. This distortion is visible or not depending on how the loop is oriented with respect to the Bragg reflection. In the images of 3.6 b) and c) dark contrast is seen inside some of the loops. Those loops that have this contrast show a narrower spacing between the dislocations of the loop meaning that the inside of the dislocation loop was diffracting. Those loops tilted the opposite direction have a larger dislocation loop spacing, since in those cases the outside of the dislocation is diffracting. The dislocation spacings are comparable to the QW thickness, consistent with the proposed loop structure inclined on {110} planes. Morphology from light scattering The three InGaNAs samples from Figures 3.4 a) and 3.5 were monitored during growth using the elastic light scattering technique at a spatial frequency of 17.5 µm 1, as described in section The light scattering (LS) signals corresponding to the three growths are shown in Figure 3.7. The time t = 0 indicates the start of quantum well growth, which took 36, 42, and 48 seconds for the 7.5, 8 and 9

51 Chapter 3. Growth and properties of dilute nitrides 36 nm quantum wells, respectively. The signal observed for each sample after the QW growth period is sensitive to the roughness of the 250 nm GaAs cap layer. As seen from x-ray and PL data, the 7.5 nm QW does not undergo relaxation and, as might be expected, there is no significant change in the LS signal after QW growth. The 9 nm sample, which showed dislocation loops in the TEM, Figure 3.6 a), displays a fast initial increase in the LS that saturates before decreasing with continued capping layer growth. The final average spacing of the dislocation loops from TEM was 80 nm, which is smaller than the LS spatial frequency. The initial increase in the LS signal is due to build up of surface roughness due to lateral migration of deposited atoms in response to inhomogeneous surface strain associated with misfit dislocations. As the layer becomes thicker the inhomogeneous strain at the growth surface drops and the surface tends to smooth leading to a reduction in the LS signal. The more complex surface strain associated with the higher density of dislocations for the 10 nm thick quantum well does not smooth as quickly as the lower dislocation density sample. Continued relaxation during growth of the capping layer may also play a role. Also, the TEM pictures for the 10 nm QW sample (Figure 3.6 b)) shows that the dislocations are more closely spaced (80 nm) than the optimal wavelength of the surface topography for light scattering, which leads to relatively low light scattering intensity. In Figure 3.8, elastic LS data is presented for growth of three bulk InGaAs and InGaAsN films. Two of the growths are InGaAs and InGaAsN films that have lattice mismatch (1.7%) identical to the QWs discussed previously. The growth of the InGaNAs layer is also repeated under identical conditions except for the application of a Bi flux of 1 x 10 6 Torr. An increase in LS signal for all three samples, assumed to correspond to strain relaxation, is observed after approximately 50 s of growth. From growth rate calibrations, this time corresponds to a thickness of approximately 10 nm, similar to the critical thickness observed in the QWs. Although the critical thickness observed here is the same for both the InGaAs and InGaNAs, it is clear that the nitrogen greatly enhances the surface roughening process. The slopes of the LS intensity vs. time for each growth at the onset of roughening are indicated in Figure 3.8, and the rate of roughening is greater by an order of magnitude for the nitrogen containing film (1.03 x 10 3 /s for InGaNAs vs 1.4 x 10 4 /s for InGaAs). From the TEM of QWs, we know that the dislocation density should be similar for InGaAs and InGaNAs films at this lattice mismatch, which suggests that the increased roughening observed here for InGaNAs is not solely due to dislocations. It may be due to the N-related instability observed by others [ATR + 01, WBGB05, LWN05].

52 Chapter 3. Growth and properties of dilute nitrides 37 3 Cap Layer Growth Scattered Light Intensity (a.u.) 2 1 QW 9 nm 10 nm nm Time (min) Figure 3.7: Elastic light scattering signal recorded during growth of three, capped InGaNAs QWs at 450 C with thicknesses as indicated. The QW growth begins at Time = 0 for a duration of less than a minute followed by a 240 nm GaAs capping layer.

53 Chapter 3. Growth and properties of dilute nitrides 38 Scattered Light Intensity (a.u.) All layers with 1.7% lattice mismatch r1654 InGaAs r1655 InGaNAs r1656 InGaNAs:Bi 3.4 x 10-4 /s 1.03 x 10-3 /s x 10-4 /s Time (s) Figure 3.8: Elastic light scattering signal recorded during growth at 450 C of three films with 1.7% mismatch past their critical thickness. The top signal corresponds to growth of a 175 nm InGaNAs layer, the middle signal corresponds to growth of a 220 nm InGaNAs layer with a Bi flux of 1 x 10 6 Torr and the bottom layer corresponds to growth of a 220 nm InGaAs layer. The Bi surfactant does not affect the critical thickness but reduces the initial rate of relaxation in the InGaNAs layer. Roughening rates at the onset of relaxation are indicated for each growth. The InGaNAs film grown with a Bi flux has a LS signal corresponding to the middle data set in Figure 3.8. The initial rate of roughening for this sample (3.4 x 10 4 /s) is three times greater than that of the InGaAs film, but still an order of magnitude lower than the same sample grown without Bi, consistent with the model of Bi as a surfactant that reduces surface roughness. 3.3 Discussion In a previous study, 1% N alloying was found to reduce the strain relaxation rate by 20% in In 0.08 Ga 0.92 As layers grown on GaAs and the relaxation occurred through the 110 {111} slip system [AST + 02]. In the present experiments, no change in critical thickness was observed with N alloying, and the dislocations are edge dislocations lying along 100 directions with a slip system 110 {110}. The presence of these atypical dislocations is attributed to the higher lattice mismatch for films in this

54 Chapter 3. Growth and properties of dilute nitrides 39 study. The reason for the difference due to N is not known, though several factors may play a role. Such factors include the higher In content in the present experiments (25-28% In compared with 8-12% in the previous work), the different types of dislocations responsible for strain relaxation in the two experiments and the fact that the growth rate was not kept fixed in the earlier experiments. In the earlier experiments [AST + 02] the In flux was fixed and the Ga flux was adjusted to obtain the desired alloy composition. This means that the growth rate of the N containing sample with 12% In was 50% lower than the growth rate for the InGaAs sample with 8% In but the same strain and no incorporated nitrogen. In the present experiments the growth rates of the buried quantum wells were kept fixed. In-situ strain measurements show that the relaxation rate is higher during growth than during growth interruptions [LCBH04], and in the same work it was suggested that the surface flux creates dislocation kinks at the surface that allow a more rapid glide of the threading dislocations required for bulk epilayer relaxation. It is possible that the lower strain relaxation rate previously observed in the N alloys was due to the lower growth rate. The use of a non-incorporating Bi surfactant during growth of InGaNAs QWs was found to have no effect on the critical thickness and dislocation density. This is surprising if point defects or dislocation kink generation at the surface are important to the relaxation process. If dislocations nucleate at the surface one might expect the surfactant, which reduces surface roughness and changes the surface adatom diffusivity, to affect the dislocation density. However, LS data indicated that the surface roughening associated with relaxation was reduced by an order of magnitude when the Bi flux was present during growth. This suggests that Bi reduces the extent of relaxation, contradictory to the other observations. It is possible that Bi may be able to suppress some of the interplay between morphology and nucleation of dislocations that was observed in previous studies of strain relaxation involving nitrogen, but more work is needed to assess this possibility. Surfactant effects will be discussed further in Chapter Summary A high degree of bowing as a function of N content is observed for the band gap of GaNAs, consistent with previous experimental results and theoretical predictions of the band gap. A degradation in photoluminescence efficiency is observed in QWs with increasing nitrogen concentrations.

55 Chapter 3. Growth and properties of dilute nitrides 40 The critical thickness for plastic relaxation is the same for InGaAs and InGaNAs QWs with the same lattice mismatch (1.7%). 1% N alloying did not have an effect on the critical thickness or dislocation density in relaxed films. The loss of coherency in these QWS is determined through observation of the loss of pendellosung fringes in high resolution x-ray diffraction patterns. Photoluminescence data and TEM micrographs confirm the presence of defects in QWs that had been assessed to be relaxed with XRD. TEM analysis reveals that the defects are edge-type misfit dislocations with line directions along 100, an unusual occurrence in FCC materials. The high lattice mismatch may provide the driving force for glide along the energetically unfavourable {110} planes required for these dislocation to move to the interface, since the 100 dislocations relieve more strain than the typical misfit dislocations oriented along 110. The use of a non-incorporating Bi surfactant during growth did not affect the critical thickness of InGaNAs QWs or bulk layers, but did reduce the rate of roughening associated with relaxation as measured by in-situ light scattering measurements. 3.5 Conclusion In conclusion, it is clear that while band gap engineering of dilute nitride semiconductors is successful, this materials system is prone to many types of defects that affect carrier transport and luminescence efficiency. The majority of the possible defects, including nitrogen clusters, rough surfaces, dislocations and cracks, are related to the strain energy associated with N when it replaces As in the lattice. Co-alloying with In is useful to some degree, but the degree of alloying necessary in order to achieve the desired shift in band gap pushes the boundaries for coherency in these structures. Phase separation and plastic relaxation must be carefully considered, and then strategically avoided to produce films with high electronic quality.

56 Chapter 4. Bismuth surfactant growth of the dilute nitrides 41 Chapter 4 Bismuth surfactant growth of the dilute nitrides In 1991, Copel and Tromp posed the following question in a journal article: Are bare surfaces detrimental in epitaxial growth? [CT91]. This inquiry was a natural progression from a 1989 discovery by the same authors that provided a breakthrough in terms of the controlled growth of semiconductors. They found that the use of a single monolayer of As actually changed the growth mode of Ge on Si from island growth to layer-by-layer growth [MRKT89], providing a new way to achieve high quality films in spite of thermodynamics. Fifteen years later, there may not be a definitive answer to this interesting question, but there is more and more evidence that the use of an appropriate surfactant is at least not detrimental, and at best can improve many of the properties desired of epitaxially grown films. In this chapter, results on the use of a Bi surfactant as a means of improving the growth of dilute nitrides will be presented and discussed, as well as a novel method of characterizing the Bi surface layer using RHEED. 4.1 Surfactant-assisted epitaxy The term surfactant is borrowed from chemistry where it is most commonly used to describe substances that modify the surface tension of liquids. In epitaxial semiconductor thin film growth where the crystal growth front is an atomically clean surface under UHV, the surface energy is similarly modified by an adsorbate; surfactant atoms such as hydrogen saturate the dangling bonds and chemically passivate the surface. Alternatively, a surfactant can be an element with a much larger atomic radius than the semiconductor of interest that surface segregates and tends to float on the surface rather than become buried under the growing film [KK95]. This was illustrated schematically in Figure 1.2. Since the interface energy between the surfactant and the semiconductor surface is likely to be significantly lower than the energy of the semiconductor-vacuum interface, it might be expected that a surfactant could have a

57 Chapter 4. Bismuth surfactant growth of the dilute nitrides 42 profound effect on surface processes during growth, both in terms of thermodynamics and kinetics. From a thermodynamic point of view, the surface free energy of a growing film, σ f, together with the strain energy determines the epitaxial growth mode. If thermodynamic equilibrium is assumed, the growth morphology should be determined by a balance of the different surface free energies involved: γ tot = γ f + γ i γ s (4.1) where the subscripts f, i, and s stand for film, interface and substrate respectively. If the total energy γ tot is less than 0, this means that energy of forming a substrate/film interface and a film/vacuum interface is energetically favourable to having a bare substrate surface, similar to the condition for wetting in liquids. In this case the growth will follow the ideal 2D layer-by-layer growth mode predicted by Frank and van der Merwe (FvM) [FvdM49]. However, in the case of heteroepitaxy, even if this condition is satisfied, the elastic strain energy due to lattice mismatch (discussed in section 3.2) must also be taken into account. If the mismatch is very large, this term is responsible for the 3D islanding transition that occurs after growth of a few monolayers in a layer-by-layer growth mode, also known as the Stranski-Krastanov, or SK growth mode. A good example of SK growth is that of InAs on GaAs, which forms islands that are also known as quantum dots. If the total energy γ tot is greater than 0, deposition will result in the immediate formation of 3D islands. The system will try to maintain an uncovered substrate surface, and minimize the area of contact between the substrate and film. This theory is due to Volmer and Weber and known as VW growth. For more information on these growth modes as pertaining to MBE, the reader is referred to, for example, Tsao s book [Tsa93]. An appropriate surfactant species necessarily segregates to the growth front to form a surface layer. Massies and Grandjean [MG93] propose that a non-reactive surfactant species bonds weakly with the semiconductor surface, forming a surface complex that reduces the bond strength between semiconductor species and thus reduces the energy barrier for adatom migration on the surface. This leads to increased surface diffusion length and tends to improve the quality of the growing crystal. This holds true for surfactant-assisted homoepitaxy where, for example, introduction of Pb as a surfactant during growth of GaAs on GaAs results in a 2D layer-by-layer growth mode at low growth temperatures that is normally observed only at high temperatures [MG93]. An increase in surface diffusion length has also been suggested for cases

58 Chapter 4. Bismuth surfactant growth of the dilute nitrides 43 in heteroepitaxy. For example the use of Bi or Sb as a surfactant during growth of InGaAs or InGaNAs alloys on GaAs results in surface smoothing and an improvement in photoluminescence emission intensity [PKHB00, TAY + 02, YBW + 05], though Sb incorporates to a small extent into the growing film. Both Sb and Bi are isoelectronic with As so are not expected to produce electronically active defects if they incorporate on a group V site, making them ideal surfactants for III-V alloys. Wixom et al. [WRS04] used patterned substrates to confirm that Sb and Bi surfactants enhance the incorporation at step edges during growth of GaAs by organometallic vapour phase epitaxy (OMVPE). Growth rates in the [110] direction were increased by close to 300%, while negligible effects were observed in [1 10] direction. In the same study, kinetic simulations were used to show that the effect may be due to a decreased barrier to hopping in the [110] direction. Other types of surfactants that are more reactive with the surface may in fact decrease the migration length of adatoms, and thus provide a means of controlling growth morphology by kinetically inhibiting certain growth modes, such as islanding. In this case the energy barrier for the exchange of an adatom with a surfactant atom must be comparable to the energy barrier to diffusion, leading to a small diffusion length and incorporation of adatoms near their point of impact [KK95]. One example is the extension of the pseudomorphic growth regime of InAs on GaAs from 1 monolayer to 6 monolayers with the use of Te as a surfactant [GME92], as well as the delaying of the critical thickness for the onset of plastic relaxation for In x Ga 1 x As [GMD + 93]. As mentioned at the beginning of the chapter, similar extension of the pseudomorphic range has been observed in the Si/Ge system using As to suppress islanding [MRKT89], and also using Sb and Bi surfactants [MRHT90]. 4.2 Experimental details for Bi surfactant assisted growth The dilute nitride films discussed in this Chapter were grown according to the standard MBE growth procedures as described in Chapter 2. Growth temperatures for the dilute nitrides were typically between 400 and 460 C, and the applied Bi surfactant flux was varied from Torr depending on the experiment. A beam of approximately 50% Bi monomers and 50% Bi 2 dimers is expected from the effusion cell at typical operating temperatures [KOT81]. The As overpressure was maintained in the 10 7 Torr range during these growths, a condition where Bi is not observed

59 Chapter 4. Bismuth surfactant growth of the dilute nitrides 44 to incorporate even at the ppm level, as determined by SIMS and x-ray diffraction [TAY + 02]. Nitrogen concentrations in the ternary GaNAs films were measured by peak splitting in the high-resolution x-ray diffraction spectra (see Chapter 2.3.1), as were the In concentrations in InGaAs films, while in InGaNAs films the nitrogen and indium concentrations were assumed to be the same as their relative concentrations in ternary InGaAs and GaNAs films grown at identical growth conditions. In a number of cases, two layers of dilute GaN x As 1 x were grown on the same substrate, with different x due to a change in conditions during the growth of the second layer. Individual layer concentrations were extracted from simulations of high-resolution x- ray diffraction measurements. Finally, in-situ RHEED and RGA experiments were performed during the growth of some samples discussed here. Revelant experimental details may be found in sections and respectively. 4.3 Effects of Bi Surfactant Morphology One of the most interesting results to be obtained during the study of Bi as a surfactant was the discovery of its dramatic smoothing capability [Ada02, TAT + 03]. Adamcyk found that Bi tends to promote a FvM and/or step-flow growth mode, even under conditions (low growth temperature, low As 2 overpressure) where anisotropic roughness along [1 10] is expected to develop. Here further evidence of the surface smoothing due to an enhancement of surface diffusion length is presented in the form of RHEED oscillations (described in section 2.2.1). Figure 4.1 a) presents RHEED oscillations corresponding to layer-by-layer growth for GaAs grown at 440 C with and without Bi surfactant. The top set of oscillations, corresponding to growth without the surfactant, are relatively weak and rapidly damped. This result indicates that the surface diffusion length at this low temperature is insufficient to sustain a pure 2D layer-by-layer growth mode, and the growing surface roughens as atoms incorporate close to the point of impact rather than diffusing to incorporate at step edges. When the same growth is performed during exposure to a Bi flux of 1 x 10 7 Torr, the RHEED oscillations are stronger and less rapidly damped, clearly showing that Bi preserves the layer-by-layer growth mode. Our data is similar to RHEED data obtained for homoepitaxy of GaAs with Pb [MG93]. In that work it was suggested that the surface diffusion length is enhanced by a reduced energy barrier for hopping due to a surfactant-induced reduction in the semiconductor-semiconductor bond strength.

60 Chapter 4. Bismuth surfactant growth of the dilute nitrides GaAs growth T subs. = 440 C (a) RHEED intensity (arb. units) No Bi Bi flux - 1 x 10-7 Torr Time (s) 200 GaAsN with Bi (BEP ~10-7 Torr) (b) T subs. = 440 C % N RHEED intensity (a.u.) % N 0.45% N No Bi % N 25 Time (s) Figure 4.1: RHEED oscillations during growth at 440 C of (a) GaAs, without Bi (top) and with a Bi flux of 1 x 10 7 Torr (bottom) and (b) GaN x As 1 x. The top set of oscillations correspond to a film with x = 1.4% N grown with a Bi flux, the bottom set correspond to a film with x = 0.01% grown with a Bi flux, and the middle two sets of oscillations correspond to films with x = 0.45%, one grown with Bi flux and the other without, as indicated. For all samples grown with Bi, the flux was 1 x 10 7 Torr.

61 Chapter 4. Bismuth surfactant growth of the dilute nitrides 46 Figure 4.1 b) shows RHEED oscillations for the growth of GaN x As 1 x at three different values of x. The pirani voltage indicating N pressure in the plasma source for the three x contents were 3.7 V, 2.2 V and 0.7 V, with the lowest voltage corresponding to the highest nitrogen concentration (see Figure 2.1). N is known to induce surface roughening during growth of GaNAs, particularly at high N concentrations [ATR + 01]. This is shown clearly by the RHEED oscillations, which show a progressively more rapid decay in intensity as x is increased from 0.01% to 1.4%. The middle two sets of oscillations in Figure 4.1 b) correspond to two films with x = 0.45%, one grown with a Bi flux of 1 x 10 7 Torr, and one grown without, as indicated on the graph. The oscillations for these two films suggest an increase in surface diffusion length similar to the results for GaAs growth presented in Figure 4.1 a). The film grown without the Bi shows weaker oscillations with a rapid decay, qualitatively similar to the top set of oscillations that corresponding to the film with x = 1.4%. Thus the roughening without Bi is similar to behaviour observed for a N concentration almost 3x greater. We conclude that the Bi surfactant reduces the tendency to roughen associated with N incorporation, also likely due to an increase in surface diffusion length Photoluminescence Photoluminescence (PL) intensity is another film property favourably affected by the presence of the Bi surfactant, as discovered previously [Ada02, TAY + 02]. Figure 4.2 shows the PL intensity for two 6 nm InGaNAs QWs grown under the same experimental conditions except for the presence of Bi during growth of the second QW, which exhibits peak PL intensity 2.4 times greater than the sample grown without Bi. Data for both samples following a rapid thermal anneal at 730 C for 60 seconds is also shown. An increase in intensity on annealing is expected for InGaNAs [DCN + 05], 1 so that observation here is not surprising, but remarkably the magnitude of the increase in PL intensity due to the Bi surfactant remains after the annealing process. From this data alone, the source of the PL improvement due to Bi is unclear. It is possible that the surfactant layer prevents the incorporation of impurities that suppress PL, or more specifically, reduces nitrogen related defects. Further insight into the improvement of the room temperature PL with Bi was gained by an investigation into the low temperature PL efficiency on a series of GaNAs 1 The blueshift of 20 nm in the peak PL emission wavelength observed after annealing is also typical for InGaAs and InGaNAs QWs, and is usually attributed to redistribution of In and/or N within the QW [DCN + 05].

62 Chapter 4. Bismuth surfactant growth of the dilute nitrides 47 RT PL intensity (arb. units) InGaNAs QWs (26% In; 1.1% N) As Grown Annealed 60s at 730 C With Bi (BEP ~10-7 Torr) Wavelength (nm) 1400 Figure 4.2: Room temperature PL from as-grown and annealed 6 nm InGaNAs quantum wells grown at 450 C, with and without Bi surfactant (Bi flux of 10 7 Torr). Max intensity increased 2.4x for the sample grown with Bi for both conditions. Dashed lines correspond to samples annealed at 730 for 60 s. samples grown with and without the Bi surfactant. Collection of PL data at low temperatures increases the luminescence efficiency and reduces thermal broadening, resulting in increased resolution in the emission spectra. The low temperature PL data for these samples was taken by D. Beaton using the pulsed 523 nm green laser discussed in section 2.3.2, and a cryostage achieving temperatures as low as 10K [Bea03]. PL spectra as a function of temperature for a sample grown with and without a Bi surfactant are shown in Figure 4.3 a) and b) respectively. Both samples were grown at 500 C, with a Bi flux of 4 x 10 6 applied during the growth of the sample with PL in Figure 4.3 b). In Figure 4.3 a), the emission spectra all show a relatively narrow peak at the band gap as well as a broader tail extending to lower energies, attributed here to localized electronic states in the bandgap. An obvious possibility is that the localized states are associated with the N clusters discussed in section 3.1, as previously reported [FNJ + 99, KZ99]. The sample grown with the Bi surfactant (spectra in Figure 4.3 b)) has a reduced low energy component in its emission spectra, relative to the narrow bandgap emission, compared with the sample grown without the surfactant in Figure 4.3 a). The

63 Chapter 4. Bismuth surfactant growth of the dilute nitrides 48 Figure 4.3: Temperature dependent PL spectra for nominally 250 nm thick GaN x As 1 x films grown under identical conditions at 500 C, except that (a) r1349 was grown with no Bi and (b) r1350 was exposed to a Bi flux of 4x10 6 Torr. A low energy tail attributed to N cluster states is seen in the spectrum of r1349 and is greatly reduced for the sample grown with the surfactant. PL data courtesy of D. Beaton [Bea03]. two samples have almost exactly the same N contents (0.54 and 0.56 %) and band gaps, as indicated by the energy of the sharp feature corresponding to the band gap emission. We interpret the difference between the two samples as an indication that the gap state density is lower in the sample grown with the surfactant. As the temperature is increased the broad low energy part of the emission spectrum in the PL declines in intensity relative to the bandgap emission Nitrogen incorporation When growth conditions are otherwise identical, use of a Bi flux results in an increase in N incorporation, as illustrated by the x-ray diffraction spectra presented in Figure 4.4. The samples were grown in sequence, both at substrate temperature of 430 C, and the only difference in growth conditions between the two samples was the presence of a Bi flux of 4 x 10 6 Torr for the sample with the bottom x-ray diffraction pattern in Figure 4.4. The x-ray pattern for the Bi assisted sample shows a greater peaksplitting between the substrate and epilayer peaks, and after simulation to determine the compositions, we find the presence of the Bi flux during growth resulted in an increase in nitrogen concentration of 36% for this particular sample. The effect of increasing Bi flux was further investigated through a series of GaNAs growths at two different growth temperatures, with data similar to that presented in Figure 4.4. It

64 Chapter 4. Bismuth surfactant growth of the dilute nitrides Diffracted Intensity ( a.u.) r1353 data simulation r1354 data simulation GaNAs:Bi 0.68% N GaNAs 0.5% N θ ( arcsec ) Figure 4.4: High resolution XRD data for two GaN x As 1 x films grown at 430 C with (r lower diffraction pattern) and without (r upper diffraction pattern) a Bi flux of 4 x 10 6 Torr. Dashed lines represent simulations from dynamical theory indicating that N content increases by 36% for the film grown with Bi. Data are offset for clarity. was discovered that the nitrogen concentration continued to increase with increasing Bi flux, saturating at a Bi flux of about 2 x 10 6 Torr. These experiments will be discussed further in section The important point is that the Bi surface layer was able to increase the N incorporation in GaNAs. Clearly, not all of the available nitrogen is incorporating during growth without the surfactant. 4.4 Understanding the surfactant effect Measuring the Bi surface coverage with RHEED In light of the beneficial modifications that the Bi surfactant was observed to have on the properties of the dilute nitrides, and with the intention of understanding the range of conditions where the surfactant effect is operational, it is of interest to measure the surface coverage of bismuth under typical growth conditions. We were able infer the bismuth surface coverage as a function of substrate temperature and Bi flux from RHEED during Bi adsorption and desorption experiments under constant As overpressure [YTT05]. Two experimental observations proved RHEED to be a

65 Chapter 4. Bismuth surfactant growth of the dilute nitrides 50 Figure 4.5: RHEED patterns of (3x1) Bi-stabilized reconstruction along 110 azimuths during growth at 370 C suitable technique to carry out these experiments. The first was that the Bi flux caused a reconstruction change from the typical (4x2) pattern observed for GaAs at temperatures below 500 C to a (3x1) reconstruction, an observation consistent with a previous Bi surfactant study [PKHB00]. Photographs of the (3x1) RHEED reconstruction for the two 110 azimuths are presented in Figure 4.5 for film growth at 370 C. This is a lower growth temperature than used for the RHEED experiments discussed here, though the reconstruction is identical. The second experimental observation was that this reconstruction change was accompanied by an increase in the RHEED specular beam intensity, which was assumed to be proportional to the Bi surface coverage. At high Bi flux, and/or low substrate temperature, the specular beam intensity saturated, which is further assumed to correspond to a Bi surface coverage of one monolayer. As the substrate temperature is decreased, the equilibrium Bi coverage increases, resulting in an increase in intensity of the specular RHEED spot. This idea is similar to the interpretation that the maxima in RHEED intensity oscillations correspond to monolayer completion, and was used previously to monitor As surface segregation during growth of Si [HTZ01]. Despite the fact that the Bi did not incorporate into the film under the growth conditions used in these experiments, there was no evidence for the formation of metallic Bi droplets on the surface, which further supports the assumption of only monolayer coverage. Neither a reconstruction change, nor an increase in RHEED specular intensity was observed when N was introduced to the surface, which suggests that N does not have the same tendency to surface segregate as Bi.

66 Chapter 4. Bismuth surfactant growth of the dilute nitrides 51 Figure 4.6 shows the specular RHEED intensity as a function of time for different substrate temperatures with the Bi shutter kept open for 30s and then closed, while maintaining a fixed Bi flux of 1.4 x 10 5 Torr. The experiment was performed, and repeated, for 8 different substrate temperatures between 425 and 585 C, but only 4 data sets are shown here for clarity. The exponential decay in the intensity when the shutter is closed is interpreted as due to thermal evaporation of the Bi from the surface. The decay rate of the RHEED intensity following the closure of the Bi shutter for each substrate temperature was determined from exponential fits to the data, and is plotted as a function of 1/T in the inset of Figure 4.6. From the Arrhenius plot, the desorption has an activation energy of 1.34 ev, which is similar to the activation energy with a value of 1.3 ev measured experimentally by Noreika et al. [NTFW82] for desorption of Bi from InSb surfaces. Figure 4.7 a) is a plot of the specular RHEED intensity taken from data similar to that presented in Figure 4.6 with the Bi shutter open, as a function of substrate temperature. The specular beam intensity decreases rapidly with substrate temperature above 500 C, which is interpreted as being due to a decrease in Bi coverage at high temperatures. The negative coverage data point at 560 C is a consequence of the corresponding data set for that temperature in Figure 4.6, where a slight decrease in intensity is observed when the Bi shutter is opened. This change in intensity may be due to a change in surface reconstruction as the surface is bombarded with Bi atoms which have some residence time on the surface but rapidly evaporate. Figure 4.7 b) shows the specular beam intensity with the substrate temperature held constant at 460 C, while the Bi flux is decreased by slowly ramping the temperature of the Bi cell down linearly from 700 to 450 C. (Ramp time is shown on the top axis in Figure 4.7 b).) In this case there is a significant decrease in RHEED intensity for Bi fluxes below 1 x 10 6 Torr A model for the surface coverage If the RHEED intensity increase is interpreted as a measure of the Bi coverage, then it is possible to fit the RHEED data as a function of temperature and flux with a Langmuir adsorption isotherm. The Langmuir isotherm uses the law of mass action to describe the equilibrium adsorption of gases on solid surfaces. Assuming the adsorbed Bi is in thermal equilibrium with the vapour, the fraction of occupied surface sites, θ, is determined by equating the rates of absorption and desorption, and can be written

67 Chapter 4. Bismuth surfactant growth of the dilute nitrides 52 open Bi close Bi RHEED Specular Intensity (a.u.) C 475 C 505 C 560 C Desorption Rate (s -1 ) /T (K) Time (s) Figure 4.6: Change in RHEED specular intensity from GaAs substrate at different substrate temperatures when exposed to a Bi flux of 1.4x10 5 Torr, and subsequent decrease in intensity when flux is removed. Inset shows desorption rate vs. inverse temperature.

68 Chapter 4. Bismuth surfactant growth of the dilute nitrides 53 Bi coverage (ML) RHEED data Langmuir model 0.0 (a) Substrate temperature ( C) 800 Time (s) RHEED Data Langmuir model Bi Coverage (ML) open Bi (b) close Bi Bi BEP (Torr) 10-8 Figure 4.7: (a) Bi coverage vs. growth temperature as inferred from change in RHEED intensity shown in Figure 4.6, and (b) Bi coverage vs. Bi flux at constant substrate temperature of 460 C from RHEED intensity as the Bi cell is ramped linearly with time from 700 to 450 C. Both are plotted with Langmuir model (see section for details).

69 Chapter 4. Bismuth surfactant growth of the dilute nitrides 54 θ = bp (4.2) 1 + bp where P is the vapour pressure of Bi at the substrate surface due to the Bi source, and b = b o exp[u/k b T], where b o is a constant, k b is Boltzmann s constant and U is the activation energy for desorption, in the limit that the surface adsorbed Bi atoms do not interact with each other. The pre-exponential factor b o arises during our derivation of Equation 4.2 by assigning 3 degrees of freedom to a Bi atom within the adsorbate layer; two translational within the site area, σ o, and one vibrational normal to the surface, and using the partition function to consider the occupancy of a single site. Thus the term b can be written explicitly as b = 1 ( ) h 2 1/2 σ o exp ω o 2πmk b T [ ] U + zǫθ k b T (4.3) where ω o is the vibrational frequency of Bi on the surface and m is the mass of a Bi atom. An attractive Bi-Bi interaction is taken into account in the mean field approximation by the zǫθ term in the exponent in Equation 4.3 [FG52], where ǫ is the Bi-Bi interaction energy, and z is the coordination number of the Bi atoms on the surface. The inclusion of the lateral interaction energy term zǫθ causes a steeper transition from partial to full coverage as a function of pressure. The modified Langmuir model with U = 1.8 ± 0.4 ev, zǫ = 0.12 ev and σ o = 0.2 nm 2 fits both the pressure and temperature dependence of the RHEED data, as indicated by the solid lines in Figures 4.7 (a) and (b). The adsorption energy U determined from the isotherm is in reasonable agreement with the activation energy of 1.34 ev determined from the temperature dependence of the Bi evaporation rate in Figure 4.6,and is also comparable with the activation energy of 1.7 ev for self desorption of Bi [Her50]. We would expect the binding energy to be dependent on the Group V/III ratio, and to increase over the determined value for Group III-rich conditions [YTT05]. In fact Bi incorporation into the bulk crystal is observed when the V/III ratio is on the order of, or less than one, and this is discussed in detail in Chapter Modelling the increase in nitrogen incorporation As discussed in section 4.3.3, an increase in the nitrogen content is observed for GaN x As 1 x films grown with Bi flux. This increase is shown graphically in Figure 4.8 for two series of GaNAs films grown at 400 and 460 C respectively. A similar effect is observed in MBE growth of GaNAs with Sb [HULR00], where the nitrogen

70 Chapter 4. Bismuth surfactant growth of the dilute nitrides Nitrogen Concentration (%) C 400 C x10-6 Bismuth flux (Torr) Figure 4.8: Nitrogen concentration vs. Bi flux for GaNAs samples grown at 400 and 460 C. Solid lines show [N] Bi = [N] o ( θ), where θ is coverage from the Langmuir model. The increase in nitrogen with Bi flux saturates at monolayer Bi coverage. incorporation is also enhanced by the presence of Sb, but the opposite behaviour is observed in GaNAs grown by organometallic vapour phase epitaxy (OMVPE), where a Bi surfactant results in a decrease in the nitrogen content of the layers [DHSS02]. Interestingly, the increase in N content with Bi flux closely parallels the Bi surface coverage and can be described by the same Langmuir isotherm as the RHEED data. In Figure 4.8, the solid lines showing the nitrogen content as a function of Bi flux, [N] Bi, are given by [N] Bi = [N] o ( θ) (4.4) where [N] o is the N concentration in the absence of Bi and θ is the Bi coverage corresponding to a particular Bi flux from Equation 4.2. The trend predicted by this expression is in good agreement with the N concentration data for both sample series. This suggests that N incorporates 55% more efficiently on regions of the surface that are covered with Bi than those that are not, and that once the flux is sufficient such that there is a full monolayer of Bi coverage, the amount of nitrogen remains constant with increased Bi flux. This result is surprising since the presence

71 Chapter 4. Bismuth surfactant growth of the dilute nitrides 56 of a non-incorporating group V surface layer might intuitively be expected to hinder the N incorporation, not increase it, due to competition between group V elements. Secondly at 500 C and below the N incorporation is independent of temperature. This would normally be interpreted as a sign that all of the reactive nitrogen sticks to the surface. The fact that the N incorporation increases with Bi means that some of the reactive N does not incorporate in the absence of the surfactant. As mentioned, these results are in sharp contrast with the OMVPE results where high coverages of surfactant were found to completely suppress the N incorporation. The radical difference between the MBE and the OMVPE results may arise from differences in the adsorbing species on the surface. In OMVPE the dimethyl hydrazine does not completely decompose and the adsorbed species is expected to be a N-C or N-H complex [DHSS02]. These species may have rather different sticking characteristics than the active nitrogen produced in the N 2 plasma source. Another possible explanation for the difference could be the fact that surface reconstruction does not occur in OMVPE because of surface saturation by H or CH 3 molecules. The surface reconstruction may allow greater N incorporation, as will be discussed further in section In-situ mass spectrometry of the surface In order to better understand the role of Bi in N incorporation into GaAs, in-situ mass spectrometry was performed during growth with a quadrupole mass spectrometer (also known as an RGA - see section 4.4.4). During a previous growth campaign where the RGA was situated behind the substrate with no line of sight to the growth surface, it had been observed that a species with mass number 89, corresponding to AsN, was present during growth of the dilute nitrides. It was suggested that a high pressure of the AsN species might be indicative of a loss of reactive N available for incorporation, through the formation of the volatile AsN on the growth surface. In addition, it was thought that the Bi surface layer might suppress the formation of AsN at the surface, resulting in the observed increase in N incorporation efficiency. To test this hypothesis, the RGA was repositioned in order to have line-of-sight to the surface. A number of experiments were performed where the partial pressure of 89 amu, or AsN, was monitored. As expected, the levels of AsN present increased from the background to measurable levels in the Torr range once N 2 and As 2 were introduced to the growth chamber, presumably due to the formation of AsN in the mass spectrometer. Igniting the plasma source further increased the AsN partial pressure

72 Chapter 4. Bismuth surfactant growth of the dilute nitrides 57 by an order of magnitude, while subsequent initiation of GaAs growth by introduction of Ga flux reduced the partial pressure by a factor of approximately three, results indicating that AsN plays a role in the growth process. The next step was to examine the effect of a Bi flux, and results from two experiments to this end are shown in Figure 4.9 a) and b). In both cases, the plasma source had been ignited and an As 2 flux was present. Figure 4.9 a) shows the effect of closing the Ga shutter (i.e. stopping crystal growth), which results in an increase in AsN available, then introducing a Bi flux to the surface, which was held at 450 C. Contrary to our expectation, the amount of AsN actually increases with the introduction of Bi. Figure 4.9 b) shows the effect of introducing a Bi flux during actual growth (i.e. Ga shutter is open) at 450 C, recovery of the original AsN level once the Bi shutter is closed, and then a repeat of the same process. Again, there is an increase in the AsN partial pressure when Bi is introduced. Clearly, the increase in N incorporation with Bi is not due to the suppression of volatile AsN formation. 4.5 Discussion It is clear from the data presented here that N incorporates into GaAs rather differently when Bi was present at the surface, as the concentration increases, and temperature dependent PL of GaNAs shows a reduction in shallow defects. If the shallow defects are associated with N clusters, the surfactant reduces the tendency of the nitrogen to cluster. A reduction in cluster states would also explain the increase in the RT PL of the InGaNAs QWs. In fact, Duca et al. [DCN + 05] studied In and N distributions in InGaNAs QWs before and after annealing with cross-sectional scanning tunnelling microscopy, and found that the annealing process resulted in a significant decrease in the number of N-N nearest neighbours. They observed that the number of N-N pairs decreased on annealing to values below those expected for a random distribution. They also observed a reduction in localized states in the band gap of InGaNAs as measured from PL. This supports the idea that if Bi reduces clustering, an increase in PL intensity is expected. As discussed in Chapter 3, GaNAs has a tendency to phase separate due to the miscibility gap. If Bi increases the surface diffusion length, one might expect that the tendency to cluster would be enhanced, not reduced. One theory that might explain our experimental observations is that of surfacereconstruction-enhanced solubility, as proposed by Zhang and Zunger [ZZ97]. This

73 Chapter 4. Bismuth surfactant growth of the dilute nitrides x10-12 (a) AsN Pressure (Torr) Ga OFF Bi ON OFF Time (s) 400 AsN Pressure (Torr) 40x Ga ON Bi ON OFF ON OFF (b) Time (s) 250 Figure 4.9: Line-of-sight RGA partial pressure of mass number 89 (AsN) under two different growth conditions at 450 C, with arrows correspond to opening and closing of Ga and Bi shutters as indicated. a) Bi flux is introduced during plasma operation, resulting in increase in AsN pressure. b) Bi flux is introduced during growth of GaNAs (plasma on, Ga shutter open), again resulting in an increase in AsN pressure.

74 Chapter 4. Bismuth surfactant growth of the dilute nitrides 59 theory is an attempt to explain why N is able to incorporate in GaAs to the degree that is observed experimentally, when the bulk thermodynamic solubility of N in GaAs is extremely low at 10 7 at%. Their first principles calculations suggest that the surface reconstruction is likely responsible for higher solubility of N on GaAs surfaces than in the bulk, as the stress state at the surface is locally altered due to the lower atomic density, and in addition, near-surface atoms have more freedom to move. This theory is supported by experimental data from Reason et al. [RMY + 04], who found that all other variables being equal, a (2x1) reconstruction resulted in the highest substitutional N incorporation (for non-surfactant assisted growth by MBE). They attribute this effect to the fact that the (2x1) reconstruction offers the highest number of available Group V sites. For the case of the Bi surfactant, it is possible that the (3x1) Bi-stabilized surface may have a higher N solubility than that of the standard As (4x2) reconstruction. If the solubility is enhanced, phase separation or clustering is also likely to be reduced. A number of other phenomena could also contribute to the increase in N incorporation. Bi may simply displace As at the surface making it easier for the N to compete with As for Ga bonds. In addition, line-of-sight mass spectrometry shows that the formation of AsN is enhanced on Bi-covered GaAs surfaces. It is possible that AsN acts as an intermediary in the incorporation of N into GaAs, and the concentration is increased due to this alternative incorporation mechanism. 4.6 Summary Under typical III-V growth conditions, and more specifically III-V dilute nitride growth conditions, Bi does not incorporate into the growing film, but segregates to the surface and acts as a surfactant. The surface morphology of dilute nitride films, which tend to be rough, is improved dramatically with the Bi surfactant, which promotes a layer-by-layer growth mode as indicated by RHEED oscillation measurements. Use of a Bi surfactant improves the RT PL of InGaNAs QWs by a factor of approximately 2.4 in both the as-grown and annealed conditions for a Bi flux of 10 7 Torr. Low temperature PL measurements indicate a reduction in mid-bandgap emission for GaNAs grown with a Bi surfactant, likely due to a reduction N cluster states.

75 Chapter 4. Bismuth surfactant growth of the dilute nitrides 60 The nitrogen incorporation in GaNAs is enhanced by up to 60% by a Bi surface layer at low N concentrations. The surface coverage of Bi as a function of substrate temperature and Bi flux, and the Bi desorption rate were inferred from measurements of the change in RHEED intensity. The Bi coverage saturates at low temperature (450 C) and high flux (10 6 Torr). A modified Langmuir model was successfully applied to describe the surface coverage results. The model assumes an attractive interaction between Bi atoms on the surface and has a binding energy for an isolated Bi atom of 1.8 ± 0.4 ev. The Langmuir model was also successfully applied to describe the increase in N concentration, which saturates once a full monolayer of Bi coverage is achieved, suggesting that N incorporates more efficiently on regions of the surface covered with Bi. Line-of-sight mass spectrometry measurements show an increase in abundance of a volatile AsN species when Bi is present during nitride growth. This species may play a role in the improvement in N incorporation efficiency. 4.7 Conclusion To conclude, use of Bi as a surfactant results in a number of benefits for the growth of dilute nitride materials. The Bi-surface complex results in increased diffusivity for adatoms, making it possible to grow films at relatively low temperatures that have surface structures similar to films grown at higher temperatures. In principle, these results should be applicable to growth of other III-V materials where material quality is a concern. Finally, the observation that N incorporation efficiency is enhanced by up to 60% for growth with Bi offers a new way of controlling N incorporation in dilute nitride semiconductors.

76 Chapter 5. Growth and properties of dilute bismides 61 Chapter 5 Growth and properties of dilute bismides In this chapter the rationale for addition of Bi to III-V semiconductors is discussed, and a summary of the work done to achieve this goal is presented. A discussion of the current state of knowledge on bismides sets the stage for the work presented here, which encompasses an investigation into the growth conditions necessary to produce high quality GaAsBi materials, as characterized by the structural and electronic properties of the material. 5.1 History of GaAsBi and related alloys The concept of introducing Bi to engineer the band gap of III-V semiconductors is not a new idea. In a 1971 Journal of Crystal Growth article, Joukoff writes Because of the semi-metallic character of InBi, one can hope to reduce the InSb gap when introducing Bi into Sb sites [JJL71]. Over the next three decades, the material InSbBi and related alloys (InAsSbBi, InAsBi) were grown by the Czochralski method [JJL71], molecular beam epitaxy [NTFW82], and metalorganic chemical vapour phase epitaxy (MOVPE) [FMCS90], due to an interest in long wavelength infrared photodetectors. The band gap of InAsBi was found to decrease at a rate of 55 mev/%bi with increasing Bi concentration, and the bandgap of InAsSbBi at 46 mev/% Bi [FMCS90]. These numbers represent unusually high bowing, suggesting the potential achievement of photodetectors operating at 12 µm with only a few percent Bi alloying. More recent papers [LKR97] assess that there has been a lack of progress in this area due to the limited solubility of Bi. In 1998, Oe and Okamoto were the first researchers to investigate the incorporation of Bi into GaAs, using MOVPE at very low growth temperatures (365 C) [OO98]. X-ray diffraction showed that the GaAs 1 x Bi x film achieved was coherent with a larger lattice constant than GaAs, with a value x of approximately 2%. As one might expect from the combination of a semimetal (GaBi) with a semiconduc-

77 Chapter 5. Growth and properties of dilute bismides 62 tor (GaAs), the material showed room temperature PL that was redshifted with respect to GaAs. The only discouraging result obtained was that small clumps of whiskers (i.e. surface defects with large aspect ratios) were visible on the surface of the film after growth, and the density of whiskers increased with growth time. This three-dimensional surface nucleation persisted through several further years of experimenting with MOVPE [Oe02], though this did not prevent further characterization of electronic and structural properties of the material by Raman spectroscopy [VOY + 01] and photoreflectance (PR) spectroscopy [YKWO03]. The PR measurements confirmed that Bi is responsible for a reduction in the bandgap with a large bowing, and also indicated that the MOVPE grown material exhibits a very weak temperature dependence of the bandgap. For example, a GaAsBi film with 2.6% Bi was found to have a temperature dependence δe g /dt = mev/k, or 1/3 of the temperature sensitivity of GaAs. This property would bode extremely well for the use of the material as the active region of semiconductor lasers for communication systems where emission wavelengths are required to remain constant under ambient temperature variations. In the meantime, several researchers at the National Renewable Energy Lab (NREL) in Colorado began to investigate the idea of isoelectronic co-doping of Bi and N in GaAs as an antidote to the degradation in electronic quality of the material experienced by the addition of N alone [MZVS01]. Provided that the N and Bi were introduced as pairs, a number of benefits could be realized in theory. First, there is the potential for strain compensation, as N is a smaller atom and Bi is larger atom than the host As atom. The strain compensation could result in a mutual increase in solubility of N and Bi. The second potential improvement would be to the mobility, analogous to an enhancement in mobility observed in other semiconductors by co-doping with both acceptors and donors [PB98]. In that particular case, oppositely charged long-range Coulomb scatterers combine to behave as single short range dipole scatterers, resulting in an enhancement of the mobility, as observed for GaN co-doped with Be and O. A similar result could be seen from N, which can form traps near the conduction band edge resulting in a long range Coulomb potential, and Bi, which is expected to behave similarly near the valence band edge [ZMW05]. An analogous effect might also be expected in terms of the strain state in the alloy, where N and Bi individually cause a strain monopole in the lattice due to their respective size differential with As. Since the strain introduced by each is of opposing sign, in next nearest neighbour configuration they would form a localized strain dipole, substantially reducing their individual disruptions. In addition, theoretical band structure

78 Chapter 5. Growth and properties of dilute bismides 63 and total energy calculations predicted that co-doping to form GaNAsBi would reduce the alloy formation energy over that of GaNAs or GaAsBi separately, and that a bandgap of 1 ev, lattice matched to GaAs, would be possible with very low nitrogen concentrations [JWZ02]. This would make the material ideal for inclusion in next generation 4-layer multijunction solar cells, where a material with these properties is needed to increase the efficiency beyond today s state of the art 3 layer solar cells (39% efficiency). Projected efficiency for a 4-layer GaAs-based solar cell including a 1 ev bandgap layer is 42%, a massive increase in solar cell terms. These ideas provided the inspiration for a foray into GaAsBi growth at the MBE Lab at UBC. After extensive experimentation with growth conditions, Tixier achieved incorporation of Bi into GaAs up to 3.6% (measured by secondary ion mass spectroscopy (SIMS)) using low growth temperatures (380 C), and very low group III/V ratios [TAT + 03]. These are conditions considered far from ideal for the growth of most GaAs related alloys. Work was also done to calibrate the lattice parameter of the new material with Rutherford backscattering (RBS) and x-ray diffraction, which was found to obey Vegard s law with extrapolation to the lattice parameter of the hypothetical zincblende alloy GaBi ( nm) [JWZ02]. A collaboration with NREL resulted in characterization of the MBE grown GaAsBi by electroreflectance (ER), and the magnitude of the bandgap bowing was determined to be 88 mev/%bi [FSM + 03], less than observed in the dilute nitrides (up to 200 mev/% N at low [N] concentration) but significantly higher that of In (12 mev/%in), even though the effect on the lattice parameter is similar (see Table 1.1). A recent theoretical paper suggests that a resonant interaction of a Bi impurity state with the valence band is responsible for the large bowing, analogous to the effect of N on the conduction band of GaAs, and finds good agreement between the experimentally observed bowing and predictions from a pseudopotential-based supercell calculation [ZMW05]. One puzzling piece of the story is that unlike the previous work by Yoshida et al. [YKWO03] using PR, the GaAsBi studied in [FSM + 03] was found by ER to exhibit a temperature dependence of the band gap almost identical to GaAs (-0.404meV/k for GaAs 1 x Bi x with x=3.1%, compared with mev/k for GaAs). The reason for this discrepancy is not known, though it may be that PR allows electron-hole pairs to thermalize into band edge localized states to a progressively greater extent at low temperatures, distorting the true temperature dependence. This weakness is also associated with PL measurements of temperature dependence. In addition to the work on GaAsBi, some progress has been made in exploring the concept of co-doping with N and Bi, both at UBC [TWY + 05] and elsewhere

79 Chapter 5. Growth and properties of dilute bismides 64 [HOFY05]. Some of the work done at UBC is discussed here in section 5.4. In addition to the reasons already discussed, it has recently been predicted [ZMW05], and then confirmed experimentally [FFM + 06], that a small concentration of Bi has a relatively large effect on the spin-orbit splitting, which makes the material potentially useful for spintronics devices. A US patent was also recently granted [Has06] that involves the use of GaAsBi, as its properties (a flat conduction band alignment with GaAs, small bandgap and potentially high electron mobility) are all desirable characteristics for reducing the threshold voltage and power consumption in GaAs-based heterojunction bipolar transistors (HBT). HBTs are used in cellular phone transmitters. In the remainder of this chapter, the work done to further investigate the feasibility of producing high quality GaAsBi and GaNAsBi with the requisite properties will be described. 5.2 Exploring the growth conditions for GaAsBi At mass number 208, Bi is the heaviest non-radioactive element in the periodic table. The Bi atom is approximately 25% larger than Ga and As and tends to surface segregate during MBE growth, making it a good surfactant as discussed in detail in Chapter 4. Unlike its neighbours Hg, Sb and Tl in the periodic table, it is also non-toxic. Under standard growth conditions for GaAs with a typical arsenic overpressure (As 2 :Ga of 7:1 at a growth temperature of 580 C) we are unable to detect Bi incorporation with SIMS. The low solubility of Bi in GaAs is most probably due to the large substitution energy, E s, reflecting strain [ZZ97]. This implies that in a competition for group V lattice sites, As will usually prevail over Bi, which explains the difficulty in Bi incorporation under typical growth conditions. However, the competition between As and Bi can be mediated by reducing the available As, and ensuring that all of the incident Bi flux sticks to the surface, accomplished in practice by carrying out the growth at temperatures below 400 C with an As/Ga ratio close to unity [TAT + 03]. Unfortunately, small deviations from the optimal growth conditions can lead to the formation of Ga or Bi droplets on the surface. Ga droplets form when the group III/V ratio is too low, and Bi droplets form when the Bi flux is too high, as is discussed in further detail in section Figure 5.1 presents AFM images of two films corresponding to the aforementioned conditions, both resulting in the formation of large droplets on the order of several hundred nanometers in diameter. Figure 5.1

80 Chapter 5. Growth and properties of dilute bismides 65 a) shows droplets formed due to high Bi flux at the end of film growth at 350 C and then immediately scanned by AFM, and Figure 5.1 b) shows the resulting surface from continued film growth at 375 C after Ga droplet formation. Due to the scale of the droplets, evidence of their formation is immediately discernible during the growth process. The sample surface becomes visibly cloudy, the RHEED pattern fades and in extreme cases becomes completely diffuse, and the surface-sensitive diffuse reflectance spectroscopy (DRS) signal used to monitor wafer temperature during growth is distorted. These results are all indications consistent with the presence of a liquid on the surface. Noreika et al. [NTFW82] also report RHEED fadeout during MBE growth of InAsBi, which they attribute to the formation of molten In-Bi alloys on the surface. For GaAsBi growth, if the droplets are Bi metal, ramping the temperature up to above 450 C after droplet formation results in recovery of the RHEED pattern and a shiny sample surface, consistent with the desorption of Bi from the surface as illustrated in Figure 4.6. If the droplets formed consist primarily of Ga metal, surface recovery is not possible as the vapour pressure of Ga is too low under accessible MBE substrate heater temperatures (up to 1000 C) to result in significant desorption [Tsa93]. In fact the compound will actually decompose into Ga liquid and As vapour before Ga will evaporate. Another complicating factor is that, while difficult to ascertain experimentally, the possibility of the droplets having a mixed Bi-Ga composition is also likely, as the Bi-Ga phase diagram shows mutual solubility over almost the entire composition and temperature range with a eutectic located at 29.5 C and close to 100% Ga [Ell65]. While growth of GaAsBi continues beneath the droplets, x-ray diffraction indicates that growth under such conditions results in compositional inhomogeneity, as the presence of liquid affects the Bi incorporation. This effect may be responsible for the absence of pendellösung fringes visible in x-ray diffraction spectra for the samples grown by Tixier. Control of droplet formation at the surface is a significant obstacle to the growth of high quality bismide material. In order to explore the conditions responsible for droplet formation with the intention of achieving greater control over the growth process, it was decided to apply diffuse light scattering (LS) as an in-situ process monitor, due to its high sensitivity to metallic droplets and other surface morphological features [YWTB06]. In the bismide growth experiments, light scattering was carried out using 244 nm UV incident light, sensitive to surface structure with a length scale 2π/q=196 nm (see section for details). The incident beam is typically aligned along the [110] direction on the wafer surface, meaning that given an appropriate angle of light detection, the scattered intensity would be greatest for

81 Chapter 5. Growth and properties of dilute bismides 66 Figure 5.1: AFM images of two GaAsBi surfaces showing large metal droplets: a) Film r1710 grown at 350 C with high Bi flux. Image shows a 5x5 µm scan with 300 nm vertical scale. b) Film r1688 grown at 375 C with droplet formation when the As flux was dropped below a critical value. Growth of GaAs was continued after the occurrence of droplets. Image shows a 10x10 µm scan with 250 nm vertical scale. development of morphological features along a perpendicular direction, [1 10] In-situ light scattering and surface morphology In Figure 5.2, the sharp onset of Ga droplet formation as detected by light scattering is shown. This data corresponds to the stepwise reduction in As 2 overpressure from the valved arsenic cracker source during growth of GaAs, then GaAsBi at 0.75 µm/hr at a growth temperature of 375 C. Figure 5.1 b) shows the AFM image corresponding to this growth, confirming that the sharp increase in the light scattering is caused by droplet formation. The LS and AFM data clearly indicate the effect of decreasing the As/Ga ratio below a critical point. The initial drop in scattered light intensity in Figure 5.2 is due to smoothing of the rough starting surface associated with thermal oxide desorption when the Ga and Bi shutters are opened during growth of a high temperature buffer layer at 550 C. The decrease in LS intensity starting just before 9000 s corresponds to surface smoothing associated with increasing the As 2 overpressure back above the rather sharp threshold for metal droplet formation. In order to ensure that the Ga droplet formation occurring when the As 2 pressure is dropped is not associated with Bi, a LS experiment identical to that presented in

82 Chapter 5. Growth and properties of dilute bismides 67 Scattered Light Intensity (au) Open Ga Open Bi Scattered light signal Chamber pressure 5x Chamber Pressure (Torr) x10 3 Time (s) Figure 5.2: In-situ diffuse UV light scattering intensity and As 2 pressure as a function of time during dilute bismide growth of r1688 by molecular beam epitaxy at 365 C. The arrow indicates the opening of the shutter on the Ga and Bi sources. The chamber pressure is a measure of the As 2 flux on the growth surface which is controlled by a mechanical valve on the arsenic cracker.

83 Chapter 5. Growth and properties of dilute bismides 68 Figure 5.2 was carried out with no Bi flux (not shown). Again, a sharp increase in scattered light intensity is observed at almost the same As overpressure when there is no Bi flux, with AFM confirming the presence of droplets. Figure 5.3 shows LS data for growth of a 220 nm dilute bismide film with 0.8% Bi at 385 C with a Bi flux of 9x10 8 Torr. This LS data shows a more gradual increase in surface roughness beginning around 3000 s. The abrupt initial increase, then decrease, in roughness during the first 1000 s of growth are once again due to thermal deoxidation and subsequent smoothing upon introduction of the Ga flux. The postgrowth surface morphology of the film corresponding to the in-situ measurements in Figure 5.3 has periodic surface ripples parallel to the [1 10] direction with a wavelength of approximately 75 nm and amplitude of 7.6 nm as shown in the 2x2 µm atomic force microscope image in Figure 5.4 c). The other two AFM images in Figure 5.4 a) and b) also show a rippled surface morphology corresponding to samples grown at the same growth temperature and Bi flux with similar in-situ LS profiles to Figure 5.3, though they show lower Bi concentrations (0% and 0.33% respectively, as measured by x-ray diffraction). Although the scale of the roughness does seem to increase for higher Bi incorporation, from these results we conclude that the surface ripples seen in Figure 5.4 are not due to the surface morphological instability sometimes observed in strained films (discussed in section 3.2.2), since even the film with no detectable Bi incorporation undergoes roughening. In addition, a film with more than twice the Bi content (1.9%) and thickness of the same order (125 nm) shows a smooth surface (rms roughness 0.2 nm) with no evidence of surface ripples (Figure 5.5). Instead we attribute the low amplitude surface ripples in Figure 5.4 to the effects of the low growth temperature and low As flux. This surface roughness is smoothed out in the presence of the higher Bi flux (4x10 7 Torr) for the sample in Figure 5.5, similar to the tendency to smooth the surface observed for non-incorporating Bi surfactant growth at higher temperatures (see section 4.3.1). The sensitivity of the anisotropy of the surface structure to As 2 overpressure has been observed previously [Bal05], even at higher growth temperatures where the surface diffusivity is significantly larger than that expected for bismide growth temperatures. A series of growths at 550 C with decreasing As 2 overpressures resulted in morphologies with increasing elongation of surface features along [1 10]. Physically, this is explained by the anisotropic mobility of the adatoms on the surface as the Ga adatoms tend to move along the As dimer chains of the surface reconstruction. At low As 2 fluxes Ga atoms tend to move easily along their preferred direction, [1 10], with

84 Chapter 5. Growth and properties of dilute bismides 69 Scattered Light Intensity (au) Open Ga Scattered light signal Chamber pressure Open Bi Ramp to 385 C Bi incorporation Roughening due to low p As 8x Chamber Pressure (Torr) Time (s) 3x10 3 Figure 5.3: In-situ diffuse light scattering intensity and As 2 pressure during growth of r1713 GaAs 1 x Bi x with x=0.8% at 385 C. The initial surface roughening and subsequent rapid smoothing is caused by the thermal oxide desorption followed by GaAs buffer layer growth. The roughening at long growth times is due to the reduction in As overpressure and is associated with the formation of the rippled surface morphology illustrated in Figure 5.4.

85 Chapter 5. Growth and properties of dilute bismides 70 Figure 5.4: 2x2 µm AFM images of three films grown at 385 C showing increasing scale of roughness along [110] (vertical direction) with increasing Bi content a) r1712, GaAs, no Bi detected by x-ray. Vertical scale is 10 nm and rms roughness is 1.1 nm b) r1711, GaAs 1 x Bi x, x=0.33%. Vertical scale is 25 nm and rms roughness is 3.8 nm c) r1713, GaAs 1 x Bi x, x=0.8%. Vertical scale is 50 nm and rms roughness is 7.7 nm. infrequent hops along [110]. At very high As 2 flux (typical MBE growth condition for GaAs) Ga diffusion tends to be two dimensional, which generates a more isotropic surface morphology. Adamcyk [Ada02] found that at 440 C a Bi surfactant was able to suppress the anisotropy associated with low As 2 pressure, likely due to a change in diffusivity of Ga adatoms as discussed in Chapter 4, and we speculate that for the film shown in Figure 5.5, the right balance between Ga, As 2 and Bi flux was achieved, enabling Bi incorporation without droplet formation or surface roughening. X-ray diffraction from the same film indicates high structural quality and room temperature photoluminescence was also visible from this sample. However, it is clear that the low As 2 pressure required to incorporate Bi presents a major challenge in terms of film quality Bismuth incorporation Figure 5.6 presents x-ray diffraction patterns for three GaAs 1 x Bi x films of varying thickness with x equal to 0.5%, 1.9% and 3.0%. An AFM image of the surface for the 1.9% Bi film was presented in Figure 5.5, and the sample with x = 3.0% is a 30 nm thick quantum well with a 290 nm GaAs cap. All three films in Figure 5.6 show pendellösung fringes in the diffraction pattern, indicating that the interfaces are smooth and compositional inhomogeneities are minimal, a major improvement over films grown in the past that showed a variation in Bi content of up to 20% through the film thickness [WTC + 04]. This improvement was obtained by using LS to find

86 Chapter 5. Growth and properties of dilute bismides 71 Figure 5.5: Atomic force microscope image of a dilute bismide film (r1698) grown at 390 C. The image area is 2x2 µm, the vertical scale is 3 nm and the rms roughness is 0.2 nm. The highly smooth surface is attributed to a higher Bi flux than that used in the growths from Figure 5.4. growth conditions for which droplets are not present. However, despite this improvement, due to the numerous variables having a large effect on the incorporation of Bi into GaAs (i.e. growth temperature, Bi flux, As/Ga ratio), the Bi composition remains difficult to control. A clear trend is observed in the amount of Bi incorporated as a function of growth temperature (higher incorporation at lower temperature), and this trend is used by other groups to control Bi content in their films. Using the [Bi] vs. growth temperature data at a fixed As/Ga ratio published by Huang et al. [HOFY05], it is calculated that the activation energy for Bi incorporation is 1.34 ev, a value that is same as the activation energy for desorption as measured experimentally in this study and presented in Figure 4.6. From the data of Huang et al. it may also be noted that the Bi concentration begins to saturate with temperature at a concentration of 4.75%. These two facts suggests that the amount of Bi incorporation is simply related to the amount of Bi present on the surface at given temperature, and should be related to the ratio given by the Bi arrival rate over the sum of the arrival and desorption rates. This can be expressed as follows: F Bi [Bi] = F Bi + c exp[ U ] (5.1) kt where F Bi is the Bi arrival rate, c is a constant, and U is the activation energy for desorption. The c exp U term represents the desorption rate. Using the Bi flux of 3 kt x 10 8 Torr used by Huang et al. to calculate the arrival rate, and extrapolating to