Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Pattern formation in lateral oxidation of aluminum-rich AlxGa1-xAs Nicoll, Christine Anne 2000

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2001-0088.pdf [ 13.67MB ]
Metadata
JSON: 831-1.0085178.json
JSON-LD: 831-1.0085178-ld.json
RDF/XML (Pretty): 831-1.0085178-rdf.xml
RDF/JSON: 831-1.0085178-rdf.json
Turtle: 831-1.0085178-turtle.txt
N-Triples: 831-1.0085178-rdf-ntriples.txt
Original Record: 831-1.0085178-source.json
Full Text
831-1.0085178-fulltext.txt
Citation
831-1.0085178.ris

Full Text

PATTERN FORMATION IN LATERAL OXIDATION OF ALUMINUM-RICH Al xGai_ xAs by C H R I S T I N E A N N E N I C O L L B.Sc , University of Waterloo, 1997 A THESIS S U B M I T T E D IN PARTIAL F U L F I L M E N T O F T H E REQUIREMENTS FOR T H E D E G R E E O F M A S T E R O F S C I E N C E in T H E F A C U L T Y OF G R A D U A T E STUDIES D E P A R T M E N T OF PHYSICS A N D A S T R O N O M Y We accept this thesis as conforming to the required standard T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A November 2000 © C h r i s t i n e Anne Nicoll, 2000 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of Physics and Astronomy The University of British Columbia Vancouver, Canada Date Abstract A high-aluminum-content ALjGai -zAs (x > 0.9) layer in an epitaxial III-V semicon-ductor heterostructure can be preferentially oxidised from an exposed edge at 375-450C in a wet N 2 ambient. In this way, buried aluminum oxide layers can be formed which are useful for optical and electronic confinement in III-V devices such as V C S E L s and distributed feedback lasers. The rate of oxidation is strongly dependent on aluminum content and is higher in the (001) directions than the (Oil). We find that an angular dependence of the form a + bsin226 accurately describes the anisotropy of the oxidation rate. This anisotropy has the interesting consequence that oxidation proceeding outward from a pinhole evolves into a "rounded-square" shape, while oxidation inward from a circular perimeter produces a square shape. Because of the strong dependence of the oxidation rate on the aluminum content, possibilities exist for lithographic control of ox-idised areas by lateral modulation of the composition of the buried layer. The aluminum content of a thin buried aluminum-rich A l s G a i _ x A s layer can be reduced by intermixing with neighbouring GaAs layers, using selective-area ion implantation and rapid thermal annealing. This makes it possible to create relatively complex oxide patterns in buried A l x G a i _ x A s layers. 11 Contents Abstract ii List of Tables v List of Figures vi Acknowledgements x 1 Introduction 1 2 Lateral Oxidation of Buried Layers of A l x G a i _ x A s 3 2.1 Background 3 2.2 Oxidation Procedure 5 2.3 Samples 6 2.4 General Observations 9 2.5 Anisotropy in the Lateral Oxidation Rate 13 2.5.1 Observations and Modeling 13 2.5.2 Origin of the Anisotropy 18 2.6 Oxidation of Thin Layers 22 2.6.1 Dots 23 2.6.2 Square and Circular Etch Perimeters 24 3 Oxide Patterning by Selective Layer Interdiffusion 28 3.1 Layer Interdiffusion Background 28 3.2 Objective 29 3.3 Survey of Previous Work on Interdiffusion 30 3.3.1 Oxidation and Intermixing 30 3.3.2 D F B lasers with selective intermixing 31 3.4 Proposed Process for Patterning Using Ion-Implantation-Enhanced Diffusion 33 3.5 Conditions for Implantation-Induced Layer Intermixing 34 3.6 Samples 36 3.7 Diffusion 39 3.8 Ion Implantation 42 3.9 Anneal 42 3.10 Patterning and Oxidation 45 3.11 Oxidation Observations 45 3.11.1 R754 Implant 1 . 45 3.11.2 R754 Implant 2 47 iii C O N T E N T S iv 3.11.3 NRC2298 Implant 1 50 3.12 X-ray Diffraction 52 4 Conclusions 57 4.1 Wet Lateral Oxidation of A l ^ G a ^ A s 57 4.2 Use of Layer Interdiffusion to Control Oxidation Rate 59 Appendix A : Gold Masking for Selective Implantation 61 Appendix B: Wet Etching 65 B . l : G a A s / A l x G a ! _ x A s 65 B.2: InGaAs/InAlAs 66 B.3: Removal of Dielectrics 67 B.4: "Stain Etch" 67 Appendix C: Electron Cyclotron Resonance Etching 68 B i b l i o g r a p h y . . 70 List of Tables 2.1 Anisotropy observed in various samples 18 2.2 Summary of oxide widths obtained from 30-minute oxidations of samples containing thin layers of Al^Gai-^As. Distances were measured perpendic-ular to wet-etched trenches which were oriented parallel to the {011}-type (cleave) planes of the samples; that is, in the "slow" direction. These measurements represent the oxide formed to the outside of the perimeters which were outlined by the trenches, in order not to include possible stress effects observed on the insides of the closed paths. Where it was possible to determine the anisotropy, that information is included in Table 2.1. . . 23 4.1 S E M cross-sections of ECR-etched R783. In each case the 100 nm Alo.9gGao.02As layer has been oxidized. In the middle image, the GaAs cap has been com-pletely removed, exposing the entire Alo.9sGao.02As layer which has then oxidized. Microwave power was 100W, D C bias -100 V , process pressure lOmTorr, and backside He 5Torr 68 v List of Figures 2.1 Schematic of the oxidation furnace setup. The N 2 gas input is fed through two valves; one for unmeasured purging, the other with a mass flow con-troller for measured flow during oxidation. There is an empty trap bottle as security against the unlikely event of liquid water from the boiler con-taminating the nitrogen line. On the output a similar trap prevents water being drawn back into the cooling oven from a flask of water whose purpose is to capture any arsine gas which may be present in the oven exhaust. . 5 2.2 Layer structures of samples used in this chapter. Layer thicknesses for all samples except R676 were determined using X-ray diffraction 7 2.3 X-ray diffraction data and simulations for samples R783, R754, and R782, used in Chapter 2. The layer thicknesses for these samples, as illustrated in Figure 2.2, were found by comparing simulated X-ray diffraction profiles with X R D data. Thus the simulated profiles shown here correspond to the layer structures in Figure 2.2 8 2.4 Scanning electron micrographs of ECR-etched trenches with unknown un-oxidized sidewall material. Both images are from a single sample that underwent a single etch, using a plasma'of Ar and BCI3, followed by oxi-dation at 417C for 20 minutes 10 2.5 Scanning electron micrograph of oxide propagating from an accidental "pinhole" through the GaAs cap of R782. Sample was oxidized at 417C for 60 minutes. Image edges are parallel to cleave axes 11 2.6 A F M surface height images showing the existence of two distinct regions within the oxidized part of a buried layer of Alo.9sGao.02As. (a) A sample which contained 500nm-thick A ^ G a ^ A s , oxidized at 425°C by Frangois Sfigakis. The outside perimeter of the oxidized area is marked by a drop in height of ~25-30nm, or ~5-6%, but another short drop occurs at the edge of a central region whose height decreases gradually toward the pinhole, (b) Sample R783, containing a lOOnm layer of Alo.9sGao.02As, oxidized for one hour at 417°C. Two regions are again present; the oxide/semiconductor boundary is marked by a vertical drop of ~3nm, and the transition into the second region by a fairly sudden drop of ~2.5nm followed by what appears to be a gradual downward slope. The A F M scan for (a) was taken by Martin Adamcyk 12 2.7 Geometric basis for modeling the oxidation of a buried A l x G a i _ x A s layer. The oxidation rate, ^p-, depends on the orientation of the oxide-semiconductor interface relative to the crystallographic axes 14 vi L I S T O F F I G U R E S vii 2.8 Numerically-generated plots illustrating the evolution of the shape of the oxide front for (a) oxidation outward from a round hole, and (b) oxidation inward from the edge of a round mesa. The "pinhole" oxide evolves into a "rounded-square" shape, while the mesa shape eventually generates a square aperture of unoxidized material in the centre 15 2.9 Simulation results overlaid on an S E M image of R676 showing regions where the buried Alo.9sGao.02As layer has oxidized from etched pinholes in the GaAs cap. The inner circle in each simulation represents the initial outline given to the computer program, and the outer perimeter represents the computed oxidation boundary. Dark outlines were generated using ^ = 1.15, lighter outlines using ^ = 1.20. The sample contains a 450-nm-thick layer of Alo.9sGao.02As and was oxidized at 425°C for 6 minutes. . . 16 2.10 Simulated oxidation from etched trench patterns, with anisotropy 1.1. The outer perimeter represents the final shape of the oxidized area result-ing from oxidation beginning at the etched edge delineated by the inner perimeter 17 2.11 Nomarski micrographs of actual oxidation from etched trench patterns similar to those in Figure 2.10. Sample R783 (lOOnm Alo.9sGao.02As), oxidized for 20 minutes at 417C. The lighter area is the oxide region. Two slightly differently-toned areas are visible in the oxide 17 2.12 Schematic cross-sections of a partially-oxidized buried layer of A l ^ G a ^ ^ A s on a (110) GaAs substrate, for oxidation propagating in opposite <001> directions, (a) A curved oxide front propagating in the [001] direction can "see" metal-terminated {111}A planes from which it is easy to remove material if the A l content x is high enough, (b) A curved oxide front propagating in the [001] direction has no access to {111}A planes 20 2.13 Schematic cross-sections of a partially-oxidized buried layer of A l x G a i _ x A s on a (110) GaAs substrate, for oxidation propagating in perpendicular <011> (slow) and <001> (fast) directions, (a) A curved oxide front propa-gating in the [Oil] direction can "see" one metal-terminated {111}A plane, but removing material from the {111} A plane will rapidly expose a {111}B plane, slowing the oxidation, (b) A curved oxide front propagating in the [001] direction has no access to either {111}A or {111}B planes 21 2.14 Nomarski optical micrograph of regions of (a) the 25 nm Alo.9sGao.02As layer of R782, and (b) (inset) the 24 nm AlAs layer of R754, oxidized at 4 0 0 ° C for 30 minutes through small wet-etched holes in the GaAs cap. . 24 2.15 (a) R754 (24nm AlAs), and (b) R782 (25nm Alo.98Gao.02As), oxidized at 3 7 5 ° C for 30 minutes 25 2.16 (a) R754 (24nm AlAs), and (b) R782 (25nm Alo.9sGao.02As), oxidized at 400°C for 30 minutes 27 L I S T O F F I G U R E S viii 3.1 X R D 6 — 29 scan of sample NRC2298, with simulation results overlaid. . 37 3.2 (a) S E M cross-section image of sample NRC2298, stain-etched to empha-size interfaces, and (b) schematic of layer structure, as determined by X R D . 38 3.3 Maximum remaining A l concentrations within high-x A ^ G a ^ ^ A s vs. dif-fusion length L for R754 and NRC2298, using simple Fickian diffusion model 40 3.4 A l profiles using error-function model for intermixing of high-x A^Gax-^As layers in (a) R754 and (b) NRC2298. Straight vertical lines indicate bound-aries of the layer as grown 41 3.5 X P S results for two samples annealed at (a) 800°C and (b) 840°C, without a pre-anneal purge to prevent oxidation damage to the surface during the anneal 44 3.6 T R I M simulation for vacancy and ion distributions for the conditions of R754 Implant 1 46 3.7 Optical micrographs showing oxidation from two regions of R754 Implant 1, annealed at 940°C for 60 seconds: a) No implant; b) 2x l0 1 4 ions/cm 2 . The dark lines are etched trenches designed to expose the AlAs for oxida-tion. The light-coloured region is where the buried AlAs has oxidized. . . 47 3.8 T R I M simulation for vacancy and ion distributions for the conditions of R754 Implant 2 48 3.9 Optical micrograph showing the trend in oxidation widths for R754 Im-plant 2 annealed at 945°C for 60s: (a) 2xl0 1 4 ions/cm 2; (b) 5x l0 1 4 ions/cm 2;(c) 7.5xl0 1 4 ions/cm 2; (d) l x l O 1 5 ions/cm 2 . The dark lines are etched trenches designed to expose the AlAs for oxidation. The light-coloured region is where the buried AlAs has oxidized. The most heavily-implanted region (d) exhibits almost no oxidation. Dark specks on surface are contamina-tion and are not intrinsic to the process 49 3.10 T R I M simulation for vacancy and ion distributions for the conditions of NRC2298 Implant 1 50 3.11 Optical micrographs of etch pits in NRC2298 after implantation and an-neal at 940°C for 60s. Sample was etched for 20 minutes in a solu-tion of 6:1:40 citric acid (50% by weight):H 2O 2(30%):H 2O. (a) 2x l0 1 3 ions/cm2;(b) 5xl0 1 3 ions/cm2;(c) l x l O 1 4 ions/cm2;(d) 2xl0 1 4 ions/cm2;(e) 5x l0 1 4 ions/cm 2 51 3.12 Effect of increasing implant dose on X R D from unannealed R754, Implant 1 (95 keV Si) 53 3.13 X-ray rocking curves of different regions of R754 Implant 1, before and after Anneals 1 and 2 (both 875°C, 30s). Baselines are offset for clarity. . 55 3.14 X-ray rocking curves of different regions of R754 Implant 1 Anneals 1 and 2 (both 875°C, 30s). Baselines are offset for clarity. As the dose increases, the fringes become more washed-out 56 L I S T O F F I G U R E S 4.1 Cross-sectional S E M images of gratings etched into GaAs with a mask of SiCv Gas flows in seem, DC bias voltages, and etch time are recorded in the figure. Microwave power was 100W, process pressure lOmTorr, and backside He 5Torr. Acknowledgements I would like to acknowledge collaborators outside of U B C : Zbigniew Wasilewski of N R C , who grew sample NRC2298; Mark Ridgway of the Australian National Univer-sity, who performed the ion implantation for the intermixing experiments; and Karen Kavanagh and Victoria Fink of Simon Fraser University, who performed the T E M . I appreciate your generosity of time, resources, and enthusiasm. Thanks to Yuval Levy and Martin Adamcyk for growing the U B C M B E samples. One must of course acknowledge one's supervisor, but I want to thank Tom Tiedje for some particular supervisory qualities. Thanks for being supportive but not heavy-handed, for being so available for discussions of physics and technical details alike, and for illustrating what those mountains out the window are for. It would be impossible to properly acknowledge every occasion where one of my col-leagues in the MBElab has taught me to use equipment or spent time pondering a problem with me. It was a luxury to have been surrounded by such intelligent, versatile, and gen-erous people. Thanks to Anders Ballestad, Ben Ruck, Jens Schmid, Jim MacKenzie, Martin Adamcyk, Sayuri Ritchie, Sebastien Tixier, Tom Pinnington, and Yuval Levy. I learned as much from your ways of problem-solving as I did from the specific knowledge you shared with me. Acknowledgement is due outside the MBElab, particularly to A l Schmalz for training, advice and maintenance and modification of equipment, and to the members of Jeff Young's lab. Thanks to Vighen Pacradouni, Jody Mandeville, and Alex Busch for sharing your expertise, and to Jeff Young for making the time to read and edit my thesis. To complete a circle, thanks to Frangois for teaching me so much about processing when I arrived at U B C , and now for being such a support in the final push to complete my thesis. x Chapter 1 Introduction High-aluminum-content A l x G a i _ x A s , when exposed to water vapour at elevated tem-perature, oxidizes readily, while GaAs and low-aluminum-content A l x G a i _ x A s does not. The product is an insulating oxide with a refractive index which is low in the infrared (n « 1.6 [1]) compared with that of GaAs (n as 3.4 - 3.6 [2]). As A l x G a i _ x A s and GaAs are nearly lattice-matched, epitaxial layers of both materials can be included in a single-crystal structure with very little strain. One can create a region of oxide, buried within a device structure, simply by exposing an edge of a buried A l I G a 1 _ I A s layer. This has afforded a flexibility in the design of GaAs-based devices requiring a means of current or optical mode confinement, notably vertical cavity surface-emitting lasers (VCSELs) . Because the buried oxide forms laterally from an exposed edge, desired regions of an Al^Gai-aAs layer can be oxidized by etching patterns through the layers above and subjecting the sample to oxidizing conditions for long enough to allow the oxide formation to propagate for the necessary lateral distance. It would be even more advantageous to be able to arbitrarily pattern the oxide in the plane of the A l x G a i _ x A s layer, without the necessity of delineating the pattern by an etch, creating trenches through the layers above. With lateral control over the aluminum content x, the selectivity of the oxidation rate in favour of high x could be used to choose which portions of the A l x G a i _ x A s layer oxidize, in the same way that the vertical composition control intrinsic to epitaxial growth constrains the oxidation to within the layer in the first place. Fine lateral control of oxidation would be of use in the fabrication of high-index-contrast optical gratings for 1 C H A P T E R 1: I N T R O D U C T I O N 2 distributed-feedback (DFB) lasers and photonic crystals. One way to effect the lateral patterning of aluminum content is by "selective layer interdiffusion." In this process the diffusion of atoms during a high-temperature anneal is locally enhanced by one of several methods, so that atoms from adjacent layers become intermixed to form a material of intermediate composition. The use of interdiffusion to induce post-growth changes in the composition of an epitaxial structure has been studied for years and has numerous applications, including the integration of different photonic components on a single wafer [3, 4, 5], altering of quantum well profiles [6], tuning of the resonant frequencies of distributed Bragg reflectors (DBR) [7], and the patterning of alloy composition within a plane to delineate waveguides [6], D F B gratings [8, 9, 10, 11, 12, 13], quantum dots, or quantum wires [14, 15]. In this work two methods are examined for the patterning of a buried A l x G a i _ x A s oxide: lateral oxidation from patterns of etched trenches and pattern formation using ion implantation induced selective layer interdiffusion. Chapter 2 is dedicated to the study of the oxidation of buried Al^Gai-^As layers from etched patterns. Observations of the oxidation of several samples are discussed. The time evolution of the shape of the oxidation front is simulated numerically, using a mathematical model for the crystalline anisotropy observed in the oxidation rate. Results of the model are compared with experiment. In Chapter 3 a process is proposed for the patterning of a buried high-index-contrast grating by ion implantation induced interdiffusion and wet lateral oxidation. A survey of relevant work in the literature is given. Optical and X-ray diffraction observations of samples at various stages in this process are presented. Some specific processing techniques are discussed in the Appendices, for reference. Chapter 2 Lateral Oxidation of Buried Layers of Al x Gai_ a As 2.1 Background The "wet, lateral" oxidation of buried layers of A l x G a 1 _ x A s with high aluminum content a; as a deliberate process has been investigated extensively since the desirable qualities of the resulting oxide were discovered during studies of the degradation of AlGaAs [16]. If a sample containing such a layer, with an edge exposed by etching or cleaving, is subjected to elevated temperatures in an ambient of N 2 and water vapour, the aluminum in the layer will begin to react with the H 2 0 to form an oxide. Oxide formation proceeds laterally through the layer from the exposed edge. The structure and properties of oxidized material depend upon many material and process parameters. As a layer of A ^ G a ^ A s is converted to oxide, a volume con-traction occurs, resulting in the reduction of layer height which is detectable by S E M and A F M . The lateral shrinkage of the material is somewhat constrained by the GaAs layers above and below, but, especially for thick layers, we have seen buckling of the entire structure [17]. Vertical contraction has been quoted at values of ~10-13% for AlAs [18, 19, 20], - 5 % [21] for Alo.9sGao.02As, and 6.7% for Alo.92Gao.osAs [21], for ex-ample. We observed a vertical drop of 5.4% at the location of the oxide/semiconductor interface by atomic force microscopy (AFM) on a sample containing a 500nm-thick layer of Alo.9sGao.02As [17]. The inclusion of gallium in the original crystal diminishes the con-traction significantly, but in all cases the contraction is significantly less than what would be expected if the material were converted to crystalline AI2O3. Even assuming the layer 3 C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A l x G a i _ x A s 4 were free to contract in all three dimensions, there should be a vertical contraction of ~20%. On the other hand, if the product of oxidation were crystalline A l ( O H ) 3 , only a ~2% height difference would be observed with shrinkage in three dimensions, or ~6.4% if all the volume reduction occurred through vertical shrinkage [17]. Indeed, the oxide is not for the most part crystalline. It is mostly amorphous, perhaps containing nanocrystallites ~4 nm [21] to ~8nm in diameter [22] of 7-AI2O3. Much of the microstructural information has been due to T E M observation. It is worth keeping in mind that ion milling during T E M sample preparation may cause changes in the sample, and densification of the oxide material has been observed under electron beam exposure [23, 24]. We found no evidence of crystallinity using X-ray powder diffraction, indicating that if there are crystals present, they are very small [1]. Devices made by oxidizing high-a; A l x G a i _ x A s rather than pure A lAs are more me-chanically robust and stable under thermal cycling [18]. It is reasonable to infer that such devices are under less strain at the interfaces, in view of the less severe contraction evident with even a very small gallium content. Much work has been done to try to quantify the dependence of oxidation rate on temperature, aluminum content, and layer thickness. See, for example, references [23, 25, 20, 26]. It is known that the rate of oxidation increases exponentially with temperature, and the higher its aluminum content x, the more readily Al^Ga^-cAs material will oxidize. Pure AlAs will oxidize significantly at temperatures lower than 370°C. A l x G a i _ x A s with x as low as 0.5 can be oxidized at temperatures of 500-525°C [27]. For thin layers the oxidation rate diminishes, and this dependence becomes very strong for layers with thicknesses of several tens of nanometers and less. Oxidation rates have been reported both to vary [26, 23] and not to vary [20] with N 2 flow rate. If the H 2 0 concentration of the oxidizing atmosphere is not the same in all groups' work, this may explain some of the differences among their data. It has been observed in the literature [18, 23, 28, 29, 30], particularly where cylindrical mesas are oxidized to form current apertures for V C S E L s [18, 28], that oxidation of very C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S O F A l x G a ! _ x A s 5 high-x A l x G a i _ x A s is noticeably anisotropic. It is interesting to note that attempts have been made to model quantitatively the oxidation rate for cylindrical mesas, without taking the anisotropy into account [28, 31]. 2.2 Oxidat ion Procedure The furnace used in this study was designed and built by Vighen Pacradouni and Frangois Sfigakis for the wet lateral oxidation of A l x G a i _ x A s and Ini_ yAlj,As, as an improvement on an earlier design by Peng Chen and Tom Tiedje. Figure 2.1 is a schematic diagram of this furnace. Further details can be found in the M.Sc. thesis of Frangois Sfigakis [17]. Figure 2.1: Schematic of the oxidation furnace setup. The N 2 gas input is fed through two valves; one for unmeasured purging, the other with a mass flow controller for measured flow during oxidation. There is an empty trap bottle as security against the unlikely event of liquid water from the boiler contaminating the nitrogen line. On the output a similar trap prevents water being drawn back into the cooling oven from a flask of water whose purpose is to capture any arsine gas which may be present in the oven exhaust. The oxidation process is carried out as follows: Once the sample has been loaded, the furnace is purged with dry N 2 at room temperature for at least one hour to remove 0 2 and H 2 0 . The use of 0 2 rather than N 2 as a carrier gas for water vapour has been C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF Al^Gai-^As 6 demonstrated to halt oxidation altogether. This may be related to the reaction of O 2 with available hydrogen to form water, preventing the formation of volatile As from A S 2 O 3 [32]. If H 2 0 is present, some oxidation may occur as the furnace temperature is ramped up. The purge is thus a precaution to ensure repeatable starting conditions. The furnace and water flask are brought to their intended process temperatures, each with an independent flow of dry N 2 . The furnace temperature during oxidation is typically ~ 3 7 0 - 4 2 0 ° C and the water is kept near 95°C. Once the boiler and furnace temperatures are stable, the dry N 2 input to the furnace is replaced by 100 seem of N 2 bubbled through the water in the boiler. The time of this switch is designated the beginning of the oxidation process. When the desired oxidation time has elapsed, the H 2 0-laden N 2 supply to the furnace is replaced by a vigorous flow of dry nitrogen. Power to all heating elements is cut at this time and the system is allowed to cool to near room temperature before the dry nitrogen flow is interrupted and the sample exposed to atmosphere. This cooling takes place over a span of several hours. The easiest way to observe the results of the oxidation process is usually to use a Nomarski Differential Interference Contrast (DIC) optical microscope. The difference between unoxidized AlGaAs and oxide is clearly visible through the GaAs cap layer that prevents the AlGaAs layer from oxidizing from the top down. Optical microscope images were acquired using either a conventional Olympus film camera or a S P O T digital camera. Image length scales were calibrated by taking images of a Digital Instruments atomic force microscope (AFM) calibration reference, consisting of a 10- ^m-spaced grid of depressed squares in platinum-coated silicon. 2.3 Samples The samples studied in this chapter were grown by Molecular Beam Epitaxy (MBE) at U B C , by Dr. Yuval Levy and Ph.D. student Martin Adamcyk. Each sample contains a layer of either pure AlAs or Alo .9 8 Ga 0 .o2As. Figure 2.2 illustrates the layer structures of C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A ^ G a ^ A s 5nm GaAs 5nm A l n 1 s G a n a s A s 5x -CI GaAs substrate R 6 7 6 5.3nm GaAs 5.3nm A I 0 2 G a 0 8 A s } 5 x - C 100nm A l 0 9 g G a O 0 2 , A s GaAs substrate R 7 8 3 24nm AlAs" 133nm GaAs GaAs substrate R 7 5 4 25nm_ 95nm GaAs 138nm GaAs 1500nm A I 0 5 7 G a 0 4 3 A s GaAs substrate R 7 8 2 Figure 2.2: Layer structures of samples used in this chapter. Layer thicknesses for all samples except R676 were determined using X-ray diffraction. the M B E growths R676, R754, R782, and R783. X-ray rocking curve or 9 —26 measurements were made using a Bede Scientific double-axis diffractometer to determine the layer thicknesses of samples R754, R782, and R783. Figure 2.3 shows the measured X-ray diffraction (XRD) data as well as simulated curves C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A l x G a i _ x A s 8 Figure 2.3: X-ray diffraction data and simulations for samples R783, R754, and R782, used in Chapter 2. The layer thicknesses for these samples, as illustrated in Figure 2.2, were found by comparing simulated X-ray diffraction profiles with X R D data. Thus the simulated profiles shown here correspond to the layer structures in Figure 2.2. C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A ^ G a ^ A s 9 generated using Philips' HRS software. The simulated curve from the R782 structure contains very fine oscillations from the thick Alo .57Gao.43As layer it contains. Since the diffractometer resolution is not sufficient to see these fringes, the simulated curve was convolved with a resolution function of 30 arcseconds full width at half maximum, before comparison with data. 2.4 General Observations Oxidation from an exposed, wet-etched trench or ridge sidewall is fairly reliable and uniform. However, oxidation from cleaved edges was found to be unpredictable and non-uniform [17]. Patterns made with electron cyclotron resonance (ECR) etching using B C I 3 , C l 2 , and Ar were observed to oxidize unpredictably. This may be due to the coating of sidewalls with GaAs transported from the floor of the trench, which may effectively seal the edge of the aluminum-containing layer from contact with oxidants. Figure 2.4 shows scanning electron micrographs of R783 with its GaAs cap removed by E C R etching and the exposed Alo.98Gao.02As oxidized from the top down. It is apparent that the walls of the etched trench are of a substance that did not oxidize. If the GaAs cap layer had been present, the Alo.9sGao.02As would have been sealed off from the oxidizing atmosphere. Not surprisingly, oxidation from the edge of an etched pattern or cleave tends to progress further from any point along the edge which appears to be delaminating locally. In this case the portion of the AlGaAs material under the delaminating region would be exposed all at once to H 2 0 vapour; the oxidation for that small region would be planar rather than lateral. Figure 2.5 is a scanning electron micrograph of a sample of R783 where a region of the buried Alo.9sGao.02As layer has oxidized from an accidental "pinhole" through the GaAs cap. The oxide extends noticeably further in the <001> directions than the <011> directions. There are also two distinct regions visible within the oxidized area in this S E M C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A l x G a i _ x A s 10 1 8 4 9 2 6 1 5 . 8 k y X 8 8 . *fc" ' * 3 > S J r » i Figure 2.4: Scanning electron micrographs of ECR-etched trenches with unknown unox-idized sidewall material. Both images are from a single sample that underwent a single etch, using a plasma of Ar and BC1 3 , followed by oxidation at 417C for 20 minutes. image: a lighter-looking region appears along the perimeter of the oxide, nearest the front, separated by a sharp boundary from the darker remainder. This is a typical C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A l x G a i _ x A s 11 1 0 4 8 7 5 2 0 . 0 k V X 3 . 0 6 iC* * I « * « V i Figure 2.5: Scanning electron micrograph of oxide propagating from an accidental "pin-hole" through the GaAs cap of R782. Sample was oxidized at 417C for 60 minutes. Image edges are parallel to cleave axes. observation among the samples in this work. Atomic force microscope (AFM) measure-ment shows that although the Alo.98Gao.02As oxide layer in the region closest to the oxide/semiconductor boundary is thinner than the original unoxidized material, the re-gion closer to the hole from which the oxidation originated is thinner yet. A n example is shown in Figure 2.6, which gives A F M scans of oxidized regions of samples containing (a) 500nm and (b) lOOnm of Alo.9sGao.02As. A large body of oxidation work has been carried out at Sandia National Laboratories. That group proposes a two-step reaction process which is consistent with their observa-tions and which may hold the explanation for our observation of two distinct regions. Using T E M , they have imaged an amorphous region of ~17nm width, distinct from the remaining material which appeared to contain small crystals, at the reaction front of an 84 nm layer of Alo.9sGao.02As oxidized for 40 C [21]. Their Raman studies indicate that during oxidation at high temperature (450°C) , both crystalline and amorphous As as well C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A l x G a ! _ x A s 12 (a) 500nm A l 0 9 8 G a 0 0 2 A s (b) 10Onm A l 0 9 8 G a 0 0 2 A s Figure 2.6: A F M surface height images showing the existence of two distinct regions within the oxidized part of a buried layer of Alo.9sGao.02As. (a) A sample which contained 500nm-thick A l x G a i _ x A s , oxidized at 425°C by Frangois Sfigakis. The outside perimeter of the oxidized area is marked by a drop in height of ~25-30nm, or ~5-6%, but another short drop occurs at the edge of a central region whose height decreases gradually toward the pinhole, (b) Sample R783, containing a lOOnm layer of Al 0 .9 8 Ga 0 .o2As, oxidized for one hour at 417°C. Two regions are again present; the oxide/semiconductor boundary is marked by a vertical drop of ~3nm, and the transition into the second region by a fairly sudden drop of ~2.5nm followed by what appears to be a gradual downward slope. The A F M scan for (a) was taken by Martin Adamcyk. as C1-AS2O3 [33] are present, and that their levels remain relatively constant during the oxidation process [32, 23]. At lower temperatures, up to 425°C, the A s 2 0 3 level is below the detection limit. A series of reactions for the oxidation process was proposed [33], wherein the AS2O3 is produced in the initial oxidation reaction, and subsequently con-verted to volatile As which then escapes, leaving behind a porous, amorphous aluminum oxide. Under conditions which lend themselves to faster oxidation, for example high temperature or high aluminum content, oxidation rates have been observed to shift from linear to parabolic [20, 28]. It is possible that under these conditions AS2O3 is produced faster than it is removed and that it forms a barrier slowing the transport of reactants to the oxidation front, accounting for the observed parabolic time dependence of the reaction [33]. Further evidence from Raman spectra [34] indicates that for planar wet C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A ^ G a ^ A s 13 oxidation at least, the first aluminum oxide produced is amorphous but crystallizes upon further heating. Perhaps the two distinct regions we have observed are due to this effect; the thicker region may be an example of As 2 0 3 - r i ch amorphous material, while the thinner region further from the front may be a zone where the AS2O3 has been removed by its conversion to a more volatile substance which then escapes. The gradual downward trend in the surface height toward the exposed edge of the layer means that the oldest oxide is also the thinnest. It may be that the originally-formed porous oxide material is Gibbsite (A1(0H) 3) that undergoes densification as it dehydrates under continued exposure to heat. This may occur continuously during the oxidation, or perhaps the second, thinner phase of oxide is formed during the cooling-off period after the H 2 0 supply is cut to halt the oxide formation. 2.5 Anisotropy in the Lateral Oxidation Rate 2.5.1 Observations and Modeling In this section we describe a mathematical model for the lateral oxidation rate of high-x A L j G a i _ x A s layers on (100) GaAs substrates. The model is linear in time, and is the simplest with the symmetry to describe the observed crystalline anisotropy. The rate of oxidation is characterized in terms of the instantaneous, local rate of progress of the oxide in the direction perpendicular to the boundary between oxidized and unoxidized regions. The rate depends upon the orientation of this boundary (or "oxidation front") at each instant in time. The mathematical formulation of the model is as follows: d ^ l = a + bsin2(2e) (2.1) where is the rate of progress of the interface, along its outward normal, n(9), at C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S O F A l x G a i _ x A s 14 Figure 2.7: Geometric basis for modeling the oxidation of a buried A l x G a i _ x A s layer. The oxidation rate, ^p-, depends on the orientation of the oxide-semiconductor interface relative to the crystallographic axes. a given time £; 0 is the angle between the direction of propagation and the [Oil] direction; a and b are constants. This model was implemented as a computer program written in M A T L A B code, with a set of points delineating the initial exposed edge of the A l x G a i _ x A s layer from which oxidation can proceed. In each time increment the angle 6 is calculated at intervals around the perimeter of the front, and the front propagates by the corresponding perpendicular distance at each location, using the model. The oxide shape is allowed to evolve in this way over numerous time steps. Figure 2.8 shows two examples of successive oxidation fronts generated by the sim-C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S O F A l x G a i _ x A s 15 Figure 2.8: Numerically-generated plots illustrating the evolution of the shape of the oxide front for (a) oxidation outward from a round hole, and (b) oxidation inward from the edge of a round mesa. The "pinhole" oxide evolves into a "rounded-square" shape, while the mesa shape eventually generates a square aperture of unoxidized material in the centre. ulation, for oxidation outward from a small circular "pinhole" and for oxidation inward from the edge of a circular mesa. Notice the similarity of the final outline of the simu-lated "pinhole" oxide to that of the observed shape in Figure 2.5. The mesa oxidation evolves toward a square unoxidized aperture, which is what has been observed in work on cylindrical VCSEL-type structures [18, 28]. The model has been applied to characterize the anisotropy of "pinhole" oxidation for several samples. Defining "anisotropy" to be the ratio of the extent of oxidation in the <001> (fast) directions to the rate in the <011> (slow) directions, we can write, in terms of our model: anisotropy = ° ~*~ (2.2) a Figure 2.9 shows the simulation results overlaid on the oxidation pattern observed by C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A l x G a i _ x A s 16 Figure 2.9: Simulation results overlaid on an S E M image of R676 showing regions where the buried Alo.9sGao.02As layer has oxidized from etched pinholes in the GaAs cap. The in-ner circle in each simulation represents the initial outline given to the computer program, and the outer perimeter represents the computed oxidation boundary. Dark outlines were generated using ^ = 1.15, lighter outlines using ^ = 1.20. The sample contains a 450-nm-thick layer of Alo.9sGao.02As and was oxidized at 425°C for 6 minutes. S E M in R676. The numerically-generated outlines fit the oxide shapes quite well, and the value for the anisotropy in this sample was found to be in the range of 1.15 to 1.20. Table 2.1 summarizes the anisotropy measured for several oxidized samples. It is difficult to say from these data whether layer thickness or compositional differences be-tween AlAs and Alo.9sGao.02As make a significant difference, though Choquette et al. have found that increasing G a mole fraction reduces the anisotropy, until for x < 0.94 it is no longer observable [18]. It does appear that our highest-temperature oxidation resulted in a slightly larger anisotropy, and this is consistent with observations in the literature that the anisotropy increases with temperature [29, 30]. As observed in the pinhole and mesa oxidation simulations, oxidation outward from C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF Al z Gai_ x As 17 [011] -[011] Figure 2.10: Simulated oxidation from etched trench patterns, with anisotropy 1.1. The outer perimeter represents the final shape of the oxidized area resulting from oxidation beginning at the etched edge delineated by the inner perimeter. Figure 2.11: Nomarski micrographs of actual oxidation from etched trench patterns simi-lar to those in Figure 2.10. Sample R783 (lOOnm Al 0 .98Ga 0.o2As), oxidized for 20 minutes at 417C. The lighter area is the oxide region. Two slightly differently-toned areas are visible in the oxide. C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S O F Al^Gai-^As 18 Sample ID A L r G a i - s A s Layer Thickness (nm) X Oxidation Temperature (C) Duration of Oxidation (min) Anisotropy a+b a R676 450 0.98 425 6 1.15-1.20 R783 100 0.98 400 60 1.05-1.11 R782 25 0.98 375 30 - 1 . 0 8 R754 24 1 400 30 1.05-1.12 Table 2.1: Anisotropy observed in various samples. convex curves evolves toward larger "rounded-square" curves, while in the case of oxida-tion from concave curves, the oxide in the "fast" directions catches up with that in the "slow" directions, forming sharp inward corners which propagate indefinitely. The oxide appears, then, to propagate essentially linearly in time, allowing the lateral transfer of features in the oxide front. This property was explored further with the oxidation from etched trench patterns. Figure 2.10 shows the initial outline of two etched trenches as well as the final simulated oxide edge, generated numerically using an anisotropy of 1.1. Figure 2.11 shows Nomarski micrographs of actual oxidized samples beginning with similar etched trenches, bearing a strong resemblance to the simulations in Figure 2.10. The persistence of sharp inner corners along the lateral oxidation front is enhanced in this case by the anisotropy of the oxidation rate. If it can be scaled down sufficiently, to define features with ~ 4 0 0 nm pitch, this property could be useful in defining side-coupled D F B gratings on a ridge waveguide laser. 2.5.2 Origin of the Anisotropy The origin of the anisotropy has been attributed to differences in the reactivity of different GaAs crystallographic planes [18]. Along any <111> direction are alternating layers of Group III and Group V atoms. These layers are arranged in pairs: within a pair, each atom is bonded to three atoms in the adjacent layer. However, there is only one bond C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A l x G a i _ x A s 19 per atom connecting the bottom layer of a pair with the top layer of the next pair down. Denoting Group III layers by the letter A and Group V layers by B, the sequence of layers seen along the [111] direction (and its equivalents) is then B A - B A - B A , and so on, and in the [111] direction it is A B - A B - A B . Thus the smooth, planar removal of material in the [ i l l ] direction requires the breakage of three bonds per As atom removed, whereas in the [111] direction it is three bonds per Ga atom and only one bond per As atom. A good discussion and illustration of this can be found in Reference [35]. We expect the process of lateral oxidation of A l x G a i _ x A s to be closely analogous to anisotropic chemical etching; in GaAs, many chemical etchants will etch the {111}B (arsenic-terminated) planes in preference to the {111}A (gallium-terminated) planes, resulting in faceted Ga-terminated sidewalls [36]. However, if the addition of gallium to the lattice causes the anisotropy to disappear, this is a strong indication that it is important to consider the differences between GaAs and AlAs before carrying the analogy too far. Studies of oxidation of AlAs layers grown on variously-oriented GaAs substrates by Koizumi et al. [29] and Vaccaro et al. [30] are very illuminating in this respect. In these works it was found that the aluminum-terminated { l l l }A-type planes oxidize faster than the arsenic-terminated {111}B planes [30], which is not surprising considering that AlAs is much more reactive in oxygen than is GaAs. For samples grown on (100) substrates they observed the same four-fold symmetry that we find in our oxidized samples, but Vaccaro et al. found that in a layer grown on a (110) substrate, a nearly trigonal symmetry was revealed. Although in (lOO)-oriented samples all the <011> directions are equivalently slow to oxidize and all <001> directions equivalently fast, in a (UO)-oriented layer the [001] direction is one of the fastest and the [001] direction one of the slowest to oxidize. Thus, the crystal symmetry direction of the propagation of the oxide is not the only deciding factor in the anisotropy in oxidation rates; the orientation of the A l x G a i _ x A s layer also has an effect. This can be better understood by noting that the A l x G a i _ x A s - o x i d e / A l x G a i _ x A s boundary has been widely reported to be convex when viewed in cross-section [25, 19] C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A l x G a i _ x A s 20 Figure 2.12: Schematic cross-sections of a partially-oxidized buried layer of A l x G a i _ x A s on a (110) GaAs substrate, for oxidation propagating in opposite <001> directions, (a) A curved oxide front propagating in the [001] direction can "see" metal-terminated {111}A planes from which it is easy to remove material if the A l content x is high enough, (b) A curved oxide front propagating in the [001] direction has no access to {111}A planes. (i.e. the oxide extends further into the centre of the A l x G a i _ x A s layer than it does at the top and bottom interfaces). This oxide "tongue" has a sharper curvature for thinner layers and it has been proposed that this is due to a surface energy effect that that inhibits oxidation at smaller layer thicknesses [25]. The exposed GaAs surface is thus not necessarily a plane oriented perpendicular to the A l x G a i _ x A s layer, but may be composed of a combination of orientations, largely those which are tilted upward or downward from the direction of oxide propagation. We first consider a layer of A l x G a ! _ x A s grown on a (110) GaAs substrate. If the oxi-dation front in the [001] direction is tilted, convex, or concave, two aluminum-terminated { l l l}A-type planes can be exposed, allowing quick oxidation (see Figure 2.12 (a)). On the other hand, if the oxidation front in the [001] direction is tilted, convex, or concave in cross-section (Figure 2.12 (b)), two arsenic-terminated {l l l}B-type planes may be exposed, but no {111}A planes are accessible. On a (lOO)-oriented sample, however, no {lll}-type planes at all can be "seen" by tilting up or down from any <001> direction, C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A h - G a ^ A s 21 Figure 2.13: Schematic cross-sections of a partially-oxidized buried layer of A l ^ G a i - z A s on a (110) GaAs substrate, for oxidation propagating in perpendicular < 0 l l > (slow) and <001> (fast) directions, (a) A curved oxide front propagating in the [Oil] direction can "see" one metal-terminated {111}A plane, but removing material from the {111}A plane will rapidly expose a {111}B plane, slowing the oxidation, (b) A curved oxide front propagating in the [001] direction has no access to either {111}A or {111}B planes. C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S O F A ^ G a ^ A s 22 and so neither [001] nor [001] direction has any obvious advantage over the other. This can explain the observation that there is a difference in rates between [001] and [001] directions for (HO)-oriented samples but not for (lOO)-oriented samples. Next we look at oxidation in <011> (slow) directions within layers grown on (100) substrates. If the cross-section of the interface is sloped upward, it will expose a different type of {111} plane than if it is sloped downward (see Figure 2.13). If some region of {lll}A-oriented surface is exposed, this may oxidize quickly, but that will expose a {111}B surface, retarding the oxidation from that point on. Thus the <011> directions are at a particular disadvantage as compared to the <001> directions, which may account for the anisotropy observed in (lOO)-oriented samples. A n example of an oxidation front in a 500nm-thick layer of Alo.9gGao.02As can be seen in the scanning electron micrograph on page 47 of the M.Sc. thesis of Frangois Sfigakis [17]; in this thick layer, there is a large portion of the interface that is oriented at very nearly 55° to the sample surface, as one would expect for a {lll}-type plane. It would be interesting to observe cross-sections of such oxide-semiconductor interfaces in various crystallographic directions and for different substrate orientations, to find whether certain planes are exposed preferentially according to the behaviour suggested here, in thick layers where interface strain effects will exert a smaller influence. The reduction in anisotropy observed with decreasing aluminum mole fraction [18] follows logically from the arguments we have offered. Since gallium does not oxidize easily, the exposure of a metal-terminated {111}A plane becomes less advantageous as more A l atoms on these planes are replaced by Ga. 2.6 Oxidation of Thin Layers Samples R782 and R754 each contain layers of high-x A l x G a i _ x A s which are less than 30nm thick. Oxidation of these layers has yielded a number of interesting observations. Both samples were patterned by electron-beam lithography, wet-etched, and oxidized C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF Al^Gai-xAs 23 simultaneously. Four oxidations were performed: Each sample was first oxidized for 30 minutes, removed from the furnace and measured. Each was subjected to a second oxidation at the same temperature, this time for 60 minutes. This routine was performed for two temperatures: 375°C and 400°C. The results of the first oxidation process are summarized in Table 2.2. Sample \ Oxidation temperature 375°C 4 0 0 ° C R782 (25 nm Alo.9sGao.02As) 1-1.3^m 1.4-2.1 fim R754 (24 nm AlAs) 1.2-2/xm 3.3-4.2//m Table 2.2: Summary of oxide widths obtained from 30-minute oxidations of samples containing thin layers of A l x G a i _ x A s . Distances were measured perpendicular to wet-etched trenches which were oriented parallel to the (011}-type (cleave) planes of the samples; that is, in the "slow" direction. These measurements represent the oxide formed to the outside of the perimeters which were outlined by the trenches, in order not to include possible stress effects observed on the insides of the closed paths. Where it was possible to determine the anisotropy, that information is included in Table 2.1. The patterns etched in these samples were dots ("pinholes"), circular trenches, square trenches, and "diamonds" (squares oriented with sides at 45° to the cleave axes). There were two sets of shapes: "large" circles and squares/diamonds with 44 diameters and sides respectively, and "small" circles and squares/diamonds with 26 /im diameters and sides respectively. The etch was significantly more uneven with the Alo.9sGao.02As sample R782. 2.6.1 Dots After oxidation for 30 minutes at 375°C, oxide formation from the holes was modest and sporadic on both samples. After the 30-minute 400°C oxidation, "pinhole" oxide forma-tion in R754 was much more regular and was shaped as expected (see Figure 2.14(b)), with an average anisotropy of 1.10 (ranging from 1.05 to 1.12 for 19 pinholes measured), and average "slow" direction oxide width of 2.6 /im. Oxidized at the same time, after 30 minutes at 4 0 0 ° C the R782 pinholes displayed disparities in shape and size, and many C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A ^ G a ^ A s 24 Figure 2.14: Nomarski optical micrograph of regions of (a) the 25 nm Alo.9gGao.02As layer of R782, and (b) (inset) the 24 nm AlAs layer of R754, oxidized at 400°C for 30 minutes through small wet-etched holes in the GaAs cap. exhibited a slight elongation in one of the <011> directions (see Figure 2.14(a)). The reason for the variation is unknown, but may be related to the irregularity of the etching. 2.6.2 Square and Circular Etch Perimeters Trenches etched in R782 oxidized fairly uniformly at the lower temperature (375°C) , except that there was more oxidation toward the outside of the shapes than toward the interior (Figure 2.15(b)). The R754 shapes were less regular. As is evident in Figure 2.15(a), the oxide front is somewhat uneven, and there is some accentuated raggedness in one of the <011>-type directions. In addition, some of the small squares exhibited a marked increase in oxidation on the interior side of the perimeter, perpendicular to the more ragged-looking oxidation direction. This can be observed in Figure 2.15(a). This did not occur with the larger shapes, and suggests an anisotropic strain in the oxidizing layer or in the cap layer which affects the oxidation of the AlAs. One possible explanation is that as the AlAs converts to oxide and the layer contracts, the ledge of GaAs over the newly-formed oxide delaminates slightly and draws water vapour in along the top A l A s / G a A s interface, speeding the conversion of the thin AlAs layer to oxide. C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A L G a i _ x A s 25 I lOum AlAs 375C A 1 0 .98 G a 0 .02 A s (b) Figure 2.15: (a) R754 (24nm AlAs), and (b) R782 (25nm Alo.9gGao.02As), oxidized at 375°C for 30 minutes. It may be that within the smaller shapes, there is more freedom for the oxide layer to contract in-plane than in larger mesas (where there is a smaller ratio of free perimeter to layer area within the closed path), and that this exerts a strain on the cap layer that is not otherwise a major issue. It is worth noting that the GaAs cap layer is C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A l x G a i _ x A s 26 much thicker than the AlAs layer in R754. That this phenomenon is not observed in R782 under the same conditions would be consistent with the previous observations that Alo.9gGao.02As contracts less than pure AlAs upon oxidation and results in a more robust oxide-containing layer structure, which is less likely to delaminate. At 4 0 0 ° C , sample R754 oxidized much more uniformly than at 375°C. However, there is still a hint of the irregularity of oxidation distance along one of the <011> directions, visible in the oxide from square trench shapes (see Figure 2.16(a)). In some of these squares, there is a slight ripple along these trench edges, suggesting a buckling delamina-tion right at the trench. Once again the Alo.9sGao.02As sample oxidized fairly regularly. Figure 2.16(a) and 2.16(b) show examples of the results of the 30-minute 4 0 0 ° C oxidation process on the two samples. It was found that in each case, the second "oxidation" resulted in no visible additional oxide formation. If the oxide at the exposed layer edge has become dense over long heating times, as suggested by our observation of the reduction in layer thickness nearer to the edge, the transport of H 2 O to the oxidation front would be' severely limited, and oxide formation would be slow or nonexistent. Choquette et al. [23] have found that samples etched and exposed to atmosphere weeks before the actual oxidation did not oxidize at a different rate from those which were oxidized immediately after etching, so exposure to atmosphere alone is not sufficient to prevent oxidation. However, Huffaker et al. [37] demonstrated that the further oxidation of partially-oxidized mesas which have been exposed to atmosphere can be prevented by a rapid thermal anneal (RTA) at ~500-6 0 0 ° C in forming gas. Samples treated in this way were protected from oxidation under the normal oxidation process conditions, and also resist etching by HC1. They suggest that the R T A treatment forms an impermeable surface oxide which seals out reactants and prevents decomposition of the porous oxide within the layer. Our samples may be sealed in the same way. From the point of view of the chemical composition of the oxide, we speculate that the hydrated Gibbsite (Al(OH) 3 ) form is porous, whereas the dehydrated forms are not. C H A P T E R 2: L A T E R A L OXIDATION OF B U R I E D L A Y E R S OF A h - G a ^ A s 27 This inhibition of oxidation in subsequent treatments may have important practical sig-nificance in that it may stabilize the internal semiconductor-oxide interface against air exposure. Chapter 3 Oxide Patterning by Selective Layer Interdiffusion The rate of lateral oxidation of a buried A lxGa i -^As layer increases with increasing aluminum content, especially as the value of x approaches 1 [20]. Thus if the aluminum content of a high-a; A ^ G a ^ A s layer can be modulated within the plane by some post-growth method, then for a particular set of process conditions some parts of the layer can be made to oxidize while others do not. This would be an interesting means to create a high-index-contrast D F B grating in a semiconductor waveguide or laser. One approach for creating this patterning is by selective-area interdiffusion: the blend-ing of interfaces between discrete epitaxially-grown layers by a local enhancement of the rate of diffusion of the atoms in that region. This can be done in various ways. 3.1 Layer Interdiffusion Background It was shown in 1981 by Laidig et al. [38] that the introduction of electrically-active impurity atoms can enhance the diffusion of atoms in a G a A s / A l A s superlattice to such an extent that under certain conditions (e.g. anneal for four hours at 575°C in an evacuated ampoule with zinc), the superlattice layers become completely alloyed. If parts of the sample surface are covered in SisN 4 , the layers beneath remain intact. This is an example of what is known as Impurity-Induced Layer Disordering (IILD). Other shallow dopants (such as Si) can be used to promote interdiffusion [39, 40]. This method for interdiffusing layers has the major drawback of requiring the presence of electrically-significant quantities of impurities in the intermixed regions [39]. 28 C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 29 There are two other popular means to effect a selective enhancement in diffusion with-out this disadvantage, commonly known as impurity free layer disordering or impurity-free vacancy diffusion, and ion-implantation enhanced layer interdiffusion (or intermix-ing) respectively. Both these methods are thought to work primarily by the generation of Group III vacancies in the crystal lattice, which diffuse within the crystal during anneal. This enhances the exchange of atoms among Group III lattice sites, notably at layer interfaces where the Group III elements are different on either side. Impurity-free layer disordering involves the capping of the sample with a layer of a dielectric substance into which gallium atoms will diffuse during an anneal, leaving behind Group III vacancies in the crystal. Ion implantation creates vacancies along the path of the energetic ions as they undergo collisions within the crystal. During subsequent anneal the vacancies diffuse as is the case with impurity-free disordering. It is worth noting that in the case of a ternary compound lattice-matched to a binary (e.g. In AlAs on InP [3]), the changing of layer compositions by blending with neighbours would introduce strain, so that the altered alloy is no longer lattice-matched to its sur-roundings. This is not a problem in the G a A s / A l G a A s system since AlAs and GaAs are nearly lattice-matched to begin with, and any blending will only reduce the mismatch. Initial process development for the intermixing experiments was done with impurity-free vacancy diffusion as the intended means for generating the Group III vacancies which would enhance the interdiffusion of A l and G a in selected areas during the anneal step. The remainder of the study focuses on the ion implantation method. 3.2 Objective The objective of this set of experiments is to achieve sufficient intermixing that within a thin layer, implanted regions will no longer oxidize when exposed to conditions that would cause an unimplanted region to oxidize. Sufficient intermixing could be defined as that which would reduce the A l concen-C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 30 tration throughout the entire depth of the A l x G a i _ x A s layer to a level at which the oxidation rate is very much slower than that of the as-grown layer (by, for instance, an order of magnitude or more). Alternatively, it may be sufficient to reduce the thickness of "oxidizable" material by significantly intermixing only some fraction of the thickness of the layer, due to the fact that for extremely thin layers, the oxidation rate decreases sharply with any reduction in layer thickness. There has been some use of post-growth intermixing to raise or lower the aluminum mole fraction of selected parts of an AlGaAs layer in order to control where oxidation will occur within that layer [41, 27, 42, 7], and there have been examples of the use of ion implantation and anneal to intermix GaAs and A l x G a i _ x A s in the pattern of a submicron-pitch grating to provide gain and/or low-contrast index coupling in a D F B laser [8, 9, 10, 11, 12, 13]. Our goal was to combine both of those concepts in order to create a buried high-index-contrast grating composed of alternating regions of A l x G a i _ x A s and insulating oxide. 3.3 Survey of Previous Work on Interdiffusion 3.3.1 Oxidation and Intermixing Massengale et al. [41] effected some lateral control of oxidation using impurity-induced vacancy diffusion. The starting material was heavily Si-doped, and contained two layers of AlAs, each surrounded by GaAs. Regions which were desired not to oxidize were protected with a C V D S i N x cap, and the sample annealed at 875°C for two hours in contact with a Si wafer. Wet oxidation was then performed at 425°C. The regions which were not capped oxidized up to ten times as fast as the regions which were capped. It was claimed that the number of Group III vacancies in Si-doped material is very large in comparison to the number found in intrinsic material, and that these vacancies were responsible for the intermixing observed, while the loss of arsenic from the uncapped re-C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 31 gions created point defects which compensated for the Group III vacancies and inhibited compositional change in the AlAs. The regions which had undergone significant inter-mixing had a lower average A l content and therefore oxidized more slowly. It seems likely however, that over the two-hour anneal time the presence of the Si impurity was more significant to the intermixing than were vacancies since vacancy-enhanced intermixing saturates after approximately 10 seconds [43]. Another example of lateral selectivity in oxidation achieved by impurity-induced layer disordering can be found in Reference [27], A region of laser material which was capped with SisN 4 remained intact during a 13-hour anneal at 850°C. The remainder of the sample was capped with C V D Si to allow diffusion of Si impurity atoms into the structure, blending several differently-composed layers into a thick region with average x~0.5, high enough so that with a very high process temperature (500-525°C) , the thick intermixed layer could be oxidized to create a wall of oxide for electrical and optical confinement. A similar process using Zn diffusion has been used by Zhou et al. [42] to create a quasi-three-dimensional photonic crystal with large feature size (~6 /xm spacing between scattering centers). They note that the use of ion-implantation induced or impurity-free disordering techniques could improve the resolution of the process. Cohen et al. [7] reduced the thickness of oxide layers in a high-index-contrast D B R stack by partially intermixing Al0.2Ga0.sAs and AlAs layers before lateral oxidation, re-sulting in a shifted resonant wavelength. This was done using a laterally-uniform implan-tation of protons at several energies to distribute vacancies throughout the stack before annealing at 900 and 950°C. There was no lateral selectivity of the intermixing in this case. 3.3.2 DFB lasers with selective intermixing Though intermixing (without oxidation) has been used by several groups to create grat-ings within D F B lasers in various material systems, the degree of intermixing attained C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 32 by the methods of these groups is difficult to ascertain. Success is normally assessed in terms of whether the grating is of a high enough quality to induce single-mode operation of the laser, which gives little clue to the applicability of the parameters used toward intermixing with the goal of inhibiting oxidation. One important piece of information gleaned from these works is that the process of ion-implantation-enhanced intermixing has been applied to create modulations in material composition with very high resolution. Three groups have presented three approaches for implementing D F B coupling using ion-implantation-enhanced intermixing. Steckl et al. [8] fabricated D B R gratings of 350 nm period in a G a A s / A l G a A s super-lattice using a 200keV S i + + focused ion beam (to 10 1 4 ions/cm 2) followed by a 950°C, 10 s anneal. The gratings were placed on either end of an unpatterned laser cavity region and provided index coupling but did not result in single-mode emission at 77K. Work at the Universitat Stuttgart has yielded single-mode gain-coupled D F B lasers in two materials systems, both using masked broad-area ion implantation and anneal. A 1.4/xm InGaAs/AlInGaAs/InP D F B laser (operated at 150K) was fabricated using a 435-nm-pitch gold grating mask for 255keV A r + implantation [9]. A room-temperature G a A s / A l G a A s laser was made by a similar process but with 170keV 6 9 G a + (up to l x l O 1 3 ions/cm 2) over a 200-260nm second-order grating [10]. The anneal was 8 7 5 ° C for 60s. Finally Konig et al. have fabricated complex-coupled InGaAsP/InP D F B lasers which have 1.55 /im C W single-mode, room-temperature operation through intermixing [11, 12, 13] and implantation-enhanced chemical etching [12, 13]. A lOOkeV G a + FIB was used to write gratings on either side of the laser ridge for side coupling. A n anneal served to complete the intermixing process for gain coupling, and an etch in hot H F selectively removed the ion-damaged InP to form high-contrast index-coupling gratings with periods 238-248nm. In this case implantation was performed at normal incidence; channelling along the [100] direction was not avoided. Al l of the above works involve modulating the composition of the active region of C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 33 the laser using ion implantation, in order to create gain and/or index coupling. The approach being explored in this work, namely the use of intermixing to pattern an oxide grating for index coupling, has the advantage that the active region is not the target of manipulation and need not necessarily be subjected to significant ion damage. 3.4 Proposed Process for Patterning Using Ion-Implantation-Enhanced Diffusion Lateral modulation of the composition of the buried Al^Gai-zAs layer is to be achieved by implanting only selected regions with Si ions before a high-temperature anneal. The areas to be implanted can be selected either by "writing" a pattern with a focused ion beam, or by masking the areas not to be implanted prior to broad-area implantation. The latter method was selected for this study. It is less complicated and more readily scaled up for commercial application if the desired pattern can be created using U V lithography. Short-period gratings can be delineated in photoresist using holographic U V lithography, and the use of broad-area implantation allows preliminary work to be done on bulk material before investigating patterning and lateral modulation. The process proposed here for the formation of a buried high-index-contrast D F B grating is as follows: 1. Sample Masking: A gold implantation mask in the form of a grating with the desired period is deposited on the sample. To prevent diffusion of gold into the GaAs cap of the sample, a barrier layer of dielectric material is deposited before the gold [4, 9]. This also allows removal of the mask by a lift-off process. The gold thickness to be used is determined by calculation of the thickness needed to prevent ions from penetrating to the sample during the next step. 2. Ion Implantation: The sample is bombarded with energetic ions which cause damage by collisions with atoms in the crystal lattice. The desired type of damage C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 34 is the Group Ill-type vacancy, which is held to be the key to the diffusion of Group III atoms in the anneal step. It is impossible to achieve only the desired defects of course; the crystal will suffer other damage during implantation. 3. Anneal: Following implantation, the mask is stripped and the sample annealed at high temperature. The Group III vacancies become mobile and hop from site to site on the Group III sublattice, resulting in displacement of G a and A l atoms from their original positions so that in the implanted regions intermixing will occur while the unimplanted (masked) parts of the sample remain relatively unaffected by this anneal, so long as the surface is adequately protected from As loss and oxidation. Aside from accomplishing the interdiffusion, the anneal step also serves to repair crystal damage from the implantation. 4. Oxidation of Intermixed Samples: Because of the x-dependence of the oxi-dation rate of A L j G a i _ x A s , a selectivity should occur in the oxidation of the now partly-intermixed sample. The regions which had been masked during implanta-tion should retain their original value of x while the regions which were implanted should be lower in aluminum (higher in gallium). It should be possible to pattern ridges on the sample, transverse to the implanted grating, expose the ridge edges by etching, and oxidize only the unimplanted regions of the original A L - G a i _ x A s layer, forming a grating of A l x G a i _ x A s alternating with Al^Gai.yAs-oxide. 3.5 Conditions for Implantation-Induced Layer Intermixing Ions travelling along the [100] axis in a zincblende crystal have fewer collisions with atoms in the lattice than do ions travelling in an arbitrary direction. This is known as channelling, and results in fewer vacancies produced and a deeper, less predictable vacancy depth profile [44]. The offset of the ion beam incident angle from the normal must of course be aligned with the mask grating direction. C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 35 The energy of the implanted ions is chosen to place the maximum in vacancy con-centration near the depth of the layer which is desired to be interdiffused. Chen and Steckl [14] suggest that maximal mixing should occur at the depth of the maximum of the second derivative of the vacancy distribution, which lies deeper than the peak in the ion distribution. However, this work was done with a Si FIB at normal incidence. Poole et al [5] assert that for "low-energy" ( « lOOkeV) implantation, intermixing observed in the G a A s / A l G a A s and InGaAs/GaAs systems at depths significantly further than the calculated ion range is due to ion channelling during implantation which results in the generation of vacancies deeper than the calculated range, and that the strongest mixing does in fact occur at the location of highest vacancy concentration. They found that for the most efficient intermixing of buried layers in these systems, the ion energy should be chosen such that the maximum of the vacancy distribution coincides with the depth of the interfaces which are desired to be blended. Minimum lateral straggle occurs with the lowest usable implant energy for a given ion species, so in that respect it may be advantageous to place the layers to be affected as near the sample surface as is practical. By contrast, the vacancy diffusion lengths found in the InGaAs/InP system appear to be high enough that the vacancy distribution is not as crucial as the total number of vacancies present [5, 3], Also, end-of-range ion damage is more persistent through the anneal than the damage along an ion track, so if possible the layer structure and ion energy should be chosen to keep the ion range out of any Q W regions or regions where crystal quality is crucial [5, 4], It has been found that excessive ion damage slows the intermixing process [5]; with higher doses, some point defects coalesce into extended defects which are not mobile and therefore not helpful in intermixing [45]. It is in fact desirable to minimize ion damage for the sake of preserving the electronic quality of the semiconductor material. It appears that the intermixing due to vacancy diffusion occurs in the first several seconds of a high-temperature anneal [45, 9], although recovery of crystal structure continues during the remainder. The anneal must be hot enough to allow interdiffusion C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 36 of the implanted regions to take place. For the G a A s / A l G a A s system, this requires a temperature > 8 5 0 ° C [8, 5]. G a A s / A l G a A s interfaces in unimplanted regions are quite stable against interdiffusion at these temperatures [46, 40]. For the materials systems involving In, the required anneal temperature is lower; anneals ranging from 6 7 5 - 8 5 0 ° C have been used to effect laterally-selective intermixing [9, 5, 12, 6]. It is crucial during anneal to prevent excessive As loss from the sample surface. This is commonly done (see, for example, References [14, 9, 10]) by placing a Si or GaAs wafer in contact with the sample, and has become known as "proximity protection." Hofsafi et al. [9] find that in samples of InGaAs and AlInGaAs on InP, performing the post-implantation anneal in an M O C V D reactor with an A s H 3 flux results in a much better surface quality than with R T A with Si proximity protection, as determined by the quality of material regrown by M O C V D on the samples. It has been noted [47] that any plasma process to which the sample is exposed (for example E C R etching) may cause damage which will affect the degree of intermixing. 3.6 Samples Two MBE-grown heterostructures were used in intermixing experiments; one of which, R754, was also used in the previous chapter. The second, labelled NRC2298, was grown by Dr. Zbigniew Wasilewski of the Institute for Microstructural Sciences at N R C in Ottawa. As noted earlier, R754 contains a layer of AlAs which was determined by X R D to be 24 nm thick. Sample NRC2298 was designed to have an even thinner oxidizing layer, specifically 20 nm of Alo.9sGao.02As. A thinner layer will require less diffusion to achieve the same reduction in oxidation. Its structure was chosen using a computer code written by Vighen Pacradouni for calculating the dispersion and mode profiles of bound modes in an arbitrary multilayer dielectric waveguide. C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION C H A P T E R 3: O X I D E P A T T E R N I N G BY S E L E C T I V E L A Y E R INTERDIFFUSION 38 Figure 3.2: (a) S E M cross-section image of sample N R C 2 2 9 8 , stain-etched to emphasize interfaces, and (b) schematic of layer structure, as determined by X R D . C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 39 3.7 Diffusion Most authors (see, for example, References [46, 48, 49, 7]), use a simple "error function" model based on Fick's second law of diffusion to predict the profile of an interdiffused interface. The rate and extent of the interdiffusion of all Group III species is described in terms of a single "interdiffusion constant" [7] which is independent of time, isotropic, and independent of A l and G a concentrations. According to this model, the formulation of the A l content profile of a partially interdiffused layer of A l x G a i _ x A s , buried in pure GaAs and having width 5 and centre at z = 0, is: where x0 is the original A l mole fraction of the layer and Ld is the single diffusion length which is used to quantify the extent to which the Group III species on either side of the interface have interdiffused. With the exact mechanisms of layer intermixing not thoroughly understood, and dif-ferent studies carried out using layers of various thicknesses, compositions, and locations within the samples, it is difficult to objectively compare experimental numbers quantify-ing the diffusion, but diffusion lengths in the literature are typically up to 4nm [7]. Figure 3.3 shows the mid-layer (maximum) remaining A l content as a function of diffusion length L by the prediction of the error function model, for each of the two structures to be used in this chapter. Figure 3.4 gives the predicted composition profiles for selected values of diffusion length. (3.1) C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 40 Highest remaining A l concentration vs L 0.55 h 0.51 i i . 1 • • « J 0 1 2 3 4 5 6 7 8 Group III diffusion length (nm) Figure 3.3: Maximum remaining A l concentrations within high-x ALjGai -zAs vs. diffu-sion length L for R754 and N R C 2 2 9 8 , using simple Fickian diffusion model. C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 41 UOIPB4 aioui iv a o3 "a? 4) CO H 03 o 03 SO .5 £ 'a * o . 3 a> W 03 £ T3 as CI . I—I 00 CO c n c o h - c o m - s r - w c N ! o o d o o c s d o U 0 I P B 4 8|0Ui iv — "3 <q += C O ^ od CO 0 5 3 q fa 2 C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 42 The oxidation rate of high-x A l x G a i _ x A s is extremely sensitive to x. For example: As compared to the oxidation rate of pure AlAs, oxidation has been reported to be 20 times slower for x = 0.92 at 440°C [18]; 40 times slower for x = 0.90 at 410°C [20]; and more than 100 times slower for x = 0.85 at 420°C [23]. In addition to the dependence on aluminum content, the oxidation rate has a thickness dependence below ~150nm [20]. Thinner layers oxidize more slowly, and this dependence gets stronger as thickness decreases [20, 23]. It has been predicted that AlAs layers <15nm should be very difficult to oxidize [26, 50]. Even if the interdiffusion does not affect the aluminum concentration at the layer midpoint, the effective thickness of the oxidizable layer may be reduced enough to inhibit oxidation. 3.8 Ion Implantation Ion implantation was performed by Dr. Mark Ridgway at the Department of Electronic Materials Engineering of the Research School of Physical Sciences and Engineering at the Australian National University. Samples were implanted unmasked with 2 8 S i at room temperature (21C) at a dose rate of 2.6xl0 1 2 ions/cm2-s. The samples were implanted with 2 8 S i ions, at 7° from normal to avoid ion channeling, which occurs at normal inci-dence. Each implanted sample was given several doses in different regions so that any effect of implantation dose on oxidation properties could be seen. 3.9 Anneal Samples were annealed in an A G Associates Heatpulse 2101 rapid thermal annealer (RTA), under a flow of Ar gas. Because unprotected GaAs decomposes and arsenic escapes at the temperatures necessary for intermixing, measures must be taken to keep the sample surface intact. Adequate purging of the chamber prior to heating is essential to remove water and C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 43 oxygen. If this step is omitted, severe pitting of the sample will occur. Pits extending more than 100 nm deep were observed even in samples protected by dielectric films (Si02, C a F 2 or SrF 2 ) and in samples whose surfaces were placed in contact with a sacrificial piece of GaAs wafer to provide As overpressure. With no pre-anneal purge a ten-second anneal at 8 0 0 ° C results in a brownish dis-colouration and slight haziness on the surfaces of two GaAs samples placed in the R T A with their surfaces together. X-ray Photoelectron Spectroscopy (XPS) measurements performed by Dr. Ken Wong of the Surface Chemistry Lab in A M P E L indicate that the surfaces were As-poor and oxidized (see Figure 3.5). With proper precautions, however, samples can be annealed and retain mirrorlike surfaces. Cleanroom engineer A l Schmalz modified the input on the R T A to carry either N 2 or Ar, in order that a long purge might be carried out in N 2 before an anneal under the purer and much more expensive Ar gas. The procedure adopted here for sample preparation and anneal is as follows: The implanted sample is first rinsed in acetone, followed by 2-propanol, then deionized water, and dried with an N 2 jet. The solvent rinse is to remove contamination the sample may have picked up from handling or packaging between shipping to Australia for implantation and annealing at U B C . This is followed by immersion in a strong HC1 solution (1:1 HC1:H 2 0) to remove oxide buildup on the GaAs surface, another H 2 0 rinse and N 2 blow-dry, and several minutes on a hotplate > 1 0 0 ° C to drive off water. The sample to be annealed is then placed face-down on a similarly-prepared piece of GaAs. Due to a visible difference in surface quality between samples annealed face-up and those annealed face-down under insufficient purge conditions, face-down is kept as the protocol. This piece of GaAs, known in the literature as a "proximity cap," helps provide arsenic overpressure during the anneal, preventing significant surface damage due to loss of arsenic. The two pieces are sandwiched between two quarter-3-inch wafers of Si and the entire assembly is held together with three clips made of bent tantalum foil. Prior to assembly of the "sandwich" the Si pieces are subjected to several high-C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 41 (a) 14 16 18 20 22 24 26 Binding energy (eV) 1 1 1 r (b) 35 40 45 50 Binding energy (cV) Figure 3.5: X P S results for two samples annealed at (a) 800°C and (b) 840°C, without a pre-anneal purge to prevent oxidation damage to the surface during the anneal. C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 45 temperature anneals while the R T A parameters are tested. The clipped sandwich structure ensures that the sample and its GaAs proximity cap are held in contact from their placement on the R T A load arm to their post-anneal removal. This is important as the automatic load arm moves quite abruptly and a free-standing pile of wafers would be upset by its motions. After being loaded into the R T A chamber, the sample sits for one hour under a purge of lOslm N 2 gas. The purge gas input is then switched to argon and a very short pre-anneal heating is performed ( ~ 4 0 0 ° C , 5s) to drive off residual H 2 0 and to warm up the R T A chamber and lamps. The desired anneal is then performed. Temperature measurement is by a thermocouple whose end is in contact with the underside of a Si wafer that serves as a holder for small samples in the R T A . The tem-perature during anneal is monitored and recorded using an analog chart recorder. 3.10 Patterning and Oxidation Prior to oxidation, samples were patterned by electron-beam lithography or U V (optical) lithography and trenches were chemically etched using a solution of citric acid and hy-drogen peroxide to remove GaAs and H F to remove high-x ALjGai-zAs . The procedures can be found in the Appendices. Patterns were etched in all regions of each sample so that oxidation of differently-dosed regions could be compared. Oxidation was performed by the same procedure as in the previous chapter. 3.11 Oxidation Observations 3.11.1 R754 Implant 1 The first implantation was done with 95keV 2 8 S i ions, into Sample R754. Figure 3.6 illustrates the vacancy and ion range distributions calculated for this set of implant C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 46 parameters using SRIM-2000 (TRIM). The vacancy peak falls well short of the AlAs layer. 1.4 H 1.2 i.o H 0.8 H 0.6 0.4 H 0.2 0.0 G a A s j A l A s G a A s Ion and vacancy distributions for 95keV implantation of Si into R754 (7 degrees from normal) Vacancies per ion per Angstrom Ions (not to scale) Boundaries of AlAs layer 1 1 I 500 1000 1500 2000 2500 Depth (Angstroms) 3000 Figure 3.6: T R I M simulation for vacancy and ion distributions for the conditions of R754 Implant 1. A piece of this implanted sample, containing three regions: unimplanted, 6x l0 1 3 ions/cm 2 , and 2xl0 1 4 ions/cm 2 , was annealed at 940°C for 30 s and oxidized for 90 min-utes at 380°C. The result was inconclusive. Al l regions oxidized considerably (~10-30 u-m) and irregularly from the etched edges, exihibiting similar characteristics to the low-temperature (375°C) oxidation of R754 in the previous chapter. In any case intermixing was not sufficient to cause a definitive change in oxidation rate. A second piece of R754 Implant 1, containing the same dose regions as the first, was annealed at 940°C, this time for 60 seconds. Because it has been suggested [7] that oxidation rates are more strongly dependent upon A l mole fraction at lower oxidation C H A P T E R 3: O X I D E P A T T E R N I N G BY S E L E C T I V E L A Y E R INTERDIFFUSION 47 50u.m a b H — - — H Figure 3.7: Optical micrographs showing oxidation from two regions of R754 Implant 1, annealed at 9 4 0 ° C for 60 seconds: a) No implant; b) 2xl0 1 4 ions/cm 2 . The dark lines are etched trenches designed to expose the AlAs for oxidation. The light-coloured region is where the buried AlAs has oxidized. temperatures, this sample was oxidized at 370°C for 3.5 hours. The oxidation in all regions was somewhat uneven, which is consistent with the observation of the previous chapter that at lower temperatures R754's thin AlAs layer oxidizes more unevenly, but still no trend becomes clear. Figure 3.7 shows oxidation from two regions. 3.11 .2 R 7 5 4 Implant 2 The second set of implants was done using lOOkeV 2 8 S i (see Figure 3.8 for T R I M calcula-tions). The vacancy peak is still short of the AlAs layer, but due to equipment restrictions implantation energies between 100 and 200keV were not possible. The four doses ranged from 2xl0 1 4 ions/cm 2 (the highest dose of R754 Implant 1), to l x l O 1 5 ions/cm 2 . Dr. Ridgway estimated that some amorphization had occurred with the heaviest three doses. Anneal was carried out at 945°C for 60s and the sample oxidized at 3 7 0 ° C for 3.5 hours. As with the previous samples, oxidation was quite uneven, but this time a trend appeared, as Figure 3.9 illustrates. This figure clearly shows that the implantation sup-presses the oxidation, which was one of the objectives of this project. Further work is required to improve the uniformity of the oxidation. The extent to which amorphization is a factor remains to be determined. C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 48 GaAs A l A s : GaAs Ion and vacancy distributions for lOOkeV implantation of Si into R754 (7 degrees from normal) Vacancies per ion per Angstrom Ions (not to scale) Boundaries of AlAs layer 1.2 H i.o H 0.8 H 0.6 H 0.4 H 0.2 H 1 ! ! 1 ! ! , , 1 , , , , p . , , , ! 1 , , , , 1 , i r 500 1000 1500 2000 2500 3000 Depth (Angstroms) Figure 3.8: T R I M simulation for vacancy and ion distributions for the conditions of R754 Implant 2. C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 49 d i * » i Figure 3.9: Optical micrograph showing the trend in oxidation widths for R754 Implant 2 annealed at 945°C for 60s: (a) 2xl0 1 4 ions/cm 2; (b) 5xl0 1 4 ions/cm2;(c) 7.5xl0 1 4 ions/cm 2; (d) l x l O 1 5 ions/cm 2. The dark lines are etched trenches designed to expose the AlAs for oxidation. The light-coloured region is where the buried AlAs has oxidized. The most heavily-implanted region (d) exhibits almost no oxidation. Dark specks on surface are contamination and are not intrinsic to the process. C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 50 0.8 H 0.6 H 0.4 H 0.2 H 0.0 — l GaAs Ion and vacancy distributions for 2()0keV implantation of Si into NRC2298 (7 degrees from normal) Vacancies per ion per Angstrom Ions (not to scale) I Locations of layer interfaces i i i i | 1 1 1 1 | 1 1 1 1 I 1 1 1 1 | 1 1 1 1 I 1 1 1 1 | 1 1 1 1 I 1 1 1 1 | 1 1 1 1 1 1 1 1 1 I 1000 2000 3000 4000 5000 Depth (Angstroms) Figure 3.10: T R I M simulation for vacancy and ion distributions for the conditions of NRC2298 Implant 1. 3.11.3 N R C 2 2 9 8 Implant 1 The waveguide structure NRC2298 was implanted at 200keV, in order to locate the vacancy peak as near as possible to the Alo.9gGao.02As layer. Figure 3.10 gives the T R I M prediction for the vacancy and ion distribution. Implant conditions were otherwise the same as for the previous samples. Five doses were administered, from 2 x l 0 1 3 to 5 x l 0 1 4 ions/cm 2 . The first two attempts to oxidize this sample were inconclusive. The sample was annealed at 940°C for 60s, patterned by U V lithography, and wet-etched. Unlike the as-grown control sample, the implanted piece did not etch nicely with a citric acid/hydrogen peroxide etch which etches very-low-aluminum A ^ G a ^ A s selectively and for which the Alo.9gGao.02As layer should have been an effective etch-stop layer. Instead, from early on C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 51 -j c . '«• 'd ' • e Figure 3.11: Optical micrographs of etch pits in NRC2298 after implantation and anneal at 9 4 0 ° C for 60s. Sample was etched for 20 minutes in a solution of 6:1:40 citric acid (50% by weight):H 2O 2(30%):H 2O. (a) 2xl0 1 3 ions/cm2;(b) 5xl0 1 3 ions/cm2;(c) l x l O 1 4 ions/cm 2;(d) 2xl0 1 4 ions/cm2;(e) 5xl0 1 4 ions/cm 2 in the etch, pits developed whose density varied with implant dose (see Figure 3.11). A n A F M measurement was taken on a region which included the "shadow" of a clip which held the sample during implantation and thus shielded a small region from ion damage. From this it appears that smooth-topped plateaux exist between pits at the same level as the etch stop layer. The pits may be located at individual damage cascades from the implantation. The first oxidation, 30 minutes at 400°C, was insufficient to yield any definitive oxi-dation, even in the control sample. The second oxidation, of the same annealed sample repatterned with freshly-etched trenches, was performed at 417C for one hour. This re-vealed that the etch pits in the implanted sample had penetrated the lower GaAs layer; several oxidation "fronts" are visible using the DIC microscope, showing that several of the Alo.65Gao.35As layers below had oxidized significantly (up to several microns). This made it impossible to discover whether the very thin Alo.9sGao.02As layer had oxidized or not. The process was performed on a different piece of the same implanted NRC2298 sample. The anneal was 90s at 940°C, the wet etch was monitored carefully by periodic A F M measurements, and the sample oxidized at 417C for one hour. However, once again C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 52 the unimplanted control sample showed no definite oxidation under DIC observation. Since the oxide layers formed from similar layers in R754 and R782 were very clearly visible using the DIC microscope, it appears that the 20 nm Alo.9sGao.02As layer in this sample is too thin to oxidize at this temperature. However, there was evidence of a compositional change in the thin Alo.9sGao.02As layer with increasing implant dose: what appears to be a uniform ~600 nm undercut from the H F etch is visible along the trench edges in the as-grown sample (figure not available) as well as in the regions of lowest dose in the implanted, annealed sample. This undercut becomes uneven with a dose of l x l O 1 4 ions/cm 2 , and is not discernible in the areas implanted with 2x l0 1 4 ions/cm 2 or more. The implanted and annealed regions showed increasing degrees of surface damage with increasing dose. Because the wet etch was more carefully controlled, there were fewer damaged locations on this sample where the buried Alo.65Gao.35As was allowed to oxidize. 3.12 X-ray Diffraction X-ray diffraction rocking curves were taken to learn how various process steps affect the layer structure of our samples. Figure 3.12 illustrates the progression of the rocking curve with increasing dose of Si at 95keV. At lower doses the interstitial Si increases the average lattice constant within a vertically-confined region, causing a broad peak which moves further left as dose increases, but the fringes that indicate the existence of a sharp interface with the AlAs layer remain. At a dose of 2xl0 1 4 ions/cm 2 , however, the peak does not shift further left, but broadens, and the original fringes vanish. It may be that this indicates the onset of amorphization in the implanted region. Figure 3.13 illustrates the extent to which the original scan features were recovered by a 30-second anneal at 875°C. Figure 3.14 compares the final post-anneal rocking curves for the various doses. The original fringes were recovered to some extent for all samples, although for higher doses they become less strong. Attempts were made to C H A P T E R 3 O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 53 10 ~i—i—i—|—i—i—i—|—i—i—i—|—i—i—i—|—i—i—i—|—i—i—i—|—i—i—i—|—i—i—i—|—i—i—i—|—i—i—i -800 -600 -400 -200 0 200 400 600 800 Relative Arcseconds Figure 3.12: Effect of increasing implant dose on X R D from unannealed R754, Implant 1 (95 keV Si). emulate the washing-out of fringes using simulations of layer structures containing various shapes of graded interface, but this was unsuccessful. This effect is due more likely to lateral inhomogeneity at the layer interfaces than to laterally-uniform intermixing. Vieu et al. [45] found that for a low dose ( lx lO 1 3 ions/cm 2) of G a + at lOOkeV, with little crystal damage evident after a 10-minute anneal at 800° C , the slightly-interdiffused G a A s / A l 3 5 G a 6 5 A s interfaces were visibly uneven under T E M observation. For a heavier dose ( lxl0 1 4 ions/cm 2 ) , the blending of interfaces was more extensive and more uniform. C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 54 The theory put forward in that work is that the vacancies which enhance interdiffusion are generated by collision cascades which, at lower doses, do not overlap, and that the distance the vacancies diffuse before recombination or formation of extended defects is not sufficient to smooth out the relatively widely separated damage cascades. This would be consistent with the uneven etching we observed in implanted N R C 2 2 9 8 (Figure 3.11). We were unable to determine the degree of intermixing from the X-ray studies. T E M studies of the interfaces by Karen Kavanagh and Victoria Fink at Simon Fraser University are inconclusive at this time. C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 55 Q 101 | i i t i i i i | i I | I I I I I I I | I I I I I I I | I I | I I I I I I I | -1200 - 800 -400 0 400 800 1200 Relative Arcseconds -1200 -800 -400 0 400 800 1200 Relative Arcseconds Figure 3.13: X-ray rocking curves of different regions of R754 Implant 1, before and after Anneals 1 and 2 (both 875°C, 30 s). Baselines are offset for clarity. C H A P T E R 3: O X I D E P A T T E R N I N G B Y S E L E C T I V E L A Y E R INTERDIFFUSION 5G -800 -600 -400 -200 0 200 400 600 800 Relative Arcseconds Figure 3.14: X-ray rocking curves of different regions of R754 Implant 1 Anneals 1 and 2 (both 875°C, 30s). Baselines are offset for clarity. As the dose increases, the fringes become more washed-out. Chapter 4 Conclusions 4.1 Wet Lateral Oxidation of Al x Gai_ x As This thesis describes pattern formation in buried G a A s / A L J G a 1 _ x A s heterostructures by wet oxidation from etched trenches. Difficulty was encountered in oxidizing pat-terns etched by electron cyclotron resonance (ECR) etching, and it is hypothesized that GaAs from the trench bottom is deposited on the sidewalls during the etch, sealing the A l x G a i _ x A s layer against oxidation. Anisotropy was evident in the lateral oxidation of high-x A l x G a i _ x A s ; the oxide extends further in the <001> directions than in the <011> directions on our samples, all of which were on (lOO)-oriented substrates. This has interesting effects on the final shape of an oxidized region. While convex oxidation fronts will remain convex as oxidation proceeds, the faster directions within a concave front will eventually catch up with the slower direction, forming sharp corners. Due to the sharpness of these corners, the periodicity of a patterned shape will persist over large oxidation distances. This contrasts with an oxidation process that is limited by diffusive transport, where the oxidation front tends to smooth out with time. This property can be used in the lateral transfer of etched patterns. One possible application would be the fabrication of side-coupled high-index-contrast D F B gratings. We proposed a mathematical model to describe this anisotropy, taking the form: ^ = a + to2(20) at 57 (4.1) C H A P T E R 4: CONCLUSIONS 58 where ^p- is the rate of progress of the interface along its outward normal, n(6), at a given time t; 6 is the angle between the direction of propagation and the [Oil] direction; a and b are constants. Implemented in a numerical simulation, this model describes the oxide shape very well, and can be used to quantify the degree of anisotropy. It was found that for a layer of Alo.9sGao.02As oxidized at 425C, the anisotropy was more pronounced than for layers of AlAs and Alo.9sGao.02As oxidized at temperatures from 375-400°C, consistent with previous observations in the literature [29, 30] that anisotropy increases with temperature. We also proposed an explanation for the crystalline anisotropy, based on the high reactivity of aluminum with oxygen and the inequivalence of the {111}A and {111}B planes in the zincblende lattice. The aluminum-terminated {111}A planes oxidize more quickly than do the arsenic-terminated {111}B planes. Depending upon the substrate orientation and direction of oxide propagation in the plane, the oxidation front within the oxidizing ALjGai - sAs layer will expose different combinations of these planes, resulting in varying rates of oxidation. Two phases were observed in atomic force microscope images of oxidized A l x G a i _ x A s samples; one region nearest the front with a constant thickness, and the remainder de-creasing in thickness as proximity to the exposed edge increased. We hypothesize that the original product of oxidation of A h - G a ^ A s (or of its A l constituent atoms) is an amorphous hydrated oxide, and that during the cooling process after oxidation this oxide becomes gradually dehydrated. Subjecting thin-layer samples R754 and R782 to a second oxidation process at 375C and 400°C after one oxidation and exposure to atmosphere did not result in an increase in the oxidation distance, suggesting that the partially or completely dehydrated oxide formed at the exposed layer edge is impermeable to water vapour. This may enhance the stability of devices made from these materials. The oxidation of thin layers from patterned etched trenches was studied. Oxidation of a thin layer of AlAs from etched trenches at very low temperature (375C) was found to be irregular while a higher temperature yielded much more uniform oxide front. A C H A P T E R 4: CONCLUSIONS 59 25-nm-thick layer of Alo.9sGao.02As, etched with the same pattern of trenches, oxidized predictably at both temperatures. This may reflect the higher volume contraction of pure AlAs on oxidation as compared to Alo.9sGao.02As, the oxide contraction causing mechanical instability during very slow oxidation. 4.2 Use of Layer Interdiffusion to Control Oxidation Rate The oxidation rate of A ^ G a ^ ^ A s within an epitaxial layer can be controlled by mod-ulating its aluminum concentration. We have proposed a process for creating a high-index-contrast D F B grating using selected-area ion implantation induced interdiffusion and wet lateral oxidation to exploit this property. Experiments showed that adequate purging to remove water vapour is crucial to the preservation of sample surface quality through high-temperature rapid thermal annealing. Samples containing thin layers of AlAs or Alo.9sGao.02 As were implanted with 2 8 S i and annealed at ~940-945C for 60-90 s. Sample R754, with the peak in predicted vacancy concentration falling well short of the 24 nm AlAs layer, required a dose of 5x l0 1 4 ions/cm 2 to yield a definite reduction in oxidation rate at low temperature (370-380°C) . This implant dose was suspected to have caused some amorphization. X-ray diffraction studies suggest a lateral inhomogeneity is present in layer interfaces of implanted, annealed samples, increasing with implant dose. Inhomogeneous intermix-ing, or damage cascades that were not fully repaired during the anneal, also caused pits to form during wet etching. These pits increased in density with increasing dose. Surface damage in general also increased with dose. It is concluded that to optimize the process, ion dose must be minimized by adjusting the implantation energy such that the peak in vacancy generation coincides with the depth of the A h - G a ^ A s layer which is to be intermixed. In turn, this layer must be thick enough to oxidize, but as thin as possible, given this restraint, to minimize the diffusion necessary to yield a selectivity of oxidation rate between implanted and unimplanted C H A P T E R 4: CONCLUSIONS 60 regions. Given that we encountered difficulty in oxidizing a 20 nm thick Alo.98Gao.02As layer, it might be better to begin with a layer of pure AlAs, to give the thinnest possible layer that will oxidize before intermixing. It may be useful to try to improve the quality of the intermixed material by the implantation of protons rather than Si ions. A higher dose of lower-energy protons might even out the inhomogeneity [45] and result in less coalescence of point defects into extended defects [51] which are more difficult to remove by annealing [52]. Appendix A: Gold Masking for Selective Implantation In order to achieve selective intermixing of a buried Al xGai_a;As layer by broad-area ion implantation, the sample must be partially masked during implantation. The purpose of the mask is to stop ions from penetrating the sample in any but the regions which are desired to be intermixed. It is this mask that will define the pattern of intermixing within the thin ALjGai-zAs layer. The primary material and thickness of the mask must meet this ion-stopping requirement. Gold is a suitable choice in this respect as it has a high density: 19.32 g/cm 3 as compared to 2.27g/cm 3 for S i 0 2 . A mask of gold need not have nearly as high an aspect ratio as one of S i 0 2 with the same capacity to stop energetic ions. To prevent diffusion of gold into GaAs, it is common to first deposit a barrier of a dielectric substance such as S i 0 2 [53] or A 1 2 0 3 [9] that can be removed in H F after implantation. The electron beam resist polymethylmethacrylate ( P M M A ) has been used for this purpose [10], and this has the advantage of allowing the gold mask to be lifted off in acetone, using an ultrasonic bath. It is not difficult, when attempting to remove a deposited oxide coating from a sample containing a high-x A l x G a i _ x A s layer with H F in an ultrasonic bath, to etch away the A l x G a i _ x A s layer, and lift off the GaAs cap layer from the substrate. In our experience P M M A can be difficult to remove from a sample surface after E C R etching, so a layer of oxide below it, that can be removed in H F after liftoff, without an ultrasonic bath, will result in a cleaner surface in the end. What appears to be a very good design can be found in Reference [9] (from top to bottom): 100 nm A u 3nm Cr 20 nm P M M A 20 nm A 1 2 0 3 61 A P P E N D I X A : G O L D M A S K I N G F O R S E L E C T I V E I M P L A N T A T I O N 62 The thin Cr layer aids the adhesion of the A u film; titanium could also be used. This mask thickness was calculated for 255keV A r + implantation, and can obviously be adjusted to suit implantation conditions. Considering that ion straggle will result in ions penetrating some lateral distance un-der the mask, the ratio of masked:unmasked area should be fairly high to preserve lateral selectivity. For example, ratios of 3:1 in a 435nm-pitch grating for 255keV A r + implan-tation [9] and 7:3 in 200-260 nm-period gratings for 170keV G a + implantation [10] have been used. In those works a negative electron beam resist was used for the patterning. The following is a process developed using A M P E L facilities: 1. Sample cleaning: If the sample has been handled it is wise to clean it before depositing the mask. A typical cleaning process would include a solvent rinse consisting of acetone and 2-propanol or ethanol, followed by de-ionized water and N 2 blow-dry, strong HC1 to remove native oxide, another rinse in de-ionized water and a final blow-dry with N 2 . The sample can be baked briefly on a hotplate > 1 0 0 ° C to drive off water from the back side, and placed immediately under vacuum to prepare for electron-beam evaporation of AI2O3. If the sample is a pristine M B E growth, this "cleaning" process may realistically introduce more contamination than it removes. In practice the amount of handling to which a sample is subjected is a serious factor in the quality of the final product, and should be minimized. 2. A1 20 3 deposition: Approximately 20 nm of A 1 2 0 3 is deposited directly on the sample's GaAs surface by electron-beam evaporation. As is usual with dielectrics, AI2O3 glows very brightly during evaporation but most of the chunks remain solid in the crucible, some pieces in the center of the beam likely fusing together. During electron-beam evaporation of any material, it is advisable to wear eye protection when looking directly at the glowing source material. 3. P M M A preparation: P M M A pre-dissolved in chlorobenzene for use as an electron beam resist must be further diluted in order to be applied in a very thin A P P E N D I X A : G O L D M A S K I N G F O R S E L E C T I V E I M P L A N T A T I O N 63 coating. Older P M M A stock will be thicker than a freshly-prepared batch, but as the layer required in this case is much thinner than when used as a resist, it will be necessary in any case to add several parts chlorobenzene to each part pre-mixed P M M A . Any attempt to substitute acetone for chlorobenzene as a thinner will be rewarded with a useless, streaked coating. It was found that the most uniform coating of a small sample is achieved by applying the P M M A solution before spinning, ensuring that the liquid flows over the sample edges to touch the spinner chuck, rather than beading on top of the sample. After the solution is applied, the sample is spun at lOOOOrpm for 40s. The sample is then baked on a hotplate at 180°C for several hours, to remove the solvents. It is important that this P M M A layer be very thin (~15-20nm, as opposed to, for instance, 100nm), or it will evolve solvents during the subsequent baking and/or E C R etching steps, and the gold mask will bubble or drift. 4. Metal deposition: Immediately following removal from the hotplate, the sample is transferred to the vacuum chamber of the electron-beam evaporation system. Approximately 3nm of T i is laid down first to aid adhesion of the Au. Titanium, when melted, wets out to a great extent, spreading over the entire inside surface of the crucible. There is a very strong possibility of the metal crawling over the wall of the crucible to make contact with the copper hearth, if the crucible is too full or if the beam current is not carefully controlled. If the titanium does touch the hearth, this can create a very effective heat sink and result in thermal stresses severe enough to shatter a carbon crucible. Aluminum exhibits similar behaviour. Gold deposition follows the titanium immediately. Gold is quite well-behaved dur-ing electron-beam evaporation, its tendency being to "ball up" upon melting rather than to spread out. The main concerns are to avoid sudden increases in beam cur-rent which may cause the accidental evaporation of a large quantity of the expen-sive source material, and to adjust the beam sweep, focus, and current to minimize A P P E N D I X A : G O L D M A S K I N G FOR S E L E C T I V E I M P L A N T A T I O N 64 "spitting" which tends to deposit undesirable micron-sized balls of gold on the mask. 5. P a t t e r n i n g : A grating pattern must now be delineated on the mask mate-rial. This can be done by conventional electron-beam lithography, using undiluted P M M A as a resist. This is spun at ~8 000rpm for 40s and baked for at least two hours at 180°C before electron-beam exposure. It was found that the electron doses used to expose P M M A on GaAs were too high for P M M A on gold and linewidths were larger in general. If the scattering of electrons by gold becomes a resolu-tion issue with P M M A , it may be necessary to use a negative resist; for the high duty-cycle grating the larger linewidth would then be acceptable. 6. E t c h i n g : The gold mask can be etched using an argon plasma in the A M P E L cleanrooms E C R etcher. The following parameters were found to be promising: 12mTorr process pressure 5Torr backside He 210 W microwave power -150 V D C bias A mask containing approximately 200 nm of gold with a P M M A layer beneath was penetrated in a three minute etch under these parameters. Using the same parameters but with -250 V D C bias resulted in carbonization of the P M M A . Appendix B: Wet Etching B.l : GaAs/Ala;Gai_xAs • C 6 H 8 0 7 (citric acid) : H 2 0 2 : H 2 0 This can be used for etching GaAs in preference to A h - G a ^ A s . We found the combination 6 C 6 H 8 0 7 : 1 H 2 0 2 : 40 H 2 O t o be a gentle, controllable etchant, suitable for removing ~100nm of GaAs in ~three minutes. This mixture stops etching at a layer of Alo.9gGao.02As or AlAs, leaving a clean, flat surface exposed to the next etchant. The etch rate was faster in broad exposed areas several microns wide than in very narrow (submicron) electron-beam-defined patterns. Selectivities and etch rates vary with the proportions of the ingredients. Refer-ence [54] is a useful guide for this etchant. • HC1 : H 2 0 2 : H 2 0 Etches GaAs. A ratio of 40 HC1 : 4 H 2 0 2 : 1 H 2 0 will etch isotropically at a high rate (>5 /xm/min) [55]. We found this too fast for controlled etching of fine patterns, as was 8 HC1 : 1 H 2 0 2 : 20 H 2 OExtreme dilution to a ratio of 8 HC1 : 1 H 2 0 2 : 100 H 2 0 resulted in a very shallow, very wide etch, unsuitable for our purposes. • H 2 S 0 4 : H2O2 : H 2 0 This is a common combination for nonselective etching of G a A s / A l x G a 1 _ x A s layers, characterized in Reference [56]. It was used to expose G a A s / A l x G a i _ x A s multilayers in References [20, 31] in the proportions 1 H 2 S04 : 8 H 2 0 2 : 80 H 2 0 and in Reference [19] in the proportions 4 H 2 S 0 4 : 1 H 2 0 2 : 1 H 2 0 . We found that the use of a selective etch for each layer gave finer control for etching shallow layers. 65 • H F Hydrofluoric acid is an excellent etchant for high-a; Al^Gai^^As, and does not attack GaAs. We have found that for AlAs and Alo.9sGao.02As, 0.6% unbuffered H F in H 2 0 is too strong for controlled etching of fine patterns, resulting in uneven etching and excessive undercut, as well as attacking P M M A . A solution of 0.06% H F in H 2 0 is satisfactory: a ~5s dip will remove a thin (—25nm) layer of AlAs or Alo.9sGao.02As. • HC1 Hydrochloric acid will attack high-a; A L j G a i ^ A s but not GaAs. If, during sample cleaning, HC1 reaches a thin layer of AlAs or Alo.9sGao.02As through any "pinholes" or scratch damage in the GaAs cap, it will create etched regions in that layer with very straight rectangular edges aligned with the {011} (cleave) axes. Williams [55] contains a summary of GaAs etch rates and sidewall profiles for the H 2 S 0 4 : H 2 0 2 : H 2 0 and HC1 : H 2 0 2 : H 2 0 systems, and Adachi and Oe [36] provide a useful survey of the effects of various etchants on (001) GaAs. Measures of citric acid refer to a mixture of 50% anhydrous citric acid to 50% H 2 0 by mass. Measures of H 2 0 2 assume a "full-strength" solution contains 30% H2O2 and 70% H 2 0 . However, H 2 0 2 does decompose over time, and the stock used in these experiments is over a year old. The recipes used in this work can be assumed to contain a lower H 2 02 : H 2 0 ratio than the measured volumes imply. B.2: InGaAs/InAlAs • C 6 H 8 0 7 (citric acid) : H 2 0 2 : H 2 0 This etches I n x ^ A L A s in preference to In^^AlyAs. Characterization of this system can be found in References [57, 54]. A solution of 10 C 6 H 8 0 7 : 1 H 2 0 2 , which according to [57] should give a selectivity of 2.5, was used to etch through —500 nm of Ini-^Al^As and ~300nm of tensile Ini-^Al^As in 8 minutes. This resulted 66 in sidewalls with a slope of approximately 55°. The higher the citric acid : H2O2 ratio, the lower the selectivity [57, 54]. B.3: Removal of Dielectrics SiG"2 and S r F 2 can be removed with H F [58]. To achieve a significant etch rate the concentration must be at least several percent (preferably more), but this is many times the 0.06% we use to etch AlAs and Alo.9sGao.02As, so care must be taken. It is inadvisable to use an ultrasonic bath with H F if the sample contains high-x A l x G a i _ x A s . B.4: "Stain Etch" It is often useful to enhance the visibility of layer interfaces under cross-sectional S E M observation. This can be achieved with a dip in 1 K 3 F e ( C N ) 6 : 1.5 K O H : 600 H 2 0 , which attacks the interfaces along a cleaved edge. A n exposure of several seconds is sufficient (see Figure 3.2 for an example of the effect of a 10 s dip on a G a A s / A L J G a 1 _ . r A s multilayer). 67 Appendix C: Electron Cyclotron Resonance Etching This Appendix consists of a collection of S E M images of samples subjected to E C R etching in the A M P E L cleanrooms PlasmaQuest E C R II system. Ar C l 2 BC1 3 Etch Time 10 seem Osccm lsccm 120 s 10 seem Osccm lsccm 180 s 9 seem 2 seem lsccm 120 s Table 4.1: S E M cross-sections of ECR-etched R783. In each case the 100 nm Alo.98Gao.02As layer has been oxidized. In the middle image, the GaAs cap has been completely removed, exposing the entire Alo.98Gao.02As layer which has then oxidized. Microwave power was 100W, D C bias -100 V , process pressure lOmTorr, and backside He 5Torr. Figure 4.1 is a survey of attempts to transfer 800-nm-pitch gratings into GaAs by E C R etching with argon and chlorine. The Si02 grating mask was fabricated by Yan Feng of the Institute for Microstructural Sciences of N R C . 68 Bibliography [1] F . Sfigakis, P. Paddon, V . Pacradouni, M . Adamcyk, C. Nicoll, A . R. Cowan, T . Tiedje, and J . F . Young, J . Lightwave Technol. 18(2), 199 (2000). [2] E . Palik, Handbook of Optical Constants of Solids (Academic Press, 1985). [3] H . Chen, R. M . Feenstra, P. G . Piva, R. D. Goldberg, I. V . Mitchell, G . C . Aers, P. J . Poole, and S. Charbonneau, Appl. Phys. Lett. 75(1), 79 (1999). [4] S. Janz, M . Buchanan, P. van der Meer, Z. R. Wasilewski, D . -X. Xu , P. Piva, I. V . Mitchell, U. G . Akano, and A. Fiore, Appl. Phys. Lett. 72(24), 3097 (1998). [5] P. J. Poole, S. Charbonneau, G . C. Aers, T . E . Jackman, M . Buchanan, M . Dion, R. D. Goldberg, and I. V . Mitchell, J . Appl. Phys. 78(4), 2367 (1995). [6] J . E . Haysom, A . Delage, J . -J . He, E . S. Koteles, P. J. Poole, Y . Feng, R. D. Goldberg, I. V . Mitchell, and S. Charbonneau, I E E E J. Quantum Electron. 35(9), 1354 (1999). [7] M . I. Cohen, H . H . Tan, and C . Jagadish, J . Appl. Phys. 85(11), 7964 (1999). [8] A . J. Steckl, P. Chen, X . Cao, H. E . Jackson, M . Kumar, and J . T . Boyd, Appl. Phys. Lett. 67(2), 179 (1995). [9] V . Hofsafi, J . Kuhn, C. Kaden, V . Harle, H . Bolay, F . Scholz, H . Schweizer, H. Hillmer, R. Losch, and W . Schlapp, Nucl. Inst. & Meth. in Phys. Res. B 106, 471 (1995). [10] C. Kaden, H.-P. Gauggel, F . E . Prins, M . Burkard, F . Adler, H . Schweizer, and G . W. Smith, Appl. Phys. Lett. 67(7), 888 (1995). [11] H . Konig, J . P. Reithmaier, A . Forschel, J . L . Gentner, and L . Goldstein, Appl. Phys. Lett. 73(19), 2703 (1998). [12] H . Konig, S. Rennon, J. P. Reithmaier, A . Forschel, J . L . Gentner, and L . Goldstein, Appl. Phys. Lett. 75, 1491 (1999). [13] H . Konig, S. Rennon, J. P. Reithmaier, and A. Forschel, in 11th International Con-ference on Indium Phosphide and Related Materials, I E E E (IEEE, 1999), pp. 29-32. [14] P. Chen and A . J . Steckl, J . Appl. Phys. 77(11), 5616 (1995). [15] A . Pepin, C . Vieu, M . Schneider, R. Planel, J . Bloch, G . B. Assayag, H . Launois, J . Y . Marzin, and Y . Nissim, Appl. Phys. Lett. 69(1), 61 (1996). •70 B I B L I O G R A P H Y 71 [16] J . M . Dallesasse, P. Gavrilovic, N. H . Jr., R. Kaliski, D. W . Nam, E . J. Vesely, and R. D. Burnham, Appl. Phys. Lett. 56, 2436 (1990). [17] F . Sfigakis, Optical Characterization of AlxGa\-xAs Oxides, M.Sc. thesis, University of British Columbia (1999). [18] K . D. Choquette, K. M . Geib, H. C. Chui, B. E . Hammons, H . Q. Hou, and T . J . Drummond, Appl. Phys. Lett. 69(10), 1385 (1996). [19] T . Takamori, K. Takemasa, and T. Kamijoh, Appl. Phys. Lett. 69(5), 659 (1996). [20] T . Langenfelder, S. Schroder, and H. Grothe, J . Appl. Phys. 82(7), 3548 (1997). [21] R. D. Twesten, D. M . Follstaedt, K. D. Choquette, and J. R. P. Schneider, Appl. Phys. Lett. 69(1), 19 (1996). [22] S. Guha, F . Agahi, B. Pezeshki, J . A . Kash, D. W . Kisker, and N. A . Bojarczuk, Appl. Phys. Lett. 68(7), 906 (1996). [23] K . D. Choquette, K. M . Geib, C. I. H . Ashby, R. D. Twesten, O. Blum, H . Q. Hou, D. M . Follstaedt, B. E . Hammons, D. Mathes, and R. Hull, I E E E J. Selected Topics in Quantum Electron. 3(3), 916 (1997). [24] F. A . Kish, S. J . Caracci, N . H. Jr., K. C. Hsieh, J. E . Baker, S. A . Maranowski, A . R. Sugg, J . M . Dallesasse, R. M . Fletcher, C. P. Kuo, T . D. Osentowski, and M . G . Craford, J . Electron. Mat. 21(12), 1133 (1992). [25] R. L . Naone and L . A . Coldren, J . Appl. Phys. 82(5), 2277 (1997). [26] B. Koley, M . Dagenais, R. Jin, G . Simonis, J . Pham, G . McLane, F . Johnson, and J . R. Whaley, J. Appl. Phys. 84(1), 600 (1998). [27] M . R. Krames, E . I. Chen, and J . N . Holonyak, Appl . Phys. Lett. 65(25), 3221 (1994). [28] S. A . Feld, J . P. Loehr, R. E . Sherriff, J . Wiemeri, and R. Kaspi, I E E E Photon. Technol. Lett. 10(2), 197 (1998). [29] K . Koizumi, P. O. Vaccaro, K. Fujita, M . Tateuchi, and T . Ohachi, J . Crystal Growth 198/199, 1136 (1999). [30] P. O. Vaccaro, K. Koizumi, K. Fujita, and T . Ohachi, Microelectronics Journal 30, 387 (1999). [31] A . C . Alonzo, X . C. Cheng, and T . C. McGil l , J . Appl. Phys. 84(12), 6901 (1998). B I B L I O G R A P H Y 72 C. I. H . Ashby, J . P. Sullivan, K . D. Choquette, K . M . Geib, and H. Q. Hou, J . Appl. Phys. 82(6), 3134 (1997). C. I. H . Ashby, M . M . Bridges, A . A. Allerman, B. E . Hammons, and H. Q. Hou, Appl. Phys. Lett. 75(1), 73 (1999). C . I. H . Ashby, J . P. Sullivan, P. P. Newcomer, N. A . Missert, H . Q. Hou, B. E . Hammons, M . J. Hafich, and A. G . Baca, Appl. Phys. Lett. 70(18), 2443 (1997). D. J . Stirland and B. W. Straughan, Thin Solid Films 31, 139 (1976). S. Adachi and K . Oe, J. Electrochem. Soc. 130(12), 2427 (1983). D. L . Huffaker, D. G . Deppe, C. Lei, and L . A . Hodge, Appl. Phys. Lett. 68(14), 1948 (1996). W . D. Laidig, J . N. Holonyak, M . D. Camras, K. Hess, J . J . Coleman, P. D. Dapkus, and J. Bardeen, Appl. Phys. Lett. 38(10), 776 (1981). D. G . Deppe, L . J . Guido, J . N. Holonyak, K . C . Hsieh, R. D. Burnham, R. L . Thornton, and T . L . Paoli, Appl. Phys. Lett. 49(9), 510 (1986). W . Xia, S. A . Pappert, B. Zhu, A. R. Clawson, P. K. L . Yu, S. S. Lau, D. B. Poker, C . W . White, and S. A . Schwarz, J . Appl. Phys. 71(6), 2602 (1992). A. R. Massengale, C. Y . Tai, M . D. Deal, J . D. Plummer, and J. J . S. Harris, in 1997 IEEE International Symposium on Compound Semiconductors, I E E E (IEEE, 1998), pp. 329-332. W . D. Zhou, P. Bhattacharya, J . Sabarinathan, and D. H. Zhu, J. Vac. Sci. & Tech. B 18(3), 1706 (2000). S.-T. Lee, G . Braunstein, P. Fellinger, K . B . Kahen, and G . Rajeswaran, Appl . Phys. Lett. 53(25), 2531 (1988). M . C. Ridgway, private communication. C. Vieu, M . Schneider, H . Launois, and B. Descouts, J . Appl. Phys. 71(10), 4833 (1992). T . E . Schlesinger and T . Kuech, Appl. Phys. Lett. 49(9), 519 (1986). D. Hofstetter, H . P. Zappe, J . E . Eppler, and P. Riel, Appl. Phys. Lett. 67(14), 1978 (1995). [48] I. Gontijo, T . Krauss, J . H . Marsh, and R. M . D. L . Rue, I E E E J . Quantum Electron. 30(5), 1189 (1994). B I B L I O G R A P H Y 73 [49] S. Biirkner, M . Maier, E . C. Larkins, W . Rothemund, E . R O'Reilly, and J. D. Ralston, Journal of Electronic Materials 24(7), 805 (1995). [50] T . Yoshikawa, H . Saito, H . Kosaka, Y . Sugimoto, and K. Kasahara, Appl. Phys. Lett. 72(18), 2310 (1998). [51] H . H . Tan and C. Jagadish, Appl. Phys. Lett. 71(18), 2679 (1997). [52] H . H . Tan, J. S. Williams, and C. Jagadish, J . Appl. Phys. 78(3), 1481 (1995). [53] C. Kaden, H.-P. Gauggel, V . Hofsafi, A . Hase, H . Schweizer, and H. Kiinzel, Appl. Phys. Lett. 65(25), 3170 (1994). [54] G . C. DeSalvo, W . F . Tseng, and J. Comas, J. Electrochem. Soc. 139(3), 831 (1992). [55] R. Williams, Modern GaAs Processing Methods (Artech House, Inc, 1990). [56] S. Iida and K. Ito, Solid State Science 118(5), 768 (1971). [57] M . Tong, A . A . Ketterson, K. Nummilia, and I. Adesida, in IEEE 4th International Conference on InP and Related Materials, I E E E (IEEE, 1992), pp. 298-301. [58] B. S. Ooi, S. G . Ayling, A . C . Bryce, and J . H . Marsh, I E E E Photon. Technol. Lett. 7(9), 944 (1995). 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0085178/manifest

Comment

Related Items