20081011_3891841_0

8 Extended Defects in Wurtzite CaN Layers: Atomic Structure, Formation, and Interaction Mechanisms PIERRE RUTERANA, ANA M. SANCHEZ, and GERARD NOUET Abstract Using high-resolution electron microscopy, atomistic modeling and image simula- tions, typical contrast was identified for the a pure edge threading dislocations in GaN layers grown by MBE on sapphire or SiC. Their atomic structure was shown to exhibit 5/7, or 8 atom cycles. A topological analysis of high angle grain bound- aries was carried out in order to determine the defect content at the interfaces. The reconstruction of some boundaries was only possible by taking into account the occurrence of structural units that exhibit 4 atom ring cycles for the disloca- tion cores. The pure edge threading dislocations were shown to be connected to misfit dislocations at the GaN/Al2O3 interface. The {1210} stacking fault has two atomic configurations in wurtzite (Ga,Al, In)N with 1/2 (10Tl> and 1/6 (2023) displacement vectors. It originates from steps at the SiC surface and it can form on a flat (0001) sapphire surface. These configurations have comparable energy in AlN, whereas the 1/2 (1011) {1210} is more stable in GaN and InN. In the investigated samples, the {1010} inversion domain boundaries exhibit two atomic configurations (Holt and V models) depending on the growth condi- tions. The samples containing Holt inversion domains have a flat surface mor- phology, whereas the V IDBs were observed in the centre of small pyramids (100 nm high) protruding at the sample surface. The Holt inversion domains were always smaller (<20nm), at higher densities (2.5xl010cm~2), whereas the V ones reach 50 nm and one order of magnitude lower density. The two atomic con- figurations have now been shown to be able to coexist inside the same sample, mainly due to interactions with the basal stacking faults. The inversion domains were found to be generated mostly at surface steps where they minimized the large misfit along the c axis (20%). 8.1 Introduction The GaN-based IH-V semiconductors have large direct band gaps which make them excellent candidates for short wavelength opto-electronic applications [I]. Among these materials, GaN and AlN are of particular interest as their solid solu- tions or superlattice structures allow bandgap engineering in the range of 3.45- 6.28 eV. As bulk crystals or wafers of these materials are not available, GaN has to be epitaxially grown on a large variety of substrates. The wurtzite allotropic struc- ture of GaN is thermodynamically the most stable, but cubic substrates have been used to stabilize the zinc blende phase since cubic semiconductors are more read- ily doped [2]. The most commonly used substrates are sapphire and a-SiC, which have hexagonal symmetry. SiC may be the most promising for its small thermal and lattice mismatch with GaN [3], although the majority of work that has been reported is on sapphire substrates which present a large lattice mismatch (16%). This was due to the good results obtained in deposition by metal organic chemi- cal vapor deposition (MOCVD) on sapphire [4] and difficulties in surface prepara- tion for SiC substrates. Recently, enhancement in the substrate cleaning proce- dure by addition of a hydrogen plasma step has allowed to overcome the latter problem and it is now possible to deposit GaN layers by electron cyclotron reso- nance (ECR) enhanced molecular beam epitaxy (MBE) [3]. The active GaN layers contain large densities of crystallographic defects, among which those that can cross the whole epitaxial layer and be detrimental to the elec- tro-optical properties are the threading dislocations [5, 6], the nanopipes [7], the inversion domains [8, 9], and the prismatic planar defects [8, 10, H]. As pointed out by many workers, the large majority of these defects is made of threading dis- locations which originate from the particular growth mode of GaN on top of the (0001) sapphire or SiC substrates. This mosaic growth mode leads to islands that are rotated mostly around the c axis and therefore are bounded by mainly a edge dislocations [12, 13]. In a conventional semiconductor, such as silicon or GaAs, such defect densities in the order of 1010 cm"2 would result in nonusable layers. In GaN, commercial LEDs are made from such layers, this has led many workers to conclude that extended defects would have a negligible electro-optical activity, such as nonradiative recombination [14]. In the case of threading dislocations, the nonradiative activity was tentatively explained by the reconstruction of the core, which eliminates the dangling bond in the 8 atom cycles [15, 16]. Unfortunately, this explanation does not hold for the laser diodes in which it was possible to in- crease the lifetime to 10,000 hours [17] only by introducing new growth techni- ques in order to have areas where the dislocation density could be lower than 106 cm"2 [18]. The {1210} planar defects have been called double positioning boundaries (DPBs) [10, 19], stacking mismatch boundaries (SMBs) [20] or inversion domain boundaries (IDBs) [8]. These faults have been investigated using high-resolution electron microscopy (HREM) and Convergent Beam Electron Diffraction (CBED), and it was shown that they are stacking faults on top of both sapphire and SiC [21, 22]. In fact, these planar defects have already been studied in the sixties and two displacement vectors have been measured by conventional microscopy [23, 24]. In ceramic AlN, Drum [24] investigated faults which intersected on the basal and prismatic {1210} planes, it was shown that the displacement vector was 1/2 (1011) and that they folded to the basal planes by leaving a 1/6 (1010) stair rod dislocation at the intersection. Almost at the same time, Blank et al. [23] were the first to study the planar defects which folded from basal to prismatic {1210} planes in wurtzite ZnS, and to interpret them as stacking faults whereas other authors considered them to be thin lamella of the sphalerite phase in CdS [25]. These prismatic faults were then shown to be growth domains and the displace- ment vector was found to be 1/6 (2023) which is the same as that of the I1 basal stacking fault in the hexagonal compact packed (hep) structure. GaN is of wurtzite structure, which is noncentrosymmetric. In such structures, defects due to polarity were reported for the first time by Aminoff and Broome [26]. Since then, several terms have been used for them, such as inversion twins [27], or antiphase boundaries [28]. Westwood and Notis [29] discussed the confu- sion between antiphase and inversion domain boundary (IDB) and showed it to be due to a too general definition of antiphase boundary that includes IDB. For example, using the two-colour symmetry theory, Pond and Holt [30] showed that the antiphase boundary described by Holt [28] was in fact an IDB. Inversion do- main boundaries have been analysed in several materials like wurtzite BeO [27] and sphalerite GaAs [31]. Two models for the atomic structure of IDBs were pro- posed by Austerman and Gehman [27] and by Holt [28], respectively. The Auster- man model has the anion sublattice continuous across the boundary whereas the cations switch from one type of tetrahedral site to the other. This model was found to agree with HREM observations in ZnO [32] as well as in ZnSe [33]. In the Holt model, the cations and anions are exchanged across the boundary, lead- ing to the formation of anti-sites or wrong bonds (A-A, B-B). Its occurrence was reported in SiC [34] as well as in GaAs epitaxially grown on Si [31]. During the last decade, many studies have been carried out in sintered AlN and essentially two morphologies were observed for the IDBs: a planar variant lying in the (0001) basal plane and a curved one [35-37]. In epitaxial GaN layers, the IDBs were observed on sapphire substrate and not on SiC as it was shown in layers grown in the same conditions on both substrates [38]. A detailed study of the polarity of GaN/SiC layers has shown that the layers are unipolar [22]. In the unique work which reported IDBs in GaN grown on SiC substrate, a thin amorphous layer was present at the interface with the substrate [39]. Various morphologies have been reported for IDBs in GaN layers grown on sapphire substrate: they can be either bounded by {1010} planes and cross the whole epitaxial layer [40], or be limited by {10T0}, {1011} and {10l2} planes and form house-shaped domains buried near the interface with the substrate [41]. They have been observed for various growth techniques: MBE, MOCVD or hybrid vapor phase epitaxy (HVPE) [9]. Many reports are now available on the IDBs atomic structure: although a translation on the basal plane has been reported [41], the most recent investigations have excluded it. In their transmission electron mi- croscopy (TEM) work, Cherns et al. [42] studied displacement fringes in GaN layers grown by MOCVD and confirmed the model with a c/2 translation, first proposed by Northrup et al. [9]. In a HREM work, on GaN layers grown by MBE, Potin et al. [43] reported the occurrence of Holt type IDBs. For the time being, there is still a controversy on the role of the defects on the device performances. The high efficiency of the light emitting diodes seems to in- dicate that most of the defects are inactive whereas the short lifetimes of laser diodes may be attributed to them. Anyway, it is clear that an improvement of the crystalline quality of the layers would allow higher performances of the devices. For the fabrication of GaN-based devices, an important step in the nitride growth progress was the use of AlN or GaN buffer layers by Akasaki et al. [44] who showed that it greatly improved the crystalline quality of GaN layers, espe- cially those grown on sapphire substrates. In the present work, an extensive report is made on the atomic scale structural analysis of the extended defects in GaN and AlN layers. The results were obtained by the HREM technique on the core structure of threading dislocations, prismatic stacking faults and IDBs. Finally, the possible formation mechanisms for each type of crystallographic defects will be discussed. 8.2 Crystallographic Considerations 8.2.1 Substrates 8.2.1.1 Sapphire The corundum form of AI2O3 belongs to the trigonal R3c spacegroup (167), taking the origin at 3c, as in the International Tables for Crystallography, the oxygen is located at (x,y,z) = (0.306,0,0.25). If this position is approximated to (x,y,z)«(1/3,0,1/4), the anion framework forms an hep lattice (Fig. 8.1) with a=0.476nm and c= 1.299 nm. The Al3+ occupy 2/3 of the octahedral sites but is located at (%,y,z) = (0,0,0.352) instead of (0,0,1/3), thus the cations are shifted by ±0.025 nm along the c axis from the ideal octahedral sites. The oxygen ion is larger than the aluminium ion (r"/r+^3), therefore the study of the steps on the substrate can be limited to the steps in the oxygen framework, leading to step heights which are multiples of c/6 (d(Ooo6) ~ 0.216 nm). The (0001) Al2O3 surfaces are oxygen terminated [45] and present steps along {1120} and {1010} planes [46]. Two crystallographically equivalent surfaces are connected by a symmetry operation of the space group; along the [0001] direction, A-A or B-B surfaces are separated by c/3, 2c/3 and c steps. A step separating two "A" surfaces will be noted for instance (A-A,c/3). Steps of height c/6, c/2 or 5c/6 separate two surfaces related by a glide symmetry operator, such steps are called demisteps, they will be noted for example (A-B,c/6) [47]. 8.2.1.2 Silicon Carbide In contrast to most of the compound semiconductors, silicon carbide (SiC) is a IV-IV alloy which exhibits a large number of polytypes. Like the other compound semiconductor materials, it is tetrahedrally coordinated and its structure can be considered as a network of corner sharing tetrahedra [48]. SiC has a number of advantages over sapphire (Table 8.1): - In the basal plane, its mismatch to GaN is only 3.5% and an AlN buffer layer allows to decrease it to less than 2.5%. - Moreover, its thermal expansion coefficients are also closer to those of GaN. - It is a semiconductor and as such exhibits a good electrical conductivity which may be used for the backside ohmic contacts. Tab. 8.1 Crystallographic data on wurtzite GaN, AlN and the substrates: sapphire and 6H-SiC Fig. 8.1 A schematic diagram of the AI2O3 sapphire unit cell, there are 6 oxygen layers in the unit cell, the dis- tances between the various atomic layers change as shown in the figure. The oxygen ions form a pseudohexago- nal lattice. The small Al ions occupy the octahedral sites. b B a A b C a B b A a Material GaN AlN SiC Sapphire a (nm) 0.3189 0.3112 0.308 0.476 a/asic (%) 3.54 1.04 a/Osapph. {%) 16.09 13.29 c (nm) 0.5185 0.4982 1.512 1.2991 c/cssc (%) 2.9 1.15 Cfcsapph. (%) 19.7 15.06 Polytype 2H 6H SC P63mc R3c The most used polytype is the 6H, but the 3C is also currently used mainly as a nucleation layer for the growth of cubic GaN on top of silicon. A comprehensive description of the tetrahedrally coordinated structures is very useful for the inves- tigation of the interface defects. The bonds describe a tetrahedron denoted T which possesses one atom species at each corner and the other atom species in its center [48]. The basal plane of this structure is defined by one face of the tetra- hedron and the bond perpendicular to this latter defines the c axis. A rotation of 180° around c produces a twinned variant denoted T (Fig. 8.2 a). The two variants are also related to each other by a mirror symmetry about a {1100} plane. A tetrahedron can occupy one of the three possible positions in the basal plane and layers of tetrahedra are then denoted T1, T2, T3, Ti, T2, T3 (Fig. 8.2 b). The structure of these materials and their different polytypes can be completely described by stacking these six tetrahedra layers. All stacking sequences are not possible and two rules must be respected to keep a corner sharing structure: (i) A tetrahedron T can be followed by another one of the same kind with the fol- lowing subscript: T1 T2 T3 and inversely for the twinned variant: T3 T2 Ti. (ii) A tetrahedron T1 must be followed by the twinned variant of the preceding subscript: T1T3 and inversely for its twinned variant: Ti T2. For example, the structures of the different polytypes, under the present study, are described by the following sequences: Tl T3 or T2 T l or T3 T'2 for the wurtzite structure, which corresponds to the polytype denoted 2H in the Ramsdel notation, Tl T2 T l T'3 for the 4H polytype, Tl T2 T3 T'2 T l T'3 for the 6H polytype, Tl T2 T3 or T'3 Tl T l for the 3C polytype. (a) (b) Atom a Atom b Fig. 8.2 Representation of the tetrahedrally coordinated materials: a the two possible tetrahe- dra; b the rules for stacking the tetrahedra. 8.2.3 Epitaxial Relationships On top of the (0001) sapphire, the GaN layers have the well-known epitaxial rela- tionship: (0001)substrate//(0001)GaN and [1120]substrate//[0Tl0]GaN. This relation leads to the continuation of the anion compact stacking (O2~ in the substrate, N3~ in the thin film). The apparent 30° rotation between the two crystals is due to the choice of the crystallographic origin (in the substrate, the origin is related to the Al3+ lattice). At the interface, the cations switch from octahedral (Al3+ in Al2O3) to tetrahe- dral sites (Ga3+ in GaN). So, an important parameter could be the coordinate poly- hedra of the cations that is between an oxygen plane and a nitrogen plane at the interface. There are three polyhedra to consider: each one corresponds to one of the three interstitial sites of the hep structure (2 tetrahedral sites plus 1 octahedral site). Fig. 8.3 Schematic diagram show- ing the fa and /?2 tetrahedral sites of GaN unit cell. If we supposed that N occupies the y sites, only one family of/? sites can be simultaneously oc- cupied by Ga atoms. 8.2.2 Epitaxial Layers GaN crystallizes in the cubic structure (sphalerite) or in the hexagonal structure (wurtzite). The latter is thermodynamically more stable and in conventional growth conditions on (0001) sapphire substrate, GaN is hexagonal (wurtzite: space group P63mc), with the lattice parameters a=0.319nm and c=0.518nm. The an- ions (N3") form an hep structure in which the cations (Ga3+) occupy half of the tetrahedral sites. The structure of a unit cell of GaN projected along [0001] is depicted schemati- cally in Fig. 8.3. The open symbols represent y sites, which are occupied by nitro- gen atoms, the Ga atoms are in the tetrahedral sites /?. These latter sites can either be at heights 3/Sc above (fii) or below (J]2) each N site, depending on the crystal polarity. Furthermore, the hep anion stacking does not have the same parameters in both materials. This leads to mismatch in the a and c directions of an hep struc- ture. In the basal plane, the mismatch is equal to: where asUb. = asub./\/3> so as to respect the 30° rotation. It was shown, after Bragg filtering of HREM images, that this lattice misfit is largely relaxed in the first monolayer of GaN, with a network of regularly spaced 60° dislocations. A residual stress of -2.1% was evaluated near the interface and related to the high density of threading defects present in the epitaxial layers [49]. In the c direction, the difference of lattice parameters leads to an even larger mis- match: where Csub. = Csub./3. This mismatch may act at the substrate surface steps as the growth takes place at adjacent terraces. In the growth on 6H-SiC, there is no rotation of the lattice, the misfits are shown in Table 8.1. 8.2.4 Bicrystallographic Analysis of lnterfacial Defects We now consider the symmetry of dichromatic complexes, i.e., configurations which can be visualized by allowing two crystals relatively rotated to interpenetrate. The configuration created by two lattices rather than two crystals is called dichromatic pattern [50]. We have used for symmetry operations and transformations the nota- tion described in the International Tables for Crystallography. Moreover, let one crys- tal be designated white (X), and the other black (//). The relative orientation is defined by the operation P=(P, p) which transforms white vectors into corresponding black ones (expressed in the white coordinate frame). P may either be a rotation, thereby relating crystals of the same polarity, or a rotoinversion relating crystals of opposite polarity. The translation p represents the shift of the black crystal origin following the operation P. This formalism enables the symmetry of dichromatic complexes to be established for any relative orientation and position of the X and ju crystals. Co- incident symmetry operations in such complexes arise when black and white crystal symmetry operations coincide and correspond to the intersection of the crystal spacegroups, i.e., 0(X) D 0(ju). Explicitly, the coincident operations in the com- plex, W, are the solutions to the following identity [51]: (i) This set will include an infinite number of translations parallel to [0001] for all ro- tations, and a 3-dimensional set also known as the coincident site-lattices (CSLs) for special misorientations, independent of any translation p. Other coincident op- erations arising for any value of rotation about [0001] may be broken by transla- tion p. In addition, colour reversing or anti-operations, which relate black features to white and vice versa will also arise. These can be identified by inspection of the multiplicity of equivalent description of P, i.e. P W(A)i. Those operations in this set which have the form of symmetry operations are antioperations in the space- group of the complex, W(c)k, i.e. (2) The space group of a dichromatic complex is then given by the extension of the coincident translation groups by the set W(c)k U W"(c)k