Surface Review and Letters, Vol. 6, No. 6 (1999) 11291141 c World Scientic Publishing Company
SiC SURFACE RECONSTRUCTION: RELEVANCY OF ATOMIC STRUCTURE FOR GROWTH TECHNOLOGY
U. STARKE, J. BERNHARDT, J. SCHARDT and K. HEINZ Lehrstuhl fr Festkrperphysik, Universitt Erlangen-Nrnberg, u o a u Staudtstr. 7, D-91058 Erlangen, Germany Received 12 September 1999
Growth of SiC wafer material, of heterostructures with alternating SiC crystal modications (polytypes), and of oxide layers on SiC are of importance for potential electronic device applications. By investigation of hexagonal SiC surfaces the importance of atomic surface structure for control of the respective growth processes involved is elucidated. Dierent reconstruction phases prepared by ex situ hydrogen treatment or by Si deposition and annealing in vacuum were analyzed using scanning tunneling microscopy (STM), Auger electron spectroscopy (AES) and low-energy electron diraction (LEED) crystallography. The extremely ecient dangling bond saturation of the SiC(0001)-(33) phase allows step ow growth for monocrystalline homoepitaxial layers. A switch to cubic layer stacking can be induced on hexagonal SiC(0001) samples when a ( 3 3)R30 phase is prepared. This might serve as seed for polytype heterostructures. Finally, we succeeded in preparing an epitaxially well matching silicon oxide monolayer with ( 3 3)R30 periodicity on both SiC(0001) and SiC(000 This initial 1). layer promises to facilitate low defect density oxide lms for MOS devices.

1. Introduction

Dierent prospective electronic applications of SiC put specic demands on the properties of SiC material used for the device production process. Generally, large scale fabrication of devices for applications such as high power, high frequency and high temperature requires high quality material to be available with good crystallinity and low defect density. This imposes a severe problem in the case of SiC due to the equal stability of dierent crystal structures (polytypes) in this material,1 as will be outlined in Sec. 2. So, for a sucient crystallinity the development of polytype grains with the corresponding grain boundaries has to be suppressed. As bulk-grown material, for example from the modied Lely growth method,2 is not good enough for an electronic application, homoepitaxially grown lms are commonly used for the device development. This requires establishing growth methods suitable for obtaining homo-polytype, well-ordered crystalline

Corresponding

lms on top of bulk-grown substrate wafers. We show in the present paper that the particular reconstruction geometry of the (33) phase on the SiC(0001) surface3 is responsible that a given polytype can be copied by attaching new material at step edges when the substrate used is cut slightly tilted with respect to the basal plane (o-axis). An important technological issue for the production of devices is the quality of oxide layers grown on SiC samples. In particular, in MOSFET devices and for passivation layers a low defect density is essential both in the oxide layer itself and at the SiC/SiO2 interface. In fact, the density of states at the interface (Dis ) is one of the most relevant factors determining the performance of a device. A major contribution to Dis likely originates from structural defects at the interface, e.g. unsaturated dangling bonds. Even though the SiC and SiO2 lattice parameters match within about 5% allowing for an epitaxial SiC/SiO2 interface in contrast to Si with a 25% mist the electronic quality of the substrateoxide

author. Fax: +49-9131-85-28400. E-mail: ustarke@fkp.physik.uni-erlangen.de PACS numbers: 68.35.Bs, 61.14.Hg, 61.16Ch, 61.72Nn, 81.05.Hd, 81.65.Mg

1130 U. Starke et al.

interface for Si is by far superior to that for SiC.4 So, obviously, better control of the oxidation or oxide deposition process is necessary in order to utilize the full potential of SiC for electronic devices. This may be possible using a ( 3 3)R30 -reconstructed oxide adlayer structure which can be induced by ex situ hydrogen treatment of SiC(0001)- or SiC(000 1)oriented samples5 and promises to represent an excellent nucleation layer for oxide growth, as shown also in the paper. Noteworthy is that the stability of SiC in dierent polytypes is accompanied by a considerable spread of gap values in their electronic band structure.6,7 So, with their lattice parameters parallel to the hexagonal bilayers being practically equal, the development of strain-free heterostructures composed of dierent polytypes appears feasible. For such an application it is necessary to develop the means to switch between dierent polytypes during growth and obtain sharp, well-dened polytype interfaces. For the formation of such a heterojunction a at surface is obviously needed during the growth process. Yet, in such cases island nucleation has been observed up to now leading to a large number of grain boundaries.8 However, we can demonstrate that by careful control of the stoichiometry during the for mation of a dierent ( 3 3)R30 superstructure characterized by a Si adatom reconstruction9 a crystal structure can be induced in the surface region which is dierent from that of the substrate.
2. Crystal Structure and Surface Termination
A silicon carbide crystal is composed of Si and C alternatingly bonded in tetrahedral coordination [cf. Fig. 1(a)]. As shown in panel (b) of the gure, these SiC clusters are arranged in a hexagonal bilayer structure similar to that in the diamond
crystal structure, yet with Si and C separately occupying the sublayers of the bilayer. The nearly degenerate total energies of the two possible mutual orientations of adjacent bilayers1013 leads to a (meta)stability of many dierent stacking sequences in SiC crystals, and thus to the so-called polytypes.1 The cubic zinc blende crystal structure shown in Fig. 2(a) is characterized by an identical orientation of all bilayers in the crystal. As the atomic geometry is repeated every three bilayers along the c axis this crystal structure is commonly denoted as 3C polytype (3 layers, Cubic). By rotating adjacent bilayers by 60 a Wurtzite crystal structure could be constructed which, for some reason, is not stable in the case of SiC. Nevertheless, SiC polytypes exist which consist of slabs of two or three identically oriented bilayers which in turn are stacked in mutually rotated orientation. In Figs. 2(b) and 2(c) the crystal structure is displayed for a 4H polytype consisting of slabs of two and a 6H polytype consisting of slabs of three identically oriented bilayers, respectively. Those polytypes have hexagonal lattice symmetry (thus the letter H) with a periodicity along the c axis of four and six layers, respectively. For simplicity we refer to the two mutual orientations of adjacent bilayers as cubic (for identical) and hexagonal (for rotated) stacking. Alternatively, the stacking sequence can be described using a notation which was introduced by Ramsdell14 and is based on assigning letters to the individual bilayers indicating their relative lateral position, as also shown in Fig. 2. From the letter combination we can draw the actual stacking sequence. A sequence of three different letters indicates identically oriented bilayers, i.e. a cubic type of stacking. A letter being repeated within three layers indicates a mutual rotation, i.e. a hexagonal type of stacking.

Fig. 1. Atomic structure of SiC: (a) tetrahedrally bonded SiC4 cluster, (b) hexagonal bilayer with Si and C in alternating tetrahedrally coordinated sites.
SiC Surface Reconstruction: Relevancy of Atomic Structure for Growth Technology 1131

a) 3C-SiC

A B C A B C

b) 4H-SiC

A B C B A B

A B C A C B

6H-SiC
Fig. 2. Crystal structure of dierent SiC polytypes displayed parallel to the (1120) plane: (a) zinc blende structure (cubic 3CSiC), (b) hexagonal 4HSiC and (c) hexagonal 6HSiC.
Fig. 3. Possible bulk-terminated surface layer stacking sequences on 4HSiC(0001): congurations S2 and S1 dier by and are denoted according to the number of identically oriented bilayers directly at the surface. Except for a 60 rotation congurations S2* and S1* are identical to S1 and S2. The layer orientation and stacking sequence is indicated by the SiC bond train parallel to the (11 projection plane. 20)
On a SiC surface parallel to the hexagonal bilayers which actually is the basal plane orientation in hexagonal lattice symmetry, dierent surface terminating stacking sequences can be present and have to be considered in the crystallographic analysis. Each individual layer of the bulk unit cell can constitute the outermost surface layer. For 4HSiC this results in four dierent possibilities, as displayed in Fig. 3. We denote shortly the dierent stacking sequences according to the depth of the bilayer orientation change, i.e. S1, S2, S1* and S2*, whereby S1* and S2* are identically to S1 and S2 except for a 60 rotation of the whole semi-innite crystal.15
Consequently, for a 6HSiC we should expect three dierent types of stacking terminations, S1, S2 and S3 and their rotated equivalents S1*, S2* and S3*.
3. Surface Analysis Techniques
In this paper we describe the crystallographic analyses of the three reconstructed phases on hexagonal SiC surfaces mentioned in the introduction which were prepared either by ex situ hydrogen treatment or by Si deposition and annealing in vacuum. The surface analytical experiments were performed in a stainless steel ultrahigh vacuum (UHV) chamber

1132 U. Starke et al.

equipped with a sample introduction stage, a scanning tunneling microscope (STM), four-grid backview low-energy electron diraction (LEED) optics and a 150 spherical sector analyzer for Auger electron spectroscopy (AES). In UHV the samples could be heated by electron bombardment and cooled to liquid nitrogen temperature. From a solid source electron beam Si evaporator the surface could be exposed to Si ux during the heating procedure. STM and AES were used to provide direct experimental information about the atomic arrangement in the topmost surface layer and the surface stoichiometry, respectively, which was necessary to reduce the number of feasible models to be considered in the crystallographic analyses. For these detailed analyses LEED spot intensities [I(E) spectra] were measured using a computer-controlled, video-based data acquisition system.16 The atomic geometries of the dierent surface phases were determined using full dynamical LEED intensity calculations and in particular the tensor LEED perturbation method.16,17 The Pendry R factor Rp 18 guided an automated search algorithm19 which identied the best-t structure including the relative weights of domains exhibiting dierent surface layer stacking. A holographic interpretation of the LEED spot intensities in addition facilitated a further reduction of the model variety in the structural analysis in the case of the (33) reconstruction.20 Samples of dierent polytype and polarity from dierent sources were used in the course of our investigations. This included pieces of bulk-grown wafers in (000 orientation and epitaxial layers grown by 1) chemical vapor deposition (CVD)21 in the case of SiC(0001) samples. Only for the LEED investigation of the (33) phase did we use a 3CSiC(111) lm sample which was grown on Si(111). Due to the absence of twinning in this lm sample and thus the exclusive presence of one bilayer orientation on this surface, the experimental LEED pattern has a reduced symmetry which allowed to obtain a larger data set on the one hand and to simplify the structure analysis on the other hand.

4. Silicate Monolayers on

SiC(0001) and SiC(000) 1

In the case of SiC the removal of preparation-induced damage such as cutting and polishing scratches can-
not be performed by the common technique of in vacuo ion bombardment (sputtering) and annealing due to the considerably dierent vapor pressure values for Si and C. At the temperatures required for a recrystallization of the sputter-damaged layers, carbonization of the surface takes place. Consequently, SiC samples have to be initially prepared by an ex situ treatment which typically consists of a thermal oxidation and the removal of the so-called sacricial oxide by etching in hydrouoric acid (HF). SiC(0001) and SiC(000 samples pre1) pared in such a manner typically display a (11) LEED pattern corresponding to the periodicity of a SiC bilayer together with some diuse background [cf. Fig. 4(a)]. LEED structure analyses of several such samples of dierent polytypes and polarity indeed determined the surface geometry to consist of unreconstructed SiC bilayers, however, covered by a submonolayer amount of oxygen or hydrogen statistically adsorbed on top of the topmost atoms of the bilayer.15,2226 An improvement of this situation can be achieved by a hydrogen etching procedure similar to a typical preparation step used before epitaxial SiC growth experiments.27,28 When the sample is annealed in a quartz tube (e.g. in a CVD reactor) to 1500C under continuous H2 gas ow at atmo spheric pressure, a ( 3 3)R30 LEED pattern with bright and sharp spots and practically no background is observed immediately after transfer of the sample into the UHV chamber without any further treatment (no HF dip, no outgassing). The quality of the LEED pattern indicates a high degree of order by far exceeding the quality of the usual (11) surfaces obtained after ex situ preparation. Notewor thy is that this kind of well-ordered ( 3 3)R30 phase can be obtained on surfaces of both polarity as shown in Figs. 4(b) and 4(c) for 4HSiC(0001) and 6HSiC(000 respectively. On both surfaces 1), AES clearly reveals the presence of oxygen. From the OKLL signal (peak-to-peak height in comparison to the Si and C signals) the amount of oxygen can roughly be estimated to about one monolayer. The ne structure of the SiLV V signal with an additional component around 65 eV indicates a SiO2 type bonding of the oxygen. Both the ( 3 3)R30 LEED pattern with its I(E) spectra and the oxygenrelated features in AES are stable against annealing up to 1000C, indicating that the oxygen atoms are part of the reconstruction pattern on the surface. In
SiC Surface Reconstruction: Relevancy of Atomic Structure for Growth Technology 1133
Fig. 4. LEED patterns (normal incidence) obtained for dierent phases on hexagonal SiC surfaces: (a) (11) phase on 4HSiC(0001) after ex situ sacricial oxidation and a subsequent HF dip, Ep = 98 eV; (b) ex situ hydrogen-etched ( 3 3)R30 -silicate reconstruction on 4HSiC(0001), Ep = 146 eV; (c) ex situ hydrogen-etched ( 3 3)R30 silicate reconstruction on 6HSiC(000 Ep = 164 eV; (d) 4HSiC(0001)-(33) phase, Ep = 135 eV; (e) incomplete 1), ( 3 3)R30 phase with streaks between the integer order spots after 20 min annealing at 1000 C, Ep = 115 eV; (f) ordered ( 3 3)R30 phase, Ep = 115 eV.

full agreement with these experimental observations, the LEED structure analyses performed for both surfaces determined a Si2 O3 monolayer to reside above an otherwise bulk-truncated crystal with convincing Pendry R factors of 0.20 for the 4HSiC(0001) and 0.14 for the 6HSiC(000 surface. This adlayer 1) is formed by a honeycomb-like arranged sublayer of two Si atoms per ( 3 3)R30 unit cell connected by twofold-coordinated oxygen atoms in a sublayer 0.47 A above the Si atoms. The Si2 O3 layer, which we denotes as the silicate layer as it strongly resembles the layer structure in sheet silicates, is found in practically identical geometry5,29 on both surface polarities [displayed in Fig. 5(a)]. As depicted also in the gure, the only signicant dierence between SiC(0001) and SiC(000 is visible in the connection 1) of the silicate layer to the topmost substrate SiC bilayer: on SiC(000 the two are directly connected 1) by a SiC bond (panel b) while on SiC(0001) a linear SiOSi bridge mediates the contact (panel c). On both surface orientations the oxygen atoms saturate
all bonds of the silicate adlayer; only one of the three Si or C bonds in the topmost substrate bilayer is not saturated with only threefold coordination. This may explain the stability of the structures in UHV and even against exposure to air ambient. The origin of the silicate adlayer reconstruction found after hydrogen etching and introduction into the UHV chamber remains unresolved from our study. We can only speculate that by the hydrogen treatment a ( 3 3)R30 periodicity is somehow impressed on the surface, which then serves as an ordered seed for the rapid oxidation resulting in the silicate type structure. The ordered seed structure obviously is necessary for the reconstruction to develop, otherwise it should have been also observed in earlier investigations of ex situ prepared surfaces. With the initial order absent oxygen adsorption and reaction proceeds statistically on all available sites, i.e. in a (11) lattice gas disorder. The lack of such an ordered seed may even be one of the reasons for the poor electronic quality of thermally oxidized

1134 U. Starke et al.

Si C O
Fig. 5. (a) Top view of the oxide structure on SiC(000 The Si2 O3 silicate adlayer consists of a honeycomb structure 1). with SiOSi bonds. In the center of the hexagons carbon atom of the topmost substrate bilayer is visible [darkone shaded area indicates the (11)-, light-shaded the ( 3 3)R30 -unit cell]. (b) Side view projection of (a) along the (01 direction (a) tilted forward). (c) Side view of the oxide structure on the SiC(0001) in (01 projection. Linear 10) 10) SiOSi bonds connect the silicate layer and the SiC substrate.
Fig. 6. Schematic plot of the possible scenario of the transformation from a silicate monolayer structure with one Si sublayer to the silicate bulk type structure (-tridymite) with two Si sublayers. The atomic displacement of one Si atom per unit cell is indicated by the arrow.
SiC Surface Reconstruction: Relevancy of Atomic Structure for Growth Technology 1135
layers on SiC, as mentioned in the introduction. The high interface state density has been attributed to disordered species between the SiC substrate and the oxide layer.3033 However, the structure of the ordered adlayer being remarkably similar to that of bulk SiO2 certainly is intuitive, leading to the speculation that it might serve as seed to deposit thicker oxide lms. Indeed, the lateral unit vector of the ( 3 3)R30 periodic lattice matches that of bulk SiO2 within 95%. The only dierence between our silicate monolayer and the bulk structure of a high temperature SiO2 phase known as -tridymite is the position of the Si atoms. In the bulk structure a silicate layer consists of three sublayers with the Si atoms alternatingly positioned below and above the oxygen atoms. Hypothetically, the silicate adlayer found on SiC can be transformed to this structure simply by shifting one of the two Si atoms in the unit cell upwards in this upper Si sublayer position. This situation is schematically drawn in Fig. 6, with the arrow depicting the displacement from the monolayer silicate position (full circle) to the bulk silicate position (open circle).
5. Dangling Bond Saturation
in the (33) Phase on SiC(0001)
A surface prepared ex situ whether it displays (11) or ( 3 3)R30 periodicity can be transformed into dierent ordered phases in UHV by annealing
with or without an additional supply of Si from the evaporator [for a survey of these phases, see Ref. 15 for SiC(0001) and Ref. 34 for SiC(000 1)]. We concentrate on two silicon-rich phases on SiC(0001), namely the (33) and the ( 3 3)R30 phase. The (33) phase on SiC(0001), which was discovered by Kaplan,35 can be prepared by annealing the sample at temperatures around 800850C under simultaneous deposition of Si from the electron beam evaporator.36 As expected from the method of preparation, AES and electron energy loss spectroscopy (EELS) indicated the surface to be enriched in Si which has to be arranged in some kind of Si adlayer.35 The reconstruction geometry of the phase was originally proposed35 to be a (33) derivative of the dimer-adatom stacking fault model (DAS), which is well known from the (77)-Si(111). However, STM images clearly rule out this possibility as there is only one prominent (adatom-like) structure visible per unit cell,23,37 as depicted in Fig. 7(a). Yet, the presence of only a single adatom type structural element per unit cell made this (33) phase a perfect candidate for the rst application of a holographic interpretation of LEED data38 to an unknown surface structure. It is a principal requirement for this technique in order to be used with an ordered superstructure that a single atom per unit cell exists in a position above the other surface atoms so that it can serve as a beam splitter dividing the incoming electron wave into an object and a reference wave.3840 A holographic reconstruction of the

Fig. 7. Experimental information for the (33) phase: (a) topographic STM image (112 A 112 ) with the unit A cell indicated in the zoomed inset (27 27 ). (b) 3D image of the holographic reconstruction showing the local A A adcluster environment (dark spheres) of the topmost single adatom (light sphere) down to the fourth layer.

1136 U. Starke et al.

surrounding of the adatom using experimental LEED patterns [cf. Fig. 4(c)] is shown in Fig. 7(b) and reveals the position of the adatom to be in a T4 type site, i.e. in a threefold hollow site on top of an additional atom immediately underneath.41 The three atoms forming the hollow site are positioned in a plane 1.3 A below, the additional atom another 1.3 A below this plane. A correlation with the adlayer picture as drawn from AES and EELS would assign these four nextneighbor atoms to be part of the Si adlayer with a single adatom on top as seen in STM and the fth adatom that can be seen in the holographic reconstruction once again underneath the adatom (2.0 A further below) to belong to the topmost SiC substrate bilayer. With this partial model of the (33) unit cell at hand, a variety of models for the complete
structure could already be ruled out.42 The remaining unresolved part of the unit cell and the detailed positions of the atoms on the surface were then determined by both quantitative LEED structure analysis and density functional theory (DFT).3,43 The LEED analysis for this complicated structure yielded a Pendry R factor of 0.19. The optimized model contains and corroborates all features drawn from the experimental evidence, i.e. the T4 adatom position, and the presence of a Si adlayer that in turn is covering the topmost SiC substrate bilayer. The structural t shows that the Si adlayer covers the complete surface without corner holes which were demanded in a model suggested by Kulakov et al.37 Due to a rotational displacement within the adlayer [cf. Fig. 8(a)] the interatomic distances can assume values between 2.31 and 2.35 , which is close to the value of A A
Fig. 8. The SiC(111)-(3 3) structure: (a) lateral relaxations within the silicon adlayer (light and dark shaded atoms) and of the trimer (open circles, thin contour) supporting the adatom (open circle, thick contour) indicated with respect to the ideal positions in the topmost SiC bilayer; (b) side view along the (11 projection. The open circle indicates 20) the adatom, and dark shading indicates the atoms revealed by LEED holography. Lateral and vertical displacements within the Si adlayer and the SiC substrate bilayer are drawn on scale.
SiC Surface Reconstruction: Relevancy of Atomic Structure for Growth Technology 1137
2.35 for an ideal SiSi bond length. In addition, as A shown in Fig. 8(b), all atoms are situated in a single layer being threefold-coordinated to their Si neighbors with 120 bond angles and onefold-coordinated to the Si atoms of the substrate bilayer. Thus, these Si atoms are eectively sp2 -hybridized and their four bonds fully saturated. The only and single remaining dangling bond per unit cell is located at the Si adatom. This provides a very eective passivation of the surface which, on the one hand, explains the stability of the (33) superstructure. On the other hand, it can also explain the good homoepitaxial growth possible under Si-rich growth conditions. It had been observed that good quality epitaxial layers of the same polytype as already present in the substrate can be obtained in CVD using a Si-rich gas mixture.44,45 In MBE experiments under Si-rich conditions which also lead to homoepitaxial growth, a (33) superstructure was observed using reection high energy electron diraction (RHEED).8,46 A successful growth process in this manner requires the substrate to be cut slightly tilted with respect to the basal plane, i.e. the so-called o-axis orientation. On such a substrate with the (33) phase present the surface passivation leads to a high mobility of incoming particles such that they can diuse along the terraces until they are attached to a step. The new material continues the periodic structure of the bilayer that is terminated at the step and thus reproduces the stacking sequence of the substrate as it is exposed at the steps. So, a step ow growth mode is established that leads to a homo-polytype epitaxy. Corroborating this picture, it had been noted that in

the case of a too small o-axis angle with correspondingly large terraces the step ow growth mode fails to develop and homoepitaxy cannot be established.47
6. Stacking Rearrangement
in the ( 3 3)R30 Si Adatom Structure
Annealing a (33) periodic surface at around 1000C for about 30 min leads to the development of a new ( 3 3)R30 phase.23,35 Alternatively this structure can be obtained by heating the ex situ prepared sample (( 3 3)R30 silicate or (11) phase) at as 950 C, shown previously.48,49 However, the two ( 3 3)R30 phases are clearly distinguishable from each other both by their composition as drawn from AES and by the LEED spot intensities. In contrast to the ex situ structures, the new phase contains no oxygen and has nearly bulklike Si/C stoichiometry. It can, in fact, also be prepared starting from any surface phase by annealing at 10001100 C under simultaneous Si deposition50 to compensate for the Si depletion caused by the heating procedure. However, the dicult balance between Si depletion and deposition makes this third preparation method rather delicate and the phase often fails to develop in perfect order. LEED structure analyses carried out for all three preparation methods on a 4HSiC(0001) sample found similar reconstruction geometries with Pendry R factors of 0.110.13. The surface is characterized by a single Si adatom in T4 position on top of a SiC substrate bilayer, as shown in Fig. 9(a). The main
Fig. 9. T4 model for the ( 3 3)R30 phase on SiC(0001) displayed in a side view projection along the [11 20] direction. (a) Si adatom fourfold-coordinated to three Si and one C atom of the topmost substrate bilayer. Geometry parameters as given in Table 1 are indicated. (b) Dierent stacking terminations denoted S1, S2 or S3 according to the number of identically oriented bilayers at the surface. Note that the S3 termination is breaking the 4H bulk stacking sequence.
% ' & $ ( " ! @ B G E D D D C 0 )
1138 U. Starke et al. Table 1. Structural parameters as dened in Fig. 9(a) determined for the ( 3 3)R30 superstructure by LEED (present work, error limits about 0.05 ), XRD and by DFT. A From LEED (this work) 1.77 0.34 2.46 From DFT Ref. 52 Ref. 53 1.75 0.22 2.42 1.71 0.25 2.41 From XRD Ref. 54 1.61 2.31

Parameters d01 () A b2 () A L01 () A
Table 2. Weights of domains with dierent surface terminating stacking sequences and Pendry R factors derived for the optimized geometries of the three dierently prepared ( 3 3)R30 structures. Pendry R factor 0.13 0.13 0.11 Surface stacking S2 S1 S3 75% 50% 20% 15% 15% 15% 10% 35% 65%
Preparation method Annealing ex situ sample Direct prep. in Si ux Annealing (33) phase
reconstruction parameters, i.e. adatom layer spacing d01 , silicon bond length L01 and a surface buckling b2 below the adatom [cf. Fig. 9(a)], compare very well for the dierently prepared surfaces (variations are below our approximate error margin,51 i.e. 0.05 ). In these geometry parameters the A resulting structure agrees well with earlier theoretical calculations using DFT13,52,53 and very recent work using X-ray diraction (XRD)54 as listed in Table 1. In fact, with this conrmation of the DFT results a long-standing problem could be resolved that originated from the experimental observation of a semiconducting nature of the surface by ultraviolet photoelectron spectroscopy (UPS),55 inverse photoelectron spectroscopy (IPS)56 and recently by scanning tunneling spectroscopy (STS).57 Theoretically this surface gap can only be explained by including large electronic correlation eects using a MottHubbard type model.58 In addition, X-ray photoelectron spectroscopy (XPS)55 and a quantumchemical theoretical approach59 had predicted dier ent models for the ( 3 3)R30 structure. This called for a crystallographic clarication. However, while in neither DFT nor XRD investigations was the surface terminating stacking sequence considered as a variable parameter, our LEED analysis nds signicant dierences in this
respect for the three preparation methods which can also explain subtle deviations found in the I(E) spectra for the three recipes. In fact, for the preparation method of annealing a (33) phase the structure cannot be tted using the S1 and S2 type stacking as discussed in Fig. 3. A correspondingly bad R factor of 0.26 could only be reduced by allowing domains with S3 stacking which should not be expected on the 4HSiC sample [cf. Fig. 9(b)]. Then, however, the R factor drops to Rp = 0.11. In the optimized structure a fraction of 65% of the surface consists of domains with this unusual stacking sequence, i.e. three identically oriented bilayers at the topmost surface which is incompatible with the 4H bulk stacking but is the basic element of 3C and 6H SiC polytypes. The area of the surface covered with these domains strongly depends on the amount of silicon exposure during the preparation of the ( 3 3)R30 phase. When prepared directly by heating in a smaller Si ux, we nd only 35% of the surface covered by S3-terminated areas; when prepared from anex situ pretreated sample by heating alone, a negligible amount of the surface displays S3 stacking.60 The respective domain weights are listed in Table 2. This latter result may be a key issue for the growth of polytype heterostructures as such a S3 surface termination might serve as seed for the

SiC Surface Reconstruction: Relevancy of Atomic Structure for Growth Technology 1139

(4747)R19.1

1190 x 1190

595 x 595

186 x 354
Fig. 10. STM image of an approximately 600-A-wide mesa in the phase transformation region. Selected patches with particular local periodicity are stepwise enlarged. The periodicity is indicated by circles and unit cell borders on the right side (Utip = 1.42 V, I = 0.4 nA).
development of a dierent polytype such as 3C or 6HSiC. It is important to note that the mechanism of the stacking rearrangement does not proceed via a rotation of a bilayer already present. This would require a large number of S1 type domains to be present in the initial surface before the ( 3 3)R30 phase develops, which to the contrary are present only with 15%. It is rather that an additional bilayer is attached to S2 type domains as a result of a severe roughening of the surface dur ing the (33) to ( 3 3)R30 transformation. In this transformation stage the LEED pattern contains streaks and additional spot as depicted in Fig. 4(e) before the ( 3 3)R30 phase develops in full order (panel f). In the same stage of the transformation in STM large mesa type structures are found with dierent local periodicities on top, as shown in Fig. 10. In both the initial (33) and the nal ( 3 3)R30 situation the surface shows large at terraces indicating the rough surface being characteristic for the phase transformation. So, it appears
that it is not primarily the reconstruction geometry that causes the stacking rearrangement. Rather, the Si-rich conditions present initially when annealing the (33) phase seem to be the key ingredient. The disappearance of the mesas with the excess Si nally desorbing must be accompanied by a considerable material transport which enables the new bilayer to form. It continues the orientation of the layers already present and thus forms a cubic stacking inconsistent with the 4H bulk structure. That is obviously caused by the excess silicon in view of the S3 termination being found only when the surface is Si-enriched during the preparation. This is supported in addition by the fact that the area of S3 stacking is reduced again when the surface is further heated in reduced Si [the method that immediux ately results in the ( 3 3)R30 structure]. So, even if the cubic stacking of the new layer may be slightly favored by the ( 3 3)R30 reconstruction geometry due to subtle energetic dierences (which we cannot decide from the present results), it is

1140 U. Starke et al.

certainly initiated by the silicon enrichment and the mesa disappearance. We recall that on an ex situ pretreated sample with only S2 and S1 domains5 no appreciable stacking rearrangement is observed when the ( 3 3)R30 phase is formed by annealing alone. This indicates that the unusual surface layer stacking requires a kinetic eect that is more important than small energy dierences.

supported by the Deutsche Forschungsgemeinschaft through SFB 292.

References

1. R. Verma and P. Krishna, Polymorphism and Polytypism in Crystals (Wiley, New York, 1966). 2. D. L. Barret, R. G. Seidensticker, W. Gaida and R. H. Hopkins, J. Cryst. Growth 109, 17 (1991). 3. U. Starke, J. Schardt, J. Bernhardt, M. Franke, K. Reuter, H. Wedler, K. Heinz, J. Furthmller, P. u Kckell and F. Bechstedt, Phys. Rev. Lett. 80, 758 a (1998). 4. J. A. Cooper, Jr., Phys. Stat. Sol. (a) 162, 305 (1997). 5. J. Bernhardt, J. Schardt, U. Starke and K. Heinz, Appl. Phys. Lett. 74, 1084 (1999). 6. W. J. Choyke, D. R. Hamilton and L. Patrick, Phys. Rev. 133, 1163 (1964). 7. G. L. Harris (ed.), Properties of Silicon Carbide, in EMIS Datareviews Series, No. 13 (Inspec, London, 1995). 8. S. Tanaka, R. S. Kern and R. F. Davis, Appl. Phys. Lett. 65, 2851 (1994); 9. U. Starke, J. Schardt, J. Bernhardt, M. Franke and K. Heinz, Phys. Rev. Lett. 82, 2107 (1999). 10. C. Cheng, V. Heine and I. L. Jones, J. Phys. CM 2, 5097 (1990). 11. C. Cheng, V. Heine and R. J. Needs, Europhys. Lett. 12, 69 (1990). 12. P. Kckell, B. Wenzien and F. Bechstedt, Phys. Rev. a B50, 17037 (1994). 13. F. Bechstedt, P. Kckell, A. Zywietz, K. Karch, B. a Adolph, D. Tenelsen and J. Furthmller, Phys. Stat. u Sol. (b) 202, 35 (1997). 14. L. S. Ramsdell, Am. Mineralogist 32, 64 (1947). 15. U. Starke, Phys. Stat. Sol. (b) 202, 475 (1997). 16. K. Heinz, Rep. Prog. Phys. 58, 637 (1995). 17. P. J. Rous, Prog. Surf. Sci. 39, 3 (1992); K. Heinz, R. Dll and M. Kottcke, Surf. Rev. Lett. 3, 1651 (1996). o 18. J. B. Pendry, J. Phys. C13, 937 (1980). 19. M. Kottcke and K. Heinz, Surf. Sci. 376, 352 (1997). 20. K. Reuter, J. Bernhardt, H. Wedler, J. Schardt, U. Starke and K. Heinz, Phys. Rev. Lett. 79, 4818 (1997). 21. N. Nordell, A. Schner and S. G. Andersson, J. o Electrochem. Soc. 143(9), 2910 (1996). 22. J. Schardt, Ch. Bram, S. Mller, U. Starke, K. Heinz u and K. Mller, Surf. Sci. 337, 232 (1995). u 23. U. Starke, J. Schardt and M. Franke, Appl. Phys. A65, 578 (1997). 24. U. Starke, J. Bernhardt, M. Franke, J. Schardt and K. Heinz, Diam. Rel. Mat 6, 1349 (1997). 25. J. Schardt, J. Bernhardt, U. Starke and K. Heinz, Surf. Rev. Lett. 5, 181 (1998). 26. M. Hollering, J. Bernhardt, J. Schardt, A. Ziegler, R. Graupner, B. Mattern, A. P. J. Stamp, U. Starke, K. Heinz and L. Ley, Phys. Rev. B58, 4992 (1998).

7. Conclusion

Three dierent reconstruction phases on hexagonal SiC surfaces were structurally analyzed using LEED, STM and AES. The detailed geometry of each structure explains or promises a direct relevancy for the technological application of SiC in electronic devices. A well-ordered Si2 O3 can be generated by an ex situ hydrogen etching treatment of the SiC samples prior to their introduction into the UHV chamber. This silicate monolayer might represent a perfect seeding layer for the deposition of well-ordered thick SiO2 lms, thus overcoming the diculties imposed by the poor quality of the SiCoxide interface conventionally obtained by thermal oxidation. The (33) phase prepared in UVH by annealing under simultaneous deposition of Si is well saturated, enabling a step ow to develop in epitaxial growth experiments which leads to good homoepitaxial and homopolytype CVD and MBE layers. Finally, a change of polytypes might be induced during growth, possibly allowing the development of polytype heterojunctions and perodic heterostructures by using the evidence from the structure analysis of the in situ prepared Si-adatom ( 3 3)R30 reconstruction where by carefully controlling the preparation procedure a surface stacking sequence can be induced that breaks the periodic stacking of the substrate polytype. These ndings indicate a strong relevancy of detailed structural properties of surface phases for growth and other technological applications that can only be understood when the structure is analyzed in detail as carried out in the present work by LEED crystallography.

Acknowledgments

We thank A. Schner and N. Nordell for providing o SiC samples and K. Christiansen for helping with the hydrogen etching procedure. This work was
SiC Surface Reconstruction: Relevancy of Atomic Structure for Growth Technology 1141 27. M. S. Saidov, Kh. A. Shamuratov and M. A. Kadyrov, J. Cryst. Growth 87, 519 (1988). 28. F. Owman, C. Hallin, P. Martensson and E. Janzen, J. Cryst. Growth 167, 391 (1996). 29. U. Starke, J. Schardt, J. Bernhardt and K. Heinz, J. Vac. Sci. Technol. A17, 1688 (1999). 30. B. Hornetz, H.-J. Michel and J. Halbritter, J. Mater. Res. 9, 3088 (1994). 31. C. Onneby and C. G. Pantano, J. Vac. Sci. Technol. A15, 1097 (1997). 32. A. Glz, G. Lucovsky, K. Koh, D. Wolfe, H. Niimi o and H. Kurz, J. Vac. Sci. Technol. B15, 1097 (1997). 33. V. V. Afanasev, M. Bassler, G. Pensl, M. J. Schulz and E. Stein von Kamienski, J. Appl. Phys. 79, 3108 (1996). 34. J. Bernhardt, M. Nerding, U. Starke and K. Heinz, Mat. Sci. Eng. B61 62, 206 (1999). 35. R. Kaplan, Surf. Sci. 215, 111 (1989). 36. The (3 3) phase on the 3CSiC(111)/Si(111) sample developed by heating alone, probably due to silicon diusion through defects in the heteroepitaxially grown lm. The structure, however, obviously is the same on 3C, 4H and 6HSiC as judged from a comparison of experimental LEED I(E) spectra. 37. M. A. Kulakov, G. Henn and B. Bullemer, Surf. Sci. 346, 49 (1996). 38. D. K. Saldin and P. L. De Andres, Phys. Rev. Lett. 64, 1270 (1990). 39. M. A. Mendez, C. Glck and K. Heinz, J. Phys. CM u 4, 999 (1992). 40. K. Heinz, U. Starke and J. Bernhardt, Prog. Surf. Sci., to be published. 41. Threefold hollow adatom position on Top of a second layer atom, thus eectively 4-fold coordinated. 42. K. Reuter, J. Schardt, J. Bernhardt, H. Wedler, U. Starke and K. Heinz, Phys. Rev. B58, 10806 (1998). 43. The DFT calculations were performed in the group of F. Bechstedt, Institut fr Festkrpertheorie und u o Theoretische Optik, Universitt Jena, Max-Wiena Platz 1, D-07743 Jena, Germany. 44. H. S. Kong, J. T. Glass and R. F. Davis, J. Appl. Phys. 64, 2672 (1988). 45. T. Kimoto, A. Itoh and H. Matsunami, J. Appl. Phys. 81, 3494 (1997). 46. A. Fissel, B. Schrter and W. Richter, Appl. Phys. o Lett. 66, 3182 (1995). 47. T. Kimoto and H. Matsunami, J. Appl. Phys. 75, 850 (1994). 48. F. Owman and P. Martensson, Surf. Sci. Lett. 330, L639 (1995). 49. U. Starke, Ch. Bram, P.-R. Steiner, W. Hartner, L. Hammer, K. Heinz and K. Mller, Appl. Surf. Sci. u 89, 175 (1995). 50. L. Li, Y. Hasegawa, T. Sakurai and I. S. T. Tsong, J. Appl. Phys. 80, 2524 (1996). 51. Models yielding Rp Rp,opt + RR can be significantly excluded based on statistical grounds, RR being the variance of the Pendry R factor.18 For our best-t model with Rp,opt = 0.107 this variance is RR = 0.016. 52. J. E. Northrup and J. Neugebauer, Phys. Rev. B52, R17001 (1995). 53. J. Pollmann, P. Krger and M. Sabisch, Phys. Stat. u Sol. (b) 202, 421 (1997). 54. A. Coati, M. Sauvage-Simkin, Y. Garreau, R. Pinchaux, T. Argunova and K. A Phys. Rev. B59, d, 12224 (1999). 55. L. I. Johansson, F. Owman and P. Martensson, Surf. Sci. Lett. 360, L478 (1996); L483 (1996). 56. J.-M. Themlin, I. Forbeaux, V. Langlais, H. Belkhir and J.-M. Debever, Europhys. Lett. 39, 61 (1997). 57. V. Ramachandran and R. M. Feenstra, Phys. Rev. Lett. (1999). 58. J. E. Northrup and J. Neugebauer, Phys. Rev. B57, R4230 (1998). 59. P. Badziag, Surf. Sci. 337, 1 (1995). 60. Based on the R factor variance,51 the error bar for determining a domain percentage is approx. 15%. Thus, 10% as obtained for the S3 termination in the case of the ex situ pretreated sample is insignicant.