Gallium-mediated homoepitaxial growth of silicon ... - APS Link Manager

2 downloads 0 Views 701KB Size Report
Gallium-mediated homoepitaxial growth of silicon at low temperatures. B. Gallas, I. Berbezier, J. Chevrier,* and J. Derrien. Centre de Recherche sur les ...
PHYSICAL REVIEW B

VOLUME 54, NUMBER 7

15 AUGUST 1996-I

Gallium-mediated homoepitaxial growth of silicon at low temperatures B. Gallas, I. Berbezier, J. Chevrier,* and J. Derrien Centre de Recherche sur les Me´canismes de la Croissance Cristalline–CNRS–Campus de Luminy, Case 913, Laboratoire associe´ aux universite´s Aix-Marseille II et III, 13288 Marseille Cedex 9, France ~Received 14 March 1996! The present work reports on the molecular-beam homoepitaxial growth of silicon with and without gallium in a low-temperature regime ~200–500 °C!. The growth mechanisms were investigated by reflection highenergy electron diffraction ~RHEED! and by high-resolution electron microscopy ~HREM!. The progressive appearance of the surface roughness was observed in situ by the continuous change in the RHEED pattern and was quantified with the damping coefficient of the RHEED oscillations. This coefficient was also used to determine the critical temperature of transition from the two-dimensional ~2D! nucleation to step-flow growth regime. HREM cross-sectional images showed that the surface kinetic roughening observed at temperatures below 300 °C was associated with a high density of periodically repeated stacking faults ~ABCBCA-type fault!. HREM comparison of layers grown with and without gallium showed that the adsorption of gallium greatly reduced the defect density and the roughness of the epitaxial layer. The Ga surfactant-mediated epitaxy thus enables us to produce epitaxial layers of high-perfection structure, defectless and with homogeneous thickness. Up to 20-nm-thick silicon epitaxial layers were grown at 300 °C on gallium-activated Si~111! with a perfect planeity ~at the atomic scale! and crystallinity ~no crystalline defect was observed at the microscopic level!. Thanks to systematic measurements of the oscillation damping, we also show that Ga does not modify the 2D nucleation–step-flow temperature transition. After analysis of different existing kinetic models, we conclude that the adsorption of gallium does not change the diffusion kinetics of silicon adatoms on the silicon substrate. Therefore, thermodynamic considerations would be invoked to explain the role of gallium on the reduction of defects and surface roughness. @S0163-1829~96!10031-X#

INTRODUCTION

In most technological applications, a major problem limiting the use of ultrathin multilayers to applications involving the growth of various heterostructures on silicon is the significant interdiffusion and segregation phenomenon generally observed at interfaces. This problem is particularly enhanced in the case of Si/Ge-doped heterostructures in which Si and Ge easily interdiffuse and dopants segregate to the surface. Until now, one has had to find experimental conditions that provide a compromise between limited interdiffusion and segregation usually obtained at low temperatures and good crystalline quality of the layers usually obtained at high temperatures. Therefore, a fundamental understanding of the progressive degradation of the layer planeity at low temperatures is a major industrial concern in molecular-beam epitaxy ~MBE! research. Indeed, a way to suppress the interdiffusion without degrading the quality of the epitaxial layers has not be found so far, although many researchers have reported experimental work on Si MBE. Considering that some other elements grow on silicon at low temperatures ~down to ambient temperature! without loss of epitaxy ~iron for example!, it was envisaged to adsorb foreign elements in order to ameliorate the Si homoepitaxial growth at low temperatures. The saturation of the Si dangling bonds by a surface-active agent modifies the bonding arrangement at the interface and changes the surface properties and consequently the growth mechanisms. If this element is a surfactant that reduces the surface free energy, it segregates to the surface and this effect prevents its incorporation into the grown layer. Gallium has been chosen because it is one of 0163-1829/96/54~7!/4919~7!/$10.00

54

the dopants most used in silicon epitaxy, and also because it is known as an efficient surface segregant for Si homoepitaxy. In this general framework, the aim of this study is to analyze the mechanisms of homoepitaxial growth at low temperatures with and without adsorption of gallium on the silicon substrate. EXPERIMENT

Silicon growth was performed on ~111!-oriented silicon substrate ~misorientation less than 0.2°) in a MBE machine that consists of a load lock and of two ultrahigh-vacuum ~UHV! chambers pumped down to a basic pressure >5310211 mbar. The epitaxial films were fabricated by evaporation of Si and Ga from effusion cells ~using thermally heated graphite crucibles!. The growth experiments were performed at different temperatures in the range 200– 500 °C with an incident silicon flux of about 9 Å /min. The substrates were prepared by a reproducible two-step process based on an ex situ chemical cleaning followed by an in situ thermal annealing. A Si buffer layer, 20 nm thick, was deposited at 750 °C prior to any epitaxial experiment. Homoepitaxial growth was performed either on the clean Si buffer layer ~111! surface or on a Ga-activated Si buffer layer ~111! surface. In the latter case, prior to the silicon deposition, the surface dangling bonds were saturated by one Ga monolayer ~ML!. Samples were then subsequently annealed under UHV at about 650 °C to desorb the Ga excess. This resulted in a sharp ( A33 A3)R30° surface reconstruction, which revealed a Ga saturation coverage of about 31 ML.1,2 After the Si epitaxy obtained at 200–500 °C, Si films were immediately cooled down to room temperature and 4919

© 1996 The American Physical Society

4920

B. GALLAS, I. BERBEZIER, J. CHEVRIER, AND J. DERRIEN

54

capped with an amorphous Si layer of about 10 nm for ex situ measurements. Structural investigations were performed in situ by reflection high-energy electron diffraction ~RHEED! and ex situ by high-resolution electron microscopy ~HREM!. RHEED patterns were recorded by a high-resolution video camera and digitalized by a computer equipped with a video signal converter card. RHEED oscillations were measured by their specular beam intensity during the silicon growth and their damping was analyzed systematically. The surface roughness ( s ) was estimated from the perpendicular width of the diffraction streaks using the following relationship: DG' 52PC/ s ,

~1!

where DG' is the full width at half maximum ~FWHM! and C a calibration constant that is determined using the distance measured between the 131 silicon streaks on the RHEED screen.4 The detailed RHEED analysis technique has already been described elsewhere and its validity checked with several experiments.3,4 RHEED intensity oscillations versus time were fitted with good agreement with the following general equation: I5Aexp~ 2lt ! sin~ v t ! ,

~2!

where A described the oscillation amplitude and l and v the damping and the pulsation of the oscillations, respectively. In order to precisely compare the role of Ga atoms on the damping of the RHEED oscillations, growth experiments with and without Ga were performed on the same substrate @and so with the same ~111! Si terrace width#. After investigation with the in situ analytical techniques, the samples were removed from UHV and prepared by the conventional cross-section technique in order to be further analyzed by HREM in a Jeol 200CX microscope. Defect densities were estimated from plane view samples at low magnification ~not shown here! with a sensitivity limit of 107 cm22 . HREM-obtained images were digitalized with a scanner pencil and the surface profiles were then extracted from these images. Such surface profiles give very precise information on the surface roughness and so are representative of the growth mechanisms. In this study the total roughness and the root mean square ~RMS! of every surface profile were computed on various areas ~about 10! of the same sample and the average of all these values was extracted. HREM images were also simulated by a multislice calculation method developed by Stadelmann.5 A supercell of 500 atoms was defined in the @111# direction to allow the modelization of repeated defects. Different types of crystalline defects were simulated for a series of defocus and thickness. Calculated images were then compared to the experimental images. RESULTS A. Si growth at low temperatures without preadsorbed Ga atoms

RHEED analysis carried out directly after the deposition of the Si buffer layer resulted in a sharp two-dimensional ~2D! streaky pattern presenting the Si 737 surface reconstruction. Such a RHEED pattern indicated that the surface

FIG. 1. Variation of the oscillations intensity vs time during homoepitaxial growth ~a! on clean Si~111! at 300 °C and ~b! on a Ga-activated Si~111! surface at 300 °C.

was clean and atomically flat. During the subsequent growth of silicon at low temperatures ~below 300 °C! this streaky pattern became spotty, representative of a rougher surface. In this regime, the growth proceeded by two-dimensional nucleation on terraces leading to a periodic variation of the step density, which induces oscillations of the specular beam intensity ~layer-by-layer growth!. One period of the oscillations corresponds exactly to the time of growth of one complete monolayer; such a phenomenon is largely used to precisely measure the growth rate.6,7 RHEED oscillations are observed in a small range of temperatures, since they are limited at low temperatures by the transition from layer-by-layer to island ~or multilayer! growth and at higher temperatures by the transition from 2D nucleation to step-flow growth. The latter regime consists of the propagation of the step train. Beyond these two limits the amplitude of the oscillations abruptly decreases to a complete disappearance. A typical evolution of RHEED oscillations, recorded during the homoepitaxial growth of silicon at 300 °C, is presented in Fig. 1~a!. At this temperature we observe an important damping of the oscillations after the growth of a few monolayers, which demonstrates a fast change in the layergrowth mode. This phenomenon is representative of the lowtemperature transition from 2D to 3D nucleation. Moreover, the surface roughness ( s ) deduced from the length of the RHEED peaks during growth ~Fig. 2! reached 6.5 Å for a total layer thickness of 150 Å. This is consistent with the RMS ('7 Å! found by HREM analysis @Fig. 3~a!#. The occurrence of a significant amount of kinetic roughness at this temperature cannot be explained by the gradual buildup

GALLIUM-MEDIATED HOMOEPITAXIAL GROWTH OF . . .

54

FIG. 2. Roughness deduced from the FWHM of RHEED diffraction Bragg peaks during homoepitaxial growth at 300 °C ~s! and 400 °C ~d!.

of roughness predicted by theoretical models of nucleation and redistribution.8,9 These theoretical studies have analyzed the random deposition of atoms on surfaces, which triggers the development of a kinetic roughness ( s ) described by

s >h b ,

~3!

where h is the film thickness.

FIG. 3. ~a! HREM cross-sectional view of the deposited silicon layer. A digitalized profile was extracted from the free surface of the layer. The surface roughness was then extracted by image analysis and compared to Fig. 2. ~b! Multislice simulation of ~a!.

4921

Depending on the underlying hypothesis ~adsorption of atoms, characteristics of the surface diffusion!, various theoretical descriptions lead to different morphology and different values of the critical exponent b . However, it is noticeable that the values of b should lie below 0.5 without ingredients other than the random deposition and the surface redistribution in the roughness modelization. Different measurements performed during the growth of Fe on Si~111! at room temperature showed that Eq. ~1! was satisfactory and the adjusted exponent b was between 0.22 and 0.3.4,10 Furthermore, the exponent of b 50.25 is close to the predicted value for kinetic roughening driven by random deposition and relaxed by surface-diffusion mechanisms.8,9 However, a clear difference was observed during the homoepitaxial growth of silicon on Si~111!; indeed, a linear dependence was obtained for the s -versus-h plot. Since most of the theoretical predictions and some experimental results for different materials give a considerably slower roughening rate, another instability would be invoked to generate a higher roughening rate during Si homoepitaxy. To date, the detailed phenomenon responsible for this roughness has not been clearly identified to our knowledge, but we suggest that the nucleation of defects during homoepitaxy could be one of the relevant origins of this roughness. Indeed, the multislice simulation @Fig. 3~b!# of the HREM image @Fig. 3~a!# demonstrates that the kinetic roughening is associated with repeated intrinsic stacking faults ~ABC/BCA-type fault!. At this growth temperature, their high density ~which can be estimated from the HREM micrographs at about 1012 cm22 ) demonstrates the very poor crystalline quality of the material grown. Moreover, we can generally see in HREM images the periodic repetition of the fault @one fault every three ~111! planes#. This periodicity leads to a 331 superstructure in the reciprocal lattice which is clearly pointed out by the Fourier transform of the HREM defect area @Fig. 4~a!# and confirmed by the RHEED pattern @Fig. 4~b!#. The diffraction evidence for these defects did not appear at the first stages of the growth but only after deposition of few monolayers. Thereafter, the intensity of the peaks originating from these defects increased continuously with the thickness of the deposited layer, assessing a strong correlation between the surface roughness ( s ) and the presence of defects. Intrinsic stacking faults are known to be fairly low-energy defects in silicon with only second-nearestneighbor stacking violation. Several researchers have studied the kinetics of nucleation of such defects and found typical stacking fault energy of the order of 65 mJ m 22 .11,12 The repetition of the stacking faults and their increasing density with layer thickness induce a continuous buildup of lattice disorder which is probably at the origin of the surface roughness. Indeed, faulted and unfaulted areas are bounded by partial dislocations that strain and distort the lattice and that consequently generate the surface roughness. The concurrent increase of the defect density and of the kinetic roughness finally leads to the growth of the amorphous silicon phase. The latter transition towards the growth of an amorphous phase has also been recently reported by Eagleshamm working on different substrate orientations.13 B. Si growth at low temperatures with preadsorbed Ga atoms

In order to improve the quality of layers grown at low temperatures, similar experiments were performed on Ga-

4922

B. GALLAS, I. BERBEZIER, J. CHEVRIER, AND J. DERRIEN

54

FIG. 5. HREM cross-sectional view of the silicon layer deposited on a Ga-activated Si~111! surface.

FIG. 4. ~a! Fourier transform of the HREM defect area. The 331 superstructure is clearly visible. ~b! RHEED pattern recorded during growth at 300 °C. Extra spots are representative of the appearance of the periodically repeated stacking faults.

activated surfaces. Previously, several authors have reported that silicon films with excellent crystallinity can grow on such surfaces, even at low temperatures.14,15 Our results show that Si~111! homoepitaxy is strongly affected by the adsorption of Ga. RHEED analysis @Fig. 1~b!# performed during the growth at 300 °C of a 150-Å-thick layer shows that the oscillation damping is much lower and, consequently, that the surface roughness is dramatically reduced. These results are confirmed by the HREM analysis. Indeed, the image in Fig. 5 displays a Si surface atomically flat without detectable roughness. Concurrently, the defect nucleation observed in pure Si/Si growth is now suppressed and the density of extended defects is too low to be observed by HREM ~i.e., lower then 107 cm22 ). This density is in the same range as that observed at high temperatures, which is, in fact, limited to impurities or substrate defects. Since defect nucleation during Si low-temperature growth is generated by a limited mobility of Si adatoms on the surface, the reduction of the defect nucleation, on the contrary, can be attributed to an increasing mobility of Si adatoms at a first approximation. Thus, it could be suggested that Ga favorably acts on the surface diffusion kinetics. While surfactants are generally introduced during epitaxy, their exact influence on the growth mechanisms is not yet understood. One necessary requirement for a surfactant is to reduce the surface free energy. However, this modification cannot explain the reduction of stacking faults and of surface roughness at low temperatures as observed in our experiment. Besides the fact that surfactants lower the surface free energy, they may also influence

positively the growth kinetics.16,17 In particular, it has been reported that elements of columns III and IV ~Ga, In, and Sn! as surfactants increase the Si diffusion length, and elements of columns V and VI ~As, Sb, and Te! decrease the diffusion length. More generally, surfactants occupying interstitial surface sites ~nonreacting surfactants! are shown to increase the surface diffusion length, whereas surfactants in substitutional sites ~reacting surfactants! are shown to decrease it.18 In the particular case of Ga-activated surfaces, it has been shown15,19 that the surface diffusion length of Si adatoms on that surface is five times larger than that on pure Si. On the contrary, others20 reported similar surface diffusion with and without Ga in pure Si/Si homoepitaxy. In order to clarify the cause of the crystalline quality improvement with preadsorption to Ga, the transition temperatures from the 2D nucleation to the step-flow regime with and without Ga were measured comparatively. In fact, the different growth-mode transitions are well evidenced by our systematic measurements of the time-corrected RHEED oscillation damping (l/ v ) at various growth temperatures. In Fig. 6 we can see a range of growth temperatures for which the damping coefficient (l/ v ) remains constant at its lower value. It corresponds to the layer-by-layer growth mode. At higher temperatures, the transition to the step-flow regime appears very abruptly, since in a range of 50 °C ~between 450 and 500 °C! the oscillations go from a minimal damping ~low l/ v ) to a total vanishing ~high l/ v ). Practically, this transition corresponds to the temperature for which the Si diffusion length is sufficiently high to allow Si adatoms to cross through the width of terraces towards the step sides ~step-flow growth mode!. Thus it gives direct information on the diffusion length magnitude. This technique has already been used to measure the diffusion length of Si adatoms during Sbmediated Si homoepitaxial growth at different coverage rates.21 In our study, the abruptness of the transition was used to precisely compare the diffusion lengths of silicon adatoms on substrates with and without preadsorbed Ga. The results in Fig. 6 show that the adsorption of Ga does not modify the temperature of this transition, since the vanishing of the oscillations occurs identically at the same temperature ~about 470 °C! with and without Ga. DISCUSSION

Although many works describe the effect of a surfactant, the exact mechanisms of surfactant-mediated growth have

54

GALLIUM-MEDIATED HOMOEPITAXIAL GROWTH OF . . .

4923

out surfactants. The relation between the diffusion energy E d and the island density is extracted from the nucleation theory.22 In the MBE regime of complete condensation, it can be expressed as N5

FIG. 6. Evolution of the RHEED oscillation damping (l/ v ) with temperature without Ga (s) and with Ga adsorbed on the surface (d). The arrows separate the temperature range where a low damping is observed ~2D nucleation regime! from those of high damping ~islanding regime at low temperatures and step-flow regime at high temperatures, respectively!.

still not been clearly assessed. Three main origins are naturally invoked to explain the surfactant action. ~a! Reduction of the surface energy. It is well known that a submonolayer of metal atoms deposited on Si produces specific surface phases ~even in the absence of a bulk alloy phase! located at the outermost surface region which are more stable than the reconstructed 737 Si~111! or 231 Si~100! surface. Different elements such as Sn, Ga, In, Sb, As, and Te are particularly suitable surfactants because they strongly reduce the surface energy of both Si and Ge. Since surfactant-terminated surfaces are energetically favorable, the surfactant will float at the surface during growth. Silicon adatoms are then easily incorporated into the substrate by site exchange with the surfactant atoms. To date, this thermodynamical behavior of a surfactant is quite clear. However, in technological applications, the major condition expected from a surfactant in a mediated growth on Si is to improve the layer crystallinity in homoepitaxial growth and to suppress uncontrolled islanding in the case of heteroepitaxial growth. Such requirements cannot be fully achieved only because of the surface-energy reduction. On the contrary, since the amount of energy required to create a surface excess is quite low, the system could easily produce roughness or islanding. ~b! Reduction of the surface diffusion. One of the simplest ways to approach the problem is to invoke the surfacediffusion variation. A surfactant should decrease the surfacediffusion length to avoid 3D growth at high temperatures and it should increase the diffusion length to improve crystallinity at low temperatures. A geometrical model was proposed18 in which the activation energy of diffusion is directly related to the bond strength between the surfactant and the substrate. It was shown that the latter is reduced by the formation of a complex when the surfactant occupies interstitial sites, while it is increased when the surfactant occupies substitutional sites. Attempted quantification of the diffusion length was performed by scanning-tunneling-microscopy ~STM! measurements of the densities of islands formed with and with-

SD R n

i/ ~ i12 !

exp

S

D

E i 1iE d , kT

where E i is the formation energy of a stable nucleus of i atoms, R the rate of nuclei formation, and n the atomic vibration frequency. Thus, in general the island density depends on E d but also on the nucleation energy E i . The latter is a combination of different parameters that describe the nucleation process ~Si-Si binding energy E b , critical size i of the stable nucleus, etc.!. In fact, E i also influences the island density, since an increasing formation energy of stable nuclei corresponds to a nucleation process energetically more favorable and thus to a higher nucleus density. Consequently, measurement of the island density cannot be in any case directly correlated to the diffusion energy. This fact probably explains the opposite behavior in diffusion mechanisms observed, for instance, for Sn and Ga in the literature.23 A similar problem is encountered when we establish a correlation between the RHEED measurement of the 2D– step-flow transition and the diffusion length. Indeed, the mean diffusion length is given by the Einstein relation l s 5Aexp

S

D

E a 2E d , 2kT

where E a is the energy of adsorption. Consequently, the diffusion length is not directly related to E d but rather to (E a 2E d ), making exact quantification of the diffusion length impossible.24 However, if we assume that this law is true and that E a can be considered as constant, the relative comparison of the diffusion lengths with and without surfactants ~carried out in this study! would give appropriate results. Now, if we compare the experimental results to the kinetic models on enhanced or inhibited diffusion of Si adatoms, some discrepancies are revealed, including ~1! the effect of Ga on the inhibition of crystalline defects despite its absence of action on the diffusion length ~Ref. 20 and our results!; ~2! the very different effect of As and Sb on the growth mechanisms, although both of them are supposed to decrease the surface diffusion;14 ~3! the important influence of the surfactant coverage level and of the surface reconstruction.25 In order to explain these discrepancies, some authors introduce new kinetic factors.26 In particular, two possible driving forces not directly related to the surface diffusion energy are generally described. (1) Reactivity modification at the step edge. Some kinetic models show that the differences between the surfactants that lead to the step-flow regime and those that lead to a highdensity 2D island growth depend on the way the surfactants bond to the surface steps and modify their reactivity. The surfactants may strongly bond at the step edge and induce a decreasing reactivity, since exchange between atoms and surfactants is no more preferred at the step edge compared to exchange between atoms directly on the terraces. This leads to a high-density 2D island regime. Or the surfactants may

4924

B. GALLAS, I. BERBEZIER, J. CHEVRIER, AND J. DERRIEN

weakly bond at the step edge. The exchange occurs preferentially at the step edge and leads to a ‘‘step-flow-like’’ regime and enhanced island coalescence. Indeed, it was demonstrated27 that it is not mandatory to have a reduced diffusion length for observing a 2D growth regime or to suppress a 3D islanding phenomenon. (2) Lowering of the Schwo¨bel barrier. Moreover, the asymmetric Schwo¨bel barrier ~smaller at ascending, higher at descending steps! encountered by adatoms at the step edge may also be strongly modified by the surfactant presence. It is suggested28 that a surfactant may reduce this asymmetry during adatom incorporation at kinks in the following way: surfactant atoms predominantly join the steps that represent energetically favorable sites ~first-principles calculations have confirmed this point29!. Then the deposited atoms must displace them in order to incorporate the crystal lattice. This gives rise to additional kinetic barriers that the adatoms have to overcome when they ascend a step. On the contrary, adatoms approaching a descending step will join the kinks by displacing the surfactant atoms with a reduced Schwo¨bel barrier. Recent experimental results confirmed the surfactantinduced lowering of the Schwo¨bel barrier during the homoepitaxial growth of Cu ~with In!,30 and Ag ~with Sb!.31,32 This effect favors an interlayer transport leading to an enhanced 2D island nucleation in homoepitaxial growth. ~c! Local dimer exchange. A geometrical model performed in the case of the growth of Ge on an As-terminated Si~001! surface proposed a two-dimer correlated exchange process that allows the elimination of four dangling bonds.33 Adsorption of additional Ge dimers in the same layer results in elimination of four other dangling bonds. This exchange mechanism provides a favorable pathway for homogeneous nucleation of 2D islands. Moreover, total-energy calculations based on first principles show that the exchange of the Ge dimer with the underlying As atoms is energetically favorable.34 The energy gained in this exchange is 1.1 eV per dimer. Therefore, every deposited adatom is rapidly incorporated and a reduced surface diffusion length is observed. So far, none of these kinetic mechanisms could explain our experimental results measured on the Ga-mediated Si homoepitaxial growth. Indeed, if we took into account these additional driving forces, we should observe a large variation in the 2D island density, which was not observed with the STM technique.20 Our RHEED and HREM experiments only prove that preadsorbed Ga atoms strongly reduce the defects appearing during Si homoepitaxy at low temperature without modifying ~a! the surface diffusion length and ~b! the 2D island density. Therefore, we suggest that besides the surfactant-induced surface energy change, other thermodynamical driving forces would be invoked in a more subtle manner to explain the defect-density reduction with Ga. Briefly, the nucleation of a faulted nucleus involves a process with subsequent steps:35 ~a! formation of a 2D nucleus

*Present address: ESRF–BP220 38042 Grenoble Cedex, France. M. Otsuka and T. Ichikawa, Jpn. J. Appl. Phys. 24, 1103 ~1985!. 2 J. Patel, J. Zegenhagen, P. Freeland, M. Hybertson, J. Golovchenko, and D. Chen, J. Vac. Sci. Technol. B 7, 296 ~1989!. 3 J. Chevrier, A. Cruz, N. Pinto, I. Berbezier, and J. Derrien, J. 1

54

in its normal position; ~b! breaking off of the nucleus from its original crystal plane site; ~c! homogeneous deformation of the nucleus; ~d! deposition on the same crystal plane of the faulted nucleus. The density of the faulted nuclei is then given by N5Kexp@ 2A * /kT # , where A * is the activation energy of the faulted nucleus. It depends on the square of the line tension ( x ) 2 and is inversely proportional to the difference @ kT2(E h 1E a ) # , where E h is the homogeneous deformation energy ~bulk and line! and E a is the adhesion energy of the faulted nucleus. At this stage we need other experiments to have access to these energetic parameters. As regards the kinetic surface roughness that develops concurrently with defects during the lowtemperature Si growth without preadsorbed Ga, one could also invoke the surface-energy modification, which is governed by at least four parameters: the free energy of the facet, the step formation energy, the kink formation energy, and the energetic interaction between steps. All of them may be greatly changed by surfactants. For example, it has been experimentally shown that surfactants differently modify the equilibrium shape of stable nuclei. In particular, Eaglesham36 showed that the 3D islands of Ge deposited on Si~001! exhibited larger $ 100% facets in the presence of Sb and larger $ 113% facets in the presence of In. These results prove the surfactant effect on surfaceenergy anisotropy. Moreover, it has been shown37 that the adsorption of In on Si~100! modifies step bunching, faceting, and step-edge roughening, depending on the coverage level and on the surface reconstruction. Again, further experiments have to be performed to clarify the role of Ga in these surface modifications. CONCLUSION

Si homoepitaxy at low temperature ~200–500 °C! leads to layers with a large density of defects and a high surface roughness as observed by our RHEED and HREM techniques. These drawbacks can be avoided by introducing a preadsorbed Ga submonolayer amount prior to the growth. The role of this surfactant is discussed within the framework of currently admitted kinetic and thermodynamic models. Our original and experimental results clearly show that kinetic models cannot explain the observation. Thermodynamic reasons should be invoked, including several hidden energetic parameters so far ignored. ACKNOWLEDGMENT

The authors thank Professor R. Kern for many helpful discussions.

Phys. ~France! I 4, 1309 ~1994!. J. Chevrier, V. Le Thanh, R. Buys, and J. Derrien, Europhys. Lett. 16, 737 ~1991!. 5 P. Stadelmann, Ultramicroscopy 21, 131 ~1987!. 6 B. A. Joyce, H. J. Neave, P. J. Dobson, and P. K. Larsen, Phys. Rev. B 29, 814 ~1984!. 4

54 7

GALLIUM-MEDIATED HOMOEPITAXIAL GROWTH OF . . .

T. Sakamoto, H. Funabashi, K. Ohta, T. Nakagawa, N. G. Kawai, T. Kojima, and Y. Bando, Superlatt. Microstruct. 1, 347 ~1985!. 8 D. Gandolfo, J. Ruiz, and V. A. Zagrebnov ~unpublished!. 9 I. Elkinani and J. Villain, J. Phys. ~France! I 4, 949 ~1994!. 10 Y. L. He, H. N. Yang, T. M. Lu, and G. C. Wang, Phys. Rev. Lett. 69, 3770 ~1992!. 11 R. Hull and J. C. Bean, Crit. Rev. Solid State Mater. Sci. 17, 507 ~1992!. 12 A. Bourret and J. Desseaux, Philos. Mag. A 39, 405 ~1979!. 13 D. J. Eaglesham, J. Appl. Phys. 77, 3597 ~1995!. 14 M. Horn-von Hoegen, J. Falta, M. Copel, and R. M. Tromp, Appl. Phys. Lett. 66, 487 ~1995!. 15 H. Nakahara and M. Ichikawa, Appl. Phys. Lett. 61, 1531 ~1992!. 16 B. Voigtla¨nder and A. Zinner, Surf. Sci. Lett. 292, 775 ~1993!. 17 H. J. Osten, J. Klatt, G. Lippert, E. Bugiel, and S. Higuchi, J. Appl. Phys. 74, 2507 ~1993!. 18 J. Massies and N. Grandjean, Phys. Rev. B 48, 8502 ~1993!. 19 U. Ko¨hler, J. E. Demuth, and R. J. Hamers, J. Vac. Sci. Technol. A 7, 2860 ~1989!. 20 B. Voigtla¨nder, A. Zinner, T. Weber, and H. P. Bonzel, Phys. Rev. B 51, 7583 ~1995!. 21 K. Sakamoto, K. Miki, T. Sakamoto, H. Matsuhata, and K. Kyoya, J. Cryst. Growth 127, 392 ~1993!. 22 A. Venables, G. D. T. Spiller, and M. Hanbu¨cken, Rep. Prog. Phys. 47, 399 ~1984!.

23

4925

M. Zinke-Allmang and L. C. Feldman, Surf. Sci. Lett. 191, 749 ~1987!. 24 S. Iwanari and K. Takayanagi, J. Cryst. Growth 119, 229 ~1992!; S. Iwanari, Y. Kimura, and K. Takayanagi, ibid. 119, 241 ~1992!. 25 J. Falta, M. Copel, F. K. Le Goues, and R. M. Tromp, Appl. Phys. Lett. 62, 2962 ~1993!. 26 T. Ide, Phys. Rev. B 51, 1722 ~1995!. 27 D. Kandel and E. Kaxiras, Phys. Rev. Lett. 75, 2742 ~1995!. 28 I. Markov, Phys. Rev. B 50, 11 271 ~1994!. 29 S. C. Wang and G. Ehrlich, Phys. Rev. Lett. 71, 4174 ~1993!. 30 H. A. van der Vegt, M. Breeman, S. Ferrer, V. H. Etgens, X. Torreles, and P. Fajardo, Phys. Rev. B 51, 14 806 ~1995!. 31 J. Vrijmoeth, H. A. van der Vegt, J. A. Meyer, E. Vlieg, and R. J. Behm, Phys. Rev. Lett. 72, 3843 ~1994!. 32 S. Oppo, V. Fiorentini, and M. Scheffler, Phys. Rev. Lett. 71, 2437 ~1993!. 33 R. M. Tromp and M. C. Reuter, Phys. Rev. Lett. 68, 954 ~1992!. 34 B. D. Yu and A. Oshiyama, Phys. Rev. Lett. 72, 3190 ~1994!. 35 R. Kern, Bull. Soc. Franc¸. Mine´r. Crist. 134, 292 ~1961!. 36 D. J. Eaglesham, F. C. Unterwald, and D. C. Jacobson, Phys. Rev. Lett. 70, 966 ~1993!. 37 L. Li, Y. Wei, and I. S. T. Tsong, Surf. Sci. 304, 1 ~1994!.