- Email: [email protected]
Experimental test of the single adatom exchange model in surfactant-mediated growth of Ge on Si(100)
Beam Interactions with Materials&Atoms
ELSEVIER
Nuclear
Instruments
and Methods
in Physics Research B 136-138 (1998) 804-809
Experimental test of the single adatom exchange model in surfactant-mediated growth of Ge on Si( 100) A.A. Bailes III ‘,*, M.A. Boshart ” Depurirwttt
b, L.E. Seiberling
a
of PII)‘.si(.s. Cbtiwrsitj~ qf Fhridu. P. 0. Bo\- 11X440, Grrit~esrill~~. FL 3261 I-X440, ” 376 Hcwttitqcy .4 wtntu. Cookc~rill~~. TN 3KSOI. (iSA
VSA
Abstract
We have tested the single adatom exchange model for surfactant-mediated growth. Using two samples with different coverages of Ge on Sb-terminated Si( IOO), we generated energy distributions of scattered MeV ions from transmission ion channeling experiments. We studied the system both after room temperature deposition of Ge and after annealing at 350°C. We then compared simulated energy distributions for the single adatom exchange model to the experimental energy distributions. No combination of temperature and coverage produced a good fit between data and simulations of this model. Before annealing. however, a model having Ge in dimer-like sites on top of undisturbed Sb dimers describes the data well for both Ge coverages. 0 1998 Elsevier Science B.V. 68.35.B~; 79.20.Rf; 79.60.Bm; 79.60.D~ K~JYNI~~.s:High energy ion scattering; Ion-solid interactions; PACS:
crystal epitaxy;
Scattering;
Channeling;
Semiconducting
surfaces:
Single
Surface structure
The creation of abrupt, flat interfaces in heteroepitaxial growth, as is required for such structures as Si/Ge superlattices, is made difficult by islanding and interdiffusion. Although the layer-by-layer (Frank-Van der Merwe) growth mode can be induced by deposition at low substrate temperature, the resulting crystallinity often is reduced. To overcome these problems, Cope1 et al. [l] introduced the idea of using a surfactant, a species that strongly segregates to the surface during growth and can kinetically inhibit islanding. Surfactants
‘Corresponding author. Tel.: 352 392-9219; fax: 352 3928586; e-mail: abailesk$phys.ufl.edu. 0168-583X/98/$19.00 0 1998 Elsevier Science B.V. All rights reserved. PIISO168-583X(97)00762-3
such as As or Sb passivate Si and Ge( 100) surfaces and lower the surface free energy when they segregate to the top layer. When deposited on the surface before the adsorbate, surfactant atoms undergo a site exchange process with the adatoms, thus rising to the top layer and reducing the diffusion length of the adatoms on the surface. This reduced diffusion length is thought to be the reason for the inhibition of islanding [2]. While surfactant-mediated growth has generated much interest, only a few studies have looked at the details of the site exchange mechanism. Based on low energy electron microscopy data for Ge grown on As-terminated Si( loo), Tromp and Reuter [2] proposed a two-dimer correlated exchange
A. A. Buiks
III et al. I Nwl.
Instr. unrl Meth. in Pll,s.
mechanism which starts with Ge adsorbing on top of, and then breaking, the surfactant dimers. Yu and Oshiyama [3] proposed a single-dimer exchange process based on total energy calculations showing that the initial adsorption site is more likely to be between surfactant dimer rows than on top of broken surfactant dimers. In a recent paper [4], we compared simulations representing these two exchange models to experimental scattered ion energy distributions. We found the Ge on Sb-terminated Si(100) system in an initial state having Ge in dimer-like sites on top of undisturbed Sb dimers. similar to an early stage of the Yu and Oshiyama model. As in the single-dimer exchange model of Yu and Oshiyama, the single-adatom exchange model proposed by Ko et al. [5] resulted from total energy calculations and starts with the adatoms adsorbing between the surfactant dimer rows. From these adsorption sites the adatoms exchange immediately, one adatom at a time, with the underlying surfactant. In all of these models, the system goes to a final state with surfactant dimers on top of near-substitutional adatoms after site exchange. In this work, we compare our experimental transmission ion channeling data to the single-adatom exchange model [5]. Ko et al. state that under the proper conditions (low growth temperature and low adatom coverage). an intermediate state with a surfactant atom residing directly above a surfactant-adsorbate heterodimer should be observable. However, we have found that this configuration of surfactant and adsorbate atoms does not fit data we have collected after sample preparation under the specified conditions, nor does it fit our data taken after annealing the sample at 350°C or using a higher Ge coverage. Comparing our data to other stages of the single-adatom exchange model was unnecessary for two reasons. First, we have tested the state that Ko et al. specify as the one that we should see for our sample preparation conditions. Second, the other states we could test are similar to parts of the Yu and Oshiyama model that we have tested previously [4]. Along with the comparison of our data to the Ko model, we show comparisons to the final state and to the state which best fits the data, Ge in dimer-like sites on top of undisturbed Sb dimers.
Rrs. B 136 138
( 1998) 804-X09
805
Samples were prepared with an ex situ dopantselective chemical etch to create a “thin window”, a piece of Si approximately 1 cm x 1 cm by 500 urn thick with a circular region of diameter -6 mm that has been thinned to 0.5 urn. We then cleaned the samples chemically and deposited a protective oxide using the Shiraki technique [6] before placing the samples in the ultra-high vacuum (UHV) chamber (base pressure -5 x 10. ‘I Torr). After thoroughly degassing the sample, we desorbed the oxide with a heat lamp. At this point, low energy electron diffraction (LEED) yielded a 2 x 1 pattern with low background and sharp spots, indicating a properly prepared Si(100) surface. We deposited Sb from a boron nitride effusion cell with the sample heated to -500°C. Exposure of the sample to several monolayers (ML) of Sb ensured saturation at -0.85 ML. The LEED pattern at this point changed to 1 x 1 with faint half-order spots. After allowing the sample to cool to near room temperature, we deposited Ge from an effusion cell at the rate of 0.15 ML/min. No noticeable change occurred in the LEED pattern for the sample with a Ge coverage of 0.15 ML, while the sample with 0.68 ML of Ge showed a LEED pattern with higher background and half-order spots no longer visible. Rutherford Backscattering (RBS) yielded coverages of 0.80 and 0.15 ML for the Sb and Ge, respectively, for the low Ge coverage sample and 0.84 and 0.68 ML for the high Ge coverage sample. More detailed descriptions of our sample preparation and experimental setup can be found elsewhere [7,8]. Upon completing the sample preparation, we transferred the sample under UHV to the scattering chamber (base pressure -5 x lo-” Torr) and placed the sample on a two-axis goniometer. A 3.5 MV van de Graaff accelerator produced a 2.5 MeV He+ beam, collimated to an angular divergence of 0.03” at the scattering chamber. With the sample in transmission geometry (adsorbates on the beam-exit side of the sample), we visually aligned the (100) axis of the sample with the beam by viewing the transmitted beam on the quartz at the end of the beamline. Comparison of Si yields with simulations has shown that the accuracy of alignment is 0.02”-0.05” for this technique. Taking as a second reference point a beam incidence -8”
806
A.A. Builes III et (11.I Nucl. Instr. und Meth. in P~z_IY.Rex B 136-13X (19981 804-809
from the (100) axis along the { 100) plane, we established the relation between lab and crystal coordinates. Thus we could readily choose a beam incidence at any tilt (I from the (100) axis and azimuth q relative to the {100) plane. In these experiments we obtained energy distributions in three directions: (0,~) = (O’.O’), (6”,0”), and (6O.30”). The first of these directions is the (100) axis, the second is the ( 100) plane, and the third is a “random” direction. in which ion energy losses approach those in amorphous media [9]. The direction chosen as the random is the result of simulating mean ion energy loss over a large range of H and cp and finding a direction where the difference between mean energy loss and random energy loss is minimal [S]. A passivated. ion-implanted solid state detector collected and analyzed the energy of the scattered ions at a scattering angle of 79”. The integrated beam current of each spectrum was limited to 24 uC (1.5 x 10” ions/cm’), which is within the range of beam dose that yields minimal irradiation damage to the Si crystal structure. Likewise we collected the axial and planar incidence data before the random incidence data because of the greater damage caused by the random incidence beam. One more step taken to minimize crystal damage was collecting before anneal and after anneal data on different spots on the thin window. The three spectra ((loo), { 100 ‘,, and random) corresponding to a given temperature and Ge coverage were taken on a single spot, however, in order to avoid problems with variations in adsorbate coverage and Si thickness. Data taken in the experiments consist of the three spectra at different beam incidences, after room temperature Ge deposition (henceforth called “before anneal”) and again after annealing the sample at 350°C for 10 min. This was done for both the low Ge coverage sample and the high Ge coverage sample for a total of 12 spectra. A simulated distribution of ion positions and energies before scattering, generated by a Monte Carlo simulation of the channeling process [lo], produced simulated energy distributions of scattered ions when overlapped with trial adatom sites [8,11]. By putting in up to five adatom sites at a time and comparing the simulated energy distributions with experimental distributions, we can test
theoretical models of the exchange mechanism in surfactant-mediated growth. Both the experimental and simulated energy distributions were normalized to the random yield; viz., for each distribution, the counts in each energy bin were divided by the total number of counts in the random distribution. Also, comparison of experimental and simulated random incidence distributions determined the thickness of the Si substrate. Before discussing the data and simulations, we need to clarify exactly what adatom sites we have tested in each of the simulations. Fig. 1 shows top views of the three states we tested. The intermediate state of the Ko model, with an Sb atom on top of an Sb-Ge heterodimer, is shown in Fig. l(a). We took the Sb atom on top to be in a substitutional site; however, moving it 0.4 A away from a substitutional site does not change the result. The Sb dimers have the experimentally measured [8] dimer bond length of 2.80 A, and we treated the Sb-Ge heterodimers as if each atom were in a homodimer. Thus, the Sb atom is in the same site as the other Sb dimer atoms on the surface, and the Ge atom is in a site that would give it the experimentally measured [12] dimer bond length of 2.60 A if it were part of a GeeGe dimer. Since this is the only part of the Ko model
t-
(b)
(4
l
Si
o Ge @ Sb
Fig. 1. Top view of the Si( 100) surface showing the three adatom configurations tested: (a) Ko model (lifted Sb atom on top of GeeSb heterodimer). (b) experimentally determined final state, (c) all dimer model (Ge in dimer-like sites on undisturbed Sb dimers). Solid circles are Si atoms, open circles are Ge atoms. and cross-hatched circles are Sb atoms. Size of circle indicates proximity to top layer.
A.A. Builes III et al. I Nucl. Instr. and Meth. in Phys. Rex B 136-138
that we tested, from this point forward use of the term Ko model will refer specifically to this state. TheJinal state, shown in Fig. l(b), consists of Sb dimers on top of near-substitutional Ge atoms, and we have used the experimentally measured [ 11,131 values for the adatom sites. The last model, shown in Fig. l(c), is the configuration we found to fit the before anneal, 0.68 ML Ge data [4]. We refer to this model, with Ge in dimer-like sites between the Sb dimer rows and Sb in symmetric dimer sites, as the all dimer model. The Sb dimer atoms are in the same sites as in the Ko model, with a dimer bond length of 2.80 A, and the Ge atoms are in sites identical to the Sb atoms except for a 90” rotation, giving them a 2.80 A dimer bond length as well. With these adatom sites for the three models, we compared our experimental energy distributions to simulated distributions for beam incidences in the (100) axial direction and along the {100) plane 6” from the (100) axis. We chose these directions because beam incidences near the (100) axis yield the greatest sensitivity to lateral positions on the (100) surface. Simulations showed that other beam incidences near the (100) axis would not add appreciably to our ability to distinguish between models. In fact, for these data the {100) planar incidence was redundant and therefore not shown. Beam incidences near other major axes were not used because of uncertainty in the : coordinates of the adatoms in the different models. Since the Monte Carlo simulation of channeling gives us a distribution of ion positions and energies, the comparison between simulation and data is based on peak height, shape, and position on the energy axis, with a xz value determining the goodness of fit. Fig. 2 shows simulated energy distributions for the three models compared to experimental data taken for the (100) beam incidence on a sample with 0.15 ML Ge on Sb-terminated Si( 100). As mentioned before, catching the system in an initial or intermediate state requires sufficiently low Ge coverage and deposition temperature. The upper set of data and simulations in Fig. 2 represents the results of our matching these conditions. While both the all dimer model and the final state fit these data reasonably well, the Ko model obvious-
-
(1998) 804-809
807
- final state all dimer model
before anneal
after anneal
2150
2250
2350
Energy (keV) Fig. 3. Experimental energy distributions (dots) compared to simulated energy distributions (lines) for the three models for 0.15 ML Ge on Sb-terminated Si( 100). Slight energy differences in the before and after anneal data arise from differences in Si thickness, as the two sets of data were taken on different spots on the sample. Only the (100) aligned spectra are shown here (see text). and each is normalized to the random yield. as described in the text.
ly does not. Upon annealing (lower part of figure), the fits between simulations of the models and experimental data do not change appreciably. We see no evidence that the Ko model describes the state of the system at this low coverage either after deposition at room temperature or after annealing at 350°C. However, the possibility exists that the transition to the final state has occurred already at this low coverage. In that case, we can say nothing about the initial states. Another possibility is that the system is in an initial state such as the all dimer model. The all dimer model and the final state both have Sb in (slightly different) dimer sites, whereas the Ge is in dimer-like sites in the all dimer model and near-substitutional sites in the final state. Thus, distinguishing between these two states requires good statistics in the Ge peak,
808
A. A. Buiks
III et ul. I Nucl. Insfr. und Mcth. in Ploys. Rrs. B 136-13X
which we do not have for the low coverage data. Depositing a higher coverage of Ge allowed us to look for a transition induced by annealing the sample, answering the question about whether we can catch the system in an initial state. Fig. 3 shows the (100) data and simulations for 0.68 ML Ge on Sb-terminated Si(100) before and after annealing. In this case, the Ge data clearly show a transition from fitting the all dimer model before annealing to fitting the final state after annealing. The Sb data also support a shift from the all dimer model to the final state upon annealing. The Ko model does not fit either the before anneal or after anneal data. The driving forces that can cause the system to go from an initial state to the final state are Ge coverage and sample temperature during growth or during subsequent annealing. Since we caught the system prior to the transition to the final state at 0.68 ML Ge before annealing, at the lower coverage of 0.15 ML Ge we expect that we also see the system in an initial state. The inability to discern a
before anneal
(190XI X04-XOY
transition in the low coverage data could indicate that we had too little Ge on the surface to reach the final state or that the final state is altered by the low coverage and the transition is subtle. From the high coverage scattering yields for the Ge and Sb peaks, we deduced that the all dimer model provides a good fit to the data. If we take the low coverage, before anneal data to be in an initial state, then that state is likely to be the all dimer model and, based on the data, cannot be the Ko model. The Ko model, in fact, does not fit any of the four data sets. In summary, using transmission ion channeling and MeV scattered ion energy distributions, we have tested the single adatom exchange model for surfactant-mediated growth. We studied the Ge/Sb/Si( 100) system at two different Ge coverages and after treatment at two different temperatures. For the higher Ge coverage, we have identified a clear transition from an initial state to a final state by depositing the Ge at near room temperature and then annealing at 350°C. The higher coverage system before annealing is in a state with Ge in dimer-like sites and Sb in undisturbed, symmetric dimer sites. After annealing the higher coverage sample, the system goes to the final state with Sb dimers on top of near-substitutional Ge atoms. At the lower Ge coverage, we cannot see a transition upon annealing the sample, but the data are consistent with the system being in the same initial state identified in the higher coverage data. The state expected by Ko et al. to be seen in the low coverage, before anneal conditions is not seen under any of the conditions we tested.
Acknowledgements after anneal
2150
We are grateful for the technical assistance of R. Johns and D. McNeill. This work was supported by the NSF (DMR) and the University of Florida Division of Sponsored Research. 2250
2350
Energy (keV) Fig. 3. Same as Fig. 2 except for the higher Ge coverage ML.
References of 0.68
[I] M. Copel. M.C. Reuter. E. Kaxiras. Rev. Lett. 63 (1989) 632.
R.M. Tromp.
Phys.
A. A. Builes III et ul. I Nucl. Insir. und Meth. in Phys. Rex B 136-138 [2] R.M. Tromp. M.C. Reuter. Phys. Rev. Lett. 68 (1992) 954. [3] B.D. Yu, A. Oshiyama, Phys. Rev. Lett. 72 (1994) 3190. [4] M.A. Boshart. A.A. Bailes III, L.E. Seiberling. Phys. Rev. Lett. 77 (1996) 1087. [5] Y.J. Ko, J.Y. Yi. S.J. Park, E.H. Lee, K.J. Chang. Phys. Rev. Lett. 76 (1996) 3160. [h] A. khizdkd, Y. Shiraki. J. Electrochem. Sot. 133 (1986) 666. [7] L.E. Seiberling. P.F. Lyman. M.W. Grant. J. Vat. Sci. Technol. A II (1993) 715. [8] M.A. Boshart, A.A. Bailes III. A. Dygo. L.E. Seiberling. J. Vat. Sci. Technol. A 13 ( 1995) 2764. [9] J.F. Ziegler, Helium: Stopping Powers and Ranges in All Elemental Matter. Pergamon Press. New York. 1977.
( 1998)804-409
809
M.A. Boshart. A. Dygo. L.E. Seiberling. Phys. Rev. A 5 1 (1995) 2637: A. Dygo. M.A. Boshart. L.E. Seiberling. N.M. Kabachnik, ibid 50 (1994) 4979: A. Dygo. M.A. Boshart, M.W. Grant, L.E. Seiberling. Nucl. lnstr. and Meth. B 93 (1994) 117. [11] M.A. Boshart, A.A. Bailes Ill. L.E. Seiberling. Surf. Sci. 348 (1996) L75. [12] M.W. Grant, D.J. Dieleman. M.A. Boshart. L.E. Seiberling. Phys. Rev. B 49 ( 1994) 16534. [13] M.W. Grant, M.A. Boshart. D.J. Dieleman. L.E. Seiberling. Surf. Sci. 316 (1994) L1088. [lo]
ELSEVIER
Nuclear
Instruments
and Methods
in Physics Research B 136-138 (1998) 804-809
Experimental test of the single adatom exchange model in surfactant-mediated growth of Ge on Si( 100) A.A. Bailes III ‘,*, M.A. Boshart ” Depurirwttt
b, L.E. Seiberling
a
of PII)‘.si(.s. Cbtiwrsitj~ qf Fhridu. P. 0. Bo\- 11X440, Grrit~esrill~~. FL 3261 I-X440, ” 376 Hcwttitqcy .4 wtntu. Cookc~rill~~. TN 3KSOI. (iSA
VSA
Abstract
We have tested the single adatom exchange model for surfactant-mediated growth. Using two samples with different coverages of Ge on Sb-terminated Si( IOO), we generated energy distributions of scattered MeV ions from transmission ion channeling experiments. We studied the system both after room temperature deposition of Ge and after annealing at 350°C. We then compared simulated energy distributions for the single adatom exchange model to the experimental energy distributions. No combination of temperature and coverage produced a good fit between data and simulations of this model. Before annealing. however, a model having Ge in dimer-like sites on top of undisturbed Sb dimers describes the data well for both Ge coverages. 0 1998 Elsevier Science B.V. 68.35.B~; 79.20.Rf; 79.60.Bm; 79.60.D~ K~JYNI~~.s:High energy ion scattering; Ion-solid interactions; PACS:
crystal epitaxy;
Scattering;
Channeling;
Semiconducting
surfaces:
Single
Surface structure
The creation of abrupt, flat interfaces in heteroepitaxial growth, as is required for such structures as Si/Ge superlattices, is made difficult by islanding and interdiffusion. Although the layer-by-layer (Frank-Van der Merwe) growth mode can be induced by deposition at low substrate temperature, the resulting crystallinity often is reduced. To overcome these problems, Cope1 et al. [l] introduced the idea of using a surfactant, a species that strongly segregates to the surface during growth and can kinetically inhibit islanding. Surfactants
‘Corresponding author. Tel.: 352 392-9219; fax: 352 3928586; e-mail: abailesk$phys.ufl.edu. 0168-583X/98/$19.00 0 1998 Elsevier Science B.V. All rights reserved. PIISO168-583X(97)00762-3
such as As or Sb passivate Si and Ge( 100) surfaces and lower the surface free energy when they segregate to the top layer. When deposited on the surface before the adsorbate, surfactant atoms undergo a site exchange process with the adatoms, thus rising to the top layer and reducing the diffusion length of the adatoms on the surface. This reduced diffusion length is thought to be the reason for the inhibition of islanding [2]. While surfactant-mediated growth has generated much interest, only a few studies have looked at the details of the site exchange mechanism. Based on low energy electron microscopy data for Ge grown on As-terminated Si( loo), Tromp and Reuter [2] proposed a two-dimer correlated exchange
A. A. Buiks
III et al. I Nwl.
Instr. unrl Meth. in Pll,s.
mechanism which starts with Ge adsorbing on top of, and then breaking, the surfactant dimers. Yu and Oshiyama [3] proposed a single-dimer exchange process based on total energy calculations showing that the initial adsorption site is more likely to be between surfactant dimer rows than on top of broken surfactant dimers. In a recent paper [4], we compared simulations representing these two exchange models to experimental scattered ion energy distributions. We found the Ge on Sb-terminated Si(100) system in an initial state having Ge in dimer-like sites on top of undisturbed Sb dimers. similar to an early stage of the Yu and Oshiyama model. As in the single-dimer exchange model of Yu and Oshiyama, the single-adatom exchange model proposed by Ko et al. [5] resulted from total energy calculations and starts with the adatoms adsorbing between the surfactant dimer rows. From these adsorption sites the adatoms exchange immediately, one adatom at a time, with the underlying surfactant. In all of these models, the system goes to a final state with surfactant dimers on top of near-substitutional adatoms after site exchange. In this work, we compare our experimental transmission ion channeling data to the single-adatom exchange model [5]. Ko et al. state that under the proper conditions (low growth temperature and low adatom coverage). an intermediate state with a surfactant atom residing directly above a surfactant-adsorbate heterodimer should be observable. However, we have found that this configuration of surfactant and adsorbate atoms does not fit data we have collected after sample preparation under the specified conditions, nor does it fit our data taken after annealing the sample at 350°C or using a higher Ge coverage. Comparing our data to other stages of the single-adatom exchange model was unnecessary for two reasons. First, we have tested the state that Ko et al. specify as the one that we should see for our sample preparation conditions. Second, the other states we could test are similar to parts of the Yu and Oshiyama model that we have tested previously [4]. Along with the comparison of our data to the Ko model, we show comparisons to the final state and to the state which best fits the data, Ge in dimer-like sites on top of undisturbed Sb dimers.
Rrs. B 136 138
( 1998) 804-X09
805
Samples were prepared with an ex situ dopantselective chemical etch to create a “thin window”, a piece of Si approximately 1 cm x 1 cm by 500 urn thick with a circular region of diameter -6 mm that has been thinned to 0.5 urn. We then cleaned the samples chemically and deposited a protective oxide using the Shiraki technique [6] before placing the samples in the ultra-high vacuum (UHV) chamber (base pressure -5 x 10. ‘I Torr). After thoroughly degassing the sample, we desorbed the oxide with a heat lamp. At this point, low energy electron diffraction (LEED) yielded a 2 x 1 pattern with low background and sharp spots, indicating a properly prepared Si(100) surface. We deposited Sb from a boron nitride effusion cell with the sample heated to -500°C. Exposure of the sample to several monolayers (ML) of Sb ensured saturation at -0.85 ML. The LEED pattern at this point changed to 1 x 1 with faint half-order spots. After allowing the sample to cool to near room temperature, we deposited Ge from an effusion cell at the rate of 0.15 ML/min. No noticeable change occurred in the LEED pattern for the sample with a Ge coverage of 0.15 ML, while the sample with 0.68 ML of Ge showed a LEED pattern with higher background and half-order spots no longer visible. Rutherford Backscattering (RBS) yielded coverages of 0.80 and 0.15 ML for the Sb and Ge, respectively, for the low Ge coverage sample and 0.84 and 0.68 ML for the high Ge coverage sample. More detailed descriptions of our sample preparation and experimental setup can be found elsewhere [7,8]. Upon completing the sample preparation, we transferred the sample under UHV to the scattering chamber (base pressure -5 x lo-” Torr) and placed the sample on a two-axis goniometer. A 3.5 MV van de Graaff accelerator produced a 2.5 MeV He+ beam, collimated to an angular divergence of 0.03” at the scattering chamber. With the sample in transmission geometry (adsorbates on the beam-exit side of the sample), we visually aligned the (100) axis of the sample with the beam by viewing the transmitted beam on the quartz at the end of the beamline. Comparison of Si yields with simulations has shown that the accuracy of alignment is 0.02”-0.05” for this technique. Taking as a second reference point a beam incidence -8”
806
A.A. Builes III et (11.I Nucl. Instr. und Meth. in P~z_IY.Rex B 136-13X (19981 804-809
from the (100) axis along the { 100) plane, we established the relation between lab and crystal coordinates. Thus we could readily choose a beam incidence at any tilt (I from the (100) axis and azimuth q relative to the {100) plane. In these experiments we obtained energy distributions in three directions: (0,~) = (O’.O’), (6”,0”), and (6O.30”). The first of these directions is the (100) axis, the second is the ( 100) plane, and the third is a “random” direction. in which ion energy losses approach those in amorphous media [9]. The direction chosen as the random is the result of simulating mean ion energy loss over a large range of H and cp and finding a direction where the difference between mean energy loss and random energy loss is minimal [S]. A passivated. ion-implanted solid state detector collected and analyzed the energy of the scattered ions at a scattering angle of 79”. The integrated beam current of each spectrum was limited to 24 uC (1.5 x 10” ions/cm’), which is within the range of beam dose that yields minimal irradiation damage to the Si crystal structure. Likewise we collected the axial and planar incidence data before the random incidence data because of the greater damage caused by the random incidence beam. One more step taken to minimize crystal damage was collecting before anneal and after anneal data on different spots on the thin window. The three spectra ((loo), { 100 ‘,, and random) corresponding to a given temperature and Ge coverage were taken on a single spot, however, in order to avoid problems with variations in adsorbate coverage and Si thickness. Data taken in the experiments consist of the three spectra at different beam incidences, after room temperature Ge deposition (henceforth called “before anneal”) and again after annealing the sample at 350°C for 10 min. This was done for both the low Ge coverage sample and the high Ge coverage sample for a total of 12 spectra. A simulated distribution of ion positions and energies before scattering, generated by a Monte Carlo simulation of the channeling process [lo], produced simulated energy distributions of scattered ions when overlapped with trial adatom sites [8,11]. By putting in up to five adatom sites at a time and comparing the simulated energy distributions with experimental distributions, we can test
theoretical models of the exchange mechanism in surfactant-mediated growth. Both the experimental and simulated energy distributions were normalized to the random yield; viz., for each distribution, the counts in each energy bin were divided by the total number of counts in the random distribution. Also, comparison of experimental and simulated random incidence distributions determined the thickness of the Si substrate. Before discussing the data and simulations, we need to clarify exactly what adatom sites we have tested in each of the simulations. Fig. 1 shows top views of the three states we tested. The intermediate state of the Ko model, with an Sb atom on top of an Sb-Ge heterodimer, is shown in Fig. l(a). We took the Sb atom on top to be in a substitutional site; however, moving it 0.4 A away from a substitutional site does not change the result. The Sb dimers have the experimentally measured [8] dimer bond length of 2.80 A, and we treated the Sb-Ge heterodimers as if each atom were in a homodimer. Thus, the Sb atom is in the same site as the other Sb dimer atoms on the surface, and the Ge atom is in a site that would give it the experimentally measured [12] dimer bond length of 2.60 A if it were part of a GeeGe dimer. Since this is the only part of the Ko model
t-
(b)
(4
l
Si
o Ge @ Sb
Fig. 1. Top view of the Si( 100) surface showing the three adatom configurations tested: (a) Ko model (lifted Sb atom on top of GeeSb heterodimer). (b) experimentally determined final state, (c) all dimer model (Ge in dimer-like sites on undisturbed Sb dimers). Solid circles are Si atoms, open circles are Ge atoms. and cross-hatched circles are Sb atoms. Size of circle indicates proximity to top layer.
A.A. Builes III et al. I Nucl. Instr. and Meth. in Phys. Rex B 136-138
that we tested, from this point forward use of the term Ko model will refer specifically to this state. TheJinal state, shown in Fig. l(b), consists of Sb dimers on top of near-substitutional Ge atoms, and we have used the experimentally measured [ 11,131 values for the adatom sites. The last model, shown in Fig. l(c), is the configuration we found to fit the before anneal, 0.68 ML Ge data [4]. We refer to this model, with Ge in dimer-like sites between the Sb dimer rows and Sb in symmetric dimer sites, as the all dimer model. The Sb dimer atoms are in the same sites as in the Ko model, with a dimer bond length of 2.80 A, and the Ge atoms are in sites identical to the Sb atoms except for a 90” rotation, giving them a 2.80 A dimer bond length as well. With these adatom sites for the three models, we compared our experimental energy distributions to simulated distributions for beam incidences in the (100) axial direction and along the {100) plane 6” from the (100) axis. We chose these directions because beam incidences near the (100) axis yield the greatest sensitivity to lateral positions on the (100) surface. Simulations showed that other beam incidences near the (100) axis would not add appreciably to our ability to distinguish between models. In fact, for these data the {100) planar incidence was redundant and therefore not shown. Beam incidences near other major axes were not used because of uncertainty in the : coordinates of the adatoms in the different models. Since the Monte Carlo simulation of channeling gives us a distribution of ion positions and energies, the comparison between simulation and data is based on peak height, shape, and position on the energy axis, with a xz value determining the goodness of fit. Fig. 2 shows simulated energy distributions for the three models compared to experimental data taken for the (100) beam incidence on a sample with 0.15 ML Ge on Sb-terminated Si( 100). As mentioned before, catching the system in an initial or intermediate state requires sufficiently low Ge coverage and deposition temperature. The upper set of data and simulations in Fig. 2 represents the results of our matching these conditions. While both the all dimer model and the final state fit these data reasonably well, the Ko model obvious-
-
(1998) 804-809
807
- final state all dimer model
before anneal
after anneal
2150
2250
2350
Energy (keV) Fig. 3. Experimental energy distributions (dots) compared to simulated energy distributions (lines) for the three models for 0.15 ML Ge on Sb-terminated Si( 100). Slight energy differences in the before and after anneal data arise from differences in Si thickness, as the two sets of data were taken on different spots on the sample. Only the (100) aligned spectra are shown here (see text). and each is normalized to the random yield. as described in the text.
ly does not. Upon annealing (lower part of figure), the fits between simulations of the models and experimental data do not change appreciably. We see no evidence that the Ko model describes the state of the system at this low coverage either after deposition at room temperature or after annealing at 350°C. However, the possibility exists that the transition to the final state has occurred already at this low coverage. In that case, we can say nothing about the initial states. Another possibility is that the system is in an initial state such as the all dimer model. The all dimer model and the final state both have Sb in (slightly different) dimer sites, whereas the Ge is in dimer-like sites in the all dimer model and near-substitutional sites in the final state. Thus, distinguishing between these two states requires good statistics in the Ge peak,
808
A. A. Buiks
III et ul. I Nucl. Insfr. und Mcth. in Ploys. Rrs. B 136-13X
which we do not have for the low coverage data. Depositing a higher coverage of Ge allowed us to look for a transition induced by annealing the sample, answering the question about whether we can catch the system in an initial state. Fig. 3 shows the (100) data and simulations for 0.68 ML Ge on Sb-terminated Si(100) before and after annealing. In this case, the Ge data clearly show a transition from fitting the all dimer model before annealing to fitting the final state after annealing. The Sb data also support a shift from the all dimer model to the final state upon annealing. The Ko model does not fit either the before anneal or after anneal data. The driving forces that can cause the system to go from an initial state to the final state are Ge coverage and sample temperature during growth or during subsequent annealing. Since we caught the system prior to the transition to the final state at 0.68 ML Ge before annealing, at the lower coverage of 0.15 ML Ge we expect that we also see the system in an initial state. The inability to discern a
before anneal
(190XI X04-XOY
transition in the low coverage data could indicate that we had too little Ge on the surface to reach the final state or that the final state is altered by the low coverage and the transition is subtle. From the high coverage scattering yields for the Ge and Sb peaks, we deduced that the all dimer model provides a good fit to the data. If we take the low coverage, before anneal data to be in an initial state, then that state is likely to be the all dimer model and, based on the data, cannot be the Ko model. The Ko model, in fact, does not fit any of the four data sets. In summary, using transmission ion channeling and MeV scattered ion energy distributions, we have tested the single adatom exchange model for surfactant-mediated growth. We studied the Ge/Sb/Si( 100) system at two different Ge coverages and after treatment at two different temperatures. For the higher Ge coverage, we have identified a clear transition from an initial state to a final state by depositing the Ge at near room temperature and then annealing at 350°C. The higher coverage system before annealing is in a state with Ge in dimer-like sites and Sb in undisturbed, symmetric dimer sites. After annealing the higher coverage sample, the system goes to the final state with Sb dimers on top of near-substitutional Ge atoms. At the lower Ge coverage, we cannot see a transition upon annealing the sample, but the data are consistent with the system being in the same initial state identified in the higher coverage data. The state expected by Ko et al. to be seen in the low coverage, before anneal conditions is not seen under any of the conditions we tested.
Acknowledgements after anneal
2150
We are grateful for the technical assistance of R. Johns and D. McNeill. This work was supported by the NSF (DMR) and the University of Florida Division of Sponsored Research. 2250
2350
Energy (keV) Fig. 3. Same as Fig. 2 except for the higher Ge coverage ML.
References of 0.68
[I] M. Copel. M.C. Reuter. E. Kaxiras. Rev. Lett. 63 (1989) 632.
R.M. Tromp.
Phys.
A. A. Builes III et ul. I Nucl. Insir. und Meth. in Phys. Rex B 136-138 [2] R.M. Tromp. M.C. Reuter. Phys. Rev. Lett. 68 (1992) 954. [3] B.D. Yu, A. Oshiyama, Phys. Rev. Lett. 72 (1994) 3190. [4] M.A. Boshart. A.A. Bailes III, L.E. Seiberling. Phys. Rev. Lett. 77 (1996) 1087. [5] Y.J. Ko, J.Y. Yi. S.J. Park, E.H. Lee, K.J. Chang. Phys. Rev. Lett. 76 (1996) 3160. [h] A. khizdkd, Y. Shiraki. J. Electrochem. Sot. 133 (1986) 666. [7] L.E. Seiberling. P.F. Lyman. M.W. Grant. J. Vat. Sci. Technol. A II (1993) 715. [8] M.A. Boshart, A.A. Bailes III. A. Dygo. L.E. Seiberling. J. Vat. Sci. Technol. A 13 ( 1995) 2764. [9] J.F. Ziegler, Helium: Stopping Powers and Ranges in All Elemental Matter. Pergamon Press. New York. 1977.
( 1998)804-409
809
M.A. Boshart. A. Dygo. L.E. Seiberling. Phys. Rev. A 5 1 (1995) 2637: A. Dygo. M.A. Boshart. L.E. Seiberling. N.M. Kabachnik, ibid 50 (1994) 4979: A. Dygo. M.A. Boshart, M.W. Grant, L.E. Seiberling. Nucl. lnstr. and Meth. B 93 (1994) 117. [11] M.A. Boshart, A.A. Bailes Ill. L.E. Seiberling. Surf. Sci. 348 (1996) L75. [12] M.W. Grant, D.J. Dieleman. M.A. Boshart. L.E. Seiberling. Phys. Rev. B 49 ( 1994) 16534. [13] M.W. Grant, M.A. Boshart. D.J. Dieleman. L.E. Seiberling. Surf. Sci. 316 (1994) L1088. [lo]