- Email: [email protected]
Kinetic growth manipulation of Si(0 0 1) homoepitaxy
Surface Science 552 (2004) 35–45 www.elsevier.com/locate/susc
Kinetic growth manipulation of Si(0 0 1) homoepitaxy Marcus Esser, Erwin Zoethout, Harold J.W. Zandvliet, Herbert Wormeester *, Bene Poelsema Solid State Physics, Faculty of Applied Physics, MESA+ Research Institute, University of Twente, P.O. Box 217, Enschede 7500 AE, The Netherlands Received 3 March 2003; accepted for publication 6 January 2004
Abstract We have confirmed in a combined diffraction and STM study that the usual kinetic growth manipulation (KGM) applied to Si/Si(0 0 1) (nucleation at relatively low temperatures and completion of monolayer growth at high temperatures) does not lead to a smooth growth front. We have identified the physical reason for this unexpected behaviour: an anti phase boundary (APB) network develops during growth, due to the 2 · 1 reconstruction of the clean Si(0 0 1) surface. The density of this APB network can be substantially reduced by application of a different and optimised KGM procedure. Following a recipe in which 1 ML of Si on Si(0 0 1) is deposited at a relatively low temperature (525 K), followed by a short anneal to 750 K, results in a surface flatness similar to that of the clean Si(0 0 1) 2 · 1 surface. Up to 10 ML of material deposited with a flash anneal after the deposition of each additional layer resulted in a surface with a negligible reduction of the in-phase and out-of-phase intensity of a reflected low energy electron beam, indicative of an almost perfectly smooth growth front. STM images support this observation. The low thermal budget of this method reduces intermixing effects in hetero-epitaxial growth of group IV semiconductor (0 0 1)-faces. 2004 Elsevier B.V. All rights reserved. Keywords: Epitaxy; Growth; Silicon; Low energy electron diffraction (LEED); Scanning tunneling microscopy
1. Introduction Novel thin film structures synthesized by MBE or CVD on silicon have found a wide variety of applications in science and technology. Smooth and abrupt interfaces between epitaxial layers are often crucial factors for the deviceÕs performance. The smoothness and abruptness of the interface leads to a conflicting demand with respect to the
*
Corresponding author. Tel.: +31-53-489-3148; fax: +31-53489-1101. E-mail address: [email protected] (H. Wormeester).
growth conditions. For hetero-epitaxial growth the growth temperature should be sufficiently low to reduce atom diffusion and hence intermixing of the layers and dopant segregation, while on the other hand the growth temperature should be sufficiently high to avoid kinetic roughening. Potential growth recipes that focus on low sample temperature during deposition while still growing smooth layers can be developed with homoepitaxial growth. Many complicating factors associated with hetero-epitaxial growth like intermixing and lattice misfit can then be neglected. For Si and Ge surfaces, the reconstruction present on these surfaces plays a major role in the final surface
0039-6028/$ - see front matter 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.susc.2004.01.012
M. Esser et al. / Surface Science 552 (2004) 35–45
morphology. The key issue in this growth study is to obtain layer-by-layer growth with the smallest possible thermal budget for Si on Si(0 0 1). This implies that after every deposited monolayer a surface that is in every aspect comparable to the clean surface has to be achieved. The Si(0 0 1) surface is stabilized by a (2 · 1) surface reconstruction where the atoms of the outermost layer couple in pairs to form dimers in order to reduce the number of dangling bonds of an ideally bulk-truncated surface. When Si atoms are deposited on Si(0 0 1) surfaces at room temperature the atoms are sufficiently mobile to combine quickly to form dimers. The dimers can be divided in two classes: dimers residing on top of the substrate dimer rows and dimers residing in the troughs between the substrate dimer rows. At temperatures slightly above room temperature (300–350 K) the on-top dimers are able to diffuse parallel to the substrate dimer rows [1]. At even higher temperatures (450 K) also diffusion perpendicular to the substrate rows becomes active [2,3]. As Si islands nucleate and grow on (0 0 1) surfaces, the dimer rows in neighbouring islands have a probability of 50% to be in the correct registry. An anti-phase boundary (APB) will form if two islands meet, and their internal dimer rows are not aligned. For the Si(0 0 1) surface there are in principle two different APBs (see insert Fig. 1), one type of APB runs parallel to the substrate dimer rows (APBA ) and the other runs perpendicular to substrate dimer rows (APBB ). The latter APB acts as a preferential nucleation site for island growth [4]. The first type of APB gives, in combination with the fact that the islands are elongated parallel to the dimer row direction, rise to the formation of trenches and double layer steps [5]. The anti-phase boundary (APB)-network [6] is of key importance for the roughness of the grown layer. In this paper it will be shown that by modulating the temperature during growth the APB network can be manipulated in such a way that smooth layers can be grown at relatively low temperatures. Before demonstrating this with the homoepitaxial growth of 15 ML, the temperature dependence of the APB network will be discussed. The size of the APB network is determined from STM, while SPA-LEED is employed to study
0
10
T [K] 500
700
900
-1
T
10 island density [a-2 ]
36
400
300
T0
(1)
T
(2)
-2
10
T
(3)
L
-3
10
T
(4)
-4
10
B -5
10
-6
10
1.0
1.5
2.0 -3
1/T [10 K
2.5 -1
3.0
]
Fig. 1. Si island densities in units of the dimer spacing a on Si(0 0 1) vs. growth temperature. Up-triangle values are calculated from the APB network. Circle values have been published by Mo et al. [8,9]. The lines are guides to the eye. T0 is the growth temperature, Tð1Þ –Tð4Þ are the flash temperatures of the SPA-LEED measurement as discussed in the text. The insert visualizes the APB network as a result of non-aligned dimer rows. The definition of the spacing L and B between APBs along and perpendicular to the substrate dimer row direction is shown.
the annealing behaviour of the APB network in detail.
2. Experimental All experiments have been conducted in ultrahigh vacuum (base pressure <1010 mbar). Two systems were used, one equipped with highresolution low energy electron diffraction (HR-or SPA-LEED manufactured by Omicron), with a resolution better than 0.1% of a Brillouin Zone (BZ). The other system was equipped with an STM-1 (Omicron) that allowed measurements at room temperature. Fresh Si samples were outgassed by resistive heating for over 24 h up to a temperature of 600 C. Final preparation was done by repeated rapid flashing to 1100 C, while the pressure was kept below 1.0 · 1010 mbar. Surfaces prepared in this way showed a FWHM of 0.5% BZ
M. Esser et al. / Surface Science 552 (2004) 35–45
3. Conventional growth The investigation was started by looking at one of the key parameters in growth, the number density of islands as a function of temperature. The APB network opens a unique way to determine the maximum island number density, especially at high coverages of grown material [6,7]. A detailed description of this method developed by Zoethout et al. was published previously [7] and is only briefly mentioned here. The random nucleation of islands leads for epitaxy on the (2 · 1) reconstructed Si(0 0 1) surface to neighbouring islands that have a probability of 50% to be in the wrong registry, i.e. they can not merge together and an APB results. The island density can be extracted from the APB network that has formed after island coalescence. However, the density extracted from the APB network is four times too low as a result of the equal chance of being in or out of registry. The island density is thus N ¼ 4=ðhLi hBiÞ, where hLi (hBi) is the averaged spacing between APBs in a direction parallel (perpendicular) to the substrate row direction (see insert in Fig. 1). In Fig. 1 a plot of N vs. 1=T in the temperature range from RT to 1000 K is shown. The dotted line refers to the work of Mo et al. [8,9], who measured the island density after deposition of 0.07 ML. The island density depends on the coverage and will reach a maximum at the time just before coalescence sets in. It is exactly this maximum number that is extracted from the APB network. The kink in the curve at a temperature around 560–600 K has been attributed to an increase of the size of the critical nucleus [7]. Note that in a nearest neighbour model an increase of the stable nucleus from one dimer to two dimers coincides energetically with island coarsening. This additional process becomes active at higher temperatures, and is due to attachment and detachment at kink sites [10–12]. The temperature at which the kink occurs of the two curves differs by
about 40 K. This is attributed to the uncertainties in sample temperature measurement. The slope of the two curves in the temperature range below 550 K is identical and is related to the diffusion process of Si dimers parallel to the dimer rows [1,8]. For the real-time SPA-LEED measurements first the appropriate energies for the in-and out-of phase conditions of the central spot have to be determined. Low energy electrons are favoured for real-time measurements as these have a higher surface sensitivity and the decrease of intensity due to the Debye–Waller effect for elevated sample temperature is smaller for low electron energies. This implies that energies with a perpendicular relative phase number of Sz ¼ 2 are desired. However, low intensity prohibits for example measurement at the real out-of-phase condition. Numerous test experiments using electron energies between 36 eV and 324 eV for the spot intensity variations during growth were performed. At an energy of 54.8 eV the best out of phase oscillations (see below) were observed, while at 92.3 eV the spot intensity remained at the highest level with the smallest fluctuation, indicative of in-phase condition. Fig. 2 shows the spot intensities measured at the chosen energies in real time during growth of Si on
1.0 normalized peak intensity []
of the specular reflected electron beam with SPALEED, while the missing dimer defect density as observed with STM is low. Si was deposited from a resistively heated Si bar.
37
0.8
0.6
0.4
0.2
0.0 0.0
0.5
1.0
1.5
2.0
time [h] Fig. 2. Real-time measurement of central spotÕs peak intensities (normalized, circle symbols: at 54.8 eV, out-of-phase condition; rhomb symbols: 93.2 eV, in-phase condition) during conventional growth of Si on Si(0 0 1) at a sample temperature 575 K.
38
M. Esser et al. / Surface Science 552 (2004) 35–45
Si(0 0 1) at a constant temperature of about 575 K. The observed oscillations are indicative of ‘‘layerby-layer’’ growth. The maxima coincide roughly with the completion of a monolayer. The low intensity of the maxima reveal that the ‘‘layer-bylayer’’ growth is by no means close to the ideal growth. We refer to this situation as ‘‘conventional’’ growth. The out-of phase intensity oscillates with time, while keeping the incident flux constant at approx. 0.05 ML/min. Due to the progressing roughening of the surface the amplitude of the oscillation is decaying from about 27% at the first two intensity maxima to 15% after about 5 grown ML. The in-phase intensity measured simultaneously shows weak oscillations, the oscillation maxima are reached after the maxima of the out-of-phase intensity oscillation. In the case of homoepitaxy on Si(1 1 1) the in-phase oscillation maxima were reached just before the out-of-phase oscillation maxima which can only be explained by a breakdown of the kinematic theory for the in-phase signal [13]. Similar effects have been measured for the growth of Pt/Pt(1 1 1) [14]. The in-phase intensity loss is due to diffuse scattering. The out-of-phase signal is in first approximation not affected. This experiment demonstrates that during the growth of Si on Si(0 0 1) growth oscillations can well be measured, but, in contrast to the homoepitaxy on Si(1 1 1), the growth is accompanied by a roughening growth front. In the next section, a novel, modified growth manipulation method will be presented that prevents this roughening.
4. Kinetic growth manipulation 4.1. Concept of two mobilities Kinetic growth manipulation (KGM) methods have been successfully used on unreconstructed metal surfaces to enhance the layer-by-layer growth. In general these methods are based on the concept of two mobilities, or better, the concept of two island densities and is extensively described and applied on unreconstructed metal surfaces by Rosenfeld et al. [25–27]. KGM aims on these surfaces specifically to reduce the consequence of
the additional energy barrier at the island edges (Ehrlich Schwoebel barrier) for downward diffusion [22–24] in order to enhance the interlayer mass transport. The concept of two mobilities consists of growing the first part of every ML at a lower temperature and afterwards continuing the growth for the rest of the ML at a higher temperature. The idea behind this is the creation of a high density of very small islands in the first phase of every MLÕs growth. This is done to increase the probability of atoms that land on top of those islands to diffuse downwards by increasing the number of attempts to descend when the temperature is higher in the later stage of every monolayerÕs growth overcoming the additional energy barrier at the island edges for downward diffusion. The more adatoms overcome the additional energy barrier and diffuse downwards, the lower is the chance for island nucleation on top of the islands. Fig. 3 shows the intensities of the central SPALEED diffraction spot at in-phase and out-ofphase conditions measured in real-time during the growth of Si on Si(0 0 1) using the concept of two mobilities. Seven monolayer of Si have been deposited; during the first half of every MLÕs deposition the sample temperature was 375 K, during the second half 575 K. The strange-looking scaling in Fig. 3 is due to the fact that at the very beginning of the experiment the sample is at the lower temperature of 375 K. Because of the Debye–Waller effect, the spot intensities are about twice as high as at 575 K. Compared to the conventional case (see Fig. 2) a slightly enhanced outof phase oscillation amplitude is recognizable (40% compared to 30% in the unmanipulated case). But what is more important is the fact that the decay of the out-of phase oscillation amplitude that was observed in the unmanipulated case is still present. This shows that the concept of two mobilities as a kinetic growth manipulation method is not able to suppress the progressive roughening of the surface in the case of homoepitaxy of the reconstructed Si(0 0 1). The reason for this is that by this approach it is not possible to overcome the problem of the building-up of a dense APB network and the corresponding preferred island nucleation in higher layers.
M. Esser et al. / Surface Science 552 (2004) 35–45
1.2
2.0
T(1) T(2)
100% @ 375 K 1.6 1.4 1.2 1.0
100% @ 575 K
0.8 0.6
normalized spot intensities []
575 K 375 K
1.8 normalized peak intensity []
39
T(3)
T(4)
1.0 0.8 0.6 0.4 0.2
0.4 0.2
0.0 0
0.0 0.00
0.25
0.50 0.75 time [h]
1.00
1.25
Fig. 3. Real-time measurement of central diffraction spot intensities during Si growth on Si(0 0 1) using the concept of two mobilities as kinetic growth manipulation method (circle symbols: 54.8 eV out-of-phase condition; rhombic symbols: 92.3 eV in-phase condition; R 1 ML/12 min). The sample temperature is 375 K in the first half of the deposition of every ML (till the minima of the out-of-phase intensity oscillations) and 575 K in the second half (till the maxima of the out-of-phase intensity oscillations). The temperature changes are marked with vertical arrows, the shorter indicate a change to 575 K while the longer indicate a change to 375 K. The scaling of the spot intensities differs from the figures before; this is due to the fact that at the very beginning of the experiment the sample is at the lower temperature of 375 K. Because of the Debye–Waller effect, at 375 K the spot intensities are about twice as high as at 575 K.
4.2. Smoothening a Si monolayer Fig. 4 shows the intensities of the central SPALEED diffraction spot at in-phase and out-ofphase conditions measured during and after the evaporation of Si (rate0.067 ML/min) on the Si(0 0 1) surface at a temperature of 525 K. At the end of the deposition, the in-phase intensity is at about 50%, the out-of-phase intensity at about 15% of it is initial value corresponding with the build-up of a rough growth front. After deposition of 1 ML of silicon the sample was flash annealed to 650 K (1), 700 K (2), 750 K (3) and 775 K (4), 1 min per flash. Already type (1) flashing is sufficient to get the in-phase intensity back to its initial value indicating a surface with just a few defects, such as
15
30 45 time [min]
60
75
90
Fig. 4. Real-time measurement of central diffraction spot intensities during Si growth on Si(0 0 1) and the following flash annealing of the grown film (circle symbols: 54.8 eV out-ofphase condition; rhomb symbols: 92.3 eV in-phase condition; R 0:067 ML/min; T 525 K). 1 ML of Si was grown and afterwards gradually flashed to 650 K (1), 700 K (2), 750 K (3) and 775 K (4) (duration of each flash: 1 min). After the flashes the sample is quickly cooled down to 525 K. In this periods the measured spot intensities rise because of the Debye–Waller effect.
point defects created during ion-sputtering. Optical anisotropy measurements of the Arþ ion bombarded surface revealed the disappearance of such defects at 670 K [15]. The out-of-phase intensity increases step by step (66% of the initial value for flash type (1), 87% (2), 97% (3), 100% (4)) with the flashes. Fig. 5(a) and (b) visualize the effect of flashing using RT STM scans of the Si(0 0 1) surface. Scan (a) was taken after evaporating a little more than 1 ML at 500 K. It is obvious that nucleation took place in the second layer before the first layer was completed. Scan (b) shows the surface with 1 ML grown at 500 K plus a 1 min flash to 660 K. Compared with scan (a) less material remains in layer 2, whereas layer 1 is filled to a higher degree. Note that scan (a) was taken at a 10 times lower deposition flux, comparable to the deposition flux used in the SPA-LEED measurements. According to Venables nucleation theory [16], this implies a 2 times larger island density. As a result, the crossover shown in Fig. 1 will occur at a temperature of about 20 K lower. This implies
40
M. Esser et al. / Surface Science 552 (2004) 35–45
Fig. 5. STM scans on Si(0 0 1) after Si evaporation (quenched to RT), RðaÞ 0:059 ML/min, RðbÞ–ðdÞ 0:59 ML/min. (a) 1 ML grown at 500 K; (b) 1 ML grown at 500 K plus flash to 660 K, 1 min; (c) 4.5 ML grown at 660 K and (d) 4.5 ML grown at 500 K applying KGM, i.e. a 1 min flash to 660 K after every completed ML. Tunnelling parameters: )2.0 V, 0.5 nA, scan width: 250 A.
that the results obtained in the STM and SPALEED are comparable. The procedure depicted in scan (b) thus shows a situation in which the island density was initially even higher than the one shown in scan (a). Scan (b) also shows that the evaporated material is distributed over two layers only. Within the kinematic scattering theory, the out-of-phase intensity Ioutofphase can be expressed as a competition of the exposed layers. For a three layer system this reduces to: Ioutofphase ¼ ½ðh0 h1 Þ ðh1 h2 Þ þ h2 I0
ð1Þ
At 1 ML coverage and only three layers involved, all the material in layer 2, h2 has to be equal to the
non-filled area of layer 1, i.e. h1 ¼ 1 h2 . The coverages h1 , h2 can than be calculated from the out-of-phase intensities: 0 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1 1 ML Ioutofphase 1@ A 1 ð2Þ h2 ¼ 4 I0 (assuming that at the out-of-phase intensity maximum a complete ML has been evaporated, see Table 1 for the results). This flash series shows that a short thermal treatment to a 200 K higher temperature results for this system in an almost perfectly closed layer. If one would try to reach a similar surface quality by growing 1 ML unmanipulated at a certain constant temperature, this
M. Esser et al. / Surface Science 552 (2004) 35–45
41
Table 1 Calculated coverages and filling factors after growth of 1 ML at 525 K and following flash types (1)–(4), for experimental data see Figs. 1 and 4 Type
Tflash /K
Ioutofphase =I0
h1
h2
fAPB
1 ML (1) (2) (3) (4)
– 650 700 750 775
15% 66% 87% 97% 100%
85% 95% 98% 99.5% 100%
15% 5% 2% 0.5% 0%
– 2.5% 2% – <0.5%
temperature would be much higher due to the formation of double steps. The reason for the development of a rough growth front is that the APB network hinders a complete filling of the first layer. This causes nucleation on top of the islands and in this way induces roughening. In the hBi-direction islands (see Fig. 1) cannot grow together and a gap of one lattice unit will remain between the islands. In contrast, an APBB does not hinder two islands merging together (see insert Fig. 1). The fraction of the ML that cannot be contained by the layer itself is defined as fAPB . fAPB can be calculated from the density of the APB network determined with STM (Fig. 1) fAPB ðT Þ ¼
1 1 þ hBiðT Þ
ð3Þ
Table 1 shows a good correspondence between the measured size of the APB network and thus the filling factor fAPB and the amount of material that cannot be incorporated in the first layer (h2 ) after annealing as determined from SPA-LEED. The situation shown in scan (b) of Fig. 5 is not as ideal as expected from the SPA-LEED measurements. This is probably due to the fact that in the STM UHV system there is no real time probe to control the deposition and thus exactly deposit 1 ML. This makes it very hard to produce a completely filled first layer. The error in the amount of deposited material after deposition of 1 ML is estimated as <1% in case of the SPA-LEED measurements and 10% in case of the STM measurements. 4.3. Layer-by-layer The experiences just described are now applied to grow thicker films in a smooth way. After the completion of each ML grown at a temperature
below the coarsening temperature the layer is first annealed to create again a smooth surface with a low defect density. Fig. 6 shows the deposition of 10 ML in this way. As soon as an oscillation maximum of the out-of-phase intensity is reached, the sample is flashed (flash types (1)–(4), see above) for 1 min. Due to the use of resistive heating, the spot intensity measurements are interrupted during flashing, whereas the silicon evaporation continues. Right after the flashes out-of-phase intensities of 50% (1), 60% (2), 75% (3) and 75% (4) of the initial values can be measured, in-phase intensities are in the range of 75–90% (1–4) of their initial values. The out-of-phase-intensities are lower than the ones measured after flashing 1 ML. This has two reasons: (1) the deposition is not interrupted, (2) it takes some time till the sample has cooled down from the flash temperature. Due to the Debye–Waller effect the spot intensities are lower at higher temperatures. Note that the increase of the spot intensities during cooling down the sample is dominated by Debye–Waller effects. Effects due to further annealing are negligible. For the last oscillation the intensities were extrapolated to obtain a value of the in-phase and out-of-phase intensity just after flashing corrected for the difference in Debye–Waller factor for the base and flash temperature. The base temperature, defined as the low temperature at which the first part of each monolayer is grown, amounts to 525 K. Fig. 6 indicates that both the in-phase and the out-of-phase signal returned to the initial levels indicating that the 10th layer has been grown very smooth indeed. We found that already the evaporating-flashing cycle can be repeated at least 15 times without noticeable decay of the spot intensities. The same growth experiments have also been performed at lower base sample temperatures (down to 425 K). The
42
M. Esser et al. / Surface Science 552 (2004) 35–45
scan (d) growth took place at 500 K with 1 min flashes to 660 K after every completed ML.
1.0
normalized peak intensites []
0.8 0.6
5. Discussion
0.4
The growth manipulation method for homoepitaxy on Si(0 0 1) that we describe in this paper has a distinctly positive effect on the smoothness of the grown layer. The oscillation amplitude of the SPA-LEED out-of-phase central spot intensity rises from 10% of the initial value to 50–75% of its initial value (see Fig. 6), where the latter value is even underestimated due to the Debye–Waller effect. That means that the growing Si layers are filled up to a much higher fraction before nucleation in a new layer takes place. Furthermore the growth manipulation results in an absence of the oscillation amplitude decay indicating that there is no progressive kinetic roughening during growth. By evaluating the STM pictures after the growth of 4.5 ML Si with and without applying the growth manipulation method we can make an overview of the distribution of the Si over the visible layers (see Fig. 7). It is already remarkable that the layer deposited at 660 K is less smooth than the layer deposited at 500 K that was flash annealed to 660 K after completion of every ML. The 4.5 ML have been deposited for the conventional growth over 5 exposed layers. Markedly small is the amount of deposit in the 7th and 8th layer in contrast to the only 3 exposed layers for the manipulated growth with only a small amount deposited in the 6th layer. The not completely filled 4th layer is probably due to the inaccuracy of the flash after 1 ML. The surface roughness value can be quantified with a parameter g, defined as [17] P grown 2 ðhn hideal2D Þ n g ¼ P n ideal3D ð4Þ hideal2D Þ2 n n ðhn
0.2 0.0 1.0
1 1 1 1 2 2 3 3 4 4
0.8 0.6 0.4 0.2 0.0
0 30 60 90 120150 180 210 time [min]
Fig. 6. Real-time measurement of central diffraction spot intensities during Si growth on Si(0 0 1) using kinetic growth manipulation (KGM; lower panel: 54.8 eV out-of-phase condition; upper panel: 92.3 eV in-phase condition; R 1 ML/15 min; T 525 K). 10 ML of Si were grown applying flash annealing to 650 K (1), 700 K (2), 750 K (3) and 775 K (4) (duration of each flash: 1 min) after every completed ML. During the flash annealing, the evaporation continues. After the flashes the sample quickly cools down to 525 K. For the last oscillation an extrapolation of the intensity to the time of the flash is made.
flashing after completion of every ML results in the same spot intensity amplitudes. It is a remarkable result for the homoepitaxy on Si(0 0 1) that especially at this low base growth temperatures even after the growth of 10 ML the top layer is as smooth as found here. The STM scans in Fig. 5(c) and (d) show the effect of the applied flashing growth manipulation. As it is hard to visualize thick rough layers with STM the grown layers are only 4.5 ML thick. In scan (c) the material is grown at 660 K (growth at 525 K resulted in a rough growth front that could not be imaged properly), in
In case of ideal 2D growth, i.e. the type of growth where nucleation in layer n þ 1 starts only if layer n is filled up completely, g equals to 0, whereas g ¼ 1 for ideal 3D growth (Poisson distribution). From our conventional growth data g ¼ 0:73 is
M. Esser et al. / Surface Science 552 (2004) 35–45
1.0
controlled growth conventional ideal 2D growth poisson growth
coverage
0.8
0.6
0.4
0.2
0.0
0
1
2
3 4
5 6 layer
7
8
9 10 11
Fig. 7. Comparison of the grown films (thickness 4.5 ML) after conventional growth at 660 K and manipulated homoepitaxy of Si(0 0 1) at a combination of 500 and 660 K: The distribution of material into different atomic layers is plotted. Layer 0 is the first bulk layer. For further comparison, the distributions for ideal 2D growth (dashed line) 3D growth (solid line) of 4.5 ML are plotted into the graph.
obtained, the controlled growth data lead to g ¼ 0:31. Even if the results from our STM measurements do not completely meet the results in terms of surface quality after growth from the SPA-LEED measurements, the g values express the positive effect of the growth manipulation method unequivocally. The kinetic growth manipulation procedure developed in this paper is markedly different from the procedure originally developed for unreconstructed metal surfaces. The latter method is based on the concept of two island densities which assumes that islands created in the initial phase can fully grow together without domain boundaries. The reconstruction of the surface as occurs for Si and Ge (0 0 1) leads to the formation of an APB network that exactly prevents this. The only way to obtain a surface after the deposition of 1 ML that resembles the starting surface, i.e. has a low APB network density, is through a substantial annealing. The complete growth at an elevated temperature is not as successfull. The islands are too large to be able to reduce the APB network density in a reasonable time [18–21]. The APB
43
network density has to be quite small as enhanced nucleation takes place at APBÕs [4]. Increasing the length scale by deposition at more elevated temperatures is not a feasible approach. Growth at increased temperatures leads to the formation of double steps which strongly reduce interlayer transport, an essential ingredient for smooth growth. Layer-by-layer growth is thus only possible at temperatures at which hetero-epitaxy of for instance Si and Ge would lead to strong intermixing. The deposition of 1 ML at a relatively low base temperature followed by a short annealing is capable of producing quite smooth layers. During the flash, the material contained in the small islands quickly redistributes towards neighbouring islands. This ripening is rather fast as a result of the small size of the islands and causes a drastical decrease of the APB network density. The temperature flash process does not lead to only a fill up of the holes in a layer with the material on top of this layer, a process that essentially does not change the density of the APB network. The inphase intensity of the relected electron beam returns after the flash to the value observed for the clean surface, indicating that the same number of defects, like steps etc, are present at the surface after the flash. This implies that the density of the APB network has to be reduced during the flash. We have found that with a base temperature for deposition of 425 K still layer-by-layer growth can be observed. At lower base temperatures the smoothness of the initial layer can not be replicated. This is probably due to the absence of diffusion of dimers perpendicular to dimer rows. The essentially 1D diffusion parallel to the dimer rows, at lower temperatures apparently induces to many APBÕs to be successfully flash annealed with a reasonable thermal budget. A short anneal to 750 K is able to smoothen a layer grown at a reduced temperature and decrease the APB network density considerably. The thermal budget of the layer, i.e. the time at which the sample was at the highest temperature is in this way at least 15–20 times lower than compared to conventional growth while up to 15 ML no decay of the in-phase and out-of-phase intensity can be observed.
44
M. Esser et al. / Surface Science 552 (2004) 35–45
The setting of the base temperature at which the majority of the material is deposited and the flash temperature at which the smooth surface is obtained can be derived from the temperature dependent island density as depicted in Fig. 1. The kink in this curve at about 580 K discriminates the useful range for both the base and flash temperature. A base temperatures at which the majority of the material is deposited above 580 K leads to much larger islands and this will require a larger thermal budget, i.e. an increased anneal time for ripening that would lead to a similar length scale of the APB network as obtained at lower temperatures. This graph also indicates a temperature of at least 580 K for the anneal. The island size also depends very strongly on the temperature and a small increase in temperature can lead to a large increase in island size. It is also important to note that for 1 ML of material double step formation is avoided. A flash temperature that reduces the number of defects on which preferential nucleation takes place to the order of the density of defects on the clean surface, including steps can be considered as the upper limit for the flash temperature in the application of this kinetic growth manipulation method.
6. Conclusions A kinetic growth manipulation (KGM) method that aims at reducing the density of the anti-phase boundary (APB) network emerging during growth experiments on the reconstructed Si(0 0 1) surface has been developed. It has been demonstrated that this KGM method can be used successfully to prevent progressive roughening of the growth front during homoepitaxy of Si(0 0 1). As a consequence, films with a thickness of many ML can be grown in a flat and smooth manner. Furthermore it has been shown, that growth manipulation methods like the concept of two mobilities that work successfully against the additional energy barrier at the island edges for downward diffusion in epitaxy of unreconstructed metal surfaces are not successful in preventing the progressive
roughening of the growth front in the case of the reconstructed Si(0 0 1) surface. The described method is probably general applicable to epitaxial growth of semiconductor and metal layers which surfaces reconstruct with at least two different domains. The reduced thermal budget is especially important if intermixing of substrate and layer has to be avoided.
References [1] B.S. Swartzentruber, Phys. Rev. Lett. 76 (1996) 459. [2] B. Borovsky, M. Krueger, E. Ganz, Phys. Rev. Lett. 78 (1997) 4229. [3] M. Krueger, B. Borovsky, E. Ganz, Surf. Sci. 385 (1997) 146. [4] U.K. Hamers, R.J. K€ ohler, J.E. Demuth, J. Vac. Sci. Technol. A 8 (1990) 195. [5] W. Wulfhekel, PhD thesis, University of Twente, 1997. ISSN 0944–2952. [6] H.J.W. Zandvliet, E. Zoethout, W. Wulfhekel, B. Poelsema, Surf. Sci. 482 (2001) 391. [7] E. Zoethout, P.W. van den Hoogenhof, H.J.W. Zandvliet, B. Poelsema, J. Appl. Phys. 92 (2002) 5785. [8] Y.W. Mo, J. Kleiner, M.B. Webb, M.G. Lagally, Phys. Rev. Lett. 66 (1991) 1998. [9] A. Pimpinelli, J. Villain, D.E. Wolf, Phys. Rev. Lett. 69 (1992) 985. [10] H.J.W. Zandvliet, H.B. Elswijk, E.J. Loenen, Surf. Sci. 272 (1992) 264. [11] N. Kitamura, B.S. Swartzentruber, M.G. Lagally, M.E. Webb, Phys. Rev. B 48 (1993) 5704. [12] C. Pearson, B. Borovsky, M. Krueger, R. Curtis, E. Ganz, Phys. Rev. Lett. 74 (1995) 2710. [13] M. Esser, H. Wormeester, B. Poelsema, Surf. Sci. 482 (2001) 1355. [14] B. Poelsema, A.F. Becker, G. Rosenfeld, R. Kunkel, N. Nagel, L.K. Verheij, G. Comsa, Surf. Sci. 272 (1992) 269. [15] D.J. Wentink, M. Kuijper, H. Wormeester, A. van Silfhout, Phys. Rev. B 56 (1997) 7679. [16] J.A. Venables, Philos. Mag. 27 (1973) 697. [17] G. Comsa, private communications. [18] F.A. Nichols, W.W. Mullins, Trans. Metall. Soc. AIME 233 (1965) 1840. [19] F.A. Nichols, W.W. Mullins, J. Appl. Phys. 36 (1965) 1826. [20] F.A. Nichols, J. Appl. Phys. 37 (1966) 2805. [21] M. Esser, K. Morgenstern, G. Rosenfeld, G. Comsa, Surf. Sci. 402 (1998) 341. [22] G. Ehrlich, F.G. Hudda, J. Chem. Phys. 44 (1966) 1039.
M. Esser et al. / Surface Science 552 (2004) 35–45 [23] R.L. Schwoebel, E.J. Shipsey, J. Appl. Phys. 37 (1966) 3682. [24] R. Kunkel, B. Poelsema, L.K. Verheij, G. Comsa, Phys. Rev. Lett. 65 (1990) 733. [25] G. Rosenfeld, R. Servaty, C. Teichert, B. Poelsema, G. Comsa, Phys. Rev. Lett. 71 (1993) 895.
45
[26] G. Rosenfeld, N.N. Lipkin, W. Wulfhekel, J. Kliewer, K. Morgenstern, B. Poelsema, G. Comsa, Appl. Phys. A 54 (1998) 772. [27] G. Rosenfeld, B. Poelsema, G. Comsa, J. Cryst. Growth 151 (1995) 230.
Kinetic growth manipulation of Si(0 0 1) homoepitaxy Marcus Esser, Erwin Zoethout, Harold J.W. Zandvliet, Herbert Wormeester *, Bene Poelsema Solid State Physics, Faculty of Applied Physics, MESA+ Research Institute, University of Twente, P.O. Box 217, Enschede 7500 AE, The Netherlands Received 3 March 2003; accepted for publication 6 January 2004
Abstract We have confirmed in a combined diffraction and STM study that the usual kinetic growth manipulation (KGM) applied to Si/Si(0 0 1) (nucleation at relatively low temperatures and completion of monolayer growth at high temperatures) does not lead to a smooth growth front. We have identified the physical reason for this unexpected behaviour: an anti phase boundary (APB) network develops during growth, due to the 2 · 1 reconstruction of the clean Si(0 0 1) surface. The density of this APB network can be substantially reduced by application of a different and optimised KGM procedure. Following a recipe in which 1 ML of Si on Si(0 0 1) is deposited at a relatively low temperature (525 K), followed by a short anneal to 750 K, results in a surface flatness similar to that of the clean Si(0 0 1) 2 · 1 surface. Up to 10 ML of material deposited with a flash anneal after the deposition of each additional layer resulted in a surface with a negligible reduction of the in-phase and out-of-phase intensity of a reflected low energy electron beam, indicative of an almost perfectly smooth growth front. STM images support this observation. The low thermal budget of this method reduces intermixing effects in hetero-epitaxial growth of group IV semiconductor (0 0 1)-faces. 2004 Elsevier B.V. All rights reserved. Keywords: Epitaxy; Growth; Silicon; Low energy electron diffraction (LEED); Scanning tunneling microscopy
1. Introduction Novel thin film structures synthesized by MBE or CVD on silicon have found a wide variety of applications in science and technology. Smooth and abrupt interfaces between epitaxial layers are often crucial factors for the deviceÕs performance. The smoothness and abruptness of the interface leads to a conflicting demand with respect to the
*
Corresponding author. Tel.: +31-53-489-3148; fax: +31-53489-1101. E-mail address: [email protected] (H. Wormeester).
growth conditions. For hetero-epitaxial growth the growth temperature should be sufficiently low to reduce atom diffusion and hence intermixing of the layers and dopant segregation, while on the other hand the growth temperature should be sufficiently high to avoid kinetic roughening. Potential growth recipes that focus on low sample temperature during deposition while still growing smooth layers can be developed with homoepitaxial growth. Many complicating factors associated with hetero-epitaxial growth like intermixing and lattice misfit can then be neglected. For Si and Ge surfaces, the reconstruction present on these surfaces plays a major role in the final surface
0039-6028/$ - see front matter 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.susc.2004.01.012
M. Esser et al. / Surface Science 552 (2004) 35–45
morphology. The key issue in this growth study is to obtain layer-by-layer growth with the smallest possible thermal budget for Si on Si(0 0 1). This implies that after every deposited monolayer a surface that is in every aspect comparable to the clean surface has to be achieved. The Si(0 0 1) surface is stabilized by a (2 · 1) surface reconstruction where the atoms of the outermost layer couple in pairs to form dimers in order to reduce the number of dangling bonds of an ideally bulk-truncated surface. When Si atoms are deposited on Si(0 0 1) surfaces at room temperature the atoms are sufficiently mobile to combine quickly to form dimers. The dimers can be divided in two classes: dimers residing on top of the substrate dimer rows and dimers residing in the troughs between the substrate dimer rows. At temperatures slightly above room temperature (300–350 K) the on-top dimers are able to diffuse parallel to the substrate dimer rows [1]. At even higher temperatures (450 K) also diffusion perpendicular to the substrate rows becomes active [2,3]. As Si islands nucleate and grow on (0 0 1) surfaces, the dimer rows in neighbouring islands have a probability of 50% to be in the correct registry. An anti-phase boundary (APB) will form if two islands meet, and their internal dimer rows are not aligned. For the Si(0 0 1) surface there are in principle two different APBs (see insert Fig. 1), one type of APB runs parallel to the substrate dimer rows (APBA ) and the other runs perpendicular to substrate dimer rows (APBB ). The latter APB acts as a preferential nucleation site for island growth [4]. The first type of APB gives, in combination with the fact that the islands are elongated parallel to the dimer row direction, rise to the formation of trenches and double layer steps [5]. The anti-phase boundary (APB)-network [6] is of key importance for the roughness of the grown layer. In this paper it will be shown that by modulating the temperature during growth the APB network can be manipulated in such a way that smooth layers can be grown at relatively low temperatures. Before demonstrating this with the homoepitaxial growth of 15 ML, the temperature dependence of the APB network will be discussed. The size of the APB network is determined from STM, while SPA-LEED is employed to study
0
10
T [K] 500
700
900
-1
T
10 island density [a-2 ]
36
400
300
T0
(1)
T
(2)
-2
10
T
(3)
L
-3
10
T
(4)
-4
10
B -5
10
-6
10
1.0
1.5
2.0 -3
1/T [10 K
2.5 -1
3.0
]
Fig. 1. Si island densities in units of the dimer spacing a on Si(0 0 1) vs. growth temperature. Up-triangle values are calculated from the APB network. Circle values have been published by Mo et al. [8,9]. The lines are guides to the eye. T0 is the growth temperature, Tð1Þ –Tð4Þ are the flash temperatures of the SPA-LEED measurement as discussed in the text. The insert visualizes the APB network as a result of non-aligned dimer rows. The definition of the spacing L and B between APBs along and perpendicular to the substrate dimer row direction is shown.
the annealing behaviour of the APB network in detail.
2. Experimental All experiments have been conducted in ultrahigh vacuum (base pressure <1010 mbar). Two systems were used, one equipped with highresolution low energy electron diffraction (HR-or SPA-LEED manufactured by Omicron), with a resolution better than 0.1% of a Brillouin Zone (BZ). The other system was equipped with an STM-1 (Omicron) that allowed measurements at room temperature. Fresh Si samples were outgassed by resistive heating for over 24 h up to a temperature of 600 C. Final preparation was done by repeated rapid flashing to 1100 C, while the pressure was kept below 1.0 · 1010 mbar. Surfaces prepared in this way showed a FWHM of 0.5% BZ
M. Esser et al. / Surface Science 552 (2004) 35–45
3. Conventional growth The investigation was started by looking at one of the key parameters in growth, the number density of islands as a function of temperature. The APB network opens a unique way to determine the maximum island number density, especially at high coverages of grown material [6,7]. A detailed description of this method developed by Zoethout et al. was published previously [7] and is only briefly mentioned here. The random nucleation of islands leads for epitaxy on the (2 · 1) reconstructed Si(0 0 1) surface to neighbouring islands that have a probability of 50% to be in the wrong registry, i.e. they can not merge together and an APB results. The island density can be extracted from the APB network that has formed after island coalescence. However, the density extracted from the APB network is four times too low as a result of the equal chance of being in or out of registry. The island density is thus N ¼ 4=ðhLi hBiÞ, where hLi (hBi) is the averaged spacing between APBs in a direction parallel (perpendicular) to the substrate row direction (see insert in Fig. 1). In Fig. 1 a plot of N vs. 1=T in the temperature range from RT to 1000 K is shown. The dotted line refers to the work of Mo et al. [8,9], who measured the island density after deposition of 0.07 ML. The island density depends on the coverage and will reach a maximum at the time just before coalescence sets in. It is exactly this maximum number that is extracted from the APB network. The kink in the curve at a temperature around 560–600 K has been attributed to an increase of the size of the critical nucleus [7]. Note that in a nearest neighbour model an increase of the stable nucleus from one dimer to two dimers coincides energetically with island coarsening. This additional process becomes active at higher temperatures, and is due to attachment and detachment at kink sites [10–12]. The temperature at which the kink occurs of the two curves differs by
about 40 K. This is attributed to the uncertainties in sample temperature measurement. The slope of the two curves in the temperature range below 550 K is identical and is related to the diffusion process of Si dimers parallel to the dimer rows [1,8]. For the real-time SPA-LEED measurements first the appropriate energies for the in-and out-of phase conditions of the central spot have to be determined. Low energy electrons are favoured for real-time measurements as these have a higher surface sensitivity and the decrease of intensity due to the Debye–Waller effect for elevated sample temperature is smaller for low electron energies. This implies that energies with a perpendicular relative phase number of Sz ¼ 2 are desired. However, low intensity prohibits for example measurement at the real out-of-phase condition. Numerous test experiments using electron energies between 36 eV and 324 eV for the spot intensity variations during growth were performed. At an energy of 54.8 eV the best out of phase oscillations (see below) were observed, while at 92.3 eV the spot intensity remained at the highest level with the smallest fluctuation, indicative of in-phase condition. Fig. 2 shows the spot intensities measured at the chosen energies in real time during growth of Si on
1.0 normalized peak intensity []
of the specular reflected electron beam with SPALEED, while the missing dimer defect density as observed with STM is low. Si was deposited from a resistively heated Si bar.
37
0.8
0.6
0.4
0.2
0.0 0.0
0.5
1.0
1.5
2.0
time [h] Fig. 2. Real-time measurement of central spotÕs peak intensities (normalized, circle symbols: at 54.8 eV, out-of-phase condition; rhomb symbols: 93.2 eV, in-phase condition) during conventional growth of Si on Si(0 0 1) at a sample temperature 575 K.
38
M. Esser et al. / Surface Science 552 (2004) 35–45
Si(0 0 1) at a constant temperature of about 575 K. The observed oscillations are indicative of ‘‘layerby-layer’’ growth. The maxima coincide roughly with the completion of a monolayer. The low intensity of the maxima reveal that the ‘‘layer-bylayer’’ growth is by no means close to the ideal growth. We refer to this situation as ‘‘conventional’’ growth. The out-of phase intensity oscillates with time, while keeping the incident flux constant at approx. 0.05 ML/min. Due to the progressing roughening of the surface the amplitude of the oscillation is decaying from about 27% at the first two intensity maxima to 15% after about 5 grown ML. The in-phase intensity measured simultaneously shows weak oscillations, the oscillation maxima are reached after the maxima of the out-of-phase intensity oscillation. In the case of homoepitaxy on Si(1 1 1) the in-phase oscillation maxima were reached just before the out-of-phase oscillation maxima which can only be explained by a breakdown of the kinematic theory for the in-phase signal [13]. Similar effects have been measured for the growth of Pt/Pt(1 1 1) [14]. The in-phase intensity loss is due to diffuse scattering. The out-of-phase signal is in first approximation not affected. This experiment demonstrates that during the growth of Si on Si(0 0 1) growth oscillations can well be measured, but, in contrast to the homoepitaxy on Si(1 1 1), the growth is accompanied by a roughening growth front. In the next section, a novel, modified growth manipulation method will be presented that prevents this roughening.
4. Kinetic growth manipulation 4.1. Concept of two mobilities Kinetic growth manipulation (KGM) methods have been successfully used on unreconstructed metal surfaces to enhance the layer-by-layer growth. In general these methods are based on the concept of two mobilities, or better, the concept of two island densities and is extensively described and applied on unreconstructed metal surfaces by Rosenfeld et al. [25–27]. KGM aims on these surfaces specifically to reduce the consequence of
the additional energy barrier at the island edges (Ehrlich Schwoebel barrier) for downward diffusion [22–24] in order to enhance the interlayer mass transport. The concept of two mobilities consists of growing the first part of every ML at a lower temperature and afterwards continuing the growth for the rest of the ML at a higher temperature. The idea behind this is the creation of a high density of very small islands in the first phase of every MLÕs growth. This is done to increase the probability of atoms that land on top of those islands to diffuse downwards by increasing the number of attempts to descend when the temperature is higher in the later stage of every monolayerÕs growth overcoming the additional energy barrier at the island edges for downward diffusion. The more adatoms overcome the additional energy barrier and diffuse downwards, the lower is the chance for island nucleation on top of the islands. Fig. 3 shows the intensities of the central SPALEED diffraction spot at in-phase and out-ofphase conditions measured in real-time during the growth of Si on Si(0 0 1) using the concept of two mobilities. Seven monolayer of Si have been deposited; during the first half of every MLÕs deposition the sample temperature was 375 K, during the second half 575 K. The strange-looking scaling in Fig. 3 is due to the fact that at the very beginning of the experiment the sample is at the lower temperature of 375 K. Because of the Debye–Waller effect, the spot intensities are about twice as high as at 575 K. Compared to the conventional case (see Fig. 2) a slightly enhanced outof phase oscillation amplitude is recognizable (40% compared to 30% in the unmanipulated case). But what is more important is the fact that the decay of the out-of phase oscillation amplitude that was observed in the unmanipulated case is still present. This shows that the concept of two mobilities as a kinetic growth manipulation method is not able to suppress the progressive roughening of the surface in the case of homoepitaxy of the reconstructed Si(0 0 1). The reason for this is that by this approach it is not possible to overcome the problem of the building-up of a dense APB network and the corresponding preferred island nucleation in higher layers.
M. Esser et al. / Surface Science 552 (2004) 35–45
1.2
2.0
T(1) T(2)
100% @ 375 K 1.6 1.4 1.2 1.0
100% @ 575 K
0.8 0.6
normalized spot intensities []
575 K 375 K
1.8 normalized peak intensity []
39
T(3)
T(4)
1.0 0.8 0.6 0.4 0.2
0.4 0.2
0.0 0
0.0 0.00
0.25
0.50 0.75 time [h]
1.00
1.25
Fig. 3. Real-time measurement of central diffraction spot intensities during Si growth on Si(0 0 1) using the concept of two mobilities as kinetic growth manipulation method (circle symbols: 54.8 eV out-of-phase condition; rhombic symbols: 92.3 eV in-phase condition; R 1 ML/12 min). The sample temperature is 375 K in the first half of the deposition of every ML (till the minima of the out-of-phase intensity oscillations) and 575 K in the second half (till the maxima of the out-of-phase intensity oscillations). The temperature changes are marked with vertical arrows, the shorter indicate a change to 575 K while the longer indicate a change to 375 K. The scaling of the spot intensities differs from the figures before; this is due to the fact that at the very beginning of the experiment the sample is at the lower temperature of 375 K. Because of the Debye–Waller effect, at 375 K the spot intensities are about twice as high as at 575 K.
4.2. Smoothening a Si monolayer Fig. 4 shows the intensities of the central SPALEED diffraction spot at in-phase and out-ofphase conditions measured during and after the evaporation of Si (rate0.067 ML/min) on the Si(0 0 1) surface at a temperature of 525 K. At the end of the deposition, the in-phase intensity is at about 50%, the out-of-phase intensity at about 15% of it is initial value corresponding with the build-up of a rough growth front. After deposition of 1 ML of silicon the sample was flash annealed to 650 K (1), 700 K (2), 750 K (3) and 775 K (4), 1 min per flash. Already type (1) flashing is sufficient to get the in-phase intensity back to its initial value indicating a surface with just a few defects, such as
15
30 45 time [min]
60
75
90
Fig. 4. Real-time measurement of central diffraction spot intensities during Si growth on Si(0 0 1) and the following flash annealing of the grown film (circle symbols: 54.8 eV out-ofphase condition; rhomb symbols: 92.3 eV in-phase condition; R 0:067 ML/min; T 525 K). 1 ML of Si was grown and afterwards gradually flashed to 650 K (1), 700 K (2), 750 K (3) and 775 K (4) (duration of each flash: 1 min). After the flashes the sample is quickly cooled down to 525 K. In this periods the measured spot intensities rise because of the Debye–Waller effect.
point defects created during ion-sputtering. Optical anisotropy measurements of the Arþ ion bombarded surface revealed the disappearance of such defects at 670 K [15]. The out-of-phase intensity increases step by step (66% of the initial value for flash type (1), 87% (2), 97% (3), 100% (4)) with the flashes. Fig. 5(a) and (b) visualize the effect of flashing using RT STM scans of the Si(0 0 1) surface. Scan (a) was taken after evaporating a little more than 1 ML at 500 K. It is obvious that nucleation took place in the second layer before the first layer was completed. Scan (b) shows the surface with 1 ML grown at 500 K plus a 1 min flash to 660 K. Compared with scan (a) less material remains in layer 2, whereas layer 1 is filled to a higher degree. Note that scan (a) was taken at a 10 times lower deposition flux, comparable to the deposition flux used in the SPA-LEED measurements. According to Venables nucleation theory [16], this implies a 2 times larger island density. As a result, the crossover shown in Fig. 1 will occur at a temperature of about 20 K lower. This implies
40
M. Esser et al. / Surface Science 552 (2004) 35–45
Fig. 5. STM scans on Si(0 0 1) after Si evaporation (quenched to RT), RðaÞ 0:059 ML/min, RðbÞ–ðdÞ 0:59 ML/min. (a) 1 ML grown at 500 K; (b) 1 ML grown at 500 K plus flash to 660 K, 1 min; (c) 4.5 ML grown at 660 K and (d) 4.5 ML grown at 500 K applying KGM, i.e. a 1 min flash to 660 K after every completed ML. Tunnelling parameters: )2.0 V, 0.5 nA, scan width: 250 A.
that the results obtained in the STM and SPALEED are comparable. The procedure depicted in scan (b) thus shows a situation in which the island density was initially even higher than the one shown in scan (a). Scan (b) also shows that the evaporated material is distributed over two layers only. Within the kinematic scattering theory, the out-of-phase intensity Ioutofphase can be expressed as a competition of the exposed layers. For a three layer system this reduces to: Ioutofphase ¼ ½ðh0 h1 Þ ðh1 h2 Þ þ h2 I0
ð1Þ
At 1 ML coverage and only three layers involved, all the material in layer 2, h2 has to be equal to the
non-filled area of layer 1, i.e. h1 ¼ 1 h2 . The coverages h1 , h2 can than be calculated from the out-of-phase intensities: 0 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1 1 ML Ioutofphase 1@ A 1 ð2Þ h2 ¼ 4 I0 (assuming that at the out-of-phase intensity maximum a complete ML has been evaporated, see Table 1 for the results). This flash series shows that a short thermal treatment to a 200 K higher temperature results for this system in an almost perfectly closed layer. If one would try to reach a similar surface quality by growing 1 ML unmanipulated at a certain constant temperature, this
M. Esser et al. / Surface Science 552 (2004) 35–45
41
Table 1 Calculated coverages and filling factors after growth of 1 ML at 525 K and following flash types (1)–(4), for experimental data see Figs. 1 and 4 Type
Tflash /K
Ioutofphase =I0
h1
h2
fAPB
1 ML (1) (2) (3) (4)
– 650 700 750 775
15% 66% 87% 97% 100%
85% 95% 98% 99.5% 100%
15% 5% 2% 0.5% 0%
– 2.5% 2% – <0.5%
temperature would be much higher due to the formation of double steps. The reason for the development of a rough growth front is that the APB network hinders a complete filling of the first layer. This causes nucleation on top of the islands and in this way induces roughening. In the hBi-direction islands (see Fig. 1) cannot grow together and a gap of one lattice unit will remain between the islands. In contrast, an APBB does not hinder two islands merging together (see insert Fig. 1). The fraction of the ML that cannot be contained by the layer itself is defined as fAPB . fAPB can be calculated from the density of the APB network determined with STM (Fig. 1) fAPB ðT Þ ¼
1 1 þ hBiðT Þ
ð3Þ
Table 1 shows a good correspondence between the measured size of the APB network and thus the filling factor fAPB and the amount of material that cannot be incorporated in the first layer (h2 ) after annealing as determined from SPA-LEED. The situation shown in scan (b) of Fig. 5 is not as ideal as expected from the SPA-LEED measurements. This is probably due to the fact that in the STM UHV system there is no real time probe to control the deposition and thus exactly deposit 1 ML. This makes it very hard to produce a completely filled first layer. The error in the amount of deposited material after deposition of 1 ML is estimated as <1% in case of the SPA-LEED measurements and 10% in case of the STM measurements. 4.3. Layer-by-layer The experiences just described are now applied to grow thicker films in a smooth way. After the completion of each ML grown at a temperature
below the coarsening temperature the layer is first annealed to create again a smooth surface with a low defect density. Fig. 6 shows the deposition of 10 ML in this way. As soon as an oscillation maximum of the out-of-phase intensity is reached, the sample is flashed (flash types (1)–(4), see above) for 1 min. Due to the use of resistive heating, the spot intensity measurements are interrupted during flashing, whereas the silicon evaporation continues. Right after the flashes out-of-phase intensities of 50% (1), 60% (2), 75% (3) and 75% (4) of the initial values can be measured, in-phase intensities are in the range of 75–90% (1–4) of their initial values. The out-of-phase-intensities are lower than the ones measured after flashing 1 ML. This has two reasons: (1) the deposition is not interrupted, (2) it takes some time till the sample has cooled down from the flash temperature. Due to the Debye–Waller effect the spot intensities are lower at higher temperatures. Note that the increase of the spot intensities during cooling down the sample is dominated by Debye–Waller effects. Effects due to further annealing are negligible. For the last oscillation the intensities were extrapolated to obtain a value of the in-phase and out-of-phase intensity just after flashing corrected for the difference in Debye–Waller factor for the base and flash temperature. The base temperature, defined as the low temperature at which the first part of each monolayer is grown, amounts to 525 K. Fig. 6 indicates that both the in-phase and the out-of-phase signal returned to the initial levels indicating that the 10th layer has been grown very smooth indeed. We found that already the evaporating-flashing cycle can be repeated at least 15 times without noticeable decay of the spot intensities. The same growth experiments have also been performed at lower base sample temperatures (down to 425 K). The
42
M. Esser et al. / Surface Science 552 (2004) 35–45
scan (d) growth took place at 500 K with 1 min flashes to 660 K after every completed ML.
1.0
normalized peak intensites []
0.8 0.6
5. Discussion
0.4
The growth manipulation method for homoepitaxy on Si(0 0 1) that we describe in this paper has a distinctly positive effect on the smoothness of the grown layer. The oscillation amplitude of the SPA-LEED out-of-phase central spot intensity rises from 10% of the initial value to 50–75% of its initial value (see Fig. 6), where the latter value is even underestimated due to the Debye–Waller effect. That means that the growing Si layers are filled up to a much higher fraction before nucleation in a new layer takes place. Furthermore the growth manipulation results in an absence of the oscillation amplitude decay indicating that there is no progressive kinetic roughening during growth. By evaluating the STM pictures after the growth of 4.5 ML Si with and without applying the growth manipulation method we can make an overview of the distribution of the Si over the visible layers (see Fig. 7). It is already remarkable that the layer deposited at 660 K is less smooth than the layer deposited at 500 K that was flash annealed to 660 K after completion of every ML. The 4.5 ML have been deposited for the conventional growth over 5 exposed layers. Markedly small is the amount of deposit in the 7th and 8th layer in contrast to the only 3 exposed layers for the manipulated growth with only a small amount deposited in the 6th layer. The not completely filled 4th layer is probably due to the inaccuracy of the flash after 1 ML. The surface roughness value can be quantified with a parameter g, defined as [17] P grown 2 ðhn hideal2D Þ n g ¼ P n ideal3D ð4Þ hideal2D Þ2 n n ðhn
0.2 0.0 1.0
1 1 1 1 2 2 3 3 4 4
0.8 0.6 0.4 0.2 0.0
0 30 60 90 120150 180 210 time [min]
Fig. 6. Real-time measurement of central diffraction spot intensities during Si growth on Si(0 0 1) using kinetic growth manipulation (KGM; lower panel: 54.8 eV out-of-phase condition; upper panel: 92.3 eV in-phase condition; R 1 ML/15 min; T 525 K). 10 ML of Si were grown applying flash annealing to 650 K (1), 700 K (2), 750 K (3) and 775 K (4) (duration of each flash: 1 min) after every completed ML. During the flash annealing, the evaporation continues. After the flashes the sample quickly cools down to 525 K. For the last oscillation an extrapolation of the intensity to the time of the flash is made.
flashing after completion of every ML results in the same spot intensity amplitudes. It is a remarkable result for the homoepitaxy on Si(0 0 1) that especially at this low base growth temperatures even after the growth of 10 ML the top layer is as smooth as found here. The STM scans in Fig. 5(c) and (d) show the effect of the applied flashing growth manipulation. As it is hard to visualize thick rough layers with STM the grown layers are only 4.5 ML thick. In scan (c) the material is grown at 660 K (growth at 525 K resulted in a rough growth front that could not be imaged properly), in
In case of ideal 2D growth, i.e. the type of growth where nucleation in layer n þ 1 starts only if layer n is filled up completely, g equals to 0, whereas g ¼ 1 for ideal 3D growth (Poisson distribution). From our conventional growth data g ¼ 0:73 is
M. Esser et al. / Surface Science 552 (2004) 35–45
1.0
controlled growth conventional ideal 2D growth poisson growth
coverage
0.8
0.6
0.4
0.2
0.0
0
1
2
3 4
5 6 layer
7
8
9 10 11
Fig. 7. Comparison of the grown films (thickness 4.5 ML) after conventional growth at 660 K and manipulated homoepitaxy of Si(0 0 1) at a combination of 500 and 660 K: The distribution of material into different atomic layers is plotted. Layer 0 is the first bulk layer. For further comparison, the distributions for ideal 2D growth (dashed line) 3D growth (solid line) of 4.5 ML are plotted into the graph.
obtained, the controlled growth data lead to g ¼ 0:31. Even if the results from our STM measurements do not completely meet the results in terms of surface quality after growth from the SPA-LEED measurements, the g values express the positive effect of the growth manipulation method unequivocally. The kinetic growth manipulation procedure developed in this paper is markedly different from the procedure originally developed for unreconstructed metal surfaces. The latter method is based on the concept of two island densities which assumes that islands created in the initial phase can fully grow together without domain boundaries. The reconstruction of the surface as occurs for Si and Ge (0 0 1) leads to the formation of an APB network that exactly prevents this. The only way to obtain a surface after the deposition of 1 ML that resembles the starting surface, i.e. has a low APB network density, is through a substantial annealing. The complete growth at an elevated temperature is not as successfull. The islands are too large to be able to reduce the APB network density in a reasonable time [18–21]. The APB
43
network density has to be quite small as enhanced nucleation takes place at APBÕs [4]. Increasing the length scale by deposition at more elevated temperatures is not a feasible approach. Growth at increased temperatures leads to the formation of double steps which strongly reduce interlayer transport, an essential ingredient for smooth growth. Layer-by-layer growth is thus only possible at temperatures at which hetero-epitaxy of for instance Si and Ge would lead to strong intermixing. The deposition of 1 ML at a relatively low base temperature followed by a short annealing is capable of producing quite smooth layers. During the flash, the material contained in the small islands quickly redistributes towards neighbouring islands. This ripening is rather fast as a result of the small size of the islands and causes a drastical decrease of the APB network density. The temperature flash process does not lead to only a fill up of the holes in a layer with the material on top of this layer, a process that essentially does not change the density of the APB network. The inphase intensity of the relected electron beam returns after the flash to the value observed for the clean surface, indicating that the same number of defects, like steps etc, are present at the surface after the flash. This implies that the density of the APB network has to be reduced during the flash. We have found that with a base temperature for deposition of 425 K still layer-by-layer growth can be observed. At lower base temperatures the smoothness of the initial layer can not be replicated. This is probably due to the absence of diffusion of dimers perpendicular to dimer rows. The essentially 1D diffusion parallel to the dimer rows, at lower temperatures apparently induces to many APBÕs to be successfully flash annealed with a reasonable thermal budget. A short anneal to 750 K is able to smoothen a layer grown at a reduced temperature and decrease the APB network density considerably. The thermal budget of the layer, i.e. the time at which the sample was at the highest temperature is in this way at least 15–20 times lower than compared to conventional growth while up to 15 ML no decay of the in-phase and out-of-phase intensity can be observed.
44
M. Esser et al. / Surface Science 552 (2004) 35–45
The setting of the base temperature at which the majority of the material is deposited and the flash temperature at which the smooth surface is obtained can be derived from the temperature dependent island density as depicted in Fig. 1. The kink in this curve at about 580 K discriminates the useful range for both the base and flash temperature. A base temperatures at which the majority of the material is deposited above 580 K leads to much larger islands and this will require a larger thermal budget, i.e. an increased anneal time for ripening that would lead to a similar length scale of the APB network as obtained at lower temperatures. This graph also indicates a temperature of at least 580 K for the anneal. The island size also depends very strongly on the temperature and a small increase in temperature can lead to a large increase in island size. It is also important to note that for 1 ML of material double step formation is avoided. A flash temperature that reduces the number of defects on which preferential nucleation takes place to the order of the density of defects on the clean surface, including steps can be considered as the upper limit for the flash temperature in the application of this kinetic growth manipulation method.
6. Conclusions A kinetic growth manipulation (KGM) method that aims at reducing the density of the anti-phase boundary (APB) network emerging during growth experiments on the reconstructed Si(0 0 1) surface has been developed. It has been demonstrated that this KGM method can be used successfully to prevent progressive roughening of the growth front during homoepitaxy of Si(0 0 1). As a consequence, films with a thickness of many ML can be grown in a flat and smooth manner. Furthermore it has been shown, that growth manipulation methods like the concept of two mobilities that work successfully against the additional energy barrier at the island edges for downward diffusion in epitaxy of unreconstructed metal surfaces are not successful in preventing the progressive
roughening of the growth front in the case of the reconstructed Si(0 0 1) surface. The described method is probably general applicable to epitaxial growth of semiconductor and metal layers which surfaces reconstruct with at least two different domains. The reduced thermal budget is especially important if intermixing of substrate and layer has to be avoided.
References [1] B.S. Swartzentruber, Phys. Rev. Lett. 76 (1996) 459. [2] B. Borovsky, M. Krueger, E. Ganz, Phys. Rev. Lett. 78 (1997) 4229. [3] M. Krueger, B. Borovsky, E. Ganz, Surf. Sci. 385 (1997) 146. [4] U.K. Hamers, R.J. K€ ohler, J.E. Demuth, J. Vac. Sci. Technol. A 8 (1990) 195. [5] W. Wulfhekel, PhD thesis, University of Twente, 1997. ISSN 0944–2952. [6] H.J.W. Zandvliet, E. Zoethout, W. Wulfhekel, B. Poelsema, Surf. Sci. 482 (2001) 391. [7] E. Zoethout, P.W. van den Hoogenhof, H.J.W. Zandvliet, B. Poelsema, J. Appl. Phys. 92 (2002) 5785. [8] Y.W. Mo, J. Kleiner, M.B. Webb, M.G. Lagally, Phys. Rev. Lett. 66 (1991) 1998. [9] A. Pimpinelli, J. Villain, D.E. Wolf, Phys. Rev. Lett. 69 (1992) 985. [10] H.J.W. Zandvliet, H.B. Elswijk, E.J. Loenen, Surf. Sci. 272 (1992) 264. [11] N. Kitamura, B.S. Swartzentruber, M.G. Lagally, M.E. Webb, Phys. Rev. B 48 (1993) 5704. [12] C. Pearson, B. Borovsky, M. Krueger, R. Curtis, E. Ganz, Phys. Rev. Lett. 74 (1995) 2710. [13] M. Esser, H. Wormeester, B. Poelsema, Surf. Sci. 482 (2001) 1355. [14] B. Poelsema, A.F. Becker, G. Rosenfeld, R. Kunkel, N. Nagel, L.K. Verheij, G. Comsa, Surf. Sci. 272 (1992) 269. [15] D.J. Wentink, M. Kuijper, H. Wormeester, A. van Silfhout, Phys. Rev. B 56 (1997) 7679. [16] J.A. Venables, Philos. Mag. 27 (1973) 697. [17] G. Comsa, private communications. [18] F.A. Nichols, W.W. Mullins, Trans. Metall. Soc. AIME 233 (1965) 1840. [19] F.A. Nichols, W.W. Mullins, J. Appl. Phys. 36 (1965) 1826. [20] F.A. Nichols, J. Appl. Phys. 37 (1966) 2805. [21] M. Esser, K. Morgenstern, G. Rosenfeld, G. Comsa, Surf. Sci. 402 (1998) 341. [22] G. Ehrlich, F.G. Hudda, J. Chem. Phys. 44 (1966) 1039.
M. Esser et al. / Surface Science 552 (2004) 35–45 [23] R.L. Schwoebel, E.J. Shipsey, J. Appl. Phys. 37 (1966) 3682. [24] R. Kunkel, B. Poelsema, L.K. Verheij, G. Comsa, Phys. Rev. Lett. 65 (1990) 733. [25] G. Rosenfeld, R. Servaty, C. Teichert, B. Poelsema, G. Comsa, Phys. Rev. Lett. 71 (1993) 895.
45
[26] G. Rosenfeld, N.N. Lipkin, W. Wulfhekel, J. Kliewer, K. Morgenstern, B. Poelsema, G. Comsa, Appl. Phys. A 54 (1998) 772. [27] G. Rosenfeld, B. Poelsema, G. Comsa, J. Cryst. Growth 151 (1995) 230.