Chapter 4

Mg/Si{111}

Abstract

The deposition of Mg on a Si{111} surface failed to produce a Si{111}√3x√3-30-Mg surface structure. Various Mg deposition-rates, substrate temperatures, Mg coverages, and anneal practices of the Mg/Si{111} system can, however, produce 1x1, 3x1, and Si{111}⅔√3x⅔√3-30-Mg surfaces. These results are in contrast to an earlier surface-extended-x-ray-absorption-fine-structure (SEXAFS) study by Vandr'e et.al.[77], that reports the formation and determination of two different Si{111}√3x√3-30-Mg structures, as well as the formation of a 3x3 structure. Given the different results of the present study and of the SEXAFS study, a summary of the SEXAFS study, of the questions raised, of the details of Stony Brook experiment, and a discussion are presented.

4.1 Introduction

Vandre et.al.[77] have recently reported a Si{111}√3x√3-30-Mg surface and analyzed it using SEXAFS. Their results are summarized here. A Si{111} wafer (n-type) was annealed in UHV at 1100 C to produce a clean Si{111}7x7 surface. Ion bombardment and AES are not mentioned as having been used. Mg was then thermally deposited on a room-temperature Si{111}7x7 surface. Increased Mg coverages caused the LEED pattern to change from (7x7) to (1x1), then to √3, and finally to extinguishment. Mg coverage was assumed to be ⅓ ml at the lowest coverage that produced a good √3 LEED pattern. Larger coverages were then determined from the size of the Mg-K edge in the X-ray absorption spectra. Anneal (300 C, 30 sec) of ⅓ ml and of higher coverages of Mg/Si{111} produced a 3x1 surface structure followed by a ten-fold decrease in the intensity of the Mg-K edge. Lower anneal temperature caused no surface-structural changes. Anneal at 1100 C caused the desorption of Mg and the restoration of the Si{111}7x7 structure.

The SEXAFS analysis concluded that ¹/₆ ml of Mg formed a T₄-√3 phase on a relaxed Si{111} surface, see Figure 1.7. The distance between the Mg atom and first(second)-layer Si atoms was 2.47(2.60)±0.04 Å. The first-layer Si atoms relaxed outwardly from the Si bulk planes by 0.2 Å and relaxed radially inward toward the Mg atoms by 0.08 Å. Higher coverages, ½ ml, produced a new √3x√3-30 phase which included both Mg atoms in the occupied T₄-sites and Mg atoms on off-centered T₁-sites between T₄-sites, see Figure 4.1. These new sites were determined to be interstitial in nature and only partially occupied. Full occupation of both sites required ⁴/₃ ml of Mg. The nearest-neighbor (n.n.) distance for Mg atoms in this phase was 3.3 Å, which was within 3% of the 3.2 Å n.n. distance of Mg atoms in hcp Mg. So this phase was remarkably similar to the basal plane of Mg. However, at 1-ml the long-range order in the surface disappeared, and the SEXAFS spectra were the same as those of bulk Mg.

The SEXAFS study raises many interesting points and questions, which invite a study of the same structures by quantitative LEED and SXD analyses:

1. If the ⅓ ml Si-√3-Mg surface is a T₄-structure, then one would expect the LEED I(V) spectra of Si-√3-Al and -Mg to be practically identical. This would allow for a quick confirmation of the T₄-structure.
2. How deep are the surface relaxations of the Si{111} atoms in the two reported √3 phases?
3. How does the Si{111}7x7 surface remove its stacking fault to form the Mg-√3 at room temperature?
4. Why does a group-II atom form a √3 structure on the Si{111}7x7 surface, especially at room temperature?
5. If Mg forms a √3 on Si, then what about Mg/Ge{111}, Be/Si{111}, etc....?
6. The proposed ⅓ ml T₄-√3x√3-30 phase is known to exist for Al, Ga, In, Pb, and Sn on Si{111} and Pb/Ge{111}, but the higher-coverage √3x√3-30 phase is quite new and different from other previously proposed high-coverage √3 structures. For example, see Figures 4.2, 4.3, 4.4, 4.5, and 4.6, trimer-based and close-packed models have been proposed or determined for the Si or Ge{111}-√3-Pb, -Au, -Bi, and -Sb surfaces, see Chapter 5 and Appendices B and C.
7. The reported 3x1 phase is also new; what is its structure?
8. Can hcp Mg grow epitaxially on Si{111}, given suitable deposition rates, temperature, etc....?
9. What is the relation between the √3 phase and the stable Mg silicide (SiMg₂)?

4.2 Experiment

The Stony Brook experiment [78] was conducted on two Si{111} samples. Both samples, nominally 10x20x0.2-mm, were cleaved from a larger phosphor-doped (8-10 Ohm•cm) wafer. The first sample was degreased in acetone and methanol, then dipped in HF to form a passive oxide surface. The second sample was treated with the Shiraki method [39]; both samples yielded the same results. The deposition of magnesium was also accomplished in two ways. The first Mg deposition-technique used 1x1x1-mm Mg pieces loaded into a quartz tube (4 mm O.D., 2 mm I.D., 50 mm Length), which was spirally wrapped with a Ta ribbon (0.025x2x250 mm). The tube was mounted on a flange provided with a shutter, electrical connections, and the ability to direct the Mg toward the Si sample. The tube was inclined 20 deg from the horizontal plane, with its opening 10-cm from the sample face, and was resistively heated by passing a current through the Ta ribbon. The temperature was monitored with an infra-red radiometer aimed at the Ta ribbon because it proved difficult to monitor the temperature of the Mg inside the quartz tube. The second deposition technique used Mg platelets (4x10x0.5 mm) mounted on resistively heated Ta foil-strips. It should be noted that Mg is easily evaporated at quite low temperatures: at 250 C the vapor pressure is approximately 10^-6 Torr. At such temperatures, heating of the sample by radiation from the Mg source was negligible. A third deposition technique using Mg in a W spiral basket was attempted and failed. Unfortunately, the Mg did not wet the W; this produced very unstable bursts of source material. The previous two techniques were, however, quite stable and allowed for reproducible depositions.

After bakeout and outgassing, the vacuum was at or below 2•10^-10 Torr, except during cleaning or deposition. After an initial Ar-ion bombardment (2 uA, 375 eV, 5•10^-5 Torr, 120 min., 30 C) and anneal (1000-1100 C, 15 min.), a Si{111}7x7 LEED pattern was obtained. Further cycles using shorter times produced cleaner and well ordered Si{111}7x7 patterns.

Sample cleanliness was monitored with AES and was typically very good as suggested by the ratios R_O < 0.0003 and R_C < 0.0015, see Table 2.1. These ratios suggested surface contamination levels of much less than 1 at.% for oxygen and less than 1 at.% for carbon [37]. No other impurities were detected.

Magnesium coverage was monitored by following the increase of the R_Mg ratio, see Equation 2.6 discussed in Chapter 2, where R_∞=[I^∞_Mg(45)]/[I^∞_Si(92)] = 0.87 and lambda_45=3.6 Å were assumed [37,38]. Note that the 45-eV Mg AES-line also coincided with a weak Si AES-line. Thus, the lowest value of the ratio R_{Mg} was 0.05. Additionally, the thickness of one layer of Mg was assumed to be equal to the n.n. distance (3.2 Å) in the hcp-Mg phase. This was based on the assumption of a close-packed Mg layer and the lack of Mg diffusion into the Si substrate. Such assumptions were most likely not correct.

We give now a summary of experimental results, which can be broken into three groups: normal, slow, and hot deposition. In the first instance, Mg was deposited at a rate of 0.35 to 1.35 Å/min. (0.11 to 0.42 LE/min.) onto a room-temperature Si{111}7x7 surface. For Mg thicknesses reaching 1.35 Å (0.42 LE, R_Mg=0.05-0.50) the 7x7 LEED pattern was gradually obscured. With increased Mg thicknesses up to 3.2 Å (1.0 LE, R_Mg=1.30) only the 1x1 pattern was visible over a very high background. This pattern sometimes sharpened and reverted to a 3x1 pattern in the course of several hours or days. Above the aforementioned coverage, the 1x1 pattern disappeared into the background showing that there was a loss of long-range order on the sample surface.

In the second case, slower deposition was made on a room-temperature substrate at a rate of less than 0.1 Å/min. (0.03 LE/min.) or on a 100-200 C substrate at the normal rate. Again, the 7x7 pattern faded away in the range of low Mg coverage, but intermediate Mg coverages caused the appearance of a mixture of 1x1 and ⅔√3x⅔√3-30 LEED patterns. Only the ⅔√3 pattern was often visible, usually the background was quite high, the beams were diffuse, and I(V) spectra could not be collected. The ⅔√3 patterns were visible for Mg thicknesses up to 4.3 Å (1.34 LE, R_Mg=2.0) and sometimes reverted to 1x1 patterns in the course of several hours of days. Thicker Mg films produced no LEED patterns. It should be noted that a ⅔√3 diffraction pattern does not include all of the integral-order beams of a 1x1 pattern.

The third case involved anneal after deposition. Heating 1x1 surfaces (300-400 C, 1 min.) produced gradual improvements in the 1x1 patterns. Longer anneals produced visible fractional-order streaks between the integral-order beams. These streaks would eventually sharpen up and produce well-ordered 3x1 LEED patterns. Anneal for up to 1 hour would not change the structure and only lead to improvements of the surface order. The magnesium coverage of this structure was typically 0.5 Å (0.16 LE), though it ranged from 0.35 Å to 1.35 Å. Anneal to higher temperatures, greater than 600 C, restored the 7x7 pattern in all three cases mentioned above, although the ratio R_Mg did not always return to 0.05 after high temperature heating (1200 C). Therefore, Ar-ion bombardment followed by anneal was often necessary to obtain clean Si{111}7x7 surfaces and to remove all magnesium from the surface region.

4.3 Discussion

4.3.1 Si{111}⅔√3x⅔√3-30-Mg

The failure to produce a Si{111}√3x√3-30-Mg surface is disappointing and puzzling, while the production of a ⅔√3 surface is even more puzzling. This raises several questions: a) was the structure misidentified by Vandr'e et.al. (or by us)? b) what is the ⅔√3-Mg? c) what can causes this lack of reproducibility between two research groups?

In the present research the ⅔√3 LEED pattern was, admittedly, initially misidentified as a √3. A second look informed us, however, that this was clearly not a √3 pattern, see Figure 4.7. We did in fact see fractional-order beams rotated 30 deg from the k_x axis. However, these beams were not those of a √3 pattern. If Vandr'e et.al. had a ⅔√3 pattern, then it is doubtful that they consistently misidentified the pattern as a √3. The two patterns are dissimilar and the unit-cell sizes are quite different. Additionally, the ⅔√3 surface cannot be reconciled with the near hcp-Mg structure determined by Vandr'e et.al..

Without LEED intensity data we cannot completely solve this problem. However, we note that the ⅔√3 phase has a surface unit-cell length of 4.43 Å, which we confirmed using LEED beam-profile measurements. SiMg₂, a silicide, has the CaF₂ structure with a lattice parameter of 6.39 Å; the surface unit-cell length on a {111} plane is 4.52 Å. This is merely 2% different from the ⅔√3 unit-mesh size. It is therefore likely that Mg is forming a slightly strained silicide on the Si{111} surface. Given the fact that the ⅔√3 phase is sometimes mixed with the 1x1 phase, a thin commensurate silicide is formed, not an ordered overlayer. We have tried to think of possible models for an ordered ⅔√3 phase adsorbate-superstructure. In Figure 4.7 one can see that the ⅔√3 phase unit-cell repeats with its corners on three different adsorption-sites (H₃, T₄, T₁) in this example or at least on three inequivalent sites in any other example. This fact makes it highly unlikely that a conventional ordered overlayer was formed; a silicide is more probable. This interpretation obviously requires further study.

Sample cleaning or preparation could play a role. For this reason, the experiment was done twice. Two different Mg deposition-techniques and two different Si samples were used; the results were the same. The first time, simple HF etching was used to form a passive oxide overlayer on the Si sample. The second time, the more complicated Shiraki cleaning method was used. Both times, the samples still required additional in-situ cleaning to remove layers of C and O from the Si sample. Both samples were cleaved from the same type of wafer, but it could be that the factory processing or dopant type had an effect. However, a similar wafer was used for the Au experiment, and no trouble occurred in forming √3-Au surfaces. This does not rule out the fact that Si-√3-Mg formation could be more sensitive. We do not know, however, of any reports stating that low-concentration phosphor-doping hinders the formation of a surface structure. Alternatively, it could be that the ion bombardment done in order to clean the surface roughened the surface and hindered the √3-phase's formation. However, ion bombardment could not be avoided for two reasons: 1) It was needed in order to reduce the carbon and oxygen contamination to the minimal levels mentioned earlier; heating alone was not sufficient. The reduction of surface carbon was necessary, because substitutional carbon can greatly affect silicon surface-structures. 2) After Mg deposition we wanted to clean the surface and begin anew. Heating did reduce the Mg surface content, but not always completely. In order to completely reduce the Mg content, ion bombardment was needed. Furthermore, we note that if ion bombardment would alter the Si surface so much that the Mg-√3 could not be formed, then serious concerns would be raised with regard to the energetics and stability of this structure. Previous experiments in this and other laboratories have produced the √3's of Al, Au, B,.... using ion-bombardment to clean the surfaces. Would the present be the only case in which ion bombardment is so detrimental? Obviously, such questions of preparation may only be solved by additional experimentation.

4.3.2 Si{111}3x1-Mg

The 3x1 structure, as observed by Vandr'e et.al., presents a greater conundrum. It could be that these authors did see a 3x1 LEED pattern and misidentified it as a 3x1 pattern. Figure 4.8a is a schematic of a Si{111}3x1 pattern. The fractional-order (integral-order) beams are represented by the open triangles (closed circles). Only one rotational-domain of the structure is presented in this figure, and we can clearly see three times as many beams in the k_x direction as in the k_y direction. Figures 4.8b and c display schematics of the other two possible rotationally-related domains. These three patterns combined together form a new pattern, see Figure 4.8d, which contains equal numbers of all three possible domains. Now, Figure 4.8e is a schematic of a 3x1 LEED pattern. It is clearly different. For example, the (⅓ , ⅓) beam is present in a 3x1 pattern and not in a 3x1 pattern. In fact, all hk beams that satisfy the equations h=±[(⅓)+m] and k=h+n, where n,m=0,1,2,.... are not in the 3x1 pattern. At first glance, we notice that the beams missing are the fractional beams of a √3x√3-30 pattern. So, it could be that Vandr'e et.al. just erred. It could also be that they had a 3x1 structure which we could not reproduce, especially if we could not reproduce a √3 phase. It could be, however, that they had both a 3x1 and a √3 pattern mixed which would yield an apparent 3x1 phase. To the trained eye this would be discernible, different domain sizes of 3x1 and √3 would present themselves as varying levels of intensity in the LEED pattern. Different concentrations of Mg, given the same anneal time, should create these different domain populations. This is not mentioned in the SEXAFS study; therefore it must be assumed that this did not occur. Also, if anneal of a 1x1 phase can produce such an apparent 3x1 phase with 3x1 and √3 components, this means that anneal can produce, at least in part, a √3 structure. We have not observed this. The above observations have been made, however, with the Si-√3-Ag and Si{111}3x1-Ag surfaces [79], i.e., an apparent Si{111}3x3-Ag-Ag surface.

Another curiosity arose when comparing the LEED I(V) spectra of the 3x1-Mg phase with those of Si{111}3x1-Na, -Li, and -Ag as collected by Fan et.al.[80], see Figures 4.9 and 4.10. The spectra measured from the 3x1-Mg surface (⅔ 0, 0 ⅔, ⅓ ⅔, ⅔ ⅓, 10, 01) are essentially identical to those from the 3x1-Na, -Li, and -Ag surfaces. Fan et.al.[80] noted the resemblance of I(V) spectra for these three systems. We must remember that the spectra arise from ordered scattering of electrons by the 3x1 unit-cell. Varying the locations, the concentrations, and the type of metal atoms in this unit cell will dramatically change the I(V) spectra. As an example, the atomic radii of Na, Li, and Ag are different. The Na-Si, Li-Si, and Ag-Si bond lengths would therefore be different. The underlying relaxation of the Si-Si bonds would differ depending on the different adsorbate bonding. This would create a different structure for each of the above metal elements. Even if two structures were crystallographically identical, except for the replacement of Na atoms with Ag atoms, the spectra would be very different. Li's electronic configuration is 1s² 2s¹; Na's electronic configuration is 1s² 2s² p⁶ 3s¹; Ag's electronic configuration is 1s² 2s² p⁶ 3s² p⁶ d¹0 4s² p⁶ d¹0 5s¹. They each have very different cores and scattering phase-shifts. Therefore, these atoms cannot be ordered inside the 3x1 unit-cell without affecting the diffracted intensities. They may induce the Si{111} surface to reconstruct, transfer charge to the surface, free-up dangling bonds, but they cannot have long-range order without changing the I(V) spectra. This also applies to the Si{111}3x1-Mg surface. If Mg and Na were in the same ordered structure, then one would expect that their I(V) spectra could be similar, since their electronic configurations differ by only a 3s² electron. However, this is not the case for Li, Ag, and Mg. The only manner in which Si{111}3x1-Na, -Li, -Ag, and -Mg surfaces could yield the same LEED I(V) spectra is if the metal atoms are not ordered in the surface unit-cell and instead induce the Si surface atoms to reconstruct. Similar results have been reported for Ge{111}√3x√3-30-Ge, -Li, and -Ag [81]; for Ge{111}3x1-Li, -Na, and -K [80]; and for Si{111}√3x√3-30--Si, -Ta, and -Ag [31] surfaces.

In conclusion, a Si{111}√3x√3-30-Mg surface was not formed and therefore its structure could not be determined. Instead, a Si{111}⅔√3x⅔√3-30-Mg surface was created by thermal evaporation of Mg on a Si{111}7x7 surface. This surface was most likely related to the {111} planes of a reacted SiMg₂ overlayer. Anneal of Si{111}1x1-Mg surfaces produced Si{111}3x1-Mg surfaces. The LEED I(V) spectra of these surfaces were identical to those of the Si{111}3x1-Li, -Na, and -Ag surfaces. The structures of these four surfaces, therefore, are the same. Mg, Li, Na, and Ag induce the Si surface to reconstruct and are not part of the superstructure.