3
3)R30Æ periodic lattice matches that of bulk SiO2 within 95%. The only dif-ference between the silicate monolayers on SiC and the bulk structure of a high temperature SiO2 phase known as β-tridymite is the posi-tion of the Si atoms. In the bulk structure a sili-cate layer consists of three sublayers with the Si atoms alternatingly positioned below and above
the oxygen atoms. Hypothetically, the sili-cate adlayer found on SiC can be transformed into this structure simply by shifting one of the two Si atoms in the unit cell upwards into this upper Si sublayer position. In that way the (
3
3)R30Æ silicate layer may be a useful seed for further growth of oxide films and help to improve the performance of SiC-MOS struc-tures.
Interface Structure of B2-FeSi on Si(111)
U. Starke; S. Walter, F. Blobner, M. Krause, S. M¨uller, and K. Heinz (Universit¨at Erlangen) The growth of iron silicide on a silicon substrate
is of considerable technological interest, since – dependent on structure and composition – it can be metallic, semiconducting or ferromag-netic. With these versatile properties, iron sili-cide offers a large variety of potential applica-tions when integrated into silicon based devices.
Yet, it appears that the structure of a thin sili-cide film can be different from the stable bulk structure of the corresponding chemical compo-sition. E.g. for 1:1 stoichiometry in the bulk phase diagram ε-FeSi is found which has the rather complex B20-structure (Strukturbericht nomenclature). In contrast, in ultrathin films growth in CsCl-type structure (B2) is formed.
Although B2-FeSi is not stable as bulk struc-ture, it grows as film of high-quality on Si(111) [von K¨anel et al., Physical Review B 45, 13807 (1992)]due to a lattice mismatch of only 1.5%.
Since B2-FeSi is metallic, films on Si are good candidates for Schottky junctions. However, the structure of the interface, which plays a crucial role for such devices, is largely unknown until now. In the present paper, the atomic structure and energetics of the B2-FeSi/Si(111) interface are investigated using quantitative low-energy electron diffraction (LEED) [Heinz, Reports on Progress in Physics 58, 637 (1995)], and density-functional theory (DFT) calculations[Kresse et al., Computational Materials Science 6, 15 (1996)].
The silicide films were prepared by deposition of Fe onto Si(111) and subsequent annealing in ultra-high vacuum (UHV). The film quality was controlled through both the quality of the LEED pattern and the film morphology as imaged by scanning tunneling microscopy (STM). Fig-ure 103(a) displays the image of a film resulting from the initial deposition of 2.60.25 ML Fe (1 monolayer (ML) Fe corresponds to 1 Fe atom per top substrate layer Si atom) and subsequent annealing. Obviously, there is layer-by-layer growth leading to a distribution of domains with two different heights (according to the average coverage being between 2 and 3 silicide lay-ers, i.e. 2 and 3 pairs of Fe and Si layers).
A quantitative evaluation using a Gaussian fit of the histogram shown in Fig. 103(b) confirms the formation of steps of approximately 1.7 ˚A height. Since the layer spacing in B2-FeSi(111) is just about half this value (0.8 ˚A), obviously only double steps exist, i.e. the surface must be terminated by either Si or Fe.
Integration of the two Gaussian peaks yields that 4015% of the substrate is covered by the thinner film domain equivalent to an initial Fe coverage of 2.40.15 ML which largely agrees with the less accurate estimation via the quartz balance. The sharp (11) LEED pattern ob-served (Fig. 103(c)) indicates, that for film, in-terface and substrate the lateral unit cell has the
Figure 103: (a) STM image for (on average) 2.4 silicide layers exhibiting domains of two different heights.
In the inset (Utip= –1.0 V) atomic resolution is provided with the real space unit cell highlighted. (b) STM histogram of (a) and its evaluation retrieving both the step height and the domain weights. The smooth curves represent Gaussian fits. (c) LEED pattern of the same state of the film at 90 eV.
same size and shape, since they are all probed by the LEED electrons at this film thickness.
From the atomically resolved STM image in the inset of Fig. 103(a) the atomic distances are de-termined as3.8 ˚A in accordance to the Si lat-tice parameter.
Figure 104: Side view of the atomic arrange-ments for the different possible configurations of the B2-FeSi/Si(111) interface. Dark spheres represent iron atoms, light spheres silicon atoms. Short bond sticks indicate two bonds directed by 60Æinto and out-of-plane.
Six different atomic configurations at the inter-face were considered as feasible models. As shown in Fig. 104, they can be classified by (i) the coordination of the Fe atom nearest to the interface and (ii) the stacking sequence of the
silicide layers relative to the substrate. The co-ordination of Fe can be 5-, 7- or 8-fold. The stacking sequence at the interface can be un-faulted (A-type) or un-faulted (B-type) so that the unit cell of the silicide is oriented like that of Si or rotated by 180Æ, respectively. Accordingly, the notations A5, B5, A7, B7, A8 and B8 are commonly used to describe the different scenar-ios at the interface.
In the quantitative LEED analysis the best-fit quality can be achieved for the B8-type interface configuration (Pendry R-factor RP= 0.159). Yet, the next best interface type (B5) with RP= 0.180 is not outside the limits of statistical errors, i.e., it cannot be excluded safely by LEED. This is due to the fact that the only qualitative difference of the B5- and B8-type interface is the missing Si layer be-tween the lowest Fe layer and the Si substrate (Fig. 104). As Si is a considerably weaker scat-terer than Fe and as the layer’s contribution is weakened by electron attenuation, only moder-ate differences can be expected for the diffrac-tion intensities of the two models. Neverthe-less, the B8-type interface provides the best fit and (except for B5) all other configurations can be ruled out. The preference for a B8-type in-terface is also found in the energy minimization calculations – and incidently, here the next best interface configuration is A8 while the B5 inter-face can be ruled out. LEED on the other hand
eliminates A8, so that the collaboration of the two methods clearly identifies B8 as the true configuration. Table 4 lists the R-factors and interface energies for∆µSi= 0 for all six mod-els considered. It should be noted that even for
∆µSi=∆HFeSithe B5 model can be savely ruled out [Walter et al., Journal of Physics: Condensed Matter, submitted].
Table 4: Best-fit Pendry R-factors RP from LEED and interface energies (for ∆µSi= 0) from DFT [eV/unit cell] for the different models tested.
Model A5 B5 A7 B7 A8 B8
RP 0.237 0.180 0.209 0.206 0.221 0.159 EInt 1.78 1.82 1.09 1.11 0.70 0.63
With this interface model the film thickness of the two domains found in the STM corre-sponds to 5 and 7 atomic layers, respectively.
The LEED intensities calculated for the two do-mains were combined for the structure analy-sis, whereby the layer spacings were optimized independently for both domains. DFT calcula-tions were carried out also for 5 and 7 atomic layer films. The layer spacings are listed in de-tail in Fig. 105(b) according to the nomencla-ture in panel (a). The substantial relaxations at the film surfaces and at the interface detected by both DFT and LEED agree rather well. Note-worthy, the relaxations are very different for the 5 and 7 layer domains. Already the sec-ond spacing above the interface (dI2) differs by 0.2 ˚A. This reflects the mutual influence of sur-face and intersur-face in the ultrathin films. How-ever, a comparison to a superlattice calculation with a thick film (Fig. 105(b)) reveals, that al-ready the interface of the 7 atomic layer do-main (3 ML Fe) is very similar to a deeply bur-ried interface. Each near interface spacing of the 7 layer film is closer to the related superlat-tice spacing than to the corresponding 5 layer value. This indicates that the decay length of the mutual interplay between surface and inter-face as carried by the strong covalent bonding in the material is of the order of about five atomic
Figure 105: (a) Best-fit model for a B2-FeSi film containing 7 atomic layers (3 ML Fe) in B8 inter-face structure. (b) layer spacings for 5 and 7 atomic layer models from LEED and DFT as well as a thick film (DFT).
Accordingly, LEED which is typically viewed as a surface technique can be used for quan-titative interface analysis on an atomic level, when homogeneous films can be prepared thin enough. The good agreement between LEED and DFT as well as the excellent LEED fit qual-ity prove, that the silicide film can be grown with a well defined and sharp interface and so Schottky junction devices seem feasible with