Structural and compositional reversible phase transitions on low-index

EUROPHYSICS LETTERS

15 December 2001

Europhys. Lett., 56 (6), pp. 822–828 (2001)

Structural and compositional reversible phase transitions on low-index Fe3 Si surfaces U. Starke 1 (∗ ), J. Schardt 1 , W. Weiss 1 , W. Meier 1 , C. Polop 1,2 , P. L. de Andres 2 and K. Heinz 1 1 Lehrstuhl f¨ ur Festk¨ orperphysik, Universit¨ at Erlangen-N¨ urnberg Staudtstr. 7, D-91058 Erlangen, Germany 2 Instituto de Ciencia de Materiales (CSIC) Cantoblanco, E-28049 Madrid, Spain (received 4 July 2001; accepted in final form 4 October 2001) PACS. 68.35.Rh – Phase transitions and critical phenomena. PACS. 68.35.Bs – Structure of clean surfaces (reconstruction). PACS. 61.14.Hg – Low-energy electron diffraction (LEED) and reflection high-energy electron diffraction (RHEED).

Abstract. – The (100), (110), and (111) surfaces of Fe3 Si have been studied by quantitative Low-Energy Electron Diffraction and Auger Electron Spectroscopy. Reversible phase transitions between the D03 and the CsCl structure develop upon annealing, triggered by a substantial reversible surface segregation of Si. On all surfaces, Si termination is preferred either by realizing a topmost Si layer whenever the choice between Si and Fe exists, or by direct Si occupation of nominal Fe sites. Similarities to epitaxial iron silicide films structural behaviour are identified.

Fe-Si alloys have attracted widespread interest for technological and fundamental reasons. For iron-rich compositions, the material exhibits ferromagnetism, and it is considered a promising candidate for the construction of magnetic devices in silicon-based integrated circuits [1]. At room temperature (RT) three homogeneous iron silicide phases of different stoichiometry exist in the bulk phase diagram. Among those, Fe3 Si is the only well-ordered phase of iron-rich composition (fig. 1). It is stable from RT up to the melting point [2]. Its crystal structure is of D03 type, where silicon atoms have maximized their mutual distance (see fig. 2), prevailing in a rather large area of the phase diagram below the ideal silicon concentration of 25%. This is through random occupation of sites C by Fe and Si which still maintains the D03 symmetry. Below about 10% Si —in a very narrow concentration regime— the D03 phase is replaced by a B2 (= CsCl) phase, where sites C and B are randomly occupied by Fe or Si [3]. Below this composition all atomic sites are randomly occupied by Fe or Si, which corresponds to the A2 (= bcc) crystal structure. On the Si-rich side immediately above the ideal concentration of the D03 structure an abrupt transition to an inhomogeneous phase region is observed. As indicated in fig. 1, the phase boundary runs from 26% at RT to 31% at 1200 ◦ C [2–4]. At (∗ ) Present address: Fritz-Haber-Institut der Max-Planck-Gesellschaft - Faradayweg 4-6, D-14195 BerlinDahlem, Germany. c EDP Sciences

U. Starke et al.: Structural and compositional reversible phase etc.

823

1600 1400

Temperature [˚C]

1200 α-FeSi 2

1000

B

950˚C ,fer rom ag.

600

A

A C

400 Fe 3Si

ε-FeSi

200 0

0

10

20

30

40

50

60

B

β-FeSi 2

70

Atomic percent Silicon

Fig. 1

C

TC

800

80

90 100

Fe (A)

Fe (B)

Si (C)

Fig. 2

Fig. 1 – Phase diagram of iron silicide (after [2]). Ordered, homogeneous phase regions are indicated by the grey shading. Fig. 2 – Unit cell of the D03 structure of Fe3 Si made up of bcc- and CsCl-type subcubes. Random occupancies of sites B and C, or all three sites, lead to CsCl or bcc, respectively.

higher Si concentration two ordered phases exist at RT, namely -FeSi and β-FeSi2 . In ultrathin films, however, different crystal phases than in bulk material are observed in the entire composition range [5,6]. While -FeSi and β-FeSi2 are substituted by other crystal structures, namely CsCl for 1:1 composition [6–8] and a still debated structure in the case of FeSi2 [9–11], ultrathin films in the Fe3 Si composition region did not develop well-ordered structures [5], but ordered phases have been reported in the literature for thicker films up to 800 ˚ A [12]. Also, on the (100) surface of bulk Fe3 Si, atomic segregation and diffusion within the surface region are observed, leading to deviations from a spatially homogeneous stoichiometry and to a change in surface periodicity [13, 14]. Knowledge of the surface structures involved should help to understand segregation and diffusion processes, both for the structural development of thin-film silicides and for surfaces related to bulk crystals. For that purpose, we present a systematic investigation of the structure and evolution of different phases on low-Miller-index surfaces of Fe3 Si. Quantitative low-energy electron diffraction (LEED) and Auger electron spectroscopy (AES) were used to analyze the phases which develop upon annealing of the samples. The samples with (100), (110) and (111) orientation were cut from a single crystal Fe3 Si. Ex situ polishing was followed by repeated cycles of sputtering and heating in ultra-high vacuum (UHV). Thereafter, they went through a prolonged annealing treatment at different temperatures between 350 and 750 ◦ C and the temperature-dependent development of new surface phases was monitored by LEED and AES. LEED intensity measurements were carried out using a 4-grid rear view LEED optics from whose luminescent screen intensity vs. energy spectra, I(E), were recorded with a computer-controlled video-LEED system [15]. AES spectra were measured using the LEED optics as a retarding field analyzer. The peak-to-peak ratio rFe/Si for Fe and Si Auger transitions at 47 eV and 92 eV, respectively, was used to describe the stoichiometry changes in the surface region. On each surface two different phases developed by annealing, one below 400 ◦ C and another around 600 ◦ C denoted in the following as low (LT) and high (HT) temperature phases, respectively. The transition between them is reversible, although rather slow. The preparation

824

EUROPHYSICS LETTERS

D03 a)

CsCl

bcc

b)

c)

(100) (1x1)

c(1x1)

D0 3 -cube

d)

c(1x1)

e)

f)

(110)

g)

( ½¼ 0½

( 1/2 x1)

(1x1)

h)

i)

(111) (1x1)

/

1./2.layer

(1x1)

Fe

(1x1)

Si

mixed

Fig. 3 – Atomic models for the three surfaces and unit cell notation consistent with the observed LEED patterns (occupations are shown by different shadings).

of the LT phase requires annealing times of about 100 min, whilst the HT phase develops after about 10 min. These large transition times allowed us to freeze the phases by a rapid quench to low temperatures (liquid nitrogen), where LEED and AES measurements were taken. As an example, quenching the sample from 600 ◦ C to 200 ◦ C took 1 min, ensuring that the LT phase could not develop with noteworthy weight. It is interesting to note that this reversible structural transition at the surface takes place in the same temperature regime as the ferromagnetic phase transition in bulk Fe3 Si (cf. fig. 1). From the analysis of the different LEED patterns we obtain the size and symmetry of the 2D unit cells. The simplest case is the LT phase of the (100) surface. Whilst for this orientation all three feasible bulk structures (bcc, D03 , CsCl) exhibit a square unit cell (see fig. 3a) consistent with the symmetry of the LEED pattern observed, its size calculated from the diffraction angles is only consistent with that of the D03 phase and we denote this phase as (1 × 1). In comparison, in the LEED pattern of the HT phase of the (100) surface, diffraction spots (n, m), where n + m is odd, are systematically extinguished, resulting in a smaller real space unit cell, namely c(1 × 1). This is consistent with the bulk-like termination of both the CsCl and the bcc structure as displayed in fig. 3b, c, and a decision through LEED can only come from the analysis of the diffraction intensities (see below). Yet, AES reveals different compositions for the LT phase (rFe/Si = 3.5) and the HT phase (rFe/Si = 2.4 at 600 ◦ C) indicating a considerable silicon enrichment at least in the surface region (notice that this region, defined by some appropriate attenuation length for electrons, is roughly twice times larger for AES than for LEED). Figure 4 displays the reversible transition from the LT to the HT phase for the (100) face as monitored by the Fe/Si AES-ratio. In addition, the stability regimes of the two phases were investigated earlier through the analysis of the


825

Auger ratio r(Fe/Si)

4.0 3.5 3.0 2.5 2.0 1.5 400

450

500

550

600

650

temperature (˚C)

Fig. 4 – Reversible transition of the (100) surface from the D03 (T ≤ 400 ◦ C) to the CsCl (500– 650 ◦ C) structure monitored by the Auger ratio between the Fe and the Si signal (the solid line is a least-squares fit to the data).

temperature dependence of their I(E) spectra [14]. From these two independent measurements we establish that the transition between the two structures starts between 400 and 450 ◦ C, and the c(1 × 1) phase is already stable between 500 and 625 ◦ C. For the sake of completeness, we should mention that a third phase of (1 × 1) symmetry is observed above 700 ◦ C. Yet, the transformation to this higher-temperature phase is not reversible and we will not consider it further in the present paper. Furthermore, such an extra phase is not found on the (110) and (111) surfaces. In the (110) surface every atomic plane contains sites A, B and C. As a consequence, bulk-like termination of the three possible bulk structures leads to different surface unit cells as displayed in fig. 3d, e, f. The experimental LEED pattern for the LT phase is compatible only with the bulk-like–terminated D03 structure (fig. 3d) for which we use again the notation (1 × 1). Upon annealing, certain spots (e.g., the (10) spot) disappear, and a new structure develops displaying a ( 12 × 1) LEED pattern which is compatible only with the CsCl structure (see fig. 3e). This HT phase is fully developed at about 575 ◦ C. Similarly to the (100) face, it is Si-enriched relative to the LT phase as judged by AES (LT: rFe/Si = 3.6; HT: rFe/Si = 2.8). For the (111) face, the size and shape of the 2D unit cell is the same for D03 , CsCl and bcc structures (see fig. 3g, h, i), and each layer contains only one type of atoms from the D03 cube, whereby layers are stacked as {ABAC}. Consequently, the observed patterns for both the LT and HT phases are of the same (1 × 1) symmetry, i.e. no spots disappear upon annealing. Nevertheless, the I(E) spectra of the phases are substantially different, and so must correspond to different structures. The Auger signal again shows that the HT phase (rFe/Si = 2.6) contains more silicon than the LT phase (rFe/Si = 3.6). The HT phase develops at about 575 ◦ C. From the experimental observations, we can summarize that the high occupational order of the D03 crystal structure is lost in the HT phase, and it is accompanied by a considerable Si enrichment in the surface region. In order to identify the atomic structure of the different phases and the occupancy of the different sites A-C, full dynamical analyses of the measured I(E) data sets were carried out. The perturbative method Tensor LEED [16], and its extensions to stoichiometry and atomic vibrations [17], were applied to vary geometrical parameters as well as the chemical occupation of sites and atomic vibrational amplitudes. An automated search procedure [18] guided by the Pendry R-factor, RP [19], was applied to find the best fit between theory and experiment. It turned out that there are no major geometrical

826

EUROPHYSICS LETTERS

reconstructions on the surfaces (except the usual multilayer relaxations and some buckling in mixed layers). Instead, the surfaces investigated are characterized by their surface termination and the phase-specific occupation of sites A-C in subsequent atomic layers (the reader should recall that, e.g., a bulk-like terminated D03 (100) surface could be terminated either by a pure iron or by a mixed layer). The best-fit R-factors retrieved for the different phases were all below 0.2, and so very convincing even for the most exigent standards used in LEED. Based on the variance of the Pendry R-factor [19], which takes into account the large experimental data base, atomic positions can be resolved to about 0.05 ˚ A or better, and the error margin for atomic concentrations is estimated to be below 20%. This accuracy level is restricted to only the top few layers by electron attenuation. In the present case this depth limit lies in the 10–15 ˚ A region. Details of the structure analyses, together with a complete listing of the layer relaxations and bucklings detected, will be published in a forthcoming paper. Here we concentrate on the identification of the above-mentioned phases (D03 or CsCl), the surface termination and the layer-dependent stoichiometry. In fig. 5 the results for the occupancies of the different sites in the top three layers of the high- and low-temperature phases are tabulated for the three faces investigated. It should be noted that in some cases the analyses found additional structural domains of different stoichiometry. However, they occupy only a small fraction of the surface, which we interpret as the slow phase transition being still incomplete. Therefore, we concentrate on the main domains. The overall appearance of the results shows that in the LT phases the D03 crystal structure is maintained, while in the HT a transition to CsCl crystal structure is observed. In the LT phases we observe the following prominent features: Whenever the surface has a choice for its layer termination (i.e. on the (100) and (111) surfaces) it chooses a layer containing Si. The Si layer concentration at least in the top two layers is always enriched compared to the respective bulk layer stoichiometry (if it is not 100% anyway). The corresponding segregation is most pronounced in the top layer of the (110) surface in which basically all original Fe atoms on B sites are substituted by Si. Furthermore, on the top layer of the (100) surface, 30% of atoms in B sites are replaced by Si. The trend for Si enrichment diminishes with increasing depth into the surface, the third or fourth layers already show bulk (i.e. D03 ) layer stoichiometry as indicated in fig. 5. Correspondingly, the LT diffraction patterns always show the full (1×1) symmetry of the D03 structure. In the HT phases, the surfaces are strongly Si enriched according to both the LEED analyses and the Auger signals, cf. figs. 4 and 5. In fact, B-positions (originally Fe in the D03 structure) are now almost completely occupied by Si. For the (100) and (111) orientations the surface prefers a termination by a full Si layer. Similarily, in the top layer of the (110) surface besides the B-positions (originally Fe) being completely occupied by Si, we find 20% vacancies on the A-positions (Fe). A slightly lower Si enrichment is detected in the next two layers, yet, the surface region has completely transformed into a CsCl-type structure. D03 order within the penetration depth of the LEED electrons can be ruled out. Both for (100) and (110) surfaces this follows from the absence of any trace for a (1 × 1) LEED pattern, while for (111) surfaces it was checked by dynamical LEED calculations. Only below this penetration limit we assume a D03 bulk as indicated by the dotted lines in fig. 5. The transitions between the different LT and HT surface structures observed, i.e. between a D03 and a CsCl phase, prove to be reversible. This is not in contradiction to the long time scales involved, which only imply that the transitions are kinetically slowed down. Reversibility rules out that surface segregation of silicon upon annealing at increasing temperatures might be caused by some initial depletion through sputtering during sample preparation and the subsequent stoichiometric recovery of the surface, as has been observed, e.g., in some iron aluminides [20]. Furthermore, the usual driving force for segregation, namely the gain


(100)

L Tphases

B(70%Fe) C(100%Si) A(70%Fe) B(100%Fe) C(100%Si)

D03

(110)

(111)

A(90%Fe) B(100%Si) C(90%Si)

C(100%Si)

A(100%Fe) B(90%Fe) C(90%Si)

B(90%Fe)

A(100%Fe) B(100%Fe) C(100%Si)

827

A(80%Fe) A(100%Fe)

D03

D03

HTphases

B(100%Si) C(100%Si) A(100%Fe) B(100%Si) C(100%Si)

CsCl

A(80%Fe+ ) B(100%Si) C(100%Si)

C(90%Si)

A(100%Fe) B(60%Si) C(60%Si)

B(70%Si)

A(100%Fe) B(80%Si) C(80%Si)

A(100%Fe) A(100%Fe)

CsCl

CsCl

D03

D03

D03

Fig. 5 – Occupancy of sites A-C for the top three layers of the different phases as resulting from quantitative LEED for the dominating structural domains. E.g., 80% Fe means that 20% of sites are occupied by Si except when the presence of vacancies is indicated by an open square. Underneath the displayed layers, the structure is D03 (LT) or (disordered) CsCl (HT), at least within the penetration depth of LEED.

in binding energy coming by the modified bond situation within the surface, cannot be made responsible as this should be largely independent of temperature. Instead, as the total energy of different structures can be very close, and as the free energy rather than the energy should be minimized, entropy effects might induce a more homogeneous distribution of the two elements within the surface (rather than the 3:1 ratio of the ideal D03 structure). However, the entropy effect is not large enough to enforce random occupation of all sites leading to (disordered) bcc. Instead, a CsCl structure develops even though the bulk phase at this concentration should not be stable (fig. 1). However, the CsCl structure happens to be the stable one developing in ultrathin silicide films at this composition. A possible explanation would be that the structural development is dictated by the interactions within the thin surface slab enriched in silicon and the driving forces may be the same as in the case of epitaxial films. Alternatively, the fact that the temperature observed for the thin-film structural transition coincides with the ferromagnetic phase transition in bulk Fe3 Si might suggest that magnetic forces influence the delicate balance between the different structures found on the Fe3 Si surfaces. However, a definite answer to the question for the driving forces for these structural developments observed can only come from a theoretical investigation. In summary, by applying quantitative LEED and AES, the temperature-dependent phases for the (100), (110) and (111) surfaces of Fe3 Si have been determined. A general tendency for surface enrichment of Si has been identified: this is a reversible phenomenon that can be controlled at will by changing the temperature. As a consequence, reversible phase transitions

828

EUROPHYSICS LETTERS

from the D03 to the CsCl structure and vice versa take place. The fact that segregation leads to a CsCl phase at higher temperatures may be correlated to the existence of the same phase in thin epitaxial films of iron silicides. The influence of magnetic forces, however, cannot be ruled out. ∗∗∗ We thank Dr. H. Viefhaus (MPI f¨ ur Eisenforschung, D¨ usseldorf) for providing the sample ingot, and Dr. H. Conrad (FHI, Berlin) for helping with ex situ preparation. We are also indebted to Deutsche Forschungsgemeinschaft (DFG) through SFB 292 and Sta315/4 as well as to Spanish DGES-PB98-524. CP is grateful for financial support by the Spanish Ministry of Education. REFERENCES ¨sges R., Carbone C., Eberhardt W., Pampuch C., Rader O., Kachel T. and Gudat [1] Kla W., Phys. Rev. B, 56 (1997) 10801. [2] Hansen M., Constitution of Binary Alloys (McGraw-Hill, New York) 1958. [3] Kubaschewski O., Iron Binary Phase Diagrams (Springer, Berlin) 1982. [4] Nishino Y., Inoue S., Asano S. and Kawamiya N., Phys. Rev. B, 58 (1993) 13607. [5] Alvarez J., Hinarejos J. J., Michel E. G. and Miranda R., Surf. Sci., 287/288 (1993) 490. ¨nel H., Ma ¨ der K. A., Mu ¨ller E., Onda N. and Sirringhaus H., Phys. Rev. B, 45 [6] von Ka (1992) 13807. [7] Kafader U., Tuilier M. H., Pirri C., Wetzel P., Gewinner G., Bolmont D., Heckmann O., Chandesris D. and Magnan H., Europhys. Lett., 22 (1993) 529. [8] Hinarejos J. J., Castro G. R., Segovia P., Alvarez J., Michel E. G., Miranda R., Rodriguez-Marco A., Sanchez-Portal D., Artacho E., Yndurain F., Yang S. H., Ordejon P. and Adams J. B., Phys. Rev. B, 55 (1997) R16065. [9] Vazquez de Parga A. L., de la Figuera J., Ocal C. and Miranda R., Ultramicroscopy, 42-44 (1992) 845. ¨ller-Gubler E., Mu ¨ller P., Stalder R. and von Ka ¨ nel [10] Sirringhaus H., Onda N., Mu H., Phys. Rev. B, 47 (1993) 10567. [11] Jedrecy N., Waldhauer A., Sauvage-Simkin M., Pinchaux R. and Zheng Y., Phys. Rev. B, 49 (1994) 4725. [12] Onda N., Sirringhaus H., Goncalves-Conto S., Schwartz C., Zehnder S. and von ¨nel H., Appl. Surf. Sci., 73 (1993) 124. Ka [13] Busse H., Kandler J., Eltester B., Wandelt K., Castro G. R., Hinarejos J. J., Segovia P., Chrost J., Michel E. G. and Miranda R., Surf. Sci., 381 (1997) 133. [14] Starke U., Meier W., Rath C., Schardt J., Weiß W. and Heinz K., Surf. Sci., 377-379 (1997) 539. [15] Heinz K., Prog. Surf. Sci., 27 (1988) 239. [16] Rous P. J., Prog. Surf. Sci, 39 (1992) 3. ¨ ll R. and Kottcke M., Surf. Rev. Lett., 3 (1997) 1651; Heinz K., Rep. Prog. [17] Heinz K., Do Phys., 58 (1995) 637. [18] Kottcke M. and Heinz K., Surf. Sci., 376 (1997) 352. [19] Pendry J. B., J. Phys. C, 13 (1980) 937. [20] Heinz K. and Hammer L., J. Phys. Condens. Matter, 11 (1999) 8377.