- Email: [email protected]
An atomic scale STM study of the Fe3O4(0 0 1) surface
Surface Science 548 (2004) 106–116 www.elsevier.com/locate/susc
An atomic scale STM study of the Fe3O4(0 0 1) surface S.F. Ceballos *, G. Mariotto, K. Jordan, S. Murphy, C. Seoighe, I.V. Shvets SFI Nanoscience Laboratory, Department of Physics, Trinity College, Dublin 2, Ireland Received 5 August 2003; accepted for publication 27 October 2003
Abstract Despite the intensive investigation into the electronic properties of magnetite, fundamental issues related to the Verwey transition and the electronic transport mechanism are not fully understood. These issues are further complicated at the surface of magnetite crystals, due to the large number of possible surface terminations. The preparation procedure plays a fundamental role in determining the O/Fe ratio, and therefore the electronic properties of a magnetite crystal. We present a detailed investigation of the influence of the preparation conditions on the morphology of Fe3 O4 (0 0 1) single crystal surfaces using AES, LEED, and STM. We show that long anneals of single crystals in UHV cause segregation of contaminants to the surface and that a series reconstructions is induced. A different pffiffiffiof surface pffiffiffi preparation procedure gives rise to a clean surface exhibiting a ð 2 2ÞR45 reconstruction. This surface is terminated at the octahedral plane and has been imaged down to the atomic scale. This provides a useful test system to study the Verwey transition at the surface. 2003 Elsevier B.V. All rights reserved. Keywords: Scanning tunneling microscopy; Iron oxide; Low energy electron diffraction (LEED); Auger electron spectroscopy
1. Introduction Magnetite has been the subject of intensive studies by the scientific community during the last decades. Efforts to understand its magnetic and electrical properties have intensified in the past few years due to its half-metallic nature, which makes magnetite an interesting material for spin electronics applications. Spin electronics almost invariably involves electron transport across or along an interface, so that the understanding of electron transport in low dimensional systems is
*
Corresponding author. Tel.: +353-1-608-2020; fax: +353-1671-1759. E-mail address: [email protected] (S.F. Ceballos).
therefore of critical importance. Several examples of electronic and spin-electronic surface studies can be found in [1–3]. Stoichiometric magnetite undergoes a metal– insulator transition at around 125 K, known as the Verwey transition temperature TV . At room temperature, the electrical conductivity of Fe3 O4 is 200 X1 cm1 and it gradually decreases with decreasing temperature. When cooled down below TV , the conductivity abruptly decreases by about two orders of magnitude [4,5]. The change of conductivity is accompanied by a change in the crystallographic structure, whose symmetry is lowered from cubic to monoclinic. Magnetite is an inverse spinel material. The crystal structure is based on a face-centered cubic (f.c.c.) unit cell, containing 32 O2 anions and 24
0039-6028/$ - see front matter 2003 Elsevier B.V. All rights reserved. doi:10.1016/j.susc.2003.10.041
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
mixed valence Fe cations, with a lattice parameter . The formula can be written as of a ¼ 8:3963 A YA ½XY B O4 , where X ¼ Fe2þ , Y ¼ Fe3þ and A and B denote tetrahedral and octahedral sites, respectively. This formula indicates that one half of the ferric Fe3þ cations occupies 8 of the 64 available tetrahedral interstices, and the other half of the ferric cations, together with an equal amount of ferrous Fe2þ cations, occupy 16 of the 32 available octahedral interstices. The (0 0 1) plane of magnetite can be viewed as a stacking sequence of two alternating layers. The A-layer contains tetrahedrally coordinated Fe3þ ions, while the B-layer is composed of rows of octahedrally coordinated Fe2þ and Fe3þ ions surrounded by oxygen ions. The separation between neighboring planes (i.e. the A–B interplanar sep, while the separation between aration) is 1.05 A successive like planes (i.e. the A–A or B–B inter. In each octahedral planar separation) is 2.10 A plane, the nearest-neighbor B-site cations form rows that run along the [1 1 0], [1 1 0], [1 0 1], [1 0 1], [0 1 1] and [0 1 1] directions. The rows in successive octahedral planes are rotated by 90 with respect to one another, giving these planes a two-fold rotational symmetry. In contrast, the arrangement of cations in the A-layers give them a four-fold rotational symmetry. Both A- and B-terminated surfaces have been reported in the literature [6–8], with no satisfactory explanation of why the (0 0 1) magnetite surface should be terminated at either plane. The preparation conditions seem to play a crucial inffiffiffideterprole ffiffiffi p mining the surface termination. A ð 2 2ÞR45 clean surface reconstruction has been observed by several groups on both natural and synthetic single crystals, and on thin films grown by molecular beam epitaxy (MBE). Tarrach et al. [7] have suggested that the top-most surface layer consists of a full monolayer of tetrahedral Fe ions. Chambers et al. [9] used X-ray photoelectron spectroscopy (XPS), X-ray photoelectron diffraction (XPD) and scanning tunneling microscopy (STM) to support the conclusion that the Fe3 O4 (0 0 1) surface is constituted by half a monolayer of tetrahedral Fe ions and that this surface is charge compensated. Wiesendanger et al. [1] imaged two distinct structures on different areas of a natural crystal of
107
magnetite; the two structures were attributed to bulk terminated tetrahedral and octahedral terminations of the surface. In contrast, Voogt et al. [6] and more recently Stanka et al. [10] have proposed a B-termination layer on a single crystal and a Fe3 O4 /MgO(0 0 1) film, respectively. In both cases, they have proposed that charge compensation is achieved by an array of oxygen vacancies accompanied by a change in the valence p state ffiffiffi of the Fe. Mariotto et al. [8] observed a ð 2 pffiffiffi 2ÞR45 reconstruction on a clean magnetite (0 0 1) single crystal surface and proposed a B-layer surface termination. They further proposed that the reconstruction is due to charge ordering of the Fe cations in octahedral positions, and not to an ordered array of vacancies as proposed by previous studies. In the present study, a contaminant-free magnetite surface was obtained using the procedure described in Section 3.1.1. The clean pffiffiffi magnetite pffiffiffi surface was found to exhibit a ð 2 2ÞR45 superlattice. Subsequent annealing for long periods in UHV lead to the segregation of contaminants to the surface. The typical contaminants observed in our crystals were potassium, sulphur and calcium. Continuous annealing of the sample results in the removal of K and S from the surface and an increase in the Ca concentration. Our data shows that the (0 0 1) contaminated magnetite surface is usually terminated at the octahedral plane (B-layer). We have observed p(1 · 1), p(1 · 2), p(1 · 3) and p(1 · 4) reconstructions on the Fe3 O4 (0 0 1) surface. Surface reconstructions induced by alkaline and alkaline-earth metals have been observed on a range of different systems, such as thin films of Fe3 O4 grown on MgO(0 0 1) substrates [11,12], thin films of Fe3 O4 grown on Pt(1 1 1) substrates [13] and TiO2 single crystals [14,15]. A p(1 · 4) reconstruction was observed by Anderson et al. [11] on a 1 lm thick film of magnetite grown on a MgO(0 0 1) substrate. A similar effect was observed by Voogt et al. [6] who reported the onset of a p(1 · 3) reconstruction on a magnetite thin film grown on MgO(0 0 1). Anderson et al. attributed the surface reconstruction to the segregation of magnesium from the substrate, leading to the formation of a MgFe2 O4 phase. Gao and Chambers [16] in a study of magnetite thin
108
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
films grown on MgO(0 0 1) suggested that the Mg ions diffuse through the iron oxide via vacancies in the octahedral cation sublattice. Once they reach the surface, they decorate the rows of octahedral B-site cations. N€ orenberg and Harding [15] have investigated the Ca-induced surface reconstruction on TiO2 . They have observed a variety of surface reconstructions, including a p(1 · 3) and a p(1 · 4). They attributed the p(1 · 3) reconstruction to Ca2þ cations replacing the Ti cations at the surface of the crystal. A p(1 · 4) reconstruction, caused by an ordered array of trenches, was found to succeed the p(1 · 3) reconstruction and it was attributed to oxygen loss from the surface.
2. Experimental A set of synthetic and natural single crystals has been used in these experiments. The natural crystals were cut from the single crystalline nugget used by Tarrach et al. [7], which originated from Zillertal, Austria. The crystals were aligned with a precision of ±1 with respect to the (0 0 1) plane and were oriented along the [0 1 0] direction. The synthetic crystals were grown employing the skull melting technique [17]. The crystals were first characterized by powder X-ray diffraction. The diffraction patterns showed good agreement with the reference spectra for magnetite, and lattice constants of 8.398 ± 0.010 and 8.406 ± 0.010 A were measured for the synthetic and natural crystals, respectively. Four-point resistance vs. temperature measurements were also performed; Verwey transition temperatures of 108 and 98 K were measured for the synthetic and natural crystals, respectively. The crystals were mechanically polished using diamond paste with grain size down to 0.25 lm before being introduced into the UHV system. This system is equipped with facilities for low energy electron diffraction (LEED), Auger electron spectroscopy (AES) and STM analysis. A detailed description of the UHV system is available elsewhere [8]. STM measurements were carried out at room temperature in constant current mode using a home-built instrument. A SCALA controller by OMICRON GmbH was used. This controller allows the user to compensate for the
thermal drift using a topographic feature of the STM image as a reference point. A 0.1 nm/s thermal drift was measured when no correction was applied. By enabling this option, thermal drift was reduced to 0.0025 nm/s. A bias voltage between +0.6 and +1 V was applied to the sample and a tunneling current of between 0.1 and 0.3 nA was typically used. A stable tunneling current could not be obtained under negative bias conditions. We have developed two different in-vacuum preparation procedures to obtain a clean surface. Both procedures lead pffiffiffi topaffiffiffi contaminant-free surface, exhibiting a ð 2 2ÞR45 superlattice, but with different surface terminations. The first procedure consisted in annealing the crystals in UHV at 990 ± 10 K. This has a two-fold effect of reducing the surface and causing the segregation of impurities at the surface. The impurities were removed from the surface by Arþ sputtering for 10 min at 1 keV (Itarget 18 lA). The crystals were then annealed in an oxygen partial pressure (the typical exposure varied between 3600 and 7200 L) to compensate for the reduction of the surface caused by the long UHV annealing. The crystals were finally annealed in UHV at the same temperature for a short period of time (typically 2–8 h). It was found that this preparation procedure consistently led to a B-terminated surface. The second procedure is identical to the first one, with the exception of annealing the crystals in hydrogen (typical exposure of 2000 L) instead of oxygen. This led to co-existing A- and B-terminations. Further annealing of the crystals in UHV at 990 ± 10 K ranging from 10 to 70 h led to a series of reconstructions such as p(1 · 2), p(1 · 3) and p(1 · 4). Three different types of STM tips made out of Ni, W and MnNi have been used. Tungsten tips are standard tips for most STM applications, although magnetic tips are more attractive due to their potential applicability to Spin Polarized STM (SP-STM) measurements [18,19]. Nickel, Tungsten and Manganese–Nickel tips have been used to image the p(1 · 2), p(1 · 3) and p(1 · 4) reconstructed surfaces, respectively. In this paper, the effect that tips with different electronic properties might have on the STM images has not been
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
109
pffiffiffi pffiffiffi 3.1. Clean Fe3 O4 surface––ð 2 2ÞR45 superlattice
, correheights that are integer multiples of 2.1 A sponding to the separation between A–A or B–B planes. However, we have determined that the surface is terminated at the B-plane, since a 90 rotation of the iron rows has been observed on [8]. neighboring terraces separated by 2.1 A Fig. 2 shows an atomically resolved STM image. periodicity A highly regular structure with a 12 A along the [1 1 0] direction is visible. Adjacent atomic . This structure was rows are separated by 6 A found to extend over a length of the order of about 40 unit cells, suggesting that long-range order had set in on the surface. An extensive analysis of these results is given by Mariotto et al. [8]. The
3.1.1. B-termination AES, LEED and STM were performed on the B-terminated, which was obtained using the procedure described in Section 2. AES analysis of this surface indicated that it was contaminant-free, while the O(509 eV)/Fe(700 eV) ratio and the Fe M2;3 VV peak shape of the AES spectra provide strong evidence that no other iron oxide phase but magnetite was present. A typical LEED pattern of the surface is shown in Fig. order pffiffi1(a). ffi pFractional ffiffiffi spots corresponding top affiffiðffi 2pffiffiffi 2ÞR45 mesh are clearly visible. The ð 2 2ÞR45 superlattice was routinely reproduced following the preparation conditions described above. The STM images are characterized by multiple rectangular terraces with step-edges oriented along the [1 1 0] and [1 1 0] directions. The terraces are separated by step
2 STM image. Rows of Fe cations aligned Fig. 2. 140 · 210 A along the [1 1 0] direction are visible. Adjacent rows are sepa. rated by 6 A
addressed; only a detailed description of the topography and morphology of the surface is given. Further experiments have been carried out on the magnetite surface using different tip materials by our group. Differences in the STM images of the clean magnetite surface when scanning with magnetic and non-magnetic tips are discussed elsewhere [20].
3. Results and discussion
Fig. pffiffiffi 1.p(a) ffiffiffi LEED pattern of a clean Fe3 O4 (0 0 1) surface taken with a primary electron energy of 78 eV. The p(1 · 1) unit cell and the ð 2 2ÞR45 superlattice are highlighted. (b) LEED pattern taken with a primary electron energy of 47 eV. Satellite spots around the primary spots are clearly seen and marked with white arrows along the [1 1 0] and [1 1 0] directions. Their separation is 1/3 of that between the integral order spots. (c) LEED pattern recorded for a p(1 · 4) surface reconstruction. The primitive unit cell is indicated with a white square labelled ‘‘1’’. The p(1 · 4) unit cell is marked with the white square labelled ‘‘2’’.
110
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
pffiffiffi pffiffiffi ð 2 2ÞR45 superlattice observed by LEED and STM was attributed to an ordering of electron charges, with the formation of Fe2þ –Fe2þ and Fe3þ –Fe3þ dimers at the B-sites. We emphasize that such long-range order is established at room temperature suggesting that the metal–insulator transition on the surface occurs at a temperature that is well above the bulk temperature TV . 3.1.2. Co-existing A- and B-terminations Co-existing A- and B-terminations were obtained by annealing in a hydrogen atmosphere as described in Sectionpffiffi2. ffi LEED pffiffiffi analysis indicated the presence of a ð 2 2ÞR45 superlattice. A representative STM image is shown in Fig. 3(a). step height has been measured between A1 A the terraces labelled 2 and 3 (as shown by the line profile in Fig. 3(b)). An atomic resolution image taken on the terrace labelled ‘‘4’’ is shown in Fig. 4. The structure exhibits a four-fold symmetry, typical of a tetrahedrally terminated surface. Therefore, terraces 3 and 4 are terminated at the A-plane. The white lines present in the smaller images in Fig. 4 denote missing rows along the [1 1 0] and [1 1 0] directions, which we attribute to missing tetrahedral Fe3þ cations. By including these vacancies in the tetrahedral surface planes, they exhibit a similar reconstruction pffiffiffi pffiffiffi to the octahedral planes, and the ð 2 2ÞR45 unit cell marked by a white square is easily identified. Although the A- and B-planes exhibit similar reconstructions, we believe that the nature of the two reconstructions is intrinsically different. As explained at the beginning of this section, we have gathered substantial pffiffiffi pffiffiffi evidence to support the claim that the ð 2 2ÞR45 superlattice observed at the B-planes is due to the ordering of electron charges at the B-sites. This cannot be the case for a tetrahedrally terminated surface, since 3þ A-sites pffiffiffi pare ffiffiffi occupied by Fe cations only. The ð 2 2ÞR45 mesh observed on the A-plane surface was explained by Chambers et al. as due to a missing Fe3þ cation per unit cell. Our results are in good agreement with this explanation. pffiffiffi pffiffiffi As summarized in Section 1, the ð 2 2ÞR45 reconstruction exhibited by the Fe3 O4 (0 0 1) surface has been attributed by some groups to A-planes at the surface and by other groups to B-
2 STM image, It ¼ 0:1 nA, Vb ¼ 1 V taken Fig. 3. 1100 · 1100 A with a MnNi tip. Terrace edges are aligned along the [1 1 0] and [1 1 0]. A line profile is taken along different terraces numbered are present except for the from 1 to 4. Steps heights of 2.1 A separation between terraces 2 and 3 where the step height with . Terraces 1 and 2 are termithe neighboring terrace is 1.05 A nated at the octahedral plane, while terraces 3 and 4 are terminated at the tetrahedral plane.
planes at the surface, leading to a certain degree of confusion. This is possibly due to the fact that the simultaneous presence of both terminations has been rarely observed. Our results provide convincing evidence that the type of termination depends on the preparation procedure. 3.2. Contaminated Fe3 O4 surface Annealing the sample for extended periods in UHV resulted in the diffusion of contaminants from the bulk to the surface. These contaminants
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
111
2 STM image, It ¼ 0:1 nA, Vb ¼ 1 V taken with a MnNi tip. Atomic rows run along the [1 1 0] and [1 1 0]. A white Fig. 4. 100 · 100 A pffiffiffi pffiffiffi square in the lower right-hand corner denotes the ð 2 2ÞR45 primitive unit cell. Missing rows can be seen along both the [1 1 0] and [1 1 0] directions on the same terrace and are indicated with white lines on the side images. This indicates that this terrace is terminated at the tetrahedral plane.
were typically calcium, potassium and sulphur, and were detected on both the natural and synthetic crystal surfaces. Ca and K in particular play a crucial role in the dynamics of the surface reconstruction of magnetite (1 0 0). These two impurities were consistently found on the surfaces of crystals obtained from different sources. A series of p(1 · 1), p(1 · 2), p(1 · 3) and p(1 · 4) surface reconstructions have been observed with increas-
Fig. 5. A series of p(1 · 1), p(1 · 2), p(1 · 3), p(1 · 4) surface reconstructions were observed on the Fe3 O4 (0 0 1) surface caused by the segregation of impurities from the bulk. The reconstructions are plotted as a function of the annealing time and of the Ca and K concentrations.
ing concentrations of Ca and K driven from the bulk to the surface by the continuous annealing. These are summarized in Fig. 5. 3.2.1. p(1 2) reconstruction AES analyis revealed a Ca concentration of 1.3%, while LEED analysis showed a streaked p(1 · 1) superlattice. The terrace edges are aligned along the [1 1 0] and [1 1 0] directions. Fig. 6(a) shows a high resolution zoom-in on a terrace. Closer inspection of the atomic rows reveals the presence of a p(1 · 2) surface reconstruction
2 STM atomic resolved image taken Fig. 6. (a) A 200 · 200 A with a Nickel tip, It ¼ 0:1 nA, Vb ¼ 1 V. The rows of atoms run 2 atomically resolved along the [1 1 0] direction. (b) 52 · 48 A STM image showing the p(1 · 2) reconstructed surface. A 6 · 12 2 square identified with the p(1 · 2) surface reconstruction is A outlined.
112
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
2 unit cell of the p(1 · 2) (Fig. 6(b)). The 12 · 6 A reconstruction is shown by a solid black square. Although a p(1 · 2) superlattice is seen on most of 2 unit cell has also been the surface areas, a 6 · 6 A found on a few patches of the surface. 3.2.2. p(1 3) reconstruction A different reconstruction was obtained when the Ca and K concentrations increased to 5.7% and 1.5%, respectively. The corresponding Auger spectrum is shown in Fig. 7. The LEED pattern displays satellite spots along the [1 1 0] and [1 1 0] directions. The satellite spots are marked with dark arrows and their separation is 1/3 of the separation between the primary spots, which cor in real space. The responds to a distance of 18 A error in the LEED measurements was calculated to be 6%, which was probably related to the position of the sample in the UHV drive and the camera optics (see Fig. 1(b)). STM images of the p(1 · 3) reconstruction are shown in Fig. 8(a) and (b). Atomic rows are visible along the [1 1 0] and [1 1 0] directions; the periodicity along these , and adjacent rows are separated by rows is 6 A 18 A. The rows are rotated by 90 on terraces , which proves separated by odd multiples of 2.1 A a B-plane terminated surface is imaged. From the studies of the clean Fe3 O4 (0 0 1) surface presented in Section 3.1.1, we conclude that the p(1 · 3) surface reconstruction is created by the diffusion of
Fig. 7. AES data show the presence of K and Ca contaminants on the surface. A Ca peak at 292 eV and a K peak at 249 eV are visible, corresponding to a 5.7% and 1.5% concentration in the near surface layers.
Fig. 8. (a) A 60 · 60 nm2 STM image of a p(1 · 3) reconstruction taken with a W tip, It ¼ 0:1 nA, Vb ¼ 1 V. Atomic rows are visible along the [1 1 0] and [1 1 0] directions; The rows are ro, tated by 90 on terraces separated by an odd multiple of 2.1 A which is evidence of a B-plane terminated surface. The rows are attributed to segregation of K on the surface. (b) 14 · 12.5 nm2 , and adjacent rows STM image showing a periodicity of 6 A . are separated by 18 A
Ca and K to the surface. The diffusion of metals such as Ca or K ions from the bulk to the surface, can be explained in terms of the size accommodation model [21]. The p(1 · 3) surface reconstruction is caused by the segregation of K and Ca ions that replace every second octahedrally coordinated Fe ion [22]. The (0 0 1) bulk terminated surface of magnetite is polar and is therefore intrinsically unstable [23,24]. A bulk terminated surface at a B-layer does not contain any Fetetra ions. Assuming an average charge of +2.5 for the Feoct ions, they share 2.5 e divided over six bonds, i.e. 5/12 e per bond. To fill O bonds with Feoct ions, each oxygen shares (2)5/12) ¼ 19/12 e . To fill the bond with the Fetetr ions, the oxygen shares (2)3/4) ¼ 5/4 e . If we consider a unit cell, it contains 4 Feoct ions and 8 oxygen atoms. The 4 Feoct ions have one dangling bond each. These bonds contain a total of (4 · 5/12) ¼ 1.67 e . Of the 8 oxygen atoms, four have one dangling octahedral bond and the other four have one dangling tetrahedral bond. Therefore, the O dangling bonds contain a total of (4 · 19/12) + (4 · 5/4) ¼ 11.33 e . The total number of electrons on the oxygen dangling bonds is (1.67 + 11.33) ¼ 13 e . This leaves the surface with a charge deficiency of 3 e , since the 8 O atoms at the surface require a total of 16 e to fill their
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
dangling bonds at an octahedral bulk terminated surface. In the case of a Kþ ion replacing an octahedrally co-ordinated Fe ion in the model proposed here for a p(1 · 3) surface reconstruction, the modified unit cell contains 5 Feoct atoms, 1 Koct atoms and 12 O atoms. Following the same electron counting procedure as previously described, a p(1 · 3) reconstructed surface has a lack of 4.75 e / unit cell. A non-autocompensated surface is also found for the case of Caþ2 replacing an octahedrally co-ordinated Fe ion. These considerations indicate that a surface exhibiting a p(1 · 3) reconstruction is not expected to be a stable one, and our experimental evidence support this model by showing that it is a metastable state. A p(1 · 3) surface reconstruction was observed on only a few occasions and after further annealing was replaced by a different reconstruction. 3.2.3. p(1 4) reconstruction For long annealing times, metallic K is gradually desorbed from the surface due to its high value of vapour pressure. The subsurface layer tends to be depleted from this impurity thus reducing the rate of its surface segregation. Longer annealing times also reduce the O/Fe ratio at the surface which leads to the formation of a p(1 · 4) surface reconstruction. Fig. 1(c) shows the LEED pattern recorded for this surface. Satellite spots are seen around the primary spots. AES spectra for the p(1 · 4) reconstructed surface, typically showed surface contamination levels of 6 1.5% for potas-
113
sium and 6% for calcium. The STM images showed well-defined trenches oriented along the [1 1 0] and [1 1 0] directions. The surface exhibits a two-fold symmetry since a rotation of the trenches by 90 was not observed on the same terrace but high steps. We only on terraces separated by 2.1 A can therefore conclude that the surface is B-layer terminated. The periodicity of the trenches on the . As annealing time terraces varies from 20 to 60 A is increased, the separation between the trenches becomes smaller and more regular. This is shown in the line profile of Fig. 9(a) where the terraces are periodicity between them equidistant with a 24 A (see Fig. 9(b)). The dynamics of this trench formation are discussed by Mariotto et al. [22], attributing the trenches to K and Ca segregation and to a reduction of the O/Fe ratio at the surface. The increased density of the trenches as a function of annealing time can be explained in terms of a size accommodation model. Continued annealing produces a Ca build-up on the surface to a point where the surface becomes saturated. To accommodate more Ca, the surface undergoes a gradual transformation resulting in the formation of trenches. This type of surface provides a greater surface area and correspondingly a larger number of surface sites to accommodate more Ca atoms with respect to the flat surface. This mechanism is analagous to that behind the oxygen-induced missing row structure observed in the initial stages of oxidation of the W(0 0 1) surface [25,26].
2 STM image It ¼ 0:1 nA, Vb ¼ 1 V. Regular trenches can be clearly seen on the surface. (b) Line profile corFig. 9. (a) 500 · 500 A periodicity between the trenches. responding to previous image showing a 24 A
114
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
An insight into the segregation mechanism of Ca atoms at the Fe3 O4 (0 0 1) surface is provided by atomically resolved STM images like the one shown in Fig. 10(a). Atomic rows oriented along periodicity, while the [1 1 0] direction exhibit a 6 A , instead of adjacent rows are separated by 5 A value expected for a bulk terminated surthe 6 A face. The maxima on adjacent rows are shifted by with respect to each other, giving rise to the 6 A
superlattice in Fig. 10(a). The line profile in Fig. 10(b) shows a large corrugation of the rows along the [1 1 0] direction. An average value of 0.4 A average was measured. This value is approximately twice the value of the corrugation measured on rows of Fe ions on the clean octahedrally terminated Fe3 O4 (0 0 1) surface. This large corrugation indicates that the atoms imaged on the narrow terraces are not Fe ions but, as indicated
2 STM image taken with a MnNi tip, It ¼ 0:1 nA, Vb ¼ 1 V. The atomic rows exhibit a 6 A periodicity along the Fig. 10. (a) A 70 · 70 A [1 1 0] direction. A 5 A separation is observed between rows along the [1 1 0] direction. The p(1 · 1) The superlattice is marked. (b) Line periodicity. (c) A schematic of the p(1 · 4) reconstruction. Oxygen vacancies are created profile along the [1 1 0] direction showing a 6 A at the surface by the extended anneal. The calcium ions replace every second Feoct ion. The p(1 · 1) and the superlattice are marked by solid and dashed lines respectively.
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
by the AES spectra, they are Ca or K atoms. It must be kept in mind that the clean and p(1 · 4) reconstructed surfaces were both scanned using the same type of tip (i.e. fabricated from MnNi alloy). Therefore, the difference in the corrugation cannot be ascribed to a tip effect, but is rather due to the different atomic species on the crystal surface. Although 1.5% of potassium was detected by AES, we propose a schematic model in Fig. 10(a) to explain the observed surface structure explicitly in terms of Ca atoms segregating at the surface. This approach is taken because STM analysis of 1 0 the surface shows the presence of a 0:5 1 incommensurate superlattice with an orthorhombic symmetry, which cannot be attributed to a potassium ferrite surface phase. The orthorhombic superlattice can be clearly seen in Fig. 10(a), with and b 2:5 A . Since lattice parameters of a 3 A CaFe2 O4 (calcium ferrite) belongs to the Pnam space group (rhombic/bipyramidal class) [27], we believe that a Ca1x Fe2þx O4 phase is present on the narrow terraces shown in Fig. 10(a). A schematic model of the reconstruction observed is illustrated in Fig. 10(c). Three rows of Ca cations can be identified on each of the narrow terraces shown in Fig. 10(a). On the sides of the Ca rows, less pronounced features can be observed. These features also form rows aligned along the [1 1 0] direction, periodicity and a corrugation of 0.2 with a 6 A . The periodicity and corrugation values indicate A that these rows are made of Fe ions in tetrahedral coordination. A further piece of evidence corroborating this explanation is given by a simple geometrical analysis of the STM images and can be appreciated from the model of Fig. 10(c). The Fetetr cations do not lie half-way between two neighboring Ca cations. By referring to Fig. 10(c), it is clear that the distance between the Fetetr cation , and the Ca cation marked ‘‘Ca-1’’ is about 5 A while the distance to the Ca ion marked ‘‘Ca-2’’ is . This is in very good agreement with the about 1 A values measured from our STM images. To check if our argument is sound, we have compared the concentration of Ca measured by AES with that predicted by our model, and we have found them in good agreement. A concentration of 4.7% cal-
115
culated from the model compares well with a 6% concentration detected by AES. Electron counting arguments show that oxygen vacancies in a B-layer would lower the polarity of the surface and make it more stable. Assuming that the oxygen vacancies are created at the O–Fetetr sites, 0.84 e /unit cell are missing to compensate the surface. This surface remains nonautocompensated, although it is more stable than the p(1 · 3) reconstructed surface or an octahedrally bulk terminated one. The p(1 · 4) surface reconstruction was observed by us on numerous occasions and it was indeed found to be a highly stable state. Without disregarding the autocompensation model proprosed by Pashley [28] and LaFemina [29] and followed after by many other authors to support their results, we remark that non-autocompensated polar-surfaces (including magnetite (0 0 1)) have been observed before [1,30– 33]. 3.2.4. p(1 1) reconstruction Further annealing of the crystals brings the Ca concentration up to a value of 9.5 at.% as shown by AES. No traces of any other contaminant are detected. At this stage a p(1 · 1) LEED pattern was observed. STM results showed that the terrace definition is almost completely lost.
4. Conclusions We have studied the (0 0 1) surface of natural and synthetic Fe3 O4 single crystals using a range of surface sensitive techniques, AES, LEED and STM. A detailed description of the preparation procedure leading to a contaminant-free magnetite surface is outlined. Diffusion of K and Ca bulk impurities and their surface segregation results in the formation of a set of surface reconstructions. The main results can be summarized as follows: pffiffiffi pffiffiffi (1) A ð 2 2ÞR45 reconstruction has been observed on a clean magnetite surface. It was found that the surface terminates at the B-plane when annealed in oxygen partial pressure, while co-existing A- and B-terminations were observed when annealing in hydrogen partial pressure.
116
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
(2) Annealing of the contaminant-free crystals in UHV induces segregation of Ca and K contaminants from the bulk to the surface. For low concentrations (less than 2%) of contaminants a p(1 · 2) surface reconstruction was observed. A p(1 · 3) reconstruction resulted after further annealing with an increase of the Ca and K on the surface. No reduction of the O/Fe ratio have been observed. After long annealing sessions the p(1 · 3) reconstruction was replaced by the p(1 · 4) reconstruction. The O/Fe ratio has decreased as function of the annealing time. A concentration of Ca and K similar to the p(1 · 3) reconstruction was measured. We can conclude that the surface reconstructions of Fe3 O4 are induced on one hand by the contamination level on the surface and the O/Fe ratio on the other hand. For a high level of contamination a blurred p(1 · 1) LEED mesh was observe and STM images showed a surface where topographic features can hardly be distinguished. To summarize, we have shown that the (0 0 1) surface of magnetite is highly sensitive to the treatment conditions. In particular, the surface reconstructions of magnetite strongly depend on alkaline and alkaline-earth metals segregating from the bulk to the surface and on the O/Fe ratio. We have observed a p(1 · 3) reconstruction that does not satisfy the electron counting argument. Although this reconstruction is a metastable state of the surface, this brings into question the principle of autocompensation and its universal applicability since other terminations were reported that are not autocompensated. Acknowledgements Financial assistance from Science Foundation Ireland (SFI) under contract no. 00/PI.1/C042 is gratefully acknowledged. References [1] R. Wiesendanger, I.V. Shvets, D. B€ urgler, G. Tarrach, H.J. G€ untherodt, J.M.D. Coey, S. Gr€aser, Science 255 (1992) 583.
[2] A. Gupta, J.Z. Sun, J. Magn. Mater. 24–43 (1999) 200. [3] J.F. Gregg, I. Petej, E. Jouguelet, C. Dennis, J. Phys. D 35 (2002) R121–R155. [4] P.A. Cox, Transition Metal Oxides, An Introduction to their Electronic Structure and Properties, Clarendon Press, Oxford, 1995. [5] R. Aragon, R.J. Rasmussen, J.P. Shepherd, J.W. Koenitzer, J.M. Honig, J. Magn. Magn. Mater. 47–54 (1986) 1335. [6] F.C. Voogt, T. Fujii, P.J.M. Smulders, L. Niesen, M.A. James, T. Hibma, Phys. Rev. B 60 (1999) 11193. [7] G. Tarrach, D. B€ urgler, T. Schaub, R. Wiesendanger, H. G€ untherodt, Surf. Sci. 285 (1993) 1. [8] G. Mariotto, S. Murphy, I.V. Shvets, Phys. Rev. B 66 (2002) 245426. [9] S.A. Chambers, S. Thevuthasan, S.A. Joyce, Surf. Sci. 450 (2000) L273. [10] B. Stanka, W. Hebenstreit, U. Diebold, S.A. Chambers, Surf. Sci. 448 (2000) 49. [11] J.F. Anderson, M. Kuhn, H. Diebold, K. Shaw, P. Stoyanov, D. Lind, Phys. Rev. B 56 (1997) 9902. [12] K.A. Shaw, E. Lochner, D.M. Lind, J. Appl. Phys. 87 (2000) 1727. [13] S.K. Shaikhutdinov, W. Weiss, R. Sch€ ogl, Appl. Surf. Sci. 161 (2000) 497. [14] L.P. Zhang, M. Li, U. Diebold, Surf. Sci. 412/413 (1998) 242. [15] H. N€ orenberg, J.H. Harding, Surf. Sci. 473 (2001) 151. [16] Y. Gao, S.A. Chambers, J. Cryst. Growth 174 (1997) 446. [17] Prof. Honig, Purdue University, Indiana, USA. [18] S. Murphy, J. Osing, I.V. Shvets, J. Magn. Magn. Mater. 198–199 (1999) 686. [19] S.F. Ceballos, G. Mariotto, S. Murphy, I.V. Shvets, Surf. Sci. 523 (2003) 131. [20] K. Jordan, S.F. Ceballos, G. Mariotto, S. Murphy, I.V. Shvets, to be published. [21] A.P. Sutton, R.W. Balluffi, Interfaces in Crystalline Materials, Oxford University Press, 1996. [22] G. Mariotto, S.F. Ceballos, S. Murphy, C. Seoighe, N. Berdunov, I.V. Shvets, Phys. Rev. B, submitted for publication. [23] P.W. Tasker, J. Phys. C 12 (1979) 4977. [24] P.W. Tasker, Philos. Mag. A 12 (1979) 4977. [25] J.A. Meyer, Y. Kuk, P.J. Estrup, P.J. Silverman, Phys. Rev. B 44 (1991) 9104. [26] M.S. Altman, E. Bauer, Surf. Sci. 347 (1996) 265. [27] P.M. Hill, H.S. Peiser, J.R. Rait, Acta Cryst. 9 (1956) 981. [28] M.D. Pashley, Phys. Rev. B 40 (1989) 10481. [29] J. LaFemina, Crit. Rev. Surf. Chem. 3 (1994) 297. [30] O. Warren, P. Thiel, J. Chem. Phys. 100 (1994) 659. [31] A. Barbieri, W. Weiss, M.V. Hove, G. Somorjai, Surf. Sci. 302 (1994) 259. [32] Q. Guo, P. Moeller, Surf. Sci. 340 (1995) L999. [33] N. Condon, F. Leibsle, P. Murray, T. Parker, A. Lennie, D. Vaughan, G. Thornton, Surf. Sci. 397 (1998) 278.
An atomic scale STM study of the Fe3O4(0 0 1) surface S.F. Ceballos *, G. Mariotto, K. Jordan, S. Murphy, C. Seoighe, I.V. Shvets SFI Nanoscience Laboratory, Department of Physics, Trinity College, Dublin 2, Ireland Received 5 August 2003; accepted for publication 27 October 2003
Abstract Despite the intensive investigation into the electronic properties of magnetite, fundamental issues related to the Verwey transition and the electronic transport mechanism are not fully understood. These issues are further complicated at the surface of magnetite crystals, due to the large number of possible surface terminations. The preparation procedure plays a fundamental role in determining the O/Fe ratio, and therefore the electronic properties of a magnetite crystal. We present a detailed investigation of the influence of the preparation conditions on the morphology of Fe3 O4 (0 0 1) single crystal surfaces using AES, LEED, and STM. We show that long anneals of single crystals in UHV cause segregation of contaminants to the surface and that a series reconstructions is induced. A different pffiffiffiof surface pffiffiffi preparation procedure gives rise to a clean surface exhibiting a ð 2 2ÞR45 reconstruction. This surface is terminated at the octahedral plane and has been imaged down to the atomic scale. This provides a useful test system to study the Verwey transition at the surface. 2003 Elsevier B.V. All rights reserved. Keywords: Scanning tunneling microscopy; Iron oxide; Low energy electron diffraction (LEED); Auger electron spectroscopy
1. Introduction Magnetite has been the subject of intensive studies by the scientific community during the last decades. Efforts to understand its magnetic and electrical properties have intensified in the past few years due to its half-metallic nature, which makes magnetite an interesting material for spin electronics applications. Spin electronics almost invariably involves electron transport across or along an interface, so that the understanding of electron transport in low dimensional systems is
*
Corresponding author. Tel.: +353-1-608-2020; fax: +353-1671-1759. E-mail address: [email protected] (S.F. Ceballos).
therefore of critical importance. Several examples of electronic and spin-electronic surface studies can be found in [1–3]. Stoichiometric magnetite undergoes a metal– insulator transition at around 125 K, known as the Verwey transition temperature TV . At room temperature, the electrical conductivity of Fe3 O4 is 200 X1 cm1 and it gradually decreases with decreasing temperature. When cooled down below TV , the conductivity abruptly decreases by about two orders of magnitude [4,5]. The change of conductivity is accompanied by a change in the crystallographic structure, whose symmetry is lowered from cubic to monoclinic. Magnetite is an inverse spinel material. The crystal structure is based on a face-centered cubic (f.c.c.) unit cell, containing 32 O2 anions and 24
0039-6028/$ - see front matter 2003 Elsevier B.V. All rights reserved. doi:10.1016/j.susc.2003.10.041
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
mixed valence Fe cations, with a lattice parameter . The formula can be written as of a ¼ 8:3963 A YA ½XY B O4 , where X ¼ Fe2þ , Y ¼ Fe3þ and A and B denote tetrahedral and octahedral sites, respectively. This formula indicates that one half of the ferric Fe3þ cations occupies 8 of the 64 available tetrahedral interstices, and the other half of the ferric cations, together with an equal amount of ferrous Fe2þ cations, occupy 16 of the 32 available octahedral interstices. The (0 0 1) plane of magnetite can be viewed as a stacking sequence of two alternating layers. The A-layer contains tetrahedrally coordinated Fe3þ ions, while the B-layer is composed of rows of octahedrally coordinated Fe2þ and Fe3þ ions surrounded by oxygen ions. The separation between neighboring planes (i.e. the A–B interplanar sep, while the separation between aration) is 1.05 A successive like planes (i.e. the A–A or B–B inter. In each octahedral planar separation) is 2.10 A plane, the nearest-neighbor B-site cations form rows that run along the [1 1 0], [1 1 0], [1 0 1], [1 0 1], [0 1 1] and [0 1 1] directions. The rows in successive octahedral planes are rotated by 90 with respect to one another, giving these planes a two-fold rotational symmetry. In contrast, the arrangement of cations in the A-layers give them a four-fold rotational symmetry. Both A- and B-terminated surfaces have been reported in the literature [6–8], with no satisfactory explanation of why the (0 0 1) magnetite surface should be terminated at either plane. The preparation conditions seem to play a crucial inffiffiffideterprole ffiffiffi p mining the surface termination. A ð 2 2ÞR45 clean surface reconstruction has been observed by several groups on both natural and synthetic single crystals, and on thin films grown by molecular beam epitaxy (MBE). Tarrach et al. [7] have suggested that the top-most surface layer consists of a full monolayer of tetrahedral Fe ions. Chambers et al. [9] used X-ray photoelectron spectroscopy (XPS), X-ray photoelectron diffraction (XPD) and scanning tunneling microscopy (STM) to support the conclusion that the Fe3 O4 (0 0 1) surface is constituted by half a monolayer of tetrahedral Fe ions and that this surface is charge compensated. Wiesendanger et al. [1] imaged two distinct structures on different areas of a natural crystal of
107
magnetite; the two structures were attributed to bulk terminated tetrahedral and octahedral terminations of the surface. In contrast, Voogt et al. [6] and more recently Stanka et al. [10] have proposed a B-termination layer on a single crystal and a Fe3 O4 /MgO(0 0 1) film, respectively. In both cases, they have proposed that charge compensation is achieved by an array of oxygen vacancies accompanied by a change in the valence p state ffiffiffi of the Fe. Mariotto et al. [8] observed a ð 2 pffiffiffi 2ÞR45 reconstruction on a clean magnetite (0 0 1) single crystal surface and proposed a B-layer surface termination. They further proposed that the reconstruction is due to charge ordering of the Fe cations in octahedral positions, and not to an ordered array of vacancies as proposed by previous studies. In the present study, a contaminant-free magnetite surface was obtained using the procedure described in Section 3.1.1. The clean pffiffiffi magnetite pffiffiffi surface was found to exhibit a ð 2 2ÞR45 superlattice. Subsequent annealing for long periods in UHV lead to the segregation of contaminants to the surface. The typical contaminants observed in our crystals were potassium, sulphur and calcium. Continuous annealing of the sample results in the removal of K and S from the surface and an increase in the Ca concentration. Our data shows that the (0 0 1) contaminated magnetite surface is usually terminated at the octahedral plane (B-layer). We have observed p(1 · 1), p(1 · 2), p(1 · 3) and p(1 · 4) reconstructions on the Fe3 O4 (0 0 1) surface. Surface reconstructions induced by alkaline and alkaline-earth metals have been observed on a range of different systems, such as thin films of Fe3 O4 grown on MgO(0 0 1) substrates [11,12], thin films of Fe3 O4 grown on Pt(1 1 1) substrates [13] and TiO2 single crystals [14,15]. A p(1 · 4) reconstruction was observed by Anderson et al. [11] on a 1 lm thick film of magnetite grown on a MgO(0 0 1) substrate. A similar effect was observed by Voogt et al. [6] who reported the onset of a p(1 · 3) reconstruction on a magnetite thin film grown on MgO(0 0 1). Anderson et al. attributed the surface reconstruction to the segregation of magnesium from the substrate, leading to the formation of a MgFe2 O4 phase. Gao and Chambers [16] in a study of magnetite thin
108
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
films grown on MgO(0 0 1) suggested that the Mg ions diffuse through the iron oxide via vacancies in the octahedral cation sublattice. Once they reach the surface, they decorate the rows of octahedral B-site cations. N€ orenberg and Harding [15] have investigated the Ca-induced surface reconstruction on TiO2 . They have observed a variety of surface reconstructions, including a p(1 · 3) and a p(1 · 4). They attributed the p(1 · 3) reconstruction to Ca2þ cations replacing the Ti cations at the surface of the crystal. A p(1 · 4) reconstruction, caused by an ordered array of trenches, was found to succeed the p(1 · 3) reconstruction and it was attributed to oxygen loss from the surface.
2. Experimental A set of synthetic and natural single crystals has been used in these experiments. The natural crystals were cut from the single crystalline nugget used by Tarrach et al. [7], which originated from Zillertal, Austria. The crystals were aligned with a precision of ±1 with respect to the (0 0 1) plane and were oriented along the [0 1 0] direction. The synthetic crystals were grown employing the skull melting technique [17]. The crystals were first characterized by powder X-ray diffraction. The diffraction patterns showed good agreement with the reference spectra for magnetite, and lattice constants of 8.398 ± 0.010 and 8.406 ± 0.010 A were measured for the synthetic and natural crystals, respectively. Four-point resistance vs. temperature measurements were also performed; Verwey transition temperatures of 108 and 98 K were measured for the synthetic and natural crystals, respectively. The crystals were mechanically polished using diamond paste with grain size down to 0.25 lm before being introduced into the UHV system. This system is equipped with facilities for low energy electron diffraction (LEED), Auger electron spectroscopy (AES) and STM analysis. A detailed description of the UHV system is available elsewhere [8]. STM measurements were carried out at room temperature in constant current mode using a home-built instrument. A SCALA controller by OMICRON GmbH was used. This controller allows the user to compensate for the
thermal drift using a topographic feature of the STM image as a reference point. A 0.1 nm/s thermal drift was measured when no correction was applied. By enabling this option, thermal drift was reduced to 0.0025 nm/s. A bias voltage between +0.6 and +1 V was applied to the sample and a tunneling current of between 0.1 and 0.3 nA was typically used. A stable tunneling current could not be obtained under negative bias conditions. We have developed two different in-vacuum preparation procedures to obtain a clean surface. Both procedures lead pffiffiffi topaffiffiffi contaminant-free surface, exhibiting a ð 2 2ÞR45 superlattice, but with different surface terminations. The first procedure consisted in annealing the crystals in UHV at 990 ± 10 K. This has a two-fold effect of reducing the surface and causing the segregation of impurities at the surface. The impurities were removed from the surface by Arþ sputtering for 10 min at 1 keV (Itarget 18 lA). The crystals were then annealed in an oxygen partial pressure (the typical exposure varied between 3600 and 7200 L) to compensate for the reduction of the surface caused by the long UHV annealing. The crystals were finally annealed in UHV at the same temperature for a short period of time (typically 2–8 h). It was found that this preparation procedure consistently led to a B-terminated surface. The second procedure is identical to the first one, with the exception of annealing the crystals in hydrogen (typical exposure of 2000 L) instead of oxygen. This led to co-existing A- and B-terminations. Further annealing of the crystals in UHV at 990 ± 10 K ranging from 10 to 70 h led to a series of reconstructions such as p(1 · 2), p(1 · 3) and p(1 · 4). Three different types of STM tips made out of Ni, W and MnNi have been used. Tungsten tips are standard tips for most STM applications, although magnetic tips are more attractive due to their potential applicability to Spin Polarized STM (SP-STM) measurements [18,19]. Nickel, Tungsten and Manganese–Nickel tips have been used to image the p(1 · 2), p(1 · 3) and p(1 · 4) reconstructed surfaces, respectively. In this paper, the effect that tips with different electronic properties might have on the STM images has not been
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
109
pffiffiffi pffiffiffi 3.1. Clean Fe3 O4 surface––ð 2 2ÞR45 superlattice
, correheights that are integer multiples of 2.1 A sponding to the separation between A–A or B–B planes. However, we have determined that the surface is terminated at the B-plane, since a 90 rotation of the iron rows has been observed on [8]. neighboring terraces separated by 2.1 A Fig. 2 shows an atomically resolved STM image. periodicity A highly regular structure with a 12 A along the [1 1 0] direction is visible. Adjacent atomic . This structure was rows are separated by 6 A found to extend over a length of the order of about 40 unit cells, suggesting that long-range order had set in on the surface. An extensive analysis of these results is given by Mariotto et al. [8]. The
3.1.1. B-termination AES, LEED and STM were performed on the B-terminated, which was obtained using the procedure described in Section 2. AES analysis of this surface indicated that it was contaminant-free, while the O(509 eV)/Fe(700 eV) ratio and the Fe M2;3 VV peak shape of the AES spectra provide strong evidence that no other iron oxide phase but magnetite was present. A typical LEED pattern of the surface is shown in Fig. order pffiffi1(a). ffi pFractional ffiffiffi spots corresponding top affiffiðffi 2pffiffiffi 2ÞR45 mesh are clearly visible. The ð 2 2ÞR45 superlattice was routinely reproduced following the preparation conditions described above. The STM images are characterized by multiple rectangular terraces with step-edges oriented along the [1 1 0] and [1 1 0] directions. The terraces are separated by step
2 STM image. Rows of Fe cations aligned Fig. 2. 140 · 210 A along the [1 1 0] direction are visible. Adjacent rows are sepa. rated by 6 A
addressed; only a detailed description of the topography and morphology of the surface is given. Further experiments have been carried out on the magnetite surface using different tip materials by our group. Differences in the STM images of the clean magnetite surface when scanning with magnetic and non-magnetic tips are discussed elsewhere [20].
3. Results and discussion
Fig. pffiffiffi 1.p(a) ffiffiffi LEED pattern of a clean Fe3 O4 (0 0 1) surface taken with a primary electron energy of 78 eV. The p(1 · 1) unit cell and the ð 2 2ÞR45 superlattice are highlighted. (b) LEED pattern taken with a primary electron energy of 47 eV. Satellite spots around the primary spots are clearly seen and marked with white arrows along the [1 1 0] and [1 1 0] directions. Their separation is 1/3 of that between the integral order spots. (c) LEED pattern recorded for a p(1 · 4) surface reconstruction. The primitive unit cell is indicated with a white square labelled ‘‘1’’. The p(1 · 4) unit cell is marked with the white square labelled ‘‘2’’.
110
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
pffiffiffi pffiffiffi ð 2 2ÞR45 superlattice observed by LEED and STM was attributed to an ordering of electron charges, with the formation of Fe2þ –Fe2þ and Fe3þ –Fe3þ dimers at the B-sites. We emphasize that such long-range order is established at room temperature suggesting that the metal–insulator transition on the surface occurs at a temperature that is well above the bulk temperature TV . 3.1.2. Co-existing A- and B-terminations Co-existing A- and B-terminations were obtained by annealing in a hydrogen atmosphere as described in Sectionpffiffi2. ffi LEED pffiffiffi analysis indicated the presence of a ð 2 2ÞR45 superlattice. A representative STM image is shown in Fig. 3(a). step height has been measured between A1 A the terraces labelled 2 and 3 (as shown by the line profile in Fig. 3(b)). An atomic resolution image taken on the terrace labelled ‘‘4’’ is shown in Fig. 4. The structure exhibits a four-fold symmetry, typical of a tetrahedrally terminated surface. Therefore, terraces 3 and 4 are terminated at the A-plane. The white lines present in the smaller images in Fig. 4 denote missing rows along the [1 1 0] and [1 1 0] directions, which we attribute to missing tetrahedral Fe3þ cations. By including these vacancies in the tetrahedral surface planes, they exhibit a similar reconstruction pffiffiffi pffiffiffi to the octahedral planes, and the ð 2 2ÞR45 unit cell marked by a white square is easily identified. Although the A- and B-planes exhibit similar reconstructions, we believe that the nature of the two reconstructions is intrinsically different. As explained at the beginning of this section, we have gathered substantial pffiffiffi pffiffiffi evidence to support the claim that the ð 2 2ÞR45 superlattice observed at the B-planes is due to the ordering of electron charges at the B-sites. This cannot be the case for a tetrahedrally terminated surface, since 3þ A-sites pffiffiffi pare ffiffiffi occupied by Fe cations only. The ð 2 2ÞR45 mesh observed on the A-plane surface was explained by Chambers et al. as due to a missing Fe3þ cation per unit cell. Our results are in good agreement with this explanation. pffiffiffi pffiffiffi As summarized in Section 1, the ð 2 2ÞR45 reconstruction exhibited by the Fe3 O4 (0 0 1) surface has been attributed by some groups to A-planes at the surface and by other groups to B-
2 STM image, It ¼ 0:1 nA, Vb ¼ 1 V taken Fig. 3. 1100 · 1100 A with a MnNi tip. Terrace edges are aligned along the [1 1 0] and [1 1 0]. A line profile is taken along different terraces numbered are present except for the from 1 to 4. Steps heights of 2.1 A separation between terraces 2 and 3 where the step height with . Terraces 1 and 2 are termithe neighboring terrace is 1.05 A nated at the octahedral plane, while terraces 3 and 4 are terminated at the tetrahedral plane.
planes at the surface, leading to a certain degree of confusion. This is possibly due to the fact that the simultaneous presence of both terminations has been rarely observed. Our results provide convincing evidence that the type of termination depends on the preparation procedure. 3.2. Contaminated Fe3 O4 surface Annealing the sample for extended periods in UHV resulted in the diffusion of contaminants from the bulk to the surface. These contaminants
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
111
2 STM image, It ¼ 0:1 nA, Vb ¼ 1 V taken with a MnNi tip. Atomic rows run along the [1 1 0] and [1 1 0]. A white Fig. 4. 100 · 100 A pffiffiffi pffiffiffi square in the lower right-hand corner denotes the ð 2 2ÞR45 primitive unit cell. Missing rows can be seen along both the [1 1 0] and [1 1 0] directions on the same terrace and are indicated with white lines on the side images. This indicates that this terrace is terminated at the tetrahedral plane.
were typically calcium, potassium and sulphur, and were detected on both the natural and synthetic crystal surfaces. Ca and K in particular play a crucial role in the dynamics of the surface reconstruction of magnetite (1 0 0). These two impurities were consistently found on the surfaces of crystals obtained from different sources. A series of p(1 · 1), p(1 · 2), p(1 · 3) and p(1 · 4) surface reconstructions have been observed with increas-
Fig. 5. A series of p(1 · 1), p(1 · 2), p(1 · 3), p(1 · 4) surface reconstructions were observed on the Fe3 O4 (0 0 1) surface caused by the segregation of impurities from the bulk. The reconstructions are plotted as a function of the annealing time and of the Ca and K concentrations.
ing concentrations of Ca and K driven from the bulk to the surface by the continuous annealing. These are summarized in Fig. 5. 3.2.1. p(1 2) reconstruction AES analyis revealed a Ca concentration of 1.3%, while LEED analysis showed a streaked p(1 · 1) superlattice. The terrace edges are aligned along the [1 1 0] and [1 1 0] directions. Fig. 6(a) shows a high resolution zoom-in on a terrace. Closer inspection of the atomic rows reveals the presence of a p(1 · 2) surface reconstruction
2 STM atomic resolved image taken Fig. 6. (a) A 200 · 200 A with a Nickel tip, It ¼ 0:1 nA, Vb ¼ 1 V. The rows of atoms run 2 atomically resolved along the [1 1 0] direction. (b) 52 · 48 A STM image showing the p(1 · 2) reconstructed surface. A 6 · 12 2 square identified with the p(1 · 2) surface reconstruction is A outlined.
112
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
2 unit cell of the p(1 · 2) (Fig. 6(b)). The 12 · 6 A reconstruction is shown by a solid black square. Although a p(1 · 2) superlattice is seen on most of 2 unit cell has also been the surface areas, a 6 · 6 A found on a few patches of the surface. 3.2.2. p(1 3) reconstruction A different reconstruction was obtained when the Ca and K concentrations increased to 5.7% and 1.5%, respectively. The corresponding Auger spectrum is shown in Fig. 7. The LEED pattern displays satellite spots along the [1 1 0] and [1 1 0] directions. The satellite spots are marked with dark arrows and their separation is 1/3 of the separation between the primary spots, which cor in real space. The responds to a distance of 18 A error in the LEED measurements was calculated to be 6%, which was probably related to the position of the sample in the UHV drive and the camera optics (see Fig. 1(b)). STM images of the p(1 · 3) reconstruction are shown in Fig. 8(a) and (b). Atomic rows are visible along the [1 1 0] and [1 1 0] directions; the periodicity along these , and adjacent rows are separated by rows is 6 A 18 A. The rows are rotated by 90 on terraces , which proves separated by odd multiples of 2.1 A a B-plane terminated surface is imaged. From the studies of the clean Fe3 O4 (0 0 1) surface presented in Section 3.1.1, we conclude that the p(1 · 3) surface reconstruction is created by the diffusion of
Fig. 7. AES data show the presence of K and Ca contaminants on the surface. A Ca peak at 292 eV and a K peak at 249 eV are visible, corresponding to a 5.7% and 1.5% concentration in the near surface layers.
Fig. 8. (a) A 60 · 60 nm2 STM image of a p(1 · 3) reconstruction taken with a W tip, It ¼ 0:1 nA, Vb ¼ 1 V. Atomic rows are visible along the [1 1 0] and [1 1 0] directions; The rows are ro, tated by 90 on terraces separated by an odd multiple of 2.1 A which is evidence of a B-plane terminated surface. The rows are attributed to segregation of K on the surface. (b) 14 · 12.5 nm2 , and adjacent rows STM image showing a periodicity of 6 A . are separated by 18 A
Ca and K to the surface. The diffusion of metals such as Ca or K ions from the bulk to the surface, can be explained in terms of the size accommodation model [21]. The p(1 · 3) surface reconstruction is caused by the segregation of K and Ca ions that replace every second octahedrally coordinated Fe ion [22]. The (0 0 1) bulk terminated surface of magnetite is polar and is therefore intrinsically unstable [23,24]. A bulk terminated surface at a B-layer does not contain any Fetetra ions. Assuming an average charge of +2.5 for the Feoct ions, they share 2.5 e divided over six bonds, i.e. 5/12 e per bond. To fill O bonds with Feoct ions, each oxygen shares (2)5/12) ¼ 19/12 e . To fill the bond with the Fetetr ions, the oxygen shares (2)3/4) ¼ 5/4 e . If we consider a unit cell, it contains 4 Feoct ions and 8 oxygen atoms. The 4 Feoct ions have one dangling bond each. These bonds contain a total of (4 · 5/12) ¼ 1.67 e . Of the 8 oxygen atoms, four have one dangling octahedral bond and the other four have one dangling tetrahedral bond. Therefore, the O dangling bonds contain a total of (4 · 19/12) + (4 · 5/4) ¼ 11.33 e . The total number of electrons on the oxygen dangling bonds is (1.67 + 11.33) ¼ 13 e . This leaves the surface with a charge deficiency of 3 e , since the 8 O atoms at the surface require a total of 16 e to fill their
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
dangling bonds at an octahedral bulk terminated surface. In the case of a Kþ ion replacing an octahedrally co-ordinated Fe ion in the model proposed here for a p(1 · 3) surface reconstruction, the modified unit cell contains 5 Feoct atoms, 1 Koct atoms and 12 O atoms. Following the same electron counting procedure as previously described, a p(1 · 3) reconstructed surface has a lack of 4.75 e / unit cell. A non-autocompensated surface is also found for the case of Caþ2 replacing an octahedrally co-ordinated Fe ion. These considerations indicate that a surface exhibiting a p(1 · 3) reconstruction is not expected to be a stable one, and our experimental evidence support this model by showing that it is a metastable state. A p(1 · 3) surface reconstruction was observed on only a few occasions and after further annealing was replaced by a different reconstruction. 3.2.3. p(1 4) reconstruction For long annealing times, metallic K is gradually desorbed from the surface due to its high value of vapour pressure. The subsurface layer tends to be depleted from this impurity thus reducing the rate of its surface segregation. Longer annealing times also reduce the O/Fe ratio at the surface which leads to the formation of a p(1 · 4) surface reconstruction. Fig. 1(c) shows the LEED pattern recorded for this surface. Satellite spots are seen around the primary spots. AES spectra for the p(1 · 4) reconstructed surface, typically showed surface contamination levels of 6 1.5% for potas-
113
sium and 6% for calcium. The STM images showed well-defined trenches oriented along the [1 1 0] and [1 1 0] directions. The surface exhibits a two-fold symmetry since a rotation of the trenches by 90 was not observed on the same terrace but high steps. We only on terraces separated by 2.1 A can therefore conclude that the surface is B-layer terminated. The periodicity of the trenches on the . As annealing time terraces varies from 20 to 60 A is increased, the separation between the trenches becomes smaller and more regular. This is shown in the line profile of Fig. 9(a) where the terraces are periodicity between them equidistant with a 24 A (see Fig. 9(b)). The dynamics of this trench formation are discussed by Mariotto et al. [22], attributing the trenches to K and Ca segregation and to a reduction of the O/Fe ratio at the surface. The increased density of the trenches as a function of annealing time can be explained in terms of a size accommodation model. Continued annealing produces a Ca build-up on the surface to a point where the surface becomes saturated. To accommodate more Ca, the surface undergoes a gradual transformation resulting in the formation of trenches. This type of surface provides a greater surface area and correspondingly a larger number of surface sites to accommodate more Ca atoms with respect to the flat surface. This mechanism is analagous to that behind the oxygen-induced missing row structure observed in the initial stages of oxidation of the W(0 0 1) surface [25,26].
2 STM image It ¼ 0:1 nA, Vb ¼ 1 V. Regular trenches can be clearly seen on the surface. (b) Line profile corFig. 9. (a) 500 · 500 A periodicity between the trenches. responding to previous image showing a 24 A
114
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
An insight into the segregation mechanism of Ca atoms at the Fe3 O4 (0 0 1) surface is provided by atomically resolved STM images like the one shown in Fig. 10(a). Atomic rows oriented along periodicity, while the [1 1 0] direction exhibit a 6 A , instead of adjacent rows are separated by 5 A value expected for a bulk terminated surthe 6 A face. The maxima on adjacent rows are shifted by with respect to each other, giving rise to the 6 A
superlattice in Fig. 10(a). The line profile in Fig. 10(b) shows a large corrugation of the rows along the [1 1 0] direction. An average value of 0.4 A average was measured. This value is approximately twice the value of the corrugation measured on rows of Fe ions on the clean octahedrally terminated Fe3 O4 (0 0 1) surface. This large corrugation indicates that the atoms imaged on the narrow terraces are not Fe ions but, as indicated
2 STM image taken with a MnNi tip, It ¼ 0:1 nA, Vb ¼ 1 V. The atomic rows exhibit a 6 A periodicity along the Fig. 10. (a) A 70 · 70 A [1 1 0] direction. A 5 A separation is observed between rows along the [1 1 0] direction. The p(1 · 1) The superlattice is marked. (b) Line periodicity. (c) A schematic of the p(1 · 4) reconstruction. Oxygen vacancies are created profile along the [1 1 0] direction showing a 6 A at the surface by the extended anneal. The calcium ions replace every second Feoct ion. The p(1 · 1) and the superlattice are marked by solid and dashed lines respectively.
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
by the AES spectra, they are Ca or K atoms. It must be kept in mind that the clean and p(1 · 4) reconstructed surfaces were both scanned using the same type of tip (i.e. fabricated from MnNi alloy). Therefore, the difference in the corrugation cannot be ascribed to a tip effect, but is rather due to the different atomic species on the crystal surface. Although 1.5% of potassium was detected by AES, we propose a schematic model in Fig. 10(a) to explain the observed surface structure explicitly in terms of Ca atoms segregating at the surface. This approach is taken because STM analysis of 1 0 the surface shows the presence of a 0:5 1 incommensurate superlattice with an orthorhombic symmetry, which cannot be attributed to a potassium ferrite surface phase. The orthorhombic superlattice can be clearly seen in Fig. 10(a), with and b 2:5 A . Since lattice parameters of a 3 A CaFe2 O4 (calcium ferrite) belongs to the Pnam space group (rhombic/bipyramidal class) [27], we believe that a Ca1x Fe2þx O4 phase is present on the narrow terraces shown in Fig. 10(a). A schematic model of the reconstruction observed is illustrated in Fig. 10(c). Three rows of Ca cations can be identified on each of the narrow terraces shown in Fig. 10(a). On the sides of the Ca rows, less pronounced features can be observed. These features also form rows aligned along the [1 1 0] direction, periodicity and a corrugation of 0.2 with a 6 A . The periodicity and corrugation values indicate A that these rows are made of Fe ions in tetrahedral coordination. A further piece of evidence corroborating this explanation is given by a simple geometrical analysis of the STM images and can be appreciated from the model of Fig. 10(c). The Fetetr cations do not lie half-way between two neighboring Ca cations. By referring to Fig. 10(c), it is clear that the distance between the Fetetr cation , and the Ca cation marked ‘‘Ca-1’’ is about 5 A while the distance to the Ca ion marked ‘‘Ca-2’’ is . This is in very good agreement with the about 1 A values measured from our STM images. To check if our argument is sound, we have compared the concentration of Ca measured by AES with that predicted by our model, and we have found them in good agreement. A concentration of 4.7% cal-
115
culated from the model compares well with a 6% concentration detected by AES. Electron counting arguments show that oxygen vacancies in a B-layer would lower the polarity of the surface and make it more stable. Assuming that the oxygen vacancies are created at the O–Fetetr sites, 0.84 e /unit cell are missing to compensate the surface. This surface remains nonautocompensated, although it is more stable than the p(1 · 3) reconstructed surface or an octahedrally bulk terminated one. The p(1 · 4) surface reconstruction was observed by us on numerous occasions and it was indeed found to be a highly stable state. Without disregarding the autocompensation model proprosed by Pashley [28] and LaFemina [29] and followed after by many other authors to support their results, we remark that non-autocompensated polar-surfaces (including magnetite (0 0 1)) have been observed before [1,30– 33]. 3.2.4. p(1 1) reconstruction Further annealing of the crystals brings the Ca concentration up to a value of 9.5 at.% as shown by AES. No traces of any other contaminant are detected. At this stage a p(1 · 1) LEED pattern was observed. STM results showed that the terrace definition is almost completely lost.
4. Conclusions We have studied the (0 0 1) surface of natural and synthetic Fe3 O4 single crystals using a range of surface sensitive techniques, AES, LEED and STM. A detailed description of the preparation procedure leading to a contaminant-free magnetite surface is outlined. Diffusion of K and Ca bulk impurities and their surface segregation results in the formation of a set of surface reconstructions. The main results can be summarized as follows: pffiffiffi pffiffiffi (1) A ð 2 2ÞR45 reconstruction has been observed on a clean magnetite surface. It was found that the surface terminates at the B-plane when annealed in oxygen partial pressure, while co-existing A- and B-terminations were observed when annealing in hydrogen partial pressure.
116
S.F. Ceballos et al. / Surface Science 548 (2004) 106–116
(2) Annealing of the contaminant-free crystals in UHV induces segregation of Ca and K contaminants from the bulk to the surface. For low concentrations (less than 2%) of contaminants a p(1 · 2) surface reconstruction was observed. A p(1 · 3) reconstruction resulted after further annealing with an increase of the Ca and K on the surface. No reduction of the O/Fe ratio have been observed. After long annealing sessions the p(1 · 3) reconstruction was replaced by the p(1 · 4) reconstruction. The O/Fe ratio has decreased as function of the annealing time. A concentration of Ca and K similar to the p(1 · 3) reconstruction was measured. We can conclude that the surface reconstructions of Fe3 O4 are induced on one hand by the contamination level on the surface and the O/Fe ratio on the other hand. For a high level of contamination a blurred p(1 · 1) LEED mesh was observe and STM images showed a surface where topographic features can hardly be distinguished. To summarize, we have shown that the (0 0 1) surface of magnetite is highly sensitive to the treatment conditions. In particular, the surface reconstructions of magnetite strongly depend on alkaline and alkaline-earth metals segregating from the bulk to the surface and on the O/Fe ratio. We have observed a p(1 · 3) reconstruction that does not satisfy the electron counting argument. Although this reconstruction is a metastable state of the surface, this brings into question the principle of autocompensation and its universal applicability since other terminations were reported that are not autocompensated. Acknowledgements Financial assistance from Science Foundation Ireland (SFI) under contract no. 00/PI.1/C042 is gratefully acknowledged. References [1] R. Wiesendanger, I.V. Shvets, D. B€ urgler, G. Tarrach, H.J. G€ untherodt, J.M.D. Coey, S. Gr€aser, Science 255 (1992) 583.
[2] A. Gupta, J.Z. Sun, J. Magn. Mater. 24–43 (1999) 200. [3] J.F. Gregg, I. Petej, E. Jouguelet, C. Dennis, J. Phys. D 35 (2002) R121–R155. [4] P.A. Cox, Transition Metal Oxides, An Introduction to their Electronic Structure and Properties, Clarendon Press, Oxford, 1995. [5] R. Aragon, R.J. Rasmussen, J.P. Shepherd, J.W. Koenitzer, J.M. Honig, J. Magn. Magn. Mater. 47–54 (1986) 1335. [6] F.C. Voogt, T. Fujii, P.J.M. Smulders, L. Niesen, M.A. James, T. Hibma, Phys. Rev. B 60 (1999) 11193. [7] G. Tarrach, D. B€ urgler, T. Schaub, R. Wiesendanger, H. G€ untherodt, Surf. Sci. 285 (1993) 1. [8] G. Mariotto, S. Murphy, I.V. Shvets, Phys. Rev. B 66 (2002) 245426. [9] S.A. Chambers, S. Thevuthasan, S.A. Joyce, Surf. Sci. 450 (2000) L273. [10] B. Stanka, W. Hebenstreit, U. Diebold, S.A. Chambers, Surf. Sci. 448 (2000) 49. [11] J.F. Anderson, M. Kuhn, H. Diebold, K. Shaw, P. Stoyanov, D. Lind, Phys. Rev. B 56 (1997) 9902. [12] K.A. Shaw, E. Lochner, D.M. Lind, J. Appl. Phys. 87 (2000) 1727. [13] S.K. Shaikhutdinov, W. Weiss, R. Sch€ ogl, Appl. Surf. Sci. 161 (2000) 497. [14] L.P. Zhang, M. Li, U. Diebold, Surf. Sci. 412/413 (1998) 242. [15] H. N€ orenberg, J.H. Harding, Surf. Sci. 473 (2001) 151. [16] Y. Gao, S.A. Chambers, J. Cryst. Growth 174 (1997) 446. [17] Prof. Honig, Purdue University, Indiana, USA. [18] S. Murphy, J. Osing, I.V. Shvets, J. Magn. Magn. Mater. 198–199 (1999) 686. [19] S.F. Ceballos, G. Mariotto, S. Murphy, I.V. Shvets, Surf. Sci. 523 (2003) 131. [20] K. Jordan, S.F. Ceballos, G. Mariotto, S. Murphy, I.V. Shvets, to be published. [21] A.P. Sutton, R.W. Balluffi, Interfaces in Crystalline Materials, Oxford University Press, 1996. [22] G. Mariotto, S.F. Ceballos, S. Murphy, C. Seoighe, N. Berdunov, I.V. Shvets, Phys. Rev. B, submitted for publication. [23] P.W. Tasker, J. Phys. C 12 (1979) 4977. [24] P.W. Tasker, Philos. Mag. A 12 (1979) 4977. [25] J.A. Meyer, Y. Kuk, P.J. Estrup, P.J. Silverman, Phys. Rev. B 44 (1991) 9104. [26] M.S. Altman, E. Bauer, Surf. Sci. 347 (1996) 265. [27] P.M. Hill, H.S. Peiser, J.R. Rait, Acta Cryst. 9 (1956) 981. [28] M.D. Pashley, Phys. Rev. B 40 (1989) 10481. [29] J. LaFemina, Crit. Rev. Surf. Chem. 3 (1994) 297. [30] O. Warren, P. Thiel, J. Chem. Phys. 100 (1994) 659. [31] A. Barbieri, W. Weiss, M.V. Hove, G. Somorjai, Surf. Sci. 302 (1994) 259. [32] Q. Guo, P. Moeller, Surf. Sci. 340 (1995) L999. [33] N. Condon, F. Leibsle, P. Murray, T. Parker, A. Lennie, D. Vaughan, G. Thornton, Surf. Sci. 397 (1998) 278.