- Email: [email protected]
One-dimensional reconstruction observed on Fe3O4(110) by scanning tunneling microscopy
surface science ELSEVIER
Surface Science 328 (1995) 237-247
One-dimensional reconstruction observed o n Fe304(110) by scanning tunneling microscopy R. Jansen ,, V.A.M. Brabers b, H. van K e m p e n a,* Research lrts'titute for Materials, University of Nijmegen, Toernooiveld 1, NL-6525 ED Nijmegen, The Netherlands b Department of Physics, Eindhoven University of Technology, NL-5600 MB Eindhoven, The Netherland~ Received 2 Decembcr 1994; acceptcd for publication 3 February 1995
Abstract
The (110) surface of magnetite Fe304 singlc crystals was studied by scanning tunneling microscopy (STM). A clean and rcgular surface was obtaincd after sputtering and annealing at 1200 K. This preparation procedure resulted in a one-dimensional reconstruction consisting of rows running in the (710) direction. The row spacing was found to var~ on different terraces, where the most frequent row separation was determined to be 25 A. Current-versus-voltage curves display a transition from semiconducting to metallic character when the tip-sample distance is reduced. These results, together with complementary low energy electron diffraction (LEED) measurements, are compared with a bulk termination of Fe304 and the other known iron-oxide phases. Keywords: Iron oxide; Low energy electron diffraction (LEED); Low indcx single crystal surfaces; Magnetic surfaces; Scanning tunneling microsca-~py; Scanning tunneling spectroscopies; Surface relaxation and reconstruction; Surface structure, morphology, roughness, a~d topography (
1. Introduction
Due to their important technological applications as catalyst materials and their corrosion resistant behavior, metal-oxide surfaces have recently drawn increasing attention. A m o n g the vast amount of oxides, those containing iron have taken up a special placc in view of their application in magnetic devices. There exists a number of different iron-oxide phases, each with different stability ranges [1]. F e l _ ~ O (wustite) crystallizes in the cubic sodiumchloride structure with a large numbcr of vacancies and with the iron cations, mainly Fe z+ ions, octahe-
' Corresponding author. E-mail: [email protected]; Fax: + 31 80 653450.
draUy coordinated to the oxygen anions: In bulk form it is only thermodynamically stable above 843 K. ot-Fe/O 3 (hematite) has the rhombohedral structure of corundum with the antiferromagnetically ordered Fe 3~ cations located in the distorted oxygen octahedral holes formed by the hexagonal closepacked O sublattice. Fe304 (magnetite) is ~halfmetallic [2,3] and ferrimagnetic and has the cubic inverse spinel structure with a lattice constant o f 8.3967 A [4]. In the fcc lattice formed by the oxygen anions Fe 2+ c a t i o n s occupy octahedral positions while the Fe 3 ~ cations are distributed between octa~ hedral and tetrahedral positions. The spins of Fe ions in octahedral and tetrahedral sites couple antiferromagnetically, resulting in a net magnetization due tO the different total magnetic moment in the two sfib-
0039-6028/95/$09.50 g3 1995 Elsevier Scicncc B.V. All rights reservcd SSDI 0 0 3 9 - 6 0 2 8 ( 9 5 ) 0 0 1 7 3 - 5
238
R. Jansen et al. / Surface Science 328 (1995) 237-247
lattices. The high room temperature conductivity of magnetite is attributed to a continuous hopping of electrons between octahedral Fe ions [5], which is frozen out below 124 K at which the Verwey phase transition occurs [6-10]. The metastable 3,-Fe203 (maghemite) has a cubic inverse spinel structure very similar to F e 3 0 4 and differs from magnetite in that it contains ordered vacancies in the octahedrally coordinated iron sublattice. Recently, several iron-oxide surfaces have been investigated by LEED and STM. The (111) termination of a-Fe203 single crystals has been studied by LEED [11,12], where depending on the oxygen partial pressure and the temperature used during annealing different LEED patterns were generated. One of them was shown to be consistent with a Fe304(111) surface termination exposing 1 / 4 monolayer of Fe atoms. This was later confirmed by a STM study on this surface prepared under similar conditions [13]. The same termination was found for (lll)-oriented Fe304 thin films, grown epitaxially onto P t ( l l l ) subStrates [12,14]. These films have also been investigated by STM [15] where for monolayer or submonolayer thicknesses the oxide was identified as F e O ( l l l ) . The only surface study of F e 3 0 4 single crystals was performed on the (001) surface [16]. Here STM and LEED results showed also that depending on the preparation procedure different surface structures can be observed. In contrast to the (111) orientation, however, these structures are consistent with the atomic positions expected for the bulk F e 3 0 4 structure, and no evidence was found for the formation of a surface layer of one of the other bulk iron-oxide phases. In this paper we report on STM and LEED measurements performed on the (110) surface of Fe304 single crystals. These investigations were motivated by the possible use of F e 3 0 4 surfaces as test systems for spin-sensitive STM. Important considerations in this respect are the high spin-polarization at the Fermi level [2,3,17], the good room temperature conductivity required for stable STM operation and the presence, of two anti-ferromagnetically coupled spin sublattices. In the bulk, F e 3 0 4 c a n be thought to consist of two different planes perpendicular to (110) with the arrangement of the Fe and O ions as given in Fig. 1. These type-Aoand type-B layers alternate with a spacing of 1.484 A between successive layers.
A 0
•
Q
0
0
Oo
o
Oo
0
Oo
o
0
0
Oo
0
Oo
0
•
O
o
6)
O
Oxygen
• o
OctahedralFe TetrahedralFe
•
010) (001)
8.4A-~
~
8.4A ~
Fig. 1. The atomic arrangement in the two types of (110) layers A and B present in bulk Fe304. The surface unit cell is 8.4 A along (001) and 6 .~ along (110). Note that the type-A layer contains both octahedral and tetrahedral Fe ions.
As shown in Fig. 1, the type-A layer contains both octahedrally and tetrahedrally coordinated Fe ions which have antiparallel magnetic moments. This is the reason why we chose to examine the (110) surface. We have investigated the effect of different preparation procedures on surface chemical composition, as well as on the structural order indicated by LEED and STM. The results will be compared with the bulk structure of F e 3 0 4 and also with the other known bulk iron-oxide phases, and the implications regarding the use of this surface as test system for spin-sensitive STM are addressed.
2. Experimental The experiments were performed in a UHV chamber with a base pressure of 5 × 10 -11 mbar, maintained by an ion pump and additional titanium sublimation pumping. The chamber is equipped with a two-grid LEED system, an electron analyzer for Auger-electron spectroscopy (AES) and a mass spectrometer for residual gas analysis. Samples can be cleaned by Ar + ion bombardment using a rastering ion gun, and annealed by electron bombardment from the back. During annealing oxygen gas can be leaked into the system, while the sample temperature is monitored by an infrared pyrometer. The complete UHV chamber is suspended from the laboratory floor by three air-damping columns, with the homebuilt STM positioned near the center of gravity on Viton isolation elements to reduce external vibrations. The STM was operated with commercial electronics and all data presented in this paper were
R. Jansen et al. / Surface Science 328 (1995) 237-247
obtained at room temperature with commercially available P t - I r tips used without further cleaning. Calibration of horizontal and vertical STM scales was achieved by using Si(111)7 X 7 and -2 X 1 reconstructed surfaces and a Au(111) sample having monatomic steps, respectively. Samples were cut from synthetic Fe304 single crystals into disks of 5 mm diameter and 1 mm thickness, exposing the (110) crystal face. This was confirmed by X-ray diffraction, which was also used to determine the direction of the (001) and (110) vectors lying in the surface plane. Samples were then polished with diamond paste, with grain sizes down to 0.25 /zm, and damped onto a Mo-holder. After rinsing in ethanol they were loaded into the UHV system where the surface chemical composition was checked by AES. Ar + ion bombardment was performed at a beam energy of 1 keV for 30 rain at
239
normal incidence. Annealing was done at temperatures ranging from 800 to 1200 K in oxygen partial pressures up to 10 - 6 mbar. After every preparation step, samples were inspected by AES to determine the degree of surface contamination, as well as to monitor changes in the relative strength of the iron and oxygen Auger signals resulting from the preparation. An absolute ratio of the Fe and O content of the surface could however not be determined. LEED measurements were used for qualitative interpretation of the atomic structure.
3. Results The first attempt to prepare a clean surface consisted of only annealing the sample. This was tried because previous studies have demonstrated that ion
Fig. 2. STM image obtained on the FesO4(ll0) surface after ion bombardmentand annealing at 850 K. Scan size 2000 A squared, tunnel current 50 pA at + 2.5 V sample voltage.
240
R. Jansen et al. / Surface Science 328 (1995) 237-247
bombardment results in a preferential sputtering of the lighter oxygen atoms [18,19]. Moreover, it was shown in a spin-resolved photoemission experiment [20] that the magnetic properties of a magnetite surface can be drastically altered by ion bombard-
ment. Annealing our samples at temperatures up to 1000 K resulted in the removal of carbon from the surface. Other contaminations like K, C1 and Ca could however not be removed at these temperatures, while at higher temperatures diffusion of Ca from
(a)
E = 96 eV
O 0
0
0
0
0
0
0
0 (oo)
O
O
0
0
0
O
O
O
(b) S = 1 0 7 e V
O I II II
o
6 |
(oo)
©
Fig. 3. LEED patterns of the F%O4(110) surface at 96 (a) and 107 eV (b) incident electron energy. Each photograph is accompanied by a hand-drawn schematic pattern, where the small black squares in (b) represent fractional order spots observed in the [001] direction.
R. Jansen et a l . / Surface Science 328 (1995) 237-247
the bulk caused an increase in the Ca contamination at the surface. On this surface no LEED spots could be seen, while STM showed no traces of any clean and ordered areas. After bombarding the surface with Ar + ions, all contamination levels were reduced to below the detection limit of the AES system, except for a few percent of Ar atoms. Comparison of the ratio of Auger signals from Fe and O showed that an oxygen-deficient surface composition was obtained. This surface was subsequently annealed at about 800 K for 1 hour, yielding a decreased Fe-to-O ratio, although the oxygen content was still lower than before ion bombardment. Also some traces of Ar were still detected, while STM images taken at this stage showed a number of crater-shaped depressions obvi-
241
ously created during the ion-bombardment process. Additional annealing for 1 hour at a slightly higher temperature (850 K) was however sufficient to remove the remaining Ar atoms and restore the original Fe-to-O ratio from before the ion bombardment. An example of the topography seen on this surface by STM is given in Fig. 2 which shows strings of varying length stacked in a disordered manner. There however appears to b e a preference for these string structures to lie in a direction close to the (110) lattice vector. Although the above preparation procedure yielded a clean surface, it did not result in a well-ordered atomic structure. Therefore, the annealing temperature was step-wise increased up to 1200 K, which gradually improved the surface regularity. This was
Fig. 4. STM image showing rows along (~10) with a regular spacing of three times the bulk 8.4 A lattice constant. Scan size 1500 squared, tunnel current 0.2 nA at + 2.0 V sample voltage.
242
R. Jansen et al. / Surface Science 328 (1995) 237-247
evident from the STM measurements as well as from the appearance and sharpening of LEED patterns. It must be noted here that the higher temperatures again caused Ca to diffuse to the surface. After a few cycles of ion bombardment and annealing, the Ca Auger signal disappeared below the noise level. Below we will present LEED and STM results gathered from several measurements on different samples, each time cleaned again before a new measurement by ion bombardment and annealing at 1200 K. In two cases the annealing was done in oxygen partial pressures of 10 -8 and 10 -6 mbar, respectively, without any noticeable effect on LEED or STM results. A brief report of the STM results obtained on this surface was already published elsewhere [21]. LEED patterns at incident electron energies of 96
and 107 eV are displayed in Figs. 3a and 3b, respectively. They show a rectangular lattice as expected for a cubic (110) surface. From the spot-spacing along the [001] direction, a periodicity of 8.0 ± 0.5 A was extracted, while a 3.0 +__0.5 A periodicity was determined for the [ i l 0 ] direction. Comparing this to the 8.397 × 5.937 ~2 unit cell expected for a bulkterminated Fe304(l10) surface (see Fig. 1), we see that the observed surface unit cell is a factor of two smaller in the (110) direction. This indicates that we do not have a bulk-terminated Fe304 surface. The LEED patterns presented in Fig. 3 do not show spots Mong the [110] direction corresponding to a 6.0 _ 0.5 A periodicity, although these were often also observed, but with a much lower intensity. In addition to these spot patterns, a number of weak fractional
Fig. 5. STM image showing resolution within the rows. The scan direction of the tip is rotated as compared to Fig. 4. Scan size 200 ,~ squared, tunnel current 50 pA at + 2.5 V sample voltage.
R. Jansen et al. / Surface Science 328 (1995) 237-247
order spots were seen along the [001] direction, corresponding to a surface periodicity of three times the bulk lattice constant of 8.4 A along (001). These spots are indicated by the small black squares in the hand-drawn schematic pattern of Fig. 3b. The three-times bulk periodicity along (001) was much more pronounced in the electronic structure probed by STM. This is evident in the image presented in Fig. 4, which was obtained at + 2 V sample bias and 0.2 n A tunnel current. It shows rows that were always found to run in the (110) direction. The distance between neighboring rows was determined from several images to be 25 + 0.5 A, in good agreement with 3 times the bulk 8.4 ~, periodicity derived from the fractional order LEED spots along [001]. The corrugation perpendicular to the rows is about 2.9 A. The image of Fig. 4 also shows the step-terrace structure that covers most of the surface area. It is characterized by large rectangularly shaped terraces separated by predominantly straight steps parallel to the rows (i.e. parallel to (110)). Analysis of the stepo heights showed that they are mostly 3.2 and 6.4 A high. Considering the bulk interlayer spacing of 1.5 A between successive planes perpendicular to (110), the measured step heights correo
e /
0.2 ~
0.1
i
°
d c b / ~ ~
a
-0.1 -0.2 p
i
-2
-1
0 sample voltage (V)
i
r
1
2
Fig. 6. I - V curves obtained on the Fe304(l10) surface for different tip-sample distances, set by stabilizing a 0.2 nA current at sample voltages equal to 2.0, 1.4, 1.0, 0.6 and 0.2 V for curves (a) to (e), respectively. The tip-sample separation thus decreases from curve (a) to (e).
243
spond to double and four-fold steps, respectivel~¢, taking the measured single step height to be 1.6 A. Since in bulk F%O 4 type-A and type-B (110) layers (see Fig. 1) are alternating, the observed double and four-fold steps connect parts of the surface with the same underlying type of (110) plane. Whether the deviation of the measured single step height from the bulk 1.5 A value is caused by a relaxation of the outermost surface layers is difficult to establish by STM due to inaccuracies in the STM distance calibration. A higher resolution image, taken in an area with the same 25 A-spaced rows, is presented in Fig. 5, where the scan direction of the STM tip was rotated as compared to the previous image. Now also atomic scale corrugation can be seen along the rows, with an average amplitude of about 0.2 A. The spacing of the protrusions along the rows is disordered for larger scales, but on a smallo scale distances are opredominantly 3, 6, 9 and 12 A, i.e. multiples of 3 A. This is in agreement with the 3 A periodicity derived from the LEED patterns. Attempts to prepare a surface with large scale ordering within the rows have not been successful yet. To determine the conductivity character of the prepared surface, we measured current-versus-voltage curves ( I - V curves) at different tip-sample distances. These are plotted in Fig. 6, where each curve is the average of 900 single curves. Since we did not succeed in detecting spatial variations in the I - V characteristics, the plotted curves can be regarded as taken at random locations at the surface. The tip-sample distance was decreased from curve (a) to curve (e) by reducing the bias voltage while maintaining the same 0.2 nA current. For a large tip-sample distance (curve (a)) a clear gap in the conductivity is found, ranging from about - 1 . 2 to + 0.8 V. When the tip-sample distance is reduced, the conductivity gap gets smaller, while it has completely disappeared for the smallest separation (curve (e)). At this separation the surface shows an almost linear metallic-like I - V curve. It must be noted here that for this small tip-sample distance, stable scanning could still be performed so that mechanical contact between tip and sampleocan be ruled out. Besides the rows with a 25 A spacing on terraces connected by even multiple steps, minor parts of the sample showed terraces separated by single or o
244
R. Jansen et al. / Surface Science 328 (1995) 237-247
three-fold steps. An image obtained in such an area is depicted in Fig. 7. Again a number of terraces having 25 A-spaced rows can be seen [22]. These are marked by the letter X and are connected with each other via even multiple steps, as seen before. In between, terraces marked with the letter Y have a different row separation, either 17 ___0.5 A or 34 + o 0.5 A , in agreement with 2 and 4 times the bulk 8.4 lattice constant along (001), respectively. Note that the row direction is the same on all terraces. Interestingly, it was found that the height of steps, connecting an X with a Y terrace, is always an odd multiple of the single step height, mostly 1.6 and 4.8 A. Based on the alternation of type-A and -B (110) planes in bulk magnetite, a single (or three-fold, etc) step involves a transition to the other type of (110) layer. It is therefore concluded that a different row
spacing is obtained depending on the underlying type of (110) layer.
4. Discussion Regarding the results there are several points to be addressed. First we will discuss the effect of the preparation steps on the surface chemical composition. Next we will compare the surface symmetry derived from LEED with that expected for a bulkterminated Fe304(l10) surface, as well as for a bulk Fe304 crystal, covered with a layer of another iron oxide phase. Then we will include the STM results and arrive at the conclusion that neither LEED nor STM results are consistent with a bulk-terminated Fe304 surface. We will then interpret the results in
Fig. 7. STM image showing terraces with different row spacing. Scan size 1000 A squared, tunnel current 50 pA at + 2.5 V sample voltage.
R. Jansen et al. / Surface Science 328 (1995) 237-247
terms of a reconstruction of the two outermost F%O 4 layers. 4.1. Chemical surface composition
By monitoring the Auger Fe-to-O ratio we have seen that Ar + ion bombardment creates a surface composition deficient in oxygen. After annealing, the oxygen content of the surface layers was enhanced again, and for sufficiently long times at temperatures above 850 K, the Fe-to-O ratio from before the ion bombardment was restored. Based on this we can exclude a transformation into one of the other ironoxide phases for the complete surface layer probed by AES. Additional annealing or annealing in 10 -6 mbar oxygen did not increase the oxygen content beyond the initial value, indicating that an equilibrium surface composition was reached. We might note here that at 1200 K the Fe304 phase is predicted from the bulk phase diagram [11] for the range of oxygen partial pressures used here. Concerning the outermost surface layer, AES cannot exclude a phase change of only this layer, partly due to the experimental accuracy and partly because some phase changes simply do not alter the Fe-to-O ratio. This, for example, happens when we replace an A type Fe304(110) layer by an F e O ( l l 0 ) layer, since these have exactly the same Fe and O content. 4.2. L E E D comparison with bulk iron-oxides
From the LEED patterns we derived the dominant periodicity to be about 8 A in the [001] direction and about 3 A along [110], which does not match with a bulk-terminated Fe304 surface. This would have produced a two times larger periodicity (i.e. 6 A) in the [-il0] direction. We would like to recall here that we did observe spots corresponding to this 6 periodicity, but always at a much lower intensity. This suggests that these originate from the bulk Fe304 structure, whereas the dominant 3 A periodicity arises from a thin oveflayer with a different symmetry. We therefore compare the dominant surface symmetry to those expected for the other bulk iron-oxide phases. For FeO(110) the surface unit cell would be 3 A along (110), but about twice as small as for Fe304 along (001). The observed LEED pattern would therefore correspond to a 2 × 1o
245
reconstructed FeO surface, rather than a bulkterminated FeO surface. When considering the 3,_ Fe203 phase, we recall that it differs from Fe304 in that it contains ordered vacancies at the octahedral Fe sites. It might be seen from Fig. 1 that the creation of these octahedral vacancies can only increase the (110) surface unit vector of T-Fe203 with respect to Fe~O4, but it can never reduce it to the observed 3 A. We can therefore also rule out a y-Fe203 surface termination. When considering an a-Fe203 surface termination, we have to remember that it has the hcp structure, with an A B A B . . . stacking of the hexagonal oxygen layers along (111), while Fe30 4 has an A B C A B C . . . stacking. Although the type of stacking can easily be changed along the (111) stacking direction, this is not possible along the (110) direction without producing lattice deformations and stress at the interface between the Fe304 bulk and an a-F%O 3 Surface layer. Such a stressed interface is unlikely to survive the annealing treatment. We thus conclude that the dominant LEED pattern does not agree with a bulk termination of any of the four known bulk iron-oxide phases. 4.3. S T M results
For the interpretation of the STM images we will take the bulk Fe304 structure as a starting point. We have seen that the observed step heights are in good agreement with the bulk interlayer spacing. Moreover, the predominant double and four-fold steps indicate that one of the two types of (110) layers is favored at the surface. Unfortunately, from a minimization of the number of dangling bonds or the electrostatic surface energy, we cannot predict whether the favored termination is the type-A or -B layer, since both have the same number of dangling bonds and are equally polar. The two types of (110) layers however produced a different spacing of the rows, which is not surprising considering the different atomic arrangement in both layers. We might note here that in FeO all planes perpendicular to (110) are equivalent, which would have resulted in single steps and an identical structure on all terraces. For the interpretation of the corrugation observed with STM, we confine ourselves to discussing only the Fe atoms, because experiments and bulk calculations have shown that in magnetite the density of
246
R. Jansen et aL / Surface Science 328 (1995) 237-247
oxygen-derived states is very low in the energy region of a few eV around the Fermi level [2,3,17]. The distances between protrusions in the (110) direction along the rows were found to be multiples of 3 A. This suggests that they originate from octahedrally coordinated Fe atoms, since in the oxygen fcc lattice, the octahedral interstitials form rows along (110) with a 2.968 A spacing between neighboring sites. The distribution of Fe among these sites does however not show ordering consistent with that of Fe304 or the other three bulk iron-oxide phases. With this identification, we see no evidence for corrugation arising from tetrahedrally coordinated Fe. The absence of tetrahedral Fe ions might be the reason for the smaller unit cell derived from LEED, for a surface unit cell consistent with LEED is obtained if we take a type-A (110) layer and remove the tetrahedral Fe ions. The absence of Fe at tetrahedral sites suggests that the surface resembles moie the structure of c~-Fe203, where only octahedral Fe ions are present. Moreover, this gives a clue to the tripled periodicity along the (001) lattice vector, since cutting the noncubic c~-Fe203 crystal along a plane equivalent to (110) in Fe304 will result in the required three-times periodicity. Although an a-Fe203 layer cannot be matched smoothly to the (110) surface of an Fe304 bulk crystal due to the different stacking of the oxygen layers, a structure that preserves the tripled periodicity and the higher oxygen content can still be formed. Such a structure is depicted in Fig. 8, where the surface is covered with tracks of oxygen and octahedral iron atoms along (110) with a width of a few atomic spacings along (001). The height of these tracks was chosen to be two atomic layers, since in STM the observed corrugation perpendicular to the rows was found to be 2.9 A. This is close to twice the bulk interlayer spacing along (110), which indicates that the formation of the rows involves two layers, instead of only the outermost one. Note that the proposed structure allows the formation of locally equivalent domains by translations in the (001) direction over once or twice the 8.4 A lattice constant. Furthermore, note that a translation over half this lattice constant would produce a different structure of the rows, with the top layer consisting of oxygen and tetrahedral Fe atoms. The presence of "a-Fe203-1ike" layers (such as
25.2A ~
l
(llO)
0*0 0,0 0,0 0.0 0°0 OoO 0°0
0,0 0*0 0,0 0"0 0,0 OeO 0*0
O,O
0-0
0*0 0-0 0,0
OoO 0*0 OeO
AB
l (i1°) < (00a)
Fig. 8. Ball representation of a Fe304(110) surface with row reconstruction. The top part shows a projection along the (110) vector, which lies in the surface plane. The lower part shows a top view on the (110) surface, where only atoms of the highest lying layer are drawn. Open circles denote oxygen ions, while octahedral and tetrahedral Fe ions are given as small black and grey circles, respectively. Note the absence of tetrahedral Fe ions in the two top layers.
shown in Fig. 8) at the surface can explain the conductivity gap found in the I - V curves observed for larger tip-sample distances, while for smaller bias voltages, electrons tunnel through these layers into the underlying magnetite. In the latter case, a linear I - V curve is found since magnetite is conducting. We would however like to stress that the above interpretation must be treated with some reserve, as more experimental details are needed for a definite conclusion. Regarding the use of this magnetite surface as test system for spin-polarized STM, more detailed information is required about the precise atomic arrangement, as well as the local surface magnetic properties. We envision that this will derive from a combination of atomic scale surface probes such as STM and LEED, and more global techniques such as spin-averaged and spin-resolved photoelectron spectroscopy and spin-polarized electron energy loss spectroscopy.
5. Summary and conclusions We have studied the (110) surface of magnetite Fe304 single crystals. A well-ordered structure was
R. Jansen et al. / Surface Science 328 (1995) 237-247
o b s e r v e d for surfaces p r e p a r e d by ion b o m b a r d m e n t and subsequent a n n e a l i n g at 1200 K. T h e obtained surface exhibits a t e r r a c e - d e p e n d e n t o n e - d i m e n s i o n a l reconstruction w h i c h w a s m o s t e v i d e n t in the electronic structure p r o b e d b y S T M . F r o m a c o m p a r i s o n w i t h the structures o f k n o w n bulk i r o n - o x i d e phases it w a s c o n c l u d e d that the surface is not a s i m p l e bulk i r o n - o x i d e termination. A m o r e detailed characterization o f the surface a t o m i c and m a g n e t i c structure is therefore r e q u i r e d if this m a g n e t i t e surface is to be u s e d as test system for spin-polarized S T M .
Acknowledgements T h e authors w i s h to thank D.L. A b r a h a m and B.J. N e l i s s e n for their contribution at the initial stage o f the experiments, and are grateful to J. H e r m s e n for his technical assistance. Part o f this w o r k w a s supported b y the Stichting F u n d a m e n t e e l O n d e r z o e k der M a t e r i e ( F O M ) w h i c h is financially supported by the N e d e r l a n d s e Organisatie v o o r W e t e n s c h a p p e l i j k Onderzoek (NWO).
References [1] Landolt-B6rnstein, Numerical Data and Functional Relationships in Science and Technology 4b, 12b and 27d, Magnetic and Other Properties of Oxides and Related Compounds (Springer, Berlin, 1970, 1980, 1991). [2] A. Yanase and K. Siratori, J. Phys. Soc. Jpn. 53 (1984) 312. [3] Z. Zhang and S. Satpathy, Phys. Rev. B 44 (1991) 13319.
247
[4] R.W.G. Wyckoff, Crystal Structures, Vol. 1-3, 2nd ed. (Krieger, Malabar, 1982). [5] T. Mizoguchi and M. Inoue, J. Phys. Soc. Jpn. 21 (1966) 1310. [6] A.J.M. Kuipers and V.A.M. Brabers, Phys. Rev. B 14 (1976) 1401. [7] A.J.M. Kuipers and V.A.M. Brabers, Phys. Rev. B 20 (1979) 594. [8] S. Iida et al., J. Appl. Phys. 53 (1982) 2164. [9] E. IOta, Y. Tokuyama, A. Tasaki and K. Siratori, J. Magu. Magn. Mater. 31-34 (1983) 787. [10] R. Aragon, D.J. Buttrey, J.P. Shepherd and J.M. Honig, Phys. Rev. B 31 (1985) 430. [11] R.J. Lad and V.E. Henrich, Surf. Sci. 193 (1988) 81. [12] A. Barbieri, W. Weiss, M.A. Van Hove and G.A. Somorjai, Surf. Sci. 302 (1994) 259. [13] N.G. Condon, P.W. Murray, F.M. Leibsle, G. Thornton, A.R. Lennie and D.J. Vaughan, Surf. Sci. 310 (1994) L609. [14] W. Weiss, A. Barbieri, M.A. Van Hove and G.A. Somorjai, Phys. Rev. Lett. 71 (1993) 1848. [15] H.C. Galloway, J.J. Benitez and M. Salmeron, Surf. Sci. 298 (1993) 127; H.C. Galloway, J.J. Benltez and M, Salmeron, J. Vac. Sci. Technol. A 12 (1994) 2302. [16] G. Tarrach, D. Biirgler, T. Schaub, R. Wiesendanger and H.-J. Giintherodt, Surf. Sci. 285 (1993) 1. [17] S.F. Alvarado, M. Erbudak and P. Munz, Phys. Rev. B 14 (1976) 2740. [18] D. Buczek and S. Sastri, J. Vac. Sci. Technol. 17 (1980) 201. [19] D.F. Mitchell, G.I. Sproule and M.J. Graham, J. Vac. Sei. Technol. 18 (1981) 690. [20] M. Aeschlimann, G.L. Bona, F. Meier, M. Stampanoni, G. Zampieri and H.C. Siegmann, Appl. Phys. Lett. 49 (1986) 824. [21] R. Jansen, B.J. Nelissen, D.L. Abraham, H. van Kempen and V.A.M. Brabers, IEEE Trans. Magn. 30 (1994) 4506. [22] In the image of Fig. 7 these rows seem to appear as two parallel ones, which differs from the structure of the rows in Fig. 5. The difference is most probably an artifact produced by a double tip in case of Fig. 7.
Surface Science 328 (1995) 237-247
One-dimensional reconstruction observed o n Fe304(110) by scanning tunneling microscopy R. Jansen ,, V.A.M. Brabers b, H. van K e m p e n a,* Research lrts'titute for Materials, University of Nijmegen, Toernooiveld 1, NL-6525 ED Nijmegen, The Netherlands b Department of Physics, Eindhoven University of Technology, NL-5600 MB Eindhoven, The Netherland~ Received 2 Decembcr 1994; acceptcd for publication 3 February 1995
Abstract
The (110) surface of magnetite Fe304 singlc crystals was studied by scanning tunneling microscopy (STM). A clean and rcgular surface was obtaincd after sputtering and annealing at 1200 K. This preparation procedure resulted in a one-dimensional reconstruction consisting of rows running in the (710) direction. The row spacing was found to var~ on different terraces, where the most frequent row separation was determined to be 25 A. Current-versus-voltage curves display a transition from semiconducting to metallic character when the tip-sample distance is reduced. These results, together with complementary low energy electron diffraction (LEED) measurements, are compared with a bulk termination of Fe304 and the other known iron-oxide phases. Keywords: Iron oxide; Low energy electron diffraction (LEED); Low indcx single crystal surfaces; Magnetic surfaces; Scanning tunneling microsca-~py; Scanning tunneling spectroscopies; Surface relaxation and reconstruction; Surface structure, morphology, roughness, a~d topography (
1. Introduction
Due to their important technological applications as catalyst materials and their corrosion resistant behavior, metal-oxide surfaces have recently drawn increasing attention. A m o n g the vast amount of oxides, those containing iron have taken up a special placc in view of their application in magnetic devices. There exists a number of different iron-oxide phases, each with different stability ranges [1]. F e l _ ~ O (wustite) crystallizes in the cubic sodiumchloride structure with a large numbcr of vacancies and with the iron cations, mainly Fe z+ ions, octahe-
' Corresponding author. E-mail: [email protected]; Fax: + 31 80 653450.
draUy coordinated to the oxygen anions: In bulk form it is only thermodynamically stable above 843 K. ot-Fe/O 3 (hematite) has the rhombohedral structure of corundum with the antiferromagnetically ordered Fe 3~ cations located in the distorted oxygen octahedral holes formed by the hexagonal closepacked O sublattice. Fe304 (magnetite) is ~halfmetallic [2,3] and ferrimagnetic and has the cubic inverse spinel structure with a lattice constant o f 8.3967 A [4]. In the fcc lattice formed by the oxygen anions Fe 2+ c a t i o n s occupy octahedral positions while the Fe 3 ~ cations are distributed between octa~ hedral and tetrahedral positions. The spins of Fe ions in octahedral and tetrahedral sites couple antiferromagnetically, resulting in a net magnetization due tO the different total magnetic moment in the two sfib-
0039-6028/95/$09.50 g3 1995 Elsevier Scicncc B.V. All rights reservcd SSDI 0 0 3 9 - 6 0 2 8 ( 9 5 ) 0 0 1 7 3 - 5
238
R. Jansen et al. / Surface Science 328 (1995) 237-247
lattices. The high room temperature conductivity of magnetite is attributed to a continuous hopping of electrons between octahedral Fe ions [5], which is frozen out below 124 K at which the Verwey phase transition occurs [6-10]. The metastable 3,-Fe203 (maghemite) has a cubic inverse spinel structure very similar to F e 3 0 4 and differs from magnetite in that it contains ordered vacancies in the octahedrally coordinated iron sublattice. Recently, several iron-oxide surfaces have been investigated by LEED and STM. The (111) termination of a-Fe203 single crystals has been studied by LEED [11,12], where depending on the oxygen partial pressure and the temperature used during annealing different LEED patterns were generated. One of them was shown to be consistent with a Fe304(111) surface termination exposing 1 / 4 monolayer of Fe atoms. This was later confirmed by a STM study on this surface prepared under similar conditions [13]. The same termination was found for (lll)-oriented Fe304 thin films, grown epitaxially onto P t ( l l l ) subStrates [12,14]. These films have also been investigated by STM [15] where for monolayer or submonolayer thicknesses the oxide was identified as F e O ( l l l ) . The only surface study of F e 3 0 4 single crystals was performed on the (001) surface [16]. Here STM and LEED results showed also that depending on the preparation procedure different surface structures can be observed. In contrast to the (111) orientation, however, these structures are consistent with the atomic positions expected for the bulk F e 3 0 4 structure, and no evidence was found for the formation of a surface layer of one of the other bulk iron-oxide phases. In this paper we report on STM and LEED measurements performed on the (110) surface of Fe304 single crystals. These investigations were motivated by the possible use of F e 3 0 4 surfaces as test systems for spin-sensitive STM. Important considerations in this respect are the high spin-polarization at the Fermi level [2,3,17], the good room temperature conductivity required for stable STM operation and the presence, of two anti-ferromagnetically coupled spin sublattices. In the bulk, F e 3 0 4 c a n be thought to consist of two different planes perpendicular to (110) with the arrangement of the Fe and O ions as given in Fig. 1. These type-Aoand type-B layers alternate with a spacing of 1.484 A between successive layers.
A 0
•
Q
0
0
Oo
o
Oo
0
Oo
o
0
0
Oo
0
Oo
0
•
O
o
6)
O
Oxygen
• o
OctahedralFe TetrahedralFe
•
010) (001)
8.4A-~
~
8.4A ~
Fig. 1. The atomic arrangement in the two types of (110) layers A and B present in bulk Fe304. The surface unit cell is 8.4 A along (001) and 6 .~ along (110). Note that the type-A layer contains both octahedral and tetrahedral Fe ions.
As shown in Fig. 1, the type-A layer contains both octahedrally and tetrahedrally coordinated Fe ions which have antiparallel magnetic moments. This is the reason why we chose to examine the (110) surface. We have investigated the effect of different preparation procedures on surface chemical composition, as well as on the structural order indicated by LEED and STM. The results will be compared with the bulk structure of F e 3 0 4 and also with the other known bulk iron-oxide phases, and the implications regarding the use of this surface as test system for spin-sensitive STM are addressed.
2. Experimental The experiments were performed in a UHV chamber with a base pressure of 5 × 10 -11 mbar, maintained by an ion pump and additional titanium sublimation pumping. The chamber is equipped with a two-grid LEED system, an electron analyzer for Auger-electron spectroscopy (AES) and a mass spectrometer for residual gas analysis. Samples can be cleaned by Ar + ion bombardment using a rastering ion gun, and annealed by electron bombardment from the back. During annealing oxygen gas can be leaked into the system, while the sample temperature is monitored by an infrared pyrometer. The complete UHV chamber is suspended from the laboratory floor by three air-damping columns, with the homebuilt STM positioned near the center of gravity on Viton isolation elements to reduce external vibrations. The STM was operated with commercial electronics and all data presented in this paper were
R. Jansen et al. / Surface Science 328 (1995) 237-247
obtained at room temperature with commercially available P t - I r tips used without further cleaning. Calibration of horizontal and vertical STM scales was achieved by using Si(111)7 X 7 and -2 X 1 reconstructed surfaces and a Au(111) sample having monatomic steps, respectively. Samples were cut from synthetic Fe304 single crystals into disks of 5 mm diameter and 1 mm thickness, exposing the (110) crystal face. This was confirmed by X-ray diffraction, which was also used to determine the direction of the (001) and (110) vectors lying in the surface plane. Samples were then polished with diamond paste, with grain sizes down to 0.25 /zm, and damped onto a Mo-holder. After rinsing in ethanol they were loaded into the UHV system where the surface chemical composition was checked by AES. Ar + ion bombardment was performed at a beam energy of 1 keV for 30 rain at
239
normal incidence. Annealing was done at temperatures ranging from 800 to 1200 K in oxygen partial pressures up to 10 - 6 mbar. After every preparation step, samples were inspected by AES to determine the degree of surface contamination, as well as to monitor changes in the relative strength of the iron and oxygen Auger signals resulting from the preparation. An absolute ratio of the Fe and O content of the surface could however not be determined. LEED measurements were used for qualitative interpretation of the atomic structure.
3. Results The first attempt to prepare a clean surface consisted of only annealing the sample. This was tried because previous studies have demonstrated that ion
Fig. 2. STM image obtained on the FesO4(ll0) surface after ion bombardmentand annealing at 850 K. Scan size 2000 A squared, tunnel current 50 pA at + 2.5 V sample voltage.
240
R. Jansen et al. / Surface Science 328 (1995) 237-247
bombardment results in a preferential sputtering of the lighter oxygen atoms [18,19]. Moreover, it was shown in a spin-resolved photoemission experiment [20] that the magnetic properties of a magnetite surface can be drastically altered by ion bombard-
ment. Annealing our samples at temperatures up to 1000 K resulted in the removal of carbon from the surface. Other contaminations like K, C1 and Ca could however not be removed at these temperatures, while at higher temperatures diffusion of Ca from
(a)
E = 96 eV
O 0
0
0
0
0
0
0
0 (oo)
O
O
0
0
0
O
O
O
(b) S = 1 0 7 e V
O I II II
o
6 |
(oo)
©
Fig. 3. LEED patterns of the F%O4(110) surface at 96 (a) and 107 eV (b) incident electron energy. Each photograph is accompanied by a hand-drawn schematic pattern, where the small black squares in (b) represent fractional order spots observed in the [001] direction.
R. Jansen et a l . / Surface Science 328 (1995) 237-247
the bulk caused an increase in the Ca contamination at the surface. On this surface no LEED spots could be seen, while STM showed no traces of any clean and ordered areas. After bombarding the surface with Ar + ions, all contamination levels were reduced to below the detection limit of the AES system, except for a few percent of Ar atoms. Comparison of the ratio of Auger signals from Fe and O showed that an oxygen-deficient surface composition was obtained. This surface was subsequently annealed at about 800 K for 1 hour, yielding a decreased Fe-to-O ratio, although the oxygen content was still lower than before ion bombardment. Also some traces of Ar were still detected, while STM images taken at this stage showed a number of crater-shaped depressions obvi-
241
ously created during the ion-bombardment process. Additional annealing for 1 hour at a slightly higher temperature (850 K) was however sufficient to remove the remaining Ar atoms and restore the original Fe-to-O ratio from before the ion bombardment. An example of the topography seen on this surface by STM is given in Fig. 2 which shows strings of varying length stacked in a disordered manner. There however appears to b e a preference for these string structures to lie in a direction close to the (110) lattice vector. Although the above preparation procedure yielded a clean surface, it did not result in a well-ordered atomic structure. Therefore, the annealing temperature was step-wise increased up to 1200 K, which gradually improved the surface regularity. This was
Fig. 4. STM image showing rows along (~10) with a regular spacing of three times the bulk 8.4 A lattice constant. Scan size 1500 squared, tunnel current 0.2 nA at + 2.0 V sample voltage.
242
R. Jansen et al. / Surface Science 328 (1995) 237-247
evident from the STM measurements as well as from the appearance and sharpening of LEED patterns. It must be noted here that the higher temperatures again caused Ca to diffuse to the surface. After a few cycles of ion bombardment and annealing, the Ca Auger signal disappeared below the noise level. Below we will present LEED and STM results gathered from several measurements on different samples, each time cleaned again before a new measurement by ion bombardment and annealing at 1200 K. In two cases the annealing was done in oxygen partial pressures of 10 -8 and 10 -6 mbar, respectively, without any noticeable effect on LEED or STM results. A brief report of the STM results obtained on this surface was already published elsewhere [21]. LEED patterns at incident electron energies of 96
and 107 eV are displayed in Figs. 3a and 3b, respectively. They show a rectangular lattice as expected for a cubic (110) surface. From the spot-spacing along the [001] direction, a periodicity of 8.0 ± 0.5 A was extracted, while a 3.0 +__0.5 A periodicity was determined for the [ i l 0 ] direction. Comparing this to the 8.397 × 5.937 ~2 unit cell expected for a bulkterminated Fe304(l10) surface (see Fig. 1), we see that the observed surface unit cell is a factor of two smaller in the (110) direction. This indicates that we do not have a bulk-terminated Fe304 surface. The LEED patterns presented in Fig. 3 do not show spots Mong the [110] direction corresponding to a 6.0 _ 0.5 A periodicity, although these were often also observed, but with a much lower intensity. In addition to these spot patterns, a number of weak fractional
Fig. 5. STM image showing resolution within the rows. The scan direction of the tip is rotated as compared to Fig. 4. Scan size 200 ,~ squared, tunnel current 50 pA at + 2.5 V sample voltage.
R. Jansen et al. / Surface Science 328 (1995) 237-247
order spots were seen along the [001] direction, corresponding to a surface periodicity of three times the bulk lattice constant of 8.4 A along (001). These spots are indicated by the small black squares in the hand-drawn schematic pattern of Fig. 3b. The three-times bulk periodicity along (001) was much more pronounced in the electronic structure probed by STM. This is evident in the image presented in Fig. 4, which was obtained at + 2 V sample bias and 0.2 n A tunnel current. It shows rows that were always found to run in the (110) direction. The distance between neighboring rows was determined from several images to be 25 + 0.5 A, in good agreement with 3 times the bulk 8.4 ~, periodicity derived from the fractional order LEED spots along [001]. The corrugation perpendicular to the rows is about 2.9 A. The image of Fig. 4 also shows the step-terrace structure that covers most of the surface area. It is characterized by large rectangularly shaped terraces separated by predominantly straight steps parallel to the rows (i.e. parallel to (110)). Analysis of the stepo heights showed that they are mostly 3.2 and 6.4 A high. Considering the bulk interlayer spacing of 1.5 A between successive planes perpendicular to (110), the measured step heights correo
e /
0.2 ~
0.1
i
°
d c b / ~ ~
a
-0.1 -0.2 p
i
-2
-1
0 sample voltage (V)
i
r
1
2
Fig. 6. I - V curves obtained on the Fe304(l10) surface for different tip-sample distances, set by stabilizing a 0.2 nA current at sample voltages equal to 2.0, 1.4, 1.0, 0.6 and 0.2 V for curves (a) to (e), respectively. The tip-sample separation thus decreases from curve (a) to (e).
243
spond to double and four-fold steps, respectivel~¢, taking the measured single step height to be 1.6 A. Since in bulk F%O 4 type-A and type-B (110) layers (see Fig. 1) are alternating, the observed double and four-fold steps connect parts of the surface with the same underlying type of (110) plane. Whether the deviation of the measured single step height from the bulk 1.5 A value is caused by a relaxation of the outermost surface layers is difficult to establish by STM due to inaccuracies in the STM distance calibration. A higher resolution image, taken in an area with the same 25 A-spaced rows, is presented in Fig. 5, where the scan direction of the STM tip was rotated as compared to the previous image. Now also atomic scale corrugation can be seen along the rows, with an average amplitude of about 0.2 A. The spacing of the protrusions along the rows is disordered for larger scales, but on a smallo scale distances are opredominantly 3, 6, 9 and 12 A, i.e. multiples of 3 A. This is in agreement with the 3 A periodicity derived from the LEED patterns. Attempts to prepare a surface with large scale ordering within the rows have not been successful yet. To determine the conductivity character of the prepared surface, we measured current-versus-voltage curves ( I - V curves) at different tip-sample distances. These are plotted in Fig. 6, where each curve is the average of 900 single curves. Since we did not succeed in detecting spatial variations in the I - V characteristics, the plotted curves can be regarded as taken at random locations at the surface. The tip-sample distance was decreased from curve (a) to curve (e) by reducing the bias voltage while maintaining the same 0.2 nA current. For a large tip-sample distance (curve (a)) a clear gap in the conductivity is found, ranging from about - 1 . 2 to + 0.8 V. When the tip-sample distance is reduced, the conductivity gap gets smaller, while it has completely disappeared for the smallest separation (curve (e)). At this separation the surface shows an almost linear metallic-like I - V curve. It must be noted here that for this small tip-sample distance, stable scanning could still be performed so that mechanical contact between tip and sampleocan be ruled out. Besides the rows with a 25 A spacing on terraces connected by even multiple steps, minor parts of the sample showed terraces separated by single or o
244
R. Jansen et al. / Surface Science 328 (1995) 237-247
three-fold steps. An image obtained in such an area is depicted in Fig. 7. Again a number of terraces having 25 A-spaced rows can be seen [22]. These are marked by the letter X and are connected with each other via even multiple steps, as seen before. In between, terraces marked with the letter Y have a different row separation, either 17 ___0.5 A or 34 + o 0.5 A , in agreement with 2 and 4 times the bulk 8.4 lattice constant along (001), respectively. Note that the row direction is the same on all terraces. Interestingly, it was found that the height of steps, connecting an X with a Y terrace, is always an odd multiple of the single step height, mostly 1.6 and 4.8 A. Based on the alternation of type-A and -B (110) planes in bulk magnetite, a single (or three-fold, etc) step involves a transition to the other type of (110) layer. It is therefore concluded that a different row
spacing is obtained depending on the underlying type of (110) layer.
4. Discussion Regarding the results there are several points to be addressed. First we will discuss the effect of the preparation steps on the surface chemical composition. Next we will compare the surface symmetry derived from LEED with that expected for a bulkterminated Fe304(l10) surface, as well as for a bulk Fe304 crystal, covered with a layer of another iron oxide phase. Then we will include the STM results and arrive at the conclusion that neither LEED nor STM results are consistent with a bulk-terminated Fe304 surface. We will then interpret the results in
Fig. 7. STM image showing terraces with different row spacing. Scan size 1000 A squared, tunnel current 50 pA at + 2.5 V sample voltage.
R. Jansen et al. / Surface Science 328 (1995) 237-247
terms of a reconstruction of the two outermost F%O 4 layers. 4.1. Chemical surface composition
By monitoring the Auger Fe-to-O ratio we have seen that Ar + ion bombardment creates a surface composition deficient in oxygen. After annealing, the oxygen content of the surface layers was enhanced again, and for sufficiently long times at temperatures above 850 K, the Fe-to-O ratio from before the ion bombardment was restored. Based on this we can exclude a transformation into one of the other ironoxide phases for the complete surface layer probed by AES. Additional annealing or annealing in 10 -6 mbar oxygen did not increase the oxygen content beyond the initial value, indicating that an equilibrium surface composition was reached. We might note here that at 1200 K the Fe304 phase is predicted from the bulk phase diagram [11] for the range of oxygen partial pressures used here. Concerning the outermost surface layer, AES cannot exclude a phase change of only this layer, partly due to the experimental accuracy and partly because some phase changes simply do not alter the Fe-to-O ratio. This, for example, happens when we replace an A type Fe304(110) layer by an F e O ( l l 0 ) layer, since these have exactly the same Fe and O content. 4.2. L E E D comparison with bulk iron-oxides
From the LEED patterns we derived the dominant periodicity to be about 8 A in the [001] direction and about 3 A along [110], which does not match with a bulk-terminated Fe304 surface. This would have produced a two times larger periodicity (i.e. 6 A) in the [-il0] direction. We would like to recall here that we did observe spots corresponding to this 6 periodicity, but always at a much lower intensity. This suggests that these originate from the bulk Fe304 structure, whereas the dominant 3 A periodicity arises from a thin oveflayer with a different symmetry. We therefore compare the dominant surface symmetry to those expected for the other bulk iron-oxide phases. For FeO(110) the surface unit cell would be 3 A along (110), but about twice as small as for Fe304 along (001). The observed LEED pattern would therefore correspond to a 2 × 1o
245
reconstructed FeO surface, rather than a bulkterminated FeO surface. When considering the 3,_ Fe203 phase, we recall that it differs from Fe304 in that it contains ordered vacancies at the octahedral Fe sites. It might be seen from Fig. 1 that the creation of these octahedral vacancies can only increase the (110) surface unit vector of T-Fe203 with respect to Fe~O4, but it can never reduce it to the observed 3 A. We can therefore also rule out a y-Fe203 surface termination. When considering an a-Fe203 surface termination, we have to remember that it has the hcp structure, with an A B A B . . . stacking of the hexagonal oxygen layers along (111), while Fe30 4 has an A B C A B C . . . stacking. Although the type of stacking can easily be changed along the (111) stacking direction, this is not possible along the (110) direction without producing lattice deformations and stress at the interface between the Fe304 bulk and an a-F%O 3 Surface layer. Such a stressed interface is unlikely to survive the annealing treatment. We thus conclude that the dominant LEED pattern does not agree with a bulk termination of any of the four known bulk iron-oxide phases. 4.3. S T M results
For the interpretation of the STM images we will take the bulk Fe304 structure as a starting point. We have seen that the observed step heights are in good agreement with the bulk interlayer spacing. Moreover, the predominant double and four-fold steps indicate that one of the two types of (110) layers is favored at the surface. Unfortunately, from a minimization of the number of dangling bonds or the electrostatic surface energy, we cannot predict whether the favored termination is the type-A or -B layer, since both have the same number of dangling bonds and are equally polar. The two types of (110) layers however produced a different spacing of the rows, which is not surprising considering the different atomic arrangement in both layers. We might note here that in FeO all planes perpendicular to (110) are equivalent, which would have resulted in single steps and an identical structure on all terraces. For the interpretation of the corrugation observed with STM, we confine ourselves to discussing only the Fe atoms, because experiments and bulk calculations have shown that in magnetite the density of
246
R. Jansen et aL / Surface Science 328 (1995) 237-247
oxygen-derived states is very low in the energy region of a few eV around the Fermi level [2,3,17]. The distances between protrusions in the (110) direction along the rows were found to be multiples of 3 A. This suggests that they originate from octahedrally coordinated Fe atoms, since in the oxygen fcc lattice, the octahedral interstitials form rows along (110) with a 2.968 A spacing between neighboring sites. The distribution of Fe among these sites does however not show ordering consistent with that of Fe304 or the other three bulk iron-oxide phases. With this identification, we see no evidence for corrugation arising from tetrahedrally coordinated Fe. The absence of tetrahedral Fe ions might be the reason for the smaller unit cell derived from LEED, for a surface unit cell consistent with LEED is obtained if we take a type-A (110) layer and remove the tetrahedral Fe ions. The absence of Fe at tetrahedral sites suggests that the surface resembles moie the structure of c~-Fe203, where only octahedral Fe ions are present. Moreover, this gives a clue to the tripled periodicity along the (001) lattice vector, since cutting the noncubic c~-Fe203 crystal along a plane equivalent to (110) in Fe304 will result in the required three-times periodicity. Although an a-Fe203 layer cannot be matched smoothly to the (110) surface of an Fe304 bulk crystal due to the different stacking of the oxygen layers, a structure that preserves the tripled periodicity and the higher oxygen content can still be formed. Such a structure is depicted in Fig. 8, where the surface is covered with tracks of oxygen and octahedral iron atoms along (110) with a width of a few atomic spacings along (001). The height of these tracks was chosen to be two atomic layers, since in STM the observed corrugation perpendicular to the rows was found to be 2.9 A. This is close to twice the bulk interlayer spacing along (110), which indicates that the formation of the rows involves two layers, instead of only the outermost one. Note that the proposed structure allows the formation of locally equivalent domains by translations in the (001) direction over once or twice the 8.4 A lattice constant. Furthermore, note that a translation over half this lattice constant would produce a different structure of the rows, with the top layer consisting of oxygen and tetrahedral Fe atoms. The presence of "a-Fe203-1ike" layers (such as
25.2A ~
l
(llO)
0*0 0,0 0,0 0.0 0°0 OoO 0°0
0,0 0*0 0,0 0"0 0,0 OeO 0*0
O,O
0-0
0*0 0-0 0,0
OoO 0*0 OeO
AB
l (i1°) < (00a)
Fig. 8. Ball representation of a Fe304(110) surface with row reconstruction. The top part shows a projection along the (110) vector, which lies in the surface plane. The lower part shows a top view on the (110) surface, where only atoms of the highest lying layer are drawn. Open circles denote oxygen ions, while octahedral and tetrahedral Fe ions are given as small black and grey circles, respectively. Note the absence of tetrahedral Fe ions in the two top layers.
shown in Fig. 8) at the surface can explain the conductivity gap found in the I - V curves observed for larger tip-sample distances, while for smaller bias voltages, electrons tunnel through these layers into the underlying magnetite. In the latter case, a linear I - V curve is found since magnetite is conducting. We would however like to stress that the above interpretation must be treated with some reserve, as more experimental details are needed for a definite conclusion. Regarding the use of this magnetite surface as test system for spin-polarized STM, more detailed information is required about the precise atomic arrangement, as well as the local surface magnetic properties. We envision that this will derive from a combination of atomic scale surface probes such as STM and LEED, and more global techniques such as spin-averaged and spin-resolved photoelectron spectroscopy and spin-polarized electron energy loss spectroscopy.
5. Summary and conclusions We have studied the (110) surface of magnetite Fe304 single crystals. A well-ordered structure was
R. Jansen et al. / Surface Science 328 (1995) 237-247
o b s e r v e d for surfaces p r e p a r e d by ion b o m b a r d m e n t and subsequent a n n e a l i n g at 1200 K. T h e obtained surface exhibits a t e r r a c e - d e p e n d e n t o n e - d i m e n s i o n a l reconstruction w h i c h w a s m o s t e v i d e n t in the electronic structure p r o b e d b y S T M . F r o m a c o m p a r i s o n w i t h the structures o f k n o w n bulk i r o n - o x i d e phases it w a s c o n c l u d e d that the surface is not a s i m p l e bulk i r o n - o x i d e termination. A m o r e detailed characterization o f the surface a t o m i c and m a g n e t i c structure is therefore r e q u i r e d if this m a g n e t i t e surface is to be u s e d as test system for spin-polarized S T M .
Acknowledgements T h e authors w i s h to thank D.L. A b r a h a m and B.J. N e l i s s e n for their contribution at the initial stage o f the experiments, and are grateful to J. H e r m s e n for his technical assistance. Part o f this w o r k w a s supported b y the Stichting F u n d a m e n t e e l O n d e r z o e k der M a t e r i e ( F O M ) w h i c h is financially supported by the N e d e r l a n d s e Organisatie v o o r W e t e n s c h a p p e l i j k Onderzoek (NWO).
References [1] Landolt-B6rnstein, Numerical Data and Functional Relationships in Science and Technology 4b, 12b and 27d, Magnetic and Other Properties of Oxides and Related Compounds (Springer, Berlin, 1970, 1980, 1991). [2] A. Yanase and K. Siratori, J. Phys. Soc. Jpn. 53 (1984) 312. [3] Z. Zhang and S. Satpathy, Phys. Rev. B 44 (1991) 13319.
247
[4] R.W.G. Wyckoff, Crystal Structures, Vol. 1-3, 2nd ed. (Krieger, Malabar, 1982). [5] T. Mizoguchi and M. Inoue, J. Phys. Soc. Jpn. 21 (1966) 1310. [6] A.J.M. Kuipers and V.A.M. Brabers, Phys. Rev. B 14 (1976) 1401. [7] A.J.M. Kuipers and V.A.M. Brabers, Phys. Rev. B 20 (1979) 594. [8] S. Iida et al., J. Appl. Phys. 53 (1982) 2164. [9] E. IOta, Y. Tokuyama, A. Tasaki and K. Siratori, J. Magu. Magn. Mater. 31-34 (1983) 787. [10] R. Aragon, D.J. Buttrey, J.P. Shepherd and J.M. Honig, Phys. Rev. B 31 (1985) 430. [11] R.J. Lad and V.E. Henrich, Surf. Sci. 193 (1988) 81. [12] A. Barbieri, W. Weiss, M.A. Van Hove and G.A. Somorjai, Surf. Sci. 302 (1994) 259. [13] N.G. Condon, P.W. Murray, F.M. Leibsle, G. Thornton, A.R. Lennie and D.J. Vaughan, Surf. Sci. 310 (1994) L609. [14] W. Weiss, A. Barbieri, M.A. Van Hove and G.A. Somorjai, Phys. Rev. Lett. 71 (1993) 1848. [15] H.C. Galloway, J.J. Benitez and M. Salmeron, Surf. Sci. 298 (1993) 127; H.C. Galloway, J.J. Benltez and M, Salmeron, J. Vac. Sci. Technol. A 12 (1994) 2302. [16] G. Tarrach, D. Biirgler, T. Schaub, R. Wiesendanger and H.-J. Giintherodt, Surf. Sci. 285 (1993) 1. [17] S.F. Alvarado, M. Erbudak and P. Munz, Phys. Rev. B 14 (1976) 2740. [18] D. Buczek and S. Sastri, J. Vac. Sci. Technol. 17 (1980) 201. [19] D.F. Mitchell, G.I. Sproule and M.J. Graham, J. Vac. Sei. Technol. 18 (1981) 690. [20] M. Aeschlimann, G.L. Bona, F. Meier, M. Stampanoni, G. Zampieri and H.C. Siegmann, Appl. Phys. Lett. 49 (1986) 824. [21] R. Jansen, B.J. Nelissen, D.L. Abraham, H. van Kempen and V.A.M. Brabers, IEEE Trans. Magn. 30 (1994) 4506. [22] In the image of Fig. 7 these rows seem to appear as two parallel ones, which differs from the structure of the rows in Fig. 5. The difference is most probably an artifact produced by a double tip in case of Fig. 7.