Local atomic order and element-speciﬁc magnetic moments of Fe3Si thin ﬁlms on MgO(001)
and GaAs(001) substrates
B. Krumme,1,*C. Weis,1H. C. Herper,1F. Stromberg,1C. Antoniak,1A. Warland,1E. Schuster,1P. Srivastava,1,†
M. Walterfang,1K. Fauth,2J. Minár,3H. Ebert,3P. Entel,1W. Keune,1,4 and H. Wende1
1Fachbereich Physik and Center for Nanointegration Duisburg-Essen (CeNIDE), Universität Duisburg-Essen,
Lotharstraße 1, D-47048 Duisburg, Germany
2Experimentelle Physik IV, Physikalisches Institut, Universität Würzburg, Am Hubland, D-97072 Würzburg, Germany
3Institut für Physikalische Chemie, Universität München, Butenandtstraße 5-13, D-81377 München, Germany
4Max-Planck-Institut für Mikrostrukturphysik, Weinberg 2, D-06120 Halle, Germany
共Received 21 May 2009; revised manuscript received 8 September 2009; published 9 October 2009兲
We investigated the magnetic as well as the structural properties of Fe3Si ﬁlms on GaAs共001兲-共4⫻6兲,
GaAs共001兲-共2⫻2兲, and MgO共001兲by x-ray magnetic circular dichroism 共XMCD兲and Mössbauer spectros-
copy. From the XMCD spectra we determine averaged magnetic moments of 1.3–1.6
Bper Fe atom on the
different substrates by a standard sum-rule analysis. In addition, XMCD spectra have been calculated by using
the multiple-scattering Korringa-Kohn-Rostoker method which allows the site-speciﬁc discussion of the x-ray
spectra. The Mössbauer spectra show a highly ordered and stoichiometric growth of Fe3Si on MgO while the
growth on both GaAs substrates is strongly perturbed, probably due to diffusion of substrate atoms into the
Fe3Si ﬁlm. Therefore, we have studied the inﬂuence of Ga or As impurities on the magnetic properties of Fe3Si
by calculations using coherent-potential approximation within the Korringa-Kohn-Rostoker method. For se-
lected impurity concentrations additional supercell calculations have been performed using a pseudopotential
DOI: 10.1103/PhysRevB.80.144403 PACS number共s兲: 75.50.Bb, 78.70.Dm, 75.70.Ak, 76.80.⫹y
Since the idea of using the spin of the electrons in addi-
tion to their charge as a carrier of information, tremendous
effort has been made to create spin-polarized currents in
semiconducting materials. One approach is to make use of
spin injection, where an electric current is spin polarized by
a ferromagnetic electrode from which it is injected into a
semiconducting material.1–3Strong support for this approach
results from the possibility of combining such devices with
the existing semiconductor technology. However, there are
still many difﬁculties to overcome before such spintronic
devices may ﬁnally be established. For a high efﬁciency in
the spin-injection process the ferromagnetic electrode should
exhibit a high degree of spin polarization and a perfect inter-
face with the semiconductor.1In recent studies, Fe3Si on
GaAs turned out to be a promising ferromagnet-
semiconductor combination. Spin injection has successfully
been demonstrated with an efﬁciency of 3%.4As the quality
of the interface directly inﬂuences the spin injection, a better
understanding of the correlation between structure and mag-
netic properties at the interface is necessary to fulﬁll the task
of improving the spin injection.
Fe3Si is a binary Heusler-type compound. Such Heusler
compounds are theoretically predicted to exhibit half-
metallic behavior involving a high degree of spin polariza-
tion up to 100%.5,6However, a spin polarization of 45⫾5%
is reported so far.7Ordered Fe3Si crystallizes in a D03struc-
ture which is described by four interpenetrating fcc lattices8
whereas randomly distributed Fe and Si atoms lead to a B2
structure. In perfectly ordered Fe3Si each fcc sublattice is
occupied by only one element, i.e., three sublattices are oc-
cupied by Fe atoms and one is occupied by Si atoms 共Fig. 1兲.
This crystal structure leads to two inequivalent Fe sites: Fe
atoms on sites Bare surrounded by eight nearest-neighbor Fe
atoms yielding a magnetic moment of 2.2
atoms on Asites are surrounded by four nearest-neighbor Fe
atoms and four Si atoms exhibiting a magnetic moment of
Bper atom.9As a further consequence of the different
surroundings, Fe atoms on different sites have different hy-
perﬁne ﬁelds which can be resolved by conversion electron
Mössbauer spectroscopy 共CEMS兲.8The lattices of Fe3Si and
GaAs共001兲match almost perfectly 共lattice mismatch 0.1%
GaAs关001兴兲. Epitaxial growth has been
carefully studied in detail.10–12 In contrast, on MgO a lattice
mismatch of 5.2% occurs when Fe3Si grows rotated by 45°
FIG. 1. D03structure of bulk Fe3Si with Fe atoms on inequiva-
lent lattice sites Aand Band Si atoms on site C.
PHYSICAL REVIEW B 80, 144403 共2009兲
1098-0121/2009/80共14兲/144403共8兲©2009 The American Physical Society144403-1
on MgO共001兲, i.e., Fe3Si关001兴
MgO关110兴.13,14 With a Curie
temperature of 840 K 共Ref. 15兲Fe3Si allows for operation at
room temperature 共RT兲. In comparison to pure Fe/GaAs共001兲
its interface is thermally more stable.16 Both are crucial facts
for the ﬁnal application in devices.
In the present paper, we take a closer look at the correla-
tion between chemical ordering and magnetic properties in
thin ﬁlms of Fe3Si. We present our combined x-ray absorp-
tion spectroscopy 共XAS兲and Mössbauer studies for Fe3Si on
GaAs in comparison to MgO. We use Fe3Si on MgO共001兲
for comparison as a quality standard for which we know that
Fe3Si ﬁlms grow highly ordered on this substrate. CEMS
allows us to characterize the chemical ordering of the Fe3Si
ﬁlms and to determine the hyperﬁne ﬁeld distributions site
dependent. Complementary, the XMCD spectroscopy reveals
the averaged magnetic moment per Fe atom. In addition,
XAS and XMCD spectra have been calculated within
multiple-scattering theory. From the calculations we obtain
site-speciﬁc spectra, densities of states, and magnetic mo-
ments. Furthermore, we have studied the inﬂuence of Ga or
As impurities on the magnetic properties of Fe3Si since we
gained evidence of an interdiffusion of substrate atoms into
the Fe3Si ﬁlm on GaAs from our CEMS measurements.
II. EXPERIMENTAL ASPECTS
Films of 80 Å Fe3Si 关57 monolayer 共ML兲兴 共Ref. 17兲
were deposited in an UHV chamber at a base
pressure of 1⫻10−10 mbar on three different substrates:
MgO共001兲, Ga-terminated GaAs共001兲-共4⫻6兲, and
As-terminated GaAs共001兲-共2⫻2兲. The Fe3Si ﬁlms on MgO
and GaAs-共4⫻6兲were prepared simultaneously so that the
growth conditions for both samples were identical. Before
introducing the substrates into the preparation chamber, we
cleaned them with propanol, the GaAs additionally with ac-
etone, and dried them in a stream of N2. To obtain the
GaAs-共4⫻6兲surface reconstruction, the substrate was sput-
tered with 500 eV Ar+ions while being heated up to a tem-
perature of 870 K for 90 min. For the GaAs-共2⫻2兲surface
reconstruction an As-capped GaAs substrate was heated up
to 820 K for 10 min. No contamination of the surface with O
or C was detectable via Auger spectroscopy after this proce-
The two elements Fe and Si were coevaporated at a sub-
strate temperature of TS=520 K from a resistively heated
crucible and by electron-beam deposition, respectively. For
the CEMS measurements, 57Fe isotopes were deposited in-
stead of natural Fe. Figure 2shows the reﬂection high-energy
electron diffraction 共RHEED兲we used to monitor the Fe3Si
ﬁlm growth on GaAs-共4⫻6兲共right column兲and MgO 共left
column兲. Both pattern sets consist of three pictures compar-
ing the crucial phases of the Fe3Si growth. The substrate
reﬂections of MgO and GaAs-共4⫻6兲are shown at the top of
the columns and were measured along the 关11
Fe3Si. On MgO the substrate reﬂections directly change into
Fe3Si共001兲reﬂections within the ﬁrst ML. This indicates a
layer-by-layer growth from the very beginning. In contrast,
on GaAs-共4⫻6兲the substrate reﬂections totally vanish when
the deposition is initiated. The ﬁrst Fe3Si reﬂections appear
only at a ﬁlm thickness of about 6 ML indicating a coales-
cence of the initial islands at that thickness. Recent in situ
studies by real-time x-ray diffraction ﬁnd that the growth of
Fe3Si on GaAs共001兲begins with islands of 6 atomic layers
height and changes to a two-dimensional layer-by-layer
growth at about 14 atomic layers.10 Finally, the samples were
capped with 20 Å Au for transportation to the measuring
We investigated XAS and XMCD spectra at the dipole
beamline PM3 at BESSY in Berlin, Germany. We detected
the XAS at the Fe L2,3 edges in total-electron yield mode by
measuring the drain current of the sample. For a simulta-
neous determination of the incoming photon ﬂux I0we re-
corded the photocurrent of a Au mesh in the incident beam.
The dichroic spectra were obtained by changing the direction
of the magnetization while the helicity of the circularly po-
larized synchrotron radiation remained constant. The easy
axis of magnetization is in plane along the 关100兴direction of
the Fe3Si.18 Thus, we measured at grazing incidence with an
angle of 20° between the photon wave vector and the surface
of the sample. The sum rules derived by Thole et al.19 and
Carra et al.20 were applied to determine spin- and orbital-
resolved magnetic moments from the dichroic signal of the
Fe atoms in the sample. This method reveals magnetic mo-
ments per Fe atom averaged over the two different Fe sites.
Before normalizing the XAS to unity, the measured spectra
have been normalized to the incoming photon ﬂux and cor-
rected for a small linear background. Saturation effects have
been considered as described, e.g., in Refs. 21–23.
CEMS was applied to characterize the chemical ordering
of the Fe3Si ﬁlms. We used a 57Co source embedded in a Rh
FIG. 2. RHEED patterns of Fe3Si共001兲ﬁlm growth on
MgO共001兲共left column兲and GaAs-共4⫻6兲共right column兲measured
during growth at 520 K with an electron energy of 15 keV. The
growth was observed along the 关100兴direction of MgO and along
0兴direction on GaAs. Both correspond to the 关11
of Fe3Si 共details see text兲.
KRUMME et al. PHYSICAL REVIEW B 80, 144403 共2009兲
matrix. The samples were mounted in a gas ﬂow proportional
counter with a He-5% CH4mixture. The CEMS were all
measured in zero external ﬁeld at room temperature. Since
the surroundings of Fe atoms of type A, Fe共A兲, differs from
that of type B atoms, Fe共B兲, their hyperﬁne magnetic ﬁelds
are different. Therefore, the individual Mössbauer spectra of
the two types of Fe atoms have a different splitting of the
lines in the corresponding sextet. Hence, the experimental
spectrum will mainly show the superposition of two such
sextets,8as is illustrated in Fig. 3for a bulklike Fe3Si ﬁlm on
MgO共001兲. The spectrum was least-squares ﬁtted using the
computer code NORMOS.24 The four-ﬁtted subspectra at the
bottom 共shifted downward for clarity兲represent the two sex-
tets of Fe atoms on the two inequivalent sites whereas the
two additional subspectra with smaller intensity originate
from a not perfectly ordered D03crystal structure. The solid
line above is the sum of the sextets, ﬁtting well the experi-
mental data that are represented by the solid circles. It is thus
possible to clearly distinguish the two dominant inequivalent
Fe sites in a Mössbauer spectrum by ﬁtting sextets with dif-
ferent hyperﬁne ﬁelds to the experimental data. The hyper-
ﬁne ﬁeld, Bhf, and isomer shift 共relative to bulk bcc Fe at
of the A-site and B-site subspectra in
Fig. 3have been determined as Bhf共B兲=30.78⫾0.02 T,
共B兲=0.090⫾0.002 mm/s, Bhf共A兲= 19.96⫾0.01 T, and
共A兲=0.250⫾0.001 mm/s. The errors given are the statisti-
cal errors. These values are in good agreement with those
reported in the literature for bulk Fe3Si alloy.25
For a more detailed analysis and a clearer inspection of
the interface properties, the Mössbauer spectra were least-
squares ﬁtted in two steps: 共i兲spectra of highly chemically
ordered Fe3Si ﬁlms can be ﬁtted satisfactorily by a procedure
described by Arita et al.,26 subspectrum 共1兲in Fig. 4.共ii兲To
account for disorder at the interfaces of the ﬁlms, an addi-
tional subspectrum 关subspectrum 共2兲in Fig. 4兴with a hyper-
ﬁne ﬁeld distribution P共Bhf兲is required in the case of
Within the ﬁrst step of the ﬁtting routine the sites are
randomly occupied by Fe and Si atoms. The ﬁt to the experi-
mental data is obtained by varying this site occupancy and
calculating the corresponding CEM spectra. Long-range,
S共D03兲,S共B2兲, and short-range,
共2兲, order parameters
are computed from the occupancy of Fe sites with different
nearest and next-nearest Si atoms that leads to the best
matching simulated spectrum 关subspectrum 共1兲兴. If after this
ﬁrst step a considerable part of the CEM spectra cannot be
satisfactorily ﬁtted, an additional subspectrum with a hyper-
ﬁne ﬁeld distribution 关subspectrum 共2兲兴 is used to describe
the remaining part in the second step, accounting for an ad-
ditional, non-Fe3Si-like Fe phase.
III. COMPUTATIONAL ASPECTS
The theoretical XMCD data are obtained from ab initio
calculations employing the spin-polarized relativistic
Korringa-Kohn-Rostoker 共KKR兲method as implemented in
the SPR-KKR code.27–29 For the exchange-correlation func-
tional the local-density approximation in the parametrization
of Vosko et al.30 has been used. For the calculations of the
corresponding XAS and XMCD spectra we used an expres-
sion based on Fermi’s golden rule which was implemented
within the SPR-KKR code.27 For the KKR calculations we
have used an angular momentum expansion up to lmax =2 and
ak-point mesh of 22⫻22⫻22 共which corresponds to 843
irreducible kpoints兲. In order to study the inﬂuence of Ga
and As impurities on the magnetic structure of Fe3Si KKR
calculations within the coherent-potential approximation
共CPA兲as well as supercell calculations employing the VASP
code and the projector-augmented wave pseudopotentials31,32
have been performed. For the latter, orthorhombic supercells
with 32 atoms have been used in which one or two Fe atoms
of type B have been replaced by Ga or As, respectively. The
generalized gradient correction 共PW91兲共Ref. 33兲has been
used for the supercell calculations. A mesh of 15⫻8⫻4k
points and an energy cutoff of 381.3 eV have been used. The
FIG. 3. 共Color online兲CEMS of bulklike Fe3Si共001兲ﬁlm 共57
ML兲on MgO共001兲least-squares ﬁtted by four sextets.
FIG. 4. 共Color online兲Measured CEMS 共solid circles兲of 57 ML
Fe3Si on different substrates at RT together with ﬁtting curves
共lines兲as described in the text.
LOCAL ATOMIC ORDER AND ELEMENT-SPECIFIC…PHYSICAL REVIEW B 80, 144403 共2009兲
experimental lattice constant of Fe3Si a=5.65 Å has been
used for all calculations containing Fe3Si neglecting possible
tetragonal distortions which may occur if Fe3Si is grown on
a substrate. In order to ensure that small tetragonal distor-
tions have no signiﬁcant inﬂuence on the magnetic moments
of the system we have performed KKR-CPA calculations in
which the c/aratio is varied by ⫾5%keeping the volume
ﬁxed. The maximum change in the orbital 共spin兲moments is
less than 2% 共4%兲. However, the average change in magnetic
moments is smaller being about 1.5%. This holds also in the
presence of impurities. Relaxation of the lattice constant due
to the Ga or As impurities has been neglected so far.
The characterization of the chemical ordering of the Fe3Si
ﬁlms with CEMS is shown in Fig. 4. Solid circles represent
the experimental data. The lines represent the subspectra 共1兲
and 共2兲used for the ﬁtting as described above, and the total
spectrum. On MgO it was adequate to ﬁt the experimental
data with a calculated spectrum for nearly perfectly ordered
Fe3Si and the ﬁtting was completed after the ﬁrst step as
described above. The spectra of Fe3Si ﬁlms on GaAs-共4
⫻6兲and GaAs-共2⫻2兲had to be ﬁtted with two subspectra
in order to account for the nonperfect order in the ﬁlm. Spec-
trum 共1兲is calculated following the work of Arita et al.26 and
spectrum 共2兲is the additional subspectrum with a distribu-
tion of magnetic hyperﬁne ﬁelds representing a non-Fe3Si
phase, which we attribute to an imperfect interface. This
could be related to an interdiffusion at the interface between
Fe3Si and the substrate. Hsu et al.34 were able to ascribe this
interdiffusion to Ga atoms. Such an interdiffusion of non-
magnetic impurity atoms would result in a lower magnetic
ordering and a reduced magnetic Fe moment. The chemical
order parameters that we obtained from our analysis are sum-
marized in Table. I. In the ﬁrst row of the table, the expected
values for perfectly ordered Fe3Si are given. In real samples
the order can be perturbed by the interface to the substrate or
a capping layer, thus lowering the order parameters. From
the CEM spectra of the Fe3Si ﬁlm on MgO we obtained
parameters S共D03兲and S共B2兲which are reduced by 17% and
22% compared to the ideal values. In the case of GaAs sub-
strates the change in the order parameters becomes even
more dramatic resulting in a deviation of up to 50%.
Comparison of the CEM spectra of the Fe3Si ﬁlms on the
Ga-terminated GaAs-共4⫻6兲and the As-terminated
GaAs-共2⫻2兲surfaces reveals more pronounced peaks in the
case of GaAs-共2⫻2兲indicating a better chemical ordering
on GaAs-共2⫻2兲than on GaAs-共4⫻6兲. In the case of
GaAs-共4⫻6兲subspectrum 共2兲has a spectral area of
20.8⫾0.2%whereas for GaAs-共2⫻2兲this value becomes
29.6⫾0.2%. This leads us to the conclusion that the disorder
in the ﬁlms on GaAs-共4⫻6兲is due to Ga interdiffusion at the
interface and that on GaAs-共2⫻2兲the interdiffusion is due
to Ga as well as As atoms.
57Fe CEMS of Fe3Si does not directly yield information
about the value of the magnetic moment of the Fe atoms.
Thus, we applied the XMCD spectroscopy to determine spin-
and orbital-resolved magnetic moments. Figure 5shows the
XAS as well as the XMCD spectra at the Fe L2,3 edges of the
Fe3Si ﬁlms on MgO and GaAs-共4⫻6兲together with a bulk-
like Fe ﬁlm as a reference. The L2,3 edges of the XAS spectra
are broader for the Fe3Si ﬁlms in comparison to the Fe ref-
erence. Concurrently, the maximum of the absorption signal
at the L3edge decreases by ⬃8%for Fe3Si on MgO共001兲
and by ⬃17%on GaAs-共4⫻6兲. Additionally, a shoulder oc-
curs 2 eV above the L3edge in the Fe3Si XAS. However, at
the L2edge the absorption intensity is nearly unchanged. The
broadening as well as the shoulder can be ascribed to a hy-
bridization of Fe and Si atoms in the lattice. Such a feature
has been observed in other Heusler systems by Kallmayer
et al.35 The white lines of Fe3Si on GaAs-共4⫻6兲and
GaAs-共2⫻2兲in Fig. 6exhibit no obvious difference indicat-
ing a very similar electronic structure of Fe3Si on various
For a detailed study of the structure the electronic density
of states, DOS, and the XAS of bulk Fe3Si have been calcu-
lated, allowing a site-speciﬁc analysis of the spectra. From
the DOS it is obvious that the two types of Fe atoms have a
different electronic structure. Atoms of type B have nearly
the same DOS as bulk Fe whereas the A-type atoms show a
TABLE I. Chemical order parameters for 57 ML Fe3Si on dif-
ferent substrates, determined from a comparison of measured
CEMS and CEMS simulations as described in the text.
Theoretical values 1 0.67 −0.33 −0.33
MgO 0.83⫾0.07 0.52⫾0.12 −0.23 −0.31
GaAs-共4⫻6兲0.68⫾0.002 0.32⫾0.1 −0.09 −0.38
GaAs-共2⫻2兲0.78⫾0.03 0.42⫾0.12 −0.38 −0.40
FIG. 5. 共Color online兲Normalized XAS 共top兲and XMCD 共bot-
tom兲spectra measured at the Fe L2,3 edges of 57 ML Fe3Si on MgO
and GaAs-共4⫻6兲compared to the spectra of a bulklike Fe sample
as a reference. The inset shows an enlargement of the XMCD signal
between the L2,3 edges. The spectra were measured at RT.
KRUMME et al. PHYSICAL REVIEW B 80, 144403 共2009兲
completely different behavior, see Fig. 7, which is related to
the fact that in the latter case only half of the nearest neigh-
bors are Fe atoms. This is in agreement with previous
investigations.36,37 The different properties of the two Fe
types are also translated to the x-ray spectra, see Fig. 8.
Similar to the experimentally observed shoulder 共cf. Fig. 5兲a
small bump P between 2 and 6 eV is observed in the theo-
retical spectrum. However, this feature occurs only in the L3
peak of the Fe共A兲atoms and seems to be caused by a hy-
bridization of an Fe dorbital with a sstate of Si, see Fig. 8.
The spectrum of the Fe共B兲atoms looks very similar to that of
bulk Fe showing no P-like structure at the L3peak. Further-
more, from Fig. 8it is obvious that the spectra of the two
types of Fe are slightly shifted against each other, i.e., the
maxima of the Fe共A兲L3peak occurs about 0.25 eV above the
L3peak of the Fe共B兲atoms. Such a shift of the XAS spectra
could not be observed directly in the experiment because the
energy difference is smaller than the linewidth of the L3
peak. The different position of the two XAS is accompanied
by a change in the number of dholes. The atoms of type A
have less dholes compared to bulk Fe, namely, ⌬h=−0.24
whereas the number of holes for Fe共B兲remains unchanged
compared to bulk Fe. The difference between the A and B Fe
atoms becomes also obvious from the XMCD. If we disclaim
Gaussian broadening to simulate the broadening caused by
the experimental measurement, a tiny positive signal occurs
in case of Fe共B兲共see inset of Fig. 8兲which is known from
bulk bcc Fe cf. Fig. 5.
Comparing the XMCD signal of the Fe reference with
those of the Fe3Si ﬁlms in Figs. 5and 6no features can be
found which allow a deconvolution of the magnetic contri-
butions of the different Fe sites. Only a decrease in the
XMCD intensity can be observed for Fe3Si ﬁlms. For Fe3Si
on MgO the XMCD signal corresponding to the L3edge is
decreased by ⬃15%. In the case of Fe3Si on GaAs-共4⫻6兲
this difference becomes ⬃31%. Even a change in the surface
reconstruction to GaAs-共2⫻2兲results in a further reduction
in the XMCD signal to ⬃38%. The diminution of the XMCD
goes along with diminishing magnetic moments obtained
from the sum-rule analysis shown in Table. II. The magnetic
moment on MgO calculated with the sum rules is 1.6
is in very good agreement with results from the literature.9
On GaAs we obtained decreased moments. Whereas the
magnetic moment on GaAs-共4⫻6兲is reduced by
B兲, which is within the estimated error bar of 10%
for the sum rules, on GaAs-共2⫻2兲we obtained a moment
decreased by ⬃19%共1.3
B兲. Taking a closer look at the
XMCD signal between the L2,3 edges reveals a change in the
spectral trend. The inset of Fig. 5shows an enlargement of
this feature. Contributions to the XMCD signal in this energy
range are ascribed to sp-hybridized states or spd-wave
mixing.23 Pure Fe has a clear positive XMCD signal in this
FIG. 6. 共Color online兲Normalized XAS 共top兲and XMCD 共bot-
tom兲measured at RT at the Fe L2,3 edges of 57 ML Fe3Si on
FIG. 7. 共Color online兲Calculated spin-resolved density of states
of Fe3Si. The inset shows the total 共not spin-resolved兲DOS of the
region above the Fermi energy 共EF=0兲being relevant for the cal-
culation of XMCD spectra.
FIG. 8. 共Color online兲Calculated XAS 共top兲and XMCD 共bot-
tom兲spectra of Fe in bulk Fe3Si 共EF=0兲. The inset is an enhance-
ment of the energy region between the L3and L2peak. In order to
make these tiny features visible we disclaim smoothing of the data.
LOCAL ATOMIC ORDER AND ELEMENT-SPECIFIC…PHYSICAL REVIEW B 80, 144403 共2009兲
range, in contrast to Fe3Si, for which the signal almost com-
pletely vanishes. So one can conclude that the sp hybridiza-
tion in Fe3Si is different to that of pure Fe. We attribute this
difference to the effect of the Si bonding with Fe which
happens by 4selectrons which can be also seen in the inset
of Fig. 7at an energy of 4 eV.
A possible explanation for the reduced magnetic moments
in Fe3Si/GaAs共001兲is the diffusion of As or Ga atoms from
the substrate in the ferromagnet. The inﬂuence of alike dif-
fusion is studied by placing Ga and As impurities in bulk
Fe3Si. Although the lattice mismatch between Fe3Si and the
GaAs substrate is rather small this should be a reasonable
model. We have used two different methods to investigate
the inﬂuence of impurities on the magnetic properties of
Fe3Si, namely, CPA method where Fe and impurity atoms
partially share the same lattice site and supercell calculations
in which single Fe atoms are replaced by Ga or As atoms.
Impurity concentrations up to 10% have been taken into ac-
count, whereby it turned out from our calculations that Ga
and As preferably occupy Fe共B兲sites. Impurities on the low
moment Fe共A兲sites are less important because one has to
effort more energy to place the impurities on Fe共A兲instead
of Fe共B兲sites. In order to ﬁgure out the size of possible
impurity concentrations in Fe3Si the mixing energy
Emix =EFe3−xYxSi −共3−x兲EFe +xEY+ESi 共1兲
has been calculated, which determines whether the alloy
Fe3−xYxSi with Ybeing Ga or As is stable against decompo-
sition. In Eq. 共1兲the energies on the right side correspond to
the calculated ground-state energies of the elements, which is
a reasonable choice in case of Ga. For As the situation is
more complex because there exist a huge number of Fe-As
alloys such that the alloy may decompose in one or more of
these alloys instead of its elements. However, Eq. 共1兲gives
evidence of the amount of the impurity concentration. In
case of Ga the mixing energies are negative if the impurity
concentration does not exceed 3.3%. The As-containing alloy
seems to be stable for all investigated concentrations, which
may be related to the above-mentioned problem of the proper
choice of the decomposition components.
In case of 3.33% Ga or As the energy difference between
the impurities sitting on Fe共B兲or Fe共A兲sites amounts to
138.5 and 182.7 meV/f.u., respectively. The same trend is
observed for the supercell calculations. An impurity concen-
tration of 4.17% 共i.e., one Fe atom is replaced by an impu-
rity兲of Ga 共As兲on Fe共A兲sites leads to a 157.4 meV/f.u.
共237.8 meV/f.u.兲higher energy compared to impurities on
Fe共B兲sites. Independent which method—CPA or
supercells—has been used the magnetic moments decrease in
equal measure with increasing impurity concentration, see
Fig. 9. Assuming an impurity concentration of 3.33% the
average spin moment is reduced by 0.12
of Ga 共As兲. The absolute values of the orbital moments ob-
tained from plain spin density-functional calculations are al-
ways too small compared to experiment. However, we ob-
serve the same trend as for the spin moments, i.e., the orbital
moments are slightly smaller compared to the value obtained
for Fe3Si. However, the above discussed results suggest that
the magnetic moments of Fe3Si/GaAs-共4⫻6兲and −共2⫻2兲
are related to alloy formation at the interface. An impurity
concentration of 3.3% seems to be sufﬁcient to decrease the
average magnetic moment by about 0.1
B, which is of the
same magnitude as the change in the measured magnetic
moments, see Table II. The even smaller magnetic moment
of the 共2⫻2兲-reconstructed surface may be related to addi-
tional As diffusion, which can be larger as compared to Ga
diffusion. Although the absolute value of the magnetic mo-
ments is decreased by Ga and As impurities this has only
minor inﬂuence on the spin polarization. The spin polariza-
tion is mainly determined by the difference of the majority
N↑共EF兲and minority density of states N↓共EF兲at the Fermi
From Eq. 共2兲we obtained a spin polarization of −42.3%for
pure Fe3Si which is close to the value known from the
literature.7For Fe2.9Ga0.1Si 共impurity concentration=3.33%兲
TABLE II. Total averaged magnetic moment per Fe atom and
ratio of orbital moment mlto spin moment msas obtained from
sum-rule analysis for 57 ML Fe3Si. We estimate the error bar in the
order of 10%.
Substrate mtot in
Fe reference 2.2 0.04
MgO 1.6 0.09
FIG. 9. 共Color online兲共a兲Spin and 共b兲orbital magnetic mo-
ments of Fe3−xYxSi 共Y=As,Ga兲depending on the impurity concen-
tration xin percent. Results marked by circles and squares obtained
from KKR-CPA calculations. Stars denote the magnetic moments
derived from supercell calculations 共VASP兲for Fe共24−n兲YnSi8
KRUMME et al. PHYSICAL REVIEW B 80, 144403 共2009兲
the spin polarization still amounts to −36.8%which is close
to the value obtained for bulk Fe3Si. The high degree of spin
polarization as well as the weak dependence between spin
polarization and interface quality in the system
Fe3Si/GaAs共001兲make this system interesting for further
In summary, we have investigated structural and magnetic
properties of Fe3Si on GaAs and MgO by combining XMCD
and Mössbauer measurements with calculations within
multiple-scattering theory. From CEMS we gain evidence for
chemical disorder of Fe3Si on GaAs substrates. Measured
and calculated XMCD spectra match well even in details of
the ﬁne structure between the L3and L2edges, thus yielding
well agreeing averaged Fe moments. As one result of the
calculations we obtain the different contributions to the
XMCD from Fe on the two inequivalent sites, which are not
distinguishable in the experimental spectra. To investigate
the inﬂuence on the magnetic properties by diffusion of sub-
strate atoms, we have carried out KKR calculations consid-
ering Ga and As impurities with various concentrations. Con-
cluding, our results indicate an interdiffusion at the interface
of Ga atoms from the substrate, although the spin polariza-
tion of Fe3Si is not dramatically affected.
We thank U. v. Hörsten for help with sample preparation
and CEMS measurements. We gratefully acknowledge sup-
port during our beamtimes by BESSY staff. Financially sup-
ported by DFG 共SFB 491 and SFB 445兲and BMBF 共Grant
No. 05 ES3XBA/5兲.
*Corresponding author; email@example.com
†Permanent address: Nanostech Laboratory, Indian Institute of
Technology Delhi, New Delhi 110 016, India.
1G. Schmidt and L. W. Molenkamp, Semicond. Sci. Technol. 17,
2S. Hövel, N. C. Gerhardt, M. R. Hofmann, F.-Y. Lo, A. Ludwig,
D. Reuter, A. D. Wieck, E. Schuster, H. Wende, W. Keune, O.
Petracic, and K. Westerholt, Appl. Phys. Lett. 93, 021117
3S. Hövel, N. C. Gerhardt, M. R. Hofmann, F.-Y. Lo, D. Reuter,
A. D. Wieck, E. Schuster, W. Keune, H. Wende, O. Petracic, and
K. Westerholt, Appl. Phys. Lett. 92, 242102 共2008兲.
4A. Kawaharazuka, M. Ramsteiner, J. Herfort, H.-P. Schönherr,
H. Kostial, and K. H. Ploog, Appl. Phys. Lett. 85, 3492 共2004兲.
5R. A. de Groot, F. M. Mueller, P. G. van Engen, and K. H. J.
Buschow, Phys. Rev. Lett. 50, 2024 共1983兲.
6V. N. Antonov, D. A. Kukusta, A. P. Shpak, and A. N. Yaresko,
Condens. Matter Phys. 11, 627 共2008兲.
7A. Ionescu, C. A. F. Vaz, T. Trypiniotis, C. M. Gürtler, H.
García-Miquel, J. A. C. Bland, M. E. Vickers, R. M. Dalgliesh,
S. Langridge, Y. Bugoslavsky, Y. Miyoshi, L. F. Cohen, and K.
R. A. Ziebeck, Phys. Rev. B 71, 094401 共2005兲.
8M. B. Stearns, Phys. Rev. 168, 588 共1968兲.
9W. A. Hines, A. H. Menotti, J. I. Budnick, T. J. Burch, T. Li-
trenta, V. Niculescu, and K. Raj, Phys. Rev. B 13, 4060 共1976兲.
10 V. M. Kaganer, B. Jenichen, R. Shayduk, W. Braun, and H.
Riechert, Phys. Rev. Lett. 102, 016103 共2009兲.
11 V. M. Kaganer, B. Jenichen, R. Shayduk, and W. Braun, Phys.
Rev. B 77, 125325 共2008兲.
12 J. Herfort, H.-P. Schönherr, and K. H. Ploog, Appl. Phys. Lett.
83, 3912 共2003兲.
13 D. Kmiec, B. Sepiol, M. Sladecek, M. Rennhofer, S. Stankov, G.
Vogl, B. Laenens, J. Meersschaut, T. Ślezak, and M. Zajkac,
Phys. Rev. B 75, 054306 共2007兲.
14 Kh. Zakeri, I. Barsukov, N. K. Utochkina, F. M. Römer, J. Lind-
ner, R. Meckenstock, U. von Hörsten, H. Wende, W. Keune, M.
Farle, S. S. Kalarickal, K. Lenz, and Z. Frait, Phys. Rev. B 76,
15 Y. Nakamura et al.,Landolt-Börnstein, New Series Vol. III/19c
共Springer, Berlin, Germany, 1988兲.
16 J. Herfort, H.-P. Schönherr, A. Kawaharazuka, M. Ramsteiner,
and K. H. Ploog, J. Cryst. Growth 666-670, 278 共2005兲.
17 We call each layer of atoms in the crystal structure 1 monolayer
共ML兲, i.e. the schematic view in Fig. 1represents 5 ML in our
18 K. Lenz, E. Kosubek, K. Baberschke, H. Wende, J. Herfort, H.-P.
Schönherr, and K. H. Ploog, Phys. Rev. B 72, 144411 共2005兲.
19 B. T. Thole, P. Carra, F. Sette, and G. van der Laan, Phys. Rev.
Lett. 68, 1943 共1992兲.
20 P. Carra, B. T. Thole, M. Altarelli, and X. Wang, Phys. Rev. Lett.
70, 694 共1993兲.
21 R. Nakajima, J. Stöhr, and Y. U. Idzerda, Phys. Rev. B 59, 6421
22 J. Hunter Dunn, D. Arvanitis, N. Mårtensson, M. Tischer, F.
May, M. Russo, and K. Baberschke, J. Phys.: Condens. Matter
7, 1111 共1995兲.
23 W. L. O’Brien and B. P. Tonner, Phys. Rev. B 50, 12672 共1994兲.
24 R. A. Brand, Nucl. Instrum. Methods Phys. Res. B 28, 398
25 M. B. Stearns, Phys. Rev. 129, 1136 共1963兲.
26 M. Arita, S. Nasu, and F. E. Fujita, Trans. Jpn. Inst. Met. 26, 710
27 H. Ebert, Rep. Prog. Phys. 59, 1665 共1996兲.
28 The MUNICH SPR-KKR package, version 3.6, H. Ebert et al., http://
29 H. Ebert, Lect. Notes Phys. 535, 191 共2000兲.
30 S. H. Vosko, L. Wilk, and M. Nussiar, Can. J. Phys. 58, 1200
31 G. Kresse and J. Furtmüller, Comput. Mater. Sci. 6,15共1996兲.
32 P. E. Blöchl, Phys. Rev. B 50, 17953 共1994兲.
33 J. P. Perdew and Y. Wang, Phys. Rev. B 45, 13244 共1992兲.
34 Y. L. Hsu, Y. J. Lee, Y. H. Chang, M. L. Huang, Y. N. Chiu, C.
C. Ho, P. Chang, C. H. Hsu, M. Hong, and J. Kwo, J. Cryst.
Growth 301-302, 588 共2007兲.
35 M. Kallmayer, H. J. Elmers, B. Balke, S. Wurmehl, F. Emmer-
ling, G. H. Fecher, and C. Felser, J. Phys. D 39, 786 共2006兲.
LOCAL ATOMIC ORDER AND ELEMENT-SPECIFIC…PHYSICAL REVIEW B 80, 144403 共2009兲
36 Kh. Zakeri, S. J. Hashemifar, J. Lindner, I. Barsukov, R. Meck-
enstock, P. Kratzer, Z. Frait, and M. Farle, Phys. Rev. B 77,
37 N. I. Kulikov, D. Fristot, J. Hugel, and A. V. Postnikov, Phys.
Rev. B 66, 014206 共2002兲.
KRUMME et al. PHYSICAL REVIEW B 80, 144403 共2009兲