Content uploaded by Subhankar Bedanta
Author content
All content in this area was uploaded by Subhankar Bedanta
Content may be subject to copyright.
IOP PUBLISHING JOURNAL OF PHYSICS: CONDENSED MATTER
J. Phys.: Condens. Matter 20 (2008) 425205 (8pp) doi:10.1088/0953-8984/20/42/425205
FeSi diffusion barriers in
Fe/FeSi/Si/FeSi/Fe multilayers and
oscillatory antiferromagnetic exchange
coupling
FStromberg
1, S Bedanta1,CAntoniak
1, W Keune1,2and H Wende3
1Fachbereich Physik, Universit¨at Duisburg-Essen (Campus Duisburg), D-47048 Duisburg,
Germany
2Max-Planck-Institut f¨ur Mikrostrukturphysik, D-06120 Halle, Germany
3Fachbereich Physik and Center for Nanointegration (CeNIDE), Universit¨at Duisburg-Essen
(Campus Duisburg), D-47048 Duisburg, Germany
E-mail: fstromberg@gmx.de
Received 11 July 2008, in final form 21 August 2008
Published 16 September 2008
Online at stacks.iop.org/JPhysCM/20/425205
Abstract
We study the diffusion of 57Fe probe atoms in Fe/FeSi/Si/FeSi/Fe multilayers on Si(111)
prepared by molecular beam epitaxy by means of 57Fe conversion electron M ¨ossbauer
spectroscopy (CEMS). We demonstrate that the application of FeSi boundary layers
successfully inhibits the diffusion of 57 Fe into the Si layer. The critical thickness for the
complete prevention of Fe diffusion takes place at a nominal FeSi thickness of tFeSi =10–12 ˚
A,
which was confirmed by the evolution of the isomer shift δof the crucial CEM subspectrum.
The formation of the slightly defective c-FeSi phase for thicker FeSi boundary layers (∼20 ˚
A)
was confirmed by CEMS and reflection high-energy electron diffraction (RHEED).
Ferromagnetic resonance (FMR) shows that, for tFeSi =0–14 ˚
A, the Fe layers in all samples are
antiferromagnetically coupled and we observe an oscillatory antiferromagnetic coupling
strength with FMR and superconducting quantum interference device (SQUID) magnetometry
for varying FeSi thickness with a period of ∼6˚
A.
(Some figures in this article are in colour only in the electronic version)
1. Introduction
Due to the strong antiferromagnetic (AF) exchange coupling
(EC) between the Fe layers in Fe/Si/Fe layered structures,
which can be even stronger than in purely metallic
multilayers [1], there has been increasing interest in these
systems during the last few years [2–4]. However, the
underlying mechanism for this unusual EC still remains
elusive. The experimental complications arise from the high
reactivity of the Fe/Si interface, even at room temperature (RT).
Many studies on the phase formation upon depositing
Fe on Si substrates were conducted in the last few years.
Depending on the thickness of the Fe layer, different phases
(stable and metastable) are formed [5–9]. The common
observations are the formation of a disordered structure for
low Fe coverages (about 3 monolayers (ML) Fe), which has
a composition close to FeSi. Upon further evaporation of
Fe an amorphous magnetic phase similar to Fe3Si evolves
(starting from about 8 ML Fe). For about 20 ML or more Fe
coverage one obtains a structure consisting of the disordered
FeSi interface, α-Fe(Si) and pure α-Fe. Upon annealing at
intermediate temperatures the metastable c-FeSi phase forms,
and after annealing at higher temperatures -FeSi and β-
FeSi2emerge. The direct observation of the metastable c-
Fe1−xSi phase at the Fe/Si interfaces was a major issue in
the notion that this phase is responsible for mediating the EC
across the Si layer in Fe/Si/Fe sandwiches or multilayers [10].
This was based on band structure calculations by Moroni
et al [11]. Their result showed a peak of the density of
states 0.2 eV above the Fermi level for both stoichiometric
0953-8984/08/425205+08$30.00 ©2008 IOP Publishing Ltd Printed in the UK1
J. Phys.: Condens. Matter 20 (2008) 425205 F Stromberg et al
c-FeSi and defective c-Fe1−xSi. The EC was thought to
be generated by sd-mixing within the Anderson model. On
the other hand, Imazono et al [12] applied SXF (soft x-ray
fluorescence spectroscopy) to detect the Fe1−xSixcomposition
in sputtered Fe/Si multilayers and proposed a model implying
the formation of amorphous FeSi2and Fe3Si in the nominally
13 ˚
A thick amorphous Si interlayers. The EC was found to be
antiferromagnetic (AF) for this Si thickness (and also for 10
and 15 ˚
A a-Si). Most interestingly, Endo et al [13]showed
that initially non-coupled Fe/Si/Fe multilayers with a thick a-
Si layer convert to an AF coupled state after annealing. The
main conclusion was that the EC is mediated by crystalline
silicides in the interlayer. Later, these researchers successfully
applied the quantum interference model [14] to describe
the temperature dependence of the bilinear and biquadratic
coupling constants for samples with different Si content,
ranging from insulating to metallic behavior [15]. A very
strong EC for nominally pure Si interlayers and a simultaneous
exclusion of the formation of metallic c-FeSi at the interfaces
was found by B¨urgler et al [16]. They observed various
phases (from Fe3Si to semiconducting FeSi2)with soft x-ray
emission (SXE) and near-edge x-ray absorption spectroscopy
(NEXAFS). Resistivity measurements perpendicular to the
film plane showed a nonlinear I–Vcharacteristic which was
assigned to an insulating or semiconducting material in the
spacer region. Using wedge-type samples B¨urgler et al could
also measure the dependence of the EC for different interlayer
thicknesses and compositions (Fe1−xSix,x=0.5–1). For
x=1 (nominally pure Si interlayers) the coupling strength is
exponentially decaying, and for x=0.5(Fe
0.5Si0.5interlayers)
it exhibits an oscillating behavior and the overall coupling
strength was reduced compared to the first case. Other
groups also prepared epitaxial Fe/Fe0.5Si0.5/Fe sandwiches or
multilayers which showed an exponential decrease of the EC
with increasing Fe0.5Si0.5thickness [17–20], in contrast to [16]
and [21]. Usually these discrepancies are ascribed to deviations
in the stoichiometry of the interlayer and the smoothness of the
interfaces which were claimed to be very good in [16,21]. It is
interesting to note how small deviations in the stoichiometry of
the interlayers from the Fe0.5Si0.5composition can have drastic
effects on the nature of the EC. For example, Croonenborghs
et al [17] worked with a nominal composition of Fe0.5Si0.5for
the interlayer, whereas Gareev et al [21] applied a composition
of Fe0.56Si0.44. In the first case an exponential decay of the
coupling strength with increasing FeSi thickness was observed,
and in the second case an oscillating behavior. Because the
strongest AF coupling was always observed for nominally
pure Si layers Gareev et al [1] rejected the original idea that
metallic iron silicides at the interfaces are responsible for the
strong EC in Fe/Si/Fe layers. For nominally pure Si the EC
should be short ranged and exponentially decaying (according
to the quantum interference model). Gareev et al [22,23]
indeed achieved a record EC of >8mJm
−2by inserting a
c-FeSi boundary layer at the bottom of Fe/Si/Fe structures.
Furthermore, the AF coupling maximum was shifted to smaller
Si thicknesses. The reason for these two results was attributed
to reduced interdiffusion and the prevention of pinholes by
the boundary layer, but direct experimental proof, at least
Figure 1. Illustration of the diffusion process at the Fe/FeSi/Si
interfaces. The 57Fe probe atoms experience different isomer shifts δ
depending on the Si content of their environment.
for the diffusion, was unavailable at this time although the
growth of all layers was shown to be epitaxial by LEED. Most
interestingly the position of the coupling maximum was not
shifted while varying the thickness of the bottom FeSi layer,
and the onset of FM coupling was also not changed [23].
The goal of the present work is to shed light on
the effectiveness of FeSi diffusion barriers in Fe/FeSi/Si
multilayers and to find the effective FeSi thickness at
which Fe diffusion into the Si layer is completely inhibited.
Furthermore we want to study at which FeSi layer thickness
the metastable c-FeSi phase forms and if additional silicide
phases are observed. Additionally, the correlation of the
exchange coupling with FeSi thickness is examined with
ferromagnetic resonance (FMR) and superconducting quantum
interference device magnetometry (SQUID). We conducted a
systematic study on Fe/Fe0.5Si0.5/Si/Fe0.5Si0.5/Fe multilayers
grown on Si(111) by molecular beam epitaxy (MBE). The
nominal Fe0.5Si0.5layer thickness was varied from 0 to 14 ˚
A.
57Fe conversion electron M ¨ossbauer spectroscopy (CEMS)
in combination with the tracer layer technique [24–28]was
applied at RT in order to investigate the diffusion of Fe and Si at
the Fe/Fe0.5Si0.5/Si boundaries, which is illustrated in figure 1.
Due to the isotope-selective detection of the conversion
electrons the atomistic information about magnetism and
structure in layers and at interfaces can be obtained with a
high depth selectivity (typically a few atomic layers) using
57Fe enriched thin layers. The various Fe–Si phases can be
well distinguished by M¨ossbauer spectroscopy because they
differ in their hyperfine interactions. The most important
M¨ossbauer hyperfine-interaction parameters are the isomer
shift δ, the quadrupole splitting EQ(observable in non-
magnetic materials) and the magnetic Zeeman splitting caused
by the hyperfine magnetic field, Bhf,atthe57Fe nucleus [29]. δ
depends on the total s-electron density at the 57Fe nuclei in the
sample, EQis a measure of the deviation from local cubic
symmetry around the 57Fe atom that leads to an electric field
gradient (EFG), and Bhf originates from the total electron spin
density at the 57Fe nucleus and is observable in magnetically
ordered phases or in materials with slow electronic Fe-3d spin
2
J. Phys.: Condens. Matter 20 (2008) 425205 F Stromberg et al
relaxation. Complementary results were obtained from x-ray
diffraction (XRD), reflection high-energy electron diffraction
(RHEED), SQUID magnetometry and FMR.
2. Experimental details
All samples were grown in ultrahigh vacuum (UHV) by MBE
under the same conditions. The base pressure was 1 ×
10−10 mbar. For each sample the relevant periods of the
multilayers were repeated five times. Therefore each sample
contains four pairs of iron layers which are separated by the
interlayers. Prior to the deposition the Si(111) substrates were
rinsed in acetone, ethanol and 20% HF acid and then loaded
into the UHV system. Finally they were heated to 900 ◦C
in UHV to remove the silicon oxide. In order to minimize
Si diffusion from the Si substrate and achieve initial epitaxial
growth, the first step in the preparation was the evaporation of
10 ˚
A thick FeSi2buffer layers which were carefully annealed
for 15 min at 420 ◦C in order to promote the formation of the
γ-FeSi2phase [30]. All samples were capped with 20 ˚
ACrfor
protection. Three types of samples were prepared which differ
in the way the 57Fe tracer layer was incorporated (see figure 2).
Samples of type A contain a 6 ˚
Athick57Fe tracer layer
directly below the 56FeSi boundary layer of thickness tFeSi,
which was varied between 0 and 14 ˚
A. Samples of type
B contain a 6 ˚
Athick56Fe layer separating the Si layer
and the 57Fe tracer layer. In samples of type C, the 57FeSi
boundary layer itself is enriched with 57Fe. The Si and Fe
thicknesses (56Fe +57Fe) were kept constant at 20 ˚
A and 21 ˚
A,
respectively, for all samples, except for samples of type B.
Samples of type A were used to study the diffusion of 57Fe into
the a-Si layer. Samples of type B were investigated to study
the diffusion of Si into the 57Fe layer and, finally, samples
of type C were analyzed to study the phase formation in the
57FeSi layer. Nominally, the substrate temperature during
deposition was kept near room temperature (∼50 ◦C). The
deposition rates for Fe were 0.05 ˚
As
−1,andforSi0.085 ˚
As
−1,
for both the pure layers and the boundary layers. The FeSi
layers were prepared as artificial multilayers (digital alloys)
with an effective FeSi rate of 0.133 ˚
As
−1.57Fe, 56Fe and
Cr were evaporated from resistively heated evaporation cells,
while Si was evaporated from an electron gun. The depleted
56Fe metal contained a residual isotopic composition of 0.2%
57Fe and had a purity of 99.94 at.%. 57Fe was isotopically
enriched to 95.5% and had a purity of 99.95 at.%. The
film thicknesses and rates were measured with independent
calibrated quartz crystal oscillators. The relative errors of the
Fe, FeSi and Si thicknesses are 4%. For the alloy, in order
to obtain the real FeSi thickness from the nominal (Fe +Si)
thickness tFeSi measured by the quartz crystal oscillators,
one has to apply a factor of 0.7 to the nominal thickness
obtained from the quartz crystals, taking proper account of
the different mass densities of Fe, Si and FeSi. The FeSi
thickness tFeSi given here is the nominal thickness, if not stated
otherwise. Additionally, the film growth was monitored by
RHEED with an e−-beam energy of 15 kV and a current of
30 μA. For CEMS at RT, a 57Co source embedded in an
Rh matrix was used. The samples were mounted in a gas-
flow proportional counter which used a He-5% CH4mixture.
Figure 2. The geometrical structure of the prepared samples of type
A, B and C.
The CEM spectra, all measured in zero external field, were
least-squares fitted with the computer program NORMOS [31].
All isomer shifts are given relative to α-Fe at RT. θ–2θhigh-
angle and low-angle x-ray diffraction (XRD) were performed
with Cu Kαradiation. The magnetometry measurements were
carried out in a commercial SQUID magnetometer (Quantum
Design MPMS). FMR was conducted at RT with a microwave
frequency of f0=9.8 GHz, and with the magnetic field
applied in the sample plane. The field was continuously varied
from 0.0 to about 0.2 T.
3. Results and discussion
3.1. Structural analysis
High-angle XRD measurements were performed to determine
the layer structure and its epitaxial relationship. All θ–2θ
scans of our samples (not shown) exhibit a weak broadened
reflection at 2θ≈45◦, which was assigned to Fe(110)
with slightly reduced lattice constants. Using thicker FeSi
layers the lattice constant deduced from the Fe(110) reflection
approaches the value of pure iron of 2.87 ˚
A. The same effect
was observed by increasing the Fe thickness itself, which
leads to the main conclusion that an increasing FeSi thickness
reduces the diffusion of Fe into the Si layer. This is supported
by the CEMS measurements (see below). Figure 3shows
typical low-angle x-ray reflectivity data of samples of type A
(tFeSi =10 ˚
A) and type B (tFeSi =0˚
A). All reflectivity curves
exhibit superstructure Bragg reflections from the multilayer
periodicity and fast oscillating Kiessig fringes due to the total
thickness interference. Both observations indicate that our
samples have a good multilayer periodicity and flat surfaces.
The enhancement of the oscillations for tFeSi =10 ˚
A indicates
that by inserting FeSi layers the interface quality is improving.
RHEED measurements during deposition of the mul-
tilayers gave interesting insights into the growth mode
and the morphology of our samples. Epitaxial island
growth was found for the very first Fe and FeSi layers
(figures 4(a) and (b)). The following Si formed an amorphous
layer (not shown). The second FeSi layer also grew in its
3
J. Phys.: Condens. Matter 20 (2008) 425205 F Stromberg et al
Figure 3. Typical low-angle x-ray reflectivity of sample type A with
a nominal FeSi thickness of tFeSi =10 ˚
A (lower curve) and sample
type B with tFeSi =0˚
A (upper curve).
Figure 4. RHEED patterns of the first sample layers (images of the
buffer layer and the amorphous Si and FeSi layers havebeen
omitted). (a) 15 ˚
A56Fe, (b) 22 ˚
AFeSi,(c)9 ˚
AFe,(d)10 ˚
AFeSi.
amorphous state (not shown). Interestingly, the following
second Fe and third FeSi layers grew in polycrystalline states,
which was concluded from the observation of Debye–Scherrer
rings (figures 4(c) and (d)).
These results demonstrate that the FeSi/Si and Si/FeSi
interfaces are inequivalent, which was also observed in [10].
Due to the epitaxial growth for the first Fe and FeSi layer
Figure 5. Typical CEM spectra of sample type C with tFeSi =6˚
A(a)
and tFeSi =20 ˚
A (b). The spectra were fitted with two subspectra: a
dominant quadrupole doublet D and a weak magnetic hyperfine field
distribution P(Bhf )(right-hand side).
it was possible to extract information about the relaxation of
their respective in-plane lattice parameters. For this purpose
we used the Bragg formula df=(ks/kf)ds, with dfand ds
being the in-plane atomic distances for film and substrate,
respectively, and kfand ksthe positions in kspace of the first-
order diffraction spots relative to the zero-order reflection for
film and substrate, respectively. For Fe, the relaxation process
was found to end at a thickness of approx. 10 ˚
A, starting from
a value of 2.73 ˚
Aofγ-FeSi2. In contrast, the relaxation of
the FeSi film is only finished up to 18 ˚
A, with a final value of
the lattice constant of 2.77 ˚
A, matching closely the value of
c-FeSi [32,33].
3.2. M¨
ossbauer spectroscopy
To investigate the structural phase of the FeSi diffusion barriers
we analyze CEM spectra of sample type C (57FeSi tracer
layers), as shown in figure 5. They were fitted with a
dominant quadrupole doublet D and a weak subspectrum with
a magnetic hyperfine distribution P(Bhf). For a thickness of
tFeSi =6˚
A (figure 5(a)) the values for the isomer shift δ
and the quadrupole splitting EQof the doublet are δ=
0.20(1)mm s−1and EQ=0.60(1)mm s−1. A doublet
with virtually identical hyperfine parameters was also detected
in Fe/Si multilayers [3]andFe/Fe
xSi1−xmultilayers [2].
However, for tFeSi =20 ˚
A (figure 5(b)) one obtains values of
δ=0.22(1)mm s−1and EQ=0.39(2)mm s−1.
In figures 5(a) and (b) the area (relative intensity) of the
subspectrum with the distribution P(Bhf)amounts to only 10%
and 18%, respectively, of the total spectral area, providing only
a weak contribution. The spectral parameters of the c-FeSi
phase are δ=0.26 mm s−1and EQ=0.15(1)mm s−1,
according to the literature [18,33–37]. This metastable phase
crystallizes in the B2 (or CsCl) structure with space group
Pm¯
3m[33]. This is a bcc lattice with Si at the origin and
Fe at (1/2,1/2,1/2). Fe and Si are both coordinated with
eight nearest neighbors of different kinds. The corresponding
bulk stable phase is -FeSi which is also cubic but has the
4
J. Phys.: Condens. Matter 20 (2008) 425205 F Stromberg et al
space group P213(T4)[33]. In the ideally cubic c-FeSi phase
no quadrupole splitting should be observed. One attempt to
explain our large EQvalue for tFeSi =6˚
A (figure 5(a)) is
the lattice deformation of the film [35], which produces an
electric field gradient, since for this small thickness the lattice
relaxation is still not yet finished, as confirmed by our RHEED
results. There exists a linear correlation [35] between lattice
mismatch and quadrupole splitting for c-FeSi films thinner or
equal to 50 ˚
Aof(−18.6%)
(mm s−1). However, taking into account our
quadrupole splitting of EQ=0.60(1)mm s−1one obtains
an unreasonable mismatch of −11.2%, which is in contrast to
the maximum attainable mismatch between c-FeSi and Fe of
−3.5%. A more reasonable explanation for our results is the
formation of a defective c-Fe0.5Si phase for thin FeSi layers. It
contains defects which consist of n=0–6 vacancies as next-
nearest neighbors of a central 57Fe atom [37]. The parameters
for n=3(δ=0.24(1)mm s−1and EQ=0.54(1)mm s−1)
are similar to ours for tFeSi =6˚
A. The strong decrease of
EQfor thicker FeSi films (figure 5(b)) and the increase of
δcan be attributed to better ordering and better homogeneity
of the crystal structure, which leads to a less disturbed c-
FeSi phase. The origin of the magnetic distribution P(Bhf)
in figure 5can be twofold. A small excess of Fe during
evaporation can lead to non-stoichiometric regions in the FeSi
layer which behave ferromagnetically. Walterfang et al [36]
observed average magnetic hyperfine fields Bhfwith values
smaller than 3 T at RT for Fe excess of 4.5 at.% in thin c-FeSi
films on MgO. This was attributed to the presence of a fraction
of a non-stoichiometric c-FexSi1−xphase in the FeSi film.
We observe much higher Bhfvalues (Bhf =20.9Tand
22.6 T for tFeSi =6˚
A and 20 ˚
A, respectively) which, in turn,
would suggest a much higher Fe excess during evaporation
than 4.5%, since Bhfscales with the Si content. Due to the
use of calibrated quartz monitors this situation is unlikely. It
is more probable that mixing at the 56Fe/57FeSi interface leads
to the formation of a ferromagnetic solution or phase similar to
Fe3Si during Fe evaporation on Si, if one exceeds a certain Fe
thickness, as described in section 1.
To elucidate the functionality of the FeSi diffusion barrier
we now analyze CEM spectra of sample types A and B (57Fe
tracer layers), as shown in figure 6. The spectra of sample
type A (figures 6(a)–(c)) were fitted with two subspectra:
a quadrupole doublet D1 and a magnetic hyperfine field
distribution P(Bhf).
The magnetic distributions P(Bhf)originate from the 57Fe
atoms at the 57Fe/56 FeSi interface which experiences an Fe-
rich environment and therefore a magnetic hyperfine field.
The spectral areas of the doublet D1 are the sum of two
contributions. The first one originates from 57Fe atoms which
diffused into the 56FeSi layer. The second one is due to
57Fe atoms which diffused through the 56FeSi layer into the
a-Si layer. One observes that, for increasing FeSi thickness
tFeSi, the relative spectral area of the doublet D1 experiences
a strong decrease: starting from 39% for tFeSi =0˚
Aand
decreasing to 21% for tFeSi =3˚
A, it finally reaches 3% for
tFeSi =14 ˚
A (figures 6(a)–(c)). The interpretation of this
result is that, for the thicker and less defective FeSi layers
(14 ˚
A), the diffusion of 57Fe is inhibited, but even for relatively
Figure 6. Typical CEM spectra of sample types A and B. The
nominal FeSi thickness for sample type A was tFeSi =0˚
A(a),
3˚
A(b)and14 ˚
A(c).The56 Fe thickness for sample type B was
tFe =6˚
A (d). Spectra of type A were fitted with two subspectra: a
quadrupole doublet D1 and a magnetic hyperfine field distribution
P(Bhf). The spectrum of type B was fitted with a magnetic hf field
distribution P(Bhf )only (right-hand side).
thin layers (3 ˚
A) which contain a higher amount of defects,
the diffusion is strongly suppressed. For the CEM spectrum
of sample type B (figure 6(d)) with a 6 ˚
A56Fe layer instead
of 56FeSi it was sufficient for the fitting to use a relatively
sharp magnetic distribution P(Bhf). The maximum of the
distribution is located at Bhf =33 T, which is identical to the
value of pure α-Fe. The sharpness of the distribution and the
absence of a doublet shows that the diffusion of Si into the
57Fe probe layer is negligible and the main diffusing species
at RT is Fe, as was also shown in [10]. CEMS measurements
on amorphous Fe1−xSixlayers [38–41]and57 Fe ion-implanted
silicon [42] showed a correlation between isomer shift δand Si
concentration x, as exhibited in figure 7. With decreasing x,
starting from high Si content, the isomer shift first drops from
∼0.2to∼0.13 mm s−1and then increases again until it reaches
the highest value of ∼0.28 mm s−1for crystalline -FeSi [33]
which corresponds to a Si content of x=0.5.
Our results for the evolution of the isomer shift δand the
relative spectral area (inset) of doublet D1 with increasing tFeSi
are shown in figure 8. Starting from tFeSi =0˚
A, δdrops
slightly from 0.23 to 0.21 mm s−1and then again increases
to 0.235 mm s−1at tFeSi =10 ˚
A. A further increase of
tFeSi is accompanied by a jump (vertical dotted line) of δto
5
J. Phys.: Condens. Matter 20 (2008) 425205 F Stromberg et al
Figure 7. Correlation between isomer shift δand silicon content xin
Fe1−xSixalloys. The dashed line is a guide for the eyes.
Figure 8. Isomer shift δof the doublet D1 (sample type A) versus
nominal FeSi thickness tFeSi . The inset displays the evolution of the
relative spectral area of the doublet D1 with increasing FeSi
thickness tFeSi .
0.27–0.28 mm s−1. Simultaneously, the relative spectral area
of D1 decreases monotonically from initially 42% down to
3% for tFeSi =14 ˚
A (inset, figure 8). The interpretation of
these results is as follows. For the region tFeSi =0–10 ˚
A
we have a superposition of quadrupole doublets with isomer
shifts of less than 0.20 mm s−1, originating from 57 Fe atoms in
the a-Si layer (Si-rich environment) and of δ=0.27 mm s−1
from the 56FeSi layer, leading to an overall isomer shift δ
0.23 mm s−1. Increasing tFeSi further inhibits the diffusion of
57Fe atoms through the FeSi layer into the a-Si. This leads
to an increase of the overall δof D1. One can compare this
progress of δdirectly with figure 7but has to keep in mind
that we are starting from high Si content, which is located
at the right-hand side of figure 7, and move to the left for
increasing tFeSi (lower Si content). The jump of δfor tFeSi ≈
11 ˚
Ato∼0.27 mm s−1corresponds to a sudden change of
the Si content in the environment of the 57Fe probe atoms to
x=0.5. This marks the critical thickness tFeSi at which no
57Fe diffuses into the a-Si layer. Since the area of D1 for
these thicknesses becomes very small (see inset figure 8)this
also means that only a negligible amount of 57Fe diffuses into
the 56FeSi layer itself. The remaining magnetic distribution
P(Bhf)corresponds to slightly disturbed α-Fe and a very small
Figure 9. Typical hysteresis loops (sample type A) at RT for
tFeSi =8˚
A (a) and tFeSi =14 ˚
A(b).
contribution of FeSi alloys with a nominal composition of
Fe55Si45 to Fe65Si35. These estimations are extracted from the
evolution of the hyperfine field distributions P(Bhf )with tFeSi
(see [36]).
3.3. SQUID magnetometry and FMR
In order to examine the strength and the sign of the exchange
coupling as a function of tFeSi in our structures we performed
SQUID and FMR measurements. Typical hysteresis loops are
showninfigure9. The first loop (figure 9(a)) for tFeSi =8˚
A
has a low remanence and large saturation field, typical for AF
coupled systems. By contrast, for tFeSi =14 ˚
A the remanence
of the hysteresis loop is higher and the sample saturates at
lower fields. This already hints at some kind of oscillations
in the exchange coupling strength. From the analysis of the
hysteresis loops at RT we obtained the parameter FAF =
1−MR/MS.MRrefers to the remanent magnetization and MS
to the saturation magnetization (MSwas taken at an external
field of 2 T for all samples). The common assumption is that
this parameter is proportional to the amount of AF coupled
regions of the sample [43]. If FAF =1 the whole sample is
AF coupled, while if FAF =0 the coupling is FM.
Care has to be taken if one extracts the strength of the AF
coupling from this parameter. However, as already pointed out
by den Broeder et al [44], this parameter is rather a measure
of an incomplete AF alignment which can be frequently
caused by pinholes, the latter causing FM coupling. A direct
confirmation of the sign of the EC can be assessed by FMR.
Due to the coupling of the Fe films across the interlayers
one observes an acoustical mode and an optical mode in the
precession of the Fe films. Since the dispersion relations are
6
J. Phys.: Condens. Matter 20 (2008) 425205 F Stromberg et al
Figure 10. FMR spectrum of sample type A (tFeSi =4.5˚
A) with two
resonance fields at Bres1 and Bres2 (f0=9.8 GHz).
different for FM or AF coupling, a distinction between these
two can be made by looking at the relative positions of the
absorption lines, taking into account that the optical mode has
always less intensity than the acoustical mode [45,46]. The
splitting of the two modes Bres provides the strength of the
coupling. A typical FMR spectrum is shown in figure 10.
Since the external field is modulated at a frequency of 100 kHz
in order to enhance the sensitivity of the measurement with
the lock-in technique, one observes differentiated Lorentzian
absorption lines.
The absorption spectrum in figure 10 was fitted with a
sum of two differentiated Lorentzian lines. The resonance
fields are found from the zero crossings of the fit (Bres1 and
Bres2, see figure 10). One observes that the optical mode
appears at higher external fields, which conclusively means
that the Fe films are AF coupled in this sample [45,46].
In fact, AF coupling was observed in all of our samples.
The strength of the coupling can be calculated via the
formula Bres =JEC/Msdfor symmetric interlayer-coupled
multilayers [46], with Bres,JEC,Msand dbeing the effective
exchange field (splitting of the acoustical and optical mode),
the exchange coupling constant, the saturation magnetization
and the thickness of one Fe layer, respectively. However, our
RHEED measurements (section 3.1) prove that we are dealing
with an asymmetric multilayer system here, where the FeSi/Si
and Si/FeSi interfaces are found to be inequivalent. As stated
explicitly in [46], in the latter case the exchange coupling
JEC is not determined any more by the simple equation for
Bres given above. Therefore we refrain from calculating
JEC from our measured Bres values. We like to emphasize,
however, that the measured Bres values provide a measure
of the relative coupling strength. The combined results of
SQUID and FMR measurements are shown in figure 11.It
is remarkable that both parameters correlate very well with
each other, which was not expected beforehand, most notably
for FAF, as discussed before. One observes a clear oscillation
of the coupling strength with maxima at tFeSi =3.5˚
Aand
tFeSi =8˚
A. There is a small damping of the oscillation, since
the FAF value at tFeSi =12 ˚
A is clearly lower than the value
at the minimum at tFeSi =6˚
A. Since our main investigation
was focused on the diffusion of Fe, which was suppressed for
thicknesses starting from 12 ˚
A, we did not investigate samples
Figure 11. FAF parameter (open diamonds) and splitting of the
resonance fields Bres (full circles) from SQUID hysteresis loops
and FMR measurements, respectively, versus nominal FeSi thickness
tFeSi.
with higher FeSi thicknesses than 14 ˚
A. The observed AF-
type oscillation is in qualitative accordance with theoretical
predictions by Herper et al [47].
It has to be mentioned that in [47], although inhomoge-
neous alloy formation at the interfaces of Fe/Si/Fe trilayers
was taken into account, the Si lattice was simulated within
a bcc structure with a lattice constant of 5.27 ˚
A. This in
turn produced a metallic Si lattice. Comparing our results
in figure 11 more closely with similar behavior observed by
B¨urgler et al [16] for pure FeSi interlayers without Si, one
observes subtle differences. The two maxima in [16]were
observed at tFeSi =18 and 39 ˚
A. We can conclude that our
first maximum is extended to a higher total thickness tSof 26 ˚
A
(tS=2tFeSi +tSi)thanin[16]. However, our second maximum
at tS=2×8˚
A+20 ˚
A=36 ˚
A is only slightly shifted in
comparison to that in [16]. A tentative interpretation of this
result could be that the interfacial disorder in our samples with
the thinnest FeSi boundary layers produces a phase shift and/or
an amplitude change of the oscillating exchange coupling.
The physical origin is an additional cosine-like term in the
exchange coupling from the specular reflection of the electron
waves at the disordered interfaces [48]. For thicker FeSi layers
this effect is negligible since the disorder is strongly reduced
due to the formation of the non-defective c-FeSi phase.
4. Conclusions
We have successfully studied the interdiffusion processes
in Fe/FeSi/Si/FeSi/Fe multilayers on Si(111) via CEMS.
We demonstrate that the insertion of FeSi boundary layers
suppresses the Fe/Si interdiffusion leading to better defined
interfaces. A complete prevention of the Fe diffusion was
found for tFeSi =10–12 ˚
A. This result allows for a reliable
investigation of interlayer exchange coupling at this or greater
FeSi thicknesses. The evolution of the slightly defective c-FeSi
phase for thicker FeSi films was confirmed by RHEED and
CEMS. All our samples exhibit oscillating antiferromagnetic
interlayer coupling, as confirmed by FMR. The oscillation
period of this coupling is approx. 6 ˚
A, which was also
observed by SQUID. Our finding is in qualitative agreement
7
J. Phys.: Condens. Matter 20 (2008) 425205 F Stromberg et al
with theoretical predictions by Herper et al [47]. Future work
will include the investigation of the coupling for variable Si
thickness, since in the present work the main focus was on the
FeSi boundary layer.
Acknowledgments
We are indebted to U v H¨orsten for his valuable technical
assistance. Fruitful discussions with M Walterfang and
R A Brand are highly appreciated. This work was supported by
the Deutsche Forschungsgemeinschaft (in part by grant nos. Ke
273/18-2 and SFB 491).
References
[1] Gareev R R, B¨urgler D E, Buchmeier M, Schreiber R and
Gr¨unberg P 2002 J. Magn. Magn. Mater. 240 235
[2] Kopcewicz M, Luci´nski T and Wandziuk P 2005 J. Magn.
Magn. Mater. 286 488
[3] Luci´nski T, Kopcewicz M, H ¨utten A, Br¨uckl H, Heitmann S,
Hempel T and Reiss G 2003 J. Appl. Phys. 93 6501
[4] Luci´nski T, Wandziuk P, Stobiecki F, Andrzejewski B,
Kopcewicz M, H ¨utten A, Reiss G and Szuszkiewicz W 2004
J. Magn. Magn. Mater. 282 248
[5] Fanciulli M, Degroote S, Weyer G and Langouche G 1997
Surf. Sci. 377 529
[6] Fanciulli M, Zenkevich A and Weyer G 1998 Appl. Surf. Sci.
123 207
[7] D´ezsi I, Fetzer C, Sz¨ucks I, Dekoster J, Vantomme A and
Caymax M 2005 Surf. Sci. 599 122
[8] Alvarez J, de Parga A L V, Hinarejos J J, de la Figuera J,
Michel E G, Ocal C and Miranda R 1993 Phys. Rev. B
47 16048
[9] Kl¨asges R, Carbone C, Eberhardt W, Pampuch C, Rader O,
Kachel T and Gudat W 1997 Phys. Rev. B56 10801
[10] Strijkers G J, Kohlhepp J T, Swagten H J M and
de Jonge W J M 1999 Phys. Rev. B60 9583
[11] Moroni E G, Wolf W, Hafner J and Podloucky R 1999
Phys. Rev. B59 12860
[12] Imazono T, Hirayama Y, Ichikura S, Kitakami O,
Yanagihara M and Watanabe M 2004 Japan. J. Appl. Phys.
43 4327
[13] Endo Y, Kitakami O and Shimada Y 1997 J. Magn. Soc. Japan.
21 541
[14] Bruno P 1995 Phys. Rev. B52 411
[15] Endo Y, Kitakami O and Shimada Y 1999 Phys. Rev. B
59 4279
[16] B¨urgler D E, Buchmeier M, Cramm S, Eisebitt S, Gareev R R,
Gr¨unberg P, Jia C L, Pohlmann L L, Schreiber R, Siegel M,
Qin Y L and Zimina A 2003 J. Phys.: Condens. Matter
15 S443
[17] Croonenborghs B, Almeida F M, Gareev R R, Rots M,
Vantomme A and Meersschaut J 2005 Phys. Rev. B
71 024410
[18] Dekoster J, Degroote S, Meersschaut J, Moons R, Vantomme A,
Botty´an L, De´ak L, Szil´agyi E, Nagy D L, Bacon A Q R and
Langouche G 1999 Hyperfine Interact. 120/121 39
[19] de Vries J J, Kohlhepp J, den Broeder F J A, Coehoorn R,
Jungblut R, Reinders A and de Jonge W J M 1997 Phys. Rev.
Lett. 78 3023
[20] Fullerton E E, Mattson J E, Lee S R, Sowers C H, Huang Y Y,
Felcher G and Bader S D 1992 J. Magn. Magn. Mater.
117 L301
[21] Gareev R R, B¨urgler D E, Buchmeier M, Olligs D,
Schreiber R and Gr¨unberg P 2001 Phys.Rev.Lett.
87 157202
[22] Gareev R R, B ¨urgler D E, Buchmeier M, Schreiber R and
Gr¨unberg P 2002 Appl. Phys. Lett. 81 1264
[23] Gareev R R, B ¨urgler D E, Buchmeier M, Schreiber R and
Gr¨unberg P 2002 Trans. Magn. Soc. Japan 2205
[24] Sauer C and Zinn W 1993 Magnetic Multilayers ed
L H Bennett and R E Watson (Singapore: World Scientific)
[25] Shinjo T and Keune W 1999 J. Magn. Magn. Mater. 200 598
[26] Macedo W A A, Sahoo B, Kuncser V, Eisenmenger J, Felner I,
Nogues J, Liu K, Keune W and Schuller I K 2004
Phys. Rev. B70 224414
[27] Stromberg F, Keune W, Kuncser V E and Westerholt K 2005
Phys. Rev. B72 064440
[28] Sahoo B, Macedo W A A, Keune W, Kuncser V E,
Eisenmenger J, Nogues J, Schuller I K, Felner I, Liu K and
R¨ohlsberger R 2006 Hyperfine Interact. 169 1371
[29] Gonser U 1975 M¨ossbauer Spectroscopy ed U Gonser
(Heidelberg: Springer)
[30] von K¨anel H, Stalder R, Sirringhaus H, Onda N and
Henz J 1991 Appl. Surf. Sci. 53 196
[31] Brand R A 1987 Nucl. Instrum. Methods Phys. Res. B28 398
[32] Al-Sharif A I, Abu-Jafar M and Qteish A 2001 J. Phys.:
Condens. Matter 13 2807
[33] Fanciulli M, Weyer G, Svane A, Christensen N E, von K¨anel H,
M¨uller E, Onda N, Miglio L, Tavazza F and Celino M 1999
Phys. Rev. B59 3675
[34] Degroote S, Vantomme A, Dekoster J and Langouche G 1995
Appl. Surf. Sci. 91 72
[35] Croonenborghs B, Almeida F M, Cottenier S, Rots M,
Vantomme A and Meersschaut J 2004 Appl. Phys. Lett.
85 200
[36] Walterfang M, Keune W, Trounov K, Peters R, R¨ucker U and
Westerholt K 2006 Phys. Rev. B73 214423
[37] Fanciulli M, Rosenblad C, Weyer G, von K¨anel H and
Onda N 1996 Thin Solid Films 275 8
[38] Elsukov E P, Konygin G N, Barinov V A and Voronina E V
1992 J. Phys.: Condens. Matter 47597
[39] Bansal C, Campbell S J and Stewart A M 1982 J. Magn. Magn.
Mater. 27 195
[40] Mangin P and Marchal G 1978 J. Appl. Phys. 49 1709
[41] Marchal G, Mangin P, Piecuch M and Janot C 1976
J. Physique Coll. 37 763
[42] S´anchez F H, van Raap M B F and Desimoni J 1990
Phys. Rev. B44 4290
[43] Inomata K, Yusu K and Saito Y 1995 Phys. Rev. Lett. 74 1863
[44] den Broeder F J A and Kohlhepp J 1995 Phys.Rev.Lett.
75 3026
[45] Zhang Z, Zhou L, Wigen P E and Ounadjela K 1994
Phys. Rev. B50 6094
[46] Lindner J, Kollonitsch Z, Kosubek E, Farle M and
Baberschke K 2001 Phys. Rev. B63 094413
[47] Herper H C, Weinberger P, Szunyogh L and Sommers C 2002
Phys. Rev. B66 064426
[48] Los V F and Los A V 2008 Phys. Rev. B77 024410
8
