PHYSICAL REVIEW B                                      VOLUME 57, NUMBER 21                                                    1 JUNE 1998-I

   Quantitative study of the interdependence of interface structure and giant magnetoresistance
                                           in polycrystalline Fe/Cr superlattices

                                                                 R. Schad*
              Laboratorium voor Vast-Stoffysika en Magnetisme, Katholieke Universiteit Leuven, B-3001 Leuven, Belgium
                      and Research Institute for Materials, KU Nijmegen, NL-6525 ED Nijmegen, The Netherlands

                                              P. Belie¨n,  G. Verbanck, and C. D. Potter
              Laboratorium voor Vast-Stoffysika en Magnetisme, Katholieke Universiteit Leuven, B-3001 Leuven, Belgium

                                                                H. Fischer
                                        Institute Laue Langevin, 38042 Grenoble Cedex 9, France

                                                      S. Lefebvre and M. Bessiere
                                       LURE, Universite´ de Paris-Sud, 91405 Orsay Cedex, France

                                              V. V. Moshchalkov and Y. Bruynseraede
              Laboratorium voor Vast-Stoffysika en Magnetisme, Katholieke Universiteit Leuven, B-3001 Leuven, Belgium
                                                       Received 15 September 1997 
                We present a quantitative characterization of the interface roughness of Fe/Cr superlattices based on specular
              and off-specular x-ray diffraction using anomalous scattering. We discuss the dependence of the amplitude of
              the giant magnetoresistance  GMR  effect, including changes in the interlayer magnetic coupling, on the
              interface structure. We observe a reduction of the GMR effect with increasing amplitude of the interface
              roughness having constant lateral correlation length. However, the physical interpretation of this clear result in
              terms of spin-dependent interface scattering remains unclear because of the unknown bulk contribution.
               S0163-1829 98 03918-6 

                       INTRODUCTION                                      configuration changes from fully antiparallel to parallel
                                                                         alignment. The latter will be easily achieved only if the ex-
   The discovery of giant magnetoresistance1  GMR  in                    ternal magnetic field is strong enough to saturate the magne-
Fe/Cr superlattices opened a new field of possible applica-              tization. The antiferromagnetic alignment at zero field, how-
tions for artificially tailored materials. The effect is explained       ever, depends  in the case of an exchange coupled
by spin-dependent scattering of the electrons at impurities or           superlattice  on the kind of the exchange coupling and on
interfaces.2­4 This spin dependence results from spin-                   superlattice imperfections in the form of pinholes. Instead of
dependent electron states and from the spin dependence of                a simple antiferromagnetic alignment, the magnetization di-
the scattering potential. For instance, the majority electrons           rections can form 90° angles between adjacent magnetic
of Fe are much stronger scattered at Cr impurities than are              layers.8 This will reduce the observed GMR by a factor of 2.9
the minority electrons.5 This leads to different resistivities           The strength of the 90° coupling is mediated by roughness of
for the parallel and the antiparallel alignment of the magne-            the interfaces10 or loose spins inside the spacer layers.11 So,
tization directions of the magnetic layers. The antiparallel             in both cases the 90° coupling indirectly links the size of the
configuration at zero strength of the external field is provided         GMR effect to the superlattice quality. Magnetic pinholes
by antiferromagnetic exchange coupling for an appropriate                will cause ferromagnetic alignment of parts of the sample
thickness of the Cr spacer layer. This configuration can be              which consequently do not contribute to the GMR effect,
forced into parallel alignment by an external field, thus re-            thus diminishing its amplitude. Not only pinholes but also
sulting in a change of the resistance. However, antiferromag-            precursors of these in the form of larger spacer layer thick-
netic exchange coupling is not a prerequisite for the obser-             ness fluctuations might lead to partially ferromagnetic align-
vation of the GMR effect since the antiparallel alignment can            ment because of local changes of the exchange coupling.
be obtained also by other methods.6,7                                    These magnetic contributions can be separated experimen-
   The burning question was and still is, how the size of the            tally from the pure electronic contributions by magnetization
GMR effect is related to the structural properties of the su-            measurements which give directly the fraction of the sample
perlattice. Here one has to distinguish between several con-             which is antiferromagnetically ordered  AFF  and the part
tributions to the GMR which are directly or indirectly linked            which does not contribute  local ferromagnetic alignment,
to the structural properties. These are contributions of  i  the         pinholes  or which contributes only partially to the GMR
magnetic structure,  ii  the spin-dependent electronic struc-             angle between magnetization directions of adjacent mag-
ture, and  iii  the spin-dependent electron scattering.                  netic layers between 0° and 180° .
   The magnetic structure is of importance because the full                 The second contribution to the GMR effect, the electronic
size of the GMR effect is only observed when the magnetic                structure, can generate a GMR effect even in defect-free


0163-1829/98/57 21 /13692 6 /$15.00                              57      13 692                      © 1998 The American Physical Society



57                                   QUANTITATIVE STUDY OF THE INTERDEPENDENCE . . .                                           13 693

point contacts with ballistic transport12 or in the limit of di-      independent ones. We present XRD spectra of high quality
luted scatterers.13,14 This contribution comes mostly from the        Fe/Cr superlattices together with simulations which deter-
asymmetry of the Fermi velocities of the two spin channels.           mine the values of parameters for the interface structure both
These band-structure effects are to some extent related to the        perpendicular to and in the plane of the interfaces.
third contribution to the GMR effect, the spin-dependent
electron scattering. On one hand, the minigaps in the band                                  EXPERIMENTAL
structure caused by the periodicity of the superlattice will be
influenced by the defects which cause the scattering. On the             The superlattices were prepared in a Riber molecular-
other hand, the spin dependence of the scattering process is          beam epitaxy deposition system  2 10 11 mbar base pres-
caused by the asymmetry of the band structure, first via the          sure  using electron-beam evaporation hearths, which were
density of states at the Fermi level, and second via the spin-        rate stabilized to within 1% by a homemade feedback control
dependent scattering potential at impurities or interfaces. The       system32 using Balzers quadrupole mass spectrometers
first contribution makes any scattering event spin dependent,          QMS . Additionally, integration of the QMS signal was
even scattering at phonons.15 Experimentally, the contribu-           used for automatic control of the shutters of the individual
tions of the electronic structure and the spin-dependent scat-        evaporation sources. In this way, a reproducible bilayer
tering cannot yet be separated since scattering is dominant in        thickness throughout the whole superlattice was ensured, as
all reported samples so far. It is this spin-dependent scatter-       well as a constant Cr thickness over all superlattices. The Fe
ing that generally receives the most attention in the literature,     and Cr layers  starting material of 99.996% purity  were
experimentally and theoretically.                                     electron-beam evaporated in a pressure of 4 10 10 mbar at
      Here two contributions have to be considered separately,        a rate of 1 Å/s on polycrystalline yttrium stabilized zirco-
the spin-dependent scattering at impurities inside the mag-           nium oxide  YSZ  substrates  typically 5 5 mm2 . In order
netic layers  referred to as bulk scattering  and the scattering      to minimize thickness inhomogeneities, the substrate was ro-
at the interfaces. Both can  in principle  cause a GMR                tated at 60 rpm during the whole growth process. The surface
effect.16­22 In combination they can even cancel each other           roughness of the YSZ substrates was evaluated ex situ by
provided that their spin asymmetry is opposite. The ideally           atomic-force microscopy  AFM . Typical rms values of the
pure cases, samples with only bulk or only interface scatter-         YSZ surface roughness were 5 Å on a 1  m2 area. After
ing, are difficult to achieve experimentally. This would re-          rinsing in isopropyl alcohol and drying in a dry N2 flow, the
quire the growth of samples with either ideally flat interfaces       substrate was annealed for 15 min at 600 °C in UHV.
or defect-free layers. However, recent experiments on Co/Cu              The superlattices consisted of ten bilayers with 22 Å of Fe
superlattices indicate that spin-dependent interface scattering       and 13 Å of Cr starting the growth with a Fe layer. The
dominates the GMR effect.23                                           interface roughness was varied by growing the samples ei-
      We therefore have a strong motivation to investigate            ther directly onto the YSZ substrates  sample numbers 5,7,9 
quantitatively the effects of interface structure  e.g., rough-       or onto a 20 Å thick Cr buffer  sample numbers
ness  on GMR. A detailed comparison with theory requires a            6,8,10,12,14,16  using substrate temperatures  TG  increas-
comprehensive and quantitative analysis of the interface              ing from 0 to 400 °C in steps of 50 °C  increasing sample
structure. The most powerful technique for this purpose is            numbers . In this way, we obtained a series of 18 Fe/Cr
x-ray diffraction24  XRD  because, first, it is a nondestructive      superlattices of which nine have been selected for this analy-
technique applied after the completion of the growth of the           sis because of their magnetic properties  see below .
sample, second it probes the whole superlattice structure as it          The structural characterization of the superlattices was
is seen by the electrons in the transport measurements and,           obtained through small angle  SA  XRD measurements using
third, it uses waves with a wavelength similar to the one of          either a synchrotron source with wavelength of 2.0753 Å  15
the electrons at the Fermi level of usual metals. Unfortu-            eV below the Cr absorption edge  or a Rigaku rotating anode
nately, ordinary XRD provides only low contrast for Fe/Cr             diffractometer at 4 kW power and with an x-ray wavelength
superlattices, because of the comparable electron densities of        of 1.542 Å (Cu K ). The following experimental XRD set-
Fe and Cr. This effect has impeded until now the quantitative         ups were used:  i  specular reflectivity measurements  or
evaluation of the XRD spectra. However, synchrotron radia-            symmetrical  -2  scans  at SA were used to determine the
tion allows the use of anomalous diffraction by choosing the          interface roughness   in perpendicular direction;  ii  rocking
wavelength close to the absorption edge of one of the atomic          curve or  -scan measurements at SA providing information
species, resulting in an enhanced contrast. Additionally, re-         about the lateral correlation length  x of the interface rough-
cent developments of theoretical models describing specular           ness and the Hurst parameter h. The lateral correlation
and diffuse x-ray scattering from superlattices25­31 allow            length is a measure for the spatial decay of the height-height
simulations of XRD spectra which are characterized by a               correlation function whereas h describes the fractality of the
high degree of agreement with the measured spectra and ac-            interface structure;  iii  offset (   ) 2  scans to study the
cordingly deliver very reliable values for the interface struc-       correlation of the interface roughness in perpendicular direc-
ture of the superlattices.                                            tion expressed by the correlation length  z . The measured
      In this paper we present the interpretation of the transport    spectra were simulated applying recently developed theories
properties of polycrystalline Fe/Cr superlattices based on a          describing specular as well as diffuse x-ray scattering at su-
quantitative analysis of their XRD data. The transport prop-          perlattices. In this model the scattered intensity is calculated
erties are characterized by high values of the GMR effect  up         by dynamical scattering in the distorted-wave Born approxi-
to 80% for samples with 10 bilayers  indicating the domi-             mation as a function of the vertical and lateral scattering
nance of spin-dependent scattering processes above spin-              vectors  qz and qx . Further details can be found in the origi-



13 694                                                   R. SCHAD et al.                                                            57

nal literature.25­31 Large-angle XRD which usually can be
employed for quantitative analysis of the interface structure
of superlattices24 cannot be used in this case because the
samples are polycrystalline with only poor texture.20 But also
in the case of high-quality epitaxial Fe/Cr superlattices33,34
the similar lattice constants of Fe and Cr are responsible for
the much less pronounced large-angle spectra compared to
the SA XRD scans. Therefore, the analysis of the SA data
generally delivers more robust values of  .
   The electrical measurements were performed in an Ox-
ford cryostat  1.5 up to 300 K  equipped with a 15 T magnet.
Resistivities were determined using a standard four-probe
Van der Pauw method. The magnetoresistance is defined as
  / s ( 0  s)/ s , where  0 is the resistivity in zero field
and  s the saturation resistivity in a magnetic field Hs paral-
lel to the interfaces. All quoted resistivity values were mea-
sured at 4.2 K.
   The magnetization measurements were performed in an
alternating gradient magnetometer. The antiferromagnetic
fraction of the samples is defined as AFF 1 (Mr /Ms)
with Mr and Ms being, respectively, the remnant and the
saturation magnetization. This AFF was used to correct the            FIG. 1. Specular SA XRD spectra of one sample measured with
magnetoresistance for small variations in the magnetic order       x-ray wavelengths of, respectively, 1.542 Å  Cu K  laboratory
of the samples by dividing    by AFF.35 This way the mag-          source  and 2.0753 Å  synchrotron source . Shown are the mea-
netoresistance data become independent of this contribution.       sured data  points  and the simulations  lines . The two simulations
                                                                   are obtained using identical input parameters except for the differ-
                                                                   ent optical constants which were taken from literature. All curves
                   RESULTS AND DISCUSSION                          are vertically offset for clarity.

   As a function of the sample growth temperature TG we            oxide. Then the simulation was optimized by adjusting the
found the best layering quality and a maximum of the GMR           vertical interface roughness    Fig. 1 lower curve . The cri-
amplitude around TG 250 °C.20 However, the reduction of            teria for assessing the quality of a simulation was the match-
the GMR towards higher TG is only caused by a decrease of          ing of the superlattice Bragg peak intensities and shapes. The
the AFF  Ref. 35  and is thus a magnetic contribution.             uncertainty of the obtained roughness value depends on the
Therefore our analysis is restricted to nine samples grown at      distinctness of the superlattice structure in the spectrum. This
lower TG where the changes of the GMR are of spin-                 varies with the roughness itself and the x-ray wavelength
dependent origin.                                                  used. Careful estimates of these uncertainties were obtained
   First, we will discuss the structural properties of the su-     by studies of the influence of   on the quality of the simu-
perlattices measured with XRD. In Ref. 20 we assessed the          lations and are used as error bars in Fig. 4.
interface quality by the peak to background intensity ratio of        Simulations taking into account possible variations of the
SA XRD rocking curves. This was, however, revealing no             interface roughness throughout the stacking of the superlat-
information over the lateral roughness length scale and addi-      tice  cumulative roughness24 or inequality of Fe/Cr and
tionally, the intuitive interpretation of the diffuse intensity    Cr/Fe interfaces  were not successful so that we can con-
can be misleading.30,31 Here we are able to present a quanti-      clude that this effect must be small or absent. In order to
tative simulation of the specular and diffuse XRD spectra          keep the number of simulation parameters limited we used
giving a comprehensive overview over the relevant interface        identical roughness for all superlattice interfaces. Addition-
structure parameters. Since not all samples could be mea-          ally, it should be noted that the obtained values of   were not
sured at the synchrotron source we first will compare simu-        influenced by a later fine adjustment of the substrate rough-
lations of the specular SA XRD data obtained using, respec-        ness  s which only determines the inter-Bragg peak intensity
tively, the synchrotron source and the laboratory source. This     and the damping of the finite-size peak oscillations. We find
is demonstrated for the sample with the most pronounced            values of  s  about 3 Å  being slightly smaller than
superlattice structure since here any deviations between           the ones measured by AFM  about 5 Å . This small differ-
simulation and measurement will become most obvious, but           ence might be caused by the different lateral length scale
of course, similar agreement is found also for the other           over which the two methods are sensitive38 and by the fact
samples  Fig. 1 . The specular data show a rich structure          that the AFM data were taken in air.
being the pronounced superlattice Bragg peaks and the                 As next step, all parameters of this simulation had served
higher frequent finite-size peaks. We produced the best simu-      as input parameters for the simulation of the spectrum mea-
lation for the spectrum measured at the synchrotron using as       sured with the Cu K  wavelength. Only the optical con-
input parameters the thicknesses of all layers  Fe, Cr, and a      stants had, of course, to be changed according to the different
top oxide layer ,36 the number of bilayers, the optical param-     wavelength used.37 Although the two measured spectra look
eters of Fe, Cr, YSZ, top oxide,37 and the roughnesses of,         very different because of the enhanced material contrast in
respectively, the substrate  determined by AFM  and the top        the case of the synchrotron data for which the wavelength



57                                      QUANTITATIVE STUDY OF THE INTERDEPENDENCE . . .                                           13 695



















                                                                         FIG. 3. SA XRD rocking curves of samples 6 and 16 with qz at
                                                                      the position of the second-order superlattice Bragg peak. Shown are
                                                                      the measured data  crosses  and the simulations  lines . All curves
                                                                      are vertically offset for clarity.

                                                                      substrate. The   values we obtain are in qualitative agree-
                                                                      ment with the peak-to-background intensity ratios derived in
      FIG. 2. Specular SA XRD spectra of all samples measured ei-     Ref. 20. However, the quantitative structure analysis by
ther with a wavelength of 1.542 Å  Cu K  laboratory source;           simulation provides values of well-defined structure param-
samples 7, 8, 10, 12, 14  or 2.0753 Å  synchrotron source; samples    eters and additionally, allows us to estimate the lateral
5, 6, 9, 16 . Shown are the measured data  crosses  and the simu-     roughness components.
lations  lines . All curves are vertically offset for clarity.           The lateral correlation of the interface roughness was
                                                                      measured by   scans at qz , the vertical wave vector, set to
was chosen close to the absorption edge of Cr, both simula-           the position of the second superlattice Bragg peak. These
tions are in excellent agreement with the data  Fig. 1 . This         measurements were done at the synchrotron, so data are
degree of agreement proves that spectra from superlattices            available for samples 5, 6, 9, and 16 which are samples
with such low material contrast as Fe/Cr can be successfully          grown at, respectively, low and high TG and with or without
simulated and quantitative roughness data can be obtained.            a Cr buffer. Two examples of measured data together with
Furthermore, it gives confidence in the structure analysis ob-        their respective simulation are shown in Fig. 3. The relevant
tained through simulations of spectra measured with the               parameters of the simulation are the lateral correlation length
laboratory source.                                                     x and the Hurst parameter h which describe the decay of the
      The specular data with their respective simulations of all      height-height correlation function. In simple terms h can be
samples are shown in Fig. 2. Deviations between simulation            taken as a measure for the jaggedness of the interfaces.28
and measurement can be observed at very small angles for all          Both parameters will be relevant for the transport properties
samples measured with the laboratory source  sample num-              since, for a given value of the roughness amplitude, they will
bers 7,8,10,12,14 . This is caused by the nonlinearity of the         determine the density of steps at the interfaces which form
x-ray detector at high intensities. The other samples had been        finally the deviations from a perfect interface, i.e., the scat-
measured at the synchrotron. Deviations in intensity between          tering centers.39 The interdependence of  x and h results in
measurement and simulation at wave vectors in-between su-             some uncertainty of their estimated values with h 0.5
perlattice Bragg peaks  most pronounced in the spectrum of             0.2. In order to limit the number of free simulation param-
sample 6  are likely caused by surface contamination. In              eters we kept h fixed at h 0.5. The lateral correlation length
principle, these long-wavelength deviations can be repro-             is about 90 Å for all samples. The roughness correlation in
duced in the simulation by introducing an extra surface layer         the z direction, expressed by  z , is likely of less direct im-
of several nm thickness and adjusting its optical parameters.         portance for the electron scattering, but it might have an
However, this does not influence the intensities of the super-        influence on the interlayer exchange coupling via thickness
lattice Bragg peaks and hence the values obtained for the             variations of the Cr layer. Variations of the exchange cou-
relevant interface roughness parameter  . Furthermore, this           pling are taken into account by the antiferromagnetic frac-
contamination layer is also unlikely to influence the electri-        tion. The samples discussed here have a constant value of
cal transport data. Therefore, we decided to keep the simu-            z 130 Å, obtained from simulations of the asymmetric
lations as simple as possible, only including the relevant lay-       (   ) 2  scans.
ers. In general, the films grown on the Cr buffer are smoother           In summary, the SA XRD analysis reveals that the struc-
than without buffer. Obviously, this Cr seed layer provides a         tural parameters of the samples discussed here vary mostly in
better template for the superlattice growth than the bare YSZ         the amplitude of the interface roughness (2.2 Å   5 Å),



13 696                                                       R. SCHAD et al.                                                        57

                                                                        taxial samples . The scattering at such big steps could be less
                                                                        spin selective than at monoatomic ones.
                                                                            ii  Since these polycrystalline samples have a high de-
                                                                        gree of bulk defects it is doubtful whether their contribution
                                                                        can be neglected. Including in the discussion bulk scattering
                                                                        which might have a spin asymmetry in the electron scattering
                                                                        there would exist a GMR effect already without any interface
                                                                        contribution. In order to explain now the observed roughness
                                                                        dependence of the GMR amplitude, the spin asymmetry of
                                                                        the electron scattering at the interfaces would have to be
                                                                        opposite to the bulk contribution. Then increasing scattering
                                                                        at the interfaces increasingly compensates the GMR effect
                                                                        stemming from the bulk scattering. The electrons which are
                                                                        less scattered at the bulk impurities would be scattered at the
                                                                        interfaces and vice versa.4
                                                                           For both scenarios increasing values of   would increase
                                                                         s and, at the same time, decrease   . However, the inter-
                                                                        face roughness dependence of the spin asymmetry of the
                                                                        interface scattering would be exactly opposite. This dilemma
                                                                        in the interpretation of the experimental data is an inherent
                                                                        problem for all samples with a non-negligible amount of
                                                                        bulk defects, in particular polycrystalline samples. A clear
                                                                        interpretation is impeded not only by the presence of such
                                                                        bulk defects but also their undefined contribution to the spin
   FIG. 4. Saturation resistivity  S  a  and magnetoresistance, cor-    asymmetry of the electron scattering and their unknown
rected for variations of the antiferromagnetic fraction,                changes in concentration and influence when varying the in-
  /AFFT  b  as a function of the interface roughness amplitude  .       terface quality. These undefined and variable bulk contribu-
Shown are the measured data  crosses  and linear best fits  lines .     tions might also account for the scatter of the transport data
 S increases with increasing   whereas    decreases.                    in Fig. 4.
with little variation in the lateral roughness parameters. This
structural information is now used to understand the trans-                                      SUMMARY
port properties. Since the interface roughness amplitude   is              We presented the interpretation of the transport properties
the structure parameter varying mostly we focus on the dis-             of Fe/Cr superlattices based on their structural properties.
cussion of  s and    as a function of    Fig. 4 . First, it has         The GMR effect reaches very high values compared to other
to be noted that the GMR   / s is rather high  up to 80%                polycrystalline samples of up to 80% for ten bilayers indi-
compared to values usually reported for nonepitaxial                    cating the importance of spin-dependent scattering processes.
samples.16­19 This indicates, in our case, that the spin-               We analyzed the structure of the high-quality Fe/Cr superlat-
dependent electron scattering      dominates the spin-                  tices by quantitative simulation of the XRD spectra revealing
independent ( s) events. Therefore our analysis is rather in-           the relevant structural interface parameters perpendicular to
dependent of uncharacterized changes in structural defects              and in the plane of the interfaces. We found a decrease of the
affecting the spin-independent background resistivity. In ad-           magnetoresistance    and   / 
dition, the transport properties show a strong variation with                                             s with increasing roughness
                                                                        amplitude. The theoretical understanding is not clear. The
  indicating a strong link between interface roughness and              decrease of the magnetoresistance could be either caused by
magnetoresistance. We observe an increase of  s and a de-               enhanced roughness increasingly scattering electrons of both
crease of    or   / s with increasing    Fig. 4 . The expla-            spin orientations with similar strength or by a compensation
nation of this might be one of the following scenarios:                 of a bulk contribution by the interface scattering having op-
    i  Neglecting any spin-dependent bulk scattering, the ob-           posite spin asymmetry to the electron scattering. Therefore, a
served roughness dependence of the magnetoresistance has                clear experimental result about the influence of the interface
to be ascribed to the changes in the interface properties in the        structure on the GMR amplitude will have to be based on
following way. The increasing interface roughness amplitude             samples with negligible bulk scattering.
reduces the spin asymmetry of the interface scattering. This
is expected for higher values of   when the minority elec-
trons are also increasingly scattered, thus reducing the spin                             ACKNOWLEDGMENTS
asymmetry of the interface scattering.4 On the other hand,                 This work was financially supported by the Belgian Con-
the pronounced superlattice Bragg peaks in the SA XRD                   certed Action  GOA  and Interuniversity Attraction Poles
spectra indicate rather smooth interfaces, certainly for the             IUAP  programs. R.S., C.D.P., and G.V. acknowledge sup-
best samples. However, the exact value of the roughness am-             port by, respectively, the European Community  Marie Cu-
plitude above which the GMR amplitude should decrease                   rie , the Research Council of the Katholieke Universiteit
with increasing   is not known. An alternative explanation              Leuven, and the Belgium Iteruniversity Institute for Nuclear
could be the possible occurrence of bigger steps at the inter-          Sciences. We are indebted to J. Barnas for helpful discussion
faces of these polycrystalline superlattices  contrary to epi-          and carefully reading the manuscript.



57                                      QUANTITATIVE STUDY OF THE INTERDEPENDENCE . . .                                                      13 697

*Author to whom correspondence should be addressed: Research               20 P. Belie¨n, R. Schad, C. D. Potter, G. Verbanck, V. V. Mosh-
      Institute for Materials, Katholieke Universiteit Nijmegen, Toer-            chalkov, and Y. Bruynseraede, Phys. Rev. B 50, 9957  1994 .
      nooiveld 1, NL 6525 ED Nijmegen, The Netherlands, Fax  31            21 J. M. Colino, I. K. Schuller, R. Schad, C. D. Potter, P. Belie¨n, G.
       0 24 3652190. Electronic mail: schad@sci.kun.nl                            Verbanck, V. V. Moshchalkov, and Y. Bruynseraede, Phys. Rev.
 Present address: Philips Optical Storage, Kempische Steenweg                     B 53, 766  1996 .
      293, 3500 Hasselt, Belgium.                                          22 J. M. Colino, I. K. Schuller, V. Korenivski, and K. V. Rao, Phys.
1 M. N. Baibich, J. M. Broto, A. Fert, F. Nguyen Van Dau, F.                      Rev. B 54, 13 030  1996 .
      Petroff, P. Etienne, G. Creuzet, A. Friederich, and J. Chazelas,     23 H. J. M. Swagten, G. J. Strijkers, G. L. J. Verschueren, M. M. H.
      Phys. Rev. Lett. 61, 2472  1988 .                                           Willekens, and W. J. M. de Jonge, J. Magn. Magn. Mater. 176,
2 R. Q. Hood, L. M. Falicov, and D. R. Penn, Phys. Rev. B 49, 368                 169  1997 .
       1994 .                                                              24
3                                                                                E. E. Fullerton, I. K. Schuller, H. Vanderstraeten, and Y.
     Y. Asano, A. Oguria, and S. Maekawa, Phys. Rev. B 48, 6192                   Bruynseraede, Phys. Rev. B 45, 9292  1992 .
       1993 .                                                              25
4 J. Barnas and Y. Bruynseraede, Phys. Rev. B 53, 5449  1996 .                   S. K. Sinha, E. B. Sirota, S. Garoff, and H. B. Stanley, Phys. Rev.
5 I. A. Campbell and A. Fert, in Ferromagnetic Materials, edited                  B 38, 2297  1988 .
                                                                           26
      by E. P. Wohlfarth  North-Holland, Amsterdam, 1982 .                       V. Holy´ J. Kuben a, I. Ohi´dal, K. Lischka, and W. Poltz, Phys.
6 J. Barnas, A. Fuss, R. E. Camley, P. Gru¨nberg, and W. Zinn,                    Rev. B 47, 15 896  1993 .
                                                                           27
      Phys. Rev. B 42, 8110  1990 .                                              V. Holy´ and T. Baumbach, Phys. Rev. B 49, 10 668  1994 .
7                                                                          28
     J. Barnas, A. Fuss, R. E. Camley, U. Walz, P. Gru¨nberg, and W.             J.-P. Schlomka, M. Tolan, L. Schwalowsky, O. H. Seeck, J. Stett-
      Zinn, Vacuum 41, 1241  1990 .                                               ner, and W. Press, Phys. Rev. B 51, 2311  1995 .
8 M. Ru¨hrig, R. Scha¨fer, A. Huber, R. Mosler, J. A. Wolf, S.             29 H. E. Fischer, H. Fischer, O. Durand, O. Pellegrino, S. Andrieu,
      Demokritov, and P. Gru¨nberg, Phys. Status Solidi A 125, 635                M. Picuch, S. Lefebvre, and M. Bessiere, Nucl. Instrum. Meth-
       1991 .                                                                     ods Phys. Res. B 97, 402  1995 .
9 C. D. Potter, R. Schad, P. Belie¨n, G. Verbanck, V. V. Mosh-             30 H. E. Fischer, H. M. Fischer, and M. Picuch  unpublished .
      chalkov, Y. Bruynseraede, M. Scha¨fer, R. Scha¨fer, and P. Gru¨n-    31 Henry E. Fischer, memoire DHDR  Diplome d'Habilitation a Di-
      berg, Phys. Rev. B 49, 16 055  1994 ; R. Schad, C. D. Potter, P.            riger les Recherches , Universite Joseph Fourier, Grenoble,
      Belie¨n, G. Verbanck, V. V. Moshchalkov, Y. Bruynseraede, M.                France.
      Scha¨fer, R. Scha¨fer, and P. Gru¨nberg, J. Appl. Phys. 76, 6604     32 W. Sevenhans, J.-P. Locquet, and Y. Bruynseraede, Rev. Sci.
       1994 .                                                                     Instrum. 57, 937  1986 .
10 J. C. Slonszewski, Phys. Rev. Lett. 67, 3172  1991 .                    33 E. E. Fullerton, M. J. Conover, J. E. Mattson, C. H. Sowers, and
11 J. C. Slonszewski, J. Appl. Phys. 73, 5975  1993 .                             S. D. Bader, Appl. Phys. Lett. 63, 1699  1993 .
12 Kees M. Schep, Paul J. Kelly, and Gerrit E. W. Bauer, Phys. Rev.        34 R. Schad, C. D. Potter, P. Belie¨n, G. Verbanck, V. V. Mosh-
      Lett. 74, 586  1995 .                                                       chalkov, and Y. Bruynseraede, Appl. Phys. Lett. 64, 3500
13 P. Zahn, I. Mertig, M. Richter, and H. Eschrig, Phys. Rev. Lett.                1994 .
      75, 2996  1995 .                                                     35 R. Schad, J. Barnas, P. Belie¨n, G. Verbanck, C. D. Potter, H.
14 I. Mertig, P. Zahn, M. Richter, H. Eschrig, R. Zeller, and P. H.               Fischer, S. Lefebvre, M. Bessiere, V. V. Moshchalkov, and Y.
      Dederichs, J. Magn. Magn. Mater. 151, 363  1996 .                           Bruynseraede, J. Magn. Magn. Mater. 156, 339  1996 .
15 C. T. Yu, K. Westerholt, K. Theis-Bro¨hl, and H. Zabel J. Appl.         36 The properties of the top oxide layer  typical thickness, rough-
      Phys. 82, 5560  1997 .                                                      ness, composition, and optical parameters  were characterized in
16 E. E. Fullerton, D. M. Kelly, J. Guimpel, I. K. Schuller, and Y.               independent experiments  R. Schad, D. Bahr, J. Falta, G. Mater-
      Bruynseraede, Phys. Rev. Lett. 68, 859  1992 .                              lik, P. Belie¨n, G. Verbanck, K. Temst, and Y. Bruynseraede, J.
17 N. M. Rensing, A. P. Payne, and B. M. Clemens, J. Magn. Magn.                  Phys. Condens. Matter 10, 61  1998 .
      Mater. 121, 436  1993 .                                              37 S. Brennan and P. L. Cowan, Rev. Sci. Instrum. 63, 850  1992 .
18 N. M. Rensing, B. M. Clemens, and D. L. Williamson, J. Appl.            38 K. Temst, M. J. Van Bael, B. Wuyts, C. van Haesendonck, Y.
      Phys. 79, 7757  1996 .                                                      Bruynseraede, D. G. de Groot, N. Koeman, and R. Griessen,
19 S. Joo, Y. Obi, K. Takanashi, and H. Fujimori, J. Magn. Magn.                  Appl. Phys. Lett. 67, 3429  1995 .
      Mater. 104, 1753  1992 .                                             39 A. Kaserer and E. Gerlach, Z. Phys. B 97, 139  1995 .