INSTITUTE OF PHYSICS PUBLISHING JOURNAL OF PHYSICS: CONDENSED MATTER

J. Phys.: Condens. Matter 14 (2002) 10063–10074 PII: S0953-8984(02)39059-3

Reversal of exchange bias in nanocrystalline

antiferromagnetic–ferromagnetic bilayers
C Prados1 , E Pina1,3 , A Hernando1 and A Montone2
1 Instituto de Magnetismo Aplicado, RENFE-UCM-CSIC, PO Box 155, Las Rozas 28230

Madrid, Spain
2 ENEA, Materials and Technology Unit, CR Casaccia, CP 2400, 00100 Rome, Italy

E-mail: epina@renfe.es

Received 1 July 2002, in final form 19 September 2002

Published 18 October 2002
Online at stacks.iop.org/JPhysCM/14/10063

Abstract
The sign of the exchange bias in field cooled nanocrystalline antiferromagnetic–
ferromagnetic bilayers (Co–O and Ni–O/permalloy) is reversed at temperatures
approaching the antiferromagnetic (AFM) blocking temperature. A similar
phenomenon is observed after magnetic training processes at similar
temperatures. These effects can be explained assuming that the boundaries
of nanocrystalline grains in AFM layers exhibit lower transition temperatures
than grain cores.

1. Introduction

Exchange anisotropy refers to the interface interaction between ferromagnetic (FM) and
antiferromagnetic (AFM) materials. When a AFM/FM thin film bilayer is cooled through the
Néel (TN ) temperature of the AFM (with the Curie temperature TC of the FM larger than TN ) in
a saturating magnetic field, a unidirectional anisotropy is induced in the FM. This anisotropy is
commonly manifested as, among other things, a hysteresis loop shifted towards fields opposite
to that applied during the cooling process, an enhancement of coercivity and a sin θ component
in the torque curve. The strength of the exchange anisotropy is usually quantified by the shift
of the hysteresis loop, which is called the exchange field, H E . Although this phenomenon was
first reported a long time ago in passivated Co particles [1], a great number of experimental and
theoretical works were afterwards developed, many of them reported in reviews [2] and [3],
due to the intriguing physics underlying this interface exchange coupling and its practical
applications. In fact, exchange anisotropy is systematically used for controlling the direction
of magnetization in magneto-electronic devices, such as spin valves in magnetic storage read
heads [4]. In addition to the above-described features, exchange-biased bilayers also can
exhibit several less usual properties, such as positive exchange bias [5, 6], magnetic training
effects [7], asymmetric magnetization reversal processes [8] and memory effects [9].
3 Author to whom any correspondence should be addressed.

0953-8984/02/4310063+12$30.00 © 2002 IOP Publishing Ltd Printed in the UK 10063

10064 C Prados et al

An intuitive picture of the exchange anisotropy mechanism has already been proposed
in the early works of Meiklejohn and Bean [1], based on an uncompensated AFM interface
and the interfacial exchange interaction between the AFM–FM spins. However, there is little
quantitative understanding of the phenomenon and the development of an ultimate theoretical
framework accounting for the different aspects of the unidirectional anisotropy remains elusive.
Different theories have been proposed since the earlier works, in which collinear magnetic
structures on both FM and AFM sides of the interface were assumed. Most of these theories,
reported in reviews [10, 11], propose more complex interfaces that better approach the actual
ones. We will focus on theoretical models adequate for polycrystalline thin film systems,
similar to those studied here. Two main trends can be distinguished in such theoretical
models explaining the exchange anisotropy, mainly concerning the magnitude of H E . Firstly,
some works propose the existence of a domain structure parallel to the interface, either in the
AFM or the FM material [12]. This approach has been used to explain different properties
related to the exchange anisotropy, such as irreversible [13] and memory [9] effects. Another
general approach assumes the formation of AFM domains perpendicular to the interface plane.
Malozemoff [14] considers a random exchange interaction at the interface due to the surface
roughness or chemical inhomogeneity at atomic scale. The interfacial random-field energy is
minimized when the AFM breaks up into a perpendicular domain structure. Within this trend
we can also include works related to exchange anisotropy in polycrystalline systems. Takano
et al [15] suggest that uncompensated spins on the surface of AFM films are responsible for
unidirectional anisotropy. The origin of this uncompensated moment lies in the finite grain
size of the AFM film. In this way, the AFM grains play a similar role to the perpendicular
AFM domains in Malozemoff’s approach.
In this paper we report measurements on thermal evolution of coercivity, Hc , and the
exchange field, H E , in nanocrystalline Co–O/permalloy (Py) and Ni–O/Py bilayers after
different cooling and training processes. Since the uncompensated moment related to the
finite size of the AFM grains seems to be the source of the exchange anisotropy in this kind
of structure, enhanced interfacial effects are expected when grain size is decreased [16]. In
particular, we report new and striking phenomena consisting in a reversal of sign in exchange
anisotropy when temperature is increased in the region just below the AFM transition. This
behaviour is related to the existence of two different magnetic transition temperatures in the
AFM layers, one of them corresponding to grain boundaries and the other to grain cores.
Finally, a simple phenomenological model is developed to account for the relation between the
described experimental results concerning the reversal of exchange bias and the nanocrystalline
nature of the samples.

2. Sample preparation and structural characterization

Co–O (20 nm)/Py (20 nm) and Ni–O (40 nm)/Py (20 nm) bilayer samples were grown by rf
magnetron sputtering onto 1 mm thick glass substrates at nominal room temperature. The
background pressure was lower than 5 × 10−7 mbar. Oxide layers were reactively sputtered
from pure Co and Ni targets. The sputtering gas during reactive deposition of layers was
oxygen up to 2 × 10−3 mbar and Ar up to 5 × 10−3 mbar. Py layers were deposited from
composite targets using Ar at 5 × 10−3 mbar. A final Cu layer was added as an oxidation
barrier. Deposition rate calibration of the individual layers was carried out by low-angle x-ray
diffraction, being 0.1 nm s−1 (Py), 0.03 nm s−1 (Co–O) and 0.07 nm s−1 (Ni–O), respectively.
These conditions were revealed as the most appropriate for growing thin films in our sample
preparation system with the required nanocrystalline structure.
Reversal of exchange bias in nanocrystalline antiferromagnetic–ferromagnetic bilayers 10065

Figure 1. Planar-view TEM image of a Co–O (20 nm) layer deposited onto 1 mm glass substrate.
The image is slightly out of focus to enhance phase contrast at the grain boundaries.

Separate control samples of both Ni and Co oxides were also prepared under identical
oxygen reactive sputtering conditions in order to carry out a deeper analysis of the crystalline
structure of the AFM layers. Hence, a 20 nm thickness Co–O layer and a 40 nm thickness
Ni–O layer were grown onto 1 mm thick glass substrates.
Structural characterization of samples was performed by transmission electron microscopy
(TEM) with a JEOL 4000FX microscope operating at 300 kV equipped with ultra-thin window
x-ray detector and imaging electron spectrometer. Samples, in plan and cross-section geometry,
were prepared with the usual method of grinding, dimpling and ion beam milling at glazing
angle. Planar-view images were taken from Co–O and Ni–O single-layer specimens revealing
a quasi-spherical grain shape with a log-normal distribution of grain diameters centred at
3.9 and 5.3 nm, respectively. Figure 1 shows a bright-field planar-view image of the Co–O
sample. Besides the clear nanocrystalline structure, the image, which is slightly out of focus
to enhance the phase contrast, shows a remarkable phase contrast at the grain boundaries. This
fact indicates compositional or structural differences between boundaries and cores which will
be reflected in the magnetic properties. The estimated thickness of boundaries in both cases is
around 0.5 nm. Cross-sectional images of bilayers were taken in order to observe the interaction
interface between AFM and FM layers as well as the lateral dimensions of grains. Figure 2
shows a bright-field cross-sectional image of a Co–O/Py sample from which a statistical value
of 14.2 nm for the grain size of Py grains has been determined. A well defined interface with a
roughness less than 3 nm is also observed. X-ray microanalysis and microdiffraction confirm
the presence of a sharp interface within the relative spatial resolutions.

3. Magnetization measurements

3.1. Thermal evolution of hysteresis

Magnetization measurements were carried out by using a Quantum Design SQUID
magnetometer. Magnetic field was always applied parallel to the film plane and magnetic
moment was measured along the field direction. Co–O/Py bilayers were cooled from 300 to
50 K in the presence of a saturating magnetic field of +1 kOe. Figure 3 shows three consecutive
10066 C Prados et al

Figure 2. Cross-sectional view TEM image of a Co–O (20 nm)/Py (20 nm) bilayer deposited onto
1 mm glass substrate.

Figure 3. Three consecutive hysteresis loops for a Co–O (20 nm)/Py (20 nm) bilayer measured
at 50 K after cooling the sample in a field Hcool = +1000 Oe. A huge training effect of shift and
coercivity is observed after the first cycle.

hysteresis loops which were measured at 50 K immediately after the field cooling process. The
shift of the loops towards negative fields and the enhanced coercivity indicate the presence of
exchange anisotropy induced by the field cooling process. The huge decrease of coercivity and
exchange field after the first consecutive loops indicates the existence of a sizeable number
of biased spins belonging to the AFM layer in metastable spin configurations. After each
magnetization cycle the AFM spins evolve to energetically favourable configurations. In this
way, the training process erases partially the initial spin distribution and, hence, it provokes a
decrease of exchange anisotropy effects. Since nanocrystalline structure implies a large ratio
of spins located at grain boundaries, where their disordered structure brings out fluctuations
in magnetic anisotropy and exchange strength, the training effect is strongly reinforced by the
reduced grain size of the AFM layer.
After the magnetic training at 50 K, hysteresis loops were measured at different
temperatures between 50 and 300 K. Figures 4(a) and (b) show the evolution of Hc and H E with
temperature for the sample Co–O (20 nm)/Py (20 nm). The first result is that the temperature
above which exchange anisotropy effects disappear is around 200 K, well below the TN of
bulk CoO (TN = 293 K). This transition temperature, called the blocking temperature, TB , is
in the lower range of previously reported values for polycrystalline CoO [2] and reflects the
influence of the reduced grain size (TB = 175 K is reported in [17] for a 20 Å CoO/164 Å Co
bilayer). Three regions can be distinguished in the thermal evolution of H E :
(i) at low temperatures the hysteresis loops are shifted towards negative fields and the absolute
value of H E decreases with the increase of temperature (this is the most usual behaviour
for a positive cooling field),
Reversal of exchange bias in nanocrystalline antiferromagnetic–ferromagnetic bilayers 10067

Figure 4. Thermal evolution of coercivity (a) and loop shift (b) for a Co–O (20 nm)/Py (20 nm)
bilayer measured after cooling the sample in a field Hcool = +1000 Oe from 300 to 50 K and three
training cycles at 50 K. Positive exchange bias is observed between 150 and 180 K.

(ii) as the temperature approaches the AFM blocking temperature, a particular effect occurs,
the sign of H E is reversed (above ∼150 K), i.e., the loops are shifted towards positive
fields, and
(iii) after reaching a maximum positive value around 170 K, the unidirectional anisotropy
disappears above 200 K.

Figure 4(a) shows the evolution of coercivity along different measurement temperatures. It
decreases monotonically with the increasing temperature and exhibits an enhanced value below
the blocking temperature of the AFM layer. A characteristic feature is observed (which is
discussed below) at temperatures where reversal of exchange field appears. Similar behaviour
is observed in nanocrystalline Ni–O/Py bilayers. In that case the samples have been field
cooled from 400 to 50 K under +1 kOe and data have been measured after training the sample
at 50 K. The measured Ni–O blocking temperature is around 375 K and positive shifted loops
are observed above 300 K (figure 5).
The experimental validity of the results presented in this work has been tested in two
ways. Firstly, the above-described hysteretic measurements have been repeated up to four
times in different runs. In this way, each experimental point in figures 4 and 5 is the result
of a statistical average over a number of different measurements. Error bars correspond to
the standard deviation and always are much smaller than the magnetic features related to the
reversal effect. In some cases, error bars are smaller than the symbols. On the other hand,
the experiments in both kinds of bilayer have been repeated cooling the samples under the
10068 C Prados et al

Figure 5. Thermal evolution of coercivity (a) and loop shift (b) for a Ni–O (40 nm)/Py (20 nm)
bilayer measured after cooling the sample in a field Hcool = +1000 Oe from 400 to 50 K and three
training cycles at 50 K. Positive exchange bias is observed between 300 and 400 K.

highest available field (+55 kOe). Results are identical, but affected by larger experimental
error due to remanent currents in the superconducting coil. This new set of measurements was
performed in order to discard minor loop effects at low temperatures.

3.2. Magnetic training at fixed temperatures

Reversal of exchange anisotropy sign has been also observed after training the samples at a
fixed temperature close to the AFM transition. The measurement procedure was as follows:
the bilayer was cooled under an applied field of +1 kOe from 300 K (400 K in the case of
bilayers with Ni–O as AFM layer) to the measurement temperature, Tm (special precautions
were taken in order to avoid undercooling), and then consecutive loops were measured with
applied fields in the +1 to −1 kOe range. The same cooling procedure was performed for each
measurement temperature. Figure 6(b) shows the evolution of H E for three consecutive loops
after cooling the sample according to the above-described procedure to temperatures between
120 and 200 K. In the low-temperature range, H E is always negative and its absolute value
decreases with consecutive cycles. However, at temperatures between 150 and 160 K, whereas
the first loop shows negative exchange anisotropy,successive cycles are shifted towards positive
fields. Immediately before the AFM transition, even the first loop exhibits a positive shift, of
which the absolute value decreases after the second one. Figure 6(a) shows the evolution of
coercivity for different measurement temperatures.
Reversal of exchange bias in nanocrystalline antiferromagnetic–ferromagnetic bilayers 10069

Figure 6. Evolution of coercivity (a) and loop shift (b) of three consecutive loops measured after
cooling the sample in a field Hcool = +1000 Oe from 300 K with measurement temperature Tm .
The sample is a Co–O (20 nm)/Py (20 nm) bilayer. Training induced reversal of the exchange
anisotropy is observed between 155 and 170 K.

4. Analysis and discussion

Positive unidirectional exchange anisotropy has been reported in thin film bilayers cooled
under large magnetic fields [18]. Those results were attributed to the combination of an AFM
exchange through the AFM–FM interface and an FM alignment of the AFM surface spins
along the large cooling field. In the present work, we have observed a change of sign of
exchange bias in bilayers cooled under moderate (but saturating) magnetic fields. Following
a similar approach to that developed in [15], we consider the antiferromagnet as a set of
crystallites, which exhibit individually a rather compensated spin structure at the interface
with the ferromagnet. The finite size of the grains implies uncompensation of magnetic
moment at the AFM/FM interface. This uncompensated moment is mainly located at the grain
boundaries of the AFM layer. Due to the structural disorder associated with the grain boundary,
anisotropy, exchange coupling and magnetic transition temperatures will be expected to be
different to those within grain cores. These particularities become sizeable in nanostructured
materials, with a large ratio of spins located at grain boundaries [19, 20]. In our case, the
structural characterization of oxide layers showed the existence of a grain boundary region
clearly differentiated from grain cores. A reduced transition temperature is expected in the
AFM grain boundaries as a consequence of the finite-size scaling effect [21]. For the sake of
simplicity, we are going to consider the interfacial AFM–FM interaction as FM direct exchange
coupling and therefore, during the cooling process, the uncompensated moment in the AFM
10070 C Prados et al

layer would align parallel to the magnetization in the FM (any other mechanism of coupling
would not modify the explanation developed below). The rest of the AFM spins will arrange
accordingly, and this particular spin configuration will remain frozen giving rise to negative
shifted hysteresis loops. The mechanism that we propose to account for the observed reversal of
exchange anisotropy assumes different blocking temperatures for boundary and core regions:
we define TB,c as the blocking temperature of the core regions and we can consider a gradient of
blocking temperatures at the boundary region with TB,b as the lower one (TB,b < TB,c ). Below
TB,b the uncompensated moment is principally located at grain boundaries and it lies within
the direction of the cooling field. When TB,b < T < TB,c the grain boundary region becomes
paramagnetic as the temperature increases. The uncompensated moment will decrease in
absolute value and eventually will reverse its sign due to the AFM nature of the magnetic order
of the oxide layer. In this framework, the thermal evolution of H E can be explained as follows.
(i) After cooling the system in a positive field,the hysteresis loops are shifted towards negative
fields at low temperature due to the positive AFM uncompensated moment. Strong training
effects are observed as shown in figure 3 due to the disordered nature of the grain boundary
region.
(ii) As temperature increases and approaches the overall magnetic transition of the AFM layer,
grain boundaries of the oxide (where the uncompensated magnetic moment is located)
becomes paramagnetic. The AFM nature of the oxide layer makes, as the grain boundary
region loses its magnetic order, the grain cores exhibit negative uncompensated magnetic
moment. Therefore, positive shift is induced in the hysteresis loop of the FM layer.
(iii) Finally, above the AFM transition temperature, exchange anisotropy disappears.
In a similar way it is possible to account for results such as those shown in figure 6. After
cooling the sample under a positive field to a given temperature close to the transition, the
first hysteresis loop will exhibit negative shift. Subsequent training of the sample will produce
strong magnetic relaxation of grain boundaries and the resulting uncompensated moment will
give positive shift of the FM loops.
We have developed a simple qualitative model in order to show how, as the boundaries
of a set of AFM grains lose magnetic order with increasing temperature, the sign of the
uncompensated moment is reversed. We have calculated the thermal evolution of the
uncompensated moment in a certain distribution of bidimensional AFM grains. Firstly, we
consider the contact interface to be a plane with an AFM arrangement of spins, as shown in
the idealized top view in figure 7. In order to make a realistic approximation we consider such
an interface to be formed by circular grains with ratio R (figure 7), reflecting the AFM grains
in the TEM images (figure 1). Within each grain the spins are arranged in a squared lattice
whose parameter will be related to the actual parameter of studied oxides. Let us define the
net moment per unit area of one of these AFM grains as


i 2 + j 2
R 2 (−1)
i+ j
mR = . (1)
π R2
The AF configuration providing a positive uncompensated moment has been selected
between the two possible spin arrangements for each grain, in order to simulate the field
cooling process carried out in the magnetic measurements. For that, the following initial
condition is taken:

(−1)i+ j > 0. (2)
i 2 + j 2
R 2

In order to simulate the increase of temperature, we eliminate the magnetic moment of

those spins placed in the grain boundary. Hence, we define a magnetic ratio Rmag in such a way
Reversal of exchange bias in nanocrystalline antiferromagnetic–ferromagnetic bilayers 10071

FM
AFM

a b
Figure 7. Schematic top view of the contact AFM–FM interface (a) and distribution of AFM grains
with circular symmetry (b) from the perspective of the right-hand figure.

that the magnetic order only exists within the grain (see figure 8). We define a new parameter
that is the relative thickness of the non-magnetic grain shell:
t f = R − Rmag (3)
where R and Rmag are referred to the lattice parameter (t f is a non-dimensional parameter).
The decrease of the magnetic diameter Rmag , i.e. the increase of t f , accounts for the progressive
magnetic transition of the grain boundary region induced by increasing temperature. According
to the results of TEM studies, we consider a log-normal grain size distribution centred on an
average ratio R0 . The total uncompensated magnetic moment of such a distribution will be
given by the following expression:
   R 2  
ln R0
M R0 = k exp − 2
(−1)i+ j . (4)
R
2σ i 2 + j 2
R 2 mag

Figure 9 shows the evolution of the total uncompensated magnetic moment as a function
of the thickness of the paramagnetic shell for sets of AFM grains with different diameter
distributions, calculated according to equation (4). Grain diameter distributions are centred
between 2.0 and 3.5 nm and the distribution width is in all cases 2.0 nm (parameters have
been obtained from the analysis of bright-and dark-field TEM images corresponding to cobalt
oxide). The initial uncompensated moment is positive and its value per unit area decreases with
the increase of the average grain diameter. As the grain boundary becomes paramagnetic, the
uncompensated moment decreases and, due to the AFM coupling within the grains, eventually
changes to negative when the thickness of the paramagnetic shell is of the order of one lattice
parameter (a = 0.427 nm for Co–O [22]). This result is in good agreement with those from
the structural analysis in which a typical value of 0.5 nm was obtained for the grain boundary
thickness. This fact indicates that the total transition of the grain boundary to the paramagnetic
state is responsible for the change of sign in the uncompensated AF moment.
Most of the magnetic quantitative features related to exchange anisotropy have not been
satisfactorily explained yet, but some qualitative aspects, such as the relation between the
magnitudes of shift [2] and coercivity enhancement [23], have been directly connected to
the strength of the interfacial coupling through the AFM/FM interface. Therefore, both
magnitudes, H E and Hc , will be a function of the uncompensated interfacial moment, and
their thermal evolution will reflect the thermal dependence of this uncompensated magnetic
10072 C Prados et al

Rmag

Figure 8. Scheme of one AFM grain with a total ratio R and a magnetic ratio Rmag . Notice that,
eliminating the spins on the boundary (white arrows), the uncompensated moment (black arrows)
is placed now on the border of the inner ratio (Rmag ).

Average grain size

3
2,0 nm
Uncompensated moment per unit area

2,5 nm
2 3,0 nm
3,5 nm

-1

-2
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0

Relative thickness of the non magnetic grain shell, tf

Figure 9. Uncompensated moment per unit area of a set of bidimensional AFM grains with different
average grain size calculated as a function of the relative thickness of the non-magnetic shell t f .
The initial spin configuration (t f = 0) has been selected to provide a positive uncompensated
moment in each grain, in order to simulate a field cooling process. Curves sketch the reversal of
the uncompensated moment induced by the magnetic transition of the grain boundary.
(This figure is in colour only in the electronic version)
Reversal of exchange bias in nanocrystalline antiferromagnetic–ferromagnetic bilayers 10073

1,5
Co-O (20 nm)/Py (20 nm)

1,0 Exc hange field der ivative

T B, c or e ~ 180 K
0,5
d H E,c /d T (a. u.)
0,0
T B, shell ~ 150 K
-0,5

-1,0 Coer c ivity der ivative

-1,5

-2,0
50 100 150 200 250 300
Temper atur e, T (K )

Figure 10. Thermal evolution of temperature derivative of exchange field and coercivity performed
from experimental data shown in figure 2. Two transition temperatures appear: the first one
(∼150 K) is related to the magnetic transition of grain boundaries and the higher one (∼180 K) to
the transition of grain cores of the AFM layer.

moment. Figure 10 shows the temperature derivative of the experimental data corresponding
to H E and Hc for CoO/Py bilayers (shown in figure 4) where two transition temperatures are
clearly distinguished. The lower one, TB,b ∼ 150 K, is identified as the blocking temperature
of the uncompensated moment of the AFM grain boundaries, and it corresponds to the onset
of the reversal of the exchange anisotropy. The higher transition temperature, TB,c ∼ 180 K,
is the blocking temperature of grain cores in the AFM layer. Between both temperatures, the
sign of the uncompensated AFM moment is opposite to that induced during the field cooling
process.

5. Conclusions

We have observed and measured a positive exchange bias in nanocrystalline Py/Co–O and
Py/Ni–O bilayers at temperatures approaching the AFM transition. This effect also appears
after magnetic training of the samples at a determined temperature. This phenomenon is
qualitatively attributed to the existence of two different transition temperatures in the AFM
grains, associated with boundaries and cores, and to the change of sign of the uncompensated
moment of the AFM grains after the magnetic transition of boundaries. The observed
effect becomes sizeable due to the nanocrystalline nature of the oxide layers. The thermal
evolution of hysteresis loop shift and coercivity provides an immediate way of determining
and discriminating both the grain boundary and grain core transition temperatures in these
nanocrystalline materials.

Acknowledgment

This work has been supported by the Spanish CICyT agency under project MAT98-0965-C04.
10074 C Prados et al

References

[1] Meiklejohn W H and Bean C P 1956 Phys. Rev. 102 1413

[2] Nogués J and Schuller I K 1999 J. Magn. Magn. Mater. 192 203
[3] Berkowitz A E and Takano K 1999 J. Magn. Mater. 200 552
[4] Dieny B et al 1991 J. Appl. Phys. 69 4774
[5] Nogués J, Lederman D, Moran T J and Schuller I K 1996 Phys. Rev. Lett. 76 4624
[6] Kiwi M, Mejia-Lopez J and Portugal R D 2000 Solid State Commun. 116 315–19
[7] Schlenker C and Paccard D 1967 J. Physique 28 611
[8] Nikitenko V I et al 1998 Phys. Rev. B 57 R8111
[9] Gökemeijer N J, Cai J W and Chien C L 1999 Phys. Rev. B 60 3033
[10] Kiwi M 1999 J. Magn. Magn. Mater. 234 552
[11] Stamps R L 2000 J. Phys. D: Appl. Phys. 33 R247–68
[12] Mauri D et al 1987 J. Appl. Phys. 62 3047
[13] Stiles M D and McMichael R D 1999 Phys. Rev. B 55 3722
[14] Malozemoff A P 1987 Phys. Rev. B 35 3679
Malozemoff A P 1998 J. Appl. Phys. 63 3874
[15] Takano K et al 1997 Phys. Rev. Lett. 79 1130
[16] Liu K et al 2001 Phys. Rev. B 63 060403
[17] Gruyters M and Riegel D 2001 Phys. Rev. B 63 052401
[18] Nogués J, Lederman D, Moran T J and Schuller I K 1996 Phys. Rev. Lett. 76 4624
[19] Navarro E, Hernando A, Yavari A R, Fiorani D and Rosenberg M 1999 J. Appl. Phys. 86 2166
[20] del Bianco L, Ballesteros C, Rojo J M and Hernando A 1998 Phys. Rev. Lett. 82 4500
[21] Lederman D, Ramos C A, Jaccarino V and Cardy J L 1993 Phys. Rev. B 48 8365
[22] Kittel C 1993 Introduction to Solid State Physics ed Reverte p 28
[23] Li Z and Zhang S 2000 Phys. Rev. B 61 R14897