You are on page 1of 21

C 3 Interlayer Exchange Coupling

Peter H. Dederichs
Institut f ur Festk orperforschung
Forschungszentrum J ulich GmbH
Contents
1 Introduction 2
2 Theoretical Background 3
2.1 Relation between Interaction Energies and Density of States . . . . . . . . . . 3
2.2 Simple 1-dimensional plane-wave model . . . . . . . . . . . . . . . . . . . . . 3
2.3 Green Function and Density of States . . . . . . . . . . . . . . . . . . . . . . 5
2.4 Interlayer Exchange Coupling . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5 Asymptotic Analysis for Large Spacer Thicknesses . . . . . . . . . . . . . . . 7
3 Ab-initio Calculations for IEC of Co/Cu/Co Layers 8
4 Reection Coefcients for Cu/Co (001) 12
5 Experimental Observations of Interlayer Coupling 15
A Change of Integrated Density of States N(E) 18
B Frozen Potential Approximation for Magnetic Interactions 19
C3.2 Peter H. Dederichs
Cu Co (Cu) Co Cu
n
Fig. 1: Interlayer exchange coupling: coupling of two ferromagnetic layers of Co atoms through
a non-magnetic Cu spacer layer.
1 Introduction
Interlayer exchange coupling denotes the coupling of two ferromagnetic layers via a non-
magnetic spacer layer; it is schematically illustrated in Fig. 1 for two Co layers, being coupled
through a non-magnetic Cu layer. The effect was discovered in 1986 by Peter Gr unberg [1],
who found an antiferromagnetic coupling for two Fe layers separated by Cr layers; in previ-
ous experiments only ferromagnetic coupling had been observed, originating from Fe pinholes
through the Cr spacer, which enforced the ferromagnetic coupling. Later Parkin [2] showed
for a large number of systems, that the interlayer coupling depends oscillatory on the spacer
thickness. The exact relation to the Fermi surface of the spacer was pointed out by Bruno and
Chappert [3]. For a recent review of interlayer exchange coupling see [4].
It is generally agreed, that the interlayer exchange coupling is an interference effect, arising
from the repeated spin dependent reection of electrons at the two interfaces between the spacer
and the ferromagnetic layers. Basically it is the two-dimensional analogue to the RKKY inter-
action between two magnetic impurities in a non-magnetic host. However in two dimensions
the effect is more spectacular, since in case of strong reections, the electrons are spatially con-
ned to the spacer layer leading to particularly strong coupling. The discovery of interlayer
coupling was soon followed by the discovery of the even more important Giant Magnetoresis-
tance (GMR) by Fert and Gr unberg in 1988. Both effects paved the way for the new elds of
Magnetoelectronics and Spintronics.
In this lecture we will focus on the basic understanding of interlayer exchange coupling (IEC) as
an interference effect and discuss the relation between the electronic structure of the magnetic
layer and the spacer layer on the one hand and the interface reectivity on the other hand, which
are the essential ingredient for the IEC. We demonstrate these relations in ab-initio calculations
for the model system Co/Cu/Co. Finally we review some experimental results on interlayer
coupling and quantum well states.
Interlayer Exchange Coupling C3.3
2 Theoretical Background
2.1 Relation between Interaction Energies and Density of States
In the independent particle approximation the total energy E
tot
is related to the density of states
(DOS) n(E) by
E
tot
=

E
F
dE(E E
F
)n(E) =

E
F
dEN(E) with N(E) =

E
dE

n(E

) (1)
Here E
tot
refers to the energy of a grand canonical system, for which the Fermi energy E
F
as
the chemical potential is the basic variable. The second equality follows by partial integration.
N(E) is the integrated density of states, counting the number of states with energies E

below
E. The interaction energy E
inter
between two defects 1 and 2 in a given host, or between two
Co layers in a Cu matrix as sketched in Fig. 1, is determined by
E
inter
= E
1+2
E
1
E
2
(2)
where E
1
and E
2
are the changes of the total energy due to the isolated layers 1 and 2 and
E
1+2
is the change due to the interacting layers 1 and 2. Therefore the interaction energy is
directly related to the corresponding changes of the density of states
n
int
(E) = n
1+2
(E) n
1
(E) n
2
(E) (3)
and n
int
(E) is directly caused by the interference of the waves scattered at layer 1 and layer
2, which modies the DOS [5].
2.2 Simple 1-dimensional plane-wave model
We expect therefore, that the DOS n
int
arises from waves which are repeatedly reected at
the layers 1 and 2. Let us denote by the phase shift of the reection coefcient r =| r | e
i
of the incident plane wave. For a complete round trip of the incident wave k scattered at layer
1 and 2 the total phase of the wave changes by = (2kD +
1
+
2
), where D is the spacer
thickness. The contribution to n
int
(E) should therefore be given by
n
int
(E)

=
2

2D
dk
dE
| r
1
| | r
2
| cos(2kD +
1
+
2
) (4)
For a constructive interference, i.e. = 2n, the DOS is increased, while for a destructive
interference n
int
< 0. Moreover n
int
should be proportional to| r
1
| and | r
2
| and a
prefactor containing the density of k-values per unit energy and the width D of the spacer. By
summing up all higher-order round-trip processes one obtains for the DOS
n
int
(E) =
4D

dk
dE

n=1
| r
1
r
2
|
n
cos n(2kD +
1
+
2
) (5)
=
2

Im

i2D
dk
dE
r
1
r
2
e
i2kD
1 r
1
r
2
e
i2kD

and for the integrated DOS:


N
int
(E) =
2

Im ln(1 r
1
r
2
e
i2kD
) (6)
C3.4 Peter H. Dederichs
from which the interaction energy E
int
can be calculated via eq. (1) by integration (and n
int
by differentiation). n
int
(E), N
int
(E) and E
int
(E
F
) are illustrated in Fig. 2 for several
r
1,2
values [5]. For small r-values (| r | 0.5) one obtains a sin-like behavior correspond-
ing to the weak interference described by the cos-function in eq. (4). For strong reection
(| r |

= 1) N(E) shows jumps of height 1 (per spin direction), signalizing that at these en-
ergies / thicknesses the interactive DOS n
int
exhibits a -function corresponding to a bound
state, in which the electrons are fully conned to the spacer. Note that the interaction energy
E
inter
, obtained by integration, is positive, meaning a repulsive interaction, if the bound state
or resonance peak is at E
F
, while a negative, i.e. attractive, interaction results if the occu-
pied and unoccupied DOS peaks are equally spaced from E
F
. These bound or resonant states
localized in the spacer are called quantum well states.
Fig. 2: The interaction contributions to the density of states n
int
(E), to the integrated density
of states, N
int
(E), and to the total energy, E
int
(E
F
), for a one-dimensional jellium model
as a function of the layer thickness D.
Interlayer Exchange Coupling C3.5
2.3 Green Function and Density of States
The Green function G(r, r

; E), dened as the causal solution of the Schroedinger equation with


a unit source term (r r

) at positions r


2
2m

2
r
+V (r) E

G(r, r

; E) = (r r

) (7)
describes the propagation of electrons of energy E, being stationary produced at the position r

,
in a potential V (r). Formally, the Green function can be represented in terms of the eigenfunc-
tions

(r) and eigenvalues E

(spectral representation)
G(r, r

; E) =

(r)

(r

)
E +i E

with +0 (8)
where the limit +0 guaranties causality. Using the identity for 1/(x +i)
1
x +i
= P (
1
x
) i(x) (9)
where P denotes the principle value and (x) the Dirac -function, and the completeness rela-
tion for the

(r)

(r

) = (r r

) (10)
the spectral representation (8) can be veried by insertion into eq. (7). Now we can prove that
the density of states n(E), dened by
n(E) = 2

(E E

) (11)
can be evaluated from the Green function by a trace operation
n(E) =
2

Im

drG(r, r; E) = 2

(EE

dr |

(r) |
2
=
2

ImTrG(E) (12)
Here we want to calculate the change n(E)in a crystal arising from a potential V due to e.g.
a defect or a layer. In symbolic notation the Green function G(E) and its unperturbed counter
part can be written as
G(E) =
1
E +i H
, G
o
(E) =
1
E +i H
o
(13)
As is shown in the appendix A, the change of the integrated DOS N(E) due to the potential
V is given by
N(E) =
2

ImTr ln (1 G
o
(E) V ) (14)
Now we consider two perturbations, i.e V = V
1
+ V
2
. As required by eq. (1) and (2), we
need for the interaction energy the change of the integrated DOS N
int
due to the interaction
between the layers 1 and 2. As shown in appendix A, the change N
int
is given by
N
int
=
2

ImTr ln(1 G
21
o
t
1
G
12
o
t
2
) (15)
which is the proper generalization of the simple model result of eq. (2). Here the transition
matrices t
1
and t
2
describe the scattering of the single layers 1 and 2 embedded in the unper-
turbed host matrix, whereas the Green functions G
12
o
and G
21
o
describe the propagation in the
host material from the layer 1 to the layer 2 and back.
C3.6 Peter H. Dederichs
2.4 Interlayer Exchange Coupling
The adaptation of this formula to the geometry of two parallel layers 1 and 2 embedded in an
ideal crystal is non-trivial; therefore only the results are given here. Firstly, the two-dimensional
periodicity of the two layers and the crystal in the plane of the layers can be exploited, such that
by using Blochs theorem the trace over all atoms in the same monolayer can be performed ana-
lytically, so that from this summation only an integral over the two-dimensional Brillouin zone
of area A
BZ
remains. The interaction energy of two spin-polarized magnetic layers embedded
in a non-magnetic spacer is then given by [6]
E
int
=
1

E
F
dE
1
A
BZ

BZ
dq

ImTr ln M

(q

; E) (16)
with M = {1 G(q

; E) R

1
(q

; E) G(q

; E) R

2
(q

; E)}
nn

LL
. Here is the spin index. Fur-
thermore M, G and R are matrices (not operators) with angular momentum indices L, L

and
monolayer indices n, n

. G(q

; E) describes the propagation of Bloch waves with in-plane q

-
vectors from one layer to the other one, while the R

(q

; E) denote the generalized reection


matrices. If we evaluate this formula for the energy difference between the ferromagnetic align-
ment of the layers (E
FM
) and the antiferromagnetic one (E
AF
) the interlayer exchange coupling
energy E
IEC
= E
FM
E
AF
is given by
E
IEC
=
1
A
BZ

E
F
dE

BZ
dq

Imln
det(1 G
o
R
+
1
G
o
R
+
2
) det(1 G
o
R

1
G
o
R

2
)
det(1 G
o
R
+
1
G
o
R

2
) det(1 G
o
R

1
G
o
R
+
2
)
(17)
Here the nominator describes the interaction for the ferromagnetic state and consists of the
contributions from the majority electrons (with reection coefcients R
+
1
, R
+
2
) and from the
minority electrons (with R

1
, R

2
) while the denominator denotes the interaction for the anti-
ferromagnetic conguration (with the mixed coefcients R
+
1
, R

2
and R

1
, R
+
2
). Very often the
reection coefcients are relatively small, and then an expansion of (17) yields
E
IEC
=
1
A
BZ

E
F
dE

BZ
dq

ImTr {G
o
R
1
G
o
R
2
} with R = R
+
R

(18)
In general we see that the interlayer exchange coupling is determined by two different quan-
tities: (i) the Green function of the nonmagnetic host, describing the propagation from one
layer to the other one and back and (ii) the spin dependent reectivity of the two layers, given
by the matrices R

1
and R

2
, which, as we will discuss in the following, are strongly spin de-
pendent, i.e. very different for majority electrons (R
+
1,2
) and minority electrons (R

1,2
). For
instance, for Co/Cu(001) the reection coefcients for minority electrons are very large while
the coefcients for the majority electrons are small (see sect. 4). Then the interlayer exchange
coupling is dominated by the energy E
FM
of the ferromagnetic conguration, and here by the
minority electrons. Moreover only for the minority electrons of the ferromagnetic conguration
pronounced quantum well states occur (see sect. 5).
In the whole section 2 we have considered only the interaction energy in the independent particle
approximation, which is not valid for interacting electron systems. However, as proven in
appendix B [17], the so-called single-particle terms arising from the expression (3) for the DOS
or (15) for the integrated DOS gives the correct result for the magnetic interaction energy E
FM

E
AF
. Therefore the expression (17) is asymptotically correct in the limit of large distances
and is also a good approximation for short distances, except when the two layers are in direct
contact.
Interlayer Exchange Coupling C3.7
2.5 Asymptotic Analysis for Large Spacer Thicknesses
In the numerical calculations shown in the following section, the interlayer coupling energy
E
IEC
is evaluated from eq. (17) by introducing an energy mesh for the occupied states and a q

-
mesh in the 2-dimensional Brillouin zone. Here a typical complication arises for larger spacer
thicknesses: the Green function G
o
(q

; E) strongly oscillates both as a function of E and q

, so
that a very high density of energy and q

-points is needed. Bruno and Chappert [3] have shown


that in the limit of large spacer thicknesses an asymptotic expansion of eq. (18) can be derived,
which is extreme importance for understanding the IEC. The method exploites the fact that for
large distances the Green function strongly oscillates of a function of E and q

, so that most
energy and q

regions give asymptotically no contributions. Signicant contributions to the


integral arise only from those E and q

regions, where the total phase describing the oscillations


is stationary (stationary phase approximation). This analysis shows, that contributions to the
IEC arise only
(i) from energies E in a small stripe E
F
O(
k
F
D
) E E
F
around the Fermi energy E
F
of
the spacer, which contracts to the Fermi surface for D .
(ii) from such q

-vectors q
c
in the 2-dim. Brillouin zone, for which the spanning vector of the
Fermi surface normal to the interface is stationary. Denoting these calipers by 2k
c
, the IEC is
asymptotically given by
E
IEC

=

c
A
c
D
2
cos(2k
c
D +
c
) (19)
Thus asymptotically the coupling energy scales as
1
D
2
times a periodic function of 2k
c
D, with
the period determined by the caliper 2k
c
. In general more than 1 critical q
c
-values can occur,
so that a multiperiodic behavior is obtained. The properties of the magnetic layers enter only in
the phase
c
and in the amplitude A
c
, which is basically quadratic in the reection asymmetry
A
c
| R
+
o
R

o
|
2
of the majority and minority channels.
Fig. 3: Cross section of the Fermi surface of Cu in the [1

10] plane passing through the origin.


The indicated arrows in the (001), (110) and (111) directions determine the periods of the
interlayer exchange coupling for layers with [001], [110] and [111] orientations.
The above formula represents an extreme simplication, since the inspection of the Fermi sur-
face of the spacer material, which is well known for most materials, allows to determine the
dependence of interlayer coupling on the spacer thickness. For free electrons the Fermi surface
C3.8 Peter H. Dederichs
is a sphere and then only one such spanning vector exists being given by the diameter 2k
F
of
the Fermi sphere. However in general more such calipers exist, as we will discuss now for the
Fermi surface of Cu, being representative for the noble metals. Fig. 3 shows a cross section
of the Cu Fermi surface [5] along the (110) plane passing through the origin. For (001) layers
two critical calipers are obtained, representing the smallest and the largest diameter of the dog-
bone in (001) direction. They give rise to oscillations with a long and a short period. Detailed
experimental studies (see section 5) yield periods which are in excellent agreement with the
information from Fermi surface studies and with ab-initio calculations of IEC. For the (110)
direction, four different periods are obtained, only one of them, i.e. the length of the dog-bone,
lies in the (110) plane and is shown in Fig.3 . In (111) direction only a single, long period is
obtained given by the diameter of the neck of the Fermi surface in (111)-direction.
3 Ab-initio Calculations for IEC of Co/Cu/Co Layers
In this section we present the results of ab-initio calculations for the Co/Cu/Co systems, which
have been obtained at Juelich. The calculations are based on density functional theory in the
local density approximation and apply the Korringa-Kohn-Rostoker (KKR) method for layered
systems. For details we refer to a recent review [7]. Many similar calculations and results
have been presented by Kudrnovsky et al. [10]. For IEC and Giant Magnetoresistance trilayers
of Co/Cu/Co(001) are a model system which shows nearly perfect epitaxial growth, which is
partly due to the small mist of the lattice constants. Moreover the Fermi surfaces are known
very accurately from de Haas-van Alphen and cyclotron resonance experiments [8]. Finally the
Fermi surface of the noble metals is relatively simple and does not depart very much from the
free-electron Fermi sphere.
Co/Cu/Co(001) trilayers: As a typical example of the results, Fig. 4 shows the calculated IEC
energies E
IEX
= E
F
E
AF
as a function of the Cu spacer thickness, given in monolayers
(ML), for three different thicknesses of the two Co-layers, i.e. 1 ML Co (dashed line), 5 ML
Co (dotted line) and 7 ML Co (solid line) [6]. Positive energies mean that the antiferromagnetic
coupling is preferred, while for negative energies the ferromagnetic alignment is stable. One
nds, that the results for 5 and 7 ML Co are very similar, but signicant differences are seen for
1 ML Co. In fact, all Co layers with a thickness of 3 or more MLs are very similar, only 1 and 2
ML Co behave differently. These differences are, however, very large: For 1 ML Co a dominant
long oscillation period with wave length
L
of about 6 ML is observed, corresponding to the
small k
c
value for the waist of the dog-bone, while for larger thicknesses the small oscillation
period of
S

= 2.6 ML is most signicant.
A t of the numerical results to the asymptotic expression with two oscillation periods
S
and

L
and analogous amplitudes A
S
and A
L
shows, that for large distances, i.e. from 8-10 ML
on, a very accurate t is obtained, but signicant deviations occur for smaller distances. Thus
the asymptotic results are valid only for rather large distances, but are nevertheless qualitatively
also meaningful for shorter spacer thicknesses. For 1 ML Co one obtains an amplitude ratio of
A
S
/A
L

= 0.75, while for large Co thicknesses A
S
/A
L
10, showing that the long period is
much less important.
In the numerical calculations we can analyze the different oscillation periods by dividing the
q

-integral over the two dimensional Brillouin zone into two areas, as is shown in Fig. 5. The
rst one is centered around the q

-value q
1
= 0, which is the critical q
c
-value for the long
period, where as the second area is the outer region of the Brillouin zone containing the four
Interlayer Exchange Coupling C3.9
Fig. 4: Calculated exchange coupling energies (IEC) for Co/Cu(001) as a function of the Cu
spacer thickness for three different thicknesses of the Co layers, i.e 1 ML Co (dashed line with
triangles), 5 ML Co (dotted line with empty squares) and 7 ML Co (full line with lled squares).
equivalent critical q
2
-values in (100) direction being related to the short period. The advantage
of this decomposition is, that the sum of both contributions still gives the exact coupling for all
Cu thicknesses and that due to the decomposition also the preasymptotic behavior is correctly
described.
Fig.6 shows this decomposition as a function of the spacer thickness for several Co thicknesses
from 5-11 monolayers. As already noted before, the short period oscillation, labeled by q
2
, has
a much larger amplitude than the long period obtained from the area around q
1
= 0. However
we also see a second, equally important difference: the amplitude and phase of the short period
is independent of the Co thickness, since all curves coincide on each other, while for the long
period the curves are signicantly different, in particular they are phase-shifted and have also
different amplitudes. This strikingly different behavior asks for an understanding, which can be
obtained from a calculation of the reection coefcients entering in the IEC-expression. This
will be discussed in the next section. In short we will nd that for the q
2
-value the Cu waves
will be totally reected at the Cu/Co interface, and do therefore note signicantly penetrate into
the Co layer, so that no important dependence on the Co thickness is found.
Results for Co/Cu/Co(110) layers: The analysis of the interlayer coupling in terms of RKKY
oscillation is slightly more complicated than for the (001) direction. For the (110) orientation
four inequivalent stationary calipers exist, of which two are associated with the same critical
q

-point [9]. The three stationary q

-points in the two dimensional Brillouin zone are: q


1
= 0
(

point, center of BZ), q


2
=

S (corner point), q
3
=

X(001). The period at

point is about 2.1
ML; the k
c
-value is given by the length of the dog-bone in Fig.3. The calipers of the

S and

X
points are not in the plane shown in Fig.3. For the

S-point there exist two stationary calipers of
2.6 and 9.6 ML. The large period corresponds to the diameter of the neck and the short one to
the distance between two adjacent neck in (110) direction. The caliper of the

X-points yields a
period of about 5.3 ML.
In Fig. 7 we show the contributions to the IEC arising from the division of the total Brillouin
zone into inuence areas about the 3 critical q

-values [11], such that the sum of all three


contribution gives the total coupling. Firstly we see that the contribution from the

X-point
is rather small and not important. For short distances the contribution q
2
from the

S-point
C3.10 Peter H. Dederichs
Fig. 5: Two dimensional Brillouin zone for the [001] orientation. Shown are also the positions
of the ve stationary points, the two inequivalent ones are located at q
1
(0, 0) =:

and q
2
=
(1.08, 0)/a. The dashed line indicates the subdivision of the 2d-BZ into areas A
i
(I = 1, 2)
around the stationary points. The shaded area denotes the irreducible part of the 2D Brillouin
zone.
Fig. 6: The q
1
and q
2
contributions to the interlayer coupling of Co/Cu/Co(001) obtained by
dividing the Brillouin zone in q
1
- and q
2
-regions as in Fig. 5 for 5 ML Co (full line), 7 ML
(dashed line) and 11 ML Co (dotted line). The q
2
-curves coincide for these Co thicknesses.
Interlayer Exchange Coupling C3.11
Fig. 7: The q
1
=

, q
2
=

S and q
3
=

X-contributions to the IEC energies of 5 ML thick
Co layers in Cu (110).
Fig. 8: Calculated exchange coupling energies (IEC) for Co/Cu (111) as a function of the Cu
spacer thickness for a Co layer thickness of 5 ML.
dominates the behavior by a large peak at 4-5 ML. Here the amplitude of the long period (9.6
ML) is bigger than the small one (2.6 ML). For larger distances the q
1
-contribution from the

-point becomes more important. This oscillation shows an interesting beating effect, which
should not be confused with an additional oscillation period. This beating is caused by the
special value of the oscillation period of 2.1 ML, which is close to an integer number (2 ML) of
layers. Thus the long wavelength modulation of the antiferromagnetic coupling of neighbouring
layers does not arises from a new period, but is an incommensurability effect.
Results for Co/Cu/Co(111) layers: For the (111) orientation the calculated interlayer coupling
is shown in Fig. 8. One can see, that for large thicknesses only one period is important. This
period of 4.3 ML is in good agreement with Fermi surface data and the critical caliper of the
Cu-neck shown in Fig. 3.
C3.12 Peter H. Dederichs
4 Reection Coefcients for Cu/Co (001)
As we have learned in sect 2, the oscillatory dependence of the interlayer coupling on the
spacer thickness is determined by the Fermi surface of the spacer. However the amplitude
of the coupling as well as the phase shifts of the oscillations are determined by the magnetic
layers, more precisely by the spin dependent, complex reection coefcient of the spacer Bloch
waves at the magnetic layer. Moreover the asymptotic analysis as well as the ab-initio results
presented in the last section show, that of all q

values in the 2-dim. Brillouin zone only those


critical values q
c
are important for which the Fermi surface caliper perpendicular to the interface
is stationary.
For the model systems Co/Cu(001), these are the two q

values q
1
= 0 and q
2
shown in Fig.
5, the corresponding calipers of which are shown in Fig. 3 and give rise to the long and short
oscillation periods. Our aim is to obtain a deeper understanding of the IEC-results presented in
Fig. 4 and Fig.6. Naturally two questions arise: (i) Why does the q
1
-contribution dominate for
1 ML Co, but the q
2
-contribution for thicker Co layers? (ii) Why does the q
1
-contribution vary
with Co thickness, but the q
2
-contribution not?
In order to answer these questions, we have to analyze the electronic structure of the Co layers.
Due to q

-conservation we can focus on the band structure and density of states for given
q

= q
1
and q

= q
2
.
Band Structures of Cu and Co for q
1
and q
2
: In the rst step, we just consider the bulk
band structures of Cu and Co for these two q

-values as a function of the k


z
-value [6]. This
is relevant for the two half-space problem: the scattering of a Cu-Cu Bloch wave at the Co/Cu
interface and the partial transmission into the Co half-space, the partial reection back into the
Cu half-space, respectively. The upper part of Fig. 9 shows for q
1
= 0 the Cu band structure
and the majority and minority bands for ferromagnetic fcc Co. The lower part of Fig. 9 shows
the corresponding bands for the q
2
-value. In Cu the relevant band at E
F
has
1
-symmetry
and can only couple to Co states of the same symmetry. Therefore all
1
-bands are denoted
in Fig. 9 by full lines, while the dashed lines indicate d-bands of other symmetry. Locally the

1
-band consists of s-, p
z
- and d
z
2-admixtures. The Cu and the Co majority bands at E
F
are
rather similar. Since the typical d-bands are below E
F
, the bands at E
F
are mostly of s- and
p
z
-character. Due to this similarity of the bands, one expects that a Cu-spin electron can easily
penetrate the interface, so that the reection coefcient is small. On the other hand for spin
electrons, the Co band at E
F
has mostly d
z
2 character, so that the reection coefcient should
be much larger.
Qualitatively the same behavior is obtained for the q
2
value. For spin, the Cu and Co majority
bands are similar, so that a small reection coefcient is expected. However the Co minority
band is very different. In fact, the Fermi energy lies in a
1
-gap, so that no allowed Co minority
Bloch waves exist, which can carry a current. Therefore at E
F
the incident Cu Bloch wave is
fully reected. On the Co side the incident wavefunction is matched to purely evanescent states,
i.e. Bloch waves with a complex k-vector which decrease exponentially and cannot contribute
to the current. Depending on the size of the imaginary part, the incident Bloch wave sees
therefore only a few Co monolayers.
Thus from the band structures we expect: (i) only the minority electrons are strongly reected,
(ii) the minority electrons at q
2
are fully reected, which, due to the short penetration into the
Co layers, explains, why in Fig. 6 the q
2
-contribution to the IEC is for thicker Co layers much
larger than the q
1
-contribution and independent of the Co thickness. However these arguments
are only obtained from the bulk band structure and cannot really be applied to the electronic
Interlayer Exchange Coupling C3.13
d d
Fig. 9: Band structures of Cu and Co in the direction k
z
perpendicular to the [001] plane for
the two critical q

-values q
1
= 0(

point, above) and q


2
= (1.08, 0)/a (below). Dashed line
denote d-states, which for symmetry reasons cannot couple to the Cu-bands at around E
F
.
structure of say 5 ML Co embedded into a Cu crystal. Therefore in the following we consider
in detail the electronic structure of small Co(001) layers in Cu, calculate the DOS of these layers
and compare them with the calculated reection coefcients for this layer thickness. This gives
a much more detailed and basic information about the IEC in these systems than the bulk band
structure.
Density of States of small Co(001) layers: Fig. 10 left shows the density of states of layers
of varying thickness [12], more precisely the DOS for the given q

-value q
1
= 0, which is
averaged over all Co atoms in the layer. The thickness ranges form 1 to 60 ML. The full lines
refer to the minority DOS, the dashed lines to the majority one, being shifted to lower energies.
For 1 ML Co(001) in Cu a broad d
z
2 resonance exists at about 0.8 eV, while for 2 ML Co
this resonance is shifted to 0.2 eV. In this case a second resonance appears at much lower
energies ( 4 eV, not shown in Fig. 10). Both resonances can be understood as bonding
C3.14 Peter H. Dederichs
Fig. 10: Left: Local density of states of (001) Co layers with thicknesses of 1 ML, 2 ML, 4 ML,
10 ML and 60 ML Co embedded in Cu. All DOS refer to the projected DOS for the stationary
point q
1
= 0, with only states of
1
-symmetry being included. Full lines: minority electrons,
dashed lines: majority electrons (shown is not the Co-DOS, but the changes of the Co DOS with
respect to the Cu DOS). Right: Reection coefcients of an incident Cu
1
-wave with q
1
= 0
on Co(001) layers for 1, 2, 4, 10 and 40 ML thickness. Full lines minority electrons, dashed
lines: majority electrons.
and antibonding d
z
2 hybrides between the two monolayers. When we increase the number of
Co layers, both the upper and lower hybrides split further, such that for 10 ML Co we have
Interlayer Exchange Coupling C3.15
ve upper resonances and ve lower ones, with a vanishing DOS in between. Thus apparently
the relatively at d
z
2-bands of the Co spin electrons shown in the band structure of Co at
around the Fermi level and at 5 eV split into more and more resonances, showing the size-
quantization of the band structure for a nite layer thickness. This is most clearly seen for 60
ML Co, where due to numerical resolution problems the smearing of the DOS leads to a typical
1-dimensional DOS, obtained by summing for given q
1
= 0 over all k
z
-values and showing van
Hove singularities at the band edges.
Reection Coefcient at Co(001) layers: Fig. 10 shows for the same sequence of Co layers
the resulting reection coefcients for Cu waves with given q
1
= 0 incidents on the Co layer
[12]. Again the full lines refer to minority waves, the dashed lines to majority ones. For 1 ML
Co we obtain for the minority waves full reection at about 0.8 eV, where the shape of the
curve resembles the form of the Co DOS. For 2 ML Co the resonance peak is shifted to 0.2
eV, the peak of the DOS, but very strong reectivity also occurs for lower energies. This is
more clearly seen for 4 and 10 ML thickness where below about 0.6 eV the incident wave is
nearly totally reected due to missing Co
1
-states for these energies. In the bulk this is just
the energy region of the bulk
1
-gap.
Amore interesting behavior is obtained above 0.6 eV, where the DOS shows the size-quantized
resonances or quantumwell states in the Co layer. For 4 ML Co we have two resonances at 0.4
eV and at E
F
which result in two peaks in the reection coefcients at these energies. In addi-
tion two energies with zero reection coefcients (antiresonances) occur in between. For 10
and 40 ML Co this behavior shows up more dramatically. In fact it is not fully reected in the
calculated results due to numerical resolution problems. One can show, that at each resonance
of quantum well state in the Co layer the scattering phase changes by , causing both a maxi-
mum of 1 in the reection coefcient, when the total phase is (n +
1
2
) and a strong resonance
exist, and a zero reection coefcient, when the total phase crosses the value n, meaning full
transmission. At higher energies, i.e. above the Co d-band edge of 0.2 eV, the reection is much
smaller and the oscillations are much wider spaced due to the strong dispersion of the sp-band
arising from the large overlap of the wave functions. Note that for the majority electrons the
same behavior of weak reectivity and widely spaced oscillations are found in the whole energy
region considered.
For the q
2
-point the behavior of the DOS and the reectivity is basically very similar. The
important difference is, that the band gap occurs in the region of the Fermi level, so that already
for a few Co monolayers full reection occurs. Only 1 and 2 ML Co show a different behavior.
5 Experimental Observations of Interlayer Coupling
After the discovery of interlayer coupling by Gr unberg [1] and its oscillatory origin by Parkin
[2] many groups have studied the interlayer coupling for a rich variety of systems. Usually the
coupling as a function of the angle between the two magnetic layers is written as
E
int
= J
1
cos J
2
cos
2

where J
1
= E
AF
E
FM
is the so called bilinear coupling and J
2
is the bi-quadratic
coupling, which, for J
2
< 0, favors a 90
o
coupling. Usually J
1
is much larger than J
2
.
For the measurement of the coupling, several methods exist [4]. For instance, starting from a
stable antiferromagnetic system, a magnetic eld is applied to align both magnetizations paral-
lel to each other in the magnetic eld. The evaluation of the remagnetization curve for a Fe/Au
C3.16 Peter H. Dederichs
Table 1: Selection of observed bilinear coupling strengths and periods collected from the liter-
ature [4, 13].
Sample Maximum strength J
1
in mJ/m
2
Periods in ML and (nm)
(at spacer thickness in nm)
Co/Cu/Co(100) 0.4 (1.2) 2.6 (0.47); 8 (1.45)
Co/Cu/Co(110) 0.7 (0.85) 9.8 (1.25)
Co/Cu/Co(111) 1.1 (0.85) 5.5 (1.15)
Fe/Au/Fe(100) 0.85 (0.82) 2.5 (0.51); 8.6 (1.75)
Fe/Cr/Fe(100) >1.5 (1.3) 2.1 (0.3); 12 (1.73)
Fe/Mn/Fe(100) 0.14 (1.32) 2 (0.33)
Co/Ru(0001) 6 (0.6) 5.1 (1.1)
Co/Rh/Co(111) 34 (0.48) 2.7 (0.6)
wedge/ Fe structure grown on a Ag-buffered GaAs(001) substrate is shown in Fig.11a [13, 14].
For small distances strong ferromagnetic coupling is observed, partly due to ferromagnetic pin-
holes and bridges. A double period structure with 2.48 ML and 8.6 ML is found, in good
agreement with Fermi surface studies. Fig. 11b shows similar results for a Fe/Au/Fe trilayer
grown on a Fe-wisker. As one can see, in this case the coupling strength is an order of magnitude
larger, but still smaller then the calculated values. This is typical for practically all experiments.
The periods agree very well with Fermi surface data and ab-initio calculations, but the ampli-
tudes are considerably smaller then calculated. In particular, the short period oscillations are
strongly suppressed, and very often they cannot be found at all. This can be easily understood
from the interface roughness. For instance, in the presence of step-like roughness, the interlayer
coupling has to be averaged over neighbouring layer distances, which strongly suppresses the
short periods. The same is true for interdiffusion, where also ab-initio calculations show that
the short periods are strongly affected.
0 10 5 15 20 25
Spacer thickness (ML)
0.00
-0.01
-0.02
-0.03
J
1

+

J
2
(
m
J
/
m
2
)
2/Q
1
2/Q
2
(a)
(a)
(b)
0.0 1.0 0.5 1.5 2.0 2.5 3.5 3.0
Spacer Thickness (nm)
Fig. 11: Coupling strength of Fe/Au-wedge/Fe(001) trilayers as a function of Au interlayer
thickness (a) on a Ag-buffered GaAs substrate [14] and (b) on an Fe (001) wisker [4]. The inset
in (a) includes ranges where the coupling is ferromagnetic (J
1
+J
2
> 0).
Interlayer Exchange Coupling C3.17
A selection of experimental observations of the IEC, i.e. the bilinear coupling strength J
1
and
the oscillation periods, is presented in Table 1 (from ref. [4], [13]).
Fig. 12: Photoemission spectra of ultrathin Cu lms on fcc-Co(100). The shadowed spectral
structures, with binding energy depending on the lm thickness, derive from the quantization of
the energy levels due to electron connement. They are observed up to 50 atomic layers, (90
Angstr om), thickness [16].
As has already been discussed in sect. 2.1, and in particular in Fig. 2, the strong reection at
the interfaces can lead to the appearance of quantum well states in the spacer, if the interfer-
ence condition leading to the localized or resonant states in satised. The occurrence of these
states has been observed for Co/Cu (001) [15]. Instead of a trilayer, a variable number of Cu
layers has been grown nearly perfectly layer by layer on fcc Co(001). Then the photoemission
spectra are measured as a function of energy for the q

-value q
1
= 0, corresponding to the long
C3.18 Peter H. Dederichs
period oscillations. Fig. 12 shows the photoemission spectra of such ultrathin Cu layers on fcc
Co(001). For comparison also the spectra of the Co(001) (below) and Cu(001) are shown. With
varying thickness up to about 50 ML Cu, the size-quantized peaks move with varying thickness
through the Fermi level. Up to 15 different quantum well levels are observed. Every 6 mono-
layers such a peak crosses the Fermi level, which is in good agreement with the long period of
5.9 ML in interlayer coupling. A measurement of the spin polarization of these states by inverse
photoemission shows, that they are to a high extent minority states, which is in line with the
relative large reection coefcient of Cu-minority electrons at the Co-layers. Fig. 13 shows the
binding energies of these quantum well states in the same thickness and energy region. As can
be seen, the experimental results [16] agree very well with the theoretical results [11].
Fig. 13: Binding energy of the quantum well states as a function of the Cu lm thickness.
Circles: experimental data [16]. Squares: theoretical data [11].
Appendices
A Change of Integrated Density of States N(E)
If one considers the Hamiltonian H = H
o
+ V with an additional potential V , the Green
functions G(E) and G
o
(E) as given by eq. (13) are related by a Dyson equation:
G(E) = G
o
(E) +G
o
(E) V G(E) (A1)
which can be proven by inserting H = H
o
+V in the identity
(E +i H) G(E) = 1 , (E +i H
o
) G = 1 +V G(E) (A2)
Interlayer Exchange Coupling C3.19
from which eq. (A1) follows by multiplication from left by G
o
. Formally (A1) can be solved
with respect to G(E) leading to
G(E) =
1
1 G
o
V
G
o
or G(E) = G
o
+G
o
V
1
1 G
o
V
G
o
(A3)
where in the last step eq. (A1) has been used. Inserting the last expression into eq. (12) we
obtain for the change of the DOS due to the potential V
n(E) =
2

Tr G
o
V
1
1 G
o
V
G
o
(A4)
Since for the trace operation the identity Tr AB = Tr BA is valid, we can combine the right G
o
in (A4) together with the left one. Then by using the equation
G
o
G
o
=

1
E +i H
o

2
=
d
dE
1
E +i H
o
=
dG
o
dE
(A5)
we obtain for n(E):
n(E) =
2

Tr
dG
o
dE
V
1
1 G
o
V
=
d
dE
2

Tr ln(1 G
o
V ) (A6)
from which eq. (14) for the integrated DOS follows by integration. In the case of two potentials
V = V
1
+V
2
the interactive part of the DOS is according to eq. (3) given by
n
int
(E) =
d
dE
2

Tr ln

(1 G
o
(V
1
+V
2
))
(1 G
o
V
1
)(1 G
o
V
2
)

(A7)
Here the argument of the logarithm can be simplied by using the identity
1 G
o
(V
1
+V
2
) = (1 G
o
V
1
)

1
1
1 G
o
V
1
G
o
V
1
G
o
V
2
1
1 G
o
V
2

(1 G
o
V
2
) (A8)
= (1 G
o
V
1
) {1 G
o
t
1
G
o
t
2
} (1 G
o
V
2
)
where the t-matrices t
1
and t
2
describe all scattering processes at the isolated layers 1 and 2,
e.g.
t
1
= V
1
1
1 G
o
V
1
(A9)
Therefore one obtains for the integrated DOS N
int
N
int
(E) =
2

Tr ln(1 G
o
t
1
G
o
t
2
) (A10)
B Frozen Potential Approximation for Magnetic Interactions
Here we show that the magnetic interaction energy, i.e. the total energy difference E
FM
E
AF
between the ferromagnetic and antiferromagnetic congurations of two magnetic layers or im-
purities, can be calculated in rst order from the single particle energies alone[17]. In density
C3.20 Peter H. Dederichs
functional theory, the total energy E{n
+
(r), n

(r)} is a unique functional of the spin densi-


ties n
+
(r) and n

(r) and obeys a variational principle, such that E{n


+
(r), n

(r)} is extremal
against small variations n
+
(r), n

(r) around the exact solutions. Furthermore, the total en-


ergy E = E
sp
+ E
dc
can be split up into single particle contributions E
sp
and double counting
terms E
dc
.
E
sp
=

E
F
dE(E E
F
) (n
+
(E) +n

(E)) (B1)
E
dc
=

dr(n
+
(r)V
+
eff
(r) +n

(r)V

eff
(r)) +W{n
+
(r), n

(r)} (B2)
W =

drn(r)V (r) +
e
2
2

drdr

n(r)n(r

)
|r r

|
+E
xc
{n
+
(r), n

(r)}
Here V

eff
(r) are the spin dependent Kohn-Sham potentials, which in a variational sense can be
replaced by trial potentials. W is the sum of the Coulomb energies and exchange correlation
energies E
xc
. V (r) is the nuclear potential and n(r) = n
+
(r) +n

(r) the charge density.


In the following we use an additional extremal property of the double counting energy E
dc
.
For xed trial potentials V

eff
the double counting energy E
dc
{n
+
(r), n

(r)} is insensitive to
variations in n

(r). In rst order one obtains


E
dc
|
V

eff
=

dr

W
n
+
(r)
V
+
eff
(r)

n
+
(r) +

W
n

(r)
V

eff
(r)

(r)

(B3)
Since the exact solution requires V

eff
= W/n

(r), the error in eq. (B3) is in fact of second


order. Therefore, E
dc
{n
+
, n

} can be calculated with some approximate spin densities n

(r)
which need not to be identical with those generated from the trial potentials V

eff
(r), which
strongly enhances the variational freedom.
For the magnetic interaction of two magnetic layers or impurities we take advantage of the
extremal properties of both E and E
dc
. First, we superimpose the changes V
1
eff
and V
2
eff
due to the isolated defects to obtain the trial potential V
eff
= V
0
eff
+ V
1
eff
+ V
2
eff
, where
V
0
eff
refers to the potential of the host crystal. Since for larger distances the interaction is weak,
the superposition is a very good rst order approximation. Furthermore V
1,2
eff
are assumed to
be localized on the magnetic layers (although this has not necessarily to be the case). Second,
for the spin densities n

(r) in E
dc
we assume that the charge density n(r) = n
+
(r) +n

(r) is
the same for both congurations. For instance, the average density of both congurations may
be taken or we may use the superposition of the changes n
1
(r) and n
2
(r) of the individual
layers, which is identical for both congurations. For this reason all troublesome Coulomb
terms in the double counting energy E
dc
are the same for both congurations and cancel each
other for the magnetic interaction energy E
FM
E
AF
. Finally, the genuine magnetic part E
m
dc
of the double counting energy depends at least quadratically on the magnetization |m(r)|. Also
here we assume that |m(r)| is the same for both congurations. This is, e.g., the case, if the
moments are well localized on the magnetic layers and if the interaction effects are small. Then
also E
m
dc
cancels in the total energy difference E
FM
E
AF
. Note that small deviations from
these approximations for n(r) and |m(r)| in E
dc
affect due to eq. (B3) the total energy only in
second order.
Thus we have shown, that the magnetic interaction energy is in rst order determined by the
single particle energies only. Therefore the expression (17) for the interlayer exchange coupling
E
IEC
is the better the weaker the interaction and becomes exact asymptotically. Note that
Interlayer Exchange Coupling C3.21
our proof applies only to the magnetic part of the interaction energies, but not to the full
interaction, for which Coulombic terms are usually very important.
References
[1] P. Gr unberg et al., Phys. Rev. Lett. 57, 2442 (1986)
[2] S.S. Parkin, N. More and K.P. Roche, Phys. Rev. Lett. 64, 2304 (1990)
[3] P. Bruno and C. Chappert, Phys. Rev. Lett. 64, 1602 (1991); P. Bruno, Phys. Rev. B52,
411 (1995)
[4] D.E. B urgler et al.; in Handbook of Magnetic Materials, vol. 13, ed. K.H.J. Buschow,
Elsevier (2001)
[5] P. Bruno, article on Theory of Interlayer Exchange Coupling in 30. IFF-Ferienkurs 1999,
Magnetische Schichtsysteme
[6] P. Lang et al., Phys. Rev. B53, 9092 (1996)
[7] N. Papanikolaou, R. Zeller and P.H. Dederichs, J. Phys.: Condens. Matter 11, 2799 (2002)
[8] M.R. Halse, Philos. Trans R. Soc. London A265, 507 (1996)
[9] M.D. Stiles, Phys. Rev. B48, 7238 (1993)
[10] J. Kudrnovsky et al., Phys. Rev. B50, 16105 (1994); Phys. Rev. B53, 5125 (1996); Phys.
Rev. B53; 15036 (1996)
[11] L. Nordstr om, P. Lang, R. Zeller and P.H. Dederichs, Phys. Rev. B50, 13058 (1994)
[12] K. Wildberger et al.; Phys. Rev. B58, 13721 (1998)
[13] D.E. B urgler, article on Zwischengitteraustauschkopplung, 30. IFF-Ferienkurs 1999,
Magnetische Schichtsysteme
[14] Q. Leng et al.; J. Magn. Magn. Mat. 126, 367 (1993)
[15] C. Carbone, article Photoemission from Magnetic Films, 30. IFF-Ferienkurs 1999,
Magnetische Schichtsysteme
[16] C. Carbone et al., Sol. State Commun. 100, 749 (1996)
[17] A. Oswald, R. Zeller, P.J. Braspenning and P.H. Dederichs, J. Phys. F: Met. Phys. 15, 193
(1985)

You might also like