See

discussions, stats, and author profiles for this publication at:

https://www.researchgate.net/publication/222467703

Slonczewski, J. C. Current-driven
excitations of magnetic multilayers. J.
Magn. Magn. Mater. 159, L1-L7
Article in Journal of Magnetism and Magnetic Materials June 1996
DOI: 10.1016/0304-8853(96)00062-5

CITATIONS

READS

3,009

669

1 author:
John Slonczewski
IBM
109 PUBLICATIONS 9,364 CITATIONS
SEE PROFILE

Available from: John Slonczewski

Retrieved on: 05 October 2016

~ H journalof
magnetism
and
magnetic
~ H materials

ELSEVIER

Journal of Magnetism and Magnetic Materials 159 (1996) L 1-L7

Letter to the Editor

Current-driven excitation of magnetic multilayers

J.C. Slonczewski

IBM Research Dit,ision, Thomas J. Watson Research Center, Box 216, Yorktown Heights, NY 10596, USA

Received 27 October 1995; revised 19 December 1995

Abstract

A new mechanism is proposed for exciting the magnetic state of a ferromagnet. Assuming ballistic conditions and using
WKB wave functions, we predict that a transfer of vectorial spin accompanies an electric current flowing perpendicular to
two parallel magnetic films connected by a normal metallic spacer. This spin transfer drives motions of the two
magnetization vectors within their instantaneously common plane. Consequent new mesoscopic precession and switching
phenomena with potential applications are predicted.
PACS: 75.50.Rr; 75.70.Cn

A magnetic multilayer (MML) is composed of

alternating ferromagnetic and paramagnetic sublayers whose thicknesses usually range between 1 and
l0 nm. The discovery in 1988 of g i a n t m a g n e t o r e s i s t a n c e (GMR) in such multilayers stimulates much
current research [1]. Although the initial reports dealt
with currents flowing in the layer planes (CIP), the
magnetoresistive phenomenon is known to be even
stronger for currents flowing perpendicular to the
plane (CPP) [2]. We predict here that the spinpolarized nature of such a perpendicular current generally creates a mutual transference of spin angular
momentum between the magnetic sublayers which is
manifested in their dynamic response. This response,
which occurs only for CPP geometry, we propose to
characterize as spin transfer. It can dominate the
Larmor response to the magnetic field induced by

* Fax: + 1-914-945-3291; email: [email protected].

the current when the magnetic sublayer thickness is

about 1 nm and the smaller of its other two dimensions is less than 10= to 10 3 rim. On this mesoscopic
scale, two new phenomena become possible: a steady
precession driven by a constant current, and alternatively a novel form of switching driven by a pulsed
current.
Other forms of current-driven magnetic response
without the use of any electromagnetically induced
magnetic field are already known. Reports of both
theory and experiments show how the exchange
effect of external current flowing through a ferromagnetic domain wall causes it to move [3]. Even
closer to the present subject is the magnetic response
to tunneling current in the case of the sandwich
structure f e r r o m a g n e t / i n s u l a t o r / f e r r o m a g n e t
( F / I / F ) predicted previously [4]. Unfortunately, theoretical relations indicated that the dissipation of
energy, and therefore temperature rise, needed to
produce more than barely observable spin-transfer
through a tunneling barrier is prohibitively large.

0304-8853/96/$15.00 Copyright 1996 Elsevier Science B.V. All rights reserved.

PH S0304-8853(96)00062-5

12

,/.C, Slo,cgewski / Journal of Magnetism and Magnetic Materials 159 (1996) L/ L7

However. the advent of multilayers incorporating

very thin paramagnetic metallic spacers, rather than
a barrier, places the realization of spin transfer in a
different light. In the first place, the metallic spacer
implies a low resistance and therefore low Ohmic
dissipation for a given current, to which spin-transfer
effects are proportional. Secondly, numerous experiments [5] and theories [6] show that the fundamental
interlayer exchange coupling of RKKY type diminishes in strength and varies in sign as spacer thickness increases. Indeed, there exist experimental spacers which are thick enough (e.g. 4 nm) for the
exchange coupling to be negligible even though spin
relaxation is too weak to significantly diminish the
GMR effect which relies on preservation of spin
direction during electron transit across the spacer.
Moreover, the same fact of long spin relaxation time
in magnetic multilayers is illustrated on an even
larger distance scale, an order of magnitude greater
than the circa 10 nm electron mean free path, by spin
injection experiments [7]. It follows, as we show
below, that interesting current-driven spin-transfer
effects are expected under laboratory conditions involving very small distance scales.
We begin with simple arguments to explain current-driven spin transfer and establish its physical
scale. We then sketch a detailed treatment and summarize its results. Finally, we predict two spin-transfer phenomena: steady magnetic precession driven
by a constant current and a novel form of magnetic
switching.
We consider the five metallic regions represented
schematically in Fig. 1. Layers A, B, and C are
paramagnetic, whilst F I and F2 are ferromagnetic.
The instantaneous macroscopic vectors hS~ and k S 2
forming the included angle 0 represent the respective total spin momenta per unit area of the ferromagnets. Now consider a flow of electrons moving
rightward through the sandwich. The works on spin
injection [7] show that if the thickness of spacer B is
less than the spin-diffusion length, usually at least
100 nm, then some degree of spin polarization along
the instantaneous axis parallel to the vector S~ of
local ferromagnetic polarization in FI will be present
in the electrons impinging on F2.
This leads us to consider a three-layer (B, F2, C
in Fig. 1) model in which an electron with initial
spin state along the direction Sj is incident from

Si~i

S2 ~,

EF=0J- . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

_K_2 ]

IA

V_

gl

F2 C

Fig. 1. Bottom: Coulomb plus locally diagonalized exchange

potential V_ versus position ( in a five-layer system composed
of paramagnets A, B, C, and ferromagnets FI and F2. The particle
flow is rightward the charged flow leftward (1~ > 0). Top: Vector
diagram of spin moments Si, 2 and their current-driven velocities
'~1.2 for magnets FI,2.

region B onto ferromagnet F2. Consider the moving

spin-quantization frame given by orthogonal unit
vectors 2.9g satisfying S 2 = S 2g and having the axis
.9 in the direction S 2 S 1. This frame generally
rotates adiabatically as determined by the rotating
vectors Si.2(t). Using ~ for the axis of spin quantization in this frame, the spin state of the electron
incident from region B is (cos 0/2, sin 0/2). The
Coulomb plus Stoner exchange potential of the magnet has the locally diagonal values V+(s~), where
is the position coordinate perpendicular to the multilayer plane unrelated to 2.9g. The subscripts _+ correspond to majority/minority-spin energy bands, respectively. Within the limitations of the WKB
parabolic-band approximation, we define the s~-components of the corresponding wave vectors k+ ({:).
Employing a unit system in which (h2/2) divided by
the electron mass is unity, these wave numbers are
given by the formula
k~= (U-k~-

V+) 'j=

(l)

where E is the constant energy of the electron and

kpis the magnitude of the conserved component kp
of the wave vector orthogonal to axis s~. We let
magnet F2 lie between se:= ~1 and se2, and place
= 0 at the center of region B. Thus we have the
equality V+ = V , and we assume k+ = k_ is real, in
paramagnetic regions well outside of magnet F2
(particularly ~ = 0 and s>> ~2). The stationary WKB
Hartree-Fock spinor wave function O = ( O + , O )

J.C. Sloncxewski/Jmlrnal of Magnetism and Magnetic Materials 159 (1996) LI L7

carrying unit particle flux for all ~> 0 may be

written

).
The method of spin currents and momentum conservation used below is widely used in deriving the
conventional exchange coupling energy written
- J S 1 S 2 [4,9]. Moreover, in that context it is shown
to be equivalent to other common methods [10].
The rightward particle flux ~ and the components q~ = (@~,b, @_) of rightward Pauli-spin ( = 2 s )
flux defined by

,;-j-f +o:--aT-,
{ d~+
q~+(sc)=q),.+i@,.=i[~-6

-6d~

(3)
d~O_
-)

(4)

satisfy general conditions of continuity. For the state

(2) the Pauli-spin flux within regions B and C approaches
@~=exp(i~['(k-k+)d~)sin0,

qb_=cos0

L3

is predicted by Eq. (6) to be ( A S ) =

(sin 0, 0, 0)/2. This is equivalent to the total absorption of the expectation value of the transverse (2)
component of spin of the electron incident on F2.
The oscillations about this mean echo predicted oscillations of conventional exchange coupling J with
magnet thickness [1 1].
In like spirit, we may consider that if the Stoner
splitting is so large as to eliminate minority-spin
electrons from the magnets (V > E), or in any case
if kp is sufficiently large, then k will be imaginary
according to Eq. (1). We assume that the thickness
of the film F2 is too great for appreciable tunneling
of minority-spin electrons. Therefore the component
reflects totally back into region B just as ~+
transmits totally into region C. Consequently, the
spin factor of the reflected wave ~ is (0, sin 0/2)
whilst that of the transmitted wave is (cos 0/2,0).
The off-diagonal nature of the matrix element (4)
dictates that scattering from F2 totally annihilates the
transverse spin. By spin conservation, it is totally
transferred to F2 without oscillations. But the transmitted electron flux is cos"`0/2, so the spin transfer
p e r transmitted electron is, instead of Eq. (6),
sin 0
ASe. - 2 c o s 2 0 / 2 ( 1,0, 0) = (tan 0 / 2 ) ( 1,0, 0).
k+-k_,

(5)

(7)

in the limit of slowly varying potential. These expressions describe the conical precession of one-electron spin about S 2 with the frequency governed by
the exchange splitting V - V + during its passage
through the magnet.
A crucial consideration is that by conservation of
angular momentum the magnet reacts to the passage
of one such electron by acquiring a change of classical momentum AS"` equal to the sum of the i n w a r d
spin fluxes from both sides of magnet F2:

The total bar on electron transmission when ~ = -~"`

causes the singularity at 0 = zr in this equation.
Briefly put, Eqs. (6) and (7) describe the complete
transfer of the transverse component of incident-electron spin to the scattering ferromagnet, except for
fluctuations due to geometrically-dependent wave
interferences. The mean of these fluctuations will be
small considering the usually broad distribution of
incident-electron directions. It follows that an electric current composed of preferentially polarized incident electrons generally causes a well-defined motion of the moment of the scattering ferromagnet.
Treatment of the total electron flow in the full
five-region system of Fig. 1 gives useful macroscopic expressions for current-driven spin transfer,
including dynamic reactions of both magnets F1 and
F2. The paramagnets A and C are considered semiinfinite. The interiors of all three paramagnets A, B,
and C have the parabolic energy-momentum expression E = k~ + kp - Q 2 where Q is the Fermi vector

AS"`. + iAS2, ~
= [qb+ (0) - q~+(~c)]/2

'(

=-~

1-exp

(f7
i

(k - k + ) d ~

))

sin0,

(6)

AS2. : = O.

The mean of the spin transfer averaged with

respect to direction of electron motion and therefore

J. ('. Slotm:ewski / Journal of Magnetism and Magnetic Materials 159 (1996) L1-L7

L4

and we take E = 0 to be the Fermi level. In our

model, V generally, varies with s~ only near the
interfaces, so we determine Q at the center {: = 0 of
region B. For the two ferromagnets, assumed to have
the sameband structures but generally different thicknesses, w e h a v e E = k e +k-+ - K + where K+ are
similarly the internal Fermi vectors for majority/
minority-spin electrons, with K+ > K .
To treat this system, we first solve the two-component Schroedinger equation in the WKB limit for a
general value of 8. Some pages of algebra are
needed to do this and evaluate the fluxes (3) and (4)
for these solutions. The common ballistic assumption
distributes the electric current in the momentum
space of paramagnet A by uniformly displacing the
forward right-hand half of the Fermi sphere a constant infinitesimal amount independently of spin. If,
for example, the smaller in-plane dimension of the
multilayer is d = 100 nm, the ballistic condition
a > > d on the mean free path a will often be well
satisfied at 80 K or less. At ambient temperatures,
where the ballistic assumption is likely poor, the
two-channel current model used in GMR theory [1,2]
generally introduces the polarization of the current
needed for spin transfer. Integration of the fluxes (3)
and (4) over occupied states provides the current
densities of charge /~ (leftward) and spin I =
(I,.,I.,.,1~) (rightward) as functions of s~, as in calculations of tunneling currents [4] and conventional
exchange J [4,9]. By momentum conservation, the
velocities of the adiabatically rotating magnet vectors are then given by

*-~1 = I(-oo)

- I(0),

~-~2= I(0)

- I(9c)

(8)

with the notation 2 = d x / d t . To minimize the number of parameters in the theory, we use the ballistic
assumption. In addition, we consistently average currents with respect to the phase factor e ik+W, where w
is the thickness of spacer B.
We define the normal energy Eno r = - k ~ available to a Fermi-level incident electron for the purpose of surmounting the potential rise within one of
the ferromagnets. The stationary states incident from
the left (paramagnet A in Fig. 1) fall into three
classes a, b, c according to the ranges o f kp defined
below. The fluxes @e and q~ are identical for states
belonging to a given one of these classes:
Class a: 0 <_ kp < K . Since E,o,. > [max V,~(~)]

=-K~
for or= _+, an electron fully transmits
through the system independently of o-. Therefore,
the aggregate incident flux J,, of Class a states
contributes to I e an amount /~a = eJ,(#: 0) and nothi n g ( I ~ = 0 ) to l a t a n y s ~.
Class b: K _ < k p < K + . Now we have - K + <
E ...... < - K 2 so both magnets transmit only electrons
of polarity cr = + along the local axis of quantization. Those with o-= - are totally reflected. Using
the WKB wave functions, one finds that the aggregate incident o-= + flux Jb contributes to l e a net
charge current
leb = eJb(4COS20/2)/( 3 + COS 8).

(9)

The corresponding spin currents are found to be

Ib(0) = Jb[(sin 0 ) / ( 3 + c o s 8)]2" + ( l b / 2 e ) e ,
/ b ( ~ ) = (/eb/Ze).

Class c: K < K + < k p .

(10)
Now we have Eno,.<

- K 2 < - K 2 and all incident electrons totally re-

flect, giving I~ = 0 and I t. = 0.

We combine the above results according to l~ =
I~ + l~b + l~c and I = I~ + I b + I~ and substitute the
latter into the second Eq. (8) to find

I~ = eJ~ + 2eJb(1 + cos 8 ) / ( 3 + cos 0 ) ,

S2 = Jb(sin 0 ) / ( 3 + cos 0)2".

(11)

One similarly finds an analogous relation for Sl.

Note from the Eqs. (11) that the ratio $2/I~ depends
on the ratio Ja/Jb. Since in practice Q is often
effectively nearly equal to K+ (see below), we
assume Q = K+ when evaluating Ja/Jb under the
ballistic assumption. Then this ratio becomes a function of one parameter, the polarizing factor P having
the conventional definition
n + - n_
e .

n++n_

K+- K
.

K.+K

(12)

Here n + are the majority/minority-state Fermi-level

spin densities in the magnets. Experimental 4 K
values for the ferromagnetic elements Fe, Co, Ni,
and Gd are P = 0.40, 0.35, 0.23, 0.14, respectively,
as obtained from tunneling between them and a
superconductor [ 12].
A result of this calculation for the five-layer
system is the relation

'~1.2 =

(Ieg/e)'~l.2

X (,~, X~2),

(13)

J.C. Slonczewski / Journal q[ Magnetism and Magnetic Materials 159 (1996) L l - L 7

where SI,2 are unit vectors si = S J S i . The scalar

function g ( > 0) is given by the formula

g=[-4+(I+p)3(3+~,'~2)/4p3/2]

-'

(14)

The absence of any film-thickness dependences in

Eqs. (13) and (14) results from our averaging the
fluxes with respect to the phase of the exponent of
the expression eik" occuring therein, involving phase
differences across the thickness w of the spacer B.
These oscillations are asymtotically negligible for
spacers thicker in order of magnitude than one atomic
layer. This averaging operation causes the conventional exchange coefficient J, determined by the
spin currents present in the absence of charge current, to vanish because such oscillations of J with
spacer thickness are symmetric about zero in meanfield calculations [6,9,10]. Only the current-driven
coupling represented by Eq. (9) remains after consistent application of this averaging procedure. The
practical consequence is that the predicted dynamics
will not be greatly diminished even when the presence of atomic steps in the crystalline interfaces
tends to nullify J This circumstance is analogous to
the persistence, both in theory and experiment, of
GMR even under such conditions that large spacer
thickness or rough interfaces make J neglible. Moreover, current-driven spin transfer is insensitive to
spacer thicknesses smaller than the mean free path
(10-30 nm at ambient temperature).
Noteworthy is the prediction of Eqs. (13) and (14)

e.1.5I
2 . 0

P=I

q~

o.~

1.o

.(.~ 0.5
o
0

0.2

0.4

0.6

0.8

1.0

O/rr
Fig. 2. Spin-transfer velocities IS].21 of ferromagnetic spin vectors

St, 2 versus included angle 0, based on Eqs. (9) and (10). The
units are It / e (1~ - current, e = electron charge). Equal polarization coefficients P of the m a g n e t s are assumed.

L5

that the five-layer dynamics are reversible with respect to sign of the electric current. It is the subsequent spin transfer back to the polarizing magnet by
reflected minority electrons discussed before Eq. (7)
which causes the polarizing magnet to react dynamically. Note also the equality [Sl[ = 182], even though
the magnets may differ in thickness. (For P < 1, it
may be special to our choice of identical magnet
parameters.) The magnitude of these velocity vectors
is plotted in Fig. 2.
Note that the spin transfer predicted by Eqs. (13)
and (14) for P < 1 generally vanishes at 0 = 0 and Jr
as in the case of Eq. (6) based on our three-layer
discussion. Its functional dependence on 0 tends to
that of Eq. (7) in the limit P ~ 1. However, its
magnitude is just half of that inferrable from Eq. (7)
because of the multiple minority-spin reflections
tending to confine electrons within the 'quantum
well' defined by the spacer, and thus share the spin
transfer among the magnets F1 and F2.
We leave to the reader the immediate geometric
proof of the relation IS~,xl= [ l ~ / 2 e l t a n ( O / 2 ) , holding for P = 1, from the conservation relation S~ + S~
= I ( - ~ c ) - I(~c) = ( / J 2 e ) ( ~ 1 - s2) and the assumption that the vectors St.2 lie within the plane
common to S 1 and So. This relation does not rest on
the WKB or other approximations, but is a logical
consequence of the 'perfect spin polarizer' concept.
The geometric relationships between these velocities dictated by the vector products in Eq. (13) are
illustrated by the vector diagram in Fig. 1. The
counter-intuitive tendency for the magnetic moments
to rotate in the same direction propeller fashion is
made consistent with angular-momentum conservation when the spin currents in regions A and C are
considered, viz. Eqs. (8). These motions of S~ and
S 2 within their common plane contrast with the
orthogonal precessions like SL = ]~JSI X S 2 dictated
by the conventional exchange Hamiltonian - J S t
S~. It is this new property of current-driven exchange
which implies the novel mesoscopic magneto-dynamics illustrated below. These motions due to spin
transfer can dominate over those due to precession
about the magnetic field H = I ~ d / 2 circulating about
the current when the smaller in-plane dimension d
satisfies the order-of-magnitude inequality d < 1 /xm
( 1 0 3 G / M s ) X ( 1 n m / w ) where w is the magnetic film thickness.

L6

J.C. S/(mczewski /Journal ( f Magnetism and Magnetic Materials 159 (1996) LI-L7

One defect of the WKB approximation is that it

allows only perfect transmission or perfect reflection
of the majority- or minority-spin component at the
normal/magnet interface, not allowing the coefficients to range between 0 and 1. A measure of the
severity of this defect is provided by first-principles
calculations of transmission probability, which we
define as T_(kp) with hkp now the transverse component of crystalline momentum, for Bloch electrons
crossing the interface from a paramagnet into the
majority-spin band of a ferromagnet [8]. In favorable
cases, Fig. 1 of this reference shows results satisfying T+ > 0.96 within the major central portion of the
space of kp describing the Fermi surface. Toward
the edge of this space, T+ decreases steeply down to
T+ < 0.04 within a narrow fringe. The mean transmissions over the Fermi surfaces in these favorable
cases have the values T + = 0.86 ( A g / F e 001), 0.84
( A u / F e 001), 0.79 ( C u / C o 001), 0.75 ( C u / C o 111),
0.66 ( C u / C o 110). These features qualitatively resemble our WKB picture with Q a little greater than
K+, for which T.(kp)= 1 in the main region k~ <
K+ and T + ( k p ) = 0 in the smaller annular fringe
region K+ < kp < Q. The case Cr/Fe(001), however, is rather unfavorable in this respect because
T+(kp) differs considerably from both 0 and 1 on
most of the Fermi surface [8].
Therefore, our use of the WKB approximation is
reasonable for certain compositions including the
favorable cases incorporating the noble-metal spacers Ag, Au, and Cu mentioned above. Our assumption of spherical Fermi surfaces satisfying Q = K+,
used to derive Eqs. (13) and (14), gives T + = I for
all kp. Therefore, it is a special case not far from the
facts indicated by the differences between the computed T+ values, quoted above, and 1. Also, from
the general smallness of transmission T into minority bands computed [8] for A g / F e (T = 0.16) and
for A u / F e (T = 0.17), we predict the peaking of
the spin-transfer rate at large 0 shown by the curves
in Fig. 2 with large values of P and reflected by the
limiting Eq. (7).
To obtain an idea of the remarkable new phenomena made possible by current-driven spin transfer,
consider an effective uniaxial anisotropy field Hu,
which includes the effect of magnet shape, and
Gilbert damping coefficient ce. The Landau-Lifshits

equation for such a single magnetic domain, modified to include the term (13), is

S" =g2 (TH. c "$2c- aS~ +

e-lleg.~l

NS2),

(is)
where "), is the gyromagnetic ratio and a fixed flame
is defined by orthogonal unit vectors a, b, c of
which c is the symmetry axis of anisotropy. For the
sake of illustration, we assume S~ is constant in time
because F1 is much thicker than F2 or has a much
larger damping, etc. In addition, we align S~ with
the anisotropy axis of S: (gl = c).
The motion of magnet F2 is found by substituting
the solution

g~=(sinO)(acoswt+bsinoot)+ccosO

(16)

into Eq. (15) under the assumption of weak damping.

Applying the condition 101 << I~ol yields the lowestorder relations o)= TH u cos 0 and

O= -(c~yHocosO+log/S~_e)sinO,
(17)
where the function g(O) ( > 0) is given by Eq. (14)
with sl "s2 = cos 0.
Qualitatively different behaviors of magnet F2
occur, depending on the sign of H u. A steady precession with constant polar angle 0. may occur for a
constant current under the condition H u < 0 making
_+c the hard directions of magnetization. Another
condition is that the quantity within parentheses in
Eq. (17) vanishes for a value of 0 different than 0
and ~-. Thus the frequency v is tuneable according
to 2Try= w=I~g/eo~S 2. Reasonable material parameters can provide v = 10 GHz using /~, = 10 ~ A
cm ~- which implies the feasibility of a monodomain sub-micron-scaled microwave-frequency
oscillator powered and tuned by a constant applied
current.
In the case H u > 0, _+c are easy directions for S,.
Under some conditions, time-dependent solutions of
Eq. (15) describe switching with 0(/) varying between orientations near the easy directions 0 = 0 and
7r. Switching away from 0 = 0 is subject to the
threshold condition I~ < - eS2ceyHu/g(O) obtained
by means of small-0 expansion. Switching away
from 0 = "z- is governed by I~ > eS2c~yHu/g(~) for
P < 1. For reasonable material parameter values,
repetitive switching by alternating 1 ns wide pulses
of applied current density on the order of 107 A

J.C. Sloncze~ ski / Journal qf Magnetism and Magnetic Materials 159 (1996) LI-L7

cm -2 is predicted. This effect may have applications

to high-speed, high-density digital storage and m e m ory.
O n e general advantage of devices based on current-driven spin transfer is the utter modesty of
current supply d e m a n d e d by the all-metal structure.
For a device diameter d, the Landauer-type ballistic
resistance [ 13] is of order R -- 2 7r -~fie- 2k v 2d - 2 T~h,
where T~.h is a characteristic overall transmission
coefficient for the layer system / F I / B / F 2 / .
Our
preferred value d = 100 n m gives R = 0 . 2 T ~ l 12.
Therefore the requirement that the lead resistance
plus the internal resistance of the current generator
exceed R is modest indeed with one exceptional
case (in need of clarification) that P is near 1 a n d 0
is near 7r. For then, T~h approaches 0 and R is
singular, totally blocking currents. The power dissipated in the multilayer for j = 107 A c m "- is only
0.2T~h 1 # W . W h e n the effective anisotropy field
opposing switching exceeds the order of 10 4 0 e ,
conservation of energy will require upward revision
of these estimates, but this subject is b e y o n d the
scope of the present article.

Acknowledgements
It is a pleasure to acknowledge helpful discussions with D. DiVincenzo, G a n g Xiao, S.S.P. Parkin,
M. Ketchen, P. Chaudhari, D.D. Stancil, J. K a u f m a n ,
R. Nesbet, S. Zhang, R. Landauer, S. Strunck, E.

L7

Marinero, D. Weller, D. Rugar, J. M a m i n , and B.A.

Jones. Grateful thanks are also due to M.D. Stiles for
discussions and access to his research results prior to
publication.

References
[l] P.M. Levy, in: Solid State Physics, Eds. H. Ehrenreich and
D. Turnbull, Vol. 47 (Academic Press, New York. 1994) p.
367: A. Fert, P. Griinberg, A. Barth~l~my, F. Petroff and W.
Zinn, J. Magn. Magn. Mater. 140-144 (1995) 1: P.M. Levy,
J. Magn. Magn. Mater. 140-144 (1995) 485.
[2] W.P. Pratt et al.. Phys. Rev, Lett. 66 (1991) 3060.
[3] g. Berger, J. Phys. Chem. Solids 35 (1974) 947: C.-Y. Hung
and L. Berger, J. Appl. Phys. 63 (1988) 4276; E. Salhi and L.
Berger, J. Appl. Phys. 76 (1994)4787.
[4] J.C. Slonczewski, Phys. Rev. B 39 (1989) 6995.
[5] B. Heinrich and J.F. Cochran, Adv. Phys. 42 (1991) 99.
[6] D.M. Edwards et al., Phys. Rev. Lett. 67 (1991) 493, 1476;
J.L. Fry et al., J. Appl. Phys. 69 (1991) 4780; P. Bruno, J.
Magn. Magn. Mater. 121 (1993) 248; M.D. Stiles, Phys.
Rev. B 48 (1993) 7238.
[7] M. Johnson, J. Magn. Magn. Mater. 140-144 (1995) 21.
[8] M.D. Stiles, to be published.
[9] K.B. Hathaway and J.R. Cullen, J. Magn. Magn. Mater.
104-107 (1992) 1840: J.C. Slonczewski. J. Magn. Magn.
Mater. 126 (1993) 374: D.M. Edwards, A.M. Robinson and
J. Mathon, J. Magn. Magn. Mater. 140 144 (1995) 517: J.C.
SIonczewski, J. Magn. Magn. Mater. 150 (1995) 13.
[10] J. d'Albuquerque e Castro, M.S. Ferreira and R.B. Muniz,
Phys. Rev. B 49 (1994) 16062.
[11] B. Bulka and J. Barna;;, J. Magn. Magn. Mater. 140-144
(1995) 491.
[12] R. Meservey and P.M. Tedrow, Phys. Rep. 238 (1994) 174.
[13] R.S. Sorbello, Phys. Rev. B 39 (1989) 4984.