Sie sind auf Seite 1von 13

PHYSICAL REVIEW B, VOLUME 65, 155322

Ballistic spin transport and spin interference in ferromagnetInAs2DESferromagnet devices


T. Matsuyama, C.-M. Hu, D. Grundler, G. Meier, and U. Merkt
Universitat Hamburg, Institut fur Angewandte Physik und Zentrum fur Mikrostrukturforschung, Jungiusstrae 11, D-20355 Hamburg,
Germany
Received 8 June 2001; published 5 April 2002
We examine the injection of electron spins from ferromagnets F into quasi-two-dimensional electron
systems 2DESs of semiconductors and employ the transfer-matrix formalism to obtain the carrier-density
dependence of the conductances across F/InAs2DES single as well as F/InAs2DES/F double junctions in
the ballistic limit. The Rashba spin-orbit interaction in the semiconductor and oblique modes in devices of
finite widths are taken into account. We distinguish between spin-valve and spin-transistor geometry, in which
the in-plane magnetization vectors in the ferromagnetic electrodes point perpendicular and parallel to the
current direction, respectively. In case of the spin-transistor geometry, we find optimum coupling with the
Rashba spin-precession state for injection straight along the channel. The distinct mismatch of majority and
minority spin subbands in the ferromagnet to the band structure of the semiconductor causes spin-dependent
scattering at the interface. This results in spin filtering. Within a Stoner model for ferromagnets spin filtering
is stronger for iron than for Permalloy parameters and can be enhanced by an additional elastic scattering
potential at the interface. In F/InAs2DES/F double junctions Fabry-Perot-type interferences are obtained. In
case of spin-valve geometry they are due to particle interference, whereas in spin-transistor geometry spin itself
is involved.
DOI: 10.1103/PhysRevB.65.155322

PACS numbers: 73.23.b, 72.25.Dc, 71.70.Ej, 73.40.Sx

I. INTRODUCTION

Electrical spin injection from a ferromagnet F into a


semiconductor Sm and spin transport in semiconductors are
topics of enormous and persistent interest stimulated by the
pioneering theoretical work of Datta and Das1 on the ballistic
spin field-effect transistor a decade ago. The efficiency of
spin injection across the interface between a metallic ferromagnet and a semiconductor has since then been discussed
controversially and has been predicted to be vanishingly
small if the bulk spin polarization of common metallic ferromagnets is regarded.2 However, very recently, theoretical
considerations worked out by different authors39 have suggested that the interface plays a major role for spin injection.
For ballistic transport across the interface the main consensus
is now that the efficiency is strongly enhanced due to the
band-structure mismatch between the ferromagnet and the
semiconductor.3,4,6 This causes spin filtering. Ideal spin filtering, i.e., a 100% spin-polarized current, is predicted for
certain combinations of epitaxially grown materials.3,4 Datta
and Das1 have proposed a new type of electronic device that
utilizes the spin of the electron in a field-effect transistor
with a quasi-one-dimensional electron system 1DES. In the
spin field-effect transistor, spin transport along the channel
and spin precession due to the spin-orbit coupling in an
asymmetric potential play the decisive roles. Polarized spins
are assumed to be injected and detected by ferromagnetic
source and drain contacts, respectively compare Fig. 1. The
modulation of the source-drain current comes about by controlled alignment of the spin direction in the semiconductor
with respect to the magnetization vector of the ferromagnetic
drain electrode. The underlying spin-orbit interaction is
known as the Rashba effect.10 It lifts the spin degeneracy in
the 1DES as a consequence of structure inversion asymmetry
of the macroscopic electrostatic potential perpendicular to
0163-1829/2002/6515/15532213/$20.00

the channel without any external magnetic field. The spin


precession and the concomitant current modulation are controlled via field effect by applying an external gate voltage
that alters this built-in asymmetric potential. The phase difference R 2m * L/ 2 of the spin precession1 between
source and drain contact is a function of the gate-voltagedependent Rashba spin-orbit coupling parameter , the electron effective mass m * , and the separation L between the
source and drain electrode. The model of Datta and Das1 has
considered a ferromagnetic single-mode injector with 100%
spin polarization and injection rate. Spin precession is obtained when the injected electron is a superposition of
spin-up and spin-down eigenstates in the semiconductor. For
the ferromagnet-semiconductor interface in a real device,
complete spin polarization is not the general case. It is ex-

FIG. 1. Sketch of the hybrid device with two ferromagnetic


contacts as spin injectors and spin detectors for a the spin-valve
and b the spin-transistor geometry. The black and white arrows
denote the electron spin in the semiconductor and the magnetization
configuration of the electrodes, respectively. The injected spin in the
spin-valve geometry is conserved in the semiconductor but goes
into a precession state in the spin-transistor geometry.

65 155322-1

2002 The American Physical Society

MATSUYAMA, HU, GRUNDLER, MEIER, AND MERKT

PHYSICAL REVIEW B 65 155322

pected that due to the band-structure mismatch at the interface and due to oblique modes in a sample of finite width,
the balanced transition into these two eigenstates is no more
guaranteed.
First experimental attempts with a true field-effect transistor incorporating a two-dimensional electron system 2DES
on p-type InAs with magnetically well-defined Permalloy
source and drain contacts11,12 have revealed a change of sign
in the gate-voltage-dependent resistance when the magnetization configuration is switched in the ballistic transport
regime.13,14 In the spin-valve geometry the resistance change
is independent of gate voltage. This behavior cannot be explained by existing models. In experiments on a quasiballistic mesoscopic 2DES where the separation between the ferromagnetic source and drain has been varied, Hu et al. have
recently observed a distinct dependence of the device resistance on the channel length that also exhibits a change in
sign.15
In this paper we consider spin injection by spin filtering
and study spin-dependent transport across F/InAs2DES
single as well as F/InAs2DES/F double junctions in the
presence of spin-orbit coupling in the ballistic regime. The
spin filtering stems from the role of the ferromagnetsemiconductor interface as spin-dependent scatterer due to
the abrupt change in the band structure.6,7 In the ballistic
regime the elastic as well as the inelastic scattering length of
the 2DES in the semiconductor exceeds the channel length.
This also holds for the spin coherence length even for InAs
with its relatively strong spin-orbit interaction.16 Hence, scattering events are reduced to the interfaces. In our calculation
we incorporate the latter as elastic scattering. In order to
handle the crossover from metallic to tunnel-junction
behavior,17 potentials as sources of additional elastic scattering at the interfaces are introduced. Their strength is characterized by a dimensionless Z parameter.18 The influence of
the carrier density and of the Z parameter on the spininjection rate and the conductance are studied utilizing realistic material parameters. We critically evaluate the physical
aspects of recently established models in the ballistic
limit.7,1921 Any kind of spin flip22 or spin relaxation23 due to
spin-spin scattering of magnetic impurities at the interfaces
and inside the channel is neglected.7,18,20,24 27 We distinguish
between spin-valve and spin-transistor geometry as sketched
in Figs. 1a and 1b. In the former case the magnetizations
and spins in the ferromagnets are aligned parallel to the
ferromagnet-semiconductor interface. Then the injected spin
direction is conserved in the 2DES. In the latter case the
magnetizations and spins point perpendicular to the interfaces. The injected spin enters a precession state and is not a
constant of motion in the channel. It is shown that the optimized coupling to the spin-precession state is given for normally injected modes, i.e., for modes injected perpendicular
to the interface. In F/InAs2DES/F double junctions FabryPerot-type interferences are obtained where spin itself is involved in the interference process.
The paper is organized as follows. In Sec. II we introduce
the Hamiltonian and the boundary conditions for the
ferromagnet-semiconductor interface. Numerical results for
the transmission coefficients and for the conductances of

single F/InAs2DES and double F/InAs2DES/F junctions


are presented in Sec. III and Sec. IV, respectively. We conclude with a summary in Sec. V.
II. BASIC ASSUMPTIONS

Our calculation is based on the model of Hu and


Matsuyama20 that unifies the picture of spin injection across
many types of single junctions: ferromagnet/semiconductor-1
ferromagnet/metal-,28 ferromagnet/superconductor-,29 dilute
-magnetic-semiconductor/semiconductor-,30 and ferromagnetic-semiconductor/semiconductor31 single junctions. The
theory holds for a degenerate electron system in a semiconductor and for an s-wave superconductor. Anisotropic band
structures can be covered as long as group velocities at the
Fermi energy can be defined for the appropriate directions of
the electron motion. Here we extend the model to 2DES and
double junctions and focus on ferromagnet-semiconductor
interfaces taking into account spin-orbit interaction in the
semiconductor.
A. Hamiltonian

In the one-band effective-mass approximation the Hamiltonian with exchange interaction H ex in the ferromagnet and
Rashba spin-orbit interaction H R in the semiconductor has
the form

1
2

V x,y,z
2 m * x

H ex x

H R x .

1
32,33

We have to consider a position-dependent effective mass


m * (x) and electrostatic potential V(x,y,z). The confinement
potential along the z axis in the inversion layer on p-type
InAs can be approximated by a triangular well V(z)eF s z
in which the surface-electric field F s is a linear function of
the electron density n s in the 2DES.34,35 The total potential in
F/InAs2DES/F junctions with potentials at the interfaces
(x0,L) is modeled by the expression
V x,y,z V 0 x xL V y eF s z
Zv F, f x xL ,

where we use the Heaviside function (x), a dimensionless Z


parameter,18,20 and the arithmetic mean v F, f of the Fermi
velocities of the two spin subbands in the ferromagnet F.
Along the x direction it has the shape of a potential barrier of
length L and of height V 0 . The origin of the energy scale lies
in the middle between the bottoms of the spin-split subbands
in the ferromagnetic regions. There the Fermi energy E f is
defined and stays fixed. Hence the height V 0 depends on the
Fermi energy inside the semiconductor. We consider only the
case of transmission; i.e., the potential barrier is always
lower than the Fermi energy. The potential V(y) is introduced to account for the finite width W of the channel. This
leads to a countable number of discrete transverse modes. So
we neither treat the pure 1D nor a quasi-1D regime but consider the full 2D characteristic of the electron gas with ob-

155322-2

BALLISTIC SPIN TRANSPORT AND SPIN . . .

PHYSICAL REVIEW B 65 155322

lique modes in the channel. We assume conservation of the


wave number k y , i.e., adiabatic mode coupling.
In the 3d transition metals and their alloys the d-band
electrons are mainly responsible for the magnetic
properties.36,37 However these electrons exhibit low mobilities. Electron transport is carried by hybridized s-band-like
itinerant d electrons at the Fermi level. The exchange interaction leads to an imbalance between spin-up and spin-down
densities of states. In this scenario the Hamiltonian in the
ferromagnet can be approximated by an isotropic effective
interaction36
H ex x h x m ,

with exchange interaction h(x) and the vector of the Pauli


spin matrices ( x , y , z ). The vector m denotes the
unity vector of the magnetization in the ferromagnetic electrodes and points in the direction of the minority spins. In
case of spin-valve and spin-transistor geometry we choose
the unit vectors mup uy and ux , respectively, to define up
magnetization. In this magnetization we describe the minority spin state as the spin state F . The spin-quantization
axis is chosen in the z direction and all spin states are expressed in the basis (1,0),(0,1) of the eigenstates with respect to the Pauli matrix z . 38
The Rashba Hamiltonian
i
H R x x x uz
2

has to be symmetrized to become a Hermitian operator20,21


because of the spatial variation of the Rashba parameter
(x). The Rashba parameter inside the semiconductor depends on the surface-electric field F s in the 2DES Refs. 10
and 39 and can be tuned via the electron density n s Refs.
35 and 40. In devices with comparatively large channel
widths, the contribution of the spin-orbit interaction due to
the lateral potential V(y) is neglected.41,42 The kinetic momentum
P x

x m * x

uz
i

m * (x)
h(x)
(x)
F s (x)

x0

0xL

xL

me
h
0
0

0.03m e
0
(n s )
F s (n s )

me
h
0
0

with the spinor wave function (x). The material parameters in each region are summarized in Table I. Nonparabolicity of the subband mass35,43,44 in InAs is ignored here.
B. Wave functions and eigenenergies

In the ferromagnetic contacts (x0, xL) we obtain


from the Hamiltonian in Eq. 1 the eigenfunctions

F r

e i(q x xq y y) q z z 1 .

The explicit form of the wave function q z (z) in the z direction is not important for our treatment. The eigenfunctions
are products of spatial envelope functions and spinors. The
orbital part of the wave function is characterized by the wave
numbers q
x , q y , and q z in the ferromagnet. The spin in the
ferromagnet is linearly polarized and independent of spatial
motion. The phase factor s in the upper spinor component
defines the spatial alignment of the spin. The lower spinor
component indicates the quantum number of the spin eigenstate. For spin-valve and spin-transistor geometry the value
of s is i and 1 and the spin of the state points in the minus
y and plus x directions, respectively. The negative sign in
case of spin-valve geometry is due to our definition mup
uy of the magnetization vector. The corresponding
eigenenergy is

differs from the canonical momentum because of the additional momentum operator in the Rashba Hamiltonian. This
has consequences for the interference properties of
F/InAs2DES/F double junctions as will be discussed in
Sec. IV. The kinetic momentum P(x) is related to the velocity operator vP/m * . As the Rashba Hamiltonian couples
the electron spin with the orbital motion, the projection of
the electron spin on any fixed quantization axis is not a good
quantum number simultaneously for all eigenspinors. Hence
the projection P x applied on the eigenfunctions of Eq. 1
has an eigenvalue only for motion parallel to the x direction.
From the continuity equation for the probability density, one
obtains the current
1
Re x P x x ,
j
m * x

TABLE I. Effective electron mass m * , exchange interaction h,


Rashba parameter , and surface electric field F s in the ferromagnetic (x0, xL) and semiconductor (0xL) regions.

E F E q z

2
2 q

2m e

h,

with the energy E q z of the wave function in the z direction


and wave vectors q
in the plane of the device. In the ferromagnet the parabolic energy dispersions for the spin subbands are shifted along the energy axis by the exchange energy h. The sign of the exchange energy reverses when the
magnetization of the ferromagnet is switched from the up to
the down direction (mdown mup ). In the free electron
model, the group velocities of the spin subbands at a given
energy,
v F E

q

me

2
EE q z h ,
me

depend on the direction of the magnetization.


In the semiconductor (0xL) we obtain from the
Hamiltonian in Eq. 1 the eigenfunctions

155322-3

MATSUYAMA, HU, GRUNDLER, MEIER, AND MERKT

PHYSICAL REVIEW B 65 155322

with the experimental results obtained for inversion layers on


p-type InAs.35 Thus the Rashba parameter increases linearly
with the density n s . Note that in the 2D case the group
velocities

v Sm
E

2 EV 0 E i0
m*

13

of both branches are equal. Using the relations E f V 0


E i0 (k f ) 2 /(2m * ) and k f (2 n s ) 1/2 one obtains the
Fermi velocity

v Sm
m*
FIG. 2. Energy dispersion of a Rashba spin-split 2D subband.
Effective magnetic field vectors BR gray arrows and eigenspins
black arrows are indicated for four representative momentum
states.

Sm
r

e i(k x xk y y) i z


e i
1

10

with 2D subband wave functions i (z). The orbital part of


the wave function is characterized by the wave numbers k
x
and k y and the index i of the 2D subband. We consider only
the lowest subband i0 in the following.24 Contrary to the
situation in the ferromagnet, the spinor is a function of the
spatial motion and the spatial alignment of the spin does
depend on the wave vector k
via the phase factor
e i

k y ik
x
k

The Rashba spin-orbit interaction causes a pseudomagnetic


field BR that is always directed perpendicular to both the
surface-electric field F s uz and the wave vector k as depicted
in Fig. 2. The spins of the eigenstates can only align either
parallel or antiparallel to the pseudomagnetic field and are
circularly polarized. The angle in Eq. 11 corresponds to
the angle between the spin orientation and the k x axis. The
energy dispersion splits into two branches

V 0 E i0
E Sm

2
2 k

2m *

k
.

12

The Rashba splitting 2 k is isotropic in momentum space


and linear in the absolute value of the momentum k . In a
triangular well34 one obtains for the subband edges E i0
F s2/3 .

(k
From the relation E Sm
f )E Sm (k f ) at the Fermi energy E f one can derive the Rashba parameter
1
( 2 /2m * )(k
f k f ). In our calculation we determine it
1/2
from the Fermi wave numbers k
f 2 (1 )n s .
Hence, the strength of the spin-orbit interaction is tuned by
the parameter n/n s that denotes the asymmetry of the
spin-subband occupation in the 2DES.7 We have chosen a
value 0.1 for all of our present calculations in accordance

2 n s

14

which retains the proportionality to the square root of the


electron density due to the relation k f n s1/2 .
C. Boundary conditions

The boundary conditions are derived from the integration


of the Hamiltonian in Eq. 1 across the interface at xx 0 :

L x
0 R x 0 ,

P x R x
m R* x

xx
0

P x L x
m L* x

15

2iv F, f Z x 0 .
xx
0

16

Here we have the average v F, f for the Fermi velocities of the


two spin subbands in the ferromagnet and the operator

11

m *

P x

m*

y .
x

The indices L and R stand for the left- and right-hand sides
of the interface, respectively. One can see from Eq. 16 that
the group velocity is important for the matching condition of
the wave functions. The same conditions are used in the
work of Hu and Matsuyama.20 For Z0, our conditions
agree with the ones of Zulicke and Schroll21 and with the
ones of Molenkamp et al.19 In the latter work, the boundary
conditions are determined by the requirement of the continuity of the flux of the wave function at the interface. The
symmetrized Rashba Hamiltonian in Eq. 4 already takes
this into account and directly leads to Eq. 16 without any
additional assumptions. Note that this equation fulfills the
conservation of the probability current at the interface for
oblique modes too. In particular, our boundary conditions
guarantee the conservation T , T , R , R , 1 of
the probability coefficients for transmission and reflection
across the ferromagnet-semiconductor interface in the presence of spin-orbit interaction. The first and second indexes
denote the spin state of the incident and outgoing electrons,
respectively. Equation 16 coincides with the well-known
boundary condition18,45 employing just the canonical momentum operator when the second term of the kinetic momentum operator P x can be neglected.6,7

155322-4

BALLISTIC SPIN TRANSPORT AND SPIN . . .

PHYSICAL REVIEW B 65 155322

With the boundary conditions in Eqs. 15 and 16 one


can derive the transfer matrices M L and M R for the interfaces at the left and right ferromagnetic contacts, respectively, in the usual way.46 49 In our case two propagating
modes for each direction have to be considered so that we
obtain 44 matrices.
The subband wave function i0 (z) varies on the length
scale k z1 determined by its inverse width. Accordingly, the
overlap q z i0 has a maximum at wave number q z k z
and vanishes rapidly for higher ones.24 This means that only
wave functions q z (z) are relevant whose wavelengths correspond to nonvanishing Fourier components of the subband
function i0 (z). The shape of the confining potential in the
2DES is less important.
The total transfer matrices M for single and double interfaces with an intervening region are composed of M L (M R )
and M L GM R , respectively, with the propagator matrix

e ik x L

ik x L

e ik x L

0
e

ik x L

P
17

The propagator matrix takes into account the phase difference between the two boundaries. The transmission and reflection amplitudes t , and r , , respectively, are obtained from the components M i j of the total transfer matrix:

t ,

t ,

t ,

t ,

M 11

M 12

M 21

M 22

18

and

r ,

r ,

r ,

r ,

M 31

M 32

M 41

M 42

M 11

M 12

M 21

M 22

1
2v F, f k

sin F
. 19

20

v Sm k x i

k ,
y

21

which will prove convenient as input parameters for the determination of the probability coefficients. They characterize
the ferromagnet and semiconductor, respectively. Both parameters are normalized with twice the value of the mean
Fermi velocity 2v F, f in the ferromagnet. The real parts of
these parameters are the x components of the normalized
group velocities v x /2v F, f in the ferromagnet and semiconductor, respectively. The approximation in Eq. 20 is justified in view of the law of reflectance which results in relatively large angles Sm inside the semiconductor. As a
consequence of wave number conservation q y k y one obtains the relation

The probability coefficients T , and R , are defined by


the ratio j out,x / j in,x of the probability currents between the
outgoing and incident electrons in the x direction. From Eq.

) t , 2 and
6 we obtain the coefficients T , ( v x / v F,x

2
R , ( v F,x / v F,x ) r , . As an example, in Fig. 3 one out
of all possible scattering processes at a single interface is
shown for an electron normally incident from the spin
subband of the ferromagnet.
From now on the sign index 1 denotes the spin
eigenstate independent of its position as super- or subscript.
A bar above the index inverts the sign. As noted above, the
phase factors si and 1 describe spin-valve and spintransistor geometry, respectively. The index m or clarifies the magnetization in the electrode. We introduce two
parameters
v F
1 q x

,
F
2v F, f m e
2v F, f

FIG. 3. Schematic dispersions with transmission and reflection


amplitudes for normal incidence (k y 0). The electron starts with
wave vector q from the spin subband in the ferromagnet. The
energy offset V 0 E i0 between the ferromagnet and the semiconductor is ignored in this sketch. The arrows point in the direction of
the group velocity.

kf
kf
sin Sm 1.
qf
qf

22

Here, we have the angle of incidence F in the ferromagnet


and the angle of refraction Sm in the semiconductor as
shown in Fig. 4. Because the Fermi wave vector k f in the
semiconductor is much smaller than the one in the ferromagnet q f , the angle of incidence in the ferromagnet has to be
virtually zero for transmission.
We emphasize that for the evaluation of the transfer matrix the group velocities of all spin subbands are required.
Therefore, in the case of the ferromagnet, instead of the parametrization with the exchange interaction h(x) the group
velocities are more convenient as input parameters. We consider only the case of low temperatures (k B TE f V 0 ) and
therefore we assume that the group velocities in the ferromagnet are virtually constant around the Fermi energy.
Therefore they can be represented by the Fermi velocities of
majority and minority spin subbands that are listed in Table
II. In the case of switching the magnetization between up and
down the Fermi velocities of the spin states are swapped.
In order to simplify the discussion we consider only up magnetization in the following discussions unless both magnetization directions are required.

155322-5

MATSUYAMA, HU, GRUNDLER, MEIER, AND MERKT

PHYSICAL REVIEW B 65 155322


TABLE II. Fermi velocities of majority and minority spins. Data
for Permalloy Ni80Fe20 are from photoemission spectroscopy and
refer to the 011 direction Ref. 37. For iron the Fermi velocities
are derived from calculated band structures Ref. 50 and are averaged over the three crystal directions.

v f ,ma j (106 m s1 )
v f ,min (106 m s1 )

FIG. 4. Schematic presentation of spin injection from the spin


subband of a ferromagnet into a semiconductor. Black arrows indicate spins. Gray arrows point into the direction of the group velocities. The values of the transmission and the reflection coefficients
are determined by the overlap of the spin states as well as by the
mismatch in the group velocities of the incident and the final state.
III. RESULTS FOR A SINGLE FInAs2DES INTERFACE

In this section we evaluate the spin injection across a


single interface. We begin with the analytical expression for
the transmission coefficient for incidence from the ferromagnetic side. Then we present numerical results for the prob-

Fe

Ni80Fe20

0.55
0.28

0.22
0.18

ability coefficients of Fe/InAs2DES and InAs2DES/Fe


junctions for spin-valve as well as for spin-transistor geometry. In the last part of this section we discuss the dependence
of the spin conductance and the spin-injection rate on the
carrier density as well as on the strength of the Z parameter.
We compare results for iron with those for Permalloy.
From the total transfer matrix we obtain with Eq. 18 the
transmission amplitude
4 sF
s
t ,
s,Z

F Re P iZ

e i

i Im P
23

for injection from the ferromagnetic side. The denominator

* iZ F P
* iZ F P
* iZ F P
* iZ e i F P iZ F P iZ
s,Z e i F P
* e i( ) s 1 P
* Ps
F P iZ F P iZ F F P P

depends on the dimensionless parameters defined in Eqs.


20 and 21. The transmission amplitude shows that the
overlap of the spinors between incident and final states is
involved in the transmission. As can be seen in Fig. 4 the
spins of the two eigenstates in the semiconductor in general
are not collinear due to the imposed conservation of the

wave number k
y k y . For weak Rashba interaction or a
small component k y the real part Re( P ) becomes larger
than the imaginary part in Eq. 23. In this case the amplitude
in Eq. 23 is determined dominantly by the overlaps of the
spinor states (s, ) and (e i , ). In order to simplify the
physical picture we can then pretend in the interpretation of
Eq. 23 as if the direct spin states in ferromagnet (s, ) and
in semiconductor (e i , ) themselves are responsible for
the overlap.
In case of normal incidence (k y 0) we obtain from Eq.
23 the transmission coefficient
s

T ,

F
P

1
1

P
F

Z
F P

4 , ,

si,

2,

s1,
25

24

for the spin-valve (si) and the spin-transistor (s1) geometry. In the spin-valve geometry there is no crossing between the spin states. Only the ratio P /F v Sm / v F and
the product F P ( v F v Sm )/(2v F, f ) 2 are relevant for the
transmission probability. Both depend on the group velocities
v F and v Sm and are independent of the spin subband in
the semiconductor. The transmission of Eq. 25 corresponds
to the expression T1/(1Z 2e f f ) with effective Z parameter
Z e f f , which is introduced in many works.7,25 We distinguish
two regimes: namely, the highly transmissive and the tunneling cases. The denominator of Eq. 25 clearly shows that in
case of weak potential barriers (Z 2 F P ) the term
F / P (1 P /F ) 2 becomes dominant and the matching
of the group velocities in the ferromagnet and semiconductor
plays the essential role for the transmission probability. Its
maximum is obtained for equal velocities ( v F v Sm ). In the
tunneling limit (Z 2 F P ) the second term of the denominator is dominant and the transmission probability changes
s
into the relation T , F P . Then the numbers of states
in the ferromagnet (F q) and in the semiconductor ( P
k) determine the transmission probability as is well known

155322-6

BALLISTIC SPIN TRANSPORT AND SPIN . . .

PHYSICAL REVIEW B 65 155322

for tunneling devices. The result in Eq. 25 is also valid for


Heusler alloys,51 in which only the majority spin subband is
populated and the mean Fermi velocity v F, f has to be substituted by the Fermi velocity of the only occupied spin subband. The expression of Eq. 25 for normal injection is directly applicable to the 1DES.
For oblique modes the transmission coefficient can be approximated in the tunneling limit by
s
T ,

2F Re P
2
Z 1cos

e i

, 26

where the overlap of the spinors becomes important. The


angles are taken between the spin and the k x axis. Contrary to normal injection the wave vectors as well as the
corresponding spins of the two spin subbands in the semiconductor are no longer collinear. Therefore the parameter
P becomes a complex number and depends on the branch
index . In this approximation we obtain for the product
F Re( P ) v F v Sm cos( )/(2v F, f ) 2 . Assuming strong
spin-orbit interaction and oblique modes with large wave
numbers k y , the imaginary part Im( P ) instead of the Z
parameter appears in Eq. 26. Due to the antagonistic relation between the real Re( P ) and the imaginary part
Im( P ) the transmission probability decreases more
strongly than with the second power of the wave number k y
according to Eq. 21.
We now discuss numerical results for the carrier and angle
dependence of the transmission and reflection coefficients.
We first consider spin injection from the ferromagnet into the
semiconductor restricting to incidence from the spin subband and to the case Z0. In case of spin-valve geometry
the normally injected electrons transmit only into the same
spin state see Fig. 5 because the spins in the ferromagnet
and semiconductor are parallel for both eigenstates see Fig.
4. The density dependence of the transmission T , is determined by the matching condition of the group velocities
between the spin subbands in the ferromagnet and semiconductor. The maximum T , 1 is observed when the condition v
F, f v Sm, f is fulfilled. For oblique modes the overlap
of the incident spin with the same spin state in the semiconductor decreases, simultaneously it increases with the
complementary state. The angle dependence of the transmission T , is symmetric with respect to the sign of the angle.
The density and angle dependence for an electron incident
from the spin subband is almost the same with the exception that the maximum of the transmission T , is located at
the slightly different density that satisfies the condition v
F, f
v Sm, f .
In case of the spin-transistor geometry the spin of the
normally incident electron is spatially orthogonal to the spins
of both eigenstates in the semiconductor see Fig. 4. Therefore the spinor of the incident electron has the same overlap
to both eigenstates in the semiconductor. As the group velocities of both eigenstates in the semiconductor are the
same, one obtains identical transmissions T , and T , in
case of normal incidence as evident in Fig. 6. The maximum
transmission T , T , 0.5 is reached when the condi-

FIG. 5. Transmission and reflection coefficients for spin injection from the spin state of a ferromagnetic contact across a Fe/
InAs2DES single interface in spin-valve SV geometry. Results
for three different angles of oblique modes defined in the semiconductor ( 0,45) are plotted for Z0 vs electron density n s in
the 2DES. The inset in the R , 0 panel illustrates the wave
vectors dotted arrows and the corresponding spins solid arrows
of an incident black as well as an outgoing gray electron.

tion v
F, f v Sm, f is fulfilled. But for oblique modes the overlaps to the two spin eigenstates in the semiconductor differ.
This leads to an antagonistic behavior in the angle dependence of the transmissions T , and T , . Also, in the
spin-transistor geometry, the transmission and reflection co-

FIG. 6. Transmission and reflection coefficients for spin injection from the spin state of a ferromagnetic contact across a Fe/
InAs2DES single interface in spin-transistor ST geometry. Results for three different angles of oblique modes defined in the
semiconductor ( 0,45) are plotted for Z0 vs the electron
density n s in the 2DES. The inset in the R , 0 panel illustrates
the wave vectors dotted arrows and the corresponding spins solid
arrows of an incident black as well as an outgoing gray electron.

155322-7

MATSUYAMA, HU, GRUNDLER, MEIER, AND MERKT

PHYSICAL REVIEW B 65 155322

FIG. 7. Transmission and reflection coefficients for spin injection from the spin state of a semiconductor across an
InAs2DES/Fe single interface in spin-valve SV geometry. Results for three different angles of oblique modes defined in the
semiconductor ( 0,45) are plotted for Z0 vs electron density n s in the 2DES. The insets in the T , and R , panels illustrate the wave vectors dotted arrows and the corresponding spins
solid arrows of the incident black as well as outgoing and reflected gray electrons.

efficients for incidence from the spin state are similar to


the ones from the state. Only the positions of their
maxima appear at slightly different electron densities.
The spin injection into a semiconductor with spin-orbit
interaction leads to a spatial spin precession when the state of

the injected electron in the semiconductor Sm t , Sm

t , Sm is composed of both eigenstates. The expectation


value of the spin in the z direction,

,
2

G s,m

e 2 Wk
f
h 2

/2

/2

s,m
T s,m
, T , cos d ,

28

Sm z Sm 2 t *, t , cos

This applies to spin-valve and spin-transistor geometry as


well as to any mode angle. If the incident angle in the ferromagnet exceeds the critical value defined by the condition
q y k , the electron is totally reflected back into its spin
subband in the ferromagnet. This means that in the 2DES the
effective number of modes in and states at the Fermi
energy is the same and given by the smaller one: namely,
Wk
f /.
Next we consider the reverse case and let the spins inject
from the semiconductor side into the ferromagnet. There are
two differences from the situation discussed above. The spin
is flipped when the momentum alters its sign. In order to
attain the maximum overlap of the spin states in the semiconductor the electron is reflected into the other spin subband as can be seen from the reflection coefficients R , in
Figs. 7 and 8. This branch crossing is a consequence of time
reversal invariance of the Hamiltonian in Eq. 1 inside the
semiconductor. Branch crossing becomes important for
F/InAs2DES/F double junctions and will be discussed in
Sec. IV in more detail. The second difference is that for the
spin-transistor geometry the density dependence of the transmission coefficients T , and T , for normal modes are no
longer equal. This is caused by the different group velocities
of the spin subbands in the ferromagnet.
Within the linear response we calculate the conductance
of a single junction for low temperatures (k B TE f V 0 ),

cos k k r
2
27

is an oscillatory function of coordinate r and difference k


k of the wave vectors. The phase-angle differences
and are determined by the phases
of the transmission amplitudes t , and by the phases
of the spinors as defined in Eq. 11, respectively. The
amplitude t *, t , reaches its maximum only when the
amplitudes t , and t , are identical and both corresponding transmissions T0.5. Accordingly, only the normal injection mode leads to the optimum spin precession state. Precession is deteriorated when the angle is increased for
oblique modes. Nevertheless, the oblique modes cause a
background in the conductance.
In case of spin injection from the ferromagnet into the
semiconductor the incident electron is only reflected into the
same spin subband (R , 0; see Figs. 5 and 6 due to the
overlap condition of the spin states within the ferromagnet.

FIG. 8. Transmission and reflection coefficients for spin injection from the spin state of a semiconductor across an
InAs2DES/Fe single interface in spin-transistor ST geometry.
Results for three different angles of oblique modes defined in the
semiconductor ( 0,45) are plotted for Z0 versus electron
density n s in the 2DES. The inset in the R , panel illustrates the
wave vector dotted arrows and the corresponding spins solid arrows of the incident black as well as the outgoing gray electron.

155322-8

BALLISTIC SPIN TRANSPORT AND SPIN . . .

PHYSICAL REVIEW B 65 155322

FIG. 10. Transmission coefficients of normally incident modes


(k y 0) across a Fe/InAs2DES/Fe double junction for the spinvalve a,b and the spin-transistor cf geometry. The curves
are calculated for a channel length L150 nm and width W
1 m and for Z0. Magnetization configurations of the Fe electrodes are indicated by arrows.

FIG. 9. Spin conductance G and spin-injection rate of a single


F/InAs2DES junction for Fe and Permalloy Py electrodes. The
dependence on carrier density in a is calculated for Z0, the
dependence on the strength of the Z parameter in b for n s
1012 cm2 .

from the angle-dependent transmissions for up (m) and


down (m) magnetization.52 Integration instead of discrete summation virtually does not change the result as there
are a high number of transverse modes involved in wide
junctions. The integration of both transmission coefficients
can be carried out by one of the angles say, because
the products dk y k cos d are equal for both spin subbands 1 in the semiconductor due to equal wave num
bers k
y k y .
The carrier dependence of the conductance G and of the
spin-injection rate (G G )/(G G ) is shown for
iron and Permalloy electrodes in Fig. 9a. All results are
identical for spin-valve and spin-transistor geometry. Switching the magnetization of the electrodes interchanges the conductances G and G and alters the sign of the injection rate

. Note that the total conductance G G does not


change. Although the transmission probabilities decrease
with electron density for higher values, the conductances and
the absolute values of the injection rate increase monotonously with density. This reflects the dominant contribution
of the mode number. Which of the two spins is more efficiently injected depends on the magnetization of the electrode. Our numerical calculations show that the influence of
oblique modes decreases the spin-injection rate by about
10% compared to the 1D case.
Comparison between iron and Permalloy shows that spin
filtering at a ferromagnet-semiconductor single junction is
stronger for iron. In our model the band structure is assumed
to be isotropic and is characterized by the Fermi velocities
v F, f of the two spin subbands. The larger their difference, the
more the transmission coefficients T , and T , differ
and the more the spin filtering increases. This effect could
also be enhanced when a half-metallic ferromagnet or an
anisotropic band structure for example, cobalt in 011
direction37 is used. In these types of materials the magnetic splitting is so large that the minority spin band is above
the Fermi energy and one can obtain 100% spin polarization.
Spin filtering also increases with the strength of the Z
parameter for both iron and Permalloy as shown in Fig. 9b.
At high Z parameters the matching condition of the Fermi
velocities between the ferromagnet and semiconductor plays
no longer a decisive role. Instead, the number of states of the
spin subbands in the ferromagnet becomes important for the
transmission coefficients.
IV. FInAs2DESF DOUBLE JUNCTIONS

In this section we discuss the transport through


F/InAs2DES/F double junctions for spin-valve as well as
for spin-transistor geometry. Numerical results are presented
for iron electrodes.
As an example, in Fig. 10 the carrier-density dependence
of the transmission coefficients across a Fe/InAs2DES/Fe
double junction is shown for normally incident modes with-

155322-9

MATSUYAMA, HU, GRUNDLER, MEIER, AND MERKT

PHYSICAL REVIEW B 65 155322

out additional interface scattering (Z0). The two left and


four right panels refer to spin-valve and spin-transistor geometry, respectively. Results for parallel and antiparallel
magnetization configurations are presented in the respective
upper and lower rows. There are always oscillations caused
by Fabry-Perot-like interferences between the two boundaries. These oscillations reflect transmission resonances that
originate from the destructive interference of electron waves
reflected from the first and second boundaries.
In spin-valve geometry there is no crossing between the
spin subbands of the two ferromagnets (T , 0). The spin
is conserved and the transmission problem is the one of two
independent electron systems with spin and states.
When we neglect the difference of the effective masses in the
ferromagnets and in the semiconductor for a moment in order to simplify the physical picture, the transmission probabilities
T

4q L q R
q L q R 2

1
1

q L2 k 2 q R2
k 2 q L q R 2

29

sin 2kL
2

can be easily calculated for each spin system. Here we have


the wave numbers q L , q R , and k of the electrons in the left
and right ferromagnets and in the semiconductor, respectively. We begin the discussion of Eq. 29 with the parallel
magnetization configuration of the electrodes. In the simplified picture both spin systems see symmetric potential barriers with different heights (q L q R and q L q R ). The principal course of the transmission shows the same density
dependence as that of a single F/InAs2DES junction and
the amplitude of the oscillations grows with carrier density in
the regime n s 0.41012 cm2 . This behavior reflects the
mismatch of the Fermi velocities at both interfaces that is
stronger for states than for states in the parallel configuration. The overall behavior is in principle described by
the prefactor of the sinusoidal function in the denominator of
Eq. 29 which is positive for all wave numbers k. As the
factor 4q L q R /(q L q R ) 2 1, the resonance peaks reach values T1 in this configuration.
In case of antiparallel electrode magnetizations we have
the relations q L q R and q L q R , i.e., different wave numbers in the left and right electrodes. Hence the two spin systems in the two ferromagnets see equivalent but asymmetric
potential barriers which nevertheless lead to identical transmission coefficients. Only when the value of the wave number k in the semiconductor is in between the ones in the
ferromagnets q L and q R does the prefactor of the sinusoidal
function become negative and the denominator of Eq. 29
become smaller than unity. Otherwise all resonances peaks
have the same height but the factor 4q L q R /(q L q R ) 2 1.
Now we turn our attention to the resonance positions.
Independent of the magnetization configuration they are determined by the condition kLn /2. As discussed in Sec. III
by means of Fig. 7, because of spin conservation an electron
incident from a spin state in the semiconductor is reflected
from its k state into a k state at the right boundary and
the other way around from a k into a k state at the left
boundary. Nevertheless, only one series of resonance peaks

is obtained despite the two spin systems in the ferromagnetic


electrodes and their magnetization configuration. The reason
is that the kinetic momentum k m * v Sm / has to be considered for the resonance condition and the group velocity in
the semiconductor is identical for both spin branches. When
the Z parameter of the interfaces is increased, the peak positions of the resonances shift as a function of electron density
without changing their periodicity. At the same time their
amplitudes become smaller. As a summary for the spin-valve
geometry we like to state that the electron transport across a
F/InAs2DES/F double junction is carried by two independent spin systems. The transmission probabilities are determined dominantly by the matching conditions of the bandstructure parameters and that separately for each of the two
spin systems.
In the spin-transistor geometry both spin states intermix in
the semiconductor region leading to spin precession.1 Accordingly, transmission into the reverse spin state of the ferromagnetic drain contact becomes possible (T , 0). The
transmission probability of the modes depends on the alignment of the spin at the end of the channel relative to that of
the detector. The evolution of the precession angle is determined by the Rashba parameter which is a function of the
electron density in our system. The precession manifests itself as a modulating background in the density dependence
of the transmission probability. Whereas in spin-valve geometry the Fabry-Perot-like resonances of the transmission
probability T , are caused by particle interference, the
large-scale variation in the spin-transistor geometry originates from the spin interference. Of course, the particle resonances are also present in Figs. 10c10f.
The total conductance of F/InAs2DES/F devices for low
temperatures (k B TE f V 0 ),
G s,m k f

e2
h

k ,
y

s,m

T , k y ,

30

is calculated within the Landauer-Buttiker formalism taking


s,m
into account the transmissions T , of the distinct spinvalve and spin-transistor geometries (si,1) as well as the
magnetization configurations (m,) of the electrodes.
The two spin states and as well as the k y modes due to
the finite width of the channel have to be summed up. Conductances G and G for Fe/InAs2DES/Fe junctions versus carrier density are shown for spin-valve as well as for
spin-transistor geometry in Figs. 11a and 11b, respectively. We focus on iron as the spin-injection rate predicted
for this material is higher than for Permalloy. The effect of
switching the magnetization between parallel and antiparallel
configurations can be seen in the upper black and gray traces
in Figs. 11a and 11b. Naturally, the effect of switching is
less pronounced if compared to the one expected in the original work1 where completely polarized ferromagnetic contacts are regarded. The degree of spin polarization in iron is
40%. This means for spin injection as well as for spin detection that both spin states in the ferromagnetic electrodes are
involved in the transport across F/InAs2DES/F junctions.
They mutually compensate spin-related effects in the conductance as can be seen from the transmission probabilites in

155322-10

BALLISTIC SPIN TRANSPORT AND SPIN . . .

PHYSICAL REVIEW B 65 155322

FIG. 11. Conductance and conductance ratio across a Fe/


InAs2DES/Fe double junction versus carrier density. The curves
are calculated for a channel length L150 nm and width W
1 m and for Z0.

Fig. 10. The density range plotted in Figs. 11a and 11b
corresponds to a variation of the spin-precession angle of
R 2 . The amplitudes of the conductance oscillations
are smaller than the ones of the transmission probabilities as
a result of averaging over all modes. The increasing mismatch of the Fermi velocity with electron density leads to
enhanced amplitudes of the oscillations. The spin-injection
rate from iron electrodes is typically of the order 10%
2 one
see Fig. 9. Accordingly, from the relation G/G
expects conductance ratios of about 1% when the magnetization of one of the electrodes is switched. But in our case,
can noticeably
depending on the density, the ratio G/G
2
differ from the value and even alter its sign. Unlike in
transport by a single mode,1 the density dependence of the
conductance dominantly reflects the increasing numbers
of modes with increasing electron density and not so much
the spin precession itself. Also the participation of both
spins from the ferromagnet and the oblique modes smear out
the conductance modulation brought about by the spin precession in the spin-transistor geometry. These background
contributions can be filtered out by subtracting two
conductances for proper magnetization configurations. Corresponding results are depicted as conductance ratios
2(G G )/(G G ) in Figs. 11a,11b and
G/G

Figs. 12a,12b. In the spin-valve geometry the ratio G/G

FIG. 12. Carrier-density dependence of the conductance ratio


across a Fe/InAs2DES/Fe junction at small densities. The curves
present enlarged views of the ones in Figs. 11a and 11b.

increases with electron density at low densities and saturates


at higher ones. In contrast to saturation, a modulation of the
conductance ratio can clearly be observed in case of spintransistor geometry. This result qualitatively agrees with experimental findings obtained for metal-oxide-semiconductor
field-effect transistors on p-type InAs with Permalloy source
and drain contacts.13,14
The large-scale variation observed in Fig. 11a recovers
the spin-valve characteristics of the ballistic spin-filter
transistor.6 Our transfer-matrix formalism takes into account
phase coherence and interference. They lead to the smallscale variations stemming from Fabry-Perot-type resonances.
At small electron densities they cause pronounced oscillating
conductance changes. Very interestingly, the conductance
change alters its sign as becomes more clear in the enlarged
Figs. 12a and 12b. The flanks of these Fabry-Perot resonances could be utilitized for switching in a kind of interference transistor. The resonance condition is tuned by the kinetic momentum k which is proportional to the group
velocity given in Eq. 14. In the absence of Rashba interaction one tunes the resonances via the relation k n s . On

155322-11

MATSUYAMA, HU, GRUNDLER, MEIER, AND MERKT

PHYSICAL REVIEW B 65 155322

p-type InAs the Rashba parameter is tuned via the carrier


density,35 but direct control of the Rashba parameter without
changing the carrier density53 also has been demonstrated.
This way the resonances can be tuned by the Rashba parameter alone.
V. CONCLUSIONS

We have presented a theory for the ballistic spin fieldeffect transistor based on the transfer-matrix formalism. We
have calculated the properties of the spin-polarized transport
for a device with a mesoscopic width, i.e., for a device realized in the reported experiments.1315 As important features
we have included in our theory the spin filtering at the
ferromagnet-semiconductor interface, the gate-controlled
spin-orbit coupling via the Rashba approximation, and the
effect of oblique modes and the spin precession along the
semiconductor channel. Introducing real band-structure parameters we have found that the spin-injection rate across
the Fe/InAs2DES interface depends on the carrier density
n s of the 2DES and can be tuned between 0.25 and
0.16 in the regime of density considered in Fig. 9a.
Here, the reason is the mismatch of the group velocities of
majority and minority spins in the ferromagnet with respect
to the group velocity in the semiconductor. Spin filtering is
enhanced by increasing the strength of the spin-independent
elastic scattering potential at the interface. In Fig. 9b the
injection rate is, e.g., enlarged by a factor of 3. The microscopic reason is that in the limit of tunneling the number of
momentum states is important for the spin-polarized transmission.
For the Fe/InAs2DES/Fe hybrid structure we have calculated the properties for the ballistic spin transport. We have
discussed two scenarios: the spin-valve and the spintransistor geometry Fig. 1. In both scenarios, for the dependence of the spin-polarized transport on the carrier density
we distinguish between a long-period and a short-period
2(G
variation in the conductance ratio G/G
G )/(G G ). In case of the spin-valve geometry, the
long-period variation found in our calculations represents the
characteristics of the ballistic spin-filter transistor.6 The formalism presented here goes further and includes interference
and coherence effects. They lead to the short-period oscillations due to Fabry-Perot resonances which are also tuned via
the electron density, i.e., by the gate voltage. Their important
impact on the spin-polarized transport is, in particular, evident in the spin-transistor geometry.
Our calculations in the spin-transistor geometry are based
on a spin field-effect transistor exhibiting a width of 1 m
and a channel length of 150 nm. They show both the effect of
spin precession along the semiconductor channel that leads

S. Datta and B. Das, Appl. Phys. Lett. 56, 665 1990.


G. Schmidt, D. Ferrand, L. W. Molenkamp, A. T. Filip, and B. J.
van Wees, Phys. Rev. B 62, 4790 2000.
3
Ph. Mavropoulos, N. Papanikolaou, and P. H. Dederichs, Phys.
1
2

and the Fabry-Perot


to the long-period variation of G/G
resonances that cause peaklike resonances. The long-period
variation of the conductance ratio shows the full period of
the Datta-Das spin precession mode in the considered regime
of density due to the gate-controlled Rashba spin-orbit interaction. Here, our formalism can answer the important question: How is an electron with spin injected across a real
ferromagnetic-semiconductor interface into the spinprecession state? We find that for normal injection the transmission probabilities from a spin subband in the ferromagnet
into both spin subbands of the semiconductor are the same
thus facilitating spin precession. However, they become different for oblique modes. This means that the coupling to the
spin-precession state is deteriorated by oblique injection. The
oblique modes contribute an undesired background to the
of several percent is
conductance. A maximum of G/G
predicted for the Fe/InAs2DES/Fe hybrid structure. In our
model calculation where we have used as far as possible
varrealistic device parameters the conductance ratio G/G
ies from positive to negative and back again at low carrier
density. In addition, Fabry-Perot resonances due to interference, which were not regarded in the original work of Datta
and Das, are superimposed. They are short-period oscillations and most pronounced in the regime of low carrier density. There, they lead to a rapid oscillation between positive
and negative conductance ratios. Since spin is involved in
these resonances, they are controlled via the spin-orbit interaction. We suggest to use the steep transfer function flank
of the Fabry-Perot resonances in an interference-transistor
device.
In the framework of our model we have also calculated
the spin-polarized transport in a Permalloy/InAs2DES/
Permalloy hybrid structure in the two scenarios. They reproduce the characteristic features discussed above. In particular, in the spin-valve geometry, the conductance change is
almost constant for higher carrier density. In the spintransistor geometry, a change of sign in the overall course of
the conductance ratio is predicted. This is the signature of
spin precession along the channel also for Permalloy electrodes. These features agree qualitatively with the reported
experimental results on metal-oxide-semiconductor fieldeffect transistor devices on p-type InAs with Permalloy
source and drain contacts.13,14
ACKNOWLEDGMENTS

We thank M. Wagner, R. Winkler, and P. Schmuser for


valuable discussions. Financial support of the Deutsche Forschungsgemeinschaft via the Sonderforschungsbereich 508
Quantenmaterialien and the NEDO International Joint Research Program Japan is gratefully acknowledged.

Rev. Lett. 85, 1088 2000.


G. Kirczenow, Phys. Rev. B 63, 054422 2001.
5
I. Zutic and S. Das Sarma, Phys. Rev. B 60, 16 322 1999.
6
D. Grundler, Phys. Rev. B 63, 161307R 2001.
4

155322-12

BALLISTIC SPIN TRANSPORT AND SPIN . . .

PHYSICAL REVIEW B 65 155322

D. Grundler, Phys. Rev. Lett. 86, 1058 2001.


H. B. Heersche, Th. Schapers, J. Nitta, and H. Takayanagi, Phys.
Rev. B 64, 161307R 2001.
9
Th. Schapers, J. Nitta, H. B. Heersche, and H. Takayanagi, Phys.
Rev. B 64, 125314 2001.
10
E. I. Rashba, Fiz. Tverd. Tela 2, 1224 1960 Sov. Phys. Solid
State 2, 1109 1960; Y. A. Bychkov and E. I. Rashba, J. Phys.
C 17, 6039 1984.
11
G. Meier and T. Matsuyama, Appl. Phys. Lett. 76, 1315 2000.
12
G. Meier, M. Halverscheid, T. Matsuyama, and U. Merkt, J. Appl.
Phys. 89, 7469 2001.
13
T. Matsuyama, G. Meier, D. Grundler, R. Kursten, and U. Merkt,
in Proceedings of the 25th International Conference on the
Physics of Semiconductors, Osaka 2000, edited by N. Miura and
T. Ando Springer, Heidelberg, 2001, p. 1803.
14
G. Meier, T. Matsuyama, and U. Merkt, Phys. Rev. B 65, 125327
2002.
15
C.-M. Hu, J. Nitta, A. Jensen, J. B. Hansen, and H. Takayanagi,
Phys. Rev. B 63, 125333 2001.
16
W. H. Lau, J. T. Olesberg, and M. E. Flatte, Phys. Rev. B 64,
161301 2001.
17
E. I. Rashba, Phys. Rev. B 62, 16 267 2000.
18
G. E. Blonder, M. Tinkham, and T. M. Klapwijk, Phys. Rev. B 25,
4515 1982.
19
L. W. Molenkamp, G. Schmidt, and G. E. W. Bauer, Phys. Rev. B
64, 121202R 2001.
20
C.-M. Hu and T. Matsuyama, Phys. Rev. Lett. 87, 066803 2001.
Note that the coordinate system is rotated with respect to the one
in the present work and that the definition of the Z parameter
differs by a factor 2.
21
U. Zulicke and C. Schroll, Phys. Rev. Lett. 88, 029701 2002.
22
P. Seba, P. Exner, K. N. Pichugin, A. Vyhnal, and P. Streda, Phys.
Rev. Lett. 86, 1598 2001.
23
A. Bournel, P. Dollfus, P. Bruno, and P. Hesto, Eur. Phys. J.: Appl.
Phys. 4, 1 1998.
24
A. Chrestin, T. Matsuyama, and U. Merkt, Phys. Rev. B 49, 498
1994.
25
N. A. Mortensen, K. Flensberg, and A.-P. Jauho, Phys. Rev. B 59,
10 176 1999.
26
J. Nitta, F. E. Meijer, and H. Takayanagi, Appl. Phys. Lett. 75,
695 1999.
27
H. X. Tang, F. G. Monzon, R. Lifshitz, M. C. Cross, and M. L.
Roukes, Phys. Rev. B 61, 4437 2000.
28
G. E. W. Bauer, Phys. Rev. Lett. 69, 1676 1992.
29
M. J. M. de Jong and C. W. J. Beenakker, Phys. Rev. Lett. 74,
1657 1995.
30
R. Fiederling, M. Keim, G. Reuscher, W. Ossau, G. Schmidt, A.
7
8

Waag, and L. W. Molenkamp, Nature London 402, 787 1999.


Y. Ohno, D. K. Young, B. Beschoten, F. Matsukura, H. Ohno, and
D. D. Awschalom, Nature London 402, 790 1999.
32
R. A. Morrow and K. R. Brownstein, Phys. Rev. B 30, 678
1984.
33
Ch. Schnittler and M. Kirilov, Phys. Status Solidi B 176, 143
1993.
34
T. Ando, A. B. Fowler, and F. Stern, Rev. Mod. Phys. 54, 437
1982.
35
T. Matsuyama, R. Kursten, C. Meiner, and U. Merkt, Phys. Rev.
B 61, 15 588 2000.
36
E. C. Stoner, Philos. Mag. 15, 1080 1933; Proc. R. Soc. London,
Ser. A 165, 372 1938.
37
D. Y. Petrovykh, K. N. Altmann, H. Hochst, M. Laubscher, S.
Maat, G. J. Mankey, and F. J. Himpsel, Appl. Phys. Lett. 73,
3459 1998.
38
C. Cohen-Tannoudji, B. Diu, and F. Laloe, Quantum Mechanics
Wiley, New York, 1977, Vol. 1.
39
E. A. de Andrada e Silva, G. C. La Rocca, and F. Bassani, Phys.
Rev. B 50, 8523 1994; 55, 16 293 1997.
40
J. Nitta, T. Akazaki, and H. Takayanagi, Phys. Rev. Lett. 78, 1335
1997; G. Engels, J. Lange, Th. Schapers, and H. Luth, Phys.
Rev. B 55, 1958 1997.
41
A. V. Moroz and C. H. W. Barnes, Phys. Rev. B 60, 14 272
1999.
42
F. Mireles and G. Kirczenow, Phys. Rev. B 64, 024426 2001.
43
U. Merkt and S. Oelting, Phys. Rev. B 35, 2460 1987.
44
C.-M. Hu, J. Nitta, T. Akazaki, H. Takayanagi, J. Osaka, P. Pfeffer, and W. Zawadzki, Phys. Rev. B 60, 7736 1999.
45
G. Bastard, Wave Mechanics Applied to Semiconductor Heterostructures Halstead, New York, 1988.
46
E. Merzbacher, Quantum Mechanics 3rd ed. Wiley, New York,
2000, p. 97.
47
A. Voskoboynikov, S. S. Liu, and C. P. Lee, Phys. Rev. B 58,
15 397 1998.
48
E. A. de Andrada e Silva and G. C. La Rocca, Phys. Rev. B 59,
R15 583 1999; A. Voskoboynikov, S. S. Liu, and C. P. Lee,
ibid. 59, 12 514 1999.
49
S. Datta, Electronic Transport in Mesoscopic Systems Cambridge
University Press, Cambridge, England, 1999.
50
F. J. Himpsel, J. E. Ortega, G. J. Mankey, and R. F. Willis, Adv.
Phys. 47, 528 1998; J. Callaway and C. S. Wang, Phys. Rev. B
16, 2095 1977.
51
R. A. de Groot, F. M. Mueller, P. G. van Engen, and K. H. J.
Buschow, Phys. Rev. Lett. 50, 2024 1983.
52
A. Majumdar, Phys. Rev. B 54, 11 911 1996.
53
D. Grundler, Phys. Rev. Lett. 84, 6074 2000.
31

155322-13

Das könnte Ihnen auch gefallen