The Rashba Hamiltonian and electron transport

Laurens W. Molenkamp and Georg Schmidt

Physikalisches Institut (EP3), Universit
at
Wurzburg, D-97074 W urzburg, Germany
arXiv:cond-mat/0104109v2 [cond-mat.mes-hall] 11 Jul 2001

Gerrit E.W. Bauer

Theoretical Physics Group, Department of Applied Physics and DIMES,
Delft University of Technology, Lorentzweg 1, 2628 CJ Delft, The Netherlands
(February 1, 2008)

This obviously cannot be correct because in the absence

The Rashba Hamiltonian describes the splitting of the con- of an external magnetic field, the spin-quantization axis
duction band as a result of spin-orbit coupling in the presence in the (isotropic) semiconductor can be rotated with the
of an external field and is commonly used to model the elec-
magnetization direction, which should therefore be with-
tronic structure of confined narrow-gap semiconductors. Due
out physical consequences.4 The problem with the calcu-
to the mixing of spin states some care has to be exercised in
the calculation of transport properties. We derive the velocity
lations of Ref. 3 can be traced to an incorrect treatment
operator for the Rashba-split conduction band and demon-
of the velocity operator, which in the presence of spin-
strate that the transmission of an interface between a ferro- orbit interaction is not simply given by ~~k/m, where m
magnet and a Rashba-split semiconductor does not depend is the effective mass of an electron and ~k its wavevector.
on the magnetization direction, in contrast with previous as- It is the purpose of our communication to clarify the is-
sertions in the literature. sues mentioned above. First, we will discuss the nature of
the eigenstates of the Rashba Hamiltonian in some detail
and derive the proper velocity operator. For comparison,
we give similar expressions for the eigenstates and ve-
Narrow-gap semiconductors, most notably InAs, play locity operator for a Stoner-Wohlfarth ferromagnet. Fi-
an important role in the rapidly evolving field of spin- nally, we calculate explicitly the transmission coefficient
tronics. As non-magnetic element in hybrid devices these between a ferromagnet and a Rashba-split electron gas
and show that the contact conductance is invariant with
materials are expected to help control the electron spin
states, just like the electron charge is controlled in con- respect to a magnetization reversal of the ferromagnet.
The Hamiltonian of an otherwise free electron sys-
ventional electronic devices. Part of this potential stems
from the natural two-dimensional electron gas (2DEG) tem, but including the Rashba spin-orbit scattering term
reads:2
on clean InAs surfaces, which allows high-quality ohmic
contacts to superconductors and ferromagnets. Another ~2
~2  
H= + i ~ E
~ ~ (1)
reason is the seminal paper of Datta and Das,1 which 2m
describes how the electrical field of an external gate elec-
trode can be used to manipulate the precession of a con- where is an effective mass parameter
duction electron spin. Essential for this mechanism is and ~ = (x, y , z ) is the vector of Pauli spin matrices.
the field-dependent spin-orbit coupling, which is rela- For a 2DEG with a confining electric field normal to the
interface E ~ = (0, 0, Ez ) :
tively large and well-established for the 2DEG on InAs.
It is now generally accepted that the spin-orbit interac- H=
tion in narrow-gap 2DEGs is governed by the Rashba 2
 2 2
  
~
Hamiltonian,2 which increases linearly with the electron E0 2 + y 2 hEz i i y
2m x x
wave vector.

~2 2 2
hEz i x + i y E0 x2 + y 2 (2)
The spin-orbit-interaction induced spin-splitting is 2m
sometimes confused with an exchange or Zeeman split- where hEz i is the expectation value over the lowest
ting. However, the latter require breaking of the time subband with energy E0 . Experimentally, one typically
inversion symmetry and are therefore fundamentally dif- observes5,6 values for hEz i on the order of 1011 eV m.
ferent from the former. It is then not surprising that The eigenstates for in-plane motion (identified by their
physical properties like exciton spin splittings or, in the
quantum numbers ~k = (kx , ky ) and s = 1) are
present context, spin-dependent transport properties of
narrow-gap hybrid devices are not well understood. In a   kx 
k
i~
is k i ky
recent paper, for example, it was argued3 that the con- ~ks (~r) = N~ks e k~r
. (3)
ductance of the interface between a ferromagnet and a 1
spin-orbit spin-split semiconductor should change on a where N~ks is a normalization factor. From this expres-
flip of the magnetization direction of the ferromagnet. sion for the eigenstates, it is immediately obvious that

1
the Rashba-split subbands are not spin-polarized.7 The is a unitary rotation matrix corresponding to a magneti-
electron energy dispersion relation E~ks reads (see Fig. zation direction of m
~ = (sin cos , sin sin , cos ) . For
1a): plane wave states with wave vector ~k and m ~ = (0, 1, 0)
the Hamiltonian for the ferromagnet
~2 h 2 2
i
E~ks = E0 + (k + skR ) kR (4)
2m
!
 ~2 2
k i

q Hncf k, = , = = 2m ~2 2
where k = kx2 + ky2 , kR = hEz i m/~2 . 2 2 i 2m k (12)
The normalization factor of the eigenfunctions can
be determined in different ways. Normalization of the is formally equivalent to that of the Rashba Hamiltonian
R  2
probability distribution d~r ~ks (~r) = 1 gives N~ks =
  !
~2 2
k i
1/ 2S where S is the area of the 2DEG. However, for a R
HR = 2m
~2 2
(13)
calculation of the transport properties it is more conve- iR 2m k
nient to normalize the states such that its currents are
unity in the transport, say x, direction. To this end we with R = hEz i (kx iky ) . In the ferromagnet the ve-
have to compute the expectation value of the current or locity operator is always diagonal in spin space, however:
velocity operator which, in the presence of the Rashba  
term, are not simply proportional to the gradient opera- px 1 0
vx,ncf (k, , ) = . (14)
tor anymore. The proper matrix representation in spinor m 0 1
space can be derived via the Hamilton equation of mo-
tion: In order to demonstrate explicitly that transport
through a Rashba semiconductor/ferromagnet junction
H H H H does not depend on the magnetization direction of the
q = ; p = ; x = vx = ; px =
p q px x (5) ferromagnet, it is sufficient to consider the simple case of
a single mode quantum point contact (Fig. 2) without
The velocity operator in the x-direction therefore reads: an additional interface potential barrier. A ferromag-
2
! net on the left side of the contact (its electronic states

1 i ~m x i hEz i will be indicated by superscript L in the following) is at-
vx = 2

. (6)
~ i hEz i i ~m x tached (at x = 0) to a Rashba semiconductor on the right

 (superscript R). In the semiconductor we have eigen-
Requiring ~ks (~r) |
vx |~ks (~r) = 1 we find states at the Fermi energy EF = 2m ~2 2
kF at wave vectors
p
r s ks = skR + kR 2 + k 2 which are taken to be positive
m 1 F
N~ks = (7) in the following. The states at the Fermi energy are right
kx 1 + s kkR 


2~
moving:


This value diverges when the group velocity vanishes, i.e.  

1
 
R ik+ x R ik x 1
for s = 1 at k = kR . k+ (x) = N e ; k (x) = N e
The above considerations for a 2DEG are only slightly i i (15)
modified for a quantum wire. For the lowest mode:
! and left moving
p2x
E0 + 2m i hE zi
p x
H (x, px ) = ~
p2x
(8)  
1
 
1
i hE
~
zi
p x E
0 + 2m R
k+ (x) = N e ik+ x R
; k (x) = N e ik x
! i i(16)
px hEz i
H m i
vx = = ~ (9)
px i hE ~
zi px
m with normalization
r s
Note that there is no Rashba level splitting in a quantum
r r
m 1 m 1
dot. N= = p
2~ ks + skR 2~ kR + kF2
2
(17)
It is instructive to compare the Rashba Hamiltonian
with that of a 2D non-collinear ferromagnet with a dis-
persion as sketched in Fig. 1b: The flux normalization reflects the identical group veloc-
ities for the two bands. The normalization is invariant
p2y
!
p2x under a unitary transformation which diagonalizes the
+ 0
Hncf = 2m 2m p2 p2y
+ U + z U Hamiltonian. We have seen above that we can interpret
0 2m + 2m
x
(10) the Rashba semiconductor as a pseudo-ferromagnet in
where 2 is the exchange splitting and which the magnetization is rotated from the z to the y
direction, and with a k-dependent exchange splitting R .
cos /2 ei sin /2
 
We simplify the situation by taking the quantization axis
U (, ) = (11)
ei sin /2 cos /2 of the ferromagnet parallel to the pseudo-magnetization

2
of the Rashba Hamiltonian by transforming the Rashba We can now calculate the conductance via the Landauer
Hamiltonian as follows: formula:
    q
U+ , HR U , = e2 e 2 4 1 + (kR /kF )
2
2 !2 2 2 G = |t |2 = 2
p2x h h
   q
0 0 1 1 + 1 + (kR /kF ) 2 (29)
2m
p2x
+ hE z i , (18)
0 2m x 1 0

yielding the following eigenstates along the quantization To calculate G , we flip the magnetization of the ferro-
(y) axis of the ferromagnet magnet on the left side, yielding as incoming state
r     
s m ikF x 0 ikF x 0
L
 
R m 1 (x) = e + r e
U k+ (x) = p , (19) ~kF 1 1 (30)
~ kR + kF 0
2 2

while transmission occurs only into

s  
R m 0
U k (x) = p . (20)
~ kR + kF 1
2 2 s  
R m ik x 0
(x) = t p e (31)
On the left side we assume first a half-metallic ferromag- ~ kR 2 + k2
F
1
net, for which the conduction electrons are either all spin
up or down with wave vector kF : Flux continuity gives:
r
m ikF x 1
  ~ L
L (x) = e (21) vx L
(x) |x=0 = (x) |x=0 = vx R
(x) |x=0

~kF 0 im x (32)
r  
m ikF x 0 ~ ~
q
L (x) = e (22) = t (k kR ) = t kR 2 + k2
F
~kF 1 m m (33)
Assuming that the spin is up on the left side, we can now
write the eigenstates of the ferromagnet in terms of the and, comparing this expression with Eq. (28), we see
reflection coefficient r : that the transmission coefficient is identical for up and
down spins. This is in contrast with the counterintuitive
r
m
     results of Ref. 3 - where, we believe, an incorrect velocity
L ikF x 1 ikF x 1
(x) = e + r e . operator has been applied.
~kF 0 0 (23) Since the effective-mass Rashba Hamiltonian of Eq. (1)
is isotropic, the interface conductance is invariant under
On the side of the Rashba-split semiconductor, we have arbitrary rotations of the magnetization direction. In
transmission for one spin direction only, which corre- addition, the above calculations may be generalized to
sponds to a wave vector k+ . transmission from a weak ferromagnet with both spins
s   occupied up to Fermi numbers k/ with
R m ik+ x 1
(x) = t+ p e . (24)
2 + k2 0
p
~ kR F e2 X 4 kR 2 + k2
F
G= 2 (34)
h
 p
=, 2 2
1 + (kR + kF ) /k2
where t+ is the transmission coefficient. The transport
coefficients are determined by the requirement of the con-
tinuity of the wave function and its flux (not simply the The interface conductance should therefore not affect
derivative) at the interface x = 0: anisotropies due to interference effects in the Datta
transistor.1
L R
(0) = (0) (25) We hope that this paper will help to dispel the confu-
sion concerning the transport properties of semiconduc-
vx L
(x) |x=0 = vx R
(x) |x=0 (26)
tors with spin-orbit interactions. We compared eigen-
The condition for flux continuity can be rewritten as states and velocity operators for two systems, a non-
magnetic 2DEG in the presence of the Rashba Hamil-
~ L tonian and a non-collinear Stoner-Wohlfarth model fer-
(x) |x=0
im x s romagnet. As expected, the transmission coefficient of
~



m an interface between a ferromagnet and a Rashba-split
= + ikR t p eik+ x |x=0 semiconductor is found independent on the magnetiza-
im x ~ kR 2 + k2
(27)
F tion direction of the ferromagnet.
~ ~
q Note that the independence of the total conductance
= (k+ + kR ) = t kR 2 + k2 (28)
m m F of a single ferromagnetic/normal metal interface on the

3
magnetization direction is quite general, but does not M. Villeret, Appl. Phys. Lett. 75, 2307 (1999); J. Phys.:
mean that the interface is not spin-selective. Indeed, Condens. Matter 12, 2833 (2000).
9
a ferromagnet does inject a net spin into the nonmag- G. Kirczenow, Phys. Rev. B 63, 05422 (2001).
10
netic material, with efficiencies that depend on the spe- R. Fiederling, M. Keim, G. Reuscher, W. Ossau, G.
cific electronic band structures.8,9 Small modulations of Schmidt, A. Waag, and L.W. Molenkamp, Nature (Lon-
a single interface conductance could be achieved in prin- don) 402, 787 (1999); Y. Ohno, D.K. Young, B. Beschoten,
ciple by forcing the magnetization vector of the ferro- F. Matsukara, H. Ohno, and D.D. Awschalom, Nature
magnet into directions which deviate from the crystal (London) 402, 790 (1999).
symmetry axes. However, in order to detect a strongly
spin-polarized interface transmission by a transport ex-
FIG. 1. Schematic representation of the conduction band
periment, an analyzing ferromagnet is essential. This is structure of (a) a semiconductor in which the spin-degeneracy
employed, of course, in the giant magnetoresistance ef- is broken by spin-orbt interaction as described by the Rashba
fect. In semiconductors, the spin-polarized current can Hamiltonian and (b) a spin-polarized exchange-split band fer-
also be detected by the circular optical polarization of romagnet (Stoner-Wohlfarth model).
the electroluminescence of a light emitting diode.10
We thank the Volkswagenstiftung for supporting a
Rontgen-Professorship of G.E.W.B. at W urzburg Uni- FIG. 2. The system under consideration: we discuss the
versity. G.E.W.B. acknowledges support by the NEDO conductance of a ferromagnet-semiconductor hybrid quantum
project NTDP-98; L.W.M. and G.S. acknowledge sup- point contact. The band structures of the two materials are
port by the DARPA SPINS programme. depicted in Fig. 1. The constriction separating the two ma-
Note added: After submission of this manuscript, we terials symbolizes the single-channel adiabatic transport we
received a preprint by Z ulicke and Schroll with similar assume in our calculation. The ferromagnet is magnetized in
results. Bruno and Pareek, cond-mat/0105506, report the y-direction and the current flows in the xdirection.
numerical calculations for the same system. In con-
trast to what we report here, the latter authors find a
small anisotropy in the transport as a function of the
magnetization angle. These anisotropies are allowed by
(Casimir-Onsager) symmetry, but they vanish for the ef-
fective mass Hamiltonian (1), which does not contain any
warping corrections which reflect the reduced symme-
try of the crystalline lattice.

1
S. Datta and B. Das, Appl. Phys. Lett. 56, 665 (1990).
2
E.I. Rashba, Fiz. Tverd. Tela 2, 1224 (1960) [ Sov. Phys.
Solid State 2, 1109 (1960)]; Yu. A. Bychkov and E.I
Rashba, Pisma Zh. Eksp. Teor. Fiz. 39, 66 (1984) [JETP
Lett. 39, 78 (1984)].
3
D. Grundler, Phys. Rev. Lett. 86, 1058 (2001).
4
This argument holds also in the presence of stray fields from
the ferromagnet which rotate invariantly with the magne-
tization.
5
J. Nitta, T. Akazaki, H. Takayanagi, and T. Enoki, Phys.
Rev. Lett. 78, 1335 (1997).
6
There is some uncertainty as to whether the experimental
data on might suffer from an interfering effect; see, e.g.,
A.C.H. Rowe, J. Nehls, R.A. Stradling, and R.S. Ferguson,
Phys. Rev. B 63, 201307(R) (2001).
7
Note that that even in the limit of hEz i = 0 the eigen-
states (3) are linear combinations of spin-up and spin-down
states. Pure spin eigenstates could be obtained in this limit
by including a Zeeman spin splitting in the Hamiltonian
which in the end is allowed to vanish.
8
M.S. Ferreira, J. dAlbuquerque e Castro, R.B. Muniz, and

4
 E

a)

kR k

b)
2
k
Ferromagnet InAs

z y
x