Spin polarizations and spin Hall currents in a two-dimensional electron gas with magnetic impurities
C. Gorini, P. Schwab, and M. Dzierzawa
Institut für Physik, Universität Augsburg, 86135 Augsburg, Germany R. Raimondi
CNISM and Dipartimento di Fisica “E. Amaldi,” Università Roma Tre, 00146 Roma, Italy 共Received 16 May 2008; published 29 September 2008兲
We consider a two-dimensional electron gas in the presence of Rashba spin-orbit coupling, and study the effects of magnetics-wave and long-range nonmagnetic impurities on the spin-charge dynamics of the system.
We focus on voltage induced spin polarizations and their relation to spin Hall currents. Our results are obtained using the quasiclassical Green function technique, and hold in the full range of the disorder parameter␣pF. DOI:10.1103/PhysRevB.78.125327 PACS number共s兲: 72.25.⫺b In the field of spintronics, much attention has recently
been paid to spin-orbit related phenomena in semiconduc- tors. One such phenomenon is the spin Hall effect, i.e., a spin current flowing perpendicular to an applied electric field.1–4 It is now well known that, for linear-in-momentum spin-orbit couplings such as the Rashba or Dresselhaus ones, the spin Hall current vanishes exactly in the bulk of a disordered two-dimensional electron gas共2DEG兲.5–8This can be under- stood by looking at the peculiar form of the continuity equa- tions for the spin, as derived from its equations of motion in operator form.9–11 For a magnetically disordered 2DEG things are, however, different, and a nonvanishing spin Hall conductivity is found.12–14Once more, a look at the continu- ity equations provides a clear and simple explanation of the effect:13a term, whose appearance is due to magnetic impu- rities, directly relates in-plane spin polarizations, induced by the electric field, to spin currents. As the former, which have been the object of both theoretical and experimental studies,15–20 are influenced by the type of nonmagnetic scat- terers considered, we forgo the simplified assumption that these areswave and take into account the full angle depen- dence of the scattering potential. Besides going beyond what is currently found in the literature, where, in the presence of magnetic impurities, the nonmagnetic disorder is either ne- glected or purelyswave, our approach also shows the inter- play between polarizations and spin currents in a 2DEG.21 We note that in the correct limits our results agree with what is found in Ref. 14. On the other hand a discrepancy with Ref. 12arises.
For the calculations we rely on the Eilenberger equation for the quasiclassical Green function in the presence of spin- orbit coupling.22 The spin-orbit energy is taken to be small compared to the Fermi energy, i.e., ␣pFⰆ⑀F—or equiva- lently ␣ⰆvF, and the standard metallic regime condition 1/Ⰶ⑀F is also assumed. Here ␣ is the spin-orbit coupling constant,pF共vF兲 the Fermi momentum共velocity兲in the ab- sence of such coupling, and the elastic quasiparticle life- time due to nonmagnetic scatterers. Our results hold for a wide range of values of the dimensionless parameter ␣pF since this is not restricted by the above assumptions. Contri- butions of order共␣/vF兲2are neglected throughout. We focus on intrinsic effects in the Rashba model; extrinsic ones,23
Dresselhaus terms,24 and hole gases25are not taken into ac- count. Finally, weak localization corrections, which could in principle play an important role,11 are beyond the scope of our present work.
The Hamiltonian of the 2DEG, confined to thex−yplane, reads
H= p2
2m−b·+V共x兲, 共1兲 with b=␣ez⫻p the Rashba internal field, the vector of Pauli matrices, andV共x兲=Vnm共x兲+Vm共x兲the disorder poten- tial due to randomly distributed impurities.26 Nonmagnetic scatterers give rise to Vnm共x兲,
Vnm共x兲=
兺
i U共x−Ri兲, 共2兲while Vm共x兲 describes magnetics-wave disorder Vm共x兲=
兺
i
B·␦共x−Ri兲. 共3兲 Both potentials are treated in the Born approximation, and the standard averaging technique is applied.
To begin with, we look at the continuity equation for the syspin polarization,13,27
tsy+x·jsy= − 2m␣js
z
y − 4
3sf
sy, 共4兲
where the second term on the right-hand side is due to mag- netic impurities. Heresfis the spin-flip time that stems from the potential共3兲 关cf. Eq.共13兲兴. Under stationary and uniform conditions the above equation implies a vanishing spin current—hence a vanishing spin Hall conductivity—unless magnetic disorder is also present, in which case instead
jsz
y = − 2
3m␣sf
sy. 共5兲
Since the out-of-plane polarized spin current is related to the in-plane spin polarization, we now use simple physical argu- ments to explain how the latter is generated by an applied voltage.15,28Since the Fermi surface is shifted by an amount
1098-0121/2008/78共12兲/125327共5兲 125327-1 ©2008 The American Physical Society
proportional to the applied electric field 共say along the x direction兲, as shown in Figs. 1共a兲 and 1共b兲, there will be more occupied states with spin up—alongy—than with spin down. In the case of short-range disorder, the total in-plane polarization can be estimated to be proportional to the den- sity of states multiplied by the shift in momentum, sy
⬃N␦p⬃N兩e兩E. Since in the present situation we are dealing with the two Fermi surfaces corresponding to the two helic- ity bands ⑀⫾=p2/2m⫾␣p, obtained from the Hamiltonian 共1兲, one expectssy⬃共N+−N−兲␦p, where, for the Rashba in- teraction, one has N⫾=N0共1⫿␣/vF兲, N0=m/2. Explicit calculations agree with this simple picture and lead to the result due to Edelstein,15 sy= −N0␣兩e兩E. When long-range disorder is considered, a reasonable guess would be to sub- stitute forthe transport timetr
→tr, 1
tr
=
冕
dW共兲关1 − cos共兲兴, 共6兲withW共兲 being the angle-dependent scattering probability, so thatsy= −N0␣兩e兩Etr. This was proposed in Ref.29; how- ever, the picture is too simplistic, and therefore the guess is wrong. As discussed in Ref.21, the propersypolarization is given bysy= −N0␣兩e兩EE, with
→E, 1
E
=
冕
dW共兲关1 − cos共2兲兴. 共7兲This particular time E, where “E” stands for Edelstein, arises from the asymmetric shift of the two Fermi surfaces, as depicted in Fig.1共c兲, due to different transport times in the two bands. It shows that contributions from both forward 共= 0兲and backward共=兲scatterings are suppressed. The
next step is to consider what happens when magnetic impu- rities are included. Relying once again on the simple picture of the shifted Fermi surface, one could argue that these have a rather small impact on the spin polarization since the spin- flip scattering time usually makes a small contribution to the total transport time. However, even when this is the case, magnetic disorder does not simply modify the total transport time but has an additional nontrivial effect. In its presence the spins do not align themselves along the internal b field since they acquire nonvanishing components in the plane orthogonal to it 关see Fig. 1共d兲兴. It is these components that give rise to a finite spin Hall conductivity. In this respect, magnetic disorder has an effect similar to that of an in-plane magnetic field: it affects the spin-quantization axis and tilts the spins out of their expected stationary direction. We now make these arguments quantitative.
The starting point is the Eilenberger equation,22which we write explicitly for a homogeneous Rashba 2DEG in linear response to a constant and homogeneous applied electric field
tgK=vF·E兩e兩⑀geqK −1
2
再
p1Fb·,e·E兩e兩⑀geqK冎
+i关b·,gK兴−i关⌺ˇ,gˇ兴K. 共8兲 The quasiclassical Green function 关gˇ⬅gˇt1t2共pˆ;x兲兴is defined as 共=p2/2m−兲
gˇ= i
冕
dGˇt1t2共p,x兲, Gˇ =冉
G0R GGKA冊
, 共9兲where Gˇ
t1t2共p,x兲 is the Wigner representation of the Green function, which has a matrix structure in both Keldysh共de- noted by the check symbol兲and spin space. Equation 共8兲 is the equation of motion for the Keldysh component—the one related to physical observables—identified by the superscript
“K,” which will be from now on implicitly assumed and thus dropped. Moreover, geqK= tanh共⑀/2T兲共geq
R−geqA兲, where geqR = −geqA = 1 −b·, indicates the equilibrium—no electric field—function.22All objects are evaluated at the Fermi sur- face in the absence of spin-orbit coupling while is the angle defined by the momentum, p=p共cos, sin兲, ande
=共−sin, cos兲. From Eqs.共2兲and共3兲one obtains the self- energy contributions
⌺ˇ
nm共p兲=nnm
兺
p⬘
兩U共p−p
⬘
兲兩2Gˇ共p⬘
兲, 共10兲and
⌺ˇ
m=nmB2 3
兺
l=1
3
兺
p lGˇ共p兲l, 共11兲wherennmandnmdenote the concentrations of nonmagnetic and magnetic impurities, respectively. In order to consider long-range nonmagnetic disorder, we first expand the non- magnetic scattering kernel in spherical harmonics of the scat- tering angle, and neglect its dependence on the modulus ofp andp
⬘
py
px py
px Ex
(a) (b)
py
px px
py
(c) (d)
FIG. 1. 共Color online兲 关共a兲 and 共b兲兴 The Fermi-surface shift,
␦p=兩e兩E, due to an applied electric field along thexdirection. The white arrows show the direction of the internal fieldb.关共c兲and共d兲兴 Shifted bands and spin polarization in stationary conditions. 共c兲 Asymmetric shift of the two bands when angle-dependent scattering is present. The long dark 共blue兲 arrows show the contributions to the spin polarization arising from a sector d of phase space.共d兲 When magnetic disorder is turned on, additional contributions or- thogonal to the internal field b appear, here shown by the short inward and outward pointing共blue兲arrows. Out-of-plane contribu- tions are also present but, for the sake of simplicity, are not shown.
nnm兩U兩2= 1
2N0关1 + 2K1cos共−
⬘
兲 + 2K2cos共2− 2⬘
兲+ . . .兴⬅ 1
2N0关1 +K共−
⬘
兲兴, 共12兲 with the nonmagnetic contribution to the elastic lifetime.Then we write the magnetic scattering kernel in terms of the spin-flip time sf,
nmB2= 1 2N0sf
. 共13兲
The complete disorder self-energy can then be written, sepa- rating itss-wave and higher harmonics contributions,
⌺ˇ=⌺ˇ
m+⌺ˇ
nm 1 +⌺ˇ
nm 2 = − i
6sf
兺
l=1 3
l具gˇ典l− i 2具gˇ典−
i 2具Kgˇ典,
共14兲 where具. . .典⬅兰d/2. . ..
The connection betweengˇand the physical observables is made by integrating over the energy⑀, which is the Fourier conjugate variable of the time differencet1−t2. For instance, the spin density is given by the angular average of the Keldysh component,30
s=seq−N0
8
冕
d⑀具Tr共g兲典. 共15兲In order to solve Eq.共8兲, it is convenient to turn it into matrix form, writingg as a four-vector
g=g00+g·, 共g兲=共g0,g兲. 共16兲 Rather than using the standard共x,y,z兲basis, we choose to rotate to 共储,⬜,z兲, the subscripts 储 and⬜, indicating, respectively, the directions parallel and perpendicular to the internal field b. Defining the rotation matrixR共兲by
冢
0xyz冣
=冢
100 − cos0 sin00 cossin00 0001冣冢
⬜0储z冣
, 共17兲one has
g
⬘
=兺
⬘=0
3
R−1⬘共兲g⬘, 共g
⬘
兲=共g0,g储,g⬜,gz兲, 共18兲K共,
⬘
兲=兺
⬘=0
3
R−1⬘共兲K共−
⬘
兲R⬘共⬘
兲. 共19兲 Expanding in harmonics, we also drop the four-vector indi- cesK共,
⬘
兲=K共a兲+ cos共−⬘
兲K共b兲+ sin共−⬘
兲K共c兲+ . . . . 共20兲 In the above we have definedK共a兲=
冢
0000 K0001 K0001 0000冣
, K共b兲=冢
2K0001 K0002 K0002 2K0001冣
,共21兲 and
K共c兲=
冢
0000 K0002 −000K2 0000冣
. 共22兲For the purpose of calculating polarizations and spin cur- rents, the higher harmonics play no role and are thus ignored.
By using geqR = −geqA = 1 −b· and performing a rotation to the new spin basis, one can write Eq.共8兲as
tg
⬘
=1ⴱ关−Mg⬘
+共N0+N1兲具g⬘
典+共N2+N3兲具Kg⬘
典兴+SE. 共23兲 The matrices appearing in Eq.共23兲readM=
冢
−ⴱ100v␣FK1 −ⴱ100v␣FK1 − 2␣001pFⴱ 2␣p001Fⴱ冣
,共24兲N0=
冢
10 1 −00 00034sfⴱ 1 −00034sfⴱ 1 −00034sfⴱ冣
, 共25兲N1= ␣
vF
冢
− 1000 −冉
1 −00034sfⴱ冊
0 00 00 00 0冣
, 共26兲N2=ⴱ
␣
vF
冢
− 1000 − 1 0 0000 0 00 00 0冣
, N3=ⴱ冢
1 0 0 00 1 0 00 0 1 00 0 0 1冣
,共27兲 whereⴱis the elastic quasiparticle lifetime, defined as
1
ⴱ⬅ 1
+ 1
sf
, 共28兲
which we now use for convenience of notation but will be later incorporated into the proper transport time. Finally, SE is the source term due to the electric field. We take this to be along thexdirection so that
SE⬅ 兩e兩vFE⑀关2 tanh共⑀/2T兲兴
冢
− cos− sincos0vv␣␣FF冣
. 共29兲Solving for theszspin current flowing alongy, we obtain
js
z y = −N0
4
冕
d⑀vF具pˆygz典= −N0
4
冕
d⑀冤
− 342msf−␣i冥
共具pˆyg⬜典−具pˆxg储典兲= −N0
4
冕
d⑀冤
− 342msf−␣i冥
具gy典=
冤
− 342msf−␣i冥
sy, 共30兲i.e., the continuity equation result 关Eq. 共4兲兴 under homoge- neous conditions. In the third line we have used Eq.共17兲to set具gy典=具pˆyg⬜典−具pˆxg储典. Similarly, one obtains the complete expression for the frequency dependentsyspin polarization
sy= −N0␣兩e兩E2共␣pF兲2
冋 冉
1tr−i冊冉
1E−i冊冉
34sf−i冊
+ 2共␣pF兲2
冉
1E+ 4 3sf
− 2i
冊 册
−1. 共31兲Besides 1/sf, there appear in the above two other different time scales,
1
tr
⬅1
共1 −K1兲+ 1
sf
, 1
E
⬅1
共1 −K2兲+ 1
sf
.
The first,tr, is the total transport time. The second,E, is the generalization of the characteristic time related to thesyspin polarization introduced in共7兲. By using Eq.共31兲in Eq.共30兲, one obtains the expression for the frequency dependent spin Hall conductivity
sH共兲= 兩e兩 4
冉
34sf−i
冊
2共␣pF兲2冋 冉
1tr−i冊冉
1E−i冊
⫻
冉
34sf−i
冊
+ 2共␣pF兲2冉
1E+ 4 3sf
− 2i
冊 册
−1.共32兲 Its real part is displayed in Fig. 2for different values of the disorder parameter ␣pF. In the limit →0, the magnitude of the spin Hall conductivity depends on the value of␣pFas well as on the ratio/sf. In the absence of magnetic impu- rities one has the known resultsH= 0. As spin-flip scattering grows, the conductivity reaches values of the order of the
“universal”兩e兩/8. This was noted already in Ref.12, where however, as pointed out in the beginning, angle-dependent scattering was not considered. Large values of ␣pFcan be achieved both in III-V and II-VI semiconducting materials.
Doping the latter with Mn allows controlling of the spin-flip time sf while only weakly affecting the electron mobility31–33 even though it is not perfectly clear whether these can appropriately be described in terms of the linear Rashba model.34 Additionally, for certain frequencies one can see crossing points关⬇0.5 and⬇2 in Fig.2共a兲兴at which magnetic disorder has no effect on the spin Hall re- sponse. Such points are well defined only when ␣pF⬇1.
For clean共␣pFⰇ1兲or dirty共␣pFⰆ1兲samples, the differ- ent curves cross each other over a progressively wider range of frequencies.
Finally, in the diffusive regime, trⰆ1,␣pFtrⰆ1, and
τ/τsf= 0.4 τ/τsf= 0.3 τ/τsf= 0.2 τ/ττ/τsfsf= 0.1= 0
αpFτ= 1 ωτ ReσsH/(e/8π)
5 4 3 2 1 0 1.2
1 0.8 0.6 0.4 0.2 0 -0.2 -0.4
τ/τsf= 0.4 τ/τsf= 0.3 τ/τsf= 0.2 τ/ττ/τsfsf= 0.1= 0
αpFτ= 5
ωτ ReσsH/(e/8π)
20 15
10 5
0 4 3 2 1 0 -1 -2 -3 (a)
(b)
FIG. 2.共Color online兲Real part of the frequency dependent spin Hall conductivity in units of the universal value 兩e兩/8 for 共a兲
␣pF= 1 and共b兲 ␣pF= 5. The different curves correspond to dif- ferent values of the ratio/sf= 0 , 0.1, 0.2, 0.3, 0.4共from top to bot- tom at the maximum of ResH兲.
assuming tr/sfⰆ1, E/sfⰆ1, one obtains the following spin-diffusion equations:
tsx= −
冉
1s+ 4
3sf
冊
sx, 共33兲tsy= −
冉
1s+ 4
3sf
冊
sy−␣N0兩e兩EEs, 共34兲
tsz= −
冉
2s+ 4
3sf
冊
sz, 共35兲where 共2␣pFtr兲2/2tr⬅1/s is the D’yakonov-Perel spin- relaxation rate, tied to Rashba spin-orbit coupling. From Eq.
共34兲the sensitivity of the in-plane spin polarization on spin- flip scattering is apparent: in the stationary limit the source
共proportional toE兲is balanced by the spin relaxation. Spin- flip scattering leaves the source unchanged, whereas it en- hances the relaxation rate so that in the end syis reduced.
In conclusion, we studied the combined effect of long- range and magnetic disorders on voltage induced spin polar- izations and the related spin Hall currents in a Rashba 2DEG.
We investigated homogeneous but nonstatic conditions from the dirty共␣pFⰆ1兲to the clean 共␣pFⰇ1兲 regime. Care is required when treating long-range disorder because of the two-band structure of the problem while magnetic impuri- ties, even in low concentrations, play a nontrivial role be- yond that of a simple redefinition of the time scales.
This work was supported by the Deutsche Forschungsge- meinschaft through SFB 484 and SPP 1285, and by CNISM under Progetti Innesco 2006.
1M. I. D’yakonov and V. I. Perel, Phys. Lett. 35A, 459共1971兲.
2J. E. Hirsch, Phys. Rev. Lett. 83, 1834共1999兲.
3S. Murakami, N. Nagaosa, and S.-C. Zhang, Science 301, 1348 共2003兲.
4J. Sinova, D. Culcer, Q. Niu, N. A. Sinitsyn, T. Jungwirth, and A.
H. MacDonald, Phys. Rev. Lett. 92, 126603共2004兲.
5J. I. Inoue, G. E. W. Bauer, and L. W. Molenkamp, Phys. Rev. B 70, 041303共R兲 共2004兲.
6E. G. Mishchenko, A. V. Shytov, and B. I. Halperin, Phys. Rev.
Lett. 93, 226602共2004兲.
7R. Raimondi and P. Schwab, Phys. Rev. B 71, 033311共2005兲.
8A. Khaetskii, Phys. Rev. Lett. 96, 056602共2006兲.
9E. I. Rashba, Phys. Rev. B 70, 201309共R兲 共2004兲.
10O. V. Dimitrova, Phys. Rev. B 71, 245327共2005兲.
11O. Chalaev and D. Loss, Phys. Rev. B 71, 245318共2005兲.
12J. I. Inoue, T. Kato, Y. Ishikawa, H. Itoh, G. E. W. Bauer, and L.
W. Molenkamp, Phys. Rev. Lett. 97, 046604共2006兲.
13C. Gorini, P. Schwab, M. Dzierzawa, and R. Raimondi, Physica E共Amsterdam兲 40, 1078共2008兲.
14P. Wang, Y.-Q. Li, and X. Zhao, Phys. Rev. B 75, 075326 共2007兲.
15V. M. Edelstein, Solid State Commun. 73, 233共1990兲.
16A. G. Aronov and Y. B. Lyanda-Geller, JETP Lett. 50, 431 共1989兲.
17Y. K. Kato, R. C. Myers, A. C. Gossard, and D. D. Awschalom, Phys. Rev. Lett. 93, 176601共2004兲.
18V. Sih, R. C. Myers, Y. K. Kato, W. H. Lau, A. C. Gossard, and D. D. Awschalom, Nat. Phys. 1, 31共2005兲.
19C. L. Yang, H. T. He, Lu Ding, L. J. Cui, Y. P. Zeng, J. N. Wang, and W. K. Ge, Phys. Rev. Lett. 96, 186605共2006兲.
20N. P. Stern, S. Ghosh, G. Xiang, M. Zhu, N. Samarth, and D. D.
Awschalom, Phys. Rev. Lett. 97, 126603共2006兲.
21M. Milletarì, R. Raimondi, and P. Schwab, Europhys. Lett. 82,
67005共2008兲.
22R. Raimondi, C. Gorini, P. Schwab, and M. Dzierzawa, Phys.
Rev. B 74, 035340共2006兲.
23W.-K. Tse and S. Das Sarma, Phys. Rev. B 74, 245309共2006兲.
24M. Trushin and J. Schliemann, Phys. Rev. B 75, 155323共2007兲.
25C.-X. Liu, B. Zhou, S.-Q. Shen, and B.-F. Zhu, Phys. Rev. B 77, 125345共2008兲.
26Had we chosen the linear Dresselhaus field, the whole calcula- tion would have gone through without substantial difference.
More in general, this holds for any internal field b, where the direction ofb does not depend on the modulus ofp.
27Several suggestions for the definition of the spin current exist in the literature. We choose a nonconserved spin current js
i
= 1/4兵i,v其,i=x,y,z,v=ⵜpH, which is consistent with the SU共2兲 covariant formulation, see, e.g., P. Q. Jin, Y. Q. Li, and F. C. Zhang, J. Phys. A 39, 7115共2006兲.
28D. Culcer and R. Winkler, Phys. Rev. Lett. 99, 226601共2007兲.
29H. A. Engel, E. I. Rashba, and B. I. Halperin, Phys. Rev. Lett.
98, 036602共2007兲.
30Within the quasiclassical formalism,g describes the dynamical part only, to which the equilibrium part has to be added.
31Y. S. Gui, C. R. Becker, J. Liu, V. Daumer, V. Hock, H. Buh- mann, and L. W. Molenkamp, Europhys. Lett. 65, 393共2004兲.
32V. Daumer, I. Golombek, M. Gbordzoe, E. G. Novik, V. Hock, C. R. Becker, H. Buhmann, and L. W. Molenkamp, Appl. Phys.
Lett. 83, 1376共2003兲.
33S. A. Crooker, D. A. Tulchinsky, J. Levy, D. D. Awschalom, R.
Garcia, and N. Samarth, Phys. Rev. Lett. 75, 505共1995兲.
34More precisely, the Rashba Hamiltonian is appropriate for nar- row quantum wells共widthⱗ6 nm兲but most likely not for wider structures, in which the so-called inverted-band structure mani- fests itself.