Intrinsic Spin Hall Effect Induced by Quantum Phase Transition in HgCdTe Quantum Wells
aa r X i v : . [ c ond - m a t . m e s - h a ll ] N ov Intrinsic Spin Hall Effect Induced by Quantum Phase Transition in HgCdTe QuantumWells
Wen Yang and Kai Chang ∗ State Key Laboratory for Superlattices and Microstructures, Institute of Semiconductors,Chinese Academy of Sciences, P. O. Box 912, 100083, Beijing, China
Shou-Cheng Zhang
Department of Physics, McCullough Building, Stanford University, Stanford, CA 94305-4045, USA
Spin Hall effect can be induced both by the extrinsic impurity scattering and by the intrinsicspin-orbit coupling in the electronic structure. The HgTe/CdTe quantum well has a quantum phasetransition where the electronic structure changes from normal to inverted. We show that the intrinsicspin Hall effect of the conduction band vanishes on the normal side, while it is finite on the invertedside. This difference gives a direct mechanism to experimentally distinguish the intrinsic spin Halleffect from the extrinsic one.
Spin-polarized transport in nonmagnetic semiconduc-tors is a crucial ingredient for realizing spintronicdevices.[1] The spin Hall effect (SHE) opens up thepromising prospect of generating spin currents in conven-tional semiconductors without applying external mag-netic field or introducing ferromagnetic elements. Re-cently the previously predicted extrinsic SHE (ESHE)[2]and the newly discovered intrinsic SHE (ISHE)[3] havebecome one of the most intensively studied subjects.The experimental observations of SHE have been re-ported by two groups[4, 5] in n -type epilayers andtwo-dimensional electron and hole gases, although theirtheoretical interpretation as extrinsic or intrinsic arestill ambiguous.[6, 7, 8] The ISHE in the 2D hole gashas no vertex correction,[6], and its existence has beenwidely accepted,[9] the existence of electron ISHE in two-dimensional systems is under substantial debate.[10, 11,12, 13, 14, 15] The current understanding is that the elec-tron ISHE in the ideal model (single-band Hamiltonianwith parabolic dispersion and linear Rashba and/or Dres-selhaus spin splitting) is exactly cancelled by the impu-rity induced vertex corrections in the clean limit,[10, 11]even for momentum dependent scattering.[12, 13, 14]Very recently quantum spin Hall effect was predictedtheoretically and observed experimentally in a narrow-gap HgTe quantum well with the unique inverted bandstructure[20]. The HgTe quantum well has a quantumphase transition when the quantum well thickness d istuned across a critical thickness d c ≈ nm . For d < d c ,the electronic structure is normal, similar to the GaAsquantum wells, where the conduction band has Γ char-acter, and the valence band Γ character. In this regime,we show that the ISHE vanishes in the conduction banddue to vertex corrections, consistent with previous re-sults. For d > d c , the electronic structure is inverted,where the conduction and the valence bands interchangetheir Γ − Γ characters. In this regime, we show thatthe ISHE is finite in the conduction band. Since the im-purity configuration is not expected to change drastically across d c , the difference of the SHE across the d c tran-sition therefore singles out the ISHE contribution. Thismechanism solves a long standing challenge of how todistinguish the ISHE from the ESHE.First we develop a unified description of Γ -electronand Γ -hole SHE based on a general N -band effective-mass theory, which remains valid over the whole rangeof Γ -Γ coupling strengths and bandgaps. Then wetake the N =8 model (the Kane model) to make a realis-tic calculation of the ISHE in CdTe/Cd x Hg − x Te quan-tum wells, taking into account the non-ideal factors in aself-consistent way and the impurity scattering inducedvertex corrections through standard diagrammatic tech-niques. The calculated ISHE agrees with previous theo-ries in the limit of weak Γ -Γ coupling, while it showsnontrivial behaviors in the strong coupling regime. Itexhibits a large [3 ∼ σ = e/ (8 π )] abrupt increase accompanying theΓ -Γ phase transition in the lowest conduction band.This large ISHE is robust against impurity scattering in-duced vertex corrections. By varying the well width orthe electric bias across the quantum well, we can switchthe electron ISHE on/off or even tune it into resonance.These operations can be realized in experimentally acces-sible conditions and they may be utilized to distinguishthe electron ISHE from the ESHE.Following the new envelope function approach,[21] theband-edge Bloch basis { Φ µ } is classified into N rel-evant bands { Φ j } and infinite irrelevant bands { Φ l } .In the N -dimensional { Φ j } subspace, the image of theHamiltonian H for a general microstructure is H jj ′ = H jj ′ + P l H jl ( E − E l ) − H lj ′ .[21] The image of an arbi-trary operator O ( = H ) can also be obtained as O jj ′ = O jj ′ + X l ( O jl E − E l H lj ′ + H jl E − E l O lj ′ ) , (1)where H and O are, respectively, the image of H and O in the { Φ µ } space. Then the images of velocity v , spin s , and spin current j βα ≡ ( v α s β + s β v α ) / α, β = x, y, z )operators are given by ~ V jj ′ = [ r , H jj ′ ] / ( i ~ ), ~ S jj ′ = h Φ j | s | Φ j ′ i , and J βα = ( V α S β + S β V α ) / -Γ coupling taken into account, they generalize theprevious theories (which neglect this coupling) to the N -band case, e.g., the widely used single-band (four-bandLuttinger-Kohn) model corresponds to N =2 ( N =4). Weemphasize that explicit consideration of the Γ -Γ cou-pling is important in determining electron ISHE, espe-cially for strong Γ -Γ coupled systems. Further, thedifferent non-ideal band structure factors arise from thesame origin (Γ -Γ coupling), so they are not indepen-dent and should be incorporated self-consistently throughexplicit consideration of the Γ -Γ coupling. We no-tice that the equation-of-motion argument[13] (valid for N =2) for the nonexistence of electron ISHE is not appli-cable to other values of N (e.g., N =4 ,
6, or 8).The above theory can be applied to study both ISHEand ESHE in a general microstructure. In the presentwork we consider ISHE only, due to its much larger mag-nitude compared to ESHE,[7, 8] especially for small elec-tron density. The linear response spin Hall conductiv-ity σ SH = e/ ( ~ A ) lim ω → (cid:2) G zxy ( ω ) − G zxy (0) (cid:3) / ( iω ), with A the sample area and G zxy ( ω ) the impurity-averaged re-tarded correlation function of J zy and V x . Using standarddiagrammatic perturbation theory, G zxy ( ω ) is evaluatedtaking into account the impurity induced self-energy cor-rections in the self-consistent Born approximation andvertex corrections in the ladder approximation (inset ofFig. 1), yielding σ SH = eπ Z ∞−∞ dω f ( ω ) lim η → + Re (cid:20) ∂P ( ω ′ + iη, ω + iη ) ∂ω ′ − ∂P ( ω ′ + iη, ω − iη ) ∂ω ′ (cid:21) ω ′ = ω , where f ( ω ) = 1 / [ e ( ~ ω − µ ) / ( k B T ) + 1], P ( z, z ′ ) =(1 / A ) Tr J zy G ( z )Γ( z, z ′ ) G ( z ′ ) , z = iω m , z ′ = iω n , G ij ( z )and Γ ij ( z, z ′ ) are, respectively, matrix elements ofthe impurity-averaged Matsubara Green’s function anddressed velocity vertex in the eigenstate basis of H . Theycan be calculated from the Dyson equation and the ver-tex equationΓ( z, z ′ ) = V x + n I R d R U ( R ) ~ G ( z )Γ( z, z ′ ) G ( z ′ ) U ( R ) ~ , where n I is the impurity concentration, and U ij ( R ) = h i | V C ( r − R ) | j i is the matrix element of the single-impurity potential V C ( r ).Now we consider the lattice-matched symmetricCdTe/Cd x Hg − x Te quantum well under electric bias.Its bandgap can be tuned in a large range by varyingthe Cd content x , the well width W , or the bias elec-tric field F , serving as an ideal workbench for study-ing ISHE under various Γ -Γ coupling strengths. Forsuch narrowgap systems, the N =8 Kane model is a good -400 -395 -390 -385 0 20 40 60 80-3-2-10123 0.0 0.1 0.2 0.3 0.4-420-410-400-390020406080 * == +(1) Self-consistent Born approx.(2) Ladder approx. = + v x * S p i n H a ll c ondu c t i v i t y S H / Fermi energy (meV)
HH1 E1 (a) E2 E ne r g y ( m e V ) k // (nm -1 ) (b) E1E2HH1HH2LH1 W=25 nmF=40 kV/cm
FIG. 1: (color online) (a) σ SH for W =25 nm and F =40 kV/cmwith (solid lines) or without (dashed lines) vertex corrections.Inset: (1) Dyson equation in the self-consistent Born approx-imation and (2) vertex equation in the ladder approximation.(b) Corresponding energy spectrum. starting point. It incorporates the aforementioned non-ideal factors non-perurbatively and self-consistently. TheDresselhaus spin-orbit coupling is neglected because itis much smaller than the Rashba effect in a narrow-gap quantum well,[23] as verified by the quantitativeagreement between theory and experiment in recent in-vestigations on the transport properties of CdTe/HgTequantum wells.[24] We also adopt the widely employedaxial approximation (good for electrons and reasonablefor holes in narrowgap systems) and short-range impu-rity potential V C ( r ) = V δ ( r ). All band parametersused in our numerical calculation are experimentally de-termined values.[24, 25] We take the temperature T =0K and, unless specified, the effective disorder strength ξ ( ≡ n I V ) = 6 . ˚A , corresponding to typical elec-tron (hole) self-energy broadening 0.1 meV (1 meV) andcollisional lifetime 6 ps (0.6 ps).First we consider the weak Γ -Γ coupling case x =0 . . Cd . Te) ≈ σ ) at theedges of the first (E1) and second (E2) conduction bands.Such behavior is greatly suppressed by the inclusion ofvertex corrections, in sharp contrast to the hole ISHE,which has an opposite sign and remains largely unaffectedby vertex corrections. Therefore in the weak couplingregime, the results of the previous theories are recovered.To explore electron ISHE in the strong couplingregime, we consider the CdTe/HgTe quantum well cor-responding to x =0. Due to the abnormal positions andeffective masses of the Γ electron and Γ light-hole bandsin the HgTe layer, the bandgap E Γ g of the quantum wellat k k =0 can be tuned by varying the well width or theelectric bias. Fig. 2(a) shows that the derivative of E Γ g is discontinuous at W ≈ , ,
24, and 28 . S p i n H a ll c ondu c t i v i t y S H / Fermi energy (meV) (c)
W=6 nmW=8 nm E g ( m e V ) Well width (nm) (a)
CdTe/HgTeF=50 kV/cm W e ll w i d t h ( n m ) Bias electric field (kV/cm)
E1-HH1HH1-E1HH1-HH2 E2-HH2HH2-E2 (b) n e ( c m - ) Fermi energy (meV)
W=6 nm W=8 nm
FIG. 2: (color online) (a) E Γ g for F =50 kV/cm. (b) Band-edge ( k k =0) phase diagram of the lowest conduction bandand highest valence band. (c) σ SH for F =50 kV/cm, W =6and 8 nm, respectively [indicated by filled circles in (a) and(b)]. Solid (Dashed) lines correspond to ξ = 6 . ˚A .Inset: electron density vs. Fermi energy. certain phase transitions. Actually, the first critical pointat W ≈ → HH1-E1.[26] Namely, the lowestconduction (highest valence) band changes from E1 toHH1 (HH1 to E1), where E (HH) denote Γ electron (Γ heavy-hole) states. Other critical points corresponds tosimilar transitions [Fig. 2(b)]. They manifest the red(blue) shift of electron states E2, E3, · · · (heavy-holestates HH1, HH2, · · · ) with increasing well width/electricbias due to weakening of the confinement/quantum con-fined Stark effect. From Fig. 2(c), we see that in the E1-HH1 phase, σ SH arising from the lowest conduction band(E1) is largely cancelled by vertex corrections, especiallyfor small Fermi energy. In contrast, in the HH1-E1 phase,the lowest conduction band (HH1) takes on pure Γ sym-metry at small wave vectors and its contribution to σ SH is largely unaffected,[27] leading to the abrupt increaseof σ SH accompanying the phase transition from E1-HH1to HH1-E1. This phase transition induced ISHE is ro-bust against impurity induced vertex corrections since itvaries only slightly when ξ is increased by an order ofmagnitude, i.e., when typical electron lifetime [mobility]decreases from 6 to 0.6 ps [3 × to 3 × cm / (Vs)]. By changing the well width, large electron ISHE canbe switched on/off, especially for small Fermi energy orelectron density [inset of Fig. 2(c)].In the above, the phase transition occurs at small crit-ical well width and the electric bias plays a minor role.When the critical well width increases, the bias electricfield induced quantum-confined Stark effect would be-come strong enough to induce the phase transition E1-HH1 → HH1-E1 and control the appearance of large elec-tron ISHE. To demonstrate this, we consider the case x =0 .
16 with Eg(Hg . Cd . Te) ≈
0. For W =25 nm, the S p i n H a ll c ondu c t i v i t y S H / Fermi energy (meV) (c)
F=90 kV/cmF=60 kV/cm W e ll w i d t h ( n m ) Bias electric field (kV/cm) E - HH HH - E HH - HH E - HH (b) E g ( m e V ) Bias electric field (kV/cm) (a)
CdTe/Hg Cd TeW=25 nm n e ( c m - ) Fermi energy (meV)
F=60 kV/cm F=90 kV/cm
FIG. 3: (color online) (a) E Γ g for W =25 nm. (b) Band-edge( k k =0) phase diagram of the lowest conduction band andhighest valence band. (c) σ SH for W =25 nm, F =60 and 90kV/cm, respectively [indicated by filled circles in (a) and (b)].Solid (Dashed) lines correspond to ξ =6 . ˚A . Inset:electron density vs. Fermi energy. bandgap E Γ g ≈
60 meV at F =0 and decreases to zeroat F ≈
75 kV/cm [Fig. 3(a)]. The discontinuities ofits derivative at F ≈ , σ SH in Fig. 3(c) shows a large increase whenthe bias electric field is tuned across the critical point.Again, the slight dependence on the disorder strength ξ manifests the robustness of the ISHE against impu-rity induced vertex corrections. The field-induced phasetransition provides a dynamic way to switch on/off theelectron ISHE, especially for small Fermi energy or elec-tron density [inset of Fig. 3(c)].Turning back to CdTe/HgTe quantum wells, Fig. 2(a)shows that the electric bias can induce the transition HH1 → E2 in the lowest conduction band or, equivalently, thetransition E2 → HH1 in the second conduction band [Fig.4(a)]. In the E2 phase, the Rashba spin splitting betweenthe two branches of the second conduction band reversesits sign at a critical wave vector k [Fig. 4(b)]. Anal-ysis shows that this behavior comes from the couplingbetween the two branches and the interface states,[28]thus it does not exist in the HH1 phase. By varying thewell width or electric bias, such behavior can be switchedon/off [Fig. 4(a)] and the critical wave vector [gray scalemap in Fig. 4(a)] or critical electron density [inset of Fig.4(c)] can be tuned, offering us the possibility to manipu-late the ISHE arising from the second conduction band.Indeed, σ SH in Fig. 4(c) exhibits a resonance when theFermi energy coincides with the spin degeneracy point.We notice that although such level-crossing inducedresonance has been predicted for the widely accepted holeISHE in p -type GaAs quantum wells (based on calcula-tions that neglect vertex corrections),[29] similar predic-tion for the much debated n -type systems still remains S p i n H a ll c ondu c t i v i t y S H / Fermi energy (meV) (c) =6.2 eV ¯ =31 eV ¯ =62 eV ¯ HH1 E2+LH1
Bias electric field (kV/cm) W e ll w i d t h ( n m ) E2HH1 E3 (a) -3.0000.100.200.28 k (nm -1 ) S p i n s p li tt i ng ( m e V ) k // (nm -1 )(b) CdTe/HgTeF=30 kV/cm k n e ( c m - ) F (kV/cm)
FIG. 4: (color online) (a) Band-edge ( k k =0) phase diagram(the gray scale map in E2 phase indicates k ) and (b) Rashbaspin splitting (at W =25 nm, F =30 kV/cm) of the secondconduction band. (c) σ SH for W =25 nm, F =30 kV/cm [indi-cated by the filled circle in (a)] and different disorder strength ξ . Inset: critical electron density for W =25 (solid line), 20(dashed line), and 15 (dotted line) nm. absent. For hole ISHE, a challenging hole lifetime & µ p & cm /(V s) is required to ob-serve the resonance.[29] For electron ISHE, the require-ment is significantly relaxed to electron lifetime & ξ .
20 eV ˚A , cf. Fig. 4(c)] or elec-tron mobility µ n & × cm /(V s). These have alreadybeen realized in previous experiments, e.g., µ n =3 . × cm / (V s) for W =7 . µ n =3 . × cm / (Vs) for W =21 nm[24] (close to the well width used in ourcalculation).In summary, we have investigated the electron ISHEin narrowgap HgCdTe quantum wells based on a unifieddescription for electron and hole ISHE. While the ISHEof the conduction band vanishes on the normal side of theΓ -Γ phase transition, a ISHE in the conduction bandcan be generated on the inverted side. It is robust againstimpurity induced vertex corrections. By changing the Cdcontent, the well width, or the bias electric field, we canswitch the ISHE on/off or tune it into resonance underexperimentally accessible conditions. Ref. [31] showsthat the spin Hall effect can be experimentally observedby the non-local transport measurements in mesoscopicsystems. We propose to carry out such measurement forboth the normal and inverted quantum wells, both closeto the transition. The difference uniquely singles out theISHE contribution.This work is supported by the NSFC Grant No.60525405, the knowledge innovation project of CAS, theNSF under grant numbers DMR-0342832 and the US De-partment of Energy, Office of Basic Energy Sciences un-der contract DE-AC03-76SF00515. ∗ Electronic address:[email protected][1] S. A. Wolf et al ., Science , 1488 (2001).[2] M. I. D’yakonov and V. I. Perel’, Phys. Lett. A , 459(1971); J. E. Hirsch, Phys. Rev. Lett. , 1834 (1999).[3] S. Murakami, N. Nagaosa, and S. C. Zhang, Science ,1348 (2003); J. Sinova et al ., Phys. Rev. Lett. , 126603(2004).[4] Y. K. Kato, R. C. Myers, A. C. Gossard, and D. D.Awschalom, Science , 1910 (2004); V. Sih et al ., Na-ture Phys. , 31 (2005); V. Sih et al. , Phys. Rev. Lett. , 096605 (2006); N. P. Stern et al ., Phys. Rev. Lett. , 126603 (2006).[5] J. Wunderlich, B. Kaestner, J. Sinova, and T. Jungwirth,Phys. Rev. Lett. , 047204 (2005).[6] B. A. Bernevig and S. C. Zhang, Phys. Rev. Lett. ,016801 (2005).[7] H. A. Engel, B. I. Halperin, and E. I. Rashba, Phys. Rev.Lett. , 166605 (2005).[8] W. K. Tse and S. Das Sarma, Phys. Rev. Lett. , 056601(2006).[9] J. Sinova, S. Murakami, S. Q. Shen, and M. S. Choi, SolidState Commun. , 214 (2006); and references therein.[10] E. G. Mishchenko, A. V. Shytov, and B. I. Halperin,Phys. Rev. Lett. , 226602 (2004); J. Inoue, G. E. W.Bauer, and L. W. Molenkamp, Phys. Rev. B, , 033104(2003); , 041303(R) (2004); D. N. Sheng, L. Sheng, Z.Y. Weng, and F. D. M. Haldane, ibid . , 153307 (2005).[11] K. Nomura, J. Sinova, N. A. Sinitsyn, and A. H. Mac-Donald, Phys. Rev. B , 165316 (2005).[12] E. I. Rashba, Phys. Rev. B , 201309(R) (2004); R.Raimondi and P. Schwab, ibid . , 033311 (2005).[13] O. Chalaev and D. Loss, Phys. Rev. B , 245318 (2005);O. V. Dimitrova, ibid . , 245327 (2005).[14] A. Khaetskii, Phys. Rev. Lett. , 056602 (2006).[15] A. G. Mal’shukov and K. A. Chao, Phys. Rev. B ,121308(R) (2005); A. V. Shytov, E. G. Mishchenko, H.A. Engel, and B. I. Halperin, ibid . , 075316 (2006);C. M. Wang, X. L. Lei, and S. Y. Liu, ibid . , 113314(2006); P. L. Krotkov and S. Das Sarma, ibid . , 195307(2006).[16] D. F. Nelson, R. C. Miller, and D. A. Kleinman, Phys.Rev. B , 7770 (1987).[17] W. Yang and K. Chang, Phys. Rev. B , 113303 (2006); , 193314 (2006).[18] ˙I. Adagideli and G. E. W. Bauer, Phys. Rev. Lett. ,256602 (2005).[19] L. Sheng, D. N. Sheng, and C. S. Ting, Phys. Rev. Lett. , 016602 (2005); W. Ren et al ., ibid . , 066603 (2006);Z. H. Qiao, W. Ren, J. Wang, and Hong Guo, ibid . ,196402 (2007).[20] B. A. Bernevig, T. L. Hughes, and S. C. Zhang, Sciences , 1757 (2006); M. K¨onig, Science Express 1148047(2007).[21] M. G. Burt, J. Phys.: Condens. Matter , 6651 (1992).[22] In obtaining ~ S and J βα , we have neglected the secondterm in Eq. (1), which is smaller than the first termby a factor H jl / ( E − E l ) ≪
1. This approximation be-comes exact when { Φ j } and { Φ l } are chosen such that h Φ j | s | Φ l i = 0, e.g., for N = 2 (single-band model), 6(six-band Luttinger-Kohn model), and 8 (Kane model). [23] G. Lommer, F. Malcher, and U. R¨ossler, Phys. Rev. Lett. , 728 (1988).[24] X. C. Zhang et al ., Phys. Rev. B , 245305 (2001); K.Ortner et al ., ibid . , 075322 (2002).[25] II-VI and I-VII Compounds; Semimagnetic Compounds ,Landolt-B¨ornstein, Group III, Vol. 41B, ed. U. R¨ossler(Springer-Verlag, Berlin, 1999).[26] N. F. Johnson, P. M. Hui, and H. Ehrenreich, Phys. Rev.Lett. , 1993 (1988).[27] Based on a modified single-band Rashba model insteadof the correct heavy-hole Hamiltonian, the conclusion of Ref. [11] differs from ours. We believe that in the in-verted band phase, single-band results should be takenwith cautious due to the strong Γ -Γ coupling.[28] Y. R. Lin-Liu and L. J. Sham, Phys. Rev. B , 5561(1985).[29] X. Dai, Z. Fang, Y. G. Yao, and F. C. Zhang, Phys. Rev.Lett. , 086802 (2006).[30] J. R. Meyer et al ., Phys. Rev. B , 2204 (1988).[31] E. M. Hankiewicz, L. W. Molenkamp, T. Jungwirth, andJairo Sinova, Phys. Rev. B70