PHYSICAL REVIEW LETTERS week ending

PRL 95, 226801 (2005) 25 NOVEMBER 2005

Quantum Spin Hall Effect in Graphene

C. L. Kane and E. J. Mele
Dept. of Physics and Astronomy, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA
(Received 29 November 2004; published 23 November 2005)
We study the effects of spin orbit interactions on the low energy electronic structure of a single plane of
graphene. We find that in an experimentally accessible low temperature regime the symmetry allowed spin
orbit potential converts graphene from an ideal two-dimensional semimetallic state to a quantum spin Hall
insulator. This novel electronic state of matter is gapped in the bulk and supports the transport of spin and
charge in gapless edge states that propagate at the sample boundaries. The edge states are nonchiral, but
they are insensitive to disorder because their directionality is correlated with spin. The spin and charge
conductances in these edge states are calculated and the effects of temperature, chemical potential, Rashba
coupling, disorder, and symmetry breaking fields are discussed.

DOI: 10.1103/PhysRevLett.95.226801 PACS numbers: 73.43.
f, 72.25.Hg, 73.61.Wp, 85.75.
d

The generation of spin current solid state systems has Unlike graphite which has a finite Fermi surface, however,
been a focus of intense recent interest. It has been argued graphene is in a critical electronic state which can be
that in doped semiconductors the spin orbit (SO) interac- strongly affected by small perturbations at low energy.
tion leads to a spin Hall effect [1,2], in which a spin current Graphene consists of a honeycomb lattice of carbon
flows perpendicular to an applied electric field. The spin atoms with two sublattices. The states near the Fermi
Hall effect has been observed in GaAs [3,4]. Murakami energy are
orbitals residing near the K and K 0 points at
et al. [5] have identified a class of cubic materials that are opposite corners of the hexagonal Brillouin zone. An ef-
insulators, but nonetheless exhibit a finite spin Hall con- fective mass model can be developed [9] by writing the low
ductivity. Such a ‘‘spin Hall insulator’’ would be of intrin- energy electronic wavefunctions as
sic interest, since it would allow for spin currents to be
generated without dissipation.
r  uAK ; uBK ; uAK0 ; uBK0  r (1)
In this Letter we show that at sufficiently low energy a
single plane of graphene exhibits a quantum spin Hall where uA;BK;K0  r describe basis states at momentum k 
(QSH) effect with an energy gap that is generated by the K, K 0 centered on atoms of the A, B sublattice. r is a
SO interaction. Our motivation is twofold. First, four component slowly varying envelope function. The
Novoselov et al. [6] have recently reported progress in effective mass Hamiltonian then takes the form,
the preparation of single layer graphene films. These films y   @
exhibit the expected ambipolar behavior when gated and H 0 
i@vF x z x  y @y  : (2)
have mobilities up to 104 cm2 =V s. Thus, the detailed ex-
perimental study of graphene now appears feasible. We Here ~ and ~ are Pauli matrices with z  1 describing
believe the QSH effect in graphene is observable below a states on the AB sublattice and z  1 describing states
low but experimentally accessible temperature. Secondly, at the KK 0  points. This Hamiltonian describes gapless
we will show the QSH effect in graphene is different from states with Eq  vF jqj. Without spin, the degeneracy
the spin Hall effects studied for three-dimensional cubic at q  0 is protected by symmetry. The only possible terms
systems in Ref. [5] because it leads to a phase which is that could be added to open a gap are proportional to z or
topologically distinct from a band insulator. The QSH z z . The z term, which corresponds to a staggered
effect in graphene resembles the charge quantum Hall sublattice potential is odd under parity (which interchanges
effect, and we will show that spin and charge currents the A and B sublattices). The z z term is even under
can be transported in gapless edge states. As a model parity, but odd under time reversal (which interchanges K
system, graphene thus identifies a new class of spin Hall and K 0 ).
insulator. It may provide a starting point for the search for The SO interaction allows for a new term, which will be
other spin Hall insulators in two-dimensional or in layered the focus of this Letter:
materials with stronger SO interaction. H  so y
z z sz : (3)
SO
SO effects in graphite have been known for over 40
years [7], and play a role in the formation of minority Here sz is a Pauli matrix representing the electron’s spin.
hole pockets in the graphite Fermi surface [8]. However, This term respects all of the symmetries of graphene, and
these effects have largely been ignored because they are will be present. Below we will explicitly construct this
predicted to be quite small and they are overwhelmed by term from the microscopic SO interaction and estimate
the larger effect of coupling between the graphene planes. its magnitude. If the mirror symmetry about the plane is

0031-9007=05=95(22)=226801(4)$23.00 226801-1 © 2005 The American Physical Society

 PHYSICAL REVIEW LETTERS week ending
PRL 95, 226801 (2005) 25 NOVEMBER 2005

preserved then this is the only allowed spin dependent term R
0, though the correction to (5) is small due to car-
at q  0. If the mirror symmetry is broken (either by a bon’s weak SO interaction.
perpendicular electric field or by interaction with a sub- In the quantum Hall effect the bulk topological order
strate) then a Rashba term [10] of the form s p z^ is requires the presence of gapless edge states. We now show
allowed, that gapless edge states are also present in graphene. We
will begin by establishing the edge states for R  0. We
y
H R  R x z sy
y sx  : (4) will then argue that the gapless edge states persist even
when R
0, and that they are robust against weak
For R  0, so leads to an energy gap 2so with Eq  electron-electron interactions and disorder. Thus, in spite
p





























of the violation of (5) the gapless edge states characterize a
 @vF q2  2so . For 0 < R < so the energy gap
2so
R  remains finite. For R > so the gap closes, state which is distinct from an ordinary insulator. This QSH
and the electronic structure is that of a zero gap semicon- state is different from the insulators discussed in Ref. [5],
ductor with quadradically dispersing bands. In the follow- which do not have edge states. It is also distinct from the
ing we will assume that R < so and analyze the spin Hall effect in doped GaAs, which does not have an
properties of the resulting gapped phase. This assumption energy gap.
is justified by numerical estimates given at the end of the For R  0, the Hamiltonian (2) and (3) conserves sz ,

ment [15]. Consider a large cylinder (larger than @vF =so )

Letter. and the gapless edge states follow from Laughlin’s argu-
The gap generated by z z sz is different from the gap
that would be generated by the staggered sublattice poten- and adiabatically insert a quantum   h=e of magnetic
tials, z or z sz . The ground states in the presence of the flux quantum down the cylinder (slower than so =@). The

current such that spin @ is transported from one end of the

latter terms are adiabatically connected to simple insulat- resulting azimuthal Faraday electric field induces a spin
ing phases at strong coupling where the two sublattices are
decoupled. In contrast, the gap parameter z z sz produces cylinder to the other. Since an adiabatic change in the
gaps with opposite signs at the K and K 0 points. This has no magnetic field cannot excite a particle across the energy
simple strong coupling limit. To connect smoothly between gap so it follows that there must be gapless states at each
the states generated by z and z z sz one must pass end to accommodate the extra spin.
through a critical point where the gap vanishes, separating An explicit description of the edge states requires a
ground states with distinct topological orders. model that gives the energy bands throughout the entire
The interaction (3) is related to a model introduced Brillouin zone. Following Haldane [11], we introduce a
by Haldane [11] as a realization of the parity anomaly in second neighbor tight binding model,
2  1-dimensional relativistic field theory. Taken sepa- X y X
rately, the Hamiltonians for the sz  1 spins violate time H  tci cj  it2 ij sz cyi cj : (6)
hiji hhijii
reversal symmetry and are equivalent to Haldane’s model
for spinless electrons, which could be realized by introduc- The first term is the usual nearest neighbor hopping term.
ing a periodic magnetic field with no net flux. As Haldane The second term connects second neighbors with a spin
showed, this gives rise to a z z gap, which has opposite dependent amplitude. ij 
ji  1, depending on the
signs at the K and K 0 points. At temperatures well below
orientation of the two nearest neighbor bonds d1 and d2 the
the energy gap this leads to a quantized Hall conductance
electron traverses in going from site j to i. ij  1 (
1)
xy  e2 =h. This Hall conductance computed by the
if the electron makes a left (right) turn to get to the second
Kubo formula can be interpreted as the topological Chern
bond. The spin dependent term can be written in a coor-
number induced by the Berry’s curvature in momentum
dinate independent representation as id1 d2 p


s. At low
space [12,13]. Since the signs of the gaps in (3) are
opposite for opposite spins, an electric field will induce energy (6) reduces to (2) and (3) with so  3 3t2 .
opposite currents for the opposite spins, leading to a spin The edge states can be seen by solving (7) in a strip
current Js  @=2eJ"
J#  characterized by a quantized geometry. Figure 1 shows the one-dimensional energy
bands for a strip where the edges are along the zigzag
spin Hall conductivity
direction in the graphene plane. The bulk band gaps at
e the one-dimensional projections of the K and K 0 points are
sxy  : (5) clearly seen. In addition two bands traverse the gap, con-
2

necting the K and K 0 points. These bands are localized at
Since spin currents do not couple to experimental probes it the edges of the strip, and each band has degenerate copies
is difficult to directly measure (5). Moreover, the conser- for each edge. The edge states are not chiral since each
vation of sz will be violated by the Rashba term (4) as well edge has states which propagate in both directions.
as terms which couple the
and  orbitals. Nonetheless, However, as illustrated in Fig. 2 the edge states are ‘‘spin
Murakami et al. [14] have defined a conserved spin szc , filtered’’ in the sense that electrons with opposite spin
allowing sxy to be computed via the Kubo formula. We propagate in opposite directions. Similar edge states occur
find that sxy computed in this way is not quantized when for armchair edges, though in that case the 1D projections
226801-2
 PHYSICAL REVIEW LETTERS week ending
PRL 95, 226801 (2005) 25 NOVEMBER 2005

1 actions and disorder distinguishes the spin filtered edge

states from ordinary one-dimensional wires, which are
E/t localized by weak disorder.
A parallel magnetic field Hk breaks time reversal and
leads to an avoided crossing of the edge states. Hk also
0 reduces the symmetry, allowing terms in the Hamiltonian
X which provide a continuously gapped path connecting the
states generated by z z sz and z . Thus in addition to
gapping the edge states Hk eliminates the topological
distinction between the QSH phase and a simple insulator.
-1 The spin filtered edge states have important consequen-
0 π/a kX 2π/a ces for both the transport of charge and spin. In the limit of
low temperature we may ignore the inelastic backscatter-
FIG. 1. (a) One-dimensional energy bands for a strip of gra- ing processes, and describe the ballistic transport in the
phene (shown in inset) modeled by (7) with t2 =t  0:03. The edge states within a Landauer-Büttiker [19] framework.
bands crossing the gap are spin filtered edge states.
For a two terminal geometry [Fig. 2(a)], we predict a
ballistic two terminal charge conductance G  2e2 =h.
of K and K 0 are both at k  0. It is interesting to note that For the spin filtered edge states the edge current density
for zigzag edges the edge states persist for so ! 0, where is related to the spin density, since both depend on nR"

they become perfectly flat [16]. This leads to an enhanced nL# . Thus the charge current is accompanied by spin accu-
density of states at the Fermi energy associated with zigzag mulation at the edges. The interplay between charge and
edges. This has been recently seen in scanning tunneling spin can be probed in a multiterminal device.P Define the
spectroscopy of graphite surfaces [17]. multiterminal spin conductance by Iis  j Gsij Vj . Time
We have also considered a nearest neighbor Rashba reversal symmetry requires Gsji 
Gsij , and from
term, of the form iz^ s dcyi cj . This violates the Fig. 2(b) it is clear that Gsij  e=4
for adjacent contacts
conservation of sz , so that the Laughlin argument no longer i and j. In the four terminal geometry of Fig. 2(b) a spin
applies. Nonetheless, we find that the gapless edge states current I s  eV=4
flows into the right contact. This
remain, provided R < so , so that the bulk band gap geometry can also be used to measure a spin current. A
remains intact. The crossing of the edge states at the spin current incident from the left (injected, for instance,
Brillouin zone boundary kx 
=a in Fig. 1 (or at k  0
for the armchair edge) is protected by time reversal sym-
metry. The two states at kx 
=a form a Kramers doublet
whose degeneracy cannot be lifted by any time reversal
symmetric perturbation. Moreover, the degenerate states at
kx 
=a  q are a Kramers doublet. This means that V I 0
elastic backscattering from a random potential is forbid-
den. More generally, scattering from a region of disorder
can be characterized by a 2 2 unitary S matrix which
(a)
relates the incoming and outgoing states: out  Sin ,
where  is a two component spinor consisting of the left
and right moving edge states L" , R# . Under time reversal V/2
in;out ! sy out;in . Time reversal symmetry therefore im-
poses the constraint S  sy ST sy , which rules out any off
diagonal elements.
Electron interactions can lead to backscattering. For 0 Is 0
y y
instance, the term u L" @x L" R# @x R# , does not violate
time reversal, and will be present in an interacting
Hamiltonian. For weak interactions this term is irrelevant
under the renormalization group, since its scaling dimen- -V/2
sion is   4. It thus will not lead to an energy gap or to
localization. Nonetheless, it allows inelastic backscatter- (b)
ing. To leading order in u it gives a finite conductivity of FIG. 2. Schematic diagrams showing (a) two terminal and
the edge states, which diverges at low temperature as (b) four terminal measurement geometries. In (a) a charge
u
2 T 3
2 [18]. Since elastic backscattering is prevented current I  2e2 =h V flows into the right lead. In (b) a spin
by time reversal there are no relevant backscattering pro- current I s  e=4
 V flows into the right lead. The diagrams to
cesses for weak interactions. This stability against inter- the right indicate the population of the edge states.
226801-3
 PHYSICAL REVIEW LETTERS week ending
PRL 95, 226801 (2005) 25 NOVEMBER 2005

with a ferromagnetic contact) will be split, with the up

(down) spins transported to the top (bottom) contacts,
generating a measurable spin Hall voltage. σ z τz s z
The magnitude of so may be estimated by treating the
microsopic SO interaction
@
FIG. 3. Feynman diagram describing the renormalization of
the SO potential by the Coulomb interaction. The solid line
VSO  s rV p (7) represents the electron propagator and the wavy line is the
4m2 c2
Coulomb interaction.
in first order degenerate perturbation theory. We thus
evaluate the expectation value of (8) in the basis of states
given in (1) treating r as a constant. A full evaluation In summary, we have shown that the ground state of a
depends on the detailed form of the Bloch functions. How- single plane of graphene exhibits a QSH effect, and has a
ever a simple estimate can be made in the ‘‘first star’’ nontrivial topological order that is robust against small
P p


perturbations. The QSH phase should be observable by
approximation: uK;K0 ;A;B r  p expiKp r
d= 3.
studying low temperature charge transport and spin injec-
Here Kp are the crystal momenta at the three corners of the tion in samples of graphene with sufficient size and purity
Brillouin zone equivalent to K or K 0 , and d is the a basis to allow the bulk energy gap to manifest itself. It would
vector from a hexagon center to an A or B sublattice site. also be of interest to find other materials with stronger SO
We find that the matrix elements have precisely the struc- coupling which exhibit this effect, as well as possible
ture (3), and using the Coulomb interaction Vr  e2 =r we
estimate 2so  4
2 e2 @2 =3m2 c2 a3  2:4 K. This is a
three-dimensional generalizations.
We thank J. Kikkawa and S. Murakami for helpful
crude estimate, but it is comparable to the SO splittings discussions. This work was supported by the NSF under
quoted in the graphite literature [8]. MRSEC Grant No. DMR-00-79909 and the DOE under
field Ez may be estimated as R  @vF eEz =4mc2 . For
The Rashba interaction due to a perpendicular electric Grant No. DE-FG02-ER-0145118.
Ez 50 V=300 nm [3] this gives R 0:5 mK. This is
smaller than so because Ez is weaker than the atomic
scale field. The Rashba term due to interaction with a [1] S. Murakami, N. Nagaosa, and S. C. Zhang, Science 301,
substrate is more difficult to estimate, though since it is 1348 (2003).
presumably a weak Van der Waals interaction, this too can [2] J. Sinova et al., Phys. Rev. Lett. 92, 126603 (2004).
be expected to be smaller than so . [3] Y. K. Kato et al., Science 306, 1910 (2004).
[4] J. Wunderlich et al., Phys. Rev. Lett. 94, 047204 (2005).
This estimate of so ignores the effect of electron-
[5] S. Murakami, N. Nagaosa, and S. C. Zhang, Phys. Rev.
electron interactions. The long range Coulomb interaction Lett. 93, 156804 (2004).
may substantially increase the energy gap. To leading order [6] K. S. Novoselov et al., Science 306, 666 (2004).
the SO potential is renormalized by the diagram shown in [7] G. Dresselhaus and M. S. Dresselhaus, Phys. Rev. 140,
Fig. 3, which physically represents the interaction of elec- A401 (1965).
trons with the exchange potential induced by so . This is [8] N. B. Brandt, S. M. Chudinov, and Y. G. Ponomarav, Semi-
similar in spirit to the gap renormalizations in 1D Luttinger metals 1. Graphite and its Compounds (North Holland,
liquids and leads to a logarithmically divergent correction Amsterdam, 1988).
to so . The divergence is due to the long range 1=r [9] D. P. DiVincenzo and E. J. Mele, Phys. Rev. B 29, 1685
Coulomb interaction, which persists in graphene even (1984).
[10] Y. A. Bychkov and E. I. Rashba, J. Phys. C 17, 6039
to so as well as similar corrections to @vF can be summed
accounting for screening [20]. The divergent corrections
(1984).
[11] F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988).
using the renormalization group (RG) [20]. Introducing the [12] D. J. Thouless, M. Kohmoto, M. P. Nightingale, and
dimensionless Coulomb interaction g  e2 =@vF we inte- M. den Nijs, Phys. Rev. Lett. 49, 405 (1982).
grate out the high energy degrees of freedom with energy [13] F. D. M. Haldane, Phys. Rev. Lett. 93, 206602 (2004).
between  and e
‘ . To leading order in g the RG flow [14] S. Murakami, N. Nagaosa, and S. C. Zhang, Phys. Rev. B
equations are 69, 235206 (2004).
[15] R. B. Laughlin, Phys. Rev. B 23, R5632 (1981).
dg=d‘ 
g2 =4; dso =d‘  gso =2: (8) [16] M. Fujita et al. J. Phys. Soc. Jpn. 65, 1920 (1996).
[17] Y. Niimi et al., Phys. Rev. B 70, 214430 (2004).
These equations can be integrated, and at energy scale ", [18] T. Giamarchi and H. Schulz, Phys. Rev. B 37, 325 (1988).
so "  0so 1  g0 =4 log0 ="2 . Here g0 and 0so are [19] M. Büttiker, Phys. Rev. B 38, 9375 (1988).
the interactions at cutoff scale 0 . The renormalized gap is [20] J. Gonzalez, F. Guinea, and M. A. H. Vozmediano, Phys.
determined by Rso so Rso . Using an effective interac- Rev. B 59, R2474 (1999).
tion g0  0:74 [21] and 0 2 eV this leads to 2Rso [21] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 93, 197402
15 K. (2004).

226801-4