Johan Thesis Chapter 2

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

5

Chapter 2

Introduction to carbon physics

Carbon is in many ways a unique element. Most importantly, it is crucial for life on earth as
we know it since the human body is to a large extent made up out of carbon. Scientifically,
the whole huge field of organic chemistry deals entirely with carbon-based compounds; and
in the field of physics carbon is one of the most intensively studied materials. There even
exists a journal named “Carbon” devoted exclusively to carbon that only in the year of
2006 consisted of over 3000 pages.
Pure carbon compounds comes in many different incarnations with different effective
dimensionalities. There are molecules and structures made up entirely of carbon atoms,
some of which have generated much excitement over the last two decades. One example
is the Fullerenes that were discovered in 1985 (Kroto et al., 1985). Carbon nanotubes is
another example. These are tubes made out of carbon, with a typical diameter of a few
nanometers (from which the name stems) and their lengths can be as large as a few mm.
Although they were discovered a long time ago,1 the really big interest came with their
rediscovery in 1991 (Iijima, 1991). Carbon nanotubes are interesting in many ways. For
example, depending on the diameter and how the tube is rolled up it can be both metallic
[i.e., the electronic excitation spectrum is gapless and the nanotube responds to an electric
field like a 1-dimensional (1D) metal] or semiconducting (i.e., the electronic spectrum has
a gap that can be used for photonic applications since the bandgap it typically in the right
range for optics). There has also been experimental demonstrations of transistors made out
of carbon nanotubes [for a recent review, see e.g. (McEuen et al., 2002) or (Avouris et al.,
2003)]. These have been shown to exhibit good characteristics, but unfortunately there is a
1
For a thorough discussion of where the credit is due, see (Monthioux and Kuznetsov, 2006).
6

problem with producing nanotubes with specific properties. A number of reviews and books
about carbon nanotubes have been produced over the years, see for example (Dressehaus
et al., 2000). Two other well-known pure carbon materials are graphite and diamond.
Graphite in particular is important enough to our topic to deserve a special section to
which we soon will turn.

2.1 Graphene

Graphite is a layered material made up of weakly coupled planar sheets of carbon atoms
that are arranged in a hexagonal lattice structure. It was noted by Wallace already in 1947
that a good starting point for studying graphite is to study the single sheet of graphite,
this is the material that we today call graphene. Graphene was for a long time assumed
not to exist (or being stable) by itself. But in 2004, in an experimental breakthrough it was
shown that atomically thin, large (∼ 10 µm) two-dimensional crystals of carbon could be
produced by a fairly simple technique (Novoselov et al., 2004). It is also worth to mention
that the same method can also be used to obtain two-dimensional crystals of other layered
materials (Novoselov et al., 2005b) . In this section, we will discuss some of the electronic
properties of graphene starting from the lattice structure and a simple tight-binding model
for the electronic motion. The discussion here will only serve as a brief review to set the
stage for the treatment of the graphene bilayer, for a more thorough treatment we refer the
reader to e.g. (Peres et al., 2006b) and references therein.

In ideal graphene the carbon atoms are arranged in a planar hexagonal lattice. Each
carbon atom has four electrons in the outer shell. Out of these three hybridize to form
the directed orbitals that dictates the hexagonal lattice structure and are responsible for
the formation of the lattice (Pauling, 1960). The remaining electrons occupy the π-orbitals
that are sticking out of the plane. These electrons are relatively free to move around in the
plane and are responsible for the low-energy electronic properties of graphene.

The hexagonal lattice is not a Bravais lattice. To fit the system into the usual Bloch state
picture it is therefore necessary to describe the system in terms of a triangular lattice with
7

two atoms in the unit cell, the two sublattices we call A1 and B1.2 A simple model for the
electronic properties of graphene consists of a nearest neighbor tight-binding Hamiltonian
considering only the π-orbital on each atom. The real-space lattice structure is depicted in
Fig. 2·1. A possible choice of the real-space lattice vectors are:

a √
a1 = (3, 3), (2.1a)
2
a √
a2 = (3, − 3). (2.1b)
2

Here a(≈ 1.4 Å) denotes the distance between nearest carbon atoms, and the vectors con-
necting atoms that are nearest neighbors are:

a √
δ1 = (1, 3), (2.2a)
2
a √
δ2 = (1, − 3), (2.2b)
2
δ3 = a(−1, 0), (2.2c)

which we take (by convention) to connect the A1-atoms to the B1-atoms. To construct
the tight-binding model for the motion of the electrons it is convenient to introduce the
operator that creates (annihilates) an electron on the lattice site at position Ri and lattice
site αj as c†αj,Ri (cαj,Ri ).3 Here α = (A, B) denotes the atom sublattice and j (j = 1)
denotes the plane. The tight-binding Hamiltonian then reads:

! ! " #
Ht.b. = −t c†A1,Ri cB1,Ri +δj + h.c. . (2.3)
Ri j=1,2,3

Here t (≈ 3 eV) is the energy associated with the hopping of electrons between neighboring
π orbitals. We now define the Fourier-transformed operators

1 ! ik·Ri
cαj,Ri = √ e cαj,k , (2.4)
N k

2
The subindex 1 is not really necessary at this point, but it is necessary later on when we need a layer
index in the graphene bilayer.
3
Because the spin is irrelevant for the independent electron problem that we are currently studying we
only consider spin-less electrons in this chapter.
8

a1
a2

δ1
A1
δ3 δ2
B1

Figure 2·1: The real-space lattice structure of graphene is that of a two-


dimensional planar honeycomb lattice. The A1 (B1) atoms are indicated
by the dark (light) circles.

where N is the number of unit cells in the system. Throughout this thesis we will use units
such that ! = kB = 1 unless specified otherwise. The tight-binding Hamiltonian in this
basis reads
!$ %
Ht.b. = ζ(k)c†A1,k cB1,k + ζ ∗ (k)c†B1,k cA1,k , (2.5)
k

where √
! $ ky a 3 %
ik·δi ikx a/2
ζ(k) = −t e = −te 2 cos( ) + e−i3kx a/2 . (2.6)
2
i

Because of the sublattice structure it is often convenient to describe the system in terms of
" #
a spinor: Ψ†k = c†A1,k , c†B1,k , in which case the Hamiltonian can be written as
 
! 0 ζ(k)
Ht.b. = Ψ†k   Ψk . (2.7)
k ζ ∗ (k) 0

The reciprocal lattice is shown if Fig. 2·2 along with the first Brillouin zone (BZ) which is
a hexagon. Alternatively one can use the diamond-shaped reciprocal unit cell shown in the
9

figure to label the states. A choice of the reciprocal lattice vectors are

2π √
b1 = (1, 3), (2.8a)
3a
2π √
b2 = (1, − 3). (2.8b)
3a

The encircled (K and K # ) corners of the BZ of Fig. 2·2 have the coordinates

2π √
K = √ ( 3, 1), (2.9a)
3 3a
2π √
K# = √ ( 3, −1). (2.9b)
3 3a

We now expand close to these corners of the BZ according to k = K + q and k = K# + q#


with the result that

3at −iπ/6+iqx a/2 3at −iπ/6


ζ(q) ≈ e (qx + iqy ) = e (qx + iqy ) = vF qei(φ−π/6) , (2.10a)
2 2
3at −iπ/6+iqx a/2 # 3at −iπ/6 # !
ζ(q# ) ≈ e (qx − iqy# ) = e (qx − iqy# ) = vF qe−i(φ +π/6) ,(2.10b)
2 2

where we have defined φ = φ(q) = tan−1 (qy /qx ). The Fermi-Dirac velocity is given by
vF = 3ta/2 in terms of the parameters in the tight-binding model. The extra phase of π/6
can be absorbed into the phases of the B1 wave-functions. Furthermore, because it is the
same for both the K point and the K # point, we will not have to worry about it even when
we are constructing wave functions that have components in both valleys.4 Thus from now
on we never write the π/6. The only time that the direction of φ is important in graphene
is for large values of |q| where the linear approximation to ζ(q) breaks down. In bilayer
graphene the direction of φ is important also for lower energies when one is considering the
so-called “trigonal distortion” that we will discuss in Chapter 3.

Alternative convention

Since two Bloch states that are separated by a reciprocal lattice vector are equivalent it is
possible to use other pairs of the corners of BZ to label the states. If we instead choose to
4
The two inequivalent corners of the BZ are often referred to as “valleys”.
10

describe the system using the pair of corners in the BZ denoted by stars in Fig. 2·2, namely
the K and K # points defined by


K = √ (0, −1), (2.11a)
3 3a
K# = −K, (2.11b)

we find

3at
ζ(q) ≈ (−qy + iqx ) = ivF qeiφ , (2.12a)
2
3at # !
ζ(q# ) ≈ (qy + iqx# ) = ivF qe−iφ . (2.12b)
2

This convention has the advantage that time reversal symmetry is easier to implement with
this particular choice. In particular time reversal takes k → −k, and there is in addition
the complex conjugate operation of an anti-unitary operator (Sakurai, 1994). Therefore
time reversal just exchange the K and the K # points in this convention.

K’
b1
K K

K’ K’
b2
K

Figure 2·2: The reciprocal lattice of graphene is a triangular lattice result-


ing in a hexagonal Brillouin zone. The two choices of the K and K # points
discussed in the text are shown as the circles and the stars in the corners of
the BZ. We also show the diamond as another choice of the reciprocal unit
cell and the approximation of covering it with two circles.

In any convention, after the proper rotation of the coordinate system the form of Hamil-
tonian close to the K point is that of a 2D Dirac Hamiltonian of massless Fermions:
 
! 0 qx + iqy
HDirac = vF Ψ†q   Ψq . (2.13)
q qx − iqy 0
11

The resulting spectrum is linear in the momentum E± (q) = ±vF q (as measured from the
K-point). A linear dispersion relation is unconventional in condensed matter systems and
has therefore generated a lot of excitement in the community. The bands are sketched in
Fig. 2·3 and gives rise to the so-called “Dirac point” at zero energy where the two bands
touch. Another interesting feature is that in neutral graphene the chemical potential is
sitting exactly at the Dirac point, this results in a Fermi point instead of the more generic
case of a Fermi surface. Because the Fermi surface is what gives rise to the stability to the
concept of the Fermi liquid (Shankar, 1994), it is expected that there will be deviations
from the Fermi liquid paradigm in graphene.

0.1

0.05
E"eV#

-0.05

-0.1

-0.1 -0.05 0 0.05 0.1


k ! "eV#

Figure 2·3: The “Dirac cone” dispersion of the quasiparticles in graphene.

Clearly the Hamiltonian in Eq. (2.13) is not valid for all points in the BZ. When the
momentum is far away from the corners of the BZ, the lattice will appear in the dispersion
and it will no longer be cylindrically symmetric. Nevertheless, to study the low-energy
properties near the Dirac point the linear spectrum should be a good approximation. But
since the band width is finite it is sometimes necessary to introduce a cutoff Λ to regularize
the theory at high energies. The linear spectrum is then assumed to be valid for momenta
such that vF q & Λ. A simple estimate of Λ is obtained by demanding that the number of
states in the BZ is conserved within the linear approximation. Thus two circles of radius
Λ should cover the BZ as sketched in Fig. 2·2. More explicitly 2πΛ2 = (2π)2 /Au , where

Au = 27a2 /2 is the area of the real-space unit cell. In the natural units where vF = 1
* √
this implies that Λ ≈ t π 3 ≈ 7 eV.
12

Like in high-energy physics we often use “natural” units setting ! = vF = 1. This


imply that energies and frequencies have the same units as momenta and wave numbers.
Therefore we often give the momenta in units of eV. For an electron (or hole) pocket with
a radius given by Q in eV, the corresponding electron (hole) density is

Q2 + Q ,2 4
n= = ≈ Q2 7.8 × 1013 cm−2 , (2.14)
π t 9πa2

which includes both the two valleys (K-points) and the two spin projections.

2.2 Graphite

Graphite is a material with a long scientific history that has been studied extensively
in the past. Numerous reviews on graphite have been written over the years, see for
example (Brandt et al., 1988; Chung, 2002). In this section, we briefly review the theory of
the graphite band structure, which is very relevant to the physics of the graphene bilayer.

Band model of graphite

In graphite the graphene planes are stacked in the A-B-configuration with the interplane
distance d ≈ 3.44 Å5 (Brandt et al., 1988). The lattice structure is sketched in Figure 2·4.
Thus there are now two planes and hence four atoms in the unit cell. Actually the real
space unit cell of graphite is the same as that of a graphene bilayer since the graphite lattice
is made up by stacking graphene bilayers. The Brillouin zone is also shown in Figure 2·4
along with the high-symmetry point K [H] at k⊥ = 0 [k⊥ = π/(2d)]. The atoms in the
unit cell are now labeled by A1, B1, A2 and B2; where 1 and 2 denotes the layer and A
and B the sublattice in each layer. The convention that we use to label the atoms is shown
in Figure 2·4 [see also Figure 3·1]. Therefore the A and B atoms in each plane are now
inequivalent, in particular the A atoms have a nearest neighbor directly above and below
it in the neighboring layer whereas the B atoms are sitting in the middle of the hexagons
5
In graphite with imperfect stacking (turbostratic graphite) the interplane spacing can vary down to
d ≈ 3.36Å.
13

H
H’
B2

K
A2 K’

H
H’
A1 B1

(a) (b)

Figure 2·4: (a) The lattice structure of graphite. (b) The hexagonal Bril-
louin zone of graphite, including the positions of the K and the H points.

of the neighboring layers.6 Using now a four-component spinor defined by

$ %
Ψ†k = c†A1,k , c†B1,k , c†A2,k , c†B2,k , (2.15)

the Hamiltonian can be written generally as

!
Hkin = Ψ†q H0 (q)Ψq . (2.16)
q

6
The advantage of this notation is that one can talk collectively about the A (B) atoms that are equivalent
in their physical properties such as the weight of the wave functions and the distribution of the density of
states etc. This notation was used in the early work of e.g., (McClure, 1957; Slonczewski and Weiss, 1958).
Many authors use a notation similar to A1 → A, B1 → B, A2 → B, e and B2 → A. e In this notation the
e and B (B)
relative orientation within the planes of the A (A) e atoms are the same, but for the other physical
properties the equivalent atoms are instead A (B) and B e (A).
e Since the other physical properties are often
more relevant for the physics than the relative orientation within the planes we choose to use the more
“natural” convention.
14

Here H0 (p) is a 4 × 4-matrix that close to the K-point can be parametrized as


 
∆ + γ Γ 2 /2 v pe iφ γ Γ −v v pΓe −iφ
 5 F 1 4 F 
 
 vF pe−iφ 2
γ2 Γ /2 −v4 vF pΓe −iφ v3 vF pΓe 

 
H0 (p) =  , (2.17)
 γ1 Γ −v4 vF pΓe iφ 2
∆ + γ5 Γ /2 vF pe −iφ 
 
 
−v4 vF pΓeiφ v3 vF pΓe−iφ vF peiφ γ2 Γ2 /2

where v3 = γ3 /γ0 and v4 = γ4 /γ0 . Microscopically, Γ ≡ 2 cos(k⊥ d) can be obtained


from a tight-binding dispersion in the direction perpendicular to the layers. The angle
φ ≡ φ(p) is obtained as in Eq. (2.10) or Eq. (2.12). Whether one should use eiφ or e−iφ
is dictated by the relative orientation of the appropriate pair of atoms projected on to the
x-y-plane as will be discussed in Chapter 3. The entries going like vF pe±iφ comes from the
in-plane graphene dispersion. Hopping terms corresponding to nearest neighboring planes
are different depending on the pair of atoms in question: γ1 (A ↔ A), γ3 (B ↔ B), γ4
(A ↔ B). γ2 (γ5 ) denotes a hopping between next nearest neighboring planes for the B
(A) atoms. Finally ∆ denotes the difference in on-site energies of the A and B atoms due
to their different crystal environments. The matrix
 
1 0 0 0 
 
0 e−iφ(p) 0 0 
 
M1 (p) =  , (2.18)
0 0 1 0 
 
 
0 0 0 eiφ(p)

can be used to perform the gauge transformation H1 = M†1 (φ)H0 M1 (φ) that moves all of
the e±iφ phase-factors (and hence the information of the orientation of the lattice) to the
γ3 -term:
 
∆ + γ Γ 2 /2 v p γ Γ −v v pΓ
 5 F 1 4 F 
 
 vF p 2
γ2 Γ /2 −v4 vF pΓ v3 vF pΓe 3iφ
 
H1 (p) =  . (2.19)
 γ1 Γ −v4 vF pΓ ∆ + γ5 Γ2 /2 vF p 
 
 
−v4 vF pΓ v3 vF pΓe−3iφ vF p γ2 Γ2 /2
15

This form has the advantage that it makes it manifest that the γ3 -term is responsible for
the “trigonal distortion” of the bands and actually breaks the cylindrical symmetry of the
bands. To make contact with the graphite literature one further performs two unitary
transformations. First, symmetric/antisymmetric combinations of the A-atoms leading to
bonding/antibonding bands are generated by the matrix:
 √ √ 
1/ 2 0 −1/ 2 0
 
 0 1 0 0
 
M2 =  √ √ . (2.20)
1/ 2 0 1/ 2 0
 
 
0 0 0 1

We further permute the bands with the matrix


 
1 0 0
0
 
0 0 −1 0 
 
M3 =  , (2.21)
0 1 0 0
 
 
0 0 0 −1

so that the transformed Hamiltonian H2 = M†3 M†2 H1 M2 M3 becomes


 √ √ 
∆ + γ Γ 2 /2 + γ Γ 0 (v Γ − 1)v p/ 2 (v Γ − 1)v p/ 2
 5 1 4 F 4 F
 √ √ 
 0 ∆ + γ5 Γ2 /2 − γ1 Γ (1 + v4 Γ)vF p/ 2 −(1 + v4 Γ)vF p/ 2
 
H2 =  .
 (v Γ − 1)v p/√2 √
(1 + v4 Γ)vF p/ 2 2
γ2 Γ /2 v3 vF pΓe 3iφ 
 4 F 
 √ √ 
(v4 Γ − 1)vF p/ 2 −(1 + v4 Γ)vF p/ 2 v3 vF pΓe−3iφ γ2 Γ2 /2
(2.22)
Except for some minor notational differences and the overall gauge transformation by (2.18)
this is the Slonczewski-Weiss-McClure model for graphite (McClure, 1957; Slonczewski and
16

Weiss, 1958), which is usually written as


 
E 0 H H ∗
 1 13 13 
 
 0 E2 H23 −H23 ∗ 
 
HSWMC = , (2.23)
H ∗ ∗ H33 
 13 H23 E3 
 

H13 −H23 H33 E3

where

1
E1 = ∆ + γ1 Γ + γ5 Γ2 , (2.24a)
2
1
E2 = ∆ − γ1 Γ + γ5 Γ2 , (2.24b)
2
1 2
E3 = γ2 Γ , (2.24c)
2
1
H13 = √ (−γ0 + γ4 Γ)eiα ζ, (2.24d)
2
1
H23 = √ (γ0 + γ4 Γ)eiα ζ, (2.24e)
2
H33 = γ3 Γeiα ζ. (2.24f)

Typical values of the parameters from the graphite literature are shown in Table 2.2, see
also (Partoens and Peeters, 2006) for a discussion on the connection between the tight-
binding parameters and those of the Slonczewski-Weiss-McClure model. The accepted

γ0 γ1 γ2 γ3 γ4 γ5 γ6 = ∆ &F
3.16 0.39 -0.02 0.315 0.044 0.038 0.008 -.024
3.12 0.377 -0.020 0.29 0.120 0.0125 0.004 -.0206

Table 2.1: Values of the Slonczewski-Weiss-McClure parameters for the


band structure of graphite. Upper row from (Brandt et al., 1988) and lower
row from (Chung, 2002).

parameters from the graphite literature results in electrons near the K-point [k⊥ = 0] and
holes near the H-point [k⊥ = π/(2d)] in the BZ as sketched in Figure 2·4. These electron
and hole pockets are chiefly generated by the coupling γ2 that in the tight-binding model
corresponds to a hopping between the B-atoms of next-nearest planes. Note that this
process involves a hopping of a distance as large as ∼ 7 Å.
17

Finally, it is interesting to note that at the H-point in the BZ, Γ = 0, and therefore the
two planes “decouple” at this point. Furthermore, if one neglects ∆ the spectrum is that
of massless Dirac fermions just like in the case of graphene. Note that in graphite A and B
atoms are different however, and that the term parametrized by ∆ will generally open up
a gap in the spectrum leading to massive Dirac fermions at the H-point. Since the value of
∆ in the literature is quite small, the almost linear massless behavior should be observed
by experimental probes that are not sensitive to these small energy scales.
The values of the parameters used in the graphite literature are consistent with a large
number of experiments. The most accurate ones are various magneto-transport measure-
ments discussed in (Brandt et al., 1988). More recently, ARPES was used to directly
visualize the dispersion of massless Dirac quasi-particles near the H-point and massive
quasi-particles near the K-point in the BZ (Zhou et al., 2006a; Zhou et al., 2006b).
The band structure of graphite has been calculated and recalculated many times over
the years, a recent reference is (Charlier et al., 1991). It is also worth to mention that
because of the (relatively) large contribution of the non-local van der Waals interaction
to the interaction between the layers in graphite, the usual local density approximation or
semilocal density approximation schemes are off by an order of magnitude when the binding
energy of the planes are calculated and compared with experiments. For a discussion of
this topic and a possible remedy see (Rydberg et al., 2003).

You might also like