Functionalized Carbon Nanotubes and Device Applications: Home Search Collections Journals About Contact Us My Iopscience
Functionalized Carbon Nanotubes and Device Applications: Home Search Collections Journals About Contact Us My Iopscience
Functionalized Carbon Nanotubes and Device Applications: Home Search Collections Journals About Contact Us My Iopscience
This article has been downloaded from IOPscience. Please scroll down to see the full text article.
2004 J. Phys.: Condens. Matter 16 R901
(http://iopscience.iop.org/0953-8984/16/29/R01)
Download details:
IP Address: 220.225.244.114
The article was downloaded on 08/01/2011 at 06:41
Please note that terms and conditions apply.
View the table of contents for this issue, or go to the journal homepage for more
Home Search Collections Journals About Contact us My IOPscience
INSTITUTE OF PHYSICS PUBLISHING JOURNAL OF PHYSICS: CONDENSED MATTER
J. Phys.: Condens. Matter 16 (2004) R901R960 PII: S0953-8984(04)77483-4
TOPICAL REVIEW
Functionalized carbon nanotubes and device
applications
S Ciraci
1
, S Dag
1
, T Yildirim
2
, O G ulseren
1
and R T Senger
1.3
1
Department of Physics, Bilkent University, 06800 Ankara, Turkey
2
NIST Center for Neutron Research, National Institute of Standards and Technology,
Gaithersburg, MD 20899, USA
3
T
UB
conduction, and valence bands crossing at the Fermi level, and exhibit 1D
quantum conduction [14, 15]. The (n. 0) zigzag SWNTs are generally semiconductor and
are only metal if n is an integer multiple of 3. Although the overall electronic structure of
SWNTs has been described by this simple picture, recent studies [16, 18] have shown much
more complicated structural dependence. For example, the (9. 0) tube is, in fact, a small band
gap semiconductor. The semiconducting behaviour of SWNTs has been of particular interest,
since the electronic properties can be controlled by doping or implementing defects in nanotube
based optoelectronic devices [1926].
Band calculations of SWNTs were initially performed by using a one-band orbital
tight binding model [11]. Subsequently, experimental data [2730] on the band gaps were
extrapolated to conrm the inverse proportionality with the radius of the nanotube [13].
Later, rst-principles calculation [31] within the local density approximation (LDA) showed
that the
states
(CB) and the singlet
and
states
mix and repel each other in curved graphene. As a result, the purely
state is
shifted downward with increasing curvature.
In gure 3(a), the doubly degenerate states (which are the valence band edge at the I
point), the doubly degenerate
state (which is in the conduction band for large R) are shown. As seen,
the shift of the singlet
mixing. This explains why the tight binding calculations predict band gaps around 1 eV for
n = 7. 8 tubes, while the self-consistent calculations predict much smaller values. Based on
orbital tight binding model, it was proposed [13] that E
g
behaves as
E
g
=
0
d
0
R
. (1)
which is independent of the helicity. Within the simple orbital tight binding model,
0
is taken
to be equal to the hopping matrix element V
pp
. d
0
is the bond length in graphene. However,
as seen in gure 3(b), the band gap displays a rather oscillatory behaviour. The relation given
in equation (1) was obtained by a second-order Taylor expansion of one-electron eigenvalues
of the orbital tight binding model [13] around the K point of the BZ, and hence it fails to
represent the effect of the helicity. By extending the Taylor expansion to the next higher order,
Yorikawa and Muramatsu [41, 42] included another term in the empirical expression of the
band gap variation,
E
g
= V
pp
d
0
R
1 + (1)
p
cos(3)
d
0
R
. (2)
which depends on the chiral angle, , as well as an index p. Here is a constant and the
index p is dened as the integer from k = n 2m = 3q + p. The factor (1)
p
comes from
the fact that the allowed k is nearest to either the K or K
/
point of the hexagonal Brillouin
zone. For zigzag nanotubes = 0, and hence the solid curves in gure 3(b) are ts to the
empirical expression E
g
= V
pp
d
0
R V
pp
d
2
0
R
2
, obtained from equation (2) by using
the parameters V
pp
= 2.53 eV and = 0.43. The agreement between the rst-principles
calculations [18] and the STS data [29, 30] is very good considering the fact that there might
be some uncertainties in identifying the nanotube in the experiment.
The situation displayed in gure 3 indicates that the variation of the band gap with the
radius is not simply 1R, but additional terms incorporating the chirality dependence are
required. Most importantly, the mixing of the singlet
state due to
the curvature and its shift towards the valence band with increasing curvature are not included
in either the orbital tight binding model or the empirical relations expressed by equations (1)
and (2). This behaviour of the singlet
is generated on the O
2
molecule to push it away from the SWNT. The magnetic moment of O
2
+ SWNT diminishes
at a distance d - 2 . Moreover, the singlet state of adsorbed O
2
leads to a bound state at
the A site at d = 1.47 , with an energy 0.45 eV above the corresponding physisorption
state (i.e. E
T
E
0
T
= 0.45 eV). A similar singlet bound state occurs also at the Z site at
R916 Topical Review
Figure 9. The electronic structure of O
2
physisorbed on an (8. 0) SWNT calculated using the spin
polarized GGA with the atomic structure fully optimized in the double cell. The spin up and spin
down bands corresponding to the triplet ground state are shown by broken and continuous lines,
respectively. The zero of energy is set at the Fermi level. E
V
indicates the top of the valence band.
The adsorption site (A site) is shown in the inset. (Reproduced from [75].)
d = 1.47 and 0.80 eV above the physisorption state of the Z site at d = 2.70 . These
singlet bound states in gure 10(b) correspond to local minima on the BornOppenheimer
surface and are separated from the more energetic physisorption states by an energy barrier.
It is noted that these states [77, 86] are neither easily accessible from the physisorption state,
nor support the hole doping picture because the band gap is 0.5 eV. However, O
2
adsorbed at
the H site of the (6. 6) SWNT behave differently. Figures 10(c) and (d) show the variation of
E
T
(d), F
d
CA
; the difference between spin unpolarized and spin polarized total energies LE
T
; the binding
energy E
u
b
obtained from spin unpolarized calculations; the binding energy E
p
b
obtained from
spin polarized calculations; the magnetic moment j per supercell corresponding to the magnetic
ground state corresponding to the adsorption of individual Ti, Mn, Mo, Au atoms on a (6, 6) SWNT.
(Reproduced from [103].)
Atom Site
d
CA
() LE
T
(eV) E
u
b
(eV) E
p
b
(eV) j (j
B
)
Ti H 2.2 0.48 2.62 1.81 1.68
Mn H 2.5 2.23 3.25 0.1 5.60
Mo H 2.3 0.2 4.34 0.1 3.61
Au T 2.3 0.02 0.41 0.28 0.79
interaction between the graphite surface and most of the atoms included is actually weak.
The curvature effect is the primary factor that strengthens the binding [18, 26, 43]. Specic
adsorbateSWNT (A + SWNT) systems are found to be in a magnetic ground state, since
LE
T
= E
u
T
[A + SWNT] E
p
T
[A + SWNT] > 0. Most of the transition metal atoms adsorbed
on the (8. 0) and (6. 6) tubes are in the magnetic ground state with LE
T
> 0, and hence they
have net magnetic moments ranging from 5.49 j
B
(for Mn) to zero magnetic moment (for Pd
Topical Review R921
and Pt). While Ni adsorbed SWNT has very low magnetic moment (0.04 j
B
), the adsorbates
such as Au, Ag and Cu have magnetic moments in the range of 0.40.6 j
B
. Atoms, such as
Na, Al, C, Si, Pb, O, S, H favour a nonmagnetic ground state when they are adsorbed on the
(8. 0) SWNT. The magnetic moment generated upon the adsorption of individual transition
atoms has important implications, and points to an issue: whether molecular magnets (or
nanomagnets) can be produced from carbon nanotubes. Addressing this issue may open an
active eld of study on SWNTs, which are coveredor substitutionally doped by transition metal
atoms according to a well dened pattern. Inclusion of transition metal elements inside the tube
is another way to obtain nanomagnetic structures. In this way, these atoms are prevented from
oxidizing. Whether a permanent magnetic moment can be generated by exchange interaction
on these transition metal coated SWNTs would be an interesting question to answer. Recently,
the magnetization and hysteresis loops of iron nanoparticles partially encapsulated at the tips
and inside of aligned carbon nanotubes have been demonstrated by experimental works [104].
Binding energies listed in table 2 are of particular interest for coating of SWNTs by metal
atoms, and hence for the fabrication of nanowires. The atoms which were observed to form
continuous and quasi-continuous coating on the SWNT (Ti, Ni and Pd) have relatively high
binding energies as compared to those atoms (Au, Fe, Pb) which remain as discrete particles
on the surface of the tube [96]. We also note that in forming a good coverage, not only
adatomSWNT interaction, but also other factors, possibly adatomadatom interaction, play
a crucial role. Good conductors such as Au, Ag, and Cu have very weak binding. On the
other hand, Na with one 3s electron on the outer shell is bound with a signicant binding
energy (E
b
= 1.3 eV). The binding energy of Mg is very weak and is only 0.03 eV at the
H site due to its outer shell (3s)
2
. For the same reason Zn by itself exhibits a similar trend
(E
u
b
= 0.05 eV) with its (4s)
2
valence structure. Owing to the weak binding the type of
the bond between Mg (Zn) and the SWNT is between chemisorption and physisorption with
d
CA
= 3.8 (d
CA
= 3.7 ). While an individual Al atom (with (3s)
2
3p valence structure)
is not bound to the graphite surface, its binding on the (8. 0) SWNT is relatively strong. This
can be explained by the curvature effect, since the binding was found to be even stronger at the
high curvature site of the SWNT under uniaxial radial deformation [26, 43]. Note also that the
binding energies E
u
b
, as well as E
p
b
and the magnetic moments of the adatom adsorbed on the
(8. 0) SWNT, came out to be consistently lower for the adatom adsorbed on the (6. 6) tube.
Perhaps this trend can also be explained by the curvature effect, since the radius of (8. 0) is
smaller than that of (6. 6) [18, 31].
The transition metal atoms with a few d electrons, such as Sc, Co, Ti, Nb, Ta, form strong
bonds with a binding energy ranging from 2.4 to 1.8 eV, and hence can be suitable for metal
coating of a SWNT. These metals can also be used as a buffer layer to formuniformcoating of
good conductors such as Au, Ag, Cu. Most of the adatoms yield the strongest binding at the H
site. Ni, Pd, Pt (column VIII elements) and Cu, Ag, Au (column I-B elements) seem to prefer
the A site. The average adsorbateC distance,
d
CA
, ranges between 1.9 (minimum) and
3.7 (maximum); most of them occur at 2.1 . The interaction of group IV elements with
the SWNT is also crucial, since they have the same valence conguration. C and Si adatoms
form rather strong bonds with the SWNT. The calculated binding energies are rather high,
i.e. E
b
= 4.2 and 2.5 eV, respectively. The Z site is energetically favourable for both C and Si
adatoms.
The character of the bond between the adatom and SWNT can be examined by charge
density analysis. For the fully relaxed, minimum total energy conguration, the SCF total
charge density (r) =
occup
n.k
+
n.k
(r)+
n.k
(r) and the difference charge density L(r) =
[A + SWNT] [SWNT] [A] can be calculated. In L(r), the total charge densities
of adatom adsorbed SWNT [A + SWNT], of bare SWNT [SWNT] and of free atom [A]
R922 Topical Review
Figure 13. Contour plots of total (r) and difference L(r) charge densities. For C the charge
densities are calculated on a plane passing through the adatom and zigzag CC bond. For others
(Na, Al, Ti) the charge density plane passes through the centre of the hexagon and adatom. In
contour plots of L(r), charge is depleted from black regions and is accumulated at white regions.
The counter-plots of (r) for [Na + SWNT] are not presented, since they do not convey relevant
information owing to the low charge density around adsorbed Na atoms. (Reproduced from [103].)
are all calculated in the same supercell with the same atomic positions. The difference charge
density conveys information about the charge rearrangements upon adsorption. Figure 13
shows (r) and L(r) which were calculated for four adatoms, namely Na, Al, C, Ti [103].
Dramatic differences in the valence electron congurations of these adatoms are reected in
the characters of the bonds they form with the SWNT.
According to the above analysis the charge is generally transferred from the adatom to
the CC bond of SWNT. The 3s valence electron of Na is weakly bound and is donated to the
Topical Review R923
conduction band of the SWNT, which is derived from the modied conduction band of the
bare SWNT. In the case of Al, which is adsorbed at the H site, electrons are transferred from
Al and nearest C atoms to the region between the Al atom and the centre of the hexagon on
the surface of the SWNT and to the CC bond. The character of the bond formed between
the C adatom and C atoms of the SWNT is reminiscent of the sp
3
covalent bonds in diamond
crystals. (r) has a maximum value at the centre of the CC bonds. Because of new bond
formation, charge is transferred to the back-bonds. The charge density of adsorbed Si exhibits
a character similar to that of adsorbed C. Directional bonds form between Si and C atoms of
the SWNT. The Ti atom adsorbed at the H site forms a bond. The d orbitals are responsible for
this bond. According to the Mulliken analysis the charge transferred from a Ti atom is 1.45
electrons for the spin unpolarized case. The interaction between the Au atom and SWNT is
weak, which results in a small binding energy. Accordingly, the charge rearrangement due to
Au adsorption is minute.
5.2. Electronic structure
An individual atom adsorbed on an SWNT may give rise to resonance states in the valence
and conduction bands, and also localized states in the band gaps. The energy states associated
with a single adsorbate form energy bands if the electronic structure is calculated within the
supercell method. Actually these bands correspond to a linear chain of adsorbates with a
1D lattice constant of the supercell, c
s
. The localized states are relevant for the doping of a
s-SWNT. Depending on their position relative to the band edges, they are specied as donor
states (if they are close to the edge of the conduction band, E
C
) or as acceptor states (if they
can occur close to the edge of the valence band, E
V
). The latter case is also known as hole
doping.
Na, Al, C and Si, which are adsorbed individually on the (8. 0) SWNT with a repeat
period of 2c, give rise to a spin paired, nonmagnetic ground state [103]. Figure 14 presents
their calculated band structures and the LDOS at the adatoms. The calculated band structure
of the SWNT with Na adsorbed at the H site forming a regular chain with a lattice parameter
(or interatomic distance) a = c
s
2c yields a half-lled band. This band is normally
the rst empty conduction band of the tube. Since the 3s valence electron of Na has low
ionization energy, it is donated to the SWNT to occupy the empty conduction band. As a
result, this empty band is gradually populated and also modied upon adsorption of Na; hence
the SWNT becomes metallic. Another alkali metal, Li, has been the focus of attention because
of SWNT ropes which were expected to be a candidate for anode material use in Li battery
applications [105]. In fact, recent experiments showed higher Li capacity in SWNTs than that
of graphite [106]. Zhao et al [107] investigated Li intercalation in (10. 0), (12. 0), (8. 0) and
(10. 10) SWNT ropes modelled by a uniform, 2D hexagonal lattice. They found that charge
is transferred from adsorbed Li to the SWNT. Furthermore, the SWNT is slightly deformed
upon Li adsorption. The band structure of the Li intercalated (8. 0) rope has indicated that the
bands of the bare SWNTs are not affected in any essential manner; but their conduction bands
get occupied above E
g
and the system becomes metallic. Their ndings are consistent with
the adsorption of Na discussed above.
Similarly, the calculated band structure of the SWNTwith Al adsorbed at the Hsite forming
a regular chain structure gives rise to a half-lled band derived from the empty conduction
band of the SWNT. Therefore, the localized state due to an adsorbed individual Al atom is a
donor state. In the case of C, a small gap occurs between the bands derived from adsorbate
states. The empty and lled bands just above and below E
F
are derived from C adatoms with
a signicant contribution from the SWNT. Si yields an almost fully occupied band in the band
R924 Topical Review
Figure 14. Energy band structures, TDOSs and LDOSs of single Na, Al, C, Si adsorbed on a
zigzag (8. 0) tube. LDOSs calculated at the adsorbate. The zero of energy is set at the Fermi level.
Na, Al and Si are adsorbed at the H site; C is adsorbed at the Z site. (Reproduced from [103].)
gap and an almost empty band at the bottom of the conduction band. The dispersion of bands
for Si adsorbed on the H site displays differences from that of the C adatom adsorbed at the Z
site. However, the overall behaviours of state densities are similar.
Topical Review R925
Figure 15. Geometric structures of adsorbed Ti chains on (14. 0) (top) and (8. 8) (bottom) SWNTs.
Along the tube axis, each unit cell contains two Ti atoms for the (14. 0) tube and one Ti atom for the
(8. 8) tube. Several unit cells are shown for the purpose of visualization. The distances between
the two neighbouring Ti atoms in the chains are 2.6 for (14. 0) and 2.7 for (8. 8) tubes.
(Reproduced from [108].)
The spin polarized band structure and TDOSs of adsorbed transition metal atoms (i.e. Au,
Mn, Mo and Ti) display a different situation due to the magnetic ground state. Au yields two
bands in the band gap, namely lled spin up and empty spin down bands. This is in compliance
with the calculated magnetic moment of 1j
B
per adsorbed Au atom. There is a small band
gap of 0.2 eV between these Au bands. It is seen that the adsorption of Au does not induce
signicant modication in the bands of (8. 0), except the Au 6s() and Au 6s() bands in the
gap; it gives rise to two sharp peaks below and above E
F
in the TDOS. That the contribution
of the SWNT states to these peaks is minute and that the band gap between the conduction and
valence bands of the SWNT is practically unchanged conrm the weak interaction between
Au and the SWNT. Filled Au 5d() and 5d() bands occur in the valence band of the SWNT
2 eV below E
F
. Three bands formed fromTi 3d() are fully occupied and accommodate three
electrons of adsorbed Ti atoms. Other Ti 3d() bands occur above E
F
, but they overlap with
the conduction band of the SWNT. The dispersive and almost fully occupied spin down band
is formed from the states of carbon and hence derived from the conduction band of the bare
SWNT. The SWNT is metallized upon Ti adsorption, since this band crosses the Fermi level
and also overlaps with the other conduction bands. This situation is in accordance with the
Mulliken analysis, which predicts the electrons to be transferred from Ti to the SWNT.
Yang et al [108] have studied the binding energies and electronic structures of Ti, Al and
Au chains adsorbed on (10. 0), (14. 0), (6. 6) and (8. 8) SWNTs using rst-principles methods.
The geometric conguration of the adsorbed metal chain is shown in gure 15. In spite of the
fact that the calculated binding energies include signicant coupling between adjacent metal
atoms, they show the same trends as the binding energies given by Durgun et al [102, 103]. Ti
is much more likely than Al and Au to form a continuous chain on the surface of SWNTs due
to much higher binding energies. Al adsorbed SWNTs in turn have higher binding energies
than the Au adsorbed ones. On the basis of the works by Bagci et al [109] and Prasen et al
[110], we believe that the stability of the Al and Au chains originates from the stability of
zigzag-free Al and Au chains. In calculating the binding energies the energy references are
crucial. It is well known that adsorbed Ti and Au have magnetic ground states leading to a
R926 Topical Review
Figure 16. Optimized atomic structure of (8. 0), (9. 0) and (6. 6) armchair SWNTs which are
uniformly covered with Ti. Corresponding densities of states for spin up and spin down electrons
are shown. (Reproduced from [7].)
net spin. Therefore, the binding energies reported by Yang et al [108] might change if one
took the magnetic states into account. The important effect of the Ti chain is to metallize the
zigzag SWNTs due to the delocalized d states crossing the Fermi level. However, the situation
is different for the Ti chain adsorbed armchair tube, since a small gap is opened owing to the
symmetry breaking.
5.3. Transition element covered or lled SWNTs
It is clear from the above discussion that individual atoms, such as Ti, Ni, Si, can form rather
strong bonds to the surface of the SWNT. Motivated by the experimental work of Zhang
et al [97] showing the continuous coating of the SWNT by Ti, Dag et al [7] investigated the
coverage of an (8. 0) tube by specic transition element atoms, such as Ti, Fe, Co, Cr, using
a rst-principles plane wave method. They found that only Ti can yield a continuous, regular
coverage of the SWNT, as well as giving rise to a dramatic modication of the atomic structure.
Upon optimization of the geometry the circular cross section of the SWNT has changed to
a square. Even more interesting is that the s-SWNT has become a highly conducting metal
due to several Ti bands crossing the Fermi level. The conductance of the Ti covered tube has
been estimated to be 16e
2
h. Furthermore, they found that the Ti covered (8. 0), (9. 0) and
(6. 6) tubes have magnetic ground states with signicant magnetic moments of 13.3, 13.7 and
9.5 j
B
cell, respectively. Figure 16 shows the structure and density of states of Ti covered
(8. 0), (9. 0) and (6. 6) tubes.
Yang et al [111] have examined the magnetism of transition metal covered or lled (9. 0)
SWNTs. They also found through ab initio calculations that these structures exhibit substantial
Topical Review R927
Figure 17. The spin polarized band structure (left: minority spin; right: majority spin) and DOS
(bottom: minority spin; top: majority spin) for Co lled SWNT as shown in the right panel. Both
plots suggest that the magnetic moment resides mostly in the narrow transition metal d band below
the Fermi level. (Reproduced from [111].)
magnetism. In particular, Co atoms packed inside a variety of carbon nanotubes offer strong
spin polarization at the Fermi level, as well as considerable magnetic moments. Figure 17
shows the spin polarized band structure and DOS for the Co lled SWNT.
6. Radial deformation of carbon nanotubes
Modication of electronic properties of condensed systems by an applied external pressure
or strain in the elastic range has been a subject of active study. However, in most cases, the
changes that one can induce by the elastic deformation are minute or even negligible due to the
rigidity of the crystals. On the other hand, the situation is rather different for SWNTs owing to
their tubular geometry [33]. Carbon nanotubes display very interesting mechanical properties.
While the MWNTs showthe greatest values for the Youngs modulus, as high as 2 TPa, SWNTs
appear to be very soft in the radial direction [112]. SWNTs are highly exible: they can sustain
elastic deformations remarkably. The softness of carbon nanotubes in the radial direction has
been observed experimentally: the contact area is attened when two nanotubes are brought
close to each other [113]. If R is large, a nanotube takes on a attened structure [114]. It has
been argued that SWNTs show linear elasticity under hydrostatic pressure up to 1.5 GPa at
room temperature.
R928 Topical Review
It has been shown that the structure and the electronic properties of SWNTs undergo
dramatic changes as a result of these deformations [23, 43, 115125]. Signicant radial
deformation of SWNTs can be realized in the elastic range, whereby the curvature is locally
changed. In this way, zones with higher and lower curvatures relative to the undeformed SWNT
can be attained for the same circumference. Hence, one expects radial deformation to induce
important modications in the electronic and conduction properties of nanotubes [24, 25, 126
129]. Tight binding calculations have indicated that a SWNT may undergo an insulator
metal transition under a uniaxial or torsional strain [24, 124, 125]. Multiprobe transport
experiments [23] on individual SWNTs showed that the electronic structure can be modied
by bending the tube, or by applying a circumferential deformation. Empirical extended H uckel
calculations [115] predicted that the conductance of an armchair SWNT can be affected by the
circumferential deformations and a band gap can develop in a metallic armchair SWNT upon
twisting.
A systematic analysis of the effect of the radial deformation on the electrical properties
has been carried out by G ulseren et al [43] using rst-principles total energy and electronic
structure calculations. In their study the radial deformation is generated by applying uniaxial
compressive stress
yy
to a narrow strip on the surface of a SWNT. In practice such a
deformation can be realized by pressing the tube between two rigid and at surfaces. As
a result, the radius is squeezed in the y direction while it is elongated along the x direction and
hence the circular cross section is distorted to an elliptical form with major and minor axes, a
and b, respectively. A natural variable for describing the radial deformation is the magnitude
of the applied strain along the two axes,
c
yy
=
R
0
b
R
0
(4)
and
c
xx
=
R
0
a
R
0
(5)
where R
0
is the radius of the undeformed (zero-strain) nanotube. The point group of the
undeformed nanotubes is D
nh
or D
nd
for n even or odd, respectively. Under the radial
deformation described above, the point group becomes C
2h
or D
2h
. However, depending on the
nanotube orientation around the tube axes, the in-plane mirror symmetry can be broken. For
the (6. 6) tube, there are several different orientations to investigate when looking at the effect
of mirror symmetry on the band crossing at the Fermi level. The rst-principles calculations
predict that elastic radial deformation does not have a noticeable effect on the rst- and second-
nearest neighbour CC distances but it induces signicant changes in the bond angles [43].
This observation is therefore important and has to be taken into account in tight binding studies
of SWNTs with radial deformation.
6.1. Elasticity
For the radial deformation, the relation between stress and strain is given by the generalized
Hookes law,
xx
= 0 = C
11
c
xx
+ C
12
c
yy
(6)
and
yy
=
F
y
A
= C
12
c
xx
+ C
11
c
yy
. (7)
Here F
y
is the restoring force applied on the surface area A, C
11
and C
12
are the in-plane elastic
stiffness constants. Assuming the validity of Hookes law, the elastic deformation energy E
D
Topical Review R929
Figure 18. (a) The strain component c
xx
= (R
0
a)R
0
along the major axis as a function of the
applied strain c
yy
= (R
0
b)R
0
along the minor axis. The slope is the in-plane Poisson ratio,
|
.
(b) Variation of the elastic deformation energy per carbon atom. (c) The restoring force on xed
carbon atoms. For the (8. 0) SWNT, the force is scaled by 0.5 since it is only on one carbon atom,
while for the other tubes it is on two carbon atoms. (Reproduced from [43].)
(i.e. the difference between the total energies of radially deformed and undeformed SWNTs)
is
E
D
= O
1
2
(1 +
2
|
)C
11
|
C
12
c
2
yy
(8)
in terms of the in-plane Poisson ratio,
|
[43]:
|
=
c
xx
c
yy
=
C
12
C
11
. (9)
Deformation energies obtained fromclassical elasticity in equation (8) are compared with
those calculated from rst principles in gure 18. Interestingly, the quadratic form obtained
from classical theory ts very well to the elastic deformation energy calculated from rst
principles. Hookes relation, and hence the elastic character of the deformations, persists up
to c
yy
= 0.25. It is also noted that the SWNT becomes stiffer as R decreases. The variation
of the restoring forces is expected to be linear in the elastic range. The restoring forces in
gure 18(c) are in overall agreement with this argument, except the deviations at certain data
R930 Topical Review
Figure 19. (a) The variation of the band gap, E
g
. (b) The density of states at the Fermi level D(E
F
)
as a function of the applied strain c
yy
. (Reproduced from [43].)
Table 4. In-plane elastic constants of SWNTs. All elastic constants are in GPa except for
|
which is unitless. C
eff
= C
11
(1
2
|
). (Reproduced from [43].)
Radius ()
|
C
eff
C
11
C
12
(7. 0) 2.76 0.904 129.88 713.36 645.15
(8. 0) 3.14 0.874 98.70 416.88 364.20
(9. 0) 3.52 0.864 91.02 319.67 270.36
(6. 6) 4.06 0.828 86.12 273.46 226.34
points due to uncertainties in the rst-principles calculations, which are amplied because the
force is a derivative quantity. Elastic constants calculated from rst principles are listed in
table 4.
6.2. Effect on the electronic structure
First-principles calculations conrm that the band gaps of zigzag tubes reduce with applied
strain and eventually vanish leading to metallization. Figure 19 summarizes the variation of
the band gap and the density of states at the Fermi level, D(E
F
), as a function of the applied
strain. For (7. 0) and (8. 0) SWNTs the band gaps decrease monotonically and the onset of
an insulatormetal transition follows with the band closures occurring at different values of
strain. Upon metallization, D(E
F
) increases with increasing strain. However, the behaviour
of the (9. 0) tube is different. Initially, the band gap increases with increasing strain, but then
Topical Review R931
Figure 20. The variation of the energy eigenvalues of states near the band gap at the I point of
the BZ as a function of the applied strain. The shaded region is the valence band. The singlet state
originating in the conduction band is indicated by squares. (Reproduced from [43].)
decreases with strain exceeding a certain threshold value and eventually diminishes. For all
these zigzag SWNTs the band gap strongly depends on the magnitude of the deformation and
E
g
is closed at 13%, 22% and 17% strain for (7. 0), (8. 0) and (9. 0) nanotubes, respectively,
whereas the armchair (6. 6) SWNT, which is normally metallic, remainedmetallic with a slowly
decreasing D(E
F
) even for signicant radial deformation. Earlier, Delaney et al [85] showed
that the
conduction and valence bands of a (10. 10) tube, which normally cross at the
Fermi level with quasi-linear dispersion, open a pseudogap in the range of 0.1 eV at certain
directions of the BZ perpendicular to the axis of the tube owing to tubetube interactions
in a rope. The opening of the gap is caused by the broken mirror symmetry. Lammert
et al [120] showed the gapping by squashing (20. 20) and (36. 0) metallic tubes, since then
circumferential regions are brought into close proximity. Uniaxial stress of a few kilobars
can reversibly collapse a small radius tube inducing a 0.1 eV gap, while the collapsed large
radius tubes are stable. In the study of Park et al [44], the band gap of the (5. 5) tube was
monotonically increasing, probably due to bilayer interactions, since the separation of the two
nearest walls of the tube became comparable to the interlayer distance of graphite.
In order to explain the band gap variation of (n. 0) tubes, the energies of a few bands
near the band gap are plotted as a function of strain in gure 20. The singlet
state in
the conduction band shifts downwards in energy much faster than the other states do with
increasing strain. This is due to the curvature increasing with increasing radial deformation.
Since the singlet
conduction
band under low strains. Once the singlet
band, which
is normally in the conduction band, falls into the band gap as a result of increased curvature.
Note that, while the band gap shows signicant change with n (for example by going from
Topical Review R933
Figure 22. (a) Variation of the binding energies E
b
of single hydrogen atoms adsorbed on a (8. 0)
zigzag SWNT as a function of the radial deformation c
yy
dened in the text. The upper curve
corresponds to H adsorbed on the high curvature site of the deformed tube. The lower curve is
for the adsorption on the low curvature site. (b) As (a) but for a single Al atom. (Reproduced
from [26].)
(8. 0) to (9. 0), E
g
changes from 0.65 to 0.09 eV), the binding energies vary smoothly with
R
1
. Increase of E
b
with decreasing R (or with decreasing n) shows that for small R the
character of the surface deviates from that of the graphene. This result has been exploited
to create different zones with different curvatures on the surface of a single SWNT by radial
deformation and hence to attain different chemical reactivities.
Radial deformation can be created by applying a uniaxial compressive stress on a narrow
strip on the surface of the SWNT [26, 43] where the circular cross section of the bare tube is
distorted to an elliptical one with major and minor axes a and b, respectively. In gure 22(a)
the variation of the binding energies of a single hydrogen atom adsorbed on the (8. 0) surface
shows that E
b
at the sharp site, i.e. near one of the ends of the major axis, increases with
increasing c
yy
(dened in equation (4)), and traces out the upper curve. The lower curve is
associated with the adsorption on the at site, i.e. near one of the ends of the minor axis.
According to the predictions of this study, at the high curvature site, E
b
has increased by
0.85 eV for c
yy
= 0.3. On the other hand, E
b
for the adsorption on the low curvature (at)
site rst decreases with increasing strain and then saturates at an energy 0.25 eV less than
that corresponding to c
yy
= 0. The difference of binding energies of the sharp and at sites
R934 Topical Review
is substantial and is equal to 1.1 eV for c
yy
= 0. This is 44% of the binding energy of H
on the undeformed SWNT. As a result of H adsorption, the sp
2
character of the bonding of
the tube has changed locally and become more like sp
3
. In gure 22(b) the binding energy
of Al exhibits a behaviour similar to that of H, despite H and Al favouring different sites on
the (8. 0) tube; E
b
at the sharp site of the deformed SWNT increases with increasing c
yy
. For
example E
b
increases by 0.80 eV for c
yy
= 0.3 which is 80% of the binding energy on
the undeformed tube. For Al adsorbed on the at site, E
b
rst decreases with increasing c
yy
,
then gradually increases. Adsorption of Al induces local changes in the atomic and electronic
structures. For example, the surface of the tube where Al is adsorbed expands.
We see that according to the rst-principles calculations the chemical reactivity of a
zigzag SWNT can be modied reversibly and variably by radial deformation. The effects of
deformation are signicantly different for the zigzag and armchair SWNTs. It is remarkable
that Al, which does not bind to the graphite surface, can be adsorbed at the high curvature site
of a zigzag tube under radial deformation with a signicant binding energy. Similarly, it has
been shown that a H
2
molecule which is bound to the surface of (8. 0) becomes physisorbed
at the high curvature site with a signicant binding energy [130]. If the physisorbed H
2
can
approach the surface as closely as 1.5 , it dissociates into two Hatoms, each of which becomes
bound to adjacent C atoms of the SWNT at the surface. We believe that the tunable adsorption
can have important implications for metal coverage and selective adsorption of foreign atoms
and molecules on the carbon nanotubes and can lead to a wide variety of applications, ranging
from hydrogen storage to chemical sensing, and new nanomaterials.
6.4. Effect of pressure on nanoropes
It is anticipated that physical properties can also be altered by intertube interactions between
nanotubes packed in a hexagonal lattice, as so-called nanoropes. Intertube interactions in
nanoropes can be probed by applying external pressure to vary the intertube distance. For
fullerenes, such high pressure studies have yielded many interesting results [131] including
new compounds such as the pressure induced polymeric phases of C
60
. Similar covalent
bonding can occur between the nanotubes in a rope. Such a property could have important
consequences for nanoscale device applications and composite materials that require strong
mechanical properties, since nanoropes consisting of interlinked SWNTs will be signicantly
stronger than nanoropes composed of vdW packed nanoropes [122].
Raman studies on SWNT ropes have been carried out up to 25.9 GPa, and have
indicated that the mode intensities and energies are not completely reversible upon pressure
cycling [132]. These results have been interpreted as irreversible pressure induced changes
in the atomic structure. Furthermore, observation of very large volume reduction and high
compressibility [119] has suggested that a microscopic volume reducing deformation other
than vdW compression can occur under high pressure. Some of these effects are tentatively
attributed to crushing or attening of nanotube surfaces in the rope, deforming the circular
cross sections to elliptical or hexagonal ones [119]. Peters et al [133] reported that a structural
phase transition occurred at 1.7 GPa. Furthermore they performed empirical uniform force
eld calculations and obtained equilibrium lattice constants of the (10. 10) nanotube rope as
a function of applied pressure P. These calculations have indicated a sudden transition of
structure at 1.75 GPa [133]. Their interpretation of the structural transition has been objected
to by Tang et al [134] on the grounds that their experiments were reversible up to P = 4 GPa.
Yildirim et al [123] were the rst to investigate the effects of pressure on the ropes based
on the rst-principles calculations. The zigzag tubes (5. 0), (6. 0), (7. 0), (9. 0) and the
armchair tube (6. 6) were included in that study. The pressure dependence was determined
Topical Review R935
Figure 23. Optimized structures of SWNT ropes. (a) (6. 0) and (b) (7. 0) nanotubes are packed by
vdW interaction under zero pressure and (c) (7. 0) one-dimensionally interlinked under pressure.
The interlinked structure in (c) has lower total energy than the vdW packed structure in (b). Lattice
parameters of the ropes, a, b and c and the angle, are shown. (Reproduced from [123].)
Figure 24. Total energy versus 2D lattice constant for (7. 0) nanotube ropes in different phases.
The top view along the axis of the rope of the different phases is shown in the inset. The zero of
energy is set to the total energy calculated for the optimized structure of the vdW packed rope at
zero pressure. (Reproduced from [123].)
by minimizing the total energy as a function of nanotube separation, i.e. E
T
(a. b), while the
other parameters, including the atom position c and = 2n, were optimized. At critical
pressure, a structural phase transformation from a vdW lattice to a new orthorhombic (space
group C
m
C
m
) lattice was observed in which the nanotubes were interlinked along the [110]
direction. The covalent bonding between nanotubes occurred at the high curvature sites as a
result of rehybridization of the carbon orbitals. Figure 23 compares the structure of a vdW
packed (7. 0) nanorope with that of a 1D interlinked one. The same structural transformation
was observed for the (5. 0), (6. 0), (9. 0) zigzag tubes.
The bonds interlinking two adjacent tubes have bond lengths comparable to those in
diamond. This indicates that the rehybridization occurs fromthe sp
2
to the sp
3
type. However,
the bond angles differ from the exact value of the tetrahedral angle and vary between 100
and 120
indicating some strain. For some (n. 0) tubes the energy difference between the
1D interlinked structure and the vdW packed structure is small. Regardless of that, the 1D
interlinked structure is stable once it is formed, since one has to overcome an energy barrier
of 0.7 eV in order to break the links, as illustrated in gure 24.
The total energies of the vdW and 1D interlinked phases cross each other at a 9 with
a small energy barrier of 46 meVunit cell (552 K) from the vdW side. The pressure
R936 Topical Review
for attaining this lattice constant is only about 0.3 GPa for the vdW phase, indicating that
polymerization of vdW (7. 0) nanoropes could occur at modest pressures and temperatures.
Conversely, 0.7 eVunit cell (or 25 meVatom) is required to go from the 1D interlinked
phase to the vdW packed structure. This energy is comparable to the 1D polymerization
energy of C
60
. Interestingly, at about a 8 there occurs another interlinked phase of the
(7. 0) nanorope, where nanotubes are interlinked along both a and b axes. This 2D interlinked
phase is not favoured energetically. Its minimum lies 1 eVunit cell above the minimum
of the 1D linked phase. However, once it is formed, it is stable and sixteen times stiffer
(d
2
Eda
2
= 13.7 eV
2
) than the vdW packed rope. The structures transform to even more
complex structures at higher pressure. The ropes of the (6. 6) armchair tubes do not form
interlinking up to 60 GPa. They are rather hexagonally distorted such that the local structure
of nanotube faces is reminiscent of that of a graphite sheet. Upon releasing the pressure
the hexagonally distorted (or polygonized) structure returns to the original one, indicating
that the deformation of the (6. 6) nanorope is elastic (see gure 25). The above structural
transformations induced by the applied pressure lead to signicant changes in the electronic
structure [123]. In spite of the fact that an individual (7. 0) tube is a semiconductor with
E
g
= 0.65 eV, the dispersion of the bands near Fermi level in the direction perpendicular to
the tube axis makes the vdW rope metallic. At high pressure (30 GPa), the (2D interlinked)
high density phase becomes a wide band gap insulator with E
g
2 eV.
7. Devices based on carbon nanotubes
Much of the research on carbon nanotubes has been carried out with the motivation of
generating devices, such as sensors, transistors and nanomagnets, and hence nding a feasible
way to contribute to the objectives of miniaturization. Attempts to design and fabricate
nanodevices from carbon nanotubes have exploited several properties discovered recently,
some of them reviewed in the previous sections. In this section, we will rst review recent
progress made in calculating I V characteristics of nanostructures and present some of the
device models or circuit elements, which are designed theoretically using SWNTs. Finally we
will discuss the devices fabricated.
7.1. Ab initio methods in transport calculations
Accurate prediction of the transport properties of nanoscale atomic or molecular systems
(nanotubes, in particular) including their currentvoltage (I V) characteristics is essential
for realization of a broad spectrum of device applications. Theoretical investigations of the
transport properties of these nanoscale systems are based on calculations with various levels of
sophistication, including phenomenological, semiclassical and quantum mechanical ab initio
methods.
Semiempirical methods [20, 135145] based on parametrized Hamiltonians, such as
tight binding or extended H uckel models, have been widely used. Since these parametrized
Hamiltonians are generally derived fromthe bulk or isolatedmolecular systems, the effects such
as external bias and gate potential are usually not accounted for. Moreover, these models do
not allow for structural relaxations to be performed. Semiempirical methods are not typically
self-consistent; however, they are relatively simple and easy to implement. These methods
provide some quantitative insight especially in cases where the charge transfer effects are
relatively small. Some applications to carbon nanotubes include designation of a quantum
dot by introducing pentagonheptagon defects in a carbon nanotube [20], investigation of
spin coherent quantum transport in carbon nanotube magnetic tunnel junctions [140], study
Topical Review R937
Figure 25. The two-dimensional crystal structure and relevant physical parameters for various
high density phases of carbon nanotubes. (a) A 2D interlinked structure of (5. 0) nanotubes. (b)
A hexagonal network of (9. 0) nanotubes, in which (9. 0) tubes are interlinked along a, b and
[110] directions. (c) A very dense structure of (7. 0) nanotubes obtained under 30 GPa. (d) The
optimizedstructure of (6. 6) armchair tubes under 53 GPa. Nanotubes are distortedin such a fashion
that the local nearest neighbour structure is reminiscent of graphite structure. d
CC
indicates the
smallest distance between two carbon atoms of nearest neighbour tubes in the rope. (Reproduced
from [123].)
R938 Topical Review
Figure 26. Upper panel: geometry of the bent (4. 4) nanotubes with bending angles (a) = 0
,
(b) 3
, (c) 6
. Lower panel: the calculated transmission function and LDOS for the corresponding
geometries. (Reproduced from [144].)
of the dynamic conductance properties of carbon nanotubes of different helicities, both with
and without defects [141], and calculation of transmission properties of extended [144] and
nite size [145] carbon nanotubes. Electronic transport properties of extended nanotubes
in the presence of distortions and mechanical deformations were studied by Nardelli [144]
using the surface Green function matching formalism and a nearest neighbour orbital tight
binding Hamiltonian [146]. The orbital bands were described via a single parameter,
V
pp
= 2.75 eV. Calculations made for small radius (4. 4) nanotubes show that bending of
the tube by small angles (3
and 6
d G
-
() (11)
where
G
-
() = G
R
-
G
A
(12)
with G
R
and G
A
= (G
R
)
-
= 2i Im[ f
l
R
l
+ f
r
R
r
] . (13)
Here j
l
and j
r
are the Fermi levels deep in the left and right electrodes such that V
b
=
(j
l
j
r
)e is the external applied bias. The Green functions are evaluated by direct matrix
R940 Topical Review
Figure 27. (a) The metallic electrodenanotube interface. A semi-innite (4. 4) armchair SWNT
is in contact with a semi-innite Al(100) metallic electrode. The device region is coupled to perfect
atomic scale electrodes which extend to reservoirs at z = . (b) Charge density contours of the
interface at equilibrium. The contours of the bulk lead match that of the device at the edges of the
device region. (Reproduced from [161].)
inversions:
G
R
() = [I H
R
l
R
r
]
1
(14)
where H[
c
( r)] is the Hamiltonian matrix of the central region within DFT and evaluated
using the charge density obtained from equation (11). The cyclic procedure summarized
above should be iterated until NEGFDFT self-consistency is achieved.
The I V characteristics of the nanoscale device are calculated using the Landauer formula
I (V
b
) =
2e
h
d T() [ f
l
( j
l
) f
r
( j
r
)] . (15)
It should be noted that the transmission function T() depends both on temperature and on the
applied bias and is expressed as
T() = Tr[
l
()G
R
()
r
()G
A
()] (16)
where
p
is the function describing the linewidth due to the coupling of the central device
region to the electrode p. It is related to the corresponding self-energies of the electrodes in
the form
p
= i(
R
p
A
p
). ( p = l. r). (17)
Topical Review R941
z-position (nm)
-200 0 200 400 600
0
1
2
3
G
(
G
)
0
c
Fibre
a
b
Figure 28. (a) A transmission electron micrograph (TEM) image of carbon nanotubes at the end
of a nanotube bre. (b) A schematic diagram of the experimental set-up. Nanotubes are lowered
under SPM control into a liquid metal surface. (c) Variation of conductance with nanotube bre
position. Plateaus are observed corresponding to additional nanotubes coming into contact with
the liquid metal. (Reproduced from [15].)
With the development of self-consistent ab initio methods based on DFT and the NEGF
formalism, the transport properties of a broad set of nanoscale systems [164168], including
C
60
, linear chains of carbon, fullerene and silicon, and gold nanowires, are studied under
nonequilibrium conditions. Some interesting features such as negative differential resistance
in I V curves and the importance of the detailed atomic structure of the electrodes are reported.
Carbon nanotubes having different electronic structures depending on their chirality and
radius are candidates for being main constituents of the future nanoelectronics technology. A
considerable amount of experimental and theoretical effort has been devoted to investigating
quantum transport properties of nanotubes. Calculations show that innitely long, perfect,
metallic single-wall carbon nanotubes have an equilibrium conductance of G = 2G
0
, where
G
0
= 2e
2
h, due to two subbands crossing at the Fermi level [11, 12]. The existence of
pentagonheptagon pair defects in innite nanotubes is shown to alter the conductance pattern
slightly [152]. The discrepancy between the theoretical calculations and the much smaller
values of the nanotube conductance measured in experiments is attributed to poor nanotube
electrode contacts.
In an interesting transport experiment performed by Frank et al [15] the conductance of
MWNTs, dipped into a liquid metal to minimize the contact resistance effects, were found to
be close to G
0
, half the expected value. Figures 28(a) and (b) show the transmission electron
micrograph (TEM) image of the nanotubes used and the experimental set-up. The conductance
is seen to increase as the nanotube bre is dipped into the liquid metal in steps of G
0
as shown
in gure 28(c). Each step corresponds to an additional nanotube coming into contact with the
metal. Although narrower, two more steps at fractional G
0
values are also observed.
An explanation of the measured conductance pattern is provided by Choi and co-
workers [153] in their ab initio calculation of the conductance of the (10, 10) SWNT with one
end immersed in a jellium metal. They showed that of the two available channels, it is mostly
band incident electrons that suffer strong back-scattering at the nanotubemetal contact,
while the
-
band is responsible for the observed conductance of the system, i.e. G = G
0
. In
another account, using a semiempirical scattering technique Sanvito et al [169] calculated the
conductance of a MWNT with a special geometry and provided a possible explanation of the
measured [15] fractional conductance values in terms of interwall interactions in a MWNT.
Later, Taylor et al [161] simulated transport properties of a SWNTmetal interface using the
R942 Topical Review
NEGFDFT formalism and a semi-innite Al electrode of nite cross section (see gure 27).
They showed that there are two transmission eigenvectors which connect the two sides of
the electrodenanotube interface. The two transmission eigenvalues T
1
and T
2
were smaller
than unity; however, their sum at the Fermi energy gave an equilibrium conductance almost
equal to G
0
for a range of distances between the nanotube and the electrode. This result is
consistent with both the nanotubejellium calculation [153] and the experiment [15]. In a
recent study Palacios et al [170] addressed the problemof conductance of nite length carbon
nanotubes with realistic metal electrodes using a rst-principles method called the Gaussian
embedded cluster method [163]. They observed resonance patterns around the Fermi level
in the conductance proles. As electrodes, elemental Ti was found to make better high
transparency contacts to carbon nanotubes compared to more traditional metals such as Al
and Au.
The electronic and transport properties of carbon nanotubes can be modied by doping
with foreign atoms. The effects of substitutional nitrogen doping in carbon nanotubes under
nonequilibriumconditions were investigated by Kaunet al [171] using NEGFDFTformalism.
Nitrogen doping of nanotubes does not cause substitutional structural relaxation; however,
the step-like equilibrium conductance pattern of the nanotube is altered due to the action of
a substantial number of nitrogen atoms as elastic scattering centres. It was found that for
doped zigzag nanotubes even a single nitrogen atom substitution increases the current ow
and for small radii tubes narrows the current gap. Periodic doping of nitrogen atoms in a
semiconducting nanotube makes the structure metallic with a nonlinear I V characteristic and
a negative differential resistance at higher biases. It is interesting that breaking of symmetry
by the impurities can raise the current ow in semiconducting tubes and reduce the current
slightly in metallic ones. All these nonequilibriumtransport properties can be understood from
the simple physical picture of alignment and misalignment of current carrying bands of the
device due to applied bias.
7.2. Device models
Several novel properties of carbon nanotubes, which have been revealed through the active
experimental and theoretical research during the last decade, have been combined in
nanodevice operation or circuit elements. These are semiconductor heterostructures or metal
semiconductor junctions; resonant tunnelling double-barrier (RTDB) and multiple-quantum-
well structures (MQWS); metallic nanorings for persistent currents and solenoids; metal
nanotube contacts and Schottky barriers; interconnects; etc.
SWNTSWNT junctions. Structural defects can affect the atomic and electronic structure
of nanotubes. The role of topological defects, such as 57 pentagonheptagon defects, has
been investigated by different groups. Charlier et al [19] carried out an extensive study on
the topological 57 defects in different directions on the surface of the (12. 0) SWNT and
revealed their effects on the atomic and electronic structure by using a recursion method
within the tight binding description of the carbon bands. They found that the 57 defect
can induce changes in diameter and chirality of the nanotube. For example, if the 57 defect
can be aligned exactly on the axis of the (n. 0) zigzag tube, the chirality remains the same,
but the diameter becomes smaller. If the 57 defect is perpendicular to the axis of the tube,
the diameter is conserved, but the zigzag SWNT becomes chiral. Chirality increases with
increasing number of 57 defects perpendicular to the axis of the tube. The 57 defects
in any orientation on the surface of SWNTs change both the diameter and the chirality. Not
only atomic or geometric structure, but also the electronic structure around the Fermi level gets
Topical Review R943
Figure 29. Energies of the discrete states of the (6. 4)m(5. 5)(6. 4) quantum dots as a function
of the number of (5. 5) unit cells m. (Reproduced from [20, 173].)
modied by the presence of 57 defects. Most importantly, nanotubes with different chiralities
and different electronic properties and diameters can be connected using such a pair defect to
formdifferent kinds of heterostructures. These heterostructures (such as metalsemiconductor
and semiconductorsemiconductor) can form rectiers or diodes and hence they can play an
important role in nanoelectronics. In fact, the calculation of electronic transport through such
junctions yielded tunnelling conductance and hence indicated that they can be used as building
blocks of nanoscale electronic devices [172]. Chico et al [173] studied the electronic structure
of (8. 0)(7. 1) and (8. 0)(5. 3) junctions, which were formed from a single (8. 0) SWNT
by introducing a single and triple 57 defect pair, respectively. For the case of the (8. 0)
(7. 1) junction, it corresponds to a heterostructure between a semiconductor having a band
gap of 0.62 eV and a semimetal. The second junction, (8. 0)(5. 3), is a semiconductor
semiconductor heterostructure with two different band gaps.
It is clear that by the controlled generation of 57 defects on a chiral SWNT one can form
a variety of SWNTSWNT junctions with different heterostructure characteristics. Chico
et al [20] went further and proposed an interesting heterostructure, namely a new type of
quantum dot based solely on single SWNTs. It is made from a series of (6. 4)m(5. 5)(6. 4)
SWNT junctions. Here the (6. 4) tube is a semiconductor; m(5. 5) indicates m unit cells of a
metallic armchair tube. The (6. 4) tube was connected to the (5. 5) tube smoothly, introducing
heptagonpentagon defects. In this way formation of undesirable interface states has been
eliminated. Normally, a metallic nanostructure between two semiconducting ones forms a
quantumdot and this has quantized electronic states. Using the empirical tight binding method
and Green function matching approach along with the transfer matrix technique, Chico et al
[20] showed that electronic states in the metallic m(5. 5) part are indeed conned and exhibit
a sharp, function-like density of states. Moreover, their level spacings vary depending on the
value of m, showing a clear quantum size effect (see gure 29).
Quantum structures generated by radial deformation. The metallization of semiconducting
SWNTs under radial deformation (discussed in section 6) has been proposed to generate
quantum structures [24, 44, 118]. The idea is simple and originates from the numerous works
done on semiconductor heterostructures (A
n
B
m
) and quantum dots [174]. It is known that
R944 Topical Review
owing to the band offsets of the semiconductor heterostructures, the energies of the band states
of one semiconductor Bmay fall into the band gap of the adjacent semiconductor A. According
to the effective mass approximation, the height (depth) of the conduction (valence) band edge
of Ameasured fromthat of B, LE
C
(LE
V
), behaves as a potential barrier for electrons (holes).
For example, m layers of B between n layers of two As forma quantumwell yielding conned
electronic states. The depth of the well and the width of the barrier and well (in terms of
number of layers n and m, respectively) are crucial parameters for monitoring the resulting
electronic properties. By repeated formation of (A
n
B
m
) one generates MQWSs or RTDBs
by combination (A
n
B
m
A
n
). Similar to what one achieves in semiconductor heterostructures,
periodic MWQSs or nite RTDBstructures can be realized on an individual s-SWNT and their
electronic properties can be variably and reversibly monitored.
Performing empirical tight binding and also rst-principles calculations, Kili c et al [24]
have shown that a MQWS is generated by applying different radial deformations at adjacent
parts of a (7. 0) s-SWNT and repeating them periodically. They considered a supercell
consisting of 16 unit cells of the (7. 0) SWNT; n of the unit cells were kept undeformed
and the remaining m = 16 n were radially deformed. Tight binding charge densities of the
band edge states for superlattices corresponding to n = 4, 8, 12 have clearly demonstrated
the connement of states and quantum well structure formation. Later, the strain induced
quantum well structures were treated by means of SCF rst-principles calculations on a
superlattice consisting of 12 (8. 0) SWNT unit cells [18]; half of them were radially strained
with c
yy
=0.16; the remaining half were left undeformedi.e. c
yy
= 0. Within the superlattice
geometry, the adjacent unstrained and strained parts were repeated yielding a periodic band gap
modulation. Applying the standard method for band line-up explained in detail below [175],
using a SCF crystal potential and referring to the energies at the top of the valence band, the
mist of the valence band edges between adjacent (i.e. undeformed and deformed) zones has
been found to be 180 meV. This indicates a staggered band line-up. As in the empirical tight
binding calculations, it has been demonstrated that quantum structures can be formed on a
single SWNT. Tombler et al [118] have shown that a quantumdot can be produced by bending
(and hence inducing radial deformation of) the SWNT at a given point.
Quantumstructure formation by modulating hydrogenation. Amodel of an electronic device
based on a similar band gap modulation and hence MQWS formation has been proposed
recently [176]. In this case the band gap modulation is provided by the modulating hydrogen
adsorption (see gure 30). The localization of states in one of the zones (either a (C
32
)
l
(i.e. specied by l) zone or a (C
32
H
32
)
q
(i.e. specied by q) zone) is based on the physical
grounds discussed above.
Formation of different zones leading to different electronic structures on the same
(8. 0) SWNT upon modulating the H adsorption, and also the effect of the superlattice
size (i.e. l, q and l + q), is further analysed by examining the electronic potential [176].
The electronic potential which consists of Hartree, exchange and ionic pseudopotentials,
V
e
(r) = V
H
(r) + V
x
(r) + V
ion
(r), is calculated self-consistently. Figure 31 shows the planar
averaged electronic potential
V
e
(z) =
1
a
s
b
s
V
e
(x. y. z) dx dy (18)
and the running average of the planar averaged potential,
V
r.e
(z), along the axis of the SWNT.
Here a
s
and b
s
are superlattice parameters. As seen in gure 31(a), the short periodicity,
l = q = 1, is not sufcient to induce two adjacent zones with markedly different electronic
structures. In gure 31(b) the differences between two zones become pronounced and LV
Topical Review R945
Figure 30. Superlattice structures, [(C
32
)
l
(C
32
H
32
)
q
] formed on a single (8. 0) SWNT by
modulating adsorption of hydrogen atoms. (a) l = q = 1; (b) l = q = 2. The positions of
64 H and 128 C atoms, as well as the superlattice parameter c
s
, are calculated from full structure
optimization by using conjugate gradient method. (Reproduced from [176].)
Figure 31. Averages of calculated self-consistent electronic potentials of the superlattices. (a)
[(C
32
)
1
(C
32
H
32
)
1
]; (b) [(C
32
)
2
(C
32
H
32
)
2
]; (c) [(C
32
)
3
(C
32
H
32
)
2
]; (d) [(C
32
)
2
(C
32
H
32
)
3
]. Dashed
and thick curves are planar averaged,
V
e
(z), and running averaged,
V
r.e
(z), respectively. Relatively
high cell averaged potential zones correspond to (C
32
)
l
. See the text. (Reproduced from [176].)
increases for the superlattice; l = q = 2. Figures 31(c) and (d) indicate that the potential,
V
r.e
(z), becomes atter as l and q increase from2 to 3 and two adjacent zones display markedly
different electronic structures.
Band line-up of zones, (C
32
)
l
and (C
32
H
32
)
q
, is a complex process involving the relaxation
of the atomic structure and charge transfer at the interfaces and also resulting in modication
R946 Topical Review
(C
32
H
32
)
q
(C
32
H
32
)
q
E
V
E
C
E
g
l
E
g
q
(C
32
)
l
Figure 32. Aschematic illustration of the energy band diagram and band offsets estimated by using
averaged self-consistent electronic potentials,
V
r.e
(z), calculated for the l and q zones, LE
B.l(q)
and band gaps, E
g1
and E
g2
, calculated for l = and q = . The band diagram indicates a
type-I band line-up appropriate for a resonant tunnelling double-barrier structure. (Reproduced
from [176].)
of the electronic potential. Experimental and theoretical methods have been proposed in
the past to determine the band offsets and hence to obtain the band diagram perpetuating
along the semiconductor heterostructures. A solution of this problem [175] involves the
following steps. First, the average electronic potentials corresponding to
V
r.e
(z) for nite
l and q, as well as for innite l and q, (C
32
)
l=
and (C
32
H
32
)
q=
, are calculated. The
actual atomic positions, especially those at the interface, and the charge transfer between
regions are of crucial importance for the band offsets and are incorporated self-consistently
in the calculated average electronic potential. Secondly, for the case of l = q =
the energy difference between the edge of the valence band E
V
and the average potential,
i.e. LE
B
[
l(q)=
= E
V.l(q)=
V
r.e
[
l(q)=
, are calculated. LE
B
for (C
32
)
l=
and (C
32
H
32
)
q=
is found to be 2.62 and 3.91 eV, respectively. In the third step, one assumes that LE
B
is
unchanged for nite l and q and determines the valence band edges of a particular superlattice
with given l and q by adding LE
B
[
l(q)=
to the corresponding average potential values
V
r.e
of the l and q zones. At the end, the energies of the valence band edges of two adjacent
zones (i.e. hydrogen-free and exohydrogenated) are xed. Then, the valence band offset
of the superlattice is LE
V
= E
V.l
E
V.q
. The energies of the conduction band edges
of the superlattice are determined by adding band gaps of different zones (calculated for
l = and q = ) to the energies of the valence band edges, i.e. E
C.l
= E
V.l
+ E
g1
and
E
C.q
= E
V.q
+ E
g2
. Here, E
g1
and E
g2
are the band gaps of hydrogenated and hydrogen-free
zones, respectively. The conduction band offset is then determined by subtracting these latter
energies, i.e. LE
C
= E
C.q
E
C.l
. Figure 32 presents a schematic illustration of the band
diagram obtained in [176].
The present discussion for the periodically repeating quantum wells can be extended to
nite systems, for example to the resonant tunnelling double barriers, made by a zone (C
32
)
l
placed between two zones of (C
32
H
32
)
q
, i.e. (C
32
H
32
)
q
(C
32
)
l
(C
32
H
32
)
q
. Metallic reservoirs
adjacent to both barriers, i.e. (C
32
H
32
)
q
, and metallic connects of the device can be achieved
by metal coating [102, 103, 109]. Also a metalsemiconductor heterostructure can be formed
by modulating the hydrogenation of an (n. n) armchair SWNT. In this way metallic connects
are provided by the hydrogen-free ends of the tube.
SWNTmetal contact. The switching of the current in a s-SWNT at room temperature by an
external electric eld has been utilized to fabricate a neweld effect transistor (FET) [99101].
The interaction between the s-SWNT and metal electrode and the Schottky barrier, +
B
, formed
Topical Review R947
0
5
5
3 6 9 12 15 18 21 24
10
V
e
(
z
)
[
e
V
]
z()
Figure 33. (a) Local densities of states (LDOSs) at different Au and C atoms of the (8. 0) zigzag
SWNT side-bonded to the Au(100) surface. The locations of atoms are shown in the inset. The
zero of energy is taken at the Fermi energy E
F
. State densities shown by empty circles and lled
diamonds correspond to six neighbouring carbon atoms. Other densities are for single atoms.
(b) Contour plots of the SCF electronic potential, V
e
(r), on a vertical plane. (c) As (b) but on a
horizontal plane bisecting s. In the dark grey regions, V
e
(r) > E
F
. (d) The variation of V
e
(z) on
a perpendicular line passing through the centre of the SWNT. (Reproduced from [179].)
thereof have been proposed as the origin of the FET operation. Model calculations have been
carried out to provide further understanding of experimental I V characteristics [177, 178].
The origin of the potential barrier formed between an s-SWNT and a specic metal
electrode has been investigated recently by using rst-principles calculations [179]. The
interaction and energy of binding between an individual Au atomand (8. 0) tube is essential for
the contact. The contact on the gold electrode is formedby placing the s-SWNTon the Au(100)
surface, so a surface Au atom faces the centre of the hexagon. The AuSWNT equilibrium
distance, s, occurred at 3.18 according to full structure optimization. This implies a weak
AuSWNTinteraction which is characterized by physisorption. The LDOSs calculated for two
Au atoms at different sites (one is at the contact just belowthe SWNThexagon; the second being
furthest from the SWNT mimics the clean Au surface) in gure 33(a) have nite state density
at the Fermi level. These two LDOSs are similar; there are only minute changes, whereas the
LDOS of the carbon atoms at the contact still has a band gap. The Fermi level lies near the
top of the valence band of the SWNT and is consistent with a small +
B
in the hole doping
picture. The LDOS at the carbon atom which is furthest from the metal electrode displays a
state distribution similar to that of the carbon atom in the contact region. The calculated total
charge density of the SWNT bonded to the metal electrode, E
M
, i.e. [SWNT + E
M
], and the
difference charge density, L = [SWNT + E
M
] [E] [SWNT], show minute charge
rearrangements. These results indicate that the weak Au electrodeSWNT interaction does
not induce any signicant changes in the electronic structure. The SCF electronic potential
between the SWNT and the Au electrode, V
e
(r), is presented on a vertical plane and also on a
horizontal plane bisecting s in gures 33(b) and (c). The shaded area shows that the electronic
potential energy at the contact yields a potential barrier +
c
= V
e
E
F
> 0. At the contact
midway between the SWNT and Au(100) surface, +
c
is calculated to be 3.9 eVthat is,
comparable with the calculated work function (+ 5 eV) of the Au slab. Figure 33(d) shows
the variation of V
e
(z) on a line passing through the centre of the SWNT and perpendicular
R948 Topical Review
Figure 34. (a) Local densities of states (LDOSs) at different Mo and C atoms of the (8. 0) zigzag
SWNT side-bonded to the Mo(110) surface. The locations of atoms are shown in the inset. The
zero of energy is taken at the Fermi energy E
F
. State densities shown by empty circles and lled
diamonds correspond to eight neighbouring carbon atoms, The others are for single atoms. (b)
Contour plots of the total SCF charge density on a vertical plane. State densities shown by empty
circles correspond to ten neighbouring carbon atoms. (Reproduced from [179].)
to the Au(100) surface. The effective potential barrier, +
c.eff
, can be even higher owing to
the increased connement of electrons in the contact region. Therefore, electrons which are
transferred from the metal to the s-SWNT have to tunnel through a potential barrier +
c.eff
(r).
The above situation is, however, different in the case of an s-SWNT side-bonded to the
Mo(110) surface, as shown in gure 34. Upon relaxation, the tube has rotated slightly so C
atoms tend to approach Mo atoms. The LDOS at the Mo atom interacting with the C atoms
of the SWNT is different from that at the clean Mo(110) surface. Moreover, the LDOS at the
carbon atomclosest to the Mo surface has nite state density at E
F
. In particular, the peak near
E
F
is associated with the CMo bond states and hence may be attributed to the metal induced
gap states (MIGS). The LDOS of the C atoms furthest from the contact has a band gap near
E
F
. This situation indicates that the site of the SWNT forming contact is conducting, while the
opposite site furthest from the contact remains semiconducting. Owing to the strong MoC
bond, the spacing between the SWNT and the Mo electrode is smaller (s = 1.96 ) than that
with the Au electrode. The strong CMo bond formation is claried by the calculated total
SCF charge density in gure 34(b). The analysis of the difference charge density L shows
Topical Review R949
the strong charge transfer to the CMo bonds. The electronic potentials calculated on the
horizontal plane bisecting s and also on a line from the SWNT to the Mo(110) surface through
the contact show that V
e
(r) - E
F
in the contact region. Accordingly, a potential barrier, +
c
,
does not form at the extended contact between the SWNT and the Mo electrode.
It appears that the above results distinguish two different types of electronic structure for
two different electrodes. A large spacing and sizable potential barrier between the tube and
metal electrode, +
c
3.9 eV, are characteristics of the contact made with the Au surface.
This explains why the devices made from Au electrodes have high contact resistance. Here,
weak coupling of electronic states may not lead to MIGS and the (8. 0) SWNT is identied
as being semiconducting even after the contact has been set with the Au surface. Because of
weak couplingand hence nite +
c
, the s-SWNTAu contact is reminiscent of the metaloxide
semiconductor (MOS) junction. A small +
B
is estimated for the p-type character. Upon radial
deformation, s decreases and eventually +
c
collapses. Similar features have been observed
recently in STM studies using multiwall carbon nanotube tips and Au(111) samples [180].
However, +
c
has disappeared due to strong coupling between the states of the s-SWNT and
Mo(110) [91]. The height of the Schottky barrier +
B
that forms at the SWNTMo(110) junction
having a nite contact region followed by a free s-SWNT depends on the position where the
Fermi level is pinned in the gap of the s-SWNT. A crude estimate based on the LDOS suggests
that +
B
0.4 eV for the p-type character. The small cross section of the tube prevents us
from identifying a band diagram across the diameter. But the bands are normally bent along
the axis of the s-SWNT and the height of the barrier is monitored via the applied gate voltage.
Finally, we note that the electronic properties of present metalSWNT junctions, in particular
the Fermi level pinning, exhibit marked differences from those of metalSi heterostructures,
which are known to be insensitive to the type of metal.
At this point, we also mention an interesting work by Buldum and Lu [181] on the
intermolecular junctions of SWNTs. They found atomic scale characteristics in the quantum
transport properties of junctions between two SWNTs. In particular, the contact resistance was
strongly dependent on the local atomic conguration of the contact. The optimal electronic
transport between nanotubes occurred when the surfaces of SWNTs were in atomic registry.
However, as illustrated in gure 35, the contact resistance can vary over several orders of
magnitude when the relative atomic conguration at the contact has been changed. Their
predictions were conrmed experimentally [182]. It has been suggested that the monitoring
of contact resistance may lead to new atomic scale switches, resistors and ampliers.
A similar effect was predicted earlier for an Al chain, which is connected to two at Al
substrates to a weaker extent. It was found that the quantumconductance of the chain depends
on the position of the Al atom(whether in a top site or a hollowsite) where the chain is attached
to the electrodes [183].
Metal nanorings. Metal nanorings are of particular interest, because they may lead to the
realization of nanosolenoids on SWNTs. The small radius of a metallic nanoring wrapping
the carbon nanotube may lead to interesting electromagnetic properties. The magnetic eld B
at the centre of the ring can be expressed in terms of the quantized angular momentum L
z
of
the electrons in the direction parallel to the tube axis
B =
j
0
eL
z
4mr
3
. (19)
where r is the radius of the nanoring. Taking the lowest possible value for L
z
and r = 5.9 ,
we estimate B to be of the order of 100 G. The current in the metal ring that can induce
such a high magnetic eld is comparable to the current attained in suspended, monatomic
R950 Topical Review
Figure 35. (a) Amodel of a four-terminal junction formed by crossing two nanotubes. The terminal
labels, rotation angle, O, and the translation directions are shown. The tubes are considered to be
rigid. The current is passing between 1 and 4 and the voltage is measured between 2 and 3. (b) The
contact resistance of the (18. 0)(10. 10) junction as a function of the rotation angle O. The tubes
are in registry at O = 30
, 90
, 150
, 60
, 120
, 180
(E
F
) D
(E
F
))(D
(E
F
) + D
(E
F
)) (P = 86%) for the spin
valve effect. Dag et al [7] have obtained a signicant net magnetic moment for Ti covered
(8. 0), (9. 0) and (6,6) SWNTs.
8. Devices fabricated using carbon nanotubes
The objective of most of the effort made so far in the research on carbon nanotubes has been to
discover a technological application. Nanoelectronic or molecular electronics considered as
ultimate miniaturization [192] has attracted much of the interest. The realization of a single-
molecule device has been challenging due to difculties in manipulating the molecule and
achieving the electrical contact. Intensive laboratory work combined with theoretical studies
during last decade has made remarkable advances in methods and techniques which were able
to produce novel electronic devices based on carbon nanotubes. These devices have kept the
great expectations for these materials alive and have continued to motivate further research.
Some of these devices will be reviewed in this section.
8.1. Transistors based on carbon nanotubes
In 1998 Tans et al [99] achieved the fabrication of a eld effect transistor (FET), which consists
of a semiconducting SWNT connected to two metal electrodes. Asemiconducting Si substrate
covered with a 300 nm layer of thermally grown SiO
2
was used as a back-gate (see gure 37).
The nanotube was able to switch froma conducting to an insulating state at roomtemperature as
Topical Review R953
Figure 38. Two-probe I V
bias
curves for various values of the gate voltage (V
gate
). Data were taken
at room temperature and in vacuum (10
4
mbar) with the voltage applied to contacts 1 (drain)
and 2 (source). A negative V
gate
leads to ohmic behaviour while a positive V
gate
results in a strong
suppression of the current at low bias voltage and nonlinear I V
bias
curves at higher bias. Inset:
conductance at V
bias
= 0 as a function of V
gate
. The conductance through this single-molecule
switch can be varied over at least six orders of magnitude and saturates at 10
6
O
1
for negative
V
gate
. (Reproduced from [99].)
a result of the voltage applied to the gate. As shown in gure 38, at V
gate
= 0, I V
bias
exhibited
a small nonlinearity. However, a pronouncedgap-like nonlinearity developed around V
bias
= 0
with increasing V
gate
. At negative V
gate
, the I V
bias
curve becomes linear with a resistance that
saturates around 1 MO. Achange of 10 Vin V
gate
resulted in modulation of the conductance by
about six orders of magnitude at V
bias
= 0. According to the electronic structure and energy
diagram of the FET modelled by Tans et al in gure 37(c) the difference in work function
will result in bending of the bands of the semiconducting SWNT. A local polarization layer
develops at the interface until the valence band edge of the SWNT aligns with the Fermi
level of the metal electrode. The band bending length was estimated to be 400 nm. It was
assumed that the Fermi level is pinned at the valence band edge of the SWNT at the contact,
since the FET has the same room temperature resistance of 1 MO and a similar temperature
dependence. This behaviour of the device, namely a mild temperature dependence, has been
attributed to the absence of a Schottky barrier. Since the proposed band diagram of the device
was reminiscent of that of the barrier injection transit time diode (BARITT), its operation is
qualitatively described in terms of the BARITT model. It has been argued that using special
geometry, reducing the contact resistance and the thickness of SiO
2
, the device would allow
a maximum frequency of the order of 10 THz. Also its gain can be raised by one order if the
thickness of SiO
2
is decreased from3000 to 50 . It has also been foreseen that the integration
of multiple devices into a circuit would be possible, by molecular self-assembly techniques.
Later, Zhou et al [98] measured I V characteristics of devices obtained with s-SWNTs
contacted by 200 thick Ni electrodes with 600 thick Au on top. The lengths of the SWNTs
between electrodes were less than 30 jm. Degenerately doped Si wafers with 5000 thick
thermally grown oxide on the surface were used as the substrate. The heavily doped substrate
is conducting at low temperature and was used as a back-gate. Electrical transport through an
s-SWNT involved thermal activation at high temperature and tunnelling behaviour through a
R954 Topical Review
reverse biased metalSWNTjunction at lowtemperature. Under high V
bias
, I V characteristics
of s-SWNTs exhibit pronounced asymmetry with respect to the bias polarity, as a result of
local gating. Different samples exhibited I V characteristics resembling that of the Si based
p metaloxidesemiconductor eld effect transistor (MOSFET). The positive voltage gain of
3 recorded in this device was almost one order of magnitude higher than the maximum gain
of 0.35 obtained by Tans et al [99]. The transconductance was measured to be 200 nA V
1
which is two orders of magnitude higher than the similar results obtained by Martel et al [101].
These values indicate great progress.
Both transistors discussed above have been realized by the side bonding of SWNTs. These
devices generally have high contact resistance and unipolar (p-type) transport characteristics.
High contact resistance between the side-bonded s-SWNT and high work-function metal
electrode is in accord with the rst-principles calculations [179]. Martel et al [101] realized
an end-bonded SWNTmetal contact. A FET was made of a semiconducting SWNT of
8000 length and 14 diameter contacting TiC and passivated with a uniform SiO
2
layer.
In contrast to the above planar devices, in the latter device the apparent barrier height for
carrier injection is modulated by the gate eld. This allowed the FET device to be ambipolar
with low contact resistance for both n- and p-type conduction. Devices showing ohmic I V
curves in both strong accumulation and inversion regimes have been attributed to the very
small Schottky barrier at the contact. The low barriers for both electrons and holes have
been attributed to the effective barrier heights being strongly modulated by the electrostatic
eld of the gate. Because of their quasi-one-dimensional structure the electronic behaviour
of nanotube contacts is found to be sensitive to the electrostatics [193]. Fermi level pinning
appears to be relatively unimportant for nanotubes end-bondedto TiC. The needle-like shape of
the metallic carbide contacts focuses the electric eld and hence the gate can induce a stronger
electric eld at the junction. As a result, whereas the true Schottky barrier height is as high as
one expects (i.e. +
B
= E
g
2), it is so thin that carriers can easily tunnel through the barrier.
The symmetry and narrowness of the Schottky barrier for both holes and electrons in a
FET based on an s-SWNT have led to an interesting application. Injection of electrons and
holes simultaneously through the narrow barriers at the source and drain contacts has been
possible at room temperature. These injected carriers have recombined and emitted photons
with energy h = E
g
and wavelength 0.81.5 jm. Because depends on the E
g
of
the s-SWNT, the wavelength of the emitted light can be adjusted easily by using an s-SWNT
with a different radius, or one can tune the emission to any wavelength by radial deformation.
This new result holds promise that SWNTs can enable nanoscale photon sources for future
optoelectronics and bre optics [194].
The charge carrier mobility is crucial for the applications such as in few- and single-
electron memories and chemical/biochemical sensors. Instead of a Schottky barrier as in
previous studies, the work by Durkop et al [195] exhibited an ohmic contact behaviour.
They presented measurements on long, ohmically contacted nanotube devices, showing eld
effect mobility in semiconducting SWNT transistors as high as 79 000 cm
2
V
1
s
1
at room
temperature, and the intrinsic mobility in s-SWNTs is even higher. These measured values
exceed those of any other semiconductor at room temperature.
Beyond the single FET, research on SWNTbased electronics has focused on the integration
of FETs on a chip, since the advance requires assembly strategies that allowprecise localization
and interconnection. Soh et al [197] achieved the synthesis of SWNTs by chemical vapour
deposition (CVD) of methane at controlled locations on a substrate using patterned catalytic
islands. The CVDsynthesis and microfabrication techniques were combined to produce a large
number of ohmically contacted nanotube devices with controllable length, placed on a Si wafer
with 1 jm thick thermally grown SiO
2
. It has been argued that this fabrication technique
Topical Review R955
allows large numbers of integrated nanotube circuits to be built reliably with a contact resistance
of 10 kO (or 20 kO two-terminal resistance in the ballistic regime). Later Franklin et al
[198] devised a method for reliable integration of long suspended SWNTs into electrically
addressable devices, which involves patterned growth of SWNTs to bridge predened Mo
electrodes. This approach affords SWNT based devices without any postgrowth processing.
In the course of active research towards nanoelectronics, Bachtold et al [199] demonstrated
logic circuits with FETs based on SWNTs. Their transistors show device characteristics such
as high gain (>10), a large onoff ratio (>10
5
) and room temperature operation.
Much recently, Keren et al [196] reported the realization of a self-assembled carbon
nanotube FET operating at room temperature where a DNA scaffold molecule provides the
address for precise localization of an s-SWNT as well as a template for the extended metallic
wires contacting it. In order to get the transistor to self-assemble, they attached a SWNT
onto a specic site on a DNA strand, by binding together with proteins. They then made gold
nanowire out of DNA molecules at each end of the nanotube. The device could be switched
on and off by applying voltage to it.
8.2. Chemical sensors
As discussed in section 3, the adsorption of atoms or molecules such as NO, NH
3
and O
2
on
s-SWNTs has been reported to change the electrical resistance and similar properties. It has
been suggested that this effect of carbon nanotubes can be used in chemical sensors to detect
specic molecules in the environment [62, 63]. Owing to their bulky size and high power
consumption, the miniaturization of chemical sensors is, in fact, very important for various
applications. Modi et al [200] achieved production of a miniaturized gas ionization sensor a
few microns in length using MWNTs. The sharp tips of MWNTs generate very high electric
eld at relatively low voltages, lowering breakdown voltages several times in comparison
to those of traditional electrodes. This feature enables compact, battery powered and safe
operation of such sensors. In this sensor, controlled dc voltage is applied between an anode
consisting of vertically aligned MWNT lm and a cathode made from Al sheet. A MWNT
lm (consisting of MWNTs 250300 in diameter, 30 jm long, with 500 separation)
was grown by CVD on a SiO
2
substrate. An individual MWNT, attached to the substrate in
the head-on position, has a tip radius of 150 , which, in turn generates a very high nonlinear
electric eld. In this way the formation of a corona is promoted at relatively lowvoltage, which
can be provided by special batteries. The breakdown voltage was found to become lower as
the interelectrode spacing was reduced. The breakdown voltage went down from 280 V (at
150 jm) to 130 V (at 25 jm separation). Such voltages can be easily obtained by series
connection of carbonzinc batteries. That makes the sensor portable. In the operation of the
sensor, it is found that while the breakdown voltage is independent of the gas concentration, the
current discharge is linearly dependent on the concentration. This sensor and the underlying
technique that it uses are very convenient, because: (i) the precise breakdown voltage provides
the ngerprint for the gas to be identied; (ii) by monitoring the self-sustaining discharge
current, the gas concentration can be determined; (iii) the response time is short [200].
9. Conclusions
Because of growing interest in nanotechnology and the quest for novel nanodevices, study of
carbon nanotubes has seen a tremendous explosion of work. Despite the large number of very
important research works, the scope of this reviewhas been focused on a limited eld covering
atomic and electronic structure of single-wall carbon nanotubes.
R956 Topical Review
In this review article, various physical properties of carbon nanotubes and specic
applications using these properties have been investigated. Our discussions have been guided
by rst-principles calculations within density functional theory, which were proven to be
appropriate and to yield accurate predictions on the atomic and electronic structure of carbon
nanotubes.
It has been shown that bare nanotubes have many important properties which are suitable
for use in potential technological applications. That a (n. m) SWNT can be metallic or
semiconducting depending on n and m offers a wide range of options. Moreover, these options
are multiplied several times by the functionalization of tubes. Here we dealt with specic
cases wherein the properties of the nanotube undergo a dramatic change upon physisorption of
foreign atoms. Not only functionalization but also radial deformation has been seen to modify
the nanotube properties. A semiconducting nanotube can be metallic under a radial strain
which changes the circular cross section into an elliptical one. Even more interesting is that
these changes can be tuned reversibly. Finally, we discussed howproperties of bare nanotubes
as well as those induced by functionalization or radial deformation can be exploited for the
design of new molecular devices. Transistors and chemical sensors fabricated by using carbon
nanotubes, as explained at the end, show that carbon nanotubes are indeed nanostructures
which may potentially allow several technological applications.
Acknowledgments
This work was partially supported by the National Science Foundation under Grant No INT01-
15021 and T
UB