Appl. Phys. A 74, 339–343 (2002) / Digital Object Identifier (DOI) 10.

1007/s003390201277
Applied Physics A
Materials
Science & Processing

Thermal properties of carbon nanotubes and nanotube-based materials

J. Hone1,∗ , M.C. Llaguno1 , M.J. Biercuk1 , A.T. Johnson1 , B. Batlogg2 , Z. Benes3 , J.E. Fischer3
1 Department of Physics and Astronomy and Laboratory for Research on the Structure of Matter, University of Pennsylvania, Philadelphia PA 19104-6272, USA
2 Bell Laboratories, Lucent Technologies, Murray Hill, NJ 079743, USA
3 Department of Materials Science and Engineering and Laboratory for Research on the Structure of Matter, University of Pennsylvania, Philadelphia PA
19104-6272, USA

Received: 17 October 2001/Accepted: 3 December 2001/Published online: 4 March 2002 –  Springer-Verlag 2002

Abstract. The thermal properties of carbon nanotubes are di- phonon bandstructure of the sheet “folds” into the 1D band-
rectly related to their unique structure and small size. Because structure of the tube. Secondly, the cylindrical shape of the
of these properties, nanotubes may prove to be an ideal mate- tube renders it stiffer than the sheet, changing the dispersion
rial for the study of low-dimensional phonon physics, and for of the lowest-lying modes.
thermal management, both on the macro- and the micro-scale. Figure 1 shows the theoretically derived low-energy
We have begun to explore the thermal properties of nano- phonon bandstructure of an isolated (10, 10) nanotube [1].
tubes by measuring the specific heat and thermal conductivity The 1D quantized nature of the bandstructure is evident: there
of bulk SWNT samples. In addition, we have synthesized are a series of 1D “sub-bands” separated at the zone center
nanotube-based composite materials and measured their ther- by energies of a few meV. There are four acoustic bands; one
mal conductivity. longitudinal (LA), two degenerate transverse (TA), and one
The measured specific heat of single-walled nanotubes “twist”, all of which have linear dispersion at low energy. The
differs from that of both 2D graphene and 3D graphite, es- high phonon band velocity (vLA = 24 km/s, vTA = 15 km/s,
pecially at low temperatures, where 1D quantization of the
phonon bandstructure is observed. The measured specific heat
shows only weak effects of intertube coupling in nanotube
bundling, suggesting that this coupling is weaker than ex-
pected. The thermal conductivity of nanotubes is large, even
in bulk samples: aligned bundles of SWNTs show a thermal
20
conductivity of > 200 W/m K at room temperature. A linear
K(T ) up to approximately 40 K may be due to 1D quantiza-
tion; measurement of K(T ) of samples with different average
nanotube diameters supports this interpretation.
Nanotube–epoxy blends show significantly enhanced 15
E (meV)

thermal conductivity, showing that nanotube-based compos-

ites may be useful not only for their potentially high strength,
but also for their potentially high thermal conductivity.
10
PACS: 62.25.+g; 63.22.+m; 61.46.+w

5
1 Specific Heat

A carbon nanotube can be thought of as a single graphene

sheet that is wrapped into a cylinder. Wrapping the sheet has 0
two major effects on the phonon bandstructure. Firstly, the 2D 0 0.1 0.2 0.3 0. 4
k(1/Å )
∗ Corresponding author. Fig. 1. Low-energy phonon bandstructure of a (10, 10) carbon nanotube [1].
Present address: Department of Physics, California Institute of Technology, The inset shows the phonon density of states (PDOS) of an isolated nano-
Pasadena CA 91125, USA tube (solid line) compared to the PDOS of graphene (dot–dashed line) and
(Fax: +1-626/683-9060, E-mail: hone@caltech.edu) graphite (dashed line)
340

vtwist = 9 km/s), coupled with the small diameter of the nano-

tube, causes the relatively large sub-band splitting.
The inset shows the low-energy phonon density of states
(PDOS) of a (10, 10) nanotube (solid line) compared to that
of 2D graphene (dot-dashed line) and 3D graphite (dashed
line). The nanotube PDOS is constant at the lowest energies,

C (mJ/gK)
and then increases stepwise with the entry of higher sub-
bands. As the system is 1D, there is a van Hove singularity at
each band edge. The graphene PDOS is large in this energy
range because of the existence of an out-of-plane transverse
mode with quadratic dispersion. The PDOS of graphite, how-
ever, is significantly reduced because of the added coupling
between neighboring layers, which renders the system 3D
rather than 2D, and shifts low-energy spectral weight upward
in energy.
Figure 2 shows the calculated low-temperature heat spe-
cific heat of an isolated nanotube. Because the PDOS is con-
stant at low energy, the specific heat displays linear tempera- Fig. 3. Predicted specific heat of a graphene sheet, an isolated nanotube,
ture dependence at low temperature because only the acoustic graphite, and a nanotube rope
modes are populated [2]. Above ∼6 K, the slope of C(T)
increases as the optical sub-bands become populated. This
linear behavior, with an increase in slope near 6 K, is the graphene sheets. Comparing 2D graphene to 3D graphite, we
expected signature for a 1D quantized phonon spectrum in can see that below approximately 60 K, interlayer coupling
single-walled nanotubes. causes C(T ) to decrease much more strongly with tempera-
As can be seen in the case of graphene and graphite, ture. The isolated SWNT displays a smaller C(T ) at low
changing the dimensionality of a system can greatly affect the temperature compared to the graphene sheet due to the ab-
low-energy PDOS and therefore the low-temperature specific sence of the quadratic out-of-plane mode-tubes, which are
heat. In graphite, the range of this effect is related to the De- stiffer to bending than sheets. Finally, the strongly coupled
bye energy of the interlayer modes, approximately 10 meV. rope curve diverges below the single-tube C(T ) below 30 K,
In nanotubes, it might be expected that bundling tubes into following a curve similar to that of graphite. It is clear from
3D crystalline arrays (“ropes”) would reduce the low-energy Fig. 3 that graphite-like coupling between neighboring tubes
PDOS. Mizel et al. [3] have calculated the low-energy phonon in a SWNT bundle should cause the signature of 1D quan-
bandstructure of SWNT bundles, and find a relatively high tization in the specific heat to be obscured. However, in the
intertube Debye energy E⊥ D , approximately 5 meV, for the case of weaker coupling, single-tube behavior might persist to
case where neighboring tubes have graphite-like (“strong”) lower temperatures.
coupling. The SWNT samples used for measurement of spe-
Figure 3 shows the predicted specific heat of 2D graphene, cific heat [4] were obtained from purified SWNT suspen-
3D graphite, 1D isolated SWNTs and 3D SWNT ropes, show- sions (tubes@rice), and subsequently purified and vacuum-
ing the dramatic effects of differing interlayer coupling. At annealed. The sample was composed of large bundles of
high temperatures, all of the specific heats are identical and SWNTs with an average tube diameter of 1.25 nm and the
mostly reflect the phonon structure of the constituent 2D residual catalyst concentration was approximately 2 at. %,
as determined by X-ray diffraction (XRD), high-resolution
transmission electron microscopy (HRTEM) and energy-
dispersive X-ray spectroscopy (EDX) measurements. After
analysis, the samples were baked at 300 ◦ C under dynamic
vacuum for three days to remove atmospheric contaminants,
and then kept under vacuum until minutes before the meas-
urements were begun. Specific heat was measured from
300 K to 2 K using a relaxation technique.
Figure 4 shows the measured specific heat of a SWNT
sample from 2 K to 300 K. The hollow circles represent the
raw data. The solid line represents the contribution from the
catalyst, based on the known specific heat of Ni and Co. Fi-
nally, the solid circles represent the corrected specific heat of
the SWNTs. Figure 5 shows the measured specific heat on
a logarithmic scale, compared to the predicted specific heat
of graphene, graphite, isolated SWNTs and strongly coupled
ropes. The measured data follows the isolated SWNT curve
all the way down to 4 K. Thus, existing models of the phonon
bandstructure of SWNTs are largely consistent with the meas-
ured data. The data diverge below the single-tube curve at
Fig. 2. Calculated single-nanotube specific heat 4 K, rather than at 30 K, as expected for the case of graphite-
 341

C (mJ/gK)

C (mJ/gK)
0 2 4 6 8
Fig. 4. Measured specific heat of a purified SWNT sample [4]. The raw data
is corrected for the contribution from the catalyst impurities to obtain the
contribution from the SWNTs Fig. 6. Measured specific heat at low temperature; fit using a simple model
to account for weak intertube interactions [4]

dot-dashed line; the solid line represents the sum of the

two contributions. E D  , E D⊥ , and E sub are taken as indepen-
dent fitting parameters, and adjusted to give the best fit to
the measured data. The values obtained are E D  = 92 meV,
E D ⊥ = 1.4 meV, and E sub = 4.1 meV.
The theoretical acoustic mode velocities for a SWNT [1]
C (mJ/gK)

translate into an effective Debye energy of 103 meV, only

slightly higher than the fitted 92 meV. Our fitted E sub
(4.1 meV), however, is considerably larger than the theoretical
single-tube value of 2.7 meV. The first optical sub-band cor-
responds to tube flattening, and should require significantly
more energy in a rope since tubes are constrained by their
neighbors; theoretical calculations [5] that take radial tube-
tube interactions into account show excellent agreement with
the experimental value. The experimental tube–tube coup-
10
ling, measured by E D ⊥ = 1.2 meV, is significantly smaller
than the theoretical value of approximately 5 meV [3] ob-
Fig. 5. Measured specific heat compared to theoretical models [4]. The data tained using coupling constants derived from graphite. The
agree with the predicted C(T ) of an isolated nanotube down to 4 K, imply-
ing a weak intertube interaction difference may be related to the lack of commensurability
between neighboring tubes, which would imply a dramatic
weakening of the corrugation in the intertube potential, so
like coupling. Therefore, we conclude that SWNTs are much that tubes in a real rope may slide or twist more freely than
more weakly coupled mechanically than might be expected; it expected.
should be possible to observe the 1D–2D transition character- The measured high on-tube Debye energy confirms, in
istic of single tubes. a bulk sample, the high Young’s modulus previously observed
The solid points in Fig. 6 show the low-temperature meas- for individual tubes [6]. The weak tube–tube coupling, how-
ured specific heat on a linear scale. The data clearly show ever, implies that the mechanical strength of SWNT ropes
a linear temperature dependence from 2 K to 8 K, with an will be relatively poor. It may be necessary to crosslink tubes
increase in slope above 8 K. This behavior is direct confirma- within a rope, or to separate them completely, in order to re-
tion of a 1D-quantized phonon spectrum in SWNTs. How- alize their near-ideal properties in high-strength composites.
ever, the linear slope does not extrapolate to zero at T = 0, as However, weak coupling may be an advantage for high ther-
would be expected for isolated SWNTs. We attribute this dis- mal conductivity. Berber et al. [7] find that strong tube–tube
crepancy to intertube coupling, which should cause T 3 -like coupling decreases the high-temperature thermal conductiv-
behavior at low temperature. ity of SWNT bundles by an order of magnitude relative to
The lines in Fig. 6 show the results of employing a sim- isolated tubes; weak coupling may imply no significant re-
ple model to simulate the behavior of weakly coupled SWNT duction in the thermal conductivity when tubes are bundled
ropes. In this model, the acoustic modes are collapsed into ropes. Similarly, in composites, the inner tubes in a rope

onto a single mode with Debye energy E D in the on-tube should be relatively unperturbed by the surrounding matrix,
direction, and transverse Debye energy E D⊥ , with a spe- which could also be an advantage for high thermal conduc-
cific heat represented by the dashed line. A single optical tivity. The issues of commensurability that were raised as an
mode enters at E sub , with specific heat represented by the explanation for the weak tube–tube mechanical coupling also
342

have implications for the electronic coupling between neigh- tization, in which only the acoustic modes of the tube carry
boring SWNTs in a rope [8]. any heat flow. However, the role of intertube contacts on the
temperature dependence of K(T ) is unknown. In order to
more definitively determine whether the linear K(T ) is due to
2 Thermal conductivity 1D quantization, we have measured K(T ) for samples with
different nanotube diameters. Because the phonon sub-band
As diamond and graphite display the highest known ther- splitting increases with decreasing tube diameter, we expect
mal conductivity at moderate temperatures, it is likely that that the linear K(T ) should extend to higher temperatures in
nanotubes should be outstanding in this regard as well. In- samples with a smaller average tube diameter.
deed, recent theoretical work [7] has predicted that the room- Figure 8 shows the thermal conductivity divided by tem-
temperature thermal conductivity of nanotubes is as high as perature for four SWNT samples [11]. The samples were
6600 W/m K. In addition, at low temperature, the thermal synthesized by laser ablation at differing oven temperatures in
conductivity should show the effects of 1D quantization just order to produce different average nanotube diameters. Two
as is seen in the specific heat. The thermal conductivity in samples synthesized at 1100 ◦ C, with an average diameter
a highly anisotropic material is most sensitive to the high- of 1.4 nm, and two samples synthesized at 1200 ◦ C, with an
velocity and high-scattering-length phonons. Therefore, it is average diameter of 1.2 nm, were measured. All four sam-
likely that even in nanotube bundles, the thermal conductiv- ples show a linear K(T ) at low temperature, as shown by
ity should directly probe on-tube phonons and be insensitive the constant value of K/T (normalized to 1 here for all sam-
to inter-tube coupling. ples). For the 1.4 nm diameter samples (open symbols), K/T
Figure 7 shows the measured temperature-dependent ther- begins to increase at approximately 35 K, while a similar in-
mal conductivity of bulk samples of SWNTs that have been crease is not seen for the 1.2 nm samples (filled symbols)
aligned by filtration in a high magnetic field [9]. In the align- until approximately 40 K. This behavior is consistent with our
ment direction, the room-temperature thermal conductivity expectations for a 1D quantized thermal conductivity. A puz-
is greater than 200 W/m K, which is comparable to a good zling inconsistency, however, is that the linear K(T ) extends
metal and within an order of magnitude of that of highly crys- to approximately 40 K while the linear C(T ) extends only
talline graphite or diamond. The thermal conductivity of un- to approximately 8 K. If the phonon scattering time is rela-
aligned samples is about one order of magnitude smaller [10]. tively constant for all modes, these temperatures should be
However, the temperature dependence of the thermal con- roughly equal. One possible explanation for this discrepancy
ductivity is roughly the same in both types of sample. Also, is that the optical sub-bands of the nanotube scatter much
in both types of sample, simultaneous measurement of the more strongly than the acoustic sub-bands, so that their in-
electrical and thermal conductivity shows that the electronic fluence on the thermal conductivity is suppressed until higher
contribution to K(T ) is negligible at all temperatures. temperatures are reached. Clearly, more experimental and
Below approximately 40 K, the thermal conductivity dis- theoretical work is necessary in order to fully understand this
plays a strictly linear temperature dependence in all samples. behavior.
This temperature dependence is likely to be due to 1D quan- Composite materials having high thermal conductivity
have a number of potential applications, particularly in heat
sinking for electronics and motors. To explore the potential
K/T (arb. units)
K (W/m-K)

Fig. 8. Thermal conductivity divided by temperature, K/T , of SWNT sam-

ples with different average diameters [11]. The range of linear K(T ), i.e.
Fig. 7. Thermal conductivity of a bulk sample of SWNTs in which the tubes constant K/T , extends to higher temperatures in samples with a smaller
are aligned by filtration in a strong magnetic field [9]. The measurement is diameter, as would be expected for a scenario of 1D quantization of the
taken in the direction parallel to the tubes phonon structure
 343

ing, with a 120% enhancement at 1% loading. In addition,

nanotubes seem to be superior to VGCF as a filler material.
This initial result demonstrates that nanotubes are, in fact,
an excellent filler for making high-thermal-conductivity com-
posites.

Acknowledgements. This work was supported by NSF Grant No. DMR-

9802560, DOE Grant No. DEFG02-98ER45701 and the Laboratory for
Research on the Structure of Matter MRSEC, No. DMR00-79909.

References

1. R. Saito, G. Dresselhaus, M.S. Dresselhaus: Physical Properties of

Carbon Nanotubes (Imperial College Press, London 1998)
2. L.X. Benedict, S.G. Louie, M.L. Cohen: Solid State Comm. 100, 177
(1996)
3. A. Mizel, L.X. Benedict, M.L. Cohen, S.G. Louie, A. Zettl, N.K. Bu-
draa, W.P. Beyermann: Phys. Rev. B 60, 3264 (1999)
Fig. 9. Thermal conductivity enhancement for epoxy samples with varying 4. J. Hone, B. Batlogg, Z. Benes, A.T. Johnson, J.E. Fischer: Science 289,
loading of SWNTs and vapor-grown carbon fibers (VGCF) [12] 1730 (2000)
5. D. Kahn, J.P. Lu: Phys. Rev. B 60, 6535 (1999)
6. W. Teizer, R.B. Hallock, E. Dujardin, T.W. Ebbesen: Phys. Rev. Lett.
of using nanotubes for such applications, we have synthe- 82, 5305 (1999)
7. S. Berber, Y.K. Kwon, D. Tomanek: Phys. Rev. Let. 84, 4613 (2000)
sized nanotube-based composites by mixing as-grown nano- 8. A.A. Maarouf, C.L. Kane, E.J. Mele: Phys. Rev. B 61, 11 156 (2000)
tube soot into industrial epoxy (Shell Chemicals Epon 862 9. J. Hone, M.C. Llaguno, N.M. Nemes, A.T. Johnson, J.E. Fischer,
epoxy resin) [12]. As a comparison, highly graphitic vapor- D.A. Walters, M.J. Casavant, J. Schmidt, R.E. Smalley: Appl. Phys.
grown carbon fibers (VGCF) were mixed into the same resin. Lett. 77, 666 (2000)
Figure 9 shows the measured room-temperature thermal con- 10. J. Hone, M. Whitney, C. Piskoti, A. Zettl: Phys. Rev. B 59, R2514
(1999)
ductivity enhancement for samples with 0–1 wt % nanotubes, 11. M.C. Llaguno, J. Hone, J.E. Fischer, A.T. Johnson: in press
and 0–2 wt % VGCF. The nanotube samples show an increas- 12. M.J. Biercuk, M.C. Llaguno, M. Radosavljevic, J.K. Hyun, A.T. John-
ing thermal conductivity enhancement with increasing load- son, J.E. Fischer: in press