Proceedings of HT–FED 2004
ASME Heat Transfer/Fluids Engineering Summer Conference
July 11–15, 2004, Charlotte, North Carolina, USA
HT–FED2004–56786
THERMAL, ELECTRICAL AND MECHANICAL characterization of carbon nanotube–epoxy
composites
1 1
Vivek Sundaram and Roop L. Mahajan
1 Department of Mechanical Engineering,
University of Colorado at Boulder,
Boulder, CO 80309-0427
mahajan@spot.colorado.edu
ABSTRACT
In this paper, we propose carbon nanotube–epoxy
composites as an underfill material for flip-chip on board
(FCOB) applications. Purified single-walled carbon nanotubes
(SWNTs) are considered, and the matrix selected is a typical
underfill epoxy used in flip-chip on board applications. Three
techniques: mechanical separation, mechanical dispersion in
alcohol and ultrasonic dispersion were explored to embed the
carbon nanotubes (CNTs) in the epoxy matrix. Dispersion of
the nanotubes in the epoxy using the three techniques was
investigated via field emission scanning electron microscopy
(FE-SEM). Of the three methods, ultrasonic dispersion
produced the best results for a uniform dispersion of CNTs in
epoxy matrix. Isothermal Dynamic Scanning Calorimetry was
then performed to ascertain whether the addition of CNTs into
the polymer matrix affected the curing characteristics of the
epoxy. The thermal, electrical and mechanical properties of
nanotube–epoxy composites were measured as a function of
SWNT content in the matrix. The results reveal improved
thermal, electrical and mechanical properties with increase in
carbon nanotube concentrations.
INTRODUCTION
Nearly 12 years ago, Sumio Iijima at NEC, Japan, first
noticed odd nanoscopic threads lying in a smear of soot. Made
of pure carbon, as regular and symmetric as crystals, these
exquisitely thin, impressively long macromolecules soon
became known as nanotubes [1,2]. Carbon nanotubes (or
CNTs) can in principle play the same role as silicon does in
electronic circuits but at a molecular scale, where silicon and
other standard semiconductors cease to work [3,4]. These
materials exhibit unusually high thermal [5] and electrical
conductivity [6,7,8] in addition to superior mechanical [9,10]
and optical properties [11]. This uniqueness of nanotubes has
instilled interest among many researchers to investigate these
novel materials. A few of the earlier studies have revealed that
the properties of CNTs are highly affected by impurities [12]
resulting from growth processes and by voids and vacancies
1,2 2
Dudley S. Finch and Stephanie A. Hooker
2 Materials Reliability Division
National Institute of Standards and Technology
Mailcode 853.00
325 Broadway, Boulder, CO 80305-3328
finchd@boulder.nist.gov; shooker@boulder.nist.gov;
formed due to the migration of atoms in the lattice structure
[13]. As a result, the potential of CNTs for different
applications has not yet been fully realized.
Recent advances and enhancements in the synthesis of
nanotubes have made possible the growth of high-purity
crystalline bundles of CNTs [14,15]. In parallel, significant
progress has also been made in determining thermal and
electrical properties of CNTs including molecular dynamics
simulations [16] and measurement of thermal conductivity of
an individual CNT using a MEMS device [17]. In our research
group, we have been exploring the potential of carbon
nanotube–epoxy composites both as a thermal interface
material and as an underfill material in flip-chip applications.
In this paper, we focus on the second application. An
objective is to determine the fraction of carbon nanotube in the
resin matrix that would result in optimum thermal, mechanical
and electrical performance. To this end, the thermal, electrical
and mechanical properties of nanotube–epoxy composites have
been measured for different weight percentages of nanotubes.
NOMENCLATURE
k – thermal conductivity, σ - electrical AC conductivity, E –
Young’s Modulus
1. SAMPLE PREPARATION
Purified single-walled carbon nanotubes (SWNTs)
purchased from Carbon Nanotechnologies Inc. were used to
prepare the nanotube–epoxy composites. The type of epoxy
used for specimen preparation is a common underfill material
used in flip-chip packages. Three techniques: mechanical
separation, mechanical dispersion in alcohol and ultrasonic
dispersion have been explored to embed the carbon nanotubes
in the epoxy matrix. As a first step, the effectiveness of each
technique at dispersing the CNTs was evaluated using high
resolution Field Emission Scanning Electron Microscopy (FE-
SEM). Typical electron micrographs for each method are
shown in figure 1.
1 Contribution of the National Institute of Standards and
Technology, not subject to copyright in the U.S.
 All specimens evaluated have 1 percent by weight of CNT
in epoxy. For each preparation method, two randomly selected
portions of the same composite sample are shown. Referring to
figure 1(a), we observe that the two portions show very
different distributions. The portion shown on the right hand
side indicates a clustered distribution of CNTs in the epoxy
matrix while another portion of the same sample shown on the
left is devoid of CNTs. Similarly, the SEM micrograph in
figure 1(b) of the sample prepared by the second method shows
nanotube aggregates in the picture on the left while the scan on
the right is again noted for absence of CNTs. However, in
figure 1(c), both portions show a much more even, non-
clustered distribution of nanotubes in the matrix. Multiple
scans of different portions of the samples produced by three
different methods confirmed the behavior described above.
Almost no CNTs CNTs seen as clusters
Figure 1(a): first method of mixing - mechanical dispersion in
epoxy matrix using mortar & pestle
CNT Aggregates Almost no CNTs
Figure 1(b): second method of mixing - mechanical mixing by
manual stirring in alcohol
CNTs more uniformly
CNTs more uniformly
distributed dispersed
Figure 1(c): third method of mixing - mechanical separation
and crushing using mortar & pestle, ultrasonic dispersion and
finally mechanical mixing of alcohol-dried CNT in the matrix
Once the superiority of the third method was established,
this preparation procedure was used to produce epoxy -CNT
composite samples for characterization as follows. We note,
however, that even though the ultrasonic dispersion gave the
best results of the three, the composites were still far from
desired dispersion. Insulated clusters could still be present
throughout the sample.
The carbon nanotubes were dispersed ultrasonically for as
long as two hours in methanol after they had been mechanically
crushed and separated usin g a mortar and pestle to promote the
formation of a stable suspension. The alcohol was allowed to
evaporate overnight leaving a black residue of carbon particles
behind. The base matrix is a high-purity, low-stress liquid
epoxy. It has a very high flow to facilitate penetration of gaps
as small as 1 mil (25 µm) under the chip. When fully cured the
material forms a very rigid and low-stress seal. The alcohol-
free CNT residue was subsequently mixed with the epoxy
mechanically using a mortar and pestle to enable uniform
dispersion. The uncured composite mixture was then
transferred to specially fabricated flexible 3110 RTV silicone
rubber molds of varying sizes (shown in figure 2) to obtain
desired samples of suitable dimensions.
mm
25.4
25.4 mm
53.98 mm
30.12 mm
5 mm
30 mm
Figure 2: Specially fabricated silicone molds for making epoxy -
CNT composites
2 Contribution of the National Institute of Standards and
Technology, not subject to copyright in the U.S.
 First, the molds were placed in an oven under vacuum and conductivity response to frequency is the same as that of the
heated at 75ºC for 20 minutes to remove trapped air from the matrix (i.e., the conductivity increases with increase in
mixture. Subsequently, the specimens were cured at 150ºC for frequency). Such a behavior is typical of insulators. It appears
two hours to get fully cured CNT-epoxy composite samples.
The mold sizes used for the different tests were: (1)
1.00E+00
Mechanical: dog-bone shaped samples of dimensions 30 mm x 1.00E -01
5 mm x 3 mm for tensile tests; (2) Electrical: 25.4 mm x 25.4 1.00E -02

AC Conductivity, S/m
mm x 0.99 mm; (3) Thermal: 54 mm x 30.1 mm x 1.5 mm thick 1.00E -03

as an interface material. 1.00E -04

0 wt% CNT
0.5 wt% CNT
1.00E -05 1 wt % CNT

2. EFFECT OF CURE KINETICS 1.00E -06

2 wt % CNT
4 wt % CNT
5 wt % CNT
Isothermal dynamic scanning calorimetry (DSC) was 1.00E -07

performed to investigate the effect of addition of carbon 1.00E -08

nanotubes to the curing reaction of the polymer matrix. The 1.00E -09

epoxy came in syringes which were stored at -100ºC in a deep 1.00E -10

freezer. At this low temperature the epoxy is in the glassy 10 100 1000 10000 100000 1000000 10000000

state, and the reaction is diffusion controlled, and therefore
extremely slow to occur. Thus, it can be safely assumed that
the epoxy was initially uncured. Five milligrams of epoxy was
used as the base matrix for each run. The epoxy was cured in-
situ in DSC aluminum pans at a constant rate of 40ºC/min from
room temperature to 150ºC and held at 150ºC for two and a half
hours. The sample was cooled at a constant rate of 40ºC/min
back to room temperature. It was found that there was no
evidence of a change in the kinetics of the curing reaction due
to the addition of carbon nanotubes to the polymer matrix.
3. ELECTRICAL CONDUCTIVITY AND PERMITIVITY
MEASUREMENTS
The DC resistance of each specimen was first measured
using a digital multimeter to determine if the specimens were
electrically conductive. Next, 2-terminal AC impedance
spectroscopy measurement was performed to evaluate the
electrical behavior of the different materials as a function of
frequency. In this measurement, an impedance analyzer was
used to acquire data on both the real and imaginary parts of the
complex impedance (Z*) and admittance (Y*) from 100Hz to
1MHz. AC conducitivity was then calculated for each
specimen using the following formula:
Frequency, Hz
Figure 3: Effect of frequency on electrical conductivity for
different weight percentages of CNTs in epoxy
figure 3 that this insulating behavior is exhibited up to a value
between 1 and 2 weight percent of CNT. For weight fraction
greater than 2%, the conductivity is unchanged with frequency
which corresponds to the behavior expected of conductors.
Secondly, examining the results at a given frequency as shown
in figure 4, we note that the conductivity increases with
increase in nanotube percentage in the epoxy matrix, and this
trend becomes asymptotic at approximately 4 weight percent.
By comparison, increase in conductivity of the 2 weight percent
samples compared to that of the material is more than 4 orders
of magnitude. At 5 weight percent, the conductivity improves
only by an order of magnitude compared to 2 weight percent
samples. Thus at higher CNT concentrations, there is no
appreciable increase in conductivity, indicating saturation.
CNT Concentration, % by weight
1.00E+00
0 1 2 3 4 5 6
1.00E-01

Y '×d 1.00E-02

σ =
AC Conductivity, S/m

(1) 1.00E-03
A
1.00E-04
where, Y ' is the real part of the complex admittance, d is the
distance between electrodes (i.e. the average thickness), and 1.00E-05

A is the contact area. Two electrical leads were attached to 1.00E-06

each specimen for measurement. The impedance analyzer was 1.00E-07

then compensated internally for the resistance of these leads, as
well as for that of the measurement fixture. Compensation was
performed prior to each measurement to reduce error. DC
measurements resulted in resistances greater than 50M Ω for 0
and 0.5 weight percent specimens. For 1, 2, 4 and 5 weight
percent CNT samples, the resistances were ~7M Ω , ~1M Ω ,
~20K Ω and ~5K Ω respectively. The clear indication is an
increase in electrical conductivity with increase in CNT
concentration.
A similar trend is revealed by the AC measurements. A
plot of the AC conductivity as a function of frequency is shown
in the figure 3. The conductivity was calculated at different
frequencies ranging from 100Hz to 1MHz. We first note that
for 0 and 0.5 percent by weight CNT composites, the
1.00E-08
1.00E-09
Figure 4: Variation of electrical conductivity as a function of
CNT weight percent in the matrix at 1000Hz
4. THERMAL CONDUCTIVITY MEASUREMENTS
As before, the samples used for the measurements were 0,
0.5, 1, 2, 4 and 5 percent by weight of carbon nanotubes in the
epoxy. A standard sandwich experiment was used to measure
the thermal conductivity as a function of weight percent for a
constant input power of 13 watts. The experimental setup
shows a solid aluminum block of dimensions 54 mm x 30.1
mm x 76.2 mm weighing about 600 grams with patch heater

3 Contribution of the National Institute of Standards and

Technology, not subject to copyright in the U.S.
attached to the top end of the block. Thermocouples were
attached at three equidistant spots, one at 6.4 mm from the top
end, a second one at 38.1 mm from the top and a third one 70
mm from the top end on the aluminum block. A square copper
plate of dimensions 127 mm2 was maintained at 8°C by
circulating water from a constant temperature bath. The weight
exerted by the aluminum block helped minimize the air gaps
between the aluminum and the copper plate.
V
Patch Heater
T1
values determined through Eq. (3) as the accurate estimates of
thermal conductivity.
The general trend is that that the thermal conductivity
increases with increase in weight percent of CNTs. At 5 weight
percent of CNT, the thermal conductivity of the composite is
approximately 200% of that of the matrix. The drop
experienced from 0 (k=0.866 W/m-K, thermal conductivity of
pure matrix) to 0.5 weight percent (k=0.579 W/m-K) is an
anomaly. We believe that this drop is due to the presence of air
bubbles that might be trapped during the preparation of the
composites due to the introduction of CNTs in the continuous,
rigid crosslinks of the thermoset epoxy.

Aluminum Thermal Conductivity Measurements

Thermocouples T2
INPUT POWER = 13 W
2
CNT-Epoxy
Interface 1.8
1.75
T3 Copper 1.6
(Constant
T4 T3’
Temperature 1.4
1.447

Thermal Conductivity, W/m.K

sink)
1.2
Plexiglass filled with
water 1
0.932
0.866
0.8
0.755

0.6 0.579

0.4
Re-circulator
0.2

0
0 1 2 3 4 5 6
Figure 5: Schematic of thermal conductivity experimental setup % by wt. of CNT in epoxy matrix

In a typical run, the copper plate is first cooled down to
8°C by circulating water through the plexiglass enclosure.
After steady state is reached, the patch heater is switched on
and set to a constant input power. At steady state, the
temperatures are recorded by the thermocouples T1, T2, T3 and
T4. The temperature, T3, is extrapolated to the tip of the
aluminum at the junction between the aluminum and the
composite. This temperature is denoted as T 3′ . Because the
temperature, T4, is measured on the surface of the copper plate,
it is directly used to calculate:
∆T = T 3′ − T 4 (2)
The thermal conductivity of the interface is then calculated as
follows:
Q×l
k= (3)
A × ∆T
where, l is the thickness of the composite interface = 1.5 mm
Figure 6: Variation of thermal conductivity as a function of
weight percent of CNTs in the matrix
We also note that the CNTs are randomly oriented in the
matrix and do not provide a continuous thermal path. As a
result, the potential thermal conductivity improvement may not
have been realized.
5. MECHANICAL PROPERTY MEASUREMENTS BY
TENSILE TESTS
Tensile tests of dog-bone shaped samples were performed
at room temperature (21ºC) and at a constant cross-head speed
of 1 mm/min on an INSTRON electro-mechanical tester. The
strain was measured with an extensometer. The gauge length is
12.5 mm ± 2.5 mm, and the thickness of the sample is 3 mm.
Most of the tensile tests were performed to the limit strain of
the extensometer while those of the 4 and 5 weight percent
composite samples were conducted until the specimens snapped
and broke. The stress-strain curves are shown in figure 7.

The results are plotted in figure 6. We note that the value of k

obtained in Eq. (3) is truly a measure of the overall
conductance that includes contact resistance at the two
interfaces. However, a comparison of the measured value for
the pure epoxy in this experiment (0.866 W/m-K) with k = 0.87
W/m-K reported in [18] clearly indicates that the thermal
resistance is almost negligible. Assuming that in the
subsequent tests at different CNT concentrations the thermal
interface resistance remains the same, it is fair to take the k

4 Contribution of the National Institute of Standards and

Technology, not subject to copyright in the U.S.
 14 9.000

8.000
12
7.000

Normalized Stress
10 6.000

5.000 1 wt. % CNT

Stress, MPa

8 2 wt. % CNT
4.000 4 wt. % CNT
6 5 wt. % CNT
3.000

4 2.000
0 wt. % of CNT
1wt. % CNT in HYSOL 1.000
2 2 wt. % CNT in HYSOL
4 wt. % CNT in HYSOL 0.000
5 wt. % CNT in HYSOL
0 0 5 10 15 20 25
0 5 10 15 20 25 Strain, % Elongation
Strain, % Elongation

Figure 7: Tensile test plot of stress-strain relationship for the Figure 8: Normalized stress-strain curves for resin composites
resin and its composites
A similar trend is observed for Young’s modulus readings
The epoxy resin is very ductile since the maximum stress presented in figure 9.
levels reached are not those of a standard epoxy resin which is
generally about 30-60 MPa. It is observed that the stress level 4.500
is increased by the addition of CNTs in the matrix acting as 4.000

Normalized Young's Modulus

reinforcements. The mechanical properties of 0, 1, 2, 4 and 5 3.500
weight percent samples are presented in Table 1.
3.000

Table 1: Young’s Modulus and yield stresses at 10% strain 2.500

2.000

1.500
Young's Yield Strength at 10%
CNTs in matrix Modulus elongation 1.000

0.500
wt. % MPa s 10% , MPa
0.000
0 E0 = 109 3 0 1 2 4 5
wt. % of CNT in epoxy
1 E1 = 235 6

2 E2 = 348
4 E4 = 396
5 E5 = 432
Normalized stress-strain curves presented in figure 8 show
a better representation of the effect of CNTs on the mechanical
behavior of the composites. In this figure, the stress of the
composite matrix with different CNT concentration is divided
by the pure matrix stress at the same strain level. It can be
concluded that the effect of CNT reinforcements is much
reduced in the case of 4 weight percent and reduces further for
5 weight percent samples.
9
Figure 9: Normalized Young’s modulus of CNT–epoxy
10.6 composites at 10% elongation
11.9
A slope decrease can be observed for 4 and 5 weight
percent specimens. The SEM images in figure 1(b) show the
presence of porosity and CNT aggregates. There are also some
zones with very high CNT concentrations (clusters). At higher
CNT percentages (4 and 5 weight percents), the cluster size and
the CNT concentrations increase resulting in more
inhomogeneity. These differences partially explain the
premature fracture that occurred in 4 and 5 weight percent
composites at 10% elongation. On a similar note, the porosity
at higher weight percents (4 and 5 weight percents, see figure
1) may also be another origin of the lower reinforcement effect
of CNTs in the resin matrix.

CONCLUSION
This paper has demonstrated the potential of carbon
nanotubes as reinforcements for a commonly used epoxy in
electronic packaging applications. It is observed that a small
quantity of CNT addition can alter the properties of a base
polymer matrix especially its mechanical and electrical
behavior. The electrical measurements indicate interesting
trends. First, the electrical conductivity increases with
increases in the weight percent of nanotubes with relatively

5 Contribution of the National Institute of Standards and

Technology, not subject to copyright in the U.S.
smaller gains at higher CNT concentrations. Secondly, the
gains are larger at lower frequencies. It is observed that the
threshold of a conducting or interconnected network was
reached at approximately 2 weight percent. Due to the lack of a
continuous thermal path across the thickness of the specimen
and due to random orientation of CNTs in the matrix, the
thermal conductivity increase is not substantial. At 5 weight
percent of CNT, the thermal conductivity increased by only a
factor of two. The addition of 1 to 5 weight percent of CNT
into the epoxy matrix has a considerable effect on the
mechanical properties with increase in the weight fraction. For
instance, the Young’s modulus and the yield strength of the 1
weight percent composite increases by approximately 100 and
200% respectively compared to the pure matrix. The decrease
in the slope from 2 to 4 weight percent of CNTs of the
normalized Young’s modulus confirms that the homogeneity of
the composite is a critical point for the mechanical behavior of
the composites, whereas inhomogeneity may have less of an
effect on electrical properties.
The results of this study indicate reasonable gains in
thermal, electrical and mechanical properties as a function of
CNT concentration. With better methods of dispersion, these
gains are expected to be substantially higher.
ACKNOWLEDGEMENTS
The authors acknowledge the financial and technical
support offered by Dr. Suraj Rawal from Lockheed Martin. We
gratefully acknowledge the support provided by the Bowman
lab group at the Department of Chemical Engineering,
University of Colorado, Boulder, for their DSC equipment.
The authors would also like to thank the National Institute of
Standards and Technology for their invaluable support toward
providing instrumentation and equipment to perform the
various tests for successful completion of this paper.
REFERENCES
[1] Dresselhaus M.S., Dresselhaus G., and Eklund P.C.,
Science of Fullerenes and Carbon Nanotubes. London:
Academic Press (1996)
[2] Iijima S., Helical microtubles of graphite carbon.
Nature (London) (1991), 354:56-8.
[3] Frank S., Poncharal P., Wang Z.L., and De Heer W.A.,
Carbon Nanotube Quantum Resistors. Science 280, 1744
(1998)
[4] Sanvito S., Kwon Young-Kyun, Tomanek D., and
Lambert C.J., Fractional Quantum Conductance in Carbon
Nanotubes. Phys. Rev. Lett. 84(9), 1974 (2000)
[5] Berber S., Kwon Young-Kyun, and Tomanek D.,
Unusually High Thermal Conductivity of Carbon Nanotubes.
Phy. Rev. Lett. 84(20), 4613 (2000)
[6] Jeon Tae-In, Kim Keun-Ju, Kang C., Oh Seung-Jae,
Son Joo-Hiuk, An K.H., Bae D.J, and Lee Y.H., Terahertz
Conductivity of Anisotropic Single-walled Carbon Nanotube
Films. App. Phy. Lett. 80(18), 3403 (2002)
[7] Sandler J., Shaffer M.S.P., Prasse T., Bauhofer W.,
Schulte K., and Windle A.H., Development of a dispersion
process for Carbon Nanotubes in an Epoxy matrix and the
resulting Electrical Properties. Polymer 40, 5967 (1999)
[8] Grimes C.A., Mungle C., Kouzoudis D., Fang S., and
Eklund P.C., The 500 MHz to 5.50 GHz complex Permittivity
spectra of Single-wall Carbon Nanoube-loaded polymer
composites. Chem. Phy. Lett. 319, 460 (2000)
[9] Yu Min-Feng, Files B.S., Arepalli S., and Ruoff R.S.,
Tensile Loading of Ropes of Single-walled Carbon Nanotubes
and their Mechanical Properties. Phy. Rev. Lett. 84(24), 5552
(2000)
[10] Krishnan A., Dujardin E., Ebbesen T.W., Yianilos
P.N., and Treacy M.M.J., Young’s Modulus of Single-walled
Nanotubes. Phy. Rev. B 58(20), 14013 (1998)
[11] Kymakis E., Alexandou I., and Amaratunga G.A.J.,
Single-walled Carbon NAnotube-Polymer Composites:
Electrical, Optical and Structural Investigation. Synthetic
Metals 9108, 1-4 (2001)
[12] Che J., Cagin T, and Goddard III W.A., Thermal
Conductivity of Carbon Nanotubes. Nanotechnology 11, 65-69
(2000)
[13] Nardelli B.M., Fattebert J.L., Orlikowski D., Roland
C., Zhao Q., and Bernholc J., Mechanical Properties, Defects
and Electronic Behavior of Carbon Nanotubes. Carbon 38,
1703-1711 (2000)
[14] Sohn J.I., Choi Chel-Jong, Lee S., and Seong Tae-
Yeon, Growth Behavior of Carbon Nanotubes on Fe-deposited
(001) Si substrates. App. Phy. Lett. 78(20), 3130 (2001)
[15] Huang Z.P., Xu J.W., Ren Z.F., Wang J.H., Siegal
M.P., and Provencio P.N., Growth of highly oriented Carbon
Nanotubes by plasma-enhanced hot filament chemical vapor
deposition. App. Phy. Lett. 73(26), 3845 (1998)
[16] Prylutskyy Y.I., Durov S.S., Oglobyla O.V., Buzaneva
E.V., and Scharff P., Molecular dynamics simulation of
Mechanical, Vibrational and Electronic properties of Carbon
Nanotubes. Computational Materials Science 17, 352-355
(2000)
[17] Kim P., Shi L., Majumdar A., and McEuen P.L.,
Thermal Transport Measurements of Individual Multiwalled
Nanotubes. Phy. Rev. Lett. 87(21), 5021 (2001)
[18] Mahajan R.L., Malhotra C.P., and Sharma R.K., An
Analytical Cure Model for underfill epoxies, ASME J.
Electronic Packaging, 124, 391-396
6 Contribution of the National Institute of Standards and
Technology, not subject to copyright in the U.S.