A Combined Process of In Situ Functionalization and
Microwave Treatment to Achieve Ultrasmall Thermal
Expansion of Aligned Carbon Nanotube–Polymer
Nanocomposites: Toward Applications as Thermal
By Wei Lin, Kyoung-Sik Moon, and C. P. Wong*
In the past twenty years, substances with low or negative thermal a novel assembly process of the ACNT–polymer TIMs is needed.
expansivities have attracted much interest because of their In this study, we report a strong anisotropy in the CTE of
signiﬁcance in electronic packaging, precision equipment, and ACNT–epoxy composites, and achieve an ultrasmall CTE in the
intelligent materials.[1–3] In electronic packaging systems, CNT-aligned direction by enhancing the ACNT–epoxy interface
mismatch in the coefﬁcient of thermal expansion (CTE) between with a combined process of in situ CNT functionalization and
various materials has become a key issue for developing the next microwave treatment. Furthermore, an assembly process is
generation of electronic packaging with higher system reliability. proposed to adapt the ACNT–epoxy TIM structure into real-life
CTE values of polymer portions are much higher than those of applications, an issue never previously clariﬁed in the literature.
silicon, ceramic, and copper metallization. These large CTE Table 1 shows the CTE measurement results for a thermally
mismatches lead to thermal-stress accumulation at contact cured ACNT–epoxy composite (TCOM), a microwave-cured
interfaces during both packaging and device performance, which ACNT/epoxy composite (MCOM), a thermally cured epoxy
triggers component failure by, for example, warpage and (TEP), and a microwave-cured epoxy (MEP). Both the TEP and
rupture. So far, the effective approach to reduce the CTE of MEP samples display the typical CTE–temperature relation for
the polymer portion has been to add ﬁllers of low or negative amorphous polymers, with the CTE being much higher above
CTEs into polymer matrices.[2,4] The low or negative CTEs of the glass transition points (Tg) than below them. No apparent
carbon nanotube (CNTs), together with their low mass density difference in CTE was observed between TEP and MEP,
and outstanding mechanical and thermal properties, renders consistent with their similar Tg values and storage moduli (G0 )
them the right ﬁller for advanced polymer nanocomposites in shown in Table 2. This excludes unexpected inﬂuences of the
microelectronic applications. One of the most important microwave radiation on the pure epoxy itself. Both the TCOM and
applications for these materials is CNT–polymer nanocomposites the MCOM samples show strong anisotropy in thermal
for thermal interface materials (TIMs) with enhanced thermal expansion. Before their glass transitions, more than 63%
conductivity and, equally importantly, reduced CTEs. Thermal reduction in the through-thickness CTE (designated as aN) with
properties of polymer composites ﬁlled with randomly dispersed regards to the CTE of the epoxy is observed, while their in-plane
CNTs have been extensively studied in the past decade. CTE values (aP) are similar to that of the epoxy. Unexpectedly, the
Unfortunately, real-life applications of these materials were aN of the MCOM above its Tg is extremely small and close to the
inhibited by their low thermal conductivity and relatively high CTE of copper, which is even smaller than that below the Tg, that
CTEs, caused mainly by the weak CNT–polymer interface.[3,6] is, a 90% reduction of the CTE of the epoxy above its Tg. In
Alternatively, researchers turned to a simple inﬁltration process to comparison, the TCOM, as usual, shows a large CTE increase at
prepare polymer composites ﬁlled with aligned carbon nanotubes Tg as the turning point. One point that we would like to emphasize
(ACNT), because the CNT alignment ensures much higher here is the special importance of the CTE minimization of TIM
thermal conductivities than a random dispersion. However, a composites at temperatures above their Tg, which seems to have
CTE study of ACNT–polymer composites has never been been neglected or considered nonfeasible in the past. As is
reported. The ACNT/polymer interface is still an issue, as such known, it is common for polymers to have, upon heating, a higher
positive CTE at their Tg. Similar phenomena have also been
observed in polymer-based composites, including CNT–polymer
[*] Prof. C. P. Wong, W. Lin, Dr. K. S. Moon
School of Material Science & Engineering
composites. For example, Wang et al. and Xu et al. observed that
and Packaging Research Center the CTE values of CNT/polymer composites at temperatures
Georgia Institute of Technology above their Tg were at least 70 times that for single-crystal silicon,
771 Ferst Drive NW and 13 times that for pure copper. Although semiconductor
Atlanta, GA 30332 (USA) devices are expected to operate below 150 8C, the thermal
nonuniformity usually referred to as hot spots, where the power
DOI: 10.1002/adma.200803548 density could be >300 W cm–2, is a factor that eventually
Adv. Mater. 2009, 21, 2421–2424 ß 2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 2421
Table 1. CTE comparisons at temperatures below and above the Tg. All
values were extracted from the slopes in quasi-equilibrium cooling curves.
aN and aP indicate the through-thickness and in-plane CTEs, respectively.
Samples aN [ppm KÀ1] aP [ppm KÀ1]
Below Tg Above Tg Below Tg Above Tg
TEP 81 191 80 186
MEP 80 188 82 191
TCOM 30 68 81 310
MCOM 23 18 80 240
determines the device reliability. The hot-spot issue will cause
the local temperature to be higher than the Tg of the TIM
composite; in this case, the intensively heated part expands much Figure 1. DMA results of TCOM and MCOM. Only cooling curves are
more than the nearby components. How can the CTE of polymer included.
composites above the Tg be reduced to, for example, a value close
to the CTE of copper ($16 ppm KÀ1) or silicon ($3 ppm KÀ1)? In avoiding degradation of intrinsic CNT properties are important
other words, how can we realize the principle of ‘1 þ 1 >2’? challenges in this ﬁeld. Therefore, a novel approach is necessary
To the best of our knowledge, this is the ﬁrst report so far on an to enhance the ACNT–polymer interface. Appropriately, the
ultralow CTE above the Tg for CNT/polymer composites. It seems combined processes of ACNT in situ functionalization and
that microwave treatment plays a key role in obtaining such an microwave curing helps to improve the ACNT–epoxy interfaces
ultralow aN of the MCOM above the Tg. During curing, while maintaining their well-aligned structures. The distinct
microwaves selectively heat up the CNTs and the polymer at differences in Tg and G0 between the TCOM and the MCOM, as
the interfaces. The fast coupling between the ACNTs, shown in Figure 1 and Table 2, give us a hint on the enhanced
functionalized sites on the ACNT surface, and the reactive CNT–epoxy interface status in the MCOM. Notably, a 6 8C
functional groups in the epoxy matrix with the oscillating increase in Tg and a great enhancement in G0 with the
electromagnetic ﬁeld is equally important. In fact, microwave variable-frequency microwave (VFM) treatment compared with
irradiation has attracted much interest in synthetic organic the pure epoxy and the thermally cured composite samples is
chemistry because of its special role in dramatically increasing observed. Consistent with the DMA results, MCOM displays a
reaction rates, and its capability of inducing chemical reactions much higher thermal conductivity than TCOM (Fig. 2), which
that cannot proceed by thermal heating alone.[11,12] It is further veriﬁes our postulation.
postulated that during the curing process in the microwave Why does the better interface lead to the ultralow aN above the
ﬁeld, the interfacial bonding between the ACNTs and the epoxy Tg? The ﬁrst possibility is the frozen, or at least partially frozen,
matrix was dramatically improved. orientation of polymer segments along the thickness direction—
As is known, a weak CNT–polymer interface has been a key the CNT-alignment direction. A bonded interface is the right
challange for CNT/polymer composites since CNTs were force to freeze the polymer segments close to the interface, even
discovered. A weak interface results in inefﬁcient load transfer, above the Tg. At temperatures below the Tg, a sideward expansion
and consequently only moderate improvement, in mechanical of the polymer molecules contributes more to the aP, while a
properties. Phonon coupling across the weak interface is also lengthwise expansion of the molecules dominates the aN. If it is
limited. This has been considered the main reason for the high
interfacial thermal resistance, and eventually, why the thermal
conductivities of CNT/polymer composites are much lower than
expected.[6e,6f,13] Conventionally, people use a wet chemical
method to functionalize the CNT surface to increase interfacial
bonding between the CNTs and the polymer matrix. However,
the CNT alignment is destroyed and the surface of the CNT is
damaged. Minimizing unnecessary structural defects and
Table 2. DMA result comparisons.
Samples Tg [a] [8C] Tg [b] [8C] G0 (100 8C) [MPa] G0 (50 8C) [MPa]
TEP 120.72 129.9 1336 1464
MEP 120.62 128.9 1339 1489
TCOM 120.86 131.15 1732 2198
MCOM 126.9 136.19 2332 2662
[a] Extracted from storage modulus curves in the quasi-equilibrium cooling process.
[b] Extracted from tan d peaks in the quasi-equilibrium cooling process. Figure 2. Thermal conductivities of TCOM, MCOM, and epoxy.
2422 ß 2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Adv. Mater. 2009, 21, 2421–2424
the case, the aN of the MCOM should be smaller than that of the
TCOM, while with a larger aP. Although it looks like the aN of the
MCOM is slightly smaller than that of the TCOM, their aP are
close to each other. It is likely that radial contraction of the ACNTs
in the MCOM obliterates this difference based on the better
interface. At temperatures above the Tg, the overall volume
expansion of the TCOM should roughly equal that of the MCOM.
Even if the polymer chains do not lose their orientation, not a big
difference is expected between the aN of the MCOM and that of
the TCOM. However, at temperatures above the Tg, the aN of the
MCOM is only a third of the TCOM. This indicates the existence
of a second mechanism that contributes to such an ultralow aN, as
Molecular dynamics simulations suggest longitudinal con-
traction of a single-walled carbon nanotube (SWNT).[5c] Low-
dimensional systems gain structural and vibrational entropy by
exploring the voids in conﬁgurational space at relatively small
energy cost, resulting in thermal contraction in the harmonic
regime at moderate temperatures. In Ref. [5c], a longitudinal
thermal contraction of a (10, 10) single-walled carbon nanotube
was found to be shared by bending, twist, and pinch modes of
the tube, whereas the volumetric contraction was dominated by
the pinch mode. A radial contraction was also expected. A
negative radial CTE of À1.5 ppm KÀ1 in SWNTs was experi- Figure 4. Illustration of the ACNT–epoxy TIM assembly process and the
mentally estimated.[5b] Although it is not yet clear whether CNT status at the interface.
multiwalled carbon nanotubes (the CNTs used in this study are
multiwalled) contract longitudinally or not, it is reasonable to
expect that they will have a much-lower CTE than epoxy. Upon
heating, there is an in-plane stretching coupled with a normal shows a much larger aP than its aN, where aP is even larger than
compression imposed on the polymer network because of the the CTE of the epoxy. The even larger aP of the TCOM is probably
ACNT anchoring effects at interfaces, where the in-plane a result of an increase in excluded volume,[3a] where the weak
stretching comes from the radial contraction of the ACNTs, CNT–polymer interfaces cannot exert the additional inﬂuences as
while the normal compression is a result of a large CTE mismatch in the MCOM.
between the ACNTs and the polymer matrix. Given that Poisson’s One may worry that the large aP will cause a big in-plane CTE
ratio of a rubbery polymer network (a crosslinked network above mismatch between the TIM and the mating surfaces. Actually, it
its Tg) is close to 0.5, which indicates nearly no volume change, is not the case for the pressure-involved TIM assembly process.
the in-plane stretching and the normal compression have Figure 4 shows the development of the CNT–epoxy TIM assembly
overlapping effects that reduce the aN. Figure 3 illustrates such process. During the TIM assembly, ACNTs lie down on the
a mechanism. Below the Tg, since segmental movements of mating surface with dramatically increased CNT coverage. For
polymer molecules are frozen, such stretching/compression example, the thickness of a ACNT 100 mm thick reduces to
forces cannot exert much inﬂuence on the epoxy network. Above $15 mm upon a 0.4 MPa compression force. The SEM images in
the Tg, they do. Although this overall inﬂuence may not be as Ref.  show clearly the cranked CNT-array structure after
strong as the interfacial anchoring effect that acts on a thin compression. In the Supporting Information, we show the top
polymer ﬁlm by a speciﬁc substrate, where even an ultranegative view of a slightly compressed ACNTarray at a mating surface after
CTE was observed, it is possible to reduce aN to 1/10 of the the mating substrate is removed. Furthermore, the mating
CTE or lower for pure epoxy. This also explains why the MCOM surface is rough, the surface roughness being simpliﬁed as
waviness in Figure 4. Therefore, on the nanometer scale, the
ACNTs within the TIM layer are no longer ‘vertically’ oriented, but
more likely adopt the orientation shown in Figure 4, where a CTE
gradient is expected. This provides an insight into the real status
of the ACNT–epoxy TIMs. The epoxy inﬁltration that accom-
panies the pressure will improve the interfacial adhesion and ﬁx
the CNT orientation so that 1) tangential stress (sT in Fig. 4) at the
mating interface is relatively small, because the mismatch
between aN and the CTE of the mating substrate is effectively
reduced; 2) normal stress (sN) is beneﬁcial in holding the
pressure to ensure close contact between the CNTs and the
Figure 3. Illustration of the mechanism for the ultrasmall through- mating surface, so that thermal conductance degradation during
thickness CTE of MCOM above its Tg. device performance can be minimized.
Adv. Mater. 2009, 21, 2421–2424 ß 2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 2423
In summary, the results in this study highlight the enhance-
 a) T. A. Mary, J. S. O. Evans, T. Vogt, A. W. Sleight, Science 1996, 272, 90.
ment of the CNT–polymer interface by the combined process of
b) R. Roy, D. K. Agrawal, H. A. Mckinstry, Annu. Rev. Mater. Sci. 1989, 19, 59.
in situ functionalization and microwave treatment, which confer c) S. Numata, N. Kinjo, D. Makino, Polym. Eng. Sci. 1988, 28, 906.
beneﬁts in CTE reduction, load transfer, and phonon transport  a) J. S. O. Evans, J. Chem. Soc. Dalton Trans. 1999, 19, 3317. b) K. Yano, A.
across the interface in ACNT–polymer TIM composites. Usuki, A. Okada, T. Kurauchi, O. Kamigaito, J. Polym. Sci. Part A: Polym.
Chem. 1993, 31, 2493.
 a) S. Wang, Z. Liang, P. Gonnet, Y. Liao, B. Wang, C. Zhang, Adv. Funct.
Mater. 2007, 17, 87. b) Y. Xu, G. Ray, B. Abdel-Magid, Compos. A 2006, 37,
 Z. Q. Zhang, C. P. Wong, IEEE Trans. Adv. Packag. 2004, 27, 515.
Vertically aligned CNT arrays 2–3 mm thick were grown on a SiO2/Si  a) Y. Yosida, J. Appl. Phys. 2000, 87, 3338. b) Y. Maniwa, R. Fujiwara, H.
substrate with 10 nm thick Al2O3 and 2 nm thick Fe ﬁlms through a thermal
Kira, H. Tou, H. Kataura, S. Suzuki, Y. Achiba, Phys. Rev. B 2001, 64,
chemical vapor deposition (CVD). The CVD growth was carried out at 750 8C,
241402. c) Y. Kwon, S. Berber, D. Tomanek, Phys. Rev. Lett. 2004, 92, 015
with a gas-ﬂow rate ratio of Ar/H2/C2H4 ¼ 380/150/150 standard cubic cm
901. d) P. M. Ajayan, Chem. Rev. 1999, 99, 1787. e) J. N. Coleman, U. Khan,
minÀ1 (sccm) . ACNTs were functionalized in situ during the CVD growth
Y. K. Gun’ko, Adv. Mater. 2006, 18, 689. f) R. H. Baughman, A. A. Zakhidov,
process, as reported in a recent communication . The ACNT arrays were
ﬂipped onto polyimide double-sided tape and then inﬁltrated with epoxy to W. A. d de Heer, Science 2002, 297, 787.
prepare ACNT–epoxy composites, followed by degassing under vacuum for  a) C. Wei, D. Srivastava, K. J. Cho, Nano Lett. 2002, 2, 647. b) E. T.
40 min. Samples were thermally cured in a common convection oven at 155 8C Thostenson, Z. F. Ren, T. W. Chou, Compos. Sci. Technol. 2001, 61, 1899.
for 40 min, or in a VFM chamber (central frequency: 6.4250 GHz) set at the c) E. S. Choi, J. S. Brooks, D. L. Eaton, M. S. Al-Haik, M. Y. Hussaini, H.
same temperature. After furnace cooling, the ACNT–epoxy samples were Garmestani, D. Li, K. Dahmen, J. Appl. Phys. 2003, 94, 6034. d) C. H. Liu, H.
peeled off the double-sided tape and polished to be double-sided, parallel, Huang, Y. Wu, S. S. Fan, Appl. Phys. Lett. 2004, 84, 4248. e) M. B. Bryning,
ﬂat, and smooth, and were cut into pieces for speciﬁc measurements. The D. E. Milkie, M. F. Islam, J. M. Kikkawa, A. G. Yodh, Appl. Phys. Lett. 2005,
samples maintain vertical alignment of the CNTs in the epoxy matrix, as 87, 161909. f) S. Shenogin, L. P. Xue, R. Ozisik, P. Keblinski, D. G. Cahill,
shown in the Supporting Information, consistent with the SEM image shown J. Appl. Phys. 2004, 95, 8136.
in the literature . The epoxy used was bisphenol-F (EPON862),  a) H. Huang, C. H. Liu, Y. Wu, S. S. Fan, Adv. Mater. 2005, 17, 1652. b) S.
with 4-methylhexahydrophthalic anhydride and 1-cyanoethyl-2-ethyl-4- Sihn, S. Ganguli, A. K. Roy, L. Qu, L. Dai, Compos. Sci. Technol. 2008, 68,
methylimidazole as the curing agent and catalyst, respectively. 658. c) T. Borca-Tasciuc, M. Mazumder, Y. Son, S. K. Pal, L. S. Schadler,
CTE was measured using a thermal mechanical analyzer (TMA, TA P. M. Ajayan, J. Nanosci. Nanotechnol. 2007, 7, 1581.
Instruments Model 2940), at a heating rate of 5 8C minÀ1. When measuring  R. F. Boyer, R. S. Spencer, J. Appl. Phys. 1945, 16, 594.
the through-thickness CTE (along the CNT-aligned direction), the  a) R. S. Prasher, J. Chang, I. Sauciuc, S. Narasimhan, D. Chau, G. Chrysler,
bulk-mode ﬁxture and specimens $1 mm thick were used. The quartz
A. Myers, S. Prstic, C. Hu, Intel. Technol. J. 2005, 9, 285. b) R. Prasher, Proc.
probe was seated normal to the specimen top surface. Two different modes
IEEE. 2006, 94, 1571. c) R. Mahajan, C. P. Chiu, G. Chrysler, Proc. IEEE.
were used to measure the in-plane CTE. The ﬁlm-mode ﬁxture was used for
2006, 94, 1476.
thin specimens. The bulk-mode ﬁxture was used for thick specimens,
 a) T. J. Imholt, C. A. Dyke, B. Hasslacher, J. M. Perez, D. W. Price, J. A.
where the quartz probe was seated normal to the side of the specimen, that
is, normal to the CNT orientation. Storage modulus and tan d were Roberts, J. B. Scott, A. Wadhawan, Z. Ye, J. M. Tour, Chem. Mater. 2003, 15,
measured using a dynamical mechanical analyzer (DMA, TA Instruments 3969. b) C. Y. Wang, T. H. Chen, S. C. Chang, S. Y. Cheng, T. S. Chin, Adv.
Model 2940) at a heating rate of 5 8C minÀ1. The single-cantilever mode Funct. Mater. 2007, 17, 1979.
was used, with the CNTs in the specimen vertically oriented, that is, along  D. D. Young, J. Nichols, R. M. Kelly, A. Deiters, J. Am. Chem. Soc. 2008, 130,
the vibration direction. A constant frequency of 10 Hz and an amplitude of 10048.
10 mm were adopted. At least ﬁve measurements were recorded to obtain  C. O. Kappe, Angew. Chem. Int. Ed. 2004, 43, 6250.
the average values in Tables 1 and 2. Effective thermal conductivities were  a) C. W. Nan, G. Liu, Y. H. Lin, M. Li, Appl. Phys. Lett. 2004, 85, 3549. b) S. T.
obtained by measuring thermal diffusivities with a Netzsch laser ﬂash Huxtable, D. G. Cahill, S. Shenogin, L. P. Xue, R. Ozisik, P. Barone, M.
apparatus (LFA447); mass densities and speciﬁc heat were measured using Usrey, M. S. Strano, G. Siddons, M. Shim, P. Keblinski, Nat. Mater. 2003, 2,
a differential scanning calorimeter (DSC, TA Instruments Model 2940). 731. c) S. Shenogin, A. Bodapati, L. Xue, R. Ozisik, P. Keblinski, Appl. Phys.
Lett. 2004, 85, 2229.
 D. Tasis, N. Tagmatarchis, A. Bianco, M. Prato, Chem. Rev. 2006, 106,
Acknowledgements  a) M. Mukherjee, M. Bhattacharya, M. K. Sanyal, Phys. Rev. E 2002, 66,
061801. b) T. Kanaya, T. Miyazaki, R. Inoue, K. Nishida, Phys. Status Solidi B
The authors acknowledge the NSF (#0621115 and #0422553) for ﬁnancial
2005, 242, 595.
support of this work, and Prof. K. Jacob and Dr. X. He at Georgia Institute of
Technology for helpful discussions. Supporting Information is available  L. Qu, L. Dai, Adv. Mater. 2007, 19, 3844.
online from Wiley InterScience or from the author.  L. Zhu, D. Hess, C. P. Wong, Nano Lett. 2005, 5, 2641.
 W. Lin, Y. Xiu, H. Jiang, R. Zhang, O. Hildreth, K. Moon, C. P. Wong, J. Am.
Received: December 1, 2008 Chem. Soc. 2008, 130, 9636.
Revised: January 8, 2009  B. L. Wardle, D. S. Saito, E. J. Garcia, A. J. Hart, R. G. de Villoria, E. A.
Published online: March 19, 2009 Verploegen, Adv. Mater. 2008, 20, 2707.
2424 ß 2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Adv. Mater. 2009, 21, 2421–2424