Carbon nanotube thermal interface material for high-brightness light-emitting-diode cooling
This content has been downloaded from IOPscience. Please scroll down to see the full text.
IP Address: 184.108.40.206
This content was downloaded on 05/10/2013 at 08:09
Please note that terms and conditions apply.
2008 Nanotechnology 19 215706
View the table of contents for this issue, or go to the journal homepage for more
Home SearchCollections JournalsAboutContact usMy IOPscience
Nanotechnology 19 (2008) 215706 (8pp)
Carbon nanotube thermal interface
material for high-brightness
K Zhang1, Y Chai2, M M F Yuen1, D G W Xiao3and P C H Chan2
1Department of Mechanical Engineering, Hong Kong University of Science and Technology,
Clear Water Bay, Kowloon, Hong Kong SAR, People’s Republic of China
2Department of Electronic and Computer Engineering, Hong Kong University of Science and
Technology, Clear Water Bay, Kowloon, Hong Kong SAR, People’s Republic of China
3Advanced Packaging Technology Limited, Hong Kong University of Science and
Technology, Clear Water Bay, Kowloon, Hong Kong SAR, People’s Republic of China
E-mail: firstname.lastname@example.org (K Zhang)
Received 13 January 2008, in final form 20 March 2008
Published 23 April 2008
Online at stacks.iop.org/Nano/19/215706
Aligned carbon nanotube (CNT) arrays were fabricated from a multilayer catalyst configuration
by microwave plasma-enhanced chemical vapor deposition (PECVD). The effects of the
thickness and annealing of the aluminum layer on the CNT synthesis and thermal performance
were investigated. An experimental study of thermal resistance across the CNT array interface
using the modified ASTM D5470 standard was conducted. It was demonstrated that the
CNT-thermal interface material (CNT-TIM) reduced the thermal interfacial resistance
significantly compared with the state-of-art commercial TIM. The optimized thermal resistance
of the CNT arrays is as low as 7 mm2K W−1. The light performance of high-brightness
light-emitting diode (HB-LED) packages using the aligned CNT-TIM was tested. The results
indicated that the light output power was greatly improved with the use of the CNT-TIM. The
usage of the CNT-TIM can be also extended to other microelectronics thermal management
(Some figures in this article are in colour only in the electronic version)
Challenges in the thermal management of an electronic
package arise from the continued increase in power dissipation
and power density of higher-power devices .
example is solid state lighting using a high-brightness light-
emitting diode (HB-LED) which is an attractive candidate
for future general illumination applications.
efficiency, reliability and color of the solid state lighting
devices strongly depend on successful thermal management
of the package.With the increase of input power and
the applications of high-density HB-LED packaging, the
requirements of improving the performance and reliability
impose a significant challenge on the thermal design of
HB-LED packages and the application of novel materials.
The luminance of HB-LEDs will reduce linearly while the
life will reduce exponentially with the increasing junction
temperature [2, 3].Therefore, proper thermal design is
imperative in keeping the LED package below its rated
operating temperature. With the multilayer structure of HB-
LED packages, the thermal resistance across material interface
remains the bottleneck in heat transfer from the LED junction
to the cooling system .
conventional TIMs used in the interface is relatively low
because the TIMs normally make up of polymer matrix with
thermally conductive fillers.
The carbon nanotube (CNT) is a promising candidate to
resistance as well as the ultra-high thermal conductivity, about
600–3000 W m−1K−1for an individual multi-walled carbon
nanotube (MWNT) [5, 6] and 15–250 W m−1K−1for bulk
vertically aligned CNT (VACNT) films [7–9].
The thermal conductivity of
© 2008 IOP Publishing LtdPrinted in the UK
Nanotechnology 19 (2008) 215706K Zhang et al
When pressure reach 0.1-0.2 Torr,
open the valve of H2(40sccm)and N2(10sccm),
Turn on plasma
Turn off plasma and heater,
close the valve of all gases
Open the valve of CH4(10sccm)
Figure 1. Schematic process procedure of CNT synthesis.
CNT composites with CNTs randomly dispersed in
matrix material possess thermal conductivity no more than
10 W m−1K−1[10–12]. Fan et al filled polymer in aligned
CNT arrays, and the thermal property of the CNT array
composite was found to be better than that of random CNT
composite .They further used an aluminum thin film
to reduce the thermal interfacial resistance of the CNT array
composite . However, the thermal interfacial resistance
was still higher than 50 mm2K W−1. It should be further
improved to reach the thermal performance target of TIM
of 5 mm2K W−1. Xu et al  and Cola et al 
prepared some dry CNT arrays.
was measured using different methods and the results were
20 mm2K W−1and 16 mm2K W−1, respectively. Tong et al
 also measured the thermal conductance of the CNT array
as9×104W m−2K−1. Alltheseexperimentalresultsindicated
that a vertically aligned CNT array without polymer matrix is
a good candidate of TIM.
In this paper, a process of CNT-TIM synthesis by
microwave PECVD is presented.
aluminum thin film was optimized, and the effect of annealing
the aluminum thin film to minimize the thermal interfacial
resistance of the CNT-TIM was studied. The CNT-TIM was
used in HB-LED packaging and the optical performance was
evaluated. The result showed that CNT-TIM can maintain the
output light power of LED devices at high input current. It
The thermal resistance
The thickness of the
indicated the effectiveness of the CNT-TIM in facilitating the
heat dissipation of HB-LED packages.
2. Experimental testing of CNT-TIM
2.1. CNT-TIM fabrication
CNTs were synthesized by the microwave PECVD process.
Microwave power of 300 W was used. Zhang et al has shown
that better CNT–substrate bonding and better CNT synthesis
quality resulted in improved thermal performance .
20 nm titanium buffer layer was sputtered on the silicon
substrate to improve its adhesion with the catalyst layer and
theeventualalignedCNT-TIM. Analuminumlayer was further
sputtered on top of the titanium layer to help activate the
catalyst particles.The thickness of aluminum layer was
optimized, and the effect of aluminum layer annealing on
CNT synthesis and thermal performance was studied.
10 nm nickel catalyst layer was further deposited by e-
beam evaporation on the top of the aluminum layer. In the
CNT fabrication process, the substrate with catalyst was first
introduced to the chamber and subjected to 40 sccm H2and
10 sccm N2 plasma treatment at 850◦C for 1 min. CNTs
started to grow with further introduction of 10 sccm CH4at the
same temperature for 5 min. The plasma and the heater were
turned off after the CNT growth. The whole system was then
cooled down naturally to room temperature. Figure 1 shows
the processing procedure of CNT synthesis.
2.2. Thermal resistance measurement
The thermal performance of the TIM is affected by the
bulk material and the interfacial conditions of the HB-LED
packages. Therefore, the thermal performance of CNT-
TIM in this study was evaluated by the thermal resistance
measurement system designed in accordance to ASTM D5470.
The testing procedure was modified by adopting a lower
contact pressure 0.15 MPa as compared with the 3 MPa
stipulated in the ASTM D5470 standard. This modification
is in line with the practice adopted in electronic packaging
processes . The testing setup isshown in figure 2(a).
Figure 2. (a) Schematic system of thermal resistance measurement; (b) sample package; (c) SEM image of grown CNT-TIM; (d) TEM of
Nanotechnology 19 (2008) 215706 K Zhang et al
Figure 3. Schematic side view of an HB-LED package with
The CNT-TIM sample package, shown in figure 2(b), has the
aligned CNT array grown on a 1 inch by 1 inch Si substrate
coupled with an aluminum alloy plate by mechanical pressure.
A load cell was used to control the mechanical contact pressure
at 0.15 MPa. Using this measurement technique, both the
thermal resistance of bulk CNT-TIM (RTIM) and the thermal
interfacial resistances (RSi–TIMand RAl–TIM) can be evaluated.
The test sample configuration was designed to assimilate that
of the real packages.
In order to compare the thermal performance of CNT-
TIM with other state-of-art TIMs, three other kinds of TIM
sampleswere prepared. Thefirstis‘air’, representingthedirect
attachment of a silicon substrate to an aluminum alloy surface
with air entrapped in between. The second is ‘commercial
silver epoxy’ having a layer of 25 µm thick commercial silver
epoxy pasted and cured between a silicon substrate and an
aluminum alloy surface. The third is a ‘metal system’ formed
by a reflow process consisting of sputtered 80 nm titanium,
300nm copper and10 µm electroplated eutectic lead/tinsolder
on both the silicon substrate and the aluminum alloy surface.
2.3. HB-LED testing with CNT-TIM
The thermal management capability of the CNT-TIM was
tested in HB-LED packages and the optical performance
of the HB-LED packages was tested against packages with
commercial TIM. A blue color 1 mm by 1 mm LED
chip was assembled on a silicon-interposer with flip-chip
technology.Figure 3 shows the cross-section of the LED
package with CNT-TIM. The flip-chip bonded LED device
was first assembled to a heat slug.
attached to a 1 inch by 1 inch heatsink using a different
Then the module was
TIM. A commonly used commercial silver epoxy TIM was
used as the benchmark TIM. The LED device was adhered
to the CNT-TIM by dispensing some room temperature fast
cured thermally conductive polymer around the CNT-TIM and
letting it cure. The detailed process flow of patterning a CNT-
TIM and assembling it to an HB-LED package is described
The output light power of different samples was measured
using a PMS-50 spectrometer from Everfine Company
Limited. The input electrical current was varied from 150
to 900 mA. Measurements were taken after the samples were
turned on for 6 min to ensure that the sample reached a thermal
3. Results and discussion
3.1. Thermal performance of CNT-TIM
The total measured thermal resistance of each TIM sample,
RTIM, included the thermal resistance of bulk TIM (RTIM)
and two thermal interfacial resistances (RSi–TIMand RAl–TIM).
It was obtained in three steps, as shown in figure 4. First,
the whole sample package was put in the thermal resistance
measurement system.The measured thermal resistance,
Rpackage, was the sum of a whole series of thermal resistances,
including RCu–Si, RSi, RSi–TIM, RTIM, RTIM–Al, RAl, and
RAl–Cu. Second, a Si chip was put in the thermal resistance
measurement system. This measured thermal resistance, RSi,
covered RCu–Si, RSi, and RSi–Cu. Among them, RCu–Siwas
assumed equal to RSi–Cuas the surface roughness conditionsof
the two Cu block surfaces and the two Si surfaces are the same.
Similarly, the measured thermal resistance RAlcan be obtained
inclusive of RCu–Al, RAl, and RAl–Cu. With the assumption that
RCu–Alequals RAl–Cu, the total thermal resistance of the TIM,
RTIM, can be calculated as follows:
RTIM= RSi–TIM+ RTIM+ RTIM–Al
where the thermal resistance of bulk Si (RSi) and bulk
aluminum alloy (RAl) can be calculated directly from their
Figure 4. Three steps to obtain the total thermal resistance of TIM.
Nanotechnology 19 (2008) 215706 K Zhang et al
Thermal resistance (mm2 K/W)
Thermal performance of different TIM
Commercial silver epoxy
CNT-TIM grown by PECVD
Figure 5. Measured total thermal resistance RTIMof different TIMs.
thickness and thermal conductivity:
=0.386 × 10−3(m)
141(W m−1K−1)= 2.74 × 10−6(m2K W−1)
4 × 10−3(m)
180(W m−1K−1)= 22.2 × 10−6(m2K W−1).
In order to minimize the error of measurement, several
samples were prepared and tested to obtain the average total
thermal resistance, RTIM, and standard deviation for each of
the four kinds of TIM, as shown in figure 5. As for CNT-
TIM, five samples were prepared and measured. The thermal
resistance, RTIM, ranged from 2 to 13 mm2K W−1. The data
deviation resulted from the error of temperature measurement
(±0.3◦C by using class A platinum resistance temperature
detectors) and copper bar fabrication (5 µm). In addition, the
samples were not exactly the same even though we used the
same preparation method for each one. The resulting average
thermal resistance of CNT-TIM is 7 mm2K W−1with a
standard deviation of 5 mm2K W−1. This is about 10% of that
of the commercial silver epoxy TIM samples. It is also much
lower thanthat of themetal system samples. Thisindicatesthat
the aligned CNT array is a promising base material for TIM.
The total thermal resistance of TIM, RTIM, depends not
only on the thermal property of TIM itself, but also on the
thermal and physical properties of the contacting members,
the contact geometry, and the contact pressure . The low
thermal resistance of CNT-TIM in this study was achieved by
havingbetter bondingbetween theCNT-TIM andthe substrate.
In the process of heating the substrate to the CNT synthesis
temperature, 850◦C, the supporting materials and the catalyst
melt to form alloys or compounds because the melting point
of nanoscale particles is typically much lower than that of the
bulk material . Therefore, the catalyst adhered better to
the substrate and thus provided better adhesion at the roots
of the CNT array. Furthermore, the flexible nanoscale tips
of CNT arrays can possibly penetrate into the troughs on
the rough interfacial surface, thus building more phonon flow
paths. In addition, the high CNT density made the CNT array
an effective heat conduction pathway and a strong structural
support system. These facilitated phonon heat transfer from
the silicon chip to the aluminum alloy plate via the CNT array
alongitsaxialdirection. Asaresult, thetotalthermalresistance
of CNT-TIM was much lowered.
3.2. Effects of annealing of substrate and catalyst
We investigated the effect of annealing of substrate in O2
on CNT synthesis.Substrates with supporting layers of
20 nm titanium and 10 nm aluminum were annealed in O2
at 550◦C for 30 min before the 10 nm nickel catalyst layer
was deposited.Figures 6(a) and (b) are AFM images of
substrates (Si/Ti/Al) with and without annealing of substrate
in O2, respectively.It was shown that after annealing the
surface of the substrates changed from a thin film structure
(with average roughness of 1.524 nm) to a relatively rough
configuration (with average roughness of 5.168 nm). High-
density uniform sized small protrusions were observed on the
substrate. The interaction surface area of the substrate was
We also investigated the effect of additional annealing
of catalyst-coated substrates in H2microwave plasma. After
the 10 nm nickel catalyst layer was evaporated on the top,
the substrates were put in the PECVD chamber and heated
to 850◦C. The catalyst-coated substrates were then annealed
for 1 min in H2 microwave plasma at this temperature
and 720 Pa.Figures 6(a) and (b) are AFM images of
catalyst-coated substrates (Si/Ti/Al/Ni) without annealing in
H2microwave plasma. Figures 6(c) and (d) are AFM images
of catalyst-coated substrates(Si/Ti/Al/Ni) withannealing inH2
microwave plasma. It is shown that the average roughness
of catalyst-coated substrates after annealing in H2microwave
plasma is higher than those without annealing.
Figures 6(e) and (f) show the results of CNT growth on
substrates with and without substrate (Si/Ti/Al) annealing in
O2, respectively. It is obvious that there is no CNT growth
on the substrate without annealing in O2, while there are
vertically aligned CNTs grown on the annealed substrate. It is
more conducive for CNT growth on relatively rough surfaces
because the neutral gases are decomposed into ions or radicals
in the plasma region with the intense high-energy electron
impact, and the ions are more reactive with the larger catalysts
where electric fields are preferentially concentrated .
Therefore, both annealing of substrates and annealing of
catalyst-coated substrates are good for aligned CNT growth.
The thermal resistances of CNT-TIM grown on substrates
with and without annealing were measured to determine the
effect of annealing on the thermal performance of CNT-TIM.
It is shown clearly in table 1 that the thermal resistance of
CNT-TIM on substrates with annealing in O2is much lower
than that of CNT-TIM on substrates without annealing in O2.
In conclusion, substrate annealing is helpful not only to CNT
synthesis but also to thermal performance improvement of
3.3. Effects of aluminum buffer layer thickness
It was found from experiments that the growth of CNTs is
sensitive to the thickness of the aluminum buffer layer. The
effect of aluminum thickness on CNT growth under the same
synthesis conditions was investigated, and the result is shown