Catalytic Chemical Vapor Deposition of
Single-Wall Carbon Nanotubes at Low
Temperatures
Mirco Cantoro,
†
Stephan Hofmann,*
,†
Simone Pisana,
†
Vittorio Scardaci,
†
Atlus Parvez,
†
Caterina Ducati,
‡
Andrea C. Ferrari,
†
Arthur M. Blackburn,
§
Kai-You Wang,
§
and John Robertson
†
Electrical Engineering DiVision, UniVersity of Cambridge, Cambridge CB3 0FA, U.K.,
Department of Materials Science and Metallurgy, UniVersity of Cambridge,
Cambridge CB2 3QZ, U.K., and Hitachi Cambridge Laboratory,
Cambridge CB3 0HE, U.K.
Received January 11, 2006; Revised Manuscript Received March 24, 2006
ABSTRACT
We report surface-bound growth of single-wall carbon nanotubes (SWNTs) at temperatures as low as 350
°
C by catalytic chemical vapor
deposition from undiluted C
2
H
2
.NH
3
or H
2
exposure critically facilitates the nanostructuring and activation of sub-nanometer Fe and Al/Fe/Al
multilayer catalyst films prior to growth, enabling the SWNT nucleation at lower temperatures. We suggest that carbon nanotube growth is
governed by the catalyst surface without the necessity of catalyst liquefaction.
Carbon nanotubes (CNTs) have been a driving force for
current advances in nanotechnology, both on an applied and
on a fundamental level. Single-wall carbon nanotubes have
shown the highest Young’s modulus and highest axial
thermal conductivity of any solid. Moreover, SWNTs have
the highest current carrying capacity of any conductor, which
makes them an attractive electronic, sensing, or heat sinking
material for nano-electromechanical systems (NEMS) and
future (hybrid) integrated circuitry.
1,2
Defect-free SWNT synthesis is generally thought to require
high temperatures (T).
3
This belief arises from the success
of high-T deposition processes. Arc-discharge, laser ablation,
and high-pressure CO conversion with inherent (local)
temperatures in the order of 1000-4000 °C have been
optimized mainly for bulk CNT production.
3
For device
fabrication, these techniques heavily rely on purification from
other carbon allotropes and an indirect postgrowth assembly
via stable suspensions. In comparison, catalytic chemical
vapor deposition (CVD) allows selective, aligned CNT
growth directly onto a substrate and thus presently is the
only economically viable process for integrating CNTs into
a device.
1,4-6
This approach, however, exposes the substrate
to the CNT growth temperature and atmosphere, which
creates a need for less aggressive, low T processing condi-
tions. Present back-end CMOS technology allows a maxi-
mum temperature of 400-450 °C, the limit being set by the
mechanical integrity of low dielectric constant intermetal
dielectrics.
7
Thermal CVD of SWNTs has been reported at 550 °Cin
furnace
8
and cold wall systems.
9
In situ environmental
transmission electron microscopy (TEM) experiments show
SWNT nucleation at 480 °C.
10
Random-network FETs have
been fabricated with SWNTs grown at 450 °C by remote
plasma-enhanced (PE) CVD.
11
However, the high temper-
atures of bulk production techniques still dominate growth
model considerations with the assumption that the catalyst
cluster has to be liquefied and that the catalyst bulk is rate-
controlling.
12,13
These considerations are also transferred to
surface-bound CVD.
14
CNT growth below 500 °Cisnot
thought to be possible based on calculations of size-corrected
melting points
14
and carbon saturation.
15
In this Letter, we report SWNT growth at temperatures
below 450 °C by thermal CVD at cold wall conditions and
demonstrate field effects in as-integrated SWNT FETs. We
use evaporated thin catalyst films, which allow accurate
patterning by standard lithography techniques and thus
compatibility with integrated circuit design. The restructuring
of a solid thin film into catalytically active islands requires
mobility and thus thermal activation, which should be
distinguished from the low-temperature limit of the actual
SWNT nucleation. We show that NH
3
or H
2
exposure
* To whom correspondence may be addressed. E-mail: sh315@
cam.ac.uk.
†
Electrical Engineering Division, University of Cambridge.
‡
Department of Materials Science and Metallurgy, University of
Cambridge.
§
Hitachi Cambridge Laboratory.
NANO
LETTERS
2006
Vol. 6, No. 6
1107-1112
10.1021/nl060068y CCC: $33.50 © 2006 American Chemical Society
Published on Web 05/02/2006
critically affects the dewetting and activation of sub-
nanometer Fe and Al/Fe/Al
16
multilayers on Si/SiO
2
sub-
strates upon temperature elevation, enabling an appropriate
catalyst nanostructuring at low temperatures. SWNT growth
is observed at temperatures as low as 350 °C upon
subsequent exposure to pure C
2
H
2
at low pressures (<10
-2
mbar). Raman analysis shows that such low temperature
growth leads to a narrower SWNT diameter and possibly
SWNT chirality distribution, confirming trends observed by
other groups.
8,17,18
Combined with previous ab initio calcula-
tions,
19
we suggest that surface, rather than bulk catalyst
effects, dominate SWNT growth kinetics in surface-bound
CVD.
We use polished, 0.01-0.02 Ω cm boron-doped Si(100)
wafers as substrates, covered with thermally grown SiO
2
(∼50-200 nm). The SiO
2
layer can prevent uncontrolled
silicide formation
20
and serves as gate dielectric for FET
fabrication. High-purity Fe, Co, and Ni catalyst films and
Al layers are deposited by thermal evaporation. The evapora-
tion rate is <1 Å/s at <10
-6
mbar base pressure. No substrate
heating is used during evaporation. The film thickness is
monitored in situ by a quartz crystal microbalance and
calibrated ex situ by atomic force microscopy (AFM, Digital
Instruments Nanoscope III and Veeco Explorer) and spec-
troscopic ellipsometry (J. A. Woollam Co., M-2000 V). The
catalyst films are patterned by optical or e-beam lithography
using S1813 or PMMA as resist. The substrates are trans-
ferred in air and loaded onto a resistively heated graphite
stage (Figure 1a). The stainless steel CVD chamber is
diffusion pumped (base pressure <10
-6
mbar) with mass-
flow-controlled gas feeds.
The temperature is continuously monitored by three
shielded thermocouples distributed across reference Si
substrates (500 µm in thickness, equivalent to samples) and
the graphite heater, Figure 1a. All temperatures indicated
refer to the substrate surface temperature measured on top
of the Si and not to the heater block (which has higher T for
a typical poor thermal contact) and thus truly represent the
catalyst temperature. It is known from general heterogeneous
catalysis that significant temperature differences between an
individual catalyst particle and its support will not exist under
realistic conditions of rate and size.
21
This holds for exo-
thermic feed-gas dissociation as well as plasma heating.
The catalyst film is heated in NH
3
(grade 5, 0.6-20 mbar)
or H
2
(VLSI grade, 0.6-100 mbar) for typically 15 min to
equilibrate the desired growth temperature. For some
samples, catalyst reconstruction is aided by a dc plasma
excitation for ∼30s(<15 W power). The chamber is then
evacuated and undiluted C
2
H
2
(AA grade, 99.6% purity) is
allowed to flow from a side gas inlet at 10
-3
-10
-2
mbar.
After a growth period of 5 min, the samples are cooled in
vacuum.
The samples are characterized by scanning electron
microscopy (SEM, LEO 1530VP FEGSEM), high-resolution
(HR) TEM (JEOL JEM 4000EX, 400 kV; FEI Tecnai F20,
200 kV), and Raman spectroscopy (Renishaw 1000 Raman
spectrometer, 514.5, 633, and 785 nm excitation). For
HRTEM analysis, SWNTs are either removed from the
substrates and dispersed onto lacey carbon TEM grids or
grown directly onto e-beam transparent, 50 nm thick silicon
nitride membranes (Agar Scientific).
Ti and Pd source and drain contacts for FET devices are
sputter-deposited or evaporated through a PMMA lift-off
mask, to create contact separations in the range of 300-800
nm. Electrical characteristics are collected at ambient condi-
tions.
SWNT nucleation requires nanometer-sized catalyst island
dimensions
22-24
and thus a very thin initial catalyst film.
25
Figure 1. (a) Graphite heater stage with shielded thermocouples. (b-f) SEM images of SWNTs thermally grown in undiluted C
2
H
2
for 5
min from (b-d) 0.3 nm Fe, (e) Al/Fe (0.5 nm)/Al (top, 0.2 nm), and (f) Al/Fe (0.3 nm)/Al (top, 0.2 nm) multilayers on Si/SiO
2
substrates.
The catalyst films are annealed in NH
3
at (b-d, f) 0.6 mbar and (e) 20 mbar, prior to SWNT growth. The process temperature is indicated
for each sample. Scale bars: 100 (b-d), 500 (e), and 200 nm (f).
1108
Nano Lett.,
Vol. 6, No. 6, 2006
Larger metal clusters can catalyze MWNTs and CNFs.
25
Parts b-d of Figure 1 show SEM images of SWNTs
nucleated from a 0.3 nm Fe film on Si/SiO
2
substrates. A
transition from long, bundled SWNTs distributed mainly
laterally on the substrate (Figure 1b,c) to a mixture of
SWNTs and more vertically extended, thicker MWNTs/
CNFs (Figure 1d) is observed with decreasing growth
temperature. The lowest temperature at which we could
detect SWNTs by HRTEM and Raman spectroscopy is 350
°C (Figure 3). Parts e and f of Figure 1 show evaporated
Al/Fe/Al multilayers after processing in similar growth
conditions to the pure Fe films (Figure 1b,d). The CNT yield
is much higher, resulting in vertical alignment due to van
der Waals interactions of neighboring nanotubes. At given
CVD conditions, preannealing the catalyst in a vacuum
(<10
-5
mbar pressure) leads to no CNT growth.
NH
3
and H
2
have often been used as diluents during CNT
growth, in particular for PECVD;
26,27
here we focus on their
role in catalyst restructuring prior to growth. Parts a-dof
Figure 2 show AFM topography and line sections of 0.3 nm
thick Fe as-evaporated and after annealing in different
atmospheres and temperatures. The AFM measurements are
performed in tapping mode at ambient conditions and
therefore include the effects of oxidation.
28
At 500 °C, the
NH
3
annealed Fe film (Figure 2b) shows smaller average
island dimensions than for vacuum annealing (Figure 2a).
Within the limits of the AFM analysis, annealing in NH
3
at
300 °C results in cluster sizes (Figure 2c) similar to the
distribution at 500 °C (Figure 2b). Sub-nanometer films
generally nucleate in a Volmer-Weber mode on SiO
2
and
thus can be discontinuous as-evaporated. Thin solid films
dewet driven by surface and elastic energy minimization;
29
the resulting islands sinter due to ripening or migration.
30
These processes tend to scale with the melting temperature.
As a rule of thumb, Huettig and Tamman temperatures are
semiempirically defined as 0.3 and 0.5 times the melting
point in K, respectively, to indicate the temperature at which
atoms become mobile at defects (269 °C for Fe) or in the
bulk (631 °C for Fe).
31
This neglects substrate interactions
20,32
and the fact that for small particles mobility occurs at much
lower temperatures.
33,34
NH
3
or H
2
exposure reduces initially
oxidized Fe and facilitates surface mobility of metal atoms
and clusters.
35
Adsorbed gases in general can act as “sur-
factants”, modifying surface energies.
36
As Figure 2 shows,
thin films form small catalyst islands at very low tempera-
tures. We find H
2
to require higher pressures for thermal
catalyst activation, which can be compensated by a weak
plasma excitation to etch and enhance catalyst reconstruc-
tion.
25,37
An Al underlayer (Tamman T 194 °C) enhances SWNT
yield (Figure 1e,f) by clustering itself, thereby increasing
the catalyst support surface area and decreasing the prob-
ability of catalyst sintering.
38
The Al surface is (at least)
partly oxidized during sample transfer after evaporation,
forming a stable alumina-catalyst interface. An (oxidized)
Al top layer additionally stabilizes small Fe clusters
16
and
thus increases SWNT yield (Figure 1e,f) compared to non-
Al-topped Al/Fe bilayers (results not shown here). Prelimi-
nary results on sub-nanometer Ni and Co films indicate less
efficient SWNT nucleation, which might be related to
different catalyst-substrate interactions for the used growth
conditions.
Figure 2. AFM topography images of (a) 500 °C vacuum (<10
-5
mbar) annealed, (b) 500 °CNH
3
(0.6 mbar) annealed, and (c) 300 °C
NH
3
(0.6 mbar) annealed 0.3 nm Fe on Si/SiO
2
. (d) AFM line sections of as-evaporated 0.3 nm Fe and samples shown in (a-c). All AFM
image dimensions are 500 × 500 nm
2
with identical z scale. (e) HRTEM image of 0.1 nm Fe on 50 nm thick silicon nitride membrane after
annealing in NH
3
(0.6 mbar) and exposure to undiluted C
2
H
2
for 5 min at 500 °C. The arrows mark a SWNT (scale bar: 10 nm).
Nano Lett.,
Vol. 6, No. 6, 2006 1109
Catalyst reduction can also occur due to hydrogen liberated
from the catalytic decomposition of C
2
H
2
.
32
For high SWNT
CVD temperatures, metal clusters are often deliberately kept
oxidized prior to growth to stabilize them against excessive
sintering. Metal oxides show a stronger interaction with
oxidized substrates and therefore lower mobility.
35
For our
conditions, however, the vacuum-annealed samples show
larger catalyst clusters (Figure 2) and no SWNT nucleation
at low temperatures. A careful monitoring of the catalyst
activation is thus essential when comparing different CVD
conditions.
Chemisorbed H can additionally lead to temperature- and
pressure-dependent catalyst surface reconstruction, thus
altering its interaction with C
2
H
2
.
39
Hydrogen penetration into
subsurface layers can loosen the catalyst surface
39
and may
give higher carbon diffusivities for low-temperature SWNT
nucleation.
19
Figure 2e shows a HRTEM image of a 0.1 nm Fe film on
an e-beam transparent silicon nitride membrane after SWNT
CVD at 500 °C. A low-density distribution of sub-3-nm Fe
particles is formed. As-nucleated SWNTs have diameters
ranging from 1.2 to 2.3 nm. It has to be emphasized that
depending on the wetting behavior the catalyst island shape
can be highly anisotropic and that the high metal mobility
leads to catalyst reshaping upon graphene formation during
CVD.
23,40
Thus the postgrowth catalyst structure indicates
SWNT nucleation sites
23
but is not representative of the initial
catalyst morphology. Within the limited number of processed
silicon nitride substrates, we observe a SWNT yield similar
to silicon oxide supports.
Figure 3a shows a HRTEM image of SWNTs nucleated
at 350 °C. We find closed SWNT tips, some of which present
a spherical catalyst particle, suggesting a tip growth model.
Even though there is a large variation in their morphology,
some SWNTs show a remarkable crystallinity, with clean
parallel walls. Parts b and c of Figure 3 compare the Raman
spectra (633 nm excitation) of samples grown at 350 and
500 °C from 0.1 nm Fe. Both spectra show radial breathing
modes (RBMs) (Figure 3b) and a structured G peak (Figure
3c).
41,42
More defects and disordered graphitic material are
present for the lowest growth temperature, as indicated by
the larger, more structured D peak, consistent with Figure
1d. Parts d and e of Figure 3 compare the relative
43
abun-
dance of metallic and semiconducting SWNTs seen by
Raman spectroscopy as a function of deposition temperature.
We derive the SWNT diameter by d ) C
1
/(ω
RBM
- C
2
),
with C
1
) 214.4 nm cm
-1
and C
2
) 18.7 cm
-142
and assign
based on the Kataura plot
44
and the known excitation
energy.
45
A RBM analysis of a large range of different
catalyst film thicknesses and CVD temperatures shows that
by lowering T and the amount of evaporated Fe we get
smaller average SWNT diameters with a narrower distribu-
tion. This is consistent with the catalyst coarsening behavior
(Figure 2) and the HRTEM analysis (Figures 2e and 3a) and
confirms trends reported for pulsed laser deposition at much
higher temperatures
46
and by CVD at 550-850 °C.
8,17,18
We
stress that due to the cutoff of our notch filter, we cannot
detect SWNT diameters >2 nm.
Figure 4 shows back-gated FET devices and typical I
ds
-
V
gs
characteristics from ∼5 µm wide strips of random-
network SWNT carpets nucleated from 0.1 nm thick Fe at
different temperatures. To selectively burn metallic CNTs
and thus increase the I
ds(on)
/I
ds(off)
ratio,
47
we apply a positive
V
gs
which turns off the semiconducting SWNTs (these show
Figure 3. (a) HRTEM image of SWNTs thermally grown at 350 °C for 5 min in undiluted C
2
H
2
from a NH
3
(0.6 mbar pressure) annealed
Al/Fe (0.3 nm)/Al (top, 0.2 nm) multilayer on Si/SiO
2
(scalebar: 10 nm). A SEM image of the sample has been shown in Figure 1f. (b,
c) Raman spectra (633 nm excitation) of SWNTs grown at 500 and 350 °C from 0.1 nm Fe on Si/SiO
2
(the peaks assigned to the Si
substrate are indicated). (d, e) Relative abundances of different SWNT diameters at 500 and 350 °C, respectively (514.5 and 633 nm
excitation). The assignment of semiconducting (S) and metallic (M) SWNTs and reference intensities (*) are indicated.
1110
Nano Lett.,
Vol. 6, No. 6, 2006
p-type characteristics in air) and then apply a large V
ds
. The
as-conditioned devices show a very consistent p-type gating
behavior, although the I
ds(on)
/I
ds(off)
ratio is not yet much
beyond 20. We are currently optimizing contact formation
48
and SWNT network density. Also a shorter gate length is
desirable to achieve high conductance through directly
bridging SWNTs
49
rather than allowing percolation conduc-
tion to define device operation.
50
An optimized (pregrowth)
catalyst nanostructuring, a lower C
2
H
2
partial pressure,
10,24
and additional diluent
11
can lower the SWNT defect density
and minimize deleterious MWNTs/CNFs contributions.
The commonly accepted CNT growth model constitutes
a rare example of a catalytic reaction in which the rate is
believed to be controlled by diffusion through the catalyst
bulk.
12-15,51
The idea of catalyst liquefaction
14
originates from
the vapor liquid solid (VLS) model originally proposed for
the growth of semiconductor whiskers.
52
A liquid catalyst
seed thereby acts as preferential adsorption site of gaseous
precursors, and the growth of a solid nanostructure is thought
to occur through precipitation at the liquid-solid interface.
We previously showed that CNF CVD is possible well below
the size-corrected melting points of mesoscale Ni, Co, and
Fe catalysts clusters,
19,53,54
even if saturated with carbon,
55
and suggested that CNF nucleation, like most processes in
heterogeneous catalysis, is dominated by the catalyst sur-
face.
19
Here we suggest that SWNT growth is mediated by
similar surface processes.
As discussed for thin film dewetting (Figure 2), small
catalyst particles can continuously change shape and surface
in a “fluidlike” behavior.
33,40
Although highly mobile, the
metal clusters are not amorphous (as one would expect for
a molten phase) but crystalline, in particular for surface-
bound, low-temperature CVD where catalyst-substrate and
catalyst-CNT interactions are relevant.
23,40
Without compar-
ing the detailed sticking coefficients of the CVD gas
precursors on substrate and catalyst, it can be argued that
CNTs grow selectively on the catalyst as it is the only
location where the barrier for thermal precursor dissociation
is low enough to support a controlled carbon flux.
19
The final
arrangement of the carbon atoms in a rolled graphene
structure is controlled by the surface of the metal particle
and its shape, which in turn can change with increasing
carbon coverage.
23,40
The smaller the catalyst cluster, the
more dominant its surface. Independent of a definition of
solid, liquid phase, or surface premelting, it can be expected
that catalytic SWNT growth is governed by surface pro-
cesses. Following our original argument,
19
recent molecular
dynamics simulations ofa1nmFenanoparticle show
negligible carbon penetration into the catalyst bulk and that
SWNT growth is fed by fast carbon diffusion on the catalyst
surface.
56
No supersaturation-segregation process is neces-
sary. Chemisorbed species such as H can significantly alter
a catalyst surface
39
and thus aid diffusive transport and
SWNT nucleation dependent on temperature and pressure.
To summarize, we showed the critical effects of NH
3
or
H
2
on Fe thin film catalyst restructuring, enabling low-
temperature SWNT CVD and opening new possibilities for
direct device integration. Conceptually, a liquid catalyst-
carbon eutectic is not a necessity; SWNT growth is domi-
nated by surface processes. The findings extend to catalytic
growth of anisotropic nanostructures in general.
57
Acknowledgment. The work was supported by the EU
project CANAPE. S.H. acknowledges funding from Peter-
house, Cambridge, V.S. from EPSRC Grant GR/S97613, and
C.D. and A.C.F. from The Royal Society. C.D. thanks FEI
Company, EPSRC, Isaac Newton Trust for the use of the
Tecnai TEM.
References
(1) Tseng, Y. C.; Xuan, P. Q.; Javey, A.; Malloy, R.; Wang, Q.; Bokor,
J.; Dai, H. J. Nano Lett. 2004, 4, 123.
(2) Sazonova, V.; Yaish, Y.; Ustunel, H.; Roundy, D.; Arias, T. A.;
McEuen, P. L. Nature 2004, 431, 284.
(3) Dresselhaus, M. S.; Dresselhaus, G.; Avouris, P. Carbon Nanotubes:
Synthesis, Structure, Properties and Applications; Springer: New
York, 2000.
(4) Dai, H. J. Acc. Chem. Res. 2002, 35, 1035.
(5) Jung, Y. J.; Homma, Y.; Ogino, T.; Kobayashi, Y.; Takagi, D.; Wei,
B. Q.; Vajtai, R.; Ajayan, P. M. J. Phys. Chem. B 2003, 107, 6859.
(6) Peng, H. B.; Ristroph, T. G.; Schurmann, G. M.; King, G. M.; Yoon,
J.; Narayanamurti, V.; Golovchenko, J. A. Appl. Phys. Lett. 2003,
83, 4238.
(7) Morgen, M.; Ryan, E. T.; Zhao, J. H.; Hu, C.; Cho, T. H.; Ho, P. S.
Annu. ReV. Mater. Sci. 2000, 30, 645.
(8) Maruyama, S.; Kojima, R.; Miyauchi, Y.; Chiashi, S.; Kohno, M.
Chem. Phys. Lett. 2002, 360, 229.
(9) Liao, H. W.; Hafner, J. H. J. Phys. Chem. B 2004, 108, 6941.
(10) Sharma, R.; Rez, P.; Treacy, M. M. J.; Stuart, S. J. J. Electron
Microsc. 2005, 54, 231.
(11) Min, Y.-S.; Bae, E.-J.; Oh, B. S.; Kang, D.; Park, W. J. Am. Chem.
Soc. 2005, 127, 12498.
Figure 4. Back-gated FETs and their I
ds
-V
gs
characteristics based on random-network SWNT carpets nucleated from H
2
plasma pretreated
(a, b) and NH
3
-annealed (c) 0.1 nm thick Fe films, exposed to 5 sccm of C
2
H
2
for 5 min at 500 °C and 420 °C. Sputtered Ti source/drain
contacts are 30 nm thick with a separation of ∼400 nm on a 200 nm thick SiO
2
gate, and the respective devices are measured after a V
ds
) 0-10 V burning sweep (V
gs
)+50 V) (a, b). Evaporated Pd source/drain contacts are 30 nm thick with a separation of ∼640 nm on
a 50 nm thick SiO
2
gate (c). The gate-leakage current is <10 pA.
Nano Lett.,
Vol. 6, No. 6, 2006 1111
