CARBON 4 7 ( 2 0 0 9 ) 2 0 8 5 –2 0 9 4

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/carbon

Dimensional control of multi-walled carbon nanotubes

in floating-catalyst CVD synthesis

Gregg S.B. McKeea, Christian P. Decka, Kenneth S. Vecchiob,*

a
Materials Science and Engineering Program, University of California-San Diego, La Jolla, CA 92093, USA
b
Department of NanoEngineering, University of California-San Diego, 9500 Gilman Drive, MC-0448, La Jolla, CA 92093, USA

A R T I C L E I N F O A B S T R A C T

Article history: Dimensional control in CVD synthesis of MWNT’s is significant and critical to a number of
Received 15 July 2008 different applications. This study examines the dimensional effect of a number of synthe-
Accepted 27 March 2009 sis variables on the products of floating-catalyst CVD, including catalyst concentration in
Available online 5 April 2009 the feedstock, nanotube growth time, and deposition substrate selection. Extensive diam-
eter surveys are performed by TEM and compared with results from thermo-gravimetric
analyses to Raman spectroscopy, offering a novel dimensional analysis of nanotubes grown
by FC-CVD methods. CNT diameters are inversely proportional to the catalyst concentra-
tion with weak correlation over the range examined and are directly proportional to growth
time. Results are combined with prior art to develop a new theory regarding catalyst parti-
cle formation over a range of catalyst concentrations. Carbon deposition occurs in two
stages, the first characterized by accelerating deposition and increases in CNT diameter
and length, the second by etching of the array and carbon deposition at a constant rate.
Deposition substrates interact directly with the catalysts to strongly influence the resulting
nanotube diameters, based upon the mobility of the catalysts on the substrate surface.
 2009 Elsevier Ltd. All rights reserved.

1. Introduction amount of literature and the wide variety of synthesis varia-

Given the extraordinary properties of carbon nanotubes
(CNT’s), from high specific surface area to tunable electric
conductivity to high specific stiffness and strength, there
has been a rush to develop applications for nanostructures
before the methods for synthesis are adequately understood.
Field emission devices [1–3], nano-electric devices [1,4–6], and
structural composites [7,8] are emerging, making use of CNT’s
as core elements. However, the synthesis of economical,
reproducible, quality nanotubes remains a significant chal-
lenge, which must be addressed before widespread adoption
of carbon-nanotube or any nanostructure-centric application
can occur. To date, studies in CNT dimensional control re-
main very unusual [9–11], especially given the enormous
tions available for CNT’s.
Floating-catalyst chemical vapor deposition (FC-CVD) syn-
thesis of CNT’s has its origins in the synthesis of vapor-grown
carbon fibers (VGCF). For many years CNT’s were observed in
VGCF products, but not recognized as anything but particu-
larly small fibers. In recent years, reports have suggested at
least some VGCF originate from or otherwise contain hollow
CNT cores [12,13]. The synthesis process and proposed
growth models for the two are nearly identical [14–16]. For
this reason, it may be expected that some of the synthesis
factors influencing the properties of resulting VGCF might
also be significant in the synthesis of CNT’s. For example,
the ratio of catalyst to carbon source has been shown to be
proportional to the fiber diameter and aspect ratio of VGCF
[16]. If the same were applicable to multi-walled nanotubes

* Corresponding author: Fax: +1 858 534 5698.

E-mail address: kvecchio@ucsd.edu (K.S. Vecchio).
0008-6223/$ - see front matter  2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.carbon.2009.03.060
2086 CARBON 4 7 ( 2 0 0 9 ) 2 0 8 5 –2 0 9 4
(MWNT), it would allow for more precise control over dimen-
sions in an area where most dimensional control is per-
formed empirically, if at all. FC-CVD is quickly emerging as
one of the most scalable and least expensive methods for pro-
ducing high quality single-walled nanotubes (SWNT) and
MWNT, thus, there is great need for a better understanding
of the variables, which define the resulting products.
Numerous studies have examined the importance of dif-
ferent growth parameters in catalyst assisted CVD growth of
carbon nanotubes; however, these reports tend to focus
exclusively on synthesis methods using thin film catalysts,
and the control of nanotube diameters in floating-catalyst
CVD has not been widely researched. With thin film catalysts,
nanotube diameters are controlled by varying the metal film
deposition thickness [17], but this approach cannot be applied
to FC-CVD, as the method of catalyst introduction is funda-
mentally different. Instead of thin films, which coalesce into
catalyst islands during heating, catalyst material is continu-
ously added during floating-catalyst growth, and catalyst par-
ticles form as reactants decompose and deposit on substrate
surfaces. As a result, nanotube diameters are affected by
parameters that do not apply to conventional thin film CVD
growth, such as catalyst solution concentrations. Of the few
reports on diameter control using FC-CVD, the conclusions
with regard to catalyst concentrations have not always been
consistent, with increasing concentrations resulting in either
reduced diameters [18], increased diameters [19–21], or show-
ing little influence [22,23]. Of these reports, some examine
only single-walled tubes [21,23], and those studying multi-
walled tubes use widely different reaction temperatures
(ranging from 675 C [19] to 1150 C [18]), and catalyst concen-
trations (ferrocene-hydrocarbon molar ratios of 1.17–15.1%
[18], 0.6–6.0% [19], 0.10–5.26% [20], and 0.44–1.73% [22]). Con-
trol of nanotube diameters is important, and the mechanisms
by which the catalyst concentration can affect nanotube
dimensions must be understood. To further investigate this
relation, an intermediate reaction temperature (750 C) and
catalyst concentration range (0.23–9.56% molar ratio) were se-
lected for this work.
This work examines several variables relevant to dimen-
sional control in vapor-grown carbon fiber synthesis and their
dimensional impact on FC-CVD synthesis of MWNT. We pro-
pose herein a mechanism to explain these apparently con-
tradicting reports on the influence of catalyst
concentrations on nanotube diameter. At lower catalyst con-
centrations (such as those used by Sinnott et al. [19] and
Singh et al. [20]) nanotube diameters increase with increasing
catalyst concentrations, while at high concentrations (used
by Bai et al. [18]), larger metal particles form, which are not
suitable for nanotube formation. In addition, other reports
have indicated a relationship between substrate material
and nanotube diameter for thin film catalyst CVD growth
[24]. As most previous parametric studies with FC-CVD used
only a single substrate material (most often quartz), different
substrates were selected in this work to see if this parameter
had any effect on diameters with FC-CVD.
Analysis is accomplished through extensive TEM diameter
surveys, as well as Raman, thermo-gravimetric, and atomic
force microscopy (AFM) analyses. Novel structure, reactivity,
and dimensional trade-offs with catalyst concentration and
growth time are identified. An analysis of the substrate choice
highlights newly observed interactions between the substrate
and catalysts delivered by the FC-CVD method.
2. Experimental
CNT’s were synthesized through the pyrolytic decomposition
of a benzene-ferrocene feedstock solution in a CVD furnace as
shown in Fig. 1. Nanotube synthesis began by flowing 99.5%
argon through the reaction chamber (a 2’’ diameter quartz
tube) as the furnace heated to the synthesis temperature.
Once the furnace reached the desired temperature (approxi-
mately 750 C at the growth zone, 120 C at the feedstock
injection point), the ferrocene–benzene feedstock solution
was fed into the chamber, where it was collected at the
mouth of the tube in an alumina crucible, approximately
300 mm from the growth zone, and vaporized by heat from
the furnace. About 100 sccm Ar (99.5%) flow carried the vapor
into the furnace, where CNT deposited onto substrates over a
length of 75 mm. After the desired growth time elapsed, any
remaining solution was removed, the furnace shut down, and
argon flow maintained until the chamber had completely
cooled.
The catalyst concentration, growth time, and substrate
type were examined separately. In each set of experiments,
all variables not under examination (furnace temperature,
growth time, substrate, etc.) were held approximately
constant.
In order to examine the significance of catalyst concentra-
tion, the ratio of ferrocene to benzene in the feedstock was
systematically varied, from 4.75 g ferrocene per liter benzene
(0.227% molar ratio) up to the room temperature solubility
limit at approximately 200 g ferrocene per liter benzene
(9.58% molar ratio). Multiple synthesis runs were made for
each catalyst to carbon ratio for purposes of analysis
duplication.
Growth time was varied from 5 min up to 240 min by con-
trolling the time over which the catalyst/ferrocene feedstock
was introduced into the furnace (at constant rate). A variety
of substrates were examined: silicon, quartz, alumina, NiO,
mullite, graphite, an alumina-mullite mixture, a machinable
ceramic with approximate composition of 55% fluorophlogo-
pite mica in a borosilicate glass matrix, and a soda-lime glass
(72.2% SiO2, 14.3% Na2O, 1.2% K2O, 6.4% CaO, 4.3% MgO, 1.2%
Fig. 1 – Schematic diagram of the FC-CVD CNT growth setup.
A solution of the catalyst (ferrocene) and carbon source (ben-
zene) is introduced into the furnace in liquid form, thermally
vaporized, and carried over the substrate where aligned CNT
arrays are deposited.
 CARBON 4 7 ( 20 0 9 ) 2 0 8 5–20 9 4
Al2O3, 0.3% SO3). All substrates were polished to at least
4000 grit prior to CNT deposition. Growth time and sub-
strate-related experiments used a catalyst to carbon ratio of
approximately 15 g/L. To measure catalyst mobility on sub-
strates, 10 nm of Fe was e-beam evaporated upon the sub-
strates in question, which were then sintered for 10 min
under approximately 150 ccm of 10% H2 90% Ar, to replicate
the reducing atmosphere present in the synthesis chamber
during nanotube growth.
Residual iron content in as-produced nanotubes was cal-
culated through complete oxidative removal of the carbona-
ceous component. Large samples (5–100+ mg) were held at
650 C in air for several hours and the total mass change re-
corded. By assuming complete conversion of the catalyst to
Fe2O3, each residual mass was corrected for oxygen mass
(Oxygen makes up 30.06% of Fe2O3 mass) and the mass of re-
tained iron calculated.
Deposition rates were calculated from the amount of car-
bon deposited on a 75 · 25 mm quartz slide placed at the
beginning of the growth zone in the reactor.
Nanotube dimension measurements were made via SEM
(Philips XL20, 15 kV accelerating voltage) or TEM (Philips
EM420, 120 kV accelerating voltage, or JEOL JEM-2011, 200 kV
accelerating voltage) observations. TEM samples were pre-
pared through sonication in MeOH for 2.5 min and dripped
onto a 200 mesh Cu grid. Micrographs were collected between
70,000 and 150,000· magnification and nanotube diameters in
each measured and tabulated. At least 250 individual and dis-
tinct nanotubes from each sample were measured to provide
a statistically accurate representation. Catalyst particle size
measurements were made by contact-mode atomic force
microscopy (Nano-R, Pacific Nanotechnologies).
Oxidation mass-loss profiles were obtained in a Perkin–El-
mer Pyris Diamond Thermo-Gravimetric/Differential Thermal
Analyzer (TG/DTA). Masses of approximately 1 mg were
placed in open-top alumina pans under a purge gas of 2%
O2, balance He. The thermal program initiated the purge
gas, raised the temperature to 200 C at 40/min, soaked for
10 min, then increased to 1000 C at the rate of 1/min (in or-
der to simulate the reaction equilibrium state and allow direct
comparison between these data and those found in other
works) [25–27]. Each sample was analyzed twice to ensure
repeatability of the results.
Raman spectroscopy (Renishaw Raman) was performed at
room temperature using an argon ion laser (514.5 nm,
0.35 mW @ sample). Data was collected from three distinct
locations of the exposed side of an aligned array of nanotubes
for each sample. Three spectra collected for each location were
normalized to the D-band (1350 cm 1) peak and averaged to-
gether to eliminate minor variations in the individual spectra.
In order to examine the relationship between catalyst con-
centrations, carbon levels in the furnace, and nanotube diam-
eters, a two-stage CVD synthesis was performed with the goal
of creating an environment of excess carbon for the cases of
high and low catalyst concentration. In the first stage of syn-
thesis, a growth feedstock with either 12.5 or 151 g ferrocene/
L benzene was used. After 2.5 min of growth, the feedstock
containing ferrocene was removed and replaced with pure
benzene for an additional 27.5 min. Dimensions of the result-
ing deposits were then found via TEM.
2087
3. Results and discussion
CNT were deposited in aligned arrays with thicknesses from
100 lm to a few millimeters (Fig. 2). Each sample was tested
as-synthesized (excepting mechanical processing for TGA &
sonication for TEM sample prep) without treatment for the re-
moval of non-nanotube impurities. Samples were observed to
be generally free of non-nanotube carbonaceous impurities at
SEM levels of magnification. At higher magnifications in the
TEM; however, certain samples were observed to have poten-
tially significant levels of carbonaceous and catalyst particle
impurities, as will be discussed later.
For the purposes of this investigation, cleaning of the CNT
by refluxing or another partial oxidation technique may have
removed excess catalyst particles and non-nanotube carbo-
naceous impurities, but would have resulted in significant
damage to the CNT structures, such that any meaningful
structural analysis would have been impossible.
As might be expected, a strong correlation between the
catalyst to carbon ratio and the retained iron content is ob-
served (Fig. 3). The relationship roughly follows a parabolic
curve, with a rapid increase in the retained iron content at
low catalyst concentrations (0–25 g/L), followed by a more
gradual increase at higher concentrations. Some variability
in iron content is evident for synthesis runs at a given catalyst
concentration, but the overall pattern shows a proportional
increase in one with the other. In only one of the many syn-
thesis runs did the retained iron content drop below
1.25 wt.% Fe, possibly suggesting a minimum iron content
for CNT formation (catalyst limited growth). At low catalyst
concentrations, a situation of excess carbon may exist (cata-
lyst limited growth), such that amounts of carbon cannot be
incorporated into the CNT and may deposit later in the fur-
nace, beyond the CNT deposition area, or be carried out in
the exhaust stream. Saturation of the retained iron content
with high catalyst concentrations may imply an upper limit
at which point there is insufficient carbon for the formation
of a nanotube from each catalyst particle (carbon limited
growth).
Fig. 2 – An example SEM micrograph showing the macrosco-
pic alignment of the CNT array subsequent to removal from
the growth substrate. The degree of alignment may be con-
sidered generally representative of all aligned arrays exami-
ned. Scale bar is 200 lm.
2088 CARBON 4 7 ( 2 0 0 9 ) 2 0 8 5 –2 0 9 4
Fig. 3 – Retained iron content plotted as a function of the fer-
rocene to benzene ratio in the synthesis feedstock. The cur-
ve shows a direct proportionality between the two variables
and is roughly parabolic up to the room temperature solubi -
lity limit of ferrocene in benzene (200 g/L).
Altering the catalyst concentration in the feedstock with-
out modifying other synthesis conditions does not appear to
directly change the resulting deposition rate (Fig. 4). Large
changes in the catalyst concentration produce no predictable
change in the yield, when performed under similar circum-
stances. Similarly, a large range of deposition rates can be ob-
tained using very similar concentrations (Fig. 4, note yield
range at 15 g/L) by varying the growth time. Data points at
concentrations other than 15 g/L had growth times of approx-
imately 90 min. Thus, it is apparent that the relationship be-
tween yield and ferrocene concentration, if any, is minimal in
comparison with the relationship between yield and growth
time.
Results of the survey of diameter distribution as a function
of retained iron content are summarized in Fig. 5. More de-
tailed distribution results for individual samples may be
found in the Supplementary information. Significant scatter
Fig. 4 – Deposition rate (yield) is plotted as a function of
catalyst concentration. Wide variations of the catalyst con-
centration produce no predictable change in yield. However,
by varying the growth time, large changes in yield may be
obtained without altering the catalyst concentration. Thus,
yield appears to be independent of the iron to carbon feed-
stock ratio, but is strongly dependent upon growth time.
in the data suggest no dependence of the diameter on the iron
content. However, if one chooses to exempt the 12 wt.% Fe
data point, a trend is weakly suggested, where the mean
diameter shows a slight decreasing trend with increasing iron
content. In either case, no trend is evident in the diameter’s
standard deviation. The diameter trend is in agreement with
previous results from Bai et al. [28], who studied catalyst con-
centrations from 1.17 to 15.1 at.%, but contrary to results by
Sinnott et al. [29], who reported on catalyst concentrations
of 0.075 at.% and 0.75 at.%. However, it should be noted that
the latter reported on only two samples/catalyst
concentrations.
Bai et al. [28] suggest that the increased catalyst content
creates a finer dispersion of catalyst particles as well as a car-
bon deficiency in the reaction zone, resulting in a smaller
mean CNT diameter. In order to test this hypothesis, a two-
stage CVD synthesis was performed with the goal of creating
an environment of excess carbon for the cases of high and
low catalyst concentration, as discussed in Section 2. CNT
dimensions resulting from these runs fell generally in line
with the other diameter measurements shown in Fig. 5: the
12.5 g/L sample showed a mean diameter of 37.2 nm and a
standard deviation of 15.3 nm; the 151 g/L sample showed a
mean diameter of 33.8 nm and a standard deviation of
19.3 nm. Histograms showing the numerical distribution of
the two samples may be found in the Supplementary
information.
The observed differences in standard deviation fall within
the scatter observed for all catalyst concentrations and can-
not be considered significant. The difference in diameters is
approximately 10% of the overall diameter and as such, falls
above the approximate margin of error of 6% for a 95% confi-
dence level. If the theory put forward by Bai et al. [28] were
accurate, the two samples should have statistically similar
diameters, but, as a difference has been observed, it must
Fig. 5 – CNT diameter and standard deviation are plotted
versus the retained iron content. A weak decreasing trend is
observed in the mean diameter as retained iron content in-
creases. Significant scatter is apparent in the standard dev-
iation and no direct relation to the retained iron content is
observed. Data points represent a simple random sample of
at least 250 data points, excepting the highest iron content
sample (12 wt.% Fe), for which 206 tubes were examined.
 CARBON 4 7 ( 20 0 9 ) 2 0 8 5–20 9 4 2089

be due to causes other than a carbon deficiency during

synthesis.
It would be generally expected that a higher catalyst con-
centration will lead to more and larger catalyst particles than
a comparatively lower catalyst concentration. It is also ex-
pected that the growth rate of the catalyst particles (the rate
at which they will encounter reactant moieties in the furnace,
incorporate them, and increase the particle’s dimensions) will
depend directly upon the dimensions of the particle at the
time in question. Once a catalyst particle’s dimension sur-
passes some threshold; however, the likelihood that a CNT
will grow from it diminishes, even as the particle’s growth
rate continues to increase. Thus, the nanotubes that form will
do so from smaller catalyst particles (on average) with com-
mensurately smaller diameters. The larger catalyst particles
will tend towards faster growth to the point at which they ex-
clude themselves due to size, leading to a smaller average
CNT diameter with the higher catalyst concentrations. On
the other hand, at very low concentrations (lower than those
explored here), the particles would be size-limited due to an
unavailability of catalyst material, leading to a proportional
relationship between catalyst concentration and CNT
diameters.
When viewed in conjunction with the two previously men-
tioned reports [28,29], a more generalized picture may be ob-
served: at very low catalyst concentrations, catalyst particle
size (and thus, CNT diameter) is directly proportional to the
catalyst concentration. As the catalyst concentration is in-
Fig. 6 – Raman studies of CNT structure are summarized in
creased; however, many catalyst particles grow to exceed
terms of the (a) G:D and (b) G’:D peak ratios, plotted versus
the threshold for CNT formation, leading to increased non-
the retained iron content in the sample. The same approx-
CNT contamination in the sample and the formation of smal-
imate minimum of peak ratios is observed across all iron
ler diameter nanotubes.
contents; however, as iron content is decreased, the maxi-
There exists considerable scatter in the mean CNT diame-
ma of the data range increases as well. This suggests the
ters for all catalyst concentrations investigated. As such,
potential, but not guarantee, for improved structures with
while the catalyst concentrations examined herein may im-
decreasing iron contents.
pact the CNT dimensions, variation within this range does
not appear to be a reliable method for controlling their
dimensions, as it is in the case of VGCF’s. Other variables ap-
pear to have significantly larger and more reliable impacts. the data range decreases, with approximately the same min-
Raman spectroscopy performed on MWNT’s generally re- imum for all iron contents measured. This suggests that
sults in a characteristic fingerprint: the D band, which ap- CNT’s with lower iron content have the potential for signifi-
pears at approximately 1350 cm 1, is associated with sp3- cantly improved structural perfection versus those with larger
hybridized carbon and disordering in the nanotube, while amounts of retained iron, but are not guaranteed to be so im-
the G band appears at approximately 1580 cm 1, a result of proved. This also contradicts the finding by Bai et al. [28] that
sp2-hybridized carbon bonding (the in-plane stretching mode impurity content in CVD generated CNT’s decreases when the
for graphene sheets) [30]. The G’ band is the second order har- catalyst to carbon source ratio is increased. However, that
monic of the D band and appears in the realm of 2700 cm 1. study appears to have been performed by SEM observations,
Other higher order peaks occur in the neighborhood of the so results may have been more representative of localized
G’ peak as well, but are generally not useful for CNT charac- areas rather than bulk materials.
terization. As CNT quality measurements are frequently per- While the as-synthesized CNT arrays were observed to be
formed through comparison of the G to D band peaks in the clean on the micron scale (Fig. 7a), nanoscale impurities were
Raman spectra [31,32], Fig. 6a summarizes the Raman spec- observed in certain samples. For example, samples using cat-
troscopy results in this fashion. However, it has been shown alyst concentrations equal to or greater than approximately
that the G-band resonance in CNT’s shows some defect sensi- 75 g ferrocene per liter benzene were often found to contain
tivity and thus, comparison of the G’ band peak (2700 cm 1) significant quantities of carbon coated catalyst particles, both
to the D band peak may be a more accurate measure of CNT isolated and attached to nanotubes, in addition to those ob-
quality [33]; G’:D peak ratios are presented in Fig. 6b. served enclosed within CNT’s (Fig. 7b). As this type of contam-
While the Raman spectroscopy results summarized in ination was generally limited to the higher catalyst ratios
Fig. 6a and b shows a wide range of G:D and G’:D peak ratios used and not observed in other samples, it is likely a direct re-
at low retained iron content, as the iron content is increased, sult of the higher concentrations used in synthesis, and corre-
2090 CARBON 4 7 ( 2 0 0 9 ) 2 0 8 5 –2 0 9 4
Fig. 7 – (a) An example SEM micrograph of a CNT sample.
The CNT’s examined appear free of impurities on the mic-
ron scale. Scale bar for this micrograph is 2 lm. (b) Samples
synthesized using the higher catalyst concentrations analy-
zed were frequently observed to contain significant quanti
ties of carbon coated catalyst particles in addition to those
observed enclosed within CNT’s. As this type of contamina-
tion was not observed at lower catalyst concentrations, this
is likely a direct result of higher catalyst concentrations. Sca-
le bar is 100 nm.
lates well with the proposed catalyst particle formation
scheme.
A summary of thermo-gravimetric oxidation analyses ap-
pears in Fig. 8. As the mass-loss curves are asymmetric over
temperature and in order to concisely compare oxidation pro-
files for many samples, the data points plotted in Fig. 8 corre-
spond to a reaction coordinate of x = 0.5 (each profile showed
only a single oxidation peak indicative of CNT burn-off). The
oxidation profiles reveal an increase in the oxidation temper-
ature with decreasing iron content. This indicates a decreased
reactivity in the CNT, which stems from at least two different
mechanisms.
The first arises from the presence of iron particles catalyz-
ing the oxidation of CNT’s [34–37]. As the amount of retained
iron increases, the amount of catalytic material is increased,
resulting in a higher density of catalytically active sites. As
the number of sites is diminished (through reduction of the
iron content), the reactivity of the structure decreases, shown
by an increase in the temperature of oxidation.
Fig. 8 – Thermo-gravimetric burn-off of CNT samples is sum
marized and plotted versus the sample iron content. In ord-
er to summarize the oxidation curve with a single number,
the temperature corresponding to oxidation of half the sam-
ple mass is used. As iron content in the sample decrea-ses,
burn-off temperatures increase, indicating less reactive str-
uctures. This change occurs through diminished defect den-
sities and less catalytic oxidation by the iron particles.
The second mechanism arises from the improved struc-
tural perfection demonstrated in the Raman spectroscopy
analyses reported here. Defects in CNT’s serve as sites of in-
creased reactivity [35,38], so as the peak ratio increases, indi-
cating a smaller density of defect structures, the reactivity of
the CNT structure will decrease. Both mechanisms serve to
decrease the reactivity of the CNT’s and thus, increase the
sample’s temperature of oxidation.
Elongation of the CNT array and carbon deposition rate ap-
pear to exhibit two distinct stages as growth time is increased
(Fig. 9a and b). Throughout the first stage (stage 1 growth), ar-
ray length increases linearly and the deposition rate acceler-
ates (as noted by the increasing slope of the fit line in
Fig. 9b). In the second stage (stage 2 growth), array length be-
gins to diminish while the mass continues to increase, but
now at a linear rate.
Deposition and growth during stage 1 may be understood
if one considers the changing morphology of the deposition
surface. Initially, only the deposition substrate is present. At
this point, carbon is deposited onto the substrate surface,
generally as nascent CNT’s, or is incorporated into their elon-
gation. As the tubes elongate, wall area is increased, thereby
increasing the available surface area for carbon deposition.
Carbon continues to be incorporated into the elongation of
the CNT’s, but secondary hydrocarbon pyrolysis leads to car-
bon deposition on the sidewalls, increasing CNT diameters
(and further increasing the wall area available for deposition).
Diameter survey results (Fig. 10) confirm an increase in CNT
diameter and standard deviation with increasing growth
time. Thus, as the available surface area for deposition in-
creases, so the rate of deposition increases. The increase in
the standard deviation with growth time may be attributed
to the constant nucleation of nanotubes. As growth time pro-
gresses, carbon deposition will occur on the surfaces of exist-
ing nanotubes, increasing their diameters, but will also
contribute to the formation of new tubes which will grow in
 CARBON 4 7 ( 20 0 9 ) 2 0 8 5–20 9 4
Fig. 9 – (a) CNT array length and (b) deposition rate are plot-
ted as a function of growth time. Both may be separated into
two phases of growth and deposition. Stage 1 growth shows
linear array elongation and accelerating deposition rate
with increasing time. Stage 2 growth shows etching of the
CNT array and continued carbon deposition, but now at a li-
near rate.
Fig. 10 – CNT diameter and standard deviation are plotted
versus the growth time during stage 1 growth. Accelerating
carbon deposition during stage 1 occurs as the surface area
available for deposition increases as CNT’s elongate. This
results in an increase in average CNT diameters as well as in
the diameter’s standard deviation as growth time increases.
diameter with passing time, thereby increasing the overall
distribution of CNT diameters with growth time.
Growth changes to stage 2 at approximately 120 min. Two
potential reasons behind this change include: (1) carbon and
2091
new catalysts can no longer reach the substrate and base of
the array due to CNT thickening or (2) all active catalyst par-
ticles have become poisoned and can no longer support
CNT growth. The decrease in the average array length in con-
cert with the increasing carbon mass also suggest preferential
etching of the top (more exposed) surface of the array and
redistribution or redeposition of that carbon into the bulk of
the array, leading to a densification of the material. Addition-
ally, carbon continues to deposit, but does not appear to be
incorporated into the tubule structures. Thus, the available
Fig. 11 – (a) Nanoscale non-nanotube carbonaceous conta-
mination was observed on larger CNT’s in several samples.
This type of contamination occurred in samples with incr-
eased growth times, across the spectrum of catalyst concen-
trations examined. Scale bar is 100 nm. (b) A bi-layer struc
ture was observed on some of the CNT’s examined, causing
significant variation in the outer diameter of the CNT, while
the inner diameter and inner layer thickness remained app-
roximately constant. Scale bar is 100 nm. (c) The inner layer
appears to be organized graphene layers, while the outer
layer appears to be a less organized form of carbon, proba-
bly amorphous. White lines indicate the approximate boun-
daries of each layer. Scale bar is 20 nm.
2092 CARBON 4 7 ( 2 0 0 9 ) 2 0 8 5 –2 0 9 4

Fig. 12 – Mean CNT diameter and standard deviation are
compared across several different deposition substrates.
Substrate impact on CNT dimensions appears to be related
to the mobility of the catalyst particles on the substrate
surface, which governs the ability of the catalysts to sinter
and agglomerate.
surface area does not significantly change as a result of con-
tinued deposition, yielding a linear increase in mass and con-
stant deposition rate. For these reasons, with increasing
growth times, synthesis products transition from aligned ar-
rays of CNT to aligned arrays of nano-fibers to a CNT-rein-
forced-carbon composite material.
Nanoscale carbonaceous contamination was observed on
larger tubes in certain samples (Fig. 11a). In this case, there
is no obvious link between the catalyst concentration ratio
used and the contaminated samples; several samples across
the range of ratios examined displayed varying degrees of car-
bonaceous deposits attached to the CNT walls. Instead, this is
a result of longer growth times, showing the impact of addi-
tional carbon deposition on CNT walls. Additionally, some
samples with this type of contamination showed an interest-
ing bi-layer structure on certain CNT’s, similar to that noted
elsewhere [20,31]. The outer layer varies significantly in thick-
ness, while the inner layer’s thickness and the tube’s inner
diameter remain approximately constant (Fig. 11b). In these
cases, the tube structure appears to be an inner set of orga-
nized graphene layers coated with a less organized outer layer
of carbon, probably amorphous (Fig. 11c), supporting the pro-
posed mechanism for stage 1 growth.
Fig. 12 summarizes the results of the diameter survey on
the various substrates tested. No deposits were observed on
graphite substrates and the NiO, soda-lime glass, and alu-
mina/mullite combination substrates did not yield well-
aligned CNT arrays. Notably, a significant difference in mean
CNT diameter is observed between substrate materials. Most
substrates tested yielded mean CNT diameters of approxi-
mately 25 nm. Alumina and quartz substrates fell signifi-
cantly outside this range, with mean diameters of 15.8 nm
and 36.1 nm, respectively.
It has been recently suggested that iron mobility may vary
significantly from one substrate material to another [39], lead-
ing to the diameter variations observed. A lower catalyst
mobility on the substrate would limit sintering and agglomer-
ation of the catalyst, thereby yielding smaller catalyst parti-
cles and a lower mean CNT diameter. Given the CNT
diameter differences observed, it is expected that iron mobil-
ity on alumina would be lower than on silicon, which, in turn,
would be lower than on quartz. Examination of 10 nm iron
thin-films deposited on these substrates and sintered for
10 min confirms this prediction (Fig. 13). Particles are modeled
as semi-ellipsoid shapes and increase in size from alumina
(radii: 275, 271, and 31 nm) to silicon (radii: 315, 280, and
45 nm) to silica (radii: 429, 364, and 69 nm).
It is proposed that the observed diffusivity difference be-
tween silica and alumina is related to the inter-atomic bond-
ing within the oxides which influences the ease with which
catalyst oxides may form as well as the interactions of the
catalyst oxide with the substrate. The bond character in alu-
mina tends to be very ionic [40], whereas the bond character
in silica tends to be mostly covalent [41]. Formation of iron
oxides, also predominantly ionic, has been observed in FC-
CVD on both silica and alumina substrates [42]. The iron
oxide may undergo ionic interactions with the alumina to
form a solid solution or potentially a new compound at the

Fig. 13 – About 5 · 5 lm AFM scans of (a) Alumina, (b) Silicon, and (c) Quartz substrates. About 10 nm of iron was e-beam
evaporated onto each of the substrates, which were then annealed at 850 C for 10 min under 10% H2, 90% Ar gas stream.
Resulting particle volumes increase, left to right, indicating increasing mobility of the catalyst on the substrate surface and
explaining the difference in CNT dimensions observed. Particles are modeled as semi-ellipsoid shapes; alumina: radii: 275,
271, and 31 nm; silicon: radii: 315, 280, and 45 nm; silica: radii: 429, 364, and 69 nm. Scale bars are 1 lm.
 CARBON 4 7 ( 20 0 9 ) 2 0 8 5–20 9 4
substrate surface (spinel structured Fe(Fe,Al)2O4) [43], pinning
the particle and thus, limiting its mobility on the surface. Fur-
ther, above 600 C amorphous (sputtered) SiO2 surfaces trans-
form to non-polar 4-membered siloxane rings [44], and would
not have similar interactions with the iron or iron oxide,
although an olivine structure with comparatively higher cata-
lyst content may be formed. Thus, the use of ionically bonded
substrates, particularly oxides with which the catalyst oxide
may form a solid solution, should favor smaller catalyst for-
mation during the sintering phase. This topic will be exam-
ined in more detail in a later paper.
4. Conclusions
Given the similarity of VGCF synthesis to FC-CVD CNT synthe-
sis, it was expected that CNT diameter would similarly depend
upon the catalyst to carbon source ratio. Catalyst particle
dimensions are traditionally understood to directly impact
the diameters of the CNT’s formed; however, in this case, a
higher catalyst concentration in the feedstock does not directly
translate to larger CNT dimensions. The results of the TEM
diameter survey indicate a relatively weak correlation between
CNT dimensions and the catalyst concentrations examined,
but with the opposite proportionality versus what was ex-
pected. It has been shown that this dependence is not a result
of carbon depletion within the reaction zone. Instead, the new-
ly proposed theory suggests that at very low catalyst concen-
trations, catalyst particle size is limited by the availability of
catalyst material and thus, directly proportional to the catalyst
concentration, but as the concentration is increased, catalyst
growth accelerates proportionately with size and many parti-
cles grow to exceed the threshold for CNT formation, leading
to increased non-CNT contamination in the sample and CNT
formation from the smaller catalysts.
It is evident that there are trade-offs for catalyst concen-
tration and growth time. As catalyst concentration increases,
diameter decreases, but stability and structural perfection de-
crease, raising the reactivity of the CNT’s. As growth time in-
creases, the overall CNT array length increases, but so do the
average diameter and its standard deviation. Carbon deposi-
tion rate appears to be independent of the catalyst concentra-
tion, but increases in two stages with the growth time, with
CNT yielded only during the first growth stage. The examina-
tion of FC-CVD in this work revealed a two-stage growth pro-
cess, which is unique to floating-catalyst synthesis. With thin
film catalysts, an initial rapid growth slows as the catalysts
become poisoned. This process can occur relatively quickly
(<1 h), and no substantial changes to nanotube lengths or
yields are observed beyond this point [45,46]. The continuous
addition of catalyst material in FC-CVD allows extended reac-
tion durations, and in this study a second growth stage was
identified (beyond some initial growth at a constant rate). In
this stage the nanotube mat stops lengthening (and actually
shrinks as products are etched away); however, its mass con-
tinues to increase. This process has not been reported previ-
ously, and differs from the catalyst poisoning found in thin
film catalyst growth. These results suggest that synthesis
durations for FC-CVD growth can be optimized to maximize
nanotube mat height and yield.
2093
Substrate choice has been newly shown to have significant
impact upon CNT diameters grown by FC-CVD, with the rela-
tive effect predictable through the catalyst’s mobility upon
the substrate’s surface. In the case of alumina, it is proposed
that catalyst mobility is limited by ionic substrate/catalyst or
substrate/catalyst-oxide interactions. These results should al-
low for improved control in the FC-CVD synthesis of CNT’s.
Acknowledgements
The authors acknowledge Dr. Ron Gronsky and Dr. Chris Ku-
mai of UC Berkeley, and Dr. Frank Talke and Ralf Brunner of
UC San Diego CMRR for assistance accessing various pieces
of analytical equipment.
Appendix A. Supplementary information
Supplementary data associated with this article can be found,
in the online version, at doi:10.1016/j.carbon.2009.03.060.
R E F E R E N C E S
[1] Sveningsson M, Lee SW, Park YW, Campbell EEB. Field
emission from multi-walled carbon nanotubes – its
applications in NEMS. AIP Conference 2005:620–3.
[2] Choi WB, Chung DS, Kang JH, Kim HY, Jin YW, Han IT, et al.
Fully sealed, high-brightness carbon-nanotube field-
emission display. Appl Phys Lett 1999;75(20):3129–31.
[3] de Heer WA, Chatelain A, Ugarte D. A carbon nanotube field-
emission electron source. Science 1995;270(5239):1179–80
(Washington, D.C.).
[4] Bandaru PR, Daraio C, Jin S, Rao AM. Novel electrical
switching behaviour and logic in carbon nanotube Y-
junctions. Nat Mater 2005;4(9):663–6.
[5] Li H, Zhang Q, Li J. Carbon-nanotube-based single-electron/
hole transistors. Appl Phys Lett 2006;88(1):013508/1–3.
[6] Johnston DE, Islam MF, Yodh AG, Johnson AT. Electronic
devices based on purified carbon nanotubes grown by high-
pressure decomposition of carbon monoxide. Nat Mater
2005;4(8):589–92.
[7] Kireitseu MV, Bochkareva LV, Hui D. Rheological behavior of
CNT-reinforced damping materials. Ann Trans Nordic Rheol
Soc 2006;14:219–25.
[8] Hernandez CD, Zhang M, Fang S, Baughman RH, Gates TS,
Kaing SK. Multifunctional characteristics of carbon nanotube
(CNT) yarn composites. In: Proceedings of the
multifunctional nanocomposites international conference,
Honolulu, HI, United States. 2006. p. mn2006 17028/1–
mn2006/5.
[9] Kuo C-S, Bai A, Huang C-M, Li Y-Y, Hu C-C, Chen C-C.
Diameter control of multiwalled carbon nanotubes using
experimental strategies. Carbon 2005;43(13):2760–8.
[10] Kim L, Lee E-M, Cho S-J, Suh JS. Diameter control of carbon
nanotubes by changing the concentration of catalytic metal
ion solutions. Carbon 2005;43(7):1453–9.
[11] Chai S-P, Zein SHS, Mohamed AR. The effect of catalyst
calcination temperature on the diameter of carbon
nanotubes synthesized by the decomposition of methane.
Carbon 2007;45(7):1535–41.
[12] Endo M, Takeuchi K, Kobori K, Takahashi K, Kroto HW, Sarkar
A. Pyrolytic carbon nanotubes from vapor-grown carbon
fibers. Carbon 1995;33(7):873–81.
2094 CARBON 4 7 ( 2 0 0 9 ) 2 0 8 5 –2 0 9 4
[13] Serp P, Figueiredo JL. A microstructural investigation of
vapor-grown carbon fibers. Carbon 1996;34(11):1452–4.
[14] Dai H. Nanotube growth and characterization. In:
Dresselhaus MS, Dresselhaus G, Avouris P, editors. Carbon
nanotubes: synthesis, structure, properties, and applications,
vol. 80. New York: Springer-Verlag. p. 29–53.
[15] Dresselhaus MS, Endo M. Relation of carbon nanotubes to
other carbon materials. In: Dresselhaus MS, Dresselhaus G,
Avouris P, editors. Carbon nanotubes: synthesis, structure,
properties, and applications, vol. 80. New York: Springer-
Verlag. p. 11–28.
[16] Dresselhaus MS, Dresselhaus G, Sugihara K, Spain IL,
Goldberg HA. Graphite fibers and filaments. New
York: Springer-Verlag; 1988. p. 382.
[17] Yamada T, Namai T, Hata K, Futaba DN, Mizuno K, Fan J, et al.
Size-selective growth of double-walled carbon nanotube
forests from engineered iron catalysts. Nat Nanotechnol
2006;1:131–5.
[18] Bai S, Li F, Yang Q-H, Cheng H-M, Bai J. Influence of ferrocene/
benzene mole ratio on the synthesis of carbon
nanostructures. Chem Phys Lett 2003;376(1–2):83–9.
[19] Sinnott SB, Andrews R, Qian D, Rao AM, Mao Z, Dickey EC,
et al. Model of carbon nanotube growth through chemical
vapor deposition. Chem Phys Lett 1999;315:25–30.
[20] Singh C, Shaffer MSP, Windle AH. Production of controlled
architectures of aligned carbon nanotubes by an injection
chemical vapour deposition method. Carbon
2003;41(2):359–68.
[21] Lupo F, Rodrı́guez-Manzo JA, Zamudio A, Elı́as AL, Kim YA,
Hayashi T, et al. Pyrolytic synthesis of long strands of large
diameter single-walled carbon nanotubes at atmospheric
pressure in the absence of sulphur and hydrogen. Chem Phys
Lett 2005;410:384–90.
[22] Aguilar-Elguézabal A, Antúnez W, Alonso G, Paraguay-
Delgado F, Espinosa F, Miki-Yoshida M. Study of carbon
nanotubes synthesis by spray pyrolysis and model of growth.
Diam Relat Mater 2006;15:1329–35.
[23] Chaisitsak S, Nukeaw J, Tuantranont A. Parametric study of
atmospheric-pressure single-walled carbon nanotubes
growth by ferrocene–ethanol mist CVD. Diam Relat Mater
2007;16:1958–66.
[24] Vander Wal RL, Ticich TM, Curtis VE. Substrate-support
interactions in metal-catalyzed carbon nanofiber growth.
Carbon 2001;39(15):2277–89.
[25] Bom D, Andrews R, Jacques D, Anthony J, Chen B, Meier MS,
et al. Thermogravimetric analysis of the oxidation of
multiwalled carbon nanotubes: evidence for the role of defect
sites in carbon nanotube chemistry. Nano Lett
2002;2(6):615–9.
[26] Pang LSK, Saxby JD, Chatfield SP. Thermogravimetric analysis
of carbon nanotubes and nanoparticles. J Phys Chem
1993;97(27):6941–2.
[27] Saxby JD, Chatfield SP, Palmisano AJ, Vassallo AM, Wilson MA,
Pang LSK. Thermogravimetric analysis of
buckminsterfullerene and related materials in air. J Phys
Chem 1992;96(1):17–8.
[28] Bai S, Li F, Yang Q, Cheng H-M, Bai J. Influence of ferrocene/
benzene mole ratio on the synthesis of carbon
nanostructures. Chem Phys Lett 2003;376(1, 2):83–9.
[29] Sinnott SB, Andrews R, Qian D, Rao AM, Mao Z, Dickey EC,
et al. Model of carbon nanotube growth through chemical
vapor deposition. Chem Phys Lett 1999;315(1):25–30.
[30] Dresselhaus MS, Dresselhaus G, Jorio A, Souza Filho AG, Saito
R. Raman spectroscopy on isolated single wall carbon
nanotubes. Carbon 2002;40(12):2043–61.
[31] Bacsa WS, Ugarte D, Chatelain A, de Heer WA. High-
resolution electron microscopy and inelastic light scattering
of purified multishelled carbon nanotubes. Phys Rev B
1994;50(20):15473–6.
[32] Tohji K, Goto T, Takahashi H, Shinoda Y, Shimizu N,
Jeyadevan B, et al. Purifying single-walled nanotubes. Nature
1996;383(6602):679.
[33] Maultzsch J, Reich S, Thomsen C, Webster S, Czerw R, Carroll
DL, et al. Raman characterization of boron-doped
multiwalled carbon nanotubes. Appl Phys Lett
2002;81(14):2647–9.
[34] Kelemen SR. The reaction of carbon and water as catalyzed
by a nickel surface. Appl Surf Sci 1987;28(4):439–74.
[35] McKee GSB, Vecchio KS. Thermogravimetric analysis of
synthesis variation effects on CVD generated
multiwalled carbon nanotubes. J Phys Chem B
2006;110(3):1179–86.
[36] Pan ZJ, Yang RT. Catalytic behavior of transition metal oxide
in graphite gasification by oxygen, water, and carbon dioxide.
J Catal 1991;130(1):161–72.
[37] Thomas JM. Microscopic studies of graphite oxidation. In:
Walker PL, editor. Chemistry and physics of carbon. New
York: Marcel Dekker, Inc.; 1966. p. 121–202.
[38] Shimada T, Yanase H, Morishita K, Hayashi J-I, Chiba T.
Points of onset of gasification in a multi-walled carbon
nanotube having an imperfect structure. Carbon 2004;42(8–
9):1635–9.
[39] Wirth CT, Hofmann S, Robertson J, Mattevi C, Cepek C,
Goldoni A, et al. In situ time-resolved XPS study of catalyst
behavior during carbon nanotube growth. In: Proceedings of
the 2008 MRS spring meeting, San Francisco, CA, USA. 2008.
[40] Lodziana Z, Topsoe N-Y, Norskov JK. A negative surface
energy for alumina. Nat Mater 2004;3(5):289–93.
[41] Shchipalov YK. Surface energy of crystalline and vitreous
silica. Glass Ceram (Trans Steklo i Keramika) 2000;57(11–
12):374–7.
[42] Liu H, Zhang Y, Arato D, Li R, Merel P, Sun X. Aligned multi-
walled carbon nanotubes on different substrates by floating
catalyst chemical vapor deposition: critical effects of buffer
layer. Surf Coat Technol 2008;202(17):4114–20.
[43] Sakata K, Honma K, Ogawa K, Watanabe O, Nii K. Interface
chemistry and bonding strength for diffusion-bonded Fe/Fe–
FeO/Al2O3 systems. J Mater Sci 1986;21(12):4463–7.
[44] Papirer E, Ligner G, Balard H, Vidal A, Mauss F. Surface energy
of silicas and c-alumina modified by heat treatment. Chem
Mod Surf 1990;3:15–26.
[45] Li Q, Zhang X, DePaula RF, Zheng L, Zhao Y, Stan L, et al.
Sustained growth of ultralong carbon nanotube arrays for
fiber spinning. Adv Mater 2006;18:3160–3.
[46] Puretzky AA, Geohegan DB, Jesse S, Ivanov IN, Eres G. In situ
measurements and modeling of carbon nanotube array
growth kinetics during chemical vapor deposition. Appl Phys
A 2005;81:223–40.