Generation of carbon nanofilaments on carbon fibers at 550°C

C A R B O N 4 7 ( 2 0 0 9 ) 3 0 7 1 – 3 0 7 8

. sc iencedi rec t . com

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

Generation of carbon nanofilaments on carbon fibers at 550 �C

Claudia C. Luhrsa,*, Daniel Garciaa, Mehran Tehrania, Marwan Al-Haika,Mahmoud Reda Tahab, Jonathan Phillipsa,c

aDepartment of Mechanical Engineering, University of New Mexico, Albuquerque, NM 87131, United StatesbDepartment of Civil Engineering, University of New Mexico, Albuquerque, NM 87131, United StatescLos Alamos National Laboratories, Los Alamos, NM 87544, United States

A R T I C L E I N F O

Article history:

Received 13 May 2009

Accepted 6 July 2009

Available online 9 July 2009

0008-6223/$ - see front matter � 2009 Elsevidoi:10.1016/j.carbon.2009.07.019

* Corresponding author: Fax: +1 505 277 1571E-mail address: [email protected] (C.C. L

A B S T R A C T

Employing a relatively new method, in which carbon structures are grown from fuel rich

combustion mixtures using palladium particles as catalyst, multi-scale diameter nanome-

ter – micrometer filament structures were grown from ethylene/oxygen mixtures at 550 �Con commercial PAN micrometer carbon fibers. The filaments formed had a diameter

roughly equal to the palladium particle size. At sufficiently high metal loadings

(>0.05 wt.%) a bimodal catalyst size distribution formed, hence a bimodal filament size dis-

tribution was generated. Relative short, densely spaced nanofilaments (ca. 10 nm diame-

ter), and a slightly less dense layer of larger (ca. 100 nm diameter) faster growing fibers

(ca. 10 lm/h) were found to exist together to create a unique multi-scale structure. A

protocol was developed such that only nano-scale fibers or a mixture of nano and sub-

micron fibers could be produced. No large range order was evident in the filaments. This

work demonstrates a unique ability to create a truly ’multi-scale’ carbon structure on the

surface of carbon fibers. This fiber structure potentially can enhance composite material

strength, ductility and energy absorption characteristics.

� 2009 Elsevier Ltd. All rights reserved.

1. Introduction

Composites (e.g. fiber reinforced polymers) failure under

high-energy loading, i.e. impact, can be traced to the limited

bond strength between the matrix and the fibers [1–3]. Mul-

ti-scale fiber systems that include high surface area ‘nano’

component clearly will have increased surface area, hence

possibly increased shear strength. The most common ap-

proach to creating a multi-scale system is simply to physically

mix carbon nanotubes into a more traditional composite con-

sisting of epoxy with embedded microscale fibers. The inclu-

sion of carbon nanotubes (CNTs) clearly toughens different

matrices [4,5]. Depositing CNTs in a brittle matrix increases

stiffness by orders of magnitude [6]. This approach to create

multi-scale composites is limited due to the difficulty of dis-

persing significant amounts of nanotubes [7,8] and it has

er Ltd. All rights reserved

.uhrs).

repeatedly been reported that phase separation occurs above

relatively low weight percent loading (ca. 3%) due to the

strong van der Waals forces between the CNTs compared with

that between the CNTs and the polymer matrix. Hence, the

nanotubes tend to segregate and form inclusions.

One means to prevent nanotubes or nanofilaments

agglomeration is to anchor one end of the nanostructure,

thereby creating a stable multi-scale structure. This is most

readily done by literally growing the CNTs directly on micron

scale fibers. Recently, CNT were grown on carbon fibers, both

polyacrylonitrile- (PAN-) and pitch-based, by hot filament

chemical vapor deposition (HFCVD) using H2 and CH4 as pre-

cursors. Nickel clusters were electrodeposited on the fiber

surfaces to catalyze the growth and uniform CNTs coatings

were obtained on both the PAN- and pitch-based carbon

fibers. Multi-walled CNTs with smooth walls and low

.

3072 C A R B O N 4 7 ( 2 0 0 9 ) 3 0 7 1 – 3 0 7 8

impurity content were grown [9]. Carbon nanofibers were also

grown on a carbon fiber cloth using plasma enhanced chem-

ical vapor deposition (PECVD) from a mixture of acetylene and

ammonia. In this case, a cobalt colloid was used to achieve a

good coverage of nanofibers on carbon fibers in the cloth [10].

Caveats to CNT growth include damage in the carbon fiber

surface due to high-temperatures (>800 �C) [9,11]. Qu et al.

[12] reported a new method for uniform deposition of CNTs

on carbon fibers. However, this method requires processing

at 1100 �C in the presence of oxygen and such high tempera-

ture is anticipated to deepen the damage in the carbon fibers.

In the present work, multi-scale filaments (herein, linear

carbon structures with multi-micron diameter are called

‘fibers’, all structures with sub-micron diameter are called ‘fil-

aments’) were created with a low-temperature (ca. 550 �C)

alternative to CVD growth of CNTs. Specifically, nano-scale fil-

aments were rapidly generated (>10 lm/h) on commercial

microscale fibers via catalytic (Pd particles) growth from a fuel

rich combustion environment at atmospheric pressure. This

atmospheric pressure process [13,14], derived from the pro-

cess called Graphitic Structures by Design (GSD), is rapid, the

temperature low enough (ca. 550 �C) to avoid structural dam-

age and the process is inexpensive and readily scalable. In

some cases, a significant and unexpected aspect of the process

was the generation of unique ‘three scale’ materials. That is,

materials with these three size characteristics were produced:

(i) micrometer scale commercial carbon PAN fibers, (ii) a layer

of ‘long’ sub-micrometer diameter scale carbon filaments, and

(iii) a dense layer of ‘short’ nanometer diameter filaments.

A similar mean to generate carbon fibers on a parent struc-

ture was published by Downs et al., who used hydrocarbon/H2

mixtures produce C filaments on commercial PAN fibers at

650 �C although employing diverse Cu–Ni alloys as catalyst

[15,16]. Such work relies on the use of a different catalyst,

reducing atmospheres and slightly higher temperatures than

the ones reported here. References to analogous processes

using diverse catalysts as iron, nickel, platinum and other al-

loys have also been published by Baker et al., Bokx et al.,

Owens et al. and Rodriguez et al. [17–20]. Terrones et al. [21]

reported on the generation of graphitic cones using Pd as a

catalyst although in such case Ar atmospheres and higher

temperatures (ca. 850–1000 �C) were employed to generate

more than fibers, open edge conical structures. Reactive gases

mixtures of different nature (other hydrocarbons or reducing

atmospheres) have been studied as well [22,23].

Our research group recently published the basic process to

generate fibers and thin films using fuel rich ethylene–oxygen

mixtures using Pd as a catalyst, that included a discussion of

the mechanisms dominating the carbon growth [24]. The pres-

ent work represent a next step in the study of this system in

which the same process aids the generation of carbon compos-

ites containing fibers of multiple size scales with the objective

of improving the material energy absorption characteristics.

2. Experimental methods

2.1. Sample preparation

The process employed for the nanofilament synthesis in-

volves three main steps; a sizing burn-off/activation process

of the commercial filaments, wet impregnation to create well

dispersed nano-scale metal catalyst particles and actual

nano-scale filament growth via exposure to a low-tempera-

ture fuel rich combustion environment.

2.1.1. Sizing removal and activationCommercial PAN carbon fibers (�7.5 lm diameter, Toho Tenax

America, Inc.) were used as the microscale support (substrate)

on which the carbon nanostructures were grown. Fibers were

chopped into 3 and 6 mm lengths. In order to remove the siz-

ing, the carbon fibers were treated at 525 �C in O2 (100 sccm

flow) for 10 min at the center of a controlled atmosphere

quartz tube (1 in. diameter) centered inside a 24 in. Lindburgh

tube furnace. The treated fibers were removed, rinsed in ethyl

alcohol and dried in air at 100 �C for 1 h. In order to ‘activate’

the carbon, the fibers were then ‘burned’ (O2, 100 sccm,

525 �C, 20 min). These two processes resulted in a net weight

loss between 10% and 15%.

2.1.2. Metal loadingThe catalyst was prepared by incipient wetness impregnation

using Pd (NO3)2Æ6H2O (99.9%, Sigma–Aldrich) as precursor with

the pretreated micron carbon fibers as support. The amount

of precursor salt, Pd(II) nitrate hydrate, required for a final

metal loading of 0.05, 0.5, 1, 5, 10 and 100 wt.% was dissolved

in de-ionized water. Once the carbon fibers were impregnated

by the solution (just enough to ‘wet’ all fiber surfaces) they

were left to dry for 12 h in air at 100 �C.

2.1.3. Filament generationAfter the drying step, fibers were placed in a sintered alumina

combustion boat and treated in the same tube furnace appa-

ratus used during the sizing removal. To assure that the cata-

lyst precursor salts were decomposed and the metal was in a

reduced state, the fibers were treated in three stages. These

processes, and all other processes, were conducted at ambi-

ent pressure (no pump). The first step required calcining in in-

ert gas at �150 sccm at 250 �C for 4 h. In the second step, the

temperature was increased to 550 �C and a reducing gas mix-

ture, 93%Ar/7%H2 introduced for 1 h. Finally, at the same tem-

perature, the system was flushed with N2 gas (600 sccm) for

1 h. This completed the pretreatments intended to produce

highly dispersed and reduced Pd particles on the fiber. The

‘GSD process’ (carbon deposition from a fuel rich combustion

mixture [13]), was initiated immediately after the reduction

process. A (fuel rich) mixture of N2 (300 sccm), oxygen

(15 sccm) and ethylene (15 sccm) was introduced while main-

taining the temperature at 550 �C. This last step was per-

formed at five different growth duration times: 1, 5, 35, 90

and 270 min for different specimens. Once the GSD process

concluded, the heating elements were turned off and the

reactor was flushed with nitrogen and then cooled. The initial

flow (�2 min) was high (1000 sccm for 2 min), but reduced to

low (50 sccm of N2) for most of the cooling process.

2.2. Sample characterization

A transmission electron microscope (TEM, JEOL 2010)

equipped with Energy Dispersive X-ray (EDX) analysis, and a

scanning electron microscope (SEM, 5200 Hitachi) were used

C A R B O N 4 7 ( 2 0 0 9 ) 3 0 7 1 – 3 0 7 8 3073

to examine the samples. A Philips powder X-ray diffractome-

ter using Cu Ka radiation was used to study crystalline struc-

tures of samples. A Netzsch STA 409 PC Luxx (TGA/DSC) was

employed to perform temperature-programmed oxidation

(TPO) of the samples. Raman spectra were obtained with a

confocal Raman microscope, using 1 mW of 514.5 nm excita-

tion light focused on the sample with a 100·, 0.9 N.A objec-

tive. Scattered light was collected in a 180� backscattering

geometry and dispersed in an f/1.8 spectrograph onto a CCD

detector. Spectra were typically obtained as the sum of ten

30 s integrations.

3. Results and discussion

A brief description of the sample following each treatment is

provided below. The burn-off process created ‘active sites’

which served as nucleation centers for metal catalyst particle

formation. From each of these particles a carbon filament

with a low degree of graphitization grew. As described in de-

tail later, the filament diameters were approximately the

same as that of the particles, generally found at the growing

end of the fiber, that catalyzed growth. Hence the distribution

of filament diameters reflected the distribution of diameters

of the catalyst particles. Indeed, for very low metal loadings

(ca. 0.05%) only filaments with diameters averaging around

10 nm grew, suggesting only nano-scale catalyst particles

were present. At higher metal loadings, bimodal fiber distri-

butions were found, suggesting a bimodal size distribution

of catalyst particles were present. Specifically at the higher

loadings, in addition to fibers of diameter 10 nm, filaments

of a much larger diameter (order of 100 nm, or ‘sub-micron’),

that grew more quickly, were found. This suggests that above

‘saturation’ a bimodal distribution of metal catalyst particles

were present; nano-scale particles and sub-micron particles.

The distinct change in structure at about 0.05 wt.% metal

loading, led to the designation of this loading level as ‘satura-

tion’ loading. In the sense employed herein, saturation implies

a level of metal loading that equals the number of chemically

active sites on the fiber surface. Apparently, on fibers with a

metal loading greater than saturation, two scales of metal

catalyst particles, nanometer and sub-micrometer formed.

At loadings below saturation, only nanometer scale metal

catalyst particles formed. In all cases, the filaments appar-

ently grew from radicals generated in the combustion

environment, with a diameter roughly equal to the size of

the catalytic particle. The net result is that for samples with

metal loadings greater than ‘saturation’, the newly grown ‘fila-

ments’ actually constituted a multi-scale system: distinct

populations of nanometer and sub-micrometer sized

filaments.

Fig. 1 – Carbon fiber after sizing removal and activation. (A)

After initial quick burn and rinse, only fragments of sizing

remain. (B) After oxygen ‘activation’, the surface is nearly

identical to that after sizing removal but slightly roughened

with fiber diameter measurably reduced (ca. 5%).

3.1. Sizing removal and burn-off

SEM observation of the samples revealed the sizing removal

step completely removed the sizing and the burn-off con-

sumed carbon evenly in the axial direction (Fig. 1). There

was no apparent change in the surface morphology as a re-

sult of burn-off indicating no apparent damage to the car-

bon surface. The average diameter for fibers before burn-

off was 7.4 lm, but burn-off yielded a net weight loss close

to 10% and reduced the average diameter by a small

amount.

3.2. Multi-scale structure

Comparison between SEM (Fig. 2) and TEM (Fig. 3) of samples

with metal loading beyond the saturation point (more below),

and treated in the GSD process (550 �C, 1:1 O2:C2H4, 35 min),

provides clear visual evidence of a complex ‘two layer’ and

‘multi-scale’ filament growth pattern. It is important to note

in examining the SEM and TEM pictures that attention to only

one type of microscopy leads to a misunderstanding of the

structure. To wit, the filament structure is composed of ‘slow’

growing filaments with a mean diameter of about 5 nm, and

‘fast’ growing filaments that have a less dense layer of larger

diameter (ca. 50–150 nm) compared to the first group of fila-

ments. A helpful analogy: there is a layer of ‘trees’ clearly

not tightly packed, and below that, at ground level, a dense

layer of ‘grass’.

Several features of the growth process are worth noting.

First, there are two scales of filaments, nano-scale (defined

Fig. 2 – SEM of sub-micron filaments. These samples were

treated in a 1:1 mixture of O2:C2H4 for 90 min at 550 �C. (A) A

‘perspective’ view of several filaments that were pretreated

PAN fibers. (B) A closer view reveals that the filaments are

generally less than 100 nm in diameter, are not straight and

tangle together to form mats. (C) Note that there appear to be

metal particles at the ‘head’ of many of the filaments.

Fig. 3 – TEM of nano-scale filaments. (A) Relatively dense

layer of filaments with an average diameter of ca. 10 nm

forms on the surface of the fibers. The average height of this

layer is less than 200 nm. (B) The Pd particles that catalyze

filament growth can clearly be seen. (C) The dramatic

difference in size of the two types of filaments can be

observed in the figure. This sample was treated under GSD

conditions for only 35 min. The metal loading weight was

0.5%.

3074 C A R B O N 4 7 ( 2 0 0 9 ) 3 0 7 1 – 3 0 7 8

as having a diameter less than 100 nm), although the distribu-

tion of nanofilaments is ‘bimodal’ in that invariably some far

larger filaments, associated with large catalyst particles, are

present. Second, the morphology of the sub-micron filaments

is not uniform. Some clearly have smooth surfaces, whereas

others grow in tight spirals or are kinked. Third, once beyond

a couple of microns in length, the sub-micron filaments are

never straight, and hence interweave to form a tangled mat.

Fourth, the filaments are associated with the metal catalyst

particles that are ‘lifted’ off the surface by the growing fiber.

C A R B O N 4 7 ( 2 0 0 9 ) 3 0 7 1 – 3 0 7 8 3075

Fifth, the ‘net’ thickness of the filament layer grew at a rate of

about 10 lm/h. A best estimate is that the individual fibers in

this layer grew twice as fast. Sixth, this type of growth was al-

ways observed for metal loading greater than about 0.05%.

However, in the event that the fiber activation step was not in-

cluded in the pretreatment, only ‘large’ diameter fibers

formed and these were found to form in poorly dispersed

‘groups’. Seventh, control studies showed that a fuel rich mix-

ture was required for filament growth. In the absence of oxy-

gen, no fibers were detected. Also, for completeness, growth

in the absence of metal was studied and there was no evi-

dence of fiber growth in this environment. Eighth, at higher

metal loadings, up to and including 100%, the structure of

the filaments was similar to that described in detail for the

0.5% metal loaded sample.

3.3. Saturation loading

In several cases metal loading was kept to relatively very low

levels. In these cases the observed morphology was much

Fig. 4 – SEM of saturation loaded sample. To allow growth

sufficient to observe with SEM, sample treated using

standard protocol, but with 270 min of exposure to GSD

conditions. (A) Clearly the fibers are not covered with sub-

micron filaments. Contrast with Fig. 2. (B) The surface

(compare with Fig. 1) is virtually completely covered with

nanofilaments. (TEM images not shown here but are similar

to those shown in Fig. 3.)

simpler (Fig. 4). A sample loaded with only 0.05% metal and

treated to the standard GSD treatment (550 �C, 1:1 O2:C2H4)

but for an extended time, 270 min, grew primarily the ‘grass’

layer, that is the filaments have a mean diameter of about

5 nm and the net length <0.5 lm is consistent with ‘slow’

growth. Some of the larger fibers are also present, but clearly

far fewer than observed at metal weight loadings of 0.5% and

higher. TEM images not shown here but are similar to those

shown in Fig. 3a and b.

3.4. Characterization

Every effort was made to characterize the structure of the fil-

aments, in particular their degree of graphitization. The

methods employed were X-ray diffraction, TEM, TPO and Ra-

man spectroscopy. XRD analysis revealed only a weak and

rather broad peak for all samples near 25.5� 2h indicating an

amorphous or turbostratic structure. TPO analysis included

examining the following four samples:

3.4.1. Sample 1 (carbon fibers (sizing removed))Commercially purchased PAN-based carbon fiber (Toho Te-

nax, Inc.) treated in a tubular furnace at 525 �C in O2 (100 sccm

flow) for 10 min. The treated fibers were removed, rinsed in

ethyl alcohol on a sieve and dried in air at 100 �C for 1 h.

3.4.2. Sample 2 (graphite)Graphite flakes that are 7–10 lm size and 99% metals basis.

3.4.3. Sample 3 (amorphous carbon)Carbon from pyrolyzed sugar made by heating commercially

purchased sucrose to 1000 �C.

3.4.4. Sample 4 (carbon fibers with filaments – Pd removed)Prepared by standard process with a growth time of 90 min

(time the sample was exposed to the fuel mixture). The sam-

ple was placed into a vial and aqua regia (�2 cc) was added.

Sample was in vial with aqua regia for 34 days. After 34 days,

about of the aqua regia was removed from the vial via a drop-

per into a waste container. Then new aqua regia was added

via a dropper. This procedure was repeated twice. After

1 day, all the aqua regia was removed (via a dropper again)

from the vial with the sample. Then the vial with the sample

in it was filled with distilled water. After this step, the water

along with the fibers was poured onto a funnel to filter the

water. With the sample on the funnel, it was rinsed 2 more

times with distilled water. Then the sample was put into a

combustion boat (covered in aluminum foil with holes

punched in) and allowed to dry overnight in air underneath

a fume hood.

Unfortunately, in each case, most of the signal came from

the original fiber. This signal tended to ‘wash-out’ any signal

from the filament growth. For example, (Fig. 5A) despite re-

peated efforts to find an ‘ideal’ protocol, the TPO only pro-

duced results complex to interpret with certainty. It was

anticipated that the most graphitic samples would burn at

the highest temperature. Yet, it was always observed that

PAN fibers, clearly not graphitic, always burned more slowly

than a pure graphite sample. It was anticipated that the most

3076 C A R B O N 4 7 ( 2 0 0 9 ) 3 0 7 1 – 3 0 7 8

amorphous sample would start to burn first. Yet, the sample

made of pyrolyzed sugar started to burn at temperatures

higher than the PAN fibers and other samples.

In a similar fashion, the Raman data was inconclusive (not

shown). Pure PAN fibers with sizing removed and fibers with

filaments, grown at both saturation and 10· saturation metal

loadings, produced indistinguishable Raman spectra. More-

over, in all cases, the G (graphite) and D (disorder) peaks were

of nearly identical relative magnitude. The only method that

appeared to yield a distinct signal was TEM. As shown in

Fig. 5B, all TEM data are consistent with the interpretation

that most of the filament growth is amorphous in character.

Carbon filaments on fiber structures were successfully

generated by a relatively low-temperature and atmospheric

pressure approach. The nature of the filaments formed were

a function of the catalytic (Pd) metal loading. At sufficiently

high metal loadings (>0.05%) multi-scale filament structures

formed in which sub-micron (ca. 100 nm diameter) filaments

grew rapidly over a layer of nano-scale (ca. 10 nm) slow grow-

ing filaments. At sufficiently low loadings (�0.05%), the sur-

face was nearly completely covered with nanofilaments and

very few sub-micron filaments formed. In all cases the fila-

ments were clearly poorly crystallized.

All filaments grew from Pd catalyst particles deposited on

the commercial PAN fiber surfaces. It is presumed, given the

low-temperature employed and the phenomenological simi-

larity with earlier studies [13,14], that the source of carbon

for filament growth was radicals generated homogeneously

(e.g. CH2). These react on the metal catalyst particles, decom-

posing readily to deposit carbon atoms (e.g. CH2! C + H2). It is

also presumed that the formation of filaments from the

deposited carbon atoms is similar to that described repeat-

Fig. 5 – (A) TPO Study – A major effort was undertaken to distingu

fast they combust. The results were not readily explained. For e

slowly, the PAN fibers with sizing removed (Sample 1) burned m

very slow burn rates. Repeatedly the samples with filaments (aft

failed to completely combust. Even the results for amorphous c

surprising. Combustion for this sample initiated at a higher tem

Nanofilament Structure. This TEM image is exemplary of many t

(ii) a catalytic metal particle is found at the tip.

edly in the literature [17,25–28]. Carbon atoms deposited at

the catalyst particle surface, driven by a chemical potential

gradient, transport through the catalyst particle, or around

its periphery, to ‘add’ to the solid carbon forming behind

the particle. In fact, this same mechanism, the so-called ‘root’

mechanism, is used to explain the growth of carbon nano-

tubes [29–32].

It has been reported that C@O, –OH and C–O–C groups in

the carbon surface change its polarity and may also lead to

reduction in stabilization period for growing carbon fibers

[33–35]. We believe that the role of oxygen relates to the pre-

viously mentioned homogeneous generation of radicals. It is

these radicals, rather than the parent hydrocarbon species

(ethylene), from which the fibers grow. In essence, the homo-

geneously formed radicals (e.g. CH, CH2) are chemically iden-

tical to the fragments presumed to form when hydrocarbons

decompose on a catalytic surface, the generally understood

first step in filament growth. Indeed, the role of radicals gen-

erated homogeneously from ethylene/oxygen mixtures has

repeatedly been shown to be remarkable. In fuel lean condi-

tions at 550 �C the radicals are known to etch platinum [36–

40]. In fuel rich conditions, such as those employed in this

study, graphitic structures form with remarkable rapidity on

platinum [41]. Thus, the notable rate of growth observed here,

only in fuel rich combustion conditions, is entirely consistent

with that observed in earlier work with both platinum and

nickel [13].

It was demonstrated that filament size can be controlled.

First, it was established that the filament size ‘matched’ the

size of the catalyst particle. Thus, control is achieved by

adjusting the incipient wetness protocol to select Pd particle

size. It was found that ‘activating’ the PAN surface by lim-

ish the relative ‘degree of graphitization’ on the basis of how

xample, although graphite is anticipated to burn the most

ore slowly than a pure graphite sample (Sample 2), even at

er Pd removal, Sample 4, by dissolution in aqua regia) always

arbon formed by the pyrolysis of sugar (Sample 3) are

perature then observed for the PAN fibers (Sample 1) (B)

hat show: (i) the nanofilaments are not well crystallized, and

C A R B O N 4 7 ( 2 0 0 9 ) 3 0 7 1 – 3 0 7 8 3077

ited combustion decreases the size of the filaments and in-

creases the homogeneity of the filament distribution for any

specific metal loading. Increasing carbon surface heteroge-

neity has long been empirically demonstrated to decrease

average metal particle size formed by wet impregnation

methods [42–45].

Work with graphite clearly shows that metal particles

will only form at steps and other defects, suggesting that

‘activation’ increases dispersion by increasing the number

of defects [46]. It is also clear that activation increases the

number of ‘primary adsorption sites’ on carbon, creating a

hydrophilic surface and leading to smaller water ‘drop’ size

during incipient wetness impregnation [47]. That is, the

more primary adsorption sites exist, the less water per site,

assuming constant water. Since the final metal particle size

is a function of the concentration of metal precursor in the

aqueous solution and the drop size, this could also partially

explain the empirical relationship between extent of ‘activa-

tion’ and degree of dispersion. Moreover, the work con-

ducted herein is consistent with the prior empirical work.

Indeed, in the absence of activation only a few large catalyst

particles were present, whereas on activated surfaces, at

low (sub-saturation) loadings, only very small particles were

present (Fig. 4).

The findings that metal loadings above an active surface

site ‘saturation’ level form on both sub-micron and nanofila-

ments are consistent with the ‘activation creates greater dis-

persion’ hypothesis, with one caveat. That is, it should be

postulated that not only does activation create hydrophilic

sites to attract liquid drops, it also forms ‘anchoring’ or bind-

ing sites for metal atoms. Further, during calcination, metal

released when the parent salt decomposes must be chemically

bound to these surface ‘active sites’ explaining the mecha-

nism that prevents subsequent sintering. However, once these

sites are ‘saturated’, additional metal on the surface is not

bound, and perhaps surprisingly does not add to the metal

particles strongly bound to active sites. Rather, it is postulated

this metal is ‘free metal’ presumably rapidly diffusing and sin-

tering by agglomeration with other ‘free’ metal. Hence, above

‘saturation’ loadings bimodal particle size distributions form:

anchored metal forms small particles, and free metal sinters

to form large particles. There is certainly ample evidence in

the literature of metal very weakly anchored to graphitic sur-

faces, leading to rapid diffusion and particle growth, per the

so-called ‘Weak Metal Support Interaction’ (WMSI) between

non-activated carbon and metal particles [48]. This in turn

leads to a bimodal filament size distribution.

During filament growth, however, even the ‘anchored’

metal appears to break free of the surface. It does not sinter

to form larger catalyst particles, but rather is ‘lifted’ off the

fiber surface by a growing filament. Characterization of the

samples confirmed a tip growth model in which the Pd

particles are carried away from the fibers’ surfaces and

constitute the main front for the sub-micron and nanofila-

ment generation. Characterization methods, particularly

TEM, suggest that the filaments are not crystalline. That

is, only short range order is evident, consistent with the

growth of turbostratic carbon. In contrast, on nickel under

the same conditions, graphitic carbon growth has been

found [13].

4. Conclusions

Multi-scale carbon structures grew catalytically from palla-

dium particle impregnated PAN fibers exposed to ethylene/

oxygen mixtures at 550 �C. Above ‘saturation loading’

(approximately 0.05% on activated PAN fibers) metal particles

were present in a bimodal distribution of nanometer (ca.

10 nm) and submicrons (ca. 100 nm) sizes. As filament size

roughly corresponds to catalyst particle size, this led to the

formation of a bimodal size distribution of filaments. Rela-

tively short, densely spaced nanofilaments (ca. 10 nm diame-

ter), and a slightly less dense layer of larger (ca. 100 nm

diameter) faster growing fibers (ca. 10 lm/h) were found to

co-exist to create a unique multi-scale structure. At metal

loadings less than saturation only the nanometer sized palla-

dium particles were present, hence only nanometer scale fil-

aments formed. All analytical techniques employed indicated

poor crystallinity of the filaments. This work demonstrates a

unique ability to create a truly ’multi-scale’ carbon structure

on the surface of carbon fibers. Potentially multi-scale struc-

tures like those produced here could enhance composite

material strength, ductility and energy absorption character-

istics. Further research is underway to investigate this

potential.

Acknowledgements

This work has been supported by Defense Threat Reduction

Agency (DTRA) Grant # HDTRA1-08-1-0017 P00001 and the

National Science Foundation (NSF) Award # CMMI-0800249.

The authors gratefully acknowledge this support. The tech-

nical help from Dr. Steve Doorn, Los Alamos National Labo-

ratory in performing Raman Spectroscopy is greatly

appreciated.

R E F E R E N C E S

[1] Garmestani H, Al-Haik MS, Dahmen K, Tannenbaum R, Li DS,Sablin S, et al. Polymer-mediated alignment of carbonnanotubes under high magnetic fields. Adv Mater2003;15:1918–21.

[2] Reda Taha MM, Shrive NG. Enhancing fracture toughness ofhigh-performance carbon fiber cement composites. ACIMater J 2001;98:168–78.

[3] Zhandarov SF, Mader E, Yurkevich OR. Indirect estimation offiber/polymer bond strength and interfacial friction frommaximum load values recorded in the microbond and pull-out tests. Part 1: local bond strength. J. Adhes Sci Tech2002;16:1171–200.

[4] Xia Z, Curtin WA, Sheldon BW. Fracture toughness of highly-ordered carbon nanotube/alumina nanocomposites. EngMater Tech 2004;126:238–44.

[5] Li D, Zhang XF, Sui G, Wu DH, Liang J, Yi XS. Toughnessimprovement of epoxy by incorporating carbon nanotubesinto the resin. J Mater Sci Lett 2003;22:791–3.

[6] Gojny FH, Wichmann MHG, Kopke U, Fiedler B, Schulte K.Carbon nanotube-reinforced epoxy-composites: enhancedstiffness and fracture toughness at low nanotube content.Compos Sci Technol 2004;64:2363–71.

3078 C A R B O N 4 7 ( 2 0 0 9 ) 3 0 7 1 – 3 0 7 8

[7] Xia Z, Riester L, Curtin WA, Li H, Sheldon BW, Liang J, et al.Direct observation of toughening mechanisms in carbonnanotube ceramic matrix composites. Acta Mater2004;52:931–44.

[8] Li YL, Kinloch IA, Windle AH. Direct spinning of carbonnanotube fibers from chemical vapor deposition synthesis.Science 2004;304:276–8.

[9] Makris TD, Giorgi R, Lisi N, Pilloni L, Salernitano E, DeRiccardis MF, et al. Carbon nanotubes growth on PAN andpitch based carbon fibres by HFCVD. Fullerenes, Nanotubes,Carbon Nanostruct 2005;13:383–92.

[10] Boskovic BO, Golovko VB, Cantoro M, Kleinsorge B, ChuangATH, Ducati C, et al. Low temperature synthesis of carbonnanofibres on carbon fibre matrices. Carbon 2005;43:2643–8.

[11] Zhu S, Su CH, Lehoczky SL, Muntele I, Iia D. Carbon nanotubegrowth on carbon fibers. Diamond Relat Mater2003;12:1825–8.

[12] Qu L, Zhao Y, Dai L. Carbon microfibers sheathed withaligned carbon nanotubes: towards multidimensional,multicomponent, and multifunctional nanomaterials. Small2006;2:1052–9.

[13] Phillips J, Shiina T, Nemer M, Lester K. Graphitic structures bydesign. Langmuir 2006;22:9694.

[14] Phillips J, Leseman ZC, Cordaro J, Luhrs CC, Al-Haik M. Novelgraphitic structures by design. In: Proceedings of 2007 ASMEInternational Mechanical Engineering Congress andExposition, WA, Seattle, USA, 2007.

[15] Downs WB, Baker RTK. Modification of the surface propertiesof carbon fibers via the catalytic growth of carbon nanofibers.J Mater Res 1995;10:625–33.

[16] Downs WB, Baker RTK. Novel carbon fiber–carbon filamentstructures. Carbon 1991;29(8):1173–9.

[17] Baker RTK, Barber MA, Harris PS, Feates FS, Waite RJ.Nucleation and growth of carbon deposits from the nickelcatalyzed decomposition of acetylene. J Catal 1972;26:51–72.

[18] Bokx PK, Kock AJHM, Boellard E, Klop W, Geus JW. Theformation of filamentous carbon on iron and nickel catalysts– I thermodynamics. J Catal 1985;96(2):454–67.

[19] Owens WT, Rodriguez NM, Baker RTK. Carbon growth onplatinum catalyst. J Phys Chem 1992;96(12):5048–53.

[20] Rodriguez NM, Kim MS, Baker RTK. Deactivation of copper–nickel catalyst due to changes in surface composition. J Catal2003;140(1):16–29.

[21] Terrones H, Hayashi T, Munoz-Navia M, Terrones M, Kim YA,Grobert N, et al. Chem Phys Lett 2001;343(3–4):241–50.

[22] Park C, Rodriguez NM, Baker RTK. Carbon deposition on iron–nickel during interaction with carbon monoxide–hydrogenmixtures. J Catal 1997;169(1):212–27.

[23] Park C, Baker RTK. Carbon deposition on iron–nickel duringinteraction with ethylene–hydrogen mixtures. J Catal1998;179(2):361–74.

[24] Atwater MA, Phillips J, Doorn SK, Luhrs CC, Fernandez Y,Leseman ZC, et al. The production of carbon nanofibers andthin films on palladium catalyst from ethylene–oxygenmixtures. Carbon 2009;47:2269–80.

[25] Baker RTK, Harris PS. In: Walker PL, Thrower PA, editors.Chemistry and Physics of Carbon, vol. 14, New York: Dekker;1978. p. 83–165.

[26] Baker RTK, Yates DJC, Dumesic JA. Coke formation on metalsurfaces. In: Baker RTK, editor. ACS Symposium Series 202,Washington, DC, USA: American Chemical Society; 1981. p. 1.

[27] Nielsen JR, Trimm DL. Mechanisms of carbon formation onnickel-containing catalysts. J Catal 1977;48:155–65.

[28] Boellaard E, Debokx PK, Kock PA, Geus JA. The formation offilamentous carbon on iron and nickel-catalysts Morphology.J Catal 1985;96:481–90.

[29] Chen CK, Perry WL, Xu H, Jiang Y, Phillips J. Plasma torchproduction of macroscopic carbon nanotube structures.Carbon 2003;41:2555–60.

[30] Gavillet J, Loiseau A, Journet C, Willaime F, Ducastelle F,Charlier JC. Root-growth mechanism for single-wall carbonnanotubes. Phys Rev Lett 2001;87:2755041–4.

[31] Saito Y, Okuda M, Fujimoto N, Yoshikawa T, Tomita M,Hayashi T. Single wall carbon nanotubes grown radially fromni fine particles formed by arc evaporation. Jpn J Appl Phys1994;Part I 33:L526–9.

[32] Gavillet J, Loiseau A, Ducastelle F, Thair S, Bernier PO,Stephan J, et al. Microscopic mechanisms for the catalystassisted growth of single-wall carbon nanotubes. Carbon2002;40:1649–63.

[33] Mori S, Suzuki M. Characterization of carbon nanofiberssynthesized by microwave plasma-enhanced CVD at low-temperature in a CO/Ar/O2 system. Diamond Relat Mater2009;18(4):678–81.

[34] Seo MK, Park SJ. Surface characteristics of carbon fibersmodified by direct oxyfluorination. J Colloid Interface Sci2009;330:237–42.

[35] Mori S, Suzuki M. Effect of oxygen and hydrogen addition onthe low-temperature synthesis of carbon nanofibers using alow-temperature CO/Ar DC plasma. Diamond Relat Mater2008;17(6):999–1002.

[36] Wu NL, Phillips J. Catalytic etching of platinum duringethylene oxidation. J Phys Chem 1985;89:591–600.

[37] Wu NL, Phillips J. Reaction enhanced sintering of platinumthin films during ethylene oxidation. J Appl Phys1986;59:769–79.

[38] Wu NL, Phillips J. Sintering of silica-supported platinumcatalysts during ethylene oxidation. J Catal 1988;113:129–43.

[39] Dean VW, Frenklach M, Phillips J. Catalytic etching ofplatinum foils and thin films in hydrogen–oxygen mixtures. JPhys Chem 1988;92:5731–8.

[40] Wei TC, Phillips J. Thermal and catalytic etching:mechanisms of metal catalyst reconstruction. Adv Catal1996;41:359–421.

[41] Wu NL, Phillips J. Carbon deposition on platinum duringethylene oxidation. J Catal 1988;113:383–97.

[42] Linares-Solano A, Rodriguez-Reinoso F, Salinas-Martinez deLecea C, Mahahjan OP, Walker PL. Platinum catalystssupported on graphitized carbon-black. 1. Characterizationof the platinum by titrations and differential calorimetry.Carbon 1982;20:177–84.

[43] Jenkins Jr RG, Walker PL, Linares-Solano A, Rodriguez-Reinoso F, Salinas-Martinez de Lecea C. Platinum catalystssupported on graphitized carbon-black 2. Characterization ofthe platinum by small-angle X-ray-scattering andtransmission electron-microscopy. Carbon 1982;20:185–9.

[44] Ehrberger P, Mahajan OP, Walker Jr PL. Carbon as a supportfor catalysts. 1. Effect of surface heterogeneity of carbon ondispersion of platinum. J Catal 1976;43:61–7.

[45] Ehrberger P, Walker Jr PL. Carbon as a support for catalysts 2.Size distribution of platinum particles on carbons of differentheterogeneity before and after sintering. J Catal1978;55:63–70.

[46] Phillips J, Dumesic JA. Iron pentacarbonyl decompositionover grafoil – production of small metallic iron particles. JPhys Chem 1980;84:1814–22.

[47] Phillips J, Kelly D, Radovic L, Xie F. Microcalorimetric study ofthe influence of surface chemistry on the adsorption of waterby high surface area carbons. J Phys Chem B2000;104(34):8170–6.

[48] Wu NL, Phillips J. XRD evidence of preferential orientation ofplatinum crystallites on graphite. Surf Sci 1987;184:463–82.