Carbon 44 (2006) 67–78

www.elsevier.com/locate/carbon

The effect of the cobalt loading on the growthof single wall carbon nanotubes by CO disproportionation

on Co-MCM-41 catalysts

Yuan Chen a,b, Dragos Ciuparu a,c, Sangyun Lim a, Gary L. Haller a, Lisa D. Pfefferle a,*

a Department of Chemical Engineering, Yale University, New Haven, CT 06520, USAb School of Chemical and Biomedical Engineering, Nanyang Technological University, 637722, Singapore

c Department of Petroleum Processing and Petrochemistry, ‘‘Petrol-Gaze’’ University, 2000 Ploiesti, Romania

Received 8 December 2004; accepted 13 July 2005Available online 28 September 2005

Abstract

Highly ordered MCM-41 mesoporous molecular sieves in which silicon was isomorphously substituted with 0.5–4 wt.% cobaltwere synthesized using an alkyl template with a 16 carbon atoms alkyl chain length. These materials were used as catalysts forthe synthesis of uniform diameter single wall carbon nanotubes (SWNT) by CO disproportionation (Boudouard reaction). TheSWNT synthesis conditions were identical for all Co-MCM-41 samples, and consisted of pre-reduction of the Co-MCM-41 catalystin hydrogen at 500 �C for 30 min followed by reaction with pure CO at 800 �C and 6 atm for 1 h (conditions previously optimizedfor 1 wt.% Co-MCM-41). The SWNT grown in the Co-MCM-41 catalysts were characterized by TGA, multi-excitation energyRaman spectroscopy and TEM. The state of the catalyst and the size of the metallic cobalt clusters formed in Co-MCM-41 duringthe SWNT synthesis were characterized by X-ray absorption spectroscopy. The mechanism controlling the diameter distribution ofthe SWNT produced is related to the size uniformity of the cobalt clusters nucleated in the Co-MCM-41 catalytic template: theSWNT growth selectivity and size uniformity is influenced by the cobalt concentration in the framework. If the cobalt is not initiallystrongly stabilized in the MCM-41 framework during template synthesis, the catalyst produces SWNT with a wider diameter dis-tribution. Co-MCM-41 catalysts with up to 3 wt.% cobalt can be used to grow SWNT with a diameter distribution similar to thatobtained with 1 wt.% Co-MCM-41, but at yields greater by a factor of approximately 2.4.� 2005 Published by Elsevier Ltd.

Keywords: Carbon nanotubes; Carbon yield; Catalyst

1. Introduction

Since their discovery in 1991, carbon nanotubes havestimulated an intense research effort due to their highmechanical strength [1] and thermal conductivity [2],their unique electronic properties [3–6], and the possibil-ity of building nanoscale molecular devices. Amongknown nanomaterials, carbon nanotubes exhibit per-haps the richest diversity of structures and structure-

0008-6223/$ - see front matter � 2005 Published by Elsevier Ltd.doi:10.1016/j.carbon.2005.07.035

* Corresponding author. Fax: +1 203 432 4387.E-mail address: lisa.pfefferle@yale.edu (L.D. Pfefferle).

property relations [7]. Their physical and chemical prop-erties are determined by the chirality, that is, the way inwhich an equivalent structure would form by a graphitesheet rolling up, in addition to the nanotube length anddiameter. The SWNT growth techniques explored so farcannot produce significant amounts of SWNT with pre-determined specific properties. The lack of purity anduniformity in length, diameter, and chirality has beena significant hindrance to the development of a success-ful technology for large scale production of electronicdevices using carbon nanotubes. With cleaning andseparation, narrow tube diameter distributions are

68 Y. Chen et al. / Carbon 44 (2006) 67–78

obtainable, but electronic properties are compromisedby contamination and introduction of defects.1

It has been suggested that the diameter of SWNT iscontrolled by the size of metal catalyst particles [8,9].Therefore, the key to the controlled growth of SWNTis to control the similar metal clusters in catalysts, avoid-ing aggregation into large particles during the high tem-perature reaction. Among the three main processes usedfor carbon nanotubes synthesis—the arc discharge, laserablation, and chemical vapor deposition (CVD)—thelast may offer more control of the tube diameter and chi-rality because it is relatively easy to control metal clustersize in solid catalysts. Several efforts have been made tocontrol the metallic particles for SWNT growth [8,10–14], but in most studies the sizes of the nanoparticlesare relatively large for SWNT growth, ranging from 1to 14 nm, which leads to the formation of SWNT witha wide diameter distribution with amorphous carbon,multi-wall carbon nanotubes (MWNT), and graphiteimpurities. We [15,16] and other authors [17] have re-ported that the increase of the cobalt particle sizes inthe catalysts during SWNT synthesis leads to widerSWNT diameter distribution and poor SWNT selectiv-ity. Resasco and coworkers found that molybdenum[18,19] and tungsten [20] can be used to stabilize the co-balt against reduction and investigated the effect of theCo to Mo ratio on the SWNT synthesis performance.Using an optimized catalyst, these authors were ableto achieve a narrow diameter distribution for the semi-conducting SWNT as determined from fluorescencemeasurements [21].

We have recently developed a catalytic system con-sisting of cobalt as the catalytic component incorpo-rated into the pore wall of MCM-41 mesoporousmolecular sieves by isomorphous substitution for sili-con, resulting in an initially nearly atomic dispersion[22]. In our SWNT synthesis process using Co-MCM-41 catalyst, the cobalt is reduced and nucleates intosub-nm metallic clusters that initiate the growth of car-bon nanotubes. The MCM-41 matrix stabilizes the co-balt against reduction, allowing formation of verysmall cobalt clusters, uniform in size, that enable growthof SWNT with diameters within ±0.05 nm [23]. The sizeof the cobalt clusters produced can be engineered bymanipulating the pore radius of curvature, that is incor-porating cobalt in MCM-41 of different pore diameters,thus allowing the growth of carbon nanotubes of uni-form, pre-selected diameter [23,24]. Other template syn-thesis parameters, notably the initial pH of the synthesissolution, affect the size and state of the cobalt clustersformed during the SWNT synthesis process [25].

All of our previous studies on the effects of the syn-thesis parameters on SWNT synthesis performance, as

1 Dr. Phaedon Avouris from IBM, private communication.

assessed by SWNT yield, purity, and diameter unifor-mity, were performed using catalysts with 1 wt.% Coloading [15,16]. The optimized pre-treatment and reac-tion conditions allowed synthesis of uniform diameterSWNT, but at a limited yield. The tubes were noted tobe significantly more defect free than those from otherprocesses1 most likely due to the milder chemical treat-ments required for purification. A possible strategy to beconsidered for the synthesis of larger yields of SWNT isto increase the metal loading in our catalyst. Since theperformance of the SWNT synthesis process was ob-served to depend mainly on the size and uniformity ofthe sub-nm metallic Co clusters formed on the catalystsurface, changing the cobalt loading may require differ-ent SWNT synthesis conditions in order to obtainSWNT with good purity and narrow diameter distribu-tion. Production of larger SWNT yields would provideincentives for the development of large scale productionof SWNT with a narrow diameter distribution. There-fore, the present contribution is focused on the investi-gation of the influence of the cobalt loading on theselectivity and the diameter distribution of the SWNTproduced as a critical step to increase the efficiency ofthe SWNT synthesis process based on the Co-MCM-41 catalyst.

2. Experimental

Co-MCM-41 samples with different cobalt loadingsof 0.5, 1.0, 2.0, 3.0 and 4.0 wt.% (as determined byinductively coupled plasma, ICP measurement at Gal-braith Laboratories, Inc.) were synthesized followingthe method described elsewhere [22]. Both the purityof the silica source, and the pH during Co-MCM-41synthesis were observed to influence the reducibility ofthe cobalt ions in the framework [25]. The catalystsemployed in these studies were synthesized using aCab-O-Sil silica source and the initial pH of the synthe-sis solution was controlled at 11.5. The physicochemicalproperties of the Co-MCM-41 samples used in thisstudy are given and discussed in detail elsewhere [22].

SWNT were synthesized by CO disproportionation.For a typical batch, 200 mg of fresh Co-MCM-41 wereloaded into a 10 mm internal diameter quartz reactorplaced in an Omega ceramic fiber radiant heater, whichallowed precise temperature control throughout the en-tire catalyst bed. Prior to exposure to CO the catalystwas heated in flowing hydrogen at one atmosphere fromroom temperature to 500 �C at 20 �C/min, and reducedisothermally for 30 min. After this pre-reduction treat-ment, the catalyst was purged with ultra high purity ar-gon at the reduction temperature, and then heated to800 �C at 20 �C/min in flowing argon. SWNT weregrown for 60 min under 6 atm CO (99.5% from Airgas).Before entering the reactor, the CO stream was passed

Fig. 1. DTG curves for as-synthesized Co-MCM-41.

Y. Chen et al. / Carbon 44 (2006) 67–78 69

through a bed of glass beads heated at 400 �C in order todecompose the Fe pentacarbonyl originating from theCO container.

A Setaram Setsys 1750 instrument was used for thethermo-gravimetric analysis (TGA) under air flow ofthe catalyst samples loaded with SWNT. Samples wereheld at 150 �C for 1 h to remove the water adsorbed inthe MCM-41 material before initiating the temperatureprogram. The weight change in the sample was moni-tored over the temperature program from 150 to1200 �C at 10 �C/min for two successive ramps; the sec-ond ramp was used as a baseline to correct the first one.A holey crucible was used to limit mass transfer interfer-ence. The total carbon yields on Co-MCM-41 weredetermined by Leco induction furnace oxidation at Gal-braith Laboratories, Inc.

Multi excitation wavelength Raman spectra of the as-synthesized SWNT, without any purification or pre-treatment, were collected with a Renishaw Ramanscopein the backscattering configuration. The spectra wereobtained using 488 nm (2.54 eV), 514.5 nm (2.41 eV),633 nm (1.96 eV) and 785 nm (1.58 eV) laser energies.

High Resolution Transmission Electron Microscopy(HR-TEM) images of SWNT were collected on a TecnaiF20 200 kV microscope from Philips. The solid sampleswere dispersed in pure ethanol by sonication andapproximately 0.05 ml of this suspension was droppedon a copper mesh coated with an amorphous holey car-bon film. The ethanol evaporated at room temperatureprior to the TEM analysis.

X-ray absorption data were collected at beam lineX23A2, National Synchrotron Light Source, Brookha-ven National Laboratory. Two 30-cm long ion cham-bers filled with pure N2 were placed collinearly withthe beam, in front and behind the sample, to measurethe intensities of the incident (I0) and transmitted (IT)beams. A third ion chamber was used to record the spec-trum of the internal reference sample consisting of a co-balt foil in order to determine the absorption edge of Co(7709 eV) for each spectrum. An approximately 45 mgsample of Co-MCM-41 loaded with carbon was pressedinto a rectangular wafer (�1.5 · 1 cm) to form 0.5-mmthick pellets. The thickness of pellets satisfied the condi-tion that absorption edge steps at the Co K absorptionedge was about 2. Extended X-ray Absorption FineStructure (EXAFS) spectra were recorded in the trans-mission mode from 200 eV below to 1000 eV above theCo K edge.

Analysis of the X-ray adsorption spectra followed theprocedures described in detail in our previous report[16]. The EXAFS spectra were calibrated to the edge en-ergy of a cobalt foil internal reference. The backgroundremoval and edge-step normalization were performedusing the FEFFIT code [26]. The theoretical EXAFSfunctions for different cobalt species (Co and Co3O4)generated by the FEFF6 program [27] were used to fit

the experimental data in order to obtain the correspond-ing Co–Co and Co–O first shell coordination numbers.

3. Results

Temperature programmed oxidation was previouslyused to distinguish among different carbon species insamples containing SWNT [28]. However, the oxidationtemperature of SWNT has been observed to vary con-siderably for samples prepared at different conditions[29,30]. The rather wide range of SWNT oxidation tem-peratures has been attributed to the differences in thecatalytic activity of metallic particles present in SWNTsamples [28]. Quantitative determination of differentcarbon species on the catalysts by TGA is not accurate.In our results, TGA was only used for qualitative anal-ysis. Differential thermal analysis (DTA) results of Co-MCM-41 loaded with carbon after SWNT growth weregiven in Fig. 1 for catalysts samples with 0.5, 1, 2, 3 and4 wt.% Co. There were several peaks between 150 and1200 �C corresponding to mass loss by carbon oxidation(negative peaks) and oxygen uptake by cobalt oxidation(positive peaks). The first negative peak centered be-tween 150 and 300 �C, overridden by a positive peak,especially for higher Co concentration samples, is as-signed to the oxidation of amorphous carbon species.During amorphous carbon oxidation, however, there isa weight gain beginning at around 200 �C resulting fromthe oxidation of the metallic cobalt particles initiallycovered by amorphous carbon. When the amorphouscarbon layer covering the cobalt particles is removed,the reduced cobalt particles become accessible to thegas phase oxygen and are oxidized. The weight gain

Fig. 2. Variation of the total carbon yield and carbon yield/cobaltloading ratio as a function of cobalt loading in MCM-41.

Fig. 3. Multi-excitation wavelength Raman spectra recorded forSWNT grown on 3 wt.% Co-MCM-41.

70 Y. Chen et al. / Carbon 44 (2006) 67–78

from cobalt oxidation increases with the increase in co-balt loading from 0.5 to 4 wt.%.

The two negative peaks around 360 and 510 �C areassigned to oxidation of SWNT. This assignment wasconfirmed by Raman spectroscopy of carbon loadedCo-MCM-41 samples previously exposed to air for30 min at 300 and 450 �C, respectively. The Ramanspectra (not shown) gave evidence of strong Ramanbreathing mode (RBM) peaks after both oxidationtreatments; however, after exposure to air at 600 �C,the RBM peaks disappeared suggesting complete oxida-tion of the SWNT in the sample at this temperature. Theposition of the peak around 360 �C shifts with the cobaltloading, most likely because the differences in the inter-actions between the SWNT and the cobalt metal clustersduring the oxidation process. Also the presence of tubesboth inside and outside the pores of the MCM-41 mayinduce oxygen mass transfer limitations that broadenedand shifted the SWNT oxidation temperatures. Themass loss peaks centered above 900 �C are assigned tothe oxidation of graphitic carbon. The intensity of thegraphite peak increased with the increase in the cobaltloading, suggesting that the higher cobalt concentrationcaused cobalt to migrate into larger particles, moreselective for graphite. It should be noted here, however,that the increase in the mass loss peaks assigned to theoxidation of SWNT is much stronger than those ofamorphous carbon and graphite, suggesting a highselectivity to SWNT. It is important to minimize gra-phitic carbon because it is the most difficult impurityto remove without destroying the nanotubes.

Because of the overlap between the mass loss peakcaused by carbon oxidation and the mass gain peak pro-duced by cobalt oxidation, the total carbon yield deter-mined by TGA is not accurate. Therefore, the carbonyields resulting in this study were determined by Lecoinduction furnace oxidation and measurement of theevolved CO2 at Galbraith Laboratories, Inc. The totalcarbon yield is plotted in Fig. 2 as a function of the co-balt loading of the catalyst. The carbon yield increasedfrom 0.98 to 5.13 wt.% when the cobalt loading in-creased from 0.5 to 4 wt.%, respectively. It should benoted here that the carbon yield obtained with the1 wt.% Co-MCM-41 in this study is much lower thanthose obtained in our previous studies [15,16]. The lowercarbon loading results most likely from the different pHlevel used in the synthesis of the Co-MCM-41 catalystsin this study, as discussed elsewhere [25]. These carbonyield levels are consistent with our expectation that theamount of carbon produced correlates with the concen-tration of the cobalt in the catalysts. As indicated inFig. 2, the ratio between the carbon yield and the cobaltloading is relatively constant. The average yield is about1.25 wt.% carbon/1 wt.% cobalt, which translates intoabout 6 carbon atoms per Co atom. Assuming each ac-tive cobalt cluster in Co-MCM-41 consists of about 20

atoms (as determined in a previous study [23]), therewill be approximately 120 carbon atoms per active Cocluster. A SWNT of about 0.8 nm in diameter and100–1000 nm in length as determined by our TEMexperiments, contains about 1000–10,000 carbon atoms.This difference, consistent with other CVD processes,suggests that there may be only a very small fractionof the Co in the catalyst which is active for SWNTgrowth.

Y. Chen et al. / Carbon 44 (2006) 67–78 71

The SWNT produced with each catalyst were charac-terized by multiple excitation wavelength Raman. TheRaman spectra collected at different laser energies forthe Co-MCM-41 sample containing 3 wt.% Co afterSWNT growth are given in Fig. 3. Independent of thelaser energy, the spectra showed three types of features:the RBM peaks between 150 and 350 cm�1 characteris-tic for SWNT, the D band at approximately 1300 cm�1

assigned to defective and disordered carbon species, andthe peak complex centered around 1600 cm�1, known asthe G band, which is characteristic for ordered carbonspecies such as carbon nanotubes and graphite. TheRBM region of the spectra for the four excitation ener-gies in Fig. 3 indicate a rather narrow distribution ofSWNT diameters. The Raman spectra of the SWNTon Co-MCM-41 with different cobalt loadings showedsimilar features with those shown in Fig. 3. The RBMregion of the spectra recorded using four different exci-tation energies for the as-synthesized SWNT samplesproduced in the Co-MCM-41 catalysts with cobalt load-ings from 0.5 wt.% to 4 wt.%, without any purificationor pre-treatment, are shown in Fig. 4. It can be seen thatthe spectra for SWNT produced in Co-MCM-41 cata-lysts with 1, 2 and 3 wt.% cobalt loadings show onemajor peak and few weak satellite peaks at each laser

Fig. 4. Raman breathing mode regions of multi-excitation energy Raman s488 nm, (b)514.5 nm, (c) 633 nm, (d) 785 nm. Stars indicate the peaks identi

energy, while the spectrum recorded for the Co-MCM-41 sample with 4% cobalt shows three clearly definedpeaks under 633 nm laser in Fig. 4c, suggesting thediameter distribution of the SWNT is altered as the co-balt loading increases beyond 3 wt.%. It should also benoted that the diameter distribution obtained with thecatalyst containing only 0.5 wt.% Co is also somewhatbroader, most likely because it is dominated by the lar-ger metallic clusters formed by the faster reduction ofthe Co3+ species present in the catalyst [24].

Rao et al. [31] first demonstrated the dependence ofthe intensity of RBM peaks on the laser excitation en-ergy, which results in the diameter-selective resonanceRaman scattering of vibrational modes in carbon nano-tubes. When the energy of incident photons matches avan Hove singularity (VHS) for the joint valence andconduction bands of SWNT (subject to selection rulesfor optical transitions), one expects to find the resonantenhancement of Raman spectra [32]. The relation be-tween electron transitions and a possible SWNT struc-ture can be described in a Kataura plot, as shown inFig. 5 using a tight binding model [33]. Semiconductingand metallic tubes are indicated by solid and open cir-cles, respectively. The diamond symbols in Fig. 5 markthe tubes corresponding to the peaks observed in the

pectra for SWNT grown on 0.5, 1, 2, 3 and 4 wt.% Co-MCM-41 (a)fied in RBM regions on 3 wt.% Co-MCM-41 sample.

72 Y. Chen et al. / Carbon 44 (2006) 67–78

multi-excitation energy Raman identified by star sym-bols in Fig. 4, with the large diamonds correspondingto the main RBM peaks and the small ones to the weak-er satellite peaks. The diameter values were determinedusing the equation proposed by Weisman and Bachilo[34]. Some of diamond symbols do not match the circlesresulting from calculations. Weisman demonstrated thatthe tight binding model underestimates the apparentscattering arising from variations of the tube chiralityat a given diameter, and underestimates the E11 transi-tion energies of the semiconducting SWNT [34], whichlikely cause the mismatch between experimental dataand calculated points.

Except for the 4 wt.% sample, the Raman spectraconfirmed the narrow diameter distribution of theSWNT observed in the TEM images shown in Fig. 6.It should be noted that the RBM peaks obtained usingthe 633 and 785 nm laser energies and correspondingto larger diameter tubes had very weak intensities,although their locations in the Kataura plot suggest theyshould be in resonance at these laser energies. This indi-cates that there are only small amounts of larger tubes inthese samples, as confirmed in our TEM experiments.These results, correlated with the carbon yields givenin Fig. 2, suggest that the increase in the cobalt loadingcan be used to increase the yield of the SWNT producedwith the Co-MCM-41 catalysts, but cobalt loadings lar-ger than 3 wt.% alter the uniformity of the diameter dis-tribution of the SWNT produced. However, both

Fig. 5. Background: Circles from calculated gap energies as a functionof SWNT diameter by tight binding (TB) model (adapted from [33]).ESðMÞ11ð22;33Þ, S refers to semiconducting tubes, M refers to metallic tubes,

11, 22, or 33 refer to the first, second or third electron transitions.Superimposed data: the diamonds indicate the diameter of the SWNTdetermined from RBM on 3 wt.% Co-MCM-41 sample in Fig. 4.Horizontal lines indicate four energies of incident photons usually usedin Raman instruments: 488 (2.54 eV), 514.5 (2.41 eV), 633 (1.96 eV)and 785 nm (1.58 eV). The dash line indicates the average diameter ofSWNT determined by TEM analysis.

template synthesis parameters and SWNT synthesisreaction conditions have been optimized only for1 wt.% cobalt catalysts produced with the C16 alkylchain length template. Narrow diameter distributionSWNT can probably be produced with high yields byCO disproportionation over 4 wt.% Co-MCM-41 cata-lysts if template synthesis parameters and SWNT syn-thesis conditions are optimized for higher cobaltloadings. However, optimization of template synthesisvariables in conjunction with catalyst pre-treatmentand SWNT growth reaction conditions are beyond thescope of the present contribution.

Many researchers have attributed the red-shift andbroadening of the Raman and near-IR spectra to theaggregation of SWNT into bundles [32,35–38]. Recently,Strano used Raman spectroscopy to elucidate the aggre-gation state of SWNT [39]. As-synthesized SWNT pro-duced with the Co-MCM-41 catalysts are aggregatedin bundles as observed in the TEM images in Fig. 6.Taking into account the broadening of the spectra re-corded with aggregated tubes, the peaks in the RBM re-gion should be even narrower than those observed inFig. 4. The 1 wt.% Co-MCM-41 in Fig. 4d indicates awider distribution compared with our previous resultsobtained with 1 wt.% Co-MCM-41 [15,16]. The widerdistribution most likely comes from the different pH le-vel used in the synthesis of Co-MCM-41 as discussedelsewhere [25], and other un-optimized catalyst andSWNT synthesis variables. Both pH and pore diameterwere observed to affect the reducibility of the Co in theMCM-41 template, and thus, the reaction conditionsneed to be optimized for each template synthesiscondition.

HR-TEM images in Fig. 6 show that Co loading af-fects both the selectivity to SWNT and diameter unifor-mity of SWNT produced. The quality of SWNT grownis correlated with the size of the cobalt clusters formed inthe catalyst. Fig. 6A and B show that 3 wt.% Co-MCM-41 produces SWNT with a uniform diameter. There arealso a few large cobalt particles in this sample. Most co-balt clusters leading to the growth of SWNT are still in-side the MCM-41 pore structure and they are difficult toimage by TEM. On the other hand, Fig. 6C and D from4 wt.% Co-MCM-41 show many large cobalt particleswith diameters in the 10–20 nm range, covered by layersof graphite. The graphite contributes to peaks above900 �C in TGA of Fig. 1 and its presence indicates that4 wt.% Co-MCM-41 has lower selectivity to SWNTthan 3 wt.% Co-MCM-41. The single SWNT inFig. 6D has a diameter of 1 nm, which is correlated withlarger diameter SWNT peaks in Raman spectra inFig. 4. It also indicates that 4 wt.% Co-MCM-41 pro-duce nanotubes with wider diameter distribution.

X-ray absorption spectroscopy was employed toinvestigate how cobalt concentration affects the forma-tion of cobalt clusters during SWNT synthesis and influ-

Fig. 6. TEM images of SWNT on Co-MCM-41, (A, B) 3 wt.% Co-MCM-41, (C, D) 4 wt.% Co-MCM-41. Tube bundles (1), carbon layers coveringcobalt particles (2).

Fig. 7. Normalized Co K-edge XANES spectra recorded for 0.5–4%Co-MCM-41 loaded with carbon after SWNT synthesis. Spectra forthe fresh Co-MCM-41 and cobalt foil are given as references. The insetshows the derivative of Co foil spectrum.

Y. Chen et al. / Carbon 44 (2006) 67–78 73

ences the diameter distribution of the SWNT produced.The EXAFS spectra near the Co K edge (7709 eV) re-corded for Co-MCM-41 samples with cobalt loadingsranging from 0.5 to 4 wt.% after SWNT synthesis underidentical pre-reduction and reaction conditions are givenin Fig. 7. The spectra recorded with a Co foil and thefresh Co-MCM-41 are also given for comparison. Theinsert of Fig. 7 shows the derivative of the Co foil spec-trum to indicate the spectral features of interest, as dis-cussed below. Note that the spectra were normalized bythe number of Co atoms in the beam as the averagechemistry per atom is being calculated.

Three features near the Co K edge provide informa-tion concerning the state of the Co-MCM-41 catalyst.The first feature is the pre-edge peak assigned to the di-pole forbidden transitions whose intensities are func-tions of the local symmetry of the Co atoms or ions.The pre-edge intensity increases when the cobalt concen-tration in MCM-41 increases. The second spectral fea-ture of interest is the energy of the main edgeevidenced by the second peak (B) in the derivative ofthe EXAFS spectrum given in the inset of Fig. 7. Theposition of the main edge varies linearly with the valence

of the Co species. As discussed in our previous contribu-tion [40], the Co main edge does not show significant

Fig. 8. Variation of the intensities of the pre-edge peak and of thewhite line with Co concentration in MCM-41.

74 Y. Chen et al. / Carbon 44 (2006) 67–78

shifts because of the narrow range of possible oxidationstates and is, therefore, difficult to assess. We have,however, successfully applied the generalized 2D corre-lation analysis technique to monitor the changesoccurring in this spectral feature in a series of dynamicX-ray absorption spectra recorded in situ during theSWNT synthesis process [41]. The third spectral featureof interest is the intensity of white line, which correlateswith the density of holes in the d orbitals/band of cobaltspecies. The white line intensity for samples after SWNTsynthesis decreases with the increase of Co concentra-tion. The intensity of the pre-edge peak observed inthe EXAFS spectra at the energy corresponding to thefirst minimum in the derivative of the Co foil depictedby (A) in the inset of Fig. 7 (i.e., 7713 eV), and that ofthe white line, determined from the minimum depictedby (C) in the inset of Fig. 7 (i.e., 7725 eV), are plottedin Fig. 8 against the cobalt loading of the sample. Theincrease in the intensity of the pre-edge feature andthe decrease of white line intensity as the concentra-tion of cobalt in the Co-MCM-41 catalyst increases isdirect evidence for the increase in the concentration ofthe reduced cobalt species in samples with higher Coloadings.

The spectra in Fig. 7 are plotted in R space in Fig. 9together with the fitting results obtained using the theo-retical EXAFS functions for reduced and oxidized co-balt species (Co and Co3O4) generated by the FEFF6program. The fitting parameters are given in Table 1.The major peaks centered at R values of approximately1.6 and 2.2 A correspond to the Co–O and Co–Co inter-actions, respectively. The spectrum for the 0.5 wt.% Co-MCM-41 showed a weak Co–Co peak suggesting a very

Fig. 9. EXAFS data for 0.5–4% Co-MCM-41 catalysts after reaction in R

small fraction, if any, of the cobalt ions in the silicaframework has been completely reduced to metal andnucleated into cobalt clusters to give rise of Co–Co peakin the EXAFS spectrum in R space. The resulting zero

space along with the fitting of the spectra with theoretical models.

Table 1Structure parameters of Co-MCM-41 catalysts from EXAFS fitting

Co, % Co–O first shell Co–Co first shell

NCo–Oa dR (A)b r2c NCo–Co

d dR (A)b r2c

0.5 3.98 ± 1.62 0.14 ± 0.02 0.88 0 – –1 3.63 ± 1.08 0.14 ± 0.02 0.89 2.82 ± 1.48 �0.03 ± 0.01 0.832 1.00 ± 0.45 0.13 ± 0.01 1.48 6.58 ± 0.81 �0.02 ± 0.01 0.803 0.52 ± 0.32 0.12 ± 0.02 1.94 7.64 ± 0.97 �0.03 ± 0.01 0.724 0.28 ± 0.15 0.12 ± 0.03 2.15 8.26 ± 0.45 �0.03 ± 0.03 0.71

a NCo–O average first shell coordination of cobalt-oxygen.b dR deviation from the effective half-path-length R (R is the interatomic distance for single scattering paths).c r2(·10�2 A2) mean-square deviation in R.d NCo–Co average first shell coordination of cobalt.

Y. Chen et al. / Carbon 44 (2006) 67–78 75

Co–Co coordination number is consistent with a smallfraction of reduced cobalt. Also note, as discussedabove, that only a small fraction of the cobalt in the cat-alyst forms clusters active for SWNT growth, consistentwith zero Co–Co coordination number for the 0.5 wt.%sample. When the cobalt concentration in MCM-41 in-creases, the intensity of the corresponding Co–Co peakincreases and the Co–O peak decreases. The same trendwas observed for the Co–Co and Co–O coordinationnumbers listed in Table 1.

2 Lim et al. work in preparation.

4. Discussion

We observed in our previous studies [15,16] that thediameter uniformity of the SWNT produced with theCo-MCM-41 catalyst is achieved by controlling the sizedistribution of the sub-nanometer Co clusters formed inthe catalyst. The cobalt ions, initially nearly atomicallydispersed into the amorphous silica wall during theMCM-41 synthesis, are stabilized in the oxidized formagainst reduction under severe reducing conditions[24]. During the SWNT synthesis process, however,the cobalt ions are gradually reduced, and reach thepore wall surface to nucleate into metallic clusters. Thediameter and diameter distribution of the SWNT pro-duced is most likely controlled by the size/state and sizedistribution of the cobalt clusters initiating their growth.The size distribution of the cobalt clusters was proposedto be controlled by the relative rates of several compet-ing physical and chemical processes affecting the struc-ture and the state of the catalyst during the SWNTsynthesis [40]. These competing processes include cobaltreduction, migration through the pore wall to the porewall surface, nucleation and growth into clusters, andcarbon deposition.

We have found in our previous in situ X-ray absorp-tion spectroscopy studies that pre-reduction of Co-MCM-41 in hydrogen at 500 �C does not reduce the co-balt ions to metallic cobalt, but only removes hydroxylgroups by heating [40]. Pre-reduction also increases thedensity of electrons in the d orbital of the Co2+ ions,

and produce intermediate Co1+ species, weakening theirinteraction with the silica framework. CO is expected toweaken the interaction between the cobalt species andthe MCM-41 pore wall making them more mobile atthe surface and accelerating their nucleation into clus-ters capable of dissociation of CO and initiation of thegrowth of SWNT. This mechanism was supported bythe strong influence of the CO pressure on the rate ofCo cluster growth and by the sharp increase in thepre-edge peaks only after the catalyst is exposed to CO[40]. However, the growth of the cobalt clusters ceasesas they become covered by carbon.

According to the mechanism discussed above, vary-ing the cobalt loading in our samples should affect boththe stability of Co in fresh Co-MCM-41 catalysts andthe rates of some of the competing processes controllingthe size distribution of the Co clusters formed duringSWNT synthesis, which in turn influence the diameterdistribution of the SWNT produced.

Co-MCM-41 samples with different cobalt loadingshave different stability against reduction, as determinedby temperature programmed reduction. While a detailedreport on the characterization of Co-MCM-41 sampleswith different cobalt loadings can be found elsewhere,2

the results of the temperature programmed reductionexperiments performed with Co-MCM-41 samples hav-ing different Co loadings are summarized in Table 2.The data in Table 2 clearly show that while for sampleswith cobalt loadings of 2 wt.% and higher, the cobaltloading does not influence the behavior of the catalystduring reduction, the samples containing 0.5 and1 wt.% Co are more resistant against reduction thanthose with higher Co loadings, therefore under identicalreaction conditions they have lower reduction rates. Itshould also be noted here that the catalysts with Coloadings of 2 wt.% and higher also showed significantlyhigher concentrations of Co3+ species initially present.These species are reduced at significantly lower temper-atures. Because the cluster size estimated from the

Table 2Summary of temperature programmed experiments for Co-MCM-41samples with different cobalt loadings

Co loading(wt.%)

Onset ofreduction(�C)

Maximumreductionrate (�C)

Fractionof Co3+ fromtotal cobalt (%)

0.5 745 915 01.0 662 795 2.62.0 621 733 7.63.0 621 733 11.74.0 621 733 14.4

76 Y. Chen et al. / Carbon 44 (2006) 67–78

Co–Co coordination numbers is a volume average of allspecies in the sample, the presence of a few large clustersin a sample may significantly affect the Co cluster sizeestimate from the Co–Co coordination number. Sincethe concentration of the Co3+ species is significantlyhigher for the samples with 2 wt.% Co loading and high-er, it is likely that the Co–Co coordination numbers forthese samples are overestimated because these easilyreducible species are reduced quickly and are likely toform a few large Co clusters, selective for amorphouscarbon and graphite, resulting in a bimodal distributionof particle sizes in the sample. This is consistent with therather large jump in the Co–Co coordination numberbetween 1% and 2% Co-MCM-41 samples, followedby a milder increase in the Co–Co coordination numberas the Co loading is further increased. However, somecontribution of the higher stability against reductionof the sample containing 1 wt.% cobalt to this steeper in-crease in the Co–Co coordination number cannot be ru-led out.

From a chemical kinetics perspective, assuming firstorder reduction rate with respect to cobalt, it is expectedthat a higher concentration of cobalt in the frameworkwould accelerate the reduction process leading to a lar-ger gradient between the concentration of reduced spe-cies at the pore wall surface and in the bulk of thepore wall. Consequently, this larger concentration gradi-ent should accelerate the diffusion of reduced speciesthrough the pore wall to the pore wall surface, causinga larger concentration of reduced cobalt species at thepore wall surface, which should further translate intofaster nucleation rates and formation of larger cobaltclusters. These expected effects were observed in the sizeof the cobalt clusters quantified in terms of Co–Co coor-dination numbers, with the Co–Co coordination num-ber increasing from 0 to 8.26 when Co loadingincreased from 0.5 to 4 wt.%, respectively. Interestingly,this cluster size increase did not produce a wider distri-bution of tube diameters until the Co loading reached4 wt.%, as determined from the multiple excitationwavelengths Raman spectroscopy and TEM character-ization of the SWNT produced, suggesting the rate ofthe process controlling the diameter of SWNT producedis not affected by the cobalt loading as much as thosecontrolling the size of the cobalt clusters.

A recent molecular dynamics study [42] of the catalystparticle size dependence on carbon nanotube growthindicated that catalysts particles containing at least 20metal (Fe) atoms have a far better tubular structure thanSWNT from smaller clusters. Because of the higherreduction temperature of the 0.5 wt.% Co-MCM-41,lower concentration of reduced cobalt atoms are avail-able at the pore wall surface, cobalt atoms are more likelystrongly anchored on the surface and nucleate slowly intoclusters. Once the cobalt cluster size reaches the optimalsize around 6–8 A, they lead to the growth of SWNT andcluster size growth ceases. The EXAFS results obtainedwith the 0.5 wt.% Co-MCM-41 sample show the weakestpre-edge intensity in Fig. 7 and the smallest Co–Co peakin the R space spectrum in Fig. 9. This is likely becausethere are just a few Co clusters in this sample large en-ough to grow SWNT, while the others are so small thatthey do not show the collective behavior characteristicfor metallic cobalt. The Co–Co coordination number ofthe 0.5 wt.% Co-MCM-41 after SWNT synthesis wasnear zero. For 1–3 wt.% Co-MCM-41, the cobalt is alsoinitially strongly stabilized in the MCM-41 framework.After reduction and nucleation, however, there are morecobalt clusters that reach the optimized size to produceduniform diameter SWNT. On the other end, the 4 wt.%Co-MCM-41 had many more Co3+ species, which aremuch easier reduced and nucleate into very large parti-cles. Higher concentrations of reduced cobalt atoms areavailable at the pore wall surface. Cluster size growthwins the competition with SWNT growth, when clusterssize reach 6–7 A, they are not covered by carbon, but in-stead they get the chance to grow larger, which leads tothe growth of SWNT with larger diameter and graphite.This scenario is supported by the strongest pre-edgeintensity in Fig. 7, the largest Co–Co peaks in theR spacespectra in Fig. 9, and the highest Co–Co coordinationnumber for the 4 wt.% Co-MCM-41 sample. Large parti-cles are more selective to graphite; once they are coveredby several layers of graphite, they cannot deposit morecarbon leading to a low carbon to cobalt weight ratio,as observed in Fig. 2. This is also consistent with the re-sults shown in Fig. 1, where the 4 wt.%Co-MCM-41 pro-duces the largest amount of graphite, and with the TEMimage in Fig. 6C and D showing the large metallic parti-cles covered by carbon.

It should also be mentioned that, since carbon yield/carbon loading from Fig. 2 is almost constant, the frac-tion of active Co clusters (those which grow SWNT)must be proportional to Co loading. However, the aver-age cobalt cluster size increases non-linearly with Coloading, as indicated by Co–Co coordination numberin Table 1. This suggests that those large particles lead-ing to the growth of graphite and amorphous carbondominate the EXAFS results. This is consistent withthe fact that the EXAFS result is a volume average ofall species in the sample, thus the presence of a few large

Y. Chen et al. / Carbon 44 (2006) 67–78 77

particles may significantly affect the Co cluster sizeestimation.

5. Conclusions

An investigation of the effect of the cobalt concentra-tion in MCM-41 on the SWNT growth was carried out.The results obtained by TGA, Raman spectroscopy andHR-TEM, combined with EXAFS spectroscopy suggestthat the cobalt concentration affected the reducibility ofthe cobalt incorporated into the MCM-41 controllingthe size of the metal clusters formed during SWNT syn-thesis and thus the selectivity to and diameter unifor-mity of SWNT produced. Cobalt clusters with anoptimal size of approximately 6–8 A lead to the growthof SWNT with uniform diameter under CO. The lowestcobalt loading MCM-41 used (0.5 wt.%) produces just afew Co clusters in the sample large enough to growSWNT. The highest cobalt concentration MCM-41 used(4 wt.%) produces large cobalt particles, which lead tothe growth of graphite. If the cobalt is not initiallystrongly stabilized in the MCM-41 framework duringtemplate synthesis, the catalyst cannot produce SWNTwith narrow diameter distribution. Incorporation ofup to 3 wt.% Co-MCM-41 produced anchored smallclusters and SWNT with a narrow diameter distribution.This was accomplished with a major increase in theSWNT loading while maintaining high selectivity toSWNT. This is important as it provides incentives forthe practical, large scale application of this SWNT syn-thesis process to produce low defect and uniform diam-eter SWNT for electronic applications.

Acknowledgements

We thank the financial support from DoE-BES forthis project, and the use of the National SynchrotronLight Source at Brookhaven National Laboratory. Weare also grateful to Sang Nyon Kim and Professor Fo-tios Papadimitrakopoulos at University of Connecticutfor the access to the multi-excitation wavelength Ramaninstrument.

References

[1] Yu MF, Files BS, Arepalli S, Ruoff RS. Tensile loading of ropesof single wall carbon nanotubes and their mechanical properties.Phys Rev Lett 2000;84(24):5552–5.

[2] Hone J, Batlogg B, Benes Z, Johnson AT, Fischer JE. Quantizedphonon spectrum of single-wall carbon nanotubes. Science2000;289(5485):1730–3.

[3] Deheer WA, Chatelain A, Ugarte D. A carbon nanotube field—emission electron source. Science 1995;270(5239):1179–80.

[4] Frank S, Poncharal P, Wang ZL, de Heer WA. Carbon nanotubequantum resistors. Science 1998;280(5370):1744–6.

[5] Collins PC, Arnold MS, Avouris P. Engineering carbon nano-tubes and nanotube circuits using electrical breakdown. Science2001;292(5517):706–9.

[6] Hu JT, Odom TW, Lieber CM. Chemistry and physics in onedimension: synthesis and properties of nanowires and nanotubes.Acc Chem Res 1999;32(5):435–45.

[7] Dresselhaus MS, Dresselhaus G, Eklund PC. Science of fullerenesand carbon nanotubes. San Diego: Academic Press; 1996.

[8] Li Y, Kim W, Zhang Y, Rolandi M, Wang D, Dai H. Growth ofsingle-walled carbon nanotubes from discrete catalytic nanopar-ticles of various sizes. J Phys Chem B 2001;105(46):11424–31.

[9] Dai H, Rinzler AG, Nikolaev P, Thess A, Colbert DT, SmalleyRE. Single-wall nanotubes produced by metal-catalyzed dispro-portionation of carbon monoxide. Chem Phys Lett 1996;260(3,4):471–5.

[10] Cheung CL, Kurtz A, Park H, Lieber CM. Diameter-controlledsynthesis of carbon nanotubes. J Phys Chem B 2002;106(10):2429–33.

[11] Choi HC, Kim W, Wang D, Dai H. Delivery of catalytic metalspecies onto surfaces with dendrimer carriers for the synthesis ofcarbon nanotubes with narrow diameter distribution. J PhysChem B 2002;106(48):12361–5.

[12] Fu Q, Huang S, Liu J. Chemical vapor depositions of single-walled carbon nanotubes catalyzed by uniform Fe2O3 nanoclus-ters synthesized using diblock copolymer micelles. J Phys Chem B2004;108(20):6124–9.

[13] Li Y, Liu J, Wang Y, Wang ZL. Preparation of monodispersedFe–Mo nanoparticles as the catalyst for CVD synthesis of carbonnanotubes. Chem Mater 2001;13(3):1008–14.

[14] Kitiyanan B, Alvarez WE, Harwell JH, Resasco DE. Controlledproduction of single-wall carbon nanotubes by catalytic decom-position of CO on bimetallic Co–Mo catalysts. Chem Phys Lett2000;317(3,4,5):497–503.

[15] Chen Y, Ciuparu D, Lim S, Yang Y, Haller G, Pfefferle L.Synthesis of uniform diameter single wall carbon nanotubes inCo-MCM-41: effect of CO pressure and reaction time. J Catal2004;226(2):351–62.

[16] Chen Y, Ciuparu D, Lim S, Yang Y, Haller G, Pfefferle L.Synthesis of uniform diameter single wall carbon nanotubes inCo-MCM-41: effect of the catalysts prereduction and nanotubegrowth temperatures. J Catal 2004;225(2):453–65.

[17] Resasco DE, Alvarez WE, Pompeo F, Balzano L, Herrera JE,Kitiyanan B, et al. A scalable process for production of single-walled carbon nanotubes (SWNTs) by catalytic disproportion-ation of CO on a solid catalyst. J Nanopart Res 2002;4(1/2):131–6.

[18] Alvarez WE, Kitiyanan B, Borgna A, Resasco DE. Synergism ofCo and Mo in the catalytic production of single-wall carbonnanotubes by decomposition of CO. Carbon 2001;39(4):547–58.

[19] Herrera JE, Balzano L, Borgna A, Alvarez WE, Resasco DE.Relationship between the structure/composition of Co–Mo cata-lysts and their ability to produce single-walled carbon nanotubesby CO disproportionation. J Catal 2001;204(1):129–45.

[20] Herrera JE, Resasco DE. Role of Co–W interaction in theselective growth of single-walled carbon nanotubes from COdisproportionation. J Phys Chem B 2003;107(16):3738–46.

[21] Bachilo SM, Balzano L, Herrera JE, Pompeo F, Resasco DE,Weisman RB. Narrow (n,m)-distribution of single-walled carbonnanotubes grown using a solid supported catalyst. J Am ChemSoc 2003;125(37):11186–7.

[22] Lim S, Ciuparu D, Pak C, Dobek F, Chen Y, Harding D, et al.Synthesis and characterization of highly ordered Co-MCM-41 forproduction of aligned single walled carbon nanotubes (SWNT). JPhys Chem B 2003;107(40):11048–56.

78 Y. Chen et al. / Carbon 44 (2006) 67–78

[23] Ciuparu D, Chen Y, Lim S, Haller GL, Pfefferle L. uniform-diameter single-walled carbon nanotubes catalytically grown incobalt-incorporated MCM-41. J Phys Chem B 2004;108(2):503–7.

[24] Lim S, Ciuparu D, Chen Y, Yang Y, Pfefferle L, Haller GL. Porecurvature effect on the stability of Co-MCM-41 and the formationof size-controllable subnanometer Co clusters. J Phys Chem B2005;109(6):2285–94.

[25] Lim S, Yang Y, Ciuparu D, Wang C, Chen Y, Pfefferle L, et al.The effect of synthesis solution pH on the physicochemicalproperties of Co substituted MCM-41. Top Catal 2005;34(1–4).

[26] Stern EA, Newville M, Ravel B, Yacoby Y, Haskel D. TheUwxafs analysis package—philosophy and details. Physica B1995;209(1–4):117–20.

[27] Ankudinov AL, Ravel B, Rehr JJ, Conradson SD. Real-spacemultiple-scattering calculation and interpretation of X-ray-absorption near-edge structure. Phys Rev B 1998;58(12):7565–76.

[28] Alvarez WE, Pompeo F, Herrera JE, Balzano L, Resasco DE.Characterization of single-walled carbon nanotubes (SWNTs)produced by CO disproportionation on Co–Mo catalysts. ChemMater 2002;14(4):1853–8.

[29] Rinzler AG, Liu J, Dai H, Nikolaev P, Huffman CB, Rodriguez-Macias FJ, et al. Large-scale purification of single-wall carbonnanotubes: process, product, and characterization. Appl Phys A—Mater Sci Process 1998;67(1):29–37.

[30] Shi ZJ, Lian YF, Liao FH, Zhou XH, Gu ZN, Zhang YG, et al.Purification of single-wall carbon nanotubes. Solid State Commun1999;112(1):35–7.

[31] Rao AM, Richter E, Bandow S, Chase B, Eklund PC, WilliamsKA, et al. Diameter-selective Raman scattering from vibrationalmodes in carbon nanotubes. Science (Washington, DC)1997;275(5297):187–91.

[32] Dresselhaus MS, Eklund PC. Phonons in carbon nanotubes. AdvPhys 2000;49(6):705–814.

[33] Kataura H, Kumazawa Y, Maniwa Y, Umezu I, Suzuki S,Ohtsuka Y, et al. Optical properties of single-wall carbonnanotubes. Synth Metals 1999;103(1–3):2555–8.

[34] Weisman RB, Bachilo SM. Dependence of optical transitionenergies on structure for single-walled carbon nanotubes inaqueous suspension: an empirical Kataura plot. Nano Lett2003;3(9):1235–8.

[35] Sugano M, Kasuya A, Tohji K, Saito Y, Nishina Y. ResonanceRaman scattering and diameter-dependent electronic states insingle-wall carbon nanotubes. Chem Phys Lett 1998;292(4–6):575–9.

[36] O�Connell MJ, Bachilo SM, Huffman CB, Moore VC, StranoMS, Haroz EH, et al. Band gap fluorescence from individ-ual single-walled carbon nanotubes. Science 2002;297(5581):593–6.

[37] Reich S, Thomsen C, Ordejon P. Electronic band structure ofisolated and bundled carbon nanotubes. Phys Rev B: CondensMatter Mater Phys 2002;65(15):155411/1–155411/11.

[38] Strano MS, Moore VC, Miller MK, Allen MJ, Haroz EH, KittrellC, et al. The role of surfactant adsorption during ultrasonicationin the dispersion of single-walled carbon nanotubes. J NanosciNanotechnol 2003;3(1/2):81–6.

[39] Heller DA, Barone PW, Swanson JP, Mayrhofer RM, Strano MS.Using Raman spectroscopy to elucidate the aggregation state ofsingle-walled carbon nanotubes. J Phys Chem B 2004;108(22):6905–9.

[40] Ciuparu D, Chen Y, Lim S, Yang Y, Haller GL, Pfefferle L.Mechanism of cobalt cluster size control in Co-MCM-41 duringsingle-wall carbon nanotubes synthesis by CO disproportionation.J Phys Chem B 2004;108(40):15565–71.

[41] Haider P, Chen Y, Lim S, Haller GL, Pfefferle L, Ciuparu D.Application of the generalized 2D correlation analysis to dynamicnear-edge X-ray absorption spectroscopy data. J Am Chem Soc2005;127(6):1906–12.

[42] Ding F, Rosen A, Bolton K. Molecular dynamics study of thecatalyst particle size dependence on carbon nanotube growth. JChem Phys 2004;121(6):2775–9.