Acta mater.

48 (2000) 3803–3812
www.elsevier.com/locate/actamat

CARBON NANOTUBE–METAL–OXIDE NANOCOMPOSITES:

MICROSTRUCTURE, ELECTRICAL CONDUCTIVITY AND
MECHANICAL PROPERTIES
E. FLAHAUT1, A. PEIGNEY1†*, Ch. LAURENT1, Ch. MARLIÈRE2, F. CHASTEL1 and
A. ROUSSET1
1
CIRIMAT, UMR CNRS 5085/ LCMI, Centre Inter-universitaire de Recherche et d’Ingénierie des
Matériaux, Université Paul-Sabatier, F 31062 Toulouse, cedex 4, France, 2Laboratoire des Verres, UMR
CNRS 5587, Université Montpellier 2, C.C. 69, place Eugène Bataillon, F 34095 Montpellier, cedex 5,
France

( Received 20 February 2000; received in revised form 28 March 2000; accepted 28 April 2000 )

Abstract—Carbon nanotube–metal–oxide composites (metal=Fe, Co or Fe/Co alloy; oxide=Al2O3, MgO or

MgAl2O4) have been prepared by hot-pressing the corresponding composite powders, in which the carbon
nanotubes, mostly single or double-walled, are very homogeneously dispersed between the metal–oxide
grains. For the sake of comparison, ceramic and metal–oxide nanocomposites have also been prepared. The
microstructure of the specimens has been studied and discussed in relation to the nature of the matrix, the
electrical conductivity, the fracture strength and the fracture toughness. The carbon nanotube–metal–oxide
composites are electrical conductors owing to the percolation of the carbon nanotubes.  2000 Acta
Metallurgica Inc. Published by Elsevier Science Ltd. All rights reserved.

Résumé—Des composites nanotube de carbone–métal–oxyde (métal=Fe, Co ou alliage Fe/Co; oxyde=Al2O3,

MgO ou MgAl2O4) ont été préparés par frittage sous charge des poudres composites correspondantes, dans
lesquelles les nanotubes de carbone, essentiellement mono- ou bi-feuillets, sont dispersés de façon très homo-
gène entre les grains métal–oxyde. Des céramiques et des nanocomposites métal–oxyde ont aussi été préparés
pour permettre des comparaisons. La microstructure des échantillons a été étudiée et discutée en relation
avec la nature de la matrice, la conductivité électrique, la résistance à la rupture et la ténacité. Les composites
nanotube de carbone–métal–oxyde sont conducteurs électriques grâce à la percolation des nanotubes de car-
bone.  2000 Acta Metallurgica Inc. Published by Elsevier Science Ltd. All rights reserved.

Keywords: Hot pressing; Scanning electron microscopy (SEM); Composites; Electrical properties; Mechan-
ical properties

1. INTRODUCTION trical percolation threshold [3] and allows one to

obtain, with only very small amounts of CNTs, an
Amongst the works actually devoted to the possible
electrical conductivity sufficient to provide an electro-
applications of carbon nanotubes (CNTs), their incor-
static discharge [7].
poration in a polymer [1–12], metal [13, 14], or cer-
amic [15–18] matrix and the characterization of the CNT–SiC composites have been prepared by Ma et
microstructure and properties of these dense com- al. [15], who mixed large multiwall carbon nanotubes
posite materials are investigated. Indeed, it is pro- (MWNTs: 30–40 nm in diameter) with SiC powder
posed that owing to their remarkable mechanical and hot-pressed the mixture. These authors reported
properties [19–21], the CNTs could advantageously an improvement of about 10% over monolithic SiC
substituted for carbon fibers as reinforcing elements both in bending strength and fracture toughness, but
in composites. Other important particularities of the microstructure of the dense materials was not
CNTs are their very high aspect ratio and their met- fully investigated. Therefore, the contribution of the
allic or semiconducting character [22, 23]. Thus, their CNTs to the mechanical properties was not clearly
addition to a polymer matrix leads to a very low elec- established in this work. The present authors have
reported a novel catalytic route for the in situ forma-
tion, in a composite powder based on alumina, of sin-
* E-mail address: [email protected] (A. Peigney)
gle-wall carbon nanotubes (SWNTs) and thin
† To whom all correspondence should be addressed. Tel.: MWNTs [16, 24]. The microstructure and mechanical
+33-61-55-62-80; fax: +33-61-55-61-63. properties of dense CNT–Fe–Al2O3 materials pre-

1359-6454/00/$20.00  2000 Acta Metallurgica Inc. Published by Elsevier Science Ltd. All rights reserved.
PII: S 1 3 5 9 - 6 4 5 4 ( 0 0 ) 0 0 1 4 7 - 6
3804 FLAHAUT et al.: CARBON NANOTUBE–METAL–OXIDE NANOCOMPOSITES

pared by hot-pressing the nanocomposite powders prepared by heating up to 1000°C in a H2 atmosphere.

have been investigated [17, 18]. In comparison to For the sake of brevity, the powders are labeled
similar carbon-free nanocomposites, the relative den- CMA1, CMA2 (CNT–Fe–Al2O3), B, MB and CMB
sities are lower, the matrix grains are smaller and the (MgO, Co–MgO and CNT–Co–MgO, respectively)
fracture strength and fracture toughness are generally and D, MD and CMD (MgAl2O4, Co–MgAl2O4 and
markedly lower. Nevertheless, SEM observations of CNT–Co–MgAl2O4, respectively). Note that attrition-
composite fractures indicated that the CNT bundles, milling was conducted on the D powder and on the
which are very flexible, locally act to dissipate some oxide solid solutions used as precursors for the
of the fracture energy. Amongst the reasons which CMA2, MD and CMD composite powders. This
could explain the lack of mechanical reinforcement attrition-milling leads to a contamination of the corre-
at the macroscopic scale, we reported the too-low vol- sponding powders by some ZrO2 particles, about 1
ume fraction of CNTs and the presence of others µm in size.
species, notably large diameter (>15 nm) filamentous The powders were uniaxially hot-pressed at 43
carbon that we call nanofibres. It was also pointed out MPa in graphite dies, in a primary vacuum, at 1500°C
that the influence of the nature of the ceramic matrix (MgAl2O4 ceramic, Al2O3– and MgAl2O4–matrix
in which the CNTs are dispersed was to be investi- composites) or 1600°C (MgO ceramic and MgO–
gated. matrix composites) with a dwell time fixed at 15 min.
Working on the powder synthesis, the present The dense specimens (20 mm in diameter and 2 mm
authors have greatly enhanced both the quantity of thick) for mechanical tests were ground with diamond
CNTs and the quality of carbon (much less suspensions. Surfaces were polished to an optical fin-
nanofibers) in the CNT–Fe–Al2O3 powders [25, 26]. ish and thermal and/or chemical etching treatments
Furthermore, novel composite powders, CNT–Fe/Co– were adjusted to reveal the grain boundaries.
MgAl2O4 and CNT–Co–MgO, have been synthesized Starting powders, polished surfaces, etched sur-
[27–30]. The aims of this work are to prepare dense faces and fracture profiles of dense specimens were
CNT–Fe–Al2O3 composites containing higher quan- observed by scanning electron microscopy (SEM).
tities of CNTs and much less nanofibers than those The average grain size of the oxide (Gm) was determ-
previously investigated [17, 18] and novel dense ined by the linear intercept method [31]. The maximal
CNT–Fe/Co–MgAl2O4 and CNT–Co–MgO com- size of the metal and ZrO2 particles (dmetal and dZrO2,
posites. For the sake of comparison, ceramics respectively) were measured on back-scattered elec-
(MgAl2O4 and MgO) and composites without CNTs tron images of polished specimens. Relative densities
(Fe/Co–MgAl2O4 and Co–MgO) are also prepared. (d%) were calculated from measurements obtained by
The microstructure and mechanical properties of the Archimedes method, using the density of graphite
these dense materials are investigated. The electrical (dgraphite=2.25 g/cm3) for CNTs. The powders and the
conductivity at room temperature is also measured, hot-pressed materials were also studied by X-ray dif-
correlated to the apparent quantity of CNTs and com- fraction (XRD).
pared with values obtained by other authors for CNT– The transverse fracture strength (sf) was determ-
polymer composites. ined by the three-point-bending test on parallel-
epipedic specimens (1.6×1.6×18 mm3) machined with
a diamond blade. The fracture toughness (KIc) was
2. EXPERIMENTAL
measured by the SENB method on similar specimens
The preparation of oxide solid solution and CNT– notched using a diamond blade 0.3 mm in width. The
metal–oxide composite powders was described in pre- calibration factor proposed by Brown and Srawley
vious papers [24–29]. By heating up to 1000°C in a [32] was used to calculate the SENB toughness from
H2–CH4 atmosphere, monophased oxide solid sol- the experimental results. Cross-head speed was fixed
utions are submitted to a selective reduction which at 0.1 mm/min. The values given for sf and KIc are
leads to the formation of transition metal particles, a the average of measures on seven and six specimens,
few nanometers in diameter, both inside and at the respectively. The electrical conductivity of dense
surface of each grain of the powder. When they reach specimens was measured at room temperature with
the required size, the surface metal particles catalyze d.c. currents on parallelepipedic specimens
the in situ formation of SWNTs and thin MWNTs (1.6×1.6×8 mm3), parallel to their length, i.e. perpen-
which gather in extensively branched bundles and are dicular to the hot-pressing axis. The current densities
extremely well dispersed as a web-like network used were lower than 160 mA/cm2.
between the matrix agglomerates (Fig. 1a–f). In
addition, some carbon is also found in the form of
graphene layers wrapping the metal particles which 3. RESULTS AND DISCUSSION
have not catalyzed the formation of CNTs. Statistical 3.1. Powder characterization
studies on HREM images of individual CNTs have
shown than over 80% have only one or two walls, It is first important to note that the metal content
and that most diameters are smaller than 6 nm [26, has been chosen in order to optimize both the quality
30]. Carbon-free metal–oxide powders are similarly of carbon and the quantity of CNTs in the different
 FLAHAUT et al.: CARBON NANOTUBE–METAL–OXIDE NANOCOMPOSITES 3805

Fig. 1. SEM images of CNT–metal–oxide nanocomposite powders: (a) CNT–Fe–Al2O3 (CMA1); (b) CNT–
Fe–Al2O3 (CMA2); (c, d) CNT–Co–MgO (CMB); (e, f) CNT–Fe/Co–MgAl2O4 (CMD).

powders. Thus, depending on the nature of the matrix, some contamination from the ZrO2 balls, which is of
a specific metal content was used (Table 1). The XRD the order of a few wt%.
patterns (not shown) of the composite powders reveal Macroscopic parameters related to the presence of
the peaks of a-Fe, e-Co or cubic-Fe/Co alloy besides carbon are reported in Table 1. The carbon content
those of the involved matrix, a-Al2O3, MgO or (Cn) is slightly lower for CMA1 (4.8 wt%) than for
MgAl2O4, respectively. For CMA1 and CMA2, CMA2 (5.7 wt%), probably owing to the presence of
cementite Fe3C is also detected. In powders CMA2, more catalytic particles on the matrix grain surfaces
D, MD and CMD, which derive from attrition-milled for the latter powder. Indeed, CMA1 and CMA2 pow-
oxides, some of the different forms of ZrO2 (cubic, ders were prepared from the same oxide solid sol-
tetragonal and/or monoclinic) are detected, showing ution, except that it was attrition-milled prior to
3806 FLAHAUT et al.: CARBON NANOTUBE–METAL–OXIDE NANOCOMPOSITES

Table 1. Some characteristics of the metal–oxide and carbon nanotube–metal–oxide nanocomposite powders

Composite Label Metal content (wt%) Cn (wt%) ⌬S (m2/g) ⌬S/Cn (m2/g)

CNT–Fe–Al2O3 CMAl Fe: 8.38 4.8 13.8 288

CNT–Fe–Al2O3a CMA2 Fe: 8.38 5.7 15.1 265
Co–MgO MB Co: 13.47 - - -
CNT–Co–MgO CMB Co: 13.47 2.8 10.4 370
Fe/Co–MgAl2O4a MD Fe/Co: 9.66 - - -
CNT–Fe/Co–MgAl2O4a CMD Fe/Co: 9.66 4.9 10.3 210

Cn: carbon content; ⌬S: surface area of carbon in 1 g of composite powder, representing the quantity of nanotubes; ⌬S/Cn: specific surface area
of carbon, representing the quality of nanotubes.
a
Specimens prepared from powders attrition-milled before reduction. Note that for CNT-containing composites, the carbon content was not taken
into account in the calculation of the metal content.

reduction for CMA2, resulting in a finer agglomerate its precursor oxide powder. Traces of cementite are
size. The carbon content for CMD (Cn=4.9 wt%) is also detected in CMD. Secondly, in CNT-containing
similar to that measured for CMA1, but it is notably composites, a wide peak corresponding to the dis-
lower in the case of CMB (Cn=2.8 wt%). As proposed tance between graphene layers (d002=0.34 nm)
elsewhere [25, 26], the surface area of carbon found appears, which reflects a higher crystallization level
in 1 g of composite powder (⌬S — Table 1) is a good of the graphenic species than in the powders. Finally,
representation of the quantity of CNTs in the powder. in all dense composites, the width of the metal peaks
⌬S is about 1.5 times higher in Al2O3–matrix com- decreases, compared to the powders, reflecting the
posites than in MgO– and MgAl2O4–matrix com- growth of the intergranular metal particles upon hot-
posites. From Cn and ⌬S, one obtains the specific sur- pressing. However, this phenomenon is less marked
face area of the carbon (⌬S/Cn — Table 1) which in CNT–metal–oxide composites, probably because it
represents what we call the quality of carbon in the is inhibited by the presence of graphene layers cover-
powder. A higher figure for ⌬S/Cn denotes more car- ing most metal particles in the powders, as well as
bon in the form of CNTs, or CNTs with a smaller by the carburization in the case of Fe particles.
diameter and/or less walls, or bundles of CNTs with All the CNT-containing composites present
smaller diameter, i.e. made up of less CNTs unachieved densifications, with relative densities
[25,26,33]. The CMB powder presents the higher (d — Table 2) 5–8% below those of the correspond-
quality of carbon (⌬S/Cn=370 m2/g) whereas the ing carbon-free metal–oxide composites. The MgO
CMD powder presents the lower one (⌬S/Cn=210 ceramic also presents a fairly poor relative density.
m2/g). Some microstructural characteristics were determined
SEM images of the powders containing CNTs are from SEM observations of etched and unetched pol-
reported in Fig. 1. The web-like network of CNT ished surfaces (Table 2). In the case of Al2O3– and
bundles is clearly observed for all powders. As men- MgAl2O4–matrix composites, the matrix grain size
tioned above, the matrix agglomerates are much (Gm) is very small (0.3–0.8 µm) and the metal par-
larger for CMA1 (>10 µm — Fig. 1a) than for CMA2 ticles size (dmetal) is not higher than 0.5 µm. Both Gm
(⬍1 µm — Fig. 1b). The matrix agglomerates for and dmetal are higher for MgO–matrix composites (3–
CMB (Fig. 1c, d) are only of a few hundreds of nano- 7 µm and 1.5–2 µm, respectively). Note that previous
meters in size. The CNT bundles in this powder are works [34] on carbon-free metal–oxide nanocompos-
very thin. In the CMD powder (Fig. 1e, f), some ites have shown that the metal nanoparticles (⬍10
matrix agglomerates are almost 2 µm in size and the nm) located within the matrix grains in the powder
CNT bundles are larger than in CMB (Fig. 1d). These remain in intragranular position in the dense material
observations are in agreement with the values of and therefore are protected against excessive growth
⌬S/Cn reported above. upon hot-pressing. Thus, the metal–oxide and CNT–
metal–oxide specimens are hybrid
3.2. Microstructure of the hot-pressed specimens
micro/nanocomposites [35] containing both
Comparing with the XRD patterns recorded on the micrometer-sized metal particles which are at inter-
powders, the XRD patterns of the hot-pressed com- granular positions and nanometer-sized intragranular
posites show only a few minor differences. Firstly, metal particles.
dense CMA1 and CMA2 contain more cementite than The comparison of the matrix grain size for the
the corresponding powders. At a high temperature, MgAl2O4 ceramic (Gm=13.0 µm) and for the
the larger Fe surface particles probably react with the MgAl2O4–matrix composites (Gm=0.5–0.8 µm) shows
graphene layers which generally cover them that the matrix grain growth during hot-pressing is
[16,17,25,29] and some CNTs may also react with the hampered by the metal particles (inter- and
Fe particles. This phenomenon is more clearly intragranular) and by the CNTs. Moreover, the spe-
marked in CMA2 because the powder contains more cific role of the CNTs in the inhibition of matrix grain
Fe surface particles owing to the attrition-milling of growth is confirmed by the average matrix grain size
 FLAHAUT et al.: CARBON NANOTUBE–METAL–OXIDE NANOCOMPOSITES 3807

Table 2. Relative density and microstructural characteristics of the hot-pressed ceramics and nanocomposites

Ceramic or composite Label d (%) Gm (µm) dmetal (µm) dZrO2 (µm)

CNT–Fe–Al2O3 CMAl 88.7 0.3 ⱕ0.5 -

CNT–Fe–Al2O3a CMA2 87.5 0.3 ⱕ0.5 ⱕ1
MgO B 90.1 5 - -
Co–MgO MB 96.6 7 ⱕ2 -
CNT–Co–MgO CMB 92.9 3 ⱕ1.5 -
MgAl2O4a D 99.7 13 - ⱕ1
Fe/Co–MgAl2O4a MD 98.2 0.8 ⱕ0.5 ⱕ1
CNT–Fe/Co–MgAl2O4a CMD 90.6 0.5 ⱕ0.5 ⱕ1

d: relative density calculated by assuming that all carbon has the density of graphite, with dgraphite=2.25 g/cm3; Gm: average grain size of the
oxide; dmetal: diameter of the larger metal particles; dZrO2: diameter of the larger ZrO2 particles in specimens prepared from powders attrition-milled
before reduction (a).

in CMA1 and CMA2 (Gm=0.3 µm), which is much The presence of carbon deposits (B in Fig. 2c) points
smaller than that (ca. 2 µm) reported for Fe–Al2O3 towards a thermal instability of CNTs in the primary
nanocomposites [34]. This is in agreement with our vacuum at 1600°C rather than towards a chemical
previous results [17, 18]. However, this effect of reaction with MgO.
CNTs is markedly lower in the CMB composites For Al2O3– and MgAl2O4–matrix materials, a com-
(Table 2), probably because, as will be shown here- parison on SEM images of the quantities of CNTs
after, most CNTs are destroyed during hot-pressing. between the powders and dense specimen is difficult,
Furthermore, SEM observations also reveal that the because in the powders many superimposed CNTs
ZrO2 particles present as a contamination in the appear on a depth of several micrometers (three
CMA2, D, MD and CMD powders are found as dimensions), whereas for the hot-pressed specimen,
slightly submicronic particles in the hot-pressed only the CNTs at the fracture surface are observed
materials. In addition, it is observed that the CNT- (two dimensions). This can lead to an underestimation
containing composites present pores which are gener- of the CNTs quantity in the latter case. However, it
ally larger than the matrix grains and are sometimes seems that the quantity of CNTs is significantly
elongated. This is a consequence of the presence of decreased during the hot-pressing, particularly for
agglomerates in the powders, which are only partially Al2O3-based composites.
crushed during hot-pressing.
3.3. Electrical conductivity
The SEM observations of the fracture surface of
CNT–metal–oxide composites (Fig. 2a–d) show the The electrical conductivity (s — Table 3) of all
presence of CNT bundles in the materials but in very CNT-containing composites is in the range 0.2–4.0
different quantities from one sample to another. For S cm⫺1 whereas the ceramics and the metal–ceramic
CMA1 (Fig. 2a), a few long CNT bundles emerge composites are insulating (s⬍10⫺9 S cm⫺1). More-
from the fracture surface but most seem to be cut near over, from one specimen to another, the values are
the surface, perhaps after some pull-out. More very fairly well correlated to the quantity of CNTs
long CNT bundles appear for CMA2 (Fig. 2b), and observed on the fracture surfaces (Fig. 2a–d). Indeed,
much more for CMD (Fig. 2d). On the contrary, it is the lower value is obtained for CMB in which the
very difficult to observe even a few CNTs on the sur- majority of CNTs seems to have been destroyed (Fig.
face fracture of CMB (A in Fig. 2c). However, some 2c). Both CMA2 and CMD composites have been
deposits probably composed of carbon are also prepared from very fine powders, thus leading to a
observed (B in Fig. 2c). Thus, it clearly appears that better dispersion of the CNT bundles within the
many CNT bundles remain in CMA1, CMA2 and matrix (Fig. 1b, c) compared to CMA1 for which the
CMD, whereas most of the CNTs have been matrix agglomerates are much larger. This probably
destroyed during the hot-pressing of CMB. This could leads to a better connectivity of the CNTs in CMA2
be a consequence either of a specific chemical inter- and CMD dense composites than in CMA1, account-
action between the CNTs and the MgO matrix or of ing for the higher electrical conductivity of the former
the higher hot-pressing temperature used for the materials (1.5–4.0 vs 0.4–0.8 S cm⫺1). Furthermore,
MgO-based specimens (1600°C vs 1500°C for the the values are significantly higher for CMA2 than for
other materials), owing to the difficulty in densifying CMD because the quantity parameter (⌬S — Table
them. A high temperature chemical interaction 1) is 50% higher for CMA2 than for CMD.
between the CNTs and the residual gas in the furnace To assess the real role of the CNTs in the electrical
is also a possibility. We also hot-pressed a CNT–Co– conductivity of CMB, the MgO matrix was dissolved
MgO composite at a lower temperature (1200°C). The by a mild HCl treatment known to not damage the
densification of the material was very low (82%) but CNTs [30] and the so-obtained carbon species were
the quantity of CNT bundles was similar or higher observed by high resolution electron microscopy. In
than that observed for CMA2 or CMD (Fig. 2b, d). agreement with SEM observations (Fig. 2c), only a
3808 FLAHAUT et al.: CARBON NANOTUBE–METAL–OXIDE NANOCOMPOSITES

Fig. 2. SEM images of the fracture surface of ceramics and nanocomposites: (a) CNT–Fe–Al2O3 (CMA1); (b)
CNT–Fe–Al2O3 (CMA2); (c) CNT–Co–MgO (CMB); (d) CNT–Fe/Co–MgAl2O4 (CMD); (e) MgO (B); (f)
Co–MgO (MB); (g) MgAl2O4 (D); (h) Fe/Co–MgAl2O4 (MD).
 FLAHAUT et al.: CARBON NANOTUBE–METAL–OXIDE NANOCOMPOSITES 3809

Table 3. Electrical conductivity (s), fracture strength (sf), fracture toughness (KIc) and fracture mode of the hot-pressed ceramics and nanocomposites

Ceramic or composite Label or reference s (S cm⫺1) sf (MPa) KIc (MPa m1/2) Fracture mode

Al2O3 [34] - 335 4.4 Intergranular

Fe–Al2O3 [34] - 630 7.2 Mixed
CNT–Fe–Al2O3 [17] - 540 3.6 Mixed
CNT–Fe–Al2O3* [18] - 295 2.7 Intergranular
CNT–Fe–Al2O3 CMAl 0.4–0.8 400 5.0 Mixed
CNT–Fe–Al2O3* CMA2 2.8–4.0 296 3.1 Intergranular
MgO B n.m 202 - Intergranular
Co–MgO MB n.m. 283 - Mixed
CNT–Co–MgO CMB 0.2 254 - Intergranular
MgAl2O4* D n.m. 308 - Transgranular
Fe/Co–MgAl2O4* MD n.m. 212 2.94 Mixed
CNT–Fe/Co–MgAl2O4* CMD 1.5–1.8 221 1.71 Mixed

a
Specimens prepared from powders attrition-milled before reduction; n.m.: not measurable; mixed: the fracture presents both the inter- and
transgranular characters.

small proportion of undamaged CNTs were observed, The characteristics of CMA1 (Table 3) can be com-
the majority of carbon being found in the form of pared with those of the composite reported in Ref.
disorganized graphene layers, which probably reflects [17], both having been prepared from powders with
a damage to the CNTs during hot-pressing at 1600°C. micrometer-sized agglomerates: the fracture strength
Moreover, similar experiments have been conducted of CMA1 is lower but its fracture toughness is higher,
on a composite hot-pressed at 1700°C and have which could be correlated to some differences in the
revealed the transformation of all the CNTs into dis- microstructure, the densification is lower for CMA1
organized graphene layers. This material had a very (88.7% vs 91%) and the matrix grains are smaller (0.3
low electrical conductivity (s⬍10⫺8 S cm⫺1). These vs about 1 µm). The higher quantity of CNTs in
results show that, for the three categories of CNT– CMA1 seems to favor the inhibition of matrix grain
metal–oxide composites, including CMB in which the growth and to hamper the densification. Conse-
quantity of CNTs seems to be very low, the electrical quently, more elongated pores subsist in the present
conductivity can be attributed to a percolation composite, which could explain the lower value of sf
phenomenon between CNT bundles. The very high (Table 3). On the contrary, the increase in the quantity
aspect ratio of the CNT bundles (>104) makes the per- of CNTs in CMA1 may result in an easier transfer of
colation possible with a very small content of CNTs. the stress and thus could account for the significant
Similarly, Sandler et al. reported, for MWNTs–poly- increase in the fracture toughness (from 3.6 MPa m1/2
mer composites, an electrical conductivity equal to for [17] to 5.0 MPa m1/2 for CMA1).
10⫺2 S cm⫺1 with only 0.1 vol% of MWNTs [3] and It has been shown [34, 35] that the hybridization
Yoshino et al. measured 0.05–0.30 S cm⫺1 with 10 of microcomposites and nanocomposites could result
and 20 vol% of MWNTs, respectively [7]. We have in a further improvement in both the fracture strength
estimated that in the present starting powders, the and fracture toughness, which was associated with a
CNTs volume fraction could be up to about 20 vol% mixed inter/trans-granular fracture mode. The fracture
but it seems to be much lower in the dense materials strength and fracture toughness of CMA1 are higher
(perhaps only a few vol% and even less in CMB), than those of the Al2O3 ceramic prepared by the same
owing to damages during hot-pressing. In the present route [34], but are lower than those of Fe–Al2O3 com-
experiments, most CNTs have only one or two walls posites [34]. Taking into account that CMA1 is poorly
and are less than 6 nm in diameter, which is favorable densified (only 88.7% vs about 98% [34]), the mech-
to an enhanced conductivity, at identical volume frac- anical properties can be considered as not so poor and
tions, in comparison with large MWNTs. Thus, we a possible contribution of the CNTs to the reinforce-
infer that the difference in the characteristics of the ment is not to be ruled out. However, it is not
CNTs explains why we obtain a higher electrical con- clearly demonstrated.
ductivity than previously reported [3, 7]. Some differ- The characteristics of CMA2 (Table 3) are compa-
ences in the CNTs structure and the resulting electri- rable to those of the composite reported in Ref. [18],
cal behavior (metallic or semiconductor) could also both having been prepared from powders derived
account for these differences. from attrition-milled oxide precursors. Compared to
CMA1 and the composite in Ref. [17], it appears that
3.4. Mechanical properties and fracture surface
the attrition-milling is detrimental to sf and KIc in the
characteristics
resulting composites (Table 3). Firstly, less intragran-
Compared to previously studied CNT–Fe–Al2O3 ular Fe particles are obtained upon reduction in the
composites [17, 18], CMA1 and CMA2 differ both composite powders because the oxide agglomerates
by a higher quantity of CNTs in the starting powder in the precursor powders are submicronic, and thus
and a better quality of carbon (much less nanofibers). their proportion is also lower in the dense specimens.
3810 FLAHAUT et al.: CARBON NANOTUBE–METAL–OXIDE NANOCOMPOSITES

In these nanocomposites, the intragranular nanopart- 13 µm (Table 2). Both MgAl2O4–matrix composites
icles generally induce the transgranular fracture (MD and CMD) have a lower fracture strength than
mode, which contributes to the mechanical reinforce- the ceramic D (Table 3) despite a great refinement of
ment [34]. In contrast, for CMA2 and for the com- the matrix grain size (Gm — Table 2). However, their
posite in Ref. [18], the fracture mode is intergranular fracture mode remains partially transgranular (Fig.
probably in part because there are too few intragranu- 2d, h) owing to the metal nanoparticles, a strong pro-
lar Fe nanoparticles. Secondly, in the CMA2 powder portion of which are included inside the matrix grain
the CNT bundles are more intimately dispersed in the in these nanocomposites as reported by Quénard et
matrix than in CMA1 and in the dense CMA2 speci- al. [40]. The CMD composite, which is very incom-
men, the pores are smaller, less elongated but more pletely densified (90.6%), shows a fracture strength
numerous than in CMA1. Consequently, it is similar to that of MD but presents a much lower frac-
observed that in CMA2 (Fig. 2b), more CNT bundles ture toughness. Indeed, many CNTs appear on the
are located into the pores and are thus not constrained fracture surface of CMD (Fig. 2d) but most emerge
and damaged during the fracture. In contrast, a higher from pores, as in CMA1 (Fig. 2b) and do not seem
proportion of CNT bundles are included in the dense to have been constrained during the fracture. Only a
parts of CMA1 and SEM observations (Fig. 2a) reveal few CNTs, included in the densified parts of the
that these CNTs have been damaged during the frac- material, have been cut near the fracture surface.
ture. Thus if CNTs could contribute to any reinforce- Thus, we infer that in CMD, as in CMA2, most CNTs
ment in this kind of nanocomposite, they would be cannot be efficient for an eventual reinforcement of
efficient only in CMA1. Moreover, the presence in the material.
CMA2 of a small quantity of micrometric ZrO2 par- Some observations of the fracture surface of the
ticles, which usually reinforce the material, cannot CMA1 composite were conducted at a higher magni-
explain the low values of sf and KIc for CMA2. fication using a field-emission gun scanning electron
Despite an uncompleted densification (90.1%), the microscope (SEM-FEG — Fig. 3). Some CNT
fracture strength measured for the MgO ceramic bundles have been stretched upon hot-pressing,
(σf=202 MPa — Table 3) is comparable to the values acquiring the shape of the matrix grains at the grain
reported by Rice et al. [36] for non-annealed hot- boundary (Fig. 3a). Other CNTs have left their
pressed MgO with a similar average grain size (5 impression at the surface of the matrix grains (Fig.
µm). The comparison with MgO–matrix nanocompo- 3b) showing that some wetting of the CNTs by the
sites shows an evolution of sf similar to that between matrix occurs at the grain boundary during hot-
Al2O3 and the Al2O3–matrix composites (Table 3). pressing. In a previous paper [17], it was shown that
Indeed, sf is higher for Co–MgO (MB) than for MgO some CNTs embedded in the matrix grain were cut
(B) but slightly lower for CNT–Co–MgO (CMB) than after some pull-out. The same effect also seems to
for MB. Comparing B and MB, the improvement of occur with CNT bundles located at matrix grain
the densification (96.6% vs 90.1% — Table 2) may boundaries in the present composites (Fig. 3c, d).
be sufficient to explain the increase in sf, but it is Moreover, small holes can be observed (h in Fig. 3c),
also associated with a transition of the fracture mode which may have been the location of CNTs extracted
from purely intergranular to mixed inter/transgranular from a grain. It is well known that, in any fiber-
(Fig. 2e, f). This may be provoked by the presence reinforced composite, one of the keys to an efficient
of intragranular metal nanoparticles, as in the Al2O3– reinforcement is that the fiber–matrix interface is not
matrix nanocomposites [34]. The lower value of sf too weak but also not too strong [41], thus allowing
for CMB, compared to MB, associated with the tran- some energy absorption under stress by the deco-
sition to a fully intergranular fracture mode, could be hesion of these interfaces which often leads to the
correlated with its lower densification but it may also fiber pull-out.
result from the presence of the disordered graphene The SEM observations show: (i) that pull-out
sheets derived from damaged CNTs. The possible phenomena occur in CNT–Fe–Al2O3 composites; (ii)
contribution of non-damaged CNTs to the mechanical that CNT bundles may be wetted by the matrix at the
properties of these MgO–matrix nanocomposites are grain boundaries; or (iii) have been included inside
to be ruled out. the matrix grains during hot-pressing. These micro-
The fracture strength equal to 308 MPa obtained scopic observations show that some of the CNT
for the MgAl2O4 ceramic (D) is higher than the values bundles, probably weakly bonded to Al2O3, could act
previously reported by other authors [37, 38], prob- to reinforce the ceramic. However, some CNTs are
ably as a consequence of the high densification of the located in the pores and therefore cannot contribute
present specimen (99.7%) but also possibly owing to to the reinforcement of the material. Another
a beneficial effect of the micrometric tetragonal ZrO2 important point for achieving a reinforcement is to
particles [39], a contamination which occurred during have a sufficient volume fraction of tubular carbon.
the attrition-milling of the starting oxide powder. The This parameter seems to be rather high in the pow-
fracture mode in D is clearly transgranular (Fig. 2g) ders: it has been previously estimated to be of the
showing that for the present material, the critical grain order of 20 vol% [18] and could be higher in the
size related to the change of fracture mode is below present CNT–Fe–Al2O3 powders. However, on SEM
 FLAHAUT et al.: CARBON NANOTUBE–METAL–OXIDE NANOCOMPOSITES 3811

Fig. 3. SEM-FEG images of the fracture of the CNT–Fe–Al2O3 (CMA1) composite showing some aspects of
the CNT–matrix interactions.

images of the dense composites (Fig. 2a–d), the vol- prepared. The CNTs, mainly single or double-walled,
ume fractions seem to be much lower and SEM-FEG have grown in situ in the starting powders and thus
images furthermore show a lot of carbon deposits at are very homogeneously dispersed between the
the matrix grain junctions (Fig. 3e, f) arising from the metal–oxide grains.
damage of CNTs upon hot-pressing, producing dis- With the Al2O3 matrix, the increase of the quantity
ordered graphene sheets which gather at grain junc- of CNTs in the powder leads to a refinement of the
tions. microstructure of the hot-pressed specimen but with-
out a significant gain in the volume fraction of CNTs.
With both the Al2O3 and the MgAl2O4 matrix, a frac-
4. CONCLUSIONS
tion of the CNTs seems to be destroyed during the
CNT–Fe–Al2O3 composites have been prepared by hot-pressing at 1500°C. When using the MgO matrix,
hot-pressing composite powders where the quantity most CNTs are destroyed during a hot-pressing at
of CNTs has greatly been increased in comparison 1600°C, but the CNTs are not damaged if the treat-
with previous works. Novel dense CNT–Co–MgO ment is limited to 1200°C. It seems that the quantity
and CNT–FeCo–MgAl2O4 composites have also been of CNTs retained in the massive composite is more
3812 FLAHAUT et al.: CARBON NANOTUBE–METAL–OXIDE NANOCOMPOSITES

dependant of the treatment temperature than of the 10. Lourie, O. and Wagner, H. D., Appl. Phys. Lett., 1998,
nature of the oxide matrix. CNT damaging produces 72, 188.
11. Lourie, O. and Wagner, H. D., Composites Sci. Technol.,
disordered graphene layers which gather at matrix 1999, 59, 975.
grain junctions. 12. Files, B. S., Proceedings of the International Conference
Probably owing to a too-low relative density (87– on Integrated Nano/Microtechnology for Space Appli-
93%), the fracture strength and the fracture toughness cations, 1–6 November. Johnson Space Center Houston,
TX, 1998.
of the CNT-containing composites are generally 13. Kuzumaki, T., Miyazawa, K., Ichinose, H. and Ito, K., J.
lower than those of the carbon-free metal–oxide com- Mater. Res., 1998, 13, 2445.
posites and only marginally higher than those of the 14. Xu, C. L., Wei, B. Q., Ma, R. Z., Liang, J., Ma, X. K. and
ceramics. Microscopical observations show that some Wu, D. H., Carbon, 1999, 37, 855.
CNTs are trapped inside the matrix grains or at grain 15. Ma, R. Z., Wu, J., Wei, B. Q., Liang, J. and Wu, D. H.,
J. Mater. Sci., 1998, 33, 5243.
boundaries and seem to be wetted by the matrix in 16. Peigney, A., Laurent, Ch., Dumortier, O. and Rousset, A.,
the case of alumina. Most of these CNTs are cut near J. Eur. Ceram. Soc., 1998, 18, 1995.
the fracture surface after some pull-out and could 17. Laurent, Ch., Peigney, A., Dumortier, O. and Rousset, A.,
contribute to a mechanical reinforcement. However, J. Eur. Ceram. Soc., 1998, 18, 2005.
18. Peigney, A., Laurent, Ch., Flahaut, E. and Rousset, A.,
this is not demonstrated at a macroscopic scale. It is Ceram. Int., 2000, 26, 677.
necessary in future works to improve the preparation 19. Treacy, M. M. J., Ebbesen, T. W. and Gibson, J. M., Nat-
process to obtain composites with a higher ure, 1996, 381, 678.
densification and including a higher volume fraction 20. Ruoff, R. S. and Lorents, D. C., Carbon, 1995, 33, 925.
of CNTs. 21. Salvetat, J. P., Briggs, G. A. D., Bonard, J. M., Bacsa, R.
R., Kulik, A. J., Stockli, T., Burnham, N. A. and Forro,
Whereas the ceramics and metal–oxide nanocom- L., Phys. Rev. Lett., 1999, 82, 944.
posites are insulators, the carbon nanotube–metal– 22. Dai, H., Wong, E. W. and Lieber, C. M., Science, 1996,
oxide composites are electrical conductors with an 272, 523.
electrical conductivity in the range 0.2–4.0 S cm⫺1 23. Ebbesen, T. W., Lezec, H. J., Hiura, H., Benett, J. W.,
Ghaemi, H. F. and Thio, T., Nature, 1996, 382, 54.
owing to the percolation of the CNTs. The values of 24. Peigney, A., Laurent, Ch., Dobigeon, F. and Rousset, A.,
the electrical conductivity are fairly well correlated to J. Mater. Res., 1997, 12, 613.
the relative quantity of CNTs, the specimens becom- 25. Peigney, A., Laurent, Ch. and Rousset, A., J. Mater.
ing insulators when the CNTs are destroyed. The so- Chem., 1999, 9, 1167.
produced disordered graphene layers do not percolate. 26. Laurent, Ch., Peigney, A., Flahaut, E. and Rousset, A.,
Mater. Res. Bull., 2000, 35(5), in press.
For the first time, it has been shown that CNTs confer 27. Govindaraj, A., Flahaut, E., Laurent, Ch., Peigney, A.,
an electrical conductivity to ceramic–matrix com- Rousset, A. and Rao, C. N. R., J. Mater. Res., 1999, 14,
posites, which retain the mechanical properties of 2567.
the ceramic. 28. Flahaut, E., Govindaraj, A., Peigney, A., Laurent, Ch.,
Rousset, A. and Rao, C. N. R., Chem. Phys. Lett., 1999,
300, 236.
Acknowledgements—The authors would like to thank Mr L. 29. Coquay, P., De Grave, E., Vandenberghe, R.E., Dauwe,
Datas for his assistance in the HREM observations, which have C., Flahaut, E., Laurent, Ch., Peigney, A. and Rousset, A.,
been performed at the Service Commun de Microscopie Elec- Acta Mater., 2000, 48, 3015.
tronique à Transmission, Université Paul-Sabatier. 30. Flahaut, E., Peigney, A., Laurent, Ch. and Rousset, A., J.
Mater. Chem., 2000, 10, 249.
31. Mendelsohn, M. I., J. Am. Ceram. Soc., 1969, 52, 443.
REFERENCES 32. Brown, W. F. and Srawley, J. E., Plane Strain Crack
1. Ajayan, P. M., Stephan, O., Colliex, C. and Trauth, D., Toughness Testing of High Strength Metallic Materials,
Science, 1994, 265, 1212. ASTM Spec. Tech. Pub., 410. ASTM, Philadelphia, PA,
2. Shadler, L. S., Giannaris, S. C. and Ajayan, P. M., Appl. 1966.
Phys. Lett., 1998, 73, 3842. 33. Peigney, A., Laurent, Ch., Flahaut, E. and Rousset, A.,
3. Yoshino, K., Kajii, H., Araki, H., Sonoda, T., Take, H. and Carbon, 2000, accepted for publication.
Lee, S., Full. Sci. Technol., 1999, 7, 695. 34. Devaux, X., Laurent, Ch., Brieu, M. and Rousset, A., in
4. Musa, I., Baxendale, M., Amaratunga, G. A. J. and Eccles- Composite Materials, eds A. T. Di Benedetto, L. Nicolais
ton, W., Synth. Met., 1999, 102, 1250. and R. Watanabe. Elsevier Science Publishers BV, Amster-
5. Curran, S., Davey, A. P., Coleman, J., Dalton, A., McCar- dam, 1992, p. 209.
thy, B., Maier, S., Drury, A., Gray, D., Brennan, M., 35. Niihara, K., J. Ceram. Soc. Jpn., 1991, 99, 974.
Ryder, K., Lamy de la Chapelle, M., Journet, C., Bernier, 36. Rice, R. W., Br. Ceram. Soc. Proc., 1972, 20, 329.
P., Byrne, H. J., Caroll, D., Ajayan, P. M., Lefrant, S. and 37. Baudin, C., Martinez, R. and Pena, P., J. Am. Ceram. Soc.,
Blau, W., Synth. Met., 1999, 103, 2559. 1995, 78, 1857.
6. Bower, C., Rosen, R., Jin, L., Han, J. and Zhou, O., Appl. 38. Fujita, M., Yoshimatsu, H., Osaka, A. and Miura, Y., J.
Phys. Lett., 1999, 74, 3317. Ceram. Soc. Jpn. Int. Ed., 1994, 103, 81.
7. Sandler, J., Shaffer, M. S. P., Prasse, T., Bauhofer, W., 39. Quénard, O., Peigney, A., Laurent, Ch. and Rousset, A.,
Schulte, K. and Windle, A. H., Polymer, 1999, 40, 5967. Mater. Res. Bull., 2000, 35, in press.
8. Lourie, O., Wagner, H. D. and Levin, N., Polymer, 1997, 40. Quénard, O., Laurent, Ch., Peigney, A. and Rousset, A.,
38, 5699. Key Engng Mater., 1997, 132-136, 944.
9. Wagner, H. D., Lourie, O., Feldman, Y. and Tenne, R., 41. Després, J. -F. and Monthioux, M., J. Eur. Ceram. Soc.,
Appl. Phys. Lett., 1998, 72, 188. 1995, 15, 209.