RAMAN SPECTROSCOPY OF POLYMER–CARBON NANOMATERIAL COMPOSITES

INTRODUCTION

Elastomers often require, on account of their low modulus and strength, inclusion of reinforcing fillers in order to get considerable improvements in the final composite properties. Owing to their very small size and large surface area, carbon nanomaterials such as graphite nanoplatelets, carbon nanotubes, or graphene have attracted extensive interest in recent years as fillers for elastomeric matrices because at low filler loadings, they significantly improve properties compared with traditional composites filled with carbon black.

These carbon nanomaterials are now considered to be ideal fillers on account of their outstanding mechanical, electrical, and thermal properties, and they find extensive applications in advanced composites for industrial materials ranging from reinforced composites to electrical sensors.

The extent of property improvement imparted by fillers to a polymeric matrix has been shown to depend on several parameters including the size of the particles, their aspect ratio, their state of dispersion, and their surface chemical characteristics, which determine the interaction between the filler and the polymer chains and thus the interface of the polymer–filler system.¹ In conventional particles, it has been now well established that the surface characteristics of the filler and the chemically active sites present on the particle surface, as well as the chemical nature of the polymer, determine the filler–matrix interactions, which can be increased only if a uniform dispersion of the fillers is achieved. The presence of agglomerates, bundles, or stack of fillers negatively affects the rupture properties of the materials, since they act as crack initiators leading to brittle samples. On the other hand, the high aspect ratio of the carbon nanomaterials allows the formation of an interconnected filler network that has a strong impact on the mechanical properties on the composite material at a much lower filler loading than carbon blacks.

As a typical example, filling effects of multiwall carbon nanotubes (MWCNTs) and carbon black (CB) on a styrene–butadiene rubber (SBR) are compared through mechanical and electrical properties (Figure 1). Figure 1a, which displays the tensile behaviors of two (SBR) composites, reveals the superior reinforcing efficiency of carbon nanotubes. Unfortunately, the deformation at rupture is significantly reduced with regard to the sample filled with carbon black because of the presence of remaining agglomerated structures acting as failure points. This shows that one of the major problems of the dispersion of MWCNTs in polymeric media arises from van der Waals interactions between individual tubes leading to the formation of aggregates and agglomerates that reduce the expected property improvements. The transmission electron microscopy (TEM) image of a SBR composite displays nanotube bundles (Figure 2) despite the processing conditions using solution blending known to ensure good dispersion of the filler in the elastomeric matrix.

View Figure

• Download as Powerpoint
• Download Figure

View Figure

• Download as Powerpoint
• Download Figure

The influence of the filler network on the viscoelastic behavior of the elastomeric composites at small strains is shown through the strain dependence of the storage modulus G′, which exhibits a nonlinear behavior called the “Payne effect” in honor of the extensive studies made by Payne and his coworkers on this effect.² The strain dependence of the dynamic properties of filled vulcanizates has been extensively studied with a special focus on the test of quantitative models intended to describe the peculiar viscoelastic behavior of filled rubbers.^3,4 The drop in the elastic modulus has been mostly explained by a breakdown of the filler network originating from filler–filler interactions, but Wang⁵ has described a model based on a filler networking formed by direct contact between aggregates or via immobilized elastomeric layers surrounding the particle surface. In the latter case, the modulus would gradually decrease with increasing distance from the filler surface. In addition, rubber may be trapped in the filler network and released upon breakdown of the filler network by increasing strain amplitude. As seen in Figure 1b, at the same filler loading, the amplitude of the Payne effect is much more important for the composite filled with carbon nanotubes than for the one filled with carbon black because filler networking is formed at a lower filler loading of MWCNTs on account of the high aspect ratio of the tubes.

In addition, the SBR/MWCNT composite displays a stronger stress-softening than the SBR/CB one (Figure 1c). This stress-softening effect, or Mullins effect, observed at high extensions, is characterized by a lowering in the stress when the filled vulcanizate is extended a second time.^6,7 This stress-softening process, which is considered to be a hysteretic mechanism related to the energy dissipated by the material during deformation, has been associated with the rupture properties. Atomic-force microscopy under strain carried out by Lapra et al.⁸ on filled vulcanizates has been shown to bring invaluable information on the stress-softening effect. It was demonstrated that the strain field is highly heterogeneous, depending on the local concentration of filler, and that the strain undergone by elastomer chains can be very high locally, in the regions where distances between aggregates are very short. The overstraining effects make chains reach their limit of extensibility at low strains and detach from the filler surface causing the loss of elastic chains.⁹

Directly connected to the formation of a filler network is the electrical conductivity imparted by the black fillers to a host polymeric matrix usually considered to be an electrical insulator. The formation of an interconnecting filler network is characterized by a sharp drop, by several orders of magnitude, in the electrical resistivity of the composites reaching the so-called percolation threshold. It has been found that changes in electrical conductivity with strain amplitude are similar to the changes in dynamic moduli with strain amplitude.¹⁰ The electrical percolation threshold is achieved at a tiny amount of conductive particles (around 0.5 phr for rubber composites filled with multiwall carbon nanotubes) while a carbon black content as high as 10–50 phr, depending on the morphology of the carbon black, is required to make an electrically conductive polymer (Figure 1d).

For CB-filled rubber compounds, the reinforcement is mainly attributed to the formation of strong filler–rubber interactions.^11–13 However, in the case of carbon nanomaterials, despite the fact that there is a general agreement about the importance of the aspect ratio of the nanoparticles, the basic processes contributing to the mechanical behavior of the composites are not fully understood regarding essentially the interface of the polymer–filler system. Raman spectroscopy, which brings information at a molecular level, can further advance our understanding of composite behavior. After recalling the basic principles of Raman spectroscopy, some specific features of the Raman spectra of carbons will be discussed in order to demonstrate, in the last part of this paper, the potential of Raman spectroscopy for the analysis of carbon nanomaterials–based composites.

BASIC PRINCIPLES OF RAMAN SPECTROSCOPY

Infrared absorption and Raman scattering are molecular spectroscopies widely used to obtain information on polymeric systems from their vibrational properties. One major advantage of Raman scattering is that it allows the analysis of thick polymer samples, while only very thin films can be examined by infrared transmission spectroscopy, since infrared radiation is readily absorbed by functional groups of the polymer.

While infrared radiation arises from a direct resonance interaction between the frequency of the infrared incident radiation and that of a particular vibrational mode, the Raman effect is an inelastic scattering of light occurring upon irradiation of a molecule with a monochromatic light (usually a laser).

The major part of the light scattered from a molecule has almost the same frequency as that of the incident beam (elastic light scattering known as Rayleigh scattering), but a small amount of radiation is scattered with a shift in frequency from that of the original incoming light (inelastic light scattering known as Raman scattering). The frequency of the scattered light can be lower (Stokes Raman scattering) or higher (anti-Stokes Raman scattering) than that of the incident light, causing the molecule to gain or to lose vibrational energy. After being brought to a virtual state by the incident photon, the molecule can relax to excited vibrational levels so that the photon is scattered with lower energy causing the Stokes–Raman shift (Figure 3). If the transition to the virtual state starts from an excited vibrational state, then the molecule can return to the initial ground state and produce a scattered photon of higher energy than that of the incident light. This scattering, called anti-Stokes Raman scattering, is weaker than the Stokes process because a molecule is predominantly in its ground vibrational state at room temperature, and as a result the Stokes scattering is by far the most important part of the spectrum.

View Figure

• Download as Powerpoint
• Download Figure

The changes in frequency between the incident photons and the scattered ones are characteristic of the vibrational modes of the sample responsible for the scattering. The access to the frequency of these modes yields information on the chemical structure, symmetry, and environment of the molecule. Indeed, the activity of the vibrations is governed by the selection rules determined from the symmetry properties of the molecule (crystal) considered. Hence, only those vibrations that produce a variation of the dipole moment of the molecule will be active in infrared absorption spectroscopy, while only the vibrations that produce a variation of the polarizability of the molecule will be active in Raman scattering. So, both techniques provide complementary information: in particular, if the molecule possesses an inversion symmetry, only the vibrational modes symmetric with respect to the inversion operation can be Raman active, and only the antisymmetric ones can be infrared active.

Raman scattering is a weak effect since the Stokes lines are typically 10⁶ to 10⁷ times weaker than the Rayleigh scattered line. Lasers, which offer high-intensity monochromatic lights, have become the unique sources used for excitations in the near-UV, visible, or near-IR ranges. In particular, laser excitation in the near-IR is well suited to lower the fluorescent level, which often spoils the Raman signal.

With the advent of confocal microscopes, Raman spectroscopy has been launched to a new dimension of spectroscopic investigations in research laboratories, becoming an important tool in chemical analysis. A confocal microscope provides the opportunity to investigate different in-depth layers of a sample and to obtain three-dimensional images, provided that the sample is transparent to the incident and scattered lights. The in-depth spatial resolution (in the order of 1–2 μm) and optical slicing of a sample are provided by a pinhole, which eliminates signals outside the focal plane. Typically, backscattering geometry is employed in Raman microscopy, where the incident/scattered light is focused/collected by the same microscope objective, making it possible to measure the Raman spectrum from the sample surface without sample preparation with a lateral spatial resolution in the micrometer range, depending on the objective and on the incident wavelength used. Confocal Raman microscopy has been extensively applied in the depth profiling and imaging of polymer films, staked structures, and inclusions. Raman microimaging has been used to access the composition and morphology of polymer blends.^14,15 Zerda et al.¹⁶ also applied this technique to study the distributions of silica filler and elastomer domains in binary and ternary polymer blends. In this study, a near infrared laser was used to excite the Raman effect in order to reduce the fluorescence level and sample degradation, which significantly limit the applicability of the Raman microimaging technique.

The weak Raman scattering process may be enhanced by several orders of magnitude when the wavelength of the exciting laser coincides with or is close to an electronic absorption of the molecule. Nevertheless, the enhancement is not observed for all the vibrational modes, and this selective scattering enhancement can be particularly useful when chromophoric species experience resonance effects contrary to the surrounding environment. Porphyrins, carotenoids, and other classes of biological molecules show high levels of resonance enhancement. A particularly interesting case is that of carbon-based materials for which a double-resonance Raman scattering process explains the strong intensity enhancement and the excitation wavelength dependence of some bands of the spectrum.

Raman spectroscopy has become a key technique for the characterization of these materials, ranging from highly oriented pyrolytic graphite (HOPG), graphene, carbon nanotubes, pyrocarbons, and carbon black. Hundreds of contributions have been published on this subject (see, e.g., Ferrari and Basko¹⁷ and the cited references for a recent review). Furthermore, these materials have offered new opportunities for the synthesis of polymer composites that possess a strong potential for a wide range of applications. Because their Raman scattering is resonantly enhanced while most polymers do not display resonance effects, they give rise to strong well-defined bands even if they are used at very small amounts in the composite. For this reason, Raman spectroscopy provides unique insights into the intrinsic properties of carbon-based materials and their interaction with the surrounding environment as well as their mechanical reinforcing efficiency in polymer composites.

RAMAN SPECTROSCOPY OF CARBON NANOMATERIALS

Carbon has several allotropes—diamond and graphite are well known—and can exist in a wide range of disordered forms. Raman spectroscopy was found to be quite sensitive to these structures.

Graphites are formed of stacked parallel two-dimensional graphene sheets with sp² hybridized carbon atoms forming a network of regular hexagons. The π orbital, which is formed by the 2p_z orbitals of carbon and distributed over the graphene sheet, imparts to this nanostructure its electrical and thermal conductive character. Adjacent graphene sheets separated by a distance of 0.335 nm are held together by weak van der Waals forces, making graphite more affected by disorder along the axis perpendicular to the graphene planes. As for layered silicates, for the use of graphite as reinforcing nanofiller for polymers, the graphene sheets should be separated and randomly dispersed in the polymer matrix to provide nanoscale reinforcement.

The Raman spectrum of a highly oriented pyrolytic graphite (HOPG), which is a grown graphite with a quasi-perfect infinite ABAB stacking of graphene layers, only displays two phonon modes of E_2g symmetry at the center of the first Brillouin zone, which are active in first-order Raman scattering. The first one is located at about 43 cm⁻¹ and corresponds to anti-phase translational motions of successive layers perpendicular to their normal.¹⁷ It is hardly observable because of its proximity with the Rayleigh line. This mode does not exist in a single layer (1L) graphene. The second one, commonly called the G band, is located at 1581 cm⁻¹ (Figure 4) and corresponds to the zone center vibration of carbon atoms against each other in the layer planes.¹⁷ This is a common feature of all graphene and carbon graphitic materials. The Raman spectrum recorded under the excitation wavelength of 633 nm also exhibits bands at 2458, 2687, and 3246 cm⁻¹ (Figure 4a); they are assigned to second-order Raman scattering due to two-phonon processes.¹⁷

View Figure

• Download as Powerpoint
• Download Figure

Graphite nanoplatelets (GNPs) obtained by exfoliation of a flaky graphite and composed of a small number of lamellae can be considered to be multilayer graphenes. With regard to HOPG, GNPs (of grade 4827 from Asbury Carbons) display an additional band at 1324 cm⁻¹ (D band) and a high frequency shoulder of the G band located around 1604 cm⁻¹, called the D′ band¹⁷ (Figure 4b). These two bands reflect the presence of disorder in the graphitic lattice. The disorder can also be quantified by measuring the I_D/I_G ratio between the disorder-induced D band and the G band.^18–22 The Raman signals are very sensitive to the degree of structural disorder (defects), and for this reason Raman spectroscopy has become one of the most informative techniques for the characterization of disorder in sp² carbon materials. The disordered modes have been shown to appear because of double-resonance Raman processes.^{17,19,23–34}

The band at 2687 cm⁻¹ in HOPG, like that at 2650 cm⁻¹ in GNPs, is called the G′ band. It corresponds to the first overtone of the disorder-induced D band, and is observable even in the absence of defects. It results from double-resonant Raman scattering by two phonons. The analysis of the G′ band has been shown to yield information on the number of layers and on the stacking order in graphene systems.^17,34 A significant change in shape and intensity of the G′ band is observed from single graphene to graphite, moving from a single peak into the double structure of graphite in which the maximum intensity is shifted to a higher frequency compared with graphene as a result of interactions between the stacked graphene layers.

The Raman spectrum of a graphene monolayer contains the G and the G′ bands analogous to graphite and located, according to Dovbeshko et al.,³⁵ at 1580 and 2643 cm⁻¹, respectively, when excited at of 633 nm.

Rolling up a graphene layer into a seamless cylinder leads to single-wall carbon nanotubes (SWCNTs). A typical Raman spectrum of SWCNTs taken from Kao and Young³⁶ is shown in Figure 5. The G mode in SWCNTs gives rise to a multi-peak feature due to the symmetry breaking when the graphene sheet is rolled.³³ Besides the presence of the D and G′ bands, the Raman spectrum of SWCNTs also displays several peaks around 180 cm⁻¹ known as the “radial breathing modes” (RBMs), in which all the C atoms are vibrating in the radial direction with the same phase (totally symmetric A mode), as if the tube was breathing.³³ This mode provides information on the nanotube diameter.

View Figure

• Download as Powerpoint
• Download Figure

Multiwall carbon nanotubes (MWCNTs) composed of multiple concentric graphite cylinders have Raman spectra close to those of defective carbon materials with a strong D band, indicating more disorder in the multilayer configuration (Figure 4c).

The first-order spectrum of a poorly graphitized material like carbon black (CB) consists of two broad bands centered at 1588 cm⁻¹ (including the unresolved G and D′ bands) and at 1332 cm⁻¹, including the D band and two other components (D₃ and D₄ bands) revealed by the deconvolution (Figure 6). The D₃ band has been assigned to amorphous carbon,^21,22,37,38 while the D₄ band has been attributed to hydrocarbon components or aliphatic moieties grafted onto the basic structural units.³⁸ A revisited interpretation of the D₄ band based on a study of a panel of selected graphitized and partially graphitized carbons has been suggested recently.³⁹

View Figure

• Download as Powerpoint
• Download Figure

Raman spectroscopy has also been used to investigate the changes in the morphology and structure of CB particles occurring during heat treatment.^37,38 Zerda et al.³⁷ report that the intensity of the amorphous Raman peak gradually decreases during the heating process (up to 3000 K) thus suggesting some organization of amorphous carbon into graphitic layers. Interestingly, the authors also show that SBR filled with untreated carbon exhibit a higher modulus than those filled with graphitized carbon black, while an opposite effect is observed for low-strain dynamic properties. The increase in the loss modulus with the heat treatment is explained by increased agglomeration between carbon black aggregates with larger crystallites. Later, Zerda and Gruber⁴⁰ focused on the kinetic processes that cause graphitization. The analysis of the Raman spectra of laser-heated carbon blacks by using a continuous wave laser heating and pulse laser heating shows that two ordering mechanisms of different reaction rates are present during the heat treatment. A Raman characterization of five CB grades carried out by Pawlyta et al.,³⁸ shows that after heat treatment (up to 2600 °C), only a partial graphitization takes place, and the graphitizability is limited by the diameter of the primary particles. Donnet et al.⁴¹ also use Raman spectroscopy to relate the difference between iodine number and nitrogen area determinations for carbon blacks to the relative importance of the disordered part of the carbon surface. A disorder index, D, defined by D = A_D/(A_G + A_D), where A_D and A_G are the intensities of the D and G bands, respectively, was determined for milled graphite samples used as model material for the more complex carbon black surface. In a study devoted to the surface activity and chemistry of thermal carbon blacks, Darmstadt et al.,⁴² combine various techniques, including Raman spectroscopy, to probe different portions of the surface and bulk of the CB particles.

FACTORS THAT INFLUENCE THE SPECTRAL FEATURES OF THE RAMAN BANDS OF CARBON MATERIALS

Pressure, temperature, strain, laser excitation energy, and orientation are known to affect the spectral features of carbon-based materials.

Temperature

A change in temperature has also been shown to strongly influence the vibrational properties of carbon materials. Huong et al.⁵⁰ report a lowering of the stretching frequencies of MWCNTs with an increase in temperature ascribed to a lengthening of the C–C distances, the shift being perfectly reversible. In a paper dealing with the intrinsic temperature effect of Raman spectra of double-walled carbon nanotubes (DWCNTs), Zhou et al.⁵¹ report downshift of all Raman bands with the increased temperature. According to the authors, the G band of DWCNTs and SWCNTs have both a nonlinear and second-order polynomial relation with the increased temperature ascribed to the contribution of the third- and fourth-order anharmonic term in the lattice potential energy with pure temperature effect. In another paper of the same group,⁵² Raman spectra of SWCNTs have been investigated in the temperature range of 100–600 K. The intensity of RBM is seen to exhibit a sharp increase at first and then a gradual decrease as the temperature changes from 200 to 600 K, explained by a temperature-dependent density of electronic states of the nanotube.

Laser irradiation can lead to a significant sample heating during the measurement process, resulting in an increase in the local temperature if heat dissipation is not efficient. The laser-heating effect on Raman spectra of SWCNTs was investigated by examining variations of individual suspended tubes with different laser power levels.⁵³ It was shown that the laser-heating effect is much more notable for suspended SWCNTs in contrast to nanotubes sitting on substrate, which is ascribed to the good absorption and inefficient heat dissipation. The substrate is thought to act as a heat sink, and the temperature of the sample does not increase if using a laser with moderate power density (such as 1 mW/1 μm²). Zhang et al.⁵⁴ have shown that the frequencies of the G peaks of SWCNTs decrease with increasing laser power density, and the change is reversible thus indicating that the tubes are not damaged during the irradiation process. But the authors also observe for unpurified nanotubes that the I_D/I_G decreases rapidly in the initial increase in laser power, then remains basically unchanged in the subsequent decrease of laser power as a result of a loss of impurities. In contrast, I_D/I_G is constant throughout the entire process of increase and subsequent decrease of laser power in the case of the purified SWCNT sample. Huang et al.⁵⁵ have studied the temperature effect induced by laser heating of carbon nanotubes, active carbon, and HOPG. A similar temperature-dependent effect was observed for CNTs and active carbon with a frequency dependence on the temperature (calculated using the Stokes and anti-Stokes lines at 120 cm⁻¹) that is not the same for the different peaks: the slope of the G′ peak is about twice that of the D peak, which is less than that of the G peak (Figure 7). The authors do not find laser-induced shifts in the HOPG sample, and this lack of dependence has been attributed to its high thermal conductivity. Further studies of the same group show that the elevated temperature of the SWCNTs and MWCNTs induced by laser heating is due to the presence of impurities, defects, and disorder that reduce the thermal conductivity of the samples.⁵⁶ In contrast, the high conductivity of HOPG and diamond allows the samples to dissipate laser-induced heating effectively.

View Figure

• Download as Powerpoint
• Download Figure

Analysis of heating effects induced by the laser power, carried out on pure MWCNTs and on MWCNTs embedded in a styrene–butadiene rubber (SBR), confirms in both cases reversible laser-induced frequency shifts. However, above a laser power of 1 mW the intensity of the D band of the nanotubes incorporated in the polymer matrix becomes lower than that of the G band (Figure 8a), causing a strong decrease of the I_D/I_G ratio (Figure 8b), which remains almost unchanged in the subsequent increase laser power and also in the decreasing laser power cycle.⁵⁷ This result, not observed in non-embedded MWCNTs, but quite similar to that obtained by Zhang et al.⁵⁴ for SWCNTs, has been attributed to a purification (irreversible graphitization) of carbon nanotubes due to overheating effects probably more important in the composite on account of the low thermal conductivity of the surrounding matrix. The rubber matrix is considered to be a thermal insulator, and its low thermal conductivity (0.120 W/m·K) is only enhanced by 70% by addition of 10 phr of MWCNTs⁵⁸ despite the high thermal conductivity of individual nanotubes, around 6000 W/(m·K) for SWCNTs and 3000 W/(m·K) for MWCNTs.⁵⁹

View Figure

• Download as Powerpoint
• Download Figure

Laser excitation energy

Another prominent feature in the Raman spectra of carbon-based materials is the dispersive behavior of the D and G′ bands, that is, their frequency changes with the incoming laser excitation frequency (Figure 9). Both bands are seen to upshift with increasing photon energy in a linear way over a wide laser energy range, with a slope around 50 cm⁻¹/eV for the D band and around 100 cm⁻¹/eV for the G′ band (about twice that of the D band) (Figure 10).

View Figure

• Download as Powerpoint
• Download Figure

View Figure

• Download as Powerpoint
• Download Figure

These peculiarities of the Raman spectra of carbon-based materials arise from their electronic properties and the mutual interaction between the electronic cloud and the carbon nuclei motions (phonons). The single one-atom-thick sheet of carbon atoms arranged in a honeycomb network in monolayer (1 L) graphene may be the basic unit of all other allotropes. In the graphene network, each atom is bonded to its three neighbors by strong covalent sp² bonds constructed from hybridization of the 2s, 2p_x, and 2p_y atomic orbitals. The overlap of the p_z orbitals forms the bonding (π) and antibonding (π*) orbitals known as the electronic valence and conduction bands. The long-range π-conjugation imparts to graphene its amazing electrical, mechanical, and thermal properties.

Figure 11 shows the electronic band structure of graphene in the first Brillouin zone. The valence and conduction bands touch at discrete points located at the corners of the first Brillouin zone, referred to as the “Dirac” points or K points.⁶⁶ The π–π* band dispersions are approximately linear around the K points, thus forming double cones as shown in Figure 11. So, optical transitions induced by incoming photons with energies in the range of ∼1 to 4 eV may occur at different points inside the first Brillouin zone around the K point, at wave vectors k depending on the incoming photon energy. Increasing photon energy makes the π–π* transition k vector move away from the K point inside the first Brillouin zone.

View Figure

• Download as Powerpoint
• Download Figure

In a simple resonance process, an electron is excited from the filled π band to the empty π* band and is scattered back to the initial state with the creation of a phonon and emission of a Raman (Stokes) radiation. This process can be Raman active only if the total momentum transfer is Δk = 0, which implies that the phonon modes concerned here are only the zone center modes (at the Γ point), that is, with wave vector q_ph = 0, if we neglect the wave vector of the incident photon, which is extremely small compared with the dimensions of the first Brillouin zone. This is indeed the case for the G band, which is a common feature of all graphitic carbons and whose frequency is independent of the excitation energy.

The second-order double-resonance processes, involving coupling of electrons and phonons away from the center of the Brillouin zone, have been widely discussed in the literature.^{17,19,23–34} In particular, in the case of one-phonon scattering (D and D′ bands), these processes are activated by the presence of structural defects (disorder) of different kinds, thus creating holes in the electronic bands. Examples of double-resonant inter-valley scattering processes are shown in Figure 12. An electron in the ground state absorbs the energy of an incident photon of energy ħω_L, resonant with a π–π* transition, promoting this electron from the valence band into the conduction band. Then, it is inelastically scattered by a phonon of energy ħω_ph and wave vector q_ph ≠ 0 and decays to a hole in the valence band by generating a photon at a frequency ħω_Sc, lower than that of the incident laser light in the Stokes process, which is the most probable. Finally, the electron is elastically scattered back to the initial state, so that the momentum transfer of the whole process is Δk = 0, as in the simple resonance. Figure 13 shows that intra-valley resonance concerns scattering within the same cone around a given K point, whereas inter-valley resonance concerns scattering between two inequivalent neighboring K and K′ points. It follows that the modulus of q_ph involved in inter-valley processes is much greater than that involved in intra-valley processes. Indeed, the D band corresponds to a TO mode situated close to the Brillouin zone boundary K point, whereas the D′ band is assigned to a phonon of the same phonon branch in the vicinity of the zone center (Figure 13a). The modulus of q_ph also changes with the energy of the incident phonon, as do the positions of the D and D′ phonon modes on the corresponding dispersion curve: the highest is the incident photon energy, the more q_ph is displaced away from the K point (D band) or from the Γ point (D′ band). As shown in Figure 13b, in agreement with Raman data, the D band is located on a dispersive part of the phonon branch (strong dispersion of the D band vs the energy of the incident photon), whereas the D′ band corresponds to a rather flat part of this branch (no detectable dispersion). It is worth noting that, in addition to the rules of conservation of energy and momentum, the Raman activity is further governed by selection rules related to the symmetry properties of the electronic states and of the phonon branches.⁶⁷ Of course, all these conditions are fulfilled in the case of the G, D, and G′ bands. Note also that similar schemes have been established for anti-Stokes Raman scattering processes.

View Figure

• Download as Powerpoint
• Download Figure

View Figure

• Download as Powerpoint
• Download Figure

In the absence of resonance, any two-phonon Raman scattering can be active, provided that the wave vectors of the two phonons involved in the process have the same modulus and opposite directions, whatever their modulus and direction. So, two-phonon Raman spectra are generally related to a weighted two-phonon density of states, resulting in the observation of broad features. Obviously, the shape of the two-phonon Raman spectra of HOPG, showing well-defined structures, does not fit with such a picture. The effect of double resonance, in this case, is to select one particular (resonant) D or D′ phonon in their dispersion branch, determined by the energy of the incoming photon, and to enhance the contribution of their harmonics, even though their activity is not conditioned by the presence of structural defects (Figure 12b). It follows that the 2D (G′) band displays a dispersion twice as large as that of the D band and that no dispersion is observed for 2D′, as is the case for D′.

Now, the splitting of the G′ band in bilayer, trilayer, and multilayer graphenes and HOPG has been related to splittings of the electronic π and π* bands into two or more components, which are effective in highly ordered carbon samples because of well-defined interactions between layers with ordered staking.^17,34 In defective materials, such as MWCTs, pyrocarbons, or carbon blacks, the electronic band structures are somehow relaxed (spread out), resulting in broadened π and π* bands. Hence, the q_ph “selection rule” versus the incident photon energy, which is effective for both one-phonon and two-phonon Raman scattering, is more or less relaxed, leading to a progressive broadening of the whole (one- and two-phonon) Raman spectra with the increasing disorder of carbon structures. In particular, as shown in Figure 4, the doublet structure of the G′ band as observed in HOPG evolves to a single broad band in MWCTs, while at the same time the D band intensity of MWCTs is strongly enhanced due to increasing disorder.

Orientation

Because of the quasi-one-dimensional structure of carbon nanotubes or the two-dimensional geometry of graphene, the anisotropic fillers are expected to be oriented in a polymer matrix upon application of an uniaxial deformation. This orientating capability is of great importance because it affects the properties of the composite.

The degree of alignment of SWCNTs in a melt-spun composite fiber of 1 wt.% SWCNTs in polymethylmethacrylate was determined by polarized Raman spectroscopy.⁶⁸ The authors observe that the spectra recorded with the analyzer axis parallel to the polarization axis of the incident light strongly depend on the angle between the nanotube axis and the incident excitation polarization, all Raman peaks decreasing gradually in intensity when the angle increases from 0 to 90°. The Raman intensity is observed to be proportional to cos⁴ θ, where θ is the angle between the SWCNT axis and the incident excitation polarization. This orientation dependence in the Raman scattering is ascribed to the resonance enhancement in the Raman spectra of SWCNTs.

Orientation of SWCNTs under strain in a silicone rubber has also been quantified by Raman spectroscopy.⁶⁹ The authors have used the method of Hwang et al.⁶⁸ to evaluate the degree of alignment from the Raman data. The results show that the degree of orientation increases upon application of a tensile strain and by 200% strain; the data are close to cos⁴ θ. Other papers report that the Raman scattering intensity of the D band decreases monotically with an increase in the angle between draw direction and the polarization direction of the incident light in SWCNT/composites.^70–72

Wood et al.⁷³ have also used Raman spectroscopy to measure the degree of alignment of the tubes in the polymer that can be aligned by flowing the polymer prior to curing or by stretching the specimen in simple tension.

Other studies dealing with the orientation of multiwall carbon nanotube composites show that the D band is the most sensitive Raman peak to the nanotube alignment.^74,75 Its intensity has been shown to be more reduced than that of the G mode when the sample is oriented perpendicular to polarized light, leading to D_///D_⊥ ratios higher than G_///G_⊥ ones upon alignment.⁷⁵ The same conclusion has been drawn in the analysis of Raman spectra of polystyrene filled with MWCNTs where the composite was drawn above the glass-transition temperature of the polymer at a draw ratio of 4 (Figure 14) then brought quickly to a temperature below Tg in order to freeze the state of orientation.⁷⁶

View Figure

• Download as Powerpoint
• Download Figure

CHARACTERIZATION OF THE POLYMER–FILLER INTERFACE

The structural characteristics of carbon nanomaterials such as their superior mechanical properties, high aspect ratio, and surface area, directly influence the efficiency of reinforcement when incorporated into polymer matrices. In addition, the polymer–filler interactions are believed to play an important role in the overall properties of the nanocomposites. In a comprehensive review of the interaction studies in carbon nanotube–polymer nanocomposites, Rahmat and Hubert⁷⁷ describe experimental approaches of interaction measurements such as wetting, probe microscopy techniques, and spectroscopy including Raman scattering.

As shown before, the Raman band frequencies decrease upon application of a uniaxial deformation as a result of an increase in the interatomic distance. By embedding carbon materials in a polymeric matrix, a change in the C–C bond and thus a shift in the Raman peaks can only occur if the nanotubes carry the strain. In other words, downshifts of the Raman shifts with applied composite strain are expected to depend on the efficiency of load transfer from the matrix to the nanotubes, which is linked to the quality of the polymer–filler interface. The reinforcement of an epoxy resin by SWCNTs and MWCNTs has been characterized by following nanotube deformation from stress-induced Raman band shifts.⁴⁷ In all cases the G′ Raman peak position moves to a lower wavenumber with tensile strain indicating that the macroscopic stress on the composite deforms the nanotubes but the rate of band shift differs depending on the state of filler dispersion in the resin and on the adhesion of the nanotube surface to the matrix.⁴⁷

Schadler et al.⁷⁸ have analyzed the behavior of MWCNT/epoxy composites both in tension and compression. It was found that the Raman peak position of the G′ band shifts significantly under compression but not in tension, indicating a poorer load transfer behavior of nanotubes in tension compared with that in compression. It was suggested that during load transfer to MWCNTs, only the outer layers are stressed in tension, whereas all the layers respond in compression (Figure 15). In a subsequent study dealing with load transfer in epoxy composites filled with SWCNTs, the same group observe a shift of 15 cm⁻¹ of the G′ band due to hydrostatic compression on the tube bundles, which is explained by a decrease of the inter-tube spacing.⁷⁹ But no shift is observed in compression, and only a slight trend to lower wavenumbers is seen in tension, suggesting that nanotubes slip within the bundles.

View Figure

• Download as Powerpoint
• Download Figure

The deformation of nanocomposites containing graphene flakes with different numbers of layers has been investigated by Gong et al.⁸⁰ It was shown that the rate of band shift of the 2D (or G′) band tends to decrease as the number of graphene layers increases. While good stress transfer has been found between a polymer matrix and monolayer and bilayer graphene, less efficient stress transfer is seen to take place for trilayer and many-layer material because of slippage between the internal graphene layers, implying lower effective Young modulus in polymer-based nanocomposites.⁸¹

For silicone filled with SWCNTs, Frogley et al.⁸² have shown that the wavenumber of the G′ band decreases linearly with strain at first and tends to a constant value by about 50% strain. The shift in the linear part of the curve is only 2 cm⁻¹ over 50% strain, which is much less than 7–18 cm⁻¹ shift after 1% strain for SWCNTs observed in stiffer polymers. The authors attribute the nonlinear behavior to a change in stress-transfer efficiency; the constant value of the Raman wavenumber implies that the stress is no longer transferred effectively to the tubes.

Upon stretching SBR filled with five parts of nanotubes uniaxially (Figure 16), no significant shift of the G′ band is observed despite the high deformation of the elastomeric composite, which means that load transfer to the nanotubes is negligible. This conclusion is consistent with that reached by Nah et al.,⁸³ who have shown that the amount of bound rubber in natural rubber filled MWCNTs is much lower with carbon nanotubes than with carbon black. Since the amount of bound rubber reflects the level of filler–rubber interaction, the authors deduce that interfacial interactions between MWCNTs are weak in comparison with the relatively strong adhesion between rubber and carbon black.

View Figure

• Download as Powerpoint
• Download Figure

In recent papers,^57,84–86 it has been shown that the dispersive behavior of the carbon filler in the polymer matrix, that is, the Raman spectra change with the energy of the incident laser, can be used to get information on the polymer–filler interface. Figure 17 shows as a typical example for the case of a poly(dimethylsiloxane) (PDMS)/GNPs composite that a dispersive behavior is still observed when the carbon nanomaterial is embedded in the polymeric medium. Actually, one would expect that an interaction of the filler surface with polymer chains will alter the resonance window. Figure 18a shows that the D band dispersion of MWCNTs embedded in a SBR matrix is quite similar to that of pure MWCNTs, indicating a weak polymer–filler interface, while that of carbon black is higher in the polymer than in the pure state (Figure 18b). No change in the dispersive behavior is found when GNPs are incorporated in silicone rubber (Figure 18c), but a higher slope is observed in the case of MWCNTs with regard to that obtained for isolated MWCNTs (Figure 18d), which is attributed to a wrapping of the PDMS chains around the tube surface, allowing CH–π interactions between the methyl groups of PDMS and the π electrons of carbon nanotubes.⁸⁷ In addition, Figure 18a and 18c show that, for each excitation wavelength, the wavenumber of the D band of the filler in the composite is higher than that in the pure filler. As discussed in the section related to the pressure effect, the upshift of the D band can be explained by compressive forces exerted by polymer chains on the filler surface.

View Figure

• Download as Powerpoint
• Download Figure

View Figure

• Download as Powerpoint
• Download Figure

From the Raman analysis, it appears that the higher reinforcing efficiency of hydrocarbon rubbers by carbon nanomaterials, as depicted in Figure 1, cannot be ascribed to interfacial bonding between the filler and the matrix. The crucial factors responsible for the considerable improvements in stiffness are, without any doubt, the high aspect ratio and the orienting capability of these anisotropic carbon structures that allow, if well dispersed, the formation of an interconnecting filler network, providing mechanical reinforcement and electrical conduction at a vey low filler loading. The results obtained for the silicone composite filled with carbon nanotubes are more likely attributed to the high flexibility of the PDMS chains that wrap the tube surface, allowing a strong filler–matrix interface. This strong interface, combined with the attributes of MWCNTS—high aspect ratio and large specific surface area—explain the high levels of mechanical reinforcement imparted to the elastomeric matrix by tiny amounts of carbon nanotubes.^88,89

There are some reports in the literature on the mechanical properties of elastomeric composites based on GNPs. Araby et al.⁹⁰ have shown that, in an ethylene–propylene–diene monomer rubber (EPDM), the reinforcing effects of GNPs are lower than those of MWCNTs at the same filler loading. In a second paper, Araby et al.,⁹¹ derive, from a fit of their experimental data to micromechanical models, an aspect ratio of 7 for GNPs in SBR/GNPs composites, which is rather low with regard to the dimensions of GNPs and could be due to a clustering of the filler particles in the polymer matrix. Galimberti et al.^92,93 report investigations carried out on a poly(1,4-cis-isoprene) filled with a nanographite (or GNPs). It is shown that the composite containing 12 phr of GNPs displays similar stress values to those exhibited by the composite filled with a same amount of CB, the highest values being observed by CNTs. The low mechanical reinforcement imparted by GNPs is attributed by the authors to a low accessibility of the nanographite surface to the polymer chains. It undoubtedly reveals a weak polymer–filler interface. In a study of reduced graphene oxide/natural rubber (NR) nanocomposites, Potts et al.⁹⁴ conclude, in agreement with the conclusion of Bokobza,⁹⁵ that the superior reinforcement capability of graphene-based materials results from a high aspect ratio, alignment during stretching, and promotion of strain-induced crystallization in the case of crystallizable elastomer like NR. The authors also mention the significant impact of the processing conditions on the morphology and the properties of the resulting composites. An interfacial bonding between carbon nanomaterials that would be probed by Raman spectroscopy can be achieved by a surface modification of the filler, which seems to be an important step to ensure a molecular level dispersion of the particles.

CONCLUSIONS

This paper reports information that can be gained from Raman spectroscopy that has proved to be one of the most powerful techniques used in extensive studies of carbon materials in the pure state or embedded in a polymer matrix. The strong resonance Raman scattering from these carbon materials makes possible their characterization even if tiny amounts are dispersed in a polymeric medium. The sensitivity of the Raman spectra to various factors such as pressure, temperature, strain, or orientation allows a better understanding of the molecular processes involved in the extent of property improvement in composite materials. Moreover, the dispersion behavior of some Raman bands, that is, their shift to higher frequencies with increasing laser excitation energy, is able to probe interfacial properties between filler particles and the elastomeric matrix.