Review Papers

Review on Raman scattering in semiconductor nanowires: I. theory

[+] Author Affiliations
Andrés Cantarero

University of Valencia, PO Box 22085, 46071 Valencia, Spain

J. Nanophoton. 7(1), 071598 (Nov 21, 2013). doi:10.1117/1.JNP.7.071598
History: Received August 7, 2013; Revised October 4, 2013; Accepted October 7, 2013
Text Size: A A A

Open Access Open Access

Abstract.  Raman scattering is a nondestructive technique that is able to supply information on the crystal and electronic structures, strain, temperature, phonon–phonon, and electron–phonon interaction. In the particular case of semiconductor nanowires, Raman scattering provides additional information related to surfaces. Although correct, a theoretical approach to analyze the surface optical modes loses critical information when retardation is neglected. A comparison of the retarded and unretarded approaches clarifies the role of the electric and magnetic polarization in the Raman selection rules. Since most III–V compounds growing in the zincblende phase change their crystal structure to wurtzite when growing as nanowires, the polariton description will be particularized for these two important crystal phases. Confined phonons exist in cylindrical nanowires and couple with longitudinal and transverse modes due to the presence of the nanowire’s surface. This coupling vanishes in the case of rotational symmetry. The boundary conditions of the electromagnetic fields on small-size nanowires (antenna effect) have a dramatic effect on the polarization properties of a Raman spectrum.

Figures in this Article

Nanowires (NWs) are quasi-one-dimensional (1-D) materials where one of the dimensions, usually in the 10- to 100-nm range, is much smaller than the other two (in the range of a few microns). They were first called nanowhiskers, but they have received other names in the literature such as nanorods or nanocolumns. They can be made of metals, polymers, or semiconductors, and the most common cross-sections are nearly cylindrical, hexagonal, triangular, or squared. The NW geometry has several advantages as compared to quantum wells or quantum dots, the most elemental one being the large surface-to-volume ratio as compared to a thin film (additional contribution of the lateral surface). Depending on the density and cross-section, an NW bunch can have 100 times more surface than a thin film of the same dimensions. This fact can be useful in the development of solar cells or chemical sensors.1 On the other hand, new concepts of heterostructures arise from the NW shape, like core/shell or axial heterostructures (quantum dots in a wire). Core/shell nanostructures can be particularly useful in the development of solar cells,2 while axial heterostructures are suitable for quantum computing.3

Semiconductor NWs can be grown by chemical vapor deposition, chemical beam epitaxy, or molecular beam epitaxy.46 Usually, they are grown using gold as a catalyst,7 but depending on the semiconductor, they can also be grown without a catalyst.5 They are commonly grown on a convenient substrate like Si or GaAs, which can be prepatterned in order to develop ordered NW arrays.1 Due to the existence of a free surface, heterostructures can be grown with larger lattice mismatch than superlattices or quantum wells.8 The reason is the relaxation of the structure at the sample surface.9 In very thin NWs, the fact that the bond distances close to the surface are larger due to relaxation effects produces changes in the electronic band structure of the material. NWs have a huge amount of applications—in photonic as light emitting diodes or lasers, solar cells, as energy harvesting materials,1,2 in transport as single-electron transistors,3,10 or in phototransport devices.11,12

One more particularity of semiconductor NWs is the fact that the balance between surface energy and bulk energy gives rise to new crystal structures, not stable when grown in bulk systems. In particular, most of the III–V semiconductors crystallizing in the zincblende (ZB) structure in bulk develop the wurtzite (WZ) structure when growing in the form of NWs;7 actually, very often, ZB staking faults appear within the WZ structure. A recent experiment on InAs NWs photoluminescence13 shows the existence of two bands in the emission at low temperatures, presumably one with origin in the ZB region and a second from the WZ region. From this variation, an energy gap of 2.45 eV for the E1 gap of InAs has been obtained. This value is in very good agreement with resonant Raman measurements14 and ab initio calculations.15 High-resolution transmission electron microscopy images clearly show the existence of these two regions. The same behavior has been observed in other semiconductors, like InP.6 A detailed band structure calculation in the WZ phase has been performed in these two compounds.15,16

Figure 1 shows a set of InN NWs grown by molecular beam epitaxy under different temperatures, nitrogen flux, and In vapor pressure conditions. We can observe how the shape of the NWs can be completely different depending on the growth conditions.

Graphic Jump LocationF1 :

Indium nitride NW morphologies obtained under varied metal flux and substrate temperature (thermocouple temperature using the 1×1cm2 with 8.88×8.88mm2 opening) and constant growth time of 300 min and nitrogen flux FN=2 sccm (cc/min at 0°C at 1 atm of pressure) and a plasma excitation power PN=500W, which corresponds to a growth rate at stoichiometry of 12.5m/min. Arrows indicate the modifications produced with the increase of the substrate temperature Ts. (The author thanks Dr. Christian Denker for providing some SEM images of his PhD dissertation.)

Finally, the electronic bands bend at the surface of the NWs as in any surface in a solid, but in the case of an NW the effect is more notable due to the large contribution of the surface compared to the inner part of the NW. Although the origin is not completely clear, some NW semiconductors have a depletion layer at the surface, some others an accumulation layer.5,17 The accumulation layer influences both the optical18 and transport properties.19

In this first work, we will focus on the theoretical development of the Raman theory, first in bulk and then particularizing for the case of NWs. In a second work, we will pay more attention to the comparison of the existing theories with the available experimental results.

In a semiconductor, most of the incident radiation is either reflected, absorbed, or transmitted; only a small amount is scattered. The elastically scattered light is called Rayleigh scattering. Actually, the elastic scattering is possible due to the existence of defects in the sample; in a nearly perfectly grown material with translational symmetry, the elastic scattering must be very low.20,21 A tiny amount of the scattered light, in general several orders of magnitude smaller than that scattered elastically, corresponds to inelastic light scattering or Raman scattering.

The Raman effect or light scattering was predicted by Smekal in 1923 (22) and measured simultaneously by Landsberg and Mandelstam,23 and by Raman and Krishnan24 in solids and liquids, respectively. Actually, the first set of authors discovered the effect previously, but the publication of the second set of authors appeared earlier in 1928. For this discovery, Raman was the Nobel laureate in physics in 1930. Raman was the only Third-World scientist who received the Nobel laureate for a research performed in an undeveloped country.

Although the Raman effect is understood usually as the scattering of light due to phonons, any kind of inelastic light scattering is called Raman scattering. The scattering by electronic excitations or magnetic excitations are also Raman scattering. In this review, we are going to limit the analysis to the Raman scattering by phonons. The classical theory of the Raman effect was developed by the same authors who discovered the effect, Mandelstam et al.,25 and the quantum theory by Tamm.26 A rigorous development of the classical theory can be found in the book of Born and Huang.27 We will follow here the development of Cantarero et al.28,29 and Trallero-Giner et al.30

In the infrared, there is a direct coupling of the electric field of the light with the phonons, but in Raman scattering, the coupling is via virtual electronic states, as it was shown many years ago by Loudon.31 If the semiconductor gap is around the energy of the laser or scattered light, the transition between electronic states becomes real and the Raman signal increases orders of magnitude. In that case, we have to take into account the width of the electronic states, and the process is usually referred as resonant Raman scattering (RRS).28 Double and triple resonances (multiple resonances with several electronic levels) can also increase, additionally, the scattering signal.32 Electric fields,33 magnetic fields,34 or high pressure35 can also be tuned to produce RRS. Excitons play an important role in first-order RRS (28) or Raman scattering by one phonon. In second-order RRS (scattering by two phonons at the Γ-point), excitons also play a fundamental role in the resonant behavior36 because the intermediate excitonic state can also resonate. Electron–phonon interaction is thus the crucial part of the resonant process, since there is a direct interaction between the electrons and the phonons, even out of resonance. Electron–phonon interaction is actually very important within the electronic transitions itself since it renormalizes the electronic gap of the materials37 and gives a broadening to the electronic states. Also, when the electrons have enough energy to emit optical phonons, the photoluminescence processes become more efficient since the emission probability increases. Electron–phonon interaction is important not only in optics but also in transport, through the different relaxation mechanisms.38

The laser wavelength or laser energy is the other crucial part of the resonant process, since the variation of the laser energy approaching the electronic transition is the basis of the RRS. As an example, the Raman signal of strontium titanate/barium titanate (STO/BTO) quantum wells when illuminated with visible light was not good enough to even see the Raman signal. But ultraviolet light has been shown to be very efficient for analyzing STO/BTO quantum wells constituted even by a single layer39 due to resonance effects. In single wires, since there is a small amount of material, resonant effects are very important15 to make the Raman effect observable.

In nonpolar semiconductors, there is no splitting between the the transverse (TO) and longitudinal (LO) optical phonons, but in polar semiconductors like GaAs, there is a splitting related to the difference between the low- and high-frequency dielectric permittivities (or the effective charges). The splitting is a measure of the polarity of the semiconductor. In polar semiconductors, plasmons interact with LO phonons giving rise to an anticrossing when the plasma frequency coincides with the LO phonon frequency, which depends on the electron concentration.40 From the measurement of the two split frequencies, the electron concentration can be obtained.40 In heavily doped samples, the electronic background is of the order of the Raman signal; electrons interact strongly with the phonons giving rise to the Fano lineshape,40,41 produced by the sum of quantum probabilities of the two processes, which can have opposite signs. Since phonons depend on the mass of the ions, Raman scattering has been used in the last years to study the isotopic composition of isotopically engineered samples4244 and phonon anharmonicities.45 Phonons are also sensitive to pressure or stress variations.46,47 In thin films and semiconductor nanostructures, Raman scattering is used very often to obtain the lattice mismatch through the phonon deformation potentials.48

Recently, with the advances in the fabrication of holographic filters (notch or edge filters), a single spectrometer is enough to study the Raman scattering in most of the materials. A good holographic filter has a cutoff of 100cm2, or even less. There are, in the market, several companies selling compact instruments, smaller than a shoebox, completely equipped connectable to a portable computer capable of obtaining reasonable Raman scattering signals at a competitive price. If we need to study phonon modes at lower frequencies, a double or triple spectrometer, in subtractive mode, is needed. Also, if we need to increase the resolution because the phonon lines are narrow, we can use a triple spectrometer in additive mode and lower the temperature to decrease the phonon anharmonicities. Most of the Raman systems are equipped with a charged coupled device, but if we really need resolution, we have to go back to the photomultipliers. In the book of Kuzmany,49 there is a complete description of the apparatus used in Raman scattering: spectrometers, gratings, and detectors.

During the last decades, part of the scientific community has been dedicated to characterize, by Raman scattering, thin films, quantum wells, quantum dots, quantum wires, or even more complex nanostructures.50 In the simple case of a thin film, if we use the typical collection, a photographic camera objective, producing a laser spot 10μm, most of the Raman signal will come from the substrate and it can mask the really important signal. We need, in that case, the use of a microscope (micro-Raman or μ-Raman system), where the laser spot can be <1μm (with a confocal microscope, we can go to the diffraction limit50). However, the use of a microscope has a couple of disadvantages. First, all the light, including the reflected light, goes through the entrance slit of the spectrometer (unless we use a notch or edge filter). The second disadvantage is that the selection rules are not completely fulfilled. If we have an objective of 100× and a numerical aperture50NA=0.9, such as that shown in Fig. 2(a), the entrance angle with a refractive index of 3 (a typical value for semiconductors) will be 17.5deg. In a typical. μ-Raman system, a confocal microscope is employed in order to perform the measurements on a specific point of the sample (below 1 μm in lateral resolution and 100 nm in depth). A confocal microscope and a motorized xyz plate allow the mapping of small areas at different depths. In Fig. 2(b), there is a scheme of a confocal system where we observe how the light enters into a sample with a smaller angle due to the difference of refractive indices. Although we do not have a complete backscattering geometry, the selection rules are fulfilled to a great extent due to the difference in refractive index.

Graphic Jump LocationF2 :

(a) Cross-section of an objective (by courtesy of Carl Zeiss Microscopy, Germany). 1. Objective thread. 2. Stop face of the objective. 3. Spring system for the specimen-protection mechanism. 4. to 7. Lens groups for the correction of image errors. 8. Correction collar for adapting to deviating cover glass thicknesses or temperatures. 9. Front lens system. 10. Front lens holder. (b) Schematic representation of a confocal setup, where we can observe the angle change inside the sample due to the differences in refractive indices.

In order to compare the Raman signal or the Raman intensity of different materials, it is convenient to use the Raman scattering efficiency (RSE), instead of the scattering cross-section, since the first amount is volume independent. In a polar binary compound semiconductor, the RSE per unit solid angle Ω can be defined as28Display Formula

where c is the speed of light in vacuum, μ the reduced mass of the ions, vc the volume of the unit cell, ωL (ωS) the laser (scattering) angular frequency, nL (nS) the refractive index at the laser (scattered) frequency, ωΓ the angular frequency of the corresponding phonon, eL (eS) the polarization vector of the laser (scattered) light, and RΓ the Raman tensor corresponding to the Γ phonon mode. The factor N(ωΓ)+1 has its origin in a quantum mechanical description of the phonon field,28 and it corresponds to the phonon emission (Stokes shift). In the case of phonon absorption (anti-Stokes shift or anti-Stokes scattering), the factor is N(ωΓ). In practice, this factor is replaced by the Bose–Einstein distribution function, and this is the reason why at the anti-Stokes frequency, the Raman scattering signal is very weak as compared to that appearing at the Stokes shift, unless the temperature is very high. In a classical description, both Stokes and anti-Stokes components have the same intensity. An exception is stimulated Raman scattering, where phonons at the anti-Stokes frequency50 are excited. For Stokes shift, even if N(ωΓ)0, the factor 1 remains and the Stokes shift can be observed at any temperature. The relative intensity of the Stokes/anti-Stokes Raman shift can be used to monitor the temperature.51

The Raman tensor is a second-order tensor, sometimes written as a third-order tensor (or even as a fourth-order tensor52) contracted with the phonon polarization (Rijkq^k, where q^ is a unitary vector in the direction of the phonon displacement). The form of the Raman tensor depends on the symmetry of the crystal and the electron–phonon interaction, since in the Raman effect, electrons play a crucial role as virtual intermediate states in the Raman process.31 For an LO phonon propagating along x ([1,0,0]) direction, the Raman tensor for the optical modes of a ZB crystal can be written in terms of the Raman polarizability α as Display Formula

and it is observable in the x(y,z)x¯ scattering configuration written in Porto’s notation.53 Actually, we can write Ryzx=Rzyx=α. The remaining nonzero Raman tensor components are Rxzy=Rzxy=Rxyz=Ryxz=α. We will come back to this point in Sec. 4. The Raman polarizability has the expression54Display Formula
where the relative phonon displacement u¯Γ is defined by the expression Display Formula
and WFI(ωL,eL,ωS,eS) is the amplitude probability of scattering of a phonon with wave number ωΓ. The remaining quantities have been defined already. The amplitude probability Display Formula
where the δ(EFEI) of the Fermi golden rule has already been taken into account. This amplitude probability corresponds to the Feynmann diagram given in Fig. 3. The remaining five terms or Feynmann diagrams31 are nonresonant, and they can be neglected. Eμ (Eν) and γμ (γν) are the energy and broadening of the electronic states (which can be accessed in a time-resolved Raman experiment55). If there are several initial and/or final states (as it is the case of several valence bands in a ZB or WZ compound), we have to sum over I,F after squares WFI. The sum over intermediate states is performed before squaring the expression. HeR and HeL are the electron-radiation and electron(hole)-phonon Hamiltonians, which can be written in the second quantization formalism.28 The important point is the electron–phonon coupling constant, which consists of two parts, one corresponding to the emission of the phonon by the electron (see Feynmann diagram in Fig. 3) and a second corresponding to the emission of the phonon by the hole. In ZB- (and WZ-) type materials, only the second option is nonzero, and the electron–phonon coupling constant is Display Formula
where Dh(r) is the deformation potential (DP) matrix as defined by Bir and Pikus.56 In the case of the Fröhlich interaction,30 the basic expressions are the same; the only change is in the electron–phonon interaction Hamiltonian. Its coupling constant, in that case, is30Display Formula
where the Fröhlich constant Display Formula
and qζ=qmζ/(me+mh) (ζ=e,h). Since Fröhlich is intraband (there is also an interband Fröhlich interaction with upper conduction bands—the electro-optic effect), we have to consider both contributions, that of the electron and that of the hole. We cannot do the dipole approximation (κLκS0) here, and the absolute value of q must be taken into account. The Raman tensor for Fröhlich interaction is diagonal. Display Formula

Graphic Jump LocationF3 :

Feynmann diagram corresponding to the Raman process described above.

It is interesting to realize that the DP is a local interaction, and thus it is related to the periodic part of the Bloch function, while the Fröhlich interaction, only possible in polar materials, is a long-range interaction related to the polarization charges in the medium, and only the envelope of the Bloch function contributes to the matrix element.30 On the other hand, since both processes are possible in the case of LO phonons, the amplitude probability must be added before squaring it, and interference effects can appear29 when the Raman polarizabilities corresponding to deformation potential (DP) and Fröhlich interaction are of the same order of magnitude, i.e., close to an electronic transition, in resonance. This is the reason we usually say that the selection rules are not completely fulfilled in resonance. During many years, the huge enhancement of Raman scattering close to the gap could not be explained, and it was attributed to unknown values of one of the constants that appear in the expression of RSE or to the existence of impurity-induced Raman scattering.57 Finally, it was demonstrated that the consideration of excitons, instead of electrons,31 as intermediate states in the Raman process could explain the absolute value of Raman scattering in the case of deformation potential,28 Fröhlich interaction,30 and also the interference between both.29 The interference is produced due to the fact that we have to add the probabilities before squaring (they correspond to different intermediate steps).

In Sec. 3, the Raman tensor of an LO phonon propagating along the [100] direction of the ZB structure was defined in Eq. (2). We also defined the remaining components of the Raman tensor using the conventional unitary vectors along x, y, and z. If one is interested in a different crystal surface, a general method to proceed is to define the following three vectors:58Display Formula

defining the crystal surface where the Miller indices are normalized (h2+k2+l2=1). The two perpendicular vectors forming a right-handed coordinate system are Display Formula

It is also convenient to define the matrix formed by these three vectors written in rows. Display Formula


Following Fig. 4, the a1 axis will be always along the x direction, i.e., the incoming wave vector of the light κL and outgoing κS are along a1. The phonon wave vector q, in backscattering configuration, will also be directed along a1. Thus, the LO phonon will vibrate in the a1 direction, while the two TO phonons will vibrate along the a2 and a3 directions. The three vectors are unitary. As commented in Sec. 3, the Raman intensity will be proportional to Display Formula


Graphic Jump LocationF4 :

Fix axes system, where we have chosen z as the direction of the cylinder axis, x as the direction of the light in backscattering configuration, and y as the perpendicular direction, forming a right-handed system. The vectors eL and eS indicate the polarization of the incoming and scattered light, respectively, and θ and ϕ the corresponding angles of rotation, with origin in the z axis with counterclockwise rotation.

In the case of an LO phonon, the Raman intensity will be proportional to Display Formula

while in the case of TO phonons, the intensity will be proportional to Display Formula

The reason of summing after squaring is that these are two different final states, but since they appear at the same energy when looking at the Stokes shift, we have to add the intensities. Of course, depending on the selection rules for a given surface, only one of them can contribute. In the case of the ZB structure, it is very simple to calculate the resulting intensity for a (11¯0) surface, for instance. In that case, Display Formula


If the polarization vectors eL and eS are both along the (110) direction, the transformed vectors Display Formula

are along the x direction. The transformed Raman matrix Ri=ARiA. Display Formula

The Raman matrices multiplied by the phonon polarization, Riai, Display Formula

in units of the Raman polarizability. The final result is |eLRLOeS|2=0, |eLRTO1eS|2=0, |eLR'TO2eS|2=αTO. It is simple to check that, for eL=eS=(1,1,1)/3, the result is |eLRLOeS|2=0, |eLRTO1eS|2=8/9αTO, and |eLRTO2eS|2=4/9αTO (the total contribution of the TO is 4/3αTO). The LO mode is forbidden for this configuration. These results are identical as that given by Cardona.52

Although we have given a general procedure, we will pay particular attention to the case where one of the axis is along the [111] direction, the NW growth direction or the cylinder axis. The reason is that this is the equivalent to the [0001] direction of the WZ structure and the results can somehow be compared. Figure 4 shows the coordinate system where the NW axis is along the laboratory z direction. In that case, it is better to select A as Display Formula

for (11¯0) and (112¯) surfaces, respectively. For a (11¯0) surface, if eL=eS=(1,1,2)/6, |eLRTO1eS|2=2/3αTO, |eLRTO2eS|2=1/3αTO, i.e., the contribution of the TO mode is αTO, while the LO mode does not contribute. If eL=eS=(1,1,2)/6, the contribution of |eLRTO2eS|2=4/3αTO. As is well known, the LO phonon is forbidden in the case of a (11¯0) surface in a ZB material. If eL=eS=(1,1,0)/2, |eLRLOeS|2=2/3αLO and |eLRTO2eS|2=2/3αTO. In the case where eL=eS=(1,1,1)/3, |eLRTO2eS|2=4/3αTO. For crossed polarizations, |eLRLOeS|2=1/3αLO.

Polar plots can be drawn by considering, in the case of (11¯0) surfaces, in agreement with Fig. 4, Display Formula


In the WZ structure, there are nine optical modes, but two of them are silent modes. The seven Raman active optical modes are drawn in Fig. 5. The first mode vibrates along the z direction of the WZ structure. The Raman tensor is Display Formula


Graphic Jump LocationF5 :

Raman allowed phonon modes in wurtzite-type crystals. The A1 mode is 1-D while the E modes are 2-D

The 1-D A1 mode is a polar mode, with a different polarizability when propagating in the z direction or in the xy plane. As stated before, group theory gives only the difference in anisotropy (a and b polarizabilities), but does not consider the LOTO splitting, which is due to the polarity of the medium. The two-dimensional (2-D) E1 modes are also polar modes, with Raman tensor Display Formula


When propagating along x, E1(x) will be LO, while it will be TO if it propagates along y. There are actually two bidimensional E1 modes, but one of them corresponds to two acoustic modes, the third acoustical mode being a 1-D A1 mode. There are also two nonpolar bidimensional E2 modes, as can be observed in Fig. 5. Both modes actually mix, but the amplitude of the heavier (lighter) atom dominates in the E2l (E2h) modes, vibrating at lower (higher) frequency. The corresponding Raman tensor can be written as Display Formula


The WZ is an anisotropic structure. At this point, it is interesting to remark that there are two kinds of anisotropic structures. In the first one, the crystal anisotropy dominates;59,60 actually ε is very different from ε. This is the case of most of the layered compounds,46 or nonpolar compounds like graphite, but the most common case is that where the Coulomb interaction dominates.35 This is the case of III–V compounds grown in the WZ structure. In recent cases where typical ZB crystals grow in the WZ structure, the material anisotropy is unkonwn, but, on one side, we know the LOTO splitting, i.e., the intensity of the Coulomb interaction, and on the other side, a small difference is expected in the crystal anisotropy since the structures are identical until second neighbors. This is very important when we want to draw polar plots in the case of WZ materials. When we rotate the polarization, the polar modes are mixed. Display Formula


Instead in a strongly anisotropic material, A1(TO) and A1(LO) [or E1(TO) and E1(LO)] mix. Since we are interested in crystals with a small anisotropy, the Raman polarizabilities a and b will be taken as equal to simplify the final expressions, i.e., Display Formula


With this approximation, we can easily obtain the Raman selection rules in backscattering from the lateral side of a single WZ-NW grown along the c-axis. Looking at the picture given by Sander et al.,61 the x axis is perpendicular to the m-plane (101¯0) [(100) plane in three-index notation], while the y axis corresponds to the a-plane (112¯0) [(110) in three-index notation]. The z axis is along c, the [0001] direction ([111] direction of the ZB).

The selection rules for a few standard configurations are summarized in Table 1. If one is interested in polar plots, we have to calculate the components of the Raman tensor. Let us do, as an example, the scattering configuration x(ζ(ϕ)ζ(ϕ))x¯ where ζ is in the zy plane. The vectors eL=(0,0,1) and eS=(0,sinϕ,cosϕ).

Table Grahic Jump Location
Table 1Selection rules for backscattering configuration from the a plane (nanowire lying in a surface) and c plane (nanowire perpendicular to the surface or a bunch of nanowires).

The Raman intensities will be, in the x(ζζ)x¯ configuration, Display Formula

We have summed the squares (without considering any interference) since they are different phonon modes or different final states. However, we have to remember that they mix, and give an intermediate frequency between the A1(LO) and E1(LO) or the A1(TO) and the E1(TO), in agreement with Eq. (25).

The number of phonon modes in a crystal depends on the number of atoms in the unit cell. As is well known from a general course of solid-state physics, if a unit cell (primitive cell) has n atoms, there are 3n3 optical modes. Depending on the complexity of the crystal, we can have from a few to tens of optical modes. For instance, in silicon, there are three optical modes, as in ZB-type materials; in WZ materials, there are nine optical modes (seven are Raman active), while in a more complex material, like Bi2S3 and isostructural layered materials, there are 57 optical phonons, 27 of them Raman active.62 In the case of an NW, there is no translational symmetry in the plane perpendicular to the NW axis. If the NW is very thin (a few nanometers of radius), we have confined modes (see Sec. 7). With increasing NW radius increases, the confined phonon modes overlap due to the finite linewidth, giving rise to an asymmetry at the low frequency region of the LO or TO phonon peaks. NWs of the order of several tens of nanometers can be considered, for most of the purposes, as a bulk material. Just as an example, in transport phonons can be treated as bulk materials even for NW diameters down to 30 to 40 nm.63 The only difference will come from the existence of a surface or an interface (core/shell NWs), giving rise to new phonon modes, the so-called surface optical (SO) modes.

Let us start our discussion analyzing bulk phonons (i.e., there is no surface) in a solid. From group theory (at q=0), there is no splitting of the optical modes; LO and TO modes have the same energy. Actually, the LOTO splitting is produced at finite q in polar semiconductors due to the coupling of the phonon field with the electromagnetic field of the light (phonon–polariton). In bulk materials, the polariton branches are derived from the solution of the Maxwell equations inside the material and the coupling of electric field produced by the phonon displacement with the electric field of the electromagnetic wave.64 From the LO to the TO frequencies, there is a forbidden region where the electromagnetic wave cannot propagate in bulk materials (the electromagnetic field is totally reflected). Actually, because of the losses (imaginary part of the dielectric function), the reflectivity is not exactly one in this region.

Although the coupling between the ionic displacement in a polar crystal and the electric field of the electromagnetic wave of the light is produced directly in the infrared region of the spectrum, it can be analyzed by Brillouin scattering,65 and this coupling is responsible for the existence of surface modes, as we will see later. In a Raman experiment in backscattering geometry, q4πnL/λL, where nL and λL are the refractive index and wavelength of the laser light, respectively. For the typical green line of an Ar laser (λL=514.5308nm), q0.08nm1. In a Brillouin scattering experiment, the complete measurement range is below this amount.

If we consider the harmonic movement corresponding to a phonon traveling along the crystal, the equation of motion will be that corresponding to the harmonic oscillator. On the other hand, if there is an electromagnetic wave in the polar crystal, there is a polarization vector proportional to the electric field, the proportionality constant being the electric susceptibility at a microscopic scale. But these fields are coupled through the ionic movement. Let u=u+u be the vector displacement between the positive and negative ion in a simple polar semiconductor (as the ZB, for instance) and w=μ/vcu, (μ/vc is basically the density ρM, the mass density). The equations to be solved in order to obtain both the electric polarization and the vector displacement due to the existence of the external field E are Display Formula


Equation (28) must fulfill at the same time the Maxwell equations, which we write in the Gaussian system, for simplicity, and assume a plane wave solution ei(k·rωt), Display Formula


Combining these equations, we arrive to the solution Display Formula


At very high frequencies, the ionic movement cannot follow the electric field, only the electrons, and we arrive to the trivial solution Display Formula


On the other hand, uncoupled transverse phonons do not carry electric field and thus vibrate at their resonant frequency ωTO, thus b11=ωTO2. Finally, in static conditions (2w/t2=0), we obtain from the coupled Eq. (28) Display Formula


The final expressions coupling the fields and atomic movement are Display Formula

Display Formula

Taking the displacement from Eq. (33), substituting in Eq. (34), and using Eq. (30) in the particular case of transverse modes (P·k=0), we arrive at the following equation: Display Formula


Using Eq. (35), we arrive at the solution Display Formula


The solution of Eq. (36) is given in Fig. 6 using the data of GaAs.42 At k=0, there are two solutions obtained from the last equation taking the limit k0: ω=ωLO and ω=(c/ε0)k. For large values of k, we obtain, taking the limit k, ω=ωTO and ω=(c/ε)k. The longitudinal solutions are obtained assuming P·k=PkE=4πPD=0. From the coupled equations, ω=ωLO. In Sec. 6.1, we will see how these equations are modified due to the existence of a limiting surface in the solid. The value of the electric field can be recovered from Eq. (28). In the case of TO modes, Eq. (28) gives E=0, and for LO modes, Display Formula


Graphic Jump LocationF6 :

Lower and upper polariton branch in bulk GaAs calculated from Eq. (36) using the data of GaAs.42

We can follow the same strategy to analyze the polariton dispersion in a crystal with the WZ structure. There are two modes, which are Raman and infrared active, A1 (a 1-D mode vibrating along the c axis of the WZ structure) and E1 (a 2-D mode vibrating in the plane perpendicular to the c axis). In a uniaxial crystal, the dielectric function is a tensor, isotropic in the plane perpendicular to c. Let us write Display Formula


We will restrict ourselves to the case where the phonons propagate along the z axis and the x axis (k is along z or x—the y direction is equivalent to x). If kz^, we have a A1(LO) mode and two degenerate E1(TO) modes (vibration along x and y). If the phonons propagate along the x (or y) direction, we will have the A1(TO) mode vibrating along z and propagating along x and two E1 modes, a mode vibrating along x and propagating along x, the E1(LO), and a second mode vibrating along y and propagating along x, the E1(TO). Obviously, these two modes are not degenerate in that case.

For the case where kz^, the electric field can be along the x axis (or y). In that case, Eq. (30) becomes Display Formula


The equations coupling the fields with the atomic movement are Display Formula


The same solution is obtained if the fields and ions vibrate in the y plane [exchanging x by y in Eqs. (39) and (40)]. The solutions are the dispersion of the two E1(TO) degenerate modes. By solving the three equations, we arrive at a solution equivalent to that of Eq. (36), replacing ωTO by ω and ε by ε. The solution can be better written in an equivalent way as Eq. (30). Display Formula


The limits k0 and k are easily extracted from these equations, ω=(c/ε0)k and ω=(c/ε)k, respectively (see Fig. 7).

Graphic Jump LocationF7 :

Lower and upper polariton branch in bulk ZnO calculated from Eq. (36) using the data of ZnO.66

Here the electric field and the movement is along z; the solution is Ez=4πPz (D=0), corresponding to the A1(LO) with frequency ω=ε0/εω. These solutions are plotted in Fig. 7 for ZnO. If k is along the x axis, the E1 mode propagating along the x direction will be longitudinal, with frequency ω=ε0/εω. The dispersion of the modes propagating along y and z are obtained by solving the equations Display Formula

Display Formula
respectively. The dispersion is shown in Fig. 7.

In the figures with the polariton branches, k has been written in wave numbers.

In the case of Raman scattering, the value of q is far away from the polariton region. In backscattering configuration and assuming that the selection rules are fulfilled (that depends on the cylinder radius), q=kLkS. For λL=514.53nm, the laser frequency in cm1 is ωL=1/λL=19,435.213cm1. The scattered frequency, for instance, for Si is ωS=19,435.213519.5=19,383.263cm1 (515.909 nm). The value of q will be q=2πnL/λL+2πnS/λS4πnL/λL0.0855nm1=855,000cm1 (in the polariton region, k15,000cm1).

Surface Optical Modes Within the Polariton Description

In finite-size samples, there are additional conditions on the electric and the magnetic fields since they must fulfill the boundary conditions at the sample surface. The existence of these boundary conditions generates additional solutions in the polariton equations. In general, there is a complete set of solutions ranging from below the TO to above the LO region, depending on the geometry.

Let us consider a single cylindrical semiconductor NW of radius a, with a dielectric permittivity ε(ω), and an electromagnetic field with time dependence eiωt. If the external medium has a dielectric permittivity εM (in case of air or vacuum, εM=1), the electric and magnetic fields outside and inside of the NW must fulfill the Helmholtz equations. Display Formula

inside the material and Display Formula
outside the material. The Helmholtz equation can be solved exactly in six coordinate systems,67 in particular, in cylindrical coordinates. In case of a hexagonal cross-section, the main conclusions from this section will not vary, but the calculation of the electromagnetic field must be done numerically due to the complexity of the boundary conditions.

Let us call k2=ε(ω)ω2/c2 and k02=εMω2/c2. Equations (44), in cylindrical coordinates, become, for a scalar function ψ, Display Formula

inside the material. Outside, we have to replace k by k0. By assuming a solution of the type R(ρ)einθeikzz, we arrive at the equation Display Formula
where x2=β2ρ2(k2kz2)ρ2 [or x2=β02ρ2(k02k0z2)ρ2 outside the crystal] and whose solutions are Bessel functions of first class [Jn(βρ)], second class [Yn(βρ)], also called Neumann functions, and third class [Hn(1)(βρ) or Hn(2)(βρ)], or Hankel functions. Inside the cylinder, the solution is of the type Jn(βρ) (analytical at the origin), and outside the cylinder, it is more convenient to take Hankel functions [Hn(1)(β0ρ)=Jn(βρ)+iYn(β0ρ)], i.e., Display Formula
where we have removed the superindex of the Hankel function for clarity. The Hankel functions can be expanded in plane waves.68Display Formula