PHYSICAL REVIEW B VOLUME 61, NUMBER 20 15 MAY 2000-II

Interpretation of Raman spectra of disordered and amorphous carbon

A. C. Ferrari* and J. Robertson
Department of Engineering, University of Cambridge, Cambridge CB2 1PZ, United Kingdom
共Received 24 November 1999兲
The model and theoretical understanding of the Raman spectra in disordered and amorphous carbon are
given. The nature of the G and D vibration modes in graphite is analyzed in terms of the resonant excitation of
␲ states and the long-range polarizability of ␲ bonding. Visible Raman data on disordered, amorphous, and
diamondlike carbon are classified in a three-stage model to show the factors that control the position, intensity,
and widths of the G and D peaks. It is shown that the visible Raman spectra depend formally on the configu-
ration of the sp 2 sites in sp 2 -bonded clusters. In cases where the sp 2 clustering is controlled by the sp 3
fraction, such as in as-deposited tetrahedral amorphous carbon 共ta-C兲 or hydrogenated amorphous carbon
共a-C:H兲 films, the visible Raman parameters can be used to derive the sp 3 fraction.

I. INTRODUCTION its higher photon energy of 5.1 eV, excites both the ␲ and
the ␴ states and so is able to probe both the sp 2 and sp 3
The great versatility of carbon materials arises from the sites, allowing a direct probe of the sp 3 bonding.15,16 Never-
strong dependence of their physical properties on the ratio of theless, visible Raman spectroscopy is widely used on
sp 2 共graphitelike兲 to s p 3 共diamondlike兲 bonds.1 There are DLC’s, and it would be very useful to have a framework in
many forms of s p 2 -bonded carbons with various degrees of which at least indirectly derive the sp 3 fraction of DLC’s.
graphitic ordering, ranging from microcrystalline graphite to The aim of this paper is to describe in detail the Raman
glassy carbon. In general, an amorphous carbon can have any process in disordered carbons. It is shown that the visible
mixture of s p 3 , s p 2 , and even s p 1 sites, with the possible Raman spectrum depends fundamentally on the ordering of
presence of up to 60 at. % hydrogen. The compositions are sp 2 sites and only indirectly on the fraction of sp 3 sites. We
conveniently shown on the ternary phase diagram, Fig. 1. give a restricted range of conditions under which it is pos-
We define diamondlike carbon 共DLC兲 as amorphous carbon
sible to use visible Raman spectroscopy to derive the sp 3
with a significant fraction of s p 3 bonds. The hydrogenated
content. To do this, we first describe the atomic and elec-
amorphous carbons 共a-C:H兲 have a rather small C-C sp 3
tronic structure of amorphous carbon and then the nature of
content. DLC’s with higher s p 3 content are termed tetrahe-
Raman scattering in disordered carbons, both of which show
dral amorphous carbon 共ta-C兲 and its hydrogenated analog
ta-C:H. Another crucial parameter is the degree of clustering unique features. We then present a three-stage model relating
of the s p 2 phase, which should be added as a fourth dimen- the visible Raman parameters to the sp 2 nanostructure and
sion in the ternary phase diagram.1 Amorphous carbons with content of disordered carbons. This is sufficiently general to
the same s p 3 and H content show different optical, elec- hold for all amorphous carbons, both hydrogenated and
tronic, and mechanical properties according to the clustering hydrogen-free. This paper focuses on the G and D peaks,
of the sp 2 phase. neglecting other features that are sometimes present, such as
Raman spectroscopy is a standard nondestructive tool for
the characterization of crystalline, nanocrystalline, and amor-
phous carbons.2–14 The Raman spectrum of diamond consists
of the T 2g 1332-cm⫺1 zone center mode.2 The Raman spec-
tra of disordered graphite show two quite sharp modes, the G
peak around 1580–1600 cm⫺1 and the D peak around 1350
cm⫺1, usually assigned to zone center phonons of E 2g sym-
metry and K-point phonons of A 1g symmetry,
respectively.3–6 The unusual fact is that G and D peaks, of
varying intensity, position, and width, continue to dominate
the Raman spectra of nanocrystalline and amorphous car-
bons, even those without widespread graphitic ordering.
The key property of interest in DLC is the s p 3 content.1
This is usually measured by nuclear magnetic resonance
共NMR兲 or electron-energy-loss spectroscopy 共EELS兲, but
these are time-consuming and destructive methods. Raman
scattering is sometimes used to probe the s p 2 /s p 3 fraction in
DLC’s. However, visible Raman spectroscopy is 50–230 FIG. 1. Ternary phase diagram of amorphous carbons. The three
times7,8 more sensitive to s p 2 sites, as visible photons pref- corners correspond to diamond, graphite, and hydrocarbons,
erentially excite their ␲ states. uv Raman spectroscopy, with respectively.

0163-1829/2000/61共20兲/14095共13兲/$15.00 PRB 61 14 095 ©2000 The American Physical Society

14 096 A. C. FERRARI AND J. ROBERTSON PRB 61

those at 1100–1200 and 1400–1500 cm⫺1, which will be

discussed elsewhere.

II. ATOMIC AND ELECTRONIC STRUCTURE OF

DISORDERED CARBONS

Disordered carbons have s p 3 and s p 2 sites. The sp 3 sites

have only ␴ states while the s p 2 sites also possess ␲ states.
It is often possible to treat ␴ and ␲ states separately. ␴ and ␲
bonds have a significantly different behavior. ␴ bonds are
two-center bond orbitals between adjacent atoms. In the
bond-orbital approximation,17,18 any property of occupied
states such as the total energy, charge density, or polarizabil-
ity can then be expressed as simply the sum of independent,
short-range terms for each bond. There are no long-range
forces in this approximation, and the electron structure de-
pends only on short-range order.
␲ states are different, because a ␲ orbital usually interacts
with ␲ states on more than one atom to form a conjugated
system such as benzene. Then, one can no longer define
unique bond orbitals. Conjugated bonds cannot now be ex-
pressed as the sum of independent, two-center bonds. Each
bond contains contributions from adjacent bonds, and this
gives rise to longer-range forces and long-range
polarizabilities.19
The medium-range order due to ␲-bonding distinguishes
disordered carbons from the ␴-bonded disordered semicon-
ductors like a-Si. ␲ bonding is maximized if the ␲ states
FIG. 2. 共a兲 Phonon dispersion 共Ref. 21兲 and 共b兲 electronic band
form pairs of aligned ␲ states, or sixfold aromatic rings or structure 共Ref. 17兲 of a single graphite layer. Similar phonon dis-
graphitic clusters of aromatic rings.17 This occurs in micro- persion of graphite is found in the ab initio calculations of Kresse,
crystalline graphite and in annealed DLC’s. However, as- Furthmuller, and Hafner 共Ref. 22兲. The bold lines from ⌫ and K
deposited DLC’s are more disordered than this ‘‘cluster mark the mapping of the E 2g and A 1g -like eigenvectors of aromatic
model.’’ 20 clusters on those of graphite, according to Mapelli et al. 共Ref. 21兲.
Figure 2共b兲 shows the band structure of a single graphite The bold line from K to M corresponds to phonons selected by the
layer.17 The ␴ and ␲ states act separately. The ␴ states lie k⫽q ‘‘quasi selection rule,’’ as shown by the dashed vertical line.
well away from the Fermi level and have gap of 6 eV. ␲ The phonons on the right of K, from K to ⌫, are also selected by the
states and empty ␲ * states form bands that touch at the k⫽q ‘‘quasi selection rule,’’ but do not correspond to modes with
Brillouin zone K. The ␲ band energies along ⌫KM in high modulation of polarizability.
nearest-neighbor tight-binding approximation are
The long-range effects in conjugated systems can be for-
E⫽⫾ ␥ 兩 1⫹2 cos共 ka 兲 兩 , 共1兲 malized by defining a mobile bond order P u v between two
orbitals19 u, v :
where ␥ is the pp ␲ interaction, a is the lattice spacing, and
k is the wave vector. In graphitic clusters, the ␲ states have
minimum band energies of roughly17
P u v ⫽2 兺 c uc v 共4兲

冉 冊
occ
2␥ a
E g⬇ ⬇2 ␥ , 共2兲
M 1/2 La and a bond-bond polarizability ⌸ u v ,wx between bonds uv and
wx:
where M is the number of aromatic rings in the cluster and
L a is the cluster diameter or in-plane correlation length.
We consider aromatic clusters as parts of a graphite su- ⳵ 2E
perlattice. Hence, from Eqs. 共1兲 and 共2兲, the energy states of ⌸ u v ,wx ⫽ , 共5兲
⳵ ␥ u v ⳵ ␥ wx
the clusters can be mapped onto those of graphite by
for wave functions ␺ ⫽ 兺 c u ␾ u , where the sums are over
a occupied states.19 E is the sum of energies of occupied states
1⫹2 cos共 ka 兲 ⫽⫺ 共3兲
La and ␥ u v is the nearest-neighbor interaction along bond uv.
The difference between ␴ and ␲ states is that for ␴ states
or sin(⌬ka)⬇a/()La) for small ⌬k, where ⌬k is the k dis- P u v ⬇0 if orbitals u and v are on different bonds, while in
tance away from K, and K⫽( 32 , 23 ) ␲ /a. Smaller clusters cor- conjugated ␲ systems, P u v ⫽0, and it decreases gradually
respond to k’s further away from K. with the separation of u and v .
PRB 61 INTERPRETATION OF RAMAN SPECTRA OF . . . 14 097

The vibrational modes of covalent solids are often mod-

eled as a valence force field of bond-stretching and bond-
bending forces. These forces are usually short-ranged for
␴-bonded systems.26 On the other hand, a valence force field
for graphite typically uses forces up to fifth-nearest
neighbors.27,28 Recently Mapelli et al.21 were able for the
first time to provide a common force field for aromatic mol-
ecules and graphite by using forces proportional to the bond
orders P u v and bond-bond polarizabilities ⌸ u v ,wx , Eqs. 共4兲
and 共5兲. As ⌸ is directly related to the electronic structure
and tight-binding parameter ␥, this method provides a formal
relationship between longer-range forces, so they are not just
adjustable parameters.

IV. RAMAN SCATTERING IN DISORDERED CARBON

Raman modes in single crystals obey the fundamental se-

lection rule q⬇0, where q is the wave vector of the scattered
phonon. In a finite-size domain, the selection rule is relaxed
to allow the participation of phonons near ⌫, with ⌬q
⬇2 ␲ /d, where d is the dimension of the crystalline domain.
Nemanich, Solin, and Martin29 共NSM兲 showed that the Ra-
man scattering intensity of a finite crystal is given by

n 共 ␻ 兲 ⫹1
I共 ␻ 兲⫽
␻ 兺
q, j
C„q, ␻ j 共 q兲 …兩 F 共 q兲 兩 2

⌫/2␲
FIG. 3. Phonon density of states of a single graphite layer ⫻ , 共6兲
共graphene兲 共Ref. 21兲 and of diamond 共Ref. 23兲. 关 ␻ ⫺ ␻ j 共 q兲兴 2 ⫹⌫ 2 /4

where C„q, ␻ j (q)… is the Raman coupling coefficient for a

III. VIBRATIONAL MODES
phonon of wave vector q and branch j, and 兩 F(q) 兩 2 is the
Figure 3 shows the vibrational density of states 共VDOS兲 wave-vector uncertainty of the phonons involved in the light
of graphite and diamond.21–23 The VDOS of diamond ex- scattering. n( ␻ )⫹1 is the boson occupation factor, and ⌫ is
tends beyond its Raman frequency, 1332 cm⫺1, to ⬃1360 the phonon lifetime broadening. In amorphous materials, the
cm⫺1. Figure 2共a兲 also shows the phonon dispersion curves wave-vector uncertainty is ⌬q⬇1/a, where a is the bond
of a graphite layer.21 The graphite VDOS extends beyond its length, and now all phonon modes can participate in the
⌫-point Raman frequency up to a band limit of ⬃1620 cm⫺1, Raman spectrum. The intensity is now given by the matrix-
due to the upwards phonon dispersion away from ⌫. Graph- element-weighted vibrational density of states according to
ite has a higher VDOS band limit than diamond because the the Shuker-Gammon formula30
sp 2 sites have stronger, slightly shorter bonds than sp 3 sites.
The DOS of alloys can be of two types. If the coupling n 共 ␻ 兲 ⫹1
I共 ␻ 兲⫽ C共 ␻ 兲G共 ␻ 兲. 共7兲
between sites is small, the alloys are in the atomic limit, so ␻
the alloy DOS resembles a compositionally weighted mix-
ture of the DOS of each component. If the coupling is strong, Here, G( ␻ ) is the VDOS of the disordered network. Equa-
then the alloy DOS and the band limits interpolate smoothly tion 共7兲 describes quite well the Raman spectra of a-Si and
between the two components. The Raman spectra suggest a-Ge, which are sp 3 bonded only, by using a broadened
that the VDOS of DLC’s are in the atomic limit, in that version of the crystalline VDOS as G( ␻ ). 26
specific features, such as the G mode of s p 2 sites, remain at The visible Raman spectra of disordered carbons are in
all s p 2 contents. Thus, the band limit does not change lin- marked contrast. The VDOS of disordered carbon with vari-
early with s p 3 content from 1600 to 1360 cm⫺1. This is ous sp 3 contents consists of smooth, broad features.31,32 In
partly because s p 2 sites tend to cluster in a s p 3 matrix in contrast, the Raman spectra of all disordered carbons are
DLC’s.17 A consequence is that the vibrations of sp 2 sites dominated by the relatively sharp G and D features of the
remain around 1600 cm⫺1 and lie above the band limit of the sp 2 sites. This could be ascribed to the much greater cross
sp 3 matrix. This causes these modes to be localized on the section of the ␲ states.7,8 Nevertheless, the prevalence of G-
sp 2 sites and above the extended modes of the s p 3 matrix.24 and D-like features, even in amorphous carbons with little
The band limit thus cannot be used as a way to get the sp 3 graphitic ordering, requires explanation.
fraction, as sometimes suggested.25 We therefore need a dif- The G mode of graphite at 1581 cm⫺1 has E 2g symmetry.
ferent approach in which the changes in the visible Raman Its eigenvector shown in Fig. 4共a兲 involves the in-plane
spectra are related primarily to the changes of the sp 2 phase bond-stretching motion of pairs of C sp 2 atoms. This mode
and only weakly to the s p 3 phase. does not require the presence of sixfold rings, and so it oc-
14 098 A. C. FERRARI AND J. ROBERTSON PRB 61

We now give a more detailed account. In particular, we

propose a physical mechanism to explain the k⫽q ‘‘quasi
selection rule;’’ we identify a different branch in the disper-
sion relation as the origin of D peak 共in contrast to Refs. 36
and 37兲. We formally show which real-space motions give
rise to the D peak, and we propose, on the basis of the
‘‘quasi selection rule,’’ an interpretation to some experimen-
tal findings.
Raman scattering is the inelastic scattering of photons by
phonons due to the change of polarization caused by the
phonon mode.38 When the photon energy is above the band
gap, electrons of all wave vectors can be excited. However,
in graphite, the band gap lies in the visible range only within
a small part of k space around the K point, Fig. 2共b兲. All
these bands have ␲ character. In this case, photons reso-
nantly excite states only at the k vector where the band gap
equals the photon energy. This sets up a polarization density
wave of this k vector. Its intensity is strong because of the
long-range polarizability of ␲ states.
The change of bond polarization with bond length is by
FIG. 4. Carbon motions in the 共a兲 G and 共b兲 D modes. Note that far the dominant term in the Raman matrix element for ␲
the G mode is just due to the relative motion of sp 2 carbon atoms states.26 This term is large for the breathing mode of sixfold
and can be found in chains as well. rings. By symmetry, for a breathing mode of a graphite
plane, the contributions from each ring add constructively
because of the long-range polarization. On the other hand, by
curs at all sp 2 sites, not only those in rings. It always lies in
symmetry, contributions from rings of other orders within a
the range 1500–1630 cm⫺1, as it does in aromatic and ole-
graphite plane tend to cancel. Thus, the polarization wave
finic molecules.33 and Raman coupling have long-range coherence for breath-
The D peak around 1355 cm⫺1 is a breathing mode of A 1g ing modes due to the ␲ bonding and the symmetry of the
symmetry involving phonons near the K zone boundary, Fig. graphite sheet. The greatest coupling is when the electron
4共b兲. This mode is forbidden in perfect graphite and only and phonon states are in phase over the range of polarization.
becomes active in the presence of disorder. The D mode is This leads to the ‘‘quasi selection rule,’’ k⫽q, for the
dispersive; it varies with photon excitation energy, even breathing modes of graphene sheets, Fig. 2. For first-order
when the G peak is not dispersive.34–37 We will see that its scattering, the fundamental selection rule must be relaxed to
intensity is strictly connected to the presence of sixfold aro- allow non-(q⫽0) phonons to contribute. This means that we
matic rings. Tuinstra and Koenig3 共TK兲 noted that the ratio also need disorder to allow the enhancement of k⫽q
of its intensity to that of the G peak varied inversely with phonons.
La : Turning to graphitic clusters, we noted above that the
electronic states of graphitic clusters of size L a can be
mapped onto the modes of graphite at wave vector k by Eq.
I共 D 兲 C共 ␭ 兲
⫽ , 共8兲 共3兲. Mapelli et al.21 showed that the eigenvectors of the main
I共 G 兲 La Raman modes of aromatic oligomers have the same symme-
try as the E 2g and A 1g Raman modes of graphite. They also
where C(515.5 nm)⬃44 Å. 2,3,37 The D peak was first attrib- showed that the eigenvectors of these oligomers or clusters
uted to a A 1g breathing mode at K, activated by the relax- can be mapped onto those of graphite phonons along the
ation of the q⫽0 selection rule.3 It was then linked to direction ⌫KM . In particular, the A 1g -type breathing modes
maxima in the VDOS of graphite at M and K points.4,27 of the aromatic clusters map onto phonons between K and
However, this does not account for the dispersion of the D (K⫺M )/2, and the E 2g -type modes map onto phonons from
position with photon energy, why the D peak overtone, seen ⌫ to (⌫⫺K)/4 共branches shown in bold in Fig. 2兲. Our rela-
even where no D peak is present, is dispersive, or why the tionship 共3兲 can be used to map the A 1g -like modes, thus
I(D)/I(G) ratio 共8兲 is dispersive.34 Phonon confinement 共6兲 providing a way to visualize the real-space motion along that
does not explain why the D mode is more intense than others branch. This indicates that aromatic clusters can be consid-
with smaller ⌬q. It also does not explain why the D mode is ered as a part of a graphite superlattice, both electronically
seen in disordered graphite with L a ⬇30 nm, 6 when the NSM and vibrationally. This simultaneous mapping means that the
formula 共6兲 would limit the participating phonons to a much behavior and dispersion of the D and G peaks in graphite
narrower ⌬q range around ⌫. also holds for aromatic oligomers and clusters in disordered
Figure 2共a兲 shows the phonon dispersion of a single layer carbon.
of graphite. Baranov et al.,35 Pocsik et al.,36 and Matthews Band and phonon dispersions are rather isotropic around
et al.37 proposed that the D peak arises as resonant Raman K. As the photon energy rises, the k⫽q selection rule selects
coupling by a strong enhancement of the Raman cross sec- a ring of phonons around the K point. The symmetric breath-
tion of the phonon of wave vector q, when it equals the wave ing modes have the highest modulation of the polarizability
vector k of the electronic transition excited by the incident and therefore have the highest Raman cross section. This
photon (k⫽q ‘‘selection rule’’兲. suggests that modes between K and M give the highest con-
PRB 61 INTERPRETATION OF RAMAN SPECTRA OF . . . 14 099

tribution to the D peak and possibly explains why the inten-

sity of D peak is higher than modes from other branches of
the dispersion curves but with the same q. Note that previous
works36,37 assigned the D peak to all the modes around K in
the lower optical branch, which touches the acoustic branch
at K. To select the correct optical branch we cannot just rely
on the coincidence between its calculated dispersion and the
experimental positions of D peak. The A 1g mode is singly
degenerate. We thus need a band singly degenerate at K and
upwards dispersing away from K. The branch chosen by
Refs. 36 and 37 leads to a doubly degenerate E mode at K.21
The upper branch in Ref. 37 disperses downwards, but the ab
initio calculations of Mapelli et al.21 and Kresse, Furth-
muller, and Hafner22 reproduce the symmetry and upwards
dispersion, Fig. 2.
Applying the k⫽q selection rule to all the phonon
branches of graphite, we can account for other features of the FIG. 5. Variation of the I(D)/I(G) ratio with L a . The broad
Raman spectrum. First, considering the ⬃2400-cm⫺1 peak as transition between the two regimes is indicated.
an overtone of the lower acoustic branch away from K, we pling coefficient, which incorporates various resonances.
can explain its redshift39 with increasing laser energy due to There is no a priori reason to choose a particular function to
the opposite dispersion of this phonon branch. Second, a fit the spectrum. Empirically, the visible Raman spectra of
Stokes shift with lower frequency than the anti-Stokes shift amorphous carbons show one or two prominent features 共the
was reported for the D peak, and vice versa for the G and D peaks兲 and some minor modulations 共usually
2400-cm⫺1 peak.39 We can now explain the slight difference around 1100–1200 and 1400–1500 cm⫺1兲. The simplest fit
in the Stokes and anti-Stokes energies due to the slope of the consists of two Lorentzians or two Gaussians. A Lorentzian
phonon and electron dispersion relations away from K. Fi- fit is often used for crystals, arising from finite lifetime
nally, in graphite or disordered graphite with a high c-axis broadening, and it is normally used for disordered graphite.
ordering, the D peak and its second-order peak are A Gaussian line shape is expected for a random distribution
doublets.6,34,39 This was originally attributed to two maxima of phonon lifetimes in disordered materials. A simple two-
in the VDOS at K and M.27 However, we attribute these symmetric-line fit is not always suitable, and one can find a
doublets to the splitting of phonon and electron branches of multipeak fit 共typically four: D,G⫹2 at ⬃1100 and ⬃1400
given wave vector by interlayer interactions when three- cm⫺1兲.
dimensional stacking occurs.21,35 This causes doublets to act The most widely used alternative to a Gaussian fit is a
as signatures of c-axis ordering.6 Breit-Wigner-Fano 共BWF兲 line for the G peak and a Lorent-
So far we implicitly assumed graphite to be the reference zian for the D peak.40–42 The BWF line has an asymmetric
to explain the Raman features in micro/nanocrystalline line shape, which should arise from the coupling of a discrete
graphite. The main consequence is that the D peak arises mode to a continuum.43 The BWF line shape is given by
from aromatic rings. Starting from graphite, at a fixed ␭,
I(D)/I(G) will increase with increasing disorder, according I 0 关 1⫹2 共 ␻ ⫺ ␻ 0 兲 /Q⌫ 兴 2
I共 ␻ 兲⫽ , 共9兲
to TK Eq. 共8兲. For more disorder, clusters decrease in num- 1⫹ 关 2 共 ␻ ⫺ ␻ 0 兲 /⌫ 兴 2
ber become smaller and more distorted, until they open up. where I 0 is the peak intensity, ␻ 0 is the peak position, ⌫ is
As the G peak is just related to the relative motion of C sp 2 assumed as the full width at half maximum 共FWHM兲 and
atoms, the I(D) will now decrease with respect to I(G) and Q ⫺1 is the BWF coupling coefficient. The Lorentzian line
the TK relationship will no longer hold, as shown in Fig. 5. shape is recovered in the limit Q ⫺1 →0. We emphasize that
For small L a , the D-mode strength is proportional to the several points should be considered with Eq. 共9兲. First, the
probability of finding a sixfold ring in the cluster, that is, BWF curve tails increasingly to lower frequencies for lower
proportional to the cluster area. Thus, in amorphous carbons Q values. This allows a BWF line to account for residual
the development of a D peak indicates ordering, exactly op- Raman intensity at ⬃1100 and 1400 cm⫺1, without two extra
posite from the case of graphite. peaks. The BWF⫹Lorentzian line pair is therefore an excel-
We can finally summarize the main factors modifying lent means to fit Raman spectra of all carbons, from graphite
C( ␻ ): to ta-C. A Lorentzian line shape is used for the D peak as it
共1兲 sp 2 sites are resonantly enhanced over s p 3 ones, is from the same family as the BWF line, while the various
共2兲 within the s p 2 matrix, q⫽k modes are enhanced over enhancement mechanisms for the D peak are consistent with
the others, and a Lorentzian. However, any wide low-frequency tail of the
共3兲 breathing modes are enhanced within q⫽k modes. BWF line will push the D peak to lower frequencies as the
disorder increases. This significantly decreases the D peak
V. SPECTRUM FITTING size compared to a two-Gaussian fit. In general, the D-peak
position will decrease with increasing disorder with the
A practical point when comparing different fitting param- BWF⫹Lorentzian fit, but will increase 共up to 1400 cm⫺1 or
eters for Raman spectra is to know the fitting procedures more兲 for the double-Gaussian fit.9,10,13 Note that the fit of
used. The Raman spectrum is a VDOS modified by a cou- the D peak and especially its position is the least accurate for
14 100 A. C. FERRARI AND J. ROBERTSON PRB 61

many amorphous carbons, because it is often only a low-

frequency shoulder of the G peak. Two factors can shift the
D peak. On one hand, smaller aromatic clusters have higher
modes21 and shift D upwards. On the other hand, a decrease
in number of ordered aromatic rings on passing from nano-
crystalline graphite to a-C lowers D and reduces its intensity,
due to softening of the VDOS.44
Another important issue from Eq. 共9兲 is that the maximum
of the BWF line is not at ␻ 0 but lies at lower frequencies:

⌫
␻ max⫽ ␻ 0 ⫹ , 共10兲
2Q
as Q is negative. We define the G position as ␻ max rather
than ␻ 0 . ␻ 0 is higher than the apparent peak maximum
because ␻ 0 is the position of the undamped mode.43 We
attribute no physical meaning to the undamped frequency but
merely view the BWF line as an efficient way to fit the data. FIG. 6. Schematic diagram of influences on the Raman spectra.
The asymmetric BWF line shape is appropriate for the G A dotted arrow marks the indirect influence of the sp 3 content on
peak due to the asymmetry of the VDOS of graphite or increasing G position.
amorphous carbons towards lower wave numbers.9 No Fano
resonance is present. Whenever reporting data from other 共3兲 a-C→ta-C (→⬃100% sp 3 ta-C, defected
papers using BWF fits, we will use ␻ max , derived by apply- diamond45兲.
ing Eq. 共10兲 to their data. Moreover, ␻ max compares directly For simplicity, we will consider the evolution of G-peak po-
with data from symmetric curve fitting. sition and I(D)/I(G). Except where differently stated, we
Finally, it is not always clear if the I(D)/I(G) ratio refer to Raman data at 514 nm.
should be the ratio of the peak heights or peak areas. Gener-
ally, groups using BWF⫹Lorentzian fits report peak height
ratios, while groups using two Gaussians report the area ra- A. Stage 1: From graphite to nanocrystalline graphite
tio. The difference is not so important for disordered graph- The main effects in the evolution of the Raman spectrum
ite, as the peak widths are similar, but this is not so for in this stage are the following.
amorphous carbons. In that case, the broadening of the D 共a兲 The G peak moves from 1581 to ⬃1600 cm⫺1.
peak is correlated to a distribution of clusters with different 共b兲 The D peak appears and I(D)/I(G) increases follow-
orders and dimensions, and thus the information about the ing the TK equation 共8兲.
less distorted aromatic rings is in the intensity maximum and 共c兲 There is no dispersion of the G mode.
not in the width, which depends on the disorder. Ring orders
other that six tend to decrease the peak height and increase
its width. Unless differently stated, in this paper we refer to
I(D)/I(G) as the ratio of peak heights.

VI. THREE-STAGE MODEL

The large amount of experimental visible Raman spectra

on amorphous carbons will be interpreted using a phenom-
enological three-stage model. Given a perfect, infinite graph-
ite sheet, we consider the introduction of a series of defects:
bond-angle disorder, bond-length disorder, and hybridiza-
tion. We neglect the possible role of hydrogen, as C-H
modes give no detectable contributions in the G and D peaks
共Sec. VIII兲. The Raman spectrum is considered to depend on
共1兲 clustering of the s p 2 phase,
共2兲 bond disorder,
共3兲 presence of s p 2 rings or chains, and
共4兲 the sp 2 /s p 3 ratio.
These factors act as competing forces on the shape of the
Raman spectra, as shown schematically in Fig. 6. We define
an amorphization trajectory6 ranging from graphite to ta-C
共or diamond兲 consisting of three stages, as shown in Fig. 7:
共1兲 graphite→nanocrystalline graphite (nc-G), FIG. 7. Amorphization trajectory, showing a schematic variation
共2兲 nanocrystalline graphite→a-C, and of the G position and I(D)/I(G) ratio.
PRB 61 INTERPRETATION OF RAMAN SPECTRA OF . . . 14 101

These effects, at a fixed wavelength, can be explained by ordering of a-C. This is expressed in effect 共b兲 by the pro-
the VDOS of graphite and phonon confinement.5,6 First, the portionality of I(D)/I(G) to M, the number of ordered rings.
shift of G is really the appearance of a second peak, D ⬘ , at In fact, I(D)/I(G) is proportional to the number and clus-
⬃1620 cm⫺1, which merges in the G peak for small grains. tering of rings, but the main disordering effect in stage 2 can
A single line fit to G⫹D ⬘ feature gives a net increase of G be taken as the decrease of number of ordered rings, since
position. The appearance of D ⬘ occurs because the relaxation the dimensions are under 20 Å. We propose a new relation
of the q⫽0 selection rule allows higher-frequency phonons, for stage 2:
as phonons disperse upwards away from ⌫; see Figs. 2 and 3.
The main structural change is passing from a monocrystal- I共 D 兲
⫽C ⬘ 共 ␭ 兲 L 2a . 共12兲
line to a polycrystalline material; there are virtually no sp 3 I共 G 兲
sites. The loss of three-dimensional ordering is indicated by
the disappearance of the doublet in the D peak and in its Imposing continuity between Eqs. 共8兲 and 共12兲, we find
second-order peak.6 There are many experimental results C ⬘ (514 nm)⬇0.0055. At low excitation energy, the D peak
showing stage 1, such as those from Lespade and is due to large aromatic clusters. Thus, combining Eqs. 共2兲
co-workers.5,6 and 共12兲, I(D)/I(G) will vary with the optical gap as
We note that there are only few experimental verifications
I共 D 兲 C⬙
of the TK equation 共8兲, where L a is known independently by ⫽ . 共13兲
x-ray diffraction 共XRD兲,2 and the minimum L a for which the I 共 G 兲 E 2g
TK equation has been directly verified is ⬃20 Å. TK as- We have verified Eq. 共13兲 by studying the clustering of sp 2
sumes that graphite becomes uniformly nanocrystalline. sites in ta-C deposited at elevated temperatures.48 It is clear
However, for a system with mixed grain sizes, with volume that this is an ideal situation in which thermal energy favors
fractions X i and dimensions L ai , the effective L a is given by the clustering of the sp 2 phase into ordered rings, and so Eq.
N 共13兲 holds. This is not so in general, especially for as-
1 1
L a,eff
⫽ 兺i X i L ai . 共11兲 deposited samples, where the ion-induced disorder in the sp 2
phase invalidates the simple relation 共2兲 between cluster size
and band gap.20 In fact, as we will discuss in Sec. VI C, in
We can thus explain why, since XRD weights more the big- going from as-deposited a-C’s to ta-C we have always
ger crystallites, the TK equation will underestimate L a due to I(D)/I(G)⬃0, but the gap increases. However, we stress
the dominant effect of small crystallites.46 that for visible Raman spectroscopy, whenever a D peak is
present 关 I(D)/I(G)⭓0.1– 0.2兴 , a decrease of the gap will
B. Stage 2: From nanocrystalline graphite to a-C always be reflected in an increase of I(D)/I(G), even if not
exactly in the form of Eq. 共13兲. We will discuss elsewhere
In this stage, defects are progressively introduced into the
the progressive insensitivity of I(D)/I(G) to the gap with
graphite layer, causing its phonon modes to soften, particu-
increasing excitation energy.
larly the G peak. The Shuker-Gammon formula 共7兲 applies,
Clear experimental examples of stage 2 are the ion im-
and the VDOS is no longer that of graphite. The end of stage
plantation of glassy carbon40,41 共g-C兲 and sputtered a-C.49,31
2 corresponds to a completely disordered, almost fully
Figure 8 plots data of McCulloch and co-workers40,41 on ion
sp 2 -bonded a-C consisting of distorted sixfold rings or rings
implantation of g-C at room temperature as a function of ion
of other orders 共maximum 20% s p 3 ). A typical example is dose 关Fig. 8共a兲兴 and at a fixed dose with increasing implan-
sputtered a-C.47 The main effects in the evolution of the tation temperature 关Fig. 8共b兲兴. We show the first and second
Raman spectrum are
stages of amorphization. The sp 3 content was checked by
共a兲 The G peak decreases from 1600 to ⬃1510 cm⫺1. EELS and it rose to 15% only at the very end of stage 2. An
共b兲 The TK equation is no longer valid: I(D)/I(G)⬀M NMR determination of sp 3 content in sputtered a-C with the
⬀L 2a . G peak at ⬃1500 cm⫺1 and I(D)/I(G)⬃0 gave sp 3
共c兲 I(D)/I(G)→0. ⬃7%. 47,31
共d兲 Increasing dispersion of the G peak occurs. The structure of a-C at the end of stage 2 consists of
mainly sp 2 sites in puckered ring-like configurations 共con-
Another effect is the absence of well-defined second-order sisting of five-, six-, seven-, and eightfold disordered rings兲,
Raman peaks, but a small modulated bump from ⬃2400 to and few if any sp 3 sites.47,44,50,51 Li and Lannin47 showed an
⬃3100 cm⫺1. absence of ordered, planar sixfold rings 关consistent with
Increasing bond-angle and bond-bending disorder and the I(D)/I(G)⬃0] and few chainlike structures.
presence of nonsixfold rings softens the VDOS.44,24 The in-
troduction of s p 3 sites into a structure composed only of
C. Stage 3: From a-C to ta-C
sixfold rings further softens the VDOS.24,44
Increasing the defects and reducing L a below 2 nm, the In passing from a-C to ta-C, the sp 3 content rises from
number of ordered rings now decreases and I(D) starts to ⬃10–20 % to ⬃85%, while the sp 2 sites change gradually
decrease. The G peak relates only to bond stretching of sp 2 from rings to chains. The ␲ states become increasingly lo-
pairs, so G retains its intensity, and I(D)/I(G) decreases calized on olefinic sp 2 chains and, eventually, sp 2 dimers
with increasing amorphization 共Fig. 5兲. The TK equation is embedded in the sp 3 matrix.24,51–54,15 The sp 2 modes lie
no longer valid. This is the usual situation with a-C. Devel- above the sp 3 modes and become localized.24 Olefinic CvC
opment of the D peak indicates disordering of graphite but bonds are shorter than aromatic bonds, so they have higher
14 102 A. C. FERRARI AND J. ROBERTSON PRB 61

FIG. 8. Variation of the G position and I(D)/I(G) ratio with

共A兲 ion dose and 共B兲 implant temperature for ion-bombarded glassy
carbon, after McCulloch and co-workers 共Refs. 40 and 41兲. FIG. 9. Variation of the G position and I(D)/I(G) ratio with
sp 3 fraction for as-deposited a-C. Data from Prawer et al. 共Ref.
42兲, this work, and Anders and co-workers 共Refs. 56 and 59兲.
vibration frequencies.33,55 The main effects in the evolution
of the Raman spectrum are as follows. man spectra 共⬃1570 cm⫺1 compared to ⬃1510 cm⫺1兲. This
共a兲 The G peak increases from ⬃1510 to ⬃1570 cm⫺1 共or dispersion can only be fully explained by contributions of
⬃1630 cm⫺1 for s p 2 dimers in ion-implanted olefinic sp 2 groups, whose higher vibration frequencies lie
diamond.45兲 above the graphite band limit.15 Tallant et al.60 suggested to
共b兲 I(D)/I(G) is very low or 0. fit the Raman spectra of as-deposited ta-C using the frequen-
共c兲 Dispersion of the G peak occurs. cies of embedded ideal five-, six-, and sevenfold sp 2 rings.
The main change, i.e., the increase of the G-peak position Such a model, even if very good fits are obtained for the
with s p 3 content, is due to the change of s p 2 configuration 514-nm Raman spectra, cannot explain the dispersion above
from rings to olefinic groups, with their higher vibrational ⬃1580 cm⫺1 seen in the uv Raman spectra and is thus in-
frequencies lying above the band limit of graphite. This ef- correct for ta-C. uv Raman spectroscopy gives not only an
fect is larger than the tendency of the G peak to fall due to evenly weighted probe of sp 3 and sp 2 sites, but also an
evenly weighted probe of ring and chain sp 2 modes, not
mixing with lower-frequency s p 3 modes. This emphasizes
biased towards sp 2 configurations of lower band gap. It
the importance of the localization of s p 2 modes above the
shows a G peak at 1660 cm⫺1 in ta-C, indicating a prepon-
sp 3 modes, which minimizes the mixing of s p 2 with sp 3 derance of chain groups.
modes. It follows that the model of Richter et al.25 does not Figure 9 shows our 514-nm Raman data on ta-C depos-
hold in practice. ited by a filtered cathodic vacuum arc 共FCVA兲, together with
The second major change is the absence of a D peak in a data of Prawer et al.42 and Anders and co-workers59,56 on
BWF fit. The G skewness falls to almost 0 at high sp 3 (t)a-C deposited by FCVA. We only included data with sp 3
content.42 Also, the G-peak width first increases and then content known by EELS. Figure 9 is a clear example of the
falls, as the G modes become localized on s p 2 dimers or transition from stage 2 to stage 3 in accordance with the
shorter s p 2 chains with a sharper length distribution. A above trends. Note the absence of the D peak in ta-C’s. Al-
single-Gaussian fit is poor, although it still gives a fair rep- though I(D)/I(G)⬃0, typically we have an increase in the
resentation of peak position and FWHM.56 gap from ⬃0.5 to ⬃2.5 eV, going from a-C to ta-C. This
It has been argued that the high frequency of the G peak would contradict Eq. 共13兲. However, this is expected since
in ta-C is due to its high compressive macroscopic stress.56,14 the gap is controlled by the ␲-electron delocalization, not
We disagree with this, as it is found that the G peak does not necessarily in well-ordered rings, but on the whole sp 2
move if the stress is removed by annealing.57–59 We verified phase. An increasing sp 2 content, even if not via an ordered
that annealing up to complete stress release induces minimal sp 2 matrix, causes a decrease of the gap reflected in a soft-
structural changes in ta-C.57 Also, the G peak of ta-C is ening of the G mode and an increase in its FWHM.
blueshifted in both uv Raman spectra 共⬃1660 cm⫺1, com- Figure 7 summarized the behavior of G peak position and
pared to ⬃1590 cm⫺1 for s p 2 -bonded a-C兲 and 514-nm Ra- I(D)/I(G) through all the three stages. It shows no unique
PRB 61 INTERPRETATION OF RAMAN SPECTRA OF . . . 14 103

relationship between the G-peak position and s p 3 content.

G-peak position can either increase or decrease with increas-
ing s p 3 and a high and low s p 3 content corresponds to the
same G position. I(D)/I(G) would discriminate, between
high and low content and, except for the first stage, in which
sp 3 is constant anyway, it would be a crucial parameter to
quantify the s p 3 phase. Figure 7 also emphasizes that most
changes of the Raman spectra are not driven by the sp 3
increase, but by the evolution of s p 2 clusters.
Figure 7 shows how we could relate the 关 G,I(D)/I(G) 兴
pair to s p 3 content. However, the situation is more complex
than described so far, as the clustering of the s p 2 phase has
to be taken directly into account, as we discuss now.

VII. THE HYSTERESIS CYCLE

The amorphization trajectory discussed above is derived

for disordering 共e.g., ion implantation兲 in relatively ordered
carbons or for room-temperature depositions. What happens
if we follow an ordering trajectory from ta-C to graphite?
Examples of an ordering trajectory are deposition at high
temperature, annealing after deposition, low-dose ion im-
plantation of ta-C, or unfiltered deposition processes. These
cases favor clustering of s p 2 sites into fairly ordered aro-
matic rings.
There are two fundamental processes: 共a兲 s p 3 sites con- FIG. 10. Amorphization trajectory, showing the possibility of
vert to s p 2 sites and 共b兲 s p 2 cluster size increases and the hysteresis in stages 2 and 3.
sp 2 phase eventually orders in rings. There are two situa-
tions. During a room-temperature deposition of ta-C, the sp 2 VIII. RELATIONSHIPS BETWEEN VISIBLE RAMAN
and sp 3 phases are linked together, forcing the s p 2 phase to SPECTRA AND THE sp 3 FRACTION IN a-C:H
evolve continuously with increasing s p 3 content, giving the
trends seen in Fig. 7. On the other hand, other treatments, More generally, if there is a relationship between sp 2 and
such as annealing or high-temperature deposition, separate sp phases, e.g., between the optical gap and sp 3 fraction,
3

the two processes so that clustering 共b兲 occurs at lower tem- we can derive sp 3 content from the visible Raman spectra.
peratures than conversion 共a兲.48,57 This causes hysteresis. We apply this idea to derive a correlation between visible
Visible Raman spectroscopy is much more sensitive to clus- Raman spectra and sp 3 content for a-C:H.
tering than conversion. The effect of the hysteresis is that The main effect of H in a-C:H is to modify its C-C net-
there is no unique relation between I(D)/I(G) or the G po- work compared to a-C of similar sp 3 content. A higher sp 3
sition and s p 3 fraction 共Fig. 10兲. Thus, we need an indepen- content is achieved mainly by H saturating CvC bonds as
dent assessment of the s p 3 fraction. Fundamentally, optical wCHx groups, rather than by increasing the fraction of
and electrical properties correlate closely with the degree of C—C bonds 共Fig. 1兲. Most sp 3 sites are bonded to
sp 2 clustering, and not directly with the s p 3 content. This hydrogen.63,64 Thus, highly sp 3 a-C:H are soft, low-density,
implies that in general visible Raman spectroscopy is not a polymeric films.63,64 In a-C:H the sp 2 sites can exist as rings
safe way to get sp 3 content. Various examples of hysteresis as well as chains. Increasing H content reduces the sp 2 clus-
can be found in the literature;14,48,57,61,62 see Fig. 11. ter size and increases the band gap. We have three bonding
We have so far neglected the presence of s p 1 bonds, regimes.1,63 At low H content, sp 2 bonding dominates and
whose C-C vibrations at 2100–2200 cm⫺1 共Ref. 33兲 lie out- the gap is under 1 eV. At intermediate H content, the C-C
side the G and D regions. Even if present in a small amount, sp 3 bonding is a maximum, the films have the highest den-
this does not change our model. sity and diamondlike character, and the gap is 1–1.8 eV. At
Generally, in an inhomogeneous material we predict the highest H contents, the sp 3 content is highest, the bonding is
TK equation to underestimate L a with respect to XRD, as for more polymeric, and the band gap is over 1.8 eV. ta-C:H
Eq. 共11兲. This gives a hysteresis even in stage 1, in that differs in that a higher sp 3 fraction occurs at a fixed, lower H
visible Raman spectroscopy is more sensitive to the smaller content of 25–30 % 共Fig. 1兲. ta-C:H has much more C-C sp 3
graphitic domains in a material not composed of grains hav- bonding than a-C:H with similar sp 3 fraction, giving a
ing a similar L a . higher density and higher hardness.65
Are there conditions for estimating s p 3 content by visible In visible Raman spectra, we can neglect all C-H modes.
Raman spectroscopy? Figures 7 and 9 show that a high The stretching modes lie above 3000 cm⫺1.64 C-H bending
G-peak position combined with a I(D)/I(G)⬃0 is a suffi- modes lie in the D-peak region,33,64 but we neglect them
cient condition to assess the s p 3 content of ta-C. In this case, because they are not resonantly enhanced. This is supported
the s p 3 content can be read off from Fig. 9共a兲. Here, a higher by a similar behavior for D and G peaks with changing ex-
G position correlates with a higher optical gap. citation energy found in a-C:H and a-C.10,11 C-H modes
14 104 A. C. FERRARI AND J. ROBERTSON PRB 61

FIG. 12. G position and I(D)/I(G) ratio vs optical Tauc gap for
as-deposited a-C:H. Data from Tamor and Vassel 共Ref. 13兲 and this
work. The precursor gases are also indicated. Note that a double-
Gaussian fit was used 共Ref. 13兲.

the difference is that in a-C:H the CvC stretching frequen-

FIG. 11. G position and I(D)/I(G) ratio data showing the hys-
cies tend to fall towards the lower values of under 1500
teresis effect. Data on as deposited ta-C, from Prawer et al. 共Ref.
cm⫺1 seen in polyacetylene,33,55 whereas in ta-C the CvC
42兲, this work, and Anders and co-workers 共Refs. 56 and 59兲. Data
stretching frequency tends to rise towards that of the embed-
on irradiated ta-C from McCulloch et al. 共Refs. 61 and 62兲; on ta-C
annealed after deposition from Ferrari et al. 共Ref. 57兲; and on ta-C
ded CvC dimer at 1630 cm⫺1. The mixing with sp 3 modes
deposited at high temperature from Chhowalla et al. 共Ref. 48兲. also helps to lower the G peak in a-C:H.
For as-deposited a-C:H, there is a general relationship
could become detectable at much higher photon energy, such between sp 2 content and optical gap;67 see Fig. 13. The line
as that in uv Raman spectroscopy. A typical signature of in Fig. 13 is a fit to the experimental data. Applying the
hydrogenated samples is the increasing photoluminescence fitting line to the data of Fig. 12, we obtain the relationship
background with increasing H content. This background between sp 3 content and Raman parameters shown in Fig.
overshadows the Raman signal of a-C:H with H content over 14. The crosses in Fig. 14 are for samples whose sp 3 content
⬃40–45 at. %66 The ratio between the slope m of the fitted was directly measured 共Ref. 68兲 by NMR or EELS 共this
linear background and the intensity of the G peak, m/I(G), work兲. They agree with the sp 3 content derived by Raman
could be used as a measure of the bonded H content.66 spectroscopy. Thus Raman spectroscopy is a valuable
We derive and explain the relation between visible Raman method to obtain sp 3 content for as-deposited a-C:H. Fig-
parameters and sp 3 content for a-C:H deposited by plasma- ures 12–14 will be improved by a further systematic study.
enhanced chemical vapor deposition 共PECVD兲. From Tamor Figure 14 also shows Raman and sp 3 data for ta-C:H
and Vassel13 we obtain a general relation between 514-nm films deposited by an electron cyclotron wave resonance
Raman parameters and the optical gap for as-deposited source from C2H2. 65 The G peak of ta-C:H is seen to lie
a-C:H 共Fig. 12兲. I(D)/I(G) is now an area ratio of a two- above that of a-C:H of similar sp 3 content 共gap兲. These data
Gaussian fit 关thus giving I(D)/I(G) up to ⬃4, in contrast to show how the transition between the second and third stage
⬃2 obtained with the intensity ratio in a BWF⫹Lorentzian also occurs in a-C:H, indicating how the three-stage model
fit兴. Figure 12共a兲 shows that the G peak falls with increasing applies to both unhydrogenated and hydrogenated carbons.
gap for a-C:H, differently from ta-C, although the gap in- It is important to note that the G peak in a-C:H and
creases with sp 3 content in both materials. The reasons for (t)a-C shows dispersion with photon energy, in both cases
PRB 61 INTERPRETATION OF RAMAN SPECTRA OF . . . 14 105

FIG. 13. Optical Tauc gap vs sp 3 content for as-deposited

a-C:H. Data from Tamor and co-workers 共䊉兲 共Refs. 13 and 68兲,
Kleber et al. 共䉱兲 共Ref. 69兲, Jarman et al. 共䉲兲 共Ref. 70兲, Li and
Lannin 共⽧兲 共Refs. 31 and 47兲, and this work 共䊏兲. An ideal point at
5 eV is set to correspond to 100% sp 3 共*兲. The line is a quadratic fit
to the data.

increasing for higher photon energies.10–12,15,16 Thus, the re-

lations between the G position and gap or s p 3 fraction of
Figs. 9, 12, and 14 apply for 514-nm excitation. The in-
creased G position with increased excitation arises from the
resonant selection of wider-band gap ␲ states from sp 2 FIG. 14. G position and I(D)/I(G) ratio vs sp 3 fraction for
groups with higher vibration frequency. This leads to a lower as-deposited a-C:H. The data are obtained applying the fit of Fig.
sensitivity of the G position and I(D)/I(G) to optical gaps 13 to data in Fig. 12. The ⫻ symbols indicate samples for which
with higher excitation energy,12 since the optical gap is due sp 3 was directly measured by NMR 共Ref. 68兲 or EELS 共this work兲.
to the more delocalized ␲-bonded structures.17 This would The ta-C:H data point 共〫兲 is shown for comparison; its sp 3 content
suggest that red Raman spectroscopy is preferable to the tra- was directly measured.
ditional green or blue spectroscopy to better exploit the abil-
ity of visible Raman spectra to follow the fine variations of the resonant enhancement of their vibrations. We are able to
optical gap on s p 2 order. classify all the available visible Raman data by considering
The width of the G peak is proportional to the bond-angle the effect of a three-stage introduction of disorder into graph-
disorder at sp 2 sites. Figure 15 plots the G width (⌬G) ite on its Raman spectrum. We showed how this description
against the optical gap for as-deposited a-C:H.13 It is seen applies both to hydrogen-free and hydrogenated amorphous
that ⌬G passes through a maximum at around 1.5 eV for carbons.
a-C:H, which corresponds to films of maximum C-C sp 3 or
‘‘diamondlike’’ content.
As the skeletal structure of a-C:H depends strongly on its
H content, we expect a strict relation between the H content
and C-C structure during annealing of a-C:H. We therefore
expect only a small hysteresis of the Raman parameters dur-
ing annealing of a-C:H compared to the case of ta-C. Thus,
relations in Figs. 12 and 14 are valid for both as-deposited
and annealed a-C:H films. However, we expect hysteresis for
ta-C:H.

IX. CONCLUSIONS

We have reviewed and critically assessed the origin and

the meaning of the D and G peaks in the Raman spectra of
graphite and amorphous carbons. We pointed out that the G
peak is due to the relative motion of s p 2 carbon atoms, while
the D peak is linked to breathing modes of rings. We showed FIG. 15. ⌬G vs the optical Tauc gap for as deposited a-C:H.
how the electronic and vibration states of s p 2 aromatic clus- Data from Tamor and Vassel 共Ref. 13兲 and this work. Note that ⌬G
ters can be mapped onto those of graphite. The Raman spec- is the width of the Gaussian, not its FWHM 共Ref. 13兲. The precur-
tra depend formally on the ordering of the s p 2 sites, due to sor gases are also indicated.
14 106 A. C. FERRARI AND J. ROBERTSON PRB 61

The ability to deduce s p 3 content from the visible Raman be derived from the 514-nm Raman spectra. A relationship
spectra depends on the linkage of s p 2 and s p 3 phases. In was given between the G-peak position, I(D)/I(G), and sp 3
H-free (t)a-C, the clustering of s p 2 only depends on sp 3 content.
content as-deposited, but generally not in films annealed, de-
posited at higher temperatures, or ion implanted. Thus, the ACKNOWLEDGMENTS
sp 3 content can be deduced from their Raman spectra only
for as-deposited ta-C. The C-C network of a-C:H depends The authors are grateful to C. Mapelli and C. Castiglioni
strongly on its hydrogen content, which links the sp 2 and for useful discussions. A. C. F. would like to thank the Eu-
sp 3 phases together. This allows the s p 3 content of a-C:H to ropean Union Marie Curie TMR for financial support.

28
*Electronic address: [email protected] R. A. Jishi, L. Venkantaraman, M. S. Dresselhaus, and G.
1
J. Robertson, Prog. Solid State Chem. 21, 199 共1991兲; Pure Appl. Dresselhaus, Chem. Phys. Lett. 209, 77 共1993兲.
Chem. 66, 1789 共1994兲. 29
R. J. Nemanich, S. A. Solin, and R. M. Martin, Phys. Rev. B 23,
2
D. S. Knight and W. B. White, J. Mater. Res. 4, 385 共1989兲. 6348 共1981兲.
3
F. Tuinstra and J. L. Koening, J. Chem. Phys. 53, 1126 共1970兲. 30
R. Shuker and R. W. Gammon, Phys. Rev. Lett. 25, 222 共1970兲.
4
R. J. Nemanich and S. A. Solin, Phys. Rev. B 20, 392 共1979兲. 31
F. Li and J. S. Lannin, Appl. Phys. Lett. 61, 2116 共1992兲.
5 32
P. Lespade, R. Al-Jishi, and M. S. Dresselhaus, Carbon 20, 427 G. P. Lopinski, V. I. Merkulov, and J. S. Lannin, Appl. Phys.
共1982兲. Lett. 69, 3348 共1996兲.
6 33
P. Lespade, A. Marchard, M. Couzi, and F. Cruege, Carbon 22, D. Lin-Vien, N. B. Colthurp, W. G. Fateley, and J. G. Grasselli,
375 共1984兲. The Handbook of Infrared and Raman Characteristic Frequen-
7
N. Wada, P. J. Gaczi, and A. Solin, J. Non-Cryst. Solids 35&36, cies of Organic Molecules 共Academic, New York, 1991兲.
543 共1980兲. 34
R. P. Vidano, D. B. Fishbach, L. J. Willis, and T. M. Loehr, Solid
8
S. R. Salis, D. J. Gardiner, M. Bowden, J. Savage, and D. Rod- State Commun. 39, 341 共1981兲.
way, Diamond Relat. Mater. 5, 589 共1996兲. 35
A. V. Baranov, A. N. Bekhterev, Y. S. Bobovich, and V. I.
9
R. O. Dillon, J. A. Woollam, and V. Katkanant, Phys. Rev. B 29, Petrov, Opt. Spektrosk. 62, 1036 共1987兲 关Opt. Spectrosc. 62, 612
3482 共1984兲. 共1987兲兴.
10 36
M. Yoshikawa, N. Nagai, M. Matsuki, H. Fukuda, G. Katagiri, H. I. Pocsik, M. Hundhausen, M. Koos, and L. Ley, J. Non-Cryst.
Ishida, A. Ishitani, and I. Nagai, Phys. Rev. B 46, 7169 共1992兲. Solids 227–230, 1083 共1998兲.
11 37
J. Wagner, M. Ramsteiner, C. Wild, and P. Koidl, Phys. Rev. B M. J. Matthews, M. A. Pimenta, G. Dresselhaus, M. S. Dressel-
40, 1817 共1989兲. haus, and M. Endo, Phys. Rev. B 59, 6585 共1999兲.
12 38
M. A. Tamor, J. A. Haire, C. H. Wu, and K. C. Hass, Appl. Phys. P. Y. Yu and M. Cardona, Fundamentals of Semiconductors
Lett. 54, 123 共1989兲. 共Springer-Verlag, Berlin, 1996兲.
13
M. A. Tamor and W. C. Vassel, J. Appl. Phys. 76, 3823 共1994兲. 39
P. Tan, Y. Deng, and Q. Zhao, Phys. Rev. B 58, 5435 共1998兲.
14 40
J. Schwan, S. Ulrich, V. Bathori, H. Erhardt, and S. R. P. Silva, J. D. G. McCulloch, S. Prawer, and A. Hoffman, Phys. Rev. B 50,
Appl. Phys. 80, 440 共1996兲. 5905 共1994兲.
15
K. W. R. Gilkes, H. S. Sands, D. N. Batchelder, J. Robertson, and 41
D. G. McCulloch and S. Prawer, J. Appl. Phys. 78, 3040 共1995兲.
W. I. Milne, Appl. Phys. Lett. 70, 1980 共1997兲. 42
S. Prawer, K. W. Nugent, Y. Lifshitz, G. D. Lempert, E. Gross-
16
V. I. Merkulov, J. S. Lannin, C. H. Munro, S. A. Asher, V. S. man, J. Kulik, I. Avigal, and R. Kalish, Diamond Relat. Mater.
Veerasamy, and W. I. Milne, Phys. Rev. Lett. 78, 4869 共1997兲. 5, 433 共1996兲.
17
J. Robertson and E. P. O’Reilly, Phys. Rev. B 35, 2946 共1987兲; J. 43
M. V. Klein, in Light Scattering in Solids III, edited by M. Car-
Robertson, Adv. Phys. 35, 317 共1986兲. dona and G. Guntherodt, Topics in Applied Physics Vol. 51
18
W. A. Harrison, Phys. Rev. B 8, 4487 共1973兲. 共Springer-Verlag, Berlin, 1982兲.
19 44
C. A. Coulson and H. C. Longuet-Higgins, Proc. R. Soc. London, D. Beeman, J. Silverman, R. Lynds, and M. R. Anderson, Phys.
Ser. A 191, 39 共1947兲; 193, 447 共1948兲. Rev. B 30, 870 共1984兲.
20
J. Robertson, Diamond Relat. Mater. 4, 297 共1995兲. 45
S. Prawer, K. W. Nugent, D. N. Jamieson, Diamond Relat. Mater.
21
C. Mapelli, C. Castiglioni, G. Zerbi, and K. Mullen, Phys. Rev. B 7, 106 共1998兲; J. D. Hund, S. P. Withrow, C. W. White, and D.
60, 12 710 共1999兲; C. Mapelli, Tesi di Laurea, Politecnico di M. Hembree, Phys. Rev. B 52, 8106 共1995兲.
46
Milano, 1998. A. Cuesta, P. Dhamelincourt, J. Laureyns, A. Martinez-Alonso,
22
G. Kresse, J. Furthmuller, and J. Hafner, Europhys. Lett. 32, 729 and J. M. D. Tascon, J. Mater. Chem. 8, 2875 共1998兲; H. Wil-
共1995兲. hem, M. Lelaurain, E. McRae, and B. Humbert, J. Appl. Phys.
23
P. Pavone, K. Karch, O. Shutt, W. Windl, D. Strauch, P. Gian- 84, 6552 共1998兲.
nozzi, and S. Baroni, Phys. Rev. B 48, 3164 共1993兲. 47
F. Li and J. S. Lannin, Phys. Rev. Lett. 65, 1905 共1990兲.
24 48
T. Kohler, T. Frauenheim, and G. Jungnickel, Phys. Rev. B 52, M. Chhowalla, A. C. Ferrari, J. Robertson, and G. A. J. Amara-
11 837 共1995兲. tunga, Appl. Phys. Lett. 76, 1419 共2000兲.
25 49
A. Richter, H. J. Scheibe, W. Pompe, K. W. Brzezinka, and I. F. Parmigiani, E. Klay, and H. Seki, J. Appl. Phys. 64, 3031
Muhling, J. Non-Cryst. Solids 88, 131 共1986兲. 共1988兲.
26 50
R. Alben, D. Weaire, J. E. Smith, and M. H. Brodsky, Phys. Rev. G. Galli, R. A. Martin, R. Car, and M. Parrinello, Phys. Rev. Lett.
B 11, 2271 共1975兲; D. Beeman and R. Alben, Adv. Phys. 26, 62, 555 共1989兲.
339 共1977兲. 51
U. Stephan, T. Frauenheim, P. Blaudeck, and J. Jungnickel, Phys.
27
R. Al-Jishi and G. Dresselhaus, Phys. Rev. B 26, 4514 共1982兲. Rev. B 49, 1489 共1994兲.
PRB 61 INTERPRETATION OF RAMAN SPECTRA OF . . . 14 107

52
K. W. R. Gilkes, P. H. Gaskell, and J. Robertson, Phys. Rev. B ings No. 498 共Materials Research Society, Pittsburgh, 1998兲, p.
51, 12 303 共1995兲. 37.
53 61
D. A. Drabold, P. A. Fedders, and P. Strumm, Phys. Rev. B 49, D. G. McCulloch, E. G. Gerstner, D. R. McKenzie, S. Prawer,
16 415 共1994兲. and R. Kalish, Phys. Rev. B 52, 850 共1995兲.
62
54
N. A. Marks, D. R. McKenzie, B. A. Pailthorpe, M. Bernasconi, D. G. McCulloch, D. R. McKenzie, S. Prawer, A. R. Merchant, E.
and M. Parrinello, Phys. Rev. B 54, 9703 共1996兲. G. Gerstner, and R. Kalish, Diamond Relat. Mater. 6, 1622
55
Y. Yacoby and E. Ehrenfreund, in Light Scattering in Solid IV, 共1997兲.
63
P. Koidl, C. Wagner, B. Dischler, J. Wagner, and M. Ramsteiner,
edited by M. Cardona and G. Guntherodt 共Springer-Verlag, Ber-
Mater. Sci. Forum 52, 41 共1990兲.
lin, 1991兲. 64
56 J. Ristein, R. T. Stief, L. Ley, and W. Beyer, J. Appl. Phys. 84,
J. W. Ager, S. Anders, A. Anders, and I. G. Brown, Appl. Phys.
3836 共1998兲.
Lett. 66, 3444 共1995兲. 65
57 N. A. Morrison, S. E. Rodil, A. C. Ferrari, J. Robertson, and W.
A. C. Ferrari, B. Kleinsorge, N. A. Morrison, A. Hart, V. Stolo- I. Milne, Thin Solid Films 337, 71 共1999兲.
jan, and J. Robertson, J. Appl. Phys. 85, 7191 共1999兲. 66
B. Marchon, J. Gui, K. Grannen, G. C. Rauch, J. W. Ager, S. R.
58
T. A. Friedmann, J. P. Sullivan, J. A. Knapp, D. R. Tallant, D. M. P. Silva, and J. Robertson, IEEE Trans. Magn. 33, 3148 共1997兲.
Follstaedt, D. L. Medlin, and P. B. Mirkarimi, Appl. Phys. Lett. 67
J. Robertson, Philos. Mag. B 76, 335 共1997兲.
71, 3820 共1997兲. 68
M. A. Tamor, W. C. Vassel, and K. R. Carduner, Appl. Phys.
59
S. Anders, J. W. Ager, G. M. Pharr, T. Y. Tsui, and I. G. Brown, Lett. 58, 592 共1991兲.
Thin Solid Films 308, 186 共1997兲. 69
R. Kleber, K. Jung, H. Ehrhardt, I. Muhling, K. Breuer, H. Metz,
60
D. R. Tallant, T. A. Friedmann, N. A. Missert, M. P. Siegal, and and F. Engelke, Thin Solid Films 205, 274 共1991兲.
J. P. Sullivan, in Covalently Bonded Disordered Thin-Film Ma- 70
R. H. Jarman, G. J. Ray, R. W. Stanley, and G. W. Zajac, Appl.
terials, edited by M. P. Siegal et al., MRS Symposia Proceed- Phys. Lett. 49, 1065 共1986兲.