Professional Documents
Culture Documents
Leoni 2001
Leoni 2001
3, 597± 623
Abstract
Polycrystalline ® lms produced via physical vapour deposition often possess a
microstructure composed of columnar oriented grains. From the mechanical
point of view, they cannot be treated as isotropic bulk solids; rather their
elastic macroscopic behaviour is only transversely isotropic. In this context, the
eŒect of texture on the elastic response of such ® lms has been investigated for
diŒerent grain interaction models. In particular, the determination of
macroscopic stress by X-ray diŒraction methods has been analysed. Traditional
grain interaction models such as those due to Voigt and to Reuss have been
compared with that proposed by Vook and Witt, compatible with the presence
of only transverse isotropy of the body (even in the absence of crystallographic
texture). By simulation, it has been demonstrated that, although texture only
moderately aŒects the values of the macroscopic mechanical elastic constants of
the transversely isotropic body, it can pronouncedly in¯ uence the results from the
X-ray diŒraction measurements. This eŒect is shown to depend strongly on the
type of grain interaction.
} 1. Introduction
Knowledge of the internal strain± stress ® elds in thin deposits is of key impor-
tance for optimizing the corresponding production process and service life. De-adhe-
sion failure and surface-quality-dependen t properties such as the hardness, wear and
corrosion resistance are highly dependent on the presence of stresses in the deposited
layer (Machlin 1995).
X-ray diŒraction is a powerful tool for the non-destructive measurement of the
internal stresses in thin ® lms. Various methods based on X-ray diŒraction have been
presented for the analysis of macroscopically mechanically isotropic (sometimes
termed quasi-isotropic (Hauk 1997)) bodies subjected to homogeneous (Noyan and
Cohen 1987, Hauk 1997, Kamminga et al. 2000) or depth-dependent (Sasaki et al.
1993, Genzel 1994, 1997, 1998, Hauk 1997, Leoni 1998, Leoni et al. 1999, Scardi et
Philosophical Magazin e A ISSN 0141± 8610 print/ISSN 1460-699 2 online # 2001 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
598 M. Leoni et al.
al. 1999) strain± stress ® elds. Special procedures, such as the crystallite group method
(for example Hauk (1997)), have been proposed for the diŒraction stress analysis of
specimens exhibiting distinct macroscopic anisotropy due to texture.
Owing to their microstructure, thin ® lms cannot usually be considered as macro-
scopically isotropic. Thin ® lms produced via physical vapour deposition provide a
good example, as they are often composed of columnar grains aligned along a
preferential direction usually parallel to the surface normal (Machlin 1995). In this
manner, they can exhibit rotational symmetry of the macroscopic (elastic) properties
with respect to the surface normal. From the microstructural crystallographic point
of view, such ® lms show a ® bre texture (for example Bunge (1982)), while, from the
mechanical (macroscopic) point of view, they can be considered as transversely
isotropic (Love 1927, van Leeuwen et al. 1999). All macroscopically isotropic bodies
are obviously also macroscopically transversely isotropic, whereas the reverse does
not necessarily hold. It should be noted that also, in the absence of crystallographic
texture only transverse isotropy on a macroscopic scale can occur (van Leeuwen et
al. 1999).
The macroscopic elastic response of a polycrystalline aggregate diŒers, in gen-
eral, from that of a single crystal. By means of suitable models, the macroscopic
elastic behaviour of a polycrystalline material (as given by the full macroscopic
compliance or stiŒness tensors) can be calculated from single-crystal elastic data.
In the absence of texture, the most common models, namely the Voigt, the
Reuss, the Neerfeld± Hill and the Eshelby± KroÈner models (for example Noyan and
Cohen (1987) and van Leeuwen et al. (1999)), predict macroscopic mechanical iso-
tropy for the aggregate . Thus, if these models are used for the description of an only
transversely isotropic untextured thin ® lm, the symmetry of the calculated macro-
scopic elastic tensor is higher than that really presented by the specimen.
An alternative grain interaction model that can be used for the analysis of a thin
® lm with the symmetry axis parallel to the surface normal has been presented by
Vook and Witt in the 1960s (Vook and Witt 1965, Witt and Vook 1968). This model
does not allow calculation of the full macroscopic elastic tensor; only two macro-
scopic elastic constants, su cient to describe the elastic behaviour of the ® lm under
rotationally symmetric loading, can be calculated. It is possible to prove that the
Vook± Witt model is not compatible with macroscopic elastic isotropy (Welzel et al.
2000).
The Vook± Witt model was originally intended for the study of thermal stres-
ses in thin ® lms. As such, it was used and in some way extended by several
investigators (Murakam i and Yogi 1985, Wieder 1995a,b,c, 1996), but in most
cases it was applied erroneously in the analysis of diŒraction data. Only recently
(van Leeuwen et al. 1999) was the model revisited, elaborated and adapted to the
analysis of residual stresses in thin ® lms using X-ray diŒraction. However, the
consequences of crystallographic texture have not been considered until now. In
the present paper, the approach by van Leeuwen et al. (1999) is extended to
textured ® lms. To demonstrate the distinct in¯ uence of texture on the strain
data obtained from a diŒraction experiment, simulations have been conducted
for various materials covering a large range of elastic (single-crystal) anisotropies.
In this way the large diŒerences become apparent between the values of the stress
components obtained by application of the traditional diŒraction methods and
the actually imposed stress ® eld. Therefore, a more appropriate analysis of dif-
fraction strain data can be proposed.
Strain± stress Welds in transversely isotropic thin Wlms 599
} 2. Theoretical basis
Figure 1. De® nition of and relations between the crystal (C), sample (S) and laboratory (L)
reference frames. (a) Relation between the L system and the intermediate L 0 system;
(b) transformation between the L 0 system and the C system through the intermediate
system L 00 ; (c) transformation between the S system and the L system through the
intermediate system S 0 .
600 M. Leoni et al.
Equation (1) represents a system of six relations between six "Sij and six ¼Sij (i.e. there
are 12 unknowns). From a mathematical point of view, the stress and strain tensors
are fully determined if six more independent equations are given. These additional
hypotheses are provided by adopting a suitable grain interaction model, that is by
assigning, to six of the unknowns, values independent of the orientation of the
crystallite in the S system.
The analysis will be focused on a transversely isotropic thin ® lm subject to a
plane state of stress. Both the symmetry axis of the elastic properties and the sym-
metry axis of the stress ® eld are chosen parallel to the sample normal. Mechanical
equilibrium requires the macroscopic average stress components perpendicular to
the sample surface to be zero. Denoting by angular brackets the volume-weighed
averaging over all occurring crystallite orientations, mechanical equilibrium for the
thin ® lm thus implies that
h¼Si3 i ˆ h¼S3i i ˆ 0: …2†
The ® lm is free to expand along its normal, and the average strain along that direc-
tion is
h"S33 i ˆ "S? : …3†
Owing to the transverse isotropy, the rotational symmetry of the stress ® eld as
expressed by
h¼S11 i ˆ h¼S22 i ˆ ¼S== ; h¼S12 i ˆ h¼S21 i ˆ 0; …4†
induces a rotational symmetry of the strain ® eld:
h"S11 i ˆ h"S22 i ˆ "S== ; h"S12 i ˆ h"S21 i ˆ 0: …5†
Five independent elastic constants su ce to describe completely the macroscopic
elastic behaviour of a transversely isotropic body (Love 1927, Hendrix and Yu 1998,
Leigh and Berndt 1999). Equations (2)± (5) do not permit the determination of all
these ® ve constants (using Hooke’s law).
However, only two elastic constants are required to describe fully the relation
between the stress and strain tensors for a transversely isotropic body subjected to a
plane state of stress (symmetry axis parallel to the sample normal): only two strain
tensor components and one stress tensor component are independent and non-zero.
Thus
"S== ˆ A¼S== ; …6†
angles will be called ¬, and ®. It is usual to associate a set of Euler angles with a
vector g ˆ …¬, , g) in the three-dimensional orientation (Euler) space G (more
details are given in appendix A). In this way, each point in the orientation space
represents a possible orientation of the C system with respect to the S system.
Given the vector g ˆ …¬; ; ®† in G, the volume fraction dV…g†=V of crystallites
in the in® nitesimal orientation range d3 g ˆ sin … † d¬ d d® around g, is provided by
the orientation distribution function (ODF) f …g† according to
dV…g† f …g† 3
ˆ dg
V 8p2
f …¬; ; g†
ˆ sin … † d¬ d d®: …8†
8p2
Owing to this de® nition, the ODF is a probability density (Matthies et al. 1987), and
therefore
………
f …g† 3
2
d g ˆ 1: …9†
G 8p
With these de® nitions, the average value of the (i; j) component of a rank-2
tensor O (in this case, " or ¼) in the S system, is given in terms of Euler angles by
the threefold integral:
………
OSij ‰ f …g†=8p2 Š d3 g
S G
hOi j i ˆ ………
‰ f …g†=8p2 Š d3 g
G
………
1
ˆ OSij f …g† d3 g
8p2 G
… 2p … p … 2p
1
ˆ 2 OSij f …¬; ; ®† sin … † d¬ d d®: …10†
8p ®ˆ0 ˆ0 ¬ˆ0
All strain and stress tensor components can be averaged in this way, but a grain
interaction model is needed for actual computation of the values (see the discussion
below equation (1) and } 2.3).
On this basis, equations (6) and (7) can be reformulated as
………
"S11 f …g† d3 g
"S== h"S11 i G
A ˆ S ˆ S ˆ ……… ; …11†
¼== h¼11 i
¼S11 f …g† d3 g
G
………
"S33 f …g† d3 g
"S? h"S33 i
Bˆ ˆ ˆ ……… G ; …12†
¼S== h¼S11 i
¼S11 f …g† d3 g
G
permitting the value for the two mechanical elastic constants A and B to be calcu-
lated if the ODF, that is the texture, is known.
As discussed in the introduction, (X-ray) diŒraction aŒords a measure of the
macroscopic deformation in solids. However, it does not provide a direct evaluation
602 M. Leoni et al.
of the average stress± strain in the whole sample, in the sense of equation (10). A
diŒraction line contains data on only a subset of the crystallites for which the
diŒracting planes are perpendicular to the chosen measurement direction. Because
only the direction of the diŒraction vector is de® ned, a degree of freedom occurs for
the diŒracting crystallites: the rotation around the diŒraction vector (denoted by the
angle ¶). Thus the diŒraction experiment is insensitive to features perpendicular to
the diŒraction vector and an average over these crystallites is observed experimen-
tally. Hence, the diŒraction strain is in general not equal to the overall mechanical
strain.
In a diŒractometer, the measurement direction, that is the orientation of the
diŒraction vector h with respect to the sample surface normal, is indicated by two
angles Á and ’. The tilt of the sample with respect to the diŒraction plane is indi-
cated by Á, while ’ denotes the rotation of the sample around its surface normal. For
a textured specimen the average strain in the measurement direction of the diŒrac-
tion experiment is given by
L
"meas
’;Á ˆ h"33 i
ˆ hsL33kl ¼Lkl i
… 2p
"L33 F …h; ¶; ’; Á† d¶
ˆ 0… 2p ; …13†
F …h; ¶; ’; Á† d¶
0
where a LS SC
i j and a i j are the direction cosines in the corresponding rotation matrices
(for cubic materials see appendix A for more details). For clarity, the angular and
direction dependences of the transformatio n matrices and of the ODF have been
indicated explicitly in equation (14).
When, for a given set of h and Á in equation (14), the number of diŒracting
crystallites is zero, then F …h; ¶; Á† ˆ 0 independent of ¶ and, obviously, the cor-
responding "meas
Á cannot be evaluated by the diŒraction experiment. The stress tensor
components ¼Skl to be used in equation (14) can be calculated only if a grain inter-
action model is adopted.
suitable rotation of sC for which literature values are available), combining the
Vook± Witt model (see also equation (15)) with Hooke’s law permits the calculation
of three non-zero stress tensor components of a crystallite in the S system according
to
0 S 10 S 1 0 S 1 0 1
s1112 sS1222 2sS1212 ¼11 "12 0
B S S S CB S C B S C def B "S C
@ s1111 s1122 2s1112 A@ ¼22 A ˆ @ "11 A ˆ @ == A …16†
sS1122 sS2222 2sS1222 ¼S22 "S22 "S==
} 3. Simulations
Based on equation (14), simulations of diŒraction measurements have been car-
ried out for diŒerent cubic materials characterized by diŒerent extents of elastic
anisotropy, de® ned by the parameter Ai :
sC1111 ¡ sC1122
Ai ˆ : …24†
2sC1212
Tungsten …Ai ˆ 1:00†, niobium …Ai ˆ 0:55† and gold …Ai ˆ 2:85† were selected for
the simulations: the respective single-crystal elastic constants corresponding to the C
system have been collected in table 1. For most of the known materials, the aniso-
tropy is in the range de® ned here by the `limiting’ values of Ai for niobium and gold.
For each material, the expected (X-ray) diŒraction strain was calculated (equation
(14)) as function of sin2 Á (the so-called sin2 Á plot), for a number of re¯ ections. The
calculations were performed assuming an average stress ¼S== ˆ 100 MPa for the ® lm.
Note that this average stress can be transferred immediately to the average strain
parallel to the substrate "S== from equation (11) and to the strain perpendicular to
the substrate "S? from equation (12), when the elastic constants A and B are known.
The sin2 Á plot is a traditional way of X-ray residual stress analysis (Noyan and
Cohen 1987) based on traditional grain-interaction models compatible with macro-
scopic isotropy. Therefore, in order to evaluate the errors arising from the adoption
of a classical grain interaction model, the sin2 Á plots simulated for the current
606
Table 1. Single-crystal elastic constants and anisotropy factor Ai for the elements used in the sin2 Á plot simulations. The stiŒness matrix components ci j
and the corresponding compliance matrix components si j are shown. The Voigt two-index notation (Nye 1957) is used: to revert to the traditional
tensorial notation, the index must be replaced as follows: 1 ˆ 11, 2 ˆ 22, 4 ˆ 12 and remembering that smn ˆ 4sijkl if m; n > 3. The values in bold
have been calculated here by tensor inversion.
¡0.73
Nye (1957)
Nb (bcc) 244.6 138.1 29.2 6.9 ¡2.49 34.2 0.55 Meyers and Chawla (1984)
Strain± stress Welds in transversely isotropic thin Wlms 607
transversely isotropic ® lm have also been tentatively ® tted according to the tradi-
tional macroscopically isotropic model for the same value of average stress ¼S== . In the
case of macroscopic isotropy, the common form of equation (14) is (for example
Noyan and Cohen (1987) and Stickforth (1966))
h"L33 i ˆ S 1hkl …h¼S11 i ‡ h¼S22 i† ‡ 12 S 2hkl …h¼s11 i cos2 ’ ‡ h¼S22 i sin2 ’† sin2 Á; …25†
where S 1hkl and 1 hkl
2 S2
denote the X-ray elastic constants (XECs) which arise from the
explicit elaboration of equation (14) for the macroscopically isotropic body. In view
of the rotational symmetry of the ® lm, there is no dependence on ’; thus
h¼S11 i ˆ h¼S22 i and
"meas
Á ˆ h"L33 i
ˆ q ‡ m sin2 Á; …26†
where ¼S== ,
as in equation (4), is the average stress in the S system, parallel to the
sample surface. As demonstrated by Stickforth (1966), the two XECs are indepen-
dent of Á for a macroscopically isotropic body. Thus, if the XECs are known, the
average stress parallel to the surface can be calculated from the slope of the straight
line de® ned by equation (26). On the other hand, for a homogeneous specimen and a
homogeneous strain± stress ® eld, the absence of such linearity in the sin2 Á plot
implies the absence of macroscopic isotropy of the body (Welzel et al. 2000).
To demonstrate the pronounced eŒects of texture on the diŒraction stress ana-
lysis, a simple Gaussian ® bre texture (Bunge 1982), with 58 pole width (half-width at
half-maximum (HWHM)) was used for two diŒerent preferred orientations, namely
h110i and h111i.
In addition to the calculation of the diŒraction strain, the elastic constants A and
B were also evaluated as functions of the texture pole width using equations (11) and
(12).
The ODF was supplied numerically for the calculation of A, B and the diŒraction
strain according to equations (11), (12) and (14) respectively. The corresponding
section of the pole ® gure, that is the peak area (integrated intensity of the re¯ ection
considered) versus sin2 Á is also shown for the diŒraction strain calculations. Note
that instrumental aberrations and sample eŒects, such as absorption , are not con-
sidered; application of proper corrections and a comparison with measured data
from vapour-deposite d ® lms will be presented elsewhere (Leoni et al. 2000). For
details on the calculations using the traditional models of Voigt, of Reuss and of
Neerfeld and Hill, see appendix B.
between the values for the elastic constants as calculated for the various grain
interaction models, in the presence or absence of texture, might be considered mod-
erate. As will be shown, the consequences of the intrinsic elastic anisotropy in the
diŒraction-stress analysis, through the eŒects of texture and of the choice of grain
interaction model, are noticeable.
The mechanical elastic constants A and B for h110i and h111i-® bre-textured
niobium ® lms are shown in ® gure 4 as a function of the texture pole width. For a
cubic material and a perfect h111i ® bre texture, it is possible to evaluate the elastic
constants A and B analytically (Kamminga et al. 2000; Vook and Witt 1965):
In this case, A and B are independent of the choice of the grain interaction model;
the values obtained according to equations (27) and (28) agree with the numerical
results shown in ® gures 4 (c) and (d). The convergence to a single value for A and B
for pole width going to zero, independent of the type of grain interaction model
610 M. Leoni et al.
Figure 4. Textured, transversely isotropic niobium ® lm: variation in the macroscopic elastic
constants A and B as functions of the texture pole width (a), (b) for a Gaussian h110i
® bre texture and (c), (d) for a Gaussian h111i ® bre texture. A key, valid for all four
plots, is presented in (a).
adopted, is not expected for the h110i texture and is indeed not observed (see ® gures
4 (a) and (b)).
The proof of Hill (1952) that the average elastic constants for a macroscopically
mechanically isotropic aggregate should fall within the limits imposed by the Reuss
and the Voigt models is in accordance with the numerical results for A and B.
However, it should be recognized that the `eŒective’ XECs need not be within the
Voigt and the Reuss bounds, as demonstrated by the sin2 Á plots for untextured (van
Leeuwen et al. 1999) and textured ® lms (this work).
The sin 2 Á plots for untextured and h110i-textured niobium ® lms, containing a
stress ¼S== of 100 MPa are shown in ® gure 5 (a) for the {211} re¯ ection. Relatively
minor diŒerences occur between the results of the various models, namely the Reuss,
the Voigt and the Vook± Witt models, in the absence of texture. However, if texture is
present, as characterized by three intensity maxima occurring in the plot of inte-
grated intensity versus sin2 Á, signi® cant diŒerences with the untextured case are
observed. Thus, a clearly nonlinear (`discontinuous ’) dependence of "meas Á on sin2 Á
is visible according to the Vook± Witt and the Reuss grain interaction models; the
eŒects in the diŒraction stress analysis of the intrinsic (single-crystal ) elastic aniso-
tropy of the material are enhanced by the presence of texture. The pairs of dotted
curves and the pairs of broken curves in ® gure 5 represent the maximum and mini-
mum values for "L33 …"meas
Á ˆ h"L33 i) as observed for variable ¶ at constant Á and at
constant (arbitrary) ’ respectively (see equation (14) for the corresponding Vook±
Witt and the Reuss grain interaction models). Independent of the texture of the ® lm,
all sin2 Á plots for the Vook± Witt and the Reuss grain interaction models and for the
Strain± stress Welds in transversely isotropic thin Wlms 611
re¯ ection concerned should lie between the corresponding pair of dotted curves.
Hence, if experimental results do not lie between these curves, then the correspond-
ing grain interaction model is not compatible with reality.
Similar observations as above can be made considering corresponding results for
the {222}, {321} and {400} re¯ ections as shown in ® gures 5 (b), (c) and (d) respec-
tively. Again, nonlinear sin2 Á curves occur according to the Vook± Witt grain inter-
action model. Note that, for the {400} re¯ ection, particularly large diŒerences occur
between the results of the various grain interaction models, whereas a relatively small
diŒerence is observed between the maximum and minimum values of "L33 at the same
Á. For the Reuss model it is possible to demonstrate that the sin2 Á plot correspond-
ing to the {00l} and {hhh} re¯ ections is always linear, independent of the texture (see
also Brakman (1983)).
An interesting comparison with tungsten is oŒered by ® gure 6, analogous to
® gure 3. The anisotropy of niobium changes the distribution of strain at constant
Á as a function of ¶, in contrast with what is observed for tungsten.
In all the presented sin2 Á plots, the strain curve exhibits a change in trend
(maxima or minima of the ® rst derivative) at a value of Á where a minimum occurs
in the texture contribution (i.e. diŒraction peak area).
Now, to resemble what is normally done in practice, the simulated data accord-
ing to the Vook± Witt grain interaction model for the case of a textured ® lm (® gure
612 M. Leoni et al.
which are obtained at Á values where re¯ ections occur belonging to the family of
crystals that represent the ® bre texture component. Thus, also strain data recorded at
speci® c Á values pertaining to other {hkl} re¯ ections of the same family of crystals
can be included in the `reduced’ sin2 Á plots. Although the strain data in such plots
may apparently be reasonably well ® tted with a straight line (equation (26)), the
results shown in table 2 demonstrate that erroneous values for the stress can be
obtained.
Commonly, re¯ ections at large Bragg angles are employed in order to minimize
errors in the lattice spacing value due to the uncertainty in peak position determina-
tion. The results obtained here show that the application of the normal procedure
(equation (26)) to the corresponding strain data can lead to quite erroneous results
(e.g. see results for the {400} re¯ ection and h110i ® bre texture in table 2).
Table 2. Apparent stress values for the niobium ® lm calculated from simulated data by using the macroscopically isotropic models (Voigt, Reuss and
Neerfeld± Hill respectively). The intercept and the slope for the straight line ® tted according to equation (26) are given together with the correlation
coe cient R. The true stress is 100 MPa. The ® bre texture is indicated by the ® bre axis; the orientation distribution around the ® bre axis is
Gaussian with a 58 HWHM.
1 hkl
2 S2 (TPa¡1 † ¼2== …MPa†
Re¯ ection Intercept Slope Neerfeld± Neerfeld±
{hkl} Texture …£10¡4 † …£10¡4 † R Voigt Reuss Hill Voigt Reuss Hill
111 h110i ¡9.98 19.4 0.999 14.14 17.10 15.62 137 113 124
112 14.4 0.998 14.14 15.17 14.27 102 95 101
M. Leoni et al.
h110i ¡8.32
123 h110i ¡8.26 14.7 0.999 14.14 15.17 14.42 104 95 102
004 h110i ¡5.02 7.95 0.999 14.14 9.39 11.05 56 85 72
111 h111i ¡10.2 17.2 0.998 16.89 17.10 17.00 102 101 101
112 h111i ¡10.0 17.0 0.997 16.89 15.17 16.03 101 112 106
123 h111i ¡9.97 16.9 0.997 16.89 15.17 16.03 100 112 106
004 h111i ¡6.22 11.4 0.998 16.89 9.39 13.14 67 121 87
Strain± stress Welds in transversely isotropic thin Wlms 615
Figure 8. Textured, transversely isotropic gold ® lm: variation in the macroscopic elastic
constants A and B as functions of the texture pole width (a), (b) for a Gaussian
h110i ® bre texture and (c), (d) for a Gaussian h111i ® bre texture. A key, valid for
all four plots, is presented in (a).
eŒects of texture and of choice of the grain interaction model are most dramatically
revealed in the diŒraction stress analysis, as shown next.
Gold and niobium possess diŒerent crystal structures (fcc and bcc respectively).
In comparison with the niobium ® lm, diŒerent re¯ ections were thus selected for the
study of the gold ® lm, namely the {111}, {311}, {400} and {331} re¯ ections. The
corresponding sin2 Á plots for the untextured and h110i-textured gold ® lms contain-
ing a stress ¼S== of 100 MPa are shown in ® gure 9. The largest diŒerences between the
various grain interaction models are observed for the {111} and {400} re¯ ections
where the Reuss model predicts a linear sin2 Á plot.
Figure 10. Textured, transversely isotropic gold ® lm (Gaussian h111i ® bre texture; 58
HWHM): lattice strain "meas Á (sin2 Á plots) and intensity (arbitrary units) versus
2
sin Á for (a) the {311} and (b) the {331} re¯ ections. A key, valid for both plots, is
presented in (a). The additional pairs of curves represent the minimum and maximum
values for "L33 observed for variable ¶ for the Reuss model (- - - -) and for the Vook±
Witt model (¢ ¢ ¢ ¢ ¢ ¢).
Figure 11. Textured, transversely isotropic gold ® lm (Gaussian h111i ® bre texture; 58
HWHM): `reduced’ sin2 Á plot for the {331} re¯ ection (strain data taken from ® gure
10 (b)) as used for the application of the traditional, macroscopically isotropic models;
straight-line ® t is according to equation (26). The relative intensity is also shown in the
® gure (Ð Ð ) (arbitrary units).
Table 3. Apparent stress values for the gold ® lm calculated from simulated data by using the macroscopically isotropic models (Voigt, Reuss and
Neerfeld± Hill respectively). The intercept and the slope for the straight line ® tted according to equation (26) are given with the goodness-of-® t
parameter R. The true stress is 100 MPa. The ® bre texture is indicated by the ® bre axis; the orientation distribution around the ® bre axis is
Gaussian with a 58 HWHM.
1 hkl
2 S2 (TPa¡1 † ¼2== …MPa†
Re¯ ection Intercept Slope Neerfeld± Neerfeld±
{hkl} Texture …£10¡4 † …£10¡4 † R Voigt Reuss Hill Voigt Reuss Hill
111 h110i ¡6.8 9.9 0.998 14.45 12.05 13.25 69 82 75
113 h110i ¡9.2 15.3 0.991 14.45 23.85 19.15 106 64 80
004 h110i ¡20.6 38.6 0.998 14.45 34.36 24.41 267 112 158
113 h110i ¡8.2 14.2 0.998 14.45 16.01 15.23 98 89 93
111 h111i ¡6.7 11.7 0.998 12.17 12.05 12.11 96 97 97
113 h111i ¡7.3 12.7 0.992 12.17 23.85 18.01 104 53 71
004 h111i ¡13.9 22.7 0.998 12.17 34.36 23.27 187 66 98
133 h111i ¡6.9 12.1 0.996 12.17 16.01 14.09 99 76 86
Strain± stress Welds in transversely isotropic thin Wlms
617
618 M. Leoni et al.
For fcc metals such as gold, h111i ® bre texture is more common than h110i ® bre
texture. The eŒects of the diŒerent textures can be observed by comparing ® gure 9
(h110i ® bre texture) and ® gure 10 …h111i ® bre texture). From ® gure 9 (b) and ® gure
10 (a) the decisive in¯ uence of the type of texture on the sin2 Á dependence of the
strain becomes evident.
As shown for the niobium ® lms, to resemble what is usually done in practice, the
simulated data can be interpreted on the basis of traditional analysis involving the
incorrect assumption of macroscopic isotropy. As an example, ® gure 11 shows the
`reduced’ (see above for the niobium ® lm) data set for the {331} gold re¯ ection (h111i
® bre texture; see ® gure 10 (b)). A linear dependence of "meas
Á on sin2 Á is imposed
according to the traditional analyses. Incorrect and highly diŒerent (from reality)
results are obtained by applying the macroscopically isotropic models to obtain a
stress value from the simulated data (table 3; note that the true stress is 100 MPa).
} 5. Conclusions
The values of the macroscopic mechanical elastic constants of a polycrystalline
body are aŒected moderately by the type of grain interaction model adopted, irre-
spective of the presence of texture. Fibre texture (with the ® bre axis perpendicular to
the specimen surface) in a polycrystalline thin ® lm made of an intrinsically aniso-
tropic material, drastically in¯ uences the dependence of the lattice strain on the
measurement direction, that is nonlinearity occurs in the sin2 Á plots. In addition,
this kind of texture induces a strong dependence of the lattice strain on the type of
grain interaction.
Correct values for the stress components in a textured thin ® lm can be obtained
by ® tting, to the measured data, the values of strain calculated as a function of the
measurement direction by means of equation (13) (general) or (14) (® bre texture).
The grain interaction in strained polycrystalline thin ® lms is more likely to be of
transversely isotropic character than of macroscopically isotropic nature. The appli-
cation of the traditional Voigt, Reuss or Neerfeld± Hill models leads to very erro-
neous values (errors as large as 50% ) for the stress derived from classical diŒraction
analyses (e.g. sin2 Á plots) if a Vook± Witt type of grain interaction, compatible with
transverse isotropy as in thin ® lms, holds. This result renders an often-used method
of diŒraction stress analysis of ® bre-textured thin ® lms, the so-called `crystallite
group method’ , unreliable.
APPENDIX A
REFERENCE FRAMES
The representation of a tensorial quantity depends on the reference frame. The
transformation of a tensor component from one to another orthogonal frame of
reference is performed by means of suitable rotations, de® ned in terms of direction
cosines. Consider two orthogonal systems A and B; the cosine of the angle between
the unit vector j in the B system and the corresponding unit vector i in the A system,
a AB
i j , is called the direction cosine. The direction cosines can be grouped in a rotation
matrix a AB . Since the rotation preserves the norm and angular relations between the
reference axes, the rotation matrix is orthonormal and thus its inverse is equivalent
to its transposed.
Strain± stress Welds in transversely isotropic thin Wlms 619
vA AB B
i ˆ a i j vj : …A 1†
T
Similarly, the inverse transformation is realized by the transposed matrix …a AB
ij † :
AB T A
vBi ˆ a BA A AB A
i j vj ˆ …a i j † vj ˆ a ji vj : …A 2†
These two rotations correspond to the traditional Á and ’ positioning of the speci-
men in a stress± texture diŒractometer.
For the transformation from the laboratory L to the crystal C system, the
intermediate reference frame L 0 , ® xed with respect to C, is introduced (van
Leeuwen et al. 1999). Cubic materials (i.e. hhkli ? fhklg† are considered, and L 3 is
chosen parallel to the hh; k; li direction. By setting L 30 parallel to L 3 , the other two
orthogonal directions (corresponding to L 10 and L 20 ) can be de® ned in the plane
(h; k; l) as the vectors (k 2 ‡ l 2 , ¡hk, ¡hl) and (0; ¡l; k); a subsequent rotation around
L 30 according to the angle ¶, accomplishes the desired a CL transformation
0 0
a CL ˆ a CL a L L
0 1
k2 ‡ l2 h
B 2 1=2 1=2
0 C
B …k ‡ l 2 † …h2 ‡ k 2 ‡ l 2 † …h2 ‡ k2 ‡ l 2 †1=2 C
B C
B ¡hk l k C
B C
ˆB C
B …k ‡ l † …h2 ‡ k 2 ‡ l 2 †1=2
2 2 1=2
…k 2 ‡ l 2 †1=2 …h2 ‡ k2 ‡ l 2 †1=2 C
B C
B C
@ ¡hl ¡k l A
…k 2 ‡ l 2 †1=2 …h2 ‡ k2‡ l 2 †1=2 …k 2 ‡ l 2 †1=2 …h2 ‡ k2 ‡ l 2 †1=2
0 1
cos ¶ sin ¶ 0
B C
B
£ @ ¡ sin ¶ cos ¶ 0C …A 4†
A:
0 0 1
Note the diŒerent convention used in this paper from that used by van Leeuwen et
al. (1999) concerning the direction for the ¶ rotation (¬ rotation in the cited paper).
620 M. Leoni et al.
h ˆ sin · sin µ;
k ˆ sin · cos µ; …A 5†
l ˆ cos ·;
(b) Rotation of S 0 about the axis S 20 through the angle to obtain S 00 : the
corresponding vector …0; ; 0† in Euler space represents the rotation matrix
0 1
cos 0 ¡ sin
B C
@ 0 1 0 A:
sin 0 cos
(c) Rotation of S 00 about the axis S 300 through the angle ® to obtain S ² C ; the
corresponding vector (0, 0, ®† in Euler space represents the rotation matrix
Strain± stress Welds in transversely isotropic thin Wlms 621
0 1
cos ® sin ® 0
B C
@ ¡ sin ® cos ® 0 A:
0 0 1
A generic rotation g ˆ …¬; ; ®† is thus a combination of the single rotations outlined
above:
g ˆ …¬; ; ®†
ˆ …0; 0; ®†…0; ; 0†…¬; 0; 0†
0 10 10 1
cos ® sin ® 0 cos 0 ¡ sin cos ¬ sin ¬ 0
B CB CB C
ˆB@ ¡ sin ® cos ® 0 A@ 0
CB 1 0 CB ¡ sin ¬
A@ cos ¬ 0C
A: …A 6†
0 0 1 sin 0 cos 0 0 1
Analogous expressions hold if the convention of Bunge (1982) is adopted for the
de® nition of the Euler angles.
APPENDIX B
model, as these tensors are set ® xed for all crystallites in equations (B 3) and (B 4)
respectively.
The elastic constants A and B describing the elastic behaviour under a rotational
symmetric strain ® eld with zero average stress components perpendicular to the ® lm
surface (same hypotheses adopted for the Vook± Witt model; see above) can be
calculated for the Voigt and the Reuss models by imposing the corresponding strain
and stress ® elds
0 1 0 1
"== 0 0 ¼== 0 0
B C B C
h"S i ˆ @ 0 "== 0 A ˆ "SVoigt ; h¼S i ˆ @ 0 ¼== 0 A ˆ ¼SReuss …B 5†
0 0 "? 0 0 0
The evaluation of A and B is then a straightforward calculation by substituting the
pair of equations (B 3) and (B 4) into equations (6) and (7). The Voigt and the Reuss
models predict macroscopically isotropic properties in the absence of texture and
transversely isotropic properties for ® bre-textured specimens (Welzel et al. 2000).
The two models cannot represent the true elastic behaviour of a polycrystalline
aggregate (apart for a body composed of a truly intrinsically elastically isotropic
material), owing to the extreme nature of the grain interaction assumptions. The
Voigt model, for example, violates mechanical equilibrium, as the stress is discontin-
uous at crystallite boundaries. Further, the macroscopic elastic constants derived
from both models are incompatible (Matthies and Humbert 1995), that is
cVoigt 6ˆ …sReuss †¡1 :
Hill (1952) proved that the Reuss and the Voigt models represent the lower and
upper bounds respectively for the elastic moduli of a macroscopically isotropic
aggregate. Therefore, it is reasonable to assume that the true average elastic con-
stants are better approximated by the arithmetic average of the values predicted by
the Reuss and the Voigt models. This approach is generally referred to as the
Neerfeld (1942)± Hill (1952) model. The Neerfeld± Hill approach is just an averaging
procedure and does not represent a speci® c grain interaction model in the sense that
speci® c grain interaction parameters can be indicated. This holds also for alternative
models, which do obey the compatibility relation c ˆ s¡1 (such as that based on the
geometric averaging of the elastic tensors (Matthies and Humbert 1995)).
References
Baron, H. U., and Hauk, V., 1988, Z. Metallkd., 79, 127.
Brakman, C. M., 1983, J. appl. Crystallogr., 16, 325; 1987, Phil. Mag. A, 55, 39.
Bunge, H.-J., 1982, Texture Analysis in Materials Science (London: Butterworth).
Diz, J., and Humbert, M., 1992, J. appl. Crystallogr., 25, 756.
Genzel, CH., 1994, Phys. stat. sol. (a), 146, 629; 1997, ibid., 159, 283; 1998, ibid., 165, 347.
Hauk, V. (editor), 1997, Structural and Residual Stress Analysis by Nondestructive Methods
(Amsterdam: Elsevier).
Hendrix, B. C., and Yu, L. G., 1998, Acta mater., 46, 127.
Hill, R., 1952, Proc. phys. Soc., 65, 349.
Kamminga, J.-D., De Keijer, Th. H., Delhez, R., and Mittemeijer, E. J., 2000, J. appl.
Crystallogr., 33, 1059.
Krigbaum, W. R., and Roe, R.-J., 1964, J. chem. Phys., 41, 737.
Ledbetter, H. M., and Naimon, E. R., 1974, J. appl. Phys., 45, 66.
Leigh, S.-H., and Berndt, C. C., 1999, Acta mater., 47, 1575.
Leoni, M., 1998, PhD Thesis, Universita di Roma `Tor Vergata’, Italy.
Leoni, M., Dong, Y. H., and Scardi, P., 1999, Mater. Sci. Forum, 321± 324, 439.
Leoni, M., Welzel, U., Lamparter, P., and Mittemeijer, E. J., 2000 (to be published).
Strain± stress Welds in transversely isotropic thin Wlms 623
Love, A. E. H., 1927, A Treatise on the Mathematical Theory of Elasticity (Cambridge Uni-
versity Press).
Machlin, E. S., 1995, Materials Science in Microelectronics, Vol. 2 (New York: Giro Press).
Matthies, S., and Humbert, M., 1995, J. appl. Crystallogr., 28, 254.
Matthies, S., Vinel, G., and Helming, K., 1987, Standard Distributions in Texture Analysis
(Berlin: Akademie).
Matthies, S., Wenk, H.-R., and Vinel, G. W., 1988, J. appl. Crystallogr., 21, 285.
Meyers, M. A., and Chawla, K. K., 1984, Mechanical Metallurgy, Principles and Applica-
tions (Englewood CliŒs, New Jersey: Prentice-Hall), pp. 57± 58.
Murakami, M., and Yogi, T., 1985, J. appl. Phys., 57, 211.
Neerfeld, H., 1942, Mitt. Kaiser-Wilhelm-Inst. Eisenforsch. DuÈ sseldorf, 24, 61.
Noyan, I. C., and Cohen, J. B., 1987, Residual Stress. Measurement by DiVraction and Inter-
pretation (New York: Springer).
Nye, J. F., 1957, Physical Properties of Crystals (Oxford University Press), p. 152.
Press, W. H., Teukolsky, S. A., Vetterling, W. T., and Flannery, B. P., 1992, Numerical
Recipes in C: the Art of ScientiWc Computing, second edition (Cambridge University
Press).
Reuss, A., 1929, Z. angew. Math. Mech., 9, 49.
Roe, R.-J., 1965, J. appl. Phys., 36, 2024.
Roe, R.-J., and Krigbaum, W. R., 1964, J. chem. Phys., 40, 2608.
Sasaki, T., Yoshioka, Y., and Kuramoto, M., 1993, Curr. Jap. Mater. Res., 10, 73.
Scardi, P., Leoni, M., and Dong, Y. H., 1999, Adv. X-ray Anal., 42, (on CD-ROM).
Stickforth, J., 1966, Tech. Mitt. Krupp, Forschungsber., 24, 89.
van Houtte, P., and de Buyser, L., 1993, Acta metall. mater., 41, 323.
van Leeuwen, M., Kamminga, J.-D., and Mittemeijer, E. J., 1999, J. appl. Phys., 86, 1904.
Voigt, W., 1910, Lehrbuch der Kristallphysik (Leipzig: Teubner).
Vook, R. W., and Witt, F., 1965, J. appl. Phys., 36, 2169.
Welzel, U., Leoni, M., Lamparter, P., and Mittemeijer, E. J., 2000 (to be published).
Wider, T., 1995a, Thin Solid Films, 256, 39; 1995b, J. appl. Phys., 78, 838; 1995c, Comput.
Phys. Comm., 85, 398; 1996, ibid., 96, 53.
Witt, F., and Vook, R. W., 1968, J. appl. Phys., 39, 2773.