- Email: [email protected]
Spinodal decomposition of Ti0.33Al0.67N thin films studied by atom probe tomography
Thin Solid Films 520 (2012) 4362–4368
Contents lists available at SciVerse ScienceDirect
Thin Solid Films journal homepage: www.elsevier.com/locate/tsf
Spinodal decomposition of Ti0.33Al0.67N thin films studied by atom probe tomography L.J.S. Johnson a,⁎, M. Thuvander b, K. Stiller b, M. Odén a, L. Hultman a a b
Dept. of Physics, Chemistry and Biology (IFM), Linköping University, SE-581 83, Sweden Dept. of Applied Physics, Chalmers University of Technology, SE-421 96, Sweden
a r t i c l e
i n f o
Article history: Received 22 September 2011 Received in revised form 15 February 2012 Accepted 23 February 2012 Available online 3 March 2012 Keywords: Titanium aluminum nitride Spinodal decomposition Atom probe tomography Thin films
a b s t r a c t Details of the phase decomposition in NaCl-structure Ti0.33Al0.67N thin films deposited by cathodic arc evaporation are studied by atom probe tomography. We demonstrate that as-deposited films are in the earliest stage of decomposition for which electron microscopy and x-ray diffraction indicate a single-phase solid solution. Annealing at 900 °C further activates spinodal decomposition of the films, although pockets of undecomposed material remain after 2 h. N preferentially segregates to the AlN and TiN domains, causing the TiAlN matrix to be understoichiometric, by the energetics of N vacancies in TiAlN. The corresponding modulation in stoichiometry implies a Kirkendall effect, caused by different Al and Ti diffusivities. © 2012 Elsevier B.V. All rights reserved.
1. Introduction TiAlN thin films are used extensively as cutting tool coatings. They exhibit age hardening as described by Hörling et al. [1,2], who also showed that the effect is caused by the decomposition of metastable cubic B1-TiAlN upon heating, such as during cutting operations. The decomposition proceeds in two steps. Firstly, Ti and Al segregates to form c-TiN and metastable c-AlN coherently strained domains, where c-AlN is stabilized by coherency forces in the first step of separation. Secondly, the c-AlN transforms, upon further annealing, into h-AlN, which is the thermodynamically stable phase of AlN at ambient pressure [3]. The apparent segregation of Ti and Al on the cubic lattice is due to the existence of a miscibility gap in the pseudobinary TiN-AlN system [2,4]. This, and a number of observations [2,5–9], point to a spinodal decomposition mechanism. The difficulty of proving the hypothesis of the operation of spinodal decomposition in the TiAlN thin film material system stems from the fact that it proceeds at very small size scales and high temperatures, for which traditional thin film microstructural characterization techniques are insufficient and results inconclusive. This situation is changing with technological progress in instrumentation. Of particular interest for this kind of problem is atom probe tomography (APT), which has the power to resolve the composition of a sample with high accuracy at high resolution [10]. Recent advances [11,12], such as the laser assisted field evaporation, local counterelectrodes and focussed ion beam (FIB) specimen preparation techniques, allow the analysis of thin films of nonconductive materials. Rachbauer et al.
⁎ Corresponding author. Tel.: + 46 13 282907. E-mail address: [email protected] (L.J.S. Johnson). 0040-6090/$ – see front matter © 2012 Elsevier B.V. All rights reserved. doi:10.1016/j.tsf.2012.02.085
[8,13] studied the separation of magnetron sputtered Ti0.5Al0.5N films by APT adding to the evidence for spinodal decomposition in the system. A more detailed description of the evolving compositional variations during the spinodal process is lacking, and furthermore, very little is known about the behavior of N during the decomposition and its influence on the same. Here, we use APT to investigate the decomposition of Ti0.33Al0.67N thin films deposited by reactive cathodic arc evaporation. The choice of this high Al content is motivated by the facts that such a composition is typical for industrial coatings and that the system is close to the center of the miscibility gap [4,14]. We discuss the separation of Ti and Al as well as the coupled changes in N stoichiometry. 2. Experimental details The films were deposited by reactive cathodic arc evaporation in a Metaplas MXR323 system with samples fixtured to a rotating drum. The substrates were polished cemented carbide (WC-Co, Seco Tools “HX”) flat inserts of dimensions 13 × 13 × 4 mm. Prior to deposition the substrates were sputter cleaned with Ar ions. During the deposition the atmosphere was pure (99.995%) N2, the substrate temperature was ~ 400 °C and a bias of −55 V was applied. The resulting films were of a B1 NaCl structure with a lattice parameter of 4.15 Å as determined by X-ray diffraction (XRD), with a total thickness of ~3 μm. Rutherford backscattering spectroscopy (RBS) gave the mean composition of the films as (Ti0.33Al0.67)N0.92, with O contamination on the level of 0.5–1 at.%. The microstructure was found to be of a dense columnar type [5]. Isothermal annealing of a sample was performed for 120 min at 900 °C in an inert argon atmosphere with a Sintervac furnace (GCA Vacuum Industries). The heat treatment parameters were chosen
L.J.S. Johnson et al. / Thin Solid Films 520 (2012) 4362–4368
3. Data analysis For the application of APT to the problem of spinodal decomposition Miller et al. [16–18] is the most recent general treatment. The main variable of interest for spinodal decomposition is the mole fraction describing the separation: x ¼ nAl =ðnAl þ nTi Þ;
ð1Þ
where n is the concentration of a species. For transition metal nitrides the stoichiometry of N may have a large effect on the properties of the alloy [19–21]; we take it as: z ¼ nN =ðnN þ nAl þ nTi Þ ≈ nN ;
ð2Þ
effectively giving (Ti1 − xAlx)1 − zNz. The two variables were computed on a regular grid of 1 nm 3 cubes by first binning all detected ions using Gaussian delocalization with the IVAS software, and then computing the field variables for each voxel. The autocorrelation functions of the field variables were computed to investigate any spatial correlations present in the data. The autocorrelation function, γ(r), for variable x was computed as: →
γð r Þ ¼
1 r ′−→r −xÞ; ∑ðx→ r ′−x Þðx→ σ 2x ðN−1Þ →
ð3Þ
r′
where x is the mean, σx the standard deviation, N the number of vox0 els, and → r ranges over all valid voxel indices in the datacube. The function was then averaged radially by integration using cylindrical coordinates: 2π
γ ðr; zÞ ¼
1 ∫ γðr cosðϕÞ;r sinðϕÞ;zÞdϕ: 2π 0
ð4Þ
Radial distribution functions (RDF) are another way to show spatial correlations. In the case that a strong spatial variation in one variable influences the one for which a RDF is sought, the RDF may be extended by taking the variation of the first variable into account and binning the RDF: let g ia(r) be the RDF of variable a for atom i and g ia its average up to the radius r0: i
i
g a ðr Þ ¼ g a ðr Þr ; r∈½0; r 0 :
ð5Þ
Then the binned RDF of variable b is: i
i
g b ðr; aÞ ¼ g b ðr Þi ; ∀i : g a ∈½a−Δa; a þ Δa:
the instrument used in this study). Danoix et al. [22] analyzed this situation in detail for the case of a two-component system. They showed that the variance of the measured atomic composition, c, of a box is: 2
σ c ¼ p0 ð1−p0 Þð1−ηÞ=n;
ð7Þ
where p0 is the true composition, η is the detection efficiency, and n is the mean number of ions per box (atomic density per box, m, times the efficiency), for reasonable detection efficiencies and box sizes (m ≥ 100). In order to test the hypothesis that a sample is homogeneous, an empirical test distribution was created by simulating the model described above. A sample was considered to consist of 40 000 sample cubes (1 nm 3 in volume) of a B1 structure of appropriate density, the variance of the composition variable was calculated for each sample. Two thousand (2000) simulations were performed, and the distribution of variances was calculated, giving a p-value resolution of 0.0005. This allowed the testing of the measured variance of the composition of samples against the null hypothesis that the observed variance could be explained by the variance inherent in the measurement, by directly calculating the probability that the observed variance was in accordance with the null hypothesis. 4. Results 4.1. Mass spectra and evaporation behavior A typical mass spectrum from a Ti0.33Al0.67N film is shown in Fig. 1, and the corresponding charge state ratios are given in Table 1. Ions of Ti (doubly and triply charged), Al (singly to triply charged), and N (singly and doubly charged) are present. Of note is the large amount of complex ions, where TiN 2 + is the dominating complex, but AlN 2 + and AlO 2 + are present as well. No peaks corresponding to either O or TiO complexes were observed. Tied to the high number of complexes were a corresponding amount of multiple evaporation events per laser pulse. Finally, although the Al 2 + and N + peaks are well separated, the tail of the Al 2 + peak overlaps with the N + peak to a significant degree, most probably due to the limited thermal conductance of the film. This slows down the cooling of the APT tip after heating by each laser pulse, which increases the time window for evaporation, which in turn leads to the broadening of the peak tails. The composition of the samples, obtained by peak decomposition using natural isotope abundances and statistical averaging, is shown in Table 2. This method uses the whole spectrum, and as such it is not applicable to the ranging of ions for the volume reconstruction, which will be worse at resolving the overlap of the Al 2 + and N + peaks (compare with Fig. 3). The metal pseudobinary ratio, x, in Ti1 − xAlxN, is close to 0.6 for both samples. This is below the expected 0.67 value as confirmed by RBS measurements. This
106 3+
Al
105
counts
according to the results of Hörling et al. [5] to produce as much c-AlN as possible before transformation to the stable h-AlN occurs. Samples for APT were prepared using the FIB lift-out technique [15] where wedges roughly 3 μm high and 1–2 μm wide are attached to posts fabricated out of a silicon wafer. The wedges were sharpened by annular milling at 30 keV and – for the last stage – 5 keV Ga + ions. The samples were analyzed in a local electrode atom probe (Imago LEAP 3000X HR) in laser pulsing mode with a pulse rate of 200 kHz, a pulse energy of 0.35 nJ, and a target evaporation rate of 0.5% (expressed as detected ions per pulse) at 65 K. The laser energy was chosen after trials to provide the optimum mass resolution. Four tips from the as-deposited state and six from the annealed state were analyzed. For reconstruction, an evaporation field of 40 V/nm, image compression factor of 1.65 and a field factor of 3.3 were used, which gave final tip radii close to or in agreement with measurements by scanning electron microscopy on the tips after evaporation.
4363
10
Al2+ N+
The field variables of interest are sensitive to the sampling nature of the technique, as the collection efficiency is less than one (0.37 for
TiN2+
4
AlN2+
N2+
1000 100 10
ð6Þ
N2+ Ti2+ Al+
Ti3+
AlO2+ 5
10
15
20
25
30
m q a.u. Fig. 1. Laser atom probe mass spectrum from Ti0.33Al0.67N.
35
L.J.S. Johnson et al. / Thin Solid Films 520 (2012) 4362–4368
a
Table 1 Charge state ratios corresponding to Fig. 1. X+
X3 +
XN2 +
X+ 2
Ti2 + Al2 + N2 +
– 5.7 0.01
2.2 4.9 –
2.6 78.4 Al: 0.30 Ti: 0.03
– – 0.01
60 0.8 50 0.6
40
y nm
X2 +
1.0
deviation is due to the overlap of Al and N peaks in the APT data, which reduces the detected amount of Al. The N and impurity O concentrations indicate close to, but slightly less than, stoichiometric films, in agreement with the RBS measurements.
30
0.4
20
Al Al Ti fraction
4364
0.2 10 0
4.2. Decomposition
0
10
20
30
40
50
0.0
60
x nm
Table 2 Composition determined from the spectra, and corresponding ±2σ intervals, all values are given in atomic percent. Sample
Ti
Al
N
O
As-deposited Annealed
20.3 ± 0.8 20.9 ± 2.0
29.1 ± 1.4 29.8 ± 2.6
49.4 ± 1.8 48.9 ± 1.7
1.2 ± 0.5 0.4 ± 0.4
1.0
b 60
0.6
y nm
40 30
0.4
20
Al Al Ti fraction
0.8 50
0.2 10 0
0
10
20
30
40
50
0.0
60
x nm Fig. 2. 2D concentration maps of the Al fraction of the metal content of Ti0.33Al0.67N on slices through reconstructed samples in the xy-plane, for the (a) as-deposited and (b) annealed states.
pure TiN or AlN is also small. Almost all of the investigated specimen is in an intermediate stage of decomposition. In order to address the question of whether the observed fluctuations in the as-deposited film were true fluctuations in the film or measurement artifacts, the estimated variance of the Al fraction was tested against the hypothesis of a homogeneous film, according to Section 3. The result is given in Table 3: the estimated standard deviation was 0.12, and the theoretical standard deviation was calculated to 0.08. The variance is significantly larger than the expected
as deposited annealed
0.15
Relative amount
The variation of the composition on the metal sublattice is shown in Fig. 2 for the as-deposited and annealed samples. The two reconstructed tips are representative for the two heat treatment states, unless specifically noted in the text, and are used throughout the rest of the figures for direct comparison. The figures show concentration maps taken from slices through the reconstructed atom probe data, and are typical for each tip. The as-deposited film (Fig. 2(a)) exhibits small fluctuations around the mean composition, with no obvious wavelength or ordered structure, throughout the whole tip. The annealed sample (Fig. 2(b)), on the other hand, shows a clear segregation of Al and Ti into domains predominantly composed of TiN or AlN. The composition has a non-zero gradient over most of the slice, except at the extremum points of fully separated TiN and AlN. The Al- and Ti-enriched domains exhibit varying morphologies, which mostly deviate significantly from the spherical case. Furthermore, the Al-rich domains show internal compositional fluctuations, whereas the Ti-rich domains are homogeneous in the sense that they are compact and without saddle points in the composition. Beside the regions of segregation there is a large region on the right side of the slice in Fig. 2(b) which extends at an angle through the tip, that appears similar in nature to the as-deposited sample, and which was only observed in this particular tip. This indicates that after the heat treatment there exist regions in the sample that have not undergone phase decomposition. Histograms of the volume composition of the metal fraction for the two runs are shown in Fig. 3. The histograms presented here are of the smoothed data used to generate the compositional maps, and as such do not show much of the instrumental variation due to the sampling. Instead, the compositional variation due to the respective up-hill diffusion of Ti and Al is dominating. The as-deposited sample shows a simple normal-like distribution around a compositional mean of x = 0.57, which deviates from the known sample composition, see Section 5.1. The annealed sample again shows its decomposed structure together with the remaining TiAlN solid solution regions, and the histogram may be understood as the sum of three distinct components. The first two correspond to the formation of Al- and Ti-enriched domains (1 and 2 in Fig. 3), while the third and middle peak (marked 3 in Fig. 3) correspond to the undecomposed regions, which are similar to what is observed in the as-deposited sample. As the heights at the ends of the histogram, corresponding to x = 0 and x = 1, are vanishingly small, the volume fraction of
0.10
0.05 3 1
2 0.00 0.0
0.2
0.4
0.6
0.8
1.0
Al Al Ti fraction Fig. 3. Histograms over volume fractions for the metal sublattice concentration of Al in Ti0.33Al0.67N in as-deposited and annealed states.
L.J.S. Johnson et al. / Thin Solid Films 520 (2012) 4362–4368
Ad
x z Annealed z
0.57 0.49 0.51
0.12 0.09 0.12
Homogeneous std. dev.
p-value
0.08 0.06 0.06
b0.0005 b0.0005 b0.0005
60
0.65
50
0.60 0.55
40
0.50
30
0.45 0.40 10
0.35
0 0
10
20
30
40
50
0.30
60
x nm
b
y nm
fluctuations in a reconstructed homogeneous film with a p-value of less than 0.0005. Accordingly, we ascribe the observed fluctuations, to a significant part, to a true fluctuation of the composition in the sample. Fig. 4 shows the autocorrelation functions of the metal pseudobinary composition for the as-deposited and annealed tips. The autocorrelation was calculated as shown in Section 3, and radially averaged in the xy-plane of the reconstruction to avoid smearing caused by slightly different length scales in the reconstruction. The autocorrelation of the as-deposited film shows only a gradual decrease as the lag increases, which indicates that the compositional variations seen above in Fig. 2(a) have no dominating length-scale. The autocorrelation for the annealed sample instead exhibits a variation with a period of ~ 25 nm, which is present up to 60 nm in lag. This period can be seen in the sample, even though the undecomposed regions act as dampeners to the signal.
20
0.70 60
0.65
50
0.60 0.55
40
0.50
30
0.45 20 0.40
4.3. Stoichiometry
10 Fig. 5 shows the variation in nitrogen content in the same slices as in Fig. 2. Both samples exhibit fluctuations in the nitrogen content, and hence the stoichiometry, as the total concentration is normalized to unity for each reconstructed cube. There is no apparent difference in the structures of the fluctuations between the two samples. From the corresponding histograms in Fig. 6 it is apparent that the mean values differ slightly, with the as-deposited tip exhibiting a lower mean of z = 0.49 versus z = 0.51 for the annealed tip. The width and shape of the distributions are, however, almost identical. The estimated standard deviations and the corresponding theoretical deviations are again given in Table 3, together with the obtained p-values for the observed deviations. The observed standard deviations, for both films, are significantly larger than expected (p-values less than 0.0005) under the assumption of homogeneity (see Section 3). The autocorrelation functions of the stoichiometry variable, z, are shown in Fig. 7 for the two runs. The autocorrelation function obtained from the as-deposited film shows a slow decrease with an extended tail, with no clear oscillations present, similar to the autocorrelation function for the Al–Ti variation in Fig. 4. The autocorrelation function
0.06
0
0.03 0.02 0.01
0 10 20 30 40 50 60 70
10
20
30
40
50
0.30
60
x nm Fig. 5. 2D-concentration maps of the N content in Ti0.33Al0.67N of slices through reconstructed samples in the xy-plane, for the (a) as-deposited and (b) heat treated states.
of the annealed film, on the other hand, exhibits a steep fall – of the order of one sampling cube (1 nm 3) – coupled with a small and rapid oscillation. Due to the abrupt nature of the fall it is probable that the oscillations are noise and not true features of the underlying data. That notwithstanding, the sharp decrease itself indicates a marked difference to the as-deposited film. Fig. 8 shows the radial distribution functions (RDF) of the N content calculated for Ti and Al sites for both the as-deposited and the annealed samples. The RDF of the as-deposited sample shows no difference between Ti and Al, except at zero offset, where N is more often found together with a Ti atom than with an Al. This is most
as deposited annealed
0.10
Relative amount
0.002 0.001 0.000 0.001 0.002 0.003
0.04
0.35
0
as deposited annealed
0.05
N concentration at.
Variable Mean Standard deviation
0.70
N concentration at.
Sample
a
y nm
Table 3 Statistical analysis of variations in metal fraction and stoichiometry for both the asdeposited and the annealed samples.
4365
0.08 0.06 0.04 0.02
0.00 0
10
20
30
40
50
60
70
r nm
0.00 0.3
0.4
0.5
0.6
0.7
N content (at. fraction) Fig. 4. Autocorrelation functions of the metal sublattice occupancy of Al in Ti1 − 0.33Al0.67N, for the as-deposited and annealed states. The inset is a magnified view of the annealed state data.
Fig. 6. Histograms over volume fractions for N concentration in Ti0.33Al0.67N in asdeposited and annealed states.
4366
L.J.S. Johnson et al. / Thin Solid Films 520 (2012) 4362–4368
1.0
as deposited annealed
N Composition (at. %)
0.8
60
0.6 0.4 0.2 0.0
50 40
Ti
20
Al
10
N
2
1 0
1
2
0 0
10
20
30
40
50
60
-4
70
r nm
likely due to the preferred evaporation of Ti as TiN–ion complexes (c.f. Fig. 1) and hence an artifact of the measurement technique. The first maximum of the Al RDF lies at slightly less than 2 Å, close to the nearest neighbor distance (200) in TiN. For the annealed film the situation is more complex; the RDF is shown binned according to the Al–Ti fraction (x) around each metal atom (see Eqs. (6–7), r0 = 0.5 nm), which gives the 3D plot shown in Fig. 8b, with metal composition on the x-axis, and the RDFs on the y–z axes. From this plot it is apparent that the metal composition determines the local N-concentration, with a nonlinear variation with the Al–Ti fraction. At the Ti-rich side (x = 0) the stoichiometry (z) is ~0.49, which decreases as x increases up to the minimum at x = 0.7 of ~ 0.47, followed by an increase again up to x = 1.0. Due to the overlap of the Al 2 + and
a 49.2 49.0 48.8 48.6 48.4 48.2
Al Ti 0
1
2
3
4
distance (nm) 4
b
3
distance (nm) 2
0 50 49 48 1.0
N concentration (at. %)
1
0.5 0.0
-2
0
2
4
Distance (nm)
Fig. 7. Autocorrelation functions of the N content in Ti0.33Al0.67N, for the as-deposited and annealed states.
N concentration (at. %)
54 53 52 51 50 49
30
Al/(Al+Ti) fraction
Fig. 8. Radial distribution functions of N around Ti and Al sites in Ti0.33Al0.67N, for the (a) as-deposited and (b) annealed states, which are binned according to the Al fraction at the site of the originating atom.
Fig. 9. Proxigram from the 3:2 Al isoconcentration surface of Ti0.33Al0.67N annealed at 900 °C, showing the mean gradients of the decomposition of Al, Ti and N. The inset shows a magnified part of the N signal.
N + peaks, the N concentration is overestimated at large Al contents. Extracting the regions with x > 0.9 and analyzing the mass spectrum gives a N concentration of 49 at.%, comparable to that at the Ti side. At lags greater than zero (distances > 0), the N concentration converges to the global level of around 48 at.% for all Al–Ti compositions, with the Al side converging slower than the Ti side, as the Al rich domains are larger in size. The two lines drawn down the graph marks the conventional RDFs that are obtained if the composition variations are not taken into account, which are clearly misleading in this case. A proxigram from the annealed sample corresponding to the isoconcentration surface of x = 0.6 is shown in Fig. 9. Due to the variation in segregation shown above, the proxigram was constructed from a part of the sample showing pronounced concentration gradients, corresponding to the lower left part of Fig. 2 (b). The Al and Ti curves show the gradient on the metal sublattice. It is asymmetric with longer tails into the Ti-rich domains than into the Al-rich domains, and it is approximately 5 nm wide. Furthermore, the Ti content in the AlN-rich domains is higher (~6 at.%) than the Al content in the TiN-rich domains (~2–3 at.%). The increase in the variance at increasing distances from the isosurface is due to the decreasing volumes of high segregation and their geometrically closed shape. The variation of the stoichiometry is most clearly seen in the N profile, where a variation is seen when going along the gradient of the metal composition, similar to that seen in the RDF curves (Fig. 8b). There is a fluctuation in the region with a large gradient in the Al and Ti contents, consisting of a sinusoidal variation with the maximum on the Ti-rich side, the minimum on the other side and the inflection point almost perfectly centered on the isoconcentration surface. Although small, of the order of 1–2 at.%, the variation in N content with metal fraction is significant with regard to errors of the measurement as the total number of binned ions is highest near the isosurface, and the corresponding error of estimation is consequently small. 5. Discussion 5.1. Evaporation behavior The field evaporation of Ti–Al–N reported here, with a relatively low laser power and, accordingly, a relatively high applied DC field, resulted in a spectrum with a large amount of ion complexes. Particularly, TiN 2 +, N2+, and AlN 2 + ions were detected. A similar behavior was reported by Ai et al. [23], who used a laser assisted 1D atom probe field ion microscope to investigate TiN thin films. Although the laser power was not reported, the applied DC field was ~30 V/ nm, in comparison with the ~40 V/nm used in this study. Ai observed large amounts of N2+ and TiN 2 +, as well as the elemental ions Ti 2 +,
L.J.S. Johnson et al. / Thin Solid Films 520 (2012) 4362–4368
Ti 3 +, and N +. Additionally, Ti2N 3 + ions were reported, which were not observed in this study. The films were not phase-pure B1 TiN, but contained a minority phase of Ti2N, which could explain the Ti2N 3 + ions, although the lower field should also promote more ion complexes. The presence of TiN complexes in the spectra might be understood with regards to the binding of Ti and N, where N almost solely binds to its nearest neighbor Ti atoms, while Ti atoms also bond with the second nearest Ti neighbors [24], and this also helps explain the large amount of multiple evaporation events that were observed here. The peaks in the spectrum in Fig. 1 are mostly well separated, with the exception of the Al 2 + and N + peaks. While clearly resolved, the tail of the Al 2 + peak overlaps the N + peak significantly. This will cause some error in the classification (“ranging”) of ions as some Al 2 + ions will unavoidably be classified as N + ions, and, to a lesser degree, N + will be wrongly classified as Ti 3 +. This leads to the overestimation of N and Ti by ~ 2–3 at.% in the reconstruction. This is a well-known issue with laser APT, and is being addressed by finer control of the zone heated by the laser in newer instruments. The tail is also significantly influenced by the thermal conductance of the sample, and as such will vary with microstructure, crystal quality, etc. [25]. 5.2. The as-deposited state The presence of compositional modulations on the metal lattice in the as-deposited sample is not surprising. Generally, TiAlN films deposited with compositions in the miscibility gap show the same trend, with increasing modulations for increasing Al content and deposition temperatures [26,27]. Films deposited by cathodic arc evaporation exhibit a single phase NaCl structure for Al fractions as high as x = 0.7. This is due to the deposition conditions with a high flux of energetic ions [27], which produces recoil mixing in the growing film [28]. Such mixing counteracts surface segregation, which is bound to take place even at the modest temperatures of deposition (~450 °C). So the small variations seen in the sample should be expected from steady-state conditions. Furthermore, the fact that no clear wavelength of decomposition was observed may be understood by arguments from the theory of spinodal decomposition. The wavelength of the composition modulation of a spinodally decomposed material is an emergent property; the wavelength with the largest driving force will outgrow all other wavelengths [29]. As no wavelength dominates in the as-deposited sample, the conclusion must be that the segregation observed here is in its very earliest stage. The statistical analysis of the stoichiometry fluctuations in the asdeposited sample (see Table 3) indicates that they are not only due to the experimental sampling, yet no dominating frequency was observed in the autocorrelation function of the N content. The RDFs of N from Ti and Al sites furthermore do not show any correlation, except for the Ti–N correlation at 0 nm offset due to the preferred evaporation of TiN complexes. Due to the high energies of the incident ions during the deposition process (~100 eV [28]), a high number of vacancies are likely present in the film [30–32]. The stoichiometry measured by the atom probe is that of the detected ions, and as such is an average of both nitrogen and metal atom vacancies. Therefore, the observed fluctuations originate from both sublattices, although the N vacancies should dominate by simple thermodynamical considerations due to their lower energy of formation. Since the variation in stoichiometry cannot be coupled to variations of the Al fraction on the metal lattice, they must be ascribed to random variations due to the deposition process. 5.3. The annealed state: decomposition The annealed sample has a clearly decomposed structure, along with parts that are essentially the same as that of the as-deposited sample.
4367
These undecomposed regions, as previously detected by XRD measurements [2,5], and then connected to an undecomposed matrix containing particles of TiN and AlN, are here found to be separate from the decomposed regions. The gradient of the metal composition is nonzero for the majority of the volume, except at the fully decomposed points, which are essentially negligible when compared to the rest of the sample (as can be seen from the histogram in Fig. 3). Moreover, the autocorrelation function in Fig. 4 shows that the decomposition has evolved a periodic structure in the sample. It is also instructive to consider that the domains are not of constant composition; there are no growth fronts where precipitates consume the matrix phase or other structures tied to a nucleation and growth mechanism. Calculations on the TiN–AlN system show the existence of a miscibility gap [4,9] and this has been observed in numerous experiments [2,5–8,26,27]. There are indications in the literature of spinodal decomposition in the TiAlN system, for example [2,6–8,26,33–35], with structures similar to those observed here. By definition, the only process that could give rise to this kind of configuration is spinodal decomposition, which is consistent with the fact that the Ti0.33Al0.67N alloy composition lies within the coherent spinodal [4]. The existence of a region of relatively homogeneous composition in the annealed sample (c.f. Fig. 2(b)) indicates that the decomposition can be locally inhibited, and thus has a microstructural origin. The mean column width of these samples is 200–300 nm [5], and while no grain boundaries were observed in this sample, the reconstructed diameter of ~ 70 nm makes it probable that at least one side would be close to a grain boundary. For sputtered Ti0.5Al0.5N Rachbauer [13] observed that the decomposition was almost entirely limited to the grain boundaries at similar heat treatments as used here. Therefore, the proximity of a grain boundary to the side of the probed volume appears to be the most likely explanation of the observed variation in decomposition activity in our annealed sample. The decomposed structure visible in Fig. 2 is not isotropic with radially-symmetric domains. Instead, the domains have a have a mean elongation along a diagonal in the figure. The theory of spinodal decomposition predicts that the degree of anisotropy in the decomposition is determined by the anisotropy of the elastic constants [36]. Both experiments and calculations of the elastic anisotropy of TiN and TiAlN show that there is a considerable anisotropy in the system [37–39], so anisotropy in the compositional variations is expected, although the lack of crystallographic information prevents further comparison with theory. 5.4. Annealed state: stoichiometry variations While both samples show fluctuations in the stoichiometry in Fig. 5 with similar variances (Table 3 and Fig. 6), the annealed sample exhibits a correlation between the N distribution and the composition of the metal lattice (c.f. Figs. 8 and 9). The N signal is more or less equal at the endpoints x = 0 and x = 1, while there is a clear drop around x = 0.7. The observed variations in stoichiometry may stem from two sources in a TiN-like material. Firstly, there will be vacancies on the N sublattice, and secondly – but to a lesser degree, due to the higher energy cost – vacancies on the metal sublattice. Both of these vacancy types will be registered as a variation in stoichiometry in the APT data. B1-TiNz is thermodynamically stable over a wide range of stoichiometries [40]. Wurtzite AlN, on the other hand, is a line compound with a very narrow composition range [41]. Alling et al. [42] calculated the formation energies of (Ti1 − xAlx)Nz for z b 1, and found that the formation energy surface is convex (has negative curvature) for values of z close to 1. These conditions offer an explanation for the observed behavior of the stoichiometry. For a twophase system, the expected outcome is stoichiometric AlN and understoichiometric TiN, which is the most energetically favorable state. This would be the case for a fully completed decomposition, while here the system is dominated by the regions with a composition
4368
L.J.S. Johnson et al. / Thin Solid Films 520 (2012) 4362–4368
gradient (c.f. Fig. 3). Modeling this by allowing a three component system to relax over the energy landscape in Ref. [42] directly leads to stoichiometric AlN, TiN, and an understoichiometric (Ti1 − xAlx)Nz, which, by the nature of the convex energy surface, lowers the total energy of the system. This finding is important, as it leads to a reduction of the strain energy of the spinodal system since it reduces the lattice mismatch in the composition gradient, and thus facilitates the decomposition. This is in agreement with the findings by Alling et al. [42]; that understoichiometric Ti1 − xAlxN should exhibit an enhanced tendency for coherent decomposition. A similar sinusoidal feature as the one obtained here (see Fig. 9) may be observed in Fig. 5 (b) of Rachbauer's work [43]. In that figure the variation is composed of one major variation, where the TiN-rich part is overstoichiometric in N and the AlN-rich part is stoichiometric (which we attribute to the line-compound nature of AlN), and a sinusoidal fluctuation with the maximum on the Ti-rich side of the gradient and the minimum on the Al-rich side. Interestingly, the modulation that we read out of Rachbauer's study is present even though their film is globally overstoichiometric with respect to N, and as such the decomposed TiN phase would become even more overstoichiometric from absorbing the N ejected from the AlN. This is a complementary case to the films studied in this work. As previously mentioned, the variations in stoichiometry may be due to vacancies on both lattices, so we should consider explanations for both cases. Attributing the modulation in Fig. 5 (b) in Ref. [43] solely to N vacancies is problematic as there is no obvious reason for the segregation of N vacancies on the level of ~ 1 at.% on the Al-rich side of the decomposition gradient. Another possibility is to ascribe the observed modulation to variations in the metal vacancy concentration. Since metal vacancies are directly involved in the spinodal decomposition (by virtue of the substitutional diffusion), dynamical effects need to be considered. By making the reasonable assumption that the Al mobility is higher than the Ti mobility (as the atomic radius of Al is smaller than that of Ti, and that the melting or dissociation temperature of AlN is lower than that of TiN, which corresponds to that the metal–N bond is weaker for Al–N than for Ti–N [38]), a Kirkendall effect is introduced. That implies a net flow of metal vacancies from the Al-rich side of the composition gradient to the Ti-rich side, leaving the Al side understoichiometric and the Ti side overstoichiometric in N. This would explain the observed variation, but would also require a significant mean concentration of vacancies to be able to support the observed maximum overstoichiometry. The main source of vacancies here is the cathodic arc deposition process, which yields a high defect density in the form of dislocations and lattice point defects [28,44]. For comparison, HfN vacancy concentrations as high as 10–15% have been observed during PVD processing [30]. 6. Conclusions Laser-assisted APT of Al-rich c-TiAlN thin films deposited by cathodic arc evaporation has been carried out for both as-deposited and annealed samples. The sample evaporation behavior is similar to that reported for TiN. Spinodal decomposition towards TiN and AlN is detected readily in the alloy sample annealed at 900 °C for 2 h. The as-deposited sample exhibits variations in composition on the metal sublattice of the order of 1–2%, because of the limited thermal activation during film deposition. A stoichiometry variation tied to the phase decomposition is detected in the annealed samples. Here, N segregates preferentially to AlN and TiN, making them closer to stoichiometric, leaving the interfacial TiAlN matrix understoichiometric, while no loss of N is detected. This effect mitigates the coherency strains between the isostructural AlN and TiN domains. This modulation in the stoichiometry along the decomposition gradient may in part be due to a Kirkendall effect, from unbalanced Al and Ti diffusivities in TiAlN.
Acknowledgments The Swedish Research Council and the VINN Excellence Center on Functional Nanoscale Materials are acknowledged for financial support. Björn Alling is acknowledged for fruitful discussions. In memoriam Bo Jansson.
References [1] A. Hörling, L. Hultman, M. Odén, J. Sjölén, L. Karlsson, Surf. Coat. Technol. 191 (2005) 384. [2] P.H. Mayrhofer, A. Hörling, L. Karlsson, J. Sjölén, T. Larsson, C. Mitterer, L. Hultman, Appl. Phys. Lett. 83 (2003) 2049. [3] Q. Xia, H. Xia, A.L. Ruoff, J. Appl. Phys. 73 (1993) 8198. [4] B. Alling, A.V. Ruban, A. Karimi, O. Peil, S.I. Simak, L. Hultman, I.A. Abrikosov, Phys. Rev. B 75 (2007) 45123. [5] A. Hörling, L. Hultman, M. Odén, J. Sjölén, L. Karlsson, J. Vac. Sci. Technol., A 20 (2002) 1815. [6] A. Knutsson, M.P. Johansson, P.O.A. Persson, L. Hultman, M. Odén, Appl. Phys. Lett. 93 (2008) 143110. [7] M. Odén, L. Rogström, A. Knutsson, M.R. Terner, P. Hedström, J. Almer, J. Ilavsky, Appl. Phys. Lett. 94 (2009) 053114. [8] R. Rachbauer, E. Stergar, S. Massl, M. Moser, P.H. Mayrhofer, Scr. Mater. 61 (2009) 725. [9] P.H. Mayrhofer, L. Hultman, J.M. Schneider, P. Staron, H. Clemens, Int. J. Mater. Res. 98 (2007) 1054. [10] T.F. Kelly, M.K. Miller, Rev. Sci. Instrum. 78 (2007) 31101. [11] T.F. Kelly, T.T. Gribb, J.D. Olson, R.L. Martens, J.D. Shepard, S.A. Wiener, T.C. Kunicki, R.M. Ulfig, D.R. Lenz, E.M. Strennen, E. Oltman, J.H. Bunton, D.R. Strait, Microsc. Microanal. 10 (2004) 373. [12] D.J. Larson, D.T. Foord, A.K. Petford-Long, T.C. Anthony, Ultramicroscopy 75 (1998) 147. [13] R. Rachbauer, S. Massl, E. Stergar, D. Holec, D. Kiener, J. Keckes, J. Patscheider, M. Stiefel, H. Leitner, P.H. Mayrhofer, J. Appl. Phys. 110 (2011) 023515. [14] P.H. Mayrhofer, D. Music, J.M. Schneider, J. Appl. Phys. 100 (2006) 094906. [15] K. Thompson, D. Lawrence, D.J. Larson, J.D. Olson, T.F. Kelly, B. Gorman, Ultramicroscopy 107 (2007) 131. [16] M.K. Miller, J.M. Hyde, M.G. Hetherington, A. Cerezo, G.D.W. Smith, C.M. Elliott, Acta Metall. Mater. 43 (1995) 3385. [17] J.M. Hyde, M.K. Miller, M.G. Hetherington, A. Cerezo, G.D.W. Smith, C.M. Elliott, Acta Metall. Mater. 43 (1995) 3403. [18] J.M. Hyde, M.K. Miller, M.G. Hetherington, A. Cerezo, G.D.W. Smith, C.M. Elliott, Acta Metall. Mater. 43 (1995) 3415. [19] L.E. Toth, Transition Metal Carbides and Nitrides, Academic Press, New York, 1971. [20] C. Shin, D. Gall, N. Hellgren, J. Patscheider, I. Petrov, J.E. Greene, J. Appl. Phys. 93 (2003) 6025. [21] L. Rogström, L.J.S. Johnson, M.P. Johansson, M. Ahlgren, L. Hultman, M. Odén, Scr. Mater. 62 (2010) 739. [22] F. Danoix, G. Grancher, A. Bostel, D. Blavette, Ultramicroscopy 107 (2007) 734. [23] C.F. Ai, J.Y. Wu, C.S. Lee, Vacuum 44 (1993) 99. [24] C. Ciobanu, D. Tambe, V. Shenoy, Surf. Sci. 582 (2005) 145. [25] J.H. Bunton, J.D. Olson, D.R. Lenz, T.F. Kelly, Microsc. Microanal. 13 (2007) 418. [26] F. Adibi, I. Petrov, L. Hultman, U. Wahlström, T. Shimizu, D. Mcintyre, J.E. Greene, J.E. Sundgren, J. Appl. Phys. 69 (1991) 6437. [27] U. Wahlström, L. Hultman, J.E. Sundgren, F. Adibi, I. Petrov, J.E. Greene, Thin Solid Films 235 (1993) 62. [28] A. Anders, Vacuum 67 (2002) 637. [29] J.W. Cahn, Acta Metall. 9 (1961) 795. [30] A.J. Perry, J. Vac. Sci. Technol., A 6 (1988) 2140. [31] G. Smith, P. Swift, A. Bendavid, Appl. Phys. Lett. 75 (1999) 630. [32] L. Tsetseris, N. Kalfagiannis, S. Logothetidis, S. Pantelides, Phys. Rev. B 76 (2007) 224107. [33] S. Lee, B. Kim, H. Kim, J. Lee, J. Appl. Phys. 80 (1996) 1469. [34] A. Kimura, H. Hasegawa, K. Yamada, T. Suzuki, Surf. Coat. Technol. 120–121 (1999) 438. [35] J.Y. Rauch, C. Rousselot, N. Martin, Surf. Coat. Technol. 157 (2002) 138. [36] J.W. Cahn, Acta Metall. 14 (1966) 1685. [37] J. Kim, J. Achenbach, P.B. Mirkarimi, M. Shinn, S.A. Barnett, J. Appl. Phys. 72 (1992) 1805. [38] F. Tasnádi, I.A. Abrikosov, L. Rogström, J. Almer, M.P. Johansson, M. Odén, Appl. Phys. Lett. 97 (2011) 231902. [39] Y. Yang, H. Lu, C. Yu, J. Chen, J. Alloys Compd. 485 (2009) 542. [40] H. Wriedt, J. Murray, J. Phase Equilib. 8 (1987) 378. [41] H. Wriedt, J. Phase Equilib. 7 (1986) 329. [42] B. Alling, A. Karimi, L. Hultman, I.A. Abrikosov, Appl. Phys. Lett. 92 (2008) 71903. [43] R. Rachbauer, S. Massl, E. Stergar, P. Felfer, P.H. Mayrhofer, Surf. Coat. Technol. 204 (2010) 1811. [44] H. Ljungcrantz, L. Hultman, J.E. Sundgren, L. Karlsson, J. Appl. Phys. 78 (1995) 832.
Contents lists available at SciVerse ScienceDirect
Thin Solid Films journal homepage: www.elsevier.com/locate/tsf
Spinodal decomposition of Ti0.33Al0.67N thin films studied by atom probe tomography L.J.S. Johnson a,⁎, M. Thuvander b, K. Stiller b, M. Odén a, L. Hultman a a b
Dept. of Physics, Chemistry and Biology (IFM), Linköping University, SE-581 83, Sweden Dept. of Applied Physics, Chalmers University of Technology, SE-421 96, Sweden
a r t i c l e
i n f o
Article history: Received 22 September 2011 Received in revised form 15 February 2012 Accepted 23 February 2012 Available online 3 March 2012 Keywords: Titanium aluminum nitride Spinodal decomposition Atom probe tomography Thin films
a b s t r a c t Details of the phase decomposition in NaCl-structure Ti0.33Al0.67N thin films deposited by cathodic arc evaporation are studied by atom probe tomography. We demonstrate that as-deposited films are in the earliest stage of decomposition for which electron microscopy and x-ray diffraction indicate a single-phase solid solution. Annealing at 900 °C further activates spinodal decomposition of the films, although pockets of undecomposed material remain after 2 h. N preferentially segregates to the AlN and TiN domains, causing the TiAlN matrix to be understoichiometric, by the energetics of N vacancies in TiAlN. The corresponding modulation in stoichiometry implies a Kirkendall effect, caused by different Al and Ti diffusivities. © 2012 Elsevier B.V. All rights reserved.
1. Introduction TiAlN thin films are used extensively as cutting tool coatings. They exhibit age hardening as described by Hörling et al. [1,2], who also showed that the effect is caused by the decomposition of metastable cubic B1-TiAlN upon heating, such as during cutting operations. The decomposition proceeds in two steps. Firstly, Ti and Al segregates to form c-TiN and metastable c-AlN coherently strained domains, where c-AlN is stabilized by coherency forces in the first step of separation. Secondly, the c-AlN transforms, upon further annealing, into h-AlN, which is the thermodynamically stable phase of AlN at ambient pressure [3]. The apparent segregation of Ti and Al on the cubic lattice is due to the existence of a miscibility gap in the pseudobinary TiN-AlN system [2,4]. This, and a number of observations [2,5–9], point to a spinodal decomposition mechanism. The difficulty of proving the hypothesis of the operation of spinodal decomposition in the TiAlN thin film material system stems from the fact that it proceeds at very small size scales and high temperatures, for which traditional thin film microstructural characterization techniques are insufficient and results inconclusive. This situation is changing with technological progress in instrumentation. Of particular interest for this kind of problem is atom probe tomography (APT), which has the power to resolve the composition of a sample with high accuracy at high resolution [10]. Recent advances [11,12], such as the laser assisted field evaporation, local counterelectrodes and focussed ion beam (FIB) specimen preparation techniques, allow the analysis of thin films of nonconductive materials. Rachbauer et al.
⁎ Corresponding author. Tel.: + 46 13 282907. E-mail address: [email protected] (L.J.S. Johnson). 0040-6090/$ – see front matter © 2012 Elsevier B.V. All rights reserved. doi:10.1016/j.tsf.2012.02.085
[8,13] studied the separation of magnetron sputtered Ti0.5Al0.5N films by APT adding to the evidence for spinodal decomposition in the system. A more detailed description of the evolving compositional variations during the spinodal process is lacking, and furthermore, very little is known about the behavior of N during the decomposition and its influence on the same. Here, we use APT to investigate the decomposition of Ti0.33Al0.67N thin films deposited by reactive cathodic arc evaporation. The choice of this high Al content is motivated by the facts that such a composition is typical for industrial coatings and that the system is close to the center of the miscibility gap [4,14]. We discuss the separation of Ti and Al as well as the coupled changes in N stoichiometry. 2. Experimental details The films were deposited by reactive cathodic arc evaporation in a Metaplas MXR323 system with samples fixtured to a rotating drum. The substrates were polished cemented carbide (WC-Co, Seco Tools “HX”) flat inserts of dimensions 13 × 13 × 4 mm. Prior to deposition the substrates were sputter cleaned with Ar ions. During the deposition the atmosphere was pure (99.995%) N2, the substrate temperature was ~ 400 °C and a bias of −55 V was applied. The resulting films were of a B1 NaCl structure with a lattice parameter of 4.15 Å as determined by X-ray diffraction (XRD), with a total thickness of ~3 μm. Rutherford backscattering spectroscopy (RBS) gave the mean composition of the films as (Ti0.33Al0.67)N0.92, with O contamination on the level of 0.5–1 at.%. The microstructure was found to be of a dense columnar type [5]. Isothermal annealing of a sample was performed for 120 min at 900 °C in an inert argon atmosphere with a Sintervac furnace (GCA Vacuum Industries). The heat treatment parameters were chosen
L.J.S. Johnson et al. / Thin Solid Films 520 (2012) 4362–4368
3. Data analysis For the application of APT to the problem of spinodal decomposition Miller et al. [16–18] is the most recent general treatment. The main variable of interest for spinodal decomposition is the mole fraction describing the separation: x ¼ nAl =ðnAl þ nTi Þ;
ð1Þ
where n is the concentration of a species. For transition metal nitrides the stoichiometry of N may have a large effect on the properties of the alloy [19–21]; we take it as: z ¼ nN =ðnN þ nAl þ nTi Þ ≈ nN ;
ð2Þ
effectively giving (Ti1 − xAlx)1 − zNz. The two variables were computed on a regular grid of 1 nm 3 cubes by first binning all detected ions using Gaussian delocalization with the IVAS software, and then computing the field variables for each voxel. The autocorrelation functions of the field variables were computed to investigate any spatial correlations present in the data. The autocorrelation function, γ(r), for variable x was computed as: →
γð r Þ ¼
1 r ′−→r −xÞ; ∑ðx→ r ′−x Þðx→ σ 2x ðN−1Þ →
ð3Þ
r′
where x is the mean, σx the standard deviation, N the number of vox0 els, and → r ranges over all valid voxel indices in the datacube. The function was then averaged radially by integration using cylindrical coordinates: 2π
γ ðr; zÞ ¼
1 ∫ γðr cosðϕÞ;r sinðϕÞ;zÞdϕ: 2π 0
ð4Þ
Radial distribution functions (RDF) are another way to show spatial correlations. In the case that a strong spatial variation in one variable influences the one for which a RDF is sought, the RDF may be extended by taking the variation of the first variable into account and binning the RDF: let g ia(r) be the RDF of variable a for atom i and g ia its average up to the radius r0: i
i
g a ðr Þ ¼ g a ðr Þr ; r∈½0; r 0 :
ð5Þ
Then the binned RDF of variable b is: i
i
g b ðr; aÞ ¼ g b ðr Þi ; ∀i : g a ∈½a−Δa; a þ Δa:
the instrument used in this study). Danoix et al. [22] analyzed this situation in detail for the case of a two-component system. They showed that the variance of the measured atomic composition, c, of a box is: 2
σ c ¼ p0 ð1−p0 Þð1−ηÞ=n;
ð7Þ
where p0 is the true composition, η is the detection efficiency, and n is the mean number of ions per box (atomic density per box, m, times the efficiency), for reasonable detection efficiencies and box sizes (m ≥ 100). In order to test the hypothesis that a sample is homogeneous, an empirical test distribution was created by simulating the model described above. A sample was considered to consist of 40 000 sample cubes (1 nm 3 in volume) of a B1 structure of appropriate density, the variance of the composition variable was calculated for each sample. Two thousand (2000) simulations were performed, and the distribution of variances was calculated, giving a p-value resolution of 0.0005. This allowed the testing of the measured variance of the composition of samples against the null hypothesis that the observed variance could be explained by the variance inherent in the measurement, by directly calculating the probability that the observed variance was in accordance with the null hypothesis. 4. Results 4.1. Mass spectra and evaporation behavior A typical mass spectrum from a Ti0.33Al0.67N film is shown in Fig. 1, and the corresponding charge state ratios are given in Table 1. Ions of Ti (doubly and triply charged), Al (singly to triply charged), and N (singly and doubly charged) are present. Of note is the large amount of complex ions, where TiN 2 + is the dominating complex, but AlN 2 + and AlO 2 + are present as well. No peaks corresponding to either O or TiO complexes were observed. Tied to the high number of complexes were a corresponding amount of multiple evaporation events per laser pulse. Finally, although the Al 2 + and N + peaks are well separated, the tail of the Al 2 + peak overlaps with the N + peak to a significant degree, most probably due to the limited thermal conductance of the film. This slows down the cooling of the APT tip after heating by each laser pulse, which increases the time window for evaporation, which in turn leads to the broadening of the peak tails. The composition of the samples, obtained by peak decomposition using natural isotope abundances and statistical averaging, is shown in Table 2. This method uses the whole spectrum, and as such it is not applicable to the ranging of ions for the volume reconstruction, which will be worse at resolving the overlap of the Al 2 + and N + peaks (compare with Fig. 3). The metal pseudobinary ratio, x, in Ti1 − xAlxN, is close to 0.6 for both samples. This is below the expected 0.67 value as confirmed by RBS measurements. This
106 3+
Al
105
counts
according to the results of Hörling et al. [5] to produce as much c-AlN as possible before transformation to the stable h-AlN occurs. Samples for APT were prepared using the FIB lift-out technique [15] where wedges roughly 3 μm high and 1–2 μm wide are attached to posts fabricated out of a silicon wafer. The wedges were sharpened by annular milling at 30 keV and – for the last stage – 5 keV Ga + ions. The samples were analyzed in a local electrode atom probe (Imago LEAP 3000X HR) in laser pulsing mode with a pulse rate of 200 kHz, a pulse energy of 0.35 nJ, and a target evaporation rate of 0.5% (expressed as detected ions per pulse) at 65 K. The laser energy was chosen after trials to provide the optimum mass resolution. Four tips from the as-deposited state and six from the annealed state were analyzed. For reconstruction, an evaporation field of 40 V/nm, image compression factor of 1.65 and a field factor of 3.3 were used, which gave final tip radii close to or in agreement with measurements by scanning electron microscopy on the tips after evaporation.
4363
10
Al2+ N+
The field variables of interest are sensitive to the sampling nature of the technique, as the collection efficiency is less than one (0.37 for
TiN2+
4
AlN2+
N2+
1000 100 10
ð6Þ
N2+ Ti2+ Al+
Ti3+
AlO2+ 5
10
15
20
25
30
m q a.u. Fig. 1. Laser atom probe mass spectrum from Ti0.33Al0.67N.
35
L.J.S. Johnson et al. / Thin Solid Films 520 (2012) 4362–4368
a
Table 1 Charge state ratios corresponding to Fig. 1. X+
X3 +
XN2 +
X+ 2
Ti2 + Al2 + N2 +
– 5.7 0.01
2.2 4.9 –
2.6 78.4 Al: 0.30 Ti: 0.03
– – 0.01
60 0.8 50 0.6
40
y nm
X2 +
1.0
deviation is due to the overlap of Al and N peaks in the APT data, which reduces the detected amount of Al. The N and impurity O concentrations indicate close to, but slightly less than, stoichiometric films, in agreement with the RBS measurements.
30
0.4
20
Al Al Ti fraction
4364
0.2 10 0
4.2. Decomposition
0
10
20
30
40
50
0.0
60
x nm
Table 2 Composition determined from the spectra, and corresponding ±2σ intervals, all values are given in atomic percent. Sample
Ti
Al
N
O
As-deposited Annealed
20.3 ± 0.8 20.9 ± 2.0
29.1 ± 1.4 29.8 ± 2.6
49.4 ± 1.8 48.9 ± 1.7
1.2 ± 0.5 0.4 ± 0.4
1.0
b 60
0.6
y nm
40 30
0.4
20
Al Al Ti fraction
0.8 50
0.2 10 0
0
10
20
30
40
50
0.0
60
x nm Fig. 2. 2D concentration maps of the Al fraction of the metal content of Ti0.33Al0.67N on slices through reconstructed samples in the xy-plane, for the (a) as-deposited and (b) annealed states.
pure TiN or AlN is also small. Almost all of the investigated specimen is in an intermediate stage of decomposition. In order to address the question of whether the observed fluctuations in the as-deposited film were true fluctuations in the film or measurement artifacts, the estimated variance of the Al fraction was tested against the hypothesis of a homogeneous film, according to Section 3. The result is given in Table 3: the estimated standard deviation was 0.12, and the theoretical standard deviation was calculated to 0.08. The variance is significantly larger than the expected
as deposited annealed
0.15
Relative amount
The variation of the composition on the metal sublattice is shown in Fig. 2 for the as-deposited and annealed samples. The two reconstructed tips are representative for the two heat treatment states, unless specifically noted in the text, and are used throughout the rest of the figures for direct comparison. The figures show concentration maps taken from slices through the reconstructed atom probe data, and are typical for each tip. The as-deposited film (Fig. 2(a)) exhibits small fluctuations around the mean composition, with no obvious wavelength or ordered structure, throughout the whole tip. The annealed sample (Fig. 2(b)), on the other hand, shows a clear segregation of Al and Ti into domains predominantly composed of TiN or AlN. The composition has a non-zero gradient over most of the slice, except at the extremum points of fully separated TiN and AlN. The Al- and Ti-enriched domains exhibit varying morphologies, which mostly deviate significantly from the spherical case. Furthermore, the Al-rich domains show internal compositional fluctuations, whereas the Ti-rich domains are homogeneous in the sense that they are compact and without saddle points in the composition. Beside the regions of segregation there is a large region on the right side of the slice in Fig. 2(b) which extends at an angle through the tip, that appears similar in nature to the as-deposited sample, and which was only observed in this particular tip. This indicates that after the heat treatment there exist regions in the sample that have not undergone phase decomposition. Histograms of the volume composition of the metal fraction for the two runs are shown in Fig. 3. The histograms presented here are of the smoothed data used to generate the compositional maps, and as such do not show much of the instrumental variation due to the sampling. Instead, the compositional variation due to the respective up-hill diffusion of Ti and Al is dominating. The as-deposited sample shows a simple normal-like distribution around a compositional mean of x = 0.57, which deviates from the known sample composition, see Section 5.1. The annealed sample again shows its decomposed structure together with the remaining TiAlN solid solution regions, and the histogram may be understood as the sum of three distinct components. The first two correspond to the formation of Al- and Ti-enriched domains (1 and 2 in Fig. 3), while the third and middle peak (marked 3 in Fig. 3) correspond to the undecomposed regions, which are similar to what is observed in the as-deposited sample. As the heights at the ends of the histogram, corresponding to x = 0 and x = 1, are vanishingly small, the volume fraction of
0.10
0.05 3 1
2 0.00 0.0
0.2
0.4
0.6
0.8
1.0
Al Al Ti fraction Fig. 3. Histograms over volume fractions for the metal sublattice concentration of Al in Ti0.33Al0.67N in as-deposited and annealed states.
L.J.S. Johnson et al. / Thin Solid Films 520 (2012) 4362–4368
Ad
x z Annealed z
0.57 0.49 0.51
0.12 0.09 0.12
Homogeneous std. dev.
p-value
0.08 0.06 0.06
b0.0005 b0.0005 b0.0005
60
0.65
50
0.60 0.55
40
0.50
30
0.45 0.40 10
0.35
0 0
10
20
30
40
50
0.30
60
x nm
b
y nm
fluctuations in a reconstructed homogeneous film with a p-value of less than 0.0005. Accordingly, we ascribe the observed fluctuations, to a significant part, to a true fluctuation of the composition in the sample. Fig. 4 shows the autocorrelation functions of the metal pseudobinary composition for the as-deposited and annealed tips. The autocorrelation was calculated as shown in Section 3, and radially averaged in the xy-plane of the reconstruction to avoid smearing caused by slightly different length scales in the reconstruction. The autocorrelation of the as-deposited film shows only a gradual decrease as the lag increases, which indicates that the compositional variations seen above in Fig. 2(a) have no dominating length-scale. The autocorrelation for the annealed sample instead exhibits a variation with a period of ~ 25 nm, which is present up to 60 nm in lag. This period can be seen in the sample, even though the undecomposed regions act as dampeners to the signal.
20
0.70 60
0.65
50
0.60 0.55
40
0.50
30
0.45 20 0.40
4.3. Stoichiometry
10 Fig. 5 shows the variation in nitrogen content in the same slices as in Fig. 2. Both samples exhibit fluctuations in the nitrogen content, and hence the stoichiometry, as the total concentration is normalized to unity for each reconstructed cube. There is no apparent difference in the structures of the fluctuations between the two samples. From the corresponding histograms in Fig. 6 it is apparent that the mean values differ slightly, with the as-deposited tip exhibiting a lower mean of z = 0.49 versus z = 0.51 for the annealed tip. The width and shape of the distributions are, however, almost identical. The estimated standard deviations and the corresponding theoretical deviations are again given in Table 3, together with the obtained p-values for the observed deviations. The observed standard deviations, for both films, are significantly larger than expected (p-values less than 0.0005) under the assumption of homogeneity (see Section 3). The autocorrelation functions of the stoichiometry variable, z, are shown in Fig. 7 for the two runs. The autocorrelation function obtained from the as-deposited film shows a slow decrease with an extended tail, with no clear oscillations present, similar to the autocorrelation function for the Al–Ti variation in Fig. 4. The autocorrelation function
0.06
0
0.03 0.02 0.01
0 10 20 30 40 50 60 70
10
20
30
40
50
0.30
60
x nm Fig. 5. 2D-concentration maps of the N content in Ti0.33Al0.67N of slices through reconstructed samples in the xy-plane, for the (a) as-deposited and (b) heat treated states.
of the annealed film, on the other hand, exhibits a steep fall – of the order of one sampling cube (1 nm 3) – coupled with a small and rapid oscillation. Due to the abrupt nature of the fall it is probable that the oscillations are noise and not true features of the underlying data. That notwithstanding, the sharp decrease itself indicates a marked difference to the as-deposited film. Fig. 8 shows the radial distribution functions (RDF) of the N content calculated for Ti and Al sites for both the as-deposited and the annealed samples. The RDF of the as-deposited sample shows no difference between Ti and Al, except at zero offset, where N is more often found together with a Ti atom than with an Al. This is most
as deposited annealed
0.10
Relative amount
0.002 0.001 0.000 0.001 0.002 0.003
0.04
0.35
0
as deposited annealed
0.05
N concentration at.
Variable Mean Standard deviation
0.70
N concentration at.
Sample
a
y nm
Table 3 Statistical analysis of variations in metal fraction and stoichiometry for both the asdeposited and the annealed samples.
4365
0.08 0.06 0.04 0.02
0.00 0
10
20
30
40
50
60
70
r nm
0.00 0.3
0.4
0.5
0.6
0.7
N content (at. fraction) Fig. 4. Autocorrelation functions of the metal sublattice occupancy of Al in Ti1 − 0.33Al0.67N, for the as-deposited and annealed states. The inset is a magnified view of the annealed state data.
Fig. 6. Histograms over volume fractions for N concentration in Ti0.33Al0.67N in asdeposited and annealed states.
4366
L.J.S. Johnson et al. / Thin Solid Films 520 (2012) 4362–4368
1.0
as deposited annealed
N Composition (at. %)
0.8
60
0.6 0.4 0.2 0.0
50 40
Ti
20
Al
10
N
2
1 0
1
2
0 0
10
20
30
40
50
60
-4
70
r nm
likely due to the preferred evaporation of Ti as TiN–ion complexes (c.f. Fig. 1) and hence an artifact of the measurement technique. The first maximum of the Al RDF lies at slightly less than 2 Å, close to the nearest neighbor distance (200) in TiN. For the annealed film the situation is more complex; the RDF is shown binned according to the Al–Ti fraction (x) around each metal atom (see Eqs. (6–7), r0 = 0.5 nm), which gives the 3D plot shown in Fig. 8b, with metal composition on the x-axis, and the RDFs on the y–z axes. From this plot it is apparent that the metal composition determines the local N-concentration, with a nonlinear variation with the Al–Ti fraction. At the Ti-rich side (x = 0) the stoichiometry (z) is ~0.49, which decreases as x increases up to the minimum at x = 0.7 of ~ 0.47, followed by an increase again up to x = 1.0. Due to the overlap of the Al 2 + and
a 49.2 49.0 48.8 48.6 48.4 48.2
Al Ti 0
1
2
3
4
distance (nm) 4
b
3
distance (nm) 2
0 50 49 48 1.0
N concentration (at. %)
1
0.5 0.0
-2
0
2
4
Distance (nm)
Fig. 7. Autocorrelation functions of the N content in Ti0.33Al0.67N, for the as-deposited and annealed states.
N concentration (at. %)
54 53 52 51 50 49
30
Al/(Al+Ti) fraction
Fig. 8. Radial distribution functions of N around Ti and Al sites in Ti0.33Al0.67N, for the (a) as-deposited and (b) annealed states, which are binned according to the Al fraction at the site of the originating atom.
Fig. 9. Proxigram from the 3:2 Al isoconcentration surface of Ti0.33Al0.67N annealed at 900 °C, showing the mean gradients of the decomposition of Al, Ti and N. The inset shows a magnified part of the N signal.
N + peaks, the N concentration is overestimated at large Al contents. Extracting the regions with x > 0.9 and analyzing the mass spectrum gives a N concentration of 49 at.%, comparable to that at the Ti side. At lags greater than zero (distances > 0), the N concentration converges to the global level of around 48 at.% for all Al–Ti compositions, with the Al side converging slower than the Ti side, as the Al rich domains are larger in size. The two lines drawn down the graph marks the conventional RDFs that are obtained if the composition variations are not taken into account, which are clearly misleading in this case. A proxigram from the annealed sample corresponding to the isoconcentration surface of x = 0.6 is shown in Fig. 9. Due to the variation in segregation shown above, the proxigram was constructed from a part of the sample showing pronounced concentration gradients, corresponding to the lower left part of Fig. 2 (b). The Al and Ti curves show the gradient on the metal sublattice. It is asymmetric with longer tails into the Ti-rich domains than into the Al-rich domains, and it is approximately 5 nm wide. Furthermore, the Ti content in the AlN-rich domains is higher (~6 at.%) than the Al content in the TiN-rich domains (~2–3 at.%). The increase in the variance at increasing distances from the isosurface is due to the decreasing volumes of high segregation and their geometrically closed shape. The variation of the stoichiometry is most clearly seen in the N profile, where a variation is seen when going along the gradient of the metal composition, similar to that seen in the RDF curves (Fig. 8b). There is a fluctuation in the region with a large gradient in the Al and Ti contents, consisting of a sinusoidal variation with the maximum on the Ti-rich side, the minimum on the other side and the inflection point almost perfectly centered on the isoconcentration surface. Although small, of the order of 1–2 at.%, the variation in N content with metal fraction is significant with regard to errors of the measurement as the total number of binned ions is highest near the isosurface, and the corresponding error of estimation is consequently small. 5. Discussion 5.1. Evaporation behavior The field evaporation of Ti–Al–N reported here, with a relatively low laser power and, accordingly, a relatively high applied DC field, resulted in a spectrum with a large amount of ion complexes. Particularly, TiN 2 +, N2+, and AlN 2 + ions were detected. A similar behavior was reported by Ai et al. [23], who used a laser assisted 1D atom probe field ion microscope to investigate TiN thin films. Although the laser power was not reported, the applied DC field was ~30 V/ nm, in comparison with the ~40 V/nm used in this study. Ai observed large amounts of N2+ and TiN 2 +, as well as the elemental ions Ti 2 +,
L.J.S. Johnson et al. / Thin Solid Films 520 (2012) 4362–4368
Ti 3 +, and N +. Additionally, Ti2N 3 + ions were reported, which were not observed in this study. The films were not phase-pure B1 TiN, but contained a minority phase of Ti2N, which could explain the Ti2N 3 + ions, although the lower field should also promote more ion complexes. The presence of TiN complexes in the spectra might be understood with regards to the binding of Ti and N, where N almost solely binds to its nearest neighbor Ti atoms, while Ti atoms also bond with the second nearest Ti neighbors [24], and this also helps explain the large amount of multiple evaporation events that were observed here. The peaks in the spectrum in Fig. 1 are mostly well separated, with the exception of the Al 2 + and N + peaks. While clearly resolved, the tail of the Al 2 + peak overlaps the N + peak significantly. This will cause some error in the classification (“ranging”) of ions as some Al 2 + ions will unavoidably be classified as N + ions, and, to a lesser degree, N + will be wrongly classified as Ti 3 +. This leads to the overestimation of N and Ti by ~ 2–3 at.% in the reconstruction. This is a well-known issue with laser APT, and is being addressed by finer control of the zone heated by the laser in newer instruments. The tail is also significantly influenced by the thermal conductance of the sample, and as such will vary with microstructure, crystal quality, etc. [25]. 5.2. The as-deposited state The presence of compositional modulations on the metal lattice in the as-deposited sample is not surprising. Generally, TiAlN films deposited with compositions in the miscibility gap show the same trend, with increasing modulations for increasing Al content and deposition temperatures [26,27]. Films deposited by cathodic arc evaporation exhibit a single phase NaCl structure for Al fractions as high as x = 0.7. This is due to the deposition conditions with a high flux of energetic ions [27], which produces recoil mixing in the growing film [28]. Such mixing counteracts surface segregation, which is bound to take place even at the modest temperatures of deposition (~450 °C). So the small variations seen in the sample should be expected from steady-state conditions. Furthermore, the fact that no clear wavelength of decomposition was observed may be understood by arguments from the theory of spinodal decomposition. The wavelength of the composition modulation of a spinodally decomposed material is an emergent property; the wavelength with the largest driving force will outgrow all other wavelengths [29]. As no wavelength dominates in the as-deposited sample, the conclusion must be that the segregation observed here is in its very earliest stage. The statistical analysis of the stoichiometry fluctuations in the asdeposited sample (see Table 3) indicates that they are not only due to the experimental sampling, yet no dominating frequency was observed in the autocorrelation function of the N content. The RDFs of N from Ti and Al sites furthermore do not show any correlation, except for the Ti–N correlation at 0 nm offset due to the preferred evaporation of TiN complexes. Due to the high energies of the incident ions during the deposition process (~100 eV [28]), a high number of vacancies are likely present in the film [30–32]. The stoichiometry measured by the atom probe is that of the detected ions, and as such is an average of both nitrogen and metal atom vacancies. Therefore, the observed fluctuations originate from both sublattices, although the N vacancies should dominate by simple thermodynamical considerations due to their lower energy of formation. Since the variation in stoichiometry cannot be coupled to variations of the Al fraction on the metal lattice, they must be ascribed to random variations due to the deposition process. 5.3. The annealed state: decomposition The annealed sample has a clearly decomposed structure, along with parts that are essentially the same as that of the as-deposited sample.
4367
These undecomposed regions, as previously detected by XRD measurements [2,5], and then connected to an undecomposed matrix containing particles of TiN and AlN, are here found to be separate from the decomposed regions. The gradient of the metal composition is nonzero for the majority of the volume, except at the fully decomposed points, which are essentially negligible when compared to the rest of the sample (as can be seen from the histogram in Fig. 3). Moreover, the autocorrelation function in Fig. 4 shows that the decomposition has evolved a periodic structure in the sample. It is also instructive to consider that the domains are not of constant composition; there are no growth fronts where precipitates consume the matrix phase or other structures tied to a nucleation and growth mechanism. Calculations on the TiN–AlN system show the existence of a miscibility gap [4,9] and this has been observed in numerous experiments [2,5–8,26,27]. There are indications in the literature of spinodal decomposition in the TiAlN system, for example [2,6–8,26,33–35], with structures similar to those observed here. By definition, the only process that could give rise to this kind of configuration is spinodal decomposition, which is consistent with the fact that the Ti0.33Al0.67N alloy composition lies within the coherent spinodal [4]. The existence of a region of relatively homogeneous composition in the annealed sample (c.f. Fig. 2(b)) indicates that the decomposition can be locally inhibited, and thus has a microstructural origin. The mean column width of these samples is 200–300 nm [5], and while no grain boundaries were observed in this sample, the reconstructed diameter of ~ 70 nm makes it probable that at least one side would be close to a grain boundary. For sputtered Ti0.5Al0.5N Rachbauer [13] observed that the decomposition was almost entirely limited to the grain boundaries at similar heat treatments as used here. Therefore, the proximity of a grain boundary to the side of the probed volume appears to be the most likely explanation of the observed variation in decomposition activity in our annealed sample. The decomposed structure visible in Fig. 2 is not isotropic with radially-symmetric domains. Instead, the domains have a have a mean elongation along a diagonal in the figure. The theory of spinodal decomposition predicts that the degree of anisotropy in the decomposition is determined by the anisotropy of the elastic constants [36]. Both experiments and calculations of the elastic anisotropy of TiN and TiAlN show that there is a considerable anisotropy in the system [37–39], so anisotropy in the compositional variations is expected, although the lack of crystallographic information prevents further comparison with theory. 5.4. Annealed state: stoichiometry variations While both samples show fluctuations in the stoichiometry in Fig. 5 with similar variances (Table 3 and Fig. 6), the annealed sample exhibits a correlation between the N distribution and the composition of the metal lattice (c.f. Figs. 8 and 9). The N signal is more or less equal at the endpoints x = 0 and x = 1, while there is a clear drop around x = 0.7. The observed variations in stoichiometry may stem from two sources in a TiN-like material. Firstly, there will be vacancies on the N sublattice, and secondly – but to a lesser degree, due to the higher energy cost – vacancies on the metal sublattice. Both of these vacancy types will be registered as a variation in stoichiometry in the APT data. B1-TiNz is thermodynamically stable over a wide range of stoichiometries [40]. Wurtzite AlN, on the other hand, is a line compound with a very narrow composition range [41]. Alling et al. [42] calculated the formation energies of (Ti1 − xAlx)Nz for z b 1, and found that the formation energy surface is convex (has negative curvature) for values of z close to 1. These conditions offer an explanation for the observed behavior of the stoichiometry. For a twophase system, the expected outcome is stoichiometric AlN and understoichiometric TiN, which is the most energetically favorable state. This would be the case for a fully completed decomposition, while here the system is dominated by the regions with a composition
4368
L.J.S. Johnson et al. / Thin Solid Films 520 (2012) 4362–4368
gradient (c.f. Fig. 3). Modeling this by allowing a three component system to relax over the energy landscape in Ref. [42] directly leads to stoichiometric AlN, TiN, and an understoichiometric (Ti1 − xAlx)Nz, which, by the nature of the convex energy surface, lowers the total energy of the system. This finding is important, as it leads to a reduction of the strain energy of the spinodal system since it reduces the lattice mismatch in the composition gradient, and thus facilitates the decomposition. This is in agreement with the findings by Alling et al. [42]; that understoichiometric Ti1 − xAlxN should exhibit an enhanced tendency for coherent decomposition. A similar sinusoidal feature as the one obtained here (see Fig. 9) may be observed in Fig. 5 (b) of Rachbauer's work [43]. In that figure the variation is composed of one major variation, where the TiN-rich part is overstoichiometric in N and the AlN-rich part is stoichiometric (which we attribute to the line-compound nature of AlN), and a sinusoidal fluctuation with the maximum on the Ti-rich side of the gradient and the minimum on the Al-rich side. Interestingly, the modulation that we read out of Rachbauer's study is present even though their film is globally overstoichiometric with respect to N, and as such the decomposed TiN phase would become even more overstoichiometric from absorbing the N ejected from the AlN. This is a complementary case to the films studied in this work. As previously mentioned, the variations in stoichiometry may be due to vacancies on both lattices, so we should consider explanations for both cases. Attributing the modulation in Fig. 5 (b) in Ref. [43] solely to N vacancies is problematic as there is no obvious reason for the segregation of N vacancies on the level of ~ 1 at.% on the Al-rich side of the decomposition gradient. Another possibility is to ascribe the observed modulation to variations in the metal vacancy concentration. Since metal vacancies are directly involved in the spinodal decomposition (by virtue of the substitutional diffusion), dynamical effects need to be considered. By making the reasonable assumption that the Al mobility is higher than the Ti mobility (as the atomic radius of Al is smaller than that of Ti, and that the melting or dissociation temperature of AlN is lower than that of TiN, which corresponds to that the metal–N bond is weaker for Al–N than for Ti–N [38]), a Kirkendall effect is introduced. That implies a net flow of metal vacancies from the Al-rich side of the composition gradient to the Ti-rich side, leaving the Al side understoichiometric and the Ti side overstoichiometric in N. This would explain the observed variation, but would also require a significant mean concentration of vacancies to be able to support the observed maximum overstoichiometry. The main source of vacancies here is the cathodic arc deposition process, which yields a high defect density in the form of dislocations and lattice point defects [28,44]. For comparison, HfN vacancy concentrations as high as 10–15% have been observed during PVD processing [30]. 6. Conclusions Laser-assisted APT of Al-rich c-TiAlN thin films deposited by cathodic arc evaporation has been carried out for both as-deposited and annealed samples. The sample evaporation behavior is similar to that reported for TiN. Spinodal decomposition towards TiN and AlN is detected readily in the alloy sample annealed at 900 °C for 2 h. The as-deposited sample exhibits variations in composition on the metal sublattice of the order of 1–2%, because of the limited thermal activation during film deposition. A stoichiometry variation tied to the phase decomposition is detected in the annealed samples. Here, N segregates preferentially to AlN and TiN, making them closer to stoichiometric, leaving the interfacial TiAlN matrix understoichiometric, while no loss of N is detected. This effect mitigates the coherency strains between the isostructural AlN and TiN domains. This modulation in the stoichiometry along the decomposition gradient may in part be due to a Kirkendall effect, from unbalanced Al and Ti diffusivities in TiAlN.
Acknowledgments The Swedish Research Council and the VINN Excellence Center on Functional Nanoscale Materials are acknowledged for financial support. Björn Alling is acknowledged for fruitful discussions. In memoriam Bo Jansson.
References [1] A. Hörling, L. Hultman, M. Odén, J. Sjölén, L. Karlsson, Surf. Coat. Technol. 191 (2005) 384. [2] P.H. Mayrhofer, A. Hörling, L. Karlsson, J. Sjölén, T. Larsson, C. Mitterer, L. Hultman, Appl. Phys. Lett. 83 (2003) 2049. [3] Q. Xia, H. Xia, A.L. Ruoff, J. Appl. Phys. 73 (1993) 8198. [4] B. Alling, A.V. Ruban, A. Karimi, O. Peil, S.I. Simak, L. Hultman, I.A. Abrikosov, Phys. Rev. B 75 (2007) 45123. [5] A. Hörling, L. Hultman, M. Odén, J. Sjölén, L. Karlsson, J. Vac. Sci. Technol., A 20 (2002) 1815. [6] A. Knutsson, M.P. Johansson, P.O.A. Persson, L. Hultman, M. Odén, Appl. Phys. Lett. 93 (2008) 143110. [7] M. Odén, L. Rogström, A. Knutsson, M.R. Terner, P. Hedström, J. Almer, J. Ilavsky, Appl. Phys. Lett. 94 (2009) 053114. [8] R. Rachbauer, E. Stergar, S. Massl, M. Moser, P.H. Mayrhofer, Scr. Mater. 61 (2009) 725. [9] P.H. Mayrhofer, L. Hultman, J.M. Schneider, P. Staron, H. Clemens, Int. J. Mater. Res. 98 (2007) 1054. [10] T.F. Kelly, M.K. Miller, Rev. Sci. Instrum. 78 (2007) 31101. [11] T.F. Kelly, T.T. Gribb, J.D. Olson, R.L. Martens, J.D. Shepard, S.A. Wiener, T.C. Kunicki, R.M. Ulfig, D.R. Lenz, E.M. Strennen, E. Oltman, J.H. Bunton, D.R. Strait, Microsc. Microanal. 10 (2004) 373. [12] D.J. Larson, D.T. Foord, A.K. Petford-Long, T.C. Anthony, Ultramicroscopy 75 (1998) 147. [13] R. Rachbauer, S. Massl, E. Stergar, D. Holec, D. Kiener, J. Keckes, J. Patscheider, M. Stiefel, H. Leitner, P.H. Mayrhofer, J. Appl. Phys. 110 (2011) 023515. [14] P.H. Mayrhofer, D. Music, J.M. Schneider, J. Appl. Phys. 100 (2006) 094906. [15] K. Thompson, D. Lawrence, D.J. Larson, J.D. Olson, T.F. Kelly, B. Gorman, Ultramicroscopy 107 (2007) 131. [16] M.K. Miller, J.M. Hyde, M.G. Hetherington, A. Cerezo, G.D.W. Smith, C.M. Elliott, Acta Metall. Mater. 43 (1995) 3385. [17] J.M. Hyde, M.K. Miller, M.G. Hetherington, A. Cerezo, G.D.W. Smith, C.M. Elliott, Acta Metall. Mater. 43 (1995) 3403. [18] J.M. Hyde, M.K. Miller, M.G. Hetherington, A. Cerezo, G.D.W. Smith, C.M. Elliott, Acta Metall. Mater. 43 (1995) 3415. [19] L.E. Toth, Transition Metal Carbides and Nitrides, Academic Press, New York, 1971. [20] C. Shin, D. Gall, N. Hellgren, J. Patscheider, I. Petrov, J.E. Greene, J. Appl. Phys. 93 (2003) 6025. [21] L. Rogström, L.J.S. Johnson, M.P. Johansson, M. Ahlgren, L. Hultman, M. Odén, Scr. Mater. 62 (2010) 739. [22] F. Danoix, G. Grancher, A. Bostel, D. Blavette, Ultramicroscopy 107 (2007) 734. [23] C.F. Ai, J.Y. Wu, C.S. Lee, Vacuum 44 (1993) 99. [24] C. Ciobanu, D. Tambe, V. Shenoy, Surf. Sci. 582 (2005) 145. [25] J.H. Bunton, J.D. Olson, D.R. Lenz, T.F. Kelly, Microsc. Microanal. 13 (2007) 418. [26] F. Adibi, I. Petrov, L. Hultman, U. Wahlström, T. Shimizu, D. Mcintyre, J.E. Greene, J.E. Sundgren, J. Appl. Phys. 69 (1991) 6437. [27] U. Wahlström, L. Hultman, J.E. Sundgren, F. Adibi, I. Petrov, J.E. Greene, Thin Solid Films 235 (1993) 62. [28] A. Anders, Vacuum 67 (2002) 637. [29] J.W. Cahn, Acta Metall. 9 (1961) 795. [30] A.J. Perry, J. Vac. Sci. Technol., A 6 (1988) 2140. [31] G. Smith, P. Swift, A. Bendavid, Appl. Phys. Lett. 75 (1999) 630. [32] L. Tsetseris, N. Kalfagiannis, S. Logothetidis, S. Pantelides, Phys. Rev. B 76 (2007) 224107. [33] S. Lee, B. Kim, H. Kim, J. Lee, J. Appl. Phys. 80 (1996) 1469. [34] A. Kimura, H. Hasegawa, K. Yamada, T. Suzuki, Surf. Coat. Technol. 120–121 (1999) 438. [35] J.Y. Rauch, C. Rousselot, N. Martin, Surf. Coat. Technol. 157 (2002) 138. [36] J.W. Cahn, Acta Metall. 14 (1966) 1685. [37] J. Kim, J. Achenbach, P.B. Mirkarimi, M. Shinn, S.A. Barnett, J. Appl. Phys. 72 (1992) 1805. [38] F. Tasnádi, I.A. Abrikosov, L. Rogström, J. Almer, M.P. Johansson, M. Odén, Appl. Phys. Lett. 97 (2011) 231902. [39] Y. Yang, H. Lu, C. Yu, J. Chen, J. Alloys Compd. 485 (2009) 542. [40] H. Wriedt, J. Murray, J. Phase Equilib. 8 (1987) 378. [41] H. Wriedt, J. Phase Equilib. 7 (1986) 329. [42] B. Alling, A. Karimi, L. Hultman, I.A. Abrikosov, Appl. Phys. Lett. 92 (2008) 71903. [43] R. Rachbauer, S. Massl, E. Stergar, P. Felfer, P.H. Mayrhofer, Surf. Coat. Technol. 204 (2010) 1811. [44] H. Ljungcrantz, L. Hultman, J.E. Sundgren, L. Karlsson, J. Appl. Phys. 78 (1995) 832.