2. Materials and Methods
In the present work, BaTiO
thin film evolution was studied at P03 at PETRA III [
11], two 2D detectors were used simultaneously. One 2D detector (Pilatus 1M, 981 × 1043 pixels, pixel size of 172
m × 172
m) was used for GISAXS at a Sample-to-Detector Distance of SDD = 4 m. The X-ray beam had an energy of 13 keV and a beam size of 40
m (horizontal) × 20
m (vertical) FWHM, and the angle of incidence (
) was set to 0.5
. This angle is above the critical angle of the materials involved. A second single-photon counting pixel detector with 487 × 619 pixel of 172
m each (Pilatus 300k, Dectris, Baden, Switzerland) was mounted on the detector arm of the diffractometer with SDD = 0.28 m covering the Q-range from 2 Å
−1–4.5 Å
−1. With this second 2D detector, we were following the evolution by GIXRPD simultaneously to the GISAXS measurement. Both detectors were in air. The measurements were performed with the following parameters. The plasma process and substrate heating was started 10 min. before opening the shutter in front of the sputtering target. The sputtering power on the BaTiO
target was 14 W/cm
2 and the gas mixture was 80% Argon (Ar) and 20% Oxygen (O
) with a total flow of 1 standard cubic centimetre per minute (SCCM), and a substrate temperature of 600
C. The distance between the sputtering target and substrate was fixed at a distance of 4 cm. The films were deposited on pre-cleaned SiO
wafers with a size of 10 × 10 mm
. A more detailed description of the measurement can be found here [
8,
10]. SiO
wafers are not the industrial standard for BaTiO
thin film deposition, but we have chosen it to avoid surface related influence on the crystallization process. A more detailed study of the influence of various parameters, such as surface structure and deposition temperature, on the crystallization behavior will be published elsewhere.
Figure 1 illustrates the principal scheme of the experiment. The X-rays arrived from the left hand side and hit the wafer inside the sputtering unit under gracing incidence and get detected with the 2D detector behind the chamber.
3. Results and Discussion
This in situ experiment recorded the film evolution continuously in reciprocal space, without interrupting the sputtering process. Exemplary patterns for the GISAXS detector can be seen in
Figure 1. To analyze these patterns in more detail, horizontal cuts were taken at the level of the Yoneda peak [
12]. The Yoneda peak occurs when the exit angle of the scattered beam is close to the critical angle due to signal enhancement from the Vineyard effect [
13]. This results in a bright band of intensity at the critical angle (
). Below the critical angle the scattering intensity decreases quadratically to zero. With the refraction index
including the real part
and the imaginary part
, it depends strongly on the wavelength
and the electron density. Therefore, it is a material dependent effect when the wavelength is constant.
These patterns (seen in
Figure 1 and
Figure 2a) show a clear evolution perpendicular to the Yoneda peak over the deposition time. This relates to density correlations and arise from a the structure factor with characteristic modulations parallel to the surface. The peaks were fitted with a pseudo-Voigt function (see
Figure 2)
, with G(x) the sum of a Gaussian peak and
L(x) the Lorentzian peak, as well as the mixing parameter
which describes the fraction of each function.
It shows that the particle distance initially appears at higher
q and shifts towards smaller
q over deposition time. This shift indicates a coalescence, i.e., mainly a fusion phenomenon of adjacent nuclei or clusters, which leads to an increase of the average center-to-center distance. From the fit, we can take the FWHM, amplitude and position. The latter depicts the center-to-center distance, as mentioned before. The amplitude indicates the amount of scattering sources. The peak width is related to the distribution and displays the disorder of cluster and particle arrangement. Therefore, the FWHM is also a strong indicator which indicates a transition from fast increasing ordering to a slower growth of a two-dimensional (2D) cluster arrangement. This means a conversion from a diffusion-mediated coalescence regime to an adsorption-driven cluster growth regime. This expansion persists until one cluster grows radially into another one by the movement of grain boundaries and not by coalescence anymore (shown in
Figure 3a–c). This means the surface is covered and the newly arriving particles contribute only in the vertical growth. This describes the early formation (nucleation, diffusion, adsorption and grain growth) for thin films as known from literature [
5,
6,
10]. The relation
enables us to track the mean cluster distance during the growth as a quantitative parameter. The peak broadening in
Figure 3a at around 120 s corresponds to the growing additional scattering source of crystalline BaTiO
particles until the crystalline particles with
= 5.8 g/cm
have taken over the whole layer.
From the development of the Yoneda peak itself, we were able to get further information about the growth dynamics. Therefore, we have taken cuts in the
-direction at the position of the Yoneda peak (see
Figure 3c). The temporal evolution of the intensity at the Yoneda peak corresponds to the surface coverage or particle density at the surface, which leads to a change in the scattering distribution. At the beginning of the deposition, a clear Yoneda peak of
can be seen. At a deposition time of 20 s, the Yoneda peak of amorphous BaTiO
, with a
= 2.9 g/cm
, appears and after a deposition time of 60 s, another Yoneda peak appears. This is a relatively broad peak that shows up at
= 5.8 g/cm
which correspond to polycrystalline BaTiO
. As mentioned above, the Yoneda peak depends on the material—on the electron density, to be precise, which changes when the crystallization launches. This is were it shows an alteration around 30 s after the start of the deposition. It starts with a broadening of the Yoneda wing in
q, leads to a second line at the density of polycrystalline BaTiO
, and ends with a disappearing line at the density of the amorphous clusters. These density findings accord with the literature values for BaTiO
[
14,
15].
During this experiment, we collected the wide angle scattering (WAX) with a second detector simultaneously. An exemplary pattern taken after 500 s of deposition is shown in
Figure 4, it shows a detector tilt and rotation corrected pattern which was azimuthal resorted using the Python package pyFAI [
16]. The software was calibrated by performing a measuring with a NIST standard.
The peak positions for cubic BaTiO
are marked and show a good agreement. Small offsets at the higher angle peaks are shown and can be explained as finite-size effect and temperature introduced stress on the wafer and the thin film. The peak broadening of the [200]-peak and the double peak structure at the [211]-peak can be explained with the occurrence of tetragonal BaTiO
. The preferred orientation arises from particles being oriented on the surface and enhancing the intensities of more defined Bragg reflections, while reducing the intensities of others. [
18] The preferred orientation is diminishing with increasing film thickness as shown in
Figure 5. This can be explained with the fast increasing film thickness, a layer-plus-island growth and the therefore decreasing influence of the surface and relaxation process due to misfit dislocations [
5,
6], but still can be seen until the end of deposition.
Figure 6 shows the peak evolution by plotting the azimuthal resorted patterns versus time during the deposition at 600
C and selected GIXRPD pattern after 100, 500 and 1000 s.
Figure 7 shows exemplary pseudo-Voigt fits of the peaks. The increasing amplitude corresponds to the quantity of diffraction sources, i.e., crystallites, and the peak width correspond to the nano-crystalline size. The azimuthal integrated patterns versus time can be fitted with a pseudo-Voigt function to extract the peak width and amplitude.
Figure 6 shows the results of five second binned fits over time for the [110] and [200] peak, these peak positions are in good agreement with literature [
17,
19]. The [200] peak occurs after about 200 s of deposition and the [110] peak after about 250 s. This is about 140 s after the appearance of the crystalline Yoneda peak. In the early crystallization stage, the amount of diffraction sources are to small too overcome the background and noise in the GIXRPD detector. With the simultaneous use of two detectors for GISAXS and GIXRPD we have followed the crystallization in detail through the whole evolution of the thin film growth. First, it appears after around 60 s in cuts along
at the Yoneda peak (see
Figure 3c) as slowly growing peak at the density of crystalline BaTiO
. After a growing amount of scattering sources through the film thickness it leads to a broadening of the peak along
direction (see
Figure 2b and
Figure 3a) at around 100 s. And finally it appears as a scattering signal in the GIXRPD detector (see
Figure 5 and
Figure 6) after around 200 s.
To confirm and refine these findings, we used the software BornAgain (BA) [
20] to simulate and fit the evolution of the thin film. For the early formation, we get the best results for small sized hemispherical shaped particle (
= 2.9 g/cm
) with a small distribution on the height of 3 nm and the radius of 3 nm for the beginning and growing over time. In the early stage, there is only a cumulative disorder of inter-particle distances along the X-ray footprint. This corresponds to the early stage of nucleation, diffusion and adsorption (see
Figure 7a–c). Over time, the mean particle diameter rises, forming small clusters growing in diameter and shrinking in distance. At this stage, the best model assumes uniform hemispherical clusters of radius R arranged in a 2D hexagonal lattice with the correlation distance D between the lattice points. Assuming that all BaTiO
deposited on this triangular area is separated into three individual uniform clusters (see
Figure 3d), this leads to the equation:
With
R the micro-domain size,
the form factor ratio
, with
and
, where t is the side length of the hexagon, with the local maximum for close packing at
. This maximum emphasizes the fully covered surface as mentioned before (see
Figure 3a).
Since we were depositing on a substrate temperature higher than the known crystallization energy for BaTiO
[
1,
2], a re-crystallization was expected to happen, and was shown in the cuts along the
q direction at about 60 s. Following our evolution modeling, we have assumed a covered surface of hemispherical clusters (
= 2.9 g/cm
) with embedded particles (
= 5.8 g/cm
). For these particles with the density of crystalline BaTiO
, the best fitting shape are cylinders with a radius of 2 nm and a height of 3 nm which grow over time (see
Figure 7d). The fitting model with the highest agreement show these cylinders growing near the interface of the substrate to the thin film. After about 100 s of deposition, the best fitting model is a 10nm thick amorphous thin film with a high surface roughness and growing hemispherical ploy-crystalline clusters (see
Figure 7f). After the crystallization of the whole thin film the arriving amorphous particle coming from the sputtering target will quickly crystallize. The high surface roughness stays until the end of the deposition which is also confirmed by the missing interference pattern in the
-cuts at the
q = 0 position shown in
Figure 3c, which is in strong agreement with the literature [
10]. A schematic of the thin film evolution during the deposition is shown in
Figure 8.