Study of TiO 2 thin film nucleation and growth on planar substrates

In the first place, it is essential to investigate the TiO2 growth mechanism on planar substrates when establishing an ALD process on porous substrates. A series of TiO₂ thin films were deposited at 350 °C using TiCl₄ and H₂O precursors on as-received silicon substrates with a native oxide layer (named ”untreated”) and on silicon which has been treated by DCSBD source in air (named ”plasma treated”). The TiO2 films have linear growth on both substrates with increasing number of ALD cycles from 3 up to 5000, as expected, with roughness of approximately 10% of the film thickness (Figure 22 a and b). However, the nucleation regime of TiO2 films on plasma treated and untreated substrates is different (Figure 22 c and d). 50 and 80 ALD cycles are needed to form a continuous film on untreated and plasma treated silicon, respectively.

Figure 22. Film thickness and roughness vs. number of ALD cycles, (a) plasma treated, (b) untreated samples; initial growth region, (c) plasma treated (d) untreated. Note: Some error bars are not visible because the range is smaller than the data symbol.

No deposited material can be detected by ellipsometry for the first 3 and 5 cycles on plasma treated and untreated substrates, respectively. There is no continuous film formation with increasing number of ALD cycles up to 50 or 80 cycles on unteated or plasma treated substrates, respectively; the ellipsometric modeling showed zero thickness. Even though no film thickness can be detected in these ranges, the roughness increases with the number of cycles on both substrates. We can propose that isolated islands are formed before coherent film formation. The difference in film growth on plasma treated and untreated samples is not significant. A minor

distinctive difference was noticed in that continuous film formation took a greater number of cycles on plasma treated samples. This can be explained by 1) the roughness on plasma treated samples is larger and perhaps, more material is needed to cover rougher surface; 2) the thickness of the native oxide differs on the untreated substrate (2.1 nm) and on plasma treated samples (3.5nm). This may affect island formation.

Further investigation of the size of individual TiO2 crystallites on substrates before the formation of a continuous film can be obtained from fitting the optical parameters of the ellipsometric model for TiO2 thin films. The absorption edge obtained for thicker TiO2 films was 3.31 eV, which corresponds to the anatase structure. This value is in a good agreement with XRD measurements of TiO2 thin films deposited at 350 °C which showed that the anatase phase was dominant. With decreasing number of ALD cycles (≤ 50 cycles), where the film grows as isolated islands, the absorption edge of TiO₂ increases in energy (Figure 23 a). This is in a good agreement with the effect of quantum confinement (QC) in the isolated islands. The increase in bandgap, giving a large blue shift in the absorption edge, due to the decreasing size of the nanocrystals can be described by the effective mass model and adjusted equation for spherical crystallites:

∆𝐸_𝑔= ^ℎ2

2𝐷²𝑚₀[¹

𝑚_𝑒^∗+ ¹

𝑚_ℎ^∗] − ^3.6𝑒2

4𝜋𝜀₀𝜀_𝑟𝐷 (6.1)

where h is Planck’s constant, m_o is the electron mass, m_e^*, m_h^* are the effective electron and hole masses, respectively, e is the electronic charge, ε0 is the permittivity of free space, εr is the relative permittivity and D is the crystallite diameter.

Based on the calculation from equation (6.1), the relation between the calculated bandgap blue shift and the diameter of spherical crystallites can be illustrated in Figure 23 (b). The hemispherical crystallite size taken from the Figure 23 (b) can be plotted vs number of ALD cycles (Figure 24). The island size increases linearly until a continuous film starts to grow. The crystal size calculated from blue shift was compared with crystallite size measured by AFM and plotted in Figure 24. AFM measurements showed somewhat larger crystallite diameter which suggests that the crystallites are not strictly hemispherical, for instance, their height is less than their radius.

Figure 23. Optical bandgap of the TiO2 films as a function of the number of ALD cycles (a), Calculated bandgap blue shift as a function of the crystallite radius (b).

Figure 24. Crystallite diameter as a function of number of ALD cycles. The error bars shown on the AFM data points are estimates from the AFM image.

If it is assumed that crystallites are hemispherical, their height can be considered as the “film thickness”. The surface roughness layer parameters can be obtained from the ellipsometry material model (srough.mat) with a mixed layer of film material and void of adjustable ratio. The details of the fitting are described in Appendix B of Paper II. The surface density of crystallites was plotted versus the number of ALD cycles (Figure 25). With increasing number of ALD cycles the crystallite density is dropped from initial value of around 3–4 × 10¹⁷ m⁻³ to approximately 2–3 × 10¹⁵ m⁻³. This can be explained by Ostwald ripening, where larger crystallites are more energetically preferred than smaller ones. It is worth mentioning that crystallite density was slightly higher on plasma treated samples. This is in good correlation with crystallite size presented on Figure 24.

Figure 25. Surface density of crystallites as a function of the number of ALD cycles.

The growth rate of crystallite diameter of approximately 1.7 Å/cycle was obtained from Figure 24. This value can give the growth rate of TiO2 film of 0.85Å/cycle, which is in contrast with the growth rate of 0.5 Å/cycle for thicker film. This difference can be explained by the fact that crystallites grow not only by the deposition of a surface layer, but also by the ripening process mentioned above.