Metal-substrate-supported tungsten-oxide nanoarrays via porous-alumina-assisted anodization : From nanocolumns to nanocapsules and nanotubes

An array of highly aligned tungsten-oxide (TO) nanorods, ∼80 nm wide, up to 900 nm long, spatially separated at their bottoms by tungsten metal on a substrate is synthesized via the self-localized anodization of aluminum followed by the porous-alumina-assisted re-anodization of tungsten in a sputter-deposited Al/W bilayer. Moreover, the pore-directed TO nanocapsules may grow, which can be electrochemically top-opened in alumina nanopores and transformed to TO nanotubes, representing unique architectures built up on tungsten substrates to date. The as-grown nanorods are composed of amorphous WO3 mixed with minor amounts of WO2 and Al2O3 in the outer layer and oxide–hydroxide compound (WO3·nH2O) with aluminum tungstate (2Al2O3·5WO3), mainly present inside the rods. Once the growing oxide fills up the pores, it comes out as an array of exotic protuberances of highly hydrated TO, with no analogues among the other valve-metal oxides. Vacuum or air annealing at 550 °C increases the portion of non-stoichiometric oxides ‘doped’ with OH-groups and gives monoclinic WO2.9 or a mixture of WO3 and WO2.9 nanocrystalline phases, respectively. The nanorods show n-type semiconductor behavior when examined by Mott–Schottky analysis, with a high carrier density of 7 × 1019 or 3 × 1019 cm−3 for the air- or vacuum-annealed samples, associated with a charge depletion layer of about 8 or 10 nm, respectively. A model for the growth of the metal-substrate-separated TO nanocapsules and tubes is proposed and experimentally justified. The findings suggest that the new TO nanoarrays with well-defined nano-channels for carriers may form the basic elements for photoanodes or emerging 3-D micro- and nano-sensors.


S.1.1. Fitting of the EIS data of the air-annealed sample
The EIS data of the air-annealed sample (Fig. 8b) were best fitted using a circuit containing three R-CPE loops connected in series.The fitted values of effective capacitances (C 1 , C 2 , and C 3 ) and parallel resistances (R 1 , R 2 , and R 3 ), corresponding to three capacitive layers connected in series, having R 1 > R 2 > R 3 , are shown as their voltage-dependencies in Fig. S1.Owing to the typical behavior of a space-charge layer formed at a semiconductor-electrolyte interface (i.e.linear 1/C 2 (V), high R, R increasing with voltage), we assign the high-resistance capacitance (C 1 in Fig. S1) to a charge depletion layer formed at the column tops.The other two capacitances and their corresponding resistances (30-60 μF•cm −2 and ~10 2 -10 4 Ω•cm 2 , respectively, when taking into account a 10% column area) fit well with double layers, due to their magnitudes and low voltage dependence of the capacitance. 1ditional notes to the fitting:  R 1 was set to a minimum possible value during fitting, however, due to a short frequency range measured, the R 1 values are less reliable,  eventually, an R(RQ)(RQ) circuit can be used for fitting of the EIS spectra at voltages >0 V, which results in the same values of R 1 and C 1 , but having a bigger fitting error.

S 1.2. Fitting of the EIS data of the vacuum-annealed sample
The EIS data of the vacuum-annealed sample (Fig. 8c) were best fitted using a circuit of one R-CPE loop, having an additional CPE in series with the resistance.The fitted values of capacitances (C 0 and C 1 ) and parallel resistance (R 1 ), are shown as voltage-dependencies in Fig. S2.The voltage-dependent behavior of parameters C 1 and R 1 fits with a space-charge layer at column tops (linear 1/C 2 (V) and high R, but R is decreasing with voltage, although an increase is expected; on the other hand, the corresponding exponent of CPE is increasing with voltage from 0.85 to 0.95, in line with the theory).The capacitance C 0 of the R(RQQ) circuit, being in series with R 1 , can be related with diffusion processes in the electrolyte (it has a low CPE exponent of 0.3 to 0.8, not shown).
Additionally, several equivalent circuits may be used for fitting of the measured data: R(RQ) is more suitable for voltages below −0.

S.2.1. Microstructural analysis
The microstructural analysis from the peak broadening was performed with the Double-Voigt Approach 2 for all phases detected (see Table S1).According to this approach, in the best case the convolution of up to four functions can be fitted: a Lorentzian and Gaussian functions for the crystallite size effect (β LS and β GS ) and Lorentzian and Gaussian (β LD and β GD ) functions for the distortion or micro strain effects.However in practice it is difficult to use the four functions simultaneously because they are highly correlated unless the diffractogram covers a wide enough 2θ range.For the most abundant phase, W, the refining was worse if we only considered the term β LS .A quick observation of the diffractograms at high angles showed that the peaks were mainly Gaussians.In fact, an attempt to fit the peaks of W with the β LS failed and was only successful with the β GD component and with a minor contribution of β LS .Both the crystallite size and the mean microstrain, e o =Δd/d, can be estimated from the four components of the peak broadening with the well-known expressions. 3 is to be expected from a thin layer shape sample, the preferred orientation or texture is an important factor to be corrected for all phases that are deposited.The non-uniform intensity of the Debye rings along the γ angle is apparent.The simplest and easiest model for correcting the texture is the March-Dollase model, 4 which in the TOPAS software allows up to three independent parameters to be fitted for a maximum of two crystallographic directions.In the present case, only one direction for α-Ti and W phases was needed whereas two directions were used for WO 3 .

S.2.2 Peak asymmetry
The presence of α-Ti was established from both the asymmetry of the low-and high-angle peaks and the presence of the small peak at ≈35º 2θ.This small peak matches the 010 reflection for α-Ti as it is shown in Fig. 7a, and it is the only reflection of α-Ti that is alone in this figure.In Fig. 7b and 7c this small peak ≈35º 2θ is not as clearly seen as in Fig 7a because of the presence of tungsten oxide peaks.However, based on comparison of the diffractograms of the three samples -as-anodized, vacuum-annealed, and air-annealed, we assume that α-Ti is present in all of them.The 022 reflection of W is also overlapped at high angles by the 022 reflection of α-Ti.
We did not include the label of such reflection in Fig. 7 of the main text just for simplicity.
The recorded diffractograms (2θ range and the instrumental resolution) do not allow us to assure the reader on the presence of stacking faults in the tungsten layer.As far as we know, the presence of stacking faults in fcc structures produces a 2θ shift of the reflections as a function of their hkl indexes.In bcc structures, like W is, the effect could be similar but we have not found any precedent in the literature that would clearly mention the asymmetry of W peaks as a consequence of stacking faults.

S.2.3. Crystallite size
The estimated values of the crystallite size (for W and α-Ti ) and microstrain (for W) only orientate on the microstructure of all the phases involved.We tried to relate the rough microstructure deduced from the experimental diffractograms with the film preparation procedure, as it is mentioned in the paper.The crystallite size reported for α-Ti assumes that this phase does not have microstrain contribution to the peak broadening.However, for the W phase it was possible to distinguish between the crystallite size and microstrain effects to the peak broadening with the Double-Voigt Approach.We do not intend to make a rigorous study on the microstructure of each phase because of the resolution limit and 2θ range of the recorded diffractograms and because there is no need for a deeper insight into this matter in the frame of this article.

Fig. S1 .
Fig. S1.Dependencies of effective capacitances and parallel resistances on voltage for the airannealed sample, obtained by fitting of the measured EIS data shown in Fig. 8b using an R(RQ)(RQ)(RQ) circuit.
3 V and both R(RQQ) and R(RQ)(RQ) are possible for voltages above −0.3V, R(RQQ) showing lower fitting errors.The R and C parameters of the main capacitive layer of all circuits are very similar, thus only the parameters of the R(RQQ) fitting are shown in Fig. S2 and are used for Mott-Schottky evaluation.

Fig. S2 .
Fig. S2.Dependencies of effective capacitances and parallel resistance on voltage for the vacuum-annealed sample, obtained by fitting of the measured EIS data shown in Fig. 8c using an R(RQQ) circuit.

Table 1 :
Structural data used from the ICSD database, refined cell parameters, crystallite size, microstrain (e 0 ), preferred orientation correction and calculated wt.% for each phase.