Crystallisation Phenomena of In2O3:H Films

The crystallisation of sputter-deposited, amorphous In2O3:H films was investigated. The influence of deposition and crystallisation parameters onto crystallinity and electron hall mobility was explored. Significant precipitation of metallic indium was discovered in the crystallised films by electron energy loss spectroscopy. Melting of metallic indium at ~160 °C was suggested to promote primary crystallisation of the amorphous In2O3:H films. The presence of hydroxyl was ascribed to be responsible for the recrystallization and grain growth accompanying the inter-grain In-O-In bounding. Metallic indium was suggested to provide an excess of free electrons in as-deposited In2O3 and In2O3:H films. According to the ultraviolet photoelectron spectroscopy, the work function of In2O3:H increased during crystallisation from 4 eV to 4.4 eV, which corresponds to the oxidation process. Furthermore, transparency simultaneously increased in the infraredspectral region. Water was queried to oxidise metallic indium in UHV at higher temperature as compared to oxygen in ambient air. Secondary ion mass-spectroscopy results revealed that the former process takes place mostly within the top ~50 nm. The optical band gap of In2O3:H increased by about 0.2 eV during annealing, indicating a doping effect. This was considered as a likely intra-grain phenomenon caused by both (In0)O•• and (OH−)O• point defects. The inconsistencies in understanding of In2O3:H crystallisation, which existed in the literature so far, were considered and explained by the multiplicity and disequilibrium of the processes running simultaneously.


Introduction
Hydrogen doped indium oxide (In 2 O 3 :H) films demonstrate an electron Hall mobility (µ e ) of over 100 cm 2 /Vs [1,2]. Such outstanding property can be achieved if the In 2 O 3 film is deposited in an amorphous state and subsequently crystallised at T > 160 • C. Its optical transmittance is superior where I-are the intensities normalised to the Cs signal and L H and L O are the relative annealing losses of hydrogen and oxygen, respectively. Despite the reduced detection limit for hydrogen in the positive ion detection mode, its concentration in water containing samples was found to be high enough for a reliable determination. Fourier-transform infrared spectroscopy (FTIR) was performed on a Vertex 70 (Bruker Optics, Ettlingen, Germany) using the rock solid interferometer, mercury (Hg) IR-source, DLaTGS IR-Detector and the KBr Beam-splitter. In 2 O 3 :H films were deposited on double-side polished Si-substrates for this investigation. XRD patterns were recorded using Cu K α radiation in different scanning modes: symmetrical Bragg-Brentano and asymmetrical detector scan. In the in-plane measurement, both incident and diffracted beams had grazing angles to the sample surface. D8 Discover (Bruker, Karlsruhe, Germany) and X'Pert MRD Pro (PANalytical, Almelo, The Netherlands) diffractometers were used for these tasks. LaB 6 powder (660c) was used as a standard for estimation of the crystallite size. The Hall mobility of charge carriers was measured with an Ecopia HMS-3000 system (Anyang-city, Gyeonggi-do, Korea) in van der Pauw geometry at room temperature. The proportionality factor was taken equal to 1. Scanning electron microscopy (SEM) was made using a Hitachi S4100 microscope (Hitachi High-Technologies Corporation, Tokyo, Japan).
Samples for cross sectional transmission electron microscopy (TEM) were prepared by gluing the thin films face-to-face, followed by mechanical and ion thinning for electron transparency. TEM investigations were performed on two microscopes. The Philips CM12 (FEI Company, Hillsboro, OR, USA) was used at 120 kV accelerating voltage for preliminary investigation. The system Zeiss LIBRA 200 FE (Carl Zeiss Microscopy, Jena, Germany) operated at 200 kV accelerating voltage was used for more detailed analysis by electron energy loss spectroscopy (EELS). This microscope is equipped with an in-column energy filter for energy filtered image acquisition. The set electron energy loss was varied from 0 to 30 eV.

Crystallinity versus Electron Mobility
The crystallinity of TCO films, i.e., the size of grains and the texture, can determine electron mobility to a different extent, which depends on the individual material. According to the band structure calculations accepted for In 2 O 3 , conduction band minima are formed by the 5 s states [20][21][22]. Because of their spherical symmetry, one may then expect no significant impact of the coherence between grains (texture) onto the grain boundary scattering. However, a limiting role of the grain boundaries themselves (grain size) illustrates the following fact: the undoped single crystalline In 2 O 3 films and In 2 O 3 single crystals demonstrate an electron mobility of more than 100 cm 2 /Vs [23,24] whereas the as-prepared polycrystalline films do not reveal µ e exceeding 50-60 cm 2 /Vs [25,26]. This fact is basically related to the difference in N e , which is known to be much lower in single crystals (~10 17 cm −3 ) as compared to the polycrystalline materials (10 19 cm −3 ) [23]. In turn, high N e values in polycrystalline In 2 O 3 are usually attributed to the so called unintentional doping caused by the presence of water in ambient air. Additional negative impact on mobility provide ionized impurities.
In this work, we intended to examine to what extent the grain size and texture determine Hall mobilities in In 2 O 3 :H films. Figure 1 presents the X-ray patterns and Hall mobility data for various as-deposited and annealed films. Here, two series of experiments were done: variation of the film-thickness (a) and variation of the total pressure during deposition (b). As one may see from the series (a), all ≥200 nm thick In 2 O 3 :H 2 O films appear to be X-ray crystalline. Diffraction maxima are, however, quite broad. As we explained above, this is due to the heating of the surface by the RF plasma. This effect increases with time due to the NIR absorption of the growing film and at some point the film becomes crystalline (see Figure S1).
polycrystalline In2O3 are usually attributed to the so called unintentional doping caused by the presence of water in ambient air. Additional negative impact on mobility provide ionized impurities.
In this work, we intended to examine to what extent the grain size and texture determine Hall mobilities in In2O3:H films. Figure 1 presents the X-ray patterns and Hall mobility data for various as-deposited and annealed films. Here, two series of experiments were done: variation of the filmthickness (a) and variation of the total pressure during deposition (b). As one may see from the series (a), all ≥200 nm thick In2O3:H2O films appear to be X-ray crystalline. Diffraction maxima are, however, quite broad. As we explained above, this is due to the heating of the surface by the RF plasma. This effect increases with time due to the NIR absorption of the growing film and at some point the film becomes crystalline (see Figure S1).
Post-deposition annealing at 180 °C for 60 min in ambient air results in further crystallisation and increase of electron mobility. The biggest gain of mobility is demonstrated by thinner, initially X-ray amorphous, films. All partially crystalline In2O3:H2O films thicker than 300 nm reveal almost the same gain of mobility after annealing. Analysing XRD patterns in Figure 1a, one can find two In2O3 phases: initially crystalline and crystallised after annealing. A coexistence of two phases is especially well visible in the 400 nm thick film. Correlating this with Hall mobility data, we conclude that exactly the post-deposition crystallisation is crucial for high mobility. Interestingly, the "highmobility" phase reveals a smaller lattice constant than the phase, which crystallises during deposition. The origin of this difference will be discussed below.
A decrease of intensity of diffraction peaks is observed for the initially crystalline In2O3 phase as a result of annealing. The most probable reason for that is active recrystallization. Figure 1b demonstrates the influence of the sputtering pressure on the crystallisation behaviour of X-ray amorphous films. Apparently, this parameter determines the film texture and Hall mobility. We attribute this effect to the amount of hydroxyl in a film. Indeed, using a fixed leak rate of the needle valve feeding water vapour, a higher ptot should result in a smaller p(H2O) and, hence, in a smaller concentration of hydroxyl in In2O3. In turn, the impact of hydroxyl on the crystalline growth of In2O3 is a known phenomenon, thus, the (400)-orientation can be suppressed if water is present in a sputtering gas [27]. This is consistent with our results. A higher process pressure itself may also contribute, as it increases the energy transfer from plasma to substrate, resulting in a higher deposition temperature [28].  Post-deposition annealing at 180 • C for 60 min in ambient air results in further crystallisation and increase of electron mobility. The biggest gain of mobility is demonstrated by thinner, initially X-ray amorphous, films. All partially crystalline In 2 O 3 :H 2 O films thicker than 300 nm reveal almost the same gain of mobility after annealing. Analysing XRD patterns in Figure 1a, one can find two In 2 O 3 phases: initially crystalline and crystallised after annealing. A coexistence of two phases is especially well visible in the 400 nm thick film. Correlating this with Hall mobility data, we conclude that exactly the post-deposition crystallisation is crucial for high mobility. Interestingly, the "high-mobility" phase reveals a smaller lattice constant than the phase, which crystallises during deposition. The origin of this difference will be discussed below. A decrease of intensity of diffraction peaks is observed for the initially crystalline In 2 O 3 phase as a result of annealing. The most probable reason for that is active recrystallization. Figure 1b demonstrates the influence of the sputtering pressure on the crystallisation behaviour of X-ray amorphous films. Apparently, this parameter determines the film texture and Hall mobility. We attribute this effect to the amount of hydroxyl in a film. Indeed, using a fixed leak rate of the needle valve feeding water vapour, a higher p tot should result in a smaller p(H 2 O) and, hence, in a smaller concentration of hydroxyl in In 2 O 3 . In turn, the impact of hydroxyl on the crystalline growth of In 2 O 3 is a known phenomenon, thus, the (400)-orientation can be suppressed if water is present in a sputtering gas [27]. This is consistent with our results. A higher process pressure itself may also contribute, as it increases the energy transfer from plasma to substrate, resulting in a higher deposition temperature [28].
Considering the cubic bixbyite structure of In 2 O 3 , one can distinguish two kinds of InO 6 octahedrons which interconnect either by corners or by edges [29]. As the octahedrons are directed along the [111] axes in the lattice, the close packed oxygen layers coincide with the {00l} planes. This explains why the <00l> oriented films might reveal a smaller electron mobility.
In our further XRD, TEM and SIMS investigations we used the~150 nm thick films. XRD measurements (see Figure S2) were performed to estimate the size of coherent scattering in both lateral (D lat ) and longitudinal (D long ) directions using Scherrer's method [30]. In case of lateral size the asymmetric in-plane XRD measurements at ψ = 89.2 • were performed. Assuming single strength and linear d hkl − sin 2 ψ dependence we estimated the residual stress in the films. A complex recalculation which takes into account the measurement conditions was performed on the basis of a procedure explained elsewhere [31]. The elastic constants for In 2 O 3 were taken from literature [32,33]. Our results are collected in Figure S2 and Table 1. Hall measurement data are also presented for the assessment. We see that the crystalline state itself does not secure high electron mobility: compare, for instance, the crystalline state III with the amorphous state II-both films were deposited in the presence of water vapour. The positive impact of water on µ e is, however, obvious (compare state I with other states). This effect is explained in literature by a passivation of dangling bonds with hydrogen that decreases the electron scattering [9]. One can see that such passivation is highly pronounced exactly in the amorphous state with the utmost amount of dangling bonds (compare state II with I or III). We observe that µ e strongly increases with the grain size, compare states III → (IV, V). Thus, both qualities: the degree of crystallinity and the passivation of dangling bonds are important for reaching a high electron mobility in In 2 O 3 .
It should be noted why the concentration of free electrons changes. The presence of water does not result in a marked change of N e (compare states I and II), but high temperature does. This effect will be discussed below.
One has to notice the presence of residual compressive stress in crystalline films, especially if crystallisation takes place during deposition (states I and III). The values presented are just an estimation with a large error, based on the measurements of only two ψ values. Two reflexions (222) and (400) with quite similar Poisson's ratio were taken into account [33].
It is a fact that in the presence of water the growing In 2 O 3 films contain hydroxyl groups [9]. Moreover, the hydroxylation apparently stipulates the amorphous state of In 2 O 3 [34]. We have, however, made some curious observations in our experiments, which cannot be easily explained. Thus, the In 2 O 3 :H 2 O films (state II in the Table 1) grow predominantly X-ray crystalline on such substrates like molybdenum-films, i-ZnO, Si-wafer, or CIGS-absorber [11]. Moreover, if any additional oxygen is injected into the sputtering gas, the films becomew crystalline and resistive.

Presence and Role of Metallic Indium
Since the films appearing as X-ray amorphous could still be nano-crystalline, we undertook TEM investigations of them. Figure 2 shows cross-sectional brightfield TEM images of an In 2 O 3 :H 2 O film deposited at p tot = 0.5 Pa on glass.

Presence and Role of Metallic Indium
Since the films appearing as X-ray amorphous could still be nano-crystalline, we undertook TEM investigations of them. Figure 2 shows cross-sectional brightfield TEM images of an In2O3:H2O film deposited at ptot = 0.5 Pa on glass. Several effects can be observed. Initially, the film is amorphous, but it changes under the electron-beam after about 20 min. A diffraction contrast becomes visible, indicating crystalline structures. At the same time, droplet-like features at the film/glass interface increase in number and size. Being aware of the inherent effects of our fabrication procedure, we presume here an effect of the sample heating by the e-beam. Basically, heating of TEM-lamella due to inelastic interactions with high energy electrons is a known phenomenon [35]. According to the literature, relatively thick In2O3:H films crystallise at 180-200 °C [1][2][3]. In our special case we deal with much thinner lamella, where the impact of surface and contact interface is apparently larger. This fact may shift the crystallisation temperature to lower values. Apart from the crystallisation effect observed, we assume that the appearance of droplet-like features is related to the presence of metallic indium. Its melting (Tm = 156.6 °C) might promote In2O3:H2O crystallisation and results in accumulation of In-droplets.
Actually, the appearance of metallic indium seems to be rather probable in our case as the metallic phase was found in In2O3 and In2O3:Sn films by other investigators [36,37]. Consideration of the In-O phase diagram and general chemistry of indium oxide/hydroxide system also supports this assumption [34,38].
We applied energy filtered TEM to detect metallic indium, as it exhibits a bulk plasmon with an energy of about 12 eV [39]. The plasmonic spectrum of In2O3 [40] could not be observed in this study. The set of TEM images obtained at distinct loss energies is presented in the Supplementary Materials ( Figure S3). Metallic indium appears as bright areas at an electron energy loss of ~12 eV. We also found that indium congregates either at the film-glass interface or localises in the bulk (see Figure  S4). The latter case is shown in more detail in Figure 3, where the correlation of crystallinity ( Figure  3a) and appearance of metallic indium (Figure 3b) can be observed. The top In2O3 layer grows crystalline due to the impact of RF plasma as we discussed above. The crystalline part consists of columnar crystallites of 20-40 nm lateral size, which is consistent with the film state III in Table 1. One can conceivably detect some porosity within this area (see Figure 3a and Figure S3). Metallic indium particles segregate exactly at the transition region between amorphous and crystalline layers ( Figure 3b). These nanoparticles were found to be crystalline ( Figure S5). Obviously, melting of indium and hence its extraction in a separate phase is associated with indium oxide crystallisation. The liquid phase is known to support even high quality crystallisation in such methods as liquid Several effects can be observed. Initially, the film is amorphous, but it changes under the electron-beam after about 20 min. A diffraction contrast becomes visible, indicating crystalline structures. At the same time, droplet-like features at the film/glass interface increase in number and size. Being aware of the inherent effects of our fabrication procedure, we presume here an effect of the sample heating by the e-beam. Basically, heating of TEM-lamella due to inelastic interactions with high energy electrons is a known phenomenon [35]. According to the literature, relatively thick In 2 O 3 :H films crystallise at 180-200 • C [1][2][3]. In our special case we deal with much thinner lamella, where the impact of surface and contact interface is apparently larger. This fact may shift the crystallisation temperature to lower values. Apart from the crystallisation effect observed, we assume that the appearance of droplet-like features is related to the presence of metallic indium. Its melting (T m = 156.6 • C) might promote In 2 O 3 :H 2 O crystallisation and results in accumulation of In-droplets.
Actually, the appearance of metallic indium seems to be rather probable in our case as the metallic phase was found in In 2 O 3 and In 2 O 3 :Sn films by other investigators [36,37]. Consideration of the In-O phase diagram and general chemistry of indium oxide/hydroxide system also supports this assumption [34,38].
We applied energy filtered TEM to detect metallic indium, as it exhibits a bulk plasmon with an energy of about 12 eV [39]. The plasmonic spectrum of In 2 O 3 [40] could not be observed in this study. The set of TEM images obtained at distinct loss energies is presented in the Supplementary Materials ( Figure S3). Metallic indium appears as bright areas at an electron energy loss of~12 eV. We also found that indium congregates either at the film-glass interface or localises in the bulk (see Figure S4). The latter case is shown in more detail in Figure 3, where the correlation of crystallinity ( Figure 3a) and appearance of metallic indium (Figure 3b) can be observed. The top In 2 O 3 layer grows crystalline due to the impact of RF plasma as we discussed above. The crystalline part consists of columnar crystallites of 20-40 nm lateral size, which is consistent with the film state III in Table 1. One can conceivably detect some porosity within this area (see Figures 3a and S3). Metallic indium particles segregate exactly at the transition region between amorphous and crystalline layers ( Figure 3b). These nanoparticles were found to be crystalline ( Figure S5). Obviously, melting of indium and hence its extraction in a separate phase is associated with indium oxide crystallisation. The liquid phase is known to support even high quality crystallisation in such methods as liquid phase epitaxy [41], metal-modulated epitaxy [42], volatile surfactant assisted chemical vapour deposition [43] and others.

Optical Properties of In2O3:H2O and In2O3:H
Optical measurements allow determining properties of the continuous matter. If a material is crystalline, we receive the information from the interior area of grains, whereas the electrical properties are cumulative. Many optical investigations of In2O3 and In2O3:H films have been undertaken [2,36,44,45]. According to S. Joseph and S. Berger, the fitting of IR optical transmission spectra by effective medium approximation reveals that no changes in In2O3 transmittance should be observed if the volume fraction of metallic indium remains less than 10%. A larger indium excess results in a markedly lower transmittance as compared to that experimentally observed [36]. The presence of a metallic phase can be also deduced from the temperature dependence of the resistivity [46]. This, however, demands an even higher volumetric content of indium for the percolation of electrical current. Figure 4 presents fitted optical spectra for the thin films. The fits were obtained with the help of the RIG-VM software developed at Fraunhofer IST [47]. A Tauc-Lorentz oscillator has been used for the fundamental absorption and a Drude term for the free electrons. A free electron mass of m* = 0.28 me has been used [23]. As one can see, the optical mobility matches well with the Hall mobility (compare with Table 1) for both film states. Some deviation is observed for the crystalline state, where the Drude term gives a somewhat smaller mobility. This can be attributed to the insufficient spectral range to fit the plasma resonance of free carriers accurately. A remarkable difference in Ne values should be noted, namely, the electrical measurements gave an approximately double concentration of free electrons, which are not visible optically. This means that we might observe an additional inter-grain source of free electrons.

Optical Properties of In 2 O 3 :H 2 O and In 2 O 3 :H
Optical measurements allow determining properties of the continuous matter. If a material is crystalline, we receive the information from the interior area of grains, whereas the electrical properties are cumulative. Many optical investigations of In 2 O 3 and In 2 O 3 :H films have been undertaken [2,36,44,45]. According to S. Joseph and S. Berger, the fitting of IR optical transmission spectra by effective medium approximation reveals that no changes in In 2 O 3 transmittance should be observed if the volume fraction of metallic indium remains less than 10%. A larger indium excess results in a markedly lower transmittance as compared to that experimentally observed [36]. The presence of a metallic phase can be also deduced from the temperature dependence of the resistivity [46]. This, however, demands an even higher volumetric content of indium for the percolation of electrical current. Figure 4 presents fitted optical spectra for the thin films. The fits were obtained with the help of the RIG-VM software developed at Fraunhofer IST [47]. A Tauc-Lorentz oscillator has been used for the fundamental absorption and a Drude term for the free electrons. A free electron mass of m* = 0.28 m e has been used [23]. As one can see, the optical mobility matches well with the Hall mobility (compare with Table 1) for both film states. Some deviation is observed for the crystalline state, where the Drude term gives a somewhat smaller mobility. This can be attributed to the insufficient spectral range to fit the plasma resonance of free carriers accurately. A remarkable difference in N e values should be noted, namely, the electrical measurements gave an approximately double concentration of free electrons, which are not visible optically. This means that we might observe an additional inter-grain source of free electrons.  Based on the measurements in the UV range, we determined optical band gaps based on the Tauc-Lorentz model. These values are presented in Table 2 for the material states I, II and V. According to the literature, there cannot be an indirect gap in pure In2O3 due to the parabolic nature of the conduction band [22]. Moreover, the minimum band gap at 2.9 eV is symmetrically forbidden and the first allowed optical transition occurs from the level ~0.8 eV below the valence band top that gives the commonly observed value of Eg ≈ 3.7 eV. As presented in Table 2, the asdeposited crystalline In2O3 films (state I) reveal considerably lower Eg as compared to the expected value for this material. Amorphous films, which probably contain water (state II), demonstrate even lower values; however, for 400 nm thick films the difference between states I and II is negligible. This is consistent with the fact of partial crystallisation observed in thicker In2O3:H2O films (see Figures 1  and 3). However, no influence of water in the state II is then noticed. We do not detect the influence of expected hydrogen doping in the post-crystallised In2O3:H films (state V) as well, because the optical Eg values observed are very close to the ones known for the pure In2O3.

Chemical Changes in In2O3
To understand if there is any chemical transformation during crystallisation and which doping mechanism is realised, we performed further investigations. According to the FTIR analysis (data are not presented in this paper) none of the OH or adsorbed water (1615-1630 cm −1 ) features were observed in crystalline films. This might mean that these species, if they exist, are concentrated mainly at the grain boundaries. The IR transmittance, however, differs considerably for different film states. Thus, an addition of water during deposition results in a significant reduction of the IR transmittance (states I and II are compared). Annealing in vacuum does not significantly change it (states II → V), whereas annealing in ambient air provides quite a strong increase of the IR transmittance (states II → IV). Taking into account our TEM/EELS results, we attribute this behaviour to the presence of metallic indium in the films deposited in the presence of water. Annealing of such films in ambient air provides more effective oxidation of indium as compared to the annealing in vacuum. Based on the measurements in the UV range, we determined optical band gaps based on the Tauc-Lorentz model. These values are presented in Table 2 for the material states I, II and V. According to the literature, there cannot be an indirect gap in pure In 2 O 3 due to the parabolic nature of the conduction band [22]. Moreover, the minimum band gap at 2.9 eV is symmetrically forbidden and the first allowed optical transition occurs from the level~0.8 eV below the valence band top that gives the commonly observed value of E g ≈ 3.7 eV. As presented in Table 2, the as-deposited crystalline In 2 O 3 films (state I) reveal considerably lower E g as compared to the expected value for this material. Amorphous films, which probably contain water (state II), demonstrate even lower values; however, for 400 nm thick films the difference between states I and II is negligible. This is consistent with the fact of partial crystallisation observed in thicker In 2 O 3 :H 2 O films (see Figures 1  and 3). However, no influence of water in the state II is then noticed. We do not detect the influence of expected hydrogen doping in the post-crystallised In 2 O 3 :H films (state V) as well, because the optical E g values observed are very close to the ones known for the pure In 2 O 3 .

Chemical Changes in In 2 O 3
To understand if there is any chemical transformation during crystallisation and which doping mechanism is realised, we performed further investigations. According to the FTIR analysis (data are not presented in this paper) none of the OH or adsorbed water (1615-1630 cm −1 ) features were observed in crystalline films. This might mean that these species, if they exist, are concentrated mainly at the grain boundaries. The IR transmittance, however, differs considerably for different film states. Thus, an addition of water during deposition results in a significant reduction of the IR transmittance (states I and II are compared). Annealing in vacuum does not significantly change it (states II → V), whereas annealing in ambient air provides quite a strong increase of the IR transmittance (states II → IV). Taking into account our TEM/EELS results, we attribute this behaviour to the presence of metallic indium in the films deposited in the presence of water. Annealing of such films in ambient air provides more effective oxidation of indium as compared to the annealing in vacuum.
SIMS was used to determine the In/O and H/O ratios across the film. As we pointed out in experimental section the DC-sputtered~150 nm thick In 2 O 3 films were analysed in this case. Figure 5a compares the In/O ratio in as-prepared and annealed films deposited with and without water. As one can see, In 2 O 3 films contain less oxygen and do not change during annealing in vacuum. According to our visual observations, In 2 O 3 films are usually darker than In 2 O 3 :H films. Therefore, we assume the In 2 O 3 to be oxygen deficient. On the contrary, very transparent In 2 O 3 :H films seem to possess a stoichiometry close to x = 3 in In 2 O x . Curiously, we observe the same oxygen enrichment profile in the upper~50 nm of both: as-deposited and annealed In 2 O 3 films. At the very surface this enrichment approaches the overall level observed in the annealed In 2 O 3 :H film. As this effect is insensitive to annealing, we most probably deal with the impact of a short exposure to the ambient air prior to the measurement. It is observed only in the case of tiny crystalline, sub-stoichiometric films, which indicates a reaction in the inter grain space.
( Figure 5c). Obviously, hydrogen-to-oxygen ratio is higher in as-deposited state. Its depth profile demonstrates several pronounced maxima, which correspond to the substrate oscillation and passing by the inlet of water vapour. This result validates our procedure of hydrogen detection, proving the satisfactory sensitivity, which we can only achieve for the water-containing films. On the other hand we realise that hydrogen detected in as-deposited films represents most likely just water. A comparison of LH and LO discloses an interesting effect: the oxygen loss remains stable over the entire film thickness, whereas hydrogen releases more actively from the top but demonstrates a stable LH in the depth. We suggest the release of H2O and H2 species from the film being annealed (Figure 5c), as only their formation in a free volatile form is chemically possible. Water can evaporate in its free form if it is contained or released from the hydroxyl groups as shown in the reaction (4) below. Hydrogen would form only in the presence of metallic indium according to the reactions: In + 3 H2O → In(OH)3 + 3/2 H2 In + In(OH)3 → In2O3 + 3/2 H2 The probability of such reactions and some supporting experimental data published will be considered in the discussion chapter below.
Considering SIMS results, we could not operate with the absolute values since we did not use any external standard. However, the qualitative suggestions made were based on internal standards-indium and oxygen. We realised that the crystallised In2O3:H film, which is a high mobility TCO to be used in various devices, might suffer from the chemical heterogeneity. (b) (c) Figure 5. SIMS depth profiles obtained on various ~150 nm thick DC sputtered In2O3 films. Indium to oxygen concentration ratios (a) were obtained for the film states I, II and V, which correspond to the Table 1. State Ia represents the UHV-annealed In2O3 film. Hydrogen to oxygen concentration ratio (b) and the percentage losses of hydrogen and oxygen (c) are compared for the as-deposited and annealed states of the film, intentionally containing water.
To observe the processes taking place on the film surface, UPS and XPS measurements were undertaken. Our XPS measurements (spectra are not shown here) reproduced the results obtained by Hans F. Wardenga, where In2O3:H2O films revealed a shoulder at the O1s emission line at about 532.6 eV binding energy [9]. This was found to correspond to the OH bonds, which disappear after annealing.
The UPS spectra acquired during a stepwise increase of temperature from the ambient level (~25 °C) up to 230 °C in UHV show the following changes in the In2O3:H2O film ( Figure 6). A ~0.4 eV shift in the secondary electron edge is observed. The secondary electron edge can be used to determine the work function of a material according to the relation Wf = Eex − EBsec. Thus, we observe here the Wf change from ~4.0 eV to ~4.4 eV that basically contradicts the doping phenomenon. It is worth noticing that the work function of the thermally deposited fully oxidised indium oxide is 5.0 eV [49]. This means that indium in the films in question has a lower oxidation state than in the stoichiometric oxide. The value of Wf is also determined to a large extent by the crystallographic orientation [50]. In our case the observed increase of a work function is probably caused by the appearance of a distant order. The significant difference relative to the fully oxidised In2O3 state can be connected to the presence of metallic indium in the films (our TEM and FTIR data). Water containing, as-deposited films (state II) have the highest oxygen content among the investigated systems. It decreases after annealing (state V) together with the increase of transparency in both UV and near IR spectral regions. We attribute this oxygen loss to the release of water, as the In/O ratio remains unchanged in the water-free In 2 O 3 sample under the same treatment. Furthermore, active oxygen diffusion in In 2 O 3 starts at temperatures above 600 • C [48].
For better understanding, we represented the measured data in the form of an H/O ratio (Figure 5b) and the percentage loss of hydrogen and oxygen (see formulas (1)) as a result of the annealing (Figure 5c). Obviously, hydrogen-to-oxygen ratio is higher in as-deposited state. Its depth profile demonstrates several pronounced maxima, which correspond to the substrate oscillation and passing by the inlet of water vapour. This result validates our procedure of hydrogen detection, proving the satisfactory sensitivity, which we can only achieve for the water-containing films. On the other hand we realise that hydrogen detected in as-deposited films represents most likely just water. A comparison of L H and L O discloses an interesting effect: the oxygen loss remains stable over the entire film thickness, whereas hydrogen releases more actively from the top but demonstrates a stable L H in the depth. We suggest the release of H 2 O and H 2 species from the film being annealed (Figure 5c), as only their formation in a free volatile form is chemically possible. Water can evaporate in its free form if it is contained or released from the hydroxyl groups as shown in the reaction (4) below. Hydrogen would form only in the presence of metallic indium according to the reactions: In + In(OH) 3 The probability of such reactions and some supporting experimental data published will be considered in the discussion chapter below.
Considering SIMS results, we could not operate with the absolute values since we did not use any external standard. However, the qualitative suggestions made were based on internal standards-indium and oxygen. We realised that the crystallised In 2 O 3 :H film, which is a high mobility TCO to be used in various devices, might suffer from the chemical heterogeneity.
To observe the processes taking place on the film surface, UPS and XPS measurements were undertaken. Our XPS measurements (spectra are not shown here) reproduced the results obtained by Hans F. Wardenga, where In 2 O 3 :H 2 O films revealed a shoulder at the O1s emission line at about 532.6 eV binding energy [9]. This was found to correspond to the OH bonds, which disappear after annealing.
The UPS spectra acquired during a stepwise increase of temperature from the ambient level (~25 • C) up to 230 • C in UHV show the following changes in the In 2 O 3 :H 2 O film ( Figure 6). A~0.4 eV shift in the secondary electron edge is observed. The secondary electron edge can be used to determine the work function of a material according to the relation W f = E ex − EB sec . Thus, we observe here the W f change from~4.0 eV to~4.4 eV that basically contradicts the doping phenomenon. It is worth noticing that the work function of the thermally deposited fully oxidised indium oxide is 5.0 eV [49]. This means that indium in the films in question has a lower oxidation state than in the stoichiometric oxide. The value of W f is also determined to a large extent by the crystallographic orientation [50]. In our case the observed increase of a work function is probably caused by the appearance of a distant order. The significant difference relative to the fully oxidised In 2 O 3 state can be connected to the presence of metallic indium in the films (our TEM and FTIR data). To observe the processes taking place on the film surface, UPS and XPS measurements were undertaken. Our XPS measurements (spectra are not shown here) reproduced the results obtained by Hans F. Wardenga, where In2O3:H2O films revealed a shoulder at the O1s emission line at about 532.6 eV binding energy [9]. This was found to correspond to the OH bonds, which disappear after annealing.
The UPS spectra acquired during a stepwise increase of temperature from the ambient level (~25 °C) up to 230 °C in UHV show the following changes in the In2O3:H2O film ( Figure 6). A ~0.4 eV shift in the secondary electron edge is observed. The secondary electron edge can be used to determine the work function of a material according to the relation Wf = Eex − EBsec. Thus, we observe here the Wf change from ~4.0 eV to ~4.4 eV that basically contradicts the doping phenomenon. It is worth noticing that the work function of the thermally deposited fully oxidised indium oxide is 5.0 eV [49]. This means that indium in the films in question has a lower oxidation state than in the stoichiometric oxide. The value of Wf is also determined to a large extent by the crystallographic orientation [50]. In our case the observed increase of a work function is probably caused by the appearance of a distant order. The significant difference relative to the fully oxidised In2O3 state can be connected to the presence of metallic indium in the films (our TEM and FTIR data).  According to the Figure 6b, vacuum annealing causes a shift of the valence band edge by~0.2 eV. All observed changes are be depicted on the energy diagram, where the UPS data are used to fix the E F and E VB levels ( Figure 7). Here we used a caption E g, min (minimum) for the fundamental band gap, which is known to be 2.9 eV for the pure In 2 O 3 [23]. If we admit any doping in our films, it can be even smaller due to the band gap narrowing phenomenon [51,52]. The optical E g, opt values obtained in this study were placed in accordance with the principle described above [22]. These energy diagrams show that the Fermi level is very close to the conduction band in both materials: amorphous and crystalline. If we admit the same fundamental band gap width for both states, then the latter would be a non-degenerate semiconductor that should not be the case for such high concentration of free electrons. Since we observed a Burstein-Moss shift as a result of the annealing, the In 2 O 3 :H likely remains degenerate due to the band gap shrinkage. Major changes happen with a level of the allowed optical transition inside the valence band, namely, it shifts markedly downwards. This can be attributed to the effect of crystallisation as the valence band contains fully occupied 2p and 2s oxygen states and empty 4d-indium states [22]. According to the Figure 6b, vacuum annealing causes a shift of the valence band edge by ~0.2 eV. All observed changes are be depicted on the energy diagram, where the UPS data are used to fix the EF and EVB levels (Figure 7). Here we used a caption Eg, min (minimum) for the fundamental band gap, which is known to be 2.9 eV for the pure In2O3 [23]. If we admit any doping in our films, it can be even smaller due to the band gap narrowing phenomenon [51,52]. The optical Eg, opt values obtained in this study were placed in accordance with the principle described above [22]. These energy diagrams show that the Fermi level is very close to the conduction band in both materials: amorphous and crystalline. If we admit the same fundamental band gap width for both states, then the latter would be a non-degenerate semiconductor that should not be the case for such high concentration of free electrons. Since we observed a Burstein-Moss shift as a result of the annealing, the In2O3:H likely remains degenerate due to the band gap shrinkage. Major changes happen with a level of the allowed optical transition inside the valence band, namely, it shifts markedly downwards. This can be attributed to the effect of crystallisation as the valence band contains fully occupied 2p and 2s oxygen states and empty 4d-indium states [22]. The observations made require a more detailed discussion of the In2O3 chemistry and possible origin of doping.

Discussion
To understand the results obtained in this work, we have to review some basic properties of indium, In2O3 and In(OH)3.

Appearance of Metallic Indium in In2O3
The electro-chemical potential of metallic indium is φ 0 = −0.3382 V [53]. This means that under normal conditions, metallic indium should not reduce protons in an aqueous solution to molecular hydrogen. Nevertheless, the potential is not too high and both reactions, reduction of metal and reduction of hydrogen, may proceed simultaneously on a competitive basis.
According to the In-O phase diagram, there is no detectable phase of oxygen non-stoichiometry [38]. If any In-excess is provided (≥0.02 at.%), there is a mixture of two phases: In2O3 and metallic In, which is solid below 156.634 °C and liquid above this temperature. Indium (III) oxide is thermodynamically very stable over a wide range of T and p(O2). According to the Ellingham diagram, one needs a p(O2) of about 10 −100 atm. in order to reduce it to metallic indium at room The observations made require a more detailed discussion of the In 2 O 3 chemistry and possible origin of doping.

Discussion
To understand the results obtained in this work, we have to review some basic properties of indium, In 2 O 3 and In(OH) 3 .

Appearance of Metallic Indium in In 2 O 3
The electro-chemical potential of metallic indium is ϕ 0 = −0.3382 V [53]. This means that under normal conditions, metallic indium should not reduce protons in an aqueous solution to molecular hydrogen. Nevertheless, the potential is not too high and both reactions, reduction of metal and reduction of hydrogen, may proceed simultaneously on a competitive basis.
According to the In-O phase diagram, there is no detectable phase of oxygen non-stoichiometry [38]. If any In-excess is provided (≥0.02 at.%), there is a mixture of two phases: In 2 O 3 and metallic In, which is solid below 156.634 • C and liquid above this temperature. Indium (III) oxide is thermodynamically very stable over a wide range of T and p(O 2 ). According to the Ellingham diagram, one needs a p(O 2 ) of about 10 −100 atm. in order to reduce it to metallic indium at room temperature. The equilibrium oxygen partial pressure at 200 • C is about 10 −55 atm. At the same time, metallic indium remains stable in air and starts to oxidise visibly only after melting. The oxidation of indium in a liquid form proceeds about five time faster as compared to the solid [54]. Aside from the main oxidation state +3, indium may have also +2 and +1 in combination with oxygen. A formal oxidation state +2 is most probably a mixture of diamagnetic +1 (5s 2 ) and +3 (5s 0 ) forms, as no experimental evidence of magnetism in reduced indium oxides was detected. The theoretical investigation of hypothetical neutral, molecular In-O clusters with different In/O ratios reveals their high instability in an ionic environment [55]. The HOMO-LUMO gap was found to depend on the metal-to-oxygen ratio in the cluster. Oxidation is likely unfavourable when the In/O ratio is larger than 1, as both vertical and adiabatic electron affinities are negative for In 2 O. In oxygen-deficient In 2 O 3 films, metallic indium may form as a result of oversaturation by cooling down after deposition at elevated temperature [36]. In this case, indium precipitates according to HRTEM and EELS in a form of 5-30 nm nano-particles independently of an indium excess. It is known that intermediate oxides disproportionate in contact with water, resulting in In 2 O 3 and metallic indium [34]. These data confirm that metallic indium readily forms if any lack of oxygen and/or water is provided.
To understand the chemical impact of water during sputtering, let us briefly consider the plasma chemistry of water. Basically, low total pressure and especially plasma excitation change the chemical activity of water. Upon photo-ionisation, water vapour becomes a weakly ionised plasma consisting of electrons and H 2 O + [56]. In the highest state of excitation, the plasma consists of e − , H + , and O( n+ ). In a general case of RF-sputtering from the ceramic target, mostly M + and MO + charged species are observed [57]. In the case of In 2 O 3 , the RF plasma should contain these species in a ratio M + /MO + of more than 30. Since this ratio depends on the M-O binding energy, we took the value known for Fe 2 [58,59]. When argon is used as a sputtering gas, almost no O + , but mostly neutral oxygen is observed [58,60]. The RF-plasma above the ZnO target has a similar content; however, oxygen species generated during DC sputtering of ZnO are O − , O 2 − , and O. The content of negatively charged oxygen species increases exponentially with the reduction of the total pressure [61]. The main difference between the RF and the DC process lies in the concentration of electrons, which is much higher in the first case. Thus, in our process, we likely deal with an intermediate oxidation state of indium in the absence of strong oxidants in the plasma, which finally yields metallic indium species in a film.

Water Containing In 2 O 3
Despite a chemical impact as hydroxylation, water stipulates the amorphous state of as-deposited films. It is known that indium (III) hydroxide tends to remain jelly or even forms a colloid in aqueous solutions rather than precipitating in a crystalline form. The main reasons for that are the donor-acceptor interaction, typical for metals having free 3d orbitals, and the hydrogen bonding in hydroxides. As it is known for the most investigated analogue-aluminium hydroxide, such parameters as concentration, temperature and pH determine the hydrolysis, peptisation, aging and, finally, crystallisation [62]. Basically, al least three processes are coupled with water release and formation of many networking chemical bonds: According to the thermal gravimetric analysis, the crystalline In(OH) 3 transforms into In 2 O 3 with water elimination, starting very slowly from T ≥ 200 • C and becomes fast at about 230 • C in an inert 1 bar atmosphere [63]. As per Le Chatelier's principle, the dissociation in vacuum likely proceeds at lower temperature.
It is known that gaseous hydrogen can also be successfully applied as a hydrogenation agent yielding high-mobility In 2 O 3 films [64]. In this case, the films were obtained by RF sputtering in an amorphous state as well and crystallised by post-deposition annealing. Thorough investigation of oxygen and hydrogen desorption from the In 2 O 3 powders with different surface areas serves us with the following observations [10]. Surface hydrogen starts to desorb in a high vacuum (p = 5 × 10 −7 mbar) already at temperatures somewhat below 100 • C. Desorption of stronger bound hydrogen starts at~150 • C. Water (6 mbar in 1 bar He) becomes an active re-oxidation agent at temperatures higher than about 250 • C, whereas dry oxygen (1 bar) actively re-oxidises the surface starting from >150 • C.
Thus we realise that hydroxylation of In 2 O 3 is most likely the reason for the amorphous state of as-deposited films. This effect can be achieved via sputtering in the presence of either hydrogen or water. Hydrogen acts even more reproducibly [64], since it probably delivers just a necessary hydroxylation without any water excess. However, we still need to understand the desorption of chemically different hydrogen. Additionally, the effect of In 2 O 3 reduction in hydrogen on its electrical conductivity should be considered.

Doping and Conductivity of In 2 O 3
It is known that oxygen deficiency in In 2 O 3 causes higher conductivity [10,46,65]. According to the impedance measurements performed on In 2 O 3 polycrystalline samples, their reduction in dry hydrogen results in a slow resistance decline, starting already at room temperature. The resistance falls sharply at a temperature somewhat below 100 • C and further decreases much slower up to its minimum at about 250 • C [10]. In the presence of water, the same dependency is observed, but the temperatures are about 50 • C higher. This change is reversible and matches the hydrogen adsorption/desorption data; however, the resistance of the re-oxidised samples was found to be 4-5 orders of magnitude lower than the initial one. It might point to the inter-grain changes, e.g., In 2 O 3 reduction, which is then encapsulated by the fully oxidised shell.
We already showed above that the concentration of electrons and hence electron mobility in In 2 O 3 films were often found to be determined by the size of crystallites [23]. In the literature this effect is attributed to the so-called unintentional doping, which is supposedly caused by the inter-grain diffusion of water from ambient air [66,67]. The mechanism of such doping, however, remains questionable for us.
Theoretical studies of this matter demonstrate quite discrepant conclusions. Some basic description of the defect chemistry in In 2 O 3 was done, using solid state chemistry [68]. However, most modern investigations being aimed to justify which point defects exist in the material are performed using density functional theory (DFT). Thus, according to J. Liu, who used the GGA + U formalism, the most stable point defects in In 2 O 3 crystals are oxygen vacancies of the anti-Frenkel [69]. According to the LDA and LDA+U functional calculations, the formation energy of V In was found to be very low in n-type In 2 O 3 [70].  [72]. These defects may only coexist with V O (no charge was noticed in the original work), which facilitate the emergence of indium donors as shallow states.
Considering the penetration of hydrogen into indium oxide, we have to take into account the classical approach of experimentally obtained ionic radii. Oxygen ions (O 2− ) with tetrahedral coordination, like in the In 2 O 3 bixbyte structure, have an effective ionic radius of 1.38 Å [73]. The OH − group would have an even smaller (1.35 Å) ionic radius in this coordination, since the proton is actually a pristine positive nucleus, which is drawn in to the negative electron shell of oxygen, thus making the Coulomb repulsion between neighbouring oxygen ions smaller. On the basis of this simple consideration it is hard to imagine that the hydrogen proton or even the neutral H atom, having a Bohr radius a 0 ≈ 0.53 Å, can replace oxygen in its site. Thus we recognise that the hydroxyl (OH − ) O • is the most probable hydrogen containing species in In 2 O 3 . It is known, however, that In +3 also shows the atomic absorption of hydrogen [74]. Hydrogen was also found to be readily adsorbed by an indium rich InP surface [75]. A large thermodynamic driving force for the neutral covalent binding between hydrogen and solid state indium dimers was identified. The existence of interstitial indium ions has also some restrictions. Thus, In +3 in octahedral coordination possesses, an ionic radius of 0.8 Å is rather large to squeeze into the cavities of the bixbyite lattice. A lower indium oxidation state, e.g., +1 (In i • ), corresponds to an even larger ionic radius. The intermediate oxidation states are furthermore electrochemically unstable (see above). However, ferromagnetism was observed in oxygen deficient InO x films annealed in UHV at 600 • C [65]. This effect was found to be accompanied by the In-In clustering and formation of highly defective glassy regions in crystalline In 2 O 3 [65]. According to Preissler and Bierwagen, the existence of doubly ionized donors best describes the ionized impurity scattering in unintentionally doped In 2 O 3 [23].
Attributing this circumstance to the existence of an indium excess, it is not unlikely to suggest such a defect as In O •• , which is the non-oxidised indium at the oxygen site. It means that we might basically have In In × − In O •• − In In × clusters with an effective In +2 oxidation state.
As for oxygen vacancies, the main disagreement in literature concerns their energy level, which represents either deep [76] or shallow states [23,67]. A practical way to discover which point defects provide conductivity in oxide materials is to measure the conductivity or better the N e dependence on p(O 2 ). The main restriction on that is the requirement of an equilibrium, which for metal oxides means quite high temperatures, far beyond the typical 200-250 • C for In 2 O 3 :H. So the measurements performed at 800 • C discovered the σ ∝ p(O 2 ) 1/10 dependence for In 2 O 3 . Authors attributed this dependence to the (In i ••• -O i ) • cluster formation [77]. In other work Hall measurements are presented for In 2 O 3 films obtained at different p(O 2 ) by RF sputtering without intentional heating [78]. Conductivity was found to be rather constant (~3 × 10 3 Ω −1 cm −1 ) at low oxygen partial pressure (<8 × 10 −4 Pa). When p(O 2 ) increases, conductivity sharply drops over some orders of magnitude that is mostly caused by a decrease of N e from~10 −19 to 10 −16 cm −3 (at p(O 2 ) ≈ 10 −3 Pa). The authors suggested oxygen vacancies as the major donor defects. In this case the N e should have revealed the slope ∝ p(O 2 ) −1/6 for the very deficient oxide and ∝ p(O 2 ) −1/4 for the almost stoichiometric one. From the data presented in this paper one can derive a N e ∝ p(O 2 ) −9 correlation, which cannot be explained by the defect chemistry. This can be rather easier attributed to the presence of a metallic indium phase. There is another important point supporting this hypothesis. The work function, measured for metallic indium, varies in the range from~3.9 to~4 eV, depending on temperature [79]. This value is very close to the one measured for In 2 O 3 :H 2 O and slightly smaller as compared to the one measured for In 2 O 3 :H films (see Figure 7). This means that free electrons can easily be injected from the In 0 outer shell into the conduction band of both oxides. We would like to point also at the very interesting effect of photo-induced change in reactively DC-sputtered amorphous In 2 O 3 films: an exposure of ≤100 nm thick films to UV light (hν ≥ 3.0 eV) resulted in a stable increase of conductivity by × 10 8 reaching σ ≥ 10 3 Ω −1 cm −1 [80]. Simultaneously, the absorption coefficient increases by up to a factor of 10 3 for hv < 1.5 eV and the absorption edge shifts by +0.1 eV. A Drude approximation of the optical absorption in the near IR region gives N e = 1.5 × 10 20 cm −3 that agrees with the Hall data. These data actually represent the pure effect of In 2 O 3 reduction without hydrogenation/hydroxylation impact. They reproduce to some extent our results; however, the Burstein-Moss shift observed in presence of hydrogen is about 0.1 larger.

High-Mobility of In 2 O 3 :H
After T. Koida, the high mobility In 2 O 3 is widely accepted to be doped by hydrogen. He also stated that the doubly ionised impurities were exchanged by singly ionised ones during the annealing process that results in about twofold reduction of N e [44]. The in-situ Hall measurements performed by H. F. Wardenga et al. during annealing of as-deposited In 2 O 3 :H films in vacuum have allowed underlining the following stages [9]. The first stage elapsing at about 160 • C is accompanied with a slight decrease of µ e , which occurs, as supposed, due to the phonon scattering being expected for the degenerated semiconductors. Within this stage N e remains settled. During further heating from 160 • C up to 200 • C, the N e increases and µ e remains unchanged. At T > 250 • C, N e starts declining fast and a strong increase of µ e takes place. The authors suggest that the driving force of the rising N e is crystallisation followed by the grain growth. In turn, the depletion (decrease of N e ) at grain boundaries is to be the reason of a measurable depletion in a material with small grains. Crystallisation and grain growth are superimposed by the decomposition of In(OH) 3 . According to the authors, the release of oxygen is responsible for the drop in carrier concentration and the grain boundaries are being saturated by hydrogen, closing dangling bonds.
We may not fully agree with this explanation mainly because of the known fact that hydrogen disappears first from the inter-grain space. We believe therefore that the dangling bonds existing at grain boundaries are most probably eliminated by the reaction (4). Moreover, it is known that the undoped single crystalline In 2 O 3 reveals electron mobility exceeding 200 cm 2 /Vs that is restricted by~270 cm 2 /Vs due to the phonon scattering [23]. On the other hand the unintentionally doped polycrystalline samples can demonstrate similarly high µ e as the hydrogen doped ones [44,81]. According to T. Koida the effective mass in In 2 O 3 :H seems to depend on N e mostly, rather than on crystallinity [44]. Many of the experimental data collected for various In 2 O 3 based systems reveal a plateau on µ e = f (N e ) dependency exactly around N e~1 0 20 cm −3 [23]. This phenomenon is also associated with a large spread of mobilities indicating additional scattering due to imperfections in the crystal for the samples with µ e < 130 cm 2 /Vs.

Conclusions
To conclude, we observed that the free charge carriers in both In 2 O 3 and In 2 O 3 :H films can appear due to the presence of In 0 . We suggest that metallic indium is present in as-deposited In 2 O 3 or In 2 O 3 :H 2 O films in a much, up to the atomic level, dispersed state. The presence of water or hydrogen during In 2 O 3 deposition at low temperature secures the amorphous state of the film. Hydroxylation of In 2 O 3 is probably the main reason for that. Crystallisation of such films starts at~160 • C when the excess of indium agglomerates, releasing in a separate nano-crystalline phase due to the melting. Melted indium species may vanish during annealing in two ways: either via evaporation and oxidation by water in UHV or via oxidation by oxygen in air. Thus, the concentration of free electrons in In 2 O 3 matrix is reduced and the near IR transparency increases. Both processes, however, do not provide high mobility. The laterally extended growth of crystallites happens when water is released as a result of the hydroxide → oxide transformation. Growing crystallites interconnect at grain boundaries by the In-O-In bonds. These factors both provide high electron mobility exceeding 100cm 2 /Vs. According to our experimental observation, annealing in air demands lower temperature (~180 • C) to provide high mobility as compared to the annealing in UHV (>220 • C). We attribute this to the higher water content in the former case. Crystallisation of the In 2 O 3 :H 2 O system is accompanied with the doping of In 2 O 3 . We expect that the "unintentional" doping differs from the "hydrogen" doping as follows.
In the first case a spontaneous injection of free charge carriers from the dispersed In 0 metallic species concentrated in the inter-grain defect-rich spaces takes place. In the second case, we likely deal with oxidation of the intra-grain In 0 defects trapped during crystallisation by Schottky vacancies: Reactions (5) and (6) describe oxidation by oxygen and water, respectively. High temperature makes water an oxidizing agent, whereas low pressure facilitates hydrogen removal. According to our SIMS results, gaseous hydrogen is removed from the film mostly from the top~50 nm layer. Metallic indium can also accumulate hydrogen in the bulk of the film. As we saw, OH O • defects most probably also exist in In 2 O 3 :H films but their formation in the absence of In 0 is not associated with any redox reaction, so in that case they do not donate electrons.
Supplementary Materials: The following are available online at http://www.mdpi.com/1996-1944/12/2/266/s1, Figure S1: SEM cross-section of the RF sputtered (p tot = 0.5 Pa) 500 nm In 2 O 3 :H 2 O film obtained on Si-substrate, Figure S2: XRD patterns for~150 nm In 2 O 3 films deposited (RF-sputtering, p tot = 0.5 Pa) on glass: comparison of crystallisation conditions. Roman numerals correspond to the film states discussed in the text. Diffraction patterns were acquired using detector-scanning at grazing incidence in the out-of-plane (a) and in-plane (b) modes, Figure S3: TEM images obtained using energy filter. The set energy is marked on each image, Figure S4