Insight into the degradation mechanisms of Atomic Layer Deposited TiO2 as photoanode protective layer.

Around 100 nm thick TiO2 layers deposited by atomic layer deposition (ALD) have been investigated as anticorrosion protective films for silicon based photoanodes decorated with 5 nm NiFe catalyst in highly alkaline electrolyte. Completely amorphous layers presented high resistivity, meanwhile the ones synthetized at 300ºC, having a fully anatase crystalline TiO2 structure, introduced insignificant resistance, showing direct correlation between crystallization degree and electrical conductivity. The conductivity through crystalline TiO2 layers has been found not to be homogeneous, presenting preferential conduction paths attributed to grain boundaries and defects within the crystalline structure. A correlation between the c-AFM measurements and grain interstitials can be seen, supported by HRTEM cross sectional images presenting defective regions in crystalline TiO2 grains. It was found that the conduction mechanism goes through the injection of electrons coming from water oxidation from the electrocatalyst into the TiO2 conduction band. Then, electrons are transported to the Si/SiOx/TiO2 interface where electrons recombine with holes given by the p+n-Si junction. No evidences of intra band gap states in TiO2 responsible of conductivity have been detected. Stability measurements of fully crystalline samples over 480 h in anodic polarization show a continuous current decay. Electrochemical impedance spectroscopy allows to identify that the main cause of deactivation is associated to the loss of TiO2 electrical conductivity, corresponding to a self-passivation mechanism. This is proposed to reflect the effect of OH- ions diffusing in the TiO2 structure in anodic conditions by the electric field. This fact proves that a modification takes place in the defective zone of the layer, blocking the ability to transfer electrical charge through the layer. According to this mechanism, a regeneration of the degradation process is demonstrated possible based on ultraviolet illumination, which contributes to change the occupancy of TiO2 electronic states and to recover the defective zones conductivity. These findings confirm the connection between the structural properties of the ALD-deposited polycrystalline layer and the degradation mechanisms, and thus highlight main concerns towards fabricating long-lasting metal oxide protective layers for frontal illuminated photoelectrodes.


INTRODUCTION
Our society faces a significant challenge regarding energy production and climate change, where renewable energy sources are called to be crucial. However, the high variability on renewables such as wind and solar power is one of the main drawbacks nowadays. Storing solar energy into chemical bonds is an interesting candidate to face this variability, the so called "solar fuels". [1][2][3][4] Among them, hydrogen produced from water electrolysis is a promising candidate as energy vector, substituting natural gas as heat source 5 or fossil fuels used for transportation. 6 leaky layer"), in alignment with the silicon valence band, following the same mechanism postulated decades ago by Campet et al. 59 Other authors that claim not identifying this mid bandgap levels in their samples have proposed a mechanism based on a MIM-like (Metal-Insulator-Metal) structure defined by a TiO2 layer between the photoabsorber and the electrocatalyst. Thus, Mei et al. 56 proposed that electrons were conducted through the conduction band for a p+-Si/Ti/TiO2/Pt structure. So, under this model, electrons from the oxygen evolution reactions through the electrocatalyst are easily injected to the conduction band of the titanium oxide layer and transferred to the internal Si/TiO2 interface at the photoanode where they recombine with the photogenerated holes. However, these works generally do not consider the polycrystalline nature of the TiO2 layer and neither their degradation mechanisms, which they are often overlooked.
In this context, this work aims at revealing the role of the amorphous / polycrystalline nature of the atomically deposited TiO2 layer, considering its defective zones, to determine the electrical conductivity pathways and their evolution over time as a consequence of the degradation mechanisms that take place, which determines the endurance to the use of the protective layers based on TiO2 in silicon photoanodes for water splitting.

EXPERIMENTAL SECTION
TiO2 coatings have been grown by ALD on p + -n silicon buried junctions and simultaneously on p + degenerately doped silicon, to simulate direct injection of carriers in dark conditions. p + -Si samples were created by cutting a degenerately doped silicon wafer (0.001 Ω·cm) in 1x1 cm 2 pieces, and 50 nm of Al were thermally evaporated as back contact.
For the p + n-Si samples, a 1 cm 2 active area was lithographically defined by SiO2 passivation on a silicon n-type wafer (0.1-0.5 Ω.cm resistivity). Boron was implanted in the defined front surface and activated by rapid thermal annealing, creating a 200 nm n + region on top of the n-type substrate. As back contact, 1 µm of Al/0.5%Cu was sputtered on top of 30 nm of Ti to form a proper ohmic contact. p + -Si and p + n-Si samples were sonicated for 5 min in a 1:1:1 isopropanol, acetone and DI water cleaning solution, followed by abundant rinsing and further 5 min sonication in DI water. Before ALD deposition, front surface was cleaned with 0.1M HF for 5 min, rinsed in DI water and immediately introduced in the ALD chamber.
A R200 Picosun Atomic Layer Deposition system was used to grow TiO2 layers. TiCl4 was selected as precursor because of its wide deposition window and lower cost compared with other metalorganic compounds. Thus, TiCl4 and H2O precursors at 19 ºC were used in successive pulses at 8 mbar in N2 flow atmosphere, with 0.1 s pulses and 10 s purges. Under these conditions, layers were grown at deposition temperatures of 100, 150 and 300 ºC for 3300 cycles, corresponding to roughly 100 nm layers. Layer thickness was measured by evaluating the reflected spectra with a Sensofar interferometer device with ± 0.2 nm error. Finally, nickel-iron (NiFe) was deposited by thermal evaporation using an Oerlikon Univex 250 equipment, with a thickness of 5 nm measured by a quartz microbalance. Samples were then soldered to a Cu wire using Ag paint and epoxy protected leaving the front area exposed ( Figure S1).

Surface morphology was observed with a Zeiss Series Auriga Field Emission Scanning Electron
Microscope (FESEM). Structural characterization was carried out by X-ray diffraction (XRD) in a D8 Advance Bruker equipment with a Cu Kα radiation source working at 40 kV and 40 mA with a 3 º offset angle. Raman spectroscopy was performed using a Horiba LabRam HR 800 with 532 nm laser light with a 100x objective and an output power of 5 mW, 10 s acquisition time and 10 accumulations. High resolution transmission electron microscopy (HRTEM), high angle annular dark field (HAADF) scanning TEM (STEM) and electron energy loss spectroscopy (EELS) spectrum imaging (S.I.) were performed using a TECNAI F20 operated at 200 kV with a point to point resolution of 0.19 nm, with a coupled GATAN QUANTUM EELS detector.
HRTEM samples were prepared using a focused ion beam (FIB) to select the desired region to image, previous protecting the surface with deposited Pt. AFM and conductivity AFM (c-AFM) measures were taken with a Park Systems XE-100 with platinum conductive cantilevers biasing the sample at +1 V. Due to the p + n-Si built-in voltage, only the samples on p + -Si substrates were measured by c-AFM. The photoelectrochemical measurements (cyclic voltammetries, chronoamperometries and electrochemical impedance spectroscopy) were obtained with a Biologic VMP-300 potentiostat using Ag/AgCl/KCl (3M) (E 0 = 0.203 VRHE) as reference electrode and platinum mesh as counter electrode. A quartz cell with flat faces was used with 100 ml of 1M KOH electrolyte and a 300 W xenon lamp with an AM 1.5G filter adjusted to 100 mW/cm 2 , calibrated using a silicon diode (Gentec-EO, XLPF12-3S-H2-DO). 20 mW·cm -2 365 nm UV light is obtained from a 200 W Hg-Xe Hamamatsu LC8 light source lamp. I-V curves were obtained in two-electrode configuration depositing 50 nm of Au as top contact.

Morphological and structural characterization
In this work, we have fabricated silicon photoanodes protected with TiO2 layers prepared by Atomic Layer Deposition (ALD) in a temperature range from 100 to 300 ºC from TiCl4 precursor, which permits a wide temperature deposition window. 60 Figure 1 shows SEM surface images of the ALD TiO2 coatings synthetized at the different temperatures. For the film obtained at 100ºC (Figure 1a), a homogeneous coating is observed with no presence of grains, whereas using deposition temperatures of 150 ºC some regions with different contrast appear, corresponding to grains embedded in an amorphous matrix (Figure 1b).
Further increasing deposition temperature up to 300 ºC (Figure 1c), yields to a highly crystalline surface with smaller grain sizes. These grains are related to the TiO2 anatase phase, as can be seen in the XRD diffraction pattern for the sample synthetized at 300ºC (Figure 1d). The TiO2 anatase phase is more favourable to OHadsorption and thus causing higher ALD growth rate. 22,61 No anatase XRD peaks or other crystalline phases are detected at 100 or 150 ºC, as previously observed in literature. 53,62  As it can be seen in the SEM image for the sample synthetized at 150ºC ( Figure 1b) the density of embedded crystals is rather low, and therefore XRD measurement does not allow the identification of the crystalline structure, being below the resolution. We used Raman spectroscopy equipped with an optical microscope, in order to carry out a more localised analysis. Optical view of the Si chip with an ALD-TiO2 layer grown at 150 ºC is shown in Figure   2a, were it is seen the presence of dark spots embedded in a homogenous grey film, which may be attributed to the submicron size crystals identified by SEM. 63 Figure 2b presents the Raman spectra of two different spots, with a clear appearance of an intense peak in the darker spot, corresponding to the Eg mode of anatase located at 141.4 cm -1 . The peak is red-shifted with respect to the expected 144 cm -1 value, 64 being a sign of tensile stress, probably caused by partial amorphousness of the interface and reduced crystalline size. 65 A Raman mapping scan of the Eg main peak is shown in the inset of Figure 2a, which allows us to correlate the dark spots seen in the optical image with anatase crystallographic order presence.
Complete analysis by micro Raman spectroscopy of TiO2 layers grown at 100, 150 and 300 ºC is shown in Figure 2c. None of the spots analyzed of the sample synthetized at 100ºC present the anatase peak, whereas is present at any spot at 300 ºC samples, confirming the existence of a fully polycrystalline layer, in accordance with the XRD results and SEM. A deeper analysis of the Raman spectra shows a more intense peak for 300 ºC layers compared with 150ºC crystalline zones, and a blue shift of the Eg mode (of 10.3 cm -1 ), that can be attributed to either surface pressure or phonon confinement effect that usually exist in nanometer-sized materials. 66 As seen in SEM images (Fig. 1c), nucleation is enhanced with increased deposition temperature and smaller grains are obtained by competitive growth.

Electrochemical characterization
In Figure 3a, cyclic voltammetries of three illuminated p + n-Si homojunctions are shown. The photoanodes, previously cleaned with HF (to eliminate the SiOx) and immediately ALDprotected with TiO2 grown at 100, 150 and 300 ºC were then coated with a 5 nm film of thermally evaporated NiFe, to act as OER catalyst. For samples grown at 100 ºC, no photocurrent can be observed. However, for samples deposited at 150ºC and 300ºC the same onset potential is obtained (around 0.96 V vs RHE) and whereas significant photocurrent is obtained for the former, highly increased currents are observed for the later. This fact, together with the saturation regime reached at more anodic potentials for the 150 ºC sample, indicate a more resistive layer. The saturation current value, 22.5 mA/cm 2 , is given by the photovoltaic quality of the prepared silicon-based homojunctions that were not optimized, and the lack of adjusted antireflective strategies. 67 In dark conditions, the same tendency observed for photoanodes is obtained in TiO2-protected p + -Si anodes: increasing deposition temperature from 150 to 300 ºC enhances OER current ( Figure 3b). The nickel-iron catalyst is necessary for efficient charge injection into the electrolyte 28, [68][69][70][71] , and if it is not present, no oxidative current is obtained in the studied range of potentials (dotted line in Figure 3b). In fact, Ni(Fe)OOH is the real catalytic phase obtained after cycling in alkaline electrolytes such as standard KOH solutions, one of the earth-abundant OER catalyst with lower overpotentials 38,[72][73][74][75][76][77][78][79][80][81] Figure S3.

Conduction pathways in TiO2-ALD protected photoanodes
Columnar grains are present in both samples synthesized at 150 and 300 ºC, but nucleation is significantly different. The HRTEM image of the sample grown at 150 ºC shows that TiO2 is crystalline only in a few-nanometer dome-like region close to the substrate, while the rest of the TiO2 layer remains amorphous. However, the one prepared at 300 ºC shows that the whole TiO2 layer is crystalline. This can be attributed to the higher deposition temperature used, which enhances nucleation and lateral crystal propagation, together with crystal competition. 41 The TiO2/NiFe interface can be analysed by HRTEM EELS. In those as-prepared samples, the ~5nm apparently homogeneous NiFe catalyst layer can be seen ( Figure S4a). After cycling in 1 M KOH ( Figure S4b), a thicker layer can be observed, and the oxygen presence throughout the whole NiFe layer is increased, indicating oxidation of the metallic layer and OHincorporation, as reported in other works, where a restructuration of the NiFe towards a three dimensional nanoflakes porous layer takes place. 88 Moreover, there is no presence of Ni or Fe incorporation in the TiO2 crystalline structure, or O or Ti concentration gradient can be detected above the EELS system resolution threshold, neither in the amorphous or crystalline structures nor the grain boundaries (Figures S5, S6).
To study the surface chemical composition of the layer, XPS analysis was performed, which is discussed in the supplementary information and Figure S7. To sum up, for as prepared samples, a 2.8:1 Ni:Fe ratio is observed, natively oxidized before anodic cycling and with higher oxygen content afterwards, caused by OHintercalation and oxidation of the coatings. Ni 2p spectra presents Ni(0) metallic peaks before but not after immersion in the electrolyte. For the cycled sample, multiple Ni(II) and Ni(III) peaks can be fitted (Figure 7a) Fermi level, pointing at a higher n-type behaviour for the sample grown at 300 ºC (larger Fermi level to valence band energies). On the other hand, no signal from intra-bandgap states is detected with a fine analysis of the 0-3 eV region in accordance with another works 28 , although the small signal-to-noise ratio could be hindering the detection ( Figure S9). Also, the oxidation of the metallic NiFe layer after immersion in the electrolyte at anodic potentials is revealed by the XPS spectra, still presenting slight free electrons behaviour. The most significant result is the excellent band alignment between TiO2 conduction band and the Ni(Fe)OOH catalyst, that justify the injection of electrons from the oxygen evolution catalyst to the TiO2 conduction band.
The results obtained can be summarized in the mechanism proposed in Scheme 1, basing the conductivity in preferential charge transport pathways. The increased conductivity pathways density present for 300 ºC deposited samples with respect to 150 ºC ones (Figure 4 e-f) must be especially highlighted, in agreement with reported data in Figure 3. In addition, several features can be underlined: 21 1) The Si-TiO2 interface has been found not to introduce significant electrical barrier to charge transport. A recombining contact for holes from the p + -Si valence band and electrons from the TiO2 preferential paths conduction band is expected, facilitated by tunnelling through the ALDcaused SiOx similar to the schemes proposed by other authors. 93 2) The transversal section to charge transport has been proven to be not homogeneous. Lower resistivity regions detected by c-AFM are attributed to more defective crystalline TiO2 regions than bulk crystal grains, forming preferential conductivity paths through the grain boundaries and defects within the crystalline structrure. 84 3) There is a favourable electron transfer from the Ni(Fe)OOH to the TiO2 conduction band. This is facilitated by the permeable to electrolyte and energetically adaptive characteristics of the catalyst due to ionic diffusion, avoiding the energetic barrier formed between TiO2 and the electrolyte and facilitating OER reaction. 69

Endurance test: degradation mechanisms and stability
Crystalline TiO2-protected p + -Si anodes and p + n-Si photoanodes were tested for stability in a 1 M KOH electrolyte at anodic potentials. The obtained current density values show significant decay for both electrodes (Figures 6 and 7). To further analyse these samples, Electrochemical Impedance Spectroscopy (EIS) was applied before and during stability test.
For the p + -Si anode in dark conditions, the current measured for more than 480 h at 1.8 V vs RHE, shows an exponential decay (Figure 6a), which indicates a degradation of the electrode.
The EIS measurement shows two clear semicircles increasing over time (Figure 6b) revealing degradation correlated with two processes in series defined by different RC time constants, which allows us to propose an equivalent electrical circuit shown in Figure 6b. As seen in Figure   S10, the first semicircle (R2/C2) is independent from the applied potential and the second one varies significantly between open circuit potential (OCP) and OER working conditions, proving a non-ohmic dependence of this second one. Our electrode has been modelled to an ohmic resistance, R1 (corresponding to all resistive system contributions given by cables, back contact and substrate and electrolyte conductivity), a R2/C2 parallel circuit in series (corresponding to the deplection region within the TiO2 protective layer) and a Randles circuit (R3/CPE3, corresponding to the electrocatalytic OER charge transfer on Ni(Fe)OOH). Fitting the proposed EIS circuit to our experiments (Table 1), a clear relation is observed between the current loss during stability tests and the semicircles increase. The current diminution rate is initially high, and it is reduced as current values decrease, pointing towards a relation between current passing and conductivity degradation. The resistance R2 corresponding to the current flow through the TiO2 layer increases logarithmically over time at 1.8 V vs RHE, meanwhile R3, the resistance attributed to charge transfer in the OER reaction, increases lineally ( Figure S11). Owing to the fact that the interfacial SiOx layer does not increase in thickness after stability test ( Figure S12), the increase of R2 over time can be related with the intrinsic self-passivating process of metal oxide surfaces, following a logarithmic rate law. 94 This is highlighting an interaction of OHions from the electrolyte and the polycrystalline TiO2 driven by the electric field due to anodic polarization. Some authors have previously reported chemical interaction of OHmigration and defective regions, nulling preferential charge transport paths. 95 The significant decrease of crystalline TiO2 conductivity has not previously been stated, but a deep review of the bibliography does not deny its plausibility. Other works have shown the possibility of OHto penetrate through a permeable Ni catalyst layer and diffuse through the TiO2 82 or ALD-TiO2 dissolution on few hours scale 95 . Oxygen mobility under polarization in TiO2 has been known for years, forming preferential paths in the form of conductive filaments. 96,97 The presence of such oxidant molecules, OH -, inside the TiO2 structure can easily interact with the states responsible of the n-type semiconductor properties and conductivity. 98 This fact has been depicted in Scheme S1a and b.
On the other hand, according to the Tafel equation describing the charge transfer at the electrode/electrolyte interface (eq. 1) 99 , overpotential ( ) depends logarithmically on the current density. Assuming both elements are in series (i2 equals i3, i), and the resistance of the TiO2 layer (R2) increases while applied potential is kept constant, the current reaching the OER reaction (measured in EIS as R3) will decrease logarithmically, giving an inverse logarithmic dependence of R3 with R2 (eq. 2), represented in Figure S13. Following these equations, if 2 ∝ log( ) as it has been demonstrated in Figure S11, and substituting in eq. 2, we obtain 3 ∝ .
In his turn, capacitance of the TiO2 semicircle is not significantly modified as expected, as it corresponds to the thickness and dielectric constant of the TiO2 and these are maintained. Measurements performed under 1 sun illumination, using 1 M KOH as electrolyte.
Similar stability test was performed with a p + -n-Si photoanode with TiO2 deposited at 300 ºC and NiFe decorated. Likewise, after 333 h at 1.3 V vs RHE photocurrent has been reduced significantly ( Figure S14).
To validate the previously described degradation mechanism, superimposed UV illumination was used to modify the density of states within the TiO2electronic structure, achieving a significant recovery of the photocurrent (Figure 7a). This can be explained by the photoconductivity effect on TiO2, 100 where the UV photons create electron-hole pairs changing the occupation of localized states, causing a significant reduction of R2 (Figure 7b, Table S1). 101 Switching off UV light gave a few-seconds decay followed by a several minutes persistent photoconductivity (PPC). These states created by oxygen desorption will eventually be compensated again by OHdiffusing inside the material, as illustrated in Scheme S1. This UVlight effect should only be expected in the near-surface region of TiO2 due to small UV photon penetration depth, precisely where higher OHinteraction would be.

CONCLUSIONS
TiO2 protective layers grown by ALD have been deposited on silicon photoanodes to avoid corrosion, while allowing charge transfer to the electrolyte to perform the oxygen evolution reaction using a NiFe earth abundant catalyst. Layers consisting of approximately 100 nm have been grown at 100, 150 and 300 ºC, corresponding to completely amorphous, partially crystalline and fully crystalline TiO2.
Completely amorphous layers presented high resistivity, meanwhile the ones synthetized at 300ºC, having a fully anatase crystalline TiO2 structure, introduced insignificant resistance, showing direct correlation between crystallization degree and electrical conductivity. The conductivity through crystalline TiO2 layers has been found not to be homogeneous, presenting preferential conduction paths attributed to grain boundaries and defects within the crystalline structure. A correlation between the c-AFM measurements and grain interstitials can be seen, supported by HRTEM cross sectional images presenting defective regions in crystalline TiO2 grains.
A conduction mechanism has been proposed assuming electrons coming from water oxidation are injected into the TiO2 conduction band in preferential regions through the electrocatalyst.