Hydrogen Absorption into Titanium under Cathodic Polarization: An In-Situ Neutron Reflectometry and EIS Study

The sample was a ∼50 nm thick Ti film, sputter-deposited on a Si(111) substrate in an Ar atmosphere. A sputtering target-grade Ti disk (American Elements, PN:TI-M-03-ST.2250, 99.9% Ti), pre-cleaned with Argon plasma, was the source of the metal. Upon removal of the substrate from the sputtering chamber and exposure to air, the native oxide grew on the Ti film. This sample preparation predated the NR experiment by ∼1 month; the oxide layer is expected to have grown to ∼4 nm in thickness by the time that we began the NR experiment.³¹ The cell, designed for performing NR and electrochemical measurements simultaneously on the same sample, has been fully described by Noël et al.¹¹ When the cell was fully assembled and filled with electrolyte, the Ti thin-film became the working electrode. All aspects of the electrochemical measurements were performed via a potentiostat (Solartron 1287) and a frequency response analyzer (Solartron 1255B) controlled by a computer running CorrWare and ZPlot software (Scribner Associates).

NR was performed on the D3 reflectometer at the NRU Reactor, Chalk River, operating with a fixed neutron wavelength of 0.237 nm. Equipped with motorized high-precision slits, the reflectometer performed 2θ/θ scans while maintaining a constant beam footprint on the sample and a fixed step size of 0.01 nm⁻¹ in the momentum transfer vector (Q_z). The incident neutron flux on the sample, B_i, was measured by positioning the detector at 2θ = 0° while varying the slits exactly as in a scan. The Q_z-dependent background was measured by offsetting the sample by 0.3° vs. the incident angle. Dividing the background-subtracted reflected intensity by B_i yielded the reflectivity on the absolute scale. The normalized data thus obtained were analyzed by least-squares fitting a layer profile model using the Parratt32 program.³³

Before initiating the electrochemistry part of the experiment, we performed NR scans on the as-prepared sample, with three different sample configurations. These were mainly done to verify that the reflectometer was functioning properly, but they also served to characterize and align the sample on the instrument. The configurations were as follows:

• (a)  
  Sample clamped to a bracket with the incident (and the reflected) beam on the air-side of the thin-film.
• (b)  
  Sample clamped to the same bracket but with the incident beam on the Si-side (substrate side) of the thin-film.
• (c)  
  Sample mounted on the empty electrochemical cell (i.e., no electrolyte solution, but the cell was otherwise complete) with the incident beam on the Si-side. This geometry is necessary when the sample is mounted on the cell, since neutron access from the air-side is blocked by the cell body.

After the measurements taken with the sample in configuration (c), the cell was filled with a solution of 0.27 M NaCl (neutral pH, argon-deaerated) in heavy water (D₂O). The area of the working electrode exposed to the solution was 62.5 cm². For the entire experiment, the working electrode potential was measured vs. a saturated calomel electrode (SCE). In this paper we exclusively applied potentiostatic polarization, with cathodic currents conventionally expressed as negative numbers.

As soon as the cell was filled with electrolyte solution, we began monitoring the open circuit potential (E_OC) of the working electrode. Subsequently, we applied a potentiostatic polarization and recorded the current through the working electrode during the first two hours, while we waited for the current to achieve a steady state. After that, we maintained the same potentiostatic bias and recorded the impedance spectra.

For electrochemical impedance spectroscopy (EIS) a sinusoidal input potential of 20 mV amplitude, scanned over a frequency range from 100 kHz to 1 mHz, was applied, in addition to a dc bias, while we recorded the complex impedance Z(ω) = Z^'(ω) + jZ^''(ω), at each angular frequency, ω, where j² = −1, and Z^' and Z^'' are the real and imaginary components of the impedance, respectively. The first set of EIS scans was performed with the potential, E, set to 0 V_SCE. The scans, each lasting 1.2 h, were repeated five times to see whether there was any trend in the run-to-run variations in the impedance data. If there was no systematic variation detected, all data collected at a single applied potential were combined (via point-by-point averaging of Z^' and Z^'' acquired at each frequency) for analysis and plotting.

The first set of NR scans in the presence of electrolyte solution was performed while the EIS scans at E = 0 V_SCE were in progress. A total of eight identical NR scans were carried out. The purpose of repeating these NR scans, each performed within ∼1.2 h (i.e., relatively quickly), was again to see if there was a trend in the run-to-run variations in the NR data.

During the remainder of the experiment we probed the sample with EIS and NR as it underwent changes in response to increasingly cathodic polarization. Measurements were performed with E held at −0.25, −0.30, −0.40, −0.45, −0.50, −0.55, −0.60, −0.65, and −0.70 V_SCE, in that sequence. The procedure during each polarization was as follows: we monitored the cell current beginning almost immediately after we set E, and started the NR scan (from high to low Q_z) within a few minutes thereafter. The NR scan was repeated twice, taking a total of 10 h. Simultaneously, the cell current was monitored for the first 2 h and then EIS scans performed repeatedly for the remaining ∼8 h.

Because these experiments demand extensive allocation of precious neutron beam time, all of the measurements reported herein were performed once, on a single Ti thin film sample. We intend to repeat this work on another nominally identical film, using H₂O solution in place of D₂O as both a confirmation of full reproducibility and as a complement to the current measurements, due to the differences in the interactions of the neutrons with H and D.

The reflectivity data obtained with all three sample configurations are of good quality, for instance, normalized reflectivity from configuration (a), shown in Figure 1 (circles), covers more than 6 orders of magnitude with little background at the high-Q_z end. The quality and dynamic range of the data obtained with configurations (b) and (c) are similar to those obtained in configuration (a).

Zoom In Zoom Out Reset image size

Open-circuit potential (E_OC) monitoring began immediately after we filled the cell. The E_OC initially read by the potentiostat was 0.035 V_SCE, but drifted to 0.005 V_SCE over the course of a few hours. Routinely observed, drifts in E_OC are usually due to slow changes, either in the electrolyte or on the surface of the working electrode. After allowing the working electrode to be exposed to the solution for several hours at open circuit, we polarized it at a series of potentiostatic values while completing the rest of the experiment. Although the initial potential applied (0 V_SCE) was slightly <E_OC at the time it was applied (0.005 V_SCE), ongoing minor changes resulted in small anodic currents being observed until the potentiostatically applied polarization was stepped to a sufficiently negative value that the current became cathodic.

The results of EIS scans, performed once the working electrode achieved a steady state, agreed with each other within experimental error, indicating that changes to the sample, if any, were too small or slow to be observed on the time scale of hours. An example of the combined EIS data collected at E = 0 V_SCE is shown in Figure 2, along with the fitted impedance curve, to demonstrate the quality of the fit to the data. The equivalent circuit model that was used to fit the data is included as an inset within Figure 2.

Zoom In Zoom Out Reset image size

Since no systematic variation was found in the repeated NR scans under the same conditions, all NR data were combined and averaged, leading to the plot shown in Figure 3. Due to the superfluous neutron scattering from D₂O, the background in the high-Q_z region was significant and the dynamic range of the data reduced (compare to Figure 1).

Zoom In Zoom Out Reset image size

As shown in Figure 4 by the constancy of the current at each applied potential, the 2 h wait was long enough to stabilize the changes in the sample in all cases except at −0.65 V_SCE and probably −0.60 V_SCE. The observation that the impedance measured by all repeated EIS scans is constant within experimental error is further evidence of stability. Also, the reflectivities observed by the two NR scans at each potential are similar (within errors) suggesting that most of the changes to the sample took place during the time it took to record the background region (Q_z > 0.1 Å⁻¹ in Figure 3) of the first scan.

Zoom In Zoom Out Reset image size

Figure 4 shows current transients recorded within the first 2 h after we set the polarization E. As expected, decreasing the potential caused the cathodic current to increase momentarily, spiking to a more negative value, and then settling over time (typically within 10 minutes) to a new steady state. Remarkable deviation from the general trend occurred at −0.65 V_SCE. Not only was the transient cathodic current much higher, but it also took a long time to stabilize: indeed, the cathodic current was still decreasing after 2 h. Closer inspection of the current transient data revealed that this behavioral change started at −0.60 V_SCE. Approximately 1 h after we set E at −0.60 V_SCE, the cell current started to rise (i.e., became more cathodic in Figure 4), and thereafter showed fluctuations larger than seen at all previous potentials. The consequences of this change are visible in the EIS results (Figure 5). The data at E = −0.55 V_SCE (panel a) show a spread that is relatively small and random in sequence, whereas those at −0.60 V_SCE (panel b) arc significantly more and exhibit a systematic trend. In fact, the trend to higher impedance over time at −0.60 V_SCE indicates a reversal in the initial trend to higher cathodic currents observed during the first 2 h of the experiment (Figure 4). This change was nearly complete by the time EIS was performed at −0.65 V_SCE (not shown), all the Nyquist plots (except the first) arcing less than those at −0.60 V_SCE with a small and random distribution. Figure 6 is a summary of all the EIS results.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Physical model for the NR data

Careful analysis of the NR data on the as-prepared sample (Figure 1) in terms of a model serves two purposes: 1) it characterizes the sample in its original state, and 2) it gives a measure of the uncertainty in the parameters obtained using the least-squares fitting procedure.

Ideally, a single model should fit all three sets of data acquired on the as-prepared sample in the three configurations employed. The model we chose after proposing and investigating numerous possibilities consists of five distinct media representing air, TiO₂, Ti metal, SiO₂ and Si substrate. Each medium is represented by a plateau, a region of constant scattering-length density (SLD or ρb, where ρ is the number density of the nuclei present and b their coherent neutron scattering length, a characteristic property of each element and isotope). The thickness, d, is infinite for the first and the last media but finite for the central media (e.g., d(Ti) for the Ti layer). The interface between two media is represented by a hyperbolic tangent function of width parameter σ. Our model is consistent with the axiom of "simplest possible," as no feature was included unless it led to a significantly better fit to the measured reflectivity.

The Parratt32 program³³ used for model refinement can fit only to a single data set at a time (with no option for coupled refinement). Consequently, independent refinements were performed on the reflectivity data for the three different configurations of the as-prepared sample, and the refined parameters compared. The parameters and values for the as-prepared sample in configuration (a) are listed in Table I. The table shows the best fit values for the layer thicknesses (d), the real and imaginary components of the scattering length density (Re{ρb} and Im{ρb}, respectively), and the interfacial roughness parameter, (σ). The calculated reflectivity corresponding to the model is shown in Figure 1a, and the layer profile of the model is given in Figure 1b. The agreement between observed and calculated reflectivity is excellent in all cases.

Figure 7 is a plot of the models representing the as-prepared sample in the three measurement configurations on a common scale (the profiles for sample configurations (b) and (c) were reversed and the origins shifted to match that of configuration (a)). The agreement is good, except at the TiO₂/air interface, which is much broader in configuration (c). In all cases, Ti is the only layer thick enough to exhibit a plateau in SLD, whereas the SiO₂ and TiO₂ layers are so thin that the interfaces on either side of the layer merge and no plateau is visible. The lower peak for TiO₂ in configuration (c) is a consequence of the merging of the two closely spaced interfaces. The magnitude of ρb observed for Ti is significantly smaller (i.e., a less negative number) than that of bulk Ti (−1.95 × 10⁻⁴ nm⁻²). This is common for sputter-deposited films, and is most probably due to the formation of vacancies or incorporation of impurities, such as oxygen, during sputtering.³² The peak ρb values for the SiO₂ and TiO₂ layers (discounting the empty cell case, configuration (c), for the latter) are surprisingly close to the bulk phase values, 3.62 × 10⁻⁴ nm⁻² and 2.63 × 10⁻⁴ nm⁻², respectively, considering that these layers are too thin even to show a SLD plateau.

Zoom In Zoom Out Reset image size

The change seen for the TiO₂/air interface in configuration (c), while suggestive of either mechanical deformation of the sample during mounting on the cell or a compositional change, is actually an artifact. Sample deformation would affect all of the layers within the sample, not just the outermost, which was not the case, and compositional change is unlikely since the sample was not exposed to any different chemical environment. Instead, the change of the interface width and the TiO₂ SLD is an erroneous result, probably due to incomplete handling of the increased background signal (not shown) that was observed in this configuration. However, this deficiency became almost irrelevant when the cell was filled with D₂O solution, Figure 8, as discussed below.

Zoom In Zoom Out Reset image size

The same model, but with the air (SLD = 0) replaced with D₂O (SLD = 6.37 × 10⁻⁴ nm⁻²) gives excellent fit to all the reflectivity measured subsequent to filling the cell. As noted above (Figure 3) the increase of background caused by D₂O is very large. Nevertheless, thickness and SLD values of all the layers refined to sensible values (see Table II). In Figure 8, the black curve is the model used to fit the reflectivity data collected in configuration (c), while the red curve is the model obtained after filling the cell (at E = 0 V_SCE). The rise in SLD from Ti (negative) to D₂O (highly positive) is so large that one broad hyperbolic tangent function appears to mask the whole Ti/TiO₂/D₂O interface region (from position ∼60 to 70 nm). As the reflectometry model is not critically sensitive to the TiO₂/D₂O interface width, the sample's outer surface did not influence any of the subsequent analyzes. For the same reason, we lost the ability to follow with NR the changes occurring in the TiO₂ layer as the cathodic polarization was applied.

Since we recorded multiple NR data sets on the same as-prepared sample (Figure 7), differences between least-squares fitted thickness, SLD, and interface width values allow estimation of the uncertainties associated with our analysis. Since parameters associated with well-resolved features should be more reliable, those associated with the Ti layer should show a higher degree of consistency. All the Ti layer thickness values, d(Ti), determined for the as-prepared sample agree to within ±0.2 nm, while the spread in SLD, ρb(Ti), is only ±2 × 10⁻⁶ nm⁻². Similar consistencies are seen for the SiO₂ layer, both in the thickness and in the SLD, but the spread in d(TiO₂) is significantly larger, suggesting an uncertainty of ±0.4 nm. Values of the interfacial width parameter, σ, differ significantly, which is not surprising since σ values are generally more difficult to determine accurately. This prevents us from drawing conclusions that depend critically on interfacial widths.

The errors above are the upper limit for uncertainties, since the sample was physically removed and remounted in a different geometry between the three measurements. However, during in-situ NR and EIS measurements the sample was not physically disturbed. Consequently the precision (but not necessarily the accuracy) of results from successive NR scans should be significantly higher.

The data from two NR scans at −0.25 V_SCE, and others at more negative potentials, were analyzed using the same 5-layer model. The fitted results from the two scans agreed very well, suggesting that steady-state was achieved promptly after we set the potential. The parameters presented here are exclusively from the second of the two NR scans.

EIS analysis

A unique feature of neutron reflectometry is that the experimentalist can sometimes control, at least in part, the sensitivity or contrast between different phases within a sample by changing their SLD, denoted ρb, where ρ is the number density (atoms per unit volume) and b the coherent neutron scattering-length of the isotope. For systems consisting of solids and liquids we usually have no means of affecting the density, ρ, but we can sometimes change the scattering length, b, by choosing the isotopes of the elements involved. The coherent neutron scattering lengths, b, for Ti and H are negative (respectively−0.344 fm and −0.374 fm), while that for D is positive, with a magnitude (+0.667 fm) approximately twice that of b(H). Since the SLD of a phase containing more than one type of nucleus is the weighted average of the ρb values for each type of nucleus present, a small amount of H, dissolved into Ti without causing the metal to swell significantly, will shift the overall ρb of the metal layer in the negative direction. On the other hand, D in Ti will cause the overall ρb to shift in the positive direction and the magnitude of the shift will be twice as large as that observed for an equal amount of H. This is why we used D₂O instead of H₂O as the electrolyte. However, this advantage comes at a cost; as seen in Figure 8, filling the cell made the SLD profile of the TiO₂ layer and the adjacent interfaces all blend into a single step. This made the parameters for the TiO₂ layer that were measured by NR after the cell was filled with heavy water untrustworthy, even though they may seem reasonable. This is not the case when the cell is filled with H₂O.³²

In contrast to NR, EIS results are isotope-independent, since even at the highest frequency measured, 100 kHz, molecular vibrational or rotational responses are undetectable. Given that the Ti layer is metallic, the resistance and capacitance probed by EIS are exclusively from the TiO₂ layer, the oxide/solution interface, and the electrolyte solution; i.e., the regions NR cannot "see". Thus, NR and EIS play complementary roles in our experiment.

The EIS data presented in Figure 6 give the overall picture of the stages the sample went through as the potential was decreased. The data fall into four groups as demarcated by the dashed boundary lines in the figure:

• Group 1: Data collected at E = 0, −0.25, and −0.30 V_SCE: These data lie above Boundary 1. No EIS measurement was made at −0.35 V_SCE, but the cell current (Figure 4) suggests that −0.35 V_SCE belongs to this group.
• Group 2: Data collected at E = −0.40, −0.45, −0.50, −0.55 V_SCE: These data lie below Boundary 1 but not far from it.
• Group 3: Not really a group since it consists of only one data set, E = −0.60 V_SCE, but treated on its own since it is unique. Having a uniquely small low-frequency impedance, these data lie below Boundary 2.
• Group 4: Data collected at E = −0.65 V_SCE and −0.70 V_SCE: The impedance measured at low frequencies increased over that seen at E = −0.60 V_SCE, but remained lower than the low-frequency impedance at less cathodic polarizations. All of these data lie above Boundary 3 but not very far from it. The cell current was high during the first 2 h at E = −0.65 V_SCE (Figure 4), but by the time of the last EIS scan the current had settled down to a lower value (as signaled by the increased impedance at the low-frequency end of the EIS scan).

The results in Figure 6, analyzed in terms of the one time-constant equivalent circuit, lead to Figure 9. As Figure 9 was derived from the data in Figure 6, it is not surprising that it shows the same groupings; however, because it emphasizes different aspects of the data than do the Nyquist plots of Figure 6, Figure 9 makes the behavioral differences that led to the group assignments much more clear. Further support for the assignments of Groups 3 and 4 is evident in the different time trends in the currents, visible in Figure 4.

Zoom In Zoom Out Reset image size

At small cathodic polarizations (up to about −0.35 V_SCE) the cell was characterized by high resistance and low capacitance, as represented by the CPE parameter, Q (Figure 9). A notable decrease of resistance and increase of Q took place between −0.3 V_SCE and −0.4 V_SCE. At −0.6 V_SCE, the resistance fell to a sharp minimum, while Q experienced an inflection in the trend to that point. At E = −0.65 V_SCE, the resistance increased a little and Q resumed an increasing trend. We can identify two potentials where significant changes took place, the first critical potential somewhere between −0.35 and −0.40 V_SCE, and another at approximately −0.60 V_SCE.

The use of a constant phase element rather than a capacitor in the equivalent circuit was necessitated by the non-ideal capacitive behavior of the electrode surface, as is commonly observed in electrochemical systems. The impedance of the constant phase element can be expressed by the equation,

where Q is the CPE parameter, ω the angular frequency, and α the CPE exponent. In this case, the CPE exponent, α, was found to be 0.958 ± 0.005. Given that in our system the observed capacitive behavior arises from the space charge layer developed in and on the very thin, flat titanium oxide layer on the sample surface, the non-idealities that necessitate the use of the CPE are most probably edge effects and oxide film defects distributed laterally across the sample surface.³⁴ It is often possible to derive an equivalent capacitance from the CPE parameter using either the two-dimensional method of Brug et al.³⁵ or the one-dimensional method of Hsu and Mansfeld³⁶ to convert the measured CPE parameter into an equivalent capacitance value C_eff; however, neither method can be properly applied to the data acquired in these experiments for reasons described below.

Our analysis was performed using the simplest circuit model that yields a good fit to the data. At potentials at which no water reduction occurs, one might expect the CPE element to correspond to the capacitance of the oxide film, and the parallel resistive element to the surface oxide film resistance. Likewise, one can easily infer that once the water reduction reaction begins there should be, in series with the oxide film resistance and capacitance, a second parallel R-CPE pair in the equivalent circuit to account for the capacitance of the electrical double layer and the kinetics of the water reduction reaction. In these experiments, however, the two R-CPE pairs cannot be observed as distinct from one another. This is not an experimental measurement quality issue, but rather a consequence of the relative values of the equivalent circuit element parameters; both the circuit employed here and a circuit with a second R-CPE pair in series can generate impedance responses that are nearly equivalent to each other and to the measured data, depending on the relative values of the circuit elements. Without independent data to help reveal the values any of the separate R or CPE parameters we cannot justify modeling the EIS results with any but the simplest equivalent circuit. Therefore the one R-CPE circuit has been used for modeling the EIS data, with the acknowledgement that the derived element values may correspond to combined parameters. The leakage resistance has been termed R_p for "polarization resistance" and the CPE termed CPE_int to denote that it corresponds to the combined space-charge response of both the oxide film and the electrical double layer at the electrode/solution interface.

The NR data, measured under cathodic polarization and analyzed by least-squares fitting the five-media model, lead to Figure 10. The most pronounced effect of the cathodic polarization is the increase of SLD of the Ti layer without significantly affecting the thickness. The expanded vertical scale of the inset shows that the SLD increased monotonically as the potential became more negative. This increase is most certainly due to absorption of D into the Ti layer. Horizontal dashed lines in the inset show that the Ti SLD demarcates cathodic polarization into the same four groups indicated by the EIS (Figures 6 and 9). This consistency between independent techniques confirms that the changes observed for the sample are real even though some are not much larger than estimated uncertainties.

Zoom In Zoom Out Reset image size

Figures 11 and 12 plot ρb(Ti) and d(Ti), respectively, as functions of polarization E. Error bars in both are given according to the uncertainties estimated by comparing the models obtained for the as-prepared sample as it was investigated in various geometries (Figure 7). The lack of random fluctuations suggests that the precision (but not necessarily the accuracy) in determining ρb and d is significantly better than indicated by the size of the error bars. These figures, together with Figures 4 and 9, form the basis for our discussion and conclusions.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Taken together, EIS and NR show the following sequence of changes to the sample:

• (1)  
  For the potentials in Group 1, no significant change is observed. The cell current remained low and was always anodic, if we discount the cathodic spike that occurred immediately after we set the potential (Figure 4). The cell resistance was high and the capacitance low (Figure 9). There is no sign of D infusion to the Ti metal (Figure 11).
• (2)  
  As the potential was set to more negative values (Group 2), the general trend was that the cell resistance decreased while the capacitance increased (Figure 9). At potentials of −0.40 and −0.45 V_SCE, the cell current remained cathodic for ∼1000 s after the spike (see Figure 4) before dropping to ∼ zero. For more negative potentials the current was persistently cathodic. The D content in the Ti increased in a step at −0.40 V_SCE, and then in smaller increments at the remainder of the Group 2 potentials (Figure 11).
• (3)  
  At −0.60 V_SCE the sample went through another abrupt change, marked by a very low cell resistance and a high capacitance (Figure 9). This is the onset of the sharp rise of the D content in the Ti (Figure 11). There is evidence that the Ti layer swelled slightly with the influx of D, but this is based on a trend, instead of the change being larger than the errors (Figure 12).

Based on these results a mechanism to explain the overall behavior of the sample can be proposed. First, we identify the reason for the sharp onset for D ingress at a specific potential (between −0.35 to −0.4 V_SCE). For our electrolyte (pH ∼ 6), the equilibrium potential for water reduction to form hydrogen gas and hydroxide ions is about −0.6 V_SCE, significantly more negative than −0.4 V_SCE. We propose that the underpotential deposition of deuterium occurs in the −0.3 to −0.4 V_SCE potential range under these conditions:

where D_ads is a surface-adsorbed D atom.³⁷

Existence of a range of potentials where the cathodic polarization is sufficient to reduce H⁺ to H_ads but not for H₂ evolution is well documented for Pt and Pd.³⁸ For Pt, underpotential-deposited H_ads atoms eventually cover the whole electrode surface, and the cell current falls to zero. On an oxide-covered Ti electrode, however, the D_ads atoms may be transported across the oxide if the layer is permeable. Whether transport occurs, and at what rate, will depend largely on the integrity of the oxide layer. To a certain extent, TiO₂ layers on regular bulk Ti are highly effective barriers against H ingress. In our case, however, since the underlying Ti is a deposited thin-film containing a high density of grain boundaries and many defects and voids (its SLD is lower than that of bulk Ti), it is conceivable that the oxide is also defective and less effective as a barrier. Consequently, D ingress into the metal may be initiated as soon as D_ads atoms are present at the surface.

Close inspection of Figure 4 supports this proposed mechanism. The current spike immediately after we increased the cathodic polarization (Figure 4) is expected, and can be attributed to interfacial capacitative charging. Ignoring surface charging, the cell current observed was exclusively anodic for the potentials in the range from 0 V_SCE to −0.3 V_SCE. By contrast, at −0.40 and −0.45 V_SCE, the current remained cathodic for ∼1000s after the spike, underwent a small jump toward zero, and then fluctuated around zero as a residual current. The residual current is leakage current or due to an electrochemical process that does not involve further loading of D into the underlying metal: otherwise, the D content in Ti would not achieve a steady state. We conjecture that loading of D into the Ti metal takes place within the first 1000s, while the current is cathodic. The two NR scans successively carried out at each potential gave similar results, indicating that the arrival of D within the metallic Ti layer is prompt (i.e., on the order of minutes, not hours) and supporting this proposal.

Additional support for the proposed mechanism, and a more accurate determination of the onset potential, comes from recent work by Zeng et al.^12,39 Samples used in that study were 20–30 nm thick TiO₂ layers directly deposited onto a Pd substrate (no metallic Ti between the oxide and the substrate). Galvanostatic measurements using a Devanathan cell showed that the H-charging current through the oxide extrapolates to zero at −0.37 V_SCE in a plot of potential vs. charging current (Figure 10 in ref. 12). The authors concluded that −0.37 V_SCE is the potential where TiO₂ becomes transparent to H.

The availability of active electronic current pathways across the TiO₂ oxide, which is normally considered an insulator at potentials positive of ∼−0.6 V_SCE, has been demonstrated at potentials more positive than ∼−0.6 V_SCE by Zhu et al.,^17,18 who used scanning electrochemical microscopy to observe the activation of localized electronically conducting sites at which cathodic current could be passed at potentials as mild as −0.04 V_SCE. The number and activity of such sites increased as the potential was made more negative. The locations of the active sites were mapped on the surface to demonstrate that they were associated with grain boundaries and triple points in the underlying metal alloys. An influence of alloying element enrichment at grain boundaries, β-phase regions, and intermetallic precipitates on the local oxide conductivity was suggested.

In our case, and in that of Zeng et al.,^12,39 the Ti material used to prepare the thin film was high purity Ti, rather than alloy material as used by Zhu et al.,^17,18 so we cannot attribute any local activity to the effects of alloying elements or impurities. That may be why we do not observe cathodic currents at potentials positive of −0.4 V_SCE, but the fact that we and Zeng et al.,^12,39 did observe cathodic currents at potentials more positive than −0.6 V_SCE suggests that even without impurities and alloying additions, the presence in the oxide film of electronic defects, vacancies, and disorder at grain boundaries can render the oxide more readily conductive locally.

Taking these observations into account, we suggest that the onset of hydrogen penetration through the oxide and into the metal is controlled by the appearance of surface-adsorbed H (D) on the oxide. The H_ads cannot be produced in significant quantities on the oxide surface unless and until the oxide becomes conductive and able to supply the electrons required to reduce water or hydrogen ions to adsorbed atoms (Reaction 5). The development of conductivity, conversion of the Ti^IV oxide to a mixed Ti^III/Ti^IV oxide, and absorption of hydrogen seem to be all simultaneous consequences of the same conversion process.¹³ Therefore, once the oxide becomes conductive, at least locally, H atoms can be produced on its surface, then incorporated into the oxide, resulting in its conversion (Reaction 2), followed by H ingress into the underlying metal.

The D content of the Ti metal layer can be determined from the layer SLD, which is a weighted average of the ρb values for the Ti and the D atoms present. At the onset of D uptake, the content of D in Ti reached an atomic ratio of ∼0.04, according to Figure 11, where the atomic ratio, is defined as (number of D atoms)/(number of Ti atoms) in the metal film. From thereon, the D content increased gradually for the remainder of the Group 2 potentials (down to −0.55 V_SCE). Since the conditions for reduction of D⁺ to D_ads exist over this entire potential range, this slowdown in D uptake needs an explanation. The critical solubility of H in Ti for hydride formation at room-temperature ranges from 0.002 to 0.004 in atomic ratio H:Ti for bulk metal,^40–42 and from 0.06 to 0.07 for evaporated Ti films.^43,44 The marked difference between bulk and thin-film is undoubtedly due to morphology: bulk metal grains are tightly packed whereas evaporated thin-films contain substantial voids and cracks. Presumably, as D reaches the metal, hydrides form readily provided the local concentration is above the solubility limit. Subsequently, already formed hydrides separate additional D from the metal surface, leading to a slowdown in further hydride formation: the hydrides formed locally at the base of the D-entry windows provided by defects in the native oxide act as diffusion barriers, effectively blocking the windows and retarding further transport of D atoms into the metal.

This behavior may also account for the discrepancy between the notion that potentials negative of ∼−0.6 V_SCE are required for significant and damaging amounts of hydrogen ingress^1,5,8,9,11 and the observation of hydrogen penetration at much more positive potentials in this work and recent others.^12,17,18,39 If local hydrogen absorption windows at grain boundaries become active, but hydrogen ingress is retarded by hydride "plugs" at their base, then the amount of hydrogen absorbed may appear to be inconsequential (e.g., below detection limit) on a bulk Ti material.

According to Figure 11, a sudden increase in hydrogen content in the Ti film begins at ∼−0.6 V_SCE. Ohtsuka et al.¹³ showed that in neutral solution, cathodic reduction of a TiO₂ film does not cause it to dissolve, but induces a change in film composition due to hydrogen absorption (Reaction 2). The amount of hydrogen absorbed is a function of the cathodic potential applied. The redox transformation within the passive oxide (Ti^IV→Ti^III) leads to the availability of multiple oxidation states and a consequent increase in the conductivity of the film. As explained by Nowierski et al.,⁴⁵ at ∼−0.6 V_SCE the oxide film on Ti acts as a wide bandgap semiconductor (with a 3.05 eV bandgap), and a slight deviation from stoichiometry (TiO_2−x) transforms the oxide layer into an n-type semiconductor. The combination of oxygen vacancies and interstitial Ti^III ions leads to trapping of electrons in an orbital band just below the conduction band edge.^14–16 When the TiO₂ is polarized to potentials below the flatband potential (∼−0.55 V_SCE), redox transformations within the passive oxide (Ti^IV→Ti^III) facilitate electron transfer across the film/solution interface and allow the reduction of protons, an electrochemical process accompanied by absorption of H into the oxide.^8,11

The combined use of EIS and in-situ NR techniques has enabled us to reconcile seemingly conflicting literature reports of two key potentials, −0.37 V_SCE¹² and −0.6 V_SCE,¹³ important in the behavior of a thin-film Ti electrode under cathodic polarization_. At the first potential, significant changes in current density, oxide resistivity and capacitance, and the SLD of the Ti are observed. These changes can be attributed to the onset of local oxide conductivity and the underpotential deposition of D_ads atoms on the electrode and their entry into the metal through conductive, defective pathways in the oxide. At the second potential, −0.6 V_SCE, a marked increase in current density and the D content in the Ti film are observed. The latter potential is recognized as the potential that can render the protective oxide ineffective as a barrier by reducing it from the original TiO₂ to TiOOH.¹³ This is considered the threshold for general hydrogen uptake, which leads to significant hydrogen absorption and embrittlement of industrial Ti structures at potentials negative of −1 V_SCE. The former potential, −0.37 V_SCE, has not previously been generally recognized as a hydrogen ingress threshold in bulk Ti materials, possibly because it has not led to hydrogen uptake that is sufficient to affect bulk Ti materials.

The research presented herein is made possible by a reflectometer jointly funded by Canada Foundation for Innovation (CFI), Ontario Innovation Trust (OIT), Ontario Research Fund (ORF), and the National Research Council Canada (NRC). The Canadian Natural Sciences and Engineering Research Council (NSERC) is acknowledged for additional research support.