Evidence of hydrogen trapping at second phase particles in zirconium alloys

Zirconium alloys are used in safety–critical roles in the nuclear industry and their degradation due to ingress of hydrogen in service is a concern. In this work experimental evidence, supported by density functional theory modelling, shows that the α-Zr matrix surrounding second phase particles acts as a trapping site for hydrogen, which has not been previously reported in zirconium. This is unaccounted for in current models of hydrogen behaviour in Zr alloys and as such could impact development of these models. Zircaloy-2 and Zircaloy-4 samples were corroded at 350 °C in simulated pressurised water reactor coolant before being isotopically spiked with 2H2O in a second autoclave step. The distribution of 2H, Fe and Cr was characterised using nanoscale secondary ion mass spectrometry (NanoSIMS) and high-resolution energy dispersive X-ray spectroscopy. 2H− was found to be concentrated around second phase particles in the α-Zr lattice with peak hydrogen isotope ratios of 2H/1H = 0.018–0.082. DFT modelling confirms that the hydrogen thermodynamically favours sitting in the surrounding zirconium matrix rather than within the second phase particles. Knowledge of this trapping mechanism will inform the development of current understanding of zirconium alloy degradation through-life.


Results
Experimental. To investigate the distribution of hydrogen and deuterium in Zircaloy samples, several (> 10 in total) regions of interest (ROIs) were imaged by NanoSIMS. Figure 1 shows the Secondary Electron (SE), 1 H − , 2 H − and 2 H − / 1 H − ratio images for three separate ROIs on the 402.2 day Zy-4 (ROIs 1,2) and 48.3 day Zy-2 samples (ROI 3). The 1 H − images reveal the grain structure of the sample; the signal variation in different grains is due to the variation in sputter rate with grain orientation 48 . Taking a ratio of the 2 H − and 1 H − signals effectively allows for differences in the ion signal due to differences in the grain to grain sputter rate to be cancelled out as the variation in sputter rates will be consistent between hydrogen isotopes. Also visible in the 1 H − image are several small, dark circular features with a lower signal compared to the surrounding metal. These dark features appear to be SPPs in the α-Zr matrix. The 2 H − images show an inhomogeneous 2 H distribution with small regions with high 2 H abundance in the α-Zr matrix beneath the oxide layer. The same distribution is also clear in the 2 H − / 1 H − ratio images where grain orientation contrast has been minimised by ratioing. Both the Zy-4 and Zy-2 samples show the same localisation of 2 H − to the SPPs, but lower 2 H − / 1 H − ratios were observed in the Zy-2 sample. The Zy-4 and Zy-2 were oxidised at different times in different autoclaves yet still show the same localisation of 2 H − to the SPPs. Table 1 shows the average size of the features visible in the 2 H − / 1 H − ratio images in Fig. 1. The sizes of these features were determined using L'Image (L.R. Nittler, Carnegie Institution of Washington) to generate circular boundaries around local maxima in the 2 H − / 1 H − distribution, with the threshold value for determining the placement of a boundary set at 20% of the global maximum 2 H − / 1 H − ratio in the metal substrate region of the ROIs in Fig. 1 (i.e. discounting any signals from the oxide and metal/oxide interface regions and any peaks in ratio below 20% of the maximum ratio). The outer edge of the circular ROIs was determined by the point at which the signal dropped below 50% of the local peak value. Additionally for each ROI shown in Fig. 1 the NanoSIMS beam width was determined using the 16-84% criterion 49 . Figure 2 shows the 1 H − , 2 H − , 56 Figure 3a,b shows SE images obtained using the NanoSIMS and an FEG-scanning electron microscope (SEM) respectively from the Zy-4 sample. The 2 H − / 1 H − ratio image for this is shown in Fig. 3c. This corresponds to the lower left corner of the 2 H − / 1 H − ratio image shown in Fig. 2 but the image has been rotated, cropped and resized to match with the SEM and Energy dispersive X-ray spectroscopy (EDS) images ( Fig. 3b-d). Figure 3d,e show EDS Cr and Fe distribution maps of the same region with the SPPs labelled A-C matching the elevated 2 H − / 1 H − ratio distribution labelled A-C in Fig. 3c. Modelling. SPP crystal structures were generated with Fe: (Fe + Cr) atomic ratios of 43.75 at% and 62.50 at%. This is representative of the SPPs found in Zy-2 and Zy-4, respectively. No disorder was imposed onto the Zr sublattice. Figure 4a shows the relative hydrogen solution enthalpy, ΔH_sol, for H to dissolve in the SPP compared to hcp-Zr, for 35 interstitial sites at each SPP composition. While the spread is significant, the relative solution  www.nature.com/scientificreports/ energy is positive for all sites, as such these results still suggest that it is thermodynamically unfavourable for H to be in the bulk of the SPPs, compared to the bulk of hcp-Zr. Analysis of the local chemistry (nearest neighbour configuration) revealed that for all simulations, H relaxed into a tetrahedral site, irrespective of starting coordinates. This matches previously observed behaviour in binary SPPs and α-Zr. Figure 1 shows that the 2 H is not evenly distributed throughout the α-Zr matrix and forms several 'hot spots' in all 3 ROIs. In ROIs 1 and 2, small dark features can be observed inside several of the areas with high 2 H − signal on the 2 H image. The 2 H − signal strength in ROI 3, the 48.3 day Zy-2 sample, is too low to resolve these features to the same degree as for ROIs 1 and 2. It is proposed that the dark features visible in several of the 2 H concentrations are SPPs, and that the 2 H sits in the matrix around the SPPs.

Discussion
The natural 2 H/ 1 H ratio is 1.56 × 10 -4 ; however, 2 H − / 1 H − ratios reach values of 1.8-8.2 × 10 -2 at the 2 H − enriched regions, which is more than 100 times higher than the natural background level. This high level of enrichment indicates that the deuterium detected originates from the oxidation process in the spiked autoclave rather than as part of the natural atmosphere present in the NanoSIMS chamber or sample preparation. The water used in the spiked autoclave was not 100% D 2 O so it was expected that elevated levels of 1 H would also be observed around each of the SPPs. Due to high levels of 1 H contamination of the cross sectioned surface from sample preparation and the vacuum chamber, elevated signals of 1 H were not observed. The counts in the 1 H image are more than an order of magnitude higher than the 2 H counts, therefore any 1 H enrichments that would be present after the autoclave experiment at a similar count rate to the 2 H signal (for the 50% D2O labelled Zr-4) are masked by the background. As the H picked up during corrosion is observed to migrate to SPPs, no 2 H enrichment is detected in regions between the SPPs. It is noted that as the samples were removed from the spiked autoclave for > 6 months before NanoSIMS analysis was conducted, deuterium present in the α-Zr matrix had sufficient time to redistribute towards more energetically favourable sites. The matrix around the SPPs can therefore be considered as a deep trap site as elevated 2 H signals were still detectable.
Two separate approaches were taken to confirm the presence of SPPs at the locations of the high 2 H/ 1 H ratio. First the NanoSIMS was used to map the distribution of 56 Fe 16 O − secondary ions in the sample. Fe − has a low ionisation rate when using a Cs + primary beam to investigate negative ions in the NanoSIMS, so 56 Fe 16 O − was used as a proxy signal for the Fe concentration. There is typically enough oxygen present in the NanoSIMS to allow for the continued oxidation of material newly uncovered by sputtering and as such the 56 (Fig. 2). However, as discussed previously, the 2 H enrichment is associated with the periphery of the SPPs rather than the SPPs themselves.
This fits with the DFT results presented here. Figure 4a shows that it is thermodynamically unfavourable for H to be accommodated inside Zy-2 or Zy-4 SPPs, whether the SPP is binary or ternary in composition, when in competition with the α-Zr matrix. It should also be noted that while these simulations have been performed without considering the effect of temperature, it is unlikely that the entropy contributions at the temperatures encountered in service (~ 300 °C) would be large enough to overcome the energy gap compared to hcp-Zr (0.70 eV). The predicted preference for hcp-Zr over SPPs is reinforced by further analysis of hydrogen's local chemistry. Figure 4b shows that H preferentially occupies sites with the largest fraction of Zr nearest neighbours. It should be noted that in the structure of Zr(Fe,Cr) 2 SPPs there are no tetrahedral sites bound by more than 2 Zr atoms. Even for these sites, favourable thermodynamics are not predicted. This information when combined with the distribution of 2 H observed in the experimental results implies that the 2 H is either trapped at the interface between the SPPs and the matrix or it is trapped in the matrix surrounding the SPPs.
Trapping of hydrogen at incoherent interfaces has been observed in other alloy systems, where incoherent interfaces between precipitates and the matrix have trapped hydrogen 40,[43][44][45][46][47]50 . However, the DFT results shown here and presented in the literature 25,41 indicate that the α-Zr matrix is the preferred site for hydrogen to inhabit.
The DFT analysis of hydrogen's local chemistry indicates that the trapping of H around SPPs is probably not due to chemical affinity with Fe and Cr. Considering that H preferentially occupies tetrahedral sites in both the SPP and the α-Zr matrix, and since H shows a clear preference for tetrahedral sites with fewer Fe and Cr nearest neighbours, it is unlikely that the trapping in that region is due to direct bonding between H and Fe or Cr. On the other hand, it is well established that H is attracted to areas of tensile strain 51 .
Tensile stress fields may be present in the region surrounding the SPP-matrix interface due to a mismatch in thermal expansion coefficients between the SPPs and the matrix 52 . The coefficients of thermal expansion for ZrCr 2 53 and ZrFe 2 54 (both commonly observed stoichiometries of SPPs 25 ) are both significantly larger than that for Zy-4 55 . As a result, when the alloy is cooled from processing or operating temperatures this results in the formation of tensile stresses in the matrix around the interface 52 . These tensile stresses could then attract hydrogen to the stressed matrix surrounding the SPPs as the system cools from elevated temperatures. The build-up of stresses as a result of thermal expansion coefficient mismatch has previously been observed for Al 2 Cu particles in an Al matrix 56 .
The enriched features in ROI 3 of Fig. 1 are noticeably smaller but greater in number, and have lower 2 H − / 1 H − ratios when compared to ROIs 1 and 2. This is also shown in Table 1 where the average feature size is smaller in ROI 3 despite the beam size and raster size being similar to ROIs 1 and 2. The 1 H images in Fig. 1 each show a distribution of small dark features in the metal substrate, these are consistent with SPPs, with the SPP distribution in ROIs 1 and 2 mostly comprised of a small number of SPPs distributed in clusters whereas the SPPs in ROI 3 are present in greater numbers and are evenly distributed across the ROI. This matches the differences in the 2 H − / 1 H − features between the ROIs. The availability of a larger number of potential trapping sites (i.e. SPPs), a shorter exposure time to 2 H 2 O (45 vs 60 days) and a lower spiking level (15 vs 50%) could explain why less 2 H is concentrated in any given site in Zy-2, compared to Zy-4, and thus why the measured 2 H − / 1 H − ratio is lower. Additionally, the presence of Ni in Zy-2 compared with Zy-4 will change the stoichiometry of the SPPs, resulting in different coefficients of thermal expansion. This would lead to different strain fields around each SPP, potentially changing the ability of the matrix around the SPP to trap hydrogen.
In order to corroborate the NanoSIMS results, EDS analysis of the same ROI analysed in Fig. 2 was undertaken (Fig. 3). In the SE images a group of features < 1 µm in size are visible and scattered throughout the Zr grains (Fig. 3a,b), three of these features have been labelled A, B, and C. These features are surrounded by large amounts of 2 H − relative to the matrix away from them (Fig. 3c). EDS analysis shows that all these features are rich in Fe and Cr signals (Fig. 3d,e). The combination of their small size and the presence of Fe and Cr in the same location confirms that these features are SPPs with associated high 2 H − signals.
The phenomenon of hydrogen trapping in the matrix around SPPs has not previously been observed. While some of this can be attributed to an inability to detect hydrogen with commonly used SEM based techniques or a lack of spatial resolution in techniques commonly used to detect hydrogen, as detailed in the introduction to this paper, this does not fully explain why segregation of H to SPP interfaces has not been observed before.
Recent work by Li et al. 34 and Liu et al. 57 used NanoSIMS to investigate the distribution of 2 H in zirconium oxides 34,57 but did not observe the same trapping behaviour as seen here. This is due to the effect of the Nano-SIMS analysis itself on the trapping sites, as beam damage and mixing from the 16 keV Cs + ion beam destroys the trapping sites below the sampling depth of the NanoSIMS, allowing 2 H to escape before being detected.
The beam damage renders it difficult to detect 2 H around the SPPs as the trapping sites are irreversibly damaged during analysis, freeing the trapped 2 H and makes the 2 H trapping appear transient when analysed. The damage to trapping sites, and consequent spreading/escape of hydrogen, is reflected in the size of the features reported in Table 1. While the nature of this irradiation damage has not yet been analysed, it seems reasonable to expect the stress state in the matrix to change as a result of ion bombardment, potentially removing or reducing the tensile stresses that attracted hydrogen. Table 1 gives a 2 H concentration size of 0.97-1.77 μm, far larger than values for typical SPP size given in the literature 58 where the modal size for SPPs in Zy-4 is 0.2 µm. This can be partly explained by the fact that the NanoSIMS beam size is much larger than the SPP but even when this is accounted for, the minimum size of www.nature.com/scientificreports/ the 2 H features is measured at 0.28-0.95 μm, still larger than the expected size for SPPs in Zy-2 and Zy-4. This size increase is the result of 2 H spreading from damaged trapping sites as a consequence of NanoSIMS analysis.
To demonstrate this effect, Fig. 5 shows five successive image acquisitions from a 5th ROI on a 283 day Zy-2 sample. In this ROI a widespread pattern of 2 H concentrations is visible around SPPs, consistent with the observed 2 H distributions seen in Figs. 1 and 2. However in this ROI there are also a number of hydrides present, these are indicated in the image, and are easily identifiable due to the high level of 2 H present and their elongated morphology which is different to the concentrations of 2 H around SPPs. With successive planes of analysis, the 2 H signals around SPPs reduce in intensity and become diffuse. A similar effect can be seen for the three hydrides as they are also damaged by the Cs + beam. The graph in Fig. 5 was produced by generating ROIs around each 2 H concentration using the method described previously with L'image, however it should be noted that any ROIs that intersected with the oxide layer or the hydrides present were discarded. The data plotted in Fig. 5 were generated by summing the total 1 H − and 2 H − counts across all the ROIs in each plane and then finding the isotopic ratio. The graph shows a relative decrease in the 2 H/ 1 H ratio of the SPPs between image planes 1 and 5, showing that 2 H is released from its trapping sites by Cs + ion bombardment. The small relative increase in the 2 H/ 1 H ratio for the surrounding Zy-4 base metal indicates that the 2 H is not only released to the vacuum but also migrates laterally onto the Zy-4 surface.
This observed depletion and dispersion of the 2 H − signal is due to the accumulation of sub-surface damage in the sample, which is a result of continued analysis with the 16 keV Cs + primary ion beam. Figure 6a plots displacements produced by a single 16 keV Cs + ion impacting a Zy-4 target with depth, calculated using quick Kinchin-Pease calculations in SRIM 59 . The damage caused by each ion impact extends far beneath the ~ 1 nm sampling depth produced by each impact 60 . As a result, damage accumulates in the sample with successive planes of analysis. Figure 6b shows how the damage rapidly accumulates with continued analysis, using similar analysis conditions to those used in this paper, before saturating at ~ 30 planes of analysis.
These results indicate that NanoSIMS images are always acquired from heavily damaged material. As it can take > 5 min to image a single plane, any hydrogen released from the damaged trap sites has time to redistribute to the surrounding α-Zr matrix before being analysed in a subsequent image plane. This is the main cause of the depletion/degradation of 2 H − signal observed in Fig. 6 and contributes significantly to the large measured size of the 2 H features given in Table 1. Furthermore, this damage explains why the present result has not been observed in previous analysis undertaken with the NanoSIMS. If the imaging conditions are not correctly chosen, all 2 H will have diffused away from the sample prior to imaging.
This effect is a particular concern for SIMS imaging of hydrogen in the α-Zr matrix as hydrogen is very mobile in the metal matrix, even at room temperature 11 . This effect is of less important when imaging hydrogen in the oxide (as in 34 ) as the hydrogen is much less mobile and will not redistribute as readily.

Conclusions
Using NanoSIMS and EDS, it has been observed that SPPs in Zy-2 and Zy-4 can act as preferential trapping sites for 2 H introduced via corrosion, with measured isotope ratios of 2 H/ 1 H = 1.8-8.2 × 10 -2 , far in excess of the natural ratio. Complementary DFT modelling of the SPPs in Zy-2 and Zy-4 has shown that H solution in the bulk of the SPPs is thermodynamically unfavourable indicating that the observed 2 H trapping occurs in the α-Zr matrix around the SPPs. This trapping is likely the result of tensile thermal mismatch stresses generated in the matrix around the SPPs by the differences in thermal expansion coefficients between the matrix and the SPP on cooling.
As current understanding of the hydrogen within Zr alloys does not take hydrogen segregation to SPPs into account, and factors such as the hydriding behaviour are crucial pieces of information from a safety perspective, this new observation can inform models used to predict the margins of safety in Zr alloy components for nuclear applications, thus ensuring safe and efficient operation.

Methods and materials
Experimental. Two different sets of zirconium alloys were analysed, one Zy-4 sample and two Zy-2 samples.
The samples were oxidised in autoclaves at Jacobs, first in a refreshed autoclave at 350 °C and 18.6 MPa for 341.0 (Zy-4), 222.4 (Zy-2) and 3.3 (Zy-2) days, before being transferred to a static autoclave spiked with 50% 2 H 2 O for 61.2 days (Zy-4 and Zy-2) and 15% 2 H 2 O for 45.0 days (Zy-2) in otherwise similar conditions. These particular samples were chosen from an available array of pre-oxidised samples made available for this study, oxidation and weight gain data was provided by Jacobs plc.
The oxidation conditions were chosen due to their similarity to the conditions found in pressurised water reactors but with a slightly elevated temperature to increase the corrosion rate. Deuterium enriched water was used to allow unambiguous identification of hydrogen species introduced during the final oxidation step. The weight gains and estimated oxide thicknesses for the samples used in this study can be found in Table 2.
Each sample was sectioned into 2-3 mm cubes and mounted in cross section using araldite resin such that the full thickness across the oxide layer was visible. The samples were prepared by mechanical grinding from 800 to 4000 grit followed by a slow polishing process using a colloidal silica (0.04 μm) suspension for > 30 min, this preparation technique is based on the method used by Yardley et al. 33 . Immediately prior to loading into the NanoSIMS, the samples were coated with 10 nm of platinum in order to prevent charging of the samples during analysis.
The analysis was performed on a CAMECA NanoSIMS 50L (Cameca, France), which is a double focussing mass spectrometer with one fixed and six movable detectors and a separate secondary electron (SE) detector. A 16 keV Cs + ion beam with a current of 1.3-1.8 pA was used to sputter the surface and generate negative secondary ions. All regions of interest were initially implanted with a high current beam to remove the Pt coating in the analysis area and increase secondary ion yield and then 5-10 sequential images were acquired from each region  SPPs in regions analysed using the NanoSIMS were identified using a Zeiss Merlin Field Emission Gun-Scanning Electron Microscope (FEG-SEM) operating at 3 kV for a high lateral resolution sufficient to detect the SPPs. SE images were obtained using the InLens detector. The SPPs were chemically identified using the Oxford Instruments Extreme detector which uses a windowless silicon-drift detector developed for low-kV energy dispersive X-ray spectroscopy (EDS) analysis.
Modelling. The solution enthalpy of H in Zr and its intermetallics, from H2 molecules, is negative 25 hydriding is a exothermic). Thus, we measure the relative enthalpy ( H sol ) between H solution in the SPPs and H solution in α-Zr matrix: where E DFT α−Zr and E DFT SPP are the DFT energies of hcp-Zr and SPP supercells (as described below), and E DFT α−Zr+H and E DFT SPP+H are the DFT energies of the same structures with the addition of one H interstitial defect (for hcp-Zr we considered only the more favourable tetrahedral site).
The ternary Zr(Fe,Cr)2 SPPs found in Zy-2 and Zy-4 are C-14 Laves phases with hexagonal structure (space group P63/mmc) and disorder only on the Fe/Cr sublattice. These were modelled using a set of special quasirandom structures (SQS) 61 . SQS reproduce, in a finite supercell, the radial correlation functions of an infinite