Deuterium supersaturated surface layer in tungsten: ion energy dependence

Properties of deuterium (D) supersaturated surface layers (DSSLs) formed in tungsten (W), such as thickness, internal microstructures, and D retention, are experimentally investigated as a function of the incident ion energy, E i. W samples were exposed to D plasmas in the PISCES-A linear plasma device in a range of E i ∼ 45–175 eV, while other plasma exposure parameters were fixed: sample temperature, T s, ∼423 K, ion flux, Γi, ∼1.2 × 1021 m−2s−1, and fluence, Φi, ∼3.0 × 1024 m−2. High-resolution, cross-sectional, transmission electron microscopy observations confirm that (1) a DSSL forms even at the lowest E i ∼ 45 eV, (2) the DSSL thickness, Δt DSSL, is found to decrease with decreasing E i from ∼11–12 nm at E i ∼ 175 eV to ∼5–6 nm at ∼45 eV, and to agree with approximately the maximum implantation depth calculated using SDTrimSP, and (3) high-density D nanobubbles with a diameter of ∼1 nm or less exist inside the DSSL, which is deemed to validate a theory-predicted vacancy stabilization process due to trapping of a solute D atom(s). Utilizing a D areal density of ∼4.2 × 1019 m−2 in the first 14 nm from the surface at E i ∼ 75 eV from nuclear reaction analysis and the measured E i dependence of Δt DSSL, our previous laser-induced breakdown spectroscopy data is updated: both dynamic and static D retention increase with decreasing E i, and the D/W atomic fraction during plasma exposure reaches ∼0.3 at E i ∼ 45 eV. A possible DSSL formation mechanism is proposed.


Introduction
Deuterium (D) supersaturation, meaning that the D concentration largely exceeds the solubility limit, can occur * Author to whom any correspondence should be addressed.
Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
in the subsurface region of tungsten (W) exposed to a D plasma, since a high-flux of D is implanted into a shallow depth of W. The mean implantation depth of D in W ranges from ∼1.3 nm at 10 eV to ∼7.9 nm at 300 eV [1]. Such a thin surface layer saturated with D is often called a D supersaturated surface layer (DSSL). Because of its unique state, both experimental [2][3][4] and theoretical [5][6][7][8][9] work have been conducted in recent years to clarify its basic properties such as the DSSL thickness, ∆t DSSL , and the D retention in DSSLs, as well as its formation mechanism and condition.
Theoretical calculations have predicted that a monovacancy (V) can trap multiple D atoms and form VD j complexes [5][6][7]10]. For instance, VD 6 is calculated to be dominant at room temperature [5][6][7]. This D trapping would increase the vacancy or VD j concentration by preventing the vacancy from recombining with a self-interstitial atom [5][6][7]. This vacancy stabilization is predicted/considered to be one of the key processes to the DSSL formation.
A high D/W atomic fraction of ∼0.1 in a DSSL was experimentally confirmed [2], after a D plasma exposure of a W sample at incident D + ion energy E i = 215 eV, sample temperature T s = 300 K, and ion flux Γ i = 9.9 × 10 19 m −2 s −1 . It was measured using the Ar-sputter nuclear reaction analysis (NRA) technique with a ∼3 nm depth resolution, which also determined ∆t DSSL ∼ 10 nm.
The threshold E i for the DSSL formation was also investigated with scanning electron microscopy (SEM) surface observations [2]. D plasma-exposed W samples exhibited blurry surface images, caused by some damage structures inside the DSSL, at E i = 115-415 eV, but not at E i = 60 eV. From depth distributions of the energy transferred in D-W collisions calculated with the SDTrimSP binary collision approximation code [11,12] and the measured ∆t DSSL ∼ 10 nm, the threshold E i for the DSSL formation was predicted to be 93 ± 23 eV, which is located between 60 eV (sharp image: no DSSL) and 115 eV (blurry image: DSSL) from the SEM surface observations. Importantly, both experimentally confirmed (115 eV) and predicted (93 eV) threshold values for the DSSL formation are much lower than the minimum required E i ∼ 930 eV for D + to displace a W lattice atom, as the threshold displacement energy, E d , for W is as low as 40 eV [13]. For clarity, the (maximum) energy transferred from D to W, E D→W , is expressed as, where m D and m W are the mass of D and W, respectively. Thus, E i = 115 and 93 ± 23 eV correspond to E D→W ∼ 5 and 4 ± 1 eV. In their proposed DSSL formation mechanism [2], it was invoked that a 'temporary' Frenkel pair is created due to an energetic D + collision at a subthreshold energy, and the vacancy is then stabilized by trapping a solute D atom before recombining with the displaced (self-interstitial) W atom. In our previous work [4], in-operando/in-situ laser-induced breakdown spectroscopy (LIBS) measurements were performed during D plasma exposure of W in the PISCES-A linear plasma device to explore dynamic, as well as static, D retention in the near-surface region within the laser ablation depth of ∼350 nm. It was found that both dynamic and static D retention showed no E i dependence in a range of E i ∼ 45-175 eV. This indicates that a DSSL should exist down to, at least, E i ∼ 45 eV, which is substantially lower than the 115 eV experimentally confirmed, and 93 ± 23 eV predicted, threshold E i values in [2].
To address this discrepancy, we further investigate the E i dependence of the DSSL formation by utilizing highresolution transmission electron microscopy (TEM), combined with focused ion beam (FIB), to detect/observe DSSLs in this work. From cross-sectional TEM observations of DSSLs, ∆t DSSL is determined as a function of E i . The amount of D retained in a W sample exposed at E i ∼ 75 eV, i.e. the D areal density, n D , is quantified using NRA. By taking into account the measured ∆t DSSL and n D , the E i dependence of the D/W atomic fraction measured using LIBS in our previous study [4] is updated. Finally, we propose a possible DSSL formation mechanism.

D plasma exposure
Before D plasma exposure, the surface of 99.95% pure W samples (Midwest Tungsten Service) with a diameter of 25.4 mm and a thickness of 1.5 mm was mirror-polished. Then, the samples were annealed at 1173 K for 30 min under vacuum. After the sample preparation, D plasma exposures of the W samples were conducted in PISCES-A [14] at four different E i ∼ 45, 75, 125, and 175 eV, which span below and above the previously determined thresholds of 115 and 93 eV. We determine E i from the potential difference between the plasma space potential, V s , and the target negative bias voltage, V b , i.e. E i = V s -V b . It should be noted that D + atomic ions have this E i , while D 2 + and D 3 + molecular ions are considered to first dissociate upon implantation, and each D carries 1/2 and 1/3 of E i , respectively.
A reciprocating single Langmuir probe system, located at ∼50 mm upstream of the sample surface, measures V s , ion flux, Γ i , electron density, n e , and electron temperature, T e . In the present experiment, measured plasma parameters are: V s ∼ −15 ± 5 V, Γ i ∼ (1.2 ± 0.1) × 10 21 m −2 s −1 , n e ∼ (0.14 ± 0.03) × 10 18 m −3 , T e ∼ 5.0 ± 1.0 eV. The ion fluence, Φ i , was ∼3.0 × 10 24 m −2 , which was reached after ∼2500 s of plasma exposure. With the measured n e and T e as well as the D 2 gas pressure P D2 ∼ 1.5 mTorr from a Baratron gauge, the dominant ion species would be D + with an ion composition of (D + , D 2 + , D 3 + ) = (0.58 ± 0.06, 0.34 ± 0.05, 0.08 ± 0.02). This ion composition is estimated using a 0D model [15] updated with the latest atomic and molecular data used in the EIRENE code [16] from [17], which solves rate balance equations. Hence, the D fluence would be calculated A W sample was mounted on a sample manipulator, and was heated by plasma loading. The sample temperature, T s , was controlled by forced air cooling to ∼423 K, measured with a thermocouple pressed to the back side of the sample. The base pressure in the PISCES-A vacuum chamber is typically around 5 × 10 −7 Torr, measured with an ion gauge.

Sample analyses
Internal microstructures of the subsurface region were examined for four W samples exposed at all E i , as well as an unexposed W sample for reference, at NIFS, Japan. Since the DSSL thickness is expected to be around 10 nm [2], microstructures inside a DSSL are even smaller. Thus, a very thin lamella specimen was carefully made out of each W sample using a FIB-SEM system (HITACHI NB5000). Special care was taken for preserving the top surface layers during FIB thinning processes. High-resolution, cross-sectional, observations of the lamella specimens were then conducted with a TEM device (JEOL JEM-2800).
Depth profiling of D in another W sample exposed at E i ∼ 75 eV was conducted using NRA at IPP Garching, Germany. Details of the setup can be found elsewhere [18]. The 3 He(D,p)α reaction was used with the 3 He beam energy scanned from 2.4 MeV to 1.8, 1.2, 0.8, 0.69, and 0.5 MeV. Both reaction products (protons and alphas) were detected at 135 • and 102 • with respect to the incoming 3 He beam, respectively. Since stopping of 3 MeV alphas is much larger than that of protons, the depth resolution near the surface is much better with alphas. To further improve the near-surface depth resolution, the sample was tilted by 5 • away from the alpha detector. To obtain a depth profile of the D areal density or the D/W atomic fraction, measured proton spectra for all six energies and alpha spectra for the three lowest energies were simultaneously analyzed with the NRADC program [19], which uses the SIMNRA program [20] to simulate the spectra. Note that the information depth at 2.4 MeV was 3.1 µm, and a depth resolution from the surface of 14 nm was achieved in this analysis.

DSSL formation, internal microstructures, and thickness
In this section, the formation, internal microstructures, and thickness of DSSLs are discussed based on high-resolution TEM images.
In figure 1, cross-sectional TEM images of the subsurface region are presented for W samples exposed at all E i (a)-(d) and an unexposed (e) for comparison. First of all, some structure modifications in the subsurface region are seen for all D plasma-exposed samples in comparison with the unexposed sample. This means that a DSSL forms even at the lowest E i ∼ 45 eV (E D→W ∼ 1.9 eV), which is substantially lower than both the experimentally confirmed E i ∼ 115 eV and the predicted E i ∼ 93 ± 23 eV [2]. It is noticed that the DSSLs at E i ∼ 125 and 175 eV look darker than those at E i ∼ 45 and 75 eV. This indicates that strong distortion fields densely accumulate in the W lattice at the higher E i due to severe damage, sometimes causing small surface modifications, which are detectable with SEM. It is, thus, possible that the SEM surface observations in [2] could detect these small surface morphology changes, but not detect less severe or mild damage at E i < 115 eV. However, the plasma exposure condition in [2] is different from the present experiment. In particular, Γ i was around an order of magnitude lower in [2], which might affect the DSSL formation. The Γ i dependence will be studied in the future.
Next, looking at internal microstructures of the DSSLs in more detail (also see the enlarged images enclosed with the solid squares), one can find many white dots. A representative dot is pointed to by the arrow in each enlarged image. These white dots are nanocavities, which are thought to be D nanobubbles, since D is detected inside the DSSLs, as presented in section 4. The size of D nanobubbles is observed to be around 1 nm or less. These D nanobubbles are considered to be VD j complexes and/or V i D j clusters. To our knowledge, these TEM observations are the first detection of D nanobubbles in DSSLs, and thus the first experimental validation of the vacancy stabilization process due to trapping of a solute D atom(s) predicted by theoretical calculations [5][6][7]. It should be noted that D bubbles have also been observed in the nearsurface region of beryllium after D ion beam irradiation at 3 and 10 keV [21,22] as well as D plasma exposure at ∼40 eV [23]. In [23], cross-sectional observations revealed bubbles exist below surface cone structures and the bubble diameter was measured to be up to ∼20-30 nm.
SDTrimSP simulations [11,12] were conducted to calculate depth profiles of the number of stopped (implanted) D atoms, as exemplified in figure 2(a) for E i = 45 eV. In the calculations, the W density was assumed to be 100%, i.e. any porosities and previously implanted D atoms were not taken into account, while a large porosity (see figure 1) and a high density of D (see figures 3 and 4 below) in the DSSLs were observed in this study. Also, possible volumetric swelling is not considered here, which can be induced by high-density D nanobubbles. This effect will be examined in the future by measuring a change in the thickness of a thin W deposited layer with a thickness of ∼50-100 nm on a substrate (e.g. molybdenum) before and after DSSL formation. Figure 2(b) plots the measured ∆t DSSL , as indicated in figure 1, as a function of E i , together with a value at E i = 215 eV from [2], which are compared to SDTrimSPcalculated implantation depths of D in W. Our measured ∆t DSSL is found to decrease with decreasing E i , and to agree nearly with the maximum implantation depth or around the 0.1 × peak concentration depth in the present plasma exposure condition.
If ∆t DSSL follows the short-dashed curve at E i < 45 eV, the threshold E i for the DSSL formation is expected to be substantially below 45 eV. Although ∆t DSSL from [2] deviates from the dashed curve, it must be noted that the DSSL in [2] formed at a different plasma exposure condition (T s = 300 K, Γ i = 9.9 × 10 19 m −2 s −1 , and Φ i = 6 × 10 24 m −2 ), and its thickness was measured with the Ar-sputter NRA (∼3 nm depth resolution) technique. However, this might also suggest the saturation of ∆t DSSL .

D retention in DSSL
The amount of D retained in the subsurface region is quantified using NRA. Based on the D amount and the measured ∆t DSSL , the E i dependence of the D retention, previously measured with LIBS [4], is updated here. Note that the lower subscripts, Dy, St, and DySt, indicate dynamic, static, and total (the sum of Dy and St) in the following. A depth, s, profile of the D/W St atomic fraction in a W sample exposed at E i ∼ 75 eV is plotted in figure 3. The D areal density, n D , is ∼4.2 × 10 19 m −2 , assuming that the D profile is uniform across the 14 nm thickness limited by the depth resolution. This converts to D/W St ∼ 0.047. However, since the measured ∆t DSSL at E i ∼ 75 eV is around 7.5 nm, the D/W St atomic fraction becomes ∼0.089 in the DSSL. Here, two assumptions were made: (1) all D atoms (∼4.2 × 10 19 m −2 ) exist within the first 7.5 nm, and (2) the W concentration is 100 at.%, i.e. the bulk W density of 6.3 × 10 28 m −3 , while the DSSL is observed to be porous with high-density D nanobubbles. Thus, both values are thought to be overestimated. Note that this D/W St atomic fraction of ∼0.089 is consistent with ∼0.094 at ∆t DSSL ∼ 10.4 nm in [2].
In the measured depth of 3.1 µm, the total n D is around 1.1 × 10 20 m −2 . Thus, the D amount in the DSSL is around 38% of the total D, which indicates the high D trapping efficiency of the DSSL. The D retention in the depth from 14 to 1500 nm may be caused by dislocations [24], which may be induced by self-interstitial W atoms created in, and migrated from, the DSSL. Since vacancies in the DSSL are stabilized with a solute D atoms(s), self-interstitial W atoms should diffuse somewhere. Likewise, the D retention may also be due to the evolution of blisters, as observed in e.g [25].
The total n D of ∼1.1 × 10 20 m −2 corresponds to ∼0.4 × 10 −4 of Φ i . This means that the majority of the incident D is not retained/trapped in the material. First of all, around 65% of the incident D is reflected from the W surface at E i ∼ 75 eV [1]. Next, solute D atoms, i.e. dynamic retention, will diffuse to the surface and be released as D 2 after plasma exposure. In addition, a DSSL may enhance the release of D from the surface during plasma exposure because of its porous structure as presented in figure 1. Namely, once bubbles are interconnected, D, as well as D 2 possibly formed in bubbles, can easily reach the surface.
In our previous analysis of the LIBS data [4], the following assumptions were made to convert a line intensity ratio (D I 656.1 nm/W I 429.4 nm) into a D/W atomic fraction: (1) a constant ∆t DSSL = 10 nm [2] over a range of E i ∼ 45-175 eV  [26] at s = 10-350 nm, i.e. up to the laser ablation depth, at E i ∼ 75 eV. As a result, no E i dependence of all the D/W components (Dy, St, and DySt) was obtained. Now, we apply the measured E i dependence of ∆t DSSL and the measured D/W St depth profile to the procedure in [4]. First, D/W St in figure 3 is integrated from s = 0-350 nm, which yields ∼1.24 × 10 −9 . Then, it is assumed that D/W Dy is constant inside the DSSL (s = 0-7.5 nm), and dynamic D retention is negligible beyond the DSSL (s = 7.5-350 nm) because  of the low solubility [27,28]. In the end, D/W Dy is obtained from, with (D I/W I) Dy /(D I/W I) St ∼ 0.76 at E i ∼ 75 eV. Thus, D/W Dy ∼ 0.13, which is higher than D/W St ∼ 0.089 unlike the previous analysis in [4]. We should note that the dynamic retention component is considered to include weakly trapped D with a de-trapping energy less than 1 eV [5,6], in addition to solute D, as discussed in [4]. Using a conversion factor from D I/W I to D/W calculated at E i ∼ 75 eV for each retention component, the E i dependence of D/W is updated also by taking into account the E i dependence of ∆t DSSL . Figure 4 demonstrates that all D/W components increase with decreasing E i due to the thinning of the DSSL. During D plasma exposure, D/W DySt reaches ∼0.3 at E i ∼ 45 eV, and is expected to further increase at E i < 45 eV.
As observed in figure 1, more severe damage is created at higher E i , which may cause a higher D retention. However, the obtained E i dependence shows the opposite trend. Considering the simple assumptions made for the LIBS data, the E i dependence of, at least D/W St , should be verified using other diagnostic techniques, including NRA and secondary ion mass spectrometry (SIMS).

Possible DSSL formation mechanism
Lastly, we propose a possible DSSL formation mechanism based on the experimental findings presented here, as well as theoretical predictions and the mechanism proposed in [2].
Our proposed DSSL formation mechanism is schematically represented in figure 5. Each step is first briefly described below: (1) Implanted D atoms in solution (interstitial) sites reduce the vacancy formation energy [29], and also stress the W matrix, reducing the threshold displacement energy [30], (2) An energetic D + ion collides with a lattice W atom, surrounded by solute D atoms, (3) A (temporary or stable) Frenkel pair is created, (4) The vacancy is stabilized by trapping a solute D atom(s) before recombining with the self-interstitial W atom [5][6][7], (5) Clustering of VD j complexes occurs [29], resulting in the formation of D nanobubbles. (6) The volume increases, stresses are further applied, and the threshold displacement energy is further reduced [30].
Concerning steps (1) and (6), a single H (or D) interstitial atom in a W matrix is predicted to reduce the vacancy formation energy by ∼25% from 2.95 to 2.23 eV [29]. This is highly expected, because of the high D/W Dy ∼ 0.1-0.2, as plotted in figure 4. It should be noted that, while the reduced vacancy formation energy of 2.23 eV with a single H interstitial atom is higher than E D->W ∼ 1.9 eV at E i ∼ 45 eV, the predicted trend is favorable to the DSSL formation. Another theoretical calculation predicts that tensile strains, leading to a volume increase, reduce the threshold displacement energy for W as well as for Mo and V [30]. For instance, a volume increase by 5.5% results in a reduction of the threshold displacement energy for W by ∼25%. This process is also likely to occur, as highdensity D nanobubbles, observed in this work, can increase the volume.
In step (3), the formation of either a temporary or stable Frekel pair depends on the effective or reduced threshold displacement energy. The vacancy stabilization in step (4) is predicted in theoretical calculations [5][6][7], and is experimentally validated in this work by observing high-density D nanobubbles, which are VD j complexes and/or V i D j clusters in step (5). This sequential mechanism from step (1) to (6) is considered to be a positive feedback loop of the DSSL formation, since the higher the density of D nanobubbles, the lower the threshold displacement energy. In addition, solute D atoms are continuously supplied by plasma exposure.

Conclusion
To gain more insight into the properties of DSSLs formed in W, we have systematically conducted experiments in PISCES-A, focusing on the E i dependence in this study. W samples were exposed to D plasmas at E i ∼ 45-175 eV, while other plasma exposure parameters, such as T s ∼ 423 K, Γ i ∼ 1.2 × 10 21 m −2 s −1 , and Φ i ∼ 3.0 × 10 24 m −2 , were fixed.
The high-resolution, cross-sectional, TEM observations revealed the DSSL formation in this E i range, even at E i ∼ 45 eV, which is substantially lower than the previously measured (115 eV) and predicted (93 ± 23 eV) values in [2]. It was also found that ∆t DSSL increases from ∼5-6 nm at ∼45 eV to ∼11-12 nm at ∼175 eV, which is almost consistent with the calculated maximum implantation depth in the present condition. With extrapolation of the curve, the threshold E i for the DSSL formation is likely to be even below 45 eV.
High-density D nanobubbles with a diameter of ∼1 nm or less were clearly observed inside the DSSLs formed at all E i . Since D nanobubbles are thought to be VD j complexes as well as V i D j clusters, our observations experimentally validate the theory-predicted vacancy stabilization process due to trapping a solute D atom(s), which plays one of the essential roles in the DSSL formation.
Our previous LIBS data on the E i dependence of the D/W atomic fraction in DSSLs [4] was updated with the NRA-measured D depth profile and ∆t DSSL determined from the TEM observations. All D/W components were found to increase with decreasing E i due to the thinning of the DSSL. This E i dependence will further be confirmed using NRA or SIMS. It is worth noting that D/W DySt reaches ∼0.3 at E i ∼ 45 eV during plasma exposure, and it is expected to become even higher at E i < 45 eV. Such a high D/W atomic fraction during plasma exposure may affect plasma-material interaction processes, such as recycling [31,32] and sputtering [33,34], in particular, in detached plasma conditions with a low T e and thus a low E i .
Our proposed mechanism for the DSSL formation consists of multiple steps, which are thought to make a positive feedback loop. The key processes are supported by the theoretical calculations, which are consistent with our experimental findings.
Trace impurities such as C + and O + can exist in our plasmas. However, the maximum transferred energy of C (O) to W at E i ∼ 45 eV is around 10 (13) eV, which is lower than E d ∼ 40 eV. This also indicates that E d needs to be effectively reduced by, e.g. D interstitial atoms [29] and stresses [30] as predicted by the theories, even for C and O impurities to displace a lattice W atom.
In a magnetic confinement fusion device, one can also expect that a DSSL forms in the surface of W plasma-facing materials. The plasma exposure parameters in this study are similar to those in the first wall region, rather than in the divertor, of e.g. ITER [35]. It should be noted that a DSSL is prone to more easily form at a higher Γ i , corresponding to the divertor region, as predicted in [7]. On the other hand, T s is higher in the divertor, reducing the solute D concentration, which would impede the DSSL formation. Thus, the Γ i and T s dependence of the DSSL formation should be clarified. Also, the DSSL growth needs to be faster than the erosion rate, as impurities are often seeded into the plasma for radiation cooling to spread heat fluxes in a confinement device. We should note that the removal of D and T (tritium) from a D/TSSL seems to be easy, e.g. with glow discharge cleaning between plasma discharges, since it is a thin (∼10 nm) surface layer.