Measurement of the high-field Q drop in a high-purity large-grain niobium cavity for different oxidation processes

The most challenging issue for understanding the performance of superconducting radio-frequency (rf) cavities made of high-purity (residual resistivity ratio $g200$) niobium is due to a sharp degradation (``$Q$-drop'') of the cavity quality factor ${Q}_{0}({B}_{p})$ as the peak surface magnetic field (${B}_{p}$) exceeds about 90 mT, in the absence of field emission. In addition, a low-temperature ($100\char21{}140\ifmmode^\circ\else\textdegree\fi{}\mathrm{C}$) in situ baking of the cavity was found to be beneficial in reducing the $Q$-drop. In this contribution, we present the results from a series of rf tests at 1.7 and 2.0 K on a single-cell cavity made of high-purity large (with area of the order of few ${\mathrm{cm}}^{2}$) grain niobium which underwent various oxidation processes, after initial buffered chemical polishing, such as anodization, baking in pure oxygen atmosphere, and baking in air up to $180\ifmmode^\circ\else\textdegree\fi{}\mathrm{C}$, with the objective of clearly identifying the role of oxygen and the oxide layer on the $Q$-drop. During each rf test a temperature mapping system allows measuring the local temperature rise of the cavity outer surface due to rf losses, which gives information about the losses location, their field dependence, and space distribution. The results confirmed that the depth affected by baking is about 20\char21{}30 nm from the surface and showed that the $Q$-drop did not reappear in a previously baked cavity by further baking at $120\ifmmode^\circ\else\textdegree\fi{}\mathrm{C}$ in pure oxygen atmosphere or in air up to $180\ifmmode^\circ\else\textdegree\fi{}\mathrm{C}$. These treatments increased the oxide thickness and oxygen concentration, measured on niobium samples which were processed with the cavity and were analyzed with transmission electron microscope and secondary ion mass spectroscopy. Nevertheless, the performance of the cavity after air baking at $180\ifmmode^\circ\else\textdegree\fi{}\mathrm{C}$ degraded significantly and the temperature maps showed high losses, uniformly distributed on the surface, which could be completely recovered only by a postpurification treatment at $1250\ifmmode^\circ\else\textdegree\fi{}\mathrm{C}$. A statistic of the position of the ``hot spots'' on the cavity surface showed that grain boundaries are not the preferred location. An interesting correlation was found between the $Q$-drop onset, the quench field, and the low-field energy gap, which supports the hypothesis of thermomagnetic instability governing the $Q$-drop and the baking effect.

and spatial distribution of the rf losses. Niobium samples have been treated together with through some of the cavity and they have been analyzed with TEM and SIMS to further understand the modifications of the oxide layer and of the impurities (oxygen and hydrogen) distribution near the surface due to such treatments.
The paper is organized as follows: a description of the experimental setup and of the typical cavity preparation procedures is given in Sect. II, while Sect. III presents the experimental results after anodization treatment, which converts the niobium metal into Nb 2 O 5 , therefore shifting the oxide/metal interface deeper inside the metal. Section IV shows the results on samples and cavity measurements after baking in pure oxygen atmosphere, in air at 120 °C, 150 °C and 180 °C, and after "in-situ" baking for shorter time. The purpose of the tests was to change the oxygen concentration near the surface and to find a correlation with the Q-drop. Section V shows the cavity test results after degassing of hydrogen, which may have entered the niobium during the previous heat treatments, and after a post-purification with Ti to remove also oxygen and nitrogen. The analysis of the experimental data is presented in Sect. VI, where various aspects such as the medium field Q-slope, the relationship between the energy gap and the Q-drop, the field dependence and spatial distribution of hot-spots heating, hot-spots at grain boundaries and anomalous losses after 180 °C air baking are discussed. Section VII provides an overall discussion of the experimental results, while conclusions are summarized in Sect. VIII.

II. EXPERIMENTAL SETUP
The cavity used for this experimental study has the same shape as the one used in the CEBAF accelerator [8]. The resonant frequency of the TM 010 mode is at 1.47 GHz for a center cell with beam pipe and the ratios of peak surface electric and magnetic fields to the accelerating field are E p /E acc = 1.78, B p /E acc = 4.43 mT/(MV/m) as calculated with the SUPERFISH code [9]. The cavity was built by standard deep drawing and electron beam welding fabrication methods, starting from two 3.175 mm thick Nb sheets, which were saw-cut from a large-grain ingot made by Ningxia, China. The residual resistivity ratio (RRR) of the sheets was approximately 330 and the Ta content was less than 150 wppm, as specified by the manufacturer.
The standard cavity preparation for rf testing consists of The cavity is attached to a test stand and evacuated to about 10 -7 mbar prior to cool-down to 1.7 K.
Before insertion in the vertical cryostat, a temperature mapping system consisting of an array of 576 100 Ω Allen-Bradley 5 × 1.5 mm 2 carbon resistors divided in 36 printed circuits boards spaced by 10°, is assembled on the cavity using Apiezon N grease. Details on the temperature mapping system can be found in Ref. [10].  Figure   1 shows the temperature mapping system as assembled on a single-cell cavity. During each rf test, the surface resistance R s0 at low field (B p ∼ 10 mT) is measured as a function of temperature between 4.3 K and 1.7 K; it is described by the following formula: where R 0 res is the temperature-independent residual resistance and R BCS0 is the low field Bardeen-Cooper-Schrieffer (BCS) surface resistance. The data are fitted to Eq. (1) with the code described in [11], based on the BCS surface impedance code developed by Halbritter [12]. The fit parameters are the mean free path, l, the energy gap at 0 K, Δ, and the residual resistance. Here the critical temperature T c = 9.25 K, the London penetration depth λ L = 42 nm and the coherence length ξ 0 = 31 nm were used as the material parameters for Nb. Measurements of Q 0 vs. B p and temperature maps at various field values are taken at 1.7 K and 2.0 K using a phase-locked loop rf system. The typical measurements errors are 3 -5% for the surface field and 7 -12% for the quality factor.
The temperature map shows the difference, ΔT(x,y) = T(x,y) -T 0 , between the local temperature T(x,y) measured by each resistor on the outer cavity surface and the He bath temperature T 0 . Each resistor is calibrated between 4.3 K and 1.7 K for every test; the He bath temperature in the cryostat is held constant by a pressure regulator during the Q 0 vs.
B p measurement and is measured by two calibrated germanium resistors near the cavity.
The ΔT measured by each thermometer is related to the power dissipated per unit area, dP/dA, as follows: where η is the thermometer efficiency, R K is the Kapitza thermal resistance at the Nb/He interface, d is the cavity wall thickness, κ is the thermal conductivity of Nb and K T represent the temperature response of a thermometer to a unit of uniform power flux.
dP/dA can be expressed as: where R E s is a surface resistance due to dielectric losses, Z 0 is the impedance of vacuum, E and H are the local surface electric and magnetic field, respectively. and parameters can be found in [13]. Another set of samples was analyzed at Jefferson Lab using a refurbished SIMS system [14] with argon primary ion beam and detection of positive secondary ions. The impact energy for Ar + was 3 keV with normal incidence angle. This corresponds to approximately the same projected range as for the system at NCSU [15]. The argon beam was rastered over a 3 mm × 2.5 mm area (with current density 85 μA/cm 2 ) and the sputtering rate is ∼ 2.5 nm/min. The base pressure of the SIMS chamber was ~ 7.5×10 -9 Torr and the measurements were taken on at least two sites for each sample.

III. ANODIZATION
Anodization was commonly used for the preparation of niobium cavities in the 1970s [16], allowing to electrochemically grow an amorphous Nb 2 O 5 film whose thickness depends mainly on the voltage applied between the cavity (anode) and a cathode. For this study, the cavity is filled with NH 4 OH (15%), and a constant voltage is applied between the cavity and a Nb rod inserted in the cavity. The current is limited to about 1 A (∼ 1mA/cm 2 ) and decays exponentially with time as the oxide thickens. The power supply is switched off when the current drops to about 100 mA. In this conditions we expect an oxide thickness corresponding to about 2 nm/V [17]. By anodizing successively at increasing voltages a cavity which was previously "in-situ" baked, we convert some of the niobium into oxide until eventually the Q-drop may re-occur. This could give some indications of the depth affected by the low-temperature baking. This A new series of test was repeated at 1.7 K and 2.0 K in the same sequence as the previous one and the Q 0 vs. B p curves at 1.7 K are shown in Fig. 2. The current vs. time was measured during each anodization step, which allowed us to calculate the total charge transfer, and therefore the thickness of the Nb 2 O 5 film (indicated between parentheses in the legend of Fig. 2), using Faraday's law. The quench field after baking was 5% -12% lower than in the first series of tests. Some field emission was present only in the test after the first "in-situ" baking, starting at E p = 45 MV/m, and caused an increase of the residual resistance by about 7.5 nΩ, which reduced the Q 0 at low field in the subsequent tests. This effect, which had been observed in the past [19], was related to the radiation damage of the oxide layer [20]. Temperature maps for some of the tests, taken at the highest field achieved, are shown in Fig   mT, in their case, and of 152 mT in our cavity, although some additional losses can be seen in Fig. 3(d).
Results from the low-field measurements of R 0 res and Δ/kT c (k is the Boltzmann constant) from both series of tests are shown in Fig. 4 and Fig. 5, respectively.
Anodization up to 123 nm thick oxide increases the residual resistance by 1 -4 nΩ while the ratio Δ/kT c is progressively reduced down to similar values obtained after BCP treatment. "In-situ" baking at 120 °C for 12 -20 h increases R 0 res by 3 -5 nΩ while increasing Δ/kT c by 7 -10%. HF rinse doesn't change Δ/kT c significantly but reduces R 0 res by 3 -15 nΩ. "In-situ" baking at 120 °C reduces also the BCS surface resistance at

IV. BAKING IN PURE OXYGEN ATMOSPHERE AND IN AIR
A. Cavity test results

Baking in pure oxygen atmosphere
In order to further investigate the role of oxygen in causing the Q-drop we baked the cavity at 120 °C in 1 atm of pure oxygen as follows: the cavity was removed from the vertical test stand, placed inside the baking oven and connected to a vacuum line. Two fine-grain (50 × 50 μm 2 average grain size) Nb samples were also inserted. The cavity was evacuated to 6×10 -9 mbar (water was the main residual gas), backfilled with research grade (99.9969% purity) oxygen up to 1 atm and baked for 12 h. A relief valve limited the oxygen pressure to 1.1 atm during baking. Afterwards the Nb samples were stored in air while the cavity was high-pressure rinsed, re-assembled, evacuated, the temperature mapping system was assembled and the test stand with the cavity was finally inserted in the vertical cryostat for the rf test. There was no significant change of performance from the previous rf test (HF rinse) both at 1.7 K and 2.0 K: the residual resistance was unchanged, Δ/kT c and quench field (which occurred at the same location) increased by about 2% and 5%, respectively. The temperature map at the breakdown field also did not show significant variations from the test before. Then the experiment was repeated and the duration of baking was increased to 48 h. As a result, Δ/kT c increased by about 3%, the quench field and location did not change. The Q 0 vs. B p plot at 1.7 K after baking for 48 h is shown in Fig. 6 while the temperature map at 142 mT is shown in Fig. 7a.

Baking in air at higher temperature
The results from the oxygen baking tests suggested that the heating temperature may have not been high enough to diffuse oxygen through the oxide into the niobium.
Therefore we decided to bake the cavity in air as follows: the cavity was removed from the test stand, the beam pipes were opened and the cavity was degreased. The cavity was then rinsed with ultrapure water and, while still wet, two niobium discs were clamped to close the beam pipe openings. Two fine-grain niobium samples were placed on the bottom disc inside the cavity. The cavity was set vertically in a frame inside the oven and baked. After baking, the cavity was degreased, high-pressure rinsed, re-assembled, evacuated, the temperature mapping system was assembled and the test stand with the cavity was finally inserted in the vertical cryostat for the rf test. It is worthwhile to notice that the thermometers' thermal response coefficient K T in Eq. (2), obtained from a fit of the low-field dissipated power from rf measurements with the power dissipated on the cavity walls from the temperature maps a , increased progressively with higher baking temperature by up to more than one order of magnitude.
After baking at 180 °C, the outer surface of the cavity looked dark gray, despite the nitrogen atmosphere on the outside, suggesting a possible increase of the Kapitza resistance due to oxidation.
In order to determine whether the additional losses were due to the oxide layer, the cavity was rinsed with HF (49%) and re-oxidized in water for a total of six times. In a first attempt, the cavity was rinsed with HF for 5 min then filled with ultra-pure water for 1 h, followed by a second HF rinse and then prepared for the rf test using the standard procedure of Section II, starting with high-pressure rinse. The test results showed a reduction of R 0 res down to the value prior to air baking (∼ 20 nΩ) and of the intensity of some hot-spots, but no change in the quench field and location and in the Q 0 vs. B p dependence. The HF rinse procedure was repeated for four more times and the only change was the location of the quench (from 180-9 to 130-12) and of some of the hota The efficiency of the thermometers is assumed to be the same for all of them, the surface resistance is assumed to be field independent and dielectric losses are neglected.
spots. The Q 0 vs. B p curve after the six HF rinses is shown in Fig. 6 and the temperature map at B p = 117 mT is shown in Fig. 7d. Although the additional losses introduced by the air baking at higher temperature did not have the same field dependence as for the Qdrop, we applied the standard "in-situ" UHV baking procedure at 120 °C for 12 h to investigate whether they could also be reduced, but we did not succeed (The Q 0 vs. B p curve is shown in Fig. 6). We tried to evaluate whether enough hydrogen from the wet surface could penetrate through the oxide layer to cause the so-called Q-disease [22] during the baking at 180 °C. In order to do that, we hold the cavity in the cryostat at a temperature of 100 K for 40 h, which would allow the precipitation of a lossy niobium hydride phase, if the hydrogen concentration was above about 5 wppm. The subsequent rf tests at 1.7 K and 2.0 K did not show any reduction of the low-field Q 0 nor any variation of the Q 0 vs. B p dependence from the previous test. Figure 8 shows the variations of R 0 res and Δ/kT c for the various tests discussed in this section.

Effect of baking duration
We also investigated the evolution of the hot-spots by "in-situ" baking for shorter baking at 120 °C for 3 h and after another chemical etch followed by air baking with the same parameters as for the "in-situ" one.

A. Niobium samples results
The results from the TEM measurements of the oxide thickness on the fine-grain polycrystalline Nb samples for different baking treatments are shown in Table I. The sample treatments prior to baking are described in Sect. II. Figure 11 shows the TEM micrograph for the sample W3. There was no significant increase of the oxide thickness for the sample baked at 120 °C for 12 h in 1 atm of oxygen with respect to the reference sample, while the increase was 0.3 ± 0.1 nm for the second oxygen bake, for longer time (48 h). More significant increase, up to 3.3 ± 0.2 nm is observed after air baking at higher temperatures. The TEM pictures showed no identifiable suboxide regions.   Table I.  Table I.

V. HYDROGEN DEGASSING AND POST-PURIFICATION
In order to investigate the cause for the increase in the number of hot-spots introduced by the air baking at 180 °C and which were not significantly reduced by additional chemical etching, we wanted to evaluate the influence of interstitial impurities by heat treating the cavity in a vacuum furnace in two steps: at 600 °C first and at 1250 °C with titanium second. The heat treatments at 600 °C for 10 h or 800 °C for 3 h remove interstitial hydrogen from niobium [23] and is commonly applied to niobium rf cavities to prevent Q-disease. For the second heat treatment, the cavity is placed in a titanium box and heated to 1250 °C. A titanium layer is evaporated on the cavity surface and acts as a getter for impurities such as oxygen, nitrogen and hydrogen. In niobium the diffusion velocity of oxygen between 1100 °C and 1300 °C is of the order of millimiters per hour, approximately six to seven times faster than for carbon and nitrogen. Therefore removal of oxygen is more likely in this temperature range [24]. Hydrogen gas released by niobium at high temperatures is absorbed readily by Ti between 25 °C and 400 °C [25]. baking at 120 °C for 3 h and the cavity quenched at B p ≅ 165 mT. An additional "in-situ" bake at 120 °C for 6 h eliminated the Q-drop up to the quench field (165 mT, at location 30-9). As it can be seen from the temperature maps after post-purification shown in Fig.   17c, it was only after this treatment that most of the cavity surface showed no distributed rf losses but only a few hot-spots in the equator region, similar to the map showed in Fig. 3b. In addition, it was only after post-purification that the value of the thermometers' thermal transfer coefficient, K T , was reduced to ∼ 8 K cm 2 /W, as before the air baking at high temperatures. Post-purification also restored the high quench field value which was obtained in the first tests, described in Sec. III. The Q 0 vs. B p curves at 1.7 K before and after baking are shown in Fig. 15. The temperature maps at the highest field before and after the second baking are shown in Figs. 16c and 16d, respectively.

Medium field Q-slope
The quality factor of superconducting bulk niobium cavities has a mild degradation between B p = 20 -90 mT, which is referred to as "medium field Q-slope".
The following field dependence of the surface resistance, R s , provided a good fit of a large set of experimental data on fine-grain, large-grain and single-crystal niobium cavities [26]: Here R s0 , γ * and R 1 res are regarded as positive fitting parameters. B c = 200 mT is the thermodynamic critical field at T = 0 of niobium and T 0 is the He bath temperature. The parameter γ * is determined by the thermal feedback provided by the BCS surface resistance and thermal impedance of the outer cavity surface [26] and by nonlinear pairbreaking [27]. The coefficient R 1 res can be due to losses produced by Josephson fluxons penetrating grain boundaries [26] or by flux flow of Abrikosov vortices oscillating at the surface. The weighted average values of R s0 , γ * , R 1 res and of the fit correlation factor r 2 from R s vs. B p fits at 1.7 K and 2.0 K obtained for a standard BCP treatment and "in-situ" baking at 120 °C for 3 -12 h are summarized in

Relation between the energy gap value and the cavity performance
The weighted average values of R 0 res , Δ/kT c and l obtained from the fits of the temperature dependence of the low field surface resistance with Eq. (1) for data after BCP treatment and after "in-situ" baking at 120 °C for 12 h are indicated in Table III. In addition, the results from the tests described in Sec. II showed that anodization and baking are two surface treatments which alter the energy gap in opposite ways: anodization of a baked cavity surface at increasing voltages progressively reduces the gap, while baking at 120 °C for longer times progressively increases it. The value of the energy gap for bulk Nb is approximately 1.6 meV [20], correponding to a ratio Δ/kT c of 2.007 b . The value of Δ/kT c obtained from the R s (T) fits is an average over a depth of the order of few times the penetration depth (∼ 40 nm between 4.3 K and 1.7 K). The lower value of Δ/kT c after BCP may suggest a space dependence (sketched in Fig. 17) characterized by a thin layer with depressed superconductivity near the surface, such as considered in the description of proximity effects [28] , which baking is able to reduce.  Vortices will then enter in the niobium at this field, which is low enough to keep the cavity thermally stable, and cause a reduction of Q 0 (Q-drop), for example by flux flow losses. The gap-value is raised by baking and so is the critical field and, in this case, vortices entering the niobium at higher field cause a thermal breakdown of the cavity.  6) with H p0 and Δ 0 as fit parameters.

Field dependence and spatial distribution of hot-spots heating
The field dependence of the heat generated at hot-spots on the cavity surface, measured by thermal maps, can be analyzed by plotting Log(ΔT) vs. Log (B p ). Neglecting dielectric losses, the logarithmic slope, n = Log(ΔT)/Log (B p ) is related to the field dependence of the surface resistance, where n = 2 represents "ohmic-type" losses for which R s is independent of B p . Since the hot-spots usually occur in the equator region of the cavity where the local surface magnetic field is within 3% of B p for thermometers at locations 5 to 12 along the meridian, the ΔT from thermometers within this region was plotted as a function of B p . Figure 19 shows the field dependence of Log(ΔΤ), which was typically observed at hot-spots after a chemical etching and after "in-situ" baking at 120 °C, along with a linear fit and the value of n. At the onset of Q-drop the dissipated power at hot-spots locations increases dramatically, as shown by the increase of n by about one order of magnitude, and "in-situ" baking at 120 °C strongly reduces these anomalous losses. The ΔT(B p ) dependence for hot-spots at fields lower than the onset of Q-drop does not differ significantly from the dependence measured on other locations of the cavity. In many cases the location of hot-spots causing the Q-drop did not correspond to the locations which showed higher losses at lower fields, due for example to higher local residual resistance.  Table IV. FIG. 20. Q 0 vs. B p data measured after post-purification (squares) and after 120 °C, 3h "in-situ" baking fitted with Eq. (7) (solid lines). The data were also shown in Fig. 15. The fact that the parameter g obtained from the fits is almost the same for 1.7 and 2K is consistent with the model [27] in which g is mostly determined by the spatial distribution of the hotspots. It has been shown in [29] that the spatial distribution of hot-spots heating during the Q-drop is not compatible with the hypothesis of a hot-spot being a small normal-conducting defect. Figure 21 shows the ΔT-distribution along the meridian for thermometers neighbouring a hot-spot, normalized to the temperature at the hot-spot location, for increasing values of rf field in the Q-drop regime. The data show an expansion of the area affected by hot-spot heating for increasing rf field and no significant difference was found between the distribution at 2.0 K and 1.7 K. This behavior was predicted by the same model [27].

Grain boundary heating
In order to investigate the role of grain boundaries on the Q-drop, we recorded after each rf test the location of hot-spots with respect to grain boundaries, which are visible by eye in large-grain cavities. The temperature mapping system was always assembled with the same orientation and a fraction of 196/576 thermometers was near (within 1 cm 2 around a thermometer) or on a grain boundary. We present the statistics on the hot-spots location for a total of 33 rf tests. Hot-spots occurred at 238 different locations and a fraction 89/238 (37%) was near or on a grain boundary. The highest number of times that a hot-spot occurred at the same location was seven. There were nine locations where the incidence of hot-spots was greater or equal than five. A fraction 5/9 (56%) of these locations were near or on a grain boundary. If we limit the statistic to hotspots which correlate with Q-drop (n > 4 for B p > 90 mT), they were found at 171 different locations and a fraction 52/171 (30%) was near or on a grain boundary. The highest number of times that these hot-spots occurred at the same location was five.
There were eight locations where the incidence of hot-spots was greater or equal than four. A fraction 4/8 (50%) of these locations were near or on a grain boundary. In conclusion, grain boundaries do not seem to be the main source of hot-spots, in agreement with a recent study on cavities with different grain size [30].
In the rf tests after post-purification it was found that the position of hot-spots changes even after "in-situ" baking at 120 °C for 3 h. We wanted to verify whether this could be associated with the diffusion of some impurity sideways, along a grain boundary, rather than in depth, but the patterns of three major hot-spots in Fig. 16c did not show clear evidence for it. Figure 22 shows, for example, a temperature map limited to few thermometers surrounding a hot-spot with the grain boundaries pattern superimposed for the rf tests after post-purification and after "in-situ" baking at 120 °C for 3 h and 6 h. after additional 120 °C, 6 h "in-situ" bake. The thick red lines represent grain boundaries and the thick black line represents the equator weld.

Heating after 180 °C air bake
Baking at 180 °C for 12 h in air caused uniformely distributed "anomalous" losses and the overall surface resistance of the cavity exhibited a strong quadratic dependence from the rf field, with coefficient γ * at 1.7 K approximately a factor of five higher than after standard "in-situ" baking. In addition, the thermal response of the thermometers increased by about a factor of ten and the outer cavity surface looked dark gray. All these may suggest an increase of the Kapitza resistance. On the other hand, baking of a cavity at 180 °C for 48 h in UHV conditions did not cause the strong R s (B p ) dependence observed in this study [4]. This may suggest that the air baking introduced a large quantity of interstitial oxygen, as shown by SIMS, which may precipitate in metallic, normal conducting NbO x (x ∼ 1) suboxides, causing high, uniformely distributed rf losses, as shown by the temperature maps. These enhanced rf losses, combined with a higher Kapitza resistance, cause a warm-up of the cavity rf surface by thermal feedback, causing the quadratic R s (B p ) dependence obtained from the rf measurement. Another unusual result from the high-temperature air bake was the presence of high losses in the iris region of the cavity and could be explained by enhanced dielectric losses. Evidence for dielectric losses was previously reported in [31]. Figure 23 shows, for example, the ΔT distribution at low field (B p = 16 mT) along a meridian, from the cavity equator to the iris. The data could be fitted with Eq. (2) with R s decreasing from 10 nΩ at the equator to For an oxide thickness, d, of 10 nm, and a relative dielectric constant, ε r , of 30, R E s = 15 nΩ yields tanδ E = 2.6×10 -5 , well within the range of measured values [20]. In the hotspots areas at the iris, R E s is approximately a factor of ten higher then in other regions of the cavity, which could be due to an increased relative dielectric constant or loss tangent due to higher defect density in the oxide.

VII. DISCUSSION
The results from the anodization experiments confirm that the changes produced by the low-temperature "in-situ" baking occur within about 30 nm deep into the niobium (corresponding to the ∼ 80 nm thick anodic Nb 2 O 5 ). These results, combined with the fact that the baking effect is stable after exposure of the cavity to air for many months and to successive high-pressure water rinsing, exclude the possibility of the Q-drop being caused by adsorbates on the surface.
The increase of residual resistance observed after "in-situ" baking, anodization and electron impact on the oxide could be related in all cases to interface losses, described in Ref. [32], due to oxygen injection and NbO x (x ∼ 1) precipitates formed at the metal/oxide interface.  [4] showed some reduction of the hydrogen concentration near the surface by "in-situ" baking. If hydrogen segregated near the surface is involved, the hydrogen degassing at 600 °C -800 °C alone does not eliminate the Q-drop, because the heat-treatment is followed by a chemical etching where some hydrogen pick-up always occur. "In-situ" baking for longer times, would allow further reduction of the surface hydrogen concentration, therefore increasing the onset of Q-drop.
One important experimental evidence emerging from this study is the correlation between the energy gap of niobium and the onset of Q-drop and quench field. It was shown in Sect. VI.2 that this relation suggests that the Q-drop onset and quench field after baking are given by the combination of the change of the critical field and of the thermal stability of the cavity, with the energy gap acting as main parameter. The lower value of the energy gap obtained by chemical treatment of niobium has been attributed in the past to interstitial oxygen and normal-conducting NbO x precipitates at the metal/oxide interface, formed during the "wet" oxidation of niobium [20]. Those conclusions were infererred mainly from measurements of superconducting properties (penetration depth) of low-RRR niobium samples. To the authors' knowledge there exist no clear evidence for such features by surface analytical techniques. In this study, the resolution of the TEM images is not sufficient to draw definitive conclusions on this aspect. The gap reduction after chemical treatment and anodization does not favor the involvement of hydrogen, since positive (anodic) polarization of the cavity reduces the hydrogen pick-up and changes of the energy gap due to hydrogen were observed only at high enough concentrations to cause Q-disease [37].
Regarding the location of the hot-spots, the grain boundaries provide some, although not dominant, contribution. It was shown in Ref. [38] that the losses due to vortex penetration along strongly coupled low-angle grain boundaries can be almost as low as losses in the bulk. This may explain why some grain boundaries show higher rf losses than others, and many hotspots observed on our thermals maps are not associated with grain boundaries at all.

VIII. CONCLUSIONS
The anodization and HF rinse tests reported in this article confirm previous results on small-grain polycrystalline cavities, for large-grain niobium cavities: "in-situ" baking affects the properties of niobium up to a depth of 20 -30 nm and the growth of a new oxide layer on a baked cavity, after HF rinse, does not restore the Q-drop. The baked cavity was resistant against the occurrence of Q-drop even after oxygen injection by additional bake-outs in a pure oxygen atmosphere and in air, at higher temperatures.
Sufficiently high oxygen concentrations, as achieved by baking in air at 180 °C for 12 h, introduced higher dielectric and interface losses and reduced the thermal transfer efficiency between niobium and the He bath. Temperature maps showed that the only treatment which allowed a complete recovery from these losses was post-purification with Ti at 1250 °C. If the baking temperature is sufficiently low (≤ 120 °C), the oxide dissociation and oxygen injection from the surrounding atmosphere are minimal and a significant improvement of the Q-drop can be achieved by baking in air.
The onset of Q-drop in the large-grain cavity is raised up to B p = 145 mT by "insitu" baking at 120 °C for 3 h and above B p = 165 mT after a total of 9 h, which is shorter than the typical 48 h baking required on fine-grain cavities. This would help reducing the cavity preparation times, which is an important aspect for large-scale cavity production projects. The highest B p -value measured in this cavity corresponds to an accelerating gradient E acc = 38.7 MV/m in a ILC-TESLA shaped cavity and to E acc = 46 MV/m in a ILC-LL shape [39], well above the value of 31.5 MV/m required for the ILC project and comparable to the best fine-grain cavities treated by electropolishing.
A correlation between the onset of Q-drop, the quench field and the energy gap emerges from the data, supporting the hypothesis of thermo-magnetic breakdown. The field and space dependence of the heat generated at hot-spots causing the Q-drop fit this description well. As for the physical origin of this phenomenon, oxygen diffusion does not seem to play a major role. Hydrogen may be involved, although more systematic studies are necessary. The investigation of hydrogen in niobium by conventional surface analytical techniques is made difficult by its high mobility in Nb and by being present as the main residual gas in the analysis chambers. A well-trusted, reliable characterization method and procedure with high enough sensitivity is necessary for clear results. More detailed studies of the oxide/metal interface are also desirable to clearly identify the presence of suboxide clusters and modifications introduced by baking. The investigation of grain boundaries in niobium, their resistance, depairing current density and dependence from the misorientation angle would be helpful in the understanding of loss mechanisms due to vortices of various kinds.