Detection of surface carbon and hydrocarbons in hot spot regions of niobium superconducting rf cavities by Raman spectroscopy

C. Cao, D. Ford, S. Bishnoi, T. Proslier, B. Albee,* E. Hommerding, A. Korczakowski, L. Cooley, G. Ciovati, and J. F. Zasadzinski Physics Department, Illinois Institute of Technology, Chicago, Illinois 60616, USA Materials Science Division, Argonne National Laboratory, Argonne, Illinois 60439, USA Superconducting Materials Department, Technical Division, Fermi National Accelerator Laboratory, Batavia, Illinois 60510, USA Department of Chemical and Biological Engineering, Northwestern University, Evanston, Illinois 60208, USA Thomas Jefferson National Accelerator Facility, Newport News, Virginia 23606, USA (Received 5 November 2012; published 26 June 2013)


I. INTRODUCTION
Superconducting radio frequency (SRF) cavities, having Q values orders of magnitude higher than normal Cu, are an enabling device for a host of linear particle accelerator applications [1].SRF cavities made from high purity Nb plates involve numerous processing steps including deep drawing, welding, acid etching/polishing, two types of vacuum annealing, and high-pressure water spraying.The connections between processing and the surface superconductivity relevant for cavity operation are still not well understood [2].There is a need to develop new analytical techniques that can easily measure the material properties at the nanoscale, i.e., within a magnetic penetration depth, $ 45 nm, of the surface.Here we show that Raman spectroscopy combined with density functional theory (DFT) offers promise as a unique and direct approach to detecting trace levels of surface inclusions such as hydrides and hydrocarbons that may be playing a critical role in the performance of an SRF cavity.
Chemical polishing of the niobium surface is a vital step in the cavity processing procedure which reduces the surface roughness and removes the layer damaged from forming; however, it was recognized several decades ago that chemical contamination is a serious drawback to these procedures [3].Numerous reports describe how cavities tested immediately after heavy chemical polishing developed a rapid drop of Q starting with zero accelerating gradient (so-called Q disease [4]), and the susceptibility of niobium to hydrogen absorption during chemical processing has been established [5].Dissolved oxygen, nitrogen, and carbon are all found in SRF niobium [6], although specifications for the niobium typically restrict their concentrations to <200 atomic ppm.The above-mentioned impurities are problematic for SRF cavities because they reduce the thermal conductivity of niobium, thereby reducing its ability to dissipate heat during rf operation [7].Furthermore, dissolved oxygen [8] and hydrogen [9,10] are known to decrease the superconducting transition temperature of niobium at concentrations of a few atomic percent.
In addition to Q disease, impurities may affect a cavity's performance by Q drop, the reduction of Q with increasing rf fields, which is not yet fully understood.A particular case study of interest is a single-cell niobium SRF cavity, fabricated using a buffered chemical polishing (BCP), that exhibited medium field (20-100 mT) Q drop [11].Localized regions of high dissipation, hot spots, developed as Q decreased, and optical and electron backscattered diffraction imaging showed clearly that such hot spots were correlated with a high areal density of etch pits, especially near grain boundaries.Cold spots had more shallow pits with significantly lower density.We might hypothesize therefore that the dissipation is related to different surface structure and chemistry inside or beneath the pits as compared to other regions of the cavity surface.It has been shown previously that pits are found in conjunction with high dislocation density [12,13] and dislocations trap impurities [14].However, a link between hot spots and the chemistry underlying the etch pits is not yet confirmed.Thus, we obtained several samples from the cavity of Ref. [11] to include in our study.
The standard spectroscopy tools of electron microscopy, namely Auger and energy dispersive x-ray (EDX), showed little difference between the regions inside and outside the pits in preliminary tests [11].One concern is that such spectroscopic tools cannot measure hydrogen, a gas which diffuses easily in Nb and is a well-known contaminant potentially affecting the quality factor Q of the cavity [5].Also, these techniques measure an average composition over a relatively large electron probing depth, >100 nm, and thus would be insensitive to subtle changes in surface oxide composition occurring over $5 nm.This can be significant as it has been shown that <3% oxygen vacancies in Nb 2 O 5 change this compound from a nonmagnetic insulator to a conductor with localized magnetic moments [15].The superconducting gap region probed by tunneling sometimes exhibits a smeared density of states consistent with pair breaking due to magnetic scattering [16].There is thus a good chance that the source of rf dissipation in a pit is tied to the local chemistry (e.g.oxides, hydrides, dissolved gases, interstitial impurities, precipitates) near the surface, which might suppress the superconductivity in a variety of ways and cause increased rf impedance and dissipation.
Raman microscopy/spectroscopy is a technique which has not previously been used for Nb SRF cavity research.Raman is a simple, fast, generally nonperturbative optical method that probes molecular vibrational modes as well as bulk phonons via inelastic scattering.The probing depth in pure Nb ($ 10-20 nm) is estimated from the skin depth of the 785 nm laser and is a reasonable fraction of the magnetic penetration depth [2].Since pure Nb itself is not Raman active, any backscattering signal must originate from surface oxides, dissolved species, or inclusions near the surface.Surface impurities might reduce the electrical conductivity and increase the Raman probing depth.Using a 50X objective lens, the laser spot size is $3 m, which is ideal for examining etch pits that range in lateral dimension from $10 m to $200 m.As will be shown, Raman spectra display striking differences between the inside and outside of certain types of etch pits.A principal discovery is the consistent observation of sharp Raman peaks inside these etch pits as well as in cold-worked regions of Nb foils.This indicates that the Raman peaks are linked to a high dislocation density.The DFT calculations show agreement of some peaks with hydride phases NbH and NbH 2 .However, the overall spectra are more consistent with the C-H modes found in chainlike hydrocarbons [17].Another group of spectra exhibit relatively broader peaks that are identified as amorphous carbon.A reexamination of the hot spot samples and Nb foils by SEM/EDX indeed showed regions with high C and O content.These results suggest that regions of high dislocation density may be trapping migrating C and O as well as dissolved hydrogen with sufficient concentration to form inclusions of stable compounds.However, there is also the possibility that stable hydrocarbons are already present in the Nb as a consequence of processing such as deep drawing.These results provide a reasonable explanation of the correlation of hot spots in SRF cavities with the observation of a high density of etch pits.

II. EXPERIMENT
All Raman measurements were done using a Renishaw, inVia, Raman microscope with a 785 nm laser source, 10 s exposure time, 100% laser power (27 mW), and a 50X objective lens.All SEM/EDX measurements were done using the Hitachi S-4700 FESEM and EDXS acquisition system, with an accelerating field of 10 kV and an emission current of 10-20 A. The probing depth is $1 m.
The types of Nb samples studied included cut-outs from a tested cavity as well as high purity Nb foils.One group of foils had been rolled from a piece of Nb plate used for SRF cavity fabrication and subsequently processed in a similar manner to that used on cavities including electropolishing and BCP etching [2,5,11].In addition we examined foils as obtained directly from Alpha Aesar (as-received noted here as unprocessed).In some cases the foils were resistively heated near the melting point of Nb (2469 C) in high vacuum (10 À7 torr).These latter foils showed single crystal regions up to a few millimeters in diameter and are referred to as recrystallized.Several of these recrystallized foils were subsequently cold worked by bending back and forth ($ 90 angle) from 20 to 50 times.We also have Raman data from six different hot spot samples cut out of a single-cell cavity [11].The hot spots displayed temperature rises $0:3-0:4 K above the He bath temperature at cavity magnetic field strengths of 90 mT.The cavity was made from a large-grain plate with a residual resistance ratio of 280 and was etched by BCP.Its performance was C. CAO et al.
Phys.Rev. ST Accel.Beams 16, 064701 (2013) 064701-2 characterized by a strong medium field Q slope and no field emission.The detailed processing history and cavity performance measurements can be found in Ref. [11].Sample numbers were those assigned at Jefferson Lab and are referred to as, for example, Jlab #12.All of the calculations were performed with the Vienna ab initio Simulation Package (VASP) [18,19], using DFT, periodic boundary conditions, and a plane wave basis set with a 400 eV kinetic energy cutoff.The generalized gradient approximation was used with the Perdew, Burke, Ernzerhof [20] exchange-correlation functional, and the core electrons were described by the projectoraugmented-wave (PAW) [21,22] pseudopotentials.The geometry was optimized for each crystal structure with a 0.25 A ˚gamma centered k-point mesh determined from the Monkhorst-Pack scheme [23].The partial occupancies for the wave functions were set by the 1st order Methfessel-Paxton method with a smearing width of 0.2 eV for metallic structures and the tetrahedron with Blochl corrections for Nb 2 O 5 .The phonon spectra were calculated using the phonon module in the MEDEA software package.This program applies the direct method [24] for calculating vibrational frequencies and uses forces calculated with VASP.Supercells extending at least 10 A ˚in each lattice direction were created to minimize the interactions between equivalent atoms in adjacent cells, and atoms were displaced 0.02 A ˚from their equilibrium positions for the force calculations.The results are presented as the phonon density of states (DOS) at wave numbers where the gamma point frequency is Raman active.We call this the Raman DOS.
The niobium hydride structures were modeled as NbH 2 (Fm-3 m, one formula unit per unit cell), NbH (cccm, four formula units per unit cell), and Nb 4 H 3 (NbH with one H removed from the unit cell containing four Nb and four H).The and 0 niobium hydride phases were modeled as body-centered cubic (bcc) niobium (16 atoms per unit cell) with three and eight hydrogen atoms, respectively, inserted into tetrahedral interstices in disordered configurations.The niobium oxide structures were modeled as NbO (Pm-3m, three formula units per unit cell), NbO 2 (l4-1, eight formula units per unit cell), and Nb 2 O 5 (P2, seven formula units per unit cell).Interstitial impurities in niobium were modeled in a 4 Â 4 Â 4 bcc niobium lattice (128 Nb atoms).Experimentally determined lattice parameters for Nb [25], NbH [26], NbH 2 [27], NbO [28], NbO 2 [29], and Nb 2 O 5 [30] are available in the literature and are <1% different from our calculated values.

III. RESULTS AND DISCUSSION
To introduce our study we show, in the lower right hand panel of Fig. 1, a relatively wide-area optical microscope image of a region on hot spot #9 that displayed a high density of etch pits.The region shown is characterized by three etch pits $20 m on edge near a grain boundary.In addition, other rough patches of similar size to the etch pits are observed.An expanded view of the region near the topmost etch pit is shown in the upper panel of Fig. 1.Representative Raman spectra are shown for the four regions indicated.For smooth regions outside the etch pits (e.g.#1), the Raman spectra are relatively flat and featureless showing only a broad peak near 1350 cm À1 .Nearly identical Raman spectra are found in region #2 which is inside a standard-type, faceted etch pit indicating no apparent differences chemically from surrounding regions as mentioned in Ref. [11].Surprisingly, the grain boundary (#3) shows little difference in the Raman spectra.
However, clear differences in Raman signal are found in regions of the type labeled #4 which appear to be much rougher than surrounding areas, but still in the areas of high etch pit density.Atomic force microscopy images of those regions showed a higher rms surface roughness, typically by a factor of $10-20.The Raman intensity is always enhanced, by as much as 50X (rescaled down for the figure), and there are well-defined sharp peaks in the region of 1000-1500 cm À1 and near 2900 cm À1 .As will be shown, these peaks are quite reproducible, being observed in each of the six hot spot samples, as well as in coldworked Nb foils.In addition, there is another group of spectra characterized by two relatively broad peaks which are again consistently found on rough patches of hot spot samples as well as cold-worked foils.The main focus of the present work is the identification of these Raman peaks.

A. Interstitial impurities
We begin by presenting the DFT modeling results for interstitial impurities in bulk bcc niobium.Common impurities found in SRF niobium (hydrogen, nitrogen, carbon, and oxygen) vibrate at wave numbers below 1500 cm À1 , as shown in Table I.Hydrogen is the only species that appears in the region of the peaks that we wish to identify in our measured Raman spectra.

B. Niobium oxide modes
To test our methods and confirm that our measured Raman spectra are not due to the native oxide coating on niobium, we calculated the Raman DOS of NbO, NbO 2 , and Nb 2 O 5 , and present the results in Fig. 2. NbO does not have any Raman-active modes.We also show our measured spectra for NbO 2 and Nb 2 O 5 powders as validation for the use of calculated Raman DOS for interpretation of Raman spectra [3].Overall there is excellent agreement between the Raman DOS and Nb 2 O 5 powder, indicating the accuracy of the VASP calculations.We note, however, that the intensities of the Nb-O bending modes between 600-700 cm À1 and the double bond stretch mode near 980 cm À1 are higher in the Raman spectra of the powder than in the calculated Raman DOS, probably due to matrix element effects.Modes below 120 cm À1 are not recorded in the Raman measurement.The presence of the 980 cm À1 mode in the NbO 2 powder indicates that oxidation of this powder has occurred after exposure to the air.This is further confirmed by the weak double peak between 600-700 cm À1 , close to the two peaks of Nb 2 O 5 .Thus the NbO 2 Raman spectrum has a significant contribution of Nb 2 O 5 modes and is best interpreted as a mixture of the two oxides.
A comparison of Figs. 1 and 2 shows that Nb oxide modes are generally not detected on SRF cavity pieces or Nb foils.This is probably due to the oxide thickness ( $ 5 nm) giving a weak backscatter signal.However, oxide modes are clearly seen on Nb powders [31] where the effective surface for Raman backscatter can be considerably larger.

C. Niobium hydride modes
The calculation results for various Nb-H complexes, including ordered hydride modes are shown in Fig. 3, along with the measured Raman spectra for a smooth and rough patch of an etched foil.It is clear that several of sharp Raman peaks observed inside pits or rough patches correlate with the phonon modes calculated for the Nb-H complexes.In particular, the NbH and NbH 2 , which have the highest Raman DOS in the calculations, correlate with the strongest observed Raman peaks at 1068, 1303, and 1450 cm À1 .
Since etch pits are a fingerprint of underlying crystal defects such as dislocations, it was hypothesized that the introduction of dislocations into a sample might generate such peaks.To test this hypothesis, several of the recrystallized foils were cold worked by bending back and forth to generate a high dislocation density.The sharp peaks typical of the Raman spectra inside of pits were reproduced throughout the cold-worked region as shown by a representative spectrum (top curve) in Fig. 3.This might suggest that the origin of the sharp Raman peaks is the trapping of impurities by dislocations present near the surface.
Similar experiments were performed on the Jlab hot spot samples.Regions viewed under the optical microscope of the Raman spectrometer appeared either as light or dark.Dark regions often corresponded to etch pits similar in dimension to those reported in Ref. [11].An example of a pit from sample #12 is shown in the photograph inset of    Fig. 4. A significant fraction of the pits showed spectra similar to those found on Nb foils and a comparison of representative spectra is shown in Fig. 4.Each of the six hot spot samples is represented and similar spectra were observed in multiple pits on each sample.There is remarkable agreement among the spectra, which indicates that the development of particular impurity complexes in the subsurface of some pits is a rather common phenomenon.Since hydride inclusions precipitating at a dislocation would likely be nanoscale, there could be significant phonon scattering which would generate disorder induced first order Raman scattering from the entire optical mode branch [32].Considering this possibility, we compared the Raman spectra to the partial DOS of the optical branches that have a Raman-active mode at the zone center.These DOS calculations are shown as the red and green lines for NbH and NbH 2 , respectively.The agreement in location and characteristic width of the calculated and measured Raman peaks is quite good and the interpretation that these were hydride phases was made in an earlier report based on limited data [31].However, such hydride complexes do not explain the peaks near 2900 cm À1 .

D. Hydrocarbon modes
High wave number activity may be indicative of a surface species, and particularly, C-H stretching modes were suspected.A literature search yielded the Raman spectrum of stearic acid [17], which could explain not only the peaks near 2900 cm À1 , but also several of the lower wave number peaks.For comparison with the literature spectrum, we removed the fluorescence background from the cold-worked foil spectrum of Fig. 3.The background was obtained by blocking the peak region and fitting the rest of the data with a polynomial, as shown in Fig. 5.
In Fig. 6, we compared our spectrum with that of stearic acid [17], which is a chain compound with 18 carbon atoms.The Raman peak frequencies and their assignments are shown in Table II  be ruled out [31], however the spectrum as a whole seems to be better explained by chain-type hydrocarbons.
To eliminate possible hydrocarbon contamination, Jlab hot spot samples and all recrystallized foils were subjected to an ozone treatment [33].The treatment was performed in an atomic layer deposition system at 100 C with five cycles of 60 seconds ozone and 2 seconds nitrogen purge.No difference was observed in Raman spectra after the treatment.This might indicate that the sharp peaks are not due to an adsorbed CH 2 layer.Furthermore, an adsorbed layer might be expected to show Raman peaks over the entire surface, but, as discussed, the Raman peaks are only observed in specific, rough pits.As another check, Fourier transform infrared (FTIR) microscopy/spectroscopy was performed on the same regions of hot spot samples and cold-worked foils and strong C-H stretch absorptions near 2900 cm À1 were found.Since FTIR has a larger skin depth for these hydrocarbon absorptions than Raman (roughly a factor of 2), it therefore probes a bit deeper and the strong absorption again points to the hydrocarbon region occurring deeper into the surface than might be expected from an adsorbed monolayer.
Other regions of the hot spot samples and cold-worked foils revealed a different, yet still reproducible, Raman spectrum as shown in Fig. 7.These spectra do not show the sharper hydrocarbon modes but rather two broad peaks near 1350 cm À1 and 1600 cm À1 which have been identified as the D band of amorphous carbon (a-C) and G band of graphite respectively [34][35][36].Given that the G band is quite broad we simply refer to these Raman spectra as a-C.We present the data from Ref. [37] in Fig. 7 where an assumed a-C layer of a few nm is present on top of NbC nanoparticles.The stronger signal/noise observed here suggests that the a-C regions are thicker than a few nm.While a detailed analysis of these spectra is ongoing it appears that there may be varying degrees of hydrogenation of the a-C and it has been reported that the strong fluorescence background in Raman as observed in Fig. 4 is linked to H saturation of a-C dangling bonds [34].
The a-C modes help explain some peculiarities of Fig. 4.While the spectrum of #12 shows only the sharp hydrocarbon modes, other spectra such as #3, #9, and #10 show the hydrocarbon peaks on top of broader peaks which appear to be the a-C modes including the G band near 1600 cm À1 .The picture emerging is that Raman is revealing amorphous carbon regions with varying amounts of hydrogenation, including, in some cases, longer chain hydrocarbons.At this time it is not clear whether the long-chain hydrocarbons have formed as a consequence of C and H migration to dislocations along with annealing or whether compounds such as stearic acid are already present in the Nb.While stearic acid is a common lubricant for deep drawing and is used in soaps in machining, it is difficult to understand how such molecules could withstand the significant etching and annealing (e.g.600 C [11]) that is used in cavity processing.

E. SEM/EDX results
The Raman measurements prompted a reexamination of hot spot samples by SEM/EDX.On Jlab #12, the faceted pits showed no difference in EDX spectra.This confirms earlier reports [11] and is consistent with the Raman results in Fig. 1 on sample #9.There are other pits that look rough under SEM that showed significantly higher carbon concentration.In the left panel of Fig. 8, we show an image of We also did EDX mapping of a grain boundary on the bent foil as shown in Fig. 11.The grain boundary did not show any difference in chemical composition which is in agreement with the Raman results discussed earlier.Only the dark patches show higher concentration of carbon and oxygen.Some of the regions even show reduced concentration of niobium indicative of large volume of carbon compounds since the probing depth is around 1 m.

IV. SUMMARY AND CONCLUSION
Raman microscopy/spectroscopy, in conjunction with DFT calculations, has been shown to be an effective probe of the Nb surface, relevant for SRF cavity development.The quality of the DFT calculations has been demonstrated by the comparison between the calculated and experimental spectra of Nb oxide powders.A focus of this work has been to use backscatter Raman to gain insight into the correlation of high etch pit density with the formation of hot spots that lead to medium field Q drop.The Raman spectra observed in particular types of macroscopic pit defects that appear very rough under an optical microscope showed reproducible features of enhanced intensity and sharp peaks, in stark contrast with the featureless spectra of surrounding regions.Other pits which appear smooth and faceted showed no difference with the surroundings.
Comparison of the measured peaks with DFT calculations as well as with results in the literature indicated the presence of carbonaceous impurities in such rough pits.The close agreement of some peaks with calculated NbH and NbH 2 modes indicates that these compounds cannot be ruled out, however, the overall Raman spectra, especially the high frequency modes near 2900 cm À1 , more closely resemble that of CH 2 chain compounds such as stearic acid.Another group of Raman spectra on hot spots and Nb foils showed the presence of amorphous carbon.Subsequent reexamination of the hot spot samples by It is suggested that considerable migration of bulk C and H has taken place, as a consequence of the combination of strain induced dislocations and annealing.The presence of oxygen in some regions probably arises from diffusion from the surface oxide.It may be possible that impuritystabilized dislocation bundles have formed in Nb cavities that are resistant to disintegration by a 600 C bake, and lead to a high density of etch pits after acid etching.Some of these bundles might contain high concentrations of C, H, and O allowing the formation of CH 2 chains due to dissolved hydrogen, or that the hydrocarbons were already present in the host Nb material.In the latter case, the presence of stearic acid (or some other fatty acid) might originate from soaps used to clean the plate or lubricants used in deep drawing.Such stable compounds might migrate to dislocations.
The connection between the C-H modes and dislocations has been reinforced by the observation of identical Raman peaks in rough patches that were created by the cold working of previously annealed Nb foils.Such rough patches reveal high C and O concentrations under SEM/ EDX identical to that found in hot spot cavity samples.The picture that emerges is that regions of high dislocation density can serve as a trap of C, O, and H or of stable, preexisting, hydrocarbon inclusions.
The surface chemistry of such macroscopic defects, being considerably different from the bulk Nb, must lead to reduced or nonexistent superconductivity causing localized enhanced dissipation (hot spots).It should be noted that in very early studies, surface carbon was suspected of severely limiting the accelerating fields of Nb SRF cavities [37] but the Raman studies here show that the excess carbon is quite localized.Such hot spot regions will generate thermal quasiparticles, thereby increasing the size of the hot spot, enhancing Rs and decreasing Q.The Raman work has thus provided a plausible explanation for the correlation of hot spots with the high density of etch pits.However, we note that not all etch pits are identical.The more standard-type faceted etch pits do not show enhanced carbon or hydrogen content.The distinction between these types of macroscopic pits, along with the formation of hydrocarbons, requires further investigation.

FIG. 1 .
FIG. 1.The right panel shows two optical images of a region near grain boundary with high density of etch pits on Jlab #9.The left panel shows corresponding Raman spectra of circled regions.
FIG.5.The background (blue line) is a polynomial fit of the peakless region of the raw data.The black line is the residue after subtracting the background.

FIG. 7 .
FIG.7.Raman spectra taken from cold-worked Nb foils and Jlab hot spots in comparison with spectrum from amorphous carbon[37].Foils 1 and 2 are from Alpha Aesar that have been annealed to the melting point in UHV and subsequently cold worked in air as described in the text.The legend shows the order of the spectra based on height of the 1300 cm À1 peak.

TABLE I .
Vibrational modes of interstitial impurity atoms in the bcc niobium lattice from DFT calculations.O, N, and C are located in octahedral lattice interstices and H is in a tetrahedral lattice interstitial site.

TABLE II .
[17]n vibrational assignments and peak frequencies for rough patches on Nb bent foil and stearic acid[17].