Electronic Structure of Ce-Doped and -Undoped Nd 2 CuO 4 Superconducting Thin Films Studied by Hard X-ray Photoemission and Soft X-ray Absorption Spectroscopy
M. Horio, Y. Krockenberger, K. Yamamoto, Y. Yokoyama, K. Takubo, Y. Hirata, S. Sakamoto, K. Koshiishi, A. Yasui, E. Ikenaga, S. Shin, H. Yamamoto, H. Wadati, A. Fujimori
aa r X i v : . [ c ond - m a t . s up r- c on ] J un Electronic Structure of Ce-Doped and -Undoped Nd CuO Superconducting ThinFilms Studied by Hard X-ray Photoemission and Soft X-ray Absorption Spectroscopy
M. Horio , Y. Krockenberger , K. Yamamoto , Y. Yokoyama , K. Takubo , Y. Hirata , S. Sakamoto ,K. Koshiishi , A. Yasui , E. Ikenaga , S. Shin , H. Yamamoto , H. Wadati , and A. Fujimori Department of Physics, University of Tokyo, Bunkyo-ku, Tokyo 113-0033, Japan NTT Basic Research Laboratories, NTT Corporation, Atsugi, Kanagawa 243-0198, Japan Institute for Solid State Physics, University of Tokyo, Kashiwa, Chiba 277-8561, Japan and Japan Synchrotron Radiation Research Institute, Sayo, Hyogo 679-5198, Japan (Dated: June 21, 2018)In order to realize superconductivity in cuprates with the T’-type structure, not only chemicalsubstitution (Ce doping) but also post-growth reduction annealing is necessary. In the case of thinfilms, however, well-designed reduction annealing alone without Ce doping can induce superconduc-tivity in the T’-type cuprates. In order to unveil the origin of superconductivity in the Ce-undopedT’-type cuprates, we have performed bulk-sensitive hard x-ray photoemission and soft x-ray absorp-tion spectroscopies on superconducting and non-superconducting Nd − x Ce x CuO ( x = 0 , .
15, and0.19) thin films. By post-growth annealing, core-level spectra exhibited dramatic changes, whichwe attributed to the enhancement of core-hole screening in the CuO plane and the shift of chem-ical potential along with changes in the band filling. The result suggests that the superconductingNd CuO film is doped with electrons despite the absence of the Ce substitution. High-temperature superconductivity in cuprates is re-alized by doping hole or electron carriers into the parentmaterial which has been widely considered to be an an-tiferromagnetic (AFM) Mott insulator. Ln CuO ( Ln :rare earth) with the T’-type structure, where Cu takesthe square-planar coordination, can be doped with elec-trons by substituting Ce for Ln . Generally, electrondoping by Ce substitutions alone cannot induce supercon-ductivity in bulk crystals of the T’-type cuprates, and anadditional procedure of post-growth annealing in a reduc-ing atmosphere is required [1]. Three distinct microscopicscenarios have been experimentally proposed for the roleof the reduction annealing [2]: i) removal of impurity api-cal oxygen atoms [3, 4] ii) creation of vacancies at the reg-ular oxygen sites [5, 6] iii) filling of Cu vacancies [7]. Whilethe exact microscopic effect remains elusive, reduction an-nealing dramatically suppresses the AFM order [8, 9] andreduces quasi-particle scattering [10], inducing supercon-ductivity in the Ce-doped samples.Furthermore, it has been demonstrated that thin filmsof T’-type cuprates can exhibit superconductivity with-out any Ce doping when properly annealed [11–14]. Theobserved decrease of the c -axis parameter by annealing[12–14] rather than its increase [7] should be attributedto the removal of apical oxygen atoms in these samples[3, 4], as the removal of apical oxygen reduces the re-pulsion between the CuO planes. The large surface-to-volume ratio of thin films helps the thorough removal ofimpurity oxygen atoms. The observation has cast doubton the fundamental assumption that the parent compoundof the cuprate superconductor is an AFM Mott insulator.Theoretical studies using the local density approximationplus dynamical mean field theory (LDA+DMFT) have in-deed predicted that the parent compound of the T’-typecuprates is not a Mott insulator but a Slater insulator in the sense that the AFM order is necessary to open the in-sulating band gap [15–17]. When discussing the electronicstructure of parent compounds and the phase diagram,however, the possibility should not be overlooked thatoxygen reduction affects the carrier concentration. In fact,annealing-induced changes in the N´eel temperature [18]and the optical conductivity [19] for bulk Nd − x Ce x CuO crystals have been interpreted as due to the doping 0.03–0.05 electrons/f.u. Recent angle-resolved photoemissionspectroscopy (ARPES) studies on Ce-doped bulk singlecrystals [9, 20] and that on insulating Ce-undoped thinfilms [21] have also shown that the electron concentrationsof annealed samples estimated from the Fermi surface areaare larger than that expected from the Ce concentrations.To understand the cuprate phase diagram and the elec-tronic states of the parent compounds, it is important tounveil the electron concentration of the superconducting(SC) and non-superconducting (non-SC) Ce-undoped T’-type cuprates.For that purpose, systematic studies of thin filmsamples with various Ce concentrations and reduc-tion/oxidization treatments are necessary. However,ARPES measurements on such thin films are hamperedby the lack of an in-situ ARPES measurement systemcombined with a molecular beam epitaxy (MBE) appa-ratus for the T’-type cuprates, which makes ARPES-quality surfaces available. This is because ARPES is asurface-sensitive technique. In this Letter, we report onmeasurements of bulk-sensitive hard x-ray photoemissionspectroscopy (HAXPES) and soft x-ray absorption spec-troscopy (XAS) of SC and non-SC Nd − x Ce x CuO thinfilms annealed or oxidized under various atmosphere. ByHAXPES, one can measure core-level shifts and hencethe chemical-potential shift, which directly probes thedoped electron concentration. By annealing Nd CuO , T (K)11010 T (K)11010 R e s i s t i v i t y ( µ Ω c m ) T (K) l (a) (c) x = 0.15 x = 0.19 As-grown
Weaklyannealed
Annealed OxidizedSC OxidizedSC x = 0 c - a x i s l eng t h ( Å ) x (d)(b) non-SCSC Nd x Ce x CuO FIG. 1. Physical properties of Nd − x Ce x CuO thin films. (a)–(c) Resistivity versus temperature for the x = 0, 0.15, and 0.19films, respectively. The inset shows a magnified plot near theSC transition for each composition. (d) c -axis lengths plottedagainst Ce concentration x . The markers are color-coded ac-cording to (a)–(c). Two solid lines trace the c -axis lengths ofSC and non-SC films whose difference may originate from thedifferent amount of apical oxygen atoms. we observed dramatic changes in the core-level spectra,which can be explained by the strong modification of core-hole screening in the CuO planes and by the chemical-potential shift caused by electron doping. The presentresults indicate the possibility that the SC Ce-undopedT’-type cuprates are doped with a significant amount ofelectrons.Nd − x Ce x CuO ( x = 0 , .
15, and 0.19) thin films withthe thicknesses of 200 nm, 100 nm, and 100 nm, re-spectively, were grown on SrTiO (001) substrates byMBE. For Nd CuO , we prepared three kinds of films:as-grown, weakly-annealed, and annealed films, amongwhich only the annealed film showed superconductivitywith T c = 25 . T c was defined as the temperaturewhere the resistivity drops to zero. Ce-doped films showedsuperconductivity without ex-situ annealing in a tubularfurnace. The T c ’s were 24.0 K and 21.5 K for x = 0 .
15 and x = 0 .
19, respectively. Oxidized non-SC films were alsoprepared for both the compositions. Conditions of anneal-ing and oxidization are described in Supplementary Infor-mation. The resistivity curves and the c -axis lengths of allthe films are plotted in Fig. 1. The difference in the c -axislength between the SC and non-SC films in Fig. 1(d) maylargely originate from the difference in the amount of api-cal oxygen atoms [12–14]. HAXPES measurements wereperformed at beamline 47XU of SPring-8 at T = 300 Kwith hν = 7 .
94 keV photons. The total energy resolutionwas determined from the Fermi edge of Au to be 0.3 eV.XAS measurements were performed in the total electronyield mode at beamline 07LSU of SPring-8 at T = 300K under the pressure better than 5 × − Torr. Twokinds of linearly polarized light, with polar angle θ = 90 ◦ ( E ⊥ c ) and θ = 30 ◦ , were used for the measurements[22].Figure 2(a) shows Cu L -edge XAS spectra for E k c I n t en s i t y ( a r b . un i t s ) E ǁ c Nd CuO Cu L -edge As-grown (Non-SC)Annealed (SC) (a) (b) Nd CuO O K -edge E ǁ c ┴ E ǁ c ┴ As-grown (Non-SC)Annealed (SC)
FIG. 2. XAS spectra of Nd CuO thin films. (a) Cu L -edgeXAS spectra for E ⊥ c (top) and E k c (bottom) polarizations.The spectra have been normalized to the intensity of the Nd M , XAS peak at 978 eV. (b) O K -edge XAS spectra for E ⊥ c and normalized to the intensity at 532–540 eV. and E ⊥ c for as-grown and annealed Nd CuO films. Thespectra for E k c were obtained by subtracting the contri-bution of E ⊥ c from the spectra measured with θ = 30 ◦ polarization. The absorption intensity is proportional tothe unoccupied density of states (DOS) of the specific ele-ment multiplied by transition matrix elements [29], whichdepends on the initial- and final-state orbital symmetryas well as incident light polarization. Thus, XAS providesinformation about the element- and orbital-specific unoc-cupied DOS. With polarization E ⊥ c , matrix element ofthe transition into the 3 d x − y orbital is three times largerthan that into the 3 d z − r orbital, while the polarization E k c allows only the transition into the 3 d z − r orbital(See Supplementary Information). The XAS spectra for E ⊥ c show an intense peak at ∼
930 eV and its in-tensity decreases with annealing by 30 %, suggesting thereduction of unoccupied DOS near E F as reported for Ce-doped samples [30–33]. In contrast, spectra for E k c remains negligibly weak. This leads to a small ratio ofthe 3 d z − r weight to the 3 d x − y weight (4 % and 2% for as-grown and annealed Nd CuO , respectively) inagreement with previous reports [31, 32]. Therefore, the3 d z − r orbitals are almost completely filled regardlessof whether annealed or not, and additional electrons goto the 3 d x − y orbitals by annealing. On the other hand,the pre-edge peak ( ∼
529 eV) in O K -edge XAS measuredwith E ⊥ c [Fig. 2(b)], which represents the transitionfrom O 1 s to in-plane O 2 p orbitals hybridized with theupper Hubbard band, shows only a slight change (5 %) incontrast to Cu L , -edge XAS for E ⊥ c . This agrees withthe changes observed by Ce doping in the previous stud-ies and suggests that the orbital character of the upperHubbard band is dominated by 3 d x − y [30–32, 34].The effect of annealing is also remarkable in HAXPESspectra. Figure 3(a) shows the Cu 2 p / core-level HAX-PES spectra of the as-grown and annealed Nd CuO films.Upon photoemission from a core level, valence electronsare attracted to the core-hole site to screen its potential.Different types of core-hole screening result in various fi-nal states appearing as fine structures in the core-level
534 532 530 528Binding energy (eV)534 532 530 528Binding energy (eV) (b) Nd 3 d As-grown Annealed (c) O 1 s As-grown Annealed (d) O 1 s CuO CuO Nd O Nd O Annealed(SC)
990 985 980 975 Nd CuO Nd CuO Nd CuO d f d f L = 0.03 α = 0.75 α (0.56)(0.54) (0.44)(0.46) I n t en s i t y ( a r b . un i t s )
945 940 935 930Binding Energy (eV) p d p d L ZRS A Satellite Main h ν = 7.94 keV T = 300 K As-grown(Non-SC)Annealed (SC)Nd CuO (a) Cu 2 p (e) Valence band x = 0As-grown(Non-SC)Weaklyannealed(Non-SC)Annealed(SC) x = 0.15Oxidized(Non-SC)SC x = 0.19Oxidized(Non-SC)SC Nd x Ce x CuO As-grown(Non-SC)
FIG. 3. HAXPES spectra of Nd − x Ce x CuO thin films. (a) Cu 2 p / core-level spectra of the as-grown non-SC (top) andannealed SC Nd CuO thin films (bottom). One observe final states where the core holes are unscreened (2 p d ), screened byan electron transferred from neighboring oxygen atoms (2 p d L ), screened by an electron transferred from neighboring CuO plaquettes thereby creating a Zhang-Rice singlet (ZRS), and screened by conduction electrons (A) (b),(c) Nd 3 d / and O 1 s core-level spectra normalized to the peak height, respectively, for the as-grown and annealed Nd CuO films. (d) O 1 s spectraof the as-grown (top) and annealed (bottom) Nd CuO films fitted to a superposition of a Voigt function (for O Nd O , green),a Mahan line shape α ) e − ( EB − E /ξ | ( E B − E ) /ξ | − α Θ( E B − E ) convolved with a Voigt function (for O CuO , orange), and another Voigtfunction (for O contamination, gray). Values in the parentheses represents the ratio of the peak area. (e) Valence-band spectraof Nd − x Ce x CuO thin films. photoemission spectra [35]. As has been discussed for theCu 2 p spectrum of insulating cuprates such as La CuO [36, 37] and Sr CuO Cl [38, 39] analyzed using multi-sitecluster calculation, the peak at the highest binding energy( E B = 940–947 eV) is the satellite due to the 2 p d finalstate with an unscreened core hole. In the main peak re-gion ( E B = 930–939 eV), the highest binding energy peakcorresponds to the 2 p d L final state, where an electronis transferred from the neighboring oxygen atoms (localscreening) leaving a hole in the oxygen ligand orbital L ,and the two peaks in the middle to the final state wherean electron is transferred from oxygen in the neighboringCuO plaquettes thereby creating a Zhang-Rice singletin the plaquettes (non-local screening). Apart from peakA at the lowest binding energy discussed below, the lineshape of the main peak resembles those of the other insu-lating cuprates with two-dimensional CuO planes [39].Upon annealing, peak A is strongly enhanced and theentire line shape dramatically changes. The binding ener-gies of the charge-transferred final state are determined bythe energy levels of the electronic states from which theelectron is transferred to screen the core-hole potential.Therefore, the enhancement of the lowest energy peak in-dicates the development of electronic states closest to theFermi level ( E F ) [36, 40] and is attributed to final stateswhere the core hole is screened by conduction electrons.The observed changes in the present Cu 2 p spectra there-fore suggest that conduction electrons were introduced byannealing, consistent with the occurrence of superconduc-tivity in the annealed Nd CuO .The effect of annealing on the Nd and O core levels of Nd CuO films is shown in Figs. 3(b)–(d). Upon anneal-ing, the Nd 3 d / peak was shifted toward higher bindingenergy by 0.22 eV [Fig. 3(b)]. As for the O 1 s peak,a shift of similar amount (0.21 eV) was observed at theedge, but the spectral line shape is also modified: the fullwidth at half maximum decreased from 1.44 eV to 1.16eV and a long high binding energy tail emerges above ∼ . s photoemission spectra of Nd − x Ce x CuO in previousstudies [41, 42]. In order to disentangle this complicatedspectral deformation, the O 1 s spectra have been ana-lyzed as follows: The peak region of the O 1 s spectrum ofthe as-grown film is rather flat, implying unresolved twopeaks with similar intensities. We attribute them to oxy-gen atoms in the Nd O layers (O Nd O ) and those in theCuO planes (O CuO ), which are contained in Nd CuO with equal amount. We calculated the oxygen binding en-ergies using a Wien2k package, and found that the O Nd O s level was located at a lower binding energy than theO CuO s level [43]. Therefore, the O 1 s spectra of theNd CuO films were fitted to a superposition of a Voigtfunction (for O Nd O ) at lower binding energies and Ma-han line shape α ) e − ( EB − E /ξ | ( E B − E ) /ξ | − α Θ( E B − E ) convolvedwith a Voigt function (for O CuO ) at higher binding en-ergies. The asymmetric Mahan line shape was assumedfor O CuO considering the core-hole screening by metallicelectrons. Another Voigt function was added to the fit-ting function to reproduce a weak contamination compo-nent at ∼
531 eV. The fitting yielded O Nd O and O CuO peaks with nearly the same area both for the as-grownand annealed films as shown in Fig. 3(d), consistent withthe initial assumption that each O 1 s spectrum consistsof two components arising from O CuO and O Nd O [45].The present analysis thus enables us to identify the effectof annealing on the two O 1 s core levels separately: TheO Nd O peak was shifted by 0.22 eV toward higher bindingenergy, and the O CuO peak became strongly asymmetric(represented by the increase of asymmetry parameter α from 0.03 to 0.75). Since a finite DOS at E F leads to apeak asymmetry, the strong asymmetry of the O CuO peakin the annealed Nd CuO film is consistent with the dra-matic enhancement of the electrical conductivity. Anotherremarkable point is that annealing shifted the O Nd O s peak by almost the same amount as the Nd 3 d / peakwithout appreciable changes in the line shapes.The shift of the core-level binding energy is given by[49] ∆ E B = ∆ µ − K ∆Q + ∆ V M − ∆ E R , (1)where ∆ µ is the change in the chemical potential, ∆ Q isthe change in the number of valence electrons, K is a con-stant, ∆ V M is the change in the Madelung potential, and∆ E R is the change in the extra-atomic screening of thecore-hole potential by conduction electrons and/or dielec-tric polarization of surrounding media. Almost identicalshifts observed for the Nd 3 d and O Nd O s core levelsindicate that ∆ V M is negligibly small because it wouldshift the core levels of the O − anion and the Nd cationin different ways. ∆ E R cannot be the main origin of theobserved shifts, either, because the increase of conductionelectrons by annealing would shift the core-level peaks to-ward lower binding energy, opposite to the experimentalobservation. Considering that the valences of Nd andO − are fixed (∆ Q = 0), we conclude that the observedshifts in Nd 3 d and O Nd O s core levels of Nd CuO are largely due to the chemical potential shift ∆ µ . Theincrease of the core-level binding energies by annealing in-dicates the increase of ∆ µ due to the addition of electrons.Having identified the chemical-potential shift caused byannealing in Nd CuO , we compare the core-level struc-ture of Nd CuO with those of Ce-doped compounds. ByCe substitution, annealing, and oxidization, the Nd andCe 3 d spectra were shifted maintaining their shape. Theline-shape changes in the Cu 2 p and O 1 s spectra werealso consistent with the above scenario [50]. The chemical-potential shift ∆ µ defined as the average shift of the Nd 3 d and O Nd O s core levels, is plotted in Fig, 4(a) againstCe concentration x . For Nd CuO , the chemical poten-tial is shifted upwards with annealing, reaching the levelof Ce-doped x = 0 .
15 and 0.19 superconductors whensufficiently annealed. The valence-band spectra shownin Fig. 3(e) and Supplementary Information are also al-most identical among the three SC films with different Ceconcentrations ( x = 0, 0.15, and 0.19), indicating thatthe electronic structure and band filling are close to eachother. C he m i c a l - po t en t i a l s h i ft ( e V ) x As-grown(Non-SC)Weakly annealed(Non-SC)Annealed (SC) Oxidized (Non-SC)SCOxidized (Non-SC)SC T c ( K ) (b)(a) FIG. 4. Chemical-potential shifts in Nd − x Ce x CuO thinfilms. (a) Chemical-potential shifts ∆ µ defined as the aver-age of the Nd 3 d and O 1 s core-level shifts plotted against Ceconcentration x . (b) T c ’s plotted against ∆ µ . The markers arecolor-coded according to (a). Thin films with shorter (longer) c -axis lengths and hence less (more) apical oxygen atoms areSC (non-SC) [See Fig. 1(d)]. The arrows connect the filmswith the same Ce concentration. The large electron concentration for the annealed SCNd CuO film can be explained if oxygen atoms are re-moved not only from the apical site [3, 4] but also fromthe regular sites (in the CuO plane and/or the Nd O layer) [5, 6], leading to the total oxygen content less thanthe stoichiometric one. On the other hand, upon oxidiza-tion of Nd − x Ce x CuO ( x = 0 . x = 0 .
15 film was already oxy-gen deficient and the vacancies were filled by oxidization.Consequently, the band filling of the T’-type cuprates isnot determined by the Ce substitution alone but by thecombined effect of Ce substitution and oxygen vacanciesand, therefore, the phase diagram for this system shouldbe made as a function of actual band filling, rather thanas a function of Ce concentration. The oxidization processalso introduced excess apical oxygen atoms, as suggestedby the c -axis elongation [Fig. 1(d)], and the x = 0 .
15 filmturned non-SC.In Fig. 4(b), the T c ’s of Nd − x Ce x CuO films are thusplotted against chemical-potential shift, which representsthe electron concentration. The T c values are not solelydetermined by the electron concentration since x = 0 . c -axis lattice parameter by oxidization is also rathersmall for the x = 0 .
19 film (0.06 %) compared to that for x = 0 .
15 film (0.12 %), indicating that only a tiny amountof apical oxygen atoms were incorporated. These resultsare consistent with previous studies where the amountof oxygen reduction decreases with increasing Ce con-centrations [3, 4, 52]. The higher concentration of Ce and the smaller c -axis lattice parameter may make theCe-overdoped samples more robust against oxygen non-stoichiometry induced by reduction annealing and oxidiza-tion. While oxygen vacancy at the regular sites increasesthe electron concentration, excess oxygen atoms at theapical site immediately destroy superconductivity. Theelectronic structure of the T’-type cuprates are thus dom-inated not only by Ce concentrations but also by oxygennon-stoichiometry.In conclusion, we have performed HAXPES and XASmeasurements on Nd − x Ce x CuO ( x = 0, 0.15, and 0.19)thin films with varying annealing atmosphere, and ob-served changes in the band-filling level among them. Theelectronic structure of SC Nd CuO was found to be in-timately linked to those of Ce-doped superconductors asthe electrons were doped into the thin films by annealingprobably through the creation of oxygen vacancies. Sincethe electron concentration and superconductivity of theT’-type cuprates are significantly affected by oxygen non-stoichiometry, the electronic structure should be discussedbased on the actual electron concentration and oxygen oc-cupancies rather than solely the Ce concentration.Fruitful discussion with K. Okada and G.R. Castro isgratefully acknowledged. Experiments were performedat SPring-8 (proposal Nos. 2015B1699, 2015B1793,2015B7401, and 2016A1210). This work was supportedby Grants-in-aid from the Japan Society of the Promotionof Science (JSPS) (grant Nos. 14J09200 and 15H02109).M.H. acknowledges support from the Advanced LeadingGraduate Course for Photon Science (ALPS) and theJSPS Research Fellowship for Young Scientists. [1] Y. Tokura, H. Takagi, and S. Uchida, Nature (London) , 345 (1989).[2] N. P. Armitage, P. Fournier, and R. L. Greene, Rev. Mod.Phys. , 2421 (2010).[3] P. G. Radaelli, J. D. Jorgensen, A. J. Schultz, J. L. Peng,and R. L. Greene, Phys. Rev. B , 15322 (1994).[4] A. J. Schultz, J. D. Jorgensen, J. L. Peng, and R. L.Greene, Phys. Rev. B , 5157 (1996).[5] G. Riou, P. Richard, S. Jandl, M. Poirier, P. Fournier, V.Nekvasil, S. N. Barilo, and L. A. Kurnevich, Phys. Rev.B , 024511 (2004).[6] P. Richard, G. Riou, I. Hetel, S. Jandl, M. Poirier, and P.Fournier, Phys. Rev. B , 064513 (2004).[7] H. J. Kang, P. Dai, B. J. Campbell, P. J. Chupas, S.Rosenkranz, P. L. Lee, Q. Huang, S. Li, S. Komiya, andY. Ando, Nat. Mater. , 224 (2007).[8] P. Richard, M. Neupane, Y.-M. Xu, P. Fournier, S. Li, P.Dai, Z. Wang, and H. Ding, Phys. Rev. Lett. , 157002(2007).[9] M. Horio, T. Adachi, Y. Mori, A. Takahashi, T. Yoshida,H. Suzuki, L. C. C. Ambolode, K. Okazaki, K. Ono,H. Kumigashira, H. Anzai, M. Arita, H. Namatame, M.Taniguchi, D. Ootsuki, K. Sawada, M. Takahashi, T. Mi-zokawa, Y. Koike, and A. Fujimori, Nat. Commun. ,10567 (2016).[10] X. Q. Xu, S. N. Mao, W. Jiang, J. L. Peng, and R. L.Greene, Phys. Rev. B , 871 (1996).[11] A. Tsukada, Y. Krockenberger, M. Noda, H. Yamamoto,D. Manske, L. Alff, and M. Naito, Solid State Commun. , 427 (2005). [12] A. Tsukada, M. Noda, H. Yamamoto, and M. Naito, Phys-ica C , 459 (2005).[13] O. Matsumoto, A. Utsuki, A. Tsukada, H. Yamamoto, T.Manabe, and M. Naito, Physica C , 924 (2009).[14] Y. Krockenberger, H. Irie, O. Matsumoto, K. Yamagami,A. Tsukada, M. Naito, and H. Yamamoto, Sci. Rep. ,2235 (2013).[15] H. Das and T. Saha-Dasgupta, Phys. Rev. B , 134522(2009).[16] C. Weber, K. Haule, and G. Kotliar, Nat. Phys. , 574(2010).[17] C. Weber, K. Haule, and G. Kotliar, Phys. Rev. B ,125107 (2010).[18] P. K. Mang, O. P. Vajk, A. Arvanitaki, J. W. Lynn, andM. Greven, Phys. Rev. Lett. , 027002 (2004).[19] T. Arima, Y. Tokura, and S. Uchida, Phys. Rev. B ,6597 (1993).[20] D. Song, G. Han, W. Kyung, J. Seo, S. Cho, B. S. Kim,M. Arita, K. Shimada, H. Namatame, M. Taniguchi, Y.Yoshida, H. Eisaki, S. R. Park, and C. Kim, Phys. Rev.Lett. , 137001 (2017).[21] H. I. Wei, C. Adamo, E. A. Nowadnick, E. B. Lochocki, S.Chatterjee, J. P. Ruf, M. R. Beasley, D. G. Schlom, andK. M. Shen, Phys. Rev. Lett. , 147002 (2016).[22] See Supplementary Information for more detailed experi-mental setup, which includes Refs. [23–28].[23] H. Wadati and A. Fujimori, J. Electron Spectrosc. Relat.Phenom. , 222 (2013).[24] Y. Takata, M. Yabashi, K. Tamasaku, Y. Nishino, D.Miwa, T. Ishikawa, E. Ikenaga, K. Horiba, S. Shin, M.Arita, K. Shimada, H. Namatame, M. Taniguchi, H. No-hira, T. Hattori, S. S¨odergren, B. Wannberg, and K.Kobayashi, Nucl. Instr. Methods A , 50 (2005).[25] M. Sacchi, F. Offi, P. Torelli, A. Fondacaro, C. Spezzani,M. Cautero, G. Cautero, S. Huotari, M. Grioni, R. De-launay, M. Fabrizioli, G. Vank´o, G. Monaco, G. Paolicelli,G. Stefani, and G. Panaccione, Phys. Rev. B , 155117(2005).[26] NIST Electron Inelastic-Mean-Free-Path database, Ver-sion 1.1, National Institute of Standard and Technology,U.S., 2000.[27] S. L. Schroeder, Solid State Commun. , 405 (1996).[28] A. Ruosi, C. Raisch, A. Verna, R. Werner, B. A. David-son, J. Fujii, R. Kleiner, and D. Koelle, Phys. Rev. B ,125120 (2014).[29] F. M. F. de Groot, J. C. Fuggle, B. T. Thole, and G. A.Sawatzky, Phys. Rev. B , 5459 (1990).[30] C. F. J. Flipse, G. van der Laan, A. L. Johnson, and K.Kadowaki, Phys. Rev. B , 1997 (1990).[31] E. Pellegrin, N. N¨ucker, J. Fink, S. L. Molodtsov, A.Guti´errez, E. Navas, O. Strebel, Z. Hu, M. Domke, G.Kaindl, S. Uchida, Y. Nakamura, J. Markl, M. Klauda,G. Saemann-Ischenko, A. Krol, J. L. Peng, Z. Y. Li, andR. L. Greene, Phys. Rev. B , 3354 (1993).[32] J. Fink, N. N¨ucker, E. Pellegrin, H. Romberg, M. Alexan-der, and M. Knupfer, J. Electron Spectrosc. Relat. Phe-nom. , 395 (1994).[33] P. G. Steeneken, L. H. Tjeng, G. A. Sawatzky, A. Tanaka,O. Tjernberg, G. Ghiringhelli, N. B. Brookes, A. A. Nu-groho, and A. A. Menovsky, Phys. Rev. Lett. , 247005(2003).[34] A. Krol, C. S. Lin, Z. H. Ming, C. J. Sher, Y. H. Kao, C. L.Lin, S. L. Qiu, J. Chen, J. M. Tranquada, M. Strongin,G. C. Smith, Y. K. Tao, R. L. Meng, P. H. Hor, C. W. Chu, G. Cao, and J. E. Crow, Phys. Rev. B , 4763(1990).[35] M. A. van Veenendaal and G. A. Sawatzky, Phys. Rev.Lett. , 2459 (1993).[36] M. Taguchi, A. Chainani, K. Horiba, Y. Takata, M.Yabashi, K. Tamasaku, Y. Nishino, D. Miwa, T. Ishikawa,T. Takeuchi, K. Yamamoto, M. Matsunami, S. Shin, T.Yokoya, E. Ikenaga, K. Kobayashi, T. Mochiku, K. Hi-rata, J. Hori, K. Ishii, F. Nakamura, and T. Suzuki, Phys.Rev. Lett. , 177002 (2005).[37] K. Maiti, J. Fink, S. de Jong, M. Gorgoi, C. Lin, M.Raichle, V. Hinkov, M. Lambacher, A. Erb, and M. S.Golden, Phys. Rev. B , 165132 (2009).[38] T. B¨oske, O. Knauff, R. Neudert, M. Kielwein, M.Knupfer, M. S. Golden, J. Fink, H. Eisaki, S. Uchida,K. Okada, and A. Kotani, Phys. Rev. B , 3438 (1997).[39] T. B¨oske, K. Maiti, O. Knauff, K. Ruck, M. S. Golden, G.Krabbes, J. Fink, T. Osafune, N. Motoyama, H. Eisaki,and S. Uchida, Phys. Rev. B , 138 (1998).[40] G. Panaccione, F. Offi, P. Torelli, G. Vanko, O. Tjern-berg, P. Lacovig, A. Guarino, A. Fondacaro, A. Nigro, M.Sacchi, N. B. Brookes, and G. Monaco, Phys. Rev. B ,125133 (2008).[41] T. Suzuki, M. Nagoshi, Y. Fukuda, K. Oh-ishi, Y. Syono, and M. Tachiki, Phys. Rev. B , 4263 (1990).[42] N. Harima, J. Matsuno, A. Fujimori, Y. Onose, Y.Taguchi, and Y. Tokura, Phys. Rev. B , 220507 (2001).[43] See Supplementary Information for details of calculation,which includes Ref. [44].[44] J. C. Slater, Int. J. Quantum. Chem. S4 , 3 (1971).[45] See Supplementary Information for more details about thefitting procedure, which includes Refs. [46–48].[46] M. Ikeda, T. Yoshida, A. Fujimori, M. Kubota, K. Ono,H. Das, T. Saha-Dasgupta, K. Unozawa, Y. Kaga, T.Sasagawa, and H. Takagi, Phys. Rev. B , 014510 (2009).[47] R. Sankari, M. Ehara, H. Nakatsuji, Y. Senba, K.Hosokawa, H. Yoshida, A. D. Fanis, Y. Tamenori, S. Ak-sela, and K. Ueda, Chem. Phys. Lett. , 647 (2003).[48] P. H. Citrin, P. Eisenberger, and D. R. Hamann, Phys.Rev. Lett. , 965 (1974).[49] S. H¨ufner, Photoelectron Spectroscopy (Springer-Verlag,Berlin, 1995), Chap. 2, p. 35.[50] See Supplementary Information for individual spectra,which includes Ref. [51].[51] A. Kotani and Y. Toyozawa, J. Phys. Soc. Jpn. , 912(1974).[52] K. Suzuki, K. Kishio, T. Hasegawa, and K. Kitazawa,Physica C , 357 (1990). r X i v : . [ c ond - m a t . s up r- c on ] J un Supplementary InformationElectronic Structure of Ce-Doped and -Undoped Nd CuO Superconducting Thin Films Studied by Hard X-rayPhotoemission and Soft X-ray Absorption Spectroscopy
M. Horio , Y. Krockenberger , K. Yamamoto , Y. Yokoyama , K.Takubo , Y. Hirata , S. Sakamoto , K. Koshiishi , A. Yasui , E.Ikenaga , S. Shin , H. Yamamoto , H. Wadati , and A. Fujimori Department of Physics, University of Tokyo, Bunkyo-ku, Tokyo 113-0033, Japan NTT Basic Research Laboratories, NTT Corporation, Atsugi, Kanagawa 243-0198, Japan Institute for Solid State Physics, University of Tokyo, Kashiwa, Chiba 277-8561, Japan Japan Synchrotron Radiation Research Institute, Sayo, Hyogo 679-5198, Japan ∗ e-mail: [email protected] 1 . Conditions for the film growth and annealing (reduction or oxidization) All the films discussed here were grown by molecular beam epitaxy (MBE). For typicalgrowth conditions, post-growth annealing under reducing atmospheres is necessary to inducesuperconductivity in Nd − x Ce x CuO . The Nd − x Ce x CuO ( x = 0 .
15, 0.19) thin films weregrown on (001) SrTiO substrates at 680 and 665 ◦ C, respectively, using ozone flow ratesof 1.0 sccm. After the growth, the films were kept for 10 minutes under ultra-high vacuum(UHV) at 600 ◦ C before being subsequently cooled to room temperature. Superconductivityin these Nd − x Ce x CuO thin films has been verified also via magnetization measurements, inaddition to resistivity measurements [Figs. 1(b) and (c) in the main text]. These supercon-ducting films are labeled as “SC”. The superconducting Nd − x Ce x CuO films were dividedand one of them was treated under oxidizing conditions using 2.0 sccm ozone at 550 ◦ C for10 minutes and cooled under UHV. These films are labeled as “oxidized”.The Nd CuO thin films were grown on (001) SrTiO substrates at 725 ◦ C using ozoneflow rate of 1.0 sccm. These films were rapidly cooled down to room temperature. Toinduce superconductivity in the Nd CuO thin films, ex-situ annealing in a tubular furnaceis necessary. One of the Nd CuO thin films was annealed in a furnace first at 675 ◦ C underthe O partial pressure of 2 . × − Torr for 1 hour, and then at 525 ◦ C under UHV for 10minutes (annealed Nd CuO ). Another Nd CuO film was annealed first at 690 ◦ C underthe O partial pressure of 1 . × − Torr for 1 hour, and then at 500 ◦ C under UHV for 10minutes (weakly annealed Nd CuO ).
2. Experimental geometry
The experimental geometry of hard x-ray photoemission spectroscopy (HAXPES) is de-scribed in Supplementary Fig. 1(a). Measurements were performed with linearly polarizedlight with grazing incidence to maximize the photoemission intensity [1], and normally emit-ted photoelectrons were analyzed. The probing depth λ is typically 5–20 nm but dependson the kinetic energy of photoelectrons E k as well as on material [2]. As for the Cu 2 p corelevel, for example, Panaccione et al . [3] derived the equation λ Cu = 0 . E . k followingexperimental estimates for metallic Cu [4] and theoretical calculations using the NIST code[5]. This yields λ Cu ∼ hν = 7 .
94 eV.2 ample~ 7 nm λ E e - h ν (a) HAXPES 1° Sample ~ 5 nm E (b) XAS Sample E h ν
30° 30° c ab c ab h ν Cu λ Cu Supplementary Figure 1 : Experimental geometries. (a) HAXPES measurements. (b) XASmeasurements. Linearly polarized light with polar angle θ = 90 ◦ ( E ⊥ c , left) and θ = 30 ◦ (right)were used. Measurements of soft x-ray absorption spectroscopy (XAS) were carried out in the totalelectron yield mode using two kind of linearly polarized light, with polar angle θ = 90 ◦ ( E ⊥ c ) and θ = 30 ◦ , as schematically illustrated in Supplementary Fig. 1(b). The probingdepth λ is typically 2–7 nm [6]. As for Cu L -edge XAS, Ruosi et al . [7] derived λ Cu ∼ Cu O thin films.
3. Matrix elements of XAS
Using Fermi’s golden rule, matrix elements for Cu L -edge absorption relevant to thepresent study with light polarized along n = (sin θ cos φ, sin θ sin φ, cos θ ) can be written asfollows [8]: h p in − plane | H ′ | d x − y i = X i = x,y h p i | H ′ | d x − y i = M if sin θ (1) h p in − plane | H ′ | d z − r i = X i = x,y h p i | H ′ | d z − r i = 13 M if sin θ (2) h p z | H ′ | d x − y i = 0 (3) h p z | H ′ | d z − r i = 23 M if cos θ, (4)where H ′ is the Hamiltonian for electron-photon interaction, and M if is a reduced matrixelement. Therefore, absorption with polarization perpendicular to the c axis ( E ⊥ c , θ = 90 ◦ )is 75 % due to the transition into the 3 d x − y orbital, and that with polarization parallel tothe c axis ( E k c , θ = 0 ◦ ) is entirely due to transitions into the 3 d z − r orbital.3 . Calculations of O 1 s binding energies The O 1 s binding energies observable by photoemission spectroscopy were estimatedfrom the calculation of Slater transition states [9] using density functional theory (DFT).Calculations were carried out using the generalized gradient approximations (GGA). Forthe 4 f orbital, since GGA generally yields unrealistically high density of states (DOS) ina narrow energy window around E F , which could affect the O 1 s binding energy, we havecarried out three different kinds of calculations:(i): Nd CuO with GGA(ii): La CuO with the lattice constant of Nd CuO with GGA(iii): Nd CuO with GGA+ U ( U f = 9 eV)in (ii), Nd is replaced by La, where 4 f levels are unoccupied, and the 4 f level is locatedfar above E F . In (iii), on-site Coulomb repulsion U = 9 eV is introduced to the Nd 4 f orbitals thereby splitting the 4 f DOS away from E F . All the calculations were performedwith spin polarization and the O 1 s binding energies were finally determined by taking theaverage of spin-down and -up binding energies. The calculated O 1 s binding energies arelisted in Table. 1. Although the magnitude of the difference between the O Nd O and O CuO s binding energies depends on the calculation method, O Nd O always resides at a lowerbinding energy and O CuO at a higher binding energy. Calculation method O Nd O (eV) O CuO (eV)(i) 527.93 528.36(ii) 527.81 528.38(iii) 527.54 528.53 Supplementary Table 1 : O 1 s binding energies estimated from DFT calculations. Details ofthe calculations are described in the text.
5. Procedure for fitting O 1 s peaks First, the O 1 s spectrum of as-grown Nd CuO was fitted to a superposition of threepeaks: Voigt function (for O Nd O ), Mahan line shape α ) e − ( EB − E /ξ | ( E B − E ) /ξ | − α Θ( E B − E ) convolvedwith a Voigt function (for O CuO ), and Another Voigt function (for contamination peak at4
531 eV). The full width at half maximum (FWHM) of the Lorentian and Gaussian in theVoigt function was assumed to be the same for O Nd O and O CuO peaks. The parameter ξ in the Mahan line shape, which should be a magnitude of the order of the Fermi energy, wasfixed at 1 eV, which was determined from nearest-neighbor hopping parameter t ∼ .
25 eVpreviously derived for Nd − x Ce x CuO ( x = 0 .
15) [10]. The fitting yielded O Nd O and O CuO peaks with the area ratio of 0.44 : 0.56, consistent with the fact that the number of oxygenatoms in the Nd O layers and the CuO planes are the same. The obtained FWHM of theLorentzian was 0.15 eV, which is close to the inherent O 1 s core-hole lifetime of 0.16 eVobserved for H O [11], while the FWHM of the Gaussian was 0.75 eV, which is larger thanthe experimental total resolution of 0.3 eV possibly due to the contribution from phonons[12].Then, the O 1 s spectrum of annealed Nd CuO was fitted to the same functional formwith the FWHM of the Lorentzian and the Gaussian fixed to the values of the as-grownfilm. The parameter ξ was again fixed at 1 eV. The area ratio between O Nd O and O CuO peaks obtained by fitting was 0.46 : 0.54, which is close to that for the as-grown film, andhence validating the result of the fitting.
6. HAXPES spectra of Nd − x Ce x CuO films In Supplementary Fig. 2, core-level HAXPES spectra for Nd CuO (as-grown, weaklyannealed, annealed) and Nd − x Ce x CuO ( x = 0 . , .
19, SC, oxidized) are plotted. TheNd and Ce peak positions are shifted between films, but their spectral line shapes arealmost identical. On the other hand, the O 1 s peaks change their shape with varying Ceconcentration and oxygen content. Because the changes in the O 1 s peaks can be understoodwithin the above scenario, we compare the low energy edges of different films as representingthe O Nd O peak position, and summarize them in Supplementary Fig, 3(a). Not only the Nd3 d and O 1 s core-level peaks, but also the Ce 3 d peaks are shifted by the same amount forevery film (The origins of the shifts of the Ce 3 d were set at the Nd 3 d shift for the x = 0 . p core-level spectrum was rather5 n t en s i t y ( a r b . un i t s )
988 986 984 982 980 978 976 974Binding Energy (eV) I n t en s i t y ( a r b . un i t s )
534 533 532 531 530 529 528 527Binding Energy (eV) x = 0 (a)Nd 3 d (c) O 1 s As-grownSC x = 0.15 AnnealedOxidizedSCOxidized x = 0.19 x = 0 As-grownSC x = 0.15 AnnealedOxidizedSCOxidized x = 0.19 Weakly annealedWeaklyannealed f f L I n t en s i t y ( a r b . un i t s )
904 902 900 898 896 894Binding Energy (eV) (b) Ce 3 d SC x = 0.15 OxidizedSCOxidized x = 0.19 f L ( j = 3/2) 4 f ( j = 5/2) I n t en s i t y ( a r b . un i t s )
945 940 935 930Binding Energy (eV) (d) Cu 2 p x = 0 As-grownSC x = 0.15 AnnealedOxidizedSCOxidized x = 0.19 Weakly annealed I n t en s i t y ( a r b . un i t s ) (e) Valence band x = 0 As-grownSCAnnealedOxidizedSCOxidized x = 0.19 Weaklyannealed x = 0.15 x = 0As-grown (Non-SC)SCAnnealed (SC)Oxidized (Non-SC)SCOxidized (Non-SC) x = 0.19Weakly annealed (Non-SC) x = 0.15 Supplementary Figure 2 : HAXPES spectra of Nd − x Ce x CuO thin films. (a) Nd 3 d / . (b)Ce 3 d . (c) O 1 s . (d) Cu 2 p / . (e) Valence band. Peak or edge positions used to determine thecore-level shifts are indicated by vertical bars. small as plotted in Supplementary Fig. 3. Considering that the lowest energy peak corre-sponds to the final state where a core hole is screened by conduction electrons, the behaviorcan be understood within a simple Kotani-Toyozawa picture [13]. When the chemical po-tential is shifted, core-hole energy increases accordingly, but energy gain by transferring anelectron from conduction band to Cu 3 d level at the core-hole site also increases by the sameamount, compensating the chemical-potential shift. The binding energy of the Cu 2 p lowestenergy edge is thus insensitive to the amount of carriers. [1] H. Wadati and A. Fujimori, J. Electron Spectrosc. Relat. Phenom. , 222 (2013).[2] Y. Takata, M. Yabashi, K. Tamasaku, Y. Nishino, D. Miwa, T. Ishikawa, E. Ikenaga, K. .30.20.10-0.1 N d , O , and C u b i nd i ng ene r g y s h i ft ( e V ) C e b i nd i ng ene r g y s h i ft ( e V ) Nd 3 d O 1 s Ce 3 d Cu 2 p x = 0.15 x = 0 x = 0.19 A s - v g r o w n ( N on - S C ) W ea k l y annea l ed ( N on - S C ) A nnea l ed ( S C ) O x i d i z ed ( N on - S C ) S C O x i d i z ed ( N on - S C ) S C Supplementary Figure 3 : Binding energy shifts of the Nd 3 d , O 1 s , Ce 3 d , and Cu 2 p corelevels. The shifts were estimated by the peak position for Nd 3 d and by the low binding-energyedge for O 1 s , Ce 3 d , and Cu 2 p . The origin of the shifts of Ce 3 d are set at the shift of Nd 3 d forthe x = 0 .
15 oxidized film.Horiba, S. Shin, M. Arita, K. Shimada, H. Namatame, M. Taniguchi, H. Nohira, T. Hattori,S. S¨odergren, B. Wannberg, and K. Kobayashi, Nucl. Instr. Methods A , 50 (2005).[3] G. Panaccione, F. Offi, P. Torelli, G. Vanko, O. Tjernberg, P. Lacovig, A. Guarino, A. Fon-dacaro, A. Nigro, M. Sacchi, N. B. Brookes, and G. Monaco, Phys. Rev. B , 125133 (2008).[4] M. Sacchi, F. Offi, P. Torelli, A. Fondacaro, C. Spezzani, M. Cautero, G. Cautero, S. Huotari,M. Grioni, R. Delaunay, M. Fabrizioli, G. Vank´o, G. Monaco, G. Paolicelli, G. Stefani, andG. Panaccione, Phys. Rev. B , 155117 (2005).[5] NIST Electron Inelastic-Mean-Free-Path database, Version 1.1, National Institute of Standardand Technology, U.S., 2000.[6] S. L. Schroeder, Solid State Commun. , 405 (1996).[7] A. Ruosi, C. Raisch, A. Verna, R. Werner, B. A. Davidson, J. Fujii, R. Kleiner, and D. Koelle,Phys. Rev. B , 125120 (2014).[8] E. Pellegrin, N. N¨ucker, J. Fink, S. L. Molodtsov, A. Guti´errez, E. Navas, O. Strebel, Z. Hu, . Domke, G. Kaindl, S. Uchida, Y. Nakamura, J. Markl, M. Klauda, G. Saemann-Ischenko,A. Krol, J. L. Peng, Z. Y. Li, and R. L. Greene, Phys. Rev. B , 3354 (1993).[9] J. C. Slater, Int. J. Quantum. Chem. S4 , 3 (1971).[10] M. Ikeda, T. Yoshida, A. Fujimori, M. Kubota, K. Ono, H. Das, T. Saha-Dasgupta, K.Unozawa, Y. Kaga, T. Sasagawa, and H. Takagi, Phys. Rev. B , 014510 (2009).[11] R. Sankari, M. Ehara, H. Nakatsuji, Y. Senba, K. Hosokawa, H. Yoshida, A. D. Fanis, Y.Tamenori, S. Aksela, and K. Ueda, Chem. Phys. Lett. , 647 (2003).[12] P. H. Citrin, P. Eisenberger, and D. R. Hamann, Phys. Rev. Lett. , 965 (1974).[13] A. Kotani and Y. Toyozawa, J. Phys. Soc. Jpn. , 912 (1974)., 912 (1974).