Interfacial reconstruction and superconductivity in cuprate–manganite multilayers of YBa2Cu3O7−δ and Pr0.68Ca0.32MnO3

Interfacial reconstruction in multilayers of YBa2Cu3O7−δ and Pr0.68Ca0.32MnO3 and its impact on superconducting properties is investigated by means of electron energy-loss spectroscopy and dc electrical transport measurements. Massive underdoping of the cuprate is found by both electron transfer from the manganite and interfacial oxygen depletion, presumably as an adaptation to compressive epitaxial stress. The decrease of the optimally doped volume becomes evident from a distinct suppression of the superconducting transition temperature. This doping effect is stronger than the impact of a competing ferromagnetic order parameter reported for similar multilayer samples with spin-polarized manganites.


Introduction
The versatile structural and electronic properties of perovskite oxides provide manifold degrees of freedom for interfacial reconstruction schemes: octahedral tiltings to balance lattice mismatch and strain [1,2], mixed transition-metal cation valence to compensate for polar discontinuities and the creation or ordering of oxygen vacancies as an alternative to compositional roughening [3,4]. Much attention has been paid to novel electronic and magnetic phases emerging in the vicinity of such interfaces between perovskites, further enriching their already complex phase diagrams [5][6][7][8][9][10]. The integration of the cuprate superconductor YBa 2 Cu 3 O 7−δ (YBCO) into thin film superlattices or heterostructures with other perovskites was the subject of a number of studies focusing on the implications for the superconducting properties. Short-scale suppression of the superconducting transition temperature was found to be related to interlayer decoupling of coherence between CuO 2 planes in ultrathin superlattices [11] and to variations in intracell distances by means of epitaxial strain [12,13]. Proximity to ferromagnetic manganites, e.g. La 0.7 Ca 0.3 MnO 3 , yields additional spin injection effects on a 10 nm scale [14,15] and conversely, a suppression of the magnetization in the manganite by stationary charge transfer across the interface [16,17]. More recently, observations of inverse spin-switch behaviour [18,19] have been discussed in terms of interfacial magnetic coupling [20,21].
In this paper, we provide an example where the oxygen degree of freedom dominates the superconducting properties of YBCO in a superlattice. In multilayers with antiferromagnetic Pr 0.68 Ca 0.32 MnO 3 (PCMO), electron energy-loss spectroscopy (EELS) reveals that a combined interfacial reconstruction composed of short-scale Ca interdiffusion, charge transfer and especially intrinsic oxygen depletion lies at the bottom of a strong suppression of superconductivity.

Experimental details
Various multilayer samples, [YBCO t Y /PCMO t P ] N with t Y = 6-45 nm, t P = 8-12 nm and N = 3-12, were grown epitaxially on polished (LaAlO 3 ) 0.3 -(Sr 2 AlTaO 6 ) 0.7 (100) or SrTiO 3 (100) substrates by pulsed laser deposition (PLD) from stoichiometric targets. Process parameters 3 were a substrate temperature of 770 • C and an oxygen partial pressure of 0.4 (0.2) mbar during deposition of YBCO (PCMO). The pulsed KrF excimer laser operated at a 5 Hz rate with a fluence of 1.5 J cm −2 at the targets, providing deposition rates of 45 (20) nm min −1 for YBCO (PCMO). Post-deposition cooling was performed in 900 mbar of pure oxygen at a rate of 15 • C min −1 with a 1-2 h oxygen loading stage at 400 • C. The multilayer stacking sequence always starts with a PCMO buffer layer, also the top layer typically consists of PCMO to prevent the rather susceptible YBCO from deterioration.
Electron-transparent lamellae for cross-section analysis were prepared with a focused ion beam and subsequent low-angle (<10 • ), low-energy (2 keV) Ar ion milling. In order to avoid edge formation or intermixing at the YBCO/PCMO interfaces, the Ar milling was performed so that the azimuthal projection of the incident beam on the cross-section was parallel to the layers.
As references for the determination of cation valence from spectroscopic data, additional electron-transparent lamellae were taken from samples of PrMnO 3 , CaMnO 3 and Pr 0.5 Ca 0.5 MnO 3 . While the former are epitaxial thin film samples prepared by ion beam sputtering and PLD, respectively, the latter is polycrystalline target material. A reference thin film of oxygen-deficit YBCO was prepared by PLD with post-deposition cooling at an oxygen partial pressure of only 0.01 mbar.
Structure analysis includes x-ray diffraction (XRD), (scanning) transmission electron microscopy ((S)TEM) and EELS. STEM and EELS analyses were performed with Brookhaven's Hitachi HD-2700 STEM with a C s -corrected probe [22]. Therein, high-angle annular dark field (HAADF) imaging comprised collection angles of 53-280 mrad, and for EELS, a spectrometer collection semi-angle of 20 or 24 mrad at a convergence semi-angle of 28 mrad was used.
Electrical transport measurements were made in the van-der-Pauw setup with a constant test current of 0.5 mA. To this end, 1 × 1 mm 2 Au contacts were sputter coated at each corner of the 5 × 10 mm 2 samples.
Magnetization measurements were made in a superconducting quantum interference device magnetometer with in-plane magnetic fields applied along a substrate edge.

Microstructure and chemical reconstruction of the interface
XRD and selected area electron diffraction reveal epitaxial growth with YBCO grown in the 00l direction (notation in space group Pmmm (no. 47)) and PCMO exhibiting twins with hh0 and 00l out-of-plane orientation, respectively (Pbnm (no. 62)), see the inset of figure 1(a). The corresponding epitaxial in-plane relations read YBCO h00 PCMO 00l or PCMO hh0 , respectively. Figure 1(a) shows a representative cross-section of a multilayer sample, here [YBCO 12nm /PCMO 12nm ] 4 . The PCMO layer labelled P exhibits a columnar 00l nano-twin in hh0 surroundings. Stacking faults, antiphase boundaries and, to a minor extent, epitaxial yttria (Y 2 O 3 ) nano-particles are found in the YBCO layers, compare, e.g., [23,24]. The microstructural features of the YBCO are also reflected in an asymmetry concerning the layer stacking sequence: PCMO grows on rather irregular YBCO surfaces (termed PonY interface in the following), but in turn mediates this roughness and provides a smooth buffer Colour-coded intensity of core-loss edges simultaneously acquired in an EELS line scan across a YonP-interface (except Y L 2,3 , which required a separate detector range and was acquired subsequently). The B-cation site between Ba and Pr is defined as the interface position (x = 0) and is indicated by the horizontal white line. The leftmost panel displays the corresponding HAADF image. (c) HAADF image of a YonP-interface at an ultrathin lamella position (t/λ < 0.2), the same zone-axis as in (a). Inset: integrated intensity of Pr and Ba M 4,5 ELNES along the line scan given by the green line. Note that the Pr signal is overlaid with the Cu L 2,3 edge. At the bottom, the interfacial atomic structure model based on the EELS data is illustrated by cation columns colour-coded as follows: Pr black, Ca yellow, Mn pink, Y blue, Ba red and Cu green.
for the subsequent YBCO layer (YonP interface). The remaining half unit cell steps and twin boundaries at the PCMO surface induce planar defects in the YBCO. The interfaces between all layers are free of misfit dislocations.

5
The chemical composition of the interfaces was investigated by means of EELS line scans. Figure 1(b) shows the spatial evolution of energy-loss near-edge structures (ELNES) from all relevant core-loss edges across a YonP interface. By fitting error-function profiles to the integrated intensities of each ELNES, we deduced the following atomic model, which is also illustrated in figure 1(c). The interface is located between perovskite A-site layers of (nominal) Pr 0.68 Ca 0.32 -O and Ba-O. The very interface, defined as x = 0 in the following, is a perovskite B-site with unresolved composite character of Mn-O 2 and Cu-O. The exact determination of Cu contributions is hindered by the low signal-to-noise ratio of the Cu L 2,3 edge and its complete overlap with Pr M 4,5 . Multiple linear least-squares fits of the interfacial O K ELNES with two reference spectra from YBCO and PCMO bulk (see figure 3(a)) support a mixed Mn/Cu scenario. A robust feature of the YonP-interface is a distinctive short-scale diffusion of Ca into the first YBCO unit cell. The inflection point of the corresponding error function profile is located at x = −0.6 nm, i.e. at the first Y-site. We will further discuss the implications of this observation below.
The morphological asymmetry of the interfaces with respect to the layer stacking sequence is also reflected in the appropriate chemical profiles. In contrast to the abrupt YonP-interfaces, PonY-interfaces are characterized by a narrow interdiffusion zone, especially concerning the heavy A-site cations, i.e. Pr and Ba. The corresponding error function profiles provide an interdiffusion zone width of 2-3 nm (indicating a decrease of the element concentrations from 90 to 10%).

Interfacial charge transfer and electronic reconstruction
An important factor for the electronic and magnetic properties of the interface-near region is given by the transfer of charge across the interface, i.e. mutual doping of adjacent layers. Access to the local doping level is provided by the ELNES of core-loss edges, since they probe transitions to the first unoccupied states above the Fermi level. In particular, the O K edge (1s→ 2p transition) can be interpreted as the site-and symmetry-projected density-of-states due to the strong screening of the remaining core-hole [25]. Figure 2 shows the doping trends in the O K ELNES of PCMO and YBCO, realized by A-site substitution and varying oxygen content, respectively. The different spectral features arise from covalent mixing of O 2p states with the various A-and B-site metal orbitals, namely transition metal 3d (label K 1 ), A-site metals Pr 4f5d, Ca 3d, Ba 4f5d, Y 4d (K 2 ) and higher sp orbitals (K 3 ) [26][27][28]. The shape, position and weight of feature K 1 are affected by the 3d band filling and the local symmetry [27,29,30], and it thus represents a useful probe of the transition metal valence [31][32][33].
Focusing on the manganite, further information on the Mn valence is provided by the Mn L 2,3 ELNES (figure 2(a)), which is dominated by excitations from (spin-orbit split) 2p→ 3d and which is subject to multiplet effects due to strong wavefunction overlap between the final state and the core-hole [34]. Both the L 3 /L 2 branching ratio and Mn L 3 -O K 1 distance have been used for Mn valence determination [33,35,36]. In order to extract systematic variations of these valence markers in Pr 1−y Ca y MnO 3 , we fitted double Gaussians to the O K 1 and K 2 ELNES features (triple for y = 1) and double Gaussians to Mn L 2,3 . To this end, in addition to background subtraction, Mn L 2,3 spectra have been corrected for post-edge continuum contributions according to the scheme proposed by Varela et al [33].
The precise doping level of the different YBCO samples displayed in figure 2(b) is undetermined. We are therefore restricted to the mere qualitative correlation between the decrease of O K 1 intensity with oxygen deficit or hole depletion, respectively. According to [31,37], the total decline of a distinct K 1 peak occurs not until δ > 0.6.
The spatial evolution of O K 1 spectral weight across an embedded YBCO layer is shown in figure 3(b). A strong suppression is obvious in the vicinity of the interfaces to PCMO with uniform recovery towards the layer centre. Comparison with the corresponding values from the reference spectra in figure 2(b) points to a massive hole depletion in a zone extending roughly 3 nm from the respective interfaces. An associated electron depletion on the PCMO side of the interface is displayed in figures 3(c)-(e): all markers indicate a mid-layer Mn valence corresponding to the nominal Ca-doping level (V Mn = +3.32) within error bars and an increase by roughly +0.2 within 1 nm from the interface.
The observed electron transfer from PCMO to YBCO corresponds to the difference in the work functions of the materials, namely φ PCMO = 4.89 eV [38] and φ YBCO = 5-6 eV ( [39] and references therein). Considering the YBCO/PCMO interface as a p-p heterojunction, the width of the space charge layer around the interface can be estimated from w = 2 0 Y P V bi (n Y + n P ) 2 /(en Y n P ( Y n Y + P n P )) [40]. Therein, the charge carrier densities of doped holes in the e g -band of PCMO (0.32/Mn) and in CuO 2 layers of optimally doped YBCO (0.5/Cu) both amount to n Y,P ∼ 5 × 10 21 cm −3 ; the static dielectric constants at room temperature are Y ∼ 100 [41] and P = 30 [42], respectively. Built-in voltages in the range V bi = 0.1-1.1 V (the work function difference) then yield total space charge layer widths of w = 0.5-1.5 nm. These values roughly correspond to the experimentally observed 1 nm zone on the PCMO side of the junction, but they are not in line with the more extended hole depleted region in the YBCO. An additional effect stems from the short-scale Ca interdiffusion that changes the Ca 2+ /Pr 3+ ratio at the interface and therefore contributes to the local increase of  the Mn valence. In YBCO, the aliovalent substitution of Y 3+ by Ca 2+ nominally yields hole doping, thus counteracting the electron transfer from the PCMO, but this effect is known to be compensated for by a concomitant oxygen loss [43]. The long-range hole depleted region in the YBCO side of the junction is most likely caused by an appropriate oxygen deficit. Further evidence for this assumption is provided by the electrical transport data presented in the following subsection.

Suppression of the superconducting transition temperature
Electrical transport and magnetization have been probed in a sample batch of multilayers with varying YBCO single-layer thickness at a constant PCMO single-layer thickness of 8 nm. We first discuss the magnetic properties of the PCMO layers in order to estimate the potential impact on superconductivity. Pr 1−y Ca y MnO 3 in the doping range 0.3 y 0.5 exhibits antiferromagnetic ordering below T N ≈ 150 K, typically incorporating a minor fraction of ferromagnetic clusters [44][45][46][47]. The magnetization curves from our multilayer samples (see figure 4(a) for a representative example) show a weak magnetic moment (m = 0.09 µ B per formula unit at 5 K in saturation), which is suppressed by a factor of ∼20 compared to single-layer values reported in the literature [48]. An equally strong suppression is reported in the literature for comparable multilayer structures of YBCO and La 0.7 Ca 0.3 MnO 3 (LCMO) [17,49]; furthermore, magnetically dead layers of 3-5 nm are observed in ultra-thin single layers of LCMO and La 0.7 Sr 0.3 MnO 3 (LSMO) on various substrates [50][51][52][53]. We conclude  figure 4(b). The straight lines are a guide to the eye, intended to visualize the trends in our data. (b) Zero-resistance temperature T 0 C of our samples compared with the corresponding data from the literature on YBCO multilayers with non-magnetic PrBa 2 Cu 3 O 7 [11,12] and ferromagnetic manganites [49,57], respectively. that the low spin polarization in PCMO has only insignificant implications for superconducting properties of YBCO.
The dc electrical resistance is displayed in figure 4(b). In all samples, it is dominated by the YBCO portion as reflected by the essentially positive dR/dT and the superconducting phase transition. The critical temperature of a reference sample with a 300 nm YBCO single layer on a 100 nm PCMO buffer (Y300) amounts to T mid C = 86.5 K (1.5 K transition width), which is slightly reduced compared with (90 ± 0.5) K for a 300 nm YBCO single layer deposited directly on a SrTiO 3 substrate under the same conditions. The resistivity at room temperature increases from 0.5 m cm in Y300 to 3.5 m cm in [YBCO 6nm /PCMO 8nm ] 9 . This resistivity variation as well as the deviations from a constant dR/dT slope above T C are consistent with an enhanced oxygen deficit in the thinner YBCO layers [54]. A direct comparison with the literature data from systematic oxygen doping of polycrystalline samples is given in figure 5(a). Therein we provide evidence that the YBCO layer thickness t Y in our multilayers corresponds to an effective oxygen deficit δ eff (t Y ), since both the critical temperature and the out-of-plane lattice constant, determined from XRD, exhibit roughly the same relation between t Y and δ of the reference samples. Accordingly, the YBCO layers in [YBCO 6nm /PCMO 8nm ] 9 are subject to a heavy deficit, δ eff (6nm) ≈ 0.7, which is consistent with (merging) interfacial 3 nm zones of strong O K 1 suppression (recall the EELS profiles in figure 3(b)).
We want to point out that for several reasons, such an oxygen deficit in YBCO layers is not related to insufficient post-deposition oxygen loading (1-2 h at 400 • C and 900 mbar O 2 ). The oxygen diffusion coefficient along 00l bulk YBCO at 400 • C amounts to 10 −16 cm 2 s −1 [55], whereas De Souza and Kilner [56] indicate a corresponding coefficient of only ∼10 −23 cm 2 s −1 in the manganite. However, independent of the incorporation of PCMO layers that possibly hinder oxygen diffusion along the out-of-plane direction, oxygen loading of YBCO essentially takes place via the lateral surfaces of the multilayer stack: the oxygen diffusion coefficient in the (a, b)-plane is known to be up to 10 6 times faster than along 00l [55] and therefore dominates the loading process, since the ratio of normal to lateral sample dimensions is equal to only ∼100 nm/1 mm = 10 −4 . Consistent with this, extending the post-deposition oxygen loading time to 16 h has no impact on either the critical temperature or the out-of-plane lattice constant (compare the samples [YBCO 20nm /PCMO 8nm ] 5 and [YBCO 20nm /PCMO 8nm ] 12 in figures 4(b) and 5(a)). Furthermore, the triangular shape of the doping profile in individual YBCO layers ( figure 3(b)) contradicts an overall top-bottom gradient in the oxygen level. Figure 5(b) gives a comparison with T C values from the literature data for related multilayer samples of YBCO with non-magnetic PrBa 2 Cu 3 O 7 (PBCO) and ferromagnetic manganites, respectively. While superconductivity prevails even in single unit cell YBCO layers in the former, the competition between superconducting and ferromagnetic order parameters leads to a T C suppression in the YBCO/LCMO and YBCO/LSMO systems. No evidence for interfacenear oxygen depletion was reported for any of these reference systems; in fact, contrary to our samples, the out-of-plane lattice constant of YBCO was found to shrink when the layer thickness is decreased [12,57]. Compared with the impact of magnetic coupling and spin polarization, the pure effect of hole underdoping due to the oxygen deficit in YBCO/PCMO predominates. However, the aforementioned, size-induced reduction of spin polarization of the manganite layers in thin heterostructures [17,49] poses a restriction on this comparison. Indeed, the impact of a fully spin-polarized 50 nm LCMO layer on the T C of a YBCO single layer proves to be heavier than the suppression seen in our multilayer samples [15].
We argue that the PCMO imparts compressive in-plane stress on the YBCO, resulting in the observed out-of-plane expansion and oxygen vacancy formation. The compensation of interfacial stress from lattice mismatch or differences in thermal expansion behaviour has already been discussed as a possible source of oxygen deficit or disorder in YBCO heterostructures [58]. Evidence for this is provided by geometric phase analysis (GPA) [59] of appropriate high-resolution TEM images. Analysis of the interface between the substrate and the PCMO buffer layer indicates full stress relaxation within the first 5-10 nm of the manganite. Taking the equilibrium lattice constant of the substrate as the reference yields a pseudo-cubic in-plane lattice constant of a pc = 0.383 nm in the relaxed portion of the PCMO buffer. Comparison with the average equilibrium in-plane lattice constant of fully oxygenated YBCO a = (a + b)/2 = 0.386 nm (a = 0.382 nm and b = 0.389 nm, taken from pure phase powder samples [60]) indeed points to compressive stress in the on-growing cuprate layer, at least during the early growth stage. The formation of oxygen vacancies in the Cu-O chains of the YBCO, thus shortening the b parameter, is a plausible mechanism to release this stress. The GPA map in figure 6 reveals a very inhomogeneous strain state in the YBCO, which is on average positive compared to the underlying PCMO buffer. The thin PCMO layer sandwiched between YBCO tends to respond to this tensile strain, which is an expression of its lower elastic modulus E PCMO < 100 [61] compared to E YBCO ≈ 200 [62]. The average strain of the two YBCO layers in figure 6 with respect to the PCMO buffer is ip ≈ +0.6%. This value corresponds to a = 0.385 nm, i.e. to a slightly compressive strain compared to the equilibrium YBCO structure. In contrast, tensile stress/strain in YBCO presumably prevails in multilayer samples with ferromagnetic manganites of higher tolerance factor, LSMO (a pc = 0.390 nm [63]) and LCMO (0.387 nm [63]), and also on PBCO (0.390 nm [64]).
Another hint of adaptation to a compressively strained environment is the incorporation of Ca into the first YBCO unit cell on top of PCMO (recall the chemical reconstruction in figure 1): Ca is known for its high solubility limit of 30% on Y-sites in YBCO [65], but substitutes for Ba only under compression [66,67].
We conclude that subtle details of the interfacial reconstruction may govern the doping level and thus the superconducting transport properties of cuprates in cuprate/manganite superlattices. In particular, control over the oxygen degree of freedom in dedicated heterostructures is not restricted to the detailed sample preparation process, see [68] for an impressive example, but further relates to intrinsic properties of specific interfaces.

Summary
Our combined study of YBCO/PCMO multilayers reveals detailed information about their atomic and electronic interfacial structure and the related superconducting transport properties. We found evidence for interfacial electron transfer from PCMO to YBCO and further intrinsic underdoping of the cuprate on a 3 nm scale by oxygen depletion, presumably as an adaptation to a compressively strained environment. The corresponding reduction in the critical temperature of superconductivity proves to be stronger than that in the case of primarily magnetic impacts reported for superlattices with ferromagnetic manganites.