Charge distribution across capped and uncapped infinite-layer
neodymium nickelate thin films

Discovery of superconductivity in Sr-doped infinite-layer nickelate thin films ¹ has drawn an upsurge of interest being infinite-layer (IL) nickelates structurally and electronically analogous to cuprates. Apart from superconductivity, recent studies have also revealed the existence of charge ordering phenomena and magnetic excitations in IL-nickelate thin films ^{2, 3, 4, 5}, the amplitudes of which mostly depend upon the particular sample preparation, $i.e.$ presence/absence of a SrTiO₃ (STO) capping-layer and doping level. In particular, charge order (CO) has been observed in uncapped-IL NdNiO₂ samples and it is not much pronounced in the capped ones, which on the contrary host dispersing magnetic excitation with a bandwidth of circa 200 meV ^{2, 4}. Such a marked dichotomy naturally raises the question of the structural stability of the uncapped IL-nickelate thin films, compared to that of the capped ones. Indeed, by considering previous reports about the perovskite nickelates ^{6, 7, 8}, one could infer on similar grounds, peculiar oxygen dynamics in IL-nickelates, such as vacancies or incomplete re-intercalation. The latter may distort the IL-crystalline structure, subsequently their electronic properties. Synthesis of IL-nickelates is obtained via a so-called topotactic reduction of the precursor perovskite RNiO₃ (RNO3, R being a rare-earth) thin films which involves removing apical oxygens in RNO3 thereby reducing it to the IL-nickelate RNiO₂ (IL-RNO2) ¹. Such a pathway involves reduction of the nominal valence of Ni from 3+ (in RNO3) to 1+ in (RNO2). This partially unstable state ^{9, 10} can be prone to several structural and/or electronic reconstructions, and for a stable IL-RNO2 an SrTiO₃ (STO) capping-layer has been used in the 2-25 nm thickness range ^{11, 12}. On these grounds, the possible structural reconstructions in the uncapped IL-nickelate thin films stays unexplored. An atomically resolved structural and spectroscopic study which compares both capped and uncapped samples is necessary to obtain new insights to delineate a better understanding of such aforementioned contrasting differences.

Here we have investigated STO-capped and uncapped IL-NdNiO₂ thin films grown onto STO single crsytals as substrate, and hereafter referred to as c-NNO2 and uc-NNO2, respectively. By combining spectroscopic and microscopic techniques, specifically hard x-ray photoemission spectroscopy (HAXPES), scanning transmission electron microscopy (STEM) with monochromated electron energy loss spectroscopy (EELS), we conclude that the measured structural and electronic modifications are linked to a specific O re-intercalation path and charge reconstruction, which could contribute to the observed CO.

We start by discussing the real space structural aspects of both c-NNO2 and uc-NNO2 samples as obtained from STEM-HAADF imaging and geometrical phase analysis (GPA) ¹³. As shown in the STEM-HAADF images in Figs.1A and B, clear structural differences are observed between these two samples. The c-NNO2 shows an IL-structure in the majority of the thin-film, except the small off tilt-like defect in the middle as reported ¹⁴. On the other hand, the uc-NNO2 shows an IL-structure near the interface, with certain periodic faint contrast stripes on the top part of the thin-film which will be discussed in more details in the following section. Fig.1C shows the o-o-p strain maps ($\epsilon_{zz}$) obtained by GPA in the c-NNO2, with reference to the STO o-o-p parameter of 3.91 Å.  A compression of around 16% giving an o-o-p parameter of 3.30 Å is observed throughout the thin-film (blue region in Fig.1C), except in the defect region in the middle, indicating overall a good infinite-layer crystalline quality of this c-NNO2. Fig.1D shows such a strain map of the uc-NNO2, where a compression of around 16% is observed near the interface, giving an infinite-layer o-o-p parameter of 3.30 Å. However, the map shows a decrease of the compression to around 7% giving an o-o-p parameter of 3.65 Å on the top (green region in Fig.1D). The region of this o-o-p expansion coincides with the region where we observe certain low contrast stripes for this uc-NNO2 sample. The HAADF image in Fig.1B for the uc-NNO2 is from a region with more extended stripes, while other regions show stripes more restrained extending of only ca. 2 nm from the surface (Supporting Information, Figure S1). The two samples show no differences in the in-plane strain ($\epsilon_{xx}$) map (Supporting Information, Figure S2), and a homogeneous in-plane parameter of 3.91 Å matching with STO is observed throughout the thin-film. The Fourier transform analysis shown in Figs.1E and F for the c-NNO2 and uc-NNO2, respectively further describes the differences observed in HAADF and GPA. The c-NNO2 shows a pair of spots along q_z around (1 0 1) r.l.u. that matches with the different o-o-p parameters of STO and the infinite-layer structure. In contrast, the uc-NNO2 shows three spots along q_z around (1 0 1) r.l.u. indicating the o-o-p parameters of 3.91 Å,  3.30 Å,  and 3.65 Å respectively from the STO, infinite-layer and the stripe region. Here, we also observe extended spots located at q_x = q_z = $\frac{1}{3}$ r.l.u., and q_x = q_z = $\frac{2}{3}$ r.l.u. The reduced lattice units (r.l.u.), is defined with in plane components a = b = 3.91 Å,  and the out-of-plane (o-o-p) lattice constant c = 3.65 Å.  We chose the o-o-p as 3.65 Å  because this is the parameter of the stripe region. Interestingly, similar faint contrast stripes can be observed in the STEM-HAADF image of other uncapped nickelate samples where CO is reported (SI Fig. S5 of ³), but were unnoticed. It further anticipates the connection between CO and these (3 0 3) stripes.

As mentioned by Krieger $et$ $al.$ in ², the CO signal decreases with the level of Sr-doping on NdNiO₂. In (Supporting Information, Figure S4), we show our STEM-HAADF and GPA analysis on a 5% Sr-doped uncapped Nd_0.95Sr_0.05NiO₂ (uc-NSNO2) sample. We don’t observe any stripes or o-o-p expansion as in the case of uc-NNO2. There are certain fluorite and Ruddlesden-Popper defects on the top of the sample, which are reportedly observed in these systems ¹¹. This also rises possible connection of CO to the presence of such charge-stripes as observed in uc-NNO2. Another aspect here is about the structural stability of undoped samples compared to the doped ones. As mentioned in ³, a comparative study between uncapped samples that are both doped and undoped, indicates a difference in the o-o-p parameter for certain undoped samples. Also, as shown in the (Supporting Information, Figure S3 and Figure S4), the Fourier transform shows no stripe intensities in the regions dominated by additional Fluorite-type defects, suggesting that defects can reduce re-intercalation processes. In the case of 5% Sr-doped sample, the presence of the dopant probably changes the chemical pathway for O re-intercalation. Doping possibly causes the stabilization of particular phases, that prevent the formation of the stripe order.

Fig.2B shows a magnified Fourier transform analysis of the HAADF image showing stripes in the uc-NNO2 (Fig.2A). Here, the observed periodicity of the stripes contributes to extended rods in reciprocal space located at q_x = q_z = $\frac{1}{3}$ r.l.u., and q_x = q_z = $\frac{2}{3}$ r.l.u., with a $\Delta\text{q}_{z}$ $\approx$ 0.24 r.l.u. In the inverse-Fourier transform analysis Fig.2C, this appears as coherent domains that have a typical area of 10 x 1.5 nm². From our observations throughout the thin-film, these coherent domains appear as quasi-2D sheets of ca. 1.5 - 2 nm extension along the o-o-p direction. When the stripes areas extend over a substantial part of the thin film, several sheets are coexisting within the depth as shown in Fig.2C, but areas with only one sheet can be also observed (Supporting Information, Figure S1). The stripe wavevector values reported here are in match with the CO wave-vector of Q = ($\pm\frac{1}{3}$, 0) r.l.u. as found in the previous studies of these samples ², and others ^{3, 5} by studying the quasi-elastic scattering intensity of resonant inelastic x-ray scattering (RIXS) experiments. Recently, a q_z resolved resonant x-ray scattering (RXS) study has been carried out on an infinite-layer PrNiO₂ thin film in which the CO has the Q = ($\pm\frac{1}{3}$, 0, 0.365) r.l.u. wavevector ¹⁵. The reported value of q_z = 0.365 r.l.u. is within the extension of the rod in Fig.2B, extending between q_z = 0.22 to 0.46 r.l.u. In the Fourier transform in Fig.1E, no such rods are observed in the reciprocal space of the c-NNO2 sample. This goes in hand with the absence of CO reported in these c-NNO2 samples ². Fig.2D shows a magnified HAADF image in the stripe region, showing a faint dark contrast along the stripes, and between the rare-earth sites. A better understanding of this region is obtained from high-resolution 4D-STEM analysis. By collecting the whole diffraction pattern at each pixel, one can obtain a high resolution real space atomic mapping with good oxygen contrast ^{16, 17}. Here, in Fig.2E, by employing such an integrated Centre of Mass (iCOM) analysis by 4D-STEM, we have a clear atomic-level mapping including oxygen. It indicates periodic partial intercalation of apical oxygen in the IL crystalline structure. If we consider a perovskite NNO3 structure, this could be interpreted as periodic apical oxygen vacancies (Vo). This finding is of paramount importance as it has been well demonstrated the influence of Vo in changing the local electronic structure especially in perovskite nickelates ⁷. Since these Vo run along the stripes, they have the same periodicity as them, contributing to intensities at Q = ($\pm\frac{1}{3}$ 0 $\pm\frac{1}{3}$) r.l.u. and Q = ($\pm\frac{2}{3}$ 0 $\pm\frac{2}{3}$) r.l.u. in the reciprocal space.

The oxygen re-intercalations observed on the stripe-region of uc-NNO2, probably induce similar hole doping effects as in ¹⁸ changing the Ni valence at these sites. The charge distribution is reflected in the spectroscopic studies, XAS showing a mixed valence and RIXS experiments already demonstrating a broken symmetry of the uc-NNO2 ² in reciprocal space. Since, now we have an understanding of the real space location of these stripes, HAXPES and monochromated STEM-EELS can probe the concomitant charge modulation as well.

HAXPES gives a depth resolved macroscopic understanding of the influence of this structural modification on the local electronic structure. Figs.3A and B shows the HAXPES data in both bulk-sensitive and surface-sensitive configurations. This is done to better discriminate signals stemming from the top and the bulk/interface regions of the samples. The bulk sensitive mode is obtained at 10^∘ incidence angle of the incoming photons, with an estimated probing depth of 10 nm at 3000 eV as obtained by SESSA simulations ¹⁹. In the surface sensitive mode, the incidence angle of 80^∘ gives a probing depth of circa 2 nm at the same photon energy to get similar energy resolution. In both conditions, we compare the Ni 2p core level for the c-NNO2 and uc-NNO2. As expected, there are strong differences between the two samples and incidence conditions of the photons.

In the bulk-mode, it can be seen that the main Ni 2p_3/2 peak shows a shift between capped and uncapped, with a shoulder at a high binding energy (HBE) and low binding energy (LBE) respectively. The separation between them is almost 1.6 eV. This difference gets stronger as we go to surface-sensitive configuration, indicating a strong difference between the top part of the uc-NNO2 with that of the c-NNO2. The shift to high binding energy in the uc-NNO2 indicates a more oxidized Ni species, closer to Ni²⁺ than to Ni¹⁺, as in the case of IL-NNO2 structure. In Fig.3C, this difference is depicted more clearly by comparing the uc-NNO2 Ni-2p spectra at 10^∘ and 80^∘. The decrease of LBE shoulder as we go more surface indicates the presence of more Ni²⁺ or higher valence at the top of the thin-film. Such a strong trend is not observed in the c-NNO2. A similar comparison of the Nd 3d photoemission spectra for both samples rendered no such peculiar differences of the core level (Supporting Information, Figure S5).

The c-NNO2 being a perfect infinite-layer thin film, hosts Ni¹⁺ in a 3d⁹ configuration. It is to note that the c-NNO2 exhibit a very weak satellite peak located at almost 9 eV at higher binding energy from the main peak. It is rather different from previously reported soft X-ray PES that was having a strong satellite at only ca. 6 eV from the main edge ²⁰. The discrepancy might occur, since our hard X-ray photoemission probes the bulk part of the capped sample, which is structurally a perfect infinite-layer. A weaker satellite peak at higher energy would indicate larger charge-transfer energy, pleading for a stronger Mott-Hubbard character²¹ of the NNO2 than the one already infered from the PES ²⁰, but more in accordance with the previously XAS and RIXS report of holed-doped system²².

In the case of uc-NNO2, as observed by 4D-STEM, there is still a partial presence of apical oxygen, resulting in hole doping that is changing the valence towards Ni²⁺. The HAXPES of Fig.3 is characterized by a strong satellite peak located at a higher binding energy around 865 eV. The Ni species may be described as a mixture of Ni d⁸ and Ni d${}^{9}\underline{L}$ configuration, where $\underline{L}$ describes a hole in the ligand orbital, resulting in satellite and main edges. Comparing the Ni 2p photoemission spectrum from the top part of the uc-NNO2 with previously reported cases for NiO ²³ and reduced SrNiO₃ nickelate ²⁴, the present case is more comparable to the latter. It indicates that the Ni ion here is not exactly as for Ni²⁺ in NiO which is characterized by a strong d^{8 23} ground state, but bear more ligand-hole contribution. HAXPES is highly sensitive to changes in local chemistry, but it also gives an averaged macroscopic signal. The strong LBE shoulder that appears together with the main peak in the uc-NNO2 at 10^∘ incidence (bulk-mode) is indeed due to this averaging of signals from Ni sites in the IL-NNO2 near the interface and the higher valence Ni at the surface as due to possible oxygen re-intercalation.

A higher spatial and energy resolved spectroscopic comparison between c-NNO2 and uc-NNO2 samples is obtained by monochromated STEM-EELS. In Fig.4, such a spatial variation at Ni-L, O-K, and Nd-M edges is compared between both samples. In Figs.4B and C, a clear difference has been observed for the Ni-L₃ edge in the uc-NNO2, as we compare the spectrum from near the substrate interface, and to the top part of the thin film. The peak maxima shows a shift to a higher energy loss at the top of the thin film. Fig.4D shows the O-K edge in this sample, where no pre-peak feature is observed near the interface as expected for an IL-NNO2, while having a stronger pre-peak feature around 527 eV going to the top of the thin film. In the case of the c-NNO2 sample, as shown in Figs.4G - I, we observe no differences for the Ni-L and O-K edges (except near the small defect area in the middle), confirming existing data¹². Interestingly, the O-K edge in the whole volume of the c-NNO2 sample shows no pre-peak feature except the central defective region, indicating the good stabilization of the IL crystalline structure.

The observed spatial spectroscopic differences in the uc-NNO2 are possibly caused by local electronic reconstructions due to oxygen intercalation into the infinite-layer structure. The shift of Ni L₃ edge to a higher energy loss indicates a hole-doping effect towards a dominant Ni²⁺.

The pre-peak in O-K edge at 527 eV stems from the hybridization between the O 2p and Ni 3d orbitals ²⁵. It is very strong in the case of perovskite NdNiO₃ with Ni being in a 3d${}^{8}\underline{L}$ ground state in the metallic phase. It almost disappears for an ideal infinite-layer NdNiO₂, where the Ni is in a 3d⁹ state, and also for NiO where the Ni is in a 3d⁸ state²³. The observation of a small pre-peak in O-K at the top part of uc-NNO2, while corresponding primary to a Ni²⁺, can be explained as these Ni²⁺ being a mixture of 3d⁸ and also some 3d${}^{9}\underline{L}$ electronic configuration, that is in accordance with the Ni 2p photoemission features in Fig.3. Similar to the Nd-3d photoemission spectra (Supporting Information, Figure S5), the Nd-M edge in Fig.4E shows no difference spatially in the uc-NNO2, and it can be attributed to the low sensitivity of rare earth elements to such local electronic/structural reconstructions.