Limits to the strain engineering of layered square-planar nickelate thin films

TL;DR

This study explores the effects and limitations of epitaxial strain on the synthesis and properties of layered square-planar nickelate thin films, revealing strain-induced defects and phase behavior that constrain strain engineering.

Contribution

It provides the first systematic investigation of epitaxial strain effects on layered nickelate thin films, highlighting the limits of strain engineering for these materials.

Findings

Compressive strain enhances film quality for Nd4Ni3O10.

Tensile strain induces rock salt faults but preserves transport properties.

Strain influences phase transition and defect formation in nickelate films.

Abstract

The layered square-planar nickelates, Nd$_{n+1}$Ni$_{n}$O$_{2n+2}$, are an appealing system to tune the electronic properties of square-planar nickelates via dimensionality; indeed, superconductivity was recently observed in Nd$_{6}$Ni$_{5}$O$_{12}$ thin films. Here, we investigate the role of epitaxial strain in the competing requirements for the synthesis of the $n=3$ Ruddlesden-Popper compound, Nd$_{4}$Ni$_{3}$O$_{10}$, and subsequent reduction to the square-planar phase, Nd$_{4}$Ni$_{3}$O$_{8}$. We synthesize our highest quality Nd$_{4}$Ni$_{3}$O$_{10}$ films under compressive strain on LaAlO$_{3}$ (001), while Nd$_{4}$Ni$_{3}$O$_{10}$ on NdGaO$_{3}$ (110) exhibits tensile strain-induced rock salt faults but retains bulk-like transport properties. A high density of extended defects forms in Nd$_{4}$Ni$_{3}$O$_{10}$ on SrTiO$_{3}$ (001). Films reduced on LaAlO$_{3}$ become insulating and form compressive strain-induced $c$-axis canting defects, while Nd$_{4}$Ni$_{3}$O$_{8}$ films on NdGaO$_{3}$ are metallic. This work provides a pathway to the synthesis of Nd$_{n+1}$Ni$_{n}$O$_{2n+2}$ thin films and sets limits on the ability to strain engineer these compounds via epitaxy.

Full text

I Introduction

The discovery of superconductivity in infinite-layer Nd0.8Sr0.2NiO2 thin films reignited an interest in the nickelates as cuprate analogues Li et al. (2019); Osada et al. (2020a); Li et al. (2020a); Osada et al. (2020b); Zeng et al. (2020); Osada et al. (2021); Li et al. (2021a); Zhou et al. (2021, 2021); Zeng et al. (2022); Ren et al. (2021); Lee et al. (2022). More broadly, the infinite-layer nickelates are the $n=\infty$ member of a homologous series of ‘layered square-planar nickelates’, $R_{n+1}$NinO2n+2 or ($R$NiO2)n($R$O2), where $R=$ trivalent rare-earth cation and $n>1$. These compounds host $n$ quasi-two-dimensional NiO2 planes separated by ($R$O2)- spacer layers, as illustrated in Fig. 1. Consequently, the layering $n$ tunes the nickel 3$d$ electron filling. Mapped onto the cuprate phase diagram, the bulk stable $n=3$ compound, Nd4Ni3O8, lies in the overdoped regime with a formal electron count of 3$d^{8.67}$. Indeed, Pr4Ni3O8 single crystals are metallic Zhang et al. (2017), as corroborated by first-principles calculations Nica et al. (2020). The $n=5$ compound, Nd6Ni5O12, has a formal electron count of 3$d^{8.8}$ aligned with optimal doping; thin films were recently found to be superconducting Pan et al. (2021). The layering $n$ also tunes the out-of-plane electronic dispersion: density-functional theory (DFT) calculations suggest that, despite their similar $d$ electron fillings, the electronic structure of Nd6Ni5O12 is more two-dimensional, and thus more cuprate-like, than that of the hole-doped infinite-layer nickelates Zhang et al. (2017); Pan et al. (2021); LaBollita et al. (2022). Proposals to further promote cuprate-like electronic structure and enhance $T_{\textrm{c}}$ in the nickelates include electron doping the lower-dimensional $n=3$ compound Zhang et al. (2017); Pan et al. (2021), decreasing the $c$-axis lattice constant Been et al. (2021), increasing compressive strain via epitaxy Lee et al. (2022); Ren et al. (2021), and applying high pressure Wang et al. (2022). Furthermore, studies of layered square-planar nickelates in powder Lacorre (1992); Poltavets et al. (2006, 2007, 2008, 2010) and single crystal form have revealed a cuprate-like Fermi surface Li et al. (2023), charge/spin stripes Zhang et al. (2016, 2019), orbital polarization Zhang et al. (2017), and large super-exchange Lin et al. (2021); Shen et al. (2022). Layered square-planar nickelate thin films thus form an exciting platform to investigate the role of dimensionality, epitaxial strain, and chemical doping in nickelate superconductivity.

The synthesis of square-planar nickelate thin films, however, remains an immense challenge. Due to their low decomposition temperatures, infinite-layer and layered square-planar nickelates are accessible only via low temperature topotactic reduction of the perovskite $R$NiO3 ($n=\infty$) or Ruddlesden–Popper $R_{n+1}$NinO3n+1 ($n>1$) parent compounds, respectively Crespin et al. (1983); Levitz et al. (1983); Lacorre (1992, 1992); Hayward et al. (1999); Hayward and Rosseinsky (2003); Poltavets et al. (2006). Furthermore, the absence of superconductivity in reduced nickelate powders Li et al. (2020b); Wang et al. (2020); Huo et al. (2022) and bulk single crystals Puphal et al. (2021) to date suggests that external stabilization by a substrate may be required to yield superconductivity. The synthesis of reduced nickelate thin films, however, is complicated by the large $\sim$2–3$\%$ increase in the in-plane lattice parameter upon reduction. To minimize compressive strain in the reduced phase, the parent compound must be synthesized under tensile strain, as shown in Fig. 1. Perovskite nickelates synthesized under $\sim$2.6% tensile strain on SrTiO3, however, exhibit strain-relieving extended defects Yang et al. (2022); Li et al. (2020a); Zeng et al. (2022); Osada et al. (2021); Lee et al. (2020). A dramatic reduction of such extended defects has recently been achieved by synthesizing the parent perovskite under more modest $\sim$1.6% tensile strain on (LaAlO3)0.3(Sr2TaAlO6)0.7 (LSAT) Lee et al. (2022); Ren et al. (2021). Extended defects can also form through the reduction process. For example, LaNiO2 / SrTiO3 exhibits $a$-axis oriented domains which relieve the 1.4% compressive strain Osada et al. (2021), while these domains are absent in NdNiO2 / SrTiO3 with 0.4% compressive strain Lee et al. (2020). Therefore, understanding the strain-dependent stability of both the parent Ruddlesden–Popper and reduced square-planar phases is essential to optimize the synthesis of layered square-planar nickelate thin films.

Here, we discuss the competing requirements for the synthesis and oxygen deintercalation of Nd4Ni3O10 thin films on LaAlO3 (001), NdGaO3 (110), and SrTiO3 (001). We focus on the $n=3$ Ruddlesden–Popper compound because both the oxidized Nd4Ni3O10 and reduced Nd4Ni3O8 compounds have been synthesized as bulk single crystals, allowing us to benchmark the strain states with bulk lattice constants. Fig. 1 tabulates the lattice mismatch, $\epsilon=(a_{sub}-a_{bulk})/a_{bulk}$, of Nd4Ni3O10 and Nd4Ni3O8 on LaAlO3, NdGaO3, and SrTiO3. First, we present the molecular beam epitaxy (MBE) synthesis of Nd4Ni3O10 on the three substrates. We show that Nd4Ni3O10 on SrTiO3 exhibits a high-degree of disorder characterized by a near-equal density of vertical and horizontal rock salt faults; consequently, we do not consider these films for reduction. When synthesized under lesser tensile strain on NdGaO3, a smaller density of vertical rock salt faults forms while mainaining relatively high quality Ruddlesden–Popper ordering. By contrast, Nd4Ni3O10 on LaAlO3 exhibits coherent ordering of horizontal rock salt layers with very few extended defects.

Next, we reduce Nd4Ni3O10 to the square-planar phase, Nd4Ni3O8, on LaAlO3 and NdGaO3. In Nd4Ni3O8 on LaAlO3, we observe regions with pristine square-planar ordering along with disordered regions where the $c$-axis cants locally by as much as 7°, likely a compressive strain relaxation mechanism Kawai et al. (2010); Osada et al. (2021). However, all reduced films on LaAlO3 are insulating. On the other hand, Nd4Ni3O8 and Nd6Ni5O12 on NdGaO3 are metallic and superconducting Pan et al. (2021), respectively, despite the extended defects in the parent compound. Finally, our density-functional theory (DFT) calculations reveal that in-plane strain alters aspects of the $R_{n+1}$NinO2n+2 electronic structure relevant to superconductivity. Our study thus demonstrates a pathway to the synthesis of a superconducting compound and sets limits on the ability to strain-engineer $R_{n+1}$NinO2n+2 thin films via epitaxy.

II Results

II.1 Density-Functional Theory Calculations

We perform DFT calculations to investigate how strain tunes the electronic structure of Nd4Ni3O8 and Nd6Ni5O12. As detailed in Supplementary Note 1, our calculations reveal the following effects:

The charge-transfer energy ($\Delta$, a measure of the $p$-$d$ splitting) increases (decreases) with compressive (tensile) strain. 2. 2.

The rare-earth density-of-states at the Fermi level increases (decreases) with compressive (tensile) strain. 3. 3.

The Ni-$d_{x^{2}-y^{2}}$ bandwidth increases (decreases) with compressive (tensile) strain.

The role of the charge-transfer energy $\Delta$ and rare-earth 5$d$ ‘spectator’ bands in nickelate superconductivity has been extensively debated Botana and Norman (2020); Kitatani et al. (2020); Karp et al. (2022); Li and Louie (2022). In the cuprates, the superexchange interaction $J\sim t^{4}/\Delta^{3}$ and single-band fermiology are widely deemed essential for superconductivity Scalapino (2012); O’Mahony et al. (2022). Thus strain-engineering $R_{n+1}$NinO2n+2 compounds provides an avenue to explore the role of the charge-transfer energy and multi- or single-band fermiology in nickelate and cuprate superconductivity. Next we discuss the synthesis of these compounds under a variety of strain states.

II.2 MBE Synthesis of Ruddlesden–Popper nickelates

The synthesis of Ruddlesden–Popper nickelate thin films shares several of the challenges encountered in the synthesis of perovskite nickelates, including the difficulty in reaching the high oxidation Ni3+ state and the formation of secondary phases like NiO Lee et al. (2020); Li et al. (2020c); Sun et al. (2021); Pan et al. (2022). There are, however, additional challenges unique to the synthesis of Ruddlesden–Popper thin films. In general, the MBE synthesis of $A_{n+1}B_{n}$O3n+1 Ruddlesden–Popper compounds requires the precise sequential deposition of $A$ and $B$ monolayers to achieve the desired composition ($A/B=(n+1)/n$) and monolayer dose Lee et al. (2014); Barone et al. (2021); Pan et al. (2022). Given ideal composition, errors in the monolayer dose times result in the formation of Ruddlesden–Popper layers with an average periodicity different from the one targeted Barone et al. (2021). Errors in composition, on the other hand, can be accommodated via the formation of extended defects such as rock salt faults: half unit cell stacking faults formed by additional $A$O inclusions Balz and Plieth (1955); Ruddlesden and Popper (1957, 1957, 1958). In homoepitaxial Srn+1TinO3n+1, for example, up to 5% excess strontium can be accommodated by the formation of additional SrO rock salt layers, while strontium deficiency results in missing rock salt layers Ohnishi et al. (2005, 2008); Brooks et al. (2009); Xu et al. (2016); Dawley et al. (2020).

In addition to the composition and monolayer dose, lattice mismatch plays a crucial role in the formation of extended defects. In compressively-strained LaNiO3 / LaAlO3 films, for example, misfit dislocations form to relax the in-plane compressive strain Bak et al. (2020); Qi et al. (2021). Synthesized under tensile strain on SrTiO3, LaNiO3 Bak et al. (2020) and NdNiO3 Yang et al. (2022) films exhibit vertical rock salt faults. These extended defects effectively increase the in-plane lattice constant and thus relieve tensile strain because the distance between rare-earth planes within a rock salt layer ($\sim$2.8 Å) is greater than the atomic layer spacing of perovskite (La,Nd)NiO3 ($\sim$1.9 Å) (see Supplementary Note 2). The density of vertical rock salt faults in infinite-layer nickelates has been decreased dramatically by decreasing the tensile strain of the parent perovskite compound Ren et al. (2021); Lee et al. (2022). In comparison to perovskites, however, Ruddlesden–Popper compounds are even more prone to the formation of vertical rock salt faults because their structure already hosts horizontal rock salt layers which can instead orient vertically to relieve the in-plane tensile strain.

Most infinite-layer nickelate films to date have been synthesized on SrTiO3, motivated by the minimal 0.4% compressive strain in the reduced state. Here, we present attempts to synthesize Nd4Ni3O10 under 2.1% tensile strain on SrTiO3. In Fig. 2(a), we present x-ray diffraction (XRD) scans of two Nd4Ni3O10 / SrTiO3 films: one with Nd/Ni $\sim 4/3$ and the other with Nd/Ni $\sim 5.1/3$, a 27.5% excess in neodymium content. We will refer to these two films as ‘stoichiometric’ and ‘neodymium-rich’, respectively. The stoichiometric film exhibits two primary features in the XRD scan: a film peak at $47.4\degree$ and a broad peak at $\sim$26$\degree$. These features are reminiscent of disordered and non-stoichiometric $R$NiO3 Satyalakshmi et al. (1993); Breckenfeld et al. (2014); Li et al. (2021a). However, the stoichiometric film exhibits none of the 00$l$ superlattice peaks expected for bulk Nd4Ni3O10. The neodymium-rich film, on the other hand, exhibits superlattice peaks that are roughly consistent with bulk Nd4Ni3O10. Thus, coherent horizontal rock salt ordering forms only after supplying excess neodymium.

We investigate the nature of this crystalline disorder by atomic-resolution scanning transmission electron microscopy (STEM) imaging of the film. Microscopic signatures of Ruddlesden–Popper disorder evident in STEM include a zigzag arrangement of $A$-cations and reduced $A$–$B$ contrast Detemple et al. (2012); Coll et al. (2017); Yang et al. (2022). As illustrated in Fig. 2(b), a vertical rock salt fault in the (001) plane leads to the relative shift of an adjacent rock salt layer by $a$/2 [011] Tokuda et al. (2011). Consequently, the neodymium and nickel atomic sites are stacked in projection along the propagation direction of the electron beam, resulting in reduced atomic number ($Z$) contrast Dawley et al. (2020) (Fig. 2(c)). We present a medium-angle annular dark field (MAADF)-STEM image of the Nd5.1Ni3O10-δ film in Fig. 2(d). The film is distinguished from the substrate by the bright contrast which arises from the heavy neodymium nuclei in the film. Within the film, however, the clean layered perovskite structure of an $n=3$ Ruddlesden–Popper phase is obscured by a high density of disordered vertical and horizontal rock salt planes. A higher magnification (HAADF)-STEM image of the film shown in Fig. 2(e) similarly indicates a high density of rock salt plane formation through the reduced $Z$ contrast. A near-equal density of vertical and horizontal rock salt faults thus forms in spite of the sequential MBE shuttering sequence encouraging horizontal rock salt layering.

Our attempts to synthesize Nd4Ni3O10 on SrTiO3 demonstrate the extraordinary difficulties in stabilizing coherent horizontal rock salt ordering under large tensile strain. High quality perovskite NdNiO3, on the other hand, can be synthesized on SrTiO3, albeit with a small density of vertical rock salt faults Lee et al. (2020); Yang et al. (2022); Bak et al. (2020). Ruddlesden–Popper nickelates, however, are more likely to form vertical rock salt faults than perovskites due to the composition — in NdNiO3, Nd/Ni $=$ 1/1, while in Nd4Ni3O10, Nd/Ni $=$ 4/3. We thus propose that the additional neodymium content in Nd4Ni3O10 under tensile strain forms strain-relieving vertical rock salt faults instead of horizontal rock salt layers, regardless of the MBE shuttering sequence. In fact, to stabilize any horizontal rock salt ordering, it is necessary to supply an excess of neodymium, as demonstrated in Fig. 2(a). Ruddlesden–Popper films are therefore more sensitive to the formation of tensile strain-relieving extended defects than their perovskite counterparts. Additionally, SrTiO3 is unique out of the substrates included in our study for its charge neutral atomic planes which form a stronger discontinuity to the charged planes in the Nd4Ni3O10 film. Spectroscopic and theoretical studies of the interface between NdNiO2 and SrTiO3 show that a similar polar discontinuity is alleviated by a unit cell thick reconstruction layer of Nd(Ti,Ni)O3 Goodge et al. (2022a), which we also observe in our Nd4Ni3O10 films on SrTiO3 (Supplementary Fig. S23). In this work we therefore focus on epitaxial strain as the primary driver of reduced crystalline quality in these films. Future studies may reveal subtle differences in polar accommodation of infinite- and several-layer nickelates. Given the high density of extended defects observed in Nd4Ni3O10 / SrTiO3, we disqualify this system from consideration for reduction and instead turn our attention to substrates which yield smaller lattice mismatch.

In an effort to decrease the density of extended defects, we next synthesize Nd4Ni3O10 under more modest 0.8% tensile strain on NdGaO3 (Fig. 1). We present an XRD scan of a Nd4Ni3O10 / NdGaO3 film in Fig. 3(a) which exhibits 00$l$ superlattice peaks consistent with bulk Nd4Ni3O10. The splitting of the $00\underline{10}$ peak may be indicative of a small error in the monolayer dose or composition Barone et al. (2021). Reciprocal space mapping in Supplementary Fig. S22 demonstrates that the film is epitaxially strained to the substrate. Cross-sectional STEM imaging in Fig. 3(c-d) provides a microscopic view of the defects. Diffraction contrast near the rock salt planes in the large field-of-view MAADF-STEM image in Fig. 3(c) highlights coherent horizontal ordering of Ruddlesden–Popper layers and additional vertical rock salt faults. A HAADF-STEM image of the film in Fig. 3(d) similarly identifies the coexistence of different Ruddlesden–Popper phases as well as vertical rock salt faults. Boxes and atomic models in Figs. 2(b) and 2(d) denote regions of well-ordered horizontal Ruddlesden–Popper layers (yellow), vertical rock salt faults (red), and local step edges between regions of mixed local $n$ phase (green). Therefore, while some vertical rock salt faults are observed, the lower tensile strain imparted by NdGaO3 reduces the density of vertical Ruddlesden–Popper faults and other extended defects that dominate Nd4Ni3O10 films synthesized on SrTiO3.

To determine whether vertical rock salt fault formation can be further mitigated, we next synthesize Nd4Ni3O10 under 0.9% compressive strain on LaAlO3. The XRD scan of a Nd4Ni3O10 / LaAlO3 film in Fig. 4(a) exhibits sharp superlattice peaks with no peak splitting. Furthermore, reciprocal space mapping in Supplementary Fig. S12 confirms the film is epitaxially strained to the substrate. More detailed structural and electronic characterization of our Ruddlesden–Popper ($n=1-5$) nickelates on LaAlO3 can be found in Ref. Pan et al. (2022). MAADF-STEM images of this film in Fig. 4(c) show nearly uniform adherence to horizontal layering with very few vertical rock salt faults. Atomic-resolution HAADF-STEM imaging in Fig. 4(d) corroborates the high degree of crystalline order with only few defects visible. Deviations from the $n=3$ Ruddlesden–Popper structure is quantified in Supplementary Fig. S7 in Ref. Pan et al. (2022). While such deviations from the targeted $n=3$ Ruddlesden–Popper structure are observed in La4Ni3O10 Drennan et al. (1982); Ram et al. (1986) and reduced Nd4Ni3O8 Retoux et al. (1998) bulk crystals, our films exhibit a small density of such deviations due to the MBE shuttering sequence that encourages the formation of the targeted Ruddlesden–Popper order.

We present a summary of the characteristic crystalline microstructure for Nd4Ni3O10 films under varying amounts of compressive and tensile epitaxial strain in Fig. 5. Strain analysis of the (101) and ($\bar{1}$01) pseudocubic lattice fringes highlight local $a$/2 (011) lattice offsets at both horizontal and vertical Ruddlesden–Popper rock salt planes. Under 0.9% compressive strain on LaAlO3, Nd4Ni3O10 exhibits coherent horizontal Ruddlesden–Popper ordering with some rock salt discontinuities but very few vertical rock salt faults (Fig. 5(a)). Under 0.8% tensile strain on NdGaO3, the horizontal Ruddlesden–Popper layering structure is largely preserved with the emergence of some vertical rock salt planes (Fig. 5(b)). The density of vertical rock salt faults in Nd4Ni3O10 increases dramatically as the tensile strain is increased to $\epsilon=+2.1\%$ on SrTiO3, with a near equal density of vertical and horizontal rock salt faults evident in Fig. 5(c). Such an increase in rock salt fault density is expected as the vertical Ruddlesden–Popper layers relieve tensile strain Bak et al. (2020, 2020); Yang et al. (2022) (see Supplementary Note 2). With these challenges in stabilizing the parent Ruddlesden–Popper compounds in mind, we next consider their oxygen deintercalation. Crucially, much of the cation disorder observed in the parent compounds is preserved through reduction due to the minimal cation mobility at typical reduction temperatures ($\sim$300°C).

II.3 Oxygen deintercalation to the layered square-planar phase

A fundamental issue in the topotactic reduction of nickelates is the metastability of square-planar phases at typical reduction temperatures ($\sim$300°C) Lacorre (1992); Zhang and Greenblatt (1995); Hayward et al. (1999); Poltavets et al. (2007, 2006); Malyi et al. (2022). Decomposition to $R_{2}$O3 ($R$ = La, Pr, Nd), NiO, and nickel metal has been reported at temperatures as low as 210°C for LaNiO2 Hayward et al. (1999) and 200°C for NdNiO2 Hayward and Rosseinsky (2003). For layered square-planar compounds, the decomposition temperatures are higher: 400°C for La4Ni3O8 Lacorre (1992) and 375°C for La3Ni2O7 Poltavets et al. (2006). This difference may be ascribed to the suppressed out-of-plane cation mobility in Ruddlesden–Popper and layered square-planar nickelates, facilitated by the rock salt $R$O or fluorite $R$O2 ‘blocking’ layers, respectively. While nickelate powders can be reduced at temperatures as low as 190°C with metal hydrides Hayward et al. (1999), nickelate thin films typically require higher reaction temperatures ($>$250°C) to stabilize the square-planar phase Kawai et al. (2010); Lee et al. (2020), possibly because films are not ground with the reductant like powders. As a result, decomposition has been observed in reduced infinite-layer thin films Onozuka et al. (2016); Ikeda et al. (2013, 2016); Lee et al. (2020); the precise identification of the decomposition products is, however, difficult in thin films in part due to the small film volume. An additional challenge associated with metal hydride reductions is the possibility of hydrogen incorporation in the lattice, an effect which has been proposed as a reason for the absence of superconductivity in some films Si et al. (2020); Malyi et al. (2022); Si et al. (2022). For example, an oxyhydride NdNiOxHy phase was reported in NdNiO3 films reduced with CaH2 Onozuka et al. (2016), but to date no experiment has linked insulating behavior with hydrogen intercalation. Solid state reduction techniques have recently been demonstrated as a promising alternative to metal hydride reductions Wei et al. (2023). Thus to minimize the formation of decomposition products and potential hydrogen intercalation during metal hydride reductions, careful optimization of reduction duration and temperature is essential.

Additionally, the dramatic expansion (contraction) of the in-plane (out-of-plane) lattice parameter through reduction further complicates the stabilization of high quality square-planar nickelates. After reduction, La1-xCaxNiO2 single crystals exhibit three orthogonally-oriented domains of the infinite-layer phase separated by micro-cracks Puphal et al. (2021). Nevertheless, La1-xCaxNiO2 and Pr4Ni3O8 Puphal et al. (2021) single crystals are metallic but not yet superconducting to date. In contrast, most reduced infinite-layer Wang et al. (2020); Li et al. (2020b); He et al. (2021); Huo et al. (2022) and layered nickelate powders Poltavets et al. (2006, 2007, 2008, 2010); Sakurai et al. (2013); Hao et al. (2021); Li et al. (2021b) are insulating — metallic (Pr,Nd)4Ni3O8 pellets were, however, achieved with reductions using a sulfur getter to promote the removal of apical oxygens Nakata et al. (2016); Miyatake et al. (2020). Thus metallicity in square-planar, or $T^{\prime}$, nickelates may be exquisitely sensitive to the oxygen content like in $T^{\prime}$ cuprates Brinkmann et al. (1996). Additional proposals for the the lack of superconductivity in bulk reduced nickelates include nickel deficiency and the nucleation of ferromagnetic nickel grains Sharma et al. (2022).

Thin films, on the other hand, have several advantages that address these issues. First, thin films host an inherent in-plane versus out-of-plane anisotropy defined by the substrate that encourages the formation of a single orientation of the reduced phase. Furthermore, the thin film geometry minimizes the size of potential ferromagnetic nickel clusters and promotes reduction uniformity due to the large surface area exposed to the reductant. Thin films, however, face a challenge absent in bulk compounds: defects may form in response to reduction-induced compressive strain. In LaNiO2 / SrTiO3, for example, macroscopic Kawai et al. (2010) and local Osada et al. (2021); Ikeda et al. (2016) $a$-axis oriented infinite-layer domains form, likely to relieve the 1.4% compressive strain Osada et al. (2021). This effect was also demonstrated in LaNiO2 films which exhibit an increasing fraction of $a$-axis oriented LaNiO2 with increasing compressive strain Ikeda et al. (2013). While strain is an important factor in defect formation, synthesis optimization of the parent compound has suppressed the formation of such defects upon reduction Osada et al. (2021); Lee et al. (2022). The reduction-induced strain-relieving mechanisms in layered nickelates may be different than those observed in infinite-layer nickelates.

With these challenges in mind, we now turn to reductions on LaAlO3, which yield the highest quality as-synthesized Ruddlesden–Popper nickelates (Fig. 5), but also the highest compressive strain in the reduced state (Fig. 1). We first reduce perovskite NdNiO3 on LaAlO3 to study how a system without horizontal rock salt layers undergoes an increase in compressive strain from 0.5% to 3.3% upon reduction. We present the XRD scans of an as-synthesized NdNiO3 and reduced NdNiO2 film on LaAlO3 in Fig. 6(a). The as-synthesized film exhibits the 002 NdNiO3 peak at $\sim$47.6$\degree$ with no second phases evident over the full scan range (Supplementary Fig. S10). In the reduced film, we observe a peak at $\sim$46.3$\degree$ along with a lower intensity peak at $\sim$56.0$\degree$, which likely correspond to the 200 and 002 peaks of NdNiO2, respectively. Incremental reduction of the film results in the immediate formation of the $a$-axis oriented NdNiO2 phase (Supplementary Fig. S10) which suggests this phase is not formed a result of over-reduction, as was observed in LaNiO2 / SrTiO3 Kawai et al. (2010). Furthermore, NdNiO2 / LaAlO3 exhibits insulating electrical transport (Supplementary Fig. S10). These results suggest that NdNiO3 undergoes large-scale $a$-axis reorientation upon reduction under high compressive strain, as illustrated in Fig. 6(b).

Next we reduce Nd4Ni3O10 on LaAlO3 to investigate how rock salt layering impacts the structural transformation to the square-planar phase. The reduction-induced increase in compressive strain of the Ruddlesden–Popper (from 0.9% to 3.2%) is comparable to that of the perovskite (from 0.5% to 3.3%). In Fig. 6(c), we present XRD scans of Nd4Ni3O10 and Nd4Ni3O8 films on LaAlO3. The parent film exhibits all expected Nd4Ni3O10 00$l$ peaks and, as shown in Fig. 6(e), a metal-to-insulator transition at $\sim$150 K consistent with previous reports Sun et al. (2021); Pan et al. (2022). Upon reduction, the Nd4Ni3O10 00$l$ peaks shift toward the Nd4Ni3O8 00$l^{\prime}$ peaks, revealing the formation of the square-planar phase. Furthermore, the lack of 200 NdNiO2 or 220 Nd4Ni3O8 peaks at $\sim$46.3$\degree$ suggests that the film has retained $c$-axis orientation through reduction, as illustrated in Fig. 6(d). Reciprocal space maps in Supplementary Fig. S12 however demonstrate that the Nd4Ni3O8 film is partially relaxed with a 3.89 Å in-plane lattice constant compared to the 3.915 Å bulk value. XAS at the oxygen K-edge in Supplementary Fig. S13 further corroborates the formation of the square-planar phase. Thus, in contrast to NdNiO3, Nd4Ni3O10 retains global $c$-axis orientation upon reduction, despite the high compressive strain on LaAlO3. Like NdNiO2 / LaAlO3, however, the reduced layered nickelate is insulating (Fig. 6(e)). In fact, all films reduced on LaAlO3 are insulating, likely due to reduction-induced structural disorder (Supplementary Note 4). Furthermore, of the three strain states studied here, $R_{n+1}$NinO2n+2 films under high compressive strain on LaAlO3 are the furthest from cuprate-like due to the larger charge-transfer energy and higher neodymium DOS at $\varepsilon_{\mathrm{F}}$ (Supplementary Note 1). Thus if optimized to a metallic state, $R_{n+1}$NinO2n+2 / LaAlO3 films could be a platform to study the role of the charge transfer energy and neodymium 5$d$ states in nickelate superconductivity.

Cross-sectional STEM measurements shown in Fig. 7 reveal the microstructure of the reduced Nd4Ni3O10 / LaAlO3 film. The large field-of-view MAADF-STEM image in Fig. 7(a) shows some decomposition near the surface and diagonal defects between crystalline regions throughout the film. Atomic-resolution HAADF-STEM imaging in Fig. 7(b) shows one such clean region with horizontal layer ordering similar to the as-synthesized film (Figs. 4 and 5(a)). The reduced square-planar structure is visible by close inspection of the atomic lattice in Fig. 7(c) which shows a more pronounced in- versus out-of-plane anisotropy of the pseudo-infinite-layer unit cell. Annular bright field (ABF)-STEM imaging in Fig. 7(d) further illustrates the reduced square-planar phase with the absence of oxygen atomic columns in the horizontal neodymium planes.

Interspersed between these regions of pristine Nd4Ni3O8, we find that the diagonal defects observed by MAADF-STEM in Fig. 7(a) are in fact regions of local $c$-axis canting. Using local wavefitting analysis Smeaton et al. (2021), in Fig. 7(d) we map the local $c$-axis orientation across one such diagonal defect, revealing regions of opposite canting of up to several degrees. Within this window of variation ($\sim\pm 10\degree$), however, the film’s $c$-axis remains globally out-of-plane. By contrast, XRD scans in Fig. 6(a) suggest that a perovskite NdNiO3 film on LaAlO3 can also be reduced to NdNiO2, but that most of the film attains $a$-axis orientation with only a tiny peak corresponding to $c$-axis orientation remaining visible. The Ruddlesden–Popper structure therefore appears to stabilize the $c$-axis orientation even under high 3.2% compressive strain, preventing the large-scale reorientation observed in the infinite-layer counterpart on LaAlO3.

Under more modest (1.4%) compressive strain, LaNiO2 / SrTiO3 exhibits a small fraction of $a$-axis reorientation along diagonal planes Osada et al. (2021), similar to those observed in the triple-layer films here. An important distinction between the infinite- and triple-layer systems is that of the heavy cation lattice and the oxygen-filled (or reduced) planes. In the perovskite / infinite-layer system, the orientation of empty oxygen planes effectively defines the local $c$-axis (orthogonal to the rare-earth planes). Without the imposition of additional symmetry by oxygen occupancy, however, the cation lattice directions are essentially equivalent within the pseudocubic approximation. Local reorientation can therefore be equivalently described as local rearrangement of occupied oxygen sites. In the Ruddlesden–Popper structure, on the other hand, the heavy cation lattice bears inherent symmetry distinctions even without the oxygen lattice, with the cation $c$-axis defined as orthogonal to the rock salt planes. Here, we observe a global preservation of this $c$-axis direction upon reduction: all the Ruddlesden–Popper layers remain in the same uniform plane from the as-synthesized to reduced phase (the energy scale of cation mobility that would be required for Ruddlesden–Popper reorientation is far above the temperature of the topotactic reduction process). How closely, then, is the oxygen lattice tied to this pre-defined cation symmetry?

Locally, we find nanometer-scale regions near the canting defects which suggest an oxygen lattice reorientation internal to the global Ruddlesden–Popper structure. Supplementary Fig. S29 shows a high-magnification image of the defect structure mapped in Fig. 7(e). The atomic overlay in the bottom right of the image – where the film is epitaxial, uncanted, and well-ordered – shows how the reduced structure can be mapped onto a rectangular sublattice with longer (shorter) in-plane (out-of-plane) atomic spacings. The rectangles outline 3 $\times$ 3 neodymium atomic sites, with the long and short dimensions colored cyan and yellow, respectively. We observe an identical (but slightly rotated) structure on one side of the canted lattice near the top left of the image. Between the two, however, atomic distances in a subset of the positively-canted region are better described by a near-90$\degree$ rotation of the 3 $\times$ 3 rectangle. Even where the planar Ruddlesden–Popper structure is preserved, the in-plane spacings are shorter than the out-of-plane spacings, suggesting the internal oxygen lattice in this region differs from elsewhere in the film. Extracting the precise atomic structure of these defects will be challenging given the higher-degree of disorder in the surrounding lattice and the small total volume they comprise. While the observation of such competing reorientation likely does not strongly impact the macroscopic properties of the films studied here, it may inspire future efforts to decouple the various atomic sublattices in these or other layered materials.

These results suggest that to mitigate strain-induced extended defects upon reduction, the compressive strain in the reduced state should be minimized. NdGaO3 is an appealing option as the compressive strain of the reduced compound is limited to 1.5%, compared to 3.2% on LaAlO3 (Fig. 1). In Fig. 8, we present the reduction of the same Nd4Ni3O10 / NdGaO3 film discussed in Fig. 3. Upon reduction, the superlattice peaks in Fig. 8(a) shift toward positions consistent with the square-planar structure. Unlike the films on LaAlO3 which partially relax upon reduction (Supplementary Fig. S12), we observe that Nd4Ni3O8 / NdGaO3 is epitaxially strained to the substrate (Supplementary Fig. S22). As shown in Fig. 8(b), the as-synthesized Nd4Ni3O10 film exhibits a metal-to-metal transition at $\sim$150 K, consistent with the bulk compound Greenblatt (1997); Li et al. (2021b). Grown instead on LaAlO3, Nd4Ni3O10 possesses a metal-to-insulator transition (Fig. 6(e) and Refs. Sun et al. (2021); Pan et al. (2022)). The reduced Nd4Ni3O8 / NdGaO3 is metallic with a resistive upturn, similar to the infinite-layer nickelates Li et al. (2019); Puphal et al. (2021) and Pr4Ni3O8 Zhang et al. (2017); Miyatake et al. (2020). More details on the reduction of Nd4Ni3O10 / NdGaO3 can be found in Supplementary Note 5. Thus, metallic and eptiaxially-strained layered square-planar nickelates can be stabilized on NdGaO3.

Cross-sectional STEM images of the reduced Nd4Ni3O8 / NdGaO3 film in Fig. 8 are presented in Fig. 9. The large-scale MAADF-STEM image in Fig. 9(a) shows Nd4Ni3O8 ordering of fair crystalline quality, disordered regions with extended defects, as well as reduction-induced disorder at the surface. We observe the same effects in the HAADF-STEM images in Fig. 9(b)-(d) which show regions of varying representative levels of crystalline order. All three regions (and especially Region 1 in Fig. 9(b)) show signs of reduced crystallinity near the film surface which are likely reduction-induced and not a result of TEM sample preparation (e.g. ion beam damage). In infinite-layer thin films, stabilizing capping layers of SrTiO3 have proven an effective way to reduce similar surface degradation Lee et al. (2020). Near the top left of Region 2 in Fig. 9(c), minor $c$-axis canting similar to that discussed in Fig. 7 can also be observed. All three regions show areas with significant disorder which may have nucleated near vertical rock-salt faults in the as-synthesized film (see Figs. 3 and 5(b)). Regions without amorphization but exhibiting vertical rock salt faults are highlighted with yellow arrows. In a relatively clean region of the film, HAADF- and ABF-STEM images in Fig. 9(e-f) show the expected $n=3$ ordering and square-planar structure. Notably, the overall crystallinity of the Nd4Ni3O8 / NdGaO3 film in Fig. 9 is higher than the crystallinity of the superconducting quintuple-layer nickelate Pan et al. (2021), likely due to the challenges in stabilizing the higher order $n=5$ phase. Both reduced $n=3$ and $n=5$ compounds on NdGaO3, however, exhibit similar reduction-induced disorder, likely nucleating at vertical rock salt regions formed during the synthesis of the parent compound.

II.4 Cation stoichiometry

Optimized Nd4Ni3O8 ($n=3$) and Nd6Ni5O12 ($n=5$) films on NdGaO3 are metallic and superconducting, respectively Pan et al. (2021). However, the metallicity of these films is highly sensitive to as-synthesized structural differences and reduction conditions (Supplementary Fig. S21 and Supplementary Fig. S2 in Ref. Pan et al. (2021)). In the infinite-layer nickelates, cation stoichiometry strongly influences metallicity and superconductivity Lee et al. (2020); Li et al. (2021a). We thus investigate the role of cation stoichiometry in the metallicity of Nd4Ni3O8 on NdGaO3. In Fig. 10(a) we present XRD scans of five Nd4Ni3O10 films with systematically varying neodymium content. The structural quality of the Nd4Ni3O10 films deteriorates as the neodymium content is varied from the optimal value, evident by the broadness and reduced intensity of the $008$ peak. Electrical transport measurements in Fig. 10(b) demonstrate that the resistivity of the Nd4Ni3O10 films is relatively insensitive to cation stoichiometry, except for the 6% neodymium-rich film which exhibits higher resistivity than the rest of the films in the series. The decrease in resistive upturn temperature with increased off-composition is consistent with the decrease in metal-insulator transition temperature with neodymium-richness in NdNiO3 Breckenfeld et al. (2014). We provide more details regarding the MBE synthesis and properties of these films in Supplementary Note 8.

Next we reduce the films shown in Fig. 10(a) and present the XRD scans in Fig. 10(c). The three films within 3% of optimal stoichiometry exhibit all Nd4Ni3O8 $00l$ peaks, while the two films that deviate from optimal stoichiometry by 6% exhibit substantially diminished XRD peak intensity. These differences in structural quality are corroborated by the resistivity measurements in Fig. 10(d). The films that deviate from optimal stiochiometry by 6% are insulating, while the films closer to optimal stoichiometry are metallic (more reduction trials are provided in Supplementary Figs. S16–S20). These results suggest that Nd4Ni3O8 films can be metallic only if the cation stoichiometry lies within $\sim$3% of the optimal value. It was similarly demonstrated that Nd1-xSrxNiO2 films are insulating if the cation stoichiometry is off by 10% Li et al. (2021a). However, we also observe metallicity in Nd4Ni3O8 films that are structurally inferior to those shown in Fig. 10 (Supplementary Fig. S21(c)). The metallicity of reduced Nd4Ni3O8 films is thus intricately dependent on a variety of structural factors including cation stoichiometry, oxygen content, and extended defect density.

III Discussion

In Fig. 6, we explore the role of rock salt layers in topotactic reductions. While the reduction of Nd4Ni3O10 yields $c$-axis oriented Nd4Ni3O8 with local $c$-axis canting, the reduction of NdNiO3 primarily stabilizes $a$-axis oriented NdNiO2. The most important distinction between NdNiO3 and Nd4Ni3O10 is the presence of rock salt layers in the Ruddlesden–Popper, which transform into fluorite-like NdO2 layers through reduction (Fig. 1). Our results in suggest that rock salt and fluorite-like spacer layers suppress $a$-axis reorientation and promote the deintercalation of apical oxygens along planes parallel to the spacer layers, as discussed in Supplementary Fig. S29. These spacer layers additionally facilitate the stabilization of the square-planar phase under much higher compressive strain than currently possible in infinite-layer systems. This capability is particularly appealing as additional compressive strain has thus far enhanced $T_{\textrm{c}}$ in infinite-layer nickelates Ren et al. (2021); Lee et al. (2022), although it is unclear to what extent this enhancement can be attributed to improved crystallinity. Additionally, oxygen diffusion is known to be anisotropic in Ruddlesden–Popper compounds, in contrast to perovskites Lee and Lee (2017); Tomkiewicz et al. (2015); the influence of rock salt layers on reduction kinetics, however, remains to be investigated.

An additional issue raised by our work is the nature of the insulating and metallic states in reduced layered nickelates on LaAlO3 and NdGaO3, respectively. We speculate that in Nd4Ni3O8 / LaAlO3, the compressive strain-induced extended defects shown in Fig. 7(e) preclude metallic transport through the film, despite the prevalence of high-crystallinity regions shown in Fig. 7(b). Nevertheless, the potential role of the twinned structural domains in LaAlO3 should be considered as well. The typical size a twin domain in LaAlO3 is $\sim$1-100 µm Burema et al. (2019). The distance between $c$-axis canting defect regions in Nd4Ni3O8 / LaAlO3, on the other hand, is $\sim$100 nm, as shown in Fig. 7(a). The length scale associated with $c$-axis canting defects is thus at least an order-of-magnitude smaller than the size of LaAlO3 twin domains. This difference in length scale suggests the $c$-axis canting defects may form independently of the twin domains.

In summary, we have synthesized Ruddlesden–Popper Ndn+1NinO3n+1 and layered square-planar Ndn+1NinO2n+2 films under multiple strain states. We show that attempts to synthesize Nd4Ni3O10 under 2.1% tensile strain on SrTiO3 result in an extremely disordered structure with a high density of vertical rock salt faults, disqualifying these films from consideration for reduction. Synthesized under 0.8% tensile strain on NdGaO3, Nd4Ni3O10 exhibits horizontal rock salt ordering with a small density of vertical rock salt faults. We synthesize the highest quality Nd4Ni3O10 films under 0.9% compressive strain on LaAlO3, with few extended defects observed. Thus compared to perovskite nickelates, Ruddlesden–Popper films are more prone to the formation of tensile strain-induced extended defects because the horizontal rock salt layers can instead orient vertically to form strain-relieving rock salt faults. Therefore, minimizing the as-grown tensile strain is crucial to decreasing the extended defect density in the reduced compounds, as was demonstrated in the infinite-layer nickelates Lee et al. (2022). Minimizing tensile strain is particularly crucial in the layered nickelates due to the increased propensity for the formation of strain-relieving extended defects compared to the infinite-layer nickelates.

We reduced the Ruddlesden–Popper nickelate films on LaAlO3 and NdGaO3 to the layered square-planar phase. In Nd4Ni3O8 / LaAlO3, we observe $c$-axis canting diagonal defects interspersed between regions of high crystalline quality. All films reduced on LaAlO3 are insulating, likely due to reduction-induced structural disorder. Reduced NdNiO2 / LaAlO3 is also insulating but, in contrast to the layered nickelates, is primarily $a$-axis oriented. Horizontal rock salt layers in the Ruddlesden–Popper structure thus suppress $c$-axis reorientation during reduction under high compressive strain. Reduced Nd4Ni3O8 films on NdGaO3, on the other hand, demonstrate high-quality layered square-planar ordering as well as regions with extended defects such as vertical rock salt faults. Despite the presence of these extended defects, Nd4Ni3O8 ($n=3$) and Nd6Ni5O12 ($n=5$) films on NdGaO3 are metallic and superconducting, respectively Pan et al. (2021). The metallic state in Nd4Ni3O8 / NdGaO3, however, can only be stabilized if the neodymium content lies within $\sim$3% of the optimal quantity. The competing requirements for the MBE synthesis of the parent compound and reduction to the square-planar phase are thus met by synthesizing the parent compound under tensile strain on NdGaO3 to accommodate the large increase in the in-plane lattice parameter upon reduction, at the cost of forming strain-induced vertical rock salt faults in the as-synthesized film.

Through our systematic study across multiple substrates, we establish a method to synthesize layered square-planar nickelate thin films and set limits on the ability to strain-engineer these compounds. Furthermore, our DFT calculations demonstrate that Nd4Ni3O8, if optimized to a metallic state on LaAlO3 and SrTiO3, could be a platform to study the role of the charge-transfer energy and rare-earth 5$d$ states in nickelate superconductivity. Finally, this work provides a comprehensive starting point from which to launch future investigations into the role of epitaxial strain, dimensionality, and chemical doping in nickelate superconductivity.

IV Methods

IV.1 Molecular beam epitaxy synthesis and CaH2 reduction

We employ ozone-assisted MBE to synthesize the Ruddlesden–Popper nickelates on LaAlO3 (001), NdGaO3 (110), and SrTiO3 (001). To calibrate the nickel and neodymium elemental fluxes, we synthesize NiO on MgO (001) and Nd2O3 on yttria-stabilized zirconia (YSZ (111)), then measure the film thickness via x-ray reflectivity Sun et al. (2022). Next, we synthesize NdNiO3 / LaAlO3 (001) and use the $c$-axis lattice constant and film thickness to refine the Nd / Ni ratio and monolayer dose, respectively Li et al. (2021a). Using the optimized neodymium and nickel shutter times from the synthesis of NdNiO3 / LaAlO3, we synthesize the Ruddlesden–Popper nickelates via monolayer shuttering. Both NdNiO3 and Ruddlesden–Popper nickelates are synthesized with \sim$$1.0\mathrm{e}{-6} Torr distilled ozone (Heeg Vacuum Engineering) and 500°-600°C manipulator temperature. The MBE synthesis conditions and calibration scheme are described in Refs. Pan et al. (2022, 2021); similar techniques were also used in Refs. Li et al. (2020c); Sun et al. (2021).

The Ruddlesden–Popper films were reduced to the layered square-planar phase using CaH2 topotactic reduction. The as-synthesized films are cut into $\sim$2.5$\times$5-mm2 pieces, wrapped in aluminum foil (All-Foils), then inserted into borosilicate tubes (Chemglass Life Sciences) with $\sim$0.1 grams of CaH2 pellects ($>$92%, Alfa Aesar). The borosilicate tube is sealed at $<0.5$ mTorr using a small turbomolecular. The sealed glass ampoule is heated in a convection oven (Heratherm, Thermo Fisher Scientific) for several hours at $\sim$290°C, with a 10°C min*-1* heating rate. After reduction, the film is rinsed in 2-butanone and isopropanol to remove CaH2 residue. Similar methods are commonly used elsewhere Lee et al. (2020); Li et al. (2021a); Pan et al. (2021).

IV.2 Structural characterization

X-ray diffraction (XRD) measurements were performed using a Malvern Panalytical Empyrean diffractometer with Cu K$\alpha_{1}$ ($\lambda=1.5406$ Å) radiation. Reciprocal space maps (RSMs) were taken with a PIXcel3D area detector. Lattice constants were calculated using Nelson-Riley fits of the superlattice peak positions Nelson and Riley (1945).

Cross-sectional STEM specimens were prepared using the standard focused ion beam (FIB) lift-out process on a Thermo Scientific Helios G4 UX FIB or a FEI Helios 660. High-angle annular dark-field (HAADF-) and medium-angle annular dark field (MAADF-) STEM images were acquired on an aberration-corrected Thermo Fisher Scientific Spectra 300 X-CFEG operated at 300 kV with a probe convergence semi-angle of 30 mrad and inner collection angles of 66 mrad (HAADF) or 17 mrad (MAADF). Annular bright field (ABF)-STEM images were acquired on an aberration-corrected 300 kV FEI Titan Themis with a probe convergence semi-angle of 21.4 mrad and 12 mrad inner collection angle. Additional HAADF-STEM measurements were performed on a Thermo Fisher Titan Themis Z G3 operated at 200 kV with probe convergence semi-angle of 19 mrad and inner collection angle of 71 mrad.

Lattice-scale disorder in the as-synthesized films is visualized by extracting modulations in the (101) and ($\bar{1}$01) pseudocubic lattice fringes using the phase lock-in method described in Refs. Goodge et al. (2022b); Fleck et al. (2022) and implemented by the Python analysis package publicly available at https://doi.org/10.34863/amcp-4s12. In particular, vertical and horizontal rock salt planes or rock salt faults appear as an apparent strong local compressive strain in the pseudocubic lattice fringes. The original HAADF-STEM images used for analysis and the raw output strain maps are provided in Supplementary Fig. S26. Furthermore, regions within each Ruddlesden–Popper layer show small negative strain values: this is due to the choice of the reference lattice spacing which is based on the average image in this Fourier-based technique and is not reflective of real elastic strain in the atomic lattice. To highlight the Ruddlesden–Popper faults, we therefore include only positive strain values in the overlays shown in Fig. 5 with the full maps shown in the Supplementary Fig. S27. The strain map transparency follows the local magnitude of the strain.

The map of local $c$-axis orientation in Fig. 7 is generated with the local wavefitting analysis described in Ref. Smeaton et al. (2021) using the 002 pseudocubic lattice fringes, cropped windows of 24$\times$24 pixels, and a window step size of 12 pixels equivalent to $\sim$0.16 nm. The map displayed in Fig. 7 is additionally smoothed by a Gaussian kernel with $\sigma$ = 5 corresponding to a distance of about 5 pseudocubic unit cells; the original STEM image, raw wavefitting output, and smoothed result are provided in Supplementary Fig. S27.

Electron energy loss spectroscopy (EELS) measurements were carried out on the same instrument described above equipped with a Gatan Continuum spectrometer and camera. Spectrum images of the films on LaAlO3 and NdGaO3 were acquired with a spectrometer dispersion of 0.3 eV per channel. Spectrum images of the film on SrTiO3 were acquired operating in DualEELS mode with a spectrometer dispersion of 0.15 eV per channel. The accelerating voltage was 300 (120) kV for measurements of the films on SrTiO3 and NdGaO3 (LaAlO3). Due to their overlapping EELS edges, two-dimensional concentration maps of the La-M4,5 and Ni-L2,3 edges are determined by non-negative least squares (NNLS) fit to the weighted sum of reference components for each edge taken from substrate (La-M4,5) and film (Ni-L2,3) regions (Supplementary Fig. S25).

IV.3 Electrical transport

Electrical transport data were primarily taken using a Quantum Design Physical Property Measurements System equipped with a 9 T magnet. Hall bars were patterned with Cr (5 nm) / Au (100 nm) contacts using shadow masks which were then defined using a diamond scribe. Resistivity data was taken down to 1.8K at $\sim$3°C/min using an AC lock-in amplifier. Hall coefficients were determined from linear fits of antisymmetrized field sweeps up to 9T. All field sweeps were taken upon warming.

Several electrical transport measurements in the Supplementary Materials were taken using a home-built electrical ‘dipstick probe’ compatible with a helium dewar. Indium contacts were soldered on the four corners of each film in a Van der Pauw configuration. AC transport measurements were taken at 17.777 Hz using SR830 lock-in amplifiers. The voltage and current were measured simultaneously to determine the resistance.

IV.4 X-ray Absorption Spectroscopy

X-ray absorption spectroscopy (XAS) was performed at the Advanced Light Source, Lawrence Berkeley National Lab, at Beamline 6.3.1. The spectra were taken in the total electron yield mode at 300K with linear horizontally polarized light. The film was oriented either normal ($I_{x}$) or at 30° grazing incidence ($I_{z}$) to the beam. For the signal acquired at grazing incidence, a geometric correction factor was applied Wu et al. (2004). The spectra are normalized to the incident x-ray flux and scaled to the same intensity at energies just below the absorption edge. Every spectrum we present is an average over eight pairs of spectra measured in normal ($I_{x}$) and grazing ($I_{z}$) x-ray incidence angle.

**Data Availability

**Source data for x-ray diffraction and electrical transport in Figs. 2-10 and high-resolution STEM images contained within the main text are provided in the Source Data file. Additional data which support the findings of this study are available from the corresponding authors upon reasonable request.

Additional information

Supplementary information is available for this paper.

Correspondence and requests for materials should be addressed to Julia A. Mundy.

Reprints and permissions information is available.