Strain-Engineered Metal-to-Insulator Transition and Orbital Polarization in Nickelate Superlattices Integrated on Silicon

creating a cuprate-like Fermi surface through heterostructuring. Bulk LNO is a paramagnetic metal lacking any long-range ordering phenomena at all Epitaxial growth of SrTiO 3 (STO) on silicon greatly accelerates the monolithic integration of multifunctional oxides into the mainstream semiconductor electronics. However, oxide superlattices (SLs), the birthplace of many exciting discoveries, remain largely unexplored on silicon. In this work, LaNiO 3 /LaFeO 3 SLs are synthesized on STO-buffered silicon (Si/STO) and STO single-crystal substrates, and their electronic properties are compared using dc transport and X-ray absorption spectroscopy. Both sets of SLs show a similar thickness-driven metal-to-insulator transition, albeit with resistivity and transition temperature modified by the different amounts of strain. In particular, the large tensile strain promotes a pronounced Ni orbital polarization for the SL grown on Si/STO, comparable to that reported for LaNiO 3 SL epitaxially strained to DyScO 3 substrate. Those results illustrate the ability to integrate oxide SLs on silicon with structure and property approaching their counterparts grown on STO single crystal, and also open up new prospects of strain engineering in functional oxides based on the Si platform. X-Ant-Em instrument operated at 300 kV. The temperature dependent resistivities were measured with a van der Pauw geometry on a Quantum Design physical property measurement system. The XAS experiments were performed on the XTreme beamline at Swiss Light Source. [45]


DOI: 10.1002/adma.202004995
Successful epitaxy of SrTiO 3 (STO) directly on silicon using molecular beam epitaxy represents a milestone toward the monolithic integration of multifunctional oxides into the mainstream semiconductor electronics. [1] However, oxide superlattices (SLs) remain scarcely explored on silicon, hindered by the difficulties in growing oxide layers on silicon with a 2D mode. Over the past two decades, forming SLs has been demonstrated to be a powerful tool to engineer oxide functionalities as well as to explore novel electronic states by manipulating the dimensionality, interfacial, and/or interlayer interactions. [2][3][4] LaNiO 3 (LNO)-based SLs have aroused enormous interest due to the possibility of creating a cuprate-like Fermi surface through heterostructuring. [5,6] Bulk LNO is a paramagnetic metal lacking any long-range ordering phenomena at all Epitaxial growth of SrTiO 3 (STO) on silicon greatly accelerates the monolithic integration of multifunctional oxides into the mainstream semiconductor electronics. However, oxide superlattices (SLs), the birthplace of many exciting discoveries, remain largely unexplored on silicon. In this work, LaNiO 3 /LaFeO 3 SLs are synthesized on STO-buffered silicon (Si/STO) and STO single-crystal substrates, and their electronic properties are compared using dc transport and X-ray absorption spectroscopy. Both sets of SLs show a similar thickness-driven metal-toinsulator transition, albeit with resistivity and transition temperature modified by the different amounts of strain. In particular, the large tensile strain promotes a pronounced Ni x y 3d 2 2 − − orbital polarization for the SL grown on Si/STO, comparable to that reported for LaNiO 3 SL epitaxially strained to DyScO 3 substrate. Those results illustrate the ability to integrate oxide SLs on silicon with structure and property approaching their counterparts grown on STO single crystal, and also open up new prospects of strain engineering in functional oxides based on the Si platform.
temperatures. The nominal Ni 3+ ion has a 3d 7 configuration with filled t 2g orbitals and one electron residing in the doubly degenerate e g orbitals. To remove the degeneracy and obtain a single band Fermi surface with only x 2 −y 2 character, two main strategies have been proposed, namely tensile epitaxial strain and quantum confinement. [7][8][9][10][11] Although lots of experimental efforts have been made by combining these two effects, the observed orbital polarization is much smaller than theoretically predicted. [9] This has been attributed to the strongly correlated nature of LNO. The on-site Hund interaction and negative charge transfer energy favor a high-spin Ni 3d 8 L (L represents an oxygen hole) state, which is not susceptible to orbital polarization. [12] Aside from orbital control, the thickness-driven metalto-insulator transition (MIT) is another long-standing issue in LNO thin films and the underlying mechanism has not been fully understood. Scherwitzl et al. reported a crossover from metallic to strongly localized behavior upon reducing the film thickness, while weak localization (WL) was observed in the intermediate thickness range. The results point to the importance of disorder in triggering the MIT. [13,14] In sharp contrast with the scenario of gradual localization of electrons, angleresolved photoemission spectroscopy revealed a sudden breakdown of the Fermi liquid-like quasiparticles at a critical LNO thickness of 2 unit cells (uc). [15] This has been ascribed to the onset of a spin/charge ordered state in the quasi 2D LNO sheet, resembling the emergent antiferromagnetic order reported in LNO/LaAlO 3 (LAO) SL with the same LNO layer thickness. [3,16] Furthermore, the buckling of NiO 2 planes and removal of apical oxygen at the surface have also been invoked as being the driving forces for the insulating states in ultrathin LNO films. [17,18] Strain effect has been extensively studied in LNO by growing epitaxial films on oxide substrates possessing different lattice constants. [10,11,[19][20][21] In particular, for LNO films grown on Si, an additional thermal strain will come into play which arises from the large differences in thermal expansion coefficients (TECs) between LNO and Si. [22,23] Although LNO has been widely used as electrodes for ferroelectric capacitors integrated on Si, [24][25][26] the impact of thermal-strain on its properties has not yet been assessed. Here, we compared the strain states, thickness-driven MITs and orbital states in LNO/LaFeO 3 (LFO) SLs grown on STO single crystal and STO-buffered Si (Si/STO) substrates. The in-plane lattice parameter of the SL on Si/STO substrate decays with increasing the layer thickness of LNO, n in uc, but remains larger than the corresponding SL fully strained to STO. Upon reducing n, the SLs on both substrates gradually turn into a strongly localized state at n = 3. The larger strain on Si/STO leads to an enhanced MIT temperature (T MI ) at n = 4 as well as increased resistivity at n = 1, 2. The electron conduction mechanisms are hardly affected by the different amounts of strain in the entire thickness range. In particular, the strain promotes a prominent Ni x y − 3d 2 2 orbital polarization for the SL grown on Si/STO, as revealed by X-ray absorption spectroscopy (XAS). Such orbital polarization, however, is negligible in the corresponding SL grown on STO.
LNO/LFO SLs were fabricated on (001)-oriented STO and Si/STO substrates using pulsed laser deposition in situ monitored by reflection high-energy electron diffraction (RHEED).
The STO buffer layer with thickness of 4 nm was prepared on Si using molecular beam epitaxy and the growth details can be found elsewhere. [27] The layer thickness of LNO was varied from n = 1 to 8 uc, while the LFO layer was set to 5 uc thick. The LNO/LFO bilayer was repeated ten times to form the SL. The spacer layer LFO is a charge-transfer insulator with a band gap of ≈2.2 eV. [28] The stable 3d 5 electronic configuration of Fe 3+ prohibits the interfacial charge transfer between LNO and LFO, [29] as confirmed by XAS measurements shown later in this paper. Regarding the crystal structure, LFO is orthorhombic with a pseudocubic lattice constant a 0 = 3.93 Å, [30] and LNO is rhombohedral with a 0 = 3.838 Å. [21] The LNO and LFO layers grow in a layer-by-layer mode on STO ( Figure S1, Supporting Information), yielding a streaky RHEED pattern and a terraced surface for the n = 1 SL (Figure 1a). By contrast, the corresponding SL grown on Si/STO shows streaks with some intensity modulations, which are caused by the increasing electron transmission through surface grains. [31] Nevertheless, the surface remains smooth with a root-mean-square roughness of ≈2.8 Å (Figure 1b). The orthorhombic symmetry of the topmost LFO layer is manifested by the half-order streaks as indicated with white arrows in Figure 1a,b. [32] Figure 1c,d shows the high-resolution X-ray diffraction (XRD) patterns of LNO/LFO SLs grown on STO and Si/STO, respectively. Two 20-uc thick LNO single films are included for reference. Clear Laue fringes can be appreciated for all the SLs on STO, attesting to their smooth surfaces. The presence of satellite peaks indicates a chemically modulated structure with well-defined interfaces between the component layers. The periodicity deduced from such satellite peaks agrees well with the designed value. For the SLs grown on Si/STO, they are (00l)-single oriented without forming secondary phases. Satellite peaks are observed only for SLs with n ≥ 3, while Kiessig fringes are hardly visible for all SLs. The in-plane epitaxial relationship was ascertained by XRD ϕ scans ( Figure S2a, Supporting Information), where the unit cell of oxide layers was rotated 45° with respect to Si. We also compared the crystallinity for SLs grown on STO and Si/STO by measuring XRD rocking curves around SL (002) pc peaks ( Figure S2b,c, Supporting Information). The SLs on Si shows a typical full width at half-maximum of ≈0.50°, six times larger than that of the SLs on STO (≈0.08°). The deteriorated crystalline quality for the SLs on Si/STO can be related to the defects in STO buffer layer as well as to the thermal strain from Si substrate. [33] For both sets of SLs, as n decreases, the SL peaks gradually shift to lower angles because of the relatively smaller lattice constant of LNO as compared to LFO. The crystal structure at atomic level has been characterized using high-resolution scanning transmission electron microscopy ( Figure S3, Supporting Information). The lattice coherence is maintained within the scanning area for both SLs grown on STO and Si/STO without forming dislocations.
The strain states of LNO single films and LNO/LFO SLs grown on STO and Si/STO are assessed based on XRD reciprocal space mappings (RSMs) as depicted in The LNO single film and SLs on STO share the same Q X value with the substrate, manifesting their coherently strained states. Thus a is derived as 3.905 Å and an in-plane tensile strain of ≈1.7% is applied to the LNO layer. By contrast, all the samples on Si/STO are relaxed from Si substrate. The reflections show a broad feature caused by the in-plane mosaicity. [34] As n increases, the SL reflection on Si/STO gradually shifts to larger Q X values, corresponding to a decreasing a (Figure 2h). This can be understood by the thin nature of the 4 nm STO buffer layer, which is largely relaxed from Si at growth temperature (≈650 °C) and unlikely to fix the in-plane lattice of the relatively thick SL. [22,34] Therefore, a of LNO/LFO SL at 650 °C is determined by the thickness ratio of LNO and LFO, and decays with n given the smaller lattice constant of LNO compared to LFO. During cooling down to room temperature, the lattice of oxide SLs follows the shrinkage of the thick Si substrate, which  has a drastically smaller TEC ≈2.6 × 10 -6 °C −1 as compared to LFO (≈11.6 × 10 -6 °C −1 ) and LNO (≈10 × 10 -6 °C −1 ). [22,35,36] As a consequence, a thermal tension is usually introduced to the oxide films grown on Si. This thermal strain, however, is negligible for the SLs grown on STO substrates. Taken together, for LNO/LFO SLs grown on Si/STO, the strain state of LNO is dictated by the interfacial strain from neighboring LFO layers together with the tensile thermal strain from Si. The contribution of thermal strain is evident for the SLs with n = 1 and 2. They show roughly the same a = 3.946 Å, greater than the bulk value of LFO (a 0 = 3.93 Å). LNO single films show a sharp and elongated spot on STO, but a rather diffuse spot on Si/STO due to its relatively poor crystallinity ( Figure S2c Notably, half-order Bragg diffractions reveal that NiO 6 octahedral rotations are suppressed along the out-of-plane direction to match the lattice of STO in tensile-strained LNO/STO film. [11,21] This is probably accompanied by an additional breathing distortion of NiO 6 octahedra as suggested by density functional calculations. [20] The strain accommodation in LNO plays a pivotal role in the Ni e g orbital occupancy as we will show below. The thickness-driven MITs in LNO/LFO SLs grown on STO and Si/STO are compared in Figure 3a,b. Here the resistances of the highly insulating LFO films grown on STO and Si/STO are outside the accessible measurement range, and the resistivity is calculated by only considering the contribution from LNO layers. The transport properties show the same tendency with decreasing n for SLs grown on the two different substrates. The SLs are metallic (dρ/dT > 0) for n > 6 and turn insulating (dρ/dT < 0) for n < 4, while temperature-driven MIT is observed at low temperature for 4 ≤ n ≤ 6. As shown in Figure S4, Supporting Information, the sheet resistances of the n = 3 SLs cross the quantum of resistance in 2D, ≈25 kΩ, indicative of a crossover from weak to strong electron localization. It is interesting to note that LNO single films grown on STO becomes insulating at a larger critical thickness of 5 uc. [13] The reduced critical thickness in LNO/LFO SLs may benefit from the neighboring LFO layer, which maintains a bulk-like coordination environment (LaO-NiO 2 -LaO) for the interfacial NiO 2 plane. [17] The room-temperature resistivity, ρ (300 K) is plotted as a function of n in Figure 4a. The SLs grown on STO and Si/STO show only slight differences in ρ (300 K) when n ≥ 3, although the SLs on Si/STO show a degraded crystallinity. However, the insulating SLs (n ≤ 2) on Si/STO show a substantially larger ρ (300 K) than the ones on STO. Note that the n = 1 SL on Si/STO is too resistive to be measured. This can be linked to the reduced dimension as well as the large amount of strain (≈2.8%) on Si substrate, both of which narrow Ni-3d bandwidth. [19] Regarding T MI depicted in the inset of Figure 4a, the n = 6 SLs show essentially the same T MI of ≈41 K on both substrates because of the small difference in strain. However, T MI is enhanced by strain from 156 K on STO to 207 K on Si/STO for the n = 4 SL.
To gain further insights into the conduction mechanisms, the transport behaviors are fitted to appropriate models as shown in Figure 4b-d. The ρ-T curves of metallic LNO films and n = 8 SLs are fitted with ρ = ρ 0 + AT α , where ρ 0 is the residual resistivity and α is determined by the dominant electron scattering process. According to Fermi liquid theory, the dominant electron-phonon (electron-electron) scattering at high (low) temperature yields α = 1 (α = 2). [38] Surprisingly, the four samples can be well fitted with α = 1.5 ± 0.05 in a wide temperature range (40-300 K). Such a T 3/2 power law is different from the sputtered LNO/STO film showing a typical Fermi liquid behavior with a T 2 dependence. [13] We notice that this T 3/2 scaling has also been observed in LNO/STO films fabricated by molecular beam epitaxy, as well as in LNO/LAO SLs. [39][40][41] The deviation from the conventional Fermi liquid behavior is ascribed to the presence of bond-length fluctuations when the system is on the verge of MIT. [42] For the n = 6 SLs, the resistivity upturn at low temperature is well described by the WL model, where the self-interference of electron waves around disorders enhances backscattering and thus increases the resistivity. [13] In 2D systems, where σ 0 is the Drude conductivity and T* is related to the mean free path for electron hopping. p is determined by the main inelastic scattering process, that is, the dominant electron-electron collision and electron-phonon scattering give p = 1 and p = 3, respectively. [13] As can be seen from Figure 4c, the SLs grown on STO and Si/STO are fitted with slopes of 1.3 × 10 -5 S and 1.38 × 10 -5 , corresponding to p ≈ 1. Therefore, the electron-electron collision dominates at low temperature, in line with previous studies on LNO/STO single films and LNO/LAO SLs. [13,40] The WL model is further supported by the presence of negative magnetoresistance ( Figure S5, Supporting Information). Interestingly, a butterfly-like feature was observed at 2 K, signifying an emergent magnetism in the system. This is beyond the scope of this paper and deserves further investigation. The insulating SLs with n = 2 and 3 ( Figure 4d) are fitted to the 2D variablerange hopping (VRH) model given by ρ = ρ 0 exp[(T 0 /T) 1/3 ]. Here T 0 can be written as T 0 = 13.8/[k B N(E F )ξ 2 ], with N(E F ) and ξ being the density of states near Fermi level and the electron localization length. Upon reducing LNO layer thickness from 3 to 2 uc, T 0 rapidly increases from 1187 K (803 K) to 2.06 × 10 5 K (6.81 × 10 5 K) for the SLs grown on STO (Si/STO), indicating that electrons become much more localized due to the increasing disorders. Note that the mean hopping energy estimated by E 0 = 1/3k B T 2/3 T 0 1/3 is larger than k B T in the fitting temperature range, validating the use of a 2D-VRH model. [40] Altogether, LNO/LFO SLs gradually evolve from a metallic to a strongly localized state as n decreases. The electron conduction mechanism in LNO is hardly altered by the strain on Si substrate.
To study the effects of the different amounts of strain on the charge and orbital states, we performed XAS measurements on the n = 4 SLs grown on STO and Si/STO as shown in Figure 5. The polarization averaged XAS spectra around Ni and Fe L edges were collected with a bulk sensitive fluorescence yield mode (Figure 5a). By comparison to the reference spectra of Fe 2 O 3 and bulk LNO, [11,43] we can confirm the stable Fe 3+ and Ni 3+ states, and exclude the presence of substantial oxygen vacancies and electron transfer across the interface. In order to probe the orbital states of Ni 3+ e g electrons, we measured the XAS spectra using σ and π linearly polarized photons. Given the overlap between La M 4 and Ni L 3 edges, we focus on the Ni L 2 edge as shown in Figure 5b,c. The linear dichroism is derived as I(ab) − I(c), with I(ab) and I(c) being the in-plane and out-of-plane polarized XAS absorptions, respectively. Here I(ab) = I(σ) and I(c) = [I(π) − I(ab)sin 2 θ]/cos 2 θ, where θ is the angle between the incoming X-ray and sample plane and set at 30° in our experiments. For the SL grown on STO, the spectra are essentially the same for σ and π polarizations and negligible X-ray linear dichroism signal is observed, suggestive of a degenerated e g state as sketched in the inset of Figure 5b. By contrast, for the SL on Si/STO, the σ-polarized spectrum shows a reduced intensity and shift to lower energy with respect to the π-polarized spectrum. This means that the in-plane x 2 −y 2 orbital has a lower energy and higher electron occupancy than the 3z 2 −r 2 orbital. [44] We also measured the spectra with the surface To further quantify this orbital polarization, the hole ratio r is calculated using the sum rule [8] where h z r − 3 2 2 and h x y − 2 2 represent the number of holes in the 3z 2 −r 2 and x 2 −y 2 orbitals, respectively. For the SL grown on Si/ STO, the hole ratio r is calculated to be ≈1.118, slightly smaller than r = 1.135 reported for LNO/DSO SL epitaxially strained to DSO substrate. [9] This is in line with the difference in strain states. The n = 4 SL on STO/Si suffers a tensile strain of ≈2.4%, as compared to the epitaxial strain of ≈2.76% imposed by DSO.
The absence of orbital polarization in LNO/LFO SL grown on STO is rather surprising considering the in-plane epitaxial strain of ≈1.7%. This is, however, in agreement with previous observations in LNO/STO single films, where the strain is accommodated by suppressed NiO 6 octahedral rotations along the out-of-plane axis, together with an emergent breathing distortion of NiO 6 octahedra. [11,20,21] The preserved isotropic crystal field hardly influences the orbital configuration of Ni 3+ ion. In stark contrast with the LNO/LFO SL on STO, we observed a robust Ni x 2 −y 2 orbital polarization in LNO/LAO SL under the same amount of strain, albeit without any energy splitting ( Figure S8, Supporting Information). It has been argued that the stronger ionicity of Al 3+ O 2than Ni 3+ O 2bond gives rise to a reduced hole density for apical oxygen at the interface, which reduces the electron occupancy of Ni 3z 2 −r 2 orbital even without crystal field splitting. [10] This scenario can account for the absence of orbital polarization in our LNO/LFO SL grown on STO by considering the small difference in ionicity between Ni 3+ O 2and Fe 3+ O 2bonds. For the SLs grown on Si/STO and DSO substrates, the larger tensile strain induces a sizable crystal field splitting, as revealed by the energy shift of Ni L 2 edge, and consequent Ni x 2 −y 2 orbital polarization.
To summarize, LFO/LNO SLs have been fabricated on STO and Si/STO substrates. The strain states evolve distinctly in the two sets of SLs as a function of the layer thickness of LNO, n. In contrast with the coherently strained state for all n on STO substrates, the in-plane lattice parameter of the SLs on Si/STO is dictated by the SL composition along with the thermal strain from Si, which gradually decays with increasing n. Both sets of SLs show MITs upon reducing n and the onset of strong electron localization occurs at n = 3. The substantially larger strain on Si/ STO leads to enhanced T MI at n = 4 and increased resistivity at n = 1, 2, but hardly influences the electron conduction mechanisms at all n. In particular, the large tensile strain promotes a pronounced Ni x y − 3d 2 2 orbital polarization in the SL grown on Si/STO, comparable to the LNO SLs epitaxially strained to DSO substrate. Our results demonstrate that thermal strain can be exploited to engineer the orbital states in functional oxides, in analogy to the conventional epitaxial strain. It is interesting to note that the amount of strain for oxide films grown on silicon can be further tuned by elevating the deposition temperature, [22] offering new freedoms to design the strain states. Finally, we hope this study can stimulate more research into the exploration of new functionalities in oxide SLs integrated on silicon.

Experimental Section
The PLD growth of LNO/LFO SLs was conducted at 650 °C at an oxygen partial pressure of 0.04 mbar. The laser fluence and frequency were set to 2 J cm −2 and 2 Hz, respectively. The samples were in situ annealed for 10 min before cooling down to room temperature. The single terminated and terraced STO substrates were obtained by a chemical etching with buffered hydrofluoric acid and a subsequent annealing in flowing oxygen at 950 °C for 90 min. The surface morphology was characterized using atomic force microscopy. The XRD measurements were conducted on the PANalytical-X'Pert material research diffractometer. The cross-sectional scanning transmission electron microscopy high angle dark field images were acquired on the  [43] and ref. [11] are included for reference. b,c) Polarizationdependent XAS spectra around Ni L 2 edge and corresponding linear dichroism (LD) for the SLs grown on STO (b) and Si/STO (c) substrates.