Controlling the magnetic interactions across heterointerfaces play a central role in spintronics [1]. Due to the entangled spin, orbital, charge, and lattice degrees of freedom in transition metal oxides, the magnetic interactions at oxide interfaces are complicated by the presence of epitaxial strain, oxygen octahedral coupling, charge transfer, orbital hybridization, electron confinement, etc [2–5]. A well-known phenomenon related to the interfacial magnetic coupling is exchange bias (EB), which usually occurs in field-cooled ferromagnetic/antiferromagnetic systems [6]. EB is characterized by a horizontal shift of the magnetic hysteresis loop along with an enhanced coercive field (H _C), and is usually interpreted by the pinning effect from uncompensated surface spins of the antiferromagnetic material. For a fully compensated antiferromagnetic surface, such as the (001) plane of a G-type antiferromagnet, a spin-flop coupling is energetically favorable [7]. The antiferromagnetic spin axis aligns perpendicular to the ferromagnetic spins to minimize the interfacial spin frustration, as revealed at La_0.7Sr_0.3MnO₃/La(Sr)FeO₃ interfaces [8–10]. Normally such spin-flop coupling is unable to induce EB, but only increases H _C, while extrinsic disorders (interface roughness, for example) can create random fields acting on the ferromagnetic spins and cause EB [11,12]. Also, an intrinsic mechanism involving Dzyaloshinskii-Moriya interaction has been proposed to explain the EB at ferromagnetic/G-type antiferromagnetic interface [10,13]. In particular, it has been reported that the orbital hybridization and superexchange interaction between Mn and Fe at the interface of La_2/3Sr_1/3MnO₃/BiFeO₃ gave rise to a spin-canted state of Fe³⁺ and a concomitant EB effect [14–16]. However, first-principles calculations ascribe the magnetism of Fe³⁺ to the interfacial Mn/Fe chemical intermixing [17], which is supported by the absence of EB at the chemically abrupt interface [16]. These results point to the crucial role of interfacial chemical profile in dictating the magnetic coupling across oxide interfaces.

Pulsed laser deposition (PLD) has been widely used to construct oxide interfaces. Although tremendous achievements have been made in this field over the past few decades, the abruptness across oxide interfaces remains a controversial issue [18]. The high temperature required for the epitaxial growth as well as the high-energy species generated in the laser plume naturally facilitate the inter-diffusion between two adjoined layers [19,20]. The resulted composition variation can profoundly influence the interface properties by introducing disorders, distorting the local structure and/or altering the valence states [16,21–24]. In this work, the effects of interfacial cation interdiffusion on the magnetic properties of LaMnO₃/LaFeO₃ (LMO/LFO) heterostructures have been studied using a combination of atomically resolved scanning transmission electron microscopy (STEM), electron energy-loss spectroscopy (EELS), bulk magnetometry and resonant x-ray reflectivity (RXR). In particular, the degree of Mn/Fe intermixing is found to depend on the growth sequence, i.e., the LMO-on-LFO interface is relatively sharp, while the LFO-on-LMO interface shows pronounced Mn/Fe intermixing over ∼8 unit cells (uc) of the interfacial region. Such structural asymmetry is further manifested by the distinct magnetic properties of the two bilayers.

The LMO/LFO bilayers were fabricated on TiO₂-terminated SrTiO₃ (001) substrates using PLD, in situ monitored by reflection high energy electron diffraction (RHEED). In order to achieve TiO₂-terminated and terraced surface, the SrTiO₃ substrates were first etched with buffered hydrofluoric acid, followed by annealing in flowing oxygen at 1050°C for 2 h. During the PLD deposition, the substrate temperature and oxygen pressure were maintained at 700°C and 0.01 mbar, respectively. The laser fluence and repetition rate were set at 1.8 J/cm² and 2 Hz, respectively. The surface morphology was verified using atomic force microscopy (AFM). The atomic resolution monochromated STEM-EELS was acquired on a Thermofisher Scientific Titan electron microscope, equipped with Cs-correction for the probe and an Enfinium spectrometer, operated at 300 kV. The bulk magnetic properties were measured using vibrating sample magnetometry on a Quantum Design physical property measurement system. The RXR experiments were performed using an in-vacuum 4-circle diffractometer at the Resonant Elastic and Inelastic X-ray Scattering (REIXS) beamline of the Canadian Light Source (CLS) in Saskatoon, Canada [25]. The measurements were carried out at 20 K with a specular reflection geometry.

Bulk LMO is an A-type antiferromagnet with antiparallel aligned ferromagnetic planes [26]. When grown on SrTiO₃, it adopts a ferromagnetic state due to the electronic reconstruction and/or oxygen non-stoichiometry [27–30]. Bulk LFO is a typical G-type antiferromagnet with a Néel temperature (T _N) up to 740 K [31]. The antiferromagnetism can be retained in its thin-film form down to a thickness of 3 uc [32]. Here, LMO _m /LFO _n bilayers have been grown on TiO₂-terminated SrTiO₃ (001) substrates using PLD, with subscripts m and n indicating the corresponding layer thickness in uc. Figure 1A shows the surface morphologies of two bilayers with reversed stacking sequence, LMO₁₀-on-LFO₁₂ and LFO₁₂-on-LMO₁₀ as sketched in the insets. Both samples show atomically flat surfaces with step terraces of one uc in height, in accordance with the layer-by-layer growth mode observed in RHEED (Supplementary Figure S1).

The magnetic properties of the two samples are compared in Figures 1B,C, together with those of 10-uc LMO and 12-uc LFO plain films for reference. The LFO film is antiferromagnetic showing no magnetic signal, while the LMO film is ferromagnetic with a Curie temperature (T _C) of ∼140 K and saturation magnetization (M _S) of ∼2.1 μ _B/Mn at 10 K. When LMO is deposited on top of an LFO buffer layer, T _C remains unchanged, but M _S drops to 1.6 μ _B/Mn. The reduced magnetization can be understood by magnetic frustrations at the ferromagnetic/G-antiferromagnetic interface with competing exchange interactions, as suggested in La_0.6Sr_0.4MnO₃/La_0.6Sr_0.4FeO₃ superlattices [33]. It is interesting to note that the T _C is unaltered by the magnetic frustrations here, which is probably due to the relatively thick LMO layer as well as the heterogenous magnetic states of LMO [34]. Moreover, Gibert et al. reported a canted antiferromagnetic state in LMO grown on a LaNiO₃ buffer layer, which has been ascribed to the modifications of Mn-O-Mn bond angles induced by LaNiO₃ [24]. Such an effect may also play a role in the suppressed magnetization of the LMO-on-LFO bilayer, given the strong octahedral rotations in the orthorhombic LFO. As can be seen in Figure 1C, the LMO-on-LFO bilayer shows H _C ∼ 890 Oe, much larger than the LMO single film with H _C ∼ 340 Oe. Meanwhile, no EB was observed after cooling to 10 K in a field of 1T (Figure 1D). These observations agree with a spin-flop coupling proposed for ferromagnetic/G-antiferromagnetic systems [12]. For the LFO-on-LMO bilayer, the temperature dependent magnetization curve shows drastically reduced T _C and M _S, as compared to the LMO film. More strikingly, a step-like hysteresis loop was observed (Figure 1C), indicating two separate magnetic switching events during the field sweeping. Similar behavior has also been observed with different layer thickness of LMO (Supplementary Figure S2) This is in stark contrast with the LMO-on-LFO bilayer. As we will show below, the distinct behaviors are associated with the asymmetric Mn/Fe intermixing at the interface, which depends on the stacking sequence.

Figures 2A,C show the STEM-EELS elemental maps of the LFO₁₂-on-LMO₇ and LMO₁₀-on-LFO₁₂ bilayers, respectively. The corresponding layer-resolved EELS intensities of La, Sr, Ti, Mn, and Fe are plotted in the upper panels of Figures 2B,D. For the LFO₁₂-on-LMO₇ interface, a strong cation intermixing of Mn and Fe ions is observed, spanning ∼8 uc across the interface. This is similar to what was observed in the LFO-on-LMO bilayer grown by molecular beam epitaxy [35]. In contrast, the LMO₁₀-on-LFO₁₂ interface is relatively sharp, with Mn/Fe intermixing confined within ∼2 uc at the interface. Such an asymmetric chemical profile has also been reported in the heterostructures of LaVO₃/STO and LMO/LaNiO₃, where the diffuse interfaces are interpreted by preferential ionic surface segregations driven by the difference in the ion radii [24,36]. However, this scenario should not be at play for the LMO/LFO interface considering the identical ionic radii of Mn³⁺ and Fe³⁺ [37]. Further theoretical studies are needed to address the puzzling interdiffusion behavior at the LMO/LFO interface by taking into accounts the interface energy and surface energies of LMO and LFO. By fitting the EELS spectra of Mn (Fe) L_2,3 edge to references of Mn²⁺ and Mn³⁺ (Fe²⁺ and Fe³⁺), the layer-resolved valence states of Mn and Fe were obtained and plotted in the bottom panels of Figures 2B,D. Apparently, no charge transfer between Mn³⁺ and Fe³⁺ occurs in both heterostructures, in line with the previous report [35]. The presence of Mn²⁺ at the LMO/STO interface (Figure 2B) is due to electron accumulation driven by the polar discontinuity [28,29,38]. In contrast, no electronic reconstructions are observed at the LFO/STO interface because of the stable 3d ⁵ electronic configuration of Fe³⁺ [39]. Instead, Nakamura et al. showed that the polar charges were screened by an emergent spontaneous polarization at the LFO/STO heterointerface [40]. Overall, our STEM measurements demonstrate the LMO-on-LFO interface being chemically sharper than the LFO-on-LMO interface, and preclude any charge transfer between LMO and LFO.

RXR measurements were performed on a LFO₁₂-on-LMO₁₀ bilayer to gain insights into the step-like magnetic switching behavior. X-ray reflectivity at off-resonant energies were first measured to extract the chemical depth profile. By tuning the x-ray energy to the Mn and Fe resonances, the depth profiles of Mn and Fe magnetization can be extracted independently. In general the on-resonance x-ray reflectivity for a particular element provides an enhanced sensitivity to the depth profile of the density of that element. On resonance, the enhanced refractive index generally leads to stronger contrast in the constructive/destructive interference in x-ray reflectivity. Figure 3A shows the x-ray reflectivity at off-resonant energies (455, 633, 698, and 841 eV). The atomic concentrations of Ti, Mn, and Fe are extracted from the simulations and shown in Figure 3C. The Mn and Fe intermix over ∼7 uc around the interface, in good agreement with the STEM results. In order to resolve the magnetic switching process, RXR measurements were performed at both saturation (0.6 T) and remanence (Rem) states using circularly left (R_L) and right (R_R) polarized light. The asymmetry curves (R_L-R_R)/(R_L+R_R) measured at the L₃ edges of Mn (641 eV) and Fe (710 eV) are depicted in Figure 3B, and the derived magnetic depth profiles are shown in Figure 3D. By applying a field of 0.6 T, we observed a pronounced Fe magnetization in the intermixed region, which peaked around Z ∼ 40 Å. According to Goodenough-Kanamori-Anderson rules, the superexchange interaction between Mn³⁺ (t _2g ³,e _g ¹) and Fe³⁺ (t _2g ³,e _g ²) are strongly ferromagnetic [41–43]. Thus, the Fe moments can be induced by interacting with the surrounding Mn³⁺ ions [17]. The Fe magnetization is limited by the small volume fraction of Fe ions for Z < 40 Å, while the increasing Fe-O-Fe antiferromagnetic interactions suppress the magnetization for Z > 40 Å. Surprisingly, the Fe moments vanished as the magnetic field was removed. This might correspond to a magnetic cluster (or superparamagnetic) state in the intermixed LaMn_1-x Fe _x O₃ [44], whose magnetization becomes randomly aligned at zero fields. Notably, the magnetization of Mn is also depressed to a great extent after removing the magnetic field, even in the region without significant Mn/Fe intermixing. This means that some domains are already reversed at zero fields in LMO. Based on our RXR results, the magnetic switching sequences are sketched in Figure 3E. The step-like hysteresis loop observed for the LFO₁₂-on-LMO₁₀ bilayer corresponds to the separating switching of the intermixed LaMn_1-x Fe _x O₃ and the “clean” LMO sublayers.

To further ascertain the above scenario, we used LaAlO₃ as a shielding layer to prohibit the interfacial intermixing between LMO and LFO. As shown in Figure 4, the F-A-M trilayer exhibits a very similar behavior to the LMO single film, with slightly higher T _C and M _S. In contrast, a considerably larger M _S was observed for the M-A-F trilayer at low temperature, implying the significant role of capping layer in enhancing the magnetization of LMO [45]. Furthermore, the three samples show slim loops at 10 K with the same H _C ∼ 400 Oe, ruling out any exchange couplings at the interface. The single loop of the M-A-F trilayer strongly suggests that the step-like switching behavior is associated with the LFO/LMO interface.

To summarize, we demonstrate an asymmetric interdiffusion at the interface between LMO and LFO, which depends on the growth sequence. The LMO-on-LFO bilayer with relatively sharp interface shows a single magnetic hysteresis loop with enhanced H _C as compared to the referenced LMO film. However, for the LFO-on-LMO case, the interfacial layer with strong Mn/Fe intermixing constitutes an extra magnetic phase. This phase switches separately from the LMO layer, yielding a step-like magnetic hysteresis loop. We further show that the strong interdiffusion at LFO-on-LMO interface can be prohibited by inserting a LaAlO₃ shielding layer. Future experiments, to elucidate on the mechanisms as well as control of interdiffusion, should include the insertion of thin buffer layers, change of the substrate terminations, variations of the thicknesses and finally deposition parameters. Our results call for more investigations into the interfacial chemical profile to interpret those emergent phenomena at complex oxide interfaces.