The large choice in chemical elements in either A- and B-site of the perovskite structure allows tailoring many parameters including lattice constants, electronic band structures, magnetic interactions, and more. In particular, manganese based oxides display a huge range of properties from total spin polarization and colossal magnetoresistance to multiferroicity, with potential use from active electrodes in electronic devices to sensors and magnetic memories. Furthermore, nanostructured thin films widen their application range as their final functional properties are determined by the very active role of interfaces and surfaces which may be efficiently modified by strain relaxation mechanisms [1–4]. Specially in the case of undoped LaMnO₃ (LMO), the appearance of ferromagnetism in heterostructures of LaMnO₃/SrTiO₃ [5] and LaMnO₃/SrMnO₃ [6], or the observation of exchange bias in LaMnO₃/LaNiO₃ superlattices [7] evidence the importance of interfacial phenomena. In addition to standard heterointerfaces in layered structures, new strategies have been developed to increase the surface to volume ratio and to enhance interfacial effects. In this sense, vertically heteroepitaxial nanocomposites (VHN) provide a vertical contact area between two immiscible compounds much larger than substrate to films. Additionally, lateral strain may be tuned to large film thicknesses [8].

Spontaneous organization of vertical heteroepitaxial nanocomposite thin films has enabled the discovery of new physical phenomena at the nanoscale and their possible use in magnetoelectric devices. For instance, large adjustable low-field magnetoresistance is observed in vertically aligned La_0.7Ca_0.3MnO₃ [9] or La_0.7Sr_0.3MnO₃ (LSMO) based nanocomposites [10, 11]. A good example of this is the system LSMO:Mn₃O₄ that has been often proposed as potential nanocomposite due to the chemical compatibility of Mn₃O₄ with the whole manganite perovskite family [12]. Indeed, if the composition ratio [La+Sr]/[Mn] < 1, a spinodal decomposition mechanism in chemical solution deposited films may lead to the spontaneous formation of LSMO:Mn₃O₄ VHNs [13]. The so-generated vertical lattice strain is found to be related to the different domain and grain boundary structures and directly impact their ferromagnetic properties [11]. In the particular case of the parent compound, i.e., undoped LaMnO₃, manganese segregation is expected to occur when La deficiency leads to a non-stoichiometric ratio of La/Mn < 0.9 [14]. There, the appearance of randomly distributed segregates of manganese oxides such as Mn₃O₄ or MnO has been reported in bulk samples [15, 16] and thin films [17]. It is important, from a fundamental point of view, to understand the formation of manganese oxide segregates, as it is a characteristic secondary phase in several REMnO₃ (RE = Rare Earth) thin films and heterostructures, thus modifying their structural quality and functional properties [18–20]. Furthermore, manganese oxides are extremely attractive for potential applications due to their electrochemical, optical, or catalytic properties [21–24]. In spite of their abundance in nature reproducible properties of manganese oxides require the fabrication of uniform structures and a large effort is devoted to produce MnO_x nanostructures (for recent Reviews see, for example, [23, 24] and references therein). In the present study, we report on the spontaneous formation of regular vertically aligned nanocolumns composed of manganese oxide embedded in antiferromagnetic LaMnO₃ matrix grown by pulsed laser deposition (PLD) on (001)-oriented single crystal perovskite substrates of LaAlO₃ (LAO) and SrTiO₃ (STO). The formation mechanisms of the manganese oxide secondary phase and their influence on functional properties of manganite thin films are discussed in detail.

Films grown in vacuum conditions exhibit a clear ordered nanocomposite surface, as evidenced by SEM images either on STO or LAO substrates (see Figures 1A, 2A, respectively). It is observed that the inclusions of the secondary phase (manganese oxide) are uniformly distributed in the LMO matrix (primary phase) with typical rectangular-based dimensions of 40 × 150 nm (on STO substrate) and 40 × 90 nm (on LAO substrate). It is worth reminding here that this spontaneous regular formation of manganese oxide in LMO thin films has been obtained by laser ablation of single target and it is linked to the low oxidation conditions during the PLD process. Films grown at higher oxygen pressures exhibit flat surfaces with no sign of phase segregation [27]. More information about the arrangement of the secondary phase may be obtained from atomic force microscopy images as the shown in Figure 1B (on STO) and Figure 2B (on LAO). While the manganite LMO phase appears to be very flat with low roughness (rms < 0.5 nm), MnO_x nanopillars appear as large inclusions exceeding the surface of the matrix by several nm. From this preliminary characterization we may estimate that secondary phase occupies around 10–15% of the total surface, as determined by quantifying black areas (different chemical composition) in the SEM images. An interesting observation is that MnO_x inclusions are mainly oriented with their rectangular sides aligned with the main axes of the substrates. Statistical analysis of the inclusions is presented in the histograms of Figures 1C and 2C. We may see that size and aspect ratio is enhanced in the case of films grown on STO substrates. This suggests that the different mismatch with substrate plays a role in the statistical size distribution of the segregates.

As mentioned above, epitaxial growth of orthorhombic stoichiometric LMO films has been studied by several groups in the past to elucidate the controversial appearance of ferromagnetic behavior. Either strain or oxygen content has been claimed to be at the origin of this anomalous magnetic behavior [28, 29, 45–49]. In LMO bulk state, oxygen off-stoichiometry (in fact, cation vacancy) is accommodated by modifying the Mn-O octahedral environment mainly resulting in a strong change in the orthorhombic lattice parameters that also influences the cooperative Jahn-Teller effect and, as a result, the magnetic ordering. Quite recently we have shown that a careful epitaxial analysis is required to fully understand the growth of LMO films under different oxygen conditions [27]. For this reason, we have performed reciprocal 2θ–ϕ area scans at grazing incidence around 220 and 004 orthorhombic in-plane reflections (corresponding to 200 reflections in pseudocubic notation). Figure 3A shows a typical area scan for LMO sample grown on LAO. Similar results were obtained previously for films grown on STO [27]. We observe the presence of two peaks occurring at different 2θ angles that are the clear signature of twinned domains with two in-plane lattice parameters. To understand this result, we need to consider that growth temperature (T_g = 1123 K) is well above Jahn-Teller transition of LMO (T_JT = 750 K) and, although still exhibiting Pbnm crystallographic structure, all orthorhombic LMO matching distances are nearly equal (a_LMO798K/√2 = 3.9468 Å, b_LMO798K/√2 = 3.9478 Å, c_LMO798K = 3.9448 Å) [50]. Thus, LMO structure is metrically cubic and a cube-on-cube epitaxial relationship with substrate is expected. During cooling, cooperative Jahn-Teller transition takes place in LMO and its MnO₆ octahedral framework is extremely distorted (Mn-O bond distances being 1.903, 1.957, and 2.184 Å). To accommodate this structural distortion to the underlying substrate lattice, film may adopt either (110) or (001) orientation. In such situation, it is useful to consider the area mismatch in the basal plane, i.e., εarea(110)=(d[110]f2−as2)/as2 for (110)-orientation and εarea(001)=(d[110]fcf/2−as2)/aS2 for (001) orientation, where f and s indicate film and substrate, respectively, and d_[110] = (a²+b²)^1/2/2. We may see that ε_area[110]_STO = −0.1% and ε_area[001]_STO = −3.7% for STO and ε_area[110]_LAO = −5.9% and ε_area[001]_LAO = −9.3% for LAO. In both cases, (110) orientation shows clearly a better matching which very likely favors this orientation. In this case, when LMO c-axis is contained in the plane of the substrate, we have two different in-plane Mn-Mn matching distances, d_[110]LMO = 3.982 Å and d_[001]LMO = c_LMO/2 = 3.834 Å (see schema in Figure 3B) leading to twinned domains and the two peaks in Figure 3A. A reliable quantitative analysis of the in-plane cell parameters has been performed by integrating the intensity of the respective area scans along the azimouthal angle phi and fitting the corresponding peaks (see Figure 3C). We have obtained values of 3.952 and 3.875 Å (on STO) and 3.947 and 3.857 Å (on LAO) for d_[110]LMO and d_[001]LMO, respectively. The resulting effective strains are approximately −1% compressive and 1% tensile for STO and −4 and −2% compressive for LAO substrate. These values of residual strain and in-plane lattice parameters indicate that the LMO thin films grown under vacuum conditions are almost in a relaxed state. In the case of the films grown on STO, such relaxed state may be understood if we calculate the corresponding diagonal basal distances of STO, d_[110]STO = 5.523 Å and film, d_LMO = 5.534 Å which show a mismatch of 0.3%. This suggests that LMO tends to accommodate to the underlying substrate along STO[110] direction, i.e., in-plane LMO c-axis is slightly rotated with respect to substrate principal axes (see scheme in Figure 3D) which is characteristic of the formation of LMO twined domains being the twin boundary plane aligned with the [110]STO direction. The formation of these twin domains takes place during the cooling down process below the Jahn–Teller transition temperature. From standard θ–2θ x-ray diffraction measurements we have also determined the pseudocubic LMO out-of-plane cell parameter, obtaining 3.985 Å for LMO on STO and a larger value of 4.006 Å for LMO on LAO. In both cases, our estimation of pseudocubic unit cell volume leads to ~61 ų, close to the expected bulk value for orthorhombic stoichiometric LMO [26]. The relative orientations of film and substrate were also observed by X-ray diffraction reciprocal space maps, as depicted in Figure 4A. The map shows the area around the main symmetrical 002 reflection of LaAlO₃ substrate. The position of the main film peak close to the substrate peak is compatible with 220 LMO, which corresponds to c_||-oriented domains. The observation of 221 LMO reflection indicates the formation of orthorhombic Pbnm LMO (which corresponds to half order 2 0 ½ reflection in cubic primitive notation). The fact that half order reflection lies along Q_x direction supports the c_||-orientation of the LMO. The two separate reflections at lower Q_z values are consistent with 004 reflections from Mn₃O₄ spinel with their c-axis tilted 9° at both sides of the vertical direction. The corresponding pole figure of this 004 Mn₃O₄ reflection is depicted in Figure 4B showing the presence of four reflections at χ ~9° separated by ϕ = 90°. This indicates the textured orientation of Mn₃O₄ inclusions within the LMO matrix.

Further confirmation that partially relaxed LMO films still preserve orthorhombic symmetry has been obtained by studying the Raman response. Figure 5A shows the Raman spectra of LMO nanostructured films at room temperature. In both spectra the same peaks are observed. Those found at frequencies of 257, 244, and 481 cm⁻¹ are associated to A_g symmetry whereas those at 308 and 611 cm⁻¹ correspond to Raman frequency of B_2g symmetry. All of them are fully compatible with those of orthorhombic LaMnO₃ reported for stoichiometric single crystals [51]. Temperature dependence of Raman response for LMO samples grown on LAO substrates is presented in Figure 5B. The most pronounced features in the Raman spectra of LMO/LAO sample in the range of 450–660 cm⁻¹ correspond to the E_g Raman mode of LAO substrate and to the modes of A_g and B_2g symmetry of orthorhombic LMO structure. Note that with lowering of the temperature, B_2g Raman mode shows characteristic softening in the antiferromagnetic phase [52], as shown in the right inset of Figure 5B. Besides, a low-intensity peak at about 640 cm⁻¹, better discerned in the Raman spectra collected at low temperatures (Figure 5B, left inset), can be attributed to the stretching vibration of Mn–O bond in the MnO₆ unit of manganese oxide, possibly nanocrystalline Mn₃O₄. Namely, due to phonon confinement effect, with decreasing mean crystallite size the A_1g Raman mode in nanostructured Mn₃O₄ gradually redshifts compared to its position in bulk crystal (~656 cm⁻¹) [53]. In the case of γ-Mn₂O₃, the main Raman feature is reported to appear at 628 cm⁻¹ and, thus, it cannot explain our results of Figure 5B [54]. So, the appearance of the Raman mode at ~640 cm⁻¹ could be related to the presence of Mn₃O₄ nanocrystallites with mean size of about 30 nm.

The microstructural analysis on the formation of the MnO_x secondary phase ordered nanostructures, their correlation with substrate/film interface and the relaxation of strain was studied in detail by cross-section TEM (Figure 6). The TEM images reveal a wedge-like shape of the nanopillars. Surprisingly, MnO_x columns do not reach the interface with the substrate. However, there is a thin LMO matrix layer of around 5 nm (Figure 6). Taking into account the huge lattice mismatch between manganese oxide and substrate, it looks energetically unfavorable to nucleate directly on top of the substrate and, instead, a thin LMO layer could act as an efficient buffer layer for drastically decreasing the strain. HAADF images (Z- contrast images) presented in Figure 6C show a different contrast between matrix (bright contrast) and nanopillar (dark contrast) indicating different cation composition. Specifically, the chemical analysis, EDS profiles and EELS maps, reveals that the dark zones in HAADF image correspond to a pure manganese oxide whereas the bright zone corresponds to LMO matrix with ratio La/Mn~1. It is noticeable the absence of La inside the inclusions. The analyses of HRTEM images of these nanopillars indicate that manganese oxide has a spinel crystal structure (compatible with Mn₃O₄) with interplanar distances of 5.2 Å in-plane and of 9.2 Å out-of-plane. Further confirmation that nanoinclusions may be associated to Mn₃O₄ phase was obtained by studying the oxygen coordination by EELS. Figure 6D shows the oxygen K edge EELS spectrum obtained at the nanoinclusions. By comparing this result with the EELS spectrum seies of the manganese oxides reported in literature, the best agreement is found for the Mn₃O₄ phase [55, 56].

With respect to the perovskite substrate, all precipitates exhibit a similar crystallographic orientation that may be understood in terms of a distorted cubic spinel. Epitaxy may be defined as (001)-oriented cube-on-diagonal (i.e., in-plane rotation of 45°) but with a slight c-axis tilting of 9°, in agreement with the above x-ray diffraction results. HRTEM images (Figure 6) also show that interfaces between LMO and Mn₃O₄ inclusions are not flat, making difficult to define any contact plane. We should note that zig-zag interfaces between Mn₃O₄ films and manganese perovskite films (in that case La_0.7Sr_0.3MnO₃) have been previously reported [57], suggesting a tendency in this system to avoid flat interfaces. Nevertheless, statistical analysis shows a tendency to form angles of 35 and 10° with the vertical at both sides of the inclusion. These angles are independent of the surrounding LMO orientation. This suggests that shape of the inclusions is completely determined during nucleation and growth at the high temperature stage, regardless of their relative orientation with the twinned LMO domains formed during cooling. The other relevant feature identified by HRTEM images and spectroscopy (EDS and EELS) analysis is the presence of an epitaxial (001)-oriented layer of a La-rich secondary phase. This layer is not directly at the interface with the substrate but embedded in the LMO matrix (see Figure 7). This secondary phase, of around 2–4 nm thick, has interplanar distances of 3.94 Å in-plane and 13.5 Å out-of-plane and may correspond to the rarely observed Ruddlesden-Popper La₂MnO₄ phase [14].

The presence of twin domains in LMO has been also confirmed by cross-section analyses as two different orientations of LMO were identified, i.e., if zone axis is [001]_LMO (noted as LMO_B) or [110]_LMO (noted as LMO_A). As the Mn₃O₄ inclusions are significantly elongated, especially for LMO grown on STO (aspect ratio ~1.7) it is interesting to determine the relative orientation of their long side with respect to the LMO matrix axis. Surprisingly, we have only observed Mn₃O₄ inclusions with their long side aligned with [001]_LMO (see Figure 8) or at the interface between [001]_LMO or [110]_LMO domains (see Figure 9). We have not identified inclusions with long axis fully embedded in [110]_LMO domains. The epitaxial orientation of the precipitates was determined by analyzing electron diffraction FFT images (middle panels in Figures 8, 9) and by taking hausmannite I4₁/amd crystal structure for Mn₃O₄. Although not strictly parallel (disorientations angle between planes are ~1.5–2°) the most accurate description for the corresponding epitaxial relationships are the following: LMOA(-11-2)[110]//Mn3O4(-202)[010]//LMOB(100)[001]LMOB(010)[001]//Mn3O4(01-1)[111]

The lower panels in Figures 8, 9 are schematized representations of both epitaxial relationships where, for clarity, the defining planes are painted in green.

Chemical phase segregation is a complex problem that involves not only thermodynamic stability of the different phases but also kinetic considerations. In general, segregation occurs when the energy cost associated with the creation of new interface and contact area is overcome by the gain in free energy by nucleating a stable secondary phase. Furthermore, in epitaxial thin films, strain associated with the mismatch between lattice parameters of film phases and substrate play an important role and segregates with selected composition may occur to minimize the overall elastic energy of the film. Previous investigations have demonstrated that such mechanism may explain the occurrence of a three phase oxide manganese nanocomposite similar to our present case [58–60]. In those works, stoichiometry unbalance caused by the spontaneous outcropping of La-Sr-O islands during the growth of La_1–xSr_xMnO₃ (LSMO) films is compensated by the appearance of a Sr₃Mn₂O₇ Ruddlesden-Popper phase. The main difference in our present study is that LMO films are almost relaxed while formation of La-Sr-O islands helped to maintain LSMO fully strained state. Thus, in LMO films, the energy gain associated with reducing the overall elastic energy of the nanocomposite should be less important than the thermodynamic tendency to form Mn²⁺ ions. It is worth noting that outcropping of secondary phase islands toward the surface of the film is governed by misfit relaxation mechanisms and the resulting topological distribution depends on film thickness and kinetic effects [59]. As a consequence, secondary phases well embedded in the matrix or outcropped toward the surface could just reflect different evolution stages of a similar segregation mechanism.

We have mentioned above that phase diagram of manganese oxides indicates a strong tendency of manganese to adopt an oxidation state of Mn²⁺ in the vacuum conditions (residual oxygen) used to grow the LMO films. The crystal radius of Mn²⁺ in octahedral coordination is 0.97 Å and, attending to the cation radius and the Goldschmidt tolerance factor for perovskites, it could be feasible for Mn to occupy the La site (La³⁺ radius being 1.172 Å) in the LMO perovskite up to the solubility ratio [14]. This mechanism has been reported for La-deficient thin films [61, 62] but, in stoichiometric films should lead to the formation of La-based oxides as La₂O₃ [14] not present in our films. Furthermore, TEM and Raman measurements suggest that LMO matrix preserves [La]/[Mn] ~ 1. Thus, in the present case, the main mechanism to accommodate Mn²⁺ ions is to form Mn₃O₄ segregates (i.e., Mn²⁺[Mn³⁺]₂O₄) and the creation of a La-rich phase as the observed La₂MnO₄ phase helps to compensate the stoichiometry unbalance. It is very likely that Mn₃O₄ particles are formed on LMO films during growth process. At that stage, cube-on-cube nucleation of the Mn₃O₄ spinel on the metrically cubic LMO is plausible. The crystallite Mn₃O₄ shape may now be seen as a reminiscence or truncated form of the typical octahedral crystal shape of spinels where {111} surfaces (forming an angle of ~35° with the vertical) have the lowest surface energy. It is not difficult now to argue that during cooling, when LMO Jahn-Teller transition takes place, Mn₃O₄ accommodates to the surrounding LMO twins by distorting its own structure. The two epitaxial relationships of hausmannite Mn₃O₄ with LMO at room temperature could be just the result of the loss of cubic symmetry, probably linked to distortions originated itself by the Mn³⁺ ion in the octahedral positions of Mn₃O₄. Thus, to fully understand this mechanism, it could be very interesting then to obtain structural information of these nanocomposites at high temperature, namely above T_JT.

Temperature dependence at 0.5 T (A) and magnetic field dependence at 10 K

The temperature and the magnetic field dependence of the in-plane magnetization were used to study the effects of the secondary phase on the magnetic properties (Figure 10). The temperature dependence of magnetization (in a magnetic field of 0.1 T) revealed two transitions, the first one is the expected antiferromagnetic transition of LMO at ~140 K and the second one is the ferrimagnetic transition of manganese oxide secondary phase at around ~40 K (Figure 10A). It is known that LMO thin films may exhibit a broad range of magnetic behaviors, from antiferromagnetic to strongly ferromagnetic, depending on the preparation conditions [27]. In our case, films are prepared in vacuum conditions leading to a residual moment below 0.3 μ_B/f.u. at 10 K which approaches the bulk value reported in the literature [63]. This residual magnetization is usually associated to a canted antiferromagnetic structure (magnetic moments pointing along b-axis but at a small angle out of ab-plane). The hysteresis loop at 10 K (Figure 10B) exhibits typical features observed in loops of thin films composed of soft magnet with small H_c and hard magnet with larger H_c [12]. However, pure antiferromagnetic LMO thin films shows almost closed hysteresis loop with a very small coercive field H_c that is usually associated to the existence of uncompensated spins at interfaces. On the other hand, manganese oxide with spinel structure is a hard ferrimagnetic with large values of μ₀H_c~0.35 T [64], whose presence in LMO matrix enlarges loop shape in our system comparing with pure undoped manganite compound. The presence of disordered phase boundaries between LMO and Mn₃O₄ and the larger vertical strain could additionally increase the pinning effects and coercivity in thin films [11].

Effects of secondary phase are revealed also by the local transport measurements shown in Figure 11. Characterization is performed by conductive AFM (cAFM) which senses the current flowing from a metallic probe to Pt electrodes deposited on top of the sample. The Mn₃O₄ secondary phase is highly insulating. However, we observe large spatial differences of the current distribution within the LMO matrix. The overall LMO primary phase in regions between inclusions shows some conduction of the order of few nA for an applied voltage of 0.5 V. Particularly, we observe that conduction is enhanced at the interface region of LMO in close proximity to the inclusions (up to a factor of 10 in these experimental conditions). Moreover, it seems that we observe that not all the four sides of each inclusion are equally conductive. This fact could be supported by TEM cross section images in Figures 6–9, showing that nanostructures exhibit a quite asymmetrical shape and thus, conduction could be differently affected at both sides of the inclusion. The driving force for the enhanced conductivity at these interfaces is far from being understood. Several mechanisms are known to increase electrical conductivity of LMO films as, for example, cationic vacancies or oxygen diffusion. Also, the effect on electrical conductivity of local strain fields is quite relevant in oxide thin films and it cannot be disregarded to explain the results in Figure 11. However, a deeper study of the transport mechanisms is beyond the scope of this manuscript.

In conclusion, we report the formation of self-organized Mn₃O₄ nanocolumns in Pulsed Laser Deposited LaMnO₃ films. We show that under vacuum growth conditions there is a strong tendency to form Mn²⁺ cations leading to the segregation of wedge-like Mn₃O₄ inclusions aligned with the main axis of the substrate. Formation mechanism at high temperatures (850°C) may be understood by considering that Mn₃O₄ nucleate directly on the nearly cubic LMO matrix. The final epitaxial relationship is determined by the subsequent Jahn-Teller transition that takes place during cooling creating LMO domains and twins and it may be defined as a distorted cubic spinel with c-axis slightly tilted from the direction perpendicular to the film plane. Partial accommodation of off-stoichiometry and residual strain may be achieved by the creation of an epitaxial La-rich Ruddlesden-Popper La₂MnO₄ phase close to the interface with the substrate. The magnetic and transport properties of the nanocomposite are strongly modified by the presence of the inclusions. In particular, an enhanced coercivity together with an increase of local conductivity close to the insulating inclusions has been observed. Our results demonstrate an easy single-step route for the preparation of nanocomposite thin films with tuned functional properties consisting of self-assembled ordered nanostructures embedded in a matrix that still preserves its structural characteristics.

AP, ZK, and JS leaded the research, data interpretation and manuscript preparation. NL, MS, and ZP were responsible for Raman results. NB, FS, and JS were responsible for microstructural analysis. LL, LB, AP, ZK, NM, CF, and BM contributed to the initial structural characterization and to the magnetic and c-AFM measurements. JR, AG, AP, and JS were responsible for sample preparation. AP was responsible for preparing the final version of the manuscript that includes the inputs and comments of all the authors.