Microscopic origins for stabilizing room-temperature ferromagnetism in ultrathin manganite layers

Abstract

La_0.7Sr_0.3MnO₃ is a conducting ferromagnet at room temperature. Combined with thin SrTiO₃ layers, the resulting heterostructures could be used as highly spin-polarized magnetic-tunnel-junction memories. However, when shrunk to dimensions below an apparent critical thickness, the structures become insulating and ferromagnetic ordering is suppressed. Interface spin and charge modulations are thought to create an interfacial dead layer, thus fundamentally limiting the use of this material in atomic-scale devices. The thickness of this dead layer, and whether it is intrinsic, is still controversial. Here we use atomic-resolution electron spectroscopy to demonstrate that the degradation of the magnetic and transport properties of La_0.7Sr_0.3MnO₃/SrTiO₃ multilayers correlates with atomic intermixing at the interfaces, and the presence of extended two-dimensional cation defects in the La_0.7Sr_0.3MnO₃ layers (in contrast to three-dimensional precipitates in thick films). When these extrinsic defects are eliminated, metallic ferromagnetism at room temperature can be stabilized in five-unit-cell-thick manganite layers in superlattices, placing the upper limit for any intrinsic dead layer at two unit cells per interface.

Colossal magnetoresistance, metal-insulator transitions, charge/orbital ordering, and half-metal ferromagnetism are only some of the intriguing phenomena that occur in manganites and have driven interest in the family of perovskite manganese oxides (ABO₃ perovskite structure, B site occupied by Mn) (1). With a Curie temperature (T_c) of ∼370 K, the half-metal La_0.7Sr_0.3MnO₃ (LSMO) has been considered a promising candidate for spintronics applications. However, although complete spin polarization in LSMO was inferred from photoemission measurements (2) and a record tunneling magnetoresistance (TMR) ratio of 1,800% was obtained at low temperature in tunnel junctions with manganite electrodes separated by a thin insulating layer of SrTiO₃ (STO) (3), the TMR decreased rapidly as the temperature increased and diminished while still far below T_c. One possible origin for the reduced TMR is the reduction of ferromagnetic ordering at the interface between the manganite and the STO. These interfacial effects have been suggested to be dominant as the LSMO layer thickness decreases, causing the magnetization and T_c to degrade and the resistivity to increase, ultimately resulting in films that are insulating at all temperatures when the layer thickness decreases below a reported critical value of 7–13 unit cells (4–11). This critical thickness for sustaining conductivity below T_c is attributed to an inherent “dead layer” at the interface between LSMO and STO (12–14). One origin for the reduced T_c and saturation magnetization of thin LSMO films compared to the bulk is believed to be a result of spin canting at the LSMO/STO interface (6, 15). Alternatively, magnetic phase separation into ferromagnetic metallic and less-ordered insulating clusters has also been proposed as the origin of the degraded properties of the LSMO when the film thickness decreases below a critical thickness (4).

These intrinsic mechanisms, however, are difficult to distinguish from extrinsic effects found in all experimentally realized heterostructures, such as crystalline defects. Recently, there has been tremendous excitement about the possibility to create a new family of heteroepitaxial devices based on perovskites (16), of which manganites are a key component. Although at some heterointerfaces, atomic precision can be achieved (17, 18), for others there may be fundamental limits for interface perfection (19, 20). Here, we use state-of-the-art atomic-scale growth and probe techniques to study the microscopic origins of these limitations in manganites. Atomic-resolution electron microscopy reveals that as the manganite layers in pulsed-layer-deposition (PLD) grown LSMO/STO multilayers become less stoichiometric (due to changes in the laser spot size), extended defects segregate near the interfaces to compensate for the A-site cation excess. The formation of these quasi-two-dimensional defects in ultrathin LSMO layers is in sharp contrast to three-dimensional precipitates that form in thick off-stoichiometric manganite films (21). The presence of such defects in the LSMO layers and of atomic-scale intermixing at the interfaces correlates with the degradation of the magnetic and transport properties of the multilayers. Whereas this effect is small in thick LSMO films (23), it is dramatic in thin films, resulting in insulating layers with strongly suppressed T_c. As the microscopic perfection of the structure is improved, however, the LSMO/STO multilayers can remain ferromagnetic and conducting below T_c at layer thicknesses below 2 nm, i.e., five unit cells, with the implication that the upper limit for the interfacial dead layer is now reduced to two unit cells or less. For these ultrathin manganite layers, we have now obtained a ferromagnetic transition temperature above room temperature in films that are also conducting.

The influence of the laser fluence on the properties of thick LSMO films grown by PLD has recently been demonstrated, showing an increase in the A-site/B-site cation ratio and a reduction of T_c as the fluence increases (23). The question then arises, to what degree these effects limit the physical properties of LSMO on the nanoscale. To explore the effect of the laser fluence on ultrathin LSMO films, two sets of (LSMO)₅/(STO)₅ multilayers were grown at oxygen partial pressures (P_O2) of 10^-3 and 10^-6 torr, varying the laser spot area but keeping the same pulse power (details in SI Text). Because the total energy is kept constant for the growth of all samples, the fluence at the target surface decreases with increasing laser spot area. The microstructure of the LSMO/STO multilayers was studied by high-angle annular dark field (HAADF) imaging in an FEI Tecnai F20-ST scanning transmission electron microscope (STEM) (24). More detailed chemical composition and bonding was probed on an atomic scale using spatially resolved electron energy loss spectroscopy (EELS) performed on an aberration-corrected Nion UltraSTEM 100 (25), that enables two-dimensional element and valence-sensitive imaging at atomic resolution (22).

Fig. 1 A–D shows cross-sectional HAADF-STEM images of the LSMO/STO multilayers grown at the same temperature and P_O2, where only the laser spot size has been varied. In HAADF imaging, the scattering intensity scales, to first approximation, with the atomic number Z as Z^1.7, so the brighter stripes in Fig. 1A correspond to the nominal five-unit-cell-thick LSMO layers and the darker ones to the STO layers (26–28). The strong dependence of the structural quality of these multilayers on the laser spot size is clearly visible. Starting with abrupt interfaces between the LSMO and the STO for the multilayers grown at larger spot sizes, the interfaces become more gradual and extended defects, some of which appear as dark lines in the HAADF images (Fig. 1 C and D), are introduced as the spot size decreases. The change in the interface abruptness becomes clear when looking at the HAADF images taken at higher magnification (insets in Fig. 1 A and D). For the multilayer grown at 10.2 × 10^-2 cm² (inset in Fig. 1A) the interfaces are atomically abrupt, whereas they are more diffuse as the spot size is reduced to 1.6 × 10^-2 cm² (inset in Fig. 1D). Contrast variations in the LSMO layer already suggest that the stoichiometry in the layer deviates from the nominal La_0.7Sr_0.3MnO₃, possibly due to cation intermixing or the presence of point defects.

Fig. 1.

Cross-sectional ADF images of (La_0.7Sr_0.3MnO₃)₅/(SrTiO₃)₅ superlattices grown at P_O2 = 1 m torr and at a laser spot size of (A) 10.2, (B) 7.5, (C) 5, and (D) 1.6 × 10^-2 cm². As the laser spot size decreases, extended defects are introduced in the LSMO layer and the interfaces become more diffuse.

Although HAADF lattice imaging, in which the contrast is dominated by the heavy A-site cations, can be very revealing, complementary information on the B-site cations can be obtained through spectroscopic imaging. With recent advances in electron microscopy, in particular the successful implementation of fifth-order aberration correction, a single core-loss EELS spectrum can now be recorded in milliseconds and a full two-dimensional spectroscopic image at atomic resolution in under a minute (22). Here, individual elemental maps were obtained by recording the Ti-L_2,3, Mn-L_2,3, and the La-M_4,5 edges simultaneously, and then subtracting the background and integrating over a part of the near-edge fine structure at each point in the image. The La elemental maps and red-green-blue false color B-site maps, obtained by combining the Ti (red channel) and Mn (green and blue channels) maps, of two multilayers grown at 7.5 and 1.6 × 10^-2 cm² are shown for comparison in Fig. 2. For both samples, the elemental maps and the corresponding concentration profiles (see Figs. S1 and S2) show clear differences between the upper and the lower manganite interfaces, confirming that the STO layers are predominantly TiO₂ terminated and the LSMO layers MnO₂ terminated. Additionally, we find that the multilayer grown with the smaller laser spot size shows less abrupt interfaces and an extended defect, marked by a white arrow in (Fig. 2D). The inferior quality of this sample can also be seen from the Ti distribution in the B-site map, where patches of weak Ti sublattice are found throughout the LSMO layers (Fig. 2C and Fig. S2A), which suggests a higher degree of Mn/Ti intermixing. In electron microscopy, it is often difficult to distinguish between true intermixing and broadening of an interface due to the shape of the electron probe (probe tails). However, from the two-dimensional B-site map shown here and the Ti map (see Fig. S2A), enhanced intermixing can be directly inferred due to the observed variation of the Ti concentration in the LSMO layers showing a weak Ti sublattice in some areas of the LSMO layers (probe tails would cause broadening of the interfaces, but a correlation of the Ti concentration maxima in the LSMO layers with the B-site sublattice is not expected). Note that sample drift (∼1.0 Å/ min) during the acquisition of a spectroscopic image causes the lattice to appear distorted.

Fig. 2.

Spectroscopic-imaging of La_0.7Sr_0.3MnO₃/SrTiO₃ multilayers grown at P_O2 = 1 m torr and at a laser spot size of (A and B) 7.5 and (C and D) 1.6 × 10^-2 cm². (A and C) La elemental maps and (B and D) red-green-blue false color B-site maps, obtained by combining the Ti (red) and Mn (green and blue) maps extracted from the spectrum images. The multilayer grown with a smaller laser spot size shows less abrupt interfaces and an extended defect, marked by a white arrow in D. The growth direction is from bottom to top.

Extended defects found in the multilayer grown at a laser spot size of 1.6 × 10^-2 cm² (Fig. 1D) were further studied using spectroscopic imaging. In the elemental maps shown in Fig. 3A, a defect is visible in the upper part of the LSMO layer. The combined Mn and Ti concentration is strongly reduced in the region outlined by the white box, suggesting B-site deficiency. However, the simultaneously recorded HAADF image does not show missing columns of atoms at the B sites, but rather a higher intensity compared to the B-site atom columns in STO or LSMO. These results suggest that the B sites are locally filled by Sr atoms (due to the absence of La in Fig. 3C), forming a rock-salt-type layer to compensate for A-site excess. High-magnification HAADF images also reveal regions in the film where single rock-salt layers are introduced (see Fig. S3). In agreement with previous results on thick LSMO films (23), the reduction of the laser spot size causes an A-site cation excess in the nominal La_0.7Sr_0.3MnO₃. In the multilayer, this excess is accommodated by the introduction of extended planar defects at or near the upper interface as observed in Fig. 3. The dependence of the STO stoichiometry on the laser fluence is much weaker, and in the LSMO/STO multilayers studied here, extended defects are only found in the LSMO layers, mostly in the upper part of the manganite layer.

Fig. 3.

Spectroscopic-imaging of an extended defect in a La_0.7Sr_0.3MnO₃/SrTiO₃ superlattice grown at P_O2 = 1 m torr and at a laser spot size of 1.6 × 10^-2 cm². (A) Ti, (B) Mn, and (C) La elemental maps extracted from a 90 × 45 pixel spectrum image, and (D) the simultaneously recorded ADF image. The white box indicates a B-site deficient region at the top of a LSMO layer.

A second defect, which has formed along the growth direction in the upper part of the LSMO layer, is shown in Fig. 4. Because of the character of the Sr EELS edges, it is more difficult to obtain a strontium elemental map at the same acquisition time as the Ti, Mn, and La maps; however, information about the position of the Sr atoms is contained in the simultaneously recorded annular dark-field (ADF) image. Here, the La map (Fig. 4C) shows missing La columns at the center of the defect, but from the ADF image these sites are clearly occupied by some other atomic species. Combined with the information from the Mn and the Ti elemental maps, we can conclude that Sr atoms fill the positions of the La sites, which again suggests local off-stoichiometry.

Fig. 4.

Spectroscopic-imaging of an extended defect in a La_0.7Sr_0.3MnO₃/SrTiO₃ multilayer grown at P_O2 = 1 m torr and at a laser spot size of 1.6 × 10^-2 cm². (A) Ti, (B) Mn, and (C) La elemental maps extracted from an 109 × 80 pixel spectrum image, and (D) the simultaneously recorded ADF image. The white open circles indicate the position of the La columns around the defect. The ADF image, which also tracks the position of the Sr atoms, shows clear atomic columns in the center of the defect, whereas the La concentration is low, suggesting that the defect was formed to accommodate for excess Sr in the LSMO layer.

The physical properties of an LSMO layer are sensitive to variations of the stoichiometry of the film, i.e., A-site/B-site ratio (23, 29). The formation of extended defects, rather than a uniform distribution of point defects implies large but local fluctuations in composition. In quasi-two-dimensional layers with a lower percolation threshold, these fluctuations could be expected to have a larger impact on long-range transport properties than in a three-dimensional structure. In order to understand the interplay of the observed structural differences and the properties of the multilayers, transport and magnetic measurements were performed. The temperature-dependent magnetization shown in Fig. 5A indicates that the magnetic properties of the multilayer vary strongly with the laser spot size. T_c as well as the saturation magnetization increases as the spot size increases from 1.6 to 10.2 × 10^-2 cm². Hence, as the superlattice structure and stoichiometry is optimized, reducing interdiffusion and the number of defects, the magnetization increases. At the same time, large changes in the temperature-dependent resistivity are observed (Fig. 5C). Multilayers grown at smaller spot sizes are insulating at all temperatures, however, as the spot size is increased to 10.2 × 10^-2 cm², the sample becomes conducting below T_c.

Fig. 5.

Temperature dependence of the (A and B) magnetization (zero field cooled, and measured warming under 1,000 Oe) and (C and D) resistivity of (La_0.7Sr_0.3MnO₃)₅/(SrTiO₃)₅ superlattices grown at various laser spot sizes as indicated in the figure. The multilayers were grown at an oxygen partial pressure of (A and C) 10^-3 torr and (B and D) 10^-6 torr. As the laser spot size increases, metallic ferromagnetism is stabilized.

As the oxygen partial pressure is decreased from 10^-3 to 10^-6 torr, similar trends with respect to the laser spot size persist. Fig. 5 B and D show the temperature-dependent magnetization and resistivity for the set of samples grown at 10^-6 torr. The saturation magnetization as well as T_c increase as the spot size increases and, accordingly, the resistivity decreases. The multilayer grown with the largest spot size shows the expected behavior for LSMO, being insulating at temperatures above T_c and metallic below T_c. The T_c of this multilayer is determined to be ∼298 K. The samples grown with smaller spot sizes, however, remain insulating at all temperatures. This result is remarkable, because it shows that metallic ferromagnetic LSMO can be stabilized, even for layer thicknesses of ∼2 nm, which is far below the previously reported critical thickness. Note that postannealing in oxygen for 10 h at up to 850 °C also had no effect on the resistivity and magnetization of the multilayers (see Fig. S4), which confirms that oxygen vacancies do not dominate the observed behavior and trends with cation stoichiometry. Ti-L_2,3 and Mn-L_2,3 EELS across the multilayers also confirms that within the experimental sensitivity of ∼5% there are no significant Ti or Mn valence changes (see Fig. S5).

In conclusion, LSMO/STO multilayers with manganite layer thicknesses of ∼2 nm can exhibit ferromagnetism with T_c above room temperature and remain metallic below T_c, if the structure is optimized by tuning not only the oxygen partial pressure and growth temperature but also the laser fluence at the target. For smaller spot sizes, hence larger fluences, the quality of the multilayer deteriorates, resulting in less abrupt interfaces, and an LSMO A-site/B-site cation ratio in excess of one. This off-stoichiometry is accommodated by the introduction of extended defects and leads to a reduction of the magnetization and an increase of the resistivity such that the multilayers remain insulating at all temperatures. The laser fluence corresponding to the smaller spot size is typical for previous PLD-grown LSMO/STO multilayers (9, 30), which might explain their inferior properties. Interdiffusion across the interface and off-stoichiometry in the LSMO layer such as that observed in Fig. 2 might also explain previous results (14) which showed that the magnetic properties of thin LSMO films can be improved by adding an additional two unit cells of LaMnO₃ between the LSMO and the STO. This work suggests that the intrinsic critical thickness has yet to be reached, even for five unit cell LSMO layers which are virtually all interface.

Varying the spot size is known to alter the cation stoichiometry in other oxide systems (23, 31). When the stoichiometry differences are accommodated by cation vacancies, cation interdiffusion across an interface becomes more likely and similar dead layers or a masking of true interfacial effects may occur.