Constructing oxide interfaces and heterostructures by atomic layer-by-layer laser molecular beam epitaxy

Advancements in nanoscale engineering of oxide interfaces and heterostructures have led to discoveries of emergent phenomena and new artificial materials. Combining the strengths of reactive molecular-beam epitaxy and pulsed-laser deposition, we show here, with examples of Sr1+xTi1-xO3+delta, Ruddlesden-Popper phase Lan+1NinO3n+1 (n = 4), and LaAl1+yO3(1+0.5y)/SrTiO3 interfaces, that atomic layer-by-layer laser molecular-beam epitaxy (ALL-Laser MBE) significantly advances the state of the art in constructing oxide materials with atomic layer precision and control over stoichiometry. With ALL-Laser MBE we have produced conducting LaAlO3/SrTiO3 interfaces at high oxygen pressures that show no evidence of oxygen vacancies, a capability not accessible by existing techniques. The carrier density of the interfacial two-dimensional electron gas thus obtained agrees quantitatively with the electronic reconstruction mechanism.

density of the interfacial two-dimensional electron gas thus obtained agrees quantitatively with the electronic reconstruction mechanism.

Main Text:
Technological advances in atomic-layer control during oxide film growth have enabled the discoveries of new phenomena and new functional materials, such as the two-dimensional (2D) electron gas at the LaAlO 3 /SrTiO 3 interface (1, 2) and asymmetric three-component ferroelectric superlattices (3,4). Reactive molecular-beam epitaxy (MBE) and pulsed-laser deposition (PLD) are the two most successful growth techniques for epitaxial heterostructures of complex oxides. PLD possesses experimental simplicity, low cost, and versatility in the materials to be deposited (5). Reactive MBE employing alternately-shuttered elemental sources (atomic layer-by-layer MBE, or ALL-MBE) can control the cation stoichiometry precisely, thus producing oxide thin films of exceptional quality (6)(7)(8). There are, however, limitations in both techniques. Reactive MBE can use only source elements whose vapor pressure is sufficiently high, excluding a large fraction of 4d and 5d metals. In addition, ozone is needed to create a highly oxidizing environment while maintaining low-pressure MBE conditions, which increases the system complexity. On the other hand, conventional PLD using a compound target often results in cation off-stoichiometry in the films (9,10). In this paper we present an approach that combines the strengths of reactive MBE and PLD: atomic layer-by-layer laser MBE (ALL-Laser MBE) using separate oxide targets. Ablating alternately the targets of constituent oxides, for example SrO and TiO 2 , a SrTiO 3 film can be grown one atomic layer at a time. Stoichiometry for both the cations and oxygen in the oxide films can be controlled. Although the idea of depositing atomic layers by PLD has been explored since the early days of laser MBE (11,12), we show that levels of stoichiometry control and crystalline perfection rivaling those of reactive MBE can be achieved by ALL-Laser MBE. The technique is effective for both non-polar (such as SrTiO 3 ) and polar materials, such as the Ruddlesden-Popper (RP) phase La n+1 Ni n O 3n+1 with n = 4. By growing LaAlO 3 films on SrTiO 3 substrates at an oxygen pressure of 37 mTorr, sufficiently high to alleviate oxygen deficiency in SrTiO 3 , we show that the properties of the 2D electron gas at the LaAlO 3 /SrTiO 3 interface are in quantitative agreement with the electronic reconstruction mechanism.
The principle of ALL-Laser MBE is schematically illustrated in Fig. 1A. The key difference between ALL-Laser MBE and conventional PLD or laser MBE is the use of separate oxide targetsinstead of using a compound target of SrTiO 3 , targets of SrO and TiO 2 are switched back and forth as they are alternately ablated by a UV laser beam. In conventional PLD or laser MBE using a compound target, all elements are ablated at once and the film grows unit cell by unit cell. In ALL-Laser MBE using separate targets, on the other hand, the film is constructed one atomic layer at a time. The number of laser pulses on each target for one atomic layer is around 100, allowing a stoichiometry control of about 1%.
It has been a common practice to control the layer-by-layer growth of thin film by recording and analyzing in real time its reflection high-energy electron diffraction (RHEED) pattern (13,14). The intensity of the specularly-reflected RHEED spot is commonly used, which oscillates depending on the step edge density of the film. One oscillation period corresponds to the deposition of one unit cell layer in the unit cell-by-unit cell growth (15). Haeni et al.
have found that the intensity of the diffracted spot can be used to control the growth of each atomic layer of SrTiO 3 films in reactive MBE with alternately shuttered growth (7). In this work, we also use the diffracted spot intensity oscillation and our results confirm that the phenomenology identified by Haeni et al. also applies to ALL-Laser MBE. Figure 1B shows the RHEED intensity oscillations as the targets of SrO and TiO 2 are alternately ablated.
Starting from a TiO 2 -terminated SrTiO 3 substrate surface, the diffracted spot intensity increases to a maximum when one monolayer of SrO is deposited; it then decreases to a minimum when one monolayer of TiO 2 is subsequently deposited. Furthermore, we have found that the specular spot also oscillates with the same period as the diffraction spot, albeit 180° out of phase, if the Kikuchi lines caused by the diffused scattering of electrons do not overlap the specular spot (see Supplementary Materials).
The RHEED intensity depends on both the surface step edge density and the surface chemistry. When all elements of the film are delivered at the same time in the unit cell-byunit cell growth, the chemistry information is averaged out and only the step edge density of the film is reflected in the RHEED intensity. For ALL-MBE or ALL-Laser MBE, the surface chemistry changes when different atomic layers are deposited sequentially; consequently both the step edge density and chemistry information can be observed. A detail discussion can be found in the Supplementary Materials.
In our experiment, the RHEED diffracted intensity oscillations along the SrTiO 3 [110] azimuth were used to calibrate and control the film growth. As shown in Fig. 1B, Sr/Ti > 1 leads to an increasing peak intensity and the appearance of a "double" peak, while Sr/Ti < 1 leads to a reduced peak intensity; Sr/Ti = 1 results in oscillation peaks with a constant intensity and shape. Furthermore, insufficient or excess pulses in each cycle cause beating of the RHEED intensity (Fig. 1C) while the intensity remains constant for 100% layer coverage ( Fig. 1D). Using the RHEED intensity oscillation combined with the calibration of laser pulses per atomic layer obtained from the film thickness measurement, the cation stoichiometry in the films can be controlled to within ±1%.  Figure 2B shows x-ray diffraction (XRD) θ-2θ scans for the films around the SrTiO 3 002 diffraction peak along with that of the SrTiO 3 substrate. When the film is stoichiometric, the XRD spectrum cannot be distinguished from that of the single crystal SrTiO 3 substrate. When the film is not stoichiometric, regardless of Sr rich or deficient, a diffraction peak from the film at a smaller angle than the substrate peak is seen, indicating a c-axis lattice expansion. The c lattice constant vs.
x is plotted for the films in Fig. 2C. Also plotted are data from films grown by reactive MBE for comparison.
The results from the two techniques are in agreement with each other. In    Since the discovery of the 2D electron gas at the LaAlO 3 /SrTiO 3 interface (2), several competing mechanisms for its origin have been proposed and intensely debated, including electronic reconstruction (19), oxygen vacancies in the SrTiO 3 substrate (20,21,22), and intermixing between the LaAlO 3 film and the SrTiO 3 substrate (23,24). According to the electronic reconstruction mechanism, because the atomic layers are charge neutral in SrTiO 3 but charged in LaAlO 3 , a diverging electric potential is built up in a LaAlO 3 film grown on a TiO 2 -terminated SrTiO 3 substrate. This leads to the transfer of half of an electron from the LaAlO 3 film surface to SrTiO 3 when the LaAlO 3 layer is thicker than 4 unit cells, creating a 2D electron gas at the interface with a sheet carrier density of 3.3×10 14 /cm 2 when LaAlO 3 is sufficiently thick. A serious inconsistency with this mechanism is that the carrier densities reported experimentally are invariably lower than the expected value (25,26) except under conditions where reduction of SrTiO 3 substrate is suspected (20,21). Oxygen vacancies in SrTiO 3 are known to contribute to conductivity, but all reported conducting LaAlO 3 /SrTiO 3 interfaces have been grown at low oxygen pressures (< 10 mTorr), and annealing in oxygen is often required (1,19,27,28); higher oxygen pressures during the PLD growth result in insulating samples (27) or 3D island growth (29). Low growth pressures can not only cause oxygen vacancies in SrTiO 3 , but can also enhance the bombardment effect due to energetic species that may lead to La-Sr intermixing at the interface (27). At present, there is no consensus on the origin of the 2D electron gas at the LaAlO 3 /SrTiO 3 interface.
With ALL-Laser MBE, we grew LaAlO 3 film one atomic layer at a time, which allowed us to produce conducting LaAlO 3 /SrTiO 3 interfaces at an oxygen pressure as high as 37 mTorr.
This high oxygen pressure helps to prevent oxygen reduction in SrTiO 3 , ensure that the LaAlO 3 films are sufficiently oxygenated, and suppress the La-Sr intermixing due to the bombardment effect. Furthermore, we grew LaAlO 3 films of different cation stoichiometry, LaAl l+y O 3(1+0.5y) , as a way to test the electronic reconstruction hypothesis. As depicted by Sato Because of the high oxygen pressure during the LaAlO 3 growth, the samples were well oxygenated. This was proven by polarization-dependent x-ray absorption spectroscopy (XAS) measurements. Figure 4D shows XAS spectra with different linear polarizations for a stoichiometric LaAlO 3 film and Fig. 4E shows the Ti L 2,3 x-ray linear dichroism (XLD) signals for different LaAl 1+y O 3(1+0.5y) stoichiometry. No Ti 3+ related features around 462 eV, characteristic of the oxygen deficient LaAlO 3 /SrTiO 3 samples (32,33), are observed. Rather, the spectra are similar to the fully oxygenated samples (32,33). Figure 4F shows the Ti L 2,3 x-ray magnetic circular dichroism (XMCD) signals obtained from opposite circularly polarized XAS spectra. Very small XMCD signals were observed, indicating very weak ferromagnetism. This again is consistent with the fully oxygen annealed LaAlO 3 /SrTiO 3 samples (32, 33). films. All of the films are conducting with sheet resistance around 10 4 Ω/□ at 300 K, in contrast to insulating films grown by PLD from LaAlO 3 compound targets at this oxygen pressure (27,34). Note that only the Al-rich LaAl 1.08 O 3.12 film shows metallic behavior in the full temperature range, while all other films show low-temperature resistivity upturns, consistent with the previous stoichiometry dependence reports (35,36). The low temperature upturn has a -lnT dependence characteristic of the Kondo effect (37). This may be attributed to the inevitable defects at the LaAlO 3 /SrTiO 3 interface, consistent with the weak magnetism shown by Fig. 4F. The black dashed line in Fig. 5D represents the threshold normal-state sheet resistance h/4e 2 , or 6.5 kΩ/□ for a superconductor-insulator transition (38). Only the Alrich samples have normal-state sheet resistance below the dashed line; these might thus exhibit superconductivity. The sheet carrier density is around 10 14 /cm 2 for all the samples, close to the expected value of 1.7×10 14 /cm 2 . The sample with a higher sheet carrier density shows a lower mobility, in agreement with previous reports (26). Previous PLD studies on the stoichiometry dependence varied the laser energy density and oxygen pressure to change the LaAlO 3 composition (28,36). In our study the films were grown under the same conditions, therefore, the stoichiometry dependence observed is free from the effects of varying deposition conditions.

Sample preparation and characterization
The (001) SrTiO 3 substrates used in this work were treated following the receipe in (39) to produce atomically flat TiO 2 -terminated surface with one unit-cell-high steps (Fig. S1A). The (001) LaAlO 3 substrates used in this work were treated following the receipe in (40) to produce atomically flat AlO 2 -terminated surface with one unit-cell-high steps (Fig. S1B). The surface morphology of a 60 nm SrTiO 3 (Fig. S1C) grown on treated SrTiO 3 substrates also shows atomically flat surface with root mean square (RMS) roughness around 0.1 nm, comparable to that of the SrTiO 3 substrate. laser pulses for each AlO 2 layer. The offstoichiometric LaAl 1+y O 3(1+0.5y) films were created by supplying 100y % more Al in each layer. After the deposition, the films were cooled to room temperature in oxygen at the same pressure as during the growth. The x-ray diffraction and reflectivity measurements were performed using a Bruker D8 Discover system with Cu K  radiation (λ = 1.5406 Å). The Leptos fitting software (Bruker AXS Inc.) was used to determine the out-of-plane lattice constant of the films from the θ-2θ scans (the substrate peak was used as the reference) and to determine the film thickness from the x-ray reflectivity (XRR) measurement.
Ultraviolet Raman spectroscopy measurements were performed in a backscattering geometry normal to the film surface using a Jobin Yvon T64000 triple spectrometer equipped with a liquid nitrogen cooled multichannel charge coupled device detector. Ultraviolet light (325 nm line of the He-Cd laser) was used for excitation. Maximum laser power density was ~0.5 W/mm2 at the sample surface, low enough to avoid any noticeable local heating of the sample. Spectra were recorded at 10 K using a variable temperature closed-cycle helium cryostat.
X-ray absorption spectroscopy (XAS), x-ray linear dichroism (XLD) and x-ray magnetic circular dichroism (XMCD) measurements were carried out at the at the elliptically polarized undulator beamline 4.0.2 of the Advanced Light Source, using the Vector Magnet end station and with an energy resolution of approximately 0.1 eV (41). Samples were cryogenically cooled to 13 K. Average probing depth in the total electron yield XAS detection mode was estimated to be approximately 5 nm, providing interface-sensitive information with minimal contribution from surface adsorbates. Measurements were carried out in near-grazing (30°) incidence geometry, enabling selective alignment of the x-ray electric field parallel to the abplane of the film for vertically-polarized light (E║ab), and almost parallel to the c-axis of the film for vertically polarized light (E║c). The XMCD spectra were obtained utilizing circularly polarized x-rays with and by alternating the direction of the applied magnetic field of 0.3 T between parallel and antiparallel directions with respect to the x-ray helicity vector.
RHEED intensity oscillation patterns of the specular spot, the diffraction spots, and the

Kikuchi line intersection
For atomic layer-by-layer film growth by reactive MBE of ALL-laser MBE, the intensity of the RHEED diffraction spot along the SrTiO 3 [110] azimuth rather than the specular spot is monitored and used to control the growth. The specular spot intensity often shows "double peak" behavior and is less predictable. This effect has been recognized and articulated by Dobson et al. as due to the superposition of the elastic specular scattered and the diffused scattering such as Kikuchi bands (42). Figures S2A shows the RHEED pattern after the deposition of a TiO 2 layer. One can clearly see that beside the specular spot and the diffraction spot, there is a bright spot due to the intersection of Kikuchi lines. Figure S2B shows the RHEED intensity oscillations corresponding to different areas of integration as marked by the various rectangles as the targets of SrO and TiO 2 were being alternately ablated. The intensity of the Kikuchi intersection spot oscillates in-phase with the diffraction spot with the same period. The specular spot also oscillates with the same period as the diffraction spot, but 180° out of phase. When the Kikuchi spot and the specular spot are integrated together, a complex intensity oscillation pattern appears. In principle, the specular spot intensity oscillation can be used for the control of atomic layer-by-layer growth; in practice, however, it often overlaps with the Kikuchi spot, making its use difficult. Growth calibration using split RHEED intensity peaks Strontium, with a larger atomic number, has a larger scattering power, or atomic form factor, than Ti (43). Therefore, in general, the deposition of the SrO layer leads to an increase in the RHEED intensity and the growth of the TiO 2 layer causes the RHEED intensity to decrease.
Surface roughness adds complexity to this simple picture and the combined effects of roughness and chemistry give rise to the experimentally observed variety of RHEED intensity oscillation patterns. As shown in Fig. 1B in the main text, an oversupply of Sr causes the RHEED intensity peak to split. By adding one half extra SrO layer on top of stoichiometric SrTiO 3 , split RHEED intensity peaks as those in Fig. S3A can be created. Four films are grown with this double peak calibration and the growth for each film is interrupted at different time as shown in Fig. S3B. Starting at time t 1 , a film with half monolayer (ML) of SrO on top of a completed TiO 2 ML is presented in Fig. S3C. The surface roughness R q of this film is 0.13 nm, slightly larger than that of a TiO 2 terminated substrate (0.11 nm), indicating an incomplete coverage. RHEED intensity for this film is as indicated at t 1 in SrO and TiO 2 (44) may take place during this period of time as well, indicated by a local minima between t 4 and t 5 . From t 5 to t 6 , a ML of TiO 2 is completed. The segregation continues and the surface structure becomes to 0.5 ML SrO on top of 1 ML TiO 2 , which is the same as t 1 . Thus, RHEED intensity for t 1 and t 6 are identical. A complete ML of SrO and TiO 2 are deposited during t1 to t6, and RHEED oscillation finishes a whole period. As long as a stoichiometric supply of SrO and TiO 2 is maintained, the split RHEED intensity pattern will remain. This effect is utilized to more accurately calibrate the numbers of laser pulses for each SrO or TiO 2 layer. Deposition rate calibration by thickness measurement Besides the beating in RHEED intensity oscillation shown in Fig. 1C, the thickness measurement of calibration films by XRR was also used to calibrate the SrO and TiO 2 deposition rates for complete monolayer coverages. Because XRR oscillations are usually not 25 observable for a stoichiometric SrTiO 3 film on SrTiO 3 substrate (45), LaAlO 3 substrate was used for the calibration films. Figure S4 shows the XRR scan and fitting curve for a SrTiO 3 film built with 154 pairs of SrO and TiO 2 layers. The thickness of the film from the XRR measurement is 62.7 nm as compared to the thickness of 154 unit cells fully relaxed SrTiO 3 60.1 nm. Assuming that the SrTiO 3 film on LaAlO 3 substrate is fully strained and considering the Poisson's ratio  (46) for SrTiO 3 , the measured thickness agrees with the designed thickness within 2% (47).