Charge, spin and orbital tuning across SrTiO3jLaMnO3 transition metal oxide heterointerfaces

(1)

Charge, spin and orbital tuning across SrT iO

3

|LaM nO

3

transition metal oxide heterointerfaces

J. J. Beekman

July 8, 2016

Bachelor’s Thesis Physics & Astronomy, University of Amsterdam (UvA) Supervisors: Mark Golden, Shrawan Mishra and Anne de Visser Institute: Van der Waals-Zeeman Instituut (WZI)

Workload: February 1st_{, 2016 till July 8, 2016, 15EC}

(2)

Abstract page. The first abstract is the formal-scientific abstract, the second one is a popular-scientific abstract in Dutch. The second abstract is meant for 6 VWO (final year of dutch secondary school) students.

Abstract

In transition metal oxides the charge, spin and orbital properties are highly correlated. At the interface between transition metal oxides, various forms of symmetry breaking patterns of the spatial, time-reversal and gauge symmetry can be produced (Hwang et al. 2012). Due to this, interfaces can exhibit properties that are not observed in bulk crystals of either the sub-strate or the overlayer material. In order to make it possible to ’engineer’ solids with desirable properties, a better understanding of these properties has to be realised, which is the focus of this thesis. Recently, Wang et al. (2015) have reported an atomically sharp transition from an antiferromagnetic to a ferromagnetic ground state for the SrT iO3|LaM nO3(n) system for a

critical thickness of n = 5 LMO unit cells (uc). Though it remains unknown what the atomic origin of this emerging ferromagnetism is. Here, X-ray absorption spectroscopy measure-ments were performed at the I09 beamline of the Diamond Light Source, UK to investigate the atomic origin of the magnetism. X-ray magnetic circular dichroism measurements show ferromagnetism for samples of n ≥ 4 uc on both the manganese and the titanium sites. From 2 to 4 uc the x-ray absorption spectroscopy data show a sharp transition in the Mn valence state from mainly divalent to a mainly trivalent state. XLD measurements showed the Mn eg electrons in LMO of 2 ≤ n ≤ 10 uc samples to have a 3z2− r2 type orbital polarisation,

which increases with increasing LMO thickness. The eg states are suggested to be

antiferro-orbitally ordered in the plane of the LMO film. With these findings, we provide a number of experimental characteristics which an emerging theoretical description of the magnetism in these systems will need to describe. As such, the research presented here plays a role in the ongoing drive to explore the potential of devices and possible technologies harnessing the special properties of transition metal oxides.

(3)

Abstract

Als je in het periodieke systeem kijkt, zie je in de derde tot en met de twaalfde kolom een groep elementen staan. Dit worden de overgangsmetalen genoemd, omdat ze een deels gevulde d-schil hebben. In mijn bachelor project heb ik deelgenomen aan een onderzoeksprogramma dat de bijzondere eigenschappen van de oxides van deze overgangsmetalen in kaart brengt. Het onderzoeksteam be-stond uit post-doc Shrawan Mishra, PhD student Georgios Araizi Kanoutas en mij als bachelor onder leiding van professor Mark Golden. De onderzochte materi-alen komen in de vorm van perovskieten voor. De zuurstofatomen hebben in deze structuur een sterke wisselwerking met het overgangsmetaal en zorgen ervoor dat elektronen van de ene metaal plaats in de structuur naar een aanliggende plaatsen kunnen springen. Dit proces, dat afhankelijk is van meerdere karakteristieken van de structuur en de keuze van het overgangsmetaal, maakt het mogelijk dat mate-rialen die weinig van elkaar verschillen toch een heel breed spectrum van gedrag kunnen vertonen. Van isolerend tot metallisch, van magnetisch tot supergeleidend. Als twee zulke perovskieten op elkaar worden gestapeld is er een nieuwe speler in het veld, namelijk het grensvlak, en dit kan tot onverwacht, interessant en mogelijk nuttige fenomenen leiden. Het onderzoek tijdens mijn bachelor project was gericht op het systeem waarbij het substraat SrT iO3 met de overlaag LaM nO3 afgedekt

werd. In dit systeem zorgt het grensvlak er voor dat de ene stof opeens een mag-neet wordt, maar alleen als het een bepaalde minimum dikte heeft. Dit gedag voor dit systeem is door Wang et al. (2015) ontdekt. Voor mijn thesis onderzocht ik wat er in dit systeem op het niveau van de atomen precies gebeurt en wat hiervan de mogelijke oorzaak is. In het onderzoek wordt onze stof bestraald met fotonen met verschillende energieën, wat er voor zorgt dat onze stof elektronen uitzendt. Dit is bekend als het foto-elektrisch effect. De elektronen worden weer ‘bijgevuld’ en door de zo gecreëerde stroom bij verschillende energieën te meten krijgen we veel informatie over de stof. Deze techniek heet Röntgen absorptie spectroscopie. Voor het onderzoek kregen we 6 dagen de tijd in de Diamond Light Source, dat is de elektronen versneller in het Verenigd Koninkrijk. We deden 24 uur per dag onder-zoek. Met onze metingen hebben we ontdekt dat het magnetisme van het snijvlak veroorzaakt zou kunnen zijn door elektronen die van het LaM nO3 oppervlak naar

het snijdvlak reizen, en dat zowel de mangaan als de titanium atomen magnetisme vertonen. Door deze bevindingen weten we meer van oxide snijdvlakken die mag-netisch zijn. In de verre toekomst zouden wij zulke nieuwe materialen op basis van oxiden kunnen inzetten om een sneller computergeheugen of een nieuwe generatie van magneettreinen te maken.

(4)

1 Introduction

Atomically sharp transition metal oxide (TMO) interfaces can exhibit properties that are not observed in bulk crystals of either the substrate or the overlayer material. Although some interfaces can fundamentally not be this sharp (Naka-gawa et al. 2006), modern thin-film deposition techniques have made it possible to grow many epitaxial complex oxide thin-films with unit cell (uc) control on the thickness of the samples (Wang et al. 2015). In TMOs, the charge, spin and or-bital properties are highly correlated. At the interface between transition metal oxides, various forms of symmetry breaking patterns of the spatial, time-reversal and gauge symmetry can be produced and new charge, spin, and orbital recon-structions can emerge. By varying the thickness of the TMO-layers these effects can sometimes be controlled, making it possible to ’engineer’ solids with desirable properties (Hwang et al. 2012). TMOs have a perovskite crystal structure, ABO3,

like the example in figure 1 where A and B are cations. The crystal can be regard-ing as a stack of alternatregard-ing sub-unit cell layers of AO and BO2, meaning that

if a material ABO3 is terminated and a material A0B0O3 is grown on top, there

are two structural variants. One with the AO/BO2/A0O/B0O2 interface and one

with the BO2/AO/B0O2/A0O interface. Non bulk-like behaviour at a TMO

inter-face was first observed at the interinter-face between the two non magnetic insulating perovskite crystals LaAlO3(LAO) and SrT iO3(ST O), which has been the subject

of much research ever since (Ohtomo and Hwang 2004; Okamoto and Millis 2004; Nakagawa et al. 2006; Chen et al. 2011; Slooten et al. 2013; Zhong et al. 2010). At this interface, two configurations are possible: the p-type LaO/AlO2/SrO/T iO2,

which is found to be insulating and the n-type AlO2/LaO/T iO2/SrO where a 2D

metallic state with a charge carrier mobility of more than 10.000cm2V−1s−1 was first measured by Ohtomo and Hwang (2004). The n-type interface was found to have a sharp transition from an insulating to a conducting ground state for a criti-cal LAO layer thickness of 4 uc. Magnetism at a non-magnetic oxide interface was first observed by Brinkman et al. (2007) at this same n-type ST O|LM O interface. With the discovery of this high-mobility electron gas and magnetism at a TMO interface a new and interesting field of solid state physics has emerged.

The possibility to reach a different ground state by adding a single atomic layer could lead to exiting applications (Wang et al. 2015). Because of this it is highly relevant to thoroughly understand and control these kinds of effects.

(6)

Figure 1: The perovskite structure displayed for a LaM nO3 unit cell (Yu et al. 2015).

This thesis focuses on understanding the promising ST O|LaM nO3(LM O)

in-terface, which was recently reported to exhibit a step-like transition from antifer-romagnetic behaviour to a ferantifer-romagnetic ground state as a function of increasing LM O thickness by Wang et al. (2015). Bulk LMO is always an antiferromagnetic insulator. The manganese (Mn) atoms in LM O are normally trivalent and the lat-tice constant is about 3.9˚A. Wang et al. have shown that ST O|LM O(n) samples, where n is the LMO thickness in unit cells, make a magnetic transition for samples above a critical thickness n = 5uc and that the system has a Curie temperature of TCurie= 115K, see figure 2a. For a 9 uc sample, a total magnetic moment of

M = 20nAm2 was found, corresponding to a magnetic moment of m = 0.83µB

per perovskite uc for a sample of 25mm2_{. Superconducting quantum interference}

device (SQUID) measurements in scanning mode have shown a uniform distribu-tion of magnetic domains forming at low temperature (LT, T < TCurie) above

the critical LMO thickness, with domain sizes decreasing with increasing sample thickness, see figure 2b.

(7)

(a) The samples show a thickness dependent mag-netic transition the Curie temperature, TCurie=

115K.

(b) Magnetic domains forming above the critical thickness for the magnetic transition. The black scaling bars indicates 100µm.

Figure 2: Findings on the ST O|LM O system by Wang et al. (2015).

It is suggested by (Wang et al. 2015) that an electronic reconstruction at the ST O|LM O(n) interface causes the ferromagnetism to emerge, as explained in sec-tion 2.2. However as of today no spectroscopy study has been done to pinpoint the origin of this magnetism on an atomic scale or test this hypothesis. In this thesis x-ray absorption spectroscopy (XAS) techniques will be used to see if evidence can be found for or against an electronic reconstruction in the ST O|LM O system. In addition, element specific measurements of the magnetic properties using x-ray dichroic techniques will be presented.

2 Theory

The most relevant theoretical groundwork is mentioned in this section. For a more rigorous theoretical background I strongly recommend to consult the provided references.

2.1 The perovskite structure

The crystal structure of TMOs is the perovskite structure, ABO3, where the B

site indicates a transition metal (TM) ion, as shown in figure 1. Here the A type cation sits in the body centred position (1/2, 1/2, 1/2), the B type cations sit in the cube corner positions (0, 0, 0) and the oxide ions sit in the mid-edge positions (1/2, 0, 0), forming oxide octahedra centred at the B sites. The 3d energy level scheme of the TM is naturally split into a set of spin up and a spin down levels due to Hund’s rule splitting. The crystal field due to the Oh crystal field due

to the six O2− ions results in the splitting of the original five-fold degenerate 3d orbitals into a three-fold degenerate t2gorbital manifold and a two-fold degenerate

(8)

eg orbital set (Hwang et al. 2012). Spin, charge and orbital properties in TMOs

are due to the interactions between the TM ions and the oxide ion octahedra and thus the effective TM-TM interaction (Hwang et al. 2012). Small distortions in the oxygen octahedra can further lower the energy of the system. This is known as the Jahn Teller (JT) effect and would split the three-fold degenerate t2gorbital

in degenerate yz and zx orbitals and a xy orbital, and the two-fold degenerate eg orbital in a 3z2− r2 and a x2− y2 orbital (Mizokawa et al. 1999), given

oc-cupation of the t2gor egorbital group with an odd number of electrons, see figure 3.

Since the Mn atoms are trivalent in LMO, the 3d orbitals are occupied by four electrons, filling the spin-split lower t2g level completely and occupying the lower

eg level with only a single electron.

Figure 3: The splitting of the 3d orbitals of high spin trivalent manganese in a perovskite lattice.

2.2 Possible origin of novel interfacial ground states

Attempts have been made to construct a theoretical model of an electronic-reconstruction for oxide interfaces, where the three key factors are the interaction strengths, bandwidths and electron densities (Okamoto and Millis 2004). The ST O|LM O interface is a polar interface. Although LMO is charge neutral, it consists of al-ternating charged (La3+_O2−₎+ _{and (M n}3+_(O2−₎

2)−layers, whereas in ST O the

layers themselves are charge-neutral. Every uc of LM O will add to the potential, leading to a potential diverging to infinity, see figure 4a, this is called the polar catastrophe. By redistributing the electrons over the solid in such a way that the overlayer of the polar oxide is terminated - top and bottom - by a layer of half of the usual charge, the divergence in the potential can be avoided.

For the case of LAO on T iO2-terminated ST O, this would correspond to one

(9)

(.../LaO+/AlO2−/LaO

+1/2_{|T iO}

2/SrO/...). In this case, the extra 1/2e− per

per-ovskite uc at the LaO|T iO2 interface can occupy low-lying, empty Ti 3d-related

states of the ST O at the interface. Considering the situation after such a charge transfer process, the 1/2e−at the top-most T iO2 layer is overcompensated for by

the charge of the (LaO)+ _{layer above it. This process continues throughout the}

LAO film, such that the electric field in the LAO oscillates around zero, figure 4b, rather than that it increases as the LAO thickness increases, figure 4a. This whole process is referred to as electronic reconstruction, in analogy to the structural re-constructions that take place in conventional polar semiconductors such as GaAs, so as to circumvent what has been called the polar catastrophe.

Figure 4: (a) An electronic reconstruction at the ST O|LAO interface. Without an electronic reconstruction. The potential diverges. (b) With an an electronic reconstruction of half an electron charge transfer, the potential does not diverge. (c) An electronic reconstruction at the ST O|LM O interface. By transferring a half electron from the surface to the interface of the system, LMO effectively self-dopes (Wang et al. 2015).

In the case of LM O on ST O, the build-in potential in the LM O is suggested to lift the valence band at the top of the LM O film above the conduction band of the LMO layer close to the interface, such that electrons are transferred within the LMO itself, resulting in a kind of self-doping, figure 4c. Wang et al. (2015) attribute the observed ferromagnetism to these additional electrons in Mn states near the interface, attributing the ferromagnetic behaviour itself to the Mn atoms near the interface.

Other possible explanations for non bulk-like interfacial ground states include atomic relaxation (Zhong et al. 2010) and the strain on the lattice at the interface (Chen et al. 2011).

(10)

3 Experimental methods

The experimental methods used in this thesis are all spectroscopy techniques. For this research a beamtime at the Diamond Light Source (DLS) synchrotron facility was granted, where all data discussed in this thesis were measured.

3.1 X-ray absorption spectroscopy (XAS)

XAS is a widely used spectroscopic technique for probing the electronic structure of the unoccupied states of a crystal. In an XAS measurement the photon energy is scanned across an energy range containing core level resonances. At these char-acteristic resonant energies, the absorption of the x-ray photon causes an electron in a core level to become exited into a higher lying, empty state, leaving behind a core hole. The core hole gets filled by a higher lying electron through Auger decay, sending out an Auger electron and leaving two holes. Through this x-ray absorption process an electron depletion is created in the system. This depletion is neutralised by a drain current flowing from ground. By measuring this drain current, one has an indirect measure for the amount of x-rays absorbed (Kegel 2014).

Figure 5: The possible 2p to 3d electron transitions for high spin trivalent manganese are indicated by the black arrows. The red arrows indicate the spin up and spin down electrons occupying the filled states.

The energies and intensities of the XAS signal reflect the possible transitions in the sample at resonant energies. The L2,3-edge transitions for high-spin trivalent

manganese are indicated in figure 5. These transitions are determined by the electronic dipole selection rules. The electronic dipole selection rules follow from Fermi’s Golden Rule. For the transition metal L-edges, the final state of the XAS process, in which there is a 2p core hole and (often) holes in the valence

(11)

and conduction band d-states, is characterised by atomic multiplets, as there is a strong, attractive interaction between the core hole and the d-electrons. These atomic multiplet effects determine the line shape of the spectra and make XAS a highly sensitive technique to the local symmetry and the electronic configuration in the sample for these kinds of compounds. All XAS measurements presented in this thesis were carried out at the L2,3-edge resonance energy, probing the empty

3d states. The XAS signal differs for the use of different polarisations. In x-ray linear dichroism (XLD), the difference between the linear horizontal (LH) and the linear vertical (LV) XAS signal is measured, and in x-ray magnetic circular dichroism (XMCD), the difference between the circular left (σ+_{) and circular right}

(σ−) XAS signal is measured. One of the main benefits of XAS is that resonant energies are atom specific, and because of this, XAS allows one to pinpoint the origin of certain properties in solids, such as magnetism.

X-ray linear dichroism (XLD) In XLD measurements the LH polarised light is polarised in-plane, and the LV polarised light is polarised out-of-plane with respect to the sample, see figure 6. The XLD signal is a measure for departures from cubic spatial symmetry at the transition metal site, and of possible orbital ordering. We distinguish in-plane lattice strain, rotations and tilts in the oxygen octahedra (Liao et al. 2016; Kan et al. 2016), A-site shifts (Mizokawa et al. 1999) and Jahn-Teller (JT) distortions (Mizokawa et al. 1999) as well as orbital polarisation (Huang et al. 2004a) as possible causes of the XLD signal in perovskite type TMOs.

(a) In plane linear horizontal polarised light (b) Out of plane linear vertical polarised light

Figure 6: Schematic view of using linear horizontal and linear vertical polarised light to conduct an XLD experiment. The blue squares indicate the perovskite octahedra.

X-ray magnetic circular dichroism (XMCD) For electronic excitations following from the absorption of unpolarised or horizontally polarised light, the spin should remain unchanged over the initial and final state, ∆s = 0. For circular polarised light however, the dipole selection rules only allow for ∆s = 1 or −1 transitions for circular left (σ+) and circular right (σ−) polarised light. In order to have a difference between the σ+ _{and σ}−

measurements and see an XMCD signal, there has to be an inequality between the spin up and spin down density of state populations, see figure 7a for a highly simplified picture. This is an intrinsic property of a ferromagnetic state. Because the L2-edge and the L3-edge have

(12)

opposite spin-orbit coupling (l − s and l + s), the spin polarisation will be opposite at the two edges (St¨ohr 1999), creating an XMCD signal like the example in figure 7b.

Figure 7: (a) In a magnetic compound the spin up and spin down density of state (DOS) popula-tions of the electrons are not equal, giving rise to an XMCD signal. Note that this is a strongly simplified picture of a DOS. (b) An example XMCD signal of the L2,3-edge of a transition metal.

The negative peak corresponds to the L3 transitions and the positive peak corresponds to the L2

transitions. Obtained and modified from St¨ohr (1999).

3.2 Diamond Light Source (DLS)

The DLS is the national synchrotron facility of the UK, located at the Harwell Science and Innovation Campus in Oxfordshire. In this facility, electrons are ac-celerated in the synchrotron ring of 561.6m circumference to relativistic energies of E = 3GeV and move through so called helicical undulators. These force the elec-trons to move in an oscillating fashion and radiate photons. Through this process a photon beam with a much higher flux than other conventional x-ray sources is created. The I09 end-station used in this research allows the user to select photon energies in the range from 230eV to 2.0keV (soft x-rays) and 2.1keV to 18keV (hard x-rays). The beam is highly monochromatic due to the use of a plane grat-ing monochromator for soft x-rays and a Si(111) double-crystal monochromator for hard x-rays, see figure 8. For XAS measurements only the soft beam was used. Depending on the phase of the undulator magnetic fields, the beam is either lin-ear polarised or circular polarised. By using a liquid helium flow the sample on the manipulator can be cooled down to Tmin= 50K (http://www.diamond.ac.uk/

(13)

Figure 8: A technical drawing of the I09 beamline (http://www.diamond.ac.uk/ 2016).

3.3 Ultra high vacuum (UHV) suitcase

Some part of the samples measured were transported in ultra high vacuum (UHV, P ≤ 10−9mbar) to the DLS I09 beamline, in order to prevent structural or stoi-chiometric changes due to transport through air. The samples transported through air are referred to as ex-situ. The samples that were transported in the suitcase are referred to as in-situ. The suitcase is displayed in figure 9.

(14)

4 Data treatment procedures

All data treatment in this thesis has been done by self-written python scripts written using the Python(x,y) 2.7.5 scientific programming package, unless stated otherwise. All python data treatment files, the raw data and the treated data can be found at https://github.com/jj1993/oxide-interfaces-spectroscopy.

4.1 XAS background and normalisation

X-ray absorption data have to be normalised carefully in order to enable a quan-titative analysis, such as the generation of a dichroic signal. In the process of radiating the sample, incoming photons can lose some energy through scattering in the solid. This means incoming photons with an energy higher than the resonant core level absorption energy can lose energy and still excite the 2p to 3d electronic transition of interest. Due to this effect, a reasonably preditable background sig-nal is observed. In the relatively short energy range used for XAS measurements, the background is almost a constant slope outside of the main-edge photon energy range, and can be regarded as a series of smooth step functions initiating at the L2 and L3 edges within the main-edge photon energy range. The procedure used

in this thesis to subtract the background from the spectra is explained here.

The background subtraction process is described in more detail in Appendix A.1. The pre-edge, main-edge and post-edge photon energy regions are defined by hand. A linear fit to the pre-edge is extrapolated over the spectrum and subtracted (figure 20b). The intensity difference between the end of the pre-edge and the start of the post-edge is assumed to be solely due to a smooth, step-function located at the main edge. The steps are taken at the centre of the L2 and L3 edges for

the manganese and titanium L2,3-edge spectra (figure 20c). After subtracting the

step-function the pre- and post-edge are at the same level and a single, fourth-order polynomial is fitted to the pre- and post-edge, extrapolated over the main edge (figure 20d). After subtracting the polynomial from the spectrum (figure 20e) the step-function is added again, since this step-function is part of the main-edge behaviour (figure 20f).

4.2 Data smoothing

After normalisation the data were smoothed. Smoothing is a procedure where the data are replaced by a close fit, in an attempt to remove the noise and to stress the main characteristics of a measurement. A spline was fitted to the data, which is a piecewise polynomial, as described by Dierckx (1999). This method avoids Runge’s phenomenon, where the use of high degree polynomials causes oscillations in the fit. An example of the degree of smoothing is displayed in figure 10.

(15)

(a) An XLD spectrum with no smoothing (b) An XLD spectrum with smoothing

Figure 10: An example of the underlying XAS traces of the degree of smoothing of the data. The Ti L2,3-edge XLD measurement for the in-situ ST O|LM O(4) sample is shown.

4.3 XMCD data treatment

For XMCD measurements the samples first have to be magnetised. In order to magnetise the samples, an in-situ magnetic field had to be applied. For this pur-pose, a small linear translator holding a pair of permanent magnets was designed by Michael Zapft from the W¨urzburg University, see figure 11. To magnetise the sam-ples, the samples were first heated to room temperature, reaching a paramagnetic state. Then the magnet was placed as in figure 11, exerting an in plane magnetic field on the sample. While keeping the sample in the magnetic field, the sample was cooled to low temperature again, making the transition to an ferromagnetic state, if it was a magnetic sample. When the permanent magnets were removed with the linear translator, the sample reaches the remanence magnetic point, this magnetisation will be denoted in this thesis as H. We distinguish a positive in-plane remanence magnetisation, denoted as H+, and an remanence magnetisation reversed in-plane with respect to H+, denoted as H−. For the determination of the spin and orbital magnetic moments (msand mo) the XMCD spectrum is

mea-sured. The area of the L3 XMCD, A, and that of the L2 XMCD, B, as shown in

figure 7b are integrated. By applying the dichroism sum rules for manganese, the spin moment ms and the orbital moment mo can be determined using equation 1

and equation 2 respectively, where C is the square of the 2p to 3d radial transition matrix element and has a value of about 10M beV (St¨ohr 1999). According to measurements done by Barriocanal et al. (2010) on ST O|LM O superstructures, a small magnetic contribution from the titanium sites can also be expected.

mspin= 2B − A C (1) morbital= − 2(A + B) 3C (2)

(16)

in the ST O|LM O system is expected to originate from the Mn 3d orbitals. The XMCD results will be validated by measuring the XMCD signal for two different orientations of the in-plane magnetic field. When magnetising the sample in a reversed magnetic field alignment, the XMCD signal should also reverse. If the signal does not reverse, this could indicate that the measured intensity differences between σ+ and σ− light are not due to ferromagnetism, and probably due to an artifact in the drain current measurements or the background subtraction and normalisation procedure. Applying the sum rule on the XMCD signal should give a positive total magnetic moment ms+ mo, since we expect a ferromagnetic

state. Comparing the calculated ms and mo values to literature is an important

validation of the XMCD results. Wang et al. (2015) have measured the bulk magnetic properties using a Quantum Design vibrating sample magnetometer and a Quantum Design magnetic property measurement system. A total saturated magnetic moment of M = 20nAm2was found for a 9 uc sample of 5mm by 5mm, this corresponds to a magnetic moment m = 0.83µB per perovskite uc.

Figure 11: Camera picture of experimental setup used to magnetise the samples in-situ.

4.4 Mn valence states from Mn-L

2,3

XAS spectra

The valence states of the Mn atoms in the LMO films were analysed to determine whether an electronic reconstruction could be taking place. In the case of such a reconstruction the average valence state of the Mn in the LMO would remain trivalent, with equal parts of tetravalent atoms at the surface of the sample and divalent atoms at the interface. Analysis of the valence states was done by compar-ison of measured XAS L2,3-edge spectra published in peer-reviewed literature from

materials with formally pure valence states to the measurements, see figure 12. To compare to the effectively isotropic published spectra, the measured in plane linear horizontal (LH) and out of plane linear vertical (LV) polarised XAS spectra of our LMO films were added as 2 ∗ LH + LV . Since the XAS signal scales with the number of possible transitions, the backgrounds (including the step function) of the model spectra were removed and their total areas were normalised to the num-ber of holes in the 3d shell. The divalent M nO is thus normalised to 5 units, the trivalent LaM nO3, as well as all measurements, are normalised to 6 units and the

tetravalent SrM nO3 is normalised to 7 units. This way the fitting of the measured

(17)

less to the area of the spectra. A depth-first algorithm (https://en.wikipedia.org/ 2016) is used to produce the spectra for all different proportional mixing possibil-ities of the model spectra shown in figure 12, accurate within steps of 5%. For the purpose of the discussion in this thesis the best fitting mix of valences will be assumed to reflect a realistic estimate of the true mix of valence states at an accuracy of 5%.

Figure 12: Isotropic manganese L2,3-edge XAS spectra for divalent M nO, trivalent LaM nO3and

tetravalent SrM nO3 as measured. Data obtained from Burnus et al. (2008).

4.5 XLD data treatment

Huang et al. (2004b) reported simulations arguing that the basic multiplet struc-ture of the LMO Mn L2,3-edge XLD signal strongly reflects the 3d orbital ordering

(OO) of the system and is insensitive to the JT distortion. According to their simulations for La0.5Sr1.5M nO4 the XLD signal is strongly characterised by OO,

wearas JT and other symmetry lowering distortions hardly contribute to the main features of the XLD signal. This means that information about the OO in LMO for different samples can be deduced by comparison to this simulation (figure 13).

(18)

Figure 13: The XLD spectra for different OO of the Mn 3d eg orbitals calculated for

La0.5Sr1.5M nO4 (Huang et al. 2004a). The upper trace is for an eg orbital polarisation of an

x2− y2_{-type and the lower for an 3z}2_{− r}2_{-type. In each case the two orbitals shown are occupied}

in an in-plane antiferro-manner for one Mn site to the next.

5 Results and discussion

5.1 Experimental

Epitaxial LMO samples were grown on (100)-oriented crystalline niobium-doped STO using pulsed laser decomposition (PLD). The samples were grown by P. Reith and J. Geessinck from the Hilgenkamp and Koster groups at the MESA+ Institute for Nanotechnology at the University of Twente, the Netherlands. All XAS mea-surements were conducted at the I09 beamline of Diamond Light Source (DLS) in Harwell, UK.

Linear horizontal, linear vertical, circular left and circular right soft x-ray ab-sorption spectra were measured at the Mn L2,3-edge (thin film), the titanium L2,3

-edge (substrate) and at the oxygen K--edge across the critical thickness, n = 5, and across the transition temperature, TCurie= 115K. Unless otherwise stated,

the measurements were performed at low temperature (LT), T = 60 ± 5K. In the following, the magnetic state, valence state and OO of manganese is discussed on the basis of XMCD, XAS and XLD data respectively. Finally magnetism in the titanium atoms is discussed on the basis of XMCD data.

All Mn-L2,3 and Ti-L2,3XMCD measurements discussed here were conducted

on ex-situ samples. All Mn-L2,3-XLD measurements discussed here were conducted

on in-situ samples. The Mn valence states were determined from Mn-L2,3 XAS

(19)

(a) The n = 10 uc sample. The XMCD signal re-verses with reversing the magnetisation direction of the sample.

(b) The n = 7 uc sample. Above the critical tem-perature (RT) there is hardly any XMCD signal (see dashed black line) and below the critical tem-perature (LT) the signal is both much stronger and reverses on reversing the magnetic field.

(c) The n = 4 uc sample.

Figure 14: The Mn Ledge XMCD spectra for 3 different ST O|LM O(n) samples.

5.2 Manganese XMCD

The XMCD spectrum of ST O|LM O(10uc) is shown in figure 14a. Because the 10uc XMCD spectrum reverses almost perfectly under reversing the sample magnetisa-tion, extraction of the magnetic moments using the sum-rules might be trustworthy. This calculation gives a spin magnetic moment ms= 0.570µB±32% and an orbital

magnetic moment mo = 0.108µB± 1.5% with a ratio of ms/mo= 5.27 ± 32% for

an estimated intensity error of 2%. Both the moand msvalues measured here are

positive, confirming a ferromagnetic state. The the total magnetic moment per Mn site m = 0.68 ± 0.18µB is quite close to the total measured saturated magnetic

moment per perovskite uc m = 0.83µB by Wang et al. (2015). Note that we are

comparing the Mn magnetic moment extracted from an Mn XMCD signal with the total saturation magnetic moment per uc averaged over the thin film. The Ti sites are expected to have a magnetic contribution, as well, according to

(20)

Barrio-canal et al. (2010). This was the first XMCD experiment ever conducted at the I09 beamline, the moment is immortalised in figure 15.

In figure 14b the XMCD spectrum for the manganese L2,3-edge of ST O|LM O(7uc)

is displayed. The room-temperature (RT) measurements show an XMCD signal that is hardly distinguishable from the noise in the spectrum, and is much less intense than the low temperature (LT) signal. This indicates that the sample is not in a ferromagnetic state at RT in keeping with the Curie temperature of 115K for these systems. This ’zero result’ shows the experimental set-up and data treat-ment procedures to be robust and artifact free. The LT measuretreat-ments, on the other hand, show a clear XMCD spectrum, which reverses sign with reversing the sample magnetisation. We do note that some of the fine structure of the XMCD spectrum of the 7 uc sample under reversed magnetisation is not simply inverted.

The Mn L2,3-edge XMCD signal for ST O|LM O(4uc) is displayed in figure 14c

and shows a clearly different line shape than the 7 and 10 uc samples. The XMCD signal is still quite strong however, indicating a ferromagnetic state for this 4 uc sample.

The slightly different characterisations of the inverted XMCD signal under re-versing the magnetic field for the manganese L2,3-edge of the 7 uc sample might be

explained by the existence of magnetic domains in the LMO samples as reported by Wang et al. (2015), which are visible in the scanning SQUID data reported in figure 2b. The magnetic field is reversed by taking the samples out of the main measurement chamber and into a preparation chamber where the sample is physically magnetised. When returning the sample to the main chamber of the end-station, there is a slight uncertainty in the position of the beam spot with respect to the sample. The footprint of the beam is approximately 20µm wide (http://www.diamond.ac.uk/ 2016), and has a tail parallel to the beam due to a grazing incidence angle used during the XAS measurements. When comparing the beam spot size to figure 2b it strikes one that the beam spot is about the same size as the domains of the 7 uc LMO sample, but would contain a number of domains for a 12 uc LMO sample. Therefore it seems possible that a combination of the larger domain size for the thinner films and a small beam offset after returning to the measurement position in the analysis chamber could lead to a difference in domain sampling, which will have a much greater effect on the (magnetically sensitive) XMCD measurements for a 7 uc sample than for a 10 uc sample.

Having considered these points, it remains to be said that the significant XMCD signal observed for the 4 uc sample shows it to be ferromagnetic, in contrast to the findings of Wang et al. (2015), in which the critical thickness for ferromagnetism is found to be 5 uc. A complete set of all Mn L2,3-edge XMCD data (all ex-situ)

(21)

Figure 15: The first ever XMCD measurement at the DLS I09 beamline (on the screen). From left to right: M. Zapf (W¨urzburg University), S. Mishra (UvA), J. J. Beekman (UvA), C. Schlueter (I09 beamline) and G. A. Kanoutas (UvA).

(a) A 60% divalent and a 40% trivalent contri-bution from the model curves are shown. The ST O|LM O(2uc) sample is also shown.

(b) The sum of 60% divalent and a 40% trivalent manganese from the model curves is shown. The ST O|LM O(2uc) sample is also shown.

Figure 16: The isotropic manganese L2,3-edge XAS spectrum for ST O|LM O(2uc)

5.3 Manganese valence states

The Mn valence states of the samples were determined with the procedure as de-scribed in section 4. This procedure was carried out for the 2 uc in-situ, 2 uc ex-situ, 4 uc in-situ, 7 uc in-situ and 10 uc ex-situ samples. The formally diva-lent (M nO), trivadiva-lent (LaM nO3) and tetravalent (SrM nO3) manganese L2,3-edge

XAS spectra are displayed in figure 12 on page 17. The isotropic XAS spectrum for in-situ ST O|LM O(2uc) and the best fit to this data using a mix of the

(22)

refer-ence spectra are shown in figure 16 for a valrefer-ence state mix of 60% divalent, 40% trivalent and 0% tetravalent Mn.

The analogous fits to the data from the other samples are displayed in Appendix A.5. A summarising overview of the valence states for the different in- and ex-situ LMO samples is displayed in figure 17. On average, the manganese atoms in the thicker LMO films are trivalent, and the 2 uc samples are more divalent. For the thicker, ferromagnetic samples the divalent and tetravalent contribution are equal within error, thus this would not contradict the expectations of the electronic reconstruction scenario described earlier. The finding that the 2 uc samples are not purely trivalent shows that the simple internal self-doping picture is insufficient, however, as until the ferromagnetic thickness is reached, the Mn valence state should remain three.

(a) The calculated valence contribution for the ex-situ samples.

(b) The calculated valence contribution for the in-situ samples.

(c) The averaged valence state for all samples.

Figure 17: The manganese valence state mixture for increasing LMO sample thicknesses. The valence states were determined by comparison with literature spectra of nominally pure valence states. The line in panel (c) is solely intended as a guide for the eye.

(23)

5.4 Manganese XLD

The Mn L2,3-edge XLD signals for the 2 uc, 4 uc and 7 uc samples (all in-situ) are

displayed in figure 18. Comparison to figure 13 shows that the orbital polarisation of the eg states is of the 3z2− r2 type, rather than the x2− y2 type. Following

Huang et al. (2004a) this would suggest that the eg states are antiferro-orbitally

ordered in the plane of the LMO film, with 3x2_{− r}2_{and 3y}2_{− r}2 _orbitals

alternat-ing from one site to the next. The XLD signal gets stronger with increasalternat-ing LMO thickness, indicating an increasing degree of orbital polarisation. A complete set of all Mn L2,3-edge XLD measurements is displayed in Appendix A.6.

(a) The in-situ 2 uc sample. (b) The in-situ 4 uc sample

(c) The in-situ 7 uc sample

(24)

5.5 Titanium XMCD

Finally, we turn to the XMCD experiments conducted at the Ti L2,3 edges. The

data for the 4 uc sample show a tiny XMCD signal (figure 19a). For the 7 uc sample, a very small, yet clear ferromagnetic signal is observed (figure 19b). An ST O|LAO(10)|LM O(10) sample was also measured, and showed no sign of ferro-magnetic Ti moments (figure 19c) indicating that the Ti ferromagnetism emerges solely due to the ST O|LM O interface. A complete set of all Ti L2,3-edge XMCD

measurements is displayed in Appendix A.3.

The Ti L2,3-edge XLD measurements are displayed in Appendix A.3. O K-edge

measurements are displayed in Appendix A.4. The physical interpretation of these measurements is not discussed here, as this is beyond the scope of this thesis.

(a) The in-situ ST O|LM O(4uc). (b) The in-situ ST O|LM O(7uc)

(c) The in situ ST O|LAO(10uc)|LM O(10uc). Note, in this case the XMCD signal is not mul-tiplied by a factor of 4.

(25)

6 Conclusions

We have confirmed a ferromagnetic ground state for ST O|LM O(n) samples for n > 5. Interestingly, the 4 uc sample which, on the basis of the behaviour of previous grown and characterised samples, was reported to not be a ferromagnet, was found to be ferromagnetic, although its XMCD line shape has characteristics differing from those of the thicker samples. A sum rule analysis of the XMCD signal for the 10 uc sample yielded a Mn magnetic moment of m = 0.68 ± 0.18µB

per Mn atom. A small contribution to the total magnetic moment from the Ti sites for this 10 uc sample is also expected, as the Ti sites for the 4 uc and 7 uc samples were qualitatively found to have a non-zero magnetic moment, as shown by XMCD measurements. The XMCD signal of the Ti atoms was found to disap-pear when LMO was grown on a LaAlO3 buffer layer of 10 uc, identifying the Ti

magnetism as a ST O|LM O interfacial property. The Mn eg electrons in LMO of

2 ≤ n ≤ 10 uc samples were found to have a 3z2− r2

type orbital polarisation, which increases with increasing LMO thickness. The eg states are suggested to be

antiferro-orbitally ordered in the plane of the LMO film.

A clear correlation between changes in the Mn valence in the LMO films and the transition from non-ferromagnetic to ferromagnetic behaviour was found. The ferromagnetic, thicker samples (n ≥ 4) showed an average valence of three, and their XAS data were consistent with equal percentages of divalent and tetravalent manganese, which would appear to fit the self-doping scenario suggested from DFT calculations in Wang et al. (2015). However, the clear deviation of the 2 uc samples from trivalency, and the equally clear proof for ferromagnetism in the XMCD data for the in-situ 4 uc film are both signals that the true origin of the ferromagnetism in thin film LM O is more complex than that presented by Wang et al. (2015). In this sense, the experimental data recorded, analysed and discussed in this bache-lors thesis are playing a key role in the ongoing development of our understanding of how thin film thickness can be used to control the magnetic ground state at the interface between TMOs.

More measurements on samples of ST O|LM O(n) for n ≤ 4 are planned for a beamtime at DLS, UK in June and should reveal more insights on the sample behaviour for thinner samples.

(26)

7 Acknowledgements

I would like to thank prof. Anne de Visser for acting as second assessor of this thesis and George Araizi Kanoutas and Stephan Bron for regularly helping me. Jaap Geessink and Pim Reith from the group of Gertjan Koster from the MESA+ institute from the University of Twente have provided us with high quality samples. Tien-Lin Lee, Pardeep Kumar Thakur, Christoph Schlueter, David Duncan and Dave McCue from the DLS I09 beamline have helped us a lot in making the most of the DLS beamtime. Michael Zapf and Judith Gabel from the group of Ralph Claessen from the W¨urzburg University also taught me a lot during the measurements at DLS. Finally and most prominently this thesis would not have been realised without the kind support of dr. Shrawan Mishra and prof. Mark Golden from the Van der Waals-Zeeman Institute (WZI).

References

Barriocanal, J. G., Cezar, J. C., Bruno, F. Y., Thakur, P., Brookes, N. B., Utfeld, C., Calzada, A. R., Giblin, S. R., Taylor, J. W., Duffy, J. A., Dugdale, S. B., Nakamura, T., Kodama, K., Leon, C., Okamoto, S., and Santamaria, J. (2010). Spin and orbital Ti magnetism at LaM nO3/SrT iO3 interfaces. Nat. Comms.,

82:2010.

Brinkman, A., Huijben, M., van Zalk, M., Huijben, J., Zeitler, U., Maan, J., van der Wiel, W., Rijnders, G., Blank, D., and Hilgenkamp, H. (2007). Magnetic effects at the interface between nonmagnetic oxides. Nature Materials, 6:493 – 496.

Burnus, T., Hu, Z., Hsieh, H. H., Joly, V. L. J., Joy, P. A., Haverkort, M. W., Wu, H., Tanaka, A., Lin, H.-J., Chen, C. T., and Tjeng, L. H. (2008). Local electronic structure and magnetic properties of LaM n0.5Co0.5O3 studied by

x-ray absorption and magnetic circular dichroism spectroscopy. Physical Review, 77:125124.

Chen, Y., Pryds, N., Kleibeuker, J. E., Koster, G., Sun, J., Stamate, E., Shen, B., Rijnders, G., and Linderoth, S. (2011). Metallic and insulating interfaces of amorphous SrT iO3-based oxide heterostructures. Nano Letters, 11:3774–3784.

Dierckx, P. (1999). Curve and surface fitting with splines, Monographs on Numer-ical Analysis. Oxford University Press.

https://en.wikipedia.org/ (2016). Depth-first search. https://en.wikipedia.org/ wiki/Depth-first_search, Retrieved on: 2016-07-07.

http://www.diamond.ac.uk/ (2016). I09 beamline at dls. http://www.diamond. ac.uk/Beamlines/Surfaces-and-Interfaces/I09, Retrieved on: 2016-16-06.

Huang, D. J., Wu, W. B., Guo, G. Y., Lin, H. K., Hou, T. Y., Chang, C. F., Chen, C. T., Fujimori, A., Kimura, T., Huang, H. B., Tanaka, A., and Jo, T. (2004a). Orbital ordering in La0.5Sr1.5M nO4 studies by soft x-ray lineair

(27)

Huang, H. B., Shishidou, T., and Jo, T. (2004b). Strong lineair dichroism in Mn L2,3 absorption predicted for orbital ordering in LaM nO3. Journal of the

Physical Society of Japan, 92:2399.

Hwang, H. Y., Iwasa, Y., Kawasaki, M., Keimer, B., Nagaosa, N., and Tokura, Y. (2012). Emergent phenomena at oxide interfaces. Nature Materials, 11:103–113.

Kan, D., Aso, R., Sato, R., Haruta, M., Kurata, H., and Shimakawa, Y. (2016). Tuning magnetic anisotropy by interfacially engineering the oxygen coordination environment in a transition metal oxide. Nature Materials, 15:432–438.

Kegel, F. (2014). Orbital reconstruction at oxide heterointerfaces: A comparative analysis from x-ray absorption data. Master’s thesis, Universiteit van Amster-dam.

Liao, Z., Huijben, M., Zhong, Z., Gauquelin, N., Macke, S., Green, R. J., van Aert, S., Verbeeck, J., van Tendeloo, G., Held, K., Sawatzky, G. A., Koster, G., and Rijnders, G. (2016). Controlled lateral anisotropy in correlated manganite heterostructures by interface-engineered oxygen octahedral coupling. Nature Materials, 15:425–431.

Mizokawa, T., Khomskii, D. I., and Sawatzky, G. A. (1999). Interplay between orbital ordering and lattice distortion in LaM nO3, Y V O3and Y T iO3. Physical

Review B, 60:7309.

Nakagawa, N., Hwang, H. Y., and Muller, D. A. (2006). Why some interfaces cannot be sharp. Nature Materials, 5:204–209.

Ohtomo, A. and Hwang, H. Y. (2004). A high-mobility electron gas at the LaAlO3/SrT iO3 heterointerface. Nature, 427:423–426.

Okamoto, S. and Millis, A. J. (2004). Electronic reconstruction at an interface between a mott insulator and a band insulator. Nature, 428:630–633.

Slooten, E., Zhong, Z., Molegraaf, H. J. A., Eerkes, P. D., de Jong, S., Massee, F., van Heumen, E., Kruize, M. K., Wenderich, S., Kleibeuker, J. E., Gorgoi, M., Hilgenkamp, H., Brinkman, A., Huijben, M., Rijders, G., Blank, D. H. A., Koster, G., Kelly, P. J., and Golden, M. S. (2013). Hard x-ray photoemission and density functional theory study of the internal electric field in SrT iO3/LaAlO3

oxide heterostructures. Physical Review B, 87:085128.

St¨ohr, J. (1999). Exploring the microscopic origin of magnetic anisotropies with x-ray magnetic circular dichroism (XMCD) spectroscopy. Journal of Magnetism and Magnetic Materials, 200:470–497.

Wang, X. R., Li, C. J., L¨u, W. M., Paudel, T. R., Leusink, D. P., Hoek, M., Poccia, N., Vailionis, A., Venkatesan, T., Coey, J. M. D., Tsymbal, E. Y., Ar-iando, and Hilgenkamp, H. (2015). Imaging and control of ferromagnetism in LaM nO3/SrT iO3 heterostructures. Science, 349:716–719.

(28)

Yu, Z., Tetard, L., Zhai, L., and Thomas, J. (2015). Supercapacitor electrode materials: nanostructures from 0 to 3 dimensions. Energy Environ. Sci., 8:702– 730.

Zhong, Z., Xu, P. X., and Kelly, P. J. (2010). Polarity-induces oxygen vacancies at LaAlO3/SrT iO3 interfaces. Physical Review B, 82:165127.

(29)

(30)

A

Appendices

A.1 Background normalisation procedure

(a) The raw data (b) A linear fit to the pre-edge is subtracted

(c) The intensity difference between the start and end point of the main-edge is interpreted as a dou-ble step function and subtracted

(d) A single fourth order polynomial is fitted to the pre- and post-edges, extrapolated over the main edge.

(e) The polynomial is subtracted from the spec-trum to get a straight background signal

(f) The step-function is added to the spectrum again. The resulting main-edge is shown

Figure 20: The XAS background subtraction procedure in steps. Background subtraction for the manganese L2,3-edge for ST O|LM O(10uc) is shown. This spectrum had the most complicated

(31)

A.2 Mn L

2,3

-edge XMCD for ST O|LM O(n) samples

(a) The 4 uc sample. an XMCD signal is observed

(b) The 7 uc sample. The XMCD signal reverses under reversing the magnetisation. The signal dis-appears at room temperature.

(c) The 10 uc sample. The XMCD signal reverses under reversing the magnetisation.

(32)

(33)

A.3 Ti L

2,3

-edge XMCD for ST O|LM O(n) and ST O|LAO(10)|LM O(10)

samples

(a) The 4 uc sample. A faint XMCD signal is observed.

(b) The 7 uc sample. An XMCD signal is ob-served. The XMCD signal reverses under revers-ing the magnetisation. The signal disappears at room temperature.

(c) The 10 uc sample with a 10 uc LAO buffer layer. No XMCD signal is observed. The XMCD signal is not multiplied.

(34)

A.4 O K-edge XMCD for ST O|LM O(n) samples

(a) The 4 uc sample. An XMCD signal is ob-served.

(b) The 7 uc sample. An XMCD signal is ob-served.

(35)

A.5 Mn valence states for ST O|LM O(n) samples

(a) XAS spectra for 2uc ex-situ sample and 45% Mn 2+, 50% Mn 3+ and 5% Mn 4+ mixture.

(b) XAS spectra for 10uc ex-situ sample and 20% Mn 2+, 50% Mn 3+ and 30% Mn 4+ mixture.

(c) XAS spectra for 2uc in-situ sample and 60% Mn 2+, 40% Mn 3+ and 0% Mn 4+ mixture.

(d) XAS spectra for 4uc in-situ sample and 20% Mn 2+, 50% Mn 3+ and 30% Mn 4+ mixture.

(e) XAS spectra for 7uc in-situ sample and 15% Mn 2+, 55% Mn 3+ and 30% Mn 4+ mixture.

Figure 24: The manganese L2,3-edge XAS spectra for ST O|LM O(n) samples and their best fitting

(36)

(37)

A.6 Mn L

2,3

-edge XLD for ST O|LM O(n) samples

(a) The 2 uc ex-situ sample. _{(b) The 2 uc in-situ sample.}

(c) The 4 uc in-situ sample.

(d) The 7 uc in-situ sample.

(38)

A.7 Ti L

2,3

-edge and O K-edge XLD for ST O|LM O(n)

samples

(a) The Ti L2,3-edge of the 2 uc sample. (b) The Ti L2,3-edge of the 4 uc sample.

(c) The Ti L2,3-edge of the 7 uc sample. (d) The O K-edge of the 7 uc sample.

Figure 26: The titanium L2,3-edge and oxygen K-edge XLD spectra for in-situ ST O|LM O(n)

Charge, spin and orbital tuning across SrTiO3jLaMnO3 transition metal oxide heterointerfaces