Key concepts in spin tunneling : amorphous ferromagnets for spintronics Paluskar, P.V.

Transcription

1 Key concepts in spin tunneling : amorphous ferromagnets for spintronics Paluskar, P.V. DOI: /IR Published: 01/01/2008 Document Version Publisher s PDF, also known as Version of Record (includes final page, issue and volume numbers) Please check the document version of this publication: A submitted manuscript is the author's version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website. The final author version and the galley proof are versions of the publication after peer review. The final published version features the final layout of the paper including the volume, issue and page numbers. Link to publication General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. Users may download and print one copy of any publication from the public portal for the purpose of private study or research. You may not further distribute the material or use it for any profit-making activity or commercial gain You may freely distribute the URL identifying the publication in the public portal? Take down policy If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. Download date: 23. Jul. 2018

2

3 Key Concepts in Spin Tunneling Amorphous Ferromagnets for Spintronics Paresh Vijay Paluskar

4 De promotiecommissie bestaat uit: prof.dr. B. Koopmans 1 e promotor, Techn. Universiteit Eindhoven prof.dr.ir. H.J.M. Swagten 2 e promotor, Techn. Universiteit Eindhoven Dr.rer.nat. J.T. Kohlhepp copromotor, Techn. Universiteit Eindhoven prof.dr. R. Coehoorn lid kerncommissie, Techn. Universiteit Eindhoven en Philips Research Laboratories dr. C.F.J. Flipse lid kerncommissie, Techn. Universiteit Eindhoven prof.dr. R.A. de Groot lid kerncommissie, Radboud Universiteit Nijmegen Dr. J.S. Moodera lid kerncommissie, Massachusetts Inst. of Tech. The work described in this thesis has been carried out in the group Physics of Nanostructures, at the Department of Applied Physics, Eindhoven University of Technology, the Netherlands. This research was supported by NanoNed, a national nanotechnology program coordinated by the Dutch Ministry of Economic Affairs. Flagship NanoSpintronics. Project number 6474/7152-1B1. The cover shows the k-resolved density of states of fcc Co at the Fermi level. Artists impression by P.V. Paluskar and data from G.A. de Wijs, J.J. Attema and R.A. de Groot (Radboud Universiteit Nijmegen).

5 Key Concepts in Spin Tunneling Amorphous Ferromagnets for Spintronics Proefschrift ter verkrijging van de graad van doctor aan de Technische Universiteit Eindhoven op gezag van de Rector Magnificus, prof.dr.ir. C.J. van Duijn, voor een commissie aangewezen door het College voor Promoties in het openbaar te verdedigen op dinsdag 1 juli 2008 om uur door Paresh Vijay Paluskar geboren te Pandharpur, India

6 Dit proefschrift is goedgekeurd door de promotoren: prof.dr. B. Koopmans en prof.dr.ir. H.J.M. Swagten Copromotor: Dr.rer.nat. J.T. Kohlhepp CIP- DATA LIBRARY TECHNISCHE UNIVERSITEIT EINDHOVEN Paluskar, Paresh Vijay Key Concepts in Spin Tunneling : Amorphous Ferromagnets for Spintronics by / Paresh Vijay Paluskar. - Eindhoven : Technische Universiteit Eindhoven, Proefschrift. ISBN: NUR 926 Trefwoorden: spinpolarisatie/supergeleiding/tunneljuncties/amorfe ferromagneten Subject Headings: spin polarization/superconductivity/tunnel junctions/amorphous ferromagnets Printed By: Universiteitsdrukkerij Technische Universiteit Eindhoven.

7 Dedicated to my family, my parents Prabha and Vijay, my wife Sonu, and my brother Parag

8 vi

9 Contents 1 Introduction to spin tunneling Spintronics in daily life Basic aspects Electronic structure of 3d TM FMs Electron and spin tunneling Contemporary notions on spin tunneling AlO x : Relevant experiments MgO: Relevant experiments Relevance of amorphous ferromagnets This thesis Bibliography Probing electronic, magnetic and structural properties Sample fabrication Substrate and substrate cleaning considerations Deposition: Sputtering Plasma oxidation Structural characterization X-ray diffraction (XRD) X-ray absorption fine structure (XAFS) High-resolution transmission electron microscopy (HRTEM) In-situ analysis of chemical and electronic properties X-ray photoelectron spectroscopy (XPS) Ultraviolet photoelectron spectroscopy (UPS) Magnetic characterization Superconducting quantum interference device (SQUID) Magneto-optical Kerr effect (MOKE) Magnetic circular dichroism (XMCD) in x-ray absorption (XAS) Measuring electronic transport Superconduction tunneling spectroscopy (STS) vii

10 viii CONTENTS Inelastic electron tunneling spectroscopy (IETS) Magnetoresistance (MR) Current in-plane tunneling (CIPT) Bibliography Magnetic properties of CoFeB Background Sample preparation Properties of Co 72 Fe 20 B Crystallization of Co 72 Fe 20 B Effect of Co 72 Fe 20 B 8 crystallization on film resistance Effect of Co 72 Fe 20 B 8 crystallization on magnetic properties Properties of Co 80-x Fe x B Crystallization of Co 80-x Fe x B 20 from XRD Effect of Co 60 Fe 20 B 20 crystallization on magnetic properties Summary Bibliography Key concepts in spin tunneling Introduction Background Objectives of this work Experimental Results Sample preparation and measurement Impact of CoFeB crystallization of its TSP Verification of crystallization at interface Comparison of calculated and measured a-cofeb Calculation: Molecular dynamics Measurements: molecular dynamics vs. EXAFS Electronic structure and TSP Fe in strongly ferromagnetic state Comparison with measured TSP Interface bonding effects Changes in electronic structure on crystallization Highly spin-polarized boron sp states Conclusions Bibliography Impact of interface crystallization on inelastic tunneling Introduction

11 CONTENTS ix Background: Interface scattering Background: Inelastic electron tunneling spectroscopy (IETS) This work Experimental Methods Sample preparation and measurement Verification of crystallization at interface Experimental Results IETS spectra: Phonon modes IETS spectra: Magnon modes Size quantization of magnon modes Zero bias anomaly Summary Bibliography Correlation between magnetism and TSP Background This work and the relevance to understanding CoFeB Sample preparation and measurement Introduction to the S P behavior Basic aspects from computational magnetism S P behavior of CoFeB TSP of CoFeB shows the S P behavior Changes in valance band structure - UPS data XAS and XMCD Orbital moment (m o ) Spin moment (m s ) and exchange splitting ( ex ) Correlation between the s and the d-bands Discussion on CoFe Conclusions A Appendix A.1 Difference between Fe and Fe 80 B 20 - XAS A.2 Band-Filling and orbital moment A.3 Orbital moment A.4 Ratio of Orbital to Spin Moment A.5 Co edge XAS and XMCD Bibliography Thermal stability of MTJs Introduction

12 x CONTENTS Background This work Experimental Results Confirmation of Mn diffusion in a MTJ Does Mn diffuse? Influence of Mn diffusion on the TSP Impact of annealing on TSP Summary Bibliography Summary 145 List of publications 148 About the author 150 Acknowledgements 151

13 Chapter 1 Introduction to spin tunneling Ferromagnets and magnetic tunnel junctions in spintronics Abstract: In this chapter 1 we will introduce some relevant aspects of the electronic structure of ferromagnets, and how spintronic devices like MTJs employ this electronic structure for device operation. Then we will introduce a few experiments from which we derive our existing notions about the physics of spin tunneling. No attempt will be made to be complete or exhaustive in this section. Instead, the reader is referred to suitable reviews which embark on such an exhaustive overview. Subsequently, we will talk about a novel ferromagnetic material CoFeB which has the potential for advancing the application of spintronic devices. In the last part of the chapter, we will outline this thesis. 1 A large part of this chapter will appear as a review in the Encyclopedia of Materials Science and Technology authored by H.J.M. Swagten and P.V.Paluskar [62]. 1

14 2 Chapter 1 Introduction to spin tunneling 1.1 Spintronics in daily life The fact that electrons have a spins, i.e., an intrinsic magnetic moment, plays an important role in our everyday life. Technologies that use these electron-spins are not completely unknown in the daily life of a common man. One example is the magnetic strip on credit cards, another is a magnetic compass which navigates automobiles. A recent discovery which uses these electron-spins in electronic devices has spurred another wave of technology in the realm of data storage and sensing. It has, in essence, revolutionized the way we carry personal digital information, and therefore, not surprisingly, has been awarded the Nobel prize for Physics in Indeed, here one refers to the giant magnetoresistance effect (GMR) and the field of research that engenders from it - spintronics / magnetoelectronics. The most popular product that uses this technology is a computer hard-disk where information is read using a GMR sensor. Due to the use of these sensors, the density of information that can be stored on a hard-disk has increased substantially, allowing the emergence of products like the i-pod. This thesis is placed in this field, where the physical effects and devices based on electron-spins are explored. Let us have a look at the essentials of spin-transport in such devices. A sketch of GMR device is shown in Figure 1.1(a). Here, two ferromagnetic layers (for example, Co or Fe) are separated by a non-magnetic layer (for example, Cu or Cr). Assume that, using an external magnetic field, the magnetization of these two layers can be aligned parallel to each other [see Figure 1.1(a)] or antiparallel to each other [see Figure 1.1(b)]. When a current flows through this trilayer, the electrons which have their spins aligned with the magnetization of the layer experience less scattering events. On the other hand, the electrons with spins pointing opposite to the layer magnetization experience more scattering events. Therefore, in a parallel configuration, there are always electrons of one spin type that can easily travel through the trilayer. In Figure 1.1(a), this would be the case with the spin pointing right, which we call a spin-up or majority electron. We call the other electron, with spin pointing left, the spin-down or minority electron. Coming to the antiparallel configuration shown in Figure 1.1(b), one notices that although the spin-up electron manages to reach the top layer, its magnetization is aligned opposite to the local magnetization. Therefore, this electron too experiences more scattering events. Now, more scattering events implies that the electrons feel a higher resistance while traversing the trilayer. Since in the antiparallel configuration, both spin-up and spin-down electrons experience more scattering events, the resistance of the trilayer in this configuration is high. On the contrary, the parallel configuration allows easy transport of spin-up electrons, and the device resistance is comparatively low. This change in resistance which depends on the relative alignment of the magnetization of the two ferromagnetic layers is called magnetoresistance, and is defined as

15 1.2 Basic aspects 3 MR = R AP R P R P 100% (1.1) where P and AP denote parallel and antiparallel configuration. The resistance of such a device is shown in Figure 1.1(c) where MR is plotted as a function of the applied external field. At very high positive or negative fields (in this case, ±30 ka/m), the layers are aligned parallel, the resistance is low and the MR is zero. However, these layers are so engineered that in external fields of 7 10 ka/m, the antiparallel configuration is achieved, and the observed resistance and MR is high. In , two groups (that of Albert Fert and Peter Grünberg) reported the observation of a such a magnetoresistance effect [1, 2]. The change in resistance observed in [Fe/Cr] n multilayers was almost 50%, which led to the name giant magnetoresistance effect. Since then, significant progress has been made in enhancing the observed effect, as well as in understanding the origin of the effect. The reader may refer to extended reviews on this topic [3, 4]. Such spin-dependent electronic transport was subsequently observed in another type of device called a magnetic tunnel junction (MTJ). In this case, the nonmagnetic spacer layer of Figure 1.1(a-b) is replaced by a thin insulator ( 25 Å thick). Given the fact that quantum mechanics allows electrons to tunnel through such a thin insulator, one may imagine that electronic transport from one ferromagnetic layer to the other across such a tunnel barrier would also lead to a magnetoresistance effect. In 1995, Moodera et al. [5] and Miyazaki et al. [6] reported such a magnetoresistance effect which is appropriately called tunneling magnetoresistance (TMR). Considering that a decade later TMR effects above 200% have been reported, the application potential of such devices has not gone unrecognized. In fact, many technological devices which envision the use of this effect have been proposed, and some are already commercially available. This thesis investigates the properties and fundamental aspects of electron, and consequently, spin tunneling in such tunnel junctions. In the rest of this chapter, we will briefly review some basic ideas in this field and introduce some existing notions which constitute the basis of our understanding of this effect. 1.2 Basic aspects Electronic structure of 3d TM FMs Elemental 3d transition metal (TM) ferromagnets (FM) like Fe, Co, and Ni and alloys derived from them have intrigued humans from time immemorial. From primeval amazement regarding the magnetic compass and its implication that the earth itself was a giant magnet, and the apparent magical power of magnets in attracting and sticking to metals like iron, to existing controversies on magnetoreception in animals and birds, and the intriguing field of planetary magnetism, humans

16 4 Chapter 1 Introduction to spin tunneling Figure 1.1: Origin of spin dependent transport. Schematic representation of spin dependent transport and the origin of the GMR effect in magnetic multilayers. (a) parallel configuration: majority electrons (spins aligned with local magnetization) traverse layers with lower scattering events as compared to minority electrons (spins aligned against local magnetization) (b) antiparallel configuration: each spin species scatters in one of the two ferromagnetic layers. Therefore, the comparative resistance in this antiparallel configuration is higher than the parallel configuration. (c) Example MR curve where the resistance is plotted as a function of the applied field. At large fields, positive or negative, the resistance is low due to the parallel configuration. Closer to zero, the trilayer is engineered to achieve antiparallel configuration in one field direction (negative field in this case). In Figure (c), note that although the MR curve is not measured on a GMR stack but a TMR stack, the primary difference is only in the magnitude of the MR effect. are yet to conquer the mysteries cast by magnetism. However, all throughout, our search for answers has been fervent, to say the least. In this section, we will try to briefly sketch the basic concepts on the question why metals like Fe, Co and Ni are ferromagnetic?

17 1.2 Basic aspects 5 The essential aspects are that electrons have intrinsic spins, and their wave functions have different spatial symmetries. These wave functions are allowed to accommodate only a certain number of electrons. When placed in a solid, the wave functions form bands which electrons occupy in k-space. In 3d TMs, the d-bands lie close to the Fermi level and may accommodate 10 electrons. Their band widths are in the order of 5 ev; much smaller than the band widths of spherically symmetric delocalized s-bands. Because of this narrow band width which needs to accommodate 10 electrons, the electronic density is high, and the Fermi surface is dominated by contributions from the d-bands. Naturally, this high density of states of 3d-electrons at the Fermi level also greatly influences the electronic and magnetic properties of the solid. The magnetic properties of 3d TMs are a consequence of the fact that the electronic wave function is required to be antisymmetric, either in its spin or spatial part. This, together with the narrow band width of 3d TMs which allows greater electron density, is the cause of a collective magnetic moment in 3d TM FMs. In order to minimize coulomb repulsion, the electrons tend to couple with their spins parallel, which forces antisymmetric spatial wave functions. This is, in essence, Hund s first rule for free atoms which renders almost 80% of the periodic table in a high spin-state. In a solid however, electrons become delocalized and the gain in exchange energy which aligns spins parallel must overcome the additional kinetic energy to put the spins in the same spin-band. Therefore, the more the electronic system becomes delocalized, the smaller the chance to display ferromagnetism. For, Fe, Co and Ni, the narrow d-band comes to rescue where the large density of states (DOS) at the E F satisfies the Stoner criterion for ferromagnetism N(E F ) I > 1, where I is the Stoner parameter and represents intra-atomic exchange and correlation effects. In other words, for these elements, the DOS [N(E F )] is large enough for parallel (ferromagnetic) coupling of spins without increasing the kinetic energy of the d-bands considerably. This energy is called the exchange splitting and is typically 1 ev. As an example the DOS of Co in a ternary alloy of CoFeB is shown in Figure 1.2 [7]. The resulting spin magnetic moment is given by m s = (N N ) µ B (1.2) that is, the difference between the occupation of the spin-up and spin-down bands. Analyzing the contribution of the various types of electrons in 3d TM FMs (different spatial symmetries of the wave function), one finds that the spin moments of the d- electrons contribute 90% to the total moment, while their orbital moment is almost completely quenched in a solid. The 4sp electrons carry no orbital moment [8, 9], and their spin moments contribute 5% to the total moment Electron and spin tunneling It is well-known that when an insulator is made very thin, of the order of a few nanometer, electrons can tunnel through this thin insulator according to laws of

18 6 Chapter 1 Introduction to spin tunneling d-dos (states/ev/atom) d d s s Energy (ev) Figure 1.2: Density of states of a ferromagnet. Representative DOS of Co in CoFeB which shows s-dos and d-dos. Here the states/ev/atom are plotted as a function of energy and E F is set to zero. The s-dos is magnified 20 times for comparison with the d-dos. Please refer to Paluskar et al. [7] or Chapter 4 for details. s-dos (states/ev/atom) quantum mechanics. Regarding the spin of these tunneling electrons, it is assumed to conserve if the electron tunnels elastically. Spin tunneling becomes relevant in the case of magnetic tunnel junctions (MTJs), where the insulator is sandwiched between two ferromagnets, as shown in Figure 1.3. In such a device, the magnitude of the tunneling current depends on the relative orientation of the magnetization of both electrodes. When the magnetization of the two electrodes is aligned parallel, a large current flows, while an antiparallel alignment of the two electrodes results in a small current. This can be understood from a few elementary arguments. (i) The tunneling current is in first order proportional to the product of the electrode s density of states at the Fermi level [N(E F )]. (ii) As we noted in the previous section, in a ferromagnet, the ground-state energy bands in the vicinity of the Fermi level are shifted in energy due to exchange splitting, yielding unequal majority and minority bands for electrons with opposite spins. (iii) Assuming spin conservation for the tunneling electrons, there are two separate currents of spin up and spin down character. As a result of these ingredients, the current between electrodes with the same magnetization direction should be higher than for oppositely magnetized electrodes. This is further illustrated in the right panel of Figure 1.3. Within this simple so-called Jullière model, the resistance change is called tunneling magnetoresistance (normalized to the lowest resistance) is given by: T MR = 2P 1P 2 1 P 1 P 2 (1.3)

19 1.2 Basic aspects 7 large current barrier E F small current N 1 maj N 1 min N 2 maj N 2 min barrier E F Figure 1.3: Spin-polarized tunneling in MTJs. Schematic illustration of the physics behind tunnel magnetoresistance. Top: for parallel aligned magnetization as sketched in the left, electrons around the Fermi level with spin-up ( ) and spin-down ( ) are allowed to tunnel from majority majority bands, and from minority minority bands. Bottom: when the magnetization of the two ferromagnets is anti-parallel, tunneling takes place for majority minority and minority majority bands, leading to a reduction of total tunneling current. In terms of electrical resistance, this corresponds to a higher resistance when the magnetization of the two layers are oppositely aligned. with P 1,2 = N maj 1,2 N maj 1,2 N maj 1,2 + N maj 1,2 (1.4) where P 1,2 is the so-called tunneling spin polarization (also called TSP in this thesis) determined by the relative difference in DOS at the Fermi level (for each electrode). However, it is crucial to realize that not all electrons present at the Fermi level can efficiently tunnel through the barrier and that this simple equation is not able to capture the physics behind a number of observations in MTJs. As we shall see later, the spherically symmetric s-like electrons which have a much lower DOS at the Fermi level dominantly tunnel through the barrier, and the interface between the insulating tunnel barrier and the ferromagnets plays an essential role. Nonetheless, this expression clearly demonstrates the presence of a magnetoresistance effect and the relevance of the magnetic character of the electrodes. Moreover, it shows that so-called half-metallic ferromagnets which have only one spin species available at the Fermi level [10], may in principle engender infinitely high TMR. Indications for such anomalous behavior have indeed observed, for instance in LaSrMnO 3 / SrTiO 3 / LaSrMnO 3 [11] and Co 2 FeAl 0.5 Si 0.5 / AlO x / Co 2 FeAl 0.5 Si 0.5 [12].

20 8 Chapter 1 Introduction to spin tunneling (a) (b) (c) (d) (e) (f) (g) (h) (i) Co Co Co Co Co Co LSMO LSMO Co / Fe Al O 2 3 Co Al O 2 3 Co Ru Co Al O 2 3 Co IrMn SrTiO Al 2 O 3 Al 2 O 3 Al 2 O 3 SrTiO 3 3 Al O Co Cu / Cr Co Co-Gd Co 2 3 Co MgO Co / Fe Normal MTJ Exchange coupled Exchange biased TMR > 0 TMR 0 TMR >> 0 TMR < 0 TMR > 0 TMR >> 0 Figure 1.4: Materials used in MTJs. (a-c) Achieving parallel or antiparallel configurations. (a) Two ferromagnetic layers with different thicknesses resulting in different coercivities. (b) Exchange coupling across Ru, where the trilayer Co / Ru / Co acts as the bottom electrode. (c) Exchange biasing the bottom Co layer with IrMn which results in the shift of the center of the hysteresis loop from zero field. (dh) MTJs engineered with various types of ferromagnetic, non-magnetic and barrier materials. These stacks were used in experiments to understand spin tunneling (see text). An important aspect for the presence of magnetoresistance is the ability to independently manipulate the direction of the magnetization of the electrodes. In other words, have easy access to a parallel or antiparallel configuration of the two magnetic electrodes. This can be accomplished by a number of methods which are schematically shown in Figure 1.4. All these methods use specific materials and their properties to change the hysteresis loop of one magnetic electrode in comparison to the other. The easiest method one can imagine is to use two different thicknesses for the two electrodes [see Figure 1.4(a)], which renders two different coercivities and switching fields. Another way to change the switching fields is to use exchange coupling across a thin metallic layer like Ru [see Figure 1.4(b)]. At certain thicknesses of Ru, it couples the two adjacent Co layers anti-ferromagnetically, and allows easy switching between the two states of the MTJ. Here, the trilayer Co / Ru / Co acts as the bottom electrode. Another method commonly used is to fix or pin the direction of one of the ferromagnetic layers with an antiferromagnet like IrMn [see Figure 1.4(c)]. In this case the hysteresis loop of the pinned layer shifts away from zero [14]. With the loop of the other electrode centered around zero, this too allows switching between the parallel and antiparallel configuration. In Figure 1.5, we show another example of a TMR measurement. Here the first type of stack shown in Figure 1.4(a) with two soft-magnetic CoFeB electrodes having different coercivities is used to create a clear distinction between the resistance levels in parallel and anti-parallel alignment of the magnetization. As the field is swept, there are sharp changes in resistances when one switches from a parallel to an antiparallel configuration or vice versa. The TMR reported here is 500% at room temperature, underlining the application potential of such a device, especially if one considers two distinctly different resistance states at two different external fields.

21 1.3 Contemporary notions on spin tunneling 9 Resistance change (%) Magnetic field (ka/m) capping layers CoFeB MgO CoFeB buffer layers substrate Figure 1.5: Example of a TMR measurement. Resistance change in a magnetic tunnel junction consisting of (Co 25 Fe 75 ) 80 B 20 / 21 Å MgO / (Co 25 Fe 75 ) 80 B 20 as shown at right. The data are taken at room temperature. The arrows at left indicate the orientation of the CoFeB magnetization. Adapted from [15]. 1.3 Contemporary notions on spin tunneling Next we will discuss some of the experiments which shed new light on the physics of magnetic tunnel junction. As mentioned in the abstract of this chapter, no attempt is made to be complete here. Please refer to the review by Swagten et al. for an exhaustive account together with a description of recent advances [13]. In 1971, Tedrow and Meservey reported the first experiments [16] on spin tunneling [see Section 2.5.1]. In their case, only one electrode was ferromagnetic (Ni), the other being a superconductor (Al). They found that though minority electrons dominate the DOS at the Fermi level of Ni, majority electrons were tunneling through the thin AlO x barrier. Later it was suggested by Hertz and Aoi (1973) [17] and by Sterns (1977) [18] that, although the dominant species of electrons at the Fermi level of transition metal ferromagnets were spin-down d electrons, they did not couple well with the states over the barrier. Instead, highly dispersive s-like electrons had a much larger overlap integral with states in the barrier which led to a larger transmission probability for these electrons. Moreover, they also realized that the interaction between the s and d-electrons (s-d hybridization) leads to a suppression of the s-dos in regions of large d-dos, which is also the case at the Fermi level of a 3d transition metal ferromagnet [17, 18]. Consequently, this induces a spin polarization of the s-dos at the Fermi energy. After these initial experiments, Jullière [19] made the first prediction of a TMR effect. Given these demonstrations and predictions in spin tunneling, mainly due to technical difficulties, it took almost 25 years to do the first successful experiment with two ferromagnetic electrodes adjacent to a tunnel barrier. Two research groups, that of Moodera et al. at MIT [5] and that of Miyazaki et al. at Tohoku Japan [6], then reported the first TMR measurements

22 10 Chapter 1 Introduction to spin tunneling on MTJs. Please note that in all these experiments AlO x was preferred as barrier material, primarily since it allowed easy growth of a pin-hole free thin barrier by natural, thermal or plasma oxidation of Al thin films. This was particularly convenient for the experiments of Tedrow and Meservey, as they used Al as a superconducting bottom electrode anyway. On the theoretical side too, there was considerable effort to model tunneling through AlO x [20, 21]. However, due to its amorphous structure which hinders ab-initio calculations, despite persistent effort, our theoretical understanding of tunneling through AlO x has remained limited [22, 23]. Therefore, many experimental attempts were made to achieve this fundamental understanding, which we will discuss below. Nevertheless, theory has provided vital evidences that the interface between the barrier and the ferromagnet, and the relevant chemistry or bonding at such an interface, is crucial for spin tunneling [22 24]. For example, using first principles calculations Belashchenko et al. predicted a sign change for the spin polarization of tunneling electrons depending on where oxygen atoms sit on a Co surface [25] AlO x : Relevant experiments Earlier, we defined TMR with a simple equation [see Eqn. 1.3] which included the spin polarization (P ) of the ferromagnetic DOS. One may imagine that P is not constant over the whole Fermi surface, and varies depending on which direction in k-space one probes, that is, on the crystallographic orientation of the electrode at the interface with the tunnel barrier. The demonstration of such a crystal anisotropy of the TMR was given by Yuasa et al. [26], who showed that the use of single-crystalline Fe electrodes of different crystal orientations in MTJs resulted in a substantially different TMR. After the demonstration of TMR in MTJs, there were various attempts to verify the simple equation 1.3 given by Jullière. As shown in Figure 1.4(d-h), many of these experiments involved inserting an additional layer at the barrier-ferromagnet interface or changing either the barrier material, or the ferromagnetic material, or both. To begin with, equation 1.3 predicts a zero TMR if any of the two electrodes has zero P. A simple test would be inserting a non-magnetic dusting layer at the barrier-ferromagnet interface and measuring TMR, as shown in Figure 1.4(e). LeClair et al. [27] showed that, surprisingly, inserting one monolayer of Cu between the bottom Co electrode and the AlO x barrier showed a finite TMR. Their results are shown in Figure 1.6(a). This indicated that a part of spin current retained its spin orientation while traversing the non-magnetic Cu layer. Moreover, while the TMR exponentially decayed with a length scale of 2.6 Å for a Cu layer, a similar layer of antiferromagnetic Cr induced an even faster exponential decay on a length scale of 1.2 Å [28]. Not only do these results clearly demonstrates the limited applicability of equation 1.3, but also the truly interfacial nature of the tunneling spin polarization P, illustrating that only a few monolayers adjacent to the tunnel barrier

23 1.3 Contemporary notions on spin tunneling 11 Norm. tunnel magnetoresistance (a) 1 Cu Cr Ru T = 10 K Thickness dusting layer (Å) (b) maj. min. T = 2 K T = 300 K NiFe Al 2 O 3 Cu(001) Co(001) Thickness dusting layer (Å) Tunnel magnetoresistance (%) Figure 1.6: Oscillations in TMR. TMR when incorporating ultrathin layers at the ferromagnet-barrier interface of a MTJ. (a) Normalized TMR data at T = 10 K for sputtered Co / X / AlO x / Co junctions, with interfacial layers X = Cu, Cr, and Ru [27 29]. (b) TMR at T = 2 K and T = 300 K as a function of the thickness of the Cu interface layer thickness in an epitaxial junction of Co(001) / Cu(001) / AlO x / NiFe. The inset schematically shows quantum well reflections for minority electrons in the Cu layer, only when propagating along k = 0; adapted from [30]. are important for tunneling. In Figure 1.6(a), one notices that although the insertion of a Ru layer at the interface also results in a exponential decay of the TMR as rapid as that due to the Cr layer, in case of the Ru layer, LeClair et al. observed a change in sign of the TMR. [29]. Although they demonstrated that the sign reversal of TMR was directly related to a change of the electrode DOS due to the interfacial mixing between Co and Ru, an alternative explanation would have been the formation of quantum well states in Ru if sharp, almost single crystalline, Co/Ru interfaces could be achieved. Later Yuasa et al. achieved such sharp interfaces between single crystalline Co (001) and Cu (001) by using molecular beam epitaxy [30]. Their MTJ stack and the corresponding TMR measured on it are shown in Figure 1.6(b). Here it is noteworthy that the amplitude of the TMR oscillation is large enough to allow the sign of the TMR ratio to alternate between positive and negative value. Yuasa et al. explained that majority electrons tunneling from NiFe into Co would transmit easily as compared to minority electrons which have a higher probability to be reflected at the Co-Cu interface. If multiple scattering occur between the Co-Cu and Cu- AlO x interfaces, the minority electrons would form resonant quantum well states (QW states) in the Cu layer, resulting in the oscillatory behavior of TMR. From the period of the oscillation, they could argue that the QW states formed in the 1 band of Cu. The importance of the dominant contribution of this highly dispersive s-like 1 band in tunneling through AlO x was reiterated by Nagahama et al. [31]

24 12 Chapter 1 Introduction to spin tunneling who fabricated single crystalline MTJs with Cr (001) inserted at the interface [see Figure 1.4(e)], similar to the work of LeClair et al. They argued that, since the band structure of an epitaxial Cr layer has no band of 1 symmetry at the Fermi level in the k = 0 direction, the electrons from one electrode can tunnel only if they are scattered at the interface of the other electrode due to the presence of the Cr layer. These above results clearly show the importance of the spherically symmetric s-like electrons in tunneling through AlO x. We will return to this point in Chapter 4. Although most ferromagnets display a positive P in conjunction with AlO x, Kaiser et al. reported that Co-Gd alloys [see Figure 1.4(f)] can exhibit both significant positive and negative P systematically depending on the alloy composition [32]. It is known that in these alloys there exist independent subnetworks of Co and Gd magnetic moments which are individually aligned ferromagnetically, but align antiferromagnetically with respect to each other. Now the sign of P depends on the orientation of the respective subnetwork magnetization with respect to the applied field. The P from either of these subnetworks will be positive when its magnetization is aligned with the applied magnetic field. However, since the moments of the other subnetwork will consequently be antiparallel to the field, it give rise to negative P. Kaiser et al. argued that the measured P is the sum of independent spin-polarized tunneling currents from the Co and Gd subnetworks, resulting in a sign change of P with alloy composition. When combined with traditional ferromagnetic materials with positive P in a MTJ, these alloys lead to a positive or negative TMR depending on the sign of Co-Gd polarization [32]. As we clarified earlier, chemical bonding at the interface has been predicted to have a great influence on P. Such bonding would influence the tunneling matrix element occurring in Fermi s golden rule which couples initial and final state wave functions depending on symmetry and overlap arguments. Consider the case of Co- Pt alloys studied by Kaiser et al. [33]. They observed that the measured P did not change after alloying ferromagnetic Co with up to 40 at.% of non-magnetic Pt, while the magnetic moment of the alloy reduced by 40% of its initial value for Co. They argued that (i) the robust magnetic moment of Co in Co-Pt alloys which did not change much from its value for pure Co and (ii) the higher tunneling rate from Co atoms at the interface as compared to Pt atoms was responsible for the robust P of Co-Pt alloys. The higher tunneling rate was argued to arise from the larger affinity of Co to bond with oxygen at the Co-Pt / AlO x interface. Kaiser et al. estimated that the tunneling probability from the Pt sites at the interface was 3.8 times lower than from the Co sites. This study suggests that it is possible to form MTJs with high P and TMR with low magnetic moment alloys by utilizing interface bonding effects and manipulating the tunneling rates of the alloy constituents [33]. Arguably the most decisive experiments demonstrating the relevance of interface bonding effects were those of Sharma et al. [34] and De Teresa et al. [35, 36]. De Teresa et al. studied MTJs with Co / I / La 0.7 Sr 0.3 MnO 3 (LSMO), where I could be SrTiO 3 (STO), Ce 0.69 La 0.31 O (CLO), or AlO x (ALO) [see Figure 1.4(g)]. In

25 1.3 Contemporary notions on spin tunneling 13 these experiments, the effective polarization of Co was found to be positive (majority electrons tunnel) with ALO as barrier, and negative (minority electrons tunnel) with STO or CLO as barrier. As the P of the STO-LSMO interface was known to be positive, the inverse TMR observed in Co / STO / LSMO junctions was a signature of a negative polarization of the Co-STO interface. This inversion of the sign of P for the the Co-STO interface with respect to the P in Co-ALO interface was confirmed by growing Co / ALO / STO / LSMO junctions [see Figure 1.4(h)] which again revealed a positive P for the Co-ALO interface. De Teresa et al. argued that the negative P of Co when the barrier is STO or CLO could be viewed as a preferential selection of electrons of d-character at the Co-STO and Co-CLO interfaces, as compared to the positive P in Co-ALO where the selection of electrons with s-character occurred at the interface. This negative P of the Co-STO interface has later been verified from first principles by Velev et al. [37]. These results again show that P, and consequently TMR should be viewed as a property predominantly determined by barrier-ferromagnet interface which is strongly influenced by the chemistry at the interface MgO: Relevant experiments As we have mentioned, due to the amorphous nature of AlO x, ab-initio studies aimed to fundamentally understand spin-dependent transport in tunnel junctions have been difficult to perform [22, 23]. Therefore, there has been a continuous effort to develop crystalline barriers which allow coherent electron transport [13]. Below the use of MgO barriers (and the observation of giant TMR) is discussed specifically due to the paramount role it plays in our fundamental understanding of tunneling and due to its technological impact on MTJs. Concept of coherent tunneling One aspect which is highly unlikely in tunneling through an amorphous barrier is k conservation of the electron wave vector. On the contrary, in a crystalline barrier, k conservation (also known as coherent tunneling) is a distinct possibility. This also implies that a wave vector selected at one interface efficiently couples to a corresponding wave vector at the other interface. Keeping in mind that P is not constant over the whole Fermi surface, and the possibility of coherent tunneling, one may imagine that using a certain electrode-barrier interface in a certain crystallographic orientation would result in efficient electron tunneling for wave functions which have specific symmetries. Among other systems, such coherent spin tunneling behavior has been theoretically predicted [38, 39] for epitaxial Fe(001) / MgO(001) / Fe(001), and later, also for other bcc ferromagnetic electrodes based on Co, and CoFe alloys. In these tunnel junctions, one describes three kinds of evanescent states ( 1, 5, 2 ) which coherently tunnel between the MgO barrier and singlecrystalline Fe electrodes [see Figure 1.7]. These 1,5,2 states are electronic states

26 14 Chapter 1 Introduction to spin tunneling Majority density-of-states Fe 1 (spd) 5 (pd) 2 (d) MgO Layer number Fe Minority density-of-states Fe MgO Fe 5 (pd) 2 (d) 2 (d) Fe Layer number Fe Figure 1.7: Origin of giant TMR in MgO based MTJs. Layer-resolved tunneling DOS for k = 0 in Fe(001) / 8 monolayers MgO / Fe(001) for majority electrons when the magnetization of the Fe layers is parallel oriented (left). Each curve is labelled by the symmetry of the incident Bloch state in the left Fe electrode, showing, for example, the slow decay of the states with 1 symmetry. The strong differences in decay is schematically illustrated in the right panel. Adapted from [38]. along the Γ X direction in k-space. The choice for Fe (001) is made on the basis of the fact that the highly dispersive 1 is present at the Fermi level only in the majority spin channel, and absent in the minority spin channel. Moreover, as shown in Figure 1.7, this band has a relatively small attenuation coefficient in MgO (001), as compared to the 5, 2 bands. In a tunnel junction, these two factors play a key role in determining the tunnel conductance for the parallel and antiparallel configuration. For instance, in the antiparallel configuration, the fact that majority 1 states efficiently tunnel through the barrier but cannot couple to the DOS of the other electrode due to the absence of such a band at the Fermi level. This is shown in Figure 1.7. In the case of bcc Co (001), the situation is even more interesting. Here, for the majority channel, only the 1 states lie at the Fermi level. Therefore, it is theoretically expected that all the states are completely reflected at k = 0 in antiparallel configuration, resulting in a giant TMR. Discovery and impact of giant TMR After a number of initial efforts to observe this enormous selectivity of the wave function symmetry in epitaxial junctions, two breakthroughs were reported. One for epitaxial (001)-oriented Fe / MgO / Fe junctions [40] and the other for highlytextured sputtered CoFe / MgO / CoFe [41], showing TMR ratios well above 100%, thereby substantially exceeding the magnetoresistance of AlO x based devices. Since then, the TMR reported for MgO based MTJs has steadily improved, in particular

27 1.4 Relevance of amorphous ferromagnets 15 by using ternary CoFeB alloys as ferromagnetic electrodes [41, 42]. It is believed that high-quality MgO can be adequately stabilized between the as-grown, amorphous CoFeB electrodes, which, after annealing at temperature up to almost 400 C, crystallize in the required bcc character. An example of a TMR measurement of around 500% at room temperature is shown in Figure 1.5 for an annealed (Co 25 Fe 75 ) 80 B 20 / MgO / (Co 25 Fe 75 ) 80 B 20 junction. Today, such junctions inspire novel ideas for various spintronics devices [43]. For example, spin-torque based MTJs where, instead of the application of an external magnetic field, the angular momentum of a spin polarized current is used to switch the magnetization of one of the ferromagnetic electrodes. Such devices aim to be the basis of future random access memories [43]. 1.4 Relevance of amorphous ferromagnets We hinted the emerging and unquestionable importance of amorphous CoFeB alloys in spintronics. Let s briefly look at amorphous alloys in general, and later, the relevance of CoFeB in particular. The first demonstration of noncrystalline Au 75 Si 25 alloy in 1960 by [44] was followed by the discovery of a stable ferromagnetic state in Fe 80 P 13 C 7 amorphous alloys by the same group in 1967 [45, 46]. These observations opened up a new avenue in both, solid state physics and materials research. The fact that many phenomena remain essentially unaltered by the absence of a periodic lattice and the consequent inapplicability of Bloch s theorem has forced a reappraisal of the theoretical framework of solid state physics [47 49]. On the materials research side, it was quickly realized that these amorphous alloys showed excellent magnetic, mechanical and corrosion resistant properties. For example, the unusually low coercivities and high resistivities of Fe-B-Si alloys allowed the reduction of core losses in power transformers by a factor of 5 over contemporary materials. Concerning mechanical properties, Inoue et al. recently demonstrated that Co 43 Fe 20 Ta 5.5 B 31.5 glassy alloys exhibit a fracture strength, and a Youngs modulus which are higher than previous values reported for any bulk crystalline or glassy alloys [50]. There are numerous other aspects like fatigue life, magnetostriction and coercivity of these alloys which make them technologically relevant; please see references [47 49] for more details. Regarding the application of amorphous ferromagnets in spintronics, to the best of our knowledge, the first use of an amorphous ferromagnetic layer was made in 1995 by Jimbo et al. [51] who reported a GMR of 5.4% in CoFeB/ Cu / Co trilayers. These CoFeB alloys were first investigated in the late 1970 s, for example by O Handley et al. and by Heiman et al. [52, 53]. Subsequently, Jimbo et al. also reported studies of exchange biased CoFeB spin valves together with an anneal study of such spin valves where they showed that annealing these trilayers up to 300 C enhanced the observed value of the GMR [54, 55]. In 2002, Kano et al. reported a TMR value of 59% in AlO x based MTJs [56]. For MTJs based on AlO x barriers, there were subsequent reports of record-high TMR of 70% (2004) and 80% (2007) at room

28 16 Chapter 1 Introduction to spin tunneling temperature by Wang et al. [57] and Wei et al. [58], respectively. Concerning MgO based MTJs, Parkin et al. reported a room temperature TMR of more than 200% in CoFe / MgO / CoFeB MTJs [41]. Since these reports there have been many reports of increasingly higher TMR values with CoFeB-MgO based MTJs [15, 42]. These alloys have also facilitated record-low switching currents in spin-torque based MTJs [59]. Consequently, they were employed to observe the novel spin-torque diode effect [60], used in junctions to measure the strength, or even the direction, of the associated spin torque [61]. In this thesis, we will venture to remind the reader about this application potential of CoFeB alloys from time to time. 1.5 This thesis From the experiments discussed above, the emerging importance of CoFeB in spintronics and its considerable impact for various spintronics applications were obvious [43] during the time of this thesis. So also was the necessity for a thorough experimental and theoretical analysis of its atomic and electronic structure and their combined impact on its tunneling spin polarization (P or TSP). Therefore, this thesis is devoted to the fundamental understanding of the properties of ternary CoFeB alloys, and is an endevour to explore open questions in spin tunneling by using these properties. After this first introductory chapter (Chapter 1) which deals with a few contemporary notions regarding spin tunneling, Chapter 2 addresses the various deposition and experimental analysis tools used in this thesis. Here, to exemplify the various techniques, a few experimental results relevant to later chapters will also be presented. In Chapter 3, we will investigate some structural aspects of CoFeB alloys. In particular, we will investigate the influence of crystallization of these amorphous alloys on their structural and magnetic properties after a single anneal. We will use this information in later chapters as a starting point for further experimental work. In Chapter 4, we will investigate the atomic and electronic structure of a single CoFeB composition from first principles. Also, we will specifically investigate the TSP of an amorphous ternary alloy, an issue never addressed before, and compare it with its crystalline counterpart. Surprisingly, we find that the TSP of the amorphous alloy is larger than its crystalline counterpart. We also show that for these amorphous alloys, the spin polarization of the s-electron DOS at the Fermi level is a very good representative of the TSP in AlO x based junctions. In Chapter 5, we probe some aspects of inelastic tunneling of electrons when a sharp contrast structural change from amorphous to crystalline electrode is induced at the barrier-ferromagnet interface. In particular, the changes in the low energy magnetic excitations induced by inelastically tunneling electrons are investigated. In Chapter 6, we will probe the correlation between magnetism and TSP in

29 1.5 This thesis 17 CoFeB alloys. Such a correlation has been an outstanding issue in spin tunneling since its first observation in We find that the amorphous CoFeB alloys are very suitable to address this issue. We will focus on properties of d-electrons probed by synchrotron radiations in relation to the properties of s-electrons probed by electronic transport measurements. Our data support the conjecture that such a correlation between the d and s-electrons may exist. Finally, in Chapter 7, we will investigate the thermal stability of MTJs and the effect of high-temperature annealing. Specifically, the role of Mn diffusion from the antiferromagnets used to exchange bias one of the ferromagnetic layers is probed. We find that though Mn diffuses after annealing, it does not seem to influence the TSP.

30 18 Chapter 1 Introduction to spin tunneling Bibliography [1] M. N. Baibich, J. M. Broto, A. Fert, R. Nguyen van Dau, F. Petroff, P. Eitenne, G. Creuzet, A. Friederich, and J. Chazelas, Giant Magnetoresistance of (001)Fe/(001)Cr Magnetic Superlattices. Phys. Rev. Lett. 61, 2472 (1988). 1.1 [2] G. Binasch, P. Grünberg, F. Saurenbach, and W. Zinn, Enhanced magnetoresistance in layered magnetic structures with antiferromagnetic interlayer exchange. Phys. Rev. B 39, 4828 (1989). 1.1 [3] A. Barthélémy, A. Fert, and F. Petroff, Giant magnetoresistance in magnetic multilayers. B. Buschow, ed. Handbook of Magnetic Materials. Elsevier, Oxford, UK, Vol. 12, Chap. 1 (1999). 1.1 [4] R. Coehoorn, Giant magnetoresistance and magnetic interactions in exchangebiased spin valves B. Buschow, ed. Handbook of Magnetic Materials. Elsevier, Oxford, UK, Vol. 15, Chap. 1 (2003). 1.1 [5] J. Moodera, L. R. Kinder, T. M. Wong, and R. Meservey, Large magnetoresistance at room temperature in ferromagnetic thin film tunnel junctions. Phys. Rev. Lett. 74, 3273 (1995). 1.1, 1.3 [6] T. Miyazaki, and N. Tezuka, Giant magnetic tunneling effect in Fe/Al 2 O 3 /Fe junction. J. Magn. Magn. Mater. 139, L231 (1995). 1.1, 1.3 [7] P. V. Paluskar, G. A. de Wijs, J. J. Attema, S. Fiddy, E. Snoeck, J. T. Kohlhepp, H. J. M. Swagten, R. A. de Groot, and B. Koopmans, Spin tunneling in junctions with disordered ferromagnets. Phys. Rev. Lett. 100, (2008) , 1.2 [8] P. Söderlind, O. Eriksson, B. Johansson, R. C. Albers, and A. M. Boring, Spin and orbital magnetism in Fe-Co and Co-Ni alloys. Phys. Rev. B 45, (1992) [9] O. Eriksson, A. M. Boring, R. C. Albers, G. W. Fernando, and B. R. Cooper, Spin and orbital contributions to surface magnetism in 3d ferromagnets. Phys. Rev. B 45, 2868 (1992) [10] R. A. De Groot, F. M. Mueller, P. G. Van Engen, and K. H. J. Buschow, New class of materials: half-metallic ferromagnets. Phys. Rev. Lett. 50, 2024 (1983) [11] M. Bowen, M. Bibes, A. Barthélémy, J.-P., Contour, A. Anane, Y. Lemaitre, and A. Fert, Nearly total spin-polarization in La 2/3 Sr 2/3 MnO 3 from tunneling experiments. Appl. Phys. Lett. 82, 233 (2003)

31 BIBLIOGRAPHY 19 [12] N. Tezuka, N. Ikeda, S. Sugimoto, and K. Inomata, Giant Tunnel Magnetoresistance at Room Temperature for Junctions using Full-Heusler Co 2 FeAl 0.5 Si 0.5 Electrodes. Jpn. J. Appl. Phys. 46, L454 (2007) [13] H. J. M. Swagten, Spin tunneling in magnetic junctions. B. Buschow, ed. Handbook of Magnetic Materials. Elsevier, Oxford, UK, Vol. 17, Chap. 1 (2007). 1.3, [14] J. Nogués, and I. K. Schuller, Exchange bias. J. Magn. Magn. Mat. 192, 203 (1999) [15] Y. M. Lee, J. Hayakawa, S. Ikeda, F. Matsukura, and H. Ohno, Effect of electrode composition on the tunnel magnetoresistance of pseudo-spin-valve magnetic tunnel junction with a MgO tunnel barrier. Appl. Phys. Lett. 90, (2007). 1.5, 1.4 [16] P. M. Tedrow, and R. Meservey, Spin-dependent tunneling into ferromagnetic Ni. Phys. Rev. Lett. 26, 192 (1971). 1.3 [17] J. A. Hertz, and K. Aoi, Spin dependent tunneling from transition metal ferromagents. Phys. Rev. B 8, 3252 (1973). 1.3 [18] M. B. Sterns, Simple explanation of tunneling spin polarization of Fe, Co, and Ni and its alloys. J. Mag. Mag. Mater. 5, 167 (1977). 1.3 [19] M. Jullière, Tunneling between ferromagnetic films. Phys. Lett. 54A, 225 (1975). 1.3 [20] J. C. Slonczewski, Conductance and exchange coupling of two ferromagnets seperated by a tunneling barrier. Phys. Rev. B, 39, 6995 (1989). 1.3 [21] E. Yu. Tsymbal, and D. G. Pettifor, Modelling of spin-polarized electron tunneling from 3d-ferromagnets. J. Phys.: Condens. Matter, 9, L411 (1997). 1.3 [22] I. I. Oleinik, E. Y. Tsymbal, and D. G. Pettifor, Structural and electronic properties of Co/Al 2 O 3 /Co magnetic tunnel junctions from first principles. Phys. Rev. B 62, 3952 (2000). 1.3, [23] E. Y. Tsymbal and K. D. Belashchenko, Role of interface bonding in spindependent tunneling. J. Appl. Phys., 97, 10C910 (2005). 1.3, [24] E. Y. Tsymbal, I. I. Oleinik, and D. G. Pettifor, Oxygen-induced positive spin polarization from Fe in the vacuum barrier. J. Appl. Phys. 87, 5230 (2000). 1.3 [25] K. D. Belashchenko, E. Y. Tsymbal, I. I. Oleinik, and M. van Schilfgaarde, Positive spin polarization in Co/Al 2 O 3 /Co tunnel junctions driven by oxygen adsorption. Phys. Rev. B, 71, (2005). 1.3

32 20 Chapter 1 Introduction to spin tunneling [26] S. Yuasa, T. Sato, E. Tamura, Y. Suzuki, H. Yamamori, K. Ando, and T. Katayama, Magnetic tunnel junctions with single-crystalline electrodes: a crystal anisotropy of tunnel magneto-resistance. Euro. Phys. Lett. 52, 344 (2000) [27] P. LeClair, H. J. M. Swagten, J. T. Kohlhepp, R. J. M. van der Veerdonk, and W. J. M. de Jonge, Apparent decay of spin polarization in Cu-dusted Co/Al 2 O 3 /Co tunnel junctions. Phys. Rev. Lett. 84, 2933 (2000) , 1.6 [28] P. LeClair, J. T. Kohlhepp, H. J. M. Swagten, and W. J. M. de Jonge, Interfacial density of states in magnetic tunnel junctions. Phys. Rev. Lett. 86, 1066 (2001) [29] P. LeClair, B. Hoex, H. Wieldraaijer, J. T. Kohlhepp, H. J. M. Swagten, and W. J. M. de Jonge, Sign reversal of spin polarization in Co/Ru/Al 2 O 3 /Co magnetic tunnel junctions. Phys. Rev. B 64, (R) (2001). 1.6, [30] S. Yuasa, T. Nagahama, and Y. Suzuki, Spin-polarized resonant tunneling in magnetic tunnel junctions. Science 297, 234 (2002). 1.6, [31] T. Nagahama S. Yuasa, E. Tamura, and Y. Suzuki, Spin-dependent tunneling in magnetic tunnel junctions with a layered antiferromagnetic Cr(001) spacer- Role of band structure and interface scattering. Phys. Rev. Lett. 95, (2005) [32] C. Kaiser, A. F. Panchula, and S. S. P. Parkin, Finite tunneling spin polarization at the compensation point of rare-earth-metal-transition-metal alloys. Phys. Rev. Lett. 95, (2005) [33] C. Kaiser, S. van Dijken, S.-H. Yang, H. Yang, and S. S. P. Parkin, Role of tunneling matrix elements in determining the magnitude of the tunneling spin polarization of 3d transition metal ferromagnetic alloys. Phys. Rev. Lett. 94, (2005) [34] M. Sharma, S. X. Wang, and J. H. Nickel, Inversion of spin polarization and tunneling magnetoresistance in spin-dependent tunneling junctions. Phys. Rev. Lett. 82, 616 (1999) [35] J. Teresa, S. Barthélémy, F. Fert, H. Contour, R. Lyonnet, F. Montaigne, H. Seneor, and A. Vaures, Inverse tunnel magnetoresistance in Co/SrTiO 3 /La 0.7 Sr 0.3 MnO 3 : new ideas on spin-polarized tunneling. Phys. Rev. Lett. 82, 4288 (1999) [36] J. Teresa, S. Barthélémy, F. Fert, H. Contour, F. Montaigne, and H. Seneor, Role of metal-oxide interface in determining the spin polarization of magnetic tunnel junctions. Science 286, 507 (1999)

33 BIBLIOGRAPHY 21 [37] V. P. Velev, K. D. Belashchenko, D. A. Stewart, M. van Schilfgaarde, S. S. Jaswal, and E. Y. Tsymbal, Negative Spin Polarization and Large Tunneling Magnetoresistance in Epitaxial Co SrTiO3 Co Magnetic Tunnel Junctions. Phys. Rev. B, 95, (2005) [38] W. H. Butler, X.-G. Zhang, T. C. Schulthess, and J. M. MacLaren, Spindependent tunneling conductance in Fe MgO Fe sandwiches. Phys. Rev. B, 63, (2001) , 1.7 [39] J. Mathon, and A. Umerski, Theory of tunneling magnetoresistance of an epitaxial Fe/MgO/Fe(001) junction. Phys. Rev. B, 63, (R) (2001) [40] S. Yuasa, T. Nagahama A. Fukushima, Y. Suzuki, and K. Ando, Giant roomtemperature magnetoresistance in single-crystal Fe/MgO/Fe magnetic tunnel junctions. Nature Mater. 3, 868 (2004) [41] S. S. P. Parkin, C. Kaiser, A. Panchula, P. M. Rice, B. Huges, M. Samant, S.- H. Yang, Giant tunneling magnetoresistance at room temperature with MgO (100) tunnel barriers. Nature Mater. 3, 862 (2004) , 1.4 [42] D. D. Djayaprawira, K. Tsunekawa, M. Nagai, H. Maehara, S. Yamagata, N. Watanabe, S. Yuasa, Y. Suzuki, and K. Ando, 230% room-temperature magnetoresistance in CoFeB / MgO / CoFeB magnetic tunnel junctions. Appl. Phys. Lett. 86, (2005) , 1.4 [43] C. Chappert, A. Fert, and F. Nguyen, The emergence of spin electronics in data storage. Nature Mater. 6, 813 (2007) , 1.5 [44] W. Klement, and R. H. Willens, P. Duwez, Non-crystalline structure in solidified gold-silicon alloys. Nature 187, 869 (1960). 1.4 [45] P. Duwez, Trans. Am. Soc. Met. 60, 607 (1967). 1.4 [46] P. Duwez, and S. C. H. Lin, Amorphous ferromagnetic phase in iron-carbonphosphorus alloys. J. Appl. Phys. 38, 4096 (1967). 1.4 [47] K. Moorjani, and J. M. D. Coey, Magnetic Glasses, (Elsevier, Amsterdam) (1984). 1.4 [48] R. Hasegawa, Glassy metals : magnetic, chemical, and structural properties, (CRC press, Boca Raton) (1983). [49] T. Egami, Magnetic amorphous alloys: physics and technological applications. Rep. Prog. Phys. 47, 1601 (1984). 1.4 [50] A. Inoue, B. Shen H. Koshiba, and H. Kato, A. R. Yavari, Cobalt-based bulk glassy alloy with ultrahigh strength and soft magnetic properties. Nature Mater. 2, 661 (2003). 1.4

34 22 Chapter 1 Introduction to spin tunneling [51] M. Jimbo, K. Komiyama, H. Matue, S. Tsunashima, and S. Uchiyama, Giant magnetoresistance effect in amorphous CoFeB sandwiches. Jpn. J. Appl. Phys. 34, L112 (1995). 1.4 [52] R. C. O Handley, R. Hasegawa, and R. Ray, C.-P. Chou, Ferromagnetic properties of some new metallic glasses. Appl. Phys. Lett. 29, 330 (1976). 1.4 [53] N. Heiman, R. D. Hempstead, and N. Kazama, Low coercivity amorphous magnetic alloy films. J. Appl. Phys. 49, 5663 (1978). 1.4 [54] S. Tsunashima, M. Jimbo, Y. Imada, and K. Komiyama, Spin valves using amorphous magnetic layers. J. Magn. Magn. Mater. 165, 111 (1997). 1.4 [55] M. Jimbo, K. Komiyama, Y. Shirota, Y. Fujiwara, S. Tsunashima, and M. Matsuura, Thermal stability of spin valves using amorphous CoFeB. J. Magn. Magn. Mater. 165, 308 (1997). 1.4 [56] H. Kano, K. Bessho, Y. Higo, K. Ohba, M. Hashimoto, T. Mizuguchi, and M. Hosomi, MRAM with improved magnetic tunnel junction material. InterMag 2002 Dig. (Amsterdam) BB04 (2002). 1.4 [57] D. Wang, C. Nordman, J. M. Daughton, Z. Qian, and J. Fink, 70% TMR at room temperature for SDT sandwiche junctions with CoFeB as free and reference layers. IEEE Trans. Mag. 40, 2269 (2004). 1.4 [58] H. X. Wei, Q. H. Qin, M. Ma R. Sharif, and X. F. Han, 80% tunneling magnetoresistance at room temperature for thin Al-O barrier magnetic tunnel junction with CoFeB as free and reference layers. J. Appl. Phys. 101, 09B501 (2007). 1.4 [59] J. Hayakawa, S. Ikeda, Y. M. Lee, R. Sasaki, T. Meguro, F. Matsukura, H. Takahashi, and H. Ohno, Current-driven magnetization switching in CoFeB/MgO/CoFeB magnetic tunnel junctions. Jpn. J. Appl. Phys. 44, L1267 (2005). 1.4 [60] A. A. Tulapurkar, Y. Suzuki, A. Fukushima, H. Kubota, H. Maehara, K. Tsunekawa, D. D. Djayaprawira, N. Watanabe, and S. Yuasa, Spin-torque diode effect in magnetic tunnel junctions. Nature 438, 339 (2005). 1.4 [61] H. Kubota, Y. Suzuki, A. Fukushima, H. Kubota, H. Maehara, K. Tsunekawa, D. D. Djayaprawira, N. Watanabe, and S. Yuasa, Quantitative measurement of voltage dependence of spin-transfer torque in MgO-based magnetic tunnel junctions. Nature Phys. 7, 37 (2007). 1.4 [62] H. J. M. Swagten, and P. V. Paluskar, Magnetic tunnel junctions. Encyclopedia of Materials Science and Technology, to appear. 1

35 Chapter 2 Probing electronic, magnetic and structural properties Experiments analyzing CoFeB Abstract: This chapter 1 presents brief but requisite information on the various experimental techniques used in this thesis. While doing so, we also present some relevant but occasionally unpublished results on materials like CoFeB and MgO obtained using some of these techniques. Most of these results will be of relevance in later chapters. The chapter is divided in five main sections: sample preparation, structural characterization, in-situ measurements of electronic properties, magnetic characterization and electronic transport. No attempt is made to be complete or exhaustive. Instead, the reader is referred to suitable references which do justice to and explain in detail the particular technique under question. 1 A part of the last section of this chapter is under review. 23

36 24 Chapter 2 Probing electronic, magnetic and structural properties 2.1 Sample fabrication We begin this chapter with the essential procedure followed for samples preparation in this thesis which mainly involves deposition of various materials and oxidation of Al thin films. Prior to this discussion, lets summarize the choice of substrates used and substrate cleaning procedures followed during this thesis Substrate and substrate cleaning considerations The tunnel junctions are deposited on glass substrates, in particular 1 mm thick barium borosilicate glass sheets provided by Corning Inc. (glass code 7059). A crucial point to be considered is the roughness of the substrate for the spin polarized tunneling measurements which use very thin (35 Å) aluminum films as electrodes. From previous experience [1], the glass substrates were found to allow easy deposition of closed Al layers, as compared to silicon wafers. This difference may be ascribed to lower surface roughness, since the glass substrates are manufactured using the fusion process [2] where the glass is slowly cooled from the liquid phase to the glass phase. On the contrary, the surface of the Si wafer consists of SiO x formed at room temperature during the first exposure of the wafer to air. For all other purposes, Si (001) substrates were used due to the easy of cleaving, cheap and wide availability, and relatively good surface smoothness. For the removal of any organic material on the substrate, we first ex-situ immersed the substrate in ammonia and placed the beaker in an ultrasonic bath for 10 minutes. Subsequently, the substrate was immersed in ethanol and the procedure was repeated. Then the substrate was placed in a closed isopropanol chamber where isopropanol was being constantly evaporated. In these ex-situ cleaning steps, ammonia dissolves organic molecules and the alcohols allow removal of residue accumulated during the ammonia dip. This ex-situ cleaned substrate was stuck with silver paint to the substrate holder and loaded in the system load-lock. The final cleaning step was in-situ plasma-cleaning in an oxygen plasma. This step allows the conversion of any residual hydrocarbons from the ex-situ cleaning procedures into volatile carbon oxides and water vapor leaving a significantly cleaner substrate after the chamber is pumped to UHV. See Section for details of the plasma-cleaning procedure Deposition: Sputtering One of the most important research tool of this thesis is the ultra-high vacuum (UHV) deposition system used - EUFORAC (Eindhoven University nano-film deposition Research and Analysis Center). A picture of this facility is shown in Figure 2.1. It consists of a 6 target sputter deposition chamber, an oxidation chamber, an organic molecular beam epitaxy (MBE), a metal MBE, an in-situ photoelectron spectroscopy characterization tool, and an in-situ scanning tunneling microscope,

37 2.1 Sample fabrication 25 5 F K J J A H + D = > A H H C = E? + D = > A H N = J E + D = > A H H = I F H J * - * - 2 H A F = H = J E : Figure 2.1: UHV deposition system. This picture shows the EUFORAC system where a large number of UHV deposition and characterization tools are implemented making this an extremely powerful nano-tool. all connected to each other via transport chambers held at UHV. The capabilities of the system in growing and analyzing various sorts of thin films go hand in hand with its versatility. For more details on the capabilities of the EUFORAC, please refer to the thesis of P. LeClair [3]. In this thesis, all the samples were grown using sputter deposition. Although, in the context of magnetic films, there have been some reports of the growth of epitaxial films using sputter deposition [4, 5], generally, sputtering implies that the

38 ) * + 26 Chapter 2 Probing electronic, magnetic and structural properties = I F K J J A A F I E J E > I D M = I J = H C A J = C A J = J A H E = ) H A ) ) N ) H = J I F D A H A > = H J = H C A J = J A H E = I D M = I I K > I J H = J A +. A * )? M C A C H M J F = I = N = J E J = H C A J = J A H E = M C A = I H E C I D = F A A? J A I A I K > I J H = J A = J I F D A H A > = H I K > I J H = J A Figure 2.2: Schematic of various deposition and oxidation techniques. (a) Sketch exemplifying sputter deposition. (b) Sketch of the shadow masks used to deposit tunnel junctions. (c) Growing wedge shaped samples for thickness dependent studies. (d) Sketch of the plasma oxidation technique. layers are either polycrystalline or amorphous depending on the material. Nevertheless, magnetic tunnel junctions and a variety of sensors based on the GMR effect are popularly and conveniently grown by sputtering. Our system is a 6 source Kurt J. Lesker sputter tool equipped with a home-built load-lock. Typical base pressure after a bake-out is mbar. However, following a target change which requires breaking vacuum, the system readily achieves a base pressure of mbar without bake-out after pumping for 48 hours. Residual partial gas pressures in the chamber can be monitored with a remote gas analyzer based on mass spectrometry. This analyzer was installed on the system during this thesis. Although exhaustive reviews on sputtering are available [6, 7], let us briefly summarize the basic physical aspects of the technique, as shown in Figure 2.2(a). The material to be deposited is produced in the form of a palette and attached to an anode which is typically held at -100 to V. When a gas, typically a nobel gas like Ar, is inserted in the UHV chamber, it gets ionized. The positively charged Ar + particles accelerate towards the target material, knocking out atoms

39 2.1 Sample fabrication 27 from the target material on impact. This bombardment with a non-reactive nobel gas is the basis idea behind sputtering. The ejected atoms then scatter out, and may be allowed to condense on a substrate of suitable choice to form a film. Normally, a removable shutter is placed between the target and the sample, allowing one to control the deposition on the sample. Due to the relatively large Ar pressures used in sputtering ( mbar), the characteristic mean free path is typically smaller than the target-substrate distance and the target atoms arrive at the substrate in a broad cone of angles. Also, the energies with which the target atoms arrive at the substrate are known to be significantly larger than those for MBE. Both these distinctions, higher energy and larger cone of incoming flux of target atoms distinguish sputtering from the MBE technique. Finally, a key advantage of sputtering is that it allows the deposition of a multitude of materials which includes binary, ternary and quaternary alloys, an essential requirement of this thesis. Typically, as shown in Figure 2.2(a), a magnet is placed behind the target material which induces a helical motion of the electrons and Ar ions. This in turn increases the ionization probability of the rest of the gas, allowing the use of low Ar pressures during deposition and increasing the deposition rate. Such a technique is known as magnetron sputtering. The need of low Ar pressures stems from the fact that only a certain purity of Ar is commercially available, and any impurities like nitrogen or oxygen will essentially react with the target atoms, and degrade the purity of the deposited film. If the correct growth modes for the deposited films can be maintained, the higher growth rates accessible to magnetron sputtering are also desirable since they too allow lower deposition times in the not so pure commercially available Ar. In order to achieve maximum purity of the deposited films, during the period of this thesis, we installed a 9N purity ( %) Ar gas filter on the existing Ar lines on the EUFORAC. For growing tunnel junctions, the shadow mask technique was used in the socalled crossed-strip geometry [see Figure 2.2(b)]. A shadow mask consists of a thin metal plate with narrow rectangular strips cut through it. The strips are typically machine-cut for widths of µm. When placed directly on the substrate, the sputtered material arrives through these slits in the mask, reproducing the shape of the strips. This is exemplified in Figure 2.2(b). Here, the bottom electrode (e.g., Al/AlO x ) is grown through the horizontal strip, while the top electrode (e.g., CoFeB/Al) is grown through the six vertical strips. This essentially produces orthogonal strips with several tunnel junction at their intersection. It is worthy to mention another typical trick used in growth of single layers for specific characterization studies of thin films. The deposition of wedges is a key asset in this respect. As shown in Figure 2.2(c), typically a metal mask is positioned between the substrate and the target, and slowly retracted across the face of the substrate as the deposition proceeds. If the growth rate at a specific set of deposition conditions is known, then the retraction of the mask at a constant velocity results in a film whose thickness changes with a uniform gradient over the distance traversed

40 28 Chapter 2 Probing electronic, magnetic and structural properties by the mask. Such a wedge can then be characterized by a variety of magnetic and spectroscopic techniques which specialize in local probing of the corresponding property. An example is the MOKE technique used for magnetic characterization of the samples, which we will look at in Section Plasma oxidation All the AlO x based tunnel junction prepared throughout this thesis employ plasma oxidation of thin Al films in an oxygen atmosphere. This is a well-established and most widely used technique in preparing the AlO x tunnel barriers. The highest TMR values reported for AlO x based MTJs invariably use this method [8]. Moreover in comparison to thermal or natural oxidation, plasma oxidation is a very quick method of oxidizing Al films. Figure 2.2(d) shows a simplified sketch of the technique. The oxidation is carried out in a UHV chamber with a typical base pressure of < mbar. For oxidation, the chamber is filled with mbar pure oxygen (purity 6N). Then a high negative voltage is applied to the ring-shaped electrode B in Figure 2.2(d) with respect to the grounded ring-shaped electrode A. The potential difference between these two electrodes is 2 kv max. This potential difference creates an oxygen plasma which allows oxidation of the sample placed directly underneath the ring-shaped electrodes. Ring C in Figure 2.2(d) is used as a shield against hot ions sputtered from ring B. Typically the high-voltage power supply is used in the current limited configuration. For plasma cleaning of the substrates, the current is set to 15 ma (V DC kv) while for plasma oxidation, the current limit is set to 7.5 ma. The distance between ring C and the sample is typically set to 37 mm. 2.2 Structural characterization Next we will discuss three different structural characterization tools used in this thesis: x-ray diffraction (XRD), x-ray absorption fine structure (XAFS) and highresolution transmission electron microscopy (HRTEM). Regarding the XAFS technique, we will briefly outline the main principle and not go into any details of the data fitting procedures used. It is appropriate to add a note of thanks to Dr. Etienne Snoeck at CEMES-CNRS Toulouse, France and Dr. Steven Fiddy at station 7.1 of Daresbury labs, UK who performed and analyzed the HRTEM and XAFS measurements on amorphous CoFeB films. These measurements were a significant contribution to the understanding of the structure of these ternary alloys, one of the primary goals of this thesis. These measurements will be discussed in Chapter 4 and Chapter 5

41 2.2 Structural characterization 29 : 4, E J A I E J O? F I 6 = J # C J # # I K H? A " # I A J A? J H 0 E C D = C A : 4, =! # " " # # # # $ F A G > I K H? A A J A? J H I = F A / 1 : ) 6 = # C 6 = #! " Figure 2.3: Example of x-ray diffraction measurements. (a) A conventional 2 θ θ scan on Ta ( 50 Å) / Mg ( 1100 Å) / Ta ( 50 Å) stack grown on a glass substrate, the geometry of which is shown in the schematic. The measured data is on the same stack. The sharp peak around 2 θ = 34.4 is due to the Mg layer, while the broad bump centered around 37.6 may arise from the Ta cap and buffer layer. (b) A glancing angle measurement. The schematic shows the geometry of the measurement. The number of fringes per degree is a measure of the thickness of the principle layer. One notes a slower beating around 2 4 arising due to the comparatively thinner Ta layers X-ray diffraction (XRD) XRD is one of the most common techniques for investigating the crystallographic structure of materials, especially crystalline materials. The set-up used for the measurements described in this thesis is a Philips PWD-3710 x-ray diffractometer using a Cu K α x-ray source. The wavelengths of the x-rays are λ Kα1 = Å and λ Kα2 = Å. The sample is exposed to the x-rays at an angle θ and the reflected rays are detected at an angle 2 θ with respect to the original beam. The basic principle of XRD is the interference of radiation with a wavelength comparable to the typical lattice spacing of the crystal when reflected from different crystal planes. Having wavelengths in the order of 0.1 nm and capable of penetrating large distances into the lattice, x-rays are ideally suited for this technique. When such a beam impinges upon a crystal, a small part of it will be reflected at each plane, and the wave vector perpendicular to the planes will determine whether the different reflected waves will interfere constructively or destructively. The perpendicular component of the wave vector can easily be varied by changing the angle of incidence of the beam [see insets in Figure 2.3(a)]. Such an analysis can also be performed for crystal planes that are not parallel to the surface of the sample. This allows for a discrimination between single-crystalline films and textured films, in which only the planes parallel to the surface have the same orientation throughout

42 30 Chapter 2 Probing electronic, magnetic and structural properties the sample. In Figure 2.3(a), the diffraction intensity is plotted as a function of 2 θ for a thick Mg film grown on and capped with Ta. The substrate used is glass. The sharp peak around 2 θ = 34.4 is due to the hcp Mg (0002) layer, while the broad bump centered around 37.6 probably arises from an expanded bcc Ta (110) layers. However, this is a conjecture based on the coherence length (see below). From the angle at which constructive interference occurs it is easy to determine the lattice spacing of Mg, d, by means of the Bragg law 2d sin θ = n λ. Here, θ is the angle of incidence relative to the planes, n is an integer and λ is the wavelength of the x-ray. For Mg (0002), the measurement suggests d = Å, which is equal to half the lattice constant of hcp Mg (a = 5.21 Å). Furthermore, by means of the Scherrer formula [9], the length, t, over which the layer stacking is coherent can be estimated. In other words, the size of the crystallites in the direction perpendicular to the sample surface can be calculated: t(å) = Kλ FWHM cos θ (2.1) with FWHM the line broadening (full width at half the maximum intensity (FWHM)) and K the so-called shape factor, which is usually about 0.9. The Scherrer formula can be understood as follows. For x-rays reaching the samples at angles close to those obeying the Bragg condition the difference in path length for rays reflected at two neighboring planes is still close to one wavelength. As a result the first plane that scatters a wave exactly out of phase compared to the top plane will already be far away inside the crystal. Thin crystals or thin crystallites may then be too small to obtain completely destructive interference, and the peak in the diffraction spectrum will get a finite width with a FWHM as determined by the Scherrer formula. From the diffraction peaks shown in Figure 2.3(a), the coherence length of Mg is found to be 550 Å, half the layer thickness of 1100 Å, while that of the peak at 37.6 is found to be 50 Å, which is the layer thickness used for the seed and cap Ta layers. This coherence length of 50 Å for Ta is in good agreement with the grain sizes measured before for Ta layers with in-situ STM [3]. This would imply that the broad peak we observe at 37.6 (d = Å) may be a compressed bcc Ta (110) peak which is known to display its highest intensity at 38.5 (d = Å). When the x-rays are incident under a very small angle with the sample surface, the technique is often called grazing incidence x-ray analysis (GIXA). GIXA is used to determine layer thicknesses for thin layers of 100 to 1000 Å. In our case, we also used it for calibrating the growth rate for our films. Figure 2.3(b) shows an example on the same Mg sample used in Figure 2.3(a). The diffraction pattern is then measured in a range of angles where it is governed by reflections from different interfaces in the sample instead of reflections from all the atomic planes. As a consequence the refractive index of the material has to be taken into account and the

43 2.2 Structural characterization 31 diffraction pattern is no longer described by Braggs law. Therefore, layer thicknesses are determined from a simulation of the (low-angle) diffraction pattern [11, 12]. In Figure 2.3(b), the number of fringes per degree is a measure of the thickness of the principle layer. One notes a slower beating around 2 θ = 2 4 arising due to the comparatively thin Ta layers. In conclusion of this section, we would also like to point to several textbooks which discuss the physics behind XRD [9, 10]; an extensive treatment is for instance given by Cullity [10] X-ray absorption fine structure (XAFS) XAFS is primarily a synchrotron radiation based technique which deals with the details of the x-ray absorption spectrum of an atom at, and above, its core level binding energies. The modulation of the absorption probability due to variation in the chemical and physical state of an atom results in the XAFS spectrum. By chemical and physical state of the atom, we mean oxidation, coordination number and distance to the atoms that are neighbors of the excited atom. The primary reason that this technique is especially relevant to this thesis which deals with amorphous ferromagnets can be articulated with the title of the seminal paper by Sanders, Stern and Lytle that went New Technique for Investigating Noncrystalline Structures: Fourier Analysis of the Extended X-Ray Absorption Fine Structure [13]. We will use this technique in Chapter 4. Lets have a look at the basic idea behind the technique. The absorption coefficient (µ) of x-rays is given by Beer s law I = I 0 e µt where I 0 is incident x-ray intensity, t is the sample thickness. The fact that µ Z 4, where Z is the atomic number is at the heart of the element specific nature of the technique. Figure 2.4(a) illustrates the basic principle responsible for the XAFS, viz., photoabsorption. When the x-ray energy is equal to the binding energy of electrons in the sample, there is a sharp rise in the absorption, as the electron is promoted to the continuum. This is the basis of many of the techniques which we will discuss later in this chapter (like XPS, UPS and XMCD). According to quantum theory, to the first order, this photoelectron can be visualized as an outgoing spherical wave centered at the excited atom. As sketched in Figure 2.4(b), this electron wave is scattered by neighboring atoms, and the new waves emanating from each scattering site are superposed to the initial outgoing wave. The interference of the initial and scattered waves at the absorbing atom affects µ. In an XAFS measurement, the energy dependence of µ at and just above these core level binding energies is probed. With increasing photon energy the wave vector of the photoelectron wave increases, leading to alternating constructive and destructive interference. Moreover, the excited electron leaves a core hole and subsequently decays after a few femtoseconds from the absorption event. The decay involves the relaxing of higher energy electrons into the core hole, radiating either x-rays (fluorescence) or electrons (Auger). Both these radiations have characteristic energies belonging to the parent atom and can be used to detect the absorption event. For a broad review of this technique, refer to [14].

44 32 Chapter 2 Probing electronic, magnetic and structural properties Figure 2.4: Principles of x-ray absorption. (a) A schematic showing the basics of x-ray absorption. The x-ray photon may be linearly or circularly polarized. When linearly polarized, the absorption from a core level to the Fermi level results in XAS spectrum. When (for example) the fluorescence yield is measured and plotted against photon energy, one refers to XAFS. If the photons are circularly polarized, the resulting difference in absorption for left and right circularly polarized light is called XMCD. (b) A schematic of the origin of the XAFS spectrum is shown. The photoelectron from the excited atom A propagates through the lattice. The wave nature of the electron interacts with other atoms which causes scattering and oscillations in the absorption probability, giving rise to the XAFS spectrum. Figure 2.5 shows the XAFS set-up on station 7.1 of the Daresbury labs. Polychromatic x-rays are produced by a synchrotron radiation source, and a desired energy band of approximately 1 ev bandwidth is then selected by diffraction from a silicon double crystal monochromator. Only those x-rays that are of the correct wavelength λ (λ = hc/e, where h is Planck s constant and c is the speed of light) to satisfy the Bragg condition nλ = 2d sinθ at the selected angle θ will be reflected from the first crystal; the others are absorbed. The parallel second crystal is used as a mirror to restore the beam to its original direction. The intensity of the monochromatic x-rays is monitored in the I 0 chamber (usually with gas ionization chambers) and then allowed to irradiate the sample. The detector used is a 9-element monolithic germanium crystal (Canberra). A typical XAFS measurement on Fe K-edge is shown in Figure 2.6(a) for FeO. This data has been adapted from [15]. Here µ is plotted as a function of photon energy. The sharp rise in µ around 7120 ev is due to x-ray absorption by 1s

45 2.2 Structural characterization 33 M D E J A > A K > A? H O I J = 5 E? D H = J H > A = 1 H A B A H A? A ' A A A J / A E J D A J A? J H + = > A H H = I = F A Figure 2.5: Schematic of the Daresbury labs EXAFS set-up. The white beam coming in from the synchrotron is monochromated using a double crystal monochromator (see text). The intensity in this mono beam and its corresponding variations are recorded by monitoring I 0. The sample is typically placed at 45 to the beam incidence. The fluorescence signal is detected using a 9-element monolithic Ge detector cooled at liquid nitrogen temperatures. electrons in Fe. As mentioned above, the ensuing oscillations are due to interference of the outgoing photoelectron with neighboring atoms. Generally, the near edge oscillations are known as XANES (x-ray absorption near edge structure) while those more than 30 ev above the edge are called EXAFS (Extended X-ray Absorption Fine Structure). In Figure 2.6(b), the basic aspects to analyze the changes in µ are shown. Here µ 0 is the change in absorption at the edge, while µ 0 (E) describes the background representing absorption from an isolated atom. The fine structure function χ(e) can then be described as [13]: χ(e) = µ(e) µ 0(E) µ 0 (2.2) Given the wave nature of the photoelectron which readily explains the EXAFS effect, converting photon energy to wave vector is a common practice. Using, k = 2m(E E 0 )/ 2 where E 0 is the absorption edge energy and m is electron mass, χ(k) is typically plotted as a function of k. See Figure 2.6(c) for an example. Clearly, the oscillations decay rather quickly with k, and to emphasize these oscillations generally a k 2 or k 3 weighting of χ(k) is used, as shown in Figure 2.6(d). The apparent differences in the oscillation frequencies in Figure 2.6(d) are modelled using the EXAFS equation derived by Sayers, Stern and Lytle [13]: χ(k) = j N j e 2k2 σ 2 j kr 2 j fj (k) sin[2kr j + δ j (k)] (2.3)

46 34 Chapter 2 Probing electronic, magnetic and structural properties Figure 2.6: Example of a EXAFS measurement along with data analysis. (a) Raw data measured on a pure Fe sample. The near edge XANES region and the EXAFS region are shown where absorption [µ(e)] is plotted against energy of monochromated beam. (b) The change in absorption [ µ 0 ] together with a simple spline background [µ 0 (E)] is sketched. (c) The EXAFS fine structure function [χ(e)] is plotted against k. (d) Typically a k 2 or a k 3 weighted χ(e) is plotted to clearly portray the EXAFS oscillations. This figure is adapted from [15] Here, f j (k) and δ j (k) are the scattering properties of the neighboring atoms, N is the coordination number, R is the distance to the neighboring atom, and σ 2 is the disorder in the distance to the neighboring atoms. The basic assumptions in this equation are that the photoelectron scatters elastically and returns back to the original excited atom before the core-hole generated by the absorption event is filled. In real systems, however, inelastic scattering, as well as core-hole lifetimes play a role. So do multiple scattering events. Modern codes used to fit the oscillations, like EXCURV98 used in this thesis (or FIFF7 8), can correct for these issues [16, 17]. A Fourier analysis of the oscillations which contain information about the different distances to the local coordination, then reveals peaks at different frequencies. Such a Fourier analysis will be performed on amorphous CoFeB in Chapter 4. These and other recent theoretical advances in EXAFS are reviewed by [18 20].

47 2.3 In-situ analysis of chemical and electronic properties High-resolution transmission electron microscopy (HRTEM) High-resolution transmission electron microscopy is a technique used for imaging the structural properties of a sample by focusing electrons on it. It is a type of transmission electron microscope (TEM). In a TEM, electrons are usually generated by a process known as thermionic emission or field emission from a filament. The electrons are then accelerated by an electric potential and focused by electrostatic and electromagnetic lenses onto the sample. The beam interacts with the sample due to differences in density or chemistry. The transmitted beam contains information about these differences, and this information is used to form an image of the sample. As opposed to conventional TEM where only the amplitude resulting from this interference is measured, a HRTEM relies on the contrast which arises from the interference in the image plane of the electron wave with itself. In other words, the phase of the electron wave also carries the information about the sample and generates contrast in the image. This phase-contrast arises due to the fact that atoms in a material cause Bragg diffraction of the electrons as they pass through the atoms, changing their relative phases upon transmission through the sample. Thus, HRTEM is also known as phase-contrast imaging. Since image formation relies on this phase-contrast, it may also be considered a weakness of the HRTEM. The image is influenced by strong aberrations of the imaging lenses in the microscope, and therefore the contrast is not intuitively interpretable. Thus, the resolution of the HRTEM is limited by spherical and chromatic aberration of the lenses. However, a new generation of aberration correctors has been able to overcome such spherical aberration and software corrections now allow the production of images with sufficient resolution below 1 Å. As mentioned earlier, in this thesis, HRTEM was performed by Dr. Etienne Snoeck at CEMES- CNRS Toulouse, France. We used a Tecnai F20 FEG microscope (FEI) fitted with a spherical aberration corrector (CEOS) and having a point resolution of 1.3 Å. The measurements are discussed in Chapter 4 and in Chapter In-situ analysis of chemical and electronic properties As mentioned in Section 2.1.2, the EUFORAC has a broad range of in-situ characterization techniques which, together with the deposition chambers, greatly enhance the possibilities for novel research. Below we will discuss two such characterization techniques on the EUFORAC which were extensively used during this thesis X-ray photoelectron spectroscopy (XPS) XPS is a spectroscopic technique which allows qualitative as well as quantitative analysis of the chemical state of atoms present in the top Å of a sample surface. In the following we will briefly discuss the basic mechanisms of XPS, which are also illustrated in Figure 2.7(a). Conceptually similar to XAFS, XPS is also

48 36 Chapter 2 Probing electronic, magnetic and structural properties based on the phenomenon of photoabsorption where the interaction of a photon with an atom allows the photon energy to be absorbed by exciting an electron above its ground state. One measures an XPS spectrum when the sample under investigation is exposed to monochromatic x-ray radiation of a known fixed energy ( ω) and the energy of the ejected photoelectrons is measured using an electron spectrometer. This process is shown with an electron from a 2p 3/2 level in Figure 2.7(a). The energy of the x-ray photons is high enough to excite electrons from the core levels to the vacuum levels. As energy is conserved in the absorption process, the kinetic energy of the photoelectrons ( ω) is equal to the difference between the photon energy (E KE ) and the binding energy (E BE ) of the atomic level. The fact that the binding energies for all elements are known with high accuracy makes the XPS a chemically sensitive technique. The electron spectrometer is capable of sampling the energy of the incoming electrons, and plotting their intensity as function of their energy. Although the penetration depth of the x-rays is of the order of several µm, XPS is very surface sensitive. This surface sensitivity of the XPS originates from the fact that only those photoelectrons which are generated in the topmost Å below the surface are able to escape from the sample. Photoelectrons generated deeper in the sample lose their kinetic energy before they arrive at the surface through inelastic scattering processes. In this thesis, a twin anode x-ray source is used on which Al K α ( ω = ev) and Mg K α ( ω = ev) lines were available. Such a twin anode source is convenient when spectral overlap occurs. In other words, when the Auger line of one element in the sample overlaps with the XPS line of another element. Before measurement, the sample is positioned with its surface perpendicular to spectrometer inlet as shown in Figure 2.7(a). In this situation the photoelectrons emitted in a direction perpendicular to the surface are detected. The surface sensitivity of XPS can be enhanced, i.e., the intensity from photoelectrons generated below the surface can be suppressed, by rotating the sample such that only photoelectrons emitted at a grazing angle from the surface are detected. In this way the effective escape depth can be reduced by a factor 3. A more complete overview of XPS and its capabilities can be found elsewhere [21]. Let us consider two examples where this technique was employed in this thesis. All the CoFeB alloys investigated in this thesis were checked for the chemical stability of the procured alloy targets as well as for their compositions. Figure 2.7(b-d) show XPS spectra for a CoFeB sample where the XPS intensity is plotted as a function of binding energy for that particular element. For the case of Co and Fe in this sample, the L 2,3 (2p 1/2 and 2p 3/2 ) edge has been chosen, while boron spectra have been measured at the K (1s) edge. One notices that the peaks coincide with the known binding energies for these elements, indicating that there is no observable charge transfer in this alloy. Furthermore, the relative area under the curve of these peaks is a good indication of the relative concentrations of the elements in the

49 In-situ analysis of chemical and electronic properties 37 = 2 D J A A? J H I F A? J H I? F O > + > = J? 1 * H I F A? J H A J A H L =? K K A L A : J A I E J O = H > K E J I F F F F!! & % ' % & % % % % % * E C A A H C O A 8 I ' & ' & $ -. 6 K C = > =? C H A B -. F D J I E J A I E J O = H > K E J I - I A? - - I A?. D A 8 H A C E B E J A H A I J -. + " &. A! * # # # "! E A J E? A A H C O A 8 * E C A A H C O A 8 Figure 2.7: Principles of photoelectron spectroscopy. (a) The basic idea of photoelectron spectroscopy is exemplified. (b-d) Representative XPS spectra for the Co and Fe L 2,3 edges and the B K edge in amorphous CoFeB are shown. (d-e) UPS data on CoFeB together with the Tougaard background substraction used. alloy. Taking into account a few empirical parameters (sensitivity factors and escape depths), and the relative intensities of the peaks, one can calculate the composition of the alloy. However, it is clear that since one probes the top 20 Å of the sample, this composition analysis is, in theory, valid only for that part of the film. Another project which was briefly taken up during this thesis was MgO as an alternative barrier [27]. MTJs using this novel crystalline barrier have shown huge TMR ratios (> 200%) at room temperature [22 25]. As we shall see in Section 2.5.3, we too observed large TMR ratios in MTJs using this barrier. In order to get more insight in the chemical properties of the MgO barriers, we used the XPS technique. As mentioned earlier, chemical bonds between atoms result in shifts of the atomic levels from their known binding energies, and therefore, provide the extremely useful opportunity to study the chemical environment of atoms. An example of this chemical characterization is shown in Figure 2.8(a). Here, oxygen

50 38 Chapter 2 Probing electronic, magnetic and structural properties Sing. crys. MgO Sputt. MgO Plasma Ox. Mg (a) (b) single crystal MgO Intensity (arb. units) MgO 2 MgO (c) (d) sputtered MgO plasma oxidized Mg Binding Energy (ev) Binding Energy (ev) Figure 2.8: XPS on MgO. (a) Plot comparing oxygen 1s XPS spectra on MgO single crystal with sputtered MgO and plasma oxidized Mg layers. Dotted lines indicate where oxygen lines for MgO and MgO 2 peaks are expected. In (b) the same XPS peak for the cleaned MgO single crystal is fit with gaussian peaks centered around energies known to be associated with MgO and MgO 2. In (c) and (d), similar curves are shown for sputtered and plasma oxidized Mg layers. By comparing their relative intensities, one may also gain assess to the stoichiometry or the sputtered and oxidized layers. For details, please refer to [27] 1s lines for three different types of MgO samples are analyzed. Spectra from a single crystalline MgO substrate are compared to a plasma oxidized Mg film and a sputtered MgO film from a MgO target. From literature, the oxygen 1s lines for MgO and MgO 2 are expected at ev and ev, respectively [26]. While the spectra for single crystalline MgO substrate is centered around the expected oxygen 1s line for MgO, this is clearly not the case for either the sputtered or the oxidized film. One notes that the plasma oxidized Mg film shows a high intensity in the region where the oxygen 1s line for MgO 2 is expected. In the case of the sputtered MgO film, there are spectral regions which show a dominant MgO peak and a slight shoulder at the MgO 2 binding energy. In Figure 2.8(b), the spectra for the single crystalline MgO substrate are fit with gaussian peaks centered around MgO and MgO 2. Using these spectra as a reference and comparing the area under these gaussians to those for the oxidized and sputtered MgO layers [Figure 2.8(c-d)], one may also gain access to the stoichiometry of these layers. For details, please refer to [27].

51 2.4 Magnetic characterization Ultraviolet photoelectron spectroscopy (UPS) The UPS technique is based on the same principle as discussed for the XPS technique. However, the difference in the energy of the impinging photons, the photon source and the region of the electronic structure which these photons probe distinguish the two techniques. In this thesis, the He-I and He-II lines which have photon energies of ω = ev and ω = 40.8 ev, respectively, were used for performing UPS. Due to this considerably lower photon energy, the UPS the probes valance band structure with better sensitivity and resolution, as compared to XPS. This also allows probing the work functions of materials with better accuracy. An example of a UPS measurement on a CoFeB sample is shown in Figure 2.7(e). Here the UPS intensity is plotted as a function of the kinetic energy of the photoelectrons. At low kinetic energies, one notices a large peak at E = E sec which decays quickly with increasing kinetic energy. This peak is due to secondary electrons which have undergone inelastic scattering processes in the sample. These electrons have low energies and might not reach the electrometer for measurement. However, the measurement of this sharp rise in intensity due to the secondary electrons (E sec, known as secondary electron cut-off) is a key to measure the work function of the material. Therefore, typically a small DC negative bias is applied to the sample to allow the measurement of these electrons. As shown in Figure 2.7(e), the work function (Φ) can the be calculated as Φ = ω (E sec E F ). We will use this method to calculate work functions in Chapter 6 Although the secondary electrons are essential to measure the work function, they may be considered a nuisance in the valance band region of the UPS spectrum. As mentioned before, they arise due to inelastic scattering events of the excited photoelectrons. A common method for eliminating the background arising from secondary electrons is substraction with the Tougaard algorithm [28 30]. This background curve is shown in Figure 2.7(e). Typically, after background substraction the UPS intensity is plotted as a function of the binding energy, with the Fermi level of the material set to zero. This curve is shown in Figure 2.7(f). 2.4 Magnetic characterization Among the most common methods for characterizing magnetic thin films are SQUID (Superconducting Quantum Interference Device) magnetometry and MOKE (Magneto- Optic Kerr Effect) magnetometry. In this thesis, we have regularly used these techniques to measure the magnetic properties of various ferromagnetic films. In addition, we also used magnetic dichroism in x-ray absorption to investigate the changes in spin and orbital moments in CoFeB films as a function of the film composition.

52 40 Chapter 2 Probing electronic, magnetic and structural properties Superconducting quantum interference device (SQUID) The principle of SQUID technique is based on the detection of magnetic flux originating from a sample. The sample is placed in a magnetic field created by a superconducting magnet. The SQUID sensor, is comprised of a superconducting ring interrupted by one or two Josephson junctions. The sample is moved slowly through a detection coil coupled to the SQUID via superconducting wires. The SQUID sensor then measures variations in the persistent current within the superconducting ring and outputs a voltage that is proportional to the magnetic moment of the sample. The SQUID output voltage, when properly calibrated using a sample of known magnetic moment, can be used to provide accurate values for the magnetization of the sample. For the SQUID-magnetometer in our laboratory, a r.f.-type MPMS-55 SQUID from Quantum Design, a sensitivity of 10 7 emu (= Am 2 ) is specified by the manufacturer. The temperature can be varied between T = 1.7 K and 400 K. The maximum attainable magnetic field amounts to 5T Magneto-optical Kerr effect (MOKE) To magnetically characterize our thin films, a home built ex-situ magneto-optical Kerr effect (MOKE) set-up was available. The MOKE relies on magneto-optical effects, the discovery of which dates back to Faraday in 1846 [31]. He observed a rotation of the polarization vector upon transmission of linearly polarized light through a medium in an applied magnetic field. Several years later Kerr found the same effect in a reflected beam of light [32]. Today, one of the most conventional type of magnetometery is MOKE using a laser. A HeNe laser with standard optical lenses allow probing the magnetic thin films locally with typical laser spots sizes around 75µm. The photon energies are typically of the order of a few ev involving excitation of electrons from occupied to unoccupied valance bands. The basic principle of this magneto-optical effect is as follows. Linearly polarized light can be expressed as a superposition of left and right circular polarized in-phase components. On interaction with the exchange split valance bands in a ferromagnet, due to electric dipole selection rules, a phase shift occurs in these components (Kerr rotation) accompanied by a change in the amplitude of these circular components (Kerr ellipticity). We will return to such magneto-optical effects in the soft x-ray regime in Section A schematic of the MOKE set-up used is shown in Figure 2.9. As mentioned earlier, a HeNe laser with optical lenses produce and focus the laser beam. The key components of the measurement set-up is the photo-elastic modulator (PEM). Quite generally, the changes in the polarization induced by changes in the magnetization of magnetic material are of the order of tens of milli-degrees, which implies that a very sensitive detection technique is required. Together with the configuration of the polarizer and analyzer, the PEM allows being sensitive to these small changes in the sample magnetization. The PEM switches the polarization of the in-

53 2.4 Magnetic characterization 41 Figure 2.9: Schematic of the MOKE set-up. A HeNe laser shines on the sample through a polarizer and a photo-elastic modulator (PEM). The reflected beam is picked up be a detector through an analyzer. Locking in to the first and second harmonic of the detector signal gives the ellipticity and rotation, respectively. cident linearly polarized light to alternate between left and right circularly polarized light. Because the right and left circularly polarized light have different magnetic refraction coefficients, as mentioned earlier, this will result in a change of the ellipticity and rotation. Thus, after reflection from the sample, the electronic signal observed at the detector contains information about ellipticity (first harmonic) and rotation (second harmonic) with respect to the frequency and phase of the PEM. As shown in Figure 2.9, the lock-in amplifier is used together with a reference from the PEM modulator to access these harmonics. On the other hand, the DC signal from the detector is a good measure of the laser power. A more comprehensive quantitative analysis of the MOKE technique and a detailed account can be found in references [33, 34]. Let us now have a look at the capabilities of the MOKE technique in characterizing magnetic thin films used in this thesis. We will adopt exchange biased CoFeB layers as an example. The layer stack consists of Si / SiO x // Cu / IrMn (0 300 Å) / CoFeB (50 and 60 Å) layers annealed in a field of 200 ka/m at 260 C [see Figure 2.10(a)]. The IrMn layer is grown as a wedge according to the description in Section Effectively, exchange bias means that the IrMn layer induces a unidirectional anisotropy in the CoFeB layer which prohibits the CoFeB layer to follow the direction of the external magnetic field until a certain threshold field is

54 42 Chapter 2 Probing electronic, magnetic and structural properties Exchange bias (ka/m) Cu CoFeB IrMn Cu (a) (c) 30 CoFeB thickness Å 50 Å IrMn Thickness (Å) H bias Magnetic field (ka/m) 20 Cu IrMn thickness 50 Å CoFeB thickness 50 Å (b) (d) IrMn Thickness (Å) MOKE signal (arb. units) Coercivity (ka/m) Figure 2.10: Capabilities of the MOKE technique. (a) An example of an exchange biased CoFeB layer is shown. The sample is Cu/IrMn (0 300 Å)/CoFeB (50 and 60 Å). (b) An example of a M-H loop measured with the MOKE set-up is shown. The shift in the MOKE loop from zero field denotes the exchange bias field. The exchange bias field is shown as a function of IrMn thickness in (c), while the changes in coercivity resulting from the exchange biasing are shown in (d). applied. This can be seen in Figure 2.10(b) which shows an example of the M-H loop measured on the sample with the MOKE set-up. An external field lower than 50 ka/m is not able to reverse the magnetization of the CoFeB layer, effectively shifting its M-H loop away from zero. The magnitude of the shift in the M-H loop with respect to zero field is called the exchange bias field. As we have seen in the the introduction [see Section 1], such an exchange biased layer is used in MTJs as a reference layer. The other ferromagnetic layer is used as a free layer with a hysteresis loop centered around zero. Since the loops of the two layers are centered around two different external fields (zero and H bias ), there are regions where the magnetic field aligns the two layers parallel or antiparallel to each other. This is one of the ways in which magnetically engineered ferromagnetic layers allow the observation of TMR in MTJs. One of the desirable properties in such MTJs is to achieve high exchange bias

55 2.4 Magnetic characterization 43 fields. Considering the capabilities of the deposition technique and the MOKE setup, the well-known procedure for studying exchange biased magnetic multilayers is to grow wedges of the layers to be studied and exploit the MOKE technique for performing magnetization measurements locally. Figure 2.10(c) shows such a measurement on the sample shown in Figure 2.10(a) where the thickness of the IrMn anti-ferromagnetic layer is varied. In Figure 2.10(c), we plot the measured exchange bias field, while in Figure 2.10(d), the measured coercivity of the CoFeB layer is plotted. Note that these two quantities are extracted from the M-H loops similar to that in Figure 2.10(b) measured for each data point show in Figure 2.10(c-d). Let s address the behavior of the exchange bias of this sample first. One notices the onset of an exchange bias field around 23 Å IrMn thickness for both the 40 and 50 Å thick CoFeB layers. While the exchange bias peaks around 43 Å IrMn for both CoFeB thicknesses, the thinner CoFeB layer shows a higher H bias. This is due to the known fact that H bias 1, where t is the CoFeB thickness [35]. However, the t behavior of the curve after the peak in H bias is not completely understood. Initially, the curve seems to fall off almost exponentially with increasing IrMn thickness, and then levels off for IrMn thicker than 100 Å. One may guess that the morphology of the IrMn layer which may be expected to change in the thin part of the wedge may be the reason behind this behavior. For example, a conjecture would be changing grain size of IrMn in this thin part of the wedge. From the perspective of the random field model, this would lead to an increased interface area and/or a increase in number of uncompensated spins at the interface with the CoFeB layers would increase. This would consequently increase H bias in this region. Such a increased H bias has also been observed in Co / IrMn layers [36]. By fitting the MOKE loops, the coercivity of the CoFeB layers can also be analyzed, as plotted in Figure 2.10(d). One observes a sharp peak in coercivity around 23 Å, corresponding to the position of the sharp rise in H bias [see Figure 2.10(c)]. This sharp increase in coercivity might be related to the increased pinning centers at the onset of exchange bias where there are regions in the film which are and are not exchange biased. As the IrMn thickness increase, however, one observes a dip in the coercivity and then it eventually levels off. However, further experiments need to be performed to address the origin of both these aspects regarding the behavior of coercivity we observe here. However, the quantities displayed in Figure 2.10(c and d), clearly demonstrate the power of the MOKE technique to probe magnetic thin films locally. These measurements also show that quasi-amorphous CoFeB layers can be exchange biased exhibiting large bias fields (H bias 50 ka/m). We will return to the magnetic properties of CoFeB layers later Magnetic circular dichroism (XMCD) in x-ray absorption (XAS) The XMCD technique is based on the changes in the absorption cross section for circularly polarized x-rays depending on the properties of the absorbing materials.

56 44 Chapter 2 Probing electronic, magnetic and structural properties Just like the MOKE, it is classified as a magneto-optical technique which relates the optical and spectroscopic properties to the magnetic state of a given system. However, it has two unique attributes which make it a powerful spectroscopic technique, distinguishing it from other common magnetic characterization methods like SQUID, MOKE or vibrating sample magnetometer (VSM). Firstly, it is element specific. Secondly, it allows a direct and independent extraction of the spin and orbital moments. We will discuss measurements employing this technique in Chapter 6. The first prediction of magneto-optical effects in XAS using circularly polarized light, i.e., XMCD, was made in 1975 by Erskine and Stern for the M 2,3 edge of Ni [37]. It took several years for the first proof of XMCD measured at the K edge of Fe by Schütz et al. in 1987 [38]. The general theory of the XMCD effect was later developed by Thole et al. [39] and Carra et al. [40]. This theory allows direct quantitative measurements of the orbital and spin moments, the direct confirmation of which was given by Chen et al. [41]. For an excellent review of this technique, please refer to [42, 43]. Next we will briefly summarize the essential aspects of the technique, and spend a few words on data evaluation. Figure 2.4 shows a sketch with the basic principles of the XMCD technique. In the simplest approximation which is widely used, XMCD can be viewed as a twostep process [42, 43]. In the first step a photon of helicity, ±1, transfers its angular momentum along the direction of the wave vector to the orbiting electron of the absorbing atom. Angular momentum conservation is the most important principle here. Since the spin of the electron cannot directly interact with the photon electric field, in the absence of spin-orbit coupling the photon can only transfer angular momentum to the orbital part of the wave function. Such is the case with excitation from core s-states or if one sums over L 2,3 (2p 1/2 and 2p 3/2 ) edges. If the core state is split by spin-orbit interaction, the sub-states are no longer pure spin states. As a result, the photon angular momentum is transferred to both the orbital and spin degrees of freedom of the excited photoelectron. In fact, a relatively large portion can be transferred to the spin generating spin-polarized photoelectrons. The magnetic properties of the absorbing solid enter in the second step, where the valance band acts as a detector for the spin and/or the orbital momentum of the excited photoelectron. If the metal is ferromagnetic, there is an inherent imbalance between the number of spin-up and spin-down electrons at the E F, and hence the valance band acts as a spin-detector. In the same way, if there exists an imbalance between occupied states of magnetic quantum number m l, the valance band acts as an orbital detector. Such is the case when the valance band is also spin-orbit split. The transition rate into the unoccupied valance states depends on the number of unavailable states with spin parallel to the photoelectron spin. One can show that XMCD is proportional to the spin-polarization of the unoccupied DOS at E F. If one defines p ± and p ± as the relative weights of the spin-up and spin-down photoelectron polarization, and d (E) and d (E) as the unoccupied DOS of the valance shells, then using Fermi s golden rule one may write the absorption cross-section for ν = ±1

57 2.4 Magnetic characterization 45 [photon helicity parallel (+) or antiparallel ( ) to the preferred axis] as [44] Γ + (E) = Φ[p +d (E) + p +d (E)] (2.4) Γ (E) = Φ[p d (E) + p d (E)] (2.5) Here Φ is a constant of proportionality. Since reversing the photon helicity changes the sign, but not the magnitude of the photoelectron polarization, the above expressions simplify since Eliminating Φ, one may write: Γ(E) Γ(E) p + = p = p (2.6) p + = p = p (2.7) = Γ + Γ = p p Γ + + Γ p + p ( ) d (E) d (E) d (E) + d (E) (2.8) In Equation 2.8, the first term on the right hand side can be written as, the photoelectron polarization P e given by while the second term on the right P e = p p p + p (2.9) P unocc = d (E) d (E) d (E) + d (E) (2.10) is the degree of the spin-polarization of the unoccupied DOS at E > E F. This equation suggests that, XMCD is just proportional to the spin-polarization of the unoccupied DOS at E F, where the constant of proportionality is the photoelectron polarization, P e. Note here that we have assumed P e to be independent of photon energy and the nature of the final state. Measurement set-up Figure 2.11 shows the measurement set-up at station 5U.1 of the Daresbury labs, UK. The basic set-up is similar to that discussed for the XAFS measurements in Section The x-rays generated in the synchrotron pass through a monochromator. Subsequently, their intensity is measured in the I 0 chamber. When they impinge on the sample, the photoabsorption is measured with the total electron yield method. This basically means that a nano-ampmeter is connected between the sample and ground, and the electrons which flow from the ground to the sample due to the removal of the photoelectrons is measured. The sample is placed within a vacuum chamber between octo-pole magnets which allow the application of a magnetic field to saturate the sample.

58 46 Chapter 2 Probing electronic, magnetic and structural properties Translation and Rotation Sample holder Sample current Sample bias UHV Low vacuum I0 monitor gold grid I0 current Bias volatage Polarized x-ray Ion pump Turbo pump Differential pumping section View ports Water-cooled electromagnet Turbo pump Sample Chamber Figure 2.11: Schematic of a XMCD set-up at the 5U.1 beamline at Daresbury labs. The beam travels through an undulator and a monchromator before it reaches the I 0 chamber and impinges on the sample thereafter. An octo-pole magnet allows the application of a magnetic field in any direction radial to the sample plane. Adapted from [45]. Data evaluation Figure 2.12(a-f) shows examples of XAS and XMCD measurements on pure Co and Fe films. The top panel shows isotropic XAS spectra for Co and Fe at the L 2,3 edge, where the absorption intensity measured with the total electron yield method is plotted as a function of photon energy. Notice that a background intensity exists between the beginning and the end of the spectra, for example in Fe at 700 ev and 735 ev. This background is subtracted using a simple function that produces two steps, one at the L 3 edge and the other at the L 2 edges. The A 3 and A 2 areas are then calculated as the area under the L 3 and the L 2 edge. Since the effect of linear dichroism is considered to be at least one to two orders of magnitude lower in transition metal compounds, these isotropic A 3 and A 2 values are used for the application of the sum rules. This implies that the isotropic spectra are the average of the XAS spectra measured for left and right circular polarized light. These individual spectra for left and right circularly polarized light are shown in Figure 2.12(c-d). If one takes the difference between these XAS spectra for left and right circularly polarized light, then the resulting spectra are called XMCD, and are

59 2.4 Magnetic characterization 47 Figure 2.12: Background substraction of XMCD data. Example for pure Fe and Co. (a-b) Isotropic XAS on Fe and Co edge at the L 2,3 edge. The integrated areas A 2,3 denoted by are used to evaluate the spin and orbital moment using the sum rules. These areas are calculated after subtracting the shown two-step background. (c-d) XAS spectra for left and right circularly polarized light (Γ ± ). (e-f) XMCD spectra which are the difference between the left and right circularly polarized light XAS shown in (c-d). Here the areas A 2,3 are also used to calculate the spin and orbital moment using the sum rules. A 3 and A 2 are denoted by.

60 48 Chapter 2 Probing electronic, magnetic and structural properties plotted in Figure 2.12(e-f). Here the L 3 edge is seen to have the opposite sign in intensity in comparison to the L 2 edge. This is due to the fact that the L 3 and L 2 edges are transitions from the 2p 1/2 and 2p 3/2 states where the spin and the orbital moments are aligned parallel or antiparallel to each other. The integrated intensity under the XMCD spectra for the L 3 and L 2 edge is named as A 3 and A 2 is then used in the sum rules discussed below. Sum rules As noted earlier, the XMCD sum rules allow quantitative determination of the spin and orbital moments. We will state these sum rules next and give the important aspects they assume or overlook for the sake of simplicity. We will employ these sum rules in Chapter 6 of this thesis. The spin sum rule is given by [40]: m s n 3d = 2 A 3 4 A 2 A 2 + A 3 7 T z n 3d (2.11) As shown in Figure 2.12(a-f), the integrated areas under the L 2,3 edges of isotropic XAS spectra are used to extract A 2,3, while the corresponding areas under the XMCD spectra are used to extract A 2,3. n 3d denotes the number of d-holes, which are unknown in the case of CoFeB. The magnetic dipole term ( T z ) in Equation 2.11 refers to the asphericity of the spin magnetization. In other words, the expectation value of T z is non-zero when there is an anisotropy in the field of the spins due to distortion of the atomic cloud, i.e., when the spin density within the atom is not spherically symmetric. This can happen since the spins are associated with electrons which may be anisotropically distributed in space due to local chemical bonds, for example. However, for the case of an amorphous system, G. van der Laan and co-workers have argued that this term can be neglected as its local contributions are expected to cancel out [46]. According to Thole et al., m o is given by the orbital sum rule [39] Approximations in sum rules m o = 4 A 3 + A 2 (2.12) n 3d 3 A 3 + A 2 There are several approximations made in deriving the sum rules. To mention a few. Transitions from 2p to 4s electrons are neglected. Exchange splitting of the core-levels is not considered.

61 2.5 Measuring electronic transport 49 The difference between the d 3/2 and d 1/2 wave functions is ignored Energy dependence of any wave function is ignored. Any asymmetric spin (charge) distribution of the core levels is ignored. We will try to put these approximations briefly into perspective. The first approximation describes the neglect of the 4s electron DOS above the Fermi level. Though much smaller than the d-dos above the Fermi level, it is known to be spin polarized [see Figure 1.2]. Another important aspect not considered by the sum rules is that the spin-up and spin-down DOS of the s and the p electrons is also exchange split depending on the positions of these levels in the electronic structure. For example, in Figure 1.2, Section 1.2, one clearly sees that the s-dos shows a significant exchange splitting from electron energies 6 ev below the Fermi level. Similar observations can be made for the p electrons. For a detailed discussion on all these approximations, please see the review of Ebert et al. [43]. 2.5 Measuring electronic transport In this section, we will outline three different techniques which were used during this thesis. The superconduction tunneling spectroscopy technique was used to measure the TSP, the MR set-up was used to measure magnetoresistance and IETS spectra, and the current in-plane tunneling set-up was used to locally probe novel tunnel barrier material. The IETS set-up was developed together by Fransisco Bloom and the author, while the CIPT technique was set-up was developed together by Roeland Huijink and the author [47] Superconduction tunneling spectroscopy (STS) Pioneered by Meservey and Tedrow, the use of superconduction tunneling spectroscopy is arguably the most reliable method in exploring spin tunneling in spintronic devices. Although this technique is extensively employed in this thesis, we will not go into great detail about the various aspects it involves. Instead the reader is referred to excellent reviews, an extended and involved review by Meservey and Tedrow themselves [48], and the other authored by Corné Kant [1] who also set-up this technique at the group Physics of Nanostructures in Eindhoven. The basis of this technique is the pioneering work of Giaever [49, 50] and later Shapiro [51] who demonstrated that electron tunneling in devices with one or two superconducting electrodes separated by a thin barrier allowed the mapping of the BCS superconducting DOS. This observation is exemplified for a SC / I / NM junction in Figure 2.13(a-b). SC denotes the superconductor, I is the insulator and NM and FM denote a normal or ferromagnetic metal, respectively. The I V characteristic of such a junction below the superconducting transition temperature

62 50 Chapter 2 Probing electronic, magnetic and structural properties Current (a) - T = 0 T = 0.1 T c (b) - T = 0 T = 0.1 T c Norm. conductance (arb. units) Norm. conductance (arb. units) (c) 2 B B P = +40 % B = 2.0 T Bias Voltage (mv) (d) Norm. conductance (arb. units) Figure 2.13: Plots showing basics of the STS technique. (a) Representative I V of a SC/I/NM junction below the superconducting transition temperature of the SC. (b) di dv of the same junction which clearly shows a gap of 2. Dotted lines in (a-b) represent measurements at 0 K. (c) The application of an external field of B = 2 T Zeeman splits the spin-up and spin-down electrons by an energy difference of 2µ B B. (d) If the NM is replaced by a FM, the inherent inequality of spin-up and spin-down electrons in the FM results in an asymmetry in the di dv. The degree of this asymmetry is a measure of the TSP of the ferromagnet. of the SC is shown in Figure 2.13(a). Note that application of a bias below a certain threshold voltage ( ), does not result in any current through the device. The threshold voltage represents a gap,, in the DOS of the SC. This gap in the DOS can be confirmed by measuring the di of such a tunnel junction. As shown dv in Figure 2.13(b), the di at temperature T = 0 K strongly resembles the BCS DOS dv predicted for a type-i superconductor [52]. The second important discovery that led to the development of the technique was made by Meservey, Tedrow and Fulde who reported that applying a field in the plane of the superconductor allowed the Zeeman-splitting of the quasiparticle DOS

63 2.5 Measuring electronic transport 51 Figure 2.14: Experimental set-up used for the STS measurements. The essential parts are a lock-in amplifier and a home-built bias voltage controller [1]. A current to voltage convertor mounted close to the junctions allows to measure the current through the junctions. in a superconductor [53]. As shown in Figure 2.13(c), this meant the separation of spin-up and spin-down electrons in the SC around the Fermi level. Generally, these spin-up and spin-down electrons occupied degenerate energy states. However, the application of a large magnetic field in the plane of the superconductor added enough energy to these individual spin systems allowing raising the energy of electrons of one spin type with respect to electrons of the other spin type. This splitting is called the Zeeman splitting and scales with 2µ B B, where B is the magnitude of the external magnetic field. All these measurements discussed above were done with the second electrode being either a SC or a NM. In 1971, Tedrow and Meservey reported the first measurements where they used a FM as the second electrode [54]. Due to the inherent imbalance in the number of spin-up and spin-down electrons at the Fermi level of a FM, one may readily imagine that it sinks and sources an unequal number of electrons of either type. As shown in Figure 2.13(d), this results in an asymmetry in the di. The degree of this asymmetry is a measure of the tunneling spin polarization (TSP) of the ferromagnet. Here it is appropriate to distinguish between the dv spin-polarization of the FM with its tunneling spin polarization. The former, the spin polarization of all the electrons at the Fermi level of a 3d transition metal FM

64 52 Chapter 2 Probing electronic, magnetic and structural properties can be written as P total = N (E F ) N (E F ) N (E F ) + N (E F ) (2.13) To their surprise, Tedrow and Meservey measured a positive spin polarization for the 3d transition metal FM they used, viz., Ni, while it was known that the dominant electrons at the Fermi level of this alloy (minority d-electrons) would lead to a negative P total. Later it was suggested by Hertz and Aoi (1973) [55] and by Sterns (1977) [56] that although, the dominant species of electrons at the Fermi level of transition metal ferromagnets were spin-down d electrons, they did not couple well with the states over the barrier. Instead, highly dispersive s-like electrons had a much larger overlap integral with states in the barrier which led to a larger transmission probability for these electrons. Interestingly, these s-like electrons also are highly spin polarized due to s d hybridization which also causes spin-up s-like electrons to be the dominant species at the Fermi level in comparison to spin-down s-like electrons. Therefore, as a first order approximation, it is not completely unreasonable to define the TSP due to these s-electrons as TSP expt TSP s = N s (E F ) N s (E F ) N s (E F ) + N s (E F ) (2.14) Recall that we have already commented on this equation in Chapter 1 and we shall return to the discussion in Chapter 4. Measurement and extraction of the TSP Let us now proceed towards measuring the TSP in real devices and extracting the TSP from these measurements. Firstly, one should realize that the Zeeman splitting (2µ B B) which is essential to measure the TSP is limited by the critical magnetic field that can be applied to the SC. This critical field strongly depends on the thickness of the Al film and is typically around B c 4 Tesla for Al films thinner than 100 Å [57]. Moreover, to resolve the sharp peaks in the SC DOS shown in Figure 2.13, 2µ B B k B T. With µ B 60µeV/T and with k B 86µeV/K, it becomes clear that a temperature below 1 K is necessary to resolve the Zeeman splitting. Therefore, a conventional cryostat cannot be used to measure the TSP accurately. In this thesis, a 3 He sorption-pumped refrigerator from Oxford Cryogenics called Heliox VL was used which employs a rare 3 He isotope instead of the common 4 He as a coolant. The reduction of the vapor pressure above liquid 3 He, allows cooling down to 0.24 K. The cryostat is also equipped with an 8 Tesla superconducting magnet for applying a field, typically 2 3 Tesla, in the plane of the tunnel junctions. Regarding

65 2.5 Measuring electronic transport 53 Normalized Conductance (di/dv) (a) B=0T Fit P: 30.0 % T: 0.36K : 0.35 mev b: : Co 2 MnSi B=2T Fit Bias voltage (mv) (b) Figure 2.15: Example of an STS measurement on Al/AlO x /Co 2 MnSi. (a) The zero field curve ( ) shows the superconducting DOS of Al. (b) The application of a magnetic field (µ 0 H > 2.0 T) results in the Zeeman-splitting of the superconducting DOS which acts as a spin analyzer for the tunneling electrons. The observed asymmetry in the intensity of the measured peaks ( ) when fit (solid lines) with Maki theory [60] reveals the TSP of Co 2 MnSi. the measurement set-up itself, Figure 2.14 sketches the essential ingredients. A conventional AC technique is used with a lock-in amplifier and home-built voltage source. This home-built voltage source uses a feedback loop which ensures the correct bias voltage is applied to the junction by comparing the actual voltage at the junction and the voltage requested by the computer [1]. Typically an AC modulation of 10 µv pp is applied to the junction. The constraint of this small AC modulation comes from the fact that the resolution of the smallest features in the spectra are limited by the broadening of the Fermi level. This broadening scales with 3.5k B T (around 300 µev/k) which implies a modulation of 10 µv pp at 0.3 K. One final comment on the measurements: the alignment of the magnetic field in the plane of the SC is crucial to perform a good measurement. Till this point, we have not discussed which material is used for the superconducting electrode, and for the barrier. For reasons that will become clear shortly, Al is used as a SC. The fact that it has a low atomic number and the fact that one is able to grow a pure, continuous ultrathin Al film with a high critical field and high critical temperature hint to its suitability. Moreover, it allows the formation of a closed AlO x tunnel barrier by means of oxidation techniques. Figure 2.15 shows a representative example of an STS measurement on an Al/AlO x /FM layer. In this case the ferromagnetic layer is a Heusler alloy of Co 2 MnSi which is predicted to have a TSP of 100%. These layers were deposited in the group of Prof. T.

66 54 Chapter 2 Probing electronic, magnetic and structural properties Miyazaki at Tohoku, Japan who have demonstrated large TMR with such Heusler alloys [58, 59]. A collaboration was initiated by the author to measure the TSP of these alloys. These results are the unpublished data from this collaboration. In Figure 2.15(a), no field has been applied in the plane of the junction. The normalized conductance ( di ) of such a junction ( ) resembles the SC DOS, just as dv shown in Figure 2.13(b). The application of a 2 Tesla field results in the Zeeman splitting of the SC DOS and the conductance curve shows asymmetric peaks ( ), just as shown in Figure 2.13(d). This asymmetry indicates that the tunneling electrons are spin polarized. To extract the TSP from this curve, Tedrow and Meservey initially used the difference in the intensity of the peaks. However, it is known that this alone does not account for some of the important parameters which influence the SC DOS which in turn introduces an error in the calculated TSP. Immediately after their initial results, Tedrow and Meservey used the Maki-Fulde theory for analyzing their results. The details of this theory can be found in here [60, 61]. In this thesis, we used a fitting program which took into account the Maki-Fulde theory, hereafter called Maki theory. Next we will have a brief look why such a theory is required and what are the additional parameters it takes into account. Firstly, a SC can be described by two separate but interacting electron systems: the Cooper pairs (also called the condensate) and the unpaired electrons (also called the quasi-particles) [62, 63]. These two electron systems interact with each other, and this interaction is called orbital depairing. Depairing arises from magnetic fields in a superconductor and, as the name suggests, allows cooper pair breaking which is detrimental for the SC. Its strength is measured by a dimensionless parameter ξ. The origin of depairing can be magnetic impurities, can be magnetic field generated by a flowing current, or, most relevant for our purpose, the applied external magnetic field [62, 63]. The action of the field is to induce an orbital motion of the electrons by the Lorentz force. This breaks up the pairs since the orbital motion is incompatible with the symmetry requirements for the Cooper pairs. The second important parameter to be considered is the spin-orbit interaction which is an interaction between the unpaired electrons themselves and allows the mixing of spin-up and spin-down quasiparticles. The electric field felt by the electrons travelling in a sea of positively charged nuclei also results in a magnetic field. The spin-orbit interaction is the interaction between the magnetic field of the electron itself and that generated by the sea of positively charged nuclei it traverses. At a scattering event, which is generally accompanied by a small spin-flip probability, the spin-orbit interaction enhances this probability depending on its own strength. This too is detrimental to the SC since Cooper pairs feel these scattering events indirectly. The strength of the spin-orbit interaction is given by another dimensionless parameter b. If one expresses the spin-flip scattering rate as 1 τ so Z 4 1 τ m where τ so and τ m are the spin-flip and momentum scattering rates, respectively, then the spin-orbit parameter can be written as b = 1 3 τ so. Now it is clear that a low atomic number (Z) element is necessary to achieve low spin-flip scattering rates in the SC.

67 2.5 Measuring electronic transport 55 As mentioned earlier, Al with Z = 13 satisfies this criterion well. Returning back to the measurements shown in Figure 2.15(a-b), we see that the lines in these figures represent fits which include the SC gap, the depairing ξ, the spin-orbit scattering b, the temperature T, the applied magnetic field B, and the tunneling spin polarization P as fit parameters. During a fit, typically only the applied magnetic field B is fixed. For this particular sample with Co 2 MnSi as the ferromagnetic electrode, the relevant fit parameters are listed in the figure. The TSP of Co 2 MnSi for this sample was found to be 30%, significantly lower than the expected 100% for a Heusler alloy, indicating that further optimization of the alloy deposition conditions was necessary Inelastic electron tunneling spectroscopy (IETS) IETS is a tunneling spectroscopy technique which probes vibrations and other low energy excitations triggered by electrons which scatter inelastically during tunneling. The technique was pioneered by Lambe and Jaklevic [64] from Ford motors, who also are known for their contribution to the invention of the DC SQUID. It was originally employed and currently used heavily for studying vibrational spectra of molecules. The technique is considered to be extremely powerful over other optical vibrational spectroscopies (like Raman or infrared spectroscopy), primarily in terms of its sensitivity: only molecules are required for obtaining spectra. In comparison, typically atoms per cm 2 required to form a monolayer of material. Later, however, the technique was employed by Tsui et al. to study magnons excited by inelastically tunneling electrons [65]. We will employ this technique in Chapter 5. The technique relies on the fact that a vibrational mode having an energy hν, may be excited by a tunneling electron at a bias voltage V, if the energy of this electron, ev, is equal to hν. This implies that the vibrational mode opens up an additional conductance channel for the electron to inelastically tunnel into. In other words, the I V characteristic of the junction shows a kink at the bias voltage V. This is shown in Figure 2.16(a). If one then measures the di, one could expect dv a step increase in the conductance, as sketched in Figure 2.16(b). The d2 I would dv 2 then show a peak. We will use this technique in Chapter 5 to show that the IETS can also be used to probe structural changes at the barrier-electrode interface. For those IETS measurements, we used a standard lock-in technique which is shown in Figure A 3 mv/711 Hz AC signal from a lock-in amplifier on top of a stable, high resolution DC voltage is used to generate a DC bias sweep accompanied by a AC modulation. The tunnel junction has a non-linear I V, and therefore, the AC modulation across the junction does not inherently remain constant. Since the AC modulation voltage influences the energy resolution of the IETS technique [66 68], = (1.7V AC ) 2 + (5.4k B T/e) 2 (2.15)

68 56 Chapter 2 Probing electronic, magnetic and structural properties Current, I (A) (a) (b) (c) I di/dv (arb. units) di dv d 2 I/dV 2 (arb. units) 2 d I 2 dv Bias voltage (V) Bias voltage (V) Bias voltage (V) Figure 2.16: IETS illustration (a) Schematic representation of an I V curve where (b) shows the resulting di dv, and (c) shows the resulting d2 I. dv 2 it is imperative to maintain it constant as a function of applied DC bias over the junction. According to Equation 2.15, at 4.2 K and 3 mv AC, this IETS energy resolution turns out to be 5.1 mv. In order to maintain this constant energy resolution, we used a feedback between the lock-in amplifier placed across the junction and that used to source the AC modulation voltage. The rest of the set-up was so designed di that the I V, and d2 I could be measured simultaneously by locking into the dv dv 2 first and the second harmonic across a precision metallic resistor (R Series ) in series with the junction. The DC current through the junction (I junc = V DMM1 R Series ) and the resistance of the junction (R junc = V DMM2 I junc ) were measured using two high-precision digital multimeters (DMM1,2)., 1 K? 8, + I K H? A 8, + 8, + 8 ) A H E A I, 4 K? 8 ) +? E? E? @ A B B H A G 4 A B B H A G + I J = J 8 ) Figure 2.17: IETS instrumentation The block diagram of the IETS set-up used. Two lock-in amplifiers and two digital multimeters enable the simultaneous di measurement of I V, dv, and d2 I. dv 2

69 2.5 Measuring electronic transport Magnetoresistance (MR) The MR measurements were performed on the same set-up where the IETS measurements were also performed. Typically, a small DC bias voltage (<10 mv) was applied to the junction, and the external magnetic field was applied to measure MR. This set-up is equipped with a standard 4 He flow cryostat from Oxford Cryogenics and magnets which can reach fields in excess of 0.5 Tesla. The other aspects and components of the measurement are identical to those described for the IETS set-up in Figure An example of a TMR measurement using this set-up on Co / AlO x / Co MTJ was shown in Figure 1.1. As mentioned earlier [see Section 2.8], MgO barriers were investigated as novel barrier materials for MTJs. Here we present the most successful results obtained with MgO barriers in CoFeB / MgO / CoFeB junctions. The sample consists of Ta (50 Å) / CoFeB (150 Å) / MgO (35 Å) / CoFeB (50 Å) / Ta (50 Å) layers annealed at 375 C. It exhibits a TMR of 90% at room temperature and 150% at 5 K. One notices that the coercivity of the thinner CoFeB layers shows a larger temperature dependence. More importantly, comparing the two measurements at different temperatures, the resistance in the parallel alignment shows a much lower temperature dependence than that in the anti-parallel alignment. This is a characteristic feature of MgO based MTJs [22]. Although, this sample showed a TMR of 90% at room temperature, this value is relatively small compared to recent values which have reported TMR above 400% at room temperature. Such a measurement was shown in the introduction, see Section 1 for details. We will briefly comment on the possible origin of this lower TMR in Section Current in-plane tunneling (CIPT) A novel way of measuring the properties of a tunnel junction is known as current in-plane tunneling (CIPT) [47, 69]. The key advantage of this method is that, by nature, the CIPT technique is a local probe. This means that if an essential property of a junction (for example, barrier thickness) is gradually changed over the sample, its effects can be monitored by conducting measurements at different spots on the same sample. In other words, painful deposition and characterization of a whole set of junctions can be replaced by a single stack. The following paragraph briefly introduces the technique itself and then we move on to a number of new measurements with this technique. Imagine four minute micron-sized probes are placed collinearly on a planar tunnel junction, and a current is sent through the sample via two of them. By measuring the induced voltage drop between the other two probes, the sheet resistance of the sample can be determined. A part of the injected current may tunnel through the barrier and flow through the bottom electrode. The distance between the probes is of crucial importance here: if the probes are very close together, all of the current flows through the top electrode, and at very large probe distances, the current is

70 58 Chapter 2 Probing electronic, magnetic and structural properties 80k Resistance (k ) 60k 40k 5 K 300 K 150% 90% 20k Magnetic field [ka/m] Figure 2.18: Example of a MgO based magnetic tunnel junction. The resistance versus applied field curve for a Ta (50 Å) / CoFeB (150 Å) / MgO (35 Å) / CoFeB (50 Å) / Ta (50 Å) junction. At a temperature of 300 K, we measure a TMR of 90% and at 5 K, a TMR of 150%. For details, please refer to [27] proportionally divided between the top and bottom electrode. In between there is a transition area in which, by measuring the sheet resistance of a tunnel junction for different probe spacings, the MTJ can be characterized. A new set-up for performing the CIPT measurement was built to probe the (magnetic field dependent) sheet resistance in a MTJ [47, 69]. This work was a part of the Master thesis of Roeland Huijink who worked together with the author on this technique. A commercial 12-finger probe was landed on a sample placed in air by using a piezo actuator (from AttoCube). The probe shown in Figure 2.19(b) was procured from Capres inc. A standard lock-in measurement, as described in Section was performed. CIPT on non-magnetic MgO based tunnel junctions To exemplify the method, we will look at a tunnel junction with non-magnetic electrodes and a MgO barrier. A wedge shaped MgO barrier is grown and the complete stack is unannealed Pt (300 Å) / MgO (0 100 Å) / Pt (50 Å). Note that MgO is grown as a wedge. The thickness of the Pt electrodes is chosen in such a way that a good contrast between the resistances of the top and the bottom electrode can be obtained. A 12-finger probe was landed at different MgO thicknesses on the wedge, and at each spot the local sheet resistance of the sample was determined. By carefully choosing the probes through which the current is sent and the voltage

71 2.5 Measuring electronic transport 59 is measured, the mean probe spacing was varied. In Figure 2.19(a), the data points represent the sheet resistance measured as a function of mean probe spacing for different MgO thicknesses. The lines are fits to the data using CIPT theory [47, 69]. We will summarize a few observations to be made in this figure below: At small mean probe spacings, for MgO thickness of Å, the sheet resistance of the sample is high. Actually, here we measure the resistance of the top electrode since the small distance between the probes allow the current to shunt over the top electrode. At the other extreme, i.e., at very large probe spacings, the resistance falls off to a low value. Here, the current is allowed to spread over a large area, and can use this large area to tunnel. One therefore effectively measures the parallel resistance of both top and bottom electrodes. For the MgO thickness of 24 Å, the measured resistance is very low (equal to the parallel resistance of both electrodes) for all probe spacings. Here, the barrier resistance is too low to measure. On the other hand, for the MgO thickness of 73 Å, the measured resistance stays high (close to the value measured for the top electrode), indicating a very high barrier resistance. Between these extreme values of MgO thickness, the sheet resistance of the sample increases for increasing MgO thickness, because a smaller and smaller part of the current is able to tunnel through the barrier. These results, along with the fits to theory, clearly demonstrate that the current tunnels for large probe spacings or thin barriers. Moreover, the technique allows the observation that the MgO barrier is not perfect, as discussed next. Figure 2.19(b) shows a second experiment on the same sample. Here the measured sheet resistance is plotted as a function of barrier thickness. The probe spacing was kept fixed at 25 µm. A low resistance is observed when the MgO layer is thin, which crosses over to a high resistance when the barrier is thicker. Again this, together with the data from Figure 2.19(a), shows that with decreasing barrier thickness the current increasingly tunnels through the barrier. However, one would expect that the resistance would increase with barrier thickness in a monotonous way. This is clearly not the case here, as a dip in the resistance is observed around 65 Å of MgO. Additionally, the current seems to tunnel at MgO thicknesses around Å, which is much too high. Unlike other materials we sputter, we obtain a very low deposition rate with our deposition parameters for MgO. Not surprisingly, the calibration of the deposition rate of MgO seems to be incorrect. This poor barrier quality, combined with the knowledge of having slightly over oxidized MgO from our XPS measurements 2.3.1, may also explain the irreproducible TMR we obtained in MgO based MTJs, see Therefore, to observe a TMR with this technique, we reverted to an AlO x barrier,

72 60 Chapter 2 Probing electronic, magnetic and structural properties Sheet resistance ( / ) (a) 73 Å 56 Å 51 Å 44 Å 33 Å 24 Å Mean probe spacing ( m) (b) MgO thickness (Å) Figure 2.19: CIPT measurement on a MgO wedge. The sheet resistance of a MgO tunnel barrier measured in an unannealed Pt (300 Å) / MgO (0 100 Å) / Pt (50 Å) junction. Note that this is a non-magnetic tunnel junction with a MgO wedge. (a) CIPT scans where sheet resistance is plotted as a function of the mean probe distance. Each scan is measured at a different MgO thickness on the wedge. (b) Inset shows the image of a 12-finger probe. It is visually noticeable that not all the probes are equidistantly placed from the probe center. The figure shows the sheet resistance plotted as a function of MgO thickness. For details, please refer to [47]. which is discussed later. Before that, we discuss results on a NiO barrier where we show that RA products over several orders of magnitude can be measured with the CIPT technique. CIPT on NiO tunnel junctions In the previous section, we showed that the MgO barriers created in our sputter system are not very reliable. In this section we therefore turn to an alternative barrier material, nickel oxide. NiO has been used in tunnel junctions [70] and single electron transistors [71] before. Like MgO, NiO too is sputtered from a single composite target, allowing the deposition of an oxide wedge, and making it an appealing candidate to test our CIPT set-up. The experimental approach, as well as the experimental goal to provide evidence of a tunnel current remain the same as for the MgO barrier. In the inset in Figure 2.20(b) the sample stack is shown. This sample was originally deposited for a different experiment [72], and is therefore quite an intricate multilayer. The exact details of this stack are not important for the CIPT measurements presented here. We will regard this junction as a stack with a bottom electrode, a wedge-shaped NiO barrier and a top electrode. A 12-finger probe was landed on the wedge at different thicknesses, and a CIPT scan was conducted at

73 2.5 Measuring electronic transport 61 Sheet resistance ( / ) Å 19.5 Å 18.0 Å 16.5 Å 15.0 Å 13.5 Å 13.0 Å 12.0 Å (a) Mean probe distance (µm) Pt [Co/Pt] n NiO [Co/Pt] n (b) NiO thickness (Å) RA ( µm) Figure 2.20: CIPT measurement on a AlO x based MTJ. (a) CIPT scans on a tunnel junction with a wedge-shaped NiO barrier. Each scan is taken at a different thickness of NiO. The lines are fits to theory. Inset: Zoom in for smaller probe spacings, showing a clear (albeit small) upturn at the smallest probe spacing even for 12 Å of NiO. (b) RA product obtained from the fits in figure a. Inset: Junction stack used for this experiment (thicknesses in Å): Pt 100/[Co 4/Pt 7/] 4 /Co 4/NiO x/[co 4/Pt 7/] 4 Co 4/Pt 20 with a wedge in the NiO [72]. For details, please refer to [47]. each of these spots. Figure 2.20(a) shows the individual CIPT scans at different NiO thicknesses, with fits to theory. Assuming the sheet resistances of both electrodes to be constant over the whole sample, in the fitting procedure they are shared as common variables between the different data sets. Just as in the case of MgO,the shape of these curves provide clear evidence for tunneling through the NiO barrier. Between the different data sets, it is seen that the the sheet resistance decreases with NiO thickness. Even at 1.2 nm of NiO, although the sheet resistance seems to be almost constant for all probe spacings, one definitively observes a clear upturn in the sheet resistance at the smallest probe spacing, which indicates tunneling behavior [see inset in Figure 2.20(a)]. For all these data points, the measurement error is smaller than the symbol size, as was verified by repeating the measurement multiple times. From the fits through the CIPT scans, the RA product can be extracted for different NiO thicknesses. The RA products so obtained are plotted on a semilog scale against NiO thickness in Figure 2.20(b). A linear fit through this data shows that the RA is exponentially dependent on the barrier thickness, one of the characteristics of ( a tunnel barrier. Simmons [73] predicts that the tunnel resistance scales 8mφ ) with d exp d, where φ is the barrier height which is assumed to be constant over the NiO layer, m is the electron mass, is the reduced Planck constant, and d is the thickness of the NiO. From this measurement φ is calculated to be 0.5 ev. This, together with the nice fits in Figure 2.20(a), provides a strong indication of

74 62 Chapter 2 Probing electronic, magnetic and structural properties (a) Sheet resistance ( / ) (b) MR CIPT (%) Mean probe spacing ( m) (c) Sheet resistance ( / ) (d) MOKE (arb.u.) (e) Pt 70 Co 30 AlOx 13 Co 50 IrMn 50 Pt H (mt) 0 Figure 2.21: CIPT measurement on a AlO x based MTJ. (a) Sheet resistance of a MTJ stack versus increasing probe spacing for both the parallel (low-resistive, ) and the antiparallel (high-resistive, ) state. The lines are fits to theory. (b) Measured MR CIPT with fit, both extracted from figure a. Note that this measured MR CIPT is different from the actual junction TMR, which is calculated to be 18%. (c) Minor MR loop measured at 39 µm probe spacing, showing the switching field of the free (top) electrode. (d) Minor MOKE loop, showing the same switching for the top electrode. (e) The junction stack used for these measurements. Numbers indicate the thickness of the layer in Å. Note that the thickness of the AlO x layer is the deposited Al thickness prior to oxidation. For details, please refer to [47]. the NiO being a tunnel barrier. To conclude this section, once again this measurement shows that the CIPT method can be employed to demonstrate tunneling through a barrier. We would like to emphasize that, by varying the barrier thickness in one single sample, the RA product can be determined over many orders of magnitude. Furthermore, if necessary, by changing the thickness of the top and bottom electrode thickness, the RA product can be measured over a broader range.

75 2.5 Measuring electronic transport 63 CIPT on an AlO x based MTJ Next we choose a MTJ to apply the CIPT method. The junction stack grown for this experiment is Pt / IrMn / Co / AlO x / Co / Pt, as shown in Figure 2.21(e). The first aim of the experiment was to confirm that we could achieve a parallel and antiparallel alignment of the two Co layers on either side of the AlO x barrier, and then perform CIPT measurements in both (parallel and antiparallel) cases. The measured RA for each magnetic configuration would then allow the determination of the TMR. To achieve these magnetic configurations, an antiferromagnet was used to pin the bottom Co layer. The stack shown in Figure 2.21(e) was annealed at 290 C for 30 minutes to get this Co layer exchange biased. We confirmed that the layers could be aligned parallel and antiparallel with an external field using the MOKE [see Section 2.4.2]. Figure 2.21(d), shows the MOKE loop. Next, we performed the CIPT measurement. After landing the 12-finger probes, we fixed the distance between the current and the voltage probes to 39 µm and measured the inner loop of the MTJ. The measured sheet resistance is shown in Figure 2.21(c) as a function of the external magnetic field. The measurement clearly shows a magnetoresistance loop. The switching fields are in good agreement with those shown in the MOKE loop of Figure 2.21(d). Next, we measured the sheet resistance as a function of the distance between the probes. The external fields were tuned to ensure the parallel and antiparallel alignement of the MTJ. This measurement is shown in Figure 2.21(a) with corresponding fits from CIPT theory [47, 69]. One immediately notices that the sheet resistance of the MTJ is higher in the antiparallel configuration than that in the parallel configuration. One may now define MR CIPT = R high R low R low, which is plotted in Figure 2.21(b). The line here signifies the difference in the theoretical fits of Figure 2.21(a). Note that we clearly measure a MR CIPT of around 4% which should be distinguished from the TMR in the MTJ. The procedure for calculating the TMR on such a MTJ is as follows: for both the parallel and antiparallel configurations in Figure 2.21(a), one obtains a RA product from the theoretical fits. The TMR is simply given by TMR = RA high RA low RA low. For this particular junction, we obtain a TMR value of 18%. This measured TMR is smaller that the typical TMR (40%) we observe in our MTJs. This is probably due to the fact that we have not optimized the oxidation procedure for the comparatively thinner Al layer we oxidized here. In normal MTJs we use an Al layer of 23 Å oxidized which is oxidized for 200 seconds, while in this measurement we oxidized a 13 Å Al film for 45 seconds. Such thin AlO x layers were chosen so that the obtained RA products were measurable with our CIPT set-up. In the next chapter, we will focus on the structural and magnetic properties of CoFeB, and the effect of crystallization on them. Since these alloys are invariably annealed when used in tunnel junctions, the effect of the anneal on the magnetic

76 64 Chapter 2 Probing electronic, magnetic and structural properties properties is a relevant aspect of their properties.

77 BIBLIOGRAPHY 65 Bibliography [1] C. H. Kant, Probing spin polarization: point-contacts and tunnel junctions. PhD Thesis, Eindhoven University of Technology (2005) , 2.5.1, 2.14, [2] J. C. Lapp, Glass substrates for AMLCD applications:properties and implications. Proc. SPIE. 3014, 1 (1997) DisplayTechnologies/content/docs/TIP101.pdf [3] P. R. LeClair, Fundamental aspects of spin-polarized tunneling: Magnetic tunnel junctions and spin filters. PhD Thesis, Eindhoven University of Technology (2002) , [4] G. R. Harp, and S. S. P. Parkin, Seeded epitaxy of metals by sputter depostion. Appl. Phys. Lett. 65, 3063 (1994) [5] G. R. Harp, and S. S. P. Parkin, Epitaxial growth of metals by sputter depostion. Thin Solid Films. 288, 315 (1996) [6] M. Ohring, The Materials Science of Thin Films. Academic Press, Florida (2002) [7] S. A. Campbell, The Science and Engineering of Microelectronic Fabrication. Oxford University Press, NY. (2001) [8] C. A. M. Knechten, Plasma oxidation for magnetic tunnel junctions. PhD Thesis, Eindhoven University of Technology (2005) [9] R. L. Snyder, X-Ray Characterization of Materials. Wiley-VCH, Weinheim. (1999) , [10] B. D. Cullity, Elements of X-ray diffraction, 2nd edn. Addison-Wesley, London. (1978) [11] V. Holý, U. Pietsch, T. Baumbach, High-Resolution X-Ray Scattering from Thin Films and Multilayers. Springer Verlag, Berlin. (1999) [12] P. F. Fewster, Elements of X-ray diffraction, 2nd edn. Rep. Prog. Phys. 59, 1339 (1996) [13] D. E. Sayers, E. A. Stern, and F. W. Lytle, Practical algorithm for background subtraction. Phys. Rev. Lett. 27, 1204 (1971) , 2.2.2, [14] D. C. Koningsberger, and R. Prins, eds., X-ray Absorption. John Wiley and sons, NY. (1989)

78 66 Chapter 2 Probing electronic, magnetic and structural properties [15] M. Newville, Fundamental of XAFS. action=attachfile&do=get&target=newville_xas_fundamentals.pdf (2004) , 2.6 [16] N. Binsted, EXCURV98: CCLRC Daresbury Laboratory computer program [17] J. J. Rehr, FEFF, Department of Physics, University of Washington. http: //leonardo.phys.washington.edu/feff/welcome.html [18] J. J. Rehr, and R. C. Albers, Theoretical approaches to x-ray absorption fine structure. Rev. Mod. Phys. 72, 621 (2000) [19] F. M. F. de Groot, and R. C. Albers, X-ray absorption and dichroism of transition metals and their compounds. J. Elec. Spec. Rel. Pheno. 67, 529 (1994). [20] J. Stöhr, NEXAFS spectroscopy. Springer Verlag, Berlin (1992) [21] D. Briggs, and M. P. Seah, eds. Practical surface analysis, 2nd edn. Wiley, Chichester. (1990) [22] S. S. P. Parkin, C. Kaiser, A. Panchula, P. M. Rice, B. Huges, M. Samant, S.- H. Yang, Giant tunneling magnetoresistance at room temperature with MgO (100) tunnel barriers. Nature Mater. 3, 862 (2004) , [23] S. Yuasa, T. Nagahama A. Fukushima, Y. Suzuki, and K. Ando, Giant roomtemperature magnetoresistance in single-crystal Fe/MgO/Fe magnetic tunnel junctions. Nature Mater. 3, 868 (2004). [24] A. A. Tulapurkar, Y. Suzuki, A. Fukushima, H. Kubota, H. Maehara, K. Tsunekawa, D. D. Djayaprawira, N. Watanabe, and S. Yuasa, Spin-torque diode effect in magnetic tunnel junctions. Nature 438, 339 (2005). [25] H. Kubota, Y. Suzuki, A. Fukushima, H. Kubota, H. Maehara, K. Tsunekawa, D. D. Djayaprawira, N. Watanabe, and S. Yuasa, Quantitative measurement of voltage dependence of spin-transfer torque in MgO-based magnetic tunnel junctions. Nature Phys. 7, 37 (2007) [26] J. S. Corneille, J. W. He, and D. W. Goodman Surf. Sci. 306, 269 (1994) [27] R. Lavrijsen, MgO based magnetic tunnel junctions. Master Thesis, Eindhoven University of Technology (2006) , 2.8, 2.18 [28] S. Tougaard, Practical algorithm for background subtraction. Surf. Sci. 216, 343 (1989)

79 BIBLIOGRAPHY 67 [29] B. Berghmans, The electronic structure of an organo-metallic interface: a test case for ultra-violet photoemission spectroscopy. Master Thesis, Eindhoven University of Technology (2006). [30] M. V. Tiba, Organo-metallic structures for spintronic applications. PhD Thesis, Eindhoven University of Technology (2005) [31] M. Faraday, Experimental search in electricity: On the magnetization of light and the illumination of the magnetic lines of force. Philosoph. Trans. Royal Soc. London 136, 1 (1846) [32] J. Kerr, On rotation of the plane of polarization by refection from a pole of a magnet. London, Edinbrugh and Dublin Philosoph. Mag. and J. of Science. 3, 321 (1877) [33] M. van Kampen, Ultrafast spin dynamics in ferromagnetic metals. PhD. Thesis, Eindhoven University of Technology (2003) [34] J. H. H. Rietjens, Precessional magnetization dynamics in micron sized ferromagnetic elements. Master Thesis, Eindhoven University of Technology (2004) [35] J. Nogués, I. K. Schuller, Exchange bias. J. Magn. Magn. Mat. 192, 203 (1999) [36] F. Dalla Longa, J. T. Kohlhepp, W. J. M. de Jonge, and B. Koopmans, Laserinduced magnetization dynamics in Co/IrMn exchange coupled bilayers. J. Appl. Phys. 103, 07B101 (2008) [37] J. L. Erskin and E. A. Stern, New Technique for Investigating Noncrystalline Structures: Fourier Analysis of the Extended X-Ray Absorption Fine Structure. Phys. Rev. B 12, 5016 (1975) [38] G. Schütz, W. Wagner, E. Wilhelm, P. Kienle, R. Zeller, R. Frahm, and G. Materlik, Magnetic K-edge absorption in 3d elements and its relation to local magnetic structure. Phys. Rev. Lett. 58, 737 (1987) [39] B. T. Thole, P. Carra, F. Sette, and G. van der Laan, x-ray magnetic dichroism as a probe for orbital magnetism. Phys. Rev. Lett. 68, (1992) , [40] P. Carra, B. T. Thole, M. Altarelli, and X. Wang, x-ray magnetic circular dichroism and local magnetic fields. Phys. Rev. Lett. 70, 694 (1993) , 2.4.3

80 68 Chapter 2 Probing electronic, magnetic and structural properties [41] C. T. Chen, Y. U. Idzerda, H.-J. Lin, N. V. Smith, G. Meigs, E. Chaban, G. H. Ho, E. Pellegrin and F. Sette, Experimental confirmation of the x-ray magnetic circular dichroism sum rules for Iron and Cobalt. Phys. Rev. Lett. 75, (1995) [42] J. Stöhr, X-ray magnetic circular dichroism spectroscopy of transition metal thin films. J. Elec. Spec. Rel. Phenom. 75, 253 (1995) [43] H. Ebert, Magneto-optical effects in transition metal systems. Rep. Prog. Phys. 59, 1665 (1996) , [44] S. W. Lovesey, and S. P. Collins, X-ray scattering and absorption in magnetic materials. Clarendon press, Oxford (1996) [45] R. Nakajima, X-ray magnetic circular dichroism spectroscopy in transition metal thin films. PhD. Thesis, Stanford University (1998) [46] K. Fleury-Frenettea, S. S. Dhesi, G. van der Laan, D. Strivay, G. Weber, and J. Delwiche, Characteristics of the iron moment in Dy-Fe and Dy-FeCo amorphous alloys studied by x-ray magnetic circular dichroism. J. Mag. Mag. Mater. 220, (2000) [47] R. C. Huijink, Current-in-plane tunneling. Master Thesis, Eindhoven University of Technology (2002). 2.5, 2.5.4, 2.5.4, 2.19, 2.20, 2.21, [48] R. Meservey, and P. M. Tedrow, Spin-polarized electron tunneling. Phys. Rep. 238, 173 (1994) [49] I. Giaever, Energy gap in superconductors measured by electron tunneling. Phys. Rev. Lett. 5, 147 (1960) [50] I. Giaever, Electron tunneling between two superconductors. Phys. Rev. Lett. 5, 464 (1960) [51] J. Nicol, S. Shapiro, and P. H. Smith, Direct measurement of the superconducting energy gap. Phys. Rev. Lett. 5, 461 (1960) [52] J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Theory of superconductivity. Phys. Rev. 108, 1175 (1957) [53] R. Meservey, P. M. Tedrow, and P. Fulde, Magnetic field splitting of the quasiparticle states in superconducting Al films. Phys. Rev. Lett. 25, 1270 (1970) [54] P. M. Tedrow, and R. Meservey, Spin-dependent tunneling into ferromagnetic Ni. Phys. Rev. Lett. 26, 192 (1971)

81 BIBLIOGRAPHY 69 [55] J. A. Hertz, and K. Aoi, Spin dependent tunneling from transition metal ferromagents. Phys. Rev. B 8, 3252 (1973) [56] M. B. Sterns, Simple explanation of tunneling spin polarization of Fe, Co, and Ni and its alloys. J. Mag. Mag. Mater. 5, 167 (1977) [57] P. M. Tedrow, and R. Meservey, Spin paramagnetic effects in superconducting Al films. Phys. Rev. B 8, 5098 (1973) [58] Y. Sakuraba, M. Hattori, M. Oogane, Y. Ando, H. Kato, A. Sakuma, T. Miyazaki, and H. Kubota, Giant tunneling magnetoresistance in Co2MnSi/AlO/Co2MnSi magnetic tunnel junctions. Appl. Phys. Lett. 88, (2006) [59] Y. Sakuraba, M. Miyakoshi, M. Oogane, Y. Ando, A. Sakuma, T. Miyazaki, and H. Kubota, Direct observation of half-metallic energy gap in Co2MnSi by tunneling conductance spectroscopy. Appl. Phys. Lett. 89, (2006) [60] K. Maki, Pauli paramagnetism and superconducting state. II. Prog. Theor. Phys. 32, 29 (1964). 2.15, [61] P. Fulde, High field superconductivity in thin films. Adv. Phys. 22, 667 (1973) [62] M. Tinkham, Introduction to superconductivity. McGraw-Hill, New York (1996) [63] A. C. Rose-Innes, and E. H. Rhoderick, Introduction to superconductivity. Pergamon Press (1978) [64] J. Lambe, and R. Jaklevic, Molecular vibrational spectra by inelastic electron tunneling. Phys. Rev. Lett. 17, 1139 (1966) [65] D. Tsui, R. Dietz, and L. Walker, Multiple magnon excitations in NiO by electron tunneling. Phys. Rev. Lett. 27, 1729 (1969) [66] P. K. Hansma, Inelastic electron tunneling. Phys. Rep. 30, 145 (1977) [67] D. G. Walmsley, and J. L. Tomlin, Compilation for inelastic electron tunneling spectra for molecules chemisorbed on metal oxides. Prog. Surf. Sci. 18, 247 (1987). [68] K. W. Hipps, and U. Mazur, Inelastic electron tunneling spectroscopy. Handbook of Vibrational Spectroscopy, (John Wiley and Sons, Chichester) (2002) [69] D. C. Worledge, and P. L. Trouilloud, Magnetoresistance measurement of unpatterned magnetic tunnel junction wafers by current-in-plane tunneling. Appl. Phys. Lett. 83, 84 (2003) , 2.5.4, 2.5.4

82 70 Chapter 2 Probing electronic, magnetic and structural properties [70] S. Maekawa, and U. Gafvert, Electron tunneling between ferromagnetic films. IEEE Trans. Mag. 18, 707 (1982) [71] H. Shimada, K. Ono, and Y. Ootuka, Ferromagnetic single electron transistor. Sol. State. Elec. 42, 1407 (1998) [72] G. Malinowski, F. Dalla Longa, J. H. H. Reitjens, and B. Koopmans, unpublished , 2.20 [73] J, G. Simmons, Generalized formula for the electric tunnel effect between similar electrodes separated by a thin insulating film. J. Appl. Phys. 34, 1793 (1963)

83 Chapter 3 Magnetic properties of CoFeB Effect of crystallization Abstract: In this chapter 1, we will discuss some basic aspects regarding the structure of CoFeB, the effect of annealing on its structure and its magnetic properties. We will use the information obtained here as a starting point for further experimental work in the later chapters. We will begin with XRD study of Co 72 Fe 20 B 8 (at. %) films to investigate their crystallization. Then we probe the influence of crystallization on their magnetic properties. Here we use a novel way to probe the effect of film crystallization on their magnetic properties, in that, we use wedge-shaped samples probed with the MOKE technique to perform a thickness dependent crystallization study. We notice that, although the coercivity of these alloys is strongly influenced by the anneal temperature and their thickness, the changes in coercivity are not a direct indication of film crystallization. In the second part of the chapter we will investigate the composition dependence of crystallization Co 80-x Fe x B 20, together with the corresponding effect on magnetic properties. Here too, we reach a similar conclusion regarding the coercivity of these alloys. Moreover, we find indications that the distribution of the grain sizes in the crystalline film, may have a strong influence in the observed behavior of the coercivity. 1 Some parts of Section 3.3 of this chapter appeared as conference proceedings in Journal of Applied Physics [20], Journal Physics D: Applied Physics [21], and Journal of Magnetism and Magnetic Materials [22]. 71

84 72 Chapter 3 Magnetic properties of CoFeB 3.1 Background After the advent of magnetic tunnel junctions (MTJ s) [1, 2], the search for novel materials to advance the performance of these devices has intensified. Amorphous ferromagnetism in transition metal-borides has been extensively studied, mainly due to the magnetism observed in these materials, and their potential industrial applications [3]. These alloys are ideal candidates as they exhibit unique magnetic and electronic transport properties. Due to their amorphous / nanocrystalline nature, these alloys exhibit low random anisotropy, making them magnetically soft and suitable for the free layer in MTJ s [4]. They have Curie temperatures well above room temperature and exhibit low crystallization temperatures, providing an additional handle to explore their fundamental properties depending on their morphology. Also, their compositions can be tuned to tailor their magnetic properties depending on the application [5]. Furthermore, it has been recently shown that MTJ s incorporating the ternary alloy CoFeB in conjunction with AlO x [6, 7] and MgO [8 10] barriers exhibit recordhigh TMR values at room temperature, emphasizing their superior electronic and transport properties. However, from the fundamental point of view, the reason behind the high TMR in CoFeB based junctions is not yet resolved. Especially, the role of the crystal structure of the ferromagnet at the barrier-ferromagnet interface, which can be transformed from amorphous to crystalline after an anneal, is yet to be addressed. In order to address these outstanding issues, in this chapter we perform a preliminary experimental analysis to investigate the impact of crystallization on the structural and magnetic properties of the metallic glass CoFeB alloys. We will use this knowledge in the next chapters where we explore the tunneling spin polarization (TSP) of amorphous ferromagnets and the impact of their crystallization on the TSP. 3.2 Sample preparation To confirm the possibility of crystallizing Co 72 Fe 20 B 8, we deposited CoFeB layers on Si // SiO x and Si // SiO x / AlO x buffer layers using DC magnetron sputtering (base pressure < 10 8 mbar) at room temperatures from a single CoFeB target and investigated their structural properties using high-angle x-ray diffraction (XRD - Cu K α ) as a function of post-deposition anneal temperature. The layers were annealed for 30 minutes in a magnetic field of 250 mt in argon atmosphere at various temperatures with a ramp up/down rate of 20 C/min. Further, we used SQUID (Superconducting Quantum Interference Device) magnetometery to measure the magnetic moment and coercivity of 700 Å thick CoFeB layers.

85 3.2 Sample preparation 73 XRD Intensity (counts/sec) ºC As-dep ºC 450ºC 425ºC 400ºC 350ºC 300ºC (º) Figure 3.1: XRD on 700 Å Co 72 Fe 20 B 8 layers. High-angle x-ray diffraction measurements on Si // SiO x / AlO x / Co 72 Fe 20 B 8 ( 700 Å) / Al as a function of post-deposition anneal temperature. As-deposited CoFeB grows amorphous/nanocrystalline on both SiO x and AlO x. Note that after the anneal the peak position remains constant at 2θ 45.3, an indication of a strongly textured film Grain Size (coherence length) Net area under peak Grain size (Å) Net area (2 *counts/sec) Anneal temperature (ºC) 0 Figure 3.2: Grain sizes in crystalline 700 Å Co 72 Fe 20 B 8 layers. The Scherrer formula along with the high angle XRD measurements yields the out-of-plane coherence length t = 0.9 λ/fwhw cos θ, which is plotted along with the net area under the peak.

86 74 Chapter 3 Magnetic properties of CoFeB 3.3 Properties of Co 72 Fe 20 B 8 We will first investigate relatively thick, 700 Å, Co 72 Fe 20 B 8 films. This composition was chosen since it is close to the minimum boron content which renders CoFeB amorphous [5], which implies that the alloy crystallizes at lower temperatures than alloys with higher boron content. Low-angle XRD measurements (not shown) reveal a smooth growth of the CoFeB layers on both SiO x and AlO x. Moreover, contrary to CoFe, as-deposited CoFeB grows quasi-amorphous / nanocrystalline on both SiO x and AlO x. A high-angle scan from 2θ = 5 to 110 reveals no peaks in the asdeposited case. In the first panel of Figure 3.1 we plot the observed diffraction intensity for a 700 Å thick Co 72 Fe 20 B 8 in a very limited 2θ range. In this region we observe peaks in annealed samples due to crystallization of Co 72 Fe 20 B Crystallization of Co 72 Fe 20 B 8 Figure 3.1 shows the high-angle XRD scans around 2θ = 45.3 measured on Si // SiO x / AlO x / Co 72 Fe 20 B 8 / Al samples. Each scan is measured on different samples grown in the same batch and annealed at respective temperatures. After an anneal of 250 C, a distinct peak appears at 2θ 45.3, corresponding to a lattice spacing d 2.00 Å. This is an indication of the onset of crystallization of CoFeB. An estimate of the structure in which CoFeB crystallizes is difficult as the 2θ peak lies close to those expected for Co and Fe. The Scherrer formula [see Equation 2.1] can be used to calculate the magnitude of the coherence length (a measure of the out-of-plane crystallite size). The values calculated are plotted in the Figure 3.2, along with the net area under the peaks. From Figures 3.1 and 3.2, the crystallization behavior of Co 72 Fe 20 B 8 might be summarized as: (i) the net area under the peak, which corresponds to the amount of crystalline phase in the film, along with the crystallite size increases gradually between anneals at 250 C and 425 C, and (ii) anneals above 425 C seem to indicate a more rapid growth of the crystallites, and more rapid crystallization. The fact that the net area under the peak increases with the anneal temperature is an indication that the change over from amorphous to crystalline structure is a gradual one. Anneals above 485 C would be necessary to find the temperature at which the films crystallize completely. However, these temperatures are considerably higher than those used in MTJs. As mentioner earlier, to confirm the exact structure of the poly-crystalline film, and to understand the behavior of the grain size as a function of anneal temperature, further investigations, probably with scanning tunneling microscopy (STM) and high-resolution transmission electron microscopy (HRTEM), would be necessary. Such HRTEM measurements are discussed in Chapter 4 and Chapter 5. For the moment, it is sufficient to say that these films crystallize in a highly textured fcc

87 3.3 Properties of Co 72 Fe 20 B Normalized resistance Å 500Å 700Å 60Å 120Å Non-isothermal anneal temperature ( C) Figure 3.3: Resistance of Co 72 Fe 20 B 8 layers. Normalized resistance of CoFeB films measured in an UHV oven (pressure during anneal < 10 7 mbar). The resistance of the films gradually decreases as the temperature increases, suggesting a change over from amorphous to crystalline CoFeB films. Note that even the 60 Å films show a dramatic decrease in their resistance. structure Effect of Co 72 Fe 20 B 8 crystallization on film resistance In the XRD study discussed above, the film thickness studied was 700 Å, which is reasonably thicker than the 50 Å films used in MTJs. Since XRD is not a suitable tool to observe the crystallization of such thin films (< 100 Å), we performed insitu four point resistance measurements in a UHV oven (pressure during anneal < 10 7 mbar) during a non-isothermal anneal. The results are shown in Figure 3.3 which plots the normalized film resistance as a function of the anneal temperature. These measurements show that the resistance of the films gradually decreases as the temperature increases, and finally levels off to low values. This suggests a change over from amorphous to crystalline CoFeB in these films as the conduction electrons encounter far less scattering events in the crystalline films as compared to the nonperiodic amorphous layers. All the details in this study, however, are yet to be understood. For instance, the initial shape of the resistance curve is not similar for films of all thicknesses. Moreover, it seems that the crystallization temperatures of the films do not show

88 76 Chapter 3 Magnetic properties of CoFeB a systematic dependence on the thickness of the film. One should keep in mind, however, that these measurements were done in completely different anneal conditions as compared to the XRD study. In this case, the temperature of the oven was constantly ramped at 2 C/min, compared to the isothermal anneals used for the XRD study. Note that the 60 Å and 120 Å films show a larger decrease in their resistance as compared to films thicker than 120 Å after anneals above 400 C. This suggests that the thicker films are not completely crystalline after anneals at such temperatures, while the thinner films (60 Å and 120 Å) show comprehensive crystallization. This conjecture is supported by HRTEM measurements on 700 Å and 60 Å films which are discussed in Chapter Effect of Co 72 Fe 20 B 8 crystallization on magnetic properties We used a novel way to probe the effect of film crystallization on the magnetic properties of CoFeB. More specifically, we deposited Å CoFeB wedges and analyzed the thickness dependence of the effect of crystallization on magnetic properties like coercivity with the MOKE technique [see Section 2.4.2]. Figure 3.4(a) shows a plot of the CoFeB coercivity (H c ) as a function of film thickness and anneal temperature measured on Al / AlO x / CoFeB (0 300 Å) / Al wedges. The MOKE loops were measured with the field aligned in the direction of the field applied during the anneal. Before we discuss the MOKE measurements of Figure 3.4(a), let us briefly recapitulate the experimental facts known about these films: (i) HRTEM measurements on films thinner that 100 Å and annealed at 450 C also confirm crystallization of the films [see Chapter 4]. (ii) XRD and HRTEM measurements on films thicker than 100 Å and annealed above 300 C indicate crystallization of the films, especially for anneal temperatures at 450 C. (iii) From XRD, we also notice that the out-of-plane grain sizes increase nonmonotonously with the anneal temperature, as shown in Figure 3.2. However, this is the case for a uniform 700 Å film, and the trend observed there might not directly applicable to wedge-shaped samples discussed in this section. (iv) From HRTEM measurements, one also notes that for films thinner than 100 Å, the out-of-plane grain size is limited by the film thickness itself. Such observations have been made in our HRTEM studies, as well as those of Takeuchi et al. [11], and will be discussed in Chapter 4. Based on these considerations, one may imagine that the crystallization induces magnetocrystalline anisotropy in the films, which will lead to an increase in coercivity. The induced coercivity will also depend on grain sizes, since these are potential domain wall pinning centers. In Figure 3.4(a), although there are many details not yet understood in the

89 3.3 Properties of Co 72 Fe 20 B 8 77 Coercivity, H c (ka/m) ºC 300Å 441ºC 433ºC 420ºC 400ºC as-dep 300ºC Co 72 Fe 20 B 8 Thickness (Å) Figure 3.4: Coercivity of Co 72 Fe 20 B 8 layers. H c vs thickness d measured with MOKE on Å thick Co 72 Fe 20 B 8 wedges, annealed at various temperature. observed behavior, it is evidently seen that the coercivity heavily depends on the thickness of the ferromagnetic film when annealed up to 500 C. Anneals below 400 C have little impact on coercivity, while anneals at higher temperatures influence the coercivity strongly, especially for thicker films. Note that for thicknesses lower than 100 Å, the sheet resistance measurements discussed above as well as our HRTEM studies have shown that the films are crystalline when annealed at temperatures around 450 C. However, for these thicknesses and anneal temperatures, we do not observe a significant increase in H c compared to its as-deposited values. These observations suggest that, although a significant increase in H c may be taken as an indication of crystallization, the opposite is not true, i.e., no increase in H c may not be taken as an indication of an amorphous film. In the next section, we turn to Co 80-x Fe x B 20 with a higher (20% at.) boron content. MOKE and XRD measurements on these alloys provide some hint on the origin of behavior of H c which we observe for Co 72 Fe 20 B 8. However, there too we notice that more experimental work is necessary to address the intricate behavior of H c.

90 78 Chapter 3 Magnetic properties of CoFeB 3.4 Properties of Co 80-x Fe x B 20 Next we will discuss a composition dependent study of the structural and magnetic properties of Co 80-x Fe x B 20 alloys. As mentioned in Chapter 1, nowadays, these alloys are regularly used in MgO based MTJs and spin-torque based tunnel junctions. Moreover, to obtain a high TMR, such junctions are annealed up to 400 C, which also results in crystallization of these alloys. However, there are only a few studies reported which address post-anneal structural and magnetic properties of these alloys [11 13] Crystallization of Co 80-x Fe x B 20 from XRD We followed the same procedure as that discussed for Co 72 Fe 20 B 8 alloys. First we performed XRD on several different CoFeB compositions by systematically varying the Fe content in the alloy. Both as-deposited (not shown) and annealed films were investigated. Figure 3.5 shows XRD patterns for d = 700 Å Co 80-x Fe x B 20 films after annealing at T a = 450 C. The vertical dotted line indicates a peak from the Si substrate, while the peaks marked by arrows are from the CoFeB alloy [11 13]. The XRD pattern at the top of the graph is a representative measurement for an asdeposited Fe 80 B 20 film. It shows no diffraction peak indicating that the as-deposited film is amorphous. We preformed such measurements on all the compositions (not shown), and found no diffraction peaks, indicating that as-deposited Co 80-x Fe x B 20 films grow amorphous. After annealing at 450 C, similar to the observation we made for Co 72 Fe 20 B 8 in Figure 3.1, we also observe a peak in the diffraction spectrum for Co 80-x Fe x B 20 alloys, indicating film crystallization. Crystallization at these temperatures has also been observed elsewhere, and is consistent with literature [13]. In Figure 3.5, one also notes a shift in position of the CoFeB diffraction peak with composition. This indicates that each composition crystallizes with a different lattice spacing, also noted by Tsunekawa et al. [14]. Moreover, a comparison of all the compositions shows that Co 60 Fe 20 B 20 exhibits the highest degree of crystallization, in other words, the highest integrated intensity of the diffraction peak. Since Co 60 Fe 20 B 20 is arguably the most widely used CoFeB composition in MgO based MTJs [10, 15], next we will focus on its magnetic behavior Effect of Co 60 Fe 20 B 20 crystallization on magnetic properties In Figure 3.6(a) we show H c of Co 60 Fe 20 B 20 as function of film thickness, d for different anneal temperatures, T a. Please note that, contrary to the Co 72 Fe 20 B 8 sample studied in Figure 3.4 where the maximum wedge thickness was 280 Å, in this case the maximum wedge thickness is 600 Å. Nevertheless, a closer comparison of both measurements show some similarity. In the present case too, a sharp increase of H c is observed for T a > 440 C. Moreover, we also observe a high sensitivity to the

91 3.4 Properties of Co 80-x Fe x B as-dep. x = 80 x = 80 x = 68 T a = 450 C Intensity (s -1 ) x = 56 x = 44 x = 32 x = 20 x = 8 Si diffraction angle ( ) Figure 3.5: XRD of Co 80-x Fe x B 20 layers. XRD pattern of 700 Å Co 80-x Fe x B 20 films annealed at 450 C. The vertical dotted line indicates a diffraction peak from the Si substrate. The arrows indicate the CoFeB diffraction peaks. anneal temperature; on increasing T a from 440 C to 475 C a significant larger part of the film shows a higher H c. Let us first analyze the MOKE loops before, in and after the sharp transition in H c. As an example, in Figure 3.6(b-d), we plot the MOKE loops for the sample annealed at 450 C. The corresponding film thickness is 100 Å (b), 200 Å (c) and 400 Å (d) indicated by the arrows in Figure 3.6(a). We see a transition from a square loop (a) to a rounded loop (c). The rounded loop is typical for a magnetization reversal that is dominated by domain wall pinning at grain boundaries. The double loop seen in Figure 3.6(b), may seem a bit puzzling. One explanation may be that there are amorphous and crystalline regions under the 75 µm laser spot which is used to perform MOKE. These individual regions may exhibit a different coercivity depending on whether they are amorphous or crystalline. The contributions from these regions may also differ depending on the anneal temperature. Moreover, the transition between the two extreme coercivities observed in Figure 3.6(b), may

92 80 Chapter 3 Magnetic properties of CoFeB H c (ka/m) 35 (a) C Co 60 Fe 20 B C (d) (c) (b) 445 C C Asdep Co 60 Fe 20 B 20 thickness (Å) 1 (b) (c) (d) M/M s (arb. u) Å Å Field (ka/m) 400 Å Figure 3.6: Coercivity of Co 80-x Fe x B 20 layers. (a) H c vs thickness d of Co 60 Fe 20 B 20 annealed at different temperatures. The arrows indicate the data points of the MOKE loops shown in (b-d). depend on the distribution of the grain sizes in the film. Coming to the sharp increase in H c, the origin of this increase is still not clear. A possible explanation may come from the random anisotropy model first introduced by Alben et al. [16, 17]. This model states that the magnetic behavior changes from soft to hard when the grain size t becomes of the order of the exchange length, L ex. For L ex > t an averaging of the magnetic crystalline anisotropy takes place which leads to a low H c. When L ex t this averaging effect vanishes and grain boundaries become magnetic domain boundaries. In this case the magnetization reversal is dominated by pinning of domain walls at grain boundaries resulting in a higher coercivity. Recall that the exchange length of traditional ferromagnets like Co and Fe is of the order of Å [18]. As a first order approximation, let us assume a similar L ex for amorphous and crystalline CoFeB. Then,

93 3.5 Summary 81 For films thicker than the exchange length (d > 150 Å), the grain sizes t of the crystallites may be bigger than the exchange length too, The resulting H c would exhibit the commonly observed 1 behavior. This may be the case for d the sample annealed 475 C when the film thickness is greater than 200 Å. For film thicknesses lower than 100 Å, as we mentioned earlier, HRTEM and MOKE measurements indicate that, although the films are crystalline, they may not show any change in H c. This may be due to the fact that the out-ofplane grain sizes (t) are known to limited by the film thickness for thin films. Since t is not comparable to the exchange length, no sharp deviation in H c is observed in comparison to the as-deposited case. In the region of the sharp transition where a double loop is observed, two aspects may play a role. Firstly, both amorphous and crystalline parts of the film may contribute to the MOKE signal arising from the 75 µm laser spot. Secondly, for a given double MOKE loop, the sharpness of the transition between the two extreme H c may be influenced by the average grain size distribution. To shed more light on this issue, further experiments with an HRTEM or with an STM are indispensable. However, these results demonstrate that these alloys presents an opportunity to tune the magnetic properties of such layers by carefully choosing thickness and anneal temperatures. Moreover, the knowledge that a single anneal can change the structure of these films completely, also provides a handle to probe the electronic properties of these alloys. 3.5 Summary In summary, we investigated the structural and magnetic properties of CoFeB. Asdeposited films grow amorphous on AlO x and SiO x, and there is a gradual crossover from amorphous to crystalline structure after anneals above 250 C. The coercivity shows a dramatic change after anneals above 400 C, but the origin of this change is not yet clear. Although these experiments confirm the crystallization of CoFeB, they do not provide any information on the structure of the barrier-ferromagnet interface. Knowing that TSP is extremely sensitive to the structure at the barrierferromagnet interface [19], we will need a tool like cross-section HRTEM to provide conclusive evidence of the crystalline nature of CoFeB at the AlO x interface. We will have a closer look at the interface and spin-transport related properties in the next chapter.

94 82 Chapter 3 Magnetic properties of CoFeB Bibliography [1] J. Moodera, L. R. Kinder, T. M. Wong, and R. Meservey, Large magnetoresistance at room temperature in ferromagnetic thin film tunnel junctions. Phys. Rev. Lett. 74, 3273 (1995). 3.1 [2] T. Miyazaki, and N. Tezuka, Giant magnetic tunneling effect in Fe/Al 2 O 3 /Fe junction. J. Magn. Magn. Mater. 139, L231 (1995). 3.1 [3] K. Moorjani, and J. M. D. Coey, Magnetic Glasses, (Elsevier, Amsterdam) (1984). 3.1 [4] R. Hasegawa, Glassy metals : magnetic, chemical, and structural properties, (CRC press, Boca Raton) (1983). 3.1 [5] T. Egami, Magnetic amorphous alloys: physics and technological applications. Rep. Prog. Phys. 47, 1601 (1984). 3.1, 3.3 [6] D. Wang, C. Nordman, J. M. Daughton, Z. Qian, and J. Fink, 70% TMR at room temperature for SDT sandwiche junctions with CoFeB as free and reference layers. IEEE Trans. Mag. 40, 2269 (2004). 3.1 [7] H. X. Wei, Q. H. Qin, M. Ma R. Sharif, and X. F. Han, 80% tunneling magnetoresistance at room temperature for thin Al O barrier magnetic tunnel junction with CoFeB as free and reference layers. J. Appl. Phys. 101, 09B501 (2007). 3.1 [8] S. S. P. Parkin, C. Kaiser, A. Panchula, P. M. Rice, B. Huges, M. Samant, and S.-H. Yang, Giant tunneling magnetoresistance at room temperature with MgO (100) tunnel barriers. Nature Mater. 3, 862 (2004). 3.1 [9] D. D. Djayaprawira, K. Tsunekawa, M. Nagai, H. Maehara, S. Yamagata, N. Watanabe, S. Yuasa, Y. Suzuki, and K. Ando, 230 % room temperature magnetoresistance in CoFeB / MgO / CoFeB magnetic tunnel junctions. Appl. Phys. Lett. 86, (2005). [10] K. Tsunekawa, D. D. Djayaprawira, M. Nagai, H. Maehara, S. Yamagata, N. Watanabe, S. Yuasa, Y. Suzuki, and K. Ando, Giant tunneling magnetoresistance effect in low resistance CoFeB / MgO (001) / CoFeB magnetic tunnel junctions for read-head applications. Appl. Phys. Lett. 87, (2005). 3.1, [11] T. Takeuchi, K. Tsunekawa, Y.-s. Choi, Y. Nagamine, D. D. Djayaprawira, A. Genseki, Y. Hoshi, and Y. Kitamoto, Crystallization of amorphous CoFeB ferromagnetic layers in CoFeB/MgO/CoFeB magnetic tunnel junctions. Jpn. J. Appl. Phys. 46, L623 (2007) , 3.4, 3.4.1

95 BIBLIOGRAPHY 83 [12] S. Cardoso, C. Cavaco, R. Ferreira, L. Pereira, M. Rickart, P. P. Freitas, N. Franco, J. Gouveia, and N. P. Barradas, Characterization of CoFeB electrodes for tunnel junctions. J. Appl. Phys. 97, 10C916 (2005). [13] F. F. Li, R. Shariff, L. X. Jiang, X. Q. Zhang, X. F. Han, Y. Wang, and Z. Zhang, Thermal stability of Ir-Mn / CoFeB / AlO / CoFeB tunnel junctions. J. Appl. Phys. 98, (2005). 3.4, 3.4.1, [14] K. Tsunekawa, Y. S. Choi, Y. Nagamine, D. D. Djayaprawira, T. Takeuchi, and Y. Kitamoto, Influence of chemical compostion of CoFeB on tunneling magnetoresistance and microstructure in polycrystalline CoFeB / MgO / CoFeB magnetic tunnel junctions. Appl. Phys. Lett. 87, (2005) [15] H. Kubota, Y. Suzuki, A. Fukushima, H. Kubota, H. Maehara, K. Tsunekawa, D. D. Djayaprawira, N. Watanabe, and S. Yuasa, Quantitative measurement of voltage dependence of spin transfer torque in MgO based magnetic tunnel junctions. Nature Phys. 7, 37 (2007) [16] R. Alben, J. J. Becker, and M. C. Chi, Random anisotropies in amorphous ferromagnets. Appl. Phys. Lett. 29, 1653 (1978) [17] G. Herzer, Grain size dependence of coercivity and permeability in nanocrystalline ferromagnets. IEEE. Trans. Mag. 26, 1397 (1978) [18] R. C. O Handley, Modern Magnetic Materials: Principles and Applications, John Wiley and Sons, New York (2000) [19] P. LeClair, J. T. Kohlhepp, H. J. M. Swagten, and W. J. M. de Jonge, Interfacial density of states in magnetic tunnel junctions. Phys. Rev. Lett. 86, 1066 (2001). 3.5 [20] P. V. Paluskar, J. T. Kohlhepp, H. J. M. Swagten, and B. Koopmans, Co 72 Fe 20 B 8 : Structure, magnetism, and tunneling spin polarization. J. Appl. Phys. 99, 08E503 (2006). 1 [21] P. V. Paluskar, J. T. Kohlhepp, H. J. M. Swagten, B. Koopmans, R. Wolters, H. Boeve, and E. Snoeck, Influence of interface structure on the tunneling spin polarization of CoFeB alloys. J. Phys. D: Appl. Phys. 40, 1234 (2007). 1 [22] H. J. M. Swagten, P. V. Paluskar, R. Lavrijsen, J. T. Kohlhepp, and B. Koopmans, tunneling spin polarization and annealing of Co 72 Fe 20 B 8. J. Magn. Magn. Mater. 310, 2012 (2007). 1

96 84 Chapter 3 Magnetic properties of CoFeB

97 Chapter 4 Key concepts in spin tunneling Amorphous vs. crystalline Co 72 Fe 20 B 8 Abstract: Recently, ternary amorphous ferromagnets have boosted the performance of MTJs and spin-torque devices. Despite their immense importance in spintronics, the question why such amorphous ferromagnets display a significant TSP? has not been asked yet. Also the role of non-magnetic metalloids like boron in the alloy TSP has neither been calculated nor experimentally probed. In this chapter 1, we provide compelling evidence to establish that, contrary to one s elementary guess, the TSP of quasi-amorphous CoFeB is larger than that of highly textured fcc CoFeB. For both amorphous and fcc structures, first principles atomic and electronic structure calculations reveal striking agreement between the measured TSP and the predicted spin polarization of s-electrons. Given the disordered structure of the ternary alloy, not only do these results strongly endorse our communal understanding of tunneling through AlO x, but they also portray the key concepts that demand primary consideration in such complex systems. 1 A large part of this chapter appeared in Physical Review Letters [44]. 85

98 86 Chapter 4 Key concepts in spin tunneling 4.1 Introduction Background Right from its inception, experimental and theoretical endeavors in electron tunneling have been dedicated to (i) the understanding of the role of the electrode and barrier electronic structure, and (ii) to the various aspects concerning the nature of the electronic wave function that govern electron tunneling. Not long after it was well-established that the density of states of a superconducting electrode was directly observable in tunneling through amorphous AlO x barriers [1 3], tunneling spectroscopy to observe the influence of the electronic structure of semi-metallic electrodes was performed [4, 5]. For ferromagnetic films, using such tunneling spectroscopies, and the fact that the Zeeman splitting of the quasi-particle density of states in a superconductor could be achieved in high magnetic fields, one aspect of their electronic structure - the tunneling spin polarization (TSP) - was measured [6]. Although some preliminary effort was undertaken to study the role of the band structure of ferromagnetic films in tunneling [7], no definitive observations were made till the advent of tunnel magnetoresistance (TMR) in magnetic tunnel junctions (MTJs) [8]. Then, Yuasa et al. [9] and LeClair et al. [10] experimentally demonstrated the influence of epitaxial Fe and textured Co films on TMR and tunneling conductance, respectively. The former established the change in TMR in Fe / AlO x / Fe MTJs by growing Fe electrodes in different crystal orientations. The latter demonstrated the change in tunnel conductance of Co / AlO x / Co MTJs at bias voltages where certain bands were known to exist in the electronic structure of fcc Co. Regarding the nature of the electronic wave functions that govern the tunneling probability, despite persistent effort, our theoretical understanding of tunneling through AlO x is limited [11, 12]. This is largely due to the amorphous nature of AlO x which hinders ab-initio calculations. One aspect, the dominance of the spherically symmetric s-like electrons in tunneling through AlO x has been experimentally demonstrated [13 15]. Yuasa et al. studied the effect of a single-crystalline Cu(001) layer inserted in a Co (001) / Cu (001) / AlO x / NiFe MTJ and observed an oscillation in the TMR as a function of this Cu layer thickness. They ascribed the origin to the spin-polarized resonant states from the spherically symmetric s-like band in the Cu layer [13]. A similar conclusion was drawn in a later study where a Cr(001) layer, which has no spherically symmetric s-like states at the Fermi level (E F ), was inserted in Fe (001) / Cu (001) / AlO x / CoFe MTJs [14]. Actually, this dominance of s-like electrons in tunneling through AlO x can be anticipated if one carefully looks at the band structure of most crystalline Al 2 O 3 phases (for e.g., α, γ, and κ). Here the conduction band inevitably comprises of Al s-states [16]. Noting that the barrier heights calculated from tunneling experiments are 1 2 ev, the calculated band gap of Al 2 O 3 is 6.0 ev, and the large spatial extent of the s-electron wave functions, one may expect a large coupling constant for evanescent s-like electrons

99 4.1 Introduction 87 at the AlO x ferromagnet interface Objectives of this work Recently, spintronics has witnessed a rapid rise in the importance of amorphous ferromagnets like CoFeB. They have contributed to huge TMR in AlO x [17] and MgO [18] based MTJs. They have also been employed to observe the novel spintorque diode effect [19] and facilitated record-low switching currents in spin-torque based MTJs [20]. Although their emerging importance in spintronics is unquestionable, neither has there been a theoretical and experimental analysis of their atomic and electronic structure, nor has the impact of these properties on their TSP been investigated. In this chapter, a two fold objective is aimed at: (i) an issue which has never been investigated experimentally or computationally to date is addressed namely the TSP of an amorphous ferromagnet. We investigate why these ferromagnets exhibit such high TMR values in the first place and how one can quantitatively describe the TSP of a ternary amorphous alloy in conjunction to an amorphous barrier. Recall here that the TSP and TMR are directly related. (ii) we explore the correlation between ferromagnet morphology, its electronic structure and their combined impact on TSP. In other words, we investigate how a distinct and major structural alteration of the ferromagnetic electrode at the interface with the barrier influences the TSP. One unique aspect crystallization of amorphous CoFeB with a single high temperature anneal ( 250 C [21, 22]) is exploited to study the structural, magnetic and TSP related properties of amorphous and crystalline CoFeB in the same sample. Indeed, such control on morphology is not achievable in elemental magnetic films. Therefore, such a distinct and major structural alteration of the ferromagnetic electrode at the barrier interface has not been studied in the physics of spin-tunneling through AlO x barriers. Moreover, such crystallization and its impact on the TSP is of great relevance in MgO based MTJs. We demonstrate that amorphous Co 72 Fe 20 B 8 shows a considerably large TSP of 53%. When a single anneal is used to intentionally transform its structure to highly textured fcc, a correlated alteration of the CoFeB electronic structure is induced. Contrary to one s primary intuition, this alteration of the electronic structure manifests in an intrinsically lower TSP for fcc CoFeB as compared to that of amorphous CoFeB. To investigate the origin of the large amorphous TSP and the subsequent decrease on crystallization, we perform first-principles atomic structure calculations on amorphous and fcc CoFeB. Extended x-ray absorption fine structure (EXAFS) measurements on amorphous CoFeB are found to be consistent with the calculated amorphous structure. Remarkably, both for amorphous and fcc CoFeB, electronic structure calculations based on this calculated atomic structure exhibit a conspicuous agreement between the spin polarization (SP) of the s-electron density of states (DOS) and experimentally measured TSP. The calculations also reveal that the B sp-states get highly spin polarized and make a significant contribution to the alloy

100 88 Chapter 4 Key concepts in spin tunneling SP. We would like to emphasize that such a quantitative agreement between theory and experiment for a complex amorphous / crystalline ternary alloy has not been reported before. Moreover, given the recent development in CoFeB based spintronic devices, first principles atomic and electronic structure calculations, especially those corroborating spin polarized tunneling experiments, have not been reported yet. Furthermore, these results endorse several other earlier concepts, for example, the high sensitivity of the tunnel conductance to the ferromagnet-barrier interface [23], and the dominance of s-electrons in tunneling through AlO x barriers [13, 14]. 4.2 Experimental Results Sample preparation and measurement The high temperature anneal used to induce CoFeB crystallization stipulates another crucial requirement on our junctions, viz., the barrier properties should not change after the anneal to ensure a meaningful comparison of the TSP of amorphous (unannealed) and crystalline (annealed) CoFeB. Contrary to alternative barriers like MgO, we have previously shown that the TSP of Co and CoFe electrodes measured in Al / AlO x / Co (CoFe) junctions annealed in UHV conditions is essentially constant for anneals up to T a = 500 C [24 26]. This indicates that no relevant change occurs in AlO x barriers with respect to TSP after such high temperature anneals. Also see Chapter 7. We deposited Co 72 Fe 20 B 8 layers of various thickness on Si // SiO x and Si // SiO x / AlO x buffer layers using DC magnetron sputtering (base pressure < 10 8 mbar) at room temperature. Their structural properties were investigated using high-angle XRD (Cu K α ) after anneals at different temperatures for 30 minutes in ultrahigh vacuum conditions (pressure 10 8 mbar region during annealing). The Paul Scherrer formula [see Section 2.2.1] was used to calculate the grain size: t(å) = 1.37 FWHM cos θ (4.1) where FWHM is the full width half maximum of the XRD peak observed at 2θ [see inset in Figure 4.1(d)]. We used SQUID (Superconducting Quantum Interference Device) magnetometery to measure the magnetic properties of these layers. HRTEM was performed using a Tecnai F20 FEG microscope (FEI) fitted with a spherical aberration corrector (CEOS) having a point resolution of 1.3 Å [see Section 2.2.3] on Si // SiO x / AlO x / CoFeB (d = 700 Å or 60 Å) / Al films annealed at 450 C. A fourier transform of the HRTEM micrograph was used to measure both the interatomic distances and the angles between lattice planes. The room temperature EXAFS measurements were performed at the 7.1 station of the Daresbury Laboratory Synchrotron using fluorescence detection with a 9 element monolithic Ge detector and fitted with the Daresbury program EXCURV [see Section 2.2.2].

101 4.2 Experimental Results 89 Normalized Conductance (di/dv) (a) P: % : 0.33 mev b : : T : 0.26K (b) 1.0 P: % 0.5 : 0.35 mev b : 0.02 : T : 0.26K Bias voltage (mv) d~120å as-dep H=0 T Maki Fit H=2 T Maki Fit d~120å T a =450 C Tunneling Spin Polarization (%) Norm. Grain Size (arb.u.) Mag Mom. (µ B /TM) XRD Intensity (arb.u.) (c) d~700å 1.0 d~60å 0.5 T = 5K SQUID Anneal Temperature ( C) 2 ( ) Å 120Å As-dep 300 C 450 C d~700å 500Å 300Å 200Å 120Å 60Å 120Å 300Å d~700å (d) Anneal temperature ( C) Figure 4.1: TSP and crystallization. (a) Representative TSP measurement for an as-deposited junction with 120 Å CoFeB. (b) Similar junction after an anneal at T a = 450 C. The zero field curve ( ) shows the Al superconducting gap while the 2.0 T ( ) curve reveals the TSP of CoFeB when fit (solid lines) with Maki theory [27]. The superconducting gap ( ), orbital depairing (ξ), spin-orbit scattering (b) and temperature (T) are fit parameters. (c) TSP of CoFeB as a function of T a and d. Inset shows magnetic moment as a function of T a for 700 Å and 60 Å films. (d) The grain size perpendicular to the film plane is normalized to d and plotted as a function of T a. Insets show actual XRD data on as-deposited and annealed 700 Å and 120 Å films.

102 90 Chapter 4 Key concepts in spin tunneling Impact of CoFeB crystallization of its TSP Superconducting tunneling spectroscopy [6], which employs the Zeeman-split quasiparticle DOS of a superconductor as a spin analyzer, was used to measure the TSP. A description of this technique is given in Section 2.5.1, as well as in references [6, 28]. Figure 4.1(a) shows a representative TSP measurement for an as-deposited 120 Å CoFeB film. Regardless of the CoFeB thickness (d), for as-deposited samples we consistently measure a TSP of 53%. However, after an anneal the measured value of the TSP is strongly dependent on d and T a. As shown in Figure 4.1(b), for a junction from the same batch, an anneal at T a = 450 C prompts a reduction in the TSP. This dependence of the TSP on CoFeB thickness and T a is shown in Figure 4.1(c). Evidently, thick films (700 Å and 500 Å) show no significant change in the TSP after anneals above the crystallization temperature ( 250 C). On the contrary, the TSP of progressively thinner films decreases systematically with the thickness of the films, especially for T a = 450 C. As to the cause for this reduction in TSP, one can rule out the formation of boron oxide at the barrier-ferromagnet interface or boron diffusion into the tunnel barrier, since (a) both these processes are expected to contribute equally to the drop in TSP, regardless of CoFeB thickness, (b) no significant change in junction resistance is observed, and (c) thermodynamically, AlO x is known to be a more stable oxide. Boron segregation away from the interface can also be safely ruled out, as one might expect such a segregation to influence the TSP regardless of CoFeB thickness. These arguments also justify the use of low B content in this work. Moreover, the magnetic moment of CoFeB, independent of its thickness, does not show any significant change after annealing [see inset, Figure 4.1(c)]. If boron would segregate, one would expect the magnetic moment to asymptotically proceed towards that of a comparable Co 80 Fe 20 alloy Verification of crystallization at interface A clue to the probable reason behind this change in the TSP of thin CoFeB films can be found in x-ray diffraction (XRD) measurements on films of corresponding thickness. In Figure 4.1(d), the grain size perpendicular to the film plane, calculated using the Paul Scherrer formula, and normalized to the film thickness, is plotted as a function of T a. This plot indicates that, in progressively thinner films, the grain sizes become comparable to film thickness after annealing. For T a = 450 C and d = 120 Å, the average grain size is almost equal to the film thickness suggesting the presence of crystalline CoFeB at the interface with the AlO x barrier. This hypothesis is substantiated by high resolution transmission electron micrographs (HRTEM). Figure 4.2(a) shows a junction with a 700 Å CoFeB layer, while Figure 4.2(b) corresponds to a 60 Å CoFeB layer, both annealed at 450 C. For the 700 Å film, a close inspection of the barrier-ferromagnet interface region shows hardly any crystalline CoFeB at the interface [see lower panels of Figure 4.2(a) for a zoom-in], though we observe CoFeB

103 4.2 Experimental Results 91 Figure 4.2: Crystallization at the interface. (a) HRTEM micrograph of a Al / AlOx / CoFeB (700 A ) / Al junction after a 450 C anneal. Hardly any crystalline CoFeB is seen at the barrier-ferromagnet interface; see lower panels in (a) for magnified interface regions. (b) Similar junction, but with 60 A thick CoFeB. Contrary to the 700 A film, almost comprehensive crystallization of CoFeB is seen here, especially at the barrier-ferromagnet interface. crystallites in the bulk of the film (not shown). In sharp contrast, we observe almost comprehensive crystallization of CoFeB in the case of the 60 A film, especially at the barrier-ferromagnet interface. Together, the XRD and HRTEM data strongly advocate that thicker films (d & 500 A ) do not crystallize completely after annealing, especially at the interface with amorphous AlOx, and consequently show a TSP similar to that of as-deposited amorphous CoFeB. On the contrary, thinner films crystallize virtually completely, and the TSP of crystalline CoFeB at its interface with AlOx manifests its intrinsic value. Note that the interface sensitivity of the TSP [23] is implicitly demonstrated within this inference. Furthermore, consistent with the observations of Takeuchi et al. [29] in crystalline films, the out-of-plane grain size is limited by the film thickness, while the in-plane grain size ( A ) is similar to that observed in thicker films. As anticipated for such a Co rich composition, high angle XRD and Fourier transform (FT) of HRTEM images also confirm that CoFeB crystallizes in a highly (111) textured fcc structure.

104 92 Chapter 4 Key concepts in spin tunneling Figure 4.3: Atomic structure. (a) Representative amorphous and (b) fcc structures. (c) Calculated prdfs for Co-Co, and (d) for Fe-Co. (e) Measured and fitted k 3 weighted EXAFS oscillations on Fe and Co K edge, (f) and corresponding FT for the amorphous films.

105 4.3 Comparison of calculated and measured a-cofeb Comparison of calculated and measured a-cofeb Calculation: Molecular dynamics Having established that the lowering of the CoFeB TSP is closely related to its crystallization, we embarked on first-principles calculations using density functional theory within the generalized gradient approximation [30]. The self-consistent electronic structure and interatomic forces were calculated with the projector augmented wave method [31, 32] using the Vienna ab-initio molecular dynamics package (VASP) [33, 34]. For reliable determination of the amorphous structure, the ensemble was heated above its melting point and equilibrated in the liquid state for time periods long enough to allow diffusion beyond one lattice spacing, and then rapidly quenched to form the amorphous state. Structural and electronic properties of two 108 atom ensembles were compared to three 54 atom ensembles for further verification and statistics. It is noteworthy that ensembles without B atoms did not quench in an amorphous structure, indicating the key role played by 7 at. % B in rendering CoFeB amorphous. In the fcc case, the atoms were randomly placed in nominal positions in a fcc lattice, and then allowed to relax. The total energy of the amorphous ensembles was invariably found to be higher than that of the distorted fcc ensembles, consistent with the fact that as-deposited amorphous films crystallize after an anneal. Self-consistent DOS calculations were first carried out on a coarse k-point mesh (4 4 4, containing the Γ point). In order to obtain reliable numbers for the SP at E F, the partial DOS (of the 2 amorphous and 2 fcc configuration in the 108 atoms cubic cell) was recalculated on a k-point mesh in the full Brillouin zone (BZ). As there was no symmetry in our cells, only time-reversal symmetry could be applied and this amounted to 504 points in the cubic BZ of the 108 atoms cells. Differences between the fine and coarse mesh SPs were observed to be small. A Gaussian smearing with a width σ = 0.1 ev was applied. The dependence on σ was checked and found to be negligible Measurements: molecular dynamics vs. EXAFS Representative structures of one amorphous and one fcc ensemble are shown in Figure 4.3(a) and 4.3(b) together with the partial radial distribution functions [prdfs - Figure 4.3(c) and 4.3(d)]. Irrespective of the size of the unit cell (108 or 54 atoms), the prdfs show no significant difference in the inter- or intra-atomic coordination up to r = 5.5 Å, indicating that a 108 ensemble is of sufficient size. To gain insight in the atomic structure of amorphous films, EXAFS measurements were performed on Co and Fe K edges. The measured and fitted data are shown in Figure 4.3(e) and the corresponding FT in Figure 4.3(f). The oscillations seen in Figure 4.3(e) are characteristic of disordered solids where usually the first coordination shell is the largest contributor to the fine structure, as is evident in the single peak dominating the FT. Keeping in mind the difficulties in fitting an amorphous structure due to

106 94 Chapter 4 Key concepts in spin tunneling the large number of possible variations in the atomic arrangements, the fit to the oscillations is well within acceptable limits. More importantly, the calculated coordination number and distance to the first and second shell from our molecular dynamics simulations are in very good agreement with the fitted EXAFS data. Apart from fitting a second Fe-B shell above 3.6 Å, the fitted third coordination shells too agree fairly well with those obtained using molecular dynamics. For the second Fe-B shell, the difficulty arises from the relatively low concentrations of the two species in the compound. The peaks in the FT around Å are generally ascribed to multi-electron excitations. Based on the confirmation that the amorphous structure calculated using molecular dynamics is consistent with the measured EXAFS data, we proceeded to the electronic structure calculations. 4.4 Electronic structure and TSP Fe in strongly ferromagnetic state The calculated Co and Fe d-dos for the amorphous and the fcc alloy [see Figure 4.4(a)] points out to a strongly ferromagnetic alloy with the majority channel completely filled. Both Fe and Co are seen to be in a strongly ferromagnetic state. This is not surprising in the case of Fe considering the self-consistent density functional calculations of Schwarz et al. [36] on Co 100 x Fe x, which show that the Fe magnetic moment increases with increasing number of Co nearest neighbors, and is largest when Fe has no Fe nearest neighbors. The difference in the shape of the Co and Fe d-dos can be understood within the bonding charge transfer model of Richter et al. [35]. We will come back to the details of this changes in the Fe electronic structure and the general shape of the Fe and Co DOS in Chapter 6. Comparing the d-dos, both for Co and Fe, the Stoner gap is observed to be slightly higher and the d band width slightly lower in the amorphous case as compared to the fcc case. The d band narrowing follows from the increase in the average Co-Co and Fe-Fe coordination in the amorphous case [Figure 4.3(c) and 4.3(d)] where the first coordination shell looses 1 atom and the second coordination shell around 3.5 Å is almost completely wiped out in comparison to the fcc case. The calculated value for the alloy magnetic moment for fcc CoFeB is virtually unchanged in comparison to amorphous CoFeB, consistent with measurements shown in Figure 4.1(c) Comparison with measured TSP Considering the amorphous nature of the barrier, one might argue that k conservation is highly unlikely in tunneling through AlO x. In the first instance, if one neglects any issue related to the barrier or interface electronic structure, the spin

107 4.4 Electronic structure and TSP 95 Co d-dos Fe d-dos (a) Amor Amor FCC FCC Co d-dos Fe d-dos (c) Co s-dos Co s-dos CoFeB s-dos (b) CoFeB s-dos Energy (ev) fine k-mesh (d) g(r) B s-dos Energy (ev) am CoB fcc r (Å) B s-dos Figure 4.4: Electronic structure. (a) Calculated element-specific d-dos for Co and Fe. (b) Total s-dos calculated on a fine k-mesh for CoFeB. (c) Co s-dos and (d) B s-dos in amorphous and fcc case. Inset in (d) shows prdf for Co-B. The y-axis units in states/ev/atom and the legend in (b) applies to all plots. polarization of s-like electrons, which have been experimentally shown [13, 14] to dominate tunneling through AlO x, is the only quantity which needs consideration. Table 4.1 shows the calculated average s-electron SP at the Fermi level (E F ) for Co, Fe and B in amorphous and fcc case. Assuming that the concentration at the interface is similar to that in the bulk, we obtain the alloy SP by weighting these individual SPs with their concentrations [6]. The last two columns of Table 4.1 compare the measured TSP to the calculated SP of the alloy. Both for the amorphous

108 96 Chapter 4 Key concepts in spin tunneling Table 4.1: Calculated s-sp and measured TSP values (in %). Struc. Co Fe B avg. SP avg. SP exp. TSP without B with B a-cofeb ±0.2 53±0.5 c-cofeb ±0.5 44±0.5 and fcc case, the calculated SPs of 50.0 ± 0.2% and 41.4 ± 0.5% are in surprisingly good agreement with the measured TSP of 53 ± 0.5% and 44 ± 0.5%, respectively. Most strikingly, the difference of 9% between the two measured TSP values is directly reflected in the calculations as well, indicating that this difference might arise from the disparity in the band structure of bulk amorphous and fcc CoFeB. We would like to emphasize that, for the 5 amorphous and 2 fcc unit cells studied, the values of the element-specific and the alloy SPs shown in Table 4.1 are remarkably similar from one unit cell to another. The errors for the calculated SP in Table 4.1 are deduced from the variations in the element-specific SPs under a coarse and a fine sampling of k-space for the two 108 atom unit cells Interface bonding effects One may argue that the ground state electronic properties of CoFeB are not really fit to describe electronic transport at an amorphous / crystalline electrode amorphous barrier interface. In other words, the quantitative agreement of the average of the element-specific spin polarizations of the s-electron DOS and the measured TSP obtained here may be termed fortuitous. Interface bonding effects, which have not been included in the above calculations, are known to have pronounced effects on the TSP [37]. These may be seen as alternative explanations for the decrease in TSP observed after anneal. We have the following arguments on these issues: (i) We did estimate the impact of the stronger interface bonding expected for B and Fe as compared to Co with oxygen at the interface, using an approach similar to Kaiser et al. [38, 39]. The ratio of the Gibbs energy of formation for the oxides of Co, Fe and B was used to calculate the effect of bonding on the increased electrontransfer coefficient at the interface [38, 39]. Here too we did not see any significant deviation from the calculated SP values of Table 4.1. Distinct evidence supporting the calculated results, and the inertness of interface, is found in the measured TSP in the thicker CoFeB films (d 500 Å). For these films it is observed that interface bonding effects do not seem to induce any change in the measured TSP before and after the anneal at 450 C. This clearly indicates that the change in the TSP on crystallization does not originate from any change in the interface, and can be seen to arise entirely from the bulk band structure. (ii) Independent first principles calculations on 5 different amorphous ensembles show the exact same numbers for the element-specific spin polarization at E F within

109 4.4 Electronic structure and TSP 97 the error bars shown in Table 4.1. The size of these ensembles was varied to have 54 or 108 atoms. Similar is the case for the fcc ensembles. (iii) Given the amorphous nature of AlO x and CoFeB, interface bonding effects are rather difficult to predict and, in reality, they are an average over the configuration space at a disordered interface. More specifically, the arrangement of each atomic species in the ferromagnet with respect to the oxygen and aluminum atoms at the barrier interface over a large region is expected to vary from site to site, and consequently very difficult to computationally investigate. Additionally, the structure of amorphous AlO x, although of significant technological relevance, is still not understood well. To treat such a complex interface containing an amorphous or disordered ternary alloy with an amorphous interface is a sort of holy grail in computational spintronics. One must realize that, realistically speaking, we are far away from such calculations as they are beyond the realm of existing capabilities. Even the current calculations which employ state-of-the-art first principles molecular dynamics and density functional calculations are extremely expensive to perform for the unit cell sizes we have used. Given (1) the very good agreement between the SP of the bulk s-dos with the measured TSP, (2) the striking agreement between the predicted and measured difference in the TSP of amorphous and fcc CoFeB, and (3) the disordered structure of both the electrode and the barrier, one might wonder if a better quantitative agreement can be accomplished by going into further complexity Changes in electronic structure on crystallization Figure 4.4(b) shows the total s-dos of amorphous and fcc CoFeB obtained with the high-resolution sampling of k-space. It confirms the higher SP of the amorphous alloy as given in Table 4.1. In Figure 4.4(c), if one compares the element specific s-dos for amorphous Co (and Fe not shown) to fcc Co (and Fe), the anti-bonding s-states of fcc Co (and Fe) are pushed towards higher energy for both spin-channels. Increased s-d hybridization due to increase in the first and second shell coordination of the fcc alloy might be responsible for this [Figure 4.3(c) and 4.3(d)]. Interestingly, the decrease in the s-electron SP of the fcc alloy might be seen to primarily ensue from this spectral shift of the anti-bonding states towards higher energy, since the E F lies on the slope of the increasing majority s-dos, while lying in the deep minimum of the minority s-dos. One notices from Figure 4.4(b) that the minority DOS also shows subtle changes, which provides a secondary contribution to the change in the s-electron SP Highly spin-polarized boron sp states The impact of s-d hybridization can be also seen in the B s-dos shown in Figure 4.4(d). In our calculations we note that (1) the B sp states are highly spin polarized (s-sp > 50%; p-sp > 25%) as noted before [40], and (2) the B sites attain

110 98 Chapter 4 Key concepts in spin tunneling a small negative magnetic moment ( 0.1 µ B ) consistent with earlier work [41, 42]. This high polarization is a direct consequence of the hybridization of the B sp states with the Co/Fe d-states forming covalent bonding states below E F and anti-bonding states above it [43]. From the prdfs of Co-B [see inset Figure 4.4(d)] and Fe-B (not shown), one notes that the first coordination shell around 2.1 Å is larger in the amorphous case as compared to the fcc case. Consequently, for amorphous CoFeB, this leads to increased sp-d hybridization and the anti-bonding s-states of B are shifted to higher energy, as seen in Figure 4.4(d). Here, however, the spin polarization compared to the fcc case increases due to the lower DOS of the minority channel. We would like to emphasize that the polarization of B s-states has a direct impact on the TSP. The fifth column in Table 4.1 shows the calculated average SP of the alloy when the B atoms are considered unpolarized. The obvious disagreement with the measured TSP is an indication of the importance of highly SP B atoms at the interface. The fact that the calculated values for the SP for B are much higher than those for Co and Fe [see Table 4.1], also indicates that boron atoms getting highly spin polarized is one of the key reasons for the high TSP displayed by these alloys. 4.5 Conclusions In summary, we show that in AlO x based junctions the TSP of amorphous CoFeB is larger than that of fcc CoFeB. First-principles calculations involving the complex atomic and electronic structure of amorphous and crystalline CoFeB yield s- electron SP values in remarkable agreement with experimental TSP values, and predict highly spin-polarized B sp states contributing to the TSP. These observations not only endorse that the electronic structure of the electrode has a marked impact on tunneling, but also corroborate an intuitive and straightforward picture for the TSP of such a ternary amorphous/crystalline alloy in conjunction with an amorphous barrier in general. We believe that the present results report on novel and robust experimental and computational results which are not only of high relevance to spintronics in particular, but also to the understanding of magnetism of alloys containing non-magnetic elements and to spin-tunneling in general.

111 BIBLIOGRAPHY 99 Bibliography [1] I. Giaever, Energy gap in superconductors measured by electron tunneling. Phys. Rev. Lett. 5, 147 (1960) [2] I. Giaever, Electron tunneling between two superconductors. Phys. Rev. Lett. 5, 464 (1960). [3] J. Nicol, S. Shapiro, and P. H. Smith, Direct measurement of the superconducting energy gap. Phys. Rev. Lett. 5, 461 (1960) [4] L. Esaki and P. Stiles, Study of electronic band structure by electron tunneling: Bismuth. Phys. Rev. Lett. 14, 902 (1965) [5] J. Hauser and L. Testardi, tunneling and band structure of semimetals. Phys. Rev. Lett. 20, 12 (1968) [6] R. Meservey and P. Tedrow, Spin-polarized electron tunneling. Phys. Rep. 238, 173 (1994) , 4.2.2, [7] J. M. Rowell, Electron tunneling into films of Ni, Pd, and NiPd alloys. J. Appl. Phys. 40, 1211 (1969) [8] J. S. Moodera, L. R. Kinder, T. M. Wong, and R. Meservey, Large magnetoresistance at room temperature in ferromagnetic thin film tunnel junctions. Phys. Rev. Lett. 74, 3273 (1995) [9] S. Yuasa, T. Sato, E. Tamura, Y. Suzuki, H. Yamamori, K. Ando, and T. Katayama, Magnetic tunnel junctions with single crystalline electrodes: a crystal anisotropy of tunnel magneto resistance. Euro. Phys. Lett. 52, 344 (2000) [10] P. LeClair, P. LeClair, J. T. Kohlhepp, C. H. van de Vin, H. Wieldraaijer, H. J. M. Swagten, W. J. M. de Jonge, A. H. Davis, J. M. MacLaren, J. S. Moodera, and R. Jansen, Band structure and density of states in Co based magnetic tunnel junctions. Phys. Rev. Lett. 88, (2002) [11] I. I. Oleinik, E. Y. Tsymbal, and D. G. Pettifor, Structural and electronic properties of Co/Al 2 O 3 /Co magnetic tunnel junctions from first principles. Phys. Rev. B 62, 3952 (2000) [12] S. Zhang, and P. M. Levy, Models for magnetoresistance in tunnel junctions. Eur. Phys. J. B 10, 599 (1999) [13] S. Yuasa, T. Nagahama, and Y. Suzuki, Spin polarized resonant tunneling in magnetic tunnel junctions. Science 297, 234 (2002) , 4.1.2, 4.4.2

112 100 Chapter 4 Key concepts in spin tunneling [14] T. Nagahama S. Yuasa, E. Tamura, and Y. Suzuki, Spin dependent tunneling in magnetic tunnel junctions with a layered antiferromagnetic Cr(001) spacer Role of band structure and interface scattering. Phys. Rev. Lett. 95, (2005) , 4.1.2, [15] M. Münzenberg, and J. S. Moodera Superconductor-ferromagnet tunneling measurements indicate sp-spin and d-spin currents. Phys. Rev. B 70, R (2004) [16] For α-al 2 O 3, please refer to A. G. Marinopoulos, and C. Elsässer, Acta. Mater. 48, 4375 (2000) and S. E Kulkova, L. Yu. Zagorskaya, and I. R. Shein, Russ. Phys. J. 48, 1127 (2005). For γ-al 2 O 3, please refer to E. Menéndez- Proupin, and G. Gutiérrez, Phys. Rev. B 72, (2005). For κ-al 2 O 3, please refer to Y. Yourdshahyan, and C. Ruberto, L. Bengtsson, B. I. Lundqvist, Phys. Rev. B 56, 8553 (1997) [17] H. X. Wei, Q. H. Qin, M. Ma R. Sharif, & X. F. Han, 80% tunneling magnetoresistance at room temperature for thin Al O barrier magnetic tunnel junction with CoFeB as free and reference layers. J. Appl. Phys. 101, 09B501 (2007) [18] D. D. Djayaprawira, K. Tsunekawa, M. Nagai, H. Maehara, S. Yamagata, N. Watanabe, S. Yuasa, Y. Suzuki, and K. Ando, 230 % room temperature magnetoresistance in CoFeB / MgO / CoFeB magnetic tunnel junctions. Appl. Phys. Lett. 86, (2005) [19] A. A. Tulapurkar, Y. Suzuki, A. Fukushima, H. Kubota, H. Maehara, K. Tsunekawa, D. D. Djayaprawira, N. Watanabe, and S. Yuasa, Spin torque diode effect in magnetic tunnel junctions. Nature 438, 339 (2005) [20] J. Hayakawa, S. Ikeda, Y. M. Lee, R. Sasaki, T. Meguro, F. Matsukura, H. Takahashi, and H. Ohno, Current-driven magnetization switching in CoFeB/MgO/CoFeB magnetic tunnel junctions. Jpn. J. Appl. Phys. 44, L1267 (2005) [21] P. V. Paluskar, J. T. Kohlhepp, H. J. M. Swagten, B. Koopmans, R. Wolters, H. Boeve, and E. Snoeck, Influence of interface structure on the tunneling spin polarization of CoFeB alloys. J. Phys. D: Appl. Phys. 40, 1234 (2007) [22] S. Yuasa, Y. Suzuki, T. Katayama, and K. Ando, Characterization of growth and crystallization processes in CoFeB/MgO/CoFeB magnetic tunnel junction structure by reflective high-energy electron diffraction. Appl. Phys. Lett. 87, (2005)

113 BIBLIOGRAPHY 101 [23] P. LeClair, J. T. Kohlhepp, H. J. M. Swagten, and W. J. M. de Jonge, Interfacial density of states in magnetic tunnel junctions. Phys. Rev. Lett. 86, 1066 (2001) , [24] C. H. Kant, J. T. Kohlhepp, H. J. M. Swagten, and W. J. M. de Jonge, Intrinsic thermal robustness of tunneling spin polarization in Al/Al 2 O 3 /Co junctions. Appl. Phys. Lett. 84, 1141 (2004) [25] P. V. Paluskar, C. H. Kant, J. T. Kohlhepp, A. T. Filip, H. J. M. Swagten, B. Koopmans, and W. J. M. de Jonge, Co 72 Fe 20 B 8 : Structure, magnetism, and tunneling spin polarization. J. Appl. Phys. 97, 10C925 (2005). [26] S. S. P. Parkin, C. Kaiser, A. Panchula, P. M. Rice, B. Huges, M. Samant, and S.-H. Yang, Giant tunneling magnetoresistance at room temperature with MgO (100) tunnel barriers. Nature Mater. 3, 862 (2004) [27] K. Maki, Pauli paramagnetism and superconducting state. II. Prog. Theor. Phys. 32, (1964). 4.1 [28] C. H. Kant, Probing spin polarization: point-contacts and tunnel junctions. PhD Thesis, Eindhoven University of Technology (2005) [29] T. Takeuchi, K. Tsunekawa, Y.-s. Choi, Y. Nagamine, D. D. Djayaprawira, A. Genseki, Y. Hoshi, and Y. Kitamoto, Crystallization of amorphous CoFeB ferromagnetic layers in CoFeB/MgO/CoFeB magnetic tunnel junctions. Jpn. J. Appl. Phys. 46, L623 (2007) [30] J. P. Perdew, J. A. Chevary, S. H. Vosko, K. A. Jackson, M. R. Pederson, D. J. Singh, and C. Fiolhais, Atoms, molecules, solids, and surfaces: Applications of the generalized gradient approximation for exchange and correlation. Phys. Rev. B, 46, 6671 (1992) [31] P. E. Blöchl, Projector augmented-wave method. Phys. Rev. B, 50, (1994) [32] G. Kresse, and D. Joubert, From ultrasoft pseudopotentials to the projector augmented-wave method. Phys. Rev. B 59, 1758 (1999) [33] G. Kresse, and J. Furthmüller, Efficient iterative schemes for ab initio totalenergy calculations using a plane-wave basis set. Phys. Rev. B, 54, (1996) [34] G. Kresse, and J. Hafner, Ab-initio molecular dynamics for liquid metals. Phys. Rev. B 47, 558 (1993)

114 102 Chapter 4 Key concepts in spin tunneling [35] R. Richter, and H. Eschrig, LCAO-CPA for disordered 3d transition metal alloys. Magnetic moment formation in NiCu and FeCo. Phys. Scrip. 37, (1988) [36] K.Schwarz, P. Mohn, P. Blaha, and J. Kübler, Electronic and magnetic structure of BCC Fe-Co alloys from band theory. J. Phys. F: Met. Phys. 14, (1984) [37] E. Y. Tsymbal, and K. D. Belashchenko, Role of interface bonding in spin dependent tunneling. J. Appl. Phys., 97, 10C910 (2005) [38] C. Kaiser, S. van Dijken, S.-H. Yang, H. Yang, and S. S. P. Parkin, Role of tunneling matrix elements in determining the magnitude of the tunneling spin polarization of 3d transition metal ferromagnetic alloys. Phys. Rev. Lett. 94, (2005) [39] C. Kaiser, A. F. Panchula, and S. S. P. Parkin, Finite tunneling spin polarization at the compensation point of rare earth metal transition metal alloys. Phys. Rev. Lett. 95, (2005) [40] R. Coehoorn, unpublished [41] J. Hafner, M. Tegze, and Ch. Becker, Amorphous magnetism in Fe-B alloys: First-principles spin-polarized electronic-structure calculations. Phys. Rev. B 49, (1994) [42] H. Tanaka, S. Takayama, M. Hasegawa, T. Fukunaga, U. Mizutani, A. Fujita, and K. Fukamichi, Electronic structure and magnetism of amorphous Co 1 x B x alloys. Phys. Rev. B 47, (1993) [43] J. Kanamori, Interplay between electronic structure and corelation through the s-d mixing in transition metal systems. Prog. Theo. Phys. Supp. 101, 1 (1990) [44] P. V. Paluskar, G. A. de Wijs, J. J. Attema, S. Fiddy, E. Snoeck, J. T. Kohlhepp, H. J. M. Swagten, R. A. de Groot, and B. Koopmans, Spin tunneling in junctions with disordered ferromagnets. Phys. Rev. Lett. 100, (2008). 1

115 Chapter 5 Impact of interface crystallization on inelastic tunneling The case of Al/AlO x /Co 72 Fe 20 B 8 Abstract: In this chapter 1, we report the change in inelastic electron tunneling spectra (IETS) for Al / AlO x / CoFeB / Al junctions when the structure of CoFeB at its interface with AlO x is intentionally changed from quasi-amorphous to highly textured fcc. While for the quasi-amorphous interface there are signs of the size quantization of magnons, the spectra for the fcc interface show distinct excitations at bias voltages associated with known surface magnon modes in fcc Co. These results demonstrate that IETS can be used as a tool to probe distinct structural changes of the magnetic electrode in tunnel junctions. 1 A large part of this chapter appeared in Applied Physics Letters [38]. 103

116 104 Chapter 5 Impact of interface crystallization on inelastic tunneling 5.1 Introduction Background: Interface scattering Inelastic electron scattering processes, especially those involving magnetic excitations, have great influence on electronic transport in spintronic devices. In tunnel junctions, exchange scattering with localized impurities at the barrier interface are known to cause zero bias anomalies [1]. Applebaum introduced a s-d exchange term in the hamiltonian and calculated a contribution to the conductance from spin flip scattering. This gives a bias independent positive contribution in zero field to the conductance and a contribution from Kondo scattering which gives a logarithmic peak in zero field [2 7]. Applebaum s theory on the origin of these anomalies has been applied to tunnel junctions [8], and has been the key to explaining Kondo scattering at a single magnetic impurity [9 11]. Another important process considered by Appelbaum [7] which is highly relevant to magnetic tunnel junctions (MTJs) involves inelastic tunneling of hot electrons by exciting magnons at the barrierelectrode interface. Since such a spin-flip scattering process provides spin-mixing contributions to the total conductance, it is known to decrease tunnel magnetoresistance (TMR) in MTJs [12, 13] Background: Inelastic electron tunneling spectroscopy (IETS) IETS is a powerful tool to isolate and identify excitation spectra from specific contributions to tunneling [14, 15], largely due to the fact that such processes have discrete threshold energies which result in peaks in d2 I [16]. It has been used to dv 2 observe that tunneling electrons excite phonons [16] and magnons [17] in the barrier. Indeed, specifically for MTJs, IETS was employed by Moodera et al. [13] to explain the decrease in TMR with increasing applied bias and by Nagahama et al. to confirm the formation of quantum well states in single crystalline Fe electrodes [18]. Since exchange scattering with a magnon is directly related to the interfacial magnon modes, which in turn depend on the interface structure, IETS can be used to probe structural changes at the barrier-ferromagnet interface. No such observation has been reported as yet, presumably due to the difficulty of establishing such a marked structural change at the interface. 5.2 This work In this chapter, we induce a distinct structural change in the ferromagnetic electrode at its interface with AlO x, and thereafter probe the changes in the magnon spectrum using IETS. CoFeB is used as a ferromagnetic electrode primarily because the asdeposited quasi-amorphous/nanocrystalline layer can be crystallized into a highly textured fcc layer by a single anneal on the same sample [19]. This allows for a straightforward comparison and a possibility of identifying specific contributions to

117 5.3 Experimental Methods 105 inelastic tunneling. Moreover, such a crystallization is of significant technological importance for MTJs based on MgO barriers too. From our IETS spectra, we notice that for amorphous electrodes, the small grain size at the interface might induce a low energy cutoff in the magnon spectrum due to size quantization effects. On crystallization in the fcc (111) texture, the increase in grain size lifts the size quantization and, we observe the appearance of a distinct peak in the spectra which might be directly related to known magnon excitations in single crystalline fcc Co. 5.3 Experimental Methods Sample preparation and measurement We prepared Al / AlO x / Co 72 Fe 20 B 8 / Al junctions with CoFeB layer thickness (60 Å) specifically chosen to maximize their crystallization, particularly at the AlO x interface after a single annealing. One set of junctions from the same batch was annealed at T a = 450 C in ultra high vacuum (pressure < 10 8 mbar during annealing). For the IETS measurements, we used a standard lock-in technique which was discussed in Section Verification of crystallization at interface To verify that CoFeB crystallized at the AlO x interface, we performed high-resolution transmission electron microscopy (HRTEM) on as-deposited and annealed samples. HRTEM was performed using a Tecnai F20 FEG microscope (FEI) having a point resolution of 1.3 Å [see Section 2.2.3]. The sample stack used was Si // SiO x / AlO x / CoFeB (d = 700 Å or 60 Å) / Al films annealed at 450 C. Figure 5.1(a) shows a junction in the as-deposited state, and (b) after an annealing at T a = 450 C. For the as-deposited junction, a close inspection shows hardly any crystalline CoFeB at the AlO x interface. On the contrary, in the case of an annealed junction, we observe almost comprehensive crystallization of CoFeB in a fcc (111) texture, particularly at its interface with AlO x. The bottom Al electrode is observed to be crystalline in both the as-deposited and annealed junctions. 5.4 Experimental Results IETS spectra: Phonon modes Figure 5.2 shows representative IETS spectra for (a) an Al / AlO x / Al junction, (b) an as-deposited Al / AlO x / CoFeB / Al junction, and (c) a similar CoFeB junction annealed at T a = 450 C. One notices that, in general, the intensity of d2 I and di dv 2 dv [Figure 5.2(f)] at positive bias, i.e., when electrons tunnel into the top electrode,

118 106 Chapter 5 Impact of interface crystallization on inelastic tunneling Figure 5.1: Does CoFeB crystallize at the interface? HRTEM on Al / AlO x / CoFeB (60 Å) junction before (a) and after (b) 450 C anneal (see lower panels for zoom-in). is larger than that at a corresponding negative bias. This is generally attributed to barrier asymmetry [23]. The sharp peaks around ±3 mv in all three junctions (Figure 5.2(a)-5.2(c), follow the arrows in 5.2(a) as guides to the eye) are generally assigned to the zero bias anomaly. The di [Fig. 5.2(f)] shows a sharp dip due to dv this anomaly. In the Al / AlO x / Al junction [Figure 5.2(a)], apart from these sharp peaks, two sets of distinct shoulders can be seen: those around ±22 and ±33 mv correspond to Al TA and LA phonon modes, respectively [24, 25], and a sharp peak around ±116 mv corresponding to the OH bending mode of aluminum hydroxide [24, 25]. Generically, the IETS spectra can be composed of contributions which are both symmetric and asymmetric with respect to the polarity of the bias voltage. This depends on the actual physical location of the excitations [26]. The presence of Al phonons around mv at positive and negative bias indicates that their creation and annihilation by an electron tunneling into or out of the Al electrode has almost equal probability. Such symmetry under bias reversal has also been observed by Han et al [27]. Although both sets of phonons, Al (22 33 mv) and OH (116 mv) are also seen in the as-deposited CoFeB junction [Figure 5.2(b)], they are not observed in the annealed CoFeB junction [Figure 5.2(c)]. The absence of the OH phonon after annealing has been noted before [28]. Presently, there is no insight in the absence

119 5.4 Experimental Results 107 Al-AlO x -Al (as-dep.) 3 mv 33 mv 22 mv 116 mv (a) 1.0 d 2 I / dv 2 (d) 33 mv 33 mv d 2 I / dv 2 (arb.u.) Al-AlO x -CoFeB (as-dep.) Al-AlO x -CoFeB (T a =450 C) 22 mv mv T a =450 C Bias Voltage (mv) Bias Voltage (mv) (b) (c) d 2 I / dv 2 Norm. di/dv (e) (f) 22 mv as-dep. 33 mv Figure 5.2: Inelastic electron tunneling spectroscopy. Representative IETS spectra for (a) as-deposited Al / AlO x / Al and (b) as-deposited Al / AlO x / CoFeB. The dotted box indicates magnified regions in (d) and (e). Note the x-axis scale breaks. (c) shows spectra for the annealed junction. (f) The di dv for as-deposited and annealed CoFeB junction. The arrows are guides to the eye. of both these sets of peaks in the annealed junctions. However, from past experiments [19, 29] one might conclude that their absence has no impact on the tunneling spin polarization of AlO x based junctions. Also, one might wonder if the absence of the Al phonons (22 33 mv) in the annealed CoFeB junction [Figure 5.2(c)] might be reflected in the superconducting properties of the Al electrode, presumably ensuing from structural or compositional changes in the films. However, we found no significant change in the superconducting gap, orbital depairing and spin-orbit scattering of superconducting Al electrodes at 0.27 K IETS spectra: Magnon modes Turning to magnetic excitations, on a closer look at the Al phonon region for the Al / AlO x / Al junction [Figure 5.2(d)], one does not observe any significant asymmetry in the intensity of the peaks under bias reversal. On the contrary, in the case of the as-deposited CoFeB junction [Figure 5.2(e)], we find that the intensity of the

120 108 Chapter 5 Impact of interface crystallization on inelastic tunneling peak at positive bias (+33 mv) is almost twice as large as that at negative bias (-33 mv). This asymmetry can be more clearly noticed if one looks at the odd and even d2 I dv 2 [16] d 2 I dv (even/odd) = d2 I 2 dv (+V ) ± 2 d2 I ( V ) (5.1) dv 2 d The even part of 2 I enhances the symmetric features, whereas these symmetric dv 2 features cancel out in the odd part, leaving the asymmetric contributions with respect to bias polarity clearly portrayed. Such techniques to look at symmetric and asymmetric features in the conductance and its derivaties were commonly used in the [see, for example, [6, 30]]. These odd and even contributions are shown in Figure 5.3. For the as-deposited CoFeB junction, while the peaks around 22 and 33 mv are readily identified in the even spectra [see square symbols in Figure 5.3(c) and 5.3(a)], one would expect them to disappear in the odd spectra if only Al phonons were involved [see square symbols in Figure 5.3(d) and 5.3(b)]. Instead, the odd spectra show a pronounced residual shoulder around 31 mv. This clearly suggests that in addition to Al phonon-assisted tunneling, for positive bias there is an added contribution to the tunnel conductance which has a threshold around +31 mv. One such possible contribution can be the onset of a sharp conduction band above the Fermi level of CoFeB which increases the phase space for the tunneling electrons at this bias. However, one does not expect such sharp changes in the electronic density of states (DOS) for a quasi-amorphous ternary alloy [19]. An alternative explanation for this higher scattering intensity is magnon-assisted tunneling. One may anticipate that in a nanocrystalline material, as the grain size decreases, the coherence length of a magnon is increasingly limited [12] leading to size quantization and the appearance of a low energy cut-off in the magnon DOS [31]. Given that amorphous alloys also follow the spin wave dispersion relation [33] ω k = Dk 2, (5.2) a simple first order estimate of this low energy cut-off can be calculated as E lc = D(k 2 xmin + k 2 ymin + k 2 zmin) 3D(π 2 /d 2 ) (5.3) where k = π/d, and D is the exchange stiffness constant. For E lc = 31 mv, and a grain size d Å calculated using the Scherrer formula on x-ray diffraction measurements, we obtain D mev Å 2. This value is in good agreement with 200 mev Å 2 measured for amorphous Co 80 B 20 [32] and 185 mev Å 2 for amorphous Co 80 P 20 [33] Size quantization of magnon modes As the grain size is expected to increase after crystallization, one might expect suppression of the size quantization effect. Open circles in Figure 5.3(b) show the odd spectra for the annealed CoFeB junction. Indeed, one notices that the peak around 31 mv is replaced by very small features at bias voltages above 20 mv, indicating

121 5.4 Experimental Results 109 (c) even (d 2 I/dV 2 ) (arb. u.) as-deposited 33 mv 22 mv (a) annealed 10 mv (d) odd (d 2 I/dV 2 ) (arb. u.) 31 mv (b) (e) Bias Voltage (mv) Figure 5.3: Even and odd IETS spectra. (a) even and (b) odd spectra for an as-deposited ( ) and annealed ( ) Al / AlO x / CoFeB junction with insets (c) and (d) showing magnified Al phonon region. Inset (e) shows comparison of the odd zero bias anomaly region. The y-axis intensities are scaled to enable comparison.

122 110 Chapter 5 Impact of interface crystallization on inelastic tunneling that the quantization due to small grain sizes is lifted by the annealing. Remarkably, one also notices the appearance of a very distinct peak around 10 mv. Phonons of CoO [27], Fe 3 O 4 [34] and B 2 O 3 [35] have been measured at much higher energies (> 45 mev). Thus, at this energy, one can rule out the formation of transition metal or boron oxide at the interface which leads to inelastic phonon-assisted tunneling. This argument is substantiated by the fact that we do not observe any significant post-anneal change in junction resistance. Moreover, the tunneling spin polarization of these junctions with thick CoFeB films ( 500 Å) does not change after the annealing [19] and one does not find a strong argument as to why oxide formation should occur for thinner films. Furthermore, the presence of a sharp conduction band edge just above the fermi level of fcc CoFeB (111) contributing to enhanced conductance in such a disordered ternary alloy is highly unlikely. Band structure calculations are concomitant with this argument [19]. We tentatively ascribe this peak to magnon excitations at the AlO x CoFeB interface. Such excitations have also been seen in single crystalline fcc Co (111) around a bias energy of 9 13 mv [36, 37]. The strong similarities with the present results are endorsed by the fact that CoFeB crystallizes in highly textured (111) fcc structure. In agreement with Balashov et al., the strong peak in the positive direction indicates that the magnon creation operator for an electron tunneling into the ferromagnet has a much larger expectation value than the corresponding coefficient for the magnon annihilation operator Zero bias anomaly Parenthetically, we look the zero bias anomaly peak which appears around 2-4 mv in the odd spectra [see Figure 5.3(e)]. As compared to the as-deposited CoFeB junction and the Al / AlO x / Al junction (not shown), this peak is much sharper and shifted to lower energies for the annealed CoFeB junction. This post-anneal change in the peak might be due to the rearrangement of the localized magnetic impurity states in the barrier. One might wonder if the shift allows distinction between impurity assisted spin-flip tunneling [8] and magnon-assisted tunneling [7]. Experiments involving the dependence of the peak position on external magnetic fields at low temperatures may shed light on this issue. 5.5 Summary In summary, we show that in Al / AlO x / CoFeB based junctions, the IETS spectra show sharp contrast depending on the structure of CoFeB at the interface. For amorphous CoFeB at the interface, we see indications of size quantization of the magnons. For fcc CoFeB at the interface, we see distinct excitations around 10 mv which could also be related to magnon-assisted spin flip tunneling. We demonstrate that IETS is a powerful tool to investigate the impact of interface structure changes in MTJs.

123 BIBLIOGRAPHY 111 Bibliography [1] A. F. G. Wyatt, Anomolous density of states in normal Tantalum and Niobium. Phys. Rev. Lett. 13, 401 (1964) [2] J. Appelbaum, s d exchange model of zero-bias tunneling anomolalies. Phys. Rev. Lett. 17, 91 (1966) [3] P. W. Anderson, Localized magnetic states and fermi-surface anomalies in tunneling. Phys. Rev. Lett. 17, 95 (1966). [4] J. Appelbaum, Exchange model of zero-bias tunneling anomolies. Phys. Rev. 154, 633 (1967). [5] J. Appelbaum, J. C. Phillips, and G. Tzouras, Microscopic theory of tunneling anomalies. Phys. Rev. 160, 554 (1967). [6] J. Appelbaum, and L. Y. L. Shen, Zero-bias conductance peak anomoly of Ta-I -Al tunnel junctions at 0.3K and 90kG. Phys. Rev. B 5, 544 (1972) [7] J. Appelbaum, Electron-magnon effects in ferromagnetic tunnel junctions. Phys. Rev. B 183, 553 (1972) , [8] L. Shen and J. Rowell, Zero-bias tunneling anomalies - temperature, voltage and magnetic field dependence. Phys. Rev. 165, 566 (1968) , [9] S. Gregory, Expterimental observation of scattering of tunneling elecron by a single magnetic moment. Phys. Rev. Lett. 68, 2070 (1992) [10] V. Madhavan W. Chen, T. Jamneala, M. F. Crommie, and N. S. Wingreen, Tunneling into a single magnetic atom - spectroscopic evidence of the Kondo resonance. Science 280, 567 (1998). [11] A. Heinrich, J. Gupta, C. Lutz, and D. Eigler, Single-atom spin-flip spectroscopy. Science 306, 466 (2004) [12] S. Zhang, P. M. Levy, A. C. Marley, and S. S. P. Parkin, Quenching of magnetoresistance by hot electrons in magnetic tunnel junctions. Phys. Rev. Lett. 79, 3744 (1997) , [13] J. S. Moodera, J. Nowak, and R. J. M. van de Veerdonk, Interface magnetism and spin wave scattering in ferromagnet-insulator-ferromagnet tunnel junctions. Phys. Rev. Lett. 80, 2941 (1998) , [14] B. Stipe, M. Rezaei, and W. Ho, Single-molecule vibrational spectroscopy and microscopy. Science 280, 1732 (1998)

124 112 Chapter 5 Impact of interface crystallization on inelastic tunneling [15] A. Heinrich, C. Lutz, J. Gupta, and D. Eigler, Molecule cascades. Science 298, 1381 (2002) [16] J. Lambe and R. Jaklevic, Molecular vibrational spectra by inelastic electron tunneling. Phys. Rev. Lett. 17, 1139 (1966) , [17] D. Tsui, R. Dietz, and L. Walker, Multiple magnon excitations in NiO by electron tunneling. Phys. Rev. Lett. 27, 1729 (1969) [18] T. Nagahama, S. Yuasa, Y. Suzuki, and E. Tamura, Quantum-well effect in magnetic tunnel junctions with ultrathin single-crystal Fe(100) electrodes. Appl. Phys. Lett. 79, 4381 (2001) [19] P. V. Paluskar, G. A. de Wijs, J. J. Attema, S. Fiddy, E. Snoeck, J. T. Kohlhepp, H. J. M. Swagten, R. A. de Groot, and B. Koopmans, Spin tunneling in junctions with disordered ferromagnets. Phys. Rev. Lett. 100, (2008). 5.2, 5.4.1, 5.4.2, [20] P. K. Hansma, Inelastic electron tunneling. Phys. Rep. 30, 145 (1977). [21] D. G. Walmsley, and J. L. Tomlin, Compilation for inelastic electron tunneling spectra for molecules chemisorbed on metal oxides. Prog. Surf. Sci. 18, 247 (1987). [22] K. W. Hipps, and U. Mazur, Inelastic electron tunneling spectroscopy. Handbook of Vibrational Spectroscopy, (John Wiley and Sons, Chichester) (2002). [23] P. LeClair, J. T. Kohlhepp, H. J. M. Swagten, and W. J. M. de Jonge, Interfacial density of states in magnetic tunnel junctions. Phys. Rev. Lett. 86, 1066 (2001) [24] J. Klein, and J. Leger, Tunneling measurement of phonon spectrum in granular Al. Phys. Lett. A 28, 134 (1968) [25] A. L. Geiger, B. S. Chandrashekhar, and J. G. Adler, Inelastic electron tunneling in Al-Al-oxide-metal systems. Phys. Rev. 188, 1130 (1969) [26] E. L. Wolf, Principles of electron tunneling spectroscopy, (Oxford university press, New York) (1989) [27] X.-F. Han J. Murai, Y. Ando, H. Kuboto, and T. Miyazaki, Inelastic magnon and phonon excitations in AlCo - AlCoO - Al tunnel junctions. Appl. Phys. Lett. 78, 2533 (2001) , [28] Y. Ando, J. Murai, H. Kuboto, and T. Miyazaki, Magnon-assisted inelastic excitation spectra of a ferromagnetic tunnel junction. J. Appl. Phys. 87, 5209 (2000)

125 BIBLIOGRAPHY 113 [29] C. H. Kant, J. T. Kohlhepp, H. J. M. Swagten, and W. J. M. de Jonge, Intrinsic thermal robustness of tunneling spin polarization in Al/Al 2 O 3 /Co junctions. Appl. Phys. Lett. 84, 1141 (2004) [30] J. M. Rowell, W. L. McMillan, and W. L. Feldmann, Phonon emission and self-energy effects in normal metal tunneling. Phys. Rev. 180, 658 (1969) [31] K. Yakushiji, F. Ernult, H. Imamura, K. Yamane, S. Mitani, K. Takanashi, S. Takahashi, S. Maekawa, and H. Fujimori, Enhanced spin accumulation and novel magnetotransport in nanoparticles. Nature Mater. 4, 57 (2005) and reference 33 therein [32] H. Watanabe, H. Morita, and H. Yamauchi, Magnetic properties of amorphous Co-B alloys. IEEE. Trans. Mag. 14, 944 (1978) [33] H. Mook, N. Wakabayashi, and D. Pan, Magnetic excitations in the amorphous ferromagnet Co 4 P. Phys. Rev. Lett. 34, 1029 (1975) , [34] L. V. Gasparov, D. B. Tanner, D. B. Romero, H. Berger, G. Margaritondo, and L. Forró, Infrared and Raman studies of the Verwey transition in magnetite. Phys. Rev. B 62, 7939 (2000) [35] Z. Wang Y. Zhao, P. Lazor, H. Annersten, and S. K. Saxena, In-situ pressure Raman spectroscopy and mechanical stability of superhard boron suboxide. Appl. Phys. Lett. 86, (2001) and references therein [36] T. Balashov, A. Takács, W. Wulfhekel, and J. Kirschner, Magnon excitation with spin-polarized scanning tunneling microscopy. Phys. Rev. Lett. 97, (2006) [37] R. Vollmer, M. Etzkorn, P. S. Anil Kumar, H. Ibach, and J. Kirschner, Spinpolarized electron energy loss spectroscopy of high energy, large wave vector spin waves in ultrathin fcc Co films on Cu(001). Phys. Rev. Lett. 91, (2003) [38] P. V. Paluskar, F. L. Bloom, E. Snoeck, J. T. Kohlhepp, H. J. M. Swagten, and B. Koopmans, Impact of interface crystallization on inelastic tunneling in Al/AlO x /CoFeB. Appl. Phys. Lett. 91, (2007). 1

126 114 Chapter 5 Impact of interface crystallization on inelastic tunneling

127 Chapter 6 Correlation between magnetism and TSP The case of Co 80 x Fe x B 20 Abstract: Although the electrons at the Fermi level (E F ) of a transition metal ferromagnet are predominantly of localized spin-down d-character, electronic transport in spintronic devices is dominated by electrons of delocalized spin-up s-like character. Such is also the case in magnetic tunnel junctions where electrons tunnel between two ferromagnets separated by an insulator. In this chapter 1, we report a correlation between the spin polarization of these tunneling electrons (TSP) and the magnetic moment of amorphous CoFeB alloys. Such a correlation is surprising since the TSP involves s-like electrons close to E F, while the magnetic moment mainly arises due to all d-electrons below E F. So far understanding this correlation remained experimentally and theoretically unaddressed. We show that magnetic dichroism provides clear and crucial evidence that such a correlation may exist, and demonstrate the tunability of the electronic and magnetic properties of CoFeB alloys which are of extreme relevance to spintronics. 1 This chapter is currently under review. 115

128 116 Chapter 6 Correlation between magnetism and TSP 6.1 Background At the very foundation of spintronics lie the facts that the conduction electrons in transition metal ferromagnets possess high mobilities and that they get highly spin-polarized as a consequence of their interaction with localized d-electrons [1]. In magnetic tunnel junctions, these s-like electrons dominate the tunneling current and are primarily responsible for the tunneling magnetoresistance effect [2 5]. Early experiments to measure the spin polarization of these tunneling electrons (TSP) in Ni 1 x Fe x alloys yielded the unexpected result that the alloy magnetic moment (µ alloy ) as well as their TSP displayed the Slater-Pauling (S P) behavior [6]. The S P behavior of µ alloy [see Figure 6.1(a)] is the well-known deviation from a linear change resulting in a maximum [7, 8] as the alloy composition changes. While this non-monotonous behavior of µ alloy is very commonly observed in transition metal compounds, their TSP exhibiting a similar curve is very surprising. This surprise stems from the fact that, while µ alloy is an integral over all states below the Fermi level (E F ) and is dominated by d-electrons, the TSP originates from transport of s-like electrons close to E F. The existence of such a correlation between µ alloy and TSP would allow the engineering and the tuning of magnetic and electronic transport properties of ferromagnetic alloys for application in spintronic devices. This correlation has been observed only occasionally in experiments [9 14]. However, it is still completely unknown if this correlation should be expected at all. Moreover, the understanding of such a correlation has been neither experimentally nor theoretically addressed, making this a fundamental, highly debated and long-standing issue in spintronics. We believe that a combined study of the element-specific electronic structure of the d-bands and the s-electron dominated TSP is the key to probe and understand this correlation. 6.2 This work and the relevance to understanding CoFeB In this chapter, we demonstrate the S P behavior of both the TSP and µ alloy of amorphous Co 80-x Fe x B 20 alloys. The measured curves of both these properties show distinct similarity in trend and provide an undisputable hint to this correlation. Remarkably, using a very simple phenomenological equation which assumes µ alloy to be directly proportional to the TSP we can estimate the alloy TSP within 5% of its measured value. This strongly supports the conjecture that a direct correlation between µ alloy and TSP may definitely exist in ferromagnetic alloys. CoFeB alloys are specifically chosen to address this issues since: (i) being amorphous, they are highly insensitive to the miscibility of their constituents. (ii) Contrary to most crystalline alloys, their atomic structure does not undergo structural transitions with their composition on the microscopic scale. Both these distinctions allow easy experimental access to their characteristic prop-

129 6.3 Sample preparation and measurement 117 erties. (iii) Given their unquestionable importance in spintronics today [15, 16], and their complex ternary amorphous nature, a comprehensive effort to understand their intrinsic properties remains to be embarked upon. Here, we report a combined investigation of both the above mentioned issues: an intuitive understanding of the correlation between µ alloy and TSP is provided together with a detailed insight on the various aspects of Co 80-x Fe x B 20 electronic structure. Since the basic mechanisms for this correlation must involve the electronic structure of the d-bands, we use x-ray absorption (XAS) and magnetic circular dichroism (XMCD) to probe their properties. These techniques demonstrate a direct observation of the S P behavior for the orbital (m o ) and spin (m s ) moments, as well as the expected changes in the exchange splitting ( ex ). Together, these observations: (i) S P behavior of m o, and (ii) S P behavior of m s and ex, provide strong evidence to establish that the alteration of the electronic structure with changing alloy composition is, through s-d hybridization, primarily responsible for the correlated behavior of µ alloy and TSP. We would also like to emphasize that such a clear observation of the S P behavior, a characteristic of most transition metal alloys, has not been established yet using the XMCD technique. Moreover, with this demonstrated tunability and insight into their magnetic, electronic and transport properties (also see Section A), we believe that CoFeB alloys open several new possibilities to engineer and enhance the performance of spin-torque devices based on junctions and nanowires. Particularly so, if one considers that their µ alloy and their TSP, which is a good representative of their conduction electron spin polarization [5], change by a factor 1.7 over the whole composition range. 6.3 Sample preparation and measurement We deposit Co 80-x Fe x B 20 layers on Si // SiO x and Si // SiO x / Al / AlO x buffer layers using DC magnetron sputtering (base pressure < 10 8 mbar) and plasma oxidation of Al. The alloys are sputtered from separate targets for each alloy composition. X- ray diffraction (XRD - Cu K α ) reveals a smooth growth of the CoFeB layers on both SiO x and AlO x in an amorphous/nanocrystalline state. Film composition is verified using in-situ x-ray photoelectron spectroscopy (XPS). µ alloy is measured using superconducting quantum interference device (SQUID) magnetometery performed at 5 K. The TSP data was measured at 0.25 K using standard lock-in technique with an ac modulation voltage of 10 µv pp. The UPS data is measured in-situ at normal emission with a He-I line (21.22 ev). The XAS and XMCD measurements are performed on Al / AlO x / Co 80-x Fe x B 20 (120 Å) / AlO x layers at station 5U.1 of the Daresbury laboratory by measuring the total electron yield. For the XMCD measurement, an external field (µ 0 H 500 mt) is applied at 45 to the photon k-

130 118 Chapter 6 Correlation between magnetism and TSP Figure 6.1: Schematic representation of the Slater-Pauling behavior for Co 100-x Fe x. (a) S P curve of µ alloy. (b) Element-specific magnetic moments of Co and Fe. Adapted from [19]. Sketched DOS of (c) weak and (d) strong ferromagnets. vector. The measured spectra are corrected for this angle and photon polarization ( 66%), which is determined on pure Co and Fe reference samples. 6.4 Introduction to the S P behavior A schematic representation of the S P curve is exemplified for Co 100 x Fe x alloys in Figure 6.1(a) as a function of the Fe content [19]. Notice that the generic shape for the total magnetic moment is simply a concentration weighted average of elementspecific moments of Co and Fe shown in Figure 6.1(b). As sketched in the density of states (DOS) of Figure 6.1(d), Co is a strong ferromagnet with its spin-up d-band completely filled. Quite generally, as the alloy composition changes, its electronic structure and its magnetic moment remains unaffected [see Figure 6.1(b)]. On the contrary, Fe being weakly ferromagnetic with both spin d-bands only partially filled [see Figure 6.1(c)] shows a substantial increase in magnetic moment as the Fe content decreases [see Figure 6.1(b)]. Eventually Fe undergoes a crossover from weak to strong ferromagnetism [see Figure 6.1(c and d)]. Note that this crossover of Fe with the associated increase in the Fe moment essentially causes the S P behavior of µ alloy [7, 8, 19] Basic aspects from computational magnetism According to self-consistent density functional calculations of Schwarz et al., this increase in Fe magnetic moment is due to a rising number of Co nearest neighbors,

131 6.4 Introduction to the S P behavior 119 d-dos (states/ev/atom) Fe Co bonding (a) -1 Fe antibonding Co Energy (ev) d-dos (states/ev/atom) d d (b) s s Energy (ev) s-dos (states/ev/atom) Figure 6.2: Origin of the Slater-Pauling behavior and s-electron spin polarization. (a) Calculated d-dos for Co and Fe atoms in amorphous Co 80-x Fe x B 20. (b) Comparison of calculated s-dos and d-dos on Co [5]. where Fe atoms having no Fe nearest neighbors exhibit the largest magnetic moment [7, 8]. Co has a slightly larger electro-negativity and its majority spin-band is almost full. On the contrary, Fe has a large exchange splitting, a large magnetic moment, and 0.3 holes in the majority spin band. Since local charge neutrality is maintained, the valance difference between the two species is realized in the minority channel, as Co majority channel is full. This effectively implies that the majority band of Co and Fe are very similar. An example of this is shown in Figure 6.2(a). Here the d-dos of Co and Fe in amorphous Co 72 Fe 20 B 8 calculated from first principles [5] are compared. In agreement with the calculations of Schwarz et al., here too one notices hardly any difference between Co and Fe majority DOS. In sharp contrast, the minority band of Fe deforms to a considerable extent, primarily due to its larger exchange splitting. One notes that the bonding part of the spin-down resonance has a larger spectral weight on the Co sites due to its relatively larger attractive potential. On the other hand, the anti-bonding part has more weight on the Fe sites. That is, there is shift in spectral weight from Fe to Co in this region. To maintain local charge neutrality, both Fe d-resonances shift downwards. This is a consequence of increased coulomb attraction (which does not depend on spin) due to reduced screening. To restore charge which increased the Co minority bonding spectra weight, a back donation of electrons from Co to Fe occurs closer to the Fermi level, causing an increase in Fe exchange splitting. Recall that the Fe majority band was not full. Thus, a self-consistent electron transfer from the Fe to Co to Fe bands occurs which finally results in an increased exchange splitting and larger magnetic moment on Fe.

132 120 Chapter 6 Correlation between magnetism and TSP S P behavior of CoFeB One may ask whether amorphous CoFeB alloys also show the S P behavior. First principles electronic structure calculations predict weak ferromagnetism in amorphous Fe 80-x B x alloys [20] and strong ferromagnetism in amorphous Co 80-x B x alloys [21]. Thus for CoFeB, one may expect that as the Fe content decreases, the Fe DOS undergoes a transition from weak to strong ferromagnetism, which would cause the S P behavior. Just as expected, Figure 6.3(a) shows that µ alloy of Co 80-x Fe x B 20 exhibits the S P curve. Such a curve has also been measured for CoFeB before [22]. A clue to the underlying mechanism for this S P behavior comes from extended x-ray absorption fine structure (EXAFS) measurements [41]. Orue et al. observe that as the Co content increases, the short range order around the Fe atoms also increases, predominantly due to the rising number of Co nearest neighbors. According to the calculations of Schwarz et al [7, 8] discussed earlier, one may infer that this rising number of Co neighbors around Fe leads to increase in the Fe moment and to the S P behavior. This argument is substantiated by first-principle calculations on amorphous Co-rich Co 72 Fe 20 B 8 where Fe is observed to be in a strong ferromagnetic state [5]. In the remainder of this chapter, we will not focus anymore on the microscopic origin of the S P curve for these amorphous alloys. Instead, we will investigate their TSP, and the changes in their electronic structure with alloy composition which affect it. 6.5 TSP of CoFeB shows the S P behavior Figure 6.3(c) shows a representative TSP data measured at 0.25 K using superconducting tunneling spectroscopy [23]. The zero field curve ( ) shows the superconducting DOS of Al. The application of a magnetic field (µ 0 H > 2.0 T) results in the Zeeman-splitting of the Al superconducting DOS which acts as a spin analyzer for the tunneling electrons. The observed asymmetry in the intensity of the measured peaks ( ) when fit (solid lines) with Maki theory [24] reveals the TSP of Co 80-x Fe x B 20. This magnitude of the TSP measured as a function of the Fe content is shown as open circles in Figure 6.3(b). Notice that the change in µ alloy [Figure 6.3(a)] over the whole composition range is around a factor 1.7. Remarkably, the TSP too is observed to change by a very similar factor. While the observed correlation in the shape of the two measured curves is not perfect, this similarity between µ alloy and the TSP is puzzling since, as mentioned earlier, µ alloy evolves from the d-electrons while the s-electrons dominate tunneling through AlO x [2 5]. Nevertheless, given this apparent correlation, if one naively assumes that the TSP and moment of Co and Fe in the alloy is the same as that in pure Co or Fe films, and that B is unpolarized [25], then one could estimate the TSP using:

133 6.5 TSP of CoFeB shows the S P behavior 121 /TM atom) alloy ( B TSP (%) Norm. di/dv (arb. u.) (a) (b) Fe content (at. %) (c) alloy TSP estimated TSP 0.5 TSP 0 T % Fit Co 24 Fe 56 B 2 T 20 Fit Bias voltage (mv) UPS Intensity (arb. units) h =21.2 ev pure Co Fe 00 Fe 08 Fe 20 Fe 32 Fe 44 Fe 56 Fe 68 Fe 80 CoFeB (d) h =21.2 ev Zharnikov et al Co 100 Fe 20 Fe 40 Fe 70 Fe 100 Co 100-x Fe x (e) Binding energy (ev) Figure 6.3: Properties of Co 80-x Fe x B 20 alloys. (a) µ alloy and (b) the TSP. (c) Representative TSP of Co 24 Fe 56 B 20 measured at 0.25 K. The µ 0 H = 2 T ( ) curves reveal the TSP of CoFeB when fit (solid lines) with Maki theory [24]. (d) UPS data on Co 80-x Fe x B 20. (e) UPS data on single crystalline fcc (100) Co 100-x Fe x alloys. Data courtesy of Dr. Wolfgang Kuch [30]. TSP = µ alloy (80 x). TSPpure Co (80 x). µ pure Co + x. TSP pure Fe + x. µ pure Fe (6.1)

134 122 Chapter 6 Correlation between magnetism and TSP The TSP values so estimated are shown as open squares ( ) in Figure 6.3(b). One notes a striking similarity with the measured TSP as well as with µ alloy. In fact, the use of this crude, and admittedly oversimplified approximation, seems to estimate the alloy TSP within 5% of its measured value. Here, TSP pure Co = 42% and TSP pure Fe = 45% [12], while µ pure Co = 1.7 µ B and µ pure Fe = 2.2 µ B [7, 8, 33, 34]. 6.6 Changes in valance band structure - UPS data In order to get some insight in the changes of the electronic structure which cause this apparent correlation between TSP and µ alloy, we measured valence band spectra using ultraviolet photoemission spectroscopy (UPS). This technique was discussed in Section It is appropriate to mention here that, depending on the energy of the photons and the growth direction of the sample, this technique probes a specific region of the Brillouin zone. Our results on CoFeB are shown in Figure 6.3(d). A systematic and pronounced impact of the changing alloy composition on the valence band structure is seen in the plot. The sharp peak around 0.5 ev for the Co-rich compositions broadens as the Fe content increases up to Fe 56 and then levels off. Now, it is well known that the UPS spectra of amorphous and single crystalline alloys are very similar to each other [26 29]. Based on these previous findings, one may compare our UPS CoFeB data to that on single crystalline (100) Co 100 x Fe x alloys from Zharnikov et al. [30], as shown in Figure 6.3(e). Indeed, one notes that the sharp peak for Co-rich single crystalline Co 100 x Fe x alloys is similar to amorphous Co 80 B 20. Moreover, this similarity extends throughout the composition dependent study. By comparing their measurements to semi-relativistic band structure calculations, Zharnikov et al. argue that this change of the UPS spectra basically arises from the change in exchange splitting and band filling as the alloy composition is varied [30]. In other words, from the intrinsic difference between the exchange splitting and band filling of Fe and Co electronic structures. Note that this difference in exchange splitting and band filling is the fundamental reason why pure Co is a strong ferromagnet and pure Fe is a weak ferromagnet [7, 8]. Therefore, based on the behavior of µ alloy of our CoFeB alloys and previous measurements of Zharnikov et al. on single crystalline CoFe samples, we tentatively ascribe this pronounced valance band spectral change to the gradual crossover from weak to strong ferromagnetism in amorphous CoFeB alloys. Later, we will provide clear evidence of the increase in exchange splitting of these alloys as the composition is varied using the XMCD technique, endorsing our above arguments. From the measured UPS spectra, one can also calculate the work functions (Φ) of these alloys (see Section 2.3.2). As can be seen in Table 6.1, the Φ measured for these amorphous alloys does not show any apparent systematic behavior, and is comparable to that measured for crystalline Co and Fe alloys.

135 6.7 XAS and XMCD 123 Table 6.1: Work functions of Co 80-x Fe x B 20 alloys. Fe content in Co 80-x Fe x B 20 (at.%) pure Co pure Fe units Φ ev 6.7 XAS and XMCD Although the UPS spectra provide a clear and direct evidence on the systematic changes occurring in the electronic structure, they are not element-specific. Such an insight would be invaluable considering that the S P behavior essentially derives from the changes in the Fe electronic structure. Therefore, we performed XAS and XMCD at the Fe L 2,3 edges, probing the Fe d-dos using synchrotron radiation. Next, we will discuss two aspects which can be measured using these techniques: (i) the orbital moment m o, and (ii) the spin moment (m s ) and exchange splitting ( ex ). The changes in these properties are interrelated. They explicitly demonstrate the transition of Fe from weak to strong ferromagnetism along with the changes occurring in the DOS at E F. Moreover, as we shall see later, this transition also provides a simple picture of a correlation between the s and d-electrons which explains the S P behavior of both µ alloy and the TSP. Figure 6.4(a) shows high-quality isotropic XAS spectra with standard background subtraction (step function [31]). The experimental details are described in Section The difference in the absorption cross-section (Γ ± ) measured for left / right (+/ ) circularly polarized ( 66%) light results in the corresponding XMCD spectra shown in Figure 6.4(b). In Figure 6.4(a-d), note that Fe 100 represents pure Fe, while Fe 0 represents Co 80 B 20 measured at the Co L 2,3 edges Orbital moment (m o ) As discussed in Section 2.4.3, according to Thole et al., m o is given by the orbital sum rule [32]: m o = 4 A 3 + A 2 (6.2) n 3d 3 A 3 + A 2 As shown in Figure 6.4(a), the integrated areas under the L 2,3 edges of isotropic XAS spectra are used to extract A 2,3, while the corresponding areas under the XMCD spectra are used to extract A 2,3 [see Figure 6.4(b)]. n 3d denotes the number of d-holes, which are unknown in the case of CoFeB. The calculated m o n 3d is plotted in Figure 6.4(c). Firstly, the absolute value of m o measured for Fe 100 ( 0.13 µ B with the known n 3d = 3.4) agrees fairly well with the value of 0.1 µ B calculated including orbital polarization [33]. Moreover, the curve in Figure 6.4(c) resembles

136 124 Chapter 6 Correlation between magnetism and TSP XAS Intensity (arb. units) -m o (x10-2 B ) / n 3d 1 0 L 3 = A 3 L 2 =A 2 Fe 100 Fe 80 Fe 68 Fe 56 Fe 44 Fe 32 Fe 20 Fe 08 (a) (b) L 3 = A 3 L 2 = A Photon energy (ev) (c) (d) A A3 6 ex Fe content (at. %) 0 XMCD intensity (arb. units) -m s (x10-1 B ) / n 3d Figure 6.4: Absorption and magnetic dichroism. (a) Background subtracted XAS for the Fe L 2,3 edges in Co 80-x Fe x B 20. (b) Corresponding XMCD spectra. (c) Orbital moment per hole, m o n 3d. (d) Spin moment per hole, m s n 3d. Inset shows A 3 A 3 which is proportional to the the exchange splitting ( ex ) [40]. Lines in c and d are guides to the eye. an inverted S P curve and implies the quenching of m o with increasing Fe content. This quenching of m o is confirmed by analyzing other ratios known to be sensitive to the spin-orbit interaction (see Appendix A). The changes in the Fe electronic structure sketched in Figure 6.1(b-d) may be shown to directly result in the observed quenching of m o. It is known that m o [n (E F ) - n (E F )], where n (E F ) is the spin-resolved total DOS at E F [33, 34, 36]. In other words, m o is directly proportional to the magnetic DOS at E F. A transition from strong to weak ferromagnetism [i.e., from Figure 6.1(d) to 6.1(c)] where the spin-up band moves towards E F would result in a decrease in [n (E F ) - n (E F )]. This will consequently result in the quenching of m o. Later we will see that these changes in the magnetic DOS at E F may also have an effect on the TSP.

137 6.8 Correlation between the s and the d-bands Spin moment (m s ) and exchange splitting ( ex ) The change in [n (E F ) - n (E F )] is expected to have a direct effect on m s which constitutes 90% of the total magnetic moment. Figure 6.4(d) shows ms n 3d calculated using the spin sum rule [37]: m s n 3d = 2 A 3 4 A 2 A 2 + A 3 7 T z n 3d (6.3) The magnetic dipole term ( T z ) is neglected as its local contributions are expected to cancel out for an amorphous system [38]. See Section for details. To begin with, the absolute value of m s for Fe 100 (2.14 µ B with n 3d = 3.4) is in excellent agreement with the magnetic moment of pure Fe [33, 34]. Most remarkably, the shape of m s n 3d is distinctly similar to that of µ Fe shown for Co 100-x Fe x in Figure 6.1(b). Recall that the shape of this curve in CoFe is associated with the transformation of Fe from a weak to a strong ferromagnet. The analogous behavior of ms n 3d in Figure 6.4(d) demonstrates that, as expected, Fe in CoFeB also undergoes a similar transformation. Accompanying this increase in m s, another signature of the S P curve would be a similar increase of ex which has been shown to be directly proportional to m s [39]. Such an increase in ex would also endorse our above arguments about the shifting of the d-bands [see Figure 6.1(b-d)] which influences the magnetic DOS at E F and m o. Now, ex has been shown to be directly proportional to the A 3 A 3 (and A 2 A 2 ) ratio [40]. In the inset of Figure 6.4(d), in agreement with the expected increase in ex m s, the A 3 A 3 ratio also increases. Furthermore, quantitatively speaking, in Figure 6.1(b) the Fe moment in Co 100-x Fe x alloys is seen to increase by 23%, i.e., from the nominal 2.2 µ B to 2.6 µ B. Remarkably, in CoFeB, m s and ex A 3 A 3 also increase by 20% and 25%, respectively [see Figure 6.4(d)]. Similar to the increase in A 3 A 3, we observe an increase in the A 2 A 2 ratio which is also proportional to ex (not shown). The absolute numbers for these ratios are also in very good agreement with those calculated by Chen et al. [40]. 6.8 Correlation between the s and the d-bands Given this crossover of Fe from weak to strong ferromagnetism, we will now address how exactly these changes in the Fe d-bands bring about the S P behavior of the s- electron dominated TSP. A clear indication comes from two independent arguments: (i) Isomer shifts essentially probe the changes in the s-electron charge density at the nucleus. In amorphous Co 80-x Fe x B 20 these isomer shifts also exhibit the S P behavior [41] due to s-d hybridization. Although these measured changes in the s- electron charge density represent all s-electrons below E F and are not spin-resolved, they directly point to the interplay between the s and d-electrons.

138 126 Chapter 6 Correlation between magnetism and TSP (ii) The spin-resolved information is observed in our measurements where the S P like changes in m o, m s and ex provide a direct insight in the underlying mechanism which causes a change in the TSP. More specifically, it is well-known that, due to s-d hybridization, the s-dos is suppressed in regions of large d-dos [5] [see sketch in Figure 6.1(c-d)]. As the Fe d-bands crossover from weak to strong ferromagnetism, the spin-up d-band gradually moves below E F. Recall that this shift in the d-band also resulted in the quenching of m o [n (E F ) - n (E F )]. As shown in Figure 6.1(c), due to this shift in the d-bands, one may also imagine an associated increase in the spin-up s-electron DOS at E F [n s(e F )]. This consequently increases the spin polarization of the Fe s-electrons defined as P Fe s = n s(e F ) n s(e F ) n s(e F )+n s(e F ). As a result, P Fe s behaves in a manner similar to the magnetic moment of Fe in Figure 6.1(b). The alloy spin polarization (P alloy s ) will consequently show the S P behavior, assuming the P Co s to remain unchanged just like the Co moment in Figure 6.1(b). Note that this increase in P alloy s will result in a corresponding increase in TSP, since the TSP is a good representative of P alloy s for these amorphous ferromagnets [5]. 6.9 Discussion on CoFe Given this information on the various aspects of CoFeB electronic structure and the coherent picture for the existence of a correlation between µ alloy and TSP, the discrepancy with the TSP measurements on Co 100-x Fe x alloys which do not seem to exhibit the S P behavior [12] may seem particularly puzzling. However, these alloys are crystalline and are known to undergo structural transitions (bcc fcc) depending on their compositions, which affect their electronic structure and may obscure a clear insight. In addition, detailed XMCD measurements which appear to be indispensable to address this issue, are yet to be performed on Co 100-x Fe x. On the contrary, the TSP of Co and Fe alloyed with Ru and V [9, 13], Ni alloyed with Cu [14], and that of NiFe alloys [6] is known to exhibit a correlation with µ alloy. XMCD measurements on these alloys would also provide more understanding on this issue. Indeed, in the case of Co 80-x Fe x B 20, the measurements presented here provide strong evidence and an intuitive insight for the existence of such a direct correlation between the d-electrons and the TSP. We would like to emphasize that the search for the existence of such a correlation between µ alloy and TSP in other ferromagnetic alloys will not only advance the application potential of spintronic devices, but also inspire computational spintronics to probe the fundamental understanding of this correlation - another issue yet to be ventured upon Conclusions In summary, we investigated the magnetism and TSP of amorphous Co 80-x Fe x B 20 films. We find that the S P behavior of the alloy magnetic moment is also seen in

139 A Appendix 127 XAS (arb. units.) Fe 100 Fe 80 B Photon energy (ev) Figure 6.5: XAS on crystalline Fe and amorphous Fe 80 B 20. Comparison of raw absorption cross-section (Γ) data for pure crystalline Fe to that of amorphous Fe 80 B 20. the s-electron dominated TSP. XMCD measurements show a crossover from weak to strong ferromagnetism in the Fe-DOS. To the best of our knowledge, this is the first observation of the S P behavior in transition metal alloys using the XMCD technique. We conclude that this crossover in the Fe-DOS, together with s-d hybridization, provides an intuitive understanding of the direct correlation between µ alloy and TSP. We also believe that the tunable electronic and magnetic properties of these CoFeB alloys allow access to engineer and advance the application potential of spintronic devices. A Appendix In this section, we will try to bring forth some very interesting observations which can be derived from the XAS and XMCD data, and try to provide additional information on the orbital moment of these alloys. These aspects are intentionally postponed to the last section of this chapter, as they fall beyond the scope of the correlation between µ alloy and TSP that we have been discussing above. A.1 Difference between Fe and Fe 80 B 20 - XAS Dipole selection rules dictate that an overwhelming majority of transitions are from the L 2 3d 3/2 final state, and from the L 3 3d 5/2 final state [42]. In other words, the integrals over the L 2 and L 3 edges of the isotropic XAS spectra, [A 2 and A 3 ] directly map the unoccupied 3d 3/2 and 3d 5/2 DOS, respectively [42]. This ability of XAS to probe the nature of the final states is illustrated in Figure 6.5 which compares

140 128 Chapter 6 Correlation between magnetism and TSP n 5/2 /n 3/2 (x10-1 ) Fe content (at. %) Figure 6.6: XAS indicates occupation of 3d 5/2 to 3d 3/2 states. The n 5/2 n 3/2 ratio, i.e., the ratio of the occupation of 3d 5/2 to 3d 3/2 states. raw absorption cross-section (Γ) data for pure crystalline Fe to that of amorphous Fe 80 B 20. While Γ L2 remains unchanged, Γ L3 which probes d-states higher in the band is seen to decrease for amorphous Fe 80 B 20. According to electronic structure calculations [20], the exchange splitting in Fe 80 B 20 is 0.6 ev smaller in comparison to that of Fe [20]. This results in increased occupation of the states higher in the Fe 80 B 20 d-dos, which may directly lead to a decrease in the absorption (Γ L3 ) to these states. However, the absorption to L 2, which probes low-lying states remains largely unchanged [20]. The lower value of A 3 A 3 ex for Fe 80 B 20 seen in the inset of Figure 6.4(d) is in good agreement with the lower exchange splitting expected for Fe 80 B 20 from the above argument. So also is the lower value for ms n 3d in Figure 6.4(d). A.2 Band-Filling and orbital moment In Figures , note that Fe 100 represents pure Fe, while Fe 0 represents Co 80 B 20 measured at the Co L 2,3 edges. The orbital moment (m o ) depends on band-filling effects, the spin moment (m s ), and short-range order which influences the crystal-field splitting [33, 34]. Bandfilling effects can also be studied using XAS. As Fe has one electron less that Co, with increasing Fe content, the gradual removal of one electron can be expected to influence the relative occupancy of the 3d 3/2 and 3d 5/2 states. Due to the relatively higher energy of the 3d 5/2 states, a preferential decrease in their occupancy is ex- ( A 3 12 A pected. This can be analyzed using the A 3 A 2 ratio, wherein n 5/2 n 3/2 = [43]. Here n 5/2 and n 3/2 stands for the number of d-holes (n 3d ) in the 3d 5/2 and 3d 3/2 states. Figure 6.6 shows the n 5/2 n 3/2 ratio. Consistent with an intuitive picture, as the Co content increases adding one electron to the system, the plot for n 5/2 n 3/2 ) suggests that the weight on the 3d 5/2 states increases. The higher value of n 5/2 n 3/2 for Fe 100 as compared to Fe 80 B 20, is in accordance with expected changes in the band-structure and the exchange splitting mentioned above.

141 A Appendix 129 (a) A 3 / A A3 A 3+A Fe content (at. %) (b) m o /m s (x10-2 ) A 3+A Fe content (at. %) Figure 6.7: Quenching of the orbital moment. (a) Inset shows the A 3 A 3 +A 2 ratio while the main figure shows the A 3 A 2 ξ mo n 3d. (b) Inset shows A 3 +A 2, also known as the r value. The main figure shows the m o m s which is independent of n 3d [46]. The values for pure Fe and Co films are 4.3 and 9.5 ( 10 2) respectively. A.3 Orbital moment A The orbital moment can also be probed by using the branching ratio 3 A 3 +A 2 [44, 45]. This calculated ratio is shown in the inset of Figure 6.7(a). Though the absolute value of the ratio is close to the expected statistical value of 0.66 [44, 45], it too shows a decrease with increasing Fe content indicating the quenching of the orbital moment. However, the branching ratio which is derived from XAS is more susceptible to background which arises due to transitions into the continuum. In general, the XMCD spectra are less prone to these issues as they inherently subtract the absorption to the continuum for left and right helicity of the light. Chen et al. used relativistic tight-binding calculations to show that the A 3 A 2 ratio derived from XMCD is very sensitive to the spin-orbit parameter (ξ) [40]. This calculated ratio is shown in Figure 6.7(a). It too shows the quenching of ξ m o very similar to the behavior of mo n 3d in Figure 6.4(c). Note that the Fe 8 data point is off in Figure 6.4(c) and Figure 6.7 primarily due to low signal to noise at this low Fe content. Söderlind et al. calculated that with increasing Fe content, m o decreased if Co 100-x Fe x was bcc structured [33]. This suggests a bcc like short range order for amorphous CoFeB. Interestingly, first-principles atomic structure calculations and EXAFS on amorphous Co 72 Fe 20 B 8 also showed a bcc-like short range order [5, 47], contrary to the fcc/densely packed structure expected for such a Co rich alloy. A.4 Ratio of Orbital to Spin Moment A 3 + A 2 The ratio of m o m s = 2 3 A 3 2 A 2 is independent of n 3d [46]. Chen et al. [31] found m o m s to be for pure bcc Fe and for pure fcc Co. Our measurements [see

142 130 Chapter 6 Correlation between magnetism and TSP XAS intensity (arb. units) 1 0 Co 100 Co Co Co 36 Co Co Co content (at.%) n Co 3d 2.4 holes (a) (b) Co content (at.%) Photon energy (ev) Co r value (arb. units) -m s ( B ) XMCD intensity (arb. units) Figure 6.8: Co edge XAS and XMCD. (a) XAS on Co edge. Inset shows A 3 +A 2 also known as the r value. (b) XMCD on Co edge. Inset shows the extracted m s where n 3d for Co is taken to be the well known value of 2.4 holes [33, 34]. Figure 6.7(b)] on amorphous Co 80-x Fe x B 20 are in excellent agreement with the work of Chen et al. on crystalline Co and Fe films. The inset in Figure 6.7(b) shows the sum of the areas under the L 2,3 edges, generally also known as the r value. The linear increase with Fe content indicates that the number of holes per Fe atom does not vary with composition. A.5 Co edge XAS and XMCD Although limited by the available beam time, we performed XAS and XMCD measurements on the Co edge for most of these alloys. These data are shown for the sake of completeness in Figure 6.8. Similar to the Fe edge, the r value (A 3 +A 2 ) on the Co edge in the inset of Figure 6.8(a) is seen to vary linearly with composition. Regarding the XMCD data shown in in Figure 6.8(b), we observe no change in the spin moment on Co atoms as the composition changes. Here, after evaluation of m o n 3d, the number of holes for Co is taken to be the well known value of 2.4 holes [33, 34]. Recall, that since Co is a strong ferromagnet, one does not expect any changes in its spin magnetic moment, as confirmed by the XMCD data of Figure 6.8(b). Moreover, the obtained value of 1.6 µ B for the spin moment of Co is in good agreement with calculations [33, 34].

143 BIBLIOGRAPHY 131 Bibliography [1] C. Chappert, A. Fert, and F. Nguyen, The emergence of spin electronics in data storage. Nature Mater. 6, 813 (2007). 6.1 [2] S. Yuasa, T. Nagahama, and Y. Suzuki, Spin-polarized resonant tunneling in magnetic tunnel junctions. Science 297, 234 (2002). 6.1, 6.5 [3] T. Nagahama, S. Yuasa, E. Tamura, and Y. Suzuki, Spin-dependent tunneling in magnetic tunnel junctions with a layered antiferromagnetic Cr(001) spacer: Role of band structure and interface scattering. Phys. Rev. Lett. 95, (2005). [4] M. Münzenberg, and J. S. Moodera Superconductor-ferromagnet tunneling measurements indicate sp-spin and d-spin currents. Phys. Rev. B 70, R (2004). [5] P. V. Paluskar, G. A. de Wijs, J. J. Attema, S. Fiddy, E. Snoeck, J. T. Kohlhepp, H. J. M. Swagten, R. A. de Groot, and B. Koopmans, Spin tunneling in junctions with disordered ferromagnets. Phys. Rev. Lett. 100, (2008). 6.1, 6.2, 6.2, 6.4.1, 6.4.2, 6.5, 6.8, A.3, 6.10 [6] R. Meservey,, D. Paraskevopoulos, and P. M. Tedrow, Correlation between spin polarization of tunnel currents from 3d ferromagnets and their magnetic moments. Phys. Rev. Lett. 37, 858 (1976). 6.1, 6.9 [7] R. Richter, and H. Eschrig, LCAO-CPA for disordered 3d transition metal alloys. Magnetic moment formation in NiCu and FeCo. Phys. Scrip. 37, 948 (1988). 6.1, 6.4, 6.4.1, 6.4.2, 6.5, 6.6 [8] K. Schwarz, P. Mohn, P. Blaha, and J. Kübler, Electronic and magnetic structure of bcc Fe-Co alloys from band theory. J. Phys. F: Met. Phys. 14, 2659 (1984). 6.1, 6.4, 6.4.1, 6.4.2, 6.5, 6.6 [9] C. Kaiser, S. van Dijken, S.-H. Yang, H. Yang, and S. S. P. Parkin, Role of tunneling matrix elements in determining the magnitude of the tunneling spin polarization of 3d transition metal ferromagnetic alloys. Phys. Rev. Lett. 94, (2005). 6.1, 6.9 [10] A. T. Hindmarch, C. H. Marrow, and B. J. Hickey, tunneling spin polarization in magnetic tunnel junctions near the Curie temperature. Phys. Rev. B 72, (2005). [11] C. Kaiser, A. F. Panchula, and S. S. P. Parkin, Finite tunneling spin polarization at the compensation point of rare-earth-metal-transition-metal alloys. Phys. Rev. Lett. 95, (2005).

144 132 Chapter 6 Correlation between magnetism and TSP [12] D. J. Monsma, and S. S. P. Parkin, Spin polarization of tunneling current from ferromagnet/al 2 O 3 interfaces using copper-doped aluminum superconducting films. Appl. Phys. Lett. 77, 720 (2000). 6.5, 6.9 [13] Kaiser, C., Ph.D. Thesis, Novel materials for magnetic tunnel junctions. Rheinisch-Wesfälischen Technichen Hochschule Aachen, and IBM Almanden Research Center, [14] Such a correlation between µ alloy and TSP has also been observed in Cu-Ni alloys by J. S. Moodera et al. (unpublished.) Private communication. 6.1, 6.9 [15] A. A. Tulapurkar, Y. Suzuki, A. Fukushima, H. Kubota, H. Maehara, K. Tsunekawa, D. D. Djayaprawira, N. Watanabe, and S. Yuasa, Spin torque diode effect in magnetic tunnel junctions. Nature 438, 339 (2005). 6.2 [16] H. Kubota, Y. Suzuki, A. Fukushima, H. Kubota, H. Maehara, K. Tsunekawa, D. D. Djayaprawira, N. Watanabe, and S. Yuasa, Quantitative measurement of voltage dependence of spin transfer torque in MgO based magnetic tunnel junctions. Nature Phys. 7, 37 (2007). 6.2 [17] W. L. O Brien, and B. P. Tonner, Orbital and spin sum rules in x-ray magnetic circular dichroism. Phys. Rev. B 50, (1994). [18] W. L. O Brien, B. P. Tonner, G. R. Harp, and S. S. P. Parkin, Experimental investigation of dichroism sum rules for V, Cr, Mn, Fe, Co, and Ni: Influence of diffuse magnetism. J. Appl. Phys. 76, 6462 (1994). [19] The absolute values of these curves represent a general trend observed in numerous calculations and neutron diffraction measurements on CoFe. For example, see I. Turek, J. Kudrnovský, V. Drchal, and P. Weinberger, Itinerant magnetism of disordered Fe-Co and Ni-Cu alloys in two and three dimensions. Phys. Rev. B 49, 3352 (1994) for calculations. See M. F. Collins, and J. B. Forsyth, Magnetic moment distribution in some transition metal alloys. Phil. Mag. 8, 401 (1963) for measurements. 6.1, 6.4 [20] J. Hafner, M. Tegze, and Ch. Becker, Amorphous magnetism in Fe-B alloys: First-principles spin-polarized electronic-structure calculations. Phys. Rev. B 49, 285 (1994) , A.1 [21] H. Tanaka S. Takayama, M. Hasegawa, T. Fukunaga, U. Mizutani, A. Fujita, and K. Fukamichi, Electronic structure and magnetism of amorphous Co 1 x B x alloys. Phys. Rev. B 47, 2671 (1993) [22] R. C. O Handley, R. Hasegawa, R. Ray, and C.-P. Chou, Ferromagnetic properties of some new metallic glasses. Appl. Phys. Lett. 29, 330 (1976)

145 BIBLIOGRAPHY 133 [23] R. Meservey, and P. M. Tedrow, Spin-polarized electron tunneling. Phys. Rep. 238, 173 (1994). 6.5 [24] K. Maki, Pauli paramagnetism and superconducting state. II. Prog. Theor. Phys. 32, 29 (1964). 6.5, 6.3 [25] B s-states in Co 72 Fe 20 B 8 are found to be highly spin polarized [5]. However, one must also note that these calculations were done on low boron content (8%) alloys, while the alloys studied presently contain substantially higher amount of boron (20%), which may or may not have an influence on the polarization achieved by the boron sp-dos. Therefore, only electronic structure calculations for the whole Co 80-x Fe x B 20 range can validate this assumption. 6.5 [26] R. Hasegawa, Glassy metals : magnetic, chemical, and structural properties. (p.85, CRC press, Boca Raton) (1983). 6.6 [27] M. Matsuura, T. Nomoto, F. Itoh, and K. Suzuki, XPS study of the valance band spectrum of amorphous Fe-B alloys. Solid State Commun. 33, 895 (1980). [28] H. J. Güntherodt et al. Electronic structure of liquid and amorphous metals. J. Phys. Paris C8, 381 (1980). [29] A. Amamou, d-band structure and alloying effects in crystalline and amorphous Zr-Co and Zr-Ni alloys. Solid State Commun. 33, 1029 (1980). 6.6 [30] M. Zharnikov, A. Dittschar, W. Kuch, C. M. Schneider, and J. Kirschner, Spin-resolved photoemission and band mapping in epitaxial fcc Fe-Co alloys on Cu(001). J. Mag. Mag. Mater 165, 250 (1997). 6.3, 6.6 [31] C. T. Chen, Y. U. Idzerda, H.-J. Lin, N. V. Smith, G. Meigs, E. Chaban, G. H. Ho, E. Pellegrin, and F. Sette, Experimental confirmation of the x-ray magnetic circular dichroism sum rules for Iron and Cobalt. Phys. Rev. Lett. 75, 152 (1995). 6.7, A.4 [32] B. T. Thole, P. Carra, F. Sette, and G. van der Laan, X-ray magnetic dichroism as a probe for orbital magnetism. Phys. Rev. Lett. 68, 1943 (1992) [33] P. Söderlind, O. Eriksson, B. Johansson, R. C. Albers, and A. M. Boring, Spin and orbital magnetism in Fe-Co and Co-Ni alloys. Phys. Rev. B 45, (1992). 6.5, 6.7.1, 6.7.2, A.2, A.3, 6.8, A.5 [34] O. Eriksson, A. M. Boring, R. C. Albers, G. W. Fernando, and B. R. Cooper, Spin and orbital contributions to surface magnetism in 3d ferromagnets. Phys. Rev. B 45, 2868 (1992) and reference 30 therein. 6.5, 6.7.1, 6.7.2, A.2, 6.8, A.5 [35] M. B. Stearns, Landolt-Börnstein, New Series, Vol. III/19a, Springer-Verlag, Berlin (1984); D. Bonnenberg, K. A. Hempel, H. P. J. Wijn, ibid.

146 134 Chapter 6 Correlation between magnetism and TSP [36] H. Ebert, R. Zeller, B. Drittler, and P. H. Dederichs, Fully relativistic calculation of the hyperfine fields of 5d impurity atoms in ferromagnetic Fe. J. Appl. Phys. 67, 4576 (1990) [37] P. Carra, B. T. Thole, M. Altarelli, and X. Wang, X-ray magnetic circular dichroism and local magnetic fields. Phys. Rev. Lett. 70, 694 (1993) [38] K. Fleury-Frenettea, S. S. Dhesi, G. van der Laan, D. Strivay, G. Weber, and J. Delwiche, Characteristics of the iron moment in Dy-Fe and Dy-FeCo amorphous alloys studied by x-ray magnetic circular dichroism. J. Mag. Mag. Mater. 220, 45 (2000) [39] F. J. Himpsel, Exchange splitting of epitaxial fcc Fe/Cu(100) versus bcc Fe/Ag(100). Phys. Rev. Lett. 67, 2363 (1991) [40] C. T. Chen, N. V. Smith, and F. Sette, Exchange, spin-orbit, and correlation effects in the soft x-ray magnetic dichroism spectrum of nickel. Phys. Rev. B 43, 6785 (1991). 6.4, 6.7.2, A.3 [41] I. Orue, M. L. Fdez-Gubieda, and F. Plazaola, The local structure from two experimental atomic probes: EXAFS and Mössbauer spectroscopies. J. Non- Crys. Solids 287, 75 (2001) , 6.8 [42] H. Ebert, J. Stöhr, S. S. P. Parkin, and M. Nilson, L-edge x-ray absorption in fcc and bcc Cu metal: Comparison of experimental and first-principles theoretical results. Phys. Rev. B 53, (1996). A.1 [43] T. L. Morrison, M. B. Brodsky, N. J. Zalluzec, and L. R. Sill, Iron d-band occupancy in Fe x Ge 1 x. Phys. Rev. B 32, 3107 (1985). A.2 [44] G. van der Laan, and B. T. Thole, Local probe for spin-orbit interaction. Phys. Rev. Lett. 60, 1977 (1988). A.3 [45] B. T. Thole, and G. van der Laan, Branching ratio in x-ray absorption spectroscopy. Phys. Rev. B 38, 3158 (1988). A.3 [46] R. Wu, and A. J. Freeman, Limitations of the magnetic circular dichroism spin sum rule for transition metals and importance of the magnetic dipole term. Phys. Rev. Lett. 73, 1994 (1994). 6.7, A.4 [47] P. V. Paluskar et. al., unpublished. A.3

147 Chapter 7 Thermal stability of MTJs Role of Mn diffusion Abstract: In this chapter 1, we will address slightly different issues. Here too, we keep the application of spintronic devices in mind and focus on some of the relevant questions about their thermal stability. We will examine the role of Mn diffusion in the thermal stability of tunneling spin polarization (TSP) by directly measuring TSP of Al / AlO x / Co / FeMn and Al / AlO x / Co 90 Fe 10 / FeMn junctions using superconducting tunneling spectroscopy (STS). Mn diffusion is considered to be one of the reasons for the degradation of the TMR of MTJs after high temperature anneals. We confirm Mn diffusion towards the barrier-ferromagnet interface in our junctions using X-ray photoelectron spectroscopy after an ultra high vacuum (UHV) 500 C anneal. Surprisingly, and in contrast to the current belief, no drop in TSP is observed using STS. We therefore conclude that, though Mn diffuses significantly, our data does not support the conjecture that this diffusion is responsible for the drop in TMR observed after post-deposition anneals above 300 C. Notice that CoFeB is not used as the ferromagnetic electrode here for the obvious reason that it is susceptible to structural changes at these anneal temperatures. These changes might preclude an unambiguous determination of Mn diffusion. 1 A large part of this chapter appeared in Journal of Applied Physics [21]. 135

148 136 Chapter 7 Thermal stability of MTJs 7.1 Introduction Background A standard TMR stack generally consists of an antiferromagnet / ferromagnet / insulator / ferromagnet multilayer, where the antiferromagnetic (AF) layer is used to pin the direction of the magnetic moment of the adjacent ferromagnetic layer The AF layer generally contains a Mn alloy (e.g., Fe 50 Mn 50, Pt 50 Mn 50 ) to allow device operation at elevated temperatures (above 150 C) [1]. Presently, one of the major areas of research in MTJ s is the miniaturization of these elements for application in MRAM and their integration with CMOS (complementary metal oxide semiconductor) processing [2]. Successful integration of the MTJ in MRAM requires the device to be thermally resistant against standard high temperature CMOS processing steps ( C) [3]. In this regard, the thermal stability of the tunneling spin polarization (TSP), which is the fundamental parameter responsible for the TMR effect, has been demonstrated by Kant et al. [see Figure 7.1(a)] [4]. They showed that the TSP in an Al / AlO x / Co (Co 90 Fe 90 ) junction does not change after anneals up to 500 C when the junctions are annealed in ultra-high vacuum (UHV). However, as can be seen in Figure 7.1(b), for magnetic tunnel junctions containing Mn-based exchange bias layers, while post-deposition anneals below 300 C enhance the TMR, those above 300 C lead to its severe degradation [5]. The physical mechanism behind this drop in TMR after anneals above 300 C is not yet completely understood. Several causes have been suggested for this drop, among which Mn diffusion from the AF layer into the ferromagnetic electrode and towards the ferromagnet-barrier interface is believed to play the principle role [5, 6]. Note that the junctions studied by Kant et al. [4] did not contain any exchange biasing layers and the measured TSP was critically dependent on the anneal conditions. While UHV anneals exhibited robust TSP values [see Figure 7.1(a)], anneals in argon atmosphere, which were routinely used in the community in that period, were shown to cause a severe degradation in TSP [see Figure 7.1(c)]. To summarize the issues, the cause for degradation TMR after anneals above 300 C is still to be determined, and Mn diffusion as well as the influence of annealing conditions needs to be investigated This work In this chapter, we combine a study of the thermal stability of TSP based on the superconducting tunneling spectroscopy (STS) technique [10] with an x-ray photoelectron spectroscopy (XPS) analysis of Mn diffusion. The spin polarization of the tunneling electrons in the MTJ is very sensitive to the interfacial density of states at the barrier-ferromagnet interfaces [8]. Any change in TSP should ensue from chemical and / or morphological changes at or near the barrier-ferromagnet interface. We demonstrate that TSP in our Al / AlO x / Co / FeMn and Al / AlO x

149 7.2 Experimental Results (a) anneal (10-9 mbar) 60 (b) anneal (10-6 mbar) 60 (c) Ar atomsphere 45 Al/AlO x /Co 90 Fe TSP (%) Al/AlO x /Co TMR (%) TSP (%) Al/AlO x /Co Anneal temperature ( C) Figure 7.1: TSP and TMR versus anneal conditions. (a) The TSP of a Al / AlO x / Co (200 Å) junction shows the thermal stability of the TSP. The intrinsic nature of the thermal stability of the TSP is demonstrated even when the Co electrode is replaced by Co 90 Fe 10 [4]. (b) However, the behavior of TMR as a function of the anneal temperature in Co 82 Fe 18 / AlO x / Co 82 Fe 18 / Ir 26 Mn 74 junctions [5] is not similar to that of the TSP. This behavior is representative for comparable junctions studied by others in similar anneal conditions. (c) The TSP of a Al / AlO x / Co (200 Å) junction when annealed in an argon atmosphere shows a drop in TSP when annealed above 300 C [4]. / Co 90 Fe 10 / FeMn junctions is thermally stable up to T a = 500 C, even when Mn diffuses towards the barrier-ferromagnet interface. Notice that Co and Co 90 Fe 10 are used as ferromagnetic electrodes, instead of CoFeB. We comment on this choice in the next section. 7.2 Experimental Results Confirmation of Mn diffusion in a MTJ Detection of the influence of Mn diffusion in a conventional tunnel junction stack, for example, one consisting of FeMn / Co / AlO x / Co / Ta, is difficult to probe experimentally with XPS, since the escape depth of the photoelectrons is much less than the standard thickness of the top Co and Ta layers. See Section for details. Therefore, we deposited Al / AlO x / Co (200 Å) / FeMn (100 Å) / Co (200 Å) layers on silicon substrates using DC magnetron sputtering (base pressure < 10 8 mbar), in-situ annealed them at 500 C in ultra high vacuum (UHV, pressure < 10 8 mbar during anneal) for 30 minutes, and then studied them with in-situ XPS (Al K α ). This stack is pictorially represented in Figure 7.2. It is reasonable to assume that if Mn diffuses to the surface of the 200 Å thick top Co layer, it would also diffuse towards the AlO x / Co interface below the FeMn layer. Also, any significant Mn accumulation near the surface of 200 Å thick top Co layer should be detectable

150 138 Chapter 7 Thermal stability of MTJs Co FeMn Co XPS Intensity (counts) AlOx Al 50k 40k 30k 20k 10k (a) (b) Co 2p 1/2 Co 2p Mn x O y 2p 3/2 1/2 Mn x O y 2p 3/2 Mn 2p 1/2 Mn 2p 3/2 as deposited annealed at 500 o C Binding energy (ev) 300k 250k 200k 150k 100k Figure 7.2: Does Mn diffuse? A sketch of the experimental junction used to probe Mn diffusion with XPS. In-situ XPS (Al K α ) intensity spectra for (a) Mn peaks observed on the 200 Å top Co layer in the Al / AlO x / Co (200 Å) / FeMn (100 Å) / Co (200 Å) stack before and after in-situ post-deposition UHV anneal at 500 C. The dash-dotted (Mn) and dashed (Mn x O y ) lines indicate the expected 2p 3/2 and 2p 1/2 peak locations. (b) Co peaks for the same sample, the intensity of the Co 2p 3/2 and 2p 1/2 peaks decreases after the anneal. by XPS. The AlO x barrier layer (which is Å thick) was formed by partially plasma oxidizing the 40 Å Al bottom electrode for 200 seconds [9]. The choice of an elemental ferromagnet is obvious. As we have seen all throughout this thesis, CoFeB is undoubtedly susceptible to structural changes in anneals above 300 C. Since the transformation from amorphous to crystalline alloy unavoidably implies grain formation, grain boundary diffusion is an additional complication which needs to be avoided in this anneal-temperature dependent study to ensure the derivation of any meaningful conclusion Does Mn diffuse? Figure 7.2(a) shows the XPS spectra measured before and after a 500 C anneal in the region where a Mn line is expected. Clearly, as-deposited samples show no evidence of Mn peaks in the intensity scan, confirming the absence of Mn at or near the surface of the 200 Å thick top Co layer. However, after the anneal, two explicit peaks appear near the energies of known Mn p-level peaks, which for pure Mn, are expected to be at ev (2p 3/2 ) and ev (2p 1/2 ), respectively [13]. This result is a proof of Mn diffusion from the FeMn layer towards the surface of the 200 Å thick top Co layer. Careful examination of the spectra show that the Mn peaks are shifted to higher binding energies, evincive of an oxidized state of Mn in

151 7.2 Experimental Results 139 XPS Intensity (counts) 70k 60k 50k 40k 30k 20k Anneal temp. ( o C) Mn/Co intensity ratio (arb. u.) MnO Mn 100 o C 200 o C 300 o C 400 o C 500 o C 500 o C =70 o 11k 10k Binding Energy (ev) Figure 7.3: Mn diffusion as a function of anneal temperature. Mn intensity in each sample measured after an increasingly higher anneal temperature. Note that the XPS intensity for measurements below an anneal temperature of 400 C is plotted on a log scale. The diffused Mn oxidizes after reaching the surface of the top Co layer. This is clearly seen in the spectra measured at gracing angle (θ = 70 ) after an anneal at 500 C which has visibly larger intensity as compared to a corresponding measurement at normal incidence. Inset shows the Mn to Co intensity ratio measured in our XPS spectra as a function of post-deposition anneal temperature. the Co layer. In our sample, the 2p 3/2 Mn oxide peak is found to be around ev. The literature values for the 2p 3/2 peaks of various manganese oxides are found to lie between ev [13, 15]. The formation of Mn oxide near the surface of the top Co layer is purely due to the background partial pressure of oxygen in the chamber, which is introduced by the degassing of adsorbed oxygen from the sample plate during the anneal. Although we do not completely exclude the possibility of oxygen driven Mn diffusion (reported for a 30 Å thick CoFe layer by [11]) towards the surface of the top Co layer, we believe that our 200 Å thick top and bottom Co layers should inhibit such a process. Figure 7.1(b) shows the corresponding XPS spectra for Co 2p 3/2 (778.1 ev) and 2p 1/2 (793.0 ev) peaks. It can be seen that the spectral intensity of peaks for the asdeposited sample is much larger than the annealed sample, confirming the decrease

152 140 Chapter 7 Thermal stability of MTJs of Co concentration near the surface of the layer, due to its displacement by Mn. However, no oxidation of Co is evident, since there is no distinguishable shift in the peaks. Co 2p 3/2 peaks are expected at ev, and those for its oxides are expected between ev [13, 15]. These results are in accordance with the fact that the (negative) free energy of formation is lowest for cobalt oxides, intermediate for manganese oxides and highest for aluminum oxides [14], making it difficult for Co to oxidize in the presence of Al and Mn. Figure 7.3 shows the Mn spectra measured on different samples each annealed at progressively higher temperatures. The influence of the anneal temperature in promoting Mn diffusion is clearly visible here. One notices a small bump appearing even at an anneal at 200 C. As we have seen in Section 2.3.1, a gracing angle incidence measurement is more sensitive to the outermost surface layers. In Figure 7.3, the gracing angle incidence measurement (θ = 70 ) performed after an anneal at 500 C shows a clear presence of Mn oxide on the surface layer, since the intensity of the Mn oxide peak is much larger than that observed in a corresponding measurement at normal incidence. By fitting these peaks to gausssians one can obtain the Mn to Co spectral intensity ratio as a function of anneal temperature. The ratio is calculated by removing the background in the measurements, fitting the peaks to expected Mn and Co peaks, and then deriving the area under the curve. This ratio is shown in the inset of Figure 7.3. It is noteworthy that although the Mn/Co ratio at or near the surface of the 200 Å top Co layer increases even after anneals at 200 C, it shows a definite kink around 300 C, which has been reported as the onset temperature for TMR collapse [5] Influence of Mn diffusion on the TSP To measure the effect of Mn diffusion on TSP, cross-striped tunnel junctions with and without FeMn were prepared similar to the XPS samples. A 60 Å Ta capping layer was added on top. The junctions have an area of µm 2 and a resistancearea product of roughly 10 5 Ωµm 2. STS measurements similar to those discussed in Section were performed on these junctions. Figure 7.4(a) shows representative measurements for Al / AlO x / Co while Figure 7.4(b) corresponds to an Al / AlO x / Co 90 Fe 10 junction at 0.26 K. The extracted TSP (38 ± 1%) for Co and (48 ± 1%) Co 90 Fe 10 junctions are in fair agreement with earlier work [18]. Figure 7.4(c) shows TSP as a function of post-deposition anneal temperature for junctions which do (closed symbols) and do not (open symbols) contain an FeMn layer. Remarkably, TSP does not suffer any degradation in response to anneal up to 500 C for both types of ferromagnets, independent of the presence of FeMn. Also, the absolute values of TSP for a particular ferromagnet does not change before and after the anneal, irrespective of Mn diffusion into the layers. This result is in qualitative agreement with the work of Kim and Moodera [19], who report that Mn concentrations as high as 30% in Al / AlO x / Co y Mn 1 y junctions have only a weak negative effect on the TSP. In addition, the anneals do not affect other junction

153 7.2 Experimental Results 141 Normalized Conductance (a) Co T a =500 C TSP 38% (b) Co 90 Fe 10 B = 0 T Fit B > 2.5 T Fit Bias Voltage (mv) TSP 48% T a =500 C (c) TSP Co 90 Fe 10 /FeMn Co 90 Fe 10 Co Co/FeMn Anneal temperature ( o C) TSP (%) Figure 7.4: Thermal stability of the TSP. Conductance of Al / AlO x / Co (a) and Al / AlO x / Co 90 Fe 10 junction (b) at 0.26 K. The zero field curve ( ) shows the Al superconducting gap while the >2.5 T ( ) curve reveals the TSP of CoFeB when fit (solid lines) with Maki theory [16]. (c) TSP measured after an in-situ 30 minute post-deposition UHV anneal in Al / AlO x / Co and Al / AlO x / Co 90 Fe 10 junctions which do (closed symbols) and do not (open symbols) contain an FeMn layer. This data shows that TSP is not affected by the presence of an FeMn layer on top of the ferromagnet. parameters such as junction resistance and the superconducting band gap of our Al electrode. The thermal robustness of TSP above 300 C (evident in Figure 7.4) is in sharp contrast with the effect of post-deposition annealing on the TMR of MTJ s. In order to clarify this apparent contradiction further experiments are indispensable. The stable TSP in our junctions suggest that the TMR degradation may not be due to a degradation of the intrinsic TSP of the AlO x / Co or AlO x / Co 90 Fe 10 system, but instead is a result of extrinsic influences, such as the diffusion of impurity atoms into the barrier or to one of its interfaces. It is notable that our junctions are annealed in UHV (base pressure 10 9 mbar), as compared to earlier work by Cardoso et al. and Lee et al. [5, 11, 12] who used vacuum chamber pressures of around 10 6 mbar. We have already shown in junctions without FeMn, UHV anneals preserve the TSP as compared to similar junctions annealed in an Ar gas environment. This Ar environment was known to have a comparatively higher partial pressure of other gaseous impurities like oxygen and nitrogen [4] Impact of annealing on TSP We now turn our attention to another interesting observation. Typically, anneals below 300 C enhances TMR. One explanation of this enhancement in TMR is an improvement of TSP due to migration of excess oxygen from the bottom ferro-

154 142 Chapter 7 Thermal stability of MTJs magnetic electrode into the AlO x barrier [20]. This stoichiometric redistribution of oxygen in the barrier results in a sharper interface and improved barrier properties. Consequently, the barrier height is expected to increase, and spin-independent tunneling processes are expected to decrease, both leading to higher TMR. Another explanation which concerns both electrodes, is the possibility of a change in the ferromagnet structure at the interface after the anneal. However, our measurements do not show an increase in TSP when our junction stack is annealed at C, which indicates that there is no change at the barrier-ferromagnet interface or in the structure of the ferromagnetic electrode which contributes to enhancement of TSP, and subsequently, TMR. Therefore, the second explanation, i.e., change in the ferromagnetic electrode structure, is not supported by our measurements. 7.3 Summary In summary, we investigated Mn diffusion in Al / AlO x / ferromagnet junctions and its effect on the TSP of the electrons tunneling from the ferromagnet. Contrary to the current belief, we have shown that TSP in Al / AlO x / Co and Al / AlO x / Co 90 Fe 10 junctions is thermally stable up to 500 C, despite of Mn diffusion towards the barrier-ferromagnet interface.

155 BIBLIOGRAPHY 143 Bibliography [1] For example, with respect to industrial and automobile sensors, see the German BMBF project Magnetoelectronic specifications led by Robert Bosch GmbH [2] S. S. P. Parkin, K. P. Roche, M. G. Samant, P. M. Rice, R. B. Beyers, R. E. Scheuerlein, E. J. OSullivan, S. L. Brown, J. Bucchigano, D. W. Abraham, Y. Lu, M. Rooks, P. L. Trouilloud, R. A. Wanner, and W. J. Gallagher, Exchange-biased magnetic tunnel junctions and application to nonvolatile magnetic random access memory (invited). J. Appl. Phys. 85, 5828 (1999) [3] S. Tehrani, J. M. Slaughter, M. Deherrera, B. N. Engel, N. D. Rizzo, J. Salter, M. Durlam, R. W. Dave, J. Janesky, B. Butcher, K. Smith, and G. Grynkewich, Magnetoresistive random access memory using magnetic tunnel junctions. Proc. of the IEEE 91, 703 (2003) [4] C. H. Kant, J. T. Kohlhepp, H. J. M. Swagten, and W. J. M. de Jonge, Intrinsic thermal robustness of tunneling spin polarization in Al/Al 2 O 3 /Co junctions. Appl. Phys. Lett. 84, 1141 (2004) , 7.1, [5] S. Cardoso, P. P. Freitas, C. de Jesus, P. Wei, and J. C. Soares, Spin-tunneljunction thermal stability and interface interdiffusion above 300 C. Appl. Phys. Lett. 76, 610 (2000) , 7.1, 7.2.2, [6] M. G. Samant, J. Lüning, J. Stöhr, and S. S. P. Parkin, Thermal stability of IrMn and MnFe exchange-biased magnetic tunnel junctions. Appl. Phys. Lett. 76, 3097 (2000) [7] G. A. Prinz, Magnetoelectronics. Science 282, 1660 (1998). [8] P. LeClair, J. T. Kohlhepp, H. J. M. Swagten, and W. J. M. de Jonge, Interfacial density of states in magnetic tunnel junctions. Phys. Rev. Lett. 86, 1066 (2001) [9] P. LeClair, J. T. Kohlhepp, A. A. Smits, H. J. M. Swagten, B. Koopmans, and W. J. M. de Jonge, Optical and in-situ characterization of plasma oxidized Al for magnetic tunnel junctions. J. Appl. Phys. 87, 6070 (2000) [10] R. Meservey and P. Tedrow, Spin-polarized electron tunneling. Phys. Rep. 238, 173 (1994) [11] C. S. Yoon, J. H. Lee, D. Jeong, C. K. Kim, J. H. Yuh, and R. Haasch, Diffusion study of the exchange-biased NiFe/MnIr/CoFe electrode in magnetic tunnel junctions. Appl. Phys. Lett. 80, 3976 (2002) , 7.2.3

156 144 Chapter 7 Thermal stability of MTJs [12] J. H. Lee, D. Jeong, C. S. Yoon, C. K. Kim, B. G. Park, and T. D. Lee, Interdiffusion in antiferromagnetic/ferromagnetic exchange coupled NiFe/IrMn/CoFe multilayer. J. Appl. Phys. 91, 1431 (2002) [13] J. F. Moulder, W. F. Stickle, P. E. Sobol, and K. D. Bomben, Handbook of X-ray Photoelectron Spectroscopy, Physical Electronics Division, second ed., (1995) [14] R. C. Weast, Handbook of Chemistry and Physics, p. D-61, CRC press, 55 th ed., (1975) [15] D. Briggs, and M. P. Seah, eds. Practical surface analysis, 2nd edn. Wiley, Chichester. (1990) [16] K. Maki, Pauli paramagnetism and superconducting state. II. Prog. Theor. Phys. 32, (1964). 7.4 [17] R. Meservey, P. M. Tedrow, and R. C. Bruno, Tunneling measurements on spin-paired superconductors with spin-orbit scattering. Phys. Rev. B 11, 4224 (1975). [18] J. S. Moodera, J. Nassar, and G. Mathon, Spin-tunneling in ferromagnetic junctions. Annu. Rev. Mater. Sci. 29, 381 (1999) [19] T. H. Kim and J. S. Moodera, Enhanced ferromagnetism and spin-polarized tunneling studies in Co-Mn alloy films. Phys. Rev. B 66, (2002) [20] R. C. Sousa, J. J. Sun, V. Soares, P. P. Freitas, A. Kling, M. F. da Silva, and J. C. Soares, Large tunneling magnetoresistance enhancement by thermal anneal. Appl. Phys. Lett. 73, 3288 (1998) [21] P. V. Paluskar, C. H. Kant, J. T. Kohlhepp, A. T. Filip, H. J. M. Swagten, B. Koopmans, and W. J. M. de Jonge, Mn diffusion and the thermal stability of tunneling spin polarization. J. Appl. Phys. 97, 10C925 (2005). 1

157 145 Summary Summary Key concepts in spin tunneling Amorphous ferromagnets for spintronics This Ph.D. thesis is devoted to the fundamental understanding of the properties of ternary CoFeB alloys, and to an endevour in exploring open questions in spin tunneling by employing these properties. These ternary alloys have gained extreme importance in spintronic devices by showing large tunneling magnetoresistance (TMR) effects with AlO x and MgO based magnetic tunnel junctions (MTJs). In view of this emerging impact of CoFeB in various spintronics applications was obvious during the time of this thesis, so also was the necessity for a thorough experimental and theoretical analysis of its atomic and electronic structure and their combined impact on its tunneling spin polarization (P or TSP). To the first order, the TSP is representative of the inequality in the number of spin-up and spin-down conduction electrons at the Fermi level, and is the fundamental parameter in most spintronic devices. Returning to CoFeB, apart from a few of its bulk magnetic properties studied in the 1980 s every other structural, magnetic and electronic aspect of these alloys remained unexplored, especially in thin films relevant to spintronics. Moreover, the realization that these as-deposited amorphous alloys undergo a radical structural transition (crystallization) after an anneal also opens up possibilities to explore issues in spin tunneling which remain hitherto unaddressed. In Chapter 1 of this thesis we portrayed a few contemporary notions regarding spin tunneling and described crucial materials, experiments, and results on such tunnel junctions. In Chapter 2, we looked at device fabrication methods and the various experimental analysis tools used in this thesis. Here, to exemplify the various techniques, a few experimental results relevant to later chapters were also presented. In particular, we described some unpublished work reporting (i) large exchange bias fields obtained for CoFeB alloys, (ii) and the development of a CIPT set-up using which we demonstrated CIPT-TMR in AlO x based MTJs and characterization of MgO and NiO barriers. In Chapter 3, we investigated some structural aspects of CoFeB alloys. We showed that as-deposited films grow amorphous and investigate the influence of crystallization of these amorphous alloys on their structural and magnetic properties after a single anneal. MOKE and XRD measurement showed that the films crystallized gradually with increasing temperature In Chapter 4, we investigated the atomic and electronic structure of a single CoFeB composition. We aimed to address two issues: (i) an issue which had never been investigated experimentally or computationally to date was addressed namely the TSP of an amorphous ferromagnet. We in-

158 Summary 146 vestigated why these ferromagnets exhibit such high TMR values in the first place and how one can quantitatively describe the TSP of a ternary amorphous alloy in conjunction to an amorphous barrier. (ii) we explored the correlation between ferromagnet morphology, its electronic structure and their combined impact on TSP. In other words, we investigated how a distinct and major structural alteration of the ferromagnetic electrode at the interface with the barrier influences the TSP. We showed that amorphous Co 72 Fe 20 B 8 shows a considerably large TSP of 53%, and contrary to one s primary intuition, this TSP value was found to be larger than that observed for fcc CoFeB. First-principles atomic structure calculations showed good agreement with extended x-ray absorption fine structure measurements on amorphous CoFeB. Remarkably, both for amorphous and fcc CoFeB, electronic structure calculations based on the calculated atomic structure exhibited a conspicuous agreement between the spin polarization of the s-electron density of states and experimentally measured TSP. We emphasize that such a quantitative agreement between theory and experiment for a complex amorphous / crystalline ternary alloy has never been reported before. Moreover, the first principles calculations also revealed that the B sp-states were highly spin polarized and made a significant contribution to the alloy spin polarization. We believe that this aspect too is of significant relevance to spin-torque based MTJs and nanowires, as well as conventional MTJs employing CoFeB alloys. In Chapter 5, we probed some aspects of inelastic tunneling of electrons when a sharp contrast structural change from amorphous to crystalline electrode was induced at the barrier-ferromagnet interface. In particular, the changes in the low energy magnetic excitations induced by inelastically tunneling electrons were investigated. For amorphous CoFeB at the interface, we saw indications of size quantization of the magnons. For fcc CoFeB at the interface, we saw distinct excitations around 10 mv which could also be related to magnon-assisted spin flip tunneling. With these observations, we demonstrated that IETS is a powerful tool to investigate the impact of interface structure changes in MTJs. In Chapter 6, we probed the correlation between magnetism and TSP in CoFeB alloys. Such a correlation had been an outstanding issue in spin tunneling since its first observation in We found that the amorphous CoFeB alloys are very suitable to address this issue. Our measurements showed that the alloy magnetic moment as well as the s-electron dominated TSP showed the Slater-Pauling behavior. XMCD measurements which probe the properties of d-electrons by synchrotron radiations show a crossover from weak to strong ferromagnetism in the Fe-DOS. To the best of our knowledge, this is the first observation of the Slater-Pauling behavior in transition metal alloys using the XMCD technique. We conclude that this magnetic crossover in the Fe-DOS, together with s-d hybridization, provides an intuitive understanding of the direct correlation between the magnetic moment and TSP. We also believe that the tunable electronic and magnetic properties of these CoFeB

159 147 Summary alloys allow access to engineer and advance the application potential of spintronic devices. Finally, in Chapter 7, we investigated the thermal stability of MTJs and the effect of high-temperature annealing. Specifically, the role of Mn diffusion from the antiferromagnets used to exchange bias one of the ferromagnetic layers was probed. We find that though Mn diffuses after annealing, it did not influence the TSP.

160 List of publications 148 List of publications Magnetic tunnel junctions H.J.M. Swagten and P.V. Paluskar Encyclopedia of Material Science and Technology, in press. Spin tunneling in junctions with disordered ferromagnets P.V. Paluskar, J.J. Attema, G.A. de Wijs, S. Fiddy, E. Snoeck, J.T. Kohlhepp, H.J.M. Swagten, R.A. de Groot, and B. Koopmans Physical Review Letters, 100, (2008). Impact of interface crystallization on inelastic tunneling in Al / AlO x / CoFeB P.V. Paluskar, F.L. Bloom, E. Snoeck, J.T. Kohlhepp, H.J.M. Swagten, and B. Koopmans Applied Physics Letters, 91, (2007). Correlation between magnetism and tunneling spin polarization P.V. Paluskar, R. Lavrijsen, M. Sicot, J.T. Kohlhepp, H.J.M. Swagten, and B. Koopmans Submitted. Controlling speed and efficiency of ultrafast demagnetization by direct transfer of spin angular momentum G. Malinowski, F. Dalla Longa, J.H.H. Rietjens, P.V. Paluskar, R. Huijink, H.J.M. Swagten, and B. Koopmans Submitted. Tunneling spin polarization and annealing of Co 72 Fe 20 B 8 H.J.M. Swagten, P.V. Paluskar, R. Lavrijsen, J.T. Kohlhepp, and B. Koopmans Journal of Magnetism and Magnetic Material 310, 2012 (2007). Influence of interface structure on the tunnelling spin polarization of CoFeB alloys P.V. Paluskar, J.T. Kohlhepp, H.J.M. Swagten, B. Koopmans, R. Wolters, H. Boeve, and E. Snoeck Journal of Physics D: Applied Physics 40, 1234 (2007).

161 149 List of publications Co 72 Fe 20 B 8 : Structure, magnetism, and tunneling spin polarization P.V. Paluskar, J.T. Kohlhepp, H.J.M. Swagten, and B. Koopmans Journal of Applied Physics 99, 08E503 (2006). Mn diffusion and the thermal stability of tunneling spin polarization P.V. Paluskar, C.H. Kant, J.T. Kohlhepp, A.T. Filip, H.J.M. Swagten, B. Koopmans and W.J.M. de Jonge Journal of Applied Physics 97, 10C925 (2005). Thermal stability of tunneling spin polarization C.H. Kant, J.T. Kohlhepp, P.V. Paluskar, H.J.M. Swagten and W.J.M. de Jonge Journal of Magnetism and Magnetic Materials 286, 154 (2004).

162 About the author 150 About the author Paresh Vijay Paluskar October 2, 1978 Born in Pandharpur, in the province Maharashtra, in India Elementary Education, Mumbai Diploma in Instrumentation Board of Technical Education, Mumbai Bachelor of Engineering, Instrumentation Dr. Babasaheb Ambedkar Marathwada University, Aurangabad Master of Science, Sensory Systems Technology Fachhochschule Karlsruhe, Germany Graduate work in the field of giant magnetoresistance and exchange biasing at the research headquarters of Robert Bosch GmbH Schillerhöhe, Stuttgart, Germany. Imprinting of different pinning directions in a PtMn and synthetic antiferromagnet based spin valve 360 angle sensor using a current pulse and showing giant magnetoresistance Courses in Physics Eindhoven University of Technology, the Netherlands University of Bielefeld, Germany Ph.D. candidate Department of Applied Physics, Eindhoven University of Technology, the Netherlands Research carried out in the group Physics of Nanostructures Key concepts in spin tunneling: amorphous ferromagnets for spintronics