pure.tue.nl › ws › files › 3709443 › 200610857.pdf · de promotiecommissie bestaat uit:...

Ultrathin Co films for magnetoresistive devices : an NMRstudyCitation for published version (APA):Wieldraaijer, H. (2006). Ultrathin Co films for magnetoresistive devices : an NMR study. Technische UniversiteitEindhoven. https://doi.org/10.6100/IR609109

DOI:10.6100/IR609109

Document status and date:Published: 01/01/2006

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 version of the article upon submission and before peer-review. There can beimportant differences between the submitted version and the official published version of record. Peopleinterested in the research are advised to contact the author for the final version of the publication, or visit theDOI 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 pagenumbers.Link to publication

General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand 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.

If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, pleasefollow below link for the End User Agreement:www.tue.nl/taverne

Take down policyIf you believe that this document breaches copyright please contact us at:[email protected] details and we will investigate your claim.

Download date: 30. Jul. 2020

https://doi.org/10.6100/IR609109

https://doi.org/10.6100/IR609109

https://research.tue.nl/en/publications/ultrathin-co-films-for-magnetoresistive-devices--an-nmr-study(c4f2464e-9358-4c9d-a309-275a7b9f1976).html

Ultrathin Co Filmsfor Magnetoresistive Devices:

an NMR study

Harm Wieldraaijer

De promotiecommissie bestaat uit:prof.dr. R. Coehoorn, lid commissie Techn. Universiteit Eindhoven

en Philips Research Laboratoriesprof.Dr.rer.nat. M. Farle, lid commissie Universitat Duisburg-Essenprof.Dr. B. Hillebrands, lid kerncommissie Techn. Universitat Kaiserslauternprof.dr.ir. W.J.M. de Jonge, 1e promotor, Techn. Universiteit EindhovenDr.rer.nat. J.T. Kohlhepp, copromotor, Techn. Universiteit Eindhovenprof.dr. B. Koopmans 2e promotor, Techn. Universiteit Eindhovenprof.dr.ir. K. Kopinga, lid kerncommissie Techn. Universiteit Eindhovenprof.em.dr. P.C. Riedi, lid kerncommissie University of St. Andrews

CIP- DATA LIBRARY TECHNISCHE UNIVERSITEIT EINDHOVEN

Wieldraaijer, Harm

Ultrathin Co films for magnetoresistive devices : an NMR study / byHarm Wieldraaijer. - Eindhoven : Technische Universiteit Eindhoven,2006. - Proefschrift.ISBN-10: 90-386-2481-6ISBN-13: 978-90-386-2481-5NUR 926Trefwoorden: magnetische dunne lagen / kernspinresonantie / magnetoweerstand /kristalstructuren / kobaltSubject headings: magnetic epitaxial layers / nuclear magnetic resonance / magne-toresistance / crystal phases / cobalt

Printed By: Universiteitsdrukkerij Technische Universiteit Eindhoven.

The work described in this thesis has been carried out in the group Physics of Nanos-tructures at the Eindhoven University of Technology, Department of Applied Physics.

The cover shows cannon balls stacked in the bcc, fcc and hcp phase [Vonnegut63].Artist impression by B. Smits and H. Wieldraaijer.

Ultrathin Co Filmsfor Magnetoresistive Devices:

an NMR study

PROEFSCHRIFT

ter verkrijging van de graad van doctoraan de Technische Universiteit Eindhoven

op gezag van de Rector Magnificus, prof.dr.ir. C.J. van Duijn,voor een commissie aangewezen door het Collegevoor Promoties in het openbaar te verdedigen op

woensdag 21 juni 2006 om 16.00 uur

door

Harm Wieldraaijer

geboren te Eindhoven

Dit proefschrift is goedgekeurd door de promotoren:

prof.dr.ir. W.J.M. de Jongeenprof.dr. B. Koopmans

Copromotor:Dr.rer.nat. J.T. Kohlhepp

Contents

1 Introduction 1

2 Techniques and Structures 52.1 Experimental Techniques . . . . . . . . . . . . . . . . . . . . . . . . . 62.2 Thin Co Films . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.3 NMR on Thin Co Films . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3 Co Structures for Magnetoresistive Devices 353.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363.2 FCC vs. HCP Co in Sputtered MTJ’s . . . . . . . . . . . . . . . . . . 373.3 BCC Co(001) as a Bottom Electrode . . . . . . . . . . . . . . . . . . . 46

4 Epitaxial Co Films on Cu(001) 654.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 674.2 Structure and Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . 684.3 Influence of the Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . 764.4 Influence of the Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

Bibliography 98References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98List of Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

Abstract 115English . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115Dutch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

Curriculum Vitae and Dankwoord 121

v

vi CONTENTS

Chapter 1

Introduction

This thesis deals with the analysis of ultrathin cobalt (Co) films by nuclear magneticresonance (NMR). Co films are useful as magnetic layers in many thin-film structures,such as magnetoresistive devices. Before providing the set-up of the thesis, a verybrief introduction in thin-film technology, magnetoresistive devices, and NMR will begiven.

The use of thin films in devices provides a much wider range of obtainable prop-erties than using only bulk materials, even though the production of thin-film devicesneeds more sophisticated technology, such as ultrahigh vacuum deposition under ac-curately controlled conditions. The versatility originates in the methods of thin-filmproduction, which often consist of depositing atoms on another material and allowingthem to form their own preferential structure. When the deposition is done veryslowly, it is possible to obtain an ideal structure with hardly any of the impuritiesor defects that often abound in bulk materials. In this way very thin films can beobtained, between one layer and several thousands of layers of atoms, that are veryflat and possess a well-defined structure.

There are many advantages of thin or ultrathin films with respect to bulk mate-rials. Firstly, there is the possibility of obtaining new structures that do not exist inthe bulk, but can be forced by growing the material onto another material on whichit will only fit in the requisite structure. These new structures may have different(electronic) properties and can be used to study the relations between structure andproperties. The thin-film method allows for the study of a wider range of structuresthan the use of alloys or precipitation and even for the creation of phases that arenot in the phase diagram [Wuttig04]. Secondly, some of the properties are directlyinfluenced by the limited film thickness, leading to a behaviour of the material that isnot possible in the bulk. Examples of this are quantum size effects and effects relatedto the electron mean free path being of the same order as the thickness. Thirdly,some thin films have a slightly different structure than the bulk, which may lead tosignificant differences in the properties [Kim02].

A particularly interesting sub-field of thin-film studies is the growth of magneticthin films. These lead to many more potential applications than normal metal films(which are mainly interesting for coatings and for enhanced chemical activity) in

1

2 CHAPTER 1. INTRODUCTION

that they make devices combining electronic and magnetic properties possible, theso-called magnetoelectronics or spintronics [Wolf01]. The most common type of mag-netoelectronic device is the magnetoresistive one, where the magnetic state of thestructure is related to the electric resistance.

An example of such a magnetoresistive device is a giant-magnetoresistance (GMR)structure consisting of a stack of alternating ferromagnetic and non-magnetic thinmetal films (for example Co and Cu). In principle all layers conduct electric currentswell. However, electrons moving through a ferromagnetic film will have their spinoriented either parallel or anti-parallel to the magnetic moment of the film. Electronswith their spin parallel can move freely through the magnet, while electrons withtheir spin anti-parallel are often scattered and experience a large resistance. If allmagnetic moments in the GMR structure have the same orientation then half of theelectrons can freely move through the structure and the electric resistance is quitelow. On the other hand, if the moment of each layer is oriented anti-parallel to itsneighbours, all electrons will regularly encounter layers in which they will be scatteredoften and as a result the measured electric resistance will be high. In this way theGMR device has two stable states that can easily be read out by measuring theelectrical resistance. Switching between the states can be performed by means of amagnetic field [Baibich88, Binasch89, Barthelemy99].

Other kinds of magnetoresistive devices are tunnel-magnetoresistance (TMR) struc-tures and spin-valve transistors (SVT) [Monsma95, Jansen01]. A TMR structure(often called magnetic tunnel junction, MTJ) consists of two magnetic electrodesseparated by a very thin (∼ 1–2 nm) insulator. Due to the thinness of the insu-lator, electrons can tunnel through the layer when a voltage is applied. The re-sulting current is strongly dependent on the alignment of the electrode magnetiza-tions [Moodera95, LeClair02b]. These layered magnetoresistive structures (GMR aswell as TMR) are very useful for sensing magnetic fields (as, for example, used toread out bits on a computer hard drive) or for creating magnetic memories.

Most magnetoresistive devices depend on thin-film technology for their creation.Since the electronic properties of the films may depend strongly on the physical struc-ture it is very important to be able to determine and analyse the exact structuralproperties of the films. Also, by varying the structure of the materials and studyingthe changes in device behaviour, the detailed physics behind the behaviour may bestudied.

Of particular interest for this kind of study are metastable metal phases: theyprovide multiple structures for a single material, effectively leading to new materialsthat can be used in the devices. They also make it possible to use different phases ofthe same element and compare their effect on the device behaviour. In this way theinfluence of the physical structure can be separated from other effects. A demand onthe metastable phases is the stability of the structure under coverage with anothermaterial since in a GMR device or MTJ these additional layers are necessary andcannot be selected at will. This necessitates the study of the structure and stability ofthe metal electrodes while they are covered by different layers instead of the commonmethod of analysing uncovered layers by surface methods.

A magnetic material commonly used in these devices is Co. Co has the advantagethat it can be stabilized in several (meta)stable phases and furthermore that there is a

3

powerful tool available for a precise analysis of the structure. This tool is 59Co nuclearmagnetic resonance (NMR). NMR can determine in which atomic environments Coatoms are present and can thus easily determine the crystalline phase of the Co, thepresence of impurities and the type of interface between the Co and the other layers.An advantage of NMR over many other characterization methods is the possibility tostudy the Co layers while they are actually inside the device layer structure instead ofonly being able to measure the structure of the Co before depositing the next layer.

In this thesis several Co-metal structures are studied by 59Co nuclear magneticresonance (NMR) in order to investigate the stability and properties of different Cophases in these layers with applications in spintronic devices or model systems inmind. The thesis follows some other theses on the subject of 59Co NMR on Costructures from the same research group [Gronckel93, Alphen95, Strijkers99].

In chapter 3 two studies are presented on the occurrence and stability of Co phasesin MTJ bottom electrodes. The first study presents results on the phases that occurby sputter deposition on two different substrates. The different phases occurring havebeen shown to lead to significant and quantitatively understandable differences in theelectronic behaviour of MTJ’s [LeClair02a]. The second part of the chapter is a studyon the applicability of bcc Co as a magnetoresistive bottom electrode. Bcc Co is astrain-induced Co phase that can be grown up to a thickness of about 2 nm, however,the stability of the phase under coverage with a different material has not been studiedprior to our investigation.

In chapter 4 an extensive study on the behaviour of thin Co films grown on Cu(001)is presented. This system provides an ultrahigh-quality strained fcc-Co phase for awide range of Co thicknesses. Due to the high structural quality some new observa-tions are possible, namely the observation of the influence of the conduction-electronpolarization on the Co hyperfine field over several atomic layers next to an interfaceand that of the presence of homogeneous electric-field gradients (EFG’s) due to thestrain in the film.

Most of the results in this thesis have been published previously as articles and thisformat has largely been kept, although this may sometimes lead to the repetition of,for example, experimental details. However, most sections have been slightly extendedwith respect to the articles, either by more detailed information that was left out ofthe articles or by a somewhat more complete analysis as in the case of section 3.2.The sections in chapter 4 are reordered with respect to the original articles in orderto improve the readability and the consistence.

4 CHAPTER 1. INTRODUCTION

Chapter 2

Techniques and Structures

Abstract: In this chapter a short overview of both the experimental methods and thetypes of samples used will be given. Nuclear magnetic resonance (NMR) is presentedin some detail while the other techniques are described briefly. A short overview of theparticular properties of thin Co films is presented, concentrating on the interfaces, thecrystal phases, and the possibility of strain. Finally, the details of 59Co NMR concern-ing the analysis of thin Co films are given, together with the thin-film properties andstructural possibilities of thin Co films. For this last part our own measurements areused as a basis. Some of the presented analyses may be an addition to the standardrepertory of 59Co-NMR characterization.

Contents

2.1 Experimental Techniques . . . . . . . . . . . . . . . . . . . 6

2.1.1 Sample Preparation . . . . . . . . . . . . . . . . . . . . . 6

2.1.2 Nuclear Magnetic Resonance . . . . . . . . . . . . . . . . 7

2.1.3 Structural Characterization Techniques . . . . . . . . . . 15

2.1.4 Other Techniques . . . . . . . . . . . . . . . . . . . . . . . 17

2.2 Thin Co Films . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.2.2 Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.2.3 Crystal Phases . . . . . . . . . . . . . . . . . . . . . . . . 20

2.2.4 Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.3 NMR on Thin Co Films . . . . . . . . . . . . . . . . . . . 24

2.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.3.2 Co Hyperfine Fields . . . . . . . . . . . . . . . . . . . . . 24

2.3.3 Measuring Interfaces . . . . . . . . . . . . . . . . . . . . . 26

2.3.4 Measuring Crystal Phases . . . . . . . . . . . . . . . . . . 27

2.3.5 Measuring Strain . . . . . . . . . . . . . . . . . . . . . . . 32

5

6 CHAPTER 2. TECHNIQUES AND STRUCTURES

2.1 Experimental Techniques

2.1.1 Sample Preparation

Molecular Beam Epitaxy

Molecular beam epitaxy (MBE) is a growth technique which operates by means oflow-energy atom beams condensing on a substrate in ultra-high vacuum (UHV). Dueto the vacuum (and the small sticking coefficients of residual impurity atoms) it isable to produce extremely pure layers in a very controllable way. The atom beam iscreated by thermal (Knudsen cell) or electron-beam (e-gun) evaporation of a material.The MBE-grown layers in this thesis were deposited with a VG Semicon V80M multi-chamber MBE system. Materials like Fe, Co, and Ni are grown from e-gun evaporatorsand others (e.g. Cu, Si, Au) are evaporated by means of Knudsen cells. The substratetemperature can be accurately set during deposition and the sample can be rotatedduring growth in order to obtain optimum homogeneity. The thickness of the filmsis monitored by a calibrated quartz-crystal microbalance. Typical growth pressure isbetter than 10−10 mbar and growth rates are in the order of 10−3 to 10−2 nm/s.

After deposition the layers can be transported in UHV to special chambers foranalysis (see section 2.1.3). MBE growth is mainly used for obtaining epitaxial layerswith well-defined and highly crystalline structures. In order to obtain an optimumquality, the substrates are usually cleaned by sputter- and anneal-treatments untilthe surface is atomically flat and clean. More details on MBE growth can be found,for example, in [Herman04].

Sputtering

A faster (and industrially more common) method for depositing materials is sputterdeposition. In this method material is removed and ejected from a target material bybombarding it with energetic particles, usually Ar+ ions in order to avoid chemicalreactions, and subsequently deposited on the sample. The sputtered layers in thisthesis were grown by DC and RF magnetron sputtering, where an Ar plasma isconfined near the target by a magnetic field. The plasma itself is generated by a large(DC or RF) voltage between the target material and a nearby metallic plate. Theapparatus used is a Kurt J. Lesker sputter chamber. The growth pressure is typically10−2 mbar (background pressure better than 10−9 mbar) and growth rates are in theorder of 10−1 nm/s.

There are many parameters that may have a strong influence on the resulting layerproperties. Generally, sputter deposition leads to less crystalline material than MBEgrowth due to the higher background pressure, the higher growth rate, and the higherenergy of the impinging material. Its advantages are that almost any material can bedeposited and that many (and thick) layers can be grown in a relatively short time.Also relevant is the fact that industrial processes often also use this procedure, allow-ing the study of structures similar to those applied in industry. Shadow masks can beused for patterning the layers in order to create, for example, the crossed-electrodeconfiguration used in magnetic tunnel junctions. A more extensive description can befound in [Ohring92, LeClair02b].

2.1. EXPERIMENTAL TECHNIQUES 7

Plasma Oxidation

The oxidic layers (AlOx) used in this thesis are created by plasma oxidation of asputtered aluminium film. The samples can be transported in UHV from the sputterchamber to the plasma-oxidation chamber. The oxygen plasma is created by fillingthe chamber with about 0.1 mbar of oxygen gas while applying a voltage of 2 kVto the electrodes. In this way a stable and static oxygen plasma is produced veryrapidly. The plasma is a rotationally symmetric DC glow discharge with a plate- anda ring-shaped electrode. The sample is grounded during oxidation in order to avoidcharge accumulation. Typical times necessary for the complete oxidation of about2 nm-Al layers are about 200 seconds. A more detailed description may be foundin [Knechten05].

2.1.2 Nuclear Magnetic Resonance

General

After its discovery, nuclear magnetic resonance (NMR) [Purcell46, Bloch46] rapidlybecame one of the most important techniques for the structural analysis of manysystems. Since then, it has become a standard and indispensable tool for a widerange of disciplines and many different types of material [Castellani02, Schultz99].

In an NMR experiment the Zeeman splitting of nuclear ground-state levels isprobed by radio-frequency (RF) fields in order to determine the effective magneticfield causing the splitting. A nucleus with nuclear spin I has 2I+1 ground-state energylevels, which are degenerate in the absence of a magnetic field. When a magnetic fieldis present, the levels become non-degenerate with equidistant energies (see figure 2.1).This means that the energy difference between any two consecutive levels is equal and,thus, the transition between consecutive levels always has the same NMR resonancefrequency ν, which is proportional to the effective magnetic field, Beff ,

νL =γ

2π|Beff |, (2.1)

with γ the gyromagnetic ratio1 [Abragam61].These resonance frequencies can be determined, for example, by measuring the

absorption of the RF intensity as a function of frequency, by exciting all frequencieswith a very short RF pulse and analysing the reaction of the system, or, alternatively,by the so-called spin-echo method [Hahn50, Slichter63], which will be described inmore detail further on.

Apart from the Zeeman splitting there may also be a quadrupolar contribution tothe splitting. Nuclei with a spin I > 1

2 have a nuclear quadrupole moment, Q, whichinteracts with the electric-field gradients (EFG’s) present in a non-cubic environment.In addition to the equal Zeeman splittings between adjacent energy levels, there is anenergy term which depends on the energy-level number. As a result, for the commoncase where the Zeeman splitting is much larger than the quadrupole splitting, thelowest two levels have the smallest splitting, the next two a slightly larger, and so on(see figure 2.1). Because of this, the resonance frequencies for a transition between

1This ratio is actually the magnetogyric ratio, but this more accurate name has fallen into disuse.


m = - 7 / 2m = - 5 / 2m = - 3 / 2m = - 1 / 2m = + 1 / 2m = + 3 / 2m = + 5 / 2m = + 7 / 2

m = - 7 / 2m = - 5 / 2m = - 3 / 2m = - 1 / 2m = + 1 / 2m = + 3 / 2m = + 5 / 2m = + 7 / 2

C o : I = 7 / 2

B = 0E F G = 0

B > 0E F G = 0

B > 0E F G > 0

Energy d e g e n e r a t e g r o u n d s t a t e

Figure 2.1: Schematic energy levels for three different cases. Co is taken as an example,having both a nuclear spin (I = 7/2) and a (negative) quadrupole moment. On the leftside the situation in the absence both of a magnetic field and an electric-field gradient isdrawn: all 2I = 8 Co ground-state levels are degenerate. In the middle the case where amagnetic field is present and electric-field gradients are absent is given. Zeeman splittingoccurs: the levels become non-degenerate, m < 0 levels move to a higher energy, m > 0levels to a lower energy. The splitting between any consecutive levels is equal as indicatedby the vertical arrows. Since the NMR frequency is proportional to the energy splittingbetween consecutive levels, all transitions occur at the same frequency and only a single lineis present in the NMR spectrum. On the right side the case where both a magnetic fieldand an electric field gradient are present is drawn. The quadrupole interaction causes the±7/2 and ±5/2 levels move up in energy and the ±3/2 and ±1/2 levels to move down. Nowthe energy splitting between two consecutive levels is different for each two levels, in fact itincreases linearly with the level number if the quadrupole interaction is much smaller thanthe Zeeman splitting. As a result, each transition has a different frequency (see verticalarrows) and the NMR spectrum consists of (in this case) seven evenly spaced lines.

consecutive levels are no longer equal for all levels, but are larger for the higherlevels. Instead of a single resonance frequency one now gets 2I frequencies that areequidistant in frequency:

ν = νL + νQ =γ

2π|Beff |+ 3eQ· ↔V ζ

2I(2I − 1)h·(

m− 12

), (2.2)

with↔V ζ the component of the EFG tensor in the direction of the total effective

field and m the nuclear-spin component in the direction of the total effective field(−I ≤ m ≤ I) [Pound50, Abragam61].

Apart from being able to measure the distribution of effective fields and, poten-tially, EFG’s in the sample, it is also possible to determine the relative numbers ofnuclei feeling a specific field. This information can be determined from the intensityof the signal. For the commonly used inductive measurements the intensity of the


signal, S, is given by

S ∝ γ3B2effI(I + 1)

T·N ∝ γν2I(I + 1)

T·N (2.3)

with T the temperature and N the number of nuclei [Panissod86]. From this equationthe relative NMR sensitivity for different nuclei can also be determined. The highestsensitivity occurs for 1H, which is one of the reasons for the general use of hydrogenNMR in medical imaging (MRI), biology and organic chemistry. Other isotopes forwhich NMR is relatively sensitive are, for example, 19F, 51V, 59Co and 93Nb. However,it has to be realized that equation 2.3 ignores some dependencies of the signal intensity,such as the signal enhancement in ferromagnetic materials (see further on) or thespecifics of the measurement method.

Spin-Echo NMR

The most common NMR methods are pulsed-RF methods, among which the spin-echotechnique [Hahn50]. In these methods only a short but intense RF pulse is appliedperpendicular to the effective field by means of a small coil. The pulse rotates thenuclear moments out of their equilibrium direction, which is parallel to the effectivefield. After the pulse each group of nuclear moments precesses around the effectivefield it feels. This precession causes an inductive voltage in the coil, the signal. Dueto the distribution of fields present, the moments will get rapidly out of phase and thesignal will decrease. The decrease is called free-induction decay (FID). By measuringthe FID, the entire effective-field spectrum can be determined at once by means of aFourier transform. However, this is only feasible if the distribution of effective fieldsis very narrow, e.g. in para- or diamagnetic materials, where an externally appliedfield is the main contribution to the effective field. If the field distribution is wider,the FID will be too fast to measure and analyse, and a different method has to beused: the spin-echo method.

In spin-echo NMR two RF pulses are applied with a short time interval, τ , inbetween. After the first pulse FID will occur because of the dephasing of the nuclearmoments, however, the second pulse flips the moments around so that they startrephasing instead. The second pulse actually mirrors the precession of the momentsand, thus, at a time τ after the second pulse they will have the same relative orien-tations as they had immediately after the first pulse. At this moment an inductivesignal is obtained, the spin echo, see figure 2.3. The echo only occurs for those nucleithat have the correct precession frequency (equal to the applied RF frequency). Byvarying the RF frequency and measuring the spin-echo intensity, the distribution ofeffective fields can be measured. A more extensive description of the principle of spin-echo NMR can be found in [Hahn50, Slichter63, Panissod97, Strijkers99]. Figure 2.3shows the effects of a 90− 180 pulse series. In experiments usually equal pulses areused (α−α instead of α−2α), which function in a similar way, but give slightly moresignal when a wide distribution of hyperfine fields is present.

When measuring spin echoes, one effect has to be taken into account, namely thatnot all nuclear moments rephase at the moment of the spin echo, but only those that inthe mean time have not interacted with one of the other spins. The presence of these


Pulse o

r Sign

al Inte

nsity

T i m et i m e t t i m e t

f i r s t R F p u l s e s e c o n d R F p u l s e

F I D s i g n a l

s p i n -e c h o

e x p o n e n t i a l e c h o d e c a yd u e t o s p i n - s p i n r e l a x a t i o n

Figure 2.2: Schematic representation of the intensity envelopes of the RF pulses (grey)and of the signals (black) as a function of time for a typical spin-echo experiment. Thepulses are actually many orders of magnitude larger than the signals. Immediately afterthe first RF pulse, a fast-decaying signal is present, the free-induction decay (FID). A timeτ after the first pulse, a second similar RF pulse is given. An additional time τ later, asignal appears again, the spin-echo. This has a smaller intensity than the FID due to theexponential decrease of the signal as a function of time caused by the interactions betweenthe spins (spin-spin relaxation).

spin-spin interactions causes an exponential decrease of the spin-echo intensity as afunction of the time, 2τ , between the first pulse and the final echo (figure 2.2). Thisdecrease is described by the characteristic time of the signal decay, which is called T2

and is a measure of the strength of the interactions, where a short T2 corresponds toa strong interaction. T2 can easily be measured by determining and plotting the spin-echo intensity as a function of τ . Another option is the use of more complicated pulsesequences, which is faster but provides a less detailed measurement [Slichter63]. Thedecay behaviour not only influences the measurement but can also be used to obtaininformation based on the spin-spin relaxation time in the sample (see section 2.3).

In principle, all NMR spectra must be corrected for possible variations in spin-spinrelaxation time with frequency. However, since the precise decay behaviour is oftennot exactly single-exponential in thin films, this is in effect hard to do and of doubtfulaccuracy. The correction has not been applied to any of the measurements in thisthesis.

The decay may be more complicated than a single-exponential decay, particularlyif the NMR resonance line is split into unresolved sub-lines. In this case, interferencebetween magnetization movements of nuclei in the different sub-lines may cause amodulation to be superimposed on the exponential decay [Hahn52] (see section 2.3.4:figure 2.8(b)).

A second effect influencing the measurement is the non-perfect alignment of thenuclear magnetization with the effective field. After a measurement the magnetizationis almost perpendicular to the field, whereas it should be parallel to it before the startof a new measurement. This implies that, between two measurements, a long enoughtime has to pass for the magnetization to return to its equilibrium direction. Thisreturn is exponential and has a characteristic time T1, the spin-lattice relaxation


z

y 'x '

Mz

B r fM

z

M

z

B r f

M

z

M

z

M

( a ) i n i t i a l s i t u a t i o n ( b ) f i r s t p u l s e ( 9 0 0 ) ( c ) d e p h a s i n g ( F I D )

( d ) s e c o n d p u l s e ( 1 8 0 0 ) ( e ) r e p h a s i n g ( f ) s p i n - e c h o

Figure 2.3: Schematic drawing of the motion of the nuclear magnetization in a rotatingframe during the application of a 90−180 pulse series. The hyperfine field at the nucleus isoriented in the z-direction and the magnetization precesses around that field. The system isusually described in a reference frame that rotates around the z-axis with a frequency equalto the precession frequency of the nuclear moments. In this description the effect of thehyperfine field is cancelled. The rotating frame has the advantage that the magnetizationbehaviour becomes very much simplified: without additional fields the average magnetizationhas a constant orientation and the effect of an RF pulse with a frequency equal to theprecession frequency is a simple rotation around one of the rotating-frame axes, instead of acomplex corkscrew movement. (a) In the initial situation, the magnetization lies along thez-axis. (b) The first RF pulse, which has the same frequency as the precession, is directedalong the x′-axis of the rotating frame. In the rotating frame this is the only field present,so the nuclei precess around this field. The pulse is kept active until the magnetizationhas rotated over 90. (c) Dephasing: due to variations of the hyperfine field with positionsome groups of nuclei precess faster than others, as a result they spread out in the rotatingframe. (d) Second RF pulse which rotates all moments around the x′-axis over 180. (e)Rephasing: after the 180 pulse, the moments precessing fastest are suddenly behind andthe slower precessing ones in front. As a result they will all move towards each other again.(f) Spin-echo: when the moments have rephased for the same time duration as they hadpreviously dephased they all lie together in the rotating frame. In the real world they areprecessing together around the z-axis, inducing a voltage in the RF coil around the x axis.


time, which in solids is usually several orders of magnitude larger than the spin-spinrelaxation time. The spin-lattice relaxation time can be measured by means of theso-called Comb sequence, in which a large number of RF pulses is given in order torandomize the magnetic moments and a certain delay time later a spin-echo pulsesequence is given. If the spin-echo intensity is plotted as a function of the delay timeafter the Comb pulses, the characteristic time, T1, can be determined. According tothe Korringa-model, this spin-lattice relaxation time is inversely proportional to thetemperature [Korringa50].

A more detailed description can be found in [Slichter63].

Spin-Echo NMR on Magnetic Materials

When NMR is used on ferromagnetically ordered materials some specific features showup. The most important one is that, contrary to the case of para- and diamagneticmaterials, the externally applied field is not the main contribution to the effectivefield. The total static field experienced by a nucleus can be given as

~Beff = ~Bappl + ~Bdip + ~Bhf (2.4)

with ~Bappl the externally applied field, ~Bdip the dipolar field resulting from the elec-tronic moments of all other atoms, and ~Bhf the hyperfine field, describing the in-teraction of the nucleus with its directly surrounding electrons. The hyperfine fieldin ferromagnetically ordered materials can be tens or even hundreds of Tesla, muchlarger than the other fields.

The external field in NMR applied on magnetic materials is usually not higherthan a few Tesla. The maximum sensitivity is obtained without an applied field(see figure 2.4), so the applied field is often, and especially in thin-film analysis,zero. Although the dipolar field from surrounding atoms (in non-cubic environments)can be felt by a nucleus, the major contribution to the dipolar field generally isthe demagnetizing field caused by the macroscopic shape of the sample. This hasa value somewhere between 0 and −µ0M , with M the magnetic moment per unitvolume, and therefore at most a few Tesla. In thin magnetic films this field is givenby ~Bdip = −µ0

~M cos θ, with θ the angle between the magnetic moment and the filmnormal. Consequently, for most films that have their magnetization in plane thedemagnetizing field is zero. When the magnetization is oriented perpendicular to thefilm, the field is equal to −µ0

~M .For calculational purposes the hyperfine field is commonly split into terms orig-

inating in the interaction with different electron shells (see section 4.3.4). However,actual values of the hyperfine field for different materials and environments are usu-ally determined empirically from reference studies since the accuracy with which theycan be calculated is not high enough. This is caused by some large and nearly equal,opposing contributions encountered in the calculations [Abragam61, Guo96].

The hyperfine field is mainly determined by the magnetic moment of the atomitself and by the magnetic moments of directly neighbouring atoms. Although smalleffects of foreign atoms at a larger distance may be observed in dilute alloys and, forexample, near perfect Cu(001)/Co interfaces (see section 4.2), the main influence isthat from the directly neighbouring atoms of the first shell. This makes NMR on


ferromagnetic materials a technique providing ideal local sensitivity without muchinfluence of, for example, long-range disorder.

For crystal structures with a uniaxial rather than a cubic symmetry the hyperfinefield may be anisotropic (mainly due to an unquenched orbital moment) and can bedescribed by:

Bhf(θ) = Bhf,iso + 12Bhf,ani(3 cos2 θ − 1) (2.5)

where θ is the angle with the symmetry axis and Bhf,iso and Bhf,ani are the isotropicand anisotropic parts of the hyperfine field. For the common angles θ = 0 andθ = 90, equation 2.5 results in: Bhf(0) = Bhf,iso + Bhf,ani and Bhf(90) = Bhf,iso −12Bhf,ani. The anisotropic part is usually quite small: a few percent of the anisotropicpart.

The main structural features that influence the hyperfine field distribution and,therefore, can be determined from NMR measurements are: the presence of foreignatoms, e.g. at interfaces or in alloys, the crystallographic structure (fcc, bcc, hcp,stacking faults, grain boundaries) and the presence of strain. Some basic informationon these structures is given in section 2.2. Details about 59Co NMR on these structuresin thin Co films are given in section 2.3, including typical values of the various fieldsfor Co in section 2.3.2.

The changes in the hyperfine field with the local environment are often in theorder of Teslas and thus a broad NMR spectrum will result. A measurement of theFID is not an option for these materials since the signal usually decays to zero withina microsecond. Therefore the spin-echo method is needed and the applied field orfrequency has to be scanned.

A final typical aspect of NMR on ferromagnetically ordered materials is the pres-ence of the so-called RF enhancement in these samples. When NMR signals wereobserved for the first time in a ferromagnetic material (Co), it was seen that the ef-fect was several orders of magnitude larger than expected [Gossard59]. Furthermore,in spin-echo experiments the RF pulse power needed to obtain a signal is found to beseveral orders of magnitude lower than expected in first instance. The effect is associ-ated with the large local fields at the nuclei, originating in the electron-spin moments( ~Bloc = ~Bdip + ~Bhf), see figure 2.4. The applied RF field causes an angular displace-ment of this local field. The transverse component of the local field is the actual RFfield felt by the nuclei and is several orders of magnitude larger than the applied RFfield. The same effect also enhances the induced signal since the precessing nuclearmagnetization induces a coherent precession of the electronic magnetization.

The actual size of this enhancement factor is determined by the size of the fieldskeeping the local field in its original direction: the applied field and the anisotropy fieldin the sample. Enhancement in a typical fcc Co film is about 102 while enhancementwithin a domain wall is even larger, about 104. Because of this, NMR signals from amultidomain sample will mainly originate in nuclei situated in the domain walls andnot in those in the domains themselves.

Another important effect of the enhancement is that its value may vary withthe structural environment of the nuclei, which means that the spectrum has to becorrected for the enhancement at each frequency. This can be done by varying theRF-pulse power and measuring the spin-echo intensity. Generally the echo intensity


zM

B l o c

B r fB r f , e f f

B a p p l + B a

x

Figure 2.4: Schematic drawing of the origin of the enhancement of the RF field. TheRF field does not work directly on the local field felt by the nuclei, but on the electronicmagnetization, ~M . The magnetization is pulled out of the equilibrium position, the angle isdetermined by the ratio between the RF field and the sum of the anisotropy field and theapplied field. The local field at the nucleus is antiparallel to the magnetization, through whichthe nuclei actually feel ~Brf,eff , which is much larger than the applied RF field, particularlyif no external field, ~Bappl, is applied.

versus the RF power in dB shows a Gaussian profile. The maximum intensity, at thetop of the Gaussian, occurs when the enhanced pulse causes a nuclear-magnetizationrotation over a specific angle, which depends slightly on the specific situation butis about 120. If this occurs at a low RF-power value, the enhancement factor islarge. If the maximum occurs at a higher RF value, the enhancement is small. Sincethe out-coming signal is enhanced by the same factor as the ingoing RF pulses, thesignal from positions with a large enhancement is overestimated. By determining therelative enhancement from the RF power at which the maximum signal is found, thiseffect can be corrected for [Gronckel93, Panissod97].

More detailed descriptions of NMR on ferromagnetic materials can be foundin [Panissod86, Jonge94, Panissod97, Riedi99]. A more detailed description of thespecifics of NMR on thin Co films is given in section 2.3.

NMR Spectrometer Used

The system used in this thesis is a home-built pulsed-NMR spectrometer. The set-up can generate RF pulses between 0.1 and 2.0 µs with frequencies in the rangeof 25 to 500 MHz. The pulses are applied to the sample by means of a small coil


wrapped directly around the sample. In order to prevent reflection of the signal, thiscoil is part of an LC-circuit with two capacitors that can be automatically tunedto an impedance of 50 Ω for frequencies between 100 and 400 MHz. The spin-echois picked up by the same coil and sent to the filtering and amplifying equipmentthrough an electronic switch. It is split into two signals with 90 phase difference anddemodulated, after which its intensity is determined by means of a Fourier transform.For more details about the electronics see [Strijkers99]. Typical spectra for thinCo films are given in section 2.3. The sensitivity of the equipment is good enoughfor single Co films of a few ML in total to be measured. The equipment can beprogrammed to automatically measure different properties (e.g. spin-echo intensity,T1, T2) as a function of frequency, field or pulse power in a single run.

The sample is placed inside a cryostat in the bore of a superconductive magnet.Fields up to 5.5 T and temperatures down to 2.0 K can be reached. Typical measure-ment durations of a frequency scan are 1 to 10 hours. Some other typical values fora measurement on a Co film are: pulse separation 5 µs, delay between two measure-ments 10 to 100 ms (at liquid He temperatures), averaging per measurement point 16– 1024 times.

2.1.3 Structural Characterization Techniques

XRD and GIXA

X-ray diffraction (XRD) is one of the most common techniques for investigating thestructure of, especially crystalline, materials. The basic principle of XRD is theinterference of waves reflected from different crystal planes. To obtain this, radiationwith a wavelength comparable to the typical lattice spacing of the crystal is neededthat also has to be able to penetrate large distances into the lattice. X-ray radiationwith wavelengths in the order of 0.1 nm ideally suits both criteria. When such a beamimpinges upon a crystal, a small part of it will be reflected at each plane, and thewave vector perpendicular to the planes will determine whether the different reflectedwaves will interfere de- or constructively. The perpendicular component of the wavevector can easily be varied by changing the angle of incidence of the beam. Fromthe angle at which constructive interference occurs it is easy to determine the latticespacing, d, by means of the Bragg law

2d sin θ = nλ, (2.6)

where θ is the angle of incidence relative to the planes, n is an integer and λ is thewavelength of the X-ray. The analysis can also be performed for crystal planes thatare not parallel to the surface of the sample. This allows for a discrimination betweensingle-crystalline films and textured films, in which only the planes parallel to thesurface have the same orientation throughout the sample, and for a more full analysisof single-crystalline materials.

Furthermore, by means of the Scherrer formula [Snyder99], the length, D, overwhich the layer stacking is coherent can be estimated:

D =Kλ

β cos θ, (2.7)


with β the line broadening (full width at half the maximum intensity (FWHM)) andK the so-called shape factor, which is usually about 0.9.

The set-up used for the measurements described in this thesis is a Philips PWD3710X-ray diffractometer using a Cu Kα X-ray source. The wavelengths of the X-rays areλKα1 = 0.15405 nm and λKα2 = 0.15444 nm. The sample is exposed to the X-raysat an angle θ and the reflected rays are detected at an angle 2θ with respect to theoriginal beam.

When the X-rays are incident under a very small angle with the sample surface, thetechnique is often called grazing incidence X-ray analysis (GIXA) and is very usefulfor thin film analysis. At these angles reflection of the beam occurs at interfaces onlyand, by means of constructive or destructive interference fringes, film thickness androughness can be analysed. More information can be found in [Holy99].

LEED

Low energy electron diffraction (LEED) is a surface analysis technique in which anelectron beam of variable energy is diffracted at the sample surface and the result-ing diffraction peaks are measured using a fluorescent screen. The incoming beam isproduced by an electron gun (typical energies 0 − 1 keV) and impinges close to thesurface normal. The elastically scattered electrons are accelerated towards a fluores-cent screen after passing an initial voltage barrier in order to reject the inelasticallyscattered ones. Due to the relatively low electron energy the penetration depth is onlya few monolayer (ML) and the LEED pattern reflects the crystallographic symmetriesof the topmost layers.

A second possibility is the measurement of the LEED intensity of a particularLEED spot as a function of the electron energy, this is called LEED I-V . A fullanalysis of the resulting I-V curves is extremely difficult due to the strong interactionsbetween the electrons and the atoms. However, a simple kinematical estimate can bemade in which peaks correspond to energies for which the Bragg law is satisfied forthe electrons. In this way the average out-of-plane lattice constant, d, of the topmostlayers can be determined by

d =

√h2π2

2me·(

dE

d(n2)

)−1

· 1cos θ

=

√37.60313 eVdE/d(n2)

· Acos θ

, (2.8)

with me the electron mass, E the energy of the nth order mirror Bragg reflection andθ the angle of the electron beam with the film normal (θ = 6 in our case). It isimportant to keep in mind that this lattice spacing (and also the lattice structure)corresponds to that of the topmost monolayers and may be significantly different fromthe lattice spacing after depositing additional material.

The LEED used in this thesis is an OMICRON backview SpectaLEED apparatusconnected to the MBE system, which can thus be used for analysis during severalstages of the growth. A much used alternative to LEED is RHEED (reflection highenergy electron diffraction), which may provide information on the roughness and in-plane lattice parameter during growth. This has, however, not been used for the pre-sented studies. More information on LEED can, for example, be found in [Conrad96].


STM

A scanning tunnelling microscope (STM) is an apparatus in which a surface is scannedby means of a very sharp metallic tip. By means of piezo-electric crystals the tip isbrought to a distance of 0.2− 1 nm from the (conducting) surface to be investigatedwhile a bias voltage is applied. A tunnelling current occurs between the tip andthe sample and, by keeping the current constant while scanning the surface, thetopography of the surface can be traced with atomic resolution. However, some careshould be taken in the interpretation since STM actually measures the electronicstructure at the surface.

STM is ideally suited for determining the crystalline structure and quality andthe roughness of a surface. Also the presence of grains can easily be observed. TheSTM images presented in this thesis were recorded with an OMICRON UHV STM 1.More details of STM can be found in, for example, [Unertl96].

XPS and AES

Both in X-ray photo-electron spectroscopy (XPS) and in Auger-electron spectroscopy(AES) electrons coming from a sample during irradiation with either X-rays or anelectron beam are observed. In XPS X-rays (in this case either Al-Kα (1486.6 eV)or Mg-Kα (1253.6 eV) radiation) are incident on the sample and ionize atoms up toa few micrometre from the surface, usually by ejecting electrons from the core levelsof the atoms. The emitted electrons have a kinetic energy which is determined bythe energy of the X-rays minus the binding energy of the electron and the samplework function. In this way the binding energy, which depends on the chemical stateof the atom, can be determined. By XPS, for example, quantitative information onthe presence of Al and AlOx in a sample can be obtained in order to determine thestate of the oxidation process.

In AES the sample is irradiated by electrons with an energy between 2 and 5 keV.These electrons also eject a core electron from an atom they impinge on, but in thiscase the ejected electron is not measured. The hole in the inner atom shell will be filledby an outer electron and the released energy may be transferred to a second outerelectron that is then ejected. By measuring the energy of these secondary electrons,the chemical state of the atom can be determined. The surface sensitivity of XPSand AES is determined by the escape depth of the emitted electrons, which is in theorder of a nanometre, depending on the electron energy. More details can be foundin [Watts03].

2.1.4 Other Techniques

SQUID

A superconducting quantum interference device (SQUID) is used to measure themagnetization of a sample. In the SQUID superconducting coils are used to determinethe magnetic flux changes when moving the sample through the coils. SQUID isvery sensitive and provides an absolute magnetic moment (sensitivity in the order of10−10 Am2). The set-up used is a Quantum-Design MPMS-5S which can apply fields


up to 5 T and measure at temperatures between 1.7 and 400 K. More information onthe SQUID can be found in [Gallop91]. More general information on magnetism andmagnetic materials is given in [Craik95, O’Handley00].

Magnetoresistance Measurement

The magnetoresistance in tunnel-junction structures has been measured by meansof a home-built set-up equipped with a 1.35 T magnet and a 1.2 to 400 K variabletemperature helium flow cryostat. The measurements are performed by means of astandard four-point contact method, in which the current is varied while measuringthe voltage. V (I), dV /dI(I), and d2V /dI2(I) could be measured simultaneously bymeans of small, constant added ac-currents and analysed by using lock-in amplifiers.A detailed description of these techniques can be found in [LeClair02a, LeClair02b].

2.2 Thin Co Films

2.2.1 Introduction

Thin films can be deposited by various techniques on almost any substrate and undera wide range of conditions. As a result many physical structures are possible for afilm of any material. Since the precise structure of a film may have a large influenceon its properties (mechanical, electrical and magnetic), it is important to be able toanalyse and control the exact structure obtained. When the precise relations betweenthe growth conditions and the resulting structure are known, the film structures canbe manipulated almost at will.

In this section several possible structural properties of thin films are described.Since the field is a very extensive one, it will not be presented here in depth, but onlya short overview of some of the possibilities will be given. Several aspects are workedout in somewhat more detail for thin Co films and some remarks on the possibilityto obtain various structures are made. The section is split into three parts, dealingin turn with what are the most important aspects of thin films from our of pointview: the interfaces, the crystal phase (including the microstructure), and the strainin films.

2.2.2 Interfaces

The outstanding property defining thin films is, of course, the presence and impor-tance of either two interfaces or one interface and a surface. Since thin films areusually deposited on top of a substrate material or on top of a preceding stack of thinfilms, the interface in contact with that material may be termed ‘bottom interface’.Since film growth proceeds from this bottom interface, it is often of decisive influenceon the structure of the entire film. By thus choosing a suitable substrate and growthconditions the properties of the entire film can be influenced.

The top of the film can be bounded either by a free surface or by a second inter-face, usually called the ‘top interface’, with an additional layer. This second interface

2.2. THIN Co FILMS 19

generally has a less strong influence than the bottom one, but may still be of struc-tural relevance, by for example transforming the physical structure of the top part ofthe film. This transformation can have a large influence on measurements since thestructure and strain of a film are often investigated in situ by means of techniquesthat are sensitive to the surface structure (e.g. LEED or RHEED) while actually thestructure after growth of additional layers is of importance. The effect may then leadto significant differences between measurement and reality.

Apart from their influence on the structure of the rest of the film, interfaces areoften of direct importance on the properties of thin films. For very thin films theamount of interface material is a significant portion of the total amount of mate-rial and the often deviating properties of the interface material may influence theproperties of the entire film. For example, the changed magnetic anisotropy of theinterface atoms in magnetic systems may lead to a perpendicular magnetization insome structures [Bloemen93]. Also, for some applications such as magnetic tunneljunctions (MTJ’s) it is the interface layers themselves that actually determine thedevice properties [LeClair00b, LeClair02b].

Various types of interfaces are possible depending on the materials present, theirthermodynamic properties, and also on the kinetic aspects of the growth. Typicalinterface types include: interfaces having chemical bonds, intermixed interfaces (wherethe concentration profiles of both types of atom gradually change from 0 to 1), sharpinterfaces (where the concentration profile changes in one atomic layer from 0 to 1)and rough interfaces (where the interface is not flat on a larger scale, independent ofit being sharp or intermixed). In order to obtain sharp interfaces it is almost alwaysnecessary to use immiscible materials.

Another important classification of interfaces is that in epitaxial or non-epitaxialones. In the first case the crystal structure on both sides of the interface is connectedby a simple relation, for example when the crystal structure of the substrate continuesin the film (see also section 2.2.4). Although the epitaxy is determined by the (bottom)interface, it is actually not an interface property but more a property of the crystalphase in the substrate and the film.

Exactly which conditions lead to which type of interface is a very complex andextensive matter and a discussion falls outside the scope of this introduction. Moredetails can be found in, e.g., [Howe97, Herman04, Wuttig04].

In thin films a clear difference between the type of the bottom and the type ofthe top interface often exists since the top interface is not only influenced by thesubstrate but also by the growth of the material itself. If the growth is not layer-by-layer, the top interface will not be flat, of course. In thin film growth there usuallyare more ways to influence the bottom interface than the top interface. One oftenprefers to have flat layers, especially in growth of layer stacks such as MTJ’s, and triesto obtain either layer-by-layer growth or a nano-crystalline material instead of, forexample, three-dimensional islands. If the layers are not flat, the thickness variationsand the absence of a flat interface may have a significant influence on the propertiesof the devices or structures. An example of this is the effect of roughness in an MTJ,which causes the electron tunnelling to occur preferentially at the thinnest parts ofthe insulating barrier instead of uniformly over the entire junction. The result isthat it is hard to describe the device behaviour since that is mainly determined by


defects (i.e. the properties of the thinnest part of the barrier) and not by the regularstructure. Examples of intermixed Co interfaces are Co/Fe and Co/Ru, while Co/Cuusually gives sharp interfaces, although this may also depend on the crystallographicorientation [Camarero96].

2.2.3 Crystal Phases

General

The atoms in an ordered solid can be arranged in different ways, or crystal phases.Especially in alloys the variety of orderings is large, but also materials consisting ofone element can be ordered in several crystal structures. In thin films each structurecan occur in different orientations with respect to the film interfaces. Apart from per-fectly crystalline stackings, an even larger series of non-perfect structures is possible.Therefore, many possible atomic structures can occur in films.

The simplest crystal structures are the face-centred cubic (fcc), hexagonally closepacked (hcp) and body-centred cubic (bcc) phases (see figure 2.5), all of which areobtainable for Co. The cubic phases may be elongated or compressed in one directionand become tetragonal phases (fct, bct) [Alippi97, Marcus99]. Dislocations [Hull84]and stacking faults (SF’s) may occur in films, and on a very local scale these can inprinciple be seen as different crystal phases [Burgers34] (see section 2.3). Of course,each crystal phase can occur in different orientations relative to the film plane. Infigure 2.5 a few examples of possible orientations are given.

In bulk material several phases intermediate between fcc and hcp exist: on defor-mation of an fcc material so-called intrinsic, extrinsic, and twin stacking faults arisethrough the slipping of crystal planes. In thin film growth many more of these phasescan occur since in the (111) orientation all possible orderings of (111) planes mayoccur (section 3.2).

Phases can be stable (i.e. have a global energy minimum), meta-stable (have alocal energy minimum) or strain-induced, in which case it can only be stabilized forvery thin films by means of epitaxy with a substrate. The bcc (or bct) phase of Co isstrain-induced (see section 3.3). It is important to note that crystal phases that aremetastable in the bulk may actually be stable in the thin-film limit due to interfaceeffects.

If the same crystalline order does not continue throughout the entire film, thematerial is not single crystalline, but granular, i.e. it consists of different grains thatare themselves single crystalline but do not fit together and are separated by grainboundaries. If the crystal phases in the different grains all have the same crystalorientation perpendicular to the film plane but not parallel to it, the material iscalled textured. In this case the registry in the perpendicular direction is conservedthroughout the film as can, for example, be measured by XRD. A final possibility isthat the grains are so small that they cannot be distinguished from the boundariesany more, in this case the material is amorphous.


Co Crystal Phases

Generally magnetic transition metals are particularly rich in metastable structurespossessing different magnetic properties since the energy difference of a structuralphase change is comparable to the energy difference of a magnetic phase change.Examples are the different phases of Mn (both structurally and magnetically differ-ent) [Wuttig04] and the Invar-effect in FeNi [Schilfgaarde99]. Also for Co a largevariety of crystal phases and structures can be readily obtained in thin films underthe right conditions (except for the amorphous phase [Suslick91]). A brief discussionof this is given in the following.

At ambient pressure bulk Co has two stable phases: hcp and fcc. The hcp phase isstable below 715 K and the fcc phase above that temperature [Landolt-Bornstein86].In ultrathin films a body-centred cubic (bcc) phase can also be stabilized [Prinz85,Wieldraaijer03]. All three phases are magnetically ordered at room temperature.Calculations [Min86, Wuttig04] show that the stability of the hcp phase originates inthe ferromagnetism: if Co would be paramagnetic the fcc phase would be more stablethan the hcp phase. Due to the fully occupied majority band, the Co magnetism isvery robust against volume variations in contrast to the case of weak magnets like Feand Mn. This results in almost equal magnetic moments for all Co phases and alsoin only a limited variation in hyperfine fields.

It is not always easy to ensure that only a single phase is present in a thin film.Particularly the fcc and hcp phases are very similar, differing only in the stackingsequence of the atomic close-packed planes, which complicates the growth of only asingle phase (as shown in section 3.2). Problems like these necessitate a careful choiceof the template and the growth conditions, as well as a sensitive analysis techniquefor determining the exact structure.

The hcp phase is probably the easiest Co phase to obtain since it is the stable bulkphase around room temperature. With increasing film thickness the additionally de-posited Co tends to grow in an increasingly hcp-like structure. Especially the fcc(111)orientation easily and gradually converges towards hcp(0001) [Riedi99, Strijkers00].However, stacking faults and small amounts of fcc usually stay present, as can imme-diately be observed by the presence of satellites or broadening in the NMR spectrum(see section 2.3). The best results have been obtained for 100 nm Co films depositedon mica substrates at a growth temperature of 773 K [Gronckel94]. These films arealmost purely hcp Co(0001) and have a high structural quality. The structure for Cothicknesses below 10 nm is, however, not well-known.

An (impure) fcc phase is also remarkably easily obtained in thin films. A widevariety of substrates shows at least partially fcc growth even at room temperature,which is far below the fcc-Co-stability temperature [Riedi99]. However, a pure fccphase is much harder to obtain. For thicker films, 80-90% pure (111)-oriented fcc hasbeen obtained for 100 nm Co deposited on mica at 873 K [Gronckel94]. Virtually allother fcc(111)-Co growths suffer from the fcc to hcp transition problem: even if forvery thin films the structure is predominantly fcc, there will be increasing amountsof hcp and stacking faults with increasing thickness.

The most commonly used substrate for fcc(111) growth is Cu(111). Many studieshave shown that the Co phase grown on this substrate is always a mixture of fcc


f c c s t r u c t u r e

[ 0 0 1 ] d i r e c t i o n [ 0 0 0 1 ] d i r e c t i o n [ 0 0 1 ] d i r e c t i o n

h c p s t r u c t u r e b c c s t r u c t u r e

Figure 2.5: The face-centred cubic (fcc), hexagonally close packed (hcp) and body-centredcubic (bcc) structures. The (111) plane for fcc and the (0001) plane for hcp is drawn (greyplanes). The face-centred tetragonal (fct) structure is an fcc structure that is elongated orcompressed in the [001] direction.

and hcp and becomes increasingly hcp-like for thicker layers [Riedi99, Strijkers00].Co deposited on Au(111) has been reported to consist of either hcp [Cesari89] orfcc Co [Jomni96]. However, no studies have been performed to show that the fccphase on Au is really pure and the studies of the hcp phase do show the presenceof small amounts of fcc [Cesari89]. Recent studies of MBE-deposited Co on Au(111)have shown a random mixture of fcc, hcp and stacking fault phases [Wieldraaijerb].Therefore, the existence of pure fcc Co on Au(111) has not been proven yet. Thecommon occurrence of stacking faults in (111)-oriented Co growth is also encounteredin section 3.2.

The other fcc orientations are easier to stabilize. Several studies have shown almostpure fcc-Co on Cu(001) for various deposition methods and thicknesses [Clarke87,Suzuki92, Thomson96, Alphen96, Riedi99, Wieldraaijer05]. This is not surprising,since the fcc to hcp transformation cannot happen by merely shifting the depositedclose-packed layers. A detailed analysis is presented in section 4.2. The (110) orien-tation has been studied less often, but the possibility of stabilizing a relatively purefcc phase up to at least 15 nm has been shown for Cu(110) [Strijkers00].

The third Co structure, bcc, has been stabilized on, e.g., GaAs, Fe and Cr in boththe (001) and (110) directions [Prinz85, Wieldraaijer03]. The critical thickness isprobably about 2−3 nm, above which the structure transforms to hcp(1120) [Kim96].Thicker layers are probably not pure but stabilized by impurities, especially on GaAs.However, this can easily be found out by NMR. The phase is actually not metastablebut strain induced by epitaxy with the substrate. A short overview of various studiesis given in the section on bcc Co, section 3.3.

2.2.4 Strain

Strain, i.e. the expansion or compression of a film in one or more directions relativeto the bulk lattice parameter, is almost always present in thin films. There is always


some kind of interaction with the substrate atoms that will exert forces on atomsin the film. If no clear epitaxial relationship with the substrate is present, theseinteractions will usually lead to strains varying strongly with position and often todisorder in the film, which induces strain in the surrounding atomic structure.

If an epitaxial relationship exists between the substrate and the film structure, a(homogeneous) strain will occur throughout a thin film since there is always some lat-tice mismatch between substrate and film. For films thinner than a critical thicknessit is energetically favourable to retain the full strain. The layers are called coherentor pseudomorph since the in-plane lattice parameters are equal for both film andsubstrate. The lattice parameter of the film is fixed in both in-plane directions. Thelattice parameter in the out-of-plane direction is not fixed in this way, but will followfrom the elasticity of the film. The lowest energy state for these systems is usuallya strain in the out-of-plane direction that is opposite in sign to that in the in-planedirections. It can be described as:

∆d

d≈ γ

∆a

a(2.9)

with d and a the out-of-plane and in-plane lattice parameters of the film, respectively,and γ an elastic property of the material [Feynman63]. In general this strain rationot only depends on the material but also on the crystallographic phase and orien-tation of the film. For instance, for fcc-Co(001) γ = −1.32 (see also equation 4.1),for fcc-Co(111) γ = −0.57 and for polycrystalline Co γ = −2σ/(1 − σ) = −0.91(σ is the Poisson ratio) [Landolt-Bornstein86]. This difference will have importantconsequences when determining the (fundamental) in-plane strain from either theout-of-plane strain (for example by XRD) or the volume strain (by NMR).

The energy cost of straining the entire layer is proportional to the thickness. Abovethe critical thickness it becomes energetically favourable for the film to become inco-herent: strain is relieved by losing the coherence between the substrate and the film.This is not always easily achievable since large amounts of atoms have to be displaced.Because of this, strain may persist over relatively large film thicknesses [Weber96],and changes in the film structure may occur on strain relaxation. For example, thefct(001) phase relaxes by the slip of (111) planes in parts of the film [Hull84, Schall04].This has important implications for thin film growth, namely that a film grown epi-taxially with a certain strain is structurally different from a film grown with a largerstrain that has been partially relieved. The presence of these dislocations can easilybe observed by means of XRD because they lower the coherence length by the lossof registry within the film. The effect can also be observed by NMR (see sections 2.3and 4.2).

To summarize, strain-relieved films will always have a lower quality than fullystrained films. This makes it, for example, almost impossible to grow fcc-Co filmswith a structural quality equal to that of bulk fcc Co. Either the crystalline qualityis lower due to non-epitaxiality with the substrate, or the structure was originallystrained due to a difference in lattice parameter with the substrate.


2.3 NMR on Thin Co Films

2.3.1 Introduction

An ideal tool for the analysis of thin Co films is 59Co NMR. It is able to provide accu-rate information on a variety of aspects of the atomic scale structure of thin, buriedfilms, without the need of long-range order in the films. The very local sensitivity hasthe advantage that the structure of thin films can be directly compared with that ofbulk materials, without the thin-film nature having a strong influence per se.

In section 2.1.2 NMR was described in a general way. In this section a descriptionof the specifics of NMR on thin Co films is given on the basis of our own results. Thesection starts with some remarks on the specific features of NMR on ferromagneticallyordered layers and the main properties that can be determined from an NMR spec-trum, i.e. the properties of the interfaces, the crystal structure and the strain in thefilm. These three properties are worked out in somewhat more detail in the next part,where the common measurements and the typical parameter values for thin Co filmsare given. Also described is the possibility to obtain information from the relaxationbehaviour of the NMR spin-echo, e.g. by observing modulations on the decay causedby the presence of an unresolved splitting of the NMR line. Since NMR measurementsare usually performed at 1.5− 4 K, all values and measurements presented will applyto NMR at those temperatures. Apart from T1 (and the measured signal intensity),the NMR parameters do not vary significantly within that temperature range.

Since the description is based on our own results, most of the details are not givenin this section, but only in the ones where the relevant measurements themselves arepresented. In the current section the results are presented in a more general wayand for the details reference to the pertaining sections is made. An exception is theanalysis of the crystal phases by means of the relaxation behaviour of the spin-echo(section 2.3.4), which is based on results not included in the remainder of the thesis.

An overview of the specific differences between Co structures in the bulk and thinCo films can be found in section 4.4.

2.3.2 Co Hyperfine Fields

NMR on Co films analyses the distribution of hyperfine fields. The main structuralfeatures that influence the hyperfine field are: the presence of foreign atoms, e.g. atinterfaces or in alloys, the crystallographic structure (fcc, bcc, hcp, stacking faults,grain boundaries) and the presence of strain. These three features will be discussedone by one in the last three parts of this section (2.3.3, 2.3.4, 2.3.5). First, however,the basic values relevant for NMR on Co films will be summarized.

An idealized model of a 59Co NMR measurement is shown in figure 2.6 and atypical example of a real NMR measurement is shown in figure 2.7. Bulk Co at liquidhelium temperatures has a hyperfine field of −(20 − 23) T, where the minus signindicates that the hyperfine field is oriented antiparallel to the magnetic moment.Typically no applied field is used, since without a field the signal enhancement ishighest. The dipole field is also zero since the Co magnetization lies in the plane ofthe film (if the magnetization would be out of plane a dipole field of −1.80 T would

2.3. NMR ON THIN Co FILMS 25

59 Co S

pin-Ec

ho Inte

nsity (

arb. un

its)

F r e q u e n c y ( M H z )1 6 0 2 0 01 8 0 2 2 0

C o w i t h3 C u

n e i g h b o u r sC o w i t h2 C u

n e i g h b o u r sC o w i t h1 C u

n e i g h b o u r s

g r a i nb o u n d a r i e s

e p i t a x i a l( 1 1 1 )

i n t e r f a c eC o w i t h4 C u

n e i g h b o u r s b u l k

f c c

h c p

S F ' ss t r a i n e dv e r s i o n o fs a m e f i l m

C u ( 1 1 1 ) / 2 0 M L C o / C us i m p l i f i e d m o d e l

Figure 2.6: Idealized model of zero-applied-field NMR spectra of Cu(111)/Co/Cu at liquid-helium temperatures showing the main effects in a simplified way. The peak at the rightside corresponds to bulk Co, i.e. Co that only has other Co neighbours. Different crystalphases can be discriminated: fcc and hcp Co and some intermediate stacking-fault phases(SF’s). A short tail towards lower frequencies contains Co at grain boundaries, and at evenlower frequencies is intensity originating in interfacial Co atoms, i.e. Co atoms that have atleast one non-Co neighbour. The signal at 168 MHz comes from Co atoms having three Cuneighbours (out of twelve), which corresponds to a perfect epitaxial (111)-Co/Cu interface.At 200, 184 and 152 MHz are Co atoms having 1, 2 and 4 Cu neighbours. Intensity atthese frequencies corresponds to small amounts of interdiffusion at the interfaces of thissystem. Finally, the effect of strain on the film is shown as a dotted line: if strain causesa lattice-volume expansion the bulk peak shifts to lower frequencies. Numerous aspects ofthis system are strongly idealized and simplified, but it may serve as an overview of somestandard capabilities of NMR.

be present). Therefore the effective field at the Co nuclei is −(20 − 23) T and withγ / 2π = 10.054 MHz/T this results via equation 2.1 in bulk resonance frequenciesof about 200 − 230 MHz. In zero-applied-field measurements, the NMR intensity iscommonly plotted as a function of the resonance frequency and not of the hyperfinefield (as in figure 2.7). This is, however, not an important difference in this case sincethe two are proportional.

Since, in principle, each Co atom provides the same nuclear moment, it contributesthe same amount of signal (after correcting for a few effects, such as the enhancementand possibly the relaxation (see section 2.1.2)) and thus the area under a peak in thespectrum is directly proportional to the number of Co atoms in the correspondingenvironment(s).


59 C

o Spin

-Echo

Intensi

ty (arb

. units)

F r e q u e n c y ( M H z )1 6 0 2 0 0 2 4 0

C o w i t h 3 C u

P d ( 1 1 1 ) / 2 . 0 n m C o / C u ( a ) ( b )

1 6 0 2 0 0 2 4 0

n e i g h b o u r s

P d ( 1 1 1 ) / 2 . 0 n m C o / F e

b u l k b u l ki n t e r f a c e P d / C o i n t e r f a c e

C o / F ei n t e r -f a c e

f c c C oh c p C oa n ds t a c k i n gf a u l t s

Figure 2.7: Zero-applied-field NMR spectra of (a) Pd(111)/2.0 nm Co/Cu and (b)Pd(111)/2.0 nm Co/Fe at T = 2 K. Both spectra clearly show the bulk pure Co part asa peak at 215 MHz, which corresponds to strained fcc Co, with a shoulder at higher fre-quencies corresponding to hcp and stacking fault phases. The wings of the spectrum are theinterfaces, Co atoms which have Pd, Cu or Fe neighbours. Two types of interface can bedistinguished in (a): 1). the peak at 166 MHz corresponds to a sharp Co(111)/Cu interface,where each Co atom has 3 Cu neighbours and 2). the tail below 200 MHz corresponds todisordered / intermixed Pd/Co and Co/Cu interfaces. In (b) there is no Cu present and theCo/Fe interface can be observed on the high-frequency side of the spectrum, since Co atomswith iron neighbours have higher hyperfine fields than pure Co [Kohlhepp02].

The absence of an external field which forces a well-defined magnetization direc-tion does not have a large influence for most measurements. For hcp Co it may beimportant since the hyperfine field of hcp Co has an anisotropic part, even though thatis relatively small (∼ 2% of the isotropic part (see section 2.3.4)). Also for strainedfcc Co an anisotropic part exists, but that is even smaller (see section 2.3.5 and 4.4).

The interactions between the Co nuclear quadrupole moment (Q59Co = (41± 1) ·10−30m2 [Dembczynski92]) and the electric-field gradients (EFG’s), which are presentif the crystal structure is of lower than cubic symmetry, are often negligible in Co films.However, in principle they will split the resonance line into 2I sub-lines (with I = 7

2for Co) (equation 2.2), and in sections 2.3.4 and 4.4 we will show that the influenceof this effect can indeed be observed.

2.3.3 Measuring Interfaces

Generally each additional foreign neighbour of a certain type gives an almost equaladditional change in the hyperfine field of a Co nucleus. This constant hyperfine field


change is not exact: foreign atoms one shell away may slightly alter the shift [Jay96b]and furthermore a dependence on the distribution of the foreign neighbours aroundthe atom is expected. This last effect is similar to the case of hcp Co for whichthe hyperfine field change for a Cu atom depends on the position [Malinowska98].However, both these effects are generally small and need only be taken into accountin some special cases. Generally, the type of interface (sharp, intermixed, etc.) canbe obtained quite directly from an NMR spectrum (figure 2.7) by using the constantshift per foreign neighbour atom. The value of these shifts have been determinedempirically from studies on dilute alloys and are known for the most commonly usedelements [Gronckel93, Riedi99]. The Co hyperfine field change per neighbouring Cuatom, for example, is 1.6 T (16 MHz).

From the NMR spectrum of a thin film, the number of Co atoms with at least oneforeign neighbour can be directly determined from the relative intensities of the bulkand interface parts of the spectrum. A slightly more advanced analysis can be per-formed by fitting the spectrum with Gaussian peaks at the hyperfine fields expectedfor each environment and comparing the resulting intensities to those resulting fromsimulated impurity distributions. In this way information about the number of in-termixed layers may be obtained [LeClair01] (see section 3.3). By this method morecomplex interface types can also be analysed [Gronckel93].

If the interfaces are sharp and epitaxial a single narrow interface peak is expected,since all the Co atoms at the atoms have the same number of foreign neighbours. Fora Cu(111)/Co interface (figure 2.7), this is indeed what we find. However, there areindications that sharp (001) interfaces can not always be observed in a straightforwardmanner (see section 4.3). Furthermore there is also the possibility of a sharp (i.e. non-intermixed) interface where the two materials do not have an epitaxial relationship.In this case, the interface does not give a sharp interface line, but the total numberof Co atoms having a foreign neighbour is still 1 ML per interface. This happens forexample in the case of the MBE growth of Co on Au(111) [Wieldraaijerb]

Finally, it has to be noted that some (electronic or magnetic) effects from theinterface may in principle reach multiple monolayers within the film and can directlyinfluence the hyperfine fields. An example of this is observed in Cu(001)/Co and ispresented in section 4.3.

2.3.4 Measuring Crystal Phases

Introduction

The different Co crystal phases can easily be distinguished by NMR since they eachhave its own hyperfine field. The analysis is usually performed by studying the regularspectrum in which the spin-echo intensity is plotted versus the hyperfine field or thefrequency. There is, however, a second possibility that may be especially useful if theintensity spectrum does not show clear features. We will report some results hereon the dynamic behaviour, which clearly discriminate the crystal structure and thestructural quality. So far, these effects have not received the attention they deservein the literature and further on in this section we will illustrate them by our own (yetunpublished) results.


59 Co S

pin-Ec

ho Inte

nsity (

arb. un

its)

H y p e r f i n e F i e l d ( T )- 2 1 . 5

B a p p l i e d = 2 . 0 T/ / c - a x i s

( a ) ( b )

59 Co S

pin-Ec

ho Inte

nsity (

arb. un

its)

D e l a y T i m e , 2 t ( m s )

B a p p l i e d = 2 . 0 T ^ c - a x i s

B a p p l i e d = 0

- 2 2 . 0 - 2 2 . 5 0 1 0 2 0 3 0

B a p p l i e d = 0

B a p p l i e d = 2 . 0 T / / c - a x i sB a p p l i e d = 0 . 3 5 T / / c - a x i s

F W H M = 0 . 2 T

F W H M = 0 . 1 T

Figure 2.8: (a) Spectra of hcp Co in different applied fields (T = 2K), showing theanisotropy of the hyperfine field and the anisotropy of the line width that is caused bythe quadrupole splitting due to the electric-field gradients inherent to the hcp structure.Because of the applied fields, the hyperfine field is plotted on the horizontal axis instead ofthe resonance frequency, which depends on the applied field. The small peak at −21.6 Tfor the zero-applied-field measurement corresponds to a small amount of material in the fccphase. The shoulders at −22.15 T in the ~B‖c-measurement and at −22.35 T in the ~B ⊥ c-measurement correspond to a stacking fault phase, that clearly also shows some anisotropy.(b) Measurements of the spin-echo decay behaviour as a function of the delay time. Themodulations on the normal exponential decay behaviour show the presence of a splitting ofthe NMR line that is too small to be resolved directly in the spectrum. Also shown are fits ofthe signal that provide the modulation frequency, which is a direct measure of this splitting,and from the general T2 can also be derived. The 0.35 T measurement clearly shows thathigher harmonic modulations may also be present.

Hyperfine Fields

Fcc Co has a hyperfine field of −21.60 T (217.2 MHz), bcc Co a hyperfine field of−19.7 T (198 MHz) and hcp Co has an anisotropic hyperfine field, which is −21.88 T(220 MHz) parallel to the c-axis and −22.68 T (228 MHz) perpendicular to the c-axis [Riedi99]. In zero-applied field the magnetization inside hcp Co does not have asingle orientation, but is ordered in magnetic domains. This may have a significanteffect on the measurements since most of the NMR signal originates in Co atomslocated inside the domain walls. This leads to the observation of a broadened hcpline around −22.38 T (225 MHz) (see figure 2.8(a)). In between the fcc and hcpfields resonances are present corresponding to Co atoms ordered in crystal phases


in between the fcc and hcp stacking. These result if the stacking of (111) planes isneither consistently in the fcc nor in the hcp sequence (see section 3.2).

Another structure that can be observed is the boundary between two physicalgrains. Due to the larger distance between them, the Co atoms at the boundary havea lower hyperfine field and usually resonate about 20 MHz below the bulk peak (seefigure 2.6). Although the grain-boundary signal is hard to separate from that of aninterdiffused interface, it can in principle be used to estimate the grain size in thesample.

Resonance-Line Widths

The typical line width of bulk fcc Co is about 0.7 MHz (FWHM). However, the typicalline width for thin films is much higher: typically 5− 10 MHz. These line widths areso common that it has even been suggested that it is probably directly due to thepseudo-two-dimensional structure of the films [Riedi99]. However, under the rightconditions thin fcc-Co films can be grown that have line widths of not much morethan 1 MHz. Furthermore the additional width with respect to the 0.7 MHz bulklines can be fully explained by the presence of a small tetragonal deformation of thefcc phase (see chapter 4, specifically section 4.4.7). In most other fcc films the largebroadening is probably caused by a distribution of strains caused by the presence ofstacking faults and dislocations.

Anisotropy and Quadrupole Splitting

Some extra remarks have to be made on the analysis of hcp Co by NMR. As remarkedbefore, the hyperfine field is anisotropic, leading to a somewhat more complex lineshape than the common single-Gaussian profile. In figure 2.8(a) NMR spectra areplotted of a 100 nm hcp-Co(0001) film grown on mica at the optimum tempera-ture [Gronckel94] in either zero applied field, in a large field applied in the plane ofthe film (i.e. perpendicular to the c-axis) or in a large field applied perpendicular tothe plane of the film. The anisotropy of the hyperfine field is clearly observable. Notethat since the hyperfine field is plotted, the direct effect of the applied field and theeffect of the demagnetizing of shape field is removed.

Apart from this anisotropy, the hcp line width can also be observed to be anisotropic:for B ⊥ c it is significantly larger than for B‖c (0.2 T vs. 0.1 T). This is caused by thepresence of electric-field gradients in the hcp structure: due to the interaction betweenthe Co nuclear quadrupole moment and these EFG’s the Co resonance line is splitinto seven equidistant lines. The splitting is too small to be resolved and thus only abroadening of the line is observed. The splitting can, however, be determined by mea-suring the spin-spin relaxation behaviour. In the case of an equidistant line splittingthis decay will no longer be exponential as usual, but will show a clear modulationsuperimposed on the exponential decay [Hahn52, Abe66, Fekete78], see figure 2.8(b).The modulation frequency is proportional to the line splitting, indicating indeed alarger splitting for the broader line. This is described in more detail in section 4.4.A similar effect may be observed for other films in which the symmetry is lower thancubic.


59 Co S

pin-Ec

ho Inte

nsity

(arb. un

its)

F r e q u e n c y ( M H z ) F r e q u e n c y ( M H z )1 6 0 1 8 0 2 0 0 2 2 0 2 0 0 2 1 0 2 2 0 2 3 0

Spin-L

attice

Relax.

Time

, T 1 (m

s)Spi

n-Spin

Relax

. Time

, T 2 (ms

) 2 0 01 6 0

1 2 0

8 0

0

1 5

1 0

5

0

2 4 0

6 0 05 0 04 0 03 0 02 0 0

01 0 0

4 0

3 0

2 0

0

1 0

( a ) ( b )

( c )

( f )( e )

( d )

M B E - g r o w n

s p u t t e r e d

M B E - g r o w n

M B E - g r o w n

s p u t t e r e d

s p u t t e r e d

b e f o r e a n n e a l

a f t e r a n n e a l





Figure 2.9: Zero-applied-field NMR measurements (T = 2K) showing (a) Au(111)/3.0 −3.5 nm Co/Au grown by MBE or by sputter deposition and (b) SiOx/30.0 nm Co before andafter annealing. In (a) the longer-range structural quality of the Co is higher for the MBE-grown sample and in (b) it is much higher for the annealed sample. This can be observed bythe narrower lines (leading to better separability between the fcc and hcp / stacking faultpeaks) and by the lower amount of grain-boundary Co (seen in the range of 200−220 MHz).Below the spin-lattice (T1) and spin-spin (T2) relaxation times are plotted as a function offrequency for the Au/Co structure (c,e) and for the SiOx/Co structure (d,f). The lines areguides to the eye. The higher-quality Co can be clearly recognized by a higher T1 and bythe presence of more features in both T1 and (more pronounced) T2 [Wieldraaijera].


Structural Analysis on the Basis of the Relaxation Behaviour

We have found that also for systems without quadrupole splitting, the relaxationbehaviour (both spin-spin and spin-lattice) can provide information on the crystalphases and particularly on the crystalline quality. Theoretical descriptions of thisbehaviour as a function of structure are complicated, especially for the zero-applied-field case, where spin-spin relaxation mainly results from domain-wall motion. Inpractice, however, there are a few observations which may be quite useful for providingqualitative information on the structural quality.

The simplest measures of longer-range quality (e.g. the presence and the size ofthe grains, etc.) come directly from the NMR spectrum as can be seen in figure 2.9(a)and (b). In this figure the spectrum of Co on Au(111), both sputtered and MBE-grown (a), and the spectrum of a Co layer on SiOx before and after anneal (b) areplotted. In both cases, the better quality film (MBE-grown in (a) and after annealin (b)) can be recognized by a smaller width of the constituent peaks, leading to abetter separability, and by a smaller amount of grain boundary signal. This meansthat the grains are larger and that the structure within a single grain becomes morehomogeneous, thus resulting in separate grains with a particular crystal structureeach.

When viewing the spin-lattice relaxation time T1 ((c) and (d)) and the spin-spinrelaxation time T2 ((e) and (f)) as a function of frequency, a clear difference is alsoobserved. In both cases T1 is about a factor of 2 smaller in the low-quality casethan in the high-quality case. This behaviour appears to be quite general and whencomparing similar systems, the one with the highest structural quality seems to havethe highest T1. Furthermore, it is remarkable that the low-quality T1 spectra donot show any frequency dependent features, in contrast to the spectra of the higherquality samples.

This effect is even more pronounced in the T2 spectra. Here the presence of featuresis the strongest sign of a high quality material with crystal phases that are not mixedbut reside at separate physical positions (figure 2.9(d,e)). For the Au(111)/Co sample(e) this can be mainly seen by the steep rise of T2 towards the grain boundary andinterface region of the spectrum. The high T2 for the high-quality sample can beexplained when one realizes that the signal in this region originates in a very smallnumber of dislocated nuclei. These nuclei will thus generally be at a large distancefrom each other. Thus, the probability of interaction between two similar nuclei issmall and T2 is high. The same effect is seen for the SiOx/Co sample (f) togetherwith a second telltale effect, namely the appearance of features. Empirically thebest-defined structural phases, i.e. those having only equivalent atoms (which canhave spin-spin interactions easily) have indeed the lowest T2. The better and moreseparated these structures are, the better resolved the features get.

This last effect is shown even more clearly in figure 2.10, where the correlationbetween the NMR spectrum and the T2 spectrum for a 50 nm thick Co film on Cu(001)is shown. The film consists of a very high-quality strained-fcc Co with some minoramounts of stacking faults. The figure shows a perfect one-to-one correlation betweenthe peaks (or shoulders) of the spectrum and the minima of T2. It is even so thatsome of the shoulders and resonances are better recognizable from the minima in the


59 Co S

pin-Ec

ho Inte

nsity

(arb. un

its)

F r e q u e n c y ( M H z )2 1 0

( a )

( b )

Spin-S

pin Re

laxatio

n Time

, T 2 ( ms

)

0

5 0

1 0 0

1 5 0

2 1 5 2 2 0 2 2 5 2 3 0

1 0 x1 0 x

Figure 2.10: (a) Zero-applied-field NMR spectrum (T = 2 K) of 50 nm of Co MBE-grownon a Cu(001) single crystal. The material consists of high-quality strained fcc Co with somestacking faults structures. (b) The spin-spin relaxation time T2 as a function of frequencyshowing the exact correlation between the peaks and shoulders in the spectrum and theminima in the T2 spectrum [Wieldraaijera].

T2 spectrum than from the NMR spectrum itself. Although more study would beneeded to use this information in a quantitative way, it is a useful technique to keepin mind when analysing structures. In several cases the relaxation behaviour is avery useful indication for structural differences or changes on a longer-range that canhardly be observed directly. As stated before, we will not use this technique in theremainder of this thesis.

2.3.5 Measuring Strain

The hyperfine field of a ferromagnetic material is also influenced by strain in thematerial. If the structure is compressed, the hyperfine field will increase and if the


structure is expanded it will decrease. This influence has been obtained from thepressure dependence of the hyperfine field (see section 4.2.4 page 73). Combining thiswith the elastic behaviour of the film, an empirical dependence of the hyperfine fieldon the volume change may be derived:

∆f

f=

∆Bhf

Bhf= −(1.13± 0.01)

∆V

V. (2.10)

Strictly speaking, this equation is only valid for isotropically strained structures.However, it is also commonly used to describe strain in thin films [Riedi99, Alphen95]which is generally strongly anisotropic (see section 2.2). The simplest way to describethe influence of this type of strain more fully, would be to separate the influence ofisotropic (volume) strain and that of deformational (volume-conserving) strain. Theeffect of the first might than still be described by equation (2.10), apart from the factthat it influences Bhf,iso, while the second might be expected to only cause a hyperfinefield anisotropy (by ‘de-quenching’ of the orbital moment). This description is usedin sections 4.2 (page 73) and 4.4.3. The expected anisotropy is indeed observed, butis found to be small and not suited for quantitative analysis. However, it remainsimportant to realize that anisotropic strain will in general lead to a (change of the)hyperfine field anisotropy.

A second effect caused by anisotropic strain is the occurrence of electric-fieldgradients that lead to a splitting of the resonance line (equation (2.2)). The firstobservation of this splitting is presented in section 4.4.

Finally, it has to be noted that the difference between a Co(001) film grownwith a certain strain and another Co(001) film, which has acquired the same strainthrough strain relief after growth (see section 2.2) can be observed by NMR. The(111) dislocation planes leading to strain relief can be observed directly as stackingfaults and the variations in strain caused by the dislocations lead to a broadening ofthe line. This effect may be at least part of the origin of the strongly enlarged linewidths common in thin films (see sections 2.3.4 and 4.4).

Chapter 3

Co Structures forMagnetoresistive Devices

Abstract: In this chapter we will report on two structural studies of thin Co filmsgrown for magnetoresistive devices, in this case for use in magnetic tunnel junctions(MTJ’s).

In the first study, ‘FCC vs. HCP Co in sputtered MTJ’s’, the physical structureof the magnetic electrodes of Co/Al2O3/Co magnetic tunnel junctions (MTJ’s) hasbeen investigated by means of 59Co nuclear magnetic resonance (NMR) and X-raydiffraction (XRD). Junctions sputtered on FeMn have an essentially fcc (111)-orientedbottom electrode, while all top electrodes and the bottom electrodes of junctionssputtered on Ta are poly-phase and poly-crystalline. The specifics of the electronicstructure of the electrode crystal phases can explain the extra features and asymmetryin the conductance versus bias behaviour observed in FeMn based junctions.

In the second study, ‘BCC Co(001) as a bottom electrode’, the applicability of thethe strain-induced bcc phase of Co in magnetoresistive devices has been studied. Tothis end, ultrathin bcc-Co(001) films and the influence of the additional layers neededfor magnetoresistive devices have been examined by means of 59Co nuclear magneticresonance (NMR). NMR is shown to be a discriminating technique for determiningthe presence of structurally and magnetically pure bcc Co. The maximum stability foruncovered and Fe-covered layers grown on Fe(001)/GaAs(001) and Fe(001)/Ge(001)is found to be about 2 nm, consistent with earlier results. Growth of an Al2O3 toplayer preserves the bcc phase, in contrast to a Cu film which causes a transformation ofthe bcc structure to fcc or hcp phases. The bcc-preserving effects of Al2O3 imply thepossibility to fabricate magnetic tunnel junctions with bcc Co(001) bottom electrodes.Although bcc Co is a force-induced structure, thin layers are shown to be stable over afew years when Al2O3 has been grown on top. Junction structures using bcc Co(001)bottom electrodes were grown and characterized.

Most of the first study has been published as ‘Influence of electrode structure onmagnetotransport in magnetic tunnel junctions’ in IEEE Transactions on Magnetics38, 2727 (2002) and have been used by P. LeClair et al. in ‘Band structure and

35

36 CHAPTER 3. Co FOR MR DEVICES

density of states effects in Co-based magnetic tunnel junctions’, Phys. Rev. Lett. 88,107201 (2002). Most of the second study has been published as ‘Growth of Epitaxialbcc-Co(001) Electrodes for Magnetoresistive Devices’ in Phys. Rev. B 67, 224430(2003).

Contents

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.2 FCC vs. HCP Co in Sputtered MTJ’s . . . . . . . . . . . 37

3.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.2.2 Experimental Procedure . . . . . . . . . . . . . . . . . . . 37

3.2.3 Results: Structure . . . . . . . . . . . . . . . . . . . . . . 39

3.2.4 Results: Tunnelling . . . . . . . . . . . . . . . . . . . . . . 43

3.2.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

3.2.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.3 BCC Co(001) as a Bottom Electrode . . . . . . . . . . . 46

3.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 46

3.3.2 Reported Results on the Stability and Analysis of BCC Co 46


3.3.4 Results: Structure . . . . . . . . . . . . . . . . . . . . . . 49

3.3.5 Results: Influence of Additional Layers . . . . . . . . . . . 54

3.3.6 Results: Tunnelling . . . . . . . . . . . . . . . . . . . . . . 59

3.3.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.3.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 62

3.1 Introduction

Tunnelling processes in magnetic tunnel junctions (MTJ’s) [Moodera95] are largelydetermined by the electronic structure of the active layers. A theoretical descriptionof spin-polarized tunnelling, based on the electronic structure, is quite feasible foridealized structures and fully epitaxial layers with sharp interfaces [Appelbaum69,Zhang99, Oleinik00]. A complete description for realistic junction structures is how-ever extremely difficult. The largest problem is the lack of crystalline order of theAl2O3 barrier commonly used in MTJ’s, which is very hard to model theoretically. Inaddition, the roughness and slight intermixing at the interfaces, and the non-epitaxyof, at least, the upper electrode which is grown on the Al2O3 barrier also pose largeproblems.

A way to circumvent these problems is to vary part of the physical and electronicstructure of the device and investigate only the corresponding changes in behaviour.

3.2. FCC VS. HCP Co IN SPUTTERED MTJ’S 37

However, since the top electrode is deposited on the amorphous barrier, it is difficultto change its structure in a defined way. An easier method is to grow well-defined,preferably single-crystalline, bottom electrodes. This is relatively straightforwardto do, since the substrate on which the bottom electrode is deposited may be variedalmost at will. The barrier and the top electrode do not have an epitaxial relationshipwith the bottom electrode and their structure will be virtually unaffected. By varyingthe structure of the bottom electrode in a controlled way, the effect of specific physicaland electronic structures on the tunnel properties can be investigated. In this way, anunderstanding of the properties influencing tunnel junctions can be obtained withoutthe need for a full theoretical description.

This procedure has been used by Yuasa et al., who have established a dependenceof the tunnelling spin polarization on the crystallographic orientation for epitaxialbcc-Fe bottom-electrodes [Yuasa00] and demonstrated resonant tunnelling in MTJ’swith epitaxial Cu-dusted fcc-Co bottom-electrodes [Yuasa02]. An advantage of usingCo electrodes for this rather than Fe would be provided by the various phases in whichCo can be stabilized in thin films.

This chapter is organized as follows. In the first section a study of the physi-cal structure of the magnetic electrodes of Co/Al2O3/Co magnetic tunnel junctions(MTJ’s) for two different substrates is presented, which has been shown to explainthe difference in tunnelling properties observed [LeClair02a]. In the second sectionthe applicability of the the strain-induced bcc phase of Co in magnetoresistive deviceshas been studied by analysing its stability under coverage with different layers.

3.2 FCC vs. HCP Co in Sputtered MTJ’s

3.2.1 Introduction

The typical layer structure of standard Co / Al2O3 / Co MTJ’s is depicted in fig-ure 3.1(a). Several buffer layers are used below the bottom electrode, the most impor-tant of which is the anti-ferromagnetic FeMn layer, which exchange-biases the bottomelectrode. The Ta and bottommost Co layer are used to induce the correct structureand orientation in the FeMn layer. The top Al layer is used to prevent oxidation ofthe top electrode. A different type of bottom electrode may be obtained by leavingout the bottommost Co and FeMn layer and growing the bottom electrode directlyon the Ta layer. These two MTJ’s have different bottom electrodes and probablysimilar top electrodes. In total there are three Co layers for which the structure is ofimportance.

3.2.2 Experimental Procedure

Ferromagnetic tunnel junctions were prepared by ultra-high vacuum dc/rf magnetronsputtering, with a base pressure of better than 5 ·10−10 mbar. The details of this fab-rication process can be found elsewhere [LeClair00b]. The MTJ’s consist of: Si(001) /SiO2 / buffer / Co dCo / Al2O3 / Co 15 nm / Al 3 nm, post-annealed in a magneticfield at 200C. The buffer consists either of Ta 5 nm / Co 7 nm / FeMn 10 nm,the so-called ”FeMn-based buffer”, or only of Ta 5 nm, the so-called ”Ta buffer”. In


buffer

( a )

A l

( b ) ( c ) ( d )

f u l l t u n n e l j u n c t i o nC o a n a l y s i s s t r u c t u r e st y p e 1 t y p e 2 t y p e 3

susbst

ratejun

ction

C oA l 2 O 3C o

C oF e M n

T aS i

A l

A l 2 O 3C o

C uF e M n

T aS i

A l

A l 2 O 3C oT aS i

A lC oA l 2 O 3C u

C uF e M n

T aS i

Figure 3.1: (a) Typical Co / Al2O3 / Co-MTJ layer structure. The narrower layers arepatterned in order to obtain the crossing electrodes forming the junction. (b-d) The modelstructures used to investigate the structural properties of the various Co layers in an MTJ.Only one Co layer is present per structure since NMR cannot distinguish between differentCo layers within one structure. Type 1 (b) corresponds to the bottom electrode of a regularjunction (”FeMn-based”). Type 2 (c) is the bottom electrode of a junction without the FeMn(”Ta-based”). Type 3 (d) corresponds to the top electrode of any of these junctions. Thick-nesses in the model structures are equal to those in the MTJ, except for the Co thickness,which is varied.

order to determine the structure of the bottom Co layer for both junction types andof the top Co layers, three types of separate macroscopic structures having only oneCo layer, were grown on the same buffers as the comparable junction layers. Sampletype 1 and 2 correspond to the bottom part of an FeMn-based and a Ta-based MTJrespectively, up to and including the Al2O3. In the FeMn-based type 1 sample, Cu isused as a seed layer for the FeMn instead of Co, but this has no effect on the resultinggrowth of the Co layer. Sample type 3 consists of a full FeMn-based MTJ where theCo seed layer and bottom electrode have been replaced by Cu (see figure 3.1).

The physical structure is determined by 59Co nuclear magnetic resonance (NMR)in combination with X-ray diffraction (XRD). The NMR experiments have been per-formed at 1.5 K in zero applied field, as described in sections 2.1.2 and 2.3. NMRmeasurements provide a way of determining the relative amounts of Co with specifichyperfine fields, which in turn gives a distribution of the Co atoms over differentstructural environments (e.g. fcc, hcp, bcc). Among the main advantages of NMRare the possibility to directly observe buried layers, the fact that no long-range orderin the layers is necessary, and that the sensitivity is sufficient for measuring singlelayers of only a few nm thick.

The differential junction conductances (dI/dV (V)) were measured using standarddc or ac lock-in techniques. For V > 0 the bottom electrode was biased positively.


Inte

nsity (

arb. un

its)

2 Q ( d e g )3 5

h c p ( 1 0 1 0 )

4 0 4 5 5 0 5 5 4 3 4 4 4 52 Q ( d e g )

( a ) ( b )1 3

1 5 . 0 n m

8 . 0 n m

5 . 0 n mh c p ( 1 0 1 1 ) f c c ( 0 0 2 )

f c c ( 1 1 1 ) / h c p ( 0 0 0 2 )F e M n( 1 1 1 )

2

Figure 3.2: X-ray-diffraction spectrum for type-1 samples, (a) for a Co thickness of 15 nmand (b) the peaks for three different Co thicknesses: 5, 8 and 15 nm. The angles of theCo diffraction peaks that lie in this region are marked. The positions 1, 2 and 3 in (b)correspond to, respectively, fcc(111) Co, hcp(0002) Co and fcc(111) Co that is strained togrow epitaxially on the FeMn(111) substrate. The measurements in (b) have been off-setvertically for clarity.

3.2.3 Results: Structure

XRD measurements of a type 1 sample with a Co thickness of 15 nm are shownin figure 3.2(a). The angles of the Co diffraction peaks that lie in this region aremarked. As shown in figure 3.2(a), 15 nm of Co in a type 1 sample shows only (111)-oriented fcc Co or (0002)-oriented hcp Co at a diffraction angle 2θ = 44.3. Theobserved diffraction peak appears to be closer to the fcc than to the hcp position.However, no conclusive assignment can be made from these measurement, since strainor stacking faults can cause small peak shifts. Moreover, the slight asymmetry in thepeak may indicate the presence of more than one Co phase. In general the fcc(111) andhcp(0002) peaks are too close together for separation in a standard XRD experiment,due to the structural similarity of the phases, both consisting of close-packed Coplanes parallel to the sample. The only other observable peak is found at 43.3 andcorresponds to the FeMn(111) buffer layer on which Co is almost lattice- matched.

Thinner Co layers show a similar XRD spectrum (see figure 3.2(b)) in which theCo peak starts as a shoulder on the FeMn peak. The peaks seem to show slighttails towards higher angles, possibly indicating the presence of more than one phase.Angles corresponding to fcc(111) and hcp(0002) are again indicated, together with‘angle 3’ corresponding to fcc(111) Co that is expanded 2.3% in-plane for lattice-matching with the FeMn. This position corresponds to an out-of-plane contractionof 1.3% (following from standard elastic theory as given in more detail in chapter 2),


59 Co S

pin-Ec

ho Inte

nsity (

arb. un

its)

F r e q u e n c y ( M H z )2 1 0 2 2 0 2 1 0 2 2 0 2 3 0

( a ) t y p e 1 ( F e M n b a s e d ) ( b ) t y p e 2 a n d 3

3 . 5 n m

5 . 0 n m

8 . 0 n m

1 5 . 0 n m

T a b a s e d5 . 0 n m

A l 2 O 3 b a s e d8 . 0 n m

f c c

S F / h c p

f c cS F / h c p

2 3 0

Figure 3.3: 59Co NMR spectra at 1.5 K in zero-applied field (a) for type-1, FeMn-based,structures for different Co thicknesses and (b) for type-2, Ta-based, and type-3, Al2O3-based,structures.

but is clearly absent from the spectrum, indicating that strain is largely relieved inthe films.

Samples of type 2 and 3, with thicknesses of 5 or 8 nm, do not show any Co peak inthe XRD spectrum, indicating poly- and/or nano-crystallinity for which the intensityis too much spread out to be observable with our equipment at these thicknesses.Thus, the samples of type 1 (FeMn-based) are always oriented with the close-packedplanes parallel to the sample surface, while both other types are polycrystalline.

Figure 3.3 shows the 59Co NMR intensity as a function of frequency, after cor-rection for ferromagnetic-enhancement effects and the frequency dependence of thesignal (see sections 2.1.2 and 2.3), for the three types of NMR samples at differentCo thicknesses. Co in the fcc phase has a resonance frequency of about 217 MHz,which can be shifted over a few MHz because of strain. Resonance frequencies be-tween 220 and 228 MHz correspond to Co in the hcp phase or in phases formed byfaulty stacking in between the fcc and the hcp phase. The difference between type-1samples, grown on FeMn, and type-2 and -3 can be seen immediately. The Co layerin type-1 samples starts as a mixture of fcc, hcp and stacking-fault phases, but thematerial grown above a thickness of 3.5 nm is almost purely in the fcc phase. TheCo layer in type-2 and -3 samples consists of an almost equal mixture of phases forany thickness. It is unstrained, in contrast to the Co in type-1 samples, which hasa volume strain of +1% for the thinnest samples decreasing to +0.5% for the 15 nmthick layer.


A +0.5% volume strain in the film, corresponds to +0.3% strain in the plane ofthe film and −0.2% perpendicular to the growth direction. This would result in adiffraction angle of 2θ = 44.3o in the XRD. For this size of strain, no significantdeviation from the unstrained fcc/hcp angles is expected, in correspondence with themeasurement.

For this type of (111)-oriented growth each time a close-packed Co(111) plane isgrown on top of the previously grown one. For each new ML, two positions withinthat plane are possible relative to that of the previous ML. In total there are threeunique in-plane positions. If we define the in-plane position of the first ML as A, thanthe ML on top can be either at position B or at position C. If the second ML is at Bthe third can be at either A or C, and so on. Many different stacking sequences arepossible in this way. The fcc phase consists of ML’s stacked ABCABCA, etc. andthe hcp phase of ABABABA. Several intermediary phases, also having intermediatehyperfine fields, are possible [Toth63].

A model by Michel et al. [Michel01] provides a way of comparing the developmentof the ratio between fcc, hcp and stacking-fault phases in this kind of system. Ac-cording to this model, within a certain grain, each new ML has a constant chance αto end up in the fcc position (i.e. ABC) and thus (1−α) for the hcp position (ABA).Thus, α = 0 gives pure fcc and α = 1 gives pure hcp. The model then assumesthat the hyperfine field of an atom is determined by the number of fcc stackings inthree ML’s. For three ML’s there are 23 stacking possibilities (relative to the pre-vious 2 ML that can be defined as AB). Three fcc stackings (ABCAB), gives purefcc (217 MHz), three hcp stackings (ABABA) pure hcp (225 MHz for the isotropicpart). Two fcc stackings and one hcp stacking (for which there are three possibilities:ABCAC, ABCBA, ABACB) gives one type of SF (≈ 220 MHz) and one fcc and twohcp another type of SF (≈ 222 MHz). Thus, for a grain with α = 0.1, 73% of theNMR intensity will be in the fcc line ((1−α)3), 24% in SF1 (3α(1−α)2), 3% in SF2(3α2(1− α)) and 0.1% in the hcp-line (α3).

Assuming the presence of two types of grains in the film, each with their own valueof α, experimental results can usually be well described within the model. First, the59Co NMR spectrum is fitted with the four lines mentioned in the previous paragraphand the relative intensities are inserted in the model. The structure can then bedescribed by three parameters: α1 and α2 (i.e. the α’s of each of the grain types)and the ratio of material in both grain types, commonly denoted as the fraction of‘fcc-like grains’, xfcc (i.e. those with smallest α). Some critical remarks on the modelare given in section 3.2.5, a more complete description can be found in [Michel01].

For samples of type 2 and 3, fits give a fraction of fcc-like grains, xfcc ≈ 50%,independent of the thickness (see figure 3.4(a)). These fcc-type grains have α = 0.16,the other grains, which are more hcp like, α = 0.7 (figure 3.4(b)). These resultsare comparable to those for a 100 nm Co film grown on a glass substrate [Gronckel],which is a model specimen of a random poly-phase, polycrystalline growth, which forsimplicity we call poly-Co. For the samples of type 1 the fraction of fcc grains as afunction of thickness, xfcc(d), is also given in figure 3.4(a). For the lowest thicknessxfcc is still relatively close to that for random growth, but for higher thicknesses upto 8 nm it increases to 76%. At the same time α for the fcc-type grains decreasesmarkedly, denoting a better quality of the fcc-stacking quality within the grains. The


FCC-l

ike Gr

ains (%

)

C o T h i c k n e s s ( n m )

1 0 08 06 04 02 00

0 . 80 . 60 . 40 . 20 . 0Mi

sstack

Fractio

n, a,

for ea

ch Gra

in Type

P o s i t i o n w i t h i n C o L a y e r ( n m r )0 4 8 1 2 0 4 8 1 2 1 6

t y p e 1 o n l y( a )

( b )

( c )

( d )

t y p e 1

t y p e 2 / 3

h c p - l i k e g r a i n s

f c c - l i k e g r a i n s

h c p - l i k e g r a i n s

f c c - l i k e g r a i n s

h c p - l i k e

f c c - l i k e

Figure 3.4: Results from the model for the Co structure. (a) percentage of fcc-like grains,xfcc(d), in films of type 1 (circles) and type 2 and 3 structures (squares), (b) the fractionsof non-fcc stacking α1 and α2 for both grain types and both structures (same symbols as(a)), (c) percentage of fcc-like grains for the extra grown material compared to the previousthickness, xfcc(∆d), for type 1 structures only, (d) α1 and α2 for both grain types in theextra grown material, where mainly α for the fcc-type grains is important, since that is themain part of the material. A lower α corresponds to a better fcc quality. α for the hcp-likegrains between 3.5 and 5 nm cannot be determined well, since the amounts of this phase arealmost negligible.

quality of the hcp-type grains stays almost constant at α ≈ 0.6. This change frompolyphase growth to almost purely fcc-growth is even more clear if the fraction offcc grains of the extra material, xfcc(∆d) is plotted when going from one thickness tothe next (see figure 3.4(c,d)). The material grown between 3.5 nm and 8 nm consistsfor more than 90% of fcc grains, provided of course that no re-ordering of previouslygrown material occurs.

Thus, the growth of Co on FeMn starts as an almost equal mixture of fcc-likeand hcp-like grains, with both relatively large amounts of stacking faults. The extramaterial grown from directly above 3.5 nm up to 8 nm consists for more than 90% offcc-grains of higher quality. So even the 3.5 nm thick layer will have predominantlyfcc-Co at the topmost interface.


Norma

lized (d

I/dV) //

B i a s V o l t a g e ( V )- 0 . 8 - 0 . 4 0 . 0 0 . 4 0 . 8

1 . 0

1 . 2

1 . 4

1 . 6

T a - b a s e dM T J

F e M n -b a s e dM T J

Figure 3.5: Conductance-voltage (dI/dV (V )) characteristics, normalized at V = 0, forparallel Co magnetizations at 5 K for an fcc(111) Co 5 nm/Al2O3/poly-Co junction (”FeMn-based MTJ”) and for a poly-Co / Al2O3 / poly-Co junction (”Ta-based MTJ”).

3.2.4 Results: Tunnelling

Since tunnelling effects are extremely interface sensitive [LeClair00b], it will be thestructure of the topmost bottom-electrode layers that will determine the conductancebehaviour. In an FeMn-based junction, with a bottom electrode of type 1 and a topelectrode of type 3, tunnelling will occur between the almost 100% fcc(111)-orientedCo of the bottom electrode and the poly-Co of the top electrode. In a Ta basedjunction on the other hand, with a bottom electrode of type 2 and a top electrode oftype 3, tunnelling will take place between two poly-Co electrodes. For the Ta basedjunction no direction-dependent specific band structure effects are expected in thetunnelling behaviour since these will all be averaged out by the presence of differentphases and orientations on both sides of the barrier. For the FeMn-based junction,density-of-states effects may be expected since one of the electrodes consists for almost100% of Co in the fcc(111) phase. The asymmetry of this junction is expected to leadto an asymmetry in the transport properties with respect to applied bias voltage.

Figure 3.5 shows the conductance-voltage (dI/dV (V ) ≡ G(V )) characteristics,normalized at V = 0, for parallel Co magnetizations at 5 K for an fcc(111) Co5 nm / Al2O3 / poly-Co junction and for a poly-Co / Al2O3 / poly-Co junction.Both junction types have a tunnel magnetoresistance of 30 to 35%. As expectedthe conductance-versus-bias curve is symmetric for the poly-Co / Al2O3 / poly-Cojunctions and the shape of the curve can be well understood by a combination of aparabolic contribution from regular elastic tunnelling [Wolf85] and a linear contribu-tion from magnon-assisted tunnelling at low bias [Zhang97]. The curve for the fcc(111)


Co / Al2O3 / poly-Co junctions however, shows a distinct asymmetry and extra fea-tures in the form of a shoulder for positive bias and a dip for negative bias. Sincethe only difference in the junctions is the physical structure of the bottom electrode,the changes and asymmetry have to be caused by features in the band structure anddensity of states of fcc(111) Co. Theoretical calculations by Davis et al. [LeClair02a]for ballistic tunnelling based on the full band structures of fcc and hcp Co [Davis01]are able to reproduce qualitatively and in part quantitatively the effects seen in theconduction properties of these junctions [LeClair02a]. As expected from the amountof fcc-like Co at the barrier interface (figure 3.4(c)), FeMn-based junctions with a Cothickness different than 5 nm generally show a smaller asymmetry. Thereby makingpossible a clear assignment of the features observed in the conductance versus biascurve for the fcc(111) Co / Al2O3 / poly-Co junctions to the electronic structure ofthe fcc(111) Co.

A similar asymmetry is observed for a tunnel junction with a bcc(001) bottomelectrode and a polycrystalline top electrode, only in that case the asymmetry ismuch smaller, corresponding to the absence of strong spin-dependent features in theelectron bands relevant for the tunnelling process (see section 3.3.6).

3.2.5 Discussion

Several remarks concerning the structural model used in section 3.2.4 and the ob-served Co structure are in order. The model uses four different Co resonance lines(fcc, hcp and 2 stacking fault lines). Three different stacking-fault resonances havebeen predicted from theoretical considerations and have also been observed [Toth63].However, the observation of three stacking-fault lines is quite uncommon, in almostall cases only two lines are observed [Riedi99]. Especially the results by De Gronckelet al. [Gronckel94] cast some doubt on the three line hypothesis and makes the exis-tence of only two SF frequencies much more probable. Secondly, the model assumesa single resonance frequency for hcp Co. However, the hyperfine field of hcp Co isanisotropic and this might result in a composite resonance line for the case of domainand domain wall resonance signals with intensity at both 220 and 228 MHz. This isclearly observed in pure hcp-Co films (see chapter 2).

In the third place it has to be noted that there is no clear experimental proof thatthe model directly corresponds to the physical reality, especially in there being twograin types with constant α. In the paper where this model was introduced [Michel01],some indications to the applicability were obtained from XRD measurements, but anunequivocal proof was not established. The fact that most measurements can bewell-described within the model is not very conclusive, since the total number of freeparameters only decreases from four to three on applying the model.

For many measurements the model parameters follow from an exact solution ofthe model equations by using the experimental line intensities. This complicates theerror analysis and gives little information about the accuracy and trustworthiness ofthe parameters. The problem can be partly solved by using a fitting procedure to findthe coefficients of the model from the NMR results instead of a algebraic solution.When fitting, the influence of the experimental errors (the uncertainty in the lineintensity) can be taken into account by using different values for each intensity with

3.3. BCC Co(001) AS A BOTTOM ELECTRODE 45

a scatter corresponding to the uncertainty in the experimental determination. It isthen also possible to apply the model to systems that do not give exact solutions,such as the difference intensity between measurements of two different thicknesses.

In this way, some discussion on the physical reality of the model is possible. Anindication of its validity is the fact that the measurements can be described by quitereasonable model parameters that show gradual and systematic variations with thick-ness. Also, the fact that the percentage of fcc-like grains as a function of the positionin the Co layer (figure 3.4(c)) stays below 100% seems to agree with the suppositionthat no reordering of previously grown layers occurs. An indication to the contrary isthe observed variation of α with the position in the layer (figure 3.4(d)), that directlyimplies that the use of one α per grain type is certainly a simplification. One couldconclude from this that the model is a useful way for getting a grip on the devel-opment of the various phases in a film, but that its true correspondence to the realphysical structure is somewhat dubious.

Another point that warrants some remarks is the result of our growth study. Wehave observed a Co fcc(111) phase that appears only above a minimum Co thickness.This result is actually quite new. The Co fcc phase is easily induced by epitaxywith the substrate in the (001) and (110) direction, but in the (111) direction theconversion to the stable hcp phase is rather easy. In practice, for example by growthon Cu(111), the fcc phase is only induced for the first few ML, after which a transitionwith thickness to hcp Co starts to occur [Riedi99, Strijkers00]. Contrary to this, wefind a Co fcc(111) phase that starts to grow after about 3 nm and is stable up toabout 8 nm, with a reasonably high quality.

It may be possible that this remarkable result is related to the strain in the films.Both for growth on Cu as on FeMn about +2% strain is induced. For growth on Cuthis strain only slowly decreases and the transition to hcp may be stimulated by this.For growth on FeMn, we find that the strain is almost fully relieved before the onsetof the fcc phase. This kind of procedure might allow for the growth of more stable Cofcc(111) films than the method of using epitaxiality with the substrate to stabilize it.

3.2.6 Conclusions

59Co NMR in combination with XRD proves to be an ideal tool for determiningthe local physical structure of sputtered, non-single-crystalline MTJ’s. The analysisshows that electrodes sputtered on FeMn start growing in a mixture of fcc and hcpgrains, but, for thicknesses exceeding about 3 nm, extra material grows for almost100% in a (111)-oriented fcc phase. This results is quite remarkable when taking intothe account the relative instability of the Co fcc(111) phase to transformations intothe hcp phase.

Electrodes sputtered on Ta or Al2O3 consist of equal mixtures of fcc and hcp-likegrains of random orientation for any thickness. Theoretical calculations based on thisdifference in growth can explain the observed extra features and asymmetry in FeMnbased MTJ’s as was shown in [LeClair02a].


3.3 BCC Co(001) as a Bottom Electrode

3.3.1 Introduction

Under specific conditions thin Co films can be stabilized in the bcc-phase. For Co, thisphase is strain-induced by epitaxy with the substrate. The first report of bcc Co, madeby Prinz [Prinz85], used growth on GaAs(110). Afterwards, bcc Co has been stabilizedin between Fe-layers and Cr-layers [Houdy91], in between Au-layers [Spiridis04] andon some alloy substrates like TiAl [Kim95] and CoSi2/Si(001) [Wetzel02].

For use of bcc Co in device structures growth of another layer on top of the bccCo, e.g. Al2O3 or Cu, is inevitable. The effect of layers of these materials on the un-derlying bcc Co is not known. For some materials, a reconstruction of the top part ofthe bcc Co into a different crystal phase may occur. This reconstruction may disturba large part of the bcc Co since typical stability limits are below 20 ML. To our knowl-edge no studies have been reported on the influence of additional layers on the bccstructure, apart from layers like Fe that are already known as bcc-inducing substrates.Device structures using Cu layers on top of bcc Co have been reported [Lew00], how-ever without verification of the stability of the bcc Co.

More recent developments have shown an increased interest in bcc Co(001) due toits expected properties in fully epitaxial junctions [Zhang04, Velev05]. Bcc Co(001)bottom electrodes have been used in fully epitaxial junctions with MgO barriers andeven higher MR ratios than for Fe based junctions have been found [Yuasa05].

In this study we investigate the influence of different top layers on the stabilityand structure of ultrathin bcc-Co layers and show that it is indeed possible to growthese layers in such a way that they can be used in semi-epitaxial MTJ’s. This resultprovides a new Co phase for use in this kind of device, which allows for a morecomplete investigation of the influence of crystal structure on tunnelling properties.We also show that Cu top layers, which are used in other device structures, do inducea structural change in the Co layer.

This section is organized as follows: in the section 3.3.2 the growth of bcc Co asreported in literature is reviewed. In the following subsections (3.3.3 and 3.3.4) theexperimental details and the analysis of our own growth of bcc Co on Fe(001) arepresented. In section 3.3.5 the influence of the material grown on top of the bcc Coon the stability of the bcc phase is investigated by means of NMR. In section 3.3.6the feasibility of a MTJ with a bcc-Co(001) bottom electrode is verified by growthand analysis of such a MTJ and in the final subsections a discussion of the results onthe stability of bcc Co and the main conclusions of this study are given.

3.3.2 Reported Results on the Stability and Analysis of BCCCo

A large number of studies has been published on the growth of bcc Co. Very dif-ferent growth and characterization techniques have been employed, and sometimesconflicting results have been obtained.

Although first calculations predicted bcc Co to be a metastable phase, more recentpapers show that the bcc phase is unstable against volume-conserving tetragonal


distortions [Liu93a, Alippi97, Fox99]. The true metastable phase is a body-centred-tetragonal (bct) phase, with a c/a-ratio of 0.92, where c is the lattice constant in thegrowth direction and a is the lattice constant in the plane perpendicular to the growthdirection. The energy barrier stabilizing this phase against tetragonal deformationinto the fcc phase, which is obtained when c/a = 1.41, is however extremely low(0.7 meV per atom). This makes it almost impossible to stabilize thick, pure bct-Cofilms. In practice all pure bcc- or bct-Co layers can only be stabilized by epitaxy onlattice-matched substrates that put a constraint on the in-plane lattice parameter fora limited thickness. We will use the expression bcc Co in this section even when thelayer is strained (c/a 6= 1), because both bcc and bct Co are in practice strain-inducedphases and all grown layers show some strain.

The fact that bcc Co is strain-induced has major consequences for its propertiesand growth. The substrate and the exact growth conditions have a strong influenceon the stability of and the resulting strain on the layers. This influence caused somecontroversy on the maximum stability of bcc Co. Papers by Prinz et al. [Prinz85,Riedi87, Mangan99] and by Bland et al. [Bland91] report bcc Co with thicknesses upto 35.7 nm and 14.5 nm by growth on GaAs(110) and (001) respectively. However,many other groups show stability limits in the range of 10 to 20 ML (i.e. less than3 nm), for various substrates (including GaAs) and crystal orientations [Houdy91,Jay96a, Bruynseraede98, Blundell93, Wu98, Kim96, Spiridis04]. This implicates thatthe stability of the grown layers has to be checked carefully every time. In section 3.3.7some more will be said on this discrepancy in the observed stability.

Reduced average magnetic moments are often observed [Prinz85, Riedi87, Bland91]in bcc Co layers grown on GaAs. Values between 1.4 and 1.55 µB per Co atom arefound. Theoretical calculations [Liu93a, Singh92], however, give 1.7 µB per Co atom.The low measured average moment appears to be caused by strong interdiffusion atthe interfaces over at least several nanometre. The moment in the centre of the layersdoes correspond to the expected value [Bland91]. This interdiffusion may improve thelong range order and diminish the number of defects, thus seeming to stabilize thebcc structure as measured by long range surface techniques like low energy electrondiffraction (LEED). It may, however, cause the local physical and magnetic structureto deviate from that of pure bcc Co. Thus, in order to verify the presence of bcc Cothat is both structurally and magnetically pure not only the structure, but also thelocal magnetic moment has always to be checked.

An ideal way to measure both properties at the same time is 59Co Nuclear Mag-netic Resonance (NMR) which provides the distribution of Co hyperfine fields, a directmeasure of both the local atomic structures and the local magnetic moments in the en-tire Co layer. NMR studies of bcc Co grown on Cr(110) [Houdy91], Fe(110) [Houdy91,Jay96a], Au [Spiridis04], and τMnAl [Bruynseraede98] show that in zero applied fieldpure bcc Co is characterized by a resonance line at a frequency of 199 MHz with afull width at half maximum (FWHM) of about 10 to 15 MHz for well-grown layers.This frequency is about 20 to 30 MHz lower than the frequencies for the fcc and hcpphases of Co. The width of the bcc-Co resonance line is comparable to the fcc- andhcp-Co line widths in typical thin films [Riedi99]. The lack of an NMR signal at199 MHz for a measured structure is a direct proof of the absence of structurally aswell as magnetically pure bcc Co.


Most groups find a maximum bcc-Co thickness of about 1 to 3 nm [Houdy91,Jay96a, Bruynseraede98, Blundell93, Wu98, Kim96, Spiridis04]. The exact value de-pends sensitively on the growth method, temperature, speed, and on the specific sub-strate. The most frequently used substrates, apart from GaAs, are Fe and Cr. Growthby molecular beam epitaxy (MBE) with substrate temperatures between room tem-perature and 200C for growth on Fe and between 150C and 350C for growth on Cris found to produce the highest stability. Lower growth temperatures usually result ina lower macroscopic quality and higher temperatures induce more interdiffusion. Thestability ranges of bcc Co on (001) and (110) surfaces are comparable, but (001) usu-ally shows slightly larger interdiffusion. Extensive NMR measurements on Co/Fe(001)multilayers [Jay96a] indicate that interdiffusion in these structures is limited to about0.6 nm. However Mossbauer measurements on similarly prepared samples show nosignificant interdiffusion [Kalska01], hereby demonstrating the strong dependence onpreparational details.

The fact that bcc Co is not truly metastable results in a large variation of theobserved strains, varying from c/a = 1.15 for 2 ML of Co on Ag(001) [Li89], viaseveral results close to true bcc with c/a = 1.0 [Kim95, Kim96] to values close to thepredicted metastable bct phase with c/a = 0.92 [Boeck96, Gazzadi99]. For similargrowth on the same substrate, different lattice parameters are observed, which meansthat the exact structure also depends strongly on the experimental details. It is thusnot possible to get the precise lattice parameters from literature. The large variety instrains, however, also creates the possibility to stabilize various phases of strained bccCo by choosing substrate and growth conditions. These differently strained phasesgive a larger choice of materials for use in device structures.

Very few studies have been performed on the influence of additional layers on thestability of bcc Co [Jay96a] and none have been reported on the influence of top layersmade of a different material than the substrate. However, for use in a device structure,the growth of extra layers on top of the bcc-Co layer is indispensable. The material ofthese extra layers can not be freely chosen, but is imposed by the desired functionality.Characterization is usually performed before deposition of a top layer. However, dueto the low maximum thickness of the bcc layers, the influence of the additional layermay be quite large, especially if consisting of a non-bcc-inducing material. It is thusbetter to characterize the covered Co layer than to use only in-situ characterizationduring growth.


The bcc-Co layers were prepared by molecular beam epitaxy (MBE) on Fe(001) bufferlayers, grown on either GaAs(001) or on Ge(001) substrates. Growth on Fe was chosenabove growth directly on GaAs because of the commonly observed interdiffusion of theGaAs. The substrate was first cleaned several times by sputter-anneal cycles until aclear (4x6)-reconstruction was observed by LEED for GaAs or a (2x1)-reconstructionin the case of Ge and no contamination could be detected by X-ray photo-electronspectroscopy (XPS) and Auger electron spectroscopy (AES). GaAs(001)-(4x6) is a Ga-rich reconstruction, which prevents large-scale interdiffusion of As into and throughthe Fe layer. The growth of Fe on Ge was initiated at room temperature and continued


after a few nanometre at a temperature of 200C. This two-step growth is necessaryin order to obtain a smooth layer, without Ge diffusing all the way to the top of thelayer. The background pressure during growth always stayed below 10−10 mbar. OnGaAs, Fe was grown at room temperature at a rate of about 1 ML per minute. Thegrowth rate was measured by a calibrated quartz-crystal microbalance. The final Fethicknesses were between 5 and 15 nm. Subsequently, thin Co layers were depositedat room temperature with varying thicknesses between 0.6 and 4 nm. These layerswere covered either with 5.0 nm Fe, with 3.0 nm Cu, or with 2.3 nm Al that wassubsequently plasma-oxidized for 200 s.

The growth quality of these layers was checked in-situ by LEED, scanning tun-nelling microscopy (STM), XPS and AES. The physical structure of the Co layers wasdetermined ex-situ by 59Co NMR. The NMR experiments were performed at 1.5 Kin zero applied field (see sections 2.1.2 and 2.3). NMR measurements provide a wayto determine the relative amounts of Co with specific hyperfine fields, correspond-ing to a distribution of Co atoms over different structural environments (e.g. fcc,hcp, bcc and structures with neighbouring foreign atoms). Among the main advan-tages of NMR are the possibility to directly observe buried layers, the fact that nolong-range order in the layers is necessary, and a sufficient sensitivity for measuringsingle layers with thicknesses well below 1 nm. This high sensitivity is of partic-ular importance for this study. In contrast to studies with Fe on both sides, it isnot possible to grow multilayers of Al2O3-capped Co layers, because Al2O3 is nota bcc-inducing material. The magnetic properties of the layers were measured by asuperconductive quantum-interference device (SQUID) and by the magneto-opticalKerr effect (MOKE). Finally, the bcc Co was implemented in a bcc-Co/Al2O3/Cotunnel-junction structure for measurements on the magneto-transport properties.

3.3.4 Results: Structure

The details of the growth of Fe on GaAs(001) or Ge(001), and of Co on Fe(001) dependstrongly on growth parameters as evidenced by the large number of growth resultsreported in the literature, which are very often at variance with each other. Thus, itis important to always accurately check the crystal quality of the grown layers.

Our first characterization is performed in-situ by means of LEED. In figure 3.6(a)and (b) the structure of the two different substrates used is presented. The Ge(001)surface is clearly (1x2)-reconstructed (a) while the GaAs(001) surface shows a (4x6)reconstruction. This reconstruction is probably a mixture of (4x2) and (2x6) recon-structions, both, however, leading to the same growth of iron including the same uni-axial anisotropy [Moosbuhler02]. Deposition of Fe gives the same results for both sub-strates, independent of the thickness. The LEED pattern of 5.0 nm Fe on GaAs(001)(figure 3.6(c)) shows sharp spots in the 〈11〉 directions and quite broad, out-of-phase〈01〉-direction spots, indicating that the iron grows in the expected (001) bcc struc-ture, although with some roughness. The growth mode, found by measuring LEEDpatterns at different energies, indicates the presence of terraces or pyramids in thestructure, in agreement with results previously found by Gester [Gester97].

LEED patterns for 0.85, 1.35 and 2.0 nm Co, grown on top of the iron, are shown infigure 3.6(d) to (f), respectively. For the two thinnest Co layers at least, the registry of


( a ) ( b )

( c ) ( d )

( e ) ( f )

Figure 3.6: LEED measurements of the growth of bcc Fe and Co on GaAs(001) or Ge(001).The electron energy is about 100 eV, but not exactly equal for all images. Orientation ofthe samples is also not the same for all images, but about equal for (c) to (f), with thesharp corner points corresponding to the first-order peaks in the 〈11〉-directions. CleanGe(001)-(1x2) (a) and GaAs(001)-(4x6) (b), GaAs(001)/5.0 nm Fe (c), GaAs(001)/5.0 nmFe/0.85 nm (d), 1.35 nm (e) or 2.0 nm Co (f).

the iron lattice continues in the Co layer. Thus the Co is also growing in the bcc(001)phase and shows a good order on the LEED coherence-length scale. However, thebackground intensity indicates the presence of quite some roughness. Apart fromthe 〈11〉 and 〈01〉 spots, a c(2x2) reconstruction is clearly observed, getting strongerfor larger thicknesses as evidenced by the innermost points in the 〈11〉 directions.As determined by Kim et al. [Kim96], the c(2x2) reconstruction corresponds to theonset of hcp Co with a growth in the (1120) direction. The in-plane structure ofthis hcp phase fits with some anisotropic compression on the bcc structure by meansof a rotation of 45. In fact, for a single layer, the only difference between thisdistorted hcp phase and the bcc phase lies in the position of the central atom ofthe hcp structure, which is not centred as it is in the bcc structure and gives rise


Inte

nsity [

00] Sp

ot (arb

. units)

E n e r g y ( e V )1 0

0 . 8 5 n m C o

n o C o

2 . 0 0 n m C o

( a ) ( b )9876543210 0 2 0 0 4 0 0 6 0 0

1 . 3 5 n m C o

2 . 0 0 n m C o+ 5 . 0 n m F e +G a A s ( 0 0 1 ) - ( 4 x 6 )

( P e a k I n d e x , n ) 20 2 0 3 0 4 0

0 . 8 5 n m C o 01 0 02 0 03 0 04 0 05 0 06 0 07 0 0

Energy

(eV)

Figure 3.7: LEED I-V curves (a) and I-V peak energy versus peak index squared (b).Curves are shown for 5.0 nm Fe on GaAs(001) and for 3 different thicknesses of Co on top,showing the small difference in out-of-plane lattice parameter between Fe and bcc Co andthe appearance of a different Co phase at 2.0 nm thickness. For clarity, the curves are shiftedon the intensity scale. The peak energies are plotted for two Co thicknesses and show twodifferent slopes (which are proportional to the out-of-plane lattice parameter) for the thickestCo layer.

to the c(2x2) reconstruction. Thus, it may very well be that this structure is anintermediary growth structure of the top of the bcc Co, which may be changed into atrue bcc structure by deposition of an extra Co layer. Indeed, we find that the c(2x2)reconstruction does not immediately give way to the hcp phase upon deposition ofextra Co. Structural characterization by NMR, which measures the bulk and notonly the surface of the layer, has proven that the c(2x2) reconstruction does indicatean intermediary growth structure and not an alteration into the hcp phase, as shownfurther on. For the 2.0 nm thick Co layer the background has become very strong,indicating a considerable disorder and the c(2x2) reconstruction is also relativelystrong. However, LEED patterns measured at a lower electron energy (50 eV) stillshows sharp spots for this thickness. Together with the fact that only part of thesamples grown with this thickness still show a discernable LEED pattern these resultspoint to 2 nm being around the maximum stable thickness for uncovered bcc Co, atleast with our growth procedure.

In figure 3.7(a) LEED I-V curves of the [00] spot measured at θ = 6 for thepure iron layer and the same three Co thicknesses shown in figure 3.6 are plotted.The two thinnest Co layers give peaks which are only slightly shifted compared tothe Fe buffer layer. The thickest Co layer, however, shows two sets of peaks, one


of which roughly corresponds to that of the thinner layers and the Fe substrate.This confirms the phase transition that is occurring at this thickness. A kinematicalestimate of the out-of-plane lattice constant is derived from the energies at whichthe consecutive single-scattering peaks are found, as shown in figure 3.7(b). Thisprocedure is justified for these layers because the absence of strong multiple-scatteringfeatures indicates the relative unimportance of those effects. Also, the sharpnessof the peaks gives an indication that the influence of relaxation at the surface isnot too large. All fits of the peak position versus the peak index squared give aninner-potential shift of about 10 eV, as expected. For the substrates an out-of-planelattice constant a = (0.564± 0.002) nm is found, which is consistent with their bulklattice constant a = 0.5654 nm. The iron buffer, however, is found to be expanded1.5% out of plane compared to the bulk value a = 0.2866 nm. This expansion doesnot significantly depend on the Fe thickness or the substrate type. By bulk elasticconstants the out-of-plane lattice parameter c = (0.291 ± 0.002) nm corresponds toan in-plane lattice parameter a = (0.281 ± 0.002) nm, which is half of the latticeparameter of the substrate. Thus Fe does not show significant relaxation for layersup to 15 nm thickness. This corresponds to results found for 1.5 nm thick Fe grownon GaAs(001)-(4x6) by Gordon et al. [Gordon00]. Other reports show fully relaxed Felayers for thicknesses above 1.5 nm [Doi02], but strained iron films with thicknessesup to 160 nm have also been reported [Schonherr01].

The out-of-plane lattice parameter of the Co grown on this strained Fe layer isfound to be c = (0.287 ± 0.002) nm, significantly larger than the unstrained latticeparameter of bcc Co (0.283 nm). This is especially remarkable since usually an out-of-plane contraction is observed for growth of Co on Fe [Kim96, Gazzadi99]. However,in these studies Co was grown on top of unstrained single-crystalline Fe(001), while inour case there is still a residual strain in the Fe layer which may have an influence onthe growth of the Co layer. In view of the various strains reported for bcc Co, stronglydependent on the exact substrate and growth conditions (see section 3.3.2), this valueis not unreasonable. For the thickest Co layer shown (2 nm), one of the two observedseries of peaks corresponds to a structure that still has the same lattice constant. Theother series gives an out-of-plane lattice constant of c = (0.261 ± 0.003) nm. Thismatches hcp Co oriented in the [1120] direction with about 4% out-of-plane strain, inagreement with earlier results [Gazzadi99].

The roughness of the Fe and Co layers found by LEED measurements is corrobo-rated by STM measurements, which show granular features with typical lateral sizesaround 5 nm. In figure 3.8(a) and (b) STM measurements of 5.0 nm Fe grown on topof GaAs(001) are shown. Figure 3.8(a) has a total area of 200 nm x 200 nm and (b) of50 nm x 50 nm. Step edges of the GaAs are still clearly visible. The height differenceswithin one terrace amount to about 0.8 nm, with a root mean square roughness of0.25 nm. The observed structure is comparable to the pyramidal features describedby Gester et al. [Gester97] showing gradual slopes instead of abrupt steps.

The deposited Co will grow on this relatively rough surface of the Fe. In order todetermine the inherent roughness and growth behaviour of bcc-Co layers on bcc Fe,these have also been grown on an Fe(001) whisker. STM measurements on a 1.5 nm-thick Co layer are shown in figure 3.8(c) and (d), having total scan areas of 200 nm x200 nm and 50 nm x 50 nm, respectively. In the large-scale picture some steps on the


( a ) ( b )

( c ) ( d )

Figure 3.8: STM images of 5.0 nm Fe(001) grown on GaAs(001) (a,b) and of 1.5 nm Cogrown on an Fe(001) whisker (c,d). The total area of figures (a) and (c) is 200×200 nm2 andof figures (b) and (d) 50× 50 nm2. The height differences on a terrace are about 0.8 nm forthe Fe(001) layer and about 0.3 nm for the Co layer. The grain-like features have a typicaldiameter of about 5 nm for both structures.

Fe(001) whisker are visible. On the small scale image the height differences are about0.3 nm with a root mean square roughness of 0.04 nm. The different 5 nm granularfeatures cannot correspond to grains with different orientations. If that would havebeen the case no LEED pattern could have been observed, since the correlation lengthfor LEED is larger than the size of these features. Thus, the crystal structures in thedifferent features have to be in registry with each other. There may, however, bevertical stacking faults separating them. The growth of Co on Fe whiskers seemsto be comparable with the growth of Fe on GaAs in physical appearance, showingcomparable feature sizes and roughness in STM, if the difference in thickness is takeninto account.

It is known that well-grown Fe layers on GaAs(001)-(4x6) show a combinationof a cubic anisotropy and of an uniaxial anisotropy caused by the interface [Zolfl97,Brockmann99]. In order to check these properties of our layers, magnetization curveshave been measured both by SQUID and by MOKE on 5 nm Fe layers grown on


Norma

lized M

agnetiz

ation

H ( k A / m )- 5

1

0

0- 1 0 1 05

H / / [ 1 1 0 ]

H / / [ 1 0 0 ]

- 1

H / / [ 1 1 0 ]

Figure 3.9: Magnetization curves of GaAs(001) / 5.0 nm Fe / 1.2 nm Co / 3.0 nm Cualong several axes as measured by SQUID, showing both cubic and uniaxial anisotropy. Thecurves have been normalized on the saturation magnetization. The measurements have beenperformed at a temperature of 5 K.

GaAs(001) with up to 2 nm Co on top. The magnetization behaviour is mainlythat of the Fe-layer as that is four times thicker than the Co layer. Typical SQUIDloops are shown in figure 3.9. As expected, there is a cubic anisotropy inherent in thestructure with 〈100〉 as the easy axes and a uniaxial anisotropy, caused by the structureof the GaAs interface, with the easy axis in the [110] direction. The combinationof these two anisotropies causes the steps in the magnetization curve for the [100]direction [Zolfl97, Brockmann99]. Apart from the size of the magnetic moment, themagnetization behaviour of the structures is the same for Fe buffers and Fe bufferswith Co layers on top. As usually found, the uniaxial anisotropy is stronger than thecubic for these thicknesses, resulting in an easy axis along the [110] direction. The[110] direction shows a hard axis loop and [100] is intermediary, showing the typicaltwo distinct switching fields found for mixed anisotropies [Gustavsson02].

3.3.5 Results: Influence of Additional Layers

Spin-echo 59Co NMR measurements on buried Co layers were performed as a direct,definitive check on the presence of pure bcc Co in the films for different additionallayers on top of the Co. As a reference and check some measurements were performedon Co layers which have been sandwiched between Fe layers. Since Fe stabilizes thebcc structure in Co, one expects the structure to be left intact and comparable to


59 C

o Spin

-Echo

Intensi

ty (arb

. units)

F r e q u e n c y ( M H z )1 4 0 1 8 0 2 2 0

( a ) ( b )

1 8 0 2 2 02 0 0

4 . 0 n m

3 . 0 n m2 . 0 n m

1 . 0 n m

2 . 0 n m

1 . 0 n m

F e c o v e r e db c c f c c S F / h c p

i d . ( p a r t )

2 4 0

Figure 3.10: 59Co NMR spectra of 1.0, 2.0, 3.0, and 4.0 nm Co layers grown on Fe(001) andcovered with Fe (a) and a cut-out of the 1.0 and 2.0 nm Co layers (b). The measurementshave been performed at 1.5 K in zero applied field. The expected positions for the bcc,fcc and hcp phase are indicated (b). The small peak at 209 MHz for 1.0 nm Co filmcorresponds to Co with 1 Fe nearest neighbour. The high frequency tail of the spectrumis virtually independent of the Co thickness. For clarity, this part of the spectrum is onlyplotted for one thickness. No difference is seen for growth on either Ge(001)/Fe(001) or onGaAs(001)/Fe(001) layers. A sudden switch in growth towards fcc/hcp is seen between 2.0and 3.0 nm thickness.

uncovered bcc-Co layers, although of course, an extra interface is created and someinterdiffusion may occur. The spectra for 1.0, 2.0, 3.0 and 4.0 nm thick single Colayers sandwiched between Fe are given in figure 3.10(a) and are comparable to earlierNMR results on Co / Fe multilayers [Jay96a]. In figure 3.10(b) the central part ofthe spectra of the 1.0 and 2.0 nm Co layers is shown in more detail.

Contrary to usual NMR procedures the spectra are not corrected for the influenceof the RF-pulse power. For the part of the spectrum below about 240 MHz, thecorrection has a negligible influence. For frequencies above 240 MHz the correctioninduces an overestimation of the signal strength, caused by the relatively high noise athigher powers. This strong influence of the noise is created by the low signal strengthin this region for a single Co layer. In order to avoid these misrepresentations thecorrection has been omitted altogether for all presented spectra.

For the 1.0 nm thick layer a sharp line at f = 199 MHz is observed (FWHM9 MHz), comprising about 40 % of the Co intensity. No lines are observed at 216,220 and 228 MHz, apart from the continuous background caused by some interdif-fusion at the interfaces. This directly indicates that the bulk part of the Co layer


is situated in a chemically, structurally and magnetically pure bcc phase. Therebydirectly proving the absence of magnetic or non-magnetic impurities or other crystalphases. The small bump found at 209 MHz corresponds to Co atoms with 1 Fe near-est neighbour (NN) [Jay96a]. At the left side of the bulk peak some intensity centredaround 185 MHz is observed, corresponding to about 0.13 nm of Co. This probablycorresponds to vertical stacking fault-like structures in the Co, where the distancebetween the Co atoms is larger than in the perfect bcc phase. For the 2.0 nm thicklayer, although the highest intensity is still found at 199 MHz, a lot of extra intensityis observed at frequencies corresponding to the fcc and hcp phases and at frequenciesin between those phases and the bcc phase. This is a clear sign that at least parts ofthe Co layer are already transforming to more stable crystal phases. For the 3.0 nmCo layer (figure 3.10(a)) a distinct peak appears at a frequency of about 220 MHz,corresponding to ordered fcc or hcp Co (exactly which is not resolvable). For the4.0 nm Co layer the intensity of this peak strongly increases while the intensity at199 MHz decreases. This indicates that part of the already grown bcc Co transformsto a close-packed structure upon deposition of extra Co.

The position of the bcc peak is influenced by strain in the layer, however, as theseinfluences depend on the volume change, strain which is almost volume conservingwill not be noticed. Besides, neighbouring peaks that are not separated in the spec-trum, corresponding for example to domain walls, may also cause an apparent shiftof the bulk line. The results are in agreement with those found earlier for Co/Femultilayers [Jay96a] with respect to the peak widths and the stability of the bcc Co,although, somewhat less interdiffusion at the interfaces is observed in our case.

As shown above and also already demonstrated by other groups, the stability ofbcc Co is not negatively influenced by growing an Fe layer on top of the bcc-Colayer. For use in device structures, however, usually Cu (GMR-structures) or Al2O3-layers (MTJ’s) are wanted on top of the bcc Co. Of these Cu might be expected tohave a deteriorating influence on the stability of bcc Co. Since Cu is known to be agood template for growth of the fcc phase of Co [Riedi99], it is quite probable thatit has a negative effect on the stability of the bcc phase, which has a very differentlattice parameter. This is indeed seen in figure 3.11, where NMR spectra of variousthicknesses of Co capped by Cu are plotted. For the lowest thicknesses (up to 1.2 nm)the maximum NMR intensity is still at a position corresponding to bcc Co, but theFWHM of the peak has already increased to 22 MHz. Besides this, the peak showsa clear high-frequency tail up to about 225 MHz. This tail is likely to be caused byCo which is transformed to the more stable fcc or hcp phase. This is confirmed bythe results for the thicker Co layers, where the main part of the signal is found togradually shift towards the equilibrium fcc or hcp frequencies (figure 3.11(b)).

For the 1.5 nm thick sample a broad peak is found at 208 MHz, a resonancefrequency with which no ordered Co phase is associated. This simply means that thebcc phase and the fcc/hcp phase of Co are not well-separated phases in this case asthey are in the case of Fe capped layers. The crystalline order on an atomic scale isvery low, resulting in a kind of random stacking and producing resonance frequenciesintermediate to those of bcc and fcc/hcp. For a 2.0 nm thick Co layer, most ofthe intensity is already observed at about 220 MHz. Already at 1.5 nm thicknessthe intensity at 200 MHz is decreasing, implying that less bcc Co is retained and


59 Co S

pin-Ec

ho Inte

nsity (

arb. un

its)

F r e q u e n c y ( M H z )1 4 0 1 8 0 2 2 0

( a ) ( b )

1 8 0 2 2 02 0 0

4 . 0 n m

2 . 0 n m1 . 5 n m

1 . 2 n m

C u c o v e r e d b c c f c c

1 . 5 n m1 . 3

1 . 21 . 0

0 . 8

0 . 7

2 4 0

S F / h c p

Figure 3.11: 59Co NMR spectra of Co layers of various thicknesses grown on Fe(001) andcovered with Cu (a) and a cut-out with several lower Co thicknesses for which the 1.2,1.3 and 2.0 nm films have been vertically off-set for clarity (b). The measurements havebeen performed at 1.5 K in zero applied field. A gradual shift towards higher frequencies isobservable over the entire thickness range.

that the bcc phase does indeed transform to other phases. The intensity of theCo-Cu interface is relatively low as compared to expectations based on the roughsurfaces and the probable presence of vertical stacking faults between them found bySTM. This is probably caused by a flattening of the Co layer and a disappearance ofthe vertical stacking faults, in the process of recrystallization towards a close-packedstructure. The flattening is understandable from the immiscibility of Co and Cu andthe reordering necessary to go from the relatively low density bcc phase to a close-packed structure. Thus, in contrast to coverage by Fe layers, Cu coverage recrystallizesthe already grown bcc Co, giving a disordered atomic arrangement in-between bccand fcc/hcp. This recrystallization most likely removes the surface roughness of theCo layer.

In figure 3.12 the influence of Al2O3 top layers, which are very important for usein magnetic tunnel junctions, is shown. The Al oxidation time was chosen in sucha way that no oxidized Co or unoxidized Al is observed by XPS. For thicknesses upto 2.0 nm a single peak at 198 MHz is observed with a FWHM of about 11 MHz,comparable to the Fe-capped layers. The intensity around 220 MHz does not growupon increasing the thickness from 1.35 to 2.0 nm, although there seems to be someintensity which may be associated with a few local defects in the sample. For theselayers the maximum stability is about 2.0 nm, since other similar samples did showa transformation around this thickness (figure 3.12(b)) just like in the case of Fe


59 Co S

pin-Ec

ho Inte

nsity (

arb. un

its)

F r e q u e n c y ( M H z )1 4 0 1 8 0 2 2 0

( a ) ( b )

1 8 0 2 2 02 0 0

2 . 0 n m

1 . 3 n m

1 . 0 n m0 . 8 n m

A l 2 O 3 c o v e r e d b c c f c c

b o t h 2 . 0 n m

2 4 0

S F / h c p

Figure 3.12: 59Co NMR spectra of 0.8, 1.0, 1.3, and 2.0 nm Co films grown on Fe(001) andcovered with Al2O3 (a) and a cut-out with the 2.0 nm Co film and a nominally equal one,which is not in the bcc phase (b). The measurements have been performed at 1.5 K in zeroapplied field. In (a) the films stay in de bcc phase for all thicknesses. From (b) it is clearthat 2.0 nm Co is about the bcc-Co stability limit since one film is still purely bcc, whilethe structure of a nominally equal film has almost completely transformed towards the fccor hcp phase.

capping. For thinner layers a clear peak is not observable, since the spectrum showsinterface intensity on both sides of the bulk line. This observation, together with theintensity from grain-boundary-like structures around 180 MHz, causes an apparentshift of the bulk frequency from 199 MHz to about 190 MHz for thinner layers. Theintensity of the Co-Al2O3 interface corresponds to about 3 ML of Co with at leastone non-Co nearest neighbour. This is caused by the roughness of the top of the Colayer, which is not reduced in this case, and probably also by the diffusion of someAl into the vertical stacking faults that exist in the Co layer. The exact structureof the Co/Al2O3 interface itself can not be determined, since it is an interface withan amorphous material, which does not give well defined local environments for theinterface Co atoms.

These results show that bcc Co can be used in device structures for thicknesses upto 2.0 nm when it is capped with Al2O3. However, devices using a Cu layer on topof the Co layer will at best have a mixed crystal phase, with non-bcc Co at the topof the structure near the Co-Cu interface. The first result is particularly fortunatefor use in magnetic tunnel junctions, where tunnelling from a bcc(001)-Co bottomelectrode is feasible. Tunnelling is particularly sensitive to the top interface of theCo layer[LeClair01], but although there is quite some interface roughness the local


A p p l i e d F i e l d , H ( k A / m )- 2 0 0 0

( a ) ( b )

2 . 0

Resista

nce (kW

)

2 0 0 - 4 0 0 4 0

Magne

toresist

ance D

R/R (%

)

2 . 1

2 . 2

2 . 3

024681 01 21 41 6

Figure 3.13: Magneto-resistance measurement of a 0.8 nm bcc-Co / Al2O3 / Co / CoOx

junction at 10 K. The observed tunnel magnetoresistance is 16.4%. The field is appliedalong the [100] direction of the bcc Co. A full scan, which destroys the exchange bias afterthe first sweep by a training effect of CoOx, is shown in (a). A minor loop, in which themagnetization of the top Co layer is kept in the equilibrium direction caused by the exchangebias, is shown in (b).

structure is still bcc. Furthermore, as demonstrated, the roughness is not extensiveenough to prevent the formation of a good MTJ. For layers of about 1 nm Co, theCo is single-crystalline with an (001) orientation of the entire layer, thus giving thepossibility to use this structure as a single-crystal bottom electrode.

3.3.6 Results: Tunnelling

Now that the stability of bcc Co against coverage with Al2O3 has been established, wecan try to create magnetic tunnel junctions with single-crystal bcc(001)-Co bottomelectrodes. We used the growth procedures given above for creating suitable bcc-Colayers of about 1.0 nm thick. GaAs(001) /5.0 nm Fe was always used as a substrate.As a top electrode Co is deposited on the Al2O3 barrier. In order to make both aparallel and an anti-parallel alignment of the electrodes possible, the top Co layerwas partly oxidized after deposition, so that an anti-ferromagnetic CoOx layer wasobtained, which causes exchange biasing of the top layer. The junction area is 300 µm× 300 µm. Details of the growth procedure are given by LeClair et al. [LeClair00b].

The resistance of a 0.8 nm bcc-Co / 2.3 nm Al2O3 / 15 nm Co / CoOx at 10 Kversus the applied magnetic field is shown in figure 3.13. A full loop, starting athigh positive fields, is plotted in figure 3.13(a). It shows both a parallel and an anti-


A p p l i e d B i a s V o l t a g e ( V )- 1 0 0 0

( a ) ( b )

0 . 4

Condu

ctance

, G º d

I/dV (10

-3 W-1 )

1 0 0 - 1 0 0 0 1 0 0

Differe

ntial T

MR, DG

/G ap (n

ormaliz

ed)

0 . 0 0

0 . 5

0 . 6

0 . 7

0 . 8

0 . 2 5

0 . 5 0

0 . 7 5

1 . 0 0

Figure 3.14: Conductance versus bias voltage at 10 K (a) for parallel (solid line) andantiparallel (dotted line) magnetizations. The parabolic signature of free-electron tunnellingis clearly present together with a low voltage magnon characteristic. (b) shows the decayof the differential TMR with applied bias voltage. Note that the voltage dependence of theTMR (∆R/Rp) differs from that of the differential TMR (∆G/Gap). The differential TMRbecomes negative for higher bias voltages, while the TMR always stays positive.

parallel alignment of the electrode magnetization and a magnetoresistance of roughly16%. The return sweep of the field does not show a fully anti-parallel alignment,due to a training effect in the CoOx [Nogues99]. However, by measuring minor loopsand not switching the top Co layer away from the direction imposed by the CoOx,magnetoresistance measurements can be performed at repeated field cycles. Thisis shown in figure 3.13(b). The steps around zero field are caused by the uniaxialanisotropy of the Fe bottom layer, which is always aligned parallel with the bcc-Cobottom electrode. The changes in resistance at about ±30 kA/m are probably causedby small rotations of the top Co magnetization. For temperatures close to roomtemperature the exchange biasing is not ideal any more, giving non-perfect alignmentof the magnetization directions. At this moment it is not clear whether the relativelylow value of the magnetoresistance is an intrinsic property of bcc(001) Co, like insome other epitaxial systems [Yuasa02], or because the growth of these preliminarybcc-Co junctions is not yet fully optimized.

Typical conductance versus bias voltage curves are given in figure 3.14(a). Theconductance curves can be described by a parabolic contribution from regular elas-tic tunnelling [Wolf85] and a linear contribution from magnon-assisted tunnelling atlow bias [Zhang97]. In figure 3.14(b) the normalized, differential tunnel magnetore-sistance is plotted as a function of bias voltage, showing the typical behaviour of atunnel-junction structure [LeClair00a]. The top electrode consists of polycrystalline


fcc and hcp Co, which has a different electronic structure than the bcc Co(001). Thiscauses a difference in bias dependence between tunnelling from top to bottom elec-trode and tunnelling from bottom to top electrode. This difference is seen as anasymmetry, as observed in the differential tunnel magnetoresistance. The asymme-try in this case is much smaller than that for the fcc(111) bottom electrode seen insection 3.2.4, corresponding to less pronounced spin-dependent features in the elec-tron bands relevant for the tunnelling process [LeClair02a]. Further optimizationand more systematic studies of these junctions may be in order, to further excluderoughness- and exchange-bias-related problems.

These magnetoresistance measurements show that tunnel junctions using bcc Coas a bottom electrode do indeed work and that a reasonable tunnel magnetoresistancecan be obtained.

3.3.7 Discussion

The stability of the bcc-Co layers of about 2 nm found in our experiment fall in therange commonly observed by different groups on a variety of substrates [Houdy91,Jay96a, Bruynseraede98, Blundell93, Wu98, Kim96, Spiridis04]. The fact that thickerlayers were not obtainable fits in well with the theoretical prediction that bcc Co isnot metastable, but strain-induced [Liu93a]. We also tried to grow bcc Co directly onGaAs(001) instead of on an Fe buffer layer. Also on these substrates we were not ableto reproduce the layers of more than 10 nm thick reported by Bland et al. [Bland91].The stability limits in the study by Bland et al., together with those in studies byPrinz et al. [Prinz85, Riedi87, Mangan99], where thicknesses larger than 30 nm wereobtained on GaAs(110), deviate strongly from those observed by almost all othergroups. It would be a very important advantage for the fabrication of bcc Co if thesethicknesses could be reproducibly obtained. What causes the difference in stabilityranges is not quite clear. NMR measurements on 2 nm Co layers grown by us onGaAs(001), with the growth parameters as described by Prinz et al. [Prinz85], eithershow that the major part of the Co is non-magnetic due to interdiffusion or that theCo is in the fcc or hcp phase.

Despite extensive reported structural analysis on the thick bcc layers, the reasonwhy the stability limit is so high for these bcc structures is still unclear. It may bethat As acts as a surfactant, floating on top of the Co, thereby stabilizing the bccstructure. A similar effect has been observed from an oxygen surfactant [Kim96], thiseffect, however, only stabilized the bcc Co up to 3 nm thickness. On the other hand, itis also possible that the thick layers are stabilized by impurities (mainly As) diffusinginto the Co. The presence of significant interdiffusion of As and Ga into these layerswas already shown directly [Xu87], but the amounts far away from the interfaceswere found to be very low (< 3%). However, the observed low average magneticmoment, 1.4-1.55 µB per Co atom [Prinz85, Riedi87, Bland91], seems to indicatethat interdiffusion may go further than expected. Impurity stabilization might alsoexplain the fact that the 35.7 nm thick bcc layer was never reproduced by Prinz etal. and that this layer was grown at a higher growth rate and less well-controlledconditions than other attempts [Liu93a].

In the mean time a study by Izquierdo et al. citeIzquierdo05 has shown that for


growth on GaAs(110), As diffusion can be replaced by Ga diffusion by the use ofantimony (Sb) as a surfactant for Co thicknesses up to 2.4 nm. The Ga diffusion hasbeen shown to be much less harmful to the magnetic properties.

All characterization results on the thick layers do show that the structure con-sists of a good quality bcc phase. However the effect of low stabilizing-impurityconcentrations is hard to determine. NMR might give a definitive answer to thequestion whether these layers consist of both structurally and magnetically purebcc Co (see section 3.3.2). The only reports of NMR measurements on these lay-ers [Riedi87, Bland91] show NMR spectra with a maximum intensity around 167-170 MHz. These values agree with the reduced average magnetic moment of 1.4-1.55 µB per Co atom in these layers. They do not, however, correspond to the bcc-Coresonance line at 199 MHz. In both cases no significant NMR signal is observed atthis frequency, which would directly seem to indicate that quite large amounts ofimpurities are located throughout the layer, thereby stabilizing the bcc structure butdisturbing the bcc-Co properties. Further NMR measurements on the bcc-Co layerson GaAs by Prinz et al. and Bland et al. may give an answer to this question.

The layers grown in this study have been shown by NMR to consist of pure bcc Co.Since bcc Co is strain-induced, there may be some worries on the stability of the layerin time, which would be detriment to its usefulness. We checked this by repeating theNMR measurements on our samples, two years after growth and found that none ofthe samples with Al2O3 on top showed any change. The bcc structure was perfectlypreserved. Also, annealing for half an hour at 150 or 200C caused no changes. Aftertwo years, the Cu-covered layers also still showed their original structure, but the Cohad become magnetically much harder, which is probably caused by oxygen diffusinginto the edges of the layer and into the vertical grain-boundary-like structures present.Thus the bcc structure in layers up to 2 nm is shown to be stable over a period ofyears.

3.3.8 Conclusions

In conclusion, we have established that 59Co nuclear magnetic resonance is an idealtechnique for a direct determination of the presence of structurally and magneticallypure bcc Co. We have employed this technique to determine the influence of additionallayers on the stability of bcc Co. Although this is very important for the use of bccCo in device structures, to the best of our knowledge no studies on this influence havebeen reported before.

After an analysis of uncovered Co layers grown on Fe(001) / GaAs(001) andFe(001) / Ge(001), which proved them to be bcc up to about 2.0 nm, the Co layerswere covered by Fe, Cu or Al2O3. Results for Fe-topped layers are in agreement withthose found in the literature for Co-Fe multilayers [Jay96a] and show no negative in-fluence on the stability range of the bcc Co. On the other hand, a coverage with Cu,a material which is often used in giant magnetoresistance devices, showed a gradualshift from the bcc phase to the fcc or hcp phase via a range of intermediate phases.This corresponds to a large-scale transformation of the bcc Co in the upper part of thelayer into a locally disordered structure for any Co thickness. Growth of Al2O3 on topof the Co, which is necessary for application in tunnel magnetoresistance structures,


did not show any transformation and left the bcc-Co layers intact up to 2.0 nm. Therelatively large amount of Co-Al2O3-interface signal indicates a rough interface andprobably some interdiffusion into grain-boundary-like structures in the Co, however,without modifying the bcc(001) structure. The bcc structure was found to be stableover a period of years when covered by Al2O3. These results imply that junctionsusing bcc-Co(001) as a bottom electrode can be grown and used to study the influ-ence of the electrode structure on tunnelling properties. Grown junctions using abcc-Co(001) bottom electrode show, although suffering from some roughness-relatedand exchange-bias problems, a magnetoresistance of 16.4% at low temperature. Itis not clear whether this relatively low value is an intrinsic property of bcc(001) Co(see section 3.3.6). A small asymmetry in the conductance vs. bias curves is observedoriginating in the different electronic structures of the top and bottom electrodes.The asymmetry is much smaller than for an fcc(111) bottom electrode.

Chapter 4

Epitaxial Co Films onCu(001)

Abstract: Co films MBE-grown on Cu(001) have been investigated extensively by59Co NMR for thicknesses up to 280 ML. The films are a model system of epitaxialgrowth in thin films and very useful for studying the inherent differences with bulkCo with respect to NMR properties. For optimum quality the films were grown onCu(001) single crystals and were also covered with Cu.

The films are shown to grow in an almost layer-by-layer fashion in a tetragonallystrained face-centred (fct) Co structure. Up to 80 ML thickness the strain is remark-ably stable and less than 10% of the full epitaxial strain is relieved. For thicker films astrong strain relief, which influences the film’s elastic properties, is found. The NMRresonance line is very narrow, indicating a very high structural quality of the film anda very homogeneous strain.

For films below 20 ML thickness the influence of the spin-dependent electron scat-tering at the interfaces is observed for at least four Co atomic layers from the interfacewith monolayer resolution. An oscillatory effect on the Co hyperfine field with a pe-riod of several monolayers is measured, corresponding to the oscillating conductionelectron polarization. This observation is unique for NMR on Co structures since itusually only resolves the interface layer itself. The observation is only possible due tothe very narrow NMR lines.

For thicker films the effects of the tetragonal strain in the film can be observeddirectly by NMR. Apart from the expected anisotropy of the Co hyperfine field, thereis also an anisotropy of the Co line width. This anisotropy is caused by the quadrupolesplitting due to the electric-field gradients (EFG’s) in the film. The electric-fieldgradient can be measured directly for the first time in thin Co films. A quantitativeagreement with the strain in the films is found.

The additional NMR resonance-line width of fct-Co films relative to bulk fcc Cocan be fully explained by the combination of direct effects of the interfaces and theeffect of the strain in the film.

Most of the contents of the second section have been published as ‘Monolayer

65

66 CHAPTER 4. EPITAXIAL Co Films ON Cu(001)

resolved oscillating hyperfine fields in epitaxial face-centered-tetragonal Co(001) films’in Phys. Rev. Lett. 93, 177205 (2004) and as ‘59Co NMR observation of monolayerresolved hyperfine fields in ultrathin epitaxial fct-Co(001) films’ in J. Magn. Magn.Mater. 286, 390 (2005).

Most of the contents of the first and third section have been published as ‘Electric-field gradients in thin face-centered-tetragonal Co films observed by nuclear magneticresonance’ in Phys. Rev. B 72, 155409 (2005).

Contents

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 67

4.2 Structure and Strain . . . . . . . . . . . . . . . . . . . . . 68

4.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 68


4.2.3 Structural Quality vs Template . . . . . . . . . . . . . . . 69

4.2.4 Strain vs Thickness . . . . . . . . . . . . . . . . . . . . . . 70

4.2.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

4.2.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 75

4.3 Influence of the Interfaces . . . . . . . . . . . . . . . . . . 76

4.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 76


4.3.3 Hyperfine Fields Measured with Monolayer Resolution . . 77

4.3.4 Origin of the Hyperfine Field Variations . . . . . . . . . . 81

4.3.5 Absence of an Interface Signal . . . . . . . . . . . . . . . 81

4.3.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

4.3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 85

4.4 Influence of the Strain . . . . . . . . . . . . . . . . . . . . 85

4.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 85


4.4.3 Hyperfine Field Anisotropy . . . . . . . . . . . . . . . . . 87

4.4.4 Quadrupole splitting . . . . . . . . . . . . . . . . . . . . . 87

4.4.5 Electric Field Gradients: Measurement . . . . . . . . . . . 92

4.4.6 Electric Field Gradients: Calculation . . . . . . . . . . . . 93

4.4.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

4.4.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 97

4.1. INTRODUCTION 67

4.1 Introduction

The physical structure of thin films generally shows significant deviations from thatof bulk material. An example of this is the strain arising from the different latticeparameters of the film material and the substrate. In magnetic thin films, whichare of strong interest due to the use of these films in various spintronic devices ormodel systems [Tsymbal03, Wolf01], the magnetic and electronic properties may bestrongly influenced by this strain and other structural aspects. This influence makesit important to accurately investigate the specific physical structure of thin films.

A possible method for studying this is the use of 59Co NMR on thin Co films. Thisallows for a direct determination of the physical structure of the film on an extremelylocal scale, thus minimizing the direct effects of the limited thickness compared to bulkmaterial. Furthermore, NMR may provide additional information on the magneticand electronic structure within the film.

So far all possible Co crystal phases have been identified in thin-film structuresby NMR measurements [Riedi99] and apart from the presence of a quite homoge-neous strain in many systems (resulting in a small shift of the resonance frequency),some noteworthy differences with bulk-Co spectra have indeed been observed. Mostremarkable is the large line width usually found for thin films. Whereas bulk fccand hcp materials have line widths of about 0.7-1.0 MHz (full width at half of themaximum intensity, FWHM) [Thomson96, Fekete78], almost all thin Co films haveline widths of 5-10 MHz.

The broadening of an NMR line can be caused by the direct influence of theinterfaces on the hyperfine field [Wieldraaijer04] and also by the electric-field gradients(EFG’s) resulting from the deformation of the structure (see section 4.4.5), but botheffects are far too small to account for the 4-9 MHz broadening. The large broadeningin thin films is almost certainly caused by an inferior structural quality and a poorhomogeneity as compared to the bulk material. This poor structural homogeneityprecludes detailed studies on the properties of thin films.

A model system that may allow for this kind of investigation is Co grown onCu(001). Under proper conditions the Co film has been shown to grow in an almostlayer-by-layer fashion and to form a face-centred-tetragonal (fct) structure of highquality over a reasonable thickness range [Schmid92, Navas93, Weber96, Alphen96,Strijkers00]. This fct-Co phase is stabilized by the relatively small lattice mismatch(−2.0%) with the Cu and is unique for thin films, since the inherent homogeneousstrain cannot be obtained in bulk Co. This system and its NMR properties areinvestigated for Co thicknesses between a few tenths and a few tens of nanometres.

The chapter is organized as follows. In section 4.2 the structural analysis ofthe Co films is presented for the entire thickness range, paying particular attentionto the strain and the structural quality. In section 4.3 an analysis of the directinfluence of the interfaces is presented, independent of the structural deformationpresent of the film. The analysis is performed by studying Co films thinner than20 ML (3.5 nm), which show a variation of the hyperfine field over several monolayersfrom the interface. In section 4.4 an analysis of the indirect influence of the interfacesis presented, namely the effects of the tetragonal deformation induced in the film.The analysis is performed by studying Co films thicker than 20 ML, which show


a measurable electric-field gradient caused by the strain. In this section, also, theresulting NMR line widths for this system are explained by combining the effects ofthe interfaces and the strain (section 4.4.7).

4.2 Structure and Strain

4.2.1 Introduction

Co deposited on Cu(001) by MBE is known to grow in an almost layer-by-layerfashion and to form a strained fcc(001) (i.e. fct(001)) structure [Schmid92, Weber96,Alphen96]. However, the precise quality and the development of the strain withthe thickness of the film may depend significantly on the precise growth conditions.Furthermore, the development of the structural quality and strain by means of NMRhave not been performed for this system yet. In this section the optimum substrateand the development of the strain with the thickness are investigated.

For maximum quality and homogeneity only single Co films are used instead ofmultilayers. These last are certainly easier to measure, but are usually of lowerstructural quality simply due to accumulating defects. Moreover, contrary to themultilayer case the Cu will not be strained, since its thickness is virtually infinite.

The section is organized as follows: after providing the experimental details (sec-tion 4.2.2), we first compare Co layers on two types of Cu(001) substrate in orderto determine which gives the optimum quality (section 4.2.3). Next, we analyse thestructure and strain of the best films by means of X-ray diffraction (XRD), scan-ning tunnelling microscopy (STM) and low energy electron diffraction (LEED) andby NMR (section 4.2.4). In the discussion section the results are compared with theliterature. The section ends with some conclusions.


Cu(001) single crystals were cleaned by sputter and anneal treatments until the surfacewas atomically flat and clean, preliminary to the deposition of Co. The Co wasMBE-grown at a sample temperature of about 50C, a background pressure below10−10 mbar and a growth rate of about 1 ML/min. The Co thickness was controlledby an accurately calibrated quartz-crystal microbalance. Co thicknesses of 4, 6, 7, 8,9, 10, 12, 15, 20, 40, 80 and 280 ML were grown and the growth quality was checkedin-situ by LEED and STM. All Co layers were subsequently covered by 4 nm of Cu toprevent oxidation and to obtain optimally homogeneous Co structures with symmetricinterfaces. The completed layer structures were analysed ex-situ by XRD and 59CoNMR. The NMR experiments were performed at 2 K in a home-built frequency-tuned spin-echo NMR spectrometer. The sensitivity of the spectrometer is sufficientfor measuring Co layers down to single-ML thickness [Strijkers99].

As a comparison Co films were also grown in similar conditions on 100 nm Cubuffer layers deposited on Si(001) substrates. The Cu buffer layer grows in a well-defined (001) orientation, although with an inferior quality than the Cu single-crystals.Co films of 14, 28, 42 and 56 ML were grown and covered with 4 nm Cu.

4.2. STRUCTURE AND STRAIN 69

59 C

o Spin

-Echo

Intensi

ty (arb

. units)

F r e q u e n c y ( M H z )2 1 5 2 2 0

f c c ( u n s t r a i n e d )

4 2 M L

2 8 M L

1 4 M L

( a ) o n C u ( 0 0 1 ) b u f f e r ( b ) o n C u ( 0 0 1 ) s i n g l e c r y s t a l

5 6 M L

2 1 5 2 2 0

f c c ( u n s t r a i n e d )

4 0 M L

2 0 M L

8 0 M L

Figure 4.1: NMR spectra showing the difference in structure and quality for Co(001) filmsgrown on two different templates. The spectra have been measured in zero applied field atT = 2 K. (a) 14, 28, 42 and 56 ML thick Co films grown on Si(001) / 100 nm Cu(001) and(b) 20, 40 and 80 ML thick Co films grow on a Cu(001) single crystal. The spectra havebeen scaled in such a way that the area under the graphs is proportional to the Co thicknessfor (a) and (b) separately. The resonance frequency is proportional to the hyperfine field atthe Co nuclei (-10.054 MHz/T).

4.2.3 Structural Quality vs Template

First the structural quality of the Co layers grown on the Cu(001) buffer is comparedwith those grown on the Cu(001) single-crystal. The quality on a strictly atomic scalecan be evaluated from the NMR resonance-line width. Such a measurement is notinfluenced by the better long-range order in the single crystal, but is only sensitive tothe local properties. Since the Co structures are nominally the same, any difference inthe NMR line width can be directly related to the presence of local defects, stackingfaults or inhomogeneous strain in the films. A larger width directly indicates aninferior structural quality.

Co NMR spectra for films grown on a Cu buffer are plotted in figure 4.1(a) andthe spectra for three films grown on a Cu single-crystal in figure 4.1(b).

The spectrum of the thinnest Co film grown on a Cu buffer consists of a singleline at 214.4 MHz, corresponding to Co in the fcc phase (217 MHz) [Riedi99]. Theshift of the resonance line is caused by a small volume strain of the Co, which will bediscussed more extensively in later sections.

With increasing film thickness the line gradually shifts to 215.7 MHz for the 56 ML


Co film, indicating that a significant part of the strain is gradually relieved in thisthickness range. Except for the thinnest film all line widths are about 2.5 MHzFWHM, indicating a similar structural quality in all films and a homogeneous strainthroughout the films.

The film strain stays homogeneous if strain relief occurs solely by means of misfitdislocations at the interface. This process is accompanied by stacking faults withinthe film [Matthews70, Schall04], which are observed as the NMR lines at 219 and222 MHz [Riedi99]. The value for the strain relief derived from the measured stackingfault density [Matthews70] is comparable to the value determined from the NMR lineshift.

The spectra for 20, 40 and 80 ML Co films grown directly on a Cu(001) singlecrystal (figure 4.1(b)) do not show this strain relief, as evinced by the constant positionof the resonance line at 214.6 MHz (very close to the frequency for the thinnest filmon a Cu buffer) and the virtual absence of stacking faults. Additionally, the linewidth is strongly reduced (1.6, 1.2 and 1.1 MHz respectively), compared to the filmsgrown on a buffer. These line widths are smaller than those of any other epitaxialthin Co film [Riedi99] and only a few 100 kHz larger than the widths observed for Cosingle-crystals [Thomson96].

Thus, Co layers on a Cu(001) single-crystal show a very high-quality, strained fccphase with hardly any strain relief or stacking faults up to at least 80 ML. This incontrast to Co layers grown on a Cu(001) buffer layer, which are also of quite highquality, but show strain relief and associated stacking faults with increasing thickness.This means that Co layers grown on single-crystal substrates are much more suitablefor the investigation of perfect thin films. We will exclusively investigate these in theremainder of this study.

4.2.4 Strain vs Thickness

STM, LEED and XRD

First, the structure of the films and the strain in the films have been investigated asa function of thickness by means of STM, LEED and XRD.

The STM measurements show that, for films up to 10 ML thickness, the Co growsin a nearly layer-by-layer mode as expected [Schmid92] (see figure 4.2(a) and (b))and not more than three layers are observed simultaneously. For thicker films, theroughness increases slightly, for a 20 ML film (c) and a 40 ML film (d), 4 and 5layers are visible respectively. Thus, although the growth is not in an ideal layer-by-layer fashion, it is quite close to it and the deviation is not expected to significantlyinfluence the film structure.

LEED patterns show a square surface structure as expected for the fcc(001) di-rection without any thickness dependence, even for still thicker films, indicating theabsence of surface reconstructions and establishing the sole presence of the (001)-orientation.

The 2% larger lattice parameter of Cu will lead to an expansion of the Co lattice inthe plane of the film. This is accompanied by an out-of-plane contraction and results ina deformed fcc phase: face-centred-tetragonal (fct) Co. If, as usual, strain relief occurs


( a ) ( b )

( c ) ( d )

Figure 4.2: STM images and LEED patterns (insets) for uncovered Co films of (a) 3 ML,(b) 10 ML, (c) 20 ML and (d) 40 ML grown on a Cu(001) single crystal. All LEED patternswere taken at an electron energy of about 78 eV. The scale of the STM images is 90x90 nm2.Note the Cu-substrate step visible in the lower left quadrant of (d).

at the interfaces, the strain can be described by a single parameter since the out-of-plane strain (ε⊥ = ∆c/c0), the volume strain (∆V/V = (c0+∆c)(a0+∆a)2/(c0a

20)−1)

and the tetragonal deformation (c/a) can all be calculated from the in-plane strain(ε‖ = ∆a/a0) by the standard elastic theory [Feynman63]. Due to the very largethickness of the Cu single crystal only the Co film will be strained, with a maximumexpansion for Co(001) of ε‖ = +1.99% at room temperature and +2.00% at liquid Hetemperatures1.

1We use for the lattice constants of fcc-Co and Cu: aCo = 0.35445 nm, aCu = 0.3615 nmat room temperature and aCo = 0.3532 nm and aCu = 0.3604 nm at liquid He tempera-ture [Landolt-Bornstein86]


The corresponding out-of-plane contraction in the fcc(001) orientations followsfrom [Feynman63]

ε⊥ = γε‖ = −2c12

c11ε‖. (4.1)

Using c11 = (2.42 ± 0.02)1011 N/m2 and c12 = (1.60 ± 0.02)1011 N/m2, one findsγ = −1.32 ± 0.03 for fcc Co at room temperature [Landolt-Bornstein86]. Thus, amaximum out-of-plane strain ε⊥ = −(2.63 ± 0.06)% and a maximum volume strain∆VCo/VCo = +(1.29± 0.03)% is expected for our Co-films.

The out-of-plane lattice constant c has been measured by means of LEED andXRD in order to determine the strain in the films.

For the top monolayers of uncovered Co films c can be determined from LEEDI-V curves measured on the [00]-spot. The energies at which the consecutive single-scattering peaks are found can yield a kinematical estimate of c. The accuracy of thisanalysis is, however, limited due to the presence of multiple scattering features. Also,the top-most uncovered Co ML’s may have a lattice parameter somewhat deviatingfrom the bulk value [Cerda93, Muller95]. The results are thus mainly useful forcomparing the strain for different thicknesses and not for obtaining its absolute values.

For Co thicknesses of 10, 20 and 40 ML, c = (0.347 ± 0.001) nm is found, whichcorresponds to ε⊥ = −(2.1 ± 0.2)%. The 80 ML Co film has ε⊥ = −1.6% and the280 ML film ε⊥ = −0.9%. Apart from the absolute values, this indicates relaxationsetting in somewhere between 40 and 80 ML and, a still significant residual strain at280 ML.

After deposition of a 4 nm Cu film the completed structure was analysed by XRD.Because of the small separation between the large peak from the Cu crystal and theCo(002) peak, only the samples with 80 and 280 ML Co films could be measuredreliably. A θ− 2θ scan of the 80 ML sample resulted in c = (0.3458± 0.0002) nm, i.e.in ε⊥ = −(2.45± 0.07)% and c/a = 0.958. This contraction corresponds to (93± 5)%of the expected full strain and indicates the presence of minimal strain relief only.The out-of-plane correlation length calculated from the line width is close to 80 ML,indicating a homogeneous strain and the virtual absence of stacking faults withinthe film. This result differs from the LEED-I-V results, which showed a significantrelaxation in the (top part of the) 80 ML film. However, this difference can easilybe attributed to the Cu coverage, which stabilizes the strain of the top part of theCo film from the top side. Due to this, the strain in our films is more stable than inuncovered films, demonstrating the inapplicability of surface methods to determinethe strain in buried layers.

However, a significant strain relief was found in the Cu-covered 280 ML film,where only 45% of the strain remained. As expected [Matthews70], this strain reliefis accompanied by stacking faults within the film, leading to a reduced correlationlength of about 110 ML. Strain relief was also observed after annealing of the 80 MLfilm. This film showed 80% and 70% of the maximum strain after annealing at 300oCand 400oC, respectively, both accompanied by some reduction in correlation length.

Both the LEED and the XRD measurements have been performed at room tem-perature, as opposed to the NMR measurements that are performed at 2 K. However,


since the temperature dependence of the Co and Cu lattice parameters is very closelysimilar, no difference in the structure or the strain of the films is expected.

Thus, in our covered 80 ML Co film not more than 10% of the epitaxial strain isrelaxed. This indicates that strain relief is minor for all films up to this thickness,agreeing with the NMR results shown in figure 4.1, where no shift of the resonance lineis observed for layers of 20, 40 and 80 ML thickness. Very thick films or annealed filmsdo show a significant relaxation accompanied by a reduction of homogeneity. The filmscan be described as pure fct Co(001), since the (tetragonal) strain is homogeneousover the film and only a single orientation is present.

NMR

By means of NMR measurements the volume strain in the film can be determined.The enlarged volume of Co grown on Cu(001) leads to a reduced hyperfine field andthus a frequency shift. Unstrained fcc Co has a resonance frequency of (217.2 ±0.1) MHz at liquid He temperatures (Bhf = −21.60 T) [Kawakami72, Thomson96,Bromer78, Strijkers99]. The relative NMR frequency change is linearly proportionalto the relative volume change: ∆f/f = ∆Bhf/Bhf = −(1.13 ± 0.01)∆V/V at roomtemperature2, with f the NMR frequency. At low temperatures the proportionalityconstant may be estimated3 to be −1.16± 0.06

However, if the cubic symmetry is broken, the hyperfine field is anisotropic [Fekete78,Riedi99]: Bhf(θ) = Bhf,iso+ 1

2Bhf,ani(3 cos2 θ−1), with Bhf,iso and Bhf,ani the isotropicand anisotropic parts of the hyperfine field, respectively, and θ the angle with the sym-metry axis, i.e. the film normal in our system (see section 2.1.2). If the effects areindependent, Bhf,iso is expected to change with the volume strain as indicated above,while Bhf,ani may be expected to be roughly proportional to the tetragonal defor-mation c/a. Since, for maximum sensitivity, thin film NMR analysis is commonlyperformed without an applied field [Riedi99], the anisotropy of the hyperfine field isusually ignored as it can only be observed by means of applied fields.

In order to separate Bhf,iso the samples were measured in a field of several Tesla,applied both in the plane of the sample and perpendicular to the sample. For theperpendicular measurements the 1.80 T demagnetization field has to be taken intoaccount. Bhf,iso = (−21.24±0.01) T is found for all films up to 80 ML thickness. Thisvalue is about 0.1 T smaller than the in-plane hyperfine field measured in zero appliedfield. The difference is caused by the anisotropy. The measured value corresponds toa constant volume strain of ∆V/V = (+1.44±0.08)%. We determine ε‖ by combiningthe volume strain with ε⊥ = (−2.45± 0.07)% as determined by XRD (section 4.2.4)and find ε‖ = (+1.95±0.08)%, confirming the almost full strain. The ratio γ betweenε⊥ and ε‖ (equation 4.1) is then γ = (−1.26 ± 0.04), in good agreement with theexpected γ = −1.32± 0.03.

2The frequency shift has been found to be linearly related to an applied isotropic pressure [Jones60,Anderson64], with ∂ ln f/∂p = (6.01 ± 0.01) · 10−12 Pa−1, with p the applied pressure. Since thevolume change is also linearly proportional to the applied pressure (∂ ln V/∂p = K−1 = 3(c11 +2c12)−1 = (5.35±0.06) ·10−12 Pa−1, with K the bulk modulus), a linear relation exists between thechange in NMR frequency and the change in volume: ∆f/f = ∆Bhf/Bhf = −(1.13± 0.01)∆V/V .

3At low temperatures the elastic constants c11 and c12 are a few percentlarger [Landolt-Bornstein86].


Elastic

Const

ant, g

I n - P l a n e S t r a i n , e / / ( % )0 . 0 0 . 5 1 . 0 1 . 5

- 1 . 2

- 1 . 3

- 1 . 4

- 1 . 5

- 1 . 6 2 . 0

( 2 8 0 M L C o )

( 8 0 M L C o )

( 8 0 M L C oa n n e a l e d 3 0 0 C )o( 8 0 M L C oa n n e a l e d 4 0 0 C )o

Figure 4.3: Measured γ versus the in-plane strain ε‖ in the film. The horizontal line andthe grey area indicate the value of γ from the literature [Landolt-Bornstein86]. The pointwith maximum ε‖ corresponds to the fully strained 80 ML Co film, the other ones to filmsshowing partial strain relief and the accompanying stacking faults.

Smaller strains are obtained for the 80 ML film after annealing at 300oC and400oC and for the 280 ML film (in order of decreasing residual strain). In figure 4.3the value of γ is plotted as a function of ε‖. Instead of the expected constant γ amarked linear dependence on the strain is observed. This effect may be related to achange of the elastic properties due to the presence of stacking faults associated withthe misfit dislocations in the film [Matthews70, Schall04]. The number of these faultsin the film increases with increasing strain relief. Although this change in γ has notbeen reported yet, the influence of dislocations on several other elastic properties iswell-known [Hirth92] and some deviations have also been observed in other thin filmsystems [Wolf88, Goudeau04].

Thus, we have observed both a change and a significant anisotropy of the hyperfinefield compared to the value for bulk fcc-Co. The anisotropy is commonly ignoredin thin film NMR analysis, but should be taken into account when performing aquantitative strain analysis. The change of hyperfine field, in combination with theXRD measurements, allows us to determine the elastic properties of the film and givesa strong indication that these properties change as the film relaxes.

4.2.5 Discussion

The results on the development of strain in the films obtained by LEED and XRDdiffer in some respects from those published in literature. Many studies have been


performed on the growth of the first few ML of Co on Cu(001) or on the growth ofCo/Cu multilayers. The number of studies on the growth of single Co films thickerthan about 10 ML is quite limited. Most studies find almost layer-by-layer growthabove 2 ML thickness and up to at least 20 ML [Li90]. However, the values for thestrain vary somewhat [Clarke87, Cerda93, Heckmann94] and the distinction betweenthe polycrystalline and single-crystalline elastic constants is not always made. Strainsare determined by LEED and the variation in results seems to confirm that care hasto be taken with this kind of measurement when using it to obtain absolute values ofthe lattice parameter for films thicker than a few ML.

Studies of even thicker layers have been performed by Weber et al. [Weber96] andSchindler et al. [Schindler00]. Schindler et al. found the presence of two Co phaseswith different strain for films thicker than 10 ML, which may be specific for theelectrodeposition method used. Weber et al. measured the in-plane lattice spacingof MBE-grown Co films up to 40 ML thickness on Cu(001) by reflection high-energyelectron diffraction (RHEED). They found a fully strained phase up to 10-15 ML ofCo depending on the growth temperature and above that thickness an exponentialrelaxation to a (extrapolated) residual strain of about 70-80% of the maximum strain.This result corresponds to ours in the prediction of large residual strains for thick films(80-85% of the full strain for 40-80 ML Co films). In our case the residual strain in thisrange is even higher due to the influence of the Cu top layer. The observed strain reliefreported for uncovered Co layers is partly a reversible property of Co surface layers,which may be undone by subsequent Cu coverage. However, the strong strain reliefin our thickest film is an irreversible event, different from the extrapolated behaviourof thinner films [Weber96].

4.2.6 Conclusions

Thin Co films have been grown on Cu(001) single-crystal substrates by MBE. Thesefilms are found to be in a homogeneously strained fct phase. Films up to 80 ML showmore than 90% residual strain relative to perfectly lattice matched Co on Cu(001).By combining XRD and NMR measurements, we find that the behaviour of the Co isdescribed well by elastic theory. The somewhat larger strain in our systems, comparedto earlier results in literature [Weber96] is probably caused by the Cu coverage layer,emphasizing the significant difference between uncovered and covered thin films. Abreakdown of the gradual strain relief is observed for films of several tens of nm’sthickness. These have a much smaller residual strain than predicted by extrapolation.

The large residual strain for films up to 80 ML thickness only occurs for directgrowth on Cu single crystals, growth on Cu buffer layers results in a strong gradualstrain relief. Our measurements indicate that strain relief not only decreases thestructural quality, as expected, but also significantly changes the elastic properties.


4.3 Influence of the Interfaces

4.3.1 Introduction

The hyperfine field of an atom in a magnetic solid is a powerful probe of the struc-tural, magnetic and electronic properties of the local environment. Although it isan extremely local probe of atomic-scale properties, some longer ranging influencesdo occur, particularly via the spin-polarization of conduction electrons, which is re-sponsible for the RKKY-coupling between distant magnetic impurities and betweeninterlayer exchange coupled magnetic layers [Fert94]. The oscillations of the Knightshift of Cu caused by this effect have been directly observed by NMR [Lang74, Jin94].

The influence of foreign atoms or interfaces on the hyperfine field of magneticatoms may also reach several atomic distances. However, the large inherent widthof the Co resonance line usually obliterates these relatively small effects. Only theinfluence of directly neighbouring foreign atoms or interfaces can usually be resolvedby Co-NMR [Riedi99].

The line width is reduced in Co films of very high structural quality, thus, forthese an observation of the longer ranged effects may become feasible. A particularlysuitable system is Co on Cu(001), since it is both an archetypal interlayer couplingsystem [Bruno91, Vries95] and a model structure for ultrahigh-quality growth bymolecular beam epitaxy (MBE) methods (see section 4.2). This system, combining astrong ferromagnet with a simple noble metal in a one-on-one epitaxial relationship,may be ideal for a study of a longer ranging influence of interfaces.

In this section the NMR observation of the effect of a Co/Cu-interface on the Cohyperfine fields up to a few atomic distances from the interface is presented. Theeffect can be resolved up to at least the fourth atomic layer in MBE grown Co filmson Cu(001), and shows a small oscillation in strength with distance. This uniqueobservation is facilitated by the virtually perfect structural quality of this system,distinctly narrowing the resonance lines in comparison to other structures. Thisprovides a resolution of the distinct hyperfine field contributions.

The section is organized as follows: after very briefly repeating a few experimentaldetails, the hyperfine fields of ultrathin Co films as measured by NMR are given(section 4.3.3). A variation of the hyperfine field with the position within the Colayer is observed. In section 4.3.4 the origin of this variation is traced to the spin-dependent electron scattering at the interfaces with the Cu. In section 4.3.5 thesurprising absence of a signal corresponding to the interfacial Co atoms is analysed. Inthe discussion section the results are compared with Mossbauer results. Furthermore,the conditions imposed on the line width in order to make these observations possibleare estimated and finally the influence of the non-perfect layer-by-layer growth arediscussed. The section ends with some conclusions.


Co layers with thicknesses of 4, 6, 7, 8, 9, 10, 12, 15, 20 ML, MBE-grown on Cu(001)single crystals, were analysed by means of 59Co NMR. The growth parameters can befound in section 4.2.2. The growth occurs in a fashion that is close to layer-by-layer

4.3. INFLUENCE OF THE INTERFACES 77

growth. The films are known to be in a high quality, very homogeneously strainedfct(001) phase. Hardly any strain relief has occurred, so the Co retains its epitaxialrelationship with the Cu-substrate and contains only negligible amounts of stackingfaults as shown in section 4.2.

4.3.3 Hyperfine Fields Measured with Monolayer Resolution

The Cu(001)/Co films studied with Co thicknesses of 20 ML (3.5 nm) or higher havean NMR resonance-line width of 1-2 MHz FWHM, much smaller than the usual 5-10 MHz found for thin fcc films and of the same order of magnitude as the line widthfor bulk fcc Co (0.7 MHz) (see section 4.2).

For Cu(001)/Co films with Co thicknesses below 20 ML the NMR line widthincreases strongly (up to more than 5 MHz FWHM) with decreasing thickness. Sincethe strain is only marginally larger for the thinnest films and the structural qualityis at least as good as for the thicker layers, this broadening can only be attributed toan apparently larger influence of the interfaces.

To study this in more detail, a more accurate measurement of the thinnest filmswas performed. The resulting spectra of the bulk Co for thicknesses of 6, 7, 8, 10,12 and 15 ML are plotted in figure 4.4(a). A deviation from a single Gaussian linebecomes apparent in this resonance line of Co atoms in one and the same environment,namely a homogeneously strained fct phase with no foreign neighbour atoms.

We will now show in detail that the spectra straightforwardly point to the ex-perimental fact that indeed distinct contributions of Co ML’s at various distancesfrom the interfaces can be distinguished. Apart from a small, very broad background(which is attributed to sample edge effects), the 6 ML film seems to consist of twosub-lines (at about 211 and 214 MHz), although these are very much merged. Theinternal structure gets more clear for 8 ML and shows two lines at approximately thesame positions as for the 6 ML film. However, additionally a third line is present atabout 216.5 MHz. For even thicker layers a constantly growing fourth line seems toappear around 215 MHz. To obtain a quantitative understanding of these features,the spectra of Co thicknesses of 6, 7, 8, 9, 10, 11, 12 and 15 ML, have been fitted(within the range of 209 to 218 MHz) with the sum of up to four Gaussian line pro-files. It is found that each spectrum can be excellently fitted by the sum of at mostfour lines of equal or very similar width. This width, of about 2 MHz FWHM, iscomparable to the total line width for much thicker Co layers, where the influence ofthe interfaces on the spectra is negligible. An example is given in figure 4.4(b), whereboth the unsmoothed spectrum of a 10 ML Co film and its fit (solid line) are plotted.

The positions of the lines vary only slightly and gradually and the relative areasdevelop in a systematic way with film thickness. The two sub-lines of the 6 MLspectrum have a roughly equal area, denoted by A. Since the two Co layers that aredirectly at the interfaces, do not lie in this ‘bulk’ region of the spectrum, an area of12A in the spectrum corresponds to one ML of Co. Analysis of the 7 ML spectrumresults in the same two lines, each with an area of A again, together with a third linewith an area of roughly 1

2A. In the 8 ML measurement, the area of this third linehas increased again to A. For the 9 ML film a fourth line of area 1

2A appears and theintensity of this line doubles for the 10 ML film, which thus is built up by four lines


59 Co S

pin-Ec

ho Inte

nsity (

arb. un

its)

F r e q u e n c y ( M H z )2 1 0 2 1 5 2 1 0 2 1 5

1 51 21 0876 M L

I

I I I I II V ( a ) ( b )

I I I I I II V

1 0 M L C or a wd a t a

Figure 4.4: (a) Smoothed NMR bulk spectra for Co layers of 15, 12, 10, 8, 7 and 6 MLthickness, respectively. The measurements have been performed in zero applied field atT = 2 K. Spectra have been scaled so that the area is proportional to the film thickness andhave been off-set vertically for clarity. The dotted lines in the graph indicate the positions atwhich the lines of the substructure are located. The inset shows the raw unsmoothed data.(b) Unsmoothed spectrum of the 10 ML film (circles) plotted together with the fit (solidline) and the four lines constituting the fit (dashed lines).

of equal area A. For the thicker layers the intensity of this fourth line increases withan area of 1

2A for each extra ML Co.We feel that these experimental facts show unambiguously that the different lines

observed indeed correspond to Co ML’s at different distances from the interface.The observable influence of the interfaces on the Co hyperfine field thus extendsmuch further than only to the interface Co layer itself as was found until now byNMR [Riedi99]. The symmetry of the films exactly results in the measured numberand intensity of the sub-lines for all thicknesses, as illustrated in figure 4.5.

With these considerations we can determine the hyperfine field of a Co layeras a function of the distance to a (001)-Cu interface. The results are plotted infigure 4.6(a). The hyperfine fields of Co atoms more than four atomic layers awayfrom the interface do not show resolvable differences any more. The average hyperfinefield of these layers is basically equal to the bulk value of the pseudomorphic Co phase,where the influence of the interface is negligible. Relative to this value the influence of


( a ) ( b )

( i n t e r f a c e l a y e r )

( i n t e r f a c e l a y e r )

l a y e r : 2 ( o r 6 )

7 ML C

oC u ( 0 0 1 ) c r y s t a l

C u t o p l a y e r

l a y e r : 5 ( o r 3 )

l a y e r : 3 ( o r 5 )l a y e r : 4 l a y e r : 6 ( o r 2 )

( i n t e r f a c e l a y e r )l a y e r : 2

Co - f

ilm

C u ( 0 0 1 ) c r y s t a l

l a y e r : 5

l a y e r : 3l a y e r : 4 l a y e r : 6l a y e r : 7

F r e q u e n c y ( M H z ) F r e q u e n c y ( M H z )

59 Co S

pin-Ec

ho Inte

nsity

2 1 0 2 1 3 2 1 6 2 1 9 2 1 0 2 1 3 2 1 6 2 1 9

( c ) ( d )f i r s t 7 M L o f at h i c k C o f i l m

7 M L C o f i l m

2 35 + 6 + 7

2 ( = 6 ) 3 ( = 5 ) 44

a t o m i cl a y e r n o .

Figure 4.5: The relation between a flat Cu(001)/Co film of a certain number of ML andthe resulting spectrum. A single interface (a) results in different resonance lines (apart fromthe interface) for the second, the third and the fourth layer and for the other layers together.A simulated spectrum of the first seven Co ML’s on Cu(001) in a very thick film (which isequivalent to the spectrum of a 14 ML Co film, apart from a factor of two in the intensity)is plotted as an example (c). A Cu-covered 7 ML films has two similar Co/Cu interfaces(b) and thus less resonance lines: since the fifth layer from the bottom is the third layerfrom the top, it has from symmetry reasons the same resonance frequency as the third layer(similarly for the sixth and the second layer) (d).

the interface can be determined. The second and third Co ML from the interface havelower hyperfine fields (-2.0% and -0.8%, respectively), while the fourth ML actuallyhas an enhanced hyperfine field (+0.7%). A very slight shift of the resonance line ofCo atoms more than four ML from the interface when going from 10 ML to 12 MLCo, seems to indicate tiny variations in the hyperfine fields of the fifth and sixth ML(at most +0.05% and -0.2%, respectively).


H y p e r f i n e f i e l d , B h f ( T )

C u ( 0 0 1 )c r y s t a l

Co ato

mic lay

er num

ber87654321

C u t o p l a y e rC u t o p l a y e r

i n t e r f a c e l a y e r i n t e r f a c e l a y e r i n t e r f a c e l a y e r

i n t e r f a c e l a y e ri n t e r f a c e l a y e r

7 M L C o

t h i c kC ol a y e r

6 M L C o

- 2 0 . 8

( a ) ( b )( c )

- 2 1 . 2 - 2 1 . 6 - 2 0 . 8 - 2 1 . 2 - 2 1 . 6 - 2 0 . 8 - 2 1 . 2 - 2 1 . 6

Figure 4.6: Hyperfine fields at T = 2 K versus Co ML position relative to a Co/Cu(001)interface at T=2 K. (a) The influence of a single interface (as measured for at least 14 MLthick Co films). (b+c) Co hyperfine fields within thin Co films (7 and 6 ML respectively),where both interfaces have a significant influence on the central part of the layer. The verticalline at 21.36 T is the hyperfine field of the pure, fct Co-phase, undisturbed by the interfaces.

Since the influence of a single interface reaches at least as far as the fourth ML, thethinnest films (below 8 ML) are influenced by both Co/Cu interfaces and the positiondependence of the hyperfine field differs from that for thicker layers (figure 4.6(b) and(c)). The effect can be well described by a simple additive influence of both interfaces.Thus, the central layer of a 7 ML Co film (figure 4.6(b)) is the fourth ML from bothinterfaces and its hyperfine field is consequently enhanced even more (+1.2%).

The assignment of the different lines to the ML position is largely straightforwardand follows from the Co thickness for which they first appear. However, for thesecond and third layer this scheme does not work, however, since films thinner than6 ML could not be measured accurately enough to provide useful information. Theassignment of these two lines follows from their spin-spin relaxation behaviour, whichis similar for all measured lines, but deviates for the one at 210 MHz, indicating astronger influence from the interface. Additionally, the hyperfine fields of the 6 MLfilm can only be understood from an additive influence from both interfaces, if the210 MHz line is assigned to the second Co atomic layer from the interface.


4.3.4 Origin of the Hyperfine Field Variations

Having determined the hyperfine field variation per ML, we will now assess which ofthe several contributions to the hyperfine field is responsible for the observed varia-tions. The hyperfine field of Co or Fe is commonly decomposed into a core, a dipole,an orbital and a 4s electron field [Guo96]. The first three terms are roughly pro-portional to the spin, dipole and orbital moments of the atom itself, respectively.The influence of an interface on these moments drops off extremely rapidly with dis-tance and is in practice mainly important for the interface layer itself. The spinmoment may show large variations in some systems, but is predicted to be quiteconstant in Co-Cu(001) [Guo96]. The only term having a longer ranged effect is the4s-conduction-electron contribution.

This term is determined by the conduction-electron spin-density which showsFriedel oscillations at some distance to an interface (or impurity). The (spin-dependent)electron scattering at the interface causes a polarization of the conduction electronswhich oscillates as a function of the distance to the interface with a period related tothe extremal vectors of the Fermi surface [Bruno91]. This effect leads to oscillatinginterlayer exchange coupling between magnetic layers separated by a non-magneticmetal and to RKKY-coupling between magnetic impurities [Fert94].

Just as these conduction-electron polarization oscillations in a non-magnetic mate-rial can be measured directly by NMR [Lang74, Jin94], we observe a similar oscillationwithin the ferromagnetic layer itself. Although we can only resolve the hyperfine fieldvariation for the second to fifth Co ML’s (and have an indication of the effect on thesixth), we still can make an estimate of the oscillation period. By fitting the hyperfinefield as a function of the distance from the interface with a sine function divided bythe distance squared, meanwhile taking into account that the second ML from theinterface may still feel some influence by other effects than the conduction-electronpolarization, we find an oscillation period of (3.4± 0.3) ML and a phase of effectivelyzero. The period depends only slightly on the power used for the distance dependenceand does indeed result in a slight decrease (on the correct order of magnitude) of thehyperfine field for the sixth ML from the interface.

Theoretically, the period is determined by the extremal wave vectors of the Fermisurface [Fert94] of, in our case, Co. The predicted and measured period of the inter-layer coupling vs. the Co-thickness in a Co/Cu(001)-multilayer [Bruno93, Bloemen94]of 3.5 ML, which should depend in the same way on the Fermi wave vectors, doesindeed agree surprisingly well with our results. The strength of the effect (0.5 T forthe second ML) is comparable to the induced hyperfine fields in Cu by an Fe interface(0.6 T for the second ML [Jin94, Drittler89]), which is caused by the same mechanismand is expected to be of a similar order of magnitude.

4.3.5 Absence of an Interface Signal

We note that the resonance of the Co interface layer itself is, due to its decreasednumber of Co neighbours, expected in a different part of the spectrum. Since theCo resonance frequency is usually reduced by 16 MHz per direct Cu neighbour,the Co/Cu(001)-interface line is expected at 152 MHz [Riedi99]. Deviating values


59 Co S

pin-Ec

ho Inte

nsity (

arb. un

its)

F r e q u e n c y ( M H z )1 4 0 1 6 0 1 8 0 2 0 0 2 2 0

1 C un e i g h b o u r( 2 0 0 M H z )

b u l k

1 0 x ( 5 . 6 M L C o / 8 M L C u )

2 C un e i g h b o u r s( 1 8 4 M H )

3 C un e i g h b o u r s( 1 6 8 M H z )

4 C un e i g h b o u r s( 1 5 2 M H z )

Figure 4.7: NMR interface spectrum of a [5.6 ML Co/8 ML Cu(001)]10 multilayer. Themeasurement has been performed in zero applied field at T = 2 K. Note that defined peaks arepresent, however, at frequencies clearly different from the usual interface peaks (indicated).

are, however, reported from experiments: Suzuki et al. [Suzuki92] and Thomson etal. [Thomson96] observe quite some intensity related to intermixing at the interfaces,but assign apparent peaks at 112 and at 133 MHz, respectively, to the perfect (001)-interface. Our measurements also show a sharp line in this region (at 120 MHz), butthe observed intensity is simply caused by some sample holder material and is alsopresent without a sample. No interface intensity is measured at all in the entire rangefrom 100 to 240 MHz.

We then tried to find the interface signal in a [5.6 ML Co/8 ML Cu(001)]10 mul-tilayer, grown under the same conditions as the single layers, but could not observeit there either. However, the growth quality was slightly worse and small amountsof non-perfect interfaces were observed, probably due to accumulating roughness to-wards the top of the multilayer stack. The signal corresponded to less than 0.5 ML perrepetition and is clearly too small for the total interface. A remarkable aspect of thissignal is its deviation from the usual additive influence of foreign neighbours [Riedi99],see figure 4.7. The first three lines found, lie at 195, 176 and 151 MHz, instead of at200, 184 and 168 MHz as expected.

The regular (001)-interface signal seems to be unobservable for some reason. Inorder to determine if this also occurs for (001)-interfaces with other materials, Cofilms have been grown with an added 5 ML dusting layer of either Ni or Fe at bothinterfaces. For this thickness, both Ni and Fe grow in a pseudomorphic fcc phase andhave no significant influence on the growth of Co or Cu. The only effect expected is


a marked change in the NMR frequency of the Co interface layer, which is predictedto be 195 MHz for a Co/Ni interface and 240 MHz for Co/Fe [Riedi99].

Although the Co bulk resonance is still very narrow for these samples, confirmingthat the structure is not influenced, again no interface signal could be detected inthe range of 100 to 300 MHz for any of the samples. This seems to indicate thatthe observation of the interface resonance may be problematic for perfect Co(001)-interfaces in general.

A possible cause for this problem is the Co spin-spin relaxation time which isknown to get extremely short for interfaces in several similar systems [Strijkers00].A too fast relaxation makes the observation of any signal in a spin-echo experimentimpossible. The measured relaxation time of the second Co atomic layer is one orderof magnitude shorter than that of the rest of the film, which may indicate the presenceof this effect. The problem can be aggravated by the potentially much larger widthof the interface peak due to the quadrupole splitting at the interface, which may bean order of magnitude larger than that of the bulk [Korecki86], as well as to thestrain relaxation at the interface at larger thicknesses. The chance that the interfacefrequency is reduced to a value lower than 100 MHz is negligible, since for the Co/Niinterface even a strongly reduced frequency should certainly lie in the measured rangeand should have been observed.

Thus, the interface for these systems seems only observable for less than perfectlayers and there are strong indications that Co at a sharp (001) interface is notmeasurable in a straightforward manner. This casts some doubts on the reportedresults in the literature [Suzuki92, Thomson96] and asks for a careful analysis of allexperimental results for (001) interfaces.

In our case, however, we can conclude (although we do not directly observe it) thatthe interface is indeed sharp and not intermixed, since intermixed interfaces do notgive these measurement problems and would have been seen if they had been present.The presence of intermixing at the interfaces would also have made the monolayerresolution of the hyperfine fields impossible.

4.3.6 Discussion

We will now discuss several aspects that have not got full attention yet. Firstly, a fewremarks will be made on Mossbauer spectroscopy and its possibilities. Next, a shortdiscussion of the line width in our films will given and finally some comments on theinfluence of the non-perfect layer-by-layer growth will be made.

Mossbauer Spectroscopy

Mossbauer spectroscopy, used as a local probe technique, has also shown the influ-ence of an interface over several atomic layers in a magnetic material [Ohnishi84,Korecki86]. An oscillating hyperfine field variation for Fe(110) surfaces and a mono-tonously increasing hyperfine field towards an Fe(110)/Ag interface were found. Sincethen the technique has been applied to other Fe interfaces [Liu93b, Zukrowski95]. TheFe hyperfine fields reported show either a monotonic variation or a 2-ML-period os-cillation. Longer periods have not been reported. This may be due to the strong


effect of the varying magnetic moment of Fe, which hides other influences on the hy-perfine field. Moreover, Mossbauer spectroscopy is not suitable for all materials andthe archetypal interlayer coupling system, Co/Cu, cannot be investigated with it.

NMR Line Widths

A critical evaluation of the line widths of various structures shows that the Cu(001)/Cosystem is close to the optimum attainable. Several effects increase the line width andmost are minimal for Co grown epitaxially on a Cu(001) single crystal.

In the first place there is the basic line width for fcc-Co, such as measured onhigh-quality fcc-Co particles [Riedi99], which is 0.7 MHz FWHM.

For non-cubic structures an electric-field gradient is present, which broadens theline by interaction with the Co nuclear quadrupole moment. For Co, the interactioncauses a splitting into seven lines which are too close together for resolution (0.2 MHzper line for fully strained Co on Cu(001)) and are observed as a broadening (seesection 4.4). Adding this effect to the inherent line width, the FWHM of about 2 MHzfor the sub-lines in our thin films is reproduced and also the reduced line width forthick Co layers, which is smaller because of the reduced strain. Since epitaxial filmsare almost always strained, this broadening cannot be avoided.

A third effect is inhomogeneous strain. It causes a frequency shift that varies overthe layer and thus results in a broader line. This effect is absent in our films: sincethe line width is not larger for thick layers, which are only partly strained, the strainmust be homogeneous. For multilayers this effect may be important, providing anargument for the use of the single layers.

A final cause of broadening is the presence of grain boundaries, dislocations orstacking faults in the film. These will always occur in non-epitaxial films and in, forexample, the (111) growth of Co, stacking faults are unavoidable. These effects arethe main cause of the typical line width of 10 MHz for thin fcc-Co films, which isconsidered one of the outstanding features of the system [Riedi99]. The influence ofany concession to structural quality is immediately observable: a reference samplegrown on a 100 nm Cu(001) buffer layer on Si(001) instead of directly on a Cu(001)single-crystal, showed the presence of stacking faults and a line broadening by about50% (see section 4.2).

Thus, the line width in our system, is close to the minimum obtainable for Co,but is also close to the maximum width that will still allow resolution of the sub-lines.Apart from this, the presence of atomically flat interfaces is a prerequisite for theobservation of the effect. This is another argument against the use of multilayers,where cumulative roughness is usually present. The observed effects are unique forsingle thin ultrahigh-quality films and furthermore constitute a test of the ultimateatomic-scale quality of such systems.

Non-Perfect Layer-by-Layer Growth

We will end by briefly discussing the influence of the non-perfect layer-by-layer growth:a nominally 10 ML Co film is exactly 10 ML thick for only about 70% of the area. Theother part consists of roughly 10 nm patches of either 9 or 11 ML thickness. These

4.4. INFLUENCE OF THE STRAIN 85

patches are so large that the atoms at the edges are not seen in the spectra, but theeffect of the patches can be observed nevertheless. Firstly, their presence rules outthe possibility of determining the hyperfine field of the nth monolayer separately bysubtracting the NMR spectrum of an n-ML thick film from that of an (n+1)-ML thickfilm. For perfectly flat films having only a single thickness, this would be the easiestway of obtaining the field of any single monolayer. However, since an n-ML thickfilms also contains significant patches of (n − 1) and (n + 1)-ML thick, the methoddoes not yield useful results and a fit of the complete spectra is the only option.

Secondly, it causes the strong increase in the widths of the sub-lines for filmthicknesses lower than about 8 ML. The origin is the influence of the second interface:for these layers, both interfaces have an effect on the central monolayers and they canno longer be viewed separately. Because of this, the central hyperfine fields of, e.g., a7 ML film are somewhat different from that of a 6 or 8 ML Co film (see figure 4.6(b,c)). Since a 7 ML Co film contains about 15% of either 6 and 8 ML patches, thesub-lines will appear broadened.

4.3.7 Conclusions

In conclusion, we have observed and resolved the effect of an interface on the Cohyperfine field in Co/Cu(001) over a distance of at least four ML. The hyperfinefield directly mimics the Friedel oscillation of the conduction-electron polarizationwithin the Co layer. The oscillation period of (3.4±0.3) ML corresponds to the valueexpected from the extremal wave vectors in the Co minority-spin Fermi surface. Theeffect is resolvable only, thanks to the ultrahigh quality of the epitaxial Co/Cu(001)systems used, which results in ideally narrow NMR lines.

Furthermore, we have found strong indications that a perfect Co(001)-interfacecannot be measured in a straightforward manner by NMR, probably due to a stronglyreduced spin-spin relaxation time.

4.4 Influence of the Strain

4.4.1 Introduction

The Co fct phase is expected to have specific NMR properties that are exclusivelyobservable if the structural quality is comparable to that of bulk fcc cobalt. Havingestablished the presence of a stable and homogeneous fct-phase with a very highstructural quality (section 4.2), the observability of these properties can be tested.We will consecutively look at the direct influence of the strain on the Co hyperfinefield, at the presence of quadrupole splitting due to the expected constant electric-field gradients (EFG’s) in the film, and at a measure of the film quality followingfrom this. Meanwhile we will analyse these features to check if they can provide extrainformation on the structure and strain in the film.

The section is organized as follows: after very briefly repeating a few experimentaldetails, the hyperfine fields of ultrathin Co films as measured by NMR are given (sec-tion 4.4.3). In section 4.4.4 the quadrupole splitting is analysed and in sections 4.4.5


59 Co S

pin-Ec

ho Inte

nsity (

arb. un

its)

H y p e r f i n e F i e l d , B h f ( T )- 2 0 . 8 - 2 1 . 0 - 2 1 . 2 - 2 1 . 4

8 0 M L C o

Bs a m p l e B / /s a m p l e

F W H M = 0 . 1 5 T F W H M = 0 . 1 1 T

Figure 4.8: NMR spectra of a Cu(001) / 80ML Co / Cu film, both for in-plane and forout-of-plane saturation. The measurements have been performed at 2 K. The anisotropyin both hyperfine field and line width is clearly observable. The small deviations from aGaussian line shape have to do with the measurement problems induced by the quadrupolesplitting present.

and 4.4.6 the the electric-field gradients (EFG’s) are determined from the measure-ment and compared with the values expected from the strain, respectively. In thediscussion section the results are compared with the literature and the precise originof the remaining NMR line broadening for these thin films relative to that of bulkfcc-Co is analysed. The section ends with an overview of the conclusions.


Co layers with thicknesses of 10, 20, 40, 80 and 280 ML, MBE-grown on Cu(001)single crystals, were analysed by means of 59Co NMR. The growth parameters canbe found in section 4.2.2. The layers are reasonably, but not perfectly, flat and areknown to be in a high-quality, very homogeneously strained fct(001) phase. Only forthe 280 ML thickness does significant strain relief occur, the other films show hardlyany strain relief (less than 10%) so the Co contains only tiny amounts of stackingfaults as shown in section 4.2.

After the initial measurement, the 80 ML Co film was annealed at 300 C and400 C and measured again in order to have a larger variety of different strains.


4.4.3 Hyperfine Field Anisotropy

As given in sections 2.1.2, 2.3.5, and 4.2, the hyperfine field in the case of anisotropicstrain can be described by Bhf(θ) = Bhf,iso + 1

2Bhf,ani(3 cos2 θ − 1), with Bhf,iso andBhf,ani the isotropic and anisotropic parts of the hyperfine field, respectively, andθ the angle with the symmetry axis, i.e. the film normal in our system. Bhf,iso

is expected to change with the volume strain, while Bhf,ani may be expected to beroughly proportional to the tetragonal deformation c/a.

In order to determine Bhf,ani the samples were measured in a field of severalTesla, applied both in the plane of the sample and perpendicular to the sample. Forthe perpendicular measurements the 1.80 T demagnetization field has to be takeninto account. Resulting spectra for an 80 ML Co film are plotted in figure 4.8. Ananisotropy of Bhf,ani = (+0.20 ± 0.01) T is found, which is not dependent on thethickness for films of 10, 20, 40 and 80 ML. The anisotropy for the 280 ML film andthe 80 ML film after annealing is significantly smaller, but could not be determinedaccurately enough for a quantitative comparison. The maximum anisotropy is aboutthree times smaller than that of hcp Co (Bhf,ani = +0.57 T).

4.4.4 Quadrupole splitting

Line-Width Anisotropy

A non-cubic environment induces electric-field gradients. A nucleus possessing aquadrupole moment (i.e. 59Co) interacts with these field gradients and a splitting ofthe NMR lines occurs [Fekete78]. In most systems this splitting is too small to beobserved directly by a spin-echo NMR experiment and will only lead to a broadeningof the resonance line. This broadening will be anisotropic because the quadrupolesplitting is anisotropic:

νq(ψ) = νq,iso +12νq,ani(3 cos2 ψ − 1), (4.2)

with νq,iso and νq,ani the isotropic and anisotropic parts of the splitting and ψ the anglewith the symmetry axis [Fekete78], in our case the film normal. The isotropic part,the so-called local-ion contribution, is relatively small for Co [Fekete78, Hutchison91]and the anisotropic part which depends on the strain will be the major contribution.

In fcc Co the total EFG is indeed known to be very small and equal to the isotropicpart of the hcp-Co EFG [Hutchison91], but for fct-Co quite significant EFG’s areexpected, dependent on the deviation from cubic symmetry.

In figure 4.8 the resulting difference in line width for the two magnetization di-rections can indeed be observed and is found to be 0.04 T (about 30% of the totalwidth).

Modulated Spin-Echo Intensity Decay

In principle, the quadrupole splitting can be directly estimated from this line-widthanisotropy, but an easier and more accurate method is available. A splitting of aresonance line into several evenly spaced components, which are too close together to


59 Co S

pin-Ec

ho Inte

nsity (

arb. un

its)

D e l a y T i m e , 2 t ( m s )0 1 0 2 0 3 0 4 0

linear i

ntensi

ty scal

e 2 0 M L C o

r e g u l a ru n m o d u l a t e d

d e c a y

Figure 4.9: NMR spin-echo intensity (linear scale) versus the time between the first pulseand the echo for a 20 ML Co film together with the fit and, for comparison, the correspondingunmodulated decay. The measurement is performed in zero applied field at T = 2 K and themagnetization is roughly parallel to the sample plane. The spin-echo decays were measuredapproximately in the centre of the NMR line, where the signal is highest.

be resolvable, can be observed as a modulation of the (otherwise exponential) spin-echo intensity decay [Hahn52, Abe66, Enokiya77, Fekete78]. These modulations canonly be observed if the EFG is homogeneous throughout the entire film, since aninhomogeneous EFG will lead to a distribution of modulation frequencies. In thinfilms the EFG is usually not homogeneous enough and the effect has until now onlybeen observed in bulk materials. For thin films only exponential or doubly exponentialspin-echo decays have been reported up to now [Riedi99].

In figure 4.9 the spin-echo intensity for a 20 ML fct-Co film is plotted as a func-tion of the time 2τ between the first pulse and the echo. The figure shows a clearmodulation of the echo decay (for comparison the corresponding unmodulated decayis also plotted) over more than one decade of intensity. The figure strongly resemblesthe spin-echo modulations observed for bulk hcp Co by Fekete et al. [Fekete78]4, al-though the damping in our case is significantly faster. This result is a direct proofthat our films are of very high quality and more homogeneous than other thin Cofilms measured up to now by NMR.

4Please note that the independent parameter in figure 2 of reference [Fekete78] is wrongly givenas 2τ , while it actually is τ and that a factor π is missing within both harmonic terms in equation(8)


59 Co S

pin-Ec

ho Inte

nsity (

arb. un

its)

D e l a y T i m e , 2 t ( m s )0 1 0 2 0 3 0 4 0

logarit

hmic i

ntensi

ty scal

e

2 0 M L C o

1 s t s p i n e c h o

2 n d s p i n e c h o

3 r d s p i n e c h o

Figure 4.10: NMR spin-echo intensity (logarithmic scale) versus the time between thefirst pulse and the echo for the multiple echoes observed. As expected, clearly differentmodulation frequencies are observed. The measurement is performed in zero applied field atT = 2 K and the magnetization is roughly parallel to the sample plane. The spin-echo decayswere measured approximately in the centre of the NMR line, where the signal is highest.

Fitting of the Modulation

The modulations can be fitted well by a function similar to that used by Fekete etal. [Abe66, Fekete78], consisting of exponentially decaying harmonics on an unmod-ulated background. In fact, we find that we need a few extensions to the formula inorder to get an accurate fit. In the first place, we do observe a third harmonic term(as seen by the small bump at about 2τ = 15µs. In the second place we find that allthree harmonics have slightly different damping times, where the higher harmonicsdamp somewhat faster (as seen by the disappearance of the ‘15µs’-bump in the nextvalley). These different damping times may also be observed in the measurement pre-sented by Fekete et al. [Fekete78], but were neglected at that time. Thirdly and mostsignificantly, we find that our entire signal shows a double exponential decay insteadof a single exponential decay. The double decay is virtually always observed in thinfilms [Riedi99] and in our case both the absolute signal and the modulation strengthhave the same double exponential decay, with a constant intensity ratio between thetwo components and an constant decay time for the fast decay. The origin of thisdouble decay time is unknown, but might lie in the presence of higher harmonics ofthe modulations that damp out very fast.

The resulting function is suitable to fit all curves for all different samples and pa-rameters. Due to the limited number of modulations (fast damping) and the relatively


59 Co S

pin-Ec

ho Inte

nsity (

arb. un

its)

D e l a y T i m e , 2 t ( m s )0 1 0 2 0 3 0 4 0

linear i

ntensi

ty scal

e

4 0 M L C o

B ^ s a m p l e

B / / s a m p l e

Figure 4.11: NMR spin-echo intensity (linear scale) versus the time between the first pulseand the echo for a 40 ML Co film, for both in-plane and out-of-plane saturation. Themeasurements have been performed at T = 2 K. An anisotropy in the modulation frequencyand thus in the quadrupole splitting is observable. The spin-echo decays were measuredapproximately in the centre of the NMR line, where the signal is highest.

large number of parameters, the absolute accuracy of the modulation frequency is notmuch better than several percent (see section 4.4.5). The use of a simplified functionleads to systematical errors in the fit and does not provide a realistic estimate of theaccuracy.

The echo modulation is accompanied by the observation of multiple spin-echoes,which have a modulation behaviour differing from the ‘normal’ first echo as shownin figure 4.10. The second and third spin-echo have a two and three times as highmodulation frequency, respectively, as predicted from theory [Abe66]. Furthermore,as expected, the second echo does not have an unmodulated off-set as seen by thevery narrow and deep valleys (taking the logarithmic intensity scale into account).

The expected anisotropy of the modulation (following from equation (4.2)) isdemonstrated in figure 4.11. Measurements on a 40 ML instead of a 20 ML Cofilm are shown because the signal to noise ratio decreases strongly on applicationof an external field. νq for an in-plane applied field is closely similar to that of thezero-field measurements, indicating that the zero-field magnetization lies almost in-plane and that the NMR signal thus results from Co in domains and not in domainwalls [Riedi99]. The absolute frequency for out-of-plane magnetization is about 50%higher than for an in-plane magnetization. This larger νq corresponds to the largerline width for that orientation (figure 4.8).


Modulations as a Measure of the Structural Quality

The observation of the modulations is a direct proof of the structural quality and thestrength of the modulations can be used as a measure of the structural quality. Byfor example taking from the fit the number of modulations (Nmod) with an amplitudelarger than 10% of the unmodulated signal, this measure can be quantified. Sincethe modulation strength increases with applied pulse power (linearly with 0.19± 0.02modulation per dBm), this measure has to be taken at the optimum RF pulse poweror corrected to that value. Our films have a constant Nmod = 3.0± 0.1. The 280 MLfilm, the annealed films, Co/Cu multilayers, and films with thin Fe dusting layerson both sides have significantly lower values of Nmod, even though in some casesthis difference cannot be seen in the line width or shows up only marginally in theXRD correlation length (section 4.2.4), which demonstrates the sensitivity of NMRto structural inhomogeneities. Nonetheless, the bulk hcp-phase measured by Feketeet al. [Fekete78] has an even better structural quality: Nmod > 7.

Disturbance of the Measurement by the Modulations

Although Nmod is somewhat dependent on the modulation frequency and may, forvery thin films, be influenced by non-local effects from the interfaces, it has theadditional advantage that it is also a measure of the disturbance of the regular NMRmeasurement by the modulations, indicating when care has to be taken with theinterpretation of the exact line width and position. In our case, the presence of themodulations is the direct origin of the relatively low accuracy encountered in theprevious sections.

The reason for this is as follows. As expected, the modulation frequency νq doesnot depend on the applied RF-power in the pulses or on the frequency. Most fitparameters show only slight variations with the NMR frequency or the pulse power.The magnitude of the modulation, however, strongly increases with the applied RFpulse power. This effect severely complicates the regular NMR measurement anddecreases its accuracy. In NMR on ferromagnetic materials both the applied pulsesand the signal are strongly enhanced in the sample. In order to correct for this, theoptimum RF power (at which the spin-echo intensity is maximum) is determined.A low optimum power indicates a large enhancement, so that the signal intensityis overestimated. By correcting for this, measurement points with a high optimumpower are increased relative to those with low optimum power. This procedure worksvery well, if the noise in the system is not too high.

In a standard spin-echo NMR experiment the applied RF pulse power is varied andthe optimum power, at which the spin-echo intensity is maximum, is determined (seesection 2.1.2). The measured intensity is then corrected for the necessary RF power:a low optimum power indicates a high signal enhancement in the sample [Riedi99].By correcting for this, the intensities of measurement points where much RF poweris needed are increased in intensity with respect to those where only little poweris needed. This becomes a problem if a strong noise is present, since this usuallyincreases with the applied power and has even more influence on the final result, dueto the correction procedure.


In these samples a similar effect occurs: the increase of the modulation amplitudewith the applied RF pulse power may affect both the maximum intensity and theoptimum power. Especially if the pulse separation time τ is in a maximum of themodulation both the intensity and the optimum power will be severely overestimatedand the standard correction procedure will further aggravate this. Since, the mod-ulations are strongest in the centre of the NMR line, the height of the line will beoverestimated and, thus, the width will be underestimated. A slight shift of the reso-nance frequency may also occur. The presence of this problem can be easily diagnosedby plotting the optimum RF power versus the frequency, if a sharp peak occurs at theposition of the resonance line, the modulations significantly affect the measurement.Measuring a full decay curve and fitting it for every frequency and RF power is not avery efficient solution, since it will make the measurement inconveniently long. Theeasiest approximate solution is to change the used pulse separation time to a timethat is somewhere in the flank of the modulation.

Empirically, the unaffected optimum RF power can be found from the phase be-haviour of the modulation. At low RF powers the phase of the lowest harmonicis approximately −π for all measured systems. Somewhere around the optimum RFpower this value changes to the phase of 0 found for high RF powers. At the optimumvalue of the RF power the phase is π/2.

However, even with these tricks, the regular NMR measurement stays disturbedand the relatively low accuracy of the exact line width and line position encounteredin the previous sections is an almost unavoidable consequence.

Thus, we have indeed observed a well-defined quadrupole splitting in our films,both by means of an anisotropic broadening of the NMR line and the observation ofspin-echo decay modulations. The behaviour of the modulations is similar to that inhcp Co and multiple echoes and a clear anisotropy are also observed. The modulationscause complications in the regular NMR measurement, but their strength is a usefuland sensitive measure of the structural quality of this type of film.

4.4.5 Electric Field Gradients: Measurement

Now, that the presence of a well-defined and directly measurable electric-field gra-dient in these films has been demonstrated, confirming the presence of the uniquelyhigh-quality fct-Co phase, the information provided by these measurements can bequantitatively analysed. In this and the following sections the quadrupole splittingand corresponding EFG as a function of the strain in the film will be investigated.

The quadrupole splitting in the in-plane direction (in-plane or zero applied field)is constant from 15 to 80 ML Co thickness, as expected because of the closely similarstrains, and is |νq(90)| = (191±3) kHz. Note that the sign of the splitting cannot bedetermined directly. An accurate value of the quadrupole splitting is best determinedfrom a zero-applied-field measurement, which has a large signal-to-noise ratio.

However, because of its anisotropy, the quadrupole splitting has to be determinedin two directions, necessitating the application of large fields. These measurements aremuch more noisy and less accurate, making it much harder to determine trustworthyvalues for each thickness. The damping is somewhat fast for accurately linking themodulation to the strain for the in-field measurements.


However, several general conclusions can be drawn. By applying equation (4.2)the quadrupole splitting can be separated into an isotropic and an anisotropic part.For both hcp and fcc Co the isotropic part is known to equal 12.7% of the totalhcp EFG [Hutchison91], corresponding to a splitting of |νq,iso| = (24.4 ± 0.6) kHz.Although two of the films show strangely deviating values (up to 50 kHz), the othervalues indeed give |νq,iso| = (24 ± 3) kHz. For the anisotropic part, values around|νq,ani| = (300 ± 30) kHz are found, where the large variation is caused by the sametwo measurements that also show the deviating νq,iso.

There are strong indications that the determination of νq(0) (out-of-plane) maybe disturbed by the fast relaxation and the noise (since it has to be measured withan applied field). It shows non-systematic variations for different samples, while thein-plane modulation, which can be measured very accurately without a field, onlyshows gradual, systematic variations with thickness or annealing.

Assuming that indeed the out-of-plane component is disturbed for two of themeasurements, we find |νq,iso| = (24 ± 3) kHz for all samples and |νq,ani| = 340, 334and 328 kHz for the 20, 40 and 80 ML film, respectively, corresponding to a verysmall and gradual strain relief. Annealing of the 80 ML film at 300C and 400Cleads to 270 and 245 kHz, respectively, corresponding to the significant relaxation ofthese films seen in section 4.3.

The anisotropic part of the electric-field gradient follows from the quadrupolesplitting by ν = (3eQVzz)/(2I(2I − 1)h), with Q the quadrupole moment. The328 kHz anisotropic splitting of the 80 ML film, thus corresponds to an EFG of|V ani

zz | = (4.61± 0.06)1018 V/m−2.Thus, although the accuracy of the out-of-plane quadrupole splitting is sometimes

doubtful due to fast relaxation and noise, the expected value of the isotropic split-ting is reproduced. By using this value, the anisotropic splittings can be accuratelydetermined from the zero-field measurements alone.

4.4.6 Electric Field Gradients: Calculation

The measurements can now be compared with the EFG’s originating from the strainin the film.

The EFG at a nucleus arises from the distribution of charge around it. Theanisotropic part of the EFG used to be calculated by viewing the crystal as composedof positive point charges at the lattices sites, together with a uniform compensatingelectron charge and calculating the so-called lattice contribution by direct summa-tion over the lattice sites [Fekete78]. The real anisotropic splitting was found tobe directly proportional to this lattice contribution for a remarkably wide range ofsystems [Raghavan76]. This result is not ideal, since large empirical anti-shieldingfactors are used while, in fact, full electronic calculations should be used [Blaha88].However, lattice summation is still a useful method for determining the influence ofstrain within one specific system.

Direct summation over the lattice sites of a homogeneous fct film reproduces theresults by De Wette [Wette61]. The influence of the finite film thickness can beneglected, since electric fields and their gradients are shielded within a few atomic


c / a

0 . 5 0

0 . 2 5

0 . 0 0

- 0 . 2 5

- 0 . 5 0

V zz (Z

e/4pe 0

a3 )latt

0 . 7 0 . 8 0 . 9 1 . 0 1 . 1 1 . 20 . 6

0 . 7 0 1( b c c )

0 . 8 0 11 . 0 0 0( f c c )f c c l a t t i c e

h c pl a t t i c e

Figure 4.12: The lattice term to the electric-field gradient (EFG), to which the real EFGis expected to be proportional, calculated both for an fcc/fct lattice (solid line) and an hcplattice (dashed line) as a function of the ratio between the out-of-plane and the in-planelattice constant (c/a). The resulting EFG has units V m−2 and Z is the ionic charge. Notethat for hcp a reduced c/a is taken, where c/a = 1 corresponds to unstrained hcp.

distances [Korecki85]. The result is given in figure 4.125.In this case this lattice summation is very suitable, since the values can be di-

rectly compared with the known quadrupole splitting for hcp Co [Fekete78]. Theonly assumption is the equality of the universal correlation constant for fcc and hcpCo [Raghavan76], which is very reasonable since this correlation is indeed univer-sal over most systems. Using this assumption the ratio between the anisotropicquadrupole splitting for fct and hcp Co is equal to the ratio between the latticecontributions for both systems and since the lattice contribution for the fct system isthe only unknown of these four, it can be determined without having to go into thedetails of the, physically somewhat dubious, anti-shielding factors.

The hcp Co anisotropic quadrupole splitting is 170 kHz and the lattice contributionV latt

zz (hcp) = (Ze)/(4πε0a3) · 0.0548 [Fekete78, Hutchison91]. For the 80 ML fct-Co

film the quadrupole splitting is 1.9 times as large and thus V lattzz (fct) = (Ze)/(4πε0a

3)·(0.106±0.002). From the calculations represented in figure 4.12 it follows that (c/a−1) = −4.65 ± 0.3% corresponding to ε‖ = ∆a/a = +2.0 ± 0.1% (equation (4.1)).This result is in excellent agreement with the measured strain (section 4.2.4) of ε‖ =∆a/a = +1.95± 0.1%.

The slight decrease in the quadrupole splitting when increasing the Co thicknessfrom 20 to 80 ML, would correspond to a strain relief of about 3% which is not

5Most publications use CGS units and thus have the EFG, qlatt, in cm−3 and the first fractionas Z/a3.


significant in view of the experimental errors. Thus, there may indeed be some strainrelief in this range [Weber96], but if there is, it is only very minor. The decrease inEFG for the annealed 80 ML films (to 75% and 65% respectively), agrees roughlywith the decrease in strain as seen from NMR line shift and XRD.

Thus, the measurements of the EFG agree well with the measured strains. For Cothicknesses of 12 ML and less, the modulation frequency increases slightly and for verythin films (6 or 7 ML Co), decreases steeply. We attribute these effects to the influenceof the line splitting due to the interfaces [Wieldraaijer04], which becomes observableat the same thicknesses, and to possible influences from the interfaces directly on theEFG [Korecki85]. This is corroborated by the observation of small changes in theEFG as a function of the NMR frequency in a 15 ML Co film. Since the differentfrequencies correspond to different distances from the interface [Wieldraaijer04], thismeans that non-local effects on the EFG are indeed observed.

Thus, we have shown by a very simple method that the strain corresponding toour measured EFG agrees excellently with the strain determined in section 4.2.4. Forextremely thin films we can observe direct influences from the interfaces, however wecannot measure them accurately enough for a quantitative analysis.

4.4.7 Discussion

Several aspects that have not received full attention yet will now be discussed. First,the structural results, the EFG measurement method and the measured EFG will becompared to reports in the literature. After that an analysis will be given of the originof the larger line width in our films compared to that of bulk fcc Co, even though thestructural quality is expected to be comparable.

Comparison with the Literature

Several other methods than spin-echo NMR are used to determine EFG’s. The easiestis continuous-wave NMR which has a high enough resolution to directly resolve thesplitting in one of the orientations in hcp Co [Enokiya77]. Its sensitivity is, however,too low to be applied to thin films.

Most EFG measurements employ radiative detection from oriented radioactivenuclei implanted in a metal matrix, e.g. perturbed angular correlation spectroscopy(PACS), nuclear magnetic resonance on oriented nuclei (NMR-ON) and modulatedadiabatic passage on oriented nuclei (MAPON) [Matthias66, Hagn82, Callaghan88].Especially MAPON is very sensitive even to small EFG’s and has been successfullyused to determine the EFG of Co in unstrained fcc Co [Hutchison91]. It has more-over the advantage that the EFG distribution and sign can be determined directly.However, the implantation of radioactive ions in our thin films may significantly in-fluence their properties and quality. The stability of fct-Co films with respect to ionimplantation, might be a useful object for future MAPON studies.

Hardly any studies have been performed concerning the influence of the strainwithin a single material (as opposed to the many studies concerning hcp metalswith different c/a ratios) and those performed do not present a quantitative anal-ysis [Hoy68, Mercader75]. Thus, we cannot compare our results on the influence of


strain directly with other experimental papers.We can, however, compare the isotropic part of the EFG in our system with

that in other Co systems. Although for specific impurity atoms an EFG depend-ing on the orientation with respect to the different crystallographic axes has beenfound [Seewald97], only an axially symmetric lattice contribution and an isotropiclocal-ion contribution have been observed for Co [Fekete78, Hutchison91]. For fcc Cothe lattice contribution is absent (due to the cubic symmetry) and the observed EFGis equal to the isotropic part in hcp Co. This agrees with the expected additivity oflattice and local ion EFG’s. We find the same absolute value for this isotropic EFG,however, of opposite sign. For hcp Co the isotropic and anisotropic parts have equalsign, for fct Co, we find them to have opposite sign. This sign change can be deriveddirectly by means of the ratio between νq(0) (out of plane) and νq(90) (in plane)and equation 4.2: if the ratio between νq(0) and νq(90) is larger than 2 then νq,iso

has the same sign as νq,ani, while if the ratio is smaller than 2, the signs are opposite.Since the anisotropic lattice contribution has the same sign for hcp Co and fct Co(both have a c/a ratio below the symmetric value, see figure 4.12), this means thatthe local ion contribution has changed sign in this system. Although we can onlyaccurately determine the quadrupole splitting without an applied field, our field mea-surements are still accurate enough to reliably determine these signs. It is not clearwhat causes this effect, but maybe the additivity of lattice and local ion contributionis somewhat less straightforward after all. Further measurements on other systems,together with a comparison with electronic-structure calculations for these systems,are necessary to study the origin of this effect.

NMR Line Widths

The observation of well-defined EFG’s proves that the films are very homogeneous.Such a film should have an NMR line width comparable to that of high-quality bulkmaterial. However, several effects inherent to strained thin films influence the linewidth and have to be taken into account when comparing widths.

As a typical example, the 20 ML Co film is taken, which has a line-width in zero-field of 0.19 T FWHM, where for accuracy reasons the FWHM is determined fromthe width of a Gaussian fit of the entire line. This value is significantly larger thanthe 0.07 T found for bulk fcc films.

Several effects contribute to this line width. In the first place there is the effectthat in zero field the magnetization of the measured lines is close to the in-planedirection but does not seem to be perfectly aligned. When applying an in-plane field(of about 2 T) a line width of about 0.16 T is found, where this is taken as perfectin-plane alignment. This effect is sizeable due to the anisotropy of the hyperfinefield, which causes small distributions of the magnetization direction to result in adistribution of peaks at slightly varying frequencies and thus a broadening.

In the second place there is the effect that the line is actually split into four linesby the direct effect of the interface on the hyperfine field, where the several linescorrespond to different distances from the interface. The main line corresponds tothe centre of the layer, but small side-lines at lower and higher frequency are causedby Co atoms in the second, third and fourth ML from the interface (see section 4.3).


Taking this into account during the fitting of the line, results in a reduced line widthfor the central layers of 0.14 T.

A third effect if a longer-ranged influence of the interfaces, this is no longer re-solvable as a line splitting as for the first four ML’s, but results in small variationsof the hyperfine field as a function of the ML position. The effect is mainly causedby oscillations of the spin polarization of the conduction electrons and drops off veryslowly. An extrapolation of this effect with the expected decay (section 4.3) leadsto variations in the line positions for the various monolayers constituting the centralpeak of up to a few hundredths of a Tesla, resulting in a broadening of about 0.02 T.

The line width of 0.12 T of an fct-Co film, where the direct influence of theinterface is removed, still has line broadening due to the quadrupole splitting. Fittingthe resulting line width seven Gaussian with intensity ratios of 7, 12, 14, 15, 14,12 and 7 as expected from the Clebsch-Gordan coefficients and the known splitting,results in a inherent line width of 0.09 T.

The inherent line width for the thin film structure is thus only slightly larger thanthat of a bulk material, indicating that broadening due to defects, local variations instrain and potentially long range effects of defects and dislocations is almost negligible.On going from this width to the expected line width for the out-of-plane orientation,which is larger due to the larger quadrupole splitting in this direction, one indeedfinds the 0.04 T increase in line width observed in this system.

The line widths in both directions for both thinner and thicker films can be under-stood in a similar way. It can thus be concluded that hardly any broadening due todefects, local strain variations or long range effects of dislocations is present and thatboth the line width and the line width anisotropy can be understood quantitatively.

4.4.8 Conclusions

The influence of the tetragonal Co deformation on the hyperfine field has been directlyobserved by NMR. This has not been possible before, since a well-defined fct phasecan only be stabilized in thin films and these never had the requisite quality up tonow. The deformation results in an anisotropic hyperfine field, an anisotropic linewidth, and a well-defined anisotropic quadrupole splitting which is observable bymodulations of the spin-echo decay. The electric-field gradient in the film, which isresponsible for the quadrupole splitting, agrees well with the value expected from thestrain. The strength of the modulation can be used as a quantitative measure of thefilm quality. For very thin films (thinner than 15 ML) a slight position dependenceof the EFG is observed due to a small direct influence from the interfaces.

It can be concluded that fct Co on Cu(001) is a very well-defined phase with analmost thickness independent strain and structural quality up to at least 80 ML Co.This system in principle provides an extra Co phase (next to hcp, fcc and bcc) thatmay, for example, be used in spintronic model systems for studying the influenceof band structure and electronic properties by varying the phase of the magneticelectrode.

Bibliography

[Abe66] H. Abe, H. Yasuoka, and A. Hirai, “Spin echo modula-tion caused by the quadrupole interaction and multiple spinechoes,” J. Phys. Soc. Jap. 21 (1966), pp. 77–89.

[Abragam61] A. Abragam, The Principles of Nuclear Magnetism (Claren-don, London, 1961).

[Alippi97] P. Alippi, P. M. Marcus, and M. Scheffler, “Strained tetragonalstates and bain paths in metals,” Phys. Rev. Lett. 78 (1997),pp. 3892–5.

[Alphen95] E. A. M. van Alphen, Nuclear Magnetic Resonance Studies ofCo-Based Multilayers, Ph.D. thesis, Eindhoven University ofTechnology (1995).

[Alphen96] E. A. M. van Alphen, A. H. J. Colaris, S. K. J. Lenczowski,H. A. M. de Gronckel, C. Schonenberger, and W. J. M.de Jonge, “Nmr study of the structural properties of electrode-posited Co/Cu multilayers,” J. Magn. Magn. Mat. 156 (1996),pp. 29–30.

[Anderson64] D. H. Anderson and G. A. Samara, “Pressure dependence ofthe 59Co nuclear magnetic resonance in hexagonal cobalt,” J.Appl. Phys 35 (1964), pp. 3043–4.

[Appelbaum69] J. A. Appelbaum and W. F. Brinkman, “Theory of many-bodyeffects in tunneling,” Phys. Rev. 186 (1969), pp. 464–70.

[Baibich88] M. N. Baibich, J. M. Broto, A. Fert, F. Nguyen Van Dau,F. Petroff, P. Eitenne, G. Creuzet, A. Friederich, and J. Chaze-las, “Giant magnetoresistance of (001)Fe/(001)Cr magnetic su-perlattices,” Phys. Rev. Lett. 61 (1988), pp. 2472–5.

[Barthelemy99] A. Barthelemy, A. Fert, and F. Petroff, “Giant magnetore-sistance in magnetic multilayers,” in “Handbook of magneticmaterials, vol. 12,” edited by K. H. J. Buschow (Elsevier, Am-sterdam, 1999), pp. 1–96.

99

100 BIBLIOGRAPHY

[Binasch89] G. Binasch, P. Grunberg, F. Saurenbach, and W. Zinn,“Enhanced magnetoresistance in layered magnetic structureswith antiferromagnetic interlayer exchange,” Phys. Rev. B 39(1989), pp. 4828–30.

[Blaha88] P. Blaha, K. Schwarz, and P. H. Dederichs, “First-principlecalculations of the electric-field gradient in hcp metals,” Phys.Rev. B 37 (1988), pp. 2792–6.

[Bland91] J. A. C. Bland, R. D. Bateson, P. C. Riedi, R. G. Graham, H. J.Lauter, J. Penfold, and C. Shackleton, “Magnetic propertiesof bcc Co films,” J. Appl. Phys. 69 (1991), pp. 4989–91.

[Bloch46] F. Bloch, W. W. Hansen, and M. Packard, “The nuclear in-duction experiment,” Phys. Rev. 70 (1946), pp. 474–485.

[Bloemen93] P. J. H. Bloemen, Metallic Multilayers, Experimental In-vestigation of Magnetic Anisotropy and Magnetic InterlayerCoupling , Ph.D. thesis, Eindhoven University of Technology(1993).

[Bloemen94] P. J. H. Bloemen, M. T. Johnson, M. T. H. van de Vorst,R. Coehoorn, J. J. de Vries, R. Jungblut, J. aan de Stegge,A. Reinders, and W. J. M. de Jonge, “Magnetic layer thick-ness dependence of the interlayer exchange coupling in (001)Co/Cu/Co,” Phys. Rev. Lett. 72 (1994), pp. 764–7.

[Blundell93] S. J. Blundell, M. Gester, J. A. C. Bland, C. Daboo, E. Gu,M. J. Baird, and A. J. R. Ives, “Structure induced magneticanisotropy behavior in Co/GaAs(001) films,” J. Appl. Phys.73 (1993), pp. 5948–50.

[Boeck96] J. de Boeck, W. van Roy, C. Bruynseraede, A. van Esch,H. Bender, C. van Hoof, and G. Borghs, “Ferromagnetic Mn-based thin films with perpendicular magnetization on semicon-ductors,” Physica Scripta T66 (1996), pp. 183–9.

[Brockmann99] M. Brockmann, M. Zolfl, S. Miethaner, and G. Bayreuther,“In-plane volume and interface magnetic anisotropies in epi-taxial Fe films on GaAs(001),” J. Magn. Magn. Mat. 198-199(1999), pp. 384–6.

[Bromer78] H. Bromer and H. L. Huber, “Nuclear magnetic resonance inferromagnetic hcp and fcc 59cobalt,” J. Magn. Magn. Mater.8 (1978), pp. 61–4.

[Bruno91] P. Bruno and C. Chappert, “Oscillatory coupling between fer-romagnetic layers separated by a nonmagnetic metal spacer,”Phys. Rev. Lett. 67 (1991), pp. 1602–5.

BIBLIOGRAPHY 101

[Bruno93] P. Bruno, “Oscillations of interlayer exchange coupling vs.ferromagnetic-layers thickness,” Europhys. Lett. 23 (1993),pp. 615–20.

[Bruynseraede98] G. Bruynseraede, G. Lauhoff, J. A. C. Bland, G. Strijkers,J. de Boeck, and G. Borghs, “Strong anti-ferromagnetic cou-pling in τMnAl/Co perpendicular magnetic superlattices onGaAs,” IEEE Trans. on Magn. 34 (1998), pp. 861–3.

[Burgers34] W. G. Burgers, “On the process of transition of the cubic-body-centered modification into the hexagonal-close-packedmodification of zirconium,” Physica 1 (1934), pp. 561–86.

[Callaghan88] P. T. Callaghan, P. J. Back, and D. H. Chaplin, “Modulatedadiabatic passage of oriented nuclei. I. The theory,” Phys. Rev.B 37 (1988), pp. 4900–10.

[Camarero96] J. Camarero, T. Graf, J. J. de Miguel, R. Miranda, W. Kuch,M. Zharnikov, A. Dittschar, C. M. Schneider, and J. Kirschner,“Surfactant-mediated modification of the magnetic propertiesof Co/Cu(111) thin films and superlattices,” Phys. Rev. Lett.76 (1996), pp. 4428–31.

[Castellani02] F. Castellani, B. van Rossum, A. Diehl, M. Schubert, K. Reh-bein, and H. Oschkinat, “Structure of a protein determinedby solid-state magic-angle-spinning nmr spectroscopy,” Nature420 (2002), pp. 98–102.

[Cerda93] J. R. Cerda, P. L. de Andres, A. Cebollada, R. Miranda,E. Navas, P. Schuster, C. M. Schneider, and J. Kirschner, “Epi-taxial growth of cobalt films on Cu(100): a crystallographicleed determination,” J. Phys.: Condens. Matter 5 (1993),pp. 2055–62.

[Cesari89] C. Cesari, J. P. Faure, G. Nihoul, K. Le Dang, P. Veillet,and D. Renard, “Magnetic properties of ferromagnetic thinsandwiches correlated to their crystallographic structure andinterface morphology: hrem and nmr study,” J. Magn. Magn.Mater 78 (1989), pp. 296–304.

[Clarke87] A. Clarke, G. Jennings, R. F. Willis, P. J. Rous, and J. B.Pendry, “A leed determination of the structure of cobalt over-layers grown on a single-crystal Cu(001) substrate,” Surf. Sci.187 (1987), pp. 327–38.

[Conrad96] E. Conrad, “Diffraction methods,” in “Handbook of SurfaceScience, vol. 1: Physical Structure,” edited by W. N. Unertl(Elsevier, Amsterdam, 1996), pp. 271–360.

102 BIBLIOGRAPHY

[Craik95] D. Craik, Magnetism: Principles and Applications (Wiley,Chichester, 1995).

[Davis01] A. H. Davis, J. M. MacLaren, and P. LeClair, “Inherent tem-perature effects in magnetic tunnel junctions,” J. Appl. Phys.89 (2001), pp. 7567–9.

[Dembczynski92] J. Dembczynski, G. H. Guthohrlein, and E. Stachowska,“Sternheimer free determination of the 59Co nuclearquadrupole moment from hyperfine-structure measurements,”Phys. Rev. A 48 (1992), pp. 2752–61.

[Doi02] M. Doi, B. R. Cuenya, W. Keuen, T. Schmitte, A. Nefedov,H. Zabel, D. Spoddig, R. Meckenstock, and J. Pelzl, “Magneticand structural properties of epitaxial thin films on GaAs(001)and interfaces,” J. Magn. Magn. Mat. 240 (2002), pp. 407–9.

[Drittler89] B. Drittler, H. Ebert, R. Zeller, and P. H. Dederichs, “Abinitio calculations of nmr sattelite data for 3d impurities inCu,” Phys. Rev. B 39 (1989), pp. 6334–41.

[Enokiya77] H. Enokiya, “Nuclear magnetic resonance and nuclear relax-ation in hcp cobalt,” J. Phys. Soc. Jap. 42 (1977), pp. 796–804.

[Fekete78] D. Fekete, H. Boasson, A. Grayevski, V. Zevin, and N. Kaplan,“Anisotropic hyperfine interactions in ferromagnetic hcp Co,”Phys. Rev. B 17 (1978), pp. 347–54.

[Fert94] A. Fert and P. Bruno, “Interlayer coupling and magnetore-sistance in multilayers,” in “Ultrathin Magnetic StructuresII, Measurement Techniques and Novel Magnetic Properties,”edited by B. Heinrich and J. A. C. Bland (Springer-Verlag,Berlin, 1994), pp. 82–116.

[Feynman63] R. P. Feynman, R. B. Leighton, and M. Sands, The Feyn-man Lectures on Physics, volume 2 (Addison-Wesley, Reading,1963).

[Fox99] S. Fox and H. J. F. Jansen, “Total energy of trigonal andtetragonal cobalt,” Phys. Rev. B 60 (1999), pp. 4397–400.

[Gallop91] J. C. Gallop, SQUID’s, the Josephson Effects and Supercon-ducting Electronics (A. Hilger, Bristol, 1991).

[Gazzadi99] G. C. Gazzadi and S. Valeri, “In-depth characterization of thebct-hcp phase transition in Co epitaxial films,” Europhys. Lett.45 (1999), pp. 501–7.

BIBLIOGRAPHY 103

[Gester97] M. Gester, C. Daboo, S. J. Gray, and J. A. C. Bland, “For-mation of pyramid-like structures in the growth of epitax-ial Fe/GaAs(001) films,” J. Magn. Magn. Mat. 165 (1997),pp. 242–5.

[Gordon00] R. A. Gordon, E. D. Crozier, D.-T. Jiang, T. L. Monchesky,and B. Heinrich, “Distorted iron films on GaAs(001)-(4x6),”Phys. Rev. B 62 (2000), pp. 2151–7.

[Gossard59] A. C. Gossard and A. M. Portis, “Observation of nuclear reso-nance in a ferromagnet,” Phys. Rev. Lett. 3 (1959), pp. 164–6.

[Goudeau04] P. Goudeau, P. Villain, T. Girardeau, P.-O. Renault, and K.-F. Badawi, “Elastic constants investigation by x-ray diffractionof in situ deformed magnetic multilayers,” Scripta Mater. 50(2004), pp. 723–7.

[Gronckel] H. A. M. de Gronckel and H. Wieldraaijer, unpublished results.

[Gronckel93] H. A. M. de Gronckel, The Nanostructure of MultilayeredFilms, a NMR Study , Ph.D. thesis, Eindhoven University ofTechnology (1993).

[Gronckel94] H. A. M. de Gronckel, P. J. H. Bloemen, E. A. M. vanAlphen, and W. J. M. de Jonge, “Structure and local mag-netic anisotropy of mbe-grown Co films,” Phys. Rev. B 49(1994), pp. 11327–35.

[Guo96] G. Y. Guo and H. Ebert, “First-principles study of the mag-netic hyperfine field in Fe and Co multilayers,” Phys. Rev. B53 (1996), pp. 2492–503.

[Gustavsson02] F. Gustavsson, E. Nordstrom, V. H. Etgens, M. Eddrief,E. Sjostedt, R. Wappling, and J.-M. George, “Magnetic mo-ment and anisotropy at the Fe/ZnSe(001) interface studied byconversion electron mossbauer spectroscopy,” Phys. Rev. B 66(2002), pp. 024405–1–5.

[Hagn82] E. Hagn, “New method for the determination of the sign andmagnitude of electric quadrupole splittings. I Theoretical con-siderations,” Phys. Rev. B 25 (1982), pp. 1521–8.

[Hahn50] E. L. Hahn, “Spin echoes,” Phys. Rev. 80 (1950), pp. 580–94.

[Hahn52] E. L. Hahn and D. E. Maxwell, “Spin echo measurements ofnuclear spin coupling in molecules,” Phys. Rev. 88 (1952),pp. 1070–84.

104 BIBLIOGRAPHY

[Heckmann94] O. Heckmann, H. Magnan, P. le Fevre, D. Chandesris, and J. J.Rehr, “Crystallographic structure of cobalt films on Cu(001):elastic deformation to a tetragonal structure,” Surf. Sci. 312(1994), pp. 62–72.

[Herman04] M. A. Herman, W. Richter, and H. Sitter, Epitaxy, Physi-cal Principles and Technical Implementation (Springer-Verlag,Berlin, 2004).

[Hirth92] J. P. Hirth and J. Lothe, Theory of Dislocations (Krieger Pub-lishing Company, Malabar, Florida, 1992), second edition.

[Holy99] V. Holy, U. Pietsch, and T. Baumbach, High-Resolution X-RayScattering from Thin Films and Multilayers (Springer Verlag,Berlin, 1999).

[Houdy91] P. Houdy, P. Boher, F. Giron, F. Pierre, C. Chappert,P. Beauvillain, K. Le Dang, P. Veillet, and E. Velu, “Magneticand structural properties of rf-sputtered Co/Fe and Co/Crmultilayers,” J. Appl. Phys. 69 (1991), pp. 5667–9.

[Howe97] J. M. Howe, Interfaces in Materials: Atomic Structure, Ther-modynamics and Kinetics of Solid-Vapor, Solid-Liquid andSolid-Solid Interfaces (John Wiley & Sons, Inc., New York,1997).

[Hoy68] G. R. Hoy and K. P. Singh, “Magnetically induced quadrupoleinteractions in FeCr2S4,” Phys. Rev. 172 (1968), pp. 514–9.

[Hull84] D. Hull and D. J. Bacon, Introduction to Dislocations (Perga-mon Press, Oxford, 1984).

[Hutchison91] W. D. Hutchison, A. V. J. Edge, N. Yazidjoglou, and D. H.Chaplin, “Comparison of electric-field-gradient distributionsbetween ferromagnetic hcp and fcc cobalt,” Phys. Rev. Lett.67 (1991), pp. 3436–9.

[Jansen01] R. Jansen, O. M. J. van ’t Erve, S. D. Kim, R. Vlutters, P. S. A.Kumar, and J. C. Lodder, “The spin-valve transistor: Fabrica-tion, characterization, and physics,” J. Appl. Phys. 89 (2001),pp. 7431–6.

[Jay96a] J. P. Jay, E. Jedryka, M. Wojcik, J. Dekoster, G. Langouche,and P. Panissod, “On the stability of bcc Co in Co/Fe su-perlattices, an nmr and xrd study,” Z. Phys. B 101 (1996),pp. 329–337.

[Jay96b] J. P. Jay, M. Wojcik, and P. Panissod, “Hyperfine field andordering in bcc CoFe bulk alloys studied by 59Co nmr andmonte-carlo simulation,” Z. Phys. B 101 (1996), pp. 471–468.

BIBLIOGRAPHY 105

[Jin94] Q. Y. Jin, Y. B. Xu, H. R. Zhai, C. Hu, M. Lu, Q. S. Bie,Y. Zhai, G. L. Dunifer, R. Naik, and M. Ahmad, “Direct evi-dence of spin polarization oscillations in the Cu layers of Fe/Cumultilayers observed by nmr,” Phys. Rev. Lett. 72 (1994),pp. 768–71.

[Jomni96] S. Jomni, N. Mliki, R. Belhi, K. Abdelmoula, M. Ayadi,and G. Nihoul, “Face centred cubic cobalt ultrathin-layers inAu/Co(111) multilayers: a study by electron diffraction andby hrem,” Thin Solid Films 370 (1996), pp. 186–91.

[Jones60] R. V. Jones and I. P. Kaminow, “Volume dependence of thenmr in ferromagnetic cobalt,” Bull. Am. Phys. Soc. 5 (1960),p. 175, abstract only.

[Jonge94] W. J. M. de Jonge, H. A. M. de Gronckel, and K. Kopinga,“Nuclear magnetic resonance in thin films and multilayers,” in“Ultrathin Magnetic Structures II, Measurement Techniquesand Novel Magnetic Properties,” edited by B. Heinrich andJ. A. C. Bland (Springer-Verlag, Berlin, 1994), pp. 279–296.

[Kalska01] B. Kalska, P. Blomquist, L. Haggstrom, and R. Wappling,“Interface roughness and magnetic moments in a Fe/Co(001)superlattice,” J. Phys. Cond. Matter 13 (2001), pp. 2963–70.

[Kawakami72] M. Kawakami, T. Hihara, Y. Koi, and T. Wakiyama, “TheCo59 nuclear magnetic resonance in hexagonal cobalt,” J.Phys. Soc. Jap. 33 (1972), pp. 1591–8.

[Kim95] S. K. Kim, F. Jona, and P. M. Marcus, “Ultrathin filmsof body-centered-cubic Co on TiAl(010),” Phys. Rev. B 51(1995), pp. 5412–7.

[Kim96] S. K. Kim, C. Petersen, F. Jona, and P. M. Marcus, “Ultrathinfilms of cobalt on Fe001 and the effect of oxygen,” Phys. Rev.B 54 (1996), pp. 2184–90.

[Kim02] T. H. Kim and J. S. Moodera, “Enhanced ferromagnetism andspin-polarized tunneling studies in Co-Mn alloy films,” Phys.Rev. B 66 (2002), pp. 104436–1–5.

[Knechten05] C. A. M. Knechten, Plasma Oxidation for Magnetic TunnelJunctions, Ph.D. thesis, Eindhoven University of Technology(2005).

[Kohlhepp02] J. T. Kohlhepp, G. J. Strijkers, H. Wieldraaijer, and W. J. M.de Jonge, “Crystalline and interfacial structure of ultrathin Colayers grown on Pd(111): a 59Co nmr study,” Phys. Stat. Sol.A 189 (2002), pp. 701–4.

106 BIBLIOGRAPHY

[Korecki85] J. Korecki and U. Gradmann, “In Situ mossbauer analysis ofhyperfine interactions near Fe(110) surfaces and interfaces,”Phys. Rev. Lett. 55 (1985), pp. 2491–4.

[Korecki86] J. Korecki and U. Gradmann, “Spatial oscillation of magnetichyperfine field near the free Fe(110) surface,” Europhys. Lett.2 (1986), pp. 651–7.

[Korringa50] J. Korringa, “Nuclear magnetic relaxation and resonance lineshift in metals,” Physica 16 (1950), pp. 601–10.

[Landolt-Bornstein86] Landolt-Bornstein, editor, Numerical Data and FunctionalRelationships in Science and Technology (Springer-Verlag,Berlin, 1986), new series, group III, vol. 18 and 19 edition.

[Lang74] D. V. Lang, D. C. Lo, J. B. Boyce, and C. P. Slichter, “Mea-surements of electron-spin density near Co atoms in Cu,” Phys.Rev. B 9 (1974), pp. 3077–85.

[LeClair00a] P. LeClair, H. J. M. Swagten, J. T. Kohlhepp, and W. J. M.de Jonge, “Tunnel conductance as a probe of spin polarizationdecay in Cu dusted Co/Al2O3/Co tunnel junctions,” Appl.Phys. Lett. 76 (2000), pp. 3783–5.

[LeClair00b] P. LeClair, H. J. M. Swagten, J. T. Kohlhepp, R. J. M. van deVeerdonk, and W. J. M. de Jonge, “Apparent spin polarizationdecay in Cu-dusted Co/Al2O3/Co tunnel junctions,” Phys.Rev. Lett. 84 (2000), pp. 2933–6.

[LeClair01] P. LeClair, J. T. Kohlhepp, H. J. M. Swagten, and W. J. M.de Jonge, “Interfacial density of states in magnetic tunnel junc-tions,” Phys. Rev. Lett. 86 (2001), pp. 1066–9.

[LeClair02a] 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 structureand density of states effects in Co-based magnetic tunnel junc-tions,” Phys. Rev. Lett. 88 (2002), pp. 107201–1–4.

[LeClair02b] P. R. LeClair, Fundamental Aspects of Spin Polarized Tunnel-ing , Ph.D. thesis, Eindhoven University of Technology (2002).

[Lew00] W. S. Lew, A. Samad, S. P. Li, L. Lopez-Diaz, G. X.Cheng, and J. A. C. Bland, “Magnetic properties of epitax-ial NiFe/Cu/Co spin-valve structures on GaAs(001),” J. Appl.Phys. 87 (2000), pp. 5947–9.

[Li89] H. Li and B. P. Tonner, “Direct experimental identification ofthe structure of ultrathin films of bcc iron and metastable bccand fcc Co,” Phys. Rev. B 40 (1989), pp. 10241–8.

BIBLIOGRAPHY 107

[Li90] H. Li and B. P. Tonner, “Structure and growth mode ofmetastable fcc cobalt ultrathin films on Cu(001) as determinedby angle-resolved x-ray photoemission scattering,” Surf. Sci.237 (1990), pp. 141–152.

[Liu93a] A. Y. Liu and D. J. Singh, “Bcc cobalt: Metastable phase orforced structure,” J. Appl. Phys. 73 (1993), pp. 6189–91.

[Liu93b] G. Liu and U. Gradmann, “Magnetic order near Fe(100) in-terfaces from mossbauer spectroscopy,” J. Magn. Magn. Mat.118 (1993), pp. 99–109.

[Malinowska98] M. Malinowska, C. Meny, E. Jedryka, and P. Panissod, “Theanisotropic first-neighbour contribution to the hyperfine fieldin hexagonal-close-packed Co: a nuclear magnetic resonancestudy of diluted alloys and multilayers,” J. Phys. Cond. Matt.10 (1998), pp. 4919–28.

[Mangan99] M. A. Mangan, G. Spanos, T. Ambrose, and G. A. Prinz,“Transmission electron microscopy investigation of Co thinfilms on GaAs(001),” Appl. Phys. Lett. 75 (1999), pp. 346–8.

[Marcus99] P. M. Marcus, V. L. Moruzzi, and S.-L. Qiu, “Tetragonal equi-librium states of iron,” Phys. Rev. B 60 (1999), pp. 369–72.

[Matthews70] J. W. Matthews, “Accomodation of misfit between single-crystal films of β-cobalt and copper,” Thin Sol. Films 5 (1970),pp. 369–78.

[Matthias66] E. Matthias and R. J. Holliday, “Nuclear magnetic resonancein polarized nuclei,” Phys. Rev. Lett. 17 (1966), pp. 897–9.

[Mercader75] R. C. Mercader and T. E. Cranshaw, “The mossbauer spec-trum of strained iron and the quadrupole coupling in α iron,”J. Phys. F: Metal Phys. 5 (1975), pp. L124–7.

[Michel01] A. Michel, V. Pierron-Bohnes, J. P. Jay, P. Panissod, S. Lefeb-vre, M. Bessiere, H. E. Fischer, and G. V. Tendeloo, “Stabili-sation of fcc cobalt layers by 0.4 nm thick manganese layers inCo/Mn superlattices,” Eur. Phys. J. B 19 (2001), pp. 225–39.

[Min86] B. I. Min, T. Oguchi, and A. J. Freeman, “Structural, mag-netic and electronic properties of Co: Evidence for magnetism-stabilizing structure,” Phys. Rev. B 33 (1986), pp. 7852–4.

[Monsma95] D. J. Monsma, J. C. Lodder, T. J. A. Popma, and B. Dieny,“Perpendicular hot electron spin-valve effect in a new magneticfield sensor: The spin-valve transistor,” Phys. Rev. Lett. 74(1995), pp. 5260–3.

108 BIBLIOGRAPHY

[Moodera95] J. S. Moodera, L. R. Kinder, T. M. Wong, and R. Meservey,“Large magnetoresistance at room temperature in ferromag-netic thin film tunnel junctions,” Phys. Rev. Lett. 74 (1995),pp. 3273–6.

[Moosbuhler02] R. Moosbuhler, F. Bensch, M. Dumm, and G. Bayreuther,“Epitaxial Fe films on GaAs(001): Does the substrate surfacereconstruction affect the uniaxial magnetic anisotropy?” J.Appl. Phys. 91 (2002), pp. 8757–9.

[Muller95] S. Muller, A. Kinne, M. Kottcke, R. Metzler, P. Bayer,L. Hammer, and K. Heinz, “In-plane lattice reconstruction ofCu(001),” Phys. Rev. Lett. 75 (1995), pp. 2859–62.

[Navas93] E. Navas, P. Schuster, C. M. Schneider, J. Kirschner, A. Cebol-lada, C. Ocal, R. Miranda, J. Cerda, and P. de Andres, “Crys-tallography of epitaxial face centered tetragonal Co/Cu(100)by low energy electron diffraction,” J. Magn. Magn. Mater121 (1993), pp. 65–8.

[Nogues99] J. Nogues and I. K. Schuller, “Exchange bias,” J. Magn. Magn.Mat. 192 (1999), pp. 203–32.

[O’Handley00] R. C. O’Handley, Modern Magnetic Materials: Principles andApplications (Wiley, Chichester, 2000).

[Ohnishi84] S. Ohnishi, M. Weinert, and A. J. Freeman, “Interface mag-netism in metal: Ag/Fe(001),” Phys. Rev. B 30 (1984), pp. 36–43.

[Ohring92] M. Ohring, The Materials Science of Thin Films (AcademicPress, London, 1992).

[Oleinik00] I. I. Oleinik, E. Y. Tsymbal, and D. G. Pettifor, “Struc-tural and electronic properties of Co/Al2O3/Co magnetic tun-nel junction from first principles,” Phys. Rev. B 62 (2000),pp. 3952–9.

[Panissod86] P. Panissod, “Nuclear magnetic resonance,” in “Topics inCurrent Physics: Microscopic Models in Physics,” edited byU. Gonser (Springer Verlag, Berlin, 1986), pp. 365–408.

[Panissod97] P. Panissod, “Structural and magnetic investigations of fer-romagnets by nmr. Application to magnetic multilayers.” in“NATO ASI series High Tech, Vol. 48: Frontiers in Magnetismof Reduced Dimension,” edited by V. G. Bar’yakhtar, P. E.Wigen, and N. A. Lesnik (Kluwer Academic, Dordrecht, 1997),pp. 225–70.

BIBLIOGRAPHY 109

[Pound50] R. V. Pound, “Nuclear electric quadrupole interaction in crys-tals,” Phys. Rev. 79 (1950), pp. 685–702.

[Prinz85] G. A. Prinz, “Stabilization of bcc Co via epitaxial growth onGaAs,” Phys. Rev. Lett. 54 (1985), pp. 1051–4.

[Purcell46] E. M. Purcell, H. C. Torrey, and R. V. Pound, “Resonanceabsorption by nuclear magnetic moments in a solid,” Phys.Rev. 69 (1946), pp. 37–38.

[Raghavan76] P. Raghavan, E. N. Kaufmann, R. S. Raghavan, E. J. Ansaldo,and R. A. Naumann, “Sign and magnitude of the quadrupoleinteraction of 111Cd in noncubic metals: Universal correlationof ionic and electronic field gradients,” Phys. Rev. B 13 (1976),pp. 2835–47.

[Riedi87] P. C. Riedi, T. Dumelow, M. Rubinstein, G. A. Prinz, andS. B. Qadri, “Hyperfine-field spectrum of epitaxially grownbcc cobalt,” Phys. Rev. B 36 (1987), pp. 4595–9.

[Riedi99] P. C. Riedi, T. Thomson, and G. J. Tomka, “Nmr of thinmagnetic films and superlattices,” in “Handbook of MagneticMaterials, Vol. 12,” edited by K. H. J. Buschow (Elsevier,Amsterdam, 1999), pp. 97–258.

[Schall04] P. Schall, I. Cohen, D. A. Weitz, and F. Spaepen, “Visualisa-tion of dislocation dynamics in colloidal crystals,” Science 305(2004), pp. 1944–8.

[Schilfgaarde99] M. van Schilfgaarde, I. A. Abrikosov, and B. Johansson, “Ori-gin of the Invar effect in iron-nickel alloys,” Nature 400 (1999),pp. 46–9.

[Schindler00] W. Schindler, T. Koop, A. Kazimirov, G. Scherb, J. Zegen-hagen, T. Schultz, R. Feidenhans’l, and J. Kirschner, “Non-coherent growth patches in pseudomorphic films: Unusualstrain relief in electrodeposited Co on Cu(001),” Surf. Science465 (2000), pp. L783–8.

[Schmid92] A. K. Schmid and J. Kirschner, “In situ observation of epitax-ial growth of Co thin films on Cu(100),” Ultramicroscopy 42(1992), pp. 483–9.

[Schonherr01] H.-P. Schonherr, R. Notzel, W. Ma, and K. H. Ploog, “Evo-lution of the surface morphology of Fe grown on GaAs(100),(311)a and (331)a substrates by molecular beam epitaxy,” J.Appl. Phys. 89 (2001), pp. 169–73.

[Schultz99] W. W. Schultz, P. van Andel, I. Sabelis, and E. Mooyaart,Brit. Med. J. 319 (1999), pp. 1596–1600.

110 BIBLIOGRAPHY

[Seewald97] G. Seewald, E. Hagn, E. Zech, D. Forkel-Wirth, A. Burchard,and ISOLDECollaboration, “Dependence of the electric fieldgradient of Ir in Fe on the direction of magnetization with re-spect to the crystallographic axes,” Phys. Rev. Lett. 78 (1997),pp. 1795–8.

[Singh92] D. J. Singh, “Magnetism in bcc cobalt,” Phys. Rev. B 45(1992), pp. 2258–61.

[Slichter63] C. P. Slichter, Principles of Magnetic Resonance (Springer-Verlag, Berlin, 1963).

[Snyder99] R. L. Snyder, “X-ray diffraction,” in “X-Ray Characterizationof Materials,” edited by E. Lifshin (Wiley-VCH, Weinheim,1999), pp. 1–103.

[Spiridis04] N. Spiridis, T. Slezak, M. Zajac, and J. Korecki, “Ultrathinepitaxial bcc-Co films stabilized on Au(001)-hex,” Surf. Sci-ence 556-568 (2004), pp. 272–7.

[Strijkers99] G. J. Strijkers, Magnetic Nanostructures, An ExperimentalStudy of Structural, Magnetic and Transport Properties, Ph.D.thesis, Eindhoven University of Technology (1999).

[Strijkers00] G. J. Strijkers, J. T. Kohlhepp, H. J. M. Swagten, and W. J. M.de Jonge, “Nuclear magnetic resonance study of epitaxialCo layers on single-crystal substrates,” Appl. Magn. Res. 19(2000), pp. 461–9.

[Suslick91] K. S. Suslick, S.-B. Choe, A. A. Cichowlas, and M. W. Grin-staff, “Sonochemical synthesis of amorphous iron,” Nature 353(1991), pp. 414–6.

[Suzuki92] Y. Suzuki, T. Katayama, and H. Yasuoka, “Nmr study offcc Co/Cu(100) and (111) artificial superlattices,” J. Magn.Magn. Mat. 104-107 (1992), pp. 1843–4.

[Thomson96] T. Thomson, P. Riedi, C. Morawe, and H. Zabel, “59Co nmrinvestigations of sputtered Co/Cu(100) and (111) multilayers,”J. Magn. Magn. Mat. 156 (1996), pp. 89–90.

[Toth63] L. E. Toth and S. F. Ravitz, “Ferromagnetic nuclear resonancein cobalt nuclei in stacking faults and twins,” J. Phys. Chem.Solids 24 (1963), pp. 1203–6.

[Tsymbal03] E. Y. Tsymbal, O. N. Mryasov, and P. R. LeClair, “Spin-dependent tunnelling in magnetic tunnel junctions,” J. Phys.:Condens. Matter 15 (2003), pp. R109–42.

BIBLIOGRAPHY 111

[Unertl96] W. N. Unertl and M. E. Kordesch, “Direct imaging and geo-metrical methods,” in “Handbook of Surface Science, vol. 1:Physical Structure,” edited by W. N. Unertl (Elsevier, Ams-terdam, 1996), pp. 361–421.

[Velev05] J. P. Velev, K. D. Belashchenko, D. A. Stewart, M. van Schilf-gaarde, S. S. Jaswal, and E. Y. Tsymbal, “Negative spin po-larization and large tunneling magnetoresistance in epitaxialCo/SrTiO3/Co magnetic tunnel junctions,” Phys. Rev. Lett.95 (2005), pp. 216601–1–4.

[Vonnegut63] K. Vonnegut, Jr, Cat’s Cradle (1963).

[Vries95] J. J. de Vries, A. A. P. Schudelaro, R. Jungblut, P. J. H. Bloe-men, A. Reinders, J. Kohlhepp, R. Coehoorn, and W. J. M.de Jonge, “Oscillatory interlayer exchange coupling with theCu cap layer thickness in Co/Cu/Co/Cu(100),” Phys. Rev.Lett. 75 (1995), pp. 4306–9.

[Watts03] J. F. Watts and J. Wolstenholme, An Introduction to SurfaceAnalysis by XPS and AES (Wiley, Chichester, 2003).

[Weber96] W. Weber, A. Bischof, R. Allenspach, C. H. Back, J. Fassben-der, U. May, B. Schirmer, R. M. Jungblut, G. Guntherodt,and B. Hillebrands, “Structural relaxation and magneticanisotropy in Co/Cu(001) films,” Phys. Rev. B 54 (1996),pp. 4075–9.

[Wette61] F. W. de Wette, “Electric field gradients in point-ion anduniform-background lattices,” Phys. Rev. 123 (1961), pp. 103–12.

[Wetzel02] P. Wetzel, P. Bertoncini, D. Berling, A. Mehdaoui, B. Loegel,D. Bolmont, G. Gewinner, C. Ulhaq-Bouillet, and V. Pierron-Bohnes, “Growth of ultrathin epitaxial FexCo1−x alloy filmson Si(001): Stabilization of metastable bcc Co,” Surf. Science499 (2002), pp. 210–8.

[Wieldraaijera] H. Wieldraaijer, J. T. Kohlhepp, and W. J. M. de Jonge, un-published results.

[Wieldraaijerb] H. Wieldraaijer and J. Korecki, unpublished results.

[Wieldraaijer03] H. Wieldraaijer, J. T. Kohlhepp, P. LeClair, K. Ha, andW. J. M. de Jonge, “Growth of epitaxial bcc Co(001) elec-trodes for magnetoresistive devices,” Phys. Rev. B 67 (2003),p. 224430.

[Wieldraaijer04] H. Wieldraaijer, W. J. M. de Jonge, and J. T. Kohlhepp,“Monolayer resolved oscillating hyperfine fields in epitaxial fct-Co(001) films,” Phys. Rev. Lett. 93 (2004), pp. 177205–1–4.

112 BIBLIOGRAPHY

[Wieldraaijer05] H. Wieldraaijer, W. J. M. de Jonge, and J. T. Kohlhepp,“Electric-field gradients in thin face-centered-tetragonal Cofilms observed by nuclear magnetic resonance,” Phys. Rev. B72 (2005), pp. 155409/1–10.

[Wolf85] E. L. Wolf, Principles of Electron Tunneling Spectroscopy (Ox-ford University Press, London, 1985), chapter 8.

[Wolf88] D. Wolf and J. F. Lutsko, “Structurally induced supermoduluseffect in superlattices,” Phys. Rev. Lett. 60 (1988), pp. 1170–3.

[Wolf01] S. A. Wolf, D. D. Awschalom, R. A. Buhrman, J. M. Daughton,S. von Molnar, M. L. Roukes, A. Y. Chtchelkanova, and D. M.Treger, “Spintronics: A spin-based electronics vision for thefuture,” Science 294 (2001), pp. 1488–1495.

[Wu98] Y. Z. Wu, H. F. Ding, C. Jing, D. Wu, G. L. Liu, V. Gordon,G. S. Dong, X. F. Jin, S. Zhu, and K. Sun, “In-plane magneticanisotropy of bcc Co on GaAs(001),” Phys. Rev. B 57 (1998),p. 11935.

[Wuttig04] M. Wuttig and X. Liu, Ultrahin Metal Films, Magnetic andStructural Properties (Springer-Verlag, Berlin, 2004).

[Xu87] F. Xu, J. J. Joyce, M. W. Ruckman, H.-W. Chen,F. Boscherini, D. M. Hill, S. A. Chambers, and J. H.Weaver, “Epitaxy, overlayer growth and surface segregationfor Co/GaAs(110) and Co/GaAs(100)-c(8x2),” Phys. Rev. B35 (1987), p. 2375.

[Yuasa00] S. Yuasa, T. Sato, E. Tamura, Y. Suzuki, H. Yamamori,K. Ando, and T. Katayama, “Magnetic tunnel junctionswith single-crystal electrodes: A crystal anisotropy of tunnelmagneto-resistance,” Europhys. Lett. 52 (2000), pp. 344–50.

[Yuasa02] S. Yuasa, T. Nagahama, and Y. Suzuki, “Spin-polarized res-onant tunneling in magnetic tunnel junctions,” Science 297(2002), pp. 234–7.

[Yuasa05] S. Yuasa, T. Katayama, T. Nagahama, A. Fukushima, H. Ku-toba, Y. Suzuki, and K. Ando, “Giant tunneling magnetore-sistance in fully epitaxial body-centered-cubic Co/MgO/Femagnetic tunnel junctions,” Appl. Phys. Lett. 87 (2005),pp. 222508–1–3.

[Zhang97] S. Zhang, P. M. Levy, A. C. Marley, and S. S. P. Parkin,“Quenching of magnetoresistance by hot electrons in magnetictunnel junctions,” Phys. Rev. Lett. 79 (1997), pp. 3744–7.

BIBLIOGRAPHY 113

[Zhang99] S. Zhang and P. M. Levy, “Models for magnetoresistance intunnel junctions,” Eur. Phys. J. B 10 (1999), pp. 599–606.

[Zhang04] X.-G. Zhang and W. H. Butler, “Large magnetoresistancein bcc Co/MgO/Co and FeCo/MgO/FeCo tunnel junctions,”Phys. Rev. B 70 (2004), pp. 172407–1–4.

[Zolfl97] M. Zolfl, M. Brockmann, M. Kohler, S. Kreuzer,T. Schweinbock, S. Miethaner, F. Bensch, and G. Bayreuther,“Magnetic films epitaxially grown on semiconductors,” J.Magn. Magn. Mat. 175 (1997), p. 16.

[Zukrowski95] J. Zukrowski, G. Liu, H. Fritzsche, and U. Gradmann, “Moss-bauer spectroscopy of Cr(110)/Fe(110)/Cr(110) sandwiches,”J. Magn. Magn. Mat. 145 (1995), pp. 57–66.

114 BIBLIOGRAPHY

List of Publications

J. T. Kohlhepp, H. Wieldraaijer, and W. J. M. de Jonge, ”Exchange anisotropy as aprobe of antiferromagnetism in expanded face-centered-tetragonal Mn(001) layers”,submitted to Appl. Phys. Lett.

H. Wieldraaijer, W. J. M. de Jonge, and J. T. Kohlhepp, ”Electric-field gradientsin thin face-centered-tetragonal Co films observed by nuclear magnetic resonance”,Phys. Rev. B 72 (2005), pp. 155409/1-10

H. Wieldraaijer, W. J. M. de Jonge, and J. T. Kohlhepp, ”59Co NMR observation ofmonolayer resolved hyperfine fields in ultrathin epitaxial fct-Co(001) films”, J. Magn.Magn. Mater 286 (2005), pp. 390-3

H. Wieldraaijer, W. J. M. de Jonge, and J. T. Kohlhepp, ”Monolayer resolved os-cillating hyperfine fields in epitaxial face-centered-tetragonal Co(001) films”, Phys.Rev. Lett 93 (2004), pp. 177205/1-4

H. Wieldraaijer, J. T. Kohlhepp, P. LeClair, K. Ha, and W.J.M. de Jonge, ”Growthof Epitaxial bcc-Co(001) Electrodes for Magnetoresistive Devices”, Phys. Rev. B 67(2003), pp. 224430/1-10

H. Wieldraaijer, P. LeClair, J. T. Kohlhepp, H. J. M. Swagten, and W. J. M. de Jonge,”Influence of electrode structure on magnetotransport in magnetic tunnel junctions”,IEEE Transactions on Magnetics 38(5) (2002), pp. 2727-9

P. Evans, C. Scheck, W. J. M. de Jonge, J. T. Kohlhepp, T. Isaac-Smith, H. Wieldraaijer,J. Williams, R. Schad, and G. Zangari, ”Epitaxial growth and magnetic anisotropyof electrodeposited Ni and Co thin films grown on n-type GaAs”, IEEE Transactionson Magnetics 38(5) (2002), pp. 2670-2

J. T. Kohlhepp, G. J. Strijkers, H. Wieldraaijer, and W. J. M. de Jonge, ”Crys-talline and interfacial structure of ultrathin Co layers grown on Pd(111): a 59CoNMR study”, Physica Stat. Sol. A 189(3) (2002), pp. 701-4

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, ”Bandstructure and density of states effects in Co-based magnetic tunnel junctions”, Phys.Rev. Lett 88 (2002), pp. 107201/1-4

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/Al2O3/Co magnetictunnel junctions”, Phys. Rev. B 64(10) (2001), pp. 100406/1-4

Abstract

Summary

Thin magnetic-metal layers, such as Co films, are of interest among other things dueto their applicability in devices combining electronic and magnetic properties. Thesestructures are generally known as magnetoelectronics or spintronics. An example isa device that has an electrical resistance that is influenced by its magnetic state. Itcan be switched between different magnetic states by the application of a magneticfield and in this way information can be stored or magnetic fields measured. Thistype of structure is called a magnetoresistive device. Examples are magnetic tunneljunctions (MTJ’s) and GMR structures (giant magnetoresistance structures). Bothconsist of thin magnetic-metal films, the electrodes, that are separated either by anultrathin insulator (the barrier in an MTJ) or by non-magnetic-metal films (in GMRstructures).

The precise properties of these devices are largely determined by the magnetic andelectronic properties of the electrodes, which in turn depend mainly on the physicalstructure of these thin magnetic-metal films. Thus, knowledge of the physical (atomic)structure is imperative, both for optimizing the properties of the device and for gaininginsight in the physical principles behind it.

In this thesis the results of a study of the physical structure of thin Co films bymeans of 59Co nuclear magnetic resonance (NMR) are presented. Co films are usedbecause they offer a variety of physical structures as they have more than one stablecrystal phase. The NMR technique is used since it is a powerful tool for analysing thelocal surroundings of Co atoms. It is suitable for measuring Co atoms buried inside astructure of several layers and does not depend on the existence of long-range atomicorder in the films.

Several studies have been performed. On the one hand, we have grown and charac-terized various thin Co films in different crystal phases that may be used as electrodesin magnetoresistive devices (chapter 3). On the other hand, while thin (Co) films areused in magnetoresistive structures, the structural properties are often derived frombulk Co. For this reason we have also investigated the inherent differences, both in thephysical structure and the NMR properties, between thin films and bulk Co causedby the finite thickness of the thin films (chapter 4).

115

116 ABSTRACT

Chapter 3: Co Structures for Magnetoresistive Devices

Firstly, as presented in section 3.2, electrodes have been grown by means of sputterdeposition on three commonly used substrates: Ta, Al2O3, and FeMn. While the firsttwo show a polycrystalline, mixed-phase Co structure, we find the third to consistof oriented Co(111) that grows almost purely in the fcc phase for some thicknessranges. The difference in electronic structure associated with this difference in physicalstructure can be shown to be the origin of an observed asymmetry in the behaviourof an MTJ with an FeMn-based bottom electrode.

Secondly, as presented in section 3.3, we have shown that the bcc-Co phase canbe used in MTJ’s, even though it is unstable, but that it can not be straightforwardlyimplemented in standard GMR structures. It is known that in thin films Co can bestabilized in the bcc phase by means of a suitable substrate and growth conditions.For use in magnetoresistive devices, however, the bcc-Co film has to be also stablewhen covered by additional layers that do not necessarily match with the bcc-Cophase.

Our bcc-Co films, grown by molecular beam epitaxy (MBE) on Fe(001), withthicknesses up to 2 nm are shown to be stable under coverage either with Fe orAl2O3. This implies that bcc Co can be used as a bottom electrode in an MTJ aswas confirmed by the growth of a few working bcc-Co-based MTJ’s. The junctionswere, however, not developed far enough for an extensive analysis. On the otherhand coverage with Cu was shown to transform the film into a disordered mixture ofbcc, fcc and hcp Co. Since Cu coverage is common in GMR-structures, this resultindicates that bcc Co can not straightforwardly be applied in these structures.

Chapter 4: Epitaxial Co Films on Cu(001)

An extensive study was performed in order to analyse the influence of the finitefilm thickness on its structure and its NMR properties. The first step was to finda system suitable for this kind of study. Generally, thin films have a much lowerstructural quality on an atomic scale than many well-grown bulk materials, whichmakes it impossible to compare them. Therefore, a thin-film structure is needed thatgrows in an almost ideal fashion without the usual dislocations, stacking faults andinhomogeneities.

We have found that MBE-grown Cu(001)/Co/Cu with a thickness between 4 and280 atomic layers (ML) provides a suitable system. Structural analysis on the films bymeans of NMR, X-ray diffraction (XRD), and low energy electron diffraction (LEED)shows them to have a structural ordering quality comparable to that of the best bulkfcc Co, but to be in a phase that is typical for thin films. This phase is a tetragonallydeformed (strained) variety of fcc Co, called fct Co. The magnitude of the strain isfound to be almost independent of the film thickness up to at least 80 ML of Co.

For thicker films a significant strain relief may occur, which is found to influencethe elastic properties of the film. Growth on a thin deposited Cu film, instead ofdirectly on a Cu single crystal, also resulted in strain relief and a strongly reducedquality. Strain relief during film growth seems to be one of the major contributionsto the relatively low quality of thin films and probably only fully strained films can

ENGLISH 117

be of high structural order.On further analysis of this system, we observed a direct influence of the interfaces

on the electronic properties of the Co atoms that was not reported before. Further-more, we were able to make an overview of the effects of the strain on the NMRproperties of the Co film.

Section 4.3: Influence of the Interfaces Our NMR spectra of Co films withthicknesses lower than 20 ML showed some unexpected features. These were foundto stem from a variation of the Co hyperfine field with the distance of the Co atomfrom an interface. The effect could be resolved over a distance of at least four ML,while usually only for the Co layer directly at the interface a deviating hyperfine fieldis found. The effect is found to be unrelated to the strain in the film, but to originatepurely in the presence of abrupt interfaces.

The modulation of the hyperfine field as a function of the distance can be explainedby the Friedel oscillation of the polarization of the conduction electrons within theCo layer. The oscillation period of (3.4± 0.3) ML corresponds to the value expectedfrom the extremal wave vectors in the Co minority-spin Fermi surface. We have thusfound a novel way of observing these polarization oscillations.

Section 4.4: Influence of the Strain We can observe the tetragonal deformationof the Co in several ways by NMR: 1) the hyperfine field is anisotropic, 2) the linewidth is anisotropic, and 3) a well-defined quadrupole splitting is observable as amodulation of the NMR spin-echo decay. The quadrupole splitting is caused byelectric-field gradients (EFG’s) in the films and is only measurable by NMR if theEFG’s are very homogeneous throughout the material. Previously, the effect hadonly been reported for bulk material.

Thus, the observation of quadrupole splitting is a measure of the structural quality.We can use the strength of the corresponding modulation as a measure of the orderin this kind of system. Information on the strain in the film may be obtained fromthe EFG. Comparison with the known strain in our films by means of a very simplemodel provides a good agreement. On the other hand, the quadrupole splitting alsobroadens the NMR line and complicates the measurements through the spin-echomodulations it causes.

As a result the NMR line width can not be used as a straightforward measureof the film quality, as it is in many thin-film structures, but is complicated by thequadrupole splitting, by the direct influences of the interfaces, and by the anisotropyof the hyperfine field. We find the line width for our Co films, after correction for theseeffects, to be comparable to that of the best bulk fcc Co, confirming the comparablequality.

118 ABSTRACT

Samenvatting

Dunne magnetische metaallagen, zoals kobalt (Co)-lagen, zijn interessant vanwege huntoepasbaarheid in structuren die elektronische en magnetische eigenschappen combi-neren. Deze structuren staan bekend als magneto-elektronica of spintronica. Eenvoorbeeld is een structuur waarvan de elektrische weerstand afhangt van de magne-tische toestand. Door middel van een magnetisch veld kan er tussen de toestandengeschakeld worden. Hiermee kan informatie worden opgeslagen of kunnen magneti-sche velden worden gemeten. Een dergelijke structuur wordt een magnetoweerstands-structuur genoemd. Voorbeelden daarvan zijn, zogenaamde magnetische tunneljunc-ties (MTJ’s) en GMR-structuren (reuze-magnetoweerstandsstructuren). Beide typesbestaan uit dunne, magnetische metaallagen, die de elektroden worden genoemd,met daartussenin ofwel een zeer dunne isolator (de barriere in een MTJ) ofwel niet-magnetische metaallagen (in GMR-structuren).

The precieze eigenschappen van dit soort structuren worden voor een groot deeldoor de magnetische en elektronische eigenschappen van de elektrodes bepaald. Dezehangen op hun beurt sterk af van de fysieke structuur van deze magnetische metaal-lagen. Dat maakt kennis van de fysieke (lees: atomaire) structuur van het grootstebelang, zowel voor het optimaliseren van de eigenschappen als om inzicht te verwervenin de fysica achter die eigenschappen.

In dit proefschrift wordt een studie gepresenteerd naar de de structuur van dunnekobaltlagen met behulp van 59Co kernspinresonantie (NMR). We gebruiken Co la-gen, omdat daar een relatief grote variatie aan atomaire structuren mogelijk is: Coheeft meerder stabiele kristalstructuren. Kernspinresonantie is een ideale techniekom dit soort structuren op een extreem lokale (atomaire) schaal te onderzoeken. Ookkobalt-atomen die zich binnen in een meerlaagse structuur bevinden kunnen wordenonderzocht en een regelmatige ordening van de atomen door de hele film heen is nietnoodzakelijk.

We presenteren hier verschillende studies. Een soort studie is het groeien en on-derzoeken van verschillende Co-kristalstructuren die zouden kunnen worden gebruiktom de de invloed van de elektronische structuur op de eigenschappen van MTJ’s ofGMR-structuren te bepalen (hoofdstuk 3). Van de andere kant hebben we ook onder-zocht wat de intrinsieke verschillen zijn tussen dunne lagen en bulk Co, wat betreftde fysieke structuur en de NMR eigenschappen, veroorzaakt door de eindige filmdikte(hoofdstuk 4). Dit laatste is van belang, omdat in magnetoweerstandsstructuren al-tijd dunne Co lagen worden gebruikt, terwijl de fysieke structuur en de eigenschappendaarvan doorgaans worden beschreven als een Co bulk-kristalfase.

Hoofdstuk 3: Geschikte kobalt-fasen voor magnetoweerstandsstructuren

Als eerste is een analyse van Co onder-elektroden voor MTJ’s uitgevoerd (zie sec-tie 3.2). De elektroden zijn gegroeid met behulp van sputter depositie op drie veelvoorkomende substraten: tantaal (Ta), aluminiumoxide (Al2O3) en ijzermanganide(FeMn). Onze analyse toonde aan dat het kobalt op de eerste twee substraten be-staat uit diverse ‘korrels’ met verschillende kristalfasen in verschillende orientaties,maar dat Co op FeMn in een georienteerde fase groeit met de dichtgestapelde (111)-

DUTCH 119

vlakken parallel aan de laag. Voor bepaalde laagdiktes vindt de groei van kobalt opdit substraat vrijwel uitsluitend in de zogenaamd fcc-fase plaats. Het kan wordenaangetoond dat het verschil in elektronische structuur dat hiermee gepaard gaat eenasymmetry in het gedrag van de MTJ veroorzaakt.

Vervolgens hebben we aangetoond dat de Co bcc-fase, ondanks zijn instabiliteit,geschikt is voor gebruik in MTJ’s, maar dat het niet in de meest gebruikte GMR-structuren kan worden toegepast (zie sectie 3.3). Het was bekend dat bcc-Co doorgebruik te maken van geschikte substraten en groeimethoden voor beperkte dikteskan worden gestabiliseerd. Om in magnetoweerstandsstructuren te kunnen wordentoegepast, moet de bcc-Co laag echter ook stabiel blijven wanneer hij wordt bedektmet een volgende laag die misschien helemaal niet stabiliserend werkt op de bcc-fase.

Onze bcc-structuren, gegroeid met moleculair bundel epitaxy (MBE) op Fe(001)en met diktes tot maximaal 2.0 nm, blijken stabiel te zijn als ze bedekt worden metijzer of met Al2O3. Deze stabiliteit maakt het in principe mogelijk om MTJ’s temaken met bcc-Co onder-elektroden, hetgeen inderdaad is aangetoond met behulpvan een eerste werkende MTJ op basis van bcc-Co. De junctie is echter niet vergenoeg doorontwikkeld voor een uitgebreide analyse. Er is ook geprobeerd om bcc-Comet koper te bedekken, maar dit bleek de films onmiddellijk te transformeren in eenongeordende mix van verschillende fasen. Aangezien bedekking met koper gebruikelijkis in GMR-structuren, betekent dit dat een standaard op bcc-Co gebaseerde GMR-structuur niet eenvoudigweg realiseerbaar is.

Hoofdstuk 4: Epitaxiale kobaltlagen op Cu(001)

We hebben een uitgebreide studie gedaan naar de invloed van de eindige laagdikte opde structuur van de laag en op zijn NMR eigenschappen. Als eerste is er een systeemgezocht dat geschikt is voor dit soort studie. Normaliter heeft de structuur van dunnelagen een veel lagere ordeningskwaliteit dan veel bulk-materialen waardoor het ergmoeilijk is om ze te vergelijken. Wat er dus nodig is, is een dunne laag die op bijnaideale wijze gegroeid kan worden, dat wil zeggen zonder de gebruikelijke dislokaties,stapelfouten en inhomogeniteiten.

Met MBE gegroeide lagen van Co en Cu op een Cu(001)-kristal bleken geschiktte zijn. Structuuranalyse met behulp van NMR, Rontgendiffractie (XRD) en lage-energie-elektrondiffractie (LEED) aan Co-lagen met diktes tot 80 atoomlagen (ML)heeft aangetoond dat ze een kwaliteit van ordening vertonen die vergelijkbaar is methet beste bulk kobalt in de fcc-fase, terwijl hun kristalfase typisch is voor dunnefilms. De kristalstructuur is een vierhoekig vervormde fcc-fase (met een balkvormigein plaats van een kubische eenheidscel), fct genaamd. Voor lagen van maximaal 80 MLdik blijkt de grootte van de vervorming vrijwel onafhankelijk van de laagdikte te zijn.

Voor dikkere lagen treedt er een duidelijke vermindering van de deformatie op, diegepaard blijkt te gaan met een verandering van de elastische eigenschappen van delaag. Ook als de lagen niet op een een-kristal, maar op een gedeponeerde koperlaagworden gegroeid, blijkt de vervorming af te nemen met toenemende dikte, hetgeenleidt tot een sterk verminderde structurele kwaliteit. Een vermindering van de ver-vorming gedurende de laaggroei lijkt een van de voornaamste oorzaken van de relatieflage structurele kwaliteit van dunne lagen te zijn. Waarschijnlijk kunnen alleen films

120 ABSTRACT

die hun initiele vervorming volledig behouden van zeer hoge ordeningskwaliteit zijn.Door verder onderzoek aan deze lagen hebben we een directe invloed van de grens-

vlakken op de elektronische eigenschappen van de kobalt-atomen waargenomen dienog niet eerder was gepubliceerd. Daarnaast konden we een overzicht van de gevolgenvan de vervorming van de laag op de NMR eigenschappen maken.

Sectie 4.3: Invloed van de grensvlakken In onze NMR-spectra van kobaltlagenmet diktes kleiner dan 20 ML namen we onverwachte karakteristieken waar. Deoorzaak hiervan bleek een variatie van het Co-hyperfijnveld te zijn als functie vande afstand van het kobalt-atoom tot een grensvlak. De variatie kon worden bepaaldvoor zeker vier atoomlagen naast het grensvlak terwijl normaliter alleen bij de laagdirect aan het grensvlak een afwijking waarneembaar is. Dit effect heeft niets metde vervorming van de kobalt-structuur in de lagen te maken, maar wordt volledigveroorzaakt door de aanwezigheid van abrupte grensvlakken.

Het hyperfijnveld vertoont een modulatie als functie van de afstand tot het grens-vlak die verklaard kan worden door de Friedel oscillatie van de polarisatie van degeleidingselektronen in de kobaltlaag. De periode van de oscillatie is (3.4± 0.3) ML,wat overeenkomt met de waarden die volgen uit een analyse van de uiterste golfvec-toren van het minderheidsspin-Fermi-oppervlak van kobalt.

Sectie 4.4: Invloed van de vervorming We kunnen verschillende invloeden vande vierhoekige vervorming van het kobalt op het NMR spectrum waarnemen: 1) hethyperfijnveld is anisotroop, 2) de lijnbreedte is anisotroop en 3) er is een goed gedefini-eerde kwadrupool-opsplitsing aanwezig. De kwadrupool-opsplitsing kan waargenomenworden doordat de NMR spin-echo-intensiteit niet exponentieel afneemt, maar gesu-perponeerd op die afname een modulatie vertoont. Dit soort kwadrupool-opsplitsingwordt veroorzaakt door elektrisch-veldgradienten in de laag en kan alleen worden ge-meten met NMR als de gradient bijzonder homogeen is door de hele laag heen. Dezevoorwaarde is er de oorzaak van dat de opsplitsing nog niet eerder voor dunne filmsis gemeld.

Dit maakt de waarneming van kwadrupool-opsplitsing tot een duidelijk teken vaneen hoge structurele kwaliteit van de laag. De sterkte van de bijbehorende modulatiekan als maat voor de kwaliteit van het systeem worden gebruikt bij dit soort lagen.Uit de waarde van de elektrisch-veldgradient kan informatie over de vervorming vanhet kristalrooster worden verkregen. Een vergelijking van de hier, met behulp van eeneenvoudig model, uit volgende waarden met de al bekende vervorming levert een goedeovereenkomst op. Van de andere kant wordt de NMR-meting ook bemoeilijkt doorde aanwezigheid van de kwadrupool-opsplitsing door de daaruit volgende modulatiesen verbreding van de NMR-lijn.

De NMR-lijnbreedte is dus niet, zoals gewoonlijk, een directe maat voor de kwa-liteit van de laag, maar wordt beınvloed door de kwadrupool-opsplitsing, de directeinvloed van de grensvlakken en de anisotropie van het hyperfijnveld. De lijnbreedtevoor onze kobaltlagen blijkt, na correctie voor de invloed van deze factoren, ver-gelijkbaar te zijn met die van het beste bulk fcc-kobalt, hetgeen de vergelijkbareordeningskwaliteit bevestigt.

121

Curriculum Vitae

Harm Wieldraaijer

February 6, 1975 Born in Eindhoven, the Netherlands

1987 - 1993 Attended Bernardinus College, Heerlengymnasium

1993 - 1999 Attended Eindhoven University of TechnologyM.Sc. in Applied Physics

Internal trainee-ship in the Theoretical Physics GroupTitle: Exchange- and correlation-energy of a hole gas withvalence band coupling

External trainee-ship at Oce-Nederland B.V., VenloTitle: Meten van de oplosbaarheid van stikstof in vloeistoffenmet behulp van gas-chromatografie

Graduate research in the Theoretical and ExperimentalAtomic Physics and Quantum Electronics GroupTitle: Design of a seeded supersonic iron source

1999 - 2006 Ph.D. candidateEindhoven University of Technology, Department of AppliedPhysics.Research carried out in the group Physics of NanostructuresThesis title: Ultrathin Co Films for Magnetoresistive Devices:an NMR study

2004 - Attending teacher training college TULO, Eindhoven

2005 - Physics teacher at secondary school B.C. Broekhin, Roermond

122

Dankwoord

Bedankt6 allemaal7!!!

6Het uitvoeren van een promotie-onderzoek en het schrijven van een proefschrift doe je nooitalleen, er is altijd een grote groep mensen die gedurende die periode van (vele) jaren van grootbelang is ofwel voor de voortgang van het onderzoek ofwel voor de geestelijke stabiliteit van depromovendus. Het is niet meer dan redelijk om de laatste woorden van een lange periode aan dezemensen te wijden

7“allemaal” = iedereen die (direct of indirect) een (positieve) bijdrage heeft geleverd aan detotstandkoming van dit proefschrift. Het grootste nadeel van veel te lang over je promotie doenis echter dat het niet meer mogelijk is om iedereen die heeft bijgedragen met name te bedanken.Daarom volgt hieronder slechts een korte, en zeker niet uitputtende selectie, van de mensen die vandirect belang zijn geweest.

In de eerste plaats natuurlijk Wim, mijn promotor. Bedankt voor het promotorschap, de velegevarieerde discussies, het kritische lezen van artikelen en proefschrift, je vergevingsgezindheid voorhet maar niet volwassen willen worden van sommige van je ‘pupillen’, het redelijk verborgen houdenvan je langzaam toenemende wanhoop over het niet aflatende ‘speelgedrag’ en met name voor jegeduld met trage danwel luie promovendi (waarvan ik experimenteel heb kunnen aantonen dat heterg groot, maar toch eindig is). Daarnaast natuurlijk ook bedankt voor de les in hoe de hardewereld in elkaar zit door middel van beloftes over grof geld voor het publiceren in toonaangevendetijdschriften die slechts ter motivatie bedoeld bleken. ;-)

Meteen daarna Jurgen, mijn directe begeleider. Bedankt voor alle motiverend bedoelde kreten envoor subtiel ”magic-finger reparaties”van onwillige opstellingen. Daarnaast natuurlijk ook voor zeerveel kennis van zaken over het groeien van lagen en voor de uiteindelijk goede samenwerking.

Gerrie, Jef en Hans wil ik bedanken voor de technische ondersteuning en daarnaast voor, respec-tievelijk, het oeverloze geouwehoer, de introductie tot Helmond Sport en voetbalvandalisme en hetenthousiasme over allerlei (nogal alternatieve) fysica. Verder natuurlijk de stagiairs Ward, Rudi enManuela voor hun bijdragen en het heliumbedrijf (Wil, Jos en Nando) voor duizenden liters vloeibaarhelium op de soms meest onmogelijke momenten.

Mensen die mij verder inhoudelijk geholpen helpen en die ik daarvoor wil bedanken, zijn: Gustav,die me de kneepjes van het NMR-vak leerde (en ook veel verteld heeft over de half-mythischeMossbauer opstelling die volgens sommige geruchten ergens in de kelder zou moeten staan en ookonder mijn verantwoordelijkheid scheen te vallen). Bart die met name in het begin van mijn promo-tie altijd tijd had om over ieder stukje fysica te discussieren en die daarnaast is blijven bijdragen inde vorm van gezelligheid en veel hulp bij het maken van de kaft van dit proefschrift. En daarnaastnatuurlijk Daniel voor de verregaande ondersteuning in de begeleiding van stagiaires! ;-)

Naast de al genoemde personen zijn er ook veel anderen die ik moet bedanken voor het maken vande sfeer in de groep, met name: Maarten en Floor (de twee andere ‘witte muizen’ van weleer: hetroemruchte samenwerkingsverband om begeleiders e.d. tot wanhoop te drijven) en Floor natuurlijkook voor heel veel gezelligheid, zowel voor, tijdens als na je verblijf in de groep, voor zwarte-pietspelen, voor begeleidertje plagen, voor..., voor..., teveel om op te noemen. En daarnaast ook nogKarel, Csaba, Coen 1 en Coen 2, Paresh, Henk, Bert en de Belgen Bart en Koen en vele, vele anderenin de loop van de jaren die ik helaas niet allemaal kan noemen.

Tot slot wil ik nog kort een paar mensen van buiten de groep noemen die toch ook indirect hebbenbijgedragen aan mijn promotie: mijn vrienden van de ’donderdagavond-groep’ (waarvan Ralph enMaikel nog niet zijn genoemd en die naast alle gezelligheid vooral genoemd moeten worden vanwegede promotiesketches en bijbehorende meer of minder verantwoorde experimenten), de bijbehorendeChinees op de Kruisstraat (zowel voor als na de transformatie), mijn andere vrienden, mijn zusFemke en last but not least natuurlijk mijn ouders, voor alle goede zorgen en omdat ze erin geslaagdzijn om de jarenlange uitloop van mijn promotie te verwerken zonder al te veel ‘feedback’.

pure.tue.nl › ws › files › 3709443 › 200610857.pdf · de promotiecommissie bestaat uit:...

Documents