IN-SITU MAGNETORESISTANCE MEASUREMENTS DURING PATTERNING OF SPIN VALVE DEVICES DEBORAH MORECROFT DOWNING COLLEGE CAMBRIDGE AUGUST 2003 A DISSERTATION SUBMITTED FOR THE DEGREE OF DOCTOR OF PHILOSOPHY AT THE UNIVERSITY OF CAMBRIDGE. Abstract This dissertation describes an experimental study on the patterning of thin films and spin valve devices. Initially the change in the magnetisation reversal of ferromagnetic Ni80Fe15Mo5 thin films was investigated as the shape anisotropy was increased using optical lithography to pattern wire arrays. These structures show a progressive increase in coercivity and a transition between single and two-stage reversal with increasing milling depth. A similar patterning technique was applied to unpinned (Ni80Fe20/Cu/Ni80Fe20) pseudo spin valve (PSV) structures in order to enhance the coercivity of one of the ferromagnetic layers. The increased coercivity induced by micropatterning changed the natural similarity of the magnetic layers and the structure exhibited a small spin valve response. These initial measurements were carried out with separate milling and electrical characterisation steps. However, it was decided that it would be ideal to design a technique to do in-situ magnetoresistance measurements during milling. This meant that the samples could be milled and characterised in the same step, leading to a much cleaner and more efficient process. In-situ magnetoresistance measurements were carried out during micropatterning of PSV devices, and the measurements showed the evolution in the electrical response as wire structures were gradually milled through the thickness. Contrary to what was expected, the structures showed a maximum spin valve response when fully milled through. The effect of further increasing the shape anisotropy by reducing the wire width, and changing the material properties in the PSV structure has also been investigated. MR measurements were taken as the temperature was increased from 291K to 493K, and the results show that the patterned PSV structures have a better thermal stability than exchange biased spin valves with an IrMn pinning layer. The experiment was extended to the nanoscale, and the results show that a significant increase in MR is not observed despite the fact that the magnetic configuration tends more towards single domain. This is thought to be due to an increase in the initial resistance of the structures. A small increase in MR was observed as the wire width was decreased from 730 to 470nm, although the spin valve response is heavily dependent on the gallium dosage density during patterning in the Focused Ion Beam (FIB). Micromagnetic simulations were carried out, which agree with the experimental results and showed the change in the magnetisation reversal from rotation to switching as the dimensions were reduced on the nanoscale. ii Preface This dissertation is submitted for the degree of Doctor of Philosophy in the University of Cambridge. Except where specific reference is made, this is entirely my own work and includes nothing that is the outcome of work done in collaboration. No part of this thesis has been or is being submitted for any other qualification at this or any other university. This dissertation does not exceed the limit of length. Some of the work in this thesis has been published as listed below. Publications • “Control of the switching properties of magnetic thin films and spin valve devices by patterning”, D.Morecroft, C.W. Leung, N.A. Stelmashenko, J.L. Prieto, D.B. Jardine and M.G. Blamire. IEEE Trans. Mag. 37 Issue: 4 Part: 1 (2001) pp. 2079 –2081. • “In-situ magnetoresistance measurements during patterning of thin films and spin valve devices”, D.Morecroft, C.W. Leung, J.L. Prieto, G.Burnell and M.G. Blamire J. Appl. Phys. 91, pp. 8575-8578 (May-2002). • “In-situ magnetoresistance measurements of nanopatterned pseudo spin valve devices”, D. Morecroft, D.J. Kang, B. Van Aken, J.L. Prieto and M.G. Blamire. J. Appl. Phys., (To be submitted). • “A study of exchange bias in spin valve devices by in-situ magnetoresistance measurements.” D. Morecroft, J.L. Prieto and M.G. Blamire. J. Magn. Magn. Mater., (To be submitted). • “Integrated magnetic field sensor based on magnetoresistive Spin Valve structure”, J.L.Prieto, N.Rouse, N.K. Todd, D. Morecroft, J. Wolfman, J.E.Evetts, M.G. Blamire. Sensors & Actuators A 94 (2001) pp. 64-68. Deborah Morecroft Device Materials Group, Department of Materials Science, Cambridge University. August 2003. iii Acknowledgements So many people have contributed in different ways towards the publication of this thesis and I feel truly indebted. I am grateful to my supervisor Dr. Mark Blamire for guiding me through the Ph.D. project and for proof reading this thesis. Special thanks go to Dr. Jose Prieto for his constant support and encouragement, for his valuable comments on this thesis, for convincing me that I could run a marathon and for being a great friend. Thanks to Dr. Bas Van Aken ‘the long legged runner’ for his enlightening simulations and for checking through this thesis. Thanks to Dae-Joon Kang for the FIB work, and for never shying away from ‘just one more chip’. Thanks to Dennis Leung for providing the initial test samples, for interesting discussions and helpful advice throughout the project. Thanks to the Head of the Device Materials Group, Prof. Jan Evetts for always having his office door open, and for inspiring me with his deep insight in materials physics. Thanks to Dr. Gavin Burnell, Dr. John Durrell and Dr. Karen Yates for contributing towards the smooth running of the lab. I would like to thank my friends in the lab for making it a fun place to work, Brian “Spaghettino” Pang, Laura “Laurina” Singh, Edgar “the flying Equadorian”, Dr. Luis “Spaniardo” Hueso, the G- gang Dr. Ashish Garg and Dr. Yee-Siau Cheng, Dr. Ruman “daddy” Tomov, Dr. Vassilka “mummy” Tsaneva, Chris Bell, Sibe Mennema, Rachel Speaks, Dr. Chiranjib Mitra, Dr. Veni Madhav Adyam, James Loudon, Dr. Noel Rutter, James Ransley, Ugi Balasumbramaniam, James Chapman, Dr. Karl Sandeman, as well as former group members Dr. Jerome Wolfman, Dr. Robert “Salsa” Hadfield, Dr. Phil McBrien, you don’t realise what a difference you made. I have been financially supported by a grant from the Engineering and Physical Sciences Research Council (EPSRC). I am grateful to Nordiko for supplying the spin valve wafers for testing. A big thank you to my Dad, the true science enthusiast of the family, for reminding me what research is all about and for patiently reading through this thesis. Thanks to my sister Jane for the cartoon in the dedication, and to her husband Andrea for recently giving me a nephew ‘Thomas’. Thanks to my Mum, sister Simona and brother Anthony for their wise counsel and for putting up with me. Last but not least, thanks to Fenners gym and Cornucopia coffee….I couldn’t have made it through without you! iv Glossary of Acronyms MR Magnetoresistance AMR Anisotropic magnetoresistance GMR Giant magnetoresistance TMR Tunneling magnetoresistance VSM Vibrating sample magnetometer AFM Atomic force microscopy PSV Pseudo spin valve SV Spin valve AFM Antiferromagnetic FM Ferromagnetic NM Non-magnetic MRAM Magnetic random access memory SDT Spin dependent tunnelling RF Radio frequency LMIS Liquid metal ion source SPM Scanning probe microscopy STM Scanning tunnelling microscopy PSD Position sensitive detector EPD End-point detection v For my parents vi Contents. Abstract ii Preface iii Acknowledgements iv Glossary of Acronyms v Dedication vi Chapter 1 Introduction 1 1.1 Spintronics 2 1.2 Magnetoresistance 3 1.3 Overview of Dissertation 8 Chapter 2 Background Reviews 12 2.1 Magnetic Materials 13 2.1.1 The Weiss Mean Field Model 13 2.1.2 Quantum Theory of Magnetism 14 2.1.3 Antiferromagnetism 16 2.1.4 The band theory of ferromagnetism 17 2.1.5 Ni80Fe20 and CoFe magnetic materials 19 2.2 Magnetisation Processes 20 2.2.1 Energy considerations and domain patterns 20 2.2.2 Magnetic anisotropy 21 2.2.3 Domain walls 27 2.2.4 Magnetisation reversal of a single domain: coherent rotation 29 2.3 Ferromagnetic Wire Structures 30 2.4 Nanomagnetism 38 Chapter 3 Spin Valves: Theory and Devices 47 3.1 Interlayer Exchange Coupling 48 3.2 The Physical Origin of GMR 50 3.3 Exchange Biased Spin Valve Structures 53 vii 3.3.1 Dependence on magnetic anisotropy 56 3.3.2 Dependence on non-magnetic layer thickness and composition 56 3.3.3 Dependence on the ferromagnetic layer thickness and composition 58 3.3.4 Dependence on the exchange bias material and thickness 59 3.3.5 Interfacial roughness and structural quality 59 3.3.6 Temperature dependence 61 3.3.7 Other exchange biased spin valve structures 61 3.4 Pseudo Spin Valve Structures and MRAM 63 3.4.1 Magnetic Random Access Memories (MRAMs) 67 Chapter 4 Experimental Techniques 74 4.1 Device microfabrication 75 4.1.1 Argon Ion Milling 76 4.1.2 D.C Sputter deposition 78 4.2 Device nanofabrication 79 4.2.1 Radio Frequency (RF) Sputter deposition 81 4.2.2 Focused Ion Beam (FIB) lithography 81 4.3 Device characterisation 85 4.3.1 Electrical characterisation 85 4.3.2 Magnetic characterisation 86 4.3.3 Structural characterisation 87 4.3.4 Magnetic Domain Imaging 90 Chapter 5 Control of the Switching properties of Magnetic Thin Films and Spin Valve Devices by Patterning 92 5.1 Motivations 93 5.2 Laterally patterned thin films 93 5.3 Patterning through the thickness of thin films 98 5.4 NiFe Pseudo Spin Valve Measurements 101 5.5 Conclusions 107 viii Chapter 6 Design, Building and Testing an In-situ Magnetoresistance Measurement Rig in an Argon Ion Miller. 110 6.1 Motivations 111 6.2 The Electrical Set-up 112 6.3 The Sample Holders 114 6.4 The Field Coils 115 6.4.1 The field coil with a soft magnetic core 115 6.4.2 Field coils without a soft magnetic core. 120 6.5 The Lock-in amplifier 121 6.6 The Voltage-Offset Amplifier 122 6.7 Signal Transmission and Noise Reduction 124 6.8 Rig calibration 126 6.8.1 The Resistance Model 128 6.8.2 In-situ milling depth calculation 130 6.9 Conclusions 132 Chapter 7 In-situ Micropatterning of Pseudo Spin Valves 134 7.1 Motivations 135 7.2 Sample preparation 135 7.3 In-situ magnetoresistance measurements 136 7.4 Milling Depth Calculations 139 7.5 Induced anisotropy and shape anisotropy 144 7.6 Thermal Stability Comparison 145 7.7 Micro-patterning of CoFe Pseudo Spin Valves 147 7.8 Conclusions 149 Chapter 8 In-situ Nanopatterning Pseudo Spin Valves. 151 8.1 Motivations 152 8.2 FIB sample preparation and analysis 153 8.3 FIB Nanopatterning the Pseudo Spin Valves 160 8.4 In-situ Nanopatterning MR Measurements 163 ix x 8.5 Micromagnetic Simulations 169 8.6 Conclusion 176 Chapter 9 Summary of Dissertation 180 Other work 183 APPENDIX A 187 CHAPTER 1 Imagination is more important than knowledge. Albert Einstein Chapter 1 Introduction 1.1 Spintronics 1.2 Magnetoresistance 1.3 Overview of the Dissertation 1 CHAPTER 1 1.1 Spintronics. Recent years have seen rapid developments in the new field of spintronics, (also called magnetoelectronics or spin electronics) [1]. This new research area combines two traditional branches of physics: magnetism and electronics. The aim is to find ways to manipulate the electron spin in transport processes. Until recently information processing technology has been based on purely charge-based devices, ranging from the vacuum tube to today’s million- transistor microchips. These devices rely on the movement of electric charges, and ignore the extra degree of freedom which is inherent to every electron: the electron spin. Magnetism and electron spin are interlinked and according to quantum theory two electrons are allowed to have the same energy as long as they have different spin directions: spin-up (↑) electrons and spin down (↓) electrons. In an ordinary electric circuit the spins are randomly oriented and have no effect on current flow. In a magnetic field the spin-up and spin-down electrons have different energies depending on their orientation with respect to the applied field. Spintronic devices manipulate spin-polarised currents by using the electron spin to control current flow. Ferromagnetic materials are particularly appropriate for these devices because they exhibit spontaneous magnetisation in the absence of an applied field, which creates an imbalance of spin populations near the Fermi level in the electronic structure. The imbalance of available energy states for spin-up and spin-down electrons leads to a difference in the resistance of the two types of electron, i.e. ferromagnetic materials can provide charge carriers that are spin polarised. In fact, the modulation of spin-polarised currents in ferromagnetic materials has received considerable attention partly due to the possible magnetoelectronic applications [2], but also due to the interesting fundamental physics. Magnetism (and hence electron spin) has always been important for information storage, and magnetic recording is by far the largest economical application of magnetism [3]. Therefore, it is not surprising that the information storage industry has provided the initial successes in spintronics technology. Computers now come fitted with high-capacity hard drives that rely on a spintronic effect, giant magnetoresistance (GMR) to read the densely packed data. Magnetic Random Access Memory (MRAM) is a strong competitor for the next new type of computer memory. MRAM will have the capability to retain the magnetic information even when the power is switched off, but unlike present forms of non-volatile memory, it will have switching rates and re-writability challenging those of conventional RAM. Another interesting area of spintronics research is looking at the possibility of achieving spin- polarised currents in semiconductors instead of metals. This would allow a wealth of existing microelectronics techniques to be co-opted and would open up the possibility of other types of devices controlled by polarized light or electric fields [4,5]. This avenue of research may lead to a new class of multifunctional electronics that combine logic, storage and communications 2 CHAPTER 1 on a single chip. Among the major achievements of spintronics to date is the understanding of spin dependent transport properties in various physical systems. These include metallic multilayers showing giant magnetoresistance, ferromagnetic tunnel junctions exhibiting spin dependent tunnelling and certain ferromagnetic oxides showing colossal magnetoresistance near the metal-insulator transitions. The transport of spin polarised currents in semiconductors is barely understood to date, but interesting first results have been achieved [6]. In 1999, two groups succeeded in injecting spin across the interface between two semiconductors – one magnetic and one non-magnetic. Laurens Molenkamp and colleagues at the University of Würzburg in Germany used the magnetic semiconductor zinc selenide as the spin-aligning ferromagnetic layer [7]. Also Hideo Ohno of Tohoku University in Japan collaborated with David Awschalom’s team at the University of California in Santa Barbara to produce manganese-doped GaAs as the ferromagnetic layer [8]. On both cases, the researchers passed their spin-polarised current into GaAs-based light-emitting diode with an efficiency of about 90 per cent. Their success was confirmed by the fact that the emitted light was circularly polarised- a direct consequence of the spin polarization of the current. There are various types of magnetoresistance (MR) depending upon the different materials and construction designs. The recent research into new MR effects has lead to an extraordinary range of solid-state structures, e.g. magnetic multilayers and tunnel junctions. The next section will give a brief overview of the various types of MR effects. 1.2 Magnetoresistance. Magnetoresistance (MR) is defined as a change in the electrical resistance of a substance in the presence of a magnetic field. The signal response of a device is often characterised by the percentage MR, as shown by Eqn. (1.1), where ∆R is the change in resistance in an applied field and R is the resistance in the absence of an applied field ( ) R % RMR ∆= (1.1) Many different forms of MR have been researched, using a variety of different materials and multilayer designs. This section will give an overview of the different types of magnetoresistance effect including ordinary MR, anisotropic MR, giant MR, tunnelling MR, colossal MR, and ballistic MR. Ordinary magnetoresistance (OMR). Consider a metal or semiconductor sample as shown in Fig. 1.1. The resistance R, measured by the voltage between A and B in the presence of a constant current Ix, is found to increase 3 CHAPTER 1 by ∆R when a magnetic field is applied. This is the transverse magnetoresistance effect, and occurs because the conduction electrons are displaced from their trajectories by the Lorentz force. In general, ∆R/R=aH2, where a is a different constant for each metal [9]. An increase in resistance is also found when the applied field is parallel to the current, but ∆R is about half as large and is known as the longitudinal MR. Ordinary MR is very small for moderate magnetic fields <1%. y Hy Ez ImIm x z Positive charge accumulated C F Ix BA E Ix b Negative charge accumulated D Fig. 1.1 Schematic diagram to show the experimental arrangement for measuring transverse magnetoresistance effect. The sample shown has a negative Hall coefficient. Hy is into the page, and Im is the current in the coils producing Hy. (After Ref[ 8]). Anisotropic magnetoresistance (AMR). The variation in resistance of typical magnetic metals at room temperature depends upon the orientation of the field with respect to the measuring current. When the applied field is parallel to the current the resistance increases with increasing field, and when the applied field and current are orthogonal the resistance decreases with increasing field. The difference in resistance at high fields represents the anisotropic magnetoresistance effect (AMR) common to all ferromagnetic metals. The physical origin of the magnetoresistance effect lies in the spin-orbit coupling: as the magnetisation Ms rotates, the electron cloud about each nucleus deforms slightly, and this deformation changes the amount of scattering undergone by the conduction electrons in their passage through the lattice [10-12]. The amount of AMR varies for different materials, for Fe AMR ∆R/R ≈0.4%, whilst for Ni70Co30 AMR ∆R/R ≈5%. Fig. 1.2 shows examples of the anisotropic magnetoresistance effect in sputtered polycrystalline films of Fe, Co, Ni, Ni81Fe19, Ni70Co30, and Ni50Co50. The change in resistance due to AMR can be described by Eqn. (1.2), where θ is the angle between the current and the magnetisation. (1.2) θ2min cosAMRtot RRR ∆+= 4 CHAPTER 1 The AMR effect has commonly been used in computer hard-disk drives and other sensors, however they are now being replaced by spin valve and GMR multilayer sensors, which have larger sensing signals. Fig. 1.2 Examples of the anisotropic magnetoresistance effect in sputtered polycrystalline films of Fe, Co, Ni, Ni81Fe19, Ni70Co30 and Ni50Co50. The full and dotted lines correspond to magnetic field applied orthogonal and parallel to the current respectively in the plane of the films. The films in each case are ≅1000Å thick (After Ref [13]). Giant magnetoresistance (GMR). In 1988 Baibich et al. observed magnetoresistance changes as large as 50% at low temperatures in Fe/Cr multilayer structures [14]. The Fe layers were antiferromagnetically coupled across the Cr spacer layers, and the large change in resistance was found to be due to the change in the relative orientation of the ferromagnetic layers from antiparallel to parallel alignment as the field was increased. In order to achieve a large resistance change it was important for the Fe layers to be antiparallel in the absence of an applied field, and it was found that the interlayer coupling oscillated between ferromagnetic and antiferrromagnetic depending on the thickness of the Cr interlayer [15,16]. The effect has since been observed for a variety of different transition metal/non-magnetic metal multilayers, including Co/Cu which displayed GMR as large as 110% at room temperature for magnetic fields ∼20kOe. Values of ∼20% or more can be obtained for a field of a few tens of Oersteds. The resistance of the structure varies with the angle between the magnetisation of adjacent magnetic layers 5 CHAPTER 1 as cosθ, and since the net moment of the structure M varies as cos(θ/2), the change in resistance due to GMR can be described by Eqn. (1.3). (1.3) ( ) ( ) 2cos1min θθ −∆+= GMRRRR The physical mechanism of GMR is extrinsic and it can be engineered, which is part of the reason why there was such intense interest in optimising the effect for sensor applications. It is best described using a two current model which assumes that the conduction electrons are sub-divided into two spin sub-bands: those with spins parallel to the magnetisation (majority) and those with spins antiparallel to the magnetisation (minority). There is a dissimilar scattering rate for the two electron channels due to a difference in the available energy states at the Fermi level. When the ferromagnetic layers are ferromagnetically coupled, the majority electrons see a low resistance channel through the entire structure, which leads to a lower resistance overall. When the ferromagnetic layers are antiferromagnetically coupled, both spin channels are equally scattered through the entire structure, leading to a higher resistance state. The GMR effect is shown schematically in Fig. 1.3. (b) (a) Fig. 1.3 Schematic of conduction in multilayer magnetic film array, showing how differential spin scattering produces a different resistance for parallel (a) and antiparallel (b) film magnetisations. Since the initial discovery of GMR in magnetic/non-magnetic multilayers, many alternative designs have been investigated to control the relative orientations of the ferromagnetic layers and to optimise the signal response and sensitivity. These will be described in more detail in Chapter 3, where spin valve theory and devices will be reviewed. The GMR effect has been of great interest to the hard-disk drive industry and IBM introduced the first GMR read-head sensor in 1997, within 10 years of its initial discovery. The major advantages over AMR are the improved signal response which enabled miniaturisation and therefore increased data storage density, and also the ability to engineer the response. The figure of merit of a GMR sensor is given by the change in the MR divided by the field interval, and biasing magnets are normally used to move the operating point of the sensor to achieve the maximum figure of merit. 6 CHAPTER 1 Tunnelling magnetoresistance (TMR). In the spin dependent tunnelling (SDT) effect, electrons tunnel across a thin (1-3nm) insulating barrier between two ferromagnetic electrodes, and the tunnelling electric current between two metal layers depends on the relative angle between the magnetisation in the two metal layers. This phenomenon was first observed by Julliere [17] in a Fe-Ge-Co junction in 1975. Slonczewski showed theoretically that the tunnelling electric conductance varies as cosθ, where θ is the angle between magnetisations in the metal layers [18]. The TMR effect can be understood using the same two channel electron spin model as for GMR. However, TMR is due to a dissimilar tunnelling conductance between asymmetric spin sub-bands, whilst GMR occurs due to spin scattering asymmetry through the entire structure. The magnitude of TMR is strongly dependent on the spin polarisation of the conduction electrons, i.e. the spin asymmetry of the density of states at the Fermi level. Recently significant room temperature MR signals have been observed, ranging from 20-40% [19,20]. These recent advances have opened up a range of potential applications for TMR devices including non- volatile tunnel junction random access memories (TJRAM) and MR sensors for very high- density magnetic storage (HD drives). However, problems related to the insulating barrier still need to be overcome including reducing the junction resistance [21,22] and improving the thermal stability. Colossal magnetoresistance (CMR). The manganese oxides of general formula RE1-xMxMnO3 (RE= rare earth, M= Ca, Sr, Ba, Pb) have remarkable interrelated structural, magnetic and transport properties induced by the mixed valence (3+-4+) of the Mn ions. In particular, they exhibit very large negative magnetoresistance, called colossal magnetoresistance (CMR), in the vicinity of the metal- insulator transition for certain compositions. Historically, the mixed-valence perovskites La1- xMxMnO3 (M= Ca, Sr, Ba) were studied in the fifties, both experimentally [23,24] and theoretically [25-28]. The experiments performed on polycrystalline samples showed antiferromagnetic (AF) insulating behaviour at low and high x values and ferromagnetic (F) metallic behaviour in a certain range of concentrations centred around x≈1/3. The research shows that the magnetic and electrical properties of the manganites are governed by exchange interactions between the Mn ion spins [25,26], although the fundamental physics is still not fully understood. The great interest in CMR materials has been partly due to their large MR, but also because they posses a high spin polarisation of the conduction electrons, which could have implications for spin injection devices. In 1993, 60% MR was reported at room temperature in La0.67Ba0.33MnO3 thin film [29], and the following year an MR effect in excess of a million percent was reported as in La0.67Ba0.33MnO3 at a temperature of 77K [30]. The 7 CHAPTER 1 experiments were then extended to other CMR oxides such as the layered manganites, La2- 2xSr1+2xMn2O7 [31] and Sr2MoFeO6 [32]. The main drawback of using CMR materials for technological applications is the need for large magnetic fields, typically in the range of several T. Also the MR tends to be very large around the Curie temperature Tc, whereas it becomes negligible at temperatures much lower or higher than Tc. Ballistic magnetoresistance (BMR). Ballistic magnetoresistance (BMR) is an effect that occurs in the conduction of spin-polarised electrons through highly constricted (≈10nm) junctions in which the spin-flip mean free path is long compared to the magnetic domain wall width. When a magnetic domain wall resides in the constriction, the electrical resistance is much larger than it is after an external magnetic field is applied to sweep out the domain wall. The resulting magnetoresistive effect is much larger than for GMR or TMR. Experimentally there is a large variation in the magnitude of the BMR, as well as a wide range in the conductance where the peak BMR occurs. García et al. observed BMR values up to ≈300% in mechanically formed nanocontacts of various ferromagnetic metals and up to 700% in electrodeposited Ni-Ni nanocontacts [33]. Verluijs et al. [34] observed 540% BMR in mechanically formed nanocontacts of the half-metallic ferromagnet Fe3O4. More recently Chopra and Hua [35] reported over 3000% BMR in Ni nanocontacts deposited on electropolished Ni tips, and even larger 100,000% BMR [36] using only mechanical pulled Ni wires to eliminate the possibility of any extraneous chemical layer being present. The results were shown to be stable over several cycles. According to Chung et al. [37], spin-ballistic transport through the nanocontact is responsible for BMR in different regimes of conductance from normal metal (Ni, Co, Fe), to semimetal (CrO2) to insulators. The behaviour is determined by the spin-scattering at the domain wall and controlled by the domain wall thickness only. This implies a universal BMR mechanism, indicating that the effect should be observable in a very large class of ferromagnetic material systems. 1.3 Overview of the Dissertation. There are currently two systems of units in widespread use in magnetism: the Gaussian or cgs system and the SI or mks unit system. The unit systems have advantages and disadvantages, and they are both used throughout this thesis. The cgs and mks systems of magnetic units have different philosophies. The cgs system took an approach based on magnetostatics and the concept of the ‘magnetic dipole’, while the mks system takes an electrodynamic approach to magnetism based on electric currents. The mks system is generally used where fundamental concepts and calculations are involved, and material properties are generally discussed in cgs 8 CHAPTER 1 units. The table in Fig. 1.4 gives the equations for some of the magnetic quantities used in this thesis in mks and cgs units, as well as the conversion factors between them. cgs units mks (SI) units B=H+4πM B in gauss (G) H in oersteds (Oe) M in emu/cm3 B=µ0(H+M) B in tesla (T) H in amperes/metre (A/m) M in tesla cgs to SI SI to cgs B: 1G=10-4T H: 1Oe= 79.6A/m M: 1emu/cm3=12.57×10-4T 1T=104G 1A/m=12.57×10-3Oe 1T=796emu/cm3 Fig. 1.4 A table to show the equations for some magnetic quantities in csg and mks units, the conversion factors are also shown. The theme of this dissertation is an experimental study of how the magnetic and electrical properties of ferromagnetic thin films and pseudo spin valve structures can be controlled by patterning. Chapter 2 gives a brief review of the general background including the theory of ferromagnetism, micromagnetics, and nanomagnetics. A review is also given of the experimental work already carried out within the field. Two different subjects are described, which are directly related to the following chapters: (1) the effect of patterning highly anisotropic structures in ferromagnetic thin films, and (2) the effect of miniaturisation to the sub-micron scale on the magnetisation reversal of ferromagnetic thin films. Chapter 3 gives a review of the work carried out on GMR structures including antiferromagnetically coupled multilayers, exchange biased spin valves and pseudo spin valves. A brief overview of the current status of MRAM is also given, which is one of the applications for pseudo spin valve devices. The experimental methods used in this work are described in Chapter 4, including microfabrication, nanofabrication, and a variety of analysis techniques. Chapter 5 describes the initial experimental results including the micropatterning and magnetic analysis of soft magnetic thin films and the micropatterning of one of the layers in a Ni80Fe20/Cu/Ni80Fe20 pseudo spin valve structure and magnetoresistance analysis. 9 CHAPTER 1 10 Chapter 6 describes the design, building and testing of the in-situ magnetoresistance measurement rig including the design of the sample holder and field coils, the voltage-offset amplifier, noise reduction measurements and system calibration. Chapter 7 gives the first experimental measurements obtained from the in-situ rig showing the change in magnetoresistance with increasing milling depth as the pseudo spin valves are patterned with highly anisotropic wire array structures. Measurements were also carried out using a different pseudo spin valve construction with CoFe/Cu/CoFe. Magnetoresistance measurements were carried out at different temperatures to compare the thermal stability of the pseudo spin valve structures with exchange-biased spin valves. Chapter 8 shows the in-situ magnetoresistance results from nanopatterning highly anisotropic structures in pseudo spin valves and compares the experimental results with micromagnetic modelling. Finally this dissertation ends with a summary of the results in Chapter 9 including a section on ‘other work’. CHAPTER 1 References [1] D.D. Awschalom, M.E. Flatté, N. Samarth, “Spintronics”, Scientific American, June (2002); M. Ziese, M.J. Thornton (Eds.), Spin Electronics, Springer-Verlag Berlin Heidelberg, (2001). [2] J.M. Daughton, A.V. Pohm, R.T. Fayfield, C.H. Smith, J. Phys. D: Appl. Phys., 32, R169 (1999). [3] R.M. White, J. Magn. Magn. Mater. 226-230, 2042-2045 (2001) [4] J.M. Kikkawa, D.D. Awschalom, Nature, 397, 139 (1999). [5] H. Ohno, D. Cgiba, F. Matsukura, T. Omiya, E. Abe, T. Dieti, Y. Ohno, K. Ohtani, Nature, 408, 944 (2000). [6] M. Ziese, Spin Electronics, Chapter 17, pg 396-415, Springer-Verlag Berlin Heidelberg, (2001). [7] R. Fiederling et al. Nature, 402, 787 (1999). [8] Y. Ohno et al., Nature, 402, 790 (1999). [9] N. Cusack, The Electrical and Magnetic Properties of Solids, Chapter 6, pg 132, Longmans (1958). [10] T. R. McGuire, R.I. Potter, IEEE Trans. Mag. MAG-11, 1018 (1975). [11] J. Smit, Physica (Utrecht) XVI, 612 (1951). [12] P. Ciureanu, “Magnetoresistive Sensors” in Thin Film Resistive Sensors, ed. by P. Ciureanu, S. Middelhoek, Institute of Physics Publishing, Bristol, pg. 253 (1992). [13] S.S.P. Parkin, Ultrathin Magnetic Structures II, Springer-Verlag Berlin Heidelberg, Chapter 2.4, 156 (1994). [14] M.N. Baibich, J.M. Broto, A. Fert, F. Nguyen-Van-Dau, F. Petroff, P. Etienne, G. Creuzet, A. Friedrich, J. Chazelas: Phys. Rev. Lett.,61, 2472 (1988). [15] S.S.P. Parkin, Phys. Rev. Lett., 67, 3598 (1991). [16] S.S.P. Parkin, R. Bhadra, and K. P. Roche, Phys. Rev. Lett., 66, 2152 (1991). [17] M. Jullière, Phys. Lett. 54A, 225 (1975). [18] J.C. Slonczewski, Phys. Rev. B., 39, 6995 (1989). [19] J. Moodera, J. Nassar, G. Mathon, Annu. Rev. Mater. Sci., 29, 381 (1999). [20] S.S.P. Parkin, K.P. Roche, M.G. Sammant, P.M. Rice, R.B. Beyers, R.E. Scheurlein, E.J. O’Sullivan, S.L. Brown, J. Bucchiganno, D.W. Abraham, Y. Lu, M. Rooks, P.L. Trouiloud, R.A. Wanner, W.J. Gallagher, J. Appl. Phys., 85, 5828 (1999); R.C. Sousa, J.J. Sun, V. Soares, P.P. Freitas, A. Kling, M.F. da Silva, J.C. Saores, Appl. Phys. Lett., 73, 3288 (1998); S. Cardoso, V. Gehanno, R. Ferreira, P.P Freitas, IEEE Trans. Magn., 35, 2952 (1999). [21] J.J. Ruigrok, R. Coehoorn, S.R. Cumpson, H. van Kesteren, J. Appl. Phys., 87, 5398 (2000). 10 CHAPTER 1 11 [22] J.J. Sun, K. Shimazawa, N. Kasahara, K. Sato, S. Saruki, T. Kagami, O. Redon, S. Araki, H. Morita, N. Marsuzaki, Appl. Phys. Lett., 76, 2424 (2000). [23] G.H. Jonker, J.H. Van Santen, Physica, 16, 337 (1950); G.H. Jonker, J.H. Van Santen, Physica, 16, 599 (1950); G.H. Jonker, Physica, 20, 1118 (1954). [24] E.O. Wollan, W.C. Koehler, Phys. Rev., 100, 545 (1955). [25] C. Zener, Phys. Rev., 81, 440 (1951). [26] P.W. Anderson, H. Hasegawa, Phys. Rev., 100, 675 (1955). [27] P.G. de Gennes, Phys. Rev., 118, 141 (1960). [28] J.B. Goodenough, Phys. Rev., 100, 564 (1955). [29] R. von Helmholt, J. Singleton, D.A. Keen, R. McGreevy, K. Samwer, Phys. Rev. Lett., 71, 2331 (1993). [30] S. Jin, T.H. Tiefel, M. McCormack, R.A. Fastnach, R. Ramesh, L.H. Chien, Science, 264, 413 (1994); M. McCormack, S. Jin, T.H. Tiefel, R.M. Fleming, J.M. Phillips, R. Ramesh, Appl. Phys. Lett., 64, 3035 (1994). [31] Y. Morimoto, A. Asamitsu, H. Kuwahara, Y. Tokura, Nature, 380, 141 (1996); Y. Morimoto, Aust. J. Phys., 52, 255 (1999). [32] K.I. Kobayashi, T. Kimura, H. Sawada, K. Terakura, Y. Tokura, Nature, 395, 677 (1998). [33] N. García, M. Muñoz, Y.W. Zhao, Phys. Rev. Lett., 82, 2923 (1999); Appl. Phys. Lett., 76, 2586 (2000); N. García, M. Muñoz, G.C. Qian, H. Roher, I.G. Saveliev, Y.W. Zhao, Appl. Phys. Lett., 79, 4550 (2001). [34] J.J. Verluijs, M.A. Bari, J.M.D. Coey, Phys. Rev. Lett., 87, 026601 (2001). [35] H.D. Chopra, S.Z. Hua, Phys. Rev. B., 66, 020403-1 (2002). [36] S.Z. Hua, H.D. Chopra, Phys. Rev. B., 66, 060401-1 (2003). [37] S.H. Chung, M. Muñoz, N. García, W.F. Egelhoff, R.D. Gomez, Phys. Rev. Lett., 89, 287203-1 (2002). CHAPTER 2 Few subjects in science are more difficult to understand than magnetism. Encyclopaedia Britannica. Chapter 2 Background Reviews 2.1 Ferromagnetic Materials 2.2 Magnetisation Processes 2.3 Ferromagnetic Wire Structures 2.4 Nanomagnetism 12 CHAPTER 2 2.1 Magnetic Materials The various types of magnetic materials are traditionally classified according to their bulk susceptibility, χ, and are listed below. (i) For diamagnetic materials the susceptibility is small and negative χ≈ -10-5, and the magnetic response opposes the applied magnetic field. Examples of diamagnets are copper, silver, gold, bismuth and beryllium. Superconductors form another special group of diamagnets for which χ≈ -1. (ii) For paramagnets the χ is small and positive and typically χ≈ 10-3-10-5. The magnetisation of paramagnets is weak but aligned parallel with the direction of the magnetic field. Examples of paramagnets are aluminium, platinum and manganese. (iii) For ferromagnets the susceptibility is positive, much greater than 1, and typically can have values χ≅50 to 10,000. They exhibit spontaneous magnetisation in zero applied field, when the atomic magnetic moments tend to align parallel to each other. Examples of ferromagnetic materials are iron, cobalt, nickel, and several rare earth metals and their alloys. (iv) Antiferromagnets are ionic compounds, i.e. oxides, sulphides, chlorides. They are similar to paramagnets in that they have a small positive susceptibility, but their susceptibility has a different dependence with temperature and they have an entirely different magnetic structure. They also exhibit spontaneous magnetisation in zero field, when the atomic magnetic moments tend to align antiparallel to each other. This section will concentrate on the properties of ferromagnetic and antiferromagnetic materials, including the Weiss mean field model, the quantum theory of magnetism, a brief review of antiferromagnetism, the band theory of ferromagnetism and a review of the properties of Ni80Fe20 and CoFe, which are the magnetic materials used in this thesis. 2.1.1 The Weiss mean field model. Around the beginning of the century Weiss developed a theory to explain the behaviour of ferromagnetic materials. He introduced the idea that elementary moments interact with one another, and the interaction can be expressed in terms of a fictitious internal field which he called ‘the molecular field’ Hm, which acts in addition to the applied field. He assumed that the molecular field was very large, its magnitude was independent of any externally applied field, and its direction was not fixed, but always parallel to the magnetisation. Weiss thought that the molecular field was caused by the magnetisation of the surrounding material and suggested that there was a proportional relationship between the molecular field and the bulk magnetisation M, Hm=γM, where γ is a constant called the Weiss constant. As soon as a field is applied, M acquires a small non-zero value, so that a non-zero molecular field appears to aid the applied field in increasing the magnetisation. In order to explain the fact that 13 CHAPTER 2 ferromagnetic materials do not remain saturated when the applied field is removed, Weiss introduced the concept of magnetic domains. He proposed that a ferromagnet is divided into regions (domains), within which the magnetisation is equal to the saturation value. The magnetisation in different domains is in different directions, so that the magnetisation of a ferromagnetic specimen could be small or even zero. Magnetic saturation (Ms) is achieved by aligning the magnetisation of each domain with the applied field. The value of Ms varies with temperature, T, decreasing from a maximum value at T=0K at first slowly, then more and more rapidly as T increases, becoming zero at the Curie temperature, Tc. Above Tc the behaviour is similar to a paramagnetic material, with the magnetisation being proportional to the field and the susceptibility decreasing with increasing temperature. χ Tc0 T Fig. 2.1 A schematic of the variation of susceptibility χ with temperature T for a ferromagnetic material according to the Curie-Weiss law. The change in the susceptibility χ with temperature is described by the Curie-Weiss law, given in Eqn. (2.1), where C is the Curie constant. The relationship is shown graphically in Fig. 2.1. cTT C −=χ (2.1) Above the Curie temperature, M is zero when there is no applied field, so that Hm is also zero. The molecular field is usually a very large value, for Fe, Hm=6.9×106 Oe, this value is much larger than any continuous field produced in the laboratory. Examples of the Curie temperatures of ferromagnetic elements are: iron, Tc=770°C, nickel, Tc=358°C, and cobalt, Tc=1130°C. 2.1.2 Quantum theory of magnetism. The physical origin of the molecular field was not understood until 1928, when Heisenberg showed that it was caused by quantum-mechanical exchange forces. He showed that the 14 CHAPTER 2 exchange force is a consequence of the Pauli exclusion principle, which states that two electrons can have the same energy only if they have opposite spins. Therefore, two hydrogen atoms can come so close together that their two electrons can have the same velocity and occupy very nearly the same small region of space, i.e. have the same energy, provided these electrons have opposite spin. If their spins are parallel, the two electrons will tend to stay far apart. The electrostatic energy is modified by the spin orientations, which means that the exchange force is fundamentally electrostatic in origin. (The term “exchange” came from the idea that the electrons on two hydrogen atoms are indistinguishable and can therefore exchange places. The interchange of electrons between the atoms introduces the additional term the exchange energy into the expression for the total energy of the two atoms). Heisenberg showed that the exchange energy plays a decisive role in ferromagnetics. If two atoms i and j have spin angular momentum Sih/2π and Sjh/2π, then the exchange energy between them is given by (2.2)φCosSSJE jiexex 2−= where Jex is the exchange integral and φ is the angle between the spins. If Jex is positive, Eex is a minimum when the spins are parallel (Cosφ =1) and a maximum when the spins are anti-parallel (Cosφ = −1). If Jex is negative, the lowest energy state results from anti-parallel spins (antiferromagnetism). Ferromagnetism is due to the alignment of spin moments on adjacent atoms, so a positive value of exchange integral is necessary. The idea that exchange forces are responsible for ferromagnetism and antiferromagnetism allowed scientists to rationalise the appearance of ferromagnetism in some metals and not in others. The curve in Fig. 2.2 is called the Bethe-Slater curve, and it shows the variation of the exchange integral with the ratio ra/r3d, where ra is the radius of an atom and r3d the radius of its 3d shell of electrons. If two atoms of the same kind are brought closer and closer together without any change in the radius r3d of their 3d shells, the ratio ra/r3d will decrease from large to small values. When the ratio is large, Jex is small and positive. As the ratio decreases and the 3d electrons approach each other more closely, the positive exchange interaction, favouring parallel spins, becomes stronger and then decreases to zero. A further decrease in the interatomic distance brings the 3d electrons so close together that their spins must become antiparallel (negative Jex). This condition is called antiferromagnetism. The curve correctly separates the ferromagnetic 3d elements iron, cobalt and nickel from the antiferromagnetic 3d elements chromium and manganese. 15 CHAPTER 2 Jex Cr Mn Fe Co Ni 0 + - rab/rd ra 3d Fig. 2.2 The Bethe-Slater curve representing the variation of the exchange integral J with interatomic spacing rab and radius of the unfilled d shell rd. The positions of various magnetic elements on this curve are indicated. [1] 2.1.3 Antiferromagnetism The susceptibility of an antiferromagnetic material varies with temperature is shown in Fig. 2.3. χ TN -θ 0 T Fig. 2.3 A schematic of the variation of χ with temperature in the paramagnetic regime of materials which undergo a transformation to antiferromagnetism. [2] As the temperature decreases, χ increases but finally goes through a maximum at a critical temperature TN called the Néel temperature. The substance is paramagnetic above TN and antiferromagnetic below it. The Curie-Weiss law also applies to antiferromagnets above their ordering temperatures. However the sign of the constant term TN in the denominator is positive so the law becomes NTT C +=χ (2.3) In many ways the Néel temperature of an antiferromagnet is analogous to the Curie temperature of a ferromagnet. Both mark the borderline temperature above which the material is disordered and below which it is ordered. However, in an antiferromagnetic material the 16 CHAPTER 2 molecular field tends to align the magnetic moments antiparallel to each other. In other words the exchange force is negative. 2.1.4 The band theory of ferromagnetism. The Curie-Weiss and Heisenberg models of ferromagnetism are based on local moment theory, i.e. the electrons are localised around the atom. However, for most of the 3d ferromagnetic materials the ‘magnetic’ electrons are outer electrons, which are free to move through the solid. Furthermore, the magnetic moments per atom are not integral multiples of the electron spin, the Bohr magneton as required by localised models. The band theory of ferromagnetism was first proposed by Stoner [3] and then independently by Slater [4]. It has been successful in explaining non-integral values of atomic magnetic moments and predicting some aspects of magnetic behaviour of the 3d series metals and alloys. In the case of ‘free’ atoms, i.e. atoms locates at large distances from one another, the electrons occupy sharply defined energy levels in accordance with the Pauli exclusion principle. However, when atoms are brought together to form a solid, their electron clouds overlap, which leads to a splitting of the energy levels. So when N atoms come together to form a solid, each level of the free atom must split into N levels, because the Pauli exclusion principle now applies to the whole group of N atoms. The extent of splitting is different for the different levels, as indicated in Fig. 2.4. 1s 2s 2p 3d 4s d d El ec tro n en er gy E Fig. 2.4 Splitting of the electron energy levels as the interatomic distance d decreases.(After Ref. [5]) 17 CHAPTER 2 In the transition elements, the outermost electrons are the 3d and 4s; these electron clouds are the first to overlap as the atoms are brought together, and correspondingly the first to split. When the interatomic distance d has decreased to d0, the equilibrium value for the atoms in the crystal, the 3d levels are spread into a band extending from B to C, and the 4s levels are spread into a much wider band, extending from A to D, because the 4s electrons are further from the nucleus. Fig. 2.5 shows a schematic of the density of states of the sp- and d- bands of ferromagnetic Fe, Co and Ni. N(E) represents the density of states and E the electron energy. The density of 3d levels is far greater than the 4s levels, because there are five 3d levels per atom with a capacity of 10 electrons, whereas there is only one 4s level, with a capacity of 2 electrons. The area under the N(E) vs E curve is equal to the total number of levels in the band. Filled energy levels cannot contribute a magnetic moment, because the two electrons in each level have opposite spin and cancel each other out. In a partially filled energy band it is possible to have an imbalance of spins leading to a net magnetic moment per atom. This arises because the exchange energy removes the degeneracy of the spin-up and spin-down half bands. Fig. 2.5 Schematic diagram of the density of states in the sp- and d- bands of ferromagnetic Fe, Co and Ni. The total numbers of electrons in the spin-down (left) and spin-up (right) bands are also shown. (After Ref [6]). The larger the exchange energy the greater the difference in energy between these two half bands. Electrons fill up the band d occupying the lowest energy levels first. If the half bands are split, the electrons can begin to occupy the spin down half band before the spin up half band is full. This usually leads to a non-integral number of magnetic moments per atom. The ferromagnetism of Fe, Co and Ni is due to spin imbalance in the 3d band, and the force creating the imbalance is the exchange force. To create a spin imbalance, one or more electrons must be raised to higher energy levels, so these must not be too widely spaced or the exchange force will not be strong enough to have an effect. The 4s electrons make no contribution to the spin imbalance, because the density of levels in the 4s band is low, which 18 CHAPTER 2 means that the levels are widely spaced. Usually it is assumed that in ferromagnetic metals the conductivity is primarily carried by electrons from the sp- bands which are broad and, as a consequence, have low effective masses. In contrast, the d- bands are narrow and have high effective masses. The d- bands play a very important role in providing final states into which the sp electrons can be scattered. The criteria for the band theory of ferromagnetism can be summarised as follows: 1) The electrons must lie in partially filled bands so that there may be vacant energy levels available for electrons with unpaired spins to move into. 2) The density of levels in the band must be high, so that the increase in energy caused by spin alignment is small. 3) The atoms must be the right distance apart so that the exchange force can cause the electron spins in one atom to align the spins in the neighbouring atom. 2.1.5 Ni80Fe20 and CoFe magnetic materials. The two ferromagnetic materials used in this thesis are NiFe alloys (Permalloy) and CoFe alloys (Permendur). The soft magnetic NiFe alloys have diverse properties and are used in many fields of electrical and electronic engineering. Their structure dependent magnetic properties, such as coercivity, permeability and hysteresis loop shape, depend on the basic magnetic constants and on the microstructure. Throughout the project NiFe alloys in the range 80% were used, which have the highest permeability and therefore a square hysteresis loop. The square hysteresis loop results because in thin sheet form the demagnetising field is insufficient to nucleate a reverse domain and the exchange coupling between the grains is sufficiently strong to maintain the material at saturation in the absence of the reversal field. They can be made with low or even zero magnetostriction, and they have low coercivity, low anisotropy, and the microstructure is isotropic. Alloy Relative permeability µmax Saturation induction Bs (T) Coercivity Hc (Am-1) Curie temperature (°C) NiFe 5000-300,000 0.7 1-8 400 CoFe 5000 2.45 160 ≈980 Fig. 2.6 A table summarising the properties of NiFe and CoFe alloys used in this thesis work. (After Ref. [7, 8]). 19 CHAPTER 2 Cobalt is the only element which when alloyed with iron causes an increase in saturation magnetisation and Curie temperature. CoFe has a low anisotropy, a high permeability and is polycrystalline. Fig. 2.6 is a table summarising the magnetic properties of NiFe and CoFe. 2.2 Magnetisation Processes. In order to understand the magnetisation state in zero field and the magnetisation reversal processes, it is important to consider that the magnetic structure is a consequence of energy minimisation. This section will start by describing the energy considerations and the resulting domain patterns, which can be calculated by micromagnetic equations. There will then be an overview of the different types of anisotropy and domain walls. The final section will be on small particle switching theory, which applies to small particles that are too small to contain regular domain walls. 2.2.1 Energy considerations and domain patterns. In 1935 Landau and Lifshitz [9] showed that the existence of domains is a consequence of energy minimisation. The total energy of a ferromagnetic material in an applied field is given by the sum of the Zeeman energy, exchange energy, anisotropy energies and magnetostatic energy as shown in Eqn (2.4). ( ) .dVEEEE stataniexZee +++∫=Etot EZee (2.4) The Zeeman energy is due to the interaction between the external field H and the magnetisation vector M, and is given by .MH ⋅−= (2.5) The exchange energy was described in the previous section, and is the energy between the atomic magnetic moments, which tends to align them parallel for a ferromagnetic material and antiparallel for an antiferromagnetic material. Eqn. (2.6) is Heisenberg’s expression for the sum of the exchange energies across a material of finite size. ,2 ji ij ij SSJ ⋅−exE = ∑ (2.6) where Jij is the exchange constant between the two adjacent classical spins with spin quantum numbers Si and Sj at the i and j sites. The magnetostatic energy per unit volume of a dipole of magnetisation M in a magnetic field H is given by (2.7),.0 dMH∫−= µEstat When subjected to its own demagnetising field Hd, which is generated by M, we can put Hd=-NdM in the integral where Nd is the demagnetising factor so that the energy becomes 20 CHAPTER 2 20 0 2 . MN dMMN d d µ µ ∫ E E = = (2.8) A single domain specimen has a large magnetostatic energy associated with it, but the energy can be minimised by breaking up the magnetisation into localised regions (domains), providing for flux closure at the ends of the specimen. Providing that the decrease in magnetostatic energy is greater than the energy needed to form magnetic domain walls, a multidomain structure will occur. This is depicted in Fig. 2.7, which shows (a) the stray field from a uniformly magnetised sample, and a sample divided into (b) two, and (c) four domains. The sample has a high uniaxial anisotropy (e.g. Co), which is shown by the absence of closure domains. (b)(a) (c) Fig. 2.7 The stray field of (a) a uniformly magnetised sample, and a sample divided into (b) two, and (c) four domains. Magnetic anisotropy can significantly influence the shape of the hysteresis loop, and this term simply means that the magnetic properties depend on the direction in which they are measured. Anisotropy is exploited in the design of most magnetic materials of commercial importance and there are various different forms. It also plays a significant role in the experimental work in this thesis, and therefore the next section will explain the relevant types in more detail. 2.2.2 Magnetic anisotropy. There are several forms of magnetic anisotropy: 1. Crystal anisotropy. 2. Anisotropy induced by magnetic annealing. 3. Shape anisotropy. 4. Exchange anisotropy. 21 CHAPTER 2 Of all the different types, only crystal anisotropy is intrinsic to the material. The other forms are extrinsic or ‘induced’. Crystal anisotropy. Crystal anisotropy is due to spin-orbit coupling, the coupling between the spin and orbital motion of the electrons. When an external field tries to reorient the spin of an electron, the orbit of that electron also tend to be re-orientated, but the orbit is strongly coupled to the lattice and tries to resist any rotation on the spin axis. The energy required to rotate the spin system of a domain away from the easy direction, which is the crystal anisotropy energy, is the energy required to overcome the spin-orbit coupling. The strength of the anisotropy in any particular crystal is measured by the magnitude of the anisotropy constants K1, K2, etc. For cubic anisotropy the anisotropy energy EK can be expressed in terms of a series expansion of the direction cosines of the magnetic saturation Ms relative to the crystal axes. If Ms makes angles a, b and c with the crystal axes, then α1, α2 and α3 are the direction cosines of these angles, ( ) ( ) ......232221221232322222110 +++++ ααααααααα KKK=EK aE = (2.9) where K0, K1, and K2 are the anisotropy constants for a given material. K2 is so small that it is normally ignored and K0 is independent of angle. For Fe, K1>0 and so the easy axis are the <100> directions. Ni has an FCC crystal structure, K1<0 which represents that the easy axes are the <111> directions. Co is HCP and is a uniaxial crystal with only one easy direction and one hard direction; [ 0001] and [10 ] respectively. For a HCP crystal, the anisotropy can be represented by the one constant approximation: 01 _ (2.10)φ2sinulK where φ is the angle of magnetisation with respect to the unique axis, which for K>0 is the easy axis, while for K<0 it is the hard axis. Crystal anisotropy can have a large effect on the hysteresis loop. Fig. 2.8 shows the magnetisation curve for iron in different crystallographic directions. The graph shows that saturation can be achieved with quite low fields in the <100> direction, which is the easy direction of magnetisation. If a field is applied along the easy axis direction saturation is completed by domain wall motion. If the field is applied along the <110> or medium direction, domain wall motion occurs at low field until the domains are aligned in the anisotropy axes closest to the field direction. Magnetisation can only increase further by rotation of the Ms vector of each domain until it is parallel with the applied field. 22 CHAPTER 2 <100> <110> <111> Easy <100> <110> Medium Hard <111> 1800 1600 1400 1200 M (e m u/ cm 3 ) 1000 800 600 400 200 0 0 200 400 600 800 1000 H (Oe) Fig. 2.8 Magnetisation curve for single crystal iron. (After Ref. [10]). This is called domain rotation and it only occurs at high fields, because the field is acting against the strong anisotropy of the crystal. When the rotation is complete the domain wall disappears and the crystal is saturated. The crystal anisotropy is important when the sample is both single crystal and single domain; otherwise the magnetisation is dependent on other factors such as domain structures and induced anisotropy as discussed below. The crystal anisotropy of NiFe alloys ≈100J/m-3, which is much smaller than for Fe ≈47,000J/m-3. Anisotropy induced by magnetic annealing. When certain magnetic materials are deposited or heated (at T0. In a small field interval close to H=0 the magnetisation of the free layer reverses, whereas the magnetisation of the pinned F layer remains fixed. Only upon the application of a large negative field (equal to the exchange biasing field Heb), the exchange biasing interaction is overcome, and the pinned ferromagnetic layer reverses direction. The GMR effect is the increase of the resistance upon the change of the angle between neighbouring magnetic layers from a parallel to an antiparallel alignment. For combinations of F- and NM-layer compositions for which the GMR effect is high, the resistance curve shows a steep increase close to zero field, where the free layer magnetisation direction switches. The resistance stays high until the pinned layer reverses at the exchange bias field. There is usually a small magnetic coupling between the free layer and pinned layer, which leads to an offset, (Hcoupl in Fig. 3.6) of the field around which the free layer switches. The magnitude of the free layer coupling is important for the functioning of a spin valve, because a low offset for the free layer enables the GMR effect to take place at a lower field, which is desirable for device application. It was found that this coupling has a strong dependence on the copper interlayer thickness [24,25]. This copper dependence has been explained [26-31] as a superposition of three different mechanisms: ferromagnetic pinhole coupling, oscillating RKKY coupling and magnetostatic (Néel or orange-peel coupling). It is important that the Cu thickness is thin enough to give a good signal response, but thick enough to minimise coupling between the ferromagnetic layers. As already explained in section 3.1, the coupling between the ferromagnetic layers oscillates with increasing thickness, and it was found that the second antiferromagnetic peak (≈2nm Cu) gives a good compromise between high signal response and minimum coupling. The main advantages of exchange-biased spin-valve structures over GMR multilayers is the high sensitivity combined with the low coercivity. The sensitivity of an MR material is defined as the slope s≡(∂R/R)/∂H of the relative resistance versus field curve, obtained at a suitable working point. Sensitivities up to 18%/(kA/m) at room temperature have been reported for spin valves using permalloy (Ni80Fe20) for the free and pinned layers, and Cu for the spacer layer [32]. The AMR effect for these films gives a sensitivity of about 5%/(kA/m). A recent design for an integrated magnetic sensor is based on a Wheatstone bridge of micropatterned spin valve resistors with the addition of a soft adjacent layer structure that controls the principle characteristics by channelling the magnetic field. The design differs 55 CHAPTER 3 from earlier devices in that the local field direction rather than the local anisotropy of the spin valve is varied around the bridge. The design offers high sensitivity, small size, a relatively easy manufacturing process and potentially low price [77]. 3.3.1 Dependence on magnetic anisotropy. For most sensor applications, switching of the free layer should preferably be free of hysteresis. Rijks et al. [32], showed that this can be accomplished for spin valves within which the free layers have an in-plane, uniaxial anisotropy, with the easy axis perpendicular to the unidirectional anisotropy axis of the pinned layer. In such systems with crossed anisotropies, with the external field parallel to the biasing direction, the magnetisation of the free layer switches by a coherent rotation. In the parallel case, magnetisation reversal is the result of domain wall movement, leading to hysteresis [32]. As a result, strong Barkhausen noise will be superimposed on the sensor output signal due to pinning and depinning of domain walls. 3.3.2 Dependence on non-magnetic layer thickness and composition. Rijks et al. studied the dependence of the MR ratio on the Cu-layer thickness in spin valves with the composition (8nm Ni80Fe20/tCu nm Cu/6nm Ni80Fe20/8nm Fe50Mn50), grown on a 3nm Ta underlayer on Si (100) and covered with a 3nm Ta cap layer [24]. Fig. 3.7 Dependence of the MR curves and the MR ratio at room temperature on the Cu-layer thickness, for (Ni80Fe20/Cu/Ni80Fe20/Fe50Mn50) spin valves [24]. Layer thicknesses are given in the text. The dashed line gives a fit to the data, using Eqn. (4.2), for the Cu thicknesses for which full antiparallel alignment is obtained. 56 CHAPTER 3 Fig. 3.7 shows the results obtained. The MR ratio reaches a maximum of almost 5% at a critical thickness ≈2nm. Below this value the Cu spacer is not sufficiently thick to allow a perfectly antiparallel alignment of the magnetisation in the two layers, which leads to a decreased GMR ratio. A detailed analysis of the resistance and magnetisation curves, in terms of competing interlayer coupling and exchange biasing, has been carried out by Rijks et al. [24]. Speriosu et al. [33] have proposed the following expression for the dependence of the MR ratio of spin valves on spacer layer thickness: ( ) . /1 /exp 00 tt t R R R R NM NMNM + −   ∆=∆ l (4.2) The factor exp represents the probability that an electron which potentially contributes to the GMR effect by crossing through the NM layer, is not diffusely scattered in the NM layer. The denominator represents the shunting effect due to the NM layer. The parameter is related to the conduction electron mean free path ( NMNMt l/− NMl ) NMλ in the NM layer. The parameter t represents the conductance in the absence of the NM layer. Speriosu’s equation can be fitted to Rijk’s data as shown by the dotted line in Fig. 3.7. Dieny proposed that for systems of interest, l is approximately equal to 1/2 0 NM NMλ [34], leading to a ≈Cuλ 32nm. Dieny et al. [15] have investigated the dependence of the MR ratio on the composition of the NM spacer material. They measured the variation in the MR ratio with NM layer thickness for structures of the type Co/NM/Ni80Fe20/Fe50Mn50, (NM=Cu, Ag and Au). Fig. 3.8 Dependence on the noble metal spacer layer thickness of the MR ratio at room temperature for spin valves with the structure Si/7nm Co/ tNM nm NM/ 4.7nm Ni80Fe20/ 7.8nm Fe50Mn50/ 1.5nm NM. ( After Ref. [15] ). 57 CHAPTER 3 58 Fig. 3.8 shows that the MR ratio decreased exponentially for all the materials as the NM layer thickness was increased. The critical thickness above which fully antiparallel alignment is achieved is about 2nm for Cu and Au, and 4-5nm for Ag. Dieny et al. suggested that coupling through pinholes might be of importance up to much higher spacer thicknesses for Ag. The extrapolated MR ratio for tNM=0 is about 50% larger for Cu than for Au. This is thought to be due to a lower transmission of electrons for F/Au than for F/Cu interfaces. Dieny et al. also concluded that the higher resistance of Au explained the smaller decay length for Au than for Cu, 3.6nm and 4.5nm respectively. In [14], Dieny et al. have reported that for F/NM/F/Fe50Mn50 spin valve structures, with F=Ni80Fe20, the spin valve effect is also observed for NM=Pt, but is not seen for NM=Ta, Al, Cr and Pd. 3.3.3 Dependence on the ferromagnetic layer thickness and composition. Fig. 3.9 shows the results obtained by Dieny et al. [35] on the dependence of the GMR ratio on the free layer thickness for F=Co, NiFe and Ni. Fig. 3.9 Variation of the magnetoresistance versus the thickness of the “free” ferromagnetic layer M(1), with M(1)=Co, NiFe, or Ni at room temperature. The lines are two-parameter fits according to Eq. (4.3) given in text. (After Ref. [35]). The spin valve structure investigated was F/Cu/Ni80Fe20/Fe50Mn50. Dieny et al. proposed the following expression for the variation of the MR ratio with the F layer thickness: ( ) . 1 exp1 00 tt t R R R R F FF + −−   ∆=∆ l (4.3) The factor (∆R/R)0 is a function of the NM-layer thickness. The factor ( )[ ]FFt l−− exp1 is the angle-averaged probability for an electron with a large mean free path to be scattered within the ferromagnetic layer before being diffusely scattered at the outer boundary. This CHAPTER 3 shows that if the ferromagnetic layers are too thin, the contrast between the mean free paths for both spin directions will be reduced, because the electron spin with longer mean free path will be scattered at the outer boundary, thus reducing the MR ratio. Again, the parameter t0 depends on the conductance of the remainder of the structure (in the absence of the F layer). 3.3.4 Dependence on the exchange bias material and thickness. The antiferromagnetic pinning layer plays an important part in the function of spin valve structures. For example for read head applications the blocking temperature (temperature where the exchange field vanishes) should exceed 300°C, to prevent accidental de-pinning of the pinned layer during head fabrication or head life [36]. Exchange biasing decreases approximately linearly with increasing temperature, down to zero at the blocking temperature. The exchange energy should be large (>0.2mJ/m2), to withstand demagnetising fields at head level. Also the corrosion resistance of the exchange layer should not be worse than that of Ni80Fe20 used as reference [37]. Fig. 3.10 is a table comparing some exchange bias materials. materials Tb (°C) exchange energy (mJ/m2) corrosion resisitance requires anneal? Fe50Mn50 150 0.13 bad no NiO 190 <0.1 good no Mn76Ir24 300 0.2-0.4∗ fair yes ∗ Bottom spin valve after anneal. Fig. 3.10 Comparison of exchange bias materials. (After Ref. [36]). The variation of the exchange bias with the thickness of the antiferromagnetic layer changes depending on the specific system, the microstructure and the measurement temperature [38]. The general trend is that for thick AFM layers, e.g. over 20nm, the exchange field is independent of the thickness of the AFM layer. As the AFM thickness is reduced, the exchange field decreases abruptly and becomes zero when the AFM layer is a few nanometers thick. The dependence of the exchange bias for varying thickness of the antiferromagnetic layer in the system Fe80Ni20/FeMn is shown in Fig. 3.11. 3.3.5 Interfacial roughness and structural quality. The influence of interfacial roughness on the spin valve characteristics has been heavily debated, a good review is given in [39]. In Fe/Cr multilayers, some studies conclude that the MR amplitude is higher for rougher interfaces [40] up to an optimal roughness [41]. However, these results seem to depend on the growth conditions. Interfacial roughness plays various 59 CHAPTER 3 roles in influencing the spin valve MR amplitude depending on the combination of ferromagnetic and nonmagnetic transition metals under consideration. Fig. 3.11 Dependence of exchange bias (square symbols) and coercivity Hc (triangular symbols) with the AFM layer thickness for Fe80Ni20/FeMn at a fixed tFM=7nm [38]. In (Fe/Cr) or (Co/Cu) multilayers for which the interfacial spin dependent scattering is very important, some interfacial roughness may increase the density of scattering centres at the interfaces and therefore lead to a larger MR. In contrast in (NiFe/Cu) multilayers or spin valve sandwiches for which the interfacial spin dependent scattering is not as important as the bulk scattering, an increase in the interfacial roughness leads to a systematic decrease in the MR amplitude [42,43]. One reason for this decrease is the formation of some paramagnetic NiFeCu alloy layers at the NiFe/Cu interfaces. These paramagnetic layers provoke significant spin-flip scattering of the incoming or outgoing conduction electrons. This spin-flip scattering leads to a loss of the ‘spin memory’ of the electrons crossing the Cu spacer layer, and therefore reduces the MR amplitude. This effect is even more pronounced at Ni/Cu interfaces [35]. If the interfaces are too rough, the mean-free path of the conduction electrons become too small compared to the period of the multilayer, and a coherent interplay of spin dependent scattering between the successive magnetic layers cannot occur efficiently. In the opposite case when the interfaces are too flat, the specular reflection of the conduction electrons on the interfaces can lead to channelling of the electrons within each layer. 60 The structural quality of the multilayer certainly influences the proportion of antiparallel alignment between the magnetisations in the successive magnetic layers. In particular, the CHAPTER 3 61 existence of pin-holes through the nonmagnetic spacer may significantly reduce the MR amplitude by inducing a local ferromagnetic coupling between the magnetic layers [3, 44-46]. 3.3.6 Temperature dependence. The variation of the MR spin valve amplitude with temperature has been investigated by various groups [47-49, 35]. Generally the MR amplitude decreases monotonically with temperature by a factor of the order of 1.5-3, depending on the system under consideration, between 4K and room temperature. The decrease in spin valve amplitude as the temperature is increased is due to two main factors: (i) the intermixing of the spin ↓ and spin ↑ currents caused by scattering in the bulk of the ferromagnetic layers, or by paramagnetic fluctuations at the FM/NM interfaces. A clear correlation has been established between the Curie temperature of a ferromagnet and the slope of the decrease in the spin valve MR versus temperature [37, 51]. The higher the Curie temperature, the less the thermal variation of the spin valve MR. Structures based on Co- (Tc=1130°C) or Fe- (Tc=770°C) rich ferromagnetic materials should therefore give less thermal variation of the spin valve MR than Ni-rich alloys (Tc=358°C). (ii) The phonon scattering (especially in the non-magnetic layer), dilutes the spin dependent scattering which gives rise to the spin valve MR and prevents the exchange of conduction electrons between the successive ferromagnetic layers through the spacer. 3.3.7 Other exchange biased spin valve structures. Several modifications to the simple, Fe-Mn-biased spin valve structures have been reported to yield higher GMR ratios. Parkin demonstrated that a very thin Co layer at the Ni80Fe20/Cu interfaces in (Ni80Fe20/Cu/Ni80Fe20/Fe50Mn50) spin valves (interface “dusting”) may lead to a drastic increase in the MR ratio [50]. As shown in Fig. 3.12, the MR ratio for the structure (5.3nm Ni81Fe19/3.2nm Cu/2.2nm Ni81Fe19/9nm Fe50Mn50/1nm Cu) was found to increase (at room temperature) from 2.9% in the absence of Co, up to 6.4% for Co thicknesses above approximately 0.6nm. Conversely, adding permalloy layers of a thickness of only 0.4nm to the Co/Cu interfaces of Co/Cu/Co/FeMn spin valves was found to decrease the MR ratio from 6.8% to 3.9%. The effect of the Co layers was strongly localised to the interfaces, by varying the distance d to the interface of 0.5nm Co layers that were buried in the permalloy layers: no significant increase of the MR ratio was found for d>0.6nm. Parkin’s experiment demonstrated the crucial role player by the interfaces in a spin valve structure. CHAPTER 3 Fig. 3.12 Dependence on the Co interface layer thickness tCo of the MR ratio for spin valves with the structure (Si/[5.3-tCo]nm Ni80Fe20/tConm Co/3.2nm Cu/[2.2-tCo]nm Ni80Fe20/9nm Fe50Mn50/1nm Cu) [50]. Fig. 3.13 Normalised pinning field versus temperature for spin valves pinned by four different antiferromagnetic materials. [51]. Although much effort has gone into improving spin valve design, there are intrinsic problems which are difficult to address. One of the problems is the blocking temperature, Fig. 3.13 shows the normalised pinning field versus temperature for spin valves pinned by four different antiferromagnetic materials [51]. The NiMn antiferromagnetic layer shows an improved normalised pinning field over a wider temperature range compared to NiO, IrMn and FeMn. Synthetic spin valves, which were first suggested by Berg et al. [52] offer increased exchange fields and blocking temperatures. They consist of an antiferromagnetically coupled (FM/NM) bilayer structure instead of a single AFM layer. The choice of the spacer layer thickness in the 62 CHAPTER 3 (FM/NM) stack enables antiferromagnetic coupling between the ferromagnetic layers. Fig. 3.14 shows a schematic diagram of a synthetic SV structure. M2 M1 Decoupler Soft detection layer Substrate FM/NM bilayer structure Fig. 3.14 Schematic diagram of a synthetic spin valve structure. The GMR bilayer structure stack at the bottom shows strong interlayer exchange coupling and defines the magnetisation direction of the pinned ferromagnetic layer. The free layer is decoupled from the pinned layer by the relatively thick spacer in the sandwich, enabling a spin valve response. An optimum GMR response is observed when the angle between the magnetisation in the ferromagnetic layers changes from completely parallel to antiparallel in an applied field. Hence the formation of domain walls or magnetisation ripple will reduce the GMR response. Synthetic spin valves show a higher thermal stability, because the antiparallel magnetisation alignment will only be destroyed when the temperature is above the Curie temperature of the FM layers, which is generally higher than the blocking temperature in AFM materials. Another advantage is that the effective moment of the pinned layer is low, and therefore the contribution to the demagnetising and coupling fields acting on the free layer is much weaker than in conventional spin valves. Synthetic spin valves have also been incorporated in dual (symmetric) spin valve structures [53]. An optimum GMR response is observed when the angle between the magnetisation in the ferromagnetic layers changes from completely parallel to antiparallel in an applied field. Hence the formation of domain walls or magnetisation ripple will reduce the GMR response. 3.4 Pseudo Spin Valve Structures and MRAM. A pseudo spin valve (PSV), unlike an exchange-biased spin valve, does not contain an AFM layer. The different switching fields for the two magnetic layers arise from dissimilar properties. The dissimilar properties can be induced by a difference in geometrical, materials, or structural factors, such as thickness, material type, or deposition conditions. A variety of PSV designs have been studied and this section will give an overview of some of the designs investigated. 63 CHAPTER 3 64 (b)(a) Fig. 3.15 (a) Hysteresis loop and magnetoresistance at room temperature of a sample with structure Si/100Å Ta/40Å NiFe/60Å Cu/40Å NiCo/50Å Ta. (b) Hysteresis loop and magnetoresistance at room temperature of a sample with structure Si/30Å Fe/62Å Ag/30Å Co/62Å Ag. [14] . Dieny et al. showed that a spin valve response could be obtained from samples consisting of ferromagnetic layers with different coercivities [14]. Fig. 3.15 (a) shows the magnetisation curves and resistivity variations for samples with the structures: Si/100Å Ta/40Å NiFe/60Å Cu/40Å NiCo/50Å Ta and Si/30Å Fe/62Å Ag/30Å Co/62Å Ag. In Fig. 3.15 (a), NiFe has a lower coercivity than NiCo, while in Fig. 3.15 (b) Fe has a lower coercivity than Co. When the field is swept from the positive to the negative saturations of the structure, the magnetic configuration changes from parallel to antiparallel magnetic alignment between the two coercive fields associated with the two types of magnetic layers. The shape of the MR is intimately related to the magnetic hysteresis loops. The MR curve can exhibit a well defined plateau if the hysteresis loops associated with the two types of ferromagnet are well separated (Fig. 3.15 (a)), or a well rounded maximum if they tend to overlap (Fig. 3.15 (b)). In this type of structure, the fields required to obtain the full spin valve MR amplitude are usually much lower than in antiferromagnetically coupled multilayers. These fields are of the order of the saturation field of the softer of the two ferromagnetic materials (typically a few Oe or tens of Oe). However, in comparison to antiferromagnetically biased spin valves, the GMR ratio tends to be much smaller (≈2%). Recent work by Schmidt et al [54] exploited the idea of designing a pseudo spin valve structure using ferromagnetic layers with very different coercivities, by investigating the use of CoTi as an ultra-soft material. The full structure was: (Ti2Å/Co8Å)2/Co 25Å/Cu- dCu/Co20Å), (8Å≤dCu≤38Å). This structure exploits the difference in coercive fields of pure CHAPTER 3 Co and the Co/Ti multilayer, and an additional Co 25Å layer was added to avoid problems with poor interface quality caused by Ti impurity. A magnetoresistance response was observed when dCu>24Å, with MR values reaching 5% with a sensitivity of 2%/Oe. Fig. 3.16 Low field magnetoresistance of Co-Cu-Co-Cu-Cr spin valves of different layering orders. [61]. Some interesting work has been carried out by Ross et al. on nano-sized pseudo spin valve dots and wire arrays [55-61]. Interference lithography was used to fabricate the pseudo spin valves with the structures: Co/Cu/Co, and NiFe/Cu/Co. It is well known that the coercive field of a thin film is inversely proportional to the thickness [62] and the first structure relies on the different coercivities due to the different thicknesses of the ferromagnetic layers to produce a spin valve effect. The latter structure relies on the different coercivities of the ferromagnetic materials. Fig. 3.16 shows the low field electrical response for Co/Cu/Co pseudo spin valve nano-dots patterned with 100nm diameter and 200nm period. As shown, the system has a very rapid low field response in MR at the switching field of 100Oe. In particular, the MR for the PSV switches from the high field to the almost full maximum MR within a narrow field range <10Oe. The maximum MR value is about 6.5%. Fig. 3.16 also shows the MR response of a similar PSV but with reverse ordering of the two Co layers. Comparing with the first PSV the only difference is that the thinner Co layer is deposited first. However, the MR response is dramatically different; the low field switching field is increased by almost a factor of three and the MR switching is more gradual. The magnetic properties of the more sensitive structure was analysed, and the results showed that the rapid low field response was due to the magnetisation reversal of the thicker ferromagnetic layer. 65 CHAPTER 3 Fig. 3.17 Magnetoresistance (solid) and hysteresis (open circles) measured on an array of 250µm long, 60nm wide wires patterned from a 6nm NiFe/3.7nm Cu/3nm Co/3nm Cu film. The field was applied parallel to the wires. [55]. Fig. 3.17 shows an array of wires patterned from a NiFe/Cu/Co/Cu multilayer with widths of 60nm and 250µm length. The wires were patterned using interference lithography and argon ion milling and 940 wires were patterned. When a field is applied parallel to the wires, the resistance follows the magnetic hysteresis loop, having a higher value when the magnetisation directions of the NiFe and Co layers are antiparallel. Further work by Castaño et al. [56] investigated how the spin valve response changes as the width of the nanopatterned pseudo spin valve wires varied from 60-150nm. The starting PSV thin film consisted of sputtered NiFe (6nm)/Cu (3.7nm)/Co (3nm)/Cu (3nm), exhibiting a room temperature giant magnetoresistance ratio of 2.5%. The effects of reducing the width of the wires are a monotonic decrease in the GMR ratio (from 2.55% to 0.74%) and the saturation magnetisation, and an increase of both the resistivity of the wires and the average switching fields of the magnetic layers. On cooling the samples to 77K, the resistivity decreased slightly and the GMR amplitude increased independently of the width of the wires. The authors suggested the presence of a disordered region at the edges of the wires as a result of processing, which increased the resistivity and decreased the saturation magnetisation as the wire width decreased. McMichael et al. used an obliquely sputtered Ta underlayer to produce a high anisotropy in one of the ferromagnetic layers in a PSV design. Anisotropy fields in excess of 1500Oe were produced in 3-5nm thick polycrystalline films of Co, many times larger than the typical 66 CHAPTER 3 67 anisotropy of polycrystalline Co films in this thickness range, and large enough for application as a robust “pinned” film in a spin valve. It has been known for many years that the oblique deposition of magnetic materials can produce uniaxial anisotropy in magnetic films [64-70]. Fig. 3.18 Magnetoresistance of a spin valve measured for fields applied at φ≈90°, near the hard axis. The inset shows the magnetoresistance curve for φ≈0°. [63]. The source of the anisotropy is an anisotropic roughness that develops perpendicular to the incidence direction during oblique deposition of the Ta underlayers. Co films deposited normally on this underlayer experience large magnetostatic fields when magnetised perpendicular to the corrugations. Fig. 3.18 shows the magnetoresistance of the 7.5nmTa/3nmCo/4nmCu/3nmCo pseudo spin valve for fields applied at φ≈90°, where φ is the angle between the applied field and the anisotropy axis. The inset shows the magnetoresistance curve for φ≈0°, and the results show a maximum MR ratio of 3.5%. 3.4.1 Magnetic Random Access Memories (MRAMs). There is a strong interest in non-volatile memory devices based on magnetic materials, Magnetic Random Access Memories (MRAMs), due to their non-volatile characteristic, radiation hardness, non-destructive read-out, low-voltage, and very large (>1015) read-write cycle capability [71]. MRAMs can be as fast as Dynamic Random Access Memories (DRAMs), and almost as small as Static Random Access Memories (SRAM) in cell size. To compete with CMOS embedded memories, they must be fabricated with <125nm features, bringing several technological issues regarding micromagnetics and fabrication issues for deep submicron elements. MRAMs also compete with Ferroelectric Random Access CHAPTER 3 Memories (FERAMs) for non-volatile memories. In this case power consumption is a major issue. They compete directly with Flash memories used where speed is not a major concern (i.e. in some storage applications). Write speed for Flash technology is in the µs range in comparison to a few ns for MRAMs. In the mid 1980s an MRAM concept was developed at Honeywell which has some common features with most modern version [72]: • Writing data using magnetic hysteresis. • Reading data using magnetoresistance of the same body where data is stored. • Memory cells integrated on an integrated circuit chip. The structure relied on sensing the AMR signal of a cobalt-permalloy alloy with normal AMR ratio of about 2%. The maximum differential resistance of the cell between ‘1’ and ‘0’ when it was read was about ¼ of the 2% magnetoresistance, or about 0.5%. In real arrays with practical sense current, this gave differential sense signals of 0.5 to 1.0mV. These sense signals allowed 16K bit integrated MRAM chips to operate with read access time of about 250ns [73]. Write times for the MRAM was 100ns and could have been faster if needed. The discovery of GMR multilayer materials gave hope for higher signals and faster read access rime. Magnetic films sandwiching a copper layer and etched into stripes showed a magnetoresistance ratio of about 6% and read access times of under 50ns [73]. However, GMR multilayer MRAM cells had serious limitations; the MRAM read access time was slower than for semiconductor memory because of the low MRAM sense signal, and the cell size was limited to 1µm due to magnetisation curling along the edges of the stripe. Hword θ2 θ1 Isense Isense “0” Storage layer “1” Iword Fig. 3.19 Pseudo spin valve cell. (Adapted from Ref [73].) The invention of the pseudo-spin valve (PSV) cell [74] significantly improved signal levels, thus improving the read access time of MRAM while maintaining densities competitive with other solid state memory technologies. Not only was nearly all of the ≈6% GMR available, 68 CHAPTER 3 69 but also the signal swing was plus or minus 6% making the difference between a ‘0’ and ‘1’ about 12% of the cell resistance. This gave eight times improvement over the original mode of operation and put MRAM on a much more even footing with semiconductor memory or read access time. Fig. 3.19 illustrates the construction of a PSV cell, where the two magnetic layers have mismatched properties. In this case, the thinner film switches at lower fields, or is the ‘soft’ film and the thicker film switches at a higher field and is the ‘hard’ film. Without switching the hard film, the soft film can be manipulated to be parallel or antiparallel to the hard film. With a sequence of word fields, which starts with a negative field and ends with a positive field, the resistance either rises or falls, depending on whether a ‘1’ or a ‘0’ is stored. With simple electronics, the difference between the initial and final resistances can be sensed and the polarity of this difference indicates whether a ‘1’ or ‘0’ is stored. PSV memory cells can be at least as narrow as 0.2µm using a 2D memory organisation, and may find initial applications as a replacement for EEPROM or flash memory where high density or fast writing is important. Spin Dependent Tunnelling (SDT) devices provide a higher percentage magnetoresistance than PSV structures and therefore have the potential for higher signals and higher speed. Recent results indicate SDT tunnelling giving over 40% magnetoresistance [75, 76], compared to 6-9% magnetoresistance in good PSV cells. One of the major disadvantages of SDT devices is the resistance of the small tunnelling cells tends to be at least several 1000s of ohms and they are subject to dielectric breakdown at the 1V to 2V level. Therefore currents of more than 1mA through the devices is not practical and the currents used to sense the state of the SDT cell probably cannot be used to aid in the switching of the cell, unlike for the PSV cell. Measurements have indicated that low time constants (<1ns) can be attained for PSV cells [75]. The intrinsic speed of SDT elements configured into a DRAM type architecture or a flip-flop like cell should provide signals of 30-40mV, which is comparable to semiconductor memory cells signal levels and should thus run at comparable speeds. SDT memory devices show promise for high performance non-volatile applications. In summary, MRAM has the potential to be as fast and dense as DRAM with the additional advantage of non-volatility. Compared with flash and EEPROMs, MRAM writes much faster and does not deteriorate with millions of write cycles. Ferroelectric RAM (FRAM) is like MRAM in that it is non-volatile and fast write and there have been some limited FRAM applications. While FRAM is a competitor to MRAM, it is likely that MRAM can be denser. CHAPTER 3 References [1] M.N. Baibich, J.M. Broto, A. Fert, F. Nguyen-Van-Dau, F. Petroff, P. Etienne, G. Creuzet, A. Friedrich, J. Chazelas: Phys. Rev. Lett.,61, 2472 (1988). [2] S. S.P. Parkin, N. More, K.P. Roche: Phys. Rev. Lett., 64, 2304 (1990). [3] S.S.P. Parkin, Z. G. Li, and D. J. Smith, Appl. Phys. Lett., 58, 2710 (1991). [4] S.S.P. Parkin, Phys. Rev. Lett., 67, 3598 (1991). [5] S.S.P. Parkin, R. Bhadra, and K. P. Roche, Phys. Rev. Lett., 66, 2152 (1991). [6] S.S.P. Parkin, Ultrathin Magnetic Structures II, Springer-Verlag Berlin Heidelberg, Chapter 2.4, 148-194 (1994). [7] D.M.Edwards, J. Mathon, R.B. Muniz, and M.S. Phan, Phys. Rev. Lett., 67, 493-496 (1991). [8] P. Lang, L. Nordström, R. Zeller, and P.H. Dederichs, Phys. Rev. Lett., 71, 1927-1930 (1993); L. Nordström, P. Lang, R. Zeller, and P.H. Dederichs, Phys. Rev. B, 50, 13058-13061 (1994). [9] R. Coehoorn, Phys. Rev. B, 44, 9331-9337 (1991). [10] P. Bruno and C. Chappert, Phys. Rev. Lett., 67, 1602-1605, 2592 (1991); P. Bruno and C. Chappert, Phys. Rev. B, 46, 261-270 (1992). [11] N. Mott, Proc. Roy. Soc. 156, 368 (1936). [12] J. Mathon, Spin Electronics, Springer-Verlag Berlin Heidelberg, Chapter 4, 71-88, (2001). [13] B. Dieny, V.S. Speriosu, S.S.P. Parkin, B.A. Gurney, D.R. Wilhoit, and D. Mauri, Phys. Rev. B 43, 1297 (1991). [14] B. Dieny, V.S. Speriosu, B.A. Gurney, S.S.P. Parkin, D.R. Wilhoit, K.P. Roche, S. Metin, D.T. Peterson, and S. Nadimi, J. Magn. Magn. Mater. 93, 101 (1991). [15] B. Dieny, V.S. Speriosu, S. Metin, S.S.P. Parkin, B.A. Gurney, P. Baumgart, and D.R. Wilhoit, J. Appl. Phys. 69, 4774 (1991). [16] R. Coehoorn, J.C.S. Kools, Th.G.S.M. Rijks and K.–M.H. Lenssen, Philips J. Res. 51, 93-124, (1998). [17] T.C. Anthony, J.A. Brug and S. Zhang, IEEE Trans. Magn. 30, 3819 (1994). [18] W.F. Egelhoff, Jr. et al. J. Appl. Phys. 78, 273 (1995). [19] P.M. Baumgart, B. Dieny, B.A. Gurney, J.P. Nozières, V.S. Speriosu and D.R. Wilhoit, U.S. Patent 5287238 (1994). [20] S. Noguchi, R. Nakatani, K. Hoshino and Y. Sugita, Jpn. J. Appl. Phys. 33, 5734 (1994). [21] W.F. Egelhoff, Jr. et al. J. Appl. Phys. 79, 5277 (1996). [22] W.F. Egelhoff, Jr. et al. J. Appl. Phys. 82, 6142 (1997). [23] Y. Sugita, Y. Kawawake, M. Satomi, H. Sakakima, Jpn. J. Appl. Phys. 37, 5984 (1998). 70 CHAPTER 3 [24] Th. G.S.M. Rijks, R. Coehoorn, J.T.F. Daemen, and W.J.M. de Jonge, J. Appl. Phys. 76, 1092 (1994). [25] K. Hoshino, S. Noguchi, R. Nakatani, H. Hoshiya, and Y. Sugita, Jpn. J. Appl. Phys. Part 1 33, 1327 (1994). [26] J.C.S. Kools, J. Appl. Phys. 77, 2993 (1995). [27] J.C.S. Kools, Th. G.S.M. Rijks, A.E.M. De Veirman and R. Coehoorn, J. Appl. Phys. 31, 3918 (1995). [28] J.L. Leal and M.H. Kryder, J. Appl. Phys. 79, 2801 (1996). [29] J.L. Leal and M.H. Kryder, IEEE Trans. Magn. 32, 4642 (1996). [30] D. Wei and H.N. Bertram, IEEE Trans. Magn. 32, 3434 (1996). [31] C.M. Park, K.I. Min, and K.H. Shin, IEEE Trans. Magn. 32, 3422 (1996). [32] Th. G.S.M. Rijks, W.J.M. de Jonge, W. Folkerts, J.C.S. Kools, R. Coehoorn, Appl. Phys. Lett. 65, 916 (1994). [33] V. Speriosu, J.P. Nozières, B.A. Gurney, B. Dieny, T.C. Huang, H. Lefakis, Phys. Rev. B 47, 11579 (1993). [34] B. Dieny, J. Magn. Magn. Mater. 136, 335 (1994). [35] B. Dieny, P. Humbert, V.S. Speriosu, S. Metin, B.A. Gurney, P. Baumgart and H. Lefakis, Phys. Rev. B 45, 806 (1992). [36] P.P. Freitas, Spin Electronics, Springer-Verlag Berlin Heidelberg, Chapter 19, (2001). [37] A. Veloso, N.J. Oliveira, and P.P. Freitas, IEEE Trans. Magn.,34, 2343 (1998). [38] J. Nogués, I.K. Schuller, J. Magn. Magn. Mater. 192, 203 (1999). [39] A. Fert and P. Bruno, Ultrathin Magnetic Structures II, Springer-Verlag Berlin Heidelberg, Chapter 2.2, 110 (1994). [40] E.E. Fullerton, D.M. Kelly, J. Guimpel, I.K. Schuller, Y. Bruynseraed, Phys. Rev. Lett., 68, 859 (1992). [41] F. Petroff, A. Barthelemy, D.H. Mosca, D.K. Lottis, A. Fert, P.A. Schroeder, W.P. Pratt, R. Loloee, S. Lequien, Phys. Rev. B.,44, 5355 (1991). [42] T.C. Huang, J.P. Nozières, V.S. Speriosu, B.A. Gurney, H. Lefakis, Appl. Phys. Lett. 62, 1478 (1993). [43] V.S. Speriosu, J.P. Nozières, B.A. Gurney, B. Dieny, T.C. Huang, H. Lefakis, Phys. Rev. B.,47, 11579 (1993). [44] R.J. Highmore, W.C. Shih, R.E. Somekh, J.E. Evetts, J. Magn. Magn. Mater. 116, 249 (1992). [45] Y. Saito, S. Hashimoto, K. Inomata, Appl. Phys. Lett. 60, 2436 (1992). [46] J. Kohlepp, S. Cordes, H.J. Elmers, U. Gradmann, J. Magn. Magn. Mater. 111, 231 (1992). 71 CHAPTER 3 [47] D.H. Mosca, F. Petroff, A. Fert, P.A. Schroeder, W.P Pratt, R. Loloe, J. Magn. Magn. Mater. 94, L1 (1991). [48] B. Dieny, V.S. Speriosu, S. Metin, Europhys. Lett., 15, 227 (1991). [49] A. Vedyayev, B. Dieny, N. Ryzhanova, J.B. Genin, C. Cowache, Europhys. Lett., 25, 465 (1994). [50] S.S.P. Parkin, Phys. Rev. Lett., 71, 1641 (1993); S.S.P. Parkin, Appl. Phys. Lett. 61, 1358 (1993). [51] S.N. Mao, N. Amin, E. Murdock, J. Appl. Phys. 83 (11Pt2), 6807 (1998). [52] H.A.M. van der Berg et al., IEEE Trans. Magn.,32 (5Pt2), 4624 (1996). [53] A. Tanaka, Y. Shimizu, H. Kishi, K. Nagasaka, H. Kanai, M. Oshiki, IEEE Trans. Magn.,35, 700 (1999). [54] M. Schmidt, F. Stobiecki, B. Szymański, Phys. Stat. Sol. (a) 196 (1), 56 (2003). [55] C. A. Ross, S. Haratani, F.J. Castaño, Y. Hao, M. Hwang, M. Shima, J.Y. Cheng, B. Vögeli, M. Farhoud, M. Walsh, H.I. Smith, J. Appl. Phys 91 (10), 6848 (2002). [56] F.J. Castaño, S. Haratani, Y. Hao, C. A. Ross, H.I. Smith, Appl. Phys. Lett. 81 (15), 2809 (2002). [57] X. Zhu, P. Grütter, Y. Hao, F.J. Castaño, S. Haratani, C. A. Ross, B. Vögeli, H.I. Smith, J. Appl. Phys 93 (2), 1132 (2003). [58] N. Dao, S.L. Whittenburg, Y. Hao, C. A. Ross, L.M. Malkinski, J.Q. Wang, J. Appl. Phys 91 (10), 8293 (2002). [59] F.J. Castaño, Y. Hao, S. Haratani, C. A. Ross, B. Vögeli, M. Walsh, H.I. Smith, IEEE Trans. Magn., (2000). [60] F.J. Castaño, Y. Hao, C. A. Ross, B. Vögeli, H.I. Smith, S. Haratani, J. Appl. Phys 91 (10), 7317 (2002). [61] J.Q. Wang, L.M. Malkinski, Y. Hao, C. A. Ross, J.A. Wiemann, C.J. O’Connor, Mater. Sci. and Eng., B76, 1 (2000). [62] J. H. Fluitman, Thin Solid Films, 16, 269 (1973). [63] R.D. McMichael, C.G. Lee, J.E. Bonevich, P.J. Chen, W. Miller, W.F. Egelhoff, Jr., J. Appl. Phys 88 (9), 5296 (2000). [64] D.O. Smith, J. Appl. Phys 30, 264S (1959). [65] T.G. Knorr, R.W. Hoffman, Phys. Rev., 113, 1039 (1959). [66] Y. Hoshi, E. Suzuki, M. Naoe, J. Appl. Phys 79, 4945 (1996). [67] K. Itoh, K. Hara, M. Kamiya, K. Okamoto, T. Hashimoto, H. Fujiwara, J. Magn. Magn. Mater. 148, 132 (1995). [68] J.M. Alameda, F. Carmona, F.H. Salas, L.M. Alvarez-Prado, R. Morales, G.T. Pérez, J. Magn. Magn. Mater. 154, 249 (1996). 72 CHAPTER 3 73 [69] K. Tanaka, S. Yamauchi, E. Makino, T. Fujii, M. Inoue, M. Izaki, IEEE Trans. Magn.,35, 2916 (1999). [70] M. Michijima, H. Hayashi, M. Kyoho, T. Nakabayashi, T. Komoda, T. Kira, IEEE Trans. Magn.,35, 3442 (1999). [71] G.B. Granley, T. Hurst, Sixth Biennial IEEE International Non-Volatile Memory Technology Conference Proceddings, pg 138 (1996). [72] J. Daughton, Thin Solid Films, 216, 162 (1992). [73] J.M. Daughton, A.V. Pohm, R.T. Fayfield, C.H. Smith, J. Phys. D: Appl. Phys., 32, R169 (1999). [74] A. Pohm, B. Everitt, R. Beech, J. Daughton, IEEE Trans. Magn.,33, 3280 (1997). [75] S. Parkin, Intermag Conf. (Kyongiu, Korea, 18-21 May 1999) paper GA-01. [76] S. Carsoso, V. Gehanno, R. Fereira, P. Freitas, (Kyongiu, Korea, 18-21 May 1999) paper FC-03. [77] J.L.Prieto, N.Rouse, N.K. Todd, D. Morecroft, J. Wolfman, J.E.Evetts, M.G. Blamire. Sensors & Actuators A, 94, 64 (2001). CHAPTER 4 Any intelligent fool can make things bigger and more complex….It takes a touch of genius---and a lot of courage to move in the opposite direction. Albert Einstein. Chapter 4 Experimental Techniques 4.1 Device microfabrication 4.2 Device nanofabrication 4.3 Device characterisation 74 CHAPTER 4 This chapter will describe the experimental techniques used in this thesis. The first part will describe the device microfabrication including optical lithography, argon ion milling and d.c. sputter deposition. The second part will describe the device nanofabrication including r.f. sputter deposition and Focused Ion Beam (FIB) lithography. The final part will describe the various characterisation techniques used including electrical characterisation using four-point resistance measurements, magnetic characterisation using the Vibrating Sample Magnetometer (VSM) and structural characterisation using Atomic Force Microscopy (AFM). Magnetic domain imaging using the Bitter technique is also described at the end of the chapter. 4.1 Device microfabrication. Microfabrication of the ferromagnetic thin films and pseudo spin valves was carried out using optical lithography in the group clean room, and the masks were designed using AutoCAD. High resolution optical exposure was done in a Karl Suss contact mask aligner (Karl Suss MJB 3), using a UV light-source. An argon ion miller was used to transfer the structure into the sample, and sputter deposition was used to deposit copper for electrical contact. Fig. 4.1 shows the typical processing steps involved in the fabrication of an array of microwires across the surface of the chip, including electrical contacts. The left part of the diagram shows the mask design (a), and the cross-section across the width is shown on the right, (b). Step 1 involves ion milling away the edges of the chip. This enables better contact between the mask and the chip for the following lithography steps. Step 2 is another milling step to fit the size of the spin valve to the size of the wire array. In step 3 copper contact pads are deposited by sputter deposition and liftoff. Holes are patterned in the photoresist where the contact pads are required. After exposure and before developing the sample is submersed in chlorobenzene for two minutes and then dried using an air gun. The chlorobenzene hardens the surface of the photoresist and creates a ledge profile, which aids the lift-off procedure [1]. The chamber was pumped to below 5x10-5mbar, and before sputtering the copper target was pre-sputtered to clean the surface. The copper was then deposited over the entire surface of the film using D.C. sputter deposition. The copper was sputtered at a pure argon pressure of 2x10-3mbar and 40W power, the copper deposition rate at this power is ≈1.56nm/s. After deposition, the resist was stripped with acetone leaving the copper contact pads. The sample holder was water cooled throughout the deposition. 75 CHAPTER 4 Step 1 w w Step 2 Step 3 Step 4 (a) (b) photoresist substrate copper spin valve Fig. 4.1 The four steps involved in the fabrication of an array of micro-wires across the surface of the sample with electrical contacts. Steps 1, 2 and 4 involve lithography and milling. Step 3 involves lithography and lift off of the sputter deposited of copper contact pads. Argon Ion Milling. After the resist was patterned, ion milling was used to physically etch away the unwanted areas of the film. Ion milling is strictly a mechanical process, it involves no chemical reactions with the sample because it uses a noble gas. It relies on striking a plasma to produce the etch species, and the plasma is initiated by applying a voltage across a gap containing the low-pressure argon gas. The Kaufman ion gun was powered with a Princeton Applied 76 CHAPTER 4 Research power supply, and the chamber was pumped down to less than 4x10-6mbar using a diffusion pump prior to milling. Argon gas with a 2% oxygen mixture was used to mill the samples at a pressure of 2x10-4mbar. The oxygen enhances the etch rate by oxidising the debris from the areas of film being removed. The beam current of the Ar+ ions was 10mA with a 500V accelerating voltage. The sample holder was water cooled and rotated during milling to improve the uniformity across the surface. The milling rate was calibrated using Atomic Force Microscopy (AFM), and was found to be approximately 10nm/min for NiFe. (c) Ar+ Redeposition Photoresist (b) Photoresist Trench Ar+ (a) Substrate erosion Photoresist erosion Ar+ Fig. 4.2 Problems that may occur during ion milling: (a) mask taper transfer, (b) trenching, (c) re-deposition of the etched material. Fig. 4.2 shows some of the problems that occurred during milling of the highly anisotropic wire arrays. Since the process erodes the mask, any taper in the masking layer will be transferred to the pattern. After etching is finished and the photoresist mask is removed, the resultant pattern is broadened, Fig. 4.2 (a). This problem can be minimised by optimising the contact during lithography and rotating during milling. An enhanced erosion rate at the edge 77 CHAPTER 4 of the pattern is called trenching, and occurs when mask erosion causes the sidewalls of the pattern to be tapered at a steep angle. Some of the low angle ions will reflect off at the tapered surface towards the pattern edge, where they cause trenches, Fig. 4.2 (b). The etched materials are not volatile and tend to be re-deposited on any surface where they come into contact, e.g. the side-wall of the patterned wire, even after the photoresist has been removed, Fig. 4.2 (c). This problem can also be minimised by rotating during milling. When the film is rigid enough the side-wall can be removed by gently scrubbing with a cotton bud. Also, although the re- deposited material is evident in the AFM images, it contributes a small fraction of the total signal from the wire array. D.C Sputter deposition. Fig. 4.3 shows a schematic diagram of the sputtering system. The sputtering flange is positioned directly above the sample holder, and a rotary pump is used to rough the system and back the diffusion pump. Sputtering relies on striking a plasma between the anode (sample holder) and cathode (target) to produce the reactive species. Once the plasma is formed, ions in the plasma are accelerated toward the target and strike the surface. When an energetic ion strikes the surface of the target, four things can happen as shown in Fig. 4.4. Gas inlet Roughing valve Sample holder To Rotary Pump Diffusion pump Gate valve Backing valve Sputtering flange Fig. 4.3 A schematic picture of D.C. sputtering system. 78 CHAPTER 4 + Reflected ions and neutrals Secondary electrons Sputtered atoms Incident ion Substrate damage Surface Implantation Sputtering Fig. 4.4 Possible outcomes for an ion incident on the surface of a wafer. (After Ref. [2]). Ions with very low energy will simply bounce off the surface. At energies less than about 10eV, the ion may adsorb to the surface, giving up its energy to phonons (heat). At energies above about 10eV, the ion penetrates into the material, transferring most of its energy deep into the target, where it changes the physical structure. These energies are typical for ion implantation. Between these two extremes, both energy transfer mechanisms occur; part of the ion energy is deposited in the form of heat, and the remainder goes into the physical rearrangement of the target material. Most of the energy transfer occurs within several atomic layers, and the atoms and clusters of atoms are ejected from the surface of the material. The atoms and clusters of atoms ejected escape with energies of 10 to 50eV. This energy provides sputtered atoms with additional surface mobility for good step coverage. Planar magnetron sputtering was used, which causes the electrons to spiral around the direction of the magnetic field lines. The orbital motion of the electrons increases the probability that they will collide with neutral species and create ions. The increased ion density increases the rate of ion bombardment of the target. A resolution of ≈1µm could be achieved by optical lithography. Although the UV light produces a wavelength of ≈400nm, the resolution is limited by the minimum mask feature size, the contact between the mask and the sample and the thickness of the photoresist. 4.2 Device nanofabrication. The samples were initially prepared for Focused Ion Beam (FIB) nanofabrication using optical lithography, ion milling, D.C. sputter deposition and R.F. sputter deposition. Fig. 4.5 79 CHAPTER 4 shows the sample preparation, which involves 4 steps: (1) optical lithography and milling to pattern the mesostructures, (2) optical lithography and D.C. sputter deposition to deposit contact pads, (3) optical lithography, R.F. sputter deposition and lift off to deposit the silica hard mask, and finally (4) D.C. sputter deposition of a gold over-layer, which enabled imaging in the FIB. Step 4 sample copper silica gold No additional lithography step for step 4 contact pads negative mask for lift off Ferromagnetic mesostructures Step 2 Step 1 Step 3 Fig. 4.5 The four steps involved in preparation of the ferromagnetic mesostructures for FIB nanofabrication. The schematic diagram shows the cross-section of the sample for each step, the mask design is shown on the right. Each step involves optical lithography, step 1 involves ion milling, step 2 D.C. sputter deposition, step 3 R.F. sputter deposition, and step 4 D.C. sputter deposition. The left part of the diagram shows the cross-section for each step, and the right shows the mask design. After depositing the copper contact pads and before depositing silica, the samples were characterised using transport measurements as described in section 4.4. After depositing silica and before carrying out the lift off, a gold layer was deposited on top of the silica. The chamber was pumped down to 4x10-6mbar using a diffusion pump, and the target was pre-sputtered for 3mins. Deposition was carried out at 2x10-3mbar and 40W power, the 80 CHAPTER 4 gold deposition rate at this power is 25nm/min. Approximately 30nm of gold was deposited on top of the silica. Afterwards, lift of was carried out to remove the unwanted silica and gold. Radio Frequency (R.F.) Sputter deposition. R.F. sputter deposition was used to deposit approximately 50nm of insulating silica on top of the ferromagnetic mesostructures. R.F. sputter deposition is used to deposit insulating materials because it prevents the material from becoming charged during deposition, and extinguishing the plasma. Optical lithography, lift off and R.F. sputter deposition was used to deposit the silica directly above the ferromagnetic mesostructures. Fig. 4.6 shows a schematic of a typical R.F. plasma system. A tuning network is used to match the impedance between the plasma and the power source. A blocking capacitor is used to dc isolate the supply from the chamber. Sources are in the radio frequency (R.F.) range, commonly 13.56MHz. The chamber was pumped down to ×10-7 mbar before deposition using a diffusion pump. The silica was deposited in argon gas at a power of 50W, a pressure of 3Pa and a deposition rate of 300nm every 25mins. Approximately 50nm of silica was deposited, and the magnetron was water cooled throughout the deposition. + - CBlocking R.F. power supply V ZSub Fig. 4.6 A schematic diagram of an R.F. plasma system. (After Ref. [3]). Focused Ion Beam (FIB) lithography. This section will serve as a brief introduction to the operation of the FIB, a more detailed account of the operation of the FIB and possible uses can be found elsewhere [4]. The Focused Ion Beam technique (FIB) was mainly developed during the late 1970s and early 1980s, and the first commercial instruments were introduced more than a decade ago [5]. Modern FIB systems are becoming widely available in semiconductor research and processing environments, as well as in failure analysis and chip-design centres. The 81 CHAPTER 4 technology enables localised milling and deposition of conductors and insulators with high precision, hence its success in device modification, mask repair, process control and failure analysis [6-10]. Focused ion beam systems effectively combine together a scanning ion microscope and a precision milling system. By scanning the ion beam over a specimen and collecting the ion beam-induced secondary-electron or -ion signal an image of the surface is formed. The position for a milled pattern can be selected from the image taken, and then the structure can be produced using Ga+ ions without the use of a patterned resist mask. One of the advantages over conventional etch methods is that the etch depth can be varied across the structure, by varying the dwell time from point to point within the pattern. Ion source Suppressor Extractor Spray aperture First lens Upper octopole Variable aperture Blanking deflector Blanking aperture Lower octopole Second lens Sample Fig. 4.7 A schematic diagram of a FIB ion column. (After Ref. [5]). Fig. 4.7 is a schematic diagram of a FIB ion column. The structure of the column is similar to that of a scanning electron microscope, the major difference being the use of a gallium ion (Ga+) beam instead of an electron beam. The ion beam is generated from a liquid metal ion source (LMIS) by the application of a strong electric field. This electric field causes the emission of positively charged ions from a liquid gallium cone, which is formed on the tip of a tungsten needle. The Ga LMIS is used because of its stability, simplicity of operation and 82 CHAPTER 4 the fact that Ga ions give good sputtering yields. A vacuum of about 1×10-7 mbar is maintained inside the column. After a first refinement through the spray aperture, the ion beam is condensed in the first electrostatic lens. The upper octopole then adjusts the beam stigmatism. For patterning the samples in this thesis, the beam voltage was 30keV and the beam current was approximately 80pA. Using the variable aperture mechanism, the beam current can be varied over four decades, allowing for both a fine beam for high-resolution imaging and a heavy beam for fast and rough milling. The blanking deflector and aperture enable blanking of the beam away from the sample into a Faraday cup. This system not only protects the sample from constant milling, but can also measure the current. The lower octopole is used for scanning the beam over the sample in a user-defined pattern. In the second electrostatic lens, the beam is focused to a fine spot, enabling a best resolution in the sub 10nm range. A secondary electron detector is used to collect the secondary particles for imaging. Ion beam is scanned over substrate Secondary electrons from substrate detector Ion beam is scanned over substrate Sputtered material from substrate substrate substrate (a) (b) Fig. 4.8 Schematic diagrams showing the principle of FIB (a) imaging and (b) milling. (After Ref. [5]). 83 CHAPTER 4 Fig. 4.8 is a schematic showing the principle of FIB imaging (a) and milling (b). During FIB imaging the finely focused ion beam is raster scanned over a substrate, and secondary particles (neutral atoms, ions and electrons) are generated in the sample. As they leave the sample, the electrons or ions are collected by the secondary electron detector. To prevent positive surface charges from building up, the samples were wire bonded to the sample holder, which was grounded. FIB imaging inevitably induces some damage to the sample. Most of the Ga+ ions that arrive at the sample surface enter the sample, leading to ion implantation. The depth of the implanted region is related to the ion energy and the angle of incidence. When the Ga+ ion energy is 30keV, the implantation depth in SiO2 is 25± 8nm [5]. Also, some milling always occurs when the ion beam is scanned across the sample surface, although the milling effect can be reduced by using a fine ion beam (fine spot and low ion current). The removal of sample material is achieved using a high ion current beam, as shown in Fig. 4.8 (b). The result is a physical sputtering of the sample material. The milling process was calibrated using ‘end-point detection’; this is a real-time graph of the average brightness in the milling area. An insulator will appear darker than a conductor since the secondary electron yield of the latter is much higher. This effect results in a typical end-point detection curve as shown in Fig. 4.9. Dose/Depth Intensity Insulator Metal Insulator Fig. 4.9 End-point detection. The graph shows the average brightness in the milling area as a function of the delivered dose. This is used to monitor the milling operation in real time. The insulator-metal-insulator transition is clear. The first part (low brightness) corresponds to the insulating dielectric layer over the metal conductor. The metal itself is the central high-intensity part. Finally, after milling through the metal, the intensity diminishes again when the underlying insulating layer is reached. This technique was used to detect the amount of time to mill through the gold and silica layers, and therefore enable milling to the correct depth to design the hard mask. The resistance of the samples was measured before and after patterning the silica hard mask in the FIB. This 84 CHAPTER 4 indicated the amount of Ga+ implantation of the spin valve; the higher the resistance above the initial value, the greater the amount of Ga+ implantation. 4.3 Device characterisation. A number of different characterisation techniques were used for evaluating the samples in this thesis work, including electrical, magnetic and structural characterisation. This section will briefly go through the different techniques used throughout this thesis work. 4.3.1 Electrical characterisation. Resistivity values of thin film materials tend to be larger than the bulk due to increased surface scattering. The problem of electron scattering from thin film metal surfaces has been of interest for more than a century, and the subject is still not completely free from controversy. In 1901 Thomson used classical physics to provide a simple way to visualise the increase in resistivity as the film thickness decreases compared to the bulk value [11]. He derived the expression given in Eqn. (4.1), which shows that as the thickness is decreased, the mean free path of the thin film is reduced and the resistivity rises; a size effect is exhibited. . 2 31ln 2 0 0    +== κ κ ρ ρ σ σ f f (4.1) Where σf and λf are the resistivity and mean free path of the thin film, σ0 and λ0 are the resistivity and mean free path of the bulk material, and κ= d/λ0, where d is the thickness of the film. A more accurate quantum theory of thin-film conductivity was developed by Fuchs in 1938 [12] and elaborated upon in the ensuing half-century by other investigators, most notably Sondheimer [13]. The Fuchs-Sondheimer (F-S) theory is in close agreement with the Thomson equation. A standard four-point measurement technique was used to measure the resistance of the thin films and trilayers. A constant current is applied across the film and the resistance is measured by detecting the voltage drop through separate connections. Copper contact pads were used to make electrical contact and distribute the current flow evenly across the ends of the thin film. The effect of the increased current density on the resistance of the nanopatterned samples was investigated by applying a constant current and studying the change in resistance with time in zero applied field. It was found that the resistance of the sample remained constant, which was a direct indication that the increased current density was not detrimental to the GMR response. 85 CHAPTER 4 Magnetoresistance measurements provide an effective method for analysing the magnetisation reversal processes of ferromagnetic thin films and trilayers. The change in resistance with applied field can give a direct indication of the orientation of the magnetisation in a thin film with respect to the current, and an indication of the relative orientation of two ferromagnetic layers with respect to each other. Coercivity, switching fields and magnetic saturation values can be extracted from the measurements. Two types of magnetoresistance measurements were carried out: in-situ during milling and ex-situ for a range of temperatures from liquid nitrogen to 493K. The set-up for in-situ magnetoresistance measurements was purpose built and is described in detail in Chapter 6. For the ex-situ measurements, the sample was wire bonded to the sample holder using an ultrasonic wire bonder. The sample holder was placed in a probe, which was positioned between the pole pieces of an electromagnet. Typically 0.5mA constant current was supplied to the sample, and the field was swept between ±100Oe. The magnetoresistance of the sample was extracted from the change in voltage as the field was swept, and the current remained constant. The percentage MR of the sample is determined by Eqn. (4.2), ( ) ( ) ( )  ×− 100 s s HR HR   HR (4.2) where R(H) is the resistance in an applied field H and R(Hs) is the resistance at the saturation field. For low temperature measurements the probe was placed in liquid nitrogen and le couple of hours before measuring to allow the temperature to stabilise. High tem measurements were carried out by placing the probe in the centre of a quartz tube, wh sealed at one end to contain the heat. The quartz tube was resistively heated using a st which was wrapped in both directions to avoid any unwanted magnetic fiel temperature was carefully monitored using a thermo-couple attached to the sample h close proximity to the sample. The furnace was placed between the pole pieces of the 4.3.2 Magnetic characterisation. Magnetic characterisation was carried out using a Vibrating Sample Magnetometer which was first described by Foner [14], and is basically a comparator, which meas difference in induction between a region of space with and without the specimen. It t gives a direct measure of the magnetisation M. Fig. 4.10 is a schematic of the mechan VSM. The sample is oscillated vertically in a region of uniform field, and if the sa driven by a loudspeaker mechanism, the frequency is usually near 80Hz and the amp 0.1-0.2mm. The AC signal induced in the pick-up coils (according to the Faraday electromagnetic induction) by the magnetic field of the sample is compared with th ft for a perature ich was eel coil, ds. The older in magnet. (VSM), ures the herefore ism of a mple if litude is law of e signal 86 CHAPTER 4 from a reference specimen, and is converted to a number proportional to the magnetic moment. The advantages of this technique are that it is non-destructive, has a high sensitivity, and is easy to operate. Pick-up Reference specimen Loudspeaker/ vibration mechanism Reference Sample electromagnet Fig. 4.10 Schematic diagram of a Vibrating Sample Magnetometer (VSM). (After Ref [15]). The system is designed to measure small samples with high accuracy (less than 4 by 4mm). For larger samples a calibrated correction is required in order to determine the magnetisation values. However, a comparative analysis of other magnetic properties including coercivity and the magnetisation reversal process is always possible. 4.3.3 Structural characterisation. Scanning probe microscopy (SPMs) consists of microscopy forms that are used to measure the properties of surfaces. There are many different types of SPMs including atomic force microscopes (AFM), scanning tunnelling microscopes (STMs), and magnetic force microscopes (MFMs). A review of SPMs can be found in [16]. This section will describe AFM used in tapping mode and contact mode, both of which were used for this thesis work. The first commercially available AFM, the NanoScope® AFM (Digital Instruments) was introduced in 1989. 87 CHAPTER 4 AFM was used throughout this thesis work to probe and map the topolography of the patterned samples using a cantilever with a sharp tip on the end. The two different modes of AFM are shown in Fig. 4.11. Contact mode uses silicon nitride tips probes, which are soft enough to be deflected by very small forces and have a high enough resonant frequency not to be susceptible to vibrational instabilities. The tip is scanned across the sample surface while monitoring the change in the cantilever deflection with a split photodiode detector. A feedback loop maintains a constant deflection between the cantilever and the sample by vertically moving the scanner at each (x,y) data point to maintain a “setpoint” deflection. By maintaining a constant cantilever deflection, the force between the tip and the sample remains constant. The distance the scanner moves vertically at each (x,y) data point is stored by the computer to form the topographic image of the sample surface. Controller electronics Feedback loop maintains constant cantilever deflection Detector electronics Scanner Cantilever & Tip Sample Split photodiode detector X,Y Z Laser (a) Cantilever & Tip Detector electronics Nanoscope IIIa controller electronics Feedback loop maintains constant oscillation amplitude Scanner Sample Split photodiode detector X,Y Z Laser (b) Fig. 4.11 Schematic diagram showing (a) AFM contact mode, and (b) AFM tapping mode. (After Ref. [16]). In tapping mode silicon probes are used, which are much stiffer than the silicon nitride probes, resulting in larger force constants and resonant frequencies. Tapping mode operates by scanning a tip attached to the end of an oscillating cantilever across the sample surface. The cantilever is oscillated at or near its resonant frequency with amplitude ranging typically from 20nm to 100nm. The frequency of oscillation can be at or on either side of the resonant frequency. The tip lightly “taps” on the sample surface during scanning, contacting the surface at the bottom of its swing. The feedback loop maintains constant oscillation amplitude 88 CHAPTER 4 by maintaining a constant RMS of the oscillation signal acquired by the split photodiode detector. The vertical position of the scanner at each (x,y) data point in order to maintain a constant “setpoint” amplitude is stored by the computer to form the topographic image of the sample surface. By maintaining constant oscillation amplitude, a constant tip-sample interaction is maintained during imaging. The x- and y-axis topographic resolution for most SPM scanning techniques, including AFM, is typically 2 to 10nm. The z-axis resolution is better than 0.1nm. Solid state laser diode Cantilever + Tip Split photodiode detector A B Output BA BA + − Fig. 4.12 The mechanism of “beam deflection” detection for both contact and tapping mode. (After Ref. [16]). Fig. 4.12 shows the mechanism of “beam deflection” detection for both contact and tapping mode. Laser light from a solid state diode is reflected off the back of the cantilever and collected by a position sensitive detector (PSD) consisting of two closely spaced photodiodes whose output signal is collected by a differential amplifier. The angular displacement of the cantilever results in one photodiode collecting more light than the other photodiode, producing an output signal (the difference between the photodiode signals normalized by their sum), which is proportional to the deflection of the cantilever. This system can detect cantilever deflection < 1Å, and the long beam path (several cm) amplifies changes in the beam angle. The advantages of contact mode include the high scan speeds, it is the only AFM technique which can obtain “atomic resolution” images, and it is particularly good for samples with extreme changes in vertical topography. The mode was used for scanning the nanopatterned samples, which required accurate vertical profile analysis over small lateral distances. 89 CHAPTER 4 90 Tapping mode was used for the larger micropatterned samples, due to the better lateral resolution. A wide variety of analysis functions are available in the off-line menu for measuring captured SPM images. The two techniques most commonly used in this thesis were: • Section analysis Depth, height, width and angular measurements can be easily made using this analysis technique. • Depth analysis The depth feature may be used to make automated depth histogram measurements over many images. 4.3.4 Magnetic Domain Imaging. The Bitter technique makes use of ‘magnetic liquids’, which are colloidal suspensions of very small magnetic particles, usually of a ferrite such as Fe3O4, in a carrier liquid, which may be water or an organic solvent. Stray fields are present near the surface of the ferromagnet, which are strongest along the lines where domain walls meet the surface. When a drop of magnetic liquid is placed on the surface, the particles tend to delineate the domain walls, making the domain pattern observable under a microscope. Bitter pattern imaging has been reviewed by Kittel and Galt [17] and Craik [18]. For the purpose of this thesis work, commercially available water-based Ferrofluids was used, which consists of roughly isometric magnetite particles of about 10nm diameter, covered with various surfactants. The sensitivity of Bitter colloids can be enhanced by an external field perpendicular to the surface of the sample. The resolution of the Bitter technique reaches easily 100nm and is limited by the particle size of non-agglomerating colloids (∼10nm). CHAPTER 4 References [1] S.A. Campbell, The Science and Engineering of Microelectronic Fabrication, Chapter 11, 274, Oxford University Press, Inc. (1996). [2] S.A. Campbell, The Science and Engineering of Microelectronic Fabrication, Chapter 12, 294, Oxford University Press, Inc. (1996). [3] S.A. Campbell, The Science and Engineering of Microelectronic Fabrication, Chapter 10, 246, Oxford University Press, Inc. (1996). [4] J. Melngailis, J. Vac. Sci. Technol. B, Vol. 5, No. 2, 469, Mar/ April (1987) [5] S. Reyntjens, R. Puers, J. Micromech. Microeng. 11, 287 (2001) [6] D.K. Stewart, A.F. Doyle, J.D. Casey Jr, Proc. SPIE, 2437, 276 (1995) [7] S. Reyntjens, D. De Bruyker, R. Puers, Proc. 1998 Microsystem Symp., 125 (1998) [8] B.W. Ward, N.P. Economou, D.C. Shaver, J.E. Ivory, M.L. Ward, L.A. Stern, Proc. SPIE, 923, 92 (1988) [9] J. Glanville, Solid State Technol. 32, 270 (1989) [10] D.K. Stewart, L.A. Stern, G. Foss, G. Hughes, P. Govil, Proc. SPIE, 1263, 21 (1990) [11] J.J. Thomson, Proc. Cambridge Phil. Soc., 11, 120 (1901). [12] K. Fuchs, Proc. Cambridge Phil. Soc., 34, 100 (1938). [13] E.H. Sondheimer, Advan. Phys., 1, 1 (1951). [14] S. Foner, Rev. Sci. Inst., 30, 548 (1959). [15] S. Foner, J. Appl. Phys., 79, 4740 (1996). [16] Digital Instruments Veeco Metrology Group, Training Notebook (1999), http://www.veeco.com/html/library.asp [17] C. Kittel, J.K. Galt, Ferromagnetic Domain Theory, in Solid State Physics Vol. 3, (F. Seitz and D. Turnbull Eds.), 437-564, Academic Press, New York (1956). [18] D.J. Craik, The Observation of Magnetic Domains, in Methods of Experimental Physics, Vol. 11, (R.V. Coleman ed.) 675-743, Academic Press, New York (1974). 91 CHAPTER 5 New ideas pass through three periods: *It can't be done. *It probably can be done, but it's not worth doing *I knew it was a good idea all along ! Arthur C. Clarke Chapter 5 Control of the Switching Properties of Magnetic Thin Films and Spin Valve Devices by Patterning. The effects of patterning highly anisotropic repeating structures in soft magnetic thin films have been examined. Arrays of wires with equal mark/space ratios were patterned in resist using optical lithography. Broad beam ion milling was used to etch narrow (3µm) and wide (4mm) regions in Ni80Fe15Mo5 thin films. The results show that the wider regions dominate the magnetisation reversal, and the coercivity increases by a factor of ten along the wire axis when the wider regions are removed. The results suggest the possibility of designing structures which can be engineered to reverse domains in narrow wires within a controlled field range. Measurements were also carried out for partial milling through the thickness of the wire array structures to assess the dependence of the magnetic properties on the degree of interconnection between the wires. These structures show a progressive increase in coercivity and a transition between single and two-stage switching with increasing milling depth. A similar micropatterning technique has been applied to unpinned (Ni80Fe20/Cu/Ni80Fe20) pseudo spin valve (PSV) structures in order to enhance the coercivity of one of the ferromagnetic layers; the increased coercivity induced by patterning changes the natural similarity of the magnetic layers and the completed structure exhibits a small spin valve response. 92 CHAPTER 5 5.1 Motivations There has been a great deal of interest in artificially “engineering” the magnetic properties of thin film magnetic microstructures by altering the lateral size [1-12]. Induced anisotropy is an important parameter for controlling the magnetisation of thin films. Although anisotropy can be induced by a variety of different techniques, the present generation of devices based on the giant magnetoresistive (GMR) effect makes use of the exchange anisotropy induced in ferromagnetic layers by contact with antiferromagnetic (AF) materials such as IrMn. A practical disadvantage of this technique is the reduction in the pinning strength as the temperature is increased as discussed in recent reviews of exchange anisotropy [13] and synthetic AF pinning in spin valves [14]. The initial experiments in this chapter have been designed to investigate how the magnetisation processes of materials can be controlled by patterning. In order to simplify the experiments, Ni75Fe20Mo5 was chosen as the material to pattern because it has a low value of crystalline anisotropy. By minimising this energy term it is possible to investigate how domain wall motion and domain rotation can be influenced by patterning. Obtaining a thorough understanding of these two mechanisms will help to design tailored magnetic devices for specific applications. The pattern chosen for these experiments consists of long thin ferromagnetic wires, which represent the extreme case of shape anisotropy. The initial experiments in this chapter investigate the magnetisation reversal of structures containing both high and low shape anisotropy regions. This allows an investigation of the gradual change from two- to one- dimensional magnetism. The second part of this chapter investigates the change in the hysteresis by partially patterning through the thickness of the pseudo spin valve (PSV), which creates shape anisotropy. The milled depth between the wires affects the magnetisation reversal by varying the degree of interconnection. The final part of the chapter investigates patterning one of the ferromagnetic layers in an unpinned PSV with the aim of introducing a difference in the coercivities of the two ferromagnetic layers in the structure. Transport and magnetisation measurements were performed in order to characterise the PSVs. These preliminary investigations were carried out using optical lithography to enable direct comparison between magnetic and transport measurements. 5.2 Laterally patterned thin films. Three 10mm by 5mm silica substrates were prepared with approximately 250nm thick Ni75Fe20Mo5 film sputter deposited on top. (The samples were sputtered under the same conditions; the sample to target distance was 90mm, the deposition power was 30W, the pressure was 0.3Pa and the deposition time was 40 minutes). The VSM was calibrated using a Ni sample of known saturation. The first sample was then placed in the VSM and hysteresis 93 CHAPTER 5 measurements were carried out for different orientations of the chip with respect to the applied field. At 0° the magnetic field was along the 10mm length of the chip, and at 90° it was along the 5mm width. The second sample was put in a beaker of acetone and cleaned for 5 minutes in the ultrasonic cleaner. It was then washed with propanol using a spray gun, and dried using an air gun. A few drops of photoresist (5214) were deposited onto the surface of the sample. The sample was spun at 5000rpm for 30 seconds to spread the photoresist across the sample, and baked at 100oC for one minute to remove any moisture. Contact photolithography was used to remove the edge bead, and then the sample was baked again to remove any moisture. The chip was patterned into an array of 5mm long by 3µm wide wires, and the spacing between the wires was 3µm. Unpatterned rectangular regions of approximately 2mm by 4mm remained at either end of the wires. Fig. 5.1 (a) shows a schematic of the patterned wire array with unpatterned regions at either end of the wires. Below the schematic diagram there is a table summarising the lithographic technique. Process step Exposure Developing Developing Time(s) Type* Time(s) Edge bead removal 20 4:1 20 Patterning wires 8 3:2 13 (a) 5 mm 5mm (b) H (0°) H (90°) * The type of developer is in terms of parts developer to parts water. Fig 5.1 (a) is a schematic diagram showing the patterned Ni75Fe20Mo5 film with a wire array structure at the centre and with unpatterned rectangular regions at the ends of the wires. The direction of the applied field at 0° and 90° is shown. (b) is a schematic diagram showing the patterned Ni75Fe20Mo5 film with the unpatterned regions at the end of the wires removed. The table summarises the lithographic technique. After lithographic patterning, the sample was loaded into the argon ion miller and the chamber was pumped down to 4×10-6mbar. The sample was positioned directly in front of the source and rotated during milling to obtain a uniform etch. Milling was carried out for 60mins to ensure that the sample was milled through the Ni75Fe20Mo5 layer to the silica substrate. The 94 CHAPTER 5 beam current was maintained at 10mA, and the accelerator voltage and discharge voltage were 70V and 41V respectively. The pressure in the chamber was 2.1×10-4mbar during milling. After milling the sample was cleaned to wash away the remaining photoresist, and examined under the microscope. Magnetic hysteresis measurements were then carried out as the orientation of the sample was changed with respect to the applied field. For the third chip, the same lithography process was carried out as shown above. After milling, a second photoresist layer was spun on using the same speed and time as above. After baking, contact photolithography was used to remove the resist from the rectangular regions on the chip. The same exposure and developing times were used as for edge bead removal. The sample was loaded again into the OAR, and milled for another 60mins to remove the unpatterned regions. A schematic diagram of the third chip is shown in Fig. 5.1 (c) with the rectangular regions at either end of the wires removed. (A small overlap of approximately 20µm remains at the end of the wires). Topography scans were carried out using the AFM, and hysteresis measurements were taken at different orientations of the sample with respect to the applied field. The AFM showed that the thickness of the magnetic film was ≈265nm. Fig. 5.2 (a) shows the magnetisation versus field for the wire array with the field applied at 30°, 70°, and 80° to the length of the wires. Only half of the hysteresis loops are shown for clarity. 0 0.5 1 1.5 2 2.5 0 30 60 90 120 150 180 Field Angle (°) Sw itc hi ng fi el d (O e) Unpatterned Patterned -6 -4 -2 0 2 4 6 -60 -40 -20 0 20 40 60 Field (Oe) M ag ne tis at io n (m em u) 30deg 70deg 80deg (a) (b) Fig 5.2 (a) Magnetisation reversal graphs for the Ni75Fe20Mo5 film with a 5mm by 3µm wire array patterned through the thickness. Only half of the hysteresis is shown for clarity. The three graphs represent three angles of applied field relative to the length of the wire array. (b) Shows the initial switching field versus applied field angle for the unpatterned and patterned Ni75Fe20Mo5 thin film. The patterned structure consisted of an array of 5mm long by 3µm wide wires. 95 CHAPTER 5 There appears to be two M/H loops superimposed upon one another, for example for the 30° hysteresis there is an initial magnetisation reversal at ≈2Oe and another change in susceptibility at ≈5Oe. The sample behaves this way because the pattern consists of two parts; a highly anisotropic array of wires and a low anisotropy unpatterned region. The lower coercivity corresponds to the magnetisation reversal of the unpatterned part, and the higher coercivity is the reversal of the ferromagnetic wires. Similar work by Lee et al. [16] studied the domain nucleation processes in mesoscopic Ni80Fe20 wire junctions. Each wire consisted of a wide region (w1=5µm) and a narrow width (w1=1,2µm), both of equal length 200µm. Magnetic force microscopy and micromagnetic calculations showed that several domain walls nucleate at the wider part and are trapped in the junction area. This implies that domain nucleation at the junction of the wire initiates magnetisation reversal in the narrow half. The hysteresis graphs in Fig. 5.2 (a) show that the susceptibility of the wires decreases with increasing angle, and the switching field of the wires increases significantly with increasing angle. Fig. 5.2 (b) shows how the initial switching field changes with the angle of the applied field for the unpatterned and the partially patterned samples. The switching field of the unpatterned film remains approximately the same ±0.15Oe as the field angle is varied between 0° and 180°. However, the partially patterned film shows a maximum switching field at 0° and 180°, and a minimum at 90°. The difference between the maximum and minimum is ≈1Oe, which indicates a small change in the switching field of the unpatterned region. The switching field (Hs) of a multidomain structure is the sum of the fields required for nucleation and propagation of the domains through the sample, given by Hs=Hn+Hp, where Hn is the nucleation field and Hp is the propagation field. In this case, Hs is changing with angle due to a change in the propagation field Hp, and the difference can be explained by considering that the domain wall movement is perpendicular to the direction of the applied field. In other words, when the field is applied along the wire axis the domain walls are pinned by the ends of the wires during the magnetisation reversal. When the field is applied perpendicular to the wire length, there is no interaction with the end of the wires and the switching field is similar to the unpatterned film. Fig. 5.3 (a) shows the hysteresis at 70° for the partially patterned and fully patterned thin films. It is possible to see that when the sample is fully patterned the coercivity increases and the hysteresis shows a single loop. The switching field of the wire array part of the partially patterned film corresponds to the coercivity of the fully patterned sample. Fig. 5.3 (b) shows the switching field versus orientation for the unpatterned, partially patterned and fully patterned samples. The coercivity of the fully patterned wire array is much larger, and also the shape of the graph with applied field angle is different. The coercivity is a minimum at 0 and 180°, a maximum and 80° and decreases sharply at 90°. The change in the coercivity can be 96 CHAPTER 5 better understood by looking at Fig. 5.4, which shows the hysteresis graphs at 0°, 80° and 90° applied field orientations. -60 -40 -20 0 20 40 60 80 -300 -150 0 150 300 Field (Oe) M ag ne tis at io n (m em u) Fully patterned Partially patterned 0 10 20 30 40 50 0 30 60 90 120 150 180 Field angle (°) Sw itc hi ng fi el d (O e) Unpatterned Fully patterned Partially patterned (b) (a) Fig 5.3 (a) Magnetisation hysteresis curves at 70° for the partially patterned and fully patterned Ni75Fe20Mo5 thin films. (b) Switching field versus angle for the unpatterned, partially patterned and fully patterned Ni75Fe20Mo5 thin films. -4 -3 -2 -1 0 1 2 3 4 -500 -250 0 250 500 Field (Oe) M ag ne tis at io n (m em u) 0deg 90deg 80deg Fig 5.4 (a) Magnetisation reversal graphs at 0°, 80° and 90° for the 3µm wire array patterned in the Ni75Fe20Mo5 thin film. At 0° it is possible to see the sharp change in magnetisation, which is normally observed when the magnetisation reversal occurs by domain wall motion. At 80° the magnetisation reversal occurs by a combination of both domain wall motion and rotation, and the overall coercivity is larger. At 90° there is a much less sharp transition, and the magnetisation reversal is dominated by rotation with increasing field until it reaches saturation. However, the reversal process is not pure rotation because there is still some area under the M/H loop. 97 CHAPTER 5 98 The saturation field at 90° is increased due to the larger demagnetising factor across the width. For simplicity, we can consider the cross section of the strip to be elliptic so that the demagnetising factors Nα can be assigned to the three principal axes [17]. These demagnetising factors are [18] Nw=t/(t+w)≈t/w across the width w; Nt=w/(t+w) ≈1 through the thickness; and zero along the axis of the wire. The applied field was in the plane of the film so Nt has no relevance, and Nw=0.08. The demagnetising field is the magnitude of the magnetic saturation multiplied by the demagnetising factor, since the gap between the wires is large enough for the magnetostatic interactions to be negligible [3]. The magnetic saturation of Ni80Fe15Mo5 is 0.63×106Am-1 [19], this produces a demagnetising field of 630Oe when the film is saturated along the width. 5.3 Patterning through the thickness of thin films. Plain Ni75Fe20Mo5 films of thickness 100nm were deposited on 10mm × 5mm Si substrates by ultra-high vacuum DC magnetron sputtering. The deposition chamber walls were cooled below -100°C, and the Ar pressure during deposition was kept constant at 0.5Pa. The substrate to target (S-T) distance was 80mm, and the system base pressure was better than 3×10-9 mbar. The thicknesses of the films were determined by the deposition of a calibration sample with the same S-T distance and Ar pressure, which was measured by AFM. 3µm 6µm 5mm 3.8mm 100nm Fig. 5.5 A schematic diagram of an array of 3µm wide wires with equal mark/space ratio. The samples were patterned into 3µm wide wires arrays with equal mark space ratios using contact optical lithography and ion beam milling, as shown in Fig. 5.5. The wires were 5mm long and the array extended over a distance 3.8mm. The same lithographic patterning technique was used as summarised in Fig. 5.1. Before milling the chamber pressure was pumped down to 4×10-6mbar. Throughout the milling the chamber was kept at a constant 2.1×10-4mbar, the beam current was maintained at 10mA, and the accelerator voltage and discharge voltage were 70V and 41V respectively. The aim of the experiments was to CHAPTER 5 investigate the effect of partially milling the gaps between the wires so that a continuous film remains (see Fig. 5.5). Successive magnetic measurement and milling sequences were performed on the samples. Magnetisation measurements were carried out using a Princeton Measurements Corporation vibrating sample magnetometer (VSM), with fields applied along the length of the wires (the easy axis). The change in the magnetisation of the 100nm Ni75Fe20Mo5 film was measured as the milling depth of the magnetic material between the wires increased. Fig. 5.6 (a) shows the change in the magnetisation versus applied field for the same sample as the amount of material between the wires was successively removed; only half of the hysteresis loop is shown for clarity. As the thickness of the residual material (t) between the wires is reduced compared to the original thickness (t0), the magnetisation begins to show a double coercivity. This indicates that the patterned structure is becoming magnetised in two stages. ) me m tio n gn et i M a -1.5 -1 -0.5 0 0.5 1 1.5 2 -40 -20 0 20 40 Field (Oe) sa ( u t/t0=1 t/t0=0.48 t/t0=0.34 t/t0=0.2 t/t0=0 (a) (b) t/t0 0 5 10 15 20 0 0.2 0.4 0.6 0.8 1 C oe rc iv e fie ld (O e) Hc1 Hc2 Fig. 5.6 (a) Magnetisation versus applied field for a Ni75Fe20Mo5 sample with decreasing interconnection between the patterned wires: t/t0 = 1 (largest saturation moment), 0.48, 0.34, 0.2 and 0 (highest coercivity). (b) Coercive field versus residual thickness between the wires for a Ni75Fe20Mo5 sample with decreasing interconnection between the patterned wires: crosses indicate the zero moment coercive field, closed symbols the field at which a second magnetic transition occurs. At very low values of t the hysteresis loop returns to a single transition, as show in Fig. 5.6 (b). The initial reversal field Hc1 increases with increasing milling depth and shows a maximum enhancement of a factor of 10. The second switching field Hc2 also increases with increasing milling depth and shows a maximum enhancement of a factor of 4. Fig. 5.6 (a) and (b) show that it is possible to greatly increase the coercivity of a permalloy thin film by patterning a highly anisotropic structure through the thickness. 99 CHAPTER 5 A simple mathematical model was used to investigate the double switching field response of the sample at large milling depths with respect to the volume magnetisation of each part of the hysteresis. The aim of the model was to determine whether the patterned and unpatterned regions of the sample were reversing independently in an applied field. Two different configurations were investigated as shown in Fig. 5.7. In Fig. 5.7 (a) the cross section of the sample is divided into β representing the patterned wire array, and α the unpatterned area. In (b) the cross section is divided into β representing the patterned wire array plus the material beneath the wires, and α the material between the patterned wires. A single switching behaviour is observed until the milling approaches half the depth, although a larger field is required to complete the reversal as the milling depth increases. =β =α (b) (a) Fig. 5.7 The two configurations of the model used to investigate the double switching response of the patterned film at large milling depths. (a) β represents the patterned wire array and α is the unpatterned region. (b) β is the patterned wire array plus the material directly below and α is the region between the wires. This could be due to the larger field required to move the domain walls past the partially patterned wires. Until half the depth is reached, the entire structure is coupled together during the magnetisation reversal. For the measurements showing a double coercivity, the relative proportion of magnetisation from the hysteresis was compared to the cross sectional areas in Fig. 5.7 (a) and (b). It was calculated that the volume magnetisation of α and β do not correspond directly with the hysteresis results for either configuration (a) or (b). In other words the model shows that the patterned and unpatterned regions of the sample do not reverse completely independently at any part of the hysteresis. However the results also show that for small milling depths the reversal is most similar to Fig. 5.7 (a) where the unpatterned regions behaves more like a continuous film. For large milling depths the magnetisation reversal is more similar to (b), where the thick wire array dominates the magnetisation reversal of the material directly beneath and α is restricted to the material between the wires, which shows an increased coercivity as the thickness is decreased. The model shows that the 100 CHAPTER 5 high field switching observed at large milling depths is most likely to be due to the reversal of the thin film layer between the patterned wires. It is also important to consider the domain structure within the Ni80Fe20 thin film as the thickness is milled away. The literature shows that at a film thickness of 100nm Bloch walls have a lower energy than cross-tie or Néel walls. As the thickness is reduced cross-tie walls become more stable until ≈50nm when Néel walls form to reduce the magnetostatic energy. The widths of Bloch and Néel walls also vary in different ways with film thickness: the thinner the film the narrower the Bloch wall and the wider the Néel wall. Materials with wider domain walls tend to have a lower coercivity, although the results in Fig. 5.6 show that the increase in coercivity due to the shape anisotropy is larger than the effect of increasing the domain wall thickness. 5.4 NiFe Pseudo Spin Valve Measurements Pseudo spin valve sandwiches were grown without a pinning layer using a production spin valve sputtering system (Nordiko 9606 Plasma Vapour Deposition (PVD) Sputter Deposition System) and with the structure: Ta(3nm)/Ni80Fe20(6nm)/Cu(2.2nm)/Ni80Fe20(15nm)/Ta(3nm). The anisotropy axes in the two ferromagnetic layers were perpendicular with respect to each other. The PSVs were diced into 10mm by 5mm samples, and then patterned into 4mm by 3.3mm rectangles using optical lithography and argon ion milling. The anisotropy axis of the thicker top ferromagnetic was along the length of the rectangular patterned structure. -0.3 -0.2 -0.1 0 0.1 0.2 0.3 -10 -5 0 5 10 Field (Oe) M ag ne tis at io n (m em u) 90 0 Fig. 5.8 Magnetisation measurements of the rectangular 4mm by 3.3mm patterned pseudo spin valve along the length of the sample (0°) and along the width (90°). 101 Magnetic hysteresis measurements were carried out on the patterned sample as shown in Fig. 5.8. Measurements were taken along the length (0°) and width (90°) of the PSV rectangle, and CHAPTER 5 the hysteresis loops show a single transition despite the presence of two ferromagnetic layers. The total coupling between the two ferromagnetic layers is the sum of the exchange interaction and the magnetostatic coupling across the nonmagnetic interlayer. Previous work has shown that magnetostatic coupling occurs in bilayers of the structure Ni80Fe20/Cu/Ni80Fe20, where the thickness of the copper interlayer is 10nm and the two ferromagnetic layers are 100nm thick [20]. The results showed the occurrence of quasi walls due to flux closure from the domain walls. The shape of the graphs in Fig. 5.8 is characteristic of the anisotropy in the thicker ferromagnetic layer, which is in the same direction as the applied field at 0° orientation. The hysteresis shows that the two ferromagnetic layers are coupled together and the magnetisation reversal is dominated by the anisotropy of the thicker (15nm) ferromagnetic layer. The Ni80Fe20(15nm) top layer of the spin valve sandwich was patterned with the same wire array pattern as in the previous section and milled to different depth to investigate the effect on the switching of the device. The spin valve structure chosen for these experiments had a deliberately thick top Ni80Fe20 layer to enable controlled milling trials. Fig. 5.9 shows a schematic of the 3µm wire array structure patterned into the top ferromagnetic layer of the pseudo spin valve, the anisotropy axes Kα and Kγ, of the layers were induced during sputter deposition. Fig. 5.9 NiFe/Cu/NiFe pseudo spin valve with a 3µm wide wire array patterned into the top ferromagnetic layer. The diagram also shows the anisotropy axes, Kα and Kβ of the two ferromagnetic layers, which were induced during sputter deposition. 102 Magnetisation measurements were taken of the sample at different milling depths through the top ferromagnetic layer. The results for the spin valve device show a similar trend to the permalloy film, the coercivity of the patterned layer increases with increasing milling depths. For large milling depths the coercivity of the top layer becomes large enough to change the natural similarity of the ferromagnetic layers, and the magnetisation shows a double CHAPTER 5 coercivity. This can be seen in Fig. 5.10 for milling depths of 14.4 nm and 16.8 nm, where the field was applied along the length of the wires. The second coercivity increases by a factor of 8 with increasing milling depth through the top ferromagnetic layer. -0.4 -0.2 0 0.2 0.4 -20 -10 0 10 20 Field (Oe) M ag ne tis at io n (m em u) Unpatterned 14.4nm 16.8nm Fig 5.10 Hysteresis measurements for pseudo spin valve samples with decreasing interconnection between the patterned wires in the upper ferromagnetic layer: milled depths are 14.4nm and 16.8 nm. At 16.8nm the structure is milled through to the Cu interlayer. The field was applied along the length of the wires. -100 -50 0 50 100 Field (Oe) Unpatterned 14.4nm 16.8nm -0.4 -0.2 0 0.2 0.4 -5 -2.5 0 2.5 5 Field (Oe) M ag ne tis at io n (m em u) Unpatterned 14.4nm 16.8nm (b)(a) Fig 5.11 Hysteresis measurements for pseudo spin valve samples with decreasing interconnection between the patterned wires with the field applied perpendicular to the length of the wires, and (a) at low fields and (b) high fields. Fig. 5.11 shows the hysteresis measurements for different milling depths with the field applied perpendicular to the length of the wires for low (a) and high (b) fields. In this case, the field is applied along the anisotropy axis of the bottom unpatterned ferromagnetic layer. At 103 CHAPTER 5 low fields (a) the hysteresis loop shows a small increase in coercivity, at higher fields (b) there is a distinct change in susceptibility before saturation. At a depth of 16.8nm, the magnetisation reversal at low fields corresponds to the switching of the unpatterned layer, which has become decoupled from the thicker ferromagnetic layer due to the difference in coercivity. The high field change in susceptibility corresponds to the reversal of the patterned ferromagnetic layer by magnetisation rotation. The magnetoresistance of the samples was also measured by passing a DC current through the wires and measuring the change in resistance using a four-point probe technique as the field was applied along the length of the patterned wires. Fig. 5.12 (a) shows the MR response of the 16.8nm milled spin valve with the field applied along the wire axis, the magnetic response is also superimposed. Fig. 5.12 (a) shows that the initial magnetisation reversal corresponds to an AMR≈0.3% which turns into a GMR≈0.45% as the field is increased. Although the relative proportions of γ and α indicate towards the magnetisation reversal of the thinner (6nm) and thicker (15nm) ferromagnetic layers respectively, it was found that the proportions do not correspond directly with the dimensions of the two layers. 0 0.2 0.4 0.6 0.8 1 -20 -10 0 10 20 Field (Oe) M R (% ) MR (%) M (memu) -20 -10 0 10 20 Field (Oe) -0.2 -0.1 0 0.1 0.2 M (m em u) MR (%) M (memu) γ α (i) (ii) (iii) (a) (b) Fig. 5.12 Magnetoresistance and magnetisation measurements of the pseudo spin valve sample milled to a depth of 16.8nm, the y-axis for the MR (%) is on the left and the y-axis for the Magnetisation (memu) is on the right. In (a) the field is applied along the wire axis. The insets γ and α indicate different regions in the magnetisation reversal. In (b) the field is applied perpendicular to the wire axis and the insets (i), (ii), and (iii) indicated different parts of the magnetoresistance measurement. 104 CHAPTER 5 Although the GMR signal indicates that the layers are decoupled, the hysteresis shows that the decoupling is only partial and antiparallel alignment between the two magnetic layers is not fully achieved. The decoupling is due to the increased coercivity of the patterned layer changing the natural similarity of the two ferromagnetic layers, and also due to the reduction in the moment of the top layer. The results show that the 6nm base layer switches at a considerably lower field than the patterned upper layer. The magnetoresistance changes sign once the unpatterned layer has reversed which is a clear manifestation of the desired GMR response. This positive resistance increase disappears once the patterned layer has reversed. Fig. 5.12 (b) shows the magnetoresistance and magnetisation when the field is applied perpendicular to the wire axis. As the field is decreased from positive saturation field (i), the magnetisation rotation of the patterned layer is shown by an AMR in the magnetoresistance measurement. The GMR peak close to zero field (ii) is the switching of the unpatterned layer. As the field is increased to negative saturation field (iii) AMR is again seen which corresponds to the rotation of the patterned layer until the structure is saturated in the opposite direction. A larger field is required to saturate the patterned layer due to the large demagnetising field. Minor loop magnetoresistance analysis was carried out on the spin valve as shown in Fig. 5.13. Transport measurements were taken up to reversal fields of -5Oe, -6Oe and -7Oe, the measurements are offset for clarity. The results show that the magnetisation reversal is completely reversible up to –5Oe. Using some simple mathematical analysis it was possible to calculate the maximum GMR and AMR of the PSV sample. The analysis makes the following assumptions: (1) Up to -5Oe the only reversal process is the magnetisation rotation of the bottom layer. (2) The MR response consists of an AMR and GMR component and can be described by the Eqn. (5.1) [ ]θθ 22 cos1 2 cos1 −−   −= AGMR (5.1) where G=maximum GMR and A=maximum AMR. The relative magnetic volumes of the top (Vt) and bottom (Vb) layers were calculated from the dimensions of the sample, where Vb+Vt=1. The normalised magnetisation of the spin valve is given by Eqn. (5.2). tb s tbs s VHV M VVM M HM +=+= )(cos)cos()( θθ (5.2) 105 CHAPTER 5 0 0.6 1.2 1.8 -10 0 10 20 30 40 Field (Oe) M R (% ) -7Oe -6Oe -5Oe Fig. 5.13 Minor loop magnetoresistance analysis carried out on the pseudo spin valve patterned through the top ferromagnetic layer. Analysis was carried out with reversal fields up to -5Oe, -6Oe and -7Oe, the graphs are vertically displaced for clarity. Where M(H) is the magnetisation for an applied field H, and Ms is the saturated magnetisation. By rearranging the equation it is possible to calculate cosθ, where θ represents the angle between the ferromagnetic layers. b t s V V M HM H − = )( )(cosθ (5.3) M R (% ) -0.2 0 0.2 0.4 0.6 0.8 - 1 0 1 2 3 4 M odel - 5Oe θ (radians) Fig. 5.14 The magnetoresistance versus the angle between the magnetisation in the two ferromagnetic layers. The M(H) values were extracted from the hysteresis loop. These values give θ(H) through 106 CHAPTER 5 Eqn. (5.3), and therefore the MR versus H plot from the -5Oe plot in Fig. 5.13, could be converted to MR versus θ as shown by the black curve in Fig. 5.14. In Fig. 5.14 the curve generated from Eqn. (5.1) and the converted MR versus θ data were superimposed, and the values of G and A were adjusted to fit the experimental results. It can be seen that the model fits well to the measured data, and the result predicts a maximum value of GMR=0.72% and AMR=0.51%. 5.5 Conclusions. The initial experiments on plain permalloy films were performed in order to assess the extent to which lateral partial patterning could enhance the coercivity of the film. The results show that when the structure consists of a patterned wire array and an unpatterned region, the magnetisation reversal is predominantly determined by the wider region for fields applied parallel to the wire axis. The results imply that domain nucleation at the junction of the wire initiates magnetisation reversal in narrow part. These results are in agreement with previous work [16], which suggests the possibility of designing structures which can be used to “launch” reverse domains in narrow wires within a controlled field range. The second set of measurements investigated partially patterning through the thickness of the ferromagnetic film. The results show that a maximum coercivity is achieved when the structure is fully milled through, which corresponds to a factor of ten enhancement in coercivity. These results show that it is possible to tailor the magnetisation reversal by patterning through the thickness as well as laterally. The final measurements in this chapter investigated applying the patterning technique to create a pseudo spin valve. The results shown here demonstrate that even relatively course micron-scale patterning of one layer in an otherwise unpinned spin valve can create a small GMR effect. When a field is applied along the wire axis, a small AMR is seen due to the magnetisation reversal of the unpatterned layer. As the field is increased a GMR response is observed because the ferromagnetic layers have become decoupled due to the increased shape anisotropy in the top layer. When the field is applied along the wire width a small GMR response is seen at low fields due to the reversal of the unpatterned layer. A large field is required to saturate the sample because of the high demagnetising field across the width of the wires. The work of McMichael et al. [15] has shown that nanoscale patterning induced by deposition onto a morphologically textured surface has a much more pronounced effect on the magnetic response and leads to a substantial GMR effect. The results indicate that it may be possible to extend this work to the nano-scale, which is more in line with the current demand for miniaturisation. Since the anisotropy generated by these processes is expected to have low 107 CHAPTER 5 108 temperature dependence, this approach to spin valve design has considerable potential for high temperature device operation. CHAPTER 5 References [1] C. Mathieu et al., Appl. Phys. Lett. 70, 2912 (1997) [2] A. O. Adeyeye, J. A. C. Bland, C. Daboo, D. G. Hasko, and H. Ahmed, J. Appl. Phys. 79, 6120 (1996) [3] A. O. Adeyeye, G. Lauhoff, J. A. C. Bland, C. Daboo, D. G. Hasko, and H. Ahmed, Appl. Phys. Lett. 70, 1046 (1997) [4] C. Shearwood, S. J. Blundell, M. J. Baird, J. A. C. Bland, M. Gester, H. Ahmed, and H. P. Hughes, J. Appl. Phys. 75, 5249 (1994) [5] S. J. Blundell, C. Shearwood, M. Gester, M. J. Baird, J. A. C. Bland, and H. Ahmed, J. Magn. Magn. Mater. 135, L17 (1994) [6] K. Hong and N. Giordano, Phys. Rev. B 51, 9855 (1995) [7] K. Hong and N. Giordano, J. Magn. Magn. Mater. 151, 396 (1995) [8] A. Maeda, M. Kume, T. Ogura, K. Kuroki, T. Yamada, M. Nishikawa, and Y. Harada, J. Appl. Phys. 76, 6667 (1994) [9] M. S. Wei and S. Chau, J. Appl. Phys. 76, 6679 (1994) [10] J. F. Smyth, S. Schultz and D. R. Fredkin, D. P. Kern, S. A. Rishton, M. Cali, and T. R. Koehler, J. Appl. Phys. 69, 5262 (1991) [11] A. D. Kent, S. von Molnar, S. Gider, and D. D. Awschalom, J. Appl. Phys. 76, 6656 (1994) [12] J. I. Martin, J. Nogues, Ivan K. Schuller, M. J. Van Bael, K. Temst, C. Van Haesendonck, V. V. Moshchalkov, and Y. Bruynseraede, Appl. Phys. Lett. 72, 255 (1998) [13] A. E. Berkowitz, K. Takano, J. Magn. Magn. Mater. 200, 552-570, (1999) [14] Y. Sugita, Y. Kawawake, M. Satomi and H. Sakakima, J. Appl. Phys. 89, 6919 (2001) [15] R. D. McMichael, C. G. Lee, J. E. Bonevich, P.J. Chen, W. Miller, and W. F. Egelhoff, Jr, J. Appl. Phys. 88 (9), 5296 (2000) [16] W. Y. Lee, C. C. Yao, A. Hirohata, Y. B. Xu, H. T. Leung, S. M. Gardiner, S. McPhail, B. C. Choi, D. G. Hasko, and J. A. C. Bland, J. Appl. Phys. 87 (6), 3032 (1996) [17] J. H. Fluitman, Thin Solid Films, 16, 269 (1973) [18] J. A. Osborn, Phys. Rev., 67, 351 (1945) [19] D. Jiles, Introduction to Magnetism and Magnetic Materials, pg 94, Chapman and Hall (1998) [20] J. L. Prieto, P. Sánchez, C. Aroca, M. Maicas, E. López, M. C. Sánchez, J. Magn. Magn. Mater. 177-181, 215 (1998) 109 CHAPTER 6 La Buena suerte la buscas, la mala suerte te encuentra. (We have to chase good luck, but bad luck finds us). José Luis Prieto Chapter 6 The Design, Building and Testing of an In-situ Magnetoresistance Measurement Rig in an Argon Ion Miller. An in situ magnetoresistance test rig has been designed and built into an argon ion milling rig. The measurements from the rig allow direct analysis of the evolution of magnetic and electrical properties of ferromagnetic structures during patterning. This chapter describes the design, building and testing of the rig including; the fully rotating sample holder, the various coils for applying a field, the four-point resistance measurement set-up, the voltage offset amplifier and the signal transmission through the measurement. 110 CHAPTER 6 6.1 Motivations The previous measurements have shown that it is possible to change the natural similarity of two ferromagnetic layers in a PSV structure by patterning one of the layers to produce a small GMR response. The milling and electrical characterisation were done in separate steps, which proved to be an inefficient method for a number of reasons. It meant that either a series of samples needed to be prepared, milled to different depths and characterised individually, or the same sample was removed after milling, characterised and then put back into the argon ion miller. The disadvantage of the first technique is that there is always some variation between the different samples, for example a variation in the material properties from deposition or some variation in the lithography. The disadvantages of the second technique include that that it is very time consuming (≈2hrs to pump down to a good vacuum before milling), degradation of the PSV by oxidation, contamination or mechanical damage from the wire bonding to make electrical contact. It was decided that it would be ideal to design a technique to do in-situ measurements during milling. This meant that the same sample could be tested in a much cleaner and more efficient process. Many more milling steps could be carried out enabling trend graphs to be obtained on the variation of the resistance R during milling and also the change in resistance ∆R in the presence of an applied field. As previously described ∆R is a direct indication of the relative orientation of the magnetisation in the two ferromagnetic layers. As the dimensions are reduced the magnetic configuration will change from multi-domain to single domain and in-situ measurements will allow direct analysis of this change. This chapter describes the design, testing and calibration of the in-situ magnetoresistance set- up which was used throughout this thesis. 111 CHAPTER 6 6.2 The Electrical Set-up. The aim of this novel experimental set-up is to enable in-situ magnetoresistance measurements during patterning by argon ion milling. In order to achieve this, a system for applying a magnetic field and measuring the change in resistance under vacuum (≈ mbar base pressure) was designed. Fig. 6.1 is a schematic diagram showing the electrical set-up. 6104 −× Voltage Amplifier Oscilloscope Function Generator Sample Holder Computer Power Supply Sample Coil Coil Cu contact V Offset I V V I Lock in Amplifier Current Source Function Generator Fig. 6.1 A schematic of the electrical set up for the in-situ magnetoresistance measurement rig. The sample is placed on the sample holder which can be fully rotated in the argon ion milling chamber and is water cooled to avoid any unwanted heating effects. A triangular wave is swept at a low frequency (0.05Hz) by the function generator through a current amplifier to the field coils on either side of the chip. The power operational amplifier LM675 was used for the 112 CHAPTER 6 current amplifier set to a gain of 22. The LM675 was chosen because it is capable of delivering output currents in excess of 3 amps, operating at supply voltages of up to 60V. The circuit diagram for the voltage amplifier is shown in Fig. 6.2. Gain=R1/R2=22 0.1µF +VCC 22k LM675 + - 1 2 LR 1 0.22µF -VEE 0.1µF R2=1k R1=22k 3 5 4 VFG Fig. 6.2. Circuit diagram for the voltage amplifier supplying current to the field coils [1]. In the circuit diagram VFG is the voltage from the function generator, and LR is the load resistance equal to the resistance of the field coils plus a resistor in series with the coils. The Tektronix digital oscilloscope measures the voltage across the series resistor to calculate the current through the coils; this gives a more accurate current measurement than measuring directly across the coils due to heating effects. Pre-calibration measurements were carried out in order to convert the current to field for the different field coils. The resistance of the chip is measured by depositing copper contact pads and using a four-point resistance measurement. A higher frequency function generator (1kHz) drives a sine wave current across the chip, and a EG&G model number 5210 lock in amplifier locks into the 1kHz signal to read the voltage data. The voltage is offset to remove the initial resistance of the chip from the measurement and ∆R is amplified and then sent to the oscilloscope, which displays the field and resistance as voltage versus voltage graphs. The data is recorded by ‘Wavestar Software for Oscilloscopes’ on the computer and later converted into field versus resistance or magnetoresistance using Microsoft Excel. During milling the beam voltage was 500V and the 113 CHAPTER 6 beam current was 10mA. The pressure in the chamber was kept at a constant 2.1×10-4mbar, and the milling process was cut-off during the measurement time (≈1min). 6.3 The Sample Holders. The sample holder was designed to incorporate the chip, the field coils and the necessary electrical connections. It was designed to fit on top of the fully rotating stage in the argon ion milling chamber. Fig. 6.3 shows the initial design for the sample holder. soft magnetic core field coil contact pads Sample holder Coil with soft magnetic core - + 40 10 40 M2M2 10 1 17 6 28 32 Fig. 6.3. The initial sample holder incorporating a sample stage, electrical contact pads, and the field coil which consisted of a coil wrapped around a soft magnetic core. The sample holder is machined from copper and the stage is 10mm by 10mm in size, enabling the 10mm by 5mm samples to be measured with the field and current parallel or perpendicular to each other respectively. The underside of the sample holder has a circular hollow with side screws to allow it to be mounted onto the rotating stage in the chamber. The contact pads were made using single-sided copper circuit board, which was patterned using optical lithography and etched using Ferric Chloride (FeCl3). The circuit board was cut and sanded to the correct shape. The field coil consists of a coil wound around a soft magnetic core, and is described in detail in section 6.4. 114 CHAPTER 6 Fig. 6.4 shows the second design for the sample holder, the main differences being the sample holder was machined from aluminium, two field coils on either side of the sample, no soft magnetic core, and the size of the sample stage. Sample holder 5.0 10 11 5.5 18 m m Coil 2 Coil 1 M3 1 9 8 6 28 32 Fig. 6.4. The second design for the sample holder incorporating a sample stage, electrical contact pads, and two field coils on either side of the stage. The sample holder was made from aluminium to reduce the load on the rotating stage. The stage was made narrower to achieve a larger, more uniform field across the chip in the absence of the soft magnetic core. The two field coils holders were made of aluminium and were light and compact. The coil configuration is not Helmholtz, but provides a uniform field on the micron scale which is sufficient for the devices. The slow rate of change of field also influenced the material choice. 6.4 The Field Coils. 6.4.1 The field coil with a soft magnetic core. The field coil designed for sample holder 1 incorporates a coil wound around a soft magnetic core. The soft magnetic core is used to concentrate the flux and distribute the magnetic field uniformly across the chip. The field between the pole pieces of the soft magnetic core is uniform over the millimetre range. This design of field coil was used to magnetise millimetre- 115 CHAPTER 6 sized samples patterned by optical lithography. Two different materials were tested for the soft magnetic core: mild steel and permalloy. The mild steel core was constructed as shown in Fig. 6.3 and was wound with 3000 turns of 150µm diameter copper wire. The resistance of the coil was 58.3Ω. A current source was used to drive a current through the coil and the field between the poles of the core was measured using a Hall probe. In this design the magnetic field produced by the coils magnetises the magnetic core, and the polarisation of the core produces a field across the chip. At a high enough applied current the magnetic field from the coils is large enough to saturate the core, this is the limit of the applied magnetic field across the chip. Current (mA) 0 50 100 150 200 250 300 350 400 0 200 400 600 800 Fi el d (O e) -100 -80 -60 -40 -20 0 20 40 60 80 100 -80 -60 -40 -20 0 20 40 60 80 Current (mA) Fi el d (O e) 1 2 3 4 Fig. 6.5 (a) A graph showing the current versus field for the field coil with 3000 copper turns and a mild steel core. The graph shows the applied current for which the magnetic core saturates. The dotted lines are guides for the eye.(b) A graph to show the amount of field produced for a given applied current by the field coil with 3000 copper wire turns and a mild steel core. The numbers indicate the direction of the applied current. Fig. 6.5 (a) shows that the mild steel core saturates at ≈ 600mA, achieving a maximum field of 350Oe. However the graph loses linearity at ≈ 400mA, showing a field of 300Oe. It is also important to study the magnetic field produced from the field coil when the current is cycled in the negative and positive direction, as shown in Fig. 6.5 (b). Before the measurement was taken the core was saturated by a permanent magnet. The current versus field graph shows hysteresis with Hc≈8Oe, due to the remanence of the mild steel core. In order for this field coil to be used in magnetoresistance measurements the hysteresis of the core needs to be pre- calibrated. This was done by dividing the field measurement into four parts as shown in Fig. 6.5 (b), and curve fitting to each part. Using this technique it was possible to mathematically predict the field from the current input. Two different methods of saturating the core were 116 CHAPTER 6 investigated: applying a large current (a) and applying a large magnetic field (b) from a permanent magnet, as shown in Fig. 6.6. The application of a large current (1A) to saturate the core showed poor reproducibility due to heating effects. When the core was saturated using a large magnetic field, the reproducibility was much better. When using this design of field coil for the in-situ experiment, the core was saturated before the measurement using a permanent magnet. (a) (b) -200 -150 -100 -50 0 50 100 150 200 -150 -100 -50 0 50 100 150 Current (mA) Fi el d (O e) -150 -100 -50 0 50 100 150 Current (mA) Fig. 6.6 Graphs to show the reproducibility of the hysteresis of the field coil with a mild steel core by saturating with (a) a 1A current through the coils, (b) a permanent magnet. -100 -80 -60 -40 -20 0 20 40 60 80 100 -80 -60 -40 -20 0 20 40 60 80 Current (mA) Fi el d (O e) 1 2 3 4 0 10 20 30 40 50 60 70 0 50 100 150 200 250 Current (mA) Fi el d (O e) (b)(a) Fig. 6.7 (a) A graph showing the current versus field for the field coil with 2100 copper turns and a permalloy core. The graph shows the applied current for which the magnetic core saturates. The dotted lines are guides for the eye. (b) A graph to show the amount of field produced for a given applied current by the field coil with 2100 copper wire turns and a permalloy core. The numbers indicate the direction of the applied current. 117 CHAPTER 6 118 A field coil with a permalloy core was constructed using the same design, 2100 copper turning were wound and the initial resistance was 32.1Ω. The magnetic field from the core was tested as the applied current was increased to investigate the saturation of the core, as shown in Fig. 6.7 (a). The permalloy core is fully saturated at 200mA, producing a maximum field of 60Oe. However, the magnetic field loses linearity at ≈ 70mA producing a magnetic field of ≈52Oe. The core was saturated with a 250mA current, and the current was swept from negative to positive to investigate the hysteresis of the permalloy core. Fig. 6.7 (b) shows that the permalloy core has nearly zero hysteresis, a linear curve fit was used to predict the field from the applied current for the in-situ measurement rig. The results from the field coils with soft magnetic core are summarised in the table shown in Fig. 6.8. Mild Steel Core Permalloy Core Saturation current (mA) 600 200 Saturation field (Oe) 350 60 Maximum linear current (mA) 400 70 Maximum linear field (Oe) 300 52 Saturation method Permanent magnet 250 mA current Hysteresis Yes, Hc≈8Oe Small Curve fit Polynomial Linear Fig. 6.8 A table summarising the properties of the mild steel and permalloy cores. The mild steel produces a larger field than the permalloy core, but it also requires a larger current which can lead to heating of the coil. It also has a larger hysteresis and requires a four part polynomial equation to convert the applied current to field. In order to optimise the reproducibility, the core is saturated by a permanent magnet before pumping down to achieve a good vacuum. The mild steel core was used when larger fields were required for the in-situ magnetoresistance measurement. The permalloy core is only capable of fields up to 60Oe, but it has close to zero hysteresis and can be saturated by applying a current at anytime during the measurement. The permalloy core was used when low fields were sufficient for the in-situ magnetoresistance measurement. Transport measurements were carried out using spin valve material supplied by Nordiko with the construction: Ta(50)/NiFe(25)/CoFe(20)/Cu(26)/CoFe(20)/IrMn(100)/Ta(50), where the thicknesses are given in Å. The wafer was diced into 10mm by 5mm sized chips, with the direction of anisotropy of the ferromagnetic layers parallel to each other and also parallel to CHAPTER 6 the length of the chip. The experiment was carried out at room temperature and pressure, with the applied field perpendicular to the current and anisotropy in the layers. The results were used to compare the in-situ magnetoresistance measurement rig to a regularly used transport measurement rig to check that the set-up was working properly. % ) Current (mA) Current (mA) 0 1 2 3 4 5 6 - 60 - 40 -20 0 20 40 60 M R ( -40 -20 0 20 40 60 (b) (a) I K H I K H (a) (b) Fig. 6.9 Measurements of applied current versus MR for the spin valve sample using mild steel (a) and permalloy (b) field coils. The inset shows the direction of the current I, anisotropy K and applied field H for clarity. 0 1 2 3 4 5 6 -150 -100 -50 0 50 100 150 Field (Oe) M R (% ) Mild steel field coil Standard rig -100 -50 0 50 100 150 Field (Oe) Py field coil Standard rig Fig. 6.10 Measurements of applied current versus MR for the spin valve sample using mild steel (a) and permalloy (b) field coils. Fig. 6.9 shows the current versus MR measurement for the spin valve sample measurement using the mild steel (a) and permalloy (b) field coils. The inset shows the direction of the current I, anisotropy K and applied field H for clarity. The hysteresis from the mild steel core can be clearly seen compared to the permalloy core. The current axes were converted into 119 CHAPTER 6 field using the calibration measurements shown in Figs. 6.5 (a) and 6.7 (a), and the in-situ transport measurement results were compared to those obtained from a standard transport measurement rig. Fig. 6.10 compares the results for the mild steel (a) and permalloy cores (b) respectively. The in-situ transport measurements for the mild steel and permalloy cores compare well to the standard measurement rig, the graphs show the characteristic bell shape seen when the field is applied perpendicular to the anisotropy axis. The permalloy core sample follows the measurement rig particularly well for low fields, and shows an increasing deviation as the field increases, which emphasises the fact that it should only be used for low field measurements. The mild steel field coil follows the standard rig measurement closely up to higher fields, some deviation is seen because the sample is not properly saturated. The in- situ magnetoresistance measurements carried out using these fields coils measured between ±50Oe in order to avoid any saturation problems. 6.4.2 Field coils without a soft magnetic core. The field coils designed for sample holder 2 incorporate two coils on either side of the sample wound around a non-magnetic aluminium holder. Insulated 80µm diameter copper wire was used, and 3250 turns were wound on each holder. The coils were connected electrically in parallel in order to reduce the resistance seen by the current amplifier. The resistance of the coils was 200Ω. A hall probe was positioned in the centre between the coils, and the field was measured as a current was supplied to the coils. y = 0.8836x 0 20 40 60 80 100 120 0 20 40 60 80 100 120 140 Current (mA) Fi el d (O e) Fig. 6.11 Current versus field graph for the field coils in sample holder 2. Fig. 6.11 shows the current versus field graph for the coils. The field increases linearly with current and can be fitted with the equation y=0.8836x. Above ≈110mA the coils start to heat up, they can produce a maximum field of ≈95Oe and have sufficient uniformity for the 120 CHAPTER 6 micron and nano sized samples. 6.5 The Lock-in amplifier. The EG&G model number 5210 lock-in amplifier is an important part of the apparatus because it enables the measurement of small signals. The lock-in provides a DC output proportional to the AC signal under investigation and it is capable of measuring very small AC signals. REFERENCE INPUT PHASE SHIFTER OUTPUT OUTPUT AMPLIFIER LOWPASS FILTER B A.BA MIXER (P.S.D.) BANDPASS FILTER INPUT AMPLIFIER SIGNAL INPUT REFERENCE TRIGGER Fig. 6.12 Block diagram of a typical lock-in amplifier [2]. Fig. 6.12 shows a block diagram of a typical lock-in amplifier. The input signal to be measured is made to appear at a reference frequency. The signal is then amplified and applied to a phase sensitive detector (P.S.D.) operated at the reference frequency, which converts the AC input into the DC signal of interest. The output of the detector includes a value representing the amplitude of the signal of interest as well as components due to noise and interference, the AC noise can be reduced by selecting the appropriate filters. The lock-in amplifier has an internal oscillator, which is used to drive the current across the sample at the reference frequency. The voltage output from the sample is fed into a differential input of the lock-in, and the voltage information across the chip is derived from the amplifier at the reference frequency. A notch filter was set to limit the 50Hz noise and the lock-in gain typically varied between unity and 100 through the measurement. The lock-in amplifier output was calibrated at different frequencies using a spin valve supplied by Nordiko with the construction: Ta(50)/NiFe(25)/CoFe(20)/Cu(26)/CoFe(20)/Cu(10)/CoFe(20)/IrMn(100)/Ta(50). The MR, or ∆R/R, was derived from ∆V/V from the lock-in at different frequencies and the results were compared to a standard transport measurement rig. At 1kHz the MR from the lock-in 121 CHAPTER 6 was 9.34%, at 750Hz it was 9.32%, and at 500Hz it was 9.14%. According to the standard measurement rig the MR was 9.29%. The calibration shows that the lock-in is measuring correctly and that the response decreases with frequency. A standard reference frequency of 1kHz was chosen for the in situ measurement rig. 6.6 The Voltage-Offset Amplifier. The output of the lock-in amplifier is connected to the input of the voltage-offset amplifier. The voltage-offset amplifier is an important piece of the apparatus because it extracts the small change in the resistance of the sample in an applied field. The output signal from the lock-in includes the initial voltage of the sample in zero field due to the initial resistance and the change in the voltage in the presence of the applied field. However, the initial resistance is much larger than the change in resistance, which makes it difficult to extract the ∆R information. The voltage-offset amplifier offsets the initial R and amplifies the ∆R to increase the sensitivity of the signal. This amplifier was designed and built specifically for the measurement system. Fig. 6.13 shows the circuit diagram for the offset-amplifier. It is powered by two 9V batteries, V1 and V2 represent the voltage input signals from the lock-in amplifier. NE5532 op-amps were used because they have low noise and low distortion [3]. The first two op-amps process the incoming voltage; V1 passes through unchanged and V2 is inverted. The third op-amp adds the incoming voltages from the first two op-amps and inverts the resulting voltage giving V2-V1, it also filters out the low frequency noise. Standard voltage regulator LM317 was used to regulate the voltage from the batteries and set the voltage range for the potentiometer, which was used to offset the initial resistance. The voltage range of the potentiometer is given by the Vout from the voltage regulator, where Vout=1.25V[1+(R2/R1)] and R2 and R1 are the shown in Fig. 6.13. The potentiometer was set to offset between 8.38V and ground. The final op-amp combines the V2-V1 input and the offset voltage and amplifies ∆R using the gain settings. There are three gain settings; ×1, ×10, and ×50, and the higher gains are filtered using capacitors to reduce the noise. The pattern for the circuit board was designed using Freehand, and optical lithography was used to expose the design onto the resist covered copper coated circuit board. The copper etching was carried out using ferric chloride FeCl3. 122 CHAPTER 6 R2 1kΩ 10kΩ 50kΩ 4.7µF 1kΩ 4.7µF 1kΩ 1kΩ 4.7µF 1kΩ 1kΩ 1kΩ 1kΩ 1kΩ 1kΩ1kΩ I -V2 V1 - + + - V1 Rchip V2 (V2-V1) - + 10µF 670kΩ 670kΩ 8.38V 10µF • 9V 9V 4.7kΩ 1kΩ R1=1kΩ • • LM317 Vout ADJ Vin -V Offset - + - + G=×50 G=×10 G=×1 4.7µF OUT Fig. 6.13 Circuit diagram for the Voltage-Offset amplifier used in the in-situ magnetoresistance measurement rig. 123 CHAPTER 6 6.7 Signal Transmission and Noise Reduction. Fig. 6.14 shows the signal transmission through the in situ magnetoresistance measurement rig. Function generator 1 sends a low frequency triangular wave (0.05Hz) to the current amplifier, which amplifies the signal ×22 and sends the current to the field coils. The oscillator sends a high frequency (1kHz) sine wave to the current source, which drives the current across the chip. Function Generator 1 δR ∝ R δR Oscillator Chip Lock-in Amp Current Amp Offset Amp Oscilloscope Fig. 6.14 Schematic diagram showing the signal transmission through the in-situ magnetoresistance measurement rig. The voltage signal from the chip is fed into the differential input on the lock-in amplifier, which locks into the change in the 1kHz frequency voltage, filters the noise and amplifies the signal. The voltage offset amplifier offsets the initial signal which is proportional to R, and amplifies the voltage proportional to ∆R. Both the field voltage and the voltage from the offset amplifier are sent to the oscilloscope, which displays both signals on the screen. The voltage proportional to R is recorded from the lock-in display screen. 124 CHAPTER 6 The MR is given in Eqn (6.1) and is calculated from: IGG VR inlockampV oscill ××=∂ −− IG VR inlock inlock ×= − − inlockampV oscill VG V R R −− × =∆ (6.1) Where Voscill= the chip voltage from the oscilloscope, GV-amp= the gain from the voltage offset amplifier, Glock-in= the gain from the lock-in, Vlock-in= the gain from the lock-in amplifier and I= the r.m.s. current from the current source. Noise reduction was carried out by using shielded BNC and Lemo B-series connections, and by minimising the length of the cables. The effect of different grounding patterns was investigated by analysing the amount of noise in the transport measurement for different configurations. An unpatterned pseudo spin valve with the structure: Ta(2nm)/Ni80Fe20(15nm)/Cu(2.2nm)/Ni80Fe20(6nm)/Ta(2nm) was used for the measurement, with the current and field applied perpendicular to each other respectively. A wide range of different configurations were tested, Fig. 6.15 is a table showing two of the configurations tested; A and B. Apparatus A B Function generator × × Current source × × Power supply for current amp × × Lock-in amplifier × × Voltage-offset amplifier √ × Oscilloscope × √ Argon ion milling chamber √ √ Apparatus grounded=√ Apparatus not grounded=× Fig. 6.15 A table showing two different grounding pattern configurations; A and B tested to minimise the noise in the in-situ magnetoresistance measurement rig. The aim of the experiment was to determine which grounding pattern gave the least amount of noise. In configuration A the apparatus was grounded through a crows foot system; the voltage-offset amplifier was grounded by connecting to the main chamber, this created a single grounding path for all the apparatus at the output of the measuring system. In 125 CHAPTER 6 configuration B there were two separate grounding paths through the oscilloscope and the main chamber. Fig. 6.16 shows the transport measurements taken for the two different configurations. -60 -40 -20 0 20 40 60 Field (Oe) -0.04 0 0.04 0.08 0.12 0.16 0.2 -60 -40 -20 0 20 40 60 Field (Oe) M R (% ) (A) (B) Fig. 6.16 Transport measurements for two different grounding pattern configurations A and B in the in situ magnetoresistance measurement rig. The results show that configuration A gives the minimum noise with an average MR noise level of 0.02%. Configuration B gives an average noise level of 0.05% MR. These results show that the level of noise in the measurement can be minimised by connecting the apparatus grounds together, and connecting the single output ground to the main chamber at the final stage of the measurement system. 6.8 Rig calibration. During the in situ measurement the shutter is used to control the bombardment of argon ions onto the sample surface. The shutter is opened to allow milling of the sample, and closed during the transport measurement to prevent any further milling or noise in the measurement. The shutter is re-opened once the data has been acquired to continue milling. It is important to close the shutter during the transport measurement, because it can take up to 4 minutes to acquire the data and the milling rate of the sample is approximately 5nm per minute. At this rate the sample would mill through in 6.5 minutes. The shutter is used to allow a series of measurements to be taken at different milling depths through the structure, typically 20 transport measurements are taken for a complete in-situ measurement. Between each milling step, the resistance in the absence of field is recorded versus the total time of milling, and this is later converted into resistance versus milling depth to give an indication of how deep the sample has been milled for each transport measurement. The effect of the movement of the 126 CHAPTER 6 shutter was calibrated by carrying out resistance versus time measurements keeping the shutter continuously open, and also periodically opening and closing the shutter to investigate the difference. Pseudo spin valve samples with the construction: Ta(3nm)/Ni80Fe20(15nm)/Cu(2.2nm)/Ni80Fe20(6nm)/Ta(3nm) were used for the experiment. The samples were patterned into 36µm wide and 100µm long rectangles using optical lithography and argon ion milling. Sputtering and lift off were used to deposit copper contact pads, and electrical contact was made by wire bonding in a four-point measurement configuration. For the experiment with the shutter continuously open, the four-point measurement was connected to a Keithley multimeter, and the resistance from the multimeter was sent to the computer. Labview software was used to record the resistance versus time, measuring and recording a resistance value every second. 20 60 100 140 180 220 260 300 340 380 0 100 200 300 400 time (s) R es is ta nc e (Ω ) Open shutter Open/close shutter 0 50 100 150 200 250 300 350 400 0 100 200 300 400 Re sis ta nc e ( Ω) shutter open shutter open/close time (s) (b) (a) Fig. 6.17 (a) Resistance versus time measurements in the argon ion miller with the shutter continuously open and with the shutter opening and closing. (b) The corrected resistance versus time graph for the measurement with the shutter opening and closing compared to the continuous milling measurement. The in-situ measurement rig was used to measure the resistance versus time with the shutter periodically open and closed. The resistance was calculated from the current supplied to the chip and the voltage display on the lock-in. Fig. 6.17 (a) shows the change in resistance versus time when the shutter is continuously open and when it is periodically open and closed. The results show that the samples are fully milled through after 313s and 345s, giving a final resistance of ≈378Ω. After this final measurement the resistances became infinite. Although the shape of the graphs is similar, there is a time delay for the increase in resistance when the shutter is opening and closing. This could be due to some plasma instability due to the plasma 127 CHAPTER 6 being close to the gun [4] or a small oxidation of the surface during the measurement because the argon gas contains 3% oxygen. Assuming that the ‘dead time’ of the shutter is a constant every time it is opened, it can be calculated by dividing the difference in the fully milled through milling time of the two samples by the number of times the shutter is opened. This calculation gives a shutter ‘dead time’ of 1.3s, the resistance versus time data was adjusted to remove the shutter ‘dead time’ for each data point. Fig. 6.17 (b) shows the corrected resistance versus time graph for the measurement with the shutter opening and closing compared to the continuous milling measurement. The two graphs are very similar. In order to reliably predict the milling depth, a simple mathematical model was used to convert the resistance versus time data into resistance versus depth milled. 6.8.1 The Resistance Model. Lee et al. used a series resistance model to predict the magnetoresistance of a multilayer structure when the current was applied perpendicular to the plane of the film [5]. In their model the total change in resistance ∆R was calculated by summing the contribution to ∆R from each of the layers. This model was adapted for current in plane and applied to these measurements in order to predict the milling depth through the spin valve structures from the resistance versus time data in the in-situ measurement. The model visualises each layer in the spin valve as a separate resistance, and the total resistance is the addition of all the resistors in parallel. TanmPyCunmPytotal RRRRR 11111 )6()15( +++= Iout Iin RPy(6nm) RCu(2.2nm) RPy(15nm) RTa(3nm) Fig. 6.18 A schematic of the resistance model for the PSV structure with the current applied in the plane of the film. The equation for calculating the total resistance of the PSV is shown. Fig. 6.18 is a schematic diagram showing the resistance model for the pseudo spin valve structure. The Ta capping layer oxidised to TaO2 which is highly resistive, and has therefore been neglected from the model. The sheet resistance of each layer is calculated using 128 CHAPTER 6 ( )AR lρ= , where ρ=resistivity, ℓ=length and A=cross-sectional area. Assuming that the milling is even across the surface, each layer acts as a variable resistor during milling. Also shown in Fig. 6.18 is the equation used to calculate the total resistance of the PSV structure. If bulk values of resistivity are used in the model the calculated resistance is much lower than the measured resistance. It is known that thin film values of resistivity can be considerably larger than the bulk value [6]. Recent work carried out in our research group investigated the difference between the bulk and thin film resistivity values for NiFe, Cu and Nb, and the results are shown in Fig. 6.19. Material RT layer resistivity (10-8Ωm) RT Bulk resistivity (10-8Ωm) NiFe 23.6 (± 2.5) ∼15 Cu 22.3 (± 2.6) 1.7 Nb 157 (± 16) 14.5 Fig. 6.19 A table to show the difference in the resistivity values for thin film and bulk metals.[7] The increase in resistivity of the materials in thin film form is due to the increased surface scattering and the increased number of defects in sputtered films compared to the bulk. The defects are in the form of grain boundaries, impurities, dislocations, etc. When the values of the layer resistivity were substituted into the model, the calculated and measured values of resistance corresponded almost exactly. According to the reference books the thin film resistivity for Ta is 25-50µΩcm [8]. The calculated resistance value of a 36µm by 100µm area of spin valve without the Ta layer was 28.1Ω. By adding the Ta layer resistance to the spin valve resistance in parallel it is possible to calculate the total resistance of the structure. totalTasv RRR 111 =+ ( )( )( )( ) 1368 96 1016.2 101001050 10310361 −− −− −− Ω×=×× ××= TaR Ω= ×+= − 49.26 1016.2 1.28 11 3 total total R R The experimental value of the full spin valve structure using a four-point measurement was 26.13Ω, which is in close agreement to the calculated resistance using the resistance model 26.49Ω. 129 CHAPTER 6 6.8.2 In-situ milling depth calculation. The resistance model described above was used to predict the resistance of the pseudo spin valve as the structure is gradually milled through the 23.2nm thickness. Fig. 6.20 (a) shows the change in resistance as the structure is milled through the 15nm NiFe layer, the Cu layer and the 6nm NiFe layer. The shaded regions on the graph indicate the different layers in the structure. ) ce ( sis ta R e (a) (b) NiFe NiFe Cu 0 50 100 150 200 250 300 350 400 450 500 0 5 10 15 20 25 Thickness milled (nm) R es ist an ce ( Ω) Shutter open Shutter open/close Model Thickness milled (nm) 0 50 100 150 200 250 300 350 400 450 0 5 10 15 20 25 n Ω NiFe Cu NiFe Fig. 6.20 (a) A graph to show the change in resistance versus thickness milled for the pseudo spin valve calculated by the resistance model. The shaded regions indicate the different layers in the structure. (b) Resistance versus milling depth calculations for the sample with the shutter continuously open, for the shutter periodically open and closed, and the resistance model. According to the model the resistance of the pseudo spin valve is 397Ω when milled through to the end of the 6nm thick NiFe layer. According to the experimental data, the final resistance for both the sample measured by continuous milling and periodic milling is ≈378Ω. When the measured resistance is fitted to the predicted milling depth from the model, the shape of the graphs can be compared, as shown in Fig. 6.20 (b). The measurement when the shutter is periodically open and closed fits quite well to the model, the data for the continuous milling sample does not fit so well. One possible source of error in the measurement is a change in the average resistivity of the structure during milling. The model assumes that the resistivity of the layers remains constant throughout the milling. Fig. 6.21 shows a graph of the inverse resistance (1/R), versus the thickness (t) of the pseudo spin valve. 130 CHAPTER 6 1/ R (Ω - 1 ) 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05 0 5 10 15 20 25 Thickness (nm) Open shutter Open/close shutter Model Fig. 6.21 A graph of inverse resistance 1/R versus thickness t with the shutter continuously open, with the shutter periodically open and closed, and as calculated by the resistance model. The model assumes that the resistivity, length and width remain constant during milling, and the graph of 1/R versus t is linear, with the gradient equal to w/ρℓ. When the shutter is opened and closed through the measurement the graph is also approximately linear, and the shape of the graph follows the resistance model quite closely. When the shutter is continuously open the 1/R versus t data shows an increased deviation from linearity. The thicknesses of the layers in the pseudo spin valve are similar or smaller than the mean free path of the electrons, typical mean free paths in metals are of the order of tens or even hundreds of interatomic distances, and for NiFe≈15nm and Cu≈20nm. According to Mathon [9] when the mean free path is much longer than the thickness of the layers, the conduction electrons sample equally layers with low and high resistivity, and the resistivity of the layers can be given by an average resistivity calculated as shown in Eqn. (6.2). dcba dcba dcba +++ +++=− ρρρρρ (6.2) where a and ρa are the thickness and resistivity of the top NiFe layer, b and ρb are the thickness and resistivity of the Cu interlayer, c and ρc are the thickness and resistivity of the bottom NiFe layer and d and ρd are the thickness and resistivity of the Ta underlayer. Using Eqn. (6.2), the average resistivity of the structure is 26.5×10-8Ωm. When the average resistivity is derived from the gradient of the graph of 1/R versus t for the resistance model, 131 CHAPTER 6 132 the value obtained is 24×10-8Ωm, which is in close agreement with the equation shown above. The experimental value of the average resistivity when the shutter is open and closed equals 22.5×10-8Ωm, and when the shutter is continuously open the resistivity equals 21.18×10-8Ωm. The deviation in linearity of the measured resistivity when the shutter is continuously open is due to the effect of the argon ions bombarding the surface during the measurement. 6.9 Conclusions. An in-situ magnetoresistance measurement rig was successfully built and calibrated to allow direct analysis of the change in electrical and magnetic properties of spin valve devices during patterning. Different designs for applying the field have been investigated and it was found that coils with a soft magnetic core were more suitable for samples on the millimetre scale. For smaller samples field coils without cores were sufficient and more convenient to use. Noise reduction was achieved with the use of a lock-in amplifier and carefully arranged grounding patterns. A voltage-offset amplifier was designed to amplify the signal response and offset the initial resistance of the sample. Various techniques were used to ensure that the rig was properly calibrated. A mathematical resistance model was developed to calculate the depth milled from the resistance of the sample. CHAPTER 6 References [1] National Semiconductor (http://www.national.com/pf/LM/LM675.html) [2] Perkin Elmer Instruments (http://www.cpm.uncc.edu/programs/tn1000.pdf) [3] Texas Instruments Dual Low-Noise Operational Amplifiers (http://www.physics.brocku.ca/faculty/razavi/231/datasheets/slos075a.pdf) [4] G.G. Lister, Low-pressure gas discharge modelling, J. Phys. D: Appl. Phys., 25, 1649- 1680 (1992). [5] W.Y. Lee, S. Gardelis, B.-C. Choi, Y.B. Xu, C.G. Smith, C.H.W. Barnes, D.A. Ritchie, E.H. Linfield, J.A.C. Bland, J. Appl. Phys. 85(9), 6682 (1999) [6] M. Ohring, The Materials Science of Thin Films, Chapter 10.2.2, 457-463, Academic Press Ltd (1992) [7] D. Leung, Ph.D. thesis, Cambridge University (2003) [8] M. Ohring, The Materials Science of Thin Films, Chapter 10.2.2, 464, Academic Press Ltd (1992) [9] J. Mathon, Spin Electronics, Chapter 4, 84, (Eds.) M. Ziese, M.J. Thornton, Springer- Verlag Berlin Heidelberg (2001) 133 CHAPTER 7 The mathematical sciences particularly exhibit order, symmetry, and limitation; and these are the greatest forms of the beautiful. Aristotle (384-322 BC) Metaphysica. Chapter 7 In-situ Micropatterning of Pseudo Spin Valves The new test rig gives the opportunity to progressively measure the effect of changing selected parameters on the samples. Arrays of micro-wires with varying aspect ratio and equal mark/space ratio were patterned in ferromagnetic trilayers with the structure: Ta(3nm)/Ni80Fe20(6nm)/Cu(2.2nm)/Ni80Fe20(15nm)/Ta(3nm). The in-situ test rig was used to analyse the change in the magnetoresistance response as the shape anisotropy of the trilayer was increased during milling. The results confirm the measurement results from Chapter 5 by showing that it is possible to partially decouple the Ni80Fe20 ferromagnetic layers. The progressive measurements in this chapter show that a maximum spin valve response is observed when the structure is fully milled. The transport measurements show that magnetisation reversal occurs in two stages, with the lower switching field corresponding to the thinner ferromagnetic layer. It is also shown that as the shape anisotropy is further increased by decreasing the width of the micro-patterned wires, the spin valve response is magnified. Highly anisotropic structures were also patterned in PSVs with the construction Ta(3nm)/CoFe(6nm)/Cu(2.2nm)/CoFe(15nm)/Ta(3nm). The CoFe ferromagnetic layers do not become decoupled during patterning in the in-situ rig. This is thought to be due to larger magnetostatic and interlayer coupling energies. The Ni80Fe20 pseudo spin valve design shows a higher thermal stability than a conventional spin valve with an IrMn antiferromagnetic pinning layer. 134 CHAPTER 7 7.1 Motivations. The characteristics of magnetic stripes have been extensively studied because of their importance in both magnetoresistance devices [1-16] and, more recently magnetoelectronic devices [17,18]. There have been numerous efforts [1-16] to investigate the magnetisation reversal and MR behaviour in ferromagnetic wires so far. Very recent studies [9-13] have demonstrated that the switching field and magnetisation reversal process depend strongly on the end shape as well as the width of the ferromagnetic wires. The shape of a wire structure has a decisive influence on the magnetic properties in the micron size range. For example, the magnetisation reversal process and MR behaviour are found to change significantly in micron-sized Ni80Fe20 modulated width [15], “elbow”-shaped and cross-shaped wire structures [16] due to the modified shape-dependent demagnetising fields. The previous chapters have shown that it is possible to decouple two ferromagnetic layers in a spin valve sandwich by patterning a highly anisotropic structure in one of the layers using optical lithography and argon ion milling. An array of 3µm wide wires with equal mark-space ratio was patterned in the top layer of the structure, whilst the bottom layer remained unpatterned. The decoupling effect was shown to increase as the material between the wires was milled away to the Cu spacer, and a small spin valve response was observed. In this chapter we present results for the in-situ magnetoresistance response of micro-patterned wire arrays in spin valve structures. The results show the evolution in the transport measurements as the milling depth between the wires is increased, and also the effect of further increasing the shape anisotropy by changing the dimensions of the array. The change in MR was also investigated using different ferromagnetic materials in the trilayer. As mentioned previously, the practical disadvantage of using exchange anisotropy to pin one of the ferromagnetic layers in a spin valve is the reduction in pinning strength as the temperature is increased [19]. Temperature measurements were carried out to compare the pseudo spin valve devices with conventional spin valves with an IrMn antiferromagnetic pinning layer. The results of this study show how the spin valve response from the micro-patterned trilayers can be optimised in terms of the material properties and the structure of the ferromagnetic layers. 7.2 Sample preparation. Pseudo spin valves with the configuration: Ta(3nm)/Ni80Fe20(6nm)/Cu(2.2nm)/Ni80Fe20(15nm)/Ta(3nm) were diced into 10mm by 5mm samples. The samples were patterned into 4mm by 3.3mm rectangles using optical lithography and argon ion milling. Copper contact pads were deposited on either side of the 4mm long rectangles as shown in Fig. 7.1, using optical lithography and lift off. Fig. 7.1 also shows the direction of anisotropy in the two ferromagnetic layers, the electrical four-point 135 CHAPTER 7 measurement set-up and the field measurement directions 0° and 90°. The anisotropy direction of the top 15nm ferromagnetic layer is along the 4mm length of the sample, which also corresponds to the 0° measurement direction. Magnetic hysteresis measurements taken at 0° and 90° showed a single hysteresis loop, which confirmed the ferromagnetic coupling between the layers, with the magnetisation reversal being dominated by the thicker top ferromagnetic layer as explained in Chapter 5. The same optical lithography technique was used as summarised in Fig. 5.1 (Chapter 5) to pattern a 3µm wide wire array with equal mark- space ratio along the 4mm length of the rectangle, (see Fig. 5.9). The samples were wire bonded in a 4-point transport measurement configuration and loaded into the argon ion miller for in-situ magnetoresistance measurements. 90° = Ni80Fe20 = Cu 0° 4mm Kα 3.3mm V Kγ⊗ I Fig. 7.1 A schematic diagram showing the patterned pseudo spin valve including the anisotropy directions of the ferromagnetic layers, dimensions, the electrical four-point measurement, and the measurement directions 0° and 90°. 7.3 In-situ magnetoresistance measurements. Previous transport measurements in Chapter 5 showed an AMR and a GMR response when the wire array structure was milled through to the copper interlayer (Fig. 5.12). Using the magnetoresistance measurement rig, the results in this chapter study the change in the MR as the structure was fully milled through. Fig. 7.2 shows the evolution in the transport properties for the 3µm wide wire array from the in-situ measurement. Graph (a) shows the initial transport measurement before any milling. An AMR of ≈0.2% is seen, and the shape of the graph shows that the response is dominated by the anisotropy of the thicker top ferromagnetic layer, which is parallel to the applied field. The AMR response agrees with the magnetisation measurement along the 0° direction. Graph (b) shows that as the milling continues the AMR 136 CHAPTER 7 decreases and a small GMR is seen, this result was observed in Chapter 5. However as the milling is continued, graphs (c) and (d) show that the AMR disappears and only GMR is seen. A maximum GMR of 0.8% is observed when the sample is fully milled through. As described in Chapter 5, the appearance of a GMR signal as the milling depth increases is due to the two ferromagnetic layers becoming decoupled. The total energy of the structure before patterning is the sum of the Zeeman energy, the ferromagnetic interlayer coupling energy, the magnetostatic energy, and the perpendicular induced anisotropy energies. Initially the interlayer coupling energy and the magnetostatic energy dominate the induced anisotropy energies in the layers and they remain coupled together in an applied field. The results in Chapter 5 showed that as the shape anisotropy was gradually increased by patterning through the top layer of the pseudo spin valve, the coercivity of the top layer increased and eventually the shape anisotropy energy became larger than the interlayer coupling. The patterning changed the natural similarity of the two ferromagnetic layers and they started to move separately in an applied field creating a small GMR signal. Originally it was expected that if one layer was patterned and the other layer remained unpatterned, a maximum GMR would be seen because the energies in the two layers would be very different. - 0.4 - 0.2 0 0.2 0.4 0.6 0.8 1 M R (% ) Field (Oe) - 40 - 20 0 20 40 - 0.4 - 60 - 0.2 0 0.2 0.4 0.6 0.8 -40 -20 0 20 40 6 0 (c) (d) (a) (b) Max GMR= 0.8% M R (% ) Field (Oe) Fig. 7.2 In-situ magnetoresistance measurements showing the evolution in transport properties when an array of 3µm wide wires is patterned in a Ni80Fe20/Cu pseudo spin valve sample. 137 CHAPTER 7 The results shown in Fig. 7.2 are surprising because they show that a maximum GMR is observed when the two ferromagnetic layers are patterned with the same high shape anisotropy structure. The effect of further increasing the induced shape anisotropy was investigated by decreasing the width of the patterned wires. The length of the patterned wires was kept at a constant 4mm, but the wire width was reduced from 3µm to 2µm and 1.5µm, creating aspect ratios of 1:1333, 1:2000 and 1:2666 respectively. Fig. 7.3 shows the change in the transport measurement as the structure was milled through according to the in-situ magnetoresistance measurements for the 2µm wide wire array. Fig. 7.3 shows that the 2µm wide wire array shows the same trend as the 3µm wide wire array; the transport measurements show an initial AMR and a final GMR when fully milled through. However the maximum GMR is 2.21%, this result indicates that increasing the shape anisotropy increases the GMR. Field (Oe) - 0.1 0.4 0.9 1.4 1.9 2.4 - 0.6 - 60 -40 - 20 0 20 40 60 M R (% ) - 40 - 20 0 20 40 60 Field (Oe) - 0.1 0.4 0.9 1.4 1.9 2.4 M R (% ) (c) (d) (a) (b) Max GMR= 2.21% Fig. 7.3 In-situ magnetoresistance measurements showing the evolution in transport properties when an array of 2µm wide wires is patterned in a Ni80Fe20/Cu pseudo spin valve sample. Fig. 7.4 shows the in-situ measurements for the 1.5µm wide wire array, they also show the same trend with a small increase in GMR of 2.36%. 138 CHAPTER 7 7.4 Milling Depth Calculations. Chapter 6 introduced a parallel resistance model, which can be used to estimate the milling depth of the sample in the in-situ magnetoresistance rig. The model assumes that each layer in the pseudo spin valve acts as a separate resistance, and the resistance of each layer is equal to R=(ρℓ)/A, where ρ=resistivity, ℓ=length and A=cross-sectional area. The total resistance of the spin valve is the sum of the resistances of all the layers added in parallel. The resistance model can be adapted to allow for the wire array structure patterned through the spin valve multilayer. Assuming that the milling is even across the surface, each layer acts as a variable resistor during milling and can be sectioned into two parts; α and β, where α represents the unpatterned part and β represents the patterned part, as shown in Fig. 7.5. Max GMR= 2.36% (a) 60 4 0 20 0 20 40 Field (Oe) - 1 - 0.5 0 0.5 1 1.5 2 2.5 M R (% ) Field (Oe) 604 0 2 0 0 2 0 40 - 0.5 0 0.5 1 1.5 2 2.5 M R (% ) (d) (b) (c) Fig. 7.4 In-situ magnetoresistance measurements showing the evolution in transport properties when an array of 1.5µm wide wires is patterned in a Ni80Fe20/Cu pseudo spin valve sample. Figure 7.5 also shows the equation used to calculate the total resistance of the structure, which assumes the parts α and β as separate resistors in the model. According to the diagram, the total resistance of the top NiFe layer is the sum of the resistance of the α part plus the total resistance of each β-wire connected in parallel. When the layer is fully milled through the resistance is equal to the sum of the β-wires, because α is completely removed. 139 CHAPTER 7 Ar+ Ta Cu β β β β β α NiFe (6nm) NiFe (15nm) TanmPyCunmPyTotal RRRR number RR 111111 )6()15( +++    +    ×= αβ Fig. 7.5 A schematic diagram showing how the resistance model is used to calculate the total resistance of the pseudo spin valve as a micro-wire array is patterned through the top layer of the structure. During milling, each layer is divided into two parts: α and β, where α represents the unpatterned part and β represents the patterned part. The total resistance of the structure is calculated as shown by the equation. Depth milled (nm) 0 5 10 15 20 25 0 5 10 15 20 25 30 Re si sta nc e (Ω ) NiFe Cu NiFe Ta 0 20 40 60 80 100 0 5 10 15 20 25 30 Depth milled (nm) C ur re nt fl ow (% ) 15nm Cu 6nm Ta NiFe Cu NiFe Ta (b)(a) Fig. 7.6 The percentage current flow through each layer in the NiFe/Cu/NiFe/Ta PSV as the 3µm wire array is milled through the structure according to the resistance model.(b) A graph to show the change in the resistance with milling depth calculated by the resistance model for the micro-patterned PSV structures. 140 Figure 7.6 (a) shows the percentage current flow through each layer in the PSV as the 3µm wire array is milled through the structure according to the calculated resistance model. CHAPTER 7 141 Although a constant current is applied to the pseudo spin valve, the change in the shape of the layers during milling changes the current distribution throughout the structure. The graph shows that most of the current flow is through the two ferromagnetic layers with less than 15% traveling through the Cu interlayer and the Ta underlayer throughout the milling. In order to achieve accurate transport measurements it is important that the current is not shunted away from the ferromagnetic layers by lower resistance channels. Also, according to Chapter 6 the mean free path of the electrons in the Ni80Fe20 layer is approximately equal to the thickness of the layer before milling, i.e. ≈15nm. In Cu the mean free path is ≈20nm, which is much longer than the 2.2nm thickness. When the thickness is equal to or less than the mean free path, the electrons are forced to be distributed through the spin valve structure whatever the total resistance of the layers. In other words they travel through the structure without being scattered, apart from at the interfaces of the layers. Although the resistance of each layer and therefore the current distribution through the structure changes during milling, the MR information is not effected because the electrons are scattered through the entire structure. Fig. 7.6 (b) shows the resistance versus depth milled of the micro-patterned PSV as calculated by the resistance model. The graph represents the change in resistance with milling for the 3µm, 2µm and 1.5µm width wires, because although the wire width is decreased, the number of wires in a given area increases and therefore the total resistance of the spin valve structure remains the same. There are no obvious transitions regions between the Ni80Fe20 and Cu layers because the thin film resistivity values are similar, as shown in Fig. 6.19 in Chapter 6. A change in gradient can be seen when the milling continues through to the Ta underlayer because the resistivity is higher. The resistance versus time data from the micro-patterning in-situ measurements was converted into resistance versus milling depth using the resistance model. The change in the shape of the magnetoresistance measurements could then be analysed with respect to the milling depth. Fig.7.7 shows how the AMR and GMR of the pseudo spin valve changed as the milling depth increased with patterning 2µm and 1.5µm wide wire arrays. The two graphs show the same trend; the AMR gradually decreases and the GMR rapidly increases with milling through the bottom ferromagnetic layer. The resistance of the sample at the peak GMR was input into the resistance model, which showed a milling depth of 23.4nm, i.e. when fully milled through. Fig. 7.8 shows how the shape of the in-situ magnetoresistance measurements can be characterised in terms of the three reversal fields H1, H2 and H3. H1 is the switching field of the AMR signal, H2 is the field representing the peak of the GMR signal and H3 is the field required to saturate the GMR signal. CHAPTER 7 NiFe 0 0.5 1 1.5 2 2.5 0 5 10 15 20 25 M R (% ) AMR GMR Milling depth (nm) Cu NiFe 2µm 0 5 10 15 20 25 Milling depth (nm) AMR GMR NiFe Cu NiFe 1.5µm Fig. 7.7 Graphs showing how the AMR and GMR of the pseudo spin valve structure changes as an array of 2µm (a) and 1.5µm (b) wide wires is patterned through the structure. H 1 H 2 H 3 M R (% ) Fig. 7.8 A graph of MR versus field demonstrating how the shape of the graphs can be characterised in terms of three reversal fields H1, H2 and H3. 2µm NiFe Cu NiFe 5 10 15 20 25 30 0 0 5 10 15 20 25 Milling depth (nm) C oe rc iv ity (O e) H1 H2 H3 0 5 10 15 20 25 H1 H2 H3 NiFe Cu NiFe1.5µm Milling depth (nm) Fig. 7.9 Graphs to show how the three reversal fields H1, H2 and H3 change as an array of 2µm (a) and 1.5µm (b) wide wires is patterned through the PSV structure. 142 CHAPTER 7 Fig. 7.9 shows how the three reversal fields vary during milling of the wire arrays for (a) the 2µm wide wire array and (b) the 1.5µm wide wire array. Figures 7.7 and 7.9 show that although the AMR decreases as the milling depth increases, the reversal field H1 remains approximately the same. The large increase in GMR as the structure is milled through the bottom ferromagnetic layer is accompanied by a small increase in reversal field H2 of the peak GMR. The most significant change in reversal field with milling depth is shown by H3, the field required to saturate the MR signal. Fig. 7.9 also shows how H1, H2 and H3 change for the 1.5µm wide wire array. The change in shape is very similar to the 2µm wide wire array, except that H2 and H3 are larger with increasing milling depth. Fig. 7.10 compares the final in-situ transport measurement of the 2µm wire array with the magnetic hysteresis after the sample was removed from the chamber. -0.5 0 0.5 1 1.5 2 2.5 -60 -40 -20 0 20 40 60 Field (Oe) M R (% ) -0.15 -0.1 -0.05 0 0.05 0.1 0.15 M ag ne tis at io n (m em u) M 3 2 1 Fig. 7.10 A graph comparing the transport and the magnetisation measurements for the fully milled through 2µm wide wire array in the pseudo spin valve structure. The numbers 1, 2 and 3 indicated three reversal steps in the magnetisation reversal, and the magnetisation M is equal to the total magnetisation of the wire array. The magnetisation measurement shows three reversal steps 1, 2 and 3. Step 1 corresponds to a small overlap between the spin valve material and the copper contact pads, which is not seen in the transport measurement. The magnetisation M represents the total magnetisation of the wire array, and steps 2 and 3 correspond to two parts in the magnetisation reversal. The value of step 1≈0.02memu and the initial magnetisation saturation before patterning of the pseudo spin valve ≈0.23memu. This shows that the overlap with the copper contact pad is ≈0.16mm on either side of the sample. The total magnetisation of the wire array ≈0.189memu, step 2 equals approximately 29% for the total magnetic material which corresponds to the 6nm thick NiFe layer. The magnetisation reversal of step three is proportional to the thickness of the 15nm thick ferromagnetic layer. Fig. 7.10 shows that the first part of the transport 143 CHAPTER 7 measurement from zero field until the GMR peak corresponds to the reversal of the bottom 6nm NiFe layer. The change in the transport measurement from H2 to H3 is due to the reversal of the 15nm thick NiFe layer. 7.5 Induced anisotropy and shape anisotropy. The effect of patterning the wires parallel and perpendicular to the induced anisotropy axis of the PSVs was also investigated. The initial measurements were carried out with the configuration shown in Fig. 5.9 (Chapter 5), where the shape anisotropy is in the same direction as the induced anisotropy in the thicker ferromagnetic layer. Fig. 7.11 shows the second configuration, where the shape anisotropy is perpendicular to the induced anisotropy during deposition. For both configurations, the field was applied along the length of the wires. Fig. 7.11 A schematic diagram showing the 3µm wide wire array with equal mark-space ratio being transferred into the pseudo spin valve during argon ion milling. The diagram also shows the anisotropy direction of the two ferromagnetic layers induced during deposition, Kα and Kγ. The induced anisotropy of the thicker ferromagnetic layer is perpendicular to the length of the patterned wires. Fig. 7.12 shows the evolution in the magnetic hysteresis as the 3µm wire array structure is patterned through the thickness of the Ni80Fe20 trilayer when the induced anisotropy in the thicker ferromagnetic layer and the shape anisotropy are (a) parallel, and (b) perpendicular. The insets in Fig. 7.12 show the relative orientation of the induced anisotropy (Kα and Kβ) in the ferromagnetic layers with respect to the shape anisotropy (Ks). In (a) the applied field and induced anisotropy in the top layer (Kα) are parallel, and a square hysteresis loop is obtained for the plain film. In (b) the applied field and Kα are perpendicular, the plain film measurement shows less hysteresis and the shape is more characteristic of magnetisation rotation. This indicates that the magnetisation reversal is dominated by the induced anisotropy 144 CHAPTER 7 in the thicker ferromagnetic layer. At a mill depth of ≈11nm, the results show that the magnetisation reversal still retains some of the characteristics of the plain film measurement, but there is a definite change in shape corresponding to the increased shape anisotropy energy, which changes the magnetisation reversal mechanism. -0.4 -0.2 0 0.2 -20 -10 0 10 20 Field (Oe) M ag ne tis at io n (m em u) Milled -20 -10 0 10 20 Field (Oe) Milled Kγ Ks Kα Kγ Ks Kα Plain 11.7nm 16.3nm Plain 10.4nm 16.9nm 0.4 (a) (b) Fig. 7.12 Hysteresis measurements for the Ni80Fe20 trilayer with a 3µm wide wire. The graphs show the evolution in the magnetic hysteresis as the anisotropic structure is patterned through the thickness when the shape anisotropy and the induced anisotropy in the thicker ferromagnetic layer are (a) parallel and (b) perpendicular. The insets show the relative orientations of the anisotropy in the thicker ferromagnetic layer Kα, the thinner ferromagnetic layer Kγ and the patterned shape anisotropy Ks. As the milling depth increases the shape anisotropy dominates and the fully milled through hysteresis measurements in (a) and (b) are similar. This result shows that the shape anisotropy energy is much larger than the anisotropy energy induced during deposition of the ferromagnetic layers. It also shows that the direction of the induced anisotropy energy doesn’t need to be considered in the pseudo spin valve design when the shape anisotropy is large. 7.6 Thermal Stability Comparison. Transport measurements were carried out at temperatures ranging between 291K and 493K to compare the thermal stability of the 2µm patterned PSV with an exchange biased spin valve of the construction: Ta(5nm)/NiFe(2.5nm)/CoFe(2nm)/Cu(2.6nm)/CoFe(2nm)/IrMn(10nm)/Ta(5nm). The conventional spin valve uses an IrMn antiferromagnetic pinning layer to create a uniaxial anisotropy in the adjacent CoFe ferromagnetic layer. The coupled NiFe and CoFe layers 145 CHAPTER 7 underneath the copper reverse their magnetisation freely in an applied field. The samples were wire bonded and placed on a sample holder positioned in the centre of a quartz tube, which was sealed at one end to contain the heat. The quartz tube was resistively heated using a steel coil, which was wrapped in both directions to avoid any unwanted magnetic fields. The temperature was carefully monitored using a thermocouple attached to the sample holder in close proximity to the sample. The quartz tube was placed between the pole pieces of a magnet and four-point transport measurements were taken as the field was swept. Fig. 7.13 shows how the shape of the MR measurements changed as the temperature increased. Only half of the measurement (from positive to negative field sweep) is shown for clarity. 0 2 4 6 8 -50 0 50 Field (Oe) M R (% ) 291K 373K 423K 493K 0 2 4 -50 -40 -30 -20 -10 0 Field (Oe) M R (% ) 291K 373K 423K 493K (a) (b) Fig. 7.13 Graphs showing how the transport measurement changes with increasing temperature for (a) the conventional spin valve with an IrMn antiferromagnetic pinning layer, and (b) the 2µm wire array patterned pseudo spin valve. At room temperature the conventional spin valve shows the characteristic shape with a sharp change in MR close to zero field, which is caused by the free layer changing direction relative to the pinned ferromagnetic layer. As the temperature is increased the MR starts to decrease, partly due to a decrease in the spontaneous magnetisation of the NiFe layer, but also due to a decrease in the pinning properties of the IrMn antiferromagnetic pinning layer. As the pinning properties are decreased, the three ferromagnetic layers become more coupled together, and the relative orientations in an applied field become similar. The Curie temperature (Tc) of the NiFe layer is between 673 and 693K, and for CoFe the Tc is much higher ≈1253K [20]. The blocking temperature of IrMn is ≈523K [21]. As the measurement temperature approaches the IrMn blocking temperature, there is a large decrease in MR and the gradient of the graph also changes. The remaining MR≈1.77% is due to the difference in coercivities of the NiFe and CoFe layers. The PSV with the 2µm wire array also shows a decrease in MR due to the decrease in spontaneous magnetisation of the NiFe, however the relative change in MR over 146 CHAPTER 7 the range of temperature is less than for the conventional spin valve. As the temperature increases the thermal energy seems to aid the magnetisation reversal and the graph becomes narrower and shifts closer to zero. Fig. 7.14 shows more clearly how the MR decreases with temperature for the conventional spin valve and the pseudo spin valve. Temperature (K) 0 0.2 0.4 0.6 0.8 1 250 300 350 400 450 500 550 R el at iv e M R IrMn SV 2µm PSV Fig. 7.14 Graphs showing how the MR ratio decreases with increasing temperature for (a) the conventional spin valve with an IrMn antiferromagnetic pinning layer, and (b) the 2µm wire array patterned pseudo spin valve. Fig. 7.14 shows that the patterned pseudo spin valve has a higher thermal stability than the conventional spin valve above about 423K 7.7 Micro-patterning of CoFe Pseudo Spin Valves. The previous results have shown that it is possible to create a PSV using NiFe and Cu, the next part of the experiment investigated whether the same effect could be observed using different materials. A pseudo spin valve with the construction: Ta(3nm)/CoFe(6nm)/Cu(2.2nm)/CoFe(15nm)/Ta(3nm) was supplied by Nordiko. The spin valve was diced into 10mm by 5mm chips with the same anisotropy orientation as in the previous experiment; the anisotropy of the top 15nm thick ferromagnetic layer was parallel to the length of the chip and the anisotropy of the 6nm layer was perpendicular to the length. A VSM measurement was carried out to investigate the magnetisation reversal of the plain film. The same optical patterning and in-situ analysis was carried out to create a 1.5µm wide wire array in the CoFe/Cu/CoFe spin valve. After patterning the sample, another VSM measurement was taken in order to compare the before patterning and after patterning magnetisation reversal. Fig. 7.15 shows the fully milled through transport measurement and both VSM measurements. 147 CHAPTER 7 -2 -1 0 1 2 3 -500 -250 0 250 500 Field (Oe) M ag ne tis at io n (m em u) -0.02 -0.01 0 0.01 0.02 M R (% ) Plain VSM Milled VSM MR CoFe Fig. 7.15 Graphs showing the magnetisation reversal of the unpatterned CoFe/Cu/CoFe spin valve along the 0° direction, the magnetisation reversal of the spin valve after patterning a 1.5µm wide wire array, and the transport measurement for the fully milled through structure along the length of the wires. The magnetisation reversal of the CoFe spin valve before patterning shows a single reversal loop, and the shape is characteristic of the anisotropy in the thicker ferromagnetic layer. This shows that the thicker ferromagnetic layer is dominating the reversal process, the same as for the unpatterned NiFe spin valve. However, the coercivity and magnetic saturation of the CoFe spin valve are larger than for the NiFe spin valve. For CoFe the Ms= 1.72Oe and Hc=7.5Oe, for NiFe Ms= 0.89Oe and Hc≈1Oe. When the 1.5µm wire array is patterned through the CoFe spin valve, the coercivity is considerably increased and the susceptibility decreases, but the hysteresis loop shows a single transition indicating that the two ferromagnetic layers are moving together. This is verified by the transport measurement, which shows a small AMR≈0.01% and the shape is characteristic of the anisotropy in the thicker ferromagnetic layer. This result shows that the induced shape anisotropy greatly increases the coercivity of the CoFe PSV, but it is not large enough to overcome the coupling between the ferromagnetic layers and allow them to move separately. As explained in Chapter 5, the coupling between the two ferromagnetic layers is due to magnetostatic energy and the interlayer coupling energy. The magnetostatic energy is directly related to the saturation induction Bs of the material, and Bs for CoFe≈2.45T and for Ni80Fe20≈0.7T. The saturation induction is 3.5× larger for CoFe than for NiFe, which is an indication that the magnetostatic energy is larger for CoFe than for NiFe. The interlayer coupling energy is related to the thickness of the non- magnetic interlayer and the exchange coupling constant Aex. The thickness of the non- 148 CHAPTER 7 magnetic interlayer is the same for both spin valves, so the interlayer coupling can be directly compared by the exchange coupling constant Aex. For CoFe Aex=3.0×10-11J/m, almost five times larger than for NiFe where Aex=6.25×10-12 J/m. This shows that the coupling between the CoFe PSV layers is stronger than for the NiFe layers due to larger magnetostatic and exchange coupling interactions. 7.8 Conclusions. hen an array of 3µm wide wires are patterned on the top layer In-situ transport measurements on CoFe trilayers did not show a spin valve response with for high density operation. Previous work has shown that w of a ferromagnetic sandwich structure and milled through to the copper interlayer, a small spin valve response is shown. The results presented in this chapter show the progression of the change in the transport measurements as the milling depth is increased to the copper layer and then through the bottom layer to the substrate. The results in Chapter 5 suggest that the AMR can be attributed to the magnetisation rotation of the unpatterned layer, and the GMR corresponds with the switching of the patterned layer. These results support this theory and also show that increasing the milling depth through the bottom layer of the spin valve structure can decrease the AMR and increase the GMR of the samples. Finally it is shown that the maximum GMR is achieved when the material between the wires is completely removed. In this case the GMR is caused by the switching of the thinner layer at lower switching fields and the thicker layer at higher switching fields. The results show that there is an increase in GMR as the anisotropy of the structure is further increased by patterning with thinner wires. Magnetic hysteresis measurements have shown that the direction of the induced anisotropy during deposition does not have a significant effect on the device operation. At large milling depths the shape anisotropy dominates over the induced anisotropy. Temperature measurements have shown that the pseudo-spin valve has a better thermal stability than a conventional spin valve with an antiferromagnetic pinning layer. According to these results optimum device performance can be achieved using materials with low magnetostatic and interlayer coupling energies. However, the interlayer coupling energy is directly related to the Curie Temperature Tc, and a high Curie temperature is required for good thermal stability of the device. Therefore a compromise between high thermal stability and low interlayer coupling energy will need to be made for this device design. The results show that the optimum device design will have high shape anisotropy. increasing shape anisotropy. This is believed to be due to larger magnetostatic and interlayer coupling energies. We anticipate that it will be possible to extend this work to the nanoscale 149 CHAPTER 7 References [1] J. H. J. Fluitman, Thin Solid Films 16, 269 (1973) [2] M. H. Kryder, K. Y. Ahn, N. J. Mazzeo, S. Schwarzi, and S. M. Kane, IEEE Trans. Magn. 16, 99 (1980). [3] C. Tsang and S. K. Decker, J. Appl. Phys. 53, 2602 (1982) [4] E. J. Ozimek and D. I. Paul, J. Appl. Phys. 55, 2232 (1984) [5] J. F. Smyth, S. Schultz, D. R. Fredkin, D. P. Kern, S. A. Rishton, M. Cali, and T. R. Koehler, J. Appl. Phys. 69, 5262 (1991) [6] C. Shearwood, S. J. Blundell, M. J. Baird, J. A. C. Bland, M. Gester, H. Ahmed, and H. P. Hughes, J. Appl. Phys. 75, 5249 (1994) [7] K. Hong and N. Giordano, Phys. Rev. B 51, 9855 (1995) [8] A. O. Adeyeye, J. A. C. Bland, C. Daboo, D. G. Hasko, and H. Ahmed, J. Appl. Phys. 82, 469 (1997) [9] K. J. Kirk, J. N. Chapman, and C. D. W. Wilkinson, Appl. Phys. Lett. 71, 539 (1997); M. Rührig, B. Khamsehpour, K. J. Kirk, J. N. Chapman, P. Aitchison, S. McVite, and C. D. W. Wilkinson, IEEE Trans. Magn. 32, 4452 (1996) [10] T. Schrefl, J. Fidler, K. J. Kirk, and J. N. Chapman, J. Magn. Magn. Mater. 175, 193 (1997) [11] J. Shi, T. Zhu, M. Durlam, E. Chen, and S. Tehrani, IEEE Trans. Magn. 34, 997 (1998); J. Shi, S. Tehrani, T. Zhu, Y. F. Zheng, and J. G. Zhu, Appl. Phys. Lett. 74, 2525 (1999) [12] J. McCord, A. Hurbert, G. Schröpfer, and U. Loreit, IEEE Trans. Magn. 32, 4806 (1996) [13] J. Gadbois, J. G. Zhu, W. Vavra, and A. Hurst, IEEE Trans. Magn. 34, 1066 (1998) [14] R. D. Gomez, T. V. Luu, A. O. Pak, K. J. Kirk, and J. N. Chapman, J. Appl. Phys. 85, 2525 (1999). [15] C. C. Yao, D. G. Hasko, Y. B. Xu, W. Y. Lee, and J. A. C. Bland, J. Appl. Phys. 85, 1689 (1999) [16] W. Y. Lee, Y. B. Xu, C. A. F. Vaz, A. Hirohata, H. T. Leung, C. C. Yao, B. C. Choi, J. A. C. Bland, F. Rousseaux, E. Cambril, and H. Launois, IEEE Trans. Magn. 35, 3883 (1999) [17] W. Y. Lee, S. Gardelis, B. C. Choi, Y. B. Xu, C. G. Smith, C. H. W. Barnes, D. A. Ritchie, E. H. Linfield, and J. A. C. Bland, J. Appl. Phys. 85, 6682 (1999) [18] F. G. Monzon, D. S. Patterson, and M. L. Roukes, J. Magn. Magn. Mater. 195, 19 (1999) [19] A. E. Berkowitz, K. Takano, J. Magn. Magn. Mater. 200, 552-570, (1999) [20] J. Evetts, Concise Encyclopaedia of Magnetic and Superconducting Materials, 216 and 352, Pergamon Press Ltd (1992) [21] J. Nogués, I. Schuller, J. Magn. Magn. Mater. 192, 203-232, (1999) 150 CHAPTER 8 Our doubts are traitors, And make us lose the good that we oft may win, By fearing to attempt. William Shakespeare. Chapter 8 In-Situ Nanopatterning of Pseudo Spin Valves. The change in the electrical properties of nano-patterned Pseudo Spin Valve (PSV) structures of the form Ta(2nm)/Ni80Fe20(15nm)/Cu(2.2nm)/Ni80Fe20(6nm)/Ta(2nm) have been studied in-situ during patterning. Magnetoresistance measurements were taken as the shape of the nanowire array was milled through the PSV structure using a SiO2 hard mask designed using Focused Ion Beam (FIB) lithography. The results show that as the shape anisotropy is increased by increasing the milling depth of the nanowires, the MR increases. Further increasing the shape anisotropy by decreasing the width of the wires also increases the MR, and the ferromagnetic materials are shown to be sensitive to the gallium ion dosage. Micromagnetic simulations confirm the experimental results and emphasise the importance of the closure domains at the end of the PSV wires in the magnetisation reversal process. 151 CHAPTER 8 8.1 Motivations Due to the recent advances in lithographic patterning techniques, it has been possible to study the magnetisation reversal and electrical properties of submicron mesostructures with a range of materials, geometries and dimensions [1-26]. From a fundamental point of view, as the particle dimensions become comparable to characteristic lengths such as the magnetic domain-wall size, interesting behaviour may be expected. The study of nanomagnetics is also important in applied physics, mainly due to the applications in data storage devices and sensors. High density data storage devices and magnetoelectronic devices are now being designed to take advantage of the properties of small magnetic structures. For example, arrays of small magnetic elements have been proposed as patterned media, in which each uniaxial element stores one bit of data [27-31]. Ferromagnetic wires can be engineered with high shape anisotropy so that the magnetisation is directed either parallel or antiparallel to the wire axis [32,33]. Magnetic logic gates have been suggested in which data can be manipulated through magnetostatic interactions between chains of magnetic elements [34]. Different mechanisms have been proposed for the magnetisation reversal of sub-micron mesostructures including coherent rotation [35], fanning [36], curling [37], or buckling [37]. More recently, experimental results have shown that other factors also need to be considered including edge roughness [38] and non-uniform magnetisation configurations [39]. There has been some discrepancy between the theory and experimental results obtained from nanopatterning PSV structures. The theory predicts that as the dimensions are reduced and the layers approach single domain, the amplitude of the GMR should increase due to the more perfect alignment of magnetisation within the layers [40]. However, the experimental results indicate that the GMR tends to decrease as the dimensions are reduced [41-43]. According to the authors, the electron scattering at the pattern edges plays a greater role in the electron transport when the pattern becomes narrow. Also, the ion-milling process can affect the transport behaviour at the edges of a pattern, which becomes significant in the narrow pattern. Both of these factors contribute towards the decrease in GMR. However, the experimental results tend to agree that as the dimensions are reduced, the coercivity of both the hard and soft layer is increased. The advantage of the in-situ magnetoresistance measurements carried out in this chapter is that they allow direct analysis of the change in the initial resistance R, and the change in the resistance in an applied field ∆R, during patterning. It is therefore possible to derive how much of the change in GMR is due to a change in the initial resistance of the structure and how much is due to the change in domain configuration. 152 CHAPTER 8 Magnetotransport measurements have been shown to offer a sensitive probe of extremely small quantities of magnetic material, for example some work by Ono et al. studied the magnetic domain wall propagation in a single submicrometer wire [33]. Therefore, it is an ideal technique for characterising the change in electrical and magnetic properties of the nanopatterned PSVs. The FIB has been shown to produce high quality magnetic nanostructures in thin films of soft magnetic materials such as permalloy [44]. Other authors have demonstrated that Ga+ ion irradiation can comprehensively modify the ferromagnetic properties of Ni80Fe20 thin films for high dosage densities [45]. This chapter starts by describing the sample preparation and analysis before nanopatterning, including transport measurements and Bitter technique imaging of the micro-patterned NiFe/Cu/NiFe mesostructures. Analysis was also carried out on CoFe/Cu/CoFe mesostructures. The next part describes the patterning in the FIB including the milling depth calibration using end-point detection, and the electrical and structural characterisation of the nanopatterned SiO2 hard mask. Then in-situ measurement results are shown for three different wire widths nanopatterned in the PSVs. The final part of the chapter compares the experimental results with micromagnetic simulation. 8.2 FIB sample preparation and analysis. Pseudo spin valves with the structure: Ta(2nm)/Ni80Fe20(15nm)/Cu(2.2nm)/Ni80Fe20(6nm)/Ta(2nm) were micro-patterned using optical lithography and argon ion milling into a series of rectangles with lengths from 100 to 600µm and widths from 2µm to 45µm. The patterned rectangles showed aspect ratios from 1:2 to 1:300. The mask design shown in Fig. 4.5, chapter 4 was used for the lithography. This is an important step because it enables mapping of the chip during patterning in the FIB, and the position of each PSV rectangle is precisely known. Copper contact pads were deposited by sputtering. Each PSV rectangle was characterised using transport measurements to check the initial response of the structures before nanopatterning. Standard four-point measurements were carried out with current I=800µA applied along the length of the chip and the field applied along both the length (0°) and along the width (90°). The anisotropy direction of the 15nm Ni80Fe20 layer was along the length of the rectangle, and the anisotropy direction of the 6nm Ni80Fe20 layer was along the width of the rectangle. Fig. 8.1 shows the transport measurements with the field applied at 0° and 90° for the 36µm wide pseudo spin valve rectangle. The measurement is clearly dominated by the AMR of the top layer, which is in agreement with the results from the previous chapters. The longitudinal AMR is 0.16%, and the transverse AMR is 1.68%. 153 CHAPTER 8 -0.4 0 0.4 0.8 1.2 1.6 2 -100 -50 0 50 100 Field (Oe) M R (% ) 0deg 90deg Fig. 8.1 MR measurements with the field applied at 0° and 90° to the direction of the current for a 100µm by 36µm NiFe/Cu/NiFe PSV rectangle. -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 M R (% ) 2.2 3.3 4.2 4.2 6.7 8.3 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 -80 -60 -40 -20 0 20 40 60 80 Field (Oe) M R (% ) 250 300 8.3 Fig. 8.2 Consecutive transport measurements for the Ni/Cu/Ni PSVs showing the gradual change from AMR to GMR as the shape anisotropy is increased by patterning higher aspect ratios from 1:2.2 to 1:300. The decreasing field sweep from positive to negative field is shown for clarity. 154 CHAPTER 8 Fig. 8.2 shows the results from the transport measurements as the shape anisotropy of the structures was increased by increasing the aspect ratio from 2.2 to 300. For clarity, only half the graph is displayed. Initially as the aspect ratio increases the AMR increases because the shape anisotropy forces a more parallel alignment between the magnetisation and the current. However, as the aspect ratio continues to increase the AMR starts to decrease and a small GMR is shown. According to previous authors an increase in the shape anisotropy can lead to a decrease in the magnetostatic coupling between the two ferromagnetic layers [46]. A decrease in the magnetostatic coupling allows the ferromagnetic layers to move individually, and at an aspect ratio of 6.7 the transport measurement clearly shows a combined AMR and GMR. For large aspect ratios there is no AMR and the GMR reaches a peak value of 0.2%. The coercivity also increases as the shape anisotropy increases. If the coercivity is taken as the peak of the GMR, it increases from -8Oe to -27Oe. The results from the previous chapters have shown that the thinner layer reverses at a lower field than the thicker NiFe layer, and the high anisotropy structures show a distinct two-step reversal corresponding to the two ferromagnetic layers. 0 0.1 0.2 0.3 0.4 0.5 2 3 4 5 6 7 8 9 Aspect ratio M R (% ) AMR GMR R Aspect ratio R (Ω ) 8 58 108 158 208 258 0 20 40 60 80 100 120 140 160 AMR GMR R (a) (b) Fig. 8.3 Graphs showing how the AMR, GMR and R change for low aspect ratios (a) and for high aspect ratios (b). Fig. 8.3 shows how the AMR, GMR and R change for low aspect ratios (a) and for high aspect ratios (b). For low aspect ratios the AMR increases, peaking at about 4, and then decreases. The R and the GMR follow the same trend and increase as the aspect ratio increases. For high aspect ratios the AMR drops to zero, and the GMR and R continue to rise gradually until the R rises steeply and the GMR drops off. This data indicates that the decrease in GMR at high ratios is related to the increase in the initial resistance. The AMR decreases when the shape anisotropy is high, due to the difficulty in forming a multi-domain 155 CHAPTER 8 structure. When the shape anisotropy is high the magnetisation will tend to align along the length of the sample in zero field instead of forming a multi-domain configuration. The initial increase in GMR is because the ferromagnetic layers become decoupled due to a decrease in magnetostatic coupling through domain walls in the layers. These results are in agreement with the previous chapters, but it is interesting to note that the shape anisotropy starts to change the magnetisation reversal process at much smaller ratios than previously thought necessary. However, it is also worth noting that the GMR is much smaller (0.2%) with the lower aspect ratio. In the previous chapter a maximum MR of 2.36% was observed for 1.5µm wide wires with an aspect ratio of 2666. A bitter technique image was taken of a patterned NiFe/Cu/NiFe PSV sample with 50µm wide tracks in the absence of an applied field as shown in Fig. 8.4. The anisotropy of the top ferromagnetic layer, Ku, is along the width of the tracks as shown in the diagram. The anisotropy of the bottom layer is along the length. The previous results have shown that the magnetisation is dominated by the top ferromagnetic layer, and the Bitter technique image shows a multi-domain structure. Since the layers are coupled, we can assume the same domain configuration in both layers. It is possible to estimate the total anisotropy energy of the PSV bilayer from the bitter technique image by summing the energy contributions of the domains and the domain walls and then minimising the total energy as shown in the calculation below. w=50µm l=250µm Ku Fig. 8.4 A Bitter Technique image of the NiFe/Cu/NiFe pseudo spin valve patterned into 50µm wide tracks. The anisotropy direction of the top ferromagnetic layer, Ku, is along the width of the tracks, and the image clearly shows the domain configuration. The total energy of the domain configuration is given by Eqn. (8.1). 156 CHAPTER 8 (8.1) ( ) 3902180 22 1 22 )sinsin(sinsin AA ttAKttAK bottombottomtoptopubottombottomtoptoputot γγ θθθθ ++ ++++= E The total energy Etot includes terms from both the top and bottom ferromagnetic layers, where Ku is the anisotropy energy in (J/m3), θ is the angle between the domain orientation and the anisotropy direction in the layer, A is the area of the domains perpendicular to the wire axis, A1 is the area of domains parallel to the wire axis, A2 is the area of the 180° domain walls, A3 is the area of the 90° domain walls. Note that A and A1 are areas in the x-y plane, while A2 and A3 area areas in the x-z plane. γ180 is the energy of the 180° domain walls (J/m2), and γ90 is the energy of the 90° domain walls (J/m2). Key =Area= =Area= 4a =Area= ( a− =Area= ( 2 22awa − 2 )tw .)ta . Repetitive unit l w a y z x x a/2 Fig. 8.5 Schematic diagram of the domain configuration showing how the total energy of the PSV was calculated by taking into account the energy of different parts of the domain structure including the domain aligned with the induced anisotropy axis, the closure domains, the 180° domain walls an the 90° domain walls. The energy of a 180° domain wall is given by Eqn. (8.2). (8.2)uex KA4180 =γ The energy of a 90° domain wall is given by Eqn. (8.3). 157 CHAPTER 8 (8.3)uex KA290 =γ Where Aex is the exchange coupling constant for NiFe, which is equal to 6.25×10-12 J/m Fig. 8.5 is a schematic diagram of the domain configuration showing how the total energy of the PSV structure was calculated by taking into account the different parts of the domain structure. The total energy of the domain configuration given in Eqn. (8.1) can be simplified because the energy contribution from the domains aligned in the anisotropy direction is zero. Therefore the total energy of the domain configuration is simplified to Eqn. (8.4). (8.4) 321 24)( AKAAKAtAAtKE uexuextopbottomutot +++= The areas A, A1, A2 and A3 can be calculated by multiplying the area of each part in the repetitive unit by the number of units (l/a), as shown below. A corresponds to the area of the domains (shown in green in Fig. 8.5) 22 2 lawl a lawaA −=    −= (8.5) A1 corresponds to the area of the closure domains (black in Fig. 8.5): 222 2 1 laaa a lA =   ××= (8.6) The area of the 180° domain walls (shown in blue): ( )    −=−×  = 12 a wlttaw a lA (8.7) The area of the 90° domain walls (shown in red): 2 4 22 4 22 3 ttaa a lA l=×      +  ×  = (8.8) Substituting these values back into Eqn. (8.4) gives the total energy according to the dimensions from the Bitter Technique image.   +      −+  +   −= 2 4214 22 ltKA a wltKAtlaKtlawlKE uexuextopubottomuTOT (8.9) 158 CHAPTER 8 To find the total anisotropy energy of the PSV, the total energy is differentiated with respect to a, and then minimised by setting equal to zero. This gives Ku as shown in Eqn. (8.10) 2 4 2 64     −= bottomtop ex u tt t a wAK (8.10) According to the Bitter technique image and the PSV construction, w=50µm, a=22.1µm, Aex=6.25×10-12J/m, t=21nm, ttop=15nm, tbottom=6nm giving Ku=22.8J/m3. A NiFe film was deposited in an applied field and the anisotropy was measured as ≈400J/m3, which is much larger than the calculated value for the PSV structure. The low anisotropy constant could be due to interlayer coupling between the two ferromagnetic layers. This coupling can be added as an extra term, J (J/m2), to the total energy of the structure ETOT given in Eqn. (8.9), as shown in Eqn (8.11), where ATOT is the total area over which the layers are coupled together and θ is the angle between the two NiFe layers. ( ) TOTuexuextopubottomuTOT AJltKAa wltKAtlaKtlawlKE θcos 2 4214 22 −  +      −+  +   −= (8.11) The surface coupling J is a minimum when the layers are parallel and maximum when they are antiparallel. We know from the magnetic hysteresis measurements in the previous chapters that the layers are parallel (cosθ=1), and according to Fig. 8.5, ATOT can be replaced by . The surface coupling can be calculated from Eqn. (8.11): Elw TOT with Ku=22.85J/m3 has to be the same energy as ETOT with Ku=400J/m3 plus the term for the interlayer coupling: ( ) JlwmJKEmJKE uTOTuTOT −=== )400(8.22 33 (8.12) When l=250µm, and w=50µm, J=3.19×10-6J/m2. If we multiply the PSV surface coupling constant J by the Cu interlayer thickness (Cu=2.2nm) to get the coupling energy per unit distance, we can compare this value with the exchange coupling constant Aex for NiFe. J×(2.2×10-9)=7.01×10-15J/m Aex=6.25×10-12 J/m and the coupling energy per unit distance between the ferromagnetic layers in the PSV≈7.01×10-15 J/m. From this calculation we can see that the exchange 159 CHAPTER 8 coupling between the atomic magnetic moments is ≈900× larger than the coupling between the ferromagnetic layers in the NiFe/Cu/NiFe PSV. 8.3 FIB Nanopatterning the Pseudo Spin Valves. RF sputter deposition was used to deposit ≈50nm of SiO2 on top of the micropatterned NiFe/Cu/NiFe spin valve structures. DC Sputter deposition was then used to deposit ≈20nm of Au to enable imaging in the FIB. Before the nanowires were patterned into the hard mask, the milling time was calibrated using end-point detection (EPD) as described in chapter 4. This technique measures the absorbed current from the specimen stage. The absorbed current changes when the ion beam interacts with different materials, and shows a striking difference between dielectric and metallic materials. For metals the absorbed current increases, whilst for dielectric materials the absorbed current decreases. This technique was used to measure the amount of time it takes to mill half way through the silica layer, which was the milling depth required for the hard mask. 120 140 160 180 200 220 240 260 0 50 100 150 200 250 3 Time (s) A bs or be d C ur re nt (p A ) Au SiO2 Ni80Fe20 Cu Ni80Fe20 00 Fig. 8.6 End-point detection calibration from the FIB for a current of 80pA and a voltage of 30keV. The dotted lines indicate milling through the different layers in the structure and the red line is the stopping distance for the silica hard mask. Fig. 8.6 shows the calibration EPD from the FIB. The dotted lines indicate milling through the different layers in the structure, the Au is milled away after the first 30s, and after 100s the milling has gone through the silica layer. The structure is fully milled through after 180s, and then the Ga ions are milling the Si substrate. The red line of the graph is the stopping time, which is the amount of time it takes to mill approximately half way through the silica layer ≈60s. This stopping time was chosen because the step height was enough carry out the in-situ 160 CHAPTER 8 milling through the spin valve before milling away the SiO2 hard mask. The structure and milling depth were verified using FIB imaging and AFM analysis as shown in Fig. 8.7. 30 .0 0 H ei gh t [ nm ] -3 0. 0 0 2.00 4.00 6.00 8.00 Length [µm] Wire Width (nm) Wire Gap (nm) Wire Height (nm) Aspect Ratio Sample 1 734 984 41 136 Sample 2 468 957 42 213 Sample 3 265 1000 41 377 Fig. 8.7 FIB image (above) showing the silica nanowire array on top of the NiFe/Cu/NiFe micropatterned mesostructure. AFM cross-section analysis (middle) was used to characterise the wire width, gap and height. A table is shown (bottom) with the dimensions of the nanopatterned wire array according to the AFM analysis. The table in Fig. 8.7 shows the width of the patterned wires, the gap between the wires and the height for three different samples taken from the AFM analysis. The values for wire width, gap and height are the average values taken from 30 measurements across the wire 161 CHAPTER 8 arrays. Three samples were prepared to study the effect of increasing the anisotropy and decreasing the wire width of the PSVs on the nanoscale. The length of the wire arrays was 100µm and the width of the wires varied from ≈260nm to ≈730nm, creating aspect ratios from 136 to 377. The gap between the wires in the arrays was deliberately made large to avoid magnetostatic coupling effects. Fig. 8.8 shows transport measurements taken directly before and after patterning the silica hard layer directly above the PSV with aspect ratio 136 in the FIB. Despite the use of the SiO2 barrier layer, the Ga ions implanted the spin valve structure and changed the natural similarity of the two ferromagnetic layers. The initial AMR of the spin valve changed to a small GMR ≈0.13% and the coercivity increased from 4.93 to 17.67Oe. Previous authors have investigated the effect of the Ga ion implantation on the magnetic properties of ferromagnetic materials [44,45]. It was found that ion doses as low as 3.0×1014 ions/cm2 at 30keV could change the magnetic properties of 15.5nm thick Ni80Fe20 ferromagnetic thin films. A dose of 1.0×1016 ions/cm2 was enough to render the material non-ferromagnetic at room temperature. The results also showed that the coercivity decreased with increasing dosage, which was due to compositional changes from ion implantation. The ion dosage for the sample shown in Fig. 8.8 below is 3.3×1016 ions/cm2, which is normally high enough to render the material non- ferromagnetic, but the PSV is protected by the silica barrier layer. -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 -80 -60 -40 -20 0 20 40 60 80 Field (Oe) Before Patterning After Patterning M R (% ) Fig. 8.8 Transport measurements of the NiFe/Cu/NiFe PSV directly before and after patterning the silica hard mask in the FIB. Initially the Ga+ implantation is not a problem because the silica layer is 50nm thick, and according to previous authors the implantation depth in silica at 30keV is 25± 8nm [49]. However, when the nanowire array is patterned in the silica, the layer is milled to half the depth. It is therefore likely that there is some ion implantation in the top (15nm) thick NiFe 162 CHAPTER 8 layer of the PSV. The shape change in the transport measurement does not agree with previous authors reports of a decrease in coercivity with increasing dosage density. However, it can be explained by the ions inducing a local anisotropy in the structure. In other words even before any milling is carried out through the spin valve, the Ga implantation through the SiO2 hard has locally changed the magnetic properties of the top ferromagnetic layer in the form of a highly anisotropic wire array. 8.4 In-situ Nanopatterning MR Measurements. The three PSV samples with nanopatterned silica hard masks were loaded into the in-situ magnetoresistance measurement rig. A current of 1.5mA was supplied to the sample, the gain of the voltage amplifier was set to ×50 and a lock-in amplifier was used to minimise the noise and maximise the sensitivity of the measurement. Fig. 8.9 shows the evolution in the transport measurements for the 265nm wide wire array with aspect ratio 377 as the structure is gradually milled through. The graphs show that as the sample is milled, the MR gradually increases reaching a peak value of approximately 1.47%. As the milling continues the MR decreases, the insets indicate the milling depth as calculated by the Resistance Model described in Chapter 6.8 and Chapter 7.4. The MR measurements reach a rounded maximum rather than plateaux, which shows that the two ferromagnetic layers are not completely decoupled. The value of the MR is comparable with results from previous authors, for example Ross et al. patterned 250µm long and 60nm wide wires from NiFe/Cu/Co/Cu film and achieved an MR of ≈0.8% [47]. Fig. 8.10 shows a comparison between the initial measurement and the maximum MR from the fully milled through structure. By overlaying the graphs it is possible to study the change in the coercivity of the soft (Hc1) and hard (Hc2) ferromagnetic layers. The coercivity of the soft layer remains almost the same, but the coercivity of the hard layer shows an obvious increase. As explained in the introduction, previous work has shown that the coercivity of ferromagnetic thin films tends to increase as the thickness is decreased [48]. However, in these measurements the initial reversal field Hc1 corresponds to the reversal of the thinner ferromagnetic layer and Hc2 to the reversal of the thicker layer. This will be explained by the micromagnetic simulations in section 7.7. These results show that the MR increases as the nanowire array is patterned through the thickness of the PSV, reaching a maximum when fully milled through. 163 CHAPTER 8 0 0.5 1 1.5 2 -60 -40 -20 0 20 40 60 Field (Oe) 0 0.5 1 1.5 2 -60 -40 -20 0 20 40 60 Field (Oe) M R ( % ) 0 0.5 1 1.5 2 -60 -40 -20 0 20 40 60 Field (Oe) M R ( % ) 0 0.5 1 1.5 2 -60 -40 -20 0 20 40 60 Field (Oe) 0 0.5 1 1.5 2 -60 -40 -20 0 20 40 60 Field (Oe) 0 0.5 1 1.5 2 -60 -40 -20 0 20 40 60 Field (Oe) M R ( % ) 0 0.5 1 1.5 2 -60 -40 -20 0 20 40 60 Field (Oe) 0 0.5 1 1.5 2 -60 -40 -20 0 20 40 60 Field (Oe) M R ( % ) (a) (f) (c) (b) (h) (g) (e) (d) Fig. 8.9 In-situ nanopatterning magnetoresistance measurements for a NiFe/Cu/NiFe PSV wire array with w≈265nm and aspect ratio 377. Graphs (a) to (h) show the evolution in the transport properties as the nanowire array structure is milled through the PSV. 164 CHAPTER 8 0 0.5 1 1.5 2 2.5 -60 -40 -20 0 20 40 60 Field (Oe) M R (% ) 0 0.1 0.2 0.3 0.4 M R (% ) Max MR Min MR Hc2 Hc1 Fig. 8.10 Graphs showing the initial and maximum MR measurements from the in-situ analysis of NiFe/Cu/NiFe PSV with wire width w=265nm and aspect ratio 377. The y-axis on the left is for the maximum MR measurement and the y-axis on the right is for the initial MR measurement before argon ion milling. Hc1 indicates the magnetisation reversal field for the soft (thinner) ferromagnetic layer, and Hc2 is the magnetisation reversal field for the hard (thicker) ferromagnetic layer. Fig. 8.11 shows the evolution in the transport measurements for the 730nm wide wire array with aspect ratio 137. The results show the same trend as for the PSV with 265nm wide wires. The MR gradually increases with increasing milling depth until it reaches a maximum of 1.1% when the nanowire array is fully milled through the PSV. The MR then decreases as the milling continues because the argon ions have milled through the silica mask and are milling the entire structure. The switching fields Hc1 and Hc2 also show the same trend. The final measurement on the 468nm wide wire array with aspect ratio 213 also showed the same trend, with a maximum MR of 2.35%. It is interesting to note that most of the increase in MR occurs when milling through the bottom (6nm) ferromagnetic layer, as shown in Fig. 8.12 for the 265nm wide wire array. Fig. 8.12 shows the change in MR, the resistance in zero field R and the change in resistance in an applied field dR as the milling depth increases through the spin valve. The resistance increases with milling depth as expected, but dR and MR increase more sharply as the milling continues past the copper interlayer (17.2nm) through the bottom ferromagnetic layer. 165 CHAPTER 8 0 0.5 1 1.5 2 -80 -60 -40 -20 0 20 40 60 80 Field (Oe) (d) 0 0.5 1 1.5 2 -80 -60 -40 -20 0 20 40 60 80 Field (Oe) M R ( % ) (a) 0 0.5 1 1.5 2 -80 -60 -40 -20 0 20 40 60 80 Field (Oe) (b) 0 0.5 1 1.5 2 -80 -60 -40 -20 0 20 40 60 80 Field (Oe) M R ( % ) (c) 0 0.5 1 1.5 2 -80 -60 -40 -20 0 20 40 60 80 Field (Oe) M R ( % ) (e) 0 0.5 1 1.5 2 -80 -60 -40 -20 0 20 40 60 80 Field (Oe) M R ( % ) (g) 0 0.5 1 1.5 2 -80 -60 -40 -20 0 20 40 60 80 Field (Oe) (h) 0 0.5 1 1.5 2 -80 -60 -40 -20 0 20 40 60 80 Field (Oe) (f) Fig. 8.11 In-situ nanopatterning magnetoresistance measurements for a NiFe/Cu/NiFe PSV wire array with wire width w≈730nm and aspect ratio 137. Graphs (a) to (h) show the evolution in the transport properties as the nanowire array structure is milled through the PSV. 166 CHAPTER 8 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 0 5 10 15 20 25 Depth milled (nm) M R a nd d R (Ω ) 0 20 40 60 80 100 120 R es ist an ce (Ω ) MR dR R Fig. 8.12 A graph showing how in MR, resistance in zero applied field R and the change in resistance in an applied field dR increase as the milling depth increases for a nanowire array with wire width ≈265nm. 0 0.5 1 1.5 2 2.5 -80 -60 -40 -20 0 20 40 60 80 Field (Oe) M R (% ) Narrow Wide Medium Wire width (nm) Aspect ratio Hc1 (Oe) Hc2 (Oe) Peak MR (%) 734 137 12 22 1.1 468 213 15 29 2.35 265 377 20 48 1.47 Fig. 8.13 Graphs of the fully milled through in-situ magnetoresistance measurements from three different nanopatterned PSV samples with narrow (265nm), medium (468nm) and wide (734nm) wires. The table below the graphs summarises the differences between the measurements. 167 Fig. 8.13 shows the fully milled through MR measurements for the three different patterned nanowire arrays; narrow (260nm), medium (460nm) and wide (730nm), and the table below CHAPTER 8 the graphs summaries the differences between the measurements. The switching fields of the soft and hard layers are Hc1 and Hc2 respectively, and are taken as the midpoint of the line. The results show that as the wire width is decreased (from 734nm to 468nm), the peak MR increased, but decreased for wire width 265nm. Also Hc1 for the medium and wide wires are very similar, whilst Hc1 for the narrow wires is larger. Generally Hc2 increases as the wire width decreases. The change in Hc1 and peak MR can be explained by looking at the Ga ion dosage density of the PSVs. For the 734nm wide array 62% of the total area was patterned, for the 468nm wide array 68% of the total area was patterned and for the 234nm array 87% of the total area was patterned. A much higher percentage of the total area for the 234nm wide array was implanted with Ga ions. Fig. 8.8 showed that the implanted Ga ions modified the transport measurement from AMR to GMR by inducing a high anisotropy structure by locally changing the magnetic properties of the top layer of the spin valve. The increased anisotropy of the top layer (hard) changed the natural similarity of the two ferromagnetic layers and the thinner (soft) layer was able to reverse more independently. The situation is similar for the 234nm wide wire array, except that the Ga ions have produced a higher anisotropy structure and therefore Hc1 and Hc2 are increased to higher values. Edge roughness can also increase the switching fields, and this becomes more significant as the wire width is decreased. 0 0.5 1 1.5 2 2.5 -80 -60 -40 -20 0 20 40 60 80 Field (Oe) M R (% ) f e d c b a Fig. 8.14 A graph showing the fully milled through MR measurement and minor loop analysis of the 468nm nanopatterned wide wire array. Fig. 8.14 shows the fully milled through MR measurement and minor loop analysis for the 468nm nanopatterned wire array. The different parts of the minor loop analysis are identified with letters from from a-f to guide the discussion. At a high positive field the MR is a minimum because the ferromagnetic layers are parallel with respect to one another. When the field is applied in the negative direction the resistance rises and the letter a indicates the 168 CHAPTER 8 reversal of the thinner (soft) ferromagnetic layer. The layers do not achieve complete antiparallel alignment because a plateaux is not seen in the MR. The change from a to b is due to the reversal of the thicker (hard) layer, at point b the thicker layer has not completely reversed. At a field of -25Oe, the direction of the applied field is changed and initially no change in MR is observed, this indicates that the magnetisation of the hard layer is irreversible. When the field goes past zero, at c the soft layer starts to rotate towards the hard layer and at point d it moves past the hard layer and completes its reversal at e. The change in MR from e to f is the hard layer aligning with the field direction. At point d the two layers move past each other, but the MR doesn’t reach the same minimum seen at saturation. This shows a spread in the magnetisation, perhaps due to closure domain like structures close to the edges. 8.5 Micromagnetic Simulations. Micromagnetic simulations were carried out to investigate the magnetisation reversal of PSV nanowires at different milling depths and wire widths, using the micromagnetic software package LLG Micromagnetics SimulatorTM. LLG Micromagnetics SimulatorTM is a full 3- dimensional simulation tool with integrated graphics that solves the Landau-Liftshitz-Gilbert- Langevin equations by relaxation and/or integration. With LLG, you can characterise micromagnetic structure and dynamics in films, on surfaces and in devices and materials. For further details of the theory behind the LLG Micromagnetics SimulatorTM, please refer to Appendix A. In order to compare the simulations with the experimental results a PSV construction was input into the simulation with similar parameters as the experiments (NiFe (15nm)/Cu (2.5nm)/NiFe (5nm)). The standard magnetic parameters for NiFe were input into the model, including Aex= 6.25×10-12 J/m and Ms=6.37×105 A/m. A NiFe film was deposited in an applied field and the induced anisotropy was measured as Ku= 400J/m3. The anisotropy axis of the top (15nm) NiFe layer was along the wire axis and the anisotropy axis of the bottom (6nm) layer was perpendicular to the top layer. The cell size was input as 10×10×2.5nm, and the magnetisation in each cell has contributions from the anisotropy, exchange, magnetostatic and applied fields. The final dimensions of the fully milled structure according to the simulation was 100nm wide by 2µm long, and Fig. 8.15 (i), (ii), (iii) and (iv) show the change in the magnetisation reversal of the two ferromagnetic layers as the milling depth is increased to depths of 2.5nm, 10.3nm, 18.0nm and fully milled through respectively. Within each milling depth simulation, the top four bars show the magnetisation direction in the top (15nm) thick NiFe layer, and the bottom 4 bars show the same for the bottom (6nm) thick NiFe layer. The colour of the layers gives a direct indication of the magnetisation direction according to the colour indicator at the middle of each depth simulation, and the 169 CHAPTER 8 arrows are a guide for the eye. Parts (a) to (d) indicate the reversal of the magnetic field from saturation in one direction to the other. Part (a) corresponds to magnetic saturation in the positive field direction (+H) and part (d) shows the magnetic saturation in the negative (-H) direction. Parts (b) and (c) show intermediate states. For the 2.5nm milling depth simulation (i), at saturation the middle of the wires are aligned parallel, but the ends show closure domains which are at 90° to the wire axis and are magnetostatically coupled antiparallel to each other. The closure domains form near the ends of the wires to minimise the magnetostatic energy, but in the bottom layer the closure domains extend further from the wire ends due to flux closure from the top (thicker) layer. The initial magnetisation reversal occurs by rotation from the closure domains, which enables the thinner layer to reverse for a smaller applied field. The anisotropy of the bottom NiFe layer is across the width of the wire, and this may also aid the magnetisation reversal. Fig. 8.15 (ii) shows the change in the magnetisation reversal process as the milling depth is increased to 10nm, which is 2/3rd of the way through the top layer. The increased shape anisotropy in the top layer restricts the magnetisation rotation, and reversal occurs more abruptly at a larger applied field. The edges rotate to 90º to the wire axis before the centre aligns with the field, and the edges of the layers appear to become coupled when the thicknesses become similar. It is interesting to note that the closure domains become coupled parallel at saturation (d). In Fig. 8.15 (iii) the simulation results are shown for a milling depth of 17.5nm, which corresponds to the complete patterning of the top ferromagnetic layer while the bottom layer remains unpatterned. The black regions at the edges of the top layer indicate the area of material removed. The shape anisotropy gives a greater contrast between the magnetisation reversal in the top and bottom layers and this is reflected in the distinctly different reversal fields in the two layers. The bottom layer still shows some magnetisation rotation, whilst the top layer reverses by switching state between (c) and (d) applied field directions. Fig. 8.15 (iv) shows the fully milled simulation, with both layers patterned with the same highly anisotropic structure. The bottom layer reverses before the top layer by switching abruptly between applied field directions (a) and (b). The top layer switches between applied field directions (c) and (d), and no magnetisation rotation is observed. 170 CHAPTER 8 (i) (d)(c)(b)(a)(d)(c)(b)(a) NiFe Cu NiFe Cu NiFe NiFe (ii) Bottom Top H- H+ Top Bottom Fig. 8.15 Micromagnetic simulations showing the magnetisation reversal in a PSV with the construction NiFe/Cu/NiFe at milling depths (i) 2.5nm, and (ii) 10nm. The top four bars in each simulation are the thicker (15nm) NiFe layer, and the bottom four are the thinner (6nm) NiFe layer. The direction of the applied magnetic field H is shown, and the colour code at the centre of each simulation indicates the direction of magnetisation within the layers. The arrows are to guide the eye, and the insets show the milling depth through the PSV structure. 171 CHAPTER 8 Top (d)(c)(b)(a)(d)(c)(b)(a) Cu NiFe NiFe Cu NiFe NiFe H- H+ Top Bottom Bottom (iv)(iii) Fig. 8.15 Micromagnetic simulations showing the magnetisation reversal in a PSV with the construction NiFe/Cu/NiFe at milling depths (iii) 17.5nm, and (iv) fully milled. The top four bars in each simulation are the thicker (15nm) NiFe layer, and the bottom four are the thinner (6nm) NiFe layer. The direction of the applied magnetic field H is shown, and the colour code at the centre of each simulation indicates the direction of magnetisation within the layers. The arrows are to guide the eye, and the insets show the milling depth through the PSV structure. 172 CHAPTER 8 15nm 6nm 6nm 15nm (ii)(i) -H +H (b) (c)(a) (d)(d)(c)(b) (a) Fig. 8.16 Micromagnetic simulations showing the magnetisation reversal in a PSV with the construction NiFe/Cu/NiFe with wire widths (i) 750nm, and (ii) 260nm. The top four bars in each simulation are the thicker (15nm) NiFe layer, and the bottom four are the thinner (6nm) NiFe layer. The direction of the applied magnetic field H is shown, and the colour code at the centre of each simulation indicates the direction of magnetisation within the layers. The arrows are to guide the eye. 173 CHAPTER 8 0.0 0.2 0.4 0.6 0.8 1.0 1.2 -800 -600 -400 -200 0 200 400 600 800 Field (Oe) M R (a rb itr ar y un its ) 2.5nm 10.3nm 18nm Milled 0.0 0.2 0.4 0.6 0.8 1.0 1.2 -800 -600 -400 -200 0 200 400 600 800 Field (Oe) M R (a rb itr ar y un its ) 750nm 470nm 260nm 100nm (b)(a) Fig. 8.17 The change in MR according to the simulation results as (a) a PSV is patterned into a 100nm wide wire structure. Milling depth results are shown from 2.5nm milling to the fully milled through structure. (b) As the PSV wire width is decreased from 750nm to 100nm. The arrows in (b) indicate the change in the peak MR as the wire width is reduced. The MR is in arbitrary units and gives a maximum of 1 when the layers are antiparallel and a minimum 0 when they are parallel. Both graphs display the field sweep from positive to negative field for clarity. The simulations show how reducing the wire width changes the magnetisation reversal characteristic. Fig. 8.16 is divided into two sections; section (i) shows the magnetisation reversal for a 750nm wide PSV wire and section (ii) shows the same for a 260nm wide PSV wire. The wider wire exhibits more magnetisation rotation before finally reversing, but in contrast the narrow wire switches abruptly, which is consistent with the milling depth simulations shown in Fig. 8.15. Once again the bottom layer reverses before the top layer because of the larger closure domains. Fig. 8.17 (a) shows the change in the MR of the 100nm wide wire as the structure is milled through as derived from the angles between the two layers in the micromagnetic simulations. The MR is given in arbitrary units, and gives a maximum value of 1 when the layers are antiparallel and a minimum value of 0 when they are parallel. The field sweep from positive to negative field is shown for clarity. The MR increases with milling depth reaching a maximum when fully milled, this is in agreement with the experimental data. The coercivity of the soft ferromagnetic layer Hc1, remains approximately the same with increasing milling depth, but the coercivity of the hard ferromagnetic layer Hc2 shows an large increase. This result is confirmed experimentally in Fig. 8.10, which compares the initial MR and fully 174 CHAPTER 8 175 milled through MR for the sample with wire width 265nm. Fig. 8.17 (b) shows the change in MR as the PSV wire width is decreased from 750nm to 100nm. As the wire width is decreased the simulations predict an increase in the MR (as shown by the arrows on the graph), due to an increase in the coercivity of the hard layer Hc2 and a decrease in the magnetisation rotation of the bottom layer. Instead of magnetisation rotation at the ends of the wires as the field is increased, the bottom layer shows more of a switching behaviour. This increases the degree of antiparallel alignment between the layers and increases the MR. In the experimental data an increase in Hc2 is observed as the wire width decreases, and there is also an initial increase in the MR as the wire width is decreased. However, the MR drops for wire width 265nm, this is thought to be due to the higher dosage of Ga ions. Length (nm) 0 45 90 135 180 0 100 200 300 400 500 A ng le (º ) Thinner Thicker 180 Length (nm) 0 45 90 135 0 100 200 300 400 500 A ng le (º ) Thinner Thicker Length (nm) 0 45 90 135 180 0 100 200 300 400 500 A ng le (º ) Thinner Thicker 0 45 90 135 180 0 100 200 300 400 500 Length (nm) A ng le (º ) Thinner Thicker 100nm 260nm 470nm 750nm Fig. 8.18 Graphs showing the spread of the angle of magnetisation at the end of the wires for 100, 260, 470, and 750nm wide PSV wires relative to the wire axis so that 90º is along the axis. Blue represents the bottom layer and red is the top layer. CHAPTER 8 The micromagnetic simulations seem to indicate that the magnetisation reversal in the ferromagnetic wires is initiated by the closure domains at the end of the wires. Although in the experimental measurements the relative area of the closure domains is smaller than in the simulations (because the wires are much longer), the change in the size of the domains due to the change in the width of the wire is a real effect. Fig. 8.18 shows the spread in the angle of magnetisation at the end of the PSV wires for 100, 260, 470 and 750nm widths. Blue represents the bottom layer and red the top layer. The magnetic moments at 90° are aligned along the length of the wire. The graphs show that the spread of the magnetic moments in the thinner layer is larger than in the thicker layer. This is due to the flux closure from the thicker layer. As the wire width increases the spread of the magnetisation produces a greater number of angles with respect to the wire axis. 8.6 Conclusion In-situ magnetoresistance measurements of nanopatterned PSVs have been carried out to investigate how the magnetic and electrical properties change with increasing shape anisotropy on the nano-scale. The measurements show that the MR increases as the milling depth between the patterned wires increases, mainly due to an increase in the coercivity of the hard ferromagnetic layer. The results also indicate that the MR increases as the wire width decreases, although the properties of the PSV are heavily dependent on the dosage of Ga ions during processing in the FIB. Minor loop analysis has shown that the magnetisation of the hard layer in the PSV nano-wires is irreversible. Micromagnetic simulations have confirmed the experimental results, and shown the increase in the coercivity of the hard layer to be due to a change in the magnetisation reversal from partial rotation at the centre of the wires to incoherent switching. The reversal of the soft layer also changes from partial rotation to switching as the dimensions are reduced. The simulations indicate that reversal initiates from closure domains at the end of the wires, which are larger in the bottom (thinner) ferromagnetic layer than in the top layer due to flux closure. The edges show a higher coercivity than the centre of the wires due to magnetostatic interactions. 176 CHAPTER 8 References [1] F.J. Himpsel, J.E. Ortega, G.J. Mankey, R.F. Willis, Advances in Physics, 47 (4), 511-597 (1998). [2] C.A. Ross, H.I. Smith, T. Savas, M. Schattenburg, M. Farhoud, M. Hwang, M. Walsh, M.C. Abraham, R.J. Ram, J. Vac. Sci. Technol. B, 17(6), 3168 (1999). [3] Y. Hao, F.J. Castaño, C.A. Ross, B. Vögeli, M.E. Walsh, H.I. Smith, J. Appl. Phys., 91(10), 7989 (2002). [4] C.A. Ross, M. Farhoud, M. Hwang, H.I. Smith, M. Redjdal, F.B. Humphrey, J. Appl. Phys. 91, 6848 (2002). [5] C.A. Ross, R. Chantrell, M. Hwang, M. Farhoud, T.A. Savas, Y. Hao, H.I. Smith, F.M. Ross, M. Redjdal, F.B. Humphrey, Phys. Rev. B, 62(18), 62 (2000). [6] C. Shearwood, S.J. Blundell, M.J. Baird, J.A.C. Bland, M. Gester, H. Ahmed, H.P. Hughes, J. Appl. Phys., 75(10), 5249 (1994). [7] R.P. Cowburn, D.K. Koltsov, A.O. Adeyeye, M.E. Welland, D.M. Tricker, Phys. Rev. Lett., 83(5), 1042 (1999). [8] R.P. Cowburn, J. Phys. D: Appl. Phys., 33, R1-R16 (2000). [9] R.P. Cowburn, D.K. Koltsov, A.O. Adeyeye, M.E. Welland, Appl. Phys. Lett., 73(26), 3947 (1998). [10] K.J. Kirk, J.N. Chapman, S. McVitie, P.R. Aitchison, C.D.W. Wilkinson, J. Appl. Phys. 87, 5105 (2000) [11] K.J. Kirk, J.N. Chapman, C.D.W. Wilkinson, J. Appl. Phys., 85(8), 5237 (1999). [12] K.J. Kirk, J.N. Chapman, C.D.W. Wilkinson, Appl. Phys. Lett., 71(4), 539 (1997). [13] S. Chou, P.R. Krauss, L. Kong, J. Appl. Phys., 79(8), 6101 (1996). [14] Y. Zheng, J.G. Zhu, J. Appl. Phys., 81(8), 5471 (1997). [15] J.F. Smyth, S. Schultz, D.R. Fredkin, D.P. Kern, S.A. Rishton, H. Schmid, M. Cali, T.R. Koehler, J. Appl. Phys., 69(8), 5262 (1991). [16] W. Wernsdorfer, B. Doudin, D. Mailly, K. Hasselbach, A. Benoit, J. Meier, J.-Ph. Ansermet, B. Barbara, Phys. Rev. Lett., 77(9), 1873 (1996). [17] J.-P. Jamet, S. Lemerle, P. Meyer, J. Ferré, B. Bartenlian, N. Bardou, C. Chappert, P. Veillet, F. Rousseaux, D. Decanini, H. Launois, Phys. Rev. B, 57(22), 14320 (1998). [18] A.O. Adeyeye, J.A.C. Bland, C. Daboo, D.G. Hasko, H. Ahmed, J. Appl. Phys., 82(1), 469 (1997). [19] A.O. Adeyeye, J.A.C. Bland, C. Daboo, J. Lee, U. Ebels, H. Ahmed, J. Appl. Phys., 79(8), 469 (1996). 177 CHAPTER 8 [20] C. Mathieu, C. Hartmann, M. Bauer, O. Buettner, S. Riedling, B. Roos, S.O. Demokritov, B. Hillebrands, B. Bartenlian, C. Chappert, D. Decanini, F. Rousseaux, E. Cambril, A. Müller, B. Hoffman, U. Hartmann, Appl. Phys. Lett., 70(21), 2912 (1997). [21] J.I. Martín, J. Nogués, I. K. Schuller, M.J. Van Bael, K. Ternst, C. Van Haesendonck, V.V. Moshchalkov, Y. Bruynseraede, Appl. Phys. Lett., 72(2), 255 (1998). [22] J.I. Martín, Y. Jaccard, A. Hoffman, J. Nogués, J.M. George, J.L. Vicent, I.K. Schuller, J. Appl. Phys., 84(1), 411 (1998). [23] W. Wernsdorfer, K. Hasselbach, A. Benoit, B. Barbara, D. Mailly, J. Tuaillon, J.P. Perez, V. Dupuis, J.P. Dupin, G. Guiraud, A. Perex, J. Appl. Phys., 78(12), 7192 (1995). [24] M. Lederman, S. Schultz, M. Ozaki, Phys. Rev. Lett., 73(14), 1986 (1994). [25] D. Grundler, G. Meier, K.-B. Broocks, Ch. Heyn, D. Heitmann, J. Appl. Phys., 85(8), 6175 (1999). [26] O. Fruchart, J.-P. Nozières, W. Wernsdorfer, D. Givord, F. Rousseaux, D. Decanini, Phys. Rev. Lett., 82 (6), 1305 (1999). [27] R.L. White, R.M.H. New, R.F.W. Pease, IEEE Trans. Magn., 33, 990 (1997). [28] S.Y. Chou, Proc. IEEE, 85, 652 (1997). [29] J.G. Zhu, X. Lin, L. Guan, W. Messner, IEEE Trans. Magn., 36, 23 (2000). [30] C.A. Ross, Ann. Rev. of Mater. Res., 31, 203 (2001). [31] S.Y. Yamamoto, R. O’Barr, S. Schultz, A. Scherer, IEEE Trans. Magn., 33, 3016 (1997). [32] T. Ono, H. Miyajima, K. Shigeto, T. Shinjo, Appl. Phys. Lett., 72(9), 1116 (1998). [33] T. Ono, H. Miyajima, K. Shigeto, K. Mibu, N. Hosoito, T. Shinjo, Science, 468, 5413 (1999). [34] R.P. Cowburn, M.E. Wellend, Science, 287, 1466 (2000). [35] E.C. Stoner, E.P. Wohlfarth, Philos. Trans. R. Soc. London, Ser. A, 240, 599 (1948); reprinted in IEEE Trans. Magn., 27, 3475 (1991). [36] I.S. Jacobs, C.P. Bean, Phys. Rev., 100, 1060 (1955). [37] E.H. Frei, S. Shtrikman, D. Treves, Phys. Rev., 106, 446 (1957). [38] J. Shi, S. Tehrani, Appl. Phys. Lett., 77(11), 1692 (2000). [39] J. Shi, S. Tehrani, T. Zhu, Y.F. Zheng, J.-G. Zhu, Appl. Phys. Lett., 74(17), 2525 (1999); T. Schrefl, J. Fidler, K.J. Kirk, J.N. Chapman, J. Appl. Phys., 85(8), 6169 (1999). [40] M. Kume, A. Maeda, T. Tanuma, K. Kuroki, J. Appl. Phys., 79(8), 6402 (1996). [41] F.J. Castaño, S. Haratani, Y. Hao, C.A. Ross, H.I. Smith, Appl. Phys. Lett., 81(15), 2809 (2002) [42] L. Kong, Q. Pan, B. Cui, M. Li, S.Y. Chou, J. Appl. Phys., 85(8), 5492 (1999). [43] J.A. Katine, A. Palanisami, R.A. Buhrman, Appl. Phys. Lett., 74(13), 1883 (1999) [44] G. Xiong, D.A. Allwood, M.D. Cooke, R.P. Cowburn, Appl. Phys. Lett., 79(21), 3461 (2001) 178 CHAPTER 8 179 [45] W.M. Kaminsky, G.A.C. Jones, N.K. Patel, W.E. Booij, M.G. Blamire, S.M. Gardiner, Y.B. Xu, J.A.C. Bland, Appl. Phys. Lett., 78(11), 1589 (2001) [46] F.J. Castaño, Y. Hao, C.A. Ross, B. Vögeli, H.I. Smith, S. Haratani, J. Appl. Phys., 91(10), 7317 (2002). [47] C.A. Ross, S. Haratani, F.J. Castaño, Y. Hao, M. Hwang, M. Shima, J.Y. Cheng, B. Vögeli, M. Farhoud, M. Walsh, H.I. Smith, J. Appl. Phys., 91(10), 6848 (2002). [48] M. H. Kryder, K. Y. Ahn, N. J. Mazzeo, S. Schwarzl, and S. M. Kane, IEEE Trans. Magn., Vol. Mag-16, NO. 1, 99 (1980). [49] S. Reyntjens, R. Puers, J. Micromech. Microeng. 11, 287 (2001). CHAPTER 9 I may not have gone where I intended to go, but I think I have ended up where I intended to be. Douglas Adams Chapter 9 Summary of dissertation This dissertation describes an experimental study of the patterning of magnetic thin films and spin valve devices. The initial work was carried out on Ni80Fe15Mo5 thin films and investigated the change in the magnetic properties as broad beam ion milling was used to etch narrow (3µm) and wide (4mm) regions. The results showed that the wider regions dominated the magnetisation reversal, and the coercivity increased by a factor of ten along the wire axis when the wider regions were removed. Measurements were also carried out for partial milling through the thickness of the wire arrays to assess the dependence of the magnetic properties on the degree of interconnection between the wires. The structures showed a progressive increase in coercivity, and a transition between single and two-stage reversal with increasing wire depth. Although the change in the magnetisation reversal with increasing shape anisotropy of ferromagnetic thin films has been well studied [1-4], this initial work was important because the aim was to apply the same patterning to the top layer of a NiFe/Cu/NiFe trilayer to achieve a spin valve response. The coercivity of the patterned ferromagnetic layer increased, which changed the natural similarity of the two layers, and a small spin valve response ≈0.45% was observed. The in-situ magnetoresistance measurement rig, which was specifically designed for these experiments, enabled direct analysis of the evolution in the electrical properties as the structure was gradually milled. Graphs of the change in the resistance in zero applied field R, and the change in resistance in the presence of an applied field dR, could be plotted versus milling depth, which gave a direct indication of whether the MR was changing due to a change in the resistance or the magnetic configuration. Contrary to what was expected the micropatterned PSV structures showed a maximum MR when fully milled. Magnetic h ial m ng b d ysteresis measurements showed a two-step reversal and confirmed that the init agnetisation reversal corresponded to the thinner ferromagnetic layer, which was surprisiecause previous work has shown that the coercivity tends to increase as the film thickness is ecreased [5,6]. The effect of further increasing the anisotropy was investigated by decreasing 180 CHAPTER 9 the wire width and the results showed that the GMR was enhanced by almost a factor of three as the wire width was decreased from 3µm to 1.5µm. These results showed that increasing the shape anisotropy increased the MR from the pseudo spin valve, and the optimum milling depth was when the structure was fully milled. The effect of changing the material properties was investigated by using the same patterning technique for CoFe/Cu/CoFe pseudo spin valve structures. A spin valve response was not observed due to the larger interlayer and magnetostatic coupling effects. According to these results optimum device performance can be achieved using materials with low magnetostatic and interlayer coupling energies. However, the interlayer coupling energy is directly related to the Curie temperature Tc, and a high Curie temperature is required for good thermal stability of the device. Therefore, a compromise between high thermal stability and low interlayer coupling energy will need to be made for this device design. Temperature measurements were carried out to compare the thermal stability of the pseudo spin valve structures with exchange biased spin valves. As the temperature was increased from 291K to 493K the MR of the exchange biased spin valve decreased to ≈1/4 of the original value. The MR of the PSV also decreased, but to only 2/3 of the original value. These results show that the patterned PSV has a better thermal stability than the IrMn exchange biased spin valve. Further experiments were carried out by reducing the dimensions to the nanoscale, which is more in keeping with the general trend towards miniaturisation of magnetic devices. Nanopatterning was carried out using a silica hard mask, which was deposited and then patterned on top of the PSV. The Focused Ion Beam (FIB) was used to nanopattern the silica hard mask, and then the pattern was transferred into the PSV by argon ion milling the entire structure. The results showed that the properties of the spin valve were strongly affected by gallium ion implantation during processing in the FIB, which created another form of anisotropy by locally changing the magnetic properties of the ferromagnetic material between the patterned wires, directly beneath the silica mask. The coercivity of the top ferromagnetic layer was increased and a small spin valve response was observed before milling in the argon ion miller. As the nanowire array was patterned through the PSV, the results showed the same trend as for the micropatterned samples; the MR increased with increasing milling depth showing a maximum when fully milled. An obvious increase in the coercivity of the hard layer was observed, whilst the coercivity of the soft ferromagnetic layer remained unchanged. The MR showed an initial increase from 1.1 to 2.35% as the wire width was decreased from 734 to 468nm, but decreased to 1.47% as the wire width was further decreased. This is thought to be due to the increased gallium dosage density, which changed the magnetic and 181 CHAPTER 9 electrical properties of the structure. Despite the change in the magnetisation reversal as the wire width was decreased from the micron to the nanoscale, an increase in MR was not observed. This is thought to be due to in increase in the initial resistance of the structures. Minor loop analysis showed that the reversal of the hard ferromagnetic layer was irreversible for the fully milled 468nm wide wire array. The Bitter technique was used to study the domain configuration in a 50µm wide PSV structure, and simple mathematical analysis showed that the coupling energy per unit distance between the layers J≈7.01×10-15J/m. The calculation showed that the exchange coupling between the atomic magnetic moments Aex, is ≈900× larger than the coupling between the ferromagnetic layers in the NiFe/Cu/NiFe PSV. Micromagnetic modelling was used to simulate the change in the magnetisation reversal in the PSV as the milling depth was increased through the thickness in 100nm wide wires. The results showed that for small milling depths, the reversal was dictated by rotation from closure domains at the ends of the wires, with the bottom (thinner) layer reversing before the top (thicker) layer. These results emphasise the importance of magnetostatic interactions and flux closure at the ends of the wires. The anisotropy axis of the bottom ferromagnetic layer was along the width of the wire, which also helped the formation of the closure domains. As the milling depth increased the reversal occurred by switching; the thinner layer reversed direction first by nucleation and propagation initiating from the closure domains. The modelling also demonstrated the change in the magnetisation reversal from rotation to switching as the wire width was decreased from 750 to 260nm. The GMR was extracted from the modelling data, and graphs were plotted to show the change in MR with milling depth and decreasing wire width according to the simulations. The simulations agreed with the experimental results showing a large increase in MR as the structure was milled through, and a small increase as the wire width was decreased on the nanoscale. Both the simulations and the experimental results showed a large increase in the reversal field of the hard layer as the milling depth is increased and wire width decreased, whilst the reversal field of the soft layer remained almost the same. References [1] A.O. Adeyeye, J.A.C. Bland, C. Daboo, D.G. Hasko, H. Ahmed, J. Appl. Phys., 82(1), 469 (1997). [2] A.O. Adeyeye, J.A.C. Bland, C. Daboo, J. Lee, U. Ebels, H. Ahmed, J. Appl. Phys., 79(8), 469 (1996). [3] T. Ono, H. Miyajima, K. Shigeto, T. Shinjo, Appl. Phys. Lett., 72(9), 1116 (1998). 182 CHAPTER 9 [4] T. Ono, H. Miyajima, K. Shigeto, K. Mibu, N. Hosoito, T. Shinjo, Science, 468, 5413 (1999). [5] J. H. Fluitman, Thin Solid Films, 16, 269 (1973) [6] M. H. Kryder, K. Y. Ahn, N. J. Mazzeo, S. Schwarzl, and S. M. Kane, IEEE Trans. Magn., Vol. Mag-16, NO. 1, 99 (1980). Other Work There has been considerable interest recently in the size dependence of exchange bias in antiferromagnetic/ferromagnetic bilayers (AFM/FM) [1-6]. This is partly due to the requirement for miniaturisation of spin valve read heads for the computer industry [7], but also due to some controversy over the fundamental understanding of how the magnetic configuration changes in the AFM/FM bilayer as the physical dimensions become comparable to characteristic length scales, such as the domain size in the FM and AFM materials. This work gives a novel insight into how the properties of spin valve devices change as the exchange bias is changed by milling through the AFM layer. The ferromagnetic layers remain unpatterned to investigate the effect of decreasing the AFM size on the pinning properties. Initially in-situ transport measurements were carried out on micropatterned spin valves to analyse the change in properties close to the AFM/FM interface. The work was then extended to patterning arrays of spin valve microwires and varying the spacing between the wires, to investigate whether it is possible to engineer localised spin valve regions. Finally the experiment was extended to the nanoscale to investigate the effect of miniaturisation. Mathematical modelling is used to help interpret the experimental results. In-situ transport measurements were carried out as the AFM layer was milled away in an exchange biased spin valve structure of the form Ta(9nm)/NiFe(6.5nm)/Cu(2.6nm)/NiFe(3nm)/IrMn(10nm)/Ta(3nm), where the AFM layer was at the top of the structure to allow direct milling. No change in MR was observed as the AFM layer was milled through the bulk. As the thickness of the AFM layer was reduced from 10nm to below 3nm a decrease in MR was observed and an applied field of 80Oe was large enough to overcome the exchange field from the AFM layer and align the ferromagnetic layers. When the AFM layer was completely milled away a small MR remained due to a combination of the AMR (the current and field were perpendicular) and a small GMR from the thickness difference of the ferromagnetic layers. The change in the MR and dR are shown in Fig. 10.1 for different thicknesses of IrMn. This result agrees with previous work [8], which indicates that the exchange anisotropy from the AFM layer is an interface effect. 183 CHAPTER 9 -0.1 0.2 0.5 0.8 -80 -40 0 40 80 Field (Oe) dR ( 550s (2.8nm) 580s (1.2nm) 595s (0.7nm) -0.5 0 0.5 1 1.5 2 2.5 -80 -40 0 40 80 Field (Oe) M R (% ) 550s (2.8nm) 580s (1.2nm) 595s (0.7nm) Ω)(Ω ) Fig. 10.1 (a) In-situ MR at milling times 550s, 580s and 595s. The number in brackets in the inset is an estimation of the thickness of IrMn remaining. (b) dR versus field for milling times 550s, 580s and 595s. Optical lithography and argon ion milling were used to micropattern wire arrays in the AFM layer of exchange biased spin valves, as shown in Fig, 10.2. Ar+ Ar+ Ar+ x Ta (3nm) Py (6.5nm) Cu (2.6nm) Py (3nm) IrMn (10nm) Ta (9nm) Fig. 10.2 Schematic cross section of patterned sample showing wire array structure and lay- up of the spin valve. The width ‘x’ of the rectangular sample varies depending upon the spacing between the wires. (b) The direction of the current is perpendicular to the applied field and the wire axis. The wire width is 3µm, and the spacing varied from 8µm to 1.5µm. The length of the sample is 100µm. The width of the wires was kept constant at 3µm, and the experiments investigated the change in MR with milling as the spacing was varied between 8µm and 1.5µm. The results show a decrease in MR as the AFM regions between the wires are milled away. This is due to a 184 CHAPTER 9 decrease in the antiparallel alignment of the ferromagnetic layers. The AFM wires produce a localised spin valve effect in the FM/NM trilayer directly beneath. The percentage decrease in MR corresponds to the area of AFM material milled away, showing a larger decrease when the spacing between the AFM wires was increased. This shows that the ferromagnetic regions between the wires are free to move in the applied field. Measurements were also taken with the wire width and spacing on the nanoscale. The results still show a localised spin valve effect, even over wire spacing as small as ≈200nm. Mathematical modelling has been used to study the magnetisation dynamics of the ferromagnetic region between the pinned nanowires as the thickness of the AFM layer directly above is reduced. The results show that as the exchange field Hex is reduced, the maximum angle of the magnetic moments is increased with respect to the pinning direction, as shown in Fig. 10.3 (a). As the field is increased the magnetic moments tend to align with the applied field, and the moments furthest away from the pinned regions reverse for smaller applied fields. As the spacing between the IrMn wires is reduced to 200nm the model shows that the moments don’t completely align with the field direction even for low values of Hex, this is because Aex is strong and the spacing is small. 0 0.5 1 1.5 2 2.5 3 3.5 0 1000 2000 3000 4000 5000 Hex (A/m) α ( ra ds ) 200nm 300nm 400nm 50 75 100 0 1000 2000 3000 4000 5000 Hex (A/m) G M R (% ) 200nm 300nm 400nm (b)(a) Fig. 10.3 (a) Maximum angle α from the exchange bias direction for the magnetic moments in a region between two pinned wires called s, where s=200nm, 300nm and 400nm, as Hex is reduced by milling away the IrMn. α=0rads indicates that the magnetic moments are aligned with the exchange bias direction. As α increases towards π rads, the moments become more aligned with H. (b) The change in GMR as the exchange field Hex is reduced according to the energy minimisation model, where s=200nm, 300nm and 400nm. A constant applied field of 5000Am-1 was used. 185 CHAPTER 9 186 The relative angles of the magnetic moments in the ferromagnetic layers were converted into GMR as shown in Fig. 10.3 (b), by assuming that the zero field resistance of the spin valve remained constant. The model shows that as the spacing between the wires becomes smaller, a larger milling depth is required before a decrease in GMR is observed. When the AFM layer is milled to the FM interface, the decrease in GMR is proportional to the area of the AFM material removed. References [1] J. Yu, A. D. Kent, and S. S. P. Parkin, J. Appl. Phys. 87, 5049 (2000) [2] J. G. Zhu, Y. F. Zheng, and X. D. Lin, J. Appl. Phys. 81, 4336 (1997) [3] Y. Shen, Y. Wu, H. Xie, K. Li, J. Qui and Z. Guo, J. Appl. Phys. 91, 8001 (2002) [4] K. Lui, S. M. Baker, M. Tuominem, T. P. Russell, I. K. Schuller, Phys. Rev. B 63, 63 (2001) [5] M. Fraune, U. Rüdiger, G. Güntherodt, S. Cardoso, P. Freitas, Appl. Phys. Lett. 77, 3815 (2000) [6] S. Zhang, Z. Li, Phys. Rev. B 65, 65 (2001) [7] W. J. Gallagher, S. S. S. Parkin, W. Lu, X. P. Bian, A. Marley, K. P. Roche, R. A. Altmann, S. A. Rishton, C. Jahnes, T. M. Shaw, and G. Xiao, J. Appl. Phys. 81, 3741 (1997) [8] J. Nogués, I.K. Schuller, J. Magn. Magn. Mater. 192, 203 (1999). APPENDIX A This appendix gives an overview of the theory behind the LLG Micromagnetic Simulator. The information was taken directly from the LLG Micormagnetics Simulator User Manual. LLG Micromagnetics SimulatorTM was founded in 1997 as a sole proprietorship by Michael R. Scheinfein, who was then a professor of physics at Arizona State University in Tempe, Arizona. In 1997, the first users of LLG were a handful of scientists working on MRAM in corporate research labs. Now, there are over 100 corporate and academic users worldwide, including Europe, Asia and North America. 187 LLG Micromagnetics Simulator User Manual 3-25 CHAPTER 3 INTRODUCTION TO USING LLG This section provides you with an overview of LLG’s design, of how to use LLG and of LLG’s functionality. For complete details and instructions, refer to the appropriate sections found later in the Manual. THREE MODULES OF FUNCTIONALITY LLG Micromagnetics Simulator has three functional modules. These modules are specified in terms of the serial pro- cess of defining the solutions to most problems, while maintaining consistency with the Windows event-driven pro- gramming interface. These modules are listed below with their corresponding menus. • Input phase: data specification (LLG Input Sheet) • Simulation phase: solution of the differential equations (LLG Simulation Sheet) • Review phase: playback of results through graphical animation (movies) (LLG Movie Viewer) INPUT PHASE: DATA SPECIFICATION The LLG Input Sheet is the central interface for coordinating input parameters, error checking and setting critical glo- bal parameters. In general, inputting data specifications is the most tedious aspect of numerical simulations. The pro- gram has been designed to allow you flexibility in customizing simulations; however, this makes the data specification phase time-consuming and increases the risk of input error. Although, as a counter measure, LLG performs exhaus- tive error checking to prevent floating-point exceptions, defining a structure and how well it models an actual material or device is ultimately the user’s responsibility. Since the program solves the Landau-Lifshitz-Gilbert equations using finite differences for exchange energies and fields, as well as boundary elements for magnetostatic self-energies and fields, the structure of interest must be defined as a grid. The program uses rectangular pixels on a Cartesian grid. Although you can change the material parameters, including eliminating magnetic material altogether, this must be done on a Cartesian grid. Once you have specified the structure and clicked the Begin Simulation button, LLG initializes all of the arrays, com- putes the demagnetization field coupling tensors and calculates the fields for any boundary conditions. Once these large arrays have been specified, the simulation phase can begin. Also, you are prompted to store the simulation parameters in several files. SIMULATION PHASE: SOLUTION OF THE DIFFERENTIAL EQUATIONS Once LLG has verified that you have input the data correctly, you have your first chance to set the graphical represen- tation of the data. Many features can be viewed interactively. Chapter 3: Introduction to Using LLG 3-26 LLG Micromagnetics Simulator User Manual REVIEW PHASE: PLAYBACK OF RESULTS THROUGH A GRAPHICALLY ANIMATED MOVIE Once a simulation is complete, you can review the results by replaying them through a graphically animated movie or you can view a domain or field file in the viewer control. Graphical representation of the data is essential to compre- hending the results of a simulation. The program provides complete two- and three-dimensional views in the form of bit- map images, contour maps and vector fields. THEORY OF OPERATION Micromagnetic structure, such as that present in surface domain walls, can be extracted with standard methods for the solution to the Landau-Lifshitz-Gilbert equation. Such methods have been given in the literature by Brown [1], LaBonte [1,2], Aharoni [3-9], Hubert [10,11], and Schabes [9,12]. The equilibrium magnetization configuration results from the minimization of the system’s free energy. The energy of a ferromagnetic system is composed of 1) the mean field exchange energy Eex between nearest neighbors characterized by the exchange coupling constant A (erg/cm); 2) the magnetocrystalline anisotropy energy EK, which describes the interaction of the magnetic moments with the crystal field characterized by the constant Kv (erg/cm 3); 3) the surface magnetocrystalline anisotropy energy Eks, which cor- rects for broken symmetry near surfaces in the interaction of the magnetic moments with the crystal field, and is char- acterized by the constant Ks (erg/cm 2); 4) the magnetostatic self-energy Es, which arises from the interaction of the magnetic moments with the magnetic fields created by discontinuous magnetization distributions both in the bulk and at the surface; 5) the external magnetostatic field energy Eh, which arises from the interaction of the magnetic moments with any externally applied magnetic fields; and 6) the magnetostrictive energy Er, which arises when mechanical stress (strains) are applied to a ferromagnetic material thereby introducing effective anisotropy into the system charac- terized by Km (erg/cm 3). The solution for the equilibrium magnetization distribution is a constrained boundary value problem in two or three spa- tial dimensions with the constraint of constant magnetization Ms. The continuous magnetization distribution of a ferro- magnet is approximated by a discrete magnetization distribution consisting of equal volume cubes (3-D) or rods (2-D). Each individual discretized magnetization cell, interior to the array, will be addressed by the (X, Y, Z) coordinates of its centroid. There are Nx cells along X, Ny cells along Y, and Nz cells along Z interior to the structure to be modeled. There is one plane (3-D) or column (2-D) of boundary cells bounding the discretized region. These boundary cells (con- ditions) can reflect the continuous uniform magnetization distribution present within the domains themselves on either side of the structure. If no boundary conditions are specified, the cells at the edges are free. In the absence of surface anisotropy, the normal derivative of the magnetization distribution at the surface is zero [2,13]. In the presence of sur- face anisotropy, the Rado-Weertman boundary conditions is used [13,14]. Fundamental to the solution of the micromagnetic equations is the assumption that the bulk saturation magnetization Ms (emu/cm 3) is constant microscopically throughout the ferromagnet. The parameter Ms represents saturation magne- tization at room temperature. For most practical systems being considered (Fe, Co or Permalloy), there is little devia- tion in Ms at room temperature from the 0 K value. The value of the magnetization vector M(r) at each point within the ferromagnet is the saturation magnetization multiplied by the direction cosines, that is M(r) = (Mx(r), My(r), Mz(r)) = Msα(r) = Ms (α(r), β(r), γ(r)). The constraint equation implied by the constant magnetization assumption is |α(r)| = 1. The individual contributions to the energies in this continuum model are calculated by integrating the energy expres- sions over the structure in question. The energy integrals below are integrated over the appropriate dimension, dV. The exchange energy Eex in the continuum approximation is given by. The exchange parameter A can be extracted from spin-wave theory [15-17], which shows that A = A'Ms 2 = DS/2V, where D is the spin-wave dispersion parameter, S is the spin per atom and V is the volume per atom. The spin-wave dispersion parameter, D, is related to the exchange constant, J, in the Heisenberg hamiltonian by D = 2JSa2, where 'a' Eex dV∫= ∇α 2 ∇β 2 ∇γ 2+ +[ ] LLG Micromagnetics Simulator User Manual 3-27 Chapter 3: Introduction to Using LLG is the lattice spacing. This relationship is true for spin-wave modes along bcc [100], bcc [110], fcc [110] and fcc [100] directions. The volume magnetocrystalline anisotropy for uniaxial (e.g. easy-axis in y) EKu, and cubic crystals EKc, is given by the following expressions, respectively, where the bulk anisotropy constants for cubic, Kc, and uniaxial, Ku, symmetry can be determined from torque magne- tometry measurements. The energy due to magnetostriction can be included in the expression for the uniaxial anisot- ropy by appropriately adjusting the value of the anisotropy constant [20]. The surface magnetocrystalline anisotropy energy EKs is given by, where the integration is along the (line increment dS in 2-D or a surface increment in 3-D) boundary at the film sur- faces. The symmetry of the surface anisotropy energy was determined by Rado [21,22].The self-magnetostatic field energy Es can be represented in a number of equivalent forms, but for these purposes the most convenient represen- tation is where the self-field Hs is determined from the negative gradient of the scalar magnetic potential, The magnetic scaler potential φ satisfies inside the ferromagnet, and LaPlace’s equation outside of the ferromagnet, and at the surface ϕ = ϕ’ and in the two-dimensional case for example. The regularity of φ’ at infinity is also required. This can be guaranteed by solv- ing for the potential using Green’s function methods. The calculation of this self- field energy is the most computation- ally intensive aspect of solving the micromagnetic equations. The external field energy Eh for an applied field of Ho is simply given as [ ]∫ −+−= 22221 )1()1( ββ uuKu KKdVE ( )[ ]∫ +++= 22222222221 γβαγαγββα ccKc KKdVE E Ks S 1 2 -- αˆ nˆ•( )2d∫= Es dV 1 2 --H∫ s αˆMs•–= H ∇φ–= ∇2 φ 4πMs ∇ αˆ•= ∇2 φ′ 0= dφ dz -----– 4πMs γ+ dφdz-----–= Eh VHo αˆMs•d∫–= Chapter 3: Introduction to Using LLG 3-28 LLG Micromagnetics Simulator User Manual To calculate the magnetic microstructure in ferromagnets, the time evolution of a magnetization configuration inside a ferromagnet, which is described by the Landau-Lifshitz-Gilbert equation, must be solved. The Landau-Lifshitz-Gilbert equation has been examined experimentally and theoretically [28,29,32,81,82], and found to yield an accurate descrip- tion of the time evolution of a magnetic moment of fixed magnitude in a magnetic field. This equation has the following form. Here, the gyromagnetic frequency γ = gωe /2 is determined from the free electron value of ωe and the spectroscopic splitting factor, g = 2. The gyromagnetic frequency γ, the damping parameter α and the magnitude of the effective fields determine the time scales of interest. For time domain simulations, the free electron gyromagnetic frequency of γ = 1.78 x 107 (Oe sec-1) is used. The damping parameter α is not well known. Values of α between 0.005 and 2.0 have been used to solve LLG. The damping parameter was not found to change the equilibrium magnetization configurations in domain walls in uniform ferromagnetic systems [23]. The effective magnetic field on each magnetic moment is deter- mined from the total system energy Etot as The effective magnetic field incorporates all the effects of exchange, anisotropy, external fields and demagnetizing fields. For the analysis of the equilibrium micromagnetic structure, the differential equation need not be integrated directly. Instead, notice that, for an equilibrium magnetization distribution, dM/dt = 0, which implies that the effective field, Heff, must be parallel to the magnetization M. The magnetization configuration can be relaxed iteratively by posi- tioning each magnetization vector (almost) along the effective field vector direction throughout the mesh. The initial condition can be selected to provide a head start for the iteration procedure. When the largest residual of a single value of (MxHeff)/|M||Heff| decreases below a convergence minimum, the iteration process is stopped. The convergence min- imum for terminating the calculation is the value of the largest relative change in the largest component of the direction cosines. This value will depend upon the size of the mesh and on the closeness to a magnetization change, such as the reorientation close to the coercive field in a hysteresis loop calculation. Equilibrium domain wall configurations determined from this energy minimization scheme agree extremely well with configurations determined by solving the Landau-Lifshitz-Gilbert equation directly [23]. For equilibrium configurations for uniform systems, the more economical energy minimization scheme can be used to determine equilibrium configurations. For more complex systems, or in the presence on grain boundaries, which may serve as nucleation sites, the solution of the Landau-Lifshitz-Gilbert equation is necessary for accurate results [18,19]. The content of this section has been extracted and slightly altered from the original version published in a section of Micromagnetics of 180 Degree Domain Walls at Surfaces, M.R. Scheinfein, J. Unguris, J.L. Blue, K.J. Coakley, D.T. Pierce, R.J. Celotta, P.J. Ryan, Phys. Rev. B43(4), 3395 (1991). 1. W.F. Brown, A.E. LaBonte, J. Appl. Phys. 36(4), 1380 (1965). 2. A.E. LaBonte, J. Appl. Phys. 40(6), 2450 (1969). 3. A. Aharoni, J. Appl. Phys. 37(8), 3271 (1966). 4. A. Aharoni, J. Appl. Phys. 38(8), 3196 (1967). 5. A. Aharoni, Phil. Mag. 26, 1473 (1972). 6. A. Aharoni, phys. stat. sol. (a) 18, 661 (1973). 7. A. Aharoni, J. Appl. Phys. 46(2), 908 (1975). 8. A. Aharoni, J. Appl. Phys. 46(2), 914 (1975). 9. M.E. Schabes, A. Aharoni, IEEE Trans. Mag. MAG-23(6), 3882 (1987). 10. A. Hubert, phys. stat. sol. 32, 519 (1969). dM dt⁄ γ 1 α2+---------------– M Heff× γα 1 α2+( )Ms --------------------------- M M Heff×( )×–= Heff ∂Etot– ∂ Ms αˆ( )⁄= LLG Micromagnetics Simulator User Manual 3-29 Chapter 3: Introduction to Using LLG 11. A. Hubert, phys. stat. sol. 38, 699 (1970). 12. M.E. Schabes, H.N. Bertram, J. Appl. Phys. 64(3), 1347 (1988). 13. G.T. Rado, J.R. Weertman, J. Phys. Chem. Solids, 11, 315 (1959). 14. G.T. Rado, Phys. Rev. B40(1), 407 (1989). 15. R. Victora, J. Appl. Phys. 62(10), 4220 (1987). 16. C. Herring, C. Kittel, Phys. Rev. 81(5), 869 (1951). 17. G. Shirane, V.J. Minkiewicz, R. Nathans, J. Appl. Phys. 39(2), 383 (1968). 18. C.C. Shir, J. Appl. Phys. 49(6), 3413 (1978). 19. R. Victora, Phys. Rev. Lett. 58(17), 1788 (1987). 20. B.D. Cullity, Introduction To Magnetic Materials (Addison Wesley Publishing Co, Reading, 1972). 21. G.T. Rado, Phys. Rev. B26, 295 (1982). 22. G.A. Prinz, G.T. Rado, J.J. Krebs, J. Appl. Phys. 53(3), 2087 (1982). 23. M.R. Scheinfein and J.L. Blue, J. Appl. Phys. 69(11), 7740 (1991).