THE FABRICATION AND MEASUREMENT OF CURRENT PERPENDICULAR TO THE PLANE MAGNETIC NANOSTRUCTURES FOR THE STUDY OF THE SPIN TRANSFER EFFECT

Transcription

1 THE FABRICATION AND MEASUREMENT OF CURRENT PERPENDICULAR TO THE PLANE MAGNETIC NANOSTRUCTURES FOR THE STUDY OF THE SPIN TRANSFER EFFECT A Dissertation Presented to the Faculty of the Graduate School of Cornell University in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy by Frank Joseph Albert January 2003

2 c Frank Joseph Albert 2003 ALL RIGHTS RESERVED

3 THE FABRICATION AND MEASUREMENT OF CURRENT PERPENDICULAR TO THE PLANE MAGNETIC NANOSTRUCTURES FOR THE STUDY OF THE SPIN TRANSFER EFFECT Frank Joseph Albert, Ph.D. Cornell University 2003 This dissertation describes the development and use of a set of techniques used to fabricate magnetic nanostructures where a current can be applied perpendicular to the plane of the thin film nano-magnets. These current perpendicular to the plane (CPP) devices consisted of alternating layers of magnetic and non magnetic thin films, with majority of the work presented in this dissertation involving magnetic trilayers. These ferromagnet (F)/ normal metal (N)/ Ferromagnetic (F) trilayers were then patterned so that the diameter of this F/N/F stack was as small as 50 nm. I will present the results of an investigation of the field and current dependence of the magnetization of these magnetic nanostructures. Passing a high current density through these devices leads to two new phenomena, the self-field effect and spin-transfer effects. In the self field effect the magnetic field associated with the high current density causes a reorientation of the domains within the nanomagnet. This dissertation describes the effect of the self field in these CPP devices as a function of current, field, and size.

4 In the spin-transfer effect, spin angular momentum is transferred from a spin polarized current to a magnetic moment. This transfer of spin angular momentum results in a torque on the magnetic moment which can, for large enough current, affect the magnetization of the local magnetic moment. For low applied field this spin-transfer effect can cause full magnetization reversal. For large magnetic field this spin-transfer effect can cause high current excitations which are consistent with spin wave excitations. I will present evidence demonstrating this spin transfer effect in Co/Cu/Co and other material systems. Furthermore I will present evidence which probes the origin of the spin transfer effect.

5 Biographical Sketch Frank Joseph Albert was born in Garden Grove, California where he lived until attending college at the University of California at Irvine. He earned earned a bachelors degree in physics in Later that same year he joined Cornell university. In 2001 he earned a masters of science degree in physics and in 2002 he completed the requirements for a doctoral degree. iii

6 To, my mother and to Amanda iv

7 Acknowledgements I wish to extend my sincere gratitude to my advisor, Robert Buhrman. I wish to thank him for the faith he had in me and for the encouragement that he has given me through the years. I appreciated his way of asking questions about experimental results that inspired me to think more deeply about physics. His insight into what makes a complete physical picture will certainly be invaluable during my career. Jordan Katine, a former Buhrman group post-doc whom I had the privilege of working with, was especially helpful early in my graduate school career. Jordan provided me with the fabrication technique that I exploited in the early part of my research. But more importantly, he provided me with a clear example of what is needed to be successful in experimental physics. His belief that success in experimental physics can only come from hard work is a lesson that will remain with me throughout my career. I am sure that if I did not work with Jordan early in my graduate school career I would not be where I am today. I wish to thank Dan Ralph who has been instrumental in much of our theoretical interpretation of my experimental results. I thank him for all the thought that he invested in my experiments. I also wish to thank him for his extra effort in reading my dissertation the weekend before my B exam so that I would be able to immediately work on his corrections. v

8 I wish to acknowledge the contribution Nathan Emley to this thesis project. Nathan invested much time in learning the experimental techniques used in this dissertation and became a tremendous help. I would not have been able to accomplish the volume of work that is presented in this thesis without his efforts. I wish to thank fellow students Andrew Perrella, Preeti Chalsani, Bill Rippard, and Phil Mather for making the lab bearable and being there to answer the occasional physics and equipment questions. I especially thank Preeti, who maintained the CHA evaporator, which was a key tool that I used during my research at Cornell. I wish to acknowledge the efforts of the staff of the Cornell Nanofabrication Facility. Without the use of the machines that they maintained or their insight into nanofabrication techniques my thesis project would not have been a success. I also wish to acknowledge the financial support provided to me by the Cornell Center for Materials Research. Finally, I wish to thank Amanda Varner for her limitless support during these last five years. vi

9 Table of Contents 1 Introduction 1 2 Electron transport through magnetic nanostructures: fundamental properties and applications Introduction Electron transport through magnetic nanostructures Giant magnetoresistance Spin valves Tunneling magnetoresistance Magnetic semiconductors Spin transfer Possible impact of spin transfer in spintronics applications Field sensors Magnetic random access memory Spin transfer based high frequency oscillators Spin transfer conclusion References The fabrication of magnetic nanostructures CPP fabrication introduction CPP device fabrication at Cornell General overview Fabrication process Device tracking and final device testing Limitations Alternative CPP device fabrication alternatives explored at Cornell Polyimide with Cr mask Polyimide with Au mask Carbon/silicon mask References vii

10 4 Early spin-transfer experiments Idea of spin transfer Incorporating the idea of spin transfer into real devices (Slonczewski model) Experimental techniques and early experimental results for spin-transfer studies Competing effects (self field) Introduction to the self-field effect Devices Current induced vortex formation Spin transfer in circular nanopillars Introduction Devices Field driven switching in circular nanopillars Low field current driven switching in circular nanopillars High field current driven excitations Summary of data from nanopillars with circular cross section Spin transfer in elongated hexagons Elongated hexagon introduction Elongated hexagon devices Critical aspects of patterning elongated hexagon device structures Field driven switching in elongated hexagon nanopillars Single nano-magnet magnetometry Spin transfer driven magnetization reversal in elongated hexagon nanopillars The reliability of current induced switching Conclusion References Quantitative studies of the spin-transfer effect Introduction Free layer thickness dependence Theoretical models of spin transfer Devices Magnetoresistance Multi-domain current switching J c + Jc vs. t free Thermal fluctuation effects Discussion Magnetic layer dependence conclusion Probing electron paths with varying spacer layer thickness Introduction Experimental setup Data and analysis viii

11 5.4 Conclusion References Exploratory work in spin transfer Introduction Different materials systems Alternate materials introduction Low coercivity materials High coercivity alloys Synthetic anti-ferromagnetic systems SAF introduction SAF devices SAF field driven switching SAF current driven switching SAF current driven switching analysis Conclusion References Conclusion 226 ix

12 List of Tables 3.1 EBL exposure doses Ion mill rates IMP process progression Spin transfer experimental techniques Spin transfer critical current densities x

13 List of Figures 2.1 GMR diagram Oscillatory GMR Oscillatory exchange coupling Two channel model Experimental geometries Spin valve schematic Dipolar coupling in spin valves MTJ schematic Spin-transfer diagram Magnetic storage diagram MRAM schematic Types of magnetic nanostructures SiO 2 with C mask process: step SiO 2 with C mask process: steps 2 and SiO 2 with C mask process: step 4 and SiO 2 with C mask process: step 6 and SiO 2 with C mask process: step SiO 2 with C mask process: step Scanning electron microscope images of nanostructures SiO 2 with C mask process: steps 10 and Ion mill planarization introduction Ion mill planarization demonstration Ion mill planarization characterization SiO 2 with C mask process: step SiO 2 with C mask process: step SiO 2 with C mask process: step SiO 2 with C mask process: step SiO 2 with C mask process: step SiO 2 with C mask process: step Problems with polyimide C/Si liftoff mask technique: steps C/Si liftoff mask technique: steps xi

14 4.1 Spin transfer in domain walls N/F/N/F/N spin transfer system Self-field device schematic R(H) for self-field devices R(I) for self-field devices Effect of pulsed current through self-field devices Effect of H on the self-field effect Circular spin transfer nanopillar schematic R(H) for circular nanopillar R(I) for circular nanopillar Spin transfer at varying applied field High field high current excitations High field high current excitations, 3-D plot and slope R(H) at high bias current Elongated hexagon device schematic Effects of backscattering in EBL CAD pattern used to expose elongated hexagon SEM image of elongated hexagon devices R(H) for elongated hexagon nanopillar Angular dependence of H c in elongated hexagon nanopillars R(H) for field applied R(I) for an elongated hexagon nanopillar I c + and Ic as a function of applied field Switching reproducibility Magnetic layer thickness dependence device schematic R/R vs. RA Current driven switching in multi-domain samples J c vs. t free Thermally activated switching diagram Pulsed current induced switching J c vs. t free with varying current ramp rates Dipolar coupling field vs. d Cu Schematic of Cu thickness dependent samples J c +, Jc, and RA vs. d Cu Schematic of Fe/Au/Fe devices R(H) for Fe/Au/Fe nanopillar R(I) for Fe/Au/Fe nanopillar Schematic of FeNiCo devices R(H) for FeNiCo alloy nanopillars R(I) for FeNiCo alloy nanopillars M versus H for a FeNiCo alloy film Current induced switching under the effects of dipolar coupling xii

15 6.9 Schematic of SAF devices Dipolar coupling field with and without SAF fixed layers R(H) for large field range R(H) for intermediate and low field ranges R(I) for SAF fixed layer nanopillar R(I) upon varying applied field xiii

16 Chapter 1 Introduction The study of electron transport in nanometer scale magnetic systems, magnetic nanostructures, is a rich and exciting field. Previous studies of magnetic nanostructures have already led to many applications, including higher sensitivity magnetic field sensors, higher data storage densities, and better data transfer rate in ground isolated networks. For the future, magnetic nanostructures may allow anything from a solid state quantum computer to a high speed non-volatile memory circuit. Nestled within the vast field of electron transport through magnetic nanostructures is the spin transfer effect, in which the magnetic state of a nano-magnet can be controlled with the application of a high current density flowing through the magnet. In Chapter 2, I will introduce some exciting discoveries in field of magnetic nanostructures, including the spin-transfer effect. I will detail some current applications based on magnetic nanostructures and will discuss the possible impact of the spin-transfer effect in these applications. Lastly I will discuss the possibility of future spin transfer based applications not possible with other magnetic nanostructure based effects. As the spin-transfer effect is primarily studied in devices where the current is run perpendicular to the plane of the magnetic layers, in Chapter 3 I will describe the fabrication technique used at Cornell to fabricate this class of device. I used nano-fabrication 1

17 2 techniques to fabricate isolated pillars, nanopillars, with electrodes oriented in such a way as to allow current to run perpendicular to the magnetic layers. Using the fabrication technique outlined in this chapter we have successfully fabricated devices ranging in size between 50 and 300 nm, which has applications for the study of a whole range of magnetic nanostructure phenomena, and not just spin transfer. In Chapter 4, I introduce the theory of the spin-transfer effect as well as early spin transfer experiments. I will compare our experimental technique outlined in Chapter 3 with previous studies of the spin-transfer effect and detail the advantages of lithographically defined pillars over these earlier techniques. I will detail the investigation of high current density effects in Co/Cu/Co nanopillars, including spin transfer as well as the more classical effect of the magnetic field produced by the current. In Chapter 5, I describe an investigation aimed at probing the origin of the spintransfer effect. Improvements in nanopillar fabrication techniques have provided enough device control so as to allow accurate device to device comparisons, allowing the systematic comparison of experiment to theory. Given this we have been able to probe the validity of the various models that seek to explain the spin-transfer effect. In Chapter 6, I describe studies that seek to extend the spin transfer effect into more technologically relevant systems than Co/Cu/Co nanopillars. The use of the spin-transfer effect is hindered by such things the required power, scalability, and magnetic switching quality. In this chapter I will describe efforts to improve spin transfer devices with reference to these and other parameters. Doing such has provided a possible solution to the problem of unwanted spin transfer based effects in nanoscale magnetic field sensors. Finally in Chapter 7 I conclude with a summary of completed spin transfer work as well as reiterate current problems with our understanding of the spin transfer phenomena.

18 Chapter 2 Electron transport through magnetic nanostructures: fundamental properties and applications 2.1 Introduction The application of magnetic nanostructures to electronics is generally called spintronics, which refers to electronic components that are sensitive to both the charge and spin of the electronic carriers. In the past decade spintronic devices have led to ever increasing magnetic storage densities and have brought a whole new class of low and high field magnetic field sensors. Also, galvanic isolators composed of spintronics devices have led to an order of magnitude improvement in data transfer rate between ground isolated networks. Spintronic devices are also being developed that promise to revolutionize solid state non-volatile memory circuits, increasing speed and decreasing power consumption. Driving the tremendous impact that spintronics applications have and will continue to make have been fundamental discoveries involving electron transport through magnetic 3

19 4 systems. In this chapter I will present an overview of some of these discoveries, as well as introduce the newly demonstrated spin-transfer effect. Lastly, I will discuss possible spin transfer based spintronic applications as well as spin transfer imposed limitations to the scalability of current spintronic applications. 2.2 Electron transport through magnetic nanostructures Giant magnetoresistance Introduction to GMR: initial observations As some magnetic systems are reduced in size to the nanometer scale their magnetic state can be controlled by means other than an applied magnetic field. A giant magnetoresistance (GMR) multilayer is one such system. In GMR multilayers, at zero applied field, the relative magnetic alignment between adjacent ferromagnetic layers in a ferromagnet (F)/ normal metal (N) multilayer (Fig. 2.1a) system can be set into a preferred direction, either parallel or antiparallel alignment, with the suitable choice of the normal metal spacer layer thickness [1, 2]. The coupling that forces this alignment, known as indirect exchange coupling, can, in the antiferromagnetically coupled case be such as to increase the saturation field, H s, of the ferromagnetic layers by more than two orders of magnitude [3]. As an applied magnetic field forces adjacent F layers parallel, a large change in resistance can also occur, R R = R antiparallel R parallel R parallel, since the resistance of the system in the parallel and antiparallel magnetic states can be quite different different (Fig. 2.1b). The first multilayer system that was shown to exhibit this large antiparallel coupling

20 5 Figure 2.1: (a) Diagram of GMR multilayer system. (b) Diagram of typical R/R as a function of applied field with the current flowing in the plane of the magnetic layers.

21 6 and large magnetoresistance were thin layers of Fe/Cr [4]. The GMR ratio, GMR ratio = R R = R(H = 0) R(H s), R(H s ) in Fe 1.9 nm/cr 0.9 nm multilayers has been measured to be up to 80% at 4 K with a saturation field greater than 8 koe for current flowing in the plane of the magnetic layers, compared to < 100 Oe for isolated Fe films of the same thickness [3]. Since the initial study of Fe/Cr multilayers, the GMR ratio and indirect exchange coupling has been explored with many different materials. Even higher GMR values were found in Co/Cu multilayers, with over 120 % GMR reported at 4K [5] while much higher values of coupling were found with Ru spacer layers, with an order of magnitude larger H s than with Cu or Cr [6]. However, despite this stronger coupling studies looking at GMR in Co/Ru multilayers have shown only a small ( 10%) change in resistance versus field [6]. As the spacer layer thickness is varied, the sign of the indirect exchange coupling, obtained from measurements of H s, and the GMR ratio were shown to oscillate as a function of thickness, with the GMR ratio and the indirect exchange coupling oscillating in phase [7] (Fig. 2.2). Thus, the indirect exchange coupling can, at zero field, force the adjacent F layers to be largely parallel or antiparallel, depending on the spacer layer thickness, resulting in small or large GMR ratios respectively. In general the strength of the coupling in Co/transition metal multilayers was shown to be proportional to the number of d and sp electrons in the N layers, which is materials dependent [6]. This implies that this coupling is electronic in origin, and not due to magnetic field interactions between the adjacent F layers.

22 7 Figure 2.2: R/R in a Cu/Co multilayers system as a function of the Cu spacer layer thickness.

23 8 Oscillatory exchange coupling origin Though the GMR and indirect exchange coupling are seen to oscillate together, as a function of spacer layer thickness, the root cause is different. The coupling is due to the occurrence of quantum size effects [8]. The spin up and spin down bands of the ferromagnet are different, having quite different band structure at the Fermi level, and in general both are different from the band structure of the normal metal spacer layer. A result of this is that some of the electron states of the spacer layer are strongly confined to be strictly within the normal metal spacer layer. The end result is that these quantum well states in the normal metal layers within the magnetic multilayer can make one magnetic state much more energetically favorable than the other. A simplification of this quantum well state model is the Ruderman-Kittel-Kasuya- Yosida (RKKY) interaction [8]. If a single fixed impurity spin exists in a metal then the conduction electrons will screen the spin state, forming spin-polarized concentric rings around the spin state (Fig. 2.3a). If a second impurity spin is added to the metal then the screening potential between of the two free spins will interact. This interaction is of the form f(x) sin(x) xcos(x) x 4. With isolated spins this interaction falls off as 1/r 3. Due to this interaction it could be energetically favorable for the two spins to align parallel or antiparallel depending on the distance between them. This RKKY treatment of isolated spins can be extended to include continuous magnetic layers by simply summing over all magnetic lattice sites (fig 2.3b). Using the RKKY interaction modified for continuous magnetic layers, oscillatory coupling can be derived. The decay of the amplitude of the interaction is now more complicated, but goes something like 1/r 2-1/r 3 where r now is the normal metal spacer thickness.

24 9 Figure 2.3: (a) Spin distribution surrounding a fixed impurity spin in a metal and (b) an extension of this spin distribution for continuous magnetic films which give rise to indirect exchange coupling.

25 10 Magnetoresistance in magnetic multilayers The root cause of magnetoresistance in magnetic multilayers is spin dependent scattering at the interface and within the F layers, which reduces the mean free path of one electron spin band of the F layer relative to that of the other spin band. In the Co/Cu system it is the minority electrons, in the down spin band of Co, that have the shorter mean free path. To go from this spin dependent scattering to magnetoresistance the simple twochannel resistor model can be employed. This simple model only treats the result of an asymmetry in the spin dependent scattering and not the origin or location of the scattering, which can occur at the F/N interface or in the interior and can be different depending on the material being used. For example, the scattering that dominates the GMR in Co occurs primarily at the F/N interface while FeNi, in addition to interfacial scattering, also has a large scattering component from the interior of the F layer [9]. The two channel model treats both the same. Here the spin up electrons and the spin down electrons are broken up into two current paths (fig 2.4). For a Co layer, the spin up conductivity is higher than the spin down conductivity due to an asymmetry in this spin dependent scattering. When the layers are parallel, one spin channel sees a low resistance while the other spin channel sees a high resistance. Since the spin up channel effectively shunts the current this results in an overall low resistance. When the layers are antiparallel, both spin channels see an intermediate resistance, a difference which provides the giant magnetoresistance. Experimental geometries used to study GMR There are two general geometries that are used to study GMR and other magnetic nanostructures. These geometries refer to the direction of the current flow with reference to the plane of the magnetic layers. In the first geometry, referred to as the current in plane (CIP) (Fig. 2.5a) geometry, the current is run along the direction of the magnetic layers.

26 11 Figure 2.4: Two channel GMR model showing (a) parallel and (b) antiparallel alignment of adjacent F layers.

27 12 The second geometry is referred to as the current perpendicular to the plane (CPP) geometry (Fig. 2.5b). For the CIP geometry all that is needed is to attach probes at two points along the multilayer, which can be separated by macroscopic distances, > 100 µm. With the CPP geometry to insure that the electron transport properties are dominated by the multilayer the current source and sink can only be separated by the thickness of the multilayer, typically < 1 µm for a large number of F/N repeats but can be < 100 nm for a small number of F/N repeats. Because of this constraint very few techniques have been used to make CPP devices, with most involving complicated microfabrication techniques [10 13]. Because of the greater ease in fabricating CIP devices, much of the early work in characterizing electron transport in GMR multilayers was done in the CIP geometry. However, the CPP mode is more useful when comparing experiment to theory. In the CPP mode the electrons must traverse all the layers and may be closely approximated by 1-D transport through the layers while in the CIP mode, the electron paths are much less well defined. Therefore, comparing CPP data to theoretical models of spin dependent transport is much easier and more direct Spin valves Although the size of the magnetoresistance in exchange coupled magnetic multilayer is impressive, for applications involving low fields this large exchange coupling prevents their implementation. For low field applications, the important parameter is R/R H s such that a small change in applied field, < 50 Oe, will result in a large change in R/R. For this purpose spin valves were created [14], which is a GMR like F/N/F system where with suitable application of field the direction of magnetization for the two F layers can be controlled independently. Changing the relative angle of magnetization between the

28 13 (a) e - (b) e - Nonmagnetic layer Ferromagnetic layer Figure 2.5: (a) CIP and (b) CPP geometries.

29 14 two F layers effectively creates a spin valve because as the angle changes so does the conductivity of the multilayer. A spin valve uses low coercivity materials where the spacer layer is such that the layers are not coupled via indirect exchange coupling, shown in Fig Typically a free and fixed layer are defined, where the coercivity of the two individual F layers are substantially different. In addition to employing different values of coercivity to distinguish between the two F layers, the fixed layer is also usually pinned in a particular direction, which is achieved through exchange biasing. By placing a ferromagnetic layer in contact with an antiferromagnetic layer, the ferromagnetic layer is biased in a direction defined by the antiferromagnet [15] (fig 2.6). Although the mechanism which causes this exchange biasing entirely known, it has been shown that the direction of this pinning can be defined by annealing the AF-F bilayer in the presence of a magnetic field above the Neél temperature for the antiferromagnet, the temperature above which the antiferromagnet becomes paramagnetic [16]. Upon cooling below the Neél temperature the antiferromagnet relaxes into its lowest energy state, which is determined by the magnetization direction of the F layer, resulting in a pinning of the F layer. Some of the various AF materials that are typically used for exchange biasing spin valves are FeMn, NiMn, PtMn, and IrMn, all of which are chosen because of their high exchange bias field, > 150 Oe, and the stability of the exchange bias field to well above room temperature [17]. Using exchange biasing to pin one F layer while leaving the other F layer effectively free makes it possible to more or less arbitrarily change the angle between the magnetization vectors of the two F layers. In GMR multilayers, since there is no preferred direction for any of the magnetic layers the quality of the magnetic ordering is controlled by the strength of the indirect exchange coupling, which in turn is controlled by the thickness of the N layers. This prevents the true determination of whether GMR is dominated by an

30 15 Figure 2.6: (a) Schematic of standard spin valve (b) Diagram of R vs. H behavior showing the effects of exchange biasing.

31 16 interaction between the current and the quantum well states in the N layers, or whether GMR is due to spin dependent scattering in the F layers. With spin valves, the ability to more or less arbitrarily control the direction of the magnetization vectors of the two F layers, allows the control of the quality of the magnetic ordering, easily placing the spin valve into near perfect antiparallel or parallel alignment. Using this property of spin valves, Speriosu et al. showed that the size of the magnetoresistance in magnetic multilayers is not a direct result of indirect exchange coupling [18]. They showed that the magnetoresistance in spin valves did not oscillate with spacer layer thickness, even though the size of the indirect exchange coupling clearly did, measured by the variations in H s of the free layer. Therefore, the magnetoresistance in magnetic multilayers can be thought of as being due mainly to spin dependent scattering at the interface and within the F layers. One potential problem for spin valve microstructures, where patterned films rather than continuous films are used, is that the magnetic layers will create magnetic fields due to magnetostatic edge charges which can affect the other layers in the spin valve stack, an effect also known as dipolar coupling. In order to screen the field from the fixed layer at the free layer, so that the functioning of the spin valve will be controlled with an external field and not by the dipolar coupling fields from within the spin valve stack, the F/AF bilayer is replaced with a F/N/F/AF (Fig. 2.7), known as a synthetic antiferromagnet (SAF) [19]. Here the indirect exchange coupling through the normal metal is tailored to cause antiparallel alignment between the two F layers. Usually nm of Ru is chosen due to its large antiferromagnetic indirect exchange coupling at this thickness. This results in the dipolar coupling fields from the two F layers acting in opposite directions. By balancing the thickness of the two F layers the net dipolar coupling field acting on the free layer can effectively be cancelled. As a result, using

32 17 this SAF structure a magnetoresistance of up to 15 % in < 50 Oe applied field has been achieved, even for sub-µm sized spin valves Tunneling magnetoresistance A slightly more recent discovery has extended low field magnetoresistance even further. In their simplest form, magnetic tunnel junctions (MTJs) are two F layers separated by a thin insulating layer (I). Current tunnels from one F electrode to the other (Fig. 2.8)a. If the tunneling process conserves spin an electron tunnels from an occupied state in one electrode must tunnel into an unoccupied state in another electrode with the same spin state. As a result the tunnelling rate is dependent on the spin-dependent density of states in the two electrodes. Thus, due to the uneven density of states between the up and down spin band in the two electrodes, the MTJ will show magnetoresistance as the magnetic orientation of the trilayer stack changes. In the simplest approximation for these CPP tunneling devices, known as the Julliere model [20], the R between parallel and antiparallel orientation is described by R R = 2P 1P P 1 P 2, where P 1 and P 2 are the electron polarization values of the ferromagnetic layers. Using Co [21] as the ferromagnetic layers this predicts a R/R of 30 %, which is larger than present CIP spin valves can provide by a factor of two. Various groups investigated these MTJs from the early 1980s using NiO, CoO, Gd 2 O 3, and Al 2 O 3 barriers. However, only small changes in resistance were observed until Moodera [22] utilized low temperature growth techniques to fabricate CoFe/Al 2 O 3 /Co MTJs with over 25 % tunnelling magnetoresistance (TMR) at 4.2 K. In this approach, which used high vacuum evaporation, the CoFe was deposited onto glass slides at 77K. Next a nm thick layer of Al was

33 18 Figure 2.7: (a) Standard spin valve showing dipolar coupling. (b) SAF spin valve showing the dipolar coupling fields from the individual F layers in the SAF acting in opposite directions, resulting in a reduced net dipolar coupling effect.

34 19 Figure 2.8: (a) Magnetic tunnel junction schematic. (b) Diagram of R vs. H behavior showing the effects of exchange biasing in MTJs.

35 20 deposited on top of the CoFe. The substrate was warmed to room temperature, whereupon the Al is oxidized by exposure to an O 2 plasma in order to form an insulating Al 2 O 3 tunnel barrier. Finally the top Co electrode was deposited at room temperature. Since the initial work of Moodera, higher values of TMR have been achieved in Al 2 O 3 based MTJs using sputtering to deposit the metallic layers, and an Ar-O 2 plasma to oxidize the Al in order to form the insulting barrier, with reported TMR values up to 48 %. Even further gains in TMR have been achieved by oxidizing the Al with the use of a Kr-O 2 plasma, with reported TMR values up to 60 % at room temperature [23]. Surely, Al 2 O 3 is not the only barrier material that can be used and other materials are currently being explored with emphasis both on extending TMR and lowering the resistance of MTJs so as to make them more compatible with current applications [24, 25] Magnetic semiconductors Recently a large amount of focus has been placed on extending spintronics to semiconductor systems, as integrating spin-dependent transport into semiconductors may speed up computers or allow new spintronics applications. Mush of this interest was motivated by a recent study of spin lifetime in certain semiconductor systems. In particular using pulses of circularly polarized light to create a spin imbalance in GaAs, Awschalom showed that spin lifetimes in that material can be can be very long, in some cases > 100 ns [26]. This and related work showed that semiconductors may be a promising materials system for spin based quantum computing applications, which will require long spin lifetimes. However, getting the spins into the semiconductor, without optical excitation, until recently has been a major challenge. Experiments measuring magnetoresistance in F/S/F trilayers, a way to probe the spin injection efficiency, have so far only produced very small

36 21 magnetoresistance values, < 0.1 % for Co/Si/Co [27]. In this approach the problem lies not in spin lifetimes in the semiconductor but in injecting the spins at the Fermi surface from the ferromagnet into the semiconductor. The difficulty of spin injection is due to the large difference in conductivity between the ferromagnet and semiconductor that prohibits efficient spin injection in a diffusive transport regime. To inject spins from the spin polarized material into the semiconductor with reasonable efficiency the spin polarized material must either be nearly 100% spin polarized or have nearly matching conductivity [28]. As an alternative reasonable spin injection has been achieved using paramagnetic semiconductors at high magnetic field [29] which works by having an interface impedance that dominates over that of the electrode. A different approach is to create a ferromagnetic semiconductor. The field of magnetic semiconductors, has been studied for quite some time, however it has been revitalized with Ohno s discovery of ferromagnetism in Mn doped GaAs up to 110 K [30]. This class of ferromagnets are called dilute magnetic semiconductors (DMS). In DMS the dopant atoms are responsible for both the charge carriers and the magnetization, which allows the coupling of magnetic states to electrical conductivity in semiconductor devices. Because the charge carriers originate from the magnetic dopants the electron polarization can be very high, with values > 85% being reported. The main problem of these DMS is the low Curie temperature, T c. The highest, T c, for these devices has been achieved with 5% Mn concentration, resulting in T c = 120 K, which is too low for many practical applications. The T c might be expected to increase with still higher Mn concentration, however a Mn concentration above 5% is found to result in clustering of the Mn atoms, i.e. a segregation of Mn atoms from uniform distribution throughout the GaAs to small < 10 nm diameter clusters of Mn. As the result, there is now a large effort underway to find alternative magnetic semiconductor materials

37 22 with Curie temperatures above room temp so that the incorporation of these materials into spintronics applications would be feasible. A few promising materials have been discovered over the last couple of years, including Mn doped GaN [31] and Co doped titanium oxide with reported T c well above room temperature, > 300 C for Co doped titanium oxide although there remains questions about magnetic clustering effects in these systems Spin transfer Spin transfer is a newly demonstrated effect in which the magnetic state of a nano-magnet can be controlled with an applied current flowing through the nano-magnet. The detailed mechanism for spin transfer remains a point of contention, with at least four theoretical models to describe the effect [32 35]. However, experimentally it has been shown that by applying a large spin polarized DC current through a magnetic nanostructure the current will exert a torque on the nano-magnet and, if large enough, cause a change in the magnetic state [36, 37]. Figure 2.9 shows a diagram of the spin-transfer effect. One magnetic layer is used as a source for spin polarized current which is then injected into the next magnetic layer. The same spin dependent scattering that is responsible for GMR aligns the spins with the local magnetic moment, via a spin torque that ultimately arises from the exchange interaction in the ferromagnet. Since the overall spin angular momentum must be conserved; an opposite torque will be exerted on the magnetic moment itself. Given enough injected current the spin torque will drive the magnetic moment into another magnetic state. For low applied field it has been well demonstrated that this can cause magnetization reversal. At high field, experimental work has not yet been able to make an absolutely clear distinction between whether the spin torque results in a partial magnetization reversal and the excitation of uniform or non-uniform spin waves. To date,

38 23 this spin transfer effect has been studied in single magnetic films, F/N/F trilayers, and magnetic multilayers. However the majority of the work has been done in CPP Co/Cu/Co pseudo-spin valves, spin valves without the pinning layer, which has been the main focus of this thesis research. Although most work has been done in Co/Cu systems there is new interest in exploring this spin-transfer effect in different systems, which has raised many interesting questions. Can spin transfer occur in systems such as magnetic semiconductors or even magnetic tunnel junctions? Will modifications in electron transport properties in these systems, as compared to the CPP Co/Cu/Co pseudo spin valve, still allow the occurrence of spin transfer effects? Furthermore, can these systems provide any advantages over previous spin-transfer systems? There is much work to be done before these questions can be answered. However, if spin transfer can occur in these other systems then that furthers the possibility that spin transfer may one day provide a new tool in spintronics applications. 2.3 Possible impact of spin transfer in spintronics applications In this section I wish to outline some possible future impacts of spin transfer in spintronics applications, both desired and undesired. Some applications may suffer unwanted spin transfer effects due to the consequences of scaling device size to smaller and smaller dimensions, therefore even without a clear road map for the implementation of spin transfer into future applications, from this potential problem it is clear that it is important to study the effects of spin transfer so as to assess its future impact on present applications. On the other hand however, spin transfer may lead to innovations in present spintronics applications or even to entirely new spintronic applications altogether.

39 24 Figure 2.9: Diagram of spin-transfer effect in F/N/F pseudo spin valves.

40 Field sensors Currently the largest market for magnetic nanostructures is in low magnetic field sensors, or more specifically for magnetic field sensors employed in high-density magnetic storage. In magnetic storage a disk of magnetic material rotates very rapidly under a read-write head that has the capability to read and write magnetic information on the disk (Fig. 2.10). Currently the magnetic material on the disk is a continuous film of a high coercivity alloy. The film is composed of 10 nm size grains, which are relatively uncoupled. This is achieved by doping the film with a nonmagnetic material that tends to migrate to the grain boundaries, which weakens the grain to grain magnetic coupling. An example of currently used magnetic media is an alloy of Co-Pt-Cr-Ta, where the Pt serves to increase the coercivity of the Co while the Cr-Ta tends to migrate to the grain boundaries, magnetically isolating separate grains [38]. The magnetic moments of the media are separated into bits and each bit is written into a 0 or 1 state that is read by the sensor. Although today the read sensor in the read-write head is composed of a 250 nm wide CIP SAF spin valve, it is unclear how these read head sensors will progress over the next decade. In the past few years the areal storage density in magnetic storage devices has been increasing at an incredible rate, > 60% per year. Since the patterned width of the sensor scales roughly with the square root of the areal density this states that sensor size has been scaling at > 25 % per year. Keeping this trend going would require sensor widths < 15 nm within 10 years. As these sensors scale down to such small dimensions alternative read head sensors may have to be employed, such as magnetic tunnel junctions and even CPP spin valves. One main constraint in scaling the lateral dimension of these sensors is the sensor resistance. In order to impedance match with silicon based amplifiers, the sensors are constrained to a resistance range between Ω. Though MTJs pro-

41 26 vide a larger R/R than CPP spin valves, which is desirable for signal to noise concerns, the high resistance of known tunnel barriers may force the use of CPP spin valves due to this impedance matching issue. If CPP spin valves are used as scaling progresses, while keeping the device resistance constant, the current required to preserve the same signal to noise ratio in these devices will remain unchanged or even increase due to increased thermal fluctuation effects. This will result in a continual increase in the bias current density of the CPP sensors as their lateral dimensions continued to be scaled down. At some point the effects of spin transfer, an unwanted parasite in field sensors, may become an increasingly important concern. This places special importance in establishing the boundaries of the spin-transfer effect so that future read-write head sensor designs can consider this potential issue Magnetic random access memory Recently a considerable effort has been placed in creating a random access memory technology using spintronics devices. The main thrust at the present time is focused on the use of magnetic tunnel junctions, an approach that has been dubbed MRAM or Mag- RAM (Fig. 2.11). In MRAM a MTJ is used such that, by a suitable application of field generated by a nearby current write line, the free layer can be forced to change its magnetization state. The MTJ then remains in this state until a suitable field is applied to reverse the magnetization of the free layer. The read process is done by passing a sense current through the junction and comparing the resistance of the MTJ to a reference. On a historical note, this technology can be considered as the micro-electronic analog of ferrite-core memory. In ferrite-core memory, rings of magnetic material act as memory elements that are inductively coupled to write and read wires. The utility of such a magnetic memory device comes from the non-volatile nature of the magnetic state. The

42 27 Figure 2.10: Read head and magnetic disk used in magnetic storage.

43 28 Figure 2.11: Schematic of typical MRAM design.

44 29 advantage of MRAM primarily comes from this non-volatile nature, coupled with fast write and read speed, and an unlimited number of write cycles. In the non-volatile memory market the current leading technology, Si based Flash RAM, is much slower than MRAM demonstrations and will fatigue, or fail, after a set number of write cycles [39]. In hopes of capitalizing on this advantage, MRAM has left the research stage and is in the development stage with products planned for release in Motorola in particular has presented working 1 Mbit chips operating at 50 ns cycle time with a 7.1 µm 2 cell size [39]. The main challenge facing the wide-scale implementation of MRAM appear to be the write cycle. In order to address a dense array of MTJ bits a crossed write-line architecture is used. Thus, to address a single bit a pulse of current is applied to each of the two perpendicular write lines. Where the write lines cross the bit will be addressed. Thus, the addressed bit will see a field with an easy axis component that is only twice the easy axis component of the field that every other bit along the write line will see. Therefore, for a successful circuit, every bit must switch at a field of H c but no bit can switch at a field of H c /2. This is called the half field problem. Achieving successful MRAM that meets this requirement is very challenging. There are two main issues when dealing with the half-field problem. First the size of the junctions in MRAM is small enough that thermal fluctuations may give too broad a spread in the switching fields of the nano-magnets [40, 41]. Second, the MRAM technology relies on the reproducibility of thin film coercivity, which is a highly non-linear and very materials sensitive quantity [42]. Small deviations in the patterned shape and local materials inhomogeneity can lead to large changes in coercivity. Both of these aspects make the job of manufacturing fully functioning defect free MRAM chips with high bit counts very challenging.