IM2654 Smart Electronic Materials Prof.

Alex Grishin 2007-10-15

MRAM:
Giant and Tunneling Magnetoresistance Phenomena

Cornelius THIELE David KOHEN Fabien GAYRARD Benoit BONDU

Table of contents
Table of contents ........................................................................................................................ 2 Giant Magnetoresistive Effect.................................................................................................... 3 Origin ..................................................................................................................................... 4 Contribution of the interface .................................................................................................. 6 GMR structures ...................................................................................................................... 6 Applications ........................................................................................................................... 7 Tunneling Magnetoresistive Effect ............................................................................................ 8 Spin Dependant Tunneling..................................................................................................... 8 Magnetic tunnel junction........................................................................................................ 9 Tunnel Magnetic resistance effect.......................................................................................... 9 MRAM ..................................................................................................................................... 11 MRAM cell: ......................................................................................................................... 11 Reading process: .................................................................................................................. 12 Writing process .................................................................................................................... 13 Architecture of MRAM ........................................................................................................ 15 Improvements....................................................................................................................... 16 Different Write Processes..................................................................................................... 16 Toggle Writing ..................................................................................................................... 16 Thermally assisted writing ................................................................................................... 17 Spin Torque Transfer ........................................................................................................... 17 Other Technology................................................................................................................. 18 History & Commercialization .............................................................................................. 18 Future of MRAM ................................................................................................................. 19 References ................................................................................................................................ 20

Giant Magnetoresistive Effect

Giant magnetoresistance (GMR) is a magnetoresistive effect, it means the change in electrical resistance of some materials in response to an applied magnetic field. When you apply a magnetic field to magnetic metallic multilayers such as Fe/Cr and Co/Cu, in which ferromagnetic layers are separated by a conducting nonmagnetic spacer layer, which is a few nm thick, it results in a significant reduction of the electrical resistance of the multilayer. This effect was found to be much larger than other magnetoresistive effects observed in metals and thus, was called giant magnetoresistance. GMR was independently discovered in 1988 in Fe/Cr/Fe trilayers by a research team led by Peter Grnberg of the Jlich Research Centre, who owns the patent, and in Fe/Cr multilayers by the group of Albert Fert of the University of Paris-Sud, who first saw the large effect in multilayers that led to its naming, and first correctly explained the underlying physics. The birth of spintronics can be associated at this discovery. Grnberg and Fert have received a number of prestigious prizes and awards for their discovery and contributions to the field of spintronics, including the Nobel Prize in Physics in 2007. The change in the resistance of the multilayer happens when the applied field aligns the magnetic moments of the successive ferromagnetic layers. Without a magnetic field the magnetizations of the ferromagnetic layers are antiparallel. When you apply the magnetic field, which aligns the magnetic moments and saturates the magnetization of the multilayer, the consequence is a decrease in the electrical resistance of the multilayer.

Fig. 1: Schematic representation of the GMR effect. (a): Resistance of the magnetic multilayer as a function of applied magnetic field. (b): Magnetization configurations (arrows) of the trilayer at various magnetic fields (c): Magnetization curve for the multilayer. One definition of the GMR ratio is: GMR = (RAP-RP)/RP with RAP and RP respectively the resistance in the antiparallel and parallel configurations. In Fe/Cr and Co/Cu multilayers the magnitude of GMR can be higher than 100% at low temperatures.

Origin
GMR can be qualitatively explained using the Mott model. Mott tried to explain, in 1936, the sudden increase in resistivity of ferromagnetic metals as they are heated above the Curie temperature. Mott developed two main ideas: First, the electrical conductivity in metals can be considered as two largely independent conducting channels. One channel corresponds to the spin-up and the other corresponds to the spin-down electrons. The channels are distinguished according to the projection of their spins along the quantization axis. In metals, the probability of the scattering processes in which the spin is conserved is stronger than the probability of spin-flip scattering processes. This means that the spin-up and spin-down electrons do not mix over long distances and, thus, the electrical conduction occurs in parallel for the two spin channels. Second, in ferromagnetic metals the scattering rates of the spin-up and spin-down electrons are quite different, whatever the nature of the scattering centers is. In fact, the band structure in a ferromagnet is exchange-split. As a consequence, at the Fermy energy the density of states is not the same for spin-up and spin-down electrons. Since scattering rates are proportional to the density of states, the scattering rates and therefore resistivities are different for electrons of different spin. With Motts principles we can understand the GMR phenomenon. We assume that the scattering is strong for electrons with spin antiparallel to the magnetization direction, and is weak for electrons with spin parallel to the magnetization direction. This reflects the difference in the density of states at the Fermi level, like says Motts second argument. For the parallel-aligned magnetic layers (fig 2a), the spin-up electrons pass through the structure almost without scattering, because their spin is parallel to the magnetization of the layers. On the contrary, the spin-down electrons are scattered strongly within both ferromagnetic layers. Indeed the magnetization of the layers is antiparallel to their spin. Since conduction is the consquence of the two spin channels, the total resistivity of the multilayer is determined mainly by the highly-conductive spin-up electrons.This resistivity is low. For the antiparallel-aligned multilayer (fig 2b), both the spin-up and spin-down electrons are scattered strongly within one of the ferromagnetic layers, because within one of the layers the spin is antiparallel to the magnetization direction. Therefore, the total resistivity of the multilayer is high.

Fig. 2: Electron transport in a multilayer for parallel (a) and antiparallel (b) magnetizations of the successive ferromagnetic layers 4

Due to the exchange splitting of the d bands, the number of occupied states is different for the up-spin and down spin electrons, giving rise to the non-zero magnetic moments of 2.2B, 1.7B and 0.6B for Fe, Co and Ni respectively. In order to distinguish between the high and low-occupied spin states, the terms majority-spin electrons and minority-spin electrons are usually used.

Fig. 3: Electronic band structures (left panels) and the density of states (right panels) of Cu (a) and Co for the majority-spin (b) and minority-spin (c) electrons.

The band structure of non-magnetic Cu is same for the spin-up and spin-down electrons. d bands are fully occupied and there is a dispersive sp band at the Fermi energy. The result is a high conductivity of Cu. The electronic structure of ferromagnetic Co is different for the two spin orientations and is characterized by the exchange-split d bands. The Fermi level lies within the sp band for the majority-spin electrons, which leads to high conductivity of majority-spin channel. The Fermi level lies, however, within the d band for the minority-spin electrons resulting in low conductivity of the minority-spin channel. In the latter case the sp electrons are strongly hybridized with the d electrons, which decreases their contribution to conduction.

Contribution of the interface

The interfaces in a magnetic multilayer play an important role in spin-dependent transport in bulk elemental ferromagnets. The band structures of the two adjacent metals are different. So there is a potential step at the interface and the transmission probability across the interface is less than 1. If the interface separates ferromagnetic and non-magnetic metals, the transmission will be spin-dependent due to the spin dependence of the band structure of the ferromagnetic layer. We can study the band structures of Co and Cu ( Fig.3). We can note similarities between the band structure of Cu (Fig.3a) and the band structure of the majority spins in Co (Fig.3b). This good band matching implies a high transmission for the majority-spin electrons across the Co/Cu interface. On the contrary, as is seen from Fig.3a and 3c, there is a relatively large band mismatch between Cu and the minority spins in Co and thus the transmission of the minorityspin electrons across the Co/Cu interface is expected to be poor. Therefore, the interfaces of the Co/Cu multilayer act as spin-filters. When the filters are aligned, the majority spin-electrons can pass through relatively easily. When the filters are anti aligned, the electrons in both spin channels are reflected at one of the interfaces. This spin-dependent transmission is an important part of the electronic transport in GMR structures.

GMR structures
GMR is observed in magnetic metallic layered structures in which it is possible to reorient the magnetic moments of the ferromagnetic layers relative to one another. There are four main structures : magnetic multilayers, spin valves, pseudo spin valves, and granular solids.

Fig. 4: Different GMR structures: magnetic multilayer (a), pseudo spin valve (b), spin valve (c) and granular thin film (d).

In magnetic multilayers (fig4.a) there is an antiferromagnetic interlayer coupling. The interlayer exchange coupling is mediated by the itinerant electrons in the metallic spacer layer and oscillates between ferromagnetic and antiferromagnetic as a function of the thickness of the nonmagnetic (NM) layer. By choosing an appropriate thickness of the spacer layer it is, therefore, possible to create an antiparallel configuration of the ferromagnetic layers and then reorient (align) the moments by an applied magnetic field. In a pseudo spin valve (fig4.b), the antiparallel alignment can be obtained due to different coercivities of the two ferromagnetic layers. The magnetic moments of the soft and hard magnetic layers switch at different values of the applied magnetic field providing a field range in which they are antiparallel and the resistance is higher. In a spin valve (fig4.c), the magnetization of one ferromagnetic layer is pinned by the exchange coupling with an adjacent antiferromagnetic (AF) layer, whereas the magnetization of the other ferromagnetic layer is free to rotate with the applied magnetic field. Although the measured values of GMR are higher in magnetic multilayers, spin valves are more attractive from the point of view of applications , because only small magnetic fields need to be applied to change the resistance. In magnetic granular solids (fig4.d), magnetic precipitates are embedded in a nonmagnetic metalic material. In the absence of the field the magnetic moments of the granules are randomly oriented. The magnetic field aligns the moments in a certain direction and thus there is a resistance drop.

Applications
The largest technological application of GMR is in the data storage industry. IBM was the first company to unveil hard disks based on GMR technology, and nowadays all disk drives make use of this technology. The GMR read head sensor in a hard disk is built using a spin valve . Other applications of GMR are as diverse as automotive sensors, solid-state compasses and non-volatile magnetic memories.

Tunneling Magnetoresistive Effect

The TMR is just a quantum mechanical tunneling effect between two ferromagnets, coupled with a magnetic scattering due to the spin of electrons. First will be explained the physical principle of the spin dependent tunneling; then the tunneling magnetic resistance will be explained using the simple Jullieres model.

Spin Dependant Tunneling

SDT is the phenomenon in which electrons can tunnel from a ferromagnet through an insulating barrier, when a voltage is applied. During the tunneling of an electron, its spin is assumed to be conserved. This phenomenon was first discovered by Tedrow and Meservey in 1970; they measured the tunneling current between a ferromagnet electrode and a superconductive metal electrode through a thin insulator layer. They demonstrated that the spin was conserved during the tunneling process and that the tunneling conductance was dependant on the degree of polarization of the magnetic electrode. The spin dependent tunneling can be explained for ferromagnetic metal quite simply. The 3d bands of this metal are exchange split: they are slightly shifted regarding their energy level. This leads to a different density of filled states at the Fermi level. The figure 1 illustrates the exchange splitting of the 3d bands in ferromagnetic metal.

Fig. 5: Energy diagram of a ferromagnetic metal

As can be seen in figure 5, the d-band is splitted, resulting in a different density of space at Ef, the Fermi level.

Magnetic tunnel junction

Magnetic tunnel junctions are just an addition to the Giant Magnetic Resistance junction where the conducting layer is replaced by an insulator layer. The thickness of this layer is rather small in order to allow electron tunneling. Complete prototype sensors were fabricated using an Al2O3 (alumina) barrier with a thickness of 1.3 nm. A simple tunnel junction can be seen below, in figure 6.

Fig. 6: Magnetic Tunnel Junction The two electrodes are made of ferromagnetic metal. Depending on the orientation of their magnetization, the resistance of the junction varies. The Tunneling MagnetoResistance (TMR) is the relative variation of the resistance, depending on the magnetization of the two electrodes. The two possible configurations are called parallel (P) and anti-parallel (AP) configuration. Nowadays magnetic tunnel junctions are made with transition metal ferromagnet and exhibit a TMR ratio up to 50% at room temperature.

Tunnel Magnetic resistance effect

The reason for the magneto-resistance effect will be explained hereafter. The spin dependent tunneling explained above that between two electrodes, electrons can tunnel through an insulator barrier. For the case of a magnetic junction, Jullieres Model is used to explain simply the basis of the phenomenon. In this model, two assumptions are made. First, the electrons conserve their spin during the tunneling. The conduction is thus the sum of the up-spin electron tunneling and the down spin electron tunneling. According to this assumption, for the parallel P, when the two ferromagnets have the same magnetization, the electrons from the majority spin-stats tunnel to the majority spin-state and the minority spinsstate to the minority spin-state. When the configuration is anti-parallel AP, the electrons from the majority spin-state tunnel to the minority spin-state and vice versa. See figure 7 for a better understanding.

Fig. 7: Spin dependant tunneling effect, in P configuration and AP configuration The second assumption of Julliere is that the conductance for a particular spin orientation is proportional to the product of the effective density of states of the two ferromagnetic electrodes. According to this, the tunneling current is not the same for the P and AP configurations. For the P configuration, the current can be written as: P configuration: AP configuration: Where , N represents the density of state of the majority spin-state n represents the density of state of the minority spin-state. The indicia l is for the left electrode, r is for the right electrode The arrows represents the spin-state. It is now clear that the current is higher for the P configuration. The TMR ratio is now defined as By defining the spin polarization of a magnet as: We can rewrite the TMR as Where Pl is the polarization of the left electrode Pr is the polarization of the right electrode

Julliere's model was used to estimate the magnitude of TMR in magnetic tunnel junctions from the known values of the spin polarization of ferromagnets obtained in experiments on superconductors. The first value of TMR calculated by Julliere was 14 % at helium temperature, for a Fe-Ge-Fe tunnel junction. Nowadays improved methods of deposition allow a higher TMR ratio, up to 50%. One of the most promising applications using TMR is probably the magnetic random access memory which will be described in the following section.

10

MRAM
MRAM cell:
In a MRAM device, each bit of information is held by one cell, also called a bit stack, composed of several thin layers. Here the TMR effect is used, not GMR, because the first is more efficient. Indeed, for the bit being correctly read, we need resistance (and thus voltage) drops which can be clearly sensed: the higher the TMR signals, the greater the separation of the low and high resistance values corresponding to the two bit states.

Fig. 8: TMR signal versus the applied magnetic field, showing the separation between the two bit states Recent breakthroughs in tunnel junction fabrication led to TMR signals close to 70% in CoFeB/AlOx/CoFeB and over 200% in junctions with a MgO tunnel barrier. But at this stage in the development of MRAM technology, MR values for integrated devices are typically in the range of 3050% (NiFe, NiFeCo). The basic MRAM cell element consists of a magnetic tunnel junction structure in which two ferromagnetic electrodes are separated by a thin insulating barrier. First, one finds a top layer of soft magnetic material (NiFe), whose magnetization can be easily inverted. To increase anisotropy, the bit stack is rectangular or oval, with the longest side parallel to the easy axis. The tunnelling junction is separated by an insulating layer, AlOx and more generally Al2O3. The magnetization of the bottom layer is fixed. Applying a magnetic field, the free layer will rotate its momentum in the same direction as the field, whereas the fixed layer magnetization will remain unchanged. One can thus arrange the configuration of the bit stack as parallel (bit = 0) or antiparallel (bit = 1).

11

Fig. 9: configuration of the magnetic tunnel junction structure depending on the applied magnetic field, showing the two bit states The bit resistance mainly determines the RC time constant and read access time, with typical cell values of 10 k allowing for ns access times, depending on the lead line capacitance. The characteristic resistance area product (R A) is determined by the tunnel barrier height and thickness t. Resistance dispersion around the central resistance values occurs due to issues in junction patterning, area dimensional control and barrier thickness non-uniformity induced by electrode roughness. It has been shown that it is possible to keep a maximum TMR signal over RA values ranging from 100106 .m2 corresponding to AlOx thickness between 9 and 20 . Much investigation is being realized for further decreases down to 1 .m2 keeping full TMR to allow the use of spin transfer induced magnetization reversal to write the bit cell. Bit cell scaling requires also lowering RA values to keep a constant bit resistance while shrinking the bit cell below the 90 nm node.

Reading process:
The device is read by measuring the effective resistance of the structure, which is a function of the state of the free layer. To do so, the free layer is connected to a conductive line, called bit line, and the bottom layer to a transistor, which is turned on. Then a current is applied in the bit line which flows into the bit stack and leaves through the transistor.

Fig. 10: Reading process in a bit cell

12

To assess the bit state the bit resistance is compared to a reference value mid-way between the bit high (Rhigh) and low (Rlow) resistance values. The inevitable resistance dispersion centred on Rhigh and Rlow must be reduced since it impacts directly on the read margin. Therefore there exists an optimum value of the read voltage, approximately 300 mV, which is a quite high value for integrated circuits nowadays.

Writing process
A bit is written by applying a magnetic field in the free layer in order to rotate its magnetization. To do so, the transistor is turned off and another conducting line is added beneath the bit cell, called digit line or word line. Applying a current in both lines, bit line and digit line, one can create two magnetic fields which, when combined, can reverse the magnetization of the free layer. The current direction in the digit line is fixed, and creates a magnetic field along the free layer hard axis of magnetization. The current direction in the bit line can be chosen, so the magnetic field generated is either positive or negative. This field is applied along the free layer easy axis of magnetization.

Fig. 11: Writing process in a bit cell It is important to notice that the two magnetic fields are necessary to change the orientation of the free layer, each one of them being too weak when applied alone.

13

Energy minimization shows that the easy axis field He required to reverse the magnetization is reduced by applying simultaneously a second perpendicular field along the hard axis Hh. The solution yields an astroid equation:

where K is the anisotropy constant and MS the saturation magnetization. The astroid plot is showed on fig. 12: switching occurs for any combination of fields for which the resulting field vector lies outside the astroid.

Fig. 12: Astroid plot showing the free layer configuration when applying only one magnetic field (blue and green points) and both (red points) The digit line field will slightly tilt the magnetization toward the hard axis, and then the bit line field will be sufficient to reverse the magnetization.

Fig. 13: Magnetization tilting when applying a hard axis field Depending on the direction of the current in the bit line the field will be either positive or negative, and then the magnetic tunnel junction will be either parallel or antiparallel. The hysteresis property ensures that once the bit is written and the fields are removed, the configuration will remain unchanged: the bit memory is non-volatile.

14

Architecture of MRAM
To design a MRAM device, the bit stacks are arranged in a rectangular array spanned by the bit and word lines, as showed in fig. 14:

Fig. 14: bit cells arranged in array By turning a bit cell transistor on and applying a current in the corresponding bit line, the designated bit, and only this one, will be read. For the writing process, the importance of the two fields is emphasized: once the selected bit cell transistor is off and currents applied in the corresponding bit and word lines, all the bit along both lines will be half-selected but the total magnetic field will be sufficient only at the intersection of them, i.e. on the selected bit cell, and only this bit will be modified.

Fig. 15: Writing process in a MRAM array Currents of the order of 5 mA are typically required to switch the state of an MTJ. These currents are considered large by integrated circuit standards and create a variety of challenges for write circuit design.

15

Improvements
To make the pinned layer even more impervious to external magnetic fields, one can replace it by two ferromagnetic layers separated by a thin coupling layer, typically Ruthenium. The magnetizations in the two layers are then strongly antiparallel coupled and it is almost impossible to influence this through an external field.

Fig. 16: Coupling Layer The stronger the magnetization, the smaller one MRAM cell can be.

Different Write Processes

The simple write process described above is the most widely adopted one. It is called the StonerWohlfarth method of selecting a bit in an MRAM cell array. However, it has some drawbacks, which is why most research for MRAM cells is currently concerning the write process. The state written with this method depends on the direction of the current in the bit line. The fact that this method requires the current direction to be switchable makes it somewhat special. Also, the generated field has an influence on nearby cells, making them more susceptible to thermal activation, which can introduce bit errors in the array.

Toggle Writing
In the toggle write process, the bit and word lines are at an angle of 45 to the easy magnetization axis of the bit. When applying a field through one or both of the lines, the magnetization of the bit will orient itself perpendicular to that field. This can be used to rotate the magnetization of the free layer. First, the bit line creates a field, then bit and word line together and then only the word line. The magnetization then aligns itself to the easy axis, which leads to an inverted initial state.

16

This means that the state of the bit has to be read before the write, because a write will always invert the current state. The advantage of this method over the StonerWohlfarth method is, that there is only one current direction.

Fig. 17: Toggle write process

Thermally assisted writing

By heating a MRAM cell, one can reduce the magnetic field strenght required to manipulate the magnetization of the free layer. When the temperature of the material approaches its Curie Point, the magnetization can change very easily. There are many approaches on how to achieve the heating of a cell. Most of them send a current through the bit or word line or even through the cell itself. Sending the current through bit/word line would also heat other cells in the the same block. It has been shown that a more complex design, where only the one cell that is written to is heated leads to better results. Because of the small magnetic fields required to switch that cell, non-heated cells are almost impervious to it.

Spin Torque Transfer

The write method that is to be researched for future MRAM cells is spin-momentum or spin-torque transfer. In structures smaller than 100nm it has been observed that a current of spin-polarized electrons can change the magnetization direction of its conducting carrier. The great advantage of this method is that it does not rely on any magnetic field to switch the bit. The magnetic moment carried by the spin of the electrons directly interacts with the ferromagnetic material.

17

Other Technology
Dynamic RAM (DRAM), which is widely used for volative memory in computers and elsewhere, is a highly integrated memory technology. A DRAM cell consists mainly of a condensator and a simple FET. DRAM cells need to be refreshed to keep the value they are holding, because the condensator loses its charge constantly. This typically happens every few dozen milliseconds. Because of this the power requirement of DRAM is rather high compared to MRAM. Static RAM (SRAM) is another volative memory technology, which is used for cache memory inside processors or other integrated devices. An SRAM cell constists typically of six FETs; four of them make up a flip-flop that holds the value and the other two couple the cell to a normal and an inverse bit line. Because the FETs have a very low capacity and therefor very low switching times this type of memory is very fast.

Flash memory is a non-volatile memory technology that is in widespread use. It used to be available only in small capacities, for memory cards or embedded products. But due to recent advances in data density, flash memory is being deployed in solid state disks as a replacement for hard disk drives. A Flash memory cell is basically a FET with two gates. A charge stored in the insulated floating gate influences the resistivity of the device. To fill the floating gate with electrons a high voltage is applied to the control gate. Electrons from the created n-channel can now jump into the floating gate. To erase the cell one applies an opposite voltage to the drain, so that electrons from the floating gate can tunnel there. Hence, each write cycle wears out the device. Current flash technology allows for millions of write cycles before the device ceases to work. Read access times for flash memory are not as good as for DRAM, and a write cycle takes significantly longer.

History & Commercialization

After the discovery of the GMR effect in 1988, it was only in 2000 that IBM and infineon established a joint research program to research MRAM. In 2002 Cypress Semiconductor, which is a company that manufactures memory chips, signed a technology exchange agreement with NVE, which owned a lot of intellectual property concerning MRAM. Cypress started selling MRAM chips in 2003, with a capacity of 128Kbit, manufactured in 180nm technology. However, the low memory density and the high cost of the chip prevented it from becoming a commercial success. In March 2005 Cypress

18

sold its MRAM subsidiary and replaced the chip in its portfolio by an SRAM chip that keeps its values through a capacitor. The former Motorola subsidiary Freescale Semiconductor, which is a big manufacturer of microchips for embedded applications, started selling a 4Mbit MRAM chip in July 2006. They do this still. Renesas, the joint semiconductor branch from Hitachi and Mitsubishi from Japan, announced in December 2005 that they wanted to start producing spin torque transfer MRAM chips in a 65nm process in cooperation with the american company Grandis very soon. Grandis is mostly a research company that sells intellectual property. They have some patents concerning spin torque transfer and how to utilize that technology for memory cells in their portfolio. However, the joint venture of Renesas and Grandis does not seem have produced a competetive and marketable chip yet. Concerning technological advances, in June 2004 Infineon unveiled a prototype chip with a capacity of 16Mbit. Sony announced in December 2005 that they had a working lab prototype of an MRAM chip that utilized spin torque transfer. In August 2007, IBM and TDK signed an agreement to work together in researching MRAM technology with the hope of reducing their costs when combining their forces on the way to a competitive MRAM chip.

Future of MRAM
Because of the superparamagnetic effect MRAM cells can only be shrinked to a certain level. When they are below a certain size, estimates are around 10nm, the magnetic orientation of the layer that stores the information cannot be retained. MRAM chips currently in production are mostly fabricated using old 180nm or 90nm technology, because the industry has been pushing Flash memory a lot more in the latest times and can't afford to use a newer fab to produce samples of a technology that is not in wide-spread use. However, semiconductor roadmaps suggest that 10nm fabrication technology will be reached at around 2015. MRAM manufacturers should have a working and competetive product ready far before that to stay in business.

19

References
P. M. Tedrow and R. Meservey, Spin-Dependent Tunneling into Ferromagnetic Nickel, Phys. Rev. Lett. 26, No. 4, 192195 (1971). T. Miyazaki, T. Yaoi, and S. Ishio, Large Magnetoresistance Effect in 82NiFe/Al Al2O3/Co Magnetic Tunneling Junction, J. Magn. Magn. Mater. 98, No. 12, L7L9 (1991). Perspectives of Giant Magnetoresistance, E.Y.Tsymbal and D.G.Pettifor,published in Solid State Physics, ed. by H. Ehrenreich and F. Spaepen, Vol. 56 (Academic Press, 2001) pp.113-237 Ricardo C. Sousa, I. Lucian Prejbeanu, Non-volatile magnetic random access memories (MRAM), C. R. Physique 6, (2005) 10131021 www.research.ibm.com/journal/rd www.physics.unl.edu

20