Encyclopedia of

Nanoscience and
Nanotechnology

www.aspbs.com/enn

Nano Spintronics for Data Storage

Yihong Wu
Department of Electrical and Computer Engineering, National University of Singapore,
Singapore 117576, and Data Storage Institute, Singapore 117608

CONTENTS the advance of nanotechnology, and the latter on an emerg-

ing field called spintronics [1–4]. The combination of these
1. Introduction two fields has played an important role in advancing the
2. Basics of Magnetic Data Storage areal density of magnetic recording from a few gigabits/in2
and Spintronics to the current level of more than 100 Gbits/in2 [5–8]. In
3. Magnetoresistance addition to hard disk drives, the technologies developed
have also been applied to magnetic random access memo-
and the Associated Read Sensors
ries (MRAMs) [9–11]. Further advance in the two fields is
4. Magnetic Aspects of Spin Valves the key to realizing terabits/in2 hard disk drives and giga-
5. Electronic Aspects of Spin Valves bit nonvolatile memories within this decade. This chapter is
intended to provide an updated review of nanometer-scale
6. Magnetic Tunnel Junctions
spintronics for data storage applications, with emphasis on
7. Magnetic Random Access Memory the application of metal-based spintronics in magnetic sen-
8. Semiconductor Spintronics sors and memories.
To have a clear picture of the scope of this review,
9. Future Trend of Data Storage
Figure 1 illustrates how spintronics is positioned in the hier-
and the Role of Spintronics
archy of various different types of data storage technolo-
Glossary gies. The data storage devices can be categorized into three
Acknowledgments major groups, that is, magnetic and optical data storage,
and solid-state memory. Each of them can be divided fur-
References
ther into several subgroups, and here, we only focus on
magnetic data storage and solid-state memory. Among the
magnetic storage devices, the hard disk drive (HDD) is
1. INTRODUCTION the dominant secondary mass storage device for comput-
Nanotechnology is one of the hottest research fields in the ers, and very likely also for home electronic products in
science and engineering arena. Of the many potential appli- the near future. The HDD is an integration of many key
cations of nanotechnology, one of the most promising ones technologies, including head, medium, head-disk interface
is in data storage, particularly the hard disk drives. This (HDI), servo, channel coding/decoding, and electromechan-
is because the physical size of the recording bits of hard ical and electromagnetic devices. Among them, the read
disk drives is already in the nanometer regime, and contin- head is the only component that has experienced the most
ues shrinking due to the ever-increasing demand for higher changes, including some revolutionary ones in terms of both
recording densities. If the pace of areal density increase the operating principle and the structural design and fabri-
is maintained at the current level for about ten years, the cation processes during the last decade. The ever-increasing
dimension of the recording bit will reach the sub-10 nm demand for higher areal densities has driven the read head
regime. At this level, both the writing and reading processes evolving from a thin-film inductive head to an anisotropic
will become extremely challenging, if not impossible. magnetoresistive (AMR) head, and recently, the giant mag-
The rapid shrinkage of bit size poses formidable chal- netoresistive spin-valve (GMR SV) head. There are two
lenges to the read sensors. This is because the sensor must different forms of GMR SVs, depending on whether the
be made smaller or at least comparable to the bit size, and at sense current flows in the film plane (CIP) or perpendicu-
the same time, its sensitivity must be improved continuously lar to the film plane (CPP). Currently, CIP SVs are domi-
so as to compensate the loss in signal-to-noise ratio due to nant, but CPP SVs are more promising for future extremely
the decrease in the bit size. The former has to rely heavily on high-areal-density heads. In addition to CPP spin valves, a

ISBN: 1-58883-001-2/$35.00 Encyclopedia of Nanoscience and Nanotechnology

Copyright © 2003 by American Scientific Publishers Edited by H. S. Nalwa
All rights of reproduction in any form reserved. Volume X: Pages (1–50)
2 Nano Spintronics for Data Storage

Read only
in Section 7. Section 8 touches briefly on the semiconduc-
Phase change
tor spintronics. Finally, we discuss the future trend of data
Magnetooptical
Inductive
Johnson storage and the role of spintronics in future data storage in
Section 9.
Holography transistor
D Optical AMR
A Spectral Head
T hole burning Floppy GMR 2 terminal 3 terminal
Media
A Tape MTJ

2. BASICS OF MAGNETIC DATA

HDI Metallic
S Magnetic HDD
T Servo

Proof's Only
O
R
DRAM
ZIP

Others
Channel
S
P
Semiconductor
Diode STORAGE AND SPINTRONICS
A I
SRAM Others
2.1. Magnetic Recording
G Solid N Transistor
E State Flash T
R Hybrid
FeRAM
Inductive
O
Monsma transistor
and Magnetic Memory
AMR Nano Spintronics N
MRAM I
GMR
for Data Storage
C
Spin FET
The hard disk drive at the system level is the integration
CRAM
MTJ
S Spin SET
of many key technologies and components, which include
RRAM Spin LED/laser

NANOTECHNOLOGY Spin RTD

the storage medium, read/write transducer, channel cod-
MTT
ing/decoding, servocontrol, head/disk interface, tribology,
and electromechanical and electromagnetic systems [12–14].
Figure 1. Spintronics and its applications in data storage and elec- In this subsection, however, we focus on the physics and
tronics. The acronyms are: DRAM—dynamic random access mem- materials aspects of magnetic recording, that is, topics
ory, SRAM—static random access memory, FeRAM—ferroelectric related to the storage medium and read/write heads [15].
random access memory, MRAM—magnetic random access memory, The thin-film medium for a hard disk drive is a stack of
CRAM—chalcogenide random access memory, RRAM—resistance
multiple layer thin films formed on either an NiP-coated
random access memory, ZIP—zip drive, HDI—head–disk inter-
aluminum alloy or a glass substrate [16]. Among the mul-
face, AMR—anisotropic magnetoresistance, GMR—giant magnetore-
sistance, MTJ—magnetic tunnel junction, FET–field-effect transistor, tiple layers of thin films, the early hard disk media only
SET—single-electron transistor, LED—light-emitting diode, RTD— employed a single magnetic layer as the recording layer,
resonant tunneling diode, MTT—magnetic tunnel transistor. which is usually a polycrystalline alloy of Co, Cr, and Pt with
additional elements such as Ta or B to improve the magnetic
magnetic tunnel junction (MTJ) with low junction resistance properties. The latest magnetic media, however, employ
is also one of the potential candidates for future read sen- more than one magnetic layer as the recording layer, such as
sors. The MTJ is particularly useful for MRAM applications. the antiferromagnetically coupled (AFC) media to improve
As is with the case of read sensors, the MRAM started from the thermal stability of the information bits recorded on the
the ferrite core design to AMR, GMR pseudospin valve, hard disk [17–19]. When recording the information onto the
and the current MTJ cell design. However, it is only after disk, a writer which can generate a sufficiently high magnetic
the introduction of MTJ that the MRAM taken off, and field is used to switch the magnetization of a localized area
is expected to be in mass production in a few years’ time. of the media to one of two fixed directions, with one of them
The GMR and MTJ sensors are generally called magneto- representing digit “1” and the other representing digit “0.”
electronic or metal-based spintronic devices, which are the Each of the localized areas can be considered as one bit.
focuses of this review. The metal-based spintronic devices Each bit consists of a number of partially exchange-coupled
are based primarily on the spatial modulation of electron magnetic grains, as shown schematically in Figure 2.
spins through using layered structures of magnetic and non- When reproducing the information from the disk, the
magnetic materials. The lack of capability in charge mod- early disk drive employs the same write head as the read
ulation in these types of structures may eventually limit
their ultimate performances in terms of both the magne-
toresistance and other functionalities. To address this issue,
writer
recently, a great deal of effort has been devoted to the devel- MAGNETIC LAYER

opment of magnetic semiconductors which allow the mod- INTERLAYER

ulation of both the spins and charges. The advance in this

reader

MAGNETIC LAYER

field may eventually lead to spintronic devices with perfor- INTERMEDIATE LAYER
shield

shield

mances superior to their metal-based counterparts. In addi- UNDERLAYER

tion to pure metal-based or semiconductor-based spintronic

SEED

devices, hybrid devices making use of both technologies also Al-Mg : NiP or GLASS

have been explored actively in recent years.

The organization of the rest of this chapter is as follows.
Section 2 gives a brief introduction to the basics of magnetic
data storage and spintronics, and explains why both the spin-
tronics and nanotechnology are crucial for magnetic data
storage. Section 3 discusses different types of magnetoresis-
Magnetic
tance effects, and their applications in read sensors. In-depth transition
reviews of the magnetic and electronic aspects of spin valves
are given in Sections 4 and 5, respectively. Section 6 deals Figure 2. Schematic of a magnetic recording head and media. The head
with the MTJs, with their applications in MRAMs discussed typically consists of an inductive writer and an AMR or GMR reader.
Nano Spintronics for Data Storage 3

head, that is, the signal is detected by measuring the voltage latter a current-perpendicular-to-plane, or CPP, sensor [25].
change across the coils induced by the flux variation when So far, CIP is dominant, but CPP is expected to play an
the head passes by the recorded media. Although the per- important role in future terabit recording systems. An alter-
formance of the inductive head had been improved contin- native to the CPP spin valve is the magnetic tunnel junction,
uously to meet the requirement of rapidly increasing areal or MTJ, in which the current also flows perpendicular to
density, its structural design and working principle did pose the plane [26–29]. The major difference between the CPP
certain limitations to the ultimate performance of this type spin valve and the MTJ is that the latter is composed of
of head. That is the reason why, in the beginning of the two ferromagnetic layers separated by an insulator instead
1990s, a new type of head which is based on the anisotropy of a metal. Therefore, the electrical conduction in MTJ is
magnetoresistance of certain magnetic materials was intro- based on quantum-mechanical tunneling. The MTJ is attrac-
duced by IBM, and later also adopted by other head and tive because its MR ratio is generally more than double that
drive companies. This type of head is widely called an AMR of spin valves. However, the primary drawback of MTJ is
head [20]. The AMR effect stems from the spin–orbit inter- that its junction resistance is generally larger than that of
action of electrons, and thus it normally appears in magnetic the CIP or CPP sensors, which may affect its performance
materials; its value, however, varies from less than 1% to a as a read sensor due to the increase in thermal noise and the
few percent, depending on the type of material. The com- decrease in bandwidth when the junction size is reduced to
mon material that is used for an AMR head is Permalloy, a certain value. In addition to this, the quantum-mechanical
of which the MR ratio is about 2–4%. It decreases almost nature of electron transport also causes shot noise which
linearly with thickness for ultrathin films (<10 nm). increases with the intensity of the output signal. Thus, the
Although the AMR head showed much better perfor- major application of MTJ so far is in MRAM. As one of
mance in terms of sensitivity and scalability than the thin- the newcomers in the memory hierarchy, the MRAM has
film inductive head, its pace of structural miniaturization the potential to replace some of the commercially available
and performance improvement was soon outpaced by the memories in the near future.
rapid advance in areal density. The rapid increase of areal Except for the inductive read head, all other read sensors
density requires simultaneously a decrease in both the bit discussed above make use of both the charge and spin prop-
length and the track width. The former requires a thinner erties of electrons. AMR stems from spin–orbit interaction,
and more sensitive sensor, while the latter requires a large while the spin valve and MTJ operate based on the spin-
magnetoresistance (MR). Unfortunately, it is difficult for dependent scattering and tunneling both inside the magnetic
the AMR head to satisfy both requirements simultaneously layers and at the magnet/nonmagnet or insulator interfaces.
because the MR ratio of the AMR head decreases rapidly In contrast to Si-based electronics, the GMR-based sensors
when its thickness is reduced to less than 10 nm [21]. On are often referred to as a subcategory of an emerging field
the other hand, in 1988, a much larger MR effect, that is, called spintronics (see Fig. 1).
GMR, was discovered in artificially made magnetic and non-
magnetic multiple layers [22 23]. The MR ratio of this type
2.2. Spintronics
of superlattice structure is more than one order of mag-
nitude larger than that of the AMR. The GMR structure, Electrons possess both charges and spins. The motion of
however, could not be applied directly to the read head as charges forms the current. The ability to control or mod-
it was because of the large magnetic field that is required ulate the charge transport has made it possible to form
to switch the magnetization of the magnetic layers from one functional devices such as diodes and transistors. This is
direction to the other. This has prompted IBM to invent a so far only possible in semiconductors instead of metals
more practical structure for a read sensor which is called a because the latter have too many electrons per unit volume;
spin valve [24]. the variation of charge distribution, if any, is limited to a
The state-of-the-art spin valve consists of a dozen thin few atomic layers at the surface that can hardly cause any
layers, the heart of which is a trilayer structure consisting of measurable change in the conductance of the metal. There-
two ferromagnetic layers separated by a nonmagnetic spacer, fore, functional electronic devices have not been realized in
which is usually copper. The signal detection principle is pure metals with dimensions larger than the mean-free path
based on the fact that the resistance of the trilayer is depen- of electrons. Although they belong to the same family of
dent on the relative orientation of the magnetization of the metals, metallic magnetic materials have another additional
two ferromagnetic layers. The resistance is high when the degree of freedom which can be used to vary their elec-
magnetizations of the two layers are in opposite directions, tronic transport properties—the spin of electrons. As the
and low when they are in the same direction. To have a lin- spin of electrons in magnetic materials can be easily manip-
ear response from the sensor, the angle between the two ulated using an external magnetic field without suffering the
magnetizations is normally set at 90
at zero field, with one electrostatic screening effect as the charges do when they
of them “pinned” at a direction perpendicular to the media are subjected to an electric field, it is possible to alter the
surface through exchange coupling with an antiferromagnet, conductivity of magnetic materials without changing the car-
and the other free to rotate in response to the fringe field rier distribution inside the material. This forms the basis of
of magnetic transitions recorded on the magnetic media. GMR-based electronics or sometimes is also called magne-
There are two different forms of spin-valve sensors, toelectronics. It is a subfield of spintronics or spinelectronics
depending on whether the current flows in the plane of [1–4].
the stack of layers or perpendicular to them. The former is The spintronics in its broader sense contains all types
called a current-in-plane, or CIP, spin-valve sensor, and the of electronics that make use of both charges and spins.
4 Nano Spintronics for Data Storage

In semiconductor-based electronics, we only control the Year

charge motion, and ignore the spins. In fact, it is diffi- 1000
2001 2004 2005 2006 2007

cult to control the spins in semiconductors because they do 1 Gb/in

100
2
500 Gb/in
2
1 Tb/in
2
10 Tb/in
2

not have specific directions under normal conditions, unless

the electrons are injected into from external sources with

Bit Size (nm)

100 36nm 25nm 8nm
80nm
a net polarization. On the other hand, in metallic mag-
netic materials, it is easy to control the spins due to the
strong ferromagnetic coupling among the spins, but it is dif-
ficult to control the charges. Therefore, a question naturally
arises here: Can we have a kind of material in which we
Proof's Only
10

can control both charges and spins? The answer is “yes,” 1

and this type of material is called a magnetic semiconduc- 0 500 1000 1500 2000
tor [30–32]. The magnetic semiconductor is usually made Areal Density (Gb/in2)
through adding magnetic impurities to host semiconductors.
It is not necessary, however, that every semiconductor can be Figure 3. Bit size as a function of areal density for square bits. Also
made magnetic using this approach because some of them shown is the year at which the areal density will be reached by assuming
still do not exhibit any magnetic properties even after they a compound annual growth rate of 60%. Inset shows the size of square
are doped with a substantial amount of magnetic impurities. bits at areal densities of 100 500 1000 10000 Gbits/in2 , respectively.
Some of them, although being magnetic, show a very low
Curie temperature [33 34]. However, the situation changed
size falls into the nanometer regime (1–100 nm) when the
drastically in recent years due to the intensive efforts made
areal density reaches 70 Gbits/in2 , which was already real-
by researchers in this field in many research organizations.
Several different types of magnetic semiconductors hav- ized about two years ago. To write information to and read
ing a Curie temperature higher than room temperature information from these small bits, the write and read heads
have been found. These include, but are not limited to, must have a size which is smaller than or at least compara-
GaN:Mn [35 36], CdGeP:Mn [37], GaP:Mn [38], TiO2 :Co ble to that of the bit. In addition to the read/write head, the
[39], and ZnO:Co [40]. It should be noted, however, that spacing between the disk and the head must also be kept as
all of these are based on preliminary experimental results; small as possible, and in a practical case, it is about one fifth
further experiments are required to verify the results. The of the bit size. Therefore, once the areal density enters the
progress was made not only in materials themselves, but also hundreds of gigabits range, the dimensions of almost all of
the application of these materials in creating new functional the key components in magnetic recording have to be in the
devices such as semiconductor-based magnetic tunnel junc- nanometer regime.
tions [41 42] and spin-injection devices [43 44]. Although The continuous shrinkage of bit size poses formidable
the current technology for a read sensor is based on metallic challenges to both media and read heads. For low-areal-
spintronics, semiconductor-based spintronics has the poten- density recording, each single bit contains tens of thou-
tial to provide sensors or storage elements with superior sands of small magnetic grains. At this level, one will not
performances for next-generation data storage devices. “feel” the difference when one or two grains are missing.
Although the read sensor for magnetic recording and stor- However, the number of grains per bit decreases monoton-
age cells for magnetic memory are based on spintronics, the ically with the areal density. At 100 Gbits/in2 , the number
storage of information in disk media is still based on clas- of grains is reduced to less than 100. This results in an
sic physics, and it does not involve spintronics. However, as increase of media noise due to the particulate nature of
the bit size continues decreasing, one day the bit size will the media. One of the possible ways to suppress the noise
approach that of atoms. At this stage, a fundamental change is to reduce the grain size of the media so that each bit
in the information storage principle might be required. One will contain a greater number of grains. However, this is
of the possible scenarios is to store information in the recip- not so straightforward because the superparamagnetic phe-
rocal space or energy domain. In this case, the spin of elec- nomenon will appear, making the information bit thermally
trons and nuclei instead of the magnetization of magnetic unstable. Although the recent advance in multiple-layered
grains will be important. This is closely related to another media has greatly enhanced the thermal stability of the
emerging field called quantum information storage [45]. media with small grains, it is still far from being a perfect
solution. The thermal stability of the media is not only just
affected by the grain size, but also very much by the grain
2.3. Nanotechnology and Magnetic
size distribution. The latter is technically more challenging
Data Storage because one must use a mass-production technique which
Nanotechnology is indispensable for all kinds of data storage can produce not just small grains, but also monodispersed
devices, be it magnetic, optical, or solid-state devices. This grains. This is the area where nanotechnology is expected
is because the common driving force for any type of data to play an important role. One of the possible ways to real-
storage technology is the areal density, provided that the bit- ize such a kind of media is to fabricate single-domain par-
cost performance also improves with the areal density. As a ticles or dots by using nanofabrication techniques [46–48].
rule of thumb, the bit size shrinks by one order when the Although one of the major applications of nanostructured
areal density increases by two orders. Figure 3 shows the magnetic materials and nanomagnetism is in storage media,
size of a square bit as a function of the areal density. The bit we will not discuss this further in this chapter because, so far,
Nano Spintronics for Data Storage 5

the media are still in the “passive” form, and do not involve 3.2. AMR Sensor
spintronics.
The immediate application of the AMR was in magnetic
Now, we turn to the read sensor, which is our main focus
recording because sensors based on the AMR effect offer
in this chapter. We are concerned about the sensor when
higher output as compared to the thin-film inductive head
its size is reduced because, first, the sensor size must be
[53 54]. Although, for Ni (Fe, Co) alloys, the largest AMR
reduced accordingly when the bit size decreases, and sec-
effect so far was found for Ni70 Co30 , 26.7% at 4.2 K and
ond, the output and impedance of the sensor are dependent
6.6% at 300 K [52], the material of choice for magnetic
on the sensor size. So far, the sensor fabrication has lever-
recording applications is permalloy because of its softness,
aged the semiconductor manufacturing processes. But this
high permeability, and low magnetorestriction. The typical
trend is going to be reversed because the size shrinkage of
AMR ratio for thin permalloy films (30–50 nm) is about 2%,
read sensors has outpaced the shrinkage of critical dimen-
although the AMR of its bulk counterpart can be as high as
sions of integrated circuits [49]. In a few years’ time, the
dimension of the sensor will approach 50 nm, at which point 4% [13 21]. When being used as a sensor, the magnetization
there may not even be a solution for the semiconductor direction is normally set at 45
with respect to the current
industry at the moment. In addition to the manufacturing direction at zero field so as to maximize the sensitivity [55].
issues, one also needs to find a way to improve the perfor- This is apparent from the fact that the first derivative of 
mance of the sensors so that the output of the sensor will is maximum when  = 45
.
not decrease significantly when its size is reduced. Figure 4 shows a schematic drawing of an AMR sen-
The above discussion demonstrates clearly the importance sor element. For the sake of simplicity, we assume that the
of nanotechnology and spintronics in magnetic recording external magnetic field points in the y-axis direction. Assum-
and memory devices. In fact, spintronics is closely related to ing that the entire element is a single-domain particle with
nanotechnology because most of the phenomena related to a uniaxial anisotropy, the magnetization direction can be
bulk and surface magnetism have a characteristic length in obtained from minimization of the energy density [13]:
the nanometer regime. In the rest of this chapter, we discuss 1
in detail the fundamental concepts and latest development = −HMs sin  + Ku sin2  + Hd Ms sin2  (3)
of metallic spintronics and its applications in magnetic data 2
storage. Note that here only the magnetostatic energy, anisotropy
energy, and demagnetizing energy are taken into account; all
other energy terms are neglected. In Eq. (3), H is the exter-
3. MAGNETORESISTANCE AND THE nal magnetic field, Ms is the saturation magnetization, Ku
ASSOCIATED READ SENSORS is the uniaxial anisotropy energy constant, Hd is the demag-
netizing field, and  is the angle between the magnetization
3.1. Anisotropic Magnetoresistance and easy axis direction. Minimizing the energy density gives
The resistivity of a nonmagnetic material is usually inde-
pendent of the magnetic field. However, this may not be H H
sin  = = (4)
the case for magnetic materials. The phenomenon that the 2Ku /Ms + Hd Hk + H d
resistivity of a ferromagnetic material depends on the rela-
tive angle between the current and magnetization direction where Hk = 2Ku /Ms is the anisotropy field. Substituting
of the material is called the anisotropic magnetoresistance Eq. (4) into Eq. (2) yields
effect, or AMR, which was discovered in 1857 by Thomson
2
[50]. A series of theoretical studies carried out a century H
H  =  −  − ⊥   (5)
later showed that the AMR effect stems from the spin–orbit Hk + H d
interaction [51]. The importance of the AMR effect was
recognized in the 1970s when a large AMR was found in For soft materials, Hk  Hd ; therefore,
a number of alloys based on iron, cobalt, and nickel [52].
2
H
Materials exhibiting a normal AMR effect show a maximum H  =  −  − ⊥   (6)
resistivity when the current is parallel to the magnetization Hd
direction ( ) and a minimum resistivity when the current is
It is apparent from the above equation that the simple AMR
perpendicular to the magnetization direction (⊥ ). A mea-
sensor element exhibits a nonlinear response to the external
sure for the size of this effect is the AMR, ratio which is
field, which cannot be used as a read sensor for magnetic
defined by
recording as it is. However, it is not difficult to realize from
 − ⊥

Proof's Only
MR =  (1)

y
M
At intermediate angles between the current and magnetiza- θ H
D I
tion direction, the resistivity of an AMR material is given by
easy axis
x
T
 = ⊥ +  − ⊥  cos2  (2) W

where  is the angle between the current and the magneti- Figure 4. Schematic of a single-domain AMR sensor element. H is the
zation direction. external field, M is the saturation magnetization, and I is the current.
6 Nano Spintronics for Data Storage

Eq. (6) that the sensor can be made linear if an additional down in thickness so as to meet the requirement of shrink-
field which is much larger than that of the external field age in bit length and reduction in Mr t of the media [21].
is added to it to make the total effective external field as These intrinsic characteristics of the AMR plus the tremen-
H  = H + HB with HB H. In the case of Hd H , which dous progress made in GMR in the early 1990s have deter-
is true for magnetic recording, Eq. (6) becomes mined the short lifetime of the AMR sensor in the history of

 hard disk drives. It was gradually replaced by the spin-valve
HB 2 2HHB sensor, which was first introduced into disk drives by IBM
H ≈  −  − ⊥  −  − ⊥   (7)
Hd Hd2 in 1997.
It shows that now the sensor responds linearly to the exter-
nal field. HB is the so-called bias field or, more precisely, 3.3. Giant Magnetoresistance
the traverse bias field because it is perpendicular to the easy In the second half of the 1970s, researchers at IBM devel-
axis direction of the sensor element. In actual sensor design, oped a technique which allows growing ultrathin films with
the strength of the bias field is chosen such that the mag- extraordinary accuracy in thickness control. It was based
netization direction at zero field is about 45
away from the on the vacuum evaporation technique, but with a several
easy axis so as to maximize the sensitivity. It is obvious from orders of magnitude lower base pressure as compared to the
Eq. (7) that the smaller the demagnetizing field, the larger normal high-vacuum evaporator. At this base pressure and
the sensitivity. appropriate partial pressures of the source materials, it is
There are many different ways to form a traverse bias. possible to have a mean-free path of the evaporated materi-
Among them, the most successful was the soft adjacent layer als that is larger than the distance between the source mate-
(SAL) bias scheme, in which a soft ferromagnetic layer is rial and the substrate for most of the molecules or atoms.
laminated with the sensing layer via a thin insulating spacer, In other words, the evaporated molecules or atoms form
as shown schematically in Figure 5 [55–58]. As the SAL is beams, and impinge directly on the substrate surface to ini-
normally chosen such that most of the current flows through tiate the growth. Therefore, the technique is generally called
the sensing layer, the magnetic field induced by the sensing molecular beam epitaxy (MBE) [62]. The word “expitaxy”
current magnetizes and saturates the SAL into one direc- was used because MBE had been mainly employed to fab-
tion (pointing upward in Fig. 5). The fringe field thus gen- ricate semiconductor materials which are normally single-
erated, in turn, provides a traverse bias to the sensing layer crystalline materials grown epitaxially on the substrates. The
itself. The SAL scheme offers several advantages, such as most attractive point of MBE is that it not only allows the
adjustable bias field, relatively uniform bias field distribu- growth of ultrathin single-layer film, but also makes it pos-
tion, and reduced demagnetizing field. Although it also has sible to grow layered structures of different materials. This
drawbacks such as the current shunting effect, it so far has opened an important field in materials science called arti-
remained the most successful engineering design. In actual ficial structures. In particular, when the layers are a few
sensors, in addition to the traverse bias, one also needs a atomic layers thick, they form a so-called superlattice struc-
longitudinal bias to stabilize the domain structure so as to ture [63]. It also made it possible to create quantum-well
reduce the Baukhausen noise caused by the domain-wall structures in which the electrons of one type of material
motion [55 56]. are confined by the potential steps formed between this
As we mentioned above, the AMR sensor, intrinsically, is material and the other types of materials at both sides.
not a linear sensor. In addition to the nonlinearity issue, it This led to tremendous advances in semiconductor devices.
also suffers drawbacks such as thermal asperity [59] and side Being inspired by the work carried out by the semiconductor
reading asymmetry [60 61]. Perhaps the most fatal short- community, researchers in the surface-science community
coming of the AMR sensor is that it is difficult to scale it started to use the same technique to study surface mag-
netism and magnetic interactions across an ultrathin anti-
ferromagnetic or nonmagnetic spacer [64]. These research
_ activities led to the discovery of giant magnetoresistance in
_
_ Fe/Cr superlattices [22]. As shown in Figure 6, the resis-
_
_
_
D tance of such a superlattice structure is high at zero field,
_
_ decreases when a magnetic field is applied in both direc-

Proof's Only _
_ tions along the sample surface, and finally saturates at a
field Hs of about 2 T. The MR ratio of superlattices with an
Ins ulato r

[Fe(3 nm)/Cr(0.9 nm)] × 60 structure was measured to be

MR
Insulator

θ
MR

M about 45% at 4.2 K, which is much larger than the AMR

SA L
SAL

effect [22]. It was proved soon afterward that the resistance

Hbias W
bias
of the superlattice structure depends on the relative orien-
I
tation of the magnetization of the adjacent magnetic layers.
d At a Cr thickness of 0.9 nm, the magnetizations of adjacent
Fe layers are antiferromagnetically coupled at zero field, but
Figure 5. Schematic of an AMR sensor using an SAL bias. The sensor are realigned into the same direction when the external field
consists of an MR element as the active layer and a soft adjacent layer is sufficiently high to overcome the exchange-coupling field.
(SAL) for traverse bias. The MR element and the SAL are separated The mechanism of GMR can be understood using the
by an insulating spacer. simple two-current model [65–67, 284, 285]. In this model,
Nano Spintronics for Data Storage 7

R/R (H=0) on the material and structural parameters of the multilayers.

1-
More details can be found in other review articles [76 77].
For a magnetic/nonmagnetic multilayer with the film plane
(Fe 30 Å/Cr 18 Å)30
parallel to the xy plane, the linearized Boltzmann equation
becomes [68–70]
Hs

Proof's Only
0.8
g ↑↓ z v g ↑↓ z v eE fo v
+ = (9)
0.7
z vz  ↑↓ mvz vx
(Fe 30 Å/Cr 12 Å)35

Hs
where
0.6
(Fe 30 Å/Cr 9 Å)60 g ↑↓ z v = f ↑↓ z v − f0 v
0.5 Hs
is the deviation of the electron distribution function from
- 40 - 30 - 20 - 10 0 10 20 30 40 the equilibrium Fermi–Dirac distribution f0 v. Here,  ↑↓
Magnetic field (kG)
represents the relaxation times for spin-up (spin-down) elec-
trons, e is the elementary charge of electrons, m is the elec-
Figure 6. Magnetoresistance of Fe/Cr superlattices. Both the current
and the applied field are along the same [110] axis in the film plane.
tron mass, v is the velocity, and E is the electrical field. The
Reprinted with permission from [22], M. N. Baibich et al., Phys. Rev. general solution of Eq. (9) is
Lett. 61, 2472 (1988). © 1988, American Institute of Physics. ↑↓
g± z v


eE ↑↓ f0 v ↑↓ ∓z
the electrical conduction of ferromagnetic materials is con- = 1 + F± v exp (10)
sidered carried out by two independent channels of spin-up m vx  ↑↓ vz 
and spin-down electrons. Due to the different density of ↑↓
states distribution of spin-up and spin-down electrons near where +− is for vz > 0 vz < 0. F± v can be obtained
↑↓
the Fermi level, the mean-free path for spin-up electrons by using the boundary conditions for g± z v at both the
↑↓
is normally larger than that of spin-down electrons due to interfaces and top and bottom surfaces. After g± z v is
the larger scattering probability of s electrons to spin-down obtained, the total current per unit length in the y direction
d-electron states (when the density of states of spin-down (assuming the field along the x direction) can be obtained
↑↓
electrons is higher than that of the density of states of spin- by first integrating −evx g± z v in v space, followed by
up electrons). When all of the magnetic layers are ferro- integrating with respect to z. The MR ratio can thus be
magnetically exchange coupled, the spin-up electrons will obtained from the difference in conductivities between the
experience less scattering when they cross the nonmagnetic parallel and antiparallel alignment of the magnetic layers.
layer entering other magnetic layers within the spin-diffusion
length, which is normally larger than the mean-free path. 3.4. Interlayer Coupling
The resistivity of the spin-up electrons in this case can be
considered as a constant over the film stack, and is denoted The underlying mechanism of GMR is the long-range mag-
by ↑↑ . Similarly, the resistivity of the spin-down electrons netic exchange coupling between transition metals across
can also be considered as a constant, and is denoted by ↑↓ . a nonmagnetic spacer, which was first observed in the
As in the simple two-current model, we ignore the spin-flip Fe/Cr/Fe(001) system [78]. Although the GMR effect was
scattering, that is, no mixing of the two conduction chan- originally observed in Fe/Cr superlattices with a fixed
nels; the total resistivity is given by P = ↑↑ ↑↓ /↑↑ + ↑↓ . Cr layer thickness, subsequent studies have revealed the
However, the situation changes when the magnetic layers oscillatory behavior of interlayer coupling in this system
are antiferromagnetically coupled. This is because both the and many others [79–81]. More importantly, these results
spin-up and spin-down electrons will experience frequent were obtained not only from high-crystalline-quality sam-
scattering when they cross the multiple layers. The two con- ples grown by MBE, but also from polycrystalline sam-
duction channels will have the same resistivity, which is given ples deposited by sputtering. This has greatly accelerated
by ↑↑ + ↑↓ /2, leading to a total resistivity of AP = ↑↑ + the research in this field because of the wide availabil-
↑↓ /4. The MR ratio is thus given by ity of sputtering systems. As one of the typical results,
Figure 7 shows the oscillatory behavior of magnetoresis-

 tance of Co/Cu superlattices observed by Parkin et al. on dc
P − AP ↑↑ − ↑↓ 2
MR = =−  (8) magnetron-sputtered samples [79]. Four well-defined oscil-
AP ↑↑ + ↑↓ lations with a period of about 1 nm are seen at both 4.2 and
300 K. This system later became the most important system
The negative sign indicates that the resistivity at saturation for metal-based spintronic devices.
state is lower than that at the zero-field state. A quantita- It was soon realized that the oscillation in GMR is a
tive treatment of the GMR effect is possible using either general phenomenon in ultrathin magnetic and nonmag-
the Boltzmann transport equation [68–70] or the quantum netic multiple layers, and its origin comes from the peri-
Kubo formula [71–73] using either the simple parabolic odical switching from ferromagnetic to antiferromagnetic
band structure or the more realistic band structures [74 75]. alignment and vice versa of the adjacent ferromagnetic lay-
These models can predict very well the dependence of GMR ers when the thickness of the nonmagnetic layer varies every
8 Nano Spintronics for Data Storage

40
300K is the susceptibility of the host materials. The integrations

∆R/R (%)
30 Si/Fe(45Å)/[Co(10Å)/Cu(tCu)]N
over q and k  are to be performed within the first Brillouin
20
10

zone, and G is a vector such that k  + q + G belongs to

Proof's Only
0 the first Brillouin zone. In the case of free electrons at zero
60 temperature, the Fermi surface is a spherical surface with a
∆R/R (%)

4.2K
40 radius defined by the Fermi wavevector k  . The probabil-
F
20 ity distribution function f  k  drops to zero sharply at the
0 Fermi surface, resulting in a logarithmic singularity at q =
0 10 20 30 40 50 80 120 160  F for .q.
tCu (Å)
2k  The Fourier transforms of .q  give oscilla-
tions to J with a period of 3 = 4F /2, and 4F is the Fermi
Figure 7. Dependence of saturation traverse magnetoresistance on Cu wavelength.
spacer layer thickness for a family of related superlattice structures of For interactions between two ferromagnetic layers (here-
the form Si/Fe(4.5 nm)/[Co(1 nm)/Cu(tCu ]N . An additional Cu layer after, we refer them to as FM1 and FM2 , the interaction
was deposited on each film structure such that the uppermost Cu layer integral may be obtained by summing over all of the pairs
was about 5.5 nm thick. The number of bilayers in the superlattice N of spins belonging to FM1 and FM2 . For the sake of simplic-
is 16 for tCu below 5.5 nm (solid and empty circles) and 8 for tCu above ity, we assume that both layers are one atomic layer thick.
5.5 nm (solid and empty circles). Adapted with permission from [80], Without the loss of generality, we first look at the interac-
S. S. P. Parkin et al., Phys. Rev. Lett. 66, 2152 (1991). © 1991, American
tion between a single spin located at position O of FM1 and
Institute of Physics.
all of the spins belonging to FM2 , which is given by

few atomic layers [82]. It was found that both the GMR and d 2 
I1O 2 = S J R O − R j  (14)
the saturation magnetic field of Fe/Cr multiple layers oscil- V0 j∈F M2
late at a period of about 1.5 nm. The latter suggests that
the relative angle of the magnetization direction of the two The above summation can be further simplified by treating
ferromagnetic layers oscillates between 0 and 180
because FM2 as a thin layer with a uniform spin distribution, that is,
it is easier to saturate the two layers when they are in par-
allel alignment, and difficult when they are in antiparallel  d 
alignment. This was soon confirmed by the neutron [83] and → d 2 R   (15)
F2
V0 F 2
Brillouin scattering experiments [84]. Furthermore, detailed
investigations have revealed the existence of two coupling
periods [85–90]. Here, R  is the in-plane component of R O − R j . The inter-
Various models have been proposed to account for the action integral is then given by
oscillatory behavior of exchange coupling of the mag-
2
netic/nonmagnetic multiple layers, which include primarily A m d2
I1O2 z ≈ S2 sin2kF z for z →∝ (16)
the quantum-well model [91–97] and the Ruderman–Kittel– V0 16,
2 z2
Kasuya–Yoshida (RKKY) interaction model [98–101]. Both
models have been able to describe the origin of the exchange which is also an oscillation function with a period of 3 =
coupling, in particular, the oscillations in the sign and 4F /2. This is too short to explain the long oscillation peri-
strength of the coupling. In what follows, we explain briefly ods observed experimentally. Subsequent studies pointed
the RKKY model by closely following the treatment by out that the contradiction comes from the invalid assump-
Bruno and Chappert [100 101] and Fert and Bruno [102]. tion that the thin FM2 layer has a continuous and uniform
The interaction between two ferromagnetic layers across spin distribution. By taking into account the discreteness of
a metallic spacer has its basis in the interaction between the spin distribution in the FM2 plane, the interaction inte-
two magnetic impurities in a host metallic material. Accord- gral is given by
ing to the RKKY model, the effective exchange interaction

between two spins Si and Sj is given by 1 A 2 2 d  +∝
I1O2 z = − S dqz
2 V0 2,3 −∝
Hij = J R i − R j Si · Sj (11) 
× expiqz z d 2 q .q  qz 
where the exchange integral is 
expiq · R   (17)
J R i − R j  R  ∈FM2


1 A 2 V0  3 The last term suggests that q cannot take any arbitrary
=−  q
d q.  exp/iq · R i − R j 0 (12)
2 V0 2,3 value but the in-plane reciprocal lattice vector G   ; other-
wise, its value will be zero. This gives an effective wavevector
Here, V0 is the atomic volume and
/2kF 2 − G2 01/2 , which is responsible for the multiperiodic-
V0  3  f  k  − f  k+
 q+ 
ity due to different G in different directions. Figure 8 shows
 G
 = 212B
.q d k (13) the cross section of the Fermi surface for an fcc (001) spacer.
2, 3 k+ q+
 G − k The first Brillouin zone (FBZ) and the auxiliary prismatic
Nano Spintronics for Data Storage 9

(111) (113) λ = 1.28

(a) J(L) —
d

(000) (002)

8 9 10 11 12 13 14 15

Proof's Only
L/d

(111) (111) (113) Figure 9. Full curve: exchange-coupling strength of two ferromagnetic
z layers across a monovalent fcc (100) metal (arbitrary units) calculated
using the continuum version of the RKKY model. Broken curve: the
(111) (113)
(b) actual coupling strength with the experimentally measured periodicity
for L = Nd. Here, L is the thickness of the spacer, d is the atomic
layer spacing in the thickness direction, and N is an integer number.
Cu Adapted with permission from [99], R. Coehoorn, Phys. Rev. B 44, 9331
(1991). © 1991, American Institute of Physics.
(000) (002)

to reduce the exchange coupling between the ferromagnetic

layers through increasing the thickness of the spacer. How-
(111) (111) (113) ever, a completely decoupled or weakly coupled multilayer
structure does not function properly as a sensor because
Figure 8. Cross section of the Fermi surface for an fcc (001) metal and of the possible inconsistent movement of all of the mag-
the corresponding vectors giving rise to oscillatory interlayer coupling netic layers under an external field. One of the possible
in both the (001) and (111) directions (indicated by the horizontal and ways to overcome the drawback of the decoupled GMR
oblique arrows, respectively). (a) Free-electron model. (b) Cu Fermi stack is to use two decoupled layers with one of them much
surface. Adapted with permission from [100], P. Bruno and C. Chappert, softer than the other, so that the former will respond to
Phys. Rev. Lett. 67, 2592 (1991). © 1991, American Institute of Physics. a small field, while the latter will only change its magnetic
state when it is subjected to a large field [105 106]. This
kind of structure, in principle, can form a good sensor, pro-
zone are represented by the solid and dashed contours. The
vided that the hard layer is not only truly hard, but also
wavevectors giving rise to oscillatory interlayer coupling are
has the same crystalline structure as that of the soft layer.
indicated by the horizontal bold arrows: (a) free-electron
Unfortunately, this kind of material is still not available at
approximation, and (b) Cu Fermi surfaces. Also shown are
present. Therefore, most of the experiments on decoupled
the vectors giving rise to oscillations in the (111) direction
bilayers have been performed on NiFe/Cu/Co or Co/Cu/Co
(oblique arrows).
systems. In the latter case, the two Co layers may have dif-
On the other hand, the long oscillation period can also
ferent coercivities, and their values are dependent on the
be understood based on the argument that the spacer thick-
individual layers, thicknesses. Although a relatively large
ness does not vary continuously in Eq. (16); rather, it only
magnetoresistance with a small saturation field has been
takes discrete values z = N + 1a with N an integer number
obtained in these kinds of structures, they were still unsuit-
[99 103 104]. Because of this discrete sampling or aliasing
able for data storage applications due to the instability of
effect, one obtains an effective period
the hard layer. However, this type of structure has served
, as a good object for studying spin-dependent transport in
3= (18) magnetic/nonmagnetic multilayers [107 108]. They can also
kF − n,/a
be used for MRAMs [109]. The magnetic stiffness of the
where n is an integer chosen such that 3 > 2a, with a hard layer may be improved by using an antiferromagnet-
being the atomic spacing of the spacer in the thickness direc- ically coupled hard/Ru/soft triple-layer structure to replace
tion. This gives a period which is much larger than 4F /2 the single-layer structure [110 111].
(see Fig. 9). To address the issues involving noncoupled magnetic mul-
tilayers, IBM invented a device called the spin valve in 1991
[24 112 113]. A typical spin valve in its simplest form con-
3.5. Spin Valve sists of two ferromagnetic layers separated by a nonmagnetic
The research on read sensors using the GMR effect com- spacer and an antiferromagnetic (AFM) layer in contact
menced soon after the GMR effect was reported. This is with one of the ferromagnetic layers (see Fig. 10). The thick-
because it gives a much higher MR ratio as compared to ness of the spacer is chosen such that there is little exchange
its AMR counterpart. However, the original GMR struc- coupling between the two FM layers. The magnetization of
ture consisting of Fe/Cr, Co/Cu, or other types of magnetic one of the FM layers which is in direct contact with the
multiple layers could not be applied to read sensors in the AFM layer is “pinned” by the latter, and thus this FM layer
original form because of their large saturation fields, and is commonly called a pinned layer. On the other hand, the
also the nonlinear response near the zero-field point. The magnetization of the other FM layer is free to rotate to
most straightforward way to reduce the saturation field is respond to an external field, and thus it is called a free layer.
10 Nano Spintronics for Data Storage

Pinning Pinned Spacer Free M

Layer Layer Layer PL
Hc2
(a) PL
Hc1 (b)

Hex MR
M FL FL
Hc1
M PL H
FL
Hc2

AFM hard Cu soft H

Hin

Figure 10. Schematic of a simple spin-valve structure.

Figure 12. Schematics of M–H (a) and MR–H (b) curves of a typi-
cal spin valve (Hex : exchange-bias field; Hin : interlayer coupling field
The typical material for the FM layer is NiFe or CoFe, while between free and pinned layer; (Hc1 FL FL
− Hc2 ): coercivity of free layer;
that for the spacer is generally copper. A thin layer of Co PL PL
(Hc1 − Hc2 ): coercivity of pinned layer).
or CoFe with high Co composition is often added at the FM
and Cu interface so as to increase the MR ratio due to its
high polarization ratio and less interdiffusion at the inter- the interlayer coupling can either be ferromagnetic or anti-
face with Cu [114 115]. The choice of AFM is an issue of ferromagnetic, depending on whether the orange peel or
high complexity due to many practical considerations, which RKKY coupling dominates. We will come back to this topic
will be discussed shortly. shortly. For the particular spin valve whose characteristics
When being used as a read sensor [116], a constant cur- are shown schematically in Figure 12, the free and pinned
rent is applied to the sensor, and its voltage change caused layers are ferromagnetically coupled. Therefore, the magne-
by the fringe-field variation from the media is detected as tization of the spin valve is at maximum, while the resistance
the readout signal. In order to increase the linear density, is at minimum at zero field. When the applied field increases
usually two shielding layers are used to “shut off” the influ- in the opposite direction of the pinning field (in this case, in
ence of neighboring bits (see Fig. 11). Depending on how the positive direction), the magnetization of the free layer
the current is applied, there are two different types of sen- was first reversed, resulting in an antiparallel alignment of
sors, that is, the current-in-plane and current-perpendicular- the free and pinned layers. This corresponds to the high-
to-plane sensors, as discussed earlier. As the name suggests, resistance state. When the applied is further increased, it
in the former case, the current flows in the plane of the sen- will eventually reverse the magnetization of the pinned layer.
sor stack (see Fig. 11), while in the latter case, the current This will realign the magnetization of the pinned layer with
flows perpendicular to the plane. The small resistance of that of the free layer, recovering the low-resistance state.
the metallic multilayers makes it quite challenging to mea- Decreasing of the applied field after the saturation in the
sure the magnetoresistance of the CPP structure. Several positive direction will cause the magnetization of the pinned
approaches have been taken to address this issue, includ- layer to be reversed first, followed by the reversal of the free
ing the use of superconducting electrodes and reduction layer, and finally, the two magnetic layers realign with each
of the lateral size using lithography [117 118] or selective other again at negative fields. In general, a large exchange
growth in the pores of nanoporous membranes [119–121]. bias field and a small interlayer-coupling field are desirable
The results obtained showed that the CPP MR ratio is for read-head applications.
much larger than the CIP MR ratio. Alternatively, the GMR The performance of a read sensor is normally charac-
can also be measured with the current-at-an-angle (CAP) terized by the size of the MR effect. It can be either the
through growing the GMR structure on grooved substrates MR ratio or the sheet resistance change :RS . The former
[122]. Since, so far, CIP sensors dominate in practical appli- is given by [24, 123–125]
cations, we discuss the CIP sensors first.


:RS :RS 1 − cosF − P 
The M–H and MR–H curves of a typical spin valve are = (19)
shown schematically in Figure 12. Due to the small thick- RS RS 0 2
ness of the spacer, there always exists an interlayer cou-
where RS is the sheet resistance and :RS the change in sheet
pling between the free and pinned layers. The nature of
resistance. :RS /RS 0 is the maximum MR ratio when the
magnetizations of free and pinned layers are switched from
parallel alignment to antiparallel alignment. F and P are
the angles between the parallel direction and the magneti-
I
zation direction of the free and pinned layers, respectively.
I
It is apparent from this equation that a linear response

Proof's Only
can be obtained through setting P to 90
. This is because
Shield

Shield

sin F is proportional to the external field, as we discussed

se
ns

in Section 3.2 for AMR sensors. In addition to its linear

or

response, the spin valve also exhibits a much higher MR

ratio as compared to the AMR sensors. However, even
though that is the case, there has been and there is still a
strong demand for the improvement of the spin-valve sen-
Figure 11. Spin valve used as a read sensor in CIP mode. sors. Figure 13 shows the estimated track width and required
Nano Spintronics for Data Storage 11

0.14 35 (d) (e)

Aspect ratio = 4
0.12 30

Required sensor sensitivity (mV/µm)

AFM PL Cu FL AFM PL1 Ru PL2 Cu FL

HL Cu SL
(f) (g)
0.10 25 (a)
Read track width (µm)

0.08

0.06
Read track width
= 0.8 × track pitch
20

15
Proof's Only
AFM PL1 Ru PL2 NOL PL2 Cu FL Cu NOL
(h)
HL Ru SL Cu FL I-HL PL Cu

Improvement of magnetic properties

(i)
FL

(b)

0.04 10 AFM PL Cu FL HCL AFM PL Cu FL HR

(j) (k)

0.02 Read track width 5

= 0.6 × track pitch
I-AFM PL Cu FL HR AFM PL NOL PL Cu FL NOL
(c)
0 0 Improvement of electronic properties
0 200 400 600 800 1000
Areal density (Gbits/in2)
Figure 14. Evolution of spin valves. (a) Pseudospin valve. (b) Advanced
single spin valve. (c) Advanced dual spin valve. (b) and (c) are the
Figure 13. Estimated read track width and required sensor sensitivities results of continuous improvements in both the magnetic (d)–(g) and
at different areal densities by assuming an aspect ratio of 4:1. The solid electronic (h)–(k) designs of the pseudospin valves. (d) Original spin-
and dashed curves are obtained by assuming the read track width to valve invented by IBM. (e) Synthetic spin valve. (f) Synthetic spin
be 80 and 60% of the track pitch, respectively. The sensitivity require- valve using a hard magnet. (g) Exchange spring spin valve using an
ment is obtained by assuming that the sensor would generate an output insulating hard magnet. (h) Spin-filter spin valve using a back layer
voltage of 1 mV. or a high-conductance layer. (i) Specular spin valve. (j) Specular spin
valve using an insulation AFM. (k) Specular spin valve using nano-
oxides. The acronyms used are: HL—hard layer, SL—soft layer, PL—
sensor sensitivity at an aspect ratio of 4:1 for different areal pinned layer, FL—free layer, PL1 (2)—first (second) portion of the
densities by assuming that the sensor has to generate an pinned layer, I-HL—insulating hard layer, HCL—high-conductance
output voltage of 1 mV. The solid and dashed curves are layer, HR—high-specularity reflective layer, NOL—nanooxide layer,
obtained by assuming a read track width of 80 and 60% AFM—antiferromagnetic layer, I-AFM—insulating antiferromagnetic
of the track pitch, respectively. From this figure, one can layer.
see that a threefold increase of the sensor sensitivity is
required for increasing the areal density from 100 Gbits/in2 eliminate Baukhausen noise induced by domain wall move-
to 1 Tbits/in2 . ment, and (4) to make the sensor thermally stable at ele-
From Eq. (19), one can see that the MR ratio of a spin vated temperature under normal operation conditions and
valve is affected by two factors of different origin. The corrosion resistive. Of course, in practical cases, one also
prefactor of the right-hand-side term of Eq. (19), that is, has to consider the ease of manufacturing and reliability of
:RS /RS 0 , is mainly determined by the spin-dependent the devices. To achieve all of these goals, one needs to set a
electronic transport properties of the spin valve, while the proper traverse and longitudinal bias for the spin valve using
rest is more or less determined by its magnetic properties. appropriate materials. The former is employed to “pin” the
As summarized in Figure 14, the improvement in spin-valve magnetization of the pinned layer as rigid as possible, while
design has proceeded from both the magnetic and electronic the latter’s role is to maintain the single-domain status of
aspects. In a certain sense, the original spin valve proposed the free layer, and at the same time, without sacrificing its
by IBM can be regarded as the first step in the improve- softness.
ment of the pseudospin valve that still remains the simplest,
yet the most attractive design should the materials problem 4.1. Traverse Bias
be solved in the future. In what follows, we first discuss the
magnetic aspect of spin valves, followed by a discussion on 4.1.1. Exchange Bias
the electronic aspect. We will refer back to Figure 14 again Traverse bias is achieved through the use of exchange bias
when necessary. (EB) formed at FM/AFM interfaces [Fig. 10 and Fig. 14(d)],
which was discovered about half a century ago [126]. The
exchange bias at the FM/AFM interface is usually estab-
4. MAGNETIC ASPECTS lished through a magnetic annealing and cooling process, in
OF SPIN VALVES which the sample is first heated up to above the Néel tem-
perature of the AFM, and then is cooled down to room tem-
Like a transistor which needs a proper bias to work properly, perature in the presence of a magnetic field with a strength
the spin valve also needs a proper magnetic bias so as to of one to a few tesla. The exchange-biased FM/AFM bilayer
achieve a linear response with minimized asymmetry, if any is characterized by a shift of the hysteresis loop in the oppo-
[13 15 55 123]. The main tasks of magnetic design of spin site direction of the exchange-bias field direction and an
valves are: (1) to set the magnetization of the pinned layer as increase of the coercivity of the FM layer, as shown schemat-
rigid as possible in the vertical direction (see Fig. 10), (2) to ically in Figure 15(a).
make the free layer as soft as possible, and to align its initial The exchange bias in AFM/FM bilayers is a very com-
magnetization direction parallel to the media surface, (3) to plex issue. Despite intensive studies, the phenomenon itself
12 Nano Spintronics for Data Storage

M
450
400
Malozemoff argued that any realistic model should not be
FM
350 HE based on the assumption of an atomically perfect uncom-
300
pensated AFM surface, and proposed a random interface

HE, HC (Oe)
AFM/FM
250
model [133]. It was shown that, for a real interface with

Proof's Only
200

H
150
Hc
finite roughness, it becomes energetically favorable for the
100
HE
HC
50
antiferromagnet to break up into domains with perpendic-
0 ular domain walls. This model yields an exchange-coupling
0 2 4 6 8 10
IrMn thickness (nm) field of
(a) (b) 
2z AAF KAF
Figure 15. (a) Schematic hysteresis curves of a single FM layer (dotted HEX = (22)
, 2 MFM tFM
line) and an FM/AFM exchange-coupled bilayer (solid line). (b) Depen-
dence of exchange-bias field on the thickness of AFM material for an
where z is the number of order unity. Therefore, the
NiFe (3 nm)/IrMn bilayer system.
exchange field predicted by both the AFM domain wall
model and the random interface model is of the same order;
is still not fully understood [127–130]. A number of theoret- this is because both are determined by the characteristic
ical models have been proposed to account for the experi- domain wall energy.
mental data, which include, but are not limited to: (1) the The parallel domain wall model requires the existence of
coherent rotation model [131], (2) the Néel model [132], a spiral spin structure in the AFM layer [140], which has
(3) the random interface model, (4) the AFM domain wall been observed recently in both the conventional FM/AFM
model [133 134], (5) the interface spin canting model [135], bilayers [141] and artificial FM/AFM bilayers [142]. The for-
(6) the random and compensated interface model, and so mation of an AFM domain wall is found to be affected by
on [136–139]. Obviously, the coherent rotation model is the the magnetization of the FM layer. The exchange coupling
simplest one which assumes that: (1) both the FM and AFM between the FM and AFM layers occurs on a domain-by-
layers are single-crystalline materials, (2) the AFM layer domain basis [143]. Due to the dominant role of the AFM
has a rigid spin configuration which is not affected by any domain walls in exchange coupling, the exchange bias is gen-
strength of an external field, and its interfacial layer is fully erally sensitive to the defects in the AFM layer.
uncompensated, and (3) the magnetization reversal of the
FM layer is accomplished through coherent magnetization
4.1.2. AFM Materials
rotation. Under these assumptions, the exchange field is
given by Above is a far too brief introduction of the exchange-
coupling mechanism at the FM/AFM interfaces. An in-depth
2J SFM · SAFM discussion of this topic can be found in several excellent
HEX = (20) review papers [127–130]. In what follows, we turn to differ-
a2 MFM tFM
ent material systems that have been developed for practical
applications in spin valves [144 145]. As discussed earlier,
where J is the exchange parameter, SFM SAFM  is the spin the key to the proper operation of a spin valve is to “pin”
of the interfacial atom of the FM (AFM) layer, MFM tFM  the magnetization of the reference layer in a direction that
is the magnetic moment (thickness) of the FM layer, and is perpendicular to the media surface, which forms a 90

a is the cubic lattice constant. Although the above equa- angle with respect to the easy axis of the free layer at zero
tion does agree with the experimental observation that HEX field. The former is achieved using the exchange bias at the
decreases with tFM , it yields a value for HEX that is by orders AF/AFM interface. To realize a reliable and linear operation
of magnitude larger than that of the experimental data. The of the spin valve, the following issues have to be considered
simple model fails to explain the asymmetry observed in the carefully when choosing the AFM materials.
hysteresis either. This has prompted the proposal of various
improved models. Among them, two models have been quite 1. Size of the exchange bias field: As a subtle rotation of
successful in explaining the observed value of HEX . Mauri the magnetization of the pinned layer will cause asym-
et al. proposed a model that is able to explain the observed metry and degradation of the output signal, the pin-
exchange bias without the need to remove the condition of ning field must be sufficiently high (>300 Oe) so as
strong interfacial coupling at the FM/AFM interface [134]. to make the reference layer “rigid” under a moderate
This is made possible by allowing the formation of parallel external field.
domain walls, either in the AFM layer or in the FM layer, 2. Thermal stability: The exchange-bias field usually
wherever the energy is lower. By assuming that the domain decreases with temperature. In what way and how fast
wall forms at the AFM side of the interface, they obtained it decreases, however, depends not only on the intrin-
the following expression for the exchange field: sic properties of the materials, but also their structural
 properties, thickness, and techniques that are used to
2 AAF KAF prepare the materials. The temperature at which the
HEX = (21)
MFM tFM exchange-bias field vanishes is called the blocking tem-
perature. A high blocking temperature (>250
C) is
where AAF KAF  is the exchange stiffness (magnetocrys- required to ensure that the sensor functions properly
talline anisotropy energy constant) of the AFM layer. not only at room temperature, but also at elevated
Nano Spintronics for Data Storage 13

temperatures. This is important because the temper- the annealing temperature is too high. Therefore, a
ature of the active region of the sensor may rise to low-temperature process is preferred.
60–120
C during normal operation. The AFM materials investigated so far include: (1) Mn-
3. Large resistivity of the AFM layer: As the AFM layer based alloys, (2) Cr-based alloys, and (3) Fe, Co, or Ni-
itself does not contribute to the MR effect, a high- based oxides [144 145]. Among them, Mn-based alloys are
resistivity material is desirable for suppressing the the most widely studied, and several of them have already
current-shunting effect. In this sense, the ideal AFM been applied to practical spin-valve sensors. The Mn-based
material should be an insulator. Unfortunately, most alloys can be roughly divided into two groups. One group
of the oxide-based AFM materials like NiO can hardly includes FeMn, IrMn, RhMn, RuMn, and the other includes
satisfy other requirements, and thus are unsuitable as NiMn, PtMn, PdMn, and some of the ternary alloys of these
AFM materials for practical device applications. elements. The crystalline structure of the first group is fcc,
4. Small critical thickness: The exchange-bias field is while that of the second group is fct (CuAu–I). FeMn/NiFe
found to be quite insensitive to the thickness of the (CoFe) is the most widely studied system which exhibits an
AFM when it is sufficiently thick. However, it starts to exchange bias of about 420 Oe for a 4 nm thick NiFe without
drop rapidly when the thickness decreases to a certain thermal annealing. However, it has a low blocking temper-
value, which is largely determined by the properties of ature of about 150
C and poor corrosion resistance, and
the AFM material. As an example, Figure 15(b) shows thus is unsuitable for read sensor applications. Another sys-
the results for an Ni80 Fe20 (3 nm)/IrMn bilayer system. tem that does not require thermal annealing is IrMn/CoFe
The onset thickness for this process is called the critical (or NiFe). It has a better corrosion resistance and smaller
thickness. Considering the temperature dependence of critical thickness as compared to FeMn. Both are desir-
the exchange field, the thickness of the AFM must be able for sensor applications. However, its moderate block-
chosen well beyond the critical thickness in practical ing temperature of 250
C limits its applications in a real
devices. However, a large AFM thickness will result in drive environment. Although, in general, a thermal anneal-
a large total thickness of the sensor, which is undesir- ing process (at 200–300
C) is required to establish the AFM
able for sensors for high-areal-density hard disk drives. phase, the (Ni, Pt, Pd) Mn family gives a much higher block-
Therefore, AFM materials with smaller critical thick- ing temperature (350–450
C), better corrosion resistance,
ness are desirable for spin-valve applications. and a higher exchange-bias field (500–800 Oe), and thus is
5. Good corrosion resistance: AFM materials with being used in real devices. The primary drawback of this
poor corrosion resistance are unsuitable for practical group of materials is their large critical thickness, which may
applications. limit their applications in ultrahigh-density recording appli-
6. Low-temperature process: The thermal annealing pro- cations. Table 1 lists the properties of some widely studied
cess may affect the performance of spin valves if AFM materials [144–160].

Table 1. Properties of widely studied AFM materials.

Hex (Oe) Required Resistivity Corrosion

Materials (NiFe 4 nma ) thickness (nm) TB 
C) (1< · cm) resistance Annealing
b b c e
FeMn 420 7 150 , 165 , 150, Poor Nob
175f , 130b
170j
IrMn 200 7d , 190/230d , 200, Gooda k Nok
7.5e , ∼250e f , 325d Moderatee
10l 130l
NiMn >390 >25b , 450  >425d ,
b
175b , Moderate Requiredb
35d , 430g , 210d Goode k
30e 400f  j
PtMn 500 30 380p , ∼200 Good Required
400f ,
310j
PdPtMn 280 25d m 300m  350d f 185d Good Required
CrMnPt 220 30a o , 380 , a o
310–345a , Good No
35d 300d 300–350o
360d
NiO 200 35b , 200b h i  190e Insulator, Good Nob
30e , ≥108b Excellente
50h
A–Fe2 O3 40–75 100 200–250, Insulator Good No
250q ,
280–300r
References: a [144], b [146], c [147], d [145], e [148], f [149], g [150], h [151], i [152], j [153], k [154], l [155], m
[156], o [157], p [158], q [159], r [160].
14 Nano Spintronics for Data Storage

4.1.3. Synthetic AFM 4.1.4. Hard/Ru/Soft Synthetic Ferrimagnet

As discussed earlier, the exchange-bias field is inversely pro- The use of AFM as the pinning layer has two drawbacks:
portional to the thickness of the FM layer. For practical (1) all of the useful AFMs reported so far for spin valves
applications, however, one needs a certain thickness for the are metals which contribute negatively to the MR due to
pinned layer so as to optimize the electrical properties of the current-shunting effect, and (2) most AFMs have a
the sensor. This affects the sensor in two aspects. One is moderate or low blocking temperature, and some of them
the reduction in the exchange-bias field, and the other is the also show poor corrosion resistance. Based on this back-
magnetostatic coupling with the free layer. Although one ground, there have been several works in which hard/Ru/soft
can use the sense current field to counterbalance the mag- synthetic ferromagnets have been used as the pinning and
netostatic field, it may sacrifice the stability of the pinned pinned layers [see Fig. 14(f)] [110 111 174]. Preliminary
layer. A novel structure which can solve both problems is the experimental results showed that these types of spin valves
synthetic ferrimagnet which consists of two antiferromagnet- have a good thermal stability [110 111]. However, it would
ically coupled FM layers via a thin Ru layer [see Fig. 16(a)]. be quite challenging to improve other properties of the spin
The corresponding synthetic antiferromagnet in which one valves due to either the mismatch of the crystalline structure
of the FM layers of the synthetic ferrimagnet is exchange of the hard and soft layer or the insufficient hardness of the
coupled to an AFM layer can be used as the pinning and hard layer.
pinned layers for spin valves [161–171].
The partial cancellation of the magnetic moments from 4.1.5. Exchange Spring
both layers gives a smaller net moment, and thus an Further simplification of the synthetic ferrimagnet would be
increased exchange-bias field and a reduced magnetostatic- the hard/soft exchange spring structure [Fig. 14(g)]. This
coupling field. In addition to these advantages, it also type of structure was proposed recently by Carey et al.
provides a better thermal stability, although the blocking using cobalt ferrite as the pinning layer [175]. A high MR
temperature itself remains the same. Shown in Figure 16(b) ratio, 12.8%, and a high pinning field, 1500 Oe, have been
is a schematic drawing of a typical spin valve using the syn- obtained without compromising the softness of the free
thetic AFM as the pinned layer [also see Fig. 14(e)]. As layer. In this particular case, the pinning layer is an insula-
the exchange bias is inversely proportional to the magnetic tor which may also help to reduce the shunting current, and
moment thickness product of the pinned layer, compared thus is desirable for obtaining a high MR ratio.
to parallel alignment, the exchange-bias field in antiparallel
alignment is enhanced by a factor of [21]
4.2. Longitudinal Bias
M1s t1 + M2s t2 The traverse bias is to set a proper direction for the mag-
fP −AP = (23)
M1s t1 − M2s t2 netization of the pinned layer. For a spin valve to function
properly, one also needs to set a proper longitudinal bias
where M1s M2s  and t1 t2  are the magnetic moment and for the free layer. This is to both obtain a good linearity
thickness of the first (second) pinned layer, respectively. On with less asymmetry and to suppress the Baukhausen noise.
the other hand, the demagnetizing field is proportional to The former is achieved through setting the magnetization
the magnetic moment thickness product of the pinned layer of the free layer at 90
with that of the pinned layer, par-
itself. Therefore, the stability of the pinned layer is effec- allel to the media surface. This, in turn, can be achieved
tively increased by a factor of through first inducing an easy axis in an appropriate direc-
tion during deposition of the free layer, and then using the
shape anisotropy to stabilize it in the same direction. How-
/M1s t1 + M2s t2 /M1s t1 − M2s t2 02  ever, the shape anisotropy alone may become insufficient as
the aspect ratio of the sensor decreases. The following addi-
Although it is not shown in Figure 16, the synthetic ferrimag- tional magnetic fields need to be taken into account when
net can also be used as the free layer, which can effectively designing the sensor: (1) the fringe field from the pinned
reduce the magnetic thickness of the free layer without sacri- layer, (2) the RKKY and orange-peel coupling field from the
ficing its physical thickness [172]. However, the longitudinal pinned layer, or so-called interlayer coupling field, (3) the
bias may affect the stability of the synthetic layer [173]. current-induced field from highly conductive layers such as
the spacer layer, and (4) other fields such as imaging fields
from different sources [123]. Figure 17 shows schematically
the flux lines of various field components exerted on the free
layer. It is worth noting that the direction and magnitude of

Proof's Only FM1

Ru
FM2
the current-induced fields are adjustable through controlling
the sensing current, whereas the various field components
from the pinned layer are fixed once the sensor fabrica-
AFM PL1 Ru PL2 Cu FL tion is completed. Therefore, in practice, it is of importance
(a) (b) to ensure that magnetic fields from sources other than the
sense current be able to be compensated later by the cur-
Figure 16. Schematic of a synthetic ferrimagnet (a) and spin valve using rent field under normal operation conditions. Ideally, all of
the synthetic ferrimagnet as a pinned layer (b). the external fields other than the signal field from the media
Nano Spintronics for Data Storage 15

Current Fringe
field of P L lead lead lead lead
field AFM AFM
AFM AFM AFM AFM
FM FM
Cu Cu
+ FM FM
+
+ Image 2 AFM AFM
Image 1 +
+ (a
(a) (b
(b)

NM lead
AFM
AFM FM
Shied 1 Shied 2 lead FM lead insulator Cu insulator
Cu FM
FM AFM
AFM lead

Proof's Only
Field from image 1 Field from image 2 (c
(c) (d
(d)

Figure 17. Different kinds of magnetic fields exerted on the free layer. lead lead lead lead
For a spin valve to function properly, it is of importance to balance out
all of the components so as to make the net field as close to zero as FM FM
possible. Here, PL (FL) stands for pinned layer (free layer). PM Cu PM PM Cu
FM
PM
FM
AFM AFM
should be reduced to zero so as to obtain a high sensitiv- (e
(e) (f)
(
ity, good linearity, and null asymmetry for the read sensor.
lead
However, this also means that the sensor is too susceptible lead lead
FM
to external disturbances. This will induce noises or base- PM PM Cu
FM insulator
line popping and a shift in the readout signal, in particular, FM
Cu
insulator
AFM
the domain formation and movement-induced Baukhausen FM PM
spacer
noise. The latter is an issue of high complexity because it AFM lead
depends on many factors, such as the material and shape (g
(g) (h
(h)

of the free layer and the process to form it and the effect
of other layers. Therefore, as in the case of AMR sen- Figure 18. Schematics of different types of longitudinal bias schemes.
(a) Patterned exchange. (b) Lead-overlaid patterned exchange.
sors, a longitudinal bias of an appropriate strength is nor-
(c) Long-distance exchange. (d) Same as (c), but in CPP mode.
mally used to suppress the multidomain formation in the (e) Contiguous junction with PM. (f) Lead-overlaid contiguous junc-
free layer of spin-valve sensors [55 116 176 177 178 179]. tion with PM. (g) Patterned synthetic ferrimagnet. (h) Parallel PM.
Most of the longitudinal biasing techniques for spin-valve The acronyms used are: PM—permanent magnet, FM—ferromagnet,
sensors stems from the earlier work on AMR sensors, and AFM—antiferromagnet, NM—nonmagnetic layer. In (a)–(d), the top
may be divided into two groups. The first group is based on and bottom AFMs are two different types of materials.
exchange bias between a ferromagnet and an antiferromag-
net [see Fig. 18(a)–(d)], and the second group is based on
the magnetostatic interaction or exchange coupling between of the permanent magnet also affects the performance of the
a ferromagnetic soft film and a permanent or hard magnet sensor. The shape and cleanness of the junction are impor-
[see Fig. 18(e)–(h)]. tant for forming low-contact-resistance junctions, which is
the key to suppressing the temperature rise of the reader
4.2.1. Permanent Magnet-Based Bias [195]. To avoid oxidation and contamination at the junc-
Contiguous Junction So far, the most widely studied tion interface, the permanent magnetic and contact are nor-
bias scheme which is also employed in real disk drives is the mally formed using a self-aligned lift-off process using a
contiguous (or abutted) junction using a permanent mag- “mushroom” resist pattern formed on top of the sensor ele-
net (typically, a CoCrPt alloy), as shown schematically in ment [178 179 184]. The CoCrPt layer is typically grown
Figure 18(e) [180–182]. The key to forming a proper bias on a Cr seed layer to have a higher coercivity, typically
in this scheme lies not only in the selection of a proper 1500–2500 Oe [196]. The uniform coverage of the seed layer
material with an appropriate thickness, but also in the con- on the junction edge is crucial to obtain a high-coercivity
trol of the junction shape between the permanent magnet film along the entire junction. Poor seed-layer coverage may
and the active element of the sensor [183–188]. The over- result in instability in the MR curve of the read sensor [197].
hanging of the permanent magnet on top of the free layer
should be minimized; otherwise, it will cause hysteresis or Lead-Overlaid Patterned Permanent Magnet The con-
kinks in the transfer curves, resulting in instability in the tiguous junction longitudinal bias is by far the most widely
readback signal [189–191]. The ratio between the Mr t of the used bias scheme in real products. The drawback of this
permanent magnet and the Ms t of the free layer or the soft scheme, however, is that the bias field usually is not uni-
layer should be optimized so as to obtain a good tradeoff form across the longitudinal direction of the sensor. It is
between sensitivity and stability. The typical value is about normally stronger at the two edges and weaker at the center.
1.1–1.4 or even higher, depending on the shield-to-shield dis- If the center portion is properly biased, then it is unavoid-
tance [123, 186, 192–194]. A small shield-to-shield distance able that the edge regions will be overbiased, leading to the
requires a stronger permanent magnet. The microstructure formation of so-called dead regions. These inactive regions
16 Nano Spintronics for Data Storage

will, in general, degrade the sensitivity of the senor. The 4.2.2. Antiferromagnet-Based Bias
influence of the dead region becomes more prominent when Patterned Exchange As with the case of permanent
the read track narrows, in particular when the sensor width magnet-based longitudinal bias design for spin valves, the
decreases to below 0.3 1m [198 199]. An alternative scheme exchange-biased designs also stemmed from the works
which can suppress the effect of the inactive region is the on MR sensors [178 179 205 206]. Among the exchange-
so-called lead overlaid structure in which the contact elec- domain stabilization designs, the most widely investigated is
trodes are extended over the abutted junction, and thus form the patterned exchange design in which a patterned layer of
a direct electrical contact with the inactive region of the antiferromagnetic material is fabricated on top of the free
sensor [Fig. 18(f)] [200]. Zhang et al. carried out a compara- layer before the electrical lead is deposited [see Fig. 18(a)]
tive study of magnetic noise in read heads with a contiguous [199 207 208 209]. The advantage of this structure over the
junction and a lead overlaid design [201]. It was found that, overlaid abutted junction structure is that the dead region
at 0.25 1m magnetic read width and 4 mA bias current, the is almost removed from the edge. In addition to this, the
magnetic noise is twice as large as Johnson noise for the lead reduction in parasitic resistance associated with the junction
overlaid design, while it is comparable to Johnson noise for improves the signal-to-noise ratio. However, in this case, one
the contiguous junction design. The higher magnetic noise may need to use additional process steps to set the pinning
is attributed to a weaker longitudinal bias field with the lead direction of the patterned AFM layer because the exchange-
overlaid design. This may affect the performance of read bias direction of longitudinal bias for the free layer is 90

heads with a lead overlaid design at higher areal densities. away from that of the traverse bias for the pinned layer (see
Fig. 19). Another potential disadvantage of this design is
Patterned Synthetic Ferrimagnet Recently, the author’s the side reading due to the portions of the sensors under-
group proposed an alternative bias design in which pat- neath the patterned AFM layers. This might be suppressed
terned magnets are placed on top of the two edges of the by increasing the exchange-bias strength and the conductiv-
free layer via an ultrathin Ru film such that the permanent ity of the electrical contacts. It might also be reduced using
magnets are antiferromagnetically coupled to the free layer the lead overlaid structure shown in Figure 18(b).
[Fig. 18(g)] [202]. In conventional overlaid permanent mag-
net design or contiguous junction design with overlapping Long-Distance Exchange The most straightforward way
areas, the domain structure near the edges of the free layer to stabilize the domains of the sensing layer of MR and
is not so stable because the magnetic field generated by the GMR heads is to add an AFM layer on top of the sensing
permanent magnet inside the free layer always changes sign layer. In fact, this has been employed in the early design of
at the edges. In the new design, however, the antiferromag- an MR head [177]. This design, however, lacks the flexibil-
netic coupling between the permanent magnet and the free ity in precise control of the bias field. It was often found
layer forms a flux closure surrounding the permanent mag- that the exchange bias is too strong, which stiffens the
net, and thus improves the domain stability near the edges. magnetization of the sensing layer from rotation into the
Micromagnetic modeling has shown that, in addition to the traverse direction, resulting in a low sensitivity. However,
domain stability, the sensitivity of the sensor with the new recently, there has been a revived interest in this design
design is also higher than that of the contiguous junction owing to the recent development of our understanding of
design. the exchange bias and the ability to control the bias field
through the insertion of an ultrathin nonmagnetic spacer at
the AFM/FM interface [210–213]. Using Cu as the spacer,
Permanent-Magnet-Based Design for CPP and MTJ Nakashio et al. and Mao et al. successfully applied this
Sensors The contiguous and lead overlaid designs are not design to a CIP spin-valve head and an MTJ head, respec-
suitable for CPP and MTJ sensors unless insulating per- tively [Fig. 18(c) and (d)] [210 211]. This type of design is
manent magnets are used to form the longitudinal bias or particularly suitable for a CPP head.
an additional insulating layer is added between the perma-
nent magnet and the electrode at two edges. Redon et al.
proposed a modified version of contiguous junction or pat-
terned permanent design for an MTJ head through remov- electrode electrode

ing the portion of the pinned layer that is overlapped with AFM AFM

the permanent magnet [203]. The results of micromagnetic Sensor element Longitudinal
modeling showed that the modified design is able to improve

Proof's Only
bias AFM
(a)
the performance of the MTJ head through suppressing the
influence of the inactive region underneath the permanent
magnet. Sin et al. proposed a design in which the longi-
tudinal bias is formed through the magnetostatic coupling
from a permanent magnet placed at the other side of the
pinned layer, as shown schematically in Figure 18(h) [204]. AFM PL1 Ru PL2 Cu FL AFM
(b)
This design is suitable for CPP and MTJ sensors because
there are no conducting materials at the two sides of the Figure 19. Cross-sectional view (a) and three-dimensional schematic
sensor element. structure (b) of spin valves with patterned-exchange longitudinal bias.
Nano Spintronics for Data Storage 17

5. ELECTRONIC ASPECTS spin-filter design [216]. It is worth noting that, in addition

OF SPIN VALVES to the enhancement of the MFP of majority electrons, the
back layer or HCL also helps to better balance the current-
Analogous to a transistor, the traverse and longitudinal induced fields at the free layer due to the redistribution of
biases discussed above are to set a proper working point for a certain portion of current to the HCL, leading to a lower
the spin valves. As is the case with transistors, once a proper asymmetry of the readout signal. For example, at a free
working point is chosen, the performance of a spin valve layer thickness of 4 nm, Ueno et al. showed that the read-
is largely determined by its intrinsic properties, that is, the out asymmetry can be reduced from 10% for a conventional
magnitude of the MR caused by the spin-dependent scat- spin valve to 3% for a spin-filter spin valve [216]. However,
tering at both the interfaces and the ferromagnetic layers. the improvement of the MR in a spin-filter design is only
Although the spin valve exhibits a much higher MR as com- up to a certain extent because the HCL also increases the
pared to the AMR sensor, the ever-increasing areal density shunting current, which in turn will reduce the MR. There-
requires continuous improvement of the performance of the fore, in most practical cases, the back layer is employed in
spin valve. To this end, a variety of spin valves have been combination with other techniques such as the nanooxide to
proposed and studied both theoretically and experimentally. improve the performance of advanced spin valves.
These include, but are not limited to: (1) a multiple spin
valve (e.g., dual spin valve), (2) a spin-filter spin valve,
(3) a specular reflection spin valve, (4) a nanooxide-added 5.2. Dual Spin Valve
spin valve, and (5) a nanooxide-added dual spin valve. In The most straightforward way to improve the MR ratio
what follows, we explain briefly the motivation and under- of a spin valve is to increase the number of free/pinned
lying mechanism for each of these different types of spin layer pairs. This has prompted the proposal and develop-
valves illustrated schematically in Figure 14. ment of dual spin valves, in which the free layer is shared
by two pinned layers at both sides (see Figs. 14(c) and
5.1. Spin Filter Spin Valve 20) [217–223]. The MR of a typical dual spin valve can be
30–60% higher than that of a single spin valve, depend-
The magnetoresistance of a CIP spin valve can be thought ing on the materials, device structure, and fabrication pro-
of as originating from the difference in the mean-free path cesses [21 217 218 224]. The fabrication of a dual spin
(MFP) of the majority and minority electrons, at least phe- valve requires the use of AFM materials which are able to
nomenologically. By making use of this phenomenon, Gur- form exchange bias both at the top and bottom of the FM
ney et al. derived a novel method to measure the MFP of layer, and with a small critical thickness. Special cautions
electrons in different materials such as Cu, AuCu, Co, NiFe, are needed in choosing the thickness of the two Cu spac-
and Fe [214]. In this technique, a back layer is added to ers because two identical spacers may make it difficult to
a spin valve with a thin free layer, and its effect on :G adjust the traverse bias field in the free layer using the sense
(=GP − GAP ) is studied systematically at different back-layer current once the sensor fabrication is completed [223].
thicknesses. Here, GP and GAP are the conductances of the
spin valve in parallel and antiparallel configurations, respec-
tively. It was found that, to a good approximation, :G can
5.3. Specular Spin Valve
be written as The dual spin valve consists of one free layer and two pinned
layers. Can we increase the MR further by simply increasing
:G = :Gf + :GB /1 − exp−tB /4+ 0 (24) the number of free and pinned layers? The answer is nega-
tive because different free layers may respond to the exter-
where :Gf is the conductance change due to the free layer, nal field in a slightly different way, which may result in extra
tB is the thickness of the back layer excluding the inactive noises in the readout signal or degradation of the sensitivity.
layer, if any, at the interface with the free layer, 4+ is the The only way to overcome this is to use antiferromagneti-
MFP of majority electrons in the back layer, and :GB is the cally coupled ferromagnetic/nonferromagnetic pairs. But as
conductance change due to the back layer when tB 4+ . discussed earlier, this type of sensor is only applicable to the
The MFPs measured by this technique for Cu, Co, NiFe, and detection of a large magnetic field. One of the possible ways
Fe are 192 55 46, and 1.5 nm, respectively. This result to increase the effective number of free layers without the
implies that the addition of a back layer with an appropri- need to add more layers physically is to increase the spec-
ate thickness can enhance the MR of a spin valve with an ular reflectivity of electrons at the outer surface of the free
ultrathin free layer (so-called back-layer effect). The idea
behind this is that the back layer helps to retain or even
enhance the MFP of majority electrons, while it has little

Proof's Only
effect on the MFP of minority electrons. Kamiguchi et al.
extended this concept to synthetic spin valves, and called
them spin-filter spin valves [see Fig. 14(h)] [215]. The back
layer is retermed a high-conductance layer (HCL). Note that
the synthetic pinned layer is not shown in the schematic
drawing in Figure 14(h) because we want to focus on the AFM PL Cu FL Cu PL AFM
pure electronic effect. At a free layer thickness of 2–4 nm,
the MR ratio can be enhanced by about 10% using the Figure 20. Schematic of a dual spin valve.
18 Nano Spintronics for Data Storage

and pinned layer, respectively. As shown schematically in (thicknesses are in units of nanometers) was only about
Figure 21, if the specular reflectivity at both surfaces is unity, 3.7%. Sugita et al. [227] and Kawawake et al. [228] grew
the trilayer can be considered effectively as the repetition A–Fe2 O3 50/Co2/Cu2/Co5/Cu0.4 (thicknesses are in units
of the same unit in an infinite number of cycles. However, of nanometers) spin valves on a polished (110) plane
the difference between the equivalent structure and the real of an A–Al2 O3 single-crystalline substrate. The MR ratio
structure of the same type is that the former only has one obtained was as high as 18%. It was further increased to
free layer physically, and thus does not suffer from the draw- 27.8% for a dual spin valve with a structure A–Al2 O3 /A–
backs caused by the existence of a plural number of free Fe2 O3 50/Co2/Cu2/Co5/Cu2/Co2/A–Fe2 O3 50 (thicknesses are
layers. in units of nanometers). In either case, however, the unidi-
rectional exchange-bias field is very small. The pinned layer
5.3.1. Specular Spin Valve Using is more or less a hard magnet instead of a soft layer pinned
Insulating Antiferromagnet by an AFM. The coercivity of the bottom Co layer is more
than 1200 Oe. The large MR ratio again was attributed to
The most straightforward way to implement this kind the enhanced specular reflection at A–Fe2 O3 /Co interfaces.
of structure is to use insulating AFM layers such In addition to oxides, metals can also be used as the
as NiO and A–Fe2 O3 to form an AFM/active layers/ capping layer to increase the MR of spin valves. Although
capping layer, AFM/active layers/AFM or AFM/active lay- the mechanism of MR enhancement is not fully under-
ers/NM/AFM sandwich structure [Fig. 14(j)]. This has stood, it is believed that the potential step at both the free
been verified experimentally to be very effective in layer/capping layer interface and the capping layer/air inter-
increasing the MR ratio [219, 225–229]. The specular face may have helped to enhance the specular reflection of
reflection enhancement effect was first described quan- electrons. Compared to oxide capping layers, the metallic
titatively by Swagten et al. in spin valves with struc- capping layer has certain limitations in thickness because too
ture NiO(50 nm)/FM1/NM1/FM2/NM2/NiO(10 nm), with thick a capping layer will cause an increase in the shunt-
FM1 = Co 2 nm NM1 = Cu 2 nm FM2 = Co (t), ing current, and thus leads to a decrease in the MR ratio
NM2 = Cu 12 nm) [225]. The second nonmagnetic layer, [230 231]. Although it has not been investigated in detail,
in this case, a 1.2 nm thick Cu, is used to isolate the this type of structure might be useful for CPP spin valves.
exchange coupling between the top NiO and the second
ferromagnetic layer (FM2), so that the top NiO func-
tions only as a pure insulator. The decoupling effect of
5.3.2. Specular Spin Valve Using
NM2 has been confirmed in the measured M–H curves, Insulating Ferromagnet
in which FM2 was found to switch at almost zero field. Recently, Carey et al. [175] and Maat et al. [232] used insu-
Using this structure, Swagten et al. obtained an MR ratio lating cobalt ferrite as pinning layers to improve the per-
as high as 25% at 10 K and 15% at room tempera- formance of spin valves. Spin valves with a high MR ratio
ture [225]. A semiclassical model has been invoked to (12.8%) and a large pinning field (about 1500 Oe) have
interpret the experimental results, which suggests that the been successfully fabricated. In this case, pinning is realized
enhancement of the MR is attributed to the enhance- through an exchange-spring type of coupling between two
ment of specular reflection at the metal/insulator interfaces. ferromagnets [see Fig. 14(g)].
Also using the same NiO oxide, but in a dual spin valve
with a structure NiO50/Co2.5/Cu1.8/Co4/Cu1.8/Co2.5/NiO 5.3.3. Nanooxide Added Spin Valve
50 (thicknesses are in units of nanometers), Egelhoff et al. As discussed above, the use of oxide AFM can greatly
obtained an MR ratio of 21.5% at room temperature [219]. enhance the MR of spin valves. These oxides, however,
The large MR ratio was attributed to both the dual spin- are unsuitable for practical applications due to their low
valve structure and the enhanced specular scattering of exchange bias field, low blocking temperature (for NiO), or
electrons at the Co/NiO interfaces. Although NiO has a large thickness required. This has led to the proposal of
desirable electric property, its low exchange bias field and inserting nanooxides at the middle of the pinned layer of
low blocking temperature make it an unfavorable choice bottom spin valves. This, in combination with the top oxide
for real read sensors. Hasegawa et al. carried out the capping layer, was found to greatly enhance the MR of spin
first study on the possibility of using A–Fe2 O3 , another valves. Since the pioneer work of Kamiguchi et al. [215],
insulating AFM, to construct specular spin valves [226]. the nanooxide effect has been studied intensively to enhance
The MR ratio that has been obtained for a spin valve both the magnetic and electronic properties of advanced
with structure glass/A–Fe2 O3 /Ni80 Fe20 6/Cu2.2/Ni80 Fe20 9/Cu4 spin valves [233–264]. The nanooxide, in general, is inserted
at the middle of the pinned layer, and in the case of syn-
FL Equivalent to
thetic spin valves, it is normally inserted at the middle of the
pinned layer that is nearer to the spacer. It is more effec-

Proof's Only
PL
Cu tive when being used together with an insulating capping
Oxide
Repetition of
layer, leading to the so-called double specular spin valve
the basic unit [see Fig. 14(k)]. The nanooxide can be formed using differ-
ent techniques such as natural oxidation, plasma oxidation,
Figure 21. The number of the free layer (FL) and pinned layer (PL) remote plasma oxidation or so-called atomic beam oxida-
can be increased effectively by increasing the specular reflectivity of tion, ion beam oxidation, and ion-assisted oxidation. The
both the top and bottom surfaces to unity. plasma- and ion-based techniques were found to be able to
Nano Spintronics for Data Storage 19

form denser oxides than natural oxidation. Under optimum IrMn6 /CoFe1.5 /NOL/CoFe1.5 /Cu(t) /CoFe1/NiFe2 /CoFe1/
conditions, the oxides improve the MR without compromis- OX/Cu1/Ta1 [hereafter, we call them nanooxide-added SVs
ing the magnetic and thermal stability and the ESD robust- (NOL-SV)] were deposited on Si(100) substrates coated
ness of spin valves. Read heads with sensitivity in the range with a 1 1m thick thermally oxidized SiO2 layer by using
of 6–10 mV/1m or higher have been demonstrated using an ultrahigh vacuum sputtering system operating at a base
nanooxide-added spin valves [257]. As the results obtained pressure of 5 × 10−10 torr. Here, NOL denotes the natural
from different works are, to a certain extent, dependent on oxidation of the CoFe layer. Oxidation was conducted in a
specific experimental conditions, the results are not always separate chamber, by exposing the fresh CoFe surface to
consistent with one another. Table 2 summarizes the repre- pure oxygen atmosphere. The exposure time was 1 min,
sentative works on specular spin valves reported so far in the but the pressure was controlled from 1 × 10−4 to 1.6 Pa.
last few years. Due to the large number of publications, the The specimens were magnetically annealed at 275
C for
author cannot guarantee that the collection is complete. In 1 h under magnetic fields of 1 T in a commercial magnetic
what follows, we use our own results as examples to describe vacuum annealing oven under a pressure of 10 × 10−6 torr.
the main effect of nanooxide. The MR was measured by using a linear four-point probe
The Role of Oxide in Pinned Layers The roles of the method. The sheet resistance was calibrated by measuring
bottom oxides are twofold. First, they reduce the surface the MR on a microstructural bridge with 400 1m length
roughness of the subsequently deposited layers, and sec- and 100 1m width.
ond, they enhance the specularity of the pinned layer. The Figure 22(a) shows the dependence of the MR ratio
improvement of the surface roughness has been confirmed and interlayer coupling field on the oxygen exposure dose
through both direct measurement of the surface roughness used to oxidize the CoFe layer. It is found that Hint
using AFM and measurement of the interlayer coupling decreases almost monotonically with the oxygen exposure
field between the free and pinned layers [249]. In general, dose, decreasing from 31 Oe in a CSV to 4 Oe in an NOL–
there are two major coupling fields between the free and SV. On the other hand, the MR ratio increases from 9 to
pinned layers of different origin: (1) orange-peel coupling, 14% with an increasing O2 exposure dose from 10−8 to
and (2) RKKY coupling. The former is dominant in sam- 0.1 min Pa, which saturates when the exposure dose is over
ples with rough interfaces, while the latter becomes more 0.1 min Pa. Shown in Figure 22(b) are the sheet resistance
important when the interfaces become atomically flat. The Rsq and its change :Rsq as a function of O2 exposure dose.
orange-peel coupling decays exponentially with the thickness Rsq decreases initially, reaches a minimum at an oxygen
of the spacer, and for a sinusoidal roughness profile, it is exposure dose of about 10−3 min Pa, and increases when
given by [265] the oxygen dose increases. The initial decrease of the sheet
resistance might be attributed to the enhancement of specu-
, 2 A2 √
Hint_OP = √ MF exp−2, 2ts /4 (25) lar scattering of electrons at the interface between the metal
2tF 4 and NOL.
Figure 23 shows the MR (a), interlayer coupling field (b),
where A and 4 are the amplitude and wavelength of the sheet resistance (c), and change of sheet resistance (d) as
roughness profile, tF and ts are the thicknesses of the free a function of Cu thickness. The enhancement of surface
and spacer layer, respectively, and MF is the magnetization smoothness can be seen from the oscillation of the inter-
of the free layer. The original concept was extended to thin layer coupling field with the Cu layer thickness in NOL–
films with a structure of hard/spacer/soft by Kools et al. in SVs, as shown in Figure 23(b) (solid circle). Also shown
1999 [266]. Although Eq. (25) is modified slightly, the expo- in the figure are the experimental results for conventional
nential dependence of the coupling field on the spacer thick- spin valves (open circle) and the fitting curve according to
ness is still the same. Eq. (25) (dotted line). Although the interlayer coupling is
On the other hand, as discussed before, the strength of still ferromagnetic in the entire range of the Cu thickness
RKKY coupling oscillates with spacer thickness: that has been investigated, the clear oscillation in NOL–

 SVs and the monotonic decrease with Cu thickness in CSVs
1 1 2, ts
Hint_RKKY ∝ sin +B (26) clearly demonstrate the effectiveness of NOL in improv-
tF ts2 3
ing the surface roughness. Figure 23(c) and (d) shows the
where 3 is the oscillation period and B is the phase factor. dependences of the sheet resistance and its changes on
Except for a few reports [244 257], the oscillation can hardly the Cu thickness, respectively. It is interesting to note that
be seen in normal spin-valve samples without nanooxides. the nanooxide-added spin valve has a smaller Rsq , but a
However, it can be observed easily in samples with nano- larger :Rsq than the conventional spin valve in the entire
oxides. The suppression of orange-peel coupling has also thickness range for Cu, leading to a larger MR ratio in the
been achieved in Cu/Co superlattices through the introduc- whole range, as shown in Figure 23(a). It is also interesting
tion of a proper amount of oxygen during the growth of the to note that the maximum of MR corresponds to the mini-
entire film stack [233 267]. mum of the interlayer coupling field. This can be understood
To confirm both effects of nanooxide, spin valves with readily because, at this thickness range, the free and pinned
the structure Ta3/NiFe2/IrMn6/CoFe3/Cu(t)/CoFe1/NiFe2/ layers are antiferromagnetically coupled if one subtracts the
CoFe1/Cu1/Ta1 [t denotes the thickness of the spacer layer; orange-peel coupling field from the total coupling field. Very
thicknesses are in nanometers, hereafter, we call them con- recently, we have been able to achieve antiferromagnetic
ventional SVs (CSV)] and SVs with the structure Ta3/NiFe2/ coupling routinely in spin valves with oxide capping layers.
20 Nano Spintronics for Data Storage

Table 2. Summary of the findings reported in nanooxide-related works.

Structure of spin valve

Research group (thicknesses are in nm) GMR (%) Hin (Oe) Hex (Oe) Remarks Ref.

1. NIST, NiO 50/Co 2.5/Cu 1.8/Co 4/ 21.5 First experimental [219]

Univ. Minnesota, Cu 1.8/Co 2.5/NiO 50 indication of
Univ. California specular reflectivity
enhancement
NiO 50/Co 2.5/Cu 1.9/Co 4/ 24.8 Oscillating between Systematic study of the [233]
Cu 1.9/Co 2.5/NiO 50 80 and −20 Oe role of background gas
(CH4 , N2 , H2 O, C, H2 ,
CO, O2 );
NiO 50/Co 2.5/Cu 2/Co 3/ 19% Oxygen was found to act as
CoO 0.4/TaO 0.4 a surfactant, improving
the properties of SV
2. Eindhoven NiO 50/Co 2/Cu 2/Co (t)/ 25 at 10 K 15 First quantitative [225]
Univ. Tech., Cu 1.2/NiO 10 at R.T. study of specular
Philips reflection effect both
experimentally and
theoretically
3. Alps glass/A–Fe2 O3 /Ni80 Fe20 6/ 3.7% First report on SV
Cu 2.2/Ni80 Fe20 9/Cu 4 using/A–Fe2 O3 as AFM
4. Matsushita A–Fe2 O3 50/Co 2/Cu 2/Co 5/ 18% A–Fe2 O3 was grown [227]
Cu 0.4 epitaxially on polished
(110) plane of an
A–Al2 O3 single
crystalline substrate
A–Fe2 O3 50/Co 2/Cu 2/ 27.8%
Co 5/Cu 2/Co 2/
A–Fe2 O3 50
Ta 3/PtMn 15/CoFe 2/Ru 0.7/ 13 1092 Good thermal stability [245]
CoFe 1/NOL/CoFe 1.5/
Cu 2.2–2.9/CoFe 1/
NiFe 1–2/NOL/Ta 3
Ta/AF (PtMn or A–Fe2 O3 )/ 12–17.3 414–723 [234]
CoFe/NOL/CoFe/Cu/
CoFe/Ta or
Ta/AF (PtMn or A–Fe2 O3 )/
CoFe/NOL/CoFe/Cu/
CoFe/NOL/CoFe/Ta
5. Nagoya Ta 3/NOL/Co 2/Cu 2/Co 2/ 21.6 A–Fe2 O3 as the AFM SV [235]
Univ., A–Fe2 O3 50/Al2 O3 with NOL is more
Matsushita sensitive to temperature
below 200 K
6. Toshiba Ta 5/NiFe 2/IrMn 7/PL (with 16 400 First report on NOL effect [215]
NOL)/Cu 2/CoFe 2/ inside the SV stack;
CoFeO 0.4/TaO 0.4 Natural oxidation;
Free layer Hc, 14Oe
underlayer/IrMn 7/CoFe/ 17 7 400 :Rs : 2.5–3 <: [243]
pin-NOL/CoFe 2/ Natural oxidation;
Cu 2.2/CoFe 2/Cu 1/ Testing head for
free-NOL 10 Gbits/in2 ;
Promising for
100 Gbits/in2
Sub/Ta 3/Ni62 Fe16 Cr22 3/ 15.7 −2 500 NOL formed by ion- [259]
AF (PtMn or IrMn)/ assisted oxidation is
Co90 Fe10 1/Co50 Fe50 more stable than those
1 (for FeCo NOL), formed by natural
or Co90 Fe10 1 (for oxidation;
CoFe NOL)/
continued
Nano Spintronics for Data Storage 21

Table 2. Continued

Structure of spin valve

Research group (thicknesses are in nm) GMR (%) Hin (Oe) Hex (Oe) Remarks Ref.

oxidation process/ Thick NOL reduces

Co90 Fe10 2/Cu 2.3/ ferromagnetic coupling
Co90 Fe10 2/Cu 1/Ta 1 across NOL
7. Alps + Toshiba Double specular (Ox–p + 17–18 1100–1500 Head level characterization [258]
Ta–Ox) SAF bottom SV including ESD
8. Fujitsu Underlayer/Pd32 Pt17 Mn51 / 12.3 Oscillating between >1000 Head for 56.1 Gbits/in2 [260]
CoFeB Ru/CoFeB/Cu/ 10 and −18 Oe
CoFeB/Cu/oxide
Underlayer/Pd32 Pt17 Mn51 / 15.1 Oscillating between Head for 106.4 Gbits/in2
CoFeB Ru/CoFeB/ 20 and −15 Oe
NOL/CoFeB/Cu/
CoFeB/Cu/oxide
Underlayer/PdPtMn/CoFe 12.4 Oscillating between >1000 Head for 106.4 Gbits/in2 [241]
Ru/CoFe/NOL/Cu/ 20 and −15 Oe
CoFe/NiFe/Cu/oxide
Underlayer/PdPtMn/ 15.1 Oscillating between Head for 56.1 Gbits/in2 [242]
CoFeB/Ru/CoFeB/ 20 and −20 Oe
NOL/CoFeB/Cu/
CoFeB/Cu/oxide
Underlayer/PdPtMn/ 12.3 +2 ∼2000 tCu = 20
CoFeB/Ru/CoFeB/Cu/
CoFeB/Cu/oxide
Underlayer/PdPtMn/ 8.87 Oscillating 938 Capping layer effect: [236]
CoFeB Ru/CoFeB/Cu(t)/ (∼−17 Oe when Ta capping: MR ratio,
CoFeB(1.5)/Cu/oxide tCu = 2–3 nm) 7.55%, interlayer
coupling field, 5.1Oe;
Oxide capping: MR
ratio, 8.87%,
interlayer coupling
field, −166%
9. Philips Ta 3.5/NiFe 2/IrMn 1/ 14 ∼800 Natural oxidation; [248]
CoFe 3/NOL/CoFe (t)/ Al2 O3 , a good top NOL
Cu 2.5/CoFe 4/NOL
Ta 3.5/NiFe 2/IrMn 8/CoFe 3/ 9.5 CoFe is preferred over Co [239]
Cu 2.5/CoFe 4/NOL because the former allows
for the formation of iron
oxide
Ta 3.5/NiFe 2/IrMn 8/ 13
CoFe 3/NOL/CoFe 3
Cu 2.5/CoFe 4/NOL
10. INESC Ta 6.7/NiFe 4.2/IrMn 9/ 12.5 3 Temperature-dependent study; [251]
CoFe 1.4/NOL/CoFe Anomalous bumps seen
1.5/Cu 2.2/CoFe 4/ in MR–T curves of
NOL/Ta NOL SV
Ta 7/NiFe 5/IrMn 9/ 13.6 <4 ∼180 Remote plasma oxidation; [240]
CoFe 1.4/NOL/CoFe 2.5/ RBS result indicates a
Cu 2/CoFe 2.6/ thickness of 1.5 nm
NOL/Ta 1–2 for NOL
11. TDK Sub/Ta 5/NiFe 2/NOL/ 11.8 (RuRh Better thermal stability of SV [237]
NiFe 1/CoFe 2/ Mn) with NOLs
Cu 2.5/CoFe 2/NOL/
CoFe/PtMn or
RuRhMn/Ta 5
9.4 (PtMn)
12. Sony Glass/Ta 3–5/NiFe 12/PtMn 14.9 Ru/Ox Good thermal stability [261]
10–20/CoFe 1.1–1.5/
Ru 0.8/NOL/CoFe 2–2.2/
Cu 1.9–3.2 CoFe 1.5–2.5/
Cu 1/Ta (Ta–O) 0.2–5

continued
22 Nano Spintronics for Data Storage

Table 2. Continued

Structure of spin valve

Research group (thicknesses are in nm) GMR (%) Hin (Oe) Hex (Oe) Remarks Ref.
2
13. IBM Al2 O3 3/NiCrFe 3/NiFe 1/ 13.8 Oscillating between ∼3000 >20 Gbits/in : [244]
PtMn 20/CoFe 2/ 25 and −12 Oe Al2 O3 + NiCrFe as the
Ru 0.8/CoFe 2.2/ seed layer;
Cu 2.2/CoFe 0.9/ Strong C111D texture;
NiFe 2.7/Cu 8/ Good thermal stability;
Al2 O3 1/Ta 6 Head level test (good
linearity & small
asymmetry −15%,
6.64 mV/1m,
20 Gbits/in2 )
14. IBM, Ta 3–4/Cu 1/IrMn 8.5/CoFe 10.4 5.1 460 Natural oxidation; [262]
San Jose (d)/NOL/CoFe (3.3-d)/ MR ratio highest when
State Univ. Cu 2.4/CoFe 1.6/ d ∼ 12 nm
NiFe 2.2/Ta 3.5
Ta 3–4/Cu 1/IrMn 7/CoFe 20.5
(d)/NOL/CoFe (3-d)/
Cu 2.4/CoFe 3/Cu 2.7/
CoFe 3/IrMn 7/Ta 3.5
15. ReadRite Ta 1.5–3/NiFeCr or NiFe 2–5/ 16.1 (plasma Weak oscillation 1000 >50 Gbits/in2 : [257]
PtMn 10–20/CoFe 1.5–2/ NOL), (>0) Plasma oxidation forms
Ru 0.8/CoFe 1/NOL/ 15 (natural 5.8 at tCu = 22 nm denser and more
CoFe 1.5–2/Cu 1.5–2.4/ NOL), uniform NOL;
(CoFe/NiFe or CoFe) 2–3/ 12.5 (non- Optimum Ta capping
Ta/TaO 1–5 NOL) layer thickness
∼1 nm;
No degradation in
thermal stability;
Head level test
(6.5–10 mV/1m,
asymmetry, <2% for
positive sense
current, good for
50 Gbits/in2 )
16. LLNL FM1/Cu/FM2 9.3 (non- Nonexchange-biased [238]
ReadRite FM1 = NiFeCo, NiFeCo/ NOL), sandwich
CoFe, CoFe, 12.9
FM2 = NiCoFe, CoFe/ (NOL
NiFe, CoFe/NiFe/CoFe at one
side), 13.9
(NOL at
both side)
17. Veeco Underlayer/NOL/NiFe 2.5/ ∼12.5 Oscillation (>0) ∼330–430 Systematic study of [256]
LLNL CoFe 1/Cu (t)/CoFe 3/ different oxides;
IrMn 7/Ta 2 MR increases from 26.5
to 36.7% in the order
of AlOx , TaOx , CuOx ,
NbOx , CrOx ,
NiFeCrOx , NiFeOx ,
CoFeOx
18. Veeco, SiO2 /Ta 5/NiFe 2/Ir20 Mn80 7/ 15.5 ∼480 PVD vs. IBD; [246]
ReadRite, CoFe 2/NOL/CoFe 2/ PtMn and IrMn;
LLNL, Toshiba Cu 2/CoFe 2/Cu 1/Ta 1 Simple and synthetic;
Alternative structures Atomic beam oxidation
(replacing Ir20 Mn80 gives the highest
by Pt50 Mn50 15, or FM coupling across
Pt50 Mn50 15/Co10 Co90 the NOL
2/Ru 0.8, or Ir20 Mn80
7/Co10 Co90 2/Ru 0.8

continued
Nano Spintronics for Data Storage 23

Table 2. Continued

Structure of spin valve

Research group (thicknesses are in nm) GMR (%) Hin (Oe) Hex (Oe) Remarks Ref.

19. Veeco + Top SV: Ta 2/NOL/ 9.8 Positive 574 Systematic study of atomic [245]
LLNL + NiFe 1/CoFe 2/ oscillation beam oxidation (ABO)
Toshiba Cu 2.4/CoFe 2/ (5–30) or remote plasma
IrMn 6/Ta 2 oxidation and ion beam
Bottom SV: Ta 5/NiFe 2/ 12.2 (ABO), Increase oxidation (IBO);
IrMn 7/CoFe (2-t)/ from ABO gives largest MR
NOL/CoFe 2/Cu 2/ 11.9 (ABO) 300 to ratio and exchange-bias
CoFe 2/Cu 1/Ta 1 350 field for bottom spin
Dual SV: Ta 2/NiFe 2/ 18.5 (ABO) valves
IrMn 6/CoFe (1.5-t)/
NOL/CoFe 2/Cu 2/ ∼300
CoFe 2.5/Cu 2/CoFe
(2-t)/NOL/CoFe 1.5/
IrMn 6/Ta 2
20. Univ. Ta 2/NOL/NiFe 2.5/CoFe 1/ 12.25 29 ∼400 Natural oxidation; [263]
Utah + Cu (t)/CoFe 3/IrMn 7/ Cu wedge SV;
LLNL Ta 1.5 MR decreases
monotonically with
Cu thickness
21. Seoul Nat. Sub/Ta 5/Ni81 Fe19 2/ 15.9 ∼250 [252]
Univ. + KIST Fe50 Mn50 8/Co90 Fe10 1.5/
NOL/Co90 Fe10 1.5/
Cu 2.6/Co90 Fe10 2.5/
Cu 2.6/Co90 Fe10 1.5/
NOL/Co90 Fe10 1.5/
Fe50 Mn50 8/Ta 5
Sub/Ta 5/Ni81 Fe19 2/ Fe50 Mn50 10.1 NOL may also act as a [253]
8/Co90 Fe10 2/NOL/Co90 Fe10 diffusion barrier for Mn
2/Cu 2.6/Co90 Fe10
1.5/Ni81 Fe19 4.5/Ta 5
22. Nal. Tsing Top SV: Ta 8/Co 5/Cu 2.1/ 7.8 Bottom SV: NOL at [264]
Hua Univ. Co 4/FeMn 10/Ta 5 FeMn/Co and NiFe/Ta
Bottom SV: Ta 8/NiFe 5/ interfaces increases
FeMn 10/Co 5/Cu 2.1/ and NOL at Co/Cu
NiFe 10/Ta 5 or Cu/NiFe interfaces
decreases the MR ratio;
Top SV: NOL at Ta/Co
interfaces increases
MR ratio
23. Inst. of Phys., Ta 3.5/NiFe 2/IrMn/6/ 15 ∼380 Bottom NOL is a mixture [255]
CAS UST, CoFe 1.5/NOL/CoFe 2/ of metal and oxide
Beijing Univ. Cu 2.2/CoFe (t)/NOL/Ta
Plymouth,
Nordiko Ta/NiFe/IrMn/CoFe NOL/ 15 ∼5 ∼400 Spin filter SV with NOL [254]
CoFe/Cu/CoFe/Cu/ (already employed by
NOL/Ta other groups)
24. DSI/NUS Top SV: Ta 3/NOL/NiFe 2/ 10.8 5–15 225–450 MR increases in the order [250]
CoFe 1.5/Cu 2.2/CoFe 3/ of CoFeOx , TaOx ,
IrMn 8/Ta 2 CuOx , AlOx
Bottom SV: Ta 3/NiFe 2/ 8.8 5–24 520–555 MR increases in the order
IrMn 6/CoFe 2.5/Cu 2.2/ of TaOx , AlOx . NiFeOx ,
CoFe 1.5/NiFe 3/Cu 2/ CoFeOx , CuOx
NOL/Ta 1
continued
24 Nano Spintronics for Data Storage

Table 2. Continued

Structure of spin valve

Research group (thicknesses are in nm) GMR (%) Hin (Oe) Hex (Oe) Remarks Ref.

Dual SV: Ta 3/NiFe 2/ 21.8 ∼280 Best bottom oxide:

IrMn 6/CoFe 1.5/NOL/ CoFeOx
CoFe 1.5/Cu 2/CoFe 1/
NiFe 1/CoFe 1/Cu 2/ Best capping oxide: AlOx
CoFe 2/NOL/CoFe 1/
IrMn 6/Ta 2
Ta 3/NiFe 2/IrMn 6/CoFe 1.5/ 15.3 Oscillating between ∼350 Both specular reflection [249]
NOL/CoFe 1.5/Cu 2.2/ 0 and 50 enhancement and
CoFe 1/NiFe 2/CoFe/NOL/ interface smoothening
Cu 1/Ta 1 effects have been
confirmed, but interface
layer coupling is still
ferromagnetic

Effects of Different Types of Oxides There are many position is more desirable in terms of performance and man-
different types of oxides that can be used to improve the ufacturability [250].
performance of the spin valves. The selection of oxides that To investigate the effect of different types of oxides, we
have been studied so far is more or less based on the avail- fabricated simple top and bottom spin valves with oxides
ability of the elements that are used to form the oxides being added at different locations, as shown schematically
in individual deposition systems. In general, one has more in Figure 24. The oxides that we have investigated include
choices for the capping layer than that for the NOLs inserted CoFeOx , TaOx , AlOx , NiFeOx , and CuOx . The effects of
in the middle of the spin valves because the latter will affect these oxides on the interlayer-coupling field Hin , exchange
the magnetic properties of the spin valves. For example, if bias Hex , and MR ratio are shown in Figure 25. The main
it is inserted in the middle of the pinned layer, it cannot results can be summarized as follows. First, the exchange
be too thick; otherwise, it will separate the pinned layer bias of the top spin valve is much smaller than that of the
into two sublayers, and result in a substantial decrease of bottom spin valve. This is caused by both the property of
the exchange bias. Although it can be made thicker if it is the AFM (in this case, IrMn) itself and the oxide-induced
inserted at the AFM/seed layer interface, it cannot be too changes in the texture of the underlying layers. The latter
thick; otherwise, it will affect the crystalline texture of the is especially severe in the case of TaOx and AlOx . CoFeOx
layers that are subsequently deposited on top of the oxide. and NiFeOx are better because they were formed from the
This eventually will lead to a decrease of exchange bias. oxidation of CoFe and NiFe, which already had a well-
Bearing all of these in mind, the author’s group has carried established texture before the oxidation took place. So does
out a systematic study on: (1) how different types of oxides Cu because it is usually used as the texture control layer for
affect the performance of the spin valve, and (2) which
15
(a) 36 (c)
12 33
Rsq(Ω)
MR (%)

30
9
27

24
6

Proof's Only
21
1.5 2.0 2.5 3.0 3.5 1.5 2.0 2.5 3.0 3.5

70 5.0
60 (b) 4.5 (d)
50 4.0
∆Rsq(Ω)

40 3.5
Hint(Oe)

30 3.0
∆Rsq(Ω)

20 2.5
10 2.0
0 1.5
1.5 2.0 2.5 3.0 3.5 1.5 2.0 2.5 3.0 3.5

tCu(nm) tCu(nm)

Figure 23. Spacer (Cu) thickness dependence of MR (a), Hint

Figure 22. (a) MR ratio (solid circle) and Hint (open circles) as a func- (b), sheet resistance Rsq (c), and sheet resistance change (d) for
tion of O2 exposure dose. (b) Sheet resistance (open squares) and two batches of SVs with the structure Ta3/NiFe2/IrMn6/CoFe3/
change of sheet resistance :Rsq (solid squares) as a function of O2 Cu(t)/CoFe1/NiFe2/CoFe1.2/Cu1/Ta1 (open symbols) and Ta3/NiFe2/
exposure dose for a series of SVs with the structure Ta3/NiFe2/IrMn6/ IrMn 6/CoFe1.5/NOL /CoFe1.5/Cu(t) /CoFe1/NiFe2/CoFe1.2/NOL/Cu1/
CoFe1.5/NOL/CoFe1.5/Cu2/CoFe1/NiFe1/CoFe1.2/NOL/Cu1/Ta1 with Ta1 (solid symbols). The dotted line in (b) was fitted by using Néel’s
a deferent O2 exposure dose. The thicknesses are in nanometers. orange-peel coupling equation. The thicknesses are in nanometers.
Nano Spintronics for Data Storage 25

Texture Free layer Texture

Free layer control layer control layer
Cu pinned layer AFM
CL AFM pinned layer Seed layer
Cu Seed layer

N xid
an e
o
o-
Na
no

Proof's Only
-o
xi Na 1 2 3 4 5 6 7 8
de no
l ay -o
xi
er d el
ay
er

(a) (
(b)

Figure 24. Schematic of a top spin valve with a nanooxide inserted at AFM PL1 Ru PL2 Cu FL Cu Ta
the free and texture control layer interface (a), and a bottom spin valve
with a nanooxide capping layer (b). Figure 26. Schematic of the nanooxide-added spin valves with the oxide
inserted at different locations. Here, PL1 and PL2 refer to the first and
second pinned layers, respectively, and FL stands for the free layer.
NiFe and CoFe. Second, the interlayer-coupling field of the
top spin valve is, in general, smaller than that of the bot- Figure 27 shows the MR–H curves of case (5), in which
tom spin valve, except for the case using AlOx . This again the spin valve has a structure Ta3/NiF2/IrMn8/CoFe2.5/
suggests that the nanooxide has the surface smoothening Ru0.8/CoFe2/NOL/Cu2.2/CoFe2/Cu1/Ta1.5 (thicknesses are
effect, as discussed earlier. The improvement is particularly in nanometers). Here, NOL stands for the nanooxide layer.
remarkable for CoFeOx . Finally, we can see that the MR The NOL was formed through natural oxidation of the sec-
ratio of the top spin valve, in general, is larger than that ond pinned layer in a separate chamber, with the thickness
of the bottom spin valve. Comparing these results, we can being controlled by the oxygen dose. After the oxidation
see that CoFeOx is, so far, the most desirable bottom oxide, was completed, the sample was brought back again to the
while AlOx seems to be a better choice for the top oxide. growth chamber for the deposition of the rest of the lay-
For other oxides to be better than CoFeOx , one must find a ers. All of the deposition and oxidation processes were
material that has a similar crystalline structure and texture accomplished in a multiple-chamber ultrahigh-vacuum sys-
as that of the CoFe layer in spin valves. tem without breaking the vacuum. As can be seen from the
figure, the MR ratio initially increases under light-oxidation
Effect of the Position of Oxide In addition to the types conditions, beginning to decrease when the oxygen exposure
of oxides, we have also carried out a systematic study on dose increases, and eventually vanishing out as the oxidation
the position of oxide, that is, which position is more suit- process continues. Figure 28 shows the detailed dependence
able for the bottom oxide. Figure 26 shows schematically of the MR ratio and the interlayer coupling on the oxy-
the different positions that have been investigated. These gen dose. It is worth noting that the interlayer-coupling field
are: (1) inside the AFM layer, (2) at the AFM/first pinned decreases and the MR ratio increases in the light-oxidation
layer interface, (3) at the Ru/second pinned layer interface, regime by increasing the oxygen dose. However, the former
(4) inside the second pinned layer, (5) at the second pinned increases sharply when the latter starts to decrease as the
layer/Cu interface, (6) inside the spacer, (7) at the Cu/free oxygen exposure dose is further increased. The behavior of
layer interface, and (8) at the top surface. In what follows, the MR ratio is easily understood, but it is not as straightfor-
we describe the results for cases (4)–(6). ward to explain the sharp increase of the interlayer-coupling
field. As the interlayer-coupling field normally decreases by
decreasing the pinned layer thickness [266], it is plausible to
25 600 postulate that the sharp increase of Hin has something to do
550
20
500 with the decrease of the spacer thickness because the sur-
Hin (Oe)

face roughness should not change so drastically. Figure 29

Hex (Oe)

15 450
400
10 350 shows the X-ray diffraction spectra of spin valves under dif-
300
5
250 ferent oxidation conditions. Curves (a)–(d) correspond to
0 200
CoFeO TaO AlO NiFeO CuO CoFeO TaO AlO NiFeO CuO
(a) (b)
10
Light oxidation
12
11 8 Without oxidation
MR ratio (%)

10 Intermediate
MR (%)

oxidation
9 6
8
Heavy
7 4
oxidation
6
CoFeO TaO AlO NiFeO CuO 2
(c)
0
-6 -4 -2 0 2 4 6
Figure 25. Effect of different nanooxides on the interlayer coupling H (kOe)
field (a), exchange bias (b), and MR ratio (c) of bottom (open cir-
cles) and top (open squares) spin valves whose structures are shown in Figure 27. MR–H curves of spin valves oxidized under different oxygen
Figure 24. doses at the surface of the second pinned layer.
26 Nano Spintronics for Data Storage

70 10
2 (a)
60 1
8
(b)
50
Hint (Oe)

Intensity (a.u.)
MR(%)
3 6
40 (c)

30 4 (d)

20
2
10 4
-8 -7 -6 -5 -4 -3 -2 -1
0
0
10 10 10 10 10 10 10 10 10
Oxygen Dose (Pa.s)

Figure 28. Dependence of the interlayer coupling field and MR ratio 0.5 2.5 4.5 6.5
of spin valves on the oxygen dose that was used to oxidize the second 2 (°)
pinned layer.
Figure 30. Low-angle X-ray diffraction spectra of spin valves oxidized
under different conditions. Cases (a)–(d) correspond to cases (1)–(4),
cases (1)–(4) shown in Figure 28. The peak at an angle of
respectively, in Figure 28.
about 41.2
is from the IrMn layer, while that at about 43.7

is from the rest of the layers. It is worth noting that the
higher degree peak splits into two new peaks in case (d): NOL/CoFe1.5/Cu2.2/CoFe2/Cu1/Ta1.5 (thicknesses are in
one at an angle of about 42.9
, and the other at about 44
. units of nanometers). In this case, the oxide was formed
Although the details are not well understood at present, it in the middle of the second pinned layer. As can be seen
seems that Cu has been oxides partially to form copper oxide from the figure, the interlayer-coupling field is almost con-
or a mixture of copper oxide and CoFe oxide. The sharpness stant at the beginning, decreases, and then increases again
of the interfaces has been investigated using low-angle X-ray when the oxygen exposure dose increases. The MR ratio,
reflectometry. As shown in Figure 30, the four curves corre- however, increases first, then decreases, and increases again.
sponding to the four different cases shown in Figure 29 can Although it is difficult to understand the behavior of the
hardly be differentiated from one another. This implies that MR, one thing that is clear is that the process window of
the change in surface roughness is unlikely responsible for adding oxide at this position is much wider than that of the
the sharp increase of the interlayer-coupling field shown in previous case; thus, it is more desirable from the point of
Figure 28. The results demonstrate that it is difficult to form view of manufacturability.
a controllable oxide at the interface between two materials if The last position to be discussed here is the middle of
both can be readily oxidized. Figure 31 shows schematically the spacer. This position has not been discussed before, and
the possible oxidation processes. Under heavy oxidation con- thus it is worth investigating how it affects the performance
ditions, the loosely absorbed oxygen atoms on the CoFeOx of the spin valve. One tends to think that this will affect
surface may result in a partial oxidation of the subsequently the performance adversely because it may reduce the mean-
deposited Cu layer. This may explain the sharp increase of free path of electrons. However, as shown in Figure 33, the
the interlayer coupling field when the oxygen exposure dose MR ratio indeed increases with the oxygen exposure dose
exceeds a critical value. that is used to oxidize the Cu layer in a spin valve with
From the above discussion, it is clear that adding NOL a structure Ta3/NiF2/IrMn8/CoFe2.5/Ru0.8/CoFe2/Cu1.2/
at the second pinned layer/Cu interface might not be NOL/Cu1.2/CoFe2/Cu1/Ta1.5 (thicknesses are in units of
a viable approach due to its narrow process window. nanometers). This might be attributed to the enhancement
The next position that has been widely studied since the of scattering probability at the free layer/Cu/pinned layer
first report of the nanooxide spin valve is in the mid-
dle of the pinned layer. Figure 32 shows the depen-
dence of the interlayer-coupling field and the MR ratio
on the oxygen exposure dose for which the spin valve
has the structure Ta3/NiF2/IrMn8/CoFe2.5/Ru0.8/CoFe1.5/

120 (a) (b)

(a)
100
(b)
Intensity (a. u.)

80

Proof's Only
(c)
60 Cu
40 (d) O
20
Co
0
Fe
38 40 42 44 46

2θ (o) (c)

Figure 29. High-angle X-ray diffraction spectra of spin valves oxidized Figure 31. Schematic illustration of the oxidation process of CoFe (b),
under different conditions. Cases (a)–(d) correspond to cases (1)–(4), and the subsequent deposition and partial oxidation of Cu (c). (a) shows
respectively, in Figure 28. the CoFe surface before oxidation.
Nano Spintronics for Data Storage 27

11.0 3.6
30
10.8
25 3.4
MR: 27.2%

Resistance (ohm)
10.6

Proof's Only
20 3.2
10.4
H int (Oe)

MR(%)
15
10.2 3.0
10
10.0
5 2.8
9.8 -600 -400 -200 0 200 400 600 800 1000 1200
0 Field (Oe)
9.6
10-8 10-7 10-6 10-5 10-4 10-3 10-2 10-1 100 101
Figure 34. MR–H of a double specular dual spin valve with a struc-
Oxygen Dose (Pa.s) ture Ta3/NiFe2/IrMn5/CoFe1/NOL/CoFe2.3/Cu2.15/CoFe2/Cu2.15/
CoFe2.3/ NOL/CoFe1.0/IrMn5/Ta1. Here, the thicknesses are given in
Figure 32. Dependence of the interlayer coupling field and the MR nanometers.
ratio of the spin valves on the oxygen dose that was used to form the
oxide at the middle of the second pinned layer.
there are some differences in how their performance can be
improved. In this section, we first explain why we still need
interfaces. In this case, the MR will increase provided that CPP spin valves for data storage, followed by a discussion
the scattering inside the spacer is spin conserved. Further on possible ways for improving the performance of CPP spin
study is required to understand the underlying mechanism. valves.
As we mentioned above, the study of the nanooxide effect The areal density of a hard disk drive is the product of
is still at an early stage. A theoretical framework has yet the linear density and track density. The former is deter-
to be formulated to guide the selection of oxides and their mined by many component and system parameters, while
positions in the spin valves. Further experimental investiga- the latter is largely determined by the capability of lithog-
tions are required to study the structural properties of the raphy techniques. In what follows, we first focus on linear
nanooxide-added spin valves. Before ending this section, we density, which is defined as
show in Figure 34 the MR–H curve of a dual spin valve with
a structure Ta3/NiFe2/IrMn5/CoFe1/NOL/CoFe2.3/Cu2.15/ 1 channel density
linear density = = (27)
CoFe2/Cu2.15/CoFe2.3/NOL/CoFe1.0/IrMn5/Ta1. After the bit length P W50
optimization of structure and processes, we have success-
fully obtained an MR ratio of 27.2%, which we believe is where P W50 is the full-width-at-half-maximum of the read-
the highest value ever reported for all-metal spin valves. out pulse from an isolated transition. The channel density is
Although the exchange-bias field is too low for practical a parameter that reflects the efficiency of the channel cod-
applications, this might be solved by replacing IrMn with ing scheme (∼2.5–3). The P W50 can be estimated using the
other AFM materials such as PtPdMn and NiMn. following equation:
 1/2
P W50 = 2g + t/22 + 4deff + a2 (28)
5.4. CPP Spin Valves
where
5.4.1. Why CPP?

 M t 1/2
So far, most of the discussion has been concentrated on CIP deff = dd + F and a = k deff r  (29)
spin valves. Although many of the fundamental concepts for Hc
CIP spin valves are directly applicable to CPP spin valves,
Here, d is magnetic spacing, g is the gap length between the
sensor and the two shields, t is the thickness of the sensor, F
10.8 is the thickness of the media, and a is the transition width.
Mr and Hc are the remnant and coercivity of the media,
10.4 respectively. k is a parameter with a value ranging from 1
to 2. The above equation suggests that there are many possi-
MR (%)

10.0
ble ways to increase the linear density, although the reality is
9.6
that none of them is an easy task. Although one can always
try to push all of the parameters to their individual upper
9.2 or lower limits, it would make more sense if one tries to
determine which parameter is of more relative importance
8.8
10-8 10-7 10-6 10-5 10-4 10-3 10-2 10-1 100
first. To this end, we have tried to find the rate of change of
P W50 with respect to different parameters by assuming that
Oxygen Dose (Pa.s) we have already achieved 100 Gbits/in2 . It turned out that
Figure 33. Dependence of the MR ratio of the spin valves on the oxy- P W50 P W50
gen dose that was used to form the oxide in the middle of the spacer. = 26 and
d Mr F/Hc 
The upper curve was obtained from the major loop, while the lower
curve was obtained from the minor loop. Both show a similar depen- P W50 P W50
= 21 (30)
dence on the oxygen dose. d g + t/2
28 Nano Spintronics for Data Storage

where the following parameters have been used for the electrode
calculation: d = 12 nm, F = 12 nm, Mr F/Hc = 125 nm, T

Proof's Only
k = 1, and g + t/2 = 45 nm. It is clear that the most sensi-

shield
CIP

shield
tive parameter is the magnetic spacing, followed by the read D

gap length, and then media parameters. If the areal density CPP
continues increasing at an annual compound rate of 60%, disk W
in fewer than five years, we will reach 1 Tbit/in2 [268]. Now,
(a) (b)
the question is: What kind of sensor is required for this level
of areal density? Figure 35. (a) Schematic drawing of CPP read head in which the shields
To answer this question, we first look at what are the can also be used as the electrodes which help to eliminate the read gap.
requirements on bit length and track width for terabit (b) Schematic of a sensor element used in both the CPP and CIP modes.
recording. Shown in Figure 13 are the estimated read track
width and required sensor sensitivities at different areal den-
sities by assuming an aspect ratio of 4:1. The solid and point of view. Note that the above discussion is only valid
dashed curves are obtained by assuming the read track width for constant current density operations. The dependence of
to be 80 and 60% of the track pitch, respectively. The sen- the output signal on the sensor width will be relaxed for the
sitivity requirement is obtained by assuming that the sen- CIP sensor when it is operated under a constant power.
sor will have to generate an output voltage of 1 mV. The
read head for terabit recording would require a bit length
as small as 12 nm, assuming that the read track width is 5.4.2. Sensitivity Enhancement
80% of the track pitch. This leads to a pulse width of about of CPP Sensors
36 nm, assuming that the channel density can be increased Both theoretical and experimental studies on the CPP–GMR
to 3. Assume that all other parameters are approaching their started shortly after the discovery of CIP–GMR in ferro-
practical limits, such as d = 5 nm, F = 5 nm, and Mr F/Hc = magnetic/nonmagnetic multilayers [280–285]. It was found
625 nm (maybe even smaller); then g + t/2 must be as small experimentally that the CPP–GMR is larger than the CIP–
as 17 nm. This means that almost no read gap is allowed GMR in multilayer samples [280, 281, 286–288]. For spin
in sensors for terabit recording. This poses formidable chal- valves, Vedyayev et al. estimated theoretically that the CPP–
lenges for CIP read sensors because one needs electrodes
GMR should also be larger than the CIP–GMR [289]. By
between the two shields. Furthermore, the sensitivity of the
starting with the Boltzmann equation, Valet and Fert derived
sensor also has to be increased by a factor of 2–3, which
an analytical equation for the CPP–GMR [284]. This was
means that the MR ratio must be around 30%. This might
later extended to the treatment of an exchange-biased spin
be possible for advanced spin valves at the sheet film level
valve with a typical structure AFM/FM/NM/FM [290 291].
(as we have already achieved an MR ratio of about 27%),
In the case where the spin diffusion length is much longer
but it would be difficult to achieve the same figure at the
than the layer thicknesses, Bass and Pratt obtained [291]
sensor level.
Based on this background, the CPP spin valve is being
studied as one of the possible candidates for high-density 4G∗F tF + HAR∗F /N 2
recording applications [269–279]. One of the advantages of A:R = (32)
2∗F tF + N tN + 2AR∗F /N + ARAF /F
the CPP spin valve over the existing current-in-plane spin
valve is that it allows the elimination of the insulating gap
layer in the read head, which will contribute to the increase where tF , tN , and tAF are the thicknesses of the ferro-
of the linear density [see Fig. 35(a)]. More importantly, the magnetic, nonmagnetic, and antiferromagnetic layers, and
output voltage of the CPP sensor does not scale linearly with ARAF /F is the specific resistances at the anitiferromagnetic
the width of the sensor as the CIP does. Assume that one layer (AF )/ferromagnetic layer (F ) interface. The bulk spin
has a read element with width W , height H, and thickness asymmetry coefficient (G), interface spin asymmetry coeffi-
T , as shown schematically in Figure 35(b); the outputs of cient H ∗F , and AR∗F /N are defined as
the sensor in the CIP and CPP modes are given by
↓ ↑ ↓ ↑
CIP F − F ARF /N − ARF /N
:Vmax  = :CIP W J G=  H=
↓ ↑ ↓ ↑
CPP F +  F ARF /N + ARF /N
:Vmax  ≈ :CPP T J  (31)
↓ ↑ ↓ ↑
 + F  ARF /N + ARF /N 
Here, :CIP and :CPP are the resistivity change due to ∗F = F  AR∗F /N = (33)
the GMR effect (nominal values including the nonactive 4 4
regions), W  and T  are the average width and thickness
↑ ↓ ↑ ↓
of the active region, and J is the current density. The above where F F  and ARF /N ARF /N  are the bulk resistivity,
equations suggest that the CPP sensor is more suitable for and specific resistances at the F /N interfaces, for the major-
ultrahigh-density recording because it is much easier to con- ity and minority electrons, respectively. The above equa-
trol the thickness than to control the lateral dimension. This tion was obtained by assuming that the specific resistances
is because the former is determined by the film thickness, between the ferromagnetic layer and antiferromagnetic layer
while the latter is determined by lithography. Therefore, the or the contacts are independent of the direction of electron
CPP sensor is also advantageous from the manufacturability spin with respect to the local magnetization direction. On
Nano Spintronics for Data Storage 29

the other hand, in another extreme case where the spin dif- obtained using this technique, the maximum value of A:R
fusion length of the ferromagnetic layer is shorter than its obtained is still below 2 m< · 1m2 .
thickness, A:R is given by [291] Among all of these techniques that have been investi-
gated, the nanooxide gives the most significant enhancement
4G∗F lsfF + HAR∗F /N 2 to A:R without the need to introduce more layers or to
A:R = (34) increase the layer thickness physically; thus, it is desirable
2∗F lsfF + 2AR∗F /N + N tN
for practical applications. However, the oxide must be suf-
ficiently thin; otherwise, it will affect the flow of electrons
where lsfF is the spin diffusion length of the F layers, and all across the spin-valve stack in the CPP mode or increase the
other notations are the same as those in Eq. (32). total resistance to an unacceptable level. However, if the
When being used as a read sensor, the output voltage oxide layer is too thin, it will lose its original functional-
of a CPP SV sensor is given by :V = AJA:R. Here, J ity as an electron-reflection or resistivity-enhancement layer.
is the current density whose maximum value is limited Therefore, using oxide alone, one may not be able to obtain
to ∼107 A · cm−2 due to electron migration and power the same effect as what has been obtained in the CIP mode.
consumption considerations, and A is the head efficiency In addition to the number of layers, A:R also increases
coefficient multiplied by the dynamic range coefficient of the with the bulk and interface spin asymmetry coefficient, as we
sensor, which is about 0.2–0.5, depending on the structure of discussed above. This might be achieved through using half
the read head. Assuming A = 02 and J = 5 × 107 A · cm−2 , metals as the electrodes, but so far, no suitable materials
an A:R of 10 m< · 1m2 is required to obtain an output have been found. Based on this background, we have pro-
voltage of 1 mV. A typical spin valve, however, can only posed a possible solution to overcome this difficulty [277].
have an A:R of about 1–2 m< · 1m2 [270 274 279]. It is known from the research work on semiconductor
Therefore, the performance of the CPP SV has to be devices that electrons can be reflected not only by a single
improved before it can be applied to read heads. From potential barrier of large height, but also by a superlattice-
Eq. (32), which applies to most of the practical cases, type of structure with moderate barrier heights at the inter-
one can see that A:R can be increased by increasing the faces, provided that the thickness of each layer is chosen
parameters in the numerator or decreasing the parameters such that it is one quarter of the de Broglie wave of electrons
in the denominator. We focus on the numerator first. The [293 294]. This type of structure is often called an electron
most straightforward way to increase A:R is to increase the Bragg reflector [Fig. 36(a)]. It is relatively easy to realize this
thickness of the F layers. However, this is not a preferred kind of structure in semiconductors because the de Broglie
approach because the free layer of a read sensor must wavelength is relatively large. In metallic materials, how-
be made as thin as possible; otherwise, it will degrade ever, the electron wavelength is close to the Fermi wave-
the sensitivity of the sensor. Although the values of other length, which is just about 0.1–0.2 nm. If this is the “true”
parameters, G H, and AR∗F /N , can also be increased, their length scale, it is almost impossible to build a Bragg reflector
contribution to the enhancement of A:R is moderate for electrons using metallic materials. Fortunately, due to
because all of these parameters are determined by the the discreteness of the atomic layers, the actual wavelength
properties of the constituent materials [272]. We are now of the electron packet, or so-called envelope functions, is
left with only one parameter in the numerator, ∗F , which is given by 2,/:K [295]. Here, :K is the magnitude of the
given by F /1 − G2 . If one can find a way to increase the
resistivity of the F layers without changing G significantly,
then A:R will increase accordingly. It would be even more mλ nλ
desirable should G be increased simultaneously. One of the 4 4
possible approaches is to add nanooxides inside the F layers
[270 274]. The nanooxide effect is twofold. One of the
effects is to increase the effective resistivity of the F layers
without significantly changing the spin asymmetry coeffi-
cient. The second effect is to enhance the specular reflection
(a)
of electrons so as to increase the number of F /N layers
effectively. Nagasaka et al. showed that the nanooxide-

Proof's Only
FM NM
added CPP spin valves exhibit an A:R 25 times larger
than that of the conventional spin valves [274]. Focusing on
the denominator, Gu et al. showed that the CPP–GMR in
permalloy-based spin valves can be enhanced by increasing
the spin-memory loss in ferromagnetic layers [292]. The Spin-down electrons
particular structure that has been investigated by Gu et al.
is Nb250/Cu10/FeMn8/Py12/Cu20/Py(t1)/FeMn(t2)/Py(12-
t1)/Cu10/Nb250, where Py stands for permalloy, layer
thicknesses are in nanometers, t2 = 05 or 1 nm, and t1 Spin-up electrons
was adjustable from 1 to 11 nm. The ultrathin FeMn layer (b)
inside the free layer was used to increase the spin memory
loss due to its extremely short spin diffusion length of about Figure 36. Schematic of a spin-independent Bragg reflector (a), and a
1 nm. Although significant improvement in A:R has been spin-dependent Bragg reflector (b).
30 Nano Spintronics for Data Storage

wavevector that spans the appropriate stationary points of reflection effect, ultrathin oxides can be added at both the
the Fermi surface. This gives a wavelength of about 1.1 nm top of the free layer and the middle of the pinned layer
for Cu, for example. The Bragg reflection will occur when [Fig. 37(b)]. The same approach can be taken for spin valves
the conditions dF M = m3F M /4 and dNM = n3NM /4 are sat- using Bragg reflectors at both the free and pinned layers
isfied. Here, dF M and dNM are the thickness of the ferromag- [Fig. 37(d)]. Further to these steps naturally will be the one
netic and nonmagnetic layers, respectively, and m and n are that adopts dual spin-valve structures, as is the case with
odd integer numbers. If one chooses m and n in the range current-in-plane spin valves. The functions of the nanooxide
of 3–4, the thickness of each layer should be well within the are twofold. First, it increases the specular reflection. Sec-
controllable range of the state-of-the-art sputtering systems. ond, it reduces the interface roughness so that the electron
Now, a question arises here: How can the electron flow interference effect can occur.
in the CPP mode if one has a perfect Bragg reflector in the A series of experiments have been carried out to verify
spin-valve stack? The point is that the quantum-well states the concept. The sheet films were fabricated using ultrahigh-
only form when all of the ferromagnetic layers are ferro- vacuum sputtering. The detailed sputtering conditions can
magnetically coupled, and they only exist for the minority be found in [249 250]. After the sheet film was fabricated,
electrons [295]. In other words, the Bragg reflector in a mag- it was patterned into elements with different sizes using
netic/nonmagnetic multiple layer is a spin-dependent reflec- the e-beam lithography and ion-milling processes. The CPP
tor [see Fig. 36(b)]. It reflects only the minority electrons, spin valves that have been fabricated include: (1) simple
but lets the majority electrons flow almost freely. This will single spin valves without nanooxides and Bragg reflectors
effectively increase the bulk and interface spin asymmetry (type A), (2) single spin valves with a Bragg reflector free
coefficient, and thus A:R of spin valves. Note that the pic- layer (type B), (3) single spin valves with both a Bragg reflec-
ture given here is based on the ballistic transport of electrons tor free layer and a nanooxide cap layer (type C), (4) dual
instead of diffusive transport; thus, Eqs. (32) and (34) might spin valves with a Bragg reflector free layer (type D), and
not be directly applicable to this case. (5) dual spin valves with both a Bragg reflector free layer
Figure 37 illustrates some of the possible structures of and nanooxides at both the top and bottom pinned layers
CPP sensors using the electron Bragg reflectors, which con- (type E). Figure 38 shows a schematic drawing of the spin-
sist of a number of pairs of CoFe and Cu. The thickness valve structures. The Bragg reflector used has a structure
of Cu can be chosen such that it facilitates a strong paral- [CoFe(1.4 nm)/Cu(1.1 nm)] × 5. The antiferromagnet and
lel coupling between the adjacent ferromagnetic layers, but pinned layers are IrMn (8 nm) and CoFe (3 nm), respec-
at the same time, it satisfies the Bragg reflection condition. tively. The top oxide was formed through the oxidation of
If the priority is to achieve ferromagnetic coupling, any dif- NiFe, and the oxide in the middle of the pinned layer was
ference in the thickness of the Cu layer to the value that formed through natural oxidation of CoFe. The total thick-
is requested by the Bragg condition can be compensated ness of the CoFe pinned layer was always kept constant.
by the thickness of the corresponding ferromagnetic layer. To ensure a reliable measurement of the change in the
The m and n are not necessary to be the same; m can resistance–area product, we first tried to determine the
be small so as to reduce the total thickness of the Bragg dependence of the junction resistance on the junction area.
reflectors. The Bragg reflectors can be used either in the Figure 39 shows the typical results for types A, B, C, and
free layer only [Fig. 37(a) and (b)] or in both the free and E spin valves. As can be seen from the figure, the junc-
pinned layers [Fig. 37(c) and (d)]. To enhance the specular tion resistance is inversely proportional to the junction area,
indicating that current distribution is quite uniform over the
NOL
entire junction. Similar results have been obtained for type
D sensors. Table 3 lists the RJ and :RJ of the five different
BR FL

Cu Cu
(a)
PL NOL
AFM AFM
AFM PL Cu FL AFM PL\NOL\PL Cu BR FL Cu PL\NOL\PL AFM
(a) (b) (e)

Proof's Only
(b)
NOL Top Lead

BR FL
Bottom Lead

AFM PL Cu BR FL Al-O insulator

Cu (c)
Bottom Lead

Cu

BR PL
AFM PL\NOL\PL Cu BR FL NOL

NOL CPP
AFM AFM (d) Pillar
Top Lead
(c) (d)
(f)
AFM PL1 Cu BR FL Cu PL2 AFM
Figure 37. Schematic of some possible CPP spin valves using both
nanooxide and Bragg reflectors. Here, BR stands for Bragg reflector, Figure 38. (a)–(e) Schematic drawings corresponding to sensor type
NOL refers to nanooxide, and FL (PL) is the free (pinned) layer. A–E. (f) Optical micrograph of typical CPP sensors.
Nano Spintronics for Data Storage 31

10
Type A Type B the spin-dependent scattering effect, both inside the ferro-
RJ=1.9Ωµm
2
1
RJ=2.0Ωµm
2
magnetic layers and at the ferromagnetic/nonmagnetic inter-
faces, while that in the latter is based on spin-dependent
2 2
1 ∆R=5.0m Ωµm ∆R=8.0m Ωµm
R (Ω)

R (Ω)
quantum mechanical tunneling across a thin potential bar-
0.1 0.1 rier. To have a sound understanding of MTJ, we first look
at the basic properties of a normal nonmagnetic tunnel
1 10 100 1 10
junction. A typical quantum-mechanical tunnel junction is a

Proof's Only
S (µm2) S (µm2)
sandwich of two conductive electrodes separated by a thin
insulating barrier layer. For a tunnel junction with a suffi-
Type C
2
Type E
RJ=40.0Ωµm
2 ciently high potential barrier, the tunneling current can be
RJ=5.0Ωµm 10
∆R=15.0m Ωµm
2
∆R=36.0m Ωµm
2 calculated using the transfer Hamiltonian approach, which
1
R (Ω)

R (Ω)
reads [297 298]
1

0.1
0.1
I =A M2 D1 E + EF 1 D2 × E + EF 2 + eV 
−
1 10 100 1 10 100
2
S (µm ) S (µm )2 × /f E + EF 1  − f E + EF 2 + eV 0dE (35)

Figure 39. Dependence of the junction resistance on the junction area where
for different types of CPP spin valves.
2 
M= J K J2∗ − J2∗ K J1  · dS
2m S 1
types of spin valves. In general, we can see that both the is the matrix element of the transfer Hamiltonian, D1 and
nanooxide and the Bragg reflector can improve the perfor- D2 are the density of states for the left and right electrodes,
mance of the spin valve. The large junction resistance is due respectively, f is the Fermi distribution function, S is the
to the contact resistance at the sensor/electrode interface. cross-section area of the junction, m is the electron mass,

We are currently developing a new technique to remove the is the Planck constant, V is the bias voltage, E is the elec-
junction resistance from the measured data. But the junc- tron energy, EF 1 (EF 2 ) is the Fermi level of the left (right)
tion resistance will not affect the accuracy of the measured electrode, and A is a constant. J1 and J2 are the wavefunc-
A:R as much because we have already shown in Figure 39 tions of the electrons at the initial and final states. In the
that the current distribution problem is negligible in these case where the voltage applied across the junction is small,
samples. one has
The above results show that a well-designed and fabri-

cated Bragg reflector may play the same role as a nano- f

f E + EF1  − f E + EF2 + eV  = − · eV  (36)

oxide, but the former is more suitable for CPP spin valves E
E=EF
because it does not increase the resistance substantially, as
dI
the oxides do. The results shown here are preliminary; we GV = 0 = ∝ M2 D1 EF D2 EF  (37)
cannot exclude other effects introduced by the insertion of dV
the laminated free layers. But we believe that this type of where G is the conductance. For tunnel junctions with ferro-
structure will become more useful in the future when suit- magnetic electrodes, Eq. (37) is still valid, except that D1 and
able magnetic semiconductors are available because it is D2 are now dependent on the spin polarization of electrons.
much easier to form Bragg reflectors using semiconductors. This is known to result in a different tunneling conductance
Although it is not in spin valves, a similar type of electron between the parallel and antiparallel alignment of the mag-
interference effect has already been observed in magnetic netizations of the two electrodes, as shown schematically in
tunnel junctions grown epitaxially on a single-crystal sub- Figure 40. The total conductance in the parallel configura-
strate [296]. tion is thus approximately given by
↑ ↑ ↓ ↓
G↑↑ V = 0 ∝ D1 EF D2 EF  + D1 EF D2 EF  (38)
6. MAGNETIC TUNNEL JUNCTIONS
and that in the antiparallel configuration is
Another type of sensor that operates in the CPP mode is
the magnetic tunnel junction (MTJ) [26–29]. The key dif- ↑ ↓ ↓
G↑↓ 0 ∝ D1 EF D2 EF  + D1 EF D2 EF 
↑
(39)
ference between the CPP GMR sensor and the MTJ is that
the electrical conduction in the former case is based on ↑ ↓ ↑ ↓
Here, D1 D1  and D2 D2  are the density of states of the
majority (minority) electrons of the two electrodes, respec-
tively. In deriving the above equation, we have assumed that
Table 3. RJ and :RJ of different types of CPP spin valves.
spin is conserved during tunneling. The junction magnetore-
Type A Type B Type C Type D Type E sistance (JMR) is defined as

RJ < · 1m ) 2
19 20 50 20 40 1/G↑↓ 0 − 1/G↑↑ 0
:RJ (m< · 1m2 ) 50 80 15 26 36 JMR = (40)
1/G↑↓ 0
32 Nano Spintronics for Data Storage

from 15 to 22% at room temperature and in fields of a few

tens of Oersteds. The microstructured and exchange-biased
curren
magnetic tunnel junctions were further developed to have
a higher MR ratio (42%) and moderate specific junction
resistance (∼60 < · 1m2 , and were applied to prototypes
of MRAMs [10]. After these pioneer works, a tremendous
G ↑↑ (0) ∝ D1↑ ( EF ) D2↑ ( EF ) + D1↓ ( EF ) D2↓ ( EF ) FM1 Al2O3 FM2 amount of effort was devoted to the development of mag-

Proof's Only netic tunnel junctions and their application in MRAMs and
read heads. In what follows, we try to summarize some of
the key issues that have been addressed in the literature for
curren magnetic tunnel junctions.

6.1. Electrodes
FM1 Al2O3 FM2
6.1.1. Spin Polarization
↑↓ ↑ ↓ ↓ ↑
G ( 0) ∝ D ( E F ) D ( E F ) + D ( E F ) D ( E F )
1 2 1 2

Figure 40. Schematic illustration of spin-dependent tunneling across an The simple theoretical model suggests that ferromagnetic
insulating barrier. materials with large spin polarizations are the key to achiev-
ing a high MR. For a transition metal ferromagnet, the spin
polarization is largely determined by whether the electrons
which can be further reduced to participating in electrical conduction have an s-like charac-
2P1 P2 teristic or a d-like characteristic, or a mixture of these two.
JMR =  (41) As the MR is determined by the spin-dependent density of
1 + P 1 P2
states at the Fermi surface, its value is largely determined
Here, by the extent to which the s and d bands appear across the
Fermi surface. However, the experimental determination of
↑ ↓
P1 = /D1 EF  − D1 EF 0//D1 EF  + D1 EF 0
↑ ↓ spin polarization for each material is quite challenging. The
↑ ↓ ↑ ↓
most straightforward way to determine the polarization is
P2 = /D2 EF  − D2 EF 0//D2 EF  + D2 EF 0 (42) to use the spin-polarized photoemission spectroscopy tech-
nique, but it lacks the necessary energy resolution. An alter-
are the spin polarizations of electrons in the two ferro- native technique that has been pioneered by Tedrow and
magnetic electrodes. This is the result that has been pre- Moservey is based on the spin-polarized tunneling across an
dicted by the Jullière model [26]. Assuming a typical value FM-superconductor tunnel junction which gives a submilli-
of 0.5 for P1 and P2 , this simple model predicts a JMR electron volt energy resolution [300]. Recently, a new tech-
of 40%. Note that an alternative definition for the tun- nique based on the point contact between the sample and
nel magnetoresistance (TMR) frequently used in the lit- a superconductor has been developed [301 302]. This tech-
erature is TMR = 2P1 P2 /1 − P1 P2 . Obviously, the TMR nique measures the polarization ratio using Andrew reflec-
is always larger than the JMR, although the actual size tion. The point contact setup allows the measurement of
of the magnetoresistance effect is the same. Although the samples in various forms, and it also does not require the use
Jullière model predicts a quite large MR, it was not until of a magnetic field. The values of P measured for some of
1995 that MR ratios of about 20% were first reported by the magnetic materials using both the point-contact Andrew
two groups for magnetic tunnel junctions at room temper- reflection technique and tunneling techniques [301–305] are
ature [28 29]. Miyazaki and Tezuka [28] reported on an shown in Table 4 (note that spin polarization of Co and Ni
Fe/Al2 O3 /Fe tunnel junction, 1 × 1 mm2 in area, that exhib- should be negative in the processes where 3d-band electrons
ited a TMR ratio of 18% at room temperature. The barrier are dominant). Among them, NiMnSb, La07 Sr03 MnO3 , and
layer was formed through natural oxidation of 5.5 nm Al CrO2 are so-called half metals, which theoretically should
in air for 24 h. At about the same time, Moodera et al. have a polarization of 100%. As shown in Table 4, most
fabricated magnetic tunnel junctions with CoFe and Co or soft magnetic materials have a polarization of 40–50%, cor-
NiFe as electrodes and oxygen-plasma oxidized Al as the responding to a JMR of about 30–40% (or a TMR of
barrier [29]. An MR ratio of 18% (in this case, JMR) has 38–67%). Theoretically, the sign of JMR can either be posi-
been obtained at low-bias voltage and room temperature in tive or negative, depending on the signs of P1 and P2 . How-
CoFe/Al2 O3 /Co junctions with an area of 6 × 10−4 cm2 . The ever, most of the experiments using NiFe, CoFe, Co, and Fe
large MR ratio was attributed to the smoothness and small as the electrodes and Al2 O3 as the barrier have produced
thickness of the electrodes. In both works, the magnetic tun- positive MRs. Sharma et al. observed a negative MR in
nel effect was observed in large-area FM/Al2 O3 /FM junc- NiFe/Al2 O3 /Ta2 O5 /NiFe composite barrier junctions, which
tions fabricated by shadow masks. Gallagher et al. reported was attributed to the different spin polarizations of NiFe at
on the fabrication of microstructured magnetic tunnel junc- the two different interfaces, one with Al2 O3 and the other
tions using a simple self-aligned lithographic process adopt- with Ta2 O5 [306]. Negative MR has also been observed by
ing an exchange-biased structure [299]. The combination De Teresa et al. in La07 Sr03 MnO3 (LSMO)/barrier/Co tun-
of photo and electron-beam lithographies made it possible nel junctions, when SrTiO3 (STO) was used as the insulating
to cover a range of junction areas spanning five orders of barrier [307]. This is in good agreement with the fact that
magnitude (10−2 –103 1m3 . The MR ratio obtained ranges the spin polarization of LSMO is positive [308], while that
Nano Spintronics for Data Storage 33

Table 4. Measured values of polarizations for different materials. transport methods. A similar type of study was also carried
out by Li et al. using high-resolution transmission electron
Materials Polarization (%) Ref. microscopy [316]. With an in situ scanning tunneling micro-
Co 42 ± 2 [301] scope, Tegen et al. directly measured the roughness of the
37 ± 2 [302] films, and found a close correspondence between the val-
Ni 465 ± 1 [301] ues for the coupling fields determined by the magnetoopti-
32 ± 2 [302] cal Kerr effect and the ones computed on the basis of the
Fe 45 ± 2 [301] measured morphology parameters [317]. The authors con-
Ni80 Fe20 37 ± 5 [301] firmed an increase of the dipole coupling between the mag-
Co50 Fe50 53± [301] netic layers with decreasing barrier thickness, as predicted by
NiMnSb 58 ± 23 [301] the theoretical model [Eq. (25)]. In addition to topographic
La07 Sr03 MnO3 78 ± 4 [301] roughness, Tiusan et al. showed that, in samples involving
La2/3 Sr1/3 MnO3 95 [303] polycrystalline magnetic films, beyond the orange-peel cou-
CrO2 90 ± 36 [301] pling, an important class of interaction is related to the dis-
Fe3 O4 −39 [304]
persion fields associated with magnetic inhomogeneities, or
Sr2 FeMoO6 60 ± 3 [305]
magnetic roughness, arising from the local anisotropy fluctu-
ations [318]. The roughness problem becomes more serious
when fabricating MTJs for read-head applications due to the
of the 3d-band density of states of Co at the Fermi level
roughness of the lower shield layer. Sun et al. reported on
is negative. However, when Al2 O3 was used as a barrier,
the use of a gas cluster ion beam to smooth the shield layer,
the MR turned back to positive, and with a very different
bias dependence. The authors concluded that the electronic and successfully obtained an RA as low as 3.6–6.5 1m2 and
structure of the barrier and the barrier–electrode interface an MR ratio of 14–18% for MTJs grown on the smoothed
has a strong influence on the spin polarization of electrons. shield layers [319].
Similar results have also been observed for magnetic tunnel
junctions using Ni40 Fe60 and Fe as the electrodes and STO as 6.1.3. Effect of Nonmagnetic
the insulating barrier [309], and in Fe3 O4 /STO/LSMO [310] Interfacial Layers
and CrO2 /natural barrier/Co [311] tunnel junctions. These A spin-dependent quantum well forms in an ultrathin non-
results imply that the barrier material and/or the electronic magnetic metal sandwiched either from both sides by mag-
structure of electrode/barrier interface play an important netic materials or from one side by a magnetic material and
role in determining the sign of the spin polarization. This the other side by an insulator or vacuum [295 320 321]. In
is consistent with the ab initio calculations of de Boer et al. the former case, quantum-well states exist only when the
on the Co(001)–HfO2 interface [312], Oleinik et al. on the magnetic layers are in parallel alignment, and in both cases,
Co/Al2 O3 /Co interface [313], and Tsymbal et al. on the Fe/O they are formed for minority electrons only. The observation
surface [314], which suggest that the polarization of tunnel- of quantum-well states, however, requires atomically sharp
ing electrons is governed by the mechanism of electronic interfaces due to the short wavelength of electrons in met-
bonding at the metal/oxide interface. als. For an ideal magnetic tunnel junction with a structure
FM/NM/insulator/FM, theories predict an oscillation in the
6.1.2. Surface Roughness (Bottom Electrode) MR effect as a function of the NM layer thickness because
One of the major problems that resulted in the failed the spin polarization of the tunneling electrons oscillates as a
attempts to produce high-MR MTJs before the early 1990s result of the resonant tunneling via the quantum-well states
is related to the surface roughness of the first or bottom [322–324]. Moodera et al. fabricated Co/Au/Al2 O3 /Ni80 Fe20
electrode on which the insulating barrier and top electrode MTJs on liquid-nitrogen-cooled glass substrates with the
are formed. If the surface roughness exceeds a certain crit- thickness of Au varying from 0.1 to 1.2 nm [325]. It was
ical value, the MTJ will fail either magnetically or elec- found that the JMR decreases with the thickness of Au,
trically or in both ways. The former is mainly caused by which might be understood as the consequence of decrease
the dipole or orange-peel coupling between the bottom of spin polarization [326–328]. But what was interesting was
and top FM electrodes, while the latter is caused by pin- that a negative JMR was observed in the thickness range of
holes formed in the thin insulating barrier. Schrag et al. 0.6–0.8 nm, which came along with an unusual bias depen-
reported on measurements of the magnitude of Néel dence. The authors have shown that both results could be
orange-peel coupling due to interface roughness in a series explained by taking into account the quantum-well states
of magnetic tunneling junction devices [315]. The samples formed in the NM layer based on the Slonczewski model
studied were Si(100) substrate/Ta/Al/NiFe/FeMn/Co(PL2)/ [329]. LeClair et al. also observed an inversion of sign of the
Ru/Co(PL1)(pinned)/Al2 O3 (barrier)/NiFe(free)/Al/Ta with MR in Co/Ru/Al2 O3 /Co tunnel junctions, but they attributed
the thickness of the barrier, free, and pinned layers vary- this to a strong density-of-states modification at the (inter-
ing from sample to sample. Results from magnetometry and diffused) Co/Ru interface [330].
transport measurements are shown to be in good agreement Yuasa et al. fabricated MTJs on single-crystal Cu
with the theoretical model of Néel. In addition, the authors substrates of the structure Co(001)/Cu(001)/Al–O/Ni80 Fe20
have also used transmission electron microscopy to directly using molecular beam epitaxy, and they observed for the first
probe the sample interface roughness, and obtain results time clear oscillation of the JMR as a function of the Cu
consistent with the values obtained by magnetometry and layer thickness [296]. High-resolution transmission electron
34 Nano Spintronics for Data Storage

microscopy revealed that both the bottom electrode and the hand, in the case of natural oxidation, it is difficult for oxy-
Cu layer were grown epitaxially on the Cu substrate, leading gen to penetrate into the grain boundaries. Therefore, the
to atomically flat interfaces between Co and Cu and Cu and as-deposited MTJs with barrier layers formed by plasma oxi-
Al–O. As expected, the oscillation period at zero bias was dation normally show a higher MR as compared to those
about 1.1 nm, which is consistent with the value observed with naturally oxidized barriers. Mitsuzuka et al. studied the
in GMR oscillation. As the spin polarization of electrons in interface structure of magnetic tunnel junctions with natu-
Co and Ni80 Fe20 was found to be positive in tunnel junc- rally oxidized AlOx barriers using X-ray photoelectron spec-
tions with Al–O as the barrier, the negative peaks of the troscopy (XPS) [390]. The MTJs studied had a structure Fe
JMR observed in these junctions were attributed to the neg- (50 nm)/AlOx /CoFe (30 nm), where the barrier layer was
ative polarization of electrons formed in the Cu quantum formed through natural oxidation of Al with a thickness
wells. ranging from 0 to 5 nm. It was found that the MR showed
a maximum value when Al is about 2–3 nm in thickness.
6.2. Barrier Materials The XPS analysis showed that an Al layer thicker than 1 nm
covers the entire surface of the lower electrode. However,
6.2.1. Different Kinds of Barriers unoxidized Al remains when Al is thicker than 1 nm. For Al
The key to fabricating a high-quality MTJ is to form layers greater than 3 nm, the MR ratio is strongly affected by
a pinhole-free and uniform ultrathin barrier layer of unoxidized Al, probably due to the decrease of spin polariza-
AlOx [28 29 298 331]. Although other types of insulating tion caused by the unoxidized Al layer. Zhang et al. reported
materials have also been investigated as the tunnel bar- on the effect of natural oxidation conditions on the per-
rier, such as Ta2 O5 [332–335], AlNx Oy [336–339], ZrAlOx formance of MTJs. It was found that natural oxidation at
[340 341], GaAs [342], GaOx [343], diamond-like carbon a lower oxygen pressure produces an MTJ with a lower
[344], ZnS [345 346], SrTiO3 [347–350, 307, 351], MgF2 RA and a smaller MR ratio as compared to those formed
[352], ZnSe [353], Cu3 N [354], MgO [355–359], HfOx by a higher pressure oxygen oxidation [368]. Rutherford
[360–362], TaOx /AlOx [306 363], so far, AlOx was found to backscattering analysis confirmed the inadequate oxidation
be the most suitable barrier material for MTJ. of Al at low oxygen pressure. Natural oxidation and UV-
assisted oxidation are suitable for formation of MTJs of a
6.2.2. Formation and Characterization of AlOx low area–resistance product using ultrathin Al layers [391].
It was reported that a two-step or multiple steps of natural
The AlOx layer can be formed using various methods, such as oxidation helps to improve the uniformity of the oxidized
natural oxidation [28, 364–370], plasma oxidation [29, 371– barrier layer [369 392].
377], ultraviolet (UV)-light-assisted oxidation [378–381], ion
beam oxidation [382], radical oxidation [383], ozone oxida- Pinholes The quality of the barrier layer plays a domi-
tion [384], atomic layer chemical vapor deposition [385], and nant role in determining the characteristics of the tunnel
other equipment-specific techniques. In general, plasma or junction [393]. Allen et al. reported on the imaging of pin-
other energy beam-based techniques are faster and produce holes by electrochemical decoration of the pinholes using
denser and more uniform oxides compared to natural oxi- Cu [394]. The results obtained were found to be consistent
dation, either in air or in pure oxygen. The property of the with the breakdown voltage analysis [395]. Dimopoulos et al.
oxide formed is also dependent on the types of the process reported that the “defective” sites of the barrier can also
gases. Using Kr–O2 plasma, Tsunoda et al. obtained a TMR be detected using a barrier impedance scanning microscopy
of 58.8% at room temperature after annealing the junc- technique [396].
tion Ta5/Cu20/Ta20/Ni–Fe5/Cu5/Mn75 Ir25 10/Co70 Fe30 2.5/Al– Inhomogeneity and Defects of the Barrier Layer Ando
O/Co70 Fe30 2.5/Ni–Fe10/Cu20/Ta5 (thickness in nanometers) et al. studied the local current distribution of MTJ with
at 300
C, while the achieved TMR ratio of the MTJ fabri- a structure Ta5/Ni80 Fe20 5/IrMn15/Co5/Al1.3-oxidation using
cated with the usual Ar–O2 plasma remained 48.6% [386]. conducting atomic force microscopy [397]. Here, thickness
This improvement is remarkable considering the fact that is given in nanometers. It was found that the current distri-
the MR obtained so far using other oxidation methods is bution can be explained by assuming a Gaussian distribution
about 50% [387]. for the barrier thickness with a mean value of 1.2 nm and
Under Oxidation of Al High-resolution transmission elec- a standard deviation of 0.1–0.15 nm. A similar study was
tron microscopy study by Bae et al. [388] revealed that also carried out by Luo et al. [398]. Rippard et al. used bal-
natural oxidation occurred preferentially through the grain listic electron emission microscopy to study thin aluminum
boundary of Al grains, leading to isotropically expanded oxide tunnel junction barriers formed both by magnetron
AlOx grains when Al is fully oxidized. On the other hand, sputter deposition and thermal evaporation [399, 400]. It
in plasma oxidation, a flat AlOx layer formed uniformly on was found that the barriers made by oxidation of evapo-
the Al layer, leading to a sharp interface with the underlying rated Al become fully formed at a thickness as small as
metallic Al layer. Ando et al. conducted a systematic study 0.6–0.7 nm, while a thicker Al (0.9–1.1 nm) layer is nec-
of the microstructure of the barrier oxide formed using dif- essary to form a continuous barrier by magnetron sputter-
ferent methods [389]. It was found that it is easier to form ing. Although the decrease of barrier thickness generally
a uniform oxide within a short time period using plasma contributes to the reduction of RA, if the barrier is too
oxidation. This is because, in addition to the full cover- thin, it will also cause reliability problems when the MTJs
age of the Al surface, the energetic oxygen species are also are used as read sensors or storage cell elements. Oliver
able to penetrate into the grain boundaries. On the other et al. reported on dielectric breakdown studies on magnetic
Nano Spintronics for Data Storage 35

tunnel junctions having ultrathin barriers [401]. The MTJs, 200–250
C. For annealing at higher temperatures, both
with a structure Ta5/PtMn25/CoFe2.2/Ru0.9/CoFe2.2/Al- the MR and the resistance–area product decrease mono-
oxidation/CoFe1/NiFe2.5/Ta15 (thicknesses are in nano- tonically, although the MR decreases at a more rapid rate.
meters), were fabricated by magnetron sputtering and MR values of ∼10% were still obtained after annealing to
patterned by deep ultraviolet photolithography to a junc- ∼350
C. Due to the thick barriers used in these particular
tion area of about 0.2 1m2 . The tunnel magnetoresistance samples, the resistance–area product is in the range 5–6
was 15–22% and RA 7–22 < · 1m2 for junctions having a M< · 1m2 .
4.75–5.5 Å Al layer oxidized naturally. The authors observed Sousa et al. reported on the thermal annealing effect
two types of breakdown: abrupt dielectric breakdown at an of exchange-biased MTJs. Two series of samples were
effective field of 10 MV/cm determined by the thickness studied: one series with resistance–area products of
of the tunnel barrier, and a gradual breakdown related to 10–13 M< · 1m2 , and the other with resistance–area
defects in the tunnel barrier. After the breakdown, a metallic products of 25–30 k< · 1m2 [405]. The former has a struc-
pinhole is created, the size of which depends on the max- ture Ta7/Cu4/Ta7/NiFe6/CoFe3/Al2 O3 /CoFe3/MnRh18 or
imum current applied to the junction. The current flowing TbCo12/Ta3, and the latter has a structure Ta7/NiFe10/
through the pinhole creates a strong circular magnetic field CoFe2/Al2 O3 /CoFe4/MnR17/Ta3. The barriers of samples
that curls the local magnetization in the free layer around with higher resistance–area products were formed deposit-
the pinhole. The subsequent free-layer reversal is very sen- ing 1.8 nm Al, followed by 90 s plasma oxidation with a
sitive to the pinhole location. The electric properties after density of 6 mW/cm2 , and those of the lower resistance–
breakdown can be well described by an ohmic resistor and a area product samples were fabricated depositing 1.1–1.3
tunnel magnetoresistor connected in parallel. Kikuchi et al. nm Al, and then followed, respectively, by 20 and 15 s
also reported similar types of phenomena in the breakdown plasma oxidation at a power density of 4 mW/cm2 . The
processes [402]. The existence of pinholes in low-junction- junction tunneling magnetoresistance was found to increase
resistance samples has been discussed in the context of a from 22–26% in high-resistance samples and 22–37% in
size-dependent breakdown voltage that decreases by increas- low-resistance samples, upon anneal up to 200–230
C.
ing the junction size. Rutherford backscattering analysis suggested that anneal
improved the asymmetry in both the oxygen distribution
inside the barrier and the junction parameters. The MR
6.3. Annealing Effect and Thermal Stability increase in lower resistance samples was attributed to the
Postgrowth annealing at temperatures in the range of 220– increase of the barrier height. The increase of the barrier
400
C was found to be effective in improving the perfor- height and decrease of the barrier thickness were also
mance of magnetic tunnel junctions, beyond which anneal observed by Nowak et al. [406].
will deteriorate the performances of MTJ [403]. The highest Ando et al. studied the annealing effect on MTJs
temperature at which the MR starts to decrease monotoni- having the structure Ta/(Cu,Pt)/Fe20 Ni80 /IrMn/Co75 Fe25 /Al-
cally with temperature is highly dependent on the materials oxide/Co75 Fe25 /Fe20 Ni80 /Ta [407]. When the Al thickness,
and device structures. This has a significant impact on the oxidation time, and annealing temperature were 0.8 nm,
application of MTJs in MRAMs because standard back-end 15 s (10 s), and 300
C (250
C), the TMR, RA, and
technology for the metallization of CMOS circuits requires TMR enhancement ratio against the as-grown samples
annealing in the forming gas at 350–450
C. Many attempts were 49% (31%), 1.1 k< · 1m2 (230 < · 1m2 , and
have been made to reveal the mechanisms for MR enhance- 70% (100%), respectively. In order to investigate the
ment at intermediate temperatures and the loss of the MR annealing temperature dependence of the TMR ratio,
at elevated temperatures. We summarize some of the results the authors measured the local electrical properties for
reported in the literature below. a Ta/Fe20 Ni80 /Pt/Fe20 Ni80 /IrMn/Co75 Fe25 /Al-oxide multilayer
using the conductive atomic force microscopy technique.
The current image became very homogeneous after anneal-
6.3.1. Effect on the Barrier Layer ing at around 300
C for 1 h. The increase of the TMR ratio
Anneal at moderate temperatures was found to improve of the junction after annealing was thus attributed to both
the quality of the barrier layer through decreasing its the increase of barrier height and the decrease of barrier
thickness, increasing its barrier height, improving its homo- height fluctuation. The degradation of TMR after annealing
geneity at interfaces with the electrodes, and reducing in excess of 350
C was due to the decrease in barrier height,
the defects inside the barrier. Parkin et al. reported on leading to the increase of leakage currents.
the thermal annealing study of an MTJ with a structure Schmalhorst et al. systematically investigated the struc-
Cr80 V20 25/Co75 Pt12 Cr13 15/Al–plasma oxidation/Co15/Al20 tural, magnetic, magnetotransport, and tunneling properties
or Co88 Pt12 15/Al20 [404]. Here, the layer thickness is in of CoFe(1.5 nm)/Ru(0.9 nm)/CoFe(2.2 nm)/Al2 O3 /Ni81 Fe19
units of nanometers. It was found that the MTJ is thermally (6 nm)/Ta (5 nm) junctions for different Al thicknesses and
stable at temperatures greater than 300
C. A comparison oxidation times after isochronal annealing up to 500
C. The
of cross-section transmission electron micrographs of an mean breakdown voltage of the junction increases with tem-
untreated sample and a similar one annealed at 350
C perature, saturates at an annealing temperature of about
indicates that the thickness of the amorphous tunnel barrier 300–350
C, and remains constant up to almost 500
C. This
is slightly decreased after annealing. The resistance–area implies that Al oxide is extremely stable, and its stability is
product and the MR of the devices initially increase slightly improved by thermal annealing. The latter might be related
for lower temperature annealing treatments up to about to the defect reduction inside the barrier layer [408].
36 Nano Spintronics for Data Storage

6.3.2. Effect on Electrodes Mn from the antiferromagnetic layer through thin exchange-
Dimopoulos et al. reported that, compared to Co/Ru/Co, biased ferromagnetic layers to the tunnel barrier is observed
Co/Ru/Co50 Fe50 is more advantageous when being used as at elevated temperatures (>300
C). Yoon et al. investigated
the pinned layer in MTJs due to its excellent thermal stabil- the Mn diffusion in an NiFe/MnIr/CoFe/AlOx multilayer
ity [409]. After successive annealing steps up to 400
C, mag- after annealing at 300
C using Auger electron spectroscopy
netic tunnel junctions with structures Cr (1.6 nm)/Fe (6 nm)/ and X-ray photoelectron spectroscopy, wherein it was found
Cu (30 nm)/Co (1.8 nm)/Ru (0.8 nm)/Co50 Fe50 (2.8 nm)/ that the magnitude of Mn diffusion is correlated to the
Al (1 nm)–plasma oxidation/CoFe (1 nm)/ Fe (6 nm)/ excess oxygen generated from the plasma oxidation of the
AlOx tunnel barrier [416]. The analysis showed that Mn
Cu (5 nm)/Cr (3 nm) as pinned layers were found to still
diffusion was driven by the preferential oxidation of Mn
present a significant tunnel magnetoresistance of nearly
at the oxide insulator interface due to its larger (negative)
20%, and an almost intact magnetic rigidity of the hard
enthalpy of formation compared to iron and cobalt oxides.
magnetic system. The improvement in thermal stability was
Saito et al. reported on the thermal stability study of
attributed to the improvement of the Ru/CoFe/AlOx inter-
dual-spin-valve-type double-tunnel junctions with structures
faces in terms of interdiffusion.
Ni–Fe/Ir–Mn/Co50 Fe50 /AlOx /Co90 Fe10 /AlOx /Co50 Fe50 /Ir–Mn/
Due to the finite and negative enthalpies of formation of
Ni–Fe and Ni–Fe/Ir–Mn/Co50 Fe50 /AlOx /Co50 Fe50 /Ni–Fe/
most transition metal oxides, the electrodes will also be par-
Co50 Fe50 /AlOx /Co50 Fe50 /Ir–Mn/Ni–Fe, fabricated using
tially oxidized during annealing, as revealed by X-ray pho- photolithography and ion-beam milling [417]. In order to
toelectron spectroscopy studies. However, the mechanism clarify the mechanism of the loss of the MR ratio and that
still remains unclear regarding how the partially oxidized of V1/2 (bias voltage at which the MR decreases to half of
interfacial layers are related to the enhancement or degra- the zero-bias value) above 320
C, the authors carried out
dation of the tunnel magnetoresistance [410]. The oxidation an XPS study in Ni81 Fe19 /Ir–Mn/Co50 Fe50 /AlOx multilayers
of the electrode might be suppressed, if it has adverse influ- annealed at various TA . The Mn oxide peaks in addition to
ences on the MTJ, by inserting a thin layer of magnetic the Mn peaks (both 2p1/2 and 2p2/3 peaks were observed)
oxide at the barrier/electrode interface. During the anneal- were observed in the temperature range ≥300
C. It was
ing process, due to the much larger and negative enthalpy of found that a small amount of Mn and Mn oxide reached the
AlOx as compared to most magnetic oxides and the larger CoFe/AlOx interface at TA = 300
C due to interdiffusion
affinity of oxygen for Al, oxygen will tend to move to the of Mn. The oxygen redistribution and homogenization occur
barrier layer instead of the electrode. Using this technique, between AlOx and Mn in the range of TA > 320–350
C,
Zhang et al. fabricated tunnel junctions with an interposed leading to an increase of defect states in the barrier. This
Fe oxide layer between the Al2 O3 barrier (tAl = 8–9 Å) increase in defect states above 320–350
C was assumed
and the top CoFe pinned layer, and obtained a large tun- to be responsible for the decrease of both the MR ratio
neling magnetoresistance of 39% after 40 min anneal at and V1/2 due to the spin-independent two-step tunneling
380
C [411 412]. The annihilation of the CoFe oxide dur- via defect states in the barrier. The Mn diffusion might be
ing postgrowth annealing was also observed by Dimopoulos suppressed by increasing the pinned layer thickness [418].
et al. in Cr (1.8 nm)/Fe (6 nm)/Cu (30 nm)/Cr (1.8 nm)/Fe
(6 nm)/Cu (30 nm)/Co50 Fe50 (1 nm)/Al–plasma oxida- 6.4. Basic Characteristics of MTJ
tion/Fe (6 nm)/Cu (10 nm)/Cr (5 nm) samples, leading to
the improvement of aluminum oxide’s stoichiometry [413]. A practical exchange-biased magnetic tunnel junction has a
Matsukawa et al. fabricated MTJs with structures of bottom structure that is very similar to that of a spin valve, except
electrode/PtMn/CoFe/Ru/ CoFe/Fe1–x Ptx /Al oxide/Fe1–x Ptx / that the Cu spacer is replaced by an ultrathin insulating layer
NiFe/top electrode (x = 005–0.75), in areas from 2 × 2 to (see inset of Fig. 41). Figure 41 shows a schematic of a typi-
30 × 30 1m2 , using conventional photolithography and the cal MTJ structure. When measuring the magnetic response,
ion-milling method. After a postgrowth anneal at 400 and a constant current is applied through one of the top and
420
C, the TMRs of MTJs with Fe1–x Ptx (x = 01–0.2) were bottom electrodes, and the voltage drops across the other
still about 40 and 30%, respectively [414]. two terminals are measured when a magnetic field is swept
within a certain range from a positive value to a negative
value. However, caution must be taken when interpreting
6.3.3. Diffusion of Mn the data because the measured MR could be affected sig-
In addition to oxidation of electrodes, overoxidation of nificantly by the current distribution effect [419–424].
Al will also cause oxidation-induced diffusion of Mn from Although the MR ratio is much higher than that of a
the pinning layer to the barrier layer during the post- typical spin valve, the MTJ does have a few drawbacks com-
growth annealing. Samant et al. investigated the thermal pared to spin valves when they are used as read sensors.
stability of MTJs with a structure Si(100)/0.5 1m thick
SiO2 /Ti5/Pd15/Ir23 Mn77 or Fe50 Mn50 14/Co80 Fe20 2.4/Al1.3–

Proof's Only
plasma oxidation/Co80 Fe20 3/Pd1.5 (thickness is given V
in nanometers) using near-edge X-ray absorption fine- I
structure spectroscopy [415]. It was found that structures
with IrMn antiferromagnetic exchange-bias layers are much AFM PL I FL

more thermally stable than similar structures with FeMn

exchange-bias layers. In either case, however, diffusion of Figure 41. Typical structure of a magnetic tunnel junction device.
Nano Spintronics for Data Storage 37

These include: (1) large junction resistance, (2) bias volt-

Normalized MR ratio
1.2
age dependence of the MR, and (3) temperature depen- 1
0.8
dence of MR. The junction RA of an MTJ can be as high 0.6
as on the order of M< · 1m2 . Although it can be lowered 0.4

by reducing the barrier thickness and height, this will gener- 0.2
0
ally result in a lower MR ratio. For example, the MR ratio 0 200 400 600

drops to about 13% when the RA is reduced to 5 < · 1m2 Bias Voltage (mV)

[425]. As an example, Figure 42 shows the magnetoresis-

tance curve of a typical MTJ with a junction area of 80 1m × Figure 43. Bias voltage dependence of the MR ratio of an MTJ
with the structure Ta20/Ni81 Fe19 20/Co90 Fe10 2/Al(1 nm)-oxide/Co90 Fe10 2/
80 1m fabricated in the author’s group. For this specific
Ni81 Fe197/Fe50 Mn50 20/Ta5. Here, the thicknesses are given in
device with a structure Ta20/Ni81 Fe19 20/Co90 Fe10 2/Al(1 nm)- nanometers.
oxidation/Co90 Fe10 2/Ni81 Fe19 7/Fe50 Mn50 20/Ta5 (the thick-
nesses of other layers are given in nanometers), the TMR
ratio is 44.6% at a dc bias of 4.3 mV and the RA is about for reducing the bias voltage dependence is to use a double-
250 k< · 1m2 [426]. Shown in Figure 43 is the bias volt- tunnel junction [439]. Using the double junctions, it is pos-
age dependence of the MR ratio of this particular MTJ. sible to raise the voltage at which the MR drops to half of
As can be seen from the figure, the MR drops to almost the low-bias value to 0.9 V, which is almost double that of
half when the bias voltage is increased to about 0.4 V. a single-tunnel junction [440]. The voltage dependence can
Although this value does vary with the MTJ structure, it also be reduced using Hf-inserted AlOx barrier layers [441].
is roughly in the range of 0.3–0.5 V for most of the MTJ In addition to voltage dependence, the MR ratio of MTJ
structures [427–435]. Although the mechanism is still not was also found to decrease with temperature [430, 405, 442–
fully understood, it is believed that it is caused by both the 446]. According to Shang et al. [442], Hagler et al. [444],
bias-induced change in the barrier profile and the energy- and Wingbermühle et al. [445], the temperature-dependent
dependent density of states for both the majority and minor- total conductance of an MTJ can be written as
ity electrons [436 437]. It is argued that magnon excitation
G = GDI T /1 + P1 T P2 T  cos0 + GIN T  (44)
in the magnetic layer might also be one of the possible rea-
sons [438]. The energy-dependent density of states model where GDI T  indicates the direct tunneling conductance
can be roughly understood as follows. From Eq. (41), one and GIN T  represents the spin-independent conductance
can find that via localized states,  is the angle between the magnetiza-
JMR 2 P1 P2  tions of the two electrodes, and P1 T  and P2 T  are the
= (43) temperature-dependent spin polarizations. The temperature
V 1 + P1 P2 2 V
dependences of GDI T  and spin polarizations [P1 T  and
which reduces to JMR/V  = 4P /1 + P 2 2 P /V , P2 T ] are due to the temperature dependences of Fermi
when P1 = P2 = P . For electrodes with positive spin distribution and magnetizations, respectively, and are given 
polarizations, the JMR will decrease with voltage when by GDI T  = CG0 T /sinCT  with C = 139 × 10−4 d/ B,
P /V  < 0. Substituting Eq. (42) into P /V  < 0, one and P1 T  = P01 1 − A1 T 3/2  P2 T  = P02 1 − A2 T 3/2 . Here,
obtains D↓ D↑ /V  − D↑ D↓ /V  < 0, which can be A1 A2  is a material-dependent parameter, d is the barrier
further simplified to  lnD↑ /D↑ /V  < 0. This implies thickness in angstroms, B is the barrier height in electron-
that the JMR will drop if the ratio between the density of volts, and G0 is the value of GDI T  at zero degrees Kelvin.
states for majority electrons and that of the minority elec-
trons decreases with the bias voltage. As we will discuss 6.5. Applications of MTJ in Read Head
shortly, the decrease of JMR with bias voltage is undesirable
Magnetic tunnel junctions can be used either as read sensors
for practical applications. One of the possible approaches
in hard disk drives [447–453] or memory cells in MRAMs.
Following are some advantages and drawbacks of MTJs
65 compared to spin valves when they are used as read sensors.
65 TMR-44.5% Voc-4.3mV
60µmx40µm
60 60

6.5.1. Large Output Signal

Resistance (Ω)
Resistance (Ω)

55

Proof's Only For an MTJ sensor element with width W , height H , and
55 50

45

50 40
thickness T , the maximum output voltage is given by
-40 -20 0 20 40
Field (Oe)
I:RA
45
:VMTJ = = J:RA (45)
W ·H
40
-0.4 -0.2 0.0 0.2 0.4
where J is the current density, and :RA is the change in
Magnetic Field (kOe)
resistance–area product of the MTJ. The maximum output
Figure 42. MR–H curve measured at room temperature for an MTJ voltage refers to the value that is obtained when the magne-
with a structure Ta20/Ni81 Fe19 20/Co90 Fe10 2/Al(1 nm)-oxide/Co90 Fe10 2/ tizations of the two FM layers are switched from a parallel
Ni81 Fe19 7/Fe50 Mn50 20/Ta5. Inset shows the minor curve. Here, the thick- alignment to an antiparallel alignment or vice versa. The
nesses are given in nanometers. Reprinted with permission from [426], actual output is normally a fraction of this due to the limited
K. B. Li et al., J. Magn. Magn. Mater. 241, 89 (2002). © 2002, Elsevier. dynamic range of the read sensor and the head efficiency
38 Nano Spintronics for Data Storage

factor. In practice, however, the output is limited by the 10000

Junction resistance (Ω)

characteristic dependence of JMR on the dc bias voltage, RA = 1Ωµm2

that is [454], 1000

:VMTJ V  = VB · MR0 · Vmax − VB /Vmax (46)

Proof's Only 100 100Ω

500Gb/in2

200Gb/in2

100Gb/in2
1Tb/in2
where VB is the bias voltage, MR0 is the MR ratio at zero 10
bias, and Vmax is the voltage at which the MR becomes zero. 0 50 100 150 200 250

In deriving the above equation, we have assumed that the Reader width (nm)
MR ratio decreases linearly with the bias voltage. The max-
imum output signal is thus given by Figure 44. Junction resistance of MTJ as a function of reader width
(see text for the assumptions that have been used to calculate the
1 resistance).
:VMTJ VB = Vmax /2 = V · MR0  (47)
2 max
Assume that MR0 = 5–10% at device level, and Vmax = 6.5.4. Noises in MTJs
800 mV; it gives an output voltage of 20–40 mV. On the In addition to thermal noise, the MTJ also suffers from shot
other hand, the maximum output signal of a CIP spin valve noise which increases with the sensing current [425 455],
with thickness T , width W , current density J , and sheet 1/f , and/or telegraph noise [457–461]. This makes the
resistance change :RS is given by signal-to-noise ratio of MTJ smaller than that of the spin
valve, even when the output signal is the same. The shot-
:VSV = J · :RS · T · W  (48) noise voltage is given by VS = 2eVRB1/2 , while that of
thermal noise is given by Vth = 4kB T RB1/2 . Here, e is the
Using typical values of J = 1×107 A/cm2 , :RS = 4 <, T = elementary charge of electrons, V is the bias voltage, R is
30 nm, and W = 1 1m, the maximum output voltage will the resistance of the sensor element, B is the bandwidth,
be given by 12 mV. Of course, the actual output signal for kB is the Boltzmann constant, and T is the temperature.
both cases when being used as a read head is much smaller Assuming V = 400 mV and T = 300 K, one can easily find
than these values due to the low head efficiency. We can see that VS /Vth = 28. The dominance of shot noise in the MTJ
that the maximum output signal of an MTJ is much higher was found by Shimazawa et al. in their front flux guide-
than that of a typical spin valve. Moreover, the output of the type magnetic tunnel junction heads [448]. George et al.
spin valve scales with its width. Therefore, the MTJ head reported that, depending on the quality of the barrier, the
becomes more advantageous than the CIP spin-valve head measured value of shot noise in an MTJ can be much lower
in terms of output signal when the track width shrinks into than that predicted by theory [462]. In tunnel junctions with
the submicron regime, which is already the case in the latest pinholes, the shot-noise voltage is reduced by a factor of
products. /RP /RT + RP 01/2 , with RT and RP the ideal tunnel resis-
tance and the lumped parallel resistance due to the con-
6.5.2. Current-Perpendicular-to-Plane ductive paths of pinholes, respectively. Unfortunately, the
decrease of shot noise due to pinholes does not necessar-
Operation ily lead to the increase of SNR because the pinholes also
As is with the case of CPP GMR sensors, it is possible to result in a decrease of JMR. Although it is not specific to
remove the insulating gap in the MTJ heads, and thus they MTJ, all types of MR heads also exhibit magnetization noise
are more suitable for higher linear density heads. which may exceed the thermal noise when the sensor volume
becomes very small [463].
6.5.3. Large Junction Resistance Using the latest MTJ heads, Mao et al. [464] demon-
The primary drawback of MTJ is its large junction resis- strated an area density of 120 Gbits/in2 . Further optimiza-
tance, in particular, when its size shrinks. The large junction tion in materials, structures, and fabrication processes may
resistance will result in both a small bandwidth and a low improve the MTJ technology further, and render them suit-
SNR of the read signal [455]. For high-frequency operations, able for higher areal density applications.
the sensor resistance is required to be less than 100 < [425].
Figure 44 shows the dependence of the junction resistance
as a function of the reader width by assuming that the sen-
7. MAGNETIC RANDOM
sor has a square shape and the resistance–area product is ACCESS MEMORY
1 < · 1m2 . Also shown in the figure are the estimated sensor
7.1. Introduction
widths for applications at different recording densities. It is
clear that the MTJ may not be suitable for read heads at Although spin valves and MTJs have been invented for
an areal density of more than 200 Gbits/in2 . Jury and Wang hard disk drive applications, they have also found important
reported on a possible approach to extend the bandwidth of applications in memory devices. Computer memories featur-
the MTJ head using a simple buffer amplifier, which leads ing the following features are always highly demanded: (1)
to an SNR improvement of 5–10 dB over a no-buffer config- nonvolatility, (2) high density, (3) short cycle time, (4) low-
uration for frequencies between 100 MHz and 1 GHz, with power consumption, (5) low cost, (6) high reliability, (7) infi-
a slight reduction of SNR at low frequencies [456]. nite lifetime, and (8) radiation hard (for special-purpose
Nano Spintronics for Data Storage 39

applications). A variety of memories have been com- bit line

mercialized. These include DRAM, SRAM, ROM, Flash MTJ

EEPROM, and FeRAM. However, none of them is able to word

Proof's Only
line
satisfy all of the requirements. Furthermore, most of these
memories (e.g., DRAM) store information using charges; it Bit line
is difficult to perform down-size scaling once it reaches a Selected
ga
te

certain dimension due to the degradation of SNR. MRAM cell source drain

has been studied intensively as one of the promising candi- Word line CMOS

(a) (b)
dates to replace some of the existing memories [9–11, 465–
473]. In the case of disk drive applications, the spin valves Figure 45. Schematic of the MRAM architecture using MTJ (a) and
and MTJs are employed as sensors to sense the magnetic memory cell design using a transistor and an MTJ (b).
fields from the information bits recorded on the hard disks.
But when being used as cell elements in memories, the sen-
sors themselves are used to store the information. The basic to the readout process of MRAMs using pseudospin valves,
structure of the memory cell is still the same as that of the wherein larger currents are needed to switch the magnetiza-
sensor, but in the former case, one makes use of the satura- tion of the free layer from one direction to the opposite so
tion region of SVs and MTJs rather than the linear region as to detect the state of the stored bit (the so-called dynamic
used in read sensors. readout process).

7.2. Different Kinds of MRAM Designs 7.3. Write Operation of MRAM

The concept of magnetic random memory itself is not new, The write operation can be understood by looking at how a
but the early magnetic memory did not become popular single-domain particle switches under two orthogonal fields,
due to both the drawbacks of its own design and the rapid as shown schematically in Figure 46. Assuming that the mag-
advances made in semiconductor memories. However, the netization switching occurs via coherent rotation, or based
concept of using two crossed wires to write cells selectively on the so-called Stoner–Wohlfarth (SW) model, the switch-
is still applicable to current MRAMs. The key difference ing field can be obtained through the minimization of the
between the early magnetic memory and today’s MRAM is energy density:
that the former used an inductive signal to detect the stor-
age state, while the latter utilizes AMR [9], GMR [474–479], = −HMs cosB −  + Ku sin2 
or the spin-dependent tunneling phenomenon [10 11 299]
to detect the signals. This has made it possible to remove where H is the external magnetic field, Ms is saturation
the coils from the memory arrays, and at the same time, magnetization, Ku is the anisotropy energy constant, and 
to increase the readout sensitivity, leading to the miniatur- and B are the angle between the easy axis and Ms and H ,
ization of the cell elements. The first AMR-based MRAM respectively. Note that here only the Zeeman energy and
was developed by Honeywell in the mid-1980s. However, its the anisotropy energy are taken into account. Minimizing
application was mainly limited to certain niche markets due the energy density gives
to its high cost and relatively poor performance attributable
to the low MR ratio. With recent advances in GMR and h⊥ = − cos3  h = sin3  or h2/3 − h⊥
2/3
= 1 (49)
MTJ technologies, renewed interest has arisen in MRAMs
since the mid-1990s. Due to the large read/write current Here,
and a relatively low SNR of large cell arrays of GMR-based H
MRAMs, most of the recent works on MRAMs are based h = H Ms /2Ku = 
on the MTJ design. 2Ku /Ms
Figure 45 shows a schematic of an MRAM design using H⊥
h⊥ = H⊥ Ms /2Ku = (50)
MTJs and the corresponding cell structure using one MTJ 2Ku /Ms
and one transistor per cell. The memory cells are formed at
the cross junctions of the bit and word lines. Each memory are the reduced parallel and perpendicular components of
cell consists of an MTJ for storing the information and a the external field. Equation (44) gives an asteroid curve, as
transistor for reading the cell selectively. The transistor can shown in Figure 47. To achieve magnetization reversal, the
also be replaced by either a diode [480 467] or can even be magnetic field vector must be such that its endpoint ends
removed from the cell, as will be discussed shortly. A typi- outside the asteroid (point B, for example). For some of the
cal MTJ cell consists of a pinning layer to pin the reference
layer, a storage layer to store the information, and a thin Hard axis
insulator to separate the two magnetic layers. The signal H⊥ H
readout is based on the spin-dependent tunneling resistance M
φ θ
of the MTJ by supplying a small sensing current to a specific
H//
cell. As the sensing current is only used to detect the rel- Easy axis

ative orientation of the magnetization of the two magnetic

layers instead of being used to switch the magnetization,
it can be kept at a relatively low level. This is in contrast Figure 46. Schematic of the SW model for a single-domain particle.
40 Nano Spintronics for Data Storage

h⊥ 170

Proof's Only
160
1

R (kΩ)
150
B
A
m 140
m
h h||
θ* 1 130
-1
-100 -50 0 50 100
Magnetic Field (Oe)
-1
Figure 48. M–H curves of four identical cells with the structure Cr/Cu/
Cr/Ta /NiFe /IrMn /CoFe/Ru/CoFe /AlOx /CoFe /NiFe/CoFe/AlOx /CoFe/
Figure 47. Asteroid curve for a single-domain particle. Ru/CoFe/IrMn/Ta.

points inside the asteroid, although one of the field com- In addition to switching field fluctuation reduction, efforts
ponents is larger than that of point B, their magnetizations have also been made to reduce the absolute value of the
still cannot be switched. This simple mechanism makes it switching field, which is proportional to the write current.
possible to write the individual cells selectively [481 482]. Currently, the write current is about a few milliamperes; it
In practice, however, the magnetization reversal process will increase to more than 10 mA when the element size
of the memory element is much more complicated than what shrinks to less than 100 nm [491]. Tehrani et al. reported
is predicated by the simple single-domain model [469]. The that the effective write current can be reduced by cover-
magnetization switching process of rectangular elements is ing the word line (Cu wire) with a high-susceptibility soft
usually unrepeatable due to the formation of edge domains magnetic layer from all sides but the side that is facing the
at the flat ends of the element. Elements with tapered MTJ cell [492]. Under optimum conditions, it is possible to
ends have been introduced to suppress the formation of reduce the write current by a factor of 2–4 [491]. Further
the edge domains [483]. Although the switching process reduction of the write current can be achieved by reducing
becomes relatively repeatable, the tapered ends result in an the distance between the word line and the MTJ cells. When
increase in the switching field. An excellent review of the the demagnetizing field is taken into account, the switch-
micromagnetics of MRAM was given by Zhu and Zheng ing field of an MRAM cell using MTJ is proportional to
[484]. Using Lorentz transmission electron microscopy, Kirk the thickness of the free layer and inversely proportional to
et al. [485 486] and Yi et al. [487] found that the switch- the size of the cell. Therefore, it is important to keep the
ing field of small magnetic elements can either increase thickness/width ratio constant during down-size scaling of
or decrease, depending on whether the ends are gently the cell. Finally, the write current can also be reduced by
curved or exhibit sharp corners. In element arrays, the reducing the shape anisotropy and increasing the anisotropy
magnetostatic interactions among the elements were also of the material itself by using, for example, synthetic ferri-
found to affect the switching process of individual elements magnets as the free layer [493].
[488 489]. A unique design that makes the switching pro-
cess repeatable, and yet is more suitable for ultrahigh- 7.4. Read Operation of MRAM
density memory is the vertical magnetoresistive random
access memory (VMRAM) [471 476 484]. The VMRAM As we mentioned in the beginning of this section, one of the
cell consists of a ring-shaped CPP spin valve with two FM key features of the MRAM is that the information is stored
layers separated by a nonmagnetic layer. The formation of a in the relative orientation of the magnetization of two sep-
stable flux-closure configuration of the magnetization makes arate magnetic layers rather than in the charged/discharged
the switching process more robust and predictable. state of capacitors. Therefore, it is more suitable for scaling
down to small structures, at least, in terms of output signal.
The formation of edge domains is also sensitive to
As discussed in the MTJ section, the maximum output signal
process-related defects in the magnetic element. This was
of an MTJ is given by
found to result in fluctuations of the switching field among
different cells [10]. As one of the examples, Figure 48 shows 1
the M–H curve for a double-tunnel-junction MRAM with :VMTJ V = Vmax /2 = V · MR0  (51)
2 max
four cells of the same size fabricated in the author’s group.
Although they were fabricated using identical processes and In principle, it is independent of the cell size, although it
were placed one next to another on the same wafer, the may decrease by reducing the size in actual devices. As the
switching fields differ from one another. In addition to the MTJs in MRAMs detect the resistance difference between
fluctuation of switching fields among different cells, the shift parallel and antiparallel states, the peak-to-peak output volt-
of the switching curve along the field axis also requires a age is in the range of 120–160 mV, which is much larger
large tolerance for the switching current [490]. The latter than the output voltage of a read sensor. The use of double-
stems from various magnetic couplings between the free and tunnel junctions may further boost this value [494]. One
pinned layer, and thus must be minimized by process opti- of the key concerns in the readout process of MRAMs at
mization. In the case of mass production, how to achieve the product level is the fluctuation of MR values among
a narrow distribution for the switching field for millions or the large number of cells. Fortunately, the rapid progress in
billions of cells is a great challenge. process development has now made it possible to suppress
Nano Spintronics for Data Storage 41

the fluctuation of MR at an acceptable level. For example, Bit line VV

Tehrani et al. reported that the mean value and standard I
MTJ
deviation of MR in their particular samples were success- +
–
Cell being
read
fully suppressed to 5 and 1%, respectively, over a 200 mm
wafer [492]. Voltage
Word line
follower

7.5. Diode-Free/Multilayer/Multilevel Designs Figure 50. Schematic diagram of a switch-free 2 × 2 MRAM cell array.
A simple peripheral circuitry is used to reduce the shunting current.
The existing 1 MTJ/1 T cell design not only makes the
MRAM costly, but also makes it difficult to reduce the
dimensions of the entire MRAM cell. A switch-free design MRAM design, the current that is needed to switch the mag-
is much more desirable for fully making use of the advan- netization is still comparable to the write current of exist-
tages of MRAMs [495 496]. Figure 42 shows an example of ing MRAMs [501–508]. Very recently, Zhuang et al. [509]
such a MRAM design using MTJs. As there is no switching demonstrated the first prototype of an electrical-field-driven
diode or transistor associated with each MTJ, all of the cells RAM using the colossal magnetoresistance effect (CMR).
are electrically mutually connected, forming a large resis- As it is based on the huge change of resistance at two differ-
tor network. This will reduce the amplitude of the readout ent states, it is also dubbed a resistance random-access mem-
ory or RRAM. For a 0.8 1m × 0.8 1m cell, the write current
signal due to the current-shunting effect, as shown schemati-
was 0.2 mA, which is much smaller than the write current
cally in Figure 49. The author’s group has proposed a simple
of MRAMs. The RRAM is particularly suitable for multi-
peripheral circuitry to reduce the shunting current [496]. As
level memories due to its large resistance change between
shown schematically in Figure 50, the simple circuit is basi-
two different states (by a factor of 10–1000). Another field
cally a voltage follower. It sets an equal electrical potential
which can be explored for creating electrical-field-driven
for adjacent bit lines so as to prevent current shunting to
memory devices is magnetic semiconductors. Ohno et al.
other paths when a selected bit is being read. Although it
recently demonstrated electrical-controlled ferromagnetism
is only shown here for a 2 × 2 cell, it can be easily scaled
in (In, Mn)As-based magnetic semiconductors, but so far, it
up to large-scale cell arrays without the need to increase the
has not been implemented in any form of functional devices
complexity of the peripheral circuit. The effectiveness of this
[510].
method has been verified experimentally [496]. The diode-
free design is particularly useful for building multiple-layer
MRAMs. In addition to multiple-layer design, multilevel or 8. SEMICONDUCTOR SPINTRONICS
multistate design also remains as one of the options for real-
izing gigabit MRAMs in the future [497–500]. In this chapter, we have focused on metal-based spin-
tronics and its application in data storage. Another
important emerging field is the spintronics that uses
7.6. Electrical Field Driven Memory nonmagnetic semiconductors, magnetic semiconductors, and
metal/semiconductor hybrid structures. A number of device
The MRAM in its present form is intrinsically a current- concepts have been proposed, and some of them have been
driven device, at least during writing. The most desirable verified experimentally. These include, but are not limited
memory should be the one such that only reading involves to: (1) the spin field-effect transistor [511], (2) the spin-valve
current, while writing is based on an electrical field effect. transistor [512], (3) the magnetic semiconductor tunnel junc-
This kind of memory is difficult to be implemented using tion [513 514], (4) magnetic semiconductor diodes [515],
metallic magnetic materials because their magnetic proper- (5) the hybrid Hall effect device [516], magnetic tunnel
ties cannot be altered by a static electric field, at least at transistors [517], and spin-dependent light-emitting devices
the moment and under normal conditions. The necessity of [518–520]. The greatest advantage of spintronics using semi-
generating a magnetic field from a current to switch the conductors is that it can be integrated into the existing
magnetization in current MRAMs increases the complexity microelectronics. Furthermore, it also offers the possibility
of the design and manufacture of such devices. Although of controlling the magnetic properties using an electric field,
magnetization switching based on “spin transfer” has been which may lead to novel nanodevices, which is not possible
studied extensively, which has the potential to simplify the using metal-based magnetic materials. Research activities in
semiconductor spintronics have expanded almost exponen-
tially in recent years. An in-depth coverage of this topic is
Bit line out of the scope of this chapter because we focus on data

Proof's Only
I V

MTJ
storage, which at the moment is still dominated by metal-
Cell being
read
based spintronics. Semiconductor-based spintronic devices
Current
may find applications in data storage, but they need to
shunting Word line overcome many technical hurdles before they can be used
in hard disk drives. This is because most of the semicon-
Figure 49. Schematic diagram of a switch-free 2 × 2 MRAM cell array. ductors can only be grown on high-quality crystalline sub-
When a sensing current is supplied to a selected cell to read out the strates instead of the AlTiC substrates used currently for
data, a part of the current is shunted to other cells, resulting in a sliders. Most of the semiconductor epilayers are also too
decrease in the readout signal. fragile to sustain the various lapping/polishing processes
42 Nano Spintronics for Data Storage

employed in current magnetic head fabrication processes. spot into and, at far field, it is determined by the diffraction-
Therefore, semiconductor-based spintronic devices are more limited performance of the optical system. Although near-
suitable for future memory and logic devices. The latest field optics can reduce the spot size to a certain extent, it
reviews of semiconductor-based spintronics can be found in is still insufficient to compete with the state-of-the-art mag-
[4 31], and the April 2002 issue of Semiconductor Science netic recording devices.
and Technology. The magnetic storage devices which have been the focus
of this chapter store the information using the spin of elec-
trons. Although the output of CIP sensors scales with the
9. FUTURE TREND OF DATA STORAGE size of the information bit, the output of CPP sensors includ-
AND THE ROLE OF SPINTRONICS ing MTJ is relatively insensitive to the size of the sensor.
However, even for the latter, the SNR scales inversely with
Most of the existing data storage devices can be categorized the bit size. Therefore, the areal density of magnetic record-
into three major groups: electronic, magnetic, and optical ing will be limited either by the SNR or by the thermal sta-
devices. The electronic storage devices store the information bility of the bit, whichever gives the lower density. Although
using charges. Therefore, the absolute number of electrons large ballistic magnetoresistance has been obtained in mag-
that are used to store one bit of information matters because netic nanocontacts [521 522], one still has to find a way to
it will determine the SNR of the readout signal. To this convert it into a useful sensor for read heads, in particular,
end, various types of capacitance enhancement techniques the linearity and low-field requirements.
have been developed, such as bottle-shaped deep trenches, If the bit size of all of these different types of storage
hemispherical grains, and three-dimensional structures over devices keeps shrinking, it will eventually approach the size
the bit line. In addition to the innovation in designs, efforts of atoms. When we reach that stage, we will probably have
are also being made to develop suitable high-K capacitor to seriously look at the information storage in reciprocal
dielectrics. In the field of flash EEPROMs, multibit and mul- space. As mentioned above, the concept of reciprocal space
tilevel technologies are also being introduced to increase the recording has already been explored in optical data storage.
storage capacity. A similar concept should also be explored in charge- and
Optical storage, on the other hand, has the most diversity spin-based memories. This will lead to the general concept
among all kinds of data storage devices. This is because, as of quantum information storage. Nanometer-scale spintron-
shown schematically in Figure 51, light as an electromagnetic ics is expected to play a bridging role to bring the clas-
wave carries more information than its static counterpart of sic data storage into the quantum information storage era
electrical and magnetic fields. Depending on which param- (see Fig. 52).
eter in the electrical field expression that is used to read or
store the information, various types of optical data storage
technologies have been developed, and some of them have GLOSSARY
been commercialized. These include: (1) read-only mem-
ory (ROM), (2) phase-change (PC) disk, (3) magnetoopti-
cal (MO) disk, (4) multiple-layer storage, (5) holographic
ACKNOWLEDGMENTS
data storage, (6) hole-burning memory, and (7) time-domain
memory. The wave nature of light makes it possible to real- The author is grateful to the members of the Nano Spin-
ize the storage of information in real space [(1)–(4)], recip- electronics Group at the Data Storage Institute, espe-
rocal space [5 6], and the time domain. If we focus only cially Dr. Kebin Li, Dr. Jinjun Qiu, Dr. Yuankai Zheng,
on real space recording, the areal density of optical stor- Dr. Zaibing Guo, Dr. Guchang Han, and Ms. Luo Ping for
age is limited physically by the fact that a certain number their assistance with some of the experimental data. He is
of photons are required to reproduce the information with also grateful to Dr. Li Wang and Mr. Yatao Shen for their
an acceptable SNR. When the bit becomes very small, the assistance with some of the literature survey work. He would
number of photons that come back from a tiny bit is lim- also like to acknowledge the support of Prof. Chong Tow
ited by the intensity of the light, which is in turn determined Chong, Dr. Thomas Liew, and Dr. Chang Quan Teck.
by the damage threshold of the storage medium. Of course,
in practice, the areal density of real space optical recording
is largely determined by how small one can focus the laser Real Space Recording Reciprocal Space Recording

100100
ROM MULTIPLE PHASE 101001

Proof's Only
|1〉 |0〉
PC LAYER SHIFT 001001
MO STORAGE HOLOGRAM
000101 Kn
.
100001 . (quantum storage)
→ → → Bits meet atoms .
E = E0 exp [ k ⋅ r − ω t + ϕ ]
→ 100101 Charges interact
K3
101000 with spins
K2 Real Space Reciprocal Space

101010 K1
10000
2π

WAVELENGTH HOLE TIME

λ

FT

MULTIPLEXING BURNING DOMAIN

a = Nλ

Classic scaling
λ

HOLOGRAM MEMORY MEMORY

λ
λ 2π
Nλ

(e.g., holographic recording)

Figure 51. A wide range of optical data storage technologies that has
been developed using the different properties of light waves. Figure 52. Future trend of data storage.
Nano Spintronics for Data Storage 43

REFERENCES 30. H. Ohno, Science 281, 951 (1998).

31. H. Ohno, F. Matsukura, and Y. Ohno, JSAP Int. 5, 4 (2002).
1. G. A. Prinz, Science 282, 1660 (1998). 32. Diluted magnetic semiconductors, in “Semiconductors and
2. Special issue on magnetoelectronics, Phys. Today (Apr. 1995). Semimetals” (J. K. Furydna and J. Kossut, Eds.), Vol. 25. Aca-
3. J. F. Gregg, I. Petej, E. Jouguelet, and C. Dennis, J. Phys. D: Appl. demic, New York, 1988.
Phys. 35, R121 (2002).
33. T. Dietl, Semicond. Sci. Technol. 17, 377 (2002).
4. M. Ziese and M. J. Thornton, Eds., “Spin Electronics.” Springer,
34. S. Sato and H. Katayama-Yoshida, Semicond. Sci. Technol. 17, 367
Berlin, 2001.
(2002).
5. J. Hong, J. Kane, J. Hashimoto, M. Yamagishi, K. Noma, and
35. M. L. Reed, N. A. El-Masry, and H. H. Stadelmaier, Appl. Phys.
H. Kanai, “12th Magnetic Recording Conference.” Minneapolis,
Lett. 79, 3473 (2001).
MN, Aug. 2001.
36. T. Sakaki, S. Sonoda, Y. Yamamoto, S. Shimizu, K. Suga, K. Kindo,
6. M. Yamagishi, J. Hong, E. N. Abarra, B. R. Acharya, I. Okamoto,
and H. Hori, Appl. Phys. Lett. 91, 7911 (2002).
and H. Komoriya, “Digest of INTERMAG Europe 2002,” Ams-
37. G. A. Medvedkin, T. Ishibashi, T. Nishi, K. Hayata, Y. Hasegawa,
terdam, 2002, paper BA01.
and K. Sato, Jpn. J. Appl. Phys. 39, L949 (2000).
7. Z. Y. Zhang, Y. C. Feng, T. Clinton, G. Badran, N.-H. Yeh,
38. N. Theodoropoulou, A. F. Hebard, M. E. Overberg, C. R. Aber-
E. Girt, S. Harkness, M. Muntteanu, H. J. Richter, R. Ranjan,
nathy, S. J. Pearton, S. N. G. Chu, and R. G. Wilson, Phys. Rev.
S. Hwang, G. C. Rauch, M. Ghaly, D. Larson, E. Singleton,
Lett. 89, 107203 (2002).
V. Vas’ko, J. Ho, F. Stageberg, V. Kong, K. Duxstad, and S. Slade,
39. Y. Matsumoto, M. Murakami, T. Shono, T. Hasegawa, T. Fuku-
“Digest of INTERMAG Europe 2002,” Amsterdam, 2002, paper
mura, M. Kawasaki, P. Ahmet, T. Chikyow, S. Koshihara, and
BA02.
H. Koinuma, Science 291, 854 (2001).
8. K. Stove, F. Liu, Y. Chen, X. Dang, P. Luo, J. Chen, K. Kung,
40. K. Ueda, H. Tabata, and T. Kawai, Phys. Lett. 79, 988 (2001).
M. Lederman, M. Re, G. Choe, and M. Zheng, “Abstract of the
41. M. Tanaka and Y. Higo, Phys. Rev. Lett. 87, 026602 (2001).
47th Annual Conference on Magnetism & Magnetic Materials,”
42. M. Tanaka, Semicond. Sci. Technol. 17, 327 (2002).
Tampa, FL, Nov. 2002, paper AG08.
43. G. A. Prinz, Phys. Today 48, 58 (1995).
9. J. M. Dauton, Thin Solid Films 216, 162 (1992); J. M. Dauton,
44. R. Fiederling, M. Keim, G. Reuscher, W. Ossau, G. Schmidt,
J. Appl. Phys. 81, 3758 (1997).
10. S. S. P. Parkin, K. P. Roche, M. G. Samant, P. M. Rice, R. B. Bey- A. Waag, and L. W. Molenkamp, Nature 402, 787 (1999).
ers, R. E. Scheurlein, E. J. O’Sullivan, S. L. Brown, J. Bucchigana, 45. T. Fujisawa, D. G. Austing, Y. Tokura, Y. Hirayama, and
D. W. Abraham, Y. Lu, M. Rooks, P. L. Trouilloud, R. A. Wanner, S. Tarucha, Nature 419, 278 (2002).
and W. J. Gallagher, J. Appl. Phys. 85, 5828 (1999). 46. J. Lohau, A. Moser, C. T. Rettner, M. E. Best, and B. D. Terris,
11. S. Tehrani, E. Chen, M. Durlam, M. Deherrera, J. M. Slaughter, IEEE Trans. Magn. 37, 1652 (2001).
J. Shi, and G. Kerszkowski, J. Appl. Phys. 85, 5822 (1999). 47. C. A. Ross, Annu. Rev. Mater. Res. 31, 203 (2001).
12. B. K. Middleton, in “Magnetic Recording Technology” (C. D. Mee 48. S. Sun, C. B. Murray, D. Weller, L. Folks, and A. Moser, Science
and E. D. Daniel, Eds.), Chap. 2. McGraw-Hill, New York, 1996. 87, 1989 (2000).
13. X. Wang and A. M. Taratorin, “Magnetic Information Storage 49. International technology roadmap for semiconductors, 2001;
Technology.” Academic, New York, 1999. http://public.itrs.net.
14. R. M. White, “Introduction to Magnetic Recording.” IEEE Press, 50. W. Thomson, Proc. R. Soc. 8, 546 (1857).
New York, 1985. 51. R. I. Potter, Phys. Rev. B 10, 4626 (1974).
15. H. N. Bertram, “Theory of Magnetic Recording.” Cambridge Uni- 52. T. R. McGuire and R. I. Potter, IEEE Trans. Magn. MAG-11, 1018
versity Press, Cambridge, 1994. (1975).
16. H. J. Richter, J. Phys. D: Appl. Phys. 32, R147 (1999). 53. R. Hunt, IEEE Trans. Magn. MAG-7, 150 (1971).
17. E. N. Abarra, A. Inamota, H. Sato, I. Okamoto, and Y. Mizoshita, 54. D. A. Thompson, AIP Conf. Proc. 24, 528 (1974).
Appl. Phys. Lett. 77, 2581 (2000). 55. J. C. Mallinson, “Magneto-Resistive Heads: Fundamentals and
18. E. E. Fullerton, D. T. Margulies, M. E. Schabes, M. Carey, Applications.” Academic, New York, 1996.
B. Gurney, A. Moser, M. Best, G. Zeltzer, K. Rubin, H. Rosen, 56. F. Jeffers, J. Freeman, R. Toussant, N. Smith, D. Wachenschwanz,
and M. Doerner, Appl. Phys. Lett. 77, 3806 (2000). S. Shtrikman, and W. Doyle, IEEE Trans. Magn. MAG-21, 1563
19. E. N. Abarra, B. R. Acharya, A. Inomata, and I. Okamoto, IEEE (1985).
Trans. Magn. 37, 1426 (2001). 57. K. Yamada, T. Maruyama, T. Tatsumi, T. Suzuki, K. Shimabayashi,
20. R. Hunt, IEEE Trans. Magn. MAG-7, 150 (1971). Y. Motomura, M. Aoyama, and H. Urai, IEEE Trans. Magn. 26,
21. B. Gurney, M. Carey, C. Tsang, M. Williams, S. S. P. Parkin, 3010 (1990).
R. Fontana, Jr., and E. Grochowski, Datatech 4, 87 (2000). 58. F. B. Shelledy, IEEE Trans. Magn. 28, 2283 (1992).
22. M. N. Baibich, J. M. Broto, A. Fert, F. Nguyen van Dau, F. Petroff, 59. K. B. Klaassen and J. C. L. van Peppen, IEEE Trans. Magn. 33,
P. Etienne, G. Creuzet, A. Friederich, and J. Chazelas, Phys. Rev. 2611 (1997).
Lett. 61, 2472 (1988). 60. N. H. Yeh, IEEE Trans. Magn. MAG-18, 1155 (1982).
23. G. Binasch, P. Grunberg, F. Saurenbach, and W. Zinn, Phys. Rev. 61. D. E. Heim, IEEE Trans. Magn. 30, 1453 (1994).
B 39, 4828 (1989). 62. “Molecular Beam Epitaxy” (A. Cho, Ed.). AIP, New York, 1994.
24. B. Dieny, V. S. Speriosu, S. S. P. Parkin, B. A. Gurney, D. R. 63. L. L. Chang, L. Esaki, W. E. Howard, and R. Ludeke, J. Vac. Sci.
Wilhoit, and D. Mauri, Phys. Rev. B 43, 1297 (1991). Technol. 10, 11 (1973).
25. A. Vedyayeva, M. Chshieva, N. Ryzhanovaa, B. Dieny, C. Cow- 64. T. Shinjo and T. Takada, in “Ferromagnetic Materials” (E. P. Wohl-
acheb, and F. Brouersc, J. Magn. Magn. Mater. 172, 53 (1997). farth, Ed.), Vol. 3. Elsevier, Amsterdam, 1987.
26. M. Jullière, Phys. Lett. 54A, 225 (1975). 65. N. F. Mott and H. Jones, “The Theory of the Properties of Metals
27. S. Maekawa and U. Gafvert, IEEE Trans. Magn. MAG-18, 707 and Alloys,” pp. 265–268. Clarendon, Oxford, 1936.
(1982). 66. M. A. M. Gijs and G. E. W. Bauer, Adv. Phys. 46, 285 (1997).
28. T. Miyazaki and N. Tezuka, J. Magn. Magn. Mater. 139, L231 67. J. Mathon, in “Spin Electronics” (M. Ziese and M. J. Thornton,
(1995). Eds.), Chap. 4, p. 75 and references therein. Springer, Berlin, 2001.
29. J. S. Moodera, L. R. Kinder, T. M. Wong, and R. Meservey, Phys. 68. R. E. Camley and J. Barnas, Phys. Rev. Lett. 63, 664 (1989).
Rev. Lett. 74, 3273 (1995); J. S. Moodera and L. R. Kinder, J. Appl. 69. J. Barnas, A. Fuss, R. E. Camley, P. Grünberg, and W. Zinn, Phys.
Phys. 79, 4724 (1996). Rev. B 42, 8110 (1990).
44 Nano Spintronics for Data Storage

70. R. Q. Hood and L. M. Falicov, Phys. Rev. B 46, 8287 (1992). 111. Y. H. Wu, K. B. Li, J. J. Qiu, Z. B. Guo, and G. C. Han, Appl.
71. S. Zhang and P. M. Levy, J. Appl. Phys. 69, 4786 (1991). Phys. Lett. 80, 4413 (2002).
72. P. M. Levy, Solid State Phys. 47, 367 (1994). 112. B. Dienyl, V. S. Speriosu, B. A. Gurney, S. S. P. Parkin, D. R.
73. P. M. Levy and S. Zhang, J. Magn. Magn. Mater. 151, 315 (1995). Wilhoit, K. P. Roche, S. Metin, D. T. Peterson, and S. Nadimi,
74. X. G. Zhang and W. H. Butler, Phys. Rev. B 51, 10085 (1995). J. Magn. Magn. Mater. 93, 101 (1991).
75. P. Zahn, I. Mertig, M. Richter, and H. Eschrig, Phys. Rev. Lett. 75, 113. B. Dieny, V. S. Speriosu, S. Metin, S. S. P. Parkin, B. A. Gurney,
2996 (1995). P. Baumgart, and D. R. Wilhoit, J. Appl. Phys. 69, 4774 (1991).
76. M. A. M. Gijs and G. E. W. Bauer, Adv. Phys. 46, 285 (1997). 114. H. Kanai, Y. Yamada, K. Aoshima, Y. Ohtsuka, J. Kane,
77. R. K. Nesbet, IBM J. Res. Develop. 42, 53 (1998). M. Kanamine, J. Toda, and Y. Mizoshita, IEEE Trans. Magn. 32,
78. P. Grünberg, R. Schreiber, Y. Pang, M. B. Brodsky, and H. Sowers, 3368 (1996).
Phys. Rev. Lett. 57, 2442 (1986). 115. S. S. P. Parkin, Appl. Phys. Lett. 61, 1358 (1992).
79. S. S. P. Parkin, N. More, and K. P. Roche, Phys. Rev. Lett. 64, 2304 116. C. Tsang, R. E. Fontana, T. Lin, D. E. Heim, V. S. Speriosu, B. A.
(1990). Gurney, and M. L. Williams, IEEE Trans. Magn. 30, 3801 (1994).
80. S. S. P. Parkin, R. Bhadra, and K. P. Roche, Phys. Rev. Lett. 66, 117. M. A. M. Gijs, J. B. Giesbers, and S. K. J. Lenczowski, Phys. Rev.
2152 (1991). Lett. 70, 3343 (1993).
81. S. S. P. Parkin, Phys. Rev. Lett. 67, 3598 (1991). 118. W. Vavra, S. F. Cheng, A. Fink, J. J. Krebs, and G. A. Prinz, Appl.
82. S. S. P. Parkin, in “Ultrathin Magnetic Structures II” (B. Heinrich Phys. Lett. 66, 2579 (1995).
and J. A. C. Bland, Eds.), pp. 148–186. Springer, Berlin, 1994. 119. A. Blondel, J. P. Meier, B. Doudin, and J.-Ph. Ansermet, Appl.
83. H. Hosoito, K. Mibu, S. Araki, T. Shinjo, S. Itoh, and Y. Endo, Phys. Lett. 65, 3019 (1994).
J. Phys. Soc. Jpn. 61, 300 (1992). 120. L. Piraux, J. M. George, J. F. Despres, C. Leroy, E. Ferain,
84. P. Grunberg, S. Demokritov, A. Fuss, R. Schreiber, J. A. Wolf, and R. Legras, K. Ounadjela, and A. Fert, Appl. Phys. Lett. 65, 2488
S. T. Purcell, J. Magn. Magn. Mater. 104–107, 1734 (1992). (1994).
85. A. Fuss, S. Demokritov, P. Grünberg, and W. Zinn, J. Magn. Magn. 121. K. Liu, K. Nagodawithana, P. C. Searson, and C. L. Chien, Phys.
Mater. 103, L221 (1992). Rev. B51, 738 (1985).
86. J. Unguris, R. J. Celotta, and D. T. Pierce, Phys. Rev. Lett. 67, 140 122. T. Ono and T. Shinjo, J. Phys. Soc. Jpn. 64, 363 (1995).
(1991). 123. E. M. Williams, “Design and Analysis of Magnetoresistive Record-
ing Heads,” p. 161. Wiley, New york, 2001.
87. J. A. Wolf, Q. Leng, R. Schreiber, P. A. Grünberg, and W. Zinn,
124. S. Yuan and H. N. Bertram, J. Appl. Phys. 75, 6385 (1994).
J. Magn. Magn. Mater. 121, 253 (1993).
125. D. E. Heim, R. E. Fontana, C. Tsang, V. S. Speriosu, B. A. Gurney,
88. A. Cebollada, R. Miranda, C. M. Schneider, P. Schuster, and
and M. L. Williams, IEEE Trans. Magn. 30, 316 (1994).
J. Kirschner, J. Magn. Magn. Mater. 102, 25 (1991).
126. W. P. Meiklejohn and C. P Bean, Phys. Rev. 102, 1413 (1956).
89. Z. Q. Qiu, J. Pearson, and S. D. Bader, Phys. Rev. B 46, 8659
127. J. Nogués and I. K. Schuller, J. Magn. Magn. Mater. 192, 203 (1999)
(1992).
and references therein.
90. M. T. Johnson, S. T. Purcell, N. W. E. McGee, R. Coehoorn, J. aan
128. A. E. Berkowitz and K. Takano, J. Magn. Magn. Mater. 200, 552
de Stegge, and W. Hoving, Phys. Rev. Lett. 68, 2688 (1992).
(1999) and references therein.
91. D. M. Edwards, J. Mathon, R. B. Muniz, and M. S. Phan, Phys.
129. R. L. Stamps, J. Phys. D: Appl. Phys. 33, R247 (2000) and refer-
Rev. Lett. 67, 493 (1991).
ences therein.
92. P. Lang, L. Nordström, R. Zeller, and P. H. Dederichs, Phys. Rev.
130. M. Kiwi, J. Magn. Magn. Mater. 234, 584 (2001) and references
Lett. 71, 1927 (1993).
therein.
93. L. Nordström, P. Lang, R. Zeller, and P. H. Dederichs, Phys. Rev.
131. W. P. Meiklejohn, J. Appl. Phys. 33, 1328 (1962).
B 50, 13058 (1994). 132. L. Néel, Ann. Phys. Paris 2, 61 (1967).
94. B. A. Jones and C. B. Hanna, Phys. Rev. Lett. 71, 4253 (1993). 133. A. P. Malozemoff, Phys. Rev. B 35, 3679 (1987).
95. J. Mathon, Murielle Villeret, R. B. Muniz, and J. d’Albuquerque 134. D. Mauri, H. C. Siegmann, P. S. Bagus, and E. Kay, J. Appl. Phys.
e Castro, Phys. Rev. Lett. 74, 3696 (1995). 62, 3047 (1987).
96. M. D. Stiles, Phys. Rev. B 48, 7238 (1993). 135. N. C. Koon, Phys. Rev. Lett. 78, 4865 (1997).
97. P. Bruno, Phys. Rev. B 52, 411 (1995). 136. T. C. Schulthess and W. H. Butler, Phys. Rev. Lett. 81, 4516 (1998).
98. Y. Yafet, Phys. Rev. B 36, 3948 (1987). 137. T. C. Schulthess and W. H. Butler, J. Appl. Phys. 85, 5510 (1999).
99. R. Coehoorn, Phys. Rev. B 44, 9331 (1991). 138. M. Kiwi, J. Mejía-López, R. D. Portugal, and R. Ramírez, Appl.
100. P. Bruno and C. Chappert, Phys. Rev. Lett. 67, 1602 (1991); Phys. Lett. 75, 3975 (1999).
P. Bruno and C. Chappert, Phys. Rev. Lett. 67, 2592 (1991). 139. S. Zhang, D. Dimitrov, G. C. Hadjipanayis, J. W. Cai, and C. L.
101. P. Bruno and C. Chappert, Phys. Rev. B 46, 261 (1992) Chien, J. Magn. Magn. Mater. 198–199, 468 (1999).
102. A. Fert and P. Bruno, in “Ultrathin Magnetic Structures II” 140. M. D. Stiles and R. D. McMicheal, Phys. Rev. B 59, 3722 (1999).
(B. Heinrich and J. A. C. Bland, Eds.), pp. 82–117. Springer, 141. F. Y. Yang and C. L. Chien, Phys. Rev. Lett. 85, 2597 (2000).
Berlin, 1994. 142. P. Steadman, M. Ali, A. T. Hindmarch, C. H. Marrows, B. J.
103. C. Chappert and J. P. Renard, Europhys. Lett. 15, 553 (1991). Hickey, S. Langridge, R. M. Dalgliesh, and S. Foster, Phys. Rev.
104. D. M. Deaven, D. S. Rokhsar, and M. Johnson, Phys. Rev. B 44, Lett. 89, 077201 (2002).
5977 (1991). 143. H. Ohldag, T. J. Regan, J. Stöhr, A. Scholl, F. Nolting, J. Lüning,
105. T. Shinjo and H. Yamamoto, J. Phys. Soc. Jpn. 59, 3061 (1990). C. Stamm, S. Anders, and R. L. White, Phys. Rev. Lett. 87, 247201
106. Y. Kawawake, H. Sakakima, Y. Irie, and M. Satomi, J. Magn. (2001).
Magn. Mater. 156, 405 (1996). 144. M. Jimbo, J. Magn. Soc. Jpn. 22, 12 (1998) (in Japanese).
107. S. S. P. Parkin, Phys. Rev. Lett. 71, 1641 (1993). 145. M. Lederman, IEEE Trans. Magn. 35, 794 (1999).
108. H. Yamamoto, Y. Motomura, T. Anno, and T. Shinjo, J. Magn. 146. T. Lin, C. Tsang, R. E. Fontana, and K. K. Howard, IEEE Trans.
Magn. Mater. 126, 437 (1993). Magn. 31, 2585 (1995). The thickness of NiFe is 5 nm.
109. A. V. Pohm, B. A. Everitt, R. S. Beech, and J. M. Daugton, IEEE 147. T. Lin, D. Mauri, N. Staud, C. Hwang, J. K. Howard, and G. Gor-
Trans. Magn. 33, 3280 (1997). man, Appl. Phys. Lett. 65, 1183 (1994).
110. A. A. Seigler, P. E. Anderson, and A. M. Shukh, J. Appl. Phys. 91, 148. A. J. Devasahayam and M. H. Kryder, IEEE Trans. Magn. 35, 649
2176 (2002). (1999). The thickness of NiFe is 25 nm.
Nano Spintronics for Data Storage 45

149. G. W. Anderson, Y. Huai, and M. Pakala, J. Appl. Phys. 87, 5726 182. S. H. Liao, T. Torng, and T. Kobayashi, IEEE Trans. Magn. 30,
(2000). The free layer consists of 5 nm NiFe and 2 nm CoFe. 3855 (1994).
150. Z. Qian, J. M. Sivertsen, J. H. Judy, B. A. Everitt, S. Mao, and 183. H. Kanai, J. Kane, K. Aoshima, M. Kanamine, and Y. Uehara,
E. S. Murdock, J. Appl. Phys. 85, 6106 (1999). IEEE Trans. Magn. 31, 2612 (1995).
151. S. Soeya, S. Tadokoro, T. Imagawa, M. Fuyama, and S. Narishige, 184. R. E. Fontana, Jr., S. A. MacDonald, C. Tsang, and T. Lin, IEEE
J. Appl. Phys. 74, 6297 (1993). Trans. Magn. 32, 3440 (1996).
152. M. J. Carey and A. E. Berkowitz, Appl. Phys. Lett. 65, 1183 (1994). 185. M. Xiao, A. J. Devasahayam, and M. H. Kryder, IEEE Trans.
153. J. P. Nozières, S. Jaren, Y. B. Zhang, A. Zeltser, K. Pentek, and Magn. 34, 1495 (1998).
V. S. Speriosu, J. Appl. Phys. 87, 3920 (2000). The free layer is 186. J. G. Zhu and D. J. O’Connor, IEEE Trans. Magn. 32, 3401 (1996).
CoFe. 187. K. Takano, N. Yamanaka, and M. Matsuzaki, IEEE Trans. Magn.
154. A. Veloso, P. P. Freitas, N. J. Oliveira, J. Fernandes, and M. Fer- 34, 1516 (1998).
reira, IEEE Trans. Magn. 34, 2343 (1998). 188. E. Champion and H. N. Bertram, IEEE Trans. Magn. 32, 13 (1996).
155. M. Saito, Y. Kakihara, T. Watanabe, and N. Hasegawa, 21, 505 189. C. Mitsumata, K. Kikuchi, and T. Kobayashi, IEEE Trans. Magn.
(1996) (in Japanese). 34, 1453 (1998).
156. M. Sano, S. Araki, M. Ohta, K. Noguchi, H. Morita, and M. Mat- 190. N. J. Zhu, G. Hiner, A. Rana, Y. M. Huai, and D. Seagle, IEEE
suzaki, IEEE Trans. Magn. 34, 372 (1998). Trans. Magn. 36, 2575 (2000).
157. S. Soeya, H. Hoshiya, M. Fuyama, and S. Tadokoro, J. Appl. Phys. 191. A. Morinaga, C. Ishikawa, T. Ohtsu, N. Miyamoto, and S. Nar-
80, 1006 (1996). ishige, IEEE Trans. Magn. 38, 2262 (2002).
158. M. Saito, Y. Kakihara, T. Watanabe, and N. Hasegawa, 21, 505 192. T. Suzuki, K. Ishihara, and H. Matsutera, IEEE Trans. Magn. 32,
(1996) (in Japanese). 3383 (1996).
159. M. Sano, S. Araki, M. Ohta, K. Noguchi, H. Morita, and M. Mat- 193. Y. K. Zheng, Y. H. Wu, and T. C. Chong, IEEE Trans. Magn. 36,
suzaki, IEEE Trans. Magn. 34, 372 (1998). 3158 (2000).
160. S. Bae, J. H. Judy, P. J. Chen, W. F. Egelhoff, Jr., and S. Zurn, 194. C. Ishikawa, K. Suzuki, K. Yoshida, Y. Sugita, K. Shinagawa,
Appl. Phys. Lett. 78, 4163 (2001). Y. Nakatani, and N. Hayashi, J. Appl. Phys. 75, 1036 (1994).
161. D. E. Heim and S. S. P. Parkin, U. S. Patent 5, 464, 185. 195. T. Imamura, H. Kanai, and J. Toda, IEEE Trans. Magn. 35, 2559
162. V. S. Sperious, B. A. Gurney, D. R. Wilhoit, and L. B. Brown, (1999).
“Digest of INTERMAG ’96,” 1996, paper AA-04. 196. J. C. S. Kools, K. Rook, H. Hegde, S. B. Sant, J. Wong, W. Xiong,
163. H. A. M. van den Berg, W. Clemens, G. Gieres, G. Rupp, W. Schel- B. Druz, A. Lam, A. Devayasaham, and I. Wagner, Thin Solid
ter, and M. Vieth, IEEE Trans. Magn. 32, 4624 (1996). Films 377–378, 705 (2000).
164. J. G. Zhu and Y. F. Zheng, IEEE Trans. Magn. 34, 1063 (1998). 197. N. Sharma, C. Rea, and W. O’Kane, IEEE Trans. Magn. 36, 2496
165. J. L. Leal and M. H. Kryder, J. Appl. Phys. 83, 794 (1999). (2000).
166. C. Tsang, M. Pinarbasi, H. Santini, E. Marinero, P. Arnett, 198. S. Mao, Z. Gao, H. W. Xi, P. Kolbo, M. Plumer, L. Wang, A. Goyal,
R. Olson, R. Hsiao, M. Williams, R. Payne, R. Wang, J. Moore, I. Jin, J. Chen, C. H. Hou, R. M. White, and E. Murdock, IEEE
B. Gurney, T. Lin, and R. Fontana, IEEE Trans. Magn. 35, 689 Trans. Magn. 38, 26 (2002).
(1999). 199. J. G. Zhu, Y. F. Zheng, and S. Liao, IEEE Trans. Magn. 37, 1723
167. J. G. Zhu, IEEE Trans. Magn. 35, 655 (1999). (2001).
168. K. Meguro, H. Hoshiya, K. Watanabe, Y. Hamakawa, M. Fuyama, 200. K. Nakamoto, Y. Kawato, M. Komuro, Y. Hamakawa, and
and H. Fukui, IEEE Trans. Magn. 35, 2925 (1999). T. Kawabe, “Digest of INTERMAG 2000,” Toronto, Canada, Apr.
169. H. Kanai, M. Kanamine, A. Hashimoto, K. Aoshima, K. Noma, 2000, paper FA-06.
M. Yamagishi, H. Ueno, Y. Uehara, and Y. Uematsu, IEEE Trans. 201. J. Zhang, Y. Huai, and M. Lederman, J. Appl. Phys. 91, 7285
Magn. 35, 2580 (1999). (2002).
170. X. Yan, D. Rao, M. R. Gibbons, C. Qian, and H. C. Tong, IEEE 202. Y. H. Wu, Y. K. Zheng, and T. C. Chong, “Digest of INTERMAG
Trans. Magn. 35, 2877 (1999). 2003,” Boston, MA, Mar. 2003, paper DD06.
171. S. Mao, A. Mack, E. Singleton, J. Chen, S. S. Xue, H. Wang, 203. O. Redon, N. Kasahara, K. Shimazawa, S. Araki, H. Morita, and
Z. Gao, J. Li, and E. Murdoc, J. Appl. Phys. 87, 5720 (2000). M. Matsuzaki, J. Appl. Phys. 87, 4689 (2000).
172. A. Veloso, P. P. Freitas, and L. V. Melo, IEEE Trans. Magn. 35, 204. K. Sin, M. R. Gibbons, S. Funada, M. Mao, D. Rao, C. Chien, and
2568 (1999). H. C. Tong, J. Appl. Phys. 89, 7359 (2001).
173. S. Araki, M. Sano, S. Li, Y. Tsuchiya, O. Redon, T. Sasaki, N. Ito, 205. C. Tsang, J. Appl. Phys. 55, 2226 (1984).
K. Terunuma, H. Morita, and M. Matsuzaki, J. Appl. Phys. 87, 5377 206. X. Shi, K. Ju, J. Hagen, C. L. Lin, C. C. Han, M. M. Chen, J. W.
(2000). Chang, P. W. Wang, and E. Teng, IEEE Trans. Magn. 33, 2896
174. H. M. A. van den Berg, G. Rupp, W. Clemens, G. Gieres, (1997).
M. Vieth, J. Wecker, and S. Zoll, IEEE Trans. Magn. 34, 1336 207. C. Ishikawa, K. Suzuki, N. Koyama, K. Yoshida, Y. Sugita, K. Shi-
(1998). nagawa, Y. Nakatani, and N. Hayashi, J. Appl. Phys. 74, 5666
175. M. J. Carey, S. Maat, P. Rice, R. F. C. Farrow, R. F. Marks, A. Kel- (1993).
lock, P. Nguyen, and B. A. Gurney, Appl. Phys. Lett. 81, 1044 208. Y. Shen, W. Chen, W. Jensen, D. Ravipati, V. Retort,
(2002). R. Rottmayer, S. Rudy, S. Tan, and S. Yuan, IEEE Trans. Magn.
176. C. H. Bajorek, L. T. Romankiw, and D. A. Thompson, U.S. Patent 32, 19 (1996).
3,840,898, 1974. 209. A. M. Mack, K. Subramanian, L. R. Pust, C. J. Rea, N. Amin,
177. R. D. Hempstead, S. Krongelb, and D. A. Thompson, IEEE Trans. M. A. Seigler, S. Mao, S. Xue, and S. Gangopadhyay, IEEE Trans.
Magn. MAG-14, 521 (1978). Magn. 37, 1727 (2001).
178. C. Tsang and R. E. Fontana, Jr., IEEE Trans. Magn. MAG-18, 210. E. Nakashio, J. Sugawara, S. Onoe, and S. Kumagai, J. Appl. Phys.
1149 (1982). 89, 7356 (2001).
179. C. Tsang, IEEE Trans. Magn. 25, 3692 (1989). 211. S. Mao, Z. Gao, H. W. Xi, P. Kolbo, M. Plumer, L. Wang, A. Goyal,
180. M. T. Krounbi, W. J. Van Gestel, and P.-K. Wang, U.S. Patent 5, I. Jin, J. Chen, C. H. Hou, R. M. White, and E. Murdock, IEEE
018, 037, 1991. Trans. Magn. 38, 26 (2002).
181. D. Hannon, M. Krounbi, and J. Christner, IEEE Trans. Magn. 30, 212. N. J. Gokemeijer, T. Ambrose, and C. L. Chien, Phys. Rev. Lett.
298 (1994). 79, 4270 (1997).
46 Nano Spintronics for Data Storage

213. N. J. Gokemeijer, T. Ambrose, C. L. Chien, N. Wang, and K. K. 244. T. Lin and D. Mauri, Appl. Phys. Lett. 78, 2181 (2001).
Fung, J. Appl. Phys. 81, 4999 (1997). 245. S. Sant, M. Mao, J. Kools, K. Koi, H. Iwasaki, and M. Sahashi,
214. B. A. Gurney, V. S. Speriosu, J.-P. Nozieres, H. Lefakis, D. R. J. Appl. Phys. 89, 6931 (2001).
Wilhoit, and O. U. Need, Phys. Rev. Lett. 71, 4023 (1993). 246. J. C. S. Kools, S. B. Sant, K. Book, W. Xiong, Faiz Dahmani, W. Ye,
215. Y. Kamiguchi, H. Yuasa, H. Fukuzawa, K. Koi, H. Iwasaki, and J. Nuñez-Regueiro, Y. Kawana, M. Mao, K. Koi, H. Iwasaki, and
M. Sahashi, “Digest of INTERMAG’99,” 1999, paper DB-01. M. Sahashi, IEEE Trans. Magn. 37, 1783 (2001).
216. M. Ueno, H. Nishida, K. Mizukami, F. Hikami, K. Tabuchi, and 247. Y. Sugita, Y. Kawawake, M. Satomi, and H. Sakakima, J. Appl.
T. Sawasaki, IEEE Trans. Magn. 36, 2572 (2000). Phys. 89, 6919 (2001).
217. P. M. Baumgart, B. Dieny, B. A. Gurney, J. P. Nozieres, V. S. 248. M. F. Gillies, A. E. T. Kuiper, and G. W. R. Leibbrandt, J. Appl.
Speriosu, and D. R. Wilhoit, U.S. Patent 5,287,238, 1994. Phys. 89, 6922 (2001).
218. T. C. Anthony, J. A. Brug, and S. Zhang, IEEE Trans. Magn. 30, 249. K. Li, Y. Wu, J. Qiu, G. Han, Z. Guo, H. Xie, and T. Chong, Appl.
3819 (1994). Phys. Lett. 79, 3663 (2001).
219. W. F. Egelhoff, Jr., T. Ha, R. D. K. Misra, Y. Kadmon, J. Nir, C. J. 250. K. Li, Y. Wu, J. Qiu, and T. Chong, J. Appl. Phys. 91, 8563 (2002).
Powell, M. D. Stiles, R. D. McMichael, C.-L. Lin, J. M. Sivertsen, 251. J. B. Sousa, J. O. Ventura, M. A. Salgueiro da Silva, P. P. Freitas,
J. H. Judy, K. Takano, A. E. Berkowitz, T. C. Anthony, and J. A. and A. Veloso, J. Appl. Phys. 91, 5321 (2002).
Brug, J. Appl. Phys. 78, 273 (1995). 252. S. H. Jang, T. Kang, H. J. Kim, and K. Y. Kim, J. Magn. Magn.
220. J.-G. Zhu, X.-G. Ye, S. W. Yuan, H.-C. Tong, and R. Rottmayer, Mater. 240, 192 (2002).
J. Appl. Phys. 79, 5886 (1996). 253. S. H. Jang, T. Kang, H. J. Kim, and K. Y. Kim, Appl. Phys. Lett.
221. H. C. Tong, X. Shi, F. Liu, C. Qian, Z. W. Dong, X. Yan, R. Barr, 81, 105 (2002).
L. Miloslavsky, S. Zhou, J. Perlas, P. Prabhu, M. Mao, S. Funada, 254. A. Al-Jibouri, M. Hoban, Z. Lu, and G. Pan, J. Appl. Phys. 91,
M. Gibbons, Q. Leng, J. G. Zhu, and S. Dey, IEEE Trans. Magn. 7098 (2002).
35, 2574 (1999). 255. F. Shen, Q. Y. Xu, G. H. Yu, W. Y. Lai, Z. Zhang, Z. Q. Lu,
222. A. Tanaka, Y. Shimizu, H. Kishi, K. Nagasaka, and M. Oshiki, G. Pan, and A. Al-Jibouri, Appl. Phys. Lett. 80, 4410 (2002).
IEEE Trans. Magn. 33, 3592 (1997). 256. M. Mao, C. Cerjan, and J. Kools, J. Appl. Phys. 91, 8560 (2002).
223. A. Tanaka, Y. Shimizu, H. Kishi, K. Nagasaka, H. Kanai, and 257. Y. Huai, Z. T. Diao, and J. Zhang, IEEE Trans. Magn. 38, 20
M. Oshiki, IEEE Trans. Magn. 35, 700 (1999). (2002).
224. B. Dieny, Europhys. Lett. 17, 261 (1992). 258. N. Hasegawa, F. Koike, K. Ikarashi, M. Ishizone, M. Kawa-
225. H. J. M. Swagten, G. J. Strijkers, P. J. H. Bloemen, mura, Y. Nakazawa, A. Takahashi, H. Tomita, H. Iwasaki, and
M. M. H. Willekens, and W. J. M. de Jonge, Phys. Rev. B 53, 9108 M. Sahashi, J. Appl. Phys. 91, 8774 (2002).
(1996). 259. H. Fukuzawa, K. Koi, H. Tomita, H. N. Fuke, H. Iwasaki, and
226. N. Hasegawa, A. Makino, F. Koike, and K. Ikarashi, IEEE Trans. M. Sahashi, J. Appl. Phys. 91, 6684 (2002).
Magn. 32, 4618 (1996). 260. J. Hong, J. Kane, J. Hashimoto, M. Yamagishi, K. Noma, and
227. Y. Sugita, Y. Kawawake, M. Satomi, and H. Sakakima, Jpn. J. H. Kanai, IEEE Trans. Magn. 38, 15 (2002).
Appl. Phys. 37, 5984 (1998). 261. T. Mizuguchi and H. Kano, IEEE Trans. Magn. 37, 1742 (2001).
228. Y. Kawawake, Y. Sugita, M. Sotomi, and H. Sakakima, J. Appl. 262. W. Y. Lee, M. Carey, M. F. Toney, P. Rice, B. Gurney, H.-C.
Phys. 85, 5024 (1999). Chang, E. Allen, and D. Mauri, J. Appl. Phys. 89, 6925 (2001).
229. S. Bae, J. H. Judy, W. F. Egelhoff, Jr., and P. J. Chen, J. Appl. Phys. 263. W. C. Uhlig, M. Mao, V. Yiu, J. Li, and J. Shi, J. Appl. Phys. 89,
87, 6980 (2000). 6937 (2001).
230. W. F. Egelhoff, Jr., P. J. Chen, C. J. Powell, M. D. Stiles, R. D. 264. C.-H. Lai, C. J. Chen, and T. S. Chin, J. Appl. Phys. 89, 6928 (2001).
McMichael, J. H. Judy, K. Takano, A. E. Berkowitz, and J. M. 265. L. Néel, C. R. Acad. Sci. 255, 1676 (1962).
Daughton, IEEE Trans. Magn. 33, 3580 (1997). 266. J. C. S. Kools, W. Kula, D. Mauri, and T. Lin, J. Appl. Phys. 85,
231. Y. Kawawake and H. Sakakima, IEEE Trans. Magn. 33, 3538 4466 (1999).
(1997). 267. S. Miura, M. Tsunoda, and M. Takahashi, J. Appl. Phys. 89, 6308
232. S. Maat, M. J. Carey, E. E. Fullerton, T. X. Le, P. M. Rice, and (2001).
B. A. Gurney, Appl. Phys. Lett. 81, 520 (2002). 268. R. Wood, IEEE Trans. Magn. 36, 36 (2000).
233. W. F. Egelhoff, Jr., P. J. Chen, C. J. Powell, M. D. Stiles, R. D. 269. R. Rottmayer and J. G. Zhu, IEEE Trans. Magn. 31, 2597 (1995).
McMichael, J. H. Judy, K. Takano, and A. E. Berkowitz, J. Appl. 270. K. Nagasaka, Y. Seyama, L. Varga, Y. Shimizu, and A. Tanaka,
Phys. 82, 6142 (1997). J. Appl. Phys. 89, 6943 (2001).
234. H. Sakakima, M. Satomi, Y. Sugita, and Y. Kawawake, J. Magn. 271. M. Takahashi, T. Funayama, K. Tateyama, M. Yoshikawa,
Magn. Mater. 210, 20 (2000). H. Iwasaki, and M. Sahashi, “Digest of MMM 2001,” 2001, paper
235. T. Kato, K. Miyashita, S. Iwata, S. Tsunashima, H. Sakakima, CB-01.
Y. Sugita, and Y. Kawawake, J. Magn. Magn. Mater. 240, 168 272. H. Yuasa, M. Yoshikawa, Y. Kamiguchi, K. Koi, H. Iwasaki,
(2002). M. Takagishi, and M. Sahashi, J. Appl. Phys. 92, 2646 (2002).
236. J. Hong, K. Aoshima, J. Kane, K. Noma, and H. Kanai, IEEE 273. A. Tanaka, Y. Shimizu, Y. Seyama, K. Nagasaka, R. Kondo,
Trans. Magn. 36, 2629 (2000). H. Oshima, S. Eguchi, and R. Kanai, IEEE Trans. Magn. 38, 84
237. Y. Tsuchiya, S. Li, M. Sano, T. Uesugi, S. Araki, H. Morita, and (2002).
M. Matsuzaki, IEEE Trans. Magn. 36, 2557 (2000). 274. K. Nagasaka, Y. Seyama, R. Kondo, H. Oshima, Y. Shimizu, and
238. M. Mao, C. Cerjan, B. Law, F. Grabner, L. Miloslavsky, and A. Tanaka, Fujitsu Sci. Tech. J. 37, 192 (2001).
C. Chien, IEEE Trans. Magn. 36, 2866 (2000). 275. M. Takahashi, K. Koi, M. Yoshikawa, T. Funayama, H. Iwasaki,
239. M. F. Gillies and A. E. T. Kuiper, J. Appl. Phys. 88, 5894 (2000). and M. Sahashi, “Digest of INTERMAG Europe 2002,” Amster-
240. A. Veloso, P. P. Freitas, P. Wei, N. P. Barradas, J. C. Soares, dam, The Netherlands, Apr. 2002, Paper CA02.
B. Almeida, and J. B. Sousa, Appl. Phys. Lett. 77, 1020 (2000). 276. K. Nagasaka, H. Oshima, Y. Seyama, Y. Shimizu, and A. Tanaka,
241. H. Kanai, K. Noma, and J. Hong, Fujitsu Sci. Tech. J. 37, 174 “Digest of INTERMAG Europe 2002,” Amsterdam, The Nether-
(2001). lands, Apr. 2002, paper AC04.
242. J. Hong, K. Noma, and H. Kanai, J. Appl. Phys. 89, 6940 (2001). 277. Y. H. Wu, K. B. Li, Z. B. Guo, J. J. Qiu, G. C. Han, and T. C.
243. H. Fukuzawa, K. Koi, H. Tomita, H. N. Fuke, Y. Kamiguchi, Chong, “Digest of INTERMAG Europe 2002,” Amsterdam, The
H. Iwasaki, and M. Sahashi, J. Magn. Magn. Mater. 235, 208 (2001). Netherlands, Apr. 2002, paper CA04.
Nano Spintronics for Data Storage 47

278. A. Matsuzono, S. Terada, H. Ono, A. Furukawa, T. Sone, S. Sasaki, 312. P. K. de Boer, G. A. de Wijs, and R. A. de Groot, Phys. Rev. B 58,
Y. Kakihara, Y. Takeda, N. Chiyokubo, and H. Matsuki, J. Appl. 15422 (1998).
Phys. 91, 7267 (2002). 313. I. I. Oleinik, E. Y. Tsymbal, and D. G. Pettifor, Phys. Rev. B 62,
279. M. Hosomi, E. Makino, I. Konishiike, N. Sugawara, and 3952 (2000).
S. Ohkawara, J. Appl. Phys. 91, 8099 (2002). 314. E. Y. Tsymbal, I. I. Oleinik, and D. G. Pettifor, J. Appl. Phys. 87,
280. W. P. Pratt, Jr., S.-F. Lee, J. M. Slaughter, R. Loloee, P. A. 5230 (2000).
Schroeder, and J. Bass, Phys. Rev. B 66, 3060 (1991). 315. B. D. Schrag, A. Anguelouch, S. Ingvarsson, G. Xiao, Y. Lu, P. L.
281. S. F. Lee, W. P. Pratt, Jr., R. Loloee, P. A. Schroeder, and J. Bass, Trouilloud, A. Gupta, R. A. Wanner, W. J. Gallagher, P. M. Rice,
Phys. Rev. B 46, 548 (1992). and S. S. P. Parkin, Appl. Phys. Lett. 77, 2373 (2000).
282. G. E. W. Bauer, Phys. Rev. Lett. 69, 1676 (1992). 316. L. F. Li, X. Y. Liu, and G. Xiao, J. Appl. Phys. 93, 467 (2003).
283. S. Zhang and P. M. Levy, J. Appl. Phys. 69, 4786 (1991). 317. S. Tegen, I. Monch, J. Schumann, H. Vinzelberg, and C. M.
284. T. Valet and A. Fert, Phys. Rev. B 48, 274 (1999). Schneider, J. Appl. Phys. 89, 8169 (2001).
285. S.-F. Lee, W. P. Pratt, Jr., Q. Yang, P. Holody, R. Loloee, P. A. 318. C. Tiusan, M. Hehn, and K. Ounadjela, Eur. Phys. J. B 26, 431
Schroeder, and J. Bass, J. Magn. Magn. Mater. 118, L1 (1993). (2002).
286. M. A. M. Gijs, S. K. J. Lenczowski, and J. B. Giesbers, Phys. Rev. 319. J. J. Sun, K. Shimazawa, N. Kasahara, K. Sato, T. Kagami,
Lett. 70, 3343 (1993). S. Saruki, S. Araki, and M. Matsuzaki, J. Appl. Phys. 89, 6653
287. M. A. M. Gijs, M. T. Johnson, A. Reinders, P. E. Huisman, (2001).
R. J. M. van de Veerdonk, S. K. J. Lenczowski, and R. M. J. van 320. J. E. Ortega, A. Närmann, K. N. Altmann, W. O’Brien, D. J.
Gansewinkel, Appl. Phys. Lett. 66, 1839 (1995). Seo, F. J. Himpsel, P. Segovia, A. Mascaraque, and E. G. Michel,
288. P. M. Levy, S. Zhang, T. Ono, and T. Shinjo, Phys. Rev. B 52, 16049 J. Magn. Magn. Mater. 203, 126 (1999).
(1995). 321. R. K. Kawakami, E. Rotenberg, E. J. Escorcia-Aparicio, H. J.
289. A. Vedyayev, M. Chshiev, N. Ryzhanova, B. Dieny, C. Cowache, Choi, T. R. Cummins, J. G. Tobin, N. V. Smith, and Z. Q. Qiu,
and F. Brouers, J. Magn. Magn. Mater. 172, 53 (1997). Phys. Rev. Lett. 80, 1754 (1998).
290. A. C. Reilly, W. Park, R. Slater, B. Ouaglal, R. Loloee, W. P. Pratt, 322. A. Vedyayev, N. Ryzhanova, C. Lacroix, L. Giacomoni, B. Dieny,
Jr., and J. Bass, J. Magn. Magn. Mater. 195, L269 (1999). Europhys. Lett. 39, 219 (1997).
291. J. Bass and W. P. Pratt, Jr., J. Magn. Magn. Mater. 200, 274 (1999). 323. J. Mathon and A. Umerski, Phys. Rev. B 60, 1117 (1999).
292. J. Y. Gu, S. D. Steenwyk, A. C. Reilly, W. Park, R. Loloee, J. Bass, 324. S. Zhang and P. M. Levy, Phys. Rev. Lett. 81, 5660 (1998).
and W. P. Pratt, Jr., J. Appl. Phys. 87, 4831 (2000). 325. J. S. Moodera, J. Nowak, L. R. Kinder, P. M. Tedrow, R. J. M.
293. T. Takagi, F. Koyama, and K. Iga, Appl. Phys. Lett. 59, 2877 (1991). van de Veerdonk, B. A. Smits, M. van Kampen, H. J. M. Swagten,
294. C. Stirtori, F. Capasso, J. Faist, D. L. Sivco, S.-N. G. Chu, and and W. J. M. de Jonge, Phys. Rev. Lett. 83, 3029 (1999).
A. Y. Cho, Appl. Phys. Lett. 61, 898 (1992). 326. S. S. P. Parkin, U.S. Patent 5,764,567, 1998.
295. F. J. Himpsel, J. E. Ortega, G. J. Mankey, and R. F. Willis, Adv. 327. J. J. Sun and P. P. Freitas, J. Appl. Phys. 85, 5264 (1999).
Phys. 47, 511 (1998) and references therein. 328. P. LeClair, H. J. M. Swagten, J. T. Kohlhepp, R. J. M. van de
296. S. Yuasa, T. Nagahama, and Y. Suzuki, Science 297, 234 (2002). Veedonk, and W. J. M. de Jonge, Phys. Rev. Lett. 84, 2933 (2000).
297. R. Meservey and P. M. Tedrow, Phys. Rep. 238, 173 (1994). 329. J. C. Slonczewski, Phys. Rev. B 39, 6995 (1989).
298. J. S. Moodera, J. Nassar, and G. Mathon, Annu. Rev. Mater. Sci. 330. P. LeClair, B. Hoex, H. Wieldraaijer, J. T. Kohlhepp, H. J. M.
29, 381 (1999). Swagten, and W. J. M. de Jonge, Phys. Rev. B 64, 100406 (2001).
299. W. J. Gallagher, S. S. P. Parkin, Y. Lu, X. P. Bian, A. Marley, K. P. 331. R. Schad, D. Allen, G. Zangari, I. Zana, D. Yang, M. Tondra, and
Roche, R. A. Altman, S. A. Rishton, C. Jahnes, T. M. Shaw, and D. X. Wang, Appl. Phys. Lett. 76, 607 (2000).
G. Xiao, J. Appl. Phys. 81, 3741 (1997). 332. M. F. Gillies, A. E. T. Kuiper, J. B. A. van Zon, and J. M. Sturm,
300. P. M. Tedrow and R. Moservey, Phys. Rev. Lett. 26, 192 (1971). Appl. Phys. Lett. 78, 3496 (2001).
301. R. J. Soulen, Jr., J. M. Byers, M. S. Osofsky, B. Nadgorny, 333. P. Rottlander, M. Hehn, O. Lenoble, and A. Schuhl, Appl. Phys.
T. Ambrose, S. F. Cheng, P. R. Broussard, C. T. Tanaka, J. Nowak, Lett. 78, 3274 (2001).
J. S. Moodera, A. Barry, and J. M. D. Coey, Science 282, 85 (1998). 334. L. S. Dorneles, R. L. Sommer, and L. F. Schelp, J. Appl. Phys. 91,
302. S. K. Upadhyay, A. Palanisami, R. N. Louie, and R. A. Buhrman, 7971 (2002).
Phys. Rev. Lett. 81, 3247 (1998). 335. M. F. Gillies, A. E. T. Kuiper, J. B. A. van Zon, and J. M. Sturm,
303. M. Bowen, M. Bibes, A. Barthélémy, J.-P. Contour, A. Anane, Appl. Phys. Lett. 78, 3496 (2001).
Y. Lemaître, and A. Fert, Appl. Phys. Lett. 82, 233 (2003). 336. M. M. Schwickert, J. R. Childress, R. E. Fontana, A. J. Kellock,
304. G. Hu and Y. Suzuki, Phys. Rev. Lett. 89, 276601 (2002). P. M. Rice, M. K. Ho, T. J. Thompson, and B. A. Gurney, J. Appl.
305. Rager, A. V. Berenov, L. F. Cohen, W. R. Branford, Y. V. Phys. 89, 6871 (2001).
Bugoslavsky, Y. Miyoshi, M. Ardakani, and J. L. MacManus- 337. J. Wang, S. Cardoso, P. P. Freitas, P. Wei, N. P. Barradas, and J. C.
Driscoll, Appl. Phys. Lett. 81, 5003 (2002). Soares, J. Appl. Phys. 89, 6868 (2001).
306. M. Sharma, S. X. Wang, and J. H. Nickel, Phys. Rev. Lett. 82, 616 338. P. Shang, A. K. Petford-Long, J. H. Nickel, M. Sharma, and T. C.
(1999). Anthony, J. Appl. Phys. 89, 6874 (2001).
307. J. M. De Teresa, A. Barthélémy, A. Fert, J. P. Contour, R. Lyonnet, 339. H. J. Shim, I. J. Hwang, K. S. Kim, B. K. Cho, J. T. Kim, and J. H.
F. Montaigne, P. Seneor, and A. Vaurès, Phys. Rev. Lett. 82, 4288 Sok, J. Appl. Phys. 92, 1095 (2002).
(1999). 340. J. Wang, P. P. Freitas, E. Snoeck, X. Battle, and J. Cuadra, J. Appl.
308. J. H. Park, E. Vescovo, H.-J. Kim, C. Kwon, R. Ramesh, and Phys. 91, 7463 (2002).
T. Venkatesan, Nature (London) 392, 794 (1998). 341. J. Wang, P. P. Freitas, E. Snoeck, P. Wei, and J. C. Soares, Appl.
309. A. Ferta, A. Barthélémya, J. Ben Youssefc, J.-P. Contoura, Phys. Lett. 79, 4387 (2001).
V. Crosa, J. M. De Teresaa, A. Hamzica, J. M. Georgea, G. Faini, 342. S. Kreuzer, W. Wegscheider, and D. Weiss, J. Appl. Phys. 89, 6751
b, J. Grolliera, H. Jaffrèsa, H. Le Gallc, F. Montaignea, F. Pail- (2001).
louxa, and F. Petroffa, Mater. Sci. Eng. B—Solid 84, 1 (2001). 343. Z. S. Li, C. de Groot, and J. H. Moodera, Appl. Phys. Lett. 77,
310. K. Ghosh, S. B. Ogale, S. P. Pai, M. Robson, E. Li, I. Jin, Z. Dong, 3630 (2000).
R. L. Greene, R. Ramesh, T. Venkatesan, and M. Johnson, Appl. 344. F. J. Cadieu, L. Chen, and B. Li, J. Appl. Phys. 91, 7968 (2002).
Phys. Lett. 73, 689 (1998). 345. M. Guth, V. Da Costa, G. Schmerber, A. Dinia, and H. A. M. van
311. A. Gupta, X. W. Li, and G. Xiao, Appl. Phys. Lett. 78, 1894 (2001). den Berg, J. Appl. Phys. 89, 6748 (2001).
48 Nano Spintronics for Data Storage

346. M. Guth, G. Schmerber, Y. Henry, and A. Dinia, J. Magn. Magn. 379. U. May, K. Samm, H. Kittur, J. Hauch, R. Calarco, U. Rüdiger,
Mater. 240, 152 (2002). and G. Güntherodt, Appl. Phys. Lett. 78, 2026 (2001).
347. Y. Lu, X. W. Li, G. Q. Gong, G. Xiao, A. Gupta, P. Lecoeur, J. Z. 380. P. Rottländer, H. Kohlstedt, P. Grünberg, and E. Girgis, J. Appl.
Sun, Y. Y. Wang, and V. P. Dravid, Phys. Rev. B 54, R8357–R8360 Phys. 87, 6067 (2000).
(1996). 381. H. Boeve, E. Girgis, J. Schelten, J. De Boeck, and G. Borghs,
348. T. Obata, T. Manako, Y. Shimakawa, and Y. Kubo, Appl. Phys. Appl. Phys. Lett. 76, 1048 (2000).
Lett. 74, 290 (1999). 382. B. F. P. Roos, P. A. Beck, S. O. Demokritov, B. Hillebrands, and
349. J. Hayakawa, K. Ito, S. Kokado, M. Ichimura, A. Sakuma, D. Ozkaya J. Appl. Phys. 89, 6656 (2001).
M. Sugiyama, H. Asano, and M. Matsui, J. Appl. Phys. 91, 8792 383. K. Shimazawa, N. Kasahara, J. J. Sun, S. Araki, H. Morita, and
(2002). M. Matsuzaki, J. Appl. Phys. 87, 5194 (2000).
350. D. C. Worledge and T. H. Geballe, Appl. Phys. Lett. 76, 900 (2000). 384. B. Park and T. D. Lee, J. Magn. Magn. Mater. 226, 926 (2001).
351. I. I. Oleinik, E. Y. Tsymbal, and D. G. Pettifor, Phys. Rev. B 65, 385. T. Schneider, H. Hegde, K. Sin, S. Funada, and S. Shi, IEEE Trans.
020401 (2002). Magn. 38, 2724 (2002).
352. S. Mitani, T. Moriyama, and K. Takanashik, J. Appl. Phys. 91, 7200 386. M. Tsunoda, K. Nishikawa, S. Ogata, and M. Takahashi, Appl.
(2002). Phys. Lett. 80, 3135 (2002).
353. F. Gustavsson, J. M. George, V. H. Etgens, and M. Eddrief, Phys. 387. X. F. Han, M. Oogane, H. Kubota, Y. Ando, and T. Miyazaki,
Rev. B 64, 184422 (2001). Appl. Phys. Lett. 77, 283 (2000).
354. D. M. Borsa, S. Grachev, and D. O. Boerma, IEEE Trans. Magn. 388. J. S. Bae, K. H. Shin, and H. M. Lee, J. Appl. Phys. 91, 7947 (2002).
38, 2709 (2002). 389. Y. Ando, M. Hayashi, S. Iura, K. Yaoita, C. C. Yu, H. Kubota,
355. D. J. Keavney, E. E. Fullerton, and S. D. Bader, J. Appl. Phys. 81, and T. Miyazaki, J. Phys. D: Appl. Phys. 35, 2415 (2002).
795 (1997). 390. T. Mitsuzuka, K. Matsuda, A. Kamijo, and H. Tsuge, J. Appl. Phys.
356. T. S. Plaskett, P. P. Freitas, N. P. Barradas, M. F. da Silva, and J. C. 85, 5807 (1999).
Soares, J. Appl. Phys. 76, 6104 (1994). 391. Y. Li and S. X. Wang, J. Appl. Phys. 91, 7950 (2002).
357. E. Popova, J. Faure-Vincent, C. Tiusan, C. Bellouard, H. Fis- 392. E. Y. Chen, R. Whig, J. M. Slaughter, D. Cronk, J. Goggin,
cher, M. Hehn, F. Montaigne, M. Alnot, S. Andrieu, A. Schuhl, G. Steiner, and S. Tehrani, J. Appl. Phys. 87, 6061 (2000).
E. Snoeck, and V. da Costa, Appl. Phys. Lett. 81, 1035 (2002). 393. B. J. Jönsson-Åkerman, R. Escudero, C. Leighton, S. Kim, I. K.
358. E. Popova, J. Faure-Vincent, C. Tiusan, C. Bellouard, H. Fis- Schuller, and D. A. Rabson, Appl. Phys. Lett. 77, 1870 (2000).
cher, M. Hehn, F. Montaigne, M. Alnot, S. Andrieu, A. Schuhl,
394. D. Allen, R. Schad, G. Zangari, I. Zana, D. Yang, M. Tondra, and
E. Snoeck, and V. da Costa, Appl. Phys. Lett. 81, 1035 (2002).
D. Wang, J. Appl. Phys. 87, 5188 (2000).
359. T. Kiyomura, Y. Maruo, and M. Gomi, J. Appl. Phys. 88, 4768
395. D. Allen, R. Schad, G. Zangari, I. Zana, M. Tondra, D. Wang, and
(2000).
D. Reed, J. Appl. Phys. 89, 6662 (2001).
360. C. L. Platt, B. Dieny, and A. E. Berkowitz, Appl. Phys. Lett. 69,
396. T. Dimopoulos, V. Da Costa, C. Tiusan, K. Ounadjela, and
2291 (1996).
H. A. M. van den Berg, J. Appl. Phys. 89, 7371 (2001).
361. D. J. Smith, M. R. McCartney, C. L. Platt, and A. E. Berkowitz,
397. Y. Ando, H. Kameda, H. Kubota, and T. Miyazaki, J. Appl. Phys.
J. Appl. Phys. 83, 5154 (1998).
87, 5206 (2000).
362. C. L. Platt, B. Dieny, and A. E. Berkowitz, J. Appl. Phys. 81, 5523
398. E. Z. Luo, S. K. Wong, A. B. Pakhomov, J. B. Xu, I. H. Wilson,
(1997).
and C. Y. Wong, J. Appl. Phys. 90, 5202 (2001).
363. L. S. Dorneles, R. L. Sommer, and L. F. Schelp, J. Appl. Phys. 91,
399. W. H. Rippard, A. C. Perrella, and R. A. Buhrman, Appl. Phys.
7971 (2002).
Lett. 78, 1601 (2001).
364. K. Matsuda, A. Kamijo, T. Mitsuzuka, and H. Tsuge, J. Appl. Phys.
400. W. H. Rippard, A. C. Perrella, and R. A. Buhrman, J. Appl. Phys.
85, 5261 (1999).
365. H. Tsuge and T. Mitsuzuka, Appl. Phys. Lett. 71, 3296 (1997). 89, 6642 (2001).
366. K. Matsuda, A. Kamijo, T. Mitsuzuka, and H. Tsuge, J. Appl. Phys. 401. B. Oliver, Q. He, X. Tang, and J. Nowak, J. Appl. Phys. 91, 4348
85, 5261 (1999). (2002).
367. H. Boeve, J. De Boeck, and G. Borghs, J. Appl. Phys. 89, 482 402. H. Kikuchi, M. Sato, and K. Kobayashi, Fujitsu Sci. Tech. J. 37, 183
(2001). (2001).
368. Z. G. Zhang, P. P. Freitas, A. R. Ramos, N. P. Barradas, and J. C. 403. M. Sato and K. Kobayashi, IEEE Trans. Magn. 33, 3553 (1997).
Soares, J. Appl. Phys. 91, 8786 (2002). 404. S. S. P. Parkin, K.-S. Moon, K. E. Pettit, D. J. Smith, R. E. Dunin-
369. K.-S. Moon, Y. Chen, and Y. Huai, J. Appl. Phys. 91, 7965 (2002). Borkowski, and M. R. McCartney, Appl. Phys. Lett. 75, 543 (1999).
370. P. LeClair, J. T. Kohlhepp, A. A. Smits, H. J. M. Swagten, B. Koop- 405. R. C. Sousa, J. J. Sun, V. Soares, P. P. Freitas, A. Kling, M. F. da
mans, and W. J. M. de Jonge, J. Appl. Phys. 87, 6070 (2000). Silva, and J. C. Soares, J. Appl. Phys. 85, 5258 (1999).
371. S. Cardoso, V. Gehanno, R. Ferreira, and P. P. Freitas, IEEE Trans. 406. J. Nowak, D. Song, and E. Murdock, J. Appl. Phys. 87, 5203 (2000).
Magn. 35, 2952 (1999). 407. Y. Ando, H. Kubota, M. Hayashi, M. Kamijo, K. Yaoita, A. Chak
372. B. F. P. Roos, P. A. Beck, S. O. Demokritov, B. Hillebrands, and Chung Yu, X.-F. Han, and T. Miyazaki, Jpn. J. Appl. Phys. 39, 5832
D. Ozkaya, J. Appl. Phys. 89, 6656 (2001). (2000).
373. K. Nishikawa, M. Tsunoda, S. Ogata, and M. Takahashi, IEEE 408. J. Schmalhorst, H. Brückl, G. Reiss, G. Gieres, and J. Wecker,
Trans. Magn. 38, 2718 (2002). J. Appl. Phys. 91, 6617 (2002).
374. K. S. Yoon, J. H. Park, J. H. Choi, J. Y. Yang, C. H. Lee, C. O. 409. T. Dimopoulos, C. Tiusan, V. da Costa, K. Ounadjela, and H. A. M.
Kim, J. P. Hong, and T. W. Kang, Appl. Phys. Lett. 79, 1160 (2001). van den Berg, Appl. Phys. Lett. 77, 3624 (2000).
375. K. S. Yoon, J. H. Park, J. Y. Yang, C. O. Kim, and J. P. Hong, 410. D. J. Keavney, S. Park, C. M. Falco, and J. M. Slaughter, Appl.
J. Appl. Phys. 91, 7953 (2002). Phys. Lett. 78, 234 (2001).
376. O. Song, Y. M. Lee, C. S. Yoon, and C. K. Kim, J. Appl. Phys. 93, 411. Z. Zhang, S. Cardoso, P. P. Freitas, P. Wei, N. Barradas, and J. C.
1146 (2003). Soares, Appl. Phys. Lett. 78, 2911 (2001).
377. K. Nishikawa, M. Tsunoda, S. Ogata, and M. Takahashi, IEEE 412. Z. Zhang, S. Cardoso, P. P. Freitas, X. Batlle, P. Wei, N. Barradas,
Trans. Magn. 38, 2718 (2002). and J. C. Soares, J. Appl. Phys. 89, 6665 (2001).
378. D. Song, J. Nowak, and M. Covington, J. Appl. Phys. 87, 5197 413. T. Dimopoulos, V. Da Costa, C. Tiusan, K. Ounadjela, and
(2000). H. A. M. van den Berg, Appl. Phys. Lett. 79, 3110 (2001).
Nano Spintronics for Data Storage 49

414. N. Matsukawa, A. Odagawa, Y. Sugita, Y. Kawashima, Y. Mori- 449. D. Song, J. Nowak, R. Larson, P. Kolbo, and R. Chellew, IEEE
naga, M. Satomi, M. Hiramoto, and J. Kuwata, Appl. Phys. Lett. Trans. Magn. 36, 2545 (2000).
81, 4784 (2002). 450. M. K. Ho, C. H. Tsang, R. E. Fontana, Jr., S. S. P. Parkin, K. J.
415. M. G. Samant, J. Lüning, J. Stöhr, and S. S. P. Parkin, Appl. Phys. Carey, P. Tao, S. MacDonald, P. C. Arnett, and J. O. Moore, IEEE
Lett. 76, 3097 (2000). Trans. Magn. 37, 1691 (2001).
416. C. S. Yoon, J. H. Lee, H. D. Jeong, C. K. Kim, J. H. Yuh, and 451. S. Araki, K. Sato, T. Kagami, S. Saruki, T. Uesugi, N. Kasahara,
R. Haasch, Appl. Phys. Lett. 80, 3976 (2002). T. Kuwashima, N. Ohta, J. Sun, K. Nagai, S. Li, N. Hachisuka,
417. Y. Saito, M. Amano, K. Nakajima, S. Takahashi, and M. Sagoi, H. Hatate, T. Kagotani, N. Takahashi, K. Ueda, and M. Matsuzaki,
J. Magn. Magn. Mater. 223, 293 (2001). IEEE Trans. Magn. 38, 73 (2002).
418. J. H. Lee, S. J. Kim, C. S. Yoon, C. K. Kim, B. G. Park, and T. D. 452. K. Machida, N. Hayashi, Y. Miyamoto, T. Tamaki, and H. Okuda,
Lee, J. Appl. Phys. 92, 6241 (2002). J. Magn. Magn. Mater. 201, 235 (2001).
419. M. A. M. Gijs, J. B. Giesbers, S. K. J. Lenczowski, and H. H. J. M. 453. K. Ishihara, M. Nakada, E. Fukami, K. Nagahara, H. Honjo, and
Janssen, Appl. Phys. Lett. 63, 111 (1993). K. Ohashi, IEEE Trans. Magn. 37, 1687 (2001).
420. S. K. J. Lenczowski, R. J. M. van de Veerdonk, M. A. M. Gijs, 454. K. Inomata and H. Sakakima, in “Giant Magneto-Resistance
J. B. Giesbers, and H. H. J. M. Janssen, J. Appl. Phys. 75, 5154 Devices” (E. Hirota, H. Sakakima, and K. Inomata), p. 133.
(1994). Springer, Berlin, 2002.
421. R. J. M. vandeVeerdonk, J. Nowak, R. Meservey, J. S. Moodera, 455. K. Shimazawa, J. J. Sun, N. Kasahara, K. Sato, T. Kagami,
and W. J. M. deJonge, Appl. Phys. Lett. 71, 2839 (1997). S. Saruki, O. Redon, Y. Fujita, T. Umehara, J. Syoji, S. Araki, and
422. K. Matsuda, N. Watari, A. Kamijo, and H. Tsuge, Appl. Phys. Lett. M. Matsuzaki, IEEE Trans. Magn. 37, 1684 (2001).
77, 3060 (2000). 456. J. C. Jury and S. X. Wang, IEEE Trans. Magn. 38, 295 (2002).
423. J. S. Moodera, L. R. Kinder, J. Nowak, P. LeClair, and R. Meservy, 457. S. Ingvarsson, G. Xiao, S. S. P. Parkin, W. J. Gallagher, G. Grin-
Appl. Phys. Lett. 69, 708 (1996). stein, and R. H. Koch, Phys. Rev. Lett. 85, 3289 (2000).
424. J. Chen, Y. Li, J. Nowak, and J. Fernandez de-Castro, J. Appl. 458. E. R. Nowak, M. B. Weissman, and S. S. P. Parkin, Appl. Phys.
Phys. 91, 8783 (2002). Lett. 74, 600 (1999).
425. H. Kikuchi, M. Sato, and K. Kobayashi, Fujitsu Sci. Tech. J. 37, 183 459. S. Ingvarsson, G. Xiao, R. A. Wanner, P. Trouilloud, Y. Lu, W. J.
(2001). Gallagher, A. Marley, K. P. Roche, and S. S. P. Parkin, J. Appl.
426. K. B. Li, Y. H. Wu, J. J. Qiu, G. C. Han, Z. B. Guo, and T. C. Phys. 85, 5270 (1999).
Chong, J. Magn. Magn. Mater. 241, 89 (2002). 460. D. S. Reed, C. Nordman, and J. M. Daughton, IEEE Trans. Magn.
427. J. S. Moodera, L. R. Kinder, T. M. Wong, and R. Meservey, Phys. 37, 2028 (2001).
Rev. Lett. 74, 3273 (1995). 461. K. S. Kim, H. J. Shim, I. J. Hwang, B. K. Cho, J. H. Seok, and
428. J. S. Moodera, J. Nowak, and R. J. M. van de Veerdonk, Phys. J.-T. Kim, J. Appl. Phys. 91, 8804 (2002).
Rev. Lett. 80, 2941 (1998). 462. P. K. George, Y. Wu, R. M. White, E. Murdock, and M. Tondra,
429. A. C. Marleyand and S. S. P. Parkin, J. Appl. Phys. 81, 5526 (1997). Appl. Phys. Lett. 80, 682 (2002).
430. J. Zhang and R. M. White, J. Appl. Phys. 83, 6512 (1998). 463. N. Smith and P. Arnett, Appl. Phys. Lett. 78, 1448 (2001).
431. Y. Lu, X. W. Li, G. Xiao, R. A. Altman, W. J. Gallagher, A. Mar- 464. S. Mao, J. Nowak, E. Linville, H. Chen, O. Heinonen, H. Cho,
ley, K. Roche, and S. Parkin, J. Appl. Phys. 83, 6515 (1998). L. Wang, P. Anderson, M. Ostrowski, B. Karr, P. Kolbo, and
432. R. S. Beech, J. Anderson, J. Daughton, B. A. Everitt, and D. Wang, Z. Zhang, “47th Annual Conference on Magnetism and Magnetic
IEEE Trans. Magn. 32, 4713 (1996). Materials,” Tampa, FL, Nov. 2002, paper DC-01.
433. N. Tezuka and T. Miyazaki, Jpn. J. Appl. Phys. 37, L218 (1998). 465. Y. Irie, H. Sakakima, M. Satomi, and Y. Kawawake, Jpn. J. Appl.
434. J. J. Sun and P. P. Freitas, J. Appl. Phys. 85, 5264 (1999). Phys. 34, L415 (1995).
435. C. Heide, A. I. Krikunov, Yu. F. Ogrin, P. E. Zilberman, and R. J. 466. J. M. Dauton, J. Appl. Phys. 81, 3758 (1997).
Elliott, J. Appl. Phys. 87, 5221 (2000). 467. R. C. Sousa, P. P. Freitas, V. Chu, and J. P. Conde, IEEE Trans.
436. G. G. Cabrera and N. García, Appl. Phys. Lett. 80, 1782 (2002). Magn. 35, 2832 (1999).
437. X. H. Xiang, T. Zhu, J. Du, G. Landry, and J. Q. Xiao, Phys. Rev. 468. J. M. Slaughter, R. W. Dave, M. DeHerrera, M. Durlam, B. N.
B 66, 174407 (2002). Engel, J. Janesky, N. D. Rizzo, and S. Tehrani, J. Supercond. 15,
438. S. Zhang, P. M. Levy, A. C. Marley, and S. S. P. Parkin, Phys. Rev. 19 (2002).
Lett. 79, 3744 (1997). 469. S. Tehrani, B. Engel, J. M. Slaughter, E. Chen, M. DeHerrera,
439. Y. Saito, M. Amano, K. Nakajima, S. Takahashi, M. Sagoi, and M. Durlam, P. Naji, R. Whig, J. Janesky, and J. Calder, IEEE
K. Inomata, Jpn. J. Appl. Phys. 39, L1035 (2000). Trans. Magn. 36, 2752 (2000).
440. Y. Saito, M. Amano, K. Nakajima, S. Takahashi, and M. Sagoi, 470. D. Wang, M. Tondra, A. V. Pohm, C. Nordman, J. Anderson, J. M.
J. Magn. Magn. Mater. 223, 293 (2001). Daughton, and W. C. Black, J. Appl. Phys. 87, 6385 (2000).
441. B. G. Park and T. D. Lee, Appl. Phys. Lett. 81, 2214 (2002). 471. J. G. Zhu, Y. F. Zheng, and G. A. Prinz, J. Appl. Phys. 87, 6668
442. C. H. Shang, J. Nowak, R. Jansen, and J. S. Moodera, Phys. Rev. (2000).
B 58, R2917 (1998). 472. K. Inomata, Y. Saito, K. Nakajima, and M. Sagoi, J. Appl. Phys.
443. U. Rüdiger, R. Calarco, U. May, K. Samm, J. Hauch, H. Kittur, 87, 6064 (2000).
M. Sperlich, and G. Güntherodt, J. Appl. Phys. 89, 7573 (2001). 473. H. Boeve, C. Bruynseraede, J. Das, K. Dessein, G. Borghs, J. De
444. T. Hagler, R. Kinder, and G. Bayreuther, J. Appl. Phys. 89, 7570 Boeck, R. C. Sousa, L. V. Melo, and P. P. Freitas, IEEE. Trans.
(2001). Magn. 35, 2820 (1999).
445. J. Wingbermühle, S. Stein, and H. Kohlstedt, J. Appl. Phys. 92, 474. Y. Irie, H. Sakakima, M. Satomi, and Y. Kawawake, Jpn. J. Appl.
7261 (2002). Phys. 34, L415 (1995).
446. J. H. Lee, K. I. Lee, W. L. Lee, K.-H. Shin, J. S. Lee, K. Rhie, 475. Y. Zheng and J.-G. Zhu, IEEE Trans. Magn. 33, 3286 (1997).
and B. C. Lee, J. Appl. Phys. 91, 7956 (2002). 476. J. G. Zhu and Y. F. Zheng, IEEE Trans. Magn. 34, 1063 (1998).
447. R. Coehoorn, S. R. Cumpson, J. J. M. Ruigrok, and P. Hidding, 477. B. A. Everitt, A. V. Pohm, R. S. Beech, A. Fink, and J. M.
IEEE Trans. Magn. 35, 2586 (1999). Daughton, IEEE Trans. Magn. 34, 1060 (1998).
448. K. Shimazawa, O. Redon, N. Kasahara, J. J. Sun, K. Sato, 478. B. A. Everitt, A. V. Pohm, and J. M. Daughton, J. Appl. Phys. 81,
T. Kagami, S. Saruki, T. Umehara, Y. Fujita, J. Yamazaki, S. Araki, 4020 (1997).
H. Morita, and M. Matsuzaki, IEEE Trans. Magn. 36, 2542 (2000). 479. R. R. Katti, J. Appl. Phys. 91, 7245 (2002).
50 Nano Spintronics for Data Storage

480. H. Boeve, J. Das, C. Bruynseraede, J. De Boeck, and G. Borghs, 503. M. Tsoi, A. G. M. Jansen, J. Bass, W.-C. Chiang, M. Seck, V. Tsoi,
Electron. Lett. 34, 1754 (1998). and P. Wyder, Phys. Rev. Lett. 80, 4281 (1998).
481. T. N. Fang and J. G. Zhu, IEEE Trans. Magn. 37, 1963 (2001). 504. J.-E. Wegrowe, D. Kelly, Y. Jaccard, Ph. Guittienne, and J.-Ph.
482. P. Shang, A. K. Petford-Long, and T. C. Anthony, J. Appl. Phys. Ansermet, Europhys. Lett. 45, 626 (1999).
91, 7703 (2002). 505. J. Z. Sun, J. Magn. Magn. Mater. 202, 157 (1999).
483. J. Gadbois and J.-G. Zhu, J. Appl. Phys. 34, 1066 (1998). 506. E. B. Myers, D. C. Ralph, J. A. Katine, R. N. Louie, and R. A.
484. J. G. Zhu and Y. F. Zheng, Top. Appl. Phys. 83, 289 (2002). Buhrman, Science 285, 867 (1999).
485. K. J. Kirk, J. N. Chapman, and C. D. W. Wilkinson, Appl. Phys. 507. E. B. Myers, D. C. Ralph, J. A. Katine, F. J. Albert, and R. A.
Lett. 71, 539 (1997). Buhrman, J. Appl. Phys. 87, 5502 (2000).
486. K. J. Kirk, J. N. Chapman, and C. D. W. Wilkinson, J. Appl. Phys. 508. J. Grollier, V. Cros, A. Hamzic, J. M. George, H. Jaffrès, A. Fert,
85, 5237 (1999). G. Faini, J. Ben Youssef, and H. Legall, Appl. Phys. Lett. 78, 3663
487. G. Yi, P. R. Aitchison, W. D. Doyle, J. N. Chapman, and C. D. W. (2001).
Wilkinson, J. Appl. Phys. 92, 6087 (2002). 509. W. W. Zhuang, W. Pan, B. D. Ulrich, J. J. Lee, L. Stecker, A. Bur-
488. F. J. Castano, Y. Hao, S. Haratani, C. A. Ross, B. Vogeli, master, D. R. Evans, S. T. Hsu, M. Tajiri, A. Shimaoka, K. Inoue,
M. Walsh, and H. I. Smith, IEEE Trans. Magn. 37, 2073 (2001). T. Naka, N. Awaya, K. Sakiyama, Y. Wang, S. Q. Liu, N. J. Wu,
489. J. Janesky, N. D. Rizzo, L. Savtchenko, B. Engel, J. M. Slaughter, and A. Ignatiev, “2002 International Electron Devices Meeting
and S. Tehrani, IEEE Trans. Magn. 37, 2052 (2001). Technical Digest,” San Francisco, CA, Sept. 2002, pp. 193–196.
490. S. E. Russek, J. O. Oti, and Y. K. Kim, J. Magn. Magn. Mater. 199, 510. H. Ohno, D. Chiba, F. Matsukura, T. Omiya, E. Abe, T. Dietl,
6 (1999). Y. Ohno, and K. Ohtani, Nature 408, 944 (2000).
491. M. Ooishi, Nikkei Electron. 839, 90 (2003) (in Japanese). 511. S. Datta and B. Das, Appl. Phys. Lett. 56, 665 (1990).
492. S. Tehrani, M. Durlam, J. Åkerman, M. DeHerrera, J. Salter, 512. D. J. Monsma, J. C. Lodder, Th. J. A. Popma, and B. Dieny, Phys.
J. Slaughter, J. Janesky, B. Engel, and N. Rizzo, Nikkei Electron. Rev. Lett. 74, 5260 (1995).
835, 127 (2002) (in Japanese). 513. T. Hayashi, H. Shimada, H. Shimizu, and M. Tanaka, J. Cryst.
493. R. C. Sousa, Z. Zhang, and P. P. Freitas, J. Appl. Phys. 91, 7700 Growth 201/202, 689 (1999).
(2002). 514. D. Chiba, N. Akiba, F. Matsukura, Y. Ohno, and H. Ohno, Appl.
494. K. Inomata, Y. Saito, K. Nakajima, and M. Sagoi, J. Appl. Phys. Phys. Lett. 77, 1873 (2000).
87, 6064 (2000). 515. J. Fabian, I. Žutić, and S. D. Sarma, Phys. Rev. B 66, 165301 (2002).
495. F. Z. Wang, Appl. Phys. Lett. 77, 2036 (2000). 516. M. Johnson, B. R. Bennett, M. J. Yang, M. M. Miller, and B. V.
496. Y. K. Zheng, X. Y. Wang, D. You, and Y. H. Wu, Appl. Phys. Lett. Shanabrook, Appl. Phys. Lett. 71, 974 (1997).
79, 2788 (2001). 517. S. van Dijken, X. Jiang, and S. S. P. Parkin, Appl. Phys. Lett. 80,
497. R. S. Beech, J. A. Anderson, A. V. Pohm, and J. M. Daughton, 3364 (2002).
J. Appl. Phys. 87, 6403 (2000). 518. R. Fiederling, M. Keim, G. Reuscher, W. Ossau, G. Schmidt,
498. W. C. Jeong, B. I. Lee, and S. K. Joo, J. Appl. Phys. 85, 4782 (1999). A. Waag, and L. W. Molenkamp, Nature 402, 787 (1999).
499. W. C. Jeong, B. I. Lee, and S. K. Joo, IEEE Trans. Magn. 34, 1069 519. Y. Ohno, D. K. Young, B. Beschoten, F. Matsukura, H. Ohno, and
(1998). D. D. Awschalom, Nature 402, 790 (1999).
500. Y. K. Zheng, Y. H. Wu, Z. B. Guo, G. C. Han, K. B. Li, J. J. Qiu, 520. B. T. Jonker, Y. D. Park, B. R. Bennett, H. D. Cheong,
H. Xie, and P. Luo, IEEE Trans. Magn. 38, 2850 (2002). G. Kioseoglou, and A. Petrou, Phys. Rev. B 62, 8180 (2000).
501. J. Slonczewski, J. Magn. Magn. Mater. 159, L1 (1996). 521. N. García, M. Muñoz, G. G. Qian, H. Rohrer, I. G. Saveliev, and
502. Ya. B. Bazaliy, B. A. Jones, and S. C. Zhang, J. Appl. Phys. 89, Y.-W. Zhao, Appl. Phys. Lett. 79, 4550 (2001).
6793 (2001). 522. S. Z. Hua and H. Deep Chopra, Phys. Rev. B 67, 060401 (2003).