Magnetically Engineered Spintronic Sensors and
Memory
STUART PARKIN, SENIOR MEMBER, IEEE, XIN JIANG, CHRISTIAN KAISER, ALEX PANCHULA,
KEVIN ROCHE, AND MAHESH SAMANT
Invited Paper
The discovery of enhanced magnetoresistance and oscillatory
interlayer exchange coupling in transition metal multilayers just
over a decade ago has enabled the development of new classes of
magnetically engineered magnetic thin-film materials suitable for
advanced magnetic sensors and magnetic random access memories. Magnetic sensors based on spin-valve giant magnetoresistive
(GMR) sandwiches with artificial antiferromagnetic reference
layers have resulted in enormous increases in the storage capacity
of magnetic hard disk drives. The unique properties of magnetic
tunnel junction (MTJ) devices has led to the development of an
advanced high performance nonvolatile magnet random access
memory with density approaching that of dynamic random-access
memory (RAM) and read-write speeds comparable to static RAM.
Both GMR and MTJ devices are examples of spintronic materials
in which the flow of spin-polarized electrons is manipulated
by controlling, via magnetic fields, the orientation of magnetic
moments in inhomogeneous magnetic thin film systems. More
complex devices, including three-terminal hot electron magnetic
tunnel transistors, suggest that there are many other applications
of spintronic materials.
Keywords—Field sensor, giant magnetoresistance (GMR),
magnetic engineering, magnetic random-access memory (MRAM),
magnetic recording, magnetic tunneling junction (MTJ), magnetic
tunneling, magnetoelectronics, magnetoresistance, oscillatory
interlayer coupling, read head, spin-dependent transport, spin
valve, spintronics.
I. INTRODUCTION
Conventional magnetic materials have long been used for
various near-ubiquitous applications including electric motors and magnetic compasses and sensors [1]. In recent years
a new class of magnetic materials has emerged based on the
microscopic generation and manipulation of spin-polarized
electrons in magnetic multilayered thin-film structures [2].
Manuscript received February 15, 2003; revised March 10, 2003. This
work was supported in part by the Defense Advanced Research Projects
Agency.
The authors are with the IBM Almaden Research Center, San Jose, CA
95120-6099 USA (e-mail: parkin@almaden.ibm.com).
Digital Object Identifier 10.1109/JPROC.2003.811807
In particular, these materials can act as extremely sensitive
magnetic field sensors, because their electrical resistance can
change in the presence of magnetic fields at room temperature by factors much larger than are possible with conventional magnetic materials. In this article we will focus on two
classes of novel materials, metallic magnetic multilayered
structures and magnetic tunnel junctions, and their applications to magnetic information storage in the form of magnetic hard disk drives and magnetic random access memory
[3]. We will also briefly discuss more complex devices based
on these materials, in particular the magnetic tunnel transistor (MTT). In order to make technologically useful devices, these materials have to be magnetically engineered so
as to control and tune their response to magnetic fields. The
discovery of oscillatory interlayer coupling in 1989 in transition metal based magnetic multilayers [4] plus the phenomenon of exchange biasing discovered much earlier in 1959
[5] together form the basis of the magnetic engineering of
many of today’s most useful magnetic nanostructures.
II. SPIN-VALVE MAGNETIC RECORDING READ HEADS
In a magnetic recording hard disk drive information is
stored by magnetizing regions within a magnetic thin film
(see Fig. 1). The transitions between these regions represent “bits” which are detected, via their fringing magnetic
fields, by the read sensor. The read sensor is part of a merged
read–write recording head which has a separate writing element. The recording head is attached to a small ceramic
“slider” which is flown on an air bearing above the recording
medium at a height of just a few nanometers. The number of
magnetic bits per unit area, the areal density, has increased
at compound growth rates (CGRs) exceeding 100% for the
past several years (see Fig. 2). These astounding increases
in storage capacity have been driven in large part by a new
generation of magnetic recording read heads based on the
phenomenon of giant magnetoresistance (GMR), which were
0018-9219/03$17.00 © 2003 IEEE
PROCEEDINGS OF THE IEEE, VOL. 91, NO. 5, MAY 2003
661
Fig. 1. Fundamentals of magnetic recording. Schematic diagram of a hard disk drive. (a)
Information is stored by magnetizing regions of a thin magnetic film on the surface of a disk. Bits are
detected by sensing the magnetic fringing fields of the transitions between adjacent regions as the
disk is rotated beneath a magnetic sensor. As the area of the magnetized region has decreased,
the read element has had to scale down in size accordingly. (b) The read element is incorporated
into a merged read/write head, which is mounted on the rear edge of a ceramic slider flown above
the surface of the rapidly spinning disk via a cantilevered suspension. A hard drive unit usually
consists of a stack of several such head–disk assemblies plus all the motors and control electronics
required for operation (for more details see, for example, [7]).
first developed and introduced by IBM in late 1997 [2]. GMR
heads are now found in virtually all hard disk drives produced
today.
From the earliest days of magnetic recording more than
50 years ago [6] there have been a succession of predictions
of the ultimate achievable recording density but so far
these have always been surpassed. As the areal density has
increased and the bit size correspondingly decreased many
662
technological challenges have been met. These challenges
[7] range from mechanical issues such as reduced flying
height and contamination sensitivity, servoing and head
stiction; read issues such as magnetic shielding and sensor
sensitivity; write issues concerning field strength and speed;
media issues including noise and data stability as bit size
approaches the super-paramagnetic limit. One of the most
important technological hurdles has been and continues to be
PROCEEDINGS OF THE IEEE, VOL. 91, NO. 5, MAY 2003
Fig. 2. Increases in areal density and shipped capacity of magnetic storage over time. Since the
invention of magnetic disk recording in the 1950s, the areal density (bits stored per square inch) of
disk drives has increased rapidly. Fueled by the development of the anisotropic magnetoresistance
(AMR) and GMR read sensors, in recent years its compound growth rate (CGR) has exceeded 100%.
Developments in the magnetic media have kept pace with those of the read sensor and the write head.
For example, antiferromagnetically coupled (AFC) recording media allow the writing of narrower
tracks on media which would otherwise be susceptible to the super-paramagnetic effect. The inset
plot shows the total capacity of hard drives shipped per year; in 2002 that shipped capacity was 10
EB worth of data. [Data provided by Ed Grochowski (private communication)].
the need to extract ever more read signal from diminishing
volumes of magnetic material.
The read signal from a GMR recording head is one to two
orders of magnitude bigger than that from prior generation
state-of-the-art read heads [8] which were based on the phenomenon of anisotropic magnetoresistance (AMR) [9]. The
resistance of conventional ferromagnetic metals varies as apwhere is the angle between the direcproximately
tion of the magnetic moment of the ferromagnet and the direction of current flow. The fringing fields from transitions
in the magnetic media are then detected by changes in resistance of the read sensor element as the media is rotated at
great speed under the sensor. These fields are quite small in
the range of 10–100 Oe. The active component of an AMR
sensor is essentially an ultrathin magnetoresistive ferromag) whose plane is
netic layer (typically permalloy,
arranged to be orthogonal to that of the magnetic media. The
sensing layer is sandwiched between and electrically isolated
from two much thicker ultrasoft magnetic layers which act as
shields to ensure that the sense layer measures flux from only
one transition. The separation of these shields plus the thickness of the sense layer determines the spatial resolution of
the read head.
AMR, usually quite small—just a few percent at room
temperature, is a consequence of bulk scattering, and
moreover, in thin films typically decreases with decreasing
film thickness as scattering from the surfaces of the film
becomes more important [10]. This modest magnitude of
AMR led to predictions that areal densities would be limited
to
Gb/in . In contrast to AMR, GMR is a much larger
effect of up to more than 100% at room temperature [11]
and is dominated by interface scattering [12], [13]. The
GMR effect was originally discovered in MBE (molecular
beam epitaxy) grown epitaxial (100) oriented Fe/Cr/Fe
sandwiches [14] and Fe/Cr multilayers [15] but the effects
were quite modest at room temperature. Shortly afterwards
it was discovered that similar effects could be found in
polycrystalline sputtered Fe/Cr multilayers [4] and subsequently very large room temperature magnetoresistance was
found in Co/Cu and related multilayers [11], [16]. These
latter materials form the basis of GMR sensors and storage
devices today. AMR and GMR are compared with other
classes of structures and materials which display significant
magnetoresistance in Fig. 3.
GMR is a result of spin-dependent scattering in inhomogeneous magnetic metallic systems (for recent reviews
see, for example, [17] and [18]). As originally discussed
by Mott in the 1930s [19], [20] current in 3d transition ferromagnetic metals is carried independently by spin-up
and spin-down
electrons. According to this two-current
model [21], [22] the scattering rates can be quite different for
electrons within these two channels. In the most naïve model
the current is considered to be carried predominantly by the
electrons but the heavier electrons, which
low-mass
PARKIN et al.: MAGNETICALLY ENGINEERED SPINTRONIC SENSORS AND MEMORY
663
Fig. 3. Types of magnetoresistance. (a) AMR results from bulk spin-polarized scattering within
a ferromagnetic metal; it is manifested as a dependence of the resistance on the angle between
an applied external field and the direction of current flowing through the material. (b) Colossal
magnetoresistance (CMR) results from interactions predominantly between adjacent atoms in
certain crystalline perovskites. (c) GMR results from interfacial spin-polarized scattering between
ferromagnets separated by conducting spacers in a heterogeneous magnetic material, such as a
magnetic multilayer or granular alloy. (d) Tunneling magnetoresistance (TMR) in magnetic tunnel
junctions results from spin filtering as spin-polarized electrons tunnel across an insulating barrier
from one ferromagnet to another. (e) Anomalous MR from domain wall effects has been observed in
single-crystal ferromagnetic Fe whiskers and patterned magnetic wires. Spin-polarized scattering of
electrons as they cross from one domain to another leads to increased resistance in a GMR-like
model, although both increased and decreased resistance has been observed in lithographically
patterned magnetic elements in the presence of domain walls. (f) Ballistic MR (BMR) is another
type of domain wall effect in the limit of very narrow constrictions where the conductance may be
quantized. Very large BMR effects have been observed but their origin, whether from spin-polarized
transport, or from magnetostrictive effects is controversial.
give rise to the ferromagnetism, provide a spin-dependent
reservoir of empty states into which these sp electrons can be
scattered. Since the density of states (DOS) of and electrons at the Fermi energy are quite distinct this rationalizes
664
the spin-dependent scattering rates. In an inhomogeneous
magnetic system where the magnetic moment direction
varies spatially the scattering rates for and electrons will
also vary in space. In particular, as shown schematically in
PROCEEDINGS OF THE IEEE, VOL. 91, NO. 5, MAY 2003
Fig. 3(c), magnetic multilayers will typically have a lower
resistance when the magnetic moments of the individual
layers are parallel (P) than when antiparallel (AP) [23].
Defining the magnetoresistance as
the largest GMR effects at room temperature of 110%
have been found in sputter deposited Co/Cu multilayers
[11]. For GMR the MR varies as the cosine of the angle
between the moments of adjacent ferromagnetic layers.
The largest GMR effects in magnetic multilayers are observed for structures containing the thinnest possible magnetic and nonmagnetic layers. This is because the GMR effect is dominated by spin-dependent scattering at the magnetic/nonmagnetic interfaces [12], [13] and, especially for
current flow in the plane of the multilayer, thickening these
layers largely results in shunting of current away from these
interfacial regions [24]. In the early days of GMR the preponderance of groups argued that GMR was largely a consequence of spin-dependent bulk scattering within the interior of the ferromagnetic layers [25] and the dominant role
of the interfacial contribution was not appreciated. Of course
in structures with very thick ( 100) ferromagnetic layers,
when the GMR effects are quite small, contributions from
bulk scattering may compare with those from interface scattering. However, for technological applications the interfacial origin of GMR makes the effect much more useful because only very thin ferromagnetic layers are needed to support large GMR effects. In particular, demagnetizing fields
associated with the magnetic moments of the ferromagnetic
layers within these devices will increase with the magnitude
of these magnetic moments so it is usually very important
to minimize the volume of magnetic material. For devices of
microscopic dimensions the self demagnetizing fields, which
depend in detail on the shape of the device, will increase the
magnetic field required to change the magnetic state of such
devices. For sensors this will eventually limit the smallest
fields which such devices can detect. For memory applications demagnetizing fields may lead to interactions between
closely packed neighboring devices which will eventually
limit the density of such memories. The understanding and
control of demagnetizing fields in magnetic nanostructures
is key to the fabrication of useful devices.
III. MAGNETIC ENGINEERING
Magnetic multilayers such as those shown schematically
in Fig. 3(c) are not by themselves useful for most sensor or
memory applications because the fields required to generate
large changes in the resistance of such multilayers are typically large and, moreover, are very sensitive to the thickness
of the nonmagnetic spacer layer. In addition, especially for
sensing applications, the maximum sensitivity to small magnetic fields is obtained for particular orientations of the magnetic moments within the sensor. For AMR read sensors [see
Fig. 4(a)] the moment of the magnetic layer must be set at
from the direction of the current flowing
an angle of
through the layer. When the moment is parallel or perpendicular to the current direction, for small changes in the moment
direction, there will be little change in resistance. Various ingenious schemes were developed to maximize the sensitivity
of AMR sensors by controlling the current flow through the
device and the quiescent moment direction [26]. Usually demagnetizing fields from a second soft ferromagnetic layer
separated from the sense layer by a nonmagnetic conducting
layer (e.g., Ta) would be used to help bias the magnetic moment of the sense layer at the correct angle. The moment
of the bias layer was itself set by the self-field of the sense
current passing through the device. The moment direction of
the sense layer was then determined by a combination of the
bias field, the magnetic anisotropy and shape demagnetizing
fields of the sense layer and the self-field of the sense current
[see Fig. 4(b)].
The magnitude of the magnetic fields required to saturate
the change in resistance of magnetic multilayers with thin
layers are very high and are determined by the intrinsic exchange coupling mediated through the nonmagnetic spacer
layers [27]. The coupling strength was discovered [4] to
oscillate between ferromagnetic (F) and antiferromagnetic
(AF) coupling with increasing spacer layer thickness for
almost all nonmagnetic transition metal spacer layers [28].
Companion oscillations in the magnitude of the GMR effect
were also found [4]. These latter oscillations simply reflect
the fact that for ferromagnetically coupled magnetic layers
in symmetric multilayer structures no relative change in
magnetic orientation of adjacent magnetic layers will result
with the application of magnetic field. By taking advantage
of the oscillatory interlayer coupling it was shown that
multilayer structures could be “spin engineered” [29] or
constructed in such a way that ferromagnetically coupled
neighboring magnetic layers could be arranged to become
antiparallel in intermediate magnetic fields. This allows both
the measurement of the ferromagnetic coupling strength
between these layers and the observation of a GMR effect
from these layers [29].
For antiferromagnetically coupled magnetic multilayers
the moments of adjacent magnetic layers naturally lie antiparallel in zero magnetic field. In a strong enough magnetic
field these moments will eventually align with the field becoming parallel to one another and so resulting in a change
in the resistance of the multilayer. The saturation field is directly related to the strength of the antiferromagnetic interlayer exchange coupling energy. The magnitude of the oscillatory interlayer coupling energy was found to vary systematically for magnetic multilayers with spacer layers formed
from the 3d, 4d, and 5d transition metals in the periodic
table, increasing in strength with d band filling and from the
5d to 4d to 3d elements [28]. For technological applications
one of the most useful elements is Ru which not only displays relatively high interlayer oscillatory exchange energy
but also, for certain ferromagnetic materials, displays antiferromagnetic coupling in the limit of ultrathin Ru layers as
thin as 2 to 3 [4], [29]. For GMR sensors, where additional
metallic layers will shunt and so dilute [24] the magnitude of
the effect, using a very thin AF coupling layer is important.
Moreover, Ru has other useful properties including excellent
thermal stability and growth habits.
PARKIN et al.: MAGNETICALLY ENGINEERED SPINTRONIC SENSORS AND MEMORY
665
Fig. 4. Comparison of AMR and GMR read sensors. (a) Geometry of read sensors: the AMR sensor
comprises a magnetically soft adjacent layer (sal) separated via a spacer layer (spac) from the free
ferromagnet layer (ff), which is exchange biased by antiferromagnetic material (af) at the ends of the
sensor element only. Current is passed parallel to the layers via a pair of current leads. The GMR
sensor comprises a free ferromagnet layer spin-coupled through a spacer layer (spac) to a pinned
ferromagnet layer (pf) which is exchange biased by an entire layer of antiferromagnetic material (af).
As in the AMR sensor, current is passed parallel to the layers via current leads. (b) Mode of operation
of AMR sensor. The rate of change of resistance with respect to field in an AMR structure is greatest
when the moment of the sense layer ff is at an angle of 45 to the direction of the sense current in the
absence of any applied external field. This angle is achieved by careful engineering of the combined
effects of shape anisotropy of the element, exchange bias from the partial antiferromagnet layer af,
and the demagnetization field from the moment of the soft adjacent layer sal induced by the self field
of the sense current flowing through the sensor. Introducing an external field such as the fringing
fields at a bit transition in the recording medium perturbs the angle of the moment of the sense angle
away from the balanced 45 angle; the resistance of the sense element changes due to AMR and,
hence, the voltage drop across the sensor changes in response to the field change. Resistance of the
. Deconvoluting the voltage changes with respect to time yields the
AMR sensor varies as
desired data stream. (c) Mode of operation of a GMR sensor. The initial angle of the moment in the
GMR sense layer ff is the vector sum of the effects of shape anisotropy, magnetostatic coupling to the
pinned layer pf and the self field of the sense current. The angle of the moment of the pinned layer pf
is set by exchange bias from the antiferromagnetic layer af. As with the AMR sensor, introducing an
external field perturbs the angle of the moment and changes the resistance of the sense element,
which is detectable as a voltage change and, hence, enables one to read the recorded data. Note that
= much greater than the
the GMR effect, however, changes as
and has a maximum
AMR effect, making the GMR sensor sensitive to much smaller field changes. Hence, using GMR
enables the detection of smaller recorded bit patterns than the AMR sensor.
cos ( )
cos( )
1R R
In order to make useful magnetic field sensors the
magnetic moments of the sensor must respond in a highly
reproducible and nonhysteretic manner to magnetic field. It
is not easy to control the magnetic field response of the many
magnetic layers in a multilayer structure, and for magnetic
recording read heads, in particular, the thickness of such
multilayers precludes their use for high density recording
applications because of the narrow gap between magnetic
shields. For sensing large magnetic fields over relatively
large areas, such as for position sensors or rotation sensors
for automobile antilock brakes, magnetic multilayers may
666
be useful [30]. For the detection of the comparatively small
magnetic fields in magnetic recording read heads the most
useful GMR device is a spin-valve sandwich structure [31].
The simplest form of this device, which is shown schematically in Fig. 5(c), is a sandwich of two ferromagnetic layers
separated by a thin Cu layer, where the Cu layer thickness is
chosen to be sufficiently thick that the oscillatory interlayer
exchange coupling field is weak. This field, which varies
inversely with the ferromagnetic layer thickness, can be
made smaller by increasing the thicknesses of these layers,
and also by engineering the thickness of Cu to be intermePROCEEDINGS OF THE IEEE, VOL. 91, NO. 5, MAY 2003
Fig. 5. Evolution of magnetically engineered multilayers (a), (b) The easy axis of the “free”
ferromagnetic layer in a magnetoresistive device is oriented based on the purpose for which it is
engineered. Field sensor devices such as read heads rely on a free layer with an easy axis at right
angles to the moment of the “pinned” layer. Impinging magnetic fields will rotate the moment away
from this position and the sensor resistance changes. On the other hand, MR devices designed
for use in memory applications will have a free layer easy axis parallel or antiparallel to that of
the pinned layer. (c) A very basic GMR/TMR (tunneling magnetoresistance) stack consisting of
a pinned ferromagnetic layer magnetically locked by exchange-bias to the interfacial field of an
antiferromagnetic layer, and a simple ferromagnetic free layer. The “spin valve” is such a stack using
a conducting spacer layer between the ferromagnetic layers. (d) In this case the pinned layer is
an element consisting of a pair of ferromagnetic layers antiferromagnetically coupled through a
ruthenium (Ru) spacer layer; the lower layer in this artificial antiferromagnet is pinned via exchange
bias as in (c). This flux closure increases the pinned layer magnetic stability and reduces coupling to
the free layer. (e) Pinned element consists of an AF-coupled pair of ferromagnetic layers acting as a
single “hard” layer. No exchange bias layer to discourage rotation of the pinned element. (f) Both the
pinned and free elements consist of AF-coupled pairs. (g) A double tunnel junction. All ferromagnetic
elements consist of AF-coupled pairs. There are two pinned ferromagnets, both exchange biased by
antiferromagnetic layers. Spin filtering occurs both as current tunnels from the first pinned layer to
the free element and again as it tunnels from the free element to the second pinned element.
diate between those which give rise to AF and F coupling
[2]. At these Cu thicknesses, corresponding to nodes in the
oscillatory coupling, another form of interlayer exchange
coupling, biquadratic coupling, is sometimes observed in
which the magnetic layers want to align themselves perpendicularly to one another [32]. Biquadratic coupling was first
PARKIN et al.: MAGNETICALLY ENGINEERED SPINTRONIC SENSORS AND MEMORY
667
observed in Fe/Cr/Fe (100) sandwiches [33]. Several models
have been proposed to account for this effect including
models in which interface roughness leads to competing F
and AF interactions so favoring an orthogonal orientation of
adjacent magnetic moments [34], [35].
The magnetics of both AMR and GMR sensors is often
engineered by means of the phenomenon of exchange bias
[36], [37]. Exchange bias, first documented in the 1950s [5],
involves the stabilization of the moment of a ferromagnetic
layer by coupling it to an antiferromagnetic (AF) layer. The
AF layer gives rise to an internal exchange field at the interface with the adjacent F layer whose direction is fixed
by cooling the F/AF couple from above the Néel temperature of the AF layer in a magnetic field sufficiently large to
magnetize the F layer in one direction. When the bilayer is
cooled below the ordering temperature of the AF layer, it is
now believed that the spins in the AF layer are blocked in
AF domains with very small uncompensated net moments
in the direction of the F layer so giving rise to a net magnetic bias field [38]–[40]. The exchange coupling of these
moments with that of the adjacent F layer means that the F
layer exhibits a unidirectional magnetic anisotropy so that its
magnetic moment direction is fixed in one direction. Since
the exchange coupling is at the F/AF interface, the external
magnetic field required to rotate the moment of the F layer increases inversely with the magnetic moment (i.e. thickness)
of the F layer.
Exchange bias has been used in magnetic recording read
heads for many years as a means of stabilizing the magnetization within the sensing layer so as to reduce Barkhausen noise
[41]. The use of exchange bias in AMR heads is illustrated
in Fig. 4(a) and (b). Both ends of the ferromagnetic sensing
layer are covered with an AF exchange bias layer which prevents the ends of the sense layer from forming flux closure
domains. The magnetic structure of these regions may otherwise change irreversibly in the presence of field or due to
thermal fluctuations which can introduce significant noise
during the read process.
In a spin-valve device one of the magnetic layers, the reference layer, is pinned by exchange bias with an antiferromagnetic layer [42]. Provided that the external field is much
weaker than the exchange bias field, the magnetic moment
of the reference F layer is essentially fixed so that only the
“free” or sensing layer in the sandwich responds to external
fields.
Optimizing the sensitivity of GMR sensors for the detection of small magnetic fields is actually somewhat more
straightforward than that of AMR sensors. The optimal
sensitivity (maximum relative change in resistance per unit
field) is realized when the magnetic moments of the adjacent
layers in a spin-valve device are orthogonal to one another
in zero external field as illustrated in Fig. 4(c). In a magnetic
recording disk drive the reference F layer is exchange biased
in a direction perpendicular to the surface of the magnetic
medium and the sense layer moment is designed to be
parallel to the medium. In advanced recording read heads
today, which are deep submicrometer in size, there will be
significant magnetic coupling of the reference and sense
668
layers through the demagnetizing fields generated from
magnetic poles at the edges of the magnetic layers. These
fields favor antiparallel coupling of these layers. In addition,
because of roughness, especially correlated roughness, of
the layers in the spin-valve sandwich, dipolar fields derived
from microscopic troughs and valleys at the reference F/Cu
interface will couple to valleys and troughs in the opposing
Cu/sense F interface, giving rise to a Néel dipolar coupling
field which favors parallel alignment of the F layers [43],
[44]. Finally, the sense current itself will generate self-magnetic fields which couple the sense and reference magnetic
layers. In spin-valve read heads these three sources of magnetic interactions are carefully balanced so that the sense
layer moment is optimally arranged with respect to that of
the reference layer moment. For areal densities in today’s
Gb/in
magnetic recording disk drives of close to
the read sensor element has a width of only 100 nm and a
height about 1–1.5 times smaller than its width of 70 nm.
These tiny dimensions make the manufacture of spin-valve
sensors a tour de force. Moreover, these tiny dimensions
lead to significant magnetic shape anisotropies resulting
from the self-demagnetizing fields of the magnetic sense
layer (and reference layer) which must also be taken into
account in engineering the optimal operating point of the
spin-valve sensor.
As mentioned above, magnetostatic interactions within a
spin-valve device are very significant and, as these devices
are shrunk further to enable the detection of ever smaller
magnetic bits at ever higher recording densities, these interactions become ever larger (typically these interactions will
scale with the inverse of length and width of the sensor). In
the simplest spin-valve sensors [Fig. 5(c)], even for thicknesses of the sense and reference layers of as little as 2–4
nm, magnetostatic interactions will give rise to such significant magnetic coupling between these layers that the device is inoperable. The solution to this problem proposed by
Parkin [45] is to replace the reference layer with an artificial
antiferromagnetic (AAF) layer comprised of a sandwich of
two F layers antiferromagetically coupled via an ultra thin Ru
layer as shown in Fig. 5(d). The net moment of the reference
layer can be set to an arbitrarily small value by varying the
relative thicknesses of the two F layers in the AAF reference
layer. The AF coupling strength through Ru layers as thin as
is very large and is about 1% of the ferromagnetic
exchange coupling within the F layers themselves. Thus, the
magnetic field required to overcome the AF coupling through
Ru can be many tesla depending on the thicknesses of the
F layers. A second most important benefit of the AAF reference layer is that the exchange bias field of the reference
layer is significantly increased, roughly in proportion to the
reduced moment of the reference layer. This makes the reference moment considerably more magnetically stable. All
spin-valve read sensors have used this AAF structure since
their first introduction by IBM in magnetic disk drives.
As shown in Fig. 5(e) one can envisage building spinvalve devices without an AF exchange bias layer at all by
using an AAF reference layer. For ultrahigh-density magnetic recording where the gap between the magnetic shields
PROCEEDINGS OF THE IEEE, VOL. 91, NO. 5, MAY 2003
Fig. 6. Standard versus AFC media. (a) Standard disk recording
media. Single layer of ferromagnetic material (one section is
depicted as granular in this illustration). Demagnetization fields at
bit transitions can be strong enough to spontaneously remagnetize
domains if the bits are small enough for thermal effects to
destabilize them (the so-called superparamagnetic limit) (b) AFC
media. Two layers of ferromagnetic material separated by a thin
ruthenium coupling layer. Domains in the top layer are stabilized
by mirror-image domains in the bottom layer, allowing smaller bit
patterns to be thermally stable.
must be further reduced, thinning the spin-valve sensor is
extremely important. Since the exchange bias layer is usually quite thick relative to the active portion of the device
(AAF/Cu/F) eliminating the AF layer would be a significant
advantage. Moreover, the AF layer, which is typically comprised of a Mn alloy such as Pt–Mn or Ir–Mn, may limit the
thermal stability of the spin valve device and so limit the
maximum processing temperature during the manufacture of
the magnetic recording head. The AAF structure may also be
used for the sense layer itself as shown in Fig. 5(f).
Another important application of magnetic engineering
based on oscillatory interlayer exchange coupling through
Ru is for high-density magnetic recording media. As shown
schematically in Fig. 6 conventional longitudinal magnetic
media is comprised of a finely grained high coercivity
magnetic alloy typically based on CoPtCr with additional
additives such as B and Ta [46], [47]. The Pt is added
to Co to increase its magnetic anisotropy. The seed and
underlayers on which the media is deposited are very
important and control the detailed microstructure of the
magnetic storage layer. One magnetic bit is designed to
be comprised of many, typically 1000, individual grains
whose magnetocrystalline anisotropy easy axes will vary
randomly in direction within the plane of the media. If
the number of grains is too small per magnetic bit (whose
typical bit length is 100 nm) this will lead to significant
variations in signal from bit to bit. Thus, the grains must be
very small compared to the bit length. As the bit length is
shrunk the grains must correspondingly shrink. Similarly,
the thickness of the magnetic media must also be reduced
in order to properly scale the magnetic recording system
and allow for smaller magnetic bits [7]. This means that the
net volume of magnetic material per grain and, thus, the
magnetic anisotropy energy will also be reduced. Thermal
fluctuations will eventually cause destabilization of the
magnetic moment of individual grains when the energy
barrier to the rotation of these moments provided by the
magnetocrystalline anisotropy energy becomes too small
[47]. The magnetic anisotropy energy can be increased
by modifying the composition and microstructure of the
magnetic media but this will require higher writing fields.
However, the maximum possible writing fields provided
by the recording write head are limited by the magnitude
of the magnetization of suitable magnetic materials. To
overcome the super-paramagnetic limit the magnetic media
can be formed from an AAF structure [47] which allows for
a smaller net magnetic moment per grain while allowing
for a greater volume of magnetic material and consequently
magnetic anisotropy in the upper magnetic layer of the AAF
structured grain [see Fig. 6(b)].
IV. MAGNETIC TUNNEL JUNCTIONS
The areal density of magnetic recording disk drives
continues to increase at such a high rate that alternatives
to GMR spin valve sensors may be needed. The maximum
useful magnetoresistance provided by spin-valve recording
read sensors today is about 16%–18%. The very largest
GMR so far observed in a spin-valve device at room temperature is 24% [48]. Much larger MR values are found
in magnetic tunneling junctions (MTJs) [3]. A MTJ is
similar to a spin-valve device but the metallic Cu spacer
is replaced by a very thin insulating tunnel barrier and the
sense current is passed perpendicularly through the device
[49]–[51]. Tunneling magnetoresistance (TMR) values as
high as 50% or more have been found at room temperature
in devices with alumina tunnel barriers [3], [52], [53].
The perpendicular current flow through a tunnel junction
device is extremely attractive for ultra high density magnetic
recording since this means that the sensor can be directly
attached to the magnetic shields which can then be used
as the electrical contacts to the device [54]. By contrast, in
conventional GMR sensors where the current flows parallel
to the layers in the device the sensor has to be electrically
isolated from the conducting magnetic shields. The isolation
layers occupy space between the shields thereby limiting the
ultimate recording density which such sensors can support.
As the sensor size is further reduced it may eventually be
possible to use metallic GMR sensors in a CPP (current perpendicular to the sensor plane) geometry since the resistance
of such devices increases inversely with the cross-sectional
area of the device. The resistance of CPP GMR sensors with
m corresponding to
Gb/in is
areas of
perhaps 10–100 times too low to provide enough read signal
today. However, the resistance of conventional CPP GMR
devices can be increased by, for example, partially oxidizing
layers within the structure [55]. It may also be possible to
replace the metals currently used in GMR spin-valve device
PARKIN et al.: MAGNETICALLY ENGINEERED SPINTRONIC SENSORS AND MEMORY
669
with less conducting materials whilst maintaining high
GMR values.
TMR sensors have some advantages over GMR sensors,
namely much larger changes in resistance with magnetic
field and improved compatibility with ultra high density
recording read heads since they are naturally CPP devices.
However, by contrast with CPP GMR devices the resistance
of TMR devices is currently too high to allow them to be
useful for high performance, high data rate disk drives.
The lowest resistance TMR junctions have resistance-area
m which means
(RA) values of as little as
that such sensors may be suitable for disk drives with areal
Gb/in . Whereas scaling to even higher
densities of
areal densities makes CPP GMR sensors more attractive,
MTJ devices become less attractive as their resistance, for
otherwise the same RA, is further increased.
V. SPIN POLARIZATION OF TUNNELING CURRENT
Both GMR and MTJ devices may be considered as spintronic devices but whereas GMR is based on spin-dependent
elastic scattering of electrons at the Fermi energy, magnetic
tunneling truly concerns the generation, manipulation and
detection of spin polarized current. In the simplest model
of spin-dependent tunneling, the TMR of an MTJ depends
on the magnitude of the spin polarization of the current tunneling through the device [49]. The spin polarization P is reand milated to the density of states (DOS) of majority
spin-polarized electrons in the ferromagnet so that
nority
. If one assumes that the tunneling
probability is simply proportional to the corresponding DOS,
as proposed by Julière, then the TMR, defined as
, is given by
where
and
are the polarizations of the two ferromagnetic layers in the
MTJ [49], [56]. In this simple model the TMR is bigger
when the spin polarization is larger. What is not entirely clear
is the meaning of the term spin polarization. Clearly, tunneling matrix elements, which depend on the symmetry of
the electronic states in the ferromagnetic metals and the detailed band structure of the insulating tunnel barrier, should
influence the tunneling probability and this will depend on
the spin state of the tunneling electrons [57], [58]. Similarly,
the dependence of the tunneling probability on the thickness
of the tunnel barrier might be expected to depend on the symmetry of the electronic states in the ferromagnetic layers so
that not only the magnitude but even the sign of the polarization of the tunneling electrons might vary with tunnel barrier
thickness [59]. Another important question is the extent to
which the spin polarization is determined by the F/ tunnel
barrier interface electronic states. None of these questions
have been satisfactorily answered so far. One problem with
tunnel junctions is the degree to which the formation of the
tunnel barrier affects the properties of the underlying ferromagnetic electrode. For example, the most common and
useful tunnel barrier is Al O which is typically formed by
first depositing a thin Al layer and subsequently oxidizing
this layer [60]. Oxidation can be carried out using a wide variety of methods ranging from energetic plasmas to natural
670
oxidation by exposure to air or oxygen. However, all of these
methods may result in partial or even complete oxidation of
the underlying ferromagnetic electrode which will strongly
influence the spin polarization and TMR of such devices.
For example, it is predicted that an Fe–vacuum interface will
have an opposite spin polarization to an Fe–O interface [61].
The spin polarization can be probed by a variety of techniques including photoemission [62], point contact Andreev
reflection (PCAR) [63] and superconducting tunneling spectroscopy (STS) [60]. Photoemission measures the overall
spin polarization integrated over the DOS at the surface of
a ferromagnet. Andreev reflection occurs at the interface
between a metal and a superconductor (SC) and results
in an increased conductance for bias voltages within the
superconducting energy gap as compared to the conductance
for bias voltages exceeding the superconducting energy gap
[64]. When the metal is ferromagnetic and the current is
spin-polarized the increased conductance is suppressed and
the degree to which it is suppressed is a measure of the
spin polarization of the ferromagnet [63]. Somewhat related
is the technique of tunneling from a F metal into a SC,
whereby the tunneling conductance is measured in a ferromagnet–insulator–superconductor (FIS) structure where
the ferromagnet and superconductor are separated by a thin
insulating layer. This technique probes the polarization at
the Fermi Energy with a contribution from the various bands
that take part in the tunneling conduction weighted by the
corresponding tunneling matrix elements [65].
Although each of these techniques is a measure of the spin
polarization of the ferromagnet’s electronic states, each is
related to the polarization in different ways so it cannot be
expected that their results can be directly compared to each
other. From a practical standpoint, the spin polarization inferred from an STS measurement, which we will call the tunneling spin polarization (TSP), is most closely related to the
TMR of an MTJ because of the similarity of the structures
involved. Hence the TSP is most relevant to spin dependent
tunneling and can be used to calculate the TMR in related
MTJ’s using Juliére’s model mentioned above [49].
Spin dependent tunneling from a ferromagnet into a superconductor was first studied by Meservey and Tedrow in the
early 1970s using Al superconducting electrodes [60]. The
superconducting transition temperature of Al must be sufficiently high that the Al layer remains superconducting with
a well defined energy gap in the presence of large magnetic
fields of the order of 1–2 T. The SC transition temperature of
Al can be increased by depositing Al layers in a poor vacuum
[60] or, alternatively, by adding small amounts of solutes,
such as Cu or Si, to the Al [66]. The latter is preferable because a poor vacuum will likely contaminate the F Al O
interface affecting the spin polarization associated with this
interface. When the Al layer is sufficiently thin it will remain superconducting when uniform magnetic fields of the
order of 1 T are applied exactly in the plane of the film. The
field Zeeman spin-splits the quasiparticle DOS in the SC so
that the SC can be used as an analyzer for the spin polarized
tunneling current from the F layer in the FIS junction. Typical STS data are shown in Fig. 7 for which the tunneling
PROCEEDINGS OF THE IEEE, VOL. 91, NO. 5, MAY 2003
Fig. 8. TSP (lines) and saturation magnetization (symbols) for
various alloys of Fe, Co, and Ni. Magnetization data from [116].
Fig. 7. Conductance curves (open symbols) and corresponding
fits (lines) for Co Fe in normal (A) and inverted structure
(B), in an applied field of 1 T. In the normal structure the SC
forms the bottom electrode whereas in the inverted structure the
ferromagnet (FM) is on the bottom. The TSP extracted from fits
to the data are included in the figure and are the same for both
structures within the experimental uncertainty.
conductance is measured as a function of bias voltage for
mV. In zero magnetic field, the conducvoltages up to
tance curves are symmetric with respect to positive and negative bias voltage and show typical BCS like peaks in conductance on either side of the superconducting tunneling energy
gap [67]. In a magnetic field the conductance curves become
highly asymmetric (see Fig. 7) and display two sets of peaks
corresponding to largely independent tunneling of majority
and minority electrons. The larger the asymmetry, the larger
is the spin polarization of the ferromagnetic metal. The magnitude and sign of the TSP are extracted by fitting the conductance curves using the superconducting DOS derived by
Maki which takes into account depairing due to the magnetic
field and mixing of the spin channels due to spin orbit scattering in the superconductor [68]. The polarization thus can
only be measured using this technique at temperatures well
below the Al superconducting transition temperature and in
T . In this regard MTJs are useful devices to
high fields
explore spin polarization values over a much wider temperature range.
Fig. 8 shows a summary of TSP values obtained from STS
studies for several Co–Fe and Ni–Fe alloys using an Al O
tunnel barrier [68]. The TSP is large and is about 45–55%
for all alloys except for those with high Ni content. The sign
of the TSP is consistent with majority spin polarized tunneling current in all cases. Also plotted in Fig. 8 are the corresponding saturation magnetic moments per atom. Interest-
ingly, these recent results differ significantly from the original results of Meservey and Tedrow who found an approximately linear relationship between the TSP and the corresponding magnetic moment for various families of Ni-based
ferromagnetic alloys [60]. Such a relationship is difficult to
understand given that the TSP and magnetic moment have
quite different physical origins. While the TSP depends on
the spin-polarized DOS very near the Fermi energy, weighted
by appropriate tunneling matrix elements, the magnetic mo. Indeed, rement depends on the integrated DOS below
cent measurements suggest that this proposed relationship
between TSP and magnetic moment is accidental and that
the decrease of the TSP for alloys with high fractions of Ni
is most likely due to the difficulty of forming a clean interface with Al O barriers [69]. The extreme sensitivity of TSP
and TMR to the F/barrier interface means that it is very important to prepare high-quality junctions with well-defined
interfaces.
In addition to alloys of Co, Fe, and Ni, Meservey and
Tedrow also measured the spin polarization of various ferromagnetic rare-earth metals using an Al O barrier [70]. They
derived small positive spin polarizations for Gd ( 13%), Tb
( 6%), Dy ( 6%), Ho ( 7%), Er ( 5%), and Tm ( 3%).
A promising class of materials that are of great interest for
use in tunneling devices are the so-called half-metallic ferromagnets. These ferromagnets are characterized by having an
absence of either majority or minority states at the Fermi energy E so that the tunneling current should be 100% spin polarized at least at low bias voltages. Thus, MTJs formed with
two half-metallic electrodes should display an infinite resistance in the antiparallel state. Among those materials that
are predicted to be half metals are CrO [71], Fe O [72],
[73], Heusler alloys, such as NiMnSb and PtMnSb [74], and
the manganites, for example, La Sr MnO [75]. There
have been efforts to measure the spin polarization of many of
these materials using all of the various techniques mentioned
above. For STS studies the tunneling current will depend
sensitively on the ferromagnet/barrier interface [76], [77] so
that it is clearly very important that the interface reflect the
half-metallic property predicted for the bulk material. For
PARKIN et al.: MAGNETICALLY ENGINEERED SPINTRONIC SENSORS AND MEMORY
671
Fig. 9. Transistor-switched MTJ memory. In this cross section, one element of an array of
transistor-switched memory cells is depicted. The Write Word line and Read Word (Control) line
extend in and out of the plane of the figure. The MTJ element of the cell is above the word line,
connected to the bit line. The switching transistor is embedded in the substrate of the cell.
example, Tanaka et al. measured a TSP of only 28% for
NiMnSb [78], which is lower than most conventional ferromagnets. However, it is reasonable to suppose that the electronic structure of the interface may well be very different
from that of the bulk, or that there is a change in stoichiometry or crystal structure at the interface. Worledge et al. investigated the TSP of La Sr MnO using an SrTiO (STO)
barrier [79] and found a value of 72%. Indeed, several years
ago very high TMR values were reported at low temperatures
in MTJ’s with manganite electrodes and an STO or
barrier [80]–[82] with TMR values of more than 970% which
corresponds to a TSP of 91% [82]. More recently, Parker
et al. reported TSP measurements on CrO which showed
nearly complete spin polarization with no evidence for minority spin states [83] although MTJs with CrO electrodes
and Co counterelectrodes have shown only small TMR effects ( 25%) corresponding to modest TSP values. These results clearly indicate that there is no fundamental limit to the
magnitude of tunneling spin polarization and tunneling magnetoresistance. This is an important distinction from GMR
for which the maximum MR is limited by shunting and other
similarly prosaic effects. However, the practical use of many
of these complex half-metallic materials in MTJ devices is
a significant challenge, as their preparation is typically difficult and their Curie temperatures are quite low in most cases.
Moreover, typically it is found that the spin polarization and
corresponding TMR decrease with increasing temperature to
very small values well below the corresponding Curie temperature.
In addition to measuring and comparing TSP values for
different ferromagnetic materials, the STS technique can
help in other ways to improve our understanding of spin
dependent tunneling. In particular, by placing the superconducting electrode either above or below the tunnel barrier,
672
the STS technique can yield information about the nature
of each of the ferromagnet/barrier interfaces independently.
Since, for example, Al O tunnel insulating barriers are
usually formed by oxidation of a thin Al layer it is crucial
to avoid over- or under-oxidation of the Al as this will
likely otherwise lead to a reduction of TMR. Fig. 7 shows
conductance curves for FIS structures with Co Fe as
either the top or bottom electrode. These data show that
the measured TSP is the same for both interfaces within
experimental uncertainty, which means that for Co Fe
the bottom interface can be prepared with the same quality
as the top interface. By contrast for Ni based alloys the
TSP is slightly lower when it is deposited as the bottom
electrode. This supports the aforementioned view, that it
is difficult to prepare a good interface between Al O and
Ni-based alloys.
VI. MTJS CONTAINING AN Fe O ELECTRODE
Fe O (magnetite) is predicted theoretically to be a negatively spin— polarized half-metal [84] and there is evidence
from photoemission that Fe O has a high degree of spin polarization [85]. Among the half-metallic ferromagnets
is attractive for applications due to its high curie temperature
of 858 K, and reasonable switching characteristics. The material belong to the ferrite spinel class of compounds sometimes written as FeO Fe O . Within this structure, there are
two coordination sites for Fe ions, octahedral and tetrahedral.
The tetrahedral ions are Fe while the octahedral ions are
half filled with Fe and half with Fe . Only the octahedral coordinated ions are involved in the conduction process.
Magnetite is, in fact not a ferromagnet, but rather is ferrimagnetic with the magnetic moment of the octahedral and
tetrahedral sites opposite to one other.
PROCEEDINGS OF THE IEEE, VOL. 91, NO. 5, MAY 2003
As mentioned above most elemental ferromagnets show
majority spin polarization. Intrinsically negative spin polarization is rare in magnetic materials but is potentially useful
for building interesting spintronic devices which require both
positive and negative spin polarizations, thereby simplifying
the magnetic engineering required. The negative spin polarization in magnetite is understood by recognizing the fact that
due to Hund’s rule the addition of an electron to the half filled
d band of the Fe will have the opposite spin polarization
to the underlying t and e electrons. It is this additional
electron which is involved in transport in bulk magnetite, and
predominantly involved in tunneling in MTJ’s. The transport
mechanism within the magnetite is one of thermally activated
hopping from an Fe site to an Fe site. As the temperature is lowered below 120 K in bulk material, the conductivity decreases by two orders of magnitude, concomitant
with a lattice distortion. It has been proposed that this phase
change, the Verwey transition, is due to charge ordering of
the spin down t electrons.
Previous work on integrating magnetite into tunnel junctions has met with limited success. Li [86] explored MTJs
with Fe O electrodes but, however, observed tiny MR effects. Seneor et al. [87] claim to have grown MTJ’s with an
Fe O electrode with a Co counterelectrode and an amorphous aluminum oxide barrier. However, they measure positive MR which is contrary to theoretical predictions. One
major obstacle to the integration of magnetite into devices
is that other phases of iron oxide readily form. In particular, Fe O is easily formed, which in the gamma phase
(maghemite) is ferromagnetic, and in the alpha (hematite) is
antiferromagnetic. Similarly, a mixed phase iron oxide can
also readily form, for which the polarization is likely to be
positive and/or very small.
MTJs containing magnetite can be formed by first depositing a series of metals such that the exposed surface
is bcc Fe (100). Then, by reactively depositing Fe in an
by
oxygen atmosphere, or by reactively depositing Al
depositing Al in an oxygen-containing atmosphere directly
onto this exposed surface, (100) Fe O is formed. An
MTJ is then completed when a barrier and ferromagnetic
counter-electrode are deposited on top of the magnetite
layer: negative TMR values of more than 15% have been
seen at room temperature. This is the first observation of a
significant inverse tunneling magnetoresistance effect in a
magnetoelectronic device at room temperature [88].
MTJs with magnetite electrodes exhibit very large increases in resistance upon cooling of a factor of twenty
or more, which is consistent with the loss of thermally
activated conduction in the Fe O electrode at the Verwey
transition. This is in stark contrast with typical MTJ’s where
the resistance only increases by about 20%–50% between
room temperature and 4.2 K.
VII. MTJ AND GMR MAGNETIC RANDOM ACCESS
MEMORY (MRAM)
Another potential application of magnetically engineered
structures is for nonvolatile MRAMs based on magnetic
storage elements. In recent years MRAM technologies have
favored arrays of individually patterned magnetic storage
cells or bits where each bit comprises a magnetic thin-film
multilayered structure. The magnetic bit is designed to
have two stable magnetic states in zero and small magnetic
fields which, usually, exhibit two different resistance values
representing “0” and “1”. Until recently such bits involved
the use of the comparatively small AMR effect. While some
of these structures are very ingenious [89], these memories
have been, not only of comparatively poor performance, but
very expensive, and thus limited in their application.
Replacing AMR bit structures with GMR spin-valve bit
structures [90] suggests some obvious advantages. Firstly,
just as for magnetic field sensor applications, the magnetic
states of the GMR bit cells are much simpler (see Fig. 5).
Secondly, the larger GMR effect should give rise to larger
signals. Since, in a first approximation, the time required to
read the state of the bit cell depends on the magnitude of the
difference in signal between its two possible states, this is
clearly an advantage. Moreover, the signal from a GMR cell
in an appropriately designed MRAM can be sufficiently large
that the bit may be read nondestructively—that is, without
changing its magnetic state. This offers, in addition to the
speed advantage, lower power consumption because the bit
does not have to be rewritten every time it is read. This is a
definite advantage, especially for mobile applications where
battery life is a critical consideration.
However, even the larger signals available from GMR
structures do not make GMR MRAM attractive for mainstream RAM applications. The conductive nature of a GMR
structure necessitates that they be wired in series with
one another. In order to achieve reasonable memory array
densities many GMR cells (of number N) would have to
be electrically connected in that series; the actual signal
available when reading any one particular cell is MR/N.
This signal is not sufficient to make GMR MRAM competitive with conventional dynamic random access memory
(DRAM) and static random access memory (SRAM) [91].
On the other hand, the fundamentally highly resistive nature of the barrier in a magnetic tunnel junction allows one to
fully utilize the high MR signal from individual MTJ storage
cells via a novel cross-point MRAM architecture [3], [92].
By connecting each MTJ element in series to a switch, for
example, a silicon diode, current only passes through a single
MTJ cell in such an arrangement and the available signal
when reading that cell is MR/1. With reasonable MR values,
such an MRAM architecture has the potential to rival that of
DRAM in density, and SRAM in speed.
A key advantage of an MTJ over a GMR device in memory
applications is that current is passed perpendicularly through
the MTJ. The electrical contacts to the MTJ cell thus essentially occupy the same space as the MTJ device itself making
the cell very small. Just as for the spin-valve structure, the
free layer is likely to be comprised of a bilayer where the
interface layer is chosen to give maximum tunneling magnetoresistance and the remainder of the free layer chosen
for small magnetorestriction or other properties for optimal
magnetic switching characteristics of the free layer or op-
PARKIN et al.: MAGNETICALLY ENGINEERED SPINTRONIC SENSORS AND MEMORY
673
Fig. 10. Reading and writing from MTJ memory array. (a)
Reading a bit. Voltage is applied to the desired Read Word line (to
enable the transistors in that word) and voltage at the desired Bit line
is measured. The transistor only turns on to provide current (with
a resultant voltage) if the MTJ in the cell is in the low-resistance
(parallel moments) state. (b) Writing bits. Current is passed through
the desired Write Word line and appropriate Bit lines. Superposition
of the fields generated by the two currents orients the moment of
the free layer in the MTJ in the desired direction. The polarity of
the moment is determined by the direction of the current in the Bit
line; current flowing one direction will flip it one way, and current
in the opposite direction will flip it the other way.
timal characteristics for processing of the MTJ memory cell.
Again, just as for the spin-valve GMR structure discussed
above there are many advantages to using an AAF reference
layer. In addition, the switching characteristics of the cell can
be further tuned by using an AF-coupled structure for the
free layer, or even by building a double tunnel junction utilizing AF-coupled elements for both pinned ferromagnetic
elements and the free ferromagnet (see Fig. 5).
Although, in principle no switch is needed, and a MTJ
memory could be constructed with all the MTJ cells in the
array connected in parallel, this leads to “sneak” currents
passing through all the MTJ cells in the array which consumes more power and reduces the available signal [91],
[93]. In Figs. 9 and 10, a more sophisticated design is shown
where an array of MTJ memory elements are shown, connected in parallel to a set of upper “bit” lines. Directly below
but electrically isolated from the MTJ is a “write word” line
orthogonal to the bit line. The other electrode of the MTJ
is connected to the source of a transistor element in a layer
below the write lines; the gate of this transistor is connected
to a “read word” line that is parallel to the write word line
above.
The metal bit and word lines are conductors through which
electrical current can be passed. A unique combination of orthogonal bit and write word lines [94] can be selected in order
674
to be able to individually address one magnetic memory cell
to set or “write” its magnetic state, by passing currents simultaneously along the corresponding write word and bit
lines (only one of the word and bit line currents need be
bipolar). The vector combination of the orthogonal magnetic
self-fields of the currents or current pulses passed through
these lines is arranged such that the magnetic state of the selected memory element at the intersection of the chosen bit
and word lines can be appropriately set. The polarity of the
written state is selected by the direction of the current in the
bipolar line.
It is important that the self-fields of these same currents
must be such that the magnetic state of the half-selected devices along the same bit and word lines is not altered; these
latter cells will, however, be magnetically disturbed and it is
very important that even after many such disturbances the
magnetic state of these cells does not “creep” either to some
intermediate state or completely reverse. The half-selected
devices are also more sensitive to thermal fluctuations
through the superparamagnetic effect. Careful engineering
of both the exchange bias and the shape anisotropy of the
MTJ cells is necessary to eliminate such instabilities.
Reading memory cells within the array is similar to writing
in that a particular cell is addressed by selecting the appropriate bit line and read word line. In reading, however, the
read word line, which is the gate terminal of the transistor
connected to the MTJ, is simply raised to a voltage sufficient
to enable the transistor to switch. If the MTJ is in the “on”
(low resistance) state, the transistor will switch on and current will flow. Since the resistance of the MTJ element can
readily be varied by many orders of magnitude the behavior
of the cell can be optimized for reduced power and/or for
maximum reading speed.
A nonvolatile magnetic random access memory using
magnetic tunnel junction storage cells in either the simple
cross-point architecture or with a transistor or other switch
per MTJ element, as originally proposed by Gallagher,
Parkin, and Scheinfein at IBM Research [3], [92], [95], is
now being seriously considered by numerous other companies (see, for example, the contribution by Slaughter et al. in
this issue which describes the very similar MRAM to IBM’s
original proposal currently being pursued by Motorola).
VIII. MAGNETIC TUNNEL TRANSISTOR
In conventional magnetoelectronic devices such as the
spin-valve and the MTJ described above, electron transport
occurs at energies near the Fermi level. It is also possible
to utilize electrons with much higher energies, or so called
hot electrons, to make interesting spintronic devices. One
such example is the spin-valve transistor (SVT) [96], [97].
The SVT integrates a spin valve base with two semiconductor substrates, serving as the emitter and the collector
of the SVT is very
respectively. The collector current
sensitive to the alignment of the magnetic moments within
the spin-valve base, resulting in very high magnetic field
sensitivity. However, the hot electron energy in the SVT is
limited by the emitter Schottky barrier height to 0.9 eV.
PROCEEDINGS OF THE IEEE, VOL. 91, NO. 5, MAY 2003
The small difference between the emitter and the collector
Shottky barrier heights leads to a low collector current ( 20
nA). To sandwich a spin valve base between two semiconductor substrates also makes the fabrication process very
complex. In the magnetic tunnel transistor (MTT), a tunnel
barrier is used as the emitter [98]–[101]. The hot electron
energy can be easily adjusted by varying the emitter/base
across the tunnel barrier, enabling the
bias voltage
MTT to explore electron transport over a wide energy
range. Moreover, large collector currents can be obtained by
applying large bias voltages [100].
The schematic band diagram of one form of the MTT is
shown in Fig. 11(a). A ferromagnetic (FM) emitter injects
spin polarized hot electrons across a tunnel barrier into a FM
base layer. Scattering in the base layer causes the electrons
to lose energy and/or change momentum. Only those electrons that maintain energy larger than the collector Schottky
and that can be transmitted into the colbarrier height
lector conduction band states contribute to . Spin-dependent hot electron scattering in the base layer preferentially
scatters electrons whose spins are antiparallel with the majority spins in the base layer and, thus, makes them less likely
depends critically on the relto be collected. As a result,
ative orientation of the magnetic moments of the emitter and
the base. This dependence can be quantified by the magnetocurrent (MC), defined as
where
and
are collector current for parallel (P)
and antiparallel (AP) alignment of the emitter and base magnetic moments respectively. The collector current also depends on the conduction band structure of the semiconductor
collector. Here we present experimental results on MTT’s
with a GaAs collector. GaAs has a direct conduction band
minimum at the Brillouin zone center ( point) and higher
energy indirect minima at L and X points in the Brillouin
zone. The energy separation between the point and the L
and X minima is 0.29 and 0.48 eV, respectively [102]. All the
conduction valleys contribute to the collection of hot electrons.
The MTT is formed by dc magnetron sputtering at room
temperature. Three shadow masks are used to form the base
layer, the emitter isolation pads, and the emitter layer, respectively [Fig. 11(b)]. A typical MTT structure is given by
Co Fe
Al O
Co Fe
GaAs
Ir Mn
Ta, where is the thickness of the
barrier is formed by plasma oxidation
base layer. The
of a thin Al film. The antiferromagnetic IrMn layer pins the
emitter magnetic moment and therefore allows independent
switching of the base magnetic moment in external magnetic
fields. The active area of the tunnel junction and the total
mm and
mm ,
area of the base layer are
respectively. To reduce the leakage current of the Schottky
barrier due to the large base area, all transport measurements
are conducted at 77 K, but in principle the MTT can operate
at higher temperatures, even well above room temperature,
if the base area is made significantly smaller. In Fig. 11(c),
(a)
(b)
(c)
Fig. 11. Schematic energy diagram for (a) an MTT with a
single FM base layer and (b) picture of the MTT fabricated with
magnetron sputtering. (c) The collector current of a typical MTT as
a function of magnetic field at 77 K.
is plotted as a function of magnetic field for an MTT with
. The relative change in for parallel and antipar.
allel alignments gives rise to an MC of 73% at
The MTT is a powerful tool to study spin dependent hot
electron transport and to probe electronic structures in metals
and semiconductors. One such example is to measure spin
PARKIN et al.: MAGNETICALLY ENGINEERED SPINTRONIC SENSORS AND MEMORY
675
Fig. 13.
base.
Fig. 12. (a) The majority (solid circles) and minority (open
circles) electron attenuation length and (b) the ratio of the two as
a function of the emitter/base bias voltage.
dependent hot electron attenuation lengths in thin FM films
within the base layer of the MTT [101]. The collector current
of the MTT can be modeled by the following formula:
where
is the tunnel current,
is the emitter polarization,
is the base layer thickness,
is the attenuation length for
majority (minority) electrons within the FM base layer, and
is the electron collection efficiency at the base/collector interface. A series of MTTs is fabricated with base
layer thicknesses varying from 18 to 120 . The collector
current of each MTT is measured at a given bias voltage (i.e.,
hot electron energy) as a function of magnetic field. By fitting
the data to the formula given above, the hot electron attenuation lengths can be extracted at this given energy. The same
measurement is then conducted at various electron energies.
Typical results are summarized in Fig. 12(a). A large spin
asymmetry in attenuation length is observed for majority and
minority electrons. The majority electron attenuation length
decreases with the electron energy. This is mainly due to
the strong energy dependence of electron-electron scattering
rate, which is the most important scattering mechanism for
majority spin electrons [103]–[108]. On the other hand, minority electrons are subject to more efficient scattering because of the abundant available states to scatter into in the
d-band near the Fermi level and additional scattering mechanisms such as spontaneous spin wave scattering [109]–[111].
As a result, the minority electron attenuation length is much
smaller than that of the majority electrons. The ratio between
the majority and minority attenuation length is plotted in
Fig. 12(b), which decreases slowly from 6.4 to 4.8 in the experiment energy range. The large spin asymmetry in attenuation length implies that the MTT is a very effective spin
676
Schematic energy diagram of an MTT with a spin-valve
filter and has the potential to be a highly spin polarized electron source for spintronic applications.
A different form of the MTT is shown in Fig. 13, where a
spin-valve structure is used as the base. A typical spin valve is
Co Fe
Cu or Au
Ni Fe .
formed by
Cu. The CoFe and NiFe layers
The emitter layer is
have different coercivities. In large magnetic fields, the magnetic moments in the base are aligned parallel. When the
field is reversed, the magnetic moment of the NiFe layer first
switches at 20 Oe, while the magnetic moment of the CoFe
layer remains in the same orientation up to 120 Oe and 50
Oe for Cu and Au spacer layers, respectively. The switching
from parallel to antiparallel alignment gives rise to a giant
MC exceeding 1200% for both structures at
(Fig. 14) [112], which is nearly two orders of magnitude
larger than the TMR of typical MTJ devices and also much
larger than the MC of the MTT with a single FM base layer.
The MTT operates in the electron energy range from 1
eV up to a few electron volts, within the energy range in
which a maximum scattering asymmetry between majority
and minority electrons is expected [106], [108]–[110], [113].
The high field sensitivity and large output current make it an
intriguing spintronic device. Moreover, spin-dependent scattering in the base layer gives rise to a nearly 100% spinpolarized current at the base/collector interface and the use
of a tunnel barrier makes the MTT free from the conductivity mismatch problem [114], [115]. As a result, the MTT
promises to be a source of highly spin-polarized electron current that may be of importance for future spintronic applications.
IX. SUMMARY
In the past decade the field of spintronics has blossomed
with the development and application of magnetically engineered thin-film spintronic magnetic field sensors. It is clear
that this field has been energized by two main developments,
the discovery of giant magnetoresistance in epitaxial antiferromagnetically coupled Fe/Cr [15] and subsequently polycrystalline Fe/Cr [4] and Co/Cu [16] magnetic multilayers,
PROCEEDINGS OF THE IEEE, VOL. 91, NO. 5, MAY 2003
interfaces, which have been proven to be of paramount importance.
ACKNOWLEDGMENT
The authors thank J. Harris and G. Solomon for their contributions to our work on the magnetic tunneling transistor.
REFERENCES
Fig. 14. Collector current measured as a function of
the magnetic field at V
= 1:6 V for an MTT with a
50 A CoFe=40 A Cu=50 A NiFe (upper panel) and a
50 A CoFe=40 A Au=50 A NiFe (lower panel) spin-valve base.
A giant MC of more than 1200% is measured for both MTTs.
and the discovery of oscillatory interlayer coupling in many
transition metal magnetic multilayers [4], [28]. The latter has
led to the magnetic engineering of magnetic thin film structures, allowing the development of artificial antiferromagnets
[29], [45] and other related structures.
Spintronic devices in the form of spin valves have already
had an enormous impact on magnetic recording hard disk
drives, allowing increases in capacity of such drives by almost two orders of magnitude in just five years since their introduction by IBM. The subsequent development of magnetically engineered magnetic tunneling junctions, whose basic
principle had been realized more than two decades earlier,
seems destined to allow for a novel magnetic random access memory with high speed and density and nonvolatility,
thereby realizing the dream of a high performance magnetic
nonvolatile memory whose quest began more than 50 years
ago. Indeed, the field of spintronics has its roots dating back
to the 1930s when it was first recognized that electrical transport in ferromagnetic metals is comprised of largely independent currents of majority and minority spin electrons. Generating, manipulating and detecting such spin-polarized current is the substance of spintronics. To date the most important applications of spintronics have involved all-metal and
tunneling devices. These devices involve thin film materials
with layers as thin as just a few angstroms. A decade ago
many would have thought this inconceivable. Spintronics has
been driven by the development of thin-film magnetic materials and, in particular, the understanding and control of their
[1] J. D. Livingston, Driving Force: The Natural Magic of Magnets. Cambridge, MA: Harvard Univ. Press, 1996.
[2] S. S. P. Parkin, “Giant magnetoresistance in magnetic nanostructures,” in Annual Review of Materials Science, B. W. Wessels,
Ed. Palo Alto, CA: Annual Reviews Inc., 1995, vol. 25, pp.
357–388.
[3] S. S. P. Parkin, K. P. Roche, M. G. Samant, P. M. Rice, R. B. Beyers,
R. E. Scheuerlein, E. J. O’Sullivan, S. L. Brown, J. Bucchigano,
D. W. Abraham, Y. Lu, M. Rooks, P. L. Trouilloud, R. A. Wanner,
and W. J. Gallagher, “Exchange-biased magnetic tunnel junctions
and application to nonvolatile magnetic random access memory,” J.
Appl. Phys., vol. 85, pp. 5828–5833, 1999.
[4] S. S. P. Parkin, N. More, and K. P. Roche, “Oscillations in exchange
coupling and magnetoresistance in metallic superlattice structures:
Co/Ru, Co/Cr and Fe/Cr,” Phys. Rev. Lett., vol. 64, pp. 2304–2307,
1990.
[5] W. H. Meiklejohn and C. P. Bean, “New magnetic anisotropy?,”
Phys. Rev. B, vol. 102, pp. 1413–1414, 1956.
[6] F. Jorgensen, The Complete Handbook of Magnetic
Recording. New York: TAB, 1996.
[7] K. G. Ashar, Magnetic Disk Drive Technology: Heads, Media,
Channel, Interfaces, and Integration. New York: IEEE Press,
1997.
[8] C. Tsang, T. Lin, S. A. MacDonald, M. Pinarbasi, N. Robertson,
H. Santini, M. Doerner, R. T. L. Vo, T. Diola, and P. Arnett,
“5 Gb/in recording demonstration with conventional AMR dual
element heads & thin film disks,” IEEE Trans. Magn., vol. 33, pp.
2866–2871, Sept. 1997.
[9] T. R. McGuire and R. I. Potter, “Anisotropic magnetoresistance in
ferromagnetic 3d alloys,” IEEE Trans. Magn., vol. MAG-11, pp.
1018–1038, July 1975.
[10] A. F. Mayadas, J. F. Janak, and A. Gangulee, “Resistivity of
permalloy films,” J. Appl. Phys., vol. 45, pp. 2780–2781, 1974.
[11] S. S. P. Parkin, Z. G. Li, and D. J. Smith, “Giant magnetoresistance
in antiferromagnetic Co/Cu multilayers,” Appl. Phys. Lett., vol. 58,
pp. 2710–2712, 1991.
[12] S. S. P. Parkin, “Dramatic enhancement of interlayer exchange coupling and giant magnetoresistance in Ni Fe =Cu multilayers by
addition of thin Co interface layers,” Appl. Phys. Lett., vol. 61, pp.
1358–1360, 1992.
[13]
, “Origin of enhanced magnetoresistance of magnetic multilayers-spin-dependent scattering from magnetic interface states,”
Phys. Rev. Lett., vol. 71, pp. 1641–1644, 1993.
[14] G. Binasch, P. Grünberg, F. Saurenbach, and W. Zinn, “Enhanced
magnetoresistance in layered magnetic structures with antiferromagnetic interlayer exchange,” Phys. Rev. B, vol. 39, pp. 4828–4830,
1989.
[15] M. N. Baibich, J. M. Broto, A. Fert, F. N. v. Dau, F. Petroff, P.
Etienne, G. Creuzet, A. Friederich, and J. Chazelas, “Giant magnetoresistance of (001) Fe/(001)Cr magnetic superlattices,” Phys. Rev.
Lett., vol. 61, pp. 2472–2475, 1988.
[16] S. S. P. Parkin, R. Bhadra, and K. P. Roche, “Oscillatory magnetic
exchange coupling through thin copper layers,” Phys. Rev. Lett., vol.
66, pp. 2152–2155, 1991.
[17] M. A. M. Gijs and G. E. W. Bauer, “Perpendicular giant magnetoresistance of magnetic multilayers,” Adv. Phys., vol. 46, pp. 285–445,
1997.
[18] E. Y. Tsymbal and D. G. Pettifor, “Perspectives of giant magnetoresistance,” in Solid State Physics, H. Ehrenreich and F. Spaepen,
Eds. New York: Academic, 2001, vol. 56.
[19] N. Mott, “Electrons in transition metals,” Adv. Phys., vol. 13, pp.
325–422, 1964.
[20] N. F. Mott and H. Jones, Theory of the Properties of Metals and
Alloys. London, U.K.: Oxford Univ. Press, 1936.
PARKIN et al.: MAGNETICALLY ENGINEERED SPINTRONIC SENSORS AND MEMORY
677
[21] J. W. F. Dorleijn, “Electrical conduction in ferromagnetic metals,”
Philips Res. Rep., vol. 31, pp. 287–408, 1976.
[22] I. A. Campbell and A. Fert, “Transport properties of ferromagnets,”
in Ferromagnetic Materials, E. P. Wohlfarth, Ed. Amsterdam, The
Netherlands: North-Holland, 1982, vol. 3, p. 747.
[23] D. M. Edwards, J. Mathon, R. B. Muniz, and M. S. Phan, “A resistor
network theory of the giant magnetoresistance in magnetic superlattices,” IEEE. Trans. Magn., vol. 27, pp. 3548–3552, July 1991.
[24] S. S. P. Parkin, A. Modak, and D. J. Smith, “Dependence of giant
magnetoresistance on Cu layer thickness in Co/Cu multilayers:
a simple dilution effect,” Phys. Rev. B, vol. 47, pp. 9136–9139,
1993.
[25] J. Bass and J. W. P. Pratt, “Current-perpendicular (CPP) magnetoresistance in magnetic metallic multilayers,” J. Magn. Magn. Mater.,
vol. 200, pp. 274–289, 1999.
[26] P. Ciureanu, “Magnetoresistive sensors,” in Thin Film Resistive Sensors, P. Ciureanu and S. Middelhoek, Eds. Bristol, U.K.: Inst. of
Physics, 1992, pp. 253–436.
[27] S. S. P. Parkin, “Giant magnetoresistance and oscillatory interlayer coupling in polycrystalline transition metal multilayers,” in
Ultrathin Magnetic Structures, B. Heinrich and J. A. C. Bland,
Eds. Berlin, Germany: Spinger-Verlag, 1994, vol. II, p. 148.
, “Systematic variation of strength and oscillation period of in[28]
direct magnetic exchange coupling through the 3d, 4d and 5d transition metals,” Phys. Rev. Lett., vol. 67, pp. 3598–3601>, 1991.
[29] S. S. P. Parkin and D. Mauri, “Spin-engineering: direct determination
of the RKKY far field range function in Ruthenium,” Phys. Rev. B,
vol. 44, pp. 7131–7134, 1991.
[30] J. M. Daughton, “GMR and SDT sensor applications,” IEEE Trans.
Magn., vol. 36, pp. 2773–2778, Sept. 2000.
[31] B. Dieny, V. S. Speriosu, S. S. P. Parkin, B. A. Gurney, D. R. Wilhoit, and D. Mauri, “Giant magnetoresistance in soft ferromagnetic
multilayers,” Phys. Rev. B, vol. 43, pp. 1297–1300, 1991.
[32] K. Pettit, S. Gider, M. B. Salamon, and S. S. P. Parkin, “Strong biquadratic coupling and antiferromagnetic-ferromagnetic crossover
in NiFe/Cu multilayers,” Phys. Rev. B, vol. 56, pp. 7819–7822, 1997.
[33] M. Rührig, R. Schäfer, A. Hubert, R. Mosler, J. A. Wolf, S.
Demokritov, and P. Grünberg, “Domain observations in Fe-Cr-Fe
layered structures,” Phys. Stat. Solidi A. Appl. Res., vol. 125, pp.
635–656, 1991.
[34] J. C. Slonczewski, “Fluctuation mechanism for biquadratic exchange coupling in magnetic multilayers,” Phys. Rev. Lett., vol. 67,
pp. 3172–3175, 1991.
[35] S. O. Demokritov, E. Tsymbal, P. Grünberg, W. Zinn, and I. K.
Schuller, “Magnetic-dipole mechanism for biquadratic interlayer
coupling,” Phys. Rev. B, vol. 49, pp. 720–723, 1994.
[36] A. E. Berkowitz and K. Takano, “Exchange anisotropy,” J. Magn.
Magn. Mater., vol. 200, pp. 552–570, 1999.
[37] J. Nogués and I. K. Schuller, “Exchange bias,” J. Magn. Magn.
Mater., vol. 192, pp. 203–232, 1999.
[38] H. Ohldag, A. Scholl, F. Nolting, S. Anders, F. U. Hillebrecht, and J.
Stöhr, “Spin reorientation at the antiferromagnetic NiO(001) surface
in response to an adjacent ferromagnet,” Phys. Rev. Lett., vol. 86, pp.
2878–2881, 2001.
[39] H. Ohldag, T. J. Regan, J. Stöhr, A. Scholl, F. Nolting, J. Lüning, C.
Stamm, S. Anders, and R. L. White, “Spectroscopic identification
and direct imaging of interfacial magnetic spins,” Phys. Rev. Lett.,
vol. 87, pp. 247201–247204, 2001.
[40] W. Zhu, L. Seve, R. Sears, B. Sinkovic, and S. S. P. Parkin, “Field
cooling induced changes in the antiferromagnetic structure of NiO
films,” Phys. Rev. Lett., vol. 86, pp. 5389–5392, 2001.
[41] C. Tsang, “Magnetics of small magnetoresistive sensors,” J. Appl.
Phys., vol. 55, pp. 2226–2231, 1984.
[42] S. S. P. Parkin, “Applications of magnetic nanostructures,” in Spin
Dependent Transport in Magnetic Nanostructures, S. Maekawa and
T. Shinjo, Eds. London, U.K.: Taylor & Francis, 2002.
[43] L. Néel, “Étude théorique du couplage ferro-antiferromagnétique
dans les couches minces,” Ann. Phys., vol. 2, p. 61, 1967.
[44] J. C. S. Kools, “Exchange-biased spin-valves for magnetic storage,”
IEEE Trans. Magn., vol. 32, pp. 3165–3184, July 1996.
[45] S. S. P. Parkin and D. E. Heim, “Magnetoresistive spin valve sensor
with improved ferromagnetic layer and magnetic recording system
using the sensor,” U.S. Patent 5 465 185, 1995.
[46] B. Lu and D. E. Laughlin, “Microstructure of longitudinal media,”
in The Physics of Ultra-High-Density Magnetic Recording, M. L.
Plumer, J. van Ek, and D. Weller, Eds. Berlin, Germany: SpringerVerlag, 1998.
678
[47] A. Moser, K. Takano, D. T. Margulies, M. Albrecht, Y. Sonobe, Y.
Ikeda, S. Sun, and E. E. Fullerton, “Magnetic recording: advancing
into the future,” J. Phys. D, Appl. Phys., vol. 35, pp. R157–R167,
2002.
[48] W. F. Egelhoff, P. J. Chen, C. J. Powell, M. D. Stiles, R. D.
McMichael, C.-L. Lin, J. M. Sivertsen, J. H. Judy, K. Takano, A.
E. Berkowitz, T. C. Anthony, and J. A. Brug, “Optimizing the giant
magnetoresistance of symmetric and bottom spin valves,” J. Appl.
Phys., vol. 79, pp. 5277–5281, 1996.
[49] M. Julliére, “Tunneling between ferromagnetic films,” Phys. Lett.,
vol. 54A, pp. 225–226, 1975.
[50] T. Miyazaki and N. Tezuka, “Giant magnetic tunneling effect in
Fe=Al O =Fe junction,” J. Magn. Magn. Mater., vol. 139, pp.
L231–L234, 1995.
[51] J. S. Moodera, L. R. Kinder, T. M. Wong, and R. Meservey, “Large
magnetoresistance at room temperature in ferromagnetic thin film
tunnel junctions,” Phys. Rev. Lett., vol. 74, pp. 3273–3276, 1995.
[52] X.-F. Han, T. Daibou, M. Kamijo, K. Yaoita, H. Kubota, Y. Ando,
and T. Miyazaki, “High-magnetoresistance tunnel junctions using
Co75Fe25 ferromagnetic electrodes,” Jpn. J. Appl. Phys., vol. 39,
pp. L439–L441, 2000.
[53] M. Tsunoda, K. Nishikawa, S. Ogata, and M. Takahashi, “60%
magnetoresistance at room temperature in Co-Fe/Al-O/Co-Fe
tunnel junctions oxidized with Kr-O2 plasma,” Appl. Phys. Lett.,
vol. 80, pp. 3135–3137, 2002.
[54] R. Dill, R. E. Fontana, S. S. P. Parkin, and C. Tsang, “Shielded
magnetic tunnel junction magnetoresistive read head,” U.S. Patent
5 898 548, 1999.
[55] K. Nagasaka, Y. Seyama, L. Varga, Y. Shimizu, and A. Tanaka,
“Giant magnetroresistance properties of specular spin valve films
in a current perpendicular to plane structure,” J. Appl. Phys., vol.
89, pp. 6943–6945, 2001.
[56] J. C. Slonczewski, “Conductance and exchange coupling of two ferromagnets separated by a tunnel barrier,” Phys. Rev. B, vol. 39, pp.
6995–7002, 1989.
[57] P. Mavropoulos, N. Papanikolaou, and P. H. Dederichs, “Complex
band structure and tunneling through ferromagnet/insulator/ferromagnet junctions,” Phys. Rev. Lett., vol. 85, pp. 1088–1091, 2000.
[58] O. Wunnicke, N. Papanikolaou, R. Zeller, P. H. Dederichs, V.
Drchal, and J. Kudrnovský, “Effects of resonant interface states
on tunneling magnetoresistance,” Phys. Rev. B, vol. 65, pp.
064425–064430, 2002.
[59] J. Mathon, “Tight-binding theory of tunneling giant magnetoresistance,” Phys. Rev. B, vol. 56, pp. 11810–11819, 1997.
[60] R. Meservey and P. M. Tedrow, “Spin-polarized electron tunneling,”
Phys. Rep., vol. 238, pp. 173–243, 1994.
[61] E. Y. Tsymbal, I. I. Oleinik, and D. G. Pettifor, “Oxygen-induced
positive spin polarization from Fe into the vacuum barrier,” J. Appl.
Phys., vol. 87, pp. 5230–5232, 2000.
[62] R. Feder, Polarized Electrons in Surface Physics, Singapore: World
Scientific, 1985.
[63] R. J. Soulen, J. M. Byers, M. S. Osofsky, B. Nadgorny, T. Ambrose,
S. F. Cheng, P. R. Broussard, C. T. Tanaka, J. Nowak, J. S. Moodera,
A. Barry, and J. M. D. Coey, “Measuring the spin polarization of a
metal with a superconducting point contact,” Science, vol. 282, pp.
85–88, 1998.
[64] G. E. Blonder, M. Tinkham, and T. M. Klapwijk, “Transition from
metallic to tunneling regimes in superconducting microconstrictions: excess current, charge imbalance, and supercurrent,” Phys.
Rev. B, vol. 25, pp. 4515–4532, 1982.
[65] I. I. Mazin, “How to define and calculate the degree of spin polarization in ferromagnets,” Phys. Rev. Lett., vol. 83, pp. 1427–1430,
1999.
[66] D. J. Monsma and S. S. P. Parkin, “Spin polarization of tunneling
current from ferromagnet=Al O interfaces using copper-doped
aluminum superconducting films,” Appl. Phys. Lett., vol. 77, pp.
720–722, 2000.
[67] E. L. Wolf, Principles of Electron Tunneling Spectroscopy. New
York: Oxford Univ. Press, 1989.
[68] D. J. Monsma and S. S. P. Parkin, “Temporal evolution of spin polarization in ferromagnetic tunnel junctions,” Appl. Phys. Lett., vol.
77, pp. 883–885, 2000.
[69] B. Nadgorny, R. J. Soulen, M. S. Osofsky, I. I. Mazin, G. Laprade,
R. J. M. v. d. Veerdonk, A. A. Smits, S. F. C. F. Skelton, and S. B.
Qadri, “Transport spin polarization of Ni Fe
: electronic kinematics and band structure,” Phys. Rev. B, vol. 61, pp. R3788–R3791,
2000.
PROCEEDINGS OF THE IEEE, VOL. 91, NO. 5, MAY 2003
[70] R. Meservey, D. Paraskevopoulos, and P. M. Tedrow, “Tunneling
measurements of conduction-electron-spin polarization in heavy
rare-earth metals,” Phys. Rev. B, vol. 22, pp. 1331–1337, 1980.
[71] K. P. Kämper, W. Schmitt, G. Güntherodt, R. J. Gambino, and R.
Ruf, “CrO —a new half-metallic ferromagnet,” Phys. Rev. Lett.,
vol. 59, pp. 2788–2791, 1987.
[72] A. Yanase and K. Siratori, “Band structure in the high temperature
phase of Fe O ,” J. Phys. Soc. Jpn., vol. 53, pp. 312–317, 1984.
[73] M. Imada, A. Fujimori, and Y. Tokura, “Metal-insulator transitions,”
Rev. Mod. Phys., vol. 70, pp. 1039–1263, 1998.
[74] R. A. d. Groot, F. M. Mueller, P. G. v. Engen, and K. H. J. Buschow,
“New class of materials: half-metallic ferromagnets,” Phys. Rev.
Lett., vol. 50, pp. 2024–2027, 1983.
[75] M. B. Salamon and M. Jaime, “The physics of manganites: structure
and transport,” Rev. Mod. Phys., vol. 73, pp. 583–628, 2001.
[76] S. S. P. Parkin, “Magnetic tunnel junction device with nonferromagnetic interface layer for improved magnetic field response,” U.S.
Patent 5 764 567, June 9, 1998.
[77] P. LeClair, H. J. M. Swagten, J. T. Kohlhepp, R. J. M. v. d. Veerdonk, and W. J. M. d. Jonge, “Apparent spin polarization decay in
Cu-dusted Co=Al O =Co tunnel junctions,” Phys. Rev. Lett., vol.
84, pp. 2933–2936, 2000.
[78] C. T. Tanaka, J. Nowak, and J. S. Moodera, “Spin-polarized tunneling in a half-metallic ferromagnet,” J. Appl. Phys., vol. 86, pp.
6239–6242, 1999.
[79] D. C. Worledge and T. H. Geballe, “Spin-polarized tunneling in
La
Sr
MnO ,” Appl. Phys. Lett., vol. 76, pp. 900–902,
2000.
[80] J. Z. Sun, W. J. Gallagher, P. R. Duncombe, L. Krusin-Elbaum, R.
A. Altman, A. Gupta, Y. Lu, G. Q. Gong, and G. Xiao, “Observation of large low-field magnetoresistance in tri-layer perpendicular
transport devices made using doped manganate perovskites,” Appl.
Phys. Lett., vol. 69, pp. 3266–3268, 1996.
[81] J. O’Donnell, A. E. Andrus, S. Oh, E. V. Colla, and J. N. Eckstein,
“Colossal magnetoresistance magnetic tunnel junctions grown by
molecular-beam epitaxy,” Appl. Phys. Lett., vol. 76, pp. 1914–1916,
2000.
[82] J. Z. Sun, D. W. Abraham, K. Roche, and S. S. P. Parkin, “Temperature and bias dependence of magnetoresistance in doped manganite thin film trilayer junctions,” Appl. Phys. Lett., vol. 73, pp.
1008–1010, 1998.
[83] J. S. Parker, S. M. Watts, P. G. Ivanov, and P. Xiong, “Spin polarization of CrO at and across an artificial barrier,” Phys. Rev. Lett.,
vol. 88, pp. 196601–196604, 2002.
[84] Z. Zhang and S. Satpathy, “Electron states, magnetism, and
the Verwey transition in magnetite,” Phys. Rev. B, vol. 44, pp.
13319–13331, 1991.
[85] Y. S. Dedkov, U. Rüdiger, and G. Güntherodt, “Evidence for the
half-metallic ferromagnetic state of Fe O by spin-resolved photoelectron spectroscopy,” Phys. Rev. B, vol. 65, pp. 064417–064421,
2002.
[86] X. W. Li, A. Gupta, G. Xiao, W. Qian, and V. P. Dravid, “Fabrication and properties of heteroepitaxial magnetite (Fe O ) tunnel
junctions,” Appl. Phys. Lett., vol. 73, pp. 3282–3284, 1998.
[87] P. Seneor, A. Fert, J.-L. Maurice, F. Montaigne, F. Petroff, and A.
Vaurès, “Large magnetoresistance in tunnel junctions with an
iron oxide electrode,” Appl. Phys. Lett., vol. 74, pp. 4017–4019,
1999.
[88] S. S. P. Parkin, A. Panchula, P. M. Rice, R. P. Sears, and B. Sinkovic,
“Inverse tunneling magneto-resistance in magnetic tunneling junctions containing ferrimagnetic Fe O electrodes,” unpublished.
[89] J. M. Daughton, “Magnetoresistive memory technology,” Thin Solid
Films, vol. 216, pp. 162–168, 1992.
[90] D. D. Tang, P. K. Wang, V. S. Speriosu, S. Le, and K. K. Kung, “Spin
valve RAM cell,” IEEE Trans. Magn., vol. 31, pp. 3206–3209, Nov.
1995.
[91] R. E. Scheuerlein, “Magneto-resistive IC memory limitations and
architecture implications,” presented at the NCE Seventh Biennial
IEEE Int. Nonvolatile Memory Technology Conf., Albuquerque,
NM, June 22–24, 1998.
[92] W. J. Gallagher, S. S. P. Parkin, R. E. Scheuerlein, and J. Kaufman,
“Magnetic random access memory (MRAM) with diode-isolated circuit architecture,” U.S. Patent 5 640 343, June 17, 1997.
[93] J. M. Daughton, “Magnetic tunneling applied to memory,” J. Appl.
Phys., vol. 81, pp. 3758–3763, 1997.
[94] R. E. Matick, Computer Storage Systems and Technology. New
York: Wiley, 1977.
[95] W. J. Gallagher, S. S. P. Parkin, Y. Lu, X. P. Bian, A. Marley, K. P.
Roche, R. A. Altman, S. A. Rishton, C. Jahnes, T. M. Shaw, and G.
Xiao, “Microstructured magnetic tunnel junctions,” J. Appl. Phys.,
vol. 81, pp. 3741–3746, 1997.
[96] D. J. Monsma, J. C. Lodder, T. J. A. Popma, and B. Dieny, “Perpendicular hot electron spin-valve effect in a new magnetic field sensor:
The spin-valve transistor,” Phys. Rev. Lett., vol. 74, pp. 5260–5263,
1995.
[97] D. J. Monsma, R. Vlutters, and J. C. Lodder, “Room temperature-operating spin-valve transistors formed by vacuum bonding,” Science,
vol. 281, pp. 407–409, 1998.
[98] K. Mizushima, T. Kinno, T. Yamauchi, and K. Tanaka, “Energydependent hot electron transport across a spin-valve,” IEEE Trans.
Magn., vol. 33, pp. 3500–3504, Sept. 1997.
[99] R. Sato and K. Mizushima, “Spin-valve transistor with an
Fe/Au/Fe(001) base,” Appl. Phys. Lett., vol. 79, pp. 1157–1159,
2001.
[100] S. van Dijken, X. Jiang, and S. S. P. Parkin, “Room temperature operation of a high output current magnetic tunnel transistor,” Appl.
Phys. Lett., vol. 80, pp. 3364–3366, 2002.
, “Spin-dependent hot electron transport in Ni Fe and
[101]
Co Fe
films on GaAs(001),” Phys. Rev. B, vol. 66, pp.
094417–094423, 2002.
[102] J. S. Blakemore, “Semiconducting and other major properties of gallium arsenide,” J. Appl. Phys., vol. 53, pp. R123–R181, 1982.
[103] J. J. Quinn, “Range of excited electrons in metals,” Phys. Rev., vol.
126, pp. 1453–1457, 1962.
[104] D. P. Pappas, K. P. Kämper, B. P. Miller, H. Hopster, D. E. Fowler, C.
R. Brundle, A. C. Luntz, and Z. X. Shen, “Spin-dependent electron
attenuation by transmission through thin ferromagnetic films,” Phys.
Rev. Lett., vol. 66, pp. 504–507, 1991.
[105] G. Schönhense and H. C. Siegmann, “Transmission of electrons
through ferromagnetic material and applications to detection of
electron-spin polarization,” Ann. Phys.-Leipzig, vol. 2, pp. 465–474,
1993.
[106] H.-J. Drouhin, A. J. v. d. Sluijs, Y. Lassailly, and G. Lampel, “Spindependent transmission of free electrons through ultrathin cobalt
layers,” J. Appl. Phys., vol. 79, pp. 4734–4739, 1996.
[107] W. H. Rippard and R. A. Buhrman, “Spin-dependent hot electron
transport in Co/Cu thin films,” Phys. Rev. Lett., vol. 84, pp. 971–974,
2000.
[108] H.-J. Drouhin, “Spin-dependent scattering in transition metals,” J.
Appl. Phys., vol. 89, pp. 6805–6807, 2001.
[109] R. Knorren, K. H. Bennemann, R. Burgermeister, and M. Aeschlimann, “Dynamics of excited electrons in copper and ferromagnetic
transition metals: Theory and experiment,” Phys. Rev. B, vol. 61, pp.
9427–9440, 2000.
[110] E. Zarate, P. Apell, and P. M. Echenique, “Calculation of low-energyelectron lifetimes,” Phys. Rev. B, vol. 60, pp. 2326–2332, 1999.
[111] R. Vlutters, O. M. J. v. t. Erve, S. D. Kim, R. Jansen, and J. C.
Lodder, “Interface, volume, and thermal attenuation of hot-electron
spins in Ni Fe and Co,” Phys. Rev. Lett., vol. 88, pp. 0272021–027202-4, 2002.
[112] S. van Dijken, X. Jiang, and S. S. P. Parkin, “Comparison of magnetocurrent and tansfer ratio in magnetic tunnel transistors with spinvalve bases containing Cu and Au spacer layers,” Appl. Phys. Lett.,
vol. 82, pp. 775–777, 2003.
[113] M. Aeschlimann, M. Bauer, S. Pawlik, W. Weber, R. Burgermeister,
D. Oberli, and H. C. Siegmann, “Ultrafast spin-dependent electron
dynamics in fcc Co,” Phys. Rev. Lett., vol. 79, pp. 5158–5161, 1997.
[114] G. Schmidt, D. Ferrand, L. W. Molenkamp, A. T. Filip, and B. J.
v. Wees, “Fundamental obstacle for electrical spin injection from a
ferromagnetic metal into a diffusive semiconductor,” Phys. Rev. B,
vol. 62, pp. R4790–R4793, 2000.
[115] E. I. Rashba, “Theory of electrical spin injection: tunnel contacts as
a solution of the conductivity mismatch problem,” Phys. Rev. B, vol.
62, pp. R16267–R16270, 2000.
[116] B. D. Cullity, Introduction to Magnetic Materials. Reading, MA:
Addison-Wesley, 1972.
PARKIN et al.: MAGNETICALLY ENGINEERED SPINTRONIC SENSORS AND MEMORY
679
Stuart Parkin (Member, IEEE) is a native of
the U.K, He received the B.A. degree from
Trinity College, Cambridge, U.K., in 1977, and
the Ph.D degree from the Cavendish Laboratory,
Cambridge, U.K., in 1980.
He was elected a Research Fellow in 1979
at Trinity College. He joined IBM Research in
1982 as a World Trade Post-Doctoral Fellow,
becoming a permanent member of the staff the
following year. Currently, he is Manager of the
Magnetoelectronics Group at the IBM Almaden
Research Center, San Jose, CA. His current work involves the study of
magnetic tunnel junctions and the development of an advanced nonvolatile
magnetic random access memory based on magnetic tunnel junction
storage cells. His earlier research interests have included organic superconductors, ceramic high-temperature superconductors and, most recently,
the study of magnetic thin-film structures and nanostructures exhibiting
giant magnetoresistance (GMR). In 1991, he discovered oscillations in the
magnitude of the interlayer exchange coupling in transition metal magnetic
multilayered systems.
Dr. Parkin shared both the American Physical Society’s International New Materials Prize (1994) and the European Physical Society’s
Hewlett-Packard Europhysics Prize (1997). He has received other awards,
including the Materials Research Society Outstanding Young Investigator
Award (1991), the Charles Vernon Boys Prize from the Institute of Physics,
London (1991), the 1999–2000 American Institute of Physics Prize for
Industrial Applications of Physics, as well as several awards from IBM.
In 2001, he was named the first Innovator of the Year by R&D Magazine.
He was elected a Fellow of the Royal Society in 2000 and is also a Fellow
of the American Physical Society and the Institute of Physics (London). In
1997, he was elected a member of the IBM Academy of Technology and
named an IBM Research Master Inventor. In 1999, he was appointed an
IBM Fellow, IBM’s highest technical honor.
Xin Jiang received the B.S. degree in physics
from Tsinghua University, Beijing, China, in
1998. He is currently working toward the Ph.D.
degree in applied physics at Stanford University,
Stanford, CA.
He is carrying out work for his dissertation
at the IBM Almaden Research Center, San Jose,
CA.
Mr. Jiang was a Stanford Graduate Fellow from
1998 to 2001.
680
Christian Kaiser received the Masters degree in
physics from the University of Aachen, Aachen,
Germany, in 2002. His Masters thesis involved
spin-dependent tunneling spectroscopy of ferromagnetic alloys.
He is currently enrolled in the Ph.D. program
at the University of Aachen, while continuing
to work on his dissertation at the IBM Almaden
Research Center, San Jose, CA. His research
involves helping to build an advanced deposition
system which will combine a wide variety of
deposition techniques with in-situ analysis. This new tool will allow the
fabrication of advanced spintronic devices using advanced complex thin
film materials.
Alex Panchula received the B.S. degree in
physics and mathematics from Iowa State University, Ames, in 1996. He is currently working
toward the Ph.D. degree in the Department of
Applied Physics, Stanford University, Stanford,
CA.
In 2000, he joined the IBM Almaden Research
Center, San Jose, CA, to work on his dissertation
on the magnetotransport of magnetic nanostructures focusing on new materials. His undergraduate research focused on crystal growth, transport, and magnetic properties of rare-earth intermetallics.
Kevin Roche received the B.A. degree in physics
from the University of California, Berkeley, in
1983.
He began work at the IBM Research San Jose
Research Laboratory in the field of FM laser
spectroscopy. He is now part of the Magnetoelectronics Group at the IBM Almaden Research
Center, San Jose, CA.
Mahesh Samant received the B.Tech. degree from the Indian Institute of Technology,
Bombay, India, in 1981, and the Ph.D. degree
from Stanford University, Stanford, CA, in
1986, both in chemical engineering. His Ph.D.
dissertation concerned structural studies of
bimetallic catalysts.
He has subsequently worked in the areas of the
electrochemistry of under-potentially deposited
monolayer metallic films, liquid crystal materials
for displays, and characterization of magnetic
thin films using X-ray magnetic circular dichroism. Currently, he is a
Scientist at the IBM Almaden Research Center, San Jose, CA, working on
spintronic materials and devices.
PROCEEDINGS OF THE IEEE, VOL. 91, NO. 5, MAY 2003