Martienssen W., Warlimont H. (Eds.). Handbook of Condensed Matter and Materials Data

Подождите немного. Документ загружается.

1052 Part 5 Special Structures

the resistivities for spins parallel and antiparallel, respec-

tively, to the magnetization direction; α>1 means that

the resistivity is smaller for the majority spins. In the

parallel conﬁguration, the spin

(up) electrons are al-

ways the majority-spin electrons and are always weakly

scattered in all layers, resulting in a resistance r for this

channel that is smaller than the resistance R for the spin

−

(down) channel. The shorting of the current by this fast

electron channel makes the resistivity low in the paral-

lel state (R

=r). In the antiparallel conﬁguration, each

of the spin directions is alternately the majority and the

minority one. The resistance is averaged for each chan-

nel, and the overall resistance R

=(r +R)/4islarger

than in the parallel state. The GMR ratio is then

GMR =

− R

(r − R)

4rR

(3.15)

This picture holds in both the CIP and the CPP geom-

etries.

An antiparallel conﬁguration can also be obtained in

multilayers in which consecutive magnetic layers have

–2

–200 0 200

2.5

2.0

1.5

1.0

0.5

M (10

–3

emu)

H in plane || EA

∆R/R (%)

H (Oe)

H in plane || EA

FeMn 10 nm

NiFe 15 nm

Cu 2.6 nm

NiFe 15 nm

Antiferro-

magnetic

Ferromagnetic

pinned

Nonmagnetic

Ferromagnetic

free

Substrate

Fig. 5.3-22 (a) Schematic illustration of the spin valve multilayer originally proposed by Dieny et al. (15 nm NiFe/2.6nm

Cu/15 nm NiFe/10 nm FeMn).

(b) Magnetization curve, and (c) relative change in resistance. The magnetic ﬁeld is applied

parallel to the exchange anisotropy (EA) ﬁeld created by the FeMn layer. The current ﬂows perpendicular to this direction.

(After [3.100])

different coercivities [3.101], or by combining hard and

soft magnetic layers.

Spin Valves

The best known structure for obtaining an antiparal-

lel arrangement is the spin valve structure, originally

proposed by Dieny et al. [3.100] (Fig. 5.3-22). A spin

valve has two ferromagnetic layers (alloys of Ni, Fe,

and Co) sandwiching a thin nonmagnetic metal (usually

Cu), with one of the two magnetic layers being “pinned”,

i. e. the magnetization in this layer is relatively insensi-

tive to moderate magnetic ﬁelds. The other magnetic

layer is called the “free layer” and its magnetization can

be changed by application of a relatively small magnetic

ﬁeld.

As the alignment of the magnetizations in the two

layers changes from parallel to antiparallel, the resis-

tance of the spin valve typically rises by 5–10%. The

goal is to obtain the result that the magnetization of the

free layer reverses in a very small ﬁeld (a few oersteds),

while the magnetization of the pinned layer remains

Part 5 3.4

Mesoscopic and Nanostructured Materials 3.4 Magnetic Nanostructures 1053

ﬁxed. Pinning is usually accomplished by using an an-

tiferromagnetic layer, such as FeMn, that is in close

contact with the pinned magnetic layer. The resultant

spin valve response is given by ∆R ∝ cos(θ

−θ

where θ

and θ

are the angles of the magnetiza-

tion directions respectively of the free layer and of

the pinned layer with respect to the direction paral-

lel to the plane of the magnetization of the medium

whose magnetization is being sensed, as shown in

Fig. 5.3-23.

The discovery of the GMR effect has created

great expectations, since this effect has important

applications, particularly in magnetic information stor-

age technology. The use of spin valve multilayers in

hard-disk read heads was ﬁrst proposed by IBM in

1994 [3.102]. The principle of operation of a magnetic

read head willbe detailed in the section on‘Applications’

below. Because theGMR effect isso important for indus-

trial applications, there have been many improvements

in recent years. The simplest type of pinned layer has

been replaced by a synthetic antiferromagnet (two mag-

netic layers separated by a very thin (1 nm) nonmagnetic

conductor, usually ruthenium) [3.94]. The magnetiza-

tions in the two magnetic layers are strongly coupled

so as to make them antiparallel, and are thus effectively

immune to outside magnetic ﬁelds. The second innova-

tion is the nano-oxide layer (NOL), which is formed on

the outside surface of the soft magnetic layer. This NOL

reduces the resistance due to surface scattering, thus re-

ducing the background resistance and thereby increasing

the percentage change in the magnetoresistance of the

structure [3.103].

Exchange layer

Spacer

LeadLead

Free layer

Pinned layer

ˆy

ˆx

Fig. 5.3-23 Schematic illustration of an IBM GMR spin

valve sensor used in read heads. M

and M

are magne-

tizations,and θ

and θ

are angles from the longitudinal

direction. The read head moves perpendicular to the

xy plane. (After [3.102])

Magnetic Tunnel Junctions

Tunneling belongs to a class of electron transport

phenomena known as quantum transport, “quantum”

because they cannot be explained unless the wave na-

ture of electrons is invoked. In the quantum mechanical

picture, an electron is associated with a wave function ψ

related to the probability Ψ that the electron can be

found in a volume dv:byΨ = ψψ

∗

dv. Quantum the-

ory predicts that a particle has a nonzero probability of

penetrating through a classically forbidden region sep-

arating two classically allowed regions of conﬁguration

space, by the process of tunneling.

The concept of tunneling has found applications in

a vast number of domains in modern physics and chem-

istry; the particular case of interest here is tuneling

in heterostructures consisting of metallic and insula-

tor ﬁlms, leading to innovative electronic devices. In

a metal–insulator–metal (M–I–M) heterostructure, the

potential barrier arises from the microscopic interactions

between the metals and the insulator layer. By means of

tunneling, an electron with an energy lower than the po-

tential barrier can be transferred from one metal to the

other, under the condition that the thickness of the poten-

tial barrier is not too large compared with the electron’s

wavelength (Fig. 5.3-24).

For a long period, up to the 1970s, the spin of

the electron was neglected in tunnel transport. In

1971, Tedrow and Meservey [3.104] conducted tunneling

measurements on junctions between a very thin super-

conducting Al ﬁlm and a ferromagnetic Ni ﬁlm, through

an Al

barrier, in a high magnetic ﬁeld. They showed

Insulating layer

Fig. 5.3-24 One-electron energy diagram for a metal–

insulator–metal tunnel junction

Part 5 3.4

1054 Part 5 Special Structures

that the tunneling current was spin-dependent. The spin

splitting of the quasiparticle density of states of the

superconductor caused by the application of the mag-

netic ﬁeld was used to analyze the spin polarization of

the tunneling current for Ni. The same method was used

by the same authors to measure the spin polarization

of electrons tunneling from a variety of ferromagnetic

ﬁlms: Fe, Co, Ni, and Gd [3.105]. A positive spin polar-

ization was obtained in all cases; +44% was found for

Fe, +34% for Co, +11% for Ni, and +4.3% for Gd.

These results demonstrated that the tunneling process

preserves the character of the electronic spin and in-

augurated the exciting research ﬁeld of spin-dependent

Soft electrode

C,1

)

Tunnel barrier

Hard electrode

C,2

)

–R

)

GMR (%) = 100

–H

C,2

–H

C,1

C,2

–H

C,2

–H

C,1

C,2

Fig. 5.3-25a–c Schematic representation of a magnetic tunnel junc-

tion

(a) and its magnetization loop (b). The resistance of the MTJ

is determined by the relative orientation of the magnetizations of

the ferromagnetic electrodes of the junction, giving rise to low-

and high-resistance states for parallel and antiparallel alignments,

respectively

(c). (After [3.106])

tunnel transport. The large amount of progress in recent

years in this particular domain has led to the construction

of innovative spintronic devices, at the heart of which

we ﬁnd the magnetic tunnel junction (MTJ).

An MTJ behaves as a spin valve. However, it

is a hybrid system comprising at least two ferro-

magnetic metals of different coercivities H

C,1

and

C,2

C,1

< H

C,2

), separated by a thin insulator

(Fig. 5.3-25a) instead of a nonmagnetic metallic spacer

as in classic spin valves. In Fig. 5.3-25b, the magnetiza-

tion loop of a magnetic tunnel junction is schematically

shown. In such a device, the current ﬂowing perpendic-

ular to the ﬁlm plane (CPP geometry) depends on the

relative orientation of the magnetizations M

and M

of the two ferromagnetic electrodes of the junction,

reaching extreme values for parallel and antiparallel

alignment (we assume that the magnetization lies in the

plane of the ferromagnetic layers). For an applied mag-

netic ﬁeld parallel to the plane of the ﬁlms with a value

H ≥|H

C,2

|, the magnetizations of the two electrodes

are aligned parallel to each other, giving rise to a low-

resistance state (R

) for the junction (Fig. 5.3-25c).

In the magnetic-ﬁeld window |H

C,1

| < H < |H

C,2

the electrodes’ magnetizations are aligned antiparal-

lel to each other, corresponding to a high-resistance

state (R

) for the junction.

High tunnel magnetoresistance (TMR) at room tem-

perature was discovered in magnetic tunnel junctions

in the mid-1990s in Fe/Al

/Fe junctions [3.107],

Fe/Al

/Ni

1−x

junctions [3.108], and CoFe/

α-Al

/Co and NiFe [3.109] junctions, although the

ﬁrst measurements at low temperature were performed

in 1975 [3.110]. The tunneling resistance is modulated

by the magnetic ﬁeld in the same way as in a spin valve,

but the advantage over a “classical” spin valve is that it

has been shown to exhibit a signiﬁcantly higher GMR

ratio (50% or more), while requiring a saturation mag-

netic ﬁeld equal to or somewhat less than that required

for a spin valve. It was proposed that these MTJs would

have potential uses as low-power ﬁeld sensors and mem-

ory elements. Because the tunneling current density is

usually small, MTJ devicestend to have high resistances.

The resistance of an MTJ depends exponentially on the

thickness of the insulating layer. Uniformity over the

insulating layer is crucial to device operation. In addi-

tion, reproducibility from MTJ element to MTJ element

is important for the proper working of arrays of such

devices.

An example of such an effect is shown in Fig. 5.3-26,

foraCo/Al

/NiFe junction. When the magnetiza-

tions of the Co and NiFe layers go from an antiparallel

Part 5 3.4

Mesoscopic and Nanostructured Materials 3.4 Magnetic Nanostructures 1055

–200 0 200

NiFe

TMR (%)

H (Oe)

Fig. 5.3-26 Tunnel magnetoresistance curve at room tem-

perature of a Co/AL

/NiFe junction. (After [3.111])

to a parallel arrangement, the resistance of the junction

decreases. In this example, the effect is 16% at room

temperature [3.111]. Most of the measurements reported

in the literature have been performed using an alumina

(oxidized aluminium) barrier.

In Jullière’s model [3.110], the TMR ratio is ex-

pressed as a function of the spin polarizations SP

and

of the two magnetic electrodes by

GMR =

− R

2SP

1+SP

where SP

↑

) − N

↓

)

↑

) + N

↓

)

(3.16)

A structure analogous to the spin valve described

above may be made by making the two metallic elec-

trodes from a half-metallic ferromagnet (HMF, SP

= 100%) and separating them with a thin layer of in-

sulator. Now, if the magnetizations of the electrodes are

opposite, no current can ﬂow across the junction, since

the electrons which might tunnel have no density of ﬁnal

states on the far side to receive them. However, if the

electrode magnetizations are parallel, a tunneling cur-

rent may ﬂow as usual. This can be explained simply by

the schematic picture in Fig. 5.3-27, which represents

the case of two half-metals (with only one spin direc-

tion at the Fermi level) separated by a thin insulating

barrier.

We thus have a spin-electronic switch whose opera-

tion mirrors that of a pair of crossed optical polarizers,

and which may be switched “on” and “off” by the ap-

plication of an external magnetic ﬁeld. If the electrodes

are not ideal HMFs, then the on/off conductance ratio

Spin

Spin Spin

Spin

Parallel

configuration, R low

Antiparallel

configuration, R infinite

Fig. 5.3-27 Schematic picture of the TMR effect in the

case of two half-metallic electrodes. The transmission of

the electrons from one layer to the other is only possible in

the parallel state, resulting in a 100% TMR effect. (Courtesy

of A. Bartélémy)

is ﬁnite and reﬂects the majority and minority densities

of states of the ferromagnet concerned. In the case of

an ideal HMF, in the antiparallel conﬁguration, trans-

mission from one electrode to the other is not allowed,

and the resistance of the junction is inﬁnite. In contrast,

in the parallel state, transmission occurs, and the resis-

tance is ﬁnite. This gives, for half-metallic electrodes,

a spin polarization equal to 1 and a maximum TMR ef-

fect of 100%. In the case of a transition metal, the effect

is more complex to analyze owing to the presence of

s and d bands at the Fermi level with majority and mi-

nority spin directions. For transition metals, the TMR

ratio is limited to 40% at liquid-helium temperature and

29% at room temperature. This is a demonstration of

how the properties of metallic systems are controlled

exclusively by the “maﬁa” of electrons at or very near

the Fermi surface, whose band-structure properties the

metal reﬂects.

The spin tunnel junctions described above depend

for their operation only on the density of states, and

not on the carrier mobility. Unlike fully metallic sys-

tems, they have a lower conductance per unit area of

device, and hence larger signal voltages (of the order

of millivolts or more) are realizable for practical values

of the operating current. Moreover, the device charac-

teristics, such as the size of the “on” resistance, the

current density, the operating voltage, and the total cur-

rent, may be tuned by varying with device cross section,

the barrier height, and the barrier width. This is just one

reason why spin tunnel junctions are very promising

candidates for the spin injector stages of future spin-

Part 5 3.4

1056 Part 5 Special Structures

electronic devices. They will also be the basis of the

next generation of MRAMs, as illustrated later in this

section.

Large TMR effects have been obtained using HMFs

as electrodes, as can be seen in Fig. 5.3-28 in the case

of a junction with La

0.7

0.3

MnO

(LSMO) electrodes

and a SrTiO

(STO) barrier [3.112, 113]. The spin po-

larization deduced from this measurement is 83% if one

assumes the same polarization for the two magnetic

electrodes. With a polarization of 100%, half-metallic

compounds would allow true on/off operations, and

would be appropriate for gates of nonvolatile logic

devices.

6×10

5×10

4×10

3×10

2×10

1×10

–1000 –500 0 500 1000

R (Ω)

(G)

T = 4.2 K

TMR = 80%

SP = 83%

LSMO

SrTiO

LSMO

2.5 nm

LSMO/SrTiO

/LSMO

Fig. 5.3-28 (a) TMR curve of an LSMO/STO/LSMO

junction at 4.2 K. The spin polarization of the LSMO

layer is 83%. After [3.112, 113].

(b) High-resolution

transmission electron microscopy (HRTEM) image of an

LSMO/STO/LSMO junction (courtesy of J. L. Maurice)

Recently, Bowen et al. [3.114] reported a TMR

ratio of more than 1800% at T = 4.2K in LSMO/

SrTiO

/LSMO fully epitaxial MTJs, from which they

deduce a spin polarization of least 95% for the LSMO

layer. This large TMR value arises both from preserving

the quality of the LSMO/STO interfaces during the up-

graded patterning process and from the use of junctions

of small size (5.6×5.6 µm

) (Fig. 5.3-29). The TMR ex-

tends to temperatures of about 280 K, an improvement

compared with previous results in the literature [3.115].

The temperature dependence of the TMR is plotted

in Fig. 5.3-30a,b for two junctions using R(H ) loops

measured at V

= 10 mV. The TMR decreases rather

quickly with increasing T , but vanishes only at tempera-

tures of about 280 K for the 2 ×6 µm

junction. TMR

ratios of 30% (Fig. 5.3-30c) and 12% are obtained at 250

and 270 K, respectively. This represents a sizable im-

provement with respect to previous results [3.112, 113]

obtained from heterostructures grown under identi-

400

350

300

250

200

150

100

2000

1500

1000

500

–400 –200 0 200 400

R (kΩ)

H (Oe)

TMR (%)

1 mm

50 µm

Fig. 5.3-29 (a) Optical images of a processed sample of

small-size LSMO/STO/LSMO magnetic tunnel junctions.

(b) R(H ) loop for a 5.6×5.6 µm

junction measured at

4.2 K. (After [3.114])

Part 5 3.4

Mesoscopic and Nanostructured Materials 3.4 Magnetic Nanostructures 1057

800

600

400

200

0 50 100 150 200 250

800

600

400

200

0 50 100 150 200 250

–500 –250 0 250 500

TMR (%)

T (K)

TMR (%)

T (K)

TMR (%)

H (Oe)

T = 250 K

Fig. 5.3-30a–c Temperature dependence of the TMR measured with V

= 510 mV for two junctions: 2 × 6 µm

(a) and

1.4×4.2 µm

(b) (the dashed lines are guides to the eye). (c) R(H ) loop at T = 250 K and V

= 10 mV, showing 30%

TMR. (After [3.114])

cal conditions. Since the results shown in Fig. 5.3-30

were obtained from LSMO/STO/LSMO junctions with

a fully strained crystallographic structure, these meas-

urements prove that strain is not a limiting factor

in the potential production of manganite-based MTJs

with sizable TMR values at relatively high tempera-

tures. High-quality interfaces that limit the disruption

of the properties of the manganite are a more important

requirement.

This example illustrates fairly well that progress

in spin electronics would not be possible without the

improvement in the manufacturing of devices on the

nanometric scale.

Another promising HMF is Fe

, because

this oxide exhibits the highest Curie temperature

= 858 K). Consequently, one can expect that

its HMF character will remain signiﬁcant at room

temperature. Figure 5.3-31 showsa cross-sectional high-

Fig. 5.3-31 HRTEM image of a fully epitaxial Fe

/Fe

MTJ grown by MBE. The Fourier trans-

form of the barrier region of the micrograph shows the

high crystalline quality of this region. (Courtesy of P. Bayle

Guillemaud and P. Warin)

5 nm

–Al

Part 5 3.4

1058 Part 5 Special Structures

resolution transmission electron micrograph of a fully

epitaxial Fe

/Al

/Fe

MJT, grown by MBE

assisted by atomic oxygen [3.116]. The interest in this

result is that the growth of the interfaces was performed

without exposing the surface layers to air, limiting the

disruption of the bulk properties. The Fourier trans-

form of the barrier region of the micrograph shows the

high crystalline quality of this region. The use of crys-

talline barriers in MTJs opens the way to investigation of

tunnel transport in periodic barriers, which will permit

a better understanding of the underlying fundamental

physics.

The record formagnetoresistance valueshas recently

been obtained by Van Dijken et al. [3.117]. These authors

report a large magnetic-ﬁeld sensitivity of the collector

current in a three-terminal magnetic tunnel transis-

tor device with spin-valve metallic base layers. Giant

magnetocurrents exceeding 3400% result from strong

spin-dependent ﬁltering of electrons traversing perpen-

dicular to the spin-valve layers at energies well above

the Fermi energy. With its giant magnetocurrent and rea-

sonable output current, the magnetic tunnel transistor

is a promising candidate for future magnetoelectronic

devices.

Applications

Magnetic Read Heads and Sensors. Information is

stored on a magnetic disk in small magnetic domains ar-

ranged in concentric tracks. The conventional read head

used to be an induction coil, which sensed the rate of

change of the magnetic ﬁeld as the disk rotated. The sig-

nal and hence the density of magnetized bits were thus

limited by the speed of rotation of the disk. Magneto-

resistive sensors do not suffer from this defect, since

they sense the strength of the ﬁeld rather than its rate

of change. They are, therefore, capable of reading disks

with a much higher density of magnetic bits. Magneto-

resistive read heads are commercially available and will

be the leading technology beyond the year 2004.

When two oppositely magnetized domains meet,

a domain wall appears, with a characteristic width of

100–1000 Å, depending on the magnetic properties of

the material. Although there is no magnetic ﬁeld emanat-

ing from the uniformly magnetized domains themselves,

uncompensated magnetic poles at the domain walls

emanate stray ﬁelds that extend out of the medium.

Magnetic read heads contain a GMR or MTJ sensor

element, which is sensitive to these domain wall stray

ﬁelds. A schematic illustration of a GMR read head is

shown in Fig. 5.3-32. The element is fabricated in such

a way that the magnetization of the free layer is paral-

Support

out

Heads

MTJ sensor

Domain wall

stray fields

Recording-medium track

Fig. 5.3-32 Schematic illustration of an MR read head.

A spin valve (“all-metal” or MTJ) is used as an ultra-

sensitive sensor for the very low magnetic ﬁeld strength

generated by the small magnetic domains (bits). When

a weak magnetic ﬁeld, such as from a bit on a hard disk,

passes beneath the read head, the magnetic orientation of

the free layer changes its direction relative to the pinned

layer, generating a change in electrical resistance due to the

GMR effect. (After [3.106])

lel to the plane of the medium. Every time the magnetic

head passes over a domain wall, the magnetic ﬁeld em-

anating from the wall pushes the magnetization of the

free layer upwards or downwards, resulting in a change

in the resistance of the MTJ device (the magnetization

of the pinned layer is ﬁxed). The design goal is to ob-

tain the maximum rate of change in the resistance for

a change in the magnetic ﬁeld.

The concept used in read heads is also applied in an-

other type of magnetic sensors, with which we can sense

the operation of a piece of machinery when a magnetic

ﬁeld is applied, as illustrated in Fig. 5.3-33. A gear made

of ferrous metal can be arranged so that it perturbs the

magnetic fringing ﬁeld originating from a permanent

magnet close to the rotating gear and the sensor MTJ

element. The MTJ element is sensitive to the changing

orientation of the magnetic ﬂux as gear teeth pass by the

magnet. Thus, the readout resistance is related to the an-

gular position of the rotating gear. Such an arrangement

is used to monitor the engine speed in an automobile.

Magnetic Random Access Memories (MRAMs). An-

isotropic-magnetoresistance elements were the ﬁrst to

be used for MRAM applications, approximately ten

years ago. However, they provided a very small mag-

Part 5 3.4

Mesoscopic and Nanostructured Materials 3.4 Magnetic Nanostructures 1059

120

180

240

300

360

Angle θ (deg)

Rotating wheel

Sensor

element

(MTJ)

Permanent

magnet

Soft electrode

Insulator

Hard electrode

Fig. 5.3-33 Principle of operation of a magnetic sensor using an MTJ element

netoresistive effect of about 2%, making their use

disadvantageous in comparison with semiconductor

memories in terms of read access time (> 250 ns). The

density and cost of this kind of MRAMs were also

not competitive with standard semiconductor memo-

ries. A major breakthrough in MRAM technology came

with the use of GMR multilayers. The magnetoresis-

tive signal is much higher – rising up to 18% – but

the resistance of the all-metal GMR element is too

small compared with the complementary metal–oxide–

semiconductor (CMOS) transistor serving as the diode

in an integrated structure. For this reason, a number

of GMR cells have to be connected in series with the

CMOS transistor, thereby decreasing the effective mag-

netoresistive signal. This is a serious drawback in the

design of a fast-access memory. MTJs are the best

candidates for MRAMs, as they provide a high mag-

netoresistive effect of 20–50% depending on the spin

polarization of the magnetic electrodes, combined with

a modiﬁable resistance, which depends on the thick-

ness of the insulator and its degree of oxidation. This

permits the utilisation of a single MTJ cell in series

with a CMOS transistor (Fig. 5.3-34). This high-density

architecture is suitable for fast-memory applications.

The idea of the MRAM is to use the magnetization

direction of the free magnetic layer for information

storage, and the resulting resistance for readout. The

resistance of the memory bit is either low or high,

depending on the relative orientation (parallel or an-

tiparallel) of the magnetization of the free layer with

respect to the pinned layer. An applied magnetic ﬁeld

can switch the free layer between the two states. In

an MRAM array, orthogonal lines pass under and over

each memory bit, carrying the current which produces

the switching magnetic ﬁeld. The idea is that the cur-

rent in one line is not sufﬁcient to switch the memory

cell; the bit will switch only if current is ﬂowing

through both of the lines that cross at the selected

a) b)

Blocking

diode

Free layer

Pinned layer

Insulator

V+

Logical “0” Logical “l”

Fig. 5.3-34 (a) Schematic representation of an MRAM cell contain-

ing a single MTJ and a blocking diode (CMOS transistor), allowing

current ﬂow in only one direction, in series.

(b) Representation of

a logical “0” by the low-resistance state R

and of a logical “1” by

the high-resistance state R

. (c) The process of writing information

on bit kn situated at the crossing of word lines k and n. The existence

of the two currents i

and i

is necessary to switch the magnetiza-

tion of the free layer. For example, in the memory cell km, which is

inﬂuenced by a single current i

, no information is stored. (d) The

process of reading information from cell kn . A bias is applied be-

tween the word lines k and n and the current through the bit cell is

measured and compared with a reference value. The determination

of the magnetic state of the junction, which can then be related to

logical “0” or “1”, is thus possible

Part 5 3.4

1060 Part 5 Special Structures

bit. The readout of the stored information in the se-

lected bit is done by applying a bias between the

lines that cross it and measuring the current that ﬂows

through it.

For the successful realization of MRAM devices us-

ing MTJs, two very critical components – the tunnel

barrier and the free layer – have to be optimized. The

tunnel barrier needs to show large-scale uniformity over

the wafer, in addition to the more or less obvious pre-

requisites of being pinhole-free and extremely smooth.

This is because the tunnel resistance depends exponen-

tially on the barrier thickness and the absolute value of

the MTJ resistance is compared with a reference cell

during the read operation. As far as the free layer is

concerned, its thickness is directly related to the value

of the magnetic ﬁeld or current required for switching

a bit. A free layer less than 5 nm thick is desirable, since

it results in a low switching ﬁeld and, as a consequence,

low power consumption.

A number of difﬁculties have to be resolved to create

successful devices. These include efﬁcient spin injection

into semiconductors and heterostructures, and a search

for new spin-polarized materials. Other effects poten-

tially important for spintronic devices include optical

and electrical manipulation of ferromagnetism, current-

induced switching and precessing of magnetization, and

the possibility of a long coherence time for optically ex-

cited spins in semiconductors. For a good overview of

the issues in spin electronics, including the prospects for

spintronic quantum devices, see [3.118].

5.3.4.2 Ultrahigh-Density Storage Media

in Hard Disk Drives

The Limits of the Present Technology

In the current storage technology, which uses continu-

ous polycrystalline thin-ﬁlm media, the signal-to-noise

ratio (SNR) of the medium is directly related to the

number of grains per bit. At storage densities on the

order of 80 Gb/in

, the grain size of the CoCrPtX al-

loys (X = Ta, P, etc.) used in the current media is on the

order of 8 nm, the number of grains per bit is about 100,

and the bit size is on the order of 250 nm×30 nm. A fur-

ther reduction in the number of grains per bit below 100

would lead to a strong decrease in the SNR. To continue

decreasing the bit cell size in continuous media (while

keeping the SNR constant), it is necessary to reduce the

grain size further. However, there is a physical limit to

how small the grain size can be before thermal varia-

tions can affect the stability of the magnetization. This

is known as the superparamagnetic limit. In CoCrPt al-

loys, the minimum grain size for which the criterion for

magnetic stability can be satisﬁed is about 8 nm, which

is the grain size used in today’s media. Further reduc-

ing the grain size requires the use of other materials

as FePt, SmCo, and FeNdB alloys. However, this im-

plies that larger write ﬁelds will be needed to switch

the magnetization of the grains. This would necessi-

tate the development of new materials which are not yet

known. In the short term, the use of antiferromagnetic-

ally coupled media has allowed the superparamagnetic

limit to be extended in combination with GMR read

heads [3.119].

The Promise of Perpendicular Media

It is foreseen that within two or three years, magnetic

recording will switch to perpendicular discontinuous

media for several reasons. In perpendicular media, as

opposed to longitudinal granular media, the smaller the

bit size, the smaller the effect of the demagnetizing ﬁeld

on each bit is. In addition, the efﬁciency of the writer can

be greatly improved, and higher-anisotropy materials

can be used. Both effects contribute to a better thermal

stability at small size grain size as compared with longi-

tudinal media. This should help us to reach densities of

the order of 200–400 Gb/in

. However, to progress be-

yond these densities, a totally different approach will be

required, which may consist of using patterned media,

i. e. media made of arrays of nanometer-scale individual

dots, in which each dot carries a bit of information. The

signiﬁcant advantage of this type of medium is that the

shape and edge of each bit are not determined by the

grain size as in continuous media, but are determined

by the patterning process used to make the dots. Con-

sequently, each dot can be made of a single magnetic

domain (1 grain per bit instead of the 100 grains per

bit in a continuous medium). The volume of a grain

can therefore be much larger in a patterned medium so

that the superparamagnetic limit is no longer a prob-

lem, at least up to several terabits per square inch. Of

course, the use of patterned media raises other ques-

tions such as fabrication at low cost, tracking, and

reproducibility.

Various techniques have been developed to pre-

pare arrays of submicrometer magnetic dots: optical,

electron-beam, or X-ray lithography followed by

etching of the magnetic material; patterning using

a focused beam [3.120]; and self-organization of nano-

particles [3.121]. So far, however, none of these

techniques seems to be able to combine the criteria of

low cost, and rapid patterning of large areas (several

square inches) down to 25 nm pitch.

Part 5 3.4

Mesoscopic and Nanostructured Materials 3.4 Magnetic Nanostructures 1061

Nanoimprinting. Nanoimprinting is a promising

method of nanopatterning proposed recently for solving

the above-mentioned drawbacks. Nanoimprint lithogra-

phy is a major breakthrough in nanopatterning because

it can produce sub-10 nm feature sizes over a large area

with a high throughput and low cost, a feat impossible

with current, conventional lithography.

Basically, the process of nanoimprinting starts by

ﬁrst making a mold by electron-beam lithography. This

may take a long time, but this mold can be subsequently

used a large number of times to imprint layers of resists

on almost any kind of ﬂat substrate. The patterned layer

of resist is next used as an etching mask to transfer the

pattern to the substrate by direct reactive ion etching

(RIE) (Fig. 5.3-35a). The original approach to preparing

arrays of magnetic dots starts from prepatterned wafers,

on top of which a ferromagnetic material is deposited.

In that case the nanopatterning of the magnetic material

results directly from its deposition on a prepatterned

wafer (Fig. 5.3-35b).

Moritz and coworkers have studied the coupling

between the dots obtained by deposition of Co/Pt

multilayers on prepatterned Si substrates [3.122, 123].

Two patterning process were compared: (a) electron-

beam lithography followed by direct RIE [3.124], and

(b) nanoimprint lithography followed by lift-off and dry

RIE of silicon [3.125, 126]. Subsequent deposition of

a magnetic material (Co

−

Pt) leads to arrays of single-

domain dots with a perpendicular magnetization.

By nanoimprinting, a dot size down to 30 nm in

diameter with a pitch of 60 nm has been obtained. This

1. Imprint.

Press

mold

Remove

mold

2. Pattern

by trans-

fer RIE

Mold

Resist

Substrate

b)a)

Fig. 5.3-35a,b Schematic principle of (a) the successive technological steps of pattern transfer by imprinting, and (b) the

deposition of a magnetic material on a prepatterned substrate. The magnetic deposit covers the top of the dots, the bottom

of the trenches, and, to a lesser extent, the sidewalls of the dots

represents a density of 200 Gb/in

. After the nanoim-

printing of the Si substrate, a magnetic material was

deposited on the prepatterned wafer by sputtering or

by some other relatively directional physical depos-

ition process. The magnetic material then covered the

top of the dots, the bottom of the trenches separating

the dots, and, to a lesser extent, the sidewalls of the

dots [3.122].

Corresponding atomic and magnetic images are

shown in Fig. 5.3-36.

a) b)

Fig. 5.3-36 Structural (a) and magnetic (b) images of an

array of single-domain dots with a 60 nm pitch prepared

by nanoimprinting and coating with a (Pt/Co) multilayer.

Images obtained by atomic and magnetic force microscopy

Inﬂuence of the Angle of Deposition.

In order for

one to be able to use these arrays of magnetic dots as

recording media, the dots should be weakly coupled

so that the magnetization of one dot can be switched

without affecting the magnetization of the neighbor-

ing dots. In the present example, the sputtered species

Part 5 3.4