Buschow K.H.J. (Ed.) Concise Encyclopedia of Magnetic and Superconducting Materials

Подождите немного. Документ загружается.

resolution of 13 eV FWHM at 6 keV and a spatial

resolution of 0.5 mmina20mm length has been ob-

tained with an epitaxial tantalum strip detector with

STJs at each end. The detector response time is lim-

ited by the diffusion of quasiparticles in the epitaxial

ﬁlm, and the lifetime of the quasiparticles trapped

near the STJs. The count rate capability of these im-

aging strips is typically less than 1000 counts per sec-

ond. Similar systems have also been shown to work

with optical photons.

3. Hybrid Detectors

Microcalorimeters and STJ detectors have been com-

bined in several interesting ways. One example is the

aluminum STJ described above with a lead absorber.

In some WIMP detectors, pads of aluminum are de-

posited on semiconducting crystals. Each aluminum

pad is coupled to a TES. When a photon or particle is

absorbed in the crystal, nonequilibrium phonons are

generated as well as electron–hole pairs. The phonons

are absorbed in the aluminum pads where they break

Cooper pairs creating excess quasiparticles. These

quasiparticles then diffuse in the aluminum and col-

lect in the TES. The quasiparticles heat the TES,

creating a current pulse in the ampliﬁer.

Another interesting hybrid detector is the normal

metal-insulator-superconductor (NIS) tunnel junc-

tion microcalorimeter. In this detector the photon or

particle heats the electrons in the normal-metal ﬁlm.

This heating causes more electrons to tunnel through

the insulating barrier into the superconductor. This

current pulse is measured using a SQUID current

ampliﬁer. While the NIS tunnel junction uses tunnel-

ing, it is the temperature rise of the electrons in the

normal metal that determines its ultimate perform-

ance. The resolution is thus limited by the microcal-

orimeter Eqn. (2) with xB1.3. The best energy

resolution obtained with an NIS tunnel junction mi-

crocalorimeter is 22 eV FWHM at 6 keV.

An interesting feature of the NIS tunnel junction is

that the tunneling of hot electrons out of the normal

metal is a cooling process. If the energy of these

electrons can be removed from the system, the tem-

perature of the normal metal can be lowered below

the bath temperature. This property has been ex-

ploited to build microrefrigerators from NIS tunnel

junctions. These microrefrigerators may be useful for

cooling detectors or other small low-power devices

from 0.3 K to 0.1 K, or from 0.1 K to as low 20 mK.

4. Superconducting Granules

A superheated superconducting granule (SSG) parti-

cle detector consists of small type-one supercon-

ducting spheres suspended in a dielectric. The spheres

are cooled below their transition temperature, and

then a magnetic ﬁeld is applied. The magnetic ﬁeld is

adjusted to a value just below the ﬁeld required to

drive the spheres into their normal state. In this con-

ﬁguration, a small temperature increase within a

sphere will cause the sphere to undergo a sudden

phase transition to the normal state. When a sphere is

in the superconducting state, the magnetic ﬁeld is

screened from its interior by the Meissner effect.

During the phase transition to the normal state, the

magnetic ﬁeld penetrates the sphere. This results in a

sudden change in magnetic ﬂux, which can be meas-

ured by coils surrounding the spheres. Small (30 mm

in diameter) spheres of tin or zinc are typically used.

At 0.1 K, the heat capacity of these spheres is small

enough that only a few electron volts of energy are

needed to produce a phase transition.

Unlike microcalorimeters which measure the total

energy deposited, SSGs measure only that the energy

deposited was above a certain threshold. Some energy

information from a constant source can be obtained

by scanning the magnetic ﬁeld. The sensitivity of the

SSG particle detector is limited by variations from

sphere to sphere in the energy required to produce a

phase transition. Various fabrication techniques have

been developed which provide a threshold accuracy

of about 2%. With this accuracy, it is possible to

distinguish between ionizing particles that induce

several spheres to undergo phase transitions, and nu-

clear recoils that excite only one sphere. Detectors

with billions of spheres are used for WIMP-search

experiments.

See also: SQUIDs: The Instrument; SQUIDs:

Ampliﬁers

Bibliography

Booth N E, Cabrera B, Fiorini E 1996 Low-temperature particle

detectors. Annu. Rev. Nucl. Part. Sci. 46, 471–532

de Korte P, Peacock T (eds.) 2000 Proceedings of the 8th

International Workshop on Low-temperature Detectors

(LTD-8). Nucl. Instrum. Methods A. 444

Frank M, Hiller L J, le Grand J B, Mears C A, Labov S E,

Lindeman M A, Netel H, Chow D, Barfknecht A T 1998

Energy resolution and high count rate performance of su-

perconducting tunnel junction x-ray spectrometers. Rev. Sci.

Instrum. 69, 25–31

Kraus H 1996 Superconductive bolometers and calorimeters.

Supercond. Sci. Technol. 9, 827–42

Moseley S H, Mather J C, McCammon D 1984 Thermal de-

tectors as x-ray spectrometers. J. Appl. Phys. 56, 1257–62

Pretzl K 2000 Cryogenic calorimeters in astro and particle

physics. Nucl. Instrum. Methods A 454, 114–27

Wollman D A, Irwin K D, Hilton G C, Dulcie L L, Newbury D

E, Martinis J M 1997 High-resolution, energy-dispersive

microcalorimeter spectrometer for x-ray microanalysis. J.

Microsc. 188, 196–223

S. E. Labov and J. N. Ullom

Lawrence Livermore National Laboratory, Livermore

California, USA

1070

Radiation and Particle Detectors

Random Access Memories: Magnetic

A variety of magnetic random access memory

(MRAM) technologies have been explored over a

period of many decades. Their advantages include an

unlimited number of read-write cycles, random access

to any address, radiation hardness, and nonvolatility

(namely, the state of the memory does not require

periodic refreshing and is maintained even when

power is removed from the memory).

MRAM was originally in use for 30–40 years as

magnetic core memory, which was then the only af-

fordable RAM, although very labor-intensive and

having a high cost per bit, before the advent of static

RAM (SRAM) and dynamic RAM (DRAM) in the

form of integrated circuits. It should be noted that

while core memory has disappeared from most com-

puters, there are still a number of applications for it

in specialized systems, primarily military and space

systems. Early interest, centered on magnetic bubble

technology, was not successful, and involved an up-

per layer of permalloy patterned into a T-I bar struc-

ture and a lower layer of a garnet material with a

single anisotropy direction perpendicular to the sur-

face. Magnetic thin ﬁlms were considered as an al-

ternative to core memory. Since about the early

1990s, there has been a renewed interest in MRAMs

by replacing thin ﬁlm with anisotropic magnetoresis-

tive (MR) bit structures. However, these nonvolatile

memory technologies are of comparatively poor per-

formance and thus limited in their applications. The

small number of suppliers would be a concern for

systems requiring long-term maintenance or likely

upgrade (Jorgensen 1979, Daughton 1992).

The demand for nonvolatile radiation-hard mem-

ory has continued with additional pressures for high-

er density, lower power, higher speeds, and lower cost

per bit. The present efforts to develop mainly three

different versions of nonvolatile solid-state memories

are a response to this. First, an MRAM using the

spin-valve (SV) effect has been shown to have a fast

switching signal (see Magnetic Recording Systems:

Spin Valves). Second, an erasable programmable

read-only memory (EPROM) employing the Hall ef-

fect to detect the fringe ﬁeld of a ferromagnetic stor-

age element has also been proposed. The third

development underway is so-called spin-dependent

tunneling (SDT) MRAM. This article reviews pro-

gress in MRAMs and outlines the prospects for fu-

ture developments.

1. MRAM Using the Spin-valve Effect

MRAM using the spin-valve (SV) effect is evolved

from that using the MR effect. A spin-valve is com-

posed of two ferromagnetic (FM) layers (such as

Permalloy or cobalt) with a spacer layer of a non-

magnetic conductor (typically copper). In principle,

in the sandwich structures exhibiting the SV effect the

resistance is lower when alternate magnetizations are

parallel than when they are antiparallel (Daughton

and Chen 1993, Matsuyama et al. 1997). The access

time of the MRAM depends on sense signal, and

currently this is limited by the 2% MR thin-ﬁlm ma-

terials used today. The improvement in sensing time

with larger signals is roughly equal to the square of

improvement in signal. This is because, for a given

signal-to-noise ratio (SNR), the sense system band-

width may be increased by the square of the sense

signal improvement. A 10-fold improvement in signal

is possible with SV materials, and this would improve

read access times of the densest MRAMs to about the

same level as DRAMs (Daughton 1992).

Figure 1 is an MRAM architecture using the

FM(hard)–conductor–FM(soft) SV sandwich pro-

posed by Wang and coworkers in 1995 (Wang and

Nakamura 1995a). This memory requires SV bit lines

and word lines. In an SV sandwich there are two

kinds of ferromagnetic layers, which possess different

coercivities. As shown in Fig. 2, this memory operates

on the general principle of storing a binary datum in

the magnetically hard layer (cobalt) and sensing its

remanent state by switching the magnetically soft

layer (NiFe) in such a way that the magnetic state of

the hard layer remains unaltered. The parallel and

antiparallel states between magnetizations will give a

different readout.

Addressing any desired element in a two-dimen-

sional array is a basic requirement for MRAMs. This

is realized by means of the ‘‘astroid-shaped’’ switching

characteristics of a uniaxial anisotropic ferromagnetic

thin ﬁlm. As depicted in Fig. 3, the switching ﬁeld in

the longitudinal direction is lowered when a trans-

verse ﬁeld is applied. The word lines and bit lines,

Figure 1

SV-MRAM architecture using SV sandwich. The word

lines and SV bit lines, which are orthogonal to each

other, are used for selection purposes. The shorting bars

between elements to reduce the bit line resistance are not

shown in this ﬁgure.

1071

Random Access Memories: Magnetic

which are orthogonal to each other, are used for se-

lection purpose (see Fig. 1). A word current passes

through one of the word lines and a bit current

through one of the bit lines. The bit current is as-

sumed to ﬂow mainly along the intermediate layer

(e.g., copper) of the SV sandwich owing to its large

conductivity. These two currents combine at one and

only one intersection point, i.e., the addressed ele-

ment. As depicted in Fig. 3(a), neither the word ﬁeld

(current) with selected value nor the bit ﬁeld (current)

with selected value is by itself able to switch the

element. In Fig. 3(b), if both ﬁelds act together, they

are able to produce a combined ﬁeld vector greater

than the switching threshold of the addressed element.

In such a way, a writing or re-writing operation is

realized (Wang and Nakamura 1996a).

Magnetization conﬁguration of an SV-MRAM

element and its expected writing and reading timing

diagram is shown in Fig. 4. During a write operation,

the sign of the word current will deﬁne the stored bit

value, ‘‘0’’ or ‘‘1.’’ In the absence of external ﬁeld

(sleep mode), the alternate magnetizations are anti-

parallel, owing to the interlayer antiferromagnetic

coupling, giving small self-demagnetizing ﬁelds and

small external stray ﬁelds on nearby memory ele-

ments, both of which could help with element stabil-

ity. During a read operation (alive mode), data

readout is performed by monitoring the voltage

response of the SV bit line against a positive word

current, whose value is not sufﬁcient to switch the

hard layer, but switches the soft layer if it opposes

the soft layer’s magnetization. A different voltage,

depending on whether a ‘‘0’’ or ‘‘1’’ is stored, should

appear across that SV bit line. In the experimental

pulse sequence (Wang and Nakamura 1995a), the

sense current applied into the bit line is 5 mA and

a sense output voltage of 8 mV appears. The read/

write energy is quite low. Furthermore, the test in-

dicates that a stable readout state involving 3 10

repeated reversals, by ﬁeld cycling, of the soft layer’s

Figure 3

How to address a desired element in a two-dimensional

array with the aid of astroid-shaped switching

characteristics? (a) Neither the word ﬁeld (with selected

value) nor the bit ﬁeld (with selected value) by

themselves are able to switch the element; (b) both ﬁelds

act together to produce a combined ﬁeld vector greater

than the switching threshold of the addressed element.

Figure 4

Magnetization conﬁguration of an SV-MRAM element

and its expected writing and reading timing diagram. In

the absence of external ﬁeld (sleep mode), the alternate

magnetizations are antiparallel due to the inter-layer

antiferromagnetic coupling.

Figure 2

Schematic drawing of SV-MRAM. This memory

operates on the general principle of storing binary

datum in the magnetically hard layer (e.g., Co) and

sensing its remanent state by switching the magnetically

soft layer (e.g., NiFe) in such a way that the magnetic

state of the hard layer remains unaltered (after Wang

and Nakamura 1995a). The parallel and antiparallel

states between magnetizations will give a different

readout (spin-valve effect).

1072

Random Access Memories: Magnetic

magnetization is achieved, without gradual decay and

eventual erasure of the hard layer’s magnetization.

Thus this SV-MRAM is conﬁrmed to have a non-

destructive readout (NDRO) property.

2. EPROM Using Hall Effect

An erasable programmable read-only memory

(EPROM) using the Hall effect to detect the direc-

tion of magnetic moment of a ferromagnetic storage

element has been proposed by Timoshkov et al ., as

shown in Fig. 5 (Timoshkov et al. 1996). The mag-

netic ﬁlm of high coercivity and perpendicular an-

isotropy is used for data storage, and the Hall

element for data reading. The insulating layer be-

tween them produces electrical isolation of the func-

tional layers. The data recording process is realized

by magnetizing the perpendicular anisotropy layer.

The magnetic ﬂux B

of this layer penetrates through

the Hall element, and the data read process is carried

out by the Hall effect in the semiconductor material.

The read time for such a memory is limited only by

the relaxation time of the Hall effect processes and is

equal to 10

12

–10

13

second. The equation for the

Hall voltage U

¼ r

ned

f ðl=aÞð1Þ

where r

is the Hall factor depending on prevalent

scattering mechanism of the carriers, 1.18–1.93; I

the sense current; e is unit electric charge, 1.6 10

19

C; n and d are the carrier concentration in the semi-

conductor (data reading) layer and its width respec-

tively; and f(l/a) is a correction function, which is

equal to 0.90–0.95 for a reasonable ratio of the length

to the width of the data reading layer l/a ¼23.

3. MRAM Using Spin-dependent Tunneling

(SDT) Effect

The SDT effect (Julliere 1975, Miyazaki and Tezuka

1995, Moodera et al. 1995, Boeck 1998) is a special

current-perpendicular-to-plane (CPP) giant magneto-

resistance (GMR) effect, based on FM–tunnel–barri-

er–FM spin-polarized junctions. The tunneling

resistance between two ferromagnetic metal layers

that are separated by a thin insulator depends on the

relative orientation of the magnetization of each layer.

Work by Wang et al. on the MRAM using the

SDT effect began in 1994 (Wang and Nakamura

1995b, 1996b), and led initially to the successful fab-

rication of a 2 2 bit SDT-MRAM, based on Co/

/80NiFe tunnel junctions. Figure 6 is its scan-

ning electron microscopy (SEM) picture and sche-

matic. The chiplet has eight contact pads, which

provides the required ﬂexibility in connecting the test

site to the peripheral select/drive circuitry. The spin-

tunneling junction dimensions varied between 2 and

50 mm. All the spin tunneling samples Co(100 nm)/

(3–8 nm)/80NiFe(100 nm) were prepared by ra-

dio frequency (rf) sputtering with argon. The inter-

mediate Al

was formed by oxidation in the

atmosphere. Uniaxial magneto-crystalline anisotropy

in ferromagnetic ﬁlms, important both for memory

storage and for the way that a bit is selected, was

induced by a magnetic ﬁeld applied during sputtering.

Contact windows were opened in the insulator to al-

low the orthogonal top leads and bottom leads to

access the junctions directly. The top leads are visible

as brighter regions in the SEM picture, whereas the

bottom leads are darker.

As shown in Fig. 7, the structure of the SDT-

MRAM is relatively simple, compared with the semi-

conductor memories. Having an active memory-cell

transistor, ﬂash EEPROMs (electrically erasable pro-

grammable ROMs) typify semiconductor memories.

The minimum element area of a tunneling cell is 4.84

, where l is the resolution limit of the lithography

used to fabricate the memory elements. Progress in

photolithography technology should enable the re-

duction of l beyond the 200 nm limit (Daughton

1992). Compared with the minimum size of a

memory-element transistor, over 10l

(also shown

in Fig. 7), the SDT-MRAM can potentially be at

least twice as dense.

The spin-tunneling junction was designed with a

circular shape to beneﬁt the formation of single-do-

main structure, based on the micromagnetic simula-

tion result (He and Wang 1999), as shown in Fig. 8.

For each MRAM element there should exist only

two possible states of magnetization, e.g., left and

right. Clearly this is the case for an elliptic or circular

Figure 5

An EPROM using the Hall effect to detect the direction

of magnetic moment of a ferromagnetic perpendicular

storage element (after Timoshkov et al . 1996).

1073

Random Access Memories: Magnetic

island. In detail, for a single domain island one can be

sure that the island switches completely in a proper

writing process. A multi-domain state found in the

square patterned island would lead to a miscellaneous

logic in a one bit per island recording system. This is

because more than two states are possible and the

island might not have switched completely after writ-

ing. As a consequence, the magnetization of the is-

land becomes unstable and the result of a next writing

process may be unpredictable.

An isolated memory element was tested. Figure 9

illustrates the R(H) response’s minor loops operating

in the mode in which only the magnetically soft layer

is switched by applying a ﬁeld of 71.6 kA m

1

and

the sample is initially saturated by þ3.2 kA m

1

Note that the MR hysteresis at zero ﬁeld represents

different magnetization combinations, thereby yield-

ing different resistance values.

The on-chip parallel write/read operations were

successfully performed. Writing the information

involves a switching of the soft layer’s magnetiza-

tion, ideally without affecting the reference (hard)

layer’s magnetization. (In a memory device, inde-

pendent switching of the magnetically soft layer is

achieved by making the other layer either magneti-

cally hard or exchange-biased by an antiferromag-

netic layer, e.g., MnFe.) Reading the information

only requires a measurement of magnetoresistance.

An output voltage change of 96 mV, as shown in

Fig. 10, between the binary bit states against a

sine-wave sense voltage of 160 mV, 1 MHz has been

observed. This is a reproducible readout state.

Ultra-dense magnetic solid-state MRAMs will

probably be limited by the minimum signal level.

As a comparison, we ﬁrst deduce the signal level in

conventional current-in-plane (CIP)-type MR or SV

MRAM. Its signal is given by

Signal

CIP

¼ fJ

max

lDr ð2Þ

Figure 6

SEM micrographs of a 2 2 bit SDT-MRAM chiplet showing the test site containing 8 contact pads and 4 circular

tunnel junctions, each 10 mm in diameter, the schematic cross-section (note the CPP structure) and a single-tunnel

junction bit. The top leads are visible as bright regions in SEM whereas the orthogonal bottom leads are darker (after

Wang and Nakamura 1996b).

1074

Random Access Memories: Magnetic

where f is a factor smaller than one which depends

on the mode of storing and sensing information, J

max

is the maximum current per cross-sectional area

allowed by thermal dissipation, and Dr is the change

of resistivity. Thus at the maximum current density,

the signal level is proportional to the element size l.

As the storage density increases, the signal level

will decrease. In contrast, the signal of SDT-MRAM

results in

Signal

CPP

¼ fJ

max

tDr ð3Þ

where t is the total thickness of the spin-tunneling

junction. As illustrated in Fig. 11, at a maximum

sense current density, the signal level of SDT-MRAM

is obviously independent of its cross-sectional area l

This result is so attractive that the density of SDT-

MRAM should not be limited by signal degradation.

It is suggested that density-independent signal level is

Figure 7

Structure comparison of the SDT-MRAM and the

semiconductor memory. In (b) a ﬁeld effect transistor

(FET) comprises a source and a drain contact to send

current through a semiconductor channel and a gate

control to modulate the carrier density in the channel.

Here l is the resolution limit of the lithography used to

fabricate the memory elements.

Figure 8

Demagnetized states of patterned islands with different

shapes. The brightness stands for the left magnetization

direction and the darkness for the right (after He and

Wang 1999).

Figure 9

Measured resistance versus applied ﬁeld R(H) for

Co/Al

/80NiFe. The applied ﬁeld is limited at

71.6 kA m

1

and the sample is initially saturated by

þ3.2 kA m

1

. Note that the MR hysteresis at zero ﬁeld

represents different magnetization combinations thereby

yielding different magnetoresistances (after Wang and

Nakamura 1995b).

1075

Random Access Memories: Magnetic

the biggest advantage of CPP-type SDT-MRAM,

compared with CIP-type MR or SV MRAM.

It is pointed out that the SDT-MRAM’s signal re-

mains constant while the CIP MRAM’s signal drops

with increasing density. However, the Johnson noise

with SDT-MRAM may become a limiting factor in

scaling the MRAM element to submicron sizes be-

cause this component of noise becomes increasingly

important and may even dominate the total noise.

What matters for practical application is SNR. Even

the larger signals available from SV structures do not

make SV-MRAM attractive for mainstream RAM

applications. In order to achieve reasonable memory

array densities many SV elements (of number n) have

to be electrically connected in series, which means

that the actual signal available when reading one

particular element is MR/n (see Fig. 1 for reference).

Thus, a severe practical problem is encountered that

ultra-dense CIP SV-MRAM will be limited by low

SNR. By contrast, as illustrated in Fig. 6, the high

MR signal from the individual junction element can

be fully utilized in a cross-point architecture by con-

necting each element in parallel. In this case, the SNR

in an n bit n bit SDT-MRAM is independent of n,

which is normally a big number. As a result, although

SNR in both CIP and CPP MRAMs falls with in-

creasing density, the SNR for the CPP SDT-MRAM

remains much larger than for the CIP type MR or

SVMRAM.

So far, the experimental results of SDT-MRAM

are only for a low-density case in that only 2 2 bit

elements are included. For a dense or ultra-dense

MRAM, the cross-talk from an addressed element to

adjacent elements may cause mis-selection and creep

problems (after many such disturbances the magnetic

state of these bits creeps either to some intermediate

state or may completely reverse). Note also that the

driving ﬁeld required to switch the layer increases as a

result of the decreased length of the element, owing to

the demagnetization effect. To address these issues, a

ﬁnite element method (FEM) was used to study the

electromagnetic behavior of SDT-MRAM (Wang

et al. 1997). Three keepered SDT-MRAM designs

have been considered, as shown in Fig. 12. A low-cost

conventional keeperless MRAM design (Fig. 12(a)) is

taken as our reference. To localize the ﬁeld, an im-

proved partially keepered MRAM design (Fig. 12(b))

differs from the reference MRAM by the addition of

partial keepers, with high permeability, above the bit

lines and underneath the word lines. The third design,

known as the continuously keepered MRAM design,

is shown in Fig. 12(c). In this design, additional con-

tinuous keepers with 0.1 mm thickness are formed. It

was hypothesized that the drive ﬁeld would be con-

ﬁned within the nearly closed keeper, effectively en-

hancing the driving ﬁeld and reducing interference

between adjacent cells. The continuously keepered

MRAM design was found to reduce the cross-talk by

a factor of ﬁve and reduce the driving current by a

factor of four, compared with a conventional keep-

erless design, which will be the most favored ap-

proach for achieving 10

bits cm

2

areal density.

Figure 10

Reproducible read waveform. An output voltage change

of 96 mV (after an ampliﬁer) between the binary bit

states against a sine-wave voltage excitation (the above)

of 160 mV, 1 MHz has been observed. The vertical scale

is 1.2 V/div and the horizontal scale is 500 ns/div.

Figure 11

Halving the dimension of an MRAM cell will cause

different results on signal amplitude. In the CPP design

of (a), signal remains; in the CIP design of (b), signal

drops by a factor of 2. Note that both (a) and (b) are at a

maximum sense current density.

1076

Random Access Memories: Magnetic

4. Conclusions

Read-access time and storage density are the twin

keys to computer data storage devices. Although new

electronic devices, processor organizations and soft-

ware systems have contributed to enormous advances

in computer technology, they would have been

worthless without the faster and denser memories

that were developed with them. The SDT-MRAM

outlined here, the most promising candidate for mag-

netic nonvolatile memory, is likely to achieve very

high densities. From Fig. 13 it can be seen that SDT-

MRAM, which is simply structured, may in some

applications potentially exhibit excellent properties

superior to semiconductor memories and/or magnetic

disks. In particular, the density-independent signal

level is the biggest advantage of CPP-type SDT-

MRAM, compared with the conventional CIP-type

MR or SV MRAM, whose signal level is inversely

proportional to the square root of the storage den-

sity. Furthermore, it has also been demonstrated that

although the Johnson noise makes a large contribu-

tion to the total noise, it does not significantly de-

grade the performance of the device, because the

SNR for the SDT-MRAM remains much larger than

for the CIP MR or SV MRAM even at high storage

density.

Another important parameter for SDT-MRAM is

the intrinsic RC time constant of the device. Because

of the insulator between electrodes, junctions also act

as a parallel plate capacitor. The reason that the RC

time constant is important for memory application is

because of speed limitations in reading data from a

memory element. The RC time constant is independ-

ent of the area of the junction and exponentially

dependent on the thickness of the barrier and can

be reduced until the processing difﬁculties with

thin barriers prevent further reduction in thickness.

For modern MRAM applications, the equivalent

Figure 12

Three keepered SDT-MRAM designs: (a) conventional

keeper-less MRAM design; (b) improved partially

keepered MRAM design; (c) further improved

continuously keepered MRAM design (after Wang et al.

1997).

Figure 13

The position of SDT-MRAM in various data storage

devices.

1077

Random Access Memories: Magnetic

resistance for SDT devices must be at the lower end

of the resistance range. Parkin et al. found that the

resistance-area products of magnetic tunnel junctions

can be varied from 10

Omm

to as low as 60 Omm

varying the aluminum thickness and properly oxidiz-

ing it (Parkin et al. 1999). Incomplete oxidation leads

to the presence of metallic aluminum in the barrier,

which results in a rapid suppression of the tunnel

junction magnetoresistance. Based on these facts, the

RC time constant for the junctions should be of the

order of nanoseconds (Wong et al. 1998, Sousa et al.

1999). Nevertheless, such a SDT-MRAM has to be

read with a silicon sense ampliﬁer. This means a

mixed silicon-tunneling structure: the X-Y decoders/

sense ampliﬁers in an array in silicon with ‘‘holes’’ to

be ﬁlled in by tunneling array. Thus, the speed of the

device will probably be limited by silicon technology.

(Today, the speed of experimental silicon devices has

improved to 1 ns and will be expected to dip below

this in future.) There is no significant access-time ad-

vantage for the tunneling design as compared to the

bipolar devices.

As well as the continuing developments which are

always to be expected in semiconductor memory

technology, it is also interesting to note other novel

technologies, such as the spin-dependent tunneling

(SDT) effect. In the immediate future DRAM will

continue as the densest semiconductor memory, but

MRAM using the SDT effect looks set to take the

lead in the medium and longer term and is a very

important development for applications where non-

volatility, higher density, radiation hardness and low-

er power are required. Prototypes are out, technology

is maturing, and the advances mentioned in this ar-

ticle would speed up the pace of the application of

MRAM, since a microstructured junction might serve

as a high-capacity and low-power substitute for con-

ventional semi-conductor memories.

Bibliography

Boeck J D 1998 Switching with hot spins. Science 281, 357–9

Daughton J M 1992 Magnetoresistive memory technology.

Thin Solid Films 162–216

Daughton J M, Chen Y J 1993 GMR materials for low ﬁeld

applications. IEEE Trans. Magn. 29 (6), 2705–10

He L, Wang F Z 1999 Size and shape effects of patterned po-

lycrystalline islands. Intermag HC-09

Jorgensen F 1979 The Complete Handbook of Magnetic Re-

cording. Blue Ridge, Summit, PA

Julliere M 1975 Tunneling between ferromagnetic ﬁlms. Phys.

Lett. A 54, 225–6

Matsuyama K, Asada H, Ikeda K, Taniguchi K 1997 Low

current magnetic-RAM memory operation with a high sen-

sitive spin valve material. IEEE Trans. Magn. 33 (5), 3283–5

Miyazaki T, Tezuka N 1995 Giant magnetic tunneling effect in

a Fe/Al

/Fe junction. J. Magn. Magn. Mater. L231–4, 139

Moodera J S, Kinder L R, Wong T M, Meservey R 1995 Large

magnetoresistance at room temperature in ferromagnetic thin

ﬁlm tunnel junctions. Phys. Rev. Lett. 74 (16), 3273–6

Parkin S S, Roche K P, Samant M G, Rice P M, Beyers R B,

Scheueriein, O’Sullivan E J, Brown S L, Bucchigano J,

Abraham D W, Lu Y, Rooks M, Trouilloud P L, Wanner

R A Gallagher W J 1999 Exchange-biased magnetic tunnel

junctions and application to nonvolatile magnetic random

access memory. J. Appl. Phys. 85(8)

Sousa R C, Freitas P P, Chu V, Conde J 1999 Vertical inte-

gration of a spin dependent tunnel junction with an amor-

phous Si diode for MRAM applications. Intermag HA-03

Timoshkov Y, Khomenok V Danko V Kurmashev V 1996 The

memory element. Russ. Fed. Pat. N.2036517

Wang F Z, Mapps D, He L, Clegg W W, Nakamura Y 1997

Feasibility of ultra-dense spin-tunneling random access mem-

ory (STram). IEEE Trans. Magn. 33 (6), 4498–512

Wang F Z, Nakamura Y 1995a A new type of memory using

GMR effect. IEICE Gen. Conf. Proc. C-502. Fukuoka, Japan

Wang F Z, Nakamura Y 1995b Perpendicular GMR random

access memory using magnetic tunneling effect. J. Magn. Soc.

Jpn. 19 (S2), 108–11

Wang F Z, Nakamura Y 1996a Design, simulation, and real-

ization of solid state memory element using the weakly cou-

pled GMR effect. IEEE Trans. Magn. 32 (2), 520–6

Wang F Z, Nakamura Y 1996b Spin tunneling random access

memory (STram). IEEE Trans. Magn. 32 (5), 4022–4

Wong P K, Evetts J E, Blamire G 1998 High conductance small

area magnetoresistive tunnel junctions. Appl. Phys. Lett. 73

(3), 384–6

F. Z. Wang

University of North London, UK

Rare Earth Intermetallics: Thermopower of

Cerium, Samarium, and Europium

Compounds

The thermoelectric power, S, is known as a transport

property that is very sensitive to details of the elec-

tronic structure at the Fermi energy. Cerium com-

pounds exhibit widely different types of anomalous

physical properties owing to the hybridization of ce-

rium 4f electrons and conduction electrons. There is

already an accumulation of experimental data for S

for a large number of cerium compounds; however,

there has been no review article concerning these

data, as far as is known. The aim of this article is to

give an overview and discussion of these data for S

for different types of cerium compounds.

This article is based on previous review papers by

Sakurai (1993), and Sakurai et al. (1996). Anomalous

physical properties of cerium compounds are of in-

terest with respect to highly correlated electron sys-

tems (see Electron Systems: Strong Correlations).

Doniach (1977) considered a competition of two in-

teractions: the Kondo interaction between the cerium

4f electrons and the conduction electrons (character-

istic temperature T

) and the RKKY interaction be-

tween the 4f electrons on different lattice sites via the

conduction electrons (characteristic temperature T

1078

Rare Earth Intermetallics: Thermopower of Cerium, Samarium, and Europium Compounds

for an antiferromagnet or T

for a ferromagnet). The

former interaction causes the cerium 4f electrons and

the conduction electrons to form a hybridized non-

magnetic singlet state for temperatures ToT

, but

the cerium 4f electrons regain their magnetic moment

for T4T

. The latter interaction gives rise to an an-

tiferromagnetic state of the cerium 4f electrons for

ToT

(or a ferromagnetic state for ToT

) and to a

paramagnetic state with a random orientation of the

magnetic moments for T

(or T

)oT.

These characteristic temperatures T

and T

(or

) are exempliﬁed in Fig. 1 as functions of the hy-

bridization strength, J, between the cerium 4f elec-

trons and the conduction electrons. We can

distinguish in Fig. 1 four regions, I–IV, where ceri-

um compounds have distinctly different physical

properties. In region I, where T

(or T

cerium compounds are in a ﬂuctuating or inter-

mediate valence state, thus exhibiting no magnetic

moment in a wide temperature range ToT

.In

region II, with T

or T

, the compounds lose

the magnetic moment of their localized Ce

3 þ

ions

owing to the hybridization between the cerium 4f

electrons and the conduction electrons, and the

effective masses of the electrons become very large

(heavy fermion) for ToT

. In region III, where T

slightly smaller than T

, cerium compounds order

magnetically with reduced ordered moments owing

to the Kondo effect. Finally, in region IV, where

, cerium compounds are in an almost unper-

turbed antiferromagnetic state with the magnetic

moment determined from crystal ﬁeld effects experi-

enced by the local Ce

3 þ

ion.

The S(T) curves for a CeNi single crystal are

shown in Fig. 2 as a typical compound of region I

together with the nonmagnetic LaNi reference sample

(Sakurai et al. 1996). The absolute values of S for

LaNi along the three principal axes of the ortho-

rhombic crystal are small, and the S(T) curves behave

nearly linearly. This behavior can be approximated

by the linear relationship S(T) ¼aT, where a ¼const.

over a wide temperature range. This formula can be

derived from the Boltzmann equation for normal

metals (see also Boltzmann Equation and Scattering

Mechanisms, and Kondo Systems and Heavy Fermi-

ons: Transport Phenomena), with the constant a ba-

sically representing conduction electron properties at

the Fermi level, described by the density of states

(DOS) function (Ziman 1960).

On the contrary, the absolute values of S for CeNi

along the three principal axes are much larger than

those for LaNi, and the S(T) curves exhibit broad

peaks with maximums around 100 K, indicating that

a is not a constant but depends on T. An enhance-

ment of the DOS takes place at the Fermi level owing

to the formation of nonmagnetic Kondo singlets at

low temperatures. Thus, a is much larger for CeNi

than for LaNi. With the increase of T, the thermal

excitation breaks the Kondo state causing a gradual

recovery of the cerium magnetic moments, hence the

value of a decreases. A gradual recovery of the Curie–

Weiss law is known to take place when T is larger

than the temperature of the maximum in S vs. T.In

this scenario the Kondo interaction predominates by

far over the RKKY interaction, and the effect of the

crystalline electric ﬁeld (CEF) has not been taken into

account. Additional examples of these types of ma-

terials are CeSn

, CePd

, CeNiAl

, and CeNi

(Sakurai et al. 1996).

The S(T) curves of CeCu

(Onuki and Komatsu-

bara 1987) and CePtIn (Fujii et al. 1987) are shown in

Figure 1

Kondo temperature, T

, and the Neel temperatures, T

plotted as a function of the hybridization, J, between

cerium 4f electron and the conduction electron band

(after Doniach 1977).

Figure 2

S of an intermediate valence compound CeNi and its

nonmagnetic counterpart LaNi (single crystals) along

three crystal axes plotted as a function of temperature, T.

1079

Rare Earth Intermetallics: Thermopower of Cerium, Samarium, and Europium Compounds