Magneto-Electric Nanostructures ??

Overview

Recent Advances in Spintronics for

Emerging Memory Devices
Seung H. Kang

The emerging field of spintronics has (Reference 3 was a pioneering exam- In recent years, technologists started
the potential to bring game-changing ple), opening an emerging field of tech- developing more sophisticated devices
opportunities to nanoelectronic tech- nology, spintronics. In 2007, the Nobel based on spintronics. A widely publi-
nologies far beyond its traditional con- Committee awarded both scientists the cized device is magnetoresistive ran-
tribution to mass storage applications Nobel Prize for Physics. dom access memory (MRAM),4,5 which
such as hard disk drives. The value is a solid-state device utilizing a mag-
proposition is timely since the domi- netic tunnel junction (MTJ) as a binary
How would you…
nant semiconductor industry is in pur- storage element (Figure 1). Magnetore-
…describe the overall significance
suit of “More-than-Moore” to extend of this paper?
sistive random access memory would
the technology roadmap or to create retain its state even when the power
This paper provides an overview
functional diversifications through en- on recent advances in tunnel was completely turned off, like Flash
hanced system platforms. This article magnetoresistance and memory, but would offer much supe-
overviews a promising spintronic de- spin-transfer-torque (STT) rior switching speed and endurance.
magnetization switching from
vice in conjunction with recent break- the perspective of the emerging
Though not proven as a viable technol-
throughs in tunnel magnetoresistance spintronic device called STT ogy, some even promoted MRAM as a
and spin-transfer-torque magnetization magnetoresistive random access “universal memory” that would have
switching. memory (MRAM). all the desirable attributes of conven-
…describe this work to a tional memory technologies.
INTRODUCTION materials science and engineering In 2006, Freescale Semiconductor
professional with no experience
Conventional electronic devices in your technical specialty?
emerged as the first company to suc-
made of silicon rely on the transport of cessfully launch a commercial MRAM
The quantum mechanical effects
electrons as charge carriers. Electrons pertaining to the magnetic tunnel product for volume manufacturing.6
have another important property, spin, junction utilized for STT MRAM The memory capacity was only 4 Mbit
which makes each electron behave like are highly dependent on material built on a legacy 180 nm complementa-
properties and microstructures.
a tiny magnet. Magnetoelectronics is Some of the tunnel junction
ry metal-oxide-semiconductor (CMOS)
a field of science and engineering in- materials, only a few atomic logic platform. However, the enabling
volving both attributes of electrons as layers thick, require elegant technology (called toggle MRAM) was
an attempt to build a useful device. This engineering to achieve a particular a remarkable by-product of advancing
crystallographic orientation and
field is also called spin-based electron- interfacial relationship. More
spintronics that included elegant MTJ
ics or, more commonly, spintronics (an technological breakthroughs engineering. Accordingly, the company
acronym for spin transport electronics). shall come from the materials won such recognitions as one of “Five
Until now, the most significant success community. Killer Patents” by MIT Technology Re-
in spintronics has been accomplished in …describe this work to a view in 2004, “The Product of the Year
high-capacity hard disk drives (HDD), layperson? Award” by the Electronic Products
in particular, in continuing advances in The semiconductor industry is magazine in 2006, and more.
facing a fundamental challenge
the read-head sensor technology. The in scaling the silicon technology
Until now, however, the commercial-
spintronic effect that played a primary toward deep nanometer nodes. ly available MRAM has not made the
role for the success is known as giant While no alternative solution impact originally anticipated. Draw-
magnetoresistance (GMR), which was is on the horizon, a promising backs pertaining to this first-generation
spintronic memory device is under
first reported in 1988 by two Europe- development by integrating
MRAM are well understood. Hence its
an research groups independently: one1 nano-magnets into the conventional applications have been limited to niche
led by Albert Fert in France and the silicon technology platform. markets. For microelectronic industries,
other2 by Peter Grünberg in Germany. This technology can provide a spintronics is still an emerging field
combination of high performance,
The discovery of GMR promptly trig- low power, and nonvolatility along with technical challenges that have yet
gered extensive R&D efforts on mag- with superior device scalability. to be overcome before being accepted
netoelectric multilayers and devices as a mainstream technology. Thank-

28 www.tms.org/jom.html JOM • September 2008

 nel barrier in conjunction with new fer-
Figure 1. A schematic il- romagnetic electrode materials.10 As
lustration of a hysteresis the TMR ratio exceeded the GMR ra-
curve showing the MTJ re-
sistance (R) as a function tio, HDD companies started develop-
of the magnetic field (H). ing TMR-based read heads. In 2004,
A binary resistance state Seagate Technology announced the
can be obtained through
making the magnetization first HDD product equipped with TMR
of the upper ferromagnet- read heads (utilizing TiOx as a tunnel
ic layer (FM1, “free” layer) barrier). Now, many HDD products in-
with respect to that of the
lower ferromagnetic layer corporate TMR read heads with AlOx,
(FM2, “pinned” layer) ei- TiOx, or MgO as a tunnel barrier.
ther parallel or antiparallel Ironically, a potentially more signifi-
at the switching field Hc+
or Hc-, respectively. cant milestone for TMR was accom-
plished not in the HDD sector, but in
the semiconductor sector. After several
fully, in the past several years, signifi- temperature. In 1997, 9 years after the years of intensive R&D,4 in 2006 Fre-
cant technology discoveries and break- discovery of GMR by A. Fert’s and P. escale Semiconductor started shipping
throughs have been achieved in solving Grünberg’s groups,1,2 IBM introduced the first commercial MRAM.6 The stor-
or mitigating key issues. This article the first HDD with read heads based on age element was made of an AlOx-
overviews such advances in spintronics a GMR spin-valve sensor. This was an based MTJ, which was integrated into a
for memory applications, particularly, industrial breakthrough, which accord- 180 nm CMOS logic platform. This
in tunnel magnetoresistance (TMR) ingly prompted the data storage density was a legacy silicon technology since
and spin-transfer-torque (STT) switch- to increase by ~100% per year. leading semiconductor technologies
ing. Understanding these two phenom- Tunnel magnetoresistance has a lon- were moving toward the 65 nm genera-
ena is critical to perform the two essen- ger history than GMR. The TMR effect tion at that time. Further, the memory
tial memory functions, reading (sens- was experimentally observed in 1975 density was merely 4 Mbit, not a re-
ing) and writing (programming), in a by M. Jullière, who reported a TMR ra- spectful capacity as a standalone mem-
way that makes MRAM more compel- tio of 14% in an Fe/Ge/Co multilayer.7 ory. Nonetheless, this was the first-ever
ling. Also discussed is the emergence But the experiment was performed at successful demonstration of TMR-de-
of the second-generation MRAM (STT 4.2 K, hence, not attracting much at- vice volume manufacturing on a stan-
MRAM) based on the STT switching tention from a practical perspective. dard silicon-technology platform. In
mechanism (Figure 2). Two decades later, two groups of sci- striking contrast to read heads based on
entists demonstrated8,9 that the TMR a single TMR or a GMR sensor, this
TUNNEL
ratio could reach 18%8 and 11.8%,9 re- MRAM integrated more than 4 million
MAGNETORESISTANCE:
spectively, at room temperature when MTJs along with even higher numbers
LEAP BEYOND HARD DISK
an amorphous Al2O3 film was used as of deep sub-micrometer transistors and
DRIVES
a tunnel barrier. Several years later, the interconnects on a single die. Further-
Historically, advances in magneto- TMR ratio reached 70.4% at room tem- more, it cleared key manufacturing
electric devices have been attributed perature by optimizing the Al2O3 tun- concerns initially raised by semicon-
to new discoveries and evolutionary
improvements in magnetoresistance
(MR). The size of MR is measured
by the fractional change in resistance,
(Rhigh–Rlow)/Rlow, which is called the
MR ratio (commonly expressed in
percentage). A sensing device with a
higher MR ratio can detect weaker and
smaller signals, which is a key to high-
er capacity and performance, for ex-
ample, for hard disk drives. The read
head introduced by IBM in 1991 was
based on anisotropic magnetoresis-
tance (AMR) with a ratio on the order
of only 1%. Though the ratio was mi-
nuscule, the AMR read head was at that a b
time regarded as a remarkable accom- Figure 2. A comparison of MRAM cell architectures. (a) Conventional MRAM: MTJ switching
plishment for HDD. In comparison, occurs under a combination of the magnetic fields generated along the bit line and the write
word line. (b) STT MRAM: MTJ switching is triggered through a directly injected current. No
the GMR ratio was much higher, typi- write word line is necessary.
cally over a range of 5–20% at room

Vol. 60 No. 9 • JOM www.tms.org/jom.html 29

 bit (either “0” or “1”) into an MTJ (low
or high resistance, respectively). While
higher TMR is desirable for ensuring
fast sensing of the stored bit, an energy-
efficient and reliable writing mechanism
is critical for realizing a competitive
MRAM technology. In this regard, the
field-induced magnetization switching
(FIMS) MRAM, such as the first com-
mercial MRAM noted above, has two
serious shortcomings. First, the pro-
Figure 3. A schematic illustration of a common MTJ shape and film stack. An MTJ has gramming current is too high because
a form of an elliptic cylinder whose cross-sectional aspect ratio (b/a) defines shape metal lines (bit line and write word line
anisotropy. An MTJ film stack necessitates a combination of an antiferromagnet and a
synthetic antiferromagnetic (SAF) layer to ensure the stability of the “pinned” magnetization.
shown in Figure 2a) need to generate a
The tunnel barrier is typically on the order of 10 Å. sufficient magnetic field for the MTJ to
switch its storage-layer magnetization.
Second, the programming current does
ductor industries such as the introduc- group achieved a TMR ratio of 220% at not scale down even as the memory cell
tion of foreign materials into a silicon room temperature.14 This was remark- shrinks. Though different variations of
fab (contamination) and process com- able progress, but there would have FIMS MRAM have been developed to
patibility (thermal budget). been a drawback in adopting the ap- mitigate these two problems, their se-
However, the TMR ratio applied to proach for practical MTJ applications. verity largely remains unsolved. Cur-
this first-generation MRAM is not high In general, the ferromagnetic reference rently, FIMS MRAM is not considered
enough for next-generation applica- layer under MgO needs to be formed as to be a scalable technology toward sub-
tions. Much higher TMR ratios are re- a component of an exchange-coupled 100 nm technology generations, and
quired to achieve higher densities and synthetic antiferromagnetic (SAF) hence, is limited to niche products.
faster data sensing, in particular, when stack, as illustrated in Figure 3. In this In contrast, there is a completely dif-
the MTJs are integrated into leading case, it was difficult to grow a (001) ferent type of switching mechanism, ap-
silicon technologies. Though depen- textured MgO film that had a coherent plicable to MRAM in particular, which
dent on specific memory macro-attri- lattice match with the underlying fer- works more favorably as the MTJ size
butes, it is projected that the TMR ratio romagnetic film. This was solved by becomes smaller. In 1996, theories de-
should be higher than 100–150% to en- D.D. Djayaprawira et al.15 who devel- veloped by J.C. Slonczewski17 and L.
sure an adequate sensing margin and oped a slightly modified MTJ stack, Berger18 independently suggested that
speed. Such a high level of TMR was CoFeB/MgO/CoFeB. Unlike CoFe, the the magnetization of a thin ferromag-
often referred to as giant TMR11 (vs. as-deposited CoFeB electrodes were netic film can be switched by direct
GMR). But the conventional AlOx- amorphous while MgO was still (001) transfer of spin angular momentum
based MTJ cannot achieve it. Thanks to oriented. Through a mechanism called from a spin polarized current (via exert-
recent progress in ultra-thin film sci- solid-phase epitaxy, however, both ing torque). This mechanism is called
ence and engineering, a more promis- CoFeB electrodes can become crystal- spin-transfer-torque (STT) or spin-mo-
ing material was discovered and has lized into a coherent (001) oriented mentum-transfer (SMT) switching, or
advanced quickly, now poised for next- CsCl-type structure, consequently re- current-induced magnetization switch-
generation spintronic applications. sulting in a high TMR ratio (230%). ing (CIMS) to be differentiated from
Theoretical studies on spin-depen- The results suggest that the TMR effect FIMS. It has been known that such
dent tunneling in MTJ found that the in MgO-based MTJs is critically gov- STT phenomenon is pronounced only
TMR ratio can become much higher erned by the interfacial states and the in nano-scale magnets.19
than those predicted by Jullière’s phe- microstructures of the films, which are The writing principle of STT MRAM
nomenological model.7 If the tunneling dependent not only on film composi- is illustrated in Figure 4. The magni-
is coherent, for example, through tions but also on processing conditions. tude of the spin-transfer effect is de-
achieving a perfectly ordered (001) ori- By far, the highest TMR ratios were re- pendent on the current density in MTJ.
ented Fe/MgO/Fe MTJ,12,13 the first ported by Y.M. Lee et al.16 (500% at Accordingly, there is a critical current
principle calculation predicted that the room temperature and 1,010% at 5 K) density (Jc) to trigger switching. From
ratio could reach ~1,000%.13 One of the in pseudo-spin valve MTJs. a perspective of practical device appli-
experimental demonstrations support- cations, Jc is in general independent of
SPIN-TRANSFER-TORQUE
ing MgO’s capability of achieving a the MTJ size, so that the actual switch-
SWITCHING: THE
high TMR ratio was realized by S.S.P. ing current scales down as the MTJ
EMERGENCE OF STT MRAM
Parkin et al.14 By fabricating a highly size shrinks. This was a breakthrough
(001) textured MgO (NaCl-type) film The history of MRAM is essentially that should immediately solve one of
along with (001) textured CoFe (CsCl- characterized by advances in writing the two major problems of the FIMS
type) electrodes via sputtering, Parkin’s mechanisms: how to program a binary MRAM. Until a few years ago, how-

30 www.tms.org/jom.html JOM • September 2008

ever, STT MRAM did not attract much is widely regarded as a leader in this plexity. These problems translate to
attention beyond a laboratory research field, reportedly targeting an ultra-high compromises in power dissipation, per-
level due to several complicating fac- density (~Gbit) MRAM.22 Perpendicu- formance, and cost for a wide range of
tors. First, Jc was rather high on the or- lar magnetic anisotropy is more desir- IC products. While the end of Moore’s
der of 107 A/cm2. Second, process tech- able than the commonly applied shape Law scaling of ICs is not imminent
nologies were not mature enough for anisotropy (Figure 3) in achieving high (perhaps to continue for the next 10
fabricating good-quality sub-100 nm bit stability,23 hence allowing a smaller years or so), its value proposition is
MTJs to demonstrate the advantages and manufacturing-friendly circular being heavily eroded by the growing
of STT switching. Finally, the scalabil- MTJ geometry. The TMR ratio needs technology and economic concerns as
ity concern was less acute at that time. to be improved, however.22 the industry moves to nanoscale tech-
Most of the MRAM developers were The critical task in STT switching is nology nodes. Finally, beginning in
busy concentrating on enabling the first still to reduce Jc. Not only does Jc de- 2005, a forward-looking concept of
FIMS MRAM product. termine the programming power con- “More-than-Moore” was addressed by
The landscape started to change rap- sumption, but also it affects the memo- the International Technology Roadmap
idly beginning in late 2005 when re- ry cell size since the current is supplied on Semiconductors publications.24,25
searchers from Sony reported test-chip by the transistor serially attached to the It proposes a technology strategy of
results from a 4 kbit STT MRAM (Sony MTJ (see Figures 2b and 4). Extensive adding value to the traditional silicon
named it Spin-RAM).20 It was integrat- research is underway including efforts technology roadmap by creating func-
ed into a 180 nm CMOS platform with on finding new MTJ materials and tional diversification through SoC or
a typical MTJ size of 100 nm s 150 nm. structures. SiP which employs non-CMOS based
The MTJ includes a stack of CoFe/Ru/ technologies.
OPPORTUNITIES: IF YOU
CoFeB/MgO/CoFeB, which achieved a Conventional semiconductor memo-
BUILD IT, THEY WILL COME
high TMR ratio of 160% and a resis- ry technologies such as SRAM, DRAM,
tance-area product (RA) of 20 7Mm2. The emergence of STT MRAM is and Flash memory are based on the
More surprisingly, they demonstrated timely as industry-wide R&D efforts storage of electron charge, which has
switching as fast as 1 ns. In addition, Jc are being accelerated in pursuit of an become challenging to manipulate. In
was promising; for example, at 10 ns, alternative memory in lieu of conven- contrast, STT MRAM utilizes modulat-
Jc was lower than 3 s 106 A/cm2. The tional memories that are nearing fun- ing electron spins (a group of spin-po-
report also included endurance reliabil- damental scaling limits. For several de- larized electrons). The storage element
ity results showing no degradation after cades, the semiconductor industry has (i.e., MTJ) is made of sophisticated
1012 cycles of repeated read and write. relied on the evolution of denser tran- nano magnets whose data state depends
On the other hand, one of the parame- sistors to provide miniaturization and on the alignment of spins. The beauty
ters not addressed by Sony’s paper was lower cost per function for integrated is in that no artificial magnetic field is
data retention. In MRAM, data reten- circuit (IC) products. As the industry is necessary to manipulate the nano-mag-
tion is assessed in bit stability against migrating IC designs to 32 nm and be- nets but that the electric current gener-
environmental conditions (e.g., thermal yond, the limits of conventional CMOS ated by a common silicon transistor is
fluctuation), which is closely related to scaling are resulting in major problems, capable of sensing and programming
the switching energy barrier. including high leakage current, perfor- the storage element (by modulating
About a year after Sony’s report, mance saturation, increased parametric two resistance states). Such resistive el-
researchers from Hitachi and Tohoku variability, and wafer processing com- ements are more scalable than charge-
University accomplished another mile-
stone in STT MRAM, this time, in den-
sity.21 The joint R&D team designed
and fabricated a 2 Mbit test chip (named
SPRAM) based on a 200 nm CMOS
platform. The MTJ size was small, 50
nm s 100 nm. The performance speci-
fication was only modest: 40 ns read
and 100 ns write. But the highlight of
this work was a consideration of circuit
design parameters to build a memory
macro. One of the applied design tech-
niques was to mitigate read disturb,
which can become a circuit reliability
issue for STT MRAM. a b c
Another noticeable area of progress Figure 4. A schematic illustration of STT MRAM programming: (a) Switching to low R,
(b) to high R, and (c) I-V curve showing a switching loop. VDD, VGS, Vc+, and Vc- are the
is in adopting an MTJ based on per- bit or source line voltage, NMOS gate-voltage, switching voltage to low R, and to high R,
pendicular magnetic anisotropy (PMA) respectively.
for STT MRAM applications. Toshiba

Vol. 60 No. 9 • JOM www.tms.org/jom.html 31

storage elements. must be lowered by further reducing
Once programmed, the spin state the MgO thickness (currently, ~10Å)
is preserved until it is reprogrammed. as the MTJ area scales down.
Hence, STT MRAM is a nonvolatile The TMR effect is also dependent on
memory like Flash memory. Though the sensing bias (voltage). As shown in
Flash memory achieves smaller cell Figure 5, the ratio is highest at zero
sizes, STT MRAM is much faster in bias (measured at near-zero volt), but
programming (and erasure of the stored drops gradually as the bias increases.
bit), more energy efficient, and more What matters for sensing, therefore, is
CMOS-integration friendly when ap- the TMR ratio at the read voltage (com-
plied as an embedded memory. In addi- monly, 100~300 mV) which is one of
tion, STT MRAM has superior endur- the primary design parameters. An op-
ance characteristics to Flash memory Figure 5. A schematic illustration of timum read voltage should be deter-
TMR dependence on the MTJ voltage.
(typically less than 104 cycles). The TMR ratio decreases as the voltage mined by considering this TMR effect
It is known as the fastest nonvolatile increases. The read voltage (Vread) in conjunction with a sensing circuit
memory currently under development, is determined by considering sensing through rigorous co-optimization of
speed (TMR, signal voltage differential)
but the performance of STT MRAM and functional reliability (e.g., Vread << silicon device and MTJ electrical pa-
cannot currently reach that of embed- Vc+). rameters.
ded SRAM for high-end applications.
Switching Current: Controlling
But, it is indeed possible that STT
Variability
MRAM will be preferable in certain cessfully commercialized, but it has not
SRAM applications because it is non- been adopted as a mainstream technol- As addressed in a prior section, re-
volatile (static power saving) and offers ogy. ducing the switching current densi-
a much smaller die area (cost saving). In the case of STT MRAM, no one ty (Jc) is arguably the most important
Based on the attributes demonstrat- has yet demonstrated a production-wor- requirement for the success of STT
ed, STT MRAM appears to be more thy memory array, which indicates that MRAM. As Jc is lowered, the dynamic
suitable as an embedded nonvolatile the technology is still at an early R&D power per cell during programming is
memory for SoC applications than as stage. Lately, however, there have been decreased. This is beneficial not only
a stand-alone commodity memory. For encouraging signs that efforts on STT for power saving but also for band-
example, in 2005, Renesas Technology MRAM are advancing beyond labora- width increase. In addition, the mem-
Corporation announced that it would tory projects. As the R&D focus shifts ory cell size can be reduced, which can
develop embedded STT MRAM for au- toward solving engineering problems, result in lower cost.
tomotive microcontrollers. Some of the a few important technological subjects It is important to note that the mi-
anticipated benefits would be high per- are worthy of attention. cromagnetic nature of STT switching
formance and reliability. Reportedly, is statistical. Hence, there is an inevita-
Giant TMR: Only a Part of the
several major commodity memory sup- ble variation in switching current when
Story
pliers are also developing this technol- a cell is repeatedly programmed.27 A
ogy. Toshiba Corporation announced in A prior section highlighted a remark- significant contribution to this behav-
2007 that the company was developing able success in achieving high TMR ra- ior results from thermal agitation. The
PMA STT MRAM for Gbit capacity tios, recently up to ~500%,16 in CoFeB/ complex spin dynamics of STT switch-
applications. MgO/CoFeB-based MTJs. For useful ing is another factor. Furthermore,
device applications, however, there is there is a cell-to-cell switching current
CHALLENGES: EFFORTS
another critical MTJ parameter that variation caused by process variations
BEYOND LABORATORIES
must be coupled with a TMR ratio. It is that affect the MTJ area and shape. As
Historically, the path for an emerg- the resistance-area (RA) product of an a design parameter, therefore, the sup-
ing technology to be realized as a main- MTJ. The RA product must be in an ac- plied programming current must deal
stream semiconductor product has been ceptable range to match device parame- with all these variations to avoid a func-
elusive. One of the reasons is that such ters of the silicon logic technology. The tional glitch or failure. But the technol-
a vital industry cannot tolerate a risk if a range needs to be 10–40 7Mm2, depend- ogy will become much less competitive
technology is not guaranteed for a clear ing on the MTJ size. Unfortunately, the if the current is set in an overly con-
success, however promising it is. For TMR ratio often starts to drop sharp- servative way, since it will make the
example, it took more than 10 years for ly as the RA product decreases below cell size larger and the programming
Flash memory to reach its prime, when ~20 7Mm.26 This suggests that optimiz- power higher. Therefore, it is crucial
portable consumer electronic products ing the TMR ratio and the RA product that development efforts directed to-
such as cell phones and digital camer- simultaneously is critical to achieve a ward switching current reduction also
as demanded a nonvolatile solid-state desirable parametric combination tar- consider the reduction of the switch-
device like Flash memory. Recently, geted for specific device applications. ing current variability. A lesson shall
FRAM (ferroelectric RAM), another This task is particularly challenging come from the semiconductor industry,
type of nonvolatile memory, was suc- for STT MRAM if the MTJ resistance whose historical success is largely at-

32 www.tms.org/jom.html JOM • September 2008

tributed to the success of building ICs numbers should be used as a figure of 2. G. Binasch et al., Phys. Rev. B, 39 (1989), pp. 4828–
4829.
by controlling physical and functional merit or a deciding factor in designing 3. P. Grünberg, “Magnetic Field Sensor with Ferromag-
variations through reliable design and a memory cell due to uncertainties in netic Thin Layers Having Magnetically Antiparallel Po-
process engineering. model parameters and, more signifi- larized Components,” U.S. patent 4,949,039 (14 Au-
gust 1990).
cantly, lack of supporting silicon data. 4. S. Tehrani et al., Proc. of the IEEE, 91 (2003), pp.
Perhaps No “Trilemma,” but Still
If data retention becomes a dominant 703–714.
a Dilemma? 5. W.J. Gallagher and S.S.P. Parkin, IBM J. Res. & Dev.
issue, one may consider the following
50 (2006), pp. 5–23.
The “trilemma of magnetic record- technological approaches. First, a syn- 6. The first MRAM product was coded MR2A16A.
ing”28 was a phrase coined by mag- thetic ferromagnetic storage layer can Freescale was formerly Motorola. In 2008 Freescale
neticians from Seagate Technologies provide larger effective volume while spun off its MRAM division as EverSpin Technolo-
gies.
working on magnetic disk recording. maintaining the effective magnetic mo- 7. M. Jullière, Phys. Lett. A, 54 (1975), pp. 225–226.
It is often quoted to describe inevita- ment per area.29,30 Second, STT switch- 8. T. Miyazaki and N. Tezuka, J. Magn. Magn. Mater.,
ble tradeoffs among the three technol- ing based on perpendicular magnetic 139 (1995), pp. L231–L234.
9. J.S. Moodera et al., Phys. Rev. Lett., 74 (1995), pp.
ogy parameters in shrinking a magnet- anisotropy MTJs can substantially in- 3273–3276.
ic storage element: signal-to-noise ra- crease the bit stability against thermal 10. D. Wang et al., IEEE Trans. Magn., 40 (2004), pp.
tio (i.e., size), writability, and “thermal fluctuation.22,23 Perpendicular magnetic 2269–2271.
11. For example, refer to S. Yuasa, “Giant Tunneling
decay” (i.e., bit stability). This is not anisotropy was already adopted by Magnetoresistance in MgO-Based Magnetic Tunnel
necessarily applicable to STT MRAM, high-density HDD, but until recently Junctions,” J. Phys. Soc. Jpn., 77 (2008), pp. 1–13
which is not driven by field-induced has been considered challenging for (031001). Also, see Reference 14.
12. W.H. Butler et al., Phys. Rev. B, 63 (2001), pp. 1–12
switching. At least, the tradeoff is less MRAM applications. Third, a circuit (054416).
forceful since shrinking the bit (MTJ technique incorporating an error-cor- 13. J. Mathon and A. Umerski, Phys. Rev. B (2001), pp.
volume) is beneficial in reducing the recting code, commonly found in vola- 1–4 (220403).
14. S.S.P. Parkin et al., Nature Mater., 3 (2004), pp.
switching current. However, there is a tile SRAM and DRAM, may mitigate 862–867.
concern with maintaining thermal sta- the concern. 15. D.D. Djayaprawira et al., Appl. Phys. Lett., 86
bility as the MTJ volume is decreased (2005), pp. 1–3 (092502).
CONCLUSION 16. Y.M. Lee et al., App. Phys. Lett., 90 (2007), pp. 1–3
and the writability is increased by low- (212507).
ering the coercivity of the storage lay- Though engineering challenges re- 17. J.C. Slonczewski, J. Magn. Magn. Mater., 159
er. Though simplified, this situation is main, potential game-changing oppor- (1996), pp. L1–L7.
18. L. Berger, Phys. Rev. B, 54 (1996) pp. 9353–9358.
often described using an energy barrier tunities of STT MRAM are currently 19. J.Z. Sun, IBM J. Res. & Dev., 50 (2006), pp. 81–
(EB) against an unwanted bit reversal driving accelerated R&D efforts. The 100.
by thermal fluctuation.29,30 The charac- value proposition is timely since the in- 20. M. Hosomi et al., IEDM Tech. Dig. (2005), pp.
459–462.
teristic reversal time T is expressed by dustry is in pursuit of more scalable 21. T. Kawahara et al., ISSCC Dig. Tech. Papers
technologies beyond the maturing (2007), pp. 480–481.
¥E ´ ¥ H M V´ semiconductor technology roadmap. 22. M. Nakayama et al., “Spin Transfer Switching in
T  T o exp ¦ B µ  T o exp ¦ k s µ , TbCoFe/CoFeB/MgO/CoFeB/TbCoFe Magnetoresis-
§ kT ¶ § 2 kT ¶ The first success of the ongoing R&D tive Tunneling Junctions with Perpendicular Magnetic
efforts is to be measured against when Anisotropy” (Presentation at the 52nd Magnetism and
where To is known as the inverse of the the technology is proven as a product Magnetic Materials Conference, Tampa, Florida, 5–9
November 2007).
precession frequency, Hk is effective differentiator with sufficient silicon re- 23. X. Zhu and J.-G. Zhu, IEEE Trans. Magn., 42
anisotropy field, Ms is saturation mag- sults. Otherwise, the technology would (2006), pp. 2739–2741.
netization, V is volume of the storage remain only promising. 24. The International Technology Roadmap for Semi-
conductors: 2005 Edition, www.itrs.net/reports.html.
layer, k is Boltzmann constant, and T is Finally, it should be noted that STT 25. The International Technology Roadmap for Semi-
temperature. A common view is that to MRAM belongs to one class of spin- conductors: 2007 Edition, www.itrs.net/reports.html.
maintain the bit stability against ther- tronics (often referred to as metal spin- 26. K. Tsunekawa et al., IEEE Trans. Magn., 42 (2006),
pp. 103–107.
mal fluctuation, it is necessary to com- tronics). Researchers have been work- 27. Y. Huai et al., IEEE Trans. Magn., 41 (2005), pp.
pensate the decrease in V (smaller MTJ ing on futuristic devices that belong to 2621–2626.
area) by increasing Hk. However, this semiconductor spintronics and quan- 28. H.J. Richter, J. Phys. D: Appl. Phys., 40 (2007), pp.
R149–R177.
can make a negative impact on Jc, espe- tum computing (quantum bits instead 29. J. Janesky et al., Appl. Phys. Lett., 85 (2004), pp.
cially in the regime where STT switch- of binary bits).31–33 Practical applica- 2289–2291.
ing is thermally activated (vs. a rapid tions of such devices are not projected 30. K. Miura et al., VLSI Tech. Dig. Tech. Papers (2007),
pp. 234–235.
spin precession). to be realized in the foreseeable future. 31. S.A. Wolf, A.Y. Chtchelkanova, and D.M. Treger,
The consequence of low EB is poor But the successful enablement of STT IBM J. Res. & Dev., 50 (2006), pp. 101–109.
data retention. Nonvolatile memory of- MRAM will bring more attention and 32. D.D. Awschalom, R. Epstein, and R. Hanson, Sci-
entific American (October 2007), pp. 84–91.
ten requires 10 years of data retention momentum into such revolutionary 33. M.E. Flatté, IEEE Trans. Elec. Dev., 54 (2007), pp.
at 80°C. Specific EB numbers have been spintronics research. 907–920.
reported to meet such requirements
(40~60 times of kT depending on the References Seung H. Kang is senior staff manager and project
leader at Advanced Technology, Qualcomm Inc.,
memory macro size and the applica- 1. M.N. Baibich et al., Phys. Rev. Lett., 61 (1986), pp. 5775 Morehouse Drive, San Diego, CA 92121;
tion). However, it is debatable if those 2472–2475. [email protected].

Vol. 60 No. 9 • JOM www.tms.org/jom.html 33