An Overview of Spin-based Integrated Circuits

Abstract - Conventional CMOS integrated circuits suffer from

serve power and scalability challenges as technology node scales
into ultra-deep-micron technology nodes. Alternative approaches
beyond charge-only based circuits. In particular, spin-based
devices or integrated circuits show promising merits to overcome
these issues by adding the spin freedom of electrons to the
electronic circuits. Spintronics has now become a hot topic in
both academics and industrials. This paper overviews the status
and prospects of spin-based integrated circuits under intense
investigation and address particularly their merits and challenges
for practical applications.
I Introduction
Thanks to the good characteristics, such as high speed and
small size, metal-oxide-semiconductor field-effect transistors
(MOSFETs) and complementary MOS (CMOS) devices [1-3]
(hereafter, both are referred to as MOS devices) are the
fundamental technologies for mainstream integrated circuits.
However the rapid and continual technology scaling progress
of the MOS devices below 40 nm drives more and more issues
and challenges, e.g., intrinsic leakage currents, dynamic power
and process variability etc, indicating the end of MOS scaling
in the near future [2, 3]. Many new advanced solutions, e.g.,
silicon on insulator (SOI) [4], have been proposed recently in
order to alleviate this dilemma. They achieve one order of
enhancement in some aspects, but they cannot overcome all
the problems induced in the deep-micron technology nodes. In
addition, the production cost including facility investment
becomes very huge. Therefore alternative scaling-independent
technologies, i.e., spin-based devices and circuits [5-7], for
improving the integrated circuit performance have attracted
considerable attention to sustain the Moore`s Law beyond the
MOS scaling limit [2, 3]. Many research groups, including
academia and industries are undergoing for this purpose.
Spintronics (or spin-electronics) is a rapidly emerging R&D
area and it shows great promise in the future integrated
circuits (both memory and logic computing) [5-7]. The basic
concept of Spintronics is to control the spin property (besides
charge) of electrons, as shown in Fig. 1, in ferromagnetic thin
films based solid state nano-devices, such as spin-valve and
magnetic tunnel junction (MTJ). It provides new concepts for
future electronics, namely spin-based integrated circuits. The
founding step of Spintronics triggered the discovery of the
giant magnetoresistance (GMR) in 1988 (awarded Nobel Prize
Physics 2007 for A. Fert and P. A. Grunberg) [8, 9], and soon
triggered the definition of the spin-valve [10]. The spin-valve
sensor was firstly commercialized by IBM in 1997 to replace
the anisotropic magnetoresistance (AMR) sensor for hard disk
drive (HDD) read heads and drove immediately the storage
areal density growth rate of 100% per year. Another big step
forward for Spintronics came from replacing the metallic
spacer layer of the spin-valve with a thin non-magnetic
insulating oxide layer (e.g., Al
2
O
3
or MgO), thus creating the
magnetic tunnel junction (MTJ) [11-12]. In such configuration,
much higher resistance difference, denoted as tunnel
magnetoresistance (TMR) ratio, can be obtained at room
temperature. The discovery of MTJ triggered intensive
research and indicated a rapid development era on spin-based
integrated circuits, such as magnetic random access memory
(MRAM) and spin-transistors [5-7]. This paper overviews the
status and prospects of the spin-based integrated circuits
(including memory and logic circuits), and focuses mainly on
the period after the discovery of the MTJ.
The rest of this paper is organized as follows: section II
introduces the fundamentals of Spintronics. The spin-based
memory and logic designs are described in section III and
section IV respectively. Finally conclusions and perspectives
are discussed in section V.
II. Fundamental of Spintronics
Spintronics is an emerging technology exploiting both the
intrinsic spin degree of freedom and the magnetic moment of
the electron, in addition to its fundamental electrical charge, in
solid state devices. The origins of Spintronics can be traced
back to the 1970s [13] and the research of Spintronics widely
Wang Kang
1,2
, Weisheng Zhao
1*
, Zhaohao Wang
1
, Jacques-Olivier Klein
1
, Yue Zhang
1
, Djaafar
Chabi
1
, Youguang Zhang
2
, Dafin Ravelosona
1
, and Claude Chappert
1
2, Electronics and information engineering,
Univ. Beihang
Beijing, 100191, China
Tel : +86 01082314978
Fax : +81 01082339478
e-mail : [email protected]
1, Institute d`Electronique Fondamentale (IEF),
Univ. Paris-Sud, CNRS,
Orsay, 91405, France
Tel : (+33)169156292
Fax : (+33)169154000
e-mail : [email protected]
Fig. 1. Comparison between the principles of conventional
electronics and Spintronics [7]. (a) Conventional electronics uses
only the electrical charges; (b) Spintronics exploits the spin
degrees of freedom of electrons.
978-1-4799-2816-3/14/$31.00 2014 IEEE
676
8S-1
emerged from the discovery of spin-dependent electron
transport phenomena in solid-state devices in the 1980s [8, 9].
The discovery of spin valve or GMR in 1988 opens a new era
in Spintronics, and later in 1995, the discovery of MTJ or
TMR boosts its intensive research and applications [11, 12].
The MTJ nanopillar, as shown in Fig. 2, is one of the most
important devices of current spin-based integrated circuits.
Both perpendicular magnetic anisotropy (PMA) and in-plane
shape anisotropy MTJs can be designed with different thin
film composition.
The MTJ states, i.e., R
P
or R
AP
, can be switched by
applying either a magnetic field or a spin polarized current.
However conventional field-induced magnetic switching
(FIMS) approaches [14] require writing current in the order of
few mA to generate the required magnetic field and induce
consequently high power consumption and hardware area.
Much of academic and industrial research efforts have been
focused on developing efficient strategies for switching the
magnetization of the MTJ. One promising method relies on
the spin transfer torque (STT), which was firstly proposed
independently by Berger [15] and Slonczewski [16] in 1996,
shows good performance. Only a small bidirectional current is
required for the MTJ switching, which simplifies greatly the
integration with the peripheral MOS circuits and therefore
allows much higher density. Unfortunately according to the
experimental measurements and theoretical models, the STT
switching mechanism is normally stochastic (see Fig. 3 and
Eqs. (2)-(5)) [17-19], thus resulting in reliability issues, which
should be addressed in the practical spin-based applications.
Depending on the relative magnitude between the write
current I
wtc
and the critical current I
C0
(see Eq. (1)), the
STT switching behaviour of the MTJ can be categorized into
two regions: precessional switching region (I
wtc
> I
C0
) and
thermal activation region (I
wtc
< I
C0
).
I
C0
= o
yc

B
g
(p
0
H
S
)E
K
I = 2o
yc

B
g
E (1)
where is Gilbert damping coefficient, y the Gyro-magnetic
constant, c the magnitude of the electron charge, p
B
the
Bohr magneton constant, g the spin polarization efficiency
factor, p
0
H
s
is the saturation field in the free layer, E
K
the
anisotropy field, I the volume of the free layer and
E = (p
0
H
S
)E
K
I2 is the barrier energy.
When the current flowing through the MTJ exceeds the
critical current I
C0
, the MTJ experiences a fast precessional
switching regime, and the switching duration and probability
can be expressed as,
Pi (t
puIsc
) = 1 -exp (-
t
pulsc
:
1
) (2)
1
:
1
= j
2
C+In(n
2
A)
[

B
P
cm(1+P
2
)
(I
wtc
-I
C0
) (3)
where t
puIsc
is the driver current pulse duration, :
1
is the
mean duration for precessional switching regime, C = u.S77
is the Euler`s constant, A= Ek
B
I is the thermal stability
factor, k
B
the Boltzmann constant, I the temperature, m
the free layer magnetic moment, P the tunneling spin
polarization of the ferromagnetic layers. Otherwise, if the
current is lower than I
C0
, the switching process can still occur
with a long pulse thanks to the thermal activation. The
switching probability and pulse duration for the thermal
activation regime can be expressed as,
d Pr(t
pulsc
)
(1- Pr(t
pulsc
)) dt
=
1
:
2
(4)
:
2
= :
0
cxp (
L
k
B
1
(1 -
I
writc
I
C0
)) (5)
where :
0
is the attempt period, :
2
the mean pulse duration
for the thermal activation regime. In the practical spin-based
integrated circuits, we generally require the MTJ operates at
precessional switching regime and choose short pulse current
with amplitude larger than I
C0
to get high speed operations.
The MTJ resistance states can be sensed with either a bias
voltage I
bus
or a directional current I
cud
due to the TMR
effect. Generally a relative larger I
bus
or I
cud
is required
to improve the sense margin (SM). However it is worth noting
that the sensed current or I
cud
should be sufficiently less
than I
C0
to avoid any read disturbance (RD) (see Eq. (6)) for
the STT-based MTJ [19-21]. In addition, larger I
bus
may
lead to TMR loss (see Eq. (7)) [21].
Pi
ds
(t
cud
) = 1 -exp (-
t
rccd
:
0
exp (-A(1 -
I
rccd
I
C0
))) (6)
TNR
rcaI
=
TMR(0)
1+v
bics
2
v

2
(7)
where I
cud
and t
cud
are the amplitude and accumulated
duration of the sensing current, TNR
rcaI
is the TMR ratio
after adding bias voltage, TNR(u) is the TMR ratio with 0V
Fig. 2. Schematic structure of PMA magnetic tunnel junction
(MTJ) nanopillar: It is composed of an oxide barrier layer
sandwiched between two ferromagnetic (FM) layers. According
to the different configuration (Parallel or Anti-parallel) of the
two FM layers, the MTJ shows low or high (R
P
or R
AP
)
resistance property and the resistance difference value is denoted
by TMR=(R
AP
-R
P
)/R
P
[5]. Fig. 3. Experimental measurement of STT stochastic switching
behaviours [18].
677
8S-1
voltage, I

is the bias voltage as TNR

rcaI
= u.S TNR(u).
The STT-based MTJ has been widely used and become the
major candidate device, at least in the short term, in the
spin-based integrated circuits. Many STT-MTJ based memory
and logic circuits or prototypes have been presented in the last
years, such as STT-MRAM and Magnetic Flip-Flop (MFF) etc.
They show many great merits, such as high speed, low power,
non-volatility and instant on/off capability, especially the good
compatibility and scalability performance to continue the
Moore`s scaling law.
III. Spin-based memory circuits
There are different methods to categorize these spin-based
memory circuits. In this section, we will overview and classify
them from the device structure point of view.
A. Spin-based HDD read head
Spin-valve sensor for HDDs [5], as shown in Fig. 4, utilized
in 1997 by IBM, is the first commercialized application of
Spintronics, which provides a sensitive and scalable read
technique for HDD and brought huge economic interests. Both
geometries, i.e., current in plane and current perpendicular to
plane (CPP) were proposed, but the CPP geometry is much
better for integration and achieving minimum dimension. In
addition, the CPP configuration is also more favorable for
reducing the recoding track width. The spin-valve sensor
increased the HDD areal recording density by three orders of
magnitude (from ~0.1 to ~100 Gbit in
2
) between 1997 and
2003. However, this growth rate started to slow down after
2003, when other problems joined the limiting spin-valve head.
The MTJ read head, with the same structure as spin-valve
shown in Fig. 4, except replacing the metallic spacer layer M
with a thin non-magnetic insulating layer (e.g., Al
2
O
3
or MgO),
was then commercialized in 2004 by Seagate. The MTJ read
head promises to achieve over 200 Gbit in
2
recording density
and 1 Gbit/second data rate.
B. Magnetoresistive Random Access Memory (MRAM)
MRAM is mainly based on the hybrid structure, i.e., the
magnetic memory elements (generally referred to the MTJ)
integrated with the MOS devices. There exist two basic array
architectures, i.e., one-transistor with one MTJ (1T-1MTJ) and
cross-point architectures [5, 22-23]. In the 1T-1MTJ array
architecture, as shown in Fig. 5, each MTJ is connected in
series with an MOS transistor, where the gate of the transistor
is connected to the word-line (WL), drain to the bit-line (BL)
crossing the MTJ and source to the source-line (SL). It is
convenient for selecting the cell for the write/read operations.
However the density of the 1T-1MTJ array architecture is less
than the cross-point architecture due to the transistor required
for each memory cell. In the cross-point array architecture, the
MTJs lie at the intersection of the WLs and BLs, as shown in
Fig. 6, where the pinned layer and free layer of the MTJ are
connected directly to the BLs and WLs respectively. This
arrangement allows for a considerable packing density, since
no contact is made to the silicon die within the cell. However
it involves several significant design challenges, such as low
data access speed and sneak currents, leading to poor
write/read performance.
The read operation of the MRAM utilizes a bias voltage
on the BL to measure the effective current, or applies a current
flowing through the cell to measure the voltage. There are
three basic decision-making methods to determine the data
value stored in the memory cell: the reference-cell method,
complementary-cells comparison method and self-referenced
method [23-26]. In the reference-cell method, each sensed
value of the memory cell is compared to a reference cell, with
resistance value R
rcI
= (R
P
+ R
AP
)2. This method should
be designed with consideration of the process variations,
especially as technology scales down to the ultra-deep-micron
Fig. 4. The spin-valve read head for hard-disk recording. (Left)
The schematic configuration oI the current in plane` geometry;
(Right) The current perpendicular to plane (CPP)` geometry [5].
Fig. 5. Schematic of MRAM structure [5]. (a) Data representation
with different MTJ configurations; (b) Schematic of the 1T-1MTJ
cell structure; (c) Schematic of the 1T-1MTJ array architecture.
Fig. 6. Schematic of the cross-point architecture for MRAM [22].
678
8S-1
technology nodes. The second method uses two MTJs, which
are always written to opposite states, to store one bit of data.
This method can improve greatly the sense margin (SM)
compared with the first one, but with lower storage density.
The self-referenced method accesses the memory cell twice,
and makes the decision based on a known value written into
the cell at the second time or based on the physical
characteristics of the MTJ R-I curves. This method requires no
hardware for the reference cells, and it is insensitive to process
variations. However the repeated sense cycles result in
considerable read latency and power consumption. In addition,
it also leads to endurance reliability issues.
The write operation of the MRAM is different based on the
witching mechanism of the MTJ, and we can classify the
MRAM family into several categories as following (see also
Fig. 7): (a) Filed induced magnetic switching (FIMS) MRAM;
(b) spin transfer torque (STT) MRAM; (c) Thermally-assisted
switching (TAS) MRAM.
FIMS-MRAM
FIMS-MRAM [14] is the first generation of MRAM that
based on the field-only writing mechanism, as shown in Fig. 7
(a), where the magnetic fields are generated by two orthogonal
current lines. In such approach, the write selectivity is based
on the combination of two perpendicular pulses of magnetic
fields, i.e., B
x
and B
y
, which may result in narrow write
margin and half-selectivity issues. In addition, high power
consumption (magnetic fields requiring write currents of about
7-10 mA) and poor scalability (selectivity problems caused by
magnetic field dispersion) limit its commercialization. Later in
2003, the selectivity problem is solved by introducing a new
write method, named toggle-switching [27], which is based on
the use of a synthetic ferrimagnetic (SF) free layer and is
commercialized in 2006 by Everspin. However this method
needs to read before write and it also consumes high write
power. Most importantly, the general downscale scalability
problem (cell size about 30 F
2
) with field writing mechanism
cannot be well overcame.
STT-MRAM
STT-MRAM [16-17, 28], as shown in Fig. 7 (b), uses a
bi-directional low current (~1-2mA/cm
2
) to switch the MTJ,
and this simplifies greatly the integration with the peripheral
MOS circuits, thus achieving very good scalability (down to
10 nm diameter) and high integrated density (with minimum
cell size of 6 F
2
). The key challenge for STT-MRAM is to
achieve low write current, high speed, good endurance as well
as long retention simultaneously, because (a) both I
C0
and the
thermal stability factor A is proportional to the barrier energy
E (see Eq. (1) and (3)), there exists a conflict between low
current and long retention; (b) to get high speed (i.e., short
write duration t
puIsc
), a high write current I
wtc
is required
(see Eq. (3) and (5)), therefore a high bias voltage I
bus
or
low resistance.area product (R.A) is needed to achieve this
purpose, as shown in Fig. 8. However, both the two solutions
lead to the oxide barrier breakdown and thus poor endurance
of the MTJ.
TAS-MRAM
To solve the dilemma between write performance and
retention, and meanwhile to continue the MRAM downsize
scalability, a new concept, named thermally assisted switching
(TAS) was proposed [29, 30]. As shown in Fig. 7 (c) and (d),
this approach uses firstly a current pulse flowing through the
cell to heat temporarily the free layer of the MTJ above its
magnetic ordering temperature by Joule dissipation, greatly
reducing the required magnetic fields or STT current for the
MTJ switching. Then a magnetic field or current is applied for
the magnetic switching of the MTJ. Finally the cell is rapidly
cooled down to the room temperature and the magnetization
of the MTJ subsequently remains frozen in the new direction.
In such technique, the switching field or current can be fixed
whatever the size of the MTJ, allowing a largest integration
density and resolving the selectivity problem (only the
addressed MTJ is heated). Moreover, the MTJ was written at
elevated temperature and stored/read at room temperature,
ensuring a high thermal stability and long data retention.
MRAM is an emerging non-volatile random access memory
technology under development since the 1990s and is still
under intensive research. Many prototypes or chips have been
proposed or commercialized in markets currently. We believe
that MRAM will eventually become dominant for all types of
Fig. 7. 1T-1MTJ cell structures of various MRAM families. (a)
Magnetic field-only FIMS-MRAM; (b) STT-MRAM; (c) TAS
FIMS-MRAM; (d) TAS STT-MRAM.
Fig. 8. To obtain high I
wtc
for fast speed, either high I
bus
or
low R.A is required, leading to poor endurance of the MTJ [21].
679
8S-1
memory due to its overwhelming merits, becoming a universal
memory one day. In summary, we compare the performance of
various MRAM families, as shown in Table I.
C. Domain-wall (DW) racetrack memory
The observation of the current-induced domain wall (DW)
motion in magnetic nanowires triggered the 3D storage
devices, i.e., racetrack memory (RM) [31-32], as shown in Fig.
9, which promises to offer ultra-high storage density by
storing the data in a U-shaped nanowire normal to the plane of
the substrate. Combining with the MTJ nanopillars as the read
and write heads, good CMOS integrability and fast data access
can be achieved. The frst RM prototype Iabricated on 90 nm
node has been presented recently; however, it is based on the
in-plane magnetic anisotropy in NiFe nanowires with intrinsic
low energy barrier, leading to insufficient data retention.
Recent progress demonstrates that the perpendicular magnetic
anisotropy (PMA) materials (e.g., CoFeB) can further improve
the storage density, access speed and power consumption of
the RM. Nevertheless many challenges should be addressed
before this device can be used in industry. One of the key
challenges to build RM is to avoid any pinning defects in the
magnetic strips, because even one single pinning defect may
prevent the DW motion of the whole track.
D. Advanced spin-based memories
With the continual academic and industrial progresses on
Spintronics, some new advanced spin-based memory concepts,
such as voltage-controlled (VC) MRAM [33], multi-level-cell
(MLC) MRAM [34] and multi-terminal structures [35, 36] etc,
are proposed recently. The VC-MRAM uses the electrical filed
through a voltage to assist or accomplish the switching of the
MTJ, resulting in lower power consumption. MLC-MRAM
formed by modifying the free layer structure of the MTJ or by
stacking multiple MTJs in parallel or series, can store multiple
bits in one memory cell, thus doubling or tripling the storage
density. The multi-terminal cell structures based on spin-orbit
torque effect: spin-hall or Rashba, as shown in Fig. 10, can
indeed provide some aspects of advantages, e.g., improved
endurance by reducing the electrical stress on the oxide barrier,
but they also induce some drawbacks, e.g., larger cell size due
to the two access MOS transistors. In summary, all these
advanced concepts are still far away to be considered as
promising approaches for practical applications due to their
maturities and technological feasibilities.
IV. Spin-based logic circuits
Classic computing model of microprocessor is based on the
Von-Neumann architecture [37], as shown in Fig. 11 (a),
which consumes both high static power (due to the volatile
cache memories based on MOS devices) and dynamic power
(due to the data traffic between the CPU core and the main
memories), limiting its further downsize scalability. For
instance, the power to access the memory for fetching the
instructions and reading/writing the data (e.g. ~1 pJ/bit/mm) is
much higher than that for performing logic operations (e.g. ~
1 fJ at 22 nm node). Therefore the spin-based logic devices
and circuits, such as hybrid MTJ/CMOS logic structures,
all-spin logic devices and spin-based transistors [38-41] etc,
are under intensive investigation. They are expected to bring
the non-volatility into the MOS circuits and then allow them
to achieve instant on/off operations (i.e., powered off when
unused and retrieved instantly on active state), reducing
TABLE I
Comparison of Various MRAM Technologies
Technology
FIMS
MRAM
STT
MRAM
TAS-FIMS
MRAM
TAS-STT
MRAM
Scalability
Poor
(>60 nm)
Very good
(<10 nm)
Good
(>40 nm)
The best
(<8 nm)
Min cell size
Large
(~30 F
2
)
Very small
(~6 F
2
)
Small
(~10 F
2
)
The best
(~4 F
2
)
Endurance 10
16
10
16
10
12
10
12
Writability Poor Very good Good The best
Power Very high Low High The best
Latency
Very long
(>20 ns)
Short
(<10 ns)
Long
(>10 ns)
The best
(<8 ns)
Fig. 9. Schematic of domain-wall (DW) racetrack memory (RM)
based on current induced DW motion.
Fig. 10. Schematic of the multi-terminal cell structures based on
the spin-orbit torque effect. (a) 3-terminal cell structure; (b)
4-terminal differential spin-hall cell structure.
Fig. 11. Schematic of logic computing architectures. (a) Classic
Von-Neumann architecture; (b) The 3D integration of spin-based
logic architecture.
680
8S-1
greatly the static power. In addition, thanks to the vertical
structure of the spin-based storage devices (e.g., MTJ), they
can be fabricated above the MOS circuits at the back-end
process. This 3D integration structure [7], as shown in Fig. 11
(b), shortens greatly the data traffic distance between the
memory and logic chips, thus accelerating greatly the logic
computing speed and saving significantly the dynamic power.
In the following, we will overview several spin-based logic
devices and structures that draw much attention currently.
A. Hybrid MTJ/CMOS logic circuits
The hybrid MTJ/CMOS logic circuits are mainly based on
the logic-in-memory structure [38, 39], as shown in Fig. 12. It
is mainly composed of four parts: a sense amplifier (S.A) to
evaluate the logic result, a non-volatile memory block (e.g.,
MRAM), a write circuit, and a volatile MOS logic block. This
type of spin-based logic circuits is the most popular one
currently and draws considerable attention due to its instant
on/off capability, zero standby power, good compatibility with
conventional computing architectures and easy integration
with the existing MOS technology process. Many hybrid
MTJ/CMOS logic circuits or prototypes have been presented
in the past few years, such as magnetic flip-flop (MFF),
magnetic full adder (MFA) and magnetic look-up table
(MLUT) etc. However they also suffer from some challenges
which should be addressed, e.g., the switching latency (several
ns) of the MTJ is much larger than those of the conventional
MOS logic circuits, which limits the computing frequency to
the order of GHz. Another severe issue is the poor sense
reliability caused mainly by the device mismatch (both MOS
and MTJs devices) of the S.As and the intrinsic stochastic
switching effects of the MTJs. Different from the memory
circuits where complex error correction circuits (ECCs) can be
employed [42], it is difficult to embed them in the logic
circuits while keeping fast speed, low area and high power
efficiency. Therefore the current efforts that concentrate on
this topic are fast-access MTJ development, high-performance
S.A design, low-cost and reliable integration process etc.
B. Domain wall based logic circuits
Besides memory circuits, domain wall (DW) motion in
magnetic nanowires provides also the ability of logic designs.
One of the demonstrations is to use geometry dependence of
the DW motion to perform logic computation [43], which uses
no MOS transistors and exhibits ultra-low power consumption.
However, these circuits utilize magnetic field to drive the DW
movement and have some critical shortcomings, such as low
speed (<100 KHz), magnetic field dissipation, scalability and
reliability issues etc, for practical applications. The current
induced magnetic DW motion becomes an effective solution
to overcome these issues [31-32]. An alternative design is
based on the DW motion racetrack memory (RM), as shown
in Fig. 13 is an example of a MFA circuit [44]. It employs also
the general logic-in-memory structure to perform the logic
computing operations. Unlike the hybrid MTJ/CMOS logic
structure, where parts of the input data are volatile provided
by the MOS circuits (see Fig. 12), in the DW motion based
logic circuits, all the input and output data are stored in the
non-volatile RMs, overcoming completely the standby power
issue and achieving a real non-volatile logic circuit. However,
all the data inputs are stored in non-volatile state, which is
difficult to be realized with a reasonable area and latency
overheads. In addition, it faces the same challenges (e.g.,
pinning defects) as introduced in the RM circuits.
C. All-spin logics
All spin logic devices (ASLD) [40], as shown in Fig. 14 (a),
which employs nano-magnets as digital spin capacitors to
store data information and spin currents (through spin transfer
torque) to communicate, realizing logic gates based on the
spin majority evaluation. As shown in Fig. 14 (b) is an
example to demonstrate the possible layouts for constructing
cascadable ASLD logic gates. The magnetization directions of
the nano-magnets can be switched between the stable states if
enough torque is exerted on them. Information stored in the
input magnet is used to generate a spin current that can be
routed along a spin-coherent channel to the output magnet,
determining its state based on the spin transfer torque effect.
The key feature of ASLD is its compactness and completeness,
because no MOS transistor is needed for the logic operations
and all the logic functions can be constructed with a minimal
set of Boolean logic gates. With such design, a full spin
Fig. 12. Schematic of the hybrid MTJ/CMOS logic circuits. (a)
Logic-in-memory structure; (b) An example of 'XOR gate.
Fig. 13. Schematic of a magnetic full adder (MFA) based on the
current induced domain wall motion in magnetic nanowires.
681
8S-1
computing system can be expected with extremely low
switching power. However this is still a theoretical prospect
currently and many issues, such as reliability and clock
control, are remaining unresolved. The most critical challenge
for this ASLD is the 'magic material with strong spin orbit
interaction and there are few experimental prototypes can
produce successfully the theoretical results so far.
D. Spin-transistors
The concept of spin-transistor has been predicted early in
the 1990s [45], but it was experimentally developed recently
thanks to the rapid progress of ferromagnetic study on spin
orbit interaction, and then a wide variety of spin transistors
based on various operating principles have been proposed so
far [46-47]. Spin-transistors, including 'spin-MOSFET and
'spin-FET devices, benefit from the similar structure as MOS
transistors, as shown in Fig. 15. In the spin-FET and its related
devices, the source and drain have the same spin alignment,
and the on/off switching operation can be achieved by spin
precession of the spin-polarized carriers injected in the
channel through spin-orbit interaction. Here the spin-orbit
interaction is controlled by the gate voltage. It is worth noting
that particular channel materials with strong spin-orbit
coupling, such as InGaAs, InAs and other III-V compounds,
are required to sufficiently induce the spin-orbit interaction.
However, the channel region of the spin-MOSFET requires a
material with low spin-orbit coupling, since spin-MOSFET
requires no spin precession of spin-polarized electrons in the
channel. In the spin-MOSFET devices, the alignment of the
drain magnetization is fixed, while that of the source can be
changed, so the gate allows current to flow from the source to
the drain without modulation. In contrast to the spin-FETs, the
cutoff state of the spin-MOSFET is simply achieved by a gate
bias condition in the same manner as an ordinary MOS
transistor. In both type of devices, the spin is injected from the
ferromagnetic source, and then transported through the
channel to the drain and electrons with spin aligned with the
drain are passed and generate current. Spin transistors provide
a potential building element for novel integrated circuits and
open a promising path for achieving real all-spin based
integrated circuits.
E. Other spin-based logic devices and circuits
There are many other spin-based logic devices and circuits,
such as spin-valve logic gates and graphene-based transistors
etc [48-52]. Among them, graphene (awarded Nobel Prize in
Physics in 2010 [51]) has been widely studied recently due to
its ultra-high charge carrier mobility and long spin diffusion
lengths, allowing it to be used as a promising channel material
in the spin transistor. In 2008, the smallest graphene-based
transistor was reported [50] to be successfully fabricated with
only one atom thick and 10 atoms wide, and later in 2011,
IBM announced that they had succeeded in creating the first
graphene-based integrated circuit [52]. Graphene has proved
generally the potentiality and capability of replacing silicon as
a mainstream semiconductor. However many issues and
challenges should be addressed before it can be widely used in
industries. One of the key challenge is that single sheet of
graphene is very hard to produce, and even harder to make on
top of an appropriate substrate.
V. Summary and Conclusions
In this invited paper, we presented a general overview of
the spin-based integrated circuits and addressed particularly its
using as memories and logics by adding the spin freedom of
electrons. The rapid progress in both physics and electronics
makes them promising to overcome the power and scalability
bottlenecks of conventional MOS-based integrated circuits,
sustaining the Moore`s Law beyond the MOS scaling limit.
Recently many prototypes or small-scale products have been
successfully presented or commercialized, however many
challenges should be relieved before they can be widely used
in practical applications. For this purpose, many efforts have
been done or are undergoing in both academics and industries.
We believe that spin-based integrated circuits will become
mainstream solution to build up the next-generation storage
and computing systems.
Acknowledgments
This work was supported by ANR-DIPMEM, ANR-MARS,
NVCPU projects and the CSC exchange program.
References
[1] S. Kang, and Y. Leblebici, CMOS digital integrated circuits,
Third Edition, McGrawHill publisher, 2002.
[2] International technology roadmap for semiconductor (ITRS),
2010. http://www.itrs.net.
Fig. 14. (a) Schematic of all-spin logic device (ASLD) structures.
(b) An example of full adder layout based on ASLDs.
Fig. 15. Schematic device structures of the spin transistors. (a)
spin-MOSFET device; (b) spin-FET device [41].
682
8S-1
[3] S. E. Thompson, S. Parthasarathy, 'Moore's law: the future of Si
microelectronics, Mater. Today, Vol. 9 No. 6, pp. 20-25, 2006.
[4] A. Marshall, and S. Natarajan, SOI design: analog, memory and
digital techniques, Springer, 2002.
[5] C. Chappert, A. Fert and F. Dau, 'The emergence oI spin
electronics in data storage, Nature Mater., Vol. 6, pp. 813-823,
2007.
[6] I. Zutic, J. Fabian, and S. D. Sarma, 'Spintronics: fundamentals
and applications, Rev. Mod. Phys., Vol. 76, PP. 323-410, 2004.
[7] W. S. Zhao, et al, 'Spin-electronics based logic fabrics, in Proc.
IEEE VLSI-SOC, 2013, in press.
[8] M. N. Baibich, et al, 'Giant magnetoresistance of
(001)Fe/(001)Cr magnetic superlattices, Phys. Rev. Lett. Vol.
61, pp. 24722475, 1988.
[9] G. Binasch, et al, 'Enhanced magnetoresistance in layered
magnetic structures with antiferromagnetic interlayer exchange,
Phys. Rev. B, Vol. 39, PP. 4828-4830, 1989.
[10] B. Dieny, et al. Giant magnetoresistance in soft ferromagnetic
multilayers. Phys. Rev. B, Vol. 43, pp. 1297-1300, 1991
[11] J. S. Moodera, L. R. Kinder, T. M. Wong, and R. Meservey,
'Large magnetoresistance at room temperature in ferromagnetic
thin film tunnel junctions, Phys. Rev. Lett, Vol. 74, PP.
3273-3276, 1995.
[12] S. S. P. Parkin, et al, 'Giant tunnelling magnetoresistance at
room temperature with MgO (100) tunnel barriers, Nature
Mater, Vol. 3, PP. 862867, 2004.
[13] M. Julliere, 'Tunneling between ferromagnetic films, Phys.
Lett. A, Vol. 54, pp. 225-201, 1975.
[14] B. N. Engel et al., 'A 4-Mb toggle MRAM based on a novel bit
and switching method, IEEE Trans. Magn. Vol. 41, pp.
132-136, 2005.
[15] L. Berger, 'Emission oI spin waves by a magnetic multilayer
traversed by a current, Phys. Rev. B., Vol. 54, PP. 9533, 1996.
[16] J. C. Slonczewski, 'Current-driven excitation of magnetic
multi-layers, J. Magn. Magn. Mater., Vol. 159, pp. L1L7.
1996
[17] D. C. Worledge, et al., 'Spin torque switching oI perpendicular
Ta|CoFeB|MgO based magnetic tunnel junctions, Appl. Phys.
Lett., Vol. 98, no. 2, pp. 022501, 2011.
[18] Y. Lakys, et al., 'SelI-enabled 'error-eree switching circuit for
spin transfer torque MRAM and logic, IEEE Trans. Magn., Vol.
48, no. 9, pp.2403-2406, Sept. 2012.
[19] L. Faber et al., 'Dynamic compact model oI spin-transfer torque
based magnetic tunnel junction (MTJ), in Proc. IEEE DTIS, pp.
130135, 2009.
[20] W. Kang, et al. 'High reliability sensing circuit for deep
submicron spin transfer torque magnetic random access
memory, Electron. Lett., Vol. 49, PP. 1283-1285, 2013.
[21] W. S. Zhao, et al, 'Failure and reliability analysis oI
STT-MRAM, Microelectron. Reliab., Vol. 52, pp. 1848-1852,
2012.
[22] W. Zhao, et al, 'Cross-point architecture for spin transfer torque
magnetic random access memory, IEEE Trans. Nanotech., Vol.
11, pp. 907-917, 2012.
[23] T. M. Maffitt, et al, 'Design considerations for MRAM, IBM J.
Res. Devel., Vol. 50, pp. 25-39, 2006.
[24] W. Zhao, C. Chappert, V. Javerliac and J. P. Noziere, 'High
stability and low power sensing amplifier for MTJ/CMOS
hybrid logic circuits, IEEE Trans. Magn., Vol. 45, no. 10, pp.
3784-3787, 2009.
[25] Z. Sun, H. Li, Y. Chen, and X. Wang, 'Voltage driven
nondestructive self-reference sensing scheme of spin-transfer
torque memory, IEEE Trans. VLSI. Syst., Vol. 20, pp.
2020-2030, 2012.
[26] G. Jeong, et al, 'A 0.24-m 2.0-V 1T1MTJ 16-kb nonvolatile
magneto-resistance RAM with self-reIerence sensing scheme,
IEEE J. Solid-State Circuits, Vol. 38, no. 11, pp. 19061910,
2003.
[27] L. Savtchenko, B. N. Engel, N. D. Rizzo, M. DeHerrera, and J.
Janesky, 'Method of writing to scalable magnetoresistance
random access memory element, U. S Patent 6545906, 2003.
[28] Y. Huai, and P. P.Nguyen, 'Magnetic element utilizing spin
transIer and an MRAM device using the magnetic element, U.S.
Patent 6920063, 2005.
[29] I. L. Prejbeanu, et al., 'Thermally assisted MRAM, Jour. Phys.
Conden. Matter, Vol. 19, PP. 165218, 2007.
[30] Z. Li, and S. Zhang, 'Thermally assisted magnetization reversal
in the presence of a spin-transIer torque, Phys. Rev. B, Vol. 69,
pp. 134416, 2004.
[31] S. S. P. Parkin, M. Hayashi, and L. Thomas, 'Magnetic
domain-wall racetrack memory, Science, Vol. 320, no. 5873, pp.
190-194, 2008.
[32] Y. Zhang, et al., 'Perpendicular-magnetic-anisotropy CoFeB
racetrack memory, J. Appl. Phys., Vol. 111, pp. 093925, 2012.
[33] N. Lei, et al., 'Strain-controlled magnetic domain wall
propagation in hybrid piezoelectric/ferromagnetic structures,
Nature Commun., Vol. 4, pp. 1378, 2013.
[34] M. Aoki, et al., 'Novel highly scalable multi-level cell for
STT-MRAM with stacked perpendicular MTJs, in Symp. VLSIT,
pp. 134-135, 2013.
[35] I. M. Miron, et al., 'Perpendicular switching oI a single
ferromagnetic layer induced by in-plane current injection,
Nature, Vol. 476, pp. 189-193, 2011.
[36] Y. Kim, S. H. Choday, and K. Roy, 'DSTT-MRAM: differential
spin hall MRAM for on-chip memories, arXiv:1305.4085v1,
2013.
[37] G. J. Myers. Advances in computer architecture. John Wiley and
Sons, 1982.
[38] H. Dery, and P. Dalal, 'Spin-based logic in semiconductors for
reconfigurable large-scale circuits, Nature, Vol. 447, PP.
573-576, 2007.
[39] W. Zhao, et al., 'High performance SoC design using magnetic
logic and memory, in VLSI-SoC: Advanced Res. Syst. Chip, Vol.
379, pp. 10-33, 2012.
[40] B. Behin-Aein, D. Datta, S. Salahuddin, and S. Datta, 'Proposal
for an all-spin logic device with built-in memory, Nature
nanotech, Vol. 5, pp. 266-270, 2010.
[41] S. Sugahara, J. Nitta, 'Spin-transistor electronics: an overview
and outlook, in Proc. IEEE, Vol. 98, pp. 2124-2154, 2010.
[42] W. Kang, et al, 'A low-cost built-in error correction circuit
design for STT-MRAM reliability improvement, Microelectron.
Reliab, Vol. 53, pp. 1224-1229, 2013.
[43] D. A. Allwood, et al., 'Magnetic domain-wall logic, Science,
Vol. 309, pp. 1688-1692, 2005.
[44] H. P. Trinh, et al., 'Domain wall motion based magnetic adder,
Electron. lett., Vol. 48, pp. 1049-1051, 2012.
[45] S. Datta, and B. Das, 'Electronic analog oI the electrooptic
modulator, Appl. Phys. Lett., Vol. 56, pp. 665667, 1990.
[46] J. Schliemann, J. C. Egues, and D. Loss, 'Nonballistic spin field
effect transistor, Phys. Rev. Lett., Vol. 90, pp. 146801, 2003.
[47] S. Sugahara and M. Tanaka, 'spin metal-oxide-semiconductor
field effect transistor using half-metallic-ferromagnet contacts
for the source and drain, Appl. Phys. Lett., Vol. 84, pp.
23072309, 2004.
[48] T. Schneider et al., 'Realization oI spin-wave logic gates, Appl.
Phys. Lett, Vol.92, 022505, 2008.
[49] B. Dlubak et al., 'Highly eIIicient spin transport in epitaxial
graphene on SiC, Nature Phys., Vol.8, pp.557-561, 2012.
[50] L. A. Ponomarenko, et al., 'Chaotic dirac billiard in graphene
quantum dots, Science, Vol. 320, pp. 356- 358, 2008.
[51] K. S. Novoselov, et al., 'Electric field effect in atomically thin
carbon films, Science, Vol. 306. pp. 666-669, 2004.
[52] Y. M. Lin, et al., 'WaIer-scale graphene integrated
circuit, Science, Vol. 332, pp. 1294-1297, 2011.
683
8S-1