Simulation of spin field effect transistors: Effects of tunneling and spin relaxation on

performance
Yunfei Gao, Tony Low, Mark S. Lundstrom, and Dmitri E. Nikonov

Citation: Journal of Applied Physics 108, 083702 (2010); doi: 10.1063/1.3496666

View online: http://dx.doi.org/10.1063/1.3496666
View Table of Contents: http://scitation.aip.org/content/aip/journal/jap/108/8?ver=pdfcov
Published by the AIP Publishing

Articles you may be interested in

Simulating realistic implementations of spin field effect transistor
J. Appl. Phys. 109, 07C306 (2011); 10.1063/1.3536460

The influence of nitrogen incorporation on performance and bias temperature instability of metal oxide
semiconductor field effect transistors with ultrathin high- k gate stacks
Appl. Phys. Lett. 93, 193506 (2008); 10.1063/1.3025420

Space charge layers in organic field-effect transistors with Gaussian or exponential semiconductor density of
states
J. Appl. Phys. 101, 024514 (2007); 10.1063/1.2424397

High-performance Zn O Zn Mg O field-effect transistors using a hetero-metal-insulator-semiconductor structure

Appl. Phys. Lett. 89, 053502 (2006); 10.1063/1.2261336

Ambipolar organic field-effect transistors with air stability, high mobility, and balanced transport
Appl. Phys. Lett. 88, 133508 (2006); 10.1063/1.2190445

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
23.29.46.242 On: Wed, 07 May 2014 04:08:48
 JOURNAL OF APPLIED PHYSICS 108, 083702 共2010兲

Simulation of spin field effect transistors: Effects of tunneling and spin

relaxation on performance
Yunfei Gao,1,a兲 Tony Low,1 Mark S. Lundstrom,1 and Dmitri E. Nikonov2
1
Department of Electrical and Computer Engineering, Purdue University, West Lafayette, Indiana 47906,
USA
2
Components Research, Intel Corporation, 2200 Mission College Blvd., Santa Clara, California 95052,
USA
共Received 1 February 2010; accepted 31 August 2010; published online 18 October 2010兲
A numerical simulation of spin-dependent quantum transport for a spin field effect transistor is
implemented in a widely used simulator, nanoMOS. This method includes the effect of both spin
scattering in the channel and the tunneling barrier between the source/drain and the channel.
Accounting for these factors permits setting more realistic performance limits for the transistor,
especially the magnetoresistance, which is found to be lower compared to earlier predictions. The
interplay between tunneling and spin scattering is elucidated by numerical simulation. Insertion of
the tunneling barrier leads to an increased magnetoresistance. Simulations are used to explore the
tunneling barrier design issues. © 2010 American Institute of Physics. 关doi:10.1063/1.3496666兴

I. INTRODUCTION the SBs and the electrostatic potential in the channel. The
magnetizations of the source and drain can be switched to be
Spin-based logic is currently being explored as a poten- parallel or antiparallel to obtain low or high resistance be-
tial beyond-complementary metal-oxide-semiconductor tween these contacts, respectively, similar to a magnetic tun-
共CMOS兲 computing technologies,1 which are presently being nel junctions 共MTJs兲.7,11 Therefore, the current flow is con-
considered to supplement CMOS field effect transistors trolled by the gate and drain bias, and also by the direction of
共FETs兲 in microprocessors. Vigorous research in spintronic the contacts’ magnetization. The switching of magnetization
devices has been carried out over the last two decades2–4 and can be performed, for example, by spin transfer torque of the
has resulted in demonstration of two-terminal giant magne- flowing current. The spinFET must be distinguished from the
toresistance 共MR兲,5,6 tunneling MR 共TMR兲 共Ref. 7兲 devices, spin modulator based on spin precession, the original spin-
and TMR devices switched by spin transfer torque.8 Such tronic device proposed by Datta and Das.12 We will not con-
devices have been commercialized in magnetic hard drives sider this device here, though some later publications called
and magnetic random access memories and have a great im- it “spinFET.”
pact on everyday life. The question arises now whether there The MR ratio, which is a key device performance metric
is a spintronic device capable of similar success in logic of a spintronic device, is defined via the resistances for par-
applications. allel 共RP兲 and antiparallel 共RAP兲 contact magnetization con-
One of the candidates is the spinFET first proposed by figurations as follows MR= 共RAP − RP兲 / RP. The identical
Sugahara and Tanaka,9 a three-terminal device that utilizes quantity 共provided a fixed voltage is applied兲 is the magne-
ferromagnetic 共FM兲 contacts in the source and drain as spin tocurrent ratio MC= 共IP − IAP兲 / IAP. In order to improve MR,
injector and detector. The spinFET is essentially a combina- high spin polarization in both source and drain contacts is
tion of two Schottky barrier 共SB兲 metal-oxide-semiconductor favorable. Half-metal FMs 共HMFs兲 were predicted,13 and
field-effect transistors 共MOSFETs兲, each implemented by later on demonstrated by experiment,14,15 to have close to
carriers with a certain spin state 共e.g., one up-spin and one 100% spin polarization of electrons, which is desirable for
down-spin兲. The transport channels for up-spin and down- the contact FM material. With the ideal performance of spin-
spin electrons 共or holes兲 are independent if no spin scattering FETs, it is further shown in Ref. 16, that nonvolatile memory
occurs but they become interconnected if spin-flip processes and reconfigurable logic circuits can be constructed using
happen. The semiconductor channel makes the spinFET these devices. Despite the theoretically predicted perfect spin
compatible with the modern CMOS technology. Relatively polarization in the bulk HMF, there has been no observation
small spin orbit coupling and negligible hyperfine interaction of high spin polarized current injected from the HMF in
give electrons a long spin lifetime in silicon,10 which makes experiments.17 It is argued that when a HMF material con-
it a good candidate for the channel material. However, spin- tacts a nonmagnetic material, a randomization layer is
FETs are also envisioned with germanium or III-V semicon- formed at the interface18 where spins of localized electrons
ductor channels. Due to the exchange splitting between the are not aligned. This inevitable non-FM layer can decrease
up-spin and down-spin bands in the FM contacts, the up-spin the injected polarization and reduce the MR ratio.19
and down-spin carriers experience different barriers between Conduction mismatch between the FM and the semicon-
the contacts and the channel. The gate controls the width of ductor is another reason for the nonideal spin injection.20 The
solution was found in inserting a tunneling barrier between
a兲
Electronic mail: [email protected]. the FM and the semiconductor.21,22 Even though the tunnel-

0021-8979/2010/108共8兲/083702/10/$30.00 108, 083702-1 © 2010 American Institute of Physics

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
23.29.46.242 On: Wed, 07 May 2014 04:08:48
 083702-2 Gao et al. J. Appl. Phys. 108, 083702 共2010兲

ing barrier resistance decreases the current, a significant en- more details see the discussion in the previous studies.31,37
hancement of the injection efficiency is obtained. The third Then we apply the NEGF approach to the spinFET with spin
factor for nonideality of spinFET is spin scattering in the scattering. The key numerical model is described and the
channel. In the presence of spin scattering, the two conduct- connection between the mathematical description and the
ing channels 共up-spin and down-spin兲 are mixed, which has a physical model is discussed.
great impact on the carrier transport and the MR ratio. All
these unavoidable imperfections of spinFETs should be taken
into account when simulating the devices and assessing their A. NEGF method
performance potential.
An experimental prerequisite to building the spinFET is The channel material is described by a Hamiltonian ma-
not just electrical injection of spin polarization in a semicon- trix 关H兴 of size N ⫻ N blocks, N being the total number of
ductor but also electrical detection of spin polarization.23,24 grid points in the transport direction. Charging effects, which
Necessary conditions for efficient spin injection-detection are a result of the interaction between carriers and the chan-
and high MR have been determined theoretically.22,25 One nel potential is incorporated via the potential matrix 关U兴.
condition is low-resistance tunneling interface between the These serve as inputs in the equation for the retarded Green’s
FM and a semiconductor. Low-resistance interface to function at a specific value of energy E
silicon26 and germanium27 have been fabricated and charac-
G共E兲 = 关EI − H − U − ⌺共E兲兴−1 . 共1兲
terized. A spinFET has been demonstrated only recently;28 it
contained HMF electrodes and was switched by the spin The self-energy accounts for noncoherent processes and con-
torque effect. tains terms due to both contacts and due to scattering pro-
There have been a large number of theoretical and simu- cesses in the device
lation studies on spin injection from FMs into semiconduc-
tors, see review.29 Spin injection into semiconductors has ⌺共E兲 = ⌺L共E兲 + ⌺R共E兲 = ⌺S共E兲. 共2兲
been studied in a classical approximation, with drift- And similarly, the in- and out-scattering functions describing
diffusion type of equations.30 The nonequilibrium Green’s the change in populations of electrons and holes due to these
function 共NEGF兲 method31 is a rigorous quantum transport incoherent processes are in the following expressions:
treatment of nanoscale devices. First, the NEGF method has
been applied in the research of MTJ devices.32–35 A spinFET ⌺in/out共E兲 = ⌺Lin/out共E兲 + ⌺Rin/out共E兲 + ⌺in/out
S 共E兲. 共3兲
was treated by NEGF,19 where the transport in the channel The spectral function 关A兴, related to the local density of
was considered as ballistic with scattering only at the source/ states, and the electron/hole correlation functions 关Gn/p兴,
drain and channel interfaces. which are proportional to the occupation numbers of elec-
The present article reports the following advances com- trons and holes in states of certain energy, are given by
pared to prior work: 共1兲 simulation of spin-dependent quan-
tum transport in a FM-semiconductor-FM structure, includ- A共E兲 = i关G共E兲 − G†共E兲兴, 共4兲
ing tunneling barriers, 共2兲 rigorous treatment of spin
scattering, both in the channel and the randomization layer, Gn/p共E兲 = G共E兲⌺in/out共E兲G†共E兲. 共5兲
共3兲 identification of realistic performance limits 共especially
MR兲 for spinFET with relevant factors of nonideality, and 共4兲 These quantities are related to the local density of states, so
implementation within a well established quantum transport they also satisfy the equation
simulator, nanoMOS.36 A共E兲 = Gn共E兲 + G p共E兲. 共6兲
The rest of this paper is structured as follows. In Sec. II,
we summarize the NEGF formalism used to describe the The strength of coupling to the left 共source兲 and right 共drain兲
carrier transport in spinFETs and more specifically focus on contacts are described by the broadening matrices which are
the mathematical treatment of spin scattering and the physi- related to imaginary parts of the corresponding self-energy
cal connection with spin lifetime in various materials. In Sec. matrices
III, we apply this method to realistically structured spinFETs ⌫L/R共E兲 = i关⌺L/R共E兲 − ⌺L/R
†
共E兲兴. 共7兲
and quantitatively shows that the spin scattering affects the
I-V characteristics and can dramatically reduce the MR ratio. The in-scattering/out-scattering matrices represent the carrier
The physical reasoning is then presented along with rigorous injection and extraction rates into/out of the channel
simulation results and two solutions to enhance the MR ratio
⌺L/R
in
共E兲 = f L/R共E兲⌫L/R共E兲, 共8兲
are proposed and examined by numerical simulations. Fi-
nally, conclusions are drawn in Sec. IV.
⌺L/R
out
共E兲 = 关1 − f L/R共E兲兴⌫L/R共E兲,

II. NUMERICAL MODEL where f L/R共E兲 is the Fermi distribution functions in each con-
tact.
The NEGF formalism is ideally suited for analyzing Scattering, no matter if it is elastic or inelastic, can be
quantum transport of carriers in nanoscale devices. In this visualized as the coupling of the channel and a reservoir.31
section we first briefly restate the main equations of the The scattering process is physically described by the in-
NEGF method necessary for understanding the results. For scattering/out-scattering matrices, which show the rate of
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
23.29.46.242 On: Wed, 07 May 2014 04:08:48
 083702-3 Gao et al. J. Appl. Phys. 108, 083702 共2010兲

electrons coming into/out of a certain state. The sum of the

two matrices gives the broadening matrix due to the scatter-
ing process

⌫S共E兲 = ⌺in
S 共E兲 + ⌺S 共E兲,
out
共9兲

from which the scattering-related self-energy can be obtained

through a Hilbert transform as

⌺S共E兲 = P 冋冕 dE⬘⌫S共E⬘兲
2␲共E − E⬘兲
−i册⌫S共E兲
2
. 共10兲

The imaginary part of 关⌺S兴 obeys the same rule as that be- FIG. 1. 共Color online兲 The spinFET schematic. The source and drain are
tween 关⌫L/R兴 and 关⌺L/R兴. The real part of 关⌺S兴 is obtained via HMF. The magnetization of the drain can be switched to obtain the parallel
the Hilbert transform, where P stands for the principal value and antiparallel configurations of the two contacts. The double metal gates
control the channel electrostatics. The source contact injects and the drain
of a singular integral, see Ref. 38 for details. contact detects spin polarized current through oxide tunneling barriers. A
The NEGF and the Poisson equation are solved self- spin randomization layer exists at the boundary of the HMF.
consistently, because the carrier density is obtained from the
NEGF equations and used to solve for the electric potential,
B. SpinFET description
while the potential is necessary in order to solve the NEGF
equations. The current is calculated once consistency is The structure of the spinFET is illustrated in Fig. 1. Cur-
reached. This is the only loop necessary for the ballistic rent flows along the transport direction x. Along the z direc-
simulation 共i.e., with zero scattering terms兲. In the scattering tion are two metal gates separated by thin dielectric layers of
case, we have to consider an additional inner self- gate oxide above and below the channel, which provide good
consistency loop to calculate the in-scattering or out- electrostatics control. We have implemented the spin-
scattering matrix 关⌺in/out
S 兴 and the electron/hole correlation dependent transport simulation based on the widely used
function 关Gn/p兴 in the NEGF formalism. As described in simulator nanoMOS.36 The width of the device in the trans-
Refs. 31 and 37, the in-scattering/out-scattering energies verse direction y is assumed to be large enough, so that the
关⌺in/out
S 兴 contains 关Gn/p兴 as the inputs. They are used, in their states with various transverse momentum 共and corresponding
turn, to calculate the contact self-energy 关⌺S兴 through Eqs. energy Ey兲 can be analytically integrated, as it is imple-
共9兲 and 共10兲, and consequently to obtain 关G兴 from Eq. 共1兲. mented in nanoMOS, see Ref. 38 for details. Therefore, un-
Once self-consistency in the inner loop is achieved, the itera- less otherwise specified, energy E in the paper refers to the
tion in the outer loop of NEGF and Poisson equations starts. longitudinal energy due to motion along the x direction. In
One way to speed up the simulation is to bypass the compu- the example mostly used in this paper, the channel length is
tationally intensive Hilbert transform in Eq. 共9兲. This is pos- set at 12 nm, the channel thickness is 3 nm, and the thickness
sible for elastic scattering, where the in-scattering/out- of both top and bottom gate oxides is 1 nm.
scattering functions depend on the Green’s functions at the As the NEGF formalism is applied to spinFET, each el-
same value of energy only. In that case, the expression of the ement in the Hamiltonian 关H兴 is a 2 ⫻ 2 matrix, with the 共1,1兲
self-energy drastically simplifies, see Ref. 38. The spin scat- element representing the onsite energy of “up”-spin state and
tering considered here is elastic, and thus, admits such a the 共2,2兲 element of the “down”-spin state, relative to a cho-
simplification. Thus, the expressions for the scattering terms sen preferred axis. Therefore, the resulting size of the Hamil-
become tonian matrix 关H兴 is 2N ⫻ 2N. The same is true for the con-
tact self-energy 关⌺L/R兴, whose elements are all zeros except
⌺S共E兲 = D共E兲G共E兲, ⌺in共E兲 = D共E兲Gn共E兲, 共11兲 for the top-left and the low-right 2 ⫻ 2 blocks. The nonzero
elements in the contact self-energy describe the coupling of
⌺out共E兲 = D共E兲G p共E兲, ⌫共E兲 = D共E兲A共E兲, up-spin and down-spin carrier states in the source/drain and
the channel

冋 册
where we introduced the scattering tensor 关D兴. In this case, a
simpler self-consistency loop is performed to calculate the − teikL/R,ua 0
, 共14兲
Green’s functions at separate values of energy, which proves 0 − teikL/R,da
to be less time consuming.
At node i of the grid, total current 共Ii兲, and current for where kL/R,u/d is the momentum of the carrier in the source/
each energy level 关Ii共E兲兴 are given by the summation over drain in the up or down-spin state, and t is the amplitude of
spin states and the integral over energies coupling between the source/drain and the channel. We as-
sume that the magnetization of the contacts is along the same
ie preferred axis; otherwise a transformation matrix has to be
Ĩi共E兲 = 兺 关Hi,i+1Gi+1,i
n
共E兲 − Hi+1,iGi,i+1
n
共E兲兴, 共12兲
s ប introduced in the above equation.31 In the following analysis,
the up-spin is set as the majority spin states and down-spin is

冕 +⬁ the minority spin states in the source contact. If the magne-

dE
Ii = Ĩi共E兲. 共13兲 tization of the contacts is parallel, the drain contact shares
−⬁ 2␲ the same spin relation, whereas for the antiparallel case the
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
23.29.46.242 On: Wed, 07 May 2014 04:08:48
 083702-4 Gao et al. J. Appl. Phys. 108, 083702 共2010兲

drain contact has exactly the opposite relation between up- dependent, it is commonly used in ferro-MTJs to increase
down and majority-minority designation. their MR.4 Drift diffusion simulations21 predict that the tun-
There are two important parameters of the FM contacts: neling barrier with carefully adjusted resistance can increase
the spin splitting ⌬S and the majority spin bandwidth E␻. The the MR of a FM/semiconductor/FM stack as well. This effect
spin splitting ⌬S is the energy difference between the bottom exists for any tunneling barrier because different states
of the minority spin band and the bottom of the majority spin within a band align close to the top of the barrier of up-spin
band. The majority spin bandwidth E␻ is defined as the en- and down-spin bands. It is especially pronounced, however,
ergy difference between the Fermi level in the FM contact for certain tunneling barrier materials such as MgO.42 In that
EF,L/R and the bottom of the majority spin band. If ⌬S is case, up-spin and down-spin states over a certain range of
larger than E␻, the FM is called a HMF. The Fermi level energy belong to different bands with different crystal sym-
crosses just one spin band in such a material. Spin polariza- metries. Therefore they tunnel with drastically different
tion close to 100% is expected in the material. But it does not probabilities. As a result, MgO additionally provides very
necessarily translate into extremely high spin injection effi- efficient spin filtering effect and increases the spin polariza-
ciency in the semiconductor. In our example we set the ⌬S tion of the injected carriers. This effect can in principle be
= 2.5 eV and E␻ = 2 eV so that 100% spin injection can be modeled by setting different height of the barrier or by dif-
achieved without the interface scattering. These values are ferent mass of carriers in the barrier for up-spin and down-
close to the real case of the ⌬1 band in CoFeB, with the spin electrons. In this paper we set different transport effec-
exception that the presence of other bands leads to less than tive masses in the tunneling layer mox = 0.378m0
100% polarization. 共corresponding to MgO as per Ref. 43兲 and in the channel
We assume that the source and the drain are made of a mch = 0.19m0. The analytical integration over transverse 共ky兲
HMF. The effects of nonideal spin polarization of carriers are momenta is exact only in the case when the solution for the
accounted for by the spin randomization layer,17,18 as shown Green’s function is independent on the transverse momen-
in Fig. 1. It is a layer at the interface of a FM and semicon- tum, which is fulfilled for constant mass along the device.
ductor, where the spins of localized electrons are not aligned The integration is approximately valid for varying mass, if
with the direction of magnetization but have random direc- the current flows in the energy range 共⬍0.1 eV兲 which is
tions. The effect of the spin randomization layer is described smaller than the characteristic potential differences in the
as the first and the last block in the scattering self-energy device 共⬃1 eV兲. We also assume a constant effective mass
关⌺S兴. The rest of the diagonal blocks in 关⌺S兴 represent spin mch in the factor for the integration over the transverse mo-
scattering in the channel, with, in general, different rates of menta. The explicit form of the Green’s functions and self
scattering. This model would fairly well describe the effects energies with spin indices can be written as a set of diagonal
of the typical FM contacts with ⬃50% spin polarization and blocks for each grid point

冉 冊 冉 冊
even lower injection efficiency. The spin scattering in the
channel, which is the main physics we intend to study here, Guu Gud ⌺uu ⌺ud
G= , ⌺= . 共15兲
has the similar effects on the device performance regardless Gdu Gdd ⌺du ⌺dd
of the contact materials, but the more realistic scenario of
normal FM contacts without the spin randomization layers at The in-scattering/out-scattering functions implement the spin
the interface should be studied and will be the subject of scattering processes via the following relation to the
future work. electron/hole density 关Gn/p兴 and a scattering tensor 关D兴, see
The SB at the FM/semiconductor interface plays a vital Ref. 44
role in spintronic transport.26 It affects the spin injection and
S,ij 共E兲 = 兺 Dijkl共E兲Gkl 共E兲,
⌺in/out 共16兲
n/p n/p
detection, and controls the performance of the devices. The
SB is captured naturally in the NEGF formalism via the en- kl
ergy differences between the conduction bands in the semi-
conductor and in the metal. with 关D兴 being the fourth-order tensors in spin indices at
The SB height of the FM/semiconductor interface in the each grid point. The above equation can be qualitatively un-
conduction band is found to vary from 0.16 to 0.7 eV in S 兴兲 or out
derstood as the rate of electrons scattering into 共关⌺in
MnAs/Si, CoFe/Si, and CoFeB/Si.39 The SB height is very of 共关⌺out
S 兴兲 the state with energy E being proportional to the
small 共⬃0.02 eV兲 between CoFe and the valence band of existing electron 共关Gn兴兲 or hole 共关G p兴兲 density. We assume
Ge; and it varies with the insertion of a tunneling barrier.40,41 here the same functional form for electrons 关Dn兴 and holes
In our simulation it is an input value that can be set to model 关D p兴. The scattering tensor can be separated into the coupling
different material combinations. We fix it at a relatively low factor and the dimensionless tensor
value of 0.1 eV throughout the simulation and focus on the
effect of the channel spin scattering on the device perfor- D共E兲 = D共E兲⌽. 共17兲
mance.
A tunneling oxide layer may be formed between the For the case of isotropic scattering, the dimensionless tensor
source/drain and the channel. It is modeled as a potential is44

冉 冊 冉 冊
barrier of width W and energy height UH, and the effective
mass in the tunneling oxide is different from that in the chan- 1 0 0 −1
4⌽ij11 = , 4⌽ij12 = , 共18兲
nel. Since the tunneling barrier has a resistance that is spin- 0 2 0 0
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
23.29.46.242 On: Wed, 07 May 2014 04:08:48
 083702-5 Gao et al. J. Appl. Phys. 108, 083702 共2010兲

4⌽ij21 = 冉 冊
0
−1 0
0
, 4⌽ij22 = 冉 冊 2 0
0 1
and the equation for the self-energy turns to,

⌺s =
4
冉
D Guu + 2Gdd
− Gdu
− Gud
2Guu + Gdd
. 冊 共19兲

To understand the scattering coupling factor D, we now re-

late it to the commonly used spin-flip time T1 共or the scat-
4
1 兲, which is more familiar to experimentalist.
tering rate T−1
The broadening function ⌫共E兲 is proportional to the spectral
FIG. 2. 共Color online兲 IDS − VDS plots for 共a兲 parallel and 共b兲 antiparallel
function A共E兲 according to Eq. 共11兲. On the other hand the configurations in ballistic transport regime. The gate voltage values are 0.7,
broadening function is related to the spin-flip time 0.5, 0.3, and 0.1 V, from top to bottom curves.

⌫共E兲 = ប␥共E兲 = ប/共2T1兲. 共20兲

are up-spin. Absence of scattering will result in ballistic elec-
For two-dimensional gas of carriers, density of states, and tron transport,19 i.e., the current reaching the drain end is also
the spin-flip time do not change with energy. Since the den- 100% up-spin polarized without losing the spin and phase
sity of states is calculated per unit area, and the spectral coherence. The electrons see different potential barriers for
function is related to the unit cell of the grid, the relation up-spin and down-spin in the drain contact of different mag-
between them is netization configurations, and therefore, produce totally dif-
m 2A共E兲 ferent I-V characteristics as shown in Fig. 2.
g2D共E兲 = = . 共21兲 The up-spin channel is the only conducting channel in
␲ ប 2 2 ␲ a xa y
ballistic transport, and it has a high barrier potential in the
Thus, we can express the coupling factor via the spin-flip drain contact under the antiparallel configuration which
time as in Ref. 38 blocks the current flow and results in a very small drain
current, as can be seen in Fig. 2共b兲 when VDS ⬍ 0.4 V. The
ប3
D= , 共22兲 voltage 0.4 V is called the turn-on voltage VON here, which is
2T1maxay defined as the drain voltage required to push the minority
where ax and ay are the grid size in x and y directions and m spin band in the drain contact below the Fermi level of the
is the mass of carriers. The spin-flip time can be related21 to source contact in the antiparallel configuration. When VDS
the spin diffusion length in a nondegenerate semiconductor ⬎ VON, the up-spin band has states between the Fermi levels
with carrier density n and resistivity ␳ of source and drain; and the current will flow, as shown in

冑
Fig. 2共b兲 for VDS ⬎ 0.4 V.
T 1k BT The MR ratio plotted in Fig. 3 shows that with an ideal
Ls = . 共23兲
e 2n ␳ ballistic electron transport a high value of MR around 1000
can be obtained. The lower bound of VDS is chosen in Fig. 3
However, for a short channel device, the current is domi-
to ensure the large MR ratio as well as a reasonable drive
nated by quantum resistance rather than resistivity of the
current of the spinFETs.
channel. We will consider cases with widely varying rates of
spin scattering. Expected spin-flip times for electrons are of
B. Scattering transport
the order of ⬃0.1 ns in silicon and ⬃1 ps in germanium.
For holes, spin-flip times are comparable to momentum scat- Now let us consider the effects of spin scattering on the
tering times and can be as fast as ⬃1 fs. In the following, device performance. First, we introduce spin scattering in the
we will refer to carriers in the transistor as electrons. Our channel only and leave out the spin randomization layer and
model can in principle be applied to holes, however, then one
would need to include several valence bands in the semicon-
ductor.

III. RESULTS
A. Coherent transport
Let us first consider the case of no spin scattering. In the
spinFET studied in this paper, the source Fermi level lies
between the majority 共up-spin兲 and minority 共down-spin兲
spin bands, with the parameters spin splitting ⌬S = 2.4 and
majority spin bandwidth E␻ = 2.0, which agrees with theoret-
ical calculation in Ref. 18. The energy difference of 0.4 eV
FIG. 3. 共Color online兲 Magnetocurrent ratio of the ballistic spinFET under
between the Fermi level and the minority spin band is big different drain bias. The dots are the data obtained as VGS = 0.5 or 0.7 V, and
enough to ensure that almost 100% of the injected electrons a fitted curve is plotted to represent the average values of the discrete dots.
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
23.29.46.242 On: Wed, 07 May 2014 04:08:48
 083702-6 Gao et al. J. Appl. Phys. 108, 083702 共2010兲

FIG. 4. 共Color online兲 Energy-position resolved current in the channel for

共a兲 up-spin and 共b兲 down-spin in the parallel configuration. Up-spins convert FIG. 6. 共Color online兲 Energy-position resolved charge density in the chan-
to down-spins as the electrons traverse the device. nel in parallel configuration. The channel spin-flip coupling constants D and
injection efficiencies are: 1 ⫻ 10−5 eV2 共100 ps spin-flip time兲 and 90% in
共a兲 and 1 ⫻ 10−2 eV2 共0.1 ps spin-flip time兲 and 60% in 共b兲. The energy band
the tunneling barrier, as designated in Fig. 1. Suppose that edge for up-spin is shown. It is obvious that the energy band edges and
charge density are different in these two scenarios.
100% up-spin polarized electrons are injected into the chan-
nel. These electrons scatter for phonons or impurities with
some probability and flip to the down-spin state as they both interface and channel spin scattering. These two factors
travel along the channel. This scattering occurs everywhere significantly affect the charge distribution in the device be-
inside the channel, as shown in Fig. 4. The closer the elec- cause different current and contact polarizations result in dif-
tron is to the drain, the higher probability it has to turn into ferent injection and extraction rates for up-spin and down-
down-spin. The amount of down-spin current increases with spin electrons. Therefore, the process creates electrons
the increase in spin-flip coupling constant D. As can be seen pile-up in the channel. In another words, the charge self-
in the following analysis, this large amount of down-spin consistent calculation is necessary in order to simulate the
electrons produced in the channel will cause current leakage realistic performance of spinFETs with sensible values of the
into the drain and will degrade the device performance. channel spin-flip time and the contacts polarization. From the
Spin scattering acts as a cause of shortened lifetime of above simulations, we observe that the spin scattering can:
electrons in the channel. In other words, the local density of 共1兲 flip the spin polarizations and create down-spin current
state will spread out in real space and broaden in energy along the channel, 共2兲 broaden the local density of states, and
space as we increase the value D. This effect is observed in 共3兲 change the energy profile in the devices.
our simulation 共Fig. 5兲 for different D which corresponding In addition to the above effects, spin scattering also af-
to different spin-flip times. fects MR of the spinFETs. Figure 7 shows that with spin
It is also seen in Fig. 5 that the band edge profiles inside scattering in the channel, the drain current in the antiparallel
the channel are different for these three values of D. It can be configuration increases dramatically even below the turn-on
understood considering that electron distribution for both up- voltage 关compare it to Fig. 2共b兲兴.
spin and down-spin electrons depends strongly on the spin- Spin scattering induced leakage in the drain current can
flip time, and thus, the modified charge density generates greatly decrease the MR of the devices, as shown in Fig. 8
various potential energy profiles according to the Poisson for three different spin-flip coupling constants corresponding
equation. This dependence shows us the importance of a self- to spin-flip times of 1, 5, and 10 ps.
consistent solution of the NEGF and Poisson equation in the The peak in MR as a function of drain bias around
presence of the scattering in the channel. It is inaccurate to VDS = 0.3 V is the result of a faster growth of IP than that of
assume that the band profiles are the same with and without IAP at intermediate values of bias. The current in the parallel
spin scattering. The charge distribution will affect the energy case, IP, increases with VDS bias almost linearly, like in a
band and vice versa. The charge self-consistency can still be regular SB MOSFET 关Fig. 7共a兲兴. For the antiparallel case
observed for different values of spin polarization, as shown 关Fig. 7共b兲兴, current IAP at lower bias VDS, mainly consists of
in Fig. 6. The band profiles and the charge density 共both
up-spin and down-spin兲 in these two cases are influenced by

FIG. 5. 共Color online兲 Energy-position resolved local density of state in the

channel in parallel configuration. The spin-flip coupling constants D are FIG. 7. 共Color online兲 IDS − VDS plots for 共a兲 parallel and 共b兲 antiparallel
2.5⫻ 10−5 eV2 in 共a兲, 2.5⫻ 10−3 eV2 in 共b兲, and 1 eV2 in 共c兲, which corre- configurations in scattering transport regime. The gate voltage values are
sponding to 40 ps, 0.4 ps, and 1 fs spin-flip times, respectively. The strong 0.7, 0.5, 0.3, and 0.1 V, from top to bottom curves. The spin-flip coupling
coupling reduces the spin-flip time, and also broadens the available states in constant is 10−3 eV2, which corresponds to 1 ps spin-flip time in the
the channel. channel.
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
23.29.46.242 On: Wed, 07 May 2014 04:08:48
 083702-7 Gao et al. J. Appl. Phys. 108, 083702 共2010兲

FIG. 10. 共Color online兲 Energy-position resolved charge density of up-spin

关共a兲 and 共c兲兴 and down-spin 关共b兲 and 共d兲兴 current, for scattering 关共a兲 and 共b兲兴,
FIG. 8. 共Color online兲 Magnetocurrent ratio of spinFETs with spin scatter-
and ballistic 关共c兲 and 共d兲兴 transport regimes. In the scattering transport re-
ing under different drain bias and with different spin-flip coupling constants.
gimes, the up-spins turn to down-spins and escape to the drain, while no
The symbols are for simulation results at VGS = 0.5 V 共solid兲 or 0.7 V 共open兲
down-spins current flows in the ballistic case.
and fitted curves are plotted to represent the median values of the two cases.

the minority spin current that occurs due to the spin-flip scat- 10共b兲兴. They escape to the drain contact thanks to the low
tering. This is confirmed by the fact that such current in the barrier between the channel and the drain. The up-spin elec-
antiparallel state is not observed in the ballistic case 关Fig. trons remain confined in the channel as in Fig. 10共a兲. Note
2共b兲兴. At very small VDS共⬍0.1 V兲 it increases linearly, as that in the up-spin quantum well, the electrons occupy cer-
shown in Figs. 7共a兲 and 9共b兲, but then saturates due to the tain eigenstates of energy. One can notice the five lowest
limited amount of electrons experiencing scattering in the modes that contain from one to five antinodes of the wave
energy range where current flows. Further increasing the VDS function, respectively 关Figs. 10共a兲 and 10共c兲兴. The energy
to 0.4 V increases the energy range where current can flow, states are wider in the case shown in Fig. 10共a兲 than in Fig.
but due to quantum confinement in the channel, only an en- 10共c兲, because of the above mentioned spin-flip coupling
ergy shift in the minority current maxima and no noticeable constant values.
increase in the current magnitude are observed, as in Figs. The interface spin randomization layer can also have the
7共b兲 and 9共b兲. This causes a plateau in the function of current same effect as the channel scattering and be detrimental to
vs. bias. As VDS increases from 0.4 to 0.7 V, the energy range the MR ratio. It has been found in the previous work that the
opens for the majority current, and the total IAP increases interface treatment at the drain side is more pertinent to
dramatically thanks to the large amount of up-spin current, as achieving high MR ratio.19 With an estimate for the coupling
seen in Figs. 7共b兲 and 9共c兲. The peak of MR occurs at the constant D = 1 eV2 and a very high spin injection polariza-
plateau of IAP at VDS = 0.3 V in this case. tion ⬃80%, the MR ratio drops drastically compared to the
Figure 9 separately plots up-spin and down-spin currents ideal case without the spin randomization layers, as indicated
in the antiparallel configuration for both ballistic and scatter- in Fig. 11.
ing cases with the bias condition VGS = VDD = 0.7 V and The spin randomization layer, also called “magnetically
VDS = 0.2 V. Before the device is turned on 共VDS ⬍ VON dead layer,” does physically exist as an amorphous layer at
= 0.4 V兲 and without spin scattering, almost 100% up-spin the FM/semiconductor interface.45 A theory—experiment
electrons injected from the source are confined in the quan- coupled study46 also suggests that the spin current could po-
tum well formed by the channel and cannot escape into the larize the localized spins within the magnetically dead layer
drain 关Fig. 10共c兲兴. The negligible down-spin current flows and has a transient effect on the terminal current. Therefore,
freely from source to the drain but contributes very little to
overall current 关Fig. 10共d兲兴. When spin scattering is turned
on, a large amount of down-spin electrons is generated 关Fig.

FIG. 11. 共Color online兲 Magnetocurrent ratio of the spinFETs with spin
FIG. 9. 共Color online兲 Current-energy plot in antiparallel configuration for scattering and interface spin scattering under different drain bias and with
up-spin 共majority, solid line兲 and down-spin 共minority, dashed line兲 current different interface spin-flip coupling constants. The channel spin-flip cou-
at different VDS as VGS = 0.7 V. Very small amount of current increase is pling is 10−4 eV2 共10 ps spin-flip time兲. The symbols are for simulation
seen as VDS increase from 0.1 to 0.4 V 关共a兲 to 共b兲兴. A large amount of up-spin results at VGS = 0.5 V 共solid兲 or 0.7 V 共open兲 and fitted curves are plotted to
current flows as VDS rises past 0.3 V as in 共c兲. represent the median values of the two cases.
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
23.29.46.242 On: Wed, 07 May 2014 04:08:48
 083702-8 Gao et al. J. Appl. Phys. 108, 083702 共2010兲

FIG. 14. 共Color online兲 Energy-position resolved charge density of the 共a兲
up-spin and the 共b兲 down-spin current in the scattering transport at the
on-state in the antiparallel configuration. The high VDS pushes down the
drain energy band, which gives a large amount of up-spin current flowing
out of drain.
FIG. 12. 共Color online兲 Spin polarization along the channel with different
interface spin-flip coupling constant DINT and same channel scattering con-
stant DCHA 共solid line兲. The empirical parameter DINT adjusts the spin injec- up-spin electron transport is similar to that in a SB FET. The
tion efficiency, while DCHA controls the spin scattering along the channel. comparison of Figs. 2共b兲 and 7共b兲 stresses the need to de-
crease the current in the antiparallel configuration as the only
it is important to understand the physics of this layer and way to improve the MR ratio. The magnitudes of up-spin and
how it affects the device performance. As it is already known down-spin current in parallel and antiparallel configurations
that this layer will randomize the injection spin, in our simu- are plotted in Fig. 13.
lation this layer is modeled with an empirical parameter The subplots 共a兲 and 共b兲 verify the dominance of the
known as the interface spin-flip coupling constant DINT, up-spin current in the parallel configuration even at relatively
which describes the coupling strength of the electron spins high spin-flip coupling 共spin-flip time of ⬃1 ps兲. The anti-
and localized spins. This tunable parameter can be adjusted parallel up-spin current increases to almost 600 ␮A / ␮m in
to achieve different spin injection efficiencies. It is also as- the on-state as seen in Fig. 13共c兲, which can be explained
sumed in this work that the localized spins are invariable and with the help of the energy-position resolved charge density
always quickly return to equilibrium after scattering with plot in Fig. 14. In the on-state with VGS = VDS = VDD = 0.7 V,
electrons. Some experiments47 result in estimates of injection the high gate bias creates a thin SB between the source and
efficiency around 30%–50%, which was used to benchmark the channel, permitting a large amount of electrons to tunnel
the interface spin-flip coupling constant in our simulation. As through. The high drain bias ensures that the bottom of the
shown in Fig. 12, the interface scattering tensor DINT con- minority electron conduction band in the drain is below the
trols injection efficiency, and the channel scattering tensor source Fermi level; and, therefore, large current flows. Be-
DCHA affects the spin dephasing along the channel. There- low the turn-on voltage 共VDS ⬍ VON兲, however, down-spin
fore, the combination of these two parameters can be used to current due to spin scattering is much larger 关Fig. 13共d兲兴 than
model the realistic spin polarization scenario of spinFETs. the up-spin current limited by the quantum well confinement,
In order to improve the MR ratio, the current in the as seen in Fig. 10共b兲. Thus the up-spin current dominates the
parallel configuration should be maximized and the current total current in the on-state of the spinFETs, and the down-
in the antiparallel configuration should be minimized. In the spin current dominates in the off-state.
parallel configuration the down-spin channel is not conduc- To decrease the high antiparallel current, two solutions
tive with or without spin scattering, because the band edge are considered here. The first one is to reduce the up-spin
profile contains a high potential wall at the drain end. The current at VDS ⬎ VON by increasing the spin splitting ⌬S in the

FIG. 13. 共Color online兲 IDS − VDS, VGS plots of up-spin

关共a兲 and 共c兲兴 and down-spin 关共b兲 and 共d兲兴 current, for
parallel 关共a兲 and 共b兲兴, and antiparallel configuration 关共c兲
and 共d兲兴. The spin-flip coupling 共兩D兩 = 10−3 eV2兲 gives
large up-spin current at the on-state and large down-
spin current at medium VDS in the antiparallel
configuration.

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
23.29.46.242 On: Wed, 07 May 2014 04:08:48
 083702-9 Gao et al. J. Appl. Phys. 108, 083702 共2010兲

FIG. 15. 共Color online兲 Energy-position resolved charge density of the 共a兲
up-spin and the 共b兲 down-spin current in the scattering regime at the on-
state. The source and drain are antiparallel configured. The high VDS pushes
down the drain energy band, but the large spin splitting blocks the electrons
from going into the drain, which reduces the current at the on-state.

drain contact. The band diagram and charge density are plot- FIG. 17. 共Color online兲 Magnetocurrent ratio of the spinFETs with various
tunneling barriers configurations under different drain bias. The tunneling
ted in Fig. 15. The large ⌬S presents a high potential wall to
barriers in the source and drain ends are of 4 nm thick and 0.6 eV high.
electrons arriving at the drain and thereby blocks the current. There are three devices simulated here: without tunneling barriers for both
The simulation indicates that the up-spin current is reduced up- and down-spins 共dashed line兲, with the same tunneling barriers for both
from 560 to 0.6 ␮A / ␮m at the same bias conditions. This is spins 共dotted line兲, and with the different barriers for both spins 共solid line兲.
due to the fact that the quantum well confines the up-spin
electrons in the channel, increasing the probability of spin = VDD = 0.7 V, the spinFETs without the tunneling barriers
scattering into down-spin states. Therefore, more down-spin have a low MR of ⬃20. It can reach 80 with the insertion of
electrons are generated in the case of a larger ⌬S in the drain the same barriers 共UHD = UHU兲 for both up- and down-spins.
and the down-spin current increases from 7 to 57 ␮A / ␮m In the case that the tunneling barrier for the down-spin is
关Figs. 14共b兲 and 15共b兲兴. But the total current drops as a result higher than that for the up-spin electrons 共UHD ⬎ UHU兲, the
of the dramatic decrease in the up-spin current. MR can increase to ⬃500.
The second method is to reduce the down-spin current at
VDS ⬍ VON by inserting a tunneling barrier between channel IV. CONCLUSION
and drain. The high antiparallel leakage down-spin current at
VDS ⬍ VON induced by spin scattering is the main cause of In this work we have demonstrated a rigorous quantum
low MR ratio 关Fig. 16共b兲兴. The tunneling potential barrier transport 共NEGF兲 simulation of spinFETs taking into account
effectively blocks the current and diminishes the leakage, as spin scattering, tunneling and SBs, and self-consistent charge
shown in Fig. 16共a兲. We simulate a 4 nm thick spin- distribution. In the ideal case without channel scattering the
dependent tunneling barrier that exhibits a higher barrier device shows very large MR ratio of the order of 103. Spin
height for down-spin and a low barrier height for up-spin scattering generates a large amount of down-spin electrons,
electrons. In the parallel configuration, the up-spin domi- which increases the current in the antiparallel configuration,
nated current changes insignificantly, while the down-spin and eventually, degrades the MR ratio to around 10 with a
leakage current in antiparallel configuration is lower. The reasonable assumption for the spin-flip time in a semicon-
effect of the spin-dependent tunneling oxide is exhibited at ductor. The MR ratio becomes even lower when the inevi-
both source and drain ends. The barrier at the source end can table spin randomization layer at the FM/semiconductor in-
filter the injected current and increase its polarization, and terface is included. As a result of our numerical study, two
the barrier at the drain end can stop the leakage down-spin solutions are proposed to improve the performance of spin-
current below the turn-on voltage and almost eliminate the FETs. The first method is to increase the energy spin splitting
current in the antiparallel configuration. Thus the MR ratio is in the drain contact in order to create a high potential barrier
⬃50⫻ higher with the spin selective tunneling oxide than to block the drain leakage current, which mainly consists of
without it. the up-spin electrons coming from the source. Another solu-
The enhancement of MR ratio by adding the tunneling tion is to insert a spin-selective tunneling oxide layer be-
barriers can be seen in Fig. 17. In the on-state that VGS tween the source/drain and the channel, which brings the MR
ratio up to ⬃500.

ACKNOWLEDGMENTS
The authors gratefully acknowledge the support of the
Nanoelectronic Research Initiative through a supplement to
the Network for Computational Nanotechnology. We express
our appreciation for the helpful discussions with Supriyo
Datta of Purdue University, Brian Doyle and George Bouri-
FIG. 16. 共Color online兲 Energy-position resolved charge density of the anoff of Intel.
down-spin current in the scattering regime. The source and drain are anti-
1
parallel configured. The tunneling barrier for the down-spin electrons be- Semiconductor Industry Association, International Technology Roadmap
tween the channel and drain can lower the total current and, therefore, for Semiconductors, Chapter, “Emerging research devices,” 2009, http://
increase MR by about 50⫻. public.itrs.net/
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
23.29.46.242 On: Wed, 07 May 2014 04:08:48
 083702-10 Gao et al. J. Appl. Phys. 108, 083702 共2010兲

2
S. A. Wolf, D. D. Awschalom, R. A. Buhrman, J. M. Daughton, S. von Borghs, IEEE Trans. Electron Devices 54, 933 共2007兲.
26
Molnar, M. L. Roukes, A. Y. Chtchelkanova, and D. M. Treger, Science B.-C. Min, K. Motohashi, C. Lodder, and R. Jansen, Nature Mater. 5, 817
294, 1488 共2001兲. 共2006兲.
3
A. Fert, Rev. Mod. Phys. 80, 1517 共2008兲. 27
Y. Zhou, M. Ogawa, M. Bao, W. Han, R. K. Kawakami, and K. Wang,
4
I. Žutić, J. Fabian, and S. D. Sarma, Rev. Mod. Phys. 76, 323 共2004兲. Appl. Phys. Lett. 94, 242104 共2009兲.
5 28
M. N. Baibich, J. M. Broto, A. Fert, F. Nguyen Van Dau, F. Petroff, P. T. Marukame, T. Inokuchi, M. Ishikawa, H. Sugiyama, and Y. Saito, Tech.
Etienne, G. Creuzet, A. Friederich, and J. Chazelas, Phys. Rev. Lett. 61, Dig. - Int. Electron Devices Meet. 2009, 09-2.
2472 共1988兲. 29
A. M. Bratkovsky, Rep. Prog. Phys. 71, 026502 共2008兲.
6
G. Binasch, P. Grünberg, F. Saurenbach, and W. Zinn, Phys. Rev. B 39, 30
I. Zutić, J. Fabian, and S. C. Erwin, IBM J. Res. Dev. 50, 121 共2006兲.
4828 共1989兲. 31
S. Datta, Quantum Transport: Atom to Transistor 共Cambridge University
7
S. S. P. Parkin, C. Kaiser, A. Panchula, P. M. Rice, B. Hughes, M. Sament, Press, Cambridge, 2005兲.
and S. Yang, Nature Mater. 3, 862 共2004兲. 32
A. A. Yanik, G. Klimeck, and S. Datta, Phys. Rev. B 76, 045213 共2007兲.
8
E. B. Myers, D. C. Ralph, J. A. Katine, R. N. Louie, and R. A. Buhrman, 33
S. Salahuddin, D. Datta, P. Srivastava, and S. Datta, Tech. Dig. - Int.
Science 285, 867 共1999兲. Electron Devices Meet. 2007, 121.
9
S. Sugahara and M. Tanaka, Appl. Phys. Lett. 84, 2307 共2004兲. 34
10 X.-G. Zhang and W. H. Butler, Phys. Rev. B 70, 172407 共2004兲.
B. Huang, D. J. Monsma, and I. Appelbaum, Phys. Rev. Lett. 99, 177209 35
D. Waldron, L. Liu, and H. Guo, Nanotechnology 18, 424026 共2007兲.
共2007兲. 36
11 Z. Ren, S. Goasguen, A. Matsudaira, S. S. Ahmed, K. Cantley, and M.
S. S. P. Parkin, K. P. Roche, M. G. Samant, P. M. Rice, R. B. Beyers, and
Lundstrom, nanoMOS, 2006. Available online at: https://
R. E. Scheuerlein, J. Appl. Phys. 85, 5828 共1999兲.
12 www.nanohub.org/tools/nanomos/
S. Datta and B. Das, Appl. Phys. Lett. 56, 665 共1990兲. 37
13 M. P. Anantram, M. S. Lundstrom, and D. E. Nikonov, Proc. IEEE 96,
R. A. de Groot, F. M. Mueler, P. G. van Eugen, and K. H. J. Buschow,
1511 共2008兲.
Phys. Rev. Lett. 50, 2024 共1983兲. 38
14
R. B. Mancoff, B. M. Clemens, E. J. Singley, and D. N. Basov, Phys. Rev. D. E. Nikonov, G. I. Bourianoff, and P. A. Gargini, http://nanohub.org/
B 60, R12565 共1999兲. resources/7772
39
15
K. Inomata, N. Ikeda, N. Tezuka, R. Goto, S. Sugimoto, M. Wojcik, and E. K. Sugiura, R. Nakane, S. Sugahara, and M. Tanaka, Appl. Phys. Lett. 89,
Jedryka, Sci. Technol. Adv. Mater. 9, 014101 共2008兲. 072110 共2006兲.
40
16
M. Tanaka and S. Sugahara, IEEE Trans. Electron Devices 54, 961 共2007兲. D. Lee, S. Raghunathan, R. J. Wilson, D. E. Nikonov, K. Saraswat, and S.
17
C. T. Tanaka, J. Nowak, and J. S. Moodera, J. Appl. Phys. 86, 6239 X. Wang, Appl. Phys. Lett. 96, 052514 共2010兲.
41
共1999兲. Y. Zhou, W. Han, Y. Wang, F. Xiu, J. Zou, R. K. Kawakami, and K. L.
18
G. A. de Wijs and R. A. de Groot, Phys. Rev. B 64, 020402 共2001兲. Wang, Appl. Phys. Lett. 96, 102103 共2010兲.
42
19
T. Low, M. S. Lundstrom, and D. E. Nikonov, J. Appl. Phys. 104, 094511 W. H. Butler and A. Gupta, Nature Mater. 3, 845 共2004兲.
43
共2008兲. W. H. Butler, X.-G. Zhang, T. C. Schulthess, and M. J. MacLaren, Phys.
20
G. Schmidt, D. Ferrand, and L. W. Molenkamp, Phys. Rev. B 62, R4790 Rev. B 63, 054416 共2001兲.
44
共2000兲. S. Datta, Proceedings of the International School of Physics “Enrico
21
E. I. Rashba, Phys. Rev. B 62, R16267 共2000兲. Fermi” Course CLX edited by A. D’Amico, G. Balestrino, and A. Paoletti,
22
A. Fert and H. Jaffrès, Phys. Rev. B 64, 184420 共2001兲. 共IOS Press, Amsterdam and SIF Bologna, 2005兲, p. 244.
23 45
H. C. Koo, H. Yi, J.-B. Ko, J. Chang, S.-H. Han, D. Jung, S.-G. Huh, and J. S. Tsay, C. S. Yang, Y. Liou, and Y. D. Yao, J. Appl. Phys. 85, 4967
J. Eom, Appl. Phys. Lett. 90, 022101 共2007兲. 共1999兲.
24 46
X. Lou, C. Adelmann, S. A. Crooker, E. S. Garlidi, J. Zhang, K. S. D. Saha, L. Siddiqui, P. Bhattacharya, S. Datta, D. Basu, and M. Holub,
Madhukar Reddy, S. D. Flexner, C. J. Palmstrøm, and P. A. Crowell, Nat. Phys. Rev. Lett. 100, 196603 共2008兲.
47
Phys. 3, 197 共2007兲. B. Huang, L. Zhao, D. J. Monsma, and I. Appelbaum, Appl. Phys. Lett.
25
W. Van Roy, P. Van Dorpe, R. Vanheertum, P.-J. Vandormael, and G. 91, 052501 共2007兲.

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
23.29.46.242 On: Wed, 07 May 2014 04:08:48