IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 59, NO.

12, DECEMBER 2012 2801

Modeling and Design of Spintronic

Integrated Circuits
Sasikanth Manipatruni, Member, IEEE, Dmitri E. Nikonov, Senior Member, IEEE, and Ian A. Young, Fellow, IEEE

Abstract—We present a theoretical and a numerical formalism

for analysis and design of spintronic integrated circuits (SPINICs).
The formalism encompasses a generalized circuit theory for spin-
tronic integrated circuits based on nanomagnetic dynamics and
spin transport. We propose an extension to the modified nodal
analysis technique for the analysis of spin circuits based on the
recently developed spin conduction matrices. We demonstrate the
applicability of the framework using an example spin logic circuit
described using spin Netlists.
Index Terms—Circuit theory, logic, magnetoelectronics, spin po-
larized transport, spintronics.

I. INTRODUCTION

S PINTRONICS, the technology of control and manipu-

lation of the spin state of electrons and nanomagnets,
is one of the most promising approaches for beyond CMOS Fig. 1. Conceptual diagram of two nodes in a circuit connected by a conduc-
logic, memory, and analog applications [1]–[5]. Several spin tance branch. (a) Two nodes connected by a scalar conductance in a regular
based devices have been proposed [6]–[19] with the possibility circuit. b) Two nodes connected by a spin conductance in a spin circuit. (c) Con-
ceptual diagram of a spin current tensor when a spin current flows in a 3-D space.
of logic-nonvolatility, intrinsic directionality, higher logical (d) Spin current tensor is reduced to a spin current vector when a direction is
efficiency (large fan-in/fan-out) and reconfigurability. Com- implied by a branch of the circuit. The linearity of the circuit implies that the
bined with novel approaches for memory hierarchy [20], [21] connecting branch is described by a 4 4 spin conductance matrix.
and logic architecture [21]–[23], spintronics may enable high
performance, normally-off (with zero standby power), and
on the physical principles for spin transport developed over
instantly-on computing engines.
the past few years, first in collinear magneto-electronics for
In the past few years there has been tremendous progress in
spin valves [35] and then noncollinear magneto-electronics
spintronic devices and integration [2], [24]–[30] propelled by
[36]–[38] for spin transfer torque devices. In particular, we use
the advances in materials and fabrication techniques. In partic-
the formalism for 4 4 spin conduction matrices, introduced
ular, the advances towards in plane, three terminal and majority
in [18], which enable the combined analysis of nanomagnets
gate spin transfer torque devices [31]–[34] have opened the pos-
connected by spin transport channels. These were utilized for
sibility of spin logic devices which enable computation to be
modeling of spin logic devices in [19]. We describe the circuit
performed entirely in the magnetic and spin states of materials.
level transport models for spintronic devices from the physics
While the proposed devices show promising trends for non-
of the spin transport through nanomagnets and nanochannels.
volatile operation, low energy-delay products, and better logical
We generalize the Kirchoff’s conservation laws for spin circuits
efficiency, the suitability of the devices as components for large
to include spin dissipation [36]. We then extend the principles
scale integration remains to be shown. In particular, significant
of modified nodal analysis (MNA) [62] to spin circuits enabling
advances in spin logic device, circuit and system design are still
Netlist based SPICE simulations.
required in order to fully understand the suitability of spin de-
vices for general purpose computing.
The goal of this paper is to outline the principles for analysis II. CONCEPTS OF VECTOR SPIN CURRENT, SPIN-VOLTAGE,
of integrated spintronic circuits so that the physics of spin trans- AND SPIN CONDUCTION MATRICES
port can be utilized by SPICE developers and subsequently by
circuit and system designers for the exploration of spintronics To develop the formalism for spin conduction, let us consider
for “beyond CMOS computing.” We base the present work a branch of generic circuit consisting of two nodes , con-
nected by a conduction element [Fig. 1(a) and (b)].
Manuscript received December 13, 2011; revised February 29, 2012; ac- We now describe the concept of vector spin current (with
cepted May 30, 2012. Date of current version November 21, 2012. This paper the three Cartesian components specified by three scalar num-
was recommended by Associate Editor A. Sheikholeslami. bers), vector spin voltage (with three Cartesian components
The authors are with the Components Research, Intel Corp., Hillsboro, OR
specified by three scalar numbers), total current vector con-
97124 USA (e-mail: [email protected]).
Color versions of one or more of the figures in this paper are available online sisting of the coulomb current and vector spin current, total spin
at http://ieeexplore.ieee.org. voltage (consisting of the coulomb voltage and vector spin
Digital Object Identifier 10.1109/TCSI.2012.2206465 voltage) [18].

1549-8328/$31.00 © 2012 IEEE

2802 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 59, NO. 12, DECEMBER 2012

A. Node of a Spin Circuit

We formally define the node of a spin circuit as a collection
of physical points in a device or a circuit where all the quantities
of interest for spin and charge transport are at equilibrium [36].
The mechanisms driving the node to the state of equilibrium are
assumed to be much faster than the dynamics of the circuit.

B. Vector Spin Current

Vector spin current in a branch of a spin circuit, is the net
vector flow of magnetic moment along the branch of the circuit. Fig. 2. Accumulation of spin up electrons is expressed in a higher spin electro-
It has the units of amperes. In general, the spin current flowing in chemical potential. Diffusion current of up spin electrons flows from a higher
a three dimensional space is a tensor [37], [39]. The spin tensor to a lower spin electrochemical potential. Equivalently, spin current flows from
is described by a direction of the flow of the charges constituting higher spin voltages to lower spin voltages.
the spin current and the direction of the net magnetic moment
(spin) of the charges along each axis of the Cartesian coordinates
[Fig. 1(c) and (d)]. However, in a circuit, the direction of the to make the definition consistent to the usual electrochemical
flow of charges is defined by the connectivity of the branch and potential for electrons, which includes the term of voltage with
therefore, the spin current flowing between two points of a spin a negative sign. Overall, this choice of signs in the definitions
circuit is a vector. of spin current and spin voltage makes most of the relationships
similar to those between charge current and voltage as in Fig. 2.
(1)

The spin current can be also related to the velocity and spin D. Total Spin Current Vector
states of the carriers in a circuit/device. The components of the The total spin current is simply the combination of the charge
vector spin current are expressed as a sum over the momentum current and vector spin current. It is a 4 1 column vector.
states of electrons, normalized to density of electrons
(5)
(2.1)
The ratio between the charge current and the magnitude of the
spin current is the current’s spin polarization ratio
where are Pauli matrices, is cross sectional area, is the
velocity component normal to it, and is the spin density matrix.
It is defined with the negative sign to reflect the negative charge (6)
of electrons, similarly to the electric charge current

(2.2)
E. Total Spin Voltage Vector

This way, the spin current corresponds to the flux of magnetic The total spin voltage vector is the combination of the
moments. Thus, in Fig. 1(a), the electrical current (blue arrow scalar columbic potential and the vector spin potential. It is a
to the right) is opposite to the flux of electrons (red arrow to the 4 1 column vector.
left. If the net spin projection is positive, then the spin current,
blue arrow in Fig. 1(b), is opposite to the flux of electrons (red (7)
arrow).
The ratio between the scalar potential and the magnitude of the
C. Vector Spin Voltage spin potential is the voltage-spin polarization ratio of a node.
Vector spin voltage at a node
(8)
(3)

can be understood intuitively as the state variable associated

with the accumulation of spins of a certain direction. It is related F. Spin Conduction Matrix
to the half-difference in the electrochemical potentials of the The concept of spin conduction matrix can be derived by pos-
electrons with their spin up and down along the direction of this tulating the linear response of current to voltage. This assump-
vector. On the other hand, this difference can be related to the tion is expected to be valid in metals, where electron density
half-difference of density of the electrons with spin up and spin is high and electric fields change slowly compared to the scat-
down, , following the Valet-Fert theory [1], [35]: tering time. In the worst case, one has to take the branches of
the circuit small enough to make it a good approximation. The
(4) spin Ohm’s law, the linear relationship of spin voltage to spin
current is
where is the total density, is the electrochemical potential.
Spin voltage is defined with the negative sign as well. It is done (9)
MANIPATRUNI et al.: MODELING AND DESIGN OF SPINTRONIC INTEGRATED CIRCUITS 2803

where G is the 4 4 conductance matrix: is directly proportional and collinear to the vector spin voltage
difference applied to the NME, i.e.,

(13)
(10) (14)

where is a scalar quantity, is the applied spin vector

voltage across the NME, g is the scalar conductance, is the
G can also be interpreted as a 4 4 tensor that scales and reori-
applied voltage difference.
ents the voltage vectors to obtain the spin current vector. There-
fore, the spin conductance matrix of a conductance element is B. Spin Conduction Matrix for a Spin Elastic Series Branch
the matrix proportionality constant relating the vector spin cur-
rent through an element with the vector spin voltage difference Using the concept of spin conduction through nonmagnetic
applied across a conductance element. In general the 16 com- elements, we write the conductance of a series resistor as [18]
ponents of a spin conduction matrix are nonzero and are set by
the magnetic and geometric properties of the spin conductance
element. (15)

III. CONSERVATION LAWS FOR SPIN CIRCUITS

We now describe the extension of the Kirchoff’s current and where we described the matrix elements earlier. A resistor with
voltage laws to spin circuits [36]. Establishing the conservation no spin flip is described fully by a single element as shown in
laws is essential to setup a unique set of equations governing the (15). However, a series resistor with spin-flip needs to include
currents and voltages. a spin flip conductance to accommodate for the loss of spin
current.
A. Kirchhoff’s Voltage Law Extension for Spin Circuits
C. Spin Conduction Matrix for a Spin-Flip Conductance
The traditional voltage law for circuits is extended straight-
forwardly to spin circuits: since the sum of voltage differences We use the concept of spin-flip conductance to handle a non-
in any closed loop is zero spin-ballistic resistor. The conductance of a spin flip resistor is
[18]
(11)

(16)
where is the set of all node pairs in a given closed loop.

B. Kirchoff’s Current Law Extension for Spin Circuits

where is the spin flip conductance to accommodate the loss
Apart from the vectorial nature of spin currents and voltages, of spin polarization. The spin flip conductance sinks the spin
the difference between spin circuits and electric circuits is that current to a virtual spin ground to emulate spin current conser-
charge is strictly conserved, but spin is not. We handle the non- vation even in presence of spin flip events.
conservative nature of the spin currents entering a node by in-
troducing a spin dissipation current to a virtual ground [19]. The D. Spin Conduction Model ( -or T-Distributed Equivalent)
traditional Kirchhoff’s current law is thus extended to spin cir- for Distributed Channel With Spin Flip
cuits as follows: the sum of the vector spin currents entering
node is equal to the total dissipated vector spin current at the The spin conduction model for distributed channels in
node. At node , the spin node current law is given by (12), -equivalent and T-equivalent networks can be expressed
using the shunt and series conductances described above. Let
where is the spin current from node to node ; is the
us consider a nonmagnetic channel connecting two nodes of
set of all nodes connected to node ; is the total spin current
a spin circuit and , Fig. 4(a). The process of spin flip
dissipated due to spin flip events happening at the node .
which causes a loss of spin current from the channel is modeled
(12) using shunt resistances which go to the spin voltage ground
, Fig. 4(b). No charge current flows through
the shunt elements since the conductance components are
zero.
IV. 4-COMPONENT SPIN CONDUCTION MATRICES FOR The equivalent conductances in the -equivalent circuit
NONMAGNETIC CHANNELS, SERIES, AND PARALLEL ELEMENTS for a nonballistic elastic channel with cross section area of
the channel , channel resistivity , the channel length
We describe spin conduction matrices for nonmagnetic ele- , spin-flip length of the channel material is shown in
ments (NME). Appendix A.
Depending on the ground conditions, a T-model [Fig. 4(c)]
A. Spin Conduction Through Nonmagnetic Elements may be more convenient. The conversion from -equivalent to
The spin conduction through a linear nonmagnetic conduc- a T-equivalent is as follows (see Appendix B for details):
tive element can be described as follows: a) the charge current
through the device is directly proportional to the applied scalar (17)
voltage difference; b) the spin vector current though the element (18)
2804 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 59, NO. 12, DECEMBER 2012

Fig. 5. Circuit model for spin transport between a ferromagnet (FM) and a
normal metal (NM). The entire FM is treated as a node with a specific vector
spin voltage.
Fig. 3. Conservation laws for spin voltages and currents. (a) The sum of loop
voltage differences is zero. (b) The sum of the physical spin currents from all
physical branches is equal to the spin flip current to a virtual ground.
4 1 spin voltages at the FM and NM be and
respectively. Here for simplicity we ignore
the spin accumulation in FM. Let be the 4 1
spin current from FM to NM and be the vector direction of the
magnet’s magnetic moment. Then, according to [38], the charge
current is

(19)

The total spin current is given by

Fig. 4. and T Equivalent circuits for a normal metal distributed channel.
(a) Normal metal channel connecting nodes and (b) equivalent circuit
representing the distributed channel (c) T equivalent circuit representing the
(20)
distributed channel. (21)
(22)
V. 4-COMPONENT SPIN CONDUCTION MATRICES FOR
MAGNETIC ELEMENTS
where is the component of the spin current parallel to the
magnetic moment and is the spin current perpendicular to
A. Spin Conduction Through Magnetic Elements the magnetic moment. The expressions for spin torque conduc-
tances , and , which are related to the spin reflection
In contrast to spin conduction through nonmagnetic elements,
and transmission properties of the interface, are described in
spin conduction through magnetic elements in general can have:
Appendix C.
a) coupling between scalar voltages and spin currents and b)
The spin conduction matrix elements can also be deduced
the spin current can be noncollinear to the vector spin voltage
from experimental properties of the FM-NM interface. G is the
difference.
value of the total conductance of the interface, is the spin se-
We next describe the conduction from a ferromagnet to a
lectivity of the FM-NM interface. is the spin transfer con-
normal metal.
ductance of the normal metal. The Sharvin conductance is the
quantum limit of spin transfer conductance. is the field-like
B. Spin Conduction at Ferromagnet and Normal Metal conductance term which is typically zero for many metal inter-
Interface faces. It is encountered at higher voltages in tunneling barriers
adjacent to ferromagnets.
Spin conduction from a ferromagnet to a normal metal can
be understood as the spin dependent current in response to spin
voltages. The spin voltage at a node is in turn a result of a spin C. Spin Conduction Matrix of a Fixed Nanomagnet
polarized population set up via spin injection from elsewhere
[35]. Microscopically, the conduction is happening via spin de- We derive the conduction matrix in a special case of the mag-
pendent reflection and transmission at the interface of the fer- netization parallel to the plane of the interface, as shown in
romagnet and the normal metal [40], [41]. Much of the for- Fig. 5. We choose the coordinate system such that the x-axis is
malism is derived from quantum transport scattering theory, see along the direction of magnetization and the other two
e.g., [42], and draws on the work from superconductive trans- axes form a right-handed coordinates. Then using projections to
port [43]. We refer the reader to [37] for a detailed physical these coordinates, (19)–(22) become
explanation.
We first derive the 4 4 conduction matrix of a ferromagnet
(23)
(FM) to normal metal (NM) interface (Fig. 5) from the spin
conduction equations [38]. The elements of the spin conduc- (24)
tion matrix can be filled with experimental properties. Let the (25)
MANIPATRUNI et al.: MODELING AND DESIGN OF SPINTRONIC INTEGRATED CIRCUITS 2805

TABLE III
NANOMAGNET PARAMETERS FOR SPIN CIRCUIT THEORY USED IN
COMBINATION WITH A MACROSPIN NANOMAGNET MODEL

Fig. 6. Circuit model for spin transport between a ferromagnet (FM) and a
normal metal (NM). The FM magnetization can point in any direction in three
dimensions as determined by the nanomagnet dynamics.

TABLE I
LIST OF VARIABLES FOR SPIN CIRCUIT THEORY

D. Spin Conduction Matrix of a Free Layer Nanomagnet

Now we need the expression of the spin conduction matrix
of the FM-NM interface in the fixed coordinate system tied to
the nanomagnet shape (e.g., elliptical, see Fig. 6) rather than
the instantaneous direction of magnetization. This derivation is
given in Appendix D. The 4 4 conduction matrix in the fixed
coordinate system given by

(28)

where R is the rotation matrix

TABLE II (29)
TRANSPORT PARAMETERS USED IN SPIN CIRCUIT THEORY

The elements of the rotation matrix R are defined by the ex-

pression of the unit vectors of the axes (X, Y, Z) tied to the mag-
netization in terms of the fixed coordinates (x,y,z)

(30)

(31)

(32)
Hence, the generalized Ohm’s law for the FM-NM interface is
[18]
VI. SELF-CONSISTENCY OF NANOMAGNET DYNAMICS
WITH SPIN CIRCUIT ANALYSIS
We now describe a coupled spin transport-magnetization dy-
namics model [76] for solving spin integrated circuits which
employ nanomagnets for spin injection. The phenomenological
(26) equation describing the dynamics of nanomagnet with a mag-
Since (19)–(22) are independent of the orientation of the inter- netic moment unit vector , the modified Landau-Lifshitz-
face, we can extend by induction that the same expression for Gilbert (LLG) equation [49], [50], with spin transfer torques in
the conduction matrix is valid for any direction of magnetiza- the form of [38] is (see Table III for parameters)
tion provided that the coordinate system has its x-axis aligned
to it, (33)

where is the electron gyromagnetic ratio; is the free space

(27)
permeability; is the effective magnetic field due to mate-
rial/geometric/surface anisotropy; is the Gilbert damping of
2806 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 59, NO. 12, DECEMBER 2012

Fig. 7. Self-consistency of nanomagnet dynamics with spin transport. (a) Ex-

ample circuit demonstrating the need for self-consistency. (b) Self-consistency
between LLG dynamics and spin transport.

the material and is the component of vector spin current per-

pendicular to the magnetization entering the nanomagnet, Fig. 8. (a) Constitution of the MNA matrix equation for normal circuits
(b) Constitution of the Spin MNA matrix equation for spin circuits.
is the total number of Bohr magnetons per magnet. can
also be rewritten as .
The positive sign of the spin torque term is related to the fact the voltages of the voltage sources and currents at the current
(Section II-B) that the spin current is has the same direction as sources. can be also be written as
the flux of magnetic moments. Implicit in the LLG equation is
the fact that absolute values of the magnetic moments of single (35)
domain nanomagnets remain constant. The noise properties of
nanomagnets play a critical role in the dynamics of the magnets where matrix has the size (n is the
[51]–[54]. See Appendix F for a description of the noise prop- number of nodes, and m is the number of independent voltage
erties and numerical methods for stochastic LLG equations. sources). Matrix has the size and is determined by
In general, the direction of the nanomagnet magnetic mo- the interconnections between the passive circuit elements. Ma-
ments of a spin circuit and the spin transport via a spin circuit trix has the size and is determined by the connection
are coupled together. The spin current entering a nanomagnet of the voltage sources. Matrix has the size and is
is defined by the conductance of the nanomagnet at the present determined by the connection of the voltage sources. ( and
angular position. This is because the equivalent conductance of are closely related, particularly when only independent sources
the nanomagnet is determined by the direction the nanomagnet’s are considered). Matrix D is and is zero if only inde-
moment. Consider an example spin circuit shown in Fig. 7(a). pendent sources are considered.
The current passing through the circuit depends on the direction We note that the spin currents entering the magnets can be
of the magnet while the direction of the magnet is modified extracted as the Cartesian components of the total spin current.
depending on the injected spin current. Hence at each instant of For a magnet connected between i and j nodes, the spin cur-
time a self-consistent solution needs to be calculated to ensure rent entering i node
accuracy. Fig. 7(b) shows the self-consistent loop between LLG
nanomagnet dynamics and spin transport [76]. The LLG solvers (36)
pass the condition of the magnets to the spin circuit and the spin
circuit solver passes the spin vector current to the LLG solver For a description of the rules/algorithm for writing matrix
at each pass of the self-consistent loop till a solution is reached. please see Appendix H [63]. Matrix can be assembled as
Self-consistency can also be addressed by using an implicit nu- shown in Fig. 8. The proposed method can handle a combination
merical solver [55]. of nonmagnetic and magnetic elements as well as dependent
and independent spin/regular voltage and current sources. For a
VII. MODIFIED NODAL ANALYSIS FOR SPIN CIRCUITS detailed description on the regular MNA algorithm and depen-
We now extend the modified nodal analysis (MNA) to spin dent sources please see [64], [65]. The solution of the spin MNA
circuits in order to provide a scalable way to analyze multi-node equation can be simplified by optimal ordering of the equations
systems. A computational method for solving spin circuits is to obtain sparsity [66] among other optimization techniques. For
essential even for few node circuits since the an algorithm for parsing a netlist to do MNA, see for example
ordering of the spin conduction matrices is crucial while cal- [65].
culating equivalent conductances (see Appendix G). Following
closely the formalism for MNA [62], the spin-MNA solves the VIII. SIMULATION OF AN EXAMPLE SPIN CIRCUIT
following equation:
We now describe an example spin circuit with embedded
(34) nanomagnets to demonstrate the effectiveness of the proposed
framework. Let us consider a nonlocal spin injection detection
where is a matrix formed based on the connectivity of the device shown in Fig. 9 [19], [31], [34]. We consider a lateral spin
circuit, location of the voltage and current sources; is the injection-detection device which has been proposed recently as
vector comprising unknown node voltages and unknown cur- an all spin logic device. The device consists of two nanomagnets
rents through the voltage sources; is the vector comprising of communicating via a nonmagnetic channel. The device operates
MANIPATRUNI et al.: MODELING AND DESIGN OF SPINTRONIC INTEGRATED CIRCUITS 2807

TABLE IV
SPIN MNA MATRICES

TABLE V
PARAMETERS USED FOR EXAMPLE CIRCUIT SIMULATION

Fig. 9. A lateral spin logic device comprising two nanomagnets and nonmag-
netic channels. Channel connecting 1–2 acts as an interconnect between the two
magnets transporting spin polarized currents. (a) Top view of spin logic device.
(b) Side view of a lateral spin logic device. (c) Netlist of the circuit that can be
parsed by a spin-MNA algorithm. (d) Circuit model of spin logic device.
A. Example Spin Circuit: Numbering the Nodes, Forming
Spin Netlist
We model the device as a spin circuit comprising of two nano-
as an inverting gate for positive applied voltages and a nonin- magnets and nonmagnetic conductive elements. The nonmag-
verting gate for negative applied voltages. netic elements model the behavior of the metal channels con-
Intuitively, the operation of the device can be explained as fol- necting the magnets to each other and to the ground. In Fig. 9,
lows: the magnets create spin polarized population densities un- we show the top view and side view of the device. We choose
derneath the magnets and setup spin diffusion currents through the node-0 to be the ground and number the remaining nodes as
the channel. The direction of this spin diffusion current is set per the convention of MNA. Node 1, 2 represent the points in
by the relative strength of the spin polarization of the carriers. the device just below the magnets representing the ends of the
For a ground terminal set near the input magnet it can be shown channel. Node 3 is common node shared by the magnets and the
that the magnet 1 acts as a fixed magnetic terminal, while the supply. We can now derive the circuit diagram for the device
second magnet responds to the spin diffused to beneath it, de- as shown in Fig. 9(d). The magnetic elements are represented
pending on the applied voltages. For a positive applied voltage, by ; the nonmagnetic channel is represented by a
the device shown in Fig. 9 acts like an inverting gate, where the -equivalent circuit as described in Section IV-D. The ground
output becomes a logical invert of the input. For negative ap- connection branch is represented by a T equivalent circuit.
plied voltages, the output becomes a copy of the input magnet’s The assumed dimensions and the list of variables are shown in
condition. The sectioned structure of the channel is required Table V. We show the netlist for the device in Fig. 9(c).
to isolate spin logic gates, where the interconnection between
gates (concatenation) is achieved via a continuous free layer B. Example Spin Circuit: Forming the MNA Equation
magnet. The nonreciprocity of (output to input signal transport)
spin logic comes from an asymmetry between input and output Using the rules described in Appendix H, we can build the
magnets. This asymmetry can be achieved via: a) an asymmetric spin-MNA equation for the circuit in Fig. 9(d) as follows: the G
overlap of the magnet such that the area of the output is smaller matrix is filled with the spin conductivity elements connecting
than the area of input magnet; b) asymmetric ground condition; to nodes 1–4. The elements representing connectivity (5,3) is
c) asymmetric spin injection efficiency (Fig. 9); d) asymmetric filled with an identity matrix such that the applied voltage at
spin damping constant. node 3 is . The row 3 represents the KCL at the node 3 and
2808 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 59, NO. 12, DECEMBER 2012

correspondingly has an identity matrix at element (3,5). See

(37) at the bottom of the page.
C. Example Spin Circuit: Self-Consistent Solution of
Spin-MNA With LLG Equations
The self-consistent solution is obtained by solving all the
nanomagnet dynamical equations consistently with the trans-
port equations (Fig. 8). The full set of equations for this example
system then becomes

(38)

(39)

Fig. 10. (a) Transient self-consistent simulation of a spin circuit device.

where the voltages are derived from (37), at the bottom of the (b) Trajectory of the magnetic moment of the nanomagnets.
page.
We numerically simulated the spin logic circuit using self-
consistent solution of the spin-MNA equation with stochastic
LLG equation. The specific parameters used for this example
are shown in Table V for reproducing the results. We note that
the stochastic nature of the LLG equations produces an inherent
variability in operation of spin devices.
We show the basic operation of the device in Fig. 10. In
Fig. 10(a) we show the dynamics of the magnetic moment of
the nanomagents for various applied voltages. During the time
from 0 ns to 2 ns, a positive voltage is applied at node 3, the
device acts as an inverting gate during this interval. The de-
vice responds over a time of and the output flips to
a state opposite of the input magnet. At 2 ns, the supply voltage
is flipped to , the device acts as a noninverting gate in
this interval and the output responds at and flips to a
state parallel to the input magnet. The trajectories of the mag-
nets’ magnetic moments are shown in Fig. 10(b).
We now extract the spin and charge currents via the cir-
cuit using the branch conductances. The spin currents along
the nanomagnet easy axis are show in Fig. 11(a). The total
energy dissipation of the device can now be calculated as
the total charge current sourced from the supply times the
supply voltage. The total electrical power of the device can be Fig. 11. (a) Spin current via input and output magnets. (b) The instantaneous
extracted as shown in Fig. 11(b). power through the spin logic device.
This circuit modeling described here shows that the spin logic
circuit, using Table V parameters, operates with transient en- year retention time will outperform CMOS logic in idle power
ergy-delay metrics of with up to a 2 GHz response by several orders of magnitude. We note the leakage power of
and with zero leakage power (ignoring the overheads). Com- the MPU units have approached 50% over the past few years
pared to a CMOS technology with a 60 nW/transistor idle power [75]. For transient energy-delay improvement, a physically real-
[74], nonvolatile spin logic with state hold power and 7 istic simulation framework, along with improved spin-electrical

(37)
MANIPATRUNI et al.: MODELING AND DESIGN OF SPINTRONIC INTEGRATED CIRCUITS 2809

transduction methods, materials, and anisotropy engineering, (A10)

can address this gap in developing device and circuit topolo-
gies to approach the fundamental performance of the spintronic where we designate ;
devices and circuits for the beyond CMOS era. From these equations, and for the cross-sectional area of
the conductor
IX. CONCLUSION
In summary, we describe a spin circuit framework combining (A11)
spin transport with generalized modified node analysis to en-
able SPICE for spintronic circuit analysis. The framework han- According to the above notation, the following relations must
dles both magnetic and nonmagnetic components with the com- be satisfied for the -network
monly used MNA methodology. This framework will enable
(A12)
study of spintronics for logic, interconnect, memory, and hybrid
integration of spin devices with CMOS circuits. The ability to (A13)
synthesize and analyze spintronic CMOS integrated circuits will (A14)
enable and accelerate the study of spintronics with the potential (A15)
attributes of nonvolatility, superior energy-delay, higher logical
efficiency, recofigurability, and suitability for novel computa- (A16)
tional architectures and logic-memory paradigms. for all values of the boundary conditions (specified by a, b).
Taking a special case of , , we arrive at the equations
APPENDIX A
G-MATRIX ELEMENTS OF A NONMAGNETIC (A17)
DISTRIBUTED CHANNEL
Spin-dependence conduction can be described in a contin- (A18)
uous medium (as opposed to a lumped-element circuit) via the
drift-diffusion equations [35]. In the following, we provide the
derivation of normal magnet (NM) spin conductance matrix in- which have the solution
troduced in [18]. We show the spin conductances for currents
along one direction (x) and one direction (s) of spin, and then (A19)
generalize it to arbitrary directions of spins. Then the drift-dif-
fusion equations for the current density J, spin current density (A20)
, voltage V, and spin voltage in a nonmagnetic material are
Hence, the conductance for the series branch of the network
(A1) is [18]

(A2)

Current continuity implies (Kirchhoff current law)

(A3)
(A21)
(A4)
The conductance for the parallel branch of the network is
where the conductivity is , and the spin diffusion length is
. A general solution for these equations is

(A22)
(A5)
APPENDIX B
where the boundary conditions set the coefficients , . The spe- TO T EQUIVALENT CIRCUIT TRANSFORMATION
cific solution for the uniform conductor of length is
Spin-dependent conductance with spin relaxation can be
equivalently represented by -shaped [Fig. 4(b)] or T-shaped
(A6)
[Fig. 4(c)] networks. Here we derive a general relation between
(A7) these two approaches.
In both cases the vector voltages at the terminals, V1 and V2
are the same. In the -network, the spin-relaxation currents are
(A8)
(B1)
(A9) (B2)
2810 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 59, NO. 12, DECEMBER 2012

and the current passing through the center conductance is conduction matrix elements of a FM as phenomenological
constants extracted from experimental properties of the
(B3) FM-NM conduction. Here, we provide a description of the
conduction matrix elements of a FM that is derived from an ab
They are related to the total currents entering and leaving the initio approach [18], [38]. The conduction matrix
network as follows:
(B4)
(C1)
(B5)

Using these relations in (C4), (C5) is convenient to express the

sum and the difference of the in- and out-currents. can also be written in terms of spin scattering conductance ele-
ments as
(B6)
(B7)

Similarly we obtain for the T-network; the current from the

middle node to the ground is related to the vector voltage at this
node (C2)
where , and are the matrix elements derived from
(B8) spin scattering at the FM-NM interface. The conduction matrix
elements of an FM-NM interface can be described in terms of
and the in- and out-currents are the reflection and transmission properties of the spin up and spin
down electrons incident from an NM to FM [38].
(B9)
(B10) (C3)
The current conservation results in
(C4)
(B11)

The voltage at the middle node is thus related (for nonzero spin
relaxation) to the in- and out-currents (C5)

(B12)
where is the conductance per spin of a ballistic channel
As before, we express via the unity matrix with ideal contacts [38]; , are the transmission coeffi-
cients for up and down spin electrons from NM to FM; ,
are the reflection coefficients of the up and down spin elec-
(B13) trons at the FM-NM interface; is the number of modes in the
NM, is the number of modes in the FM. The number of modes
(B14)
in a metal can in-turn be written from the metal’s Fermi
Since the two expressions for in- and out-currents should be wave vector [48].
equivalent, the following relations between nonzero conduc- It has been argued that , are close to zero for many
tances must hold: material systems [38], which simplifies the spin torque conduc-
tance to to
(B15)
(B16) (C6)

Simple algebraic manipulations permit the expression of the

T-conductances in terms on -conductances APPENDIX D
G-MATRIX OF FREE FM LAYER
(B17)
DERIVATION FOR FREE MAGNETIC LAYER CONDUCTION
Note that this derivation is not trivially reduced to the results of MATRIX: Here we derive the G-Matrix for a free layer FM with
the traditional electronic network theory, because the conduc- an arbitrary magnetic moment direction (see Fig. 7). Let
tances are matrices rather than scalars. be the angle of the magnetic moment with z-axis and be the
angle of the projection of , with x-axis. We can write in
APPENDIX C co-ordinate system xyz as
G-MATRIX ELEMENTS OF FM EXPRESSED AS SPIN REFLECTION
AND TRANSMISSION COEFFICIENTS (D1)
4 4 CONDUCTION MATRIX ELEMENTS OF A Let us choose a new coordinate system XYZ such that the is
FERROMAGNETIC METAL IN CONTACT WITH A NORMAL collinear with the new X-axis. (Refer to Section V)
METAL EXPRESSED IN TERMS OF SPIN REFLECTION AND
TRANSMISSION COEFFICIENTS: Earlier, we described the (D2)
MANIPATRUNI et al.: MODELING AND DESIGN OF SPINTRONIC INTEGRATED CIRCUITS 2811

APPENDIX F
(D3) STOCHASTIC LLG EQUATIONS
(D4) THERMAL NOISE OF NANOMAGNETS: The dynamics
of nanomagnets are strongly affected by the thermal noise.
In the new coordinate system
Thermal noise in a nanomagnet manifests as fluctuations to the
(D5) internal anisotropic field [51]–[54]. The thermal noise can be
considered as a result of the microscopic degrees of freedom
where is the matrix described in Section V. Let us substitute of the conduction electrons and the lattice of the ferromagnetic
element [51].
(D6) At room temperature T, the thermal noise is described by a
(D7) Gaussian white noise (with a time domain Dirac-delta auto-cor-
relation). The noise field acts isotropically on the magnet. In
and rearrange to obtain the current, voltage relation in the xyz presence of the noise, the LLG equation can be written as
co-ordinate system. We obtain

(D8)
(F1)
Hence, the conductance matrix for an FM with magnetic mo-
ment along an arbitrary direction is given by where we modified (33) by adding temperature dependence. The
internal field is described as
(D9)

(F2)
APPENDIX E (F3)
CONVERSION FROM SPINOR BASIS TO VECTOR BASIS
(F4)
Conversion Between Spinor Spin Current/Voltage Basis to
4 Component Vector Current/Voltage Basis: The derivations
for magneto-electronic circuit theory are often performed in the The initial conditions of the magnets should also be random-
spinor basis for the electrons [39]. For convenience we list the ized to be consistent with the distribution of initial angles of
conversion from spinor basis to Cartesian vector basis as well magnet moments in a large collection of magnets. At tempera-
as 4-component current basis. The current in a spinor basis can
ture T, the initial angle of the magnets follows [52]
be written as [39]

(F5)
(E1)

where is the Pauli spin matrix NUMERICAL METHODS FOR STOCHASTIC LLG EQUATIONS:
An accurate choice of the method for integration of the sto-
(E2) chastic LLG equation is essential since: a) the stochastic differ-
ential equations (SDE) require careful handling of the order of
integration [67], [68]; b) a multiplicative white noise requires
which yields
an appropriate choice of calculus [53], [69]. The appropriate
model for direct integration of SDE are usually first order in-
(E3) tegration methods such as Euler & Heun. Even though higher
order methods have been proposed in the literature, the accu-
Hence, the 4-component current vector can be derived from the racy and applicability for realistic SDEs have been questioned
spinor current as follows: [67], [68]. Hence, a first order integration method with a fixed
time step is generally recommended [53].
Secondly, Stratonovich calculus is used for interpreting the
multiplicative white noise. We used a mid-point integration
(E4) method [53] to apply the Stratonovich calculus while inte-
grating the LLG equation. The discretized integration rule is

Similarly, 4-component vector spin voltage can be derived from

spinor voltage as follows: (F6)

where . The variance of the noise varies depending

(E5)
on the time step size. The discretization was performed inter-
nally using a matlab implicit self-consistent solver.
2812 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 59, NO. 12, DECEMBER 2012

APPENDIX G REFERENCES
EQUIVALENT CONDUCTANCES SERIES AND PARALLEL
[1] S. A. Wolf et al., “Spintronics: A spin based electronics vision for the
The equivalent conductance of a two spin conductance ele- future,” Science, vol. 294, pp. 1488–1495, 2001.
ments connected in parallel is [2] C. Chappert, A. Fert, and F. N. Van Dau, “The emergence of spintronics
in data storage,” Nature Mater., vol. 6, pp. 813–813, 2007.
[3] “Report on ITRS emerging research devices,” [Online]. Avail-
(G1) able: http://www.itrs.net/links/2009itrs/2009chapters_2009ta-
bles/2009_ERD.pdf
(G2) [4] K. Bernstein, R. K. Cavin, W. Porod, A. Seabaugh, and J. Welser,
“Device and architecture outlook for beyond CMOS switches,” Proc.
The rule for voltage division says, voltage across conductance IEEE, vol. 98, no. 12, pp. 2169–2184, Dec. 2010.
[5] S. A. Wolf, L. Jiwei, M. R. Stan, E. Chen, and D. M. Treger, “The
, 2 is promise of nanomagnetics and spintronics for future logic and uni-
versal memory,” Proc. IEEE, vol. 98, no. 12, pp. 2155–2168, Dec.
(G3) 2010.
[6] S. Datta and B. Das, “Electronic analog of the electro-optical modu-
lator,” Appl. Phys. Lett., vol. 56, no. 7, pp. 665–667, 1990.
[7] R. P. Cowburn and M. E. Welland, “Room temperature mag-
netic quantum cellular automata,” Science, vol. 287, no. 5457, pp.
APPENDIX H 1466–1468, 2000.
WRITING SPIN MNA EQUATIONS, NOTATIONS [8] A. Ney, C. Pampuch, R. Koch, and K. H. Ploog, “Programmable com-
puting with a single magnetoresistive element,” Nature, vol. 425, no.
Rules for writing the spin MNA equations are very similar to 6957, pp. 485–487, 2003.
the rules for regular MNA [63], [65]. However, care should be [9] S. Sugahara and M. Tanaka, “A spin metal-oxide-semiconductor
field-effect transistor using half-metallic-ferromagnet contacts for
taken to include the spin current conservation law described in the source and drain,” Appl. Phys. Lett., vol. 84, no. 13, pp.
Section III. Also, 4 4 identity matrices are to be used where 2307–2309, 2004.
appropriate to indicate the voltage sources. [10] D. E. Nikonov and G. I. Bourianoff, “Spin gain transistor in ferromag-
netic semiconductors: The semiconductor Bloch equations approach,”
To apply the spin-MNA method to a circuit with n nodes with IEEE Trans. Nanotechnol., vol. 4, no. 2, pp. 206–214, 2005.
m voltage sources, the steps are: [11] J. Wang, H. Meng, and J.-P. Wang, “Programmable spintronics logic
device based on a magnetic tunnel junction element,” J. Appl. Phys.,
1. Number the nodes: Select a reference node (usually the vol. 97, no. 10, pp. 10D509-1–10D509-3, 2005.
ground, numbered 0) and name the remaining n–1 nodes. [12] D. A. Allwood, G. Xiong, C. C. Faulkner, D. Atkinson, D. Petit, and
Also label currents through each current source. R. P. Cowburn, “Magnetic domain-wall logic,” Science, vol. 309, no.
5741, pp. 1688–1692, 2005.
2. Name the currents: Assign a name to the current through [13] A. Khitun and K. L. Wang, “Nano scale computational architectures
each voltage source flowing from positive node to negative with spin wave bus,” Superlatt. Microstruct., vol. 38, no. 3, pp.
node of the source. 184–200, 2005.
[14] S. Lee, S. Choa, S. Lee, and H. Shin, “Magneto-logic device based on a
3. Apply spin KCL: Apply Spin current conservation law at single-layer magnetic tunnel junction,” IEEE Trans. Electron Devices,
each node with current into the node to be positive. vol. 54, no. 8, pp. 2040–2044, 2007.
[15] S. Matsunaga et al., “Fabrication of a nonvolatile full adder based on
4. Write an equation for spin voltage at each spin voltage logic-in-memory architecture using magnetic tunnel junctions,” Appl.
source. Phys. Exp., vol. 1, pp. 091301-1–091301-3, 2008.
5. Rearrange the equations to the form of (36). [16] B. Behin-Aein, D. Datta, S. Salahuddin, and S. Datta, “Proposal for an
all-spin logic device with built-in memory,” Nat. Nanotechnol., vol. 5,
6. Invert the matrix and multiply with to obtain the volt- no. 4, pp. 266–270, 2010.
ages of the nodes and current of the sources. [17] D. E. Nikonov, G. I. Bourianoff, and T. Ghani, “Proposal of a spin
7. Using conductance matrix to calculate the relevant torque majority gate logic,” IEEE Electron Device Lett., vol. 32, no. 8,
pp. 1128–1130, Aug. 2011.
currents. [18] B. Behin-Aein, A. Sarkar, S. Srinivasan, and S. Datta, “Switching en-
ergy-delay of all spin logic devices,” Appl. Phys. Lett., vol. 98, pp.
123510-1–123510-3, 2011.
APPENDIX I [19] S. Srinivasan, A. Sarkar, B. Behin-Aein, and S. Datta, “All spin logic
device with inbuilt nonreciprocity,” IEEE Trans. Magn., vol. 47, no.
COMMENT ON EXTENSION FOR SEMICONDUCTOR 10, pp. 4026–4032, Oct. 2011.
SPIN TRANSPORT [20] S. Borkar and A. A. Chien, “The future of microprocessors,” Commun.
ACM, vol. 55, no. 5, pp. 67–77, May 2011.
We briefly comment on the extendibility of the present ap- [21] V. Saripalli, G. Sun, A. Mishra, Y. Xie, S. Datta, and V. Narayanan,
proach to semiconductor spin transport [70]. The major mod- “Exploiting heterogeneity for energy efficiency in chip multiproces-
ifications required to handle semiconductors using conduction sors,” IEEE J. Emerg. Sel. Topics Circuits Syst., vol. 1, no. 2, pp.
109–119, Jun. 2011.
matrix approach are as follows: a) conductance matrix for de- [22] C. Augustine, G. Panagopoulos, B. Behin-Aein, S. Srinivasan, A.
scribing drift-diffusion spin current [71]; b) description of dilute Sarkar, and K. Roy, “Low-power functionality enhanced computation
architecture using spin-based devices,” in Proc. IEEE/ACM Int. Symp.
magnetic semiconductors; c) appropriate definition of spin cur- Nanoscale Archit. (NANOARCH), Jun. 8–9, 2011, pp. 129–136.
rents in presence of spin orbit coupling [72], [73]. [23] R. Venkatesan, V. K. Chippa, C. Augustine, K. Roy, and A. Raghu-
nathan, “Energy efficient many-core processor for recognition and
mining using spin-based memory,” in Proc. IEEE/ACM Int. Symp.
ACKNOWLEDGMENT Nanoscale Archit. (NANOARCH), Jun. 8–9, 2011, pp. 122–128.
[24] M. Hosomi, H. Yamagishi, T. Yamamoto, K. Bessho, Y. Higo, K. Ya-
The authors acknowledge the discussions with Prof. S. Datta, mane, H. Yamada, M. Shoji, H. Hachino, C. Fukumoto, H. Nagao,
and H. Kano, “A novel nonvolatile memory with spin torque transfer
S. Srinivasan, and B. Behin-Aein, at Purdue University and magnetization switching: Spin-RAM,” IEDM Tech. Dig., pp. 459–462,
Prof. S. Salahuddin at UC Berkeley. Dec. 2005.
MANIPATRUNI et al.: MODELING AND DESIGN OF SPINTRONIC INTEGRATED CIRCUITS 2813

[25] T. Kawahara, R. Takemura, K. Miura, J. Hayakawa, S. Ikeda, Y. Lee, [53] M. d’Aquino, C. Serpico, G. Coppola, I. D. Mayergoyz, and G. Bertotti,
R. Sasaki, Y. Goto, K. Ito, I. Meguro, F. Matsukura, H. Takahashi, H. “Midpoint numerical technique for stochastic Landau-Lifshitz-Gilbert
Matsuoka, and H. Ohno, “2 Mb spin-transfer torque RAM (SPRAM) dynamics,” J. Appl. Phys., vol. 99, no. 8, pp. 08B905-1–08B905-3.
with bit-by-bit bidirectional current write and parallelizing-direction [54] F. M. Spedalieri, A. P. Jacob, D. E. Nikonov, and V. P. Roychowdhury,
current read,” ISSCC Tech. Dig., pp. 480–481, Feb. 2007. “Performance of magnetic quantum cellular automata and limitations
[26] R. Beach et al., “A statistical study of magnetic tunnel junctions for due to thermal noise,” IEEE Trans. Nanotechnol., vol. 10, no. 3, pp.
high-density spin torque transfer-MRAM (STT-MRAM),” IEDM 537–546, May 2011.
Tech. Dig., pp. 306–308, Dec. 2008. [55] Support—Differential equations in MATLAB [Online]. Available:
[27] T. Kishi et al., “Lower-current and fast switching of a perpendicular www.mathworks.com/support/tech-notes/1500/1510.html#implic
TMR for high speed and high density spin-transfer-torque MRAM,” [56] S. Y. Jang, C.-Y. You, S. H. Lim, and S. R. Lee, “Annealing effects on
IEDM Tech.Dig., pp. 309–312, Dec. 2008. the magnetic dead layer and saturation magnetization in unit structures
[28] K. Ono et al., “SPRAM with large thermal stability for high immunity relevant to a synthetic ferrimagnetic free structure,” J. Appl. Phys., vol.
to read disturbance and long retention for high-temperature operation,” 109, no. 1, pp. 013901-1–013901-5, Jan. 2011.
in Symp. VLSI Tech. Dig., Jun. 2009, pp. 228–229. [57] C. Bilzer, T. Devolder, K. Joo-Von, G. Counil, C. Chappert, S. Cardoso,
[29] K. Lee and S. H. Kang, “Development of embedded STT-MRAM and P. P. Freitas, “Study of the dynamic magnetic properties of soft
for Mobile system-on-chips,” IEEE Trans. Magn., vol. 47, no. 1, pp. CoFeB films,” J. Appl. Phys., vol. 100, no. 5, pp. 053903-1–053903-4,
131–136, Jan. 2011. Sep. 2006.
[30] A. Driskill-Smith et al., “Latest advances and roadmap for in-plane and [58] G. D. Fuchs, J. C. Sankey, V. S. Pribiag, L. Qian, P. M. Braganca, A.
perpendicular STT-RAM,” in Proc. 3rd IEEE Int. Memory Workshop G. F. Garcia, E. M. Ryan, L. Zhi-Pan, O. Ozatay, D. C. Ralph, and R.
(IMW), May 22–25, 2011, pp. 1–3. A. Buhrman, “Spin-torque ferromagnetic resonance measurements of
[31] T. Yang, T. Kimura, and Y. Otani, “Giant spin-accumulation signal and damping in nanomagnets,” Appl. Phys. Lett., vol. 91, no. 6, pp. 062507-
pure spin-current-induced reversible magnetization switching,” Nat. 1–062507-3, Aug. 2007.
Phys., vol. 4, pp. 851–854, 2008. [59] X. Liu, Z. Wenzhe, M. J. Cartter, and G. Xiao, “Ferromagnetic res-
[32] J. Z. Sun et al., “A three-terminal spin-torque-driven magnetic switch,” onance and damping properties of CoFeB thin films as free layers in
App. Phys. Lett., vol. 95, pp. 083506.1–083506.3, Aug. 2009. MgO-based magnetic tunnel junctions,” J. Appl. Phys., vol. 110, no. 3,
[33] A. Vogel, J. Wulfhorst, and G. Meier, “Enhanced spin injection and pp. 033910-1–033910-5, Aug. 2011.
detection in spin valves with integrated tunnel barrier,” Appl. Phys. [60] Y. Kim, S. C. Oh, W. C. Lim, J. H. Kim, W. J. Kim, J. H. Jeong, H. J.
Lett., vol. 94, pp. 122510–122510, 2009. Shin, K. W. Kim, K. S. Kim, J. H. Park, S. H. Park, H. Kwon, K. H. Ah,
[34] Y. Fukuma, L. Wang, H. Idzuchi, S. Takahashi, S. Maekawa, and Y. J. E. Lee, S. O. Park, S. Choi, H. K. Kang, and C. Chung, “Integration
Otani, “Giant enhancement of spin accumulation and long-distance of 28 nm MJT for 8–16 Gb level MRAM with full investigation of
spin precession in metallic lateral spin valves,” Nat. Mater., vol. 10, thermal stability,” in Proc. Symp. VLSI Technol. (VLSIT), Jun. 14–16,
pp. 527–531, 2011. 2011, pp. 210–211.
[35] T. Valet and A. Fert, “Theory of the perpendicular magnetoresistance [61] M. Munakata, A. Shin-Ichi, and M. yagi, “B-concentration dependence
in magnetic multilayers,” Phys. Rev. B, vol. 48, pp. 7099–7113, 1993.
on anisotropy field of CoFeB thin film for gigahertz frequency use,”
[36] A. Brataas, Y. V. Nazarov, and G. E. W. Bauer, “Finite element theory
IEEE Trans. Magn., vol. 41, no. 10, pp. 3262–3264, Oct. 2005.
of transport in ferromagnet-normal metal systems,” Phys. Rev. Lett.,
[62] H. Chung-Wen, A. Ruehli, and P. Brennan, “The modified nodal ap-
vol. 84, pp. 2481–2484, 2000.
proach to network analysis,” IEEE Trans. Circuits Syst., vol. CAS-22,
[37] M. D. Stiles and A. Zangwill, “Anatomy of spin-transfer torque,” Phys.
no. 6, pp. 504–509, Jun. 1975.
Rev. B, vol. 66, pp. 014407-1–014407-14, 2002.
[63] R. A. DeCarlo and P. M. Lin, Linear Circuit Analysis: Time Domain,
[38] A. Brataas, G. E. W. Bauer, and P. J. Kelly, “Non-collinear magneto-
Phasor and Laplace Transform Approaches. New York: Oxford
electronice,” Phys. Rep., vol. 427, no. 4, pp. 157–255, Apr. 2006.
Univ. Press, 2001.
[39] C. Cohen-Tannoudji, B. Diu, and F. Laloe, Quantum Mechanics.
[64] V. Litovski and M. Zwolinski, VLSI Circuit Simulation and Optimiza-
Weinheim, Germany: Wiley-VCH, 1977, vol. 2.
[40] M. D. Stiles, “Spin-dependent interface transmission and reflection in tion. Norwell, MA: Kluwer Academic, 1997.
magnetic multilayers,” J. Appl. Phys., vol. 79, pp. 5805–5810, 1996. [65] An algorithm for modified nodal analysis [Online]. Available: http://
[41] D. C. Ralph and M. D. Stiles, “Spin transfer torques,” J. Magn. Magn. www.swarthmore.edu/NatSci/echeeve1/Ref/mna/MNA3.html
Mater., vol. 320, pp. 1190–1190, 2008. [66] R. Berry, “An optimal ordering of electronic circuit equations for a
[42] M. Cahay, M. McLennan, and S. Datta, “Conductance of an array of sparse matrix solution,” IEEE Trans. Circuit Theory, vol. CT-18, no.
elastic scatterers: A scattering-matrix approach,” Phys. Rev. B, vol. 37, 1, pp. 40–50, Jan. 1971.
pp. 10125–10136, 1988. [67] J. Wilkie, “Numerical methods for stochastic differential equations,”
[43] Y. V. Nazarov, “Circuit theory of andreev conductance,” Phys. Rev. Phys. Rev. E, vol. 70, 2004, Art. 017701.
Lett., vol. 73, pp. 1420–1423, 1994. [68] K. Burrage, P. Burrage, D. Higham, P. Kloeden, and E. Platen, “Com-
[44] S. Dubois et al., “Evidence for a short spin diffusion length ment on Numerical methods for stochastic differential equations,”
in permalloy from the giant magnetoresistance of multilayered Phys. Rev. E, vol. 74, 2006, Art. 068701.
nanowires,” Phys. Rev. B, vol. 60, pp. 477–484, 1999. [69] D. J. Higham, “An algorithmic introduction to numerical simulation of
[45] J. Bass and W. P. Pratt Jr., “Spin-diffusion lengths in metals and stochastic differential equations,” SIAM Rev., vol. 43, p. 525, 2001.
alloys, and spin-flipping at metal/metal interfaces: An experimen- [70] H. Ohno, “A window on the future of spintronics,” Nat. Mater., vol. 9,
talist’s critical review,” J. Phys., Condens. Matter, vol. 19, pp. pp. 952–954, 2010.
183201-1–183201-50, 2007. [71] A. M. Roy, D. E. Nikonov, and K. C. Saraswat, “Electric field effects
[46] T. Kimura and Y. Otani, “Large spin accumulation in a permalloy- in semiconductor spin transport—A transfer matrix formalism,” IEEE
silver lateral spin valve,” Phys. Rev. Lett., vol. 99, pp. 196604- Trans. Magn., vol. 47, no. 10, pp. 2746–2749, Oct. 2011.
1–196604-4, 2007. [72] J. Shi, P. Zhang, D. Xiao, and Q. Niu, “Proper definition of spin current
[47] T. Kimura, T. Sato, and Y. Otani, “Temperature evolution of spin re- in spin-orbit coupled systems,” Phys. Rev. Lett., vol. 96, 2006, Art.
laxation in NiFe/Cu lateral spin valve,” Phys. Rev. Lett., vol. 100, pp. 076604.
066602-1–066602-4, 2008. [73] H.-J. Drouhin, G. Fishman, and J.-E. Wegrowe, “Spin currents in semi-
[48] C. Kittel, Introduction to Solid State Physics, 8th ed. Hoboken, NJ: conductors: Redefinition and counterexample,” Phys. Rev. B, vol. 83,
Wiley, 2005. 2011, Art. 113307.
[49] J. C. Slonczewski, “Current-driven excitations of magnetic multi- [74] U. E. Avci, R. Rios, K. Kuhn, and I. A. Young, “Comparison of per-
layers,” J. Magn. Magn. Mater., vol. 159, pp. L1–L7, 1996. formance, switching energy and process variations for the TFET and
[50] L. Berger, “Emission of spin waves by a magnetic multilayer traversed MOSFET in logic,” in Proc. Symp. VLSI Technol. (VLSIT), Jun. 14–16,
by a current,” Phys. Rev. B, vol. 54, pp. 9353–9358, 1996. 2011, pp. 124–125.
[51] W. F. Brown, “Thermal fluctuations of a single-domain particle,” Phys. [75] N. Kim et al., “Leakage current: Moore”s law meets static power,”
Rev., vol. 130, pp. 1677–1686, 1963. Computer, vol. 36, no. 12, pp. 68–75, 2003.
[52] J. Z. Sun, “Spin angular momentum transfer in current-perpendicular [76] S. Salahuddin and S. Datta, “Self-consistent simulation of quantum
nanomagnetic junctions,” IBM J. Res. Develop., vol. 50, no. 1, pp. transport and magnetization dynamics in spin-torque based devices,”
81–100, Jan. 2006. Appl. Phys. Lett., vol. 15, 2006, Art. 153504.
2814 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 59, NO. 12, DECEMBER 2012

Sasikanth Manipatruni (M’07) was born in Ian. A. Young (M’78–SM’96–F’99) was born in
Vizianagaram, India. He graduated from Indian Melbourne, Australia. He received the B.S. and M.S.
Institute of Technology (IIT) Delhi at the top of degrees in electrical engineering from the University
his class in electrical engineering in 2005 with the of Melbourne, Australia, in 1972 and 1975, respec-
best hardware thesis project award. He received tively, and the Ph.D. degree in electrical engineering
the Ph.D. degree from Cornell University, Ithaca, from the University of California, Berkeley, in 1978.
NY, working with Prof. Michal Lipson in silicon Prior to Intel, Dr. Young worked on analog/digital
photonics and opto-mechanics. During his Ph.D., he integrated circuits for Telecommunications products
was co-inventor of several silicon photonic devices at Mostek Corporation (United Technologies Corpo-
including the first 18 Gb/s micro-ring modulator, ration). He is a Senior Fellow and Director of Ex-
first GHz poly-silicon modulator, hitless broadband ploratory Integrated Devices and Circuits group in
EO switches and integrated silicon nanophotonic link. Intel Components Research, Technology and Manufacturing Group at Intel Cor-
He was previously at General Electric (GE) Global Research Center working poration, Hillsboro, OR. He is responsible for defining and developing future
in nanophotonic interconnects for massively parallel magnetic resonance circuit directions with emerging novel devices and identifying leading options
imaging (MRI) and demonstrated the first analog-RF optical MRI system at for devices and interconnects to manufacture solid-state integrated circuits in
3T magnetic fields. He joined Intel, Hillsboro, OR, in 2011. He is a Research the post-CMOS era. He joined Intel in 1983. Starting with the development of
Scientist in the Exploratory Integrated Devices and Circuits group in Intel Com- circuits for a 1 Mb DRAM, and the world’s first 1 64 K SRAM, he then led
ponents Research, Technology and Manufacturing Group at Intel Corporation. the design of three generations of SRAM products and manufacturing test vehi-
He is working on emerging novel devices to identify technology options for cles, and developed the original phase locked loop (PLL) based clocking circuit
beyond CMOS logic technologies. He has more than 10 patent applications in in a microprocessor while working on the 50 MHz Intel 486™ processor design.
nano-photonics and MRI, and 40 peer reviewed journal and conference papers. He subsequently developed the core PLL clocking circuit building blocks used
Dr. Manipatruni was a KVPY national science fellow of the Indian Insti- in each generation of Intel microprocessors through the 0.13 3.2 GHz Pen-
tute of Sciences (IISc) (1999–2001); worked at Swiss Federal Institute of Tech- tium 4. Dr. Young has developed a number of optimization metrics for process
nology (ETH), Zurich in 2004; and Inter University center for astronomy and technology development, including the transistor performance metric that pro-
astrophysics (IUCAA) in 2001. He serves as a peer reviewer for OSA, IEEE, vided a link between processor performance and basic transistor parameters, as
and Nature Photonics. His business plan for fab-less photonics was awarded well as back-end metal interconnect architecture. He holds 55 patents in inte-
the first place in Asia-Pacific at Asia Moot Corp 2008 and selected as an out- grated circuits and has authored or coauthored over 50 technical papers.
standing product at World Moot Corp 2009. Dr. Young was a member of the Symposium on VLSI Circuits Technical
Program Committee from 1991 to 1996, serving as the Program Committee
Chairman in 1995/1996, and the Symposium Chairman in 1997/1998. He was
a member of the ISSCC Technical Program Committee from 1992 to 2005,
Dmtri E. Nikonov (M’99–SM’06) received the serving as the Digital Subcommittee Chair from 1997 to 2003, the Technical
M.S. degree in aeromechanical engineering from Program Committee Vice-chair in 2004 and Chair in 2005. He was Guest Editor
the Moscow Institute of Physics and Technology, for the April 1997, April 1996, and December 1994 issues of the IEEE JOURNAL
Russia, in 1992 and the Ph.D. degree in physics from OF SOLID-STATE CIRCUITS. He has served as an elected member of the SSCS
Texas A&M University, College Station, in 1996, Adcom from 2006 to 2009. He received the International Solid-State Circuits
where he participated in the demonstration of the Conference’s 2009 Jack Raper Award for Outstanding Technology-Directions,
world’s first laser without population inversion. for the paper entitled “Optical I/O Technology in Tera-Scale Computing.”
He joined Intel Corporation in 1998 and is
presently a Research Scientist in the Components
Research group in Hillsboro, OR. He is responsible
for simulation and benchmarking of beyond-CMOS
logic devices and for managing joint research programs with universities on
nanotechnology and exploratory devices. From 1997 to 1998 he was a research
engineer and lecturer at the Department of Electrical and Computer Engineering
of University of California Santa Barbara. In 2006 he was appointed Adjunct
Associate Professor of Electrical and Computer Engineering at Purdue Uni-
versity. He has 60 publications in refereed journals in quantum optics, lasers,
nanoelectronics and spintronics, and 35 issued patents in optoelectronics,
integrated optic and spintronic devices.
Dr. Nikonov was a finalist of the Best Doctoral Thesis competition of the
American Physical Society in 1997.