investigation of chirality dependence of carbon nanotube-based ring oscillator
TRANSCRIPT
Investigation of chirality dependence of carbon nanotube-based ring oscillator
Yaser Mohammadi Banadaki, Safura sharifi, Ashok Srivastava*
Division of Electrical and Computer Engineering
Louisiana State University
Baton Rouge, LA 70803, U.S.A. *[email protected]
ABSTRACT
Carbon nanotube (CNT)-based integrated circuits including CNT field effect transistors (FET) and CNT
interconnects are a promising emerging technology. In this paper, we have designed and simulated a single-
walled CNT-based ring oscillator (RO) to investigate the chirality dependence of oscillation frequency using
a numerical quantum-based model of CNT FET with 5 nm gate length and a compact diffusive model for
CNT interconnects. We have obtained oscillation frequency of 25 GHz assuming contact resistance of
24𝐾𝛺, which reduces by ~%8.5 and ~%10 for 1 nm increase in CNT chiral diameter of zigzag CNT FET
and 1𝜇𝑚 metallic CNT interconnects between inverter stages, respectively.
Keywords: Single-walled carbon nanotube ring oscillator; Field effect transistors; Interconnects.
INTRODUCTION
The International Technology Roadmap for
Semiconductors (ITRS) [1] has mentioned future
technological and intrinsic device difficulties of
scaling Si MOSFETs in CMOS technologies
enforcing the semiconductor industry to explore the
use of novel materials and devices. Some of the
most promising alternatives for emerging VLSI
integrated circuits have been categorized under
“Carbon-based Nanoelectronics” such as CNT FET
[2-5] or the graphene-nanoribbon FET [6-7].
Basically, a single-walled CNT is a rolled-up sheet
of graphene discovered in 1991 by Iijima [8]. A
SWCNT offers high mechanical strength, high
carrier density, high thermal and electrical
conductivity with near ballistic transport. CNT FET
has similar current-voltage characteristics as that of
Si MOSFETs while the complementary design can
be fabricated using identical transistors leading to
less fabrication processes for designing CNT-based
integrated circuits [9]. Carbon nanotubes can be
either metallic or semiconducting depending on
rolling direction. In a CNT-based circuit, the
semiconducting and conducting CNTs are used for
transistors and for interconnects, respectively.
Recently, the experimental reports [10, 11] reveal
that the speed of CNT growth has the chirality
dependence developing the ability to selectively
grow CNT with specific chirality. This has
significantly influenced on the characteristics of
CNT device and interconnects urging to be
investigated in designing CNT-based integrated
circuits [12, 13] such as the ring oscillator, which
has been the purpose of this paper.
A ring oscillator is the ultimate test circuitry for
new materials and a key building block in radio
frequency, digital electronic and optical
communication systems for the frequency
translation of information signals, channel selection
and time reference [14, 15]. Javey et al have
fabricated the first complementary SWCNT-based
3-stages RO with the CNT length and diameter of 3
µm and 2 nm obtaining the oscillation frequency of
∼220 Hz [16]. IBM has announced the fabrication
of high performance CNT-based 5-stages RO with
the CNT length and diameter of 18 µm and 2nm,
respectively, which results in oscillation frequency
of 52 MHz [17]. A seven-stage CNT-based RO has
also been proposed and simulated for the CNT FET
with the gate length of 200 nm and the oscillation
frequency of 3.75 MHz [9]. As Moore’s law
requires shrinking the channel length of transistors
below 10nm within the next decade, an important
next step would be predicting and exploring the
characteristics of CNT-based integrated circuits
with regard to physical parameters of CNT such as
its chirality. Thus, a simulation approach needs to
consider the quantum confinement of the tube
circumferential direction and quantum tunneling
along the channel [18]. The non-equilibrium
Green’s function (NEGF) formalism [18-22] is a
strength simulation method which fully treats the
quantum phenomena such as the tube confinement,
quantum tunneling and contact effect leading to a
very accurate extraction of chirality dependence of
few-nanometer CNT channels. In this case, the
length of semiconducting CNT is usually less than
its mean-free-path which allows for ballistic
simulation of carriers transport in CNT FETs. For
interconnects in VLSI design, Cu has much less
mean-free path than a metallic CNT and suffers
from electron surface scattering, grain-boundary
scattering, electromigartion and void formation [23-
27]. Thus, the application of metallic CNTs for
interconnects along with CNT FETs can increase
the performance of VLSI circuits.
In this paper, we have investigated the chirality
dependence of CNT-based ring oscillator using a
numerical ballistic model for CNT FETs and a
compact diffusive model for CNT interconnects.
The arrangement of the paper is as follows; in
Section 2, the gate-all-around CNT FET structure
has been modeled by solving the Poisson and
Schrödinger equations within the NEGF formalism.
The current-voltage characteristics and
corresponding voltage transfer characteristics of the
CNT FET-based inverter have been obtained as a
function of the chirality of few nanometer CNT
channels. In section 3, the semi-classical model of
CNT interconnect has been introduced for a metallic
single-walled CNTs based on our previously
developed one-dimensional fluid model of
electronic transport considering scattering
mechanisms in long conducting CNT [24]. The
equivalent transmission model of a metallic
SWCNT isolated above a ground plane consists of
circuit elements distributed along its length
including magnetic and kinetic inductance, quantum
and electrostatic capacitances, and a non-linear
resistance resulting from acoustic and optical
phonon scatterings. Section 4 is devoted to the
simulation of CNT-based RO circuits, where
semiconducting and metallic CNTs have been used
as transistors and interconnects between inverters,
respectively. The obtained data from the device-
level simulation of CNT FETs and the model of
CNT interconnect have been introduced into SPICE
circuit simulator. The multi-stage CNT-based RO
has been simulated to explore the chirality
dependence of the oscillation frequency as a
fundamental physical parameter of CNT FETs and
CNT interconnects in designing a CNT-based RO
circuits.
Figure 1. (a) The 3-D schematic view of gate-all-around
CNTFETs. LD, LG and LS are the lengths of drain, gate and
source where the spaces between the regions are due to
visibility purpose and assumed zero in the simulations. tox and
D are the insulator thickness and the diameter of
semiconducting CNT corresponding to chirality n=3k+1 (k is
positive integer), respectively. (b) The atomic configuration in
real space and corresponding conceptual configuration in
mode space approaches. The x arrows are the direction along
the CNT FET device, c and kc are the circumferential
directions in real space and reciprocal space, respectively.
Coupling parameters between the nearest neighbor carbon
rings in mode space are shown with t and b2q, respectively.
CNT FET
Figure 1(a) shows a three-dimensional (3-D)
schematic view of the gate-all-around CNTFETs,
where the semiconductor zigzag nanotube has been
used in the channel with chirality n=3k+1 (k is
positive integer) [13]. The source and drain regions
are heavily doped and the metallic coaxial gate
manipulates the carrier transport in CNT channel.
The proposed CNT FET can be modeled
accurately using NEGF formulism, which can treat
quantum-mechanical confinement, reflection, and
tunneling [20]. The CNT FET was simulated by
converging an iterative procedure, where the
Poisson and Schrödinger equations are solved self-
consistently using NEGF formulism [19-22].
The gate-all-around CNT transistor results in
invariant potential and charge density around the
nanotube. The electrostatic potential 𝑈𝑖 along the
CNT can be determined by 2-D Poisson equation as
follows,
∇2Ui(r, x) = −q
ε[p(xi) − n(xi) + ND
+ − NA−] (1)
where ND+and NA
− are the doping profile of ionized
donor and acceptor, respectively and p(xi) and n(xi)
are hole and electron density distribution which are
updated using the calculated NEGF formulism in
iterative procedure. The retarded Green’s function
is given by,
𝐺(𝐸) = [(𝐸 + 𝑖η+)𝐼 − 𝐻 − ΣS − ΣD]−1 (2)
where E is the energy, η+ is an infinitesimal positive
value, I is the identity matrix and H is the
Hamiltonian matrix for an isolated CNT channel. ΣS
and ΣD are the self-energy matrices of the source
and drain, which describe how the ballistic channel
couples to the contacts. The mean-free-path of
nanotube is around hundreds of nanometers at room
temperature which allows near ballistic carriers
transport. In addition the carrier transport can be
describe in terms of the circumferential modes
known as “Mode space approach” [18, 19] since the
potential variation around coaxial-gated nanotube is
small. In other word, the periodic boundary
conditions kcC = 2πq , where C is the
circumference of the nanotube and q is the angular
quantum number, mathematically allows to
decouple the two dimensional nanotube lattice of a
(n,0) zigzag nanotube into n-uncoupled one-
dimensional mode space lattices as shown in
Fig.1(b). This reduces the size of the Hamiltonian
matrix and the corresponding computational cost.
By constructing two different types of rings along
the tube, the Hamiltonian matrix for the qth
subband
can be calculated as follows,
𝐻 =
[
𝑈1
𝑏2𝑞
𝑏2𝑞
𝑈2
𝑡
0
𝑡
𝑈3
𝑏2𝑞
⋱
0
𝑡
𝑈𝑁−1
𝑏2𝑞
𝑏2𝑞
𝑈𝑁 ]
𝑁×𝑁
(3)
where b2q = 2tcos(πq/n) and t are coupling
parameters between the nearest neighbor carbon
rings, and N is the total number of carbon rings
along CNT device. The diagonal elements Ui’s are
the electrostatic potential at the ith
carbon ring which
obtained from Eq. (1).
Corresponding to boundary conditions of the
Schrödinger equation, the source self-energy matrix
has only non-zero value in the (1, 1) element given
by,
Σ𝑆(1,1) =
(E−U1)2+t2+b2q2 ±√[(E−U1)2+t2+b2q
2 ]2−4(E−U1)2t2
2(E−U1) (4)
By replacing U1 with UN, the drain self-energy
matrix with non-zero value in the (N, N) element
can be calculated using an equation similar to
Eq.(4). The level broadening of source and drain
can be obtained using the self-energies of the
corresponding contacts as follows,
ΓD(S)(E) = i[ΣD(S)(E) − ΣD(S)+ (E)] (5)
where ΣD(S)+ is the Hermitean conjugate of ΣD(S)
.
The electron and hole correlation functions
𝐺 𝑛(𝑝)(𝐸) can be calculated by,
Gn(E) = G(E)Σin(E)G+(E) (6a)
Gp(E) = G(E)Σout(E)G+(E) (6b)
where Σ𝐷(𝑆)𝑖𝑛 = ΓD(S)fD(S) and Σ𝐷(𝑆)
𝑜𝑢𝑡 = ΓD(S)(1 −
fD(S)) are in(out) scattering functions for drain
(source) contact couplings, respectively, which are
the level broadening weighted by Fermi distribution
function fD(S) . The induced electron and hole
distributions on carbon ith
ring can be calculated
using the following equations,
n(xi) =4
∆x∫
Gj,jn (E)
2π
+∞
Em(j)dE (7a)
p(xi) =4
∆x∫
Gj,jp
(E)
2π
Em(j)
−∞dE (7b)
where Em is the midgap energy and the factor of four
is due to the two spin states and the twofold valley
degeneracy. The calculated p(xi) and n(xi) have been
applied to Poisson equation to calculate 𝑈𝑗𝑛𝑒𝑤 and
the corresponding Hamiltonian in the self-consistent
iterative loop. Once the algorithm converges, the
transmission coefficient T(E) and the corresponding
current can be calculated,
T(E) = trace(ΓS(E)G(E)ΓDG+(E)) (8)
I =4e
h∫ T(E)[f(E − EFS) − f(E − EFD)]
+∞
−∞dE (9)
where h is the Plank constant, f is the Fermi-Dirac
distribution and EFS and EFD are the Fermi energy
levels in the source and drain regions. The details of
the quantum transport calculation of CNT-based
field effect transistor can be followed in [18-20].
The drain-source currents of the proposed CNT
FETs have been obtained using NEGF model for
different chirality (n=3k+1) from 7 to 28. Figure
2(a) and 2(b) show the chirality dependence of
drain current as a function of drain voltage and gate
voltage, respectively. Increasing the chirality leads
to the increase in drain-source current due to the
contribution of more quantum channels
corresponding to the angular quantum number q in
equation kcC = 2πq . By intersecting the current-
voltage characteristics of n- and p-type CNT FETs,
the voltage transfer characteristic (VTC) of
complementary CNT FET-based inverter can be
obtained as a function of chirality as shown in Fig.
3(a). The VTC curves of low-chirality CNT FETs
have sharp transition from ON-state to OFF-state
while the non-linearity in VTC curves decreases by
increasing chirality of CNT FETs. The current of
CNT FET-based inverter as a function of input
voltage corresponding to VTC curve has been
shown in Fig. 3(b). Increasing the chirality from 7
to 28 leads to four orders of magnitude increase in
the maximum current of inverter from ~5.7 ×10−9𝐴 to ~4.2 × 10−5𝐴.
Figure 2. The chirality dependence of CNT FET
characteristics. (a) the IDS-VDS and (b) the IDS-VGS
characteristics as a function of CNTFET chirality from n=7 to
25. The device geometries of the CNT FETs are tox = 2nm,
LG = 10nm and LD = LS = 10nm , where the dielectric
constant is assumed εox=16.
Figure 3. (a) Voltage transfer characteristic of CNTFET-based
inverter using NEGF model for different chirality (n,0), (b) the
current of CNTFET-based inverter as a function of input
voltage for different chirality (n,0) and (c) Schematic of
CNTFET-based inverter. The device geometries and
parameters of the CNT FETs are kept as mentioned in Fig. 2.
CNT INTERCONNECT
A metallic CNT has less problem with
electromigartion and void formation and exhibits the
high thermal and electrical conductivity as well as
mechanical properties [23, 24]. Electrons in a Cu
interconnect not only have much less mean-free path
than a CNT interconnect, but also suffer from
electron surface scattering and grain-boundary
scattering leading to increasing its resistivity. These
performances introduce a metallic CNT interconnect
as a promising alternative to current Cu interconnect
in VLSI technology [1].
Srivastava et al. [25] have been developed the
semi-classical model for a conducting metallic
SWCNT using one-dimensional fluid model. The
equivalent circuit model of a metallic SWCNT
isolated above a ground plane consists of circuit
elements distributed along its length as shown in fig.
4(a). The inductance (L) consists of the series of
magnetic inductance (LM) and kinetic inductance
(LK). The magnitude of magnetic inductance is
logarithmically sensitive to the ratio of distance
between nanotube and the ground ( tins ) and
nanotube diameter (Din) given by,
LM =μ
2πln(
tins
Din) (10)
The kinetic inductance is dominant inductance
with several orders of magnitudes and thus plays a
vital role in high frequency applications, which is
contrary to macroscopic long thin wires with
relatively larger magnetic inductance. The kinetic
inductance is given by,
Lk ≈πℏ
e2vF (11)
where ℏ is reduced Plank constant.
Figure 4. (a) Transmission line model of a SWCNT
interconnect. The inductance (L) consists of the series of
magnetic (LM) and kinetic (LK) inductances. CE and CQare the
electrostatic and quantum capacitance. Rc is the contact
resistance, which is assumed equal to zero in figures 5-7 while
its dominant effect is investigated in figure 8. R(V, Din) is the
resistance of CNT interconnect which depends on voltage
across the interconnect and CNT diameter, (b) Equivalent
capacitance of a CNT interconnect with the length of Lin =1 μm as a function of chirality (n,n) and (c) Resistivity of a
CNT interconnect with the length of Lin = 1 μm as a function
of voltages and its chirality (n,n).
The capacitance also consists of quantum
capacitance of the nanotube and electrostatic
capacitance between a wire and a ground plane. The
quantum capacitance is due to the Pauli Exclusion
Principle in quantum electron gas of a metallic
SWCNT interconnect given by,
CQ ≈e2
πℏvF≈ 100 aF/μm (12)
Similar to the magnetic inductance, the
electrostatic capacitance depends on the insulator
thickness and nanotube diameter while its
contribution in calculating equivalent capacitance is
important and given by,
CE ≈2πε
ln(tinDin
) (13)
The increase of equivalent capacitance as a
function of chirality is shown in Fig 4(b). The
resistance in the equivalent transmission model has
the voltage (V) and diameter (Din ) dependence,
which can be determined by the effective mean free
path of electron in a metallic SWCNT interconnect
(λeff) considering acoustic and optical phonon
scattering mechanisms. The spontaneous scattering
lengths for emitting an optical phonon (λop) and
acoustic phonon (λac) have diameter dependence,
which can be estimated as follows [5],
λop = 56.4 Din (14)
λac =400.46×103Din
T (15)
In order to calculate the effective mean free path,
three scattering lengths needs to be calculated for
optical phonon scattering, which are optical phonon
absorption (λop,abs), optical phonon emission for the
absorbed energy ( λop,emsabs ), and optical phonon
emission for the electric field across the SWCNT
length (λop,emsfield ) as follows [24],
λop,abs = λopNop(300)+1
Nop(T) (16)
λop,emsabs = λop,abs +
Nop(300)+1
Nop(T)+1λop (17)
λop,emsfield =
ℏωop−KBT
qV
Lin
+Nop(300)+1
Nop(T)+1λop (18)
where Nop = 1 [1 + exp (ℏωop KBT⁄ )]⁄ is the
optical phonon occupation in Bose-Einstein
statistics, KB is the Boltzman constant, ℏωop is
optical phonon energy and Lin is the CNT
interconnect length. The effective mean free path
and the corresponding resistance can now be
calculated by Matthiessen’s rule as follows,
1
λeff=
1
λac+
1
λop,emsabs +
1
λop,emsfield +
1
λop,abs (19)
R(V, Din) =R0
4
Lin
λeff (20)
where R0 ≈ πℏ 2e2⁄ ≈ 12.9 kΩ is quantum
resistance. Figure 4(c) shows the voltage and
diameter (interconnect chirality) dependence of the
resistance in the equivalent transmission model for a
CNT interconnect with the length of Lin = 1 μm.
The calculated resistance of SWCNT interconnect is
highly sensitive to the value of voltage across the
tube, in which the dependence increases by
decreasing chirality. The increase in chiral value
leads to the increase in the number of energy bands
in energy-wavevector diagram of the metallic CNT
enhancing the electron current capability. By
increasing chiral value, the equivalent capacitance
increases while the resistance decreases depending
on voltage across the tube, which results in a RC
delay and consequently variation in oscillation
frequency of CNT-based RO.
CNT-BASED RING OSCILLATOR
In a ring oscillator, the input of each inverter is the
output of the previous inverter, closing the ring by
returning the output of the last inverter to the first
stage. The cascade combination of inverters results
in sequentially change of logic state and the
corresponding oscillation depending on propagation
delay (𝑡𝑝) in voltage switching between successive
inverter stages. Figure 5(a) shows the schematic of
the multi-stage RO, where semiconducting and
metallic CNTs have been used as transistors and
interconnects between inverters, respectively. The
developed analytical or numerical model of a nano-
meter device and interconnect can be incorporated
in SPICE using Verilog-AMS [28] or SPICE
elements [29, 30]. Here, we have incorporated the
device-level data of CNT FET-based VTC curve
and the non-linear resistance of CNT interconnects
using GTABLE and ETABLE elements in SPICE
as shown in Fig. 5(b). The former is a four
terminals element for initially modulating the
current-controlled by voltage, which has been
modified for non-linear resistance of CNT
interconnects by connecting the input and output
terminals in pair creating a two-terminal element
[29].
Figure 6(a) shows the output waveforms and its
Fourier transform of 3-stages CNT-based RO for
CNTFET (7, 0) and CNT interconnect (27, 0). The
circuit has a large signal oscillation at about 220
GHz and the odd components are dominant in the
frequency spectrum of the signal. The frequency of
a RO is given by f = 1 (2tpN)⁄ , where N is the
number of stages. The proposed RO, N=3, results in
an average propagation delay of 0.75 pico-second
for each stage due to its intrinsic CNT FET and
CNT interconnect assuming the perfect electrode
contacts (Rc = 0) in the model. The short channel
length of CNT FET (LG = 5 𝑛𝑚) and high intrinsic
mobility of CNT results in low propagation delay of
inverter stages [16]. Figure 6(b) shows the output
waveforms of the proposed RO for the chiralities of
CNT FET equal to n=7 and 28. The square wave
has been obtained for chirality n=7 while changing
to the sinusoidal wave for n=28.
Figure 5. (a) Circuit schematic of multi-stage CNT-based ring
oscillator and (b) SPICE schematic of CNT interconnect and
CNT FET-based inverter showing the correct connection of
GTABLE and ETABLE elements.
Figure 6. (a) Output waveforms and its Fourier transform of 3-
stage CNT-based RO for CNTFET (7,0) with 𝜀𝑜𝑥 = 4 and
CNT interconnect (27,0) and (b) Output waveforms of 3-stage
CNT-based RO for chiralities of CNTFET equal to 7 and 28
for 𝜀𝑜𝑥 = 25. The dielectric thickness of CNTFET (tox) and
the length of interconnect (Lin) are assumed 2nm and 100nm,
respectively. The lengths of drain (LD), gate (LG) and source
(LS) regions are 5nm in all the simulations.
The oscillation frequency decreases by
increasing CNT FET chirality as shown in Fig 7(a).
The oscillation frequency of 3-stage RO has been
obtained f = ~250 GHz reducing to ~ 215 GHz for
the change of chirality, n, from 7 to 28. The
diameter of CNT can be related to chirality by
𝑑 = √3𝑎𝑐𝑐𝑛 𝜋⁄ , where 𝑎𝑐𝑐 = 0.142 𝑛𝑚 is inter-
atomic distance between two nearest-neighbor
carbon atoms. Thus, by changing the diameter of
semiconducting CNT about 1.64 nm, the oscillation
frequency is reduced by ~ %14.6 leading to
straightforward estimation of frequency reduction
by ~ % 8.5 per 1 nm increase in diameter of zigzag
CNT. The oscillation frequency is determined by
the propagation time delay of CNT-based inverter
stages and by the number of stages used in the ring.
The latter creates the multi-phase outputs as each
stage provides a phase shift of 𝜋 𝑚⁄ , where m is
number of stages. Figure 7(b) and 7(c) show the
frequency of CNT-based RO as a function of
chirality for ring oscillators with m = 5 and m = 9,
respectively. For all three types of ring oscillators,
the oscillation frequency is decreased by increasing
the CNT chirality. However, the trends are different
as the change of frequency by chirality becomes
more linear for ring oscillator with higher number
of stages.
Figure 7. Frequency of CNT-based RO as a function of
chirality for ring oscillators with (a) three, (b) five and (c) nine
stages. The simulation parameters are LD = LG = LS = 5nm,
𝑡𝑜𝑥 = 2𝑛𝑚, 𝜀𝑜𝑥 = 16, Lin = 100 n𝑚 and nin = 27.
Figure 8. (a) Frequency of 3-stage CNT-based ring oscillator as
a function of chirality for different dielectric thicknesses when
the dielectric constant of CNTFETs are 𝜀𝑜𝑥 = 4 and (b)
Frequency of 3-stage CNT-based ring oscillator as a function
of chirality for different dielectric constants when the dielectric
thickness of CNTFETs are 𝑡𝑜𝑥 = 2𝑛𝑚 . ( Lin = 100 n𝑚
and nin = 27).
Figure 9. Frequency of 3-stage CNT-based RO as a function
of chirality for ring oscillators for a conducting CNT
interconnect with different chiralities when the contact
resistance is Rc = 24KΩ. The simulation parameters are LD =LG = LS = 5nm,𝑡𝑜𝑥 = 2𝑛𝑚, 𝜀𝑜𝑥 = 16, Lin = 1μ𝑚 and nin =27.
Figure 8(a) shows the oscillation frequency of 3-
stage CNT-based ring oscillator as a function of
CNT FET chirality for different dielectric
thicknesses (tox ). The other circuit parameters as
well as the chirality of CNT interconnect have been
kept constant in the simulation. For tox = 1nm, the
oscillation frequency decreases from 245 GHz to
217 GHz by increasing CNT FET chirality form n =
7 to 22 (blue curve). For CNT FET chirality more
than n=25, the oscillation of CNT-based RO is
eliminated, which results from reducing non-
linearity of VTC curves of CNT-based inverters. By
increasing the dielectric thickness, the non-
oscillation chiralities have been limited to lower
values n=19, 16, and 13 corresponding to the
dielectric thickness of tox = 1.5nm, 2nm
and 2.5 nm, respectively. The oscillation frequency
is more sensitive to CNT FET chirality for higher
dielectric thickness resulting in higher reduction of
frequency from n=7 to n=10 at tox = 2.5nm (black
curve). Also, at the same chirality n=7, the
oscillation frequency of 245 GHz at tox =1nm decreases to ~232 GHz at tox = 2.5 nm due to
the reduction of gate controllability on the
modulation of CNT FET channel by increasing the
thickness of gate oxide. Figure 8(b) shows the
frequency of 3-stage CNT-based RO as a function of
chirality for the dielectric constants of εox = 10, 16
and 25. For εox = 10 , the oscillation frequency
decreases from 250 GHz to 215 GHz by increasing
CNT FET chirality form n = 7 to 28 (blue curve).
Increasing the dielectric constant results in a shift of
the whole curve to higher oscillation frequencies.
Until now, we have assumed a perfect electrode
contacts (Rc = 0) in the model shown in Fig. 4(a) to
provide a proof-of-concept for the chirality
dependence of oscillation frequency regardless of
total contact resistance (2Rc ) of a single-walled
CNT which has been reported different values
between 10 kΩ to 100 kΩ for Pt electrodes [26].
Figure 9 shows the frequency of 3-stage CNT-based
RO as a function of CNT FET chirality for
conducting single CNT interconnects with different
chiralities, in which we have simulated the
proposed RO assuming Rc = 24KΩ . Taking into
account contact resistances, results in an order of
magnitude reduction in the oscillation frequency.
However, the trend of frequency reduction as a
result of increasing the CNT FET chirality remains
the same as in the case of perfect electrode contacts
in the previous simulations. The figure shows the
effect of different chiralities of a conducting
SWCNT interconnect on the oscillation frequency.
Increasing the chirality of SWCNT interconnect has
the same influence on oscillation frequency as the
increase of the CNT FET chirality. The oscillation
frequency reduces from 25 GHz to 22 GHz for
increasing the CNT FET chirality from n=7 to 28
with the constant chirality of CNT
interconnect nin = 27 and with the length of
Lin = 1μ𝑚 between inverter stages. In the same
way, the oscillation frequency reduces from 25 GHz
to 20.3 GHz for increasing the chirality of SWCNT
interconnects from nin = 6 to 30 with the constant
chirality of CNT FET n=7. The reduction of
oscillation frequency has been estimated by ~10%
per 1 nm increase in diameter of 1μ𝑚 metallic CNT
interconnects.
CONCLUSION
We have designed and simulated a carbon nanotube-
based ring oscillator, where semiconducting and
metallic CNTs have been used as transistors and
interconnects between inverters, respectively. We
have investigated the chirality dependence of the
proposed ring oscillator using a numerical ballistic
model based on NEGF formalism for CNT FETs
with the gate length of 5 nm and a compact diffusive
model for CNT interconnects. We have found that
the oscillation frequency of all 3-, 5-, and 7-stages
CNT-based ring oscillator decreases by increasing
the chirality of both (n, 0) CNT FETs and (n, n)
CNT interconnects while the trends becomes more
linear by increasing the number of stages. The
frequency reduction has been estimated by ~8.5%
per 1 nm increase in diameter of zigzag CNT and
~10% per 1 nm increase in diameter of 1μm
metallic CNT interconnects. We have found that the
oscillation frequency is reduced by increasing
dielectric thickness and decreasing dielectric
constant leading to the lack of oscillation for high-
chirality CNTFETs. Assuming the contact resistance
of Rc = 24KΩ , the oscillation frequency of ~25
GHz has been obtained for 5 nm (7, 0) zigzag CNT
FETs and 1μm (6, 6) metallic CNT interconnects.
REFERENCES
[1] International Technology Roadmap for
Semiconductors: Emerging Research Devices,
Edition 2009.
[2] Shulaker, M. M., Van Rethy, J., Wu, T. F.,
Suriyasena Liyanage, L., Wei, H., Li, Z., ... &
Mitra, S. (2014). Carbon Nanotube Circuit
Integration up to Sub-20 nm Channel
Lengths. ACS nano, 8(4), 3434-3443.
[3] Graham, A. P., Duesberg, G. S., Hoenlein, W.,
Kreupl, F., Liebau, M., Martin, R., ... & Unger,
E. (2005). How do carbon nanotubes fit into the
semiconductor roadmap?. Applied Physics
A, 80(6), 1141-1151.
[4] Franklin, A. D., Luisier, M., Han, S. J.,
Tulevski, G., Breslin, C. M., Gignac, L., ... &
Haensch, W. (2012). Sub-10 nm carbon
nanotube transistor. Nano letters,12(2), 758-
762.
[5] Franklin, A. D., & Chen, Z. (2010). Length
scaling of carbon nanotube transistors. Nature
nanotechnology, 5(12), 858-862.
[6] Banadaki, Y. M., & Srivastava, A. (2013,
August). A novel graphene nanoribbon field
effect transistor for integrated circuit design.
In Circuits and Systems (MWSCAS), 2013
IEEE 56th International Midwest Symposium
on(pp. 924-927). IEEE.
[7] Srivastava, A., Banadaki, Y. M., & Fahad, M. S.
(2014). (Invited) Dielectrics for Graphene
Transistors for Emerging Integrated
Circuits. ECS Transactions,61(2), 351-361.
[8] Iijima, S. (1991). Helical microtubules of
graphitic carbon. nature, 354(6348), 56-58.
[9] Fregonese, S., Maneux, C., & Zimmer, T.
(2011). A compact model for dual-gate one-
dimensional FET: application to carbon-
nanotube FETs. Electron Devices, IEEE
Transactions on, 58(1), 206-215.
[10] Rao, R., Liptak, D., Cherukuri, T., Yakobson,
B. I., & Maruyama, B. (2012). In situ evidence
for chirality-dependent growth rates of
individual carbon nanotubes. Nature
materials, 11(3), 213-216.
[11] Koziol, K. K., Ducati, C., & Windle, A. H.
(2010). Carbon nanotubes with catalyst
controlled chiral angle. Chemistry of
Materials, 22(17), 4904-4911.
[12] Dass, D., Prasher, R., & Vaid, R. (2013).
Single Walled CNT Chirality Dependence for
Electrical Device Applications. The African
Review of Physics,8.
[13] Franklin, A. D., Koswatta, S. O., Farmer, D. B.,
Smith, J. T., Gignac, L., Breslin, C. M., ... &
Tersoff, J. (2013). Carbon nanotube
complementary wrap-gate transistors. Nano
letters, 13(6), 2490-2495.
[14] Mandal, M. K., & Sarkar, B. C. (2010). Ring
oscillators: Characteristics and
applications. Indian Journal of Pure & Applied
Physics, 48, 136-145.
[15] Docking, S., & Sachdev, M. (2004). An
analytical equation for the oscillation frequency
of high-frequency ring oscillators. Solid-State
Circuits, IEEE Journal of, 39(3), 533-537.
[16] Javey, A., Wang, Q., Ural, A., Li, Y., & Dai, H.
(2002). Carbon nanotube transistor arrays for
multistage complementary logic and ring
oscillators. Nano Letters, 2(9), 929-932.
[17] Chen, Z., Appenzeller, J., Solomon, P. M., Lin,
Y. M., & Avouris, P. (2006, June). High
performance carbon nanotube ring oscillator.
In Device Research Conference, 2006 64th (pp.
171-172). IEEE.
[18] Arefinia, Z., & Orouji, A. A. (2009). Quantum
simulation study of a new carbon nanotube
field-effect transistor with electrically induced
source/drain extension.Device and Materials
Reliability, IEEE Transactions on, 9(2), 237-
243.
[19] Guo, J., Datta, S., Lundstrom, M., & Anantam,
M. P. (2004). Toward multiscale modeling of
carbon nanotube transistors. International
Journal for Multiscale Computational
Engineering, 2(2).
[20] Datta, S. (2005). Quantum transport: atom to
transistor. Cambridge University Press.
[21] Pourfath, M., Kosina, H., & Selberherr, S.
(2008). Numerical study of quantum transport
in carbon nanotube transistors. Mathematics
and Computers in Simulation, 79(4), 1051-
1059.
[22] Deretzis, I., & La Magna, A. (2008). Electronic
transport in carbon nanotube based nano-
devices. Physica E: Low-dimensional Systems
and Nanostructures,40(7), 2333-2338.
[23] Chen, F., Joshi, A., Stojanović, V., &
Chandrakasan, A. (2007, September). Scaling
and evaluation of carbon nanotube
interconnects for VLSI applications.
In Proceedings of the 2nd international
conference on Nano-Networks (p. 24). ICST
(Institute for Computer Sciences, Social-
Informatics and Telecommunications
Engineering).
[24] Mohsin, K. M., Srivastava, A., Sharma, A. K.,
& Mayberry, C. (2013). A thermal model for
carbon nanotube
interconnects. Nanomaterials, 3(2), 229-241.
[25] Srivastava, A., Sharma, A. K., & Xu, Y.
(2010). Carbon nanotubes for next generation
very large scale integration
interconnects. Journal of Nanophotonics,4(1),
041690-041690.
[26] Pop, E., Mann, D. A., Goodson, K. E., & Dai,
H. (2007). Electrical and thermal transport in
metallic single-wall carbon nanotubes on
insulating substrates.Journal of Applied
Physics, 101(9), 093710.
[27] Burke, P. J. (2004, December). Carbon
nanotube devices for GHz to THz applications.
In Optics East (pp. 52-61). International
Society for Optics and Photonics.
[28] Das, S., Bhattacharya, S., & Das, D. (2011,
November). Modeling of carbon nanotube
based device and interconnect using
VERILOG-AMS. In Advances in Recent
Technologies in Communication and
Computing (ARTCom 2011), 3rd International
Conference on (pp. 51-55). IET.
[29] Sharifi, M. J., & Banadaki, Y. M. (2009,
August). A SPICE large signal model for
resonant tunneling diode and its applications.
In COMPUTATIONAL METHODS IN
SCIENCE AND ENGINEERING: Advances in
Computational Science: Lectures presented at
the International Conference on Computational
Methods in Sciences and Engineering 2008
(ICCMSE 2008) (Vol. 1148, No. 1, pp. 890-
893). AIP Publishing.
[30] Sharifi, M. J., & Banadaki, Y. M. (2010).
General SPICE models for memristor and
application to circuit simulation of memristor-
based synapses and memory cells. Journal of
Circuits, Systems, and Computers, 19(02), 407-
424.