See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224651367

Analysis of CNTFET physical compact model

Conference Paper · October 2006

DOI: 10.1109/DTIS.2006.1708733 · Source: IEEE Xplore

CITATIONS READS
34 354

6 authors, including:

C. Maneux S. Fregonese
University of Bordeaux French National Centre for Scientific Research
196 PUBLICATIONS   1,279 CITATIONS    203 PUBLICATIONS   1,448 CITATIONS   

SEE PROFILE SEE PROFILE

T. Zimmer H. Cazin d’Honincthun

University of Bordeaux Institute of Fundamental Electronics
338 PUBLICATIONS   2,135 CITATIONS    12 PUBLICATIONS   213 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

ANR-Moorea View project

FP7 Grade View project

All content following this page was uploaded by C. Maneux on 22 May 2014.

The user has requested enhancement of the downloaded file.

Analysis of CNTFET physical compact model
C. Maneux1, J. Goguet1, S. Frégonèse1, T. Zimmer1,
H. Cazin d'Honincthun2, S. Galdin-Retailleau2
1
IXL - CNRS UMR 5818 - ENSEIRB-Université Bordeaux 1, 351 cours de la Libération
F-33405 TALENCE Cedex
2
IEF - CNRS UMR 8622 - Bât 220, Université Paris-Sud, Centre d'Orsay,
F-91405 Orsay Cedex
Corresponding author e-mail: maneux@ixl.fr

Abstract— On the basis of acquired knowledge, we present a scattering mechanisms (phonons, impurities) have been
DC compact model designed for the conventional CNTFET (C- theoretically evaluated in a reduced number of cases. A
CNTFET) featuring a doping profile similar to n-MOSFET. second serious difficulty is the experimentally observed
The specific enhancement lies on the implementation of a dispersion of device characteristics. This dispersion is
physical based calculation of the minima of energy conduction generally assigned to the variability of metal/nanotube
subbands. This improvement allows a realistic analysis of the
interfaces at the molecular scale. So, to obtain predictive
impact of CNT helicity and radius on the dc characteristics.
The purpose is to enable the circuit designers to challenge circuit simulation results, it is mandatory to precisely
CNTFET potentialities for performing logical or analogical understand transport phenomena in CNTFET at the
functionalities within complex circuits. molecular scale. Hence, considering a single walled,
semiconducting carbon nanotube as the channel of a CNT-
Keywords— Carbon nanotubes FET (CNTFET), ballistic FET including source, drain, gate electrodes, the circuit-
1-D theory, compact modeling. compatible model should describe the transistor one-
dimensional (1-D) electrostatics in its ballistic limit of
I. INTRODUCTION performance. Moreover, for convenient and efficient circuit
simulation, the circuit-compatible model has to be suitable
C ARBON nanotubes are promising candidates for
passive or active elements in post-CMOS
nanoelectronics. During the last decade, the group of
for a wide range of CNTFETs diameters ranging from 0.6 to
3 nm and for varied chiralities as long as they are
semiconducting. Although such model derives from
Ph. Avouris brought experimental evidences that CNTFETs
required approximations and simplifications to develop
could have better performances than ultimate silicon
analytical expressions, strong foundation of the underlying
MOSFETs [1]. Similar results were obtained in the group of
physics of operation [3] determine the accuracy of the
H. Dai [2]. For now, many experimental groups are
circuit transfer (dc) characteristics together with the
studying CNT based devices, covering various topics:
transient response.
- the influence of the metal used for the contacts to
This paper described the DC operation of a ballistic 1-D
control charge injection efficiency,
CNTFET compact model. Section II describes the carbon
- their frequency dependent behaviours,
nanotube within a transistor configuration and the equations
- their new promising functionalities.
associated with its intrinsic operation. The implementation
However, to predict the ultimate performances of these
of an accurate physical based calculation of the minima of
novel nanodevices, and to further offer guidance and cost
energy conduction subbands is consistent with an
reduction of the technological development, accurate and
improvement of the description of the total charge on the
reliable simulation tools appears as key issues. Indeed, as
nanotube. As this charge induces a voltage drop across the
the micro-electronics companies show an increasing
gate insulator, it directly impacts on the dc characteristics.
concern with post-CMOS technologies, there is a strong
Section III presents an analysis of the dc characteristics as a
demand for simulation tools. In particular, companies that
function of the total charge on the CNT defined according
develop circuit simulation softwares try to include
to its helicity and radius.
modelling of nanodevices-based circuits in their available
products. For CNT devices, as well as for other molecular
electronics devices, it is necessary to develop new models,
since the standard approximations and models used for
MOS electronics may loose their applicability range. A first
difficulty arises from the fact that intrinsic transport
properties of CNTs are much less understood than their bulk
semi-conductors counterparts, widely used in
microelectronics. The energy band structure strongly
depends on the nanotube helicity and radius, and the

  

 II. DESCRIPTION OF CNTFET -- Depending on the SWNT helicity and radius, the
periodic boundary conditions impose restrictions on
available states [10], which results in a discrete set of
A. Carbon nanotube within a transistor configuration energy subband structure (Fig. 2b)).

For now, the transistor configurations based on single- E

walled carbon nanotubes (SWNTs) are various: the
Schottky barrier CNTFET (SB-CNTFET) [4], the
conventional CNTFET (C-CNTFET) featuring a doping
profile similar to n-MOSFET [5], the dual-gate CNTFET
µS
exhibiting n- or p-type unipolar behavior tunable by qVDS a)
electrostatic doping [6] and the tunneling CNTFET [7]. The
subject of this paper is the conventional CNTFET. Its µD
structure albeit simple provides a behavior like normal
MOSFET with yet ballistic transport [8]. In this device,
depicted in Fig. 1a), the ungated regions (source and drain k
regions) are heavily doped and it operates on the principle -qVCNT
of barrier-height modulation by application of the gate
potential (Fig. 1b)). 8
6
Gate 4 ∆3
S D ∆2
a) 2 ∆1
E (eV)
High K
0
n+CNT
Gate
b)
-2
G -4
VGS=0 -6
Ec Increase VGS -8
S D -1 -0.5 0 0.5 1
EG/2
µS b) k

Fig. 2. The SWNT energy versus wave-number (E-k) diagram. a) the

+k -k source and drain Fermi level have been shown by µ S and µ D. b) ∆p with
qVds p=1,2,3,… are the minima of the energy subbands.
µD
B. Model description of the C-CNTFET
Fig. 1. Conventional CNTFET, C-CNTFET.a) Schematic device features
with a high K dielectric. b) The band diagram with, at VG=0V, the barrier As VG, the gate bias, modulates the top of the energy
height at the source-channel junction equal to EG/2.
barrier; it lowers the channel potential by an amount VCNT
and causes accumulation of charge in the channel, QCNT.
In this case, the on-state current is limited by the amount
This charge QCNT induces a voltage drop VG-VCNT across
of charge introduced in the channel by the gate and, for
the high K insulator which causes the energy bands to be
source-drain distances shorter than 150nm, the transport is
lowered by VCNT. This self consistent loop in C-CNTFET
assumed to be free of significant scattering and thus
operation can not be handled in a circuit-compatible model,
essentially ballistic at both high and low voltages. Hence,
necessarily based on a linear analytical development.
the description of current flow through the CNT lies on (i)
Hence, the compact model proposed by A. Raychowdhury
the features of ballistic transport and (ii) the specific
[5] lies on the calculation of the channel potential VCNT
electron confinement along the tube circumference [9]:
through fitting parameters of the channel charge, QCNT.
-- Since the current remains constant throughout the
So, before detailing our contribution to this model, a
channel, the current is calculated at the top of the energy
description of a few main equations is required.
barrier corresponding to the beginning of the channel.
-- At the top of the barrier, electrons coming from the
source fill up the +k states and the electrons coming from
the drain fill up the –k states (Fig. 2a)).
 The modulation of the channel potential, VCNT through where e is the electron charge, kB is the Boltzmann
the gate bias, VG, results from the presence of the charge in constant, h is the Planck constant and T the temperature.
the channel, QCNT: Since it is not possible to obtain an analytical closed-
Q CNT form expression for the integral given in (2), A.
VCNT = VGS − (1) Raychowdhury has proposed the following fitting
C
1 1 1 expression:
where = + , COX being the gate capacitance
C C OX C Q
VCNT = VGS for VGS < ∆1
and CQ the quantic capacitance. (4)
= VGS − α(VGS − ∆1 ) for VGS ≥ ∆1
The total charge on the nanotube, QCNT depends on the
number of carrier in the channel, nCNT which results from
2
the specific electron confinement along the tube with α = α 0 + α1 VDS + α 2 VDS , where α0, α1 and α2 are
circumference mentioned above [11][12]. As a fitting parameters depending on the gate capacity and the
consequence, its expression is the sum of the energy radius and helicity of the CNT. As an example, for a gate
subband contributions: capacity of 17pF/m and a radius CNT of 3nm, α 0 = 0,31 ,

 α 1 = −0,36V −1 , α 2 = 0,1V −2 .
+∞ 
+∞ g0  1
n CNT = ∑ ∫0
p =1 2 
 1 + exp
2
(
z + ∆ p − VCNT )2 C. Calculation of subband minima, ∆p
 k BT

(2) The structure of a CNT can be seen as the result of the

 rolling up of a graphene sheet [10] with the width defined
 G
1  dz by a lattice vector C (Fig. 3). The rolling up is
+
(
z 2 + ∆ p − VCNT )2 + VDS 

geometrically characterized by the indices n and m which
1 + exp  specify the diameter and the helicity angle of the CNT and
k BT  determine its fundamental properties (5).
with
8 G G G
- g0 = [12], where a is the distance C = n a1 + m a 2 (5)
3πaVppΠ
between 2 carbon atoms (around 0,142 nm),
VppΠ is the energy between carbon atom Nanotube
nearest-neighbour (2,5 eV),
- ∆ p is the minima of the pth energy subband.

- z given by z 2 + E C2 − E C = E , where E is he
electron energy referenced from EC which is
replace by E Cp = ∆ p − VCNT since (i) all energy
a1
subbands are considered and (ii) the energy
subbands are lowered by VCNT.
T
a2 C
The drain current equation derives from the Landauer
[10] formula which describes the ballistic transport with
perfect contacts. Its expression represents the sum of the
energy subband contributions of two terms. The first term
represents the occupation of the +k states from the source
up to µS and the other represents the occupation of the –k G
Fig. 3. Rolling up an infinite strip of graphene with width C . The lattice
states from the drain up to µD; the difference between µS
G
and µD results from the bias VDS applied between source vector C defines uniquely the nanotube. The unit cell of the tube is
and drain (Fig. 2a)): G K
defined by vector C and the primitive translation T of the nanotube.
  − ∆ P + VCNT  
 ln1 + exp  
4ek B T +∞   k BT  The rolled-up nanotube is a one-dimensional periodic

I= ∑ 
h p =1   (3) structure with a unit cell which is a rectangle defined by the
G G
 − ln1 + exp − VDS − ∆ P + VCNT  chirality vector C and the primitive translation T . The 2D-
  k BT 
  Brillouin zone of the nanotube unit cell is a rectangle in the
G G
reciprocal space defined by the vectors C' and T'
 G G G G G
(with C C' = 1 , C T' = 0 ) as depicted in Fig. 4. In the By stating K in the reciprocal space, the equation (8)
G can be written:
nanotube, the values of the electron wave-vector, K are
restricted additionally by the rotational boundary condition:
E (k1 , k 2 ) = VppΠ 1 + 2 cos(2πk 1 ) + 2 cos(2 πk 2 ) + 2 cos(2 π(k 1 − k 2 ))
G G
K C = 2 πq (6) (9)

where q=0, 1, 2, …, N-1, with N the number of carbon where k1 and k 2 have the following expressions:
pairs in the unit cells. q  2n + m 
- k1 =  ,
N  d R 
N=
(
2 * n2 + m2 + n m ) (7) q  2m + n 
dR - k2 =  .
N  d R 
with d R = gcd(2 m + n, 2 n + m) .
G Hence, the values of E (k1 , k 2 ) are successively
Therefore, the allowed values of K lie on N equidistant calculated using (9). Then, the lowest values are ordered to
lines parallel to the tube axis (dotted lines in Fig. 4). select the minima of the 1st, 2nd,…, pth energy subbands, i.e.
b1 ∆1, 2,...,p .

T’ C’ D. CNTFET Parasitic mechanisms

A common feature of CNTFETs fabricated to date is the

b2 presence of a Schottky barrier at the nanotube–metal
junctions. These energy barriers severely limit transistor
Tube axis conductance in the ‘ON’ state, due to the non-ideal
Fig. 4. Graphene Brillouin zone. The allowed values of the electron wave- electrical contacts of source and drain electrodes (Fig. 5).
vector in a nanotube lie on lines parallel to the tube axis (shown in dotted Thermionic emission and tunnelling, involved in transport
lines). across the SBs, are seen as parasitic mechanisms and, on the
whole, are modelling through series resistances lumped as
The band structure of the rolled-up nanotube can be RS and RD. Their values, roughly a few tens of kΩ, greatly
obtained by zone-folding the band structure of the graphene depend on the SWNT diameter and on the metal work
sheet. The zone-folding method used in this work is similar function [16] [17] [18].
to that presented in [14] extended to all chiralities and
diameters. It gives reasonable predictions for the band
structure for radii larger than 6 Å [15]. A convenient
analytical approximation is proposed to describe the minima Intrinsic (off-state)
energy of the subbands of semiconducting CNTs.

G G G G G G G
Thermionic emission
() G
E K = VppΠ 1 + 2 cos K(a 1 − a 2 ) + 2 cos K (a 1 ) + 2 cos K (a 2 )
(8) Tunnelling

where G G
G G
- a 1 and a 2 are defined using i and j orthogonal Intrinsic and Extrinsic
vectors: (on-state)
3 1
a1 = − ai + a j ,
2 2 Fig. 5. Conduction and Valence energy band diagram considering the on-
3 1 state and the off-state for the intrinsic CNT and the extrinsic CNT.
a2 = − ai − a j ,
2 2
G
- K takes successive values corresponding to the
G
condition stated in (6) where C is defined using
(5) with the n and m parameters.
 E. Model implementation ID (µA)
175
Finally, the compact model of the 1-D CNTFET with the 150
VDS = 0,8 V 5 subbands
4 subbands
implementation of the minima subbands calculation leads to 125 3 subbands
the flow diagram presented Fig. 6. It has been implemented
100 2 subbands
in ADS 2004A simulation tool using VerilogA as
description language. This program calculates the drain 75 1 subband

current for given VGS and VDS of an n- type C-CNTFET 50

including access serial resistances. Its field application is 25

restricted to static regime operation: capacitances are not 0
taken into account. The gate current is null so the serial gate 0 0.2 0.4 0.6 0.8 1 1.2

resistance is not taken into account. It allows to simulate a VGS (V)

device with nanotube diameter ranging from 0,8 to 3 nm Fig. 7. Drain current of a conventional CNTFET featuring n = 10, m = 11
where the SWNT helicity and radius are taken into account (diameter equal to 1.4246 nm), with low series resistances.
through n and m chirality vector parameters. Hence, the
minima subbands are pre-computed and it allows estimating When taking into account the series resistances, equal to
the number of subbands required. The parameters α0, α1 and 25kΩ (commonly used), four subbands are enough to
α2 are data enter parameters. describe the drain current for gate bias and drain bias lower
than 1V. With these conditions (Fig. 8), the drain current, ID
deceases from 85% for the biases previously mentioned
(VGS=1V, VDS=0.8V). A further decrease is observed when
the total charge on the nanotube, QCNT is introduced through
the fitting expression of α (dashed line). Indeed, the more
VGS increases, the more the influence of the total charge on
the nanotube is important and the more the drain current
decreases compared with its value calculated without taking
into account the charge.
ID (µA)

VDS =
25 ---- With charge
Fig. 6. Calculation steps for the analytical model. Through n and m 1,3V
____ Without charge
parameters, the SWNT helicity and radius are enter data of the model. The 20 1,1V
ID (VGS,VDS) characteristics are obtained thanks to fitting parameters (α0, 0,9V
α1, α3) of the charge into the channel which determine the channel 15
0,7V
potential VCNT. 10 0,5V
5 0,3V
0,1V
III. ANALYSIS OF DC CHARACTERISTICS 0
0 0,2 0,4 0,6 0,8 1
VGS (V)
For a conventional CNTFET featuring n=10, m=11
(diameter equal to 1.4246nm) and an insulator capacity of ID (µA)
VGS =
17pF/m, the fitting parameters are α0=0.62, α1=0.275V-1, ---- With charge 1,1V
α2=0.015V-2. 20 ____ Without charge
0,9V
Fig. 7 shows the drain to source current, ID as a function 15
of the applied gate bias, VGS for the contribution of 1, 2, .., 5 0,7V
subbands. Once the gate bias exceeds 0,7V, the curves are 10

no more alike, denoting the filling up of the second and 5

0,5V
more subbands and contributing to the total drain current 0,3V
increase. Note that, when taking into account the three 0 0,1V
subbands, IDS increases from 40% for VGS=1V compared 0 0,2 0,4 0,6 0,8 1 1,2

with the first subband. Thus, for this diameter and chirality, VDS (V)
the second and third subband contributions are significant. Fig. 8. A conventional CNTFET featuring n=10, m=11 (diameter equal to
However, as this result is obtained with a low value of the 1.4246 nm), with series resistances equal to 25kΩ :
Up : Drain current versus gate bias with drain bias as a parameter.
series resistance, this instance corresponds to a future Down : Drain current versus source-drain bias with gate bias as a
mature technology. parameter.
 Fig. 9 shows the ID-VDS for VGS equal to 1V and 0.5V. REFERENCES
The simulated C-CNTFETs feature three diameters (1, 1.4 [1] Ph. Avouris, MRS Bull. 29, 403, 2004.
and 1.8nm) and, for each diameter, three different chiralities [2] A. Javey, J. Guo, Q. Wang, M. Lundstrom, H. Dai, Nature 424, pp.
has been calculated (Zig-zag, near-armchair and chiral). The 654-657, 2003.
[3] S. Fregonese, G. Avenier, C. Maneux, A. Chantre, T. Zimmer, “A
corresponding n and m parameters are summarized in the compact model for HBT on thin film SOI,” IEEE Trans. Elec. Dev.,
table inset in the figure. The simulation has taken into vol. 53, N° 2, Feb 2006.
account the series resistances equal to 25kΩ. [4] J. Appenzeller, J.Knoch, M. Radosavljevic, Ph. Avouris,
For VGS equal to 1V as for 0.5V, we observed that the “Multimode transport in Schottky-barrier carbon-nanotube field-
effect transistors,” Phys. Rev. Lett., Vol. 92, N° 22, pp. 226802-1-
SWNT diameter has a first order impact on the drain current 226802-4 2004.
value. For example, for VGS=0.5V and VDS=1V, ID ≅ 2.5, 6.5 [5] A. Raychowdhury, S. Mukhopadhyay, K. Roy, “A Circuit-
and 8.5mA for diameters=1, 1.4 and 1.8nm respectively. Compatible Model of Ballistic Carbon Nanotube Field-Effect
The increase of the drain current with respect to the Transistors,” IEEE Transactions on Computer-Aided Design of
Integrated Circuits and Systems, vol. 23, N° 10, pp. 1411-1420,
diameter is directly associated with the decrease of the NT 2004.
band gap energy with its diameter [12]. The helicity angle [6] Yu-Ming Lin, J. Appenzeller, J. Knoch, Ph. Avouris, “High-
of the CNT, from zig-zag to near armchair, has also a performance carbon nanotube field-effect transistor with tunable
significant impact on the drain current, albeit less important polarities,” IEEE Trans. On Nanotech., Vol. 4, N° 5, 2005.
[7] J. Appenzeller, Yu-Ming Lin, J. Knoch, Z. Cheng, Ph. Avouris,
than the diameter one. For example, for a SWNT diameter “Comparing carbon nanotube transistors – The ideal choice : A novel
equal to 1nm and for VGS=0.5V and VDS=1V, ID ≅ 2.5 for tunnelling device design,” IEEE Trans. Elec. Dev., Vol. 52, N° 12,
zig-zag SWNT and ID ≅ 3mA for a chiral one. 2005.
[8] S.J. Wind, J. Appenzeller, Ph. Avouris, “Lateral scaling in carbon-
Id (µA) nanotube field-effect transistors,” Phys. Rev. Lett., vol. 91, N°5, pp.
058301-1-058301-4, Aug. 2003.
Diameter 1nm 1.4nm 1.8nm
Chirality n-m n-m n-m [9] Ph. Avouris, J. Appenzeller, R. Martel, S. J. Wind, “Carbon nanotube
20 Zig-zag 0-13 0-17 0-23 Electronics,” Proceedings of the IEEE, vol. 91, pp. 1772-1784, 2003.
Chiral
Near armchair
3-11 5-15
10-11
8-18 [10] R. Saito, M.S. Dresselhaus, G. Dresselhaus, Physical properties of
7-8 13-14
Vgs = 1 V
carbon nanotubes (Imperial College Press, London, 1998).
15 [11] S. Datta, Electronic Transport in Mesoscopic Systems, Cambridge
Studies in Semiconductor Physics and Microelectronic Engineering,
vol. 3, Cambridge University Press, Shaftesbury Road, Cambridge,
10
Vgs = 0,5 V CB2 2RU, 1997.
[12] S. Datta, Quantum transport – Atom to transistor, Cambridge
University Press, Shaftesbury Road, Cambridge, CB2 2RU, 2005.
5 [13] J. Guo, M. Lundstrom, S. Datta, “Performance projections for
ballistic carbon nanotube field-effect transistors,” Applied Physics
Letters, vol. 80, N° 17, pp. 3192-3194, 2002.
0 [14] H. Cazin d’Honincthun, S. Galdin-Retailleau, J. Sèe, P. Dollfus,
0 0.2 0.4 0.6 0.8 1 1.2
VDS (V) “Electron-phonon scattering and ballistic behavior in semiconducting
Fig. 9. Drain current versus source-drain bias with gate bias as a carbon nanotubes,” Applied Physics Letters 87, 172112, 2005.
parameter. The C-CNTFETs feature three diameters (1, 1.4 and 1.8nm) and [15] V. N. Popov and L. Henrard, Phys. Rev. B 70, 115407, 2004.
three different chirality for each diameter (Zig-zag, near-armchair and [16] Ali Javey, Jing Guo, Qian Wang, Mark Lundstrom and Hongjie Dai,
chiral) ing n=10, m=11, with series resistances equal to 25kΩ. “Ballistic carbon nanotube field-effect transistors,” Nature , Vol. 424,
7 Aug. 2003.
[17] A. Javey, J. Guo, D. B. Farmer, Q. Wang, D. Wang, R. G. Gordon,
M. Lundstrom, H. Dai, “Carbon Nanotube Field-Effect Transistors
IV. CONCLUSION with Integrated Ohmic Contacts and High-K Gate Dielectrics,” Nano
Letters, vol. 4, N° 3, pp. 447-450, 2004.
From the DC operation of a C-CNTFET compact model, we [18] C. Zhou, J. Kong, H. Dai, “Electrical measurements of individual
have implemented a physical based calculation of the semiconducting single-walled carbon nanotubes of various
minima of energy conduction subbands. This calculation, diameters,” Applied Physics Letters, vol. 76, N° 12, pp. 1597-1599,
obtained from a zone-folding method, is pre-computed 2000.
within the model flow diagram. The analysis of the related
simulated dc characteristics has shown that the NT diameter
and helicity angle have a significant impact on the current
drain values although not at the same proportion: the former
being more important.

View publication stats