CARBON NANOTUBES: DEVICE PHYSICS, RF CIRCUITS,
SURFACE SCIENCE, AND NANOTECHNOLOGY
A DISSERTATION
SUBMITTED TO THE DEPARTMENT OF ELECTRICAL ENGINEERING
AND THE COMMITTEE ON GRADUATE STUDIES
OF STANFORD UNIVERSITY
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
Deji Akinwande
November 2009
http://creativecommons.org/licenses/by-nc/3.0/us/
This dissertation is online at: http://purl.stanford.edu/hk621tj0645
© 2010 by Deji A Akinwande. All Rights Reserved.
Re-distributed by Stanford University under license with the author.
This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 United States License.
ii
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Philip Wong, Primary Adviser
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Robert Dutton
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Yoshio Nishi
Approved for the Stanford University Committee on Graduate Studies.
Patricia J. Gumport, Vice Provost Graduate Education
This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.
iii
iv
Abstract
In this dissertation, we report on our research to advance the understanding and
development of carbon nanotubes into a functional nanotechnology that will enable
novel applications and products. Our research approach is interdisciplinary in nature
involving progress on many fronts including material synthesis, device physics and
compact modeling, circuits, and monolithic integration with silicon substrates for
optimum performance and integration. Specifically, this dissertation reports new
results advancing the science and technology of carbon nanotubes including: i)
analytical development of the physics of CNTs providing direct insight into the
electro-physical properties, ii) experimentally-verified analytical current-voltage and
capacitance-voltage characteristics of CNTs facilitating circuit design and analysis, iii)
elucidation of the material science of nanotube synthesis enabling large-scale
synthesis of aligned growth, and iv) monolithic integration of CNTs and CMOS for
hybrid nanotechnology for future nanoelectronics.
v
Acknowledgement
The great scientist Isaac Newton remarked that he had seen farther only
because he stood on the shoulders of giants. In my case, I have completed this
marathon with satisfaction only because I was coached by the very best trainers. In
this light, the role of my primary trainer and thesis supervisor Professor H.-S. Philip
Wong in supporting and refining my research ideas and publications cannot be
overstated. It is the ideal dream of many to define the research directions of personal
interest, and have the freedom to pursue such defined scholarship at their pleasure and
discretion of time, pace, and place without any worries of research funding. I was very
fortunate to realize this dream courtesy of a blank check from Prof. Wong.
Similarly Professor Yoshio Nishi has played a profound role in enabling not
only my research but also those of my nanotube colleagues that came after me by
granting me unfettered access to the 4” carbon nanotube furnace in his lab starting
from the autumn of 2006. In the beginning, it was very difficult to grow nanotubes by
the CVD method. With great persistence over time, that nanotube furnace has now
produced some of the finest results in the field.
Professor Bob Dutton has been in some way my unofficial mentor providing
me with a global perspective and critique of my nanotube research from time to time
thereby inadvertently forcing me to have a much clearer vision of the big picture.
Additionally, his advice and perspective on academic careers were invaluable in my
successful job search.
Stanford University boasts some of the best teachers and intellectual thinkers.
I have been very fortunate to have touched the cloak of these teachers and increased
vi
my intelligence by several tens of dB. The most influential ones in my intellectual
development and education were Professors Jim Plummer (device fabrication), Brad
Osgood (Fourier transform), Tom Lee (RF circuits), Boris Murmann (circuit design),
and Hari Manoharan (solid-state physics). My deep gratitude to them and all my other
teachers for essentially teaching me how to think analytically.
No complex device research is possible without a supporting cast of staff
members. Dr. Jim McVittie has always been very generous with his time and broadly
helpful in many areas ranging from plasmas to vacuum systems of which I had no
prior training. I am also indebted to the staff of the Stanford nanofabrication facility
(SNF) and Stanford nanocharacterization laboratory (SNL). Key SNF staff members
include James Conway (E-beam), Paul Jerabak (retired), and Mahnaz Mansourpour
(Litho). SNL staff members include Chuck Hitzman (XPS), and Bob Jones (SEM). I
also want to thank Tom Carver (iron evaporation) and Pauline Prathers (wire bonding)
of Ginzton lab for their routine assistance and excellence.
Several funding agencies have played a direct role in supporting my graduate
education and research. My personal thanks goes out to my fellowship partners
including Ford foundation, Alfred P. Sloan foundation, and the Stanford future-faculty
DARE fellowship. Dr. Lozano of the School of Engineering was instrumental in
helping me secure both the Ford and Alfred P. Sloan fellowships. Research was
supported in part by the Focus Center Research Program (FENA and C2S2), and the
Toshiba Corporation.
Lastly, I acknowledge my colleagues in the nanotube club for free exchange of
stimulating ideas and extended collaboration. The club members include Gael Close,
Cara Beasley, Nishant Patil, Albert Lin, Lan Wei, Helen Chen, Jiale Liang, Arash
Hazeghi, and Jerry Zhang. Personal thanks also extend to Yuan Zhang and Sangbum
Kim who have been my Yes men on every request I ask of them.
I have truly enjoyed my time at Stanford, the beauty and cool of the campus,
the lovely artwork in CIS, and the long walks across the quad. Recreational activities
such as seminars, Ginzton happy hours and cromem study breaks, soccer, horse-back
riding, and music made my living experience very memorable.
vii
Dedicated To My Beloveth:
- Rythmm, David, and the blessed clan of gentle hearts
- Mommy and Daddy, the unspoken words carry the most weight
viii
Contents
Chapter 1 : Introduction..................................................................................................1
1.1 Brief Introduction to Carbon Nanotubes ................................................................................1
1.2 Applications of Carbon Nanotubes.........................................................................................2
1.3 Description of Chapters..........................................................................................................6
Chapter 2 : Physical and Electronic Structure................................................................7
2.1 Chirality: A Concept to Describe Nanotubes .........................................................................9
2.2 The Carbon Nanotube Lattice ..............................................................................................11
2.3 Carbon Nanotube Brillouin Zone .........................................................................................19
2.4 Tight-Binding Dispersion of Chiral CNTs ...........................................................................21
2.5 Band Structure of Armchair Nanotubes ...............................................................................23
2.6 Band Structure of Zigzag Nanotubes and the Derivation of the Bandgap............................26
2.7 Conclusion............................................................................................................................30
Chapter 3 : Device Physics...........................................................................................31
3.1 Preface..................................................................................................................................31
3.2 An Analytic Derivation of the Density of States, Effective Mass and Carrier Density for
Achiral Carbon Nanotubes..............................................................................................................32
3.2.1 Introduction.................................................................................................................32
3.2.2 Density of States .........................................................................................................34
3.2.3 Group Velocity and Effective mass ............................................................................41
3.2.4 Carrier Density............................................................................................................43
3.2.5 Conclusion ..................................................................................................................51
3.3 Analytical Ballistic Theory of Carbon Nanotube Transistors: Experimental Validation,
Device Physics, Parameter Extraction, and Performance Projection ..............................................55
3.3.1 Introduction.................................................................................................................56
3.3.2 Analytical Ballistic Theory .........................................................................................57
3.3.3 Analytical Surface Potential .......................................................................................61
ix
3.3.4 Experimental Validation and Device Physics .............................................................62
3.3.5 Parameter Extraction and Performance Projection .....................................................68
3.3.6 Conclusion ..................................................................................................................75
3.4 References ............................................................................................................................77
Chapter 4 : RF Analog Circuits ....................................................................................82
4.1 Preface..................................................................................................................................82
4.2 Analysis of the Frequency Response of Carbon Nanotube Transistors................................83
4.2.1 Introduction.................................................................................................................83
4.2.2 Overview of Prior RF Characterization ......................................................................84
4.2.3 CNFET Transfer Function (S21)..................................................................................85
4.2.4 Effect of Pad Capacitance ...........................................................................................94
4.2.5 Design Guidelines and Layout Issues for High Frequency CNFET ...........................95
4.2.6 Conclusion ................................................................................................................102
4.3 Carbon Nanotube Quantum Capacitance for Non-Linear Terahertz Circuits.....................103
4.3.1 Introduction...............................................................................................................103
4.3.2 CNT Capacitor Circuit Model...................................................................................104
4.3.3 Quantum Capacitor Analog Circuits.........................................................................109
4.3.4 Conclusion ................................................................................................................117
4.4 References ..........................................................................................................................118
Chapter 5 : Material Synthesis and Surface Science ..................................................121
5.1 Preface................................................................................................................................121
5.2 Introduction ........................................................................................................................123
5.3 Experimental Methods........................................................................................................124
5.4 Results and Discussion .......................................................................................................126
5.4.1 Surface Physics .........................................................................................................128
5.4.2 Surface Chemistry.....................................................................................................130
5.4.3 Illustrative Surface Science Model and Full Wafer Synthesis ..................................137
5.5 Conclusion..........................................................................................................................140
5.6 References ..........................................................................................................................141
Chapter 6 : Silicon – CNT Hybrid Nanotechnology ..................................................144
6.1 Preface................................................................................................................................144
6.2 Introduction ........................................................................................................................146
6.3 Hybrid Monolithic Chip .....................................................................................................149
6.4 Conclusion..........................................................................................................................155
6.5 References ..........................................................................................................................156
Closing Remarks ........................................................................................................158
x
7.1 Research Contributions ......................................................................................................158
7.2 List of Publications.............................................................................................................159
7.3 Future Work .......................................................................................................................162
xi
Figures
Figure 1.1: Growth in CNT articles in the publications of the major scientific societies. .........................3
Figure 1.2: United States patents issued and patent applications containing the phrase carbon nanotubes
in the patent abstract. ..............................................................................................................5
Figure 2.1: Illustration of the two families of carbon nanotubes. ..............................................................8
Figure 2.2: Example of a chiral object.....................................................................................................10
Figure 2.3: The three types of single-wall CNTs.....................................................................................12
Figure 2.4: An illustration to describe the construction of a CNT from graphene. .................................14
Figure 2.5: Brillouin zone of a (3,3) armchair CNT (shaded rectangle)..................................................20
Figure 2.6: Band structures for a) (10,4) and b) (10,5) nanotubes within ±±±±3eV......................................24
Figure 2.7: Band structure for the (8,8) armchair nanotube. ...................................................................25
Figure 2.8: Band structures for a) (12,0) and b) (13,0) carbon nanotubes. ..............................................27
Figure 2.9: The bandgap of semiconducting carbon nanotubes calculated using Eq. (23) (solid line)
compared to exact NNTB computation showing good agreement........................................29
Figure 3.1: An illustrative energy bandstructure of a semiconducting CNT. ..........................................37
Figure 3.2: The effective mass of the first conduction band....................................................................44
Figure 3.3: Illustrative plot of the DOS and Fermi-Dirac distribution at equilibrium. ............................47
Figure 3.4: The semiconducting zigzag CNT intrinsic carrier density. ...................................................49
Figure 3.5: The semiconducting zigzag CNT carrier density. .................................................................50
Figure 3.6: CNFET devices covered in this work illustrated with an n-type...........................................59
Figure 3.7: a) Two subband E-k dispersion of a CNT.............................................................................60
Figure 3.8: Experimental results and analytical (Eq. (4)) predictions for a ballistic 50nm p-type CNFET
(device identical to Figure 3.6b). ..........................................................................................63
Figure 3.9: Illustration of DOPS in a ballistic (n-type) CNFET..............................................................64
Figure 3.10: Improved agreement between experiment and analytical ballistic theory...........................66
Figure 3.11: VT exactly computed from ABT model (Table 3.1, Eq. (3))...............................................69
Figure 3.12: Deep subthreshold current and linear fit for extracting the CNT bandgap..........................71
Figure 3.13: Performance assessment for ballistic CNFET for logic applications. .................................73
xii
Figure 3.14: ABT model calculations for key analog metrics for a 50nm CNFET. ................................74
Figure 4.1: CNFET simplified small signal common-source model. ......................................................86
Figure 4.2: Illustrated Bode plots of four possible CNFETs frequency response. ..................................93
Figure 4.3: Representative layouts of a carbon nanotube transistor. .......................................................97
Figure 4.4: Frequency response of the gain of common-source CNFETs plotted for several values of n.
Gate width (We) is 5µm.........................................................................................................99
Figure 4.5: f-3dB of the common-source CNFET (with n=1000) plotted as a function of the gate
width/finger for the same dimensions as in Table 4.3.........................................................101
Figure 4.6: Optimum device geometry for CNT quantum capacitance circuits. ...................................106
Figure 4.7: Two-port circuit models for a CNT quantum capacitance device.......................................107
Figure 4.8: Measured NLC capacitance (including the quantum, Cq, and geometric, Cg, contributions) vs
top-gate voltage of a CNT device with source/drain shorted [26]. .....................................110
Figure 4.9: Proposed carbon nanotube quantum capacitance frequency doubler circuit. ......................114
Figure 4.10: Proposed carbon nanotube quantum capacitance frequency mixer circuit........................116
Figure 5.1: Images and device data from initial CNT synthesis. ...........................................................127
Figure 5.2: AFM surface scans of iron catalyst on single-crystal quartz...............................................129
Figure 5.3: a) XPS survey scan after sub-monolayer Fe evaporation on quartz sample........................131
Figure 5.4: a) Si 2p XPS core level spectra after sub-monolayer Fe evaporation. ................................133
Figure 5.5: XPS comparison of initial and optimized pretreatment procedures. ...................................136
Figure 5.6: A proposed illustrative model of the catalytic surface physics and chemistry during the
catalyst pretreatment procedure. .........................................................................................138
Figure 5.7: SEM images of CNT grown on full 4” quartz wafer...........................................................139
Figure 6.1: The distribution of aligning semiconducting nanotubes (sCNTs), and metallic nanotubes
(mCNTs) to lithographic defined source and drain pads. ...................................................148
Figure 6.2: a) Cross-section of a silicon nMOS with an integrated CNFET. ........................................150
Figure 6.3: DC electrical characteristics of CNT and nMOS transistors. ..............................................152
Figure 6.4: AC characteristics of nMOS-CNT integrated cascode amplifier. .......................................154
xiii
Tables
Table 2.1: Table of parameters and associated equations for carbon nanotubes .....................................17
Table 2.2: Table of specific values for selected carbon nanotubes..........................................................18
Table 3.1: ABT metrics and new extraction techniques ..........................................................................70
Table 4.1: Gain and output power for a range of transconductances.......................................................90
Table 4.2: S21 (-3dB) frequencies for a range of tbot ................................................................................96
Table 4.3: Capacitances per finger for design example.........................................................................100
1
1 Chapter 1
INTRODUCTION
1.1 Brief Introduction to Carbon Nanotubes
Carbon nanotubes (CNTs) are arguably the most fascinating new crystalline
material discovered in the past twenty years. The basic properties fueling the intense
research and economic interest are its ability to conduct electrical and thermal current
efficiently. In this respect, carbon nanotubes have been demonstrated to be the one of
the best electrical and thermal conductors known to man. Furthermore, its one-
dimensional (1D) nature with all the carbon atoms exposed on the surface makes it a
natural candidate for sensor applications.
The complete elucidation of the physical structure of carbon nanotubes is
widely credited to Sumio Iijima who in 1991 (and later in 1993) reported high-
resolution detailed electron microscope images of carbon nanotubes. Shortly thereafter,
theoretical work by several groups determined that CNTs intrinsically have unique and
diverse properties including diameter-tunable electrical, optical, and thermal
properties. Since then, subsequent research in the fundamental understanding and
potential applications of carbon nanotubes have risen substantially over the past two
decades as is evident in Figure 1.1, which is a graph of the publication growth of CNT
related articles.
Historically, much of the momentum driving the explosive growth of research
on nanotubes has been for transistor applications for very large-scale integrated
(VLSI) circuit technology. This was largely motivated by the experimental
observation of ballistic electron transport over relatively long lengths in carbon
2
nanotubes plus its higher current density and projected faster speed compared to
silicon transistors. However, we argue that carbon nanotubes still been relatively new
on the scene deserve to be studied for their own sake as well, for a variety of reasons,
including a necessary need to learn new paradigms about what nature offers in one-
dimensions (1D), and the potential for new applications beyond the reach of
conventional semiconductors. Hence, a comprehensive understanding of carbon
nanotubes is clearly of broad significance and interest that transcends the historical
motivation.
As a basic introduction, there are two families of carbon nanotubes (CNTs);
single-wall carbon nanotubes and multi-wall carbon nanotubes. A single-wall carbon
nanotube is a hollow cylindrical structure of carbon atoms with a diameter that ranges
from about 0.5nm-5nm and lengths of the order of micrometers to centimeters. A
multi-wall carbon nanotube (MWCNT) is similar in structure to the single-wall CNT
but has multiple nested or concentric walls with the spacing between walls comparable
to the interlayer spacing in graphite, approximately 0.34nm. Due to their large length
to diameter ratio, carbon nanotubes are considered (1D) nanomaterials which lead to
unique electronic properties unobservable in larger solid-state materials.
1.2 Applications of Carbon Nanotubes
In contemporary fundamental and applied science research, the potential
applications and the perceived broader impact are undoubtedly the primary drivers for
expanding the research enterprise. This has certainly been the case for nanotube
research. The unique unprecedented properties of carbon nanotubes such as its perfect
tubular structure, outstanding electrical and thermal conductance, tunable optical
properties, and superior mechanical strength and toughness have generated great
excitement leading to the pursuit of both fundamental insights of the beauty of nature
in reduced-dimensions of condensed matter, and the novel applications and
technological breakthroughs that can be developed.
3
0
100
200
300
400
500
600
700
1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
Year
nu
mb
er
of a
rtic
les
IEEE
ACS
AIP/APS
Figure 1.1: Growth in CNT articles in the publications of the major scientific societies.
IEEE and ACS represents the Institute of Electrical and Electronics Engineer and
American Chemical Society respectively. Also, AIP/APS represents the American
Institute of Physics and American Physical Society respectively.
4
In essence, the exploration of nanotubes and other nanomaterials is to learn about their
nature and their interaction with fields and matter that will allow us to design and
synthesize carbon nanotubes (CNTs) devices for next generation transformative
products. This endeavor has brought together many parties across several boundaries
of knowledge, from nanomedicine to nanoscience to nanotechnology. In short, it plays
a part in just about any discipline that has a nano prefix.
To put carbon nanotubes in some broader perspective; over the last decade,
nanotube applied research and development in academic and industrial labs across the
world have enjoyed a substantial increase reflecting a rise in the deeper understanding
of the material. Figure 1.2 which shows the increase in carbon nanotube patent
applications and patents issued in the United States is an indicator of the growing
effort to employ nanotubes in innovative applications. Invariably, many of the
applications of CNTs take advantage of its inherent nano-scale dimension, large
surface to volume ratio, and unique combination of electrical, optical, thermal and
structural properties. Some of the major applications include:
i. Field-effect transistors on conventional or flexible substrates for high-
performance electronics
ii. Sensors for real-time high-resolution monitoring of chemical and biological
molecules
iii. Nano-electromechanical systems for physical sensing or scanning probe
microscopy
iv. Field emission electron sources for display and imaging technology
v. Reinforced composites with engineered physical properties
vi. Hydrogen storage for low-emissions energy-efficient vehicles.
In the course of this dissertation, we addressed a variety of topics leading to a
greater understanding of carbon nanotubes especially for electronic applications and a
clearer direction of how to realize a hybrid nanotechnology that integrates carbon
nanotubes for arbitrary applications.
5
0
50
100
150
200
250
300
350
2000 2001 2002 2003 2004 2005 2006 2007 2008
Year
U.S
. P
ate
nts
Applications
Issued
Figure 1.2: United States patents issued and patent applications containing the phrase
carbon nanotubes in the patent abstract.
6
1.3 Description of Chapters
In the present work, our research has focused on elucidating the solid-state
physics, transistor device physics, material synthesis, and analog circuit properties of
carbon nanotubes. In addition, we developed a silicon-nanotube nanotechnology.
These are topics that cross many boundaries of knowledge including applied physics,
material science, and electrical engineering.
Chapter 2 provides a basic introduction to the physical and electronic structure
of CNTs. Chapter 3 explores in detail the device physics and charge transport
concluding with a ballistic model of nanotube field-effect transistors. Chapter 4
presents a small-signal model of CNT transistors for RF applications, and discusses
potential non-linear circuits exploiting the CNT quantum capacitance. Chapter 5
reports on the material synthesis of perfectly-aligned CNTs on quartz substrates. And
we conclude by demonstrating a hybrid nanotechnology which is based on the
integration of CNTs with conventional silicon technology in Chapter 6. A summary of
the dissertation contribution and some important challenges for future work regarding
CNTs are enumerated in Chapter 7.
7
2 Chapter 2
INTRODUCTION
Carbon nanotubes (CNTs) are arguably the most fascinating new crystalline
material discovered in the past twenty years. The basic properties fueling the intense
research and economic interest are its ability to conduct electrical and thermal current
efficiently. In this respect, carbon nanotubes have been demonstrated to be the one of
the best electrical and thermal conductors known to man. Furthermore, its one-
dimensional (1D) nature with all the carbon atoms exposed on the surface makes it a
natural candidate for sensor applications.
There are two families of carbon nanotubes (CNTs), single-wall carbon
nanotubes and multi-wall carbon nanotubes as shown in Figure 2.1. A single-wall
carbon nanotube is a hollow cylindrical structure of carbon atoms with a diameter that
ranges from about 0.5nm-5nm and lengths of the order of micrometers to centimeters.
A multi-wall carbon nanotube (MWCNT) is similar in structure to the single-wall
CNT but has multiple nested or concentric walls with the spacing between walls
comparable to the interlayer spacing in graphite, approximately 0.34nm. The ends of a
CNT are often capped with a hemisphere of the buckyball structure.
We begin this chapter by introducing the concept of chirality which is the main
idea used to describe the physical and electronic structure of CNTs. Then the physical
structure of CNTs is elucidated conceptually as a folding operation of the graphene
sheet resulting in three distinct configurations of single-wall carbon nanotubes. The
chapter concludes by examining the electronic band structure of carbon nanotubes.
8
Figure 2.1: Illustration of the two families of carbon nanotubes.
a) An ideal single-wall CNT with an hemispherical cap at both ends, and b) a multi-
wall CNT. In general, carbon nanotubes are much longer than depicted here, and the
MWCNTs can have up to several dozen walls.
a)
b)
9
2.1 Chirality: A Concept to Describe Nanotubes
Chirality is the key concept used to identify and describe the different
configurations of CNTs and their resulting electronic band structure. Since the concept
of chirality is of fundamental importance and often unfamiliar to engineers, let us take
a moment to introduce the concept of chirality before discussing how it is applied to
describe CNT structure. The term chirality is derived from the Greek term for hand
and it is used to describe the reflection symmetry between an object and its mirror
image.1 Formally, a chiral object is an object that is not super-imposable on its mirror
image, and conversely, an achiral object is an object that is super-imposable on its
mirror image. At this point, a visual illustration is often invaluable in making these
definitions vividly clear. For example, consider the left hand; its mirror image is the
right hand and we find that it is not possible to super-impose the two hands or images
such that all the features coincide precisely as illustrated in Figure 2.2. Therefore, the
human hand is a chiral object. Now, consider a circle as another example, its mirror
image is also an identical circle which super-imposes precisely on top of the original
image. Therefore, a circle is an achiral object. In a general usage, chirality is invoked
to highlight the presence or lack of mirror symmetry that provides intuition about
understanding phenomena.
Understanding the concept of chirality is essential because it is used to classify
the physical and electronic structure of CNTs. The carbon nanotubes that are super-
imposable on their mirror image are classified as achiral CNTs, and all other
nanotubes that are not super-imposable are classified as chiral CNTs. Moreover,
achiral CNTs are further classified as armchair CNTs or zigzag CNTs depending on
the geometry of the nanotube circular cross-section.
1 Care must be taken when discussing the mirror image of an object as it relates to chirality. For the example of the left hand, the plane containing the hand should be normal to the mirror.
10
Figure 2.2: Example of a chiral object.
a) The left hand and its mirror image (right hand). b) It is not possible to super-impose
the left hand on the right hand; therefore the human hand is chiral.
11
To briefly summarize, there are three types of single-wall carbon nanotubes
which are chiral CNTs, armchair CNTs, and zigzag CNTs of which the latter two are
achiral and their symmetry often makes them easier to explore and gain broad insight.
The three types of single-wall CNTs and their associated geometrical cross-sections
are shown in Figure 2.3.
2.2 The Carbon Nanotube Lattice
We introduced the concept of chirality to classify the different types of carbon
nanotubes in the previous section but it was not at all clear why CNTs arrange to form
a chiral or an achiral geometry. Fortunately, it is actually fairly easy to understand the
origin of the different types of CNTs by considering that a carbon nanotube results
from folding or wrapping of a graphene sheet. To see how the folding operation works,
we start from the direct lattice of graphene and then define a mathematical
construction which folds graphene’s lattice into a carbon nanotube. Moreover, this
mathematical folding construction directly leads to a precise determination of the
primitive lattice of carbon nanotubes, which is required information in order to derive
the CNT band structure. It is very important to keep in mind that the folding of
graphene to form a carbon nanotube is simply a convenient conceptual idea to study
the basic properties of CNTs. In actuality, CNTs naturally grow as a cylindrical
structure often with the aid of a catalyst which does not involve folding of graphene in
any physical sense.
Figure 2.4a shows the honeycomb lattice of graphene and the primitive lattice
vectors a1 and a2, defined on a plane with unit vectors x and y .
=
2,
23
1aa
a ,
−=
2,
23
2aa
a (1)
where a is the underlying Bravais lattice constant, ccaa −= 3 =2.46Å, and ac-c is the
carbon-carbon bond length (~1.42Å). Also a1 ⋅ a1 = a2 ⋅ a2 = a2, a1 ⋅ a2 =a
2/2, and the
angle between a1 and a2 is 60°.
12
Figure 2.3: The three types of single-wall CNTs.
a) A chiral CNT, b) an armchair CNT, and c) a zigzag CNT. The cross-sections of the
latter two illustrations have been highlighted by the bold lines showing the armchair
and zigzag character respectively.
13
With reference to Figure 2.4a, a single-wall carbon nanotube can be
conceptually conceived by considering folding the dashed line containing primitive
lattice points A and C with the dashed line containing primitive lattice points B and D
such that points A coincide with B, and C with D to form the nanotube shown in
Figure 2.4b. The carbon nanotube is characterized by three geometrical parameters,
the chiral vector Ch, the translation vector T, and the chiral angle θ as shown in Figure
2.4a. The chiral vector is the geometrical parameter that uniquely defines a CNT, and
|Ch|=Ch is the CNT circumference. Ch is defined as the vector connecting any two
primitive lattice points of graphene such that when folded into a nanotube these two
points are coincidental or indistinguishable. For the particular exercise of Figure 2.4,
the chiral vector is the vector from point A to B, Ch =3a1+3a2 = (3, 3). In general,
( )mnmnh ,21 =+= aaC , (n, m are positive integers, nm ≤≤0 ) (2)
and the resulting carbon nanotube is described as an (n, m) CNT.
Important observations regarding the type of CNT can be deduced directly
from the values of the chiral vector. Notice that the (3,3) CNT of Figure 2.4 leads to
an armchair nanotube. By extension, all (n,n) CNTs are armchair nanotubes. The case
when Ch is purely the along the direction of a1, (Ch = (n,0)) can be visually seen (from
the cross-section along the chiral vector) to result in zigzag nanotubes. All other (n,m)
CNTs leads to chiral nanotubes. The diameter (dt) of a carbon nanotube is derived
from its circumference Ch.
πππ
22 mnmnaCd
hhht
++=
⋅==
CC (3)
Notably, different chiralities can produce the same nanotube diameter, and as a result,
the diameter is not a unique parameter for characterizing carbon nanotubes. For
example, a (19,0) and a (16,5) nanotubes both have exactly the same diameter of
1.49nm. The equivalence of the diameter among dissimilar nanotubes has important
implications for the electronic properties.
14
x
y
a1
a2
T(1,-1) T
3a1 3a2
Figure 2.4: An illustration to describe the construction of a CNT from graphene.
a) Wrapping or folding the dashed line containing points A and C to the dashed line
containing points B and D results in the (3,3) armchair carbon nanotube in b) with
θ=30°. The CNT primitive unit cell is the cylinder formed by wrapping line AC onto
BD and is also highlighted in b).
15
This equivalence is employed later to show that the electronic properties of
CNTs are more strongly dependent on their diameter than on chirality. That is,
nanotubes with different chiralities but the same diameter have more or less the same
electronic properties. A case is point is the bandgap of nanotubes which is strongly
diameter dependent with little or no chirality dependence (see § 2.6).
The other two geometrical parameters (T and θ) can be derived from the chiral
vector. For instance, the chiral angle is the angle between the chiral vector and the
primitive lattice vector a1.
221
1
2
2
|||| mnmn
mnCos
h
h
++
+=
⋅=
a
a
C
Cθ (4)
The chiral angle can be viewed as describing the tilt angle of the hexagons relative to
the tubular axis. Due to the six-fold hexagonal symmetry of the honeycomb lattice,
unique values of the chiral angle are restricted to °≤≤ 300 θ . For the particular
exercise of Figure 2.4, θ =30°. In general, all armchair nanotubes have a chiral angle
of 30°, and for all zigzag nanotubes, θ =0°.
In order to determine the primitive unit cell of the CNT, we need to consider the
translation vector which defines the periodicity of the lattice along the tubular axis.
Geometrically, T is the smallest graphene lattice vector perpendicular to Ch. As can be
seen from Figure 2.4, T = (1,-1) for all armchair nanotubes. Similarly, the translation
vector for all zigzag nanotubes can be visually deduced to be T = (1,-2). More broadly,
the translation vector can be computed from the orthogonality condition Ch ⋅ T =0. Let
T = t1 a1+ t2 a2, where t1 and t2 are integers. Therefore,
0)2()2( 21 =+++=⋅ nmtmnth TC (5)
Determining the acceptable solution for t1 and t2 requires a subtle interplay involving
mathematical analysis and visual insight. There are two orthogonal directions (±90°)
relating T to Ch, and solving for either direction leads to an equivalent solution for the
translation vector. Let’s restrict the direction to the +90° as shown in Figure 2.4a.
Then according to the orientation definition of the lattice vectors a1 and a2, t1 must be
a positive integer and t2 must be a negative integer for T to be +90° with respect to Ch.
16
With this visual insight, one set of integers that satisfy Eq. (5) is (t1,t2) = (2m+n,-2n-m).
However, deeper thinking reveals that there are several set of integers that are also
solutions of Eq. (5). For instance, consider an (8, 2) CNT, (t1,t2) = (12,-18) is a
solution, but so are (t1,t2) = (12,-18)/2, (t1,t2) = (12,-18)/3, and (t1,t2) = (12,-18)/6. The
actual acceptable solution that leads to the shortest translation vector is (t1,t2) = (12,-
18)/6 = (2,-3), where the factor of 6 is the greatest common divisor of 12 and 18.
Hence, the acceptable solution for Eq. (5) is
+−
+==
dd g
mn
g
nmtt
2,
2),( 21T (6)
where gd is the greatest common divisor of 2m+n and 2n+m. The length of the
translation vector is
d
t
d
h
g
d
g
CT
π33=== |T| (7)
The chiral and translation vectors define the primitive unit cell of the carbon
nanotube which is a cylinder with diameter dt and length T. Some auxiliary results that
are useful to compute include the surface area of the CNT unit cell, the number of
hexagons per unit cell, and the number of carbon atoms per unit cell. The surface area
of the CNT primitive unit cell is the area of the rectangle defined by the Ch and T
vectors, | Ch × T |. The number of hexagons per unit cell (N) is the surface area divided
by the area of one hexagon.
d
h
d
h
ga
C
g
mnmnN
2
222
21
2)(2|
||=
++=
×
×=
aa|
TC (8)
This simplifies to N=2n for both armchair and zigzag nanotubes. Since there are two
carbon atoms per hexagon, there are a total of 2N carbon atoms in each CNT unit cell.
A summary of the geometric parameters and associated equations for carbon
nanotubes are listed in Table 2.1. Specific values of the geometric parameters for
selected nanotubes ranging in diameter from 1nm to 3nm are shown in Table 2.2.
17
Table 2.1: Table of parameters and associated equations for carbon nanotubes
),(21 mnmnh =+= aaC ),( nnh =C )0,(nh =C
22|| mnmnaC hh ++== C naCh 3= anCh =
22 mnmna
d t ++=π
3π
and t =
π
and t =
222
2
mnmn
mnCos
++
+=θ o30=θ o0=θ
2122
aadd g
mn
g
nm +−
+=T 21 aa −=T 21 2aa −=T
d
h
g
CT
3|== T| aT =
)2,2( mnnmgcdg d ++≡ ng d 3= ng d =
3aT =
d
h
ga
CN
2
22= nN 2= nN 2=
Note: The primitive basis vectors a1 and a2 are defined according to Eq. (1). cCNT
stands for chiral CNTs, aCNT for armchair nanotubes, and zCNT for zigzag nanotubes.
18
Table 2.2: Table of specific values for selected carbon nanotubes
Note: The bandgap (Eg) is computed from the tight-binding band structure of carbon
nanotubes which is discussed in § 2.4 and§ 2.6.
19
2.3 Carbon Nanotube Brillouin Zone
The electronic or band structure of carbon nanotubes is derived from the
Brillouin zone of graphene which is essentially the Fourier reciprocal of the direct
lattice. The folding of the direct lattice of graphene to form a nanotube manifests
electronically in that the band structure of CNTs are one-dimensional (1D) line-cuts of
the Brillouin zone of graphene. Figure 2.5a shows the Brillouin zone of a (3,3) CNT
overlaid on the Brillouin zone of graphene. The lines which represent the band
structure of the carbon nanotube are basically (1D) cuts of graphene’s reciprocal
lattice.
The Brillouin zone of carbon nanotubes is described by two important vectors
Ka and Kc. Ka is the reciprocal lattice vector along the nanotube axis, and Kc is along
the circumferential direction, both given in terms of the reciprocal lattice basis vectors
of graphene (b1, b2),
=
aa
ππ 2,
3
21b ,
−=
aa
ππ 2,
3
22b (9)
Employing the expressions for Ch, T, and N in Table 2.1, the wave vectors can be
algebraically derived.
)(1
21 bb nmN
−=aK (10)
)(1
2112 bb ttN
+−=cK (11)
The length of the reciprocal lattice wave vectors are inversely proportional to the CNT
lattice dimensions, i.e., |Ka| =2π/T, and |Kc| =2π/Ch. In the theory of CNTs, Ka is
considered to take on continuous values because it relates to the length of the CNT
which is ideally infinitely long. On the other hand, Kc exists as discrete quantities
because it reflects the physical (and quantum) confinement along the circumferential
direction.
20
a3
4π
x
y
Figure 2.5: Brillouin zone of a (3,3) armchair CNT (shaded rectangle).
The CNT Brillouin zone is overlaid on the reciprocal lattice of graphene. The numbers
refer to j=0,1,..,5 for a total of N=6 1D bands in the CNT Brillouin zone. j is a line or
subband index. The central hexagon is the first Brillouin zone of graphene, and the
high-symmetry points (Γ, M and K) of graphene’s Brillouin zone are also indicated. b)
The high-symmetry points of a line representing a CNT 1D band is illustrated.
21
It is customary to define a general wave vector (k) which describes the location
of any point on the discretized Brillouin zone of carbon nanotubes. This general
Brillouin zone wave vector is what will be used to compute the allowed energies in the
band structure of carbon nanotubes and is given by
<<−=+=
Tk
TNjj
Tk
ππ
π- and ,1,...,1,0 ,
/2 c
aK
Kk (12)
where k is a continuous wave vector that describes the continuous points along the
axial direction, and j is a discrete variable that indexes each line cut as shown in
Figure 2.5a. In summary, each value of j corresponds to a line or 1D band with wave
vectors (k) ranging from -π/T to +π/T. This is one of the most important properties of
the CNT Brillouin zone.
Additional results obtained from the study of the CNT Brillouin zone reveals
that 1/3 of all carbon nanotubes are metallic while the remaining 2/3 are
semiconducting. The metallic nanotubes include all armchair chiralities. This insight is
derived geometrically by investigating which nanotubes has lines that intersect the K-
point (or Fermi energy) of graphene. This can be formally stated mathematically by
requiring that the angle between jKc and ΓK vector be the chiral angle.
2222 3
)2(2||
2||
mnmna
mnCos
mnmna
jj
++
+=ΓΚ≡
++=
πθ
πc
K (13)
which is satisfied only when j=(2n+m)/3. This leads to the celebrated condition that a
carbon nanotube is metallic if (2n+m) or equivalently (n-m) is an integer multiple of 3
or zero,2 otherwise the CNT is semiconducting.
2.4 Tight-Binding Dispersion of Chiral CNTs
The band structure of carbon nanotubes can be determined from the nearest
neighbor tight-binding (NNTB) energy dispersion of graphene. This is sometimes
2 j=(2n+m)/3 is equivalent to j=((n-m)/3)+((n+2m)/3) which results in an integer value for j when n-m is a multiple of 3.
22
referred to as zone-folding because the energy bands of CNTs are line cuts or cross-
sections of the bands of graphene. It follows that the entire Brillouin zone of CNTs
can be folded into the first Brillouin zone of graphene. The zone-folding technique is a
powerful yet simple method to determine the electronic properties of carbon nanotubes
and the performance of CNT devices. However, the zone-folding and NNTB method
is limited as it does not account for several phenomena which are particularly
pronounced for small diameter (dt<1nm) nanotubes and high energy excitations. As
such, the NNTB band structure is primarily useful for CNTs with dt>1nm operating at
low energies,3 which fortunately covers the majority of electronic and sensor
applications of carbon nanotubes.
It is now timely to introduce the high symmetry points of CNTs to aid us in
discussing its band properties. High symmetry points are specific functions of
geometry. In the case of graphene, the Brillouin zone has an hexagonal geometry and
there are three points of symmetry: Γ, M and K (see Figure 2.5a). For carbon
nanotubes, the Brillouin zone is composed of N lines. By convention, the high
symmetry points of a line are the center and end of the line which are labeled Γ and X
respectively. It follows that each line will have a Γ-point and two X-points (see Figure
2.5b).
The band structure of carbon nanotubes can be computed by inserting the
allowed wave vectors into the energy dispersion relation for graphene. The tight-
binding energy (E) dispersion relation of graphene is:
)2
(cos4)2
cos()2
3cos(41)( 2
yyx ka
ka
ka
E ++±=± γk (14)
where the (+) and (-) signs refer to the conduction and valence bands respectively, and
γ ~3.1eV will be employed unless otherwise stated. k will now refer to the CNT
arbitrary Brillouin zone wave vector given by Eq. (12), and can be rewritten in terms
of its x and y components as ykxk yx ˆˆ +=k where
3 The phrase low energies refers to energies not far away from the Fermi energy.
23
3
333
2
)()(32
h
hx
C
mnkaCmnajk
−++=
π (15)
22
)(2)(3
h
hy
C
mnajCmnakk
−++=
π (16)
In general, for any (n,m) CNT, there will be N valence bands (E≤0) and N conduction
bands (E≥0). Figure 2.6 shows the band structures for (10,4) metallic and (10,5)
semiconducting chiral carbon nanotubes. The Fermi energy (EF) is by convention
defined to be 0eV. The semiconducting (10,5) CNT has a bandgap (Eg) ~0.86eV at the
Γ-point. We will show later in § 2.6 that the bandgap is inversely proportional to the
diameter, Eg ~0.9 (nm.eV)/dt, where dt is in nanometers.
In the subsequent sections, we will explore the band structure of the highly
symmetric achiral nanotubes to elucidate general properties of metallic and
semiconducting CNTs.
2.5 Band Structure of Armchair Nanotubes
The Brillouin zone wave vector (Eq. (12)) for armchair CNTs expressed in the
x and y coordinates is ykxanj ˆˆ)3/2( += πk . Substituting into Eq. (14) yields the
energy dispersion (Eac) for armchair nanotubes.
)- and ,12,...,1,0( ,)2
(cos4)2
cos()cos(41),( 2
ak
anj
kaka
n
jkjEac
πππγ <<−=++±= (17)
The band structure for an (8,8) armchair nanotube is shown in Figure 2.7, revealing an
energy degeneracy at ka=±2π/3, where the valence band touches the conduction band.
In general, the energy degeneracy at 0eV is common to all armchair CNTs and hence
armchair CNTs are metallic. Additionally, the lowest and highest energy subbands of
the valence and conduction bands are non-degenerate at arbitrary k-values with all
other subbands having a two-fold degeneracy. Noticeably, all the subbands have a
large degeneracy of 2n at the zone edge (ka=±π) corresponding to Eac=±γ.
24
Tπ
Tπ−
Tπ
Tπ−
-pi/T 0 pi/T
Figure 2.6: Band structures for a) (10,4) and b) (10,5) nanotubes within ±3eV.
The CNT diameters are 0.98nm and 1.04nm respectively. The metallic CNT shows a
band degeneracy at 0eV and k=±2π/3T. The semiconducting CNT has a bandgap of
~0.86eV.
25
-pi/T 0 pi/T-9
-6
-3
0
3
6
9
aπ−
aπ
Figure 2.7: Band structure for the (8,8) armchair nanotube.
The dispersion contains 16 1D subbands in the valence and conduction bands each.
For all armchair CNTs, the valence band touches the conduction band at ka=±2π/3
which explains their metallic properties. The thin lines are for the non-degenerate
subbands while the thick lines are for doubly degenerate subbands.
26
2.6 Band Structure of Zigzag Nanotubes and the
Derivation of the Bandgap
Zigzag carbon nanotubes are perhaps the most attractive type of nanotube to
explore because of the presence of either metallic or semiconducting behavior. They
also possess high symmetry leading to simple analytical expressions for many of the
solid state properties. The energy dispersion of zigzag CNTs can be obtained from the
Brillouin zone wave vector (Eq. (12)) which reduces to
yan
nkajx
an
nkaj ˆ2
32ˆ232 +
+−
=ππ
k (18)
Substituting into Eq. (14) yields the energy dispersion (Ezz) for zigzag nanotubes.
)33
- and ,12,...,1,0(,)(cos4)cos()2
3cos(41),( 2
ak
anj
n
j
n
jkakjE zz
ππππγ <<−=++±= (19)
The band structure for the metallic (12,0) and semiconducting (13,0) nanotubes are
shown in Figure 2.8. In general, when n is a multiple of 3, the zigzag CNT is metallic,
otherwise it is semiconducting. The lowest subbands have a two-fold degeneracy for
both metallic and semiconducting zigzag nanotubes.
In semiconductor theory, the bandgap (Eg) is of fundamental importance in
determining its solid state properties and also electronic transport in device
applications. The band index for the first subband (j1) for semiconducting zigzag
nanotubes is 2n/3 rounded to the nearest integer, which expressed mathematically is
31
32
32
round1 +≈
= n
nj (20)
where round(·) is a function that converts its argument to the nearest integer, and the
right-hand side expression is a linear approximation to the staircase-like round
function.
27
-pi/T 0 pi/T-9
6
3
0
3
6
9
Energy (eV)
a3π
a3π−
-9
-6
-3
0
3
6
9
Energy (eV)
a3π
a3π−
Figure 2.8: Band structures for a) (12,0) and b) (13,0) carbon nanotubes.
The (12,0) CNT is metallic while the (13,0) CNT is semiconducting due to the
bandgap at k=0. The thin lines indicate non-degenerate subbands while the thick lines
are for doubly degenerate subbands.
28
The linear approximation is actually exact for semiconducting zigzag nanotubes when
n-1 is an integer multiple of 3. Substituting the linear approximation for j into Eq. (19),
the bandgap for semiconducting zigzag CNTs is
++≈
nEg 33
2cos212
ππγ (21)
For relatively large n, π/3n is a small perturbation around 2π/3; therefore, a 1st-order
Taylor series expansion about 2π/3 in the cosine argument leads to
nn
nEg
32
3
2
3
22
πγ
πππγ =
−
+≈ (22)
The chiral index is related to the CNT diameter as shown in Table 2.1 for zCNT
thereby simplifying Eq. (22) to
t
ccg
d
aE −≈ γ2 (23)
which is the frequently employed bandgap-diameter relation for carbon nanotubes.
Numerically, Eg (eV)~0.9/dt(nm), a useful formula for quickly estimating a
nanotube’s bandgap in eV. Figure 2.9 demonstrates the accuracy of Eq. (23)
compared to exact NNTB bandgap computation. The figure includes the bandgaps of
all semiconducting carbon nanotubes with chiral indices ranging from (7,0) to (29,28).
Remarkably, even though Eq. (23) was derived for zigzag semiconducting
CNTs, it is equally accurate in estimating the bandgaps of arbitrary chiral nanotubes.
This case in point serves to illustrate the utility of the highly symmetric zigzag
nanotubes as an excellent vehicle for exploring the general properties of arbitrary
chiral nanotubes. It also illustrates that the bandgap of semiconducting nanotubes are
primarily dependent on the CNT diameter and not its specific chirality.
29
0.5 1 1.5 2 2.5 3 3.5 40.2
0.6
1
1.4
1.8
∝
Figure 2.9: The bandgap of semiconducting carbon nanotubes calculated using Eq.
(23) (solid line) compared to exact NNTB computation showing good agreement.
This plot includes all semiconducting CNTs with chiral indices ranging from (7,0) to
(29,28). The bandgap predictions for nanotubes with diameters<1nm should be
considered crude estimates only because the NNTB computation is inaccurate for sub-
nm CNTs due to curvature effects.
30
2.7 Conclusion
The physical and electronic structure of carbon nanotubes has been elucidated in
this chapter. Carbon nanotubes are classified based on the symmetry of its physical
structure resulting in armchair, zigzag, and chiral nanotubes, where the armchair and
zigzag types enjoy a higher symmetry than chiral nanotubes. The high symmetry of
achiral nanotubes is of practical convenience particularly for analysis and insight due
to the simple expressions for the energy dispersions. Fundamentally, carbon nanotubes
can be understood as a folding or wrapping of a graphene sheet. As such, both its
physical and electronic properties are derived from graphene. Invariably, a good
understanding of the physical and electronic structure of graphene is required in order
to fully appreciate the behavior of electrons in carbon nanotubes.
Electronically, carbon nanotubes can be either metallic or semiconducting, and
this diversity makes CNTs very attractive for a wide variety of applications including
interconnects, transistors, and sensors. The electronic structure of nanotubes has been
understood mostly from a relatively simple nearest-neighbor tight binding model
which has so far proved to be particularly useful in describing the low energy behavior
of charge carriers in nanotube devices with diameters greater than about 1nm. A key
property of the band structure is the horizontal and vertical mirror symmetry. The
vertical mirror symmetry also known as electron-hole symmetry holds within low
energy excitation. The analytical expression of the dispersion of electrons in
nanotubes is the foundation for much of the working theory of nanotube electronic
properties and device behavior.
31
3 Chapter 3
DEVICE PHYSICS
3.1 Preface
The device physics of a material is essential in elucidating the properties of the
material and assessing its potential for a variety of applications. Carbon nanotubes
(CNTs) been quasi one-dimensional materials represents a different paradigm than
bulk solids including a diameter dependent bandgap, weak electron-phonon coupling
leading to ballistic transport over relatively long lengths (mean free path~1um), and
high-current density. The relatively long mean free path makes metallic nanotubes
ideally attractive as interconnects in integrated circuits, while the ballistic transport
and high-current density suggests that semiconducting nanotubes will offer superior
transistor performance compared to silicon field-effect transistors (FETs).
Our objective was to study the fundamental solid-state physics of CNTs
starting from the density of states and derive expressions for many of the important
electronic properties ultimately leading to the development of an analytical theory for
ballistic transport in carbon nanotube FETs. Such an analytical ballistic theory is of
the utmost importance if carbon nanotube FETs are to become ubiquitous in future
integrated circuits because it enables compact/spice modeling, and the routine design
of circuits which are basic parts of the ecosystem of any semiconductor technology.
All the work was conceived and performed by the author with Jiale Liang
collaborating on the surface potential calculations in §3.3.3, and Soogine Chong
simulating the ITRS CMOS data used in Figure 3.14.
32
3.2 An Analytic Derivation of the Density of States,
Effective Mass and Carrier Density for Achiral
Carbon Nanotubes
(Reproduced with permission from D. Akinwande, Y. Nishi, and H.-S. P. Wong, “An Analytic
Derivation of the Density of States, Effective Mass and Carrier Density for Achiral Carbon Nanotubes”,
IEEE Trans. Elect. Devices, vol. 55, 2008)
Deji Akinwande, Yoshio Nishi, and H.-S. Philip Wong
CENTER FOR INTEGRATED SYSTEMS AND DEPARTMENT OF ELECTRICAL
ENGINEERING
STANFORD UNIVERSITY, STANFORD, CA, 94305, USA
Abstract: An analytical electron density of states for zigzag and armchair single-wall
carbon nanotubes are derived in this paper. The derivation originates from the tight-
binding energy dispersion relation for carbon nanotubes and reveals the essential
physics such as periodic van Hove singularities and its dependence on chirality. The
density of states derivation is exact and contains no additional approximations or
assumptions except those inherent in the nearest neighbor tight-binding model. In
addition, we derive analytical expressions for the group velocity, effective mass and
non-degenerate equilibrium carrier density.
3.2.1 Introduction
The electron density of states (DOS) of a crystalline solid is fundamental to
describing the electrical, optical, thermal, and mechanical properties of the solid [1].
Carbon nanotubes (CNTs) are one of the most exciting quasi one-dimensional (1D)
solids that exhibit fascinating electrical, optical and mechanical properties such as
33
high current density, large mechanical stiffness, and field emission characteristics [2-
4].
The DOS needed to explore the many properties of CNTs are currently
typically determined theoretically using one of two techniques. One technique
numerically computes the DOS from the tight-binding energy dispersion relation [2, 5],
and the other uses the approximate analytical (so-called) universal DOS [6]. The
former technique is necessary for quantitative accuracy especially since no equivalent
analytical DOS expression has been reported in the literature and moreover, the tight-
binding model is scaleable to include more neighbors. The latter scales out the
chirality dependence and is reportedly useful for estimates within 1eV of the Fermi
energy for any CNT [6]. Other approximate DOS equations that are functionally
correct have been previously published and provide good qualitative insight on the
CNT DOS dependency on energy [7, 8].
In this letter, we derive an exact analytical DOS for achiral CNTs from the
nearest neighbor tight-binding (NNTB) energy dispersion. The analytical DOS
explicitly reveals the essential physics such as van-Hove singularities (vHs), its
periodicity with respect to band index, and its dependence on chirality. In addition, the
results derived in this letter for achiral semiconducting and metallic CNTs are
generally applicable to the chiral equivalent of the same type with similar diameters
for energies within the vicinity of the Fermi energy. This similarity in the DOS and
electrical properties between achiral and chiral CNTs of similar diameter has been
observed experimentally and in numerical studies [6, 9-11]. Therefore, the analytical
DOS derived in this letter is useful for gaining insight into the device physics without
resorting to numerical calculations as we show (for example) in the subsequent
derivation of the effective mass and carrier density.
Furthermore, the analytical carrier density will find widespread use in
determining basic transport in CNT devices operating in equilibrium or quasi-
equilibrium without having to perform extensive time consuming numerical
simulations. It can also be utilized in the development of fast compact models for
circuit simulation.
34
3.2.2 Density of States
In this section, we derive the DOS of zigzag and armchair carbon nanotubes
and related properties. The 1D NNTB energy dispersion relation for zigzag (n,0) and
armchair (n,n) CNTs with band index q are respectively [2, 5],
n
q
n
qaKqKE ozz
ππγ 2cos4cos
2
3cos41),( ++±= (1)
2cos4cos
2cos41),( 2 aK
n
qaKqKE oac ++±=
πγ (2)
where q is an integer that ranges from 1 to 2n. γo is the nearest neighbor overlap
energy nominally between 2.5 and 3.2 eV, K is the reciprocal lattice wavevector, and a
is the graphene Bravais lattice constant (~2.46Å). The negative and positive prefix are
the band structures for the π (valence) and π* (conduction) bands respectively. The
intrinsic Fermi energy is 0 eV.
The simple NNTB energy dispersions do not include the effects of nanotube
curvature which has been computed to be significant for small CNTs with sub-1nm
diameters [12]. Additionally, the overlap parameter (s in [2]) which reportedly has a
near zero value has been set to zero in [2, 5] for simplicity to arrive at (1) and (2). As a
result of the simplifications intrinsic to the NNTB model, its primary use is for
accurately predicting the lowest sub-bands of CNTs with diameters greater than 1nm
which covers most electronic and semiconductor applications.
A. Zigzag Carbon Nanotubes
For clarity, we shall first derive the DOS for zigzag tubes. We consider an
infinitely long CNT for analytical simplicity. Moreover, published calculations
suggest little departure in the electronic structure for CNTs as short as sub-10 nm [13-
15]. For an infinitely long CNT, the DOS (normalized per unit length) can be
expressed as
35
∑=
=n
q
qEgEDOS2
1
),()( (3a)
where E
K
K
EqEg
∂
∂=
∂
∂=
−
ππ
11),(
1
(3b)
We solve the right hand equation directly because ∂K/∂E leaves an expression which
is a function of E. The wavevector for a zigzag CNT is
−−±= − )
2cos(23)sec(
41
cos3
2),(
2
21
n
qE
n
q
aqEK
o
zz
π
γ
π (4)
Differentiating with respect to energy, computing (3b) and organizing the resulting
denominator, the qth DOS restricted to the irreducible Brillouin zone (BZ) (half of the
first BZ) is,
))((
||
3
4),(
222
21
2 EEEE
E
aqEg
vhvh
zz
−−=
π
α (5)
for Ecb ≤ E ≤ Ect for the conduction band, and Evb ≤ E ≤ Evt for the valence band. Ecb
and Ect are the bottom and top of the conduction band respectively, and Evb and Evt are
the bottom and top of the valence band respectively. The zone degeneracy (α)
accounts for the contribution from the other half of the first BZ. Specifically, α=1 if
E=energy at BZ center (Γ point in the CNT reciprocal space), and α=2 otherwise. Evh1
and Evh2 are the zigzag vHs energies which define the energy space where the DOS is
finite and real.
|)cos(21|1
+±=
n
qE ovh
πγ (6a)
|)cos(21|2
−±=
n
qE ovh
πγ (6b)
The energies of the vHs are chirality dependent and periodic with period 2n
which is expected since there are identically 2n 1D bands in the CNT BZ. Evh1 is
located at the Γ point, and Evh2 is located in the second BZ. Several workers have
previously examined the properties of the vHs and their dependence on diameter and
chirality indirectly through computation [16, 17]; in (6), we show their explicit
36
analytical dependence. These two singularities generally reflect the two
minima/maxima present in a periodic sinusoid. In the case of a CNT, it is the quasi-
sinusoidal energy dispersion that leads to the vHs.
In order to deduce an equation for the bottom and top of the bands, another
energy definition is required.
)(cos41 2
n
qE oXBZ
πγ +±= (7)
EXBZ is the energy at the boundary of the first BZ (X point in the CNT reciprocal
space). This is calculated by setting K=π/a√3 in (1). For clarity, an illustrative graph of
a CNT bandstructure is shown in Figure 3.1, indicating the Γ and X points. Depending
on the sub-band, the bottom or top of the band are either at the Γ point or the X point.
Therefore,
vtXBZvhcb EEEE −== ),min( 1 (8a)
vbXBZvhct EEEE −== ),max( 1 (8a)
where min and max are the minimum and maximum of their arguments respectively.
The total number of states between E and E+dE is the definite integral of (3a)
dEE
E
n
qvhvh
vhvhzz
EEEE
EEE
adEEDOS
+
=
−∑
−−
−−= |
))((2
2tan
3
2)(
2
1 222
21
2
22
21
21
π
α (9)
37
-4
-3
-2
-1
0
1
2
3
4
E (
eV
)
Figure 3.1: An illustrative energy bandstructure of a semiconducting CNT.
Shown in the irreducible Brillouin zone (half of the 1st BZ) indicating the important
symmetry points, Γ and X. In this case, the bandstructure is for a (19,0) CNT and only
the first two sub-bands in the conduction, and valence bands are shown for clarity.
Γ X K (wavevector)
38
For metallic zigzag CNTs (n is a multiple of 3), the bands degenerate with the Fermi
energy are dispersionless and one Fermi band index is qF=2n/3 (located at K=0), with
a four-fold energy degeneracy. Therefore, the DOS at the Fermi energy for all metallic
zigzag CNTs is
119102~3
8)( −−= eVmx
aEDOS
o
Fzzπγ
(10)
which corroborates the result reported by Saito et. al. [2].
B. Some Properties of Zigzag DOS
It is worthwhile to examine some important features of the DOS analytically,
in order to gain intuition and understanding of the electro-physical properties of
carbon nanotubes. We derive three important results including the energy difference
between the first and second sub-bands, the energy bandwidth between the top and
bottom of the band for the first sub-band, and the dependence of the bandgap on the
diameter of the nanotube. The derivation focuses exclusively on the conduction band;
however, the results apply equally to the valence band due to electron-hole symmetry
[18].
First we re-derive the widely known relationship between the bandgap and
diameter for semiconducting zigzag CNTs directly from the DOS (a derivation based
on Taylor series expansion of the dispersion has been previously reported in [7]). An
expression for the minimum or first band index (q1) is,
3
1
3
2
3
21 +≈
= n
nroundq (11)
where round is a function that converts its argument to the nearest integer, and the
right hand expression in (11) is a linear approximation to the staircase round function.
The linear approximation is actually exact at some points because such points are
periodically coincident with the staircase round function. For reference, the exact
NNTB bandgap from (6a) is shown in (12a), and a first approximation of the bandgap
(Eg) obtained by substituting (11) for q in (12a) is shown in (12b).
|)cos(21|2
+=
n
qE og
πγ (12a)
39
|)33
2cos(212|
++≈
nE og
ππγ (12b)
For relatively large n, π/3n is a small perturbation around 2π/3; therefore a first-order
Taylor series expansion of (12) centered at 2π/3 in the cosine argument is sufficiently
accurate. As a result,
t
ccooog
d
a
nn
nE γ
πγ
πππγ 2
32
3
2
3
22 ==
−
+≈ (13)
where the relation dt = accn√3/π has been employed, and acc is the CNT carbon-carbon
length (~1.42Å).
Secondly, we derive the energy difference between the bottoms of the first and
second sub-bands. This energy difference is of value in later sections when
considering the number of sub-bands that contribute to carrier density and electrical
transport in CNTs. The bottom of the second sub-band (Ecb(q2)) is
|)cos(21|)()( 2212
+==
n
qqEqE ovhcb
πγ (14)
Using (11), q2 ~ q1-1=2n/3-1, and performing a Taylor series expansion of (14) similar
to that for (12b), the second sub-band can be simplified to
gocb En
qE =≈3
2)( 2π
γ (15)
Therefore, the energy difference (E∆12) is
2)()( 1212
g
cbcb
EqEqEE =−=∆ (16)
And lastly, we derive the relationship between the top and bottom of the conduction
band which we term the BZ energy bandwidth (EBZB). This energy bandwidth will
prove to be important in later calculations for the carrier density.
)()()( 111 qEqEqE cbctBZB −= (17)
By means of (6a) and (7) for the bottom and top respectively, and (11) for q1, and a
similar Taylor series expansion as discussed above, the BZ bandwidth is simplified to:
40
oooBZBnn
qE γγππ
γ 4.12632
12)( 1 =≈
−+≈ (18)
which is a good approximation for n as small as 13 (diameter ~1nm).
C. Armchair Carbon Nanotubes
Following the same procedure as the zigzag carbon nanotube, the DOS from
the qth sub-band for an armchair CNT restricted to the irreducible BZ is,
+−−
−−−
=
22
12
12
122
12 )()()(
||4),(
cvhcvhvh
ac
AEEAEEEE
E
aqEg
π
α (19)
for Ecb ≤ E ≤ Ect for the conduction band, and Evb ≤ E ≤ Evt for the valence band. Evh1
is an armchair vHs, and Ac1 and Ac2 are armchair energy parameters.
|sin|1n
qE ovh
πγ±= (20a)
−±= )cos(452
n
qE ovh
πγ (20b)
+−= )cos(21
n
qA oc
πγ (20b)
+= )cos(22
n
qA oc
πγ (20c)
where Evh2 is another armchair vHs that is not explicit in the (19). It is necessary to
calculate the energy at the BZ boundaries to determine the bottom and top of the bands.
The Γ point energy is computed by setting K=0 in (2), and the X point energy by
setting K=π/a.
+±=Γ )cos(45
n
qE oBZ
πγ (21a)
oXBZE γ±= (21b)
41
For armchair CNTs the bottom of the conduction band could be either within the BZ
or at the X point depending on the band index q, and the top of the band is the
maximum between the Γ and X points.
vbBZoct EEE −== Γ ),max(γ (22a)
vtvhcb EEE −== 1 n-floor(n/2)≤q≤n+floor(n/2) (22b)
vtocb EE −== γ n-floor(n/2)>q>n+floor(n/2) (22c)
where floor rounds its argument to the lowest integer.
There are a total of three vHs for armchair CNTs (equations (20a, 20b, 21a),
which are the poles of (19)). Evh1 oscillates between ±γo and can be considered the low
energy armchair singularities. The absolute value of the other two van Hove energies
are always ≥ γo and can be thought of as high energy singularities. The specific
functional form of the van Hove energies for armchair CNTs are different from that of
zigzag CNTs and it maybe possible to exploit this difference to uniquely determine the
chirality of carbon nanotubes from experimental DOS measurements or other
properties that reflect the DOS such as quantum capacitance [18]. However, the
perfect periodicity of the vHs (assumed in this paper) is noticeably broken at high
energies where the NNTB model is least accurate [5, 17].
Since all armchair CNTs are metallic (even with curvature effects included [2]),
the DOS at the Fermi energy is of great significance. One Fermi band index is qF=n
(located at K=±2π/3a), with a two-fold energy degeneracy. The calculated armchair
Fermi DOS is identical to the metallic zigzag DOS (10), further confirming the
general conclusion that all metallic CNTs have the same Fermi DOS.
3.2.3 Group Velocity and Effective mass
The electron group velocity is
),(
1|),(|
1
qEgE
KqEv
π
α
hh=
∂
∂=
−
(23)
42
where ћ is the reduced Plank’s constant. The DOS should be replaced with gzz or gac
for zigzag or armchair CNTs respectively. α is necessary in the numerator of (23) to
normalize for the zone degeneracy. Using (10) and normalizing for the sub-band
degeneracy, the Fermi velocity for a zigzag metallic CNT is ~9.6x105 m/s [19].
Similarly, armchair CNTs have the same Fermi velocity as above due to the
underlying symmetry of graphene’s Fermi pockets.
The electron (or hole) effective mass is a widely used concept in
semiconductor physics and have been shown to be useful for modeling and studying
the low energy electrical transport in CNTs [20, 21]. Instead of the standard formula
m*=ћ2/(∂2E/∂K2) which is a function of K, we derive it explicitly as a function of
energy and reuse the results above for the DOS. Equating the particle force to the
wave force, the effective mass (m*) is
11
),(*−
∂
∂
∂
∂
∂
∂=
∂
∂
∂
∂=
∂
∂=
K
E
EE
K
v
E
E
K
v
KqEm
gg hhhh (24)
where vg is the group velocity taken from the first part of (23). Substituting the DOS
for the energy gradients, a general equation for m* for a 1D band is
122
),(1
),(),(*
−
∂
∂=
qEgEqEgqEm πh (25)
Plugging in (5) and performing the necessary differentiation and simplifications
including normalizing for the zone degeneracy, the zigzag CNT m* is
( )422
21
3
2
2
3
16),(*
EEE
E
aqEm
vhvh −=
h (26)
which is generally not constant. Of great interest is the effective mass for the first sub-
band where Evh2≈Evh1+2γo and Evh1=Eg/2 for the conduction band. The resulting
effective mass for the lowest sub-band is
( )422
3
2
2
116)4(3
256),(*
EEE
E
aqEm
ogg −+=
γ
h (27)
where E≥Ecb=Eg/2. At the band minimum (E=Ecb), m* simplifies to
43
( )go
g
o
cbE
E
aqEm
+=
γγ 23
4),(*
2
2
1h (28)
For the 1.5nm (19,0) CNT which has been considered to be optimum for transport [4,
22], m*=0.053mo (γo=2.7eV, Eg~0.51eV, mo is rest mass) which is in good agreement
to an estimate of 0.058mo derived from a parabolic approximation to the (19,0) energy
dispersion [20]. A plot of the effective mass for several zigzag CNTs are shown in
Figure 3.2.
3.2.4 Carrier Density
The carrier density is an important thermodynamic concept useful for
determining the equilibrium and quasi-equilibrium properties such as electrostatics
and transport in semiconductors [23]. We derive the equilibrium electron carrier
density for semiconducting zigzag carbon nanotubes directly from the density of states
and Fermi-Dirac distribution. Due to symmetry of the band-structure, the hole carrier
density will be identical to electron carrier density. Subsequently, we quantify the
accuracy of the analytical equations with comparison to numerical results.
A. Carrier Density Calculation
The electron carrier density (n) per sub-band is the total number of occupied
states in the sub-band
∫=
top
bot
E
E
zz dEqEgEFDqn ),()()( (29)
where Etop and Ebot are the top and bottom of the sub-band respectively. For the lowest
sub-bands Etop and Ebot are equal to EXBZ and Evh1 respectively (we show later, that only
the lowest sub-bands contribute to carrier density). FD(E) is the Fermi-Dirac
distribution
kT
EE F
e
EFD−
+
=
1
1)( (30)
and EF is the Fermi energy, k is Boltzmann’s constant and T is the temperature.
44
0 0.05 0.1 0.15 0.2 0.25 0.30
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Effe
ctiv
e m
ass (
m*/
mo)
Energy (eV)
d=2nm
d=1.5nm
d=1nm
Figure 3.2: The effective mass of the first conduction band.
Plotted for 1nm (13,0), 1.5nm (19,0), and 2nm (25,0) carbon nanotubes. γo=2.7eV.
The energies have been normalized such that the band minimum is at 0eV.
45
To our best knowledge, (29) does not have an exact analytical solution. However,
under certain assumptions and approximations, an analytical solution is accessible. We
first make two approximations to simplify the integral.
1. The top of the band (Ect) is much greater than the bottom of the band (Ecb), and
due to the rapidly decaying Fermi tail, there is negligible error taking the upper
limit of the integral to infinity. This approximation is justified by inspection of
(18).
2. Using similar reason as above, the rapidly decaying Fermi tail implies that only
states with energies close to the band bottom contribute to the carrier density,
as a result, the relevant energies can be considered much smaller than the top
of the band. Mathematically speaking, Evh22-E2≈ Evh2
2 in (5) for gzz.
A graphical illustration of the DOS and FD distribution are shown in Figure 3.3. These
approximations simplify the carrier density integral to:
∫∫∞
−
∞
−−+
=−
+
=1 2
2
1
12
12
2 )1(1
1
3
4
)(1
1
3
4)( dt
t
t
eEa
EdE
EE
E
eEa
qnFzzt
vh
vh
vhEvhkT
FEEvh π
α
π
α (31)
where we have made some substitutions for simplicity (t=E/Evh1, z=Evh1/kT, zF=EF/kT).
However, this simplified integral does not have a closed form analytical
solution. Consequently, we make the so called non-degenerate assumption which
states that the Fermi energy is 3kT below the band bottom (EF ≤ Ecb-3kT), therefore the
Fermi-Dirac can be approximated by a Maxwell-Boltzmann distribution [23]. The
resulting integral equation
∫∞
−
−=
12
2
1
)1(3
4)( dt
t
tee
Ea
Eqn ztkT
E
vh
vhF
π
α (32)
has an integrand that has the form of a modified Bessel function of the 2nd kind. This
function is illustrated in Figure 3.3c. The integral has a solution (available in standard
mathematical handbooks [24])
)(3
4)( 1
12
1
kT
EKe
Ea
Eqn vhkT
E
vh
vhF
π
α= (33)
46
where K1(z) is the modified Bessel function of the 2nd kind of order one. K1(z) does
not have a closed form solution; however K1/2(z) and K3/2(z) (orders ½ and ³/2
respectively) have closed form solutions. A closed form approximation for K1(z) can
be obtained by taking the average of K1/2(z) and K3/2(z).
zez
zK −=2
)(2/1π (34a)
zezz
zK −+= )1(2
)(32/3
π (34b)
zez
z
zKzKzK −+
=+
≈2
)21(
22
)()()(
32/32/1
1π (34c)
Therefore, the carrier density for the qth sub-band is
kT
EE
vhvh
vhvhF
eE
kT
aE
kTEqn
1
12
1
322
)(−
+=
πα (35)
and the total electron carrier density (n) is a summation over all sub-bands. However,
because the 1D sub-bands are discretely spaced and the tail of the Fermi-Dirac
distribution is exponentially decaying, only the first sub-band contributes appreciably
to the total electron carrier density for the non-degenerate case [25]. A quantitative
justification for the one band approximation is presented in Appendix A. As a result,
the total electron carrier density is approximately the first sub-band contribution times
the band degeneracy which is two for zigzag CNTs.
kT
EE
c
vhF
eNqnn
1
2)(2 1
−
=≈ (36a)
12
1
322
vhvh
vhc
E
kT
aE
kTEN
πα
+= (36b)
where Nc is the effective density of states for the conduction band.
47
Energ
y
DOS (arb. units) 0 0.5 1
Fermi-Dirac (FD)
DOS*FD (arb. units)
Figure 3.3: Illustrative plot of the DOS and Fermi-Dirac distribution at equilibrium.
(a) Conduction and Valence band DOS. (b) FD for electrons (solid) and holes (dashed).
(c) DOS multiplied by the FD (has the form of a modified Bessel function of the 2nd
kind) showing that essentially only the first sub-band contributes to carrier density.
48
Eqn. (36) can be further simplified by substituting Evh2≈Evh1+2γo and Evh1=Eg/2.
kT
EE
c
gF
eNn 2
2
2−
= (37a)
gog
g
cE
kT
E
kTE
aN
πγα
344
+
+= (37b)
The intrinsic electron carrier density (EF=0) is
kT
E
ci
g
eNn 22−
= (38)
Finally, the electron carrier density can be expressed in terms of the intrinsic carrier
density.
kT
E
i
F
enn = (39)
Likewise, the hole carrier density (p) can be expressed as
kT
E
i
F
enp
−
= (40)
Equations (36)-(40) have essentially the same form as for 3D bulk
semiconductors [23]. Interestingly, the form of the carrier density and its dependence
on bandgap, temperature, and Fermi energy can be deduced simply by dimensional
scaling of the 3D bulk carrier density, specifically by scaling the effective density of
states (also known as the partition function in statistical mechanics [26]) from 3D to
1D as we show in Appendix B.
B. Accuracy of Analytical Carrier Density
The accuracy of the analytical non-degenerate carrier density calculation
compared to numeric nearest neighbor tight-binding simulations shows good
agreement as plotted in Figure 3.4 and Figure 3.5. Specifically, the comparison is
documented as the temperature, Fermi energy, and diameter vary over their practical
values. For the intrinsic case, the |error| is less than 0.6% using (33), and less than
5.5% using (36) for diameters within 1nm to 3nm. The error grows slightly to a
maximum of ~3.5% when the Fermi level is 3kT below the conduction band.
49
150 200 250 300 350 400 450 500
10-4
10-2
100
102
104
106
Intr
insic
Ca
rrie
r D
en
sity
(n i)
[1/c
m]
T (K)
d=1nm
d=1.5nm
d=2nm
d=3nm
Figure 3.4: The semiconducting zigzag CNT intrinsic carrier density.
Calculated using (33) with subband degeneracy of two. Compared with NNTB
simulations, the maximum |error| <0.6% for temperatures from 100°C to 500°C. Using
(36) instead of (33), the maximum error is 1.3% for d=1nm, 2.3% for d=1.5nm, 3.7%
for d=2nm, and 5.2% for d=3nm. The additional small error results from
approximating the modified Bessel function.
50
2 3 4 5 6 7 8
103
104
105
Ca
rrie
r D
en
sity (
n)
[1/c
m]
(Ecb
-EF)/kT
d=1nm
d=1.5nm
d=2nm
d=3nm
2 3 4 5 6 7 8
0
1
2
3
4
5
6
7
8
9
err
or
(%)
(Ecb
-EF)/kT
d=1nm
d=1.5nm
d=2nm
d=3nm
Figure 3.5: The semiconducting zigzag CNT carrier density.
Calculated using (33) versus the Fermi Level (top graph) at room temperature. The
bottom graph shows the error compared with the NNTB simulations. Within the non-
degenerate range (Ecb-EF>3kT), the maximum error is <3.5%. Using (36) in place of
(33), the maximum error is <6% for Ecb-EF>3kT.
51
A question of interest is: what is the intrinsic carrier density of CNTs
compared to bulk semiconductors? If the carrier densities of Figure 3.4 are normalized
to the nanotube cylindrical volume, a ~1.5nm zigzag CNT (bandgap is approximately
half of silicon) at room temperature has ni~7x1015/cm3, which is 5 orders of magnitude
larger than silicon, and two orders larger than germanium. However, the range of
which the carrier density of CNTs can be varied in practice is not well understood at
the moment and requires further research.
3.2.5 Conclusion
In conclusion we have derived an analytical density of states for achiral carbon
nanotubes based on the tight-binding dispersion. The DOS results for semiconducting
zigzag CNTs is applicable to semiconducting chiral CNTs with similar diameters at
least in the vicinity of the Fermi level. The same extension holds between the achiral
and chiral metallic CNTs. Furthermore, the analytical DOS offers insight into the
behavior of CNTs and will promote better compact modeling of CNT devices with
faster simulations at the circuit level compared to numerical tight-binding or first
principle schemes.
Additionally, the effective mass and equilibrium carrier densities were derived
from the DOS. This will enable an accurate analytical description of transport and
electrostatics in non-degenerate CNT devices and will conceivably be the starting
point of hand analysis and design of CNT circuits.
Acknowledgement
The authors would like to thank James Fodor and Professor Jing Guo for useful
suggestions. This work is supported in part by the Focus Center Research Program
(FENA and C2S2). Additional support from the Ford foundation and the Alfred P.
Sloan foundation are much appreciated by one of the authors (D.A).
Appendix A
JUSTIFICATION FOR ONE-BAND APPROXIMATION
52
The relative contribution of the second sub-band to the non-degenerate carrier
density is discussed in this appendix. For example, consider the second sub-band
electron carrier density normalized to the first sub-band,
kT
qEE
kT
qEE
c
c
vhF
vhF
e
e
qN
qN
qn
qn)(
)(
1
2
1
2
11
21
)(
)(
)(
)(−
−
= (A.1)
From (36b) the ratio of the effective density of states is
)(
)(
)(
)(
)(2
)(2
)(
)(
21
11
22
12
11
21
1
2
qE
qE
qE
qE
kTqE
kTqE
qN
qN
vh
vh
vh
vh
vh
vh
c
c
+
+= (A.2)
Substituting for the van-Hove energies following the technique developed in part B of
§3.2.2, Evh2(q)≈Evh1(q)+2γo, Evh1(q2)≈2Evh1(q1), and Evh1=Eg/2, the ratio simplifies to
)2(
)4(
)(22
)2(
)(
)(
1
2
og
og
g
g
c
c
E
E
kTE
kTE
qN
qN
γ
γ
+
+
+
+= (A.3)
Further approximations such as Eg<<2γo (for diameters down to ~1nm), and kT<< Eg
(for temperatures up to air breakdown temperature ~600C [27, 28], and diameters up
to 3nm) reduces the ratio to a constant.
2)(
)(
1
2 ≈qN
qN
c
c (A.4)
The key observation is that the effective DOS are approximately equal and certainly
within the same order of magnitude. The Fermi energy cancels out in the exponential
ratio and making the necessary simplifications, the relative contribution of the second
sub-band is,
kT
Eg
eqn
qn 2
1
2 2)(
)(−
≈ (A.5)
Quantitatively, this implies that for temperatures up to 500°C and diameters up
to 3nm (Eg~0.26eV), the second sub-band contribution is about an order of magnitude
less than the first sub-band carrier density (~0.07 smaller). For more practical
semiconducting CNTs with diameters of ~1.5nm operating at room temperature, the
second sub-band contribution is about four orders of magnitude less. Higher sub-band
53
carrier density contribution will be even much less than the second sub-band,
consequently, they can be ignored for non-degenerate carbon nanotubes.
Appendix B
THERMODYNAMIC DERIVATION OF CNT CARRIER DENSITY
Dimensional scaling of the carrier density of 3D bulk semiconductors of the
zinc blende lattice can reproduce the functional form of the carrier density for a
semiconducting 1D solid such as carbon nanotubes. Consider the non-degenerate
electron carrier density for a general 3D bulk semiconductor such as silicon.
kT
EE
DcD
cF
eNn
−
= 3,3 (B.1)
2/32/333
2/3
23
3, )()(2
2 kTmAh
kTmN DD
DDc =
=
π (B.2)
where Nc,3D is the electron effective density of states for the 3D semiconductor, m3D is
the effective mass, and A3D is a general constant of no specific importance in this
discussion.
In statistical mechanics, the effective density of states is a type of partition
function, an important quantity that encodes the statistical properties of a system in
thermodynamic equilibrium [26]. The relevant properties it encodes in the context of
carrier density are the bandstructure (vis-à-vis the effective mass) and temperature.
The power of three in (B.2) comes from the three dimensions of space in bulk
materials. For 1D materials the power of three reduces to one.
2/12/111, )()( kTmN DDc ∝ (B.3)
Equation (25) can be substituted for the effective mass for semiconducting CNTs.
( ) o
g
go
g
D
E
E
Em
γγ 221 ∝+
∝ (B.4)
where Eg is assumed to be much less than 2γo, a valid approximation for CNTs with
diameters down to ~1nm (Eg~0.7eV). The 1D effective density of states can now be
expressed as,
54
kTEAN gDDc 11, = (B.5)
A1D is a constant appropriate for 1D semiconductors.
The exponential term in n3D contains no dimensional information and
accordingly, carries over to the 1D case. Finally, the thermodynamically scaled 1D
non-degenerate electron carrier density is,
kT
EE
DckT
EE
DcD
vhFcF
eNeNn
1
1,1,1
−−
== (B.6)
Comparison with the derived quantities in §3.2.4 ((36) repeated here for
clarity)
12
1
12
1
322
322
vhvh
vh
vhvh
vhc
E
kT
aE
E
E
kT
aE
kTEN
π
α
πα ≈
+= (B.7)
Substituting Evh2≈Evh1+2γo and Evh1=Eg/2, and realizing that practical bandgaps are
much smaller than 4γo, the preceding equation simplifies to
kTEAkTE
EaN g
g
og
c ≈+
=πγ
α
3)4(4 (B.8a)
kT
EE
c
vhF
eNn
1
2−
= (B.8b)
where A is a constant.
We conclude that the derived quantities (B.8) are functionally identical to the
thermodynamically scaled quantities.
55
3.3 Analytical Ballistic Theory of Carbon Nanotube
Transistors: Experimental Validation, Device Physics,
Parameter Extraction, and Performance Projection
(Reproduced with permission from D. Akinwande, J. Liang, S. Chong, Y. Nishi, and H.-S. P. Wong,
“Analytical Ballistic Theory of Carbon Nanotube Transistors: Experimental Validation, Device Physics,
Parameter Extraction, and Performance Projection”, Journal of Applied Physics, vol. 104, 2008)
Deji Akinwande, Jiale Liang, Soogine Chong, Yoshio Nishi, and H.-S. Philip Wong
CENTER FOR INTEGRATED SYSTEMS AND DEPARTMENT OF ELECTRICAL
ENGINEERING
STANFORD UNIVERSITY, STANFORD, CA, 94305, USA
Abstract: We have developed a fully analytical ballistic theory of carbon nanotube
field effect transistors enabled by the development of an analytical surface potential
capturing the temperature dependence and gate and quantum capacitance
electrostatics. The analytical ballistic theory is compared to the experimental results
of a ballistic transistor with good agreement. The validated analytical theory enables
intuitive circuit design, provides techniques for parameter extraction of the bandgap
and surface potential, and elucidates on the device physics of drain optical phonon
scattering and its role in reducing the linear conductance and intrinsic gain of the
transistor. Furthermore, a threshold voltage definition is proposed reflecting the
bandgap-diameter dependence. Projections for key analog and digital performance
are discussed.
56
3.3.1 Introduction
Carbon nanotubes (CNTs) are fascinating tubular cylindrical arrangement of
carbon atoms with diameters (dCNT) on the order of 1 to 5nm [29]. These quasi one-
dimensional solids are being actively explored for a variety of applications including
interconnects [30, 31] and transistors for electronics, optics, and sensors, [32-34]
either on monolithic substrates or integrated with conventional silicon technology for
hybrid applications [35-37]. In order to obtain maximum electrical performance for
large scale integration, ballistic carbon nanotube field effect transistors (CNFETs)
shorter than the mean free path (mfp) of acoustic phonons (~0.5um) are currently the
most attractive semiconducting nano-devices [21, 25, 33]. Much of the progress in
widely advancing a semiconductor depends on the ability to routinely predict its
performance accurately and rapidly. Furthermore, an analytical theory is required in
order to design circuits intuitively by examining key physical parameters that
determine performance. Up till now, this has been challenging for CNFETs since the
accurate models are numerical requiring a self-consistent quantum mechanical
formalism that is usually time consuming [7]. For example, the device simulation data
provided by [33] (shown in Fig. 5a) required ~7.5hours on a 3GHz computer, while
the computing time for typical analytical models are below one second. Previously
reported CNFET analytical ballistic theories were not fully analytical because an
iterative numerical scheme was required to solve for the surface or chemical potential
[21, 38-40].
In this article, we develop a fully analytical ballistic theory (ABT) for the
current-voltage (I-V) of CNFETs enabled by an analytical surface potential (φs), which
captures the essential dependence on temperature, and oxide and quantum
capacitances. The analytical ballistic theory is compared to experimental data [33]
with remarkably strong agreement, and leads to clear insights and device physics such
as: 1) a simple electrical method to extract the CNT bandgap (Eg), 2) elucidation of the
role of drain optical phonon scattering (DOPS) in reducing the CNFET conductance in
the linear region. Further studies of the validated ABT leads to a series of results
including a proposed threshold voltage (VT) definition for an intrinsic CNFET that is
57
not based on carrier inversion, showing a simple ~1/dCNT relation, and the derivation
of a set of analytical analog and digital performance metrics. This work is a significant
advancement over prior CNT ABT models [21, 38-41] due to the fully analytical
surface potential, the introduced concept of DOPs, the derivation of a set of analytical
analog and digital performance metrics, and the demonstrated experimental
validation. The ABT results developed in this article applies to all ballistic CNFETs
and enables compact modeling for circuit design and simulation [39-41], as well as
qualitative and quantitative insight to the device physics.
3.3.2 Analytical Ballistic Theory
In practical intrinsic CNFETs with metal or doped Ohmic contacts (Figure 3.6),
only the lowest two subbands contribute to current [25, 42], resulting in the two
subband approximation for the current of a ballistic CNFET which is the net flux of
forward (positive velocity) and backward (negative velocity) traveling carriers [21].
The I-V is developed here for an n-type transistor, where the contribution of holes are
considered negligible.
[ ]∫∞
−
+−−=
secbE
DFFD eVEEFEEFEMEvEgdEeIϕ
),(),()()()(
[ ]∫∞
−
−−
secbE
DFF eVEEFEEFEMEvEgdEeϕ2
),(),()()()( (1)
where the former (latter) is the contribution from the 1st (2nd) subband. F(·) is the
Fermi-Dirac distribution, EF is the Fermi level defined at the source, and the drain
Fermi level is lowered by the drain voltage (VD). Ecb(=Eg/2) and Ecb2 are the bottom of
the first and second subbands respectively, and the effect of φs is to shift the bands as
illustrated in Figure 3.7a. g(E) and v(E) are the density of states and group velocity per
subband respectively, with v(E)=2/(hg(E)). h is the Planck’s constant, e is the
electrical charge constant, and M is the subband degeneracy (M=2). Simplifying for
v(E) and M in Eq. (1) results in
[ ] [ ]∫∫∞
−
∞
−
−−+−−=
secbE
DFF
secbE
DFFD eVEEFEEFdEh
eeVEEFEEFdE
h
eI
ϕϕ 2
),(),(4
),(),(4 (2)
which yields:
58
+
++
+
+=
−−−
−−
−−−
−−
kT
secbEDeVFE
kT
secbEFE
qkT
secbEDeVFE
kT
secbEFE
q
D
e
e
eR
kT
e
e
eR
kTI
)2(
)2(
)(
)(
1
1ln
1
1ln
ϕ
ϕ
ϕ
ϕ
(3)
kT is the thermal energy, Rq is the quantum resistance (Rq=h/4e2≈6.45kΩ), and
kT/eRq~4µA. Eq. (3) is the essential analytical I-V for a ballistic CNFET as a function
of the terminal voltages, material properties, and temperature. For convenience, the
source Fermi level is considered the reference potential (EF=0).
At low operating energies (where contribution of the 2nd subband is negligible),
Eq. (3) can be further reduced to the first subband ballistic current.
+−+=
−−−
)1ln()1ln( 2
22
2
2
kT
gEDeVse
kT
gEse
qD ee
eR
kTI
ϕϕ
(4)
The latter term (backward carriers) is significant in the linear region and the
former term (forward carriers) determines the saturation current (ION). We can
calculate a rough estimate for the maximum ION achievable from the first subband by
reasonably assuming the maximum eφs~1.5Ecb. In this case, the maximum ION from
the 1st subband ~12uA-36uA for dCNT from 3nm to 1nm respectively at room
temperature. The explicit gate voltage (VG) and gate capacitance dependence is
captured in the surface potential which is developed analytically in the next section.
The analytical ballistic current developed here does not include electron-electron
interactions which have been shown to be inconsequential by Maslov and Stone for an
ideal Luttinger liquid ballistic wire [43]. For a p-type CNFET, Eq. (3) or (4) is
modified by reversing the polarity of φs, and VD.
59
Figure 3.6: CNFET devices covered in this work illustrated with an n-type.
a) A top-gated CNFET with doped CNT ohmic contacts and its band diagram. Ecb is
the bottom of the conduction band, and Evt is the top of the valence band. b) A self-
aligned CNFET with Ohmic metallic contacts, and its band diagram (k=15, tox=8nm
for the experimental and analytical CNFET in this work).
60
Figure 3.7: a) Two subband E-k dispersion of a CNT.
Ecb2 is the band minima for the 2nd subband. The effect of VG is to shift the bands with
respect to EF and induce φs. b) Analytical φs (s=1.41V-1 & t=0.76) compared to self-
consistent computation showing good agreement (typical peak error <5%).
61
3.3.3 Analytical Surface Potential
The CNT surface potential is controlled by the gate voltage through a
capacitive divider network consisting of the gate oxide capacitance (Cox) and the
quantum capacitance (Cq) [18, 44, 45]. Specifically, φs determines the directional
charge carriers in the CNT channel.
Gsq
ox
oxs V
CC
C
2
)(ϕϕ
+
= (5)
where Cq/2 is the quantum capacitance per carrier direction, and VG is the actual gate
voltage under flat band conditions. In general, VG=VG′-VFB, where VG′ is the external
gate voltage, and VFB is the flat band voltage which depends on the work function
difference as usual. Cq depends on φs exponentially [46, 47], hence Eq. (5) is
transcendental and is typically solved for φs numerically in a self-consistent manner.
To aid in deducing an approximate analytical form for φs, we investigate the
quantum and semi-classical limits. At low gate voltages, the CNFET is in the quantum
capacitance limit (Cq<<Cox), and φs≈VG, with unity slope. At higher gate voltages,
both capacitances are comparable (semi-classical limit) and the slope becomes non-
linear and less than unity. Based on this insight, we propose a suitable analytical form
for the surface potential (after our recent work [46])
+−
+−−=
tVsH
tVsV GGG
s
||1
1||
111
ϕ (6)
where H(·) is the Heaviside unit step function, and s and t are bandgap, Cox and
temperature dependent parameters which we derive in the Appendix. Eq. (6) compared
to self-consistent computation shows good agreement with peak errors typically less
than 5% (Figure 3.7b). A more rigorous mathematical analysis of φs provided in the
Appendix reveals that Eq. (6) is actually the first order Taylor approximation of Eq.
(5) in the semi-classical limit.
62
3.3.4 Experimental Validation and Device Physics
The value of analytical theories depends on how successful the theory predicts
experimental observations. Employing an experimental ballistic p-CNFET previously
reported by Javey [33], we compare Eq. (4) showing only good agreement in the
saturation region (Figure 3.8). The analytical parameters used are identical to the
values reported for the experimental CNFET (dCNT=1.7nm, Eg=0.5eV, and device
geometry shown in Figure 3.6b), including a parasitic source/drain metal resistance of
~2kΩ each, which is close to the reported estimate of ~1.7kΩ. The flatband voltage
was not reported for the experimental CNFET, but VFB=0.28V was found most useful
in reproducing the measurements. In the linear region, the analytical theory (Eq. (3) or
(4)) significantly overestimates the current and conductance. We propose that the
reduced conductance of the experimental CNFET in the linear region is due to drain
optical phonon scattering (DOPS) which is explained graphically in Figure 3.9.
63
Figure 3.8: Experimental results and analytical (Eq. (4)) predictions for a ballistic
50nm p-type CNFET (device identical to Figure 3.6b).
s=1.6V-1, t=0.82 for the analytical φs. Agreement is good in the active region, but
weak in the linear region due to the DOPS effect. Experimental results courtesy of
[33].
64
Figure 3.9: Illustration of DOPS in a ballistic (n-type) CNFET.
a) At low VD, the length of the triangle well (l1) is comparable to the mfp of optical
phonon scattering (lop~10nm), thereby reducing ID. At higher VD, the well is
sharpened with a vanishing effect on ID. This device physics can be captured by a
(DOPS) parameter (α) which accounts for the reduced ballistic current.
65
In brief, at low drain bias (linear region), the length of the high energy
triangular quantum well at the drain contact is comparable to the mfp of optical
phonon scattering resulting in carrier scattering and a reduced current. At higher drain
bias (saturation region), the triangular well is sharpened with a length smaller than the
optical phonon mfp thereby having a vanishing effect on the current. These concepts
have been previously alluded to by Natori [48] and Guo [49] but not incorporated into
the ABT.
We introduce a DOPS parameter (0<α≤1) to modify the effective drain voltage
in Eq. (4) since optical phonon scattering at the drain contact implies a localized
resistance and therefore, the effective VD on the ballistic CNT channel is decreased.
+−+=
−−−
)1ln()1ln( 2
22
2
2
kT
gEDVese
kT
gEse
qD ee
eR
kTI
αϕϕ
(7)
The DOPS corrected ballistic current (Eq. (7)) is primarily justified by the remarkably
strong agreement with the experimental data in all major regions of transistor
operation (sub-threshold, linear, and saturation) as shown in Figure 3.10. Detailed self-
consistent quantum mechanical simulations reported by Guo [33] for the experimental
device are included in Figure 3.10a for reference, demonstrating the strength of the
simple ABT in accurately reproducing the experimental and self-consistent simulation
results. Furthermore, analytical results based on the approximation that Cq≈quantum
capacitance for metallic CNTs (Cq,metallic~300aF/um) are included in Figure 3.10b
showing gross errors in all regions except the linear region. The errors particularly in
the active and subthreshold regions render the approximation (sometimes employed in
compact modeling) of little value in accurate modeling or performance benchmarking
of CNFETs.
66
Figure 3.10: Improved agreement between experiment and analytical ballistic theory.
a) ID-VDS including the DOPS parameter (α=0.5) leads to an improved agreement
between analytical theory (Eq. (7)), and experimental results and quantum mechanical
simulations. b) Likewise ID-VGS also results in strong agreement with experimental
data in active, linear and subthreshold regions. The dashed line (VD=-0.1V) is the
analytical predictions if Cq≈Cq,metallic is assumed as is sometimes employed in compact
modeling, which leads to gross errors in all regions except the linear region.
67
To provide clarity in the use of the term ‘ballistic transport’ in transistors, we
make two distinctions. 1) AP-ballistic transport defined as transport with no acoustic
phonon scattering but possibly optical phonon scattering (α<1), and 2) OP-ballistic
transport (or fully ballistic transport) defined as transport with no optical phonon nor
acoustic phonon scattering (α=1). Further research is warranted to investigate how
the DOPS parameter scales with the device dimensions with the potential that the
results of such investigations might lead to optimizing the device structure to minimize
the DOPS effect.
The threshold voltage for intrinsic CNFETs has been vague from the outset
because of a lack of carrier inversion as is the case in conventional doped
semiconductors. We propose a definition for VT that is universally applicable to all
CNFETs and extractable from measurements (same definition previously proposed by
Wong [50] for thin oxide FETs). The proposed VT is the VGS for which ∂gm/∂VGS is a
maximum, where gm is the transconductance (∂ID/∂VGS). This mathematical definition
is similar to the physical definition used for conventional MOSFETs but avoids
extraction ambiguities related to short channel effects. In the quantum capacitance
limit (φs=VG), the analytical VT for n-CNFET is
−+≈
+
)24(121
ln 22 kT
DVe
kT
DVe
kT
DVegE
T eeekTV
ααα
(8)
where the approximation is due to the neglect of exponentials with negative arguments
in the intermediate equations which have a negligible effect compared to exponentials
with positive arguments. A 1st-order Taylor series of Eq. (8) yields VT=(½e)Eg+offset,
revealing that VT is linear with bandgap and hence inversely proportional to dCNT
(since Eg~1/dCNT) [42]. This is confirmed in Figure 3.11 for the VT computed exactly
from the ABT model for a ballistic 50nm n-CNFET, and compared to a simple
VT~(½)dCNT relation showing good agreement. It is worthwhile to note that the 50nm
CNFET (after Figure 3.6b) is operating in the semi-classical limit, showing that Eq.
(8) though derived for the quantum capacitance limit is also useful for accurately
estimating VT when Cq is comparable to Cox.
68
A suite of relevant analytical equations for the ballistic CNFET including
saturation and sub-threshold current, gm, and gds are provided in Table 3.1.
3.3.5 Parameter Extraction and Performance Projection
The bandgap which is a fundamental material property of semiconductors can
be extracted from experimental measurements using the analytical ballistic equation.
Specifically, with knowledge of VFB, Eg can be accurately extracted from the ballistic
ID-VGS in deep subthreshold with VDS→VDD (see Table 3.1, Eq. (8)). This opens up an
electrical technique in applied physics for measuring the temperature and diameter
dependent Eg. Using this technique, the extracted Eg~0.494eV as shown in Figure 3.12
for Javey’s experimental CNFET, in close agreement to the reported 0.5eV [33]. The
close agreement between the extracted Eg and the reported Eg for the experimental
device also serves to justify the value of VFB (=0.28V). Alternatively, with knowledge
of Eg, the flatband voltage can be extracted.
Likewise, with knowledge of Eg, the surface potential can be extracted from
ION, and ∂φs/∂VGS extracted from gm as VDS→VDD (see Table 3.1, Eq. (1) and Eq. (5)
respectively). Furthermore, the DOPS parameter, α, can be extracted directly from gds
(see Table 3.1, Eq. (7)). In principle, with knowledge of just the CNT diameter,
temperature, and device geometry, the ballistic I-V reveals the pertinent device
physics of the CNFET (i.e., all the relevant device parameters are extractable from
measurements).
69
Figure 3.11: VT exactly computed from ABT model (Table 3.1, Eq. (3)).
For a ballistic 50nm n-type CNFET (derived from Figure 3.6b), showing weak
dependence on drain voltage, and a simple ~1/dCNT relation.
70
Table 3.1: ABT metrics and new extraction techniques
−≈
−−−
kT
gEDSeVGSeV
kT
gEGSeV
q
sth eeeR
kTI 2
22
2
2 α
gFBGSsth
qEVVeI
kT
eRkT −−= )(2)ln(2
qds
Rg
α=0
GS
s
qkT
gEse
kT
gEDSeVsemVR
ee
g∂
∂
+
−
+
=−−−
ϕϕαϕ
1
1
1
1
1
2
2
2
22
)1ln( 2
2
kT
gEse
qON e
eR
kTI
−
+≈
ϕ
imumV
gwhenVV
GS
mTGS max, =
∂
∂=
Comments/Conditions
ON current in
saturation region
Sub-threshold current showing VGS and VDS
dependence. Ideal slope is 60mV/decade at 300K
Analytical
transconductance
Analytical channel
conductance
Accurate analytical VT in the quantum capacitance
limit (φs=VGS)
New technique to extract α from measured gds (VDS→0,VGS→VDD)
New technique to extract Eg
from measured Isth for nonzero VFB (VDS→VDD)
Proposed definition for VT
for intrinsic CNFETs
Analytical equations for ballistic n-type CNFET
)]24(121
ln[ 22 kT
DeV
kT
DeV
kT
DeVgE
T eeekTV
ααα
−+≈
+
12
22
)1( −
−+
+= kT
seDSeVgE
q
ds eR
g
ϕαα
#
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
71
Figure 3.12: Deep subthreshold current and linear fit for extracting the CNT bandgap.
The device is the experimental p-CNFET (see Table 3.1, Eq. (8) for n-CNFET). The
intercept=2eVFB-Eg, with VFB=0.28V (model value used in Figure 3.8 and Figure 3.10)
yields Eg~0.494eV, in good agreement with the reported CNT Eg of ~0.5eV in [33].
The departure from the dashed line is a result of the CNFET gradually turning ON.
72
Much of the excitement about CNFETs is their potential performance
advantages over silicon transistors. The performance metrics chosen here are
independent of CNT density in the channel unless otherwise stated explicitly. To aid
in assessing performance, key digital and analog performance metrics of a ballistic n-
CNFET (device structure similar to Javey’s experimental p-CNFET) are estimated.
The primary digital metric is the intrinsic CV/I delay which is shown in Figure 3.13a
indicating that 10nm gate length (Lg) CNFET satisfies the high performance (HP)
international technology roadmap for semiconductors (ITRS) [51] 25nm logic node
(Lg=10nm) using double-gate MOSFETs. The CNFET high-k gate dielectric is 4nm
thick with an equivalent oxide thickness (EOT) of ~1nm which is identical to the
electrical EOT for 25nm ITRS HP logic for a fair comparison. Further scaling of the
CNT channel length reduces the capacitances linearly but has a negligible effect on the
ballistic current resulting in an overall linear reduction of the intrinsic delay. However,
parasitic capacitances have to be negligible in order to enjoy the intrinsic performance
[52-54]. Also the ON/OFF current ratio (ION/IOFF) and ION are shown in Figure 3.13b
conveying that CNFETs with dCNT~1-3nm can satisfy both the ION and ION/IOFF ITRS
roadmap, however flatband voltage engineering maybe required to realize ION/IOFF and
arrays of CNTs simultaneously needed to meet the desired ION. For example, to meet
the ION for 25nm ITRS, a 1.5nm CNFET (VFB=0V) needs ~40CNTs/um. The ABT
model does not include ambipolar conduction which can degrade the ION/IOFF. To
mitigate ambipolor effects, source/drain engineering can be employed to achieve the
desired unipolar behaviour [55].
The primary analog metrics are gm, gm/ID, and gm/gds (intrinsic gain), which are
shown in Figure 3.14 for a 50nm ballistic n-CNFET. A peak gm of ~50uS/CNT is
achievable in moderate saturation, and the CNFET is more efficient in terms of gm/ID
when compared to the simulated ITRS transistor with a similar gate length as shown in
Figure 3.14b. The effect of DOPS is evident in Figure 3.14c showing a much degraded
intrinsic gain which nonetheless, outperforms the highest gm/gds comparable CMOS
[56]. It is imperative that DOPS has to be mitigated (e.g. by drain contact engineering)
for maximum analog performance.
73
Figure 3.13: Performance assessment for ballistic CNFET for logic applications.
a) Intrinsic delay of a 10nm CNFET at a fixed VG-VT=0.2V to ensure saturation
(VDD=0.8V, I=ION, C is a series combination of oxide and quantum capacitances). The
delay is weakly dependent on diameter, and satisfies 25nm ITRS HP logic node
(Lg=10nm). b) ABT model (Eq. (7)) for ION and ION/IOFF (VDD=0.8V, α=1), and also
showing the ITRS low operating power (LOP) ION/IOFF for the same VDD. The ITRS
LOP was chosen for b) because of its more stringent ION/IOFF requirement compared to
the ITRS HP technology.
74
Figure 3.14: ABT model calculations for key analog metrics for a 50nm CNFET.
a) transconductance (VD=0.5V). b) gm efficiency (VD=VG, α=1) showing a profile
similar to a comparable ITRS CMOS but larger by up to a factor of ~2, and identical
to an ideal BJT in subthreshold. c) Intrinsic gain metric (dCNT=1.5nm, VD=0.5V)
showing significant degradation with drain optical phonon scattering, and also
included for reference is the highest reported intrinsic gain for 32nm CMOS
technology (the reported device had an Lg=60nm) from [56], and the simulated gain
from the comparable ITRS CMOS device.
75
3.3.6 Conclusion
We have developed an analytical ballistic theory for carbon nanotube field
effect transistors showing good agreement with state of the art experimental data. The
strong agreement was achieved without including electron-electron interactions
indicating that one electron excitations dominate the ballistic transport. The developed
ABT leads to clear insights and techniques such as Eg and φs extraction from the I-V,
and a simple ~1/dCNT relation for VT. The role of DOPS in reducing the linear
conductance and the intrinsic gain has also been elucidated. These results are
immediately applicable for the applied physics and compact modeling of CNT devices.
In addition, the validated ballistic equations will aid in the CNT device and
circuit design enabling the selection of optimum diameter and device geometry
tailored to the desired application.
Acknowledgements
We thank Prof. Ali Javey (University of California, Berkeley) for the experimental
results and Prof. Jing Guo (University of Florida) for the quantum mechanical
simulations (both reported in [33]). This work is supported in part by the Focus Center
Research Program (FENA and C2S2). S.C is additionally supported by the Samsung
Fellowship, and J.L is additionally supported by Gerald L. Pearson Memorial
Fellowship. D.A is additionally supported by the Ford foundation and the Stanford
DARE graduate fellowships.
Appendix
SERIES DERIVATION FOR THE S AND T PARAMETERS OF THE ANALYTICAL SURFACE
POTENTIAL
In this appendix we solve Eq. (5) analytically for the surface potential by
means of a 1st-order Taylor series expansion resulting in an analytical form that
justifies Eq. (6). This leads to identification of the temperature, Cox, and diameter
76
dependent s and t parameters in terms of the Taylor series coefficients. For the 1st-
subband, Cq is given analytically by [46]
x
kTeAkT
eNeC
kT
gEsekTsegE
kT
gEsekTkTsegE
oq 2
2)2(
)22)(2(
2)2(
)222)(2(
22
+
=−+−
−+−−
βϕβαϕ
βϕβαϕ
−+−+
−+−
)2(2)2(
)22)(2(
sgkT
gEsekTsegE
eEkTkTAkTe ϕββα
βϕβαϕ
(A1)
where No is the effective density of states [42]. A, α, and β are empirical constants in
the analytical CNT carrier density theory determined to be 0.63, 0.88, 2.41x10-3
respectively [46, 47], with an average error <10% for dCNT~1-3nm and temperature
~300K±100K when comparing Eq. (A1) to numerical tight-binding computation for
Cq. A more comprehensive derivation and discussion is provided in reference [46].
The 1st-order Taylor series of Eq. (5) about φs= mEg is
Gss VSS )( 10 ϕϕ −≈ (A2)
with S0 and S1 being the 0th-order and 1st-order series coefficients respectively, which
are analytical but not explicitly written here because they are lengthy and can be easily
derived by the interested reader. It is important to emphasize that S0 and S1 explicitly
depend on Cox, bandgap, temperature, and m. Ideally if Cq was linear and symmetric
about eφs= 0.5Ecb then m would be ~0.25/e, however, because Cq is quasi-linear and
generally not symmetric, a value of m~0.30/e-0.33/e was found to give the best
agreement with self-consistent computation, with a weak dependence on dCNT and
temperature. Finally, simplifying Eq. (A2) yields the analytical φs.
00
1 1
SV
S
S
V
G
Gs
+
≈ϕ (A3)
which is identical to Eq. (6) in the semi-classical region, with s=S1/S0 V-1 and t=1/S0.
77
The step function in Eq. (6) is simply employed as a convenient mathematical
tool to merge the solutions of the surface potential in the quantum capacitance and
semi-classical limits of operation.
3.4 References
[1] N. W. Ashcroft and N. D. Mermin, Solid State Physics. New York,: Holt Rinehart and Winston,
1976.
[2] R. Saito, G. Dresselhaus, and M. S. Dresselhaus, Physical properties of carbon nanotubes.
London: Imperial College Press, 1998.
[3] W. B. Choi, D. S. Chung, J. H. Kang, H. Y. Kim, Y. W. Jin, I. T. Han, Y. H. Lee, J. E. Jung, N.
S. Lee, G. S. Park, and J. M. Kim, "Fully sealed, high-brightness carbon-nanotube field-
emission display," Applied physics letters, vol. 75, pp. 3129-31, 1999.
[4] A. Javey, R. Tu, D. B. Farmer, J. Guo, R. G. Gordon, and H. J. Dai, "High performance n-type
carbon nanotube field-effect transistors with chemically doped contacts," Nano Letters, vol. 5,
pp. 345-8, 2005.
[5] S. Reich, C. Thomsen, and J. Maultzsch, Carbon nanotubes : basic concepts and physical
properties. Weinheim ; Cambridge: Wiley-VCH, 2004.
[6] J. W. Mintmire and C. T. White, "Universal Density of States for Carbon Nanotubes," Physical
Review Letters, vol. 81, p. 2506, 1998.
[7] S. Datta, Quantum transport : atom to transistor. Cambridge, UK ; New York: Cambridge
University Press, 2005.
[8] A. Jorio, A. G. Souza Filho, G. Dresselhaus, M. S. Dresselhaus, R. Saito, J. H. Hafner, C. M.
Lieber, F. M. Matinaga, M. S. S. Dantas, and M. A. Pimenta, "Joint density of electronic states
for one isolated single-wall carbon nanotube studied by resonant Raman scattering," Physical
Review B, vol. 63, p. 245416, 2001.
[9] C. T. White and J. W. Mintmire, "Density of states reflects diameter in nanotubes," Nature, vol.
394, pp. 29-30, 1998.
[10] A. Hazeghi, T. Krishnamohan, and H. S. P. Wong, "Schottky-barrier carbon nanotube field-
effect transistor modeling," IEEE Transactions on Electron Devices, vol. 54, pp. 439-45, 2007.
[11] J. Deng and H.-S. P. Wong, "A Compact SPICE Model for Carbon Nanotube Field Effect
Transistors Including Non-Idealities and Its Application Part II: Full Device Model and
Circuits Performance Benchmarking," IEEE Trans. Electron Devices, vol. 54, pp. 3195-3205,
Dec. 2007 2007.
78
[12] O. Gülseren, T. Yildirim, and S. Ciraci, "Systematic ab initio study of curvature effects in
carbon nanotubes," Physical Review B, vol. 65, p. 153405, 2002.
[13] R. A. Jishi, J. Bragin, and L. Lou, "Electronic structure of short and long carbon nanotubes
from first principles," Physical Review B, vol. 59, p. 9862, 1999.
[14] A. Rochefort, D. R. Salahub, and P. Avouris, "Effects of Finite Length on the Electronic
Structure of Carbon Nanotubes," J.PHYS.CHEM.B, vol. 103, pp. 641-646, 1999.
[15] D. J. Klein, H. Y. Zhu, T. G. Schmalz, A. Rubio, and N. H. March, "Geometric boundary
effects on the electronic properties of finite carbon nanotubes," Journal of Physics and
Chemistry of Solids, vol. 59, pp. 417-423, 1998.
[16] H. Kataura, Y. Kumazawa, Y. Maniwa, I. Umezu, S. Suzuki, Y. Ohtsuka, and Y. Achiba,
"Optical properties of single-wall carbon nanotubes," Synthetic Metals, vol. 103, pp. 2555-
2558, 1999.
[17] R. Saito, G. Dresselhaus, and M. S. Dresselhaus, "Trigonal warping effect of carbon
nanotubes," Physical Review B, vol. 61, p. 2981, 2000.
[18] S. Ilani, L. A. K. Donev, M. Kindermann, and P. L. McEuen, "Measurement of the quantum
capacitance of interacting electrons in carbon nanotubes," Nature Physics, vol. 2, pp. 687-691,
2006.
[19] F. Kreup, A. P. Graham, M. Liebau, G. S. Duesberg, R. Seidel, and E. Unger, "Carbon
nanotubes for interconnect applications," in IEEE IEDM Tech. Dig., 2004, pp. 683-686.
[20] S. O. Koswatta, N. Neophytou, D. Kienle, G. Fiori, and M. S. Lundstrom, "Dependence of DC
characteristics of CNT MOSFETs on bandstructure models," IEEE Transactions on
Nanotechnology, vol. 5, pp. 368-372, 2006.
[21] K. Natori, Y. Kimura, and T. Shimizu, "Characteristics of a carbon nanotube field-effect
transistor analyzed as a ballistic nanowire field-effect transistor," Journal of Applied Physics,
vol. 97, pp. 034306-7, 2005.
[22] A. Raychowdhury, A. Keshavarzi, J. Kurtin, V. De, and K. Roy, "Carbon Nanotube Field-
Effect Transistors for High-Performance Digital Circuits;DC Analysis and Modeling Toward
Optimum Transistor Structure," IEEE Transactions on Electron Devices, vol. 53, pp. 2711-
2717, 2006.
[23] R. F. Pierret, Semiconductor Device Fundamentals. Reading, Mass.: Addison-Wesley, 1996.
[24] M. Abramowitz and I. A. Stegun, Handbook of mathematical functions with formulas, graphs,
and mathematical tables, 10th printing, with corrections ed. Washington, D.C.: U.S. Dept. of
Commerce : U.S. G.P.O., 1972.
[25] J. Guo, A. Javey, H. Dai, and M. Lundstrom, "Performance analysis and design optimization
of near ballistic carbon nanotube field-effect transistors," in IEEE IEDM Tech. Dig., 2004, pp.
703-706.
[26] C. Kittel and H. Kroemer, Thermal Physics, 2d ed. San Francisco: W.H. Freeman, 1980.
79
[27] K. Hata, D. N. Futaba, K. Mizuno, T. Namai, M. Yumura, and S. Iijima, "Water-Assisted
Highly Efficient Synthesis of Impurity-Free Single-Walled Carbon Nanotubes," in Science. vol.
306, 2004, pp. 1362-1364.
[28] I. W. Chiang, B. E. Brinson, R. E. Smalley, J. L. Margrave, and R. H. Hauge, "Purification and
characterization of single-wall carbon nanotubes," J.PHYS.CHEM.B, vol. 105, pp. 1157-61,
2001.
[29] S. Iijima and T. Ichihashi, "Single-shell carbon nanotubes of 1-nm diameter," Nature, vol. 363,
pp. 603-605, 1993.
[30] A. Naeemi, R. Sarvari, and J. D. Meindl, "Performance comparison between carbon nanotube
and copper interconnects for gigascale integration (GSI)," IEEE Electron Device Letters, vol.
26, pp. 84-86, 2005.
[31] G. F. Close, S. Yasuda, B. Paul, S. Fujita, and H. S. P. Wong, "A 1 GHz Integrated Circuit
with Carbon Nanotube Interconnects and Silicon Transistors," Nano letters, vol. 8, pp. 706-709,
2008.
[32] P. Avouris, Z. Chen, and V. Perebeinos, "Carbon-based electronics," Nature Nanotech., vol. 2,
pp. 605-615, 2007.
[33] A. Javey, J. Guo, D. B. Farmer, Q. Wang, E. Yenilmez, R. G. Gordon, M. Lundstrom, and H. J.
Dai, "Self-aligned ballistic molecular transistors and electrically parallel nanotube arrays,"
Nano Letters, vol. 4, pp. 1319-22, 2004.
[34] J. Li, Y. Lu, Q. Ye, M. Cinke, J. Han, and M. Meyyappan, "Carbon Nanotube Sensors for Gas
and Organic Vapor Detection," Nano Letters, vol. 3, pp. 929-933, 2003.
[35] D. Akinwande, S. Yasuda, B. Paul, S. Fujita, G. Close, and H.-S. P. Wong, "Monolithic
Integration of CMOS VLSI and CNT for Hybrid Nanotechnology Applications," IEEE Trans.
on Nanotechnology, pp. to be published,, Sept. 2008.
[36] Y. C. Tseng, P. Xuan, A. Javey, R. Malloy, Q. Wang, J. Bokor, and H. Dai, "Monolithic
Integration of Carbon Nanotube Devices with Silicon MOS Technology," Nano Letters, vol. 4,
pp. 123-127, 2004.
[37] T. S. Cho, K. J. Lee, J. Kong, and A. P. Chandrakasan, "A Low Power Carbon Nanotube
Chemical Sensor System," in IEEE Custom Integrated Circuits Conf., 2007, pp. 181-184.
[38] D. Jimenez, J. J. Saenz, B. Iniquez, J. Sune, L. F. Marsal, and J. Pallares, "Unified compact
model for the ballistic quantum wire and quantum well metal-oxide-semiconductor field-
effect-transistor," Journal of Applied Physics, vol. 94, pp. 1061-1068, 2003.
[39] S. Fregonese, H. Cazin d'Honincthun, J. Goguet, C. Maneux, T. Zimmer, J. P. Bourgoin, P.
Dollfus, and S. Galdin-Retailleau, "Computationally Efficient Physics-Based Compact
CNTFET Model for Circuit Design," IEEE Transactions on Electron Devices, vol. 55, pp.
1317-1327, 2008.
80
[40] J. Deng and H.-S. P. Wong, "A Compact SPICE Model for Carbon-Nanotube Field-Effect
Transistors Including Nonidealities and Its Application—Part I: Model of the Intrinsic
Channel Region," IEEE Trans. on Electron Devices, vol. 54, pp. 3186-3194, 2007.
[41] A. Balijepalli, S. Sinha, and Y. Cao, "Compact modeling of carbon nanotube transistor for
early stage process-design exploration," in International symposium on Low power electronics
and design: ACM, 2007, p.2.
[42] D. Akinwande, Y. Nishi, and H.-S. P. Wong, "An Analytical Derivation of the Density of
States, Effective Mass and Carrier Density of Achiral Carbon Nanotubes," IEEE Trans. on
Electron Devices, vol. 55, pp. 289-297, 2008.
[43] D. L. Maslov and M. Stone, "Landauer conductance of Luttinger liquids with leads," Physical
Review B, vol. 52, p. R5539, 1995.
[44] D. L. John, L. C. Castro, and D. L. Pulfrey, "Quantum capacitance in nanoscale device
modeling," Journal of Applied Physics, vol. 96, pp. 5180-5184, 2004.
[45] D. Akinwande, Y. Nishi, and H.-S. P. Wong, "Analytical Model of Carbon Nanotube
Electrostatics: Density of States, Effective Mass, Carrier Density, and Quantum Capacitance,"
in IEDM Technical Digest: IEEE, 2007, pp. 753-756.
[46] J. Liang, D. Akinwande, and H.-S. P. Wong, "Carrier Density and Quantum Capacitance for
Semiconduting Carbon Nanotubes," Journal of Applied Physics, vol. 104, p. 064515, 2008.
[47] D. Akinwande, J. Liang, and H.-S. P. Wong, "Analytical Degenerate Carrier Density and
Quantum Capacitance for Semiconducting Carbon Nanotubes," in Device Research
Conference Digest: IEEE, 2008, pp. 117-118.
[48] K. Natori, "Ballistic MOSFET reproduces current-voltage characteristics of an experimental
device," IEEE Electron Device Letters, vol. 23, pp. 655-657, 2002.
[49] J. Guo and M. Lundstrom, "Role of phonon scattering in carbon nanotube field-effect
transistors," Applied Physics Letters, vol. 86, pp. 193103-3, 2005.
[50] H.-S. Wong, M. H. White, T. J. Krutsick, and R. V. Booth, "Modeling of transconductance
degradation and extraction of threshold voltage in thin oxide MOSFET's," Solid-State
Electronics, vol. 30, pp. 953-968, 1987.
[51] The International Technology Roadmap for Semiconductors, p. available online:
http://www.itrs.net/.
[52] D. Akinwande, G. F. Close, and H.-S. P. Wong, "Analysis of the Frequency Response of
Carbon Nanotube Transistors," IEEE Trans. on Nanotech., vol. 5, pp. 599-605, 2006.
[53] B. C. Paul, S. Fujita, M. Okajima, and T. Lee, "Impact of geometry-dependent parasitic
capacitances on the performance of CNFET circuits," IEEE Electron Device Letters, vol. 27,
pp. 380-382, 2006.
81
[54] L. Wei, J. Deng, and H. S. P. Wong, "1-D and 2-D Devices Performance Comparison
Including Parasitic Gate Capacitance and Screening Effect," in IEEE IEDM Tech. Digest, 2007,
pp. 741-744.
[55] Y. M. Lin, J. Appenzeller, and P. Avouris, "Ambipolar-to-Unipolar Conversion of Carbon
Nanotube Transistors by Gate Structure Engineering," Nano letters, vol. 4, pp. 947-950, 2004.
[56] S.-Y. Wu, C. W. Chou, C. Y. Lin, M. C. Chiang, C. K. Yang, M. Y. Liu, L. C. Hu, C. H.
Chang, P. H. Wu, H. F. Chen, S. Y. Chang, S. H. Wang, P. Y. Tong, Y. L. Hsieh, J. J. Liaw, K.
H. Pan, C. H. Hsieh, C. H. Chen, J. Y. Cheng, C. H. Yao, W. K. Wan, T. L. Lee, K. T. Huang,
K. C. Lin, L. Y. Yeh, K. C. Ku, S. C. Chen, H. J. Lin, S. M. Jang, Y. C. Lu, J. H. Shieh, M. H.
Tsai, J. Y. Song, K. S. Chen, V. Chang, S. M. Cheng, S. H. Yang, C. H. Diaz, Y. C. See, and
M. S. Liang, "A 32nm CMOS Low Power SoC Platform Technology for Foundry Applications
with Functional High Density SRAM," in IEEE IEDM Tech. Digest, 2007, pp. 263-266.
82
4 Chapter 4
RF ANALOG CIRCUITS
4.1 Preface
One of the promising applications of carbon nanotubes (CNT) is for high-
frequency high-performance analog circuits at radio frequencies (RF) to optical
frequencies. This promise is based on the intrinsic transit frequency of CNT transistors
which is estimated to be in the vicinity of 1THz. Additionally, some of the unique
properties of CNTs such as the voltage-dependent quantum capacitance make them
especially attractive for novel non-linear analog circuits which form the backbone of
signal sources which are the heart of communication systems. However, realizing
CNT circuits that operate at RF and beyond will only be possible if the otherwise
undesirable parasitic elements that exist in any circuit geometry are taken into
consideration and minimized.
This work provides an analysis of nanotube analog circuits operating at high-
frequencies. This includes the performance analysis of nanotube RF amplifiers in the
presence of parasitic capacitances. In this light, design guidelines are provided to
minimize the effects of parasitics and optimize performance. In addition, the
description of a novel quantum capacitance based multiplier is elucidated showing that
it is possible to obtain THz signal generation circuits with carbon nanotubes using
existing fabrication techniques on insulating substrates.
All the work was conceived and performed by the author. Gael Close provided
very fruitful exchange of ideas which led to an optimum nanotube transistor geometry
shown in Figure 4.3c.
83
4.2 Analysis of the Frequency Response of Carbon
Nanotube Transistors
(Reproduced with permission from D. Akinwande, G. F. Close, and H.-S. P. Wong, “Analysis of the
Frequency Response of Carbon Nanotube Transistors”, IEEE Trans. Nanotech., vol. 5, 2006)
Deji Akinwande, Gael F. Close, and H.-S. Philip Wong
CENTER FOR INTEGRATED SYSTEMS AND DEPARTMENT OF ELECTRICAL
ENGINEERING
STANFORD UNIVERSITY, STANFORD, CA, 94305, USA
Abstract: The characterizations of carbon nanotube transistors at high frequencies
have so far been hindered by large parasitic and extrinsic capacitances. In this paper,
we present a quantitative analysis of the limitations imposed by probe pad parasitics
on single wall carbon nanotube transistor characterization at GHz frequencies. Our
analysis reveals the various kinds of frequency responses that can be expected to be
measured. Furthermore, we present design guidelines and a suitable device layout to
achieve gain and bandwidth at GHz frequencies.
4.2.1 Introduction
Carbon nanotube field effect transistors (CNFETs) have been extensively
characterized at DC [1-5], with emphasis on performance metrics such as the I-V
transfer curve, maximum on-current’s and minimum off-current’s, transconductance,
and subthreshold slope. Much of this attention has been motivated by the near ballistic
conduction, and high current and transconductance density of carbon nanotube (CNT)
transistors [1].
However, due to parasitic loading issues, the intrinsic high frequency
performance of CNFETs has yet to be successfully measured above a few GHz [6-11].
84
It has become necessary to understand the frequency response of CNFETs that is
achievable with current CNTs and the effect of parasitic capacitances on the measured
frequency response in order for RF CNFET research to move forward as a viable high
frequency technology.
In this paper, we provide a brief overview of the published frequency response
of CNFETs available in the literature up till now. We examine the various kinds of
frequency profiles of CNFETs that can be expected to be measured. In addition, we
present a quantitative analysis that shows the limitations imposed by probe pad
parasitic capacitance at RF. Finally, we discuss CNFET device layout issues for
improved frequency response. We will use the f-3dB as a practical figure of merit to
quantify the bandwidth of CNFETs. The f-3dB is defined as the frequency at which the
low frequency gain (S21, defined in later section) drops by -3dB.
4.2.2 Overview of Prior RF Characterization
CNFETs are very small devices with diameters on the order of a few
nanometers and lengths of tens of nanometers [1]. The CNT devices are often
contacted with metallic pads for on-wafer probing and characterization. These probe
pads can also serve as bonding pads for connection to off-chip circuits. Due to the
large size of the probe pads relative to the CNT device, the parasitic capacitance of the
pads can dominate the measured frequency response and inhibit measurement of the
intrinsic response of CNT transistors.
Several attempts to measure the RF response of CNTs operating as transistors,
rectifiers, and resistors have been reported in the literature [6-11]. Singh et al. reported
large-signal frequency measurements on CNFETs that were limited by gate and drain
pad overlap capacitance to a maximum of 200 MHz [6, 7]. Frank and Appenzeller
used RF to DC rectification to infer the frequency response of CNT transistors [8].
The maximum frequency they observed was limited to 580 MHz due to probe pads
and measurement setup limitations. Using a similar idea as [8] and with smaller probe
pads, Rosenblatt et al. used CNFETs as rectifiers up to GHz frequencies [9]. They
observed rectification behavior up to 50 GHz, though the measured output signal was
rolling off with frequency after a few GHz due to measurement setup and probe pads.
85
Small signal s-parameter measurements up to 10 GHz were reported by Huo et
al. on a CNT bundle consisting of semiconducting and metallic nanotubes [10]. The
actual response of the CNT bundle is observable up to about 500 MHz, and afterwards
a correction technique is used to subtract the effects of the probe pad capacitances and
infer the frequency response to 10 GHz. Due to the fact that the CNT bundle consisted
of metallic and semiconducting tubes and the absence of a description of their
correction technique, it is not clear what the frequency response of the device will be
if the device consists of semiconducting nanotubes only.
Even conceptually fundamental experiments with CNT resistors have faced
difficulty at RF. Li et al. attempted measurements of CNT resistance at 2.6 GHz [11]
with no conclusive data. In all of the above work involving CNFETs, no small-signal
voltage gain has been demonstrated yet due to the low output current of a single
carbon nanotube.
4.2.3 CNFET Transfer Function (S21)
A. CNFET Small-Signal Model
To understand the frequency response of CNFETs, we employ a bilateral
small-signal model including probe pad capacitances, terminal capacitances and the
voltage controlled current source.
The probe pads are usually laid out in a coplanar ground-signal-ground (GSG)
arrangement or in a signal-ground (SG) arrangement to facilitate on-wafer RF
measurements [12]. For CNT devices, the probe pads often sit on a dielectric layer
(SiO2) which is supported by a heavily doped silicon or high-resistivity substrate to
minimize substrate losses. A schematic of a CNFET with SG probe pads and the
simplified small signal model are shown in Figure 4.1.
86
Figure 4.1: CNFET simplified small signal common-source model.
(a) schematic of CNFET with SG probe pads. (b) simplified small signal model. Cgs,
and Cds, are the small signal capacitances (intrinsic+extrinsic). ro is the small signal
output resistance. Cgd is the gate to drain capacitance and it is basically the (extrinsic)
overlap capacitance (Cov) between the gate and source/drain electrodes. All the
electrodes are assumed to have negligible electrode resistance.
87
The pad parasitics consist of two capacitors to ground and a feed forward
capacitor (Cff) between input and output probe pads. Cpad represents the shunt parallel-
plate capacitance through the bottom oxide for a heavily doped substrate, or the
capacitance through the bottom oxide and substrate for a high resistivity substrate. Ct
represents the small fringing capacitance from signal pad to topside ground pad, and
will be neglected in the following analysis because the associated spacing between
signal and ground pad is much larger for Ct than for Cpad at least for heavily doped
substrates. For insulating substrates, both capacitances may need to be retained for
accurate analysis.
B. S21 Derivation
For an analysis of the transfer function of the CNFET in common-source mode
as shown in Figure 4.1, we use the 2-port S-parameter method [13], which is a
conventional technique at high frequencies. Port 1 is the gate to source terminal, and
port 2 is the drain to source terminal. To make the analysis tractable, we will use the
variable Cgs,tot to denote the total capacitance between gate and source, which includes
parasitic (probe pads), extrinsic (overlap) and intrinsic gate to source capacitances.
Likewise for Cds,tot. The Cgd,tot would comprise the intrinsic and extrinsic (overlap)
capacitance contributions plus the parasitic feed forward capacitance Cff. Assuming
the intrinsic gate to drain capacitance ≈ 0 as in conventional solid state FETs (in active
region), and a negligible Cff (we will return to this at the end), Cgd,tot=Cgd=Cov. The
overlap capacitance Cov models both the capacitance that results from direct overlap of
gate and source/drain electrodes and the fringe capacitance between the electrodes
even when there is no direct overlap between electrodes such as in self-aligned
processes [2].
The forward transfer function in a constant impedance system or S214 is defined
as the ratio of the output signal to the forward traveling input signal.
4The S21 derived in this paper is for the unmatched CNFET. For the purpose of achieving gain and bandwidth, it is not necessary to match to Z0. However, for most commercial applications (which is not the case for CNFETs at this stage in its research), matching would be required.
88
++==
1
gs
gs
2
1
221
V
V
V
V
V
VS (1)
where Vgs is the total voltage between the gate and source, and is the sum of the
incident and reflected waves.
)1( 11 Γ+= +VVgs (2)
Γ1 is the reflection coefficient at port 1. To calculate Γ1, it is necessary to determine
the input impedance of the network, which is simply the parallel combination of the
impedance due to Cgs,tot and the Miller impedance [14]. The Miller impedance (ZM) is
)1(
1
11)1(
)(1
0
0
,00
,0
ZgjwC
Zg
CjwZZgjwC
CCjwZZ
mgd
m
totds
mgd
totdsgd
M+
≈
+++
++= (3)
where w is the angular frequency of the signal, and Z0 is the characteristic impedance
of the incident signal which is typically 50 Ω. The full Miller impedance is
approximated to first order by a simple capacitance (Cgd[1+gmZ0]) which is often
referred to as the Miller capacitance in analog circuit theory. This approximation has
been verified to be valid for the frequencies of interest in this paper, that is, the -3dB
cutoff frequency of the circuit transfer function.
The input impedance (Z1) is approximately the parallel sum of two
capacitances and Γ1 can be expressed as
])1([1
])1([1
0,0
0,0
01
011
gdmtotgs
gdmtotgs
CZgCjwZ
CZgCjwZ
ZZ
ZZ
+++
++−=
+
−≈Γ (4)
Vgs can now be solved for in terms of circuit elements.
+
+++= 1
0,0 ])1([1
2V
CZgCjwZV
gdmtotgs
gs (5)
To solve for V2 we make the simplification that the load resistance is essentially the
characteristic impedance (Z0<< ro) presented by a network (s-parameter) analyzer.
gs
totdsgd
m
gd
m VCCjwZ
g
Cjw
ZgV)(1
1
,002
++
−
−≈ (6)
89
Substituting (5) and (6) into (1), S21 is computed as,
)](1][)1((1[
12
S,00,0
0
21totdsgdgdmtotgs
m
gd
m
CCjwZCZgCjwZ
g
CjwZg
+++++
−−
≈ (7)
Further simplification can be made by noting that the maximum gm reported to
date for a near-ballistic CNFET is 240 µS [2], which means gmZ0 is at least about two
orders of magnitude less than 1 for present CNFETs in a 50Ω system. Therefore,
|)](1)][(1[
12
||S|,0,0
0
21gdtotdsgdtotgs
m
gd
m
CCjwZCCjwZ
g
CjwZg
++++
−
≈ (8)
The neglect of gmZ0 in the denominator of (7) is justified since CNFETs will continue
to have low absolute transconductance until the fabrication challenge of placing
sufficiently large numbers of tubes in parallel is resolved (this issue is discussed in a
later section).
C. Low Frequency S21 Behavior [w→0]
The low frequency |S21| is approximately 2Z0gm. Next, we examine the range
of achievable S21. In the literature gm has ranged from <1 µS to about 240 µS for
Javey’s CNFET which has multiple gate fingers [2]. The gate bias voltage has often
been on the order of a volt or less. If we define small-signal input5 as 10 mV or less,
we can calculate the maximum input power to be -30 dBm. The maximum small-
signal output power is simply:
dBdBmP |S|30 212 +−= (9)
5 10mV is a rather high value for small-signal CNT excitations. However, it is a useful number because it provides a sort of maximum value to use for analysis.
90
Table 4.1: Gain and output power for a range of transconductances
gm |S21|dB *P2 (dBm)
1 µS -80 -110
10 µS -60 -90
100 µS -40 -70
240 µS -32.4 -62.4
* Small-signal input power is set to -30 dBm.
91
The values of |S21| and P2 for the experimentally achieved range of gm are listed
in Table 4.1. It is important to observe that CNFETs in common-source mode will
ordinarily provide no gain for the CNT transistors that have been reported in the
literature so far. To increase the voltage gain, many CNTs have to be placed in parallel
to get more current and a higher overall gm. Alternatively, narrow-band impedance
transforming networks [14] such as resonant circuits and transformers can be used to
boost gain. However, in this paper, we examine only the intrinsic performance
achievable with CNFETs in 50Ω systems.
Additionally, for gm’s less than 10µS, the output power will be close to or in
the noise-floor of widely used network analyzers or obscured by parasitic feed forward
paths between the input and output ports.
D. S21 Pole and Zero Analysis
Different frequency profiles can be obtained from (8) depending on the relative
weight of all the capacitances involved. To aid in identifying the possible profiles, we
perform a pole/zero analysis [15] on the S21. There is one zero frequency and two pole
frequencies that make up the transfer function. The zero frequency is
gd
mz
C
gf
π2
1= (9)
and the two pole frequencies are
)(2
1
,01
gdtotgs
pCCZ
f+
=π
(10)
)(2
1
,02
gdtotds
pCCZ
f+
=π
(11)
These three frequencies represent discrete frequencies where the S21 response changes
noticeably.
The Bode plots of four possible frequency profiles are illustrated in Figure 4.2.
In conventional solid-state amplifiers, the first pole frequency is typically the -3dB
frequency (if it is a dominant pole), and often implies a monotonic decline in the
frequency response from low frequencies up until f-3dB. For CNFETs, the -3dB
92
frequency may not imply such a monotonic response due to the zero that may be
significant and can occur before the first pole frequency. Comparing fp2 against fp1,
gdtotds
gdtotgs
p
p
CC
CC
f
f
+
+=
,
,
1
2 (12)
we conclude that fp2 ≥ fp1 since Cgs,tot ≥ Cds,tot. In the same way, comparing fz against
fp1,
+=
gd
gdtotgs
m
p
z
C
CCZg
f
f ,0
1
(13)
we observe that the zero frequency can be less than the pole frequency which is almost
never the case for conventional solid-state common-source amplifiers, except for very
narrow transistor (<< 1µm) . The reality that the zero frequency can be less than the
pole frequency of CNFETs is fundamentally due to the lack of intrinsic RF gain
(gmZ0) for current CNFETs, i.e., gmZ0<<1. Indeed Singh et al. observed this feed-
forward zero in their measurements [6, 7].
Furthermore if we compare the transit frequency (ft=gm/[2π(Cgs+Cgd]) to the
pole frequency, we have
+
+=
gdgs
gdtotgs
m
p
t
CC
CCZg
f
f ,0
1
(14)
which leads to the surprising observation that the transit frequency can be less than the
first pole frequency due to the lack of intrinsic gain for present CNFETs. For analog
RF transistors, ft is often interpreted as the upper limit on the realizable f-3dB bandwidth.
For a transistor with no gain (essentially an attenuator), the transit frequency (though
well defined as the unity current gain frequency) may not be as meaningful a figure of
merit for RF CNFET amplifiers in 50Ω systems, since f-3dB bandwidths greater than ft
is achievable for the specific case of gmZ0<<1.
93
Figure 4.2: Illustrated Bode plots of four possible CNFETs frequency response.
(a) The first pole occurs before the zero. (b) The two poles are (approximately)
degenerate and occur before the zero. (c) The zero occurs before the poles. (d) Same
as in (c) but the two poles are (approximately) degenerate.
94
4.2.4 Effect of Pad Capacitance
Many published measurements of CNFETs have been limited by pad
capacitances [8-10]. This is the case when Cgd is negligible leading to a response
similar to Figure 4.2b, and the resulting S21 is
]1][1[
2
,0,0
021
totdstotgs
m
CjwZCjwZ
ZgS
++
−≈ (15)
However, it has not yet been determined nor reported in the literature what the value
of Cds,tot is for CNFETs. Nevertheless, it is reasonable to take Cgs,tot as the upper bound
on Cds,tot (i.e., Cds,tot ≤ Cgs,tot) as in conventional transistors. This assumption is
certainly true when the pad capacitance dominates the input and output capacitance.
With this simplification, S21 can be reduced to
|]1[|
2||
2,0
021
totgs
m
CjwZ
ZgS
+≈ (16)
Cgs can approximately be written as
ov
top
cnttopo
cntQgs Cdt
llCC +
++≈
− )/21(cosh
21
επε (17)
where lcnt and d are the length and diameter of the CNT respectively, and εtop and ttop
are the dielectric constant and thickness of the top oxide (or high-k dielectric)
respectively. εo is the permittivity of air. The first and second terms models the
quantum (CQ=0.4 fF/µm) and electrostatic capacitance respectively [16]. The third
term is the overlap or fringe capacitance and is best determined by electrostatic
simulation in order to get the most accurate values.
It is useful to compute a practical range of values for the -3dB frequency of
(16). For thin bottom oxides (tbot≤1µm), Cpad typically dominates the input and output
capacitances of a single CNFET. The -3dB frequency is
pad
dBCZ
f0
31.0
≥− (18)
95
The inequality arises due to the upper bound placed on Cds,tot in the preceding analysis.
Cpad can be calculated using the parallel-plate approximation since the width and
length of probe pads are often much greater than the bottom oxide thickness.
bot
botopad
t
AC
εε= (19)
where A is the area of the pad, and εbot (3.9) is the oxide dielectric constant of SiO2. A
widely used CASCADE® recommended general purpose pad size is 100 µm x 100 µm
with a 150 µm pitch [17]. The bottom oxide thickness has ranged from about 10nm to
about 1µm for CNFETs that have so far been reported in literature. For these range of
values, the calculated pad capacitances and -3dB frequencies are listed in Table 4.2.
Above the pole frequency, the signal rolls off approximately -12dB/octave. If
CASCADE recommended minimum probe pad sizes of 25 x 35 µm are used, the pole
frequencies in Table 4.2 will theoretically increase by a maximum factor of 11.4.
However, that increase may not fully materialize in reality because the assumption of
negligible Cgd may no longer be valid.
It is clear from Table 4.2 that very thick oxides (or high resistivity substrates)
can enable measurements of CNFETs with f-3dB bandwidths in the GHz range even
with the probe pad parasitic capacitance included provided that Cgd is negligible.
4.2.5 Design Guidelines and Layout Issues for High Frequency CNFET
In order to characterize CNFET with gain up to a certain frequency, we can use
as design variables the oxide thickness, probe pad dimensions, length of tubes, and
number of gate fingers (n) of the CNFET. A practical device layout for RF
applications is shown in Figure 4.3c. Cgs, Cgd, and gm are assumed to scale
proportionally with the number of fingers, and the substrate is heavily doped.
96
Table 4.2: S21 (-3dB) frequencies for a range of tbot
tbot Cpad f-3dB
10 nm 34.5 pF 58 MHz
100 nm 3.45 pF 580 MHz
1 µm 345 fF 5.8 GHz
97
ga
ted
rain
source
source
gate
dra
in
ga
te
so
urc
e
S D
G
Top oxide
SiO2
Low-loss or Insulating
substrate
CNT a)
b)
d)
c)
Figure 4.3: Representative layouts of a carbon nanotube transistor.
a) Cross-section of a self-aligned CNFET. b) Top view of a conventional CNFET [2]
with SG electrode arrangement showing three fingers. c) Top view of the proposed
improved CNFET with GSG electrode arrangement and a zoomed-in version showing
four fingers. An extra metal layer is needed to seperate the source and drain electrodes.
d) A 3-D view of the CNFET showing just one finger. Wg,eff determines the
transconductance while We determines the gate overlap capacitances. We→Wg,eff to
minimize overlap capacitance.
98
This layout has the advantage of providing multiple gate fingers and a high
transconductance, and is a slightly modified version of [2]. Compared to [2], the gate
width per finger of the improved layout must be kept to a minimum because any extra
width beyond the effective gate control width (Wg,eff in Figure 4.3c) adds no further
transconductance but increases the gate overlap or fringe capacitance therefore
reducing the bandwidth of the device. Unlike conventional transistors, gm for CNFETs
does not scale with gate width beyond Wg,eff.
A frequency plot of |S21| from (7) as a function of the number of fingers is
shown in Figure 4.4, using a transconductance of 30µS/tube [2], and the values listed
in Table 4.3. It is evident from the plots that CNFETs with gain may have bandwidths
in the low GHz frequencies. Also, an important observation is that hundreds of gate
fingers are needed in order for the CNFET to possess gain in a 50 Ω system.
Table 4.3 reveals that the intrinsic capacitances/finger (quantum and electrostatic gate
to source values) of the nanotube itself are negligible compared to the extrinsic and
parasitic capacitances. This is a consequence of the fact that the gate electrode width
(5µm) for the conventional layouts (e.g. in [2]) is three orders of magnitude larger than
the CNT diameter (3nm) leading to relatively large extrinsic overlap capacitance. This
has been the case so far for published CNFETs largely due to the ease of fabricating
micrometer wide gate widths.
For this set of dimensions, switching to a high resistivity silicon substrate does
not have a significant influence on the -3dB bandwidth because for low n, the zero due
to Cgd dominates the response and for large n, the pole due to the total gate to source
overlap capacitance dominates. These capacitances are independent of the substrate.
If nanometer scale gate width/finger is used (e.g. in the improved layout, Figure 4.3c
with We≈Wg,eff), the overlap capacitances between the gate and electrodes would drop
significantly and the measurable bandwidth would increase appreciably. Figure 4.5
shows the bandwidth for n=1000 and for the same set of dimensions as in Figure 4.4
but with the gate width/finger swept from 10nm to 10µm.
99
107
108
109
1010
1011
S2
1 (
dB
)
Frequency (Hz)
n=1000
n=1
n=10
n=50
n=100
-60
-50
-40
-30
-20
-10
0
10
n=500
Figure 4.4: Frequency response of the gain of common-source CNFETs plotted for
several values of n. Gate width (We) is 5µm.
100
Table 4.3: Capacitances per finger for design example
Parameter Description Value*
Cov Gate-Source overlap
capacitance
1 fF
Cgd Gate-Drain overlap
capacitance
1 fF
Cgs,Q Gate-Source
quantum capacitance
0.02 fF
Cgs,ES Gate-Source
electrostatic
capacitance
0.016 fF
* The dimensions used are Wpad=50µm, lcnt=50nm, tbot=2µm, We=5µm, εtop=15,
ttop=8nm, and d=3nm. The gate overlap capacitance was determined by electrostatic
simulation to be 0.2fF/µm using a device geometry similar to [2].
101
101
102
103
104
8
9
f-3
dB
(H
z)
Gate width/finger (nm)
n=1000
10
10
10
10
10
11
Figure 4.5: f-3dB of the common-source CNFET (with n=1000) plotted as a function of
the gate width/finger for the same dimensions as in Table 4.3.
102
The transconductance and the capacitance values in Table 4.3 are taken to be
the same with the exception of the gate overlap capacitance which scales in proportion
to the gate width (We). For this set of dimensions, the -3dB bandwidth scales linearly
as the gate width/finger reduces from 10µm to 1µm and quasi-linearly between 1µm
and 100nm due to the growing impact of the pad capacitance. Below 100nm, Cpad is
the dominant contributor to Cgs,tot and the bandwidth becomes roughly independent of
gate width/finger.
In general, two factors can be identified as responsible for limiting the
improvement in bandwidth that can be measured as the gate width/finger is scaled
down. One is the parasitic feed-forward capacitance (Cff in Figure 4.1a) between the
input and output probe pads which appear in parallel with Cgd and hence sets a
minimum limit on the gate to drain capacitance. Another factor is Cpad which
contributes two poles to the frequency response and limits the bandwidth even if the
gate to source/drain overlap and feed forward capacitances are eliminated altogether.
4.2.6 Conclusion
We have analytically examined the various frequency profiles that can be
expected from current state of the art CNFETs. Our analysis shows the effects of
probe pad and overlap capacitances on the measurable -3dB bandwidth of carbon
nanotube FETs.
The potential of CNTs as a viable RF transistor technology will depend on
demonstration of CNFETs with gain at GHz frequencies. For this to be achieved,
many nanotubes would have to be fabricated precisely in parallel, which is currently a
matter of active research. Moreover, techniques to minimize the gate overlap
capacitance will have to be explored so as to achieve GHz bandwidths.
Acknowledgments
Many thanks to Professors Hongjie Dai and Boris Murmann at Stanford University for
useful discussions. This work is supported in part by the Powell Foundation and the
FENA MARCO Focus Center Research Program.
103
4.3 Carbon Nanotube Quantum Capacitance for Non-
Linear Terahertz Circuits
(Reproduced with permission from D. Akinwande, Y. Nishi, and H.-S. P. Wong, “Carbon Nanotube
Quantum Capacitance for Non-Linear Terahertz Circuits”, IEEE Trans. Nanotech., vol. 8, 2009)
Deji Akinwande, Yoshio Nishi, and H.-S. Philip Wong
CENTER FOR INTEGRATED SYSTEMS AND DEPARTMENT OF ELECTRICAL
ENGINEERING
STANFORD UNIVERSITY, STANFORD, CA, 94305, USA
Abstract: Analog circuit applications of the non-linear carbon nanotube quantum
capacitance such as frequency doublers and mixers are proposed. We present a
balanced circuit implementation and derive the transconductance conversion gain for
the non-linear carbon nanotube quantum capacitor circuit. The balanced topology
results in robust circuit performance that is insensitive to extrinsic capacitances and
parasitic resistances and is immune to the resistance of metallic nanotubes that may
be in the channel. The ballistic quantum capacitance is useful up to several terahertz,
making it suitable for low-noise THz sources. Additionally, the fundamental bandwidth
and performance limitations imposed by the quantum conductance and inductance are
discussed.
4.3.1 Introduction
Carbon nanotubes (CNTs) are tubular cylindrical arrangement of carbon atoms
with diameters on the order of 1 to 5 nm [18]. They have characteristic quantum
features such as finite density of states and van-Hove singularities due to the unique
physics of one-dimensional solids. These quantum features are real and measurable
and manifest in electrical circuits as quantum conductance, quantum capacitance and
the kinetic inductance [16, 19, 20]. The impact of the quantum conductance and
104
capacitance on electronic circuits continues to enjoy increasing understanding and as
such have been included in comprehensive, compact, and RF models of CNT
transistors [21-25]. Recent measurements of the quantum capacitance have confirmed
theoretical predictions [26, 27] and raised interesting questions as to what kind of
circuits in analog, digital, and memory systems could utilize the rich features of the
quantum capacitance for original and existing applications. One important circuit
function is to use the non-linear property of the quantum capacitance for terahertz
(THz) frequency multipliers and sources with applications ranging from
communications to security [28].
In this paper, we present the first analytic exploration and circuit simulations of
non-linear circuits exploiting the CNT quantum capacitance such as frequency
multipliers and mixers including parasitic effects. Previous work had reported mixing
with CNT transistors, taking advantage of the non-linearity inherent in a classical
square-law transistor [9]. Here, our focus is on using the novel quantum capacitance
directly and we present a circuit model, practical circuit implementations, and
discussion of the performance metrics and bandwidth limitation imposed by the finite
quantum conductance and kinetic inductance. When operating in the ballistic region,
the CNT dissipates no energy in the channel, hence generates no thermal noise,
resulting in intrinsically low-noise multipliers. Sustained exploration of the quantum
capacitance for circuit applications is a fruitful path for advancing CNTs as a state of
the art nano-electronic device.
4.3.2 CNT Capacitor Circuit Model
In linear capacitors, the charge is linear with voltage resulting in a constant
capacitance. The semiconducting CNT quantum capacitance is a non-linear capacitor
(NLC) which essentially reflects the CNT density of states [20], in contrast to a
MOSCAP or classical diode varactor which is due to a dynamic geometric capacitance.
As such, NLCs are naturally useful for a variety of non-linear circuits [14, 29]
including varactors, and frequency multipliers.
The CNT NLC is electrically visible and measurable in the standard top-gated
transistor geometry [26] illustrated in Figure 4.6, where ideally the CNT device will be
105
embedded in an integrated circuit and avoid direct exposure to the large parasitic
capacitances of probing or bonding pads. Additionally, the source and drain are
shorted together to form a two-terminal device which we henceforth call the CNT
NLC device for clarity in the rest of this work. For sufficient current drive and
performance at high frequencies, the NLC device will contain a dense array of
perfectly aligned identical semiconducting nanotubes of the order of a 100 CNTs/µm,
which is about four times greater than current state of the art peak densities grown on
crystalline quartz substrates [30]. It is quite fortunate that perfectly aligned arrays of
CNTs (Figure 4.6b) grow naturally on quartz substrates, which is the substrate of
choice for low-loss, temperature stable GHz and THz integrated circuits [31].
An equivalent circuit model of the CNT NLC device (shown in Figure 4.6) is a
lumped RLC model consisting of a voltage dependent capacitor (or varactor), a fixed
geometric capacitance, the kinetic inductance, series resistors and extrinsic
capacitances as shown in Figure 4.7a. For the case of a thick insulating substrate, the
substrate capacitances are negligible since the electric fields from the source/drain
electrodes are significantly screened by the gate electrodes and vice-versa due to their
relatively closer proximity. The screening effect simplifies the model by neglecting
the substrate capacitances as shown in Figure 4.7b. For clarity of discussion, we call
the model in Figure 4.7b the screened RLC model, and that of Figure 4.7a the
unscreened RLC model. We discuss quantitatively in §4.3.3 that the electrode to
substrate capacitance of the same order as the NLC has a negligible effect on circuit
performance.
106
Figure 4.6: Optimum device geometry for CNT quantum capacitance circuits.
(a) A top-gate CNT device cross-section with an intrinsic channel, useful for non-
linear analog circuits. The thin high-k dielectric enhances the contribution of the
quantum capacitance to the total gate capacitance, and the CNT extension regions are
heavily doped to minimize the contact resistance. Also, the substrate is insulating and
thick to minimize substrate capacitances. (b) On the bottom left is a top view of (a)
with dense aligned identical CNTs which are needed in the device to maximize the
current drive and performance at high frequencies. The gate (G), source (S), and drain
(D) are of nanoscale dimensions to reduce coupling capacitances, and the source and
drain are shorted together to make a two-terminal device. (c) Scanning electron image
of aligned arrays of nanotubes grown on quartz substrates.
107
Cg
Cq
Rg
Rc
Re
2Cgs
Cg_sub
2Cs_sub
Lk
Intrinsic
CNT model
Cg
Cq
Rg
Rc
Re
2CgsLk
a) b)
Figure 4.7: Two-port circuit models for a CNT quantum capacitance device.
(a) A lumped circuit model for the CNT NLC device including the quantum, Cq, and
geometric, Cg, capacitances, the contact (Rc), and the kinetic inductance (Lk) which are
the intrinsic contributions to the model. The extrinsic and parasitic components
include, the gate resistance (Rg), the source/drain interconnect resistance (Re), the gate
to source capacitance (Cgs), the gate to substrate capacitance (Cg_sub), and the source to
substrate capacitance (Cs_sub). For thick insulating substrates, the substrate
capacitances are negligible due to screening effects resulting in the simplified screened
RLC model shown in (b).
108
The goal of this work is to explore the non-linear analog circuit applications of
nanotubes NLC, and estimate the performance metrics. For that purpose, it is
worthwhile to briefly discuss the contact resistance (Rc) and kinetic inductance (Lk) of
the nanotube extensions. In the ideal case of a defect free CNT that is shorter than the
mean free path (typically >0.5µm) [32], the contact resistance approaches the ballistic
value of Rq/2, where Rq=6.5kΩ and the factor of two is due to the shorting of the
source and drain electrodes. An additional benefit of the shorting is that any metallic
CNT in the array is automatically shorted and its resistance becomes irrelevant.
Moreover, as long as the ratio of semiconducting to metallic CNTs is large enough,
the constant quantum capacitance of metallic CNTs contributes negligibly to the total
Cq. In the ballistic limit, the CNT generates no thermal noise since Rq is not an Ohmic
resistance but a fundamental quantum resistance [33], which is needed in the model to
account for the time constant associated with Cq and Lk, governed by the energy time
uncertainty relations [34]. In a non-ballistic CNT device, the electrical performance
and noise will degrade accordingly.
The value of the kinetic inductance for semiconducting CNTs is currently a
matter of research and generally non-constant. However, it is reasonable to expect the
inductance to be of the same order as that of a metallic CNT, which is ~3.2nH/µm.
Since the intrinsic CNT model is capacitive at frequencies below the self resonant
frequency (wSRF=(LkCNLC)-1/2, where CNLC=Cq||Cg), the precise variation of the kinetic
inductance is not that important at lower frequencies. As a result, we use a constant
value of 3.2nH/µm for Lk and this allows us to estimate the maximum frequency where
the screened RLC model is valid. At frequencies beyond wSRF, the precise value of the
inductance and its variation with gate voltage will have to be taken into account.
The critical element for non-linear analog circuit applications is the quantum
capacitance, and moreover the series combination of Cq and Cg generally retains the
non-linear property of Cq. In this work we use measured data for the NLC (CNLC) to
represent practically achievable values for the CNT capacitance leading to reasonable
estimates for circuit performance. Figure 4.8 shows the measured CNLC vs. Vg (gate
voltage) of a semiconducting CNT reflecting the quantum capacitance [26], which is
109
generally non-analytical. However, for small-signals where the gate voltage is suitably
restricted, it is possible to describe the C-V by simple analytical equations. An
example of a linear C-V approximation of the first sub-band is shown superimposed
on Figure 4.8. The linear C-V leads to a charge density that depends quadratically on
voltage. It is worthwhile to note that this linear C-V region is the most suitable for
non-linear analog circuits. At high gate voltages, the capacitance saturates to the
geometric capacitance. The linear C-V is also potentially useful for voltage controlled
oscillators (VCO), and favorable in varactor tuned phase locked loop circuits.
4.3.3 Quantum Capacitor Analog Circuits
In this section we explore the use of the quantum capacitor for two specific
non-linear analog circuits, one is the frequency doubler and the other is the mixer. For
hand analysis which requires simplicity, we calculate analytically the ideal
performance for the CNT NLC device ignoring all resistors, the kinetic inductance,
and the extrinsic Cgs of the screened RLC model (Figure 4.7b). Afterwards, we
simulate actual circuit implementations including the non-idealities. Because the NLC
dominates the circuit dynamics, the agreement between simulation and hand analysis
is strong.
There are two intrinsic time constants or poles of importance in the screened
RLC model impacting circuit performance dramatically. For simplicity of notation, we
will use the radial frequency (w) and actual frequency (f) interchangeably, where
w=f/(2п). One pole is the self resonant frequency, fSRF. Using a value of ~30aF/µm for
CNLC (center of the linear C-V region of Figure 4.8), and 3.2nH/µm for Lk, fSRF ~0.52
THz µm. The other pole is due to the RC time constant defined as fRC=(2пRcCNLC)-1
~1.6 THz µm with ballistic values of 3.25KΩ for the contact resistance. For a 100nm
channel length which is the CNT length used in this work, fSRF ~5.2 THz and fRC ~16
THz. Shorter lengths result in even higher poles, which indicate that the CNT NLC
device offers larger bandwidth compared to conventional solid-state Schottky barrier
diodes [28, 35].
110
0
10
20
30
40
50
60
70
80
-2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
Vtg
Cto
t (a
F/u
m)
Measured Data [Ilani et at.]
Linear Fit
Figure 4.8: Measured NLC capacitance (including the quantum, Cq, and geometric, Cg,
contributions) vs top-gate voltage of a CNT device with source/drain shorted [26].
For convenience the C-V has been horizontally re-centered such that the bottom is at
0V. The linear fit has a slope m=122aF/Vµm, and intercept Co=-26aF/µm.
111
For frequencies much less than the dominant pole, the screened RLC model can be
approximated by the single non-linear capacitor (CNLC) for hand analysis with
negligible error, which greatly reduces the circuit’s differential equations to an
algebraic analysis.
In the linear region, the non-linear capacitor (CNLC) is
ogNLC CtmVtC +≈ )()( (1)
where m is the slope of the linearized C-V curve, and Co is the offset. The dynamic
gate current is:
dt
tdVC
dt
tdVmdVC
dt
d
dt
dQti
g
o
g
gNLCg
)()(
2
1)(
2
+=∫== (2)
The first term leads to interesting non-linear behavior because of the quadratic
dependence on voltage, while the latter term is the classical displacement current.
A. Quantum Capacitor Frequency Doublers
Frequency doublers are circuits that double the input frequency. Consider an
input sinusoid
DCog VwtVtV += )sin()( (3)
where VDC is required to bias the device at the center of the linear region, and Vo is the
amplitude. The gate current per nanotube is
)cos()()2sin(2
)( 2wtwVCmVwtmV
wti ooDCog ++= (4)
The first term is the desired doubled signal and the latter term is the feed-through and
is typically cancelled by using balanced circuits [14, 29]. For N identical CNTs, the
total gate current is Nig(t). For maximum current, the capacitor should be fully driven,
i.e., Vo should swing over the entire linear range of the C-V. The primary figure of
merit (FOM) is the transconductance conversion gain or loss (Gc) which is the
magnitude ratio of the doubled output signal to the input signal.
o
g
g
c mVw
wV
wiG
2|)(|
|)2(|== (5)
For maximum conversion gain, the device would ideally have a large slope (m)
and be operated at high narrowband frequencies with the maximum voltage swing. It
112
is useful to define a quantum transconductance conversion gain (Gq), which occurs in
the quantum limit when CNLC≈ Cq, and EF≈ eVg, where EF is the CNT Fermi energy or
chemical potential and e is the electronic charge. The quantum limit is approached
when Cg >> Cq,peak, where Cq,peak at room temperature is of the same order as the
metallic CNT quantum capacitance (~0.3fF/µm [20]). Practically, this could be
achieved with a 3nm thick TiO2 gate dielectric (εr≈80), resulting in Cg≈2fF/µm for a
1.5nm wide CNT. In the quantum limit, Gq can be considered a FOM that provides
design intuition and is an upper bound on the achievable gain. For this case, m in (1)
can be obtained by using the quantum capacitance equation derived in [27].
)cosh(4
)( 22
kT
eVe
kT
NeCVC
gkT
E
cqgNLC
g−
=≈ (6)
where Nc is the effective density of states in units of 1/meter, kT is the thermal energy,
and Eg is the bandgap energy . The C-V slope is
)()tanh(| DCqDC
q
g
q
V VCkT
e
kT
eVC
kT
e
V
Cm
DC≈=
∂
∂= (7)
where the hyperbolic tangent has been approximated as unity for almost all practical
values of its argument since VDC will typically be hundreds of millivolts and several
times greater than kT (e.g tanh(>2)>0.96). The input signal amplitude must always be
a fraction of the bandgap, eVo=αEg, where α is less than one. Therefore, from (5)
)(2 DCq
g
q VCkT
EwG α= (8)
which shows that circuit performance is explicitly related to the normalized band gap.
Furthermore, the dependence of (8) on the quantum capacitance provides insight on
the gain and bandwidth product (Gq fSRF) in the quantum limit (assuming the self
resonant frequency is the dominant pole),
k
g
k
qg
SRFqLkT
E
L
C
kT
EwfG
π
α
πα
84<= (9)
The inequality arises by noting that w<wSRF/2. Eqn. (9) reveals that the largest gain
bandwidth is obtained with the largest ratio of Eg/Lk.
A practical implementation of a frequency doubler utilizing a CNT NLC is
shown in Figure 4.9. This circuit is widely known as a balanced circuit, [14, 29]
113
because the differential and common mode output currents can be separated and
routed differently. In the specific output connection of Figure 4.9, the common-mode
(desired) output currents are coherently routed to the output load (RL) while the
differential mode (unwanted) currents cancel. For an input oscillating at 500GHz, and
the output at 1THz, the simulated output power of -51.9dBm compares very well with
the estimate of -52.3dBm using (5). The power conversion loss (input power-output
power) is -13.9dB. The CNLC slope and intercept are 12.2aF/V and -2.6aF per nanotube
respectively (from Figure 4.8). The density of identical CNTs used in the balanced
circuit is 100 CNTs/µm and the NLC device width is 100nm yielding a total of 10
CNTs/device.
It is worthwhile to discuss the values of the extrinsic capacitance Cgs and the
parasitic resistances Re and Rg used in Figure 4.9. Due to the balanced topology of the
circuit, the currents flowing through Cgs is out of phase (differential mode) and
therefore cancels at the output load. As a result, the effect of Cgs is negligible and we
chose a value that is equal to the average CNLC for convenience. However,
transformers are not perfect in practice; therefore there will be some residual Cgs
current at the load. The magnitude of this current will depend on the degree of
imbalance in the transformer and the value of Cgs. For this reason it is typical to
include a bandpass filter to further suppress undesired signals. Likewise, the parasitic
resistances have negligible effect on the output signal because their practical values
are typically orders of magnitude smaller than the CNT contact resistance. In the
circuit simulation, Re and Rg are set to 10Ω each, a reasonable value for small footprint
nanoscale dimensions.
114
1 2 30 4
-2
0
2
-4
4
1 20 3
-90
-70
-110
-50
Figure 4.9: Proposed carbon nanotube quantum capacitance frequency doubler circuit.
(a) Schematic of a balanced frequency doubler utilizing a 0.1µm long CNT quantum
capacitor. The non-linear admittance (Ycnt) represents the screened RLC model of
Figure 4.7b. L1 (24pH), L2 (1.6nH), and C1 (2.9fF) are for resonant matching to the
source (Rsrc=50Ω). The center tapped balun is an ideal 1:1 transformer and RL is the
50Ω load. CNLC is given by (1) with parameters m=12.2aF/V and Co=-2.6aF at
VDC=0.47V. The source (Vsrc) oscillates at 500GHz with an amplitude Vo=0.2V. (b)
Agilent ADS™ harmonic balance simulated waveform. The insert is the output
spectrum with -51.9dBm at 1THz, and total Gc=80.2uS. The finite harmonics above
1THz are due to the contact resistance and kinetic inductance.
115
Furthermore, the substrate capacitance Cs_sub of the unscreened RLC model
also has negligible effect on the conversion gain provided its frequency pole (fs_sub) is
much larger than the operating frequency. As a rule of thumb, a good circuit design
would require the pole frequency to be much larger than the dominant circuit
frequency pole, for example fs_sub≥10fSRF ≈50THz. This results in an (open circuit time
constant) estimate of Cs_sub≈30aF≈10CNLC(VDC). This value of substrate capacitance
leads to less than 3% reduction in Gc. Similar values of Cg_sub≈10CNLC(VDC) causes
about 10% drop in the conversion gain.
B. Quantum Capacitance Frequency Mixers
The purpose of a frequency mixer is to take two inputs at different frequencies
and output two signals that are the arithmetic sum and difference of the input
frequencies. For an input that consists of the information carrier frequency (wc) and
the local oscillator (wlo, wlo>wc),
DCloloccg VtwVtwVtV ++= )sin()sin()( (10)
the output current computed from (2) is the sum of currents from the fundamental
oscillations, the second harmonic oscillations, and the oscillations arising from mixing.
The first two terms are undesired and cancelled by using a balanced circuit [14]. The
mixing current (imx) is:
])sin()()sin()[(2
)( twwwwtwwwwVmV
ti clocloloclocloc
mx −−−++= (11)
The upper sideband and lower sideband conversion gains are respectively,
loloc
cg
locmxusb mV
ww
wV
wwiG
2
)(
|)(|
|)(| +=
+= (12a)
loclo
cg
clomxlsb mV
ww
wV
wwiG
2
)(
|)(|
|)(| −=
−= (12b)
For most applications only one of the two output frequencies is selected. To obtain
maximum conversion gain, the same considerations mentioned above for frequency
doublers hold as expressed in (8) and (9). Figure 4.10a is a schematic of a double
balanced mixer, utilizing the CNT quantum capacitance.
116
0.5 1.0 1.5 2.0 2.50.0 3.0
-125
-100
-75
-150
-50
Freq. (THz)
Power (dBm)
RL
out
Ycnt
Vlo
Rsrc
Ycnt
Ycnt
Ycnt
Vc
-Vc
C1
L2L1
-Vlo
Rsrc
C1
L2L1
a)
b)
Figure 4.10: Proposed carbon nanotube quantum capacitance frequency mixer circuit.
(a) Schematic of a double balanced upconverter frequency mixer utilizing a 0.1µm
long CNT quantum capacitor. The non-linear admittance (Ycnt) represents the screened
RLC model of Figure 4.7b. CNLC is given by (1) with parameters m=12.2aF/V and
Co=-2.6aF at VDC=0.47V which is coupled to Vc. The two input local oscillators are
driven differentially at 500GHz (|Vlo|=200mV), and likewise for the two input carrier
sources at 10GHz (|Vc|=50mV). The differential inputs are necessary to provide
rejection for undesired output currents. The center tapped baluns are ideal 1:1
transformers, and the 50Ω input matching network (L1, C1, L2) have identical values
as in Figure 4.9. (b) Agilent ADS™ harmonic balance simulated output spectrum in
good agreement with analytical estimates from (5) for the desired mixer output at
490GHz and 510GHz. The undesired harmonics and spurs (<-40dBc) are due to the
contact resistance and kinetic inductance.
117
The carrier oscillates at 10GHz, the local oscillator is at 500GHz, and the two
mixer output signals are at 490GHz and 510GHz. The CNLC slope and intercept are
12.2aF/V and -2.6aF per tube respectively (from Figure 4.8) for a channel length of
0.1µm. The density of identical CNTs used in the circuit is 100 CNTs/µm and the
NLC device width is 100nm yielding a total of 10 CNTs per device.
The output spectrum of the mixer obtained by harmonic balance simulation is
shown in Figure 4.10b. The simulated output power of about -58dBm is within 0.5dB
of the estimate from (12). The effects of the parasitic capacitances and resistances
discussed for the doubler circuit also hold for the mixer circuit with negligible effect
on output spectrum for operating frequencies much less than the poles of the parasitic
capacitances.
4.3.4 Conclusion
Non-linear circuits such as frequency doublers and mixers exploiting the
quantum capacitance of carbon nanotubes have been analyzed, showing good
agreement with circuit simulations based on a measured CNT quantum capacitor
device. The CNT non-linear circuits show robust transconductance conversion gain
that is immune to extrinsic capacitances and parasitic resistances. The functionality of
the proposed CNT circuits is achievable with current technology involving aligned
nanotube arrays on quartz substrates. The main potential of quantum capacitor non-
linear devices is the high achievable bandwidths in the THz range, and the vanishing
thermal noise under ballistic transport, making them ideally suitable for THz sources.
As a result, it is expected that the CNT quantum capacitor will be a future
nanoelectronic alternative to solid state diode multipliers.
Acknowledgements
The authors would like to thank Dr. Shahal Ilani and Prof. Paul McCuen (Cornell
University) for providing CNT quantum capacitance data [26]. This work is supported
in part by the Focus Center Research Program (FENA and C2S2). Additional support
118
from the Ford foundation and the Alfred P. Sloan foundation are much appreciated by
D. A.
4.4 References
[1] P. Avouris, J. Appenzeller, R. Martel, and S. J. Wind, "Carbon nanotube electronics," Proc. of
the IEEE, vol. 91, pp. 1772-1784, 2003.
[2] A. Javey, J. Guo, D. B. Farmer, Q. Wang, E. Yenilmez, R. G. Gordon, M. Lundstrom, and H. J.
Dai, "Self-aligned ballistic molecular transistors and electrically parallel nanotube arrays,"
Nano Letters, vol. 4, pp. 1319-22, 2004.
[3] J. Chen, C. Klinke, A. Afzali, and P. Avouris, "Self-aligned carbon nanotube transistors with
charge transfer doping," Applied Physics Letters, vol. 86, pp. 123108-3, 2005.
[4] J. U. Lee, P. P. Gipp, and C. M. Heller, "Carbon nanotube p-n junction diodes," Applied
physics letters, vol. 85, pp. 145-147, 2004.
[5] A. Javey, Q. Wang, W. Kim, and H. Dai, "Advancements in complementary carbon nanotube
field-effect transistors," in IEEE IEDM Technical Digest, 2003, pp. 31.2.1-31.2.4.
[6] D. V. Singh, K. A. Jenkins, J. K. A. Appenzeller, D. K. A. Neumayer, D. A. Grill, and D. H. S.
P. Wong, "Frequency response of top-gated carbon nanotube field-effect transistors," IEEE
Transactions on Nanotechnology, vol. 3, pp. 383-7, 2004.
[7] D. V. Singh, K. A. Jenkins, and J. Appenzeller, "Direct measurements of frequency response
of carbon nanotube field effect transistors," Electronics Letters, vol. 41, pp. 280-282, 2005.
[8] J. Appenzeller and D. J. Frank, "Frequency dependent characterization of transport properties
in carbon nanotube transistors," Applied Physics Letters, vol. 84, pp. 1771-1773, 2004.
[9] S. Rosenblatt, H. Lin, V. Sazonova, S. Tiwari, and P. L. McEuen, "Mixing at 50 GHz using a
single-walled carbon nanotube transistor," APL, vol. 87, p. 153111, 2005.
[10] X. Huo, M. Zhang, P. C. H. Chan, Q. Liang, and Z. K. Tang, "High frequency S parameters
characterization of back-gate carbon nanotube field-effect transistors," in IEEE IEDM
Technical Digest, 2004, pp. 691-694.
[11] S. Li, Z. Yu, S.-F. Yen, W. C. Tang, and P. J. Burke, "Carbon Nanotube Transistor Operation
at 2.6 GHz," Nano Letters, vol. 4, pp. 753-756, 2004.
[12] Cascade Microtech Application Note, "Introduction to Bipolar GHz Measurement Techniques"
[13] D. M. Pozar, Microwave engineering, 3rd ed.: John Wiley, 2005.
[14] T. H. Lee, The design of CMOS radio-frequency integrated circuits, 2nd ed.: Cambridge
University Press, 2004.
[15] A. S. Sedra and K. C. Smith, Microelectronic circuits, 5th ed.: Oxford University Press, 2004.
119
[16] P. J. Burke, "An RF circuit model for carbon nanotubes," IEEE Trans. on Nanotechnology, vol.
2, pp. 55-58, 2003.
[17] Cascade Microtech Application Note, "Mechanical Layout Rules for Infinity Probes".
[18] S. Iijima and T. Ichihashi, "Single-shell carbon nanotubes of 1-nm diameter," Nature, vol. 363,
pp. 603-605, 1993.
[19] S. Luryi, "Quantum capacitance devices," Applied Physics Letters, vol. 52, pp. 501-503, 1988.
[20] D. L. John, L. C. Castro, and D. L. Pulfrey, "Quantum capacitance in nanoscale device
modeling," Journal of Applied Physics, vol. 96, pp. 5180-5184, 2004.
[21] J. Deng and H.-S. P. Wong, "A Compact SPICE Model for Carbon Nanotube Field Effect
Transistors Including Non-Idealities and Its Application Part II: Full Device Model and
Circuits Performance Benchmarking," IEEE Trans. Electron Devices, vol. 54, pp. 3195-3205,
Dec. 2007 2007.
[22] J. Guo, A. Javey, H. Dai, and M. Lundstrom, "Performance analysis and design optimization
of near ballistic carbon nanotube field-effect transistors," in IEEE IEDM Tech. Dig., 2004, pp.
703-706.
[23] D. Akinwande, G. F. Close, and H.-S. P. Wong, "Analysis of the Frequency Response of
Carbon Nanotube Transistors," IEEE Trans. on Nanotech., vol. 5, pp. 599-605, 2006.
[24] L. C. Castro, D. L. John, D. L. Pulfrey, M. A. P. M. Pourfath, A. A. G. A. Gehring, and H. A.
K. H. Kosina, "Method for predicting f/sub T/ for carbon nanotube FETs," IEEE Trans. on
Nanotechnology, vol. 4, pp. 699-704, 2005.
[25] K. Natori, Y. Kimura, and T. Shimizu, "Characteristics of a carbon nanotube field-effect
transistor analyzed as a ballistic nanowire field-effect transistor," Journal of Applied Physics,
vol. 97, pp. 034306-7, 2005.
[26] S. Ilani, L. A. K. Donev, M. Kindermann, and P. L. McEuen, "Measurement of the quantum
capacitance of interacting electrons in carbon nanotubes," Nature Physics, vol. 2, pp. 687-691,
2006.
[27] D. Akinwande, Y. Nishi, and H.-S. P. Wong, "Analytical Model of Carbon Nanotube
Electrostatics: Density of States, Effective Mass, Carrier Density, and Quantum Capacitance,"
in IEDM Technical Digest: IEEE, 2007, pp. 753-756.
[28] T. W. Crowe, W. L. Bishop, D. W. Porterfield, J. L. Hesler, and R. M. Weikle, "Opening the
terahertz window with integrated diode circuits," IEEE Journal of Solid-State Circuits, vol. 40,
pp. 2104-2110, 2005.
[29] S. A. Maas, Nonlinear Microwave and RF Circuits, 2nd ed.: Artech House, 2003.
[30] C. Kocabas, H.-S. Kim, T. Banks, J. A. Rogers, A. A. Pesetski, J. E. Baumgardner, S. V.
Krishnaswamy, and H. Zhang, "Radio frequency analog electronics based on carbon nanotube
transistors," PNAS, vol. 105, pp. 1405-1409, February 5, 2008 2008.
120
[31] D. A. Williams, "Millimeter-wave components and subsystems built using microstrip
technology," IEEE Trans. on Microwave Theory and Techniques, vol. 39, pp. 768-774, 1991.
[32] M. S. Purewal, B. H. Hong, A. Ravi, B. Chandra, J. Hone, and P. Kim, "Scaling of Resistance
and Electron Mean Free Path of Single-Walled Carbon Nanotubes," Physical Review Letters,
vol. 98, p. 186808, 2007.
[33] S. Datta, Quantum transport : atom to transistor. Cambridge, UK ; New York: Cambridge
University Press, 2005.
[34] Y. Aharonov and D. Bohm, "Time in the Quantum Theory and the Uncertainty Relation for
Time and Energy," Physical Review, vol. 122, pp. Pages: 1649-58, 1961.
[35] U. R. Pfeiffer, C. Mishra, R. M. Rassel, S. Pinkett, and S. K. Reynolds, "Schottky Barrier
Diode Circuits in Silicon for Future Millimeter-Wave and Terahertz Applications," Microwave
Theory and Techniques, IEEE Transactions on, vol. 56, pp. 364-371, 2008.
121
5 Chapter 5
MATERIAL SYNTHESIS AND SURFACE SCIENCE
5.1 Preface
It is self-evident that for any new material to be widely utilized in a broad
variety of applications, it must undoubtedly be relatively easy to manufacture at a
high-quality and with a reasonable idea of its as-made properties. This requirement
applies to carbon nanotubes (CNTs) as well; plus the additionally need for synthesis of
high-density parallel arrays of nanotubes. High-density arrays of identical or similar
nanotubes are essential for high-current and high-performance digital and analog CNT
electronics.
In this light, this work explores the surface science of catalyst dynamics to
elucidate the role that the catalyst treatment, the catalyst-substrate interaction, and the
process conditions play in the synthesis of carbon nanotubes. The obtained insight
paved the way for the synthesis of dense arrays of nanotubes with average and peak
densities of ~5 CNTs/um and ~30 CNTs/um respectively. In addition, the obtained
understanding of the surface science gained from spectroscopy and other experimental
methods was condensed into an illustrative model which explains the surface physics
and surface chemistry of catalyst-surface evolution leading to the growth of CNTs.
All the work was conceived and performed by the author with Nishant Patil
and Albert Lin collaborating on the full-scale wafer growth of dense horizontally
aligned carbon nanotubes shown in Figure 5.7.
122
Surface Science of Catalyst Dynamics for Aligned Carbon
Nanotube Synthesis on a Full-Scale Quartz Wafer
(Reproduced with permission from D. Akinwande, N. Patil, A. Lin, Y. Nishi, and H.-S. P. Wong,
“Surface Science of Catalyst Dynamics for Aligned Carbon Nanotube Synthesis on a Full-Scale Quartz
Wafer”, Journal of Physical Chemistry C, vol. 113, 2009)
Deji Akinwande, Nishant Patil, Albert Lin, Yoshio Nishi, and H.-S. Philip Wong
CENTER FOR INTEGRATED SYSTEMS AND DEPARTMENT OF ELECTRICAL
ENGINEERING
STANFORD UNIVERSITY, STANFORD, CA, 94305, USA
Abstract: The surface chemistry and physics of evaporated metallic iron catalysts on
crystalline quartz substrates is elucidated via detailed microscopy and spectroscopy
for the purpose of synthesizing dense aligned carbon nanotube (CNT) arrays. It was
found that the as-evaporated sub-monolayer iron was predominantly an oxidized
nanoporous film at room temperature in air. Subsequently, at high temperatures
(~865°C), hydrogen gas appears to play an important role in de-wetting of the surface
in a short duration and facilitating discrete iron nanoparticle formation which is
critical for the growth of self-aligned individual single-wall CNTs by catalytic
chemical vapor deposition. Additionally, residual carbon contamination of the surface
in the high vacuum evaporator was identified and mitigated by an improved surface
pretreatment resulting in dense CNT synthesis at a wafer-scale. This work provides
detailed understanding of the physical chemistry of the catalytic surface and
represents a step forward in repeatable synthesis of dense aligned CNT arrays
uniformly at a large wafer scale.
123
5.2 Introduction
The growth of long aligned carbon nanotubes (CNTs) on arbitrary wafer sizes
is highly desirable for large scale integrated sensors and nanoelectronics [1-3]. While
there exists some methods to align CNTs in-situ during growth, the most attractive in
terms of alignment accuracy and resulting CNT density is the catalytic chemical vapor
deposition (CVD) process on crystalline quartz or sapphire substrates [2, 4]. However,
preserving the CNT density on arbitrary wafer sizes with repeatable and uniform
density dispersion has been challenging. Furthermore, it is highly desirable to
significantly increase the state-of-the-art average density of aligned CNTs from
current estimates of 5-10 CNTs/um [2, 5] in order to satisfy high performance
nanoelectronic requirements and minimize device size [6-8]. As a result, it has become
necessary to understand the surface science of catalyst dynamics as a first step towards
optimizing the catalyst and surface pretreatment for dense perfectly aligned CNTs
uniformly dispersed at the wafer scale.
In this paper, a systematic investigation employing both microscopy and
spectroscopy at various stages of the synthesis procedure is undertaken to elucidate the
surface physics and chemistry of evaporated metal catalysts on crystalline quartz
substrates. It is resolved that the as-evaporated sub-monolayer iron exists as an
oxidized nanoporous film at room temperature, with a surface roughness of ~1.7Å
correlating with the 2±0.5Å estimate from the in-situ crystal monitor in the evaporator.
It is important to form isolated catalyst nanoparticles to obtain individual growth of
aligned CNTs [9, 10]. This was previously achieved by a long anneal in air at 900°C
for 1.5hrs [5]. We found that the formation of isolated nanoparticles can be similarly
achieved with an H2 anneal (~865°C) in a much shorter duration (~6mins) at sub-
atmospheric pressures. Therefore, hydrogen appears to play a critical role in
promoting rapid de-wetting of the nanoporous film, resulting in a dispersed array of
discrete catalyst nanoparticles. The time savings significantly increases the throughput
of synthesizing aligned CNTs which is an important consideration for high-volume
applications. Also, high-resolution x-ray photoelectron spectroscopy (XPS) reveals
that the nanoparticles are anchored to the substrate by an interfacial layer of iron
124
silicide which would favor a CNT root growth mode [11]. In-situ temperature
dependent XPS suggests that the as-evaporated oxidized iron transitions to metallic
iron at increasing temperatures consistent with the thermodynamic phase diagram of
oxidized iron [12]. It is concluded that at the growth temperature (using the synthesis
procedure outlined in the ‘experimental methods’ section), the nanoparticle is most
likely predominantly metallic iron which serves as the catalyst for the CVD synthesis
of CNTs.
In addition, unintentional amorphous carbon deposition onto the quartz surface
from an oil-based mechanical pump used during the pumping cycles of the high-
vacuum evaporator was identified and observed to contaminate the catalytic surface
which is detrimental to the yield, uniformity, and very likely the repeatability of
carbon nanotube growth employing iron catalysts for methane CVD. From XPS
studies, the contamination of the surface most likely occurs by the formation of Fe-C
chemical states. An improved surface pretreatment involving extended calcination and
reduction at high temperatures was found useful in desorbing carbon from Fe-C
species and mitigating the effects of amorphous carbon contamination. These results
emphasizes that nanoscopic understanding of the physical chemistry of the catalytic
surface, and knowledge of deposition systems and chambers [13] are similarly
important factors in achieving high density, repeatable and uniformly dispersed self-
aligned CNTs on a full quartz wafer.
5.3 Experimental Methods
A. CNT Synthesis.
Stable temperature (ST) cut single-crystal quartz wafers were annealed in O2 at
900°C for 8hrs in an atmospheric pressure furnace in order to heal the surface of
defects [5]. A sub-monolayer of iron (>99.95% purity) was evaporated at a pressure of
~10-7 Torr at a rate of ~0.2Å/sec for a total thickness of ~2Å as measured by the in-
situ precision crystal monitor in the mechanically pumped planetary electron-beam
evaporator (planetary rotation is ~30 revolutions/minute). The design of the evaporator
125
includes a cold trap to minimize hydrocarbons from back-streaming from the oil-based
pump into the deposition chamber. The carbon nanotube synthesis procedure consists
of two steps: a surface pretreatment step, and a growth step. The pretreatment begins
with a calcination of the sample in synthesized air (20% : 80%, O2 : Ar) for ~30mins at
550°C (will be termed initial calcination) in a CVD hot-wall horizontal furnace.
Afterwards, the furnace is pumped down to a base pressure of ~25mTorr and the
temperature is ramped up to ~865°C in Argon at a flow-rate of 1000sccm, followed by
an H2 anneal for 6mins at a flow-rate of 120sccm and a pressure of ~140Torr. The
growth results are not overly sensitive to the anneal time, for example, we find an H2
anneal of 10mins yields similar results. The CNT growth step commences with the co-
flow of H2/CH4 at a flow-rate of 120sccm/1100sccm and a sub-atmospheric pressure
of ~320-340 Torr for nominally 15mins at ~865°C. Subsequently, the furnace was
typically cooled to room temperature.
The surface properties were investigated at different stages of the surface
pretreatment step to shed light on the behavior of the catalyst film on the quartz
surface. In order to mitigate the detrimental effects of amorphous carbon deposition on
the sample surface during evaporation, an improved surface pretreatment is employed
to achieve the full-wafer results discussed in section 3.3. The improved surface
pretreatment involves an extended calcination to 625°C, followed by H2 anneal as the
temperature ramps up to 865°C.
Slight departures from this prescribed synthesis procedure also results in the
growth of carbon nanotubes as shown in the supporting information. In general,
optimization of the synthesis procedure is required for different surface pretreatment
in order to obtain the best results.
B. CNT Characterization.
Samples were characterized using a FEI XL30 Sirion thermal field emission
scanning electron microscope (SEM) with acceleration energies of ~1-5 KeV, a Veeco
Multimode atomic force microscope (AFM) operated in tapping mode at room
temperature, a PHI 5000 VersaProbe XPS system for measurements at room
126
temperature, and a Surface Science Instruments S-Probe XPS system for in-situ
temperature dependent measurements. Binding energy (BE) assignment and envelope
deconvolutions of the spectra were accomplished using the PHI MultiPak spectral
software. Both XPS chambers were maintained in ultra-high vacuum (UHV base
pressure < 4x10-10 Torr), and charge compensation (electron and ion beams) was
employed in the VersaProbe system to prevent accumulation of charge on the
insulating quartz substrate. Electrical measurements were done in a Cascade attoguard
probe station at room temperature in air.
5.4 Results and Discussion
The single-wall carbon nanotubes obtained with the synthesis procedure on
small quartz samples (<1cm2) with pre-patterned iron stripes are shown in Figure 5.1a-
b. These CNTs obtained using the initial calcination step are very well aligned with
average densities of ~5-10 CNTs/um and local densities as high as 30 CNTs/um in
agreement with previously reported results [5, 14]. The mean nanotube diameter
determined from AFM measurements of a large number of CNTs is ~1.2nm with a
standard deviation of ~0.3nm [15]. Figure 5.1c-d show the SEM image of a device
fabricated for two-port electrical measurements and the measured resistance
respectively, confirming that the CNTs are electrically active. The low-energy
resistance follows the expected Landauer model [16] and is predominantly due to the
electrical contribution of the metallic CNTs (the precise ratio of metallic to
semiconducting CNTs is not known for the specific synthesis procedure). Figure 5.1e
is the SEM image of initial growth results obtained on a 4” quartz substrate with good
alignment but low average densities (~0.1-0.3 CNTs/um).
The difficulty in reproducing the CNT densities obtained on small sample sizes
motivated a detailed systematic investigation of the surface science in order to
understand the catalytic surface and optimize the surface pretreatment for higher
densities on arbitrary wafer sizes.
127
20um100nm
20um
1
10
100
1000
0.01 0.1 1 1020um
e)
a) b)
d) c)
Figure 5.1: Images and device data from initial CNT synthesis.
a) SEM image of aligned single-wall CNTs grown on a small sample with average
densities of 5-10 CNTs/um. b) High-resolution AFM scan showing peak densities of
~30 CNTs/um. c) SEM image of a patterned CNT device for resistance measurements.
d) Resistance vs. length showing a constant-then-linear dependence on length. e)
Initial results of CNTs grown on a full 4” quartz wafer with poor densities of ~0.1-0.3
CNTs/um.
128
5.4.1 Surface Physics
Atomic force microscopy was used to elucidate the physical evolution of the
catalytic surface throughout the surface pretreatment process as shown in Figure 5.2.
The annealed crystalline blank quartz substrate initially has negligible texture as seen
in Figure 5.2a. After sub-monolayer iron evaporation, the surface reveals a disordered
nanoporous film with a root-mean-square (rms) roughness of ~1.7Å (Figure 5.2b). The
surface roughness is in good agreement with the evaporators’ in-situ thickness crystal
monitor measurement of 2±0.5Å, indicating that the surface roughness reflects the
evaporated thickness of sub-monolayer films. Calcination of the quartz sample and
temperature ramp-up in Argon to ~865°C improves the texture revealing a
polycrystalline nanoporous film (pore size ~ 8-14 nm) with an rms roughness of ~2.3Å
(Figure 5.2c).
Subsequent H2 anneal for ~6mins results in de-wetting of the surface leading to
a dispersed array of isolated nanoparticles with heights of ~3-4nm as shown in Figure
5.2d-f. The specific identity of the catalyst nanoparticles (either iron or iron oxide)
will be addressed in §5.4.2. A previously published surface pretreatment procedure for
aligned CNT growth employed a 900°C anneal for 1.5hrs in air to form isolated
nanoparticles [5]. The presence of H2 at low pressures (~140Torr) clearly appears to
rapidly facilitate isolated particle formation most likely by reducing the surface energy
for de-wetting and enhancing particle diffusion in a timescale that is an order of
magnitude shorter compared to the previous work [5]. Based on the AFM
investigation, it is concluded that the physical effect of the surface pretreatment is to
form a dispersed array of catalytic nanoparticles from an initial nanoporous film.
Further efforts are being devoted to more precisely understand the role of low-pressure
H2 in expediting particle formation.
129
Figure 5.2: AFM surface scans of iron catalyst on single-crystal quartz.
a) Blank quartz with negligible surface roughness (after 900°C O2 anneal). b) After
sub-monolayer Fe evaporation (surface roughness of nanoporous film ~1.7Å). c) After
calcination and Argon anneal to ~865°C (roughness ~1.7Å). d) After H2 anneal at
865°C for 6mins showing a dispersed array of isolated nanoparticles. e) A zoom-in
scan for section analysis (black horizontal line) to determine particle height which is
shown to be ~3-4nm in f). The scale bar is 100nm.
130
5.4.2 Surface Chemistry
In order to gain a clear knowledge of the identity of the chemical species
present on the catalytic surface, a detailed systematic XPS investigation was
conducted at different stages of the surface pretreatment procedure. Figure 5.3a shows
the XPS survey scan after Fe evaporation showing the expected peaks corresponding
to iron, silicon, and oxygen peaks from the quartz substrate. Also present on the
surface was carbon which was not intentionally deposited and represents
contamination. The origin of this carbon was not initially known, perhaps coming
from the ambient environment, the evaporation chamber, or the high temperature
anneal furnace. To determine the source of the carbon contamination, a controlled
experiment was performed between two samples. A sample labeled ‘blank’ is an
annealed quartz sample without Fe evaporated, while another labeled ‘controlled’ was
co-processed identically to the former but subsequently placed in the evaporation
chamber. The chamber was pumped down to base pressure, however no Fe was
evaporated. Both samples were housed in the same sample holder at all times and
experienced the same environment except when the controlled sample was placed in
the evaporator.
It is concluded from the XPS comparison (Figure 5.3b) that the excess carbon
originates from the evaporator which uses an oil-based mechanical pump. The process
of unintentional carbon deposition most likely involves residual back-streaming of
hydrocarbons from the pump oil into the evaporation chamber, and subsequently
adsorbing onto the sample surface. This is a stochastic process that is uncontrolled and
generally varies with evaporation.
131
Binding Energy (eV)
Binding Energy (eV)
Intensity (arb. units)
controlled
sample
blank
sample
8
3
Intensity (arb. units)
a)
b)
Figure 5.3: a) XPS survey scan after sub-monolayer Fe evaporation on quartz sample.
The elemental peaks have been assigned as shown. The presence of carbon on the
surface is unintentional. b) XPS investigation between a ‘blank’ sample (no Fe
evaporated) and a ‘controlled’ sample (with Fe evaporated) revealing that the
unintentional carbon originates from the evaporation chamber.
132
The next question regards addressing the role of carbon on the surface, and the
interaction between the Fe catalyst and the substrate. Figure 5.4a is the Si 2p core level
stacked spectra taken after different stages of the pretreatment procedure. At room
temperature, the observed predominant peak is due to SiO2 (the quartz substrate) with
a smaller peak on the right shoulder corresponding to the expected negative shift due
to Fe-Si interactions and in agreement with previous XPS studies of sub-monolayer
interactions between Fe and Si [17, 18]. At higher temperatures the intensity of the Fe-
Si increases and becomes the main observed peak after 865°C pretreatment which is
interpreted as the formation of an interfacial layer [19], stabilizing and anchoring the
nanoporous iron film to the substrate [11]. Physically, the interfacial layer represents
Fe nanoparticles partially embedded in the substrate, a view supported by reported
transmission electron microscopy studies on the structure of Fe particles on oxidized
substrates [20, 21]. Additionally, it is worthwhile to note the absence of a SiC peak
which has a characteristic BE ~100eV [22], indicating that the unintentional carbon is
not reacting with Si in a detectable amount at the relevant temperatures of interest in
this work.
Figure 5.4b is the core level XPS spectrum at the binding energies
characteristic of C 1s. The envelope has been deconvolved into four peaks using a
Gaussian-Lorentzian distribution with a ‘goodness of fit’ parameter Χ2<0.5 indicating
a good fit [23]. The four peaks have been assigned to the most likely carbon species
based on the characteristic binding energy of elemental C (BE~284.5eV) [22] and the
positive (negative) energy shifts associated with the more (less) electronegative carbon
species. The spectrum was taken after 865°C H2 anneal of the sub-monolayer
evaporated Fe on crystalline quartz (the spectrum taken immediately after Fe
evaporation which is not shown was very similar). Noticeable is a strong peak due to
Fe-C (reported BE of ~282.6eV) [24] which has been previously observed to form
from evaporated sub-monolayer carbon on iron at room temperature [25]. The strong
Fe-C peaks suggest that the unwanted amorphous carbon contaminates the surface by
carbidization of iron which is interpreted as detrimental to the efficiency of the metal
catalyst for single-wall CNT growth using methane CVD.
133
Figure 5.4: a) Si 2p XPS core level spectra after sub-monolayer Fe evaporation.
i-25°C, ii-after 550°C calcination, iii-after 865°C Ar anneal, iv-after 865°C H2 anneal.
A negative chemical shift is observed at the high temperatures implying the formation
of an iron silicide interfacial layer. b) C 1s core level spectrum after 865°C H2 anneal.
The spectrum has been deconvolved into four peaks. The prominent Fe-C peak
indicates that the unintentional carbon results in carbidization of Fe. c) XPS spectra
for the Fe 2p1 and 2p3 core levels. The intensity ratio in d) shows a transition from
oxidized Fe to metallic Fe at high temperature. The intensity ratios for Fe2O3 and
metallic Fe have been included as a reference.
134
The spectra in Figure 5.4c address the issue of determining the chemical state
of the nanoparticles. Are the particles predominantly oxidized iron or metallic iron
preceding CNT growth? The sample was placed in the ultra-high vacuum XPS
chamber and high-resolution scans were taken in-situ at three different temperatures to
study the chemical evolution of the catalyst. While the temperature dependent spectra
were taken in UHV conditions, and the actual pretreatment conditions are at sub-
atmospheric pressures, it is expected that the results obtained in UHV generally apply
to the actual conditions particularly at the highest temperature. This is because the
presence of inert argon gas in the actual pretreatment is not likely to play a role in
changing the chemical nature of the catalyst. The XPS intensity ratio (Fe 2p3 peak
normalized to the Fe 2p1 peak) shown in Figure 5.4d is a useful method to distinguish
between metallic iron and its oxidized states [22, 26].
The ratio shows a transition from oxidized iron at room temperature to metallic
iron at higher temperatures consistent with the vanishing solubility of oxygen in solid
iron from the phase diagram of iron-oxygen systems [12]. At the pre-growth
temperature of ~865°C, an extrapolation of the intensity ratio concludes that the
nanoparticles exist mostly as metallic Fe. This is certainly the most plausible chemical
state for Fe during the actual H2 reduction pretreatment step, and the conclusion is
supported by an experimental study of the complete reduction of oxidized iron in a
reducing gas at 850°C [27]. The catalytic nanoparticle is active in its metallic state [28,
29]. However, carbidized iron has a similar intensity ratio as metallic iron, therefore it
is likely the two species are present on the surface just before CNT growth. One specie
is the desired Fe nanoparticle which is an efficient catalyst, and the other is the carbon
contaminated Fe.
It is worthwhile to clarify the role of carbidized iron during CNT growth.
Detailed time-dependent transmission electron microscopy (TEM), x-ray diffraction,
and in-situ XPS studies have shown that iron carbide (predominantly in the Fe3C
phase) forms during the initial phase of CNT growth [11, 28, 30]. However, the
catalytic nanoparticle was metallic Fe preceding the CNT growth step. The premature
formation of carbidized iron during catalyst preparation (before CNT growth begins)
135
is what has been elucidated in this section, and it has been interpreted as degrading
catalytic efficiency. The exact nature of the Fe-C interaction (possibilities include iron
carbide compounds and/or Fe combined with organic groups) during the catalyst
pretreatment could not be determined from XPS because the possible Fe-C
interactions have very similar energy shifts. Additionally, the precise process of how
carbidization during catalyst pretreatment degrades the catalyst is yet unclear and
warrants further studies possibly using environmental TEM and in-situ time-resolved
XPS [28].
To mitigate the effects of carbon contamination, an improved surface
pretreatment (discussed in §5.3) was employed and the resulting spectra shown in
Figure 5.5b. The Fe core level spectra for the initial pretreatment procedure are shown
in Figure 5.5a. The peak at room temperature is assigned to Fe-C species which forms
in the UHV evaporator as the carbon contaminants interact with the evaporated Fe.
After calcination, the peaks due to the more electronegative oxidized iron species
(binding energies typically ~710.4-711 eV) [22, 26, 31] become visible as a result of
the higher oxidation activity at ~550°C. After H2 anneal, the oxidized iron is reduced
to metallic iron. However, the metallic iron re-oxidizes when exposed to air as the
sample is retrieved from the anneal furnace for insertion into the XPS chamber, which
explains the presence of oxidized iron in the XPS spectrum. The improved surface
pretreatment procedure which includes a higher temperature calcination step is more
effective (than the original procedure) in reversing Fe-C interaction (Figure 5.5b). This
occurs by the decomposition of the metastable carbidized iron at the higher calcination
temperature (a previously observed effect) [27] and the increased oxidation activity. A
longer H2 reduction step (~12mins) was subsequently employed to compensate for the
increased oxidation. The improved pretreatment procedure gave consistently higher
CNT yield (~5-10 CNTs/um) at the wafer scale.
136
Norm
alized Intensity
Norm
alized Intensity
Norm
alized Intensity
Figure 5.5: XPS comparison of initial and optimized pretreatment procedures.
a) Fe XPS core level spectra of the initial pretreatment procedure (after sub-monolayer
iron evaporation). i-25°C, ii-after 550°C calcination, iii-ramp up to 865°C and H2
anneal. The insert is a magnification of the Fe 2p3 peak (709-713 eV) showing the
presence of Fe-C species at room temperature (BE~710eV). At higher temperatures
the peaks due to the more electronegative oxidized iron species becomes visible. b)
Comparison between the Fe 2p3 spectrum of the initial pretreatment procedure and the
improved pretreatment procedure. The improved pretreatment procedure has a
decreased presence of Fe-C groups.
137
5.4.3 Illustrative Surface Science Model and Full Wafer Synthesis
A proposed model of the physical chemistry of the surface dynamics during
catalyst pretreatment is illustrated in Figure 5.6 (in the absence of surface
contamination). In brief, the as-evaporated sub-monolayer iron is a disordered
nanoporous film that is readily oxidized when exposed to the ambient environment.
Subsequent calcination and anneal serves to form an interfacial layer of iron silicide
which anchors the film to the substrate. Also, the crystalline structure of the
nanoporous film is improved and a growing fraction of the film is metallic iron.
During high temperature H2 reduction, de-wetting of the film occurs leading to a
dispersed array of isolated catalytic Fe nanoparticles.
Utilizing the improved surface pretreatment in the catalyst preparation (to
mitigate contamination) improved the density and uniformity of well-aligned carbon
nanotubes grown on full quartz wafers as demonstrated in Figure 5.7. The CNT
density and uniformity were reproducible on several quartz wafers processed in series
using the improved synthesis.
138
Figure 5.6: A proposed illustrative model of the catalytic surface physics and
chemistry during the catalyst pretreatment procedure.
At the outset of CNT growth, the catalyst is in the form of metallic iron nanoparticles
anchored to the substrate by an interfacial silicide. The precise stoichiometry of the
chemical states is not known.
139
T
C
F
L R
T
C
F
L R
T
C
F
L R
Top Center
FlatLeft Right
2um
Figure 5.7: SEM images of CNT grown on full 4” quartz wafer.
The growth was achieved using the improved synthesis procedure with significantly
higher density compared to the initial results of Figure 5.1f, and uniform density
dispersion across the wafer. Scale bar is 50um unless otherwise noted.
140
5.5 Conclusion
The surface chemistry and physics of the dynamics of evaporated iron catalysts
during surface pretreatment has been elucidated for the purpose of gaining
understanding of the surface science, and growth of well-aligned carbon nanotubes on
crystalline quartz substrates. It was found that the as-evaporated sub-monolayer iron
was predominantly an oxidized nanoporous film at room temperature in air. A
subsequent surface pretreatment procedure converted the oxidized nanoporous film to
a dispersed array of discrete metallic iron nanoparticles which served as the catalyst
for high density aligned CNT growth on a full-scale wafer. Moreover, contamination
of the surface during sample processing using oil-based vacuum systems was found to
be detrimental for high-density growth of CNTs. For optimum growth results, care has
to be taken to design a surface pretreatment procedure that mitigates contamination or
employ non-oil based vacuum systems.
This work represents a step forward in understanding the physical chemistry of
the catalytic surface, and synthesis of uniformly dispersed dense aligned CNT arrays.
Further work involves determining the maximum density of aligned CNTs achievable
from crystalline quartz substrates, which is a current question of interest in the
material synthesis of carbon nanotubes.
Acknowledgments
This work was supported in part by the Focus Center Research program (FENA and
C2S2), a Stanford graduate fellowship (N.P.), and Ford foundation and Stanford
DARE graduate fellowships (D.A). Extended discussions with Cara Beasley, Tom
Carver, Chuck Hitzman, and Dr. Jim McVittie were fruitful for this work. The CNT
processing and characterization were performed at National Science Foundation
supported facilities (Stanford Nanofabrication Facility and Stanford
Nanocharacterization Laboratory).
141
5.6 References
[1] P. Avouris, J. Appenzeller, R. Martel, and S. J. Wind, "Carbon nanotube electronics," Proc. of
the IEEE, vol. 91, pp. 1772-1784, 2003.
[2] S. J. Kang, C. Kocabas, T. Ozel, M. Shim, N. Pimparkar, M. A. Alam, S. V. Rotkin, and J. A.
Rogers, "High-performance electronics using dense, perfectly aligned arrays of single-walled
carbon nanotubes," Nature Nanotech., vol. 2, pp. 230-236, 2007.
[3] P. Qi, O. Vermesh, M. Grecu, A. Javey, Q. Wang, H. Dai, S. Peng, and K. J. Cho, "Toward
Large Arrays of Multiplex Functionalized Carbon Nanotube Sensors for Highly Sensitive and
Selective Molecular Detection," Nano Letters, vol. 3, pp. 347-351, 2003.
[4] S. Han, X. Liu, and C. Zhou, "Template-Free Directional Growth of Single-Walled Carbon
Nanotubes on a- and r-Plane Sapphire," Journal of Am. Chem. Soc., vol. 127, pp. 5294-5295,
2005.
[5] C. Kocabas, S. J. Kang, T. Ozel, M. Shim, and J. A. Rogers, "Improved Synthesis of Aligned
Arrays of Single-Walled Carbon Nanotubes and Their Implementation in Thin Film Type
Transistors," Journal of physical Chemistry. C, vol. 111, pp. 17879-17886, 2007.
[6] D. Akinwande, G. F. Close, and H.-S. P. Wong, "Analysis of the Frequency Response of
Carbon Nanotube Transistors," IEEE Trans. on Nanotech., vol. 5, pp. 599-605, 2006.
[7] A. Naeemi and J. D. Meindl, "Design and Performance Modeling for Single-Walled Carbon
Nanotubes as Local, Semiglobal, and Global Interconnects in Gigascale Integrated Systems,"
IEEE Transactions on Electron Devices, vol. 54, pp. 26-37, 2007.
[8] L. Wei, J. Deng, and H. S. P. Wong, "1-D and 2-D Devices Performance Comparison
Including Parasitic Gate Capacitance and Screening Effect," in IEEE IEDM Tech. Digest,
2007, pp. 741-744.
[9] C. Kocabas, M. Shim, and J. A. Rogers, "Spatially Selective Guided Growth of High-Coverage
Arrays and Random Networks of Single-Walled Carbon Nanotubes and Their Integration into
Electronic Devices," Journal of Am. Chem. Soc., vol. 128, pp. 4540-4541, 2006.
[10] M. A. Ermakova, D. Y. Ermakov, A. L. Chuvilin, and G. G. Kuvshinov, "Decomposition of
Methane over Iron Catalysts at the Range of Moderate Temperatures: The Influence of
Structure of the Catalytic Systems and the Reaction Conditions on the Yield of Carbon and
Morphology of Carbon Filaments," Journal of Catalysis, vol. 201, pp. 183-197, 2001.
[11] J. F. AuBuchon, C. Daraio, L. H. Chen, A. I. Gapin, and S. Jin, "Iron Silicide Root Formation
in Carbon Nanotubes Grown by Microwave PECVD," Journal of Physical Chemistry. B, vol.
109, pp. 24215-24219, 2005.
142
[12] L. S. Darken and R. W. Gurry, "The System Iron--Oxygen. II. Equilibrium and
Thermodynamics of Liquid Oxide and Other Phases," Journal of American Chemical Society,
vol. 68, pp. 798-816, 1946.
[13] A. Reina, M. Hofmann, D. Zhu, and J. Kong, "Growth Mechanism of Long and Horizontally
Aligned Carbon Nanotubes by Chemical Vapor Deposition," Journal of Physical Chemistry.
C, vol. 111, pp. 7292-7297, 2007.
[14] L. Ding, D. Yuan, and J. Liu, "Growth of High-Density Parallel Arrays of Long Single-Walled
Carbon Nanotubes on Quartz Substrates," Journal of Am. Chem. Soc., vol. 130, pp. 5428-5429,
2008.
[15] N. Patil, A. Lin, E. R. Myers, K. Ryu, A. Badmaev, C. Zhou, H.-S. P. Wong, and S. Mitra,
"Wafer-Scale Growth and Transfer of Aligned Single-Walled Carbon Nanotubes," IEEE
Trans. on Nanotech., June 2008 2009.
[16] M. S. Purewal, B. H. Hong, A. Ravi, B. Chandra, J. Hone, and P. Kim, "Scaling of Resistance
and Electron Mean Free Path of Single-Walled Carbon Nanotubes," Physical Review Letters,
vol. 98, p. 186808, 2007.
[17] K. Rührnschopf, D. Borgmann, and G. Wedler, "Growth of Fe on Si (100) at room temperature
and formation of iron silicide," Thin Solid Films, vol. 280, pp. 171-177, 1996.
[18] B. Egert and G. Panzner, "Bonding state of silicon segregated to alpha -iron surfaces and on
iron silicide surfaces studied by electron spectroscopy," Physical Review B, vol. 29, p. 2091,
1984.
[19] B. A. Orlowski, B. J. Kowalski, K. Fronc, R. Zuberek, S. Mickevicius, F. Mirabella, and J.
Ghijsen, "Study of Fe/Si multilayers by photoemission spectroscopy," Journal of Alloys and
Compounds,, vol. 362, pp. 202-205, 2004.
[20] Y. J. Jung, Y. Homma, T. Ogino, Y. Kobayashi, D. Takagi, B. Wei, R. Vajtai, and P. M.
Ajayan, "High-Density, Large-Area Single-Walled Carbon Nanotube Networks on Nanoscale
Patterned Substrates," Journal of Physical Chemistry. B, vol. 107, pp. 6859-6864, 2003.
[21] G.-H. Jeong, A. Yamazaki, S. Suzuki, Y. Kobayashi, and Y. Homma, "Behavior of catalytic
nanoparticles during chemical vapor deposition for carbon nanotube growth," Chemical
Physics Letters, vol. 422, pp. 83-88, 2006.
[22] J. F. Moulder, W. F. Sticle, P. E. Sobol, and K. D. Bomben, Handbook of x-ray photoelectron
spectroscopy Eden Prairie, Minn.: Physical Electronics, 1995.
[23] K. H. Richter, "Determination of numerical data from polychromatic ESCA spectra and their
verification," Journal of Electron Spectroscopy and Related Phenomena, vol. 60, pp. 127-153,
1992.
[24] G. Panzner and W. Diekmann, "The bonding state of carbon segregated to [alpha]-iron
surfaces and on iron carbide surfaces studied by electron spectroscopy," Surface Science, vol.
160, pp. 253-270, 1985.
143
[25] A. Wiltner and C. Linsmeier, "Formation of endothermic carbides on iron and nickel," Phys.
Stat. Sol., vol. 201, pp. 881-887, 2004.
[26] W. Temesghen and P. Sherwood, "Analytical utility of valence band X-ray photoelectron
spectroscopy of iron and its oxides, with spectral interpretation by cluster and band structure
calculations," Analytical and Bioanalytical Chemistry, vol. 373, pp. 601-608, 2002.
[27] J. Zhang, O. Ostrovski, and K. Suzuki, "Effect of Temperature on Cementite Formation by
Reaction of Iron Ore with H2-CH4-Ar Gas," Metallurgical and Materials Transactions B, vol.
31, pp. 1139-1142, 2000.
[28] S. Hofmann, R. Sharma, C. Ducati, G. Du, C. Mattevi, C. Cepek, M. Cantoro, S. Pisana, A.
Parvez, and F. Cervantes-Sodi, "In situ observations of catalyst dynamics during surface-
bound carbon nanotube nucleation," Nano letters, vol. 7, pp. 602–608, 2007.
[29] A. Khasanov, J. He, J. Gaillard, K. Yang, A. M. Rao, C. M. Cameron, J. M. Schmeltzer, J. G.
Stevens, and A. Nath, "The role of gamma-iron nanoparticulates in the growth of carbon
nanotubes," Applied physics letters, vol. 93, 2008.
[30] H. Yoshida, S. Takeda, T. Uchiyama, H. Kohno, and Y. Homma, "Atomic-Scale In-situ
Observation of Carbon Nanotube Growth from Solid State Iron Carbide Nanoparticles," Nano
letters, vol. 8, pp. 2082-2086, 2008.
[31] T. Fujii, F. M. F. de Groot, G. A. Sawatzky, F. C. Voogt, T. Hibma, and K. Okada, "In situ
XPS analysis of various iron oxide films grown by NO_ 2-assisted molecular-beam
epitaxy," Physical Review B, vol. 59, pp. 3195-3202, 1999.
144
6 Chapter 6
SILICON-CNT NANOTECHNOLOGY
6.1 Preface
It has gradually become clear in the course of the past five years that
integrating carbon nanotubes with conventional very large scale integrated (VLSI)
technology such as silicon (Si) is the least disruptive and most attractive technology
entry option for nanotubes in contrast to the alternative which would be an all CNT
stand-alone nanotechnology. Integrating nanotubes with silicon VLSI leverages all the
benefits of a mature well-understood technology with the added advantages that CNTs
offer in terms of high-performance electronics. Furthermore, nanotubes can provide
added functionality such as molecular sensing which when integrated on a Si VLSI
chip for peripheral electronics enables practical sensor products.
For this purpose, monolithic integration of carbon nanotubes with commercial
silicon VLSI technology was demonstrated for the first time. A post-silicon fabrication
process was developed to place nanotubes at defined locations on a VLSI chip and
subsequently interconnected to silicon integrated circuits. The vehicle circuit
employed for the demonstration was a cascode amplifier consisting of a silicon
transistor connected to a carbon nanotube transistor showing operation in the MHz
range.
All the work was conceived and performed by the author in collaboration with
Shinichi Yasuda, Bipul Paul, Gael Close, Shinobu Fujita, and H.-S. Philip Wong. The
author designed the CMOS circuit, developed the process for integrating nanotubes
with Si-CMOS and did all the fabrication and experimental work.
145
Monolithic Integration of CMOS VLSI and CNT for Hybrid
Nanotechnology Applications
(Reproduced with permission from D. Akinwande, S. Yasuda, B. Paul, S. Fujita, G. Close, and H.-S. P.
Wong, “Monolithic Integration of CMOS VLSI and CNT for Hybrid Nanotechnology Applications”,
IEEE Trans. Nanotech., vol. 7, 2008)
Deji Akinwande1, Shinichi Yasuda2, Bipul Paul2, Shinobu Fujita2, Gael Close1, and
H.-S. Philip Wong1
1CENTER FOR INTEGRATED SYSTEMS AND DEPARTMENT OF
ELECTRICAL ENGINEERING, STANFORD UNIVERSITY, STANFORD, CA,
94305, USA 2ADVANCED LSI TECHNOLOGY LABORATORY, TOSHIBA CORPORATION,
KAWASAKI, JAPAN
Abstract: We integrate carbon nanotube (CNT) fabrication with standard commercial
CMOS VLSI fabrication on a single substrate suitable for emerging hybrid
nanotechnology applications. This co-integration combines the inherent advantages of
CMOS and CNTs. These emerging applications include CNT optical, biological,
chemical, and gas sensors that require complex CMOS electronics for sensor control,
calibration, and signal processing. We demonstrate the successful co-integration on a
single chip with a vehicle circuit; a two transistor cascode megahertz amplifier
utilizing both silicon nMOS and CNT transistors with a total power consumption of
62.5µW.
146
6.2 Introduction
Single wall carbon nanotubes are continually been explored for a wide variety
of applications including electronics, optics, and sensing. The electronic and optical
applications exploit the metallic and semiconducting properties of CNTs for use as
interconnects, transistors, and optical devices [1-4]. Sensing applications take
advantage of the easy access to all the carbon atoms at the surface of the nanotube, and
the ability to functionalize the surface to interact with specific bio-chemical species
[5-7]. Despite the promising uses of CNTs and recent progress including the synthesis
of perfectly aligned dense nanotubes [8], and the use of chemical sorting techniques to
achieve monodispersed chirality [9, 10], the fabrication of CNTs (with controlled
chirality and predictable performance) at a very large-scale integration (VLSI) with
the necessary portfolio of devices (transistors, diodes, resistors, capacitors, etc) on a
single substrate has not been achieved. In the meantime, significant progress in CNT
applications can be made by augmenting CNT devices with standard VLSI circuits to
form a post CMOS-CNT monolithic integrated circuit (CCMIC) process. The CMOS
circuits can provide regulated power supplies, load capacitors and other peripheral
devices for high performance CNT field effect transistors (CNFETs).
Likewise, CMOS VLSI can supply advanced signal processing electronics and
analog to digital conversion of CNT sensor output for robust and optimum sensor
performance [11]. Additionally, CNT nano-electromechanical devices are been
explored as switches in integrated CMOS memory and logic circuits to minimize
power dissipation in nanoscale VLSI [12]. The variety of existing and proposed
applications of co-integration suggest that the application space employing monolithic
integration of CNT and CMOS is likely to continue expanding as CNT synthesis
becomes more mature.
There has been prior success integrating MOSFETs and CNT devices on a
single chip. Tseng et al. reported an nMOS decoder circuit useful for device
characterization of CNT arrays grown by catalytic chemical vapor deposition (CCVD)
[13]. Likewise, Lin et al. fabricated an inverter by integrating an nMOS and a p-type
CNFET with a shared polysilicon gate [14]. In both works, the nMOS and the CNFET
147
were fabricated in-situ in a custom laboratory process that did not include
complementary transistors and other VLSI devices. Additionally, monolithic high
temperature CCVD synthesis is not compatible with standard CMOS processing
which requires precise control of thermal cycles, impurity profiles, and utilizes low
temperature (<400°C) metallic interconnects.
In this work, we report a CCMIC post-CMOS process that integrates CNTs
using dielectrophoresis (DEP) [15] on top of a standard commercial 0.25µm VLSI
technology. The fundamental advantage of this technique is that the VLSI fabrication
and CNT processing are independent, that is, the CMOS can be fabricated in a
scaleable VLSI foundry with the advantages of cost, flexibility and predictable
performance. Single-wall or multi-wall carbon nanotube suspensions are dispersed on
the topmost layer of a planarized VLSI chip or wafer at room temperature (details of
the DEP dispersion technique are provided in [16]). In brief, a dichlorobenzene
suspension of single-wall CNTs obtained from Helix materials containing an unknown
mixture of semiconducting and metallic CNTs (diameter~1.3nm, purity~90%) are
dispersed on a wafer using a 500kHz, 8V continuous wave. The yield of aligning
CNTs between lithographically defined source and drain pads is 91% based on
electrical continuity measurements of three hundred devices fabricated in a laboratory
cleanroom for yield characterization purposes.
Further electrical measurements on the aligned CNTs reveal that 40% are
purely semiconducting, with ~43% containing both semiconducting and metallic
nanotubes, and some devices containing purely metallic CNTs (~17%) as shown in
Figure 6.1. It is possible to refine the nanotube suspensions by using density
differentiation [9] or aromatic polymer [10] solution based techniques to obtain purely
semiconducting CNTs. Though relatively nascent and the subject of ongoing
laboratory research, these solution based techniques are very promising and make
dielectrophoresis quite attractive for future hybrid monolithic integration of CMOS
and CNTs. It is conceivable in the future that a custom cleanroom-compatible DEP
equipment can be made to automate placement of a moderate amount of
monodispersed nanotubes on a VLSI wafer.
148
0
5
10
15
20
25
30
sCNTs sCNTs+mCNTs mCNTs
Co
un
ts
Figure 6.1: The distribution of aligning semiconducting nanotubes (sCNTs), and
metallic nanotubes (mCNTs) to lithographic defined source and drain pads.
The yield of aligning to back-gated CNFETs using the dielectrophoresis procedure is:
purely sCNTs ~40%, purely mCNTs ~17%, and devices containing both mCNTs and
sCNTs are ~43%. The sCNTs were determined from the ON/OFF current ratio which
was typically > 103. The mCNTs show no gate control, and devices containing both
have a current ratio <102.
149
6.3 Hybrid Monolithic Chip
The CCMIC process is demonstrated with an integrated two transistor
megahertz amplifier employing both a silicon nMOS and a CNT transistor. The chip
fabrication consisted of two parts, the first part being the standard VLSI fabrication
and the second is the CNT integration. The VLSI fabrication was done in a standard
tsmc 0.25µm silicon foundry with a portfolio consisting of complementary transistors,
diodes, passive devices, and three metal layers. The second metal of the three metal
stack was used as a backgate for CNT transistors. The intermetal dielectric (IMD2 in
Figure 6.2a which is a 1um SiO2 served as the backgate oxide. Since the CNFET
current scales weakly with gate oxide thickness due to the inverse logarithmic
dependence between gate capacitance and oxide thickness [2, 17], considerable gate
control and current is possible even for a thick oxide. Using the existing IMD as gate
oxide is a reasonable compromise between simplicity of integration and performance.
The passivation layer above the topmost metal had large openings in order to
gain access to the underlying CMOS devices for subsequent connection to nanotubes.
The wafer was diced afterwards into 5x5 mm2 chips for post processing to integrate
CNTs. Handling and processing small chips is very difficult, as a result, the 5x5 mm2
chips were mounted on top of a blank 4” carrier wafer with epoxy for automated full-
wafer processing.
150
Figure 6.2: a) Cross-section of a silicon nMOS with an integrated CNFET.
The foundry process is a three layer process including an interlayer dielectric (ILD)
and two inter-metal dielectrics (IMD). Access to the nMOS transistors and underlying
VLSI devices are provided by vias to the topmost metal. The CNT integration consists
of a three mask post-processing involving alignment of CNT arrays using DEP, source
and drain palladium (Pd) evaporation, and etching to make isolated CNT transistors
within the array. Metal2 in the foundry process is used as a backgate (BG) for the
CNFETs. b) Photograph of the completed CCMIC chip showing three columns where
arrays of CNFETs are located and used as cascode transistor amplifiers. c) A zoomed
view of the dashed region in b) showing an nMOS connected to three fingers of CNTs.
d) A scanning electron microscope image of a CNT aligned to gold DEP electrodes.
151
The CNT integration (shown in Figure 6.2) is a three mask optical lithography
(OPL) fabrication process. The first mask is for the deposition of metallic pads used
for DEP and also to define the fifteen active area or sites where CNTs are
subsequently aligned simultaneously by shared DEP pads. The second mask defines
the palladium source and drain electrodes for good contact to the CNTs and to prevent
the nanotubes from moving during further processing, as well as defining the
interconnects between CNT devices and the underlying CMOS devices. The final
mask is for selective etching of the shared metallic pads used for DEP in order to
make isolated CNT transistors. Each isolated CNT transistor typically has ~1-3
nanotubes bridging the gap between the source and drain with a channel length of
~2um. Having multiple nanotubes that are all semiconducting is beneficial for
increased current and transconductance (gm) since the total gm is the sum of the gm of
each individual tube in the CNFET.
However, the presence of a metallic nanotube in the CNFET is detrimental for
the DC and AC performance. The metallic nanotube draws current without providing
any transconductance leading to an inefficient transistor with higher DC power
dissipation. At AC frequencies, the metallic nanotubes act to shunt the small signal
output resistance of the CNFET, thereby reducing the voltage gain of the combined
cascode transistor. For optimum performance, it is imperative that metallic nanotubes
be avoided or removed [18], which is still a matter of ongoing research. Additionally,
palladium was used in this work for near ohmic contact to the CNTs [19]. A poor
contact leads to high Schottky barriers at the source and drain, reducing the injection
velocity, current and transconductance, and hence must be mitigated for optimum
circuit performance.
152
-2.0E-06
-1.6E-06
-1.2E-06
-8.0E-07
-4.0E-07
0.0E+00
-1.5 -1 -0.5 0
VDS (V)
I DS (
uA
)
VGS=-1.5V
VGS=-1V
a) 0
-0.4
-0.8
-1.2
-1.6
-2.0
1.E-10
1.E-09
1.E-08
1.E-07
1.E-06
1.E-05
-1.5 -1 -0.5 0
VGS (V)
-ID
S (
A)
VDS=-1.5V
b)
S~110mV/dec
10-5
10-6
10-7
10-8
10-9
10-10
0.0E+00
1.0E-04
2.0E-04
3.0E-04
4.0E-04
5.0E-04
6.0E-04
7.0E-04
0 0.5 1 1.5 2 2.5
VDS (V)
I DS (
uA
)
700
600
500
400
300
200
100
0
VGS=1.25V
VGS=1V
VGS=0.75V
0
1
2
3
4
Figure 6.3: DC electrical characteristics of CNT and nMOS transistors.
a) Drain current (IDS) vs. drain voltage (VDS) of a p-type CNFET as the gate voltage
(VGS) steps in -0.5V increments from 0V to -1.5V. b) IDS vs. VGS of the same CNFET
with VDS=-1.5V. ON/OFF ratio is ~104, S~110mV/dec, and extracted peak gm~1uA/V.
c) nMOS IDS vs. VDS characteristics with VGS in steps of 0.125V from 0.5V to 1.25V.
The nMOS channel length is 0.25um with a 4um width. d) Statistics for the ON/OFF
current ratio for the NMOS-CNT cascode transistors evaluated at a common bias
(VDD=2.5V).
c) d)
a) b)
153
In this letter, we discuss the nMOS-CNT cascode transistor operating as an
amplifier which is used as a vehicle to demonstrate the hybrid monolithic integrated
technology. Figure 6.3 shows the DC electrical data of a CNT and a typical nMOS
transistor with overall good characteristics. The CNT had a drain current ON/OFF
ratio of ~104, and a subthreshold slope (S) of about 110mV/decade typical of back-
gated CNFETs. The small signal peak transconductance (gm) was ~1uA/V with an
output resistance (rout) of 5MΩ at VD=-1.5V. Figure 6.3d shows the statistics for the
ON/OFF current ratio for the limited number of working NMOS-CNT cascode
transistors. A more comprehensive set of statistics was not feasible due to the small
number of devices on the space-limited CMOS chip.
The CCMIC chip was subsequently packaged and mounted on a printed circuit
board (PCB) where on-board input and output bias-tees routed DC supply voltages
and AC signals to the nMOS-CNT cascode amplifier. The cascode amplifier operated
as a 12dB band-pass amplifier at ~2.2 MHz. Figure 6.4 shows the circuit schematic
and AC amplifier gain. The bandwidth is limited by the load capacitance of the
external test circuit, and the AC performance of the CMOS-CNT cascode is less than
that of an all CMOS cascode because of the higher CMOS transconductance. However,
the purpose of integrating CNT with CMOS is to take advantage of the unique benefits
of carbon nanotubes. For example, the CMOS-CNT cascode amplifier could
potentially be used as a low-power sensor circuit, where the AC gain and center
frequency vary in response to chemical or gas induced changes in the resistance of the
exposed CNT.
154
-6
-2
2
6
10
14
1.7 1.9 2.1 2.3 2.5
Frequency (MHz)
Ga
in (
dB
)
Figure 6.4: AC characteristics of nMOS-CNT integrated cascode amplifier.
a) The oscilloscope trace showing the input/output waveforms. The signals are at
2.2MHz with the output ~12 dB greater than the input. b) The gain vs. frequency of
the cascode amplifier using the circuit elements in c). The bandpass response is due to
the inductor capacitor external load network at the output of the test circuit. c)
Schematic of the printed circuit board (PCB) test circuit and the simplified input
source and oscilloscope load networks; all denoted by dashed boxes. The input source
operated with VS=10mV and RS=50Ω. The scope provided CL=12pf and RL=1MΩ
load. The PCB components were Ci=Co=1uF, Li=1mH, and Ld=150uH. The amplifier
consumed 62.5µW from 2.5V Vdd, and a drain current of 25µA.
c)
b) a)
155
6.4 Conclusion
The hybrid technology of integrating CMOS with CNTs is a very new area of
applied research, and this work represents the first monolithic co-integration of single
wall CNTs on a standard commercial CMOS process. However, much more sustained
research is required to evaluate the feasibility of a hybrid monolithic process for a
wide variety of existing and future applications. In this work, we have demonstrated a
CCMIC process by integrating carbon nanotubes on top of a commercial VLSI
substrate using the dielectrophoretic effect. An integrated nMOS-CNT cascode
amplifier worked successfully as a vehicle circuit at ~2.2MHz with a gain of 12dB.
The technique is scaleable down to the latest nanoscale VLSI technology.
Moreover, the hybrid integration is also scaleable to integrate multi-wall CNTs,
nanowires, and other one dimensional tubular structures [16]. Further work involves
improving the yield of semiconducting CNTs and exploring novel applications that
can take advantage of a monolithic CMOS-CNT co-integration.
Acknowledgments
The CNT post-processing and fabrication were completed at the Stanford
Nanofabrication Facility, an NSF supported facility. This work was supported in part
by the Toshiba Corporation, and the Focus Center Research Program (FENA and
C2S2). G.C was additionally supported by an Intel Graduate Fellowship, and D.A was
additionally supported by a Ford Foundation and Alfred P. Sloan graduate fellowships.
156
6.5 References
[1] P. Avouris, Z. Chen, and V. Perebeinos, "Carbon-based electronics," Nature Nanotech., vol. 2,
pp. 605-615, 2007.
[2] A. Naeemi, R. Sarvari, and J. D. Meindl, "Performance comparison between carbon nanotube
and copper interconnects for gigascale integration (GSI)," IEEE Electron Device Letters, vol.
26, pp. 84-86, 2005.
[3] D. Akinwande, G. F. Close, and H.-S. P. Wong, "Analysis of the Frequency Response of
Carbon Nanotube Transistors," IEEE Trans. on Nanotech., vol. 5, pp. 599-605, 2006.
[4] C. Kocabas, H.-S. Kim, T. Banks, J. A. Rogers, A. A. Pesetski, J. E. Baumgardner, S. V.
Krishnaswamy, and H. Zhang, "Radio frequency analog electronics based on carbon nanotube
transistors," PNAS, vol. 105, pp. 1405-1409, February 5, 2008 2008.
[5] E. S. Snow, F. K. Perkins, E. J. Houser, S. C. Badescu, and T. L. Reinecke, "Chemical
Detection with a Single-Walled Carbon Nanotube Capacitor," Science, vol. 307, pp. 1942-
1945, March 25 2005.
[6] P. Qi, O. Vermesh, M. Grecu, A. Javey, Q. Wang, H. Dai, S. Peng, and K. J. Cho, "Toward
Large Arrays of Multiplex Functionalized Carbon Nanotube Sensors for Highly Sensitive and
Selective Molecular Detection," Nano Letters, vol. 3, pp. 347-351, 2003.
[7] J. Li, Y. Lu, Q. Ye, M. Cinke, J. Han, and M. Meyyappan, "Carbon Nanotube Sensors for Gas
and Organic Vapor Detection," Nano Letters, vol. 3, pp. 929-933, 2003.
[8] S. J. Kang, C. Kocabas, T. Ozel, M. Shim, N. Pimparkar, M. A. Alam, S. V. Rotkin, and J. A.
Rogers, "High-performance electronics using dense, perfectly aligned arrays of single-walled
carbon nanotubes," Nature Nanotech., vol. 2, pp. 230-236, 2007.
[9] M. S. Arnold, A. A. Green, J. F. Hulvat, S. I. Stupp, and M. C. Hersam, "Sorting carbon
nanotubes by electronic structure using density differentiation," Nature Nanotech., vol. 1, pp.
60-65, 2006.
[10] A. Nish, J.-Y. Hwang, J. Doig, and R. J. Nicholas, "Highly selective dispersion of single-
walled carbon nanotubes using aromatic polymers," Nature Nanotech., vol. 2, pp. 640-646,
2007.
[11] T. S. Cho, K.-J. Lee, J. Kong, and A. Chandrakasan, "Low Power Carbon Nanotube Chemical
Sensor System," in IEEE Custom Integrated Circuits Conf., 2007.
[12] R. S. Chakraborty, S. Narasimhan, and S. Bhunia, "Hybridization of CMOS With CNT-Based
Nano-Electromechanical Switch for Low Leakage and Robust Circuit Design," IEEE
157
Transactions on Circuits and Systems I: Fundamental Theory and Applications, vol. 54, pp.
2480-2488, 2007.
[13] Y. C. Tseng, P. Xuan, A. Javey, R. Malloy, Q. Wang, J. Bokor, and H. Dai, "Monolithic
Integration of Carbon Nanotube Devices with Silicon MOS Technology," Nano Letters, vol. 4,
pp. 123-127, 2004.
[14] H. Lin, Y. W. Park, and S. Tiwari, "A Compact Single-Walled Carbon Nanotube Transistor
Integrated with a Silicon MOSFET Using a Single Common Gate," in Mater. Res. Symp.
Proc., 2007, pp. 963, 0963-Q14-04.
[15] C. Chen, D. Xu, E. S. kong, and Y. Zhang, "Multichannel Carbon-Nanotube FETs and
Complementary Logic Gates With Nanowelded Contacts," IEEE Electron Device Letters, vol.
27, pp. 852-855, 2006.
[16] G. Close and H.-S. P. Wong, "Fabrication and Characterization of Carbon Nanotube
Interconnects," in IEEE IEDM Tech. Dig., 2007, pp. 203-206.
[17] S. Heinze, M. Radosavljevic, J. Tersoff, and P. Avouris, "Unexpected scaling of the
performance of carbon nanotube Schottky-barrier transistors," Physical Review B, vol. 68, p.
235418, 2003.
[18] J. Deng, N. Patil, K. Ryu, A. Badmaev, C. Zhou, S. Mitra, and H.-S. P. Wong, "Carbon
Nanotube Transistor Circuits: Circuit-Level Performance Benchmarking and Design Options
for Living with Imperfections," in IEEE ISSCC Tech. Dig., 2007, pp. 70-588.
[19] A. Javey, J. Guo, Q. Wang, M. Lundstrom, and H. J. Dai, "Ballistic carbon nanotube field-
effect transistors," Nature, vol. 424, pp. 654-7, 2003.
158
7 Closing Remarks
7.1 Research Contributions
The focus of my dissertation has been on the science, technology, and
applications of carbon nanotubes. I conceived, defined and performed a series of
original studies that crossed the boundaries from material synthesis, device physics,
circuits, and nanotechnology. The broad range of original studies have led to several
significant results including:
i- Analytical development of the physics of CNTs providing insight and basic
understanding of the electro-physical properties.
ii- Experimentally verified analytical current-voltage and capacitance-voltage
characteristics of CNT transistors facilitating rapid circuit design and analysis
for the first time.
iii- Elucidation of the material science of nanotube synthesis which enabled the
first demonstration of high-quality manufacturing of perfectly-aligned CNT
arrays for VLSI integrated circuits.
iv- Monolithic integration of CNTs and CMOS with demonstrated CMOS-CNT
analog & digital circuits for future nanoelectronics.
159
7.2 List of Publications
The journal and conference publications resulting from my doctoral dissertation are
listed below.
Journal Publications
1. D. Akinwande, N. Patil, A. Lin, Y. Nishi, and H.-S. P. Wong, "Surface
Science of Catalyst Dynamics for Aligned Carbon Nanotube Synthesis on a
Full-Scale Quartz Wafer," Journal of Physical Chemistry. C, vol. 113, pp.
8002-8008, 2009.
2. D. Akinwande, J. Liang, S. Chong, Y. Nishi, and H.-S. P. Wong, "Analytical
Ballistic Theory of Carbon Nanotube Transistors: Experimental Validation,
New Device Physics, Parameter Extraction, and Performance Projection,"
Journal of Applied Physics, vol. 104, pp. 124514, 2008.
3. D. Akinwande, Y. Nishi, and H.-S. P. Wong, "Carbon Nanotube Quantum
Capacitance for Non-Linear Terahertz Circuits," IEEE Trans. on
Nanotechnology, vol. 8, pp. 31-36, 2009.
4. D. Akinwande, S. Yasuda, B. Paul, S. Fujita, G. Close, and H.-S. P. Wong,
"Monolithic Integration of CMOS VLSI and Carbon Nanotubes for Hybrid
Nanotechnology Applications," IEEE Trans. on Nanotechnology, vol. 7, pp.
636-639, 2008.
5. J. Liang, D. Akinwande, and H.-S. P. Wong, "Carrier Density and Quantum
Capacitance for Semiconduting Carbon Nanotubes," Journal of Applied
Physics, vol. 104, pp. 064515, 2008.
6. D. Akinwande, Y. Nishi, and H.-S. P. Wong, "An Analytical Derivation of the
Density of States, Effective Mass and Carrier Density of Achiral Carbon
Nanotubes," IEEE Trans. on Electron Devices, vol. 55, pp. 289-297, 2008.
7. D. Akinwande and H.-S. P. Wong, "A Composite Circuit Model for NDR
Devices in Random Access Memory Cells," IEEE Transactions on Electron
Devices, vol. 54, pp. 776-783, 2007.
160
8. D. Akinwande, G. F. Close, and H.-S. P. Wong, "Analysis of the Frequency
Response of Carbon Nanotube Transistors," IEEE Trans. on Nanotechnology,
vol. 5, pp. 599-605, 2006.
Conference Publications
1. X. Chen, K.-J. Lee, D. Akinwande, G. F. Close, S. Yasuda, B. Paul, S. Fujita,
J. Kong, and H.-S. P. Wong, "High-Speed Graphene Interconnects
Monolithically Integrated with CMOS Ring Oscillators Operating at 1.3GHz,"
in IEDM Technical Digest: IEEE, 2009, paper 23.6.
2. D. Akinwande, J. Liang, and H.-S. P. Wong, "The Device Physics of
Experimentally Validated Analytical Theory of Transport in Ballistic Carbon
Nanotube Transistors," in American Physical Society APS Meeting, abstract
#A36.001, 2009.
3. X. Chen, D. Akinwande, and H.-S. P. Wong, "Contact Resistance, Electrical
Breakdown and Temperature-Dependent Conductance of Multi-Wall Carbon
Nanotubes," in American Physical Society APS Meeting, abstract #C1.152,
2009.
4. D. Akinwande, S. Yasuda, B. Paul, S. Fujita, G. Close, and H.-S. P. Wong,
"Monolithic Integration of CMOS VLSI and CNT for Hybrid Nanotechnology
Applications," in European Solid State Device Research Conference, 2008. pp.
91-94.
5. D. Akinwande, Ada Poon, Teresa Meng, H.-S. P. Wong, “Adventures with
Carbon Nanotubes: 3D Inductors, Capacitors & Antennas, and THz ICs",
DARPA/Army Workshop, ARL, Maryland, August 2008.
6. D. Akinwande, J. Liang, and H.-S. P. Wong, "Analytical Degenerate Carrier
Density and Quantum Capacitance for Semiconducting Carbon Nanotubes," in
Device Research Conference Digest: IEEE, 2008, pp. 117-118.
161
7. S. Yasuda, D. Akinwande, G. F. Close, H. S. P. Wong, B. C. Paul, and S.
Fujita, "Fault-Tolerant Circuit for Carbon Nanotube Transistors with Si-CMOS
Hybrid Circuitry," in IEEE Conf. on Nanotechnology, 2008. pp. 684-687.
8. D. Akinwande, N. Patil, A. Lin, Y. Nishi, and H.-S. P. Wong, "Full Scale
Wafer Growth of Perfectly Aligned Carbon Nanotubes and Aligned Growth
Characterization," in Mater. Res. Symp. Spring Meeting, San Francisco, CA,
2008.
9. D. Akinwande, Y. Nishi, and H.-S. P. Wong, "Analytical Model of Carbon
Nanotube Electrostatics: Density of States, Effective Mass, Carrier Density,
and Quantum Capacitance," in IEDM Technical Digest: IEEE, 2007, pp. 753-
756.
162
7.3 Future Work
We do not attempt to predict the future here, only to highlight some of the
most significant challenges facing the routine employment of carbon nanotubes in
practical applications. It is notable that much progress has been made in advancing the
understanding and applications of carbon nanotubes over the past 15yrs and
particularly in the past 10yrs. Nonetheless, there are still significant obstacles that
must be overcome for carbon nanotubes to be a mainstream technology for devices,
circuits, sensors, and future novel applications. The major challenge at the moment is
the controlled growth of high-quality carbon nanotubes of specific properties. For
instance, specific chirality growth of high-quality nanotubes will immediately
revolutionize the field and accelerate the deployment and commercialization of carbon
nanotechnology. Additionally, improvements in the density of aligned nanotubes to a
value of ~100 CNTs/um (20x greater than current average densities of 5 CNTs/um)
will bring nanotube transistors much closer to reality in future VLSI technology.
An area of CNT research that has yet to lift-off is the experimental
demonstration of high-performance nanotube circuits operating at GHz frequencies
and beyond. Important circuits for communication systems such as oscillators,
amplifiers, and mixers using ballistic carbon nanotubes is expected to bring to light the
potential advantages nanotubes offer compared to silicon-based circuits. Advances in
CNT circuits will undoubtedly benefit a range of applications of growing importance
including sensor electronics, flexible electronics, and bioelectronics.
Even if all these advancements are realized, it is still most likely that carbon
nanotube and silicon technologies will have to co-exist in an agreeable manner to
leverage the synergies that both technologies can offer. As a result, the continued
development of a hybrid nanotechnology employing conventional (mature) silicon
electronics combined with the high-functionality of nanotube devices promises to be a
very rewarding path for the next revolution involving wafer-scale intelligent systems
for healthcare, energy, communications and automation.
Top Related