Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
The adsorption and selective deposition of molecular and nanocluster ions on carbon based devices
(USC Thesis Other)
The adsorption and selective deposition of molecular and nanocluster ions on carbon based devices
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
The Adsorption and Selective Deposition of Molecular and
Nanocluster Ions on Carbon Based Devices
by
Patrick Joseph Edwards
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
PHYSICS
August 2021
Copyright 2021 Patrick Joseph Edwards
ii
Dedication
To my wife Jennifer
and
my parents Patti and Jeffrey
iii
Acknowledgements
I am eternally thankful to Professor Vitaly Kresin for being a truly great mentor to me over my
time at USC. He has taught me so much about science, relationships, decision making, philosophy,
and life in general. Vitaly is an incredible physicist, but more than that he is an incredible leader,
who handles the task of training young scientists masterfully. When needed, he was always
available for guidance. Over the last year, I have really missed our regular afternoon chats,
discussing what going on around the lab. These were never planned like a group meeting, I just
enjoyed swinging by his office at the end of the day and he always made time for me. It always
amazes me how trusting Vitaly is with his students. He gives us so much liberty to choose project
ideas that truly interest us and encourages new and creative solutions to the problems we face
together. Not to mention the freedom he gives to operate and tinker with these large, complex, and
(frankly) expensive machines. For me, this trust was incredibly motivating and empowering as a
young researcher. He thought I could handle it which made me start to believe that I could. Vitaly
was patient with me during this entire journey, quick to aid me when I was in a bind, and always
understanding of time taken for what he always said are more important things (health, time with
family, etc.). Projects did not always pan out as we intended, but the focus was always placed on
doing work the right way, adjusting to what we were seeing, and not forcing results. His sense of
calm during the whole experience was inspiring. The lessons I have learned for him over the years
will last a lifetime. I could not have had a better advisor.
Special thanks to Dr. Adam Bushmaker, for awarding me the opportunity to work at the
Aerospace Corporation and entrusting me with such a rich and interesting project. Thank you for
always being an advocate for me while at Aerospace. He is always considering my long-term
iv
development and looking out for opportunities for me to learn new things. His guidance the last 3
years has been invaluable and contributed greatly to much of the work written here.
I would also like to thank the remaining members of my committee, Prof. Moh El-Naggar, Prof.
Jahan Dawlaty, and Prof. Eli Levenson-Falk. I have spent many combined hours barging into their
offices with random questions and thoughts. I was never turned away. They are all so very
generous with their time and expertise, both of which are extremely valuable commodities. I am
further appreciative to Prof El-Naggar for allowing me to use his lab space so freely, particularly
the atomic force microscope and probe-station. Without these resources, many of my results would
not have been possible.
I am also grateful to have been able to work closely with many fellow students of the Kresin lab
throughout the years! I learned valuable lessons from the more senior students as I got acclimated
at USC. For their guidance, I thank Dr. Avik Halder, Dr. Daniel Merthe, Dr. Malak Kohjasteh, Dr.
Chuanfu Huang, and Dr. Lorenz Kranabetter (I suppose Lorenz isn’t technically ours, but we
adopted him). I also had the great gift of sharing many hours in the lab with my excellent
contemporaries, John Niman and Benjamin Kamerin. They are supremely skilled and handle their
work with dedication and professionalism. Working alongside them has been a highlight of my
graduate school experience. Finally, I was given the opportunity to work with an amazingly
talented group of undergraduate students! I appreciate very much Will Liang, Diego Hernandez,
Atef Sheekhoon, and Elizabeth Zhou. They were all so willing and excited to lend a hand, even
for jobs that are not glamourous. I very much look forward to hearing all about their graduate
school journeys in the future.
Starting early in my time at USC I was lucky to work closely with students of the El-Naggar lab,
from whom I learned so much. For that, I am grateful for Dr. Sahand Pirbadian, Marko Chavez,
v
and Christina Cole. Without their help, my long hours at the AFM would have been significantly
more confusing and much less productive. Similarly, I am thankful for both James Farmer and
Darian Hartsell of the Levenson-Falk lab, for many great carpool conversations, advice on device
fabrication, and SEM imaging help (James).
One of the lesser talked about, but highly valuable resources at USC is our vibrant and diverse
graduate student community. There are so many wonderful people I have gotten to meet during
this process, that I do not have the space to name them all here. Whether they struggled through
the 1
st
year with me, helped with GASP events, attended visiting weekends, lent a hand with
research, or simply commiserated over our shared experience of learning to become better
scientists, they have my thanks. I have gained lifelong friends here and I simply could not have
finished this without them.
I would like to recognize two groups who contributed to both my work personally and the
Physics research department as a whole. First, thank you to Don Wiggins and all the members of
the USC machine shop for their prodigious skill and fantastic advice on my equipment designs.
Secondly, I very much appreciate the administrative group for the physics department, namely
Christina Tasulis Williams, Lisa Moeller, Betty Byers, and Kimberly Burger. Thank you for all
the hard work keeping the department moving, and for assisting me during my time at USC.
I am very thankful to my family and close friends, for their many years of support and
encouragement. Thank you for putting up with my disappearing for weeks/months at a time while
I either studied for the next big scary exam or focused on lab work. I am very much looking forward
to seeing you all with greater regularity very soon!
I am immensely grateful of my loving wife Jennifer Edwards. She has worked incredibly hard,
forgoing her studies, and working full time to allow me the freedom and opportunity to study for
vi
the better part of a decade now. She has always unwaveringly trusted in our long-term plan, and
tirelessly supported me while I was working through the problem of the moment. Thank you so
much for enabling me to pursue my passion to its fullest. This accomplishment is as much hers as
it is mine.
I am also appreciative of my mother Patti Klein, whom I was supremely fortunate to have had
as my first teacher. Times may have changed, but the early 90’s were not nearly as accepting a
time to the idea of “home schooling” and it was a big choice to pull her kids out of conventional
school. While it might not be for everyone, she provided the best environment for me. She always
attentively tailored my instruction in such a way as to foster a love for learning and asking
questions. I can’t thank her enough for doing so, and for her guidance throughout my life.
This work is dedicated to the memory of my father figure Jeffrey Klein, who we lost suddenly
early this year. Jeffrey was one of the people most looking forward to this with me. He invested
tremendous amounts of his time and energy into my development during my formative years. He
was not only an excellent stepdad, but a really great friend. He went out of his way to show me
that he believed in me and foster self-efficacy whenever he could. He taught me that the greatest
gift you can give anyone is an opportunity, and he gave me many. He saw great value in giving
someone the chance they needed to earn something for themselves, and he relished the idea that
he could help instill confidence in others. I will forever treasure our time together. Thank you,
Fwee.
vii
Table of Contents
Dedication ...................................................................................................................................... ii
Acknowledgements ...................................................................................................................... iii
List of Figures ............................................................................................................................... xi
Abstract ..................................................................................................................................... xviii
1 Carbon based materials and devices .................................................................................... 1
1.1 Structure and electrical properties of Graphene and CNTs .............................................. 2
1.2 Fabrication methods and basic electrical properties of Graphene and CNT devices ....... 8
Growth of SWCNT’s ................................................................................................ 8
Mechanical exfoliation and CVD growth of graphene ............................................. 9
Device lithography techniques ................................................................................ 10
Basic device properties and electrical response ...................................................... 14
2 Atomic nanoclusters and applications to nanoscale devices ............................................ 19
2.1 Shell structure in small metallic clusters ........................................................................ 19
2.2 Finite-size properties of clusters .................................................................................... 20
2.3 Integration of metallic clusters into Carbon based devices ............................................ 25
3 Experimental set-ups for various works ............................................................................ 30
3.1 Device fabrication .......................................................................................................... 30
Suspended SWCNT’s via CVD on pre-patterned trenches .................................... 30
E-beam patterning and contacting of CVD Graphene ............................................ 33
viii
3.2 Cluster generation and deposition .................................................................................. 37
Machine overview ................................................................................................... 37
DC magnetron gas aggregation source ................................................................... 38
Size selection with source parameter adjustment and quadrupole filtering ............ 40
Substrate mounting and electrical connection ........................................................ 44
3.3 Atomic Force Microscope (AFM).................................................................................. 48
3.4 Substrate preparation ...................................................................................................... 52
SiO2 cleaning methods ........................................................................................... 52
HV heat treatment of graphene ............................................................................... 57
3.5 Gas Flow Chamber for ionized gas source..................................................................... 63
4 Clusters deposition on carbon based devices .................................................................... 67
4.1 Selective deposition of ionic Ag nanoclusters on graphene ........................................... 67
Introduction ............................................................................................................. 67
Initial imaging of Silver on “as fabricated” graphene device ................................. 68
Graphene vs. SiO2 at various cluster coverages and beam conditions ................... 71
SEM imaging of silver cluster coated devices ........................................................ 74
Graphene Islands and Mica comparison ................................................................. 78
High coverage limit on graphene devices: Potential mechanism and future
supporting simulations ........................................................................................................... 82
Discussion and future work .................................................................................... 83
ix
4.2 Clusters on CNT devices ................................................................................................ 88
Introduction ............................................................................................................. 88
The importance of CNT suspension ....................................................................... 89
Imaging considerations when planning experiments .............................................. 91
Decorating suspended nanotubes with Ag clusters ................................................. 92
Discussion and future work .................................................................................... 95
5 Experiments involving gaseous ions on semi-conducting SWCNTs ............................. 101
5.1 Water Assisted Desorption and Solvation of Ions on SWCNTs .................................. 101
Abstract ................................................................................................................. 102
Detectable ion adsorption on SWCNT’s............................................................... 103
Surface adsorbed water layers in on Carbon surfaces .......................................... 106
Experimental procedures for device desiccation and results ................................ 107
Monte Carlo simulation of overlapping event correction factor ........................... 110
Expected behavior of surface adsorbed dopants obscured by water desiccation .. 113
Evidence to surface/water proton hopping in clustered adsorption events ........... 116
Determination of the binding adsorption energy of N2+ on CNT’s Surfaces ...... 121
Conclusion and Acknowledgements ..................................................................... 122
5.2 Water-Assisted Ionic Defect Desorption Dynamics on Isolated Carbon Nanotubes ... 125
Introduction ........................................................................................................... 125
Sweeping and gas exposure procedure and results ............................................... 126
x
Voltage preference of ionic events: gate-controlled ion repulsion ....................... 128
Finite element analysis, fermi energy landscape analysis, and high |𝑽𝒈 | ............ 131
Evidence for surface migration in anomalous recoveries ..................................... 135
A comparison of event variety as a function of CNT desiccation ........................ 137
The effect of ∆𝝁 on intermediate 𝑽𝒈 curves ........................................................ 140
Summary ............................................................................................................... 144
Future work ........................................................................................................... 145
References .................................................................................................................................. 147
xi
List of Figures
Figure 1 – Carbon Allotropes: Cartoon displaying four structural forms of carbon. (Clockwise
from top left) Graphene, a 2D sheet of carbon atoms with a honeycomb lattice. Graphite stacked
layers of graphene weakly bound by out-of-plane bonding. The Buckminsterfullerene, 60 carbon
atoms forming a cage-like spherical structure. Lastly, the carbon nanotube (single walled here), a
cylindrical formation of a graphene sheet. ...................................................................................... 1
Figure 2 – Graphene lattice structure: (left) Hybridized orbitals of the carbon atom in
graphene. (middle) Graphene honeycomb lattice with Bravais basis. (right) Reciprocal lattice
space with a shaded first Brillouin zone
5
. ....................................................................................... 3
Figure 3 – Graphene dispersion relation: (left) Two- and (right) three-dimensional dispersion
relation for graphene. (right, zoom) Low energy regime showing a linear dispersion. ................. 4
Figure 4 – Deriving CNT properties from Graphene solutions: (left) Carbon nanotube lattice
vectors superimposed on the graphene lattice, depicting the chiral and translational vectors used
to define the rolling of the graphene sheet. (right) 1D line-cuts of the graphene Brillouin zone
due to quantization in the circumferential direction of the CNT. ................................................... 6
Figure 5 – CNT band structures: (a) (10,4) and (b) (10,5) nanotube band structure exhibiting
the metallic and semiconducting nature of CNT’s respectively. .................................................... 7
Figure 6 – Overview of lithography techniques: (a) a film of resist is applied to the substrate
surface. The resist is then exposed (b) in a patterned manner where metal deposition is desired.
(c) Exposure weakens the resist material in those areas, making them soluble in a developer
solution. (d) After development, the substrate surface is exposed in the patterned areas. (e) Metal
is then deposited over the whole substrate. (f) Stripping of the remaining resist film removes
unwanted metal deposition. .......................................................................................................... 12
Figure 7 – Graphene FET: (left) Cross-section of a graphene-FET
31
. (right) a microscope
image of a graphene Hall bar with characteristic back-gate dependence curve
8
. ......................... 15
Figure 8 – CNT FET: (left) Suspended CNT’s FET design
36
. (right) Back gate characteristics of
a metallic and semiconducting CNT FET
33
. ................................................................................. 16
Figure 9 – Summary of metallic nanoclusters: (left) Close-packed small atomic nanoclusters
exhibiting geometric shell structure. (middle) Sodium abundance spectra collected by the Knight
group at UC Berkeley. (right) A self-consistent Jellium model for electron energy levels of a
Na40 nanocluster, resulting in shell structure.
61
............................................................................ 19
Figure 10 – Plasmonic clusters: Theoretically investigated plasmonic nanocluster array in
which light entering from a source excites a local surface plasmon that propagates through the
cluster array before exiting out the drain contact
71
. ...................................................................... 21
Figure 11 – Cluster arrays for tunneling junctions and SETs: (left) Schematic illustration of
various nanocluster based arrays for use in novel devices/materials with tunable properties
63
.
(right) Cartoons of recent successful works that have achieved nanocluster based SETs
80,81
...... 22
Figure 12 – Gas phase SC Al clusters: Experimental design used for the detection of a SC
phase transition in small Aluminum clusters. Clusters were generated with a DC magnetron gas
aggregation source. Before being ionized by a tunable laser, cluster travel through a thermalizing
set to a fixed temperature by a helium cold head. Ionized clusters are separated and measured via
time-of-flight spectroscopy to determine the photoionization yield as a function of cluster size. 23
Figure 13 – Photoionization spectroscopy results: (left) Evolution of a bulge like feature
arising in closed shell Al66 clusters. (middle, right) Differential yield peak increasing with
decreasing temperature suggesting an electronic phase transition near 100 K. ............................ 24
xii
Figure 14 – Clusters on nanotubes: (left) Atomic deposition on suspended CNTs showing poor
wettability of the deposited metal, a good sign for supporting isolated clusters
97
. (right) Clusters
deposited on HBN nanotubes that display tunneling current when biased via STM probes
99
. .... 28
Figure 15 – SC nano-islands on graphene: Experimental realization of Tin/Graphene
proximity superconducting devices by both Han et al.
106
(left) and Kessler et al.
103
(right). ....... 29
Figure 16 – Micro-trench design leading to suspended CNT-FETs: A series of Microscope
and SEM images depicting the micro-trench device design with CNT suspended between
catalytic particles. ......................................................................................................................... 31
Figure 17 – Defining The graphene channel via EBL and argon etching: Schematic
representation of the patterning process used to define the graphene channel. PMMA is first spun
onto the entire wafer surface (a) before being patterned with the channel in the EBL machine (b).
After development only the exposed rectangle remains (c). Argon plasma sputtering is then
performed, removing all excess graphene and leaving only the area protected by the rectangle
after PMMA stripping (d). ............................................................................................................ 34
Figure 18 – Patterning and deposition of surface contacts: Schematic representation of the
patterning process used to define the graphene surface contacts. Again, PMMA is spun onto the
entire wafer surface. After alignment with the graphene channel (outlined for visibility), the
surface contacts are patterned with EBL machine (a). Post development, the contact shapes are
removed from the PMMA down to the substrate surface (b). Metal is then deposited over the
entire wafer filling the recessed areas (c). Stripping the PMMA off the wafer surface leaves
behind the contact metal that mad good contact to the channel and substrate surface (d). .......... 35
Figure 19 – Completed device imaging: Optical microscopy imaging of a completed graphene
device showing an isolated graphene channel on an SiO2 substrate with Ti/Au surface contacts.
....................................................................................................................................................... 36
Figure 20 – Illustration of cluster deposition system: Overview of the entire deposition system
displaying the path of the cluster beam at each point of generation through the source,
quadrupole and deposition chambers. Vacuum pump connections at the bottom of the source and
deposition chambers are not shown. ............................................................................................. 38
Figure 21 – DC magnetron source head: (left) Bare magnetron source head showing the
copper covered magnetic stage. (right) Assembled magnetron head after Ag metal target
sputtering showing the distinctive circular trench etched out of the target surface. ..................... 39
Figure 22 – Quadrupole design and application to beam characterization: (right) Schematic
of a quadrupole mass filter. (left) Relative intensity spectrum of clusters read by detection grid
at the end of the quadrupole rods by sweeping through stably transmitted cluster sizes. ............ 42
Figure 23 – Example of beam dynamics with shifting gas parameters: The evolution of a Mg
nanocluster beam profile as a function of He flow input with fixed Ar flow (150 sccm) collected
by Malak Khojasteh during initial system characterization experiments. .................................... 43
Figure 24 – Sample mounting system: (a) Overview of a simple mount capable of holding
several samples. (b) Beam collimator used to regulate the beam diameter and expose a single
device at a time. (c) Sample mount on the linear stage showing the connection of the beam
current wire. .................................................................................................................................. 46
Figure 25 – Sample mounting system for individual biasing: Overview of mount for
electrically isolated samples. Each sample sits on a wired stainless-steel puck that is separated
from the L-bracket by ceramic standoffs. Each puck has its own collimation opening. .............. 47
Figure 26 – Non-contact AFM: Schematic representation of an atomic force microscope in
non-contact mode showing details for both frequency modulated (FM) and amplitude modulated
xiii
(AM, dashed lines) operation
130
. Here the tip is driven into a set oscillation over the sample
surface and feedback loop is utilized to correct for variation in the oscillation amplitude (AM)
induced by tip interactions with a varying surface profile. ........................................................... 49
Figure 27 – Contact mode AFM: Schematic representation of an atomic force microscope in
contact mode. The tip is brought into direct contact with the sample surface until interaction
forces cause a deflection in the tip cantilever. A feedback loop is utilized to correct for deviations
of the tip deflection with respect to a set-point value while moving across the sample
128
........... 51
Figure 28 – AFM results of dirty SiO2 surface: Representative AFM topography scans of SiO2
surfaces prior to cleaning procedure. Surface contaminants are seen randomly distributed over
the surface and vary in size from 5-50 nm in height. .................................................................... 54
Figure 29 – AFM results of clean SiO2 surface: Representative AFM topography scans of SiO2
surfaces after to sonication cleaning procedure. (left) A characteristic result showing near total
removal of surface contaminates. The small white dots on the surface are sub 1 nm in height,
which could be attributed to peaks in the surface roughness of the substrate surface. (right)
“Worst case” cleaning result which still shows a significant reduction in surface contaminants
size, now ~2 nm for the largest spots. ........................................................................................... 56
Figure 30 – Heating flange: Heating unit used to HV bake graphene samples. Samples are
mounted to a copper heating chuck which is heated by two Watlow cartridge heaters inserted
into the chuck body. Thermocouple probes at either end of the check measure the surface
temperature. (Photo credit: John Niman) ...................................................................................... 58
Figure 31 – PMMA residue on graphene device channel: AFM topography imaging of the
graphene channel at the step edge (SiO2 region to the left) prior it HV bake. Devices suffered
from a pronounced PMMA ridge along the channel edge and significant PMMA spotting
throughout. .................................................................................................................................... 60
Figure 32 – Graphene channel after HV bake: Post bake AFM imaging showing a reduction
in the ridge along the graphene edge and the opening of large areas in the central channel region.
....................................................................................................................................................... 61
Figure 33 – HV baking effect on CVD graphene wafers: Representative AFM topography
images before (top) and after (bottom) HV baking resulting in a reduction and redistribution of
the surface contamination. ............................................................................................................ 62
Figure 34 – Schematic of chamber for CNT gaseous ion interaction experiments: (left)
Computer generated model of the sealed gas chamber with device socket, automated TEC for
temperature dependent measurements, and
90
Sr source mounting electrode. (right) A Photograph
of wired, surface mounted device with accompanying RTD. ....................................................... 64
Figure 35 – Ag clusters on graphene device with surface contamination: AFM imaging of
dilute Ag cluster deposition along the graphene device edge. A large amount of PMMA
decontamination is observed on the sample surface along with an unexpected cluster coverage
difference. ..................................................................................................................................... 70
Figure 36 – Mixed sample cluster coverage results: AFM imaging comparison of 8 nm Ag
cluster deposition on graphene (top) and SiO2 (bottom) at fixed scan width depicting the many-
fold increase in cluster deposition on the graphene samples. ....................................................... 73
Figure 37 – SEM imaging of SiO2 wafer: Exploratory SEM characterization of the SiO2
surface after Ag cluster deposition. Images confirm the reduction in cluster coverage is not an
artifact of AFM scanning. A film-like contamination is observed surrounding all deposited
clusters. ......................................................................................................................................... 75
xiv
Figure 38 – SEM imaging of graphene wafer: Exploratory SEM characterization of the
graphene surface after Ag cluster deposition. Ag clusters are observed on the graphene surface as
small white dots and a complete absence of deposition is noted along the exposed SiO2
underlayer. .................................................................................................................................... 76
Figure 39 – SEM imaging of graphene device: A series of SEM images of the graphene device
which received ~10 nm cluster deposition. In agreement with the SiO2 imaging we see the
clusters joined by film-like contamination. The distribution of cluster deposition on the oxide
area of the device also appears to correspond to tight areas where PMMA would be more
difficult to remove......................................................................................................................... 77
Figure 40 – Increased Ag coverage on graphene: Graphene regions continue to exhibit greatly
increased and even deposition and higher coverages. There is no evidence of aggregation of
clusters along the graphene step edge, a hallmark of surface diffusion. ....................................... 79
Figure 41 – Electrically isolated cluster doped graphene: Clusters can be seen on completely
electrically isolated pieces of graphene. This suggest that a surface charging electrostatic
mechanism is not the dominant factor in the selective deposition................................................ 80
Figure 42 – Graphene/Cluster archipelago: A large scale example of the above figure. Here a
large patch of the graphene sheet has been degraded during transfer and formed a scattered
collection of small graphene islands which still support clusters selectively. .............................. 81
Figure 43 – Ag decorated Mica: Insulating Mica samples co-mounted with CVD-G show
similar cluster coverages ruling out electrostatically dominated mechanisms. ............................ 82
Figure 44 – Heavy cluster deposition: Approximately 6 ML deposition of Ag clusters onto
graphene devices as a function of post fabrication cleaning treatment. The deposition displays an
extremely high level of surface selectivity of the cluster dopants on devices subjected to HV
baking. ........................................................................................................................................... 83
Figure 45 – Amorphous SiO2 surface: MD simulation from Zandkarimi et. al.
148
displaying the
surface landscape of amorphous SiO2 which is decorated with a multitude of ~1-1.5 nm pits. ... 85
Figure 46 – Dilute Cluster coverage on surface supported CNTs: Before and after AFM
images showing the successful size selected deposition of Al nanoclusters on CNT's ................ 90
Figure 47 – High cluster coverage on surface supported CNTs: An example of heavy
coverage resolution difficulties. It is possible, but very time consuming to resolve the CNT after
deposition. ..................................................................................................................................... 91
Figure 48 – 3nm cluster deposition on suspended CNT: Suspended CNT subjected to 3nm
cluster deposition. Clusters appear to have coalesced on the CNT surface, coming together to
form larger irregular particles ~15-20 nm in diameter. ................................................................ 93
Figure 49 – 5nm cluster deposition on suspended CNT: Suspended CNT subjected to 5 nm
cluster deposition. Here we find the clusters (~7.5-10 nm) are closer to the intended sizes. This
result suggests a lower surface mobility of larger incoming clusters on the CNT surface. .......... 94
Figure 50 – Dai Group atomic deposition on CNTs: A comparison of atomic metal deposition
for various materials onto suspended CNTs without (with) and with (right) a 1 nm thick Ti
pretreatment layer. The addition of Ti layer limits aggregation along the CNT surface, which
could prove vital for the deposition of smaller nanoclusters. ....................................................... 96
Figure 51 –PPMS system and breakout of sample chamber: Schematic overview of the
PPMS which allows us to mount a wired sample onto a puck that fits into a specially designed
helium cryostat capable of cooling to below 2 K. ........................................................................ 99
Figure 52 – PPMS Sample puck: Prototypes for device sample pucks to be used in low-
temperature transport measurements .......................................................................................... 100
xv
Figure 53 – Sample puck socket testing: Niobium test wire soldered into the sample puck and
measured down to 2 K showing a SC transition at expected Tc, confirming good thermal
transport through the socket. ....................................................................................................... 100
Figure 54 – Heading and TOC figure for publication results in ACS nano Nov. 2020 ...... 102
Figure 55 – Ionic interaction with CNT device conduction: (a) Cartoon representation of
ionic surface dopant on a CNT device channel with accompanying plot of the Coulomb potential
barrier in the valence band along the length of the CNT. (b) Repeated gate voltage dependent
transport samples of a device depicting both nominal p-type behavior (without the adsorbed ion
species) and large reduced state sweeps (with the adsorbed ion species). (c) Exemplary data,
collected at 𝑉𝑔 =0, depicting three large switching events in device conduction. The black box
outlines the zoomed plot of the first switching event (d) which shows sharp transitions between
the two conduction states of the device. The reduced CNT conduction state is a result of ion
adsorption at the CNT surface and thus its duration corresponds with the ion residence time on
the surface. .................................................................................................................................. 105
Figure 56 – Evolution of the mean ion residence time (𝝉𝒊𝒐𝒏 ) over the course of the multiple
experimental protocols: (a) 𝜏𝑖𝑜𝑛 versus temperature, with arrows showing the direction of
temperature sweeps for the initial heating phase (orange), the cooling phase (blue), and
secondary heating phase (purple). Data taken below 51° C after the initial heating phase is
permanently locked into the low conduction state (red) due to overlapping events. The plotted
points in these cases represent the total experiment run time at these temperature set-points. (b)
𝜏𝑖𝑜𝑛 at 25° C plotted versus experiment stage, including data taken after a 3 day recovery period
in atmospheric conditions. (c) 𝜏𝑖𝑜𝑛 evolution in three hour intervals over the course of ~60
continuous hours of dry gas desiccation. .................................................................................... 109
Figure 57 – Exemplary data before and after vacuum desiccation: Time dependent
conduction data taken before (a) and after (b) vacuum desiccation showing a large increase in the
reduced conduction state, corresponding to increased ion/surface residence. ............................ 110
Figure 58– Mean residence time of adsorbed species on CNT surfaces: Plots of 𝐸𝑞 2 with
respect to temperature (a) and adsorption energy (b) where 𝜏 0 = 100 𝑐𝑚 − 1, the radial
breathing mode of the CNT. Both plots display the very large range of expected mean adsorption
times that are highly sensitive to surface adsorption energy. These range from nanoseconds for
weakly physiosorbed particles (0.2 eV), to many days for chemisorbed particles (1.0 eV). ..... 116
Figure 59 – Rapid re-adsorption events due to ion hopping between CNT surface and
lowest lying water layer: (a) Exemplary data depicting clustered switching events due to
frequent reabsorption events near the CNT surface. Data can be broken into three distinct stages
depicted in (b). Stage 1: the device is functioning nominally in a high conduction state with
either no ion presents in the system, or the ion totally screened by large amounts of surface
adsorbed water. Stage 2: an N2
+
has landed on the device, been reduced by the surface adsorbed
water, and the resulting H3O
+
or H
+
has adsorbed on the CNT surface. This shunts the device
conduction, leaving it in a low conduction state. Stage 3: frequent switching events caused by
rapid H
+
hopping between the CNT surface and lower lying water layers resulting in the partial
recovery of the device. ................................................................................................................ 118
Figure 60 – Comparison of event frequency with and without adsorbed water: Clustered
switching events histograms depicting the time between adsorption events (TBE) on a
logarithmic scale for events with (a) and without (b) adsorbed water, removed by vacuum
desiccation. The presence of the unexpected peak at lower time bins is evidence of rapid ion
desorption and re-adsorption between the CNT surface and adsorbed water layers that is not
xvi
present after vacuum desiccation. When fit by 𝑟𝑖𝑜𝑛 , as calculated from the data mean and
median TBE we find poor agreement in the pre-desiccated device due to a large shift towards
shorted times caused by rapid ion re-adsorption. ........................................................................ 120
Figure 61 – Adsorption energy determination: Fit of heated trial after vacuum desiccation
resulting in an adsorption energy between 756-781 meV for N2
+
ions on the CNT surface. In
each fit the value of 𝜏 0 was selected to correspond with upper and lower bounds of the CNT
radial breathing mode. As a result, the adsorption energy of the ion is the only free parameter.
..................................................................................................................................................... 122
Figure 62 – Summary of ionic interaction events before and after vacuum desiccation:
Forward (a) and backward (b) gate voltage sweeps that detected an ionic interaction at the CNT
device surface. Forward (c) and Backward (d) events after vacuum desiccation plotted with 10%
opacity for ease of viewing the high volume of events. ............................................................. 128
Figure 63 – Histograms of the adsorption and desorption event location during back gate
voltage sweeps along with time series events captured at various 𝑽𝒈 : Histograms showing
the abundance of adsorption (a) and desorption (b) events as a function of back gate voltage
displaying the clear preference of certain back gate voltages for various events. ...................... 130
Figure 64 – Summary of time series data with back-gate variation: Event count rate (top)
and time series data (bottom) events captured at +.2,0 and -.2 V back-gate bias displaying a
strong effect on ion adsorption efficiency. ................................................................................. 131
Figure 65 – Negative Fermi Energy at select 𝑽𝒈 : Negative fermi energy landscape of the CNT
conduction channel at 𝑉𝑔 ranging from (-4,4) Volts. Curves transition in concavity from up to
down as the bake-gate sweeps from positive to negative bias, which corresponds to a transition in
ion surface binding favorability. ................................................................................................. 133
Figure 66 – Summary of channel conductance simulations as a function of adsorber ion
location: a) Plotting of simulated device conductance at three ion adsorption sites along the
channel length (red) and once in the absence of a surface ion (black). The blue trace illustrates
how an ion trajectory like those represented from the data in b) is explained by the ion travelling
down the conduction channel toward the source/drain contacts. c) a corresponding illustrative
cartoon of ion motion at each stage of the reduced state curves. ................................................ 136
Figure 67 – Event variety as a function of CNT hydration state: Backward sweeping events
captured before (a) and after (b) device desiccation highlighting the three categorical regions in
the pre-desiccation device. Nominal curves (red), central channel adsorbed ions (cyan) and
maximally reduced conduction due to adsorption near the contacts (black). ............................. 138
Figure 68 – Survey of ion adsorption location on conduction: Conduction curves as various
ion adsorption sites along the CNT channel highlighting the sensitivity of the CNT to ions near
the contact pads. Note that the curves for 0.5 and 0.45 channel length lie on top of one another.
Significant reduction in maximum on-state current isn’t observed until ~10% channel length.
(Thesis note: This is a preliminary plot the illustrate the effect which will made more presentable
prior to any attempt at publication) ............................................................................................. 139
Figure 69 – Modified fermi energy landscape with ∆𝝁 = 𝟎 , 𝑬𝒈 = 𝟎 . 𝟔 𝒆𝑽 : Negative fermi
energy curves showing the shift in the pinned values at the CNT channel bounds. The curve at
zero back-gate bias here has become flat, but even in this extreme, does not transition to concave
down. ........................................................................................................................................... 142
Figure 70 – Modified fermi energy landscape with ∆𝝁 = −𝟎 . 𝟓𝟓 , 𝑬𝒈 = 𝟏 . 𝟐 𝒆𝑽 : Negative
fermi energy curves for large bandgap devices. The increase in the bandgap results in a vertical
stretching of the curves while shifting the concavity transition only slightly compared to figure 8.
xvii
The large increase in the minimum value of the zero-bias curve could account for occasional
functioning devices that do not exhibit ionic interaction events................................................. 143
xviii
Abstract
The work presented here summarizes my efforts over the last 6 years to incorporate size-selected
metal nanoclusters into pre-existing carbon-based devices. This work aims to benefit the fields of
both cluster physics and device physics which are well positioned to assist one another by joining
well known techniques from both disciplines. From the cluster perspective, graphene and carbon
nanotube field effect transistors serve as excellent candidate substrates for supporting cluster
deposition. These devices are well characterized and highly electrically tunable via electrostatic
gating. Additionally, the two-dimensional carbon lattice surface is relatively inert and does not
disturb the shape of deposited clusters appreciably. These traits combined to provide an ideal
system for probing the properties of small metallic clusters that are well known in gas phase but
untested on supporting surfaces. For device physics, clusters present themselves as highly flexible
surface dopants for tailoring custom device behavior. They are extremely versatile in that they can
be made of a wide verity of different materials, and screened very precisely by size, charge, and
energy with well-known molecular beam techniques. To that end, this thesis contains my attempts
to better understand surface interactions of carbon-based device with surface dopants of both
cluster and molecular ions.
In Chapter 1, I cover some of the historical background and theoretical underpinnings for the
two-dimensional carbon materials. I introduce method of growing and isolating graphene and
carbon nanotubes as well as some basics in device fabrication and expected electrical
characteristics.
Chapter 2 in presents a short discussion of the discovery and electron shell structure in small
metallic nanoclusters. This is followed by a survey of potential cluster-based applications that
highlight the usefulness of originating clusters onto various surfaces and devices-like
xix
nanostructures. Finally, there is a section considering the direct integration of clusters onto
graphene and carbon nanotube field effect transistors which covers the current state of the field
and the encouraging results already published towards this goal.
In Chapter 3, I break down the specific experimental techniques used for the results obtained in
this work. This includes the fabrication of all devices used, details on the cluster beam deposition
system, an overview of the apparatus used in gaseous ion experiments, and techniques used for
both sample preparations and characterization.
Chapter 4 contains two cluster experiments. The first displays the ability of nanoclusters to
selectively coat a graphene surface, without the need for additional lithographic steps. This is
followed by preliminary successful results on our groups capability to coat suspended CNT-FET
with size selected nanoclusters.
Chapter 5 contains two experiments investigating the dynamics of gaseous ions interacting with
the conduction channel of suspended CNT-FET. I was very lucky to be able to take part in this
project and build an ongoing collaboration with Adam Bushmaker at the Aerospace Corporation.
Our investigations yielded interesting results pertaining to the interaction strength of molecular
ions at device surfaces, as well as single defect effects on device characteristics.
The projects contained in the thesis was generously supported by the National Science
Foundation (NSF), the Army Research Office (ARO), the Aerospace Corporation’s Independent
Research and Development Program, and the Aerospace Corporation University Partnership
Program.
Finally, a note to future graduate students:
It’s impossible to cover everything I’ve learned the last six years in just one document. I could
probably write for another year and still not get everything I wanted on the page. Sadly, time is
xx
finite. Thus, several topics that provided great learning experiences, but didn’t produce any direct
results, hit the cutting room floor over the last couple months. I mention this in the off chance
that a future student reads this abstract before rushing off to find the section they are desperately
hoping answers their question (oh boy, I’ve been there). It’s so easy, when reading a paper, or
thesis, to see the end results cleanly laid out and convince yourself that it all followed a straight
path like it’s organized on the page. This can be discouraging when your project is going
sideways and you have no idea why, let alone how to fix it. The long-winded point I am trying to
make here is that the path to finishing a PhD can be a bit rocky and winding. Progress on a
project does not monotonically increase. Also, grinding long hours of work in the lab doesn’t
always translate immediately into results. I know it’s hard to be patient, but remember this thing
is a marathon and not a sprint. Work hard and be kind to yourself. Your smart, and you’ll figure
it out. The result sections below make up more than half this thesis, but the data I collected to
produce them was gathered in only a few excellent experiment days. You can never know which
experiments will fall on a good day; you just have to try to be ready for them when they come
along.
1
1 Carbon based materials and devices
Figure 1 – Carbon Allotropes: Cartoon displaying four structural forms of carbon. (Clockwise from top left)
Graphene, a 2D sheet of carbon atoms with a honeycomb lattice. Graphite stacked layers of graphene weakly bound
by out-of-plane bonding. The Buckminsterfullerene, 60 carbon atoms forming a cage-like spherical structure. Lastly,
the carbon nanotube (single walled here), a cylindrical formation of a graphene sheet.
Since their respective physical discoveries in 2004 and 1991, graphene
1
and carbon nanotubes
2,3
have garnered the interest of scientists and engineers for their unique mechanical and electrical
properties. Below is a short discussion of the basic physics involved in these carbon allotropes
that will serve to introduce their electrical properties as it pertains to our device applications.
Various summaries of this type of discussion can be found in many references. I have done my
2
best to cite my favorites throughout, but you will find most to be fairly similar. I have not re-
invented the wheel here, but instead covered what I have found from the more concise and direct
treatments. The goal for this section is to simply introduce the properties of these wonderful carbon
materials with an eye on their usefulness as electronic devices. We will begin with a short
discussion of the lattice structure of graphene, from which the electronic dispersion relation and
band structure of both arises. From here we will use the fact that carbon nanotubes (CNTs) are a
cylindrical representation of a graphene sheet, about some chiral angle, to cheat a little and
determine the CNT band structure with little added effort. Following this is a very brief overview
of the historical growth methods of these materials, devices fabrication techniques, and ideal basic
functionality of both Graphene and CNT Field Effect Transistors (FET).
One small note: Throughout this thesis I will often denote Carbon nanotubes as CNTs. Unless
otherwise specified, I am always referring to “single-walled” carbon nanotubes (sometime labeled
SWCNTs). “Multi-walled” nanotubes, possessing nested cylindrical carbon tubes, are very
interesting in tier own right, we simply do not use them here. Similarly, when I refer to graphene,
it is always monolayer graphene. I have always thought this to be the convention, but I have seen
some in the literature use “graphene” to mean anything of a countable number of layers, and then
specify single layers as monolayer.
1.1 Structure and electrical properties of Graphene and CNTs
The electrical properties of graphene are a direct result of the lattice structure. Graphene has a
honeycomb lattice. This is the result of the hybridization of the electron orbitals in neighboring
carbon atoms that restructure to form an energy efficient configuration. For this case the four
valence electrons in carbon (2s, 1px, 1py) redistribute themselves to form three planar sp
2
hybridized orbitals and a perpendicular 2pz orbital, each with one electron. The planar sp
2
electrons
3
are tightly bound and form the strong covalent bonds that make up the honeycomb lattice, leading
to the often-discussed high mechanical strength of monolayer graphene
4
. The weakly bound out-
of-plane 2pz electrons are the main contributors to the electrical properties of the material.
Figure 2 – Graphene lattice structure: (left) Hybridized orbitals of the carbon atom in graphene. (middle)
Graphene honeycomb lattice with Bravais basis. (right) Reciprocal lattice space with a shaded first Brillouin zone
5
.
We can construct a Bravais lattice from the honeycomb lattice of graphene by the conventional
selection of basis atoms A and B (Figure 2, middle)
5
. This results in a unit cell described by the
basis vectors 𝒂 1
= ( √ 3𝑎 2 ⁄ , 𝑎 2) ⁄ & 𝒂 2
= ( √ 3𝑎 2 ⁄ , −𝑎 2) ⁄ , where the lattice constant 𝑎 =
√ 3𝑑 𝐶 −𝐶 = 2.46 Å and the inter carbon spacing 𝑑 𝐶 −𝐶 = 1.42 Å. The corresponding reciprocal
lattice is presented in the right panel of Figure 2 with the first Brillouin zone shaded and symmetry
points labeled. For the sake of brevity, I will merely state that the reciprocal lattice structure can
be implemented to solve for the energy dispersion relation as is done for many metals. For
graphene this can be computed by numerically
6
solving the time-independent Schrodinger
equation or analytically with a nearest-tight-binding model approximation
7
. The resulting
dispersion relation for both positive and negative bands is,
𝐸 ( 𝒌 )
±
= ±𝛾 √
1 + 4 Cos (
√ 3𝑎 2
𝑘 𝑥 ) Cos (
𝑎 2
𝑘 𝑦 ) + 4Cos
2
(
𝑎 2
𝑘 𝑦 )
Where 𝛾 ~ 3.1𝑒𝑉 ∝
𝑣 𝑓 𝑎 is the carbon-carbon interaction energy, and 𝑣 𝑓 is the Fermi velocity. The
two- and three-dimensional plots for this dispersion relation can be seen below (Figure 3
5,8
). The
4
feature of note here is the location of the fermi level. For a system of N unit cells, the 2N available
electrons (one from each of the two atoms per unit cell) exactly fills the N available states in the
valence band. We see that this places the fermi level right at the intersection of the 𝜋 bands, directly
at the K-points.
Figure 3 – Graphene dispersion relation: (left) Two- and (right) three-dimensional dispersion relation for
graphene. (right, zoom) Low energy regime showing a linear dispersion.
We now consider the states about the fermi level to gain a picture of the low energy
behavior of the material, which is where electrical properties for device applications occur. By
expanding the dispersion relation for small k values, about the K points, one obtains the linear
dispersion relation,
𝐸 ( 𝒌 )
±
= ±ℏ𝑣 𝑓 |𝒌 | = ±ℏ𝑣 𝑓 𝑘
With k in spherical coordinates. From here we can calculate the density of states (DOS) about
the fermi level
9
,
ℊ( 𝐸 ) =
2
𝜋 |𝑘 (
𝑑𝐸 𝑑𝑘 )
−1
| =
2
𝜋 ( ℏ𝑣 𝑓 )
2
|𝐸 |
with this we see that the DOS is zero at the fermi level (E=0), but there is no band-gap. This
means that the conduction and valence bands of the material touch, but only at a singular point,
5
where there are no available states. This classifies graphene as a semi-metal. Before we move onto
the device applications, we can first apply the known band structure for graphene to quickly derive
the electronic properties of carbon nanotubes (CNT’s).
One can think of a CNT as simply a sheet of graphene rolled into a hollow cylinder. Though
this is not how they are made in practice, it is a very helpful thought experiment for determining
the CNT band structure. This is done by defining the translation vector, 𝑻 and the chiral vector 𝑪 ℎ
,
on the surface of the graphene lattice (Figure 4,left)
10
. The chiral vector connects two points on
the graphene lattice that, after the rolling of the sheet, will sit on top of one another. It is described
in term of the lattice vectors as 𝑪 ℎ
= 𝑛 𝒂 1
+ 𝑚 𝒂 2
( 𝑛 , 𝑚 𝜖 ℤ) , thus yielding the carbon nanotube
naming scheme of an “(n,m)-nanotube”, like that of the (4,2) tube in Figure 4 (left). The translation
vector 𝑻 denotes the smallest lattice vector along the length of the tube, defining the primitive unit
cell. One can then construct the axial and circumferential reciprocal lattice vectors (𝑲 𝑎 & 𝑲 𝑐 ) by
the reciprocity condition
9
, and solve the periodic boundary conditions to obtain the allowed
wavevectors. For the sake of space, I will state the resulting conditions on both the axial ( 𝑘 ) and
circumferential ( 𝑞 ) wavevectors, then discuss their results
5
.
𝑘 =
2𝜋 𝑁 𝑢𝑐
𝑇 𝑙 , 𝑙 = 0,1, … , 𝑁 𝑢𝑐
− 1
𝑞 =
2𝜋 𝑪 ℎ
𝑗 , 𝑗 = 0,1, … , 𝑗 𝑚𝑎𝑥
Beginning with the k-values, we can see that they are quantized with a period inversely
proportional to the length of the tube 𝐿 = 𝑁 𝑢𝑐
𝑇 , where 𝑁 𝑢𝑐
is the number of unit cells in the
nanotube and 𝑇 is the axial length of each unit cell. For “long” nanotubes ,as is nearly always the
case in practice, 𝑁 𝑢𝑐
≫ 1 and we are left with continuous range of k-values centered (
𝜋 2
, −
𝜋 2
) on
the first Brillouin zone. More interestingly, the q-values are separated by the much larger gap of
6
2𝜋 𝑪 ℎ
, resulting a discrete set of 1D line-cuts of the graphene reciprocal lattice depicted in Figure 4
10
(right). These lines define cross-sectional cuts of the 3d graphene dispersion relation that make up
the dispersion of the CNT which generated them.
Figure 4 – Deriving CNT properties from Graphene solutions: (left) Carbon nanotube lattice vectors
superimposed on the graphene lattice, depicting the chiral and translational vectors used to define the rolling of the
graphene sheet. (right) 1D line-cuts of the graphene Brillouin zone due to quantization in the circumferential direction
of the CNT.
One could now apply these quantization rules to the dispersion relation of graphene and derive
general solutions for the band structures of all ( 𝑛 , 𝑚 ) -CNT’s. Here, we will make a qualitative
observation, given what we know about the graphene band diagram. We found earlier that the
valence and conduction bands of graphene touch at only the K-points, and this leads to its semi-
metallic nature. It then follows naturally that if one of the CNT cut-lines crosses a K-point on the
Brillouin zone, its bands will touch and will be metallic (Figure 5, left)
5
. Otherwise, there will be
a bandgap and the CNT will be a semiconductor (Figure 5, right)
5
.
7
Figure 5 – CNT band structures: (a) (10,4) and (b) (10,5) nanotube band structure exhibiting the metallic and
semiconducting nature of CNT’s respectively.
The exact bands for each unique nanotube will differ greatly, but figures above act as
representatives of each of the two potential electronic outcomes and will be very similar to others
of their type (especially close to the Fermi level). The one exception being that the size of the
bandgap in the semiconducting CNT’s shrinks with growing diameters as 𝐸 𝑔 ≅
0.9( 𝑛𝑚 ∗𝑒 𝑉 )
𝑑 , which
agrees qualitatively with our derivation. As the diameter of the tube increases, so does the
circumferential vector. This shortens the distance between the q-states, making it more likely that
a line-cut will fall nearer the K-points. This results in a shorter gap.
One final note: Work has been done on the statistical analysis of possible ( 𝑛 , 𝑚 ) combinations,
and their resulting electronic properties. It was found that for randomly generated ( 𝑛 , 𝑚 ) that
roughly 1/3 of the resulting CNTs will be metallic and 2/3 will be semiconducting. To date, there
is no method for fabricating a particular CNT chirality selectively. While the various growth
methods (discussed below) do not perfectly sample all the possible ( 𝑛 , 𝑚 ) combinations, this 1/3
to 2/3 ratio is taken as a general expectation.
8
1.2 Fabrication methods and basic electrical properties of Graphene and CNT
devices
With an introduction of the most basic structural and electrical properties on graphene and
CNT’s addressed, we now consider how researchers go about growing and isolating these
materials.
Growth of SWCNT’s
Most methods to produce CNT’s share a few common ingredients which are crucial to growth.
An inert environment free of air, a source of carbon (methane, ethanol, etc,), a catalytic metal for
the tubes to grow off of (often Fe, Co, or some alloy), and lastly some injection of energy. This
last one is where the various methods differ the most. The original method that lead to the
discovery of the first multi-walled nano-tube
2
and later SWCNT’s
2,11
was that of arc discharge.
Here a high current was passed between small carbon electrodes that had been packed with a
catalytic metal in an inert gas environment of either He, or Ar/methane (depending on catalyst
used). The vaporizing of the catalytic metal produced a soot like substance deposited on the
electrodes, which upon investigation, was found to contain CNT’s. Since this discovery
researchers have established several alternative methods, the two most popular being laser
ablation
12,13
, and chemical vapor deposition (CVD)
14
. All three methods have been significantly
studied and have been utilized to fabricate a wide range of SWCNT distributions including isolated
individuals, aligned arrays, and dense films.
Given the interest of this thesis, a short explanation of the CVD growth of CNT’s follows, as it
has been the most successful at producing single, isolated CNT’s for use in electronic devices. It
should be noted that CVD as a general growth method is very versatile with wide-ranging
application that include a multitude of different procedures. The specifics or the experiment design
9
can vary greatly depend on desired result, even just when considering different CNT applications.
As such the following is only meant as a simple introduction that is mostly tailored to the devices
used in the following work.
Chemical vapor deposition (CVD) refers to a technique where a molecular gas containing the
desired atomic species for deposition is heated to very high temperatures over a catalytic substrate.
This environment leads to the decomposition of the gas molecules, breaking free the goal atomic
species, which is then deposited on the substrate. The growth of CNT’s takes place in a quartz tube
furnace. Prior to loading, the substrate upon which CNT’s are to be grown is first coated with
catalytic nanoparticles (usually metal oxide alloys). These many faceted nanoparticles act as
collection sites for carbon atoms that are generated by the decomposition of a carbon feedstock
gas which is flown into the quartz tube at very high temperatures in the presence of Argon and
Hydrogen. By varying the temperature, durations, and gas flow rates of such a system, researchers
have been able to accurately control both the growth direction and density of carbon nanotubes on
a surface
15–17
.
Mechanical exfoliation and CVD growth of graphene
The chronologically first method for the isolation of graphene is perhaps one of the greatest
displays of the beauty in simplicity. Researchers achieved the isolation of “few” and single layer
graphene via the “Scotch tape method” which was later referred to as “Mechanical exfoliation”
1
(
Nobel prize winning achievement). Here, a piece of tape is first stuck to the surface of graphite.
Upon removal, many layers of graphene are left stuck to the tape. With successive applications
and peelings of tape, researchers were eventually able to whittle the stacks of graphene down to
very thin layers. These layers are then deposited onto the desired substrate by sticking the tape
containing the few layer graphene onto the desired surface and dissolving the tape away in a bath
10
of acetone. Now just because something is simple, doesn’t mean it is easy. This technique, which
I have simplified in writing, takes a good deal of preparation and finesse but is capable of
producing graphene flakes on the order of about 10 µm in size
18
. These flakes are then generally
transferred onto Si/SiO2 wafers and are etched into convenient shapes before being surface
contacted via electron beam lithography (see next section for details). While capable of producing
good quality graphene, this method is time consuming and has an extremely low throughput with
all of this work producing just a single (albeit high quality) device.
An alternative to this method is CVD growth of graphene which is done in a similar fashion as
the process described above for CNT’s and has been extensively reviewed
19,20
. The difference
being the catalytic substrates used are specially prepared metal films (popularly Cu or Ni). These
films are first annealed in hydrogen at high temperatures to increase metal domain sizes on the
surface. These large domains, act as the collection surfaces of carbon atoms during CVD, creating
“stained glass” like films of graphene at wafer scale sizes
21
. These large films of graphene can then
be fashioned into arrays of functioning devices via lithographic and etching techniques, which cut
out rectangular channels of graphene at regular intervals before electrically contacting them. This
has the obvious benefit of producing more devices per unit time invested in fabrication. Since there
is not specific targeting of single grains of graphene but instead cutting out of a sheet with irregular
domains, potential devices might be non-functional or electrically degraded due to defects
22,23
.
Even with this failure mechanism, the higher yield of this technique dwarfs that of mechanical
exfoliation.
Device lithography techniques
The methods for the growth of CNT’s and Graphene onto sample substrates discussed above are
just this first step in the fabrication of a functioning carbon-based device. Whether before or after
11
the addition of the CNT/Graphene channel some form of metal will need to be deposited to make
good electrical contact. To achieve this in a controlled manner one of two standard lithographic
techniques is applied, either electron beam lithography (EBL) or photolithography. Very generally,
in both cases a wafer is first coated with a “resist” material. This resist forms a thin film on the
substrate surface and happens to be sensitive to the exposure of whichever excitation source one
happens to be using. These sources being either a focused beam of electrons in EBL or a UV lamp
in photolithography. Exposure of the resist in either case causes a chemical change in the resistive
film. For “positive” resists, exposed areas become soluble in a developer solution and can be
washed away to create openings for deposition to reach the substrate surface. In “negative” resists,
the opposite is true, and exposed areas are made resilient to the developer wash and are the only
features left behind. These techniques have a very broad range of applications throughout the
fabrication of microelectronics and can be used in tandem with other techniques and complex
multi-step processes for much more than just laying down surface contact patterns. Details of their
full capabilities be found in many textbooks and reviews
24–28
, but for our uses here the methods
are best described by way of a very simple example, with some pictures.
Consider the use of positive resist to pattern and deposit metal onto a substrate shown in Figure
6 below as a series of profile views at every step of the process.
12
Figure 6 – Overview of lithography techniques: (a) a film of resist is applied to the substrate surface. The resist is
then exposed (b) in a patterned manner where metal deposition is desired. (c) Exposure weakens the resist material in
those areas, making them soluble in a developer solution. (d) After development, the substrate surface is exposed in
the patterned areas. (e) Metal is then deposited over the whole substrate. (f) Stripping of the remaining resist film
removes unwanted metal deposition.
We begin (a) by spin coating a thin film of resist onto the substrate surface. Next (b) the resist
is exposed in only the areas we wish to deposit metal. This exposure modifies the resist layer in
those areas (c, green), making the soluble and easily removed by bathing the wafer in a developer
solution (d). Next a thin layer of metal is now deposited over the whole substrate via electron-
beam physical vapor deposition
29
(e). Very briefly, in a vacuum chamber a focused beam of high
energy electrons is directed at a small crucible containing the metal to be deposited. The intense
beam of electrons heats the metal to boiling, filling the chamber with a metal vapor. This vapor
then condenses onto the surfaces of both the chamber and whatever substrates are mounted within,
depositing a thin film of metal everywhere. By controlling the beam intensity, and the deposition
time, operators can deposit films of varied thickness. Finally, the remaining resist is “stripped” off
the substrate surface by a solvent bath (and occasional sonication). The removal of the resist takes
all the unwanted metal along with it, leaving only the deposition on the pre-patterned areas (f).
13
This concludes a very general outline of steps required to deposit contact metals onto a substrate
surface. We now consider the pros and cons of each source of exposure and some of the detail
required for step (b) in Figure 6.
During EBL, a beam of electrons from a modified Scanning Electron Microscope
30
(SEM) is
used to draw the design quite literally onto the resist surface. This technique is quite precise as the
SEM interface allows the user to image the surface and select a specific target location prior
lithography. After finding the area of interest computer aided design (CAD) software can then be
employed to design completely custom structures for the focused electron beam to trace out on the
resist, making this application very versatile. This ability to locate a particular surface feature upon
which to pattern tailored design makes this the best available method for patterning contacts onto
small mechanically exfoliated graphene flakes, that must first be tracked down on the substrate
surface. The main drawback of EBL is that it is only economically used for small scale
applications. Drawing large patterns onto very big wafers would take a good deal of time, and in
many applications would not require the high resolution of the SEM. That being said, it is possible
to use the EBL method to pattern small arrays of devices onto CVD graphene wafers with a multi-
step processes, and this will be discussed in some detail later in section 3.1.2. The vast majority a
large-scale design work, particularly at the industrial level, is done via photolithography. As it
sounds, the user now uses an intense beam of light to activate a photo-resistive film on the substrate
surface. Unlike the EBL, which is focused to a point, a photo lamp is not able to selectively target
specific areas on the substrate surface. Instead, researchers fabricate their designs into “photo-
masks” which are often metal sheets with the layer design cut out by detailed milling or etching
process. This mask then sits over the substrate surface, blocking the incoming light everywhere
except for where the design features should be patterned. The need for a mask, and thus a pre-
14
defined design, does make this technique a bit less flexible than the E-Beam. On the other hand,
the masks are completely reusable, and with careful alignment can be used in multistep processes
on large scale wafers to produce a high fabrication throughput.
Basic device properties and electrical response
With a basic sketch of electronic properties of the materials and basic device fabrication in place,
we can now introduce the carbon-based field effect transistor. There are many possible device
geometries to consider while researching the application of these materials. We will focus on
simple back-gated Field-Effect Transistor (FET) structures
10,31–33
as they are the most used in
research settings. The reason for this is their “ease” of fabrication (a term used lightly) and
increased quality. These devices are not great candidates for upscaling to industry needs but serve
as great systems for the study of general device properties and dopant interactions.
Reviewing quickly, monolayer graphene for use in devices is isolated in one of two ways. It is
either mechanically exfoliated via the “scotch tape” method
1
or grown onto a copper substrate by
chemical vapor deposition (CVD)
19
. In either case, the resulting graphene is then transferred
(directly or by PMMA lift-off) to a Si/SiO2 substrate where the doped silicon layer can serve as a
back-gate electrode. Finally, metal source and drain contacts are lithographically defined to
produce a functioning device like that seen in Figure 7
8,31
below.
15
Figure 7 – Graphene FET: (left) Cross-section of a graphene-FET
31
. (right) a microscope image of a graphene
Hall bar with characteristic back-gate dependence curve
8
.
As we discovered in section 1.1, graphene is a gapless semi-metal conductor. This is made
evident in the above conductance curve of a functioning device (Figure 7, right). Here the electric
field generated by the back-gate voltage controls the available charge carrier population by shifting
the Fermi level about the K-point. We see a seamless transition from hole (𝑉 𝑔 < 0) to electron
(𝑉 𝑔 > 0) conduction along the channel. This nature makes large scale graphene an unlikely choice
for industrial applications that usually takes advantage of the “off” state in semiconductors for
logic circuits. The benefit of the graphene-FET is that it provides a highly tune-able system that
exhibits very large carrier mobilities (routinely 10
4
cm
2
V
-1
s
-1
and as high as 10
6
in carefully
controlled systems). Additionally, the conducting channel is exposed and easily modified by
surface adsorbates. This makes it a fantastic candidate for studying charge transfer and carrier
mobility characteristics of surface dopants. It should be noted that the gapless nature of graphene
degrades as the width of the channel shrinks. This has inspired efforts into the growth of graphene
nano-ribbons with controlled band-gaps for use in more traditional transistor applications as well
as research
18,31,34,35
.
16
CNT-FETs, like graphene, are fabricated onto Si/SiO2 substrates creating a similar device stack
as that seen in Figure 7. One upside to the CVD growth of CNTs is that they can be grown directly
onto the device substrate (or contacts), without the need of any transferring. This technique has
also been refined to grow suspended nanotube devices like that shown schematically in (Figure
8
33,36
). Devices grown in this way suffer from less defects (that form by growing the tube along a
rough surfaces) and less carrier scattering (by trapped charges in substrate defects) than their
surface mounted counterparts. Devices such as these are used in the following results sections and
more detail on their fabrication will be covered in section 3.1.1.
As expected, CNT-FET’s come in two varieties, metallic and semiconducting. Typical back-
gate transport characteristics for each are shown below. For the semiconducting device we most
often see a p-type behavior with on/off ratios as high as 10
7
. This hole transport stems from the
work function of the contacting material (generally Au, Pd, Pt) that aligns itself closer to the
valence band of the CNT
37–39
.
Figure 8 – CNT FET: (left) Suspended CNT’s FET design
36
. (right) Back gate characteristics of a metallic and
semiconducting CNT FET
33
.
Conductance along the CNT channel is often described in the framework the a Landauer
transport model
5,40
as,
𝐺 =
2𝑒 2
𝑁 𝑐 ℎ
𝑇 ℎ
17
Where 𝑁 𝑐 ℎ
is the number of degenerate energy sub-bands (2 for low bias and room temperature)
and 𝑇 is the transmission coefficient of the charge-carriers. The transmission coefficient can be
thought of as the probability of an electron (or hole) to travel through the channel without
scattering. If 𝑇 = 1, we recover the formula for ballistic condition in one dimension. This has been
observed in CNT’s at low temperatures where the mean free path (MFP) of the electron exceeds
the length of the channel. Here the resistance of the device is limited by only by electron scattering
at the contacts that is required by their potential difference
41,42
. In most practical situations
electrons travel in a diffusive regime ( 0 < 𝑇 < 1). Here, the transmission can be determined by
its relation to the electron MFP for various scattering mechanisms. Device resistance as a function
of length, temperature and bias voltage can be expressed as,
𝑅 ( 𝐿 , 𝑇 , 𝑉 ) = 𝑅 𝑐 + 𝑅 𝑞 𝐿 𝑙 𝑚 ,𝑒𝑓𝑓 ( 𝑇 , 𝑉 )
where 𝑅 𝑐 and 𝑅 𝑞 are the contact and unit quantum resistance respectively. The total effective
electron MFP 𝑙 𝑚 ,𝑒𝑓𝑓 ( 𝑇 , 𝑉 ) can be determined by adding the MFPs of the contributing scatters
according to Matthiessen’s rule, for example,
𝑙 𝑚 ,𝑒𝑓𝑓 = (
1
𝑙 𝑎𝑐
+
1
𝑙 𝑜𝑝
+
1
𝑙 𝑑 )
−1
With the various MFPs representing those for acoustic phonons, optical phonons, and the effect
of defects respectively. For rigorous calculations of device resistance (and therefore conductance)
the challenge becomes determining the functional form of these various MFPs in the range of
device conditions of interest (bias, length, and temperature). A good deal of work has been done
resulting in fairly complete models for device conductance that appropriately represents
experimentally obtained results
37,43–46
(note this isn’t a complete list of references, just a selection
18
I have found useful). At low biases and room temperature it is enough to account for only
acoustical phonon scattering (~
1
𝑇 ) for back-of-the-envelope calculations
5
.
Above we have quickly reviewed the effect of intrinsic scattering on transport in CNT devices.
Of interest in research settings is the incredible sensitivity of transport in CNT’s from extrinsic
scatters. These include substrate interactions via trapped charges
47–49
and most importantly surface
adsorbed species. This sensitivity arises from the reduced channel dimension. Any disturbing
potential that seeps into the channel greatly effects the flow of charge carriers who have no way
of avoiding the barrier to conduction. This has led to the development of CNT-FET’s as gas and
biological sensors
50–54
. By monitoring source-drain current fluctuations, researchers are very often
capable of detecting single adsorbate interactions with can be molecular species or individual
ions
55–58
. This sensitivity makes these devices very interesting not only for industrial applications,
but also as test systems for investigating fundamental physics at the nanometer scale. This is work
I have been lucky enough to contribute to and will be covered in section 5.
19
2 Atomic nanoclusters and applications to nanoscale devices
Here I introduce the history and basic electrical properties of small metallic nanoclusters,
which are the focus of the cluster-based experiments in the following results sections. I briefly
cover several emergent finite-size effects of small metal clusters that motivate the goal of
incorporating such clusters into novel micro/nano electronic devices. Finally, I propose the use
of carbon-based device architectures as prime ideal systems for the further investigation of
surface supported nanoclusters.
2.1 Shell structure in small metallic clusters
The study of clusters, small groupings of atoms, makes up a rich field that bridges the gap
between physics at the scales of individual atoms and that of bulk materials (Figure 9, left). In the
interest of time and specificity, we will restrict our focus to the discussion of small metallic
clusters. Research in this field gained extensive interest following the discovery and subsequent
investigation of shell structure in alkali clusters by the group of Walter D Knight as UC Berkley
in 1984
59,60
. While studying sodium clusters in the gas phase, researchers found large peaks in the
abundance spectra where certain cluster sizes were unexpectantly preferable (Figure 9, middle).
Figure 9 – Summary of metallic nanoclusters: (left) Close-packed small atomic nanoclusters exhibiting
geometric shell structure. (middle) Sodium abundance spectra collected by the Knight group at UC Berkeley. (right)
A self-consistent Jellium model for electron energy levels of a Na 40 nanocluster, resulting in shell structure.
61
20
Their proposed explanation was the existence of shell structure, very much by analogy to the
model for the electron levels in the hydrogen atom. Here the valence electrons of the small metallic
clusters become delocalized from their constituent atoms and organize to form quantized energy
levels. By applying various central potentials Knight’s group found that (as in the case of the
hydrogen atom) one can analytically solve the resulting separable differential equations for the
electron energy levels (Figure 9, right). With this method they found they could model the size
and filling order of the resulting energy levels to help explain their results
62
. The peaks in
abundance coincided with the completed filling of energy shells in the cluster and thus represented
a more stable configuration. The total number of electrons at shell fillings are often referred to as
“magic numbers” and their corresponding clusters “magic clusters”. As the number of atoms in a
cluster grows larger, the sharp transitions in structure due to electron energy levels begin to fade,
giving way to geometric shell structure where clusters self-assemble in to many faceted symmetric
shapes to minimize their energy (Figure 9, left). The realization of quantized shell structure of in
small metallic clusters has led them to be extensively investigated with focus on both modeling
and experimentally probing this finite-size effect and its consequences on cluster properties
60,61,63–
70
.
2.2 Finite-size properties of clusters
The ordered structure of atomic clusters gives rise to a wide range of interesting phenomena,
and I will mention just a few here. Due to their electronic confinement, small metal clusters have
been predicted and show to be incredibly light sensitive, with an easily induced surface plasmon
resonance, which has garnered much interest in the field of photonic and plasmonic applications
71–
74
. For example, many have theorized assembling nanocluster arrays such as that seen in Figure
10
71
below which use photon emitters to excite a localized surface plasmon that propagates along
21
an array of metal nanoparticles. These systems could be used to better understand how
nanoparticles could be incorporated into novel single photon or single plasmon devices for
exhibiting quantum control in nanoscale circuits. Optically active clusters have also received much
interest as fluorescent quantum dots, particularly in medical community as contrast agents in
bioimaging
75–77
.
Figure 10 – Plasmonic clusters: Theoretically investigated plasmonic nanocluster array in which light entering from
a source excites a local surface plasmon that propagates through the cluster array before exiting out the drain contact
71
.
Similar cluster arrays and have been proposed for single electron transistors (SET) or tunneling
junction field effect transistors (TFETs) which use a single (or few) nanoclusters to bridge the
gap between two metal contacts
78–81
(Figure 11). The ordered structure of electrons in small few-
nm clusters allows for the excitation of a single excited state in the transistor junction even at
room temperatures. Recent successes have realized single functioning devices and continue to
encourage further work in developing methods of fabrication for such systems at larger scales
22
that could open the door to a new generation of low power transistors and sensors for everyday
use.
Figure 11 – Cluster arrays for tunneling junctions and SETs: (left) Schematic illustration of various nanocluster
based arrays for use in novel devices/materials with tunable properties
63
. (right) Cartoons of recent successful works
that have achieved nanocluster based SETs
80,81
Another large field of research focuses on nanoclusters as catalysts for various chemical
reactions. Clusters speed up such processes due to their varying electron affinities, increased
surface-to-bulk ratios, or many reacting surfaces in the form of faceted edges. Here, metal
clusters (particularly noble metals) have been shown to be highly reactive in processes such as
the splitting of water, the oxidation of Carbon Monoxide (CO), and energy creating reactions in
fuel cells
82,83
.
One predicted consequence of shell structure that provides significant motivation for my work
is the possibility of a drastic increase in the superconducting transition temperature ( 𝑇 𝑐 )
84–86
. This
property stems from the high degeneracy in the electron levels near the fermi level for some
clusters. For example, the highest occupied level for a closed shell “magic” cluster has a
degeneracy of 2( 2𝑙 + 1) , where 𝑙 is the angular momentum quantum number used to index the
23
levels. The high degeneracy causes a peak in the density of states of the cluster which is
proportional to the coupling constant 𝜆 . This in turn, enhances both the gap parameter Δ and 𝑇 𝑐 of
the cluster. The combination of these two factors leads to a favorable environment for high 𝑇 𝑐
superconductivity in clusters with both numerous highest occupied levels and small energy gaps
to their next lowest unoccupied level.
Figure 12 – Gas phase SC Al clusters: Experimental design used for the detection of a SC phase transition in
small Aluminum clusters. Clusters were generated with a DC magnetron gas aggregation source. Before being ionized
by a tunable laser, cluster travel through a thermalizing set to a fixed temperature by a helium cold head. Ionized
clusters are separated and measured via time-of-flight spectroscopy to determine the photoionization yield as a
function of cluster size.
Evidence for this enhanced 𝑇 𝑐 was observed in small aluminum clusters in previously published
research by Dr. Avik Halder and Dr. Vitaly Kresin at USC
87–89
. To do this they designed a cluster
beam photoionization spectroscopy experiment shown schematically above (Figure 12)
87
. Here,
the abundance of ionized closed shelled Al66 clusters were and measured in the gas phase as a
function of temperature by passing them through a small thermalizing tube attached to the source
aperture. A system such as this allows for the mapping of the electron energy levels via a
mechanism similar to that of scanning-tunneling spectroscopy on bulk systems
90
. By varying the
temperature of the clusters and monitoring the change in the ionization yield as a function of
24
ionization energy, they managed to effectively trace out the electron density of states of the clusters
in flight
87,88,91
.
Figure 13 – Photoionization spectroscopy results: (left) Evolution of a bulge like feature arising in closed shell
Al 66 clusters. (middle, right) Differential yield peak increasing with decreasing temperature suggesting an electronic
phase transition near 100 K.
The key result for this experiment was the formation of a bulge in ionization yield curve for the
“magic” closed shell cluster Al66 (Figure 13)
88
. Investigating the differential yield curve, they
noticed formation of a large peak in the density of states with decreasing temperature. These peaks
as a function of temperature are plotted in Figure 13 (right) and suggests an electronic transition
occurring around ~100K, nearly 2 orders of magnitude above the SC transition in bulk Aluminum
(~1.2K). This feature can be explained by considering the DOS as a function of T for a
superconducting material. Well above 𝑇 𝑐 , we have a normal metal phase with a continuous DOS
as a function of energy. As the metal transitions into a SC-state the formation of a bandgap appears
in the DOS reflecting the energy gain of the electron pairs. That opening forces all the states that
used to sit in band gap region down, compressing them into a lower level that is a now more
populated peak. This gap continues to grow as the temperature is further lowered past 𝑇 𝑐 , yielding
the growing peak in the DOS.
25
2.3 Integration of metallic clusters into Carbon based devices
With the wide potential applications of atomic nanoclusters established, I now aim the focus of
this thesis on the deposition of size-selected nanoclusters in organized arrays onto selected
substrates in an effort to investigate their potential as a controlled surface dopant onto pre-existing
device architectures. The organized deposition of size-selected nanoclusters would achieve two
goals. Firstly, it would allow the use a slew of surface science techniques and equipment to further
investigate cluster properties as a function of size. This direct probing of properties is crucial
determine the potential usefulness of clusters in novel device fabrication. While much work has
been done to investigate cluster properties in the gas phase, with in-flight spectroscopic techniques,
much work is still needed to characterize the preservation of cluster properties upon deposition.
When deposited, surface interactions have the potential to skew the interesting properties observed
in unperturbed gas-phase clusters. Thus, work must be done to discover the optimal deposition
techniques and supporting substrates that best retain the symmetric structure of nanoclusters that
leads to emergence of such wonderful finite-sized effects. Secondly there is great motivation from
a novel device physics perspective. Ionized clusters, which can be generated from many different
materials and precisely screened by their size, could be used as highly selective and controllable
dopants, imparting their charge or other properties into pre-existing device architectures that are
already well understood leading. Gas phase cluster deposition offers the ability to precisely tailor
dopants to achieve custom made device characteristics. Gas phase deposition also has the added
benefit over popular solution-based cluster suspensions in that there is no use of chemical additives
(surface functionalized ligands) required to keep clusters isolated. These ligands have the potential
to effect cluster properties if left attached, and their removal (through high temperature annealing)
is potentially damaging to the cluster of supporting surfaces. In conclusion, cluster surface
26
functionalization of well-known designs not only has practical applications of the development of
new interesting devices but could also serve as sandbox systems to further study cluster properties.
While I was not able to achieve a direct measure of surface supported superconducting (SC)
nanoclusters in my time at USC. Work was done building an infrastructure toward that eventual
end, by experimenting with our ability to organize clusters into linear arrays or two-dimensional
networks onto device surfaces. Being still largely motivated by cluster superconductivity, an
eventual goal for arranging such systems would be to observe Josephson tunneling. That is,
tunneling of cooper pairs from one superconductor to another through either an insulating barrier
or a “weak link’ contact. Theoretical calculations of such an array of clusters has shown the
capability of supporting supercurrent
85,92,93
. In specific cases, like resonant tunneling between
identical clusters, there are predictions of a large increase in supercurrent over standard
superconducting systems. These theoretical systems are certainly idealized and require some work
on engineering considerations including cluster soft-landing, inert substrates determination, and
high-precision mass selection. That being said, progress toward this goal is also expected to be
taken stepwise, with even partial fulfilment of theoretically proposed gains in Tc and supercurrent
making the investigation worthwhile.
Another possible avenue to achieve an experimental measure of a SC-state would be to organize
cluster assemblies onto a conductive substrate and take advantage of the proximity effect. This is
the process in which a SC-state is induced in a normal conductor that is in direct contact with the
superconducting metal (or in our case, decorated with SC clusters). This system is often modeled
as the tunneling of both cooper pairs and normal metal electrons through a potential barrier at the
interface between the materials
94,95
. A low-temperature transport measurement of such a system
would show direct evidence of the transition via a precipitous drop in resistance.
27
Either of the above described routes would not only allow us to further investigate the nature of
SC clusters from a fundamental physics standpoint, but also make progress toward the
development of a tunable cluster-based devices harnessing other finite-sized properties. To this
end, we must now consider what substrates provide the best environment to investigate both the
surface morphology and various transport properties of deposited nanoclusters. This substrate
should fulfill certain criteria going forward. Firstly, it should disturb the cluster as little as possible.
As mentioned previously, cluster properties are the direct result of structural symmetry. Thus, we
hope to avoid any strong cluster/surface interaction might appreciably deform the energy levels of
the cluster appreciably. In a similar vein, a particularly flat surface would allow us to best
investigate how clusters sit and move on the surface. This is of great interest when attempting to
organize them into tunneling arrays, where one hopes to avoid uncontrollable cluster aggregation
and coalescence. Lastly, there is a heavy preference placed on those surfaces that have pre-existing
and well-studied devices architectures, this way time can spend integrating the clusters as opposed
to designing a device from the ground up. Finally, for its use in proximity superconducting studies,
the surface should be conducting.
With all of this in mind, we have turned toward carbon-based surfaces focusing especially
on CNT and Graphene FETs. Weak surface bonding of these materials makes them a natural choice
to support clusters. There is already some decent evidence to support this theory as well. Previous
STM measurements along with MD simulations have agree that small (Ag561) clusters soft-landed
on a C60 monolayer are not greatly deformed by the surface
96
. While work performed by Hongji
Dia’s group at Stanford has shown that atomic deposition of various metals coating suspended
CNT’s suffer from a lack of wettability to the tube surface
97,98
, causing the sputtered atoms to
coalesce and form small nanoparticles (Figure 14, left). This aversion to coating the tube is a
28
serious draw-back in the efforts to evenly coat nanowires with a dopant metal but is very
encouraging for our work where we look to land discrete clusters along the surface. Similar work
has been done by Yoke Khin Yap’s group at Michigan Tech. with gold clusters deposited onto
hexagonal boron nitride (HBN) nanotubes with in-situ electrical measurements
99
. While not yet
observed on a functioning device, these cluster were capable of supporting normal tunneling
current when contacted via STM probe tips (Figure 14, right).
Figure 14 – Clusters on nanotubes: (left) Atomic deposition on suspended CNTs showing poor wettability of the
deposited metal, a good sign for supporting isolated clusters
97
. (right) Clusters deposited on HBN nanotubes that
display tunneling current when biased via STM probes
99
.
Another large benefit of these materials is that there has already been a huge investment of research
hours towards device design, fabrication, and characterization techniques
100,101
. Though
commercial scalability is still out of reach, the field now has several well-defined methods for
generating quality devices at acceptable yields for research use. Finally, there is some very
encouraging work both theoretically
102
and experimentally showing that with both metallic-CNT
and single layered-graphene devices are capable of being coupled to superconducting dopants via
29
the proximity effect
102–109
. These feats were achieved using SC-metallic contacts in the case of
CNT’s and small SC-islands in that of graphene. Illustrations of this work on Tin/graphene hybrid
devices are shown below in (Figure 15)
Figure 15 – SC nano-islands on graphene: Experimental realization of Tin/Graphene proximity superconducting
devices by both Han et al.
106
(left) and Kessler et al.
103
(right).
With the above cited works in mind, there is now both significant motivation and good
experimental foundation for the further development of cluster/carbon based superconducting
systems. The following sections contain the experimental methods and preliminary results our
group achieved toward the integration of clusters into carbon-based device architectures during
my time at USC. While there is still work to be done before fully-fledged transport measurements
can be undertaken. I was able to establish a supply chain for carbon devices for our group, through
collaboration and personal training, as well as perform preliminary experiments analyzing how
deposited clusters interact with carbon devices.
30
3 Experimental set-ups for various works
3.1 Device fabrication
Suspended SWCNT’s via CVD on pre-patterned trenches
CNT field-effect transistors (CNT-FETs) are grown via chemical vapor deposition (CVD) on
pre-patterned single gate device architectures fabricated at the UCSB Nanofabrication Facility.
The contact pattern consists of a 500 nm deep trench etched in a Si/SiO2/Si3N4 substrate.
Lithographically defined Pt electrodes are then deposited on either side, and the bottom of the
trench by electron beam physical vapor deposition (EBPVD) to function as source/drain and gate
electrodes respectively. Utilizing a “Common gate” design, a single long trench is used to separate
30 pairs of source/drain deposited at regular intervals (see Figure 16). This pattern is then repeated
many times onto a 4 inch wafer before being cut into individual die measuring approximately 0.5
cm square. Patterning of the devices prior to CNT growth is a critical step that serves two main
purposes. Firstly, it allows for the suspension of the CNT’s over the defined micro-trench which
separates the CNT channel from the substrate surface which has be shown to be the source of noise
in the conduction channel due to charge trapping in the dielectric layer
47–49
. Secondly, with the
source and drain contacts already present at the time of growth, the formation of the CNT bridging
the contacts is the final step in the process of producing a fabricating device. This leaves no
additional patterning, depositing, or cleaning prerequired which could potentially damage or effect
an otherwise pristine device.
31
Figure 16 – Micro-trench design leading to suspended CNT-FETs: A series of Microscope and SEM images
depicting the micro-trench device design with CNT suspended between catalytic particles.
Before being shipped, a layer of photoresist is deposited onto the patterned contacts to protect
the electrodes. This resist later is also patterned at the tips of the contacts on either side of the
trench to allow for the selective deposition of the Ferric Nitrate (Fe(NO3)3) CVD growth catalyst.
This deposition is done by drop casting the catalyst solution onto the device several hours before
growth. Just before the growth process the resist layer is removed with and Acetone and IPA,
carrying away any additional unwanted catalyst. Samples are then set into a glass boat and loaded
into a Thermo Scientific (Lindberg Blue M) tube furnace with automated temperature PID
controls. An MKS multi-gas flow controller (647C) is used to maintain the appropriate flow of
argon, hydrogen, and ethanol feedstock gas throughout the course of deposition. Since CNT’s are
32
known to grow along the direction of gas flow, samples are placed with the trench perpendicular
to this flow to encourage CNT’s to bridge the gap
15
.
After growth, electrical characterization measurements of each junction of a die were conducted
using one of two similar probe stations located at either USC (Singatone 1100 series) or the
Aerospace corporation. These stations both offer (at least) four low noise, micro-manipulated,
probe tips for surface contacting devices on the die surface. These probes are wired to a semi-
conductor parameter analyzer (Agilent 4156C, or similar) which an incredibly versatile
voltage/current source capable of a number of microelectronic characterization techniques. Our
uses were comparatively simple. Constant source (~100 mV), back-gate sweeping (± 4 V)
measurements while monitoring the source/drain current determines if the junction remains open
or has been bridged by either a metallic (M) or semi-conducting (SC) CNT, as discussed in section
1.2.4 (see Figure 8). Through many iterations of development by Professor Cronin’s group, and
collaborators, the CVD process has been tuned to achieve the suspension of a single, isolated CNT
between the source and drain contacts with relatively high efficiency. In my experience, each die
containing 30 junctions generally results in several functioning devices. Devices as described have
been used in many studies and have been shown to be of extremely high quality without extraneous
defects generated by postprocessing or noise stemming from substrate interactions.
49,55–57,110–114
After characterization, die with functioning devices are then mounted to either ceramic chip
packages for use in ionic interaction experiments, or AFM sample pucks which act as a ground
planes for cluster coating experiments. In either case, device contacts are wire-bonded using an
ultrasonic wedge-to-wedge bonding system
115
to allow for electrical control by external sources
depending on the planned experiment (described later).
33
E-beam patterning and contacting of CVD Graphene
Graphene devices were fabricated in a cleanroom environment at the Aerospace corporation in
collaboration with Dr. Sean Stuart. The process begins with a commercially produced 4” wafer of
CVD grown graphene on an Si/SiO2 substrate (285 nm oxide layer) purchased from Grolltex Inc.
This wafer is hand cut using a diamond pen into approximately 1 cm square die. Each die is then
patterned with a 5x5 array of four probe graphene field effect transistors (GFETs) via a multi-step
EBL process detailed below. In the interest of space and clarity, I will be covering this process
briefly for just a single device. This only omits one small detail in that the devices are actually
fabricated 25 times in parallel on each die. Given that the whole recipe requires multiple trips in
and out of the EBL, we need to have a way to realign the wafer appropriately between each
fabrication step. To do this, the first step in the fabrication process is to deposit alignment markers
at regular intervals. This is done in the very same fashion as the surface contacts that I go over in
step-by-step detail below, only that the shapes are a bit different (visible in Figure 19 to follow).
In practice, computer software is programed to recognize these shapes, which sit off to the side of
the device location. By identifying a few of the markers in the array, the software is then able to
triangulate the exact location of the device patterning without having to target it directly. This is
important because any exposure to the beam of incoming electrons would begin to cure the resist
material (Polymethyl methacrylate, PMMA for short). The precise timing of the exposure is very
important to the accuracy of the final pattern. Therefore, these markers and the recognition
software are critical in saving the potential overexposure while hunting down the proper location
of the device by hand.
34
Figure 17 – Defining The graphene channel via EBL and argon etching: Schematic representation of the
patterning process used to define the graphene channel. PMMA is first spun onto the entire wafer surface (a) before
being patterned with the channel in the EBL machine (b). After development only the exposed rectangle remains (c).
Argon plasma sputtering is then performed, removing all excess graphene and leaving only the area protected by the
rectangle after PMMA stripping (d).
The die begins with Graphene covering the entire surface which we must first shape into the
conduction channel of the device. To do this a positive PMMA resist spin-coated onto the die
surface (Figure 17a) and baked in on a hot plate in a nitrogen environment at ~180 C for 5 mins.
Next an EBL system is used to define the graphene channel by exposing all but the aera we wish
graphene to remain Figure 17b . After development we are left with a small rectangle (30 by 100
microns) of resist on a field of graphene Figure 17c. The die is then loaded into argon plasma
sputtering system (Gatan PECS) which bombards the die surface with argon ions that etch away
all the graphene the is not covered by the protective PMMA layer. After stripping the PMMA away
35
we are left with a rectangle of isolated graphene surrounded by an insulating SiO 2 underlayer
(Figure 17d).
Figure 18 – Patterning and deposition of surface contacts: Schematic representation of the patterning process
used to define the graphene surface contacts. Again, PMMA is spun onto the entire wafer surface. After alignment
with the graphene channel (outlined for visibility), the surface contacts are patterned with EBL machine (a). Post
development, the contact shapes are removed from the PMMA down to the substrate surface (b). Metal is then
deposited over the entire wafer filling the recessed areas (c). Stripping the PMMA off the wafer surface leaves behind
the contact metal that mad good contact to the channel and substrate surface (d).
36
With the channel defined, we must now deposit surface contacts for the application of a voltage
bias along the channel length. With an eye on future experiments involving low-temperature
conductance measurements as well as investigating charge transfer of surface dopants, we chose a
four-probe contact layout. This layout will allow for the accurate resistance measurements of the
device ignoring the influence of contact and lead resistance. This is done by again applying a layer
of PMMA to the die surface and patterning the four-probe contacts over the existing channel
(outlined in Figure 18a). Once developed, gaps in the resist allow for deposited metal to make
good electrical contact with the device surface (Figure 18b). The die is then loaded into an electron
beam physical vapor deposition system where Titanium (10 nm) and Gold (50 nm) are deposited
everywhere (Figure 18c, both contacts and graphene outlined for visibility). The thin layer of Ti
serves as a “adhesion layer” for the gold deposition, which does not stick well to the SiO2 or
graphene of its own (more on that later in Chapter 7). Finally, the remaining PMMA is stripped
from the surface by sonicating the die in TCE for ~1 min, leaving behind the contacts on the
graphene channel (Figure 18d).
Figure 19 – Completed device imaging: Optical microscopy imaging of a completed graphene device showing an
isolated graphene channel on an SiO 2 substrate with Ti/Au surface contacts.
37
Figure 19 above shows an optical microscope image of a completed device. Here we see some
slight delamination in the right most contact pad. This issue was seen throughout several devices
on the first round of device fabrication and is thought to be due to either insufficient resist
exposure/washing prior to metal deposition, or the need for a thicker adhesion layer of titanium.
Unfortunately, this first round of devices also suffered from a large number of defects in the
graphene channel, coupled with an overabundance of PMMA residue on the surface. With these
factors acting as both charge scatters
22,23
and dopants
116–118
we were unable to recover the ideal
back gate conduction curve shown in section 1.2.4, with a near-zero charge neutrality point. As
such these devices are not usable for the surface charge doping experiments as we had planned.
We are confident that future iterations of devices can be fabricated to mitigate these issues. We
simply ran into some unfortunate timing in that COVID 19 closures shut down lab operations
shortly after this first round of fabrication. Even still we did manage to find reliable methods for
cleaning the device surfaces (section 3.4.2) to a sufficient level as to make them useful for
morphological studies of graphene/nanocluster interactions.
3.2 Cluster generation and deposition
Machine overview
The Nanogen 50 deposition system used for the generation of clusters was produced by Mantis
LTD, a now closed manufacturer of DC magnetron sources. The complete system contains a DC
magnetron gas-aggregation source, quadrupole mass analyzer and large deposition chamber. Each
of these components will be addressed separately in the following sections and can be seen in the
figure below (modified version from ref
119
).
38
Figure 20 – Illustration of cluster deposition system: Overview of the entire deposition system displaying the
path of the cluster beam at each point of generation through the source, quadrupole and deposition chambers. Vacuum
pump connections at the bottom of the source and deposition chambers are not shown.
All cluster beam deposition experiments occur under vacuum conditions to achieve this both the
source and deposition chambers are outfitted with turbo molecular pumps which are, in turn,
backed by belt driven mechanical pumps. In standard loading conditions, typical system pressure
reached 10
-8
– 10
-9
torr which is significantly greater vacuum than is strictly required to run the
magnetron source (~10
-6
torr). Once plasma and carrier gases are supplied to the chamber, typical
operating pressures reach 10
-3
– 10
-4
torr for the duration of deposition.
DC magnetron gas aggregation source
The system we used for the generation of clusters consists, first and foremost, of a DC magnetron
gas aggregation source. Here, like in traditional atomic magnetron sputterers
120,121
, we have a
covered magnetic stage upon which sits a metal target (Various metals usually from ACI Alloys)
of whatever material is to be sputtered (Figure 21, left). This magnetic stage is then biased with
respect to an anode cup that fits over the stage leaving a just small gap to remain electrically
isolated. Argon gas is passed over the target while a DC electric field is applied to spark an argon
plasma that is magnetically confined just over the target surface. Heavy Ar
+
ions are guided by the
39
magnetic field lines into the target surface “sputtering” off target material atoms. The magnetic
orientation is generally structured with one pole placed cylindrically around the edge of the source
with the opposite in the center. The inward funneling field lines are used to trap secondary
electrons near the surface which greatly increases negative ion production as atomic seed are
sputtered off the target and pass through the dense electron region. While the specific details of
the magnetic design are often proprietary from one company to another, these basics remain
consistent and the evidence for them can be seen by examining a slightly used target that shows
the characteristic circular trench forming where sputtering has occurred (Figure 21, right).
Figure 21 – DC magnetron source head: (left) Bare magnetron source head showing the copper covered magnetic
stage. (right) Assembled magnetron head after Ag metal target sputtering showing the distinctive circular trench etched
out of the target surface.
What makes our system different than sputtering heads used for atomic deposition or ion
implantation is the addition of an aggregation region, and differential pumping through a small
40
diameter hole which together create a beam of cluster. After atoms are sputtered off the target
surface, they are swept downstream by the carrier gases through an aggregation region that is
cooled by a surrounding liquid nitrogen jacket. The carrier gases being the ever-present Argon and
often additional of a Helium. In the aggregation region the target atoms undergo two and three
body collisions with each other and/or the LN2 cooled carrier gas molecules (He) to exchange
kinetic energy and form small cluster seeds
63
. Larger clusters are then grown by either the addition
of metal atoms on to the cluster seeds or by collisions of existing small clusters. From here, further
clustering occurs via an increase in collisions as the gas/cluster mixture is expanded into a chamber
pumped at higher vacuum (lower pressure) via a small diameter collimating skimmer at the end of
the aggregation region. It should be noted that while our system is designed to generate
predominantly negative cluster ions, a distribution of ion species and neutrals is still expected.
Size selection with source parameter adjustment and quadrupole filtering
The source used in the following studies can produce numerous cluster size distributions whose
peak size may vary from roughly 1-10 nm in diameter. The size distribution of our cluster beam
can be both measured and filtered with the use of a quadrupole mass analyzer
122,123
(Figure 22,
left)
61
mounted just past the skimmer leading from the aggregation chamber. The principles of
operation for the quadrupole mass analyzer have been described heavily with many great reviews
and textbooks on the study, even in the context of cluster beam science
61,124–126
.
Briefly, the cluster ions travel down the center of four parallel rods whose alternating pairs are
driven with equal but opposite voltages of the form Φ
0
= 𝑈 + 𝑉𝑐𝑜𝑠 ( 𝜔𝑡 ) , where U and V are the
DC and AC (peak) voltages respectively. The electric fields applied by these voltages results in
forces being felt by the ions as they attempt to pass through the central region parallel to the rods.
These forces result in the equations of motion for an ion of charge 𝑒 and mass 𝑚 ,
41
𝑑 2
𝑥 𝑑 𝑡 2
+
2𝑒𝑥
𝑚 𝑟 0
2
( 𝑈 + 𝑉𝑐𝑜𝑠 ( 𝜔𝑡 ) ) = 0
𝑑 2
𝑦 𝑑 𝑡 2
−
2𝑒𝑦
𝑚 𝑟 0
2
( 𝑈 + 𝑉𝑐𝑜𝑠 ( 𝜔𝑡 ) ) = 0
where 𝑟 0
is the radius of the circle defining the central region between the rods. With some
substitutions, these differential equations take the form of a Mathieu equation which has a known
solution that describes the combination of parameters required for an ion maintain a stable pass
through the rods without be driven off the beam path. The stability conditions that we can control
easily are the magnitudes of the applied voltages 𝑈 and 𝑉 , along with the AC frequency (𝜔 ) . By
adjusting these values, it is then possible to allow the passage of only a small range of ions
depending on their charge to mass ratio (
𝑒 𝑚 ) . With our source conditions the population of both
doubly charged and neutral ion species is very low, and thus we are effectively screening by mass
only. With the known density and atomic mass of the clustering material, it is then possibly to
convert these masses to cluster diameters. In practice, the beam yield as a function of cluster mass
is measured by setting a fixed ratio of (
𝑈 𝑉 ) while slowly varying the AC frequency to step through
the stably transported masses. By collecting a portion of the beam current with a biased grid at the
exit of the quadrupole rods, it is then possible to plot the relative beam intensity as a function of
cluster mass (or size) in real time (Figure 22, right).
42
Figure 22 – Quadrupole design and application to beam characterization: (right) Schematic of a quadrupole
mass filter. (left) Relative intensity spectrum of clusters read by detection grid at the end of the quadrupole rods by
sweeping through stably transmitted cluster sizes.
The quadrupole allows us to take a (relatively) thin vertical slice of the cluster size distribution
that then pass only that size onward to the deposition chamber. This is obviously very useful, but
the quadrupole itself is limited in that it can only select from what is already present in the beam
population. Meaning, if you would like to deposit 3 nm clusters, you need to make sure there are
some in the beam to begin with or setting the quadrupole to 3 nm will give you nothing.
Adjustments to the general size and shape of the beam distribution are done my adjusting the
various input parameters to the magnetron source during sputtering and it is this portion of
operating these sources that starts to feel a little more like art than science. The distribution of
cluster sizes achieved by such a source is a function of many, often interwoven, parameters. These
include, sputtering power, aggregation region length, aggregation region temperature, Ar/He gas
flow, and the source chamber pressure just to name a few. Figure 23 serves and as a representation
of the large range of effect produced by merely one of the above system inputs
127
. The exact
shifting of the beam as a function of one of these parameters varies from one material to another,
and only experience can guide you fully, but there are general trends that are consistent. Let’s
consider a simple example of two changes to the beam.
43
Figure 23 – Example of beam dynamics with shifting gas parameters: The evolution of a Mg nanocluster beam
profile as a function of He flow input with fixed Ar flow (150 sccm) collected by Malak Khojasteh during initial
system characterization experiments.
In general, increasing the aggregation region length (by moving the magnetron head back, away
from the exit skimmer) shifts the cluster beam distribution towards larger cluster sizes. This is
because the cluster seeds now spend more time in the aggregation region, undergoing more
collisions, leading to larger clusters (on average). Consider now, the addition of He gas to the mix.
This very mobile gas species is quite efficient at thermalizing the clusters to the approximate
temperate of the LN2 cooled walls, thereby reducing the internal cluster kinetic energy and making
them more likely to stick together upon collision. This cooling of the clusters leads to more
44
clustering across all size regime as a result. This is true in practice, adding some He does greatly
improve the beam yield of all sizes, but it also shifts the beam distribution to smaller sizes, not
larger as you might expect. This secondary effect technically revolves around the aggregation
length mentioned earlier, because the He has the side-effect of increasing the beam velocity. A
faster beam traverses the aggregation region more quickly, leaving less time for collisions,
resulting in smaller clusters (on average). This is just one example of how various parameter
changes can affect the beam distribution and it serves as an illustration of how various parameters
can interact, leading to the need to compensate one way or another. A detailed survey of the effects
of these parameters on our system for the growth of manganese clusters was carried out previously
by Dr. Malak Khojasteh and Professor Kresin
127
and acts as a starting off point for parameter
settings in the first stages of experimentation. I have also kept notes on various useful parameters
for the deposition of Al, Cu, and Ag. While it can take some time to fully explore the parameter
space for a particular material, the results of such investigations have luckily been shown to be
largely reliable and repeatable. Thus, once you have a combination that generates a sufficient yield
of a particular size of interest, you are relatively set for future experiments.
Substrate mounting and electrical connection
Once the beam has been size selected it then travels into the deposition chamber where it
coincides with one of several custom designed sample mounts. For the work presented here the
two most predominantly used mounts are depicted in Figure 24 and Figure 25 below. Both mounts
are adhered to an L-bracket that connects them to a linear translation stage. This stage is mounted
to a custom 8 inch conflat flange that is attached to the chamber perpendicular to the beam path.
This allows the linear stage to travel through the beam, exposing one sample at a time. Figure 24
shows a very simple mount capable of holding several samples in series (Figure 24a). Samples are
45
adhered to a stainless-steel plate with carbon tape leaving one open space on the mount for the
cluster beam to deposit that does not collide with any sample. This “test spot” is used to measure
the beam ion current in real time by wiring the mount (through a flange feed through) to a Keithly
6487 pico-ammeter/voltage source. This ion current can then be converted into cluster flux and
further into a coverage rate for determining the time of deposition for each substrate. For accurate
coverage calculations the cross-sectional beam area must be known accurately. To control this,
and to ensure that only one sample is exposed at a time, a collimation shield with a known aperture
is mounted in front of the sample mount (Figure 24c). This shield is kept electrically isolated from
the samples by mounting it directly to the flange supporting the linear stage and placing a ceramic
spacer bead on the sample mount to ensure no direct contact is made. Minor experimental note:
the wire attached to the sample mount is enameled copper wire which is only exposed at the very
end that sits in contact with the mounting plate (Figure 24c), so there is no risk of small movements
shorting the ion current to the chamber wall.
46
Figure 24 – Sample mounting system: (a) Overview of a simple mount capable of holding several samples. (b) Beam
collimator used to regulate the beam diameter and expose a single device at a time. (c) Sample mount on the linear
stage showing the connection of the beam current wire.
An additional sample mount is presented in Figure 25 which allows for complete electrical
isolation of each sample, should they need to be biased/measured separately. Upon the L-bracket
are locations to attach a series of 1-inch conductive sample pucks. Sample wafers are adhered to
the surface of the pucks via carbon tape. The pucks are electrically isolated from the chamber, the
sample bracket, and each other by ceramic standoffs. Each puck now has its own copper wire
running to corresponding feedthroughs of ion current measurements. The trade-off to this
configuration is that each sample puck takes up more space and thus the stage is only able to
accommodate 2 samples (and one beam test spot). This limit is set by the linear travel of the
translation stage and could be revisited with an upgrade to that component in the future. Lastly,
47
this mount also possesses a collimation shield with corresponding openings for each sample puck
(Figure 25, bottom).
Figure 25 – Sample mounting system for individual biasing: Overview of mount for electrically isolated samples.
Each sample sits on a wired stainless-steel puck that is separated from the L-bracket by ceramic standoffs. Each puck
has its own collimation opening.
As of the time of this writing I have one additional sample mount that is still in development.
This mount consists of a custom built PCB board adorned with three ceramic chip carrier sockets
identical to those used in gaseous ion experiments discussed shortly in section 3.5. These sockets
are re-usable and designed to hold devices that are wire-bonded to standard ceramic chip
carriers. The chip carrier electrical connections will then be routed to a 25-pin electrical
48
feedthrough via a vacuum safe cable, where we will be able to connect electrical equipment to
take in-situ transport measurements of cluster coated samples without breaking vacuum. Since
many cluster materials of interest risk oxidation when exposed to air, this development could
greatly broaden the scope of our system.
3.3 Atomic Force Microscope (AFM)
Morphological studies were performed with atomic force microscopy using a Cypher
Environmental AFM manufactured by Asylum Research. This is yet another piece of machinery
that is capable of a vast number of different techniques. I am trained up in comparatively few of
them and use it mostly for relatively simple surface imaging. For the interested reader there are a
number of excellent, and free, resources available that cover all the many AFM imaging modes
available along with scanning probe microscopy more generally
128–132
. For a simple start and a
background to the basic physics, I like Atomic Force Microscopy by Voigtländer
128
. It spends time
on system design and technical elements in detail that can get glossed over in some other works,
like piezoelectric devices, control electronics, and types of artifacts. As a random aside, I found
these references through the Springer book catalog. If are a graduate student and haven’t’
discovered this yet go take a look. USC has its own link that gives access to a huge library of free
textbooks. They aren’t all amazing to be honest, but they are free, and there is a high volume of
them. Chances are you can find something useful. Lastly, another good resource are the system
user manual and applications guide provided by Asylum themselves. There are several
underdeveloped sections, but they have gotten a lot better over the years. Always check to be sure
you have the most recent version.
49
Figure 26 – Non-contact AFM: Schematic representation of an atomic force microscope in non-contact mode
showing details for both frequency modulated (FM) and amplitude modulated (AM, dashed lines) operation
130
. Here
the tip is driven into a set oscillation over the sample surface and feedback loop is utilized to correct for variation in
the oscillation amplitude (AM) induced by tip interactions with a varying surface profile.
Experiments involving surface imagine of devices and size/coverages characterization of
clusters were conducted in AC tapping mode (Figure 26
130
). Here a piezo stage drives a small
cantilever tip into resonance near a sample surface. The amplitude of the oscillatory motion of the
tip is measured by tracking the motion of a laser reflected off the tip surface onto a photodetector.
As the tip is brought closer to the sample surface it feels a combination of two forces
133
. Initially,
further from the surface, long range attractive forces dominate, pulling the tip toward the surfaces
slightly during each oscillation. Closer to the surface short range repulsive forces overtake this
attraction and push the tip away from the sample. These dynamics effect both the amplitude and
the phase of the tips oscillatory motion and can be utilized along with the known tip and driving
characteristics to gain a good deal of insight regarding the properties of a sample surface. For
50
topography measurements, a setpoint amplitude is chosen that is less than the driving amplitude
sustainable near the surface. As the tip is then lowered until the interactions with the surface
dampen the amplitude to the setpoint value. Now as the tip is traced over the sample and changes
in topography will register as changes in the tip oscillation amplitude. Passing over an object
protruding from the surface will cause further dampening of the amplitude, while passing over and
valley of hole relaxes the restrictions on the tip by the surface causing an increase towards the
driving amplitude. These changes in amplitude are tracked in real time by monitoring the
photodetector signal. This signal is then fed into a feedback PID loop that adjusts tip-to-sample
distance to keep the amplitude constant. Thus, when going over a protrusion, the amplitude does
momentarily decrease, but is quickly corrected for by moving the sample stage away from the tip.
Operations like this continue while moving over the whole sample. By recording and plotting the
various tip-to-sample distance changes one effectively maps out the topography of the sample
surface.
With all these tip/surface interactions going on, one thing to consider is the amount of force we
are applying to the surface during imaging. Even though we are not in constant contact with the
surface, I have found that some clusters are weakly bound enough to some surfaces that tapping
mode settings can sweep them around on the surface. To mitigate this one must attempt to image
in “attractive mode” as opposed to “repulsive mode”. In short, this comes down to choosing a drive
and setpoint amplitudes such that the tip remains in the regime where long range attractive forces
are the dominant tip-surface interactions. That is close enough to feel the effects of variations in
the surface, but not so close as to be repelled by them. This can be easier said than done given all
the interwoven parameters of AFM imaging, and certain sample/tip combination make it
impossible. Repulsive mode is comparatively easier obtain since you can always push harder
51
(higher drive, lower setpoint) to get there, and the interaction is self-correcting since you are
pushing up against something that is pushing back. Tips will wear down quickly, and you can
damage your sample if you go over the top, but reasonable starting points are easy enough to find.
Assuming a balance can be struck, attractive mode is still capable of high resolution with less
forces applied to fragile samples
134
. Given that setting very so much from sample to sample, there
is little use in my citing exact numbers here, but good starting points will be using more flexible
tips with larger radii of curvature (i.e. less sharp), driven at lower amplitudes.
Figure 27 – Contact mode AFM: Schematic representation of an atomic force microscope in contact mode. The
tip is brought into direct contact with the sample surface until interaction forces cause a deflection in the tip cantilever.
A feedback loop is utilized to correct for deviations of the tip deflection with respect to a set-point value while moving
across the sample
128
Another option worth mentioning is “Contact mode”, which is a slightly simpler method to think
about. Here the tip is not driven into oscillations. Instead, it pressed into direct contact with the
52
sample surface which is measured by recording the deflection of the tip as it bends in response
(Figure 27
128
). Considering the tip stiffness with a few calibration measurements it is then possible
to convert this deflection measurement into the force applied in the surface by the tip. By assigning
a deflection (and therefore a force) setpoint, one can use a feedback loop to trace the topography
of the surface in the same manner as tapping mode. I didn’t tend to do this, favoring instead the
less intrusive tapping or non-contact mode for this. I did however use this during nanomanipulation
experiments and to take electrical measurements (with a conductive tip) at various points on
graphene/cluster hybrid surfaces. By setting a conservative deflection so as to not disturb the
surface (or break the tip), one is able to push weakly bound clusters around on the surface in an
effort to organize them into various formations or land with a fixed force at various locations on a
device for comparative electrical measurements (See section Error! Reference source not f
ound.).
3.4 Substrate preparation
Over many attempts to deposit clusters onto various surfaces, it became apparent that the initial
cleanliness of any substrate is of great importance. The two main substrates used in the following
results chapters are Si/SiO2 wafers that are either blank or supporting a monolayer of graphene.
The following two sections detail the experimental procedures I found successful for removing the
majority of surface contamination prior to cluster deposition.
SiO2 cleaning methods
Even though the oxide wafers used for our work are fabricated in a cleanroom environment,
there is nonetheless some level of surface contamination by the time we attempt to use them. Small
amounts of surface junk could be left-over from the fabrication processes, picked up from the
plastic case the sit in, or just deposited from the air as we walk around the lab. Given that moving
53
our whole experiment to a cleanroom is impractical, a (relatively) simple method for cleaning is
needed. To be clear, there will always be some level of surface contamination, especially in our
non-sterile environment. The aim is to develop a process that leads to a high level on confidence
that we can identify clusters on a substrate post deposition and not simply confuse them for a field
a residue (or vice-versa).
The previous procedure for our group has been to apply a steady 15-20 second stream of acetone,
followed by IPA (in a fume hood) to a newly cut piece of oxide wafer. This had been described to
me as a standard practice in most fabrication settings, with maybe the addition or substitution of
methanol after the acetone rinse. This method seemed to work decently well, for basic size
selection testing depositions. Thought there we occasionally areas on the wafer with large amounts
of 10-50 nm contaminates that led to concerns over the accuracy of the quadrupole size selection
(which was a legitimate issue in the past), and questions over potentially less visible remnants
effecting deposition. I began AFM scanning the wafer surfaces prior to depositions and found a
large distribution of surface cleanliness, with some areas being relatively clear and others looking
almost as if there had been a deposition (Figure 28). This level of potential debris is slightly
concerning since the larger collections could be easily confused as clusters. Given the scope of
debris seen, there is also the potential for less visible films of residue (from solvents like acetone
especially) having an effect on the cluster/surface binding.
54
Figure 28 – AFM results of dirty SiO2 surface: Representative AFM topography scans of SiO 2 surfaces prior to
cleaning procedure. Surface contaminants are seen randomly distributed over the surface and vary in size from 5-50
nm in height.
To mitigate this debris experimentation was done with several different solvent washing
procedures and results were compared with spot check AFM imaging. This is by no means an
exhaustive set of trials, and there is definitely some room for streamlining the process. Trialing
different washes and checking each a few times with the AFM can be quite time consuming. This
procedure is the result of conservatively probing the solution space and finding a result that worked
for our uses. Still, it works well, and provides a jumping off point for future adjustments.
It was determined that successive sonication of the silicon oxide wafers produces the best result
with the following procedure below:
30 minutes of sonication in Acetone (repeat twice)
30 minutes in Methanol (repeat twice)
30 minutes in IPA
30 minutes in DI water
55
For each step the small oxide wafer was placed face down in a small test tube. At each transition
the contents of the tube are poured into a small petri dish and the wafer is transferred to the next
tube with tweezers. The wafer can occasionally get stuck on the side of the glass where it can be
washed out with a spray bottle of whatever like solution was used in that step.
The acetone here is used to remove organic contaminants and any oily residues from the oxide
surface. It has the unfortunate downside that it evaporates quickly and can leave some of the
previously dissolved junk behind on the surface. To avoid this, it is advised to remove it from the
petri dish under a flow of fresh acetone to keep it wet (from a spray bottle). Then just be sure to
transfer it promptly to the next test tube. This “transfer under a flow of solvent” type of technique
is also good to mitigate the transfer of contaminates that can collect on the surface of fluids from
the environment, that could then move to the wafer as you pull it up through the fluid/air interface.
This sort of issue is less of a problem in cleanroom settings, but in the lab, it is advised to cover
your glassware to avoid too much dust settling.
The following solvent steps are used to eliminate acetone residue from the surface, which can
be pervasive and stubborn to remove. It is often seen in AFM imaging (examples below) as small
particulates on the wafer surface that I postulate is dissolved contaminates stuck to the surface by
sticky acetone residue that dried out. These solvent steps are the area of the procedure with the
highest potential for optimization, as I essentially found something that worked and committed to
it. Both methanol and IPA are cited as effective at the task in literature and fabrication circles. In
my experience, methanol does a slightly better job in a fixed amount of time. Here I also prefer
the two separate washes at 30 minutes over a single wash at 1 hour. Swapping to a new test-tube
with pure washing solvent helps to further dilute the acetone residue in the mix.
56
The final water step is just used to remove and last specks of broken wafer that chips up from
the sides with sloppy grabbing of the wafer with tweezers. It is very effective at this, and afterwards
samples are dried with a nitrogen gun before being mounted onto AFM pucks.
Figure 29 (left) shows a typical resulting surface after the cleaning procedure, while Figure 29
(right) depicts an uncommon “worst case scenario”. While the surfaces are not always completely
pristine, even in the “worst case” there is a significant decrease in the number of debris spots and
size (now ~2 nm). There are, very occasionally, some completely isolated much larger spots (> 50
nm, not pictured here) observed on the cleaned wafers. This seems to suggest that the smaller
debris spots have collected in the cleaning process. These large irregularly shaped deposits are rare
and less of an issue for our purposes since they are not easily confused for clusters.
Figure 29 – AFM results of clean SiO2 surface: Representative AFM topography scans of SiO 2 surfaces after to
sonication cleaning procedure. (left) A characteristic result showing near total removal of surface contaminates. The
small white dots on the surface are sub 1 nm in height, which could be attributed to peaks in the surface roughness of
the substrate surface. (right) “Worst case” cleaning result which still shows a significant reduction in surface
contaminants size, now ~2 nm for the largest spots.
57
HV heat treatment of graphene
The vast interest in the fabrication of graphene devices has made the cleaning of graphene
surfaces is an extensive area of current research. The need for such work arises because of the
logistics corresponding to the fabrication process. Namely the use of PMMA is used both in
transferring the graphene from the growth substrate to the SiO 2 surface and, during the EBL to
define the surface contacts
19
. PMMA contamination leads to several undesired effects when
considering cluster coating devices surfaces. Firstly, as with SiO2 contamination, the size and
shape of PMMA residue can be very similar to that of the deposited clusters, potentially making
them difficult to distinguish post-deposition. Additionally, the residue island can act as collection
sites for deposited clusters, causing them to collect in bunches around the edges instead of evenly
coating the smooth surface. Lastly, as mentioned in section 3.1.2, large amounts of PMMA residue
on the device surface acts as a charge dopant
116–118
, greatly shifting the charge neutrality point of
the device. Several methods of reducing the amount of residue on have been shown to be successful
including high-current cleaning
117
, mechanical scraping
118,135
(with and AFM tip), chemical
etching
136
, and annealing in various environments
137–139
.
58
Figure 30 – Heating flange: Heating unit used to HV bake graphene samples. Samples are mounted to a copper
heating chuck which is heated by two Watlow cartridge heaters inserted into the chuck body. Thermocouple probes
at either end of the check measure the surface temperature. (Photo credit: John Niman)
Given our access to vacuum chambers and already having had a heating stage, we opted for
high vacuum annealing. Thus, the removal of the PMMA from both the graphene devices and the
CVD graphene substrates was performed in this way. Samples were loaded onto a custom heating
unit (Figure 30) designed by a previous student of the group that is built onto an 8 inch conflat
flange that seals into the deposition chamber (original design by Dr. Malak Khojasteh and Dr.
59
Kresin). This system consists of a copper mounting block with three pre-drilled cylindrical through
holes along its length that match the diameter of Watlow cartridge heaters (#E3A50-E12H), which
are intern wired to separate electrical feedthroughs the flange surface. For the purposes of this
experiment we found that two heating elements provided sufficient power to reach and maintain
the target temperature, so the third was omitted. The copper block itself is attached to a linear
manipulator via internally threaded ceramic cylinders to keep it thermally isolated from the flange
itself. Prior to loading, graphene samples of interest are first adhered to stainless steel AFM pucks
with silver paste and left overnight to cure. These pucks are then mounted to the copper heating
block by way of stainless-steel mounting screws and washers (using the washers to carefully
overlap areas on the AFM puck not obscured by the sample, see image). Upon loading the system
is evacuated overnight and generally reaches an internal pressure on the order of ~10
-9
torr in the
deposition chamber.
Voltage is supplied to the heaters by a Variac DC power supply which is set to ~45 V. The
application of the power is managed by an Omega PID controller which is connected to the heaters
in parallel. After some trial and error with various heating temperatures and baking durations we
found the best results were achieved by heating the system to ~550 C for ~24-30 hrs. I note that
both measurements are fairly approximate. After some tuning, the PID controller was able to hold
a relatively stable temperature +/- 3-5 C, but there are some temperature variations over the copper
chuck due to the heater placement. This is monitored by two different thermocouple sensors (also
wired to the flange) which show approximately 25 C difference from one end of the chuck to the
other. Considering further that these sensors are on the copper surface, and the samples are not
mounted directly, but instead to the AFM pucks, one might expect some further variations at the
sample. As for the timing listed, that is the period from when the heating controller reads 550 C to
60
when it is turned off. This is not an exact measure of the heating time because the system (being
in vacuum) takes several hours to cool back down to room temperature. For our purposes the
exactness of these parameters is not necessarily crucial so long as we can achieve consistent and
reliable results. I mention these factors here only to point out potential for future optimization,
should someone determine it is worth the time investigating it.
Figure 31 – PMMA residue on graphene device channel: AFM topography imaging of the graphene channel at the
step edge (SiO 2 region to the left) prior it HV bake. Devices suffered from a pronounced PMMA ridge along the
channel edge and significant PMMA spotting throughout.
Figure 29 above shows AFM imaging before heat treatment of the graphene devices displaying
the degree of PMMA surface contamination. All graphene devices show a 10-20 nm high ridge of
PMMA at the edge on the graphene and a network of smaller dotted residue dots/smears covering
the graphene surface. The PMMA ridge likely due to an overcuring of the EBL resist used to define
the graphene channel which has hardened it, making it more difficult to remove. The ridge shown
in Figure 29 is less complete than others observed, which will show in section 4.1.2. Upon HV
baking we see the removal of a large proportions of PMMA residue. This is most easily seen in
61
the reduction of the barrier at the graphene channel (Figure 32, left) which now varies from ~2-10
nm in height and is more discontinuous. There is also a significant reduction in the excessive
spotting of PMMA throughout the surface. Figure 32 (right) shows a large-scale image of a more
central region of the channel where large open areas are now visible. What residue remains on the
graphene has collected as small spots along the crack defects. This suggests that the applied heat
causes the PMMA to aggregate prior to desorption, similar to what we observed in the cleaning of
SiO2.
Figure 32 – Graphene channel after HV bake: Post bake AFM imaging showing a reduction in the ridge along the
graphene edge and the opening of large areas in the central channel region.
In addition to the graphene devices we also ran deposition experiments on non-contacted
monolayer graphene/SiO2 wafers. They also have a doped Silicon base layer as always, but I will
refer to them as SiO2 wafers going forward. Here graphene more-or-less covers the entire wafer
with the occasional opening revealing the SiO2 underlayer. Having been fabricated by CVD these
graphene layers were still contaminated with PMMA during the transfer process the deposited
them onto the SiO2 substrate
19
. Figure 33 shows AFM topography images before (top) and after
62
(bottom) HV baking of the surface. Prior to cleaning large 100-150 nm islands of PMMA are
observed randomly deposited over the graphene. After baking, the PMMA islands have been
universally reduced to 60-100 nm in height and find themselves generally located along the crack
defects. This reduction and redistribution of the PMMA again leaves large open areas of clean
graphene.
Figure 33 – HV baking effect on CVD graphene wafers: Representative AFM topography images before (top) and
after (bottom) HV baking resulting in a reduction and redistribution of the surface contamination.
63
In conclusion, we have achieved an improved graphene surface landscape through HV baking.
There is still certainly room for improvement and adjustments could be made to the temperature,
duration, and/or gas environment in an effort to obtain even cleaner results. For our purposes in
the short term (and considering the time restrictions of the last year), the open areas of graphene
seen here were deemed sufficient for morphological investigations of cluster/graphene
depositions.
3.5 Gas Flow Chamber for ionized gas source
Experiments investigating the interaction of surface adsorbed ionic molecules with suspended
SWCNT’s were conducted in a custom gas flow chamber that was originally designed by Adam
Bushmaker of the Aerospace corporation. I made some minor structural modifications to the pre-
existing apparatus to re-assemble it into working order after it had been out of use for some time.
I added additional layers of functionality and automation to the existing operating interface to
allow for fully automated temperature dependent data collection, along with seamless recording
of long period time series data to capture extended ionic interaction events.
64
Figure 34 – Schematic of chamber for CNT gaseous ion interaction experiments: (left) Computer generated
model of the sealed gas chamber with device socket, automated TEC for temperature dependent measurements, and
90
Sr source mounting electrode. (right) A Photograph of wired, surface mounted device with accompanying RTD.
Above is a schematic representation of the experimental set-up which has been described
previously
58
. During measurements devices are housed in a custom rectangular gas flow chamber
(Figure 34, left). Along one side (as oriented above, during operation it is rotated 90 deg) is a
Swagelok connection that routes to an inlet valve which opens the chamber to gas flow from a
muli-line gas manifold up stream. This manifold (not pictured) is essentially a mechanical
switchboard the provides seamless selection and exchanging of different gasses at any given time.
On the remaining sides there is a gas outlet that is terminated with a 1 PSI one-way valve (to
prevent backflow) and a KF fitting connection that runs to a small Pfeiffer (HiCube) turbo pump
for chamber evacuation. The chamber is sealed on the top face with a custom KF flange that
supports a small electrode upon which a 100 uCi
90
Sr radioactive source is mounted. High energy
beta decay electrons from this source are used to ionize the volume of gas inside the chamber.
65
Positive gaseous ions are then driven towards the device mount by a voltage applied to the flange-
mounted electrode, in a drift chamber setup. The distance from the emitting face of the
90
Sr source
to the CNT sample was approximately 7.4 mm. Drift chamber ion current versus electrode plate
voltage for the
90
Sr source shows saturation (majority of generated ions swept out of chamber
before recombination) at around 4 V. The back face is routed with an O-ring groove that is sealed
against a custom printed circuit board that was shown to be capable of holding vacuum down to
<1 mTorr (gauge limited measurement). This circuit board routes SMB socket connections,
through grounding switches, to a ceramic chip carrier socket that that receives the wire bonded
CNT-FET packages (Figure 34, right). The central region of this socket has been routed out and
filled with a copper thermal chuck/heat pipe that makes good thermal contact with the chip package
by way of small aluminum clamps screwed to the socket face. This copper chuck is then connected
to a PID controlled thermoelectric hot/cold plate (Newport 3040 TEC) allowing for device
temperature variations from roughly -10 to 72° C. The unit itself is capable of a broader range of
temperatures but those listed are measured at the surface level of the chip by a co-mounted
resistance temperature detector (RTD).
For all current sampling experiments the devices were operated near the subthreshold regime by
applying 100 mV source-drain bias (𝑉 𝑠𝑑
) while holding the gate electrode (𝑉 𝑔 ) at a fixed value.
This fixed value is just ground for the result discussed in the following chapter 5 but was alternated
between +/- 0.25 V so a small segment of data discussed in chapter 6. Source-drain current was
then recorded in the presence of gaseous ions. During the first round of thermal desiccation
experiments this was done using an Agilent 4156C semiconductor parameter analyzer with a
sampling frequency 250 Hz in 40 second intervals. For the gas and vacuum desiccation trials, along
with later thermal trials (producing the fitted data in chapter 5) the bias were held constant and
66
data was recorded continuously by running the device current through a Stanford Research
Systems low-noise preamplifier (SR570) before sending it to a Tektronix DPO 3034 oscilloscope
with a sampling frequency of 5,000 Hz. The memory buffer of the Oscilloscope was then
periodically captured at regular intervals and an overlap cutting algorithm was applied in real time
to construct an interrupted live stream of the data. This equipment switch was made to better
capture data containing longer device switching events that were discovered to occur frequently
after device desiccation (discussed in later results). Both pieces of equipment above, along with
the thermoelectric cooler, were integrated into a Labview VI which allows for programmable and
fully automated data collection. After collection, data was analyzed via a Python script to identify
the time and duration of each large switching event.
Argon gas used in this system was ultra-high purity (UHP, 99.999%) and dry nitrogen gas used
was harvested by the evaporation of UHP liquid nitrogen (𝐿𝑁
2
) stored on-site. Thus, we consider
the lower bound on the evaporated nitrogen to be at UHP level.
67
4 Clusters deposition on carbon based devices
4.1 Selective deposition of ionic Ag nanoclusters on graphene
Quick note for the thesis: I will start will the chronological ordering of the events here. This is
certainly too winding for a paper and will absolutely be adjusted going forward to trim it down
into something more presentable. For now, it gets this section written up with some additional
context and I think that it might actually be more instructive to a future student to see the whole
process of observing a feature in data and iterating on an experiment to isolate the effect.
Again, here I refer to the Si/SiO2 wafers used in this study as simply SiO2 wafers, since we are
once concerned at the moment with the substrate that makes contact to the graphene and clusters.
In the future one could still back gate the wafers through the doped Si backing layer.
Introduction
In this section I follow the (roughly) chronological story of our investigation into the selective
adsorption of Ag nanoclusters onto graphene devices. It began with a preliminary deposition on
Graphene field effect transistors (FETs) I made in collaboration with Dr. Sean Stuart at the
Aerospace Corporation. Upon cluster deposition it was noted that there appeared to by a heavier
coverage on the graphene channel itself when compared to the surrounding SiO2 substrate. This
data was slightly muddied by the presence of fabrication debris throughout the surface, though did
appear to be some correlation between local debris sites and collections of clusters on the SiO2.
This led to a series of experiments aimed at achieving very clean graphene and SiO2 surfaces for
comparative studies of Ag cluster deposition in the hopes of identifying the source of the coverage
difference on the two surfaces.
68
The original intent of the Ag cluster depositions on graphene was multi-faceted. First,
exploratory depositions such as these were hoped to allow us to investigate how stable clusters
were on the graphene surface, and if they would uncontrollably aggregate or coalesce together. If
this was successful, a secondary goal was to deposit a dilute cluster coverage to experiment with
nanomanipulation techniques available of the AFM. These could allow clusters to be pushed
around into a linear array starting on an electrically active graphene piece (surface contacted to
allow for biasing) and running off onto an insulating surface. If successful, I would then be able
to use electrical AFM techniques to pass current through the cluster chain in the hopes of
observing tunneling current for varying cluster sizes and chain lengths. This secondary objective
was the reasoning for the selection of Ag as our sputtering material since it is conductive,
surprisingly inexpensive (by comparison to gold), and potentially non-oxidizing upon exposure
to air (though there is some debate here). After our initial depositions, I became a little side-
tracked investigating the disparity in cluster coverage that resulted in the work presented here.
Preliminary nanomanipulations has been conducted though and that is an ongoing project.
Finally, all of this is in the context of eventually studying the effects of cluster dopants on
graphene devices. Both to see the effect on device characteristics, and to use graphene FETs as
systems to better understand the properties of surface supported size-selected clusters. As
covered below, the presence of surface debris made electrical measurements on the initial run of
fabricated devices impossible, but considerations have already been made for future devices to
avoid these issues.
Initial imaging of Silver on “as fabricated” graphene device
After fabricating the first round of Graphene devices (section 3.1.2) a trial deposition was
preformed depositing approximately 20% coverage of ~10nm Silver clusters onto one array of
69
graphene FET’s. Unless otherwise mentioned, all imaging was performed on a single central
device on the die as it was the only one that was held at electrical ground during deposition. AFM
imaging of the graphene surface shows a modest coverage of clusters decorating the graphene
surface (Figure 35). It was discovered after deposition that Silver at 10 nm diameter is nearing the
practical upper screening limit for the quadrupole of our system, thus explaining the resulting
cluster distribution of 10-20 nm. This preliminary investigation was not conducted at the highest
resolution possible for the quadrupole and successful size selection at more favorable mass ranges
in dilute coverages were achieved to confirm our capability (shown later).
70
Figure 35 – Ag clusters on graphene device with surface contamination: AFM imaging of dilute Ag cluster
deposition along the graphene device edge. A large amount of PMMA decontamination is observed on the sample
surface along with an unexpected cluster coverage difference.
There are several other features of note in the imaging. Of particular interest at the time was the
presences of a film-like debris field spread over both the graphene channel and the SiO2 substrate
that was discussed in section 3.4.2. This substance has been identified as PMMA residue left
behind in the fabrication process with signatures displayed in the large heighted strip along the
edge of the graphene, the grey pools on both the graphene and SiO2, as well as the circular “halos”
on the graphene surface. This level of residue advocates for some potential reworking of the
71
fabrication process (more washing steps, adjustments the PMMA curing temperature/time, etc.),
which will be revisited in future fabrication iterations. For now, we point out the impetus for this
chapter, which is the dichotomy between the cluster coverages on SiO2 and graphene areas of the
sample. This is an extremely unexpected result given our previous cluster deposition experiments.
In the past, we had been successful at depositing copper and aluminum clusters on both SiO2
wafers and CVD graphene (CVD-G) samples separately. The differences on this occasion is that
we have them on the sample surface with clear delineation of the two substrates for comparison.
While there are usually SiO2 “pits” in CVD-G where the graphene is not continuous, prior to now
these areas we not deeply investigated due to our focus being the support of clusters on graphene
surfaces.
One candidate explanation of the phenomena was thought to be thermal surface diffusion of the
clusters from the SiO2 to the Graphene channel post deposition. Another being that there could be
some electrostatic interaction (inflight or after deposition) that attracts the ionized clusters from
the insulating oxide over to the grounded graphene. We set out to perform follow up depositions
on both CVD-G and SiO2 wafers with new cleaning methods to isolate these potential mechanisms,
and determine the role (if any) of surface contamination (sections 3.4.1 and 3.4.2).
Graphene vs. SiO2 at various cluster coverages and beam conditions
The first trials involving the cleaned substrates consisted of what was called a “mixed sample”
wherein a small wafer of CVD-G was mounted next to one of SiO2 on the same sample puck and
exposed to the cluster beam path simultaneously. This was done to test the possibility of surface
migration of the clusters post deposition. With a small gap between the wafers it would be
impossible for a cluster to run itself off the side of the SiO2 wafer and onto the graphene. These
samples were tested in three different experiments each of which focused on depositing 6-8nm Ag
72
clusters in various coverage regimes and gas flow conditions. Coverages varied from 0.05 to ~0.2
ML at fixed gas flows (70//30 Sccm Argon to Helium) and roughly 15% coverage at multiple
Argon/Helium ratios (140/70, 140/140, 70/70, 70/140 and 70/10). These samples we also placed
in at fixed beam conditions in reverse orientations to rule out any systematic beam inhomogeneity.
In every trial the coverage of clusters deposited onto the graphene significantly outnumbered that
of the SiO2, in agreement with the above deposition on the graphene device. Figure 36 contains
comparison imaging of the Graphene (top) and SiO2 (bottom) at the same deposition conditions.
Here the scan width of the image is fixed (5 µm) to illustrate the large difference in area coverage.
73
Figure 36 – Mixed sample cluster coverage results: AFM imaging comparison of 8 nm Ag cluster deposition on
graphene (top) and SiO 2 (bottom) at fixed scan width depicting the many-fold increase in cluster deposition on the
graphene samples.
With these results in mind, we can likely rule out a purely diffusion-based mechanism for the
lack of surface adsorption of the clusters on the SiO2 surface. Without any graphene to diffuse
towards, one would expect to see some form of deposition on the oxide wafers. Even if we allow
for diffusion off the SiO2 wafer toward the edges, one expects to see a collection of clusters near
the boundaries that would get trapped in the rough edges of the chip. Not only are theses not
74
present, but major evacuation of clusters from the central area on the wafer would require some
~3 mm of diffusion. This seems unrealistically high at room temperature, even for a very weak
interaction energy.
SEM imaging of silver cluster coated devices
Obtaining AFM images of the above dilute coverages of clusters was initially found to be quite
difficult due to the apparent ease of disturbing clusters on the surface with the AFM tip. Initial
scans performed with similar system settings used in the past showed images of largely clean
graphene surfaces, followed by frequent tip damage and image streaking. This led to a (justified)
concern that the initial repulsive mode scan settings were pushing clusters around and “sweeping”
the area clean. To mitigate this, work was done to find new system settings that would keep the
tip/surface interaction in “attractive mode” (section 3.3) where the forces felt by the surface by the
tip are much lower. By greatly reducing the scanning area, tip speed, and driving amplitude the
successful imaging of clusters on both the baked graphene and cleaned SiO2 was achieved. Even
still there was some concern that these measures could be insufficient (for the SiO2 surface in
particular) and supposed “effect” was simply sampling bias created by our surface measurement
technique. To rule this out, additional surface imaging was conducted by scanning electron
microscopy (SEM) in collaboration with Professor Levenson-Falk’s group at USC. Below are a
series of SEM images of both the graphene and SiO2 surfaces which confirm the differences in
cluster coverages depicted in our AFM images (SEM operation credit to James T. Farmer).
The first result of the SEM imaging is the confirmation that indeed few clusters are present on
the SiO2 substrates, and their absence is not simply the result of sweeping most clusters away with
the AFM tip. The second qualitative note is that the areas of the SiO2 substrate that do show clusters
appear to show them surrounded by some surface contaminate film which is visible as a film like
75
constrast around the dark cluster shperes in all four panals of Figure 37. This could be acetone
residue that survived the cleaning process or some oil contamination picked up in the vacuum
system of the deposition chamber. In either case, the correlcation between this and the clusters on
the surfaces suggests that these surface contaminates trap the incomming clusters, that would
otherwise not settle on the clean SiO2 surface.
Figure 37 – SEM imaging of SiO2 wafer: Exploratory SEM characterization of the SiO 2 surface after Ag cluster
deposition. Images confirm the reduction in cluster coverage is not an artifact of AFM scanning. A film-like
contamination is observed surrounding all deposited clusters.
As a sanity check, the adjoining graphene samples were also examined (Figure 38) and were
confirmed to show greater cluster coverage as expected. Also visible are occasional gaps in the
graphene (crack defects or empty pits) which exposes the underlying SiO2. These patches do not
show any sign of either surface contamination or cluster coverage. These empty areas serve as
76
futher evidence that the conamination is a key to the deposition of the clusters on the SiO2. It also
confirms that the HV baking of the SiO2 subtrate (on the CVD-G) produces a cleaner result than
the repeated sonication technique used for blank SiO2 wafers. The fact that the baking is superior
to sonication is not particularly suprising, but the confirmation that the baking produces this clean
of a surface is useful information.
The opposite contrast of the clusters on SiO2 (dark spots) relative to the graphene (light spots),
is likely due to charging if the clusters on the surface by the incomming electron beam. Recalling
that the data being collected from the sample during SEM imaging is secondary electron emmision,
the clusters on graphene have their knocked off electrons replensished by the conducting substrate
while those on the insulsulating graphene do not causing them to show as dark spots, void of signal.
Figure 38 – SEM imaging of graphene wafer: Exploratory SEM characterization of the graphene surface after Ag
cluster deposition. Ag clusters are observed on the graphene surface as small white dots and a complete absence of
deposition is noted along the exposed SiO 2 underlayer.
Also imaged via the SEM was the first “as processed” graphene deivce that received cluster
deposition (Figure 39). Here we see a confirmation of a higher coverage of clusters on the graphene
channel, and that the presence of clusters on the SiO2 regieons is again joined by patches of image
contrast consistent with a thin layer of contaminate. In this scenario the surface contaminate is
77
almost certainly reminant PMMA that we also observe in the AFM imaging (Figure 35). On the
device, the distribution of clusters on the SiO2 surfaces apears to vary from one spot to another.
Higher proportions in cramped areas nearing the junction between the gold electrical contacts and
the graphene channel (Figure 39, bottom left panel) where stubborn PMMA can collect over many
lithographic steps and is difficult to desolve.
Figure 39 – SEM imaging of graphene device: A series of SEM images of the graphene device which received ~10
nm cluster deposition. In agreement with the SiO 2 imaging we see the clusters joined by film-like contamination. The
distribution of cluster deposition on the oxide area of the device also appears to correspond to tight areas where PMMA
would be more difficult to remove.
With the these imaging results, it is now clear that the morphelogical nature of the SiO2 surface
plays an important roll on its ability to accept cluster dopants and that the cleanliness of the surface
appears to be a key component. Moving forward we focus only on the CVD-G and graphene device
78
samples as we have confirmed their cleaning method (HV baking) to be superior to wafer
sonication. With a purely surface diffusion mechanism ruled out, it is now best to consider the
SiO2 areas of these graphene substrates as the pose a direct comparrison, having both been baked.
Graphene Islands and Mica comparison
Following up in the above results we attempted to push the envelope with slightly higher cluster
coverages and a focus on observing the cluster deposition on and around the SiO2 “pits” of the
CVD-G substrate. Yet again, in all cases we see a drastically higher coverage on the graphene
portions of the substrate. Moreover, the increase in coverage continues to appear randomly
distributed over the graphene surface (Figure 40). Another piece of information against a purely
diffusion-based effect is a lack of any cluster aggregation along the step edge of the graphene, as
is often seen in literature with metal clusters on HOPG
140,141
. It would be expected that a mobile
cluster would become preferentially trapped along the interface between the two materials where
it would be energetically costly to move onto the graphene surface. Instead we see what appears
to be an even coverage on the graphene regions of the substrate.
79
Figure 40 – Increased Ag coverage on graphene: Graphene regions continue to exhibit greatly increased and even
deposition and higher coverages. There is no evidence of aggregation of clusters along the graphene step edge, a
hallmark of surface diffusion.
Figure 41 and Figure 42 shows an interesting and consistent feature throughout the data. Here,
even completely isolated patches of graphene continue to show the preferred cluster coverage.
Given that the incoming clusters are negatively charged, a possible explanation of the deposition
preference could have been an electrostatic attraction (either in flight or on surface) towards the
grounded graphene areas. This could have been from either direct attraction to the grounded areas,
or more likely the charging up of the insulating oxide regions by exposure to the ion beam, which
contains not only the cluster ion, but some percentage of charged inert gas species. These isolated
patches are disconnected from the main sheet of graphene on the wafer surface (which is grounded
to the AFM puck via silver paste) making them electrically isolated. Thus, any deposition on these
pieces would accumulate charge, and presumably set up a barrier to further cluster deposition,
same as any insulator. At heavier coverages we would then expect less cluster deposition on these
islands. On the contrary, upon comparison to other more contiguous graphene regions we find an
80
even coating of clusters throughout. Leading us to conclude that an electrostatically dominated
mechanism is unlikely.
Figure 41 – Electrically isolated cluster doped graphene: Clusters can be seen on completely electrically isolated
pieces of graphene. This suggest that a surface charging electrostatic mechanism is not the dominant factor in the
selective deposition.
81
Figure 42 – Graphene/Cluster archipelago: A large scale example of the above figure. Here a large patch of the
graphene sheet has been degraded during transfer and formed a scattered collection of small graphene islands which
still support clusters selectively.
This case is made a second (and perhaps more convincing) way by depositing the very same
cluster beam onto a Mica substrate. Here a disc of Mica was cleaved in air and mounted adjacent
to a wafer of CVD-G in the same manner as the “mixed” samples mentioned earlier. Much like
that of the graphene substrate, the mica surface possesses an organized crystal structure and
provides a very flat surface upon which to deposit clusters
142,143
. It is different in that, like the SiO2,
it is electrically insulating. Upon imaging we see agreement in coverage between the mica and the
jointly mounted graphene substrates (Figure 43), surface charging and its following
electrostatically dominated mechanisms are now considered unrealistic.
82
Figure 43 – Ag decorated Mica: Insulating Mica samples co-mounted with CVD-G show similar cluster coverages
ruling out electrostatically dominated mechanisms.
High coverage limit on graphene devices: Potential mechanism and future supporting
simulations
Armed with the results of the previous trials, it was now time to test the limits of the observed
effect while isolating the last obvious component of the experimental procedures, namely the
cleaning of the oxide surface via vacuum baking. To test this, ~7.5 nm Ag clusters were deposited
onto two pairs of graphene device substrates at 3 ML and 6 ML coverages. For each coverage one
device was left “as processed” while the other was first HV baked to remove PMMA residue. After
deposition, samples were taken to the AFM where the graphene edge of the channel was imaged
to show a direct comparison in cluster coverage. As mentioned previously, the first round of
fabricated devices possessed a large shift in the charge neutrality point, making them not useful
for measuring surface charge doping. With that aside, they are still electrically active, and the
defined graphene channel allows for significantly easier locating of a graphene/SiO 2 interface for
deposition comparison imaging (when compared to hunting around for pits on a CVD-G wafer).
83
Figure 44 – Heavy cluster deposition: Approximately 6 ML deposition of Ag clusters onto graphene devices as a
function of post fabrication cleaning treatment. The deposition displays an extremely high level of surface selectivity
of the cluster dopants on devices subjected to HV baking.
Figure 44 summarizes the results for the 6 ML deposition, which is qualitatively the same as the
3 ML data, only with more clusters. Displayed clearly in the imaging is stark contrast in surface
coverage as a function of pre-treatment. In the “as processed” device we see the Ag deposition is
continuous throughout, nearly to the point of obscuring the graphene step edge. In contrast, the
cleaned device shows a heavy coverage of Ag on the graphene channel, with very little clusters on
the surrounding oxide.
Discussion and future work
Taken as a whole, our results point to the treatment of the SiO2 surface prior to deposition as the
dominant factor in determining the successful adsorption of silver clusters. Having investigated
and ruled out the potential for diffusion or electrostatic dominated mechanisms, we now consider
the physical details of the surface landscape. These results suggest that the SiO2 itself is inherently
resistant to the adsorption of Ag clusters. Poor adhesion, though not this extreme, has been
84
observed in atomic deposition of silver thin films onto oxide surfaces
144–147
. Though the exact
chemical interaction if poorly understood, it is relatively common practical knowledge among
those in the device fabrication industry that when depositing Ag or Au onto oxide wafers (usually
to make electrical contacts) it is necessary to first deposit a thin “adhesion layer”. This is a metal
that better sticks to the oxide substrate (Cr or Ti) and serves as an intermediary, supporting further
deposition of the Ag (or Au)
144
. Without an adhesion layer in place, silver thin films degrade in
ambient conditions and delaminate from the surface completely
145
.
While it may not be permanent, atomic scale deposition is at least able to temporarily coat the
oxide surface in a way that cluster deposition is not. This leads us to believe that the effect we
observe is due to the size of the incoming cluster as it relates to the surface landscape. The
amorphous surface of SiO2 has been well studied both directly and in simulation settings and has
been found to display a surface roughness of ~1-1.5 nm. This surface roughness is well illustrated
in simulations conducted by Professor Alexandrova’s group at UCLA (Figure 45)
148
. Here it is
shown that the oxide surface consists of a series of random pits and valleys on the nanometer scale.
For everyday purposes, this is quite flat, but as a dopant size approaches the order of the surface
morphology this can greatly affect the adhesion.
85
Figure 45 – Amorphous SiO2 surface: MD simulation from Zandkarimi et. al.
148
displaying the surface landscape
of amorphous SiO 2 which is decorated with a multitude of ~1-1.5 nm pits.
We propose that when incoming particles are smaller than that of the surface roughness, they
can fall into the nooks and crannies on the surface filling them in and initially forming a film on
across the surface. As is the case with atomic silver deposition. As time passes, this film coarsens
and forms large silver domains on the surface which display low adhesion to the SiO2 surface and
delaminate
145
. This is supported in two ways; first MD simulations have been conducted that show
a high delamination energy at the interface between SiO2 and noble metal film layers
147
. Secondly,
the result of various silver deposition coverages on SiO2 surfaces has been investigated with SEM
results showing a poor wettability of silver onto SiO2 substrates
146
. Here high levels of atomic
deposition >1ML do not form a contiguous film on the SiO2, but instead suffers from silver
aggregation and coalescence on the surface forming a distribution of islands at temperatures much
lower than the melting point of silver.
The incoming silver nanoclusters are slightly larger than the surface features (6-10 nm). Due to
their spherical shape, they already possess little contact surface area when lying on an ideally flat
86
surface. The ridged nature of the SiO2 surface exacerbates this leading to very low adsorption. If
there are not some additional features to the surface landscape, like PMMA contamination, then
they desorb. The rare clusters that do survive on the SiO2 could be energetic outliers whose
incoming velocity was in the high end of the distribution
149
, causing some deformation and stick
to the surface. Alternatively, there could exist surface extraneous surface defects in the oxide
surface that arise from ion bombardment during the graphene channel etching. Argon sputtering
of SiO2 has been shown to reduce the aggregation of atomic silver deposition previously
146
.
With the methods shown above, we have successfully coated a graphene device channel with
Ag, showing extremely high selectivity. This selectivity is of particular importance when
comparing the desorption mechanisms observed between atomic and cluster depositions. Thick
atomic films used to coat devices as seen here would cover the entire surface at first, but with high
risk of lifting off the graphene surface as the contiguous film delaminates for the surrounding SiO2.
To avoid this, depositions would need to occur along with some form of lithography patterning.
This is not only a time-consuming process, but as seen in our data and elsewhere, leaves behind
residues that effect the electrical nature of the graphene. With our method, the device is cleaned
and characterized, then deposition on the two surfaces are decoupled leaving the Ag to freely coat
the graphene channel, without any need for additional processing steps. This cluster deposition is
also a valuable tool for the fabrication of Ag/graphene hybrids on wafers containing many devices
in close proximity. We have shown the capability of heavily coating a graphene surface without
any risk of the coating layer electrically shorting neighboring devices. While not investigated for
clusters yet, similar poor adhesion to SiO2 has been reported for both Au and Cu as well
144
. We
have also found one other work that has sought to embrace this adhesion deficiency to fabricate
Cu multi-level interconnects via atomic deposition and a novel method involving SiO2 layer
87
growth between metal deposition steps
147
. The above results open the door for follow-up
experiments with these and other materials to determine the limits of the effect and potential cluster
doped graphene complexes which are of growing interest, particularly involving Ag
150
.
Work must also be done going forward to optimize the device fabrication and HV baking such
that electrical measurements are made practical. Without the spurious charge doping contributed
by the PMMA residues one could use the alterations of the charge neutrality point of the device
after deposition to investigate the charge donation of the cluster dopants. A series of experiments
as a function of cluster material, size, and coverage could then be undertaken to characterize atomic
nanoclusters as highly controllable dopants for tailoring custom graphene device characteristics.
88
4.2 Clusters on CNT devices
Introduction
Below I present some preliminary results on the functionalization of CNT’s with metallic
nanoclusters. This work attempts to investigate the ability of a CNT to support a cluster deposited
from the gas phase. With the eventual goal of assembling cluster arrays, we consider the CNT
channel of a suspended CNT-FET as a potential stencil for cluster deposition. Heavy cluster
deposition could then result in a line of size selected clusters along the conduction channel,
conveniently integrated in the deceive and ready for electrical transport measurements. There has
been previous work in similar systems (mentioned in section 2.3) by both Dai and Yap groups,
investigating atomic deposition on CNTs and cluster deposition on (much wider) HBN nanotubes
respectively. While this work has been encouraging neither of these depositions occurred on
functioning devices with an eye on eventual transport measurements. In comparison to Dai’s work
we now consider cluster deposition. If successful, we open the door to many permutations of
cluster sizes and materials to investigate cluster properties. Yap’s work is encouraging to this end,
suggesting that, if integrated into a device, clusters could exhibit tunneling conduction at room
temperature. Due to lab closures (leading to a device shortage) we have not be able to obtain
electrical conduction measurements yet, but I believe we have laid the groundwork for such
measurements soon. Below we show the capability of landing clusters onto a suspended CNT
devices. This is a crucial first step, that was not necessarily a given result considering most clusters
are near or larger than the CNT diameter. With the energies involved in deposition, there was
always some concern that a cluster incident on a suspended CNT might break upon impact or
deflect in lieu of adhering to the surface.
89
The importance of CNT suspension
The main feature of importance in the devices made for this work (section 3.1.1) is the
suspension of the CNT over an open trench. This is critical to allow for enough cluster deposition
to potentially decorate the CNT itself (a very small target), without shorting the device out
completely. While our preliminary work has focused on simply decorating the tube surface, any
plan to do so that eliminates the functionality of a device is a non-starter. This issue is well
illustrated by depositions performed on surface supported devices that were generously given to
our group by Phil Collins at UC Irvine (Zot! Zot! Zot!). In early experiments aiming at dilute
cluster coverages (Figure 46, Al clusters) one can clearly resolve the nanotube with AFM imaging
and see that there are several clusters along the CNT length, where it meets the SiO2 surface. There
are even a few occasions were a cluster sits on top of the tube, without falling off to the side,
suggesting that supporting a series of clusters along a device is potentially viable.
90
Figure 46 – Dilute Cluster coverage on surface supported CNTs: Before and after AFM images showing the
successful size selected deposition of Al nanoclusters on CNT's
At higher coverages an issue become apparent via AFM imaging. First there is that imaging the
CNT is now very difficult as finding it among a dense field of clusters is problematic in even mild
depositions and impossible at 1ML+ coverages. Secondly, and most importantly, very heavy
cluster coverages nearly guarantee the shorting of the source and drain contacts with potentially
many branching conduction pathways. Given that the goal of the project is to take transport
measurements of a single orderly array/chain of clusters, this is result is not desirable.
91
Figure 47 – High cluster coverage on surface supported CNTs: An example of heavy coverage resolution
difficulties. It is possible, but very time consuming to resolve the CNT after deposition.
This experience led to the push toward obtaining suspended CNT-FET’s which was eventually
successful via a collaboration with the Aerospace corporation and Professor Steve Cronin’s group
at USC. The advent of suspending the tube over an open trench acts as a sort of stencil for higher
tube coatings. Excess clusters fall into the deep trench etched into the oxide supporting substrate
away from the level of the tube and contacts. This leaves the CNT both exposed and easy to find,
while also maintaining the electrical integrity of the device.
Imaging considerations when planning experiments
One small note on the difference in imaging of the substrate supported devices vs. suspended
devices. Given the small diameter of the CNT and the depth of the trench, AFM imaging is not
practical. This is because the tip itself has great difficulty tracking the height difference when it
approaches the CNT, preferring instead to reach down into the trench. This leads to the tip crashing
into the CNT and dragging it around (in my limited, though sadly direct experience). It may be
possible that there exists a magic combination of small tip radius, slow scanning speed, high
92
feedback, and a good deal of luck that could lead to nice AFM resolution of these devices. On the
other hand, that combination could take a very long time to find, you would certainly break many
tips/devices, and the scanning electron microscope (SEM) exists. Assuming one can schedule time
on the SEM, it is the faster and easier option. As with everything, there is a drawback to this. There
is a reaction with the electron beam of an SEM with organic residue often coating samples and in
vacuum systems
30
. This reaction results in the deposition of amorphous carbon essentially
everywhere you strike the with the beam. This deposition has adverse effects on the conduction of
CNTs. Thus, if you image a device, it is no longer useful for transport measurements. This can
obviously be mitigated once a reliable process for CNT coating has been established and there is
confidence that a cluster chain has been formed. Simply get all the transport measurements done
first, and image after. For now, we chose to image first due to the risk of device breakage during
electrical measurements. Due to the effort involved in fabricating the devices and depositing the
clusters, we wished to guarantee that we got some information from the early experiments.
Thesis note for future student: Surface mounted devices do work quite well for these dilute
coverages. I would actually go as far as to say they are preferred, due to their ease of AFM imaging
with less risk of device damage. We are still early in the process of investigating cluster/CNT
interactions. If for some application it is found that a very low doping of clusters on a device is all
that is needed, and the interaction with the underlying substrate in not important, then AFM
imaging is the better option. Maybe when trialing out a new material.
Decorating suspended nanotubes with Ag clusters
In this instance, two sample substrates each containing suspended CNT’s were exposed to Ag
cluster deposition of two different sizes, ~5 nm and ~3.2 nm. After deposition, samples were
removed from the vacuum chamber and imaged using an SEM located at the Aerospace
93
Corporation in El Segundo with the assistance of Dr. Sean Stuart. Exemplary images showing
clusters decorating the suspended CNT’s of the two samples are contained in Figure 48 and Figure
49.
Figure 48 – 3nm cluster deposition on suspended CNT: Suspended CNT subjected to 3nm cluster deposition.
Clusters appear to have coalesced on the CNT surface, coming together to form larger irregular particles ~15-20 nm
in diameter.
Considering first the sample subjected to ~3.2 nm clusters in Figure 48. We see that there are
small collections of deposited metal on the CNT surface. By eye, the collections of metal appear
to vary slightly in size and shape. From the scale bar on the SEM these deposits are approximately
15-20 nm wide which is significantly larger than the selected ~ 3nm. The non-uniformity of cluster
shape, along with the increase in size suggests that there is cluster aggregation and coalescence on
the CNT surface post deposition, in agreement with the result of atomic deposition on CNTs
reported by the Dai group
97,98
(covered further below). This idea is further supported by comparing
the clusters on the CNT to those in the background of the image, which decorate the bottom of the
trench. While the lack of focus makes it difficult to determine their exact dimensions, they are
uniformly smaller than the deposits on the CNT itself. Being less mobile on the comparatively
94
rough Pt gate electrode surface, these clusters have not been able to join forces to form larger
deposits.
Figure 49 – 5nm cluster deposition on suspended CNT: Suspended CNT subjected to 5 nm cluster deposition.
Here we find the clusters (~7.5-10 nm) are closer to the intended sizes. This result suggests a lower surface mobility
of larger incoming clusters on the CNT surface.
In some contrast, the device exposed the 5 nm deposition appears to show smaller resulting
clusters of roughly 7.5-10.5 nm (as small as 6.6nm) which are more regularly decorating the tube
surface. This is a curious result that suggests that the initial dopant size can have a great effect on
the result of devices decoration. Here, it appears that the larger clusters are less mobile on the CNT
surface and aggregate less, leading to higher tube coverages.
95
Discussion and future work
As mentioned above, this result agrees qualitatively with work done by the Hongjie Dai group
up at Stanford university studying atomic deposition on CNT’s. Here the CNT’s in question were
grown onto TEM grids which consists a metallic mesh grid, generally backed by a lacey carbon
layer. CNT’s grown on this grid were suspended over the holes in the mesh, allowing for atomic
metal deposition (via EBPVD) without obscuring the CNT’s themselves. Upon imaging,
researchers found that the atomic deposition of certain metals collected, forming self-assembled
islands on the tube surface (Figure 50, left). This coalescence implies weak bonding between the
deposited metal and the CNT surface. This is fortunate news for isolated clusters, to a point. While
it is good to know that a cluster will not break up, and solder itself to the CNT, one would like the
ability to deposit smaller clusters without worrying about them merging in the surface. For a next
step we can consider further work by Dai’s group which showed that the deposition of a ~1 nm
thick Ti film onto the CNTs prior to atomic metal deposition helped to inhibit aggregation,
allowing for relatively smooth coatings (Figure 50, right). This is an interesting result and a
potential option for us going forward that could allow for smaller cluster deposition without
aggregation on the surface.
96
Figure 50 – Dai Group atomic deposition on CNTs: A comparison of atomic metal deposition for various
materials onto suspended CNTs without (with) and with (right) a 1 nm thick Ti pretreatment layer. The addition of Ti
layer limits aggregation along the CNT surface, which could prove vital for the deposition of smaller nanoclusters.
Now, there is very obviously still a lot of work to be done to investigate this further. Not least
of which is simply a follow-up experiments with similar conditions to show reproducibility. After
that I would love to expand both the deposited clusters material and size to further probe the surface
interactions. Our other experiences have shown that Ag doesn’t seem to like adsorbing to surfaces
in general, so it is possible that other metals clusters might not suffer as much surface aggregation,
allowing us to deposit smaller sizes. If not, then we can investigate the addition of a Ti pre-
treatment layer. We were limited early in the process by both the number of functioning devices
to deposit on and the number of samples we could load into the chamber at one time. These
problems have been mitigated now going forward. Firstly, our group has been given access and
training to fabricate the devices through the collaboration with Professor Cronin’s group
(something I hope will continue in my absence) and I have made a small stock of them for future
97
depositions. Secondly, I have redesigned a simple substrate holder with a larger mounting area,
which now accommodates ~7 samples (depending on the wafer size) that will allow for more
efficient deposition runs.
Once higher coverages of clusters are achieved on the on a CNT with an appropriate coverage
to facilitate cluster-cluster conduction I envision transport measurements becoming the priority
over any sort of imaging (this might need to be with larger clusters at first, due to the surface
aggregation). In this scenario we could deposit on either or semiconducting or metallic CNT
devices. Use of semi-conducting tube which could be gated into their “off-state” could allow for
the study of the current carrying capacity of clusters chains under various positive biases. Studies
of charge doping as a function of cluster size in semi-conducting nanotube devices could also be
done with the more dilute coverages we have already achieved. This measurement was only
avoided out of practicality before now. I only had time (just before COVID shutdowns) to run the
first deposition that produced the results seen here. At the time it seemed more prudent to get the
SEM imaging and guarantee some morphological results, instead of risking a potential device blow
out during measurement that would leave us with no new information. Showing that the CNT
surface itself can indeed catch and support a cluster opens up the vary large solution space that is
various sputter-able materials at a range of sizes. To this end I have begun construction on an in-
situ device measurement sample mount. Once complete this mount will allow for three wire-
bonded samples to be loaded into the deposition chamber simultaneously all with electrical
measurement capabilities. With this functionality, we would be able to periodically check for
cluster conduction along an “off-state” suspended CNT during cluster deposition. Not only does
this allow confirmation of successful nanocluster array assembly, but preliminary electrical
characterization could be done prior to unloading as well. Without unloading the samples, we
98
greatly reduce the risk of device breakage, and open the door to investigating cluster species which
might more easily oxidize when exposed to air.
For metallic CNT devices, I would love to attempt the idea that “started it all’ so to speak. That
being the deposition of potentially super-conducting (SC) clusters for low temperature transport
measurements. At low coverages, one could investigate the potential of a SC state being induced
in the CNT conduction channel from the cluster dopants via the proximity effect. This has been
done in CNT devices before with the use of SC sour-drain contact material
107–109
, showing that the
induction of a SC state itself is possible. Doing so with surface supported clusters could serve as a
unique system to determine the critical temperature of surface supported SC clusters. There is
some concern, that for long CNT’s one might instead achieve a coulomb blockade in the CNT
conduction channel, which would shunt the device. If this is the case, then one would look toward
the potential for higher cluster coverages that could support cluster-cluster conduction as the CNT
channel transitions to in insulating state that lower temperatures.
Some of the groundwork towards the low temperature transport measurements mentioned above
has been accomplished up until this point. The plan as it stands now is to have these measurements
performed with the use of a Dynacool Physical Property Measurement System (PPMS). This is
essentially a cryostat with a fully integrated electrical measurement system. There is also the
freedom to employ a custom electrical measurement set-up should we determine that the onboard
suite of resistance and differential conductance packages do not suit our needs effectively. Samples
will be loaded onto the surface of specially modified “sample pucks” (Figure 51). These pucks
will be outfitted with a permanently wired chip carrier socket for quick and easy swapping of
various devices that are wire-bonded into corresponding ceramic chip carriers.
99
Figure 51 –PPMS system and breakout of sample chamber: Schematic overview of the PPMS which allows us
to mount a wired sample onto a puck that fits into a specially designed helium cryostat capable of cooling to below 2
K.
Lastly, progress has been made in retrofitting the PPMS sample puck to support our samples.
Figure 52 shows two prototype models. The first features a ceramic chip carrier mounted directly
to the surface of the puck. This allows for the use of a larger carrier to fit correspondingly larger
devices. While the second takes advantage of a permanently mounted chip carrier socket for
repeated use of smaller devices. Both designs are electrically isolated from the puck surface by
layers of cigarette paper dipped into GE-varnish. There was some concern that the various layers
of insulation coupled with the carrier/socket thickness would result in either inaccurate
temperature readings or longer equilibration times to each desired temperature set-point. To test
this, a small length of Niobium wire was tested in a 4-wire resistance measurement configuration
(Figure 53). The resulting measurement confirmed the expected phase transition at ~9K
confirming that there is negligible thermal power loss in our design.
100
Figure 52 – PPMS Sample puck: Prototypes for device sample pucks to be used in low-temperature transport
measurements
Figure 53 – Sample puck socket testing: Niobium test wire soldered into the sample puck and measured down to
2 K showing a SC transition at expected Tc, confirming good thermal transport through the socket.
101
5 Experiments involving gaseous ions on semi-conducting SWCNTs
I was very fortunate over the course of my PhD to receive an opportunity to collaborate with
Adam Bushmaker at the Aerospace Corporation as well as Professor Steve Cronin’s group at USC.
These collaborations lead to many great learning experiences including a development of my skills
in fabricating and characterizing carbon-based devices. In addition to learning new hands on skills
I was given the opportunity to contribute to an ongoing project studying the interactions of gaseous
ions with the conduction channel surfaces of CNT-FETs. Over the course of a couple summers
and intersessions several successful experiments were conducted investigating the change of the
adsorption lifetime of ionic species on the CNT surface with varying system conditions as well as
the effect of adsorbed ions on device performance. The sections below summarize or results to
date including one publication and some preliminary results of ongoing study.
5.1 Water Assisted Desorption and Solvation of Ions on SWCNTs
The contents of this section consist largely on work I was fortunate enough to have published in
ACS nano along with my contributing co-authors Dr. Bo Wang, Professor Stephen Cronin, and
Dr. Adam Bushmaker. Some minor edits have been made to integrate this work, including an
expansion of the methods in section 3.5 above. Supporting information for the work can be found
online though ACS (https://pubs.acs.org/doi/10.1021/acsnano.0c05638?goto=supporting-info).
102
Figure 54 – Heading and TOC figure for publication results in ACS nano Nov. 2020
Abstract
We have investigated the change in mean residence time of gaseous ions adsorbed on the surface
of suspended carbon nanotube field-effect transistors (CNT-FETs) with and without native surface
water layers that exists in atmospheric conditions. Devices were characterized electrically before
and after dehydration by thermal, dry gas, and vacuum desiccation, and in each scenario were
found to have substantially higher mean ion residence times. It is proposed that water molecules
103
native to the CNT surface in ambient conditions provide a reduction pathway for incoming gaseous
ions, yielding hydronium ions (H3O
+
). This is supported by the appearance of frequent clustered
re-adsorption events in the presence of surface water, caused by the rapid hopping of H
+
between
the device surface and the lowest water layer, which are not present in data collected from
desiccated devices. After desiccation of the device, a thermal trial was conducted to determine the
adsorption energy of N2
+
ions on the CNT surface. This work has profound implications for our
understanding of wetting in one-dimensional systems, and the chemistry of ion chemisorption and
solvation on the surfaces of materials in general.
Detectable ion adsorption on SWCNT’s
As semiconductor device structures continue to scale toward the limit of one-dimensional
channels, it is becoming increasingly necessary to investigate how surface dopant dynamics effect
charge conduction in restricted environments. To that end, carbon nanotube field effect transistors
(CNT-FETs) act as ideal experimental platform to further study these effects. The inherently 1D
nature of the CNT-FET along with its great versatility through electrostatic gating makes it a highly
sensitive probe for the detection of molecular/ionic interactions with the CNT surface. CNT
devices have already been used successfully in the development of gas and biological sensors in
recent years via their extremely high surface-to-volume ratio, which makes them electrically
sensitive to surface adsorbates.
51,54
Devices that make use of single suspended CNTs especially
have the capability of monitoring individual ionic interactions with device surfaces without the
noise induced by substrate defects.
49,55–57,110–113
In agreement with previous reports
55–57
we have found that, in the presence of positive gaseous
ions generated by a radiation exposure, suspended CNT-FET devices undergo large switching
events in conduction characteristics. These events are attributed to the interaction of individual
104
gaseous ions with the surface of the CNT conduction channel.
57
As ions approach the CNT surface,
an electrostatic potential barrier forms in the valence band (Figure 55a), hindering charge carrier
transport thereby reducing the current, shifting the gate voltage threshold, and increasing the
subthreshold swing of the device which can be seen in the repeated back gate voltage sweeps
depicted in Figure 55b. Unlike coulomb scattering in 2D MOSFETs,
114,151
holes in the valence
band have no unobstructed path to avoid the electrostatic barrier in the one-dimensional CNT
channel. This results in the large (1-2 orders of magnitude) drop in the conduction of the devices,
a far greater effect than the traditional gate screening induced by substrate defect traps in traditional
surface supported CNT-FETs,
47,48
which occur further from the conduction pathway.
In an effort to further characterize these events, we have monitored device conduction at a fixed
back gate voltage ( 𝑉 𝑔 ) , effectively monitoring a device at a vertical slice of Figure 55b (𝑉 𝑔 =0) as
it evolves over time (Figure 55c). While doing this in the presence of gaseous ions we observe the
sharp transitions from nominal device conduction to a reduced conduction state caused by ionic
adsorption. From this data we are then able to extract the individual ion residence times on the
CNT surface by recording the duration of the reduced conduction state (Figure 55d).
105
Figure 55 – Ionic interaction with CNT device conduction: (a) Cartoon representation of ionic surface dopant
on a CNT device channel with accompanying plot of the Coulomb potential barrier in the valence band along the
length of the CNT. (b) Repeated gate voltage dependent transport samples of a device depicting both nominal p-type
behavior (without the adsorbed ion species) and large reduced state sweeps (with the adsorbed ion species). (c)
Exemplary data, collected at 𝑉 𝑔 =0, depicting three large switching events in device conduction. The black box outlines
the zoomed plot of the first switching event (d) which shows sharp transitions between the two conduction states of
the device. The reduced CNT conduction state is a result of ion adsorption at the CNT surface and thus its duration
corresponds with the ion residence time on the surface.
106
Surface adsorbed water layers in on Carbon surfaces
In further experiments building on these previous results
55–57
, using meticulously controlled
sample chamber environmental conditions and vastly expanded measurement dynamic range and
duration, we have found that the durations of these large transient ion adsorption events are
strongly affected by the existence of surface adsorbed water. Atmospheric water has long been
known to coat surfaces in ambient conditions
152–157
and often serves as a highly persistent
contaminate of the internal surfaces of ultra-high vacuum systems, where it is so tightly bound that
high vacuum baking or UV treatment is often required for its removal.
158
Though originally
thought to be hydrophobic in nature, CNTs and graphene are no exception to this phenomena.
159–
164
The existence of surface water and other liquid solvents has been extensively investigated in
CNT and graphene device systems and has been linked to changes in many device
characteristics.
163–171
By measuring redshifts in the photoluminescence spectrum of suspended
CNT devices (like those detailed in our work) Homma et al.
163
showed that the presence of water
layers on the CNT surface is sensitive to environmental water vapor pressure, undergoing what
appears to be a first order phase transition in surface coverage about a critical pressure. This finding
is supported by MD and DFT simulations investigating the formation of water on graphene and
CNT surfaces
159–164
that indicate multiple layers of self-assembled hydrogen bond networks up to
11.3 Å thick.
163
The interplay between the dynamics of ion reduction/solvation via water and the rate and lifetime
of ion adsorption events on the surface of CNTs is particularly important to the development and
understanding of future CNT device applications, and for the understanding of ion-surface
interactions in general. This work seeks to both better our understanding of reliability in nanoscale
device structures for future low power electronic applications, and also to display the efficacy of
107
this system in identifying the individual adsorption energies of gaseous species at electrically
tunable surfaces. This second assertion is key to bettering our fundamental understanding of ion
doped surfaces, with applications to the study of ion dynamics in mitigating or tuning electrostatic
discharge (ESD) at charged surfaces,
172,173
surface modifications of various carbon based and
polymer substrates using corona discharge,
174,175
and directed cold atmospheric plasma treatments,
which are of growing importance as a therapeutic method in the medical community.
176
With
measurements of these water/ion/CNT interaction events at the level of the individual adsorbed
particle, this work is well positioned to increase this understanding.
Experimental procedures for device desiccation and results
Several experimental protocols were followed in this work. In a thermal desiccation experiment
the temperature of the sample mount was increased in an initial heating phase to 71° C while the
sample was in dry N2 gas purge. The sample was then cooled back to room temperature, and then
the temperature was increased again. In a gas desiccation trial, the sample was exposed to dry N2
at room temperature for long periods of time, hereby drying the CNT surface. Finally, the sample
was exposed to vacuum for one hour, at room temperate. In each case, switching events in device
conduction were measured over the course of the experiment and compared with data collected in
ambient conditions.
Figure 56a shows the evolution of mean ion residence time (𝜏 𝑖𝑜𝑛 ) over the course of the thermal
desiccation phases during which water was evaporated off the CNT surface. During Heating Phase
1 (orange), the device was heated from 25° C through pre-set temperature setpoints (25° C, 37° C,
51° C, 63° C, 71° C). Source current sampling data was collected in dry N2 at each temperature
for approximately 7 hours, with 25,710 switching events measured. Accompanying the increase in
temperature is a decreasing trend in 𝜏 𝑖𝑜𝑛 . After the initial heating phase, the temperature points
108
were retraced cooling the device to room temperature (blue). During this phase, 𝜏 𝑖𝑜𝑛 shows a
drastic increase at all temperatures, showing the sensitivity of ion residence to the presence of
adsorbed water on the device. Finally, the device was reheated (purple) and 𝜏 𝑖𝑜𝑛 again decreased
with increasing temperature, while values at each temperature step increase slightly further but
remain consistent with the newly dehydrated state. After Heating Phase 1, for measurements at
25° C and 37° C the source current remained locked into the low conduction state throughout the
entire 7 hours measurement time. The residence times reported in Figure 56a at these temperatures
correspond to this duration, where the 𝜏 𝑖𝑜𝑛 has increased to the point of consistently overlapping
events. Figure 56b plots 𝜏 𝑖 𝑜𝑛
at 25° C during experimental stages, and after three days atmospheric
exposure, showing a nearly complete recovery of the original (shorter) 𝜏 𝑖𝑜𝑛 .
109
Figure 56 – Evolution of the mean ion residence time (𝝉 𝒊𝒐𝒏 ) over the course of the multiple experimental
protocols: (a) 𝜏 𝑖𝑜𝑛 versus temperature, with arrows showing the direction of temperature sweeps for the initial heating
phase (orange), the cooling phase (blue), and secondary heating phase (purple). Data taken below 51° C after the initial
heating phase is permanently locked into the low conduction state (red) due to overlapping events. The plotted points
in these cases represent the total experiment run time at these temperature set-points. (b) 𝜏 𝑖𝑜𝑛 at 25° C plotted versus
experiment stage, including data taken after a 3 day recovery period in atmospheric conditions. (c) 𝜏 𝑖𝑜𝑛 evolution in
three hour intervals over the course of ~60 continuous hours of dry gas desiccation.
An alternative method utilized to remove water from the CNT surface was desiccation in a dry
gas environment. Here data was collected continuously for approximately 60 hours in a constant
110
flow of dry N2. Data containing over 16,000 switching events was then separated into segments of
approximately 3 hours in length and analyzed to determine the evolution of the 𝜏 𝑖𝑜𝑛 over the course
of the experiment. This analysis is summarized in Figure 56c showing a steady increase in 𝜏 𝑖𝑜𝑛 as
a function of desiccation time.
Finally, we compare data collected before and after vacuum desiccation. The device was held in
vacuum for one hour at 22° C which has been previously shown to be an effective method for
water removal.
163
The effect of desiccation on individual ionic residence can be seen in the data
shown in Figure 57, where we note the drastic increase in all but one ionic residence. Analysis of
approximately 294 minutes of data with 728 measured events leads to a rise in 𝜏 𝑖𝑜𝑛 from ~270 +/-
92 ms initially to ~774 +/- 117 ms after desiccation.
Figure 57 – Exemplary data before and after vacuum desiccation: Time dependent conduction data taken before
(a) and after (b) vacuum desiccation showing a large increase in the reduced conduction state, corresponding to
increased ion/surface residence.
Monte Carlo simulation of overlapping event correction factor
Previous work by Bushmaker et al.
57
analyzed data taken from repeated 40 second sampling
periods of device conduction in the presence of a Co-60 source in a dry gas environment. Each 40
111
second sampling trail was immediately followed by approximately 10 seconds of “dead time”
where the device was held at ground while the data measured in the preceding trail was saved to
file. These repeated samples possessed both high resolution and a low noise floor, which allowed
for the detection of consistent multi-level behavior in the low conduction state signal. These
multiple signal levels corresponded to quantized increases in device resistance and were thus
attributed to multiple ions concurrently adsorbed to the CNT surface which appear to act
completely independently. Analysis of the event frequency as a function of the number of
interacting ions was well fit to the expected Poisson distribution for non-interacting events. Thus,
ions adsorbed on the surface of a device can be thought of as series resistors, where each
subsequent drop in device conduction (rise in resistance) is attributed to the addition of a single
ion to the system. In this study, we focus on the collection and characterization of the exact
durations of each ionic interaction. Given this, we adjusted our experimental setup, such that it
would be capable of continuously measuring the device. This is very important in the case of
dehydrated devices where the event duration becomes larger, that is, on order or even exceeding
the 40 second sampling limit of previous trials. In these cases, any events that fell within the “dead
time” or on the boundaries of the sampling trial (starting before or ending after each sample) would
be lost due to our inability to accurately determine the event duration. By continuously measuring
the device, throughout the entire experiment we can accurately determine event start and end times,
without risk of them falling partially or entirely between separate sampling trials.
The caveat to this transition is that the experimental setup did not possess the low current
resolution to resolve multiple-ion events. This can lead to a discrepancy between the measured
duration of a switching event and the actual ionic residence time in high duty cycle data, where
ion arrival rates are high relative to residence times, causing some individual events to overlap.
112
For ion adsorption duty cycles higher than ~2%, the probability of overlapping events becomes
significant, resulting in an over-estimation of ion residence based purely on the amount of time
spent in the low current state. Given the broad range and evolution of individual residence times
measured over the course of the experiments, it is difficult to set the ion arrival rate to avoid event
overlap over the entirety of a multiday experiment while also encouraging enough events to collect
meaningful statistics. To address this, we performed a Monte Carlo simulation over a large range
of simulated data set sizes, measured duty cycles, and expected residence times to determine an
ion residence duration correction factor (𝐶𝐹
𝑑𝑢𝑟𝑎𝑡𝑖𝑜𝑛 ) . This correction factor converts the measured
mean switching event duration, which could be inflated as function of multiple ionic interactions
at high duty cycles, into the mean ion residence time (𝜏 𝑖𝑜𝑛 ) for individual ion adsorption. We found
this factor to be,
𝐶𝐹
𝑑𝑢𝑟𝑎𝑡𝑖𝑜 𝑛 =
𝜆 𝑒 𝜆 − 1
( 1)
Where 𝜆 is the standard Poisson parameter, and 𝑑𝑢𝑡𝑦 𝑐𝑦𝑐𝑙𝑒 = 1 − 𝑒 −𝜆 . Greater detail of this
simulation along with figures pertaining to its validity under various conditions can be found in
the supporting information file (S2-S5). Results presented for 𝜏 𝑖𝑜𝑛 that were obtained at duty
cycles higher than 2% have been corrected with the above factor. The average number of ions on
the CNT at a given time can be calculated from the Poisson distribution. For reference, the data
shown in Figure 56a had duty cycles ranging from a low of 0.0975 %, corresponding to an average
number of ions on the CNT of 0.00098 ions, to a high of 76.8% (excluding locked-down
measurements), corresponding to an average number of ions on the CNT of 1.46 ions, with a
corresponding correction factor CFduration = 0.441.
113
Expected behavior of surface adsorbed dopants obscured by water desiccation
To understand the effect of thermal desiccation on the ion residence, we must first consider how
the mean ion residence time is expected to change as a function of surface temperature. The
standard treatment for adsorption kinetics of non-interacting particles at solid surfaces describes
the mean residence time of an adsorbed ion to be
𝜏 𝑖𝑜𝑛 = 𝜏 0
𝑒 ( −𝐸 𝑎 /𝑘 𝑏 𝑇 )
( 2)
where 𝐸 𝑎 is the particle/surface activation energy of adsorption (referred to as the adsorption
energy hereafter) and 𝜏 0
is related to period of oscillation of the surface.
152–154
Thus, with
increasing temperature of the system one should expect a drop in the residence time of adsorbed
species.
This inverse relationship between residence and temperature is consistent with each phase of the
experiment presented in Figure 56a, except for the data collected in Heating Phase 1 at 71° C,
where we see a small rise 𝜏 𝑖𝑜𝑛 . We credit this anomalous data point to the simultaneous desiccation
of the device due to prolonged heating and exposure to dry N2 during Heating Phase 1. Here, the
desiccation effect overcomes the temperature dependence of desorption, resulting in a minor
increase in residence time. While most noticeable at this final data point, this desiccation effect as
present at every point of the first heating phase and it is this convolution of temperature effects
with the desiccation the makes fitting this data to 𝐸𝑞 2 impossible. In a device without this
competition between temperature and desiccation, one should expect the subsequent thermal
cycling to re-trace the same curve along a plot relating 𝜏 𝑖𝑜𝑛 to the system temperature. Instead, our
data shows the large influence of adsorbed water when comparing the data taken at fixed
temperatures after thermally cycling the device. As described earlier, Figure 56a shows that the
114
result of the first heating phase leads to a drastic increase in 𝜏 𝑖𝑜𝑛 at all temperatures displaying the
direct result of dehydrating the tube on the mean residence of adsorbed ions. The further small
increase in 𝜏 𝑖𝑜𝑛 after the second heating phase is attributed to the further desiccation of the device
surface by extended periods in a dry N2 environment at elevated temperature. This behavior was
reproduced in tests on greater than 10 separate devices, without exception. Experiments were also
repeated in the same device after re-exposure to atmospheric conditions for several days (Figure
56b). Here the device recovers its normal behavior at room temperature, with shorter ion-
adsorption lifetimes when exposed to ionized gasses. The device can then be re-heated and again
exhibits longer ion-adsorption lifetimes. The increase of the mean ion residence, along with the
recovery of devices when re-exposed to atmospheric conditions are common features throughout
all three experimental procedures.
In the data of Figure 56a there is not only a convolution of temperature and desiccation alone,
but also the competition between two desiccation mechanisms, that of applied heat and dry gas
exposure. During heating cycles this is unavoidable since the flow of dry gas acts as the source of
ionizable particles with which switching events are generated. We are however able to monitor the
effect of dry gas desiccation directly on a pristine device at room temperature as depicted in Figure
56c. Here the desiccation is seen more gradually where water removal is expected to be slower
due to the room-temperature, atmospheric pressure conditions. This effect does not appear to
saturate even after 60 hours of gas desiccation. Thus, we show the ion residence time increase
depends greatly on the desiccation technique, with heated desiccation resulting in larger changes
than room temperature gas or vacuum desiccation.
We observe a correlation between the desiccation time and the time required to return to nominal
behavior. This feature could be due to incomplete water desorption from the metal contact pads,
115
where it is more tightly bound than the CNT surface, due to hydrophilic polar oxygen and hydroxyl
terminations of the metal surface. Water on the metal contacts could act as anchor points on either
side of the suspended CNT for the assembly of water networks across the tube surface. Still present
after shorter desiccation exposures, these anchors could allow for faster re-wetting of the CNT
upon exposure to atmosphere. Even still, gas desiccation trials lasting just 24 hours at room
temperature required more than 40 hours in ambient conditions for full device recovery.
We note that one could consider the ionization or removal of some other surface contaminate,
like O2 or organic polymers, as potential contributors to the presented data. We believe this to be
unlikely given the expected adsorption/desorption kinetics for those species. Considering the case
of O2 or other atmospheric contaminates, it is theorized that such molecules would have very weak
surface adsorption energies
177,178
Given the extremely short expected residence time for such
weakly bound contaminates (Figure 58) one would also expect them to desorb quickly from the
surface (on the order of fractions of a second in conservative estimations). Once the device is
transferred to a pure dry N2 gas environment there is no way for the desorbed contaminates to be
replenished and thus their effect would drop of much faster than that which we see in our data,
which is observed to be continually evolving over many hours in each experimental protocol. In
the case of bound organics, we first bring attention to the method of device fabrication, which is
done via chemical vapor deposition (CVD) and without any post processing after CNT growth.
This method removes major sources of contamination that are found using alternative methods
such as drop casting ligand supported CNTs, and/or lithographic resist residues that are
unavoidable when defining device structures after CNT growth. We do acknowledge that
migration of contaminates onto device surfaces has been shown by storage in polymer carrying
containers.
179
While these contaminates could reside on the device surface, these molecules are
116
known to have very high adsorption energies.
180
Thus, in contrast to the weakly bound atmospheric
contaminates, the removal of these polymers by gas desiccation would take many days (Figure
58).
Figure 58– Mean residence time of adsorbed species on CNT surfaces: Plots of 𝐸𝑞 2 with respect to temperature
(a) and adsorption energy (b) where 𝜏 0
= 100 𝑐𝑚
−1
, the radial breathing mode of the CNT. Both plots display the
very large range of expected mean adsorption times that are highly sensitive to surface adsorption energy. These range
from nanoseconds for weakly physiosorbed particles (0.2 eV), to many days for chemisorbed particles (1.0 eV).
Evidence to surface/water proton hopping in clustered adsorption events
These results demonstrate that native water layers play an important role in the dynamics of
particle-surface interactions at the smallest scales. Our data suggests an interaction between water
molecules and the incoming gaseous ion species that act as a relaxation mechanism for ionic
adsorption at the CNT surface, resulting in shorter ion residences. Recent work by T. Inaba and Y.
Homma
165
details the shift and degradation of photoluminescence from CNT devices (similar to
those in this work) when exposed to atmospheric ions in air. This observation is attributed to the
chemisorption of hydronium molecules (H3O
+
) on the CNT shifting the local dielectric function
near the device surface. Similarly, our results are well described by the generation of hydronium
117
ions within the adsorbed water layers by the reduction of the incoming flux of N 2
+
ions. In the
presence of humidity, atmospheric ions (N2
+
, O2
+
) are effectively reduced, transferring charge from
atmospheric water, generating hydronium molecules as a byproduct.
181–184
It is highly likely that
this happens as N2
+
, O2
+
ions interact with the surface water layers on CNTs. In aqueous
environments, hydronium ions become solvated by the surrounding water molecules
185–187
resulting in the formation of hydronium core water clusters of the form (H3O
+
)[H2O]n. This
hydronium, or the spare proton associated with it, is highly mobile in solution via a combination
of thermal diffusion and proton hopping along adjacent water molecules known as the Grotthuss
mechanism.
188,189
In this framework, our results prior to the removal of the water from the device
show the interaction of the CNT-device with either the hydronium ion itself, or with a free H
+
transferring directly to the CNT surface. Evidence for these processes has been presented in MD
simulations on graphene and other hydrophobic surfaces.
190–194
We note specifically two studies.
First simulations by Cole et al.
194
which show that, when distributed in an electrolytic solution
over a graphene sheet, H3O
+
ions are found preferentially near the graphene/water interface at a
separation distance equal to the height of the first water layer. Secondly, simulations of single
Hydronium ions over graphene surfaces by Mohammadi et al.
190
show direct proton transfer back
and forth between the carbon lattice and an adjacent water molecule.
Evidence for interaction between ions and the adsorbed water layers was found by further
analyzing a shift in event frequency as a function of device desiccation. The presence of a host
water layer, in which solvated ions/hydronium can reside between switching events, leads to
desorption/re-adsorption events at a rate independent of, and much higher than, the gas-phase ion
arrival time. We observe this feature in the form of frequent clustered switching events, an example
118
of which is shown in Figure 59 along with an illustration of the ion positions resulting in the device
conduction states.
Figure 59 – Rapid re-adsorption events due to ion hopping between CNT surface and lowest lying water
layer: (a) Exemplary data depicting clustered switching events due to frequent reabsorption events near the CNT
surface. Data can be broken into three distinct stages depicted in (b). Stage 1: the device is functioning nominally in
a high conduction state with either no ion presents in the system, or the ion totally screened by large amounts of surface
adsorbed water. Stage 2: an N 2
+
has landed on the device, been reduced by the surface adsorbed water, and the resulting
H 3O
+
or H
+
has adsorbed on the CNT surface. This shunts the device conduction, leaving it in a low conduction state.
Stage 3: frequent switching events caused by rapid H
+
hopping between the CNT surface and lower lying water layers
resulting in the partial recovery of the device.
Without the complication of any water interaction in the system one expects the arrival of
incoming ions to follow a standard Poisson processes.
152–154
Here, the expected time between
detectable events (TBE) follows an exponential distribution 𝑒 −𝑡 ∙𝑟 𝑖𝑜𝑛 , where 𝑟 𝑖𝑜𝑛 is the rate of
incoming ions (in counts per second, cps) and is calculated directly from the data as 𝑟 𝑖𝑜𝑛 =
1 𝑚𝑒𝑎𝑛 ( 𝑇𝐵𝐸 ) ⁄ . Histograms showing the distribution of times between observed events for those
collected before and after the vacuum desiccation of a CNT-device are plotted in Figure 60.
119
Comparing these plots, we note the anomalous peak in the shortest time bin in the presence of
water (Figure 60a), well above the exponential trend of the remaining data. A fit of the pre-
desiccation distribution to 𝑟 𝑖𝑜𝑛 = 0.0465 𝑐𝑝𝑠 , as calculated from the mean of the data, shows a
poor fit that underestimates shorter time bins, while sloping too sharply for the remaining data.
This stems from the shift toward smaller times by rapid re-adsorption mentioned above, and can
be better visualized by considering the relationship between 𝑟 𝑖𝑜𝑛 and the median time for an
exponential distribution ( 𝑟 𝑖𝑜𝑛 = 𝐿𝑁 ( 2) 𝑚𝑒𝑑𝑖𝑎𝑛 ( 𝑇𝐵𝐸 ) ) ⁄ . This drastically higher 𝑟 𝑖𝑜𝑛 , stemming
from a small median value, fits the first two bins quite well, but slopes far too quickly for larger
times. For a true exponential distribution, the value of 𝑟 𝑖𝑜𝑛 should be the same, regardless of the
mean or median derivation. This is the case for the post-desiccation histogram, where the mean
and median values differ by only 0.00012, which produces indistinguishable fit lines, where the
mean 𝑟 𝑖𝑜𝑛 = 0.0382 𝑐𝑝𝑠 is shown in Figure 60b. We propose that the peak in the pre-desiccation
data arises from the multiple rapid ion re-adsorptions (Figure 59) as they switch between the tube
surface and the lowest water layer, where they are partially screened and no longer follow the
Poisson rate constant dictated by the source, but are instead much quicker herby shifting the mean
and median values toward lower times (higher 𝑟 𝑖𝑜𝑛 ) . Fitting this data without the anomalous peak
produces a good fit to an exponential distribution (Figure 60a black) and predicts an 𝑟 𝑖𝑜𝑛 =
0.0249 𝑐𝑝𝑠 which is in agreement with the post-desiccation results. In contrast, after water
removal (Figure 60b) we find good agreement with the expected distribution for all data as N2
+
ions adsorb and desorb with no interference due to the water layers.
120
Figure 60 – Comparison of event frequency with and without adsorbed water: Clustered switching events
histograms depicting the time between adsorption events (TBE) on a logarithmic scale for events with (a) and without
(b) adsorbed water, removed by vacuum desiccation. The presence of the unexpected peak at lower time bins is
evidence of rapid ion desorption and re-adsorption between the CNT surface and adsorbed water layers that is not
present after vacuum desiccation. When fit by 𝑟 𝑖𝑜𝑛 , as calculated from the data mean and median TBE we find poor
agreement in the pre-desiccated device due to a large shift towards shorted times caused by rapid ion re-adsorption.
It is possible that the observed adsorption/desorption events are a mix of many dynamic
interactions at the surface, including direct ion adsorption, hydronium adsorption, and direct proton
transfer, which are greatly affected by the lower lying adsorbed water layers. Ions in the presence
of water near the surface not only feel the adsorption interaction of the surface itself, but also the
solvation potential of the available water which lowers its effective activation energy. As the
system is slowly desiccated (either by heat or dry gas) there becomes less water available to solvate
the surface ion leading to a steady increase in 𝜏 𝑖𝑜𝑛 during desiccation. We hope to gain a deeper
understanding of this phenomenon in further experiments by systematically varying the system
humidity, device gate bias, and surface adsorbed species. One could also consider an in-situ
measurement with concurrent Raman measurement of the CNT radial breathing mode (RBM). The
121
RBM has already been shown to be sensitive to surface adsorbates in a known manner
26
and could
be used to get a direct measure of the surface coverage in real time. We will also look to apply a
dynamically changing incoming ion current to account for the increased residence of the
dehydrated CNTs and avoid event overlapping.
Determination of the binding adsorption energy of N2+ on CNT’s Surfaces
With the knowledge that desiccation of the device surface is required in order to capture the N2
+
ionic interactions, another heated trial was conducted on a separate device after ~16 hours of
vacuum desiccation at 30 °C followed a slow heating phase from 30 °C to 42 °C over the course
of 3 days. Data was then collected from 44 °C to 56 °C, where the change in 𝜏 𝑖𝑜𝑛 is noticeable, but
not so large as to risk overlapping events. Figure 61 presents 𝜏 𝑖𝑜𝑛 determined for each temperature
setpoint along with a fit to 𝐸𝑞 2 where the adsorption energy was determined to be between 756-
781 meV depending on the selection of 𝜏 0
, which is consistent with chemisorption on the CNT
surface, significantly higher strength then that of a neutral N2.
178
Here, 𝜏 0
of 𝐸𝑞 2 was chosen in
correspondence with the out-of-plane oscillations of the CNT surface which are expressed by the
radial breathing mode. The bounds for this value were selected to be 100 𝑐𝑚
−1
or 250 𝑐𝑚 −1
, in
agreement with other studies of single-walled CNTs.
164
With this parameter fixed to each bound
in Figure 61, the adsorption energy is the only free fitting parameter.
122
Figure 61 – Adsorption energy determination: Fit of heated trial after vacuum desiccation resulting in an adsorption
energy between 756-781 meV for N 2
+
ions on the CNT surface. In each fit the value of 𝜏 0
was selected to correspond
with upper and lower bounds of the CNT radial breathing mode. As a result, the adsorption energy of the ion is the
only free parameter.
Conclusion and Acknowledgements
In conclusion, we have measured the mean residence time of gaseous ions on the surfaces of
CNT devices before, during and after thermal, gas, and vacuum desiccation. In all cases,
dehydration of the CNT surface led to an increase in the mean residence time when compared to
devices in ambient conditions, with thermally assisted desiccation resulting in increases of several
orders of magnitude. Ion residence times are measured from large switching events induced by the
electrostatic potential of the ionic species near the restricted conduction channel of the CNT. This
123
shift in the ion residence time is attributed to adsorbed water layers which, when present, cause
solvation-assisted desorption of the ion from the CNT surface, and reduce the incoming N2
+
ions,
generating hydronium ions (H3O
+
), whose spare proton is free to travel through the surface water
layers. This claim is supported by the appearance of frequent clustered events, whose switching
rate greatly exceeds that defined by the system, and which disappear after device desiccation. After
thorough desiccation of a device surface, ionic events were monitored during a heated trial and a
fit of the adsorption energy of N2
+
on the CNT surface was determined to be between 756-781
meV. These results illustrate the capability of this system for studying a multitude of ion/liquid
chemical reactions and adsorbed ionic defects at surfaces in the single dopant regime for varying
device and environmental conditions, which is of growing importance as electronic device design
and applications continue to scale towards smaller dimensions.
Acknowledgments:
The authors gratefully acknowledge V.V. Kresin, J Dawlaty, M.Y. El-Naggar and J.A Chaney
for helpful discussions regarding interpretation of the data. Funding: The portion of this work
performed at The Aerospace Corporation was supported under The Aerospace Corporation’s
Independent Research and Development Program and the University Partnership Program. A
portion of this work was performed in the UCSB nanofabrication facility, part of the NSF funded
NNIN network.; Author contributions: P.J.E, B.W, and S.B.C fabricated CNT FET samples.
P.J.E. and A.W.B designed the study, procured the experimental hardware, and collected and
analyzed the data. A.W.B performed Monte Carlo analysis. P.J.E and A.W.B wrote the paper. All
authors commented on the manuscript. Competing interests: Authors declare no competing
interests.; and Data and materials availability: The data that support the findings of this study
are available from the corresponding author, A.W.B., upon reasonable request.
124
Supporting Information: The supporting information file is available online through ACS
publishing website and contains a brief section of text describing the details of the Monte Carlo
calculation referenced above. This is accompanied by five figures (S1-S5), one for displaying the
experimental design, and four pertain to the results of the Monte Carlo simulation.
125
5.2 Water-Assisted Ionic Defect Desorption Dynamics on Isolated Carbon
Nanotubes
As a leading cavate for this thesis chapter I note that while I collected and analyzed the main
thrust of the sweeping data presented here, the sampling data presented in was collected by my
collaborator and advisor on this ongoing project Dr. Adam Bushmaker. I have subsequently
continued the analysis on this data set and present it here since we believe it helps to further explain
the effect gate bias repulsion on incoming gaseous ions. In a similar vein, the finite element
analysis source code described below was also written by Dr. Bushmaker and published in part
previously in the absence of any ionic interaction
46,195
. I have gained familiarity with the source
code and have begun some modifications to analyze the following data (negative fermi landscape
plotting, event strength as a function of ion location, etc.), but this has been a collaborative effort
with Dr. Bushmaker also and he is deserving of much credit for it.
Introduction
Following up on the results of the previous section we have begun to investigate the ionic
switching events as a function of applied back-gate bias (𝑉 𝑔 ). In the preliminary results presented
here, we have found that the application of such a bias, as is required in the normal operation of
nano-scale field effect transistors, has a great effect on the relaxation dynamics of individual ionic
surface defects. Characterizing the effects of ionic surface defects on with applied bias will give a
clearer picture of the on-state stability of devices while in use. The results presented below are still
preliminary and further data collection and analysis is required to finishing wrapping up the
project.
126
Sweeping and gas exposure procedure and results
During exposure to ionizing radiation in a dry argon environment, continuous back gate voltage
(𝑉 𝑔 ) sweeps were collected for ~17.33 hours which resulted in 28,720 total sweeps (from -4 V, up
though +4 V and back) of the device. Upon analysis, a total of 241 ionic interaction events were
detected. These events have been separated out according to sweeping direction and plotted in
Figure 62a and b. We note that in the forward sweeping direction (from -4 to +4 V) events occur
largely at the bounds of the on-state voltages of device conduction, appearing significantly less
often in the middle of the negative gate voltages. There is also an evident voltage preference in the
backward sweeping direction (+4 to -4 V) where events begin either in positive gate voltages or
very near the threshold region of the device. In all cases these events come to an end in the negative
voltage region, with no events continuing to the sweep termination at -4 volts. Many events that
do stretch into the negative voltage regime exhibit an anomalous recovery trajectory, wherein the
conduction current unexpectedly dips prior to recovering its nominal value in the absence of a
surface adsorbed ion. This size of this feature varies but is often quite large, exhibiting dips in
current of greater than an order of magnitude. This is unlike the two-level conduction switching
observed in data sampling a fixed (grounded) 𝑉 𝑔 which followed the expected Arrhenius relation
as a function of temperature for a weakly chemisorbed surface interactions, suggesting that the
addition of the bias complicates the desorption mechanism in such a way to cause further shunting
of device conduction in the presence of surface adsorbed water in ambient conditions.
After the initial exposure the device was vacuum desiccated for ~10.5 hrs to remove atmospheric
water from the CNT surface. It is note that this experiment was conducted prior to the knowledge
that heating during desiccation is seemingly required to fully remove water from the CNT surface.
Thus, we cannot be certain that, after 10.5 hrs of vacuum desiccation, the tube is fully dehydrated.
127
In fact, we know from the gas desiccation experiments of the previous section that there is certainly
some left even after even longer exposure, potentially anchored to the contacts pads at the ends of
the CNT channel. Still 10.5 hours has been shown
58,163,164
, to reduce the level of surface water to
a highly detectable level. Dehydrated devices were then exposed again to ionizing radiation in the
presence of dry Ar for ~ 16 hrs producing a total 25,551 voltage sweeps. The 4,653 detected events
from this period are again separated out according to sweep direction and plotted in Figure 62 (c
& d). We note a drastic increase in the total number of events that has led to the sweeps being
plotted at 10% opacity for ease of viewing. These events also exhibit a greater variety of reduced
state curves. Suggesting, a greater array of ionic surface interactions that were not possible in the
presence of an abundance of adsorbed water.
128
Figure 62 – Summary of ionic interaction events before and after vacuum desiccation: Forward (a) and
backward (b) gate voltage sweeps that detected an ionic interaction at the CNT device surface. Forward (c) and
Backward (d) events after vacuum desiccation plotted with 10% opacity for ease of viewing the high volume of events.
Voltage preference of ionic events: gate-controlled ion repulsion
A main feature of the events captured prior to desiccation of the device surface in Figure 62 (a
and b) is the apparent voltage preference of the adsorption and desorption event frequency with
respect to 𝑉 𝑔 . We attribute this to an ionic repulsion and attraction effect of the applied back gate
voltage. The application of a negative gate bias has two effects on the device environment as seen
by the incoming ions. First, as the gate (which sits below the suspended CNT channel) becomes
highly negatively charged, it creates an attractive potential relative to the Ar
+
ions. This reduces
the number of ions likely to interact with the CNT surface as some will be driven into the gate
129
contact that may otherwise have found their way to the CNT surface if given the opportunity to
thermally drift around in the absence of the field. Secondly, any amount of negative gating induces
hole carrier conduction along the CNT channel. This p-type nature is dictated by the work-function
potential difference between the CNT and the Pt source/drain contacts
38,39
. In this state the
increased number of hole charges in the device conduction channel act to repel incoming gaseous
ions resulting the lack of events initialed in the negative bias regime during both sweeping
directions.
The voltage preference of events is further illustrated in histograms of the event start and end points
of both sweeping directions in Figure 63 (a and b). Considering the backward sweeping direction,
we notice that nearly all adsorption events (a, yellow) begin very near zero 𝑉 𝑔 and end promptly
afterword as the bias is swept into negative voltages (b, yellow) where the hole carriers increase
and the attractive potential of the gate becomes larger. The forward sweeping direction shows
adsorption events occurring in one of two regimes. Much like the backward sweeps, ions are free
to interact with the device near zero 𝑉 𝑔 values where the surface repulsion is weak. Additionally,
we see an abundance of events that begin at 𝑉 𝑔 = -4 V. These events stem from and experimental
artifact. That is that each sweep of the device begins and ends at -4 V, after which there is a short
period (~2 seconds) of “dead time” where the most recent sweep data is saved to file. During this
saving interval all device terminals are momentarily set to ground. Here, ions are no longer repelled
from the device and are occasionally adsorbed. Once the next sweep begins the device is found to
already be in the depressed state until the ion is quickly repelled which can be seen in the large
peak for both adsorption and desorption at large negative 𝑉 𝑔 (a and b, blue). Lastly, the large
rightmost bin of the desorption events in the forward sweeping direction (b, blue) is due to the off-
state current of the device being lower than the noise floor setting of the instrument used to collect
130
the data. Here events begin at some 𝑉 𝑔 < +0.5 𝑉 and end at some point during the remainder of
the positive voltage sweep. We are simply unable to pinpoint exactly where the desorption occurs
other than to say it has done so by the time the current becomes detectable again during the
backward trace. The dynamic range of the semiconductor parameter analyzer in use is roughly six
orders of magnitude at the sampling rate required to collect the data. The large on/off ratio of the
devices in use, which is desirable for practical applications, leads to the cut-off at positive 𝑉 𝑔 where
the device current becomes miniscule.
Figure 63 – Histograms of the adsorption and desorption event location during back gate voltage sweeps
along with time series events captured at various 𝑽 𝒈 : Histograms showing the abundance of adsorption (a) and
desorption (b) events as a function of back gate voltage displaying the clear preference of certain back gate voltages
for various events.
The ionic repulsion characteristic is further supported experimentally by a time series device
sampling experiment summarized in Figure 64. Here a device was subjected to a Co-60 ionizing
gamma radiation source in the presence of dry 𝑁 2
. Over the course of the experiment the back-
gate voltage was held at fixed positions for timed intervals, beginning at zero bias before
alternating between positive and negative 0.2 V. The evolution of event frequency has been
analyzed by keeping a running average of the event counts per 24 seconds. The tabulated count
131
rate displays a strong positive correlation between event frequency and applied bias with
significantly more events at +0.2 V when compared to zero bias. This trend continues as events
appear to be greatly reduced by even modest applications negative bias, which increases hole
carrier conduction and thus the repulsive nature of the CNT surface.
Figure 64 – Summary of time series data with back-gate variation: Event count rate (top) and time series data
(bottom) events captured at +.2,0 and -.2 V back-gate bias displaying a strong effect on ion adsorption efficiency.
Finite element analysis, fermi energy landscape analysis, and high |𝑽 𝒈 |
To investigate the gate induced repulsive feature of the devices along with the unique relaxation
mechanisms of the in the backward sweeping events a finite element analysis simulation for a
SWCNT-FET was conducted. Unlike more widely available FET modeling packages that consider
only uniform electrical characteristics along the length of the CNT. This simulation breaks the
CNT channel into discrete elements and applies the Landauer model for 1D ballistic conduction
42
.
This discretization allows for the physically relevant quantities (voltage, resistance, and Fermi
132
energy) to vary spatially along the channel length, which in turn allows us to capture local surface
effects on device conduction via the addition of an ionic potential added at various positions along
the device. As mentioned in the pre-amble to this section, a more complete description of the model
has been described previously for devices in the absence of ionized defects
46,195
. The following is
a brief review of the concept with the ionic addition.
Applying the Landauer model with standard approximations due to the finite element nature of
both the contact capacitance and the applied source-drain bias, one arrives at an equation for the
Fermi energy of the system along the channel length 𝑥
𝑞 𝐸 𝐹 ( 𝑥 ) =
𝑄 ( 𝐸 𝐹 ( 𝑥 ) )+ ∆𝜇 ( 𝐶 𝑠 + 𝐶 𝑑 )+ ( 𝑉 𝐺 − 𝑉 ( 𝑥 ) ) 𝐶 𝐺 𝐶 𝛴 − 𝑈 𝑖𝑜𝑛 ( 𝑥 ) ( 1)
Here 𝐶 𝑠 , 𝐶 𝑑 , and 𝐶 𝐺 are the source, drain, and gate capacitances respectively. 𝐶 Σ
is the sum total
capacitance. ∆𝜇 is the relative surface potential difference between the CNT and the source-drain
contacts. 𝑄 ( 𝐸 𝐹 ( 𝑥 ) ) is the linear charge density in terms of the Fermi function (𝑓 ( 𝐸 ) ) and density
of states (𝐷 ( 𝐸 ) ) ,
𝑄 ( 𝐸 𝐹 ( 𝑥 ) ) = ∫ [1 − 𝑓 ( 𝐸 ) ]
0
−∞
𝐷 ( 𝐸 ) 𝑑𝐸 − ∫ 𝑓 ( 𝐸 ) 𝐷 ( 𝐸 ) 𝑑𝐸
∞
0
( 2)
and 𝑈 𝑖𝑜𝑛 ( 𝑥 ) is the potential of the adsorbed ion as seen by the cylindrical CNT channel
𝑈 𝑖𝑜𝑛 ( 𝑥 ) =
1
4𝜋 𝜀 0
𝑞 √( 𝑥 − 𝑥 0
)
2
+ 𝑑 2
( 3)
where 𝑥 0
is the location of the adsorbed ion along the CNT channel, and 𝑑 is the radial distance
of the ion off the CNT center-line axis (chosen to be 0.75 nm).
133
Solving for 𝑉 ( 𝑥 ) and substituting known experimental values for ∆𝜇 , 𝐶 𝑠 , 𝐶 𝑑 , and 𝐶 𝐺 , one is then
able to solve for 𝑉 ( 𝑥 ) numerically by generated look-up tables for 𝐸 𝐹 ( 𝑥 ) at an array of 𝑄 ( 𝐸 𝐹 ( 𝑥 ) )
values relevant to device operation.
Figure 65 – Negative Fermi Energy at select 𝑽 𝒈 : Negative fermi energy landscape of the CNT conduction channel
at 𝑉 𝑔 ranging from (-4,4) Volts. Curves transition in concavity from up to down as the bake-gate sweeps from positive
to negative bias, which corresponds to a transition in ion surface binding favorability.
Considering first the negative Fermi energy landscape (that seen by the hole carriers) along the
tube length in Figure 65, we see the curvature of the energy landscape changes as a function of the
applied 𝑉 𝑔 . The value of 𝑉 𝑔 about which the curvature changes is dependent on the selection of
some device specific parameters, namely the relative surface potential difference ∆𝜇 . How the
choice of this value effects plots about the curvature transition region (nearer 𝑉 𝑔 = 0) will be
134
discussed in more detail in a following section, but those chosen here are representative of common
device characteristics. For larger applied biases |𝑉 𝑔 | > 1 𝑉 , regardless of ∆𝜇 , there are some
features that can be gleaned about the gate dependent event frequency observed in the data. Large
positive biases produce negative and concave up energy curves, where ionic adsorption is
favorable. There likely exists a trade-off point on the higher side of the positive gate bias regime,
where the attractive Fermi energy environment is overshadowed by the repulsive field of the gate
contact underneath. Given the noise floor limitations with data collection mentioned earlier, it is
not possible to find this point with our data. The time series data at small positive biases, along
with the existence of backward sweeping events having already started as the voltage descends
from the positive values serve as proof that there is some intermediate regime where the field from
the gate contact is not so strong as to sweep ions away.
In contrast, large negative gate biases produce a positive and concave down energy landscape.
In this environment the presence of many hole carriers in the conduction channel make it less
favorable for an ion to adsorb to the CNT surface. This, in conjunction with the attractive potential
of the ion to the gate itself, accounts for the lack of ionic events initiated at high negative gate bias.
The curvature itself (sloping down on either side) also indicates an energetic preference toward
the source drain contact pads in the event that an ion lands near, or migrates towards, the CNT
channel ends, where the fermi energy is fixed to the relative contact potential difference ∆𝜇 (more
on this in a moment). This is investigated further shortly.
The lower lying negative back gate 𝑉 𝑔 < 1 𝑉 curves are positive and concave up. Observing the
event frequency on our data we see that events can occur at small negative voltages. These events
suggest that there is a particular level of hole doping required to make ions binding to the surface
unfavorable. For this device, this level appears to be around -1 V, here events can begin to be seen
135
in the forward sweeping direction. This would correspond to a surface energy of ~200-250 meV.
These curves are not tailored to this device specifically though, so those numbers are not precise,
they are mentioned only to illustrate the relationship between the data and the energy curves. In
the backward sweeping direction ions that have landed on the tube in more favorable conditions
are quickly repelled as the gate is swept to negative voltages and the hole doping increases.
Desorption events occurring at larger negative gate voltages represent those with the anomalous
recovery curves seen in Figure 62 and are the topic of the next section.
Evidence for surface migration in anomalous recoveries
Incorporating the numerical solutions to the above equations into the Landauer formalism
detailed in section 1.2.4 it is possible to solve for the conductance of each element along the CNT
channel and plot the simulated device conductance as a function of 𝑉 𝑔 . Figure 66a shows the results
for a representative batch of these simulations for a device in the absence of ionic interference as
well as with adsorbed ions at three different positions along the CNT channel. The right-most red
curve in Figure 66a represents an ion adsorbed at the exact center of the CNT channel. Due to the
mirrored symmetry of the device, this is the maximum conduction possible with and ion adsorbed
and all adsorption sites of equal distance (+/-) 50 % produce identical reduced state curves. For
example, the curve representing an ion at 8 % of the CNT length is identical one produced by an
ion at 92 % length. While only three curves were plotted for illustrative purposes, it is noted that
a continuum of potential curves exists between those presented.
136
Figure 66 – Summary of channel conductance simulations as a function of adsorber ion location: a) Plotting
of simulated device conductance at three ion adsorption sites along the channel length (red) and once in the absence
of a surface ion (black). The blue trace illustrates how an ion trajectory like those represented from the data in b) is
explained by the ion travelling down the conduction channel toward the source/drain contacts. c) a corresponding
illustrative cartoon of ion motion at each stage of the reduced state curves.
These simulations reflect the large effect of ion location on the magnitude of the current drop
observed in the channel (see cartoon in Figure 66c). This is due screening of the electric field
generated by the back gate at the source/drain contacts. Near the contacts the effect of the applied
𝑉 𝑔 is lower, as observed by the Fermi energy curves in Figure 65 which begin to slope downwards
end the channel ends. The weaker field in those regions is proportionally less effective at canceling
out the potential of the adsorbed ion, resulting in a greater reduction in the current for adsorption
sites near the contacts. By comparing this series of reduced conduction curves to a sampling of the
anomalous recoveries observed in the backward sweeping data (Figure 66b) we see evidence for
137
the mobility of the ion on the CNT surface in the presence of adsorbed water. Considering the
representative blue trace superimposed onto Figure 66a, an ion that originally adsorbs nearer the
middle of the CNT surface begins to follow the right-most reduced state curve. As 𝑉 𝑔 is swept to
higher negative voltages, the ion appears to move toward the source drain contacts along the
continuum of potential reduced state curves before either being thermally desorbed or running off
the channel and onto the contact entirely. It is important to remember that when the device is in a
hydrated state, the incoming argon ions are efficiently reduced by the surface water layers
producing a hydronium ion (H3O
+
). Thus, the ion that eventually reaches the CNT surface through
the layers is a highly mobile proton capable of hopping quickly between water molecules via the
Grottuss mechanism
188
.
A comparison of event variety as a function of CNT desiccation
Considering now the difference in event variety as a function of pumping. Here we see more
evidence of water “lubricating the CNT” so to speak, causing the ions to become more mobile.
Before desiccation of the device we see significantly less variety in the event size and shape. Due
to the short ion residence times on the hydrated device we do not see any complete curves of the
reduced conductions state drawn out by a single event. Instead the distribution of possible curves
is sampled in pieces by the shorter events and becomes evident when plotting all events together.
Here sweeps fall largely into 3 categories depicted in Figure 67.
1) Nominal sweeps with no ion (the vast majority) (red)
2) Depressed sweeps with an ion attached that all seem to be in the same general central
curve (cyan)
3) The bottoms of the odd recoveries stitched together (black)
138
Figure 67 – Event variety as a function of CNT hydration state: Backward sweeping events captured before (a)
and after (b) device desiccation highlighting the three categorical regions in the pre-desiccation device. Nominal
curves (red), central channel adsorbed ions (cyan) and maximally reduced conduction due to adsorption near the
contacts (black).
where the plotting between these traces is simply transitions from one to the other. Explaining this
lack of variety, we again look to the fermi energy landscapes. As mentioned in the previous section,
the bottom trace (black) represents all the events where an ion either lands in or migrates toward
the downward sloping regions at the bounds and is swept off to the contacts. The middle trace
appears to represent ions that adsorb at sites located nearer the center of the channel. These ionic
events begin at 𝑉 𝑔 > 0, where the fermi energy curves are universally concave up. Given the high
mobility of the solvated ion in the adsorbed water layers they are more easily ushered along the
channel by the energy landscape and find themselves adsorbing more frequently near the center at
the energy minimum.
Compare this with the post pumping where we still have a good deal of nominal curves which is
expected. While there are still some ion adsorbed events that happen more often than others, it
appears to be more of a thick band of more frequent events as opposed to a singular event type
seen in the hydrated state. Lastly there is a background after this this band of sparse but present
139
really strong events at many levels. Reduced state curves here are largely continuous as well,
generated by longer single events as is expected when the ion residence is higher on the dry device.
There may be some organized distribution here, but it’s tough to tell on inspection (this is a current
area of analysis). Looking at the conduction simulations for various ion positions begins to explain
this situation.
Figure 68 – Survey of ion adsorption location on conduction: Conduction curves as various ion adsorption sites
along the CNT channel highlighting the sensitivity of the CNT to ions near the contact pads. Note that the curves for
0.5 and 0.45 channel length lie on top of one another. Significant reduction in maximum on-state current isn’t observed
until ~10% channel length. (Thesis note: This is a preliminary plot the illustrate the effect which will made more
presentable prior to any attempt at publication)
Figure 68 above shows that for many ion adsorption sites there is little change in the shape of the
reduced state curve. In fact, for the main central region of the channel (middle 10%), the curves
are essentially identical. It isn’t until the ion begins to approach the source/drain contacts (0.1L)
that we begin to see sizeable effects. For example, everything between the right most blue curve
140
and the pale blue hexagons contains the central 60% of the channel. There is a minor (compared
to other reduced states like 45% and 50%) horizonal shift in the curve, but still little change in the
maximum on-state current. This is just a quick first pass at this sort of analysis where only a few
adsorption sites have been plotted to illustrate the point. Even still, this agrees quite well with our
observed data in the dry CNT. Without adsorbed water ions are now free to land on a wider variety
of surface sites along the tube, filling in the areas between the two reduced state curves seen in the
hydrated state.
It is noted that there are a couple anomalous recoveries in the post pumping data, though
significantly less as a function of total events. These are thought to be from the incomplete
desiccation of the CNT channel near the contacts, where water may remain anchored to the
contacts. This is supported by work in the preceding thesis subchapter (5.1) that determined that
further desiccation can be obtained with applied heat in vacuum. In the rare event that an ion lands
in this water it will be reduced as in the case for the pre-desiccated device and be more likely to
run off the channel, being already near (or within) the sloped region of the energy landscape.
The effect of ∆𝝁 on intermediate 𝑽 𝒈 curves
As mentioned above there are a few experimental values that dictate the level and transition
point of the curvature in the fermi energy curves shown above. This becomes important when
considering the landscape near the device on/off transition, namely around 𝑉 𝑔 = 0. The main
parameter in question is the relative contact potential difference defined as,
∆𝜇 = −
𝐸 𝑔 2
+ 𝛥 𝑊
where 𝐸 𝑔 is the semiconducting bandgap of the CNT, and 𝛥𝑊 is the work function difference
between the CNT and the Pt source drain contacts. Now, the difficulty present in initial analysis
141
is that these quantities very from device to device, and with some system condition. For example,
the bandgap is tube chirality dependent. We have chosen the moderate value of 0.6 eV for this
qualitative analysis. This is deemed to be a reasonable lower bound for the value given the large
on/off ratio of the devices in question. For a definitive value this would need to be determined for
each device separately prior to running an experiment. This is possible by utilizing Raman
spectroscopy to measure the radial breathing mode (𝜔 𝑅𝐵𝑀 ) of the CNT
196
, but potentially
uneconomical for a large batch of devices, some of which will inevitable be lost to breakage. Also,
as mentioned in the previous chapter, the adsorbed water on the CNT surface has been shown to
shift 𝜔 𝑅𝐵𝑀 slightly making it a moving target as the device dehydrates in the dry gas environment.
𝛥𝑊 acts similarly. While we have an approximate experimental value for this component, it too
can shift during an experiment. In fact, not only is the CNT surface sensitive to environmental
changes, but the work function of metal contact surfaces have been shown to shift due to
humidity
197
, particle adsorption
198–200
(gas or growth catalyst), or charging by incoming ions and
electrons from the radioactive source. With all of this in mind, it is more productive probably to
simply discuss the range of shifts seen in the curves and not treat any one set as gospel, but instead
take what information we can from them. At least until such a time as we find ourselves capable
of determining these parameters more precisely (this is still a preliminary result).
142
Figure 69 – Modified fermi energy landscape with ∆𝝁 = 𝟎 , 𝑬 𝒈 = 𝟎 . 𝟔 𝒆𝑽 : Negative fermi energy curves showing
the shift in the pinned values at the CNT channel bounds. The curve at zero back-gate bias here has become flat, but
even in this extreme, does not transition to concave down.
Figure 69 represents the effects of the most extreme case of parameter selection on the fermi
energy landscape, just to show the range of potential outcomes. This specific combination of
parameters may not be physically likely, they are only intended to show the level of level of
potential change between the extremes. In the top plot, we have kept the bandgap the same (0.6
eV) but raised the work function 𝛥 𝑊 such that ∆𝜇 = 0. ∆𝜇 is the point at which the energy is
“pinned” at the contacts. Thus, adjusting this shifts the pinned points up and down along the edges,
as we now see all curves are symmetric about zero energy. For each curve, this shift effects the
energy near the contacts more than the central region of the channel, resulting in a curvature change
for the lower energy bands 𝑉 𝑔 < 1 𝑉 , and the flattening of at 𝑉 𝑔 = 0 𝑉 . Initial analysis of the
anomalous events shows that the downturns in current only begin at voltages less than approx. -
143
0.5 volts, with most occurring at much more negative voltages comparatively. This tipping point
is reflected in the energy landscape as the transition to concave-down curves. Without knowing
the exact value of ∆𝜇 for our device, it is impossible to predict this transition in advance. Extreme
examples like Figures 8 and 9 very roughly bound this transition zero and -3 V, which contains
our observed value of -0.5 V. It also shows that, under realistic device parameters, the 𝑉 𝑔 = 0 𝑉
curve never transitions to downward concavity, which is reflected in our data
Figure 70 – Modified fermi energy landscape with ∆𝝁 = −𝟎 . 𝟓𝟓 , 𝑬 𝒈 = 𝟏 . 𝟐 𝒆𝑽 : Negative fermi energy curves for
large bandgap devices. The increase in the bandgap results in a vertical stretching of the curves while shifting the
concavity transition only slightly compared to figure 8. The large increase in the minimum value of the zero-bias curve
could account for occasional functioning devices that do not exhibit ionic interaction events.
Alternatively, in the bottom plot the 𝛥𝑊 was returned to 0.05 while the band gap was doubled
to the high, but still certainly realistically obtainable value of 1.2 eV. This has stretched the curves
out vertically compared to Figure 65, leaving the curvature transition roughly unchanged. What
144
has significantly changed the minimum/maximum value of the of the energy for each curve,
regardless of concavity. Shifting the energy curves in this way alters the range of gate bias values
that lead to favorable ion binding/repelling. Taking for example the minimum value 𝑉 𝑔 = 0 curve
which sits ~100-150 meV higher for the larger bandgap simulation. This increase in hole doping
could be enough to repel an ion that would otherwise bind at zero bias on a lower bandgap CNT.
This sensitivity to CNT bandgap could explain observations we have made of the occasional
device, that is seemingly fully functional, but does not exhibit switching events at zero bias.
Summary
In summary, we have shown that the rate of incoming ions interacting with the CNT device is
dependent on the applied back gate voltage 𝑉 𝑔 . This is explained by the electrostatic interaction
between the incoming positive ions with the back-gate contact as well as the increasing hole carrier
concentration in the CNT channel. This is further confirmed by analyzing negative fermi energy
along the CNT channel as negative back gate voltage increases.
In the hydrated device state, there are mobile protons in the water layers that are less adhered to
the CNT surface and free to roam in the water, influenced by the energy landscape at the device
surface. At positive voltages, the curvature of the energy landscape (concave up) would tend to
direct the ion toward the middle of the CNT, where there is little variation in event strength vs ion
position. It is more easily moved by the landscape because it is a smaller proton that is lighter and
has hopping as an alternative (and faster) movement option. This combination produces a thin
band of reduced state curves for the wet tube. As the gate sweeps to negative voltages the curvature
of the landscape transitions to concave down, where ions that find themselves near the channel
ends (by thermal diffusion, ion hopping, electromigration, or otherwise) are swept toward the
contacts. Being that they are not adsorbed long, and are swept more easily to the contacts, we don’t
145
see ions adsorbed long enough to produce full curves of the reduced states. Instead we only see
them upon stitching all the events together which have effectively sampled the landscape in many
bite sized pieces, showing both central and edge adsorption curves.
For the dehydrated tube, we have the argon ions landing on the tube. This ion is more tightly
bound to its adsorption site, and unable to move along the tube without water. Once it lands, it
sticks, and is not funneled toward the center of the tube by concave up curves, nor is it easily sent
to the contacts by concave down curves. The distribution of events obtained correlate with the
sensitivity of event strength to ionic position. For a large swath of ion locations there would be
little variation in the reduced state strength, producing our band of more frequent weaker event
curves. The now more prevalent stronger events are made possible by the ion staying in place
longer. In the edge region it is possible there is some residual water that is anchored to the contacts
(this experiment was pumped but not baked). If an ion should land in the residual water, already
near the contact, at a point of negative bias current, then it would still produce an anomalous
recovery event, accounting for the very rare times we see this post pumping.
Future work
There is still clearly some additional work to be done on this project. First and foremost, more
data should be collected across separate devices. Similar experiments should also be conducted
with thermal annealing in vacuum for long period to ensure a fully dehydrated device. Further
analysis of the migration mechanism of the proton in the water layer is required. This is important
to determine what forces act on the ion such that lead to it be potentially near the channel edges as
the concavity shifts. Random thermal diffusion, ion hopping, electromigration, or electrostatic
effects of the source/drain bias are all viable as of now and some work needs to be done to pare
those options down.
146
We could also look into determining accurately the values contributing to the value of ∆𝜇 for
each device used in experiments to better construct the fermi energy landscape and compare the
concavity transition and critical voltage points to the observed data to better determine the
ion/CNT adhesion strength. This could then be compared to a thermal annealing experiment for
Argon ions on the CNT similar to the of the Nitrogen ions depicted in the previous section
58
.
147
References
1. Novoselov, K. S. et al. Electric Field Effect in Atomically Thin Carbon Films. Science
306, 666–669 (2004).
2. Iijima, S. Helical microtubules of graphitic carbon. Nature 354, 56–58 (1991).
3. Iijima, S. & Ichihashi, T. Single-shell carbon nanotubes of 1-nm diameter. Nature 363,
603–605 (1993).
4. Lee, C., Wei, X., Kysar, J. W. & Hone, J. Measurement of the Elastic Properties and
Intrinsic Strength of Monolayer Graphene. Science 321, 385–388 (2008).
5. Wong, H.-S. P. & Akinwande, D. Carbon nanotube graphene device physics. (Cambridge
University Press, 2010).
6. Machón, M., Reich, S., Thomsen, C., Sánchez-Portal, D. & Ordejón, P. Ab initio
calculations of the optical properties of 4-Å-diameter single-walled nanotubes. Phys. Rev.
B - Condens. Matter Mater. Phys. 66, 1–5 (2002).
7. Wallace, P. R. The Band Theoy of Graphite. Phys. Rev. 71, 622–634 (1947).
8. Castro Neto, A. H., Guinea, F., Peres, N. M. R., Novoselov, K. S. & Geim, A. K. The
electronic properties of graphene. Rev. Mod. Phys. 81, 109–162 (2009).
9. Ashcroft, N. W. & Mermin, N. D. Solid state physics. (Holt, Rinehart and Winston, 1976).
10. Avouris, P., Chen, Z. & Perebeinos, V. Carbon-based electronics. Nat. Nanotechnol. 2,
605–615 (2007).
11. Bethune, D. S. et al. Cobalt-catalysed growth of carbon nanotubes with single-atomic-
layer walls. Nature 363, 605–607 (1993).
12. Guo, T. et al. Self-assembly of tubular fullerenes. J. Phys. Chem. 99, 10694–10697
(1995).
13. Guo, T., Nikolaev, P., Thess, A., Colbert, D. T. & Smalley, R. E. Catalytic growth of
single-walled manotubes by laser vaporization. Chem. Phys. Lett. 243, 49–54 (1995).
14. José-Yacamán, M., Miki-Yoshida, M., Rendón, L. & Santiesteban, J. G. Catalytic growth
of carbon microtubules with fullerene structure. Appl. Phys. Lett. 62, 657–659 (1993).
15. Huang, S., Cai, X. & Liu, J. Growth of millimeter-long and horizontally aligned single-
walled carbon nanotubes on flat substrates. J. Am. Chem. Soc. 125, 5636–5637 (2003).
16. Park, S., Vosguerichian, M. & Bao, Z. A review of fabrication and applications of carbon
nanotube film-based flexible electronics. Nanoscale 5, 1727–1752 (2013).
148
17. Mubarak, N. M., Abdullah, E. C., Jayakumar, N. S. & Sahu, J. N. An overview on
methods for the production of carbon nanotubes. J. Ind. Eng. Chem. 20, 1186–1197
(2014).
18. Geim, A. K. & Novoselov, K. S. The rise of graphene. Nat. Mater. 6, 183–191 (2007).
19. Zhang, Y., Zhang, L. & Zhou, C. Review of Chemical Vapor Deposition of Graphene and
Related Applications. Acc. Chem. Res. 46, 2329–2339 (2013).
20. Bhuyan, M. S. A., Uddin, M. N., Islam, M. M., Bipasha, F. A. & Hossain, S. S. Synthesis
of graphene. Int. Nano Lett. 6, 65–83 (2016).
21. Lee, Y. et al. Wafer-scale synthesis and transfer of graphene films. Nano Lett. 10, 490–
493 (2010).
22. Chen, J.-H., Cullen, W. G., Jang, C., Fuhrer, M. S. & Williams, E. D. Defect Scattering in
Graphene. Phys. Rev. Lett. 102, 236805 (2009).
23. Ferreira, A. et al. Transport properties of graphene with one-dimensional charge defects.
EPL (Europhysics Lett. 94, 28003 (2011).
24. Plummer, J. D., Deal, M. D. & Griffen, P. B. Silicon VLSI technology. (Prentice Hall,
2000).
25. Shelley, M. D. Microfluidic Techniques. 321, (Humana Press, 2005).
26. Franssila, S. Introduction to Microfabrication. Introduction to Microfabrication (John
Wiley & Sons, Ltd, 2010). doi:10.1002/9781119990413
27. Seisyan, R. P. Nanolithography in microelectronics: A review. Tech. Phys. 56, 1061–1073
(2011).
28. Chen, Y. Nanofabrication by electron beam lithography and its applications: A review.
Microelectron. Eng. 135, 57–72 (2015).
29. Singh, J. & Wolfe, D. E. Nano and macro-structured component fabrication by electron
beam-physical vapor deposition (EB-PVD). J. Mater. Sci. 40, 1–26 (2005).
30. Reimer, L. Scanning Electron Microscopy. 45, (Springer Berlin Heidelberg, 1998).
31. Schwierz, F. Graphene transistors. Nat. Nanotechnol. 5, 487–496 (2010).
32. Avouris, P., Appenzeller, J., Martel, R. & Wind, S. J. Carbon nanotube electronics. Proc.
IEEE 9, 1772–1784 (2003).
33. Fuhrer, M. S. Single-Walled Carbon Nanotubes for Nanoelectronics. in Advanced
Semiconductor and Organic Nano-Techniques 293–343 (Elsevier, 2003).
doi:10.1016/B978-012507060-7/50020-9
149
34. Bonaccorso, F., Sun, Z., Hasan, T. & Ferrari, A. C. Graphene photonics and
optoelectronics. Nat. Photonics 4, 611–622 (2010).
35. Donnelly, M., Mao, D., Park, J. & Xu, G. Graphene field-effect transistors: the road to
bioelectronics. J. Phys. D. Appl. Phys. 51, 493001 (2018).
36. Cao, J., Wang, Q., Wang, D. & Dai, H. Suspended carbon nanotube quantum wires with
two gates. Small 1, 138–141 (2005).
37. Hazeghi, A., Krishnamohan, T. & Wong, H. S. P. Schottky-barrier carbon nanotube field-
effect transistor modeling. IEEE Trans. Electron Devices 54, 439–445 (2007).
38. Javey, A., Guo, J., Wang, Q., Lundstrom, M. & Dai, H. Ballistic carbon nanotube field-
effect transistors. Nature 424, 654–657 (2003).
39. Lin, Y. M., Appenzeller, J., Knoch, J. & Avouris, P. High-performance carbon nanotube
field-effect transistor with tunable polarities. IEEE Trans. Nanotechnol. 4, 481–489
(2005).
40. Imry, Y. & Landauer, R. Conductance viewed as transmission. Rev. Mod. Phys. 71, S306–
S312 (1999).
41. Rahman, A., Jing Guo, Datta, S. & Lundstrom, M. S. Theory of ballistic nanotransistors.
IEEE Trans. Electron Devices 50, 1853–1864 (2003).
42. Datta, S. Electronic transport in mesoscopic systems . (Cambridge University Press,
1995).
43. Rahman, A., Guo, J., Datta, S. & Lundstrom, M. S. Theory of ballistic nanotransistors.
IEEE Trans. Electron Devices 50, 1853–1864 (2003).
44. Zhao, Y., Liao, A. & Pop, E. Multiband mobility in semiconducting carbon nanotubes.
IEEE Electron Device Lett. 30, 1078–1081 (2009).
45. Bushmaker, A. W. et al. Memristive behavior observed in a defected single-walled carbon
nanotube. IEEE Trans. Nanotechnol. 10, 582–586 (2011).
46. Bushmaker, A. W., Amer, M. R. & Cronin, S. B. Electrical Transport and Channel Length
Modulation in Semiconducting Carbon Nanotube Field Effect Transistors. IEEE Trans.
Nanotechnol. 13, 176–181 (2014).
47. Liu, F., Wang, K. L., Li, C. & Zhou, C. Study of Random Telegraph Signals in Single-
Walled Carbon Nanotube Field Effect Transistors. IEEE Trans. Nanotechnol. 5, 441–445
(2006).
48. Wang, N. P., Heinze, S. & Tersoff, J. Random-Telegraph-Signal Noise and Device
Variability in Ballistic Nanotube Transistors. Nano Lett. 7, 910–913 (2007).
150
49. Sharf, T., Kevek, J. W., Deborde, T., Wardini, J. L. & Minot, E. D. Origins of charge
noise in carbon nanotube field-effect transistor biosensors. Nano Lett. 12, 6380–6384
(2012).
50. Helbling, T. et al. Sensing NO2 with individual suspended single-walled carbon
nanotubes. Sensors Actuators B Chem. 132, 491–497 (2008).
51. Collins, P. G., Bradley, K., Ishigami, M. & Zettl, A. Extreme Oxygen Sensitivity of
Electronic Properties of Carbon Nanotubes. Science 287, 1801–1804 (2000).
52. Goldsmith, B. R., Coroneus, J. G., Lamboy, J. A., Weiss, G. A. & Collins, P. G.
Scaffolding carbon nanotubes into single-molecule circuitry. J. Mater. Res. 23, 1197–
1201 (2008).
53. Stampfer, C. et al. Nano-electromechanical displacement sensing based on single-walled
carbon nanotubes. Nano Lett. 6, 1449–1453 (2006).
54. Kong, J. et al. Nanotube molecular wires as chemical sensors. Science 287, 622–625
(2000).
55. Bushmaker, A. W. et al. Single Event Effects in Carbon Nanotube-Based Field Effect
Transistors Under Energetic Particle Radiation. IEEE Trans. Nucl. Sci. 61, 2839–2846
(2014).
56. Bushmaker, A. W. et al. Radiation Induced Single Ion Surface Effects in Nanoelectronic
Circuits. IEEE Trans. Nucl. Sci. 62, 2926–2932 (2015).
57. Bushmaker, A. W. et al. Single-ion adsorption and switching in carbon nanotubes. Nat.
Commun. 7, 1–8 (2016).
58. Edwards, P. J., Wang, B., Cronin, S. B. & Bushmaker, A. W. Direct Measurement of
Water-Assisted Ion Desorption and Solvation on Isolated Carbon Nanotubes. ACS Nano
acsnano.0c05638 (2020). doi:10.1021/acsnano.0c05638
59. Knight, W. D. et al. Electronic Shell Structure and Abundances of Sodium Clusters. Phys.
Rev. Lett. 52, 2141–2143 (1984).
60. Knight, W. D., de Heer, W. A., Clemenger, K. & Saunders, W. A. Electronic shell
structure in potassium clusters. Solid State Commun. 53, 445–446 (1985).
61. de Heer, W. A. The physics of simple metal clusters: experimental aspects and simple
models. Rev. Mod. Phys. 65, 611–676 (1993).
62. de Heer, W. A., Knight, W. D., Chou, M. Y. & Cohen, M. L. Electronic Shell Structure
and Metal Clusters. Solid State Phys. - Adv. Res. Appl. 40, 93–181 (1987).
63. Johnston, R. L. Atomic and Molecular Clusters. (2002). doi:10.1201/9781420055771
151
64. Heer, W. A. De. Confinement and Size Effects in Free Metal Clusters. in Metal Clusters
at Surfaces (2000).
65. Ekardt, W. Work function of small metal particles: Self-consistent spherical jellium-
background model. Phys. Rev. B 29, 1558–1564 (1984).
66. Ekardt, W. Dynamical polarizability of small metal particles: Self-consistent spherical
jellium background model. Phys. Rev. Lett. 52, 1925–1928 (1984).
67. Saunders, W. A., Clemenger, K., de Heer, W. A. & Knight, W. D. Photoionization and
shell structure of potassium clusters. Phys. Rev. B 32, 1366–1368 (1985).
68. Knight, W. D. et al. Alkali metal clusters and the jellium model. Chem. Phys. Lett. 134, 1–
5 (1987).
69. Saunders, W. A. Fission and liquid-drop behavior of charged gold clusters. Phys. Rev.
Lett. 64, 3046–3049 (1990).
70. Kresin, V. V. & Knight, W. D. Quantized Electronic States in Metal Microclusters:
Electronic Shells, Structural Effects, and Correlations. in Pair Correlations in Many-
Fermion Systems 245–261 (Springer US, 1998). doi:10.1007/978-1-4899-1555-9_16
71. Lee, C., Tame, M., Lim, J. & Lee, J. Quantum plasmonics with a metal nanoparticle array.
Phys. Rev. A - At. Mol. Opt. Phys. 85, 1–18 (2012).
72. Tame, M. S. et al. Quantum plasmonics. Nat. Phys. 9, 329–340 (2013).
73. Koenderink, A. F. Plasmon nanoparticle array waveguides for single photon and single
plasmon sources. Nano Lett. 9, 4228–4233 (2009).
74. Zundel, L. & Manjavacas, A. Finite-size effects on periodic arrays of nanostructures.
JPhys Photonics 1, (2018).
75. Zhang, Y. et al. Ultrasmall Au nanoclusters for biomedical and biosensing applications: A
mini-review. Talanta 200, 432–442 (2019).
76. Lin, C. A. J. et al. Synthesis of fluorescent metallic nanoclusters toward biomedical
application: Recent progress and present challenges. J. Med. Biol. Eng. 29, 276–283
(2009).
77. Knoblauch, C., Griep, M. & Friedrich, C. Recent advances in the field of
bionanotechnology: An insight into optoelectric bacteriorhodopsin, quantum dots, and
noble metal nanoclusters. Sensors (Switzerland) 14, 19731–19766 (2014).
78. Weller, H. Transistors and light emitters from single nanoclusters. Angew. Chemie - Int.
Ed. 37, 1658–1659 (1998).
79. Feldheim, D. L. & Keating, C. D. Self-assembly of single electron transistors and related
152
devices. Chem. Soc. Rev. 27, 1–12 (1998).
80. Zheng, H. et al. Room temperature Coulomb blockade effects in Au
nanocluster/pentacene single electron transistors. Nanotechnology 26, (2015).
81. Zharinov, V. S., Picot, T., Scheerder, J. E., Janssens, E. & Van De Vondel, J. Room
temperature single electron transistor based on a size-selected aluminium cluster.
Nanoscale 12, 1164–1170 (2020).
82. Du, Y., Sheng, H., Astruc, D. & Zhu, M. Atomically Precise Noble Metal Nanoclusters as
Efficient Catalysts: A Bridge between Structure and Properties. Chem. Rev. 120, 526–622
(2020).
83. Liu, L. & Corma, A. Metal Catalysts for Heterogeneous Catalysis: From Single Atoms to
Nanoclusters and Nanoparticles. Chem. Rev. 118, 4981–5079 (2018).
84. Kresin, V. Z. & Ovchinnikov, Y. N. Shell structure and strengthening of superconducting
pair correlation in nanoclusters. Phys. Rev. B - Condens. Matter Mater. Phys. 74, 1–11
(2006).
85. Kresin, V. Z. & Ovchinnikov, Y. N. Giant’ strengthening of superconducting pairing in
metallic nanoclusters: large enhancement of Tc and potential for room-temperature
superconductivity. Phys. - Uspekhi 51, 427–435 (2008).
86. Kresin, V. Z. Superconducting state : mechanisms and properties . International series of
monographs on physics vol. 161 (Oxford University Press, 2014).
87. Halder, A., Liang, A. & Kresin, V. V. A Novel Feature in Aluminum Cluster
Photoionization Spectra and Possibility of Electron Pairing at T ≳ 100 K. Nano Lett. 15,
1410–1413 (2015).
88. Halder, A. & Kresin, V. V. Spectroscopy of metal ‘superatom’ nanoclusters and high- Tc
superconducting pairing. Phys. Rev. B - Condens. Matter Mater. Phys. 92, 1–9 (2015).
89. Halder, A. & Kresin, V. V. Nanocluster ionization energies and work function of
aluminum, and their temperature dependence. J. Chem. Phys. 143, 164313 (2015).
90. Fischer, Ø., Kugler, M., Maggio-Aprile, I., Berthod, C. & Renner, C. Scanning tunneling
spectroscopy of high-temperature superconductors. Rev. Mod. Phys. 79, 353–419 (2007).
91. Halder, A., Huang, C. & Kresin, V. V. Photoionization yields, appearance energies, and
densities of states of copper clusters. J. Phys. Chem. C 119, 11178–11183 (2015).
92. Ovchinnikov, Y. N. & Kresin, V. Z. Theoretical investigation of Josephson tunneling
between nanoclusters. Phys. Rev. B - Condens. Matter Mater. Phys. 81, 1–6 (2010).
93. Ovchinnikov, Y. N. & Kresin, V. Z. Cluster-based superconducting tunneling networks.
Phys. Rev. B - Condens. Matter Mater. Phys. 85, 8–10 (2012).
153
94. McMillan, W. L. Tunneling Model of the Superconducting Proximity Effect. Phys. Rev.
175, 537–542 (1968).
95. Tinkham, M. Introduction to superconductivity . (McGraw Hill, 1996).
96. Duffe, S. et al. Softlanding and STM imaging of Ag 561 clusters on a C 60 monolayer.
408, 401–408 (2007).
97. Zhang, Y. & Dai, H. Formation of metal nanowires on suspended single-walled carbon
nanotubes. Appl. Phys. Lett. 77, 3015–3017 (2000).
98. Zhang, Y., Franklin, N. W., Chen, R. J. & Dai, H. Metal coating on suspended carbon
nanotubes and its implication to metal–tube interaction. Chem. Phys. Lett. 331, 35–41
(2000).
99. Lee, C. H. et al. Room-temperature tunneling behavior of boron nitride nanotubes
functionalized with gold quantum dots. Adv. Mater. 25, 4544–4548 (2013).
100. Jia, X., Campos-Delgado, J., Terrones, M., Meunier, V. & Dresselhaus, M. S. Graphene
edges: a review of their fabrication and characterization. Nanoscale 3, 86–95 (2011).
101. Wang, X. & Shi, Y. CHAPTER 1. Fabrication Techniques of Graphene Nanostructures. in
1–30 (2004). doi:10.1039/9781782623380-00001
102. Feigel’man, M. V., Skvortsov, M. A. & Tikhonov, K. S. Theory of proximity-induced
superconductivity in graphene. Solid State Commun. 149, 1101–1105 (2009).
103. Kessler, B. M., Girit, Ç. Ö., Zettl, A. & Bouchiat, V. Tunable superconducting phase
transition in metal-decorated graphene sheets. Phys. Rev. Lett. 104, 1–4 (2010).
104. Allain, A., Han, Z. & Bouchiat, V. Electrical control of the superconducting-to-insulating
transition in graphene-metal hybrids. Nat. Mater. 11, 590–594 (2012).
105. Bretheau, L. et al. Tunnelling spectroscopy of Andreev states in graphene. Nat. Phys. 13,
756–760 (2017).
106. Han, Z. et al. Collapse of superconductivity in a hybrid tin-graphene Josephson junction
array. Nat. Phys. 10, 380–386 (2014).
107. Kasumov, A. Y. et al. Supercurrents through single-wailed carbon nanotubes. Science
284, 1508–1511 (1999).
108. Kasumov, A. et al. Quantum transport through carbon nanotubes: Proximity-induced and
intrinsic superconductivity. Phys. Rev. B - Condens. Matter Mater. Phys. 68, 1–16 (2003).
109. Morpurgo, A. F., Kong, J., Marcus, M. & Dai, H. Gate-Controlled Superconducting
Proximity Effect in Carbon Nanotubes. Science 286, 263–265 (1999).
110. Bushmaker, A. W., Deshpande, V. V., Bockrath, M. W. & Cronin, S. B. Direct
154
Observation of Mode Selective Electron−Phonon Coupling in Suspended Carbon
Nanotubes. Nano Lett. 7, 3618–3622 (2007).
111. Deshpande, V. V. & Bockrath, M. The one-dimensional Wigner crystal in carbon
nanotubes. Nat. Phys. 4, 314–318 (2008).
112. Deshpande, V. V. et al. Mott insulating state in ultraclean carbon nanotubes. Science 323,
106–110 (2009).
113. Bushmaker, A. W., Deshpande, V. V., Hsieh, S., Boekrath, M. W. & Cronin, S. B. Direct
observation of born - Oppenheimer approximation breakdown in carbon nanotubes. Nano
Lett. 9, 607–611 (2009).
114. Ralls, K. S. et al. Discrete Resistance Switching in Submicrometer Silicon Inversion
Layers: Individual Interface Traps and Low-Frequency (1/f?) Noise. Phys. Rev. Lett. 52,
228–231 (1984).
115. Long, Y., Twiefel, J. & Wallaschek, J. A review on the mechanisms of ultrasonic wedge-
wedge bonding. J. Mater. Process. Technol. 245, 241–258 (2017).
116. Suk, J. W. et al. Enhancement of the electrical properties of graphene grown by chemical
vapor deposition via controlling the effects of polymer residue. Nano Lett. 13, 1462–1467
(2013).
117. Moser, J., Barreiro, A. & Bachtold, A. Current-induced cleaning of graphene. Appl. Phys.
Lett. 91, 1–4 (2007).
118. Lindvall, N., Kalabukhov, A. & Yurgens, A. Cleaning graphene using atomic force
microscope. J. Appl. Phys. 111, (2012).
119. Grammatikopoulos, P., Steinhauer, S., Vernieres, J., Singh, V. & Sowwan, M.
Nanoparticle design by gas-phase synthesis. Adv. Phys. X 1, 81–100 (2016).
120. Kashtanov, P. V., Smirnov, B. M. & Hippler, R. Magnetron plasma and nanotechnology.
Physics-Uspekhi 50, 455 (2007).
121. Kelly, P. . & Arnell, R. . Magnetron sputtering: a review of recent developments and
applications. Vacuum 56, 159–172 (2000).
122. Paul, W. & Steinwedel, H. Notizen: Ein neues Massenspektrometer ohne Magnetfeld.
Zeitschrift für Naturforsch. A 8, 448–450 (1953).
123. Paul, W., Reinhard, H. P. & von Zahn, U. Das elektrische Massenfilter als
Massenspektrometer und Isotopentrenner. Zeitschrift für Phys. 152, 143–182 (1958).
124. Douglas, D. J. Linear quadrupoles in mass spectrometry. Mass Spectrometry Reviews 28,
937–960 (2009).
155
125. Dawson, P. H. Quadrupole mass spectrometry and its applications. (Elsevier Scientific
Publishing Company, 1976).
126. Campana, J. E. Elementary theory of the quadrupole mass filter. Int. J. Mass Spectrom.
Ion Phys. 33, 101–117 (1980).
127. Khojasteh, M. & Kresin, V. V. Influence of source parameters on the growth of metal
nanoparticles by sputter-gas-aggregation. Appl. Nanosci. 7, 875–883 (2017).
128. Voigtländer, B. Atomic Force Microscopy. Springer Series in Geomechanics and
Geoengineering (Springer International Publishing, 2019). doi:10.1007/978-3-030-13654-
3
129. Celano, U. Electrical Atomic Force Microscopy for Nanoelectronics. NanoScience and
Technology (Springer International Publishing, 2019). doi:10.1007/978-3-030-15612-1
130. Bhushan, B. Scanning Probe Microscopy in Nanoscience and Nanotechnology. (Springer
Berlin Heidelberg, 2010). doi:10.1007/978-3-642-03535-7
131. Bhushan, B. Scanning Probe Microscopy in Nanoscience and Nanotechnology 2.
(Springer Berlin Heidelberg, 2011). doi:10.1007/978-3-642-10497-8
132. Gan, Y. Atomic and subnanometer resolution in ambient conditions by atomic force
microscopy. Surf. Sci. Rep. 64, 99–121 (2009).
133. Goodman, F. O. & Garcia, N. Roles of the attractive and repulsive forces in atomic-force
microscopy. Phys. Rev. B 43, 4728–4731 (1991).
134. San Paulo, A. & García, R. High-resolution imaging of antibodies by tapping-mode
atomic force microscopy: Attractive and repulsive tip-sample interaction regimes.
Biophys. J. 78, 1599–1605 (2000).
135. Choi, W., Shehzad, M. A., Park, S. & Seo, Y. Influence of removing PMMA residues on
surface of CVD graphene using a contact-mode atomic force microscope. RSC Adv. 7,
6943–6949 (2017).
136. Her, M., Beams, R. & Novotny, L. Graphene transfer with reduced residue. Phys. Lett.
Sect. A Gen. At. Solid State Phys. 377, 1455–1458 (2013).
137. Lin, Y. C. et al. Graphene annealing: How clean can it be? Nano Lett. 12, 414–419
(2012).
138. Algara-Siller, G., Lehtinen, O., Turchanin, A. & Kaiser, U. Dry-cleaning of graphene.
Appl. Phys. Lett. 104, (2014).
139. Xie, W., Weng, L.-T., Ng, K. M., Chan, C. K. & Chan, C.-M. Clean graphene surface
through high temperature annealing. Carbon N. Y. 94, 740–748 (2015).
156
140. Francis, G. M., Kuipers, L., Cleaver, J. R. A. & Palmer, R. E. Diffusion controlled growth
of metallic nanoclusters at selected surface sites. J. Appl. Phys. 79, 2942–2947 (1996).
141. Carroll, S. J., Seeger, K. & Palmer, R. E. Trapping of size-selected Ag clusters at surface
steps. Appl. Phys. Lett. 72, 305–307 (1998).
142. Senden, T. J. & Ducker, W. A. Surface roughness of plasma-treated mica. Langmuir 8,
733–735 (1992).
143. Ostendorf, F. et al. How flat is an air-cleaved mica surface? Nanotechnology 19, 305705
(2008).
144. Nicolet, M.-A. Diffusion barriers in thin films. Thin Solid Films 52, 415–443 (1978).
145. Khanna, V. K. Adhesion-delamination phenomena at the surfaces and interfaces in
microelectronics and MEMS structures and packaged devices. J. Phys. D. Appl. Phys. 44,
(2011).
146. Romanyuk, A., Steiner, R., Mack, I., Oelhafen, P. & Mathys, D. Growth of thin silver
films on silicon oxide pretreated by low temperature argon plasma. Surf. Sci. 601, 1026–
1030 (2007).
147. Aono, T., Iwasaki, T., Yoshimura, Y. & Nakayama, Y. Method for Fabricating Multilevel
Interconnection Structures Using Differential Delamination Energies between Metal and
Silicon Oxide Thin Films. Electron. Commun. Japan 99, 41–49 (2016).
148. Zandkarimi, B. et al. Alloying with Sn Suppresses Sintering of Size-Selected Subnano Pt
Clusters on SiO2with and without Adsorbates. Chem. Mater. 32, 8595–8605 (2020).
149. Ganeva, M., Pipa, A. V., Smirnov, B. M., Kashtanov, P. V. & Hippler, R. Velocity
distribution of mass-selected nano-size cluster ions. Plasma Sources Sci. Technol. 22,
(2013).
150. Shtepliuk, I. et al. Probing the uniformity of silver-doped epitaxial graphene by micro-
Raman mapping. Phys. B Condens. Matter 580, 411751 (2020).
151. Kirton, M. J. et al. Individual defects at the Si:SiO 2 interface. Semicond. Sci. Technol. 4,
1116–1126 (1989).
152. De Boer, J. H. The Dynamical Character of Adsorption. (Clarendon Press, 1968).
153. Adamson, A. W. & Gast, A. P. Physical Chemistry of Surfaces. (Wiley-Inerscience,
1997).
154. Israelachvili, J. N. Intermolecular and Surface Forces. (Academic Press, 2011).
155. Razouk, R. I. & Salem, A. S. The Adsorption of Water Vapor on Glass Surfaces. J. Phys.
Colloid Chem. 52, 1208–1227 (1948).
157
156. Hu, J., Xiao, X.-D., Ogletree, D. F. & Salmeron, M. Imaging the Condensation and
Evaporation of Molecularly Thin Films of Water with Nanometer Resolution. Science
268, 267–269 (1995).
157. Hamada, I., Lee, K. & Morikawa, Y. Interaction of water with a metal surface: Importance
of van der Waals forces. Phys. Rev. B 81, 115452 (2010).
158. Danielson, P. Desorbing Water in Vacuum Systems: Bakeout or UV? R&D Mag 43, 57–
59 (2001).
159. Ma, J. et al. Adsorption and diffusion of water on graphene from first principles. Phys.
Rev. B 84, 033402 (2011).
160. Iwasaki, D., Suzuki, Y. & Watanabe, K. First-principles study of the dielectric functions
of carbon nanotubes with adsorbed water. Appl. Phys. Express 10, 045101 (2017).
161. Akaishi, A., Yonemaru, T. & Nakamura, J. Formation of Water Layers on Graphene
Surfaces. ACS Omega 2, 2184–2190 (2017).
162. Maekawa, Y., Sasaoka, K. & Yamamoto, T. Structure of water clusters on graphene: A
classical molecular dynamics approach. Jpn. J. Appl. Phys. 57, 035102 (2018).
163. Homma, Y. et al. Photoluminescence Measurements and Molecular Dynamics
Simulations of Water Adsorption on the Hydrophobic Surface of a Carbon Nanotube in
Water Vapor. Phys. Rev. Lett. 110, 157402 (2013).
164. Chiashi, S. et al. Adsorption effects on radial breathing mode of single-walled carbon
nanotubes. Phys. Rev. B 91, 155415 (2015).
165. Inaba, T. & Homma, Y. Atmospheric ions as the origin of photoinduced degrading of
photoluminescence from a single-walled carbon nanotube. Appl. Phys. Lett. 107, 071907
(2015).
166. Kim, W. et al. Hysteresis Caused by Water Molecules in Carbon Nanotube Field-Effect
Transistors. Nano Lett. 3, 193–198 (2003).
167. Na, P. S. et al. Investigation of the humidity effect on the electrical properties of single-
walled carbon nanotube transistors. Appl. Phys. Lett. 87, 093101 (2005).
168. Bradley, K., Gabriel, J. C. P., Briman, M., Star, A. & Grüner, G. Charge Transfer from
Ammonia Physisorbed on Nanotubes. Phys. Rev. Lett. 91, 218301 (2003).
169. Levesque, P. L. et al. Probing Charge Transfer at Surfaces Using Graphene Transistors.
Nano Lett. 11, 132–137 (2011).
170. Zahab, A., Spina, L., Poncharal, P. & Marlière, C. Water-vapor effect on the electrical
conductivity of a single-walled carbon nanotube mat. Phys. Rev. B 62, 10000–10003
(2000).
158
171. Inoue, S., Tomita, Y., Kokabu, T. & Matsumura, Y. Principles of detection mechanism for
adsorbed gases using carbon nanotube nanomat. Chem. Phys. Lett. 709, 77–81 (2018).
172. Oda, T. & Ito, Y. Studies on Electrostatic Surface Discharges on Corona-Charged
Polymer Surfaces. IEEE Trans. Ind. Appl. 26, 656–661 (1990).
173. McCarty, L. S., Winkleman, A. & Whitesides, G. M. Ionic electrets: Electrostatic
charging of surfaces by transferring mobile ions upon contact. J. Am. Chem. Soc. 129,
4075–4088 (2007).
174. Leclercq, B., Sotton, M., Baszkin, A. & Ter-Minassian-Saraga, L. Surface modification of
corona treated poly(ethylene terephthalate) film: adsorption and wettability studies.
Polymer (Guildf). 18, 675–680 (1977).
175. Zaldivar, R. J., Nokes, J., Steckel, G. L., Kim, H. I. & Morgan, B. A. The effect of
atmospheric plasma treatment on the chemistry, morphology and resultant bonding
behavior of a pan-based carbon fiber-reinforced epoxy composite. J. Compos. Mater. 44,
137–156 (2010).
176. Weltmann, K.-D. & von Woedtke, T. Plasma medicine—current state of research and
medical application. Plasma Phys. Control. Fusion 59, 014031 (2017).
177. Ulbricht, H., Moos, G. & Hertel, T. Physisorption of molecular oxygen on single-wall
carbon nanotube bundles and graphite. Phys. Rev. B 66, 075404 (2002).
178. Zhao, J., Buldum, A., Han, J. & Lu, J. P. Gas molecule adsorption in carbon nanotubes
and nanotube bundles. Nanotechnology 13, 195–200 (2002).
179. Reich, D. F. Sample Handling. in ToF-SIMS: Surface Analysis by Mass Spectrometry
(eds. Vickerman, J. C. & Briggs, D.) (IM Publications and Surface Spectra Limited,
2001).
180. Godo, M. et al. Adsorption Behavior of Chemical Contaminants by Molecular Simulation.
J. IEST 46, 98–102 (2003).
181. Sekimoto, K. & Takayama, M. Influence of needle voltage on the formation of negative
core ions using atmospheric pressure corona discharge in air. Int. J. Mass Spectrom. 261,
38–44 (2007).
182. Narcisi, R. S. & Bailey, A. D. Mass spectrometric measurements of positive ions at
altitudes from 64 to 112 kilometers. J. Geophys. Res. 70, 3687–3700 (1965).
183. Shahin, M. M. Mass‐Spectrometric Studies of Corona Discharges in Air at Atmospheric
Pressures. J. Chem. Phys. 45, 2600–2605 (1966).
184. Pavlik, M. & Skalny, J. D. Generation of [H3O]+.(H2O)n clusters by positive corona
discharge in air. Rapid Commun. Mass Spectrom. 11, 1757–1766 (1997).
159
185. Markovitch, O. & Agmon, N. Structure and Energetics of the Hydronium Hydration
Shells. J. Phys. Chem. A 111, 2253–2256 (2007).
186. Sobolewski, A. L. & Domcke, W. Hydrated hydronium: a cluster model of the solvated
electron? Phys. Chem. Chem. Phys. 4, 4–10 (2002).
187. Sobolewski, A. L. & Domcke, W. ab Initio Investigation of the Structure and
Spectroscopy of Hydronium−Water Clusters. J. Phys. Chem. A 106, 4158–4167 (2002).
188. Agmon, N. The Grotthuss mechanism. Chem. Phys. Lett. 244, 456–462 (1995).
189. Lee, D. H., Choi, C. H., Choi, T. H., Sung, B. J. & Kang, H. Asymmetric Transport
Mechanisms of Hydronium and Hydroxide Ions in Amorphous Solid Water: Hydroxide
Goes Brownian while Hydronium Hops. J. Phys. Chem. Lett. 5, 2568–2572 (2014).
190. Mohammadi, S. S. et al. Density Functional Theory and ab Initio Molecular Dynamics
Investigation of Hydronium Interactions with Graphene. Energy Procedia 110, 518–522
(2017).
191. Yin, J., Zhang, Z., Li, X., Zhou, J. & Guo, W. Harvesting Energy from Water Flow over
Graphene? Nano Lett. 12, 1736–1741 (2012).
192. Iuchi, S., Chen, H., Paesani, F. & Voth, G. A. Hydrated Excess Proton at
Water−Hydrophobic Interfaces. J. Phys. Chem. B 113, 4017–4030 (2009).
193. Zimmermann, R., Freudenberg, U., Schweiß, R., Küttner, D. & Werner, C. Hydroxide and
hydronium ion adsorption — A survey. Curr. Opin. Colloid Interface Sci. 15, 196–202
(2010).
194. Cole, D. J., Ang, P. K. & Loh, K. P. Ion Adsorption at the Graphene/Electrolyte Interface.
J. Phys. Chem. Lett. 2, 1799–1803 (2011).
195. Bushmaker, A., Amer, M. & Cronin, S. Computational modeling of channel length
modulation in carbon nanotube field effect transistors. in 2014 IEEE Aerospace
Conference 1–8 (IEEE, 2014). doi:10.1109/AERO.2014.6836430
196. Jorio, A. et al. Structural (n, m) determination of isolated single-wall carbon nanotubes by
resonant Raman scattering. Phys. Rev. Lett. 86, 1118–1121 (2001).
197. Musumeci, F. & Pollack, G. H. Influence of water on the work function of certain metals.
Chem. Phys. Lett. 536, 65–67 (2012).
198. Jaeckel, R. & Wagner, B. Photo-electric measurement of the work function of metals and
its alteration after gas adsorption. Vacuum 13, 509–511 (1963).
199. Gossenberger, F., Roman, T., Forster-Tonigold, K. & Groß, A. Change of the work
function of platinum electrodes induced by halide adsorption. Beilstein J. Nanotechnol. 5,
152–161 (2014).
160
200. Hückstädt, C. et al. Work function studies of rare-gas/noble metal adsorption systems
using a Kelvin probe. Phys. Rev. B - Condens. Matter Mater. Phys. 73, 1–10 (2006).
Abstract (if available)
Abstract
The work presented here summarizes my efforts over the last 6 years to incorporate size-selected metal nanoclusters into pre-existing carbon-based devices. This work aims to benefit the fields of both cluster physics and device physics which are well positioned to assist one another by joining well known techniques from both disciplines. From the cluster perspective, graphene and carbon nanotube field effect transistors serve as excellent candidate substrates for supporting cluster deposition. These devices are well characterized and highly electrically tunable via electrostatic gating. Additionally, the two-dimensional carbon lattice surface is relatively inert and does not disturb the shape of deposited clusters appreciably. These traits combined to provide an ideal system for probing the properties of small metallic clusters that are well known in gas phase but untested on supporting surfaces. For device physics, clusters present themselves as highly flexible surface dopants for tailoring custom device behavior. They are extremely versatile in that they can be made of a wide verity of different materials, and screened very precisely by size, charge, and energy with well-known molecular beam techniques. To that end, this thesis contains my attempts to better understand surface interactions of carbon-based device with surface dopants of both cluster and molecular ions.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Fabrication, deposition, and characterization of size-selected metal nanoclusters with a magnetron sputtering gas aggregation source
PDF
Nanoelectronics based on gallium oxide and carbon nanotubes
PDF
Atomic and molecular clusters at ultra-low temperature
PDF
A measurement of the dielectric properties of water nanoclusters
PDF
A technique for assessing water cluster sizes by pickup momentum transfer and a new source with loading system for lithium clusters
PDF
Temperature-dependent photoionization and electron pairing in metal nanoclusters
PDF
Electronic and optoelectronic devices based on quasi-metallic carbon nanotubes
PDF
Bogoliubov quasiparticles in Andreev bound states of aluminum nanobridge Josephson junctions
PDF
Interfacial polarization and photobasicity: spectroscopic studies of the electrode environment and photochemical effects
PDF
Carbon material-based nanoelectronics
PDF
Carbon nanotube nanoelectronics and macroelectronics
PDF
Integration of cost effective luminescent materials into organic light emitting diodes (OLEDs) and molecular approaches to improve the device efficiency
PDF
Experiments in electrostatic deflection of doped helium nanodroplets
PDF
One-dimensional nanomaterials: synthesis and applications
PDF
GaAs nanowire optoelectronic and carbon nanotube electronic device applications
PDF
From cables to biofilms: electronic and electrochemical characterization of electroactive microbial systems
PDF
Tailoring compositional and microstructural complexity in nanostructured alloys
PDF
Carbon nanotube macroelectronics
PDF
In-situ characterization of nanoscale opto-electronic devices through optical spectroscopy and electron microscopy
PDF
Microbially synthesized chalcogenide nanostructures: characterization and potential applications
Asset Metadata
Creator
Edwards, Patrick Joseph
(author)
Core Title
The adsorption and selective deposition of molecular and nanocluster ions on carbon based devices
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Physics
Degree Conferral Date
2021-08
Publication Date
08/03/2021
Defense Date
08/01/2021
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
carbon nanotubes,cluster deposition,device physics,field effect transistors,graphene,interfacial water,ionic surface dopants,magnetron sputtering,microelectronics,nanoclusters,OAI-PMH Harvest
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Kresin, Vitaly (
committee chair
), Bushmaker, Adam (
committee member
), Dawlaty, Jahan (
committee member
), El-Naggar, Mohamed (
committee member
), Levenson-Falk, Eli (
committee member
)
Creator Email
Patrick.j.Edwards@aero.org,pjedward@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-oUC15674428
Unique identifier
UC15674428
Legacy Identifier
etd-EdwardsPat-9980
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Edwards, Patrick Joseph
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Repository Email
cisadmin@lib.usc.edu
Tags
carbon nanotubes
cluster deposition
device physics
field effect transistors
graphene
interfacial water
ionic surface dopants
magnetron sputtering
microelectronics
nanoclusters