The Digitisation of the Sputter

Deposition Process of Transparent

Conductive Oxides by Implementing
Artificial Intelligence

A thesis submitted in partial fulfillment of the

requirements of London South Bank University
for the degree of Doctor of Philosophy
by

MD ABUL HASNATH
h"ps://orcid.org/0000-0002-9170-8797

1st Supervisor: Professor Hari Upadhyaya

2nd Supervisor: Dr Ali Salimian
 Dedication

I dedicate this thesis to the memory of my beloved mother, the late Rohima Begum who faced
extreme hardship before her passing in September 2017.
Abstract
Plasma-based sputtering is extensively employed to fabricating thin film Transparent Conductive
Oxides (TCOs), a category of semiconducting material used for a wide variety application from flat panel
display to energy harvesting devices. Methods of evaluating the plasma i.e. glow discharge has been
greatly studied requiring complex theoretical physics which is not viable for applied/materials scientist
who frequently use this method of deposition at an operational level.

The aim of the project was to explore new methods of characterizing the plasma sputtering process
to evaluate the possibility of simplifying the monitoring and assessment of the sputtering process. The
first method involved monitoring the RF-based plasma sputtering process through optical spectroscopy
and characterizing the discharge based on its specific colour. The 2nd method involved implementing
Artificial intelligence/Machine learning and feeding the emission spectrum of the plasma extracted from
an array of depositions to a deep learning model to evaluate the accuracy of predicting not only the
properties of the deposited TCO film but also the growth process conditions. Implementation of such
methods pave the way for the design of a digital shadow for plasma-based deposition in the material
engineering industry.

Spectral data from the plasma was obtained by placing an in-vacuum collimator optic probe (Plasus)
which featured a unique honeycomb structure capturing photons whilst simultaneously trapping
sputtering particles and preventing gradual coating of the collimator’s quartz window. The spectrometer
was programmed to calculate the area under the peak of the spectral range based on predesignated
segments of the spectrum. In addition to this, the light collected from the plasma was also guided to a 2nd
spectrometer (Jeti) that calculates the chromaticity index of the light.

The colour of the plasma source was deduced based on conventional chromaticity index analysis and
it was compared to the direct spectral data plots of the emission peaks to investigate the possibility of
characterizing the plasma based on its specific colour. This technique was demonstrated to be a viable
potential for evaluating the plasma in terms of providing information regarding the stability of the plasma,
chamber pressure and plasma power. A linear relationship between the colour functions and the plasma
power was observed, while the stability of the sputtering plasma can be assessed based on the plasma
colour functions. The colour functions also follow a unique pattern when the working gas pressure is
increased.

The spectral properties and colour functions of a radio frequency (RF)-based sputtering plasma
source was also monitored during consecutive sputter deposition of Indium doped zinc oxide (IZO) thin
films under argon and argon/hydrogen mix. The effect of target exposure to the hydrogen gas on charge
density/mobility and spectral transmittance of the deposited films was investigated. Consecutive
exposure to the hydrogen gas during the deposition process progressively affects the properties of thin
films with a certain degree of continuous improvement in electrical conductivity while demonstrating that
reverting to only argon from argon/hydrogen mix follows a complex pathway. Preparation of highly
conductive zinc oxide thin films without indium presence was exhibited eliminating the need for the
expensive indium addition.

The complexity of the reactive sputtering of highly conductive zinc oxide thin films in the presence
of hydrogen at room temperature was investigated. A hypothesis was put forward regarding importance

i
of precise geometric positioning of the substrate with respect to the magnetron to achieve maximum
conductivity. The electrical properties of the deposited zinc oxide thins films based on their position on
the substrate holder relative to the magnetron were examined.

Machine Learning/Deep learning models were incorporated to examine the accuracy of predicting
a single feature (sheet resistance) of thin films of indium-doped zinc oxide deposited via plasma sputter
deposition by feeding the spectral data of the plasma to the deep learning models. It was shown that
Artificial Neural networks could be implemented as a model that could predict the sheet resistance of the
thin films as they were deposited, taking in only the spectral emission of the plasma as an input.

The spectral emission data from the plasma glow of various sputtering targets containing indium
oxide, zinc oxide, and tin oxide were obtained. These spectral data were then converted into two-
dimensional arrays by implementing a basic array-reshaping technique and a more complex procedure
utilizing an unsupervised deep-learning technique, known as the self-organizing-maps method. The two-
dimensional images obtained from each single-emission spectrum of the plasma mimic an image that can
then be used to train a convolutional neural network model capable of predicting certain plasma features,
such as impurity levels in the sputtering target, working gas composition, plasma power, and chamber
pressure during the machine operation. It was demonstrated that that the single-array-to-2D-array
conversion technique, coupled with deep-learning techniques and computer vision, can achieve high
predictive accuracy and can, therefore, be fundamental to the construction of a sputtering system’s digital
twin.

ii
Acknowledgements
I would like to thank God first and foremost for allowing me to complete this thesis.

I would like to express my deepest gratitude, appreciation and thanks to my supervisors, Professor
Hari Upadhyaya and Dr Ali Salimian for their continuous guidance and support throughout this journey. I
would have struggled to complete the doctorate if it were not for them. I would like to thank Ali specifically
for his mentorship, getting me involved in his research projects and sharing his knowledge and technical
expertise. I would also like to thank my fellow colleague, Miss Cova Pardo Sanchez for her endless support
& friendship over the years. The labs and equipment would still be in a right mess were it not for her.

I would also like to thank my former supervisors Dr Jeremy Ball and Professor Hari Reehal for
accepting me as a PhD student to take on this PhD (though it has taken many twists and turns from the
initial aim) and introducing me to the world of thin film semiconducting materials and vacuum-based
technology.

Last but not least, many thanks goes to my father for constantly motivating me and sharing words of
encouragement in times of difficulty throughout the process.

iii
Abbreviations
AC Alternating Current

AI Artificial Intelligence

ANN Artificial Neural Network

AZO Aluminium Doped Zinc Oxide

BM Burnstein Moss Shift

CBM Conduction Band Minimum

CIE XYZ Commission on Illumination XYZ colour space

CIGS Copper Indium Gallium Sulphide

CNN Convolutional Neural Network

CVD Chemical Vapour Deposition

CZTS Copper Zinc Tin Sulphide

DBD Dielectric Barrier Discharge

DC Direct Current

DFT Density Functional Theory

DNN Deep Neural Network

EA Electron Affinity

EBSD Electron Backscattered Diffraction

ECR Electron Cyclotron Resonance

EDAX Energy-dispersive X-ray spectroscopy

EF Fermi Level

EI Ionisation Energy

EM Electromagnetic

FEG SEM Field Emission Gun SEM

FWHM Full Width Half Maximum

GPIB General Purpose Interface Bus

GZO Gallium Doped Zinc Oxide

HIPIMS High-Power Impulse Magnetron Sputtering

IR Infrared

iv
ITO Indium Doped Tin oxide

IZO Indium Doped Zinc Oxide

LTE Local Thermal Equilibrium

MES Mean Squared Error

MO Metal Oxide

MS Magnetron Sputtering

ND Electron Density

NH, NAr Degrees of Dissociation of Hydrogen & Argon

NIR Near Infrared

NIST National Institute of Standards and Technology

OES Optical Emission Spectroscopy

PCA Principal Component Analysis

PECVD Plasma Enhanced CVD

PLD Pulsed Laser Deposition

PVD Physical Vapour Deposition

RF Radio Frequency

Rs Sheet resistance

SEM Scanning Electron Microscopy

SOM Self Organising Map

SVD Singular Value Decomposition

SVM Support Vector Machines

SVS Scientific Vacuum System Ltd

TCO Transparent Conductive Oxides

UV Ultraviolet

VBM Valence Band Maximum

VH Hall Voltage

VIS Visible Light

WF Work Function

v
Table of Contents
Chapter 1 Introduction ...................................................................................................... 1
1.1 Overview ............................................................................................................................. 1
1.2 Aims & Objectives ................................................................................................................ 2
1.3 Structure of thesis ................................................................................................................ 3
Chapter 2 Literature Review .............................................................................................. 4
2.1 Transparent Conductive Oxides ............................................................................................ 4
2.1.1 Introduction ............................................................................................................................................ 4
2.1.2 Electrical Conductivity............................................................................................................................. 4
2.1.3 Optical Transparency ............................................................................................................................ 10
2.1.4 Hydrogen Doping .................................................................................................................................. 15
2.1.5 Fabrication methods ............................................................................................................................. 16
2.1.6 Conclusion............................................................................................................................................. 17
2.2 The Glow Discharge ............................................................................................................ 18
2.2.1 Brief History .......................................................................................................................................... 18
2.2.2 Introduction to glow discharge and applications.................................................................................. 19
2.2.3 DC glow discharge ................................................................................................................................. 21
2.2.4 Radio Frequency Discharges ................................................................................................................. 24
2.2.5 RF Magnetron Sputtering...................................................................................................................... 27
2.2.6 Sputtering Efficiency ............................................................................................................................. 32
2.2.7 Summary ............................................................................................................................................... 34

2.3 Characterisation and Modelling of Plasma .......................................................................... 35

2.3.1 Introduction .......................................................................................................................................... 35
2.3.2 Novel approach to characterising the plasma ...................................................................................... 37
2.3.3 Modelling Background .......................................................................................................................... 40
2.3.4 Modelling plasma with artificial neural networks. ............................................................................... 42
2.3.5 Summary ............................................................................................................................................... 44

Chapter 3 Experimental Methods and Characterisation Techniques ................................. 45

3.1 Experimental set-up ............................................................................................................ 45
3.1.1 Experiments for the optical analysis of the Glow Discharge via colour characterisation ..................... 46
3.1.2 Experiments for the synthesis of Highly conductive Zinc Oxide films................................................... 47
3.1.3 Experiments for investigating the effect of substrate position relative to plasma ............................... 48
3.1.4 Experiments for implementing Machine Learning concepts to the sputter deposition process .......... 50
3.2 Characterisation Techniques ............................................................................................... 62
3.2.1 Four point probe ................................................................................................................................... 62
3.2.2 Hall Effect .............................................................................................................................................. 63
3.2.3 Spectrophotometry............................................................................................................................... 66
3.2.4 Stylus Profilometer ............................................................................................................................... 67
3.2.5 Scanning Electron Microscopy (SEM).................................................................................................... 67
3.2.6 EDS ........................................................................................................................................................ 68
3.3 Summary ............................................................................................................................ 69

vi
Chapter 4 Optical Analysis of RF Sputtering Plasma through Colour Characterization ...... 71
4.1 Results ................................................................................................................................ 71
4.1.1 Chromaticity of the Plasma Emissions .................................................................................................. 71
4.1.2 Optical Emission Spectroscopy of the Plasma Emissions ...................................................................... 78

4.2 Discussion ........................................................................................................................... 83

4.3 Conclusion .......................................................................................................................... 85
Chapter 5 Highly Conductive Zinc Oxide Films Prepared Using RF Plasma Sputtering Under
Reducing Atmosphere ........................................................................................................... 86
5.1 Results and Discussion ........................................................................................................ 86
5.1.1 Film Morphology, Composition and Crystallinity Characteristics ......................................................... 87
5.1.2 Electrical Properties .............................................................................................................................. 89
5.1.3 UV-Vis Spectroscopy ............................................................................................................................. 94
5.1.4 Spectral and Colour Function Analysis of the Plasma ........................................................................... 96
5.2 Conclusion .......................................................................................................................... 99
Chapter 6 Investigation on substrate position relative to Plasma during the sputter
deposition of Hydrogen Doped Zinc Oxide ........................................................................... 101
6.1 Results and Discussion ...................................................................................................... 101
6.1.1 Electrical Properties ............................................................................................................................ 101
6.1.2 Film Thickness and Optical Properties ................................................................................................ 105
6.1.3 Analysis of the Plasma Emissions ........................................................................................................ 107
6.2 Conclusion ........................................................................................................................ 112
Chapter 7 Implementation of Machine Learning/Deep learning concepts: A step towards
the digitisation of the sputter deposition process ................................................................ 113
7.1 Results and Discussion ...................................................................................................... 113
7.1.1 Artificial Neural Networks to predict sheet resistance of IZO thin films deposited via Plasma
deposition ........................................................................................................................................................ 113
7.1.2 Implementing Supervised and Unsupervised Deep-Learning Methods to Predict Sputtering Plasma
Features ........................................................................................................................................................... 123
7.2 Conclusion ........................................................................................................................ 130
Chapter 8 Summary, Highlights and Future Work .......................................................... 132
8.1 Summary .......................................................................................................................... 132
8.2 Highlights of the Research ................................................................................................ 133
8.3 Future work ...................................................................................................................... 134
References…………………………………………………………………………………………………………………………135

vii
Chapter 1 Introduction

1.1 Overview
Transparent Conductive Oxides (TCO) are such type of semiconductors, which have attracted a
significant amount of attention due to their vast application areas. They are an essential part of various
optoelectronic devices, such as light emitting diodes (LEDs), thin-film solar cell modules, flat panel displays
and flexible electronics. Physical vapor deposition (sputtering) under vacuum is one of the proven
industrial methods of depositing thin films of these materials. During the sputter deposition of materials,
the sputtering conditions dictate the arrangement of the atoms on the substrate and will ultimately define
relevant functional parameters of the as deposited material such as bandgap. Researchers who apply this
technique for TCO preparation usually report their findings by stating the growth condition of the
sputtering process, such as chamber pressure, plasma power and the gas composition of the chamber
during deposition. However, the properties of TCO coatings will vary from one sputter system to another
depending on the dimensions of the chamber, size of the targets and a few other design related issues,
even if the sputtering conditions are maintained, according to an earlier report. As a result, several trial-
and-error experimental runs will most likely be carried out to refine a particular machine to produce a
specific desired thin film coating with certain features and functionality. This means that, to fully digitize
the sputter deposition process, significant constrains will arise. This project focused on studying the
spectral emissions of the plasma, using them to predict the properties of the thin films deposited as well
as its corresponding process parameters. Surely the plasma in the sputtering procedure is the core of the
reaction and fundamental to the deposition of the thin film. The diagnosis of laboratory plasma is typically
performed by optical emission spectroscopy (OES), through which a number of analytical techniques are
established to determine certain plasma properties, such as electron density, plasma temperature,
element recognition and qualification of elements present in the plasma. Such fundamental plasma
physics-related investigations are complex; extensive expertise, as well as time and effort, are required to
assess them.

One tool that can be exploited to digitize the sputter deposition process is Artificial
intelligence/Machine Learning. This has great potential to generate huge business value in the
semiconductor industry at all levels, from research & chip design to production through sales. The rise of
computers alongside Artificial Intelligence (AI) in recent years has had a major impact on science and
technology and ultimately changing lives and society. The latest advancement of computer technology
has resulted in the production of high speed computers with high performance graphics cards and
improved storage, bandwidth and data transaction speed; these have provided the tools and platforms
to develop, test and implement intelligent systems. A number of AI techniques have evolved over the past
decade to model the cognitive aspects of human behaviour; these include perception, reasoning,
communication and learning. The development of more advance technique including Artificial Neural
Networks (ANNs) and Evolutionary Algorithms (EA), which drew inspiration from the complex functions
of the brain and the evolution of biological life. Artificial intelligence techniques have significant potential
applications in various fields; including disaster recover, traffic control, space exploration, business
management and computer games. The ultimate goal of Artificial Intelligence is to bring about automation
that are capable of acting reliably on behalf of humans. Such systems would have great advantages in

1
domains where human intervention/errors needs to be minimised. Figure 1.1 shows the significant
traction gained in Artificial Intelligence in terms of publications and patents since its inception. The
number of patents in the AI field increased at an exponential rate, surging by an average of 28% a year
between 2012 and 2017 [1].

Figure 1.1. Growth of AI patent families and scientific publications.

1.2 Aims & Objectives

The aim of the project was to explore new methods of characterizing the plasma sputtering process
to evaluate the possibility of simplifying the monitoring and assessment of the sputtering process. The
first method involved monitoring the RF-based plasma sputtering process through optical spectroscopy
and characterizing the discharge based on its specific colour. The 2nd method of evaluation involved
employing an array of Machine Learning/Deep learning models to accurately predict certain
characteristics of thin film TCOs deposited via plasma-based sputtering and ultimately create a digital
shadow for the system. This project aimed to explore implementing an alternative approach in plasma
diagnostics during the sputter deposition process to assess the qualitative parameters of thin films as they
were deposited.
The objective is as follows:
• Collection of the emission spectra generated from a large number of plasma-based sputter
depositions
• Electrical and optical characterization of the resultant TCOs films which includes measuring
the sheet resistance, resistivity & transmittance.
• Employing several supervised and unsupervised Machine learning models created by Python
within a TensorFlow framework library which includes ANN, CNN and SOM.
• Using the spectral data associated with a deposition and its corresponding characterization
data (i.e. sheet resistance and resistivity) to train the array of machine learning models

2
 • Quantify the predictions of the electrical properties and growth conditions outputted by
deep learning models (not only for regressions but also classification analysis) by
implementing various loss functions.

1.3 Structure of thesis

Chapter 1 is the introduction to the thesis in which discusses the motivation, aim and objectives of
the research conducted. Chapter 2 provides a literature review & overview of the working principles/key
mechanisms of Transparent Conductive Oxides; it also presents some historical methods of characterising
& modelling the glow discharge. Furthermore, a novel and simple approach to characterising the
discharge was put forward. Following on from this, Chapter 2 presents a brief introduction of artificial
intelligence and key working principles associated with machine learning and more specifically deep
learning techniques and it can be implemented to simplify the evaluation of the glow discharge. Chapter
3 discusses the experimental set-up that was utilised for the fabrication of various TCOs for this research
project and the array of optical and electrical characterisation techniques used to characterise the thin
film samples. Chapter 4 explores deeper into the study of the characteristics of plasma in vacuum-based
deposition systems from a unique perspective; its colour. the stability of the emission characteristics of
an RF plasma via various statistical analysis methods is investigated. Chapter 5 presents the preparation
of highly conductive zinc oxide thin films without indium or aluminium doping and as such eliminating the
need for the expensive indium addition. Chapter 6 investigates the complexity of the reactive sputtering
of highly conductive zinc oxide thin films in the presence of hydrogen at room temperature. This chapter
reports on the importance of precise geometric positioning of the substrate with respect to the
magnetron to achieve maximum conductivity. Chapter 7 demonstrates the implementation of deep
learning models to examine the accuracy of predicting a single feature viz. sheet resistance of thin films
of indium-doped zinc oxide deposited via plasma sputter deposition by feeding the spectral data of the
plasma to the deep learning models. Chapter 7 also looks at the conversion of the spectral emission data
from the plasma glow of various sputtering targets containing Indium Oxide, Zinc Oxide and Tin Oxide into
two dimensional arrays; this is achieved by implementing basic array reshaping technique and a more
complex procedure of utilizing an unsupervised deep learning technique. Chapter 8 concludes the
research project and summarises the findings in each set of experiments and also presents the work
currently being carried out on the implementation of deep learning techniques on further deposition
processes such as plasma-based sputtering of indium doped tin oxide (ITO) to further demonstrate the
power of artificial intelligence.

3
Chapter 2 Literature Review

2.1 Transparent Conductive Oxides

2.1.1 Introduction
As mentioned, Transparent conducting oxide (TCO) thin films are applied in various optoelectronic
devices, mobile phone screens, flat panel displays, light emitting diodes and most importantly in the
fabrication of thin film-based solar cell devices, and plasma sputter deposition is one of the main
established methods of applying these coatings as reported in various studies [2-8]. The nature of these
materials brings about adjustments of dual functionality parameters viz. conductivity and transparency,
which requires complex efforts and understanding for the optimization of various sputtering conditions.
However, these characteristics are directly related to the complex electronic band configuration of these
materials and have been a subject of great attention worldwide [9]. A band gap of above 3.3 eV and
electrical conductivity in the range of 104 S·cm−1 (Siemens per centimetre) are the fundamental aspects of
such materials. Various oxides behave as TCOs among which indium tin oxide (ITO) is the dominant
commercial type [10]. Another TCO material of choice is indium zinc oxide (IZO), with applications such as
transparent electrodes in heterojunction thin film solar cell structures [11-15]. One of the TCO materials
that has recently gained interest is zinc doped indium oxide (IZO) due to the robust nature of its resistance
against moisture. The parasitic absorption in TCOs as the front window electrode in solar cell
configurations such as silicon heterojunctions is a leading factor in undermining cell efficiency. This can
be resolved by the application of TCO materials that demonstrate high mobility irrespective of their charge
density [2]. Due to the presence of multiple layers in solar cell designs such as CIGS (Copper Indium Gallium
di-Selenide, CuInGaSe2) and perovskite devices [3,4], which are sensitive to ambient moisture and
temperature, deposition of TCO materials at low temperatures is highly desirable, and hence, IZO stands
as a potential TCO material for these applications.

2.1.2 Electrical Conductivity

To gain a grasp of the conductivities of the TCOs, a comparison between some of the materials in terms
of carrier concentration and mobility. The electrical conductivity is a product of carrier concentration n,
elementary electric charge e, and mobility µ :

𝜎 = 𝜂𝑒𝜇 (2.1)

A typical metal such as Ag possesses a high electrical conductivity because of the high carrier
concentration in the order ~1023 cm-3 but interestingly the mobility does not stand out among the
conductors like p-doped Si despite the low carrier concentration of the latter. Even if Si is doped heavily
with carrier concentrations upwards of 1019 cm-3, the conductivity won’t be as high has an Ag metal
because of the reduction in mobility by carrier scattering. ITO displays conductivity that is roughly 1/10th
of that of Ag because of its relatively smaller carrier concentration despite possessing similar carrier
mobilities [16]. Figure 2.1 shows a plot of various conducting materials and how their carrier
concentrations and mobilities differ from one another. Metals as a whole are found in the top left region
indicating extremely high carrier concentrations but relatively low mobilities. In contrast, typical doped

4
semiconductors are plotted in the bottom right area due to its low carrier concentration and high mobility.
It can be seen that TCOs lie somewhere in between metals and doped semiconductors as their mobilities
is not too dissimilar from those of typical metals such as Ag. TCOs are essentially metals with a small
carrier concentration and this type of material can be described as degenerate semiconductor. These
types of materials are independent of temperatures for the generation of charge carriers i.e. the thermal
activation energy is not necessary [16]. Degeneracy of a semiconductors simply implies that there’s is an
overlap of energy bands and the Fermi energy is close to the edge of the conduction band.

Figure 2.1. Carrier density and mobility for some metals, semiconductors and TCOs [16]

Doping of a material increases the number of charge carriers and in turn the conductivity. This can
be achieved by substitutional doping i.e. creating vacancies and implanting interstitial. Though an
enhancement in the mobility can boost the conductivity, it depends on intrinsic scattering mechanism
which act as a limitation as the carrier density increases, thus mobility it is an important parameter that
has an influence on the total conductivity.

Figure 2.2 shows the typical electronic structure of metal oxide (MO) illustrated in the form of
molecular orbitals and energy bands. An essential feature of TCOs is the low absorption in the visible
range; this can be attributed to the minimum bandgap of 3.0 eV. The top of the valence band mostly
comprised of oxygen 2p bands; the bottom of the conduction band is occupied by a single and highly
dispersed metal band. The conduction band possess antibonding features and the bottom of the valence
band has bonding features as counterpart. As seen in the figure, top of the valence band possess non-
bonding features as the non-bonding oxygen 2p orbitals possess low energy and remain without changing
the energies thus a large bandgap can be obtained in oxides.

5
Figure 2.2 Schematic molecular orbital diagram and band structure of oxides [16]

In intrinsic stoichiometric oxides, it cannot exhibit high level of conductivity as well as transparency.
However, Substitutional doping via cationic donors or anion vacancies lead to the generation of charge
carriers altering the electronic band structure of the material. To improve the donor density, the donor
states combine with the conduction band at the critical density which can be determined by Mott’s
criterion as expressed by Eq. 2.2 [17]:

"#/%
𝑛! < 𝑐𝑎&∗ (2.2)

Where 𝑎&∗ is the effective Bohr radius and c is a constant. The value of c is ≈ 0.25. If this criterion is
satisfied (i.e., if the density of electrons is sufficiently high) the material becomes conductive and free
electron behaviour is observed and the material is said to be degenerate. The Fermi Energy Ef is then
determined by the high occupied state in the conduction band and can be expressed as:

ℏ! # !
𝐸! = $%"∗
(2.3)

The band structure of Tin oxide can be calculated by the parabolic functions of k in the near the edges
of the bandgap. Figure 2.3 shows the schematic diagram of the band structure for most wide bandgap
metal oxide semiconductors. These types of materials can be described as direct band semiconductors as
the valence band maximum and conduction band minimum can both be found at k=0.

6
Figure 2.3. Schematic diagram of the band structure of (a) undoped and (b) doped wide band gap metal oxide
semiconductor near the top of the valence band and bottom of the conduction band. Grey area represents occupied
states [18]

Doping introduces states in the bandgap either just below the conduction band or above the valence
band. This is depicted in figure 2.4 whereby the band structure, density of states, fermi distribution of
electron energy levels and concentration of majority charge carriers for intrinsic, n and p type materials
are shown. At 0 oK, the Fermi level of an intrinsic material is found in the middle of the bandgap (figure
2.4a). There is high probability of finding an electron in the states of the valence band; in contrast the
chance of finding an electron in the conduction band states is 0. There is 50% chance of finding an electron
at the Fermi level. For doped semiconductors, there will be a shift in the Fermi level as doping will have
an impact on the loss or gain of electrons. For n type materials (figure 2.4b), the impurity introduces donor
levels just below the conduction band whereby electrons are the majority carriers. For p-type, impurities
possess energy levels just above the valence band whereby electrons are thermally excited into these
acceptor levels leaving behind positively charge vacancies in the lattices thus the majority carriers in this
situation will be holes. Upon exposure to light, the generated charge carriers trigger a shift in free energy
and thus perturbating the equilibrium in the semiconductor.

7
Figure 2.4. Schematics depicting the band structure, density of states, Fermi-Dirac distribution of energy levels (for a
non-zero temperature), and the charge carrier concentrations for (a) intrinsic, (b) n-type, and (c) p-type
semiconductors. [19]

Defects will also have an impact on the conductivity. Both defects and doping have the same effect
on transparent insulators such as SnO2, CdO, In2O3 and ZnO. In addition to doping, the addition of vacant
oxygen sites also contribute towards the conductivity of these materials. Figure 2.5a shows the doping
sites occupied by tin (Sn) in an indium oxide (In2O3) lattice [20]. The occupation of the interstitial site by
an Sn atom results in donating an electron widening the bandgap and ultimately resulting in the synthesis
of Indium Tin Oxide (ITO). The bandgap structure of undoped and Sn doped Indium Oxide is illustrated in
figure 2.5b. Occupied states are just below the conduction band typically found in an n-type
semiconductors. The free electrons in the conduction band come from the oxygen vacancies [20]. Thus,
ITO are highly degenerate n-type semiconductors.

In the band model proposed by Fan [21] (i.e., the reference model used to date) the Indium oxide
possesses a direct wide bandgap, leading to the prevention of inter-band transitions in the visible range
allowing for high transparency. As mentioned previously, conduction band stems mainly from Indium 5s
orbitals and the valence band from the oxygen 2p electrons. Due to the n-type doping of the Tin
impurities, the Fermi energy is located a few eV below the conduction band [22]. For low doping density,
the formation of donor states occur just below the conduction band and the Fermi energy lies between
the donor level and the conduction band minimum. For high doping levels there is an increase in donor
density [23], and the donor states merges with the conduction band at the previously mentioned critical
density nc; for ITO this is ~3.43x1019cm-3.

8
Figure 2.5. (a) Sn doping sites in an In2O3 lattice. (b) Schematic energy-band model for tin doped indium oxide, where
Ef, Ev and Ec is the Fermi energy, the valence band energy and the conduction band energy respectively. Left: low
doping level, right: high doping level.

Scattering mechanisms

Mobility can be thought of as the ease in which charge carriers move through the material. It is
defined in terms of the average scattering time (𝜏) and the carrier effective mass (𝑚∗ ) as:

&'
𝜇 = %∗ (2.4)

Equation 2.4 shows that the mobility increases with decreasing scattering time and/or a reduction
in effective mass. In general, the inhibition of the carrier movement will result in a reduction in mobility
and conductivity. In crystalline TCOs, there are a number of scattering mechanisms that take place.
According to Matthiessen’s rule [24], resistivity arises from independent scattering processes which
are additive, such as scattering due to impurities, defects, and grain boundaries which is expressed in
equation 2.5 :
( ( ( (
=) + +⋯ (2.5)
)$%$&' ()*+,($- )./0/"$ )12

Ionised impurity scattering has the most impact on these structures. TCOs deposited at lower
temperatures exhibit lower crystallinity and high doping concentrations results in the formation of neutral
complexes. In such cases, grain boundary scattering and neutral impurity scattering should also be
accounted for.

At high conducting electron density, the carrier transport is limited mainly by In ionised impurity
scattering. i.e., the interaction between the ionised impurities and free electrons is said to be coulombic
which provide an intrinsic source of scattering to the doped structure. With increasing dopant
concentration, the resistivity reaches a lower limit it is not able to surpass whereas the optical window

9
gets narrower. The mobility and hence the resistivity of TCOs (ITO, SnO2, ZnO) are limited by ionised
impurity scattering for carrier concentrations above 1020cm-3 [25]. Above the critical Mott density
threshold, the conduction band is inhabited by impurities and associated electrons leading to the
formation of degenerate electron gas which in turn increases the ionized-impurity scattering and
ultimately lowering the mobility [23]. As discussed, the scattering that arises from the homogenous
distribution of the ionised dopant atoms is one of the possible effects that reduce mobility.

In grain boundary scattering, there is a presence of atomic layers of disordered atoms creating a large
number of defects. This leads to the entrapment of charge carriers i.e., immobilisation. The traps become
electrically charged which leads to the formulation of a potential energy barrier thus reducing the mobility
of free carriers from moving one crystallite to another. The potential barrier for electrons is determined
by its height. [26].

Resistivity well
As mentioned, Introduction of defect sites in solids also allow for the scattering of electrons; thus one
needs to find a balance between the increased in conduction and scattering by defect sites. As a result
there is a conductivity maximum or a resistivity minimum for TCOs; this ‘resistivity well’.It has been
reported that there is a correlation between the apparent as-deposited colour of ITO films and the film
resistivity and O2 and Sn doping concentration [23]. For low doping concentrations, the resistivity is high
due to the low carrier concentrations and low for high doping due to scattering from oxygen vacancies
which reduces the mobility. This resistivity well trend was also observed as function of the sputtering
power density. Increase in the power density generates ionised impurity centres inducing the scattering
of free electron carriers and leading to reduction of mobility. The gradual increase in carrier concentration
between power density of 1 and 3 W/cm3 is due to the increase in the quantity of substituted Sn ions in
the lattice. Further rise in sputtering power results in the carrier concentration plateauing which is due to
the uptake of Sn ion substitution and reduction oxygen vacancies from the ITO film crystallinity.

Many studies have reported that there is a strong correlation between resistivity and
annealing/deposition temperatures displaying similar curve trends as that of the power density which can
also be explained by the same mechanism and as such undergoes similar physical transformation. In
addition, varying the oxygen flow also presents the resistivity well effect; with relatively low oxygen
content, high carrier densities are observed and vice versa, high mobility is observed with increasing
oxygen content due to the enhanced crystallisation. Since both mechanisms oppose one another in terms
effects, a minimum resistivity can be seen with varying levels of oxygen flow. To summarise, the resistivity
and physical appearance of the ITO films displays trends as a function of a number of parameters such as
doping content, deposition power, substrate temperature and oxygen flow.

2.1.3 Optical Transparency

There are some metal oxides that are electrically insulating but optically transparent in the 380 to 780nm
wavelength range more specifically in the visible region; MgO and Al2O3 are such examples. Thus, this
property makes them ideal for fulfilling the role of a window in optical applications. The optical
transparent range of various materials can be seen in figure 2.7.

10
Figure 2.7. Optical transparent regions in a rage of materials [16]

Typical oxide-based compounds exhibit high level of transparency in the 300nm to 1200nm
wavelength range which includes the visible region. Fluorides such as LiF and CaF2 possess a significantly
larger window range than oxide particularly in the deep ultraviolet region. In Contrast, the window range
of conventional semiconductors such as Si and Ge can only be found in the IR region. Therefore, it is
expected for TCOs to exhibit transparent properties in the visible range. Figure 2.8 illustrates typical
transmission spectrum of a single crystal oxide.

Figure 2.8. Transmission spectrum of a typical oxide [16]

As can be seen in figure 2.8, the absorption edge is dictated by the vibration in the lattice in the IR
region; Furthermore, the electron excitation from the valence band to the conduction band results in the

11
fundamental absorption in the ultraviolet region. Optical absorption and reflection are impacted by
introduction of new carriers into the parent material of TCOs. The collective motion of the carriers exhibit
plasma-like properties in conductors. Upon the irradiation of an electromagnetic wave, the carriers
oscillate at the frequency of the light which is referred to as plasma oscillation; this leads to light being
reflected at the surface of the conductors. However, upon exposure to light at higher frequency, the
charge carriers cannot match the fast electric field oscillation of the light and as a result, leads to the
transmission of light through the conductors without causing plasma oscillation/reflection at the surface.
The maximum threshold frequency that permits the oscillation of the plasma is known as the plasma
frequency, 𝜔( and can be expressed by equation 2.6 [27]. The plasma frequency then can be converted
the wavelength as shown in equation 2.7:

+& !
𝜔* = , (2.6)
3%

$-. ,3 %
𝜆* = ) (2.7)
& +

Where n, m and 𝜀& and c represents the carrier concentration, electron rest mass, permittivity, and
speed of light in a vacuum respectively. For a metal with high carrier concentration such as Al (>1022 cm-
3
), the threshold wavelength can be found in the vacuum ultraviolet region and the reflection of visible
light takes place at the surface [28]. On the other hand, the carrier concentration of ITO is one order of
magnitude small than Al metal, the threshold wavelength shifts to the IR region of the spectrum, which
leads to transparency in the visible region as ITO possesses a large enough bandgap (>3.5eV) to exhibit
such behavior. As the carriers make the optical window of oxides narrower in the IR region, the maximum
carrier of TCOs should be less than 1021cm-3. Furthermore, the carriers also have an influence of the optical
window in the ultraviolet region as well. As illustrated in figure 2.10, the bandgap can be calculated by
plotting (hva)2 against the energy in eV whereby, h, v, and a are Plank’s constant, frequency and
absorption co-efficient respectively. With increasing carrier concentration, there is a shift of absorption
edges towards the higher side of energy, and a subsequent increase the bandgap ultimately widening the
optical window in the UV region. This is associated with the occupation of the conduction band by the
carrier electrons as displayed in the schematic inset in figure 2.10.

12
Figure 2.10: Blue shift of fundamental absorption by band filling [16]

Burnstein-Moss shift

To discuss, develop and engineer the making of these materials it is vital that the mechanism by
which these materials demonstrate transparency and conductivity is well-understood and can be
repeated during experimental trials. In n-type TCOs, the bandgap for the optical transition from the
valence to the unoccupied conduction band becomes bigger than the original bandgap Eg due to carrier
electrons occupying the bottom of the latter. This phenomenon is known as the Bernstein moss (BM shift)
as illustrated in figure 2.11 which in simple terms refers to widening the optical bandgap of the material

13
Figure 2.11. Optical widening by the effect of the Moss-burstein shift

The Fermi level, work function (WF), ionization energy (IE), energy gap (Eg), and electron affinity (EA)
are fundamental parameters that need to be addressed when discussing TCOs [29-30]. These parameters
define the electronic interface between the TCO surface and other materials and control the charge
exchange and transport across these materials. Figure 2.12 illustrates a simple energy diagram of a
semiconductor exhibiting the above-mentioned parameters.

Figure 2.12. (a) A simple flat band representation of the energy band at the surface of a semiconductor (b) the
schematic diagram showing the optical widening of the TCO bandgap using the Moss-Burstein shift, EgMB.

As electrons are naturally bound to the solid, they are prevented from escaping outside of the
material into the vacuum via an energy barrier that culminates at the vacuum level (Evac). Thus, Evac is the

14
energy level of an electron resting within close vicinity outside the solid. At this vicinity, the electron is
experiencing the full impact of the surface dipole. To get an electron from the VBM into the Evac region, a
certain amount of energy is required, and this is referred to as the ionization energy, IE hence when
thinking of n-type doping of a TCO material or a semiconductor, the ideal doping material should have
low EI. If an electron drops from the Evac region into the CBM level, it will release some energy, and this is
referred to as the electron affinity (EA) of the material [30]. The work function (Φ) is defined as the energy
required to take an electron from the Fermi level and place it in the Evac region. In a semiconductor, the Φ
is dependable on the Evac, and the Fermi level (EF) is governed by the doping concentration, density of
states, carrier density and temperature.

The parameters discussed in this section fundamentally depend on the technique by which the
material is produced and prepared. It is ultimately the arrangements of the atoms next to each other that
defines the final material’s composition and properties. Hence, the preparation technique and operation
method of arranging the constituting atoms of a TCO will govern the quality of the produced transparent
conductive film.

2.1.4 Hydrogen Doping

While ITO is the most widely used TCOs, Zinc Oxide and related compositions are an alternative and,
in some cases, the preferred materials. A big proponent of the deployment AZO is that it combats the
global shortage of Indium. AZO is extensively being employed as TCOs in thin film solar cells such as CIGS
and CZTS, in addition to its use as electrical contacts for laser diodes and LEDs. The n-type semiconductor
possesses a direct bandgap of 3.37 at room temperature. The high conductivity of the thin film material
was initially debated to be due to the oxygen vacancies and/or Zn interstitials [31], however it has been
proposed that the unintentional substitutional Hydrogen impurities are the primary driving factor for the
high conductivity [32]. ZnO films are usually doped with Al (AZO), Gd (GZO), excess Zn, or In to obtain
highly conductive and transparent compounds [33,34]. This material is also often alloyed with other TCO
materials and can be a potential alternative as it has the cheapest and most abundant atomic constituents
with no significant toxicity associated with elements such as cadmium or lead. However, pure zinc oxide
does not possess the required electrical properties to match those of IZO or ITO. Doping zinc oxide with
hydrogen is a cheaper option that can significantly improve the conductivity of this material, and this was
initially observed in 1950s [35,36].

The mechanism by which hydrogen incorporation into the zinc oxide lattice promotes n-type
conductivity was theoretically investigated by van de Walle by first principal calculations based on density
functional theory (DFT) [37]. The behaviour of interstitial hydrogen in zinc oxide was studied by van de
Walle, and the formation energy of H+, H−, H0 and H2 states in bonding centre and anti-bonding locations
were evaluated, where H+ demonstrated the lowest formation energy. The study showed that with
incorporation of the hydrogen, an O–H bond forms along the direction of the original O–Zn bond. The
strength of the O–H bond is the main driving force for stabilisation of the low energy H+ configuration. van
de Walle concluded that H+ is the stable charge state for all Fermi level positions in ZnO and acts as a
donor [38]. The formation energy of the H+ was shown to be low enough to allow for large solubility of
hydrogen in ZnO. It is interesting to note that hydrogen is amphoteric in other semiconductors, although
not in ZnO. This difference is due to the strong O–H bond, which lowers the formation energy of H+.
Parallel to this, a complex consisting of an oxygen vacancy and a hydrogen atom acts as a shallow donor.

15
Such vacancies are low energy defects [38] and neutral, but the hydrogen turns them into a shallow donor,
and as such, a hydrogen located close to the centre of such oxygen vacancies acts as a substitutional
impurity [37]. These theoretical observations were then verified by electron nuclear double resonance
and Muon spin spectroscopy [39,40]. The Fermi energy level where the positive and negative charge
states are equal in energy occurs above the conduction band minimum in ZnO [41]. Thus, overall it can be
seen that in a ZnO system, hydrogen acts as donor atom as H+, providing the lattice with an extra electron
as a charge carrier, which in turn makes ZnO an n-type material and influences the conductivity of the
material.

However, apart from carrier density, the mobility of the charge carriers is an even more important
feature of a potential conductor. Mobility is the measure of the velocity of carriers under an electric field
and is inversely proportional to the effective mass of the charge carrier. The effective mass is inversely
proportional to the second derivative of energy with respect to the k-vector in an E–k plot indicating how
the electron states are spaced in k-space. E-k plot is the curvature of the energy band in a way that the
larger curvature of the energy band translates to a smaller effective mass of the charge carrier [42]. The
energy value computed in the E–k plot is governed by the orbital overlap of adjacent atoms, where the
larger the orbital overlap is, the larger will be the drift of the charge carrier.

Parallel to the electrical properties of hydrogen-doped zinc oxide, several reported studies indicate
doping of ZnO with hydrogen gives ferromagnetic properties to ZnO [43-45].

Hydrogenated ZnO nanoparticles are reported to demonstrate ferromagnetism because due to low
formation energies, the formation of Zn vacancies and OH bonding by hydrogen is favoured in the
hydrogenation process, leading to a magnetic moment of ~0.57 µB (Bohr magneton), and the origin of
ferromagnetism is explained through the hydrogenation of the ZnO polar surface [46,47].

There are some experimental reports on electrical conductivity improvements achieved in the ZnO
system with the addition of hydrogen with respect to TCO materials. However, such reports only exist at
the research state, and no commercial TCO based on pure ZnO doped with hydrogen is present in the
market as far as literature is concerned. This work was on enhancing the electrical properties of IZO and
ZnO systems by repeated exposure of the sputtering target to an argon/hydrogen mixture as the working
gas [48] which will be presented in chapter 6. In one of the set of experiments interesting complexities
associated with reactive hydrogen/argon sputtering of un-doped ZnO targets was obersved, which is
believed to be associated with the reported magnetic properties of hydrogenated ZnO particles, and this
report outlines the mentioned observations.

2.1.5 Fabrication methods

In order to optimize the performance and the cost-effectiveness of TCO films, a wide array of
synthesis techniques have been developed over the years. Among these are chemical vapor deposition
(CVD), sputtering techniques (DC, RF, magnetron), evaporation (reactive, thermal or e-beam), spray
pyrolysis and pulsed laser deposition (PLD). Certain synthetic routes are chosen for TCO manufacturing
due to their thickness uniformity, low production costs or high throughput. Each fabrication technique
can have an influence on the intrinsic behaviour differently including structure morphology and
composition of the thin film which in turn dictates the electrical and optical properties. Physical Vapor
Deposition (PVD) techniques, such as Magnetron Sputtering (MS) or Pulsed Laser Deposition (PLD) allow

16
for the synthesis of TCOs with extremely low resistivity and sheet resistance (i.e. specific resistivity per
unit thickness) in the most common TCOs. Sputter deposition has been favoured over other techniques
in certain instances as it’s able to meet industrial requirements such as fast deposition and high
throughput. Other advantages are the low complications in the growth of films, possibility of growing films
on relatively large substrates, and reproducibility [49,50]. In the case of ITO, the best available sputtered
thin films commonly have resistivities slightly lower than 10-4 Ω cm, with sheet resistances typically smaller
than 10 Ω/ square [50]. With PLD even lower resistivity values can be obtained (down to 7 · 10 Ω-5 cm [52])
with the drawback of decreasing the deposited area [53].

2.1.6 Conclusion

In summary, the possibility of a semiconductor being both transparent and conductive is due its large
bandgap so that it can avoid absorbing visible light (~3eV). The gap is located between parabolic oxygen
2p and metal s bands, resulting in the formation of the valence and conduction band respectively. Free
charge carriers are generated by introducing dopants which are responsible for the conductivity. Electron
densities are generally in the order of 1020cm-3 for n-type TCOs yielding degenerate electron systems.
Scattering mechanisms dominate the electrical conductivity which are strongly related to the electron
concentrations. The mobility is influenced by the total contribution of all electron scattering which
deterring the upper limit of conductivity. The ionised impurity scattering is considered to have the biggest
impact for single crystalline materials. Conversely, as the structural nature of the thin film decreases,
scattering via grain boundaries become more significant in polycrystalline films. The transparency window
for TCOs is exhibited in the UV and near IR region. The bandgap determines the window in the short
wavelength region and the plasma frequency dictates the window in the long wavelength region. Both
values depend on the electron concentration and thus can be tuned to use in specific applications. The
electrical conductivity of typical n-type TCOs is around 104 whilst display transmission as high as 90% in
the visible region. The conductivity of TCOs of a p-type nature is generally lower than the n-type
counterpart by at least a factor of 1000. Although the figure of merit is a measure of the TCO’s electrical
and optical performance, the thin film TCOs are engineered to possess properties tailored for specific
applications. There are a wide variety of deposition techniques for these types of materials, sputtering
being widely utilised to yield highly transparent and conductive thin film TCOs. Each type of growth will
have its own unique influence on the intrinsic properties as thin film structure and composition. It is these
properties (not the deposition method itself) that will have an emphasis on the electrical and optical
performance of TCOs.

Plasma sputter deposition technique does not behave like a standard chemical reaction or a process
with defined steps. For example, researchers who apply this technique for TCO preparation usually report
their findings by stating the condition of the sputtering process, such as chamber pressure, plasma power
and the gas composition of the chamber during deposition. However, from one sputter machine to
another, depending on the dimensions of the chamber, the size of the targets, and a few other design
related issues, TCO coatings with different properties are obtained, even if the sputtering conditions are
maintained according to an earlier report. The sputter deposition process will be discussed in greater
detail in the next section (2.2) from the constituents of a plasma discharge to the different types of
discharge used to power the plasma.

17
2.2 The Glow Discharge

2.2.1 Brief History

The term ‘plasma’ was first introduced by Irving Langmuir in the 1920s [54]. Many physicists studied
the behaviour of DC electrical discharges in gasses at low pressure during the 19th century [55]. Various
experiments concerning tubes under vacuum were conducted to portray the strange behaviour of the
plasma.

Applying an electric or magnetic field to the plasma leads to the preferential hearing of charged
particles such as electrons and ions keeping the neutral gas atoms at low temperatures. Electrons with
low mass in particular are easily accelerated to ionisation energies of the gaseous atoms; the kinetic
energies of these electrons range from 1-10eV (equivalent to ~104-105 K). Due to the low gas temperatures
of the plasma discharge, it is often called cold plasma. Plasmas of this nature are not in thermal
equilibrium since [56]:

𝑇) ≫ 𝑇*+, & 𝑇) > 𝑇. > 𝑇* (2.8)

Whereby Te, Tg and Ti are temperatures of the electrons, neutral gaseous atoms and ions respectively.

DC and RF glow discharges are two of the most common types of electrically induced plasma where
the electric field of the latter is coupled with either capacitively or inductively to the electrons of the
discharge. [56]

The scientific evaluation of plasma discharges was explored in the latter half of the 19th century which
paved the way for the advancement made in vacuum technology. It was observed that when air was
pumped out of glass vessels in an high potential environment, the high voltage electric spark between
two electrodes created a variety of colourful ribbon like discharges as the gas pressure was lowered in a
progressive manner. In 1994, Faraday discovered the formation of a dark space near the anode of the
discharge tube as the gas pressure of was reduced demonstrating the presence of dark discharge of
electricity in gases [55].

Technological advances in quartz/glass tube construction and progression of electrode material

played a vital part in the investigation of the various types of the glow discharges [56]. In mid 19th century,
Geissler came up with a technique for the assembly of glass discharge tubes with metal electrodes and
introducing a mercury pump. Plucker noticed that effect of phosphoresce in the glass with the reduction
of gas pressure. As a result, multiple investigations were conducted to identify the root cause of this effect.
The use of X-ray was explored, furthermore the measurement of the electron charge to mass ratio was
made operating discharge tubes at low gas pressures (10−2 to 10-3mmHg). J.J. Thomson and F.W Aston
discovered that the ‘Positively charged rays’ could be passed from a glow discharge through a hole in the
cathode which led to the development of the mass spectrometry of ions [55]. The in-depth research work
conducted on plasma glow discharges played a pivotal role in the development of modern physics [56].

18
2.2.2 Introduction to glow discharge and applications
The term ‘glow discharge’ is derived from the plasma’s luminosity. This is produced because electron
surpass the threshold energy and density threshold to generate a visible light via recombination and
excitation collisions [55]. Glow discharges can be found in a wide variety of applications such as gas
fluorescence lamps, neon lights and flat panel plasma screen displays. Aside from the light industry, its
use can be found in the microelectronic industry and the material processing technology. Such use include
treatment and etching of surfaces (for the fabrication of integrated circuits (IC) etc), modification of
polymer and the deposition of thin film coatings. Other forms of glow discharge such as DC parallel plate
plasma reactors, electron bombardment plasma source, are also employed in industrial applications.

The glow discharge is fundamentally produced by applying a potential difference (ranging from 1V to
few kV) between two electrodes in vessel or chamber filled win an inert gas under reduced pressure (from
a few mTorr to atmospheric pressure) [57]. The electrons are then accelerated away from the cathode
and the collision with the gas atoms and electrons goes up as a result of the potential difference. The
collisions may lead to processes such as excitation, ionisation and dissociation. Excited species generated
from collisions can get demoted to lower energy levels by the emissions of light which is responsible for
the ‘glow’ in the discharge. Secondary electrons are released when ions accelerate towards the cathode
and bombarde its surface. These types of electrons are then accelerated away from the cathode surface
giving rise to more ionisation collisions and subsequently yielding ion-electron pairs; the ion-electron
multiplication process enables the plasma discharge to sustain itself [57].

Different types of glow discharge.

There are a range of glow discharges the most basic type being the direct current (DC) glow discharge.
A constant current is produced by applying a potential different between the cathode and anode. This
could lead to issues if one of the electrodes in non-conducting such as a build up of charge and resulting
burn-out of the glow discharge [57]. A capacitively coupled radio-frequency (cc rf) can be applied to
overcome by this issue i.e. applying an alternating voltage between the two electrodes. The accumulation
of the charge during one half cycle is neutralised by the opposite charge accumulated during other half
cycle giving rise to an alternating current glow discharge. This can be described as short discharges being
struck consecutively whereby the cathode and anode constantly change roles. An important feature of
alternating current (AC) glow discharge operating at atmospheric pressure is the dielectric barrier
discharge (DBD) i.e. the electrodes are typically coated by a dielectric barrier [57].

Another type of glow discharge is one that is sourced with a pulsed current. The pulsed glow
discharge consists of intermittent glow discharges (in the milli/microsecond range) which is then followed
by an afterglow which is typically characterised by a lengthy time period in comparison. This type of
discharge is advantageous because a high peak electrical power can be attained for a low average power;
this results in high peak efficiencies for various applications [57].

The magnetron discharge is a well-known type of discharge in which a magnetic field is applied
whereby the electrons are subjected to helical circulation around the magnetic field lines resulting in more

19
ionisation. This type of discharge is typically generated in low pressure high current environments in
comparison to conventional discharges.

Several new types of glow discharge have emerged whereby they are characterised by low pressure
and high plasma densities. The key difference lies in the application of the electrical power; it is applied
by electromagnetic induction and not by applying a potential difference between two electrodes that’s
found in the conventional glow discharges. This is known as inductively coupled discharge whereby, as
the name suggests, there is an inductive coupling between the RF power and plasma. The other type of
plasma discharge is struck with microwave power whereby the microwave discharge and magnetic field
gives rise to ECR conditions [55], the operating principle of which will be briefly touched on.

Plasma sheath
Deposition system used in this project confine the plasma within a vacuum system resulting in the
plasma species interacting with elements contained in the system. As electrons have higher thermal
velocities than ions [59], their interaction with the walls occur faster than that of the ions. This leads to
the development of a negative bias at the chamber walls attracting ions that give rise to a potential
difference; this phenomena is known as plasma sheath (figure 2.13)

Figure 2.13 Interactions of charged particles with chamber wall giving rise to a plasma sheath

The sheath is in a dynamic balance; recombination of ions at the chamber surface occur as well as
further surface recombination by electrons possessing high enough energy to overcome the energy
barrier of the plasma sheath. Debye shielding prevents the potential to be distributed throughout the
plasma, so the sheath confines the more mobile species within the plasma [60]. The potential difference
of the sheath region can be calculated via the following equation 2.9:

𝑉!" = 𝑉# − 𝑉$ (2.9)

Where Vsh is the potential of the sheath, Vp is the plasma potential which oscillates with the waveform
for RF and/or microwave plasmas. Vf is the floating potential which is the value held by an isolated
substrate within the plasma. Bombardment of the substrate by the ionised growth species and carrier gas
electrons take place at the sheath potential. This is due to the potential of the substrate attaining a
dynamic equilibrium with equal flux of both positive and negative species as a result of its isolation. There

20
is a presence of sheath conditions in all surfaces of a plasma system. While the surfaces will be grounded
(i.e. 0V potential difference), it is possible to floating potential across desired components of the system
such as the sample holder, a technique which was not explored in this project.

2.2.3 DC glow discharge

The glow discharge via DC has been historically paramount to the evaluation of the properties of the
plasma medium and application. The typical setup consisted of a long glass cylinder with an anode and
cathode electrode on either side with an operating pressure ranging from 10mTorr to 10 Torr; a typical
voltage of a few hundred volts was found between the two electrodes maintain the discharge. A summary
of the range of parameters used for the glow discharge can be found in table.

Figure 2.14. illustrates the I-V characteristics of the DC low pressure electrical discharge tube. It can
be seen that the current across the Townsend discharge region soars up with an increase in voltage
through the dark discharge regions. When the breakdown voltage at point E is reached, the gas will
breakdown at voltage Vb provided that the internal resistance of the power supply is relatively low. The
discharge will then transition into the normal glow discharge.

Figure 2.14: Typical V/I plot of a glow discharge including key characteristics such as breakdown voltage, the voltage
current characteristic and the structure of the discharge depend on the geometry of the electrodes, the gas used, the
pressure and the electrode material [58]

The normal glow discharge region is between F and G whereby the voltage is somewhat independent
of the total current passing through the discharge tube; the current density approaching the electrode
neither depends on the total current. This essentially means that in the normal glow discharge area, there
is only small point of constant between the plasma and the cathode surface at low current. As the current
increases, the total cross section of the cathode is occupied by the plasma-surface contact; this occurs
until point G, the boundary of the abnormal glow discharge is reached whereby the plasma fully covers

21
the cathode surface, to deliver the total amount of current required at a constant current density. As the
total current increases, the voltage also rises so that the cathode current density it above its setpoint and
provide the required current. At point H, due to the heating of the electrodes, there is a thermionic
emission of electrons from the cathode. This is then followed by a glow-to-arc transition if the DC power
supply’s internal resistance is relatively low [57].

The qualitative characteristics of a normal glow discharge

Figure 2.15 shows a classical electrical discharge in the normal glow region. Michael Faraday made
the first observation of the glow discharge characteristics in the 1830s thus these have been named after
their respective investigators [55]. The appearance of the glow discharge can be categorized as follows:

Figure 2.15: Typical characteristics of a normal glow discharge

- The cathode

The emission of secondary electrons at the surface of the cathode is pivotal for glow discharge to
sustain itself. The Townsend theory of gas discharge takes into account the bombardment of ions at the
cathode surface as a secondary source of electrons [61]; it defines the coefficient of secondary electron
emission as the number of electrons emitted from the cathode per ion bombardment thus the current
density is defined by:

-Aston dark space

The Aston dark space is essentially a thin region near the cathode with a strong presence of electric
field and negative space charge. It is occupied by slow electrons which will be accelerated from the
cathode. The density and energy of these electrons are too low to excite the gas hence the presence of
dark space. The initial stray electrons and secondary electrons overcome the ions in this region due to the
negative space charge [55].

22
-Cathode glow

The region next to the cathode flow contains a relatively high density of ions known as the cathode
glow. The electrons in this region possess sufficient amount of energy for collision with the neutral atoms
to occur and excite them. It often illuminates orange/reddish due to the emission excited atoms and/or
incoming positions ions approaching the cathode and being sputter off the surface. The axial length of
this region is dependant upon the type of gas and pressure. The cathode glow can from time to time
conceal the Aston dark space by clinging onto the cathode.

-Crookes (or Hittorf) darkspace / Cathode Fall

This Crookes dark space is region in which there is a presence of a moderate electric field and where
a good portion of voltage drop happens. It contains a positive space charge and relatively high ion density
[55]

-Negative Glow

The intensity in this region is the brightest in the entire discharge. The negative glow is most intense
near the cathode side with a low electric field in comparison. It is longer than the cathode with the
electrons carrying the entire current in this region. The bright light output is owed to the electrons being
accelerated towards the cathode resulting in ionisation and intense excitation. These electrons eventually
slow down and no longer provide energy for excitation as they approach the faraday dark space. The
electron number density in the negative glow region is approximately 1016 electrons/m3 [55]

-Faraday Dark Space

In this region the energy and density of the electrons is low as a result of recombination and diffusion
to the walls. In addition to the small net space charge, the axial electric field is also relatively minute [55]

-Positive Column

The positive column is defined as a quasi neutral region with a relatively low electric field, in the
range of 1 V/cm and is just about sufficient enough to maintain the required degree of ionisation near the
cathode end. The length of this region varies with the distance between the electrodes at a constant
pressure and constant voltage drop as opposed the lengths of the other regions which maintained and
unaffected by this. The electron number density in this region ranges from 1015 to 1016 electrons/m3 with
electron kinetic temperature of 1 to 2 eV. A long uniform glow is observed in the positive column; other
phenomena seen include moving striations which are spontaneously triggered and ionisation waves which
are a caused by disturbance. A glow discharge in the positive column illuminates a pinkish to blue colour
[55].

-Anode Glow and Dark space

This brightness in this region is marginally more intense than the positive column. It is a the end of
the positive column and is not always present. This is the boundary of the anode sheath and can be found
on the right end of the positive column; however, its presence may not always be observed. The anode
dark space lies in between the anode glow and the anode itself. Electrons travelling for the positive
column to the anode gives the anode sheath a negative space charge. Furthermore, the electric field in

23
the anode space is more profound than the positive column. Positive ions accelerate away from the anode
leaving a negative charge density which subsequently results in a voltage drop [55].

2.2.4 Radio Frequency Discharges

DC electrical discharges discussed above extract ‘true current’ from the power supplier via the flow
of ions and electrons to the electrodes. Physical processes in such discharges are dominated by axial DC
electric field. One of the most ground-breaking developments made in this space is the introduction of
radio frequency to power the glow discharge. When the applied frequency becomes high enough, the
oscillation period will be relatively similar to the time it takes charged particles to move across the sheath
between the plasma and electrode. Whilst it is very similar to DC glow discharge in many ways, the
differentiating factor between the two is that the rf counterpart can be applied to nonconductive
insulating samples and is not just limited to electrically conductive materials. In addition to this, there are
other more subtle differences between the two such as number densities, energy distributions of charged
particles which can affect the efficiencies of important physical processes such as excitation and
ionisation.

While DC discharges are relatively cheaper to operate, the RF glow discharge is superior as it provide
several advantages such as a wider range of operating parameters, more stable plasma, less prone to
surface oxidation and more enhanced sputtering depths [62]. Glow discharge processes use for treating
insulating materials are often powered by a high frequency in the MHz range. For an AC discharge,
conventional frequency of roughly 50Hz was not found to be an effective for processing insulators; this is
due to the mismatch in the insulator charge up time and the periodicity of the AC supply leading to short-
lived discharges with electrodes successively changing polarities. [63]

RF Plasma Interaction
Radio Frequency power coupled to the plasma through either oscillating magnetic, electric fields or
both. The extent of the interaction between the oscillating fields and the plasma depends on a number of
parameters such as the frequency of the oscillation, electron plasma frequency and electron collision
frequency in the plasma.

The electron plasma frequency is a function of the electron number and the critical density given by:

/ )"
𝜔() = 2𝜋𝑣() = 60 !1 (2.10)
# !

If the incident electro-magnetic (EM) frequency is below this, the inertia of the electrons is low
enough such that they illicit a response to the electric field in the incident EM wave and therefore the
electrons are able to absorb the energy from it. The absorption of this energy may convert to heat via
collisions, or be re-radiated in the from the reflective EM wave from the surface of the plasma.

If the frequency of the incident EM radiation is higher than the electron plasma frequency, then the
electron is not fully able to respond to the incident EM wave as its inertia is too low. This leads to the
propagation of the EM wave through the plasma as a quasi-optical dielectric medium, whereby the
interaction between the incident EM radiation and the individual electrons is relatively insignificant [64].

24
 The RF plasma discharge is highly utilised in the sputtering of a wide range of thin film semiconducting
materials. The principle operation of this technique involves the bombardment of a source material with
energetic ions, typically inert gas ions, such as Argon (Ar+). The forceful collision of these gas ions results
in the ejection of surface atoms from the target material under an RF plasma condition generated at a
frequency of 13.56MHz. This results in the deposition a thin film of the target material onto the substrate.

The RF frequency is coupled to the target material via an impedance matching network and blocking
capacitor as shown in fig 2.16. The blocking capacitor makes sure that the time average net current to
each electrode equates to 0. As RF power is applied, there is an oscillation of electrons between the two
electrodes resonating with the frequency of the rf power.

Figure 2.16 Schematic diagram of an RF sputtering system

The AC voltage/RF power supply in sputtering oscillate at a frequency that is adequate. Too low a
frequency leads to the neutralisation of the target by the positive ions during the negative half-cycle and
electrons neutralising during the positive half. In this case, sputtering does not occur as a result of the lack
of ion sheath. Because electrons have a smaller effective mass than ions, they respond to the change in
target polarity more rapidly than ions [65]. At the typical industrial radio frequency of 13.56MHz, the ion
mobility is too low for it to oscillate and therefore most ions can be essentially found in the central region
of the plasma [64]. During the positive half of the cycle, the target will attract many electrodes acting as
an ion; conversely the counter-electrode will not acquire as many ions the negative half of the cycle. This
will lead to both electrodes possessing a positive bias with respect to the plasma [64]. At some point, the
target will no longer capture as many electrons due to its negative bias curing the positive half cycle. It
will start acquiring enough ions to negate the gain in electrons during the negative half cycle. The target
has a net negative DC bias drawing in the process gaseous ions and generating an ion sheath which leads

25
to the sputtering and deposition of the target material [66]. Figure 2.17 shows the voltages in RF
sputtering, with the target potential shown in green.

Figure 2.17: Voltages in RF sputtering [66]

It is important to mention the symmetry of the process; if the target and substrate are of the same
size and magnitude then there will be complete symmetry, and neither will acquire a DC negative bias
relative to each other. Conversely in an arrangement whereby the cathode and anode are not
symmetrical, the chamber, in addition to the substrate, can also play the role of the anode.

!2 " #
= ""3# (2.11)
!3 2

The ratio of the voltages, with respect to the plasma, measured at the target, Vt, and substrate, Vs,
are proportional to the that of the area of the substrate and target area whereby n ranges from 1.5-4 and
is dependent upon the experimental set-up [67,68]. Therefore, an increased substrate/anode/chamber
area raises the relative target potential. The sputtering of the anode and chamber walls will be
insignificant If the area ratio is high enough.

One of the major concerns with these earlier sputtering processes was the non-uniformity in the
plasma density which subsequently lead to non-uniform film deposition/thickness. Such processes are not
acceptable for the deposition of uniform films in the ever growing electronic and optical coating
industries.

26
2.2.5 RF Magnetron Sputtering.
One of the most notable advancements made in sputtering is the addition of magnetrons and it has
been regarded as the one of the most effect processes for the synthesis of an array of thin film materials.
By applying a magnetic field during the glow discharge sputter deposition process the electrons are
trapped within the discharge for a longer period of time yielding the generation of more ions for the same
electron density. The probability of ionising a gas atom during their displacement from the cathode to
anode increases as the electron trajectory i.e. the mean free path is elongated. Reducing the discharge
pressure and the cathode sheath. By this manner, ions can reach the cathode with almost maximal
discharge voltage and the sputtered atoms can reach the substrate with minimal collision. A more stable
high-density discharge is achieved thus enhancing the sputtering process efficiency. Historically the
process was limited by low deposition rates, high substrate heating and low ionization efficiencies.
However, in recent times, these limitations have been overcome by the development of magnetron
sputtering system.

The process entails generating a strong magnetic field in the vicinity of the target area causing the
traversing electrons to spiral along the magnetic flux lines near the target. This environment limits the
plasma to the target area without damaging the thin film layer being deposited onto the substrate, whilst
maintaining its stoichiometry and uniformity. Adding the magnets behind the target results in the ejection
of electrons at an angle that is non-normal to the incidence to the target to orbit the magnetic field. This
allows for retention of the electrons within the plasma sheath as opposed to losing them to grounded
chamber surfaces which increases the ion generation rate and subsequently the sputter yield. Figure 2.18
shows the schematic diagram of a basic R.F. magnetron sputtering system and the basic components
involved. Ionised Ar atoms bombard a sputtering target which leads to ejection of molecules/atoms from
the target forming thin layers onto a substrate.

Figure 2.18. Schematic diagram of RF Magnetron Sputtering [69]

27
 Magnetron Types
The magnets can come in various forms, one such example is the post magnetrons; a uniform
magnetic B runs parallel to the surface of the cylindrical target. The most widely used type of magnets are
planar magnets. As mentioned, the presence of the magnetic field influences the trajectory of the emitted
electrons from the target forcing them to orbit in circular motion and bringing them back to the cathode
shown in figure 2.19. This can be described by the Lorentz equation which is as follows:

𝐹⃗ = 𝑞 (𝐸*⃗ + 𝑣⃗ × 𝐵
*⃗) (2.11)

with q the negative charge of the electron, E the electric field over the dark region, B the magnetic field,
and 𝑣⃗ the velocity of the electron.

Figure 2.19 Electrons in the presence of a uniform magnetic field follow a circular motion according to the Lorentz
force. Magnetic field points into the page

The Lorentz force on the electron depends on its velocity and magnetic field strength and is
orthogonal to both their directions. The first component of the electron motion is the movement
pattern with respect to the magnetic field lines. A 2nd component is the gyration of the electrons around
the magnetic field lines with Larmor radius, 𝑟4 ; this is the distance between the central point and
charged particle i.e. electron and is defined as:

%&^
𝑟$ = (2.12)
|'|𝐁

Where m is the mass of the charged particles, 𝑣 is the velocity perpendicular to the magnetic field B
and q is the electrical charge unit.

A third component that is accounted for with respect to the electron trajectory is the E x B Hall drift
which occurs around the post in a helical motion and is perpendicular to the electric and magnetic fields.

When an electric field is applied to the charged particles, which are already under the influence of a
magnetic field, the movement of the particles will alter slightly from the circular motion around the central

28
point. The electric field will cause the particles to deviate and thus inducing drift in a more helical fashion
perpendicular to the electric field and the magnetic field. This is the third component that is accounted
for with respect to the electron trajectory known as the Hall drift i.e. E x B. This drift component is
independent of m, v and p. The electric field E increases with v in one half of its orbit cycle while acting
against it in the second phase. This is illustrated in figure 2.20:

Figure 2.20: Movement of charged particles in a constant electric field

In summary, electrons are moving continuously in one direction around the target and a constantly
subjected to acceleration and deceleration close to the surface of the target.

Planar magnets are the most widely used type of magnets. The easy ability to design air-to-air
systems for flat substrates make the planer magnets very convenient. The permanent magnets are
typically placed behind the target in which there is a central disc magnetic pole and an annular pole
allowing for a circular symmetry with respect to the magnetic field lines as shown in figure 2.21:

29
Figure 2.21: Schematic diagram of a Planar magnetron

The magnetic field oriented radially and direct above the cathode surface while the electric field
decreases away from the surface. Similar to post magnetron set-up, the resulting effect is cycloid orbits
above the cathode surface. The presence of a vertical component of the magnetic field forces the electron
to move towards the central region of the magnets. For a circular planar magnetron, the E x B drift will be
parallel to the target resulting in electrons following a circular path around the target resulting in a high
plasma density along this path. Figure 2.22 shows 2 common planar magnetron arrangements. The
configuration of the magnet yields maximum ionisation in the region between the magnets thus most of
the ions will be formed at this position. This uniformity leads to a preferential sputtering of the target
around a circular region known as ‘racetrack’ [65] and as a results typical target utilisation is less than 50%
even with an optimised magnet design as shown in figure 2.23. The placement of permanent magnets
behind the sputter target as shown in results in the B field lines creating a tunnel in front of the target.

30
Figure 2.22: Common planar magnetron magnet arrangements

Magnetron Balance
The magnet configuration in planar and rotating cylindrical magnetrons allow for the entrapment of
electrons thus generating a high ion density in the vicinity of the target for a given pressure. With an
optimum design of system geometry and operating pressure, Electrons are lost from the target area when
their energy drops below the ionisation threshold; thus there is a presence low energy electrons in the
bulk of the plasma near the substrate. Whilst advantageous to the glow discharge process, in some
instances delivering more energy to the substrate also has its benefits. By using an unbalanced magnetron,
it has been reported that ion bombardment of the growing films could be increased, which makes the
films denser and preferentially orients the crystallites in the growing films. Changing the magnetic field
strength of the inner compared to the outer magnet i.e. unbalancing the magnetic configuration, allows
for the electrons to escape the magnetic trap and follow the magnetic field lines towards the substrate.
This yields a strong electron flux in the vicinity of the substrate. The electron flux and subsequent ion flux
can be tuned by adjusting the magnetic field strength balance between the inner and outer magnets.
Quantifying the ‘K’ ratio between the magnetic flux inner and outer magnets is one way to characterise
the balance:

)562 ∫9 + ,-562
562 :562
𝐾= = (2.13)
)78 ∫9 +:78 ,-78
78

𝐵; represents the magnetic field strength perpendicular to the magnet, and S is the area of the
magnet [70].

31
 Shifting the position of the inner and outer magnets with respect to the surface of the target result
reduces the discharge voltage which modifies the sputter rate of the target at constant current. The ion-
to-atom ratio at the substrate, an important parameter during thin film growth, is modified through the
combination of changing position of both magnets. A simpler approach to unbalance the circuitry of the
magnetron that’s highly tuneable is to implement a external Helmholtz coil which increases the field
strength pf the outer magnetic. [71]. This drastically increases the incident ion/atom ratio at the substrate
by opening a leak in the plasma trap. Furthermore, the high ion/atom ratios can be attained by using very
low ion energies (<10-20) which for the most part do not surpass the film damage threshold. Just as
significantly, the substrate bias and the external magnetic field independently control the ion energy and
ion flux incident at the substrate respectively. This means that the advantages of an ion beam deposition
system can be obtained via high rate magnetron sputtering.

Figure 2.24 illustrate the two types of unbalanced magnetron configurations can be achieved. Type
1 possess a stronger central magnet whereby the ion and electron fluxes to the substrate is reduced. Type
II has stronger outer magnets increasing the ion and electron fluxes to the substrate which as discussed
yields higher densities and subsequently promotes beneficial ion bombardment effects [72]. Ion currents
in the range of 3 to 10 mA/cm2 have been reported [72,73].

Figure 2.24: a) Balanced magnetron b) Unbalanced Magnetron type I c) type II

2.2.6 Sputtering Efficiency

Numerous variables have an effect on the outcome of the sputtering process e.g. the RF power, gas
pressure, deposition time, substrate-to-target height, angle of the target with respect to substrate,

32
substrate temperature, type of substrate. All of these factors influence the properties of the resulting thin
film [74]. Sputter yield and energy of the bombarding ions are of high importance and are primarily
dictated by the plasma power [75].

The sputter yield is defined as the rate of deposition and is a measure of the ratio of atoms ejected
from the target to high-energy ions incident on the target which is expressed by equation 1.3.8 [76,77]

(#/%012 45 161781, 9:28;7$1<)

𝛶 = (#/%012 45 ;#7;,1#8 9:28;7$1<) (2.14)

The sputter efficiency is usually measured in Υ per watt of power applied. The binding energy of the
target is threshold that must be met or surpassed for the ejection of an atom to occur from its surface.
This threshold is dependant upon the effect mass of the target atoms and bombarding ions.; It has been
reported that lower binding energies result in more efficient sputtering [76] Ions must possess enough
energy upon collision with the target for sputtering to take place which is approximately 25eV. If ions fail
to possess this level of energy, sputtering cannot take place irrespective of the incident number of ions.
At high energy levels, a reduction in sputtering rates may also be seen as a result of the ions penetrating
into the target to deep. The energetic incident electrons on the target do not affect the sputtering process
because its energy levels are too small to stimulate the collision cascade [78].

The momentum collision cascade is the process by the energetic ion incident on the surface of the
targets causes a sequency of interatomic collisions to occur and subsequently resulting in an atom being
ejected from the target which is illustrated in figure 2.25.

Figure 2.25 Schematic diagram of momentum collision cascade [79]

33
 At the fundamental level, sputtering is a momentum transfer process and therefore sputter yield
depends on the transfer of momentum between the high energy ions and the target. It has been reported
that there is a correlation between the rate of sputtering and the atomic masses of the target material
relative to the bombarding ions; highest rates of sputtering occur when the atomic assess are similar.
Conversely the greater the difference in atomic mass, the sputtering is not as efficient. Using the noble
gas that has the closest atomic mass to the sputter target ensures maximum efficiency in this regard [76].

Powell et al [76] made the observation of four different physical outcomes of the collision cascade
depending on the energy of the incident particles. Between 0 and 50 eV, sputtering only takes place in
highly dense plasma whereby the yield is very poor [80-81]. Commercial PVD sputter systems utilise an
energy within the moderate range of 50-1000eV. An ion incident upon the target with energies from 1000
and 50, 000 eV results in dense target-atom cascade whereby the atom-to-atom bonds are ruptured.
Energy levels greater than 50,000 eV causes a deep implantation of the ions into the target leading to few
atoms being sputtered and a subsequent poot sputter yield [76].

The angle of incident of the bombarding ions also has been reported to have an effect on the sputter
yield [82-86]. Particles bombarding the target at 90° emit less sputtered atoms compared to ions incident
at shallower angles [85]. Shallow angles of the collision cascade take place near the surface of the target
which leads to higher change of the sputtering of an atom. By the same token, if the angle is too shallow,
the ion will simply deflect of the target surface and thus no sputtering will take place. The ideal sputtering
angle is between 45° and 50° [84-86]. In RF sputtering, the inclination of the target does not improve the
angle of the incidence of the ions because the plasma sheath is parallel to the target and the ion impact
on the target surface is always normal.

Powell et all identify two types of transparent that take place with varying chamber pressure i.e.
ballistic transport and diffusive transport. Ballistic pressure is observed at low pressures (1.33 to 1.33x10-
5
mbar). The sputtered atoms traverse the chamber directly to the substrates with minimal collision to
decelerate them. As a results, these atoms arrive at the substrates with high energies [82-86] and implant
themselves into the substrate, leading to a thin film that’s finely grained with sufficient adhesion to the
substrate. At pressures of tens of millibars or above, diffusive transport occurs whereby the collision
between the sputtered atoms cause a loss in energy which in turn increases the electrostatic forces. The
implications of this include deposition that’s more likely to be inhomogeneous and a thin film that consists
of large grain sizes.

Other factors, such as target positioning, can alter the properties of the plasma and thin films
produced [87]. This occurs when other gasses, apart from the main working gas, are present in the
chamber, or when any object within the chamber emits atoms in the form of gas. The presence of such
gases in the system should alter the property of the plasma emission.

2.2.7 Summary
The electric glow discharge is formed by passing current through a noble gas such as Ar at a currents
ranging from 100 V to several kV. Electrons gain sufficient amount of energy to generate visible light by
excitations collisions; a phenomenon which is most easily attained by the Interaction between the anode
and cathode generating a complex circuit current that produces the glow. The striking of glow discharges
involves formation of restricted electric fields and plasma sheaths at each of the electrodes. Ionisation is

34
a critical part of a glow discharge under equilibrium conduction. DC power is source that could be used to
power the glow discharge which is frequently used to deposit electrically conductive target materials.
However, its limitations when it comes to depositing dielectric materials i.e. non conducting insulating
target materials that can take on polarised charge. RF sputtering is employed to deposit such materials
as it alternates the electrical potential of the current neutralising the charge build-up. Magnets have been
widely implemented to confine the plasma near the target allowing for high ionisation efficiencies, better
sputter yield and more uniform deposition. Commercial PVD sputter systems utilise an energy within the
moderate range of 50-1000eV for satisfactory sputter yield. Microwave power sources can also be used
to strike the plasma whereby a resonant condition is created for the electrons which leads to highly
efficient ionisation. The next chapter will discuss look into what sort of work has been done with respect
to characterising and modelling of the plasma and the need for a simple evaluation of the glow discharge.

2.3 Characterisation and Modelling of Plasma

2.3.1 Introduction
In this chapter, a brief overview of the various methods that are used to characterise and model the
glow discharge is presented; it will also present need for simplified evaluation of the discharge such that
applied/materials scientist can get a ‘snapshot’ of the state of plasma and how it correlates with specific
process parameters rather than delving into complex theoretical physics.

As discussed in the previous chapter, the plasma in the sputtering procedure is the core of the
reaction and is fundamental to the deposition of the thin film coatings such as TCOs that are used in wide
range of applications [88]. By evaluating the plasma, A set standard for deposition procedures can be
defined, regardless of the machine being used. By considering the sputtering target surface and the
plasma as a light source, this concept can be explored However, characterising plasma through optical
spectroscopy is nothing new. Currently, plasma physicists study and investigate plasma by means of
Emission spectroscopic techniques [89–95]. The diagnosis of laboratory plasma is usually carried out by
optical emission spectroscopy (OES), through which numerous analytical techniques are established to
determine the plasma properties, such as electron density, plasma temperature, elements recognition,
and quantification of elements that are present in the plasma.

The radiative behavior of the atomic constituents of any plasma can only be predicted if the expected
populations of possible states are known and the atoms obey the Boltzmann distribution for every
possible state and radiation energy density. Subsequently, it can be assumed that a thermal equilibrium
condition is present [91–96]. However, creating such a condition in the laboratory would be almost
impossible, and hence the concept of local thermal equilibrium (LTE) condition is considered. LTE can be
described as a state where the Boltzmann and Saha equation, which governs the distribution of energy
level excitation and ionization temperature, are equal to the Maxwell–Boltzmann distribution of free
electron velocities [95–97]. Achieving the LTE condition depends on defining the plasma by a common
temperature T and the existence of sufficient large electron density. McWhirter proposed a criterion
where there is a need for a critical amount of electron density for LTE conditions to exist. This criterion is
demonstrated by Equation 2.16 where Ne is the electron density, T is the plasma temperature, and ∆E is
the energy gap [98–101].

35
 #
𝑁) ≥ 1.6 × 10#< 𝑇 < (∆𝐸)% cm% (2.16)

The temperatures of ions and electrons are directly proportional to their random average kinetic
energy, and the distribution of velocities for each particle is governed by the Maxwell distribution when
thermal equilibrium conditions apply. Under the LTE conditions, the same temperature is assumed for
electrons, ions, and atoms in the plasma and the plasma temperature is referred to as the temperature
of the electron [91]. Various methods of plasma temperature determination exist, however, the simplest
method by far is the Ratio method, which utilises the intensity of two spectral lines for calculating the
temperature of the plasma. The intensity of a spectral line associated with a transition is expressed by
Equation 2.17. In this equation, Iij is the intensity of transition from state i to state j, Aij is the transition
probability between the two states, c is the speed of light, N is the number density of electrons, h is the
plank constant, gj is the statistical weight of the upper level (j) and λij is the wavelength corresponding to
the transition, Ej is the energy of level (j), T is the excitation temperature and k is the Boltzmann constant,
and the U(T) is referred to as the partition coefficient, which accounts for the probability and degeneracy
of the two states (i and j).

ℎ𝑐𝐴.= 𝑔= 𝑁 " >$

𝐼.= = 𝑒 ?@ (2.17)
𝜆.= 𝑈(𝑇)

From the spectral data obtained, two spectral lines can be chosen from the same species and their
ionization stages, e.g., Ar I, where there is a large difference in the upper energy level. By taking a ratio
(I1)/(I2) of the intensities of the two selected lines, the constants cancel each other out which yields
Equation 2.18, where I1 and I2 are the intensity of comparative peaks due to their respective ij transition
[102].

𝐼# 𝑔# 𝐴# 𝜆< ["B>% ">"C]

= 𝑒 ?@ (2.18)
𝐼< 𝑔< 𝐴< 𝜆#

Solving Equation 2.18 will give the value of T in electron volts. This method is the simplest method of
calculating the temperature of the plasma, and its accuracy is conditional upon using two lines with a
maximum difference in their upper energy states.

As an alternative to the ratio method discussed, a method that is referred to as the Boltzmann method
can be applied to calculate the temperature of the electron. This method also utilises Equation 2.17;
however, with a rearrangement. The equation is rearranged to give Equation 2.19. Taking the natural
logarithm of both sides of Equation 2.19 would lead to Equation 2.20 [91,99–102].

𝜆.= 𝐼.= 𝑁 " >$

L M= 𝑒 ?@ (2.19)
ℎ𝑐𝐴.= 𝑔.= 𝑈(𝑇)

𝜆.= 𝐼.= 1 𝑁
ln L M = − R𝐸= S + ln ( ) (2.20)
ℎ𝑐𝐴.= 𝑔.= k𝑇 𝑈(𝑇)

36
 By plotting the first term of Equation 2.20 against Ej, the slope of the plot can reveal the value of 1/kT,
from which value of T can be extracted. The electron density can be calculated by using various methods,
for example, by applying the Stark broadening relationships. The Stark broadening is caused by the electric
field of electrons and ions interfering, which results in the broadening of the spectral line. The interference
of the mentioned electric fields causes fluctuations in the field of plasma as the radiating atoms are
surrounded by the interfering electrons and ions. The electric field of the electrons or ions causes a
perturbation of energy levels that are close to the continuum, while simultaneously affecting the
externally applied electric field, which ultimately causes the observed spectral broadening. The magnitude
of this broadening in terms of its full width half maximum (FWHM) ∆λ1/2 is given by Equation 2.21.

#
𝑁) 𝑁) F # 𝑁) (2.21)
∆λ#/< = 2ω W #E X + 3.5𝐴 W #E X (1 − 1.2𝑁G "% )ω( #E )
10 10 10

In this equation, ω is the electron impact parameter, A is ionic impact broadening parameter, Ne is
the electron density, and ND is the Debye shielding parameter. The first term in this equation is associated
with broadening that is caused by electrons and the second term is associated with the broadening that
is caused by ions [103]. As the emissions from argon species is considered for the measurements,
eliminate the second term of the equation that relates to the other ions can be eliminated to obtain
Equation 2.22 [102–104]. Hence, rearranging the equation and obtaining the ∆λ1/2 value can calculate the
Ne.

𝑁)
∆λ#/< = 2ω( #E ) (2.22)
10

All of the methods mentioned so far are some of the OES methods of measuring the electron density
and electron temperature in a plasma. These methods are familiar among plasma physicists; however,
when it comes to the application of sputtering plasma and thin film analysis, using the above-mentioned
techniques will require the necessary expertise and precision spectroscopy and data interpretation. Many
scientists or operators who use plasma deposition for the preparation of thin film coatings would
ultimately require a machine that would be easy to use and produce consistent thin film coatings. The
objective is to simplify the operation of the sputtering technique for the users and possibly propose a
simpler monitoring diagnostic methodology for the plasma deposition process.

2.3.2 Novel approach to characterising the plasma

This work reports on a novel strategy of the in-situ characterization of plasma in a sputtering process
by translating the emission spectra into a colour indexing (chromaticity) pattern that is able to quickly
monitor any changes in the plasma character with a change in the operating parameter conditions during
the course of the process, which in turn provides a better handle to reflect the instabilities of the tool and
its optimizing parameters. This is a radical conceptual proposition that can potentially simplify the
evaluation of the sputtering plasma for applied scientists by theorising a unique characterisation
parameter for particular plasma deposition conditions, just like a fingerprint. As an analogy can be drawn
between this work and the human genome that defines each individual person based on the composition

37
and arrangement of billions of complex genetic coding embedded within each person’s DNA, complex
laboratory procedures can be used to sequence and encode the genetic makeup of individuals; however,
a simpler approach to distinguish people from one another is to check their fingerprint or iris patterns,
rather than conducting the complex genetic complexity that defines the individual. Indeed, that is the
objective of this communication. By considering the plasma as a light source, the focus on its most
apparent character; its colour; in doing so, patterns can potentially be identified that may relate its colour
to certain fundamental parameters that are involved in generating the plasma, such as the driving power,
the working gas, and its pressure. Plans are underway to further investigate the relationship between the
thin film coatings that are deposited via this technique and the plasma colour with reports in due time.
The colour of any light source can be mathematically described through what is known as colour spaces.
A colour space is a completely-specified scheme for describing the colour of light, ordinarily using three
numerical values (called coordinates). An important colour space, defined by the International
Commission on Illumination (CIE, the initials of its French name), is the CIE XYZ colour space [105].
However, it is wholly defined in terms of human’s perception of light through their eyes. If two instances
of light appear to a viewer to be the same colour, they are the same colour. Colour is usually recognized
by the viewer as having two aspects:
• Luminance, as an indication of the “brightness” of the light.
• Chromaticity, the property that distinguishes colours.
Colour is not a primary physical property, like the temperature or pressure of a gas. Colour is related
to the energy of photons that make up the light. Plotting the intensity of the photon energies (power) that
are associated with the relevant wavelength will produce a spectral plot. The intensity and distribution of
the emitted wavelengths that ultimately shape the plot determine the chromaticity of the light; its overall
“vertical scale” determines its luminance [105]. In fact, there can be several instances of light with
different spectrums, which nevertheless have the same colour. This situation is called metamerism.
Through mathematics, the XYZ colour spaces are used to construct the x,y colour function, which is
presented in figure 2.28A, where the specified area under the defined emission spectrums are used to
calculate the coordinates in figure 2.28B.

Figure 2.28 (A) The x,y International Commission on Illumination (CIE) chromaticity index. (B) The CIE XYZ colour
function diagram.

38
 A luminance-chromaticity space (figure 2.28B) is then constructed by defining two values; x and y, as
follows:
𝑋
𝑥= (2.23)
𝑍+𝑌+𝑋

𝑌 (2.24)
𝑦=
𝑍+𝑌+𝑋

In these equations 2.23 and 2.24 , the X, Y, and Z represent the area under emission peaks in the
regions. By plotting the x against y, the CIE x, y chromaticity diagram is constructed and is used to give a
specific sense to a particular colour (figure 2.28B). These peaks represent the emission wavelengths from
the light source at the specific wavelengths.

Another photometric parameter is the radiance and spectral radiance of the plasma light. The
radiance indicates how much radiant flux is emitted by a surface when it is received by an optical system
looking at that surface. The radiance value is calculated by measuring the Radiant flux of an emission
source, which is the radiant energy emitted per unit time. Spectral radiance expresses radiance as a
function of frequency or wavelength and the spectral plot of the emission is based on the radiance value
as a magnitude of intensity [105].

Within a sputtering plasma, a significant number of energy transitions are occurring, leading to a
complex emission spectrum. These emissions ultimately define and construct the spectral plot of the
plasma as a light source that is observed in plasma. The efficiency and characteristics of the sputtering
process depend on the characteristics of the plasma, which in turn depend on the density of the particles
and energies within the plasma [106]. Through complex procedures, some of which have already
discussed, the electron temperature and electron density, meta-stable atoms, and ions can be deduced
and estimated from the spectral plot [107–110]. If the colour parameters can be identified and linked to
the existing plasma analysis techniques, it can benefit applied scientists to perform and conduct thin film
depositions and can have an easier method of assessing the specifics of the plasma being used.

Due to the extensive optoelectronic applications, high interest in transparent conductive oxides, such
as ITO, AZO, and IZO, and their deposition characteristics via plasma sputtering exists and there is plenty
of literature available from various research groups [111,112]. These materials are of high interest to
many photovoltaic researchers, and hence this study focussed on evaluating the light from a sputtering
magnetron plasma source that was fitted with an IZO target and the objective here is to take a radical
diversion from the complexity of the conventional plasma characterisation techniques and observe the
colour of the plasma and explore the feasibility of exploiting the colour characteristics of the plasma as an
indicator of the coating process. A significant number of energy transitions occur in the UV and IR region
of the spectrum that are not accounted for when constructing a conventional chromaticity diagram;
however, this work is an initial focus on clarifying the feasibility of such a radical approach, after which
further efforts can be invested on to modelling a plasma colour analysis system that can provide a shortcut
into the characterisation of a sputtering plasma.

39
2.3.3 Modelling Background

As mentioned above, characterising a plasma via optical spectroscopy has been conducted, typically
by optical emission spectroscopy by which numerous analytical techniques are implemented to determine
certain properties such as electron density, plasma temperature, elemental recognition and qualification
of elements present in the plasma. Investigating such complexities requires in depth knowledge and
understanding of plasma physics not to mention time and effort. With the work carried out on
characterising the plasma by its light, it serves a steppingstone for the simplified evaluation of the plasma
discharge. As such, a deeper study on the modelling of a plasma was carried out to investigate whether
alternative methods could be employed in diagnosing the plasma that further reduces the barrier
between a materials/applied scientists’ understanding and operation of plasma-based systems.

It is important to know that modelling of the magnetron sputtering discharge has been reported on
multiple occasions since the 1980s which can be categorised as either purely mathematical models or
analytical models. Theoretical models can be subclassified into kinetic models, fluid models or one that is
based on plasma physics. Regardless, these models need to integrate calculations of the electric field
based on the applied external voltage as well as the distribution of the charged plasma species.
Fundamentally, the sputtering process involves the ejection of particles from the target, particle transport
and the condensation onto the surface of the substrate. So the complete model of the process would
need to discharge physics, plasma physics and interactions at the material surface. Below is a summary of
some of the models that have been employed:

1. Analytical models operate via a simple analytical formula that describes the behaviours of the
glow discharge parameter e.g. current and voltage. Whilst they are simple and easy to calculate,
their accuracy is quite poor and can only be applied to specific ranges of deposition conditions
[113]
2. Pathway models was originally developed to determine the ionized fraction of the material
arriving at the substrate and provide an explanation for the low deposition rate in some
discharges. It is based on monitoring the sputtered material & working gas species within the
discharge for a better understanding of process in pulsed magnetron sputtering [114]
3. Fluid models are based on continuity and transport equations for the wide array of discharge
species as well as Poisson equation so that a self-consistent electric field distribution is obtained.
It defines the plasma as a continuum. Whilst it is computationally simple, its validity to describe a
magnetron sputtering discharge has been scrutinised [115]
4. Ionisation region models are volume-averaged and time dependant. This region is defined as the
bright glowing plasma near the surface of the target. This time evolution of neutral & charged
species as well as the electron temperature in a pulsed magnetron sputter discharge can be
determined based on this model. There are some constraints to this model associated with certain
experimental parameters and reaction system setup. The model needs to be adapted to an
existing discharge and then fitted to reproduce the measured discharge current and voltage
waveforms [116,117].
5. Hybrid models makes use of the precision of kinetics models whilst taking advantage of the
computational simplicity of fluid models. The secondary electrons are emitted from the surface
of the cathode target and are subjected to high energy acceleration within the cathode sheath.
The electrons can be classified as either ‘fast’ electrons whereby their energy exceeds the
threshold for inelastic collisions, or slower electrons. These two groups of electrons are treated
with a kinetic Monte Carlo model and a fluid model respectively. In the hybrid method, ions &

40
 bulk electrons and fast electrons are treated by the fluid description and particle model
respectively [118] though this approached has been critiqued by Kolev and Bogaerts [119]
6. Simulations pertaining to Direct Monte Carlo involve following many plasma species which are
represented by test particles. The movement of these particles is impacted by applied forces and
collides with other particles. These simulations have been used to prediction the spatial
distribution of the ionisation [120] and ion trajectories [121] in a planar magnetron sputtering.
7. Boltzmann solver is a numerical model based on solving the Boltzmann equation to obtain the
electron energy distribution within the discharge. While this is a well established model with a
high degree of accuracy the Boltzmann equation takes into account the Lorentz force term
resulting in complex mathematical calculations. Thus, this method has only been use to
successfully model a cylindrical magnetron sputtering discharge with a coaxial inner cathode and
an outer anode [122-125].
8. Monte Carlo collisional simulations is a variation of the Monte Carlo simulation. Newton’s law is
applied to calculate the trajectories of various species and their collisions are characterised by
assigning random numbers [126]. In addition, Poisson’s equation is used to calculate the electric
field distribution self consistently from the displacement of the charged species. This method
provides spatial distribution of the charged species that’s mapped onto a grid, the electric field
across the discharge showing the distribution of the charge density which can be used to
determine the electric field distribution. This approach a powerful numerical method pertaining
to the analysis of the sputtering discharge. However it is heavily dependant upon immense
computational power as it tries to dissect the behaviour of the charged species in great detail
along with solving the Poisson equation [127].

The modelling methods mentioned all require a deep understanding and extensive knowledge in the
field of theoretical plasma physics and statistical mechanics. However, implementing such detailed
knowledge and complex mathematical models is not feasible at operational levels and industrial scale
deposition. Moreover, the distribution of the ionisation processes in magnetron sputtering discharges
does not appear to be uniform along the racetrack. Such processes commonly exhibit Inhomogeneous
plasma with distinct regions of increased intensity of light that seem to float along the racetrack. The
presence of rotating dense plasma known as spokes has been reported over the span of a few decades.
[128-130].

The plasma may seem inhomogeneous as the spokes are independent of the configuration of the
magnetrons i.e. the phenomena has been observed with rectangular, linear & circular magnetron targets
[131-134]. At higher discharge currents, where power densities are above 3kW, the inhomogeneities
observed become smoother to yield a homogenous plasma azimuthally [135,136]. This reason for this is
because of increased electron heating which is a result of a combination of secondary electrons
acceleration and pure ohmic heating [137].

The incorporation of reactive gas yields the formation of compounds on the surface of the target
which is known as surface poisoning; this detrimentally affects other sputtering discharge parameters like
secondary electron emission yields, sputter yield and the plasma composition close to the target. In some
instances, it can lead to a change in the shape of the spoke [138] and subsequent reduction in discharge
voltage upon the introduction of reactive gas; this has a correlation with a rise in the secondary electron
emission [139,140]

In addition to the presence of spokes, the plasma may also show signs of other instabilities. More
specifically, the plasma may oscillate in a direction such that it is normal to the target surface which is

41
known as breathing instability [141,142]. It is common for spokes and the breathing instability to
superimpose on each other [141].

Electrons can cause localised excitation and ionization as a result of reaching higher potential and
subsequently becoming energized. This implies that images of spokes reflect that of the potential
distribution [143]. Held et recently reported that the spokes had a high plasma density, electron
temperature and plasma potential that the surrounding plasma [144]

2.3.4 Modelling plasma with artificial neural networks.

A new & comparatively simplified method for modelling the plasma with the intention to impose all
the associated complexities and calculations to an artificial neural network has been studied . With the
assistance of modern computational and graphical power and artificial intelligence based methodologies
pertaining to data assessment, This work investigates on implementing a novel approach in the evaluation
of a plasma during the sputter deposition process to assess the qualitative parameters of thin film
materials as deposition was taking place. There is strong reason to believe that artificial intelligence and
deep learning methods have the potential and power to fully automate and create a digital shadow for
the sputtering deposition process on an industrial scale.

A digital shadow is a historical or retrospec•ve representa•on of a physical en•ty, process or system

while a digital twin is a virtual representa•on which has the poten•al to formulate solu•ons for problems
that cannot be resolved with the physical system alone. A shadow is created by aggrega•ng and analysing
historical data collected from the physical asset or system over •me. Unlike a digital twin, a digital shadow
does not provide real-•me simula•on or monitoring capabili•es. Instead, it serves as a repository of past
performance data and insights. Digital shadows are o€en used for post-analysis, trend analysis, historical
data comparison, and decision-making based on past experiences. They can help iden•fy pa•erns,
anomalies, and trends in historical data, facilita•ng informed decision-making and future planning. In
summary, while both digital twins and digital shadows involve digital representa•ons of physical en••es
or processes, digital twins focus on real-•me simula•on and monitoring for predic•ve analysis and
op•miza•on, whereas digital shadows are retrospec•ve representa•ons used for historical analysis and
decision-making based on past data.

Artificial Neural Networks (ANN) are powerful, scalable, versatile and highly complex which serve as
the core of deep learning methods. The introduction of high level application programming interfaces
(APIs) such as KERAS and TENSORFLOW in python programming language has made it convenient to
implement ANNs. The origins of ANNs go as far back as the 1940s but due to their complexity and need
for significant computational power, their application was heavily limited. With the rise of modern
computing power and business entities like Google and Facebook alongside the accessibility of large banks
of digital data, there is an emergence of ANNs which are outperforming existing machine learning models
like support vector machines (SVM) which were once market leader as far as popularity and application
goes. Historical theoretical limitations pertaining to ANNS like the model getting trapped in local minima
function can be circumvented. Ironically, Multidimensional data possessing sophisticated function
gradients are least affected by the local minima restriction.

The units that are part of the construction of ANNs hypothetically mimic the behaviour of biological
neurons and their interactive connections. However, biological neurons are mysterious and complex by
nature which is yet to be fully comprehended by neurologists. Thus, the term neuron in this AI context is

42
only used for description purposes and not to mistaken as being equivalent to an actual neuron in terms
of design and status. The term perceptron was invented to describe the constituents in ANNs in 1957 by
Frank Rosenblatt. The perceptron calculates a weighted sum of its inputs and applies a function to that
weighted sum. Figure 2.29 displays how a single perceptron operates in basic mathematical terms. A
Detailed review of artificial neural networks and deep learning can be found in [145,146]

Figure 2.29. A simple representation of the operation of single perceptron fed with three data (X0,X1,X2) inputs to
give an output (yˆ)which need to be close to a known (y) value.

ANNs essentially are based on a collection of connected perceptrons and comprise of input layers,
hidden layers and output layers. Each perceptron has a weight and threshold associated with it. When an
ANN consists of deep stack of hidden layers it is known as deep neural network.

The process involves feeding multidimensional data (vector) to the ANN/DNN via input neurons and
gets passed through the hidden layers leading to an outcome/value from the output layers. The model
will compare the output value to known experimental outputs and train itself by shifting the weights
(figure 2.30) associated with the perceptron each layer. The model learns by a process known as back
propagation which was first introduced by David Rumelhart [146] which at its core incorporates gradient
descent for the calculation of gradient of the networks error. The perceptron will apply weights to the
inputs, sum up and apply a function (σ) on the outcome, and will adjust the weights through a forward
and backward loop by implementing a loss function (error factor) until the selected weights result in yˆ
being as close as possible to y value. A basic understanding of the principles of deep learning method was
provided. As this is an exponentially developing field in artificial intelligence and computer science,
detailed discussion of the complex theories and mechanism of deep learning is beyond the scope of this

43
work. This project explores the feasibility of applying deep learning models to sputter depositions process
as with the objective of making it more digital.

Figure 2.30. A simplified illustration of the backpropagation process and adjustment of weights by a perceptron.

The objective of this work was to push all the mentioned complexities into the black box of an
artificial neural network. With modern computing powers and artificial intelligence (AI)-based data
assessment methodologies. The aim was to explore implementing an alternative approach in plasma
diagnostics during the sputter deposition process to assess the qualitative parameters of thin films as they
were deposited. Artificial intelligence and deep learning can potentially be the answer, if the objective is
to fully automate and digitise the industrial-scale sputtering process.

2.3.5 Summary
Conventional methods of plasma evaluation and diagnosis involve complex plasma physics and
determination of certain properties such as electron density, plasma temperature, type (and quantity) of
elements via numerous analytical techniques. A proposal was put forward that plasma can be
characterized by its light which can provide a general overview for the everyday applied/materials
scientist. The results of colour characterization of the plasma will be discussed in the next chapter.
Following on from this work the modelling history of the plasma was detailed. Various methods exist,
however all require a deep understanding and extensive knowledge in the field of theoretical plasma
physics and statistical mechanics. Implementing such complex mathematical models is not feasible at
operational levels and industrial scale deposition. An relatively simplified method for modelling the
plasma involving Artificial Intelligence has been postulated whereby all the associated complexities &
calculations are subjected to an artificial neural network. This work paves the way for the digitization and
automation of the sputter deposition process.

44
Chapter 3 Experimental Methods and Characterisation Techniques

This chapter provides a detailed description of the experimental set-up employed for the wide array
of experiments conducted during this project. Though the same set-up was used, each set of experiments
differed from one another and will be detailed in this chapter. In addition, the array of techniques used
to characterise the thin film samples will be discussed.

3.1 Experimental set-up

The sputtering machine used for these experiments was a V6000 unit that was manufactured by
scientific vacuum systems limited (SVS Ltd., Wokingham, UK) with a vacuum chamber of ~40 cm × 40 cm
× 40 cm. The magnetron of the V6000 unit was fitted with a six-inch 99.99% pure, Zinc Oxde & Indium Zinc
Oxide (IZO) target material (of varying compositions) with copper indium back bond for efficient thermal
dissipation. The working gas used in the experiments was a 95% argon and 5% hydrogen (Ar + H) single-
source mixed-gas cylinder. Spectral data from the plasma was obtained by placing an in-vacuum collimator
optic probe that was made by Plasus GmbH (Mering, Germany). The probe was installed on the
magnetron, so that it horizontally collected light from ~1.5 cm away from the surface of the target and at
a distance of 4 cm from the edge of the target (Figure 3.1 A). The unique feature of the optic collimator is
the honeycomb structure of the photon inlet, which traps the sputtering particle and prevents gradual
coating of the collimators quartz window. This is highly important, as any coating of the inlet quartz
window will undermine the reliability of the data obtained. Thus, A test was carried out on the quartz lens
for any possible coating by running a continuous 5 hours deposition experiment followed by dismantling
the collimators (Plasus GmbH, Mering, Germany) and using UV/VIS spectroscopy (Bibby Scientific Ltd,
Staffs, UK) to check the inlet quartz window for any coatings formed to ensure the data were not
undermined (Figure 3.1A). The Plasus Emicon Spectrometer (Mering, Germany) that generates a detail
spectral plot of the emission was used for detailed spectral analysis. The software coupled to the Emicon
spectrometer was programmed to calculate the area under the peak of the spectral range based on
predesignated segments of the spectrum. In addition to this, the light collected from the plasma was also
guided to a 2nd spectrometer; a Jeti Specbos 1201 spectrometer (Jena, Germany) that calculates the
chromaticity index of the light, the Jeti was programmed, so that, for each measurement, it took 20
readings and inputted the average.

45
Figure 3.1 (A) In-vacuum collimator optics was placed adjacent to the magnetron and the emission data were
transferred to the spectrometers via quartz fiber, the distance between the collimator and target edge was 4
cm. (B) The Collimator. (C) Magnetron with the in-vacuum collimator from the V6000 tool

Figure 3.2. (A) UV/VIS transmission spectroscopy (320 to 900 nm) to ensure the lens was not coated during the
experiments. (B) Optic collimator with a honeycomb structure

During each deposition, soda-lime glass slide substrates were thoroughly cleaned and were placed
in the chamber, with the substrate stage rotating at a speed of 20 rpm. Prior to loading, all substrates
were washed and sonicated with soap water, acetone and ethanol separately to ensure a clean surface
before the thin film deposition.

A wide range of spectral data was collected by fitting an array of targets in the magnetron and depositing
with varying plasma power and gas flows which will be further outlined subsequently. The 1st set of
experiments involved the colour characterization of the plasma during the sputter deposition of a range
of TCO targets to not only diagnose the plasma but also to attempt to investigate the relationship between
the plasma and the resultant thin film material.

3.1.1 Experiments for the optical analysis of the Glow Discharge via colour characterisation
Using the IZO target fitted onto the magnetron, a series of tests were carried out by igniting and
running the plasma under various conditions. The working gas, working gas pressure, and radio frequency
(RF) plasma power were varied and data were collected. Two types of working gases were used viz. pure
Argon (Gas Ar) and 95% Argon: 5% Hydrogen (Gas ArH). Table 3.1 defines the experimental regime
undertaken. Test 1: where the stability of the plasma was monitored at 100 W under 1.9 × 10 −3 mbar of

46
working gas pressure, under Ar and ArH gas; Test 2: The spectral emission of the plasma was monitored
while increasing the power of stabilized plasma from 100 to 300 W under 1.9 × 10 −3 mbar of Ar at a rate
of 100W per minute; Test 3: The spectral emission of an stabilized plasma was monitored at 100 W plasma
power under various Ar and ArH gas pressures at a rate of 5 x 10-4 per minute. (Note: separate spectral
analysis of the plasma power was carried out up to 350 W).

Table 3.1. The experimental regime conducted to obtain the spectral results from the plasma light source
under various conditions in 3 separate test groups.
Test Regime Variable Parameter Power (RF) Working Gas Pressure (mbar)

Test 1 Time 100 W Ar, ArH 1.9 × 10−3

Test 2 Plasma power 100 W → 300 W Ar 1.9 × 10−3

Test 3 Gas pressure 100 W Ar, ArH 1.2 × 10−3 → 7 × 10−3

3.1.2 Experiments for the synthesis of Highly conductive Zinc Oxide films
A range of IZO and ZnO samples were prepared for this experiment. For the IZO samples (S series):
The magnetron of the V6000 unit was fitted with a six-inch diameter 99.99% pure indium zinc oxide (IZO)
target (In60-Zn40 at %) material with copper indium back bond. Using this target, ‘seven’ IZO thin film
depositions were carried out. The growth conditions of these samples can be found in table 3.2. All the
experiments were carried out at ambient temperature for two hours under a 100W RF plasma power and
a chamber pressure of 2.3 × 10−3 mbar.
Table 3.2 outlines the range of deposition conditions used to synthesise the 7 IZO samples

Ar+H Prior exposure to Prior exposure to Ar

Sample ID Working Gas
(sccm) Ar+H Plasma (hr) Plasma (hr)

S1 9 ArH 0 0

S2 9 ArH 2 0

S3 9 ArH 4 0

S4 9 ArH 6 0

ArH then
S5 7 8 0
switched to Ar

S6 7 Ar 0 2

S7 7 Ar 0 4

For ZnO samples (H series): The magnetron was fitted with a six-inch diameter 99.99% pure zinc oxide
(ZnO) target material with copper indium back bond. Using this target, ‘three’ ZnO thin film depositions

47
were prepared using 150 W RF plasma power with a working gas pressure of 2.3 × 10−3 mbar for exactly 3
hours under the protocol outlined in table 3.3

Table 3.3. outlines the range of deposition conditions used to synthesise the array of H:ZnO samples

Ar+H Prior exposure to Prior exposure to Ar

Sample ID Working Gas
(sccm) Ar+H Plasma (hr) Plasma (hr)

H0 9 ArH 0 0

H3 9 ArH 3 0

H6 9 ArH 6 0

3.1.3 Experiments for investigating the effect of substrate position relative to plasma
The focus of this experiment was to present the observations made during the deposition highly
conductive zinc oxide films via RF sputtering with varying positions relative to the discharge. For this
experiment, a set of chamber pressures were used to deposit films associated with 13.9, 10 and 6 sccm
Ar + H flow rates resulting in working chamber pressures of 0.3, 0.23 and 0.15 Pa. The depositions were
carried out under 100, 150 and 200 watts of RF (radio frequency) plasma power under each flow rate.

The substrates were placed on the substrate holder of the machine, as depicted in figure 3.3, with
position E indicating a substrate on the edge of the substrate holder, and C indicating the centre of the
substrate holder position. The substrate was rotating at a speed of 10 rpm in all experiments. Using this
configuration of substrate positions, thin films of ZnO were deposited under various amounts of plasma
power and Ar + H flow rates. Table 3.2 presents the regime by which samples were prepared and named,
where each deposition was carried out for a period of 3 hours. In practice, glass samples were placed in
between the C and E positions; however, here the focus was primarily on detailed observations associated
with the E and C positions. The gas flow raw is represented as: high(H) flow rate (13.9 sccm), medium(M)
flow rate (10 sccm) and low (L) flow rate (6 sccm). For this experiment, 2 collimator probes were used to
collect the spectral emission data; one was positioned in the centre of the substrate holder and the 2nd
one was placed at the edge of the holder as depicted in figure 3.4.

48
Figure 3.3. During each of the nine deposition trials, two soda lime glass samples were placed on the substrate holder
(220 mm diameter), one in the centre and one on the edge (left). To measure the thickness gradient from the centre
to the edge, a bridge between two masks was designated on an intact glass substrate for profilometric measurements;
thickness of the samples was measured at the (10–110 mm) distance from centre, as depicted in this figure (right).

Table 3.4. Nine depositions were carried out under various plasma power and gas flow rates:

Position on Plasma Power

Sample ID Gas flow rate (sccm)
Substrate Holder (W)

E100H E 100 13.9(H)

E100M E 100 10(M)

E100L E 100 6(L)

E150H E 150 13.9(H)

E150M E 150 10(M)

E150L E 150 6 (L)

E200H E 200 13.9(H)

E200M E 200 10(M)

E200L E 200 13.9(L)

C100H C 100 13.9(H)

C100M C 100 10(M)

C100L C 100 6(L)

C150H C 150 13.9(H)

49
 C150M C 150 10(M)

C150L C 150 6(L)

C200H C 200 13.9(H)

C200M C 200 10(M)

C200L C 200 10(L)

Figure 3.4. Schematic diagram illustrating positions of 2 collimators

3.1.4 Experiments for implementing Machine Learning concepts to the sputter deposition
process
At this stage of the research, a transitioned was made from using the colour characterisation method
to utilising Machine Learning/deep learning algorithms for the diagnosis and monitoring of the plasma.
The experiments conducted generated a substan•al volume of data, comprising emission data across a
broad wavelength spectrum at specific intervals. Each individual reading comprised a numerical array
represen•ng emission peaks at par•cular wavelengths. To facilitate the management of the overall
dataset, random spectra were selected for training the neural network model for each sample. As
elucidated in Chapter 2, computer vision-based deep learning techniques were u•lized to transform the

50
1D array linked to the spectral data into a 2D tensor. Further elabora•on on the processing of the acquired
raw data will be provided in the subsequent subsec•ons.

3.1.4a Artificial Neural Network to predict the sheet resistance of IZO thin films deposited
via Plasma deposition

Figure 3.5 presents the experimental stages of this trial where the spectral data emitted from the
plasma was captured via a collimator and optical fibre during the sputter deposition of the thin films of
Indium doped zinc oxide (IZO). The sheet resistance value was obtained using the 4-point probe
measurement system post synthesis.

Figure 3.5. Illustration of the sputtering process and the experimental concept. (a) Plasma deposition; (b) Method of
collection of spectral data; (c) emission spectrum; (d) Thin film coating; and (e) Characterisation of sheet resistance

Overall, 114 thin film samples were fabricated. The deposition time and RF power for each run was
1 hour and 300W respectively at room temperature. The substrate was rotating at 10 rpm. The working
gas was an Argon/Hydrogen mix (Argon 95%, Hydrogen 5%) from a single cylinder source. The parameter
that was varied during the coating of the samples was the chamber working pressure which, as mentioned
earlier, can significantly alter the micro-environment of the plasma and sputtered atoms. Thin film
depositions were carried out under the following working pressures: 1 × 10−3, 1.5 × 10−3, 1.9 × 10−3, 2.1 ×
10−3, 2.7 × 10−3, 3.3 × 10−3 and 4.1 × 10−3 mbar.

The data obtained from the spectral map and 4-point probe measurements were plugged into
equation described in 3.5 which will be described in further detail in this chapter. The ‘x’ values
represented the emissions at each 0.2nm step of the spectral data and the y value represented the sheet
resistance of the film.

ANN Modelling
Figure 3.6 provides a visual representa•on of the concepts underlying direct ar•ficial neural
network (ANN) models. In one approach, the spectral data were transformed into vectors, each
containing the area under peaks spaced 100nm apart. Alterna•vely, the emission intensity at 0.2nm

51
intervals within segments of the plasma spectrum was forma•ed. Subsequently, the forma•ed data was
inpu•ed into a deep network of perceptrons. Addi•onally, the sheet resistance value of the TCO,
specifically IZO under examina•on, was associated with each spectral data vector and incorporated into
the neural network to ini•ate the learning process.

Figure 3.6. Illustration of the theoretical concept. (A) In this approach, spectral data of the plasma in form of point-
by-point spectral intensity values (B) in this approach, spectral data of the plasma in the form of area under the peaks
at certain intervals

The model is also fed with the sheet resistance values of IZO, physically measured to enable the
model to learn from the spectral data to predict the sheet resistance value. Once the model was fed with
enough data with known sheet resistance values and sufficiently trained, it was then exposed to spectral
data without the known outcome to predict the thin film conductivity. This was entirely based on emission
spectrum of the plasma during the deposition process of the film.

CNN Modelling
In a separate set of experiments, the spectral emission of the plasma was converted into an image
by translating the vectors associated with the emission intensity into a matrix. The matrix was then
rendered into an image whereby each of the spectral points represented a pixel of the image. These
images were then fed into convolutional neural network (CNN) so that the learning process could take
place and subsequently predict the sheet resistance values of the films with respect to its specific
deposition conditions. CNNs derive their names from the type of hidden layers it consists of i.e. its
convolutional and pooling layers to function as opposed to the use of activation functions in conventional
feed-forward neural networks. CNN is a technique that allows for the extraction of visual features from a
two dimensional (2d) array in small chunks. Each perceptron in a convolutional layer is responsible for a
small cluster of layers in the preceding layer and the bounding filter that extracts features and determines
the cluster of neurons/perceptrons from the initial signal called a kernel. In the experiments that were
conducted, the plasma emission spectrum was converted into an image and certain type of features were
extracted from this image using an array of random kernels in a CNN model at which point these were
converted into a flat single vector and subsequently used for regression analysis (figure 3.7). The kernels
were selected randomly during the model training and were learned via gradient descent process.
Ultimately, the convolutional method was designed to downsize the dataset and pick out specific features.
Mathematical operation and structure of these models was beyond the scope of this work, but further
discussion can be found in [147-150].

52
Figure 3.7. Illustration of implementation of a convolutional neural network linked to a regression multilayer ANN
model for predicting the sheet resistance of the thin films from an image generated from the spectral emission of the
plasma.

Several samples were prepared under the above pressures and the spectral data from the plasma
were gathered. After each deposition, the sheet resistance of the films was measured using a Jandel
RM3000 four-point probe system, the operating principle of which will be described later on.

The data were tabulated such that each row represented a sample, The last column denoted the
calculated sheet resistance, while the other columns, arranged from start to end, represented the signal
intensity calculated by the spectrometer at each emission wavelength point with a precision of 0.2 nm.

Figure 3.8 illustrates the spectral peaks of the plasma under four different pressures in a compara•ve
manner. It visually presents the emission peak spectra and their corresponding intensi•es, comparing 2.1
mbar against 3.7 mbar and 1 mbar against 4.1 mbar. The goal is for the model to assimilate the spectral
characteris•cs of the plasma and forecast the sheet resistance of the indium-doped zinc oxide thin films
solely based on these spectral features, thereby obvia•ng the need for complex plasma physics models.

Figure 3.8. Spectral emission from the sputtering plasma under various chamber pressures. Comparing chamber
pressures of 2.1 mbar and 3.7 mbar (left) and comparing 1 mbar and 4.1 mbar chamber pressures (right).

The direct image generated from the spectra was initially utilised to train various convolutional
neural network (CNN) models. However, none of these models could effectively learn from the image in
its raw form, as CNN models operate more efficiently with standardized values. Subsequently, the pixel
values of the image were normalized using two different methods: min-max scaling and standard scaling.

53
 H"IJK (H)
𝑥 (min max 𝑠𝑐𝑎𝑙𝑒𝑑) = INO(H)"IJK (H) (3.1)

H"µ
𝑥(𝑠𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑠𝑐𝑎𝑙𝑒) = (3.2)
s

Therefore, the following approaches for preparing and applying the data to train the neural network
models were pursued: (a) an ANN-based integral approach with 50 nm spectral windows, (b) an ANN-
based spectral approach and (c) a CNN-ANN-based spectral approach that deviated into computer vision
and image recognition. In this approached, the spectral data was transformed into two-dimensional scaled
images, incorporating an advanced computer vision techniques with deep learning convolutional neural
network architecture for image recognition model. This was coupled to a neural network model for
regression analysis (Figure 3.9).

Figure 3.9. Conversion of the spectral plot of the plasma into 2D images formed from scaling the spectral values. (a)
the original spectra, (b) image directly formed from original spectra, (c) image formed from the standard scaling of
the original spectra, (d) an image formed from the min max scaling of the original spectra.

In the preceding step, as each image was technically a matrix of size (50, 90), singular value
decomposition (SVD) to every single image was applied. The SVD decomposed the image matrix into three
matrices: a U matrix, which spanned the column space of the image; a V matrix, which spanned the row
space of the image; and a diagonal matrix holding singular values that were scalers associated with
consecutive columns of U and V. The whole image was able to be reconstructed from the out product of
U and V scaled by the diagonal singular value matrix. This is illustrated in figure 3.10.

54
Figure 3.10. The singular value decomposition and reconstruction of a matrix.

The vectors within U and V were unit vectors. The corresponding singular value served as a scalar,
providing magnitude to the outer product of these vectors. The condi•on number of a matrix reflects
the dispersion of informa•on within it and can be computed by dividing the highest singular value of the
matrix by its lowest singular value.

Deep learning necessitates vast datasets. Large quan••es of data, comprising feature columns and
corresponding output columns, are essen•al for training these models. Therefore, the primary objec•ve
of this research was to accumulate a sufficiently large dataset by con•nuously opera•ng a spu•ering
deposi•on system, capturing spectral data, and measuring an output value associated with the
experimentally prepared samples—in this instance, the sheet resistance of the thin films. The data u•lized
in this chapter were obtained through con•nuous opera•on of the machine over a one-year period. The
substan•al progress made in deep learning models owes much to the extensive ongoing research efforts
of data science researchers, which have made these models powerful and straighŒorward to implement.
Nevertheless, the founda•onal requirement for their implementa•on remains large datasets,
underscoring the centrality of data acquisi•on in this research endeavour.

3.1.4b Experiments for Implementing unsupervised deep learning methods to predict

sputtering plasma features
The next set of experiments conducted is a follow-up of initial research work carried out on applying
deep learning concepts to the sputter deposition process. Unsupervised learning; a more intricate and
multifaceted machine learning concept is explored whereby the spectral data collected from the plasma
is used to trained models without any well-labelled data or known outcomes. Once the model was trained,

55
attempts were made to use it for accurately predicting the growth conditions of the plasma depositions
associated with a particular emission spectrum.

The spectral emission data from the plasma glow of various sputtering targets containing indium oxide,
zinc oxide, and tin oxide were obtained as detailed in section. The plasma was generated at various power
and chamber pressures. These spectral data were then converted into two-dimensional arrays by
implementing a basic array-reshaping technique and a more complex procedure utilizing an unsupervised
deep-learning technique, known as the self-organizing-maps method. The two-dimensional images
obtained from each single-emission spectrum of the plasma mimic an image that can then be used to train
a convolutional neural network model capable of predicting certain plasma features, such as impurity
levels in the sputtering target, working gas composition, plasma power, and chamber pressure during the
machine operation.

In these experiments, seven different targets were fitted:

1. Pure zinc oxide,

2. Pure indium oxide
3. IZO (indium oxide, zinc oxide) 70:30 ratio
4. IZO (indium oxide, zinc oxide) 60:40 ratio
5. IZO (indium oxide, zinc oxide) 50:50 ratio
6. IZO (indium oxide, zinc oxide) 40:60 ratio
7. ITO (indium oxide, tin oxide) 90:10 ratio
With each of these targets fitted into the magnetron, a plasma was generated under various
operating conditions, whereby power, gas-flow rate, and working gas purity (using either pure argon or
argon with 5% hydrogen content) were altered. With each experiment, collected the spectral data
associated with plasma glow was collected.

The primary objective was to investigate how accurately a deep-learning model (convolutional
neural network models) trained with the spectral data would predict the following parameters using
only the spectra of a sputtering plasma:

1. Generation power;
2. Gas-flow rate;
3. Level of indium and zinc content in the target;
4. Presence of hydrogen in the chamber;

The spectral data from the plasma were collected by fitting seven different targets in the magnetron and
collecting spectral data under varying plasma powers and gas-flow rates, as depicted in Table 3.5. Similar
to the 1st deep learning experimental method, the spectral data were collected via a collimator fitted next
to the magnetron. At each single point, spectral data were collected, to have a collection of spectral data
associated with plasma operating conditions. Overall, a total of 588 spectral data plots were selected for
deep-learning studies.
Table 3.5 Different Plasmas generated using a combination of a different deposition parameters.

Material ITO, IZO, ZnO & InO

56
 Working gas Ar & Ar/H

Plasma power (W) 100, 150 200

Gas flow (sccm) 3, 5, 6, 7, 8, 9 & 10

Data collection
These experiments generate a large quantity of numerical data. The spectral emissions were
collected from 204 nm to 1104 nm at 0.2 nm intervals, which means each single reading was an array of
4500 numbers representing emission peaks at particular wavelengths. For each sample, 2 random spectra
were selected for training the neural network model. Therefore, the overall data set was a matrix of 588
rows and 4500 columns.

As the intent was to use computer-vision-based deep learning to create the model, there was a need
to convert the 1-dimensional array associated with the spectral data into a 2dimensional tensor.

Two approaches were taken to convert the 1-dimensional array into 2D:

1. Basic cut and stacking. As illustrated in 3.11a, the 4500 long array associated with the
spectra was cut into 50 segments and stacked to form a matrix of 50 × 90 pixels, as illustrated
in 3.11b.
2. By implementing an unsupervised machine-learning technique called self-organizing map
(SOM), forming a 2-dimensional array of 90 × 90 pixels. The SOM illustrated in figure 3.12 is an
algorithm that was introduced by Teuvo Kohonen, acting as a non-supervised clustering algorithm
capable of presenting multidimensional feature data into a two-dimensional map of clusters
[151].

57
Figure 3.11 (a)The basic cut-and-stack method of converting a 1-dimensional array to a 2-dimensional array. (b)The
spectral data were converted into a 2-dimensional matrix.

Figure 3.12 Using the SOM unsupervised deep-learning model to generate a 2-dimensional array from the spectral
data. The 1-dimensional spectral data array was fed to SOM model and projected onto a 2-dimensional feature map.

It is important to highlight the fact that prior to forming the 2-dimensional matrices from the spectral
data, they were scaled via min-max scaling method as per equation 3.1.

Once the 2-dimensional images were formed, a singular value decomposition (SVD) on the 2-
dimensional matrices was carried out and then reconstructed the matrix by first layer of the SVD model
as discussed previously.

Therefore, every spectral datum that related to a particular target at particular RF plasma power and
gas-flow rate and specific working gas was converted into four separate images. This family of images was
then used to train a convolutional neural network (CNN) model. Figure 3.12 illustrates the formation of
four separate images from every spectral datum. Given that there were 588 spectral data, it quadrupled
as there were 4 image groups (588X4), which were used separately to train a CNN.

58
Figure 3.12 Each spectral datum was converted into four images; a min-max-scaled 2-dimensional array and its SVD
reconstruction of first layer and a SOM-generated 2-dimensional array and its SVD reconstruction of the first layer.
The arrows represent the routes to the methods of conversion (original and SOM).

To appreciate how difficult it was to distinguish the glow of the plasma under various conditions,
some plots of the plasma spectrum are presented in the following figures: Figure 3.13 presents the
emission spectrum associated with 9-, 7-, and 3-standard-cubic-centimeter-per-minute argon flow rates
overlayed (A) and also in a segregated format (B). In figure 3.14, the effect of the presence of 5% hydrogen
in the argon gas in overlay and segregated form is presented. In figure 3.15, the emission spectrum
associated with the four different target materials is presented. These spectra reveal some differences if
precise attention is paid. However, these examples are presented here to illustrate the complexity and
precision of the spectra in detail.

Figure 3.13. (A) Overlay of spectral data at 9, 7, and 3 sccm. (B) Segregation of the spectra.

59
Figure 3.14. (A) Overlay of spectral data relating to pure argon and argon with 5% hydrogen impurity. (B) Segregation
of the same spectra. (C) Section of the spectra magnified to illustrate the miniscule variation in the spectral data.

Figure 3.15. (A) Overlay of spectral data associated with four example target materials. (B) Segregation of the spectra.

The objective was for the CNN model to be able to calculate certain parameters, such as the gas-
flow rate, plasma power, and presence of hydrogen, through computer vision assessment of these
spectra. This is further illustrated in figure 7.9, where a section between 400 and 600 nm is magnified to
illustrate the difference between spectra of the same target material at same plasma power and pressure,
but with 5% hydrogen impurity in one.

The CNN Model

The four classes of image forms were used separately to train the CNN model. The structures of the
CNN model for regression analysis and classification analysis were the same except for the final layer. The
structure of the CNN model is presented in figure 3.16. The image data were submitted to two steps of
convolution and max pooling, consecutively, prior to being flattened into a single vector. The vector was
then used to train an artificial neural network model with four dense layers containing 5000, 5000, 100,
and 20 neurons each, all bearing a rectified linear-unit function.

For regression analysis, a final dense layer with a single neuron was used for predicting the gas-flow
rates and plasma power. The loss function used for the model training was mean squared error.

For classification analysis, the final dense layer with a single neuron was set up with a sigmoid
activation function. The loss function used for model training was binary cross entropy.

60
 Both the regression and the classification models utilised an Adam optimiser with a learning rate of
0.001.

Figure 3.16. The structure of the CNN model.

A vital part of this study was its focus on the treatment of the numerical data prior to model training.
It is a well-known fact that for all deep-learning models, the data need to be scaled or standardised.

However, a secondary step in scaling the data was carried out. For each emission spectrum, the peak
intensities were summed, and every intensity to the resolution of 0.2 nm was divided by the total sum.
The importance of this step is vital to the performance of the models. This operation was performed
specifically because the temperature of ions and electrons in the plasma are directly proportional to their
random average kinetic energy, while Maxwell distribution governs the distribution of velocities for each
particle when thermal equilibrium conditions apply. This has been well detailed in chapter 2.2.

This method is the simplest approach to calculate the temperature of the plasma, and its accuracy
is conditional on using two lines, with a maximum difference in their upper energy states. Therefore, there
are hidden patterns associated with each individual plasma when ratio of peak intensities is considered.
Therefore, by implementing the operation of representing each 0.2-nanometer peak intensity as a fraction
of the total intensities, it provides the neural network model with hidden features related to the
microenvironment of the plasma for the model training.

The glow from the plasma is a representation of the plasma ingredients, its constituents, and their
states. It can be assumed that a vast amount of information is embedded in the plasma glow. The objective

61
conducting the experiments associated with deep learning is to extract and exploit this information and
correlate it with the parameters that are implemented during plasma formation. As these parameters
ultimately govern the quality and properties of the thin films produced via this method, this is a step
toward defining a digital shadow for the sputter-deposition system.

3.2 Characterisation Techniques

3.2.1 Four point probe
The 4-point probe is a simple apparatus is used to quickly determine the sheet resistance of a sample.
As the name suggests, it consists of 4 probes which are aligned with one another (and 1mm spacing in
between) and it measures the resistivity of semiconductor samples by applying a fixed current (I) through
the two outer probes and measuring the voltage (V) through the two inner probes [152]. This is illustrated
in figure 3.17.

Figure 3.17. Schematic diagram of four-point probe setup

The measurement is taken in both forward and reverse mode, where the direction of the current is
altered allowing an accurate reading. This allows the measurement of the samples’ sheet resistance which
can then be used to determine resistivity. Resistivity (r) (W.cm) for a semi-infinite volume is given by eq.
2.1 [153].
P
r = 2ps h Q i (3.3)

Where ‘s’ is the spacing in between the probes and ‘t’ is the thickness of the sample. This applies for
bulk samples where the sample thickness is greater than the probe spacing in which case the current
spreading from the outer probe is modelled as a sphere.

There are a number of factors that can affect the accuracy of the measurements. The probes must
make ohmic contact with the material; if the material has a very low resistance and it is significantly thick,
it will be immeasurable for the instrument due to the limited range of current source. Thus, for conducting
materials, only accurate readings can be taken off of thin film samples. In addition to the thickness, the
shape of the material can also contribute towards measurement errors. It is important to implement a
correction factor to the samples sheet resistance to take into account the ratio of probe spacing to film

62
thickness/sample diameter. This factor can be determined via a thickness correction factor plot found
elsewhere [154].

In the most cases, thickness will be significantly less than that of the probe spacing (thickness, t<<s)
in which case the correction factor is applied as the current will spread out in rings rather than in sphere:

P R
r = 4.53t h Q i , h,i ≥ 0.5 (3.4)

For this project, the Jandel Model RM3 4-point probe system was employed to measure the sheet
resistance.

3.2.2 Hall Effect

The principle of operation of the Hall effect is based on the van der Pauw method. It is a measurement
technique that accurately measures the resistivity of an arbitrarily shaped thin film with 4 small ohmic
contacts placed on the corners of the sample.

Figure 3.18. Schematic diagram of a van der Pauw configuration used to determine characteristic resistances RA and
RB [154]

Van der Pauw method demonstrates that there are two characteristic resistances RA and RB
corresponding to the terminals as shown in figure 3.18. The characteristics resistances are determined
by passing current through 1 set of contacts and measuring the voltage from the other two. As per the
van der Pauw configuration, RA and RB are calculated via the following expressions [154].

P&'
𝑅S = Q%"
(3.5)

63
 P%&
𝑅T = Q"'
(3.6)

RA and RB are related to the sheet resistance via the following equation:

UV( UV*
𝑒𝑥𝑝 h− V)
i + 𝑒𝑥𝑝 h− V)
i=1 (3.7)

From this, the bulk resistivity ρ can be determined via:

𝜌 = 𝑅, 𝑡 (3.8)

To perform the van der Pauw method, the following requirements have to be fulfilled:

• Sample is flat and uniformly thick

• Sample is homogenous and isotropic
• Contacts placed at circumference/edge of sample
• Area of contact should be at least an order of magnitude smaller than that of sample
• Surface does not contain any pinholes.

The Hall effect describes the behaviour of the free carriers in a semiconductor material upon the
application of not only an electric but also a magnetic field. The objective of the hall effect is to determine
the sheet carrier density ns, by measuring the Hall voltage. If a constant electric current flows through a
conductor in the presence of a magnetic field perpendicular to the plane of the sample, there is a
transverse force exerted onto the moving charge carriers by the magnetic field which pushes them to one
side of the conductor. This leads to a build-up of charge at both sides of the conductor which balances the
magnetic influence, producing a measurable voltage between the two sides of the semiconductor [152].
This measurable transverse voltage is known as the Hall voltage. In addition to the carrier density, the
measurement of this Hall voltage is also used to determine the nature of conductivity and the carrier
mobility.

64
Figure 3.19. Schematic diagram of van der Pauw configuration used in the determination of the hall voltage VH [155]

Figure 3.19 provides a visual representation of the Hall effect. To measure the Hall voltage, VH, a
current, I13, is passed through opposing pair of contacts i.e. 1 and 3 and the Hall voltage (V24) is measured
across the other pair contacts i.e. 2 and 4 as the sample is subjected to a perpendicular magnetic field.
Once the VH is determined, the sheet carrier density/concentration can be calculated via the following
formula:

QT
𝑛, = (3.10)
W|P+ |

from the known values of I, B, and q where I is current, B is magnetic field strength q is the elementary
charge.

The Hall effect system at LSBU consist of discrete hardware of bespoke design allowing for accurate
measurements. Whilst the majority of the hardware is commercially available, the configuration of the
system was designed and custom-built consisting of the following components:

• An Oxford instruments, water cooled variable pole geometry electromagnet (Model N100) which was
responsible for generating the magnetic field. This was powered by a PC controlled power supply
(Sorenson XG-100-15) and is capable of achieving a peak field strength of 1 tesla.
• A custom PC controlled switching unit (developed in-house) to allow for automated reversal of the
magnet polarity.
• Keithley measurement system which provided current sourcing, monitoring and voltage
measurements which were PC controlled via a GPIB cable. This consisted of:
-A model 7065 Hall effect matrix switching card, housed in a model 7001 switch mainframe which
automates the multiple source and measuring configurations applied to samples under
examination

65
 - A model 6220 current source capable of supplying current ranging from 100fA to 100mA allowing
for measurements of samples with wide resistivity ranges.

-A model 2182A nano-voltmeter capable of measuring voltages in the range of 1nV to 100V with
a high degree of accuracy.

• Control of the various hardware components and data logging were achieved via a bespoke software
solution by Semimetrics Ltd allowing for almost complete automation of the Hall effect measurement
process upon configuring the initial conditions.

3.2.3 Spectrophotometry
Spectrophotometry is the quantitative measurement of the reflection and transmission properties of
a material as a function of wavelength. A spectrophotometer is an instrument that is used to measure the
intensity of wavelengths in a spectrum (UV-VIS-NIR) of light compared with the intensity of light from a
standard source. This beam is chopped and diffracted to discreet wavelengths. The light beam is passed
through a sample and the intensity of light that reaching the detector is measured [156].

Figure 3.20. Working principle of double beam spectrophotometry [157]

A typical spectrophotometer such as the Varian Cary 500 is capable of producing and detecting light
from the UV range, through the visible and into the short-wave IR regions (175-3300nm). It makes use of
a pair lamps i.e., tungsten and deuterium and a series of shutters and grating to generate the
monochromatic light in nm steps. A double beam UV visible spectrophotometer is typically used to
determine the transmission and reflection properties of the samples as seen in figure 3.20. The light
source strikes the monochromator which selects the light beam with specific analytical wavelengths from
the source. This is achieved by allowing the polychromatic light to enter the slit of the monochromatic
instrument and only allowing light with a single wavelength (consisting of a small region of the
electromagnetic spectrum) to pass through via the exit slit. As mentioned, light is passed through the
system with one beam being fed directly to a detector as a reference measurement and then the second
being directed to the measurement area. This beam is then navigated through an array of fixtures
depending on the requirement of the measurement before it returns to a 2nd detector. The detector
converts the arrival of photons in the light rays into an electrical signal which allows for the measurement
of transmission and reflection and subsequent calculation of absorption [158].

66
 Optical properties of the thin films were carried out via a UV-VIS spectrometer (Bibby Scientific Ltd,
Staffs, UK) by using the absorption data. Tauc plots were made to measure the optical bandgap of the
material. The plot was made by plotting the hv(a-axis) against the (α-hv)1/2 (y-axis), where the α is the
absorp•on coefficient of the material.

3.2.4 Stylus Profilometer

A surface profilometer was employed to measure the thickness of the samples. The mechanically
based profilometer possesses a stylus which detects the surface and traverses along it whilst maintaining
contact in order to acquire changes in the surfaces’ height. A feedback loop monitors the mechanical force
from the sample that is exerted onto the probe as it scans along the system. The profilometer uses a
feedback system to keep the arm at a fixed setpoint (i.e. torque) and adjustments in the z-position of the
arm holder to maintain the setpoint can be used to reconstruct the surface [159] which can be seen in in
figure 3.21.

Figure 3.21. Schematic diagram of a basic stylus profiler

The samples were tested for film thickness via a profilometer (Veeco, NY, USA); this was carried out
by scanning the stylus of the profilometer over a one mm strip of deposited films on a soda-lime glass
substrate. A removable strip of Kapton tape was used to shield a particular area of the sample from film
deposition creating a step which the stylus profilometer scanned over for the measurement of film
thickness.

3.2.5 Scanning Electron Microscopy (SEM)

A scanning electron microscopy is a powerful instrument used to direct an electron beam from the
cathode at the surface of a sample (mounted onto an anode) which raster scans the surface to form an
image. The SEM provides information about the surface morphology of the sample [160]. The interaction
between the electrons in the beam and the surface of the samples results various being produced which
can be used to obtain in-depth information about the surface topography and composition.

67
 The SEM works by producing highly energised electrons at the top of the column and accelerating
them down which then pass through a series of magnetic lenses and apertures to yield a focused beam of
electrons until it finally reaches the surface of the sample. The stage that the sample is mounted on is
situated in the chamber area and in most cases both the column and the chamber are operated in a
vacuum; the level of vacuum depends on the design of the microscope.

The position of the electron beam is controlled by a pair of scan coils which sits above the objective
lens; these coils allow the beam to scan the surface of the sample in a raster fashion revealing information
about a defined area on the sample and depending on the signal produced from the sample-electron
interaction, the information is collected by appropriate detectors [161, 162].

These primary electrons, upon impacting the surface, transfer their energy into the surface in a tear
drop shaped area of interaction. As the electrons in the beam interact with the sample, various signals are
ejected from the sample i.e. secondary electrons, backscattered electrons and characteristic X-rays as
shown in figure 3.22. These signals are collected by several detectors to produce an image displayed on
the computer screen. The sample is penetrated by the electron beam by a few microns which varies with
the accelerating voltage and density of the sample.

Figure 3.22. Schematic diagram of the sample-electron interaction ejecting various signals

Since the SEM relies on a beam of electrons to image, it is essential that the samples being investigate
are charge free as it can result in distortion and/or image drift during high resolution image capture,
which is take at a slower raster scan rate. If the sample is of a non-conductive nature then it is usually
sputter-coated with thin film of gold (<5nm) beforehand to ensure a conductive path is created without
comprising surface features [163].

3.2.6 EDS
Energy-dispersive X-ray spectroscopy (EDS) is a powerful tool that allows the analysis of the
elemental composition of the desired thin film sample. The principle of operation is the capacity of high

68
energy electromagnetic radiation (i.e. X-rays) to eject core electrons (i.e. inner shell electrons) from an
atom in the sample. This principle is known as Mosley’s Law which states that there is a direct correlation
between the frequency of light release and the atomic number of the atom.

Ejecting these inner shell electrons from the atom will results in a vacant site or a hole which can be
occupied by a higher energy electron; it will release energy as it relaxes into the hole. The energy that is
released during the relaxation process is unique to each element found on the periodic table and the
bombardment of a thin film sample with X-rays permits the identification of elements and to which degree
these are present.

A schematic diagram illustrates the principle of operation of EDS (figure 3.23). The letters K, L and M
signify the ‘n’ value that electrons in that shell occupy. For K electrons, which occupy the innermost shell,
n=1 while α and β denote the size of the transition. For example, relaxation of an electron from L to K is
described as Kα while going from M to K would be defined as Kβ transition. This method used to describe
the process as a whole is known as Siegbahn notation.

Figure 3.23. Schematic diagram illustrating the ejection and relaxation of electrons in an atom [164]

The films prepared via the experiments outlined in 3.1.2 were consequently characterised using a
Zeiss Supra VP35 Field Emission Gun Scanning Electron Microscope equipped with an EDAX Octane Super
Energy Dispersive Spectrometer (Ametek, Darmstadt, Germany) (Energy-dispersive X-Ray spectroscopy,
EDS) to evaluate the homogeneity and morphology of the film and gather indication on the film chemical
composition, in particular the In/Zn ratio (samples S1 to S7) or level of contaminants in the pure ZnO films
(samples H0, H3 and H6). The films were also investigated using diffraction techniques, both electron
backscatter diffraction (within the Zeiss Supra FEG-SEM equipped with an EDAX DigiView EBSD camera,
Darmstadt, Germany).

3.3 Summary
An array of thin film IZO and hydrogen doped ZnO samples with varying growth conditions were
deposited via RF magnetron sputtering. For experiments involving the analysis of the plasma via the colour
characterisation method, the glow discharge during deposition was exposed to two different
spectrometers as discussed in section 2.1, one to calculate the chromaticity of light and the other to

69
acquire the emission spectrum. For experiments pertaining to machine learning, the spectral emission of
the plasma was converted into a format that was interpretable for a number machine learning/ deep
learning models. The well-trained models were given the task to predict certain features of the resultant
thin film samples and growth conditions associated with a particular emission spectrum. The thin film
samples were subjected to surface, opto-electronic and electrical characterisation.

70
Chapter 4 Optical Analysis of RF Sputtering Plasma through Colour
Characterization

In this chapter, as mentioned in the literature review (Chapter 2.3), a novel strategy of the in-situ
characterisation of plasma in a sputtering process has been reported by translating the emission spectra
into a colour indexing (chromaticity) pattern that is able to quickly monitor any changes in the plasma
character with a change in the operating parameter conditions during the course of the process, which in
turn provides a better handle to reflect the instabilities of the tool and its optimising parameters. The
materials and methods employed for this experiment can be found in 3.1.1.

4.1 Results
4.1.1 Chromaticity of the Plasma Emissions
The emission characteristics in terms of the whole spectral plot and the chromaticity index were
obtained for each test group.

Test 1, from the moment that the plasma is ignited, the spectral data can be obtained from it.
However, by plotting the chromaticity x, y values of the plasma light from a cold start, these values will
keep varying for up to 2 h. After 2 h, the variance in the x and y values that were obtained will significantly
reduce; this is the point the plasma can be considered to be stabilized or more stable. figure 4.1 presents
this observation. The x and y data obtained from the plasma during the first hour of ignition at 100 W for
a magnetron fitted with Indium Zinc Oxide (IZO) target under gas Ar (A,B) and gas ArH (C,D) along with
data obtained 2 h after ignition. During this 1 h interval the plots remain relatively stable. The stabilized
data were gathered over an additional thirty minutes period. As it can be seen (A,C) prior to stabilisation
the x,y coordinates are scattered across the plot area until the plasma is stabilized at which point the plots
fall within close proximity (B,D).

(A) (B)

(C) (D)

71
 Figure 4.1. The x and y data obtained from the plasma during the first hour of ignition at 100 W for a
magnetron fitted with Indium Zinc Oxide (IZO) target under gas Ar (A,B) and gas ArH (C,D).

To demonstrate the complexity of relying on the conventional methods, such as the peak intensity
ratio method, to monitor the progress of the plasma toward stability, the spectral plot of the 100 W Argon
plasma over a period of one hundred and thirty minutes post-plasma ignition are presented in figure 4.2.
As an example, particular transitions that were previously evaluated by other authors for measuring
electron temperature and density were chosen to measure these data [165]. The ratio plot that is
presented in figure 4.2A is related to the intensity ratio of 404.44 nm (Ar-I) transition 3s23p5(2P°1/2)5p
2[3/2] 2 to 3s23p5(2P°3/2)4s 2[3/2]° 1 and 426.62 nm (Ar-I) transition 3s23p5(2P°3/2)5p 2[3/2] 2 to
3s23p5(2P°3/2)4s 2[3/2]° 1. The ratio plot presented in figure 4.2B is related to the intensity ratio of
480.60 nm (Ar-II) transition 3s23p4(3P)4p 4P° [5/2] to 3s23p4(3P)4s 4P [5/2] and the 470.23 nm (Ar-I)
transition 3s23p5(2P°3/2)5p 2[1/2] 1 to 3s23p5(2P°1/2)4s 2[1/2]° 1. It can be seen that it will be
challenging to apply the peak intensity method as an instant real time monitoring technique. This
highlights the novelty of using colour function as a mean to monitor the plasma stability as the
chromaticity plot of the x and y values clearly demonstrates a linear path toward the stable x, y
chromaticity values.

(A) (B)
Figure 4.2. The intensity ration of four peaks are presented over 130 min. (A) The intensity ratio of the (Ar
I) 404.44 nm and the (Ar I) 426.62 nm emission peaks. (B) The intensity ratio of the (Ar I) 470.4 nm and the
480.60 nm emission peaks.

Test 2, the RF power that was used to ignite and run the plasma is the source of energy for ionising
the working gas and driving the sputtering process. Higher energy will result in more excited and
metastable atoms to be created, and consequently will increase the bombardment of the target surface.
During this part of the investigation, after stabilising the plasma at 100 W, the power was increased up to
300 W ( at a rate of 100W per minute) and data was collected at ~9 s intervals under argon. A clear linear
relationship can be observed between the chromaticity value and the plasma power, as presented in
figure 4.3. The values of x and y decrease with increasing the RF power and a shift toward blue in a linear
manner is also observed.

72
 Relationship between plasma power and
chromaticity
0.28 100 W

0.26

0.24
Y

0.22 300 W

0.2

0.18
0.3 0.31 0.32 0.33 0.34 0.35
X

Figure 4.3. The relationship between the power of plasma (radio frequency (RF)) and the x and y values.

In Test 3, an interesting pattern is observed by increasing the pressure of the chamber under each
Gas (Ar or ArH), after the plasma has been stabilized, which is highlighted in figures 4.4 and 4.5. The test
was started with the lowest pressure achievable based on the mass flow controller of the machine (1.2 ×
10−3 mbar) and then the pressure was increased. With increasing the pressure, the x and y values start to
increase, but, at a certain point, a loop is made (a U-turn shape) and it starts to decrease with higher
pressure indicating that the colour is initially shifting toward white and then reverting toward blue.

Table 4.1. The x and y data obtained by Jeti spectrometer from the plasma at 100 W, and 2 h after stabilization
for a magnetron fitted with IZO target under gas Ar at various pressures

Pressure
x y
(x10-3 mbar)

1.20 0.3396 0.2634

1.50 0.3439 0.3668

2.00 0.3481 0.2678

2.50 0.3498 0.2691

3.00 0.3500 0.2677

3.50 0.3505 0.2665

4.00 0.3505 0.2648

4.50 0.3500 0.2640

5.00 0.3487 0.2625

5.50 0.3485 0.2614

73
 6.00 0.3479 0.2606

(A)

(B)
Figure 4.4. (A). As the Argon pressure is increased, the x and y values increase until about 2.5 × 10−3 mbar
and then start to decrease. (B) The x and y plot demonstrates a U-turn formation.

Table 4.2. The x and y data obtained by Jeti spectrometer from the plasma at 100 W, and 2 h after
stabilization for a magnetron fitted with IZO target under gas ArH at various pressures

Pressure x y

74
(x10-3 mbar)

1.20 0.3361 0.2609

1.50 0.3429 0.2663

2.00 0.3480 0.2689

2.50 0.3497 0.2691

3.00 0.3502 0.2677

3.50 0.3506 0.2672

4.00 0.3512 0.2670

4.50 0.3495 0.2646

5.00 0.3489 0.2642

5.50 0.3490 0.2624

6.00 0.3488 0.2615

6.50 0.3482 0.2612

7.00 0.3470 0.2590

(A)

75
 (B)
Figure 4.5. (A). As the Argon Hydrogen mix pressure is increased, the x and y values increase until about
2.5 × 10−3 mbar and then start to decrease. (B). The x and y plot demonstrates a U-turn formation, indicating
that the colour is initially shifting toward white and then reverting toward blue .

The spectral plot of the plasma emission under various pressures clearly shows an increase in the
intensity of some of the transition lines with increasing the pressure, which ultimately affects the x and y
reading (for example, the data obtained from IZO target and Argon gas at various pressures is presented
in figure 4.6).

(A)

76
 (B)
Figure 4.6. The effect of working gas pressure observed on the spectral plot (200 to 600 nm) of the plasma
with IZO target and Argon gas only at various working gas pressures.

When compared to the x and y plots in figures 4.4 and 4.5, it can be observed from the spectra that
there is an increase in the intensity of certain peaks which makes the assessment of the data complex
with further evaluation. All of these increases in intensity can be summarised as presented in figures 4.4
and 4.5 by evaluating the colour of the plasma discharge. The inset in the second section of the plot
demonstrates the complexity of analysing every single peak for the amount of slight increase in one of
the peaks. (A): spectral plot 200 to 400nm; (B): spectral plot from 400 to 600nm.

The rise in the intensity of all the peaks is simply due to the larger number of excited atoms and well
known, however, the exact mechanism by which the x, y values make such a U-turn pattern is possibly
related to complex switching of the electronic transitions, and hence it is related to the area under
emission peaks of certain transitions across the spectrum and these peak areas are related to the
microenvironment of the plasma and the interaction of the plasma constituents, distribution of energy,
electron temperature, and charge densities involved, but the exact mechanism needs further attention
and research, this suggests that there is a possibility of correlating the colour to the plasma’s physical
state.
This is currently being investigated by the authors through the monitoring of the above-mentioned
factors affecting the plasma constituents and will be reported separately. Here, the core of the current
communication is a focus on the currently observed data in terms of colour indications.

Finally, another particular observation was made; once the IZO target material is exposed to gas ArH
and under a stable plasma, switching from ArH to Ar does not lead into a direct swift stable reading, rather
for a period up to ~1 h, the values of x and y fluctuate. The fluctuation is only observed when going from
gas ArH to Ar, and not from Ar to ArH under the same protocol, which indicates the possible incorporation
of hydrogen into the target material similar to the target poisoning process, however this cannot be
verified prior to close investigation of the target material. figure 4.7 presents this observation.

77
 Figure 4.7. Chromaticity (x, y) diagram of:. (A) Switching from Ar to ArH occurs swiftly. (B) Switching from
ArH to Ar,

The plasma colour demonstrates a colour change as the x,y values change over the course of ~1 h
prior to stabilisation in a linear fashion after the target has been exposed to Argon/Hydrogen mixture.
This observation can be interpreted as either the rate by which hydrogen is disappearing from the
chamber and/or possible formation of target hydrogen compounds at the surface of the target
(target poisoning by hydrogen) and the time that is required for the target to revert back its original
state while switching from Ar to ArH does not demonstrate any pattern, rather a swift instant jump
is observed.

4.1.2 Optical Emission Spectroscopy of the Plasma Emissions

In the previous section, The focus was on the chromaticity (colour) of the emission, parallel to these
studies Results from optical emission spectroscopy of the plasma was obtained, which are associated with
the experiments involving only the Argon gas as a comparative guide to enable to discuss the observations
seen through chromaticity analysis.

During Test 1, the stability of the plasma was monitored through the colour functions, at the same
time, the optical emission data of the spectra was obtained and divided the spectrum to seven sections:
a, b, c, d, e, f, and the UV. Figure 4.8 presents these sections; the UV covered emissions from 300–400
nm, a: 400–430, b: 431–449, c: 450–500, d: 500–600, e: 600–700, and f: 700–800 nm. The area under each
region was calculated and monitored. The area via integral under each peak region was then used for
further analysis of the data to monitor the variations of the area as a dependent of various experimental
parameters that was already discussed; duration of emission stability, RF power applied to the magnetron,
and the various operating Argon pressures.

Figure 4.8. The emission spectra of the plasma were segmented into seven sections. The area under each
region was calculated and monitored.

78
 (A)

(B)
Figure 4.9. The emission spectrum and the associated Argon, Indium, Zinc and Oxygen states within far
infra-red (A) and the UV section (B) of the spectrum obtained from the sputtering plasma.

79
(A)

(B)

(C)

80
 Figure 4.10. The emission spectrum and the associated Argon, Indium, Zinc, and Oxygen states within the
400–700 nm section of the spectrum obtained from the sputtering plasma. Spectral range, (A):400-500nm,
(B)500-600nm, (C):600-700nm.

The area under all of the regions was summed, and then the ratio of the area under each section of
the study was calculated as a percentage of the total area under the peaks to visualise how the emissions
from the associated transition changed during plasma stabilisation and when the parameters, i.e., power
and Argon pressure, were changed. In figure 4.11, the stability of the plasma over time is monitored over
a total period of 5000 s at intervals 1000 s, e.g., T1: 0 to 1000 s, T2: 1000–2000 s, etc.
It can be seen that some of these regions demonstrate a progressive increase in their emission ratio
when compared to the overall ratio, while some demonstrate a reduced ratio of emission to the whole
emission intensity. Ultimately, the area under the peak readings tend to move toward stabilisation, and
is this very much in agreement with mentioned observation associated with the chromaticity data.

Figure 4.11. Plot demonstrating the ratio of peak in (a) 400-430nm (b) 431-449nm (c) 450-500nm (d) 500-
600nm (e) 600-700nm (f) 700-800nm wavelength range to the total spectral area under the peak during
the first 5000 s from the ignition of the plasma. The time series is divided into five sections of 1000 s.

Figure 4.12 presents the effect of the RF power applied to the magnetron on the emission peak areas
of the spectral segments, where it is clearly noticeable that an increase in the emission ratio is observed
in sections a, b, and c, while d, e, and f demonstrate a reduction. There seems to be a linear trend in both
cases. This is interesting and it matches the observation of the chromaticity indices, which also
demonstrated a linear relationship with plasma generating power.

81
 Figure 4.12. Plot demonstrating the ratio (to the total spectral cover from 300 to 800 nm) of the area under
the peak in each of the segments with RF plasma power for the IZO deposition.

Figure 4.13 presents the effect of the chamber pressure on the regions of the emission spectrum and
the ratio of the peak areas to the overall area covered. Just like the previous cases, it can be seen that the
different sections of the spectrum demonstrate different trends in terms of increase or decrease of the
peak areas. Once again, each section is demonstrating either an increasing or decreasing ratio of emission
areas to the overall spectral emission area under the peaks. Although the UV emission region (300–400
nm) and the f section of the spectrum (700 to 800 nm) are not used for calculating the chromaticity
indicates, the data were presented here to demonstrate the spectral property of the plasma at those
regions. The results for the UV section of the spectrum are separately presented in figure 4.14. Where the
UV emissions peak area demonstrate an increasing trend over the 5000s period and demonstrate a
linearly increasing relationship with the plasma power, the higher chamber pressure seems to reduce the
ratio of the UV emissions to the overall emission within the spectrum.

Figure 4.13. These data demonstrate the ratio (to the total spectral cover from 300 to 800 nm) of the area
under the peak in each of the segments in response to varying the chamber pressure for fabrication of IZO.

82
Figure 4.14. (a): Plot stability test over 5000 seconds period, (b): Effect of plasma power on UV emission ratio,
(c): Effect of pressure on UV emission ratio.

Figure 4.14 demonstrates the ratio (to the total spectral cover from 300 to 800 nm) of the area under
the peak in the UV section (300–400 nm) of the spectrum. The UV emissions demonstrate an
increasing trend over the 5000 s period, followed by a slight decline and they demonstrate a linearly
increasing relationship with the plasma power. The higher chamber pressure seems to be reducing
the ratio of the UV emissions to the overall emission within the spectrum.

4.2 Discussion
This study demonstrated how the complex analysis of the emission spectrum can be summarised by
applying the colour coordinates of the plasma light. The calculation of the colour functions is extracted
through the area that is covered under the emission peaks of the plasma light that are affected by various
parameters. Higher plasma power simply provides more energy toward the ionization of the working gas
atoms. A higher density of ionized atoms, in turn, give a higher flux of emission and, as such, increase the
intensity of the spectra. By increasing the power of the plasma, the number and momentum of the argon
ions and sputtering particles are both increased, which results in higher bombardment rate of the target
surface and higher mobility of the atoms at the deposited thin film surface. The higher mobility of the
atoms will reduce the stress at the thin film surface, and hence facilitates further crystallisation [166].

The discharge current linearly increases with increasing the plasma power, irrespective of the
chamber pressure. The energy of the sputtered atoms arriving at the surface of the substrate increases
with increasing the plasma power. Parallel to this, the effect of high energy bombardment of the substrate
surface is also increased, providing thermal energy at the surface atoms to become more mobile. Overall,
the higher plasma power should lead to more crystalline deposition of thin films. However, excessive
power can have an adverse effect by causing degradation of the thin film surface, resulting in high defect

83
density [167]. Based on the colour functions of the plasma, a linear relationship with the plasma power
has been demonstrated, hence giving this technique the potential of defining a relationship between the
film properties and the x and y colour coordinates, which will be potentially be explored this further in
future work.

If all of the peaks would increase at the same rate with increasing the power, the chromaticity
parameters should not change, but in figure 4.3 There is a linear trend in how the x and y values are
plotted against each other. This indicates that as the plasma power is increased and different energy
bands tend to respond to the higher magnitude of the power. This is currently under investigation and
the reason that section 400 to 500 nm section of the spectrum divided into three sections to progress
one’s understanding of this observation; however, as yet, the findings are not conclusive and hence no
further explanation or claims can be made at present. However, from an applied scientist or an operator
of a plasma deposition system, the relationship observed in figure 4.3 is a simple observational method if
in depth an operator does not require theoretical knowledge. The broadening of the spectral lines, as
discussed in the introduction, are ultimately related to the electron density via the stark broadening
process.

The colour functions are calculated from the area under the emission peaks, hence this broadening
is taken into consideration during the colour coordinate measurements and present a simplified form of
data for the operator, who may be only interested in basic monitoring protocols and not necessarily wish
to indulge in to the complex statistical mechanics of the plasma. A similar argument can be applied for
the emission line shapes which are strongly influenced by the interaction of the radiating atoms or ions
with surrounding particles. All of the spectral data that were obtained during these experiments were
carried out using the same distance between the collimator head and the plasma. It is important to
acknowledge that this distance is highly important and it can influence the spectra of the plasma and it is
part of the ongoing research objectives to investigate various distances of the collimator and the plasma.

Through the data presented in figure 4.2, it has also been demonstrated that, when trying to identify
the stability of the plasma through the line ratio method, there may be confusion as certain peaks may
increase or decrease, irrespective of their initial magnitude and ratio, while the proposed colour index
values can provide a simple route to follow. Figure 4.10 presents some of the possible transitions that
need assessment when using the conventional line intensity techniques, while, through colour coordinate
calculations, the overall area of the peaks is considered, hence simplifying the monitoring process for a
user with limited need of extensive plasma physics knowledge. However, the colour coordinates do not
take into consideration the emissions in the UV and far-red part of the spectrum. Hence, an alternative
colour function for this application can be envisaged. If this method is matured in the future, then it can
be beneficial for applied scientists and those who utilise plasma sputtering for thin film coatings and
depositions. The colour of the plasma is derived by the quantity and state of various meta-stable atoms
and ions that constitute it. In an analogical overview, a simplified shortcut can be established to
characterise the plasma by being able to identify the relationship between the colour of the plasma and
physical events within the plasma, just as the fingerprint pattern is used to characterise people rather
than their individual genome sequence.

Higher chamber pressure lowers the voltage at which the plasma can be ignited and maintained. This
is because higher chamber pressure increases the probability of the ionisation process of argon. Higher
chamber pressure on the other hand means that more argon atoms are available to be energised to
bombard the target surface; however, a higher concentration of argon atoms means that there will be
more collision between the sputtered atoms or ions and argon atoms. These collisions can lead to lesser

84
energetic atoms arriving at the surface of the substrate [168]. A correlation can clearly be seen between
the x and y coordinates of the chromaticity index of the plasma light and certain parameters, such as
plasma stability, the working gas and the associated chamber pressure and the plasma power that affect
the x and y values. This highlights a clear potential of pursuing and investigating the proposed concept
and, if successful for a given target material to be sputtered, regardless of the manufacturer of the
sputtering deposition machine, matching the colour parameters can indicate exact similar plasma
operating conditions. Meanwhile, theoretically, there is a possibility of operating a plasma under two
different conditions and getting similar colour values, which has been partially investigated, where, by
altering the chamber pressure and plasma power, similar x or y values can be mimicked, but having an
identical x and y values at the same time so far has not been observed, this itself will be a future
experimental protocol.
The x and y coordinates can give an excellent, easy indication of the point at which the plasma is
stable when compared to more complex calculation methods. The stabilisation process itself is an
interesting concept to consider, slight variation in parameters that are required for driving plasma can
create unstable conditions, and the colour factor can be an excellent indicator of this.

4.3 Conclusion
This work demonstrated that, under unique chamber pressure and plasma power, there can be a
distinct x and y value that can be utilised by the operators of the machine to create a library of the thin
film property relevant to particular x and y coordinate values, making the process of monitoring the thin
film quality significantly easy. However, what has been reported here is just the beginning of a concept
that bears potential. The colour coordinates that are driven from the plasma here do not take into
consideration the UV and IR emissions. The work presented in this chapter has also been peer-reviewed
and published [169].
Additionally, in future efforts, the relationship between the coating properties and the plasma
photometric need to be established. Future efforts on further investigation of this concept require
establishing the relationship between the chromaticity parameters and the plasma’s inner parameters,
such as electron temperature and charge density, which enables formulating a reliable relationship
between the mentioned factors. Thereby, the first step in any future development of this concept is to
prepare and create an alternative colour coding that will take the UV and IR emissions into account and
will create a colour index beyond one’s vision.

85
Chapter 5 Highly Conductive Zinc Oxide Films Prepared Using RF
Plasma Sputtering Under Reducing Atmosphere

The previous chapter presented relationships between the plasma operating parameters such as
plasma power and chamber pressure can be established with the colour functions of the plasma was
demonstrated. The colour functions are driven from the area under the peaks of the emission spectrum
of any light source, and by treating the plasma as a light source A proposal was put forward that it can be
evaluated based on its colour. A significant observation has been demonstrated regarding the effect of
hydrogen on shifting of the colour functions and how these functions behaved once the sputtering target
was exposed to hydrogen. This study takes a closer look at this methodology with an emphasis on detailed
monitoring of the sputtering plasma during IZO preparation with argon (Ar) and argon/hydrogen (5%) gas
mix (Ar + H) as the working gas. An investigation on the effect of consecutive exposure of pure ZnO target
to (Ar + H) on the electrical properties of ZnO films without any target material doping as a precursor was
carried out. The sample preparation and experimental methods have been outlined in 3.1.2.

5.1 Results and Discussion

Evaluation of the EDS system accuracy and best practice for the chemical analyses of the films was
performed using InAs of MAC reference sample #3601. The analyses of the standards were performed at
20 and 10 kV to evaluate the differences in the quantitative analyses (table 4.1) and select the optimal
conditions. Only atomic percent was under scrutiny because the analyses aimed to identify the ratio of
In/Zn in the IZO films following the different deposition conditions. Columns highlighted in grey are the
columns indicating the ratio between the In/Zn cations in the films and In/As cation in the reference
material used for the evaluation of the analytical best practice.

It can be observed in table 5.1 that the system used tends to overestimate the lighter elements both
at 20 and 10 kV. When the carbon build-up is considered in the quantification, the atomic % calculation
seems to show better agreement with the certified values. The correlation of the quantification at 10 and
20 kV seems to be consistent, however, further overestimation of the lighter elements can be observed
at 10 kV.

If this is applied to the films it can be inferred that the measurement at 20 kV is closer to the
composition of the films, however, potential underestimation of indium could be applied. In the reference
material at 20 kV, a wider scatter of results can be seen, however, the results are constantly closer to the
certified results (ratio 1, atomic % 50/50) while the results are more constant at 10 kV, however they are
further from the certified values overall.

Theoretically, the results at 10 kV might be more representative of the ratio of In/Zn in the films since the
volume of the interaction of the beam is more limited to the surface of the samples, hence, the final
spectrum includes only a very limited amount of the elements present in the glass slide which forms the
substrate of the film. However, the indication obtained by the presence of carbon build-up in the analysis
indicates that an overall evaluation of all elements present in the volume of interaction might be closer
to the truthful composition of the material, hence, the data obtained at 20 kV will be used as a
characteristic of the films under analyses.

86
Table 5.1. Energy-dispersive X-ray spectroscopy (EDS) investigation of the films (S1 to S7) at 10 and 20 kV

5.1.1 Film Morphology, Composition and Crystallinity Characteristics

Secondary electron images of samples S1 to S7 showed a high homogeneity of the film surface
(figure 5.1 A & B). Some samples show oriented linear marks (samples S6 and S7). All samples (using
both Ar + H gas and Ar only) show areas with flower-like structures, identifiable as formation islands
(figure 5.2 C & D) with dimensions varying between 1 and 10 microns.

87
Figure 5.1. (A) The smooth homogenous surface of S6; (B) homogenous distribution of the film in sample S7 (C)
secondary electron image of sample S2 (D) secondary electron image of sample S5

The compositions of the thin films obtained at 20 kV (as discussed above, the validity of the results
obtained at different kV was assessed using a reference material to identify which one is most
representative, and 20 kV was deemed to be so) show an indium/zinc ratio of 2.3 to 1 to 3.1 to 1, different
(higher) from the In2O3/ZnO composition obtained by Sheng et al. However, a closer investigation of the
results indicates that the presence of the indium ratio in samples deposited with hydrogen is higher than
in those deposited with Ar gas alone and sample S5, which represented the switch from Ar + H to Ar gas,
shows the lowest In/Zn ratio. Investigation of the crystallinity of the thin films, both using X-ray and
electron backscatter diffraction showed that the IZO thin film is amorphous, in agreement with the above-
mentioned work of Sheng et al. [170].

Pure ZnO samples (H0, H3 and H6) show an even more distinct granularity (figure 5.2 A–C) with
grains of approximately 100 nm size and heterogenous shape; however, sample H0 shows some defects
and a larger less compact grain distribution (figure 5.2B) and no “flower-like” structure is visible in any
of the ZnO films.

88
Figure 5.2 (A). In lens image of H6; (B) in lens image of H3; (C) in lens image of H0 showing larger less homogenous
grains and fast carbon build-up during the SEM investigation.

Investigation of the composition of the ZnO films using EDS analyses shows a low concentration of
nickel and copper, aside from the zinc and oxygen and from the elements deriving from the substrate
glass (Ca, K, Si, Al, Mg).

The nickel is constantly detected at 0.2 wt % in H0 and H3, while iron varies from not detected to 0.2
wt % in these two samples. In H6, nickel is still detected from 0.2 to 0.3 wt %, while iron varies from not
detected in the majority of the areas analysed to below 0.2 wt %. Overall, the composition does not show
dramatic differences between the three samples, while morphologically, H0 shows larger grains and a less
uniform deposition compared to the other two samples, and the distribution of the grains improves also
from H3 to H6, which shows the most uniform aspect (figure 5.2A–C).

The ZnO films are fully crystalline, showing the hexagonal ZnO zincite phase (PDF 01-075-1533) with
a (101) preferential orientation.

5.1.2 Electrical Properties

The IZO and ZnO coated substrates were tested for their electrical properties via a four-point probe
and Hall effect measurement equipment. The Hall effect samples were 10 mm by 10 mm with contact
pads connecting at ~1 mm into the sample from the corners. Figure 5.3 demonstrates the average sheet
resistance of the four substrates belonging to the S1–S7 (IZO) and H0, H3, H6 (ZnO) samples along with
the standard deviation of the data. The conductivity of the samples S1–S4 clearly shows improvement
with each consecutive coating under the Ar + H gas. Sample S5 was prepared by switching the gas from
Ar + H to Ar gas from the beginning of the process and a very poor electrical conductivity in this sample
was observed, while the consistency of the results among the four substrates in the S5 group is
significantly poorer than in the other. As for the samples S6 and S7, which were prepared under Ar gas
following the processing of S5, there is an improvement in conductivity, but this does not reach the
performance of the samples prepared under Ar + H gas (S1–S4). It can be seen that each consecutive
deposition process under Ar + H gas gradually improves the sheet resistance of the deposited thin films
from S1 to S4. Sample S5 is the sample that was deposited by reverting to Ar gas only. Sample S5 has very
poor conductivity. From the standard deviation of the obtained sheet resistance data, it can be seen that
during the transition period (S5) even the samples from the same batch demonstrate poor continuity of
the property (Figure 3a). The sheet resistance of the three ZnO samples demonstrates the improvement
in sheet resistance of the films due to prolonged exposure to Ar + H gas (Figure 5.3b).

89
Figure 5.3. (a) The average sheet resistance of the S series samples and (b) H series samples

The ZnO only samples (H0, H3, H6) also clearly demonstrate the effect of consecutive exposure of
the target material to the Ar + H gas. These samples, as discussed, were all prepared with Ar + H gas.
Sample H0 was prepared with a fresh target, while H3 and H6 were prepared when the target was exposed
to 3 and 6 h of Ar + H exposure, respectively. It can be seen that the electrical conductivity of the samples
prepared improves concerning the prior exposure of the gas to Ar + H gas. The H6 sample, which was
prepared with longer period of prior exposure to Ar + H gas, was superior to H0 and H3 in terms of
electrical properties.

The resistivity of the samples and the thickness of the films are demonstrated in figure 5.4. Apart
from the sample S1, the rest of the films demonstrate a similar trend in terms of film thickness; however,
the resistivity of the samples highlights what has already been observed and discussed in figure 5.3. (a)
The resistivity of the films obtained using the Hall-effect measurement technique demonstrates that each
consecutive deposition process under Ar + H gas gradually improves the electrical conductivity of the films,
while the S5 sample, deposited during the transition from Ar + H to Ar gas, shows poor results. (b) The

90
thickness of the films does not show a significant difference, however, the first sample deposition, where
the target was first exposed to hydrogen, demonstrates a lower yield. (c) Prolonged exposure to Ar + H
gas in improving the resistivity of ZnO based samples. (d) All ZnO samples demonstrate similar film
thickness.

Figure 5.4. (a) Resistivity of S series samples (b) Film thickness of S series samples (c) Resistivity of H series samples
(d) Film thickness of H series samples.

In figure 5.5, One can see that the S1–S4 samples have similar mobility values, while the S5 is the poorest,
and S6 and S7 are close to the samples coated under Ar + H gas. However, the carrier concentration data
suggest that consecutive coating under Ar + H gas further improves the carrier concentration. The carrier
concentration in the S4 sample is 5 times higher than the S6 and S7 samples; the S5 transition sample
demonstrates the poorest results. The standard deviation values of the average sheet resistance give an
indication of the error associated with electrical properties, considering the error associated with mobility
and carrier concentration; the ‘S’ series samples show standard deviation values in the range of (0.33–
0.48) for mobility values, with the exception of S5 sample having a standard deviation of 2.8 from the four
samples prepared for each series. The standard deviation of carrier concentration in these samples was
in the range of 3.9 × 1018 for S1–S5 and ~3.2 × 1017 for S6 and S7. The H series samples show standard
deviation values of (0.25–0.33) for mobility and (3.1–3.9 × 1018) for the carrier concentration error in these
samples.

91
Figure 5.5. (a) Carrier Concentration of S series samples. (b) Mobility of S series samples. (c) Carrier Concentration of
H series samples and (d) Mobility of H series samples

Charge carrier mobility and the carrier concentration are important in determining the electrical
conductivity of a material. The mobility of the carriers is a function of carrier recombination time within
the material, while the carrier concentration is related to the carrier density. The carrier density in turn is
a product of density of states and probability of occupancy.

Both IZO and ZnO samples demonstrate electrical conductivity improvement with consecutive
exposure of the target material to Ar + H gas. However, it can be seen that in the case of the IZO sample,
the improvement can be traced to enhancement of the carrier concentration values as the mobility values
are not significantly affected. This indicates that Ar + H gas processing does not impact the carrier
recombination time significantly; however, the increase in carrier concentration indicates the density of
states at band level or probability of occupancy of these levels is altered.

In contrast, ZnO samples demonstrate an increase in both mobility and carrier concentration values.
It is important to mention that there is a trade-off between increasing carrier concentration and carrier
mobility; as carrier concentration increases the probability of charge carrier scattering increases, and as
such, overall conductivity would be affected, hence, as is indicated from the results shown, there is
potentially room for further improvement on the ZnO sample until this trade-off is observed. This
observation can be further explored to determine the maximum achievable conductivity.

When electrons are traveling through a periodic solid, their movement will be affected by the local
forces within the crystal. Thus, a term is used to define the mass of the electrons under that environment,
which is referred to as the effective mass. The effective mass, in turn, will be influencing the mobility of
the electron (charge carrier). The orbital overlap between the metal cation and the oxygen in the host
lattice (e.g., ZnO) is reported to be an important factor in determining the electron mass because metal

92
oxides with predominant s-character of the cation at the CBM are correlated with lowest effective
electron masses [171]. The results indicate that the presence of hydrogen in the sputtering gas enhances
the mobility of the electrons. This is possibly achieved by hydrogen influencing the oxygen content of the
lattice and enhancing the s-characteristic of the indium and zinc. The hydrogen radicals generated in
plasma are a strong reducing species [172], and they can withdraw oxygen from the crystal during the
deposition. Oxygen vacancy has been believed to play the role of doubly charged donor in In2O3 [173,174].

On the other hand, hydrogen atoms can be incorporated in the solid film when they occupy the
interstitial site of the crystal. The residing hydrogen forms a hydroxyl bond which may play the role of
scattering centre for charge carriers [175] A separate study on ITO (Indium Tin Oxide) has demonstrated
that due to the withdrawal of oxygen during the deposition of ITO films, the hydrogen in plasma can be
so effective for achieving lower resistivity that the films were prepared without any sintering [176]. As
discussed, the samples were prepared without additional sintering, and achieving highly conductive films
without the thermal energy requirement of sintering offers significant commercial benefits.

The hydrogen impurity can be either interstitial or substitutional. Theoretical studies reported by van
de Walle have demonstrated that hydrogen present in ZnO lattice can exclusively behave as an electron
donor [177], and this theoretical study is verified by electron nuclear double resonance and Muon spin
spectroscopy [178,179].

Considering interstitial locations, computing for formation energies associated with interstitial
hydrogen in ZnO indicates that hydrogen is in H+ state and the H0 and H− are not feasible in the ZnO
system [179]. The H+ in the ZnO system will occupy locations within the lattice where it can bind to an
oxygen atom and form an O–H bond. This in turn will lead to lattice relaxations resulting in zinc and oxygen
experiencing a slight move from their location. Regarding substitutional hydrogen, it is located in the
proximity of a nominal oxygen position within the lattice and behaves as a shallow donor [180]. The donor
behaviour of the hydrogen impurity in ZnO clearly explains the conductivity observed in ZnO systems and
improvement of IZO conductivity.

The electrical conductivity improvements in IZO and ZnO thin films have been observed with the
introduction of the hydrogen into the sputtering process are highly important considering the cost of
indium metal and its economic impact at industrial scale TCO production. This means that at industrial
scale levels, by applying hydrogen in the process, thinner films of IZO can be produced while the desired
electrical conductivity is maintained, and in the case of ZnO, further improvements would eliminate the
cost of indium. Also, hydrogen doped ZnO films have been developed at room temperature bears some
exclusive processing advantages viz. polymer substrates, unlike aluminium doped ZnO films of desired
conductivity which are generally optimised at around 300 oC.

For a TCO material to be suitable for most electrode applications, the electrical resistivity (ρ) should
be low (10−4 to 10−5 ohm·cm), the charge carrier mobility (µ) maximized to ~60 cm2 V−1 s−1 and the carrier
concentrations be kept at about 2 × 1021 cm−3 to minimize optical absorption [181]. The charge mobility
presented in the results does not match these expectations yet, however, the improvement with using
hydrogen could be exploited further by altering the amount of hydrogen, plasma power and chamber
pressure, which will be pursued in future work.

The addition of the hydrogen during the sputtering procedure has been reported to cause a drop in
the ionization discharge and Ar ion densities [182-185]. Some have also reported yield reduction with

93
introducing hydrogen [186,187] due to the low mass of hydrogen, but this was not observed in this study;
this can be explained by the formation of ArH + ions. Within the plasma, ArH + ions are formed which
possess high kinetic energy during target bombardment compared to Ar ions [187]. The film thickness
analysis shows that the presence of hydrogen in the plasma has not affected the yield, and the
bombardment of the target surface by the ArH + heavy ions could be an explanation for this.

5.1.3 UV-Vis Spectroscopy

The optical transparency of the films was examined using UV-Vis spectroscopy, and the results are
presented in figure 5.6. As the four samples within each deposition procedure (S1, S2, H6) demonstrated
a similar result, only ‘one’ sample from each set was used for these measurements (the results do not
represent average values). Samples coated under Ar + H demonstrate very similar results with a clear
indication of a wider bandgap allowing more light in the 320–430 nm region, and the samples coated
under Ar are also very similar within their group. The S5 sample, which so far has been demonstrating
poor performance in electrical properties, is no different than the S6 and S7 samples, which were also
coated under Ar gas. The ZnO based samples (H series) demonstrate a wider transmission spectrum
covering the shorter wavelengths, however, their band gap value is smaller than the IZO samples.

94
Figure 5.6. The UV-Vis spectral analysis of the samples in % transmittance is presented for; (a) IZO samples; (b) ZnO
samples.

Through the absorption data obtained, by applying a Tauc plot, the optical bandgap of the films was
calculated, and the data are presented in figure 5.7. The samples S1–S4, which were prepared under Ar +
H, demonstrate a wider optical band gap, while the samples S5–S7 demonstrated a smaller optical
bandgap. Here, the S5 sample again demonstrates smaller optical wavelength. The H series samples (ZnO)
demonstrate an increase in optical band gap values as a result of prolonged exposure of the target to ArH
gas.

95
Figure 5.7. The optical band gap of; (a) the deposited IZO films. (b) ZnO samples.

5.1.4 Spectral and Colour Function Analysis of the Plasma

The emission characteristics in terms of the whole spectral plot and the chromaticity index were
obtained for each test group over two hours. Figure 5.8 represents the x and y coordinate values of the
plasma colour during the transition from Ar gas to Ar + H and vice versa. As previously reported, the
shifting of the x and y coordinates during Ar to Ar + H gas are swift compared to the data obtained when
switching from Ar + H to Ar, as was the case during the preparation of the S5 sample. One can see that it
takes more than one hour for the x and y coordinates to reach their initial value. This indicates that the
plasma during this transition period is not stable and is changing.

The colour function approach, although still immature, bears the potential of becoming an ideal
industrial-scale real-time monitoring system of the process control due to its simplicity of understanding
for the operators. Although the colour function method of monitoring the plasma as a diagnostic tool is
demonstrating useful applications, from the presented results significant variations in the 300–400 nm of

96
the spectrum obtained from the plasma was observed, which is not accounted for in the constructing of
the x and y chromaticity coordinates.

Figure 5.8. (A) The chromaticity colour functions, x and y, are plotted during the transition from Ar to ArH as the
deposition of the S1 samples is undertaken. (B) The switch to the Ar gas from the ArH during the S5 sample deposition

This indicates that the introduction of hydrogen not only affects the plasma and the specifics of the
deposited IZO and ZnO thin film, rather one can conclude that hydrogen is altering the target material
itself and possibly forms surface compositions on the target or adsorption at the surface of the target that
tend to affect the coating procedure. This can only be confirmed by surface analysis of the target post
deposition, which will be investigated by the authors in the future.

Given the specific characteristics of the S5 sample and the shifting of the colour functions during the
transition from Ar + H to Ar, the focus shifted to presenting and examining the plasma emissions during
the S5 deposition process. The spectral data of the S5 sample at every 100 nm interval from 300 to 900
nm at various stages (from 1 s after the deposition process to 100, 1000, 3000 and 6000 s) of the process
of the samples were examined. For simplicity, the 300 to 400 nm segment of the emission spectra is
presented in figure 5.9 as an example, and the rest of the spectral segments are discussed. From the
spectral data it is observed that some emission peaks demonstrate an increase, decrease or no change
through the deposition process.

97
Figure 5.9. The detailed emission spectra of the plasma between 300 and 400 nm during the deposition of the S5
sample are presented.

The NIST database (National Institute of Standards and Technology, Gaithersburg, MD, USA) can be
used to identify and refer to some of these peaks. Figure 5.9 represents the 300–400 nm segment of the
emission spectra of the plasma. Here, for example, One can see that the emission at 303.94 nm belonging
to oxygen (II) in association with the transition 2s22p2(3P)3d 4D 5/2 to 2s22p2(3P)5f D 2[3]° 7/2 is
gradually increasing along with the emission at 326.09 nm belonging to oxygen (III) in association with the
transition 2s22p(2P°)3p 3D 2 to 2s22p(2P°)3d 3F° 3. In figure 5.9, a decrease in the 357.07 nm emission
peak belonging to argon (II) in association with the transition 3s23p4 (1D) 4p 2D° 3/2 to 3s23p4(3P) 5d 4P
3/2 was also observed. There is also an increase in the 360.58 nm belonging to argon (II) in association
with the 3s23p4(3P)3d 2P 3/2 to 3s23p4(1D)4p 2D° 5/2 transition.

In the 400–500 nm segment of the emission spectra of the plasma, there is an increase at 410.2 nm
which is typically associated with an oxygen (II) transition with the2s22p2(3P)3p 4P° 1/2 to 2s22p2(3P)3d
4D 1/2 transition. There is also another peak at 451.07 nm, which is associated with Ar (I)
3s23p5(2P°1/2)4s 2[1/2]° 1 to 3s23p5(2P°3/2)5p 2[1/2] 0 transition, which also demonstrates a gradual
increase.

In the 500–600 nm range, a decrease at 560.6 nm belonging to argon (I) in association with the
3s23p5(2P°3/2)4p 2[1/2] 1 to 3s23p5(2P°3/2)5d 2[1/2]° 1 transition was found. There are many other
peaks which can also be seen increasing during the preparation of the S5 sample in this region.

In the 600–900 nm range, results indicate that variations in peak intensity or the area covered under the
spectrum are not significantly affected.

There have been some reports on the emission line intensity variations when hydrogen is introduced
into an argon discharge along with ionization variations with hydrogen introduction [188,189]. However,
all these studies require time-consuming efforts and data analysis at micro levels in terms of interpreting
the data, which may not be easy to apply during the deposition process.

98
As hydrogen is added to the Ar magnetron discharge, the plasma parameters in the discharge begin to
change. Studies on plasma behaviour during the preparation of TiO2 films have shown that the addition
of hydrogen to oxygen-containing argon discharge leads to a continuous decrease of electron density and
a corresponding increase of electron temperature from 6.30 to 6.74 eV [190]. The increase in electron
temperature can be accounted for by the reduction of the value of electron density in the discharge.

For a good approximation in low-pressure discharge, the emission intensity of a particular line of an
element is considered to be proportional to the density of that particular species [191,192]. Comparing
the line intensity ratio of selected transitions, one can estimate the degree of dissociation and ionization
of particular species present in the discharge. The line intensity ratios of IH/IAr can be used to determine
the degree of dissociation in hydrogen plasma using the following relation [193]:
Q
𝑁Y = 𝑘 Q + (4.1)
(,

where NH and NAr are the degrees of dissociation of hydrogen and argon, and k is a parameter related to
the rate coefficients for direct excitation of the corresponding Ar and H atoms and the radiation
transition probabilities and lifetimes of the corresponding excited states respectively.

As discussed earlier, an increase in the intensities associated with certain argon and oxygen
transitions during the Ar + H gas switch to Ar only can be seen. The detailed study of the plasma is beyond
the scope of this research; however, from these results, it can be seen that that once the target material
is exposed to the Ar + H gas, the process of reverting the plasma condition to a prior state is time
dependent. The sample S5, which was prepared during such a process, is the result of a continuously
changing plasma condition, and as such, the properties of the film are directly affected. The colour
function approach, as demonstrated in figure 4.8, also verifies this. Hence, the colour function approach
can be considered a simpler approach toward monitoring the status of the plasma when significant details
are not required.

5.2 Conclusion
It has been demonstrated that the deposition of IZO and ZnO film under a reducing atmosphere can
enhance the electrical property of these thin films by enhancing the carrier concentration of IZO films and
increasing the carrier concentration and mobility in ZnO films. The key findings of the work carried out
are as follows:

1. Most importantly, there is clear evidence that consecutive exposure of the target material to
Ar+H gas will to some extent add further improvements to the electrical properties. A target
that has been previously exposed to Ar+H gas seems to be altered at the surface, as reverting
to Ar gas only for deposition will initially produce a thin film with inconsistent electrical
properties.
2. The improvements obtained with introducing hydrogen during the sputter deposition of IZO
can minimise the consumption of indium, as thinner films with identical electrical properties
can be deposited. In the case of further improvements with the electrical properties of pure
ZnO thin films, introduction of hydrogen can potentially lead to the elimination of indium. As
such, there can be significant commercial benefits in depositing the ZnO based family of TCOs
under Ar+H gas mix, if the need for indium and high-temperature sintering is avoided;
however, further improvement and assessment are required.

99
 3. One can visualize the status of the plasma stability via the application of the colour function
analysis. During the sputter deposition of IZO under argon, when the target has been
previously exposed to hydrogen, An unstable plasma yields an IZO film with very poor
performance, resulting also in the lowest concentration of indium. The fact that the
chromaticity index values (x and y) do not swiftly fall back to the values expected for argon
indicates that for the surface modifications caused by hydrogen to revert to the original state,
a layer from the target has to be removed, leading to a film with poor charge carrier and
density properties.

The colour function evaluation of the plasma can be a beneficial observational tool for monitoring
the plasma during system operations. However, further work is required to mature this method of plasma
monitoring and assessment. Hence, the next task will be to investigate the plasma emissions and, if
necessary, formulate a new colour function that takes into account the emissions in the 300–400 nm
region. This will lead to constructing a new colour coordinate system that will specifically define the colour
of plasma, similar to how the colours in the visible spectrum are defined by the x and y index values. The
research work and findings presented in this chapter have also been published as a journal publication
[194].

100
Chapter 6 Investigation on substrate position relative to Plasma
during the sputter deposition of Hydrogen Doped Zinc Oxide

Following on from the previous chapter where the enhancement of the electrical properties of IZO
and ZnO systems by repeated exposure of the sputtering target to an argon/hydrogen mixture as the
working gas was investigated. In this set of experiments as detailed in chapter 3.1.3, interesting
complexities associated with reactive hydrogen/argon sputtering of un-doped ZnO targets have been
identifited, which is believed to be associated with the reported magnetic properties of hydrogenated
ZnO particles, and this report outlines these observations.

As described in chapter 3, the recently reported on implementing a novel approach through which
the sputtering plasma can be given a unique fingerprint based on the colour of the plasma, which is
identical to the method by which colour functions of a light source are expressed in a “x” and “y”
coordinate system to define the unique colour of the light source. This technique was applied for
monitoring the plasma during these experiments and attempted to explain the effects observed by
evaluating the colour parameters of the plasma in association with the electrical properties of the ZnO
thin films deposited via reactive hydrogen/argon sputtering.

6.1 Results and Discussion

6.1.1 Electrical Properties
The ZnO coated substrates were tested for their electrical properties via a four-point probe and Hall
effect measurement equipment. Although no study was carried out on the crystal structure of the
deposited thin films; in the previous report/chapter, where samples of zinc oxide thin films deposited in
the presence of hydrogen were examined, the ZnO films were fully crystalline, showing the hexagonal ZnO
zincite phase (PDF 01-075-1533) with a (101) preferential orientation.

The results indicated that the presence of hydrogen in the sputtering gas induces n-type conductivity
by enhancing the concentration and mobility of the charge carriers (electrons). However, It can be seen
that there is significant variation when comparing results between the samples positioned at the edge
and centre of the substrate holder, as indicated in figure 6.1.

101
Figure 6.1. Carrier concentration and mobility were measured for ZnO samples deposited in the centre (left) and the
edge (right) of the substrate holder under various plasma power and chamber pressures.

Charge carrier mobility and the carrier concentration are important in determining the electrical
conductivity of a material, and it can be seen that high carrier concentration and mobility values are
associated with the samples at the edge of substrate holder, while depositions at the centre demonstrate
smaller values, and as such the overall conductivity of the edge samples are much higher. The mobility of
the carriers is a function of carrier recombination time within the material while the carrier concentration
is related to the carrier density. The carrier density in turn is the product of density of states and
probability of occupancy. As previously discussed, the hydrogen impurity behaves as a donor in these

102
samples. Associating higher conductivity to hydrogen integration into the lattice based on the discussed
literature, the results can possibly indicate lower hydrogen content in the films prepared at the centre of
the substrate holder, although this hydrogen content needs to be measured in future work to be perfectly
conclusive. As discussed, substrates were placed between the edge and centre regions. These samples
were tested via a four-point probe measurement and a conductivity gradient was found when moving
from the edge to the centre of the sample holder, the electrical property of the thin films became smaller.
Based on the mentioned ongoing studies comparing IZO and ZnO, this is only observed in ZnO samples.
To demonstrate a comparison for this report, the experiment with an IZO target fitted on the magnetron
was repeated, and only the sheet resistance of the films was measured for presentation in this report.
These results, presented in 6.2, indicated that IZO does not behave like ZnO.

Figure 6.2. Plot of Sheet resistance vs distance away from centre for IZO and ZnO samples

As can be seen from the results presented in figure 6.2, the electrical properties of the deposited
films significantly varied for ZnO, depending on the position for the substrate, while for IZO the variation
was significantly less elaborated (comparing the associated edge and centre samples). Results presented
in chapter 5 demonstrated the enhanced electrical properties of the IZO when deposited under the argon–
hydrogen mix. These results combined with the observations presented in 6.2 suggest that one way to
explain this is to assume that more hydrogen doping is occurring at the edge of the substrate compared
to the centre, unless the thickness of the coating was causing this observation. However, that thickness is
not a relevant explanation for this observation as justified in the following section.

The mobility of electrons through a solid is affected by local forces within the crystal structure. This
mobility is then interpreted as the mass of the electrons within that crystal environment. In a ZnO crystal
structure, the orbital overlap between the Zn cation and the oxygen anions can be an important factor in

103
determining the above-mentioned electron mass due to the s-character of the cation at the conduction
band [195]. It can be hypothesized that when hydrogen is present as a sputtering gas, it may interact with
the oxygen content of the lattice and result in localised lower oxygen contents, which will consecutively
lead to predominant s-characteristics. This hypothesis is based on previously reported studies
demonstrating the hydrogen to be a strong reducing gas in a plasma [196], while withdrawing oxygen
from the In2O3 crystals during the deposition process has been reported to play a role of a doubly charged
donor [197,198]. However, this cannot clearly explain the variation observed in edge and centre samples.
Sheet resistance of edge and centre prepared ZnO samples are presented in figure 6.3.

Figure 6.3. (A)The sheet resistance (Ω square) of the ZnO samples deposited on the Centre and (B) Edge part of the
substrate holder under various deposition conditions.

104
 6.1.2 Film Thickness and Optical Properties
The thickness of the coatings obtained via the profilometer in line with the sample preparation
described in chapter 3 are presented in figure 6.4 for all the ZnO samples.

Figure 6.4. The coating thickness associated with the position of the substrate on the substrate holder (at a distance
from the centre) under the varying deposition parameters, power and gas flow rate.

From figure 6.4 One can see that the films are actually slightly thinner with the position of samples
closer to the edge of the substrate holder.

The addition of the hydrogen during the sputtering procedure has been reported to cause a drop in
the ionization discharge and Ar+ ion densities [199-201]. Some have also reported yield reduction with the
introduction of hydrogen [202-204] due to the low mass of hydrogen, and this may explain why there was
a slightly thinner coating on the edge region if hydrogen bombardment of the target and its consecutive
incorporation into the lattice behaved selectively, depending on the position of the sample.

Hence, the original assumption can be considered where it was stated that more hydrogen atoms
are doped into ZnO lattice at the edge of the substrate holder and oxygen vacancy alone is not a
determining factor.

For a TCO material to be suitable for most electrode applications, as well as the electrical resistivity
(ρ), the optical absorption should be minimised [205]. The overall sheet resistance of the ZnO samples
based on their position on the substrate holder is presented in figure 6.1.

The evaluation of all the ZnO samples with UV–Vis spectroscopy, and the absorption properties of
the films are presented in figure 5.7.

105
Figure 6.5. The transmittance spectra of the ZnO thin films obtained via UV–Vis spectroscopy. All samples
demonstrate above 80% transmittance between 400 nm and 1000 nm.

The data presented in figure 6.3 and figure 6.5 clearly demonstrated the potential of ZnO as an
alternative TCO without the need for indium or any other metal doping at room temperature, if the
position of the substrate with respect to the plasma is considered. The optical band gap of the deposited
thin films was obtained via the Tauc plot method, the results are presented in Table 6.1 for edge- and
centre-deposited ZnO thin films. The samples prepared on the edge also demonstrate a slightly higher
band gap.

Table 6.1. The optical band gap of the ZnO samples deposited at various conditions, comparing the results obtained
from samples positioned on the edge and centre of the substrate holder.

106
 Centre Samples Optical Band Gap (eV) Edge Samples Optical Band Gap (eV)

C100H 3.3 E100H 3.4

C150H
Coatings 2021, 11, x FOR PEER REVIEW
3.30 E150H 3.4 10 of 18

C200H 3.20 E200H 3.37

C100M 3.30
Table 2. The optical band gap of the ZnO samplesE100M 3.3
deposited at various conditions, comparing the
results obtained from samples positioned on the edge and centre of the substrate holder. The sam-
ples prepared on the edge also demonstrate a slightly higher band gap.
C150M 3.29 E150M 3.35
Centre Optical band gap (eV) Edge Optical band gap (eV)
C200M C100H 3.32 3.3 E200M
E100H 3.4 3.39
C150H 3.300 E150H 3.4
C100L C200H 3.25 3.200 E100L
E200H 3.38 3.35
C100M 3.300 E100M 3.37
C150L 3.20 E150L 3.39
C150M 3.290 E150M 3.3
C200L C200M 3.22 3.320 E200M
E200L 3.39 3.35
C100L 3.250 E100L 3.35
C150L 3.200 E150L 3.39
C200L 3.220 E200L 3.35
6.1.3 Analysis of the Plasma Emissions
Average 3.264 3.37
SD 0.047 0.032
The plasma emissions at the surface of the target were examined as described in the experimental
section. A comparative study with ofantheIZO
3.3. Analysis target
Plasma fitted on the magnetron to evaluate the variation in
Emissions
plasma emissions comparing We theexamined
two targets (IZO and
the plasma ZnO) was
emissions carried
at the surfaceout, considering
of the that with
target as described the IZO
in the
sample, a drastic variation of electrical properties associated with sample position was not
experimental section. We carried out a comparative study with an IZO target fitted on the
magnetron to evaluate the variation in plasma emissions comparing the two targets (IZO
observed(figure 6.2). and ZnO), considering that with the IZO sample we were not observing a drastic variation
of electrical properties associated with sample position (Figure 4).
The emissions from theThe plasma generate
emissions numerous
from the emission
plasma generate peaks,emission
numerous and thepeaks,
ratioandof the
these
ratiopeaks
of is
classically associated withthese
plasma
peaksconditions. However,with
is classically associated although peak intensities
plasma conditions. However, varied from
although centre
peak in- to
edge, observations purely basedvaried
tensities on thefromspectral
centre to plot
edge, did not indicate
observations purely an easily
based identifiable
on the spectral plotvariation
did
not indicate an easily identifiable variation between the two materials. To illustrate this,
between the two materials. To illustrate this, the data representing the emission of the plasma at 100
the data representing the emission of the plasma at 100 watts under a 10 sccm flow rate
watts under a 10 sccm flow rateZnO
for the forand
theIZOZnOtargets
and IZO targets are
are presented presented
in Figure 8. in figure 6.6.

Figure 8. The spectral emission of the plasma at 100 watts of power under a 6 sccm flow rate at the edge (blue) and centre
Figure 6.6. The
(orange) spectral
of the emission
target for of the
the IZO (left) andplasma at 100
ZnO (right) watts of power under a 6 sccm flow rate at the edge (blue) and
targets.
centre (orange) of the target for the IZO (left) and ZnO (right) targets.
In Figure 8 we can see that the intensity of various peaks is higher at the edge of the
target surface compared to the centre region. These emissions are associated with various
atomic transition in the plasma. The increase in intensity at the edge is due to the way that
the magnetron operates, given the magnets installed in it, leading to an area with higher
ion bombardment rate [42]. It is possible to analyse every peak intensity and to obtain
certain parameters such as the electron temperature from the data, which is a time inten-
sive approach and beyond the scope of this report. Instead, we implemented our simpler
107
 In figure 6.6 One can see that the intensity of various peaks is higher at the edge of the target surface
compared to the centre region. These emissions are associated with various atomic transition in the
plasma. The increase in intensity at the edge is due to the way that the magnetron operates, given the
magnets installed in it, leading to an area with higher ion bombardment rate [206]. It is possible to analyse
every peak intensity and to obtain certain parameters such as the electron temperature from the data,
which is a time intensive approach and beyond the scope of this report. Instead, A simpler assessment
was applied which was based on the colour function parameters of the plasma at the centre and edge
region of the target based on the previous work. This simple approach was applied to evaluate an overall
change in the plasma condition. The colour functions of the plasma light at the centre and edge regions
were calculated and are presented in figure 6.7. Using this approach, the difference in plasma conditions
became more apparent to the naked eye. The colour functions could represent the overall macro
environment of the plasma constituents without indulging in deep physics and statistical mechanics
associated with evaluating a plasma.

Figure 6.7. The x and y chromaticity index of the plasma light associated with the array of deposition conditions

From the results presented so far, although one can see certain variation between the emission
patterns associated with ZnO and IZO targets, the variations are interesting but do not seem to be
conclusive for explaining the observed variation in electrical conductivity of ZnO and IZO as presented

108
in figure 5.4. An in-depth peak ratio analysis can be performed to gain insights into the physical
parameters of the plasma, such as electron temperature; however, in the current report the focus was
just on describing the observations and such studies require a separate dedicated study report.

Up to this point, It can be assumed that the hydrogen impurity incorporation within a ZnO or IZO
lattice follows a certain reduced gradient from the edge to the centre. The hydrogen impurity can be
either interstitial or substitutional. The H+ in the ZnO system will occupy locations within the lattice, where
it can bind to an oxygen atom and form an O–H bond. Considering substitutional hydrogen, it is located
in the proximity of a nominal oxygen position within the lattice and behaves as a shallow donor. The donor
behaviour of the hydrogen impurity in ZnO clearly explains the conductivity observed in ZnO systems and
improvement of IZO conductivity.

This implies that there is a need to consider the state of the materials during their transport from
the target surface to the substrate. From these results, and reported studies discussed, one can assume
that the ZnO particles that form on the substrate surface have more hydrogen content at the edge location
compared to the centre region of the substrate holder. This observation can possibly be explained
considering the following concepts and assumptions:

• Firstly, hydrogen ions (along with argon ions) bombard the surface of the target material, resulting
in ejection of surface atoms from the surface of the target, while some of these ions will also
integrate into the target material.
• Secondly, the integration of the hydrogen into the surface of the target material as discussed and
based on referenced literature may lead to formation of regions at the target surface possessing
complex magnetic behaviour.

One way of explaining the observations reported in this study is to consider previously reported
studies where angular dependency of thin films depositions via the sputtering process were investigated.
During the sputtering deposition process it has been shown that geometrical shadowing of an incident
beam by the existing protruding parts of the growing surface profoundly affect the deposition’s
morphology [207]. This is referred to as competitive shadowing, which can affect the properties of the
thin films, in particular step coverage and microelectronic properties [208]. In a more recent study, Hippler
et al. [209] carried out a study on the angular dependence of plasma parameters and thin film properties
focusing on titanium and titanium oxide layers deposited via HIPIMS. They performed separate reactive
(argon/oxygen) and non-reactive depositions regimes using a Langmuir probe and substrates at various
angular positions (0°, 30°, 60° and 80°) with respect to the target [209]. In their study, significant variations
in electron density were reported depending on the reactive or non-reactive experimental regime, with
the non-reactive regime showing up to five times larger electron density values. They also measured the
mean electron energy (pulse time dependent), which demonstrated similar values between the two
regimes. However, most interesting are the values they demonstrated depending on the substrate
position with respect to the target, which finds significant relevance with the observations found in this
work. Both electron density and mean electron energy values are reduced at wider incident angles on the
target surface. It should be noted that the work in this study needs to be repeated mimicking the
experimental protocols carried out by Hippler et al., something that appears very interesting and will be
pursued in the future. Considering the observations reported here, such studies will give a significant
perspective in understanding the formation of the conductive zinc oxide films in presence of hydrogen
during the sputtering process. However, at this stage, based on the above reported research, if

109
incorporation of the hydrogen into the target material leads to the formation of surfaces at the target
with complex magnetic behaviours, then it may result in the target material itself interfering with the
magnetron’s magnetic fields and as such creating electron density and electron energy values dependent
on angular incidence on the target surface. The time dependency studies on plasma potential reported by
Hippler et al. at various angular positions can also be influenced by the target surface gaining magnetic
properties as well as due to the overall interaction of electron and ions in such a complex magnetic
environment.

Parallel to the above discussion, it has been reported that the sputtered flux from an ion bombarded
target surface is composed of atoms, polyatomic molecules and clusters [210-212], and various models
have been proposed to describe the sputtering of small clusters and molecules [213-219]. If one was to
assume that during the sputtering process, molecular ejection from the target surface occurs, there will
be a flux of both pure ZnO and hydrogen doped ZnO leaving the target surface with various speeds and
randomly dispersed direction vectors. As discussed, hydrogenated ZnO nanoparticles have been shown
to demonstrate ferromagnetism.

Hypothetically, the hydrogen doped ZnO particles possessing ferromagnetic properties are affected
by the magnetic field of the magnetron’s magnets. This effect creates additional vector forces that
ultimately promote the ferromagnetic hydrogen doped ZnO to the edge of the target, while the non-
ferromagnetic pure ZnO particles are condensed randomly across the substrate holder. This model is
illustrated in figure 6.8, where a very simple approach can be followed which considers a certain number
of small particles with a magnetic moment along a magnetic field between the N and S poles of the
magnetron’s magnet. Upon surface bombardment of the target with Ar/H ions, the target atoms acquire
kinetic energy for travel to the surface of the substrate. If these particles possess ferromagnetic properties
as hypothesized in this report, then there is dipole–dipole interaction that needs to be considered in this
model; an attractive and repulsive force vector will exist as these magnetic particles interact with the field
and their geometrical orientation within space. This vector either will combine with the initial vector or
will counter it depending on the particles’ positions, according to equations 6.1–6.4.

𝑈 = −𝑀𝐵𝑐𝑜𝑠q (6.1)

µ /Z
𝐵 = Fp 𝑥 V'
(6.2)

𝑇 = −𝑀𝐵𝑠𝑖𝑛q (6.3)

[\
𝐹𝑜𝑟𝑐𝑒 = − [V (6.4)

110
 However, this work needs to be followed up with simulation studies and more in
depth experimental work to verify the hypothesis proposed here, and the authors are
planning explore this study further accordingly. The magnetron in this study had a bal-
anced magnetic set up; our plan for future studies will be to carry out the experiments
with unbalanced magnetic and no magnet set up magnetron systems.

Figure 12. A simple approach of considering a certain number of small particles with a magnetic moment along a magnetic
field between the N and S poles of the magnetron’s magnet. The angle between the centres of the magnets and their
position defines how the larger magnet affects them.
Figure 6.8. A simple approach of considering a certain number of small particles with a magnetic moment along a
magnetic field between the N and S poles of the magnetron’s magnet. The angle between the centres of the magnets
4. Conclusions
and their position defines how the larger magnet affects them.
This study primarily is a report of observations associated with deposition of zinc
Considering the oxide
interaction
films inofthe
twopresence
hypothetical magnets
of hydrogen alfawith
along and argon
beta, in
andthese
will equations,
require furtherU is studies
the
to validate the assumptions and hypothetical statements in this report,
potential energy of a magnet “alpha” exposed to the magnetic field of magnet “beta”. B is the magnetic particularly similar
experiments to that described by Hippler et al. We demonstrated that it is possible to pro-
field of the magnet alpha, M is the magnetic momentum, θ is the angle between the axial angle of the
duce highly conductive ZnO thin films doped with hydrogen at room temperature. The
magnetic moment of films a magnetic dipole and the magnetic field generated by the magnetron magnet, R is
possess acceptable optical band gap and as such demonstrate above 80% transmit-
the distance betweentance the centres
and best ofsheet
the two magnets
resistance of and
~18 FΩissquare
force imposed on alpha.
(E200 samples). Considering
However, these
the procedure
equations, It can be seen that the angular geometry of one magnet to another as well
requires certain precisions in terms of substrate positioning and its trajectory withas its location inthe
space can lead to positive to negative values of force. Considering this and assuming the hypothetical
magnetron.
magnetic particles depicted in We figure
also 6.8, the difference
demonstrated the potentialbetween the point
of assessing c1/b1 andplasma
the sputtering c2/b2 viais the
colour
magnetic field gradient. In the central position, the magnetic field gradient is higher than in the edge as
function analysis, based on evaluating the specific ratios of area under emission peaks
a tool for monitoring the status of plasma during sputtering procedures. This method does
positions. Then, the attractive magnetic force will be higher in the centre than in the edges. Particles at
not require precise plasma analysis and complex calculations when detailed study of the
points b2 and c2 will plasma
experience
is nota required
more complex magnetic
and instead environment
provides countering
the operator their to
with an easy velocity awayob-
understand
from the target surface. The gas
serving flow rate,
parameter which
as an technically
indicator translates
of sputtering to the
plasma working chamber pressure
status.
and the plasma power, which translates to
The hypothesis the kinetic
proposed on theenergy of the sputtered
ferromagnetic properties particles, will affect
of the sputtered ZnOthepar-
process such that at higher chamberfurther
ticles requires pressures, the probability
examination and willofrequire
particledetailed
collisions increases,
analysis of thewhile higher
hydrogen dop-
plasma power leads to higher energy particles, which will be less influenced by the magnetic complexityour
ing variations with balanced and unbalanced magnetron configurations to validate
described. The effect conjecture. The integration of hydrogen into the target is an area that should be explored
of higher kinetic energy overcoming the magnetic complexity can be seen in figure
further to assess this hypothesis in future attempts. It is an interesting topic to explore by
6.3, where higher plasma power results in deposition of more conductive films at the edge region of the
us and other researchers interested in the field.
substrate holder. Higher chamber pressure would also result in more particles interacting with the
magnetrons magneticAuthor
field, and as such may
Contributions: reduce the effect
Conceptualization, of the field with
A.S.; methodology, A.S.; respect
software,to sputtered
A.H.; validation,
particles; however, such conclusions require extensive analysis of the magnetic fields during the
H.U., A.S. and A.H.; formal analysis, A.S; investigation, A.S.; resources, H.U.; data curation, A.A.;
writing—original draft preparation, A.S.; writing—review and editing, A.A. and A.H; visualization,
experiment, which is beyond the scope of this report.
A.S.; supervision, H.U.; project administration, A.S.; funding acquisition, H.U. All authors have read
and agreed to the published version of the manuscript.
The following hypothesis to explain the experimental observations can be proposed: during the
sputtering process, due to complex magnetic dipole interactions, hydrogen doped ZnO particles leaving
the target surface tend to accumulate on the edges of the substrate holder region. This results in thin
films with better electrical conductivity, while at the centre, ZnO particles that do not possess magnetic

111
properties are more predominant, and as such, samples prepared in the centre location of the substrate
holder are less conductive.

This process occurs both in IZO and ZnO targets, hence the samples prepared at the edge of the
substrate demonstrate higher conductivity due to the higher hydrogen doping. In IZO samples, however,
since the presence of indium already acts as a carrier concentration enhancer, a slight improvement in
conductivity can be seen; however, in ZnO samples, because only hydrogen acts as the electron donor in
the lattice, the large variation of conductivity gradient from the edge to the centre was noticed.

However, this work needs to be followed up with simulation studies and more in depth experimental
work to verify the hypothesis proposed here, and the authors are planning explore this study further
accordingly. The magnetron in this study had a balanced magnetic set up; the plan for future studies will
be to carry out the experiments with unbalanced magnetic and no magnet set up magnetron systems.

6.2 Conclusion
This study primarily is a report of observations associated with deposition of zinc oxide films in the
presence of hydrogen along with argon and will require further studies to validate the assumptions and
hypothetical statements in this report, particularly similar experiments to that described by Hippler et al.
it was demonstrated that it is possible to produce highly conductive ZnO thin films doped with hydrogen
at room temperature. The films possess acceptable optical band gap and as such demonstrate above 80%
transmittance and best sheet resistance of ~18 Ω square and an average thickness of 850nm (E200
samples). However, the procedure requires certain precisions in terms of substrate positioning and its
trajectory with the magnetron. The work carried out and subsequent findings presented in this chapter
has been published in a journal [220].

The potential of assessing the sputtering plasma via colour function analysis was also demonstrated,
based on evaluating the specific ratios of area under emission peaks as a tool for monitoring the status of
plasma during sputtering procedures. This method does not require precise plasma analysis and complex
calculations when detailed study of the plasma is not required and instead provides the operator with an
easy to understand observing parameter as an indicator of sputtering plasma status.

The hypothesis proposed on the ferromagnetic properties of the sputtered ZnO particles requires
further examination and will require detailed analysis of the hydrogen doping variations with balanced
and unbalanced magnetron configurations to validate this conjecture. The integration of hydrogen into
the target is an area that should be explored further to assess this hypothesis in future attempts. It is an
interesting topic to explore by other researchers interested in the field.

112
Chapter 7 Implementation of Machine Learning/Deep learning
concepts: A step towards the digitisation of the sputter deposition
process

Moving away from characterisation of the plasma through its colour in the previous chapters, this
chapter focuses on the modelling of the plasma discharge by applying Artificial Intelligence as a
monitoring and predictive tool. There is reason to believe that artificial intelligence and deep learning can
be the answer if the objective is to fully automate and digitise the industrial-scale sputtering process. A
brief introduction on AI/Machine learning has been provided in in Chapter 2.3. A more detailed
explanation on how specific deep learning technique was applied to the sputter deposition process as well
the materials involved can be found in chapter 3.1.4.

7.1 Results and Discussion

7.1.1 Artificial Neural Networks to predict sheet resistance of IZO thin films deposited via
Plasma deposition
The objective of this research carried out was to push all the mentioned complexities into the black
box of an artificial neural network (ANN) and a convolutional neural network (CNN). With modern
computing powers and artificial intelligence (AI) based data assessment methodologies, it was aimed to
explore implementing an alternative approach in plasma diagnostics during the sputter deposition process
to assess the qualitative parameters of thin films as they were deposited.

The Integral Approach

The area under the spectral peaks from 300nm to 900nm in 50nm intervals was calculated yielding
a data frame with 12 columns linked to these areas. This resulted in a data frame with 517 rows and 12
columns associated with peak area values, a vector of thin film sheet resistance values represented the
13th column. Figure 6.6 presents the structure of the neural network model for the integral approach to
model construction; it also shows the predictions made by the model for sheet resistance of 24 samples
that were randomly chosen. The model is consisted of five dense layers. Layers 1 to 5 have 30, 20, 8, 4
and 1 neuron/s, respectively, generating a total of 1219 parameters requiring computation. The model’s
R2 value is 0.795. The Adam optimizer was used in this model [221], with a learning rate of 0.01 and Huber
loss function [222].

113
Figure 7.1. The structure of the neural network model designed for training with the integral approach.

The model was given the task to predict the sheet resistance for of 24 samples which was compared with
the actual value calculated via the four-point probe system. The R2 value for the models’ predictions was
0.795 with a mean squared error of 0.7.

When attempts were made to re-run the samples, the model demonstrated signs of instability; this could
be as a result of the size of the data frame being too small. Large datasets are required to train neural
networks and the dataset used was relatively smaller, although sizeable from a practical material
engineering point of view, than the standard datasets employed in machine learning and dataset.

The Spectral Approach

The goal of the spectral approach was to utilized the peak intensity values in the 195nm-1104nm
spectrum range with a 0.2nm resolution. This yielded a data frame consisting of 113 rows and 4552
columns. Any negative values from the spectrometer signal were deemed as noise and thus set to 0. The
columns of the data frame depcited the ratio of the intensity of the spectral signal at each wavelength
over the total sum of the intensity. The structure of the model and its predictions are illustrated in figure
7.2. The model consisted of eight dense layers. Layers 1 to 8 have 4551, 2000, 500, 300, 50, 8, 4 and 1
neuron/s, respectively, generating a total of over 30 million parameters requiring computation. The
model’s R2 value is 0.153.

114
Figure 7.2. The structure of the neural network model designed for training with the spectral approach.

From the results presented in figure 6.7, One can see that the model was significantly bigger and
more complex than the previous model managing the integral data frame. The model was given the task
to handle over 30 million parameters compared to the 1219 parameters of the previous model, which was
computationally very demanding. Despite this, the model’s statistical performance with R2 value of 0.15
was significantly better than the 0.795 R2 value of the previous model.

A number of neural network models were created to manage the spectral data frame. Since the data
frame was composed of large dimensions, a technique known as Principal Component analysis (PCA) was
utilized in efforts to reduce this [223]. PCA, a form of feature engineering, reduced the number of features
columns from 4552 to a mere 12; it was able to retain hidden factors from the dataset and defined data
using less components, which is explained the discrepancy in the data. Thus, it resulted in the reduction
of computational complexity. These 12 principal components were fed in the neural model as input
features. The model structure and the predictions made on physically measured samples are shown
in figure 7.3. The model is consisted of seven dense layers. Layers 1 to 7 have 10, 10, 10, 10, 8, 4 and 1
neuron/s, generating a total of 349 parameters demanding computation; the model’s R2 value is 0.883.
The model reported here displayed a dominant performance when compared to the previous models.

115
Figure 7.3. The structure of the neural network model designed for training with the combined principal component
analysis and spectral approach.

As depicted in figure 7.3, in the PCA/spectral data frame integra•on approach, the model exhibited
superior performance compared to the integral approach experiment, achieving an R2 value of 0.883,
notably surpassing the performance of prior models.

The Image Recognition Approach

The spectral data, spanning from 200 to 1100 nm with a resolu•on of 0.2 nm, resulted in an array
comprising 4500 data points. This array underwent transforma•on into a matrix with shape of (50, 90).
Subsequently, this matrix was translated into an image format, where each value of the matrix
corresponded to a pixel value within a 50-pixel x 90-pixel image. This process is depicted in figure 7.4.

Figure 7.4. Converting a spectral plot (a) into an image (b). The spectral data are initially summed up and each peak
value is calculated by dividing the original peak value over the sum. It is then converted into a matrix of (50, 90) shape.

116
 The image derived directly from the spectra was ini•ally employed to train several convolu•onal
neural network models. However, none of these models could effec•vely learn from the image in its
original format, as CNN models typically operate more efficiently with standardized values. Subsequently,
the pixel values of the image were scaled using two dis•nct normaliza•on methods: min-max scaling and
standard scaling, as outlined in Chapter 3.1.4. New images were then generated based on the normalized
pixel values of the original image. Two examples of these normalized images are illustrated in Figure 7.5.

Figure 7.5. The original image (top) is converted to an image from normalised values via standard scaling (bottom
left) and min-max scaling methods (bottom right).

Subsequently, the normalized images were u•lized to train the model for predic•ng the sheet
resistances of the IZO films deposited through the corresponding plasma process. Following training, the
normalized images were subjected to similar convolu•onal models for tes•ng purposes. The outcomes of
these tests are showcased in Figures 7.6 and 7.7. Interes•ngly, the min-max scaled data yielded the most
accurate predic•ons of sheet resistance, as evidenced by the results and corresponding R2 values.
Specifically, the standard scaled images displayed R2 values of 0.642, whereas the min-max scaled images
exhibited significantly higher R2 values of 0.897.

117
Figure 7.6. The structure of the convolutional neural network model designed for training with standard scaled values
of the pixels in image of the spectra.

The model in figure 7.6 is consisted of two convolutional layers and two max-pooling layers which are
then flattened and fed to a deep network with five dense layers. Where layers 1 to 5 had 5000, 5000, 100,
20, and 1 neuron/s, generating a total of 57,518,453 parameters requiring computation, the model’s
R2 value was 0.642.

118
Figure 7.7. The structure of the convolutional neural network model designed for training with the min-max scaled
values of the pixels in images of the spectra.

The model in figure 7.7 is consisted of two convolutional layers and two max-pooling layers which are
then flattened and fed to a deep network with five dense layers. Where layers 1 to 5 had 5000, 5000, 100,
20, and 1 neuron/s, generating a total of 57,518,453 parameters requiring computation, the model’s
R2 value was 0.897.

The findings thus far suggest that employing min-max scaling on the data enhances the CNN model's
ability to accurately discern pa•erns and hidden features within the images. At this stage, each image,
characterized by a matrix of shape (50, 90), underwent the Singular Value Decomposi•on (SVD) process,
as detailed in Chapter 3.1.4. Despite both standard scaled and min-max scaled image matrices exhibi•ng
similar condi•on numbers, there were discrepancies in the distribu•on of informa•on across singular
values when SVD was applied to the images. Figure 7.8 illustrates the typical distribu•on of singular
values, known as the "scree plot," for each layer of decomposed standard scaled and min-max scaled
matrices. In a min-max scaled matrix, the first singular value (represen•ng the maximum value)
encapsulated a higher percentage of the total sum of values compared to the standard scaled
decomposed matrices.

119
Figure 7.8. The scree plot of singular values for (A) min-max scaled images and (B) standard scaled images.

In the subsequent phase aimed at enhancing the CNN model's accuracy, the spectral image matrices
were systema•cally decomposed layer by layer (a total of 50 layers for each image). Subsequently, the
CNN model was trained using images reconstructed from smaller layers—or even individual layers—
derived from the SVD process. Given the underperformance of standard-scaled whole images in previous
trials (refer to figure 7.6, R2 value 0.642), efforts were directed towards achieving improved performance
with min-max scaled values. However, contrary to expecta•ons, an image reconstructed from the first
single layer of standard-scaled images decomposed by SVD surpassed all other CNN training endeavours,
yielding an impressive R2 value of 0.935. Notably, this model demonstrated excep•onal stability across
repeated training sessions. Conversely, reconstruc•ng an image from the first single layer of a min-max
scaled image yielded poor results in training the CNN model, with an R2 value of 0.74.

This discovery proved highly intriguing: while the original standard-scaled images performed
inadequately in training the CNN model, the min-max scaled originals exhibited notable success.
Moreover, the reconstruc•on of an image from the first single layer of standard-scaled images
outperformed all models in terms of both accuracy and stability. Consequently, given that these
experiments were conducted across various chamber pressures, the reconstructed images from the first
layer of the singular value decomposed original images were converted into a spectral format. These
findings are illustrated in figures 7.9 and 7.10, where each column represents four randomly selected
spectral reconstruc•ons of plasma emissions from the first layer of SVD of the original image at specific
chamber pressures.

120
Figure 7.9. Four randomly selected conversions of the first layer of the image decomposition via SVD, associated with
process chamber pressures into a spectral format. For standard scaled images. x-axis represents emission wavelengths
and y-axis is emission intensity.

Figure 7.10. Four randomly selected conversions of the first layer of the image decomposition via SVD associated with
process chamber pressures into a spectral format. For min-max scaled images, x-axis represents emission wavelengths
and y-axis is emission intensity.

Table 7.1. Prediction of the thin film sheet resistance from a CNN model learning from the first layer of a singular
value decomposed image of a plasma spectra (standard scaled).

Predicted Real Relative Error

7.598 7.5 1.31

13.955 13.5 3.37

7.360 7.5 1.87

121
 10.832 11.4 4.98

9,967 9.6 3.82

7.487 7.5 0.18

10.893 11 0.97

10.255 9.6 2.32

9.377 9.6 2.77

10.996 10.7 2.77

10.518 10.7 1.71

10.457 10.7 2.27

7.467 7.5 0.45

13.073 13.5 3.16

11.098 11.4 2.65

9.937 10 0.63

7.329 7.5 2.27

10.551 11.4 7.44

10.290 10.7 3.83

10.045 9.8 2.50

11.811 10.7 10.38

11.457 11 4.32

10.474 10 4.74

Table 7.1 showcases the predictions derived from the standard scaled image data by the CNN model.
An explanation can be made for the observed training performance associated with the standard scaled
first layers images when looking at figures 7.9 and 7.10 in comparison. Detecting specific features in a
particular image is the basis of the operation of CNNs. These features (and their detection with the black
box of a DNN) lack intuition and can sometimes be mysterious. As a result, standard scaling of images,
derived from the original plasma spectrum, resulted in feature consolida•on within specific column and
row spaces of the image. Conversely, min-max scaling led to a distribu•on of features across alterna•ve
row and column vector spaces. Nonetheless, considering the long-term objec•ve of inves•ga•ng the
applicability of deep learning models in transi•oning toward digi•zed spu•ering processes, the model's
performance was deemed sa•sfactory. Nevertheless, this presents an intriguing avenue for further
explora•on, par•cularly by researchers with a primary focus on the prac•cal engineering aspects of deep
learning models.

122
 Table 7.2 encapsulates the overall performance of all experiments conducted in this study, in terms
of their predic•ve accuracy. Notably, the CNN model approach achieved the highest R2 value. The stability
metric reflects the repeatability of consecu•ve R2 values produced by the model, indica•ng the degree of
devia•on from the maximum R2 obtained. A devia•on of 0.15 points is categorized as poor stability, 0.1
as moderate, and 0.05 as good, when the models were repeated ten •mes.
Table 7.2. The overall performance of the models and the data approach employed in experiments.

7.1.2 Implementing Supervised and Unsupervised Deep-Learning Methods to Predict

Sputtering Plasma Features

As detailed in chapter 3.1.4. This set of experiments reports the follow up of the previous set of
experiments whereby supervised and unsupervised deep-learning methods were implemented i.e. CNN
to predict the sputtering growth parameters by feeding the spectral data of the plasma to it.

Overall, 2352 images were used (588 images per method of 2-dimensional matrix formation) which
is associated with various targets—working gas pressures, plasma power, and hydrogen impurity—that
were present in the argon gas. 80% of the data was randomly selected for training the models and used
the remaining 20% to evaluate the prediction of the models and their accuracy.

Regression models
A very interesting initial observation made was that the models were trained best on the data sets
for which the peak-intensity ratio operations were carried out. In fact, without this step, standard scaling
alone was insufficient to train the models, as extremely poor results and prediction were observed.
However, the min-max scaling resulted in desirable accuracies, as in a sense it converts the data points to
a minimum–maximum range of 0 to 1. However, once the peak-intensity ratio operation was carried out,
standard scaling and min-max scaling were equally good for training the models for desirable accuracy.
However, all the results presented in this report are associated with data that were min-max scaled. The
results for each set of samples and experiments were investigated separately.

When presenting the results (figures 7.11-7.14), ‘Original’ refers to the min-max-scaled stack-
overlay-formed images turned into a 2d matrix. ‘Original_SVD’ refers to the first-layer reconstruction of
the original image after SVD decomposition. ‘SOM’ refers to images generated from the spectra via the
self-organizing map method and ‘SOM_SVD’ refers to the first-layer reconstruction of the SOM image
after SVD decomposition. The R-squared (R2) and mean squared error (MSE) of the models predictions

123
are also presented. The results of predicting the zinc concentration in the target are presented in figure
7.10. It can be seen that SVD reconstructions of the SOM-formed images performed less effectively.

Figure 7.11. The real value of the zinc content in the target (wt%) on the ‘y’ axis and the prediction of the zinc content
(wt%) of the target composition in the ‘x’ axis (A) Original, (B) SOM image, (C) SOM_SVD, (D) Original_SVD images

The results of predicting the indium concentration in the target are presented in figure 7.12. Once
again, One can see that the SVD reconstruction of the SOM-formed images performed less
effectively. Figure 7.13 and figure 7.14, respectively, demonstrate the predictions of the gas-flow rate
and plasma power.

124
Figure 7.12. The real value of the indium content in the target (wt%) on the ‘y’ axis and the prediction of the indium
content (wt%) of the target composition in the ‘x’ axis. (A) Original, (B) SOM, (C) SOM_SVD image, (D) Original_SVD
images.

125
Figure 7.13. The real values of gas-flow rate (sccm) during the sputtering process are presented in ‘y axis and the
predicted values in ‘x axis. (A) Original, (B) SOM, (C) SOM_SVD, (D): Original_SVD

126
Figure 7.14. The real values of plasma RF power (Watts) during the sputtering process are presented in ‘y axis and
the predicted values in ‘x axis. (A) Original, (B) SOM image, (C): SOM_SVD, (D) Original_SVD

Out of all the results, it can be seen that the SVD reconstruction of the SOM model (SVDS) produced
the least accurate predictions. The original stack-overlay method of reconstructing a two-dimensional
array from the one-dimensional array of the spectra seems to be sufficient for training the CNN model for
accurate predictions.

Classification model
When the model was used as a classifier rather than a regressor to predict the presence of hydrogen
impurity and impurity in the indium-oxide-based target, excellent results with all forms of two-
dimensional conversion techniques were obtained, as presented in figure 7.15 and figure 7.16. In both
cases, The SVD reconstruction of the SOM method demonstrated one inaccurate classification, while all
other models performed with 100% accuracy. The accuracy metrics associated with the hydrogen
presence and indium-oxide target purity are presented in Table 1, Presenting the accuracy metrics of the
CNN model’s performance trained with images formed from the four methods of two-dimensional image
formation; presence of hydrogen in the working gas and impurity in the indium-oxide target.

127
Figure 7.15. Confusion matrices demonstrating the accuracy of the two-dimensional images to generate images for
training ANN model for classification, on presence of hydrogen impurity in the sputtering working gas.

128
Figure 7.16. Confusion matrices demonstrating the accuracy of the two-dimensional generated images for training
CNN model for classification of 100% purity of indium-oxide sputtering target.

Table 7.3. Confusion matrix of model classification associated with Hydrogen presence and impurity in the indium
oxide target.

129
7.2 Conclusion
The integra•on of deep learning models into spectral analyses of spu•ering plasma offers the
poten•al to digi•ze the spu•ering process. This implies that during deposi•on, specific film proper•es
could be es•mated solely by observing the characteris•cs of the plasma. Through various modelling
approaches, as outlined in Table 7.2, the CNN model approach yielded the highest R2 value with
commendable stability. The study highlighted that result accuracy depended on both the quan•ty of data
points and the number of process parameters considered. Addi•onally, the modelling approach adopted
played a crucial role, par•cularly when focusing on specific electrical proper•es such as sheet resistance
of the TCO. Undoubtedly, future endeavours will encompass explora•on of addi•onal sample features,
including charge carrier concentra•on, mobility, crystal structure parameters, and op•cal band gap,
aiming for enhanced op•miza•on and control. Future research will also entail adjus•ng other
experimental parameters, such as sample rota•on speed, incident angle of the magnetron to the
substrate, and their rela•ve distance, to achieve highly accurate predic•ons of the func•onality of the
deposited films.

The results presented in this chapter (figures 7.11-16) clearly demonstrate the potential of deep
learning and, in particular, computer-vision-based CNN models as the tool of choice to be utilised in
associating the spectra emissions of plasma to the operating conditions of sputtering units. Plasma power
and gas-flow rates do indeed affect plasma glow and its spectral characteristics. Higher plasma power can
lead to higher energy ions and higher rates of excitation, which also affect the property of the thin film
deposited while demonstrating a unique glow feature in the plasma emission. Higher gas-flow rates
increase the chamber pressure, resulting in a greater flux of atomic collisions and variation in the ion
distribution. This, in turn, affects the property of the thin film deposited. The models presented were
shown to be successful at predicting various features associated with the process, such as (wt%) content
of the target ingredients in figure 7.11 and figure 7.12 and the gas-flow rate and plasma power, as
presented in figure 7.13 and figure 7.14. The model also demonstrated the ability to predict the presence
of an additional gas (hydrogen) as a 5% impurity in the argon working gas and to assess the purity of the
target itself, as presented in figure 7.15 and figure 7.16. In the experiments carried out, a static CNN
construction was trained with the images that were formed from the four described two-dimensional
images. By further adjusting the CNN model, higher accuracies are achievable. It was shown that the
spectral data can be easily used to form an image that can benefit from the power of modern computer-
vision deep-learning models. This simple approach even outperformed the more complex SOM method
of two-dimensional array formation.

All these parameters are critical in defining the properties of the thin films produced via this form of
plasma sputtering. When neural networks can associate the spectral features of the plasma with a
numerical assessment of these critical operating features, they can also be implemented to predict the
properties of thin films. This was demonstrated by focusing on a single sputtering target (indium zinc
oxide) and assessing the electrical property of the produced thin films. The range of the target materials
was extended to further validate the observations seen in section 7.1.

It was acknowledged that further improvements in model construc•on could be achieved with
access to addi•onal data, par•cularly from other spu•er systems, which unfortunately were not available.
Nonetheless, the work conducted within these constraints represents a significant step towards

130
establishing correla•ons between digital datasets and the desired func•onality of TCO films with a
reasonable level of accuracy. Consequently, this research holds significance for optoelectronic industries
employing physical vapor deposi•on techniques , to produce high-yield products.

In conclusion, the presented results emphasise the need for the establishment of a global repository
where data generated at various research levels by diverse research groups, including material engineers
and scien•sts, can be shared for analysis and model training. In the long run, this ini•a•ve will reduce
research and development costs and expedite the advancement of novel materials and technologies,
ul•mately benefi•ng industries as they transi•on towards efficient and cost-effec•ve scaling-up of their
produc•on processes.

Chapter 7.1 demonstrated how these models can be used to predict the physical properties of thin
films, with a focus on a single material. In Chapter 7.2, it has been shown that these models are capable
of predicting the nature of plasma at a high level of precision and, therefore, one can conclude that these
can be instrumental in predicting various other material properties associated with thin films produced
via sputtering. The deep learning trial studies carried out and subsequent findings in this chapter have
also been presented in the format of publications [224,225].

Proficient researchers specializing in data science, ar•ficial intelligence (AI), and machine learning
are con•nuously refining algorithms, loss func•ons, backpropaga•on models, and op•miza•ons to
advance this field as a universal tool for various scien•fic domains. Consequently, the models showcased
in this study have the poten•al for further refinement to a•ain greater accuracy. Therefore, integra•ng
AI processes, as demonstrated in this inves•ga•on and its correla•on to func•onal proper•es, could
greatly enhance the digi•zed manufacturing process in the realm of thin film deposi•ons.

131
Chapter 8 Summary, Highlights and Future Work

8.1 Summary
The thesis is an attempt to bring together the research work that has explored and demonstrated
two novel and simple approaches viz. the colour characterisation method and the application of Artificial
Intelligence/Machine Learning to the diagnosis and evaluation of the glow discharge during RF magnetron
sputtering. The key findings of the work are summarized below:

In chapter 4, the photometric properties of a radio frequency (RF)-based sputtering plasma source
were monitored through optical spectroscopy. The colour of the plasma source was deduced based on
conventional chromaticity index analysis and it was compared to the direct spectral data plots of the
emission peaks to investigate the possibility of characterising the plasma based on its specific colour and
exploring the potential of defining a new method by which the plasma sputtering process can be
addressed based on the plasma colour parameters. The intention of this investigation is to evaluate the
possibility of simplifying the monitoring and assessment of the sputtering process for applied scientists
operating plasma sputter deposition systems. a viable potential for this technique was demonstrated in
terms of providing information regarding the stability of the plasma, chamber pressure, and plasma
power; however, further work is underway to verify and assess a relationship between the quality of the
thin film coating and the colour characteristics of the deposition plasma. Here, the focus was solely on the
feasibility of such an approach and demonstrate interesting observations. There was a clear linear
relationship between the colour functions and the plasma power, while the stability of the sputtering
plasma can be assessed based on the plasma colour functions. The colour functions also follow a unique
pattern when the working gas pressure is increased.

Chapter 5, showcases the spectral properties and colour functions of a radio frequency (RF)-based
sputtering plasma source was monitored during consecutive sputter deposition of zinc doped indium
oxide (IZO) thin films under argon and argon/hydrogen mix. The effect of target exposure to the hydrogen
gas on charge density/mobility and spectral transmittance of the deposited films was investigated. It was
demonstrated that consecutive exposure to the hydrogen gas during the deposition process progressively
affects the properties of thin films with a certain degree of continuous improvement in electrical
conductivity while demonstrating that reverting to only argon from argon/ hydrogen mix follows a
complex pathway, which has not been reported previously in such detail to one’s knowledge. It was then
demonstrate that this effect can be used to prepare highly conductive zinc oxide thin films without indium
presence and as such eliminating the need for the expensive indium addition. As demonstrated, the
complexity observed in emission spectra can be simply identified by monitoring the colour of the plasma
through its colour functions, making this technique a simple real-time monitoring method for the
deposition process.

Chapter 6 investigates the complexity of the reactive sputtering of highly conductive zinc oxide thin
films in the presence of hydrogen at room temperature. The importance of precise geometric positioning
of the substrate with respect to the magnetron to achieve maximum conductivity was studied. The results
were examined which entailed studying the electrical properties of the deposited thins films based on
their position on the substrate holder relative to the magnetron. By considering early reports by other
researchers on the angular dependency of plasma parameters and the effect of hydrogen doping on
electric and magnetic properties of hydrogen-doped zinc oxide, a hypothesis was put forward regarding

132
the hydrogenated ZnO compounds exhibiting ferromagnetic properties near the edge; further tests are
required to be carried out to verify this hypothesis. Overall, this report presents the preparation highly
conductive zinc oxide thin film coatings via RF sputtering with hydrogen presence along with argon as the
sputtering gas.

Moving away from evaluation of the plasma via colour characterisation, the feasibility of applying
artificial intelligence for the modelling of the glow discharge in chapter 7 was explored. Deep learning
models were implemented to examine the accuracy of predicting a single feature (sheet resistance) of
thin films of indium-doped zinc oxide deposited via plasma sputter deposition by feeding the spectral data
of the plasma to the deep learning models. It was demonstrated that artificial neural networks could be
implemented as a model that could predict the sheet resistance of the thin films as they were deposited,
taking in only the spectral emission of the plasma as an input with the objective of taking a step toward
digital manufacturing in this area of material engineering. This research work is concluded in by presenting
the spectral emission data from the plasma glow of various sputtering targets containing indium oxide,
zinc oxide, and tin oxide were obtained. The plasma was generated at various power and chamber
pressures. These spectral data were then converted into two-dimensional arrays by implementing a basic
array-reshaping technique and a more complex procedure utilizing an unsupervised deep-learning
technique, known as the self-organizing-maps method. The two-dimensional images obtained from each
single-emission spectrum of the plasma mimic an image that can then be used to train a convolutional
neural network model capable of predicting certain plasma features, such as impurity levels in the
sputtering target, working gas composition, plasma power, and chamber pressure during the machine
operation. The single-array-to-2D-array conversion technique, coupled with deep-learning techniques
and computer vision, was shown to achieve high predictive accuracy and can, therefore, be fundamental
to the construction of a sputtering system’s digital shadow.

8.2 Highlights of the Research

The research work conducted has evolved from the studies conducted for the basic understanding of
the plasma characteristics in a sputtering chamber and its relationship with the vacuum processing
parameters using advanced spectroscopy and predictive modelling tools. The optimisation of the
sputtering process in pinpointing a process condition in vacuum process industry has always been a holy
grail and is limited to finding recipes for each vacuum instrument, the research work was conducted with
a motivation to find these limiting factors drawing its origins from the plasma source itself. Whilst each
chapter registers the conclusion of the studies conducted along with the summary of the work in the
preceding section, the highlights of the results can be presented as following.

• The characteristic colour of RF plasma has been successfully linked with the chromaticity index
and was demonstrated for the first time a linear relationship between the colour functions and
the plasma power, whilst a defined unique patterns for the stability of the plasma and the effect
of variations of chamber gas pressure, thus making this technique a simple real-time monitoring
method for the deposition process.
• A highly conductive (18 ohms/sq) and transparent (>80% transmission in the visible range) ZnO
films have been obtained at room temperature using hydrogen doping, making it one of the useful
TCOs for low temperature processing required for perovskite and organic solar cells.
• A correlation has been established for the first time of extensive set of electrical sheet resistance
of IZO thin films data and the RF plasma glow characteristics, which has also been validated by

133
 modelling tools involving artificial intelligence viz. deep learning and computer vision methods.
These have achieved high predictive accuracy and thus can be applied to construction of
sputtering system’s digital shadow for high value and high yield manufacturing.

8.3 Future work

Extensive work on incorporating deep learning methods in the sputter deposition process to further
establish the design of a digital shadow is underway. This includes feeding the spectral emission data of
the glow discharge into a CNN deep learning model to predict an array of electrical properties (such as
carrier concentration & mobility) of a range of thin film TCOs. Plasma sputtering is a highly complex
process which involves deep theoretical physics to explain which is a limiting factor for applied/materials
scientist who want further to explore this type of deposition technique; This gap in knowledge can by
addressed by incorporating AI to establish a relationship between the plasma sputtering process and the
properties of thin films. The application of AI does not have to be limited to the deposition of TCOs only
but can also be applied to a wide variety of thin film semiconducting materials; one such example would
be photoactive materials. The implementation of AI i.e. Machine Learning and Deep learning methods in
the manufacture of thin films materials (and the semiconductor industry as a whole) can bring forth great
prospects leading to enhanced productivity, efficiency and performance whilst reducing costs.

Since Frank Rosenbla• proposed the concept of the perceptron in the 1960s [226], followed by the
ground-breaking contribu•ons of Hinto, LeCun, and Bengio in the 1980s [227], the field of computer
science has witnessed con•nuous advancements. Presently, with leading companies like Google and
Facebook ac•vely leveraging its capabili•es, there's a surge in the development of increasingly intricate
deep-learning models, expanding the horizons of numerical analysis. Consequently, there's a growing
necessity for researchers in tradi•onal domains such as material science to embrace and apply this
technology in their work.

Integra•ng deep learning models into plasma-assisted thin-film deposi•on processes could result in
the crea•on of a digital replica for spu•er-deposi•on, a cri•cal step in producing numerous thin films used
in various everyday products. The primary challenge lies in acquiring the vital data. Deep learning models
exhibit enhanced accuracy and efficiency when trained on substan•al datasets. Incorpora•ng these
models into established engineering or materials-engineering domains will necessitate strategies for
accessing or compiling structured data. Therefore, efforts aimed at genera•ng an extensive repository of
spectral emissions linked to the spu•ering of diverse materials will play a pivotal role in establishing a
digital counterpart for plasma-spu•ering processes.

134
References

1. Gonzalez Vazquez, I.; Milasi, S.; Carretero Gomez, S.; Napierala, J.; Robledo Bottcher, N.; Jonkers, K.;
Goenaga, X. The Changing Nature of Work and Skills in the Digital Age (EUR 29823 EN); 2019.
2. Ginley, D.S.; Hosono, H.; Paine, D.C. Handbook of Transparent Conductors; Springer: Berlin, Germany, 2011;
pp. 353–425.
3. Ginley, D.S.; Bright, C. Transparent Conducting Oxides. MRS Bull. 2000, 25, 15–18.
4. Hosono, H. Recent progress in transparent oxide semiconductors: Materials and device application. Thin
Solid Films 2007, 515, 6000–6014.
5. Ellmer, K. Past achievements and future challenges in the development of optically transparent
electrodes. Nat. Photonics 2012, 6, 809–817.
6. Granqvist, C.G. Transparent conductors as solar energy materials: A panoramic review. Sol. Energy Mater.
Sol. Cells 2007, 91, 1529–1598.
7. King, P.; Veal, T.D. Conductivity in transparent oxide semiconductors. J. Phys. Condens. Matter 2011, 23,
334214.
8. Calnan, S.; Tiwari, A.N. High mobility transparent conducting oxides for thin film solar cells. Thin Solid
Films 2010, 518, 1839–1849.
9. Morales-Masis, M.; Nicolas, S.M.D.; Holovsky, J.; Wolf, S.D.; Ballif, C. Low temperature high-mobility
amorphous IZO for silicon heterojunction solar cells. IEEE J. Photovolt. 2015, 5, 1340–1347.
10. Khachatryan, H.; Kim, D.J.; Kim, M.; Kim, H.-K. Roll-to-Roll fabrication of ITO thin film for flexible
optoelectronics appli-cations: The role of post-annealing. Mater. Sci. Semicond. Process 2018, 88, 51–56.
11. Morales-Masis, M.; De Nicolas, S.M.; Holovsky, J.; De Wolf, S.; Ballif, C. Low-Temperature High-Mobility
Amorphous IZO for Silicon Heterojunction Solar Cells. IEEE J. Photovoltaics 2015, 5, 1340–1347.
12. Martins, R.; Almeida, P.; Barquinha, P.; Pereira, L.; Pimentel, A.; Ferreira, I.; Fortunato, E. Electron transport
and optical characterization amorphous indium zinc oxide films. J. Non Cryst. Solids 2006, 352, 1471–1474.
13. Taylor, M.P.; Readey, D.W.; Van Hest, M.F.A.M.; Teplin, C.W.; Alleman, J.L.; Dabney, M.S.; Gedvilas, L.M.;
Keyes, B.M.; To, B.; Perkins, J.D.; et al. The remarkable thermal stability of amorphous In-Zn-O transparent
conductors. Adv. Funct. Mater. 2008, 18, 3169–3178.
14. Park, Y.R.; Nam, E.-K.; Boo, J.-H.; Jung, D.-G.; Suh, S.-J.; Kim, Y.-S. Hydrogenated In-doped ZnO thin films for
the new anode material of organic light-emitting devices: Synthesis and application test. Bull. Korean Chem.
Soc. 2007, 28, 2396–2400.
15. Park, Y.R.; Kim, J.; Kim, Y.S. Growth and characteristics of hydrogenated In-doped ZnO thin films by pulsed
DC magnetron sputtering. Appl. Surf. Sci. 2009, 256, 1589–1594.
16. Hosono, H.; Ueda, K. Transparent Conductive Oxides Springer Handbook of Electronic and Photonic
Materials; 2017; pp 1391–1404.
17. Blakemore, J. S. Semiconducting and Other Major Properties of Gallium Arsenide. J.Appl. Phys. 1982, 53.
18. Dixit, R.; Gupta, S.; Kumar, P.; Sikarwar, S.; Yadav, B. C. Preparation and Properties of Transparent
Conducting Oxide (TCOs) Thin Films : A Review. Int. J. Innov. Res. Sci. 2017, 6(2).
19. https://physics.stackexchange.com/questions/497212/why-is-the-carrier-distribution-density-of-states
different-in-doped-semicondu
20. Sze, S. M. Physics of Semiconductor Devices, 2nd Edition.; John Wiley and Sons: New York,1981.
21. Fan, J. C. C.; Goodenough, J. B. X-ray Photoemission Spectroscopy Studies of Sn-doped Indium-oxide Films.
J. Appl. Phys. 1977, 48 (8), 3524–3531.
22. Giusti, G. Deposition and Characterisation of Functional ITO Thin Films, University of Birmingham, 2011.
23. Ma, Z.; Li, Z.; Liu, K.; Ye, C.; Sorger, V. J. Indium-Tin-Oxide for High-Performance Electro-Optic Modulation.
Nanophotonics 2015, 4 (1), 198–213.
24. Kasap, S.; Koughia, C.; Ruda, H. E. Springer Handbook of Electronic and PhotonicMaterials; Capper, P., Safa,
K., Eds.; 2017.
25. Bellingham, J. R.; Phillips, W. A.; Adkins, C. J. Electrical and Optical Properties of Amorphous Indium Oxide.
J. Phys. Condens. Matter 1990, 2 (28), 6207–6221.

135
26. Dixit, R.; Gupta, S.; Kumar, P.; Sikarwar, S.; Yadav, B. C. Preparation and Properties of Transparent
Conducting Oxide ( TCOs ) Thin Films : A Review. Int. J. Innov. Res.Sci. 2017, 6 (2)
27. Kittel, C.; McEuen, P.; Sons, J. W. &. Introduction to Solid State Physics; John Wiley & Sons, 2019.
28. Hass, G.; Hunter, W. R.; Tousey, R. Influence of Purity, Substrate Temperature, and Aging Conditions on the
Extreme Ultraviolet Reflectance of Evaporated Aluminum. J. Opt. Soc. Am. 1957, 47 (12), 1070–1073.
29. Dixon, S. C.; Scanlon, D. O.; Carmalt, C. J.; Parkin, I. P. N-Type Doped Transparent Conducting Binary Oxides:
An Overview. J. Mater. Chem. C 2016, 4 (29), 6946–6961.
30. Kahn, A. Fermi level, work function, and vacuum level. Mater. Horiz. 2016, 3, 7–10
31. Braunstein, G.; Muraviev, A.; Saxena, H.; Dhere, N.; Richter, V.; Kalish, R. P Type Doping of Zinc Oxide by
Arsenic Ion Implantation. Appl. Phys. Lett. 87 (19), 192103–1921033.
32. Yoshimura, R. Proceedings of the 35th Annual Technical Conference of the Society of Vacuum Coaters. 1992,
362.
33. Zeigler, Y.; Fischer, D. 49th Annual Technical Conference Proceedings. In Roll-to-roll Production of Thin Film
Silicon Solar Cells on PET Film Substrate: A Smart Energy Option of the Future; Society of Vacuum Coaters:
Washington, D.C., 2006; pp 609–613.
34. Menezes, C. Fourth E.C. Photovoltaic Solar Energy Conference : Proceedings of the International
Conference; D. Reidel Pub. Co. ; Distributed in the U.S. by Kluwer Boston: Stresa, 1982.
35. Mollwo, E. The effect of hydrogen on the conductivity and luminescence of zinc oxide crystals. Z.
Phys. 1954, 138, 478–488.
36. Thomas, D.G.; Lander, J.J. Hydrogen as a Donor in Zinc Oxide. J. Chem. Phys. 1956, 25, 1136–1142.
37. Van de Walle, C.G. Hydrogen as a Cause of Doping in Zinc Oxide. Phys. Rev. Lett. 2000, 85, 1012–1015.
38. Kohan, F.; Ceder, G.; Morgan, D.; Van de Walle, C.G. First-principles study of native point defects in
ZnO. Phys. Rev. B 2000, 61, 15019. [Google Scholar] [CrossRef][Green Version]
39. Hofmann, D.M.; Hofstaetter, A.; Leiter, F.; Zhou, H.; Henecker, F.; Meyer, B.K.; Orlinskii, S.B.; Schmidt, J.;
Baranov, P.G. Hydrogen: A Relevant Shallow Donor in Zinc Oxide. Phys. Rev. Lett. 2002, 88, 045504.
40. Cox, S.F.J.; Davis, E.A.; Cottrell, S.P.; King, P.J.C.; Lord, J.S.; Gil, J.M.; Alberto, H.V.; Vilão, R.C.; Duarte, J.P.;
De Campos, N.A.; et al. Experimental Confirmation of the Predicted Shallow Donor Hydrogen State in Zinc
Oxide. Phys. Rev. Lett. 2001, 86, 2601–2604.
41. Van de Walle, C.G. Hydrogen as a shallow centre in semiconductors and oxides. Phys. Status Solidi
b 2003, 235, 89–95.
42. Hosono, H.; Ueda, K. Transparent Conductive Oxides. In Springer Handbook of Electronic and Photonic
Materials; Springer: Berlin, Germany, 2017.
43. Yan, X.; Tian, L.; Tan, X.; Zhou, M.; Liu, L.; Chen, X. Modifying oxide nanomaterials’ properties by
hydrogenation. MRS Commun. 2016, 6, 192–203.
44. Li, T.; Ong, C.S.; Herng, T.S.; Yi, J.; Bao, N.; Xue, J.M.; Feng, Y.P.; Ding, J. Surface ferromagnetism in
hydrogenated-ZnO film. Appl. Phys. Lett. 2011, 98.
45. Park, J.K.; Kwon, H.-J.; Lee, C.E. NMR Observation of Mobile Protons in Proton-Implanted ZnO Nanorods. Sci.
Rep. 2016, 6, 23378.
46. Robaina, O.V.; Cabrera, A.F.; Meyer, M.; Romano, R.M.; Cruz, A.F.; Torres, C.E.R. Room-Temperature
Ferromagnetism Induced by High-Pressure Hydrogenation of ZnO. J. Phys. Chem. C 2019, 123, 19851–
19861.
47. Esquinazi, P.; Hergert, W.; Spemann, D.; Setzer, A.; Ernst, A. Defect-Induced Magnetism in Solids. IEEE Trans.
Magn. 2013, 49, 4668–4674.
48. Salimian, A.; Hasnath, A.; Anguilano, L.; Onwukwe, U.; Aminishahsavarani, A.; Sachez, C.; Upadhyaya, H.
Highly Conductive Zinc Oxide Based Transparent Conductive Oxide Films Prepared Using RF Plasma
Sputtering Under Reducing Atmosphere. Coatings 2020, 10, 472.
49. Song, S.; Yang, T.; Liu, J.; Xin, Y.; Li, Y.; Han, S. Rapid Thermal Annealing of ITO Films. Appl. Surf. Sci. 2011,
257 (16), 7061–7064.
50. Vatakketath, R. K. Investigation on the Transparent Conducting Oxide (TCO) Material Used in CIGS Thin Film
Solar Cell in Midsummer AB, Uppsala University, 2020.
51. Shigesato, Y.; Takaki, S.; Haranoh, T. Electrical and Structural Properties of Low Resistivity Tin-Doped Indium
Oxide Films. J. Appl. Phys. 1992, 71 (7), 3356–3364.

136
52. Suzuki, A.; Matsushita, T.; Aoki, T.; Yoneyama, Y.; Okuda, M. Pulsed Laser Deposition of Transparent
Conducting Indium Tin Oxide Films in Magnetic Field Perpendicular to Plume. Jpn. J. Appl. Phys. 2001, 40
(Part 2, No. 4B), L401–L403.
53. Gondoni, P. Nanostructured Transparent Conductive Oxides for Photovoltaic Applications, Politecnico di
Milano, 2014.
54. Langmuir, I. Oscillations in Ionized Gases. Proc. Natl. Acad. Sci. 1928, 14, 627–637.
55. Roth, J. R. Industrial Plasma Engineering Volume 1: Principles; Institute of Physics Publishing: Knoxville,
1995; Vol. 1.
56. Budtz-jørgensen, C. V. Studies of Electrical Plasma Discharges, Aarhus University, 2001.
57. Llewellyn-Jones, F. The Glow Discharge and an Introduction to Plasma Physics. Am. J. Phys. 1967, 35 (5),
453.
58. https://www.plasma-universe.com/electric-glow-discharge/
59. Chen, F. F. Erratum to: Introduction to Plasma Physics and Controlled Fusion; 2018.
60. Jeon, M.; Kamisako, K. Synthesis and Characterization of Silicon Nanowires Using Tin Catalyst for Solar Cells
Application. Mater. Lett. 2009, 63 (9), 777–779. ??
61. Depla, D.; Mahieu, S.; Greene, J. E. Sputter Deposition Processes. In Handbook of deposition technologies
for films and coatings : science, applications and technology; Martin, P. M., Ed.; Elsevier: Oxford, 2010; pp
253–296.
62. Lieberman, M. A.; Lichtenberg, A. J. Principles of Plasma Discharges and Materials Processing, 2nd Ed.;
Wiley, 1994.
63. http://www.glow-discharge.com/rf_vs_dc.htm 91
64. Chapman, B. Glow Discharge Processes: Sputtering and Plasma Etching; Wiley, 1980.
65. Waite, M. M.; Chester, W.; Glocker, D. a; Incorporated, I.; York, N. Sputtering Sources. Soc. Vac. Coaters
2010, 42–50.
66. [(Greene, J. E. Review Article: Tracing the Recorded History of Thin-Film Sputter Deposition: From the 1800s
to 2017. J. Vac. Sci. Technol. A Vacuum, Surfaces, Film. 2017, 35 (5), 05C204.
67. Parsons, R. II-4 - Sputter Deposition Processes; Vossen, J. L., Kern, W. B. T.-T. F. P., Eds.; Academic Press:
San Diego, 1991; pp 177–208.
68. Glocker, D. A.; Shah, S. I. Handbook of Thin Film Process Technology; Handbook of Thin Film Process
Technology; Institute of Physics Pub., 2002.
69. Maurya, D.; Sardarinejad, A.; Alameh, K. Recent Developments in R.F. Magnetron Sputtered Thin Films for
PH Sensing Applications—An Overview. Coatings 2014, 4 (4), 756–771.
70. Svadkovski, I. V.; Golosov, D. A.; Zavatskiy, S. M. Characterisation Parameters for Unbalanced Magnetron
Sputtering Systems. Vacuum 2002, 68 (4), 283–290.
71. Petrov, I.; Adibi, F.; Greene, J. E.; Sproul, W. D.; Münz, W. -D. Use of an Externally Applied Axial Magnetic
Field to Control Ion/Neutral Flux Ratios Incident at the Substrate during Magnetron Sputter Deposition. J.
Vac. Sci. Technol. A 1992, 10 (5), 3283–3287.
72. Window, B.; Savvides, N. Unbalanced Dc Magnetrons as Sources of High Ion Fluxes. J. Vac. Sci. Technol. A
1986, 4 (3), 453–456.
73. Kelly, P. J.; Arnell, R. D. Magnetron Sputtering: A Review of Recent Developments and Applications. Vacuum
2000, 56 (3), 159–172.
74. Tchenka, A.; Agdad, A.; Vall, M.C.S.; Hnawi, S.K.; Narjis, A.; Nkhaili, L.; Ibnouelghazi, E.; Ech-Chamikh, E.
Effect of RF Sputtering Power and Deposition Time on Optical and Electrical Properties of Indium Tin Oxide
Thin Film. Adv. Mater. Sci. Eng. 2021, 2021, e5556305.
75. Abdulameer, M. Simulation of plasma properties in magnetron sputtering with Kr gas. Int. J. Appl. Or Innov.
Eng. Manag. 2013, 2, 290–296.
76. Powell, R.A.; Rossnagel, S.M. (Eds.) Chapter 2 Physics of sputtering. In Thin Films; Elsevier: Amsterdam, The
Netherlands, 1999; Volume 26, pp. 23–49
77. [Oechsner, H. Sputtering—A review of some recent experimental and theoretical aspects. Appl. Phys. 1975,
8, 185–198.
78. Fridman, A. Plasma Chemistry; Cambridge University Press: Cambridge, UK, 2008.
79. ‘The Technology - Plasma Quest’. https://www.plasmaquest.co.uk/the-technology/ (accessed Apr. 06,
2022).

137
80. Kenmotsu, T.; Wada, M.; Hyakutake, T.; Muramoto, T.; Nishida, M. Enhanced sputtering yields of carbon
due to accumulation of low-energy Xe ions. Nucl. Instrum. Methods Phys. Res. Sect. B Beam Interact. Mater.
At. 2009, 267, 1717–1720.
81. Yagisawa, T.; Makabe, T. Modeling of dc magnetron plasma for sputtering: Transport of sputtered copper
atoms. J. Vac. Sci. Technol. Int. J. Devoted Vac. Surf. Film 2006, 24, 908.
82. Wehner, G. Influence of the Angle of Incidence on Sputtering Yields. J. Appl. Phys. 1959, 30, 1762–1765.
83. van Hattum, E.D.; Palmero, A.; Arnoldbik, W.M.; Rudolph, H.; Habraken, F.H.P.M. On the ion and neutral
atom bombardment of the growth surface in magnetron plasma sputter deposition. Appl. Phys. Lett. 2007,
91, 171501.
84. Stamate, E. Spatially Resolved Optoelectronic Properties of Al-Doped Zinc Oxide Thin Films Deposited by
Radio-Frequency Magnetron Plasma Sputtering Without Substrate Heating. Nanomaterials 2019, 10, 14.
85. Bairagi, S.; Järrendahl, K.; Eriksson, F.; Hultman, L.; Birch, J.; Hsiao, C.-L. Glancing Angle Deposition and
Growth Mechanism of Inclined AlN Nanostructures Using Reactive Magnetron Sputtering. Coatings 2020,
10, 768.
86. Güttler, D.; Abendroth, B.; Grötzschel, R.; Möller, W.; Depla, D. Mechanisms of target poisoning during
magnetron sputtering as investigated by real-time in situ analysis and collisional computer simulation. Appl.
Phys. Lett. 2004, 85, 6134–6136.
87. Tian-Ye, N.; Jin-Xiang, C.; Lei, L.; Jin-Ying, L.; Yan, W.; Liang, W.; You, L. A comparison among optical emission
spectroscopic methods of determining electron temperature in low pressure argon plasmas. Chin. Phys.
2007, 16, 2757–2763.
88. Lewis, B.G.; Paine, D.C. Applications and processing of transparent conducting oxides. MRS Bull. 2000, 25,
22–27.
89. Crintea, D.L.; Czarnetzki, U.; Iordanova, S.; Koleva, I.; Luggenhölscher, D. Plasma diagnostics by optical
emission spectroscopy on argon and comparison with Thomson scattering. J. Phys. D 2009, 42, 045208.
90. Zhu, X.; Pu, Y. Optical emission spectroscopy in low-temperature plasmas containing argon and nitrogen:
Determination of the electron temperature and density by the line-ratio method. J. Phys. D 2010, 43,
403001.
91. Trevizan, L.C.; Santos, D., Jr.; Samad, R.E.; Vieira, N.D., Jr.; Nunes, L.C.; Rufini, I.A.; Krug, F.J. Evaluation of
laser induced breakdown spectroscopy for the determination of micronutrients in plant materials.
Spectrochim. Acta Part B Atomic Spectrosc. 2009, 64, 369–377.
92. Unnikrishnan, V.K.; Alti, K.; Nayak, R.; Bernard, R.; Khetarpal, N.; Kartha, V.B.; Santhosh, C.; Gupta, G.P.;
Suri, B.M. Optimized LIBS setup with echelle spectrograph-ICCD system for multi-elemental analysis. J.
Instrum. 2010, 5, P04005.
93. Bastiaans, G.J.; Mangold, R.A. The calculation of electron density and temperature in Ar spectroscopic
plasmas from continuum and line spectra. Spectrochim. Acta Part B Atomic Spectrosc. 1985, 40, 885–892.
94. Iordanova, S.; Koleva, I. Optical emission spectroscopy diagnostics of inductively-driven plasmas in argon
gas at low pressures. Spectrochim. Acta Part B Atomic Spectrosc. 2007, 62, 344–356.
95. Gurnett, D.A.; Bhattacharjee, A. Introduction to Plasma Physics: With Space and Laboratory Applications;
Cambridge University press: Cambridge, UK, 2005.
96. Bellan, P.M. Fundamentals of Plasma Physics; Cambridge University Press: Cambridge, UK, 2008.
97. Unnikrishnan, V.K.; Alti, K.; Kartha, V.B.; Santhosh, C.; Gupta, G.P.; Suri, B.M. Measurements of plasma
temperature and electron density in laser-induced copper plasma by time-resolved spectroscopy of neutral
atom and ion emissions. Pramana 2010, 74, 983–993.
98. McWhirter, R.W.P. Plasma Diagnostic Techniques; Huddlestone, R.H., Leonard, S.L., Eds.; Academic Press:
Academic Press, New York, NY, USA, 1965.
99. Diwakar, P.; Hahn, D. Study of early laser-induced plasma dynamics: Transient electron density gradients
via Thomson scattering and Stark Broadening, and the implications on laser-induced breakdown
spectroscopy measurements. Spectrochim. Acta Part B Atomic Spectrosc. 2008, 63, 1038–1046.
100.Shaikh, N.M.; Rashid, B.; Hafeez, S.; Jamil, Y.; Baig, M.A. Measurement of electron density and temperature
of a laser-induced zinc plasma. J. Phys. D 2006, 39, 1384.

138
101.Tawfik, W.; Mohamed, Y.; Askar, A. Study of the matrix effect on the plasma characterization of heavy
elements in soil sediments using LIBS with a portable Echelle spectrometer. Prog. Phys. 2007, 1, 46–52.
102.Hong, Y.J.; Kwon, G.C.; Cho, G.; Shin, H.M.; Choi, E.H. Measurement of electron temperature and density
using stark broadening of the coaxial focused plasma for extreme ultraviolet lithography. IEEE Trans. Plasma
Sci. 2010, 38, 1111–1117.
103.Diao, C.Y.; Chen, C.S.; Man, B.Y.; Wang, C.; Fu, H.B. Influence of distances between the lens and the target
on the characteristic of laser induced lead plasma. Eur. Phys. J. D 2011, 63, 123–128.
104.Konjević, N.; Lesage, A.; Fuhr, J.R.; Wiese, W.L. Experimental Stark widths and shifts for spectral lines of
neutral and ionized atoms (a critical review of selected data for the period 1989 through 2000). J. Phys.
Chem. Ref. Data 2002, 31, 819–927.
105.Wyszecki, G.; Stiles, W.S. Color Science Concepts and Methods, Quantative Data and Formulae, 2nd Ed.;
Wiley: New York, NY, USA, 1982.
106.Conrads, H.; Schmidt, M. Plasma generation and plasma sources. Plasma Sources Sci. Technol. 2000, 9, 441.
107.Alder, J.; Bombelka, R.; Kirkbright, G. Electronic excitation and ionization temperature measurements in a
high frequency inductively coupled argon plasma source and the influence of water vapor on plasma
parameters. Spectrochim. Acta Part B Atomic Spectrosc. 1980, 35, 163–175.
108.Bogaerts, A.; Neyts, E.; Gijbels, R.; Van der Mullen, J. Gas discharge plasmas and their applications.
Spectrochim. Acta Part B Atomic Spectrosc. 2002, 57, 609–658.
109.Wagatsuma, K.; Hirokawa, K. Characterization of atomic emission lines from argon, neon, and nitrogen glow
discharge plasmas. Anal. Chem. 1985, 57, 2901–2907.
110.Angeli, J.; Bengtson, A.; Bogaerts, A.; Hoffmann, V.; Hodoroaba, V.D.; Steers, E. Glow discharge optical
emission spectrometry: Moving towards reliable thin film analysis—A short review. J. Anal. Atomic
Spectrom. 2003, 18, 670–679.
111.Ablikim, M.; An, Z.H.; Bai, J.Z.; Berger, N.; Bian, J.M.; Cai, X.; Cao, G.F.; Cao, X.X.; Chang, J.F.; Chen, C.; et al.
Design and construction of the BESIII detector. Nucl. Inst. Methods Phys. Res. 2010, 614, 345–399.
112.Wang, F.H.; Yang, C.F.; Liou, J.C.; Chen, I.C. Effects of hydrogen on the optical and electrical characteristics
of the sputter-deposited Al2O3-doped zno thin films. J. Nanomater. 2014, 2014, 857641.
113.Palmero, A.; van Hattum, E.; Arnoldbik, W.; Habraken, F. Argon plasma modelling in a RF magnetron
sputtering system. Surf. Coat. Technol. 2004, 188–189, 392–398.
114.Christie, D.J. Target material pathways model for high power pulsed magnetron sputtering. J. Vac. Sci.
Technol. A 2005, 23, 330–335.
115.Kolev, I. Particle-In-Cell-Monte-Carlo Collisions Simulations for a Direct Current Planar Magnetron
Discharge. Ph.D. Thesis, University of Antwerp, Antwerp, Belgium, 2007.
116.Brenning, N.; Huo, C.; Lundin, D.; Raadu, M.; Vitelaru, C.; Stancu, G.; Minea, T.; Helmersson, U.
Understanding deposition rate loss in high power impulse magnetron sputtering: I. Ionization-driven
electric fields. Plasma Sources Sci. Technol. 2012, 21, 025005.
117.Huo, C.; Raadu, M.; Lundin, D.; Gudmundsson, J.; Anders, A.; Brenning, N. Gas rarefaction and the time
evolution of long high-power impulse magnetron sputtering pulses. Plasma Sources Sci. Technol. 2012, 21,
045004.
118.Shidoji, E.; Ohtake, H.; Nakano, N.; Makabe, T. Two-Dimensional Self-Consistent Simulation of a DC
Magnetron Discharge. Jpn. J. Appl. Phys. 1999, 38, 2131–2136.
119.Kolev, I.; Bogaerts, A. Numerical Models of the Planar Magnetron Glow Discharges. Contrib. Plasma
Phys. 2004, 44, 582–588.
120.Sheridan, T.; Goeckner, M.; Goree, J. Pressure dependence of ionization efficiency in sputtering
magnetrons. Appl. Phys. Lett. 1990, 57, 2080–2082.
121.Goeckner, M.; Goree, J.; Sheridan, T. Monte Carlo simulation of ions in a magnetron plasma. IEEE Trans.
Plasma Sci. 1991, 19, 301–308.

139
122.Passoth, E.; Behnke, J.; Csambal, C.; Tichý, M.; Kudrna, P.; Golubovskii, Y.; Porokhova, I. Radial behaviour of
the electron energy distribution function in the cylindrical magnetron discharge in argon. J. Phys. D Appl.
Phys. 1999, 32, 2655–2665.
123.Porokhova, I.; Golubovskii, Y.; Bretagne, J.; Tichy, M.; Behnke, J. Kinetic simulation model of magnetron
discharges. Phys. Rev. E 2001, 63, 056408.
124.Porokhova, I.; Golubovskii, Y.; Behnke, J. Anisotropy of the electron component in a cylindrical magnetron
discharge. I. Theory of the multiterm analysis. Phys. Rev. E 2005, 71, 066406.
125.Porokhova, I.; Golubovskii, Y.; Behnke, J. Anisotropy of the electron component in a cylindrical magnetron
discharge. II. Application to real magnetron discharge. Phys. Rev. E 2005, 71, 066407.
126.Birdsall, C. Particle-in-cell charged-particle simulations, plus Monte Carlo collisions with neutral atoms, PIC-
MCC. IEEE Trans. Plasma Sci. 1991, 19, 65–85.
127.Revel, A.; Minea, T.; Costin, C. 2D PIC-MCC simulations of magnetron plasma in HiPIMS regime with external
circuit. Plasma Sources Sci. Technol. 2018, 27, 105009.
128.Tozer, B. Rotating plasma. Proc. Inst. Electr. Eng. 1965, 112, 218.
129.Wasa, K.; Hayakawa, S. Formation of Rotating Plasma in Crossed Field. J. Phys. Soc. Jpn. 1966, 21, 738–743.
130.Wilcox, J.; Cooper, W.; DeSilva, A.; Spillman, G.; Boley, F. Swirls Produced in a “Crowbarred” Rotating
Plasma. J. Appl. Phys. 1962, 33, 2714–2715.
131.Anders, A. Localized heating of electrons in ionization zones: Going beyond the Penning-Thornton paradigm
in magnetron sputtering. Appl. Phys. Lett. 2014, 105, 244104.
132.Panjan, M.; Loquai, S.; Klemberg-Sapieha, J.; Martinu, L. Non-uniform plasma distribution in dc magnetron
sputtering: Origin, shape and structuring of spokes. Plasma Sources Sci. Technol. 2015, 24, 065010.
133.Anders, A.; Yang, Y. Direct observation of spoke evolution in magnetron sputtering. Appl. Phys.
Lett. 2017, 111, 064103.
134.Anders, A.; Yang, Y. Plasma studies of a linear magnetron operating in the range from DC to HiPIMS. J. Appl.
Phys. 2018, 123, 043302.
135.Andersson, J.; Ni, P.; Anders, A. Smoothing of Discharge Inhomogeneities at High Currents in Gasless High
Power Impulse Magnetron Sputtering. IEEE Trans. Plasma Sci. 2014, 42, 2856–2857.
136.Arcos, T.; Layes, V.; Gonzalvo, Y.; Gathen, V.; Hecimovic, A.; Winter, J. Current–voltage characteristics and
fast imaging of HPPMS plasmas: Transition from self-organized to homogeneous plasma regimes. J. Phys. D
Appl. Phys. 2013, 46, 335201.
137.Šlapanská, M.; Hecimovic, A.; Gudmundsson, J.; Hnilica, J.; Breilmann, W.; Vašina, P.; von Keudell, A. Study
of the transition from self-organised to homogeneous plasma distribution in chromium HiPIMS discharge. J.
Phys. D Appl. Phys. 2020, 53, 155201.
138.Hecimovic, A.; Corbella, C.; Maszl, C.; Breilmann, W.; von Keudell, A. Investigation of plasma spokes in
reactive high power impulse magnetron sputtering discharge. J. Appl. Phys. 2017, 121, 171915.
139.Depla, D.; Mahieu, S. Reactive Sputter Deposition; Springer Series in Materials Science; Springer:
Berlin/Heidelberg, Germany, 2008; Volume 109.
140.Marcak, A.; Corbella, C.; de los Arcos, T.; von Keudell, A. Note: Ion-induced secondary electron emission
from oxidized metal surfaces measured in a particle beam reactor. Rev. Sci. Instrum. 2015, 86, 106102.
141.Yang, Y.; Zhou, X.; Liu, J.; Anders, A. Evidence for breathing modes in direct current, pulsed, and high power
impulse magnetron sputtering plasmas. Appl. Phys. Lett. 2016, 108, 034101.
142.Yang, Y.; Tanaka, K.; Liu, J.; Anders, A. Ion energies in high power impulse magnetron sputtering with and
without localized ionization zones. Appl. Phys. Lett. 2015, 106, 124102.
143.Anders, A.; Ni, P.; Andersson, J. Drifting Ionization Zone in DC Magnetron Sputtering Discharges at Very Low
Currents. IEEE Trans. Plasma Sci. 2014, 42, 2578–2579.
144.Held, J.; Maaß, P.; Gathen, V.; Keudell, A. Electron density, temperature and the potential structure of
spokes in HiPIMS. Plasma Sources Sci. Technol. 2020, 29, 025006.
145.Schmidhuber, J. Deep learning in neural networks: An overview. Neural Netw. 2015, 61, 85–117.

140
146.Hinton, G.; Rumelhart, D. Neural Network Architectures for Artificial Intelligence; American Society for
Artificial Intelligence: San Mateo, CA, USA, 1988.
147.Introduction to Convolutional Neural Networks. 2017. Available
online: https://cs.nju.edu.cn/wujx/paper/CNN.pdf
148.Lecun, Y. Gradient Based Learning for Document Recognition. 1998. Available
online: http://yann.lecun.com/exdb/publis/pdf/lecun-01a.pdf
149.He, K.; Zhang, X.; Ren, S.; Sun, J. Spatial Pyramid Pooling in Deep Convolutional Networks for Visual
Recognition. IEEE Trans. Pattern Anal. Mach. Intell. 2015, 37, 1904–1916.
150.Suárez-Paniagua, V.; Segura-Bedmar, I. Evaluation of pooling operations in convolutional architectures for
drug-drug interaction extraction. BMC Bioinform. 2018, 19, 39–47.
151.Cottrell, M.; Olteanu, M.; Rossi1, F.; Villa-Vialaneix, N. Self-organizing maps, theory and applications. Rev.
Investig. Oper. 2018, 39, 1–22.
152.Fridman. Measurements with the Model 4200-SCS. Appl. Note 2007, No. 2475, 2005 2007.
153.Valdes, L. B. Resistivity Measurements on Germanium for Transistors. Proc. IRE 1954, 42 (2), 420–427
154.Illinois, E. GT2 - Four Point Probe Correction Factor a (thickness),” Theory and Fabrication of Integrated
Circuits. http://fabweb.ece.illinois.edu/gt/gt/gt2.aspx (accessed 2022-08-19).
155.Gutiérrez, M. P.; Li, H.; Patton, J. Thin Film Surface Resistivity. 2002, 0–24.
156.Vo,K.Spectrophotometry.
https://chem.libretexts.org/Core/Physical_and_Theoretical_Chemistry/Kinetics/Reaction
_Rates/Experimental_Determination_of_Kinetcs/Spectrophotometry (accessed 2022-08-19).
157.Multispectral Colorimetric system (diagram)
158.Tkachenko, N. V. Chapter 3 - Lasers for Spectroscopy Applications; TKACHENKO, N. V. B. T.-O. S., Ed.; Elsevier
Science: Amsterdam, 2006; pp 39–59.
159.Instruments, N. How A Profilometer Works. http://www.nanoscience.com/technology/optical-profiler-
technology/how-profilometer- works (accessed 2017-05-01).
160.Reimer, L. Introduction. In Scanning Electron Microscopy; Springer Berlin Heidelberg: Berlin, 1998; pp 1–12.
161.Sandeman, K.; Coombe, P. Raman Spectroscopy and Scanning Electron Microscopy.
https://doitpoms.admin.cam.ac.uk/ (accessed 2017-05-05).
162.Everhart, T. E.; Thornley, R. F. M. Wide-Band Detector for Micro-Microampere Low-Energy Electron
Currents. J. Sci. Instrum. 1960, 37 (7), 246–248.
163.Echlin, P. Handbook of Sample Preparation for Scanning Electron Microscopy and X-Ray Microanalysis, 1st
Ed.; Springer New York, N, 2009.
164.https://en.wikipedia.org/wiki/Energy-dispersive_X-ray_spectroscopy#/media/File:EDX-scheme.svg
165.Qayyum, A.; Ikram, M.; Zakaullah, M.; Waheed, A.; Murtaza, G.; Ahmad, R.; Majeed, A.; Khattak, N.A.D.;
Mansoor, K.; Chaudhary, K.A. Characterization of argon plasma by use of optical emission spectroscopy and
Langmuir probe measurements. Int. J. Mod. Phys. B 2003, 17, 2749–2759.
166.Kwak, D.; Park, M.; Sung, Y. Discharge power dependence of structural and electrical properties of Al-doped
ZnO conducting film by magnetron sputtering (for PDP). Vacuum 2008, 83, 113–118.
167.Yan, L.; Rath, J.; Schropp, R. Electrical properties of vacuum-annealed titanium-doped indium oxide films.
Appl. Surf. Sci. 2011, 257, 9461–9465.
168.Sultan, M.S. Effect of gas pressure and flow rate on the plasma power and deposition rate in magnetron
sputtering system. Res. J. Nanosci. Eng. 2018, 2, 1–8.
169.Salimian, A.; Haghpanahan, R.; Hasnath, A.; Upadhyaya, H. Optical Analysis of RF Sputtering Plasma through
Colour Characterization. Coatings 2019, 9, 315
170.Sheng, J.; Lee, H.-J.; Oh, S.; and Park, J.S. Flexible and high-performance amorphous indium Zinc oxide
thin-film transistor using low-temperature atomic layer deposition. ACS Appl. Mater. Interfaces 2016,
8, 33821–33828
171.Hautier, G.; Miglio, A.; Waroquiers, D.; Rignanese, G.-M.; Gonze, X. How does chemistry influence
electron effective mass in oxides? A high-throughput computational analysis. Chem. Mater. 2014, 26,
5447–5458.

141
172.Wallinga, J.; Arnold Bik, W.M.; Vredenberg, A.M.; Schropp, R.E.I.; van der Weg, W.F. Reduction of Tin
Oxide by Hydrogen Radicals. J. Phys. Chem. B 1998, 102, 6219.
173.De Wit, J.H.W. Electrical properties of In2O3. J. Solid State Chem. 1973, 8, 2, 142.
174.Dewit, J.H.W.; Vanunen, G.; Lahey, M. Electron concentration and mobility in In2O3. J. Phys. Chem.
Solids 1997, 38, 819–824.
175.Luo, S.; Kohiki, S.; Okada, K.; Kohno, A.; Tajiri, T.; Arai, M.; Shoji, F. Effects of Hydrogen in Working
Gas on Valence States of Oxygen in Sputter-Deposited Indium Tin Oxide Thin Films. Acs Appl. Mater.
Interfaces 2010, 2, 663–668.
176.Dixon, S.C.; Scanlon, D.O.; Carmalt, C.J.; Parkin, I.P. n-Typed doped transparent conductive binary
oxides: An overview. J. Mater. Chem. C 2016, 4, 6946–6961.
177.Van de Walle, C.G. Hydrogen as a Cause of Doping in Zinc Oxide. Phys. Rev. Lett. 2000, 85, 1012–1015.
178.Hofmann, D.M.; Hofstaetter, A.; Leiter, F.; Zhou, H.; Henecker, F.; Meyer, B.K.; Baranov, P.G.
Hydrogen: A Relevant Shallow Donor in Zinc Oxide. Physical Review Letters, 2002, 88, 045504.
179.Cox, S.F.J.; Davis, E.A.; Cottrell, S.P.; King, P.J.C.; Lord, J.S.; Gil, J.M.; Alberto, H.V.; Vilo, R.C.; Piroto
Duarte, J.; Ayres de Campos, N.; et al. Experimental Confirmation of the Predicted Shallow Donor
Hydrogen State in Zinc Oxide. Phys. Rev. Lett. 2001, 86, 2601–2604.
180.Van de Walle, C.G. Hydrogen as a shallow center in semiconductors and oxides. Phys. Status Solidi
2003, 235, 89–95.
181.Knewstubb, P.F.; Tickner, A.W. Mass spectrometry of ions in glow discharges. I. apparatus and its
application to the positive column in rare gases. J. Chem. Phys. 1962, 36, 674.
182.Gordon, M.H.; Kruger, C.H. Non-equilibrium effects of diluent addition in a recombining argon
plasma. Phys. Fluids B: Plasma Phys. 1993, 5, 1014.
183.Meulenbroeks, R.F.G.; van Beek, A.J.; van Helvoort, A.J.G.; van de Sanden, M.C.M.; Schram, D.C.
Argon-hydrogen plasma jet investigated by active and passive spectroscopic means. Phys. Rev. E
1994, 49, 4397– 4406.
184.Mason, R.S.; Miller, P.D.; Mortimer, I.P. Anomalous loss of ionization in argon-hydrogen plasma
studied by fast flow glow discharge mass spectrometry. Phys. Rev. E 1997, 55, 7462–7472.
185.Tabares, F.L.; Tafalla, D. Sputtering of metallic walls in Ar/H2 direct current glow discharges at room
temperature. J. Vac. Sci. Technol. A 1996, 14, 3087–3091.
186.Budtz-Jorgensen, C.V.; Kringhoj, P.; Bottiger, J. The critical role of hydrogen for physical sputtering
with Ar–H2 glow discharges. Surf. Coat. Technol. 1999, 116, 938–943.
187.Smithwick, R.W., III; Lynch, D.W.; Franklin, J.C. Relative ion yields measured with a high-resolution
glow discharge mass spectrometer operated with an argon/hydrogen mixture. J. Am. Soc. Mass
Spectrom. 1993, 4, 278–285.
188.Saito, M. The relationship between relative sensitivity factors and ionization potential in dc glow
discharge mass spectrometry using Ar/0.2 vol.% H2 mixture. Anal. Chim. Acta. 1997, 355, 129–134.
189.Saikia, P.; Saikia, B.K.; Bhuyan, H. Study on the effect of hydrogen addition on the variation of plasma
parameters of argon-oxygen magnetron glow discharge for synthesis of TiO2 films. Aip Adv. 2016, 6,
045206(1–9).
190.Kakati, H.; Pal, A.R.; Bailung, H.; Chutiya, J. Effect of oxygen on the characteristics of radio frequency
planar magnetron sputtering plasma used for aluminum oxide depositionJ. Appl. Phys. 2007, 101,
083304(1– 7).
191.Coburn, J.W.; Chen, M. Optical emission spectroscopy of reactive plasmas: A method for correlating
emission intensities to reactive particle density. J. Appl. Phys. 1980, 51, 3134–3136.
192.Lavrov, B.P.; Pipa, A.V.; Röpcke, J. On determination of the degree of dissociation of hydrogen in non-
equilibrium plasmas by means of emission spectroscopy: I. The collision-radiative model and
numerical experiments. Plasma Sources Sci. Technol. 2006, 15, 135–146.
193.Clay, K.J.; Speakman, S.P.; Amaratunga, G.A.J.; Silva, S.R.P. Characterization of a-C: H: N deposition
from CH4/N2 rf plasmas u
194.Salimian, A.; Hasnath, A.; Anguilano, L.; Onwukwe, U.; Aminishahsavarani, A.; Sachez, C.; Upadhyaya, H.
Highly Conductive Zinc Oxide Based Transparent Conductive Oxide Films Prepared Using RF Plasma
Sputtering Under Reducing Atmosphere. Coatings 2020, 10, 472.

142
195.Hautier, G.; Miglio, A.; Waroquiers, D.; Rignanese, G.-M.; Gonze, X. How Does Chemistry Influence Electron
Effective Mass in Oxides? A High-Throughput Computational Analysis. Chem. Mater. 2014, 26, 5447–5458.
196.Wallinga, J.; Arnoldbik, W.M.; Vredenberg, A.M.; Schropp, R.E.I.; Van Der Weg, W.F. Reduction of Tin Oxide
by Hydrogen Radicals. J. Phys. Chem. B 1998, 102, 6219–6224.
197.De Wit, J. Electrical properties of In2O3. J. Solid State Chem. 1973, 8, 142–149.
198.Dewit, J.H.W.; Vanunen, G.; Lahey, M. Electron concentration and mobility in In2O3. J. Phys. Chem.
Solids 1997, 38, 819–824.
199.Gordon, M.H.; Kruger, C.H. Non-equilibrium effects of diluent addition in recombining argon plasma. Phys.
Fluids B Plasma Phys. 1993, 5, 1014.
200.Meulenbroeks, R.F.G.; Van Beek, A.J.; Van Helvoort, A.J.G.; Van De Sanden, M.C.M.; Schram, D.C. Argon-
hydrogen plasma jet investigated by active and passive spectroscopic means. Phys. Rev. E 1994, 49, 4397–
4406.
201.Mason, R.S.; Miller, P.D.; Mortimer, I.P. Anomalous loss of ionization in argon-hydrogen plasma studied by
fast flow glow discharge mass spectrometry. Phys. Rev. E 1997, 55, 7462–7472.
202.Tabares, F.L.; Tafalla, D. Sputtering of metallic walls in Ar/H2 direct current glow discharges at room
temperature. J. Vac. Sci. Technol. A 1996, 14, 3087–3091.
203.Budtz-Jorgensen, C.V.; Kringhoj, P.; Bottiger, J. The critical role of hydrogen for physical sputtering with Ar–
H2 glow dis-charges. Surf. Coat. Technol. 1999, 116, 938–943.
204.Smithwick, R.W., III; Lynch, D.W.; Franklin, J.C. Relative ion yields measured with a high-resolution glow
discharge mass spectrometer operated with an argon/hydrogen mixture. J. Am. Soc. Mass
Spectrom. 1993, 4, 278–285.
205.Knewstubb, P.F.; Tickner, A.W. Mass Spectrometry of Ions in Glow Discharges. I. Apparatus and Its
Application to the Positive Column in Rare Gases. J. Chem. Phys. 1962, 36, 674–683.
206.Luo, S.; Kohiki, S.; Okada, K.; Kohno, A.; Tajiri, T.; Arai, M.; Ishii, S.; Sekiba, D.; Mitome, M.; Shoji, F. Effects
of Hydrogen in Working Gas on Valence States of Oxygen in Sputter-Deposited Indium Tin Oxide Thin
Films. ACS Appl. Mater. Interfaces 2010, 2, 663–668.
207.Bales, G.S.; Bruinsma, R.; Eklund, E.A.; Karunasiri, R.P.U.; Rudnick, J.; Zangwill, A. Growth and Erosion of
Thin Solid Films. Science 1990, 249, 264–268.
208.Horkel, M.; van Aeken, K.; Eisenmenger-Sittner, C.; Depla, D.; Mahieu, S.; Leroy, W.P. Experimental
determination and simulation of the angular disruption of the metal flux during magnetron sputter
deposition. J. Phys. D Appl. Phys. 2010, 43.
209.Hippler, R.; Hubicka, Z.; Cada, M.; Ksirova, P.; Wulff, H.; Helm, C.A.; Stranak, V. Angular dependence of
plasma parameters and film properties during high power impulse magnetron sputtering for deposition of
Ti and TiO2 layers. J. Appl. Phys. 2017, 121, 171906.
210.Hofer, W.O. Angular, energy, and mass distribution of sputtered particles. In Sputtering by Particle
Bombardment III; Behrisch, R., Wirrmaack, K., Eds.; Springer: Berlin/Heidelberg, Germany, 1991; Volume
64, p. 15.
211.Urbassek, H.M.; Hofer, W.O. Sputtering of molecules and clusters. Mat. Fys. Medd. 1993, 43, 97–125.
212.Sundqvist, B.U.R. Desorption of organic molecules from solid and liquid surfaces induced by particle impact.
In Sputtering by Particle Bombardment III; Behrisch, R., Wirrmaack, K., Eds.; Springer: Berlin/Heidelberg,
Germany, 1991; Volume 64, p. 257.
213.Gerhard, W.; Villalba, V.; Maggiolo, A.R. A model calculation of the neutral molecule emission by sputtering
processes. Eur. Phys. J. B 1975, 22, 31–39.
214.Können, G.P.; Tip, A.; De Vries, A.E. On the energy distribution of sputtered dimers. Radiant. Eff. 1974, 21,
269–274.
215.Konnen, G.P.; Tip, A.; de Vries, A.E. On the energy distribution of sputtered Clustrs. Radiant. Eff. 1975, 26,
23–29.
216.Honda, F.; Lancaster, G.M.; Fukuda, Y.; Rabalais, J.W. SIMS study of the mechanism of cluster formation
during ion bombardment of alkali halides. J. Chem. Phys. 1978, 69, 4931–4937.
217.Galera, R.; Blais, J.; Bolbach, G. Molecular sputtering and damage induced by kiloelectron ions in organic
monolayer—Metal systems. Int. J. Mass Spectrom. Ion Process. 1991, 107, 531–543.
218.Haff, P.K.; Watson, C.C.; Yung, Y.L. Sputter ejection of matter from Io. J. Geophys. Res. Space Phys. 1981, 86,
6933–6938.

143
219.King, B.V.; Ziv, A.R.; Lin, S.H.; Tsong, I.S.T. Mass distribution of ejected molecules and clusters in
nanocascade sputtering processes. J. Chem. Phys. 1985, 82, 3641–3645.
220.Salimian, A.; Hasnath, A.; Aminishahsavarani, A.; Upadhyaya, H. Hypothesis on the Influence of the
Magnetic Behaviour of Hydrogen Doped Zinc Oxide during Its Plasma Sputtering Process. Coatings 2021, 11,
222
221.Aparnev, A.; Barten’ev, O. Analyzing the Loss Functions in Training Convolutional Neural Networks with the
Adam Optimizer for Classification of Imagess. Vestn. MEI 2020, 2, 90–105.
222.Balasundaram, S.; Prasad, S. Robust twin support vector regression based on Huber loss function. Neural
Comput. Appl. 2019, 32, 11285–11309.
223.Beattie, J.; Esmonde-White, F. Exploration of Principal Component Analysis: Deriving Principal Component
Analysis Visually Using Spectra. Appl. Spectrosc. 2021, 75, 361–375.
224.Salimian, A.; Aminishahsavarani, A.; Upadhyaya, H. Artificial Neural Networks to Predict Sheet Resistance
of Indium-Doped Zinc Oxide Thin Films Deposited via Plasma Deposition. Coatings 2022, 12, 225.
225.Salimian, A.; Haine, E.; Pardo-Sanchez, C.; Hasnath, A.; Upadhyaya, H. Implementing Supervised and
Unsupervised Deep-Learning Methods to Predict Sputtering Plasma Features, a Step toward Digitizing
Sputter Deposition of Thin Films. Coatings 2022, 12, 953
226.Rosenblatt, F. Perceptron Simulation Experiments. Proc. IRE 1960, 48, 301–309.
227.LeCun, Y.; Bengio, Y.; Hinton, G. Deep learning. Nature 2015, 521, 436–444.

144