Ludwig Maximilians Universität

Fakultät für Physik

Master Thesis in Physics

Optimization of Graphene Devices for

Stimulation and Recording of Neural Activity

Ramon Garcia Cortadella

September 2016

Supervisor: Prof. Dr. Jose A. Garrido

Tutor: Prof. Dr. Bert Nickel
Advanced Electronic Materials and Devices Group
Contents
List of Figures 5

Abstract 8

1 Introduction 9
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Scope of this work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Theoretical background 12
2.1 Stimulation and recording of neural activity . . . . . . . . . . . . . . 12
2.1.1 Neuron biophysics . . . . . . . . . . . . . . . . . . . . . . . . 12
2.1.2 Device-neuron electrical coupling . . . . . . . . . . . . . . . . 16
2.1.3 Recording principles . . . . . . . . . . . . . . . . . . . . . . . 19
2.1.4 Extracellular stimulation principles . . . . . . . . . . . . . . . 22
2.2 Graphene for recording and stimulation . . . . . . . . . . . . . . . . 26
2.2.1 Electronic properties of graphene . . . . . . . . . . . . . . . . 26
2.2.2 Intrinsic noise in graphene transistors . . . . . . . . . . . . . 32
2.2.3 Graphene-based materials for capacitive stimulation . . . . . 34

3 Modelling 36
3.1 Graphene-Electrolyte interface . . . . . . . . . . . . . . . . . . . . . 36
3.2 Frequency response of electrolyte-gated devices . . . . . . . . . . . . 40
3.3 Dependence of transconductance on device geometry . . . . . . . . . 52
3.3.1 Optimization of graphene transistors . . . . . . . . . . . . . . 52
3.3.2 Circular transistors . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3.3 Effect of partial modulation in terms of transconductance . . 56
3.3.4 Effect of partial modulation in terms of noise . . . . . . . . . 59

4 Devices Fabrication and Experimental Methods 62

4.1 Fabrication of graphene devices . . . . . . . . . . . . . . . . . . . . . 62
4.1.1 Growth of graphene by chemical vapor deposition . . . . . . 62
4.1.2 Wet transfer of chemical vapour deposition graphene . . . . . 63
4.1.3 Fabrication of macroscopic devices . . . . . . . . . . . . . . . 65
4.1.4 Design and fabrication of graphene microtransistors . . . . . 66
4.2 Characterization setups and techniques . . . . . . . . . . . . . . . . . 77
4.2.1 Raman spectroscopy . . . . . . . . . . . . . . . . . . . . . . . 77
4.2.2 Electrochemical characterization . . . . . . . . . . . . . . . . 80
4.2.3 Electrical characterization . . . . . . . . . . . . . . . . . . . . 84

5 Experimental Results and Validation of Numerical and Analytical

Models 88
5.1 Topography and Raman analysis . . . . . . . . . . . . . . . . . . . . 88
5.2 Frequency response of graphene devices . . . . . . . . . . . . . . . . 92
 5.2.1 Frequency response of rectangular graphene electrodes . . . . 92
5.2.2 Impedance from source to drain . . . . . . . . . . . . . . . . . 95
5.2.3 From transistor to electrode frequency response . . . . . . . . 97
5.3 Electrochemical characterization of graphene-based porous materials 99
5.4 Electrical characterization of circular transistors . . . . . . . . . . . 103
5.4.1 Basic characterization . . . . . . . . . . . . . . . . . . . . . . 103
5.4.2 Transconductance of circular and rectangular graphene tran-
sistors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

6 Summary and outlook 114

7 Appendix 116
7.1 Rectangular electrodes impedance . . . . . . . . . . . . . . . . . . . 116
7.2 From noise contributions of partitions to total device noise . . . . . . 117
7.3 Circular electrodes impedance . . . . . . . . . . . . . . . . . . . . . . 117
7.4 Capacitive coupling between the Si substrate and contacts/graphene
sheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
7.5 Numerical modelling of g-SGFET-neuron interface . . . . . . . . . . 119

References 124

8 Acknowledgments 132
List of Figures

2.1 Action potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.2 Equivalent Hodgkin-Huxley electric circuit . . . . . . . . . . . . . . . 17
2.3 Cell-cleft and device scheme . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Circuit of electrode signals amplifier . . . . . . . . . . . . . . . . . . 20
2.5 Solution-gated field-effect transistor . . . . . . . . . . . . . . . . . . 22
2.6 Amplifier circuit for transistor signals recording . . . . . . . . . . . . 23
2.7 Gouy-Chapman-Stern model . . . . . . . . . . . . . . . . . . . . . . 24
2.8 Unit cell of graphene . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.9 Dispersion relation of graphene . . . . . . . . . . . . . . . . . . . . . 28
2.10 Dispersion relation and occupation probability . . . . . . . . . . . . 29
2.11 Charge carrier density . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.12 Contributions to interface capacitance in SLG . . . . . . . . . . . . . 31
2.13 Noise amplitude and S/N ratio vs Uds . . . . . . . . . . . . . . . . . 34
3.1 Comsol simulation of graphene . . . . . . . . . . . . . . . . . . . . . 37
3.2 Comparison of FEM and secant methods . . . . . . . . . . . . . . . . 38
3.3 Fermi level in a g-SGFET channel . . . . . . . . . . . . . . . . . . . 39
3.4 Fermi level along the channel . . . . . . . . . . . . . . . . . . . . . . 40
3.5 AC response of electrodes . . . . . . . . . . . . . . . . . . . . . . . . 41
3.6 Equivalent circuit of a graphene electrode . . . . . . . . . . . . . . . 42
3.7 Effect of a resistance in series on the EIS . . . . . . . . . . . . . . . . 45
3.8 Iz(x) and U (x) for a rectangular electrode . . . . . . . . . . . . . . . 45
3.9 AC response of a graphene transistor . . . . . . . . . . . . . . . . . . 46
3.10 Equivalent circuit of a graphene transistor . . . . . . . . . . . . . . . 47
3.11 Effect of Rs on the source to drain impedance . . . . . . . . . . . . . 49
3.12 Effect of the electrolyte resistance on the impedance . . . . . . . . . 49
3.13 Effect of the electrolyte resistance on the source-to-drain impedance 50
3.14 Effective transconductance vs frequency . . . . . . . . . . . . . . . . 52
3.15 Transconductance for increasing channel lengths . . . . . . . . . . . 54
3.16 Scheme of circular and rectangular g-SGFETs . . . . . . . . . . . . . 55
3.17 Transconductance of both circular and rectangular transistors . . . . 56
3.18 Definition transistor partitions . . . . . . . . . . . . . . . . . . . . . 58
3.19 Additive property of the transconductance . . . . . . . . . . . . . . . 59
3.20 Additive property of noise . . . . . . . . . . . . . . . . . . . . . . . . 61

5
4.1 Graphene growth steps . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2 Transfer of graphene CVD . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3 Preparation of macroscopic devices . . . . . . . . . . . . . . . . . . . 67
4.4 Schematic of the DUT and Probe configurations . . . . . . . . . . . 68
4.5 Fabrication details of circular geometry . . . . . . . . . . . . . . . . 69
4.6 Lift of process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.7 Etching of graphene . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.8 Misalignment of metal layers . . . . . . . . . . . . . . . . . . . . . . 73
4.9 SU8 insulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.10 Definition of Probes and DUTs structure . . . . . . . . . . . . . . . 76
4.11 Gold contamination and cracks on SU8 . . . . . . . . . . . . . . . . . 77
4.12 Raman spectrum of graphene . . . . . . . . . . . . . . . . . . . . . . 79
4.13 Processes for different Raman bands . . . . . . . . . . . . . . . . . . 80
4.14 Potentiostat circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.15 Cyclic voltammetry applied and recorded signal . . . . . . . . . . . . 82
4.16 Electrochemical Impedance Spectroscopy . . . . . . . . . . . . . . . 83
4.17 Chronopotentiography . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.18 Source-to-drain impedance measurement setup . . . . . . . . . . . . 85
4.19 Frequency-dependent transconductance measurement setup . . . . . 86
4.20 DC response measurement setup . . . . . . . . . . . . . . . . . . . . 87
4.21 Tip station . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.1 Raman of graphene on Cu and on SiO2 . . . . . . . . . . . . . . . . 89
5.2 Raman of graphene on Cu and on SiO2 . . . . . . . . . . . . . . . . 89
5.3 Optical images of graphene porous materials . . . . . . . . . . . . . . 90
5.4 SEM images of porous graphene . . . . . . . . . . . . . . . . . . . . 91
5.5 Raman analysis of porous graphene materials . . . . . . . . . . . . . 91
5.6 Cyclic voltammetry of graphene . . . . . . . . . . . . . . . . . . . . . 93
5.7 Effect of Rs measured on graphene electrodes . . . . . . . . . . . . . 93
5.8 Imedance spectra of rectangular graphene electrodes . . . . . . . . . 94
5.9 Parameters extracted from graphene electrodes . . . . . . . . . . . . 95
5.10 Effect of Rs measured on graphene transistors . . . . . . . . . . . . . 96
5.11 Imedance spectra of rectangular graphene transistors . . . . . . . . . 96
5.12 Parameters extracted from graphene transistors . . . . . . . . . . . . 97
5.13 Equivalent transconductance versus frequency for macrotransistors . 98
5.14 Comparison of cyclic voltammograms . . . . . . . . . . . . . . . . . . 100

6
5.15 Cyclic voltammetry of reduced graphene oxide and capacity extraction100
5.16 Chronopotentiometry of graphene based porous electrodes . . . . . . 101
5.17 Electrochemical impedance spectra of graphene based electrodes . . 102
5.18 Electrochemical impedance spectra of 200nm thick rGO . . . . . . . 104
5.19 Non-homogeneous current injection . . . . . . . . . . . . . . . . . . . 105
5.20 Ids vs Uds and Ugs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.21 Cleaning of graphene . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.22 Device transconductance . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.23 Transconductance of circular and ractangular transistors in Probes . 109
5.24 Measured Ids vs Ugs curves and maximum gm in DUTs . . . . . . . 110
5.25 Simulation of the effect tracks resistance . . . . . . . . . . . . . . . . 111
5.26 Photomicrograph of SU8 layer broken . . . . . . . . . . . . . . . . . 111
5.27 Complete simulation of the effect of the channel length . . . . . . . . 112
5.28 Simulated Ids vs Ugs curves and maximum gm in DUTs . . . . . . . 113
7.1 Impedance spectra of circular electrodes . . . . . . . . . . . . . . . . 118
7.2 Parameters measured in circular graphene electrodes . . . . . . . . . 119
7.3 Effect of substrate on the source-to-drain impedance . . . . . . . . . 120
7.4 FEM simulation of an action potential . . . . . . . . . . . . . . . . . 122
7.5 Comsol simulation of the graphene electrode-neuron coupling . . . . 123

7
Abstract

Electric coupling between nerve tissue and electronic devices is the usual approach
used for recording infromation from neurons and stimulating especific resposes in
the nervous system. In this work, graphene electrodes for stimulation and graphene
solution-gated field-effect transistors (g-SGFETs) for recording are modelled. First,
a model of single layer graphene (SLG) based on the finite elements method (FEM)
is developed. Using this model, the frequency response of graphene devices can be
simulated. Besides, the analytical expressions describing the same frequency effects
are derived by modelling the graphene sheet as a ladder circuit. Both the numerical
analysis and the analytical expressions are in good agreement and indicate that
the electrical response of graphene at high frequencies strongly depends on the
sheet resistance. This dependence implies non-homogeneous current flow through
the graphene-electrolyte interfase which has a strong impact on the applicability of
graphene electrodes for neural stimulation. Secondly, the numerical and analytical
analysis of graphene transistors leads to a better understanding of the relationship
between the geometry and sensitivity of the devices. From the derived observations
a new design based on a circular geometry is conceived. This design improves
the transconductance which can be achieved in a limited spatial scale. Both the
frequency effects and the design rules for improving the sensitivity of devices are
validated. The high frequency effects are observed experimentally in macroscopic
electrodes and transistors. These devices are fabricated using not only single layer
graphene but also porous graphene materials which present much higher values of
charge injection capacity and sheet resistance. The design rules are validated by
microfabricating circular and rectangular transistors on a flexible substrate. Their
performance in terms of transconductance is evaluated, showing the advantages and
disadvantages of using a circular geometry.

8
1 Introduction

1.1 Motivation

In the last decades, huge developments have been achieved in the field of neurotech-
nology. New technologies for brain computer interface have brought new hope to
patients suffering from physical impairments [1], deafness [2] or muteness [3] among
others. These applications rely either on the stimulation of the nerve tissue for
communicating sensory information to the central nervous system or on the sensing
of neural activity to reconstruct the intentions of the patient.

The detection of electrical activity in the brain can be performed in a non-invasive

way by electroencephalography (EEG). This technology is extensively used for clin-
ical applications [4], but it provides low spatial and temporal resolution due to the
damping and filtering of the signal by the skull. In order to overcome the limita-
tions in terms of spatial resolution, more invasive techniques can be employed. Micro
electrode arrays (MEA) provide high temporal and spatial resolution by placing the
sensors directly onto the cortex of the brain. In this way, action potentials from in-
dividual neurons have been recorded [5]. MEAs can also be used for inducing neural
stimulation by generating ionic currents in the extracellular environment. The volt-
age shift produced by these currents lead to the depolarization of cell membranes
which in turn might elicit an action potential. Both the recording and stimulation
by MEAs is limited by the impedance of the electrode-electrolyte interface. In order
to achieve high spatial resolution, the size of the devices must be reduced, leading
to high interface impedance. Large impedances decrease the sensitivity or charge
injection capacity of the electrodes. In order to improve the performance of MEAs,
low impedance electrodes based on novel materials have been developed [6].

The electric stimulation of neural activity is restricted to the use of electrodes.

However, recording does not necessarily rely on electrode technology. In order
to overcome the limitations imposed by the interface impedance, an alternative
approach based on field-effect devices was proposed [7]. Solution-gated field-effect
transistors (SGFETs) present a channel in contact with the electrolyte. The voltage
fluctuations in the electrolyte couple capacitively with the channel, thereby shifting
its Fermi level and modifying the density of carriers. By applying a voltage between
two points in the channel a current which is proportional to the conductivity of

9
the channel is generated. In this way, an in situ preamplification of the signal is
performed. In order to achieve high sensitivities, the devices must present high
carrier mobilities and a strong capacitive coupling with the electrolyte. Besides
their sensitivity, the applicability of these implantable technologies is limited by
the rejection of the body [8]. The new generation of sensors and actuators must
present high biocompatibility, high mechanical and chemical stability as well as a
high flexibility.

In the search for a suitable material, graphene is an outstanding candidate. Due

to its bidimensional nature graphene presents a very large flexibility, always ham-
pered by the rigidity of the substrate on which it is transferred. It also presents
good biocompatibility [9] and it is chemically inert. Finally, graphene presents a
remarkable carrier mobility in the order of 105 cm2 /Vs at room temperature [10]
due to its peculiar band structure. On the other hand, graphene presents a rel-
atively low interface capacitance (≈ 2 µF/cm2 ). This low value strongly hampers
the application of graphene for neural stimulation, while it does not limit signifi-
cantly the sensibility of graphene solution-gated field-effect transistors (g-SGFETs)
[11]. Despite high sensitivities have been achieved, flexible g-SGFET transistors
still have to outperform the current MEAs technology and fully demonstrate their
potential for recording action potential signals with sub-cellular and sub-threshold
resolutions.

1.2 Scope of this work

In this thesis, special emphasis is placed on the modelling of graphene based devices
in order to provide new design rules for improving the sensitivity of g-SGFETs and
understanding the frequency response of graphene electrodes. The dependence of
the device properties on the size and geometry of the device is studied by deriving
analytical expressions and performing numerical analysis. A model based on the
finite elements method (FEM) is developed for describing the graphene-electrolyte
interface and the modulation of graphene conductivity. This model allows to sim-
ulate the response of graphene devices under conditions which could be hardly
described analytically or measured experimentally. In addition, the simulations
provide a reliable guide for the derivation of analytical models.

Firstly, an analysis of the frequency response of macroscopic devices is presented

10
leading to compact expressions describing the impedance of graphene electrodes
and the impedance from drain to source of g-SGFETs. The impact of these high
frequency effects on the applicability of graphene electrodes for neural stimulation
is assessed. Second, the dependence of the sensitivity of g-SGFETs on the channel
width (W ) and length (L) is explored and a simple design rule is derived in order
to find the optimum W/L ratio. From this design rule it is found that in order
to increase the sensitivity of g-SGFETs, the channel can be folded within the area
where the electrical signal has to be recorded. Exploiting this observation, a new
design of g-SGFETs based on a circular geometry is conceived. Finally, the effect
of partially modulating the channel is investigated. The derivation of an analytical
model provides a simple but meaningful relation between the area affected by signal
to be recorded and the sensitivity of the transistor. It also allows for the calcula-
tion of the noise in graphene transistors which present a non-homogeneous channel
conductance.

These theoretical observations are then validated experimentally. For this purpose
both macroscopic devices and microscopic g-SGFETs are fabricated. The former are
produced on a rigid Si/SiO2 substrate using both single layer graphene and porous
materials based on exfoliated graphene and reduced graphene oxide. The large size
of the devices allows for a clear observation of the frequency response predicted by
the models and a first approach to the characterization of the electrochemical prop-
erties of the graphene based porous materials. Additionally, microscopic g-SGFETs
with circular and rectangular geometries are fabricated on a flexible polyimide sub-
strate using the technology recently developed in our group [12].

This thesis is divided in four sections: section 2 provides an introduction to rel-

evant theoretical concepts in order to provide a context and background for fully
understanding the work presented in the following sections. Section 3 presents the
development of the analytical and numerical models used for describing new effects
in graphene based devices. Section 4 describes the experimental methods used for
the validation of the models, including both a description of the characterization
techniques and the fabrication processes. Finally, section 5 presents the experimen-
tal results which allow to validate most of the theoretical observations.

11
2 Theoretical background

In this section, the theoretical basis for understanding the physics behind the record-
ing and stimulation of neural activity are given. First, the electrical properties of
electrogenic cells, their coupling with electronic devices and the fundamental record-
ing and stimulation principles are discussed. Second, the properties of graphene are
presented, placing special emphasis on their electrical properties which play an im-
portant role in the applicability of this material in neurotechnology.

2.1 Stimulation and recording of neural activity

Before proposing novel materials and device designs for recording and stimulation
of electrical activity in nerve tissue, it is important to understand the physics be-
hind the three systems involved; electrogenic cells, sensors and actuators, and the
electronics used for the amplification of signals and generation of voltage pulses.
First, the electrical properties of electrogenic cells are described by presenting the
origin and significance of transmembrane potentials. Second, the coupling between
electrogenic cells and electronic devices is presented by introducing the point con-
tact model (PCM). Finally, the recording and stimulation principles are introduced.
Recording based on SGFETs is compared with the commonly used recording based
on electrodes pointing out the advantages of the former. The concept of extracellu-
lar stimulation is also introduced and the requirements in terms of electrochemical
properties of the electrodes are detailed.

2.1.1 Neuron biophysics

The human brain contains above 1011 neurons with approximately 2.1014 contacts
between them [13]. Within the neurons, signal transmission is performed by the
propagation of ionic signals along the axons. The so called action potentials can
propagate from one cell to another through an electric synapse or elicit the delivery
of neurotransmitters in chemical synapses. If we aim to establish a direct commu-
nication between electronic devices and neurons, the dynamics and nature of action
potentials (AP) has to be well understood.

Neurons, as the rest of cells, are enveloped by a cellular membrane, mainly composed
by lipids and proteins [14]. The cellular membrane is an important structure for

12
all living cells. It defines the volume of the cell, separating the cytosol from the
extracellular space and regulating the concentration of chemical species inside and
outside the cell. For this purpose, the cellular membrane has a different permeability
to different kinds of chemical species. The cellular membrane is constituted by a
lipid bilayer, the main components of which are phospholipids [15]. Phospholipids
present a polar head and a hydrophobic tail, and due to their amphiphilic character
they form stable bilayers in aqueous media. While the permeability of lipid bilayers
to hydrophobic compounds is relatively high, it is very high to most hydrophilic
compounds. In general, the more hydrophobic and smaller the molecules are, the
faster they can diffuse through the membrane. In addition, it is worth to point out
that lipid bilayer membranes are highly impermeable to charged species. Therefore,
the regulation of ion concentrations has to be performed by specific membrane
proteins [16]. These proteins can be classified as ion channels or ion transporters.
The former allow ions to diffuse down electrochemical gradients while the latter
require an energy input from the metabolic system for moving ions against an
electrochemical gradient. The induced differences in ion concentrations and the
selective membrane permeabilities are responsible for an electric potential difference
between the inner and outer part of the cellular membrane [17].

The concentration gradients between the intracellular and extracellular environ-

ments produce a flux of ions through the cell membrane. The flow of charged species
in favour of concentration gradients ultimately produces an electric potential across
the membrane which counteracts the chemical gradients. In this way, the equilib-
rium is reached when the electric force counteracts the effect of the concentration
gradient. Equalizing these two contributions it is possible to find the potential drop
across the cell membrane due to the concentration gradient of a single kind ionic
species [18]:

dci dU
− Di = −µi ci (2.1)
dx dx

Here, Di stands for the diffusion coefficient of an ionic species i, ci represents the ion
concentration and µi for the electrical mobility. U stands for the electric potential.
This expression can be rewritten by using the Nernst-Einstein relationship (i.e.
Di = µi RT /z i F ) to obtain the following expression [18]:

13
 dU RT 1 dci
=− i i (2.2)
dx z F c dx

Here, R is the gas constant, T the temperature, F the Faraday constant and z i the
charge of an ion i. This equation can be integrated with respect to x across the
membrane to give the well-known Nernst equation [18]:

i i i RT [ci ]in
Ueq = Uin − Uout =− ln (2.3)
zi F [ci ]out

i is the equilibrium potential usually called the Nernst potential. If more ionic
Ueq
species are included in this picture, the problem becomes more complicated. The
potential generated by the equilibrium of one ionic species affects the ionic currents
through the membrane of the other species. Therefore, a metastable equilibrium is
reached in which the flow of different species compensate each other. The flux of
individual ion species in this state slightly changes the total concentration of ions
inside and outside the cell. In order to keep the resting state permanently the ion
transporters must be active [19], continuously pumping ions through the membrane,
and therefore keeping the concentration gradients. The transmembrane potential in
rest ). It does not
the metastable equilibrium is referred as the resting potential (UM
depend only on the concentration of ions but also on the permeability (P i ) of the
membrane to each of them. In the case of a completely impermeable membrane, no
flux of ions can be established and therefore no equilibrium potential exists. In this
way, the ions with larger permeabilities dominate the equilibrium potential. The
expression which describes the resting potential was developed by David Goldman
in 1943 and, for the relevant case in which K+ , Na+ and Cl- are the main permeant
ions, it can be written as [17]:

rest RT P K [K]out + P N a [N a]out + P Cl [Cl]in

UM = log K (2.4)
F P [K]in + P N a [N a]in + P Cl [Cl]out

If the membrane is only but completely permeable to one ion, the Nernst equation
is recovered from the Goldman equation. Otherwise, if the membrane is permeable
to several ions, the transmembrane potential depends on the permeability to each
ionic species. This effect is one of the keys to understand the communication of
neurons through firing of action potentials. Action potentials are the mechanism of

14
neurons for transmitting information in the form of electric pulses. The first phase
of an AP is a rapid depolarization of the cellular membrane, it is followed by a rapid
repolarization back to the resting potential. These potential changes are induced by
changes in the permeability of the membrane to different ions. The following table
shows typical concentrations of ion species in the cytosol and extracellular space for
mammalian neurons.

Concentration (mM)
Ion Equilibrium potential (mV)
Intracellular Extracellular
Potassium (K+ ) -85.6 140 5
Sodium (Na+ ) from +86.5 to +58.3 from 5 to 15 145
Chloride (Cl- ) from -85.1 to -33.4 from 4 to 30 110
Calcium (Ca2+ ) from +118.3 to +127.2 0.0001 from 1 to 2

Table 2.1: Concentration of ion species in and out of mammalian neurons. Equilib-
rium potential of individual ionic species are calculated using Nernst equation for
T=298 K. Concentration values from [17].

The Nernst potential of K+ is very negative due to the high concentration of potas-
sium inside the cell. On the other hand, the high concentration of Na+ outside the
cell tends to drive the membrane to positive values. In the resting state, the per-
meability to potassium is much larger than the permeability to sodium, therefore
the resting potential is close to the Nernst potential of K+ (i.e. from -60 to -70 mV
in case of neurons). If the permeability to sodium is increased, the membrane po-
tential increases quickly towards the sodium equilibrium potential. In other words,
Na+ ions suddenly enter the neuron shifting its potential towards a positive value.
During an action potential, the sodium channels remain in the open state for a short
time of approximately 1 ms before becoming closed again. Fig. 2.1 indicates the
phases of an action potential and the permeability to Na+ and K+ in each of them.

The changes of permeability are achieved through the opening and closing of voltage-
gated ion channels. These transmembrane proteins have a voltage-sensor domain
which changes its conformation depending on the electric field across the cell mem-
brane [20]. The opening of ion channels occurs above a certain threshold voltage;
in the case of sodium channels, this voltage is induced by the same action potential
which propagates along an axon. In the case of potassium channels, it is the polar-
ization induced by the sodium intake which activates the channels. The dynamics
of the channels and its dependence with the membrane potential were modelled
mathematically by Hodgkin and Huxley [21] (see appendix 7.5). The Hodgkin and

15
Figure 2.1: Action potential simulated from the Hodgkin-Huxley model (described
in the next paragraph and in appendix 7.5). The phases of an action potential as
well as the evolution of the permeability to different ions are indicated in the graph.
The light colour indicates the phases with high permeability to K+ and the strong
colour indicates the phases at which the membrane is more permeable to Na+ .

Huxley model (HH model) takes into account not only the conductance from ion
channels but also small contributions from leakage and capacitive currents, which is
significant for the typical capacity of lipid bilayers (≈ 1 µF/cm2 ). The conductance
of the membrane to the different ions (Gi ), the capacitance of the membrane (CM )
i )
and the resistive leakage (GL ) as well as the equilibrium potential of each ion (Ueq
can be used within the electric model shown in Fig. 2.2.

2.1.2 Device-neuron electrical coupling

In order to establish a communication between nerve tissue and electronic devices,

an interface between the two systems has to be built. In both cases, the charge
carriers in nerve tissue are ions while in conventional electronic devices the carriers
are electrons and holes. A proper interface has to accomplish the effective conversion
of ion based signals to electronic signals. This process can be achieved either through
reduction-oxidation (red-ox) reactions at the device-electrolyte interface or through
a capacitive coupling.
The interface can be modelled by an equivalent electric circuit. The point contact
model (PCM) simplifies the interface to the equivalent electric circuit shown in Fig.
2.3. This model can be used to describe the coupling between the cell inner potential
(UM ) and the potential in the cleft (UJ ) which separates the cell from the electronic
device [22].

16
Figure 2.2: Equivalent electric circuit of the cell membrane. The capacity of the
membrane (CM ), the conductance to each ion (i) is considered separately (Gi ) as
well as the Nernst potentials for each of them (Ueqi ). The conductance for each ion

depends on the membrane potential as described by the HH model. In addition,

the leakage through the membrane is taken into account (GL ).

Applying Kirchhoff’s laws at the relevant nodes, the relation between the voltage
differences and the circuit parameters can be found. The voltage in the cleft can be
expressed as follows:

dUS dUJ
dUM dUJ

gJ (UE − UJ )+cint − + cM −
dt dt dt dt
X
i
+ gL (UM − VJ ) + gAM (UM − UJ − U0i ) = 0 (2.5)
i

Here, gJ is the conductance of the cleft, cint the capacitance of the device-electrolyte
interface, cM the membrane capacitance and gAM i the specific ion conductivities.
All these variables are normalized by the area of the attached membrane (AAM )
which is the part of the membrane separated from the device by the cleft (see Fig.
i is the Nernst potential of each ion species.
2.3). Finally, Ueq

17
Figure 2.3: Equivalent circuit of the cell-electrolyte and electrolyte-device interfaces.
Thicker lines illustrate the device and the cell which is impaled by a patch-clamp
pipette. A current Iinj can be injected by an electrode in the pipette. Both free
and attached membranes can be distinguished in the figure. The potentials in the
cleft (UJ ), in the cell (UM ), in the device (US ) and in the electrolyte bath (UE ) are
also indicated. The equivalent electric circuit accounts for the membrane electrical
properties as well as the cleft conductance (GJ ) and the interface capacitance (Cint ).

The potential inside the neuron (UM ) can be expressed as:

   
dUM dUE X
i i
Iinj = AF M cM − + gL (UM − VJ ) + gF M (UM − UE − Ueq )
dt dt
i
   
dUM dUJ X
i i
+ AAM cM − + gL (UM − VJ ) + gAM (UM − UJ − Ueq )
dt dt
i
(2.6)

The part of the cell membrane which is not in contact with the device is referred
as the free membrane (F M ). In Eq. 2.6, the properties of the F M are normalized
by its area (AF M ). In Eq. 2.6, the effect of a patch-clamp pipette in current clamp
can be simply modelled by adding a current source term (Iinj ). gFi M and gAMi must
be distinguished in order to account for possible non-homogeneous density of ion
channels along the membrane.

18
Solving these equations analytically, taking into account the dynamics of the ionic
channels (from the HH model) and without doing approximations is an immense
challenge. Nevertheless, they can be easily solved numerically (see appendix 7.5).
Some approximations can be made in order to understand the different kind of
signals to be recorded by the electronic device[22].
A simple situation occurs when the attached membrane has no voltage-gated ion
channels and the current injected by the cell in the cleft is low. Then, the last term
in Eq. 2.5 can be neglected and the derivative of UM is much larger than that of
UJ . Besides, the potential at the electrolyte can be set to ground (UE = 0) and the
potential in the device set to a constant value (dUS /dt=0). Under these conditions,
Eq. 2.5 simplifies to Eq. 2.7 which describes the dependence between the voltage
in the cleft and in the cell:

dUM
gJ UJ = gL UM + cM (2.7)
dt

If the leakage conductivity dominates the capacitive currents, the signal at the cleft
will be proportional to UM . On the other hand if the capacitive currents dominate,
UJ will be proportional the derivative of UM .
Extracellular stimulation of neurons requires a voltage shift in the cleft so that the
ion channels are opened. For instance, this can be achieved by injecting current with
the device. A crucial question is how long this voltage changes have to be applied
in order to properly activate the ion channels. The required pulse duration depends
on its strength and the exact relation between strength and duration can be found
by solving Eq. 2.6 numerically, taking into account the HH model for the membrane
conductance [23]. The solution to this problem shows that the strength-duration
relationship for intra and extracellular stimulation presents the same tendency. In
both cases, the required strength increases as the pulse duration (τ ) decreases,
approaching an asymptotic 1/τ relation for very short pulses. On the other hand,
it approaches a constant strength for long pulses (the so-called rheobase [24]). As
an example, the required strength for stimulating a typical neuron is around 20 mV
for a monophasic pulse of 1 ms [23].

2.1.3 Recording principles

The voltage signals induced in the cleft due to ionic currents generated by cells
can be recorded with electrodes. The recording is based on measuring the voltage
induced in the electrode as a result of the voltage fluctuations induced by the neural
activity. Fig. 2.4a shows the scheme of the preamplifier needed to record the poten-
tial at the cleft. As expressed in Eq. 2.8, the output signal (Uout ) is proportional to
the voltage divider between the input impedance (Zin ) of the operational amplifier
and the electrode-electrolyte interface impedance Zint .

19
 Zin
Uout = Uin (2.8)
(Zin + Zint )

Here, the input signal Uin stands for the voltage fluctuation in the cleft.
If the size of the electrode has to be reduced to the size of few or even a single neuron,
the impedance, which is inversely proportional to the area of the device, increases
dramatically. A low ratio between the electrode impedance and the amplifier input
impedance leads to a low signal amplification. Fig. 2.4b shows the schematic of
the noise sources contributing to the final output noise (eout ). In the circuit, the
electrode-electrolyte interface is modelled as a capacitor (Cint ). This assumption is
discussed in section 2.1.4. On the other hand, the resistor RB has to be added in
order to provide a path for the amplifier’s bias current[25]. Eq. 2.9 gives the relation
between the noise and the circuit parameters [25]. It can be observed that the noise
increases if Cint decreases. This increase in the noise combined with the attenuation
of the signal for low Cint values (Eq. 2.8) implies that large capacitances are needed
in order to obtain a high signal to noise ratio (S/N ).

2
RB 2
1 Zin
e2out = e2RB + I 2
amp + e2
amp + e 2
i
1 + (2πRB Cint f )2 1 + (2πRB Cint f )2 (Zin + Zint )2
(2.9)
The amplifier current noise and voltage noise are referred as eamp and Iamp respec-
tively. f represents frequency and ei the voltage fluctuations at the electrolyte.

(a) (b)

Figure 2.4: a) Schematic of the preamplifier for electrode-based recordings. b)

Schematic of the noise sources contributing to the total output noise in electrode
based recording. Iamp stands for the current input noise of the amplifier, eamp
stands for the voltage input noise of the amplifier, ei for the voltage fluctuations
in the extracellular medium and eRct for the noise of the resistor in the equivalent
circuit.

The relation between size and impedance constrains the spatial resolution of record-

20
ings based on electrodes. Despite single action potentials have been recently recorded
with electrodes based on novel materials [26], the spatial resolution or the S/N could
be improved if the constriction imposed by the interface impedance was avoided.
Changing to a different recording principle would allow to record action potentials
signals with sub-cellular resolution (e.g. from single ion channels) and improve the
S/N ratio. Solution-gated field-effect transistors (SGFET) offer a completely dif-
ferent recording principle [27] [28], relying on the modulation of a resistive channel
which is in direct contact with the electrolyte solution. Fig. 2.5 shows a scheme
of a SGFET. The detection of signals is achieved through the modulation of the
resistance of a semiconducting channel. The channel is contacted from the two sides
through a metal-semiconductor junction in order to apply a drain-to-source poten-
tial. The gate potential is set with a reference electrode which couples capacitively
with the channel through the channel-electrolyte interface. This coupling allows to
control the Fermi level in the channel. In order to minimize leakage currents to
the gate, the metal tracks and contacts have to be insulated from the electrolyte.
For that purpose a thin polymeric layer is defined on the devices, keeping the ac-
tive channels uncovered. To avoid the presence of uncovered parts of the metallic
contacts, a short region of the channel around the contacts is also insulated. This
covered area is called the access region and it represents an additional term to the
resistance in series (Rs ) of the device. Rs also includes the contact resistance (Rc )
of the metal-semiconductor interface and the resistance of the metallic tracks (Rt ).
The sensitivity of the device, quantified as the transconductance (gm ), is defined as
the change of drain-to-source current due to a potential change in the electrolyte.
The transconductance can be expressed as in Eq.2.10.

dIds dG W dn
gm = = Uds = Uds µe (2.10)
Ugs Ugs L Ugs

In the last equality, the conductanductance of the channel (G) is expressed through
the density of charge carriers (n) in the channel, the carrier mobility (µ) and the
geometrical parameters W and L standing for the width and length of a rectangular
channel respectively. The constants Uds and Ugs represent the drain-to-source and
the gate-to-source voltage respectively and e the charge of the electron.

The changes of current from drain to source (Ids ) are then measured with a current-
to-voltage converter. Fig. 2.6a shows the scheme of this converter and Eq. 2.11
indicates the output voltage as a function of the drain-to-source current:

Uout = −Ids Rg = −Uin gm Rg (2.11)

Here, Rg is the value of the resistor displayed in Fig. 2.6a which is part of the
amplifier system. This current is proportional to the voltage signal at the channel-
electrolyte interface and to the transconductance of the transistor. It is noteworthy

21
Figure 2.5: Illustration of a solution-gated field effect transistor. On the left hand,
a top view is presented, the channel (black) is connected to the drain and source
which are covered by a passivation layer (orange). On the right hand a side view
is shown; the contacts (yellow) are connected to the channel (black) which stands
on the substrate (gray). The channel and passivation layer are in contact with the
electrolyte in which the potential is fixed by a reference electrode.

that the output signal is not directly limited by the size of the device, because neither
is the transconductance. The schematic in Fig. 2.6b shows the noise sources for
the current to voltage converter and Eq. 2.12 the relation between the output noise
and circuit parameters. In the optimal case, the S/N takes the value of the signal
amplitude over the intrinsic noise of the device. The intrinsic noise (Ii ) strongly
depends on the properties of the transistor. The nature of intrinsic noise in graphene
devices is described in detail in section 2.2.2.

Rg 2 Rg + Rch 2
e2out = (eds2 + Rch 4T KB )( ) + eamp ( 2
) + Rg 4T KB + Iamp Rg2 + Ii2 Rg2
Rch Rch
(2.12)
Here, KB stands for the Boltzmann constant and T for the temperature. Rch
represents the resistance of the transistor channel and eds the noise of the voltage
source supplying (Ugs ).
It is clear from Eq. 2.11 that if the transconductance of the device is increased, the
signal at the output increases but the amplifier noise does not. Having a signal above
the amplifier noise level is possible even for extremely small devices, as long as the
transistors present a large transconductance. In section 2.2, the transconductance
of the graphene devices and its relation with the intrinsic noise is discussed.

2.1.4 Extracellular stimulation principles

As described before, an electrogenic cell can be stimulated so that an action potential

is elicited. Applying a voltage pulse with the required strength and duration requires

22
 (a) (b)

Figure 2.6: a) Schematic of the preamplifier for transistor-based recording. b)

Schematic for the noise sources contributing to the total output noise in transistor-
based recording. Iamp stands for the current input noise of the amplifier, eds stands
for the drain-to-source voltage supplier noise, eamp stands for the operation amplifier
voltage noise, Ii stands for the intrinsic noise of the transistor and URch for the
channel resistance noise respectively.

the injection of intense current pulses into the cleft during a sufficient time. For
this purpose, electrodes are the only devices which can be used. Electrodes are
typically divided into two classes, polarizable and non-polarizable electrodes. Ideally
polarizable electrodes are properly modelled by an electrical capacitor, meaning that
when an overpotential is applied, the interface gets charged and in the steady state,
the current tends to zero. On the other hand, non-polarizable electrodes can be
modelled by a resistor; the resistive behaviour holds as long as the relation between
the applied voltage and the current is linear.
When a conductive material is brought in contact with an electrolyte, an electric
potential difference between the bulk of both phases is induced. This voltage dif-
ference, called the Galvani potential is generated by the formation of an electric
double layer. In the side of the electrode, charge carriers accumulate at the surface,
while in the aqueous phase ions accumulate in close contact with the electrode. The
structure of this double layer has been extensively described, with major contribu-
tions from Helmholtz, Gouy, Chapman or Stern. The Helmholtz model considers
a rigid structure, constituted by two charged planes. The first is called the inner
Helmholtz plane (IHP), it is composed of non-solvated ions in close contact with
the electrode. The second layer is referred to as the outer Helmholtz plane (OHP)
in which solvated ions reside in a more distant position. This model does not take
into account the thermal fluctuations of the ions. On the other hand, the Gouy-
Chapman model considers that the ions are distributed according to Boltzmann
statistics. In this way, the double layer is presented as a diffuse distribution of
charge in the electrolyte phase (Fig. 2.7a). An interesting midpoint is found in the

23
Stern model in which the ions are constrained to a rigid plane in the immediate
vicinity of the electrode while the ions outside this plane are dispersed as in the
Gouy-Chapman model [29] (Fig. 2.7b). Knowing the distribution of ions and the
electric permittivity at the interface, the double layer capacitance can be calculated
as for a parallel plate capacitor (Cdl ) [30]. The latter symbol has to be distinguished
from the interface capacitance (Cint ) which is a more general term accounting for
other possible capacitive effects at the interface (see section 2.2.1).

(a) (b)

Figure 2.7: Illustration of the electrolyte double layer described by the most relevant
models.a) Diffuse layer of the Gouy-Chapman model. The voltage profile follows
an exponential decay. b) Gouy-Chapman-Stern model. A linear voltage drop in
the Helmholtz plane is followed by an exponential drop at the diffuse double layer.
From [30].

When an electric potential is applied externally between the electrode and elec-
trolyte, the equilibrium is disrupted and charge transfer between the two phases
may occur. However, even if the transfer of charge is thermodynamically favourable,
reactions might happen with a very small rate. The reaction rate constant can be
calculated by the transition state theory [29]. If the activated complex resembles
the reactant, the energy change produced by an external electric potential does not
change much the activation energy. On the other hand, if it resembles the product,
the activation energy is shifted by approximately (eη) where η is the applied over-
potential. This dependence is modelled by the parameter α which indicates how
close to the reduced species the activated complex is [29]. The dependence between
the charge transfer current (j) and η is described by the Butler-Volmer equation
(Eq. 2.13).

24
 j = j0 e(1−α)F/RT η − e−αF/RT η
 
(2.13)

If no external potential is applied, the cathodic and anodic currents counteract each
other. These current at the equilibrium is the so-called exchange current density
(j0 ).
Eq. 2.13 describes a strong non-linear dependency between the applied voltage
and the current. Nevertheless, in the low overpotential limit this equation can
be linearized to find an equivalent ohmic resistance, the so-called charge transfer
resistance (Rct ).

F
j= j0 η (2.14)
RT
If j0 is large, for instance due to a low activation energy, large currents will flow
when an overpotential is applied. In this case, the electric behaviour of the electrode
will be dominated by the charge transfer processes and the electrode is classified as
non-polarizable. On the other hand, if charge transfer kinetics at the interface is
slow, the charge and discharge of the double layer capacitance might dominate the
electric response of the electrode. In the extreme case, the electrodes are classified
as ideally polarizable electrodes.
For achieving high and long-lasting current pulses the use of non-polarizable elec-
trodes presents an important advantage, namely the possibility of achieving large
voltages by simply increasing the applied overpotential. Nevertheless, for extracellu-
lar stimulation, ideally polarizable electrodes are advantageous to avoid electrolysis
of water and reduction or oxidation of the electrode and biological redox systems
[17]. If the species involved in the reactions are immobilized on the metal surface,
the reactions are reversible but if the chemical species spread into the electrolyte
these processes are irreversible. Irreversible faradaic reactions might lead to the
accumulation of toxic products in the living tissue as well as harmful shifts of pH
[17].
The injection of charge by polarizable electrodes can be achieved by applying voltage
ramps at the electrode. The slope of the applied ramp and the capacitance of the
interface determine the intensity of the injected current, as described by the current-
voltage relation of an electric capacitance (Eq. 2.15):

dU (t)
I(t) = C (2.15)
dt
In order to achieve high pulses with a certain electrode, the slope of the voltage
ramp must sufficiently large. Given a certain pulse duration, increasing this slope
implies a larger difference between the initial and final voltage at the electrode-
electrolyte interface. Even for an almost ideally polarizable electrode, there is a

25
voltage at which the faradaic reactions dominate the capacitive behaviour. The
safe voltage range is determined by the electrochemical potential window (EW)
of the electrode in the physiological solution and it limits the applicable voltage
slope for a certain pulse duration. In fact, the suitability of electrode materials
for extracellular capacity is usually measured in terms of charge injection capacity,
defined as the charge that can be injected without polarizing the electrode above
the electrochemical potential window (i.e. Cint EW ) [31].

2.2 Graphene for recording and stimulation

Graphene is a monoatomic layer of carbon atoms structured in a honey-comb lat-

tice. Graphene is the basic structure which composes other carbon allotropes such
as graphite, carbon nanotubes or fullerenes. The properties of graphene where the-
oretically explored many years before its isolation [32][33], but it was in 2004 when
Geim and Novoselov exfoliated high quality graphene from graphite, allowing for
the proper characterization of its properties [34] [35]. With its unique properties,
graphene is an interesting material for science and a promising material for tech-
nology. The application of graphene is being explored in many research fields such
as energy storage, plastic nanoelectronics or neurotechnology and its real applica-
bility arose from its large scale production by chemical vapour deposition [36]. In
addition to its electronic properties [37], graphene presents outstanding mechanical
properties [38] including a high flexibility due to its bidimensional nature. In this
section, the attention is focussed on the electronic properties of graphene for its
application in the field of bioelectronics.

2.2.1 Electronic properties of graphene

Graphene presents a trigonal planar symmetry with two atoms per unit cell (Fig.
2.8). This symmetry arises from the sp2 hybridization of the carbon orbitals which
leads to three in plane σ bondings between adjacent atoms.
The pz orbitals, oriented perpendicularly to the carbon atoms plane, dominates
the transport properties of the material due to their strong overlapping. The first
description of the band structure of graphene was based on the tight binding ap-
proximation and was developed as an approach to the band structure of graphite
[32]. The derived dispersion relation can be expressed as:

E± (k) − ED = ±t 3 + f (k) − t0 f (k)

p
(2.16)

√
√   
3 3
Here, f (k) = 2cos( 3ky a)+4cos 2 ky a cos 2 kx a . t and t’ stand for the nearest
neighbour hopping energy and the next nearest-neighbour hopping energy respec-
tively. k represents the momentum of electrons and a the distance among nearest

26
Figure 2.8: Unit cell of graphene.a) Direct√lattice in the real-space.
√ The vectors in
blue represent the lattice vectors (a1 = (3, 3)a/2, a2 = (3, − 3)a/2). b) √Brillouin
zone is depicted
√ (red) as well as the reciprocal-lattice vectors (b1 = (1, 3)2π/3a,
b2 = (1, − 3)2π/3a). The points K and K 0 in the Brillouin zone are the reciprocal-
lattice points where the so-called Dirac point is found. As described in the main
text, many specific properties of graphene originate at these points.

carbon atoms. If t0 equals zero, the dispersion relation is symmetric for the bonding
and antibonding π bands (i.e. + and - signs in Eq. 2.16). Both bands get in contact
at the K and K 0 points of the Brillouin zone. The term ED stands for so-called
Dirac energy; the energy at which both bands meet.

Expanding this expression around the K and K 0 points of the Brillouin zone, the
dispersion relation can be linearized Eq. 2.17. Fig. 2.9 shows the energy dispersion
relation and the so called Dirac cone. It is also important to note that no band gap
appears in between the valence and conduction bands. The continuity between the
two bands, with a very low density of states around the Dirac point, is the reason
for the classification of graphene as a semimetal.

E± (q) ≈ ±vF | q | +O[(q)2 ] (2.17)

Here, q is the momentum measured relatively to K and K 0 and O[(q)2 ] stands for
the expansion to the second order about the Dirac point. vF was found equal to
106 m/s [32].

The linear dispersion relation implies a vanishing effective mass (m∗ ) at the Dirac
point. From the relation between carrier mobility (µ) and effective mass [39], an
infinite mobility is predicted in the ideal case (Eq. 2.20b). Nevertheless, measured

27
Figure 2.9: Dispersion relation in the first Brillouin zone (left) with the energy in
units of t calculated for t = 2.7 eV and t0 = −0.2t. (Right) Zoom of the energy
dispersion relation at the Dirac point (K 0 ). Picture from [37].

mobilities reached 140000 cm2 /(V s) at room temperature [10] due to nonidealities.

µ = eτ /m∗ (2.18)

Here, τ is the average scattering time and e the charge of an electron.

The density of states can be precisely calculated from Eq. 2.16 to find that it is
also symmetric with respect to energy around the Dirac point if the hopping energy
t0 is zero. Despite its actual value is not very well known, some works have shown
values of 0.2t ≤ t0 ≤ 0.2t [40]. From eq. 2.17, an approximated expression can be
found for describing the density of states close to the Dirac point.

2 |E|
ρ(E) = (2.19)
π (~vF )2

The expected value of occupied states can be calculated by integrating the density
of states multiplied by the occupation probability (i.e. Fermi-Dirac distribution)
over the energy of the conduction (valence) band to find the number of conducting
electrons (holes).
Z ∞
n(T ) = fF Dn (E)ρ(E)dE (2.20a)
0
Z 0
p(T ) = fF Dp (E)ρ(E)dE (2.20b)
−∞

28
The Fermi-Dirac distribution for electrons and holes is defined as:
 E − EF
−1
fF Dp = 1 − fF Dn = 1 − exp +1 (2.21)
kB T

(a) (b)

Figure 2.10: Dispersion relation at the Dirac point and Fermi-Dirac distribution
centered at the Fermi energy. a) Diagrams corresponding to a non-doped graphene,
with Fermi energy equal to the energy at the Dirac point. b) Diagrams for an
n-doped graphene.

Fig. 2.11 shows the dependence of the carrier density with the Fermi level at
room temperature. Graphene presents a lower conductivity around the Dirac point,
where the density of states is lower. In addition, it presents an ambipolar behaviour
centered at the Dirac point. The combination of high mobilities with an efficient
modulation of the charge carrier density makes graphene an excellent candidate for
the realization of field-effect transistors for sensing [41].
The density of charge carriers (n) in a graphene field-effect device can be approxi-
mated by Eq. 2.22 in which the carrier density is calculated through the capacitive
coupling with the gate [42].:

q
n(U ) ≈ n20 + [Cint (Ugs − UD )/e]2 (2.22)

UD = ED /e is the potential at which the density of charge carriers is the minimal

(n0 ) [43]. This minimum charrier density arises from disorder and thermal excitation
[44][45].
Nevertheless, the carrier density in graphene does not increase linearly with an ap-
plied voltage. Indeed, a quantum capacitance which corrects this tendency has to
be taken into account in series with the double layer capacitance. The quantum
capacitance is an effect common to all parallel plate capacitors based on semicon-
ductor materials with low density of states. This effect arises from the low density of
states which prevents the ”normal” charging of the semiconductor (normal meaning

29
Figure 2.11: Integrated charge carrier density for electrons and holes. Note the
low density of carriers and the symmetric total density of carriers around the Dirac
point.

equal to the charging of metals). The quantum capacitance was first described for a
2-dimensional electron gas (2DEG) [46], but its calculation can be extended to other
systems such as graphene [47][48]. Eq. 2.23 stands for the quantum capacitance in
graphene under the condition eUQ  kT [48]:

s 2
2e2 eUQ
CQ ≈ √ √ + nimp (2.23)
~vF π ~vF π

Here, UQ is defined as UQ = Ugs Cdl /(Cdl + CQ ) (i.e. the voltage that drops in the
quantum capacitance) and nimp represents the effective density of charged impurities
[48]. At high voltages, Ugs drops mainly in the double layer capacitance, therefore
the increase of UQ with Ugs decreases for high polarizations. Fig. 2.12 shows the
two contributions to the total interface capacitance. It is possible to see how CQ
does not increase linearly with Ugs .

Knowing the definition of the quantum capacitance, the capacitance of the graphene-
electrolyte interface in Eq. 2.22 can be defined as the double layer capacitance in
series with the quantum capacitance:

Cint ≡ CQ Cdl /(CQ + Cdl ) (2.24)

30
Figure 2.12: Contributions to the interface capacitance, calculated from the previous
model. The double layer capacitance is considered constant.

From Eq. 2.22 and Eq.2.23 the conductivity of solution-gated graphene can be de-
scribed. Nevertheless, since the conductivity of the channel depends on the voltage
along the channel (Uch (x), see Eq. 2.25) its value is not homogeneous along the
channel. Therefore, the total resistance cannot be integrated because the conduc-
tivity depends on the potential and the potential on the conductivity. In order to
solve the voltage profile in the channel a numerical equation has to be employed
[49].

∆xIds
U (x + ∆x) = U (x) + p
2
(2.25)
eµW n0 + [Cint (U )[Ugs − U (x) − UD ]/e]2

Assuming a constant voltage along the channel, the transconductance of g-SGFETs

can be calculated. Merging Eq. 2.10 and 2.22:
W dn W
gm = Uds µe ≈ Uds µCint (Ugs )Ugs (2.26)
L dUgs L

From the previous equation, the importance of the coupling capacitance in the
transconductance of g-SGFETs has to be emphasized. The interface capacitance
of single layer graphene (SLG) with the electrolyte ranges from 1 to 4 µF/cm2 ,
presenting a minimum value around the Dirac point due to the effect of the quantum
capacitance. The combination of high mobilities with a relatively high capacitance
allowed for the fabrication of g-SGFETs with a transconductance of up to 4 mS/V,

31
almost an order of magnitude above Si based SGFETs [11].

2.2.2 Intrinsic noise in graphene transistors

As discussed in section 2.1.3, increasing the transconductance of the SGFETs is of

major importance to overcome the noise level of the current-to-voltage converter.
Nevertheless, increasing the transconductance will also increase the intrinsic noise
of the device. In this section, the various contributions to the intrinsic noise of
graphene SGFETs are detailed in order to show the exact advantage of increasing
the transconductance.
The sources of noise in electronic devices are numerous and in some cases of unknown
nature. The main sources are classified as thermal noise (also called white noise or
Nyquist noise), shot noise and flicker noise. Thermal noise appears in both passive
and active elements such as resistors or transistors from thermal fluctuation of the
charge carriers. Its power density spectrum is constant over the frequency spectrum;
this is the reason why it is also called white noise. Nyquist showed that this noise
source is equivalent to an q rms noise current source in series with the resistance
with an amplitude In = 4kBRT ∆f , where R stands for the resistance and ∆f for
the bandwidth [50]. Thermal noise in g-SGFETs can be calculated for reasonable
channel resistances (e.g. for 1kΩ, SI = 10−23 A2 /Hz, where SI stands for the
current power spectrum). It will be shortly shown that this amplitude is far below
the contributions from other sources.
Another typically important contribution to noise in electronic devices is the shot
noise which arises from the discrete nature of charge carriers. The power density of
shot noise is proportional to the current amplitude and for typical g-SGFET current
values it is well below other contributions (i.e. SI = 10−22 A2 /Hz for a current of
300µA).
Finally, in the frequency range of interest for the recording of neural activity (<
10kHz), flicker noise dominates the current power spectrum. The current power
spectrum can be expressed as:
1
SI = a β (2.27)
f

Where β usually approaches 1.

1/f noise appears in many kinds of devices and its nature is a controversial topic.
In graphene, 1/f noise has been attributed to both charge fluctuations and mobility
fluctuations [51]. In the following discussion the focus is centered on a charge-noise
model for solution-gated graphene transistors [52].
In this model, two main contributions to the flicker noise are distinguished; the
first arises from charge fluctuations in the graphene-electrolyte interface and the

32
second from a voltage independent resistance (Rs ) [53]. The former dominates at
low carrier density while the latter dominates at high carrier density. Regarding the
first term, the charge fluctuations (δq) induce a voltage noise (en ) in the graphene
channel through the capacitive coupling (en = cδq int
). Here, the total capacitance
of the graphene-electrolyte interface cint = Cint A is proportional to the area of the
device (A). On the other hand, δq is proportional to the square root of the area[52].
In turn, these voltage fluctuations induce a current noise which can be calculated
from the transconductance of the device. In the second term, the current noise for
the resistance in series is SI = AS Ids2 . Considering the dependence of the noise

amplitude on Uds and normalizing by the ratio RS /Rtrt , SI = αS Ids 4 [52] [54] where

αS is taken as free parameter of the model. Then, the parameter a of Eq. 2.27 can
be written as:

 2
δq 4
a = SI (1Hz) = gm + αS Ids (2.28)
cint

If δq is assumed unavoidable or intrinsic to the system, the S/N ratio is maximized

if the first term dominates the current power spectrum.
2
gm
S/N = δq 2
(2.29)
4 +S
cint gm + αS Ids I,amp

Here, SI,amp equals the contributions to the noise from the amplification stage. gm
is proportional to Uds , thus increasing Uds is essential for pushing the signal noise
above the amplifier noise. Nevertheless, Ids is also proportional to Uds , therefore
if it is increased above a certain limit the noise is not dominated by the amplifier
but by the second term of Eq. 2.28. Fig. 2.13 displays the S/N vs Uds at 1 Hz for
typical transistor and voltage-to-current converter parameters.

Recalling the general expression of gm for g-SGFETs (Eq. 2.26), it is important to

keep µ, Cint and the geometrical factor W/L as large as possible. Otherwise, the
4 which can strongly reduce the S/N .
optimal Uds implies a large αS Ids

33
Figure 2.13: Effect of Uds on the noise. The amplitude of the signal to be recorded is
considered 1 mV, the resistance of the I/U converter (Rg ) is set to 10 kΩ, δq/cint =1.2
10−10 V and αS = 0.022 A−2 (from [55]), W = 20 µm, L = 10 µm, and the gm =
3 mS/V. Finally, the sheet resistance of graphene (Rsh = 2 kΩ). In blue, S/N values
calculated from Eq. 2.27. In black, SI values calculated from the Eq. 2.28.

2.2.3 Graphene-based materials for capacitive stimulation

Despite its excellent properties for the realization of SGFETs, the suitability of
single layer graphene for stimulation electrodes is questionable. As discussed in
section 2.1.4, materials for extracellular stimulation must be inert and present large
interface capacitances. SLG presents a very inert interface with physiological solu-
tion but its capacitance is low in comparison with commonly used materials. This
low capacitance leads to a relatively low charge injection capacity (Qinj ). Table
2.2 shows a summary of commonly used or promising materials for extracellular
stimulation, including the results for SLG obtained in this thesis.
In order to decrease the interface impedance of electrodes, a recurrent strategy is to
increase the specific surface area of materials by increasing the porosity or roughness
[6]. This strategy has also been employed for the fabrication of supercapacitors
for high power energy storage devices [56]. Interestingly, carbon based materials,
including graphene [57][58], have been in the focus of attention. The materials
developed for other applications are valuable in the field of neurotechnology.
In this thesis, two graphene based porous materials developed by collaborators from
the Catalan Institute of Nanoscience and Nanotechnology (ICN2) and the Italian

34
 Material Mechanism Qinj (mC cm-2 ) EW (V) vs Ag/AgCl
Pt and PtIr alloys Faradaic/capacitive 0.05-0.15 -0.6 to 0.8
Activated IrOx Faradaic 1-5 -0.6 to 0.8
Thermal iridium oxide Faradaic ∼1 -0.6 to 0.8
Sputtered IrOx Faradaic 1-5 -0.6 to 0.8
Tantalum/Ta2 O5 Capacitive ∼0.5
Titanium nitride Capacitive ∼1 -0.9 to 0.9
PEDOT Faradaic 15 -0.9 to 0.6
SLG Capacitive 0.0018 -0.4 to 0.7

Table 2.2: Summary of the charge injection capacity for common or promising
materials for neural stimulation [6]

Institute of Technology (IIT) will be characterized and their suitability for neural
stimulation assessed.

35
3 Modelling

In this thesis, special attention is paid to the modelling of graphene-based devices

and more precisely to the implications of geometrical characteristics on the per-
formance of the devices. The study is based both on the derivation of analytical
expressions that describe relevant phenomena and the definition of models based on
numerical methods.
It is interesting to highlight the advantages and drawbacks of the two approaches;
on the one hand, the analytical expressions are easy to handle and provide good
understanding of the phenomena under study. In this case, the steps from funda-
mental physical laws to the final expression consist of few understandable algebraic
calculations. The final expression is an equation which can predict the behaviour
of a system with a single calculation. On the other hand, the analytical expres-
sions become more difficult to derive as the complexity of the system increases,
stretching the ability of the scientist who aims to solve the problem. The numerical
methods provide means for solving much more complicated problems by increas-
ing the number of computations required. Nevertheless, numerical methods do not
provide exact solutions but mere approximations, they provide little insight into
the nature of the solution and for every slight modification of the conditions a new
costly computation is required.
In the following sections, the two aforementioned approaches are exploited. The nu-
merical methods are used when the analytical expressions cannot be derived. This is
the case when the geometrical characteristics of the problem make the problem ex-
cessively complicated or when, due to crossed dependencies, an iterative calculation
is necessary. Furthermore, simulations based on numerical methods represent an
important tool for supporting the derivation of analytical expressions, showing the
expected behaviour before doing real experiments. The simulations are performed
using Comsol Multiphysicsr , a software based on the finite elements method (FEM)
[59].

3.1 Graphene-Electrolyte interface

The development of graphene-based devices for sensing in aqueous environments

requires a good description of the graphene-electrolyte interface. In section 2.2,
the current understanding is presented, showing the challenges that the crossed
relations between quantum capacitance, Fermi energy and channel conductivity
represent. This problem involving crossed relations is solved by using the Comsolr
environment. The FEM is based on the partition of the physical system, creating
a mesh of elements in which the differential equations can be solved using some
simplifications [60]. The solution in each element of the mesh has to be consistent
with the solution in neighbour elements. In this way, the relation found between
the quantum capacitance and the Fermi level is the only solution which is consistent

36
among mesh points and with the boundary conditions (i.e. potential at the transis-
tor terminals Ud , Us and Ug ). In Comsolr , the physical equations to be satisfied can
be imposed either on a domain or a boundary. The model developed in this thesis
is based on three main domains or layers (3.1). The first one consists of a dielec-
tric layer, with controlled conductivity and relative permittivity. It reproduces the
behaviour of a parallel plate capacitor with a capacitance equals Cdl . The second
layer is stacked on the first one and it accounts for the quantum capacitance. In
this case, the electric permittivity is adjusted so that, given the thickness of the
layer, the electric capacitance is equal to the quantum capacitance. By defining the
dielectric layers in different domains, the potential drop in CQ and in Cdl can be
measured independently. UQ influences the value of the CQ (Eq. 2.23 and Ugs the
conductivity of graphene by defining the density of carriers (Eq. 2.22). The in-plane
conductivity of graphene is modelled by a third layer or domain, the conductivity
of which depends is proportional to the charge carriers density. The potentials UQ
and Ugs are measured in each mesh element independently, therefore the interface
capacitance and conductivity along the graphene channel are non-homogenous.

Figure 3.1: Structure of the domains defined in Comsolr for simulating the
graphene-electrolyte interface. The three layers defined for modelling the graphene
are distinguished by colours, Cdl (purple), CQ (red) and n(Ugs ) (green).

The DC response predicted by the model can be validated by another method of

numerical analysis developed in Matlabr1 ([61]). Here, the simulation is based on
the bisect method2 . The channel is divided in n parts and the quantum capacitance
is set at the boundary (source, x = 0) where the polarization is equal to (Ugs ).
Given this value of CQ , the Fermi level in the second partition (x = 0 + ∆x) can
be calculated and the process is repeated until the solution for the last division is
found. Then, the potential in the channel at the final partition (Uch (x = L)) is
compared with the boundary condition at the end of the channel (drain voltage).
If the solution does not match with the conditions, the process is repeated with a
slightly different initial quantum capacitance until a matching solution is found.

1
Model developed by Eduard Masvidal, co-worker at the CNM (Barcelona)
2
See the Matlabr function bisect

37
The solution found by this method has been compared with the FEM based calcu-
lation. In Fig. 3.3a, the DC response of a g-SGFET (Ids vs Ugs ) is compared for
the two models, showing a perfect matching between the two calculation methods.
The simulations shown in Fig. 3.3a as well as the rest of simulations presented in
this thesis are performed for the graphene properties shown in table ?? (unless the
opposite is specified).

Figure 3.2: Numerical calculation of the Ids vs Ugs curve for Uds = 0.1 V and the
parameters in table 3.1. It can be seen that both the FEM and the bisect based
method lead to the same solution

Parameters Values
Mobility 1200 cm2 /Vs
Double layer capacitance ( μF/cm2 ) 3 μF/cm2
Density of impurities (nimp ) 1 .1011 cm-2
Minimum number of carriers (n-11 ) 0.7 .1012 cm-2
Dirac voltage (VD ) 0V
Contact resistance (RD ) 0.003 Ωm

Table 3.1: Parameters employed for the simulation of graphene devices

38
An interesting prediction has been derived from these models, which cannot be easily
measured experimentally. Due to the non-homogeneous quantum capacitance along
the channel, the conductivity is also non-homogeneous. This leads to a non-linear
drop of the Fermi level along the channel. In Fig. 3.3a the potential in the channel
(Uch ) is shown for different polarization voltages (Ugs ) and in Fig. 3.3b the potential
in the centre of the channel (Uch (x = 15 µm)) vs Ugs .

Fig. 3.3b shows the abrupt change of potential in the centre of the graphene channel
that occurs around the Dirac point. When Ugs − UD = Uds , the polarization of the
channel with respect to UD at the drain is zero (Ugs − UD − Uds = 0) while at
the source it is equal to Ugs − UD . In this scenario, the conductivity close to the
source is much larger than at the drain, and therefore Uch (x) drops rapidly close
to the drain and flattens around the source. Alternatively, if Ugs − UD = Us = 0,
the opposite case occurs, Uch (x) drops abruptly close to the source and it flattens
around the drain. Therefore, the potential at the centre of the channel is close to
the drain voltage. If Ugs is far from Us or Uds , the polarization is approximately
constant along the channel and the voltage drop along the channel is approximately
linear (see Fig. 3.3a). Note that if the Dirac point is not at zero, the described
non-linearity is centered around UD .

(a) (b)

Figure 3.3: Potential distribution in the channel for Uds = 0.2 V. a) Uch vs channel
position simulated for different polarization voltage with Matlabr and Comsolr .
b) Potential in the centre of the channel (Uch (x = 15 µm)) plotted against the
polarization potential (Ugs ).

The symmetry around the Dirac point arises from the ambipolar behaviour of
graphene. If Ugs − UD equals Uds (for Uds > 0) the majority charge carriers are
electrons (Fig. 3.4 top), but if Ugs − UD equals Us the majority charge carriers
are holes (3.4 bottom). In Fig. 3.4, the Fermi energy is depicted at different posi-
tions along the channel. Keeping in mind the relation between carrier density and
conductivity, the non-linear voltage drop shown in Fig. 3.3 can be understood.

39
Figure 3.4: The dispersion relation around the Dirac point is shown at different
positions in the channel (at the source, drain and centre of the channel). It is
possible to see the origin of the symmetry between the Ugs values ( Ugs − UD = Ud
and Ugs − UD = Us ). Here UF ≡ EF .

3.2 Frequency response of electrolyte-gated devices

While the model in Matlabr is specially oriented to solve the DC response of g-

SGFETs, the FEM model is not limited to time independent problems. In Comsolr ,
simulations can be run for different kinds of problems such as stationary, time
dependent, as well as for harmonic oscillations or small signal perturbations. This
flexibility of the software can be exploited to explore the AC response of the devices.
In the present work, the numerical analysis was useful for guiding the derivation of
analytical expressions and their subsequent validation.
Graphene presents a relatively high sheet resistance (Rsh ) in comparison with bulk
electrodes. When graphene is contacted in the electrode configuration, the Rsh
has a significant effect on the impedance spectrum of the electrode. The equivalent
circuit that describes the impedance of a graphene electrode is a distributed elements
circuit (Fig. 3.5). Charge carriers flow along the graphene sheet before charging the
distributed capacitive elements; the sum of the resistances through the graphene
from the contact to the capacitive element in series with the impedance of the
capacitive element constitute the total impedance of a certain current path. At very
low frequencies, the impedance of the current paths is dominated by the capacitive

40
elements. Thus, all paths contribute equally to the total current (Fig. 3.5 top).
Increasing the frequency leads to a reduction of the impedance of the capacitive
elements with respect to the resistive elements along each current path. As a result,
the different paths do not contribute equally to the current (Fig. 3.5 middle).
The higher the frequency, the more relevant this effect is, and the current will
preferentially flow through current paths that are closer to the contact (Fig. 3.5
bottom).

Figure 3.5: Schematic representation of the AC response of graphene electrodes.

The black circuit elements represent the elements through which most of the cur-
rent flows, whereas the gray elements represent those through which the current is
comparatively low. From the still unpublished paper [62].

The potential and the current at each point in this equivalent circuit can be sim-
ulated by Comsolr , but it can also be described by an analytical expression. The
equations for two different electrode geometries were derived, namely for a rectangu-
lar and a circular electrode design. Fig. 3.6 shows the equivalent circuit with special
emphasis on the boundary conditions and model definitions. Fig. 3.6, describes a
differential of the total sheet resistance (rsh ) projected on one dimension. It is
defined differently for the two geometries; rsh [Ω] = Rsh dx/W for the rectangular
geometry (where W is the width of the graphene sheet) and rsh [Ω] = Rsh dr/2πr for
the circular geometry. The differential of impedance zint ), can be expressed through
the interfacial impedance per unit area3 (Zint ). Similarly as discussed above, the dif-
ferential impedance is defined differently for the rectangular and circular geometries.
Zint Zint
In the rectangular case, zint [Ω] = W dx and for the circular geometry zint [Ω] = 2πrdr .
The equations governing these systems are derived for the general case, in which the
impedance of the interface can be represented by any equivalent circuit. Relevant
3
Zint is an intensive magnitude with units of Ωm2 .

41
cases, which will be used for fitting the experimental data, include modelling the
interface using a constant phase elements (CPEs) and CPEs in parallel with charge
transfer resistance elements (Rct ). Constant phase elements account for non-ideal
capacitive behaviour of heterogeneous electrode/electrolyte interfaces [63].

Figure 3.6: Equivalent circuit for the graphene electrodes; including the boundary
conditions (U (x = L) = Upert ) and (dU (x)/dx x=0 = dU (r)/dr r=0 = 0) where L is
the length of the electrode in the rectangular case and the radius of the electrode in
the circular case. The differential elements of the total channel resistance rsh and
the differential elements of the total interface impedance zint are also shown.

The equivalent circuit shown in Fig. 3.6 is usually called the ladder circuit and it
is recurrent in electric engineering problems. The general solution for any Zint can
be found elsewhere [64][65] (see Eq. 7.7 in appendix 7.1 for a full derivation). The
system of equations which governs the system allows to solve the potential between
the channel and the reference electrode (U (x) and U (r)). Here, x represents the
position along rectangular graphene sheets and r the distance from the centre in
circular electrodes.
In the case of circular electrodes, the sheet resistance depends on the distance to the
centre, nevertheless Rsh and Zint are considered homogeneous along the graphene
sheet. The two equations that describe the system can be written as follows:

2πr dU (r, ω)
i(r, ω) = − (3.1a)
Rsh dr

di(r, ω) 2πr
=− U (r, ω) (3.1b)
dr Zint (ω)

By merging eq. 3.1a and eq. 3.1b, the following transformed version of the Bessel
equation which was given by Bowman [66] is obtained:

42
 d2 U (r, ω) dU (r, ω) Rsh 2
r2 2
+r − r U (r, ω) = 0 (3.2)
dr dr Zint (ω)
The solution is a linear combination of first order Bessel functions of the first (Jn )
and second (Yn ) kind:

U (r, ω) = C1 J0 (br, ω) + C2 Y0 (br, ω) (3.3)

Where b2 ≡ −Rsh /Zint

The coefficients C1 and C2 can be determined by using the boundary conditions.

The change of potential at the centre of the electrode is zero and the potential at
the metal contacts (r=L) is equal to the amplitude of AC signal (Upert ).

dU (r, ω)
=0 (3.4a)
dr r=0

U (L, ω) = −Upert (ω) (3.4b)

Using the following property;

dY0 (br, ω)
lim =∞ (3.5)
r→0 dr

allows to determine the coefficients C1 and C2 .

C2 = 0 (3.6a)

−Upert (ω)
C1 (ω) = (3.6b)
J0 (bL, ω)

Hence, the solution to the differential equation is:

−Upert (ω)J0 (br, ω)

U (r, ω) = (3.7)
J0 (bL, ω)

The impedance of the electrode can be then derived by dividing the input voltage
by the current through the contact (I(0, ω)):

−Upert (ω)Rsh −Rsh J0 (bL, ω)

Ztot (ω) = + Rs = + Rs (3.8)
2πr dUdr
(r,ω)
r=L 2πLbJ1 (bL, ω)

43
 dJ0 (br,ω)
Here dr = bJ1 (r, ω) was used.
The impedance expressed in Eq. 3.8 takes into account the series resistance (Rs )
from contacts, wires and access region (i.e. region of the channel coated by an
insulating layer). The impedance spectrum is strongly dependent on the value of
this series resistance. The first term in Eq. 3.8, which accounts for the impedance
of the distributed elements circuit, is never purely real or imaginary. When the
frequency goes to zero, the imaginary part dominates over the real part (Fig. 3.7a).
As the frequency increases, the imaginary part of the impedance decreases. When
the value of the imaginary part approaches the value of the real part, both decrease
with frequency, following the same dependence. The real part diminishes due to
the shortening of the effective channel (see Fig. 3.5) and the imaginary part due
to the decrease with frequency of the capacitive impedance. This effect is trans-
lated into a constant phase of 45o above a certain cut-off frequency. However, the
series resistance adds a constant pure real term to the impedance which dominates
the other terms pushing the phase back to 0o at high frequencies (Fig.3.7b). The
contribution of the sheet resistance of graphene at high frequencies makes possible
to extract the sheet resistance of graphene from an electrode impedance spectrum.
The expression derived for the impedance of circular electrodes was used to fit the
experimental data presented in the unpublished paper [62] and shown in appendix
7.3. The experiments performed in this thesis were based on rectangular electrodes.
Therefore, the equations employed from here on correspond to the rectangular case
(see appendix 7.1). Fig. 3.7a shows the comparison between the impedance spec-
trum obtained using the analytical expressions and the results obtained using the
FEM model. Both calculation methods present very similar results. Besides, Fig.
3.7b displays the phase of the impedance for different Rs , showing the strong effect
of this parameter on the impedance spectrum.
It is important to evaluate the possible impact of the distributed elements on extra-
cellular stimulation since it implies that current injection is not homogeneous along
the graphene sheet. At high frequencies, most of the current is injected through the
edges of the electrode. The current injected (di(x)/dx of Eq. 7.2) can be derived
by merging Eq. 7.4 and 7.6. Fig. 3.8a shows the module of the current injected
into the electrolyte versus the position x along a rectangular electrode for different
frequencies. Fig. 3.8b shows the potential along the channel which is proportional
to the injected current (U (x) = −di(x)/dxW/Zint ). The calculations correspond to
a graphene electrode of width W = 4 mm and a length L = 6 mm.

q q
dix W  −
Rsh
x
Rsh
x

iz = =− C1 e Zext + C2 e Zext (3.9)
dx Zext

In addition to the impedance in the electrode configuration, the impedance from

drain to source in the transistor configuration can be computed analytically or
numerically. The equivalent circuit which describes the impedance of the channel
is also a distributed elements circuit. Charge carriers flow along the graphene,

44
 (a) (b)

Figure 3.7: Impedance spectrum of graphene electrodes calculated using both the
analytical expressions and FEM model. a) Real and imaginary parts of equation
3.8 and module considering a Rs equal to zero. The calculated values are validated
with the results of the FEM simulations. b) Phase of the impedance for Rs = 0 Ω
using the analytical expression and the FEM model, phase for Rs = Rsh /100 and
for Rs = Rsh /10.

(a) (b)

Figure 3.8: Current and voltage along a graphene electrode. Calculations performed
for Upert = 10 mV and different frequencies. a) Module of iz along the channel. b)
U (x)/Upert along the channel.

however, by charging the distributed capacitive elements, part of the current can
flow through the electrolyte towards the opposite terminal. At low frequencies, the
impedance of the capacitive elements is very high and no current flows through
the electrolyte (Fig. 3.9 top). Increasing the frequency leads to a reduction of the
impedance of the capacitive elements with respect to the resistive paths through
graphene. As a result, paths with a smaller resistive component contribute more to
the total current (Fig. 3.9 centre). With increasing frequency, this effect becomes

45
more relevant, and at some point most of the charges cross the electrolyte-graphene
interface close to the contacts and flow through the electrolyte to the other contact
after crossing back the electrolyte-graphene interface (Fig. 3.9 bottom).

Figure 3.9: Schematic representation of the g-SGFETs AC response. The black

circuit elements represent the current paths through which the current is high in
comparison with the rest of the elements. Additionally, the gray elements cor-
respond to current paths for which the current is comparatively low. From the
unpublished paper [62].

An expression for the impedance can be derived for the general case in which the
interface impedance (Zint ) can be described by any equivalent circuit.

The voltage and current must satisfy the following equations:

Rsh dU (x, ω)
i(x, ω) = − (3.10a)
W dx

W di(x, ω)
U (x, ω) = − (3.10b)
Zint (ω) dx

By merging Eq. 3.10a and Eq. 3.10b, the following differential equation is obtained
which defines U (x, ω).

Rsh d2 U (x, ω)
U (x, ω)2 − =0 (3.11)
Zint (ω) dx2

46
Figure 3.10: Equivalent circuit for the graphene electrodes; the boundary conditions
are specified and the variables and parameters defined.

The solution of this differential equation can be written as follows:

r r
R Rsh
− Z sh(ω) x Zint (ω)
x
int
U (x, ω) = C1 (ω)e + C2 (ω)e (3.12)

From the boundary conditions (eq. 3.13a and 3.13b):

U (x = 0) = 0 (3.13a)

U (x = L) = −Upert (ω) (3.13b)

The integration constants can be derived:

−U pert(ω)
C1 (ω) = −C2 (ω) = p
(3.14)
2sinh Rsh /Zint (ω)L

The solution of the differential equation is therefore:

p

−Upert (ω)sinh Rsh /Zint (ω)x
U (x, ω) = p
(3.15)
sinh Rsh /Zint (ω)L

Similar to the electrode configuration, the impedance of the transistor is:

s !
Upert (ω)Rsh p R sh
Z(ω) = = Zint (ω)Rsh /W 2 tanh L + Rs (3.16)
W dU (x, ω)/dr x=0 Zint (ω)

It is worth to point out that Zint (ω)Rsh /W 2 = Zint (ω)/(W L)Rsh L/W = Zext Rch .

47
Here, Zext [Ω] and Rch [Ω] are the interface impedance and channel resistance of the
entire graphene sheet. On the other hand, the term in the hyperbolic tangent can
also be expressed in terms of the extensive properties Zext [Ω] and Rch [Ω]:

s v
L
r
u Rsh W
u
Rsh 2 2
Rch
L = t Z (ω) L = (3.17)
Zint (ω) int Zext
(W L)

The impedance spectrum modelled by Eq. 3.16 strongly depends at high frequencies
on the value of the series resistance. As in the electrode case, the first term, which
accounts for the impedance of the distributed elements circuit, is never purely real
or imaginary. Considering the interface graphene-electrolyte as a pure capacitor,
when the frequency tends to zero, the imaginary part goes to zero while the real
part approaches the channel resistance value. At high frequencies, both real and
imaginary parts decrease: the real part due to the opening of alternative paths
through the electrolyte and the imaginary part due to the expected decrease of the
capacitive impedance with frequency. Again, at high frequencies both take the same
value and decrease with frequency (Fig.3.11a). Therefore, a phase of 45 o above a
certain cut-off frequency is also expected for this system. Nevertheless, the series
resistance adds a constant pure real term to the impedance, dominating over the
other terms at high frequencies and pushing the phase back to 0 o . In contrast to
the electrode configuration, in the transistor configuration the balance between the
effect of the distributed elements circuit and the contact resistance leads to a peak
of the phase (Fig. 3.11b). This peak occurs at the same frequency as the drop of
the module, which goes from a resistance corresponding to the channel plus series
resistances to a resistance that corresponds to the sum of the electrolyte and the
series resistance. Fig 3.11 displays the impedance calculated using both the FEM
model and the analytical expression, showing a close matching between them.
Hence, from the frequency dependent impedance in the transistor configuration, it
is possible to obtain not only the resistance of the channel but also the capacitance
of the interface with the electrolyte. Furthermore, it is possible to distinguish the
contact resistance from the channel resistance.
The derivation of the transistor impedance is based on the same assumptions as the
derivation of the electrode impedance. The resistance of the electrolyte is neglected
and the channel is considered homogeneous both in the electrode and transistor con-
figurations. The FEM based model can be used here to validate these assumptions.
Fig. 3.12 shows the impedance spectra calculated by the FEM based model in which
the effect of the electrolyte is taken into account and compared with the effect of a
series resistance. In the electrode configuration, the electrolyte (with conductance
Ge ) is connected in series between the graphene electrode and the reference.
In the transistor configuration, the impedance at low frequencies is composed by
the contact and channel resistances, but at high frequencies it approaches the series

48
 (a) (b)

Figure 3.11: Effect of the Rs on the source to drain impedance spectra. a) Real
and imaginary parts of transistor impedance considering an Rs equal to zero. b)
Phase of the impedance given different Rs values.

(a) (b)

Figure 3.12: Effect of the electrolyte conductance, calculated using the FEM model.
The effect of a Ge = 0.7 S/m with Rs = 0 is compared with a resistance in series
Rs = 400 Ω with Ge = 100 S/m. The impedance spectra simulated for a low Ge =
100 S/m with Rs = 0 is shown as a reference. The module is shown in (a) and the
phase in (b).

resistance. Fig. 3.13 shows the impedance spectra calculated using the FEM based
model for the different Ge and Rs values. It can be observed how the effect of contact
and electrolyte resistances can be confused in the phase spectrum. Nevertheless,
the effect on the impedance module can be clearly distinguished. The analytical
model attributes the constant impedance found at high frequencies to the contact
resistance. In order to fit the impedance modulus at low frequencies, the channel
resistance is underestimated to compensate the electrolyte resistance and, therefore,
the capacitance is overestimated for fitting the data close to the cut-off frequency.

49
 (a) (b)

Figure 3.13: Effect of the electrolyte conductance on the source-to-drain impedance.

The effect of a Ge = 0.7 S/m with Rs = 0 is compared with a resistance in series
Rs = 2 kΩ with Ge = 100 S/m. Simulated values for a low Ge = 100 S/m with
Rs = 0 is shown as a reference. The module is shown in (a) and the phase in (b).

The response of the transistor in terms of channel impedance is not directly related
with the sensitivity of the device. The magnitude used to quantify the sensitivity is
the transconductance. Its evolution with frequency can be studied analytically or
simulated in Comsolr . From the definition of transconductance stated before (Eq.
2.10), and including the expression for the charge carrier density in graphene (Eq.
2.22), the transconductance can be rewritten as follows:

s  2 !
W d Cint Uint
gm = Uds µe n20 + (3.18)
L dUgs e

If the polarization voltage (Ugs ) is large, the term n0 can be neglected. In this
regime, the transconductance is proportional to Uint , which at the same time is
proportional to the voltage divider comprised of the interface impedance and the
resistance in series (i.e. resistance of contacts, electrolyte and graphene sheet).

d Zint (ω)
gm ∝ Cint Ugs (3.19)
dUgs Zint (ω) + Rs

For a purely capacitive interface (Zint (ω) = 1/jωCint ), the transconductance drops
with frequency after the cut-off frequency given by Eq. 3.20:

1
ωc1 = (3.20)
Rs Cint

However, as Zint decreases, the capacitive current through the interface increases.

50
This current contributes to the total current collected in the drain; thus, for all
practical purposes, it is equivalent to the current due to the modulation of the
eq
channel. Therefore, an equivalent transconductance (gm ) can be defined taking this
capacitive current into account:

eq dG 1
gm = Uds + (3.21)
dUgs 2Zint

Here, G is the channel conductance. The second term accounts from the direct
injection of current through the interface4 . The factor 1/2 in the second term
comes from the equal distribution of current from gate to source and gate to drain.
At low frequencies, the effective transconductance is dominated by the modulation
of the channel (showing the advantage of SGFETs over conventional electrodes).
Nevertheless, when the frequency is high enough, the effective transconductance is
predominantly caused by the capacitive currents. The transition frequency from
one regime to the other is defined as the second cut-off frequency ωc2 by equalizing
the two terms of the right-hand side of Eq. 3.21.

Uds µ
ωc2 = (3.22)
L2

These observations can be verified by making use of the FEM based model. Fig.
3.14 shows the effective transconductance (solid lines) and the contribution of the
capacitive current (dashed lines) for three different values of L. It can be seen that
the current contribution through the electrolyte saturates at high frequencies, when
the impedance is dominated by Rs . Additionally, ωc1 can be observed for small
channel lengths, when ωc1 < ωc2 .

4
For reasons of clarity, the effect of Rs is neglected.

51
Figure 3.14: Effective transconductance (solid lines) and its contribution from the
capacitive current (dashed lines) calculated for three different channel lengths (6mm,
0.6 mm and 0.06 mm) using the FEM model. ωc2 is observable for the two longer
lengths and ωc1 for the shortest. Uds = 0.1 V and Ugs = 0.5 V were used for these
simulations.

3.3 Dependence of transconductance on device geometry

Achieving a high transconductance was shown to be essential for preamplifying the

signal above the amplifier noise level. In addition, this transconductance cannot
4
be accomplished by simply increasing Uds because of the intrinsic noise has a Uds
dependence. This section aims to explore the dependences between the transcon-
ductance and the geometry of the devices.

3.3.1 Optimization of graphene transistors

Given certain graphene properties (i.e. µ, nimp , Cdl or n0 ), the transconductance

of the device can only be improved by increasing the W/L ratio. The dimensions
of the cells constrain the W value, leaving L as the only parameter to be changed.
From the definition of transconductance given before (Eq. 2.10), an infinitely small
L leads to an infinite transconductance. Nevertheless, the effect of the resistance in
series (Rs ) must be taken into account. If the channel is too short, the resistance
of the transistor is dominated by Rs , which screens the modulation of the channel
conductance (Ggr ). In this section, an expression for the optimal channel length is
derived as follows:

52
 d Ggr 1/Rs
gm = Uds (3.23)
Ugs Ggr + 1/Rs
 
dGgr 1 Ggr Rs
= Uds −
dUgs Rs Ggr + 1 (Rs Ggr + 1)2
(3.24)

Eq. 3.24 can be rewritten so that the dependency with L is made explicit. Rewriting
Ggr as σ W
L , where σ is the conductivity of graphene, leads to:

 Rs σW 
W dσ 1 L
gm = Uds Rs σW
−
2 (3.25)
L dUgs L +1 Rs σW
+ 1
L

The optimal length (Lopt ) can be defined as the length which maximizes gm :

dgm ! −1 2Rs σW
=0= 2 2
+ (3.26a)
dL (Rc σW + Lopt ) + Lopt ) (Rc σW + Lopt )3

Lopt = Rs σW (3.26b)

Here, Rs has three contributions; from the tracks, from the graphene-Au interface
(∝ 1/W 2 ) and from the access region (∝ 1/W 2 ). From Eq. 3.26b, it can be
concluded that unless the resistance in series is dominated by the tracks resistance,
the optimal length does not increase significantly with W.
The previous expression has limited use as a design rule because both the channel
length and Ugs must be optimized simultaneously. At this point, the optimal Ugs
cannot be found analytically since there is no analytical expression describing the
conductivity of graphene with respect to polarization voltage. Nevertheless, the
simulations based on numerical methods allow to find the optimal values for a given
W and graphene properties.
The previously described model based on the bisect method [61] (section 3.1) was
used to solve the Ids vs Ugs curves and extract the maximum transconductance. This
process is combined with a channel length sweep from which the optimal Ugs and
L are obtained. In Fig. 3.15, the transconductance at the optimum Ugs is plotted
versus the channel length (L) and it is compared with the transconductance at one
fixed Ugs . The maximum transconductance obtained from Eq. 3.26b is indicated
by a dot in the graph, showing the close matching with the simulation in which Ugs
is fixed. Here, the transconductance is expressed in mS/V (i.e. normalized to Uds .
Changing the polarization voltage affects the conductivity of graphene and conse-
quently the optimum L. Adjusting the polarization voltage for each length allows

53
Figure 3.15: Transconductance for different channel lengths. Black: Transconduc-
tance obtained at the optimum polarization voltage. Red: Transconductance ob-
tained at Ugs = 0.1 V. The blue dot indicates the optimal length calculated by the
analytical model. For this simulation, W = 120 µm, the access region equals 5 µm
long and the tracks resistance is 50 Ω. The resistance of the Au-Graphene interface
is 0.003 Ωm. Here, the transconductance is expressed as in mS/V (transconductance
normalized to Uds )

to increase the transconductance significantly, especially when the channel length

is large and increasing the conductivity is not critical.

3.3.2 Circular transistors

In the previous section, the basic strategies for improving the sensitivity of the
transistors were discussed. An alternative approach for increasing the transconduc-
tance is to abandon the rectangular geometry so that the transistor can be folded
within the area of interest. In this way, the length of W can be increased without
extending the device out of the region of interest. This strategy can improve the
transconductance significantly if the optimum channel length is significantly smaller
than the width W .

In this section, the design of circular transistors is explored for this purpose. The
transconductance of circular and rectangular transistors is compared, keeping W
equal to the diameter of the circular transistors. Fig. 3.16 shows a picture of both
the circular and rectangular devices.

54
Figure 3.16: Scheme of circular (left) and rectangular (right) transistors. The two
transistors have an effective W = 2Rout .

To calculate the properties of the circular devices, the narrowing of the channel
around the inner contact has to be taken into account. The outer perimeter is
2πRout , while the inner perimeter is 2πRin . Since the conductance of the graphene
channel is proportional to W , it is smaller around the inner contact. To resolve this
issue, the model based on Matlabr presented before [61] was modified so that the
conductivity of each channel division is proportional to 2πR (where R is the radius
from the centre of the device). If the effective width of the channel was 2πRout = πΦ,
where the diameter Φ equals the width W of an equivalent rectangular transistor, a
transconductance π times larger than in the rectangular case is expected. However,
the limited radius of the inner contact and the fact that the outer contact must be
perforated (see Fig. 3.16), reduces the improvement to a factor always smaller than
π.

Fig. 3.17 shows the transconductance vs the channel length (Rout − Rin for the
circular and L for the rectangular transistors). The calculations are performed for
different transistor sizes and for both the rectangular and circular geometries. The
simulations show a visible increase of the transconductance in the circular case. The
optimum length for circular transistors is always smaller than for the rectangular
transistors, this shortening compensates the narrowing of the channel around the
inner contact.

55
Figure 3.17: Simulated transconductance for both circular (solid line) and rectangu-
lar (dashed line) transistors of different sizes. For all simulations the access region
equals 5 µm (for each contact) and the tracks resistance equals 50 Ω. The resistance
of the Au-Graphene interface equals 0.003 Ωm.

3.3.3 Effect of partial modulation in terms of transconductance

In order to detect neural signals from single or few cells, the size of the devices
must be in the same range as the size of the cells. In this section, the dependence
between the spatial scale of the sensor and the signal to be recorded are studied.

The definition of transconductance given before (Eq. 2.10) is only valid if the
entire channel is affected by the signal to be detected. Otherwise, Ugs has a spatial
dependence and the effective transconductance has to be calculated taking into
account the partial modulation of the channel. When the aim is to detect the
electrogenic activity of few neurons, the size of the devices has to be in the same
scale as the signal to be recorded, otherwise the S/N decreases.

In the following, the effective transconductance due to the modulation of a small

area in the channel is derived. Consider a graphene channel of width W and length

56
L, and that the signal is only applied to a region of dimensions w and l, with w < W
and l < L (Fig. 3.18a).
The effective transconductance can be calculated by dividing the channel in n x
m areas of dimensions w and l (see Fig. 3.18a). The partitions can be designated
by the superscript i (for their position along the channel) and j (for their position
perpendicular to the channel). The transconductance can be derived taking into
account that the partition which is modulated has n − 1 partitions in series and
m − 1 in parallel. Since the conductance of several elements in parallel equals the
sum of conductances:

m
dIds d d X ij
gm = = Uds G(Ugs ) = Uds G (Ugs ) (3.27)
dUgs dUgs dUgs
j

If only one partition is modulated, the last derivative is zero for all rows j except
one. The conductance in this row can be rewritten, taking into account that the
modulated element has n−1 elements in series. Defining the modulated region with
i = 1 (without loss of generality):

 −1
1 1 1 1 1
G(Ugs ) = + + ... + i + ... + n−1 + n (3.28)
G1wl (Ugs ) G2wl Gwl Gwl G wl

The previous equation can be rewritten as:

 −1
1 n−1 Gwl (Ugs )
G(Ugs ) = 1 + = Gwl (Ugs )
. (3.29)
Gwl (Ugs ) Gwl (n − 1) +1
Gwl

Combining equations 3.27 and 3.29 leads to:

 
dGwl (Ugs ) 1 n − 1 1
gm = Uds − 2
= gmwl 2 (3.30)
dUgs n n n

gmwl stands for the transconductance of an isolated graphene channel of dimensions

w and l (i.e. gmwl = Uds wl dU
dσ
gs
) . The factor 1/n2 describes the loss of gm produced
by the resistance of the elements in series with the modulated partition.
The previous calculation can be repeated for a more complex case, in which not
only one region of size w, l modulated but pj regions in each row j are affected by
the modulation (Fig. 3.18b).
Here, Eq. 3.27 applies as well, but dG(Ugs )/dUgs has to be recalculated. For that
purpose the dependence of G with Ugs has to be defined. Here, the modulated
partitions in each row receive the index k and the unmodulated ones are designated
with the index q:

57
 (a) (b)

Figure 3.18: Transistor of width W and length L divided in n columns and m rows.
a) Perturbation applied on a single region of w, l affected by the signal (the index
i and j denote the columns and rows respectively). b) The black squares represent
the areas where the signal is applied (n = 7, m = 9). In this case, more than one
region of size w, l is modulated (i.e. p4 = 4).

n,m pj
m X n−pj −1 m  −1
1 1 pj n − pj
Gij
X X X X
G(Ugs ) = wl (Ugs ) = + = +
i,j j k Gkj
wl (Ugs ) q Gqj
wl j
Gwl (Ugs ) Gwl
(3.31)
Following the same reasoning as before and by merging Eq. 3.27 with 3.31:

m m
d X Gwl (Ugs )Gwl X pj
gm = Uds = gmwl (3.32)
dUgs pGwl + (n − p)Gwl (Ugs ) n2
j j

Here, gmwl is the transconductance of an isolated graphene area of dimensions w x

l. Eq. 3.32 shows that the effect of applying a signal to different areas is additive.
If the whole channel is affected equally by the signal (∀j : pj = n), the expression
for the transconductance of a completely modulated device is recovered (Eq. 3.33):

n m mw d W d
gm = mgmwl = gmwl = µe σ= µe σ = gmW L (3.33)
n2 n n l dUgs L dUgs

The additive property shows that if the signal to be detected affected only one
region of area S, and the device had an area A, the effective transconductance of

58
the device for amplifying the signal would be gm S/A. The previous derivations
are based on the assumption that all the partitions have the same conductance
Gwl . This assumption is not always valid, for instance, in circular transistors the
narrowing of the channel around the inner contact leads to a smaller conductance in
this region. The additive property in this kind of devices can be validated using the
FEM model. Fig. 3.19a shows the structure of the Comsolr model for a circular
transistor divided in 5 partitions5 of same length (∆R). The modulation can be
applied either to each of them independently or to all of them simultaneously. Fig ??
shows the partial transconductance from each of the partitions and the sum of them,
which are compared with the total transconductance of the transistor calculated by
modulating the whole channel simultaneously. In addition, the previously derived
equations do not consider the effect of a resistance in series with the device, the
simulations presented in Fig. 3.19 are performed with a certain access resistance
(blue region in Fig. 3.19a). The results show the validity of the additive property
for channels of non-homogenous conductance and with a resistance in series.

(a)

Figure 3.19: Simulation using the FEM model of partially modulated non-
homogeneous channels. a) Structure of the channel in the Comsolr divided in
5 partitions (red) and with inner and outer access regions (blue). b) The additive
i represent the transconductance from the par-
effect in terms of gm is validated. gm
tition i and gmtot the transconductance obtained by modulating the entire channel

simultaneously.

3.3.4 Effect of partial modulation in terms of noise

In section (2.2.2), it was presented that the low frequency current noise has two
contributions; the first accounts for charge fluctuations at the graphene-electrolyte
5
The number of partitions is chosen arbitrarily, just for showing the additive or non-additive
property of gm .

59
interface and the second for the effect of a series resistance not affected by the polar-
ization. Here, the effect of the voltage fluctuations (δqwl /cwl ) applied on different
partitions is shown. From the additive property derived in the previous section,
the total current (Ids ) is the sum of the contributions from all parts. The voltage
fluctuations or voltage uncertainty on each of these regions is (δqwl /cwl ). The total
noise in Ids can be calculated from the propagation of uncertainty:

 2  2  2  2
dIds δqwl 11 1 dIds δqwl ij 1
SI = g + ... + g + ...
dI11 cwl mwl n2 dIij cwl mwl n2

dIds 2 δqwl nm 1 2
   
...+ g (3.34)
dInm cwl mwl n2

If the whole channel is affected by the charge fluctuations (∀j : pnj = n), Eq. 3.34
can be rewritten as:

 2
δqwl 1
SI = nm gmwl 2 (3.35)
cwl n

This expression is equivalent to the expression for the total device charge noise
presented in section 2.28 (see appendix 7.2).

Eq.3.35 is only valid if all the partitions have the same transconductance. If the
factor (δqwl /cwl )gmwl (1/n2 ) is not the same for all partitions, the total noise is not
proportional to the square of the sum of partial transconductances. Instead, it is
equal to the sum of squared partial transconductances (Eq. 3.36, this outcome is es-
sential for the calculation of the noise power density of devices with non-rectangular
geometries):

m X
X n m X
X n 2
2
SI ∝ (Gmij ) =6 Gmij (3.36)
j i j i

Eq. 3.34 can be also used to calculate the noise of transistors with a non-homogeneous
conductance. For this purpose, the partial transconductance of each partition has
to be known. The values presented in Fig. 3.19 can be used in order to show
that the contributions to the noise from different partitions in a non-homogenous
channel is non-additive (Eq. 3.36). The same simulation can be performed for a
rectangular transistor in order to show that in this case the partial contributions to
the noise can be added (Fig. 3.20a).
√ The amplitude of the voltage fluctuations can
be expressed as; δqwl /cwl ) = k/ S[V ] where S stands for the area of the partition.

60
 (a) (b)

Figure 3.20: The different contributions to the noise from 5 partitions of the channel
are shown for the rectangular case (a) and the circular case (b). The partitions in
the rectangular case present the same length (∆L), the partitions in the circular
case are shown in Fig. 3.19a. The noise contributions can be added in the case
of homogeneous channels, but not for non-homogeneous channels. Here, SItot is
calculated from the total area of the devices and the transconductance gm tot found

by modulating the entire channel simultaneously.

2
The current noise power density can be rewritten as; SI (f = 1 Hz) = gm 2 k . Fig.
S
3.20 shows the current noise power density at 1 Hz and for a trivial k = 1[V m].
From these results, it can be observed that non-homogenous conductances lead to
an increase of the noise produced by voltage fluctuations at the channel-electrolyte
interface.
P i
The term SI displayed in Fig. 3.20 is the noise predicted from Eq. 3.34 and it
is shown to be larger than the noise predicted by the charge noise model presented
in section 2.2.2 and [52].

61
4 Devices Fabrication and Experimental Methods

For the validation of the models previoulsy described, two different types of devices
were fabricated. On the one hand, macroscopic transistors and electrodes were
prepared using both SLG and porous materials based on graphene. On the other
hand, microtransistors with circular and rectangular geometries were fabricated. In
the first part of this section, the design and fabrication of the devices is presented.
In the second part, the characterization techniques and setups are presented.

4.1 Fabrication of graphene devices

In this section, the design and fabrication procedure of both macroscopic devices
and microscopic g-SGFET is presented. The graphene was grown by chemical vapor
deposition and transfered by wet etching transfer. The fabrication of macroscopic
devices is simplified in order to avoid contamination of graphene. On the other hand,
microscopic g-SGFETs are fabricated on 4 inches wafers using standard semicon-
ductor technology techniques.

4.1.1 Growth of graphene by chemical vapor deposition

Graphene was initially isolated by mechanical exfoliation from graphite [34]. Exfo-
liated graphene presents outstanding properties; however, its production is hardly
scalable to large areas. Graphene grown by chemical vapor deposition (CVD) over-
comes this limitation while maintaining high cristalline quality [67]. The graphene
used in this thesis6 was grown on a copper substrate which catalizes the decom-
position of CH4 . Carbon atoms are then adsorbed on the copper surface, where
they can diffuse until creating a new nucleation point or attaching to preexisting
grains [68]. The growth on copper has been reported to be especially suitable for
the production of SLG due to the low solubility of carbon in Cu and the strong
reduction of the catalitic effect after the first graphene layer is grown.
This process is carried out in a so-called “hot-wall reactor” which consist of a heating
chamber containing a quartz glass tube. The copper foil is introduced into the tube
where the temperature and atmosphere can be controlled. Firstly, the temperature
in the oven is increased to 1015 ◦ C while a high argon7 flow is kept (400 sccm).
This step allows to remove the copper oxide layer as well as other contaminants
from the Cu surface. In addition, it allows the reconstruction of the surface, a
crucial step for lowering the number of nucleation sites and producing larger crystal
grains [69]. After the annealing, the flux of Ar is increased to 1000 sccm, methane8

6
SLG was grown by collaborators from the group of Garrido at the WSI (TU München)
7
HiQ Argon 5.0, purity≥99.999%, Linde AG
8
HiQ Methane 4.5, purity≥99.995%, Linde AG

62
and hydrogen9 are introduced in the tube with a flow of 0.2 sccm and 200 sccm
respectively. After 180 minutes the quartz tube with the copper foil is removed
from the oven, allowing for a fast cooling process under the same pressure and flow
of H2 and Ar (Fig. 4.1).

Figure 4.1: Growth of graphene by chemical vapor deposition. The diagram shows
the temperature and gas fluxes during the different phases of the CVD graphene
growth process. In the first step (from 0 to 60 min) the temperature is increased to
1015o C while keeping a high argon flux (400 sccm). The second step (gray) takes
180 minutes and it is when graphene is grown. A 0.2 sccm flux of methane and
200 sccm flux of hydrogen is injected in the tube while the flux of argon is increased
to 1000 sccm. In the last step the copper is cooled down by extracting the tube
from the oven while the flux of argon is reduced to zero.

4.1.2 Wet transfer of chemical vapour deposition graphene

After the growth, the graphene is transferred onto the desired substrate. The trans-
fer is a critical step due to the fragility of graphene and the risk of chemically and
mechanically modifying it. The protocol followed is based on the process reported
in [70] and [71]. First, poly (methyl methacrylate) (PMMA)10 is spin coatted at
2000 rpm for 50 s on the graphene/copper foil. Here, the PMMA layer provides
mechanical strength to graphene. After drying the polymer overnight at room tem-
perature, the copper can be etched by a 0.5 M FeCl3 and 1 M HCl solution (Fig.4.2).
9
HiQ Hydrogen 5.0, purity≥99.999%, Linde AG
10
PMMA 950 A2, MicroChem Corp

63
Before completely removing the copper, the graphene grown on the back side of the
copper foil has to be removed. For this pupose, the back side of the copper foil
comes into contact with the etching solution for 2 minutes and then is flashed with
deionized water (DI water)11 . After repeating this procedure three times, the foil
is suspended by surface tension forces on the etching solution for 7 hours so that
the copper is completely etched 12 . Then, the graphene/PMMA sheet is placed in
1 M HCl for 2 minutes and in DI water for 3 hours in order to remove possible
FeCl3 residues. Afterwards, the graphene/PMMA sheet is fished with the desired
substrate and it is heated on a hot plate; in this way, by removing teh water be-
tween graphene and the substrate the adherence of graphene to the surface increases
significantly. The drying process is divided in two steps; firstly a low temperature
(≈ 40 ◦ C) is applied for few hours (better overnight); then a rapid temperature
ramp up to 90 ◦ C is applied in 30 min. For the transfer onto SiO2 , it was observed
that the applied temperature ramp must reach higher values (≈ 170 ◦ C) in order to
ensure a good attachment of graphene. Finally, the PMMA layer can be removed
by dissolving it in acetone at 55 ◦ C.

Figure 4.2: Schematic representation of wet etching transfer of CVD graphene grown
on Cu foil. The PMMA is spinned on the graphene/copper foil. After the back-side
graphene is flushed and the copper etched, graphene is cleaned in DI water. Fi-
nally the graphene/PMMA foil is fished using the desired substrate and the PMMA
removed.

11
Conductivity of DI water equal to 0.07 µS/m
12
Any contact of the PMMA with the etching solution must be avoided in order to prevent
damage of the PMMA

64
4.1.3 Fabrication of macroscopic devices

Macroscopic transistors based on SLG and electrodes based on both SLG and porous
graphene materials were fabricated in order to validate the distributed elements
models derived in section 3.2. In addition, the macroscopic electrodes are suitable
for the electrochemical characterization of the novel porous materials investigated
in this thesis. The fabrication process avoids any photolithography step, minimizing
the risk of modifying the material properties. Additionally, the large area of the
electrodes ensures large currents, which are easy to measure and make the char-
acterization less dependent on local imperfections of the materials. Therefore, the
dimensions of the macro electrodes were set to W = 4 mm x L = 10 mm in or-
der to observe the dominance of the distributed elements effect at relatively low
frequencies.
The devices were built on SiO2 thermally grown on a 4 in Si wafer. The oxide layer
was required for a good insulation between the conductive Si substrate and the
electrodes. The thickness of SiO2 was set to 2 µm in order to minimize parasitic
capacitances. Wafers were cut into small square pieces of 1.5 cm side and cleaned.
In order to ensure a good cleaning, the samples were submerged in acetone for
15 min, afterwards they were flushed with isopropyl alcohol in order to avoid fast
evaporation of acetone which could leave some dirt on the samples. Finally, the
samples where cleaned for 5 min in ethanol and flushed with DI water in order to
remove rests of ethanol.
The contacts of the devices were defined by e-beam evaporation13 of a Ti (10 nm)
and a Au (100 nm) layer. An aluminum shadow mask was used for the definition
of the contacts area (Fig. 4.3). In order to achieve good adhesion of the Ti/Au
layer on SiO2 , the surface was activated by oxygen plasma14 (2 min at 250W). After
the definition of the contacts and before the transfer of graphene the surface was
again activated. In this case, the Si/SiO2 substrate pieces were placed in piranha
solution15 for 15 min. Piranha is extremely oxidizing, and therefore it etches any
rest of organic compounds on the surface. In addition, it activates the surface
by generating a high density of siloxane (≡Si-O-Si≡) groups [72] which are highly
reactive, forming strong C-O bonds with graphene [73]. After the activation, SLG
was transfered on the sample, ensuring a maximal overlapping with both contacts.
Finally, a second layer of Ti (10 nm) and Au (100 nm) was deposited in order improve
robustness of the contact.
In the case of graphene-based porous materials, no transfer process is required
since the materials are directly inkjet printed on the substrate. The two materials
studied in this work are based on exfoliated graphene (eG)16 [74] and on reduced

13
ATC ORION Deposition system
14
PS 210 Microwave plasma system, PVA TePla
15
3:1 mixture of H2 SO4 with H2 O2
16
Material provided by the group of Francesco Bonaccorso from the Italian Institue of Technolgy

65
graphene oxide (rGO)17 [75]. The printing of the eG based material was performed
directly on the SiO2 while the rGO was printed on a polymeric substrate which
was then attached to the SiO2 substrate. Due to the impossibility of performing
plasma activation before the deposition of metals, the contacts were defined with
Ag paint18 which showed good adhesion with the substrate.
Finally, Cu wires were soldered to bond the metallic pads and the samples sealed
with PDMS 19 to ensure electric insulation. The PDMS was cured at 150 ◦ C for
15 min and a polyacrylate ring was attached over the active area of the electrode in
order to contain the electrolyte. Fig. 4.3 illustrates the described process and the
final devices.

4.1.4 Design and fabrication of graphene microtransistors

In this work, circular and rectangular micro g-SGFETs were fabricated in order
to validate the predicted increase of transconductance for the circular geometry.
Additionally, if a 2 to 3-fold increase can be achieved, the S/N ratio of the transistors
for sensing of neuronal electrical activity in vivo is expected to improve significantly.
Therefore, the circular and rectangular transistors are produced in two different
configurations. The first one is optimized for device testing and it is hereafter
refered to as Device Under Test (DUT) configuration (Fig. 4.4 left). The second
configuration si optimized for the measurement of neural activity in vivo. The
sensors are placed on the tip of a probe separated from the contacts by an arm
of ≈ 1.5 cm which allows for the proper placement of the sensors on the cerebral
cortex of animals (Fig. 4.4 right). This configuration is from here on refered as
In-vivo Probe. The aim of this work is not to perform in vivo experiments but to
evaluate possible technological problems related with this configuration and the use
of circular transistors in order to pave the way for future in vivo measurements.
In the DUT configuration, the densitiy of devices is optimized so that many transis-
tors with different characteristics can be fabricated and studied in a small area. The
structure is divided in four regions; the transistors in each of them are connected
to different sides of the DUT (Fig. 4.4 left). Each of these regions, consists of 20
transistors. Both circular and rectangular transistors are included in the design so
that their performance can be compared. As disscused in section 3.3.2, using the
circular transistors design it is possible to achieve a 2 to 3-fold increase of transcon-
ductance while keeping the size of the devices constant. Therefore, the performance
of rectangular transistors of channel width W has to be compared with that of
circular transistors of diameter (Φ = W ). Two different widths and diameters are
considered for the design, and for each of them the channel length is varied from the

17
Material provided by the group of Arben Mercoçi from the Catalan Institute of Nanotechnology
18
Silver conductive paint, RS Pro
19
SYLGARD R 184, Sigma-Aldrich

66
Figure 4.3: Scheme of the preparation of macroscopic devices. On the left hand
side the process for the SLG devices is detailed and on the right hand side the
preparation of porous graphene electrodes is illustrated. The inserted figures on the
left correspond to the shadow mask used for defining the metalic contacts (up) and
the Si/SiO2 with the defined contacts (middle). The two figures at the bottom of
the figure correspond to the final devices based on SLG (left) and porous graphene
(right).

minimum value allowed by the microfabrication resolution (≈5µm) to a maximum

length above the optimum length (Lopt ) predicted in section 3.3.2.

The resolution of the microfabrication process is mainly limited by the minimal

feature size of the photolithography masks. Two different masks were used; on
the one hand a polyester film with a soft photographic emulsion gel allowing a

67
Figure 4.4: Schematic of the Device Under Test (DUT) configuration (left) and the
Probe configuration (right). The zoomed areas display the circular and rectangular
transistors included in the design. The green rectangles indicate the position of the
contact pads. In the case of DUTs the contacts are simetrically placed at the four
sides of the structure.

8 µm resolution20 was used for defining the metal layers, the graphene layer and
the structure of the devices (see Fig. 4.10). On the other hand, a chromium mask
defined on soda lime glass21 with a resolution 2-3 µm was used for defining the
SU8 layer which insulates the metal tracks from the electrolyte. By using this
combination of masks, a good trade off between resolution and fabrication costs
was achieved.
In case of circular transistors, the length of the channel (L) is defined as Rout − Rin
and its maximum value is limited to the value of Rout (i.e. the difference of radii
cannot exceed Rout ). In addition, the radius of the inner contact cannot be smaller
than 15 µm due to the resolution of the photolithography masks used. Therefore, the
maximum channel length (Lmax ) for circular transistors was (Rout −15 µm−Laccess ).
Here, the minimum length of the access region (Laccess ) is also limited by the reso-
lution of the photomasks; in this case it is limited by the chromium mask to 5 µm22 .
20
Film 10” x 12” Photomask grade 4, PHOTODATA
21
4” Chrome Photomask, grade 1, PHOTODATA
22
The differences from the mask resolution arise from the challenge of producing circular shapes.

68
Besides, the outer contact has to present an opening so that the inner contact can be
connected. In order to keep the circular geometry, the graphene channel is defined
as illustrated in Fig. 4.5, ensuring that there is no graphene in regions where the
radial symmetry is distorted. In this way, the performance of the transistor can be
simulated as for a complete circle, but the total current has to be multiplied by a
factor θ/360 o , where θ represents the angle covered by the graphene channel. The
design has to ensure that the SU8 passivation layer covers the metal tracks which
connects the inner contact. In addition, the edges of the graphene channel must be
uncovered by the SU8 (see Fig. 4.5). Otherwise, paths of unmodulated graphene in
parallel with the channel would significantly affect the response of the transistor.

Figure 4.5: Schematic of a circular transistor. The angle θ stands for the angle of
the circular device covered by graphene. Inner and outer radius (Rin , Rout respec-
tively) and the diameter (Φ) are defined. It must be pointed out that the graphene
channel ends before the SU8 window in order to avoid parallel paths through un-
modulated graphene (red shape indicates the uncovered area). This picture includes
the limitations in the design due to the photomask resolution (i.e. a separation of
5µm between graphene and metal tracks, 5µm of access region and tracks width of
10µm).

In order to measure the tendencies shown in Fig. 3.17, the evolution of gm with
L has to be measured for a suficiently large range of L. The minimum value of
L is 5 µm, limited by the photomasks resolution, and the largest is Lmax defined
previously. If the condition Lmax ≥ 2Lopt has to fulfilled in order to properly
measure the predicted tendencies in gm vs L, the smallest diameter Φ which allows
for such a wide range of L values is Φ = 80 µm. Therefore, circular and rectangular
transistors with W = Φ = 80 µm and 120µm were fabricated. The transconductance
of these transistors is expected to be close to the one shown in Fig. 3.17. Table

69
4.1 summarizes the geometrical parameters chosen for the transistors in each of the
DUT compartments.

W, Φ (µm) Geometry L, Rout -Rin (µm)

rectangular 5 10 20 40 80
80
circular 5 8 12 16 20
rectangular 5 10 20 60 120
120
circular 5 10 18 30 40

Table 4.1: Geometrical parameters of the transistors fabricated in the DUT config-
uration. In all circular transistors θ = 280 ◦ .

In the In-vivo Probes configuration, circular and rectangular transistors were fab-
ricated for four different values of W (or Φ in case of circular transistors). All
these transistors were fabricated for the predicted optimal channel length (Lopt ). If
the channel length is not extended to the maximum value Lmax , transistors with a
smaller size (W=Φ=60 µm) can be fabricated. Table 4.2 summarizes the parameters
of the four In-vivo Probes fabricated for different transistor dimensions.
W, Φ (µm) Geometry L, Rout -Rin (µm) θ (degrees)
rectangular 20
60
circular 10 290
rectangular 20
80
circular 13 209
rectangular 20
100
circular 15 221
rectangular 20
120
circular 18 224

Table 4.2: Geometrical parameters chosen for the fabrication of circular and rect-
angular transistors in the In-vivo Probe configuration.

Microfabrication of g-SGFETs

Circular and rectangular g-SGFETs were fabricated on a flexible polyimide (PI)

substrate exploiting the technology recently developed in our group [12]. The main
fabrication steps are detailed in the following points:
1) Polyimide is provided in its precursor form (polyamic acid) dissolved in N-
methyl-2-pyrrolidone23 . This solution is spin coated on a 4 ” Si wafer in four
steps; 50 s at 750 rpm, 70 s at 1000 rpm (this parameters are chosen so that
23
Microchem PI 2611

70
 a 10µm thick film is obtained). The rigid substrate has to be previously
activated in oxygen-plasma at 1.4 mbar and 400 W for 15 min and heated to
150 ◦ C to remove water.
After the coating, the polyimide is thermally cured in a N2 atmosphere on
a hot plate by applying a 2.5 ◦ C/min ramp util 200 ◦ C are reached. After
30 min at 200 ◦ C the temperature is increased to 350 ◦ C for one hour; finally,
the sample is cooled down to room temperature.
2) 1st metal layer: It is defined by a lift-off process. Firstly, the photosensible
resist (AZ 5214E) is spin coated on the polyimide substrate so that the final
thicknes is ≈ 1.7 µm. The AZ can be used either as a positive or negative
resist. For the lift-off processes, a positive (clear field) photomask is used,
therefore AZ has to be inverted. Firstly, it is baked at 90 ◦ C for 1 min and
then exposed to UV light for 4.5 s with the photolithography mask covering
the structure of the metal layer. Afterwards, it is baked at 120 ◦ C for 2 min in a
process refered as reversal bake by which the non-exposed polymer undergoes
cross linking between chains. Then it is exposed again for 30 s to UV light in
a so-called flat exposure, for which no photomask is used. Finally, the AZ is
developed in the AZ developer24 for 45 s and the process is then stopped by
rinsing the wafer in DI water cascade for 1 min.
After the first photolithography, a 10 nm thick Ti and a 100 nm thick Au layers
were deposited using an e-beam evaporator. The metal layer attaches to the
PI substrate in the regions where AZ is removed, but it is deposited on the
AZ elsewhere. The metal on the resist can be removed by dissolving the AZ
with acetone and sonicating for short periods of few seconds until only the
metallic structure remains.
3) Graphene layer: It is defined after the first metal layer. Graphene can be
transfered on the wafer by the procedure described in section 4.1.2. After the
transfer, graphene has to be etched by oxygen reactive ion etching (RIE). The
graphene active area of the transistor channels is protected from the RIE with
a photodefined AZ film. For this process a clear field mask is used, therefore
the AZ is used as positive resist (Fig. 4.7). If the AZ is not inverted the
photolithography simplifies to a first pre-exposure bake (90 ◦ C for 1 min) and
a 35 s exposure to UV light. After the development of AZ, oxygen RIE is
applied25 in order to etch the uncovered graphene.
4) 2on metal layer: Ni/Au (20 nm/200 nm) is deposited following the proce-
dure described in the step 2. The double layer of metal improves the quality
of the contact. After this step, an slight misalignment of ≈ 3 µm between the
two metal layers was observed (Fig. 4.8). The reason for utilizing Ni instead
of Ti is that the interaction between graphene and nickel is very strong due
24
AZ 726 MIF
25
RIE treatment for 30 s using an O2 flow of 50sccm and a RF1 of 500 W and RF2 of 5 W

71
Figure 4.6: Lift-off process used for defining the metal tracks and contacts.

to hybridization of the metal d-electrons with the π-orbitals of graphene [76].

72
Figure 4.7: Complete process for defining the graphene channels after the transfer.
Graphene is represented by the thin black layer drawn on the PI substrate and
Ti/Au contacts.

Figure 4.8: This image, taken with an optical microscope, shows the misalignment
between the two metal layers due to an imperfect alignment of the photomasks. The
zoomed image on the right shows the line (in green) from which the brightness profile
is extracted (right, bottom plot). The brightness allows to observe the misalignment
between the two layers which is around 2 µm.

5) The insulation of the device was performed using a biocompatible polymer.

The photoepoxy resist SU826 was used for covering all the wafer except the
26
SU8 2002, MicroChem

73
 active area of the graphene channel and the metallic contact pads. The SU8
layer is spincoated at 800 rpm for 30 s. Then, a pre-exposure bake is performed;
first at 65 ◦ C for 3 min and afterwards the temperature is increased to 95 ◦ C
(5 ◦ C/min) and baked for 8 min more.
After the pre-exposure bake, the SU8 can undergo a photolithography step
which stimulates the cross-linking of polymeric chains (14 s at 11W/cm2 ). This
process is accelerated by a post-exposure bake at 65 ◦ C for 2 min (increase
from room temperature at 5 ◦ C/min), then at 95 ◦ C for 3 min (increase of
5 ◦ C/min) and finally cooled down at room temperature. The SU8 is then
immersed in the developer27 for 40 s and rinsed in isopropyl alcohol for 1 min
to stop the development. The process is finished with a hard-bake at 120 ◦ C
for 20 min in order to close possible cracks. After the hardbake, the thickness
of the SU8 was measured to be 4.3 µm. Fig. 4.9 shows the complete process
as well as an optical image of the resulting device. It is possible to observe
that due to the misalignment of the two metal layers, the access region has
been reduced to 1-3 µm.

27
mr-Dev600, micro resist technology

74
Figure 4.9: Fabrication Process flow for the definition of an insulating SU8 layer
on top of the transistors. Photomicrograph (down-right) shows the resulting device
in which the SU8 covers the metallic contacts and leave the active area uncovered.
Despite the misalignment between metallic layers these are completely insulated.

6) The structure: The In-vivo Probes and DUTs have to be defined on the
SU8 and PI so that de devices can be pealed off from the rigid substrate. For
this purpose a dark field photomak is employed and the AZ is inverted so that
it works as a negative photoresist. After the structure is defined with AZ on
the SU8 layer, a protective Al layer (300 nm thick) is deposited. This layer
serves for shielding the SU8 layer and the graphene channel which should not
be etched by RIE. It is necessary to protect these regions with Al and not
only with AZ, as in the case of the graphene etching, because here the RIE
exposure is much longer (≈ 45 min). The SU8 is especially hard to remove,

75
 therefore, a gas mixture containing fluorinated compounds must be used [77].

Figure 4.10: The structure of the In-vivo Probes and DUTs has to be defined on
the SU8 and PI films. Here, the complete process is depicted. First the AZ is
photodefined for a lift-off of Al. The deposited metallic layer protects the graphene
channel and the polymers in those regions where they should not be etched by the
RIE process. Finally the Al is removed and the devices pealed off.

The misalignments observed for the two metal layers reduces the access region but
in any case metallic parts were observed to remain uncovered. This reduction of
the access region is expected to imply a lower access resistance; its effect will be
analyzed in the following section (5.4.2).

76
Two relevant fabrication problems were observed. In the first place, the lower quality
of polymeric photomasks led to an imperfect lift-off of the Ti/Au and Ni/Au layers.
The photomasks presented some dark spots which prevented the proper exposition
of AZ to UV light (Fig. 4.11a). Secondly, despite a hardbake was performed after
the development of SU8, some cracks could be observed on the insulating layer (Fig.
4.11b).

(a) (b)

Figure 4.11: Optical images showing microfabrication problems. a) A rectangular

g-SGFET with gold particles remaining after the lift-off process can be observed.
b) The cracks on the SU-8 observed after the hardbake are shown.

4.2 Characterization setups and techniques

In this section, the techniques and setups employed for the characterization of
graphene based materials and the fabricated devices are described. First, an intro-
duction to Raman spectroscopy and to the main electrochemical characterization
techniques is given. Then, the setups used for the electrochemical and electrical
characterization of the graphene materials and devices are described.

4.2.1 Raman spectroscopy

Raman spectroscopy allow for the determination of the vibrational and rotational
modes of a polarizable system. The polarizability of an atomic structure (α) quanti-
fies the amplitude of a dipole induced by an external electric field. The polarizabil-
ity depends on the atomic symmetries, and therefore vibrational modes of atomic
structures are reflected on the time dependent α. The dipole moment induced by
an electric field can be expressed through the normal coordinates as follows[78]:

77
 1  δα   
µ(t) =α0 E0 cos(ω0 t) + q0 E0 cos (ω0 − ωq )t
2 δq q0

1  δα   
+ q0 E0 cos (ω0 + ωq )t (4.1)
2 δq q0

Here, ω0 is the frequency of the incident field and ωq the shift induced by the
interaction with vibrating atomic structures. E0 stands for the amplitude of the
incident electric field and q is a representative normal coordinate.

From the previous equation, it is clear that the induced dipole will oscillate at three
different frequencies. These dipole oscilations are responsible for the emission of a
secondary radiation [79]. The first term, describing a vibration at a frequency ω0 ,
corresponds to a Rayleight scattering processes and it does not contain information
about the structure of the irradiated matter. The second term describes a secondary
radiation with a lower frequency than that of the incident field (Stokes scattering).
The frequency shift corresponds to the frequency of the vibrational modes of the
molecules, therefore it contains valuable information about the atomic structure.
Finally, the third term presents a possitive shift in the frequency of oscillation anti-
Stokes scattering. It presents the same resonant frequencies as the Stokes scattering
but its intensity is much lower due to the lowest occupation probability of this higher
energy states.

In crystalline lattices, the vibrations can be described as phonons. An incident

photon with energy (~ω0 ) can either scatter elastically with electrons (Rayleigh
scattering) or inelastically (Stokes or anti-Stokes scattering). If an electron-hole
pair state is generated, these scattering processes generate new relaxation mecha-
nisms [80] which lead to the emission of photons. In this way, information about
the phononic and electronic states can be obtained, and therefore structural and
compositional information. The selection rules for the emission of Raman scattered
radiation is that the energy and momentum must be conserved:

~ωk0 = ~ωk ± ~ωq (4.2a)

k0 = k ± q (4.2b)

k0 and k represent the wave vector of the emitted and incident photons respectively
while q stands for the phonon wave vector. The terms ~ωk , ~ωk and ~ωq stand for
the energy of the respective photons and phonons. The + sign represent Stokes
processes and the - sign the anti-Stokes processes.

Raman spectroscopy is an excellent technique for studying graphene properties.

It is non-destructive, and if combined with a confocal microscope, it can provide
high spatial resolution. In graphene, Raman spectroscopy allows for the deter-

78
mination of the density of defects, number and orientation of layers and density
of dopants[80]. The phonon-electron interactions occur only with Raman active
phonon modes. When defects or impurities are introduced into graphene, addi-
tional scattering events between electrons and defects can occur. These processes
activate Raman bands which in the absence of defects remain inactive due to the
selection rule ∆q = 0 (see Fig. 4.13).

The Stokes energy shift in graphene leads to two main peaks in the Raman spectrum,
namely the G(1580 cm−1 ) and 2D (2690 cm−1 ) peaks. These two peaks are Raman
active even in the absence of defects. The ratio between the 2D and the G peaks
can be used to assess the number of graphene layers. The stacking of multiple
graphene layers implies the splitting of the 2D peak into other modes which lead to
a broader and shorter 2D peak [81]. Therefore, the G peak (named after its first
observation in graphite) presents the maximum relative intensity for graphite. For
single layer graphene I2D /IG ≥ 2, but for bilayer graphene I2D /IG < 1 [82]. Finally,
the D+D” band (2463 cm−1 ) can be observed in graphene and related materials such
as graphite. It is attributed to the combination of a D phonon and a phonon of the
longitudinal acoustic branch and it is defect independent. Similarly, the 2D’ band
(3250 cm−1 ) is also defect independent.

The most representative defect activated band is the D peak (1350 cm−1 ). It is
attributed to the interaction of electrons with in-plane transversal optical phonons
(iTO) and defects which activate relaxation mechanisms satisfying momentum con-
servation through scattering (see Fig. 4.13 right). Therefore, the appearence of the
D peak is a clear indication of the presence of defects in graphene [80]. Another
band related with the density of defects is the D+D’ band (2940 cm−1 ).

Figure 4.12: Raman spectrum of single layer graphene on SiO2 . The peaks discussed
in the main text are indicated.

79
Figure 4.13: Energy dispersion relation for electrodes in graphene and processes
involved in the emission of secondary radiation for three Raman bands. In the left
side the processes involved in the G band are depicted. In the middle, the same
processes are indicated for the 2D band and finally, in the right side, the events
related with the D band are shown. The blue line indicates the excitation of an
electron due to the absorption of a photon. The red line represents the emission
of a photon during the electron-hole recombination. The dotted arrows indicate
phonon emission events while the dashed line indicates the scattering of an electron
with a deffect of the lattice. It is important to emphazise that the emission of
a photon in the D band would not be allowed if no scattering with defects was
possible. All electron-phonon scattering processes drawn in this scheme involve in-
plane transversal optical phonons. In case of the G band, the phonon consists of the
in plane displacement in opposite directions of the two atoms of the atomic basis.
The phonons involved in the D band and the 2D band originate from the breathing
mode of six-atom rings [80].

4.2.2 Electrochemical characterization

The electrochemical characterization was performed by means of a potentiostat28 .

This equipment allows to measure the current flowing through a working electrode
(WE) as a response to a voltage applied with respect to a reference electrode [83].
The potential of the reference must be constant in order to properly measure the
voltage drop at the WE. However, reference electrodes tipically have relatively large
impedance and its electrochemical potential shifts significanly with current changes.
This problem is minimized by using a counter electrode (CE) which supplies the
required current so that no current flows through the reference. In this way, cur-
rents through the reference electrode are avoided. Fig. 4.14 displays a simplified
schematic of a potentiostat. The electrochemical cell depicted in the figure is con-
nected to the potentiostat which consists of three amplifiers. The two amplifers with
the symbol x1 represent a gain 1 operational amplifier (op-amp) which provides a
28
SP-300 potentiostat, Biologic

80
voltage equal to the difference between the input voltages. The op-amp expressed
as electrometer in Fig. 4.14 measures the voltage difference between the reference
electrode and the WE. This signal is compared with the desired applied voltage in
the control amplifier which injects current so that the difference between them is
zero. Finally, the current injected into the cell flows through Rm because of the
high input impedance of the op-amps, the voltage drop across Rm can be measured
by the I/U converter.

Figure 4.14: Simplified schematic of a potentiostat. The electrochemical cell is

represented by the circle in which the reference, working and counter electrodes are
immersed. The two operational amplifiers with the x1 symbol stand for a gain 1
operational amplifier which supplies at the output the voltage difference between
the inputs. The amplifier on top stands for a servo amplifier, which provides a
current so that the difference between Uin and UW E − EAg/AgCl is zero [83].

Knowing the current across the WE and the voltage drop imposed on the WE, it is
possible to perform multiple studies in order to characterize the electrochemical re-
sponse of the electrodes. In this thesis three techniques were used; cyclic voltamme-
try (CV), chronopotentiometry (CP) and electrochemical impedance spectroscopy
(EIS). The difference between them lies mainly on the voltage signal applied (Upert )

81
and in the way the response is analyzed.
1) Cyclic voltammetry: A triangular voltage signal is applied betwen the ref-
erence electrode and the WE (Fig. 4.15a). The current generated is recorded
and plotted versus the voltage. Depending on the electrical and electrochem-
ical properties of the electrode the current generated presents different fea-
tures. CV is used for studying the reactivity of the electrode materials and
electrolyte. If a red-ox reaction occurs above a certain overpotential a current
peak can be observed around this voltage. On the other hand, if the ca-
pacitive response of the electrode dominates, a constant current is measured
(i ∝ CdV /dt). In this case positive currents are observed for positive volt-
age ramps and negative currents for negative voltage ramps. This leads to a
measurable histeresis directly related with the interfacial capacitance of the
electrode. Besides, the resistance in series with the electrode can be measured
from the time required for reaching maximum capacitive currents (τ = RC).

(a) (b)

Figure 4.15: Cyclic voltammetry experiments. a) A triangular voltage signal is ap-

plied between WE and reference. The red line shows 5 cycles at scan rate 100 V/s
and the black line 3 cycles at scan rate 50 V/s. b) The current, which depends on
the scan rate is plotted versus the electrochemical potential. A flat current versus
potential indicated capacitive behavior. Doubling the scan rate approximately dou-
bles the current. The steep current increase at high potentials indicates de limits
of the EW.

2) Electrochemical Impedance Spectroscopy: It relies on the application

of an AC signal at a certain DC polarization between reference and WE. This
signal contains the frequencies of interest (e.g. few frequency points per decade
in the range of interest) (Fig. 4.16a). The voltage variations across the WE
are then recorded in time domain and transformed into the frequency domain.
This signal in the frequency domain can be then divided by the amplitude of
the input signals in order to obtain the impedance versus frequency. Usually
the impedance is plotted in terms of module and phase (Fig. 4.16b). This

82
 spectra can be fitted with the equations from equivalent electric circuits such
as capacitors, resistors, CPEs, or combinations of them. From the fitting it is
possible to extract equivalent parameters and to know which electrochemical
process dominates the response of the electrode.

(a) (b)

Figure 4.16: Applied signal between WE and reference and processed electrochemi-
cal impedance spectroscopy data. a) A superposition of signals with all the frequen-
cies of interest (in this case 0.15, 1, 11, 110, 1100, 6000 and 150000 Hz), usually few
frequency points per decade with an amplitude of few tends of mV (in this exam-
ple 25 mV). b) The response in terms of current is recorded and transformed into
the frequency domain by Fourier transform. The amplied amplitude is divided by
the amplitude of the electric current spectrum leading to the impedance spectrum.
The impedance is a complex magnitude and the module and phase are usually rep-
resented separately. Here, the impedance measured for the signal applied in (a)
is shown. It can be observed that the phase goes from a capacitive (-90 ◦ ) to a
resistive (0 ◦ ) response while the module of the impedance decreases. The phase ap-
proaches 0 ◦ as the resistance in series (Rs ) dominates the impedance of the interface
capacitance.

3) Chronopotentiometry: A current pulse is imposed on the WE. This flux

generates a voltage change at the WE which is ploted against time. The
voltage signal is a constant ramp if the response of the electrode is purely
capacitive. If the response is simply resistive, a certain voltage drop occurs.
A resistance in series with the electrode (e.g. contact resistance) produces a
certain voltage drop at the begining and end of the pulse, the so-called ohmic
drop. In this way, both the capacitance and the resistance in series can be
obtained from the slope of the voltage signal and the initial and final ohmic
drops respectively.

83
 (a) (b)

Figure 4.17: Applied current pulse (iinj ) and reference and recorded potential (WE-
EAg/AgCl ). a) The simplest chronopotentiometric study is presented which consists
of a monophasic square current pulse. b) The response in terms of potential is
recorded and plotted versus time. In this figure the main features of the response
are indicated in red. The initial potential drop (ohmic drop) is proportional to
the reisistance in series (Rs ), then the voltage increases linearly with an slope pro-
portional to the interface capacitance and finally the capacitor discharges after the
ohmic drop following a negative exponential dependence.

4.2.3 Electrical characterization

In this work, the impedance from drain to source and transconductance versus fre-
quency are measured for macroscopic g-SGFETs. On the other hand, the transcon-
ductance of circular and rectangular microscopic g-SGFETs is measured in both
the In-Vivo Probe and DUT configuration with different setups.
In order to measure the impedance from drain to source, an EIS analysis is per-
formed. However, in this case the potentiostat is replaced by an impedance ana-
lyzer29 , which samples the applied and output signals in order to provide the real
and imaginary parts of the impedance. Fig. 4.18 illustrates the measurement setup.
It can be observed that the input signal is not applied between the reference elec-
trode and the WE but between drain and source. Here, no counter electrode (CE)
is required in order to avoid a shift of the reference potential because the total
impedance between source and drain has to be measured. This is also the reason
for not utilizing a potentiostat. It is worth to emphasize that an 8 MΩ resistor is
connected in series with the gate or Ag/AgCl electrode. This resistor is necessary
in order to minimize the flow of current through this parallel path, otherwise the
impedance between drain and source could not be precisely measured. The current

29
Solartron 1260

84
flowing through the reference generates a voltage drop in this resistor. The voltage
between gate and drain has to be measured simultaneously and the value of the
voltage source Ugs adjusted so that the final voltage drop between gate and source
is the desired polarization.

Figure 4.18: Schematic of the source-to-drain impedance measurement setup. Ugs

is swept in order to measure the changes in channel resistance and capacitance with
the polarization voltage. The resistor in series with the reference electrode (8 MΩ)
minimizes the contribution of the current path through the reference electrode to
the source-to-drain impedance.

The dependence of the transconductance with frequency has been measured with a
home built setup30 . It consists of a current-to-voltage converter and a signal gener-
ator controlled with Matlabr . A superposition of sinusoidal signals containing all
the frequencies of interest is applied between source and gate. This signal modu-
lates the conductivity of the channel leading to small changes in Ids . Therefore, the
transconductance can be obtained from dividing the current fluctuations in the fre-
quency domain by the applied voltage signal. This setup will be used to verify the
observation that at high frequencies the current through the graphene-electrolyte
interface overcomes the Ids generated by the modulation of the channel (as dis-
cussed in section 3.2). For this purpose the gate current is also measured. Fig. 4.19
shows a simplified scheme of the measurement setup including an op-amp of gain 1
which supplies at the output the voltage difference between the inputs (x1). This
amplifier allows to measure the current through the reference electrode and from
that to calculate its contribution to the equivalent transconductance defined in Eq.
3.21.

30
Built in the CNM (Barcelona) by Anton Guimerà

85
Figure 4.19: Frequency-dependent transconductance is measured by applying a sig-
nal containing all the frequencies of interest between source and gate. The current
(Ids ) response is then measured and transformed to the frequency domain. The
amplitude for each frequency is divided by the amplitude of the applied AC signal
in order to obtain the gm versus frequency response. The current through the gate
is also measured. The op-amp indicated with the symbol x1 represents a gain 1
operational amplifier. The value of Rm must be lower than the sum of reference
electrode impedance and graphene-electrolyte interface impedance in order to avoid
damping of the actual leakage current.

The transconductance of circular and rectangular g-SGFETs is measured from the

DC response of the devices, calculating the derivative of Ids with respect to Ugs . The
transistors in the In-vitro Probe configuration can be characterized using a home
built setup, precisely described in [84]. It allows for the simultaneous measurement
of the response of 32 transistors connected to the setup using the connector shown
in Fig. 4.21a. The setup is based on a current-to-voltage converter on a printed
circuit board (PCB) which preamplifies the signal. The PCB incorporates a low-pass
filter which removes high frequency noise from the measurements (Fig. 4.20). The
signals from the 32 channels are communicated to a DAQ-card31 which performs a
second amplification and digitalization of the data. The DAQ-card also applies the
polarization voltages Uds and Ugs in a process controlled through a LabWindowsr
homemade software32 .

31
DAQ NI USB 6363
32
Program written by Damià Viana and Andrea Bonaccini

86
Figure 4.20: Schematic of the DC response characterization is shown. Ids and Igs can
be measured for different Ugs and Uds polarization voltages in order to characterize
the modulation of the transistor current. The capacitor in parallel with the resistor
of the current-to-voltage converter constitutes a low-pass filter, in this case with a
cut-off frequency of 15.9 kHz. The output signals are sent to the DAQ-card.

The DUTs could not be connected to the DAC-card adquisition system33 . Therefore,
a semiautomatic probe system34 was used for stablishing the connection with a
semiconductor parameter analyzer35 . Fig. 4.21b shows the tips aligned with the
contact pads of a DUT; the alignment had to be performed for each transistor
individually. A small polymeric vessel36 to contain the electrolyte was attached
onto the DUT. In the electrolyte, the Ag/AgCl reference electrode was submerged.

(a) (b)

Figure 4.21: Connections of the In-vivo Probe and DUT configurations. a) The
In-vivo Probes are connected to the PCB through a homebuilt adapter, shown in
the picture. b) The probe system is aligned with the source of a transistor and the
drain, which is common to all transistors. The reference electrode is placed in the
electrolyte, which is contained in a small PDMS vessel.

33
No connector was available
34
SUSS PA200 Semiautomatic Probe System
35
HP 4155B, Hewlett Packard
36
µ-Dish, micro-insert 4 well, Ibidi

87
5 Experimental Results and Validation of Numerical
and Analytical Models

This section discusses the results obtained from the characterization of the macro-
scopic devices and the microtransistors. The data is compared with the predictions
given by the models presented in section 3.2. First the graphene based materi-
als are characterized by confocal Raman spectroscopy and scanning electron mi-
croscopy (SEM). The frequency response of SLG macroscopic electrodes and tran-
sistors (W=4 mm, L=6 mm) is studied and the experimental data fitted with the
distributed elements models (DE models). Then, the electrochemical properties of
graphene-based porous materials are presented. The DE model of a rectangular
electrode (see appendix 7.1) is used to evaluate the impact of non-homogeneous
current injection on the use of these materials for neural stimulation. Finally, the
DC response of circular and rectangular g-SGFETs microfabricated on a flexible
substrate is discussed. The experimental results provide means for validating the
relationship between geometrical parameters and sensitivity predicted by the ana-
lytical and numerical models.

5.1 Topography and Raman analysis

The properties of the CVD graphene used for the fabrication of both micro and
macro devices as well as the graphene-based porous materials has been studied by
confocal Raman microscopy37 .
Fig. 5.1 shows the Raman spectra of SLG on copper measured at the possitions
indicated with crosses in Fig. 5.1. It can be observed that the peak D, indicative
for the density of defects, is not visible. The average ratio of maximum intensities
2D/G, indicative of the number of stacked layers is equal to 1.45 ± 0.41 which cor-
responds to a mixture of SLG and multilayer graphene (Fig. 5.1) [82]. The Raman
signal of graphene obtained on Cu is noisy due to the intense and broad signal of
Cu38 which has to be subtracted from the data.

37
Confocal Raman microscope WITec-alpha300 R. A laser light source with λ=488nm was used
38
For a laser light of 488 nm

88
Figure 5.1: Raman spectroscopy data of graphene on Cu measured with a 50 am-
plification objective. a) Copper foil and the points where Raman specta were taken
(marked with crosses). b) Few representative Raman spectra recorded at the points
indicated in (a).

After the transfer on SiO2 it is possible to observe the presence of some regions
with multiple layers of graphene (Fig. 5.2). The ratio of maximum intensities 2D/G
indicates that the Raman signal from the multilayer graphene region resembles the
signal of graphite [85].

Figure 5.2: Raman spectroscopy data of graphene on SiO2 . a) Mapping of the ratio
between maximal intensities of the 2D and G peaks. b) Raman spectra taken at
the points marked with crosses.

The graphene-based porous materials are provided with different thicknesses. In

89
this study, two rGO samples are studied with nominal thickness of 30 and 200 nm
and one eG sample with a nominal thickness39 of 80 nm. In Fig.5.3 the active area
of the three samples is shown.

(a) (b) (c)

Figure 5.3: Optical images of the three sealed porous graphene samples. Defined by
a yellow line the area of material not covered by PDMS a) rGO porous electrode
with a nominal thickness of 30 nm and active area of 0.24 cm2 . b) rGO porous
electrode with a nominal thickness of 200 nm and active area of 0.28 cm2 . c) eG
electrode with a thickness of ≈80 nm, active area of 0.12 cm2 .

For a better observation of the porous structure, SEM images were obtained with an
electron beam energy of 20 kV and 10000 magnification (Fig. 5.4). Due to the low
conductivity of rGO samples, the deposition of a 3 nm thick gold layer was required
for minimizing charging of the material.
The SEM images show flakes of approximately 3 µm for the eG based structure and
0.5 µm for the rGO based material. These flake sizes are smaller than a typical
neuron (≈ 20 µm), therefore, the roughness in the scale of a neuron and in the
macroscopic scale is approximately the same. Then, the electrochemical properties
in the scale of neurons and in macroscopic electrodes is the same. The difference in
terms of flake size between the two materials suggests that the surface area of the
rGO electrodes is larger, and therefore a larger charge injection capacity is expected.

39
These three approximate values were given by the suppliers of the materials

90
 (a) (b)

Figure 5.4: SEM images of porous graphene electrodes. a) SEM image of a porous
electrode based on eG. b) SEM image of a porous electrode based on rGO.

(b)
(a)

Figure 5.5: Spectra of porous graphene-based materials. a) Porous material based

on eG flakes. b)Porous material based on printed rGO.

The graphene based porous materials have also been characterized by Raman spec-
troscopy. Fig. 5.5 shows the average spectra over a large area of 2 mm x 2 mm
for both the rGO and eG based materials. In both spectra it can be observed
the presence of the 2D and G peaks with very low 2D/G ratio corresponding to
multilayer graphene, actually approaching the ratio of the graphite spectrum [85].
rGO presents intense D and D+G peaks which correspond to defect activated bands
[80] typical of this material [86] [87]. On the other hand, the eG porous material
presents an spectrum similar to pure graphite, with a relatively low D peak and a
visible D+D” peak which is not related with defects [80].

91
5.2 Frequency response of graphene devices

Measurement of the interface capacitance in graphene electrodes have been recently

performed in aqueous electrolytes[88] and other environments [48] [89] using elec-
trochemical impedance spectroscopy. The capacitance of the graphene has been
shown to depend on the voltage, due to the effect of the quantum capacitance. The
capacitance is an important parameter since it dominates the impedance for a wide
range of frequencies. Nevertheless, the impedance of SLG electrodes and SGFETs
is influenced by the sheet resistance as described by the DE models presented in
section 3.2 and appendix 7.1. Here, the effects occuring at high frequencies found
by electrochemical impedance spectroscopy are presented. The obtained spectra
were fitted with the DE models which closely match the experimental data.
The impedance spectra were measured at different polarization voltages, which pre-
sumably modulate the interface capacitance as well as the sheet resistance. Previous
to the characterization of the frequency response of SLG devices it is important to
determine the electrochemical potential window (WE). The EW indicates the range
of polarizations at which no significant faradic charge transfer occurs, and therefore
only the Ugs values within the EW can be used in in-vivo applications. Fig. 5.6
displays the cyclic voltammetry of SLG in the electrolyte used for the frequency
response characterization40 ; the results show a capacitive response between UW E vs
Ag/AgCl =-0.2 and 0.6 V. It has to be emphasized that the polarization defined as
the gate-to-source potential (Ugs ) is equal to minus UW E vs Ag/AgCl.

5.2.1 Frequency response of rectangular graphene electrodes

The impedance in the electrode configuration was measured for different polarization
voltages (Ugs ). From the spectra, the dependence of the capacitance (Cint ) and the
sheet resistance (Rsh ) on the polarization voltage (Ugs ) could be extracted.
In order to better interpret the the impedance spectra obtained for different po-
larizations. The spectrum of a SLG electrode was first measured with a 1 kΩ and
a 10 kΩ resistor connected in series at Ugs = UD = 0.12 V. From this data, shown
in Fig. 5.7, it is possible to observe the effect of the resistance in series (Rs ) on
the appearence of the distributed elements effect in the impedance spectra. If Rs
is high in comparison to Rsh , the effect of the distributed elements is attenuated
because Rs dominates the spectrum at high frequencies, at which the resistive and
capacitive components of the electrode impedance have the same value. In other
words, Rs adds a purely real term to the impedance which lowers the phase from
45 ◦ to 0 ◦ (see Fig.3.8 in section 3.2). As shown in Fig. 5.7, for large Rs values the
data can be well described by a discrete (i.e. non-distributed) equivalent circuit
constituted by a CPE. On the other hand, if Rs is comparatively low, the effect of

40
500 mM PBS at a pH value of 7

92
Figure 5.6: Cyclic voltammetry of SLG measured in PBS (500 mM at a pH value
of 7) and for a scan rate of 200 mV/s. The vertical red lines indicate the limits of
the electrochemical potential window.

(a) (b)

Figure 5.7: Impedance spectra of rectangular graphene electrodes. Module (a) and
phase (b) of the impedance measured with two different resistances connected in
series (1 kΩ and 10 kΩ).

the distributed elements clearly affects the impedance spectrum and the DE model
equations are required in order to properly fit the data.

The impedance spectrum of SLG electrodes was then measured for different Ugs val-
ues, from -0.055 V to 0.345 V in steps of 0.025 V, and without any resistor connected
in series. Fig. 5.8 shows the spectra measured close to the Dirac point (0.12 V) and
away from the Dirac point (-0.055 V). The effect of the distributed elements can be

93
 (a) (b)

Figure 5.8: Impedance spectra of rectangular graphene electrodes. Module (a)

and phase (b) of the impedance measured at two different polarization voltages
(−0.055 V and 0.12 V).

observed close to the Dirac point because Rsh is around its maximum, and there-
fore Rs is relatively low with respect to Rsh . Away from the Dirac point, Rs is
comparatively large and thus the effect of the distributed elements is attenuated.

94
 (a) (b)

Figure 5.9: The capacitance (a) and sheet resistance (b) of a SLG electrode ex-
tracted from fitting the data with the model which accounts for the effect of the
distribution of elements.

The experimental data was fitted with the DE model of a rectangular electrode.
The interface impedance Zint was modelled by constant phase elements with an
α = 0.95 and the factor Q is considered equal to Cint . In the low frequency range,
the capacitive impedance dominates the spectrum. In this regime, Rsh does not
affect the impedance spectrum and therefore the data can be fitted even with a dis-
crete equivalent circuit model in order to extract the value of Cint . Similarly, the Rs
can be extracted from the high frequency range, where it dominates the spectrum.
Besides, in order to fit the data at intermediate frequencies it is necessary to use
the DE model. Fitting the impedance spectrum with this model allows to extract
Rsh . Precisely determining Rs can be challenging due to the difficulties of mea-
suring the impedance at high frequencies. Deviations in Rs might lead to a wrong
determination of Rsh since the relation between them determines the appearence of
the distributed elements effect. A more precise way of obtaining Rsh is measuring
the impedance from drain to source in the transistor configuration.

5.2.2 Impedance from source to drain

The impedance from source to drain in the transistor configuration was also mea-
sured for different polarization voltages Ugs in order to measure the dependence of
the Rsh and the Cint with Ugs .
The DE model describes a drop in the module of the impedance from Rsh + Rs
to Rs and a peak in the phase at the same frequencies. If Rs is low, the phase
approaches -45 ◦ , however the purely real term Rs pushes the phase down to 0 ◦ (see
Fig. 3.11 in section 3.2). Fig. 5.10 shows the effect of connecting a resistance in
series (150 Ω and 500 Ω) with the transistor. The impedance spectra were obtained

95
 (a) (b)

Figure 5.10: Impedance spectra of rectangular graphene transistors. Modulus (a)

and phase (b) of the impedance measured with two different resistances connected
in series (150 Ω and 500 Ω). For both measurements Ugs = 0.12 V.

(a) (b)

Figure 5.11: Impedance spectra of rectangular graphene transistors. a and b) Mod-

ulus and phase of the impedance measured for two different polarization voltages
(−0.055 V and 0.12 V)

for a Ugs = UD = 0.12 V. It can be observed how the peak in the phase spectrum
is lower for higher Rs values. Here, the mismatch at high frequencies between
the experimental data and the fitted model is tentatively assigned to the effect of
parasitic capacitances between the contacts and substrate (see appendix 7.4).

The impedance was measured at the same Ugs values as in the electrode configu-
ration. Fig. 5.11, shows the impedance measured close and away from the Dirac
point (0.12 V and -0.055 V respectively). Increasing the polarization voltage (Ugs )
decreases the sheet resistance, and therefore the effect of Rs becomes more relevant.

96
 (a) (b)

Figure 5.12: The capacitance (a) and sheet resistance (b) have been extracted from
fitting the data with the model which accounts for the effect of the distribution of
elements.

Fig. 5.12 shows the extracted Rsh and Cint values using the DE model of a rect-
angular transistor. The interface impedance Zint was modelled by constant phase
elements with an α = 0.95 and the factor Q of the CPEs is considered equal to
Cint . From the low frequency range, Rsh can be precisely determined and from the
height of the peak in the phase and the module at high frequencies Rs is extracted.
The value of the interface capacitance determines the cut-off frequency at which
the module of the impedance drops from Rsh + Rs to Rs ; therefore, Cint can be
obtained from the fitting.

The parameters extracted from the measurements in the electrode and in the tran-
sistor configurations present large differences between them. These differences in-
dicate that the Rsh values extracted from the electrode configuration and the Cint
values from the transistor configuration have to be interpreted carefully. Deviations
from the ideal case which could affect the precise determination of these parameters
include parasitic capacitances from the contacts with the substrate, non-perfectly
rectangular geometries or holes and cracks in the graphene sheet.

5.2.3 From transistor to electrode frequency response

eq
The evolution of the equivalent transconductance (gm defined in Eq. 3.21) with
eq
frequency was measured for the macroscopic SLG transistors. An increase of gm
can be observed at high frequencies due to the capacitive currents through the
eq
graphene-electrolyte interface. The measurements do not show a drop of gm at
low frequencies, indicating that ωc1 > ωc2 (defined in Eq. 3.20 and Eq. 3.22) as
expected for long channels:

97
 1 Uds µ
ωc1 = > ωc2 = (5.1)
Rs Cint L2

Fig. 5.13 shows the effect of changing the polarization voltage (Ugs ). An increase of
Ugs − UD leads to an increase of the quantum capacitance and therefore a decrease
of the interface impedance. In addition, the transconductance is proportional to
Cint due to the modulation of the channel conductivity (Eq. 2.26). Thus, the two
contributions to the equivalent transconductance (i.e. from the capacitive currents
and from the modulation of the channel conductivity) increase equally leading to
no change in ωc2 (see Eq. 3.22). In Fig. 5.13, the total equivalent transconductance
eq eq
gm is shown. The contribution to gm from Igs is also displayed, showing that this
term dominates at high frequencies. It is also possible to see the saturation of the
equivalent transconductance due to the resistance in series with the channel which
limits Igs .

eq
Figure 5.13: Evolution of gm with frequency measured at two different polarization
voltages for a transistor of W = 4 mm and L = 6 mm. The empty squares represent
the capacitive term contributing to the effective transconductance calculated from
eq
the Igs current. The filled squares indicate the total gm measured from the Ids
current.

98
5.3 Electrochemical characterization of graphene-based porous ma-
terials

The electrochemical properties of the electrodes is of major importance for their

use in extracellular stimulation. In this section, the electrochemical potential win-
dow is measured by cyclic voltammetry and compared with the one of SLG. The
interface capacity is also extracted from the cyclic voltammetry and compared with
the values obtained from chronopotentiometry. From this technique, the contact re-
sistance is also extracted. Finally, electrochemical impedance spectroscopy is used
for measuring interface capacitance, charge transfer resistance, contact resistance
and sheet resistance, making use of the DE model for rectangular electrodes. The
properties of the graphene based porous materials are compared with the values of
SLG which is used as benchmark.
In Fig. 5.14, the cyclic voltammogram for the three samples is plotted and compared
with the CV of SLG. Both the two porous materials and SLG show a capacitive be-
havior with an EW ranging from -0.5 V to 1 V (UW E vs Ag/AgCl) for all the porous
materials but a smaller range (from -0.2 V to 0.7 V) for the SLG. The comparative
figure shows the large difference in terms of interface capacitance between the SLG
and the rest of materials which show much larger values. Furthermore, the 30nm
thick rGO electrode presents current levels similar to the eG based electrode, how-
ever, it is important to take into account that the thickness of the rGO electrode is
almost 3 times smaller. Indeed, the current levels for the 200 nm thick rGO electrode
are much larger than for the eG. The strong dependence of the current injection on
the film thickness indicates that the electrolyte can penetrate into the porous struc-
ture of the electrode. Finally, it is worth to emphasize that the electrodes based on
rGO exhibit a larger resistance in series. This resistance is observed through the
large charging constant of the electrodes which leads to the non-rectangular shape
of the cyclic voltammograms. The long time constant reduces the voltage range of
the current plateau at which a constant capacitive current can be measured.
In order to extract Cint , the CV was performed at different scan rates (200, 100 and
50 V /s) for 5 cycles each. The currents within the plateau (Ipl ) are obtained from
difference between charging and discharging currents averaged over all the cycles:
∆IEW/2
Ipl = (5.2)
2

The values measured at WE-EAg/AgCl =0.2 V have been fitted with the equation i =
Cint dU/dt in order to obtain the capacitance from the slope of the linear regression.
Fig. 5.15b shows the results of the CV and the fitting for the 200 nm thick rGO
electrode. Despite of the effect of the resistance in series, the current is dominated
by the capacitive current, showing a linear relationship between current and scan
rate. Table 5.1 summarizes the results obtained from all samples in terms of capacity
and charge injection capacity (i.e. Cint EW ). The results are comparable with those

99
Figure 5.14: Cyclic voltammograms of the two rGO samples (30 and 200 nm), eG
based material and SLG taken at a scan rate of 200 mV/s. Large differences in
terms of charge injection capacity can be observed. The red squares indicate the
zoomed data for the electrodes displaying lower interface capacitances.

of table 2.2.

(a) (b)

Figure 5.15: Cyclic voltammetry analysis of 200 nm thick reduced graphene oxide
electrode. a) Cyclic voltammetry data obtained for different scanning rates (200,
100 and 50 V/s). b) Linear fitting of the current measured for three different scan-
ning rates

The results found by cyclic voltammetry were complemented by chronopotentiome-

try. A current pulse with a period of 100 ms and intensity Iinj = 2 µA was injected
through the graphene based electrodes. The voltage drop (UW E vs Ag/AgCl) was
recorded for 300 ms (Fig. 5.16) for both the rGO and the eG electrodes. The ohmic
drop appears due to the contact resistance; inmediately after imposing a current,
shifts the voltage by a value equal to Rs Iinj . On the other hand, the slope of the
voltage shift can be used to obtain Cint . The resulting values are included in Fig.
5.16.

100
 C (µF/cm2 ) EW (V) Qinj (mC/cm2 )
SLG 1.60 -0.4 to 0.7 0.00176
30 nm rGO 113 -0.5 to 1 0.169
200 nm rGO 434 -0.5 to 1 0.651
80 nm eG 52 -0.5 to 1 0.078

Table 5.1: Summary of interfacial capacitance, charge injection capacitance and

electrochemical window obtained for the cyclic voltammetry for all the characterized
electrodes.

Figure 5.16: Chronopotentiogram of the porous graphene electrodes. The interface

capacitances are indicated under the voltage ramp induced by the constant 2 µA
injected current. The contact resistance obtained from the ohmic drop are also
shown in the diagram.

The impedance spectra (EIS) were obtained for the graphene based porous materials
and SLG. An AC signal of 10 mV with an offset UW E vs Ag/AgCl ranging from
0.25 V to 0.9 V was applied and the data averaged over 5 measurements. Here, the
impedance spectra allow to measure Cint , Rsh , Rs and the charge transfer resistance
(Rct ). At low frequencies, the impedance of the capacitive elements at the interface
is very large, therefore the impedance becomes dominated by the charge transfer
resistance (see Fig. 5.18). If the charge transfer resistance is relatively low, it
dominates the impedance at non-excesively low frequencies and its effect can be
measured. Its effect is translated into a frequency-independent impedance module
and a drop in the phase back to 0 ◦ at low frequencies. The impedance at high
frequencies is dominated by Rs because the capacitive impedance of the interface
decreases with frequency. In microscopic devices, the value of Rs has contributions
from the graphene-metal contact and from the access resistance. Oppositively, in

101
the macroscopic electrodes studied in this section, the value of Rs is expected to be
dominated by the access resistance. The reason is that while the contact resistance
decreases with the width (W ) of the device, the access region is approximately the
same because Laccess is also larger in macroscopic devices (≈ 1 mm). Therefore, a
proportional relation between the material resistivity and the resistance in series is
expected for the macroscopic electrodes.
Fig. 5.17 reveals the large differences both in terms of capacitance and series resis-
tance between the studied electrodes. In the case of rGO, the 30nm thick electrode
presents an impedance at low frequencies which is approximately one order of mag-
nitude higher than for the 200 nm thick electrode. Interestingly, a difference of one
order of magnitude is also observed for the series resistance. The reason for this cor-
relation is that both the interface capacitance and the conductivity of the material
are proportional to its thickness.

(a) (b)

Figure 5.17: Electrochemical impedance spectra of graphene based electrodes. Mod-

ule (a) and phase (b) are shown.

Finally, the spectra present the effect of the previously discussed distribution of
elements (section 3.2). From this effect, the sheet resistance of the material can
also be extracted. Fig. 5.17 shows the spectra for the rGO and eG based materials
as well as for SLG electrodes. The conductivity of the graphene porous materials
could not be modulated within the EW. Nevertheless, the spectra change with the
polarization voltage because of the effect of Rct at low frequencies. Fig.5.18 shows
the impedance spectra for two different polarizations (WE-EAg/AgCl =0.25 V and
0.9 V ). This data was fitted with the DE model for rectangular transistors and
with a discrete equivalent circuit; in both cases the interface was modelled with
a CPE in parallel with the charge transfer resistance. Due to the large interface
capacitance and sheet resistance of these materials, the effect of the distribution of
elements is evident at a frequency much lower than for SLG (≈10 Hz).
Fig. 5.18 shows that in order to avoid a wrong determination of the interface
capacitance in electrodes with a large Rsh , the Cint has to be measured at sufficiently

102
low frequencies for which the resistance of the channel can be ignored. Apart from
the impact on the characterization of material properties on large electrodes, the
impact of the distribution of elements on the non-homogeneous charge injection has
to be assessed. The current flowing through the electrode-electrolyte interface can
be calculated using Eq. 3.9. Fig. 5.19 shows the module of the current flowing in
the z direction (into the electrolyte) at different positions along the electrode. The
plotted values have been calculated using the parameters extracted for the 200 nm
thick rGO electrode at WE-EAg/AgCl =0.9 V (i. e. C=269 µF/cm2 , Rsh =12 kΩ,
α = 0.89, Rct=200 kΩ and Rc =1.8 kΩ).
Fig. 5.19a shows the dramatic inhomogeneity of the injected current even at rel-
atively low frequencies (≈10 Hz). Nevertheless, when the size of the electrode is
reduced, the current tends to flow homogeneously due to the smaller influence of
the sheet resistance. Fig. 5.19b shows the modulus of iz calculated for a square
electrode of side 500 µm. If an electrode of side 50 µm is considered, no significant
inhomogeneity occurs below 10 kHz.

5.4 Electrical characterization of circular transistors

In this section, the characterization of both the circular and rectangular micro-
transistors is described. The g-SGFETs characteristics reported in the literature
and the predictions provided by the numerical models developed in this work are
shown to closely match the experimental data. Firstly, the basic characterization
of g-SGFETs is presented showing the main features of Ids vs Ugs curves. In the
second place, the cleaning effect in the transistors is shortly discussed and finally
the size and shape dependencies of the transistors are discussed and compared with
the simulated values.

5.4.1 Basic characterization

In order to measure the dependence of Ids with Ugs and Uds the characterization
setup described in section 4.2.3 was used. In this way, all the 14 transistors in
each probe could be characterized simultaneously. The head of the probes as well
as the reference electrode (Ag/AgCl) were submerged in the electrolyte (10mM
PBS at pH=7). The gate to source voltage was swept from -0.3 to 0.7 V in steps
of 20 mV and the drain to source voltage from -0.1 to 0.1 V in steps of 0.05 V.
After changing the Ugs the measurement is taken after 500 ms; after changing Uds
the delay time is 200 ms. This time is required in order to ensure that the interface
capacitance is charged, and therefore the steady state is reached. Fig. 5.20 shows the
relationship between Ids , Ugs and Uds for a circular transistor of diameter Φ=80 µm.
The figure on the left shows the linear relation between Ids and Uds where the slope
indicates the conductance of the channel. The linear relation does not hold if the
Uds is suficiently large so that it could change the conductivity along the graphene

103
 (a)
(b)

(c) (d)

Figure 5.18: Electrochemical impedance spectra of the 200nm thick rGO electrode
recorded at two different polarization voltages. a) Module of the impedance is fitted
with the DE model. b)Phase of the impedance is fitted with DE model. c). The
module is fitted with a CPE in parallel with a resistance. d) The phase is fitted
with a CPE in parallel with a resistance.

channel (see section 3.1). In Fig. 5.20a deviations from the linear relationship can
be observed. The change of conductivity with Ugs can be better observed in Fig.
5.20b. Here, the ambipolar transport of graphene can be clearly observed in the
modulation of Ids with Ugs showing the Dirac point around Ugs =0.12 V. The shift in
the Dirac point observed for different Uds is equal to Uds /2. This is approximately
the average shift of the Fermi level in the graphene channel induced by Uds . These
features can be described by Eq. 2.25 and simulated using the graphene model
based on FEM presented previously (see section 3.1).

The Dirac point for pristine graphene is stimated around 0.1 V from the difference
between the graphene work function (4.6 eV [90]) and the electrochemical potential
of the reference electrode (4.7 eV). The curves shown in Fig. 5.20b show a Dirac
voltage very close to the estimated value, nevertheless these curves were measured
at the 15th cycle (one cycle corresponds to a complete Ugs and Uds sweep). The

104
 (a) (b)

Figure 5.19: Current injected into the electrolyte at different electrode positions
and for different frequencies, (Upert = 10mV ). The position (x) is normalized by
the channel length L a) Modulus of iz for the 200 nm thick rGO electrode with
parameters L = 6 mm and W = 4 mm. b)Module of iz for an hypothetic 200 nm
thick rGO electrode with parameters L = 500 µm and W = 500 µm.

(a) (b)

Figure 5.20: DC response of a circular g-SGFET with Rout = 40 µm and Rin =

27 µm. a) Ids vs Uds for different Ugs values. b) Ids vs Ugs for different Uds values.

repetitive cycling of the potential between the graphene and the reference was shown
to produce a “cleaning effect” shifting the Dirac point from values corresponding
to p-doped graphene to more negative values of Ugs . Fig. 5.21a shows the drift of
the Dirac point and the increase of leakage current (Igs ) with the cycle number. It
is worth to emphasize the lowering of the minimum current with the cycle number.
This effect could be modelled by a decrease of the minimum charge carriers density
(n0 ). The relation between n0 and the presence of impurities has been presented
elsewhere [43]. Possible sources of contamination in our process include rest of
PMMA employed during transfer, photoresist, or gold particles not removed during
the lift-off process [91]. The latter is especially likely since optical images revealed

105
the presence of gold particles not removed due to imperfections in the photolithog-
raphy masks (Fig. 4.11a). Nevertheless, a detailed study should be performed to
prove whether the gold contamination is responsible for the p-doping of graphene
or not.

(a) (b)

Figure 5.21: Cleaning of graphene with the cycle number a) Evolution of Ids vs
Vgs with the cycle number. Inserted figure (shift of the Dirac voltage vs cycle
number). b)Evolution of Igs with the cycle number. Remember that Ugs is equal
to minus UW E vs Ag/AgCl. Igs changes sign around 0.18 V independently of the
cycle number.

From the DC characterization presented previously (Fig. 5.20) the transconduc-

tance of the devices can be calculated as the derivative of Igs with respect to Ugs
(Fig. 5.22). The transconductance around the Dirac point approaches zero because
the dependence between conductivity and Ugs is dominated by n0 . From Eq. 2.22
it is clear that the current around the Dirac point is constant with Ugs . Away from
the Dirac point, the current increases linearly with Ugs and the transconductance
approaches a maximum value which is proportional to the interface capacitance and
the carrier mobility (µ) of the carriers (Eq. 2.26). As introduced previously (section
3.3.1), the transconductance does not reach this value due to the effect of Rs . Fur-
thermore, as the conductivity of the graphene increases, the total resistance of the
transistor becomes dominated by the contact resistance and the transconductance
decreases towards zero.

5.4.2 Transconductance of circular and rectangular graphene transis-

tors

From the previous basic understanding of the transconductance measurements, the

effect of the geometry on the transconductance can be studied. For this purpose four
different probes were fabricated containing both circular and rectangular transistors.
In each probe, the diameter Φ of circular transistors was equal to the width W of

106
Figure 5.22: Transconductance of a circular g-SGFET of Rout = 40 µm and Rin =
27 µm measured during the 15th cycle and normalized by Uds .

the rectangular ones. In this way, the transconductance achievable within a limited
spatial extension can be evaluated. The chosen sizes were Φ = W = 60, 50, 40 and
30 µm. The channel lengths are the predicted optimum values in order to achieve
the highest gm (see section 4.1.4 for a detailed description of the design).

Additionally, DUTs were fabricated in order to measure the dependence between

channel length and gm for different transistor sizes. Each DUT contains circular
transistors of Φ = 80 µm and 120 µm and rectangular transitors with the equivalent
W (i.e. W = 80 µm and W = 120 µm). For each of these configurations, the
transistors were fabricated with five different channel lengths. In order to provide
a significant average, four transistors were fabricated for each of these geometrical
parameters (see section 4.1.4 for a precise description of the design).

The In-vivo Probes were measured as described in section 4.2.3 using a preamplifi-
cation stage on a PCB connected to a DAQ-card. This system allows to measure up
to 36 transistors simultaneously, therefore all the transistors in the probes (see Fig.
4.4) could be measured simultaneously during 15 cycles. On the other hand, the
80 transistors in the DUT configuration were characterized individually using the
probe system introduced in section 4.2.3. The electrochemical “cleaning” described
in section 5.4.1 could not be performed on these transistors because of the difficulty
of applying many cycles to each transistor individually. In order to properly char-
acterize the transistors it is important to center the Ids vs Ugs curve near Ugs =0 V,
otherwise the leakage currents dominates the measured currents. In order to shift
the Dirac point, the pH of the PBS (100mM KCl) was set at 2.7. It is well known
that graphene in aqueous solution is sensitive to pH; by changing the pH from 7 ot

107
2.7, a shift of 70 mV in the Dirac point is expected [92].
From the current-voltage curves, the transconductance was extracted by calculating
dIds /dUgs . Fig. 5.23 shows the averaged gm for both circular and rectangular tran-
sistors in each of the four In-vivo Probes. It is possible to observe an approximately
2-fold increase of the transconductance in the circular transistors. Nevertheless, it
has to be emphasized that the values of the transconductance are much lower than
the predicted ones (see section 3.3.2).
Fig. 5.24a shows the averaged Ids vs Ugs characteristics of circular transistors with
Φ = 120 µm and different channel lengths. Fig. 5.24b shows the average of the
transconductance as a function of channel length for each of the transistor config-
urations. It is possible to observe very large transconductance values, significanly
above the expected values from the simulations presented in section 3.3.2. Addi-
tionally, the optimal channel lengths are shorter than theoretically expected and
the decrease of the transconductance at short channel lengths cannot be observed.
Table 5.2 summarizes the measured maximum transconductances in the In-vivo
Probe configuration. The results are compared with the values predicted in section
3.3.2 showing that the experimental values are much lower than expected.

Simulated Measured
W, Φ (µm) Geometry L, Rout -Rin (µm)
max(gm ) (mS/V) max(gm ) (mS/V)
circular 10 2.21 0.5
60
rectangular 20 1.2 0.2
circular 13 3.34 0.39
80
rectangular 20 1.6 0.18
circular 15 4.46 0.51
100
rectangular 20 2.1 0.29
circular 18 5.4 1.23
120
rectangular 20 2.5 0.52

Table 5.2: Comparison of experimental and predicted maximum transconductances

for the transistors of the In-vivo Probe configuration. The simulated results are
recovered from section 3.3.2 taking into account the incomplete circular geometry
(θ).

Table 5.3 shows the measured and predicted values of maximum transconductance
for the transistors in the DUT configuration. The maximum of transconductance is
taken for the optimum channel length (L, Rout − Rin) and optimum Ugs .
The small values of gm in the In-vivo Probes can be explained by the large track
resistances (Rt ) measured, which were between 400 Ω and 1 kΩ depending on the
device. These values largely exceed the expected 50 Ω used for simulating the per-
formance of the devices in section 3.3.2. Fig. 5.25 shows the effect of adding a
1 kΩ resistance in series with a circular and a rectangular transistor of diameter and
width 120 µm. The simulations show a large drop in the transconductance from the
values predicted in section 3.3.2 to approximately the measured values. The origin

108
 (a) (b)

(c) (d)

Figure 5.23: Transconductance of circular (black) and rectangular (red) transistors

normalized by Uds . a) Comparison of averaged transconductance obtained from
circular transistors (Φ = 60 µm, Rout − Rin = 10 µm) and rectangular transistors
(W = 60 µm, L = 20 µm). b)Comparison of averaged transconductance obtained
from circular transistors (Φ = 80 µm, Rout − Rin = 13 µm) and rectangular tran-
sistors (W = 80 µm, L = 20 µm).c) Comparison of averaged transconductance ob-
tained from circular transistors (Φ = 100 µm, Rout − Rin = 15 µm) and rectangular
transistors (W = 100 µm, L = 20 µm). d) Comparison of averaged transconduc-
tance obtained from circular transistors (Φ = 120 µm, Rout − Rin = 18 µm) and
rectangular transistors (W = 120 µm, L = 20 µm). The averaging was performed
over four circular and four rectangular transistors.

of the high tracks resistances has to be further investigated.

Another indication that the high track resistance is responsible for the small transcon-
ductance observed in the In-vivo Probe configuration is found by the characteriza-
tion of circular and rectangular transistors in the DUTs. In this configuration the
tracks are significantly shorter and its resistance was measured to be approximately
25 Ω. In any case, the transconductance values measured in the DUTs are larger

109
 (a) (b)

Figure 5.24: DC characterization of rectangular and circular g-SGFETs in the DUT

configuration. a) Measured Ids vs Ugs curves of circular transistors with Φ =
120 µm and different channel lengths. The Ids vs Ugs curves were averaged over all
equal transistors (errorbars show ±σ). b) Maximum transconductances measured
in the hole transport regime for the different transistor geometrical parameters and
normalized by Uds . The average was performed over the values from the four equal
transistors of each configuration (errorbars show ±σ).

Simulated Measured
W, Φ (µm) Geometry
max(gm ) (mS/V) max(gm ) (mS/V)
circular 3.0 15
80
rectangular 1.6 6
circular 4.7 21
120
rectangular 2.5 12

Table 5.3: Experimental and predicted maximum transconductances for the tran-
sistors in the DUT configuration. The maximum value is taken considering the bias
Ugs and the geometrical parameters L and Rout − Rin . The simulated results are
recovered from section 3.3.2 taking into account the incomplete circular geometry
(θ).

than the values predicted in section 3.3.2. The reason for this increased transcon-
ductance is probably a short access region which leads to small resistances in series.
This small value is justified by the misalignment between metal layers observed after
the lift-off of the second metal layer (see Fig. 4.8) and also by the fact that the SU8
was partially detached during the measurements (Fig. 5.26). The poor adhesion of
the SU8 might also be the explanation for the increase of leakage current with the
cycle number shown in Fig. 5.21b. In future experiments, an additional hardbake
step in an annealing oven could be perform in order to improve the adhesion of SU8.

The evolution of the Ids vs Ugs curves with the channel length can be simulated for
an access region of Laccess = 1.5 µm in order to better understand the experimental

110
 (a) (b)

Figure 5.25: Simulation of the effect of a large track resistance by adding a 1 kΩ

resistance in series with a circular (rectangular) transistor of diameter (W ) 120 µm
a) Effect on Ids vs Ugs curves. b) Effect on gm .

Figure 5.26: The SU8 layer on a circular transistor is shown. The red rectangle
indicates the place where the break of the insulating layer can be observed. The
entire circle of SU8 covering the inner contact was detached during the measurement.

data. Fig. 5.27 shows the curves for a circular transistor (Φ = 120 µm) with different
channel lengths calculated using the model adapted from [61] (see section 3.3.2).
The simulations show the increase of Ids for short channel lengths; in the limit of a
zero channel length, the minimum current is only limited by the contact resistance
(Fig. 5.27 Rout − Rin = 0.05 µm) leading to tiny modulation of the total transistor
resistance. Oppositively, if the channel length is increased above the optimum value,
gm decreases again due to the decrease of Igs with the channel length. The slope
of Ids with respect to Ugs (i.e gm ) presents a maximum for an intermediate channel

111
length (Lopt ). If the resistance in series (Rs ) was larger the optimum length would
be larger too.

Figure 5.27: Simulated Ids vs Ugs curves for a complete range of channel lengths
(indicated in the legend). Simulations performed for a circular transistor of Φ =
120 µm with a channel length ranging from the maximum (58.5 µm) to a small value
(0.05 µm).

Fig. 5.28 shows the Ids vs Ugs curves and the maximum transconductance simulated
for the same geometrical parameters as shown in Fig. 5.24 and for a Laccess =
1.5 µm. The simulations are not given as a fitting of the data but as supplementary
information in order to show the predictive power of the derived FEM based model.

112
 (a) (b)

Figure 5.28: Simulation of the DC response of rectangular and circular g-SGFETs

in the DUT configuration. a) Simulated Ids vs Ugs curves for a circular tran-
sistor of Φ = 120 µm and varying channel lengths. The access region is set to
Laccess = 1.5 µm. b) Maximum transconductances simulated for different tran-
sistor geometrical parameters and normalized by Uds . The access region is set to
Laccess = 1.5 µm.
6 Summary and outlook

In this work, the principles behind recording and stimulation of neural electric
activity have been presented. In this way the idea of using SGFETs instead of
electrodes for recording has been reassessed, highlighting the importance of the
transconductance of the devices.
The suitability of graphene for the fabrication of g-SGFETs had already been re-
ported. Here, new design rules for improving their sensitivity are derived and the
applicability of graphene-based materials for neural stimulation is evaluated. For
this purpose, a simulation environment based on a finite elements model (FEM)
has been implemented in Comsol Multiphysics r .This model allows to simulate
the electric properties of a graphene sheet with complete freedom in terms of size
and geometry. Additionally, both stationary and time dependent problems can be
solved.
This model was first used for guiding the derivation of the equations describing the
frequency response of graphene electrodes and transistors. The sheet resistance of
graphene is relatively large in comparison with bulk electrodes. Therefore, graphene
electrodes have to be modelled by an equivalent electric circuit constituted by a dis-
tribution of resistive and capacitive elements. The equations governing this system
lead to a non-homogeneous current injection into the electrolyte. This effect has
been studied both in SLG and graphene based porous materials. The study of the
latter has demonstrated that the distribution of elements can be relevant not only
in SLG but also in other electrode materials with a relatively large Rsh /Zint ratio.
The porous materials have also shown large charge injection capacities, close to
the typical values of currently used materials. Despite their high sheet resistance,
the current injection is homogeneous for the frequencies and sizes used for neural
stimulation.
The dependence between the thickness of the materials and their interface capaci-
tance indicates that the charge injection capacity of this materials should be mea-
sured per unit of material volume. However, increasing the thickness of the elec-
trodes could be inefficient for neural stimulation. If the porous structure is thick
with respect to the cell-electrode cleft, the current injected from deep layers of the
porous structure will produce little voltage shifts in the cleft. In order to evaluate
the impact of this effect, a FEM model of an electrogenic cell has been developed in
Comsol r . Appendix 7.5 describes this model, based on a previous work [93], which
describes the electrical properties of cells through the Hodgkin-Huxley equations.
This model accounts for spatial dependencies such as the distance between the cell
and an hypothetical electrode. The coupling of a simulated cell with a model of the
graphene porous structures could be an interesting approach to study the effect of
the thickness of the porous structure.
Achieving effective extracellular neural stimulation with high spatial resolutions is a

114
huge challenge. In order to overcome the limitations imposed by the charge injection
capacity of electrodes,
The first developed model of a single layer graphene sheet has been used, in combi-
nation with other numerical methods [61] and analytical models, for studying the
dependence between gm and the geometrical parameters of g-SGFETs. The relation
between the area of the channel affected by the signal to be recorded and the gate
sensitivity has been derived, showing a linear dependence as long as the properties
of the channel are homogeneous.
Previous works have shown that given a spatial scale of interest (e.g. the size of a
neuron) and a Rs value, there is an optimum W/L value which leads to a maximum
transconductance [94]. In this work a simple expression for the evaluation of this
optimum value is given. Additionally, it was observed that if the optimum L is much
smaller than W , the channel can be folded within the area of interest. This folding
allows to fabricate devices with longer W values, further increasing the W/L ratio.
This principle has been validated by fabricating circular and rectangular g-SGFETs
and comparing their transconductance. The sensitivity of the fabricated devices
closely matches the values predicted by the simulations; values up to 20 mS/V have
been achieved for circular transistors of diameter equal to 120 µm while rectangular
transistors of the same size (W = 120 µm) presented sensitivities of 13 mS/V. In
order to further increase the gm of the devices, the resolution of the microfabrication
techniques should be improved. In this way, the access region could be minimized
and the optimum W/L ratio increased.
At this point, it is important to emphasize that the ultimate goal of a signal record-
ing setup is not to provide a high amplification of the signal but a high S/N ratio.
High transconductances are required in order to preamplify the signal above the
current-to-voltage converter noise. However, increasing the S/N by increasing gm
is limited by the intrinsic noise of the device. It has been reported that an impor-
tant contribution to the intrinsic noise arises from the charge fluctuations at the
graphene-electrolyte interface, which generate a noise signal SI ∝ gm 2 . The design

strategies presented in this work are therefore relevant in order to reach the opti-
mum case in which the noise is dominated by the, in principle unavoidable, charge
fluctuations at the interface. This observation should be taken into account in order
to avoid pursuing the largest transconductance unnecessarily.
Despite the progress reported in this thesis, it has to be proven whether g-SGFETs
can outperform MEAs based recordings or not. In future works, the noise in high
transconductance devices should be characterized in order to determine whether it is
dominated by the charge fluctuations at the graphene-electrolyte interface (SI ∝ gm 2 )

or by the resistance in series (SI ∝ Ids4 ). If the charge fluctuations term dominates,

the S/N ratio would be maximum and no improvement of the transconductance

would be necessary any more. Otherwise, the resolution of the photolithography
steps should be further increased allowing for smaller access regions and higher W/L
ratios. If the length of the access region can be reduced the contact resistance might

115
be dominated by the resistance of the metal-graphene interface. Understanding the
nature of this resistance is of the major goals in order to optimize the design of the
contacts and achieve high S/N ratios. In the field of extracellular neural stimulation,
further research is needed in order to increase the charge injection capacity of MEAs
leading to higher spatial resolutions. The porous materials characterized in this
thesis present promising results, however, further research is needed in order to
determine the effect of the electrodes thickness on the charge injection capacity and
on their efficacy for stimulating neurons. For this purpose, the model presented in
appendix 7.5 could be used. In addition, in case of large electrodes or when applying
short pulses (<0.1 ms) the DE model described in this thesis should be used in order
to assess the effect of non-homogeneous current injection.

116
7 Appendix

7.1 Rectangular electrodes impedance

The equations for the rectangular geometry are derived from the system of equations
(Eq. 7.1 and Eq. 7.2). The variables are defined in section 3.2.

Rsh dU (x, ω)
i(x, ω) = − (7.1)
W dx

W di(x, ω)
U (x, ω) = − (7.2)
Zint (ω) dx

By merging Eq. 7.1 and Eq. 7.2, the following differential equations is obtained,
which defines U (x, ω).

Rsh d2 U (x, ω)
U (x, ω)2 − =0 (7.3)
Zint (ω) dx2

The solution of this differential equation can be written as follows:

r r
R Rsh
− Z sh(ω) x Zint (ω)
x
int
U (x, ω) = C1 (ω)e + C2 (ω)e (7.4)

From the boundary conditions (Eq. 7.5b):

U (0) = Upert (7.5a)

dU (x, ω)
=0 (7.5b)
dx x=0

the integration constants can be derived:

r
Rsh
− L
q
Rsh Zint !
Zint e
C1 (ω) = Upert − C2 (ω) = Upert 1 − q q  (7.6)
2 ZRint
sh
cosh Rsh
Zint L

The impedance of the device can be calculated by dividing the applied voltage
(Upert ) by the current that flows at x = 0 (obtained from Eq. 7.1). The total
impedance can be expressed as:

117
 s 
Upert p Rsh
Ztot + Rs = Zint (ω)Rsh /W 2 cotgh L + Rs (7.7)
Zint (ω)
− dU dx
(x,ω) W
Rsh
x=0

Where Rs represents the series resistance (with contributions from the contact, ac-
cess region and electrolyte resistance). It is noteworthy that Zint (ω)Rsh /W 2 =
Zint (ω)/(W L)Rsh L/W = Zext Rch . Here, Zext [Ω] and Rch [Ω] are the interface
impedance and channel resistance of the entire graphene sheet. On the other hand,
the term in the hyperbolic cotangent can also be simplified to the extensive prop-
Rsh Rsh L/W
erties Zext [Ω] and Rch [Ω]; Zint (ω) = Zint (ω)/(W ∗L) .

7.2 From noise contributions of partitions to total device noise

The contribution of different partitions to the total device noise is calculated in Eq.
3.34. It can be shown that this expression leads to the total noise predicted by the
charge model (Eq. 2.28) described in section 2.2.2.
The transconductance gmwl in section 3.3.4 is proportional to w/l while the transcon-
ductance of the entire channel gmW L is proportional to W/L, therefore:

Lw
gmwl = gmW L (7.8)
Wl

The voltage fluctuations term (δqwl /cwl )2 is inversely proportional to the area of the
partition (i.e. wl) [52]. On the other hand, (δqW L /cW L )2 is inversely proportional
to W L. Thus:
WL
(δqwl /cwl )2 = (δqW L /cW L )2 (7.9)
wl
Eq. 3.35 can be then rewritten as follows:

1 2WL 2 L2 w2 1
nm(δqwl /cwl )2 gm
2
= nm(δqW L /cW L ) g m
wl
n4 wl WL
W 2 l 2 n4

mwL3
= (δqW L /cW L )2 gm
2
= (δqW L /cW L )2 gm
2
(7.10)
WL
n3 l 3 W WL

7.3 Circular electrodes impedance

The analytical model derived in section 3.2 for describing the impedance of circular
graphene transistors was used to fit the experimental data. The experimental data

118
was obtained by collaborators of our group and the following results are included
in the manuscript version of the unpublished paper [62]. The circular electrodes
present a diameter of 1 cm and where fabricated on sapphire. Sapphire is a good
electrical insulator and therefore its influence on the electrochemical properties prop-
erties of the graphene electrode should be negligible.

Fig. 7.1 shows the impedance spectra close and away from the Dirac point (UD =
0 V) and the fitting performed with the distributed elements model (DE model) for
circular electrodes derived in section 3.2.

(a) (b)

Figure 7.1: Module (a) and phase (b) of the impedance of a circular electrode. The
plotted spectra were measured close to (Ugs = 0 V) and away from (Ugs = 0.45 V)
the Dirac point and fitted using the DE model.
119
Fig. 7.2 shows the interface capacitance and sheet resistance extracted from the
fitting.

(a) (b)

Figure 7.2: Extracted interface capacitance (a) and sheet resistance (b) measured
at different Ugs values obtained by fitting the DE model of a circular electrode.

7.4 Capacitive coupling between the Si substrate and contacts/-

graphene sheet

Single layer graphene as well as graphene based porous materials were studied on
Si/SiO2 substrates. In order to minimize the effect of the substrate, a 2 µm thick
SiO2 layer was thermally grown. The thickness of this layer has to be large in order
to avoid capacitive coupling between the Si layer underneath and the graphene and
contacts on top.
Fig. 7.3 shows the impedance spectra from drain to source for a macroscopic (W =
4 mm, L = 6 mm) graphene transistors with and without electrolyte in contact with
the channel. If no electrolyte is present, the current is forced to flow through the
graphene. However, in Fig. 7.3 it is possible to observe how at high frequencies
the module of the impedance decreases. This decrease can be attributed to the
effect of alternative current paths through the Si substrate which become available
due to the capacitive coupling between Si and the contacts/graphene sheet. This
coupling affects the impedance measurements in the presence of electrolyte at high
frequencies. In Fig. 7.3, it is possible to observe an increase of the phase at high
frequencies which cannot be explained by the DE model.

7.5 Numerical modelling of g-SGFET-neuron interface

In this section, a model of an electrogenic cell based on previous works [93] is de-
scribed. The model was developed using the Comsolr modelling environment. The

120
 (a) (b)

Figure 7.3: Impedance from source to drain of a macroscopic (W = 4 mm, L =

6 mm) graphene transistor on Si/SiO2 measured with and without electrolyte (in
the latter Ugs =0 V). (a) shows the module of the impedance and (b) the phase.

geometry of the cell can be defined as desired, however the use of an axisymmetric
structure was found essential for lowering the computational costs to the level of
modest desk computers with a RAM of 4 MB. The conductance of the cell mem-
brane is divided in several contributions; one for each specific kind of ion channels
(Gi ), one for the non-specific leakage (GL ) through the membrane and finally one
accounting for the capacitance of the membrane (CM ) (see Fig. 2.2). Therefore,
the current through specific ion channels can be defined as Gi (UM )(UM − UM rest ),

the leakage currents as GL (Um − UM rest ) and the capacitive currents as C dU /dt.
M M
The voltage dependent conductance for each of the ion species can be calculated by
the Hodgkin and Huxley equations [21]. These equations describe the dynamics of
the ion channels through the opening and closing probability which depends on the
transmembrane voltage (UM ). For example, the potassium channels are constituted
by four domains, and therefore the channels are considered open if the four domains
are in the open state. The probability of being in the open state for each of them
is referred to as n. The probability that the channel is open is therefore n4 . The
change of n with time can be described by first order kinetics and it is a balance
between a positive term proportional to (1 − n) and a negative term proportional
to n:

dn(t)
= αn (1 − n) − βn n (7.11)
dt

Here, αn and βn are the rate constants, and are dependent on UM . Their dependence

121
with UM was found empirically for squid giant neurons to be [21]:

0.01(10 − (UM − UM rest ))

αn =   (7.12a)
rest )
10−(UM −UM
exp 10 −1

rest )

−(UM − Um
βn = 0.125exp ) (7.12b)
80

The conductance of the membrane related to potassium ions can be written as:

+
GK (t, UM ) = Gk+ n(t, UM )4 (7.13)

The dynamics for other ion channels were modelled similarly. For sodium channels,
two kinds of subunits are considered: the so-called activation (m) and inactivation
(h) parameters. As for the potassium channels, the dynamics of m and h are
governed by first order kinetics with αm and αh as opening rate constants and βm
and βh as closing rate constants [21].

The cell structure defined in the Comsolr divides the model in two domains, the
intracellular and extracellular domains. The boundary between the two domains
is defined as the cell membrane. On this boundary, the Hodgkin-Huxley equations
are solved at each mesh element. The difference of the voltage at the boundary
in the intracellular domain and in the extracellular domain is defined as UM and
+
used to define the rate constants. In this way, the evolution of GK (t, UM ) and
+
GN a (t, UM ) with time is found. The net current through the membrane can be
expressed as:

+ +
itot = Gk (UM )(UM −UM
rest
)+GN a (UM )(UM −UM
rest
)+GL (Um −UM
rest
)+CM dUM /dt
(7.14)

This current is modelled by current source at the boundary between the intracellular
and extracellular domains. The injected current drives the cell to the resting po-
tential. If the potential difference is disturbed from the resting potential, an action
potential is elicited. Fig. 7.4 shows a simulation of an action potential for a circular
cell with the first rate constants determined by Hodgkin and Huxley. In order to
elicit the action potential, the transmembrane potential is initially set at -50 mV,
approximately 10 mV above the resting potential. Fig. 7.4b shows the evolution of
n, m and h during the action potential. It is possible to observe the fast opening
of the Na+ channels from the increase of m. This opening leads to a fast intake of
Na+ which shifts the membrane potential towards positive values. The K+ channels
open more slowly, allowing the exit of ions to the extracellular environment. This
opening, in combination with the deactivation of the Na+ , produces a repolarization

122
of the neuron.

(a) (b)

Figure 7.4: The FEM model based on the Hodgkin Huxley equations can reproduce
the electrical properties of neurons. a) Action potential elicited due to an initial
UM = −50 mV, 10 mV above the resting potential. b) The opening probability of
ion channels can be visualized through the simulated n, m and h values.

This model can be coupled with an extracellular current source simulating the effect
of an electrode. In this way, the suitability of an specific electrode for stimulating
electrogenic cells can be simulated. Fig. 7.5 shows the Comsolr model of the cou-
pling between a neuron and a graphene electrode. Here, the neuron was defined
as a semisphere of radius 10 µm on the graphene sheet with a 100 nm thick cleft in
between. Graphene was defined using the parameters indicated in table 3.1 used
for all the simulations performed in this thesis. Applying a voltage ramp at the
graphene electrode, a capacitive current is injected into the cleft. The shift gen-
erated in the voltage cleft ultimately generates a change in UM , and therefore an
action potential is elicited. However, it is possible to see from the simulations 7.5b
that in order to generate an action potential, the voltage ramp must be applied
over 1 ms and its slope must be around 20 kV/s. This leads to a final voltage drop
at the electrode of around 20 V, a value far above the EW of graphene in physio-
logical solution. This simulation indicates the difficult applicability of graphene for
extracellular stimulation.

If the interface capacitance of graphene is increased by two orders of magnitude by

using porous graphene materials (see table 5.1), the required voltage drop should
decrease to 0.2 V which is within the EW. At this point the effect of having a
porous structure should be assessed. If the thickness of the porous material is in
the order or larger than the thickness of the cleft, the current injected through deep
layers of the electrode could generate small changes in the cleft. This effect could
be simulated with the model presented in this annex, introducing a distributed
injection of current instead of having a flat graphene surface.

123
 (a) (b)

Figure 7.5: The model of a cell introduced previously is coupled to the model of a
graphene electrode. a) Structure of the model. The electric potential is represented
by the colour scale. The cell is defined as a semisphere on the graphene sheet.
The model is axisymmetric with respect to the red axis displayed in the figure (r
indicates the distance to the axis). The small green square indicates the position
of the zoomed region shown in the inset. The cleft has a thickness of 100 nm and
the cell a radius of 10 µm. b) An action potential is elicited by applying a voltage
ramp to the graphene electrode. A voltage ramp with a slope of 20 kV/s is applied
during 1 ms (voltage from 0 V to 20 V). This pulse generates a shift in UM which
elicits an action potential.
References
[1] M. Ortiz-Catalan, R. Brånemark, B. Håkansson, and J. Delbeke, “On the via-
bility of implantable electrodes for the natural control of artificial limbs: Review
and discussion,” Biomedical engineering online, vol. 11, no. 1, p. 1, 2012.
[2] N. R. Peterson, D. B. Pisoni, and R. T. Miyamoto, “Cochlear implants and
spoken language processing abilities: Review and assessment of the literature,”
Restorative neurology and neuroscience, vol. 28, no. 2, pp. 237–250, 2010.
[3] E. F. Chang, “Towards large-scale, human-based, mesoscopic neurotechnolo-
gies,” Neuron, vol. 86, no. 1, pp. 68–78, 2015.
[4] M. Teplan, “Fundamentals of eeg measurement,” Measurement science review,
vol. 2, no. 2, pp. 1–11, 2002.
[5] M. E. Spira and A. Hai, “Multi-electrode array technologies for neuroscience
and cardiology,” Nature nanotechnology, vol. 8, no. 2, pp. 83–94, 2013.
[6] S. F. Cogan, “Neural stimulation and recording electrodes,” Annual Review of
Biomedical Engineering, vol. 10, no. 1, pp. 275–309, 2008, pMID: 18429704.
[7] P. Fromherz, A. Offenhäusser, T. Vetter, and J. Weis, “A neuron-silicon
junction- a retzius cell of the leech on an insulated- gate field-effect transis-
tor,” Science, vol. 252, no. 5010, pp. 1290–1293, 1991.
[8] T. D. Y. Kozai, A. S. Jaquins-Gerstl, A. L. Vazquez, A. C. Michael, and X. T.
Cui, “Brain tissue responses to neural implants impact signal sensitivity and
intervention strategies,” ACS Chemical Neuroscience, vol. 6, no. 1, pp. 48–67,
2015, pMID: 25546652.
[9] N. Li, Q. Zhang, S. Gao, Q. Song, R. Huang, L. Wang, L. Liu, J. Dai, M. Tang,
and G. Cheng, “Three-dimensional graphene foam as a biocompatible and con-
ductive scaffold for neural stem cells,” Scientific reports, vol. 3, 2013.
[10] L. Wang, I. Meric, P. Y. Huang, Q. Gao, Y. Gao, H. Tran, T. Taniguchi,
K. Watanabe, L. M. Campos, D. A. Muller, J. Guo, P. Kim, J. Hone,
K. L. Shepard, and C. R. Dean, “One-dimensional electrical contact to a
two-dimensional material,” Science, vol. 342, no. 6158, pp. 614–617, 2013.
[11] L. H. Hess, M. Seifert, and J. A. Garrido, “Graphene transistors for bioelec-
tronics,” Proceedings of the IEEE, vol. 101, no. 7, pp. 1780–1792, July 2013.
[12] B. M. Blaschke, M. Lottner, S. Drieschner, A. B. Calia, K. Stoiber,
L. Rousseau, G. Lissourges, and J. A. Garrido, “Flexible graphene transistors
for recording cell action potentials,” 2D Materials, vol. 3, no. 2, p. 025007,
2016.
[13] G. Buzsaki, Rhythms of the Brain. Oxford University Press, 2006.
https://books.google.es/books?id=ldz58irprjYC

125
[14] L. Tamm, “5.1 biophysics of membranes,” in Comprehensive Biophysics,
E. H. Egelman, Ed. Amsterdam: Elsevier, 2012, pp. 1 – 2. http:
//www.sciencedirect.com/science/article/pii/B9780123749208005014
[15] H. Petrache, “5.2 lipid bilayer structure,” in Comprehensive Biophysics,
E. H. Egelman, Ed. Amsterdam: Elsevier, 2012, pp. 3 – 15.
http://www.sciencedirect.com/science/article/pii/B9780123749208005026
[16] A. B, J. A, L. J, and et al, molecular biology of the cell th edition. Garland
Science, 2002.
[17] Purves and Dale, Neuroscience. Sinauer, 2008.
[18] G. G. Matthews, “Derivation of the nernst equation,” Cellular Physiology of
Nerve and Muscle, Fourth Edition, pp. 208–211.
[19] L. J. e. a. Alberts B, Johnson A, Molecular Biology of the Cell. 4th edition.
Garland Science, 2002, ch. Ion Channels and the Elecrical Properties of Mem-
branes.
[20] S. Oiki, “6.3 gating dynamics of the potassium channel pore,” in Comprehensive
Biophysics, E. H. Egelman, Ed. Amsterdam: Elsevier, 2012, pp. 31 – 67.
http://www.sciencedirect.com/science/article/pii/B9780123749208006147
[21] A. L. Hodgkin and A. F. Huxley, “A quantitative description of membrane
current and its application to conduction and excitation in nerve,” pp. 500–44,
Aug 1952.
[22] I. Willner and E. Katz, Bioelectronics: From Theory to Applications,
ser. Bioelectronics: From Theory to Applications. Wiley, 2006. https:
//books.google.co.uk/books?id=RunxEkuHxzIC
[23] D. Boinagrov, J. Loudin, and D. Palanker, “Strength–duration relationship for
extracellular neural stimulation: Numerical and analytical models,” Journal
of Neurophysiology, vol. 104, no. 4, pp. 2236–2248, 2010.
[24] W. Irnich, “The terms chronaxie and rheobase are 100 years old,” Pacing and
clinical electrophysiology, vol. 33, no. 4, pp. 491–496, 2010.
[25] E. Spinelli and M. Haberman, “Insulating electrodes: a review on biopotential
front ends for dielectric skin–electrode interfaces,” Physiological Measurement,
vol. 31, no. 10, p. S183, 2010.
[26] D. Khodagholy, J. N. Gelinas, T. Thesen, W. Doyle, O. Devinsky, G. G.
Malliaras, and G. Buzsaki, “Neurogrid: recording action potentials from the
surface of the brain,” Nature neuroscience, vol. 18, no. 2, pp. 310–315, 2015.
[27] P. Bergveld, “Development of an ion-sensitive solid-state device for neurophys-
iological measurements,” IEEE Transactions on Biomedical Engineering, vol.
BME-17, no. 1, pp. 70–71, Jan 1970.

126
[28] M. Voelker and P. Fromherz, “Signal transmission from individual mammalian
nerve cell to field-effect transistor,” Small, vol. 1, no. 2, pp. 206–210, 2005.

[29] P. Atkins and J. De Paula, Atkins Physical Chemistry. Oxford University

Press, 2002. https://books.google.es/books?id=9ZUIngEACAAJ

[30] H. Wang and L. Pilon, “Accurate simulations of electric double layer

capacitance of ultramicroelectrodes,” The Journal of Physical Chemistry C,
vol. 115, no. 33, pp. 16 711–16 719, 2011.

[31] B. He, Neural Engineering, ser. Bioelectric Engineering. Springer US, 2007.
https://books.google.es/books?id=iw9pi1otD1EC

[32] P. R. Wallace, “The band theory of graphite,” Phys. Rev., vol. 71, pp.
622–634, May 1947.

[33] D. P. DiVincenzo and E. J. Mele, “Self-consistent effective-mass theory

for intralayer screening in graphite intercalation compounds,” Phys. Rev. B,
vol. 29, pp. 1685–1694, Feb 1984.

[34] K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V.

Dubonos, I. V. Grigorieva, and A. A. Firsov, “Electric field effect in atomically
thin carbon films,” Science, vol. 306, no. 5696, pp. 666–669, 2004.

[35] K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson,

I. V. Grigorieva, S. V. Dubonos, and A. A. Firsov, “Two-dimensional gas of
massless dirac fermions in graphene.” Nature, vol. 438, no. 7065, pp. 197 –
200, 2005.

[36] D. Akinwande, L. Tao, Q. Yu, X. Lou, P. Peng, and D. Kuzum, “Large-area

graphene electrodes: Using cvd to facilitate applications in commercial touch-
screens, flexible nanoelectronics, and neural interfaces.” IEEE Nanotechnology
Magazine, vol. 9, no. 3, pp. 6–14, Sept 2015.

[37] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K.

Geim, “The electronic properties of graphene,” Rev. Mod. Phys., vol. 81, pp.
109–162, Jan 2009.

[38] C. Lee, X. Wei, J. W. Kysar, and J. Hone, “Measurement of the elastic

properties and intrinsic strength of monolayer graphene,” Science, vol. 321,
no. 5887, pp. 385–388, 2008.

[39] P. YU and M. Cardona, Fundamentals of Semiconductors: Physics and

Materials Properties, ser. Advanced texts in physics. Springer Berlin
Heidelberg, 2005, no. v. 3. https://books.google.es/books?id=W9pdJZoAeyEC

[40] S. Reich, J. Maultzsch, C. Thomsen, and P. Ordejón, “Tight-binding

description of graphene,” Phys. Rev. B, vol. 66, p. 035412, Jul 2002.

127
[41] B. Zhan, C. Li, J. Yang, G. Jenkins, W. Huang, and X. Dong, “Graphene
field-effect transistor and its application for electronic sensing,” Small, vol. 10,
no. 20, pp. 4042–4065, 2014.
[42] I. Meric, M. Y. Han, A. F. Young, B. Ozyilmaz, P. Kim, and K. L.
Shepard, “Current saturation in zero-bandgap, top-gated graphene field-effect
transistors,” Nat Nano, vol. 3, no. 11, pp. 654–659, Nov. 2008.
[43] S. Adam, E. H. Hwang, V. M. Galitski, and S. Das Sarma, “A self-consistent
theory for graphene transport,” Proceedings of the National Academy of
Sciences, vol. 104, no. 47, pp. 18 392–18 397, 2007.
[44] J. Martin, N. Akerman, G. Ulbricht, T. Lohmann, J. v. Smet, K. Von Klitzing,
and A. Yacoby, “Observation of electron–hole puddles in graphene using a
scanning single-electron transistor,” Nature Physics, vol. 4, no. 2, pp. 144–148,
2008.
[45] S. Samaddar, I. Yudhistira, S. Adam, H. Courtois, and C. Winkelmann,
“Charge puddles in graphene near the dirac point,” Physical review letters,
vol. 116, no. 12, p. 126804, 2016.
[46] S. Luryi, “Quantum capacitance devices,” Applied Physics Letters, vol. 52,
no. 6, pp. 501–503, 1988.
[47] T. Fang, A. Konar, H. Xing, and D. Jena, “Carrier statistics and quantum
capacitance of graphene sheets and ribbons,” Applied Physics Letters, vol. 91,
no. 9, p. 092109, 2007.
[48] J. Xia, F. Chen, J. Li, and N. Tao, “Measurement of the quantum capacitance
of graphene,” Nat Nano, vol. 4, no. 8, pp. 505–509, Aug. 2009.
[49] C. Mackin, L. H. Hess, A. Hsu, Y. Song, J. Kong, J. A. Garrido, and T. Pala-
cios, “A current x2013;voltage model for graphene electrolyte-gated field-effect
transistors,” IEEE Transactions on Electron Devices, vol. 61, no. 12, pp. 3971–
3977, Dec 2014.
[50] F. Mazda, Electronics Engineer’s Reference Book. Newnes, 1989. https:
//books.google.es/books?id=imRFAAAAYAAJ
[51] A. A. Balandin, “Low-frequency 1/f noise in graphene devices,” Nat Nano,
vol. 8, no. 8, pp. 549–555, Aug. 2013.
[52] I. Heller, S. Chatoor, J. Männik, M. A. G. Zevenbergen, J. B. Oostinga,
A. F. Morpurgo, C. Dekker, and S. G. Lemay, “Charge noise in graphene
transistors,” Nano Letters, vol. 10, no. 5, pp. 1563–1567, 2010, pMID:
20373788.
[53] J. Tersoff, “Low-frequency noise in nanoscale ballistic transistors,” Nano
Letters, vol. 7, no. 1, pp. 194–198, 2007, pMID: 17212463.

128
[54] L. H. Hess, “Graphene transistors for biosensing and bioelectronics,” Ph.D.
dissertation, Technische Universität München, 2014.
[55] P. Knecht, “Nanoscale graphene transistors for sensing applications,” Master’s
thesis, Technischen Universität München, 2016.
[56] A. González, E. Goikolea, J. A. Barrena, and R. Mysyk, “Review on
supercapacitors: Technologies and materials,” Renewable and Sustainable
Energy Reviews, vol. 58, pp. 1189 – 1206, 2016.
[57] Q. Ke and J. Wang, “Graphene-based materials for supercapacitor electrodes
– a review,” Journal of Materiomics, vol. 2, no. 1, pp. 37 – 54, 2016.
[58] X. Zhang, H. Zhang, C. Li, K. Wang, X. Sun, and Y. Ma, “Recent advances in
porous graphene materials for supercapacitor applications,” RSC Adv., vol. 4,
pp. 45 862–45 884, 2014.
[59] R. Pryor, Multiphysics Modeling Using COMSOL?: A First Principles Ap-
proach, ser. Infinity Science Series. Jones & Bartlett Learning, 2011.
https://books.google.es/books?id=u rL6wBD6IC
[60] P. SESHU, TEXTBOOK OF FINITE ELEMENT ANALYSIS. PHI Learning,
2003. https://books.google.es/books?id=qoXBR13xIQcC
[61] E. Masvidal, “Modelling and fabrication of graphene solution-gated field-effect
transistors for electrophysiological applications,” Master’s thesis, Universitat
Autònoma de Barcelona, 2016.
[62] S. Drieschner, A. Guimerà, R. Cortadella, D. Viana, E. Makrygiannis,
B. Blaschke, J. Vieten, and J. A. Garrido, “Frequency Response of Electrolyte-
gated Graphene Electrodes and Transistors.”
[63] M. E. Orazem and B. Tribollet, Electrochemical impedance spectroscopy. John
Wiley & Sons, 2011, vol. 48.
[64] Y. Tsividis and J. Milios, “A detailed look at electrical equivalents of uniform
electrochemical diffusion using nonuniform resistance–capacitance ladders,”
Journal of Electroanalytical Chemistry, vol. 707, pp. 156 – 165, 2013.
[65] K. Takagi, I. Jikuya, G. Nishida, B. Maschke, and K. Asaka, “A study on the
discretization of a distributed rc circuit model,” in ICCAS-SICE, 2009, Aug
2009, pp. 677–680.
[66] E. W. Weisstein, “Bessel differential equation,” From MathWorld–A Wolfram
Web Resource.
[67] X. Li, W. Cai, J. An, S. Kim, J. Nah, D. Yang, R. Piner, A. Velamakanni,
I. Jung, E. Tutuc, S. K. Banerjee, L. Colombo, and R. S. Ruoff, “Large-area
synthesis of high-quality and uniform graphene films on copper foils,” Science,
vol. 324, no. 5932, pp. 1312–1314, 2009.

129
[68] H. Kim, C. Mattevi, M. R. Calvo, J. C. Oberg, L. Artiglia, S. Agnoli,
C. F. Hirjibehedin, M. Chhowalla, and E. Saiz, “Activation energy paths
for graphene nucleation and growth on cu,” ACS Nano, vol. 6, no. 4, pp.
3614–3623, 2012, pMID: 22443380.
[69] J. Cho, L. Gao, J. Tian, H. Cao, W. Wu, Q. Yu, E. N. Yitamben, B. Fisher,
J. R. Guest, Y. P. Chen, and N. P. Guisinger, “Atomic-scale investigation of
graphene grown on cu foil and the effects of thermal annealing,” ACS Nano,
vol. 5, no. 5, pp. 3607–3613, 2011, pMID: 21500843.
[70] X. Li, Y. Zhu, W. Cai, M. Borysiak, B. Han, D. Chen, R. D. Piner,
L. Colombo, and R. S. Ruoff, “Transfer of large-area graphene films for
high-performance transparent conductive electrodes,” Nano Letters, vol. 9,
no. 12, pp. 4359–4363, 2009, pMID: 19845330.
[71] S. Wang, D. Mao, A. Muhammad, S. Peng, D. Zhang, J. Shi, and Z. Jin,
“Characterization of the quality of metal–graphene contact with contact end
resistance measurement,” Applied Physics A, vol. 122, no. 7, pp. 1–7, 2016.
[72] S. Kaya, P. Rajan, H. Dasari, D. C. Ingram, W. Jadwisienczak, and
F. Rahman, “A systematic study of plasma activation of silicon surfaces
for self assembly,” ACS Applied Materials & Interfaces, vol. 7, no. 45, pp.
25 024–25 031, 2015, pMID: 26509331.
[73] Y.-J. Kang, J. Kang, and K. J. Chang, “Electronic structure of graphene and
doping effect on sio2 ,” Phys. Rev. B, vol. 78, p. 115404, Sep 2008.
[74] A. Capasso, A. D. R. Castillo, H. Sun, A. Ansaldo, V. Pellegrini, and
F. Bonaccorso, “Ink-jet printing of graphene for flexible electronics: An
environmentally-friendly approach,” Solid State Communications, vol. 224,
pp. 53 – 63, 2015.
[75] L. Baptista-Pires, C. C. Mayorga-Martı́nez, M. Medina-Sánchez, H. Montón,
and A. Merkoçi, “Water activated graphene oxide transfer using wax printed
membranes for fast patterning of a touch sensitive device,” ACS Nano, vol. 10,
no. 1, pp. 853–860, 2016, pMID: 26691931.
[76] A. Dahal and M. Batzill, “Graphene–nickel interfaces: a review,” Nanoscale,
vol. 6, no. 5, pp. 2548–2562, 2014.
[77] G. Hong, A. S. Holmes, and M. E. Heaton, “Su8 resist plasma etching and its
optimisation,” Microsystem technologies, vol. 10, no. 5, pp. 357–359, 2004.
[78] T. Dieing, O. Hollricher, and J. Toporski, Confocal Raman Microscopy,
ser. Springer Series in Optical Sciences. Springer Berlin Heidelberg, 2011.
https://books.google.es/books?id=Cv3hu6D0i30C
[79] C. V. Raman and K. S. Krishnan, “A new type of secondary radiation,” Nature,
vol. 121, pp. 501–502, 1928.

130
[80] A. C. Ferrari and D. M. Basko, “Raman spectroscopy as a versatile tool for
studying the properties of graphene,” Nature nanotechnology, vol. 8, no. 4, pp.
235–246, 2013.
[81] J. Jang, New Developments in Photon and Materials Research, ser. Materials
science and technologies. Nova Science Publishers, Incorporated, 2013.
https://books.google.es/books?id=WQW7mAEACAAJ
[82] M. Seifert, “Polycrystalline graphene: Growth and functionalization,” Ph.D.
dissertation, Technische Universität München.
[83] A. A. N. EC08, “Basic overview of the working principle of a potentiostat/gal-
vanostat (pgstat)–electrochemical cell setup,” 2011.
[84] B. B, “Graphene solution-gated field-effect transistors,” Diploma thesis, Tech-
nische Universität München, 2014.
[85] E. Corro, “Respuesta mecánica del grafito bajo condiciones mecanicas ex-
tremas,” Ph.D. dissertation, Universidad Complutense de Madrid, 2011.
[86] S. Stankovich, D. A. Dikin, R. D. Piner, K. A. Kohlhaas, A. Kleinhammes,
Y. Jia, Y. Wu, S. T. Nguyen, and R. S. Ruoff, “Synthesis of graphene-based
nanosheets via chemical reduction of exfoliated graphite oxide,” Carbon,
vol. 45, no. 7, pp. 1558 – 1565, 2007.
[87] C. Fu, G. Zhao, H. Zhang, and S. Li, “Evaluation and characterization of re-
duced graphene oxide nanosheets as anode materials for lithium-ion batteries,”
Int. J. Electrochem. Sci, vol. 8, pp. 6269–6280, 2013.
[88] H. Ji, X. Zhao, Z. Qiao, J. Jung, Y. Zhu, Y. Lu, L. L. Zhang, A. H. Mac-
Donald, and R. S. Ruoff, “Capacitance of carbon-based electrical double-layer
capacitors,” Nature communications, vol. 5, 2014.
[89] C.-H. Kim and C. D. Frisbie, “Determination of quantum capacitance and
band filling potential in graphene transistors with dual electrochemical and
field-effect gates,” The Journal of Physical Chemistry C, vol. 118, no. 36, pp.
21 160–21 169, 2014.
[90] S. M. Song, J. K. Park, O. J. Sul, and B. J. Cho, “Determination of
work function of graphene under a metal electrode and its role in contact
resistance,” Nano Letters, vol. 12, no. 8, pp. 3887–3892, 2012, pMID:
22775270.
[91] Y. Wu, W. Jiang, Y. Ren, W. Cai, W. H. Lee, H. Li, R. D. Piner, C. W.
Pope, Y. Hao, H. Ji, J. Kang, and R. S. Ruoff, “Tuning the doping type and
level of graphene with different gold configurations,” Small, vol. 8, no. 20, pp.
3129–3136, 2012.
[92] A. Bonaccini, “Graphene solution-gated field-effect transistors on flexible sub-
strate,” Master’s thesis, Technischen Universität München, 2015.

131
[93] S. Joucla, A. Glière, and B. Yvert, “Current approaches to model extracellular
electrical neural microstimulation,” Frontiers in computational neuroscience,
vol. 8, p. 13, 2014.

[94] C. Mackin, L. H. Hess, A. Hsu, Y. Song, J. Kong, J. A. Garrido, and T. Pala-

cios, “A current voltage model for graphene electrolyte-gated field-effect tran-
sistors,” IEEE Transactions on Electron Devices, vol. 61, no. 12, pp. 3971–3977,
Dec 2014.

132
8 Acknowledgments

I want to acknowledge the support given by the Ludwig-Maximilians-Universität

and the Catalan Institute of Nanoscience and Nanotechnology and their flexibility
at the moment of defining an internship agreement. I want to thank my supervisor
Prof. Jose A. Garrido for his allways available help in terms of both scientific and
management questions. I am also very grateful to my tutor Prof. Bert Nickel who
has allways helped me when needed.
I also want to thank the members of our group and the Biomedical Applications
group at the National Microelectronics Centre (CNM). I am especially grateful to
Andrea Bonaccini and Eduard Masvidal for their patience and commitment intro-
ducing me to the microfabrication techniques, to Dr. Anton Guimerà for the long
and fruitful discussions about the modelling and design of experiments and to Damià
Viana and Igor Kunze for introducing me to the work of the team and helping with
the correction of this thesis.

133
Declaration of Originality

I hereby declare that this thesis is my own work, and that I have not used any
sources and aids other than those stated in the thesis.
München, 03.10.2016