ALL-DIAMOND NEURAL PROBES FOR ELECTRICAL RECORDING AND
ELECTROCHEMICAL DETECTION
By
Michael William Varney

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Electrical Engineering – Doctor of Philosophy
2014

ABSTRACT
ALL-DIAMOND NEURAL PROBES FOR ELECTRICAL RECORDING AND
ELECTROCHEMICAL DETECTION
By
Michael William Varney
Scientists have long been trying to understand the workings of the brain by measuring
electrical neural activity. The invention of the micro-electrode array (MEA), or neural
probe, was a great step improving neuroscientists’ capabilities of recording neural activity.
Neural probes allow for repeatable and customizable sizing and spacing of electrodes
that is out of reach of wire electrodes. This, combined with improvements in recording
hardware, on-probe electronics, and action potential detection algorithms, has brought
great advancements in the fields of neuroscience and brain-machine interfaces.
The original neural probes were fabricated using silicon, due to the technology
available at the inception of neural probes. However, many researchers have fabricated
neural probes out of various combinations of materials. The most popular materials seem
to be silicon-based materials and polymers such as polyimide, SU-8, and Parylene C.
One material of particular interest, due to its combination of advantageous chemical and
mechanical properties, is polycrystalline diamond. Diamond is a good material for neural
probes because it has a high Young’s modulus (~1011 Pa), it has good biocompatibility, it
is resistant to fouling, and it is chemically inert. The main obstacle to developing diamondbased neural probes is that diamond’s beneficial properties also make it difficult to
micromachine, making it difficult to develop processes for probe fabrication.

One major problem with diamond probe fabrication is that diamond is incompatible
with many materials, due to the high temperatures involved in depositing diamond.
However, because diamond has a large band gap (~5.5 eV), it is a good candidate for
single material MEMS (SMM). SMM devices use only one material for their structural
material, insulators, semiconductors, and conductors. This is accomplished by selectively
doping the material (diamond) to create different electrical properties for different parts of
the structure. By using the SMM process to fabricate neural probes, many of the problems
with incompatible materials are resolved. However, SMM fabrication has its own set of
issues associated with it.
The goal of this work is to develop a process to fabricate single-material diamond
neural probes, fabricate the neural probes, and test the neural probes. A process for
fabrication of single-material diamond has been presented and details on the fabrication
of the probes has been provided. Furthermore, these single-material diamond neural
probes have been used for in-vivo electrical neural recordings and in-vitro electrochemical
detection. The electrical recordings were also analyzed using a continuous wavelet
transform method in order to improve action potential detection. A comparison is also
made between silicon-based neural probes, diamond-based neural probes, and singlematerial-diamond neural probes.
This work marks the first time that SMM neural probes have been fabricated and
tested, the first time that a functional SMM device was tested, and the first time that a
SMM device was used in vivo.

ACKNOWLEDGMENTS

I would like to thank my entire family for their patience, understanding, and support
during this research. Additional thanks to all of my friends for their years of support.
I would like to thank my advisor, Dr. Dean M. Aslam, for his guidance,
encouragement, and support through this research. I would also like to thank my guidance
committee Dr. Tim Hogan, Dr. Wen Li, and Dr. Donna H. Wang for their advice and
support. Thank you to Dr. Norbert Mueller for his financial support through some of this
research. Additional thanks to all of the staff in the ECE department – I could not have
done this research without you.
I would like to thank the members of my research group, especially Dr. Hoyin Chan,
Dr. Zongliang Cao, Dr. Abed Janoudi, and Sean Hatch, for their assistance and
productive discussions.
This work was supported in part by the Engineering Research Centers Program of
the National Science Foundation under Award Number EEC-9986866.

iv

TABLE OF CONTENTS

LIST OF TABLES .......................................................................................................... viii
LIST OF FIGURES .......................................................................................................... ix
CHAPTER 1: Research Motivation and Goals ................................................................ 1
1.1 Introduction ..................................................................................................... 1
1.2 Objectives of this Work ................................................................................... 3
1.3 Dissertation Outline ........................................................................................ 6
REFERENCES ..................................................................................................... 7
CHAPTER 2: Diamond Microfabrication Technologies ................................................. 11
2.1 Introduction ................................................................................................... 11
2.2 Diamond Deposition ..................................................................................... 12
2.3 Conductivity of Undoped Diamond ............................................................... 16
2.4 Patterning Diamond Films ............................................................................ 20
2.5 Conclusion .................................................................................................... 23
REFERENCES ................................................................................................... 24
CHAPTER 3: Neural Probe Fabrication ........................................................................ 30
3.1 Introduction ................................................................................................... 30
3.2 Diamond Neural Probe Fabrication .............................................................. 32
3.3 Single-material Diamond Neural Probe Fabrication ...................................... 46
3.4 Conclusion .................................................................................................... 59
REFERENCES ................................................................................................... 60
CHAPTER 4: Electrochemical Detection Experiments .................................................. 62
4.1 Introduction ................................................................................................... 62
4.2 Results.......................................................................................................... 65
4.3 Conclusion .................................................................................................... 72
REFERENCES ................................................................................................... 73
CHAPTER 5: Electrical Neural Recording Experiments ................................................ 75
5.1 Introduction ................................................................................................... 75
5.2 In-Vivo Neural Recording.............................................................................. 76
5.3 Analysis of Neural Recordings using Threshold Detection Method .............. 86
5.4 Conclusion .................................................................................................... 93
REFERENCES ................................................................................................... 94

v

CHAPTER 6: Action Potential Detection in Low Signal-to-Noise Applications ............. 97
6.1 Introduction ................................................................................................... 97
6.2 Action Potential Detection Techniques ....................................................... 101
6.2.1 Traditional Techniques .................................................................. 101
Thresholding ............................................................................... 101
Band-Pass Filtering ..................................................................... 102
6.2.2 Advanced Techniques .................................................................. 103
Discrete Wavelet Transform ........................................................ 103
Power Detection .......................................................................... 104
Continuous Wavelet Transform ................................................... 105
6.3 Methods ...................................................................................................... 106
6.3.1 Simulating Neural Signals ............................................................. 106
6.3.2 DWT Method ................................................................................. 111
6.3.3 MTEO Method............................................................................... 111
6.3.4 CWT Method ................................................................................. 111
6.3.5 Resolving Action Potentials .......................................................... 111
6.4 Results........................................................................................................ 113
6.5 Conclusion .................................................................................................. 123
REFERENCES ................................................................................................. 124
CHAPTER 7: Analysis of Neural Recording Experiments using the Continuous
Wavelet Transform Method .............................................................................. 126
7.1 Introduction ................................................................................................. 126
7.2 Application of Continuous Wavelet Transform Technique and
Thresholding .......................................................................................... 128
7.3 Results........................................................................................................ 130
7.4 Conclusion .................................................................................................. 160
CHAPTER 8: Summary and Recommendations ......................................................... 162
8.1 Summary of Contributions .......................................................................... 162
8.2 Future Research Areas .............................................................................. 164
CHAPTER 9: Education: Woodcreek Elementary Outreach Program ......................... 166
9.1 Introduction ................................................................................................. 166
9.2 TASEM Summer Camp .............................................................................. 168
9.3 Woodcreek Elementary School-Based TASEM Program ........................... 171
9.4 Results........................................................................................................ 177
9.5 Conclusion .................................................................................................. 181
REFERENCES ................................................................................................. 182

vi

LIST OF TABLES

Table 2-1: Growth parameters for boron-doped-diamond films. ................................... 13
Table 2-2: Growth parameters for undoped-diamond films........................................... 19
Table 2-3: Etch parameters for diamond films. ............................................................. 21
Table 3-1: Fabrication process for diamond neural probes. .......................................... 41
Table 3-2: Fabrication process for single-material diamond neural probes. ................. 56
Table 4-1. Substances electrochemically detected using diamond electrodes. ............ 62
Table 7-1: Comparison of average detection rates for each probe type using the
basic threshold detection method and the CWT detection method................... 158
Table 9-1: TASEM summer camp statistics. ............................................................... 170

vii

LIST OF FIGURES

Figure 1-1: List of tasks completed in this work. ............................................................. 5
Figure 2-1: SEM of diamond grown by MPCVD. ......................................................... 14
Figure 2-2: Raman spectrograph of diamond grown by MPCVD. ............................... 15
Figure 2-3: Graph of diamond film growth rate versus additional O2 flow in
diamond growth environment. ............................................................................ 18
Figure 2-4: SEM of diamond patterned using RIE. ....................................................... 22
Figure 3-1: First part of diamond neural probes fabrication process. (a) p++ Si
doping of backend. (b) Deposition of dielectric layer. (c) Growth of
undoped diamond. (d) Deposition of dielectric layer. (e) Growth of doped
diamond. (f) Patterning of doped diamond. ........................................................ 33
Figure 3-2: Second part of diamond neural probe fabrication process. (g)
Removal of backside oxide. (h) Patterning of undoped diamond. (i)
Deposition and patterning of dielectric layer. (j) Deposition of Ti/Au
interconnects. (k) Deposition and patterning of SiO2 top insulating layer.
(l) Probe release via EDP. .................................................................................. 34
Figure 3-3: Fabricated diamond neural probes: before release (top-left), being
manipulated by forceps (top-right), and on the back of a US one-cent
coin (bottom)....................................................................................................... 40
Figure 3-4: SMM probe fabrication process. (a) Si/SiO2 Substrate. (b)
Undoped and doped diamond growth. (c) Doped diamond patterning.
(d) Undoped diamond growth. (e) Probe patterning. (f) Insulation
patterning. (g) Metal pad deposition. (h) Optional electrode
modification. (i) Probe release. ........................................................................... 47
Figure 3-5: Optical micrograph of probe after first etch step (top) and
scanning electron micrograph of probe after first etch step (bottom). ................. 50
Figure 3-6: Optical micrograph of probe after second growth step (top) and
scanning electron micrograph of probe after second growth step
(bottom). ............................................................................................................. 51
Figure 3-7: Optical micrograph of probe after second etching step. ............................. 52

viii

Figure 3-8: Optical micrograph of a probe after exposing the electrodes
and pads. ............................................................................................................ 52
Figure 3-9: Optical micrograph of probe after pad deposition (top).
Scanning electron micrograph of probe after pad deposition (bottom). .............. 53
Figure 3-10: Comparison of CV curves using different working electrodes
in 1M HClO4, highlighting the improvement of fluorine-terminated
diamond over hydrogen-terminated diamond. Scan rate is 0.2 V/s. ................... 55
Figure 3-11: Image of a released SMM diamond probe on a fingertip. ........................ 55
Figure 4-1. Cyclic voltammogram comparing the potential windows of
oxygen-terminated diamond and fluorine-terminated diamond in
1 M KCl. ..................................................................................................... 66
Figure 4-2: Cyclic voltammograms of 1M KCl performed by 30 electrodes
on 6 different probes. .......................................................................................... 68
Figure 4-3: Impedance spectroscopy of 6 electrodes on a single singlematerial probe. Spectroscopy was performed in vitro, using a 1M
KCl solution. The image on the right acts as a legend, showing
which colors correspond to which electrodes on the fabricated
probe. ................................................................................................................. 69
Figure 4-4: Family of background-subtracted cyclic voltammograms of
krebs solution with varying amounts of norepinephrine (NE),
demonstrating a lower detection limit of 5 nM or less. The inset
shows the cyclic voltammograms before subtracting the krebs
solution background. .......................................................................................... 71
Figure 5-1: Anesthetized guinea pig with bolt affixed to skull. ...................................... 76
Figure 5-2: Anesthetized guinea pig with neural probe inserted in right
audio cortex and speaker placed at left ear. ....................................................... 77
Figure 5-3: Neural recording from the audio cortex of a guinea pig taken
with a single-material diamond neural probe with boron-dopeddiamond Electrode.............................................................................................. 78
Figure 5-4: Neural recording from the audio cortex of a guinea pig taken
with a diamond-based neural probe with a boron-doped-diamond
electrode. ............................................................................................................ 78
Figure 5-5: Neural recording from the audio cortex of a guinea pig taken
with a silicon-based neural probe with an iridium oxide electrode. ..................... 79

ix

Figure 5-6: Electrical neural recordings from the audio cortex of a guinea
pig taken with a single-material diamond neural probe with borondoped-diamond electrodes. ................................................................................ 80
Figure 5-7: Electrical neural recordings from the audio cortex of a guinea
pig taken with a diamond-based neural probe with boron-dopeddiamond electrodes. ........................................................................................... 81
Figure 5-8: Electrical neural recordings taken from the audio cortex of a
guinea pig with a silicon-based neural probe with iridium oxide
electrodes. Electrodes 1 – 8. .............................................................................. 82
Figure 5-9: Electrical neural recordings taken from the audio cortex of a
guinea pig with a silicon-based neural probe with iridium oxide
electrodes. Electrodes 9 – 16. ............................................................................ 83
Figure 5-10: Detected action potentials versus frequency of applied audio
stimulus. ............................................................................................................ 85
Figure 5-11: Histograms showing detected action potential times relative
to the onset of stimulus for a diamond SMM probe with borondoped diamond electrodes. ................................................................................ 88
Figure 5-12: Histograms showing detected action potential times relative
to the onset of stimulus for a diamond-based neural probe with
boron-doped-diamond electrodes. ...................................................................... 89
Figure 5-13: Histograms showing detected action potential times relative
to the onset of stimulus for a silicon-based neural probe with iridium
oxide electrodes. Electrodes 1 – 8...................................................................... 90
Figure 5-14: Histograms showing detected action potential times relative
to the onset of stimulus for a silicon-based neural probe with iridium
oxide electrodes. Electrodes 9 – 16.................................................................... 91
Figure 5-15: Average firing rate versus frequency of applied audio stimulus
for a single-material diamond neural probe. The hashed line
represents the background average firing rate when no stimulus was
applied. ............................................................................................................... 93
Figure 6-1: Graphs of neural recordings taken from the audio cortex of a
guinea pig using (a) an iridium oxide electrode and (b) a borondoped-diamond electrode. .................................................................................. 99

x

Figure 6-2: Demonstration of how thresholding is performed, with the top
row showing the input signal, the second row showing the threshold
being applied, and the bottom row showing the detected neural
spikes, for (a) single thresholding, (b) absolute thresholding, and
(c) double thresholding. .................................................................................... 102
Figure 6-3: Graphs of (a) a simulated neural spike train and (b) an
individual simulated neural spike. ..................................................................... 107
Figure 6-4: Power spectra of (a) a simulated neural spike train, (b)
generated white noise, (c) generated pink noise, and (d) recorded
neural signal shown in Fig. 6-1, from a boron-doped-diamond
electrode. .......................................................................................................... 109
Figure 6-5: Graph of simulated neural spike train with non-stationary pink
noise added with a base SNR of -20 dB. .......................................................... 110
Figure 6-6: True positives detected action potentials (out of 10) versus
SNR for simulated neural signals with added (a) stationary white
Gaussian noise, (b) non-stationary white Gaussian noise, (c)
stationary pink Gaussian noise, and (d) non-stationary pink
Gaussian noise. The cost ratio variable, l, was set to 0, where the
cost of omissions was equal to the cost of false positives. Blue
represents the DWT method, red represents the CWT method, and
green represents the MTEO method. ............................................................... 114
Figure 6-7: False positives detected versus SNR for simulated neural
signals with added (a) stationary white Gaussian noise, (b) nonstationary white Gaussian noise, (c) stationary pink Gaussian noise,
and (d) non-stationary pink Gaussian noise. The cost ratio variable,
l, was set to 0, where the cost of omissions was equal to the cost of
false positives. Blue represents the DWT method, red represents
the CWT method, and green represents the MTEO method. ........................... 115
Figure 6-8: Ratio of false positives to true positives versus SNR for
simulated neural signals with added (a) stationary white Gaussian
noise, (b) non-stationary white Gaussian noise, (c) stationary pink
Gaussian noise, and (d) non-stationary pink Gaussian noise. The
cost ratio variable, l, was set to 0, where the cost of omissions was
equal to the cost of false positives. Blue represents the DWT
method, red represents the CWT method, and green represents the
MTEO method. ................................................................................................. 116

xi

Figure 6-9: True positives (out of 10) versus SNR for l values with cost
ratios ranging from .001 to 1 (cost of omissions is higher than cost
of false positives). Blue represents the DWT method, red represents
the CWT method, and green represents the MTEO method. ........................... 118
Figure 6-10: True positives (out of 10) versus SNR for l values with cost
ratios ranging from 1 to 1000 (cost of omissions is lower than cost
of false positives). Blue represents the DWT method, red represents
the CWT method, and green represents the MTEO method. ........................... 119
Figure 6-11: False positives versus SNR for l values with cost ratios
ranging from .001 to 1 (cost of omissions is higher than cost of false
positives). Blue represents the DWT method, red represents the
CWT method, and green represents the MTEO method. ................................. 120
Figure 6-12: False positives versus SNR for l values with cost ratios
ranging from 1 to 1000 (cost of omissions is lower than cost of false
positives). Blue represents the DWT method, red represents the
CWT method, and green represents the MTEO method. ................................. 121
Figure 6-13: Figure of merit versus SNR for each method for simulated
neural signals with added (a) stationary white Gaussian noise, (b)
non-stationary white Gaussian noise, (c) stationary pink Gaussian
noise, and (d) non-stationary pink Gaussian noise. The figure of
merit is calculated by (1 + [omissions]) * (1 + [false positives]) and
taking the lowest value across all values of l. Blue represents the
DWT method, red represents the CWT method, and green
represents the MTEO method. ......................................................................... 122
Figure 7-1: Neural recordings from the auditory cortex of a guinea pig
taken with a silicon-based neural probe (channels 1-8). The stimulus
periods are highlighted in yellow. Action potentials detected using
the thresholding technique are shown with blue stems. Action
potentials detected using the CWT method are shown with red
stems. ............................................................................................................... 131
Figure 7-2: Neural recordings from the auditory cortex of a guinea pig
taken with a silicon-based neural probe (channels 9-16). The
stimulus periods are highlighted in yellow. Action potentials detected
using the thresholding technique are shown with blue stems. Action
potentials detected using the CWT method are shown with red
stems. ............................................................................................................... 132

xii

Figure 7-3: Expanded, 5-second view of neural recordings from the
auditory cortex of a guinea pig taken with a silicon-based neural
probe (channels 1-8) from Fig. 7-1. The stimulus periods are
highlighted in yellow. Action potentials detected using the
thresholding technique are shown with blue stems. Action potentials
detected using the CWT method are shown with red stems. ............................ 133
Figure 7-4: Expanded, 5-second view of neural recordings from the
auditory cortex of a guinea pig taken with a silicon-based neural
probe (channels 9-16) from Fig. 7-2. The stimulus periods are
highlighted in yellow. Action potentials detected using the
thresholding technique are shown with blue stems. Action potentials
detected using the CWT method are shown with red stems. ............................ 134
Figure 7-5: Neural recordings from the auditory cortex of a guinea pig
taken with a diamond-based neural probe (channels 1-8). The
stimulus periods are highlighted in yellow. Action potentials detected
using the thresholding technique are shown with blue stems. Action
potentials detected using the CWT method are shown with red
stems. ............................................................................................................... 135
Figure 7-6: Expanded, 5-second view of neural recordings from the
auditory cortex of a guinea pig taken with a diamond-based neural
probe (channels 1-8) from Fig. 7-5. The stimulus periods are
highlighted in yellow. Action potentials detected using the
thresholding technique are shown with blue stems. Action potentials
detected using the CWT method are shown with red stems. ............................ 136
Figure 7-7: Neural recordings from the auditory cortex of a guinea pig
taken with a single-material diamond neural probe (channels 1-6).
The stimulus periods are highlighted in yellow. Action potentials
detected using the thresholding technique are shown with blue
stems. Action potentials detected using the CWT method are shown
with red stems. ................................................................................................. 137
Figure 7-8: Expanded, 5-second view of neural recordings from the
auditory cortex of a guinea pig taken with a single-material diamond
neural probe (channels 1-6) from Fig. 7-7. The stimulus periods are
highlighted in yellow. Action potentials detected using the
thresholding technique are shown with blue stems. Action potentials
detected using the CWT method are shown with red stems. ............................ 138
Figure 7-9: Histograms showing threshold-detected action potential times
relative to the onset of stimulus for a silicon-based neural probe with
iridium oxide electrodes. Electrodes 1 – 8. ....................................................... 140

xiii

Figure 7-10: Histograms showing threshold-detected action potential times
relative to the onset of stimulus for a silicon-based neural probe with
iridium oxide electrodes. Electrodes 9 – 16. ..................................................... 141
Figure 7-11: Histograms showing threshold-detected action potential times
relative to the onset of stimulus for a diamond-based neural probe
with boron-doped-diamond electrodes. ............................................................ 142
Figure 7-12: Histograms showing threshold-detected action potential times
relative to the onset of stimulus for a single-material neural probe
with boron-doped-diamond electrodes. ............................................................ 143
Figure 7-13: Histograms showing CWT-detected action potential times
relative to the onset of stimulus for a silicon-based neural probe with
iridium oxide electrodes. Electrodes 1 – 8. ....................................................... 144
Figure 7-14: Histograms showing CWT-detected action potential times
relative to the onset of stimulus for a silicon-based neural probe with
iridium oxide electrodes. Electrodes 9 – 16. ..................................................... 145
Figure 7-15: Histograms showing CWT-detected action potential times
relative to the onset of stimulus for a diamond-based neural probe
with boron-doped-diamond electrodes. ............................................................ 146
Figure 7-16: Histograms showing CWT-detected action potential times
relative to the onset of stimulus for a single-material neural probe
with boron-doped-diamond electrodes. ............................................................ 147
Figure 7-17: Histograms comparing detection using the thresholding
method and the CWT method for a silicon-based neural probe with
iridium oxide electrodes. Electrodes 1 – 8. The number of detected
action potential times relative to the onset of stimulus is shown for
thresholding in blue (darker) and for CWT in cyan (lighter). ............................. 149
Figure 7-18: Histograms comparing detection using the thresholding
method and the CWT method for a silicon-based neural probe with
iridium oxide electrodes. Electrodes 9 – 16. The number of detected
action potential times relative to the onset of stimulus is shown for
thresholding in blue (darker) and for CWT in cyan (lighter). ............................ 150
Figure 7-19: Histograms comparing detection using the thresholding
method and the CWT method for a diamond-based neural probe
with boron-doped-diamond electrodes. The number of detected
action potential times relative to the onset of stimulus is shown for
thresholding in blue (darker) and for CWT in cyan (lighter). ............................. 152

xiv

Figure 7-20: Histograms comparing detection using the thresholding
method and the CWT method for a single-material diamond neural
probe with boron-doped-diamond electrodes. The number of
detected action potential times relative to the onset of stimulus is
shown for thresholding in blue (darker) and for CWT in cyan (lighter). ............. 154
Figure 7-21: Bar graphs comparing average detection rates across all
good electrodes using the thresholding method and the CWT
method for each probe type. The number of detected action
potential times relative to the onset of stimulus is shown for
thresholding in blue (darker) and for CWT in cyan (lighter). ............................. 157
Figure 9-1: Student team working on programming their robot. ................................. 171
Figure 9-2: Students demonstrating the robot they designed, built and
programmed. .................................................................................................... 173
Figure 9-3: Student presenting his project in front of other students and
parents. ............................................................................................................ 174
Figure 9-4: Students teleconferencing with Professor Dean Aslam, from
Michigan State University, who is presenting a demonstration. ........................ 176
Figure 9-5: Preferred career of polled TASEM program graduates. ........................... 178
Figure 9-6: Histogram of ratings of the Woodcreek TASEM program by
Woodcreek graduates. ..................................................................................... 180

xv

CHAPTER 1:
Research Motivation and Goals

1.1

Introduction
Neural probes have been of great interest over the past decade due to their

overwhelming potential for use in neuroprosthesis [1,2], the treatment of neurological
disorders [3], or further unraveling the mysteries of the brain. Over the past half-century,
neural probes have evolved from individual insulated wires to precision-micromachined,
three-dimensional arrays of electrodes (for a review of early neural micro-electrode arrays
[MEAs], see Kovacs et al. [4]). Modern neural probes are fabricated from a variety of
materials, the most popular being silicon and polymers such as polyimide, SU-8, and
Parylene C.
Many research groups are developing unique processes for creating neural probes
[4-16]. However, silicon can be an undesirable probe material as, without modification, it
has been found to have poor flexibility, solubility in water, and can induce undesirable
glial responses [17]. Many groups have solved this problem through coating silicon with
biocompatible materials. As an alternative to coating silicon, undoped diamond has been
explored as a substrate material due to its structural properties and high biocompatibility
[18].
Probes often have different materials serving as electrodes. All neural probes use an
inert, metallic (or metalized) substance as their electrodes. Common electrode materials
for electrical sensing applications include gold, platinum and iridium. Iridium-oxide
electrodes and boron-doped poly-C electrodes have both been shown to be useful for the
1

detection of neurotransmitters [19,20]. However, iridium-oxide electrodes are used as
quasi-reference electrodes, whereas boron-doped-diamond electrodes are used as
working electrodes. Boron-doped poly-C electrodes could potentially be used for both
electrochemical detection and for recording electrical signals. Chan et al. [21,22]
developed a process for fabricating neural probes using undoped poly-C as the structural
material and boron-doped poly-C as the electrode material. Non-diamond materials are
used for interconnects and electrical-insulation layers. This probe has been used to
measure electrical neural activity, as well as to electrochemically detect neurotransmitters
in vitro.
This work is focused on the development of a neural probe process using a single
material MEMS (SMM) concept [23-25]. SMM is a subset of MEMS technologies where
MEMS devices are fabricated from a sole semiconductor with selective doping to achieve
the electrical behavior desired. A microsystem merely has to contain single-material
subsystems or devices to be considered an SMM device. Advantages associated with
SMM design include simplified mechanical design, reduced mask count, and reduced
equipment requirements.
Through SMM, the fabrication cost can be reduced by using as few masks and
materials as possible. Poly-C is good candidate for SMM because of its unique
combination of physical properties and the feasibility of selectively growing poly-C to be
a semiconductor or insulator.
The SMM concept can be applied to neural probe fabrication. Neural probes can be
fabricated using as few as three or four masks and one material. In SMM diamond neural
probe, undoped diamond (107 ohm-cm) serves not only as a structural layer, but also as

2

the insulating layer. Highly boron-doped diamond (10-3 ohm-cm) is utilized for
interconnects and electrodes.

3

1.2

Objectives of this Work
In this work, there are 3 main objectives which all represent seminal work. These

objectives can be further broken down into 11 tasks, as seen in Fig. 1-1. The objectives
of the work are as follows:
1. Development and fabrication of SMM diamond neural probes – Developing a
process and relevant techniques in order to produce SMM neural probes.
Furthermore, fabricating SMM neural probes which are the first of their kind, and
the first SMM BioMEMS device. This objective can be divided into 3 tasks:
a. Fundamentals of diamond micromachining
b. Diamond probe fabrication
c. SMM diamond probe fabrication – seminal
2. Demonstration of functional SMM diamond neural probes – Using SMM diamond
probes for both in-vitro electrochemical detection and in-vivo neural recording
applications. This is the first time that SMM diamond probes have been tested and
the first SMM device to ever be used in vivo. This is divided into 2 tasks:
a. In-vitro electrochemical detection experiment using SMM diamond neural
probes – seminal
b. In-vivo neural recording experiments using SMM diamond neural probes –
seminal
3. Determining viability of SMM diamond neural probes – Determining whether the
SMM diamond probes can be used in neural recording experiments. Comparing
the performance of the SMM diamond probes to other probe types. This objective
can be broken down into 5 tasks:
4

a. Analysis of neural recordings using basic thresholding technique (window
discriminator) for silicon, diamond, and SMM diamond neural probes seminal
b. Study of action potential detection methods for low SNR applications
c. Analysis of neural recordings using the continuous wavelet transform
method for silicon, diamond, and SMM diamond neural probes – seminal
d. Comparison between results using basic thresholding technique and
advanced continuous wavelet transform technique for silicon, diamond and
SMM diamond neural probes – seminal
e. Performance comparison between silicon probes, diamond probes, and
SMM diamond probes – seminal

Figure 1-1: List of tasks completed in this work.

5

1.3

Dissertation Outline
This dissertation is divided into 9 Chapters. This first chapter serves to introduce the

work presented, enumerate the objectives of the work, and outline the remainder of the
dissertation. The second chapter details the fundamentals of diamond microfabrication
including diamond deposition, creating highly insulating undoped diamond, and
patterning of diamond films. The third chapter explains and chronicles the development
and fabrication of diamond-based neural probes and SMM diamond neural probes. The
fourth chapter describes electrochemical testing performed using the SMM diamond
neural probe. Chapter 5 describes neural recording experiments performed using the
SMM diamond neural probes and their preliminary results using a basic thresholding
technique. Chapter 6 explores advanced action potential techniques using the discrete
wavelet transform (DWT), the continuous wavelet transform (CWT), and the Teager
energy operator (TEO). The performance of each of these methods is compared for low
SNR applications. The seventh chapter applies the CWT method to the results from
Chapter 5. The results are then used to compare the CWT method to the basic
thresholding method in Chapter 5 and to compare the performance between probe types.
Chapter 8 summarizes the contributions of this work and proposes future research areas.
The final chapter discusses K-12 outreach work that was performed in tandem with the
rest of the work described in this dissertation.

6

REFERENCES

7

REFERENCES

[1] X. Navarro, T.B. Krueger, N. Lago, S. Micera, T. Stieglitz and P. Dario, “A critical
review of interfaces with the peripheral nervous system for the control of
neuroprostheses and hybrid bionic systems,” Journal of the Peripheral Nervous
System, vol. 10, pp. 229-258, 2005.
[2] M.A. Lebedev and M.A.L. Nicolelis, “Brain-machine interfaces: past, present and
future,” Trends in Neurosciences 2006, 29, 536-546.
[3] E. Moro and E.A. Lang, “Criteria for deep-brain stimulation in Parkinson's disease:
review and analysis,” Expert Review of Neurotherapeutics, vol. 6, pp. 1695-1705,
2006.
[4] D.T. Kewley, M.D. Hills, D.A. Borkholder, I.E. Opris, N.I. Maluf, C.W. Storment and
G.T.A. Kovacs, “Plasma-etched neural probes,” Sens. Actuators A, vol. 58, pp. 2735, 1997.
[5] S. Takeuchi, T. Suzuki, K. Mabuchi and H. Fujita, “3D flexible multichannel neural
probe array,” J. of Micromech. and Microeng., vol. 14, pp. 104-107, 2004.
[6] D.R. Kipke, “Implantable neural probe systems for cortical neuroprosthesis,”
Proceedings of the Int. Conf. of the IEEE Eng. in Med. and Bio. Soc., pp. 5344-5347,
2004.
[7] Q. Bai and K.D. Wise, “Single-unit neural recording with active microelectrode arrays,”
IEEE Trans. Biomed. Eng., vol. 48, pp. 911-920, 2001.
[8] P. Norlin, M. Kindlundh, A. Mouroux, K. Yoshida and U.G. Hofmann, “A 32-site neural
recording probe fabricated by DRIE of SOI substrates,” J. of Micromech. and
Microeng., vol. 12, pp. 414-419, 2002.
[9] P.K. Campbell, K.E. Jones, R.J. Huber, K.W. Horch and R.A. Normann, “A siliconbased three-dimensional neural interface: manufacturing process for an intracortical
electrode array,” IEEE Trans. Biomed. Eng., vol. 38, pp. 758-768, 1991.
[10] M.K. Kindlundh, P. Norlin and U.G. Hofmann, “A neural probe process enabling
variable electrode configurations,” Sensors and Actuators B: Chemical, vol. 12, pp.
51-58, 2004.
8

[11] T.H. Yoon, E.J. Hwang, D.Y. Shin, S.J. Parl, S.J. Oh, S.C. Jung, H.C. Shin and S.J.
Kim, “A micro-machined silicon depth probe for multichannel neural recording,” IEEE
Trans. Biomed. Eng., vol. 47, pp. 1082-1087, 2000.
[12] W.L.C. Rutten, H.J. VanWier and J.H.M. Put, “Sensitivity and selectivity of intraneural
stimulation using a silicon electrode array,” IEEE Trans. Biomed. Eng., vol. 38, pp.
192-198, 1991.
[13] K. Chueng, L. Gun, K. Djupsund, D. Yang and L.P. Lee, “A new neural probe using
SOI wafers with topological interlocking mechanism,” Proceedings of IEEE Int. Conf.
on Microtechnologies in Medicine and Biology, pp. 507-511, 2000.
[14] P.S. Motta and J.W. Judy, “Multielectrode microprobes for deep-brain stimulation
fabricated with a customizable 3-D electroplating process,” IEEE Trans. on Biomed.
Eng., vol. 52, pp. 923-933, 2005.
[15] J. Wu, L. Yan, W.C. Tang and F.G. Zeng, “Micromachined electrode arrays with formfitting profile for auditory nerve prostheses,” Proc. of IEEE Eng. in Medicine and
Biology, vol. 5, pp. 5260-5263, 2005.
[16] K.D. Wise, A.M. Sodagar, Y. Yao, M.N. Gulari, G.E. Perlin and K. Najafi,
“Microelectrode, micro-electronics, and implantable neural Microsystems,” Proc. of
the IEEE, vol. 96, pp. 1184-1202, 2008.
[17] D.H. Szarowski, M.D. Andersen, S. Retterer, A.J. Spence, M. Isaacson, H.G.
Craighead, J.N. Turner and W. Shain, “Brain responses to micromachined silicon
devices,” Brain Research, vol. 983, pp. 23-35, 2003.
[18] L. Tang, C. Tsai, W.W. Gerberich, L. Kruckebeu and D.R. Kania, “Biocompatibility of
chemical-vapour-deposited diamond,” Biomaterials, vol. 16, pp. 483-488, 1995.
[19] R.K. Franklin, M.D. Johnson, K.A. Scott, J.H. Shim, H. Nam, D.R. Kipke and R.B.
Brown, “Iridium oxide reference electrodes for neurochemical sensing with MEMS
microelectrode arrays,” IEEE Sensors, pp. 1400-1403, 2005.
[20] G.M. Swain and R. Ramesham, “The electrochemical activity of boron-doped
polycrystalline diamond thin film electrode,” Anal. Chem., vol. 65, pp. 345-351, 1993.
[21] H.-Y. Chan, M. Varney, D.M. Aslam K.D. and Wise, “Fabrication and characterization
of all-diamond microprobes for electrochemical analysis,” Proceedings of IEEE Int.
Conf. on Nano/Micro Engineered and Molecular Syst., pp. 532-535, 2008.
9

[22] H.-Y. Chan, D.M. Aslam, S.H. Wang, G.M. Swain and K.D. Wise, “Fabrication and
testing of a novel all-diamond neural probe for chemical detection and electrical
sensing applications,” In Proceedings of IEEE Int. Conf. on Micro Electro Mechanical
Systems, pp. 244-247, 2008.
[23] Z. Cao and D.M. Aslam, “MEMS structures using polycrystalline diamond singlematerial micro technologies,” Presented at the IEEE Nano Electro Mechanical
Systems, Xiamen, China, 2010.
[24] Z. Cao, M.W. Varney and D.M. Aslam, “Single-material MEMS using polycrystalline
diamond,” Tech. Digest MEMS, pp. 51-54, 2010.
[25] M.W. Varney, Z. Cao and D.M. Aslam, “Fabrication and testing of a novel all-diamond
neural probe for chemical detection and electrical sensing applications,” Presented
at the IEEE Nano Electro Mechanical Systems, Xiamen, China, 2010.

10

CHAPTER 2:
Diamond Microfabrication Technologies

2.1

Introduction
In order to create useful MEMS devices using polycrystalline diamond (poly-C),

fundamental techniques of preparing, growing, and etching poly-C must be learned or
developed. The same can be said of any materials to be processed, however diamond
has some peculiarities that make these fundamentals more critical than some other
materials. For example, diamond growth typically takes place in a high temperature
environment, making many materials and processes incompatible. Also diamond is nearly
impossible to etch chemically requiring aggressive patterning techniques for even simple
patterning. Once the diamond microfabrication fundamentals are understood, neural
probe fabrication (as well as other diamond MEMS device fabrication) becomes possible.
This Chapter will cover the growth of diamond films, resolving aberrant conductivity of
undoped diamond films, and the patterning of diamond films.

11

2.2

Diamond Deposition
Deposition of diamond on non-diamond surfaces requires a seeding step before

diamond growth. This step spreads small diamond particles over the surface to give the
diamond film places to begin to grow from. Typical seeding methods are spraying or
spinning diamond-loaded fluids [1,2] or diamond-loaded water [3], spinning of diamondloaded photoresist [4], surface abrasion [5-7] followed by sonication of diamond-loaded
solution [8], or bias-enhanced nucleation [9-11]. In this study, diamond probes are
fabricated on top of SiO2, so nucleation with diamond-loaded water is ideal as it is
selective to hydrophilic surfaces. The diamond-loaded water used was purchased from
Advanced Abrasives Corporation with an average particle size of 1 µm. The solution was
diluted so that the diamond content was 5% of the volume. Before seeding, the Si/SiO2
substrate is exposed to an oxygen plasma in a plasma asher at a power of 250 W and a
pressure of 500 mTorr to remove any organic residue. (Alternatively a brief etch in 9:1
H20:HF has also been used.) The diamond solution was also prepared using an ultrasonic
bath to agitate the solution for 30 minutes to ensure an even distribution of diamond
particles. Once the solution and substrate are prepared, diamond solution was pipetted
onto the substrate until about 70% of the substrate was covered. At this point, the wafer
was spun at 4000 RPM for 45 seconds to evenly distribute diamond particles over the
substrate. The resulting seeding density was approximately 1011 nuclei/cm2.
Diamond growth was achieved using microwave plasma-assisted chemical vapor
deposition (MPCVD). The system used was a Wavemat MPDR 313EHP with a 2.45 GHz,
5 kW Sairem GMP60KSM microwave power supply. The system chamber measured 9”
in diameter with a quartz bell jar of 5” in diameter. Growth was done in a hydrogen-rich
12

plasma with methane used as a carbon source. Before the growth process, the MPCVD
chamber was pumped to a vacuum of less than 20 mTorr absolute pressure. Once the
chamber had been evacuated, gas flow was initiated and the pressure in the chamber
was maintained at 40 Torr absolute. The main constituent of the growth plasma was H2,
as such H2 had the highest flow rate of 100 sccm. Once the pressure had stabilized, the
MPCVD was started and the output power was slowly increased to 2.3 kW as the
temperature increased to about 700 °C. The source of carbon for diamond growth is CH4
which is next added at a rate of 1.5 sccm. The peak substrate temperature reached during
growth was ~750 °C. A summary of growth conditions can be seen in Table 2-1.
Table 2-1: Growth parameters for boron-doped-diamond films.

Parameters

Values

Microwave Power

2.3 kW, 2.45 GHz

Deposition Pressure

40 Torr

Target Substrate Temperature

750 °C

Gases Used for Growth

Growth Rate

Hydrogen [H2]

100 sccm

Methane [CH4]

1.5 sccm

TMB [0.098% B(CH3)3:H2]

10 sccm

0.10 – 0.15 µm/h

13

Typical diamond growth was performed using H2 and CH4 at a flow rate of 100 sccm
and 1.5 sccm, respectively. Trimethylboron (TMB) [B(CH3)3] can also be included in the
growth environment in order to dope the diamond. The range of sheet resistivity of
diamond is between 10-3 to 102 Ω-cm depending on the amount of TMB present during
growth. Fig. 2-1 shows an example SEM of diamond grown using this method. The quality
of the diamond was verified using Raman spectroscopy. Fig. 2-2 shows an example
Raman spectrograph of samples of diamond grown.

1µm
Figure 2-1: SEM of diamond grown by MPCVD.

14

Figure 2-2: Raman spectrograph of diamond grown by MPCVD.

15

2.3

Conductivity of Undoped Diamond
In order to use diamond for single-material MEMS, it is important to be able to

produce diamond films with high electrical resistivity. In theory, undoped diamond should
have a band gap of ~5.5 eV and the resistivity of natural diamond has been shown to be
~1014 – 1016 Ω-cm [12-14]. However, undoped diamond films grown in a 100:1.5 H2:CH4
environment, as described in section 2.2, were found to have a resistivity on the order of
103 - 105 Ω-cm. The mechanism causing conduction in as-grown diamond films has been
frequently investigated [15-28]. Probable conduction mechanisms suggested include
conduction along the grain boundaries [20,21,25], conduction due to impurities (B, N, or
Si) [19,26], and conduction due to hydrogen traps [28]. Additionally, the surface of asgrown diamond films is typically conductive due to the presence of non-diamond carbon.
However, this surface layer conductivity is easily mitigated by briefly etching the diamond
film or through annealing. There is no consensus on what the primary mechanism of bulk
conduction is in as-grown diamond films. However, the main contributor in the undoped
diamond films in this study is likely unintended boron doping, as the same growth system
is used for highly-doped, lightly-doped, and undoped diamond films.
In order to prevent boron incorporation in undoped diamond films, O 2 was
experimentally added to the growth environment. It is known that O 2 in the growth
environment can increase the resistivity of as-grown diamond films. However, O2 in the
growth environment is also a very effective etchant of diamond. In order to determine
what amounts of O2 were appropriate so that the growth rate of diamond films would not
be significantly impacted by the addition of diamond, a study was conducted where the
amount of O2 in the growth environment was varied and the growth rate of diamond films
16

was measured. In this study, diamond films were grown with growth gases of H2:CH4 with
flow rates of 100:2 sccm. The growth rates for various flowrates of O2 can be seen in Fig.
2-3. It can be seen in Fig. 2-3 that there is a significant decrease in diamond growth rate
if more than 0.5 sccm of O2 is added to the growth gases. For the remainder of this work,
a ratio of 4:1 CH4:O2 was used in the growth of undoped diamond. Typical growth
conditions for undoped diamond films can be seen in Table 2-2. The addition of O2 to the
growth environment increased the resistivity of undoped diamond films by approximately
6 orders of magnitude, from ~103 Ω-cm to >109 Ω-cm. Presumably the oxygen accounts
for this increase in resistivity because the oxygen prevents unwanted boron from
depositing in the diamond lattice. Oxygen may also reduce the number of sp 2 carbon
bonds present in the finished diamond.

17

0.25

Growth Rate [ m/h]

0.2

0.15

0.1

0.05

0

0

0.5

Flow Rate of O2 [sccm]

1

1.5

Figure 2-3: Graph of diamond film growth rate versus additional O2 flow in diamond growth environment.

18

Table 2-2: Growth parameters for undoped-diamond films.

Parameters

Values

Microwave Power

2.3 kW, 2.45 GHz

Deposition Pressure

40 Torr

Target Substrate Temperature

750 °C

Gases Used for Growth

Growth Rate

Hydrogen [H2]

100 sccm

Methane [CH4]

1.5 sccm

Oxygen [O2]

0.38 sccm

~0.10 µm/h

19

2.4

Patterning Diamond Films
Conventional diamond patterning is done using dry etching tools, as wet etching of

diamond is impractical. One of the common methods of etching is reactive ion etching
(RIE) [29-53]. RIE has been thoroughly investigated for patterning diamond films. Typical
gases used to etch are O2 and CF4. Pure O2 etching can be problematic due to columnar
structures being formed [34-36, 45, 47-53]. These structures have been attributed to
differences in chemical composition of diamond grains and grain boundaries, intrinsic
crystal defects, and unintentionally sputtered micro-masks. Adding CF4 to the O2 etch
recipe will increase the etch rate and avoid the formation of the diamond columns [30]
[40]. However, the presence of fluorine in the plasma can damage the underlying layers
once the diamond is partially or entirely etched. In some processes, when the underlying
layers must be preserved, a multi-step etching process can be adopted [50-53]. In a multistep process, the increased etch rate of CF4 can be taken advantage of at the beginning
of the etch process. When the etching approaches the underlying layers, less-destructive
plasma can be used to finish the etching, such as O2 or H2 plasma.
For the purpose of this study, the layers beneath the diamond films (SiO 2 and Si) are
sacrificial and need not be preserved. So the etch recipe used in this work used CF4
plasma. Since this work deals requires thicker diamond films to be etched, the RIE
process requires a metal masking material. The metal that was used was Al as it has a
lower etch rate in the specific etch recipe used for etching diamond. Al was patterned
using liftoff; PR 9260 was used for the liftoff process because its higher thickness allows
for thick layers of Al to be patterned. After developing the photoresist, the photoresist
residue was removed using O2 plasma in a March Asher with a power of 100 W for 3
20

minutes. Next the Al was deposited with a thickness of 1 µm using an Enerjet evaporator.
Liftoff was performed in PRS 2000 which was agitated by ultrasound. Once the Al
masking layer was prepared, the doped diamond was etched in a Plasmatherm 790 RIE
system. The etch process was done at a pressure of 50 mTorr absolute, with 5 sccm CF4
and 20 sccm Ar. The RF etch power was 300 W. The etch rate varies across the sample
but is typically in the range of 0.9-1.5 µm/h. Typical etch parameters used in this work can
be seen in Fig. 2-3. Fig. 2-4 shows an etched diamond film. It can be seen that the
remaining SiO2 layer beneath the diamond film is no longer smooth due to the etching
with CF4 plasma.
Table 2-3: Etch parameters for diamond films.

Parameters

Values

Microwave Power

300 W

Pressure

50 Torr

Gases Used for Etching

Etch Rate

Tetrafluoromethane [CF4]

5 sccm

Argon [Ar]

20 sccm

0.9 – 1.5 µm/h

21

Diamond

SiO2
Figure 2-4: SEM of diamond patterned using RIE.

22

2.5

Conclusion
In this chapter, surface micromachining techniques for fabricating single-material

diamond neural probes were discussed. Diamond growth via MPCVD was described and
the pre-growth nucleation process was discussed. A resolution of the issue of
unintentional conductivity of undoped diamond was presented where oxygen was added
to the diamond growth environment. Typical growth recipes for undoped and doped
diamond films were given. Finally, the patterning process of diamond was described and
a typical etch recipe was presented.

23

REFERENCES

24

REFERENCES

[1] G. S. Yang, M. Aslam, K. P. Kuo, D. K. Reinhard and J. Asmussen, "Effect of
ultrahigh nucleation density on diamond growth at different growth-rates and
temperatures," J. of Vacuum Science & Technology B, vol. 13, pp. 1030-1036,
1995.
[2] A. P. Malshe, R. A. Beera, A. A. Khanolkar, W. D. Brown and H. A. Naseem, "Initial
results of a novel pre-deposition seeding technique for achieving an ultrahigh
nucleation density for CVD diamond growth," Diam. and Rel. Mat vol. 6, pp. 430434, 1997.
[3] Y. X. Tang and D. M. Aslam, “Polycrystalline Diamond Technology and
Piezoresistive Sensor Application for Cochlear Prosthesis,” Ph.D. dissertation,
Dept. Elect. Eng., Mich. State Univ., East Lansing, MI, 2006.
[4] A. Masood, M. Aslam, M. A. Tamor and T. J. Potter, "Techniques for patterning of
CVD diamond films on non-diamond substrates," J. of the Electrochemical Society,
vol. 138, pp. L67-L68, 1991.
[5] A. A. Morrish and P. E. Pehrsson, "Effects of surface pretreatments on nucleation
and growth of diamond films on a variety of substrates," Appl. Phy. Lett., vol. 59, pp.
417-419, 1991.
[6] L. Demuynck, J. C. Arnault, C. Speisser, R. Polini and F. LeNormand, "Mechanisms
of CVD diamond nucleation and growth on mechanically scratched and virgin
Si(100) surfaces," Diamond and Related Materials, vol. 6, pp. 235-239, 1997.
[7] P. A. Dennig, H. Shiomi, D. A. Stevenson and N. M. Johnson, "Influence of substrate
treatments on diamond thin-film nucleation," Thin Solid Films, vol. 212, pp. 63-67,
1992.
[8] E. J. Bienk and S. S. Eskildsen, "The effect of surface preparation on the nucleation
of diamond on silicon," Diam. and Rel. Mat., vol. 2, pp. 432-437, 1993.
[9] F. S. Lauten, Y. Shigesato and B. W. Sheldon, "Diamond nucleation on unscratched
SiO2 substrates," Appl. Phy. Lett., vol. 65, pp. 210-212, 1994.
[10] M. J. Chiang and M. H. Hon, "Positive dc bias-enhanced diamond nucleation with
high CH4 concentration," Diam. and Rel. Mat, vol. 10, pp. 1470-1476, 2001.
[11] M. D. Irwin, C. G. Pantano, P. Gluche and E. Kohn, "Bias-enhanced nucleation of
diamond on silicon dioxide," Appl. Phy. Lett.,, vol. 71, pp. 716-718, 1997.
[12] Halliday, Resnick, Walker, Fifth Edition of Fundamentals of Physics Extended,
Wiley, New York, 1997.
25

[13] H.O. Pierson, Handbook of Carbon, Graphite, Diamond and Fullerenes –
Properties, Processing and Applications, William Andrew Publishing/Noyes, 2004.
[14] C. Beck and M. Ahmed, Diamond – Chemically Vapour Deposited Diamond, Azom,
2004.
[15] Y. Muto, T. Sugino and J. Shirafuji, “Electrical conduction in undoped diamond films
prepared by chemical vapor deposition,” Applied Physics Letters, vol. 59, pp. 843845, 1991.
[16] B. Huang, D.K. Reinhard, “Electric field – dependent conductivity of polycrystalline
diamond thin films,” Applied Physics Letters, vol. 59, pp. 1494-1496, 1991.
[17] M. Werner, O. Dorsch, A. Hinze, E. Obermeier, R.E. Haper, C. Johnston, P.R.
Chalker and I.M. Buckley-Golder, “Space-charge-limited current flow and trap
density in undoped diamond films,” Diamond and Related Materials, vol. 13, pp.
825-828, 1993.
[18] T. Sugino, Y. Muto, J. Shirafuji and K. Kobashi, “Electrical conduction mechanisms
in polycrystalline chemically vapour-deposited diamond films,” Diamond and
Related Materials, vol. 2, pp. 797-802, 1993.
[19] S. Jin and T.D. Moustakas, “Electrical conductivity studies of diamond films
prepared by electron cyclotron resonance microwave plasma,” Applied Physics
Letters, pp. 2354-2356, 1993.
[20] A.K. Kulkarni, A. Shrotriya, P. Cheng, H. Rodrigo, R. Bashyam and D.J. Keeble,
“Electrical properties of diamond thin films grown by chemical vapor deposition
technique,” Thin Solid Films, vol. 253, pp. 141-145, 1994.
[21] A.K. Kulkarni, K. Tey and H. Rodrigo, “Electrical characterization of CVD diamond
thin films grown on silicon substrates,” Thin Solid Films, vol. 270, pp. 189-193,
1995.
[22] G. De Cesare, S. Salvatori, R. Vincenzoni, P. Ascarelli, E. Cappelli, F. Pinzari and
F. Galluzzi, “On the electrical properties of polycrystalline diamond films on silicon,”
Diamond and Related Materials, vol. 4, pp. 628-631, 1995.
[23] A.K. Sikder, A.P. Jacob, T. Sharda, D.S. Misra, M. Pandey, D. Kabiraj and D.K.
Avasthi, “DC electrical conductivity of chemical vapour deposited diamond sheets: a
correlation with hydrogen content and paramagnetic defects,” Thin Solid Films, vol.
332, pp. 98-102, 1998.
[24] L. Wang, Y. Xia, J. Ju and W. Zhang, “Electrical properties of chemical vapor
deposition diamond films and electrical response to X-ray,” Diamond and Related
Mateirals, vol. 9, pp. 1617-1620, 2000.
[25] B.J. Lee, B.T. Ahn, J.K. Lee and Y.J. Baik, “A study on the conduction path in
undoped polycrystalline diamond films,” Diamond and Related Materials, vol. 10,
pp. 2174-2177, 2001.
26

[26] Q. Su, J. Lu, L. Wang, J. Liu, J. Ruan, J. Cui, W. Shi and Y. Xia, “Electrical
properties of [100]-oriented CVD diamond film,” Solid-State Electronics, vol. 49, pp.
1044-1048, 2005.
[27] E.J. Correa, Y. Wu, J. Wen, R. Chandrasekharan and M. Shannon, “Electrical
conduction in undoped ultrananocrystalline diamond thin films and its dependence
on chemical composition and crystalline structure,” Journal of Applied Physics, vol.
102, 2007.
[28] B.R. Stoner, J.T. Glass, L. Bergman, R.J. Nemanich, L.D. Zoltal and J.W.
Vandersande, “Electrical conductivity and photoluminescence of diamond films
grown by downstream microwave plasma CVD,” Journal of Electronic Materials, vol.
21, pp. 629-634, 1992.
[29] G.S. Sandhu, and W.K. Chu, “Reactive ion etching of diamond,” Applied Physics
Letter, vol. 55, no. 5, pp.437-438, 1989.
[30] O. Dorsch, M. Werner and E. Obermeier, “Dry etching of undoped and boron doped
polycrystalline diamond films,” Diamond & Related Materials, vol. 4, pp.456-459,
1995.
[31] C. Vivensang, L. Ferlazzo-Manin, and M.F. Ravet, “Surface smoothing of diamond
membranes by reactive ion etching process,” Diamond & Related Materials, vol. 5,
pp. 840-844, 1996.
[32] Gopi M.R. Sirineni, H.A. Naseem, A.P. Malshe, and W.D. Brown, “Reactive ion
etching of diamond as a means of enhancing chemically-assisted mechanical
polishing efficiency,” Diamond and Related Materials, vol. 6, pp. 952-958, 1997.
[33] H. Shiomi, “Reactive ion etching of diamond in O2 and CF4 Plasma, and fabrication
of porous diamond for field emitter cathodes,” Japanese Journal of Applied Physics,
vol. 36, pp. 7745-7748, 1997.
[34] E.-S. Baik, and Y.-J. Baik, “Aligned diamond nanowhiskers,” Journal of Materials
Research, vol. 15, no. 14, pp. 923-926, 2000.
[35] E.-S. Baik, Y.-J. Baik, S.W. Lee, and D. Jeon, “Fabrication of diamond nanowhiskers,” Thin Solid Films, no. 377-378, pp. 295-298, 2000.
[36] E.-S. Baik, Y.-J. Baik, and D. Jeon, “Control of diamond micro-tip geometry for field
emitter,” Thin Solid Films, no. 377-378, pp. 299-302, 2000.
[37] T. Shibata, Y. Kitamoto, K. Unno, and E. Makino, “Micromachining of diamond film
for MEMS applications,” Journal of Microelectromechanical Systems, vol. 9, no. 1,
pp. 47-51, 2000.
[38] R. Otterbach, U. Hilleringmann, and K. Goser, “Reactive ion etching of CVDdiamond for sensor devices with Al feature size of 100 nm,” Industrial Electronics
Society Conference, vol.3, pp. 1873-1877, 2000.
27

[39] Y. Nishibayashi, Y. Ando, H. Saito, T. Imai, T. Hirao, and K. Oura, "Anisotropic
etching of a fine column on a single crystal diamond," Diamond and Related
Materials, vol. 10, pp. 1732-1735, 2001.
[40] P.W. Leech, G.K. Reeves, and A. Holland “Reactive ion etching of diamond in CF 4,
O2, O2 and Ar-based mixtures,” Journal of Materials Science, vol. 36, no. 14, pp.
3453-3459, 2001.
[41] Y. Fu, H. Du, and J. Miao, “Patterning of diamond microstructures on Si substrate
by bulk and surface micromachining,” Journal of Materials Processing Technology,
132, pp. 73-81, 2003.
[42] X. Wang, G.D. Hong, J. Zhang, B.L. Lin, H.Q. Gong, and W.Y. Wang, "Precise
patterning of diamond films for MEMS application," Journal of Materials Processing
Technology, vol. 127, pp. 230-233, 2002.
[43] Y. Ando, Y. Nishibayashi, K. Kobashi, T. Hirao, and K. Oura, “Smooth and high-rate
reactive ion etching of diamond,” Diamond and Related Materials, vol. 11, pp. 824827, 2002.
[44] Y. Ando, Y. Nishibayashi, A. Sawabe, “‘Nano-rods’ of single crystalline diamond,”
Diamond and Related Materials, vol. 13, pp. 633-637, 2004.
[45] C.Y. Li, and A. Hatta, “Effect of metal coating on the formation of diamond whiskers
in O2 RF plasma,” Diamond & Related Materials, vol. 15, pp. 357-360, 2006.
[46] G. Ding, H. Mao, Y. Cai, Y. Zhang, X. Yao, and X.L. Zhao, “Micromachining of CVD
diamond by RIE for MEMS applications,” Diamond & Related Materials, vol. 14, pp.
1543-1548, 2005.
[47] C.Y. Li, and A. Hatta, “Preparation of diamond whiskers using Ar, O2 plasma
etching,” Diamond & Related Materials, vol. 14, pp. 1780-1783, 2005.
[48] C.Y. Li, and A. Hatta, “Nanowhiskers formation by radio frequency Ar/O2 plasma
etching of Al coated diamond films,” Thin Sold Films, vol. 515, pp. 4172-4176,
2007.
[49] M.D. Stoikou, P. John, and J.I. Wilson, "Unusual morphology of CVD diamond
surfaces after RIE," Diamond and Related Materials, vol. 17, pp. 1164-1168, 2008.
[50] H. Y. Chan, D. M. Aslam, J. A. Wiler, and B. & Casey, “A novel diamond
microprobe for neuro-chemical and-electrical recording in neural prosthesis,”
Microelectromechanical Systems, Journal of, vol. 18(3), pp. 511-521, 2009.
[51] H.Y. Chan, “Polycrystalline CVD Diamond Probes for use in In Vivo and In Vitro
Neural Studies,” Ph.D. dissertation, Dept. Elect. Eng., Mich. State Univ., East
Lansing, MI, 2008.

28

[52] Z. Cao and D. Aslam, “Fabrication technology for single-material MEMS using
polycrystalline diamond,” Diamond and Related Materials, vol. 19(10), pp. 12631272, 2010.
[53] Z. Cao, “Single-material MEMS using Polycrystalline Diamond,” Ph.D. dissertation,
Dept. Elect. Eng., Mich. State Univ., East Lansing, MI, 2011.

29

CHAPTER 3:
Neural Probe Fabrication

3.1

Introduction
As no testing could be conducted without probes, the most critical step in this work

must be the fabrication of the single-material diamond neural probes. But before singlematerial diamond probes could be fabricated, the previous diamond probe technology
had to be repeated in order to learn the processing techniques required and to produce
samples for comparison. This chapter discusses the fabrication process for making
diamond neural probes and single-material diamond neural probes. In order to
understand the purpose of this work, it is important to understand the differences between
the diamond neural probes fabricated in this (and previous) work and the single-material
diamond neural probes that are unique to this work.
Single-material diamond probes are fabricated entirely out of polycrystalline diamond.
However, other materials are used during fabrication and for the devices’ electrical
interface. The six-electrode variant of the single-material diamond probes are
approximately 99.6% diamond by volume with the remaining portion making up the metal
pads on the backend of the probe. Previous “all-diamond” neural probe research [1-4]
were <30% diamond by, using the same probe geometry. The previous “all-diamond”
probes included a Si backend, multiple SiO2 insulating layers, and Au interconnects. Even
discounting the Si backend, the front portion of these probes, which is inserted into the

30

brain, is ~50% diamond by volume. It is immediately apparent that the current research
differs dramatically from earlier research in terms of process and materials used.
While the final single-material diamond probe is comprised almost entirely of
diamond, other materials are utilized during fabrication. Probes are fabricated on top of a
Si substrate with a 1 µm layer of SiO2. Furthermore, Al is used as a masking layer during
each of the diamond patterning steps. Finally, as previously stated, Cr and Au are used
to make the pads on the backend of the probe for the electrical interface. After each etch
step, the Al masking layer is removed, and after the final step, the Si/SiO 2 substrate is
removed.
In this chapter, the fabrication processes for each probe type is discussed. Each
process is illustrated and detailed process steps for each method are presented. In
addition, micrographs of the single-material diamond neural probes are shown. The
design and layout of the diamond-based neural probes has been presented in [4], the
SMM diamond neural probes share the same layout with a reduced number of masks.

31

3.2

Diamond Neural Probe Fabrication
Diamond probe fabrication is accomplished in 12 major steps, beginning with an n-Si

wafer. The Si wafer is selectively doped p++ in order to form the thick Si backend of the
probe (Fig 3-1a). The p++-Si is used as a chemical etch stop during the final probe release
step. After doping the Si, a layer of SiO2 is deposited on the wafer (Fig 3-1b). The wafer
is then seeded with diamond and undoped diamond is grown (Fig. 3-1c). Next an
insulating layer of oxide is deposited on the undoped diamond (Fig 3-1d). The wafer is
again seeded and doped diamond is grown (Fig 3-1e). Next, two diamond patterning
steps are performed. First the doped diamond is patterned in order to form electrodes
(Fig. 3-1f). The backside oxide is etched as it is an unwanted masking material for probe
release (Fig. 3-2g). Then, the underlying undoped diamond is patterned to form the shape
of the probes (Fig. 3-2h). A dielectric layer of SiO2 is deposited and patterned (Fig. 3-2i).
Next, a Ti/Au layer is deposited and patterned using liftoff to form interconnects and
backend pads (Fig. 3-2j). SiO2 is deposited as the top insulating layer and is patterned
using a buffered HF solution (Fig. 3-2k). The final fabrication step is releasing the probes,
which is done by EDP (Fig. 3-2l).

32

p++-Si

a

n-Si
SiO2

b
Undoped diamond

c
SiO2

d
Doped diamond

e

f

Figure 3-1: First part of diamond neural probes fabrication process. (a) p++ Si doping of backend. (b)
Deposition of dielectric layer. (c) Growth of undoped diamond. (d) Deposition of dielectric layer. (e)
Growth of doped diamond. (f) Patterning of doped diamond.

33

g

h
SiO2

i
Ti/Au

j
SiO2

k

l
Figure 3-2: Second part of diamond neural probe fabrication process. (g) Removal of backside oxide. (h)
Patterning of undoped diamond. (i) Deposition and patterning of dielectric layer. (j) Deposition of Ti/Au
interconnects. (k) Deposition and patterning of SiO2 top insulating layer. (l) Probe release via EDP.

34

The first major step in the fabrication of diamond neural probes is defining the
backend of the probe by selectively doping the n-Si wafer heavily with B (p++). This is
done so that the backend of the probes is not etched away during the release step in
EDP. The reason for keeping a layer of Si underneath the backend of the probes is to
make sure that the probes could be handled and bonded without breaking the backend
of the probes. Silicon-based probes have a thicker backend as well for this same reason.
In order to selectively dope the silicon wafer, a layer of oxide was first grown on the wafer.
Before growing the oxide, an RCA clean was performed to remove any metallic or organic
contaminants from the surface of the wafer. The oxide was patterned with buffered HF
using the first mask. Another RCA clean was performed to remove the photoresist and
any contaminants. Deep boron diffusion and drive-in are performed in separate furnaces.
After the boron doping, the masking oxide layer was removed in HF.
The next step in the fabrication process is dielectric deposition. The purpose of this
step is to prevent out-diffusion of boron during the diamond growth steps, which would
lead to unintentional doping of the undoped diamond film. Before depositing oxide,
another RCA clean was performed to remove impurities. This step can be skipped if the
oxide deposition is done immediately after the previous step, as HF will remove any metal
or organic residue from the wafer. To complete this step, LPCVD oxide was deposited, at
450 °C, on the wafer.
Before diamond growth can be achieved, diamond nanoparticles must be spread
over the substrate, a process known as diamond seeding (or nucleation). Seeding was
performed using a 0.5% solution of ~0.1 µm diamond particles in water. Before seeding
can effectively be performed on SiO2, the surface needs to be refreshed to become
35

hydrophilic. This was done by briefly etching the SiO2 surface in diluted hydrofluoric acid
for a short time (<30 seconds). The diamond solution must also be prepared, which has
been done using an ultrasonic bath to agitate the solution for 30 minutes to ensure an
even distribution of diamond particles. Once the solution and substrate are prepared,
some diamond solution was pipetted onto the substrate until about 70% of the substrate
was covered. At this point, the wafer was spun at 500 RPM for 5 seconds and then 2000
RPM for 30 seconds to evenly distribute diamond particles over the substrate. Diamond
growth was accomplished using microwave-plasma-assisted chemical vapor deposition
(MPCVD). Before the growth process, the MPCVD chamber was pumped to a vacuum of
less than 20 mTorr absolute pressure. Once the chamber has been evacuated, gas flow
was initiated and the pressure in the chamber was maintained at 40 Torr absolute. The
main constituent of the growth plasma was H2, as such H2 has the highest flow rate of
100 sccm. Once the pressure has stabilized, the MPCVD was started and the output
power was slowly increased to 2.3 kW as the temperature increases to about 750 °C. The
source of carbon for diamond growth was CH4 which was next added at a rate of 1.5
sccm. Undoped diamond was allowed to grow for 20 hours to form the structural backing
of the probes. After the diamond growth process there was a layer of undoped diamond
that was approximately 3 µm thick. The quality of diamond was experimentally verified
using Raman spectroscopy to verify the main peak of Raman shift was at 1332 cm-1, as
expected.
Next another layer of oxide was deposited on top of the undoped diamond. This
dielectric layer prevents conduction between interconnects through the undoped
diamond. This is necessary due to the poor resistivity (~102-103 Ω-cm) of the undoped
36

diamond. A layer of PECVD oxide with a thickness of ~1 µm was deposited on the wafer.
Annealing of the oxide was done using rapid thermal processing, 10 minutes at 650 °C,
10 minutes at 750 °C, and finally 3 minutes at 970 °C.
Next doped diamond was grown. Because the top layer of the wafer was oxide, the
wafer must first be seeded as described before for undoped diamond growth. The growth
parameters in the MPCVD are almost the same, except the gases used are different. In
doped diamond growth the gases used are H2, CH4, and trimethlyborane [B(CH3)3] diluted
in H2 to 0.005% as a boron source. The flow rates used are 100 sccm of H 2, 1 sccm of
CH4, and 2 sccm of diluted B(CH3)3. The diamond was grown for 13 hours and the
resulting layer was approximately 1 µm thick.
The next fabrication step is patterning the doped diamond layer to form the electrodes
of the probes. Diamond was patterned using RIE with Al as a masking layer. For this
mask, photoresist 1827 was patterned using the second mask. Then 1 µm of Al was
deposited on the sample. The Al was patterned through a lift-off process. Diamond was
then etched in O2 plasma at 50 mTorr with 300 W of RF power for 2 hours. Etching
diamond in pure O2 plasmas leaves columnar microstructures of diamond called diamond
whiskers. These diamond whiskers were removed by immersing the sample in diluted HF
for 10 seconds. Finally, the Al mask was removed using Al Etch A, a combination of
H3PO4, HNO3, and C2H4O2.
Next the backside oxide was etched. This step is important to the probe fabrication
because oxide has a very low etch rate in EDP, which is used to release the probes.
Leaving the oxide on the backside of the wafer would make the release process take

37

much longer. The front side of the wafer was protected by spinning photoresist 9260. The
oxide was etched using buffered HF for 15 minutes. The photoresist was then removed
using acetone.
The next fabrication step is to pattern the undoped diamond to from the probe
sheaths. Again, an Al mask was used for the RIE of diamond. However, for this step the
mask was patterned using wet etching of Al rather than lift-off. First, 1 µm of Al was
deposited on the wafer. Then the negative photoresist 5214 was patterned using mask 3.
The Al was patterned using Al Etch A and the photoresist was removed using acetone.
RIE was performed with a more aggressive recipe that had 20 sccm of O 2 and 5 sccm of
CF4, at 50 mTorr with 300 W of RF power, for 3 hours. The Al mask was then removed
using Al etch A.
Next a dielectric layer of oxide was deposited and patterned. 1 µm of PECVD oxide
was deposited on the wafer. The second mask (boron-doped diamond) was used in order
to expose the electrodes. For this mask, photoresist 9260 was used as a mask. The oxide
was etched by RIE using a typical SiO2 etching recipe for 50 minutes. The remaining
photoresist residue was removed using acetone and ashing in O 2 plasma. The dielectric
layer was also patterned using the third mask (undoped diamond). For this mask,
photoresist 5214 had to be used. However, since 5214 is a much thinner photoresist than
9260, the patterning of photoresist and etching of oxide had to be done twice. The 5214
would be etched completely in 50 minutes of RIE, so there were two separate RIE
sessions of 25 minutes each. The photoresist residue between RIE and after RIE was
removed using acetone and O2 plasma ashing.

38

At this point, additional steps could be taken to deposit electrodes of different metals.
In the past, probes had been fabricated with Pt and Ag electrodes in addition to borondoped diamond electrodes for electrochemical detection applications. These electrodes
would be deposited using a lift-off process as described before. Each electrode type
requires an additional mask.
Next the metal interconnects and contact pads were deposited on the probes. This
was done with a lift-off process using photoresist 9260. The photoresist was patterned
with the fourth mask (interconnects and pads). Ti and Au were deposited on the probes
with thickness of 500 Å and 4000 Å respectively. The Ti layer was necessary because it
has better adhesion to diamond and creates an ohmic contact with boron-doped diamond.
Lift-off was performed in acetone.
The next fabrication step is the deposition and patterning of the top oxide insulating
layer. This layer protects the interconnects so that only the electrodes and backend pads
are exposed. 2 µm of PECVD oxide was deposited on the sample. Photoresist 9260 was
patterned using the fifth mask (top insulator). The oxide layer was etched using RIE with
a standard oxide etch recipe for 90 minutes. The remaining photoresist residue was
removed using acetone and O2 plasma ashing.
The final fabrication step is releasing the probes. This is done by first thinning the
wafer using HF and nitric acid. Before thinning the wafer, the front side is protected by a
layer of wax. The wafer was then thinned in the acid mixture for 30 minutes. Then the
wax was removed and the probes were released by dissolving the remainder of the n-Si
using ethlyenediamine pyrocatechol (EDP). Finally the released probes were thoroughly

39

rinsed. At this point the diamond neural probe fabrication is complete and the probes are
ready for packaging and testing. Pictures of these probes can be seen in Fig. 3-3. The
complete fabrication process is detailed in Table 3-1.

Figure 3-3: Fabricated diamond neural probes: before release (top-left), being manipulated by forceps
(top-right), and on the back of a US one-cent coin (bottom).

40

Table 3-1: Fabrication process for diamond neural probes.
Process Step

Detail

Pre-furnace
Clean

Recipe: DWD/TCA
Parameter Table: DWDSKIN
4 hours, 1000 °C

Spin 1813

Photoresist
thickness ≈1.4
µm

Spinner

4000 RPM, 30 seconds

Hotplate

1 minute, 95 °C

MA6

5 seconds

Develop
Photoresist

Wet Bench

MIF 318, 1 minute

Oxide Etch

Wet Bench

BHF, 17 minutes

Pre-furnace
Clean

Wet Bench

RCA Clean

Deep Boron
Diffusion

A2 tube

Drive-in

A4 tube

Mask 1: BDF

Strips deep
boron masking
oxide

Pre-furnace
Clean
LPCVD Oxide

Oxide thickness
≈1.2 µm

Refresh Oxide
Surface

Dielectric
Deposition

RCA Clean of n-Si wafer

B2 tube

Oxide Etch

Undoped
Diamond
Growth

Wet Bench
Oxide thickness
≈1.2 µm

Expose

Dielectric
Deposition

Parameters

Grow Oxide

Soft Bake

Pattern Probe
Backend

Equipment

Recipe: BEDEP99
Parameter Table: BORON99
5 hours, 1175 °C
Recipe: N2ANL/OX
Parameter Table: N2ANL/OX
5 hours, 1175 °C

Wet Bench

HF:H2O (1:1), 7 minutes

Wet Bench

RCA Clean

D4 tube

Recipe: HTO
3 hours 17 minutes, 450 °C

Wet Bench

HF:H2O (1:7), 5 seconds

Diamond
Nucleation

0.5% Diamond
powder loaded
water

Spinner

500 RPM for 5 seconds,
2000 RPM for 30 seconds

Diamond
Growth

Diamond
thickness ≈3 µm

MPCVD

H2:CH4 (100:1.5 sccm),
2.3 kW, 20 hours, 750 °C

PECVD Oxide

Oxide thickness
≈1 µm

PECVD

Annealing

RTP

10 minutes, 650 °C

Annealing

RTP

10 minutes, 750 °C

Annealing

RTP

3 minutes, 970 °C

41

Table 3-1 (cont’d)
Process Step

Detail

Refresh Oxide
Surface
Doped
Diamond
Growth

Wet Bench

HF:H2O (1:7), 5 seconds

0.5% Diamond
powder loaded
water

Spinner

500 RPM for 5 seconds,
2000 RPM for 30 seconds

Diamond
Growth

Diamond
thickness ≈1 µm

MPCVD

H2:CH4:TMB (100:1:2),
2.3 kW,13 hours, 750 °C

Spin 1827

Photoresist
thickness ≈3 µm

Spinner

3000 RPM, 30 seconds

Hotplate

1 minute, 95 °C

MA6

17 seconds

Wet Bench

MIF 318, 1 minute

Enerjet
Evaporator

Al 10000 Å

March Asher

O2, 100 W, 1 minute

Lift-off

Wet Bench

Acetone, 2 hours

RIE

PlasmaTherm
790

O2 20 sccm, 300 W, 50 mTorr,
2 hours

Wet Bench

HF:H2O (1:7), 10 seconds

Wet Bench

Al Etchant Type A

Spinner

4000 RPM, 30 seconds

Oxide Etch

Wet Bench

BHF, 15 minutes

Remove
Photoresist

Wet Bench

Acetone

Expose

Mask 2: BDP

Develop
Photoresist
Masking
Layer
Deposition
Ashing

Al thickness ≈1
µm

Remove
Diamond
Whiskers
Remove Al
Mask
Spin 9260
Etch Backside
Oxide

Parameters

Diamond
Nucleation

Soft Bake

Pattern Doped
Diamond

Equipment

Photoresist
thickness ≈6 µm

42

Table 3-1 (cont’d)
Process Step

Detail

Equipment

Parameters

Masking
Layer
Deposition

Al thickness ≈1
µm

Enerjet
Evaporator

Al 10000 Å

Spin 5214

Photoresist
thickness ≈1.6
µm

Spinner

3000 RPM, 30 seconds

Hotplate

1 minute, 95 °C

MA6

4 seconds

Hard Bake

Hotplate

1.5 minutes, 115 °C

Flood Expose

MA6

1 minute

Develop
Photoresist

Wet Bench

AZ300, 1 minute

Pattern Mask

Wet Bench

Al Etchant Type A

Remove
Photoresist

Wet Bench

Acetone

RIE

PlasmaTherm
790

O2:CF4 (20:5 sccm), 300 W,
50 mTorr, 3 hours

Remove Al
Mask

Wet Bench

Al Etchant Type A

Soft Bake
Expose
Pattern
Undoped
Diamond

Mask 3: UDP

PECVD Oxide

Oxide thickness
≈1 µm

PECVD

Spin 9260

Photoresist
thickness ≈6 µm

Spinner

4000 RPM, 30 seconds

Hotplate

2.5 minutes, 95 °C

MA6

35 seconds

Develop
Photoresist

Wet Bench

AZ400:H2O (4:1), 2 minutes

RIE

PlasmaTherm
790

Recipe: n_SiO2, 50 minutes

Remove
Photoresist

Wet Bench

Acetone

March Asher

O2, 100W, 100 seconds

Spinner

3000 RPM, 30 seconds

Hotplate

1 minute, 95 °C

MA6

4 seconds

Hard Bake

Hotplate

1.5 minutes, 115 °C

Flood Expose

MA6

1 minute

Soft Bake
Expose

Dielectric
Deposition
and Patterning

Mask 2: BDP

Ashing
Spin 5214

Photoresist
thickness ≈1.6
µm

Soft Bake
Expose

Mask 3: UDP

43

Table 3-1 (cont’d)
Process Step

Detail

Equipment

Parameters

Develop
Photoresist

Wet Bench

AZ300, 1 minute

RIE

PlasmaTherm
790

Recipe: n_SiO2, 25 minutes

Remove
Photoresist

Wet Bench

Acetone

March Asher

O2, 100W, 100 seconds

Spinner

3000 RPM, 30 seconds

Hotplate

1 minute, 95 °C

MA6

4 seconds

Hard Bake

Hotplate

1.5 minutes, 115 °C

Flood Expose

MA6

1 minute

Develop
Photoresist

Wet Bench

AZ300, 1 minute

RIE

PlasmaTherm
790

Recipe: n_SiO2, 25 minutes

Remove
Photoresist

Wet Bench

Acetone

Ashing

March Asher

O2, 100W, 100 seconds

Spinner

4000 RPM, 30 seconds

Hotplate

2.5 minutes, 95 °C

MA6

35 seconds

Develop
Photoresist

Wet Bench

AZ400:H2O (4:1), 2 minutes

Ashing
Metal
Deposition
Lift-off

March Asher
SJ20
Evaporator
Wet Bench

O2, 100 W, 1 minute

Spinner

4000 RPM, 30 seconds

Hotplate

2.5 minutes, 95 °C

MA6

35 seconds

Develop
Photoresist

Wet Bench

AZ400:H2O (4:1), 2 minutes

Ashing
Metal
Deposition
Lift-off

March Asher
Enerjet
Sputterer
Wet Bench

O2, 100 W, 1 minute

Ashing
Spin 5214
Dielectric
Deposition and
Patterning

Photoresist
thickness ≈1.6
µm

Soft Bake
Expose

Spin 9260

Mask 3: UDP

Photoresist
thickness ≈6 µm

Soft Bake
Pattern Counter
Electrode
(Optional
Electrochemical
Probe Step)

Expose

Spin 9260
Pattern
Reference
Electrode
(Optional
Electrochemical
Probe Step)

Mask 4: CEP

Photoresist
thickness ≈6 µm

Soft Bake
Expose

Mask 5: REP

44

Ti/Au 500/4000 Å
Acetone, 2 hours

Ti/Au/Ag 100/80/4000 Å
Acetone, 2 hours

Table 3-1 (cont’d)
Process Step

Detail

Equipment

Parameters

Photoresist
thickness ≈6 µm

Spinner

4000 RPM, 30 seconds

Hotplate

2.5 minutes, 95 °C

MA6

35 seconds

Develop
Photoresist

Wet Bench

AZ400:H2O (4:1), 2 minutes

Ashing
Metal
Deposition
Lift-off

March Asher
SJ20
Evaporator
Wet Bench

O2, 100 W, 1 minute

Spin 9260
Soft Bake
Expose
Pattern Metal
Interconnects

Mask 6: IP

Probe Release

Acetone, 2 hours

PECVD Oxide

Oxide thickness
≈2 µm

PECVD

Spin 9260

Photoresist
thickness ≈6 µm

Spinner

4000 RPM, 30 seconds

Hotplate

2.5 minutes, 95 °C

MA6

35 seconds

Develop
Photoresist

Wet Bench

AZ400:H2O (4:1), 2 minutes

RIE

PlasmaTherm
790

Recipe: n_SiO2, 90 minutes

Remove
Photoresist

Wet Bench

Acetone

March Asher

O2, 100W, 100 seconds

Wet Bench

HF-Nitric, 30 minutes

Wet Bench

EDP, 4 hours, 115 °C

Soft Bake
Top Oxide
Insulator

Ti/Au 500/4000 Å

Expose

Mask 7: TIP

Ashing
Wafer
Thinning
Probe
Release

45

3.3

Single-material Diamond Neural Probe Fabrication
SMM probe fabrication is accomplished is 7-8 major steps, depending on the

application. A schematic of these steps can be seen in Fig. 3-4. The substrate is a Si
wafer coated in 1 µm of SiO2 (Fig. 3-4a). The first step is growing undoped and then
doped diamond (Fig. 3-4b). Next the doped diamond layer is patterned to form the
electrodes, interconnects, and pads (Fig. 3-4c). Then an additional layer of undoped
diamond is grown over the probes (Fig. 3-4d). Next the probe shapes are defined by
patterning the undoped diamond all the way down to the substrate (Fig. 3-4e). Then the
electrodes and pads are exposed by etching the top layer of undoped diamond (Fig. 34f). Next, the metal pads are deposited on top of the exposed doped diamond pads (Fig.
3-4g). Optionally, the electrodes can be selectively modified for other applications (Fig. 34h). Finally, the probes are released from the substrate by chemically etching the SiO 2
from underneath the probes (fig. 3-4i).

46

Figure 3-4: SMM probe fabrication process. (a) Si/SiO2 Substrate. (b) Undoped and doped diamond
growth. (c) Doped diamond patterning. (d) Undoped diamond growth. (e) Probe patterning. (f) Insulation
patterning. (g) Metal pad deposition. (h) Optional electrode modification. (i) Probe release.

Before SMM diamond probes can be fabricated, an appropriate substrate must be
prepared. Almost any material with a melting point above 900 °C could be used. However,
other useful characteristics are a thermal expansion coefficient similar to poly-C,
resistance to out-diffusion of potential diamond dopants, mechanical strength, and it must
be able to be chemically etched. As such, sapphire would appear to be an ideal candidate.
For the purpose of cost reduction, a 750 µm thick, n-type Si wafer with a 1 µm layer of
SiO2, hand-cut to a size of 5 cm by 5 cm, was used.
47

Before diamond growth can be achieved, diamond nanoparticles must be spread
over the substrate, a process known as diamond seeding (or nucleation). Seeding was
performed using a 5% solution of diamond particles in water. As such, before seeding
can effectively be performed on Si or SiO2, the surface needs to be modified to become
hydrophilic. This can be done by briefly etching the Si or SiO 2 surface. Typically diluted
hydrofluoric acid is used for a short time (<30 seconds) to achieve this aim. Alternatively,
O2 plasma has been used for similar results. The diamond solution must also be prepared,
which has been done using an ultrasonic bath to agitate the solution for 30 minutes to
ensure an even distribution of diamond particles. Once the solution and substrate are
prepared, some diamond solution is pipetted onto the substrate until about 70% of the
substrate is covered. At this point, the wafer is spun at 4000 RPM for 45 seconds to evenly
distribute diamond particles over the substrate.
Diamond growth is accomplished using microwave-plasma-assisted chemical vapor
deposition (MPCVD). Before the growth process, the MPCVD chamber is pumped to a
vacuum of less than 20 mTorr absolute pressure. Once the chamber has been evacuated,
gas flow is initiated and the pressure in the chamber is maintained at 40 Torr absolute.
The main constituent of the growth plasma is H2, as such H2 has the highest flow rate of
100 sccm. Once the pressure has stabilized, the MPCVD is started and the output power
is slowly increased to 2.3 kW as the temperature increases to about 700 °C. The source
of carbon for diamond growth is CH4 which is next added at a rate of 1.5 sccm. Undoped
diamond is allowed to grow for 20 hours to form the structural backing of the probes. In
order to improve the probe performance by reducing the conductivity between
interconnects, oxygen is added to the growth environment. The addition of .38 sccm of
48

O2 to the growth environment increases the resistivity from about ≈102-103 Ω-cm to ≈107
ohm-cm at a cost of reduced the growth rate. O2 is added to the growth environment for
4 hours. Once the undoped diamond layer is done growing, doped diamond is grown by
removing O2 from the growth environment and adding trimethylboron [B(CH3)3] (TMB)
(diluted in H2 to .005%) to the growth environment at a rate of 10 sccm. Doped diamond
growth continues for about 4 hours. At the end of diamond growth, the flow of CH 4 and
TMB is ended and the hydrogen plasma removes the imperfect surface layer for 15
minutes. After the diamond growth process there is a layer of undoped diamond that is
approximately 5 µm thick and a layer of doped diamond that is approximately 1 µm thick.
The quality of diamond is experimentally verified using Raman spectroscopy to verify the
main peak of Raman shift is at 1332 cm-1, as expected.
The next step in the fabrication process is to etch the doped diamond layer to form
the electrodes, interconnects, and pads. Diamond is patterned using reactive ion etching
(RIE). The RIE process for etching thicker diamond layers requires a metal masking
material. The metal that is used is Al as it has a lower etch rate in the specific etch recipe
used for etching diamond. Al is patterned using liftoff; PR 9260 is used for the liftoff
process because its thickness allows for thick layers of Al to be patterned. After
developing the photoresist, the photoresist residue is removed using O 2 plasma in a
March Asher with a power of 100 W for 3 minutes. Next the Al is deposited with a
thickness of 1 µm using an evaporator. Liftoff is performed in PRS 2000 which is agitated
by ultrasound. Once the Al masking layer is prepared, the doped diamond is etched in a
Plasmatherm 790 RIE system. The etch process is done at a pressure of 50 mTorr
absolute, with 5 sccm CF4 and 20 sccm Ar. The RF etch power is 300 W. The etching is
49

performed for 75 minutes to ensure that the doped layer is completely etched. The etch
rate varies across the sample but is typically in the range of 0.9-1.5 µm/h. The etching is
verified by measuring the resistance between interconnects. If the doped diamond layer
is completely etched, the resistance between interconnects cannot be measured using a
multimeter. Once the etching is done, the Al mask layer is removed using Al etch A.
Micrographs of a probe after the first etch step can be seen in Fig. 3-5.

Figure 3-5: Optical micrograph of probe after first etch step (top) and scanning electron micrograph of
probe after first etch step (bottom).

The next step in fabrication is growth of undoped diamond to form the top insulating
layer. As there is a preexisting diamond layer, seeding for diamond growth is
unnecessary. The growth conditions are the same as the previous undoped diamond
growth with gas flow rates of 100 sccm, 1.5 sccm, and .38 sccm for H 2, CH4, and O2,
respectively. Undoped diamond is grown for about 4 hours, resulting in a layer that is
approximately 1 µm thick. Micrographs of a probe after the second undoped diamond
growth can be seen in Fig. 3-6.

50

Figure 3-6: Optical micrograph of probe after second growth step (top) and scanning electron micrograph
of probe after second growth step (bottom).

Next, the shape of probes is defined by RIE. An Al mask is made using the same
process as above, except its thickness is 1.5 µm. The RIE conditions are also the same
except the etch time is increased to 4 hours in order to etch through both undoped layers.
The CF4 plasma also aggressively etches the SiO2 layer underneath the diamond.
Knowing this, it can be determined whether etching is complete by measuring the
resistance between two points off of the probes. If the diamond is completely etched, as
well as the underlying SiO2, then there will be a low resistance between any two points
on the surface of the n-Si wafer. Once complete etching has been verified, the Al mask
is removed, again using Al etch A. An optical micrograph of a probe after the probe shape
has been defined can be seen in Fig. 3-7.

51

Figure 3-7: Optical micrograph of probe after second etching step.

After patterning the probe shape, the electrodes and pads are exposed by etching
the undoped diamond layer above them. For this process, a 1 µm thick Al mask is made
using the negative photoresist PR 5214, rather than PR 9260. The RIE recipe remains
unchanged and the etch time is 75 minutes. The etch process can be verified by
measuring the resistance between an electrode and the pad it’s connected to. If the
undoped layer is completely removed, there is a low resistance between the pad and
electrode. The Al mask is removed by Al etch A. An optical micrograph after this step can
be seen in Fig. 3-8.

Figure 3-8: Optical micrograph of a probe after exposing the electrodes and pads.

At this point, the single-material diamond probes are entirely made of diamond. The
next step in the process is to add the metal pads on the backend for use in wire bonding.
The metal pads are made up of two different metal layers. The top metal layer is Au

52

because it is easy to wire bond to and it is resistant to the HF etch that releases the
probes. However, Au does not adhere well to diamond, and if only gold is used, the metal
pads will peel off of the diamond during the bonding process. To alleviate this problem,
and to achieve an ohmic contact between the metal and doped diamond, a layer of Ti or
Cr is put underneath the Au. Ti and Cr form a thin layer of TiC and CrC, respectively,
which give an ohmic contact and good adhesion to diamond. In this process, Cr is used
because it is resistant to etching in HF. The metal layers are patterned using liftoff. The
photoresist used is again 9260. The metal layers are deposited by evaporation with
thicknesses of 50 nm and 500 nm for Cr and Au. Micrographs of a probe after depositing
pads can be seen in Fig. 3-9.

Figure 3-9: Optical micrograph of probe after pad deposition (top). Scanning electron micrograph of probe
after pad deposition (bottom).

At this point in fabrication the electrodes can selectively be modified for various
applications. For the purpose of electrochemical detection it is of interest to modify the
electrodes to be terminated with F, rather than H or O. The reason is because the
performance of the diamond as a working electrode in an electrochemical cell improves
53

dramatically, as seen in Fig. 3-10. In the figure, BDD represents a hydrogen-terminated
boron-doped-diamond electrode. F-BDD represents a fluorine-terminated boron-dopeddiamond electrode. The Pt and Au electrodes are provided as a performance reference.
In the figure, a cyclic voltammogram is presented that shows the potential window of
several different working electrodes. It is apparent from Fig 3-10 that the potential window
of the F-BDD electrode is significantly wider than the BDD electrode (as well as the Au
and PT electrodes). This means that an F-BDD working electrode could be used in
analytical electrochemistry experiments to detect more analyte types than the other
working electrodes. The electrodes on the SMM diamond neural probe can be modified
to be fluorine-terminated by exposing them the CF4 plasma for a short time (< 1 minute).
Other modifications of the surface of diamond are possible for other applications, but have
not been thoroughly explored.
The final step of single-material diamond probe fabrication is releasing the probes by
etching the SiO2 underneath the probes. The etching of SiO2 is accomplished using HF.
After release the probes are thoroughly rinsed in deionized water for 20 minutes and then
dried. An example of a released probe can be seen in Fig. 3-11. The full process flow with
detailed steps can be found in Table 3-2.

54

Figure 3-10: Comparison of CV curves using different working electrodes in 1M HClO4, highlighting the
improvement of fluorine-terminated diamond over hydrogen-terminated diamond. Scan rate is 0.2 V/s.

Figure 3-11: Image of a released SMM diamond probe on a fingertip.

55

Table 3-2: Fabrication process for single-material diamond neural probes.
Process Step

Undoped
and Doped
Diamond
Growth

Detail

Equipment

Parameters

Refresh Oxide
Surface

n-Si wafer with 10
µm of SiO2

March Asher

O2, 100 W, 1 minute

Diamond
Nucleation

5% Diamond
powder loaded
water

Spinner

4000 RPM for 45 seconds

MPCVD

H2:CH4 (100:1.5 sccm), 2.3
kW, 20 hours, 750 °C

Undoped
Diamond
Growth
High-resistivity
Undoped
Diamond
Growth
Boron-doped
Diamond
Growth

Undoped
Diamond
thickness ≈5 µm

MPCVD

H2:CH4:O2 (100:1.5:0.38
sccm), 2.3 kW, 4 hours, 750 °C

Doped Diamond
thickness ≈1 µm

MPCVD

H2:CH4:TMB (100:1.5:10
sccm), 2.3 kW, 4 hours, 750 °C

Spin 9260

Photoresist
thickness ≈6 µm

Spinner

4000 RPM, 30 seconds

Hotplate

2.5 minutes, 95 °C

MA6

35 seconds

Develop
Photoresist

Wet Bench

AZ400:H2O (4:1), 2 minutes

Ashing

March Asher

O2, 100 W, 1 minute

Enerjet
Evaporator

Al 10000 Å

Lift-off

Wet Bench

PRS 2000, 1 hour with
ultrasonifcation

RIE

PlasmaTherm
790

Ar:CF4 (20:5 sccm), 300 W, 50
mTorr, 75 minutes

Wet Bench

Al Etchant Type A

MPCVD

H2:CH4:O2 (100:1.5:0.38
sccm), 2.3 kW, 4 hours, 750 °C

Soft Bake
Expose

Pattern
Doped
Diamond

Masking Layer
Deposition

Undoped
Diamond
Growth

Remove Al
Mask
High-resistivity
Undoped
Diamond
Growth

Mask 1: Borondoped Diamond
Electrodes and
Interconnects

Al thickness ≈1
µm

Undoped
Diamond
thickness ≈1 µm

56

Table 3-2 (cont’d)
Process Step

Detail

Equipment

Parameters

Photoresist
thickness ≈6 µm

Spinner

4000 RPM, 30 seconds

Hotplate

2.5 minutes, 95 °C

MA6

35 seconds

Develop
Photoresist

Wet Bench

AZ400:H2O (4:1), 2 minutes

Ashing

March Asher

O2, 100 W, 1 minute

Enerjet
Evaporator

Al 15000 Å

Lift-off

Wet Bench

PRS 2000, 1 hour with
ultrasonifcation

RIE

PlasmaTherm
790

Ar:CF4 (20:5 sccm), 300 W, 50
mTorr, 4 hours

Remove Al
Mask

Wet Bench

Al Etchant Type A

Spinner

3000 RPM, 30 seconds

Hotplate

1 minute, 95 °C

MA6

4 seconds

Hard Bake

Hotplate

1.5 minutes, 115 °C

Flood Expose

MA6

1 minute

Develop
Photoresist

Wet Bench

AZ300, 1 minute

Ashing

March Asher

O2, 100 W, 1 minute

Enerjet
Evaporator

Al 10000 Å

Lift-off

Wet Bench

PRS 2000, 1 hour with
ultrasonifcation

RIE

PlasmaTherm
790

Ar:CF4 (20:5 sccm), 300 W, 50
mTorr, 75 minutes

Remove Al
Mask

Wet Bench

Al Etchant Type A

Spin 9260
Soft Bake
Expose

Pattern
Undoped
Diamond
Probe Shape

Masking Layer
Deposition

Spin 5214

Mask 2: Undoped
Diamond Probe
Sheath

Al thickness ≈1
µm

Photoresist
thickness ≈1.6
µm

Soft Bake
Expose

Pattern
Undoped
Diamond
Probe
Electrodes
and Pads

Masking Layer
Deposition

Mask 3: Undoped
Diamond
Electrodes and
Pads

Al thickness ≈1
µm

57

Table 3-2 (cont’d)

Process Step

Detail

Equipment

Parameters

Photoresist
thickness ≈6 µm

Spinner

4000 RPM, 30 seconds

Hotplate

2.5 minutes, 95 °C

MA6

35 seconds

Develop
Photoresist

Wet Bench

AZ400:H2O (4:1), 2 minutes

Ashing

March Asher

O2, 100 W, 1 minute

Metal
Deposition

Enerjet
Evaporator

Cr/Au 500/5000 Å

Lift-off

Wet Bench

PRS 2000, 1 hour with
ultrasonifcation

RIE

PlasmaTherm
790

CF4 20 sccm, 300 W, 50
mTorr, 30 seconds

Wet Bench

HF, 30 minutes

Wet Bench

DI Water rinse, 20 minutes

Spin 9260
Soft Bake
Expose

Pad Deposition

Electrode
Modification
(Optional)
Probe Release

Mask 4: Metal
Pads

Probe
Release
Probe
Cleaning

In order to effectively use the probes, they must be bonded to a PCB after release.
Al wire was used for bonding. After bonding, the wires and backend of the probe are
covered in nonconductive epoxy so that the probe can be immersed in liquid during testing
or use. This fabrication represents the first SMM neural array ever fabricated.
Furthermore it is the first SMM bioMEMS device ever fabricated, and one of the first
functional SMM devices ever produced.

58

3.4

Conclusion
Both diamond neural probes and single-material diamond neural probes have been

successfully fabricated. Detailed fabrication processes were discussed for each probe
type. Micrographs of the single-material diamond neural probe fabrication process were
also shown. Some of the advantages of SMM fabrication are demonstrated as the
complexity of probe fabrication is significantly reduced for the SMM diamond neural probe
fabrication. While the diamond-based probe fabrication has a 5-page table showing the
fabrication detail and requires a minimum of 6 masks, the SMM diamond probe fabrication
has a 2.5 page table showing the fabrication details and requires a minimum of 4 masks.
This work represents a milestone in SMM development as it was the first SMM
microelectrode array, the first SMM bioMEMS device, and one of the first functional SMM
devices ever fabricated.

59

REFERENCES

60

REFERENCES

[1] H.-Y. Chan, M. Varney, D.M. Aslam K.D. and Wise, “Fabrication and characterization
of all-diamond microprobes for electrochemical analysis,” Proceedings of IEEE Int.
Conf. on Nano/Micro Engineered and Molecular Syst., pp. 532-535, 2008.
[2] H.-Y. Chan, D.M. Aslam, S.H. Wang, G.M. Swain and K.D. Wise, “Fabrication and
testing of a novel all-diamond neural probe for chemical detection and electrical
sensing applications,” Proceedings of IEEE Int. Conf. on Micro Electro Mechanical
Systems, pp. 244-247, 2008.
[3] H.-Y. Chan, D. M. Aslam, J. A. Wiler, and B. & Casey, “A novel diamond microprobe
for neuro-chemical and-electrical recording in neural prosthesis,”
Microelectromechanical Systems, Journal of, vol. 18(3), pp. 511-521, 2009.
[4] H.-Y. Chan, “Polycrystalline CVD Diamond Probes for use in In Vivo and In Vitro
Neural Studies,” Ph.D. dissertation, Dept. Elect. Eng., Mich. State Univ., East
Lansing, MI, 2008

61

CHAPTER 4:
Electrochemical Detection Experiments

4.1

Introduction
One property of diamond that is of particular interest for electrochemical applications

is its large electrochemical potential window (∼4 V). Diamond has been used in many
electrochemical detection applications, as can be seen in Table 4-1 [1]. For biomedical
applications, diamond has been used to detect neurotransmitters, such as: dopamine [24], norepinephrine [2, 5], epinephrine [2], serotonin [6-8], histamine [8], nicotinamide
adenine dinucleotide (NADH) [9] and adenosine [10].
Table 4-1. Substances electrochemically detected using diamond electrodes.

Organic Substances
adenosine, ascorbic acid, caffeine, carbamate
pesticides, catecholamines, cephalexin,
chlorophenols, chlorpromazine, p-cresol, cysteine,
dopamine, formaldehyde, flavonoids, glucose,
glutathione, guanosine, histamine, indoles, NADH,
nitrophenol, nucleic acids, oxalic acid,
penicillamine, phenol, polyamines, purine,
pyrimidine, serotonin, sulfa drugs, tetracycline
antibiotics, theobromine, theophylline, tiopronin
and xanthine

Inorganic Substances
azide anion, hydrazine, hydrogen
peroxide, iodide, nitrate, nitrite,
dissolved oxygen, dissolved ozone,
peroxodisulfate, sulfate, sulfide,
Ag+, As(III), Cd2+, Cu2+, Hg+, Mn2+,
Ni2+, Pb2+, Sn4+ and Zn2+

Other studies on capillary electrophoresis to detect neurotransmitters using borondoped-diamond have revealed a detection limit down to 10 nM [2, 3]. Sarada et al. [8]
explored the electrochemistry of histamine and serotonin with boron-doped diamond
using cyclic voltammetry, hydrodynamic voltammetry and low inject analysis. The borondoped poly-C electrode in histamine had a signal-to-noise ratio (SNR) an order-of62

magnitude greater than that of the glassy carbon electrode, a linear dynamic range of 3–
4 orders of magnitude and a detection limit of about 1 µM. Comparatively, they found that
the detection limit of 5-hydroxytryptamine was about 10 nM.
Rao et al. [9] studied nicotinamide adenine dinucleotide (NADH) using boron-doped
poly-C. They found a lower detection limit of 10 nM, while maintaining a SNR of at least
seven. Furthermore, they were able to distinguish between NADH and ascorbic acid when
the concentrations were comparable. Herlambang et al. [6] detected neurotransmitters
using an overoxidized poly pyrrole-modified, boron-doped-diamond microfiber electrode.
They found detection limits of 500 fM for dopamine and 600 fM for serotonin.
Boron-doped diamond is an excellent candidate as a working electrode in an
electrochemical cell for chemical detection, as well as having several other notable
applications in electrochemistry. As such, electrochemical applications for diamond and
single-material microprobes are very promising. Diamond electrodes offer a benefit over
other electrode types typically used in neural probe applications. Other probes, such as
the silicon-based probe, typically use Au or Pt as a working electrode. Iridium oxide
electrodes have also been used as quasi-reference electrodes. As such, the probes with
boron-doped diamond electrodes can measure more types of analytes at a higher
sensitivity than silicon-based probes. The scope of this work is limited to detection of
neurotransmitters. However, diamond microprobes could be useful for many
electrochemical applications where electrode placement is critical or electrode spacing
resolution is critical.

63

This work is unique because it is the only electrochemical detection of
neurotransmitters that has been performed using a SMM diamond probe with borondoped-diamond electrodes as a working electrode.

64

4.2

Results
In order to detect a neurotransmitter using analytical electrochemistry, an

electrochemical cell must be established and a test must be performed which can be
compared against known standards for different neurotransmitters. The single-material
neural probe, or MEA, was used with a commercial Ag/AgCl in 1M KCl reference
electrode and a commercial Pt counter electrode. The working electrode was borondoped diamond. A counter electrode could be added to the single-material probe design,
however, an independent reference electrode is required. Diamond (even with different
terminations) does not meet the prerequisites for a reference electrode, as it does not
have a well-known and stable equilibrium electrode potential independent of electrolytes
present in the electrochemical cell. Cyclic voltammetry measurements were performed
on the electrochemical cell using a CH Instruments model 750C bipotentiostat.
In an effort to optimize the diamond electrodes for electrochemical detection, some
were modified to have a fluorine-terminated surface, rather than the typical oxygenterminated surface, by being exposed to CF4 plasma. Cyclic voltammetry was performed
on a KCl solution to compare the potential window of the oxygen- and fluorine-terminated
diamond working electrodes. It has been found that the potential window of the BDD
working-electrode can be altered significantly by modifying the surface of the diamond.
Fluorine-terminated diamond has a much greater potential window than oxygenterminated diamond. As seen in Figure 4-1, the potential window of a fluorine-terminated
electrode extends much further on the reduction side of the voltammogram. This is likely
due to the extreme hydrophobicity of the fluorine-terminated BDD [11] preventing

65

hydrogen evolution on the surface of the electrode. This means that more analytes could
be detected by means of reduction using this electrode.

Figure 4-1. Cyclic voltammogram comparing the potential windows of oxygen-terminated diamond
and fluorine-terminated diamond in 1 M KCl.

To test repeatability of the electrochemical properties of fabricated probes, several
probes were used to perform cyclic voltammetry of 1M KCl solution to determine the

66

potential window of each electrode on each probe. Several electrodes were omitted (6
out of 36 electrodes) from two of the probes because there was no electrical connection
to the electrodes (presumably from complications in fabrication or packaging). The results
of the other 30 electrodes can be seen in the voltammograms in Fig. 4-2. It can be seen
that the potential window of different electrodes and different probes is very repeatable.
All of the electrodes have a potential window of ~4.4 V in 1M KCl.
In addition to testing the potential window, the impedance of each electrode was also
measured. The impedance spectroscopy was performed using the same bipotentiostat
system with the probes immersed in a 1M KCl solution. The measured impedance of each
electrode on a single probe can be seen in Fig. 4-3. Measurements from additional probes
are very comparable to the probe shown. The impedances are comparable between
different electrodes, with the average impedance at 1 kHz being 31.6 kΩ with a standard
deviation of 2.8 kΩ.

67

Figure 4-2: Cyclic voltammograms of 1M KCl performed by 30 electrodes on 6 different probes.

68

Figure 4-3: Impedance spectroscopy of 6 electrodes on a single single-material probe. Spectroscopy was
performed in vitro, using a 1M KCl solution. The image on the right acts as a legend, showing which
colors correspond to which electrodes on the fabricated probe.

69

In order to test the detection of neurotransmitters, an experiment was designed to
add small amounts of norepinephrine to Krebs solution and then perform cyclic
voltammetry. The experiment was performed using electrodes from a diamond probe as
the working electrode, a commercially available Pt counter electrode, and a commercially
available Ag/AgCl reference electrode. The measured current in the experiment was then
compared to the current from a cyclic voltammogram of pure Krebs solution. In the
voltammograms in Fig. 4-4 the change in current from adding norepinephrine to a pure
Krebs solution can be seen. The inset shows the cyclic voltammograms before
subtracting the background current. The figure shows that there is a clear difference
between 1 nM, 5 nM and 15 nM concentrations of norepinephrine, indicating a
quantification limit in the 5 nM range. This detection limit is consistent with other results
using diamond as a working electrode [2, 3, 8, 9]. This detection limit should be sufficient
for in vivo biomedical applications as typical neurotransmitter levels are several orders of
magnitude larger in the mM range [12]. The figure also demonstrates an extremely low
background current, a major advantage of the diamond electrode. Performing this cyclic
voltammetry with other fabricated diamond probes has comparable results. The cyclic
voltammetry method was chosen to test the electrochemical probe because it is easy to
interpret and is applicable to testing for small concentrations of analytes. This method
could potentially be used to detect other analytes with the same probe. Peaks will be
detected at different voltages on the CV curve which can be used to identify several
analytes and their concentrations so long as the redox voltages are distinct. It is likely that
some a priori knowledge of the biological system would be required to isolate several
neurotransmitter concentrations. Similar diamond probes have been used to detect many

70

types of analytes including several neurotransmitters, as described in Section 4.1 of this
dissertation.

Figure 4-4: Family of background-subtracted cyclic voltammograms of krebs solution with varying
amounts of norepinephrine (NE), demonstrating a lower detection limit of 5 nM or less. The inset shows
the cyclic voltammograms before subtracting the krebs solution background.

71

4.3

Conclusion
It has been demonstrated that diamond probes can be used for some electrochemical

applications. The potential window of the fluorine-terminated boron-doped-diamond
electrodes is ~4.4 V in 1M KCl and is very repeatable between different electrodes and
probes. The impedance of the electrodes is also repeatable and the average impedance
at 1 KHz is ~30 KΩ. The diamond working electrodes have been used, in vitro, to quantify
the detection limit of the neurotransmitter norepinephrine in Krebs solution. It appears
that diamond microprobes (or similar devices) may be of use in electrochemical detection
applications where controlled electrode size and spacing are critical.

72

REFERENCES

73

REFERENCES

[1] A. Kraft, “Doped diamond: A compact review on a new, versatile electrode material,”
Int. J. Electrochem. Sci., 2, pp. 355-385, 2007.
[2] D. Shin, B.V. Sarada, D.A. Tryk, and A. Fujishima, “Application of diamond
microelectrodes for end-column electrochemical detection in capillary,” Anal. Chem.,
vol. 75, pp. 530-534, 2003.
[3] J. Cvacka, V. Quaiserova, J. Park, Y. Show, A. Muck, and G.M. Swain, “Boron-doped
diamond microelectrodes for use in capillary electrophoresis with electrochemical
detection,” Anal. Chem., vol. 75, pp. 2678-2687, 2003.
[4] P.S. Siew, K.P. Loh, W.C. Poh, and H. Zhang, “Biosensing properties of
nanocrystalline diamond film grown on polycrystalline diamond electrodes,” Diamond
Relat. Mater., vol. 14, pp. 426-431, 2004.
[5] J. Park, V.Q. Mocko, K. Peckova, J.J. Galligan, G.D. Fink, and G.M. Swain,
“Fabrication, characterization, and application of a diamond microelectrode for
electrochemical measurement of nore-pinephrine release from the sympathetic
nervous system,” Diamond Relat. Mater., vol. 15, pp. 761-772, 2006.
[6] O. Herlambang, B.V. Sarada, and T.N. Rao, “Miniturization of diamond microsensor
system for in vivo detection,” Chem. Sens. B, vol. 18, pp. 121-123, 2002.
[7] J.M. Halpern, S. Xie, G.P. Sutton, B.T. Higashikubo, C.A. Chetek, H. Lu, H.J. Chiel,
and H.B. Martin, “Diamond electrodes for neurodynamic studies in Aplysia
californica,” Diamond Relat. Mater., vol. 15, pp. 183-187, 2006.
[8] B.V. Sarada, T.N. Rao, D.A. Tryk, A. Fujishima, “Electrochemical oxidation of
histamine and serotonin at highly boron-doped diamond electrodes,” Anal. Chem.,
vol. 72, pp. 1632-1638, 2002.
[9] T.N. Rao, I. Yagi, T. Miwa, D.A. Tryk, A. Fujishima, “Electrochemical oxidation of
NADH at highly boron-doped diamond electrode,” Anal. Chem., vol. 71, pp. 25062511, 1999.
[10] S. Xie, G. Shafer, C.G. Wilson, and H.B. Martin, “In vitro adenosine detection with a
diamond-based sensor,” Diamond Relat. Mater., vol. 15, pp. 225-228, 2006.
[11] G. Sine, L. Ouattara, M. Panizza, and Ch Comninellis, "Electrochemical behavior of
fluorinated boron-doped diamond," Electrochemical and solid-state letters 6, no. 9,
D9-D11, 2003.
[12] A. Scimemi and M. Beato, “Determining the neurotransmitter concentration profile at
active synapses,” Molecular neurobiology, vol. 40, no. 3, pp. 289-306, 2009.
74

CHAPTER 5:
Electrical Neural Recording Experiments

5.1

Introduction
Single-material neural probes were designed for in-vivo electrical recording of neural

signals. As such, the ultimate test of their efficacy is being used to record neural activity.
In order to validate the recordings, the results must match the expected results of an
experiment. Two neural recording experiments were performed on a pigmented guinea
pig (Cavia porcellus) at the Kresge Hearing Research Institute at the University of
Michigan. The animal use procedures were approved by the University of Michigan. For
one experiment, data was recorded from silicon probes, diamond probes, and singlematerial diamond probes, and each can be compared. For the other recording
experiment, only single-material diamond probes were used. In this analysis, only a basic
threshold method will be used to detect action potentials. The detection was performed
by the recording equipment, a Plexon Neural Data Acquisition System, while the
experiment was performed. These analyses represent the capability of single-material
diamond probes using traditional recording and detection methods. Neural recordings
have been previously performed using diamond probes, but this work is the first time that
single-material diamond probes have been used.

75

5.2

In-Vivo Neural Recording
In order to test the SMM neural probe, the probe was surgically implanted into a live

guinea

pig’s

auditory

cortex.

This

marked

the

first

implantation

of

a

SMM device into a living animal. Before the experiment, the guinea pig was anesthetized
and a bolt was affixed to the skull of the animal to hold the skull (and brain) in place (Fig.
5-1). Once the animal was restrained, a partial craniotomy was performed above the right
auditory cortex. A ground wire was attached to the pericranium far from the implantation
site. A speaker was placed in the animal’s left ear (corresponding to the right auditory
cortex). Finally the packaged neural probe was inserted through the dura matter into the
auditory cortex (Fig. 5-2).

Figure 5-1: Anesthetized guinea pig with bolt affixed to skull.

76

Figure 5-2: Anesthetized guinea pig with neural probe inserted in right audio cortex and speaker placed at
left ear.

In the experiment, the anesthetized guinea pig was subjected to 200 ms, widefrequency audio-pulses at a frequency of 2 Hz. During the experiment, the electrical
activity at all electrodes was recorded. In addition, the recording equipment also detected
action potentials using a threshold technique. Fig. 5-3 shows a close-up example of an
electrical signal recorded from a single-material diamond probe. For the purpose of
comparison, Fig. 5-4 shows a similar recording from a diamond neural probe and Fig. 55 shows a recording from a silicon-based neural probe. Recordings from all electrodes
on each probe type can be seen in Fig. 5-6, Fig. 5-7, and Fig. 5-8.
77

All-Diamond SMM Probe
600

400

Amplitude [ V]

200

0

-200

-400

-600

0.5

1

1.5

2

2.5

Time [s]

3

3.5

4

4.5

5

Figure 5-3: Neural recording from the audio cortex of a guinea pig taken with a single-material diamond
neural probe with boron-doped-diamond Electrode.

Diamond Probe
600

Amplitude [ V]

400

200

0

-200

-400

-600

0.5

1

1.5

2

2.5

Time [s]

3

3.5

4

4.5

5

Figure 5-4: Neural recording from the audio cortex of a guinea pig taken with a diamond-based neural
probe with a boron-doped-diamond electrode.

78

Silicon Probe
1000

Amplitude [ V]

500

0

-500

-1000

0.5

1

1.5

2

2.5

Time [s]

3

3.5

4

4.5

5

Figure 5-5: Neural recording from the audio cortex of a guinea pig taken with a silicon-based neural
probe with an iridium oxide electrode.

79

Site 1

[V]

500
0

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

500

Site 2

[V]

-500

0
-500

Site 3

[V]

500
0
-500

Site 4

[V]

500
0
-500

Site 5

[V]

500
0
-500

Site 6

[V]

500
0
-500

Time [s]

Figure 5-6: Electrical neural recordings from the audio cortex of a guinea pig taken with a single-material
diamond neural probe with boron-doped-diamond electrodes.

80

Site 1

[V]

500
0
-500
1

2

3

4

5

6

7

8

1

2

3

4

5

6

7

8

1

2

3

4

5

6

7

8

1

2

3

4

5

6

7

8

1

2

3

4

5

6

7

8

1

2

3

4

5

6

7

8

1

2

3

4

5

6

7

8

1

2

3

4

5

6

7

8

Site 2

[V]

500
0

400
200
0
-200
-400

Site 3

[V]

-500

Site 4

[V]

500
0
-500

Site 5

[V]

500
0
-500

Site 6

[V]

500
0

500

Site 7

[V]

-500

0

500
0
-500

Site 8

[V]

-500

Time [s]

Figure 5-7: Electrical neural recordings from the audio cortex of a guinea pig taken with a diamond-based
neural probe with boron-doped-diamond electrodes.

81

Site 1

[V]

1000
0
-1000
-2000
1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

Site 2

[V]

2000
0
-2000

Site 3

[V]

1000
0
-1000

Site 4

[V]

2000
0

2000
1000
0
-1000

Site 5

[V]

-2000

Site 6

[V]

1000
0
-1000

Site 7

[V]

2000
0
-2000

Site 8

[V]

2000
0
-2000

Time [s]

Figure 5-8: Electrical neural recordings taken from the audio cortex of a guinea pig with a silicon-based
neural probe with iridium oxide electrodes. Electrodes 1 – 8.

82

Site 9

[V]

2000
0
-2000
1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

Site 10

[V]

2000
0

500

Site 11

[V]

-2000

0
-500

Site 12

[V]

2000
0
-2000

Site 13

[V]

2000
0
-2000

Site 14

[V]

1000
0

Site 15

1000
0
-1000
-2000

Site 16

[V]

[V]

-1000

1000
0
-1000
-2000
-3000

Time [s]

Figure 5-9: Electrical neural recordings taken from the audio cortex of a guinea pig with a silicon-based
neural probe with iridium oxide electrodes. Electrodes 9 – 16.

83

The recordings show peak values every 500 ms, corresponding to the frequency of
the applied stimulus. However, the recorded signals from the diamond probes and the
single-material diamond probes appear to have a lower SNR than what is often presented
for silicon-based neural probes with other electrode materials. Low SNR could be due to
a myriad of reasons, such as: a large distance between the neurons and electrodes (probe
placement), inappropriately sized electrodes (diameter was 30 µm), high impedance per
unit area of the electrode (electrode impedance of ~30 kΩ at 1 kHz), or poor insulation
between recording sites (>1 MΩ). The SNR issue can be alleviated through signal
processing. The SMM neural probes have an SNR which is comparable to other diamond
electrodes (e.g., Figs. 5-4 and 5-7), but slightly worse. The slightly poorer SNR can be
accounted for by the lower conductivity of doped diamond, as compared to metals
typically used for probe interconnects. Further work has been done in signal processing
to detect action potentials in these relatively noisy recordings.
A second recording experiment was performed where the stimulus was a range of
different single-frequency tones. The neuron firing rate at each electrode site was
recorded for each tone. Tones ranged from 1,000 Hz to 30,000 Hz and were played in a
random order. The amplitude of the audio stimulus was also randomized between 37.7
and 87.7 dB SPL. The experiment was repeated 30 times. The results of the experiment
can be seen in Fig. 5-10. Fig. 5-10 shows the sum of detected action potentials versus
audio frequency for all 30 iterations. The expected result of this experiment is that certain
electrode sites should be sensitive to certain frequencies because the neurons being
recorded have higher firing rates with those audio frequencies [1-3]. However, the action
potential detection algorithm was a simple thresholding technique (window discriminator)

84

that was inadequate to accurately detect action potentials due to the poor SNR of the
recordings. The result is that there is little correlation between detected firing rate and
applied audio stimulus. The activity was higher for lower frequency stimuli for all recording
sites. There is no discernable difference between recording sites.

220

210
200
190
180
170
160
150

210

Action Potential Sum

Action Potential Sum

220

Action Potential Sum

Probe 3

Probe 2

Probe 1

200
190
180
170
160

130
120
110
100

150

10

10

10

Audio Frequency [Hz]

80
70
4

3

4

3

4

3

90

10

10

Audio Frequency [Hz]

10

Audio Frequency [Hz]

Probe 6

Probe 5

Probe 4
50
45
40
35
30
25

60

Action Potential Sum

Action Potential Sum

Action Potential Sum

55

50
40
30

60
50
40
30

20
3

10

4

10

Audio Frequency [Hz]

4

3

10

10

Audio Frequency [Hz]

4

3

10

Figure 5-10: Detected action potentials versus frequency of applied audio stimulus.

85

10

Audio Frequency [Hz]

5.3 Analysis of Neural Recordings using Threshold Detection
Method
In the initial recording experiment, a six-electrode SMM probe was implanted into the
auditory cortex of a guinea pig. During the experiment, a wide-band audio stimulus was
played through a speaker placed at the ear of the guinea pig. The stimulus lasted 200 ms
and was applied every 500 ms. During the experiment, the electrical activity at all 6
electrodes was recorded. In addition, the recording equipment also detected action
potentials using a threshold technique. Action potentials were detected if the recorded
signal went above or below user-set thresholds.
The expected result from this experiment is to see a correlation between the
application of stimuli and the firing rate of neurons. Specifically, as stimulus is applied,
the firing rate is expected to increase, especially at the onset of stimulus. It is expected
to see a large number of action potentials within the first 50 ms of stimulus, and few during
the period where no stimulus is applied. It is known that specific neurons are often tuned
to certain frequencies, and entire regions of the auditory cortex can be mapped to tuned
frequencies, a technique known as tonotopy [1-3]. However, the stimuli in this experiment
are wide band stimuli, so all neurons should be sensitive to the stimuli. In order to
determine how many action potentials were detected during the stimuli and at the
beginning of stimuli, the action potential detection times were compared with the times
that stimuli started. For every action potential detected, the time after the onset of stimulus
was recorded, a number between 0 and 500 ms. In order to visualize this information, the
action potential times were plotted as a histogram, with 25 ms bins. These results can be
seen in Fig. 5-11 for each of the 6 electrodes for the single-material diamond neural probe.
86

Fig. 5-12 has histograms for the diamond-based neural probe. Fig. 5-13 and Fig. 5-14
have the histograms from the silicon-based neural probe. In the figures, stimulus was
applied from 0 to 200 ms.
For the single-material diamond neural probes, it is apparent that the number of
action potentials occurring near the onset of stimulus (first 50 ms) is much higher than the
rest of the recording. These results are consistent with the expected result; there is a
sharp increase in firing rate at the onset of stimulus. For the diamond-based neural
probes, there is an increase of firing rate during the stimulus in most of the recordings.
Several electrodes (4, 5, and 6) show little correlation to the stimulus and fewer detected
spikes. It is presumed that either these electrodes were not near neurons or (more likely)
there was a poor connection due to complications in fabrication or packaging. The siliconbased neural probes show a sharp increase in firing rate at the onset of stimulus and an
increased firing rate through the duration of the stimulus. Again, several electrodes
appear to have poor connections (electrodes 6, 11, and 14). These results all coincide
with the expected results of the experiment which suggest that all 3 probes may be viable
for use in neural recording experiments. However, the poor signal-to-noise ratio of the
single-material diamond and diamond-based neural probes may cause some action
potentials to be missed and a more thorough detection algorithm may be prudent.

87

Site 1

Site 2
100

Spikes Detected

Spikes Detected

60
40
20
0

0

100

200

300

400

50

0

500

Time After Onset of Stimulus [ms]

0

100

Site 3

400

500

400

500

400

500

100

Spikes Detected

Spikes Detected

300

Site 4

80
60
40
20
0

0

100

200

300

400

50

0

500

Time After Onset of Stimulus [ms]

0

100

Site 5

300

Site 6
Spikes Detected

80

50

0

200

Time After Onset of Stimulus [ms]

100

Spikes Detected

200

Time After Onset of Stimulus [ms]

0

100

200

300

400

60
40
20
0

500

Time After Onset of Stimulus [ms]

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 5-11: Histograms showing detected action potential times relative to the onset of stimulus for a
diamond SMM probe with boron-doped diamond electrodes.

88

Site 1

Site 2
40

Spikes Detected

Spikes Detected

40
30
20
10
0

0

100

200

300

400

30
20
10
0

500

Time After Onset of Stimulus [ms]

0

100

Site 3
Spikes Detected

Spikes Detected

5

0

100

200

300

400

Time After Onset of Stimulus [ms]

0

100

200

300

500

400

500

400

500

Time After Onset of Stimulus [ms]

Site 6
15

Spikes Detected

Spikes Detected

400

5

Site 5

5

0

100

200

300

400

10
5
0

500

Time After Onset of Stimulus [ms]

0

100

200

300

Time After Onset of Stimulus [ms]

Site 7

Site 8
60

Spikes Detected

30

Spikes Detected

500

10

0

500

10

20
10
0

400

15

10

0

300

Site 4

15

0

200

Time After Onset of Stimulus [ms]

0

100

200

300

400

40
20
0

500

Time After Onset of Stimulus [ms]

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 5-12: Histograms showing detected action potential times relative to the onset of stimulus for a
diamond-based neural probe with boron-doped-diamond electrodes.

89

Site 1

Site 2
60

Spikes Detected

Spikes Detected

30
20
10
0

0

100

200

300

400

40
20
0

500

Time After Onset of Stimulus [ms]

0

100

Site 3
Spikes Detected

Spikes Detected

10
5
0

100

200

300

400

0

500

Time After Onset of Stimulus [ms]

0

100

200

300

500

400

500

400

500

Time After Onset of Stimulus [ms]

Site 6
3

Spikes Detected

Spikes Detected

400

20

Site 5
40
20

0

100

200

300

400

2
1
0

500

Time After Onset of Stimulus [ms]

0

100

200

300

Time After Onset of Stimulus [ms]

Site 7

Site 8
40

Spikes Detected

60

Spikes Detected

500

40

60

40
20
0

400

60

15

0

300

Site 4

20

0

200

Time After Onset of Stimulus [ms]

0

100

200

300

400

30
20
10
0

500

Time After Onset of Stimulus [ms]

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 5-13: Histograms showing detected action potential times relative to the onset of stimulus for a
silicon-based neural probe with iridium oxide electrodes. Electrodes 1 – 8.

90

Site 9

Site 10
40

Spikes Detected

Spikes Detected

40
30
20
10
0

0

100

200

300

400

30
20
10
0

500

Time After Onset of Stimulus [ms]

0

100

Site 11
Spikes Detected

Spikes Detected

10
5
0

100

200

300

400

500

400

500

400

500

10

Time After Onset of Stimulus [ms]

0

100

200

300

Time After Onset of Stimulus [ms]

Site 14
2

Spikes Detected

Spikes Detected

400

20

Site 13
40
20

0

100

200

300

400

1.5
1
0.5
0

500

Time After Onset of Stimulus [ms]

0

100

200

300

Time After Onset of Stimulus [ms]

Site 15

Site 16
30

Spikes Detected

60

Spikes Detected

500

30

0

500

60

40
20
0

400

40

15

0

300

Site 12

20

0

200

Time After Onset of Stimulus [ms]

0

100

200

300

400

20
10
0

500

Time After Onset of Stimulus [ms]

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 5-14: Histograms showing detected action potential times relative to the onset of stimulus for a
silicon-based neural probe with iridium oxide electrodes. Electrodes 9 – 16.

91

The second recording experiment also came from the auditory cortex of a guinea pig.
In this experiment, however, different frequency stimuli were applied rather than wideband audio stimuli. Detected action potentials were recorded for each electrode. The
expectation is that the firing rate should increase for certain audio frequencies as neurons
are often tuned to certain audio frequencies. Furthermore, different electrodes should be
sensitive to different audio ranges.
In order to determine the firing rate for different frequencies, the number of action
potentials detected for a particular frequency are summed across the 30 repetitions of the
experiment. The stimulus time for a particular frequency was also summed from all 30
repetitions. The firing rate was calculated by dividing the total number of detected action
potentials for a given frequency by the total amount of time the frequency was applied. In
addition, the average firing rate when no stimulus was applied was calculated in a similar
fashion. The average firing rate versus applied stimulus frequency can be seen in Fig. 515. The hashed horizontal line represents the average firing rate with no stimulus applied
and can be considered the background measurement. The results show an increase in
neural activity when the stimulus is applied. The firing rate is consistently higher during
stimulus than when no stimulus is applied. However, the firing rate does not appear to be
frequency dependent. For all electrodes, the firing rate seems to be slightly higher for
lower frequencies. For example, on electrode 6 there seems to be a peak at about 2 kHz
with a firing rate of about 7.5 Hz, whereas the average firing rate about 4 kHz seems to
be about 4.5 Hz. While there appears to be some correlation between frequency and
firing rate, it does not appear to be significant. As such the experimental results are

92

inconsistent with the expected results. This is likely due to poor SNR of the original
recording and the simple thresholding method used to detect action potentials.

15
10
5

25
20
15
10
5
0 3
10

4

10

Audio Frequency [Hz]

25
20
15
10
5
0 3
10

4

10

Audio Frequency [Hz]

30

Average Firing Rate [Hz]

20
15
10
5
4

10

Audio Frequency [Hz]

Average Firing Rate [Hz]

30

25

25
20
15
10
5
0 3
10

4

10

Audio Frequency [Hz]

Electrode 6

Electrode 5

Electrode 4
30

Average Firing Rate [Hz]

Average Firing Rate [Hz]

Average Firing Rate [Hz]

Average Firing Rate [Hz]

20

0 3
10

30

30

25

0 3
10

Electrode 3

Electrode 2

Electrode 1
30

4

10

Audio Frequency [Hz]

25
20
15
10
5
0 3
10

4

10

Audio Frequency [Hz]

Figure 5-15: Average firing rate versus frequency of applied audio stimulus for a single-material diamond
neural probe. The hashed line represents the background average firing rate when no stimulus was
applied.

93

5.4

Conclusion
Single-material diamond, diamond-based, and silicon-based neural probes have

been surgically implanted into the audio cortex of a guinea pig for in-vivo electrical
recording. This is the first time that a single-material neural probe has been tested in vivo
and the first time that a single-material MEMS device was tested in vivo. A wide-band
audio stimulus was applied to the ear of the guinea pig, and the resulting neural activity
was recorded with each type of probe. As expected, each type of probe showed a rise in
neural firing rate during the stimulus, while the single-material diamond and silicon-based
neural probes showed a significant rise at the onset of stimulus. This demonstrates that
the single-material diamond probe can successfully record action potentials in vivo. A
second experiment was performed with the single-material diamond neural probes where
the frequency response of the neurons was tested. There was an increase in firing rate
during the stimuli. However, there was not an appreciable frequency dependency for any
of the recording sites.

94

REFERENCES

95

REFERENCES

[1] F.C. Hellweg, R. Kock, and M. Volrath, “Representation of the cochlea in the
neocortex of guinea pigs,” Exp. Brain Res., vol. 29, pp.467–474, 1977.
[2] H. Redies, U. Sieben, and O.D. Creutzfeldt, “Functional subdivisions in the auditory
cortex of the guinea pig,” J. Comp. Neurol., vol. 282, pp. 473–488, 1989.
[3] J. G. Arenberg, S. Furukawa, and J.C. Middlebrooks, “Auditory cortical images of
tones and noise bands,” Journal of the Association for Research in Otolaryngology,
vol. 1, no. 2, pp. 183-194, 2000.

96

CHAPTER: 6
Action Potential Detection in Low Signal-to-Noise Applications

6.1

Introduction
Recordings of electrical signals from neurons are critical in developing brain-machine

interfaces and neuroprosthetics, studying neural disorders, or researching other
neuroscience applications. Unfortunately, these recordings are typically buried within
noise signals, as can be seen in the real neural recordings in Fig. 6-1. In an ideal
experiment, electrodes can be positioned near neurons and a very high signal-to-noise
ratio (SNR) can be achieved. However in most real experiments, complications arise
negatively impacting SNR. In chronic implantations, this can be due to movement of the
electrode or biofouling of the electrode. This can result in significant degradation of SNR
that can cause traditional action potential detection methods (thresholding [1]) to be
ineffective. Biofouling can affect recording action potentials more than just affecting the
amplitude of the spike. The layer of protein that adheres to the electrode surface adds not
only a resistive component to the impedance of the electrode (lowering the detected
voltage), but also a capacitive component which affects the spectral composition of the
action potential. This can affect the performance of band-pass filtering as a method of
reducing noise, as the range of frequencies of interest can change during chronic
implantation. There are additional reasons why electrode arrays could have poor SNR,
including poor electrode placement (it is improbable that every electrode in an array would
be ideally placed), sparsely located neurons, electrode material, or tissue damage during

97

surgical implantation. To compensate for the low SNR inherent in neural recordings, many
techniques have been developed to attempt to detect action potentials from noisy signals.

98

(a)
1500

Amplitude [ V]

1000

500

0

-500

-1000

0.2

0.4

0.6

0.8

1

Time [s]

1.2

1.4

1.6

1.8

2

1.2

1.4

1.6

1.8

2

(b)
600

400

Amplitude [ V]

200

0

-200

-400

-600

0.2

0.4

0.6

0.8

1

Time [s]

Figure 6-1: Graphs of neural recordings taken from the audio cortex of a guinea pig using (a) an iridium
oxide electrode and (b) a boron-doped-diamond electrode.

99

Some techniques can be extremely useful in certain situations where the expected
action potential shape is known before the experiment, or the action potential detection
can be supervised, and a template devised during the experiment. However, in chronic
implantations where the shape of the action potentials might change, or in experiments
where supervision is impractical, these methods (matched filtering [2] or principal
components [3]) are not sufficient. Instead, methods that are unsupervised and require
no a priori knowledge are of the greatest interest. The methods being considered involve
power detection [4], the discrete wavelet transform (DWT) [5-7], and the continuous
wavelet transform (CWT) [8]. The goal of this work is to compare these methods by using
them to detect simulated neural spikes in varying amounts (and types) of noise. This work
differs from previous work because the same method for identifying action potentials is
used for each transform method (the DWT, CWT, and energy transform). This work
should identify which unsupervised method(s) are best suited for use in detecting action
potentials in very poor SNR applications.

100

6.2

Action Potential Detection Techniques

6.2.1 Traditional Techniques
Thresholding
The simplest method of detecting action potentials is thresholding. In thresholding,
the signal level is compared to a preset threshold (requiring supervision), and anytime the
signal exceeds the threshold, an action potential is detected. Additional thesholding
techniques include absolute thesholding and double thesholding (Also known as the
window discriminator). In absolute thresholding, the absolute value of the signal is taken
before thresholding (any DC component is typically removed before taking the absolute
value). Thresholding and absolute thresholding don’t take any account of frequency or
event length, they depend solely on voltage. Double thresholding depends on event
length, as well as the amplitude, of the recorded signal. In double thresholding, the signal
must exceed a first threshold, then fall beneath a second threshold, within a certain period
of time. Examples showing each thresholding method can be seen in Fig. 6-2.
Thresholding can be unsupervised, being set from estimates of the mean and standard
deviation of the noise. More advanced detection techniques always rely on some form of
thresholding for ultimately detecting action potentials.

101

(a)

(b)

(c)

Figure 6-2: Demonstration of how thresholding is performed, with the top row showing the input signal, the
second row showing the threshold being applied, and the bottom row showing the detected neural spikes,
for (a) single thresholding, (b) absolute thresholding, and (c) double thresholding.

Band-Pass Filtering
Band-pass filtering takes advantage of the fact that action potentials are highly
stereotyped and have a particular power spectrum. If the power spectrum of the action
potentials is known, band-pass filtering can be used to remove noise from other
frequencies. This is very useful when there is little overlap between the power spectra of
the action potential signal and the noise signal. However, this is typically not the case.
There are typically two large sources of noise in the recorded signal. The first source of
noise comes from the equipment being used to record the signal. This noise is typically
adequately modeled by white noise, and band-pass filtering can be used to sufficiently
filter this noise in all but the most extreme noise levels. The second source of noise comes
from the recording environment, made up of signals from nearby neurons and the
superposition of many far-away neurons. Unfortunately since this noise is largely
generated from action potentials, the power spectrum is not white, but resembles the
spectrum of individual action potentials. Due to the related spectra, band-pass filtering is
not very effective at removing this noise.

102

6.2.2 Advanced Techniques
Discrete Wavelet Transform
Several techniques have been developed that use the discrete wavelet transform
(DWT). The DWT is a transformation technique similar to the discrete Fourier transform
(DFT), but whereas the DFT converts temporal information into frequency information,
the DWT retains temporal information as well as obtaining frequency information. This is
accomplished by the use of wavelets in the transformation rather than the waves
(sinusoids) used in the Fourier transform. The wavelet transformation is defined by:
∞

𝑊(𝛼, 𝜏) = ∫ 𝑥(𝑡)
−∞

1
𝛼 1⁄2

𝜓(

𝑡−𝜏
) 𝑑𝑡
𝛼

Where W(α,τ) are the wavelet coefficients for scale (α) and translation (τ), Ψ(t) is the
mother wavelet, and x(t) is the original signal. For the DWT, dydadic scales and
translations from a discrete set (α = 2j; τ = k*2j; j,k ∈ Z) are necessary so that the basis
functions (Ψ[α,τ]) form an orthonormal basis. This means that the data at each scale is
unique in that the original signal can be reconstructed from the approximations at all of
the scales. Different levels of the DWT give different resolutions of temporal versus
frequency information. As the frequency resolution increases (low frequencies), the
temporal resolution decreases. Whereas at high frequencies, the frequency resolution is
poor, and the temporal resolution is increased. The DWT works as an action potential
detector because at some scales (levels) the frequency spectra of the wavelet match the
frequency spectra of the action potential. One technique uses a non-linear combination
of multiple DWT scales to detect action potentials [5]. This technique has excellent results
under certain conditions, but requires a priori information about the action potentials being
103

detected and predictable noise. Another DWT technique uses a sum of multiple DWT
scales [6]. This technique has better overall performance than the previous technique as
shown by [5]. Other techniques include array denoising using the DWT [7], and a
simplified version of the DWT known as wavelet lifting (WL). One of the most promising
aspects of the DWT method is the simplicity of implementation. The WL method further
simplifies the implementation making it an ideal candidate for real-time detection for
implantable devices. One drawback of the DWT is that it is negatively impacted by nonstationary noise.
Power Detection
Power detection is also known as energy detection and involves calculating the
instantaneous power of the signal and comparing it to the mean and standard deviation
of the power to detect neural spikes. The power detection methods of interest use the
Teager energy operator (TEO) (also known as the Teager-Kaiser energy operator) and
the multi-resolution Teager energy operator (MTEO) [4]. The TEO in discrete time is
simply:
𝑃[𝑛] = 𝑥 2 [𝑛] − 𝑥[𝑛 − 1]𝑥[𝑛 + 1]
Where P is the power of the signal x. The output from an energy operator is typically
windowed. The TEO can also be changed to work at different resolutions. The TEO at
different scales is typically known as the kTEO, which is calculated by:
𝑃[𝑛] = 𝑥 2 [𝑛] − 𝑥[𝑛 − 𝑘]𝑥[𝑛 + 𝑘]
Where k indicates the resolution of the operator. The MTEO uses multiple kTEO’s in order
to detect action potentials. This results in a much more robust action potential detector.
104

The limitation of the MTEO is in cases of severe SNR, where the MTEO has more false
positives than other techniques.
Continuous Wavelet Transform
The CWT is almost identical to the DWT, even the definition is the same:
∞

𝑊(𝛼, 𝜏) = ∫ 𝑥(𝑡)
−∞

1
𝛼 1⁄2

𝜓(

𝑡−𝜏
) 𝑑𝑡
𝛼

The difference between the DWT and CWT lies in the selection of scales. The DWT
is for discrete signals with dydadic scales and translations, ensuring the original signal
can be recovered from the transform and the basis of the transform is orthonormal. The
CWT has discrete signals, scales, and translations, but they scales and translation need
not be dydadic, leading to redundancy in the wavelet coefficients and inability to recreate
the original signal. The translations are discrete and determined by the sampling rate, i.e.
τ=n/fs where n is the sample number and fs is the sampling frequency. The scales are
also discrete, but determined by the user depending on the expected length of the action
potential. Action potential lengths are highly regular, although the lengths vary in different
animals and brain areas. The CWT only considers events with lengths similar to the
expected action potential. This serves to limit much of the noise from other frequencies.
For this reason, the CWT is expected to have the fewest false positives in poor SNR
applications.

105

6.3

Methods

6.3.1 Simulating Neural Signals
In order to determine which action potential detection method performs the best for
low SNR applications, a sample signal needs to be generated in order to test the detection
methods. For this purpose, simulated action potentials were used, which can be seen in
Fig. 6-3. The simulated signal had a sampling frequency of 10 kHz.

106

(a)
1

0.5

Amplitude

0

-0.5

-1

-1.5

-2
0

0.2

0.4

0.6

0.8

1

Time [s]

1.2

1.4

1.6

1.8

2

6

7

8

9

10

(b)
1

0.5

Amplitude

0

-0.5

-1

-1.5

-2
0

1

2

3

4

5

Time [ms]

Figure 6-3: Graphs of (a) a simulated neural spike train and (b) an individual simulated neural spike.

107

To thoroughly test the detection methods, several forms of noise were added to the
signal. Two different colors of noise were considered. White noise was considered
because it accurately represents noise generated from recording hardware. Pink noise
was also considered since it approximates the power spectrum of noise from the
recording environment. The power spectra of the clean signal, white noise, and pink noise
can all be seen in Fig. 6-4. It is apparent that action potential detection should be more
difficult in pink noise. In addition to two colors of noise, non-stationary noise was also
considered. To generate non-stationary noise, the amplitude of the noise was varied
throughout a short (10 second) simulated neural signal. An example of non-stationary
pink noise added to the simulated spike train can be seen in Fig. 6-5. So the four forms
of noise considered are stationary white noise, non-stationary white noise, stationary pink
noise, and non-stationary pink noise.

108

(a)

(b)

(c)

800

600

100

50

800

500

Power

Power

Power

1000

700

150

400
300
200

1000 2000 3000 4000 5000

0

Frequency (Hz)

400
200

100
0

600

1000 2000 3000 4000 5000

0

Frequency (Hz)

1000 2000 3000 4000 5000

Frequency (Hz)

(d)

5

x 10
9
8
7

Power

6
5
4
3
2
1

0

1000

2000

3000

4000

5000

6000

Frequency (Hz)

7000

8000

9000

10000

Figure 6-4: Power spectra of (a) a simulated neural spike train, (b) generated white noise, (c) generated
pink noise, and (d) recorded neural signal shown in Fig. 6-1, from a boron-doped-diamond electrode.

109

3

2

Amplitude

1

0

-1

-2

-3

0.2

0.4

0.6

0.8

1

Time [s]

1.2

1.4

1.6

1.8

2

Figure 6-5: Graph of simulated neural spike train with non-stationary pink noise added with a base SNR of
-20 dB.

To test each detection method, 1000 trials were conducted for each form of noise
over a range of different SNRs. The SNR was determined by the ratio of the power of the
clean signal to the power of the noise. The SNRs used ranged from -10 dB to -20 dB in 1
dB increments. The final variable varied was a cost measure, l. This variable represents
the ratio of cost of false alarms to the cost of omission of real spikes. The range
considered covered ratios of less than .001 and greater than 1000. In total, 396,000 trials
were performed.

110

6.3.2 DWT Method
As seen by the spectrum of the simulated neural spikes (Fig 6-4), the principle
frequencies of interest range from about 250 Hz to 1250 Hz (.8 to 4 ms). With a sampling
rate of 10 kHz, the first 4 dydadic scales of the DWT correspond to 5000 Hz, 2500 Hz,
1250 Hz, and 625 Hz (.2, .4, .8, and 1.6 ms). These are the scales considered in the DWT
algorithm to detect action potentials. The wavelet used for the DWT was the
biorthonormal 1.5 wavelet, as its shape is reminiscent of a biphasic action potential.

6.3.3 MTEO Method
For the MTEO detection method, 4 kTEO filters were used (k=5, 10, 15, and 20
[corresponding to frequencies of 2000, 1000, 667, and 500 Hz and event lengths of .5, 1,
1.5, and 2 ms]). Each kTEO filter was windowed.

6.3.4 CWT Method
For the CWT detection, again 4 scales were used, this time corresponding to .5, 1,
1.5, and 2 ms (or 2000, 1000, 667, and 500 Hz; the same as the MTEO method). The
wavelet used for the CWT was the biorthonormal 1.5 wavelet.

6.3.5 Resolving Action Potentials
The method of detecting action potentials at each scale, combining the decisions at
individual scales, and estimating the spike arrival times is described by Nenandic et al.
[8], whose method was based on work from Donoho et al. [9]. At each individual scale,
the noise level is estimated from the mean, median, and standard deviation of the wavelet
coefficients. Any coefficients exceeding a threshold are marked. The coefficients from
111

each scale exceeding their noise threshold are all combined and if the event length is less
than the longest expected neural spike event length, it is considered an action potential.
The cost variable l is used in setting the threshold at each scale. The results from the
detection algorithm were then compared to the times of simulated action potentials. The
number of true positives, false positives, and omissions were tracked for all trials.

112

6.4

Results
For every combination of cost variable l (-0.2, -0.15, -0.1, -0.05, 0, 0.05, 0.1, 0.15,

0.2), SNR (-10, -11, -12, -13, -14, -15, -16, -17, -18, -19, -20), and noise type (white
Gaussian noise [WGN], non-stationary white Gaussian noise [NSWGN], pink Gaussian
noise [PGN], non-stationary pink Gaussian noise [NSPGN]), 1000 simulations were run
for each detection method (DWT, CWT, and MTEO). The number of correct detections
(true positives) from each trial was recorded and averaged, along with the number of false
alarms (false positives). The number of simulated spikes in each test was 10, so the
number of omissions is simply the number of true positives subtracted from 10.
For the case where the cost of omissions is equal to the cost of false positives (l=0)
the average number of true positives versus the SNR is plotted in Fig. 6-6 for each of the
noise types. In Fig. 6-6 (and subsequent figures), the DWT method’s results are plotted
in blue, the CWT method’s results are plotted in red, and the MTEO method’s results are
plotted in green. Fig. 6-7 shows the number of false positives for the same equal cost
case. In terms of number of true positives, it would appear that the MTEO method has
the best performance at low SNR, while the CWT method had the best performance in
higher SNR. In terms of number of false positives, the DWT method has the fewest under
a SNR lower than -14 dB for white noise and lower than -12 dB for pink noise. In signals
with relatively little noise (-10 dB) the MTEO method has the fewest false positives in
white noise. However, in poor SNR applications, the MTEO method results in a staggering
number of false positives for all noise types. To help determine which wavelet method
has better performance in low SNR applications, the ratio of false positives to true

113

positives was evaluated, seen in Fig. 6-8. The false positive to true positive ratio is
comparable between the two methods.
(b)
10

9

9

8

8

7

7

True Positives

True Positives

(a)
10

6
5
4

6
5
4

3

3

2

2

1
-20

-18

-16

-14

SNR [dB]

-12

1
-20

-10

-18

-16

9

9

8

8

7

7

6
5
4

-14

-12

-10

5
4
3

2

2

1

1
-16

-10

6

3

-18

-12

(d)
10

True Positives

True Positives

(c)
10

0
-20

-14

SNR [dB]

-14

SNR [dB]

-12

0
-20

-10

-18

-16

SNR [dB]

Figure 6-6: True positives detected action potentials (out of 10) versus SNR for simulated neural signals
with added (a) stationary white Gaussian noise, (b) non-stationary white Gaussian noise, (c) stationary pink
Gaussian noise, and (d) non-stationary pink Gaussian noise. The cost ratio variable, l, was set to 0, where
the cost of omissions was equal to the cost of false positives. Blue represents the DWT method, red
represents the CWT method, and green represents the MTEO method.

114

(a)

(b)

80

100
90

70

80
60

False Positives

False Positives

70
50
40
30

60
50
40
30

20
20
10
0
-20

10
-18

-16

-14

SNR [dB]

-12

0
-20

-10

-18

-16

-12

-10

-14

-12

-10

(d)
120

100

100

80

80

False Positives

False Positives

(c)
120

60

40

20

0
-20

-14

SNR [dB]

60

40

20

-18

-16

-14

SNR [dB]

-12

0
-20

-10

-18

-16

SNR [dB]

Figure 6-7: False positives detected versus SNR for simulated neural signals with added (a) stationary
white Gaussian noise, (b) non-stationary white Gaussian noise, (c) stationary pink Gaussian noise, and (d)
non-stationary pink Gaussian noise. The cost ratio variable, l, was set to 0, where the cost of omissions
was equal to the cost of false positives. Blue represents the DWT method, red represents the CWT method,
and green represents the MTEO method.

115

(a)

(b)
12

9

Ratio of False Positives to True Positives

Ratio of False Positives to True Positives

10

8
7
6
5
4
3
2
1
0
-20

-18

-16

-14

SNR [dB]

-12

10

8

6

4

2

0
-20

-10

-18

-16

(c)

-10

-14

-12

-10

16

Ratio of False Positives to True Positives

Ratio of False Positives to True Positives

-12

(d)

16
14
12
10
8
6
4
2
0
-20

-14

SNR [dB]

-18

-16

-14

SNR [dB]

-12

14
12
10
8
6
4
2
0
-20

-10

-18

-16

SNR [dB]

Figure 6-8: Ratio of false positives to true positives versus SNR for simulated neural signals with added (a)
stationary white Gaussian noise, (b) non-stationary white Gaussian noise, (c) stationary pink Gaussian
noise, and (d) non-stationary pink Gaussian noise. The cost ratio variable, l, was set to 0, where the cost
of omissions was equal to the cost of false positives. Blue represents the DWT method, red represents the
CWT method, and green represents the MTEO method.

In order to establish which method has the best performance, additional cost ratio
cases were evaluated. Figs. 6-9 and 6-10 show the number of true positives detected by

116

each method over a wide range of cost ratios. Figs. 6-11 and 6-12 show the number of
false positives detected by the methods for the same range of cost ratios. It would appear
that overall the CWT method appears to perform the best at low SNR, while MTEO might
be the best at high SNR. In order to quantify which method had the best performance, a
new figure of merit was devised. The figure of merit was calculated by the equation:
𝐹𝑖𝑔𝑢𝑟𝑒 𝑜𝑓 𝑀𝑒𝑟𝑖𝑡 = (1 + [# 𝑜𝑓 𝑂𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠]) ∗ (1 + [# 𝑜𝑓 𝐹𝑎𝑙𝑠𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠])
This figure of merit has a minimum value of 1 when there are no omissions or false
positives. The values for the figure of merit were calculated for each value of l and each
SNR. The minimum values across all l values were taken and plotted against SNR in Fig.
6-13. Fig. 6-13 shows very clearly that the CWT method has the best overall performance
over the entire range of SNR evaluated and for all noise types considered. However, all
methods have comparable performance in cases of high SNR (-10 dB). It is also worth
noting that all methods have considerably poorer performance in the presence of pink
noise, as compared to white noise, with poor SNR.

117

NSWGN, l=-0.2

0
-20

NSWGN, l=-0.15

-15

SNR [dB]

5

0
-20

-10

WGN, l=-0.1

WGN, l=-0.05

-15

SNR [dB]

WGN, l=0
True Positives
-15

SNR [dB]

-10

0
-20

-10

-15

SNR [dB]

-10

-10

NSPGN, l=-0.05

-15

SNR [dB]

5

0
-20

-10

-15

SNR [dB]

-10

NSPGN, l=0
10

5

0
-20

-15

SNR [dB]

10

10

5

0
-20

5

PGN, l=0

10

5

-15

SNR [dB]

5

0
-20

-10

-10

NSPGN, l=-0.1

PGN, l=-0.05
True Positives

-15

SNR [dB]

-15

SNR [dB]

10

10

NSWGN, l=0

10

True Positives

0
-20

-10

5

0
-20

-10

SNR [dB]

0
-20

-10

5

NSWGN, l=-0.05

5

0
-20

-15

SNR [dB]

10

True Positives

True Positives

True Positives

0
-20

-10

-15

5

PGN, l=-0.1

5

-10

NSPGN, l=-0.15

10

True Positives

True Positives

True Positives
-15

SNR [dB]

10

0
-20

SNR [dB]

0
-20

-10

10

5

0
-20

-15

-15

SNR [dB]

10

5

NSWGN, l=-0.1

10

0
-20

-10

10

True Positives

5

-15

SNR [dB]

5

PGN, l=-0.15

10

True Positives

True Positives

10

0
-20

0
-20

-10

True Positives

WGN, l=-0.15

-15

SNR [dB]

5

True Positives

-10

True Positives

5

10

True Positives

-15

SNR [dB]

NSPGN, l=-0.2

10

True Positives

True Positives

True Positives

5

0
-20

PGN, l=-0.2

10

True Positives

WGN, l=-0.2
10

-15

SNR [dB]

-10

5

0
-20

-15

SNR [dB]

-10

Figure 6-9: True positives (out of 10) versus SNR for l values with cost ratios ranging from .001 to 1 (cost
of omissions is higher than cost of false positives). Blue represents the DWT method, red represents the
CWT method, and green represents the MTEO method.

118

WGN, l=0

NSWGN, l=0

0
-20

NSWGN, l=0.05

-15

SNR [dB]

WGN, l=0.1

WGN, l=0.15

0
-20

-10

WGN, l=0.2

-15

SNR [dB]

NSWGN, l=0.2
True Positives

-15

SNR [dB]

-10

-15

-15

SNR [dB]

SNR [dB]

5

0
-20

-10

-15

SNR [dB]

-10

NSPGN, l=0.2
10

5

0
-20

-10

NSPGN, l=0.15

PGN, l=0.2

-10

-15

SNR [dB]

10

10

5

0
-20

0
-20

-10

5

0
-20

-10

10

5

-15

SNR [dB]

5

PGN, l=0.15

5

-10

NSPGN, l=0.1

10

True Positives

True Positives
-15

SNR [dB]

10

True Positives

0
-20

-10

10

5

0
-20

-15

SNR [dB]

-15

SNR [dB]

10

5

NSWGN, l=0.15

10

True Positives

True Positives

5

0
-20

-10

SNR [dB]

0
-20

-10

10

True Positives

True Positives

True Positives
-15

SNR [dB]

-15

5

PGN, l=0.1

10

5

0
-20

SNR [dB]

0
-20

-10

-10

NSPGN, l=0.05

5

NSWGN, l=0.1

10

0
-20

-15

-15

SNR [dB]

10

True Positives

5

0
-20

-10

0
-20

-10

10

True Positives

5

-15

SNR [dB]

5

PGN, l=0.05

10

True Positives

True Positives

10

0
-20

0
-20

-10

True Positives

WGN, l=0.05

-15

SNR [dB]

5

True Positives

-10

True Positives

5

10

True Positives

-15

SNR [dB]

NSPGN, l=0

10

True Positives

5

0
-20

PGN, l=0

10

True Positives

True Positives

10

-15

SNR [dB]

-10

5

0
-20

-15

SNR [dB]

-10

Figure 6-10: True positives (out of 10) versus SNR for l values with cost ratios ranging from 1 to 1000 (cost
of omissions is lower than cost of false positives). Blue represents the DWT method, red represents the
CWT method, and green represents the MTEO method.

119

NSWGN, l=-0.2

0
-20

WGN, l=-0.15

50
-15

SNR [dB]

150
100
50
0
-20

-10

WGN, l=-0.1

-15

SNR [dB]

100
50
0
-20

-10

WGN, l=-0.05

50

-15

SNR [dB]

50
0
-20

-10

-15

SNR [dB]

20

SNR [dB]

-10

-15

SNR [dB]

-10

-15

SNR [dB]

-10

100
50
0
-20

-10

-15

SNR [dB]

-10

NSPGN, l=0
150

100
50
0
-20

-15

SNR [dB]

NSPGN, l=-0.05

PGN, l=0

50

0
-20

50

150

150

False Positives

False Positives

40

100

0
-20

-10

50

NSWGN, l=0
100

60

-15

SNR [dB]

100

0
-20

-10

-10

150

PGN, l=-0.05

100

WGN, l=0

-15

50

150

False Positives

False Positives

False Positives

100

-15

SNR [dB]

NSPGN, l=-0.1

100

0
-20

-10

150

80

False Positives

-15

SNR [dB]

50

200

150

NSWGN, l=-0.05

150

SNR [dB]

100

0
-20

-10

200

False Positives

False Positives

False Positives

50

-15

150

PGN, l=-0.1

150

100

0
-20

SNR [dB]

-10

NSPGN, l=-0.15

100
0
-20

-10

-15

SNR [dB]

200

200

NSWGN, l=-0.1

150

0
-20

-15

100
0
-20

-10

300

False Positives

False Positives

False Positives

100

-15

SNR [dB]

200

PGN, l=-0.15

200

150

0
-20

0
-20

-10

NSWGN, l=-0.15

200

0
-20

-15

SNR [dB]

100

False Positives

-10

False Positives

100

200

False Positives

-15

SNR [dB]

200

300

False Positives

100

NSPGN, l=-0.2

300

False Positives

False Positives

False Positives

200

0
-20

PGN, l=-0.2

300

False Positives

WGN, l=-0.2
300

-15

SNR [dB]

-10

100
50
0
-20

-15

SNR [dB]

-10

Figure 6-11: False positives versus SNR for l values with cost ratios ranging from .001 to 1 (cost of
omissions is higher than cost of false positives). Blue represents the DWT method, red represents the CWT
method, and green represents the MTEO method.

120

WGN, l=0

NSWGN, l=0

0
-20

WGN, l=0.05

-15

SNR [dB]

60
40
20
0
-20

-10

WGN, l=0.1

-15

SNR [dB]

20
0
-20

-10

WGN, l=0.15

-15

SNR [dB]

0
-20

-10

WGN, l=0.2

-15

SNR [dB]

SNR [dB]

-10

20
10
-15

SNR [dB]

-10

False Positives
-15

SNR [dB]

-10

60
40
20
0
-20

-10

-15

SNR [dB]

-10

NSPGN, l=0.2
60

40
20
0
-20

-15

SNR [dB]

NSPGN, l=0.15

PGN, l=0.2

30

0
-20

20

80

60

False Positives

False Positives

10

40

0
-20

-10

20

NSWGN, l=0.2

20

-15

SNR [dB]

40

0
-20

-10

40

-15

20

60

PGN, l=0.15

20

-10

NSPGN, l=0.1

60

40

-15

SNR [dB]

80

40

0
-20

-10

False Positives

False Positives

10

30

False Positives

-15

SNR [dB]

60

20

SNR [dB]

50

0
-20

-10

60

NSWGN, l=0.15

30

False Positives

False Positives

False Positives

False Positives

20

-15

80

40

-10

NSPGN, l=0.05

PGN, l=0.1

60

40

0
-20

SNR [dB]

0
-20

-10

-15

SNR [dB]

100

50

NSWGN, l=0.1

60

0
-20

-15

50
0
-20

-10

100

False Positives

False Positives

False Positives

20

-15

SNR [dB]

100

PGN, l=0.05

80

40

0
-20

0
-20

-10

NSWGN, l=0.05

60

0
-20

-15

SNR [dB]

False Positives

-10

50

False Positives

-15

SNR [dB]

100

False Positives

20

50

150

False Positives

40

NSPGN, l=0

150

False Positives

60

0
-20

PGN, l=0

100

False Positives

False Positives

80

-15

SNR [dB]

-10

40
20
0
-20

-15

SNR [dB]

-10

Figure 6-12: False positives versus SNR for l values with cost ratios ranging from 1 to 1000 (cost of
omissions is lower than cost of false positives). Blue represents the DWT method, red represents the CWT
method, and green represents the MTEO method.

121

(a)

(b)

120

160
140

100

80

Figure of Merit

Figure of Merit

120

60

40

100
80
60
40

20

0
-20

20

-18

-16

-14

SNR [dB]

-12

0
-20

-10

-18

-16

(c)

-10

-14

-12

-10

300

250

Figure of Merit

200

Figure of Merit

-12

(d)

250

150

100

50

0
-20

-14

SNR [dB]

200

150

100

50

-18

-16

-14

SNR [dB]

-12

0
-20

-10

-18

-16

SNR [dB]

Figure 6-13: Figure of merit versus SNR for each method for simulated neural signals with added (a)
stationary white Gaussian noise, (b) non-stationary white Gaussian noise, (c) stationary pink Gaussian
noise, and (d) non-stationary pink Gaussian noise. The figure of merit is calculated by (1 + [omissions]) *
(1 + [false positives]) and taking the lowest value across all values of l. Blue represents the DWT method,
red represents the CWT method, and green represents the MTEO method.

122

6.5

Conclusion
Accurate action potential detection is important in neuroscience studies. Extracellular

recordings tend to have poor SNR due to electrode placement, biofouling, and outside
noise sources. Traditional methods of detecting action potentials were reviewed and
several methods of detecting action potentials in low SNR applications were explored
(DWT, CWT, MTEO).
In order to determine which method had the best performance in low SNR
applications, Monte Carlo simulations were performed to assess the performance of each
method at various SNR. In addition, four types of noise were used in these trials,
stationary and non-stationary white Gaussian noise, and stationary and non-stationary
pink Gaussian noise. The MTEO method had the worst performance of the 3, except in
cases with high SNR, where the performance was comparable. Between the DWT and
CWT, the CWT method has better results than the DWT (more significantly in white noise
than pink noise). However, the results are comparable, which means DWT might still be
a better method for real-time spike detection, since it’s algorithm is simpler to implement;
in addition, the DWT can be used to re-create the original signal making it also an effective
compression method for transmitting neural data. In post-processing of neural signals,
the CWT appears to be the best choice of unsupervised action-potential detection in low
SNR applications.

123

REFERENCES

124

REFERENCES

[1] J. P. Welsh and C. Schwarz, “Multielectrode recording from the cerebellum,” in
Methods for Neural Ensemble Recordings, M. A. L. Nicolelis, Ed. Boca Raton, FL:
CRC, ch. 5, pp. 79–100, 1999.
[2] I. N. Bankman, K. O. Johnson, and W. Schneider, “Optimal detection, classification,
and superposition resolution in neural waveform recordings,” IEEE Trans. Biomed.
Eng., vol. 40, no. 8, pp. 836–841, Aug. 1993.
[3] M. Abeles and M. H. Goldstein, “Multispike train analysis,” Proc. IEEE, vol. 65, pp.
762–773, 1977.
[4] J. H. Choi, H. K. Jung, T. Kim, “A new action potential detector using the MTEO and
its effects on spike sorting systems at low signal-to-noise ratios,” IEEE Trans.
Biomed. Eng., vol. 53, no. 4, pp. 738–7346, April 2006.
[5] K. H. Kim, S. J. Kim, “A wavelet-based method for action potential detection from
extracellular neural signal recording with low signal-to-noise ratio,” IEEE Trans.
Biomed. Eng., vol. 50, no. 8, pp. 999–1011, Aug. 2003.
[6] V. Shalchyan, W. Jensen, D. Farina, “Spike detection and clustering with unsupervised
wavelet optimization in extracellular neural recordings,” IEEE Trans. Biomed. Eng.,
vol. 59, no. 9, pp. 2576–2585, Sept. 2012.
[7] K. G. Oweiss, D. J. Anderson, “Noise reduction in multichannel neural recordings using
a new array wavelet denoising algorithm,” Neurocomputing, vol. 38–40, pp. 16871693, June 2001.
[8] Z. Nenandic, J. W. Burdick, “Spike detection using the continuous wavelet transform,”
IEEE Trans. Biomed. Eng., vol. 52, no. 1, pp. 74–87, Jan. 2005.
[9] D. L. Donoho and I. M. Johnstone, “Ideal spatial adaptation by wavelet shrinkage,”
Biometrika, vol. 81, pp. 425–455, 1994.

125

CHAPTER: 7
Analysis of Neural Recording Experiments using the Continuous
Wavelet Transform Method

7.1

Introduction
The efficacy of an acute neural recording device is best measured by its ability to

accurately detect action potentials. In the neural recordings performed with diamondbased and single-material diamond neural probes (Chapter 5), it was found that there is
a significant amount of noise in the recorded signals. The poor signal-to-noise ratio (SNR)
of these recordings suggests that it would be hard to detect action potentials accurately
using a simple thresholding technique. For this reason, different detection techniques
were explored to determine which method performs the best with very poor SNR
recordings (Chapter 6). The results showed that wavelet techniques far outperformed
power detection techniques in noisy recordings. In addition, it was found that the
continuous wavelet transform (CWT) method had the best overall detection rates. The
aim of this chapter is to apply the CWT method to detect action potentials in neural
recordings taken using silicon-based, diamond-based, and single-material diamond
neural probes. The results will be analyzed to compare the thresholding and CWT
detection techniques and to compare the efficacy of the three different probe types. The
expectation is that all of the probe types will detect more action potentials using the CWT
method instead of the basic thresholding technique, but that the probes with poorer SNR
(the diamond-based probe and especially the single-material diamond probe) will benefit
more. However, in order to demonstrate the efficacy of the diamond-based and single126

material diamond neural probes, it must be demonstrated that the CWT method allows
detection rates comparable to the silicon-based neural probes.

127

7.2 Application of Continuous Wavelet Transform Technique and
Thresholding
The CWT detection was performed as described previously in Chapter 6. Six scales
were used, corresponding to 0.5, 1, 1.5, 2, 2.5, and 3 ms (or 2000, 1000, 667, 500, 400,
and 333 Hz), and the wavelet used was the biorthonormal 1.5 wavelet. For each scale,
the noise level was estimated using the mean, median, and standard deviation of the
wavelet coefficients. The noise threshold was set using a cost variable l being equal to
zero, meaning the cost of false positives and the cost of omissions were considered equal.
Coefficients exceeding the noise threshold were combined across all four scales. Events
occurring within a typical event time (1.75 ms) of each other were combined. Events
occurring within a typical refractory period (4.5 ms), but not within a typical event time
(1.75 ms) were discarded. The remaining events, the detected action potentials, were
compared with the onset of stimulus periods to determine their arrival time relative to the
stimulus. These arrival times relative to the onset of stimulus were used to generate
histograms similar to those from the analysis of neural recordings using the thresholding
method in Chapter 5.
The basic thresholding method was reapplied similarly to Chapter 5. However, in
Chapter 5, the thresholds were set manually in order to subjectively maximize the
performance of the detection. While this does not invalidate the results from Chapter 5, it
would make it impossible to make an objective comparison of the detection methods, as
the CWT method uses automated selection of coefficient thresholds; the results would be
biased toward one of the methods (most likely the basic thresholding method). One way
to solve this inequality in methods would be to set the coefficient thresholds manually for
128

each of the scales on each electrode (180 thresholds). However, this could still lead to
some bias, as manual thresholds are ultimately susceptible to human error. Instead, the
basic thresholding technique was re-performed using the same automated noise
threshold determination used in the CWT method. That is, the noise threshold was
estimated using the mean, median, and standard deviation of the absolute value of the
signal, with a cost ratio of zero. Again, a cost ratio of zero indicates that the cost of a false
positive is equal to the cost of an omission. When the signal exceeded the noise
threshold, an event was marked. Events occurring within a typical event period (1.75 ms)
were merged, while events occurring within a typical refractory period (4.5 ms) were
discarded. The remaining events were compared with the stimulus periods applied to
determine the arrival time of the action potentials relative to the onset of stimulus. Again,
these arrival times were sorted into histograms like the ones described in Chapter 5.

129

7.3

Results
The basic thresholding and CWT detection methods were applied to recordings from

silicon-based neural probes with iridium oxide electrodes, diamond-based neural probes
with boron-doped-diamond electrodes, and single-material diamond probes with borondoped-diamond electrodes, which have all been described previously. The silicon-based
probes had 16 electrodes, the diamond-based probes had 8 electrodes, and the singlematerial diamond probes had 6 electrodes. The first results are plots of the recorded
signals with markers (stems) indicating where the basic thresholding method detected
action potentials and where the CWT method detected action potentials. In each of the
plots, detections by basic thresholding are shown by blue stems above the recorded
signal, while detections by CWT are shown by red stems below the recorded signal. Figs.
7-1 and 7-2 show the recorded signals and the detections by thresholding and CWT for
each of the 16 electrode sites on the silicon-based neural probes with iridium oxide
electrodes for the entire recording length. In addition, Figs. 7-3 and 7-4 show expanded
views of 5 seconds of the recording time. Fig. 7-5 shows the recorded signals and the
detections by thresholding and CWT for the entire recording length of the diamond-based
neural probe with boron-doped-diamond electrodes. Fig. 7-6 shows a 5-second close-up
of the same recordings. Fig. 7-7 shows the recordings and detected action potentials of
the entire recording length from the single-material diamond probe with boron-dopeddiamond electrodes. Fig. 7-8 shows the expanded view for a 5-second window of these
recordings.

130

Figure 7-1: Neural recordings from the auditory cortex of a guinea pig taken with a silicon-based neural
probe (channels 1-8). The stimulus periods are highlighted in yellow. Action potentials detected using the
thresholding technique are shown with blue stems. Action potentials detected using the CWT method are
shown with red stems.

131

Figure 7-2: Neural recordings from the auditory cortex of a guinea pig taken with a silicon-based neural
probe (channels 9-16). The stimulus periods are highlighted in yellow. Action potentials detected using the
thresholding technique are shown with blue stems. Action potentials detected using the CWT method are
shown with red stems.

132

Figure 7-3: Expanded, 5-second view of neural recordings from the auditory cortex of a guinea pig taken
with a silicon-based neural probe (channels 1-8) from Fig. 7-1. The stimulus periods are highlighted in
yellow. Action potentials detected using the thresholding technique are shown with blue stems. Action
potentials detected using the CWT method are shown with red stems.

133

Figure 7-4: Expanded, 5-second view of neural recordings from the auditory cortex of a guinea pig taken
with a silicon-based neural probe (channels 9-16) from Fig. 7-2. The stimulus periods are highlighted in
yellow. Action potentials detected using the thresholding technique are shown with blue stems. Action
potentials detected using the CWT method are shown with red stems.

134

Figure 7-5: Neural recordings from the auditory cortex of a guinea pig taken with a diamond-based neural
probe (channels 1-8). The stimulus periods are highlighted in yellow. Action potentials detected using the
thresholding technique are shown with blue stems. Action potentials detected using the CWT method are
shown with red stems.

135

Figure 7-6: Expanded, 5-second view of neural recordings from the auditory cortex of a guinea pig taken
with a diamond-based neural probe (channels 1-8) from Fig. 7-5. The stimulus periods are highlighted in
yellow. Action potentials detected using the thresholding technique are shown with blue stems. Action
potentials detected using the CWT method are shown with red stems.

136

Figure 7-7: Neural recordings from the auditory cortex of a guinea pig taken with a single-material diamond
neural probe (channels 1-6). The stimulus periods are highlighted in yellow. Action potentials detected using
the thresholding technique are shown with blue stems. Action potentials detected using the CWT method
are shown with red stems.

137

Figure 7-8: Expanded, 5-second view of neural recordings from the auditory cortex of a guinea pig taken
with a single-material diamond neural probe (channels 1-6) from Fig. 7-7. The stimulus periods are
highlighted in yellow. Action potentials detected using the thresholding technique are shown with blue
stems. Action potentials detected using the CWT method are shown with red stems.

138

Several observations can be made from Figs. 7-1 through 7-8. It is evident from the
recordings that the SNR is different between probe types. The silicon-based probes have
the highest SNR, while the single-material diamond probes have the lowest SNR. It can
also be seen that several electrodes do not have any discernible signal on the siliconbased probe and the diamond-based probe. It can be seen that the number of detections
is higher using the CWT method for all of the probe types. Both the basic thresholding
and CWT methods appear to detect action potentials both during stimulus and outside of
stimulus. It is not apparent whether a higher percentage of detections occur within
stimulus for the thresholding or CWT method. In order to determine how each of the
methods affects the number of detections within stimulus and the arrival times relative to
the onset of stimulus, the histograms of arrival times are required. Figs. 7-9 and 7-10
show the histograms of spike arrival times relative to the onset of stimulus using the basic
thresholding method for each of the electrodes on the silicon-based neural probe. Fig. 711 shows the histogram of spike arrival times relative to onset of stimulus using the basic
thresholding method for the diamond-based neural probes. Fig. 7-12 shows the histogram
of action potential arrival times relative to the onset of stimulus using the basic
thresholding technique for the single-material diamond neural probes. Figs. 7-13 and 714 show the histograms of event arrival times relative to the onset of stimulus using the
CWT method for the silicon-based neural probe. Fig 7-15 shows the histogram of arrival
times relative to the onset of stimulus using the CWT method for the diamond-based
neural probe. Fig 7-16 shows the histogram of event arrival times relative to the onset of
stimulus using the CWT method for the single-material diamond neural probe.

139

Site 2
50

40

40

Spikes Detected

Spikes Detected

Site 1
50

30
20
10
0

0

100

200

300

400

Time After Onset of Stimulus [ms]

30
20
10
0

500

0

100

40

40

30
20
10

0

100

200

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

40

30
20
10

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

40

30
20
10

300

200

300

500

400

500

Time After Onset of Stimulus [ms]

Site 8

40

200

400

10

50

100

500

20

Site 7

0

300

30

50

0

200

Time After Onset of Stimulus [ms]

Site 6

40

200

400

10

50

100

500

20

Site 5

0

400

30

50

0

300

Site 4
50

Spikes Detected

Spikes Detected

Site 3
50

0

200

Time After Onset of Stimulus [ms]

400

Time After Onset of Stimulus [ms]

30
20
10
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 7-9: Histograms showing threshold-detected action potential times relative to the onset of stimulus
for a silicon-based neural probe with iridium oxide electrodes. Electrodes 1 – 8.

140

Site 10
50

40

40

Spikes Detected

Spikes Detected

Site 9
50

30
20
10
0

0

100

200

300

400

Time After Onset of Stimulus [ms]

30
20
10
0

500

0

100

40

40

30
20
10

0

100

200

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

40

30
20
10

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

40

30
20
10

300

200

300

500

400

500

Time After Onset of Stimulus [ms]

Site 16

40

200

400

10

50

100

500

20

Site 15

0

300

30

50

0

200

Time After Onset of Stimulus [ms]

Site 14

40

200

400

10

50

100

500

20

Site 13

0

400

30

50

0

300

Site 12
50

Spikes Detected

Spikes Detected

Site 11
50

0

200

Time After Onset of Stimulus [ms]

400

Time After Onset of Stimulus [ms]

30
20
10
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 7-10: Histograms showing threshold-detected action potential times relative to the onset of stimulus
for a silicon-based neural probe with iridium oxide electrodes. Electrodes 9 – 16.

141

Site 2
10

8

8

Spikes Detected

Spikes Detected

Site 1
10

6
4
2
0

0

100

200

300

400

Time After Onset of Stimulus [ms]

6
4
2
0

500

0

100

8

8

6
4
2

0

100

200

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

8

6
4
2

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

8

6
4
2

300

200

300

500

400

500

Time After Onset of Stimulus [ms]

Site 8

8

200

400

2

10

100

500

4

Site 7

0

300

6

10

0

200

Time After Onset of Stimulus [ms]

Site 6

8

200

400

2

10

100

500

4

Site 5

0

400

6

10

0

300

Site 4
10

Spikes Detected

Spikes Detected

Site 3
10

0

200

Time After Onset of Stimulus [ms]

400

Time After Onset of Stimulus [ms]

6
4
2
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 7-11: Histograms showing threshold-detected action potential times relative to the onset of stimulus
for a diamond-based neural probe with boron-doped-diamond electrodes.

142

Site 1

Site 2
20

Spikes Detected

Spikes Detected

20
15
10
5
0

0

100

200

300

400

Time After Onset of Stimulus [ms]

15
10
5
0

500

0

100

Site 3
Spikes Detected

Spikes Detected

10
5

0

100

200

300

400

500

Time After Onset of Stimulus [ms]

400

500

400

500

15
10
5
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Site 5

Site 6
20

Spikes Detected

20

Spikes Detected

400

20

15

15
10
5
0

300

Site 4

20

0

200

Time After Onset of Stimulus [ms]

0

100

200

300

400

Time After Onset of Stimulus [ms]

15
10
5
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 7-12: Histograms showing threshold-detected action potential times relative to the onset of stimulus
for a single-material neural probe with boron-doped-diamond electrodes.

143

Site 2
50

40

40

Spikes Detected

Spikes Detected

Site 1
50

30
20
10
0

0

100

200

300

400

Time After Onset of Stimulus [ms]

30
20
10
0

500

0

100

40

40

30
20
10

0

100

200

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

40

30
20
10

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

40

30
20
10

300

200

300

500

400

500

Time After Onset of Stimulus [ms]

Site 8

40

200

400

10

50

100

500

20

Site 7

0

300

30

50

0

200

Time After Onset of Stimulus [ms]

Site 6

40

200

400

10

50

100

500

20

Site 5

0

400

30

50

0

300

Site 4
50

Spikes Detected

Spikes Detected

Site 3
50

0

200

Time After Onset of Stimulus [ms]

400

Time After Onset of Stimulus [ms]

30
20
10
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 7-13: Histograms showing CWT-detected action potential times relative to the onset of stimulus for
a silicon-based neural probe with iridium oxide electrodes. Electrodes 1 – 8.

144

Site 10
50

40

40

Spikes Detected

Spikes Detected

Site 9
50

30
20
10
0

0

100

200

300

400

Time After Onset of Stimulus [ms]

30
20
10
0

500

0

100

40

40

30
20
10

0

100

200

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

40

30
20
10

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

40

30
20
10

300

200

300

500

400

500

Time After Onset of Stimulus [ms]

Site 16

40

200

400

10

50

100

500

20

Site 15

0

300

30

50

0

200

Time After Onset of Stimulus [ms]

Site 14

40

200

400

10

50

100

500

20

Site 13

0

400

30

50

0

300

Site 12
50

Spikes Detected

Spikes Detected

Site 11
50

0

200

Time After Onset of Stimulus [ms]

400

Time After Onset of Stimulus [ms]

30
20
10
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 7-14: Histograms showing CWT-detected action potential times relative to the onset of stimulus for
a silicon-based neural probe with iridium oxide electrodes. Electrodes 9 – 16.

145

Site 2
30

25

25

Spikes Detected

Spikes Detected

Site 1
30

20
15
10
5
0

0

100

200

300

400

Time After Onset of Stimulus [ms]

20
15
10
5
0

500

0

100

25

25

20
15
10
5
0

100

200

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

20
15
10
5
400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

200

300

500

Time After Onset of Stimulus [ms]

Site 8
25

20
15
10
5
300

400

5

25

200

500

10

30

100

400

15

Site 7

0

300

20

30

0

200

Time After Onset of Stimulus [ms]

Site 6
25

300

500

5

25

200

400

10

30

100

500

15

Site 5

0

400

20

30

0

300

Site 4
30

Spikes Detected

Spikes Detected

Site 3
30

0

200

Time After Onset of Stimulus [ms]

400

Time After Onset of Stimulus [ms]

20
15
10
5
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 7-15: Histograms showing CWT-detected action potential times relative to the onset of stimulus for
a diamond-based neural probe with boron-doped-diamond electrodes.

146

Site 1

Site 2
80

Spikes Detected

Spikes Detected

80
60
40
20
0

0

100

200

300

400

Time After Onset of Stimulus [ms]

60
40
20
0

500

0

100

Site 3
Spikes Detected

Spikes Detected

500

40
20

0

100

200

300

400

Time After Onset of Stimulus [ms]

400

500

400

500

60
40
20
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Site 5

Site 6
80

Spikes Detected

80

Spikes Detected

400

80

60

60
40
20
0

300

Site 4

80

0

200

Time After Onset of Stimulus [ms]

0

100

200

300

400

Time After Onset of Stimulus [ms]

60
40
20
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 7-16: Histograms showing CWT-detected action potential times relative to the onset of stimulus for
a single-material neural probe with boron-doped-diamond electrodes.

The expected result from this experiment is to see increased neural activity during
stimulus (0 – 200 ms), with a peak toward the beginning of the stimulus period. From Figs
7-9 and 7-10, it can be seen that using the basic thresholding technique we can strongly
see the expected result on electrodes 2, 4, 5, 7, 8, 10, 12, 13, and 15 of the silicon-based
neural probe. On electrodes 1, 9, and 16 we can see a trend showing a higher firing rate
147

near the onset of stimulus, but the results are not as strong as the other electrodes.
Electrodes 3, 6, 11, and 14 show no correlation to the expected result. From Figs. 7-13
and 7-14, it can be seen that the CWT method results in a strong correlation to the
expected result on electrodes 1, 2, 4, 5, 7, 8, 9, 10, 12, 13, 15, and 16. Electrodes 6, 11,
and 14 still show no correlation to the expected result, but electrode 3 bears a weak
correlation to the expectation. For the silicon-based probe, the CWT detection method
outperforms the basic thresholding method because several electrodes give results that
are closer to what is expected. This can be more clearly seen in Figs. 7-17 and 7-18 which
show combined histograms for the thresholding and CWT detection methods using the
silicon-based neural probe. In the histograms, the number of detections by thresholding
are shown by the blue (darker) bars, while the number of detection by CWT are shown
by the cyan (lighter) bars. Figs. 7-17 and 7-18 show that the CWT detection method not
only results in more action potentials being detected, but also in the results aligning more
with the expected results for the experiment. Even though the silicon-based probe
recordings do not suffer from poor SNR, they still appear to benefit from using the CWT
detection method.

148

Site 2
50

40

40

Spikes Detected

Spikes Detected

Site 1
50

30
20
10
0

0

100

200

300

400

Time After Onset of Stimulus [ms]

30
20
10
0

500

0

100

40

40

30
20
10

0

100

200

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

40

30
20
10

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

40

30
20
10

300

200

300

500

400

500

Time After Onset of Stimulus [ms]

Site 8

40

200

400

10

50

100

500

20

Site 7

0

300

30

50

0

200

Time After Onset of Stimulus [ms]

Site 6

40

200

400

10

50

100

500

20

Site 5

0

400

30

50

0

300

Site 4
50

Spikes Detected

Spikes Detected

Site 3
50

0

200

Time After Onset of Stimulus [ms]

400

Time After Onset of Stimulus [ms]

30
20
10
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 7-17: Histograms comparing detection using the thresholding method and the CWT method for a
silicon-based neural probe with iridium oxide electrodes. Electrodes 1 – 8. The number of detected action
potential times relative to the onset of stimulus is shown for thresholding in blue (darker) and for CWT in
cyan (lighter).

149

Site 10
50

40

40

Spikes Detected

Spikes Detected

Site 9
50

30
20
10
0

0

100

200

300

400

Time After Onset of Stimulus [ms]

30
20
10
0

500

0

100

40

40

30
20
10

0

100

200

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

40

30
20
10

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

40

30
20
10

300

200

300

500

400

500

Time After Onset of Stimulus [ms]

Site 16

40

200

400

10

50

100

500

20

Site 15

0

300

30

50

0

200

Time After Onset of Stimulus [ms]

Site 14

40

200

400

10

50

100

500

20

Site 13

0

400

30

50

0

300

Site 12
50

Spikes Detected

Spikes Detected

Site 11
50

0

200

Time After Onset of Stimulus [ms]

400

Time After Onset of Stimulus [ms]

30
20
10
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 7-18: Histograms comparing detection using the thresholding method and the CWT method for a
silicon-based neural probe with iridium oxide electrodes. Electrodes 9 – 16. The number of detected action
potential times relative to the onset of stimulus is shown for thresholding in blue (darker) and for CWT in
cyan (lighter).

150

Fig. 7-11 shows that using the basic thresholding technique we can see a correlation
to the expected result on electrodes 1 and 8 on the diamond-based neural probe. On
electrodes 2 and 7, there may be a correlation, but not enough action potentials were
detected. Electrodes 3, 4, 5, and 6 show no action potentials being detected. From Fig.
7-15, it can be seen that the CWT method results in a strong correlation to the expected
result on electrodes 1, 2, and 8. Electrodes 3, 4, 5, and 6 still show no correlation to the
expected result, but electrode 7 shows some correlation to the expectation. For the
diamond-based probe, the CWT detection method outperforms the basic thresholding
method because the detection rate is much higher and results that are closer to
expectations. This can be clearly seen in Fig. 7-19, which shows the combined
histograms for thresholding and CWT detection methods using the diamond-based neural
probe. In the histograms, the number of detections by thresholding are shown by the blue
(darker) bars, while the number of detection by CWT are shown by the cyan (lighter) bars.
Fig. 7-19 shows that the CWT detection method results in more than twice as many action
potentials being detected, and also the results more closely reflect the expected results
for the experiment. As expected, the poor SNR of the diamond-based neural probe is
significantly affected by the CWT detection method.

151

Site 2
30

25

25

Spikes Detected

Spikes Detected

Site 1
30

20
15
10
5
0

0

100

200

300

400

Time After Onset of Stimulus [ms]

20
15
10
5
0

500

0

100

25

25

20
15
10
5
0

100

200

300

400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

20
15
10
5
400

Time After Onset of Stimulus [ms]

0

500

0

100

Spikes Detected

Spikes Detected

200

300

500

Time After Onset of Stimulus [ms]

Site 8
25

20
15
10
5
300

400

5

25

200

500

10

30

100

400

15

Site 7

0

300

20

30

0

200

Time After Onset of Stimulus [ms]

Site 6
25

300

500

5

25

200

400

10

30

100

500

15

Site 5

0

400

20

30

0

300

Site 4
30

Spikes Detected

Spikes Detected

Site 3
30

0

200

Time After Onset of Stimulus [ms]

400

Time After Onset of Stimulus [ms]

20
15
10
5
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 7-19: Histograms comparing detection using the thresholding method and the CWT method for a
diamond-based neural probe with boron-doped-diamond electrodes. The number of detected action
potential times relative to the onset of stimulus is shown for thresholding in blue (darker) and for CWT in
cyan (lighter).

152

Fig. 7-12 shows the results from the single-material diamond neural probe using the
basic thresholding technique. It can be seen that all of the electrodes have comparable
performance. As expected there is a sharp rise in neural activity shortly after the onset of
stimulus. However, after the initial increase in activity, there is no increase in activity
during the remainder of the stimulus, as expected. It is not clear why there is a lack of
activity during the remainder of the stimulus. From Fig. 7-16, it can be seen that the CWT
method results in many more detected action potentials than the thresholding method.
However, there is still an inexplicable lack of increased activity during the stimulus other
than immediately after the onset of stimulus. For the diamond-based probe, the CWT
detection method outperforms the basic thresholding method because the detection rate
is much higher. This can be clearly seen in Fig. 7-20, which shows the combined
histograms for thresholding and CWT detection methods using the single-material
diamond neural probe. In the histograms, the number of detections by thresholding are
shown by the blue (darker) bars, while the number of detection by CWT are shown by the
cyan (lighter) bars. Fig. 7-20 shows that the CWT detection method results in more than
five times as many action potentials being detected. As expected, the very poor SNR of
the single-material diamond neural probe is significantly affected by the CWT detection
method.

153

Site 1

Site 2
80

Spikes Detected

Spikes Detected

80
60
40
20
0

0

100

200

300

400

Time After Onset of Stimulus [ms]

60
40
20
0

500

0

100

Site 3
Spikes Detected

Spikes Detected

500

40
20

0

100

200

300

400

Time After Onset of Stimulus [ms]

400

500

400

500

60
40
20
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Site 5

Site 6
80

Spikes Detected

80

Spikes Detected

400

80

60

60
40
20
0

300

Site 4

80

0

200

Time After Onset of Stimulus [ms]

0

100

200

300

400

Time After Onset of Stimulus [ms]

60
40
20
0

500

0

100

200

300

Time After Onset of Stimulus [ms]

Figure 7-20: Histograms comparing detection using the thresholding method and the CWT method for a
single-material diamond neural probe with boron-doped-diamond electrodes. The number of detected
action potential times relative to the onset of stimulus is shown for thresholding in blue (darker) and for
CWT in cyan (lighter).

The CWT detection method provides benefits to all three of the probe types that were
used in this experiment. For the silicon-based neural probe, the CWT method allows more
action potentials to be detected and improves the results of several electrodes that had
lower SNR. For the diamond-based neural probe, the CWT more than doubled the
154

number of action potentials detected and doubled the number of viable electrodes. For
the single-material diamond neural probe, the CWT method increased the number of
detected action potentials by more than five-fold. In order to directly compare the benefits
to using the CWT method over the basic thresholding method for the three probe types,
the results were averaged for all of the good electrodes on each probe. Electrodes which
showed little or no correlation to the stimulus were ignored (electrodes 3, 4, 5, and 6 on
the diamond-based probe and electrodes 3, 6, 11, and 14 on the silicon-based probe). In
addition, rather than comparing the number of action potentials detected, the rate at which
action potentials was detected was compared. This makes a direct comparison much
easier as the length of the recordings for each of the probe types was significantly
different. The results are presented in Fig. 7-21 showing the average firing rate of
detected action potentials across all of the electrodes for each of the probe types. The
detected firing rate using the thresholding method is in blue (darker), while the detected
firing rate using the CWT method is in cyan (lighter). The top graph shows the results for
the silicon-based neural probe with iridium oxide electrodes, the middle shows the results
for the diamond-based neural probe with boron-doped-diamond electrodes, and the
bottom graph shows the results for the single-material diamond neural probe with borondoped-diamond electrodes. From Fig. 7-21 it can be seen that using the CWT method
instead of the basic thresholding technique results in ~65% more action potentials being
detected, with ~50% more being detected during stimulus, for the silicon-based neural
probe. For the diamond-based neural probe, ~410% more action potentials are detected,
with ~345% more being detected within stimulus, using the CWT technique instead of the
thresholding technique. Using the CWT method, rather than thresholding, ~875% more

155

action potentials are detected, with ~700% more detected during stimulus. As expected,
using the CWT method instead of the basic thresholding method dramatically improves
the detection rate of action potentials, especially when analyzing recordings from the
diamond-based and single-material diamond neural probes.

156

Silicon-based Neural Probe

Average Firing Rate [Hz]

30
25
20
15
10
5
0
0

50

100

150

200

250

300

350

400

450

500

350

400

450

500

350

400

450

500

Time After Onset of Stimulus [ms]
Diamond-based Neural Probe

35

Average Firing Rate [Hz]

30
25
20
15
10
5
0
0

50

100

150

200

250

300

Time After Onset of Stimulus [ms]
Single-material Diamond Neural Probe

35

Average Firing Rate [Hz]

30
25
20
15
10
5
0
0

50

100

150

200

250

300

Time After Onset of Stimulus [ms]

Figure 7-21: Bar graphs comparing average detection rates across all good electrodes using the
thresholding method and the CWT method for each probe type. The number of detected action potential
times relative to the onset of stimulus is shown for thresholding in blue (darker) and for CWT in cyan
(lighter).

157

The silicon-based neural probe with iridium oxide electrodes is considered sufficient
for acute neural recording activities. In order for the diamond-based and single-material
diamond neural probes to be considered effective for acute neural recordings, their
performance must be comparable to that of the silicon-based probe. The performance will
be measured using three variables: the peak average-detection-rate shortly after the
onset of stimulus, the average detection rate over the whole stimulus period, and the
average detection rate over the entire recording period. As the sites that the recordings
were taken are different, it is not expected that the results will be exactly the same.
However, the results should be comparable, e.g. within 30% of each other. These results
have been calculated for both the basic thresholding technique and the CWT method and
can be found in Table 7-1.
Table 7-1: Comparison of average detection rates for each probe type using the basic threshold detection
method and the CWT detection method.

Probe Type
Silicon-based
Neural Probe
Diamond-based
Neural Probe
Single-material
Diamond
Neural Probe

Peak
AverageDetection detectionMethod
rate
Threshold 21.9 Hz
CWT
29.3 Hz
Threshold 12.1 Hz
CWT
32.4 Hz

Average Detection Average
Rate
during Detection Rate
Stimulus
for Recording
10.5 Hz
5.8 Hz
15.7 Hz
9.5 Hz
4.1 Hz
1.8 Hz
18.3 Hz
9.4 Hz

Threshold 4.7 Hz
CWT
32.6 Hz

0.8 Hz
6.1 Hz

0.3 Hz
3.3 Hz

From Table 7-1, it can be seen that when using the basic thresholding method for
detecting action potentials, the diamond-based and single-material diamond neural
probes far underperform the silicon-based neural probe due to their poor SNR. The
detection rates for the diamond-based neural probe are roughly half of the detection rates
158

of the silicon-based neural probe when using the basic thresholding technique. The
single-material diamond probe performs even worse, having well under a quarter of the
detection rates of the silicon-based neural probe when using the basic thresholding
method. However, when using the CWT method to detect action potentials, the results
are much different. The silicon-based neural probes have a higher detection rates, as
previously noted. The diamond based neural probe has comparable (and even higher)
detection rates to the silicon-based neural probe. This is evidence that the diamondbased neural probe can be used in acute neural recording experiments as its performance
is comparable to the accepted silicon-based neural probe. The single-material diamond
neural probe has a comparable peak average-detection-rate (~10% higher) to the same
detection rate from the silicon-based probe when using the CWT method. But, the
average detection rate during stimulus and the average detection rate for the entire
recording are ~65% lower than the rates for silicon-based neural probe. While the results
from the single-material diamond neural probe show a strong correlation to the onset of
stimulus, they do not show the increased neural activity during the full extent of the
stimulus. Therefore, it is inconclusive whether the single-material diamond neural probes
are suitable for use in acute neural recording applications.

159

7.4

Conclusion
In Chapter 5, it was found that the poor SNR of the diamond-based and single-

material diamond neural probes made them non-ideal for neural recording applications
when using a basic thresholding technique. In Chapter 6, different action potential
detection methods were explored with poor SNR recordings in mind. It was found that the
CWT method provided the best performance in simulated, low-SNR applications. To
determine the efficacy of the CWT method on real recordings, the method was applied to
signals recorded from the audio cortex of a guinea pig taken from thee different probe
types: the silicon-based neural probe with iridium oxide electrodes, the diamond-based
neural probe with boron-doped-diamond electrodes, and the single-material diamond
neural probe with boron-doped diamond electrodes. In addition, the thresholding method
was reapplied using the same automated noise estimation technique used in the CWT
method. Because of this, it was possible to directly compare the basic thresholding
detection and the CWT detection methods for each of the probe types. It was found that
the CWT method enhanced the results from all three probe types, significantly improving
the detection rates for the diamond-based (410% higher) and single-material diamond
(875% higher) probes. In addition to higher detection rates, the results from the CWT
detection method more closely resembled the expected results from the experiment for
several electrodes on both the silicon-based and diamond-based neural probes. Finally,
it was demonstrated that the diamond-based neural probes are suitable for use in acute
neural recording experiments when using the CWT detection method because the
detection rates for the diamond-based and silicon-based probes were nearly identical for
both probe types. The efficacy of the single-material diamond neural probes in acute
160

neural recording experiments is still in question, as the peak detection rate shortly after
the onset of stimulus was comparable to the silicon-based and diamond-based probes,
but the detection rate through the rest of the stimulus and after the stimulus period was
negligible.

161

CHAPTER 8:
Summary and Recommendations

8.1

Summary of Contributions
The development and testing of a single-material diamond neural probe, the first of

its kind, has been presented. Polycrystalline diamond micromachining technologies
required for this aim have been explored and described. The technologies that have been
reported include diamond seed nucleation, diamond film deposition, increasing the
resistivity of undoped diamond, and diamond film patterning.
Diamond-based neural probes and single-material diamond neural probe have both
been successfully fabricated. The single-material diamond neural probe is the first neural
probe to ever be fabricated from a single material, that is its structural material, electrode
material, interconnect material, and insulating material are all polycrystalline diamond. It
is also the first functional single-material MEMS device ever reported. Detailed fabrication
processes have been described and documented for both the diamond-based neural
probes and the single-material diamond neural probes.
Single-material diamond probes were briefly demonstrated being used for
electrochemical detection. This shows that diamond-based and single-material diamond
probe technology could be developed for electrochemical applications. Silicon-based,
diamond-based, and single-material diamond neural probes were all surgically implanted
into a live guinea pig’s audio cortex and electrical neural activity was recorded on each
probe type. This was the first time that a single-material MEMS device was ever surgically
162

implanted into an animal. Using a basic thesholding technique, the electrical recording
results were analyzed. While there was a correlation between the stimulus and the results
for the diamond-based and single-material diamond probes, their performance was not
as good as the silicon-based probe due to their poor signal-to-noise ratio.
Advanced action potential detection techniques were explored in order to identify
which technique(s) would be best suited for detecting action potentials from recordings
with poor signal-to-noise ratios. It was found that wavelet techniques had the best
performance with the continuous wavelet transform technique slightly outperforming the
discrete wavelet technique. The recordings from each of the probe types was then
analyzed using the continuous wavelet transform detection technique. It was found that
using this technique, significantly more action potentials could be detected from the
diamond-based and single-material diamond neural probes. In addition, the diamondbased neural probe had detection rates that were as good as the widely accepted siliconbased neural probe. However, the results from the single-material diamond probe were
inconclusive as its detection rate of action potentials shortly after the onset of stimulus
was comparable to the silicon-based probe, the detection rate during the remainder of the
stimulus was significantly lower than the silicon-based and diamond-based neural probes.
Further study is needed to determine the efficacy of single-material diamond neural
probes.

163

8.2

Future Research Areas
In order to develop diamond-based or single-material diamond neural probes for

commercialization, several critical studies must be performed.
1. Improvements must be made in probe performance. The conductivity of doped
diamond is too low for the single-material diamond probe design. This poor
conductivity is likely responsible for the poor SNR in electrical recording of neurons
and presents complications for electrochemical applications. Modifying the probe
design to have shorter interconnects or larger interconnect cross-sectional area
may help improve this issue. However, material development would be a more
ideal solution.
2. Chronic experiments must be performed. All of the in vivo experiments performed
to date have been acute implantations. The long term effects of surgically
implanting a diamond neural probe into living brain tissue must be explored. It is
anticipated that the body’s response to the diamond probe will be more positive
than the silicon probe due to the high biocompatibility of diamond. However, no
experiments have been performed to determine the long-term effects of diamond
probe implantation or the viability of boron-doped-diamond electrodes in chronic
electrical recording applications.
3. Further in vivo recording experiments are required. While the diamond-based
neural probe has been demonstrated to have comparable recording performance
to silicon-based probes, more tests are needed to confirm this in other vectors
besides guinea pigs. The single-material diamond probe has not been shown to
have comparable performance to silicon-based probes, yet. Further study is
164

needed, most likely of the next generation of single-material probes, to determine
whether the recording performance is suitable for neuroscience experiments.

165

CHAPTER 9:
Education: Woodcreek Elementary Outreach Program

9.1

Introduction
Capable engineers are becoming more crucial to the future of this society. In order

to prepare students to enter the work world with the necessary math, science and
problem-solving skills, it is important to get them interested in math, science and
engineering concepts at a young age. Traditionally, science, technology, engineering and
math (STEM) topics have been taught using well-defined problems, involving top-down
approaches where the decisions on teaching methods are made by educational experts
and organizations. In the real world, however, problems are not well defined, and learning
methods vary greatly among students. Consequently, a restructuring of school science
around real-world problems has been suggested by educational experts and
organizations [1, 2], which has led to a number of studies focusing on inquiry [3-6].
Notable examples of inquiry-based studies are Design-Based Science (DBS) [7-16] and
Learning By Design (LBD) [17-19]. Recently, new learning techniques have been
developed including the use of technology to increase student interest. The TechnologyAssisted Science, Engineering and Mathematics (TASEM) program represents such
techniques and has shown efficacy in instilling interest in the field of engineering, while
also introducing basic and advanced concepts in STEM areas [20]. The TASEM program
consists of:

166

• STEM topics being explained using Lego® robotics
• Students directly interacting with university students and faculty
• Students designing and building their own STEM projects
• Students developing presentation skills through project presentations

The current TASEM program does not have a large enough footprint. In the last few
years, the hands-on TASEM modules have been modified and applied with Woodcreek
Elementary School's third grade classes to establish that TASEM can serve a more
diverse audience. This paper presents the details of the Woodcreek TASEM program,
along with statistics supporting the effectiveness of the program.

167

9.2

TASEM Summer Camp
The TASEM program has mainly been offered through summer camps at Michigan

State University. The summer program consists of four two-hour sessions over
consecutive days. The program is very flexible, as it is adaptable to any age of student.
It has been used on students as young as kindergarteners, and as old as schoolteachers;
the TASEM program has been applied through workshops for science teachers in
Michigan (~60 teachers) and New York (~90 teachers). The TASEM program has shown
that it is capable of overcoming geographic limitations by offering summer sessions
remotely. Remote sessions were held in the Oakland, Michigan area in 2005-2006 and
at the University of South Florida in 2006. As the program is not fully funded by grants,
the number of students influenced is limited due to the inherent cost to the students. This
prevents many students coming from lower socioeconomic demographics from
participating in the summer courses. This issue was addressed in 2004 by modifying the
program to be used in Ovid-Elsie, a rural area to the north of Lansing, Michigan. It was
demonstrated that a modified version of the program could be offered at a lower cost to
the students.
The TASEM summer program was designed to raise student interest in STEM areas.
The program uses Legos®, something the students already understand and enjoy, in
order to interest students and make it easier for them to relate to more advanced topics.
The level of complexity of the summer program varies by student as the program is
designed to fit the needs of individual students. There are several categories of STEM
topics that students can choose to focus on, such as nanotechnology, static charges,
robotics or microcontroller programming [21, 22]. At the beginning of the program,
168

students are placed into small groups depending on their topics of interest and ability
level. Different groups can have very different experiences as they explore their subject
of interest under the guidance of instructors, who are mainly Ph.D. students and
undergraduate students supervised by an engineering professor. The expertise and
knowledge offered by these instructors sets this program apart from similar programs
across the nation. Throughout the sessions, the students will progress through activities
specific to their topic of interest. In the last two sessions, students typically prepare a
project demonstrating what they have learned about their subject. The students will then
present their projects at the end of the program to show their grasp of the material.
The main goal of the TASEM program is to get learners of all ages excited about
STEM topics. Table 9-1 shows a summary of the statistics relating to the TASEM summer
camp program over four previous years. The table shows that the program has had a
significant impact on many students. On average, more than half of the students
participating in the summer camp program participate for more than one year.
Furthermore, nearly 10% of the students involved enroll in more than one session each
summer. These statistics show that the students are excited by, and interested in, the
camps. On average, 70% of the students participating in the summer camps are in
elementary school. This means that the program is reaching younger students and getting
them interested in STEM subjects at an early age. The number of students enrolled each
summer has consistently been above 100 students. It is worth noting that 2006 had a
record number of students due to additional advertisement in local newspapers,
supplementing the normal advertisement in local schools. Furthermore, starting in 2007,

169

the number of students per group was decreased from three to two in order to improve
the individual student’s learning experience.
Table 9-1: TASEM summer camp statistics.

Year

Number of
Participants

% in
Elementary
School

% in
Middle
School

% in High
School

% of Students
Returning from
Previous Years

% of Students Taking
More Than One Session
Per Year

2005

110

70.9%

26.4%

2.7%

47.3%

7.3%

2006

174

69.5%

26.4%

4.0%

57.5%

5.7%

2007

130

69.2%

23.8%

6.9%

57.7%

15.4%

2008

101

70.3%

27.7%

2.0%

41.6%

7.9%

Average

128.75

70.0%

26.1%

3.9%

51.0%

9.1%

170

9.3

Woodcreek Elementary School-Based TASEM Program
In order to broaden the impact of TASEM, the summer program has been modified

from a one-week summer program to an in-school program. This new program has been
offered to third graders at Woodcreek Elementary School in Lansing, Michigan. The
Lansing school district is comprised mostly of low-income families and has shown below
average standardized test scores in the fields of science and math. The student body at
the school is extremely diverse. Approximately 80% of the students are African American,
while 6% are white, 6% are Asian and 4% are Hispanic, with the remaining 4% belonging
to other ethnic groups.

Figure 9-1: Student team working on programming their robot.

The program uses the Lego® Mindstorms system and Lego® bricks and gears to
explain mathematical concepts such as multiplication, ratios and fractions, and
171

scientific/engineering principles including electricity, force, magnetism and simplified
concepts in electronics. The program has been integrated into the school curriculum and
consists of weekly one-hour sessions held during regular school hours throughout the
school year. These sessions are conducted by a Ph.D. student, another researcher
(typically a post-doctorate or an undergraduate student) and the science specialist at the
school, under the guidance of a professor of engineering from Michigan State University.
The program has been offered each year since 2003 to two third-grade classes with about
25 students per class. Students were divided into groups of two or three and each group
was assigned a laptop for use in programming Lego® Mindstorms robots, as seen in Fig.
9-1. Low cost modules consisting of Lego® gears, beams and axles were used to
introduce the concept of ratios and fractions as well as to develop the students’ logical
skills by experimenting with multiple gears connected in various configurations. Among
the scientific concepts introduced using these modules were the concepts of rotation,
speed, energy and energy conversion. The students could relate these concepts to other
principles they learn in their science classes. Using specially designed instruction sheets
along with the Mindstorms computer-aided tutorials, students learn how to build and
program the robots. Different students achieve different levels of proficiency in
programming depending on their level of interest. However, most students are able to
learn and use basic programming concepts using the Mindstorms program. The students
then apply these concepts to design, build and program their own robot, as seen in Fig.
9-2. At the end of the program, the students make presentations (Fig. 9-3) to the entire
school and their parents, on the function of their robots. These presentations demonstrate
the students’ understanding of the concepts taught throughout the program.

172

Figure 9-2: Students demonstrating the robot they designed, built and programmed.

The Woodcreek sessions differ from the summer TASEM sessions in several
important ways. The in-school program is more narrowly structured than the summer
program, with a focus on gradually building the students’ problem-solving skills,
teamwork, and mathematical skills. This is possible because of the extended duration of
the in-school program, which allows students to spend more time on individual tasks and
topics. With a total instruction time of about 35 hours, the school program gives students
the opportunity to explore many STEM areas, and provides an improved learning
experience overall. Another key difference is that much of the learning is done through
planned lectures and activities that emphasize elementary-level math and science,
instead of allowing students to independently select topics of interest, as in the summer
program. This ensures that all of the students have the same learning opportunities and
stay on track. A third important difference between the two programs concerns the

173

makeup of the participating group of students. Presumably, only students who want to
learn about robotics and other STEM topics enroll in the summer program. The summer
program also requires students to pay for enrollment. Because of this, only students who
actively seek the program and whose parents are willing to pay the enrollment fee may
participate. In contrast, in-school program is offered to every third grader, with no financial
burden on the students. This broadens the reach of the program and leads to the inclusion
of students who may not yet be interested in STEM topics. This allows the TASEM
program to reach students who might not otherwise become interested in STEM learning
and can be expected to have a larger impact than the summer program in that respect.

Figure 9-3: Student presenting his project in front of other students and parents.

The in-school TASEM program has several unique aspects that set it apart from many
other STEM learning activities. The most important aspect of this program is that it
primarily uses graduate student researchers as instructors and gives the elementary
174

students an opportunity to directly interact with a university professor via
videoconferencing, Fig. 9-4. Interaction with university students and faculty allows for a
level of learning beyond that found in a normal elementary school classroom. Another
major difference between this and other programs is that this program is offered entirely
during regular school hours, rather than through after-school sessions. This makes the
sessions mandatory for all third graders. This is beneficial as it ensures a maximum
impact, as it includes students who may otherwise be unable or unwilling to attend afterschool sessions. Another aspect that makes this program different than other STEM
learning programs is the focus on student presentations. At the end of the program, there
is a school-wide presentation where third graders present their projects to other students.
At these presentations there are also parents, educators, administrators and media
present. This allows the students to demonstrate gained knowledge to others, and
increases other students’ interest in STEM areas. The unique attributes of the Woodcreek
Elementary TASEM program result in an excellent STEM learning environment.

175

Figure 9-4: Students teleconferencing with Professor Dean Aslam, from Michigan State University, who is
presenting a demonstration.

176

9.4

Results
The Woodcreek Elementary TASEM program has been very successful over the

seven years since 2003. This is apparent when considering the anecdotal evidence
collected from teachers at the school. The magnet school’s science and engineering
specialist reports that the program has been, and continues to be, the largest selling point
of Woodcreek Elementary School because students, and their parents, are so excited
about the program. She goes on to explain that the program is an effective tool in
students’ character development as it encourages and trains them to use “team skills” as
they develop their robots. She reports that 75% of the student body actively ask questions
and tell anecdotes about the program outside of the allotted class time. A fifth grade
teacher has noticed a difference between students who have gone through the program,
and students who have not. As this particular school has a large turnover rate, only about
half of the current fifth grade class participated in the third grade TASEM program. She
says that students who have gone through the program exhibit a better sense of
teamwork, and are much more focused and intelligently engaged in STEM discussions.
The half of the class who went through the program has an A-/B+ grade average in math
and science, and a good average in their citizenship grade. The other half who did not
participate in the TASEM program has a B-/C+ average in math and science and average
or poor grades in citizenship. A third grade teacher has noticed that students pay more
attention and show a deeper understanding of certain math and science concepts when
taught through the TASEM program. She also estimates that 95% of her third graders are
greatly enthused by the program. In addition to strong anecdotal evidence, standardized
test scores indicate that the TASEM program has had a positive impact on the students.
177

For example, scores from the fourth grade Michigan Educational Assessment Program
(MEAP) [23] over three consecutive years before the TASEM program was introduced
(2002-2004), show an average of 39.8% of students performing satisfactorily in math,
while 11.4% of students scored in the lowest scoring bracket. Over the past three years
(2007-2009), on average 91.2% of students were scored as performing adequately in
mathematics, and no students scored in the lowest scoring bracket. Overall, the TASEM
program has had a positive impact on students at Woodcreek Elementary.

Veterinarian
Pediatric

Nurse

Nurse

17%

8%
NBA Player

17%

8%
Non-STEM;
17%

Computer

Lawyer

Engineer
17%

8%

Doctor
25%

Figure 9-5: Preferred career of polled TASEM program graduates.

To measure the long-term efficacy of the Woodcreek TASEM program, a smallsample study was conducted at Dwight Rich Middle School in the Lansing School District.
Unfortunately, due to the volatile nature of the school district, only twelve sixth- and
seventh-grade Woodcreek graduates were available for feedback on the program. These
twelve were given a questionnaire comprised of four questions by their current teachers,
178

asking what profession they wanted to practice when they grew up. As seen in Fig. 9-5,
83% of the students chose professions in STEM fields. The students were also asked
what they remembered most about third grade. Out of the students, 67% indicated that
the TASEM program was what they remembered most, with 25% indicating that they
remembered their teacher most, and one student indicated that math was what they
remembered most. The students were presented a figure with different sized gears on it
connected in sequence, with an indication that the first gear was turning clockwise. They
were asked to indicate which direction each gear would turn and which gear would turn
the fastest. Three quarters of the students were able to answer this question correctly,
which is significant because gears are not taught in their middle school curriculum. The
final question asked the students to rate the TASEM program on a scale of one to five.
The results can be seen in Fig. 9-6 with the average being between good and excellent.
Due to the small sample size (12 out of a population of ~100), and lack of control study,
it is impossible to infer statistically significant conclusions from this data. However, these
statistics do show that the program was successful as the graduated students: (a) are
interested in STEM topics after going through the program, (b) vividly remember the
TASEM program four years later, (c) are able to solve a problem based on material
learned from the TASEM program and (d) highly rate the program itself.

179

Figure 9-6: Histogram of ratings of the Woodcreek TASEM program by Woodcreek graduates.

180

9.5

Conclusion
In this paper, the Woodcreek Magnet Elementary School TASEM program has been

presented. This program has shown that it is capable of reaching students of different
socioeconomic and cultural backgrounds. The program has a high potential for impact on
students. The program can easily be expanded to an entire school district and to other
school districts. The key components to introducing this program in other schools are
computers, Lego Mindstorm kits, and dedicated instructors. Instructors could be trained
to teach the program anywhere in the country, and teleconferencing can be used to show
demonstrations, deliver supplemental lectures and provide student interaction with
university faculty. Also, the TASEM program can be modified to suit different purposes
for different schools or grade level. The program's effectiveness has been shown through
teacher testimonials, as well as indications from previous students. Data collected from
program graduates indicate that they are still interested in STEM topics, they remember
the program well, they remember material presented through the program and they retain
a high opinion of the program. Finally, a method for enacting this program remotely has
been suggested.

181

REFERENCES

182

REFERENCES

[1] M. Lipman, Thinking in Education, New York, NY: Cambridge University Press, 1991.
[2] P.C. Blumenfeld, R.W. Marx, H. Patrick, J.S. Krajcik, and E. Soloway, “Teaching for
understanding,” in International handbook of teachers and teaching, B. J. Biddle, T.
L. Good & I. F. Goodson, Eds. The Netherlands: Kluwer Academic Publishers, pp.
819-878, 1997.
[3] Cognition & Technology Group at Vanderbilt, “The jasper series as an example of
anchored instruction: Theory, program description, and assessment data”,
Educational Psychologist, vol. 27, no. 3, pp. 291-315, 1992.
[4] J.S. Krajcik, P.C. Blumenfeld, R.W. Marx, K.M. Bass, J. Fredricks, and E. Soloway,
“Inquiry in project based science classrooms: Initial attempts by middle school
students,” The Journal of the Learning Sciences, vol. 7, no. 3-4, pp. 313-350, 1998.
[5] D.E. Penner, R. Lehrer, and L. Schauble, “From physical models to biomechanics: A
designbased modeling approach,” The Journal of the Learning Sciences, vol. 7, no.
3-4, pp. 429-449, 1998.
[6] N.B. Songer, “Exploring learning opportunities in coordinated networkenhanced
classrooms: A case of kids as global scientists,” Journal of the Learning Sciences,
vol. 5, pp. 297–328, 1996.
[7] D. Fortus, R.C. Dershimer, J. Krajcik, R.W. Marx and R. Mamlok-Naaman, “Design
based science and student learning,” J. Research Sci. Teaching, vol. 41, pp. 10811110, 2004.
[8] P. Roberts, “The place of design in technology education,” in Innovations in science
and technology education, D. Layton Ed. New York, NY: UNESCO, pp. 27–38, 1995.
[9] K. Baynes, Designerly Play, Loughborough, UK: Loughborough University of
Technology, 1994.
[10] H.A. Simon, The Sciences of the Artificial, Cambridge, MA: MIT Press, 1999.
[11] L.L. Bucciarelli, Designing Engineers, Cambridge, MA: MIT Press, 1994.
[12] M. Davis, P. Hawley, B. McMullan, and G. Spilka, Design as a Catalyst for Learning,
Alexandria, VA: Association for Supervision and Curriculum Development, 1997.
[13] International Technology Education Association, Standards for Technological
Literacy, Reston, VA: International Technology Education Association, 2002.

183

[14] National Research Center, National Science Education Standards, Washington, DC:
National Academy Press, 1996.
[15] National Research Center, Scientific Research in Education, Washington, DC:
National Academy Press, 2002.
[16] E.L. Chiapetta, T.R. Koballa Jr., and A.T. Collette, Science Instruction in the Middle
and Secondary Schools, 5th ed. Upper Saddle River, NJ: Prentice-Hall, 2002.
[17] J.L. Kolodner, Case-based Reasoning, San Mateo, CA: Morgan Kaufmann, 1993.
[18] J.L. Kolodner, D. Crismond, J. Gray, J.K. Holbrook, and S. Puntambekar, “Learning
by design from theory to practice,” Int. Conf. of the Learning Sciences, Atlanta, GA,
pp. 16-22, 1998.
[19] H.S. Barrows, How to Design a Problem-based Curriculum for the Preclinical Years,
New York, NY: Springer, 1985.
[20] D.M. Aslam, “Technology assisted science, engineering and mathematics (TASEM)
education at all levels using K-Ph.D. concept,” Proc. ASEE Annual Conf., Chicago,
IL, p. 92, 2006.
[21] D.M. Aslam and A. Shao, "Nanotechnology learning modules using technology
assisted science, engineering and mathematics,” Proc. ASEE Zone 1 Conf., West
Point, NY, 2008.
[22] D.M. Aslam, Z. Cao, and C. Rostamzadeh, "Innovative engineering education using
programmable Lego robotic VD Graaf generators,” Proc. ASEE Zone 1 Conf., West
Point, NY, 2008.
[23] State of Michigan. (2011). MDE – Michigan Educational Assessment Program
(MEAP)
[Online].
Available:
http://www.michigan.gov/mde/0,1607,7-14022709_31168---,00.html

184