SENSING AND ACTUATION OF BEAD-TAGGED BIOMATERIALS ON STANDARD
CMOS SUBSTRATES
By
Faisal T. Abu-Nimeh

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Electrical Engineering
2011

ABSTRACT
SENSING AND ACTUATION OF BEAD-TAGGED BIOMATERIALS ON
STANDARD CMOS SUBSTRATES
By
Faisal T. Abu-Nimeh
Magnetic molecular-level sensing and manipulation are emerging as lab-on-chip platforms.
These platforms entail low-cost, low-power, high efficiency, and portable implementations.
Biomaterials are usually attached to magnetic beads and used in bio-analysis applications
such as sorting, counting, purification, and assembly. Some of the potential applications are
2D biological or artificial tissue assembly at the micro-scale level.
Here, we present the design and demonstration of a self-contained device for sensing and
manipulating biomaterials tagged with magnetic beads. The core elements of the device
consist of all-integrated programmable magnetic coil arrays for pseudo-parallel sensing and
actuation, which are capable of maneuvering small (bead-bound) bio-objects individually
and larger ones collaboratively. Our design does not require any external magnetic sources.
It relies on the magnetic field generated by planar on-chip coil arrays. The coil arrays are
selectively and dynamically controlled. Each element, composed of the coil and its logical
control circuitry, can detect bio-objects in the order of 1µm diameter, or manipulate them
using eight-level programmable AC or DC magnetic fields. All array sensing and actuation
components are shared and multiplexed to reduce the overall imprint. The components are
isolated and tuned to work at 900MHz by incorporating high-speed switching (up to 40MHz)
for seamless pseudo-parallel execution.
In addition, we present a new and unique on-chip biomedical proof-of-concept application. Adopting trends in neuroscience, we employ the magnetic beads to initiate and elongate

iii
neuronal axons in vitro on-chip. This application domain will assist in better understanding and studying the process of neuroregeneration, propel future clinical applications, and
provide the foundation of techniques needed to wire “neuronal circuits” using living cells.

To my family.

iv

ACKNOWLEDGMENTS

I would like to express my sincere gratitude to my parents, my wife Zaina, and my family
who have provided me with everything I needed and have been at my side throughout this
endeavor. I am deeply thankful for their support and love, without that I certainly would
not have been able to accomplish and finish this work. And thanks to my daughter Raneem
who showed me a new perspective on life.
First, I would like to thank my committee members. Most importantly Prof. Fathi Salem,
my advisor, whom I worked with since my capstone project during my bachelor’s degree. His
kind personality and vast knowledge gave me the energy to continue this labor to the end.
Thanks to Profs. Mason, Xing, and Chakrabartty for the discussions and support. Also, I
would like to thank Prof. Kyle Miller, who enlightened me about neuronal cultures and gave
me the opportunity to have a hands-on experience in magnetic beads.
Second, I would like to thank Miller’s lab members, especially, Phillip Lamoureux, who
constantly provided me with the things I needed for my on-chip biological experiments.
Thanks to Prof. Wen Li for allowing me to use her lab equipment to conduct my experiments.
Also, I am grateful to AMSaC, Microtechnology, and AMI lab members for their discussions
and help throughout my journey.
Finally, I would like to thank everyone who helped me throughout this journey. I couldn’t
list all the names here, however, It would have been impossible to reach this point without
their support. I am very grateful to all who gave me part of their time, knowledge, and
help.

v

TABLE OF CONTENTS

List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
List of Figures

ix

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1 Introduction to Biomedical designs on standard CMOS substrates
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Literature Review and Contributions . . . . . . . . . . . . . . . . . . .
1.2.1 Review of the state-of-the-art . . . . . . . . . . . . . . . . . . .
1.2.2 Main contribution of this thesis . . . . . . . . . . . . . . . . . .
1.3 CMOS scalability, reliably, and compatibility . . . . . . . . . . . . . . .
1.3.1 Scalability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.2 Reliably . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.3 Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4 Point-of-Care testing and personalized medicine . . . . . . . . . . . . .
1.5 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2 Superparamagnetic Beads
2.1 Introduction . . . . . . . . . . . .
2.2 Physical and chemical properties
2.3 Forces . . . . . . . . . . . . . . .
2.3.1 Gravity . . . . . . . . . .
2.3.2 Magnetic Force . . . . . .
2.3.3 Drag Force . . . . . . . .
2.3.4 DLVO Forces . . . . . . .
2.3.5 Other Forces . . . . . . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

3 Integrated Circuits and Systems Design and Implementation
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 DC Manipulation . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.1 Programmable Current Source . . . . . . . . . . . . . . .
3.2.2 Digital Logic . . . . . . . . . . . . . . . . . . . . . . . .
3.2.3 Actuator . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.4 Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.4.1 Bead Transport . . . . . . . . . . . . . . . . . .
3.2.4.2 Pseudo-parallel Manipulation . . . . . . . . . .
3.3 AC Manipulation . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.1 Class E Power Amplifier . . . . . . . . . . . . . . . . . .
vi

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

x

.
.
.
.
.
.
.
.
.
.

1
1
2
2
5
5
6
7
10
11
13

.
.
.
.
.
.
.
.

14
14
15
16
16
17
18
19
22

.
.
.
.
.
.
.
.
.
.

25
25
26
27
29
31
34
43
44
50
50

3.4

3.5

3.3.2 Digital Logic
3.3.3 Actuator . . .
3.3.4 Arrays . . . .
RF Sensing . . . . .
3.4.1 Cross Coupled
3.4.2 Digital Logic
3.4.3 Arrays . . . .
Tri-mode System . .

. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
LC Oscillator
. . . . . . . .
. . . . . . . .
. . . . . . . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

54
55
59
63
63
76
78
82

4 Magnetic Bead Control
4.1 Micro-Beads Motion . . . . . . . . . . . .
4.1.1 First-Order Translational Motion .
4.1.2 First-Order Rotational Motion . . .
4.1.3 Second-Order Translational Motion
4.1.4 Second Order Rotational Motion .
4.1.5 First-Order vs. Second-Order . . .
4.2 On-Chip Coil Model . . . . . . . . . . . .
4.3 Control Design . . . . . . . . . . . . . . .
4.4 Control Based on Energy Functions . . . .

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

88
89
89
91
91
92
92
92
94
97

5 IC Packaging
5.1 Introduction . . . . . . . . . . . . . . . . .
5.2 PDMS . . . . . . . . . . . . . . . . . . . .
5.3 Standard . . . . . . . . . . . . . . . . . . .
5.4 Custom . . . . . . . . . . . . . . . . . . .
5.4.1 Wire Bonding Insulation . . . . . .
5.4.2 Enhanced Wire Bonding Insulation

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

100
100
101
102
103
104
105

.
.
.
.
.
.

111
111
112
114
115
115
118

6 On-Chip Axonal Initiation and Elongation
6.1 Introduction . . . . . . . . . . . . . . . . . .
6.2 Biophysical Model . . . . . . . . . . . . . .
6.3 Experiments . . . . . . . . . . . . . . . . . .
6.3.1 Neuronal Culture . . . . . . . . . . .
6.3.2 Elongation on Dish . . . . . . . . . .
6.3.3 On-Chip Axonal Elongation . . . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

7 Conclusions and Future Work
128
7.1 Summary of Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
7.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
A Maxwell VBS to create and analyze on-chip sensors and actuators for 0.5
CMOS process
131
vii

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

viii

145

LIST OF TABLES

1.1

Comparison with current state-of-the-art in magnetic bead manipulation . .

3

3.1

Coupling coefficients and mutual inductances . . . . . . . . . . . . . . . . . .

36

3.2

Inductors used in this work and their parameters . . . . . . . . . . . . . . .

71

ix

LIST OF FIGURES

1.1

The decreasing rate of transistor dimensions is 0.7x every two years. For
interpretation of the references to color in this and all other figures, the reader
is referred to the electronic version of this dissertation. . . . . . . . . . . . .

7

Drosophila neuronal cultures grown on a CMOS chip treated with polyornithine has a higher yield than neurons grown on plain silicon nitride. The
image was taken 24 hours after placement on chip. . . . . . . . . . . . . . . .

11

1. A disposable strip is used by a health provider. 2. The strip is calibrated
using a portable hand held device. 3. Point of care testing at the site of the
patient is conducted. 4. Real-time results and measurements are reported. .

12

Forces affecting beads in aqueous experiments. The gravity force is not shown
and it is in the same direction as Fmag , Fmag is the magnetic force, FN is the
normal, Fel is the electrostatic double layer , FvdW is the van der Wals, and
Fdrag is the drag force. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

16

Summary of forces acting on a bead with various radii. The parameters in the
equations estimated as: g = 9.82m/s, B1 = 1.43e−2 T , B1 = 187.75T /m,
B2 = 6.25e−4 T , B2 = 14.42T /m, ∆χ = 1, d = 10e−9 m, η = 1, v =
1e−6 m/s, A = 3.4e−21 , λ = 90e−9 m, = 7e−10 , k = 1.38e−23 J/K, T =
298K, NA = 6.02e23 , e = 1.602e−19 C, i = 0.03mol/l, ψs = −3e−3 V , and
ψb = −26e−3 V . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

24

3.1

System-level block diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . .

27

3.2

Variable power bi-directional current sink with a simplified coil path. The
’e’ signal is an “enable” signal which can be used to switch on/off the entire
DC actuation path. The components on the left of the dashed line are shared
across the entire array to save area. . . . . . . . . . . . . . . . . . . . . . . .

28

Post layout simulation of one of the programmable current sources implemented on chip. The magnetic field direction can be changed using the positive current stairs or the negative ones. . . . . . . . . . . . . . . . . . . . . .

29

1.2

1.3

2.1

2.2

3.3

x

3.4

A dedicated AND gate and a transmission gate are used to isolate signals
from/to each coil in the array. . . . . . . . . . . . . . . . . . . . . . . . . . .

30

FEM simulation of the 30x30 coil’s magnetic flux density at the surface of the
chip and near the metal winding. . . . . . . . . . . . . . . . . . . . . . . . .

32

FEM simulation of the coil’s magnetic force generated on various beads using
1.5mA using the 30µm x 30µm coil. The beads were placed on different spatial
locations on the surface of the chip. . . . . . . . . . . . . . . . . . . . . . .

33

FEM simulation of the 80x80 coil’s magnetic flux density at the surface of the
chip and near the metal winding. . . . . . . . . . . . . . . . . . . . . . . . .

34

FEM simulation of the coil’s magnetic force generated on 2.8µm bead using
the 80µm x 80µm coil. The bead was placed in the upper left corner (edge)
and the center locations on the surface of the chip (i.e. Z=0). . . . . . . . .

35

Layout of stacked coil array. The array is centered in the middle and surrounded by digital switches from the left and right sides. The DC current
source along with the thermometer decoder are placed on the top. The digital
row/column decoders are placed at the bottom. . . . . . . . . . . . . . . . .

37

3.10 Photograph of 1.5mm x 1.5mm CMOS die as an open cavity device. . . . . .

38

3.11 Concept drawing showing how an AFM system can be used to measure the
magnetic field generated by the coil array. The cantilever flexes according the
magnitude of the field generated on the magnetized tip. The optical sensor
detects the laser deflection caused by the movement. . . . . . . . . . . . . .

40

3.12 Magnetic Force Microscopy measurement of a 7µm x 7µm region on top of
the coil. The coil was switched on in the middle of the scanning process. . .

41

3.13 A montage of a series of video frames showing how a magnetic bead is attracted (downward) toward a coil then moved to the right by switching on the
neighboring coil. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

43

3.14 A montage of a series of video frames showing how magnetic beads are transported from one location to another. . . . . . . . . . . . . . . . . . . . . . .

45

3.15 A montage of a series of video frames showing how magnetic beads can draw
the letters M S U sequentially using pseudo-parallel scheme. . . . . . . . . .

46

3.5

3.6

3.7

3.8

3.9

xi

3.16 A micrograph of the coils closely stacked with three magnetic beads on top.
The first frame on the left shows a bead “1” traveling to the highlighted coil
when it is switched on. Bead “2” is also influenced by the movement and the
magnetic field in the coil and in the interconnects. (50x magnification) . . .

49

3.17 A variable gain class E power amplifier tuned at 900MHz. The dashed box is
a simplified version showing how the same coil can be shared for sensing and
actuation. In actuation mode, the PAE is connected to a single coil, the other
end of the coil is connected to ground. . . . . . . . . . . . . . . . . . . . . .

51

3.18 The smallest current passing through the actuator during a regular switching
period. The peak-to-peak value is about 150µ. In this view, two actuators are
turned on and off (only one is shown) while the rest of the array is completely
shut off. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

52

3.19 PAE output spectrum tuned at 900MHz. In order to minimize harmonic
distortion outside the band of interest, the actuator impedance as well as the
LC filter tank associated with the PAE are tuned to 900MHz . . . . . . . . .

53

3.20 Min-sized, three stage ring oscillator with custom W/L to create an average
delay of 0.10422 ns. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

54

3.21 The average output current for different switching periods (20ns to 29ns). The
magnitude is maximum at 28ns and 21ns switching periods. . . . . . . . . .

55

3.22 The center cell labelled L5 (blue curve) is being switched on and off every 250ns. The green curve depicts the logical switching activity on the
transmission-gate switch and the other curves (shown in red) represent the
currents going through the 8 neighboring coils. The magnitude of the current
going through L5 dominates over the rest, in contrast of using a single switch
which produces much more noise (not shown). . . . . . . . . . . . . . . . . .

56

3.23 Three layered tiny coil 3D model. Each metal layer is designed as a “U” shape
object. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

57

3.24 [3D Side view] Simulated magnetic field shown in green/blue as a result of
passing 1.5mA through it (3 metal layers highlighted in pink). The B-field
near the surface is approximately 0.4mT . . . . . . . . . . . . . . . . . . . .

57

3.25 The forces (in Newton) generated on the cantilever tip by a single on-chip coil
using different power levels. The error bars represents the variation in spring
constants provided by the cantilever’s fabrication house. . . . . . . . . . . .

58

xii

3.26 System block diagram. The global control is responsible for managing all
building blocks to create the desired sensing or manipulation profile. . . . . .

60

3.27 An annotated picture of the fabricated die: The RF components are placed
far away from the array to reduce substrate injection and coupling. The I/O
pads are removed from the 3 sides to permit easier testing. . . . . . . . . . .

61

3.28 Comparison between two layout topologies. . . . . . . . . . . . . . . . . . . .

62

3.29 A micrograph snippet of the 6x6 array using 6x6 microns coils. The oscillator,
PAE, and DC source are not shown. (20x magnification) . . . . . . . . . . .

63

3.30 Cross coupled LC oscillator. The output frequency of the oscillator is also
tuned at 900MHz. The variable capacitor is implemented as a varactor bank.

65

3.31 Computed coil’s inductance based on the location of the bead in the X-Y
plane. The coil is 6x6 microns, its center is located at (0,0) in the X-Y plane.

66

3.32 Zoomed coil’s inductance just above the tiny coil. The location of the bead
in the X-Y plane. The coil is 6x6 microns, its center is located at (0,0) in the
X-Y plane. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

67

3.33 Layout of improved sensing on-chip inductor. Three metal layers were used.
Metal3 for winding, Metal2 for exit bridge, and Metal1 for grounding. . . . .

68

3.34 Inductor characteristic graphs. Inductance vs. Frequency, Q-value, versus
Frequency, and Resistance vs. Frequency. . . . . . . . . . . . . . . . . . . . .

69

3.35 Schematic of the circuit level model of the inductor. Enhanced PI-model. . .

70

3.36 PSS simulation result. The phase noise and output noise are shown on the
right figure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

72

3.37 Transient response when the varactor is digital set at maximum and minimum
capacitance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

73

3.38 On-chip counter output using an oscilloscope. . . . . . . . . . . . . . . . . .

75

xiii

3.39 The blue curve represents the output when the chip is submerged in distilled
water with no beads present. When a bead comes nearby the sensing sight a
downshift of resonance frequency is observed. When the bead lies on top of
the sensing site the oscillation is lost (not shown). . . . . . . . . . . . . . . .

75

3.40 Tank resistance in ohms which needs to be compensated for using the active
devices. This resistance includes the digital switches and the varactors. . . .

77

3.41 A DC sweep simulation of a PMOS device where the width of the device is
incremented while keeping the source-drain current fixed. VDD is 3 volts. . .

77

3.42 A simulation of 3 differently sized NMOS transistors to find the W/L ratio
which will create a proper gm to overcome the inductor’s. . . . . . . . . . . .

78

3.43 An AC simulation running at vdd=1.8v to find the conductance of Inductor4,
two analog multiplexers, and two analog switches. . . . . . . . . . . . . . . .

80

3.44 A DC simulation running at vdd=1.8v to find the transconductance (gm) of
a NMOS transistor. The source is connected to 0.7V to realize the body effect. 81
3.45 A DC simulation running at vdd=1.8v to find the transconductance (gm) of
a PMOS transistor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

81

3.46 A DC simulation running at vdd=1.8v to find the voltage at which the tail
NMOS transistor provides 15mA. . . . . . . . . . . . . . . . . . . . . . . . .

82

3.47 5 Coils in the array are scanned to sense magnetic beads. One coil detects a
bead when the output frequency is downshifted. The other coil stop oscillating
(curve not shown), and the other three coils have the same frequency output.

83

3.48 A 3x3 coil array with improved inductors Q-value and sensing isolation. . . .

84

3.49 System block diagram for tri-mode operation. . . . . . . . . . . . . . . . . .

85

3.50 A micrograph of the tri-mode system. . . . . . . . . . . . . . . . . . . . . . .

86

3.51 The frequency response of the large switch. The output voltage divided by
the input voltage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

87

4.1

Comparison between 1st-order transitional motion model vs. 2nd-order. In
the 2nd-order the bead’s inertia is included. . . . . . . . . . . . . . . . . . .
xiv

93

4.2

Visualization of a 80x80 micron on-coil as used on the chip, in the Cartesian
space. A, B, C and D indicate the four sections of the mathematical model. .

94

4.3

Control Strategy for second order bead dynamics. . . . . . . . . . . . . . . .

95

4.4

Process reaction curve of a bead moving from one coil’s center to another,
based on a second order system. For the Zieger-Nichols tuning, a tangent is
drawn approximately in the inflection point. TL = 30sec and a = 39 · 10−6 .

96

4.5

Simulation result of the response to a step function at t = 1 sec. . . . . . . .

96

5.1

Standard ceramic package with PCB board for testing. . . . . . . . . . . . . 101

5.2

Standard plastic package with drainage for testing. . . . . . . . . . . . . . . 102

5.3

Packaging of the CMOS die on top of a PCB probe. . . . . . . . . . . . . . . 104

5.4

Packaged CMOS die on top of a prototype PCB. . . . . . . . . . . . . . . . . 105

5.5

Epoxy covering bonding wires. . . . . . . . . . . . . . . . . . . . . . . . . . . 106

5.6

Micrograph of CMOS die submerged in fluid with magnetic beads on top.
The image illustrates how the epoxy covers part of the die and made 3 coils
unusable. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

5.7

Top view of chip1 insulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

5.8

Top view of chip2 insulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

5.9

Final packaged device. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

6.1

The axon is treated as a series of dashpots (G). Where F is the force or tension
applied or generated at the growth cone and the friction dashpot η represents
the attachments to the substrate. . . . . . . . . . . . . . . . . . . . . . . . . 113

6.2

Axon’s model simulation, without mass addition, shows a 11µm long axon
being stretched for 1 hour with 1nN tension force. After 15min the length is
11.9. After 30min the length is 12.97. After 45min the length is 14.22. At
the end of the simulation the axon’s length is 15.71. A total growth of 4.71
microns. The diameter of the axon is thinning because mass addition was not
realized. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
xv

6.3

Three electromagnetic placed around the sample. . . . . . . . . . . . . . . . 116

6.4

Magnetic bead manipulation using three of the shelf electromagnetic. . . . . 117

6.5

Custom stage where the electromagnetic is placed in plane with the sample.

6.6

Forces generated from electromagnet EM100. . . . . . . . . . . . . . . . . . . 119

6.7

Montage of two magnetic beads trying to tow axons grown on a dish. . . . . 120

6.8

Montage of a magnetic bead trying to tow axons grown on a dish. The axon
actually pulls the bead toward its body. . . . . . . . . . . . . . . . . . . . . . 121

6.9

A group of neuronal cells grown on a treated surface of Silicon Nitride. . . . 123

119

6.10 A neuronal cell with multiple axon on top of the coil array. . . . . . . . . . . 123
6.11 On-Chip axonal elongation setup. The open-cavity device makes it easier to
deposit small droplets of embryos. The coverslip is placed on top of the plastic
package to reduce vibrations and evaporations. . . . . . . . . . . . . . . . . 124
6.12 On-Chip axonal elongation, initial axon state where the bead is bond to the
east of the axon. The bead is not visible under fluorescent filters. . . . . . . 126
6.13 On-Chip axonal elongation, final axon state where it elongated 1.7µm in 20
min. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.14 On-Chip axonal initiation, initial axon state where the bead is bond to the
west of the axon. The bead is not visible under fluorescent filters. . . . . . . 127
6.15 On-Chip axonal initiation, final axon state where it branched out after 30 min.127

xvi

Chapter 1
Introduction to Biomedical designs on
standard CMOS substrates

1.1

Introduction

Conventional micro and nano magnetic particles have become a unique tool for analyzing
biomaterials. Since the mid 1800’s several USA patents like [1, 2, 3] appeared on magnetic
separation and arrogation. Most techniques used until the early 1990’s were based on bulky
equipment and large magnetic particles. The introduction of nano superparamagnetic beads
few decades ago revolutionized the process of biomedical analysis and became a new tool in
numerous scientific and clinical applications, e.g., cell counting, protein isolation, infection
detection, sorting, in-vitro diagnostic (IVD) immunoassays, and more. During the same
period of time, electronic design and fabrication processes became more advanced and utilized
complementary metal-oxide-semiconductor (CMOS) to drive low-cost, low-power, and high
efficiency chips.
1

In this thesis, we design, develop, and test a platform based on standard CMOS technology to create a high-precision, economical, and high throughput lab-on-chip (LOC) device
suitable for sensing and actuation of bead-tagged biomaterials. The platform is also utilized
to demonstrate a unique biomedical application, on-chip axonal elongation.

1.2
1.2.1

Literature Review and Contributions
Review of the state-of-the-art

Current state-of-the-art designs of magnetic bead sensing and actuation include capillaries
or microfluidic channels and pumps [4, 5, 6] to guide the biomaterial samples into chambers
or fenced regions for manipulation. Lee et al. [4, 7, 8] started with a simple ring trap and
a matrix of wires to trap and move a single bead to a desired location, then they designed
a hybrid system using a Polydimethylsiloxane (PDMS) microfluidic structure on top of a
CMOS die to sequentially move multiple beads to a certain location using an 8x8 coil array.
Also, [9] introduced a device that is capable of moving a collection of beads, enclosed by
capillaries, using a line of integrated coils. Other approaches, e.g., [10], used mechanical
structures such as spin valve arrays where the magnetic bead can be trapped, or use micromagnetic-tweezers to guide a bead through a pre-defined path, e.g., [11]. Additionally, [12]
built a line of coils on a Printed Circuit Board (PCB) device. Their device also used a
microfluidic channel on top of the PCB to guide the beads from one coil to another.
The main limitation in these implementation is that there is no capability of (pseudo)parallel movement of beads. Prior work have demonstrated only sequential bead movement
(at most every 10-20 m seconds [4], which means other beads would change position over
2

Table 1.1: Comparison with current state-of-the-art in magnetic bead manipulation
This
[4]
[8]
[9]
[12]
Work
Cur40mA
20mA
11mA
40mA
400
rent
A/mm2
Source
Shared
Shared Dedicated
Coil
80µm
20µm
25µm
400µm
3.5mm
Array
8x8
8x8
5x5
Line
Line
Bead
2.8µm
8.5µm
8.5µm
15µ
1µm
TechCMOS
CMOS
SiGe
CMOS
PCB
nology
0.5µm
0.18µm
3M
0.35µm
3M2P
6M1P
4M2P
Manip- Multiple Multiple
Single
Single
Single
ulation
PseudoSequeParallel
ntial
Freq.
40MHz
50Hz
Force
600pN
40pN
50pN
fN
Package Standard µFluidic µFluidic µFluidic µFluidic
PDMS
PDMS

the coil array). In this work, we apply a force to multiple beads (pseudo-)simultaneously in
the order of (1/40M=) 250n seconds. Thus one can move multiple beads at several locations
(pseudo-)simultaneously. Additionally, we distinguish our design from the aforementioned
present-state-of-the-art, as summarized in Table 1.1, by introducing a new programmable
manipulation platform using high-speed switches on a closely stacked coil array, which can
simultaneously transport and move magnetic beads in a pseudo-parallel fashion. The coils
are carefully laid out and stacked to generate overlapping magnetic fields, using larger DC
currents, to assure coil-to-coil transport. And non-overlapping fields, using smaller DC
currents, to assure valid trapping. Additionally, the pseudo-parallel manipulation platform
allows the programmable array to individually move multiple small objects or collaboratively
transport larger ones.
On-chip magnetic bead sensors usually require an external magnetic source to generate
3

the reference magnetic field [13, 14]. Or they use the CMOS Hall effect to sense magnetic
particles using “Hall” arrays and chambers [15, 16]. Additionally, [17] uses a photo-diode
configuration to detect beads lying above the center of the coil. The main limitation of this
setup is that the beads are detected only when they pass through a very small region (namely,
center of the coil). Moreover, [18, 19, 20] starts with a single LC cross coupled oscillator
to detect beads and expands it to an array of sensors by replicating the active transistors
for four sensing sites. In this work, we share a single pair of cross coupled active transistors
across the entire array by utilizing high speed switches for pseudo-parallel sensing.

Finally, state-of-the-art designs have been limited to sensing only [18], actuation (manipulation) only [4], or designs which require a user intervention or special micro-fluidic
packaging. Here, we present programmable compact coil arrays for dynamic pseudo-parallel
sensing and manipulation by multiplexing a shared core (specifically the core consist of crosscoupled active transistors for enabling oscillation and sensing, a DC current source, and an
AC currentsource). All components of our design are integrated on the same die and used
for sensing and/or actuation without the need for specialized post processing, magnetized
materials, or specialized packaging. Our integrated platform introduces new possibilities to
sense and manipulate [21, 22] multiple small magnetic particles, at different spatial locations,
in pseudo-parallel fashion or large single particles with a collaborative execution scheme of
the array of coils. Each “cell” in the array can generate 8 programmable magnetic field
levels or sense magnetic beads with a diameter of at least 1µm; thus, giving the user more
functionality and flexibility in a unified standard package that does not require microfluidic
pumps and structures. Nevertheless, It is possible to integrated PDMS structures on top of
the simplified packaging if the targeted application requires such setup.
4

1.2.2

Main contribution of this thesis

The list below itemizes the contributions which differentiates this work from current stateof-the-art:

1. Tri-mode lab-on-chip device capable DC actuation, AC actuation, and RF sensing of
magnetic beads using a shared micro-coil array and shared DC, AC, and RF cores.
2. Pseudo-parallel manipulation scheme using high-speed switches (up to 40MHz) for
simultaneous maneuvering and transport of magnetic beads.
3. Novel biomedical application: on-chip axonal elongation using magnetic beads.
4. Simplified open-cavity design to alleviate micro-fluidic complexity and facilitate onchip biological cultures studies.

1.3

CMOS scalability, reliably, and compatibility

The main driver of the Integrated Circuits (ICs) industry are microprocessor chip makers
who mainly target consumer electronics, personal computers, laptops, and smart phones.
However, due to their mass production, availability, and price competitiveness, CMOS dominated the semiconductor business for the past two decades. As a result, the prices of CMOS
devices went down as their production went up. Consequently, other industries including
biomedicine have been taking advantage of this well established technology and utilized its
scalability, reliability, and compatibility to deliver miniaturized biomedical instruments for
the masses.
5

1.3.1

Scalability

Feature size and channel length in CMOS have been shrinking steadily. According to the
so-called Moore’s law, the number of transistors in integrated circuits approximately doubles
every two years. The largest semiconductor chip maker Intel, e.g., has kept up with this pace
for the past 40 years. Fig. 1.1 shows the feature size of the transistor verses the production
year. As the number of transistors double every two years more functionality is being added
to integrated circuit chips, thus, making them more capable. This makes CMOS technology
more appealing to other industries like biomedicine.

Biomedical applications are quite sophisticated they often require a direct interface with
organic matter such as cells, chemical compounds, or living tissue. Such requirements not
only demand a high degree of circuit intelligence but also precise power, temperature, and
size specifications. Using CMOS allows us to exploit the large number of transistors available
on a single die to increase functionality, drive down the cost while meeting specifications and
demands.

Most digital designs can be trivially reduced in size using common frequency and channel
scaling techniques without the need to completely redesign the integrated circuitry. On
the contrary, analog designs don’t follow the same trend and often require further design
considerations and adjustments. Consequently, they don’t scale as easy as their digital
counter parts. In biomedical applications, mixed signal (Analog/Digital/RF) designs are
dominant due to the sophisticated interface between the living cells and the electronics
connected to them. Thus, CMOS scaling can be utilized in the digital portion of the design.
6

Figure 1.1: The decreasing rate of transistor dimensions is 0.7x every two years. For interpretation of the references to color in this and all other figures, the reader is referred to the
electronic version of this dissertation.

1.3.2

Reliably

Most biomedical applications require stable functionality over a very long period of time. For
example, heart pacemakers are expected to live seven to ten years depending on the implant.
Thus, lifetime and longevity are crucial aspects when considering such devices. CMOS has
proven itself over the years to produce reliable high-yield chips. Most devices designed in
CMOS have an average life span of ten years. The most common reasons which cause
physical failure in CMOS chips are electromigration, time-dependent gate oxide breakdown
(TDGOB), and hot carriers injection (HCI).
Electromigration is the gradual displacement of the ions in a conductor as a result of the
current flowing through it. This effect is very important in our application because of the
use of on-chip coils where each coil is supplied with a variable electrical current to generate
a distinct magnetic field. When high current densities are used, integrated circuits are more
7

likely to fail earlier than expected. Using a mathematical model developed by Jim Black of
Motorola, one can predict the mean time to failure (MTTF) for a semiconductor circuit due
to electromigration. The model is known as Black’s equation[23]:
Q

M T T F = Aj −n e kT ,

(1.1)

where A is a predefined constant based on experimental data, j is the current density, n is
a model parameter, Q is the activation energy, k is the Boltzmann constant, T is the temperature in Kelvins. Therefore, careful measures and design techniques should be followed
when using large current densities j to avoid electromigration. It is apparent that current
density j and the temperature T are the main factors in the design process which affect
electromigration. Therefore, for an interconnect to remain intact at high temperatures, the
maximum current density of the conductor must necessarily decrease. These considerations
are discussed in Section 3.2.1.

Time-dependent gate oxide breakdown, which is also referred as Time Dependent Dielectric Breakdown, is the failure where the gate oxide breaks down as a result of long-term low
electric field. The breakdown is caused by the development of a conducting path through the
gate to the substrate due to electrons tunneling as a result of operating MOS at a voltage
higher than the one specified by the foundry. When there is a conducting path between the
gate and the substrate, it is not possible to control the current flow between the drain and
source by means of the gate voltage; consequently, creating faulty logic in the circuit which
renders the (biomedical) device useless. Hu et al. [24] created a mathematical model which
8

fits the TDGOB of several devices:

t
·t
tBD = BD1 BD2 ,
tBD1 + tBD2

(1.2)

where tBD1 and tBD2 are the lifetimes predicted by the E-model (thermochemical, the
breakdown occurs due to field enhanced bond breakage) and the 1/E model (hole-induced
breakdown) respectively and can be approximated as follows:

tBD1 = t1 · exp (−γ · Eox ) ,

(1.3)

tBD2 = t2 · exp (G/Eox ) ,

(1.4)

and

all fitting constants (t1 , t2 , and γ) can be obtained from [24]. Accordingly, the design has to
adhere to foundry design rules for gate voltages and proper operating temperature to guarantee proper life span for the device. It is important to note that these approximate models
start deviating when the thickness of the oxide becomes very thin e.g. 5nm. Therefore, one
cannot engineer a device to fit these models; thus, the best practice is to follow the guidelines
supplied by the process developer and foundry when setting voltage specifications.
Hot carrier injections are electrons or holes which have gained high kinetic energy while
being accelerated by a strong electric field in a MOS transistors. Because of this high
kinetic energy, hot carriers could get injected in places where they should not be, thus,
making the transistor behave different that expected. The transistor’s transconductance
slowly degrades causing the threshold voltage to change, hindering it from forming a channel
in the drain region. This causes digital circuits (as well as threshold sensitive analog circuits)
9

to fail prematurely due to logical errors caused by the threshold shift. It is difficult to
mathematically model hot carrier damage rates. Most models are not sufficiently accurate
in predicting such rates. The reader is referred to [25] for more details on HCI and other
reliability mathematical models.

1.3.3

Compatibility

There are several points where CMOS has to be compatible with biomedicine. First, low
power: when CMOS was initially introduced, the complementary logic made it so appealing
because an engineer had to design half the circuit and the other half is simply the complement. This methodology also allows half the circuit to be switched on while the other is
completely off. As a result, the power is substantially reduced compared to other technologies like BJTs (Bipolar junction transistors). Low power designs make it possible to create
devices that can run for years without replacing the power source, e.g., battery. Moreover,
they can be designed to harvest power from their surroundings without requiring bulky or
additional components; thus, making them very attractive to the biomedical applications.
Second, biochemical compatibility: most CMOS substrates are protected and passivated
with Silicon Dioxide (SiO2 ) and Silicon Nitride (Si3 N4 ). These two chemical compounds are
employed to create biocompatible platforms for growing cultures[26, 27] or create implantable
devices[28, 29].
In this work, we compare growing neuronal cultures on plain silicon nitride and on pretreated silicon nitride. We also measure the growth yield and choose a suitable protocol for
subsequent experiments. Fig. 1.2 shows an image of neuron cultures created from drosophila
embryos grown on a CMOS coil array. Section 6.3 details the process of growing neurons
10

Figure 1.2: Drosophila neuronal cultures grown on a CMOS chip treated with polyornithine
has a higher yield than neurons grown on plain silicon nitride. The image was taken 24 hours
after placement on chip.
on-chip and how to provide a more compatible substrate for better yield.

1.4

Point-of-Care testing and personalized medicine

Point of care testing (POCT) is medical testing near the site of patient care. It allows
quicker medical response, convenience, and better results. Most POCT devices are built to be
mobile, disposable or reusable, and support a wide range of tests without the need for bulky
laboratory equipment, see Fig. 1.3[30]. The key objective of POCT is to provide economical
and quick devices that can be deployed in remote areas or any place where sophisticated
11

Figure 1.3: 1. A disposable strip is used by a health provider. 2. The strip is calibrated
using a portable hand held device. 3. Point of care testing at the site of the patient is
conducted. 4. Real-time results and measurements are reported.

lab equipment are not present or available at hand. As a result, the proper treatment
can be implemented at the same time and place, leading to an improved clinical outcome
for the patient and the medical provider.

A good example of POCT and personalized

medicine is the use of portable blood glucose meters, which allow diabetic patients to actively
monitor their blood sugar at their convenience without the need to leave their homes or see a
specialist. Consequently, achieving faster results, better care and awareness, and lower cost.
Designing such devices require small footprint, low power, low cost, and high throughput.
In this work, we utilize the CMOS features discussed in 1.3 to create a lab-on-chip platform
capable of detecting and manipulating magnetic bead-tagged biomaterials. This device can
be used for research and development or clinical studies.
12

1.5

Thesis Organization

The thesis is organized as follows: Chapter 1 introduces the overall technical subject, lists
CMOS features, example applications, challenges, and motivation. A summary of current
state-of-the art is also discussed in this chapter. In chapter 2, we describe and introduce
magnetic beads, list their physical and chemical properties, and the forces acting on a magnetic bead suspended in a fluidic sample. Chapter 3 is where we introduce our manipulation
and sensing devices with supportive simulations and lab experiments. Each component is
explained and discussed separately in a dedicated section to show our scientific contribution
at the circuit and at the system levels. In Chapter 5, our customized packaging method
is summarized and supported with a lab experiment to show its effectiveness. Chapter 6
demonstrates and introduces a unique biomedical application “on-chip axonal elongation”
using magnetic beads where our device is used to accelerate the growth rate of an axon in
vitro. Finally, Chapter 7 lists our contributions and lays out future steps needed to improve
this effort.

13

Chapter 2
Superparamagnetic Beads

2.1

Introduction

Micro- and nano-sized magnetic particles have been used extensively in biomedical analysis in
recent years due to their compactness, versatility, and reliability [31, 32, 16, 33, 34]. Most of
the time, the application of interest imposes the bead’s size, coating, and iron mass content.
First, magnetic bead diameters come in different sizes ranging from 100s [35] of nano meters to a few micro meters [36]. However, the target biomaterial size (e.g. protein, blood cell,
etc) decides the appropriate diameter of the bead. Second, the iron mass content (e.g. 15.26
mg/mL for the Adembeads [37]) is determined by the amount of force needed to maneuver
the biomaterial, for example, the more iron content the better the magnetic manipulation.
Lastly, the biochemical coating provides flexibility in terms of binding to different antibodies,
peptides, etc. Therefore, the targeted bio-material, the surrounding (fluidic) medium, and
the substrate help identify the kind of coating to be used in an experiment or an application.
These characteristics have a direct impact on the forces affecting the magnetic beads in an
14

aqueous environment.
The magnetic beads used and/or modeled in this work are Dynabeads MyOne (diameter 1µm), Dyabeads M-280 (2.8µm diameter), Dyabeads M-450 (4.5µm diameter), and
AbCTM (6µm diameter). The selection criteria were based on the size and the iron mass
content of the bead, where the target bio-material size was around 1 to 20µm and the forces
needed were in the nano- and pico- newtons ranges.

2.2

Physical and chemical properties

Typical superparamagnetic micro and nano spheres (beads) are made of Silicon Dioxide
(SiO2 ) or polystyrene matrices and iron oxide crystals. They are usually synthesized with
different compositions which include magnetite (Fe3 O4 ), maghemite (Fe2 O3 , γ-Fe2 O3 ), or
ferromagnets like Fe and Co.
There are several methods and protocols [38, 39, 40] to create coated magnetic beads.
The most common methods are coprecipitation and thermolysis. There are other methods
like microemulsion and hydrothermal which are used less nowadays. First, coprecipitation
is done by adding a base under inert atmosphere around room temperature to aqueous
Fe2+ /Fe3+ salt solution to get iron oxides like Fe3 O4 or γ-Fe2 O3 . However, the challenge
in this process is to control the particle size and achieve an accepted particle distribution.
Second, thermolysis is thermal decomposition caused by heat which breaks chemical bonds
in the compound (organometallic) undergoing decomposition.
15

Figure 2.1: Forces affecting beads in aqueous experiments. The gravity force is not shown
and it is in the same direction as Fmag , Fmag is the magnetic force, FN is the normal, Fel
is the electrostatic double layer , FvdW is the van der Wals, and Fdrag is the drag force.

2.3

Forces

Now we introduce and discuss the major forces, shown in Fig. 2.1, a bead encounters in
aqueous experiments:

2.3.1

Gravity

The main force responsible for sedimenting the magnetic bead is the gravitational force.
This force is simply given as
FG = mg,
where m is the mass of the bead, g is the gravitational acceleration.
16

(2.1)

2.3.2

Magnetic Force

A (magnetic) bead in a magnetic field, say H, becomes magnetized and acquires a moment
m given by
m = Vm M = Vm (∆χH),

(2.2)

where Vm is the volume of the bead, M is known as the magnetization of the bead, and ∆χ
is the difference in the relative magnetic susceptibility between the magnetic bead and the
surrounding medium. It is important to note that the bead’s magnetization becomes saturated in practice, and thus the magnetic moment saturates independent of further increases
in the applied field (H). As a result, the amount of magnetic force generated by passing DC
currents through the coils is limited by the magnetic moment of the bead.

Now the magnetic force acting on the (magnetized) bead is due to the field generated by
the on-chip spiral coils. We begin with the magnetic potential energy, say U , for trapping a
magnetic bead, given as (see [41])
1
U = − m · B,
2

(2.3)

where m is the magnetic moment as defined in (2.2) and B is the (external) magnetic flux
density. The moment of (2.2) is re-expressed as

m=

Vm (∆χ)
B,
µ0

(2.4)

where µ0 is vacuum permeability, and B is the assumed acquired magnetic field by the bead.
The magnetic force is thus defined as the spatial gradient descent of the potential energy,
17

i.e.,
Fm = − U =

1 Vm (∆χ)
B2,
2 µ0

(2.5)

where B 2 is the dot product of the magnetic field vector with itself. Note that, when the
relative susceptibility is positive the generated magnetic force acts in the negative direction
of the gradient.

2.3.3

Drag Force

Any object moving through a fluid encounters a hydrodynamic resistance which opposes its
motion. This resistance is called the drag force, and it is directly related to the viscosity of
the medium the object is traveling in. Assuming that the bead is a perfect sphere moving
in a medium with a low Reynolds number (R0 ), the drag force is approximated as

Fd = −6πrh ηvs · fD ,

(2.6)

where rh is the hydraulic radius of the bead, η is the dynamic viscosity of the fluid, vs
is the velocity of the bead relative to the fluidic environment, and fD is the drag coefficient. Temperature plays a significant role in keeping the viscosity of the medium constant.
The dependence of viscosity on temperate can be captured using different models [42, 43].
Consequently, the application of interest, which determines the medium and the targeted
bio-sample should enforce careful temperature control in order to maintain the pre-calculated
viscosities and drag forces.
The drag coefficient has a significant effect on the bead while it is moving near a solid
wall (i.e. the substrate surface of the chip). An expression introduced in [44] models the
18

relationship between the drag coefficient, the radius, and the (minimum) distance (say, d)
between the surface and the bead as

fD = 1 −
−

9
16

45
256

rh
1
+
rh + d
8

rh
rh + d

4

rh
rh + d

rh
rh + d

−

1
16

3

(2.7)
5 −1

.

It is noted that the drag coefficient equals 1 far away from any surface, and increases as it
approaches a surface. To capture the significance of this coefficient, assume that the distance
(d) between the bead and the surface approaches zero; as a result of that, the drag force
increases more than twofold.
Using this expression and (2.6), one can estimate the expected amount of drag force
which would hinder the movement of the bead/bio-material prior to the fabrication of the
CMOS coil array. Therefore, the specifications of the coil (i.e. the number of turns and
track width) and DC circuity (i.e. the amount of current passed through the coil) have to
be adjusted carefully to overcome this drag force. Other parameters like the radius of the
bead and its velocity can be adjusted by changing the type of bead and/or microfluidic flow.

2.3.4

DLVO Forces

DLVO forces result from a theory named after Derjaguin, Landau, Verwey, and Overbeek
[45, 46]. They separately suggested that interactions between small particles in aqueous
solutions can be explained by the exchanges between two forces: (1) the van der Waals
force [47] which is the total dipole-dipole force, dipole-induced dipole force, and dispersion
forces and (2) the electrostatic double-layer force [48] where two parallel layers of charge
19

surround the surface of the chip. The first layer is the surface charge and it comprises ions
adsorbed directly onto the object due to chemical interactions. The second layer is made of
ions attracted to the surface charge by the electric or coulomb force, thus, isolating the first
layer.

First, the van der Waals force entails biochemical bindings (i.e. of antibodies) as a form
of non-covalent bonds by means of Lifshitz-van der Waals interactions [49]. Additionally, the
van der Waals energy interaction between a spherical shape (i.e the bead) and a flat surface
(i.e. the surface of the chip) is approximated by

W =−

Arh
,
6d2

(2.8)

where A, the Hamaker constant of the bead, is dependent on the material of the bead, the
surface, and the medium. d is the (minimum) distance away from the surface of the chip (in
the vertical z-direction ez ). The van der Waals force is modeled as
ˆ

FvdW = W

1
· ˆz ,
e
1 + 14d/λret

(2.9)

where λret is the characteristic wavelength of interaction [50]. It becomes significant when
the beads are close to each other or to the chip’s surface.

The second DLVO force is the electrostatic force. It is induced through adsorption of
ions present in the liquid and dissociation of surface groups (silanol groups for glass or silica
surfaces). When the double-layers of two approaching objects overlap (e.g., bead and chip’s
surface) an electrostatic force is created; it can be either repulsive or attractive [51, 52]. This
20

force can be written as

Fel =

2π κrh
2ψs ψb exp (−κd)
1 − exp (−2κd)
2
2
e
ψs + ψb exp (−2κd) · ˆz ,

where

(2.10)

is the permittivity of the fluid medium, ψs is the surface potential of the chip, ψb is

the surface potential of the bead, and κ is the Debye-H¨ckel inverse double-layer thickness,
u
which is defined as
κ=

kT
2000NA e2 i

−1

,

(2.11)

where k is the Boltzmann constant, T is the thermodynamic temperature in Kelvins, NA is
the Avogadro constant, e is the charge, and i is the ionic strength of the electrolyte medium
(Mole/l).

The van der Waals forces are initially attractive, however, the electrostatic double-layer
force is either attractive or repulsive depending on the surface charge. A net repulsive
DLVO force is desired between the beads and the surface of the chip to ensure stability of
the colloid[53]. Lack of such stability might result in beads aggregations and/or adhesions to
the chip’s surface. Finally, the ionic strength i and the pH of the medium are important in
order to predict the DLVO forces affecting the beads. For example, in order to successfully
grow cell cultures (e.g. neurons) on the surface of the chip, the medium surrounding the
culture will mandate a very limited range of pH and ionic strengths. Consequently, these
limitations will make the coating of the beads and the surface of the chip as one of the major
issues to be addressed in order to grow a successful culture as well as diminish undesired
DLVO forces.
21

2.3.5

Other Forces

There are also a collection of forces which cause random Brownian motion. Brownian motion
causes the random movement of small objects while they are suspended in liquid. This motion
should be taken into account when trapping or tethering magnetic beads at a specific location
using on-chip coils. In this scenario, the magnetic bead “drifts away” or diffuses from the
point of interest if there is no magnetic force trapping it in place (e.g, when the coil is
momentarily switched-off to save power or actuate at another spatial location). Therefore,
to estimate the random distance traveled by the bead and the amount of time it takes the
bead to escape from the magnetic trap, we compute the diffusion constant, say D. To this
end, one uses the Stokes-Einstein equation [54] for spherical particles moving through liquids
with low (Reynolds number) R0 as follows:
k T
D= B ,
6πηr

(2.12)

where kB is Boltzmann’s constant, T is the thermodynamic temperature in Kelvins, and r
is the radius of the bead. Accordingly, the total brownian distance traveled is computed as
the root mean-square displacement (RMSD) [55] as

x2 =

where x2 is the averaged

n
i=0

qi Dt,

(2.13)

x2 (sum of all square displacements), t is the time in seconds,
i

and qi is a numerical constant which depends on dimensionality. For example, if the bead
is restricted to move in one dimension (i.e., 1D) then q1 = 2, for 2D, q2 = 4, and for
3D q3 = 6, etc. As a result, it is obvious from (2.12) and (2.13) that the displacement is
22

inversely proportional to the square root of the radius of the bead. In this work, the bead is
considered to move in the X-Y plane only and it doesn’t experience a brownian displacement
larger than the radius of the on-chip coils. Finally, the switching period (for the pseudoparallel execution) of a coil should be smaller than the time it takes a bead to diffuse from
the center of the coil (trap) to its outer edge. Therefore, instead of keeping a coil switched
on all the time to trap a certain bead, one can optimally select a switching period to trap
a bead before it diffuses out of range; thus, saving power and reducing the amount of heat
generated.
Finally, Fig. 2.2 summarizes the forces discussed above in terms of the radius of the
magnetic bead used. First, the gravity force is mainly canceled by the buoyancy and is
neglected when transporting beads on the surface of the chip. Second, two magnetic forces
(Magnetic1 and Magnetic2) were generated from different coil variants to illustrate the range
of forces, these can be increased by incrementing the amount of current (therefore B) through
the coil. Third, the drag force becomes more prominent for nano-scale sized beads. In this
work, the size of the bead is in the micro-range (i.e. 10−6 ). Fourth, the van der Waals
force (FvdW) and electrostatic double-layer force (FeL) make up the DLVO forces which
can hinder the movement and transportation of the beads when configured incorrectly.

23

10

Force (N)

10

10

10

10

5

Gravity
Magnetic1
Magnetic2
Drag
FvdW
FeL

10

15

20

25

10

9

10

8

10

7

10

6

10

5

10

4

Bead Radius (m)
Figure 2.2: Summary of forces acting on a bead with various radii. The parameters in the
equations estimated as: g = 9.82m/s, B1 = 1.43e−2 T , B1 = 187.75T /m, B2 = 6.25e−4 T ,
B2 = 14.42T /m, ∆χ = 1, d = 10e−9 m, η = 1, v = 1e−6 m/s, A = 3.4e−21 , λ = 90e−9 m,
= 7e−10 , k = 1.38e−23 J/K, T = 298K, NA = 6.02e23 , e = 1.602e−19 C, i = 0.03mol/l,
ψs = −3e−3 V , and ψb = −26e−3 V .

24

Chapter 3
Integrated Circuits and Systems
Design and Implementation

3.1

Introduction

Low cost lab-on-chip platforms are establishing a new venue for clinical diagnostics specifically in point-of-care testing (POCT) and in-field deployment. They are becoming more
attractive due to their portability, inexpensiveness, disposability, robustness, reliability, and
high throughput. By using very small sample volumes (in the order of 1-100µL) one can
identify, diagnose, or separate samples with very high precision using superparamagnetic
beads.
Binding biomaterials to magnetic beads is relatively simple. It requires a standard
preparation protocol which depends on the surface coating of the bead. Manufacturers
[37, 56, 35, 36] offer variety of coatings to enable easier binding and faster deployment. For
further discussion on the use of magnetic beads in analyzing biological and chemical samples,
25

the reader is referred to a recent review by M. Gijs et al. [5].

3.2

DC Manipulation

In this section we introduce the design of 8x8 magnetic manipulation coil-arrays, fabricated
in standard 0.5µm CMOS technology, suitable for bio-medical applications. The spacing
between any two coils is 0.9µm, which is the smallest DRC (Design Rule Checking) value
permissible in this technology. By placing digital switches and controllers to the outer sides of
the array, one can stack the coils of the array closely to enable shaping the overall magnetic
field profile and consequently the forces over the array. The system-level block diagram
depicted in Fig. 3.1 consists of four major components: the 8x8 coil array, 3-to-8 row and
column decoders, a variable power bi-directional current source, and the global digital logic
control. First, the global control is responsible for: (a) coordinating the generated magnetic
fields by providing the correct signals to the decoders and the current source, (b) providing
a flexible interface to allow programmable off-chip algorithms for control and manipulation,
(c) fast switching with a period of 25n seconds for pseudo-parallel collaborative or single-coil
manipulation. Second, the row/column decoders and digital logic (a) facilitate single coil
selection (b) minimize interconnect coupling and noise, and (c) lower the overall number
of I/O pins of the package. Third, an eight-level bi-directional current source to generate
different magnetic flux densities and directions for refined maneuvering. The last component
is the actual coil-array where electric current is passed through a pre-selected coil location
to generate an electromagnetic field using on-chip multi-layered inductors.
26

Figure 3.1: System-level block diagram.

3.2.1

Programmable Current Source

NMOS transistors are used to sink currents through a coil. The resistance of the path
(including the coil) as well as the width/length of the transistor set the amount of current
which can be sunk, thus, eliminating the need for a dedicated resistor to set a reference
current. However, the maximum current drawn has to conform with the technology’s current
density limits in order to keep the substrate at room temperature and protect the metal layers
and oxides from overheating and melting.
We have cascaded a set of eight NMOS transistors in parallel to create a binary controllable current sink. A simplified depiction of this configuration is shown in Fig. 3.2. It
operates as follows: a user selects a single coil from the 8x8 array using row and column decoder signals. Next, the direction (polarity) of the current is selected (e.g. counter-clockwise
¯
or clockwise) using S and S. The parallel NMOS block as well as the current direction
27

Figure 3.2: Variable power bi-directional current sink with a simplified coil path. The ’e’
signal is an “enable” signal which can be used to switch on/off the entire DC actuation path.
The components on the left of the dashed line are shared across the entire array to save area.

(polarity) switches are shared across the whole array to reduce real-estate. An eight bit
word en[0:7] is used to adjust the amount of current required for actuation. Fig. 3.3 shows
a post-layout simulation, with parasitic capacitance and resistance extracted, of one of the
programmable current sinks implemented. When the signal S is high the direction of the
current is positive (counter-clockwise for the spiral coil) and vise versa.

The non-linearity of the steps are a result of the parallel NMOS arrangement. The forces
generated from the coil and exerted on the beads are inherently nonlinear, regardless of
these steps, see (2.5). This non-linearity is exploited to create larger forces for the initial
motion/magnetization of the bead then refined to smaller steps as we increase the amount
of current sunk by the parallel NMOS transistors. If linear current sources are desired, one
can carefully resize the parallel NMOS transistors in a custom design.
28

Figure 3.3: Post layout simulation of one of the programmable current sources implemented
on chip. The magnetic field direction can be changed using the positive current stairs or the
negative ones.

3.2.2

Digital Logic

To individually access each coil in the 8x8 array, two 3-to-8 decoders link any inductor load to
the programmable current source. The decoders allow each input to enable one of the eight
rows or columns available. For pseudo-parallel manipulation schemes, a high-speed control
signal (switching up to 40 MHz) is used to link a specific cell to the driving circuity and it
consists of a dedicated transmission gate as well an AND gate, see Fig. 3.4. If cross-talk
is tolerated in an application, a single NMOS switch can be used to minimize the size and
logic of the control circuitry [21]. However, due to the noise and coupling generated by the
neighboring coils, using NMOS switches exhibits a negative influence on the performance of
the array.
In Fig. 3.2, the components on the left of the dashed line are shared across the whole array
to save space on the die. The components on the right are intentionally duplicated for every
single coil in the array. When designing a coil array’s schematic, it might be intuitively
appealing to use a shared ground rail to interconnect all coils. However, it is crucial to
29

Figure 3.4: A dedicated AND gate and a transmission gate are used to isolate signals from/to
each coil in the array.

separate and isolate the ground paths to minimize ground bouncing [57]. The array operates
in a pseudo-parallel scheme using high speed switches which hammer the ground node with
large and fast changing currents. This causes the ground node to move, thus, affecting
the performance of the system with unintentional transients, electromagnetic interference
(EMI) radiations, and inductive coupling with neighboring coils. The main cause of this
repercussion is due to the rapid changing of the magnetic flux (Φm ) in the interconnects.
However, the magnetic flux is directly proportional to the magnetic field passing through a
loop area or a boundary as
A·d ,

Φm =

(3.1)

c

where the magnetic flux Φm is defined as the line integral of the magnetic vector potential A
passing through a surface bounded by contour c traced along an infinitesimal vector element
d . Therefore, in order to reduce the flux, which causes ground bouncing, the path back to
ground should be reduced. As a result, we separated the ground path for each coil using
two dedicated switches. This makes the overall loop area around (A) shrink, causing the
total magnetic flux to be reduced. Additionally, a second measure is employed in the layout
to reduce noise by creating a patterned metal1-poly ground shield [58] beneath each coil in
the array. This reduces coil-substrate coupling and increases the performance of the coil
(namely, the coil’s Q-value).
30

3.2.3

Actuator

A square shape spiraling inductor is selected in this design to meet “Non-Manhattan-shape”
DRC requirements for this CMOS technology. In another coil array design, a similar coil
array is created with different specifications. The side length of a single coil is approximately
80µm and it produces larger forces and utilizes larger DC currents.
To foresee the forces and the magnetic flux densities generated by each coil we used
a Finite-Element-Method (FEM) simulator to precisely model the CMOS substrate, metal
layers, and passivation layers, using the process’ proprietary specifications. We consequently
obtained the most predictive results as depicted in the sample of Fig. 3.5. Here, the XAxis represents the distance from the center of the 30µm x 30µm coil, the Y-Axis is the
magnitude of the magnetic flux density (B). Note that the generated field, without a bead
present on top of the coil, will have two maxima near the coil’s center (shown via the solid red
curve) because the spacing between the metal windings cannot be reduced below the DRC
requirements. However, for distances at and above 2µm (shown via the dash-dotted green
curve), this effect disappears. Consequently, a 2µm silicon oxide and silicon nitride layers,
provided by the fabrication house, are applied. Such layers also serve as a protection covering
for the platform. While the silicon nitride layer is bio-compatible with some experiments,
other alternative chemicals like Parylene-C can also be deposited post fabrication to meet
the application’s requirements. Moreover, when adding a 1µm bead on the surface of the
chip, aligned to the center of the coil, the magnetic flux density is increased and focused
toward the bead. This increase is due to the coupling between the coil and the bead. Also,
the result magnetic field is not perfectly centered because the coil’s winding doesn’t perfectly
end at the center.
31

Magnitude of Magnetic Flux Density (T)

2.2

x 10

3

No Bead, B @ Coil
No Bead, B @ Surface
1µ m Bead, B @ Coil
1µ m Bead, B @ Surface

2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
15

10

5

0

5

10

15

Distance from the center of the coil along the x axis ( µm)
Figure 3.5: FEM simulation of the 30x30 coil’s magnetic flux density at the surface of the
chip and near the metal winding.

32

(0,0)

(15,15)

Figure 3.6: FEM simulation of the coil’s magnetic force generated on various beads using
1.5mA using the 30µm x 30µm coil. The beads were placed on different spatial locations on
the surface of the chip.

To predict the forces acting on the bead, we use different Invitrogen DYNAL R magnetic
bead [36] sizes, which are modeled in the FEM simulator using the physical specifications
as well as the B-H hysteresis curves provided by the manufacturer. Using the simulator,
a current of 1.5mA is passed through the 30µm x 30µm coil generating different magnetic
flux densities. The force acting on the surface of the bead is calculated, using the “virtual
work” method and depicted in Fig. 3.6. The beads were placed on the surface of the chip
(i.e Z=0) and then moved in the X-Y plane. From a top-view perspective, a bead located
at (0,0) is placed at the upper left corner of the coil and experiences the weakest magnetic
force. A bead located at (15,15) is placed at the center of the coil and experiences the largest
magnetic force.
Similarly, for 80µm x 80µm the magnetic flux density is simulated, shown in Fig. 3.7,
33

Magnetic Flux Density Magnitude (T)

0.025
Z
Z
Z
Z
Z

0.02

Distance: 0um
Distance: 0.5um
Distance: 1um
Distance: 1.5um
Distance: 2um

0.015

0.01

0.005

0

3

2

1

0

1

Distance from Coil’s center (m)

2

3
x 10

5

Figure 3.7: FEM simulation of the 80x80 coil’s magnetic flux density at the surface of the
chip and near the metal winding.
using the same FEM simulator. The DC current is 40mA and the largest magnetic flux
around the center is 17mT. The amount of force generated from these coils is much larger
than the previous one. A second simulation is conducted to measure the magnetic force,
shown in Fig. 3.8, on a 2.8µm beads at the surface of the chip. The bead is placed at the
center of the coil and at the upper left corner. The smallest and largest forces at the corner
are (38.86fN at 1mA) and (62.18pN at 40mA) respectively.

3.2.4

Arrays

To target larger forces (up to 1nN) a larger on-chip coil actuator is designed as discussed
in section 3.2.3. The 0.5µm bulk CMOS technology used supports three metal layers. The
34

1

x 10

0.9

9

Bead @ Center
Bead @ Edge

Magnetic Force (N)

0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0

5

10

15

20

25

30

35

40

DC Current (mA)
Figure 3.8: FEM simulation of the coil’s magnetic force generated on 2.8µm bead using the
80µm x 80µm coil. The bead was placed in the upper left corner (edge) and the center
locations on the surface of the chip (i.e. Z=0).

35

Table 3.1: Coupling coefficients and mutual inductances
K=-9.19m K=-35.52m K=-9.19m
M=-10.33p

M=-39.95p

M=-10.33p

K=-38.28m

Center

K=-38.28m

M=-43.06p

Coil

M=-43.06p

K=-9.19m

K=-35.52m

K=-9.19m

M=-10.33p

M=-39.95p

M=-10.33p

upper most two metal layers (metal3 and metal2) are used to create the coil’s windings and
the lower most layer (metal1) is used for routing (e.g. interconnecting the coil’s two ports
to their corresponding switches), see Fig. 3.9. Careful routing techniques are required to (a)
minimize the coupling across interconnects (b) accommodate the largest number of coils in
the array while keeping the distance between any two coils at minimum, namely 0.9µm (c)
shorten the routing path to keep the interconnect resistance at minimum. Finally, there is
a trade-off between the size of the coil, the number of metal layers available in the CMOS
technology, and the maximum current density allowed per via. Thus, one has to balance
between all these parameters to achieve the desired force and array size.
Owing to the fact that the coils are closely stacked, mutual inductances and coupling
coefficients must be considered when building the entire array. In order to quantify these
effects, a 3x3 sub-array snippet is formed of coil sizes of 30µm x 30µm. Table 3.1 lists the
mutual inductance values (M ) and coupling coefficients (K) computed by ASITIC [59] as
seen by the coil in the center of the 3x3 snippet.
Integrating all modules together on the same die requires careful examination in terms
of routing, isolation, and coupling among neighboring coils. Moreover, the modular design
should consider distributing heat equally across the entire array to maintain consistent interactions with the bio-objects of interest. The system, shown in Fig. 3.10, has three different
36

Figure 3.9: Layout of stacked coil array. The array is centered in the middle and surrounded
by digital switches from the left and right sides. The DC current source along with the
thermometer decoder are placed on the top. The digital row/column decoders are placed at
the bottom.

37

Figure 3.10: Photograph of 1.5mm x 1.5mm CMOS die as an open cavity device.
8x8 arrays; each one uses different routing and isolation techniques, DC driving circuitry,
and configuration. This prototype is fabricated in 0.5µm 3-Metal-2-Poly CMOS. It consumes
9mW using a 5v power supply. We used the 1.5mmx1.5mm die to accommodate all three
8x8 arrays. Each coil array occupies 248µm x 248µm and requires a 100µm x 260µm digital
overhead (switches, decoders, etc). A second prototype, also fabricated in 0.5µm 3-Metal2-Poly CMOS, has similar specifications but larger coils, improved grounding and isolation,
and larger DC currents to generate larger forces. This prototype, however, consumes 31mW
using the same power supply voltage running at power level (000) and 200mW when running
at full power (i.e. 111).
Two different experiments are conducted to quantify the forces generated by the 30µm
x 30µm coils and one experiment for the 80µm x 80µm coils. The first experiment is to
test that the coils are capable of generating magnetic forces on micro-scale objects before
38

submergence into fluids. Therefore, we used an Atomic Force Microscopy (AFM) system by
“Nanosurf” called “EasyScan 2 AFM” [60] with a magnetized cantilever (MAGT) [61] fabricated by “AppNano”. The cantilever’s tip height is around 15µm and its radius of curvature
(ROC) is around 35nm. The principle used in measuring the force is called Magnetic Force
Microscopy (MFM) where the stray magnetic field generated by the coil array interacts with
the magnetized tip by moving the flexible cantilever up/down as shown in Fig. 3.11. Using
this technique the magnetic force or force gradients are measured for the coil array using
AC or DC (current) modes. For this experiment, the DC “Static” mode was used in the
measurements.
Measuring a force using a cantilever is based on Hooke’s law where the cantilever acts
like a spring with a constant k and responds to a force F (in equilibrium) as follows

z=

F
,
k

(3.2)

where z is (proportional to) the deflection of the lever. As a result, when measuring the
deflection using the laser sensor, one can directly measure the force generated by the coil
array acting on the tip. Also, the force F can be computed as in section 2.3.2 using (2.5)
but the magnetic moment m is the moment of the cantilever and B is the stray magnetic
field from the coil array.
Because we are using the DC mode, the sensitivity of the measurement is limited by
the deflection sensor’s noise. There is an additional pink noise associated with the type of
the detector being used (e.g. piezoresistive) and the 60Hz power line/fluorescent light noise.
These factors should be considered while studying MFM measurements. Nowadays, most
commercial AFMs come with dedicated filters to eliminate and reduce these noise sources.
39

Figure 3.11: Concept drawing showing how an AFM system can be used to measure the
magnetic field generated by the coil array. The cantilever flexes according the magnitude
of the field generated on the magnetized tip. The optical sensor detects the laser deflection
caused by the movement.

To measure the magnetic force generated by the 30µm x 30µm coils, the cantilever’s tip
is placed on top of a coil, specifically, the upper left corner using a micro-positioner stage.
Next, a 7µm x 7µm region, the maximum area allowed using the current AFM system, is
scanned while keeping the AFM’s Z-controller fixed at a pre-specified height. During the
scan, the deflection of the cantilever is recorded and then multiplied by the spring constant to
produce the force as depicted in Fig. 3.12. In this experiment, the single coil was switched on
when the AFM scan reached half of the 7x7 block. Thus, the right half of the figure (colored
mainly in green) depicts the coupling forces between the cantilever’s magnetic tip and the
coil (while it was switched off). In contrast, the left half of the figure (colored mainly in
blue) depicts when the coil is switched on and exerting a magnetic force on the cantilever’s
tip, which caused it to deflect forward.
Second, a lab experiment is conducted to validate the transport forces acting on a single
bead in the X-Y plane. The setup is very similar to the one shown in Fig. 5.3. While the
40

Figure 3.12: Magnetic Force Microscopy measurement of a 7µm x 7µm region on top of the
coil. The coil was switched on in the middle of the scanning process.

41

30µm x 30µm coils are switched off, small droplets (5µL) of a diluted 1µm Tosylactivated
magnetic beads solution are inserted in the PBS solution using a pipette. Then a small (1
minute) grace period is observed in order for the beads to settle in the fluid or on the surface
of the chip. Using a microscope, one visually locates a magnetic bead on the 8x8 array
and programs a specific coil to attract the bead toward it. Fig. 3.13 shows multiple frames
taken from a video sequence while a bead is being manipulated to the right. In frames 4,
and 5 the bead is attracted closer to the surface by switching on the coil to the left. In
the remaining frames, the bead is transported to the right by switching off the coil on the
left and switching on the coil on the right. The total distance traveled in this experiment
was 16µm in 3.3 seconds, which is an average velocity of 4.85µm/s. The magnetic force
generated in this experiment cannot be compared with the simulations conducted in section
3.2.3 because the bead did not fully move to the center of the coil. This is due to several
possible factors: The X-Y transitional forces are not strong enough to influence the bead,
the bead found a preferred chemical binding site and attached itself to it, or the equilibrium
forces acting on the bead are not maximized at the center of the coil. This latter possibility
is caused by the fact that the main vdd/gnd rails (power supply and ground rails), which are
exposed, are contributing to the transition. Careful vdd-gnd shielding is required to alleviate
this influence.

The last experiments are performed on the 80µm x 80µm coil array. The setup is very
similar to the previous experiment except that the medium used here is deionized water, and
the beads’ size is 2.8µm in diameter with Protein-G coating. The 2.8 micron beads have
more iron content than the one used in the previous experiment.
42

Figure 3.13: A montage of a series of video frames showing how a magnetic bead is attracted
(downward) toward a coil then moved to the right by switching on the neighboring coil.
3.2.4.1

Bead Transport

A proof of concept of bead transport is demonstrated in a supplementary video at (http://
www.msu.edu/~asd1815/videos.html titled [on transport]) and as a montage in Fig. 3.14.
Initially, the beads are randomly moving on top of the coil array then the upper right coil is
switched on. The beads that lie above the coil immediately move to the center, and the beads
at the neighboring coil are transported to the right. Next, the upper right coil is switched off
and its neighbor is switched on. The beads that aggregated on top of the upper right coil are
now transported to the neighboring coil. Note, that a collection of beads attached to a debris
lying on top of the third coil, from the right, is also transported. Lastly, the neighboring
coil below is switched on and the beads are transported to the next row. The forces in this
experiment were chosen to affect the immediate neighboring coils only. If the bead doesn’t
lie in that area then it would not be transported. Finally, a bead lying at the edge of the coil
reached the center in 1.75s. Therefore, the average velocity is approximated to be 22.8µm/s
and the average magnetic force is 602pN. The debris’ average velocity from the third coil to
the second is approximately 2.1µm/s and the average magnetic force is 55pN. These forces
43

are comparable with the simulation results provided in Fig. 3.8, i.e. the bead lying at the
edge initially faces a 62pN magnetic force and quickly reaches 940pN. Therefore, on average
a bead on the edge of the 80µm x 80µm coil will approximately encounter an average of
600pN.

3.2.4.2

Pseudo-parallel Manipulation

In addition, we demonstrate more abilities of these platforms in another video at (http:
//www.msu.edu/~asd1815/videos.html titled [on msu]) and as a montage in Fig. 3.15.
The movie clip demonstrates two features: (1) pseudo-parallel manipulation; where the beads
are collectively used to draw the letter “M” simultaneously. This parallel movement of the
beads is a result of careful coil-array layout and isolation, dedicated high-speed switches,
and precise magnetic forces that do not span beyond the pre-specified spatial location. (2)
Pre-defined bead diffusion delay, which allows the user to program the array to wait for a
specified amount of time for the beads to diffuse in order proceed with the next operation.
This feature is observed after drawing “M” and before drawing “S”, and after drawing “S”
and before drawing “U.” Algorithm 1 explains how the coil array is used to draw the letters
in the word MSU. The switching speed for this experiment was chosen to be 16MHz, the
byte codes (e.g. LT RM ) represents the spatial location of the coil to be switched on. The
length of the byte code represents the total number of coils to be switched on in pseudoparallel fashion. For example, to draw the letter “M” 21 coils in the array are activated
simultaneously. If the coil location is present in the byte code then it will be switched on
and actuated “REFRESH RATE” times to trap the beads for a longer period of time in
that position. Once the letter “M” is drawn the, procedure “CALL DIFFUSION DELAY”
is executed to allow the beads to diffuse before drawing the next letter.
44

Figure 3.14: A montage of a series of video frames showing how magnetic beads are transported from one location to another.

45

Figure 3.15: A montage of a series of video frames showing how magnetic beads can draw
the letters M S U sequentially using pseudo-parallel scheme.

46

Algorithm 1 Draw MSU On-Chip
DEFINE LT RM [21] = 0xC0, 0xF 0, 0xC1, 0xC9, ...
DEFINE LT RS [30] = 0xC8, 0xD0, 0xD8, 0xE0, ...
DEFINE LT RU [20] = 0xC8, 0xF 0, 0xC9, 0xF 1, ...
DEFINE REFRESH RATE = 100
DEFINE REFRESH DELAY = 2ms
DEFINE SWITCHING FREQ 16MHz
if LET T ER = M then
LOOPCOU N T ⇐ LEN GT H(LT RM )
BY T E 2 OU T P U T ⇐ LT RM
else if LET T ER = S then
LOOPCOU N T ⇐ LEN GT H(LT RS )
BY T E 2 OU T P U T ⇐ LT RS
else if LET T ER = U then
LOOPCOU N T ⇐ LEN GT H(LT RU )
BY T E 2 OU T P U T ⇐ LT RU
end if
while LET T ERS 2 DRAW = 0 do
for REF RESH = 0 → REF RESH RAT E do
for i = 0 → LOOPCOU N T do
OU T P U T ⇐ BY T E 2 OU T P U T [i]
end for
end for
CALL DIFFUSION DELAY( )
end while

47

Unlike the previous experiment on the 30µm x 30µm array, the one on the 80µm x
80µm array does not suffer from weak bead transitions because: (1) there are two dedicated
switches per coil as shown in Fig. 3.2 whereas the previous experiment used a single switch
with the second coil node (i.e ground node) was directly shared across the entire array, (2)
the vdd/gnd rails are shielded using auxiliary metal layers and placed as far as possible from
the array, (3) the amount of DC current is increased in this experiment to generate larger
magnetic forces in order to move and trap magnetic beads at the center, and (4) the medium
(nonionized water) and magnetic bead coating (Protein-G) reduced the amount of DLVO
forces between the beads and the surface of the chip.

Another a lab experiment is conducted on the 6x6 micron coils to validate the translational forces of a bead in the X-Y plane. Before conducting the experiment, the bonding
wires between the CMOS die and plastic package are coated with Parylene-C because of
its biocompatibility, insulation capabilities, and cost effectiveness [62]. Also, it serves as
protection layer on the surface of the chip from a wide range of chemicals. The Parylene deposition process took 1 hour 31 minutes and 40 seconds. The Dimer mass was 3.1348 grams.
Next, while the coils are switched off, small droplets (5µL) of a diluted 2.8µm Protein-G
Dyna beads solution are injected in the deionized water solution, on top of the chip, using a
pipette. Then a small (20 seconds) grace period is observed in order for the beads to settle
on the surface of the chip. For testing purposes a microscope is utilized so that the user
can visually locate a magnetic bead on the array. Nevertheless, to improve the clarity and
focus of the microscope a Corning glass coverslip is placed on top of the plastic package
to prevent liquid surface tension, evaporation, and vibration. Next, the user programs a
specific coil to attract the bead toward it. Fig. 3.16 shows two frames taken from a video
48

Figure 3.16: A micrograph of the coils closely stacked with three magnetic beads on top.
The first frame on the left shows a bead “1” traveling to the highlighted coil when it is
switched on. Bead “2” is also influenced by the movement and the magnetic field in the coil
and in the interconnects. (50x magnification)

sequence while a bead is being manipulated. Initially the highlighted coil is switched on,
namely, row=5 and column=2. Then the bead labeled “1” moves diagonally to reach the
center of the highlighted coil after 11 seconds. The total distance traveled by bead “’1” is
7µm in 11 seconds. It is noted that the bead labeled “2” is also influenced by the magnetic
field generated by the coil and possibly the local interconnects, however, the magnetic forces
acting on bead “2” are relatively weak. To minimize such interconnect influences between
coils in the same rows and columns in the array, we have shielded the array, when possible,
using the top most metal layer i.e. metal3.

Finally, there is a trade-off between the size of the coil, the number of metal layers
available in the CMOS technology, and the maximum current density allowed per via. Thus,
one has to balance between all these parameters to achieve the desired force and array size.
49

3.3

AC Manipulation

The operational concepts of the AC coil array can be explained as follows. Using a technique analogous to switching regulators (switched-mode power supplies), the average output
current of the amplifier can be controlled by switching the ring oscillator on/off at select durations. The durations are specific intervals that contain different fractions of the oscillation
periods to provide different DC components. Thus, the DC magnetic field, and correspondingly the DC force, can be selectively commanded to generate forward (positive) or backward
(negative) actuation on magnetized objects. In one targeted application, namely actuating a
magnetic bead in fluid, it is anticipated that the high frequency components would be filtered
out by the fluidic medium and primarily low and DC components would contribute to the
actuation of the bead. Thus non-contact actuation of magnetized beads is easily achieved
with the current integrated platform. The actual strategies of switching operations for best
(sensing and/or actuation) performance is relegated to software in order to gain flexibility
for broad choices of the sensing and control space.

3.3.1

Class E Power Amplifier

The class E power amplifier (PAE) [63] provides an efficient design of the driving engine for
the array. In order to increase the amplifier’s overall power output and to digitally control
it, one replaces the single NMOS switch with a parallel NMOS configuration [64]. Fig.
3.17 shows the parallel NMOS switches (we use minimum size W/L). The eight inputs are
connected to a 3-to-8 thermometer decoder to reduce the number of I/O pins required to
control the PAE, and to facilitate binary-word increase/decrease in magnetic field actuation
[65]. In Fig. 3.17, RFC represents a large on-chip inductor that provides a DC path from
50

Figure 3.17: A variable gain class E power amplifier tuned at 900MHz. The dashed box is
a simplified version showing how the same coil can be shared for sensing and actuation. In
actuation mode, the PAE is connected to a single coil, the other end of the coil is connected
to ground.

51

Figure 3.18: The smallest current passing through the actuator during a regular switching
period. The peak-to-peak value is about 150µ. In this view, two actuators are turned on
and off (only one is shown) while the rest of the array is completely shut off.
the supply and it is approximated as an open circuit at RF. The inductance L2 is designed
to be 30nH and C2 is determined such that the PAE is tuned to 900MHz. Using the lowest
output power, the PAE generates 150µA (peak-to-peak) when driving a single coil actuator.
In Fig. 3.17, RFC represents a large on-chip inductor that provides a DC path from the
supply and it is approximated as an open circuit at RF. The inductance L2 is designed to be
30nH and C2 is determined such that the PAE is tuned to 900MHz. The smallest current
generated is 150µA (peak-to-peak, see Fig. 3.18) when the PAE is driving the coil actuator.
In this case, only one transistor is being switched ON and OFF, the other 7 transistors are
completely OFF. The PAE output spectrum is depicted in Fig. 3.19.
A simple ring oscillator, tuned at 900MHz, is chosen in this design to reduce the complexity of the system and minimize real-estate. Ring oscillators are known to fluctuate with
temperature changes and drift with time, therefore, if necessary, one can replace it with a
Phase-Locked-Loop synthesizer. This platform is intended for temperature controlled bio52

Figure 3.19: PAE output spectrum tuned at 900MHz. In order to minimize harmonic distortion outside the band of interest, the actuator impedance as well as the LC filter tank
associated with the PAE are tuned to 900MHz

materials where the chip is assumed to operate around room temperature. The number of
stages (gates) is chosen to be three, and the W/L sizes of each gate are adjusted to obtain
an average delay of 0.10422ns tolerant to different variations in the fabrication process. The
output frequency is given approximately as

f=

1
2 ∗ N ∗ Td

(3.3)

where f is the frequency, N is the number of stages and Td is the (average) time delay per
stage. Accordingly, f is calculated to be approximately 900MHz.
The Class E Power Amplifier described above is controlled by a 3-to-8 thermometer
decoder, where the smallest possible input ‘000’ translates to ‘00000001’ output. This keeps
the PAE turned on at all times unless the oscillator is switched off. Consequently, we have
replaced the inverter in the first stage of the oscillator with an NAND gate, as shown in Fig.
53

Vdd

Enable

Output
Enable

gnd

Figure 3.20: Min-sized, three stage ring oscillator with custom W/L to create an average
delay of 0.10422 ns.
3.20, to allow switching off the oscillator and consequently, the PAE. Fig. 3.21 shows the
average (DC component) of the periodic output of the PAE as a result of different switching
intervals.

3.3.2

Digital Logic

Each cell in the 16x16 array needs to be accessed individually. Therefore, two 4-to-16
decoders are used to selectively link any inductor load to the driving circuits (the cross
coupled LC oscillator and PAE). The decoders operate as in Random Access Memory (RAM)
designs. Each input enables one of the sixteen rows and columns available. In order for the
driving circuits to access a single coil cell, a high-speed control signal (switching at 40 MHz)
is used to link a specific cell to the driving circuity. A dedicated switch consisting of 10
transistors is used to connect and isolate each coil in the array. The switch is built using an
AND-gate for column/row selection and a tranmission-gate to pass the input/output signals.
54

Average Output Current

4

x 10

5

2
0
2
4

2

4
6
Power Levels

8

20ns
21ns
22ns
23ns
24ns
25ns
26ns
27ns
28ns
29ns

Figure 3.21: The average output current for different switching periods (20ns to 29ns). The
magnitude is maximum at 28ns and 21ns switching periods.
If cross-talk and wire-loading is tolerated in an application, a much simpler approach
can be adopted to minimize the size and logic of the control by replacing the 10 transistor
switches with a single NMOS transistor. However, due to the noise and coupling generated by
the neighboring cells, using NMOS switches exhibits negative influence on the performance
of the array. Fig. 3.22 shows the performance of the 10-transistor transmission-gate switches
in minimizing the cross-talk. In this simulation, a 3x3 block in the array was monitored.
The center cell in the block is switched ON/OFF every 250 ns while all its 8 neighbors are
switched OFF. The peak-to-peak current for the 8 neighboring cells is negligible (shown in
red in the Figure) compared to the current of the center cell (shown in blue).

3.3.3

Actuator

The magnetic field generated in the array is proportional to the size of the inductor, the
current passing through it, and the medium surrounding it. In this paper, the on-chip
55

Figure 3.22: The center cell labelled L5 (blue curve) is being switched on and off every 250ns.
The green curve depicts the logical switching activity on the transmission-gate switch and
the other curves (shown in red) represent the currents going through the 8 neighboring coils.
The magnitude of the current going through L5 dominates over the rest, in contrast of using
a single switch which produces much more noise (not shown).

inductors were prototyped using all available 3 metal layers to constitute the smallest possible
size in OnSemi 0.5µm CMOS technology. Each coil measures 6x6 microns in the plane
with an inductance of 14.42495 pH and a resistance of 0.52µ Ohms. Fig. 3.23 depicts
the tiny inductor. The 6x6 microns inductor was simulated to ensure the desired range
of magnetic field output. The magnetic flux density generated is characterized (Fig. 3.24)
with the assistance of an EM simulator (Ansoft Designer 4) using the proprietary technology
parameters to ensure precision. The simulation shows that using the largest possible PAE
current produces a magnetic flux close to 0.4mT.
Furthermore, a test circuit was fabricated to quantify the ability of these tiny inductors
to manipulate magnetized objects. The test circuit consists of a 3-to-8 thermometer decoder
controlling a PAE driving a mini-sized 6x6 micron coil. The passivation layer covering the
coil has been removed to allow precise measurements. To measure the generated magnetic
56

Figure 3.23: Three layered tiny coil 3D model. Each metal layer is designed as a “U” shape
object.

2.5634e-7
6.1650e-4
1.4088e-3
Figure 3.24: [3D Side view] Simulated magnetic field shown in green/blue as a result of
passing 1.5mA through it (3 metal layers highlighted in pink). The B-field near the surface
is approximately 0.4mT

57

Generated Force (N)

3

x 10

7

2.5
2
1.5
1
0.5
0
0

2

4

6

8

PAE Power Levels
Figure 3.25: The forces (in Newton) generated on the cantilever tip by a single on-chip
coil using different power levels. The error bars represents the variation in spring constants
provided by the cantilever’s fabrication house.

field of the fabricated coil, we have used an Atomic Force Microscopy (AFM) system with
a magnetized tip (MAGT) fabricated by AppNano. The cantilever’s tip height is around
15µm and its radius of curvature (ROC) is around 35nm. The cantilever’s tip was placed on
top of the “exposed” test chip.
The magnetic field of the coil has been detected by recording the measured deflection
of the cantilever using the AFM optical laser sensor. The force exerted on the cantilever’s
tip is then computed. For each power level, we operated the AFM system to conduct 256
measurements which were then averaged. Therefore, a total of 2048 experimental runs were
executed to measure 8 different (cantilever deflection) force levels. The experiment runs were
separated by a 5 second timeout period to eliminate possible correlation between different
driving power levels. In Fig. 3.25, the middle curve depicts the nominal magnetic fieldgenerated forces for the 8 driving power levels.
58

3.3.4

Arrays

Integrating all components together on the same die requires careful considerations in terms
of routing, isolation, and coupling among neighboring coil inductors. Moreover, careful
considerations are due in terms of placement of RF and digital circuits with noisy sources,
forming the array shape according to the targeted applications. The modular design should
consider distributing heat equally across the entire array to maintain consistent interactions
with the bio-objects of interest. Future versions will include on-chip temperature sensors.
The overall system, shown in Fig. 3.26, has 3-degrees of flexibility: the shape of the array,
different magnetic field levels per cell, and the array control scheme e.g. spiral, pyramidical,
etc. Accordingly, with this flexibility, one can create collectively an arbitrary magnetic
potential surfaces for manipulation. For example, it might be desired to generate a bowl
shaped magnetic field using a square array. To that end, one can program the digital
controller to set the edges of the array at maximum power ‘111’ and reduce the power
linearly along the X and Y axes in a gradient fashion toward the center of the square. More
elaborate details on these schemes are the subject of future reports. As the switching control
speed is about 40 MHz, time sharing is in fact used to power the desired configuration via
multiplexing digital control, thus, requiring a single cross coupled LC oscillator for sensing
and a single PAE for actuation.
A prototype array was fabricated in OnSemi 0.5µm 3M2P technology. For the first
conservative design, we used the 1.5mmx1.5mm size for an array of 16x16 inductor cells. For
different design rule CMOS technologies, one may adjust the design easily to accommodate
the specific design rule constraints in selecting the small inductors and the corresponding
arrays. The prototype is an asymmetrical 16x16 array suitable for singular and collaborative
59

Figure 3.26: System block diagram. The global control is responsible for managing all
building blocks to create the desired sensing or manipulation profile.

object manipulation with an estimated power consumption of 67mW. The distance between
the center of any two neighboring cells on the same row is 93.450µm and the distance between
two neighboring cells on the same column is 78.00µm. The array filled out the entire upper
half of the die as shown in Fig. 3.27. The lower portion of the design, contains the PAE and
digital control circuitry. The LC oscillator was not included in this run. The components
have been intentionally separated and isolated to reduce coupling (cross-talk) and improve
the performance of the actuation. The control logic can be applied directly to the chip from
an external device using the I/O pins. Furthermore, the pad-frames on the upper half of the
die have been removed to minimize the noise generated from the bonding wires and to allow it
to be accessed from the upper half (for example via an Atomic Force Microscopy for testing)
without damaging or shorting the I/O pins. The digital circuity and switching overhead
60

Figure 3.27: An annotated picture of the fabricated die: The RF components are placed far
away from the array to reduce substrate injection and coupling. The I/O pads are removed
from the 3 sides to permit easier testing.

occupies 51.450µmx34.5µm for this configuration. Finally, for future designs, the size of the
actuators can be increased to increase magnetic fields to accommodate larger objects and
beads. Accordingly, the driving circuitry and the PAE would need to be adjusted to properly
to drive the new inductor loads.
A second design includes a symmetrical 16x16 array with some improvements. The
difference between the two layout techniques is shown in Fig. 3.28. The digital overhead
is reduced and occupies 38.4µmx32.7µm only. This is due to using a redesigned switching
circuitry with a more compact layout. Another enhancement is that the overall area of
the array is now 542.4µmx536.7µm which is 36% smaller than the one above. Finally,
the distance between any two neighboring actuators (vertical or horizontal) is 33.6µm. This
61

Figure 3.28: Comparison between two layout topologies.
compact symmetrical design allows easier controller programming, smaller die size and equal
heat distribution. The power consumption for this array is estimated to be 60mW. Finally,
it is possible to modify these arrays and move all the digital components to the die sides by
creating a wide bus for the interconnects. This permits the coil array to be closely stacked
and compact. However, careful component isolation to minimize coupling and noise would
be challenging.
Last, a third design (shown in Fig. 3.29) includes a symmetrical 6x6 array with some
improvements. First, the digital circuitry is moved to the side and the coils in the array are
stacked as close as possible. Second, a redesigned digital switching circuitry is implemented
with a more compact layout. The entire digital block occupies 200µm x 180µm. Third, the
top most metal layer is used as a shield across the array to prevent stray magnetic fields
from interfering with the manipulation process. Also, a poly layer is used to reduce substrate
coupling with the coil array. Finally, the distance between any two neighboring actuators
(vertical or horizontal) is 6.3µm. This compact symmetrical design allows easier controller
programming, smaller die size and equal heat distribution. The power consumption for this
array is 7.25mW in actuation mode and 1.9mW in sensing mode.
62

Figure 3.29: A micrograph snippet of the 6x6 array using 6x6 microns coils. The oscillator,
PAE, and DC source are not shown. (20x magnification)
The shape and size of the array can be adjusted to meet the requirements of the application of interest. With the spatial array designs, one can create variable magnetic field
profiles to enable steering one or multiple beads floating in liquid above the surface of the
die to a final spatial destination. As an example, one may program the global controller to
set the center coil of the array on maximum power ‘111’ and reduce the power along the X
and Y axes in a gradient manner toward the edges to create a pyramidal field shape. In this
manner, if the control strategy steers a bead to a maximum magnetic field, then the bead(s)
will be steered to the center of the array.

3.4
3.4.1

RF Sensing
Cross Coupled LC Oscillator

CMOS inductors’ quality factor are very limited and since the linewidth of the impedance of
the LC tank is determined by the quality factor, it is not possible to measure the impedance
to achieve the sensitivity required for bio-sensing. However, for on-chip LC cross coupled
63

oscillators the phase noise profile[18] will be significantly compressed compared to the tank
impedance due to the virtual damping phenomena[66]. Therefore, a cross coupled LC oscillator is adopted to detect small magnetized particles. The active MOS components and the
varactors (symbolized as variable capacitor) in Fig. 3.30 are shared across the entire array
to save real-estate. The analog multiplexers facilitate the selection between actuation and
sensing modes. Using the output frequency of the cross coupled LC oscillator, it is possible
to monitor the change in resonance frequency when a magnetized particle in the order of
1µm is placed on top of the coil of interest. The method used in detection relies on the
effective inductance of the vertical coil and its Q-value. For normal operations, the oscilla√
tion resonates at frequency f = 1/(2π LC1 ) = 900MHz. Nevertheless, when introducing
a new medium, e.g a magnetic bead, on top of the vertical coil, its effective inductance is
going to increase – resulting in a loss of resonance or a decrease in the oscillating frequency.
Therefore, a simple integrated frequency counter is used to detect that change.
The 6x6 microns vertical inductor of 14.42pH requires a 2.168nF capacitor to oscillate
at 900MHz (for this technology a simple poly-poly2 capacitor occupies 0.887 fF/µm2 ) which
is quiet large. Therefore, a custom poly-poly2 capacitor, appropriately sized cross coupled
MOS transistors, and a tank of digitally controlled varactors are constructed for the LC
oscillator. The W/L of the MOS transistors are adjusted for valid parasitics capacitance
values. Also, the varactor tank is digitally controlled using 8-bits to calibrate the oscillator for different sensing paths. Moreover, the number of I/Os are reduced using a 3-to-8
thermometer decoder.
To verify the sensing abilities of the coils, we incorporate the Invitrogen MyOne Dynabead’s model, using its physical specifications as well as the B-H hysteresis curves provided
64

Figure 3.30: Cross coupled LC oscillator. The output frequency of the oscillator is also tuned
at 900MHz. The variable capacitor is implemented as a varactor bank.

65

Figure 3.31: Computed coil’s inductance based on the location of the bead in the X-Y plane.
The coil is 6x6 microns, its center is located at (0,0) in the X-Y plane.

by the manufacturer, into an EM simulator (Ansoft HFSS). Then, we simulate the vertical
coil with the bead suspended in distilled water. The simulation is swept over the spatial
coil surface from (-12,-12)µm to (12,12)µm, in 3µm steps. The EM simulator subsequently
recomputed the inductance of the vertical coil depending on the position (distance) of the
bead with respect to the center of the coil. Fig. 3.31 shows that the maximum inductance
occurs when the bead is located just above the coil. To verify this behavior another higher
resolution simulation was carried out where the bead was re-positioned from (-3,-3) to (3,3)
in 1µm steps and the maximum inductance was found at (-2,-2) as shown in Fig. 3.32. The
simulations took one week to complete using Dual Intel Xeon X5460 3.16GHz 64bit with
64GB of RAM. Several simulations have shown that the oscillation frequency decreases below
900MHz (in shifts ranging from 0.5MHz to 5MHz). Such shift ranges are easily detectable
by the frequency counter.
66

Inductance (pH)
16.8
3
16.7
2
16.6
1

16.5
16.4

1

Y

0

16.3
16.2

2

16.1
3
3

2

1

0

1

2

3

X

Figure 3.32: Zoomed coil’s inductance just above the tiny coil. The location of the bead in
the X-Y plane. The coil is 6x6 microns, its center is located at (0,0) in the X-Y plane.
The main advantage of this sensor is its relative small size compared to the sensed object
of interest. However, due the fact that the Q value of this inductor is very low, the sensitivity
is negatively affected. Therefore, another larger sensor adopted and is depicted in Fig. 3.33.
In this design, the upper most metal layer (Metal3) is used for creating the spiral winding,
Metal2 is used as an I/O bridge, Metal1 is used a ground, and Poly is used as a ground
shield. Using patterned ground shielding we improved the Q value of the inductor by 2
folds. The resistance, Q, inductance plot is shown in Fig. 3.34. The circuit level model is
shown in 3.35 and values of each component is expanded in table 3.2. This table summarizes
most of the inductors used in this work. All parameters are self explanatory, however, SRF,
is the self-resonance frequency of the coil.

67

P2
P2
P1
P1

Figure 3.33: Layout of improved sensing on-chip inductor. Three metal layers were used.
Metal3 for winding, Metal2 for exit bridge, and Metal1 for grounding.

68

7.54

L vs. Freq

900

Q vs. Freq

R vs. Freq

57.8

(Ohm)

900MHz,7.53nH

Y0 [m ]

(nH)

900MHz,740.3m
M

900MHz,57.52Ohm

57.2

7.52
400

100
7.5
0

1.0

0

Freq. (GHz)

1.0

56.6
0

1.0

Figure 3.34: Inductor characteristic graphs. Inductance vs. Frequency, Q-value, versus Frequency, and Resistance vs. Frequency.
69

PORT1 ls1
coxp1
csbp1

cp
ls11 rs11 rs22 ls22
rs1

ls2

rs2
coxt1
csbp2
rsbt1
GND

rsbp1
csbt1

PORT2
coxp2
rsbp2

Figure 3.35: Schematic of the circuit level model of the inductor. Enhanced PI-model.
The phase noise and oscillation are simulated using Period-Steady-State Analysis (PSS)
and the results of the improved phase noise is shown in Fig. 3.36. The oscillation range is
894.8MHz when the varactor input is 0000 0000 and 906.9MHz when the varactor input is
1111 1111. The frequency ranges are shown in 3.37. The startup time for the oscillator is
estimated to be 13ns.

70

Table 3.2: Inductors used in this work and their parameters
Parameter
Turns
Inner Radius
Metal Width
Spacing
Frequency
Inductance
Q-Value
SRF
Max Size
Top Metal
Shield/GND
cp
ls1
rs1
ls11
rs11
ls2
rs2
ls22
rs22
coxp1
coxt1
coxp2
csbp1
csbt1
csbp2
rsbp1
rsbt1
rsbp2

Inductor1
11.75
3.6µm
3.1µm
1.4µm
900MHz
7.675nH
1.07
8.97GHz
113.5µm
Metal3
TRUE
300pF
3nH
1KΩ
3nH
1KΩ
3nH
1KΩ
3nH
1KΩ
500fF
500fF
500fF
20fF
20fF
20fF
1KΩ
1KΩ
1KΩ

Inductor2
10.75
18.6µm
1.8µm
0.9µm
900MHz
10.07nH
0.379
14.27GHz
95.5µm
Metal2
FALSE
1.74fF
5.08nH
111.9Ω
856.3nH
217.6Ω
5.08nH
111.9Ω
856.3nH
217.6Ω
18.1fF
40.26fF
22.16fF
7.41fF
16.48fF
9.06fF
1.5KΩ
678KΩ
1.23KΩ

71

Inductor3
9.5
15.5µm
1.8µm
1.2µm
900MHz
7.08nH
0.328
18.07GHZ
89.3µm
Metal2
FALSE
1.68fF
3.54nH
91.33Ω
0.587nH
177.66Ω
3.53nH
91.3Ω
0.587nH
177.66Ω
15.75fF
35.3fF
19.54fF
6.88fF
15.42fF
8.54fF
1.63KΩ
729.63Ω
1.32KΩ

Inductor4
4
63.4µm
15.1µm
1.5µm
900MHz
5.5nH
6.539
11.47GHz
277.35µm
Metal3
TRUE
2.52fF
990pH
2.1Ω
126pH
2.02Ω
990pH
2.1Ω
126pH
2.02Ω
31.1fF
83.3fF
24.4fF
13.3fF
24.3fF
7.05fF
714Ω
390Ω
1.34Ω

Inductor5
2.5
54.9µm
20.0µm
2.0µm
900MHz
2nH
5.337
26.28GHz
243.3µm
Metal3
TRUE
2.52fF
959pH
2.07Ω
134pH
1.96Ω
959pH
2.07Ω
134pH
1.96Ω
32.4fF
86.8fF
25.4fF
13.4fF
24.69fF
7.15fF
705Ω
384Ω
1.32KΩ

3

V+ & V-

Output Noise
Phase Noise
50

V+

50

2

V (V)

1

V/sqrt(Hz) (dB)

0

-1

150

-2

-3

100
V/sqrt(Hz) (dBc/Hz)

100

.25

.5
.75
Time (ns)

1.0

1.25

.

3
10

150

5
6
4
7
10
10
10
10
Relative Freq. (Hz)

8
10

Figure 3.36: PSS simulation result. The phase noise and output noise are shown on the right figure.

72

frequency (MHz)

2.5

V (V)

2
1.5
1
.5
5

0

10

15

Time (ns)

20

0

25

Freq. when varactor input is 0000 0000

3

5

10

15

20

25

5

10

15

20

25

Time (ns)
V+, Transient response
when input is 0000 0000

2.5
2
V (V)

950
900
850
800
750
700
650
600
550

Freq. when varactor input is 1111 1111

frequency (MHz)

950
900
850
800
750
700
650
600
550

V+, Transient response
when input is 1111 1111

1.5
1
.5

0

5

10

15

Time (ns)

20

25

0

0

Time (ns)

Figure 3.37: Transient response when the varactor is digital set at maximum and minimum capacitance.

73

A test circuit is fabricated to validate the simulation results. For the 6x6 coils the LC
oscillator failed to start. The possible reasons of this failure are: it might be due to the
very small Q-value of the 6x6 inductor, or it could be the way the coils are arranged in
the array which might have caused the resistance of the path to the coil to become larger
than the negative resistance of active MOS transistors. A second test circuit has another
alternative sensing site (coil) with a dedicated LC to verify different oscillation schemes.
The output of the alternative LC oscillator is connected to integrated counter which divides
the high-frequency output of the LC oscillator down to 4.5MHz as shown in Fig. 3.38. The
chip was running at VDD=1.8v and consumed an average of 1.5mA while in resonance. The
peak-to-peak output of the oscillator is 1.3v. The output of the integrated counter is fed to a
logic analyzer where it translated signals above 0.7V to logical one and below 0.7V as logical
zero. A total of 4096 samples are recorded by the logic analyzer using a sampling frequency
of 200MHz. The frequency spectrum of the output is shown in Fig. 3.39. Therefore, there
are three possible states for the LC oscillator (a) oscillating at the resonance frequency (b)
oscillating at a lower resonance frequency (c) no oscillation at all.

The detection approach can be summarized as follows (i) initially the entire chip is
submerged in liquid. (ii) all coils across the array are sequentially switched and connected
to the cross coupled LC oscillator for calibration. The varactors are then adjusted using
the 8-bit digital control to tune the resonance frequency for the entire array based on the
counter’s output. (iii) when a bead lands near a coil, a shift in the resonance frequency will
occur as seen in figure (3.39), curve labeled “Bead Present.” (iv) alternatively, when a bead
lands on top of the coil the oscillation might stop and the output of the oscillator would
be a DC value close to VDD e.g. 1.8v. In general, the detector will output a logical “1”
74

Figure 3.38: On-chip counter output using an oscilloscope.

0.6
Bead Present
No Bead Present
0.5

|Y(f)|

0.4

0.3

0.2

0.1

0

5

10
15
Frequency (MHz)

20

25

Figure 3.39: The blue curve represents the output when the chip is submerged in distilled
water with no beads present. When a bead comes nearby the sensing sight a downshift of
resonance frequency is observed. When the bead lies on top of the sensing site the oscillation
is lost (not shown).

75

when there is no oscillation i.e. a bead exists above the coil and a logical “0” when there is
oscillation. In future implementations, the correlation in frequency shifts across neighboring
coils can be utilized to extract additional high-resolution information about the exact spatial
location of the bead in the entire array.

3.4.2

Digital Logic

The digital logic used for sensing is the same logic used for the DC and AC modules. The high
speed switches are utilized to connect any coil in the array to the sensing tank. A digitally
controlled varactor is used to calibrate the resonance frequency when the device starts up.
The varactor is made up using back to back PMOS transistors with binary increasing widths.
A 3-8 thermometer decoder is used to adjust the capacitance needed for the tank to oscillate
at a certain frequency. The resistance of the path is shown in Fig. 3.40 and it has to be
compensated by a negative resistance using the active devices (NMOS, PMOS) in the cross
coupled configuration.
The negative resistance for a single transistor can be approximated using first order
square law equations. Or it can be simulated using Level=49 transistor SPICE Models. The
transconductance (gm) of a PMOS device with Id = 1mA is shown in Fig. 3.41. There are
many parameters to adjust in order to get proper oscillation. For example, the transistors
has to be sized (W/L) correctly to guarantee stable oscillation. Using the DC simulation in
the figure a designer has to calculate the total resistance of the tank (positive + negative).
The net resistance has to be more negative for stable operations.
The value of gm is usually chosen to be 1.5 to 3 times [67, 68] more than the calculated
or simulated values. Fig. 3.42 depicts the tanks conductance vs. three differently sized
76

55
50

Resistance (Ohms)

45
40
35
30
25
20
15
10
0

0.5

1
Frequency (GHz)

1.5

2

Figure 3.40: Tank resistance in ohms which needs to be compensated for using the active
devices. This resistance includes the digital switches and the varactors.

7

Transconductance (S)

6
5
4
3
2
1
0
0

50
100
150
Width Multiplier (Initial Width = 1.5µ m)

200

Figure 3.41: A DC sweep simulation of a PMOS device where the width of the device is
incremented while keeping the source-drain current fixed. VDD is 3 volts.

77

0.25

Conductance (S)

0.2

Tank & Aux
M=2K
M=3K
M=4K

0.15

0.1

0.05

0 8
10

9

10
Frequency (Hz)

10

10

Figure 3.42: A simulation of 3 differently sized NMOS transistors to find the W/L ratio
which will create a proper gm to overcome the inductor’s.
NMOS transistors. The variable M represents the multiplier used in the simulation for the
transistor. For example, if M=1000 then the actual width of the transistor is 1.5 microns *
1000 = 1500 microns. The simulation shows that for a frequency range between 800MHz and
1GHz the dashed blue cover, labeled, M=4K or 4,000 the gm of the transistor overcomes
that of the tank. In order to layout these huge transistors, multi-fingered transistors are
realized.

3.4.3

Arrays

Sensing arrays are designed the same way as the DC, and AC arrays. However, other
considerations are due to guarantee oscillation on all inductors/coils in the array. First,
the inductor itself has to be designed correctly as previously discussed. The Q-value of
the inductor has to be between 5-10, the higher the better. The conductance of the active
78

components of the LC-cross-coupled oscillator has to compensate for the worst case inductor
routing. Therefore, a simulation is performed to calculate the impedance of the entire path
to the coil array. Fig. 3.30 depicts the simplified path, however, each coil has two analog
multiplexers and two analog switches connected to its ports. To measure the impedance
and the conductance, an AC source is connected to the tank of a 3x3 sensing coil array.
Fig. 3.43 shows the conductance that needs to be compensated at 900MHz. The coil used
in this simulation is Inductor4 in table 3.2 The transistors’ W/L in the switches and the
multiplexers are configured wide (width is 43.95 microns and the multiplier is 20) to reduce
the resistance of the path. Additionally, once they are shared with other modes (i.e. DC
mode) they have to accommodate large amounts of currents. The active components (NMOS
and PMOS) has to be sized accordingly to produce a gm that is 1-3 times larger than the
tank’s. Fig. 3.44 shows a DC simulation of an NMOS transistor with its source connected to
DC voltage of 0.7mV to compensate for the NMOS’s body effect. The width of the transistor
is swept from 50 to 60 (x 1.5µm) and the drain source current is fixed at 15mA. Fig. 3.44
shows the DC simulation of the PMOS transistor which matches the NMOS’s gm. However,
the source is not connected to a DC voltage, it is connected directly to ground. The width
of the transistor is swept from 95 to 105 (x 1.5µm) and the drain-source current is also
fixed at 15mA. Finally, in order to fix the active transistors’ source-drain current at 15mA a
tail-current source is realized as a NMOS transistor. The width of the transistor is 9µm and
the width multiplier is 100. Fig. 3.46 shows a DC simulation on the tail NMOS transistor
to generate 15mA. Approximately, by applying 1V at the gate the transistor would provide
15mA IDS. Additionally, the tail transistor is not connected to a current mirror to facilitate
testing. A user can connect a programmable voltage source to the tail NMOS transistor to

79

0.14

Conductance (S)

0.12

0.1

0.08

0.072 at 900MHz
0.06

0.04

0.02
0

0.5

1

1.5

2

Frequency (GHz)

Figure 3.43: An AC simulation running at vdd=1.8v to find the conductance of Inductor4,
two analog multiplexers, and two analog switches.

adjust the amount of current driven through the tank.
A lab experiment is conducted to test the sensing capability of the coil arrays. The
chip is submerged in deionized water with no beads present initially. Then two beads were
placed on 2 neighboring coils in the array. The array is set to sense the location of two
beads. Coil (0,0) doesn’t have any bead, coil (0,1) has a bead lying on top of it, coil (0,2)
has a bead lying close to the center, coil (0,3) and (0,4) don’t have beads on them. Fig.
3.47 shows the output of the frequency counter connected to the LC-Oscillator. The coils
are multiplexed to detect beads. Coils (0,0), (0,3), and (0,4) don’t show any major shift
in frequency therefore, there is no bead present. Coil (0,1) shows a downshift in frequency,
and (0,2) is not oscillating at all which means there is a bead on it. The mean frequency of
oscillation when there is no bead present is 4.8017 MHz with a standard deviation of 0.0704.
The downshift in frequency when adding a magnetic bead on top of the coil is approximately
80

0.226

Transconductance (S)

0.224
0.222
0.22
0.218

0.2168 at Width=53
0.216
0.214
0.212
50

52

54

56

58

60

Width * 1.5e 6 (m)

Figure 3.44: A DC simulation running at vdd=1.8v to find the transconductance (gm) of a
NMOS transistor. The source is connected to 0.7V to realize the body effect.

0.221

Transconductance (S)

0.22
0.219
0.218
0.217

0.2169 at Width=100

0.216
0.215
0.214
0.213
95

100

105

Width * 1.5e 6 (m)

Figure 3.45: A DC simulation running at vdd=1.8v to find the transconductance (gm) of a
PMOS transistor.

81

70
60

IDS (mA)

50
40
30
20
10
0
0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

Gate Voltage (V)

Figure 3.46: A DC simulation running at vdd=1.8v to find the voltage at which the tail
NMOS transistor provides 15mA.
0.7127 MHz. The detection algorithm is described at the end of subsection 3.4.1. Finally,
the sensing sensitivity in the 8x8 is not performing as expected. There are some issues when
scanning the entire array for beads. There are some inductors in the array which refuse to
start oscillating even when there is no bead present on top. This is due to, low Q-value of
inductors, long interconnects between the active component and the coils, coupling between
neighboring coils. These issues are further investigated in a dedicated chip, shown in Fig.
3.48, that is currently being fabricated.

3.5

Tri-mode System

A dedicated CMOS chip to run all actuation and sensing modes, on the same die, is fabricated. It supports three modes of operation (Sensing, AC actuation, DC actuation) on
82

0.6
Bead Present
No Bead Present
No Bead Present
No Bead Present

0.5

|Y(f)|

0.4

0.3

0.2

0.1

0

5

10

15

20

25

Frequency (MHz)

Figure 3.47: 5 Coils in the array are scanned to sense magnetic beads. One coil detects a
bead when the output frequency is downshifted. The other coil stop oscillating (curve not
shown), and the other three coils have the same frequency output.

the same chip. A system level block diagram is shown in Fig. 3.49, the array size can be
chosen arbitrarily based on the available real-estate on the die. The column/row decoders
need to scale according to the size of the array. If the array is symmetrical, standard cell
decoders can be used to access each cell in the array. In order to reduce, the number of
pad-frames, bonding wires, and I/Os 3-8-thermometer decoders are utilized. Each module
of the sub-components (DC/AC/SENSING) accepts an 8-bits level configuration. The DC
current source can generate 8-level DC currents, the class-e amplifier can generate 8-power
outputs, and the oscillator has an 8-bit varactor to adjust the capacitive load in order to tune
its output frequency. All of these modules are multiplexed using a wide analog multiplexer
as discussed in the DC, AC, RF sections before. Finally, the global control is responsible for
coordinating all the components to create seamless pseudo-parallel operation, where a user
83

Figure 3.48: A 3x3 coil array with improved inductors Q-value and sensing isolation.

84

Figure 3.49: System block diagram for tri-mode operation.
can sense and actuate objects up to 40MHz.
In order to switch between different modes we utilized the high-speed switches designed in
the previous section as shown in Fig. 3.4. The high-speed switches use the smallest possible
transistor length for both NMOS and PMOS transistors. The width for the transistors is
43.95 microns and the multiplier is 20. The gain for the switch is plotted in Fig. 3.51. At
900MHz the gain is 99.55%, which implies that the amplitude of signal passed through it is
not affect it.
Moreover, when switching between the DC and RF sensing modes, the startup time
for the oscillator is considered the limiting factor. However, the startup time for the LC
oscillator implemented, as discussed in section 3.4.1, is 13ns which is more than enough in
a 40MHz (250ns) cycle. Therefore, it is possible to simultaneously manipulate and monitor
magnetic beads by quickly alternating between the DC and RF sensing modes.
All of the components discussed in the AC, DC, and RF sections where carefully inte85

Figure 3.50: A micrograph of the tri-mode system.
grated and fabricated. A die photograph is shown in Fig. 3.50. The size of the die is 1.5mm
x 1.5mm the size of the array is approximately 740µm x 740µm, the digital overhead is
400µm x 1000µm, and the share DC, AC, and RF modules are 460µm x 900µm.
Finally, the lab experiments in the previous sections were mainly conducted using the
tri-mode chip except for the AC experiments and the 30x30 µm DC experiment. The 80x80
coils where we did the DC pseudo-parallel manipulation and RF sensing were in fact part of
the tri-mode system.

86

1.005

1

Vout/Vin

0.995

0.99

0.985

0.98

0.975
0

0.5

1

1.5

2

Frequency (GHz)
Figure 3.51: The frequency response of the large switch. The output voltage divided by the
input voltage.

87

Chapter 4

Magnetic Bead Control

We have introduced in previous chapters multiple platforms to manipulate and sense magnetic micro-beads in the size order of 1µm. These platforms use DC and/or AC electric
currents for manipulation, on-chip RF oscillators for sensing, and an open-cavity device for
low-cost packaging. In this chapter, we develop a second-order dynamic model for the bead
and implement a tailored closed-loop PID controller for precise micro-bead manipulation
using the on-chip sensors and actuators. The second order model includes the micro-beads’
inertia, crucial for precise transport, which is often neglected in the literature, e.g., [11, 4, 17].
Also, the controller does not use any visual feedback, as in [11], or external optical sources,
as in [17], for its sensing modality. In contrast, the entire coil array is utilized by selectively
addressing any single coil and using such coil in time-multiplexing mode to sense and actuate
the bead(s) for seamless pseudo-parallel operations.
88

4.1

Micro-Beads Motion

Standard Cartesian coordinates are adopted so that the coil array (and thus the substrate
surface) form the x − y-plane, and the z-axis is normal to the substrate surface. Standard
Spherical coordinates may also be invoked as commonly used in the literature.

4.1.1

First-Order Translational Motion

1D (1-dimension) and 3D first-order models for translational motion of a bead on the surface
of a micro-coil array are now presented. The 1D model is a result of reducing the 3D model
by limiting the translational movement to the x-direction only. The equations developed for
the first-order model are based on [4]. We begin with the magnetic potential energy, say U ,
for trapping a magnetic bead, given as

1
U = − m · B,
2

(4.1)

where m is the magnetic moment vector of the bead and B is the (external) magnetic flux
density. The moment is expressed as

m=

χV
B,
µ0

(4.2)

where V is the volume of the bead and χ its magnetic susceptibility. Therefore, the magnetic
force becomes
Fm = − U =

1 χV
2 µ0

B2,

(4.3)

Next, any object moving through a fluid encounters a hydrodynamic resistance which
89

opposes its motion. This resistance is called the drag force, and it is directly related to the
viscosity of the medium the object is traveling in. Assuming that the bead is a perfect sphere
moving in a medium with a low Reynolds number (R0 ), the drag force Fd is approximated
as
Fd = −6πrηv · fD ,

(4.4)

where r is the radius of the bead, η is the dynamic viscosity of the fluid, and fD is the drag
coefficient. v is the velocity of the bead relative to the fluidic environment and is expressed
as
v = vb − vf ,

(4.5)

where vb is the velocity of the bead and vf is the velocity of the fluidic flow. Since there is
no fluidic flow in open cavity devices, v is equal to the absolute velocity of the bead.

Finally, the normal force Fn on a bead with mass mb is


Fn =





mb 






0 
0
−g




,




(4.6)

where g is the gravitational acceleration. As a result, the total forces acting on the bead is

Fm + Fd + Fn = 0.

(4.7)

The 3D equation can be simplified to 1D since the plane of interest is x − z-plane [4],
90

therefore, the final form of translational velocity for the controller is

x=
˙

4.1.2

m
6πηrfD

∂Bx
∂Bz
sin θ +
cos θ
∂x
∂x

(4.8)

First-Order Rotational Motion

Rotation is caused by torque, which is the magnetic torque Nm and the torque caused by
viscous drag Nd . When modeling the 3D rotational motion we assume the rotational axis
˙
along the three Cartesian axes are: (i) β (i.e. the rotation along the x-axis), (ii) γ (i.e. the
˙
rotation along the y-axis), and (iii) κ (i.e. the rotation along the z-axis.) Therefore, the
˙
rotation around the x − z-plane is

˙
θ=

4.1.3

15
m
˙
3 (Bx cos θ − Bz sin θ) − 256r x
8πηr

(4.9)

Second-Order Translational Motion

According to Newtons second law, a body of a known mass subjected to a force undergoes
an acceleration. This acceleration is often neglected due to the mico-bead’s small mass, as a
result, the inertia of the bead is ignored. For precise control strategy a second order model
is developed to accommodate the inertia. The sum of all forces applied on a bead is

F = mb · a,
91

(4.10)

where mb the mass of the bead and a the acceleration of the bead. Therefore, the 1D
equation of x [69] is
¨

x=
¨

4.1.4

m
mb

∂Bx
∂Bz
6πηrfD
sin θ +
cos θ −
x
˙
∂x
∂x
mb

(4.11)

Second Order Rotational Motion

The second order rotational torque is assumed too small to be included in the control strategy,
therefore, it is neglected.

4.1.5

First-Order vs. Second-Order

Now, we compare the first-order dynamic model of a bead to a second-order model when
the bead is traveling between two wires. The distance between the two wires is 10 µm.
Initially the bead is trapped on “wire 1” by supplying 100mA DC current through it. Then
at t=1 “wire 1” is switched off and “wire 2” is switched on by passing 100mA through it.
Fig. 4.1 depicts the simulation result of this operation. Clearly, there is a difference in the
behavior of the bead when modeled with inertia. For the 1st-order model the bead almost
instantaneously reaches 90% of its destination to “wire 2” when it is switched on, while it
takes the 2nd-order model almost two seconds to reach the same point.

4.2

On-Chip Coil Model

The CMOS technology used in this work does not allow “non-manhattan” shaped metals.
Consequently, square spiral coils are developed in this design. In order to compute the
magnetic flux density, thus, the magnetic force, we apply the Biot-Savart’s law for each wire
92

1.5

x 10

5

1.4
1.3

Distance (Meters)

1.2
1.1
1
0.9
0.8
0.7
First Order
Second Order

0.6
0.5
0

1

2
3
Time (Seconds)

4

5

Figure 4.1: Comparison between 1st-order transitional motion model vs. 2nd-order. In the
2nd-order the bead’s inertia is included.

93

Figure 4.2: Visualization of a 80x80 micron on-coil as used on the chip, in the Cartesian
space. A, B, C and D indicate the four sections of the mathematical model.
segment in the coil as derived in [70]. The coil is divided to four sections {A,B,C,D} as
shown in Fig. 4.2. The total magnetic field is the sum of all wires in all sections.

4.3

Control Design

Feedback control is widely used in automation and robotic applications to stabilize an actuation task. A feedback system can be described as an output variable (y) which is measured
by a sensor and fed back via a controller to be adjusted to reach a desired value. In other
words, the output variable (y) is the bead’s current location, influenced by a noise source
(FT ), is compared to the desired destination of the bead (yd ) and then passed to the controller to adjust the system’s actuation parameters, e.g. current through a coil, to move the
bead to the desired location. This operation is shown as a block diagram in Fig. 4.3.
Any minute object suspended in a liquid platform experiences Brownian motion. Therefore, the noise source here is best observed as Brownian motion of the bead. This motion
is caused by a collection of forces. In this implementation, the main cause of this motion is
assumed to be the thermal force and it is considered to be a white Gaussian noise source
94

yd

+-

e

PID u

F
T
++

y

SENSE
Figure 4.3: Control Strategy for second order bead dynamics.
[11].
Most design methods for controllers require complete models of the dynamic process. The
Ziegler-Nichols tuning of PID is adopted here to simplify the approach [69]. Fig. 4.4 shows
Ziegler-Nichols tuning of a bead moving from one coil’s center to another. A simulation is
performed in Fig. 4.5 to show the second-order model of the bead, using Ziegler-Nichols
tuning method, while being manipulated between two neighboring coils in the array. The
bead doesn’t reach its exact final destination due to a tolerance range set by the controller.
If the bead escapes that range, due to diffusion, the PID controller will adjust the current
though the coil to trap the bead back in place.
Finally, the entire coil array shares a single PID controller to move a bead from one
location to the other. Initially the coil array is scanned and calibrated using the on-chip
coils. Then the array and the PID controller are programmed to move a bead from a specific
location (i.e. row=0,column=0) to another (i.e. row=3,column=3). The PID controller
is switched to control the first target coil (i.e. row=1,column=1), then, when the bead
is attracted (trapped) there the PID controller is “handed over” to the next coil for the
following trapping operation until the bead reaches its final destination. For simplicity, we
assume (i) the bead’s path is always simple i.e. from a coil’s center to another, (ii) there are
no obstacles on the surface of the chip to block a bead from reaching its destination. We
95

5

x 10

5

x (m)

4
3
2
1
0

10

20

30
40
Time (seconds)

50

60

Figure 4.4: Process reaction curve of a bead moving from one coil’s center to another, based
on a second order system. For the Zieger-Nichols tuning, a tangent is drawn approximately
in the inflection point. TL = 30sec and a = 39 · 10−6

Distance (m)

5

x 10

5

4
3
2
1
0

Bead’s Location
Desired Location
2

4
6
Time (seconds)

8

10

Figure 4.5: Simulation result of the response to a step function at t = 1 sec.

96

plan to report on path finding and obstacle avoidance strategies in a future publication.

4.4

Control Based on Energy Functions

The following modeling considers the magnetized force interaction when a magnetized object
of interest (bead) is initially magnetized by a first coil on the array, which then is switched
off, followed by a neighboring second coil switching on. Let the induced magnetic moment of
1
the (magnetic) bead by the first coil is given as Mk − 1 = µ χV Bk − 1, where (the physical
0

constants) µ0 , χ, V are respectively the permeability, the susceptibility and volume of the
bead due to the spatial coil located at k − 1[4]. Thus the generated magnetic potential for
a bead which acquired the moment of magnetic field generated from location k − 1 and now
is acted upon by spatial coil location k is given as the vector dot product of the 2 adjacent
magnetic fields.

1
U = − Mk−1 · Bk
2
1 1
= − ( χV Bk−1 ) · Bk .
2 µ0

(4.12)

As the forces from the spatial coils are generated sequentially, it is thus assumed that the
field Bk−1 is in steady state when the new field Bk is generated from the adjacent coil k.
Thus, the translational dynamics of the bead is given as [22, 17, 4]

m¨ = Fd + Fm
x

(4.13)

= (−6aπηrB (x − vf )) +
˙
97

1 1
χV (Bk−1 )Bk )
2 µ0

We relegate the definition of the (constant, physical) parameters to [4]. For the purpose of
control, the equations are expressed in the compact form

m¨ = Γ − dx + b( Bk )Bk−1 ,
x
˙

(4.14)

where Γ is a vector determined by the fluid flow, d is a (damping constant) determined by
the flow friction to the bead, and the scalar b is signifying scaling of the magnetic force
generated. The rotational dynamics of the magnetized bead is expressed compactly in the
form
¨
˙
I θ = Nd + Nm = −dr (θ + β x) + (Bk−1 × Bk ),
˙

(4.15)

where the parameters are compressed physical constants. It is known (theoretically and
experimentally) that the rotational dynamics are fast in comparison to the translational
dynamics and are often taken to be in the steady state of the induced moments as the
bead quickly lines up with the adjacent magnetic field of spatial coil k. As an initial step for
analysis, we focus on the translation dynamics with the rotational dynamics already assumed
in the steady state.
The energy associated with the translational dynamics are the sum of the potential and
kinetic energies as [22]
1
˙ ˙
W (x, x) = (xT x) − ΓT x + U,
˙
2

(4.16)

The control signal is the generated magnetic field Bk which is radiating from the spatial coil
k and is controlled by the current strength in the coil k. It is noted that the presence of
fluidic flow renders the vector Γ non-zero, and with the consideration of the inertia of the
bead, the dynamics are non-gradient. Thus methods that exploit the gradient in devising
98

control are not applicable. However, in practice, when the mass of the bead is neglected, the
dynamics become gradients and the bead can be steered by the potential surface shaping.
The platform array in effect discretize the potential surface spatially and the coil elements
of the array define spatial regions of the potential field. Thus, an object can be steered
along arbitrary paths along the platform surface. The energy function in (4.16) captures
the dynamics of the physical system with control and thus represents the constraint in an
optimization setting. In the optimization, one can derive the control sequence that would,
e.g., steer a single or a population of beads to form a prescribed probability distribution or
profile.

99

Chapter 5
IC Packaging

5.1

Introduction

All of the components are integrated on the same die and fabricated using standard 0.5µm
3M2P CMOS technology then packaged using standard ceramic packages (shown in Fig. 5.1)
or plastic packages (shown in Fig. 5.2). Contrary to other state-of-the-art implementations,
our method doesn’t need any special post-processing, such as; microfluidic structures to
channel or manipulate bio-materials. It is designed to operate as an “open cavity” device
where the entire surface of the chip is exposed to the sample or material of interest. This
reduces the complexity of the system, makes it cheaper to manufacture, and more portable
by eliminating microfluidic overhead and the micro-pumps associated with them. However,
it is still possible to add these structures if the targeted applications requires such setup.
Using an open-cavity design introduces new challenges, in which, some are shared with
microfluidics. The challenges are: insulating the wire bonds attaching the CMOS die with the
ceramic/plastic package, increasing biocompatibility of the die’s surface, and maneuvering
100

Figure 5.1: Standard ceramic package with PCB board for testing.
the sample of interest on an open surface.

5.2

PDMS

Microfluidic channels are often created using Polydimethylsiloxane (PDMS). This compound
is commonly referred to as Silicone. In Bio-MEMs, silicon wafers are used to design channels
then PDMS is poured over them and left to cure. Nevertheless PDMS is typically hydrophobic, therefore, its surface is usually exposed to RF Plasma techniques to make it hydrophilic.
Once the surface bonds are treated and ready a cover slip (glass) can be placed on the acti101

Figure 5.2: Standard plastic package with drainage for testing.
vated side to seal the PDMS channel structure. Thus, creating a micro-fluidic device. These
devices are heavily used in lab-on-chip and other rapid analysis platforms.
In this work, we deferred from using PDMS and microfluidic channels to keep the overall
design simple. However, we utilized PDMS to insulate the PCB board after assembly. We
used a product of SYLGARD 182 SILICONE ELASTOMER KIT from Dowcorning. The
mixture was prepared according to the standard protocol described in the datasheet. Then
it was poured on the PCB to cover the traces and the soldering points with the header pins.
The board was exposed to 100◦ C for 5 minutes to expedite the curing process.

5.3

Standard

Standard packages were used to reduce the cost and complexity of the final device. Initially
a standard DIP40 ceramic package was utilized, however, the package size and the cavity
clearance were not suitable for AFM measurements and on-chip biomedical experiments.
Second, a standard LCC28 ceramic package was explored, however, the cavity size made the
102

wire-bonding insulation difficult. Third, a standard QFN28 plastic package was used, the
size of the cavity was large so both AFM and biological experiments were conducted. The
plastic package was easily perforated to create proper drainage, however, the only problem
was soldering the package to a standard PCB. Since it is a lead-less package, it required
an oven to solder the device on the PCB. Therefore, to reduce the costs, we decided not to
use this package. Finally, a standard plastic package LQFP64 was chosen due to its large
cavity, long leads, and plastic edges. The low profile cavity made it easy to conduct AFM
experiments and create a relatively deep well to contain liquids suitable for lab-on-chip and
biomedical experiments.

5.4

Custom

First, a printed circuit board (PCB) is used to mount the CMOS die directly on the edge of
an NeuroNexus A-series [71] probe. Then it is placed in a Corning dish under the microscope
as in Fig. 5.3. The bonding wires are protected using EpoxyTechnology 353 ND and 353
NDT. The three on-chip arrays are exposed and submerged in Phosphate buffered saline
(PBS) solution inside a dish. Then small droplets of magnetic beads are applied on the
surface of the chip. The experiment results are discussed in Section 3.2.4.
Second, a LQFP64 package is soldered on a prototype PCB as shown in Fig. 5.4. The size
of the PCB is 1.6” x 1.6” x 0.0625”. The PCB board is chosen to fit under the objective of the
microscope without requiring much space. The cavity size is 8.402mm x 8.402mm, the depth
of the cavity is 0.635mm and the height of the CMOS die is between 0.127 and 0.381mm.
The bonding wire arc’s highest point is estimated to be 0.5mm. These values allowed us
to create a “well” around the CMOS die that is suitable for lab-on-chip experiments. The
103

Figure 5.3: Packaging of the CMOS die on top of a PCB probe.
bonding wires were covered with different compounds for insulation. Also, the depth of the
cavity is large enough to fill it with liquids and cover it with cover slip. The following subsections discuss the insulating techniques that were using on the CMOS die and the bonding
wires.

5.4.1

Wire Bonding Insulation

Simple wire bonding insulation is initially deployed. A product from Loctite (5 minute
epoxy) is used and applied to the bonding wires as directed in the user guide. The mixture
is applied directly on the bonding wires and manually manipulated to cover the wires and
the pads. Fig. 5.5 shows the final insulated bonding wires. The bubbles in the epoxy are a
result of rapid mixing before applying the compound. Fig. 5.6 shows a close-up of the epoxy
on the die. Three coils in the array were covered because (1) there is a bonding wire close
104

Figure 5.4: Packaged CMOS die on top of a prototype PCB.
to the array (i.e lower left), (2) the epoxy flowed downward to the die because of surface
tension.
This chip was functional for approximately 2 months then it failed possibly due to leakage.
There were several factors which caused this chip to fail. First, the edges of the CMOS die was
regularly exposed to organic chemicals, water, chlorine, as well as UV light. These chemicals
caused the edges to chip after numerous exposures. Additionally, extreme UV light exposure
changed some of the properties of the semiconductor which eventually changed the threshold
voltage on the DC current supplies causing them to fail or operate beyond their capacities.

5.4.2

Enhanced Wire Bonding Insulation

A more sophisticated approached is developed to better insulate the CMOS die and the
wire bondings. First, the device is is coated with a 2µm layer of Parylene-C using Specially
105

Figure 5.5: Epoxy covering bonding wires.

Figure 5.6: Micrograph of CMOS die submerged in fluid with magnetic beads on top. The
image illustrates how the epoxy covers part of the die and made 3 coils unusable.

106

Coating Systems PDS 2010. The deposition time was 1 hour and 31 minutes and 40 seconds.
The dimer mass was 3.1348 grams. Second, the bonding wires and the CMOS die were
covered by EpoxyTechnology 353 NDT or EpoxyTechnology 353NDT-5. Note that, the
viscosity and the thixotropic index of the epoxy are crucial parameters. Choosing a higher
viscosity makes the application procedure difficult, and choosing a low viscosity makes it
hard to control the flow, thus, covering the entire die. We found that a viscosity (20 RPM
and 23◦ C) between 5000 and 15000 cPs and thixotropic index between 2.1 and 3.6 fit best
for this procedure. The procedure of applying the epoxy is described as follows:

• Before mixing the epoxy it might be desirable to add a black pigment to make the
color of the epoxy black in order to improve the Epifluorescence microscopy images.
• After mixing the epoxy, a small amount (approximately 5mL) is applied on the bonding
wires and around the CMOS die. The entire cavity can be filled if need be. However,
it is desirable to keep the epoxy level below the edges of the package.
• The epoxy might start flowing toward the center of the chip. However, the flow should
be minimal.
• Dry Nitrogen is slowly blown toward the center of the chip to prevent any slow flow
and to create a dam surrounding the die.
• The pressure of the dry nitrogen is slowly increased until the epoxy flows outwards.
This creates a dam around the CMOS die. At this point the epoxy should stay at its
current position due to its high viscosity.
• The device is placed in preheated vacuum oven (Fisher Scientific Isotemp Vacuum
Oven Model 281A) set at 50◦ C then the vacuum is increased to, approximately, 30
inches of Hg. The epoxy will rise when subjected to such pressure, thus, creating a
107

Figure 5.7: Top view of chip1 insulation.
larger dam around the CMOS die, which creates the open-cavity needed for cultures,
or bead manipulations.
• While the device is still under pressure, the temperature is gradually increased to
100◦ C to cure the epoxy.
• Once the temperature reaches 100◦ C. The vacuum is released progressively and the
epoxy is watched carefully until it cures. 353NDT cures in 10 minutes under 100◦ C,
however, due to preheating and vacuum pressure, it is highly recommended to reduce
that time and carefully watch the amber color of the epoxy.
The results of the aforementioned method is shown in Fig. 5.7 for the first chip and Fig.
5.8 for the second chip. For the first chip, there were debris that prevented the epoxy to flow
back. However, for the second chip the epoxy covered the chip properly.
Finally, a standard corning dish is perforated and installed on the PCB board as shown
in Fig. 5.9. The dish is insulated with standard 5 minute epoxy to prevent leaks. The same
108

Figure 5.8: Top view of chip2 insulation.
353NDT epoxy can be also used to seal the dish for better biocompatibility. The die can be
attached directly to PCB as shown in the figure or it can be encapsulated with a standard
plastic package as previously discussed. If the fabrication house provides free wire-bonding,
it might be desirable to use pre-packaged CMOS dies to reduce costs. However, the package
leads and thermal package (if present) need to be sealed and insulated accordingly. PDMS
can be used, after Parylene-C, coating to submerge the entire PCB while keeping the die
and the surface of the package exposed.
The new packaging scheme is built to accommodate neuronal cultures. The dish is filled
with media and then small droplets of neuronal embryos are deposited in the center i.e. on
top of the chip. The culture is then placed to in an incubator to grow. For more details, see
Chapter 6.

109

Figure 5.9: Final packaged device.

110

Chapter 6
On-Chip Axonal Initiation and
Elongation

6.1

Introduction

Forces have a key role on the growth and development of neurites. For example, external
application of forces can double the elongation rates of axons without disrupting the transmission of action potentials or causing axonal thinning. Current methods for applying forces,
namely using force calibrated glass needles[72, 73, 74], require excellent manual dexterity,
elaborate steps for needle calibration, and continuous observation using fluorescents and
bulky optics. This limits both the throughput and the application of this general approach
to users with a high degree of training.
In this thesis, we present our contributions toward a fully integrated device suitable for
autonomous axonal elongation and monitoring on top of the surface of a standard CMOS
substrate. The system consists of a programmable on-chip coil array capable of generating
111

customized electromagnetic forces, in the order of nano Newtons, to synergistically control
and manipulate superparamagnetic beads bound to neurites. The electromagnetic forces
utilize singular and collaborative maneuvering to elongate axons in 3-Dimensions. Each coil
in the array is also time-shared to a cross coupled LC oscillator to sense and detect the spatial
position of the bead and the cell body, and thus, scrutinizing the movement and growth of
the axon. Therefore, the system previewed provides an automated, high precision, and highresolution platform to better understand and study neuroregeneration in vitro, which will
propel future clinical applications and provides the foundation of techniques needed to wire
neuronal circuits using living cells.

6.2

Biophysical Model

The process of axonal elongation has been described to occur in three stages [73]:
• An initial elastic stretch.
• Delayed stretching.
• Constant elongation rates.
This behavior is modeled by Dennerll et al. [73] using three elements where the axon
behaves as a spring, a Voigt element, and a dashpot. All elements are connected in series.
The combination of the three elements are also called a Burgers element [73]. This model
illustrates the effects of tensile stress on the elongation process, but it does not include
the axon’s distal deformations and/or the effects of adhesions between the axon and the
substrate. On the other hand, there is Aeschlimann model, where the axon is modeled as (1)
series of springs that stretch elastically and (2) a growth dashpot at the end where the new
mass is added during elongation. This model includes the spring and fluid like behaviors
112

Figure 6.1: The axon is treated as a series of dashpots (G). Where F is the force or tension
applied or generated at the growth cone and the friction dashpot η represents the attachments
to the substrate.

associated with axonal elongation. Finally O’Toole et al. [75] model axons as a series of
dashpots that act like a viscoelastic fluid, as show in Fig. 6.1. They derive a model that
includes the effects of tension generation at the growth cone as well as the dissipation along
the length of the axon due to adhesions to the substrate.

The force profile of the axon, f , is a function of the distance from the cell body, x, and the
length of the axon, L(t). A constant force F0 applied at the growth cone causes distension at
each point along the length. Therefore, the mathematical model [75] for the force is realized
in 6.1 and force the axon’s length in 6.2.

f (x, L(t)) =

F0 cosh x η/G

,

(6.1)

cosh L(t) η/G

L(t) =

η/G arcsinh sinh L0 η/G exp
113

F0 t
G

,

(6.2)

Axon’s Cross sectional Area (µm2)

3.5
t=0
t=15min
t=30min
t=45min
t=60min
3

2.5

2
0

5

10
Axon’s Length (µm)

15

Figure 6.2: Axon’s model simulation, without mass addition, shows a 11µm long axon being
stretched for 1 hour with 1nN tension force. After 15min the length is 11.9. After 30min
the length is 12.97. After 45min the length is 14.22. At the end of the simulation the axon’s
length is 15.71. A total growth of 4.71 microns. The diameter of the axon is thinning because
mass addition was not realized.
where G is the growth dashpot parameter and η is the constant of friction. During the
elongation, G varies along with the cross-sectional area while η remains constant. Fig. 6.2
shows a simulation of this model [75] where the axon is 11µm long and a constant force (F0 )
of 1nN is applied at the growth cone.

6.3

Experiments

Several lab experiments were conducted to elongate axons using magnetic beads. Initially,
we used off the shelf components i.e. electromagnets, and Neodymium magnets. Next, we
114

employed the on-chip micro-coil arrays to manipulate the beads, attach them to axons, and
initiate and elongate them.

6.3.1

Neuronal Culture

Drosophila neurons are isolated as discussed in [76]. Cells are grown and imaged at 25C in
L-15 medium (Sigma Chemical Co., St. Louis, MO) then supplemented with 0.6% glucose, 1
mM glutamine, 100 U/ml penicillin, 136 mg/ml streptomycin sulfate, 10% fetal calf serum,
and N9 growth supplement [77]. This is the standard L-15+N9 based media developed for
the culture of chick sensory neurons by Lamoureux et al. [78].

6.3.2

Elongation on Dish

First, we tried to use off-the-shelf components to manipulate magnetic beads to bind and
elongate axons. We used a product of APW Company 1.0 Diameter Round Electromagnets
EM100 that can hold up to 25 pounds. Three electromagnets were used to maneuver the
beads on the dish and were placed around the sample as shown in Fig. 6.3. Initially, all
three electromagnetic are switched off then the one on the west is switched on, then the east,
then the north. This movie is depicted in a montage in Fig. 6.4. The video is available at
(http://www.msu.edu/~asd1815/videos.html titled [off manipulation]). Here, the beads
are pointing toward the maximum magnetic field emitted by the electromagnetic. The
electromagnets are not lying in the same plane as the sample (dish), therefore, the beads
are pointing “upwards” i.e. normal to the dish. This weakens the manipulation forces and
causes the distal region of the axon to detach itself from the substrate and retreat.
Moreover, a custom microscope stage was fabricated in order to dip the electromagnets
115

Figure 6.3: Three electromagnetic placed around the sample.

116

Figure 6.4: Magnetic bead manipulation using three of the shelf electromagnetic.
117

into the stage for maximum transitional forces. The center of the electromagnet is placed
in plane with the culture dish, as depicted in Fig. 6.5. The forces generated from the
electromagnet vs. the distance (from its center) is shown in Fig. 6.6. The forces decrease
cubically as the distance increases. Additionally, having an electromagnetic close to the
sample creates problems when using on stage incubators because the user has to move the
magnetic source father away from the sample, thus, reducing the force. Or when larger
dishes are used the magnetic forces become weaker for the targets at the center of the dish.
Multiple experiments were conducted to elongate axons using these electromagnets. Two
experiments are shown in Fig. 6.7 and Fig. 6.8. The videos are available at (http://
www.msu.edu/~asd1815/videos.html titled [off exp1] and [off exp2]). Phase-microscopy
was used to take these photographs. The magnetic beads were manufactured by Invitrogen
and were coated with Texas red rabbit HRP (horseradish peroxidase) antibody to facilitate
the binding between beads and the axon’s membrane. In the first montage, two 1.05 µm
beads were attached to an axon. The electromagnetic placed at the south was turned on at
maximum power. The magnetic force was not capable of pulling/towing the beads toward
the south. In the second, montage 6.8 a single bead was attached to an axon but it couldn’t
pull it due to the weak force generated by the electromagnet. The bead actually moved
along the axon due to mass additions and transport from the cell body.

6.3.3

On-Chip Axonal Elongation

Axonal elongation experiments are implemented on-chip i.e. in vitro to eliminate the limitations seen in the previous section. The limitations are mainly: manipulation resolution,
manipulator size, weak magnetic force, and temperature.
118

Figure 6.5: Custom stage where the electromagnetic is placed in plane with the sample.
8
7

Force (nN)

6
5
4
3
2
1
0
0

5

10

15
20
Distance (mm)

25

30

Figure 6.6: Forces generated from electromagnet EM100.
119

Figure 6.7: Montage of two magnetic beads trying to tow axons grown on a dish.

120

Figure 6.8: Montage of a magnetic bead trying to tow axons grown on a dish. The axon
actually pulls the bead toward its body.

121

The setup is very similar to the previous section, however, instead of using a culture dish
we utilized the chips designed and packaged in the previous chapters. The chips were coated
and treated with different compounds to increase the growth yield on the silicon substrate.
First, for the chips which were previously coated with Parylene-C for insulation and packaging, it was impossible to grow drosophila neurons with the techniques used in sections 6.3.2.
This is due to the hydrophobic nature of Parylene-C. As a result, we treated the surface
with Poly-L-Ornithine (SIGMA) for 30 minutes prior to platting i.e. placing the embryos on
the surface of the chip. Additionally, on some chips we activated the Parylene-C with UV
(ultra-violate) light (15 min) and horse/calf serum (3 hours) [79] prior to platting. The result
cultures are shown in Fig. 6.9 and 6.10. Based on more than 20 on-chip trials, we estimate
that we achieved 20% increase in neuronal growth with these techniques. Epifluorescent
microscopy was used to take these photographs. The cultures were pictured 18 hours after
platting and they were stained with a dye made by Invitrogen, Vybrant DiI. The staining
procedure is similar to what is in the datasheet, however, few modifications were adopted to
improve the imaging and the life-span of the axons:
• Dilute 5µL of DiI in 1mL L-15 media. The media has to be supplement free i.e. no
serum.
• Remove the media from the culture and add the diluted medium above.
• Incubate for 45 minutes at room temperature.
• Remove the diluted media and carefully wash with PBS 1-3 times.
• Add supplemented L-15 media back again.
• Wait for the cultures to recover (15 minutes) and start imaging and elongation.
In order to elongate the axons, in this experiment we used 2.8µm beads which has more
122

Figure 6.9: A group of neuronal cells grown on a treated surface of Silicon Nitride.

Figure 6.10: A neuronal cell with multiple axon on top of the coil array.

123

Figure 6.11: On-Chip axonal elongation setup. The open-cavity device makes it easier to
deposit small droplets of embryos. The coverslip is placed on top of the plastic package to
reduce vibrations and evaporations.
iron content that the previous bead (1.05µm). These beads were coated with the Protein-G
then they were deposited as small droplets (5µL) onto the surface of the chip. The package
was then sealed with a coverslip, depicted in Fig. 6.11, to improve the clarity and focus
by preventing surface tension, evaporation, and vibration. Also, we used the new packaging
scheme shown in Fig. 5.9.
Before starting the elongation experiment a grace period (10 min) is observed in order
for the beads to bind to the axons and settle at the surface of the chip. The user can use the
124

sensing scheme on-chip to detect where the beads are located or one can use the microscope
in the bright-field mode to locate them. If none of the beads were attached to the axons,
the user can use the on-chip manipulation mode to move the beads close to an axon to help
attaching it to the axon. The grace period (10 min) is observed to create a “good” bond
between the bead and the membrane of the axon.
In order to observe the axonal elongation procedure, the beads have to be labeled with a
fluorescent dye that is similar, in its wavelength, as the DiI. If the dye is not available then
the user has to switch between bright-light filter and fluorescent filters. Fig. 6.12 shows the
initial state of the axon before elongation. Fig. 6.13 shows the axon after being elongated
with a magnetic bead. The bead is not visible under the fluorescent filter. The experiment
was conducted 24 hours after egg lay, and 19 hours after platting. The axon grew 1.7µm
in 20 minutes. The length was measured them using imagej[80]. The force of elongation
is estimated to be 700pN, based on the previous simulations in section 3.2.4. Additionally,
the elongation rate is double the normal rate and it is within the simulated growth for 1nN
towing force as shown in Fig. 6.2.
Finally, a second experiment is conducted to initiate a branch out of an axon/bundle. In
this experiment the bead is attached to the side of the axon and most likely it caused it to
initiate a new branch. Fig. 6.14 shows the initial state of the axon and Fig. 6.15 shows the
new branch created. The estimated towing force is 400pN.

125

Figure 6.12: On-Chip axonal elongation, initial axon state where the bead is bond to the
east of the axon. The bead is not visible under fluorescent filters.

Figure 6.13: On-Chip axonal elongation, final axon state where it elongated 1.7µm in 20
min.

126

Figure 6.14: On-Chip axonal initiation, initial axon state where the bead is bond to the west
of the axon. The bead is not visible under fluorescent filters.

Figure 6.15: On-Chip axonal initiation, final axon state where it branched out after 30 min.
127

Chapter 7
Conclusions and Future Work

7.1

Summary of Contributions

A lab-on-chip programmable magnetic field generator is implemented, tested, and demonstrated. It can efficiently detect and maneuver biomaterials using magnetic beads without
the need of external magnets or special fabrication and packaging processes. Multiple CMOS
chips are fabricated and tested for DC/AC actuation and RF sensing. The platforms are
suitable for single or collaborative object sensing and manipulation. The designed system
features variable magnetic fields, programmable control scheme, different planar inductor
designs and array shapes, sizes, and implementations. With these platforms it is possible
to dynamically create an unlimited number of magnetic force profiles and contours needed
to interact and sense different bio-objects for different bio-applications. Moreover, a novel
biomedical application is introduced as a proof of concept. On-chip axonal elongation and
initiation is demonstrated by binding magnetic beads to the axons and towing them with
localized magnetic forces using the mico-coil array.
128

7.2

Future Work

Future implementation should include on-chip feedback control to precisely maneuver and
manipulate beads. The sensing mechanism used here has to be improved by using a CMOS
technology with RF capabilities such as IBM 0.13µm or better. The current 0.5µm CMOS
technology suffers in producing inductors with high Q-values. In order to create Q¿5 using the
current technology the size of the inductor has to be larger than 400µm which is unfeasible for
such applications. Additionally, on-chip temperature sensor should be included to guarantee
that the heat generated from the coils does not negatively influence the biological sample on
top, and does not alter the oxide layers when using higher currents. Finally, the packaging
scheme can be improved to increase reliability by using better insulators and miniaturized
3D printers.
Possible applications for this coil array is (1) micro-robots where the bead can be functionalized to act as a robot controlled by the mico-coil array. These small robots can be
used to target specific objects in vitro. (2) On-chip imaging where the array can be used to
scan for dielectric material and image their location, the resolution of the imager depends
on the size of the coil array and spatial resolution of the LC-CC sensor. A non-uniform
gradient can be explored when designing such coil, therefore, any point in the coil can be
distinguished based on the shift in frequency at that spatial location. Neighboring coils can
be used to correlate the location of the target using the pseudo-parallel scheme introduced
in this work. (3) Micro-heaters, the coil array can be used to heat and incubate biological
materials by supplying large currents through the coils. A feedback control can be used to
adjust/set and program the temperature externally.

129

APPENDICIES

130

Appendix A
Maxwell VBS to create and analyze
on-chip sensors and actuators for 0.5
CMOS process

1 ’ C o p y r i g h t ( c ) 2011 , F a i s a l T. Abu−Nimeh
’ All rights reserved .
3 ’ L i c e n s e : BSD 3−Clause License , See l i c e n s e . t x t
’ CSANN Lab
5 ’ Michigan S t a t e U n i v e r s i t y
’ 80 x80 s p i r a l
7 ’ 20100803

9 Dim oAnsoftApp
Dim oDesktop
131

11 Dim o P r o j e c t
Dim oDesign
13 Dim o E d i t o r
Dim oModule
15
’ Current i n mAmpes
17 Dim Current
’ O u t t e r S p a c i n g S i z e or I n d u c t o r Length
19 Dim S i z e

21 Const boxName = ”Box”
Const U n i t s = ”um”
23 Const ViaWidth = 0 . 6
Const ViaSpacing = 0 . 6
25 Const MetalWidth = 1 . 8
Const MetalSpacing = 1 . 2
27 Const SNT = 1 ’ v a l u e s b e l o w s h o u l d be t a k e n from NDA documents
Const SOXT = 1
29 Const M3T = 1
Const M3M2 = 1
31 Const M2T = 1
Const M2M1 = 1
33 Const M1T = 1

132

Const M1PLY = 1
35 Const PLYT = 1
Const TBPSG = 1
37
Dim N
39
Current = 40 ’mA
41 S i z e = 80 ’ Length i s 80um
N = Int ( S i z e / ( 2 ∗ ( MetalWidth+MetalSpacing ) ) ) ’ Number o f Turns
43 ’ MsgBox (” l o o p s ” & CStr (N) )
Set oAnsoftApp = CreateObject ( ” AnsoftMaxwell .
MaxwellScriptInterface ”)
45 Set oDesktop = oAnsoftApp . GetAppDesktop ( )
oDesktop . NewProject
47 Set o P r o j e c t = oDesktop . G e t A c t i v e P r o j e c t
o P r o j e c t . I n s e r t D e s i g n ” Maxwell ” , ” Maxwell Model1 ” , ”” , ””
49 Set oDesign = o P r o j e c t . S e t A c t i v e D e s i g n ( ” Maxwell Model1 ” )
Set o E d i t o r = oDesign . S e t A c t i v e E d i t o r ( ”3D Modeler ” )
51 Set oModule = oDesign . GetModule ( ” BoundarySetup ” )

53 o E d i t o r . SetModelUnits Array ( ”NAME: U n i t s Parameter ” , ” U n i t s :=” , ”um
” , ” R e s c a l e :=” , f a l s e )
Dim Succ

133

55 ’ Phase 1 i s c l o c k w i s e w i n d i n g
’ Phase 0 i s c o u n t e r
57 Succ = c r e a t e S p i r a l ( ” m e t a l 3 ” , 1 ,M3T, 0 )

59 ’ v i a 2 −− use l a s t segment p o s i t i o n , add l e n g t h o f segment t o Y
and o f f s e t v i a s i z e from r i g h t
o E d i t o r . CreateBox Array ( ”NAME: BoxParameters ” ,
61

” X P o s i t i o n :=” , CStr ( S i z e −((N−1)∗ MetalSpacing+MetalWidth ∗(N) )+
ViaSpacing )&”um” ,
” Y P o s i t i o n :=” , CStr (N∗( MetalWidth+MetalSpacing )+ViaSpacing )&”um”
,

63

” Z P o s i t i o n :=” , CStr( −1∗(M3M2+M2T) )&”um” ,
” XSize :=” , CStr ( ViaWidth )&”um” , ” YSize :=” , CStr ( ViaWidth )&”um” , ”
Z S i z e :=” , CStr (M3M2+M2T+M3T)&”um” ) ,

65 Array ( ”NAME: A t t r i b u t e s ” , ”Name:=” , ” v i a 2 ” , ” PartCoordinateSystem :=
” , ” Global ” , ” MaterialName :=” , ”aluminum” , ” S o l v e I n s i d e :=” , t r u e
)

67 ’ b r i d g e
o E d i t o r . CreateBox Array ( ”NAME: BoxParameters ” ,
69

” X P o s i t i o n :=” , CStr ( S i z e −((N−1)∗ MetalSpacing+MetalWidth ∗(N) ) )&”
um” ,
” Y P o s i t i o n :=” , CStr (N∗( MetalWidth+MetalSpacing ) )&”um” ,

134

71

” Z P o s i t i o n :=” , CStr( −1∗(M3M2+M2T) )&”um” ,
” XSize :=” , CStr ( S i z e /2 )&”um” , ” YSize :=” , CStr ( MetalWidth )&”um” ,
” Z S i z e :=” , CStr (M2T)&”um” ) ,

73 Array ( ”NAME: A t t r i b u t e s ” , ”Name:=” , ” m2bridge ” , ”
PartCoordinateSystem :=” , ” Global ” , ” MaterialName :=” , ”aluminum”
, ” S o l v e I n s i d e :=” , t r u e )

75 ’ Create one c o n d u c t i n g p a t h
o E d i t o r . Unite Array ( ”NAME: S e l e c t i o n s ” , ” S e l e c t i o n s :=” , ” metal3 1 ,
via2 , m2bridge ” ) , Array ( ”NAME: UniteParameters ” , ” K e e p O r i g i n a l s :=
” , false )
77
’ E x c i t e me
79 ’ oModule . A s s i g n C u r r e n t Array (”NAME: Current1 ” , ” Faces :=” , Array
( 6 5 8 ) , ” Current :=” , CStr ( Current )&”mA” , ” I s S o l i d :=” , f a l s e , ”
P o i nt o u t o f t e r m i n a l :=” , f a l s e )
’ oModule . A s s i g n C u r r e n t Array (”NAME: Current2 ” , ” Faces :=” , Array ( 4 0 )
, ” Current :=” , CStr ( Current )&”mA” , ” I s S o l i d :=” , f a l s e , ” Point
o u t o f t e r m i n a l :=” , t r u e )
81
’ Create P a s s i v a t i o n
83 o E d i t o r . CreateBox Array ( ”NAME: BoxParameters ” ,

135

” X P o s i t i o n :=” , ”0um” , ” Y P o s i t i o n :=” , ”0um” , ” Z P o s i t i o n :=” , CStr
( −1∗(M3M2+M2T+M2M1+M1T+TBPSG) )&”um” ,
85

” XSize :=” , CStr ( S i z e )&”um” , ” YSize :=” , CStr ( S i z e )&”um” , ” Z S i z e :=
” , CStr (SOXT+M3T+M3M2+M2T+M2M1+M1T+TBPSG)&”um” ) ,
Array ( ”NAME: A t t r i b u t e s ” , ”Name:=” , ”SOX” , ” PartCoordinateSystem
:=” ,

87

” Global ” , ” MaterialName :=” , ” s i l i c o n d i o x i d e ” , ” S o l v e I n s i d e :=” ,
true )
o E d i t o r . CreateBox Array ( ”NAME: BoxParameters ” ,

89

” X P o s i t i o n :=” , ”0um” , ” Y P o s i t i o n :=” , ”0um” , ” Z P o s i t i o n :=” , CStr (
M3T+SOXT)&”um” ,
” XSize :=” , CStr ( S i z e )&”um” , ” YSize :=” , CStr ( S i z e )&”um” , ” Z S i z e :=
” , CStr (SNT)&”um” ) ,

91

Array ( ”NAME: A t t r i b u t e s ” , ”Name:=” , ”SNT” , ” PartCoordinateSystem
:=” ,
” Global ” , ” MaterialName :=” , ” s i l i c o n n i t r a t e ” , ” S o l v e I n s i d e :=” ,
true )

93
’ Create a i r box
95 o E d i t o r . CreateBox Array ( ”NAME: BoxParameters ” ,
” X P o s i t i o n :=” , CStr(−1∗ MetalSpacing )&”um” , ” Y P o s i t i o n :=” , CStr
( −1∗( MetalSpacing+S i z e ) )&”um” , ” Z P o s i t i o n :=” , ”−15um” ,

136

97

” XSize :=” , CStr ( S i z e +2∗MetalSpacing )&”um” , ” YSize :=” , CStr (3∗
S i z e +2∗MetalSpacing )&”um” , ” Z S i z e :=” , ” 30um” ) ,
Array ( ”NAME: A t t r i b u t e s ” , ”Name:=” , ” a i r b o x ” , ” Color :=” , ” (200
200 200) ” , ” PartCoordinateSystem :=” , ” Global ” , ” MaterialName
:=” , ” a i r ” , ” Transparency :=” , 1 , ” S o l v e I n s i d e :=” , t r u e )

99
’ Create non o v e r l a p p i n g l a y e r s
101

o E d i t o r . S u b t r a c t Array ( ”NAME: S e l e c t i o n s ” , ” Blank P a r t s :=” , ”
a i r b o x ” , ” Tool P a r t s :=” ,
” m e t a l 3 1 ” ) , Array ( ”NAME: S u b t r a c t P a r a m e t e r s ” , ”
CoordinateSystemID :=” , −1, ” K e e p O r i g i n a l s :=” ,

103

true )
o E d i t o r . S u b t r a c t Array ( ”NAME: S e l e c t i o n s ” , ” Blank P a r t s :=” , ”SOX” ,
” Tool P a r t s :=” ,

105

” m e t a l 3 1 ” ) , Array ( ”NAME: S u b t r a c t P a r a m e t e r s ” , ”
CoordinateSystemID :=” , −1, ” K e e p O r i g i n a l s :=” ,
true )

107
’ Create Setup and P l o t s
109 Set oModule = oDesign . GetModule ( ” A n a l y s i s S e t u p ” )
oModule . I n s e r t S e t u p ” M a g n e t o s t a t i c ” , Array ( ”NAME: Setup1 ” , ”
MaximumPasses:=” , 1 0 , ” MinimumPasses:=” ,

137

111

2 , ” MinimumConvergedPasses :=” , 1 , ” P e r c e n t R e f i n e m e n t :=” , 3 0 , ”
S o l v e F i e l d O n l y :=” ,
f a l s e , ” P e r c e n t E r r o r :=” , 1 , ” S ol ve M at ri xA tL as t :=” , t r u e , ”
UseOutputVariable :=” ,

113

f a l s e , ”PreAdaptMesh:=” , f a l s e , ” N o n L i n e a r R e s i d u a l :=” , 0 . 0 0 1 , ”
MuNonLinearBH:=” ,
t r u e , ”ComputeHc:=” , f a l s e , ”HcNonLinearBH:=” , t r u e )

115 Set oModule = oDesign . GetModule ( ” F i e l d s R e p o r t e r ” )
oModule . C r e a t e F i e l d P l o t Array ( ”NAME: Mag B1” , ” SolutionName :=” ,
117

” Setup1 : LastAdaptive ” , ” QuantityName:=” , ”Mag B” , ” P l o t F o l d e r
:=” , ”B” , ” UserSpecifyName :=” ,
0 , ” U s e r S p e c i f y F o l d e r :=” , 0 , ” I n t r i n s i c V a r :=” , ”” , ” PlotGeomInfo
:=” , Array ( 1 ,

119

”Volume” , ” O b j L i s t ” , 4 , ”Box1” , ” m e t a l 3 1 ” , ”SOX” , ”SNT” ) , ”
F i l t e r B o x e s :=” , Array (
1 , ”” ) , Array ( ”NAME: PlotOnVolumeSettings ” , ” P l o t I s o S u r f a c e :=” ,
t r u e , ” CloudDensity :=” ,

121

0 . 5 , ” P o i n t S i z e :=” , 1 , ” Refinement :=” , 0 , Array ( ”NAME:
Arrow3DSpacingSettings ” , ” ArrowUniform :=” ,
t r u e , ” ArrowSpacing :=” , 0 , ” MinArrowSpacing :=” , 0 , ”
MaxArrowSpacing :=” , 0 ) ) )

123

138

125 Function c r e a t e S p i r a l (Name , Phase , Thickness , Z)
Dim pCount , L e f t R i g h t , UpDown, StrX1 , StrX2
127

pCount=1

129

For L e f t R i g h t = 0 To (N−1)
i f Phase = 0 then

131

StrX1 = CStr ( S i z e −L e f t R i g h t ∗ ( 2 ∗ ( MetalWidth+MetalSpacing ) ) )
StrX2 = CStr ( S i z e −L e f t R i g h t ∗ ( 2 ∗ ( MetalWidth+MetalSpacing ) ) +(
MetalWidth+MetalSpacing ) )

133

else
StrX1 = CStr ( S i z e −L e f t R i g h t ∗ ( 2 ∗ ( MetalWidth+MetalSpacing ) ) +(
MetalWidth+MetalSpacing ) )

135

StrX2 = CStr ( S i z e −L e f t R i g h t ∗ ( 2 ∗ ( MetalWidth+MetalSpacing ) ) )
end i f

137

o E d i t o r . CreateBox Array ( ”NAME: BoxParameters ” , ” X P o s i t i o n :=” ,
CStr ( L e f t R i g h t ∗( MetalWidth+MetalSpacing ) )&”um” , ” Y P o s i t i o n
:=” , CStr ( L e f t R i g h t ∗( MetalWidth+MetalSpacing ) )&”um” , ”
Z P o s i t i o n :=” , CStr (Z)&”um” ,
” XSize :=” , StrX1&”um” , ” YSize :=” , CStr ( MetalWidth )&”um” , ”
Z S i z e :=” , CStr ( T h i c k n e s s )&”um” ) ,

139

Array ( ”NAME: A t t r i b u t e s ” , ”Name:=” , Name & pCount , ”
PartCoordinateSystem :=” , ” Global ” , ” MaterialName :=” , ”
aluminum” , ” S o l v e I n s i d e :=” , t r u e )

139

pCount=pCount+1
141

o E d i t o r . CreateBox Array ( ”NAME: BoxParameters ” , ” X P o s i t i o n :=” ,
CStr ( L e f t R i g h t ∗( MetalWidth+MetalSpacing ) )&”um” , ” Y P o s i t i o n
:=” , CStr ( S i z e −( L e f t R i g h t ∗ MetalSpacing+MetalWidth ∗(
L e f t R i g h t +1) ) )&”um” , ” Z P o s i t i o n :=” , CStr (Z)&”um” ,
” XSize :=” , StrX2&”um” , ” YSize :=” , CStr ( MetalWidth )&”um” , ”
Z S i z e :=” , CStr ( T h i c k n e s s )&”um” ) ,

143

Array ( ”NAME: A t t r i b u t e s ” , ”Name:=” , Name & pCount , ”
PartCoordinateSystem :=” , ” Global ” , ” MaterialName :=” , ”
aluminum” , ” S o l v e I n s i d e :=” , t r u e )
pCount=pCount+1

145

Next

147

For UpDown = 0 To (N−1)
i f Phase = 0 then

149

StrX1 = CStr (UpDown∗( MetalWidth+MetalSpacing ) )
else

151

StrX1 = CStr ( ( UpDown+1) ∗( MetalWidth+MetalSpacing ) )
end i f

153

o E d i t o r . CreateBox Array ( ”NAME: BoxParameters ” , ” X P o s i t i o n :=” ,
CStr (UpDown∗( MetalWidth+MetalSpacing ) )&”um” , ” Y P o s i t i o n :=” ,
CStr (UpDown∗( MetalWidth+MetalSpacing ) )&”um” , ” Z P o s i t i o n :=”
, CStr (Z)&”um” ,

140

” XSize :=” , CStr ( MetalWidth )&”um” , ” YSize :=” , CStr ( S i z e −UpDown
∗ ( 2 ∗ ( MetalWidth+MetalSpacing ) ) )&”um” , ” Z S i z e :=” , CStr (
T h i c k n e s s )&”um” ) ,
155

Array ( ”NAME: A t t r i b u t e s ” , ”Name:=” , Name & pCount , ”
PartCoordinateSystem :=” , ” Global ” , ” MaterialName :=” , ”
aluminum” , ” S o l v e I n s i d e :=” , t r u e )
pCount=pCount+1

157

o E d i t o r . CreateBox Array ( ”NAME: BoxParameters ” , ” X P o s i t i o n :=” ,
CStr ( S i z e −(UpDown∗ MetalSpacing+MetalWidth ∗(UpDown+1) ) )&”um”
, ” Y P o s i t i o n :=” , StrX1&”um” , ” Z P o s i t i o n :=” , CStr (Z)&”um” ,
” XSize :=” , CStr ( MetalWidth )&”um” , ” YSize :=” , CStr ( S i z e −UpDown
∗ ( 2 ∗ ( MetalWidth+MetalSpacing ) ) −(MetalWidth+MetalSpacing ) )&”
um” , ” Z S i z e :=” , CStr ( T h i c k n e s s )&”um” ) ,

159

Array ( ”NAME: A t t r i b u t e s ” , ”Name:=” , Name & pCount , ”
PartCoordinateSystem :=” , ” Global ” , ” MaterialName :=” , ”
aluminum” , ” S o l v e I n s i d e :=” , t r u e )
pCount=pCount+1

161

Next
i f ( pCount > 1 ) then

163

Boxes = Name & ”1”
pieces = 2

165

Do While p i e c e s <pCount
Boxes = Boxes & ” , ”& Name & p i e c e s

141

167

p i e c e s=p i e c e s +1
Loop

169

o E d i t o r . Unite Array ( ”NAME: S e l e c t i o n s ” , ” S e l e c t i o n s :=” , Boxes ) ,
Array ( ”NAME: UniteParameters ” , ” K e e p O r i g i n a l s :=” , f a l s e )
End i f

171

createSpiral = true
End Function

173
Function c r e a t e S h i e l d (Name , Thickness , Z)
175

Dim pCount , s h i e l d N , numBars
pCount=1

177
s h i e l d N = Int ( S i z e / ( MetalWidth+MetalSpacing ) )
179

’ MsgBox (” l o o p s ” & CStr ( s h i e l d N ) )
For numBars = 0 To ( s h i e l d N −1)

181

o E d i t o r . CreateBox Array ( ”NAME: BoxParameters ” , ” X P o s i t i o n :=” , ”
0um” , ” Y P o s i t i o n :=” , CStr ( numBars ∗( MetalWidth+MetalSpacing )
)&”um” , ” Z P o s i t i o n :=” , CStr (Z)&”um” ,
” XSize :=” , S i z e&”um” , ” YSize :=” , CStr ( MetalWidth )&”um” , ” Z S i z e
:=” , CStr ( T h i c k n e s s )&”um” ) ,

183

Array ( ”NAME: A t t r i b u t e s ” , ”Name:=” , Name & pCount , ”
PartCoordinateSystem :=” , ” Global ” , ” MaterialName :=” , ”
aluminum” , ” S o l v e I n s i d e :=” , t r u e )

142

pCount=pCount+1
185

Next
createShield = true

187 End Function

143

BIBLIOGRAPHY

144

BIBLIOGRAPHY

[1] H. Townsend, “Method of magnetic separation,” 1891, US Patent 453,317.
[2] C. T. T. R. H. Sanders, “Process of magnetic separation and apparatus therefore,” 1893,
US Patent 499,253.
[3] X. Wetherill, “Method of and apparatus for magnetic separation,” 1896, US Patent
555,792.
[4] H. Lee, Y. Liu, D. Ham, and R. Westervelt, “Integrated cell manipulation system—
cmos/microfluidic hybrid,” Lab on a Chip, vol. 7, no. 3, pp. 331–337, 2007.
[5] M. Gijs, F. Lacharme, and U. Lehmann, “Microfluidic applications of magnetic particles
for biological analysis and catalysis,” Chemical Reviews, vol. 110, no. 3, pp. 1518–1563,
Mar 2010.
[6] Y. Moser, T. Lehnert, and M. Gijs, “On-chip immuno-agglutination assay with analyte
capture by dynamic manipulation of superparamagnetic beads,” Lab on a Chip, Jan
2009.
[7] H. Lee, A. Purdon, and R. Westervelt, “Micromanipulation of biological systems with
microelectromagnets,” Magnetics, IEEE Transactions on, vol. 40, no. 4, pp. 2991–2993,
2004.
[8] H. Lee, Y. Liu, R. Westervelt, and D. Ham, “Ic/microfluidic hybrid system for magnetic
manipulation of biological cells,” Solid-State Circuits, IEEE Journal of, vol. 41, no. 6,
pp. 1471–1480, 2006.
[9] U. Lehmann, M. Sergio, S. Pietrocola, E. Dupont, C. Niclass, M. Gijs, and E. Charbon,
“Microparticle photometry in a cmos microsystem combining magnetic actuation and in
145

situ optical detection,” Sensors and Actuators B: Chemical, vol. 132, no. 2, pp. 411–417,
2008.
[10] W. Altman, J. Moreland, S. Russek, and V. Bright, “Optimization of spin-valve parameters for magnetic bead trapping and manipulation,” Journal of Magnetism and
Magnetic Materials, vol. 322, no. 21, pp. 3236–3239, 2010.
[11] Z. Zhang, Y. Huang, and C.-H. Menq, “Actively controlled manipulation of a magnetic microbead using quadrupole magnetic tweezers,” Robotics, IEEE Transactions
on, vol. 26, no. 3, pp. 531 – 541, 2010.
[12] A. Rida, V. Fernandez, and M. Gijs, “Long-range transport of magnetic microbeads
using simple planar coils placed in a uniform magnetostatic field,” Applied Physics
Letters, 2003.
[13] S.-J. Han, H. Yu, B. Murmann, N. Pourmand, and S. Wang, “A high-density magnetoresistive biosensor array with drift-compensation mechanism,” Solid-State Circuits
Conference, 2007. ISSCC 2007. Digest of Technical Papers. IEEE International, pp.
168 – 594, Feb 2007.
[14] Y. Liu, N. Sun, H. Lee, R. Weissleder, and D. Ham;, “Cmos mini nuclear magnetic
resonance system and its application for biomolecular sensing,” Solid-State Circuits
Conference, 2008. ISSCC 2008. Digest of Technical Papers. IEEE International, pp.
140 – 602, Feb 2008.
[15] K. Skucha, P. Liu, M. Megens, J. Kim, and B. Boser, “A compact Hall-effect sensor
array for the detection and imaging of single magnetic beads in biomedical assays,”
Solid-State Sensors, Actuators and Microsystems Conference (TRANSDUCERS), 2011
16th International, pp. 1833–1836, 2011.
[16] P.-A. Besse, G. Boero, M. Demierre, V. Pott, and R. Popovic, “Detection of a single
magnetic microbead using a miniaturized silicon hall sensor,” Applied Physics Letters,
vol. 80, no. 22, pp. 4199–4201, 2002.
[17] U. Lehmann, M. Sergio, E. Dupont, and E. Labonne, “Actuation and detection of
magnetic microparticles in a bioanalytical microsystem with integrated cmos chip,”
Nanosystems Design and Technology, Jan 2009.
[18] H. Wang, Y. Chen, A. Hassibi, A. Scherer, and A. Hajimiri, “A frequency-shift CMOS
magnetic biosensor array with single-bead sensitivity and no external magnet,” SolidState Circuits Conference - Digest of Technical Papers, 2009. ISSCC 2009. IEEE International, pp. 438–439, 2009.
146

[19] H. Wang, C. Sideris, and A. Hajimiri, “A frequency-shift based CMOS magnetic biosensor with spatially uniform sensor transducer gain,” Custom Integrated Circuits Conference (CICC), 2010 IEEE, pp. 1–4, 2010.
[20] H. Wang, S. Kosai, C. Sideris, and A. Hajimiri, “An ultrasensitive CMOS magnetic
biosensor array with correlated double counting noise suppression,” Microwave Symposium Digest (MTT), 2010 IEEE MTT-S International, pp. 616–619, 2010.
[21] F. Abu-Nimeh and F. Salem, “Integrated magnetic array for bio-object sensing and
manipulation,” IEEE Biomedical Circuits and Systems Conference, Nov 2010.
[22] ——, “Electro-magnetic sensing and actuation array on silicon substrate platforms,”
IEEE Nanotechnology Materials and Devices Conference, Oct 2010.
[23] J. Black, “Electromigration—a brief survey and some recent results,” Electron Devices,
IEEE Transactions on, vol. 16, no. 4, pp. 338 – 347, 1969.
[24] C. Hu and Q. Lu, “A unified gate oxide reliability model,” Reliability Physics Symposium
Proceedings, 1999. 37th Annual. 1999 IEEE International, pp. 47 – 51, 1999.
[25] J. Keane, X. Wang, D. Persaud, and C. Kim, “An all-in-one silicon odometer for separately monitoring hci, bti, and tddb,” Solid-State Circuits, IEEE Journal of, vol. 45,
no. 4, pp. 817–829, 2010.
[26] A. Neumann, T. Reske, M. Held, K. Jahnke, C. Ragoß, and H. R. Maier, “Comparative
investigation of the biocompatibility of various silicon nitride ceramic qualities in vitro,”
Journal of Materials Science: Materials in Medicine, vol. 15, pp. 1135–1140, 2004.
[27] P. George, A. Lyckman, D. LaVan, A. Hegde, Y. Leung, R. Avasare, C. Testa, P. Alexander, R. Langer, and M. Sur, “Fabrication and biocompatibility of polypyrrole implants
suitable for neural prosthetics,” Biomaterials, vol. 26, no. 17, pp. 3511–3519, 2005.
[28] A. Hoogerwerf and K. Wise, “A three-dimensional microelectrode array for chronic
neural recording,” Biomedical Engineering, IEEE Transactions on, vol. 41, no. 12, pp.
1136 – 1146, 1994.
[29] T. Jochum, T. Denison, and P. Wolf, “Integrated circuit amplifiers for multi-electrode
intracortical recording,” Journal of Neural Engineering, vol. 6, no. 1, p. 012001, Jan
2009.
147

[30] Philips.com. (2011, Jan) Next generation point-of-care testing. [Online]. Available: http://www.newscenter.philips.com/main/standard/news/backgrounders/2010/
20100107 handheld diagnostics.wpd
[31] A. Bausch, F. Ziemann, A. Boulbitch, and K. Jacobson, “Local measurements of viscoelastic parameters of adherent cell surfaces by magnetic bead microrheometry,” Biophysical Journal, Jan 1998.
[32] G. P. Hatch and R. E. Stelter, “Magnetic design considerations for devices and particles used for biological high-gradient magnetic separation (hgms) systems,” Journal of
Magnetism and Magnetic Materials, Jan 2001.
[33] Q. A. Pankhurst, J. Connolly, S. K. Jones, and J. Dobson, “Applications of magnetic
nanoparticles in biomedicine,” Journal of Physics D: Applied Physics, vol. 36, no. 13,
p. R167, 2003.
[34] X. Zhu, K. Han, and G. Li, “Magnetic nanoparticles applied in electrochemical detection
of controllable dna hybridization,” Anal Chem, vol. 78, no. 7, pp. 2447–9, Apr 2006.
[35] MicroMod. (2010, Dec) Nanomag-d. [Online]. Available: http://www.micromod.de/
scripts/p1.asp?sid=567178210&lng=e
[36] Invitrogen. (2010, Nov) Dynal beads myone. [Online]. Available: http://products.
invitrogen.com/ivgn/product/35601
[37] Ademtech. (2010, Dec) Bio-adembeads protein g. [Online]. Available:
//www.ademtech.com/products.aspx

http:

[38] A. Wooding, M. Kilner, and D. Lambrick, “Studies of the double surfactant layer stabilization of water-based magnetic fluids,” Journal of colloid and interface science, vol.
144, no. 1, pp. 236–242, 1991.
[39] A.-H. Lu, E. L. Salabas, and F. Sch¨th, “Magnetic nanoparticles: Synthesis, protection,
u
functionalization, and application,” Angewandte Chemie International Edition, vol. 46,
no. 8, pp. 1222–1244, 2007.
[40] Y. Jun, J. Lee, and J. Cheon, “Chemical design of nanoparticle probes for highperformance magnetic resonance imaging,” Angewandte Chemie International Edition,
vol. 47, no. 28, pp. 5122–5135, 2008.
148

[41] H. Lee, D. Ham, and R. M. Westervelt, CMOS biotechnology.
York Inc, 2007.

Springer-Verlag New

[42] C. Seeton, “Viscosity–temperature correlation for liquids,” Tribology Letters, vol. 22,
no. 1, pp. 67–78, 2006.
[43] P. Kennedy, Flow analysis of injection molds. Hanser Munich, 1995.
[44] R. Wirix-Speetjens, W. Fyen, K. Xu, J. D. Boeck, and G. Borghs, “A force study of
on-chip magnetic particle transport based on tapered conductors,” Magnetics, IEEE
Transactions on, vol. 41, no. 10, pp. 4128–4133, 2005.
[45] E. Verwey, J. Overbeek, and K. van Nes, Theory of the stability of lyophobic colloids:
the interaction of sol particles having an electric double layer. Elsevier New York, 1948.
[46] B. Derjaguin and L. Landau, “Theory of stability of highly charged lyophobic sols and
adhesion of highly charged particles in solutions of electrolytes,” Acta Physicochim.
URSS, vol. 14, pp. 633–52, 1941.
[47] V. Parsegian, Van der Waals forces: a handbook for biologists, chemists, engineers, and
physicists. Cambridge University Press, 2006.
[48] S. Behrens and D. Grier, “The charge of glass and silica surfaces,” The Journal of
Chemical Physics, vol. 115, p. 6716, 2001.
[49] C. J. Van Oss, M. K. Chaudhury, and R. J. Good, “Interfacial lifshitz-van der waals
and polar interactions in macroscopic systems,” Chemical Reviews, vol. 88, no. 6, pp.
927–941, 1988. [Online]. Available: http://pubs.acs.org/doi/abs/10.1021/cr00088a006
[50] J. Gregory, “Approximate expressions for retarded van der Waals interaction,” Journal
of Colloid and Interface Science, vol. 83, no. 1, pp. 138–145, 1981.
[51] G. Kar, S. Chander, and T. S. Mika, “The potential energy of interaction between
dissimilar electrical double layers,” Journal of Colloid and Interface Science, vol. 44,
no. 2, pp. 347 – 355, 1973.
[52] E. Reiner and C. Radke, “Double layer interactions between charge-regulated
colloidal surfaces:
pair potentials for spherical particles bearing ionogenic
surface groups,” Advances in Colloid and Interface Science, vol. 47, pp. 59
– 147, 1993. [Online]. Available: http://www.sciencedirect.com/science/article/
B6V5F-44J18J8-29/2/539f491f25c63a96eba33f9b82331d48
149

[53] C. Liu, T. Stakenborg, S. Peeters, and L. Lagae, “Cell manipulation with magnetic
particles toward microfluidic cytometry,” Journal of Applied Physics, vol. 105, no. 10,
p. 102014, 2009.
[54] M. Jackson, Molecular and cellular biophysics. Cambridge University Press, 2006.
[55] R. Tilley, Understanding solids: the science of materials. Wiley, 2004.
[56] BangsLabs. (2010, Dec) Polymer microspheres. [Online]. Available:
bangslabs.com/products/polymer microspheres

http://www.

[57] H. Kaeslin, Digital integrated circuit design: from VLSI architectures to CMOS fabrication. Cambridge University Press, 2008.
[58] T. Lee, The design of CMOS radio-frequency integrated circuits. Cambridge University
Press, 2004.
[59] A. M. Niknejad. (2010, May) Analysis and simulation of spiral inductors and
transformers for ics. [Online]. Available: http://rfic.eecs.berkeley.edu/$\sim$niknejad/
asitic.html
[60] Nanosurf. (2011, Jan) The nanosurf easyscan 2 afm. [Online]. Available:
//www.nanosurf.com/index.cfm?content=040220&spr=en

http:

[61] AppNano. (2011, Jan) Technical specifications sheet, probe model: Magt-hm. [Online].
Available: http://www.appnano.com/images/pdfs/MAGT-HM.pdf
[62] L. Li, W. Qureshi, X. Liu, and A. Mason, “Amperometric instrumentation system with
on-chip electrode array for biosensor application,” Biomedical Circuits and Systems
Conference (BioCAS), 2010 IEEE, pp. 294 –297, nov. 2010.
[63] F. Abu-Nimeh and F. Salem, “Digital class e power amplifier for the rf band,” Mechatronics and Automation, Proceedings of the 2006 IEEE International Conference on,
pp. 1548 – 1552, May 2006.
[64] P. Cruise, C.-M. Hung, R. Staszewski, O. Eliezer, S. Rezeq, K. Maggio, and D. Leipold,
“A digital-to-rf-amplitude converter for gsm/gprs/edge in 90-nm digital cmos,” Radio
Frequency integrated Circuits (RFIC) Symposium, 2005. Digest of Papers. 2005 IEEE,
pp. 21 – 24, Jun 2005.
150

[65] F. Abu-Nimeh and F. Salem, “Cmos integrated electro-magnetic force actuator,” Circuits and Systems, 2006. MWSCAS ’06. 49th IEEE International Midwest Symposium
on, vol. 2, pp. 395 – 398, Jul 2006.
[66] D. Ham and A. Hajimiri, “Virtual damping in oscillators,” in Custom Integrated Circuits
Conference, 2002. Proceedings of the IEEE 2002, 2002, pp. 213–216.
[67] A. Hajimiri, “Concepts and methods in optimization of integrated LC VCOs,” SolidState Circuits, IEEE Journal of, vol. 36, no. 6, pp. 896–909, 2001.
[68] A. Hajimiri and T. Lee, “Design issues in CMOS differential LC oscillators,” Solid-State
Circuits, IEEE Journal of, vol. 34, no. 5, pp. 717–724, 1999.
[69] J. Barthel, “Motion Control of Magnetic Beads on a CMOS Chip,” Unpublished Report,
2010.
[70] A. Beyzavi and N. Nguyen, “Modeling and optimization of planar microcoils,” Journal
of Micromechanics and Microengineering, vol. 18, p. 095018, 2008.
[71] NeuroNexus. (2011, Jan) Probe packages. [Online]. Available:
neuronexustech.com/

http://www.

[72] J. Zheng, P. Lamoureux, V. Santiago, T. Dennerll, R. E. Buxbaum, and S. R. Heidemann, “Tensile regulation of axonal elongation and initiation.” Journal of Neuroscience,
vol. 11, no. 4, pp. 1117–1125, Apr. 1991.
[73] T. J. Dennerll, P. Lamoureux, R. E. Buxbaum, and S. R. Heidemann, “The cytomechanics of axonal elongation and retraction,” The Journal of cell biology, vol. 109, no.
6 Pt 1, pp. 3073–3083, Dec. 1989.
[74] J. Zheng, R. E. Buxbaum, and S. R. Heidemann, “Measurements of growth cone adhesion to culture surfaces by micromanipulation.” The Journal of cell biology, vol. 127,
no. 6 Pt 2, pp. 2049–2060, Dec. 1994.
[75] M. O’Toole, R. Latham, R. M. Baqri, and K. E. Miller, “Modeling mitochondrial dynamics during in vivo axonal elongation.” Journal of theoretical biology, vol. 255, no. 4,
pp. 369–377, Dec. 2008.
[76] B. Sicaeros and D. K. O’Dowd, “Preparation of neuronal cultures from midgastrula
stage Drosophila embryos.” Journal of visualized experiments : JoVE, no. 5, p. 226,
2007.
151

[77] S. Heidemann, M. Reynolds, K. Ngo, and P. Lamoureux, “The culture of chick forebrain
neurons,” Methods in cell biology, vol. 71, pp. 51–65, 2003.
[78] P. Lamoureux, S. R. Heidemann, N. R. Martzke, and K. E. Miller, “Growth and elongation within and along the axon.” Developmental neurobiology, vol. 70, no. 3, pp.
135–149, Feb. 2010.
[79] E. Delivopoulos, A. F. Murray, N. K. MacLeod, and J. C. Curtis, “Guided growth of
neurons and glia using microfabricated patterns of parylene-C on a SiO2 background.”
Biomaterials, vol. 30, no. 11, pp. 2048–2058, Apr. 2009.
[80] M. Abr`moff, P. Magalh˜es, and S. Ram, “Image processing with imagej,” Biophotonics
a
a
international, vol. 11, no. 7, pp. 36–42, 2004.

152