CMOS UNIVERSAL REAL-TIME LABEL-FREE DNA ANALYSIS SYSTEM-ON-CHIP by Hamed Mazhab Jafari A thesis submitted in conformity with the requirements for the degree of Doctorate of Philosophy Graduate Department of Electrical and Computer Engineering University of Toronto Copyright c ⃝ 2013 by Hamed Mazhab Jafari
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
As described in Section 1.2, many conventional DNA-based pathogen detection sys-
tems rely on the gold-standard optical signal readout. Such systems are costly, require
bulky optics, typically require PCR for high sensitivity, and are often slow due to the
serial nature of the optical signal readout. DNA analyzers with non-optical parallel elec-
tronic readout, such as those described in Section 1.3, are a promising cost-effective,
highly sensitive and fast alternative to optical methods, often with their own limitations.
CHAPTER 1. INTRODUCTION 18
Another such method is known as amperometric electrochemical DNA sensing.
Amperometric DNA sensing methods have gained interest in recent years but are not
yet as developed as the optical or pH-based DNA sensing techniques. They have the
potential to reduce the instrumentation cost and provide portable diagnostic platform
for DNA analysis and detection. The main advantages of amperometric sensing versus
gold-standard optical are the elimination of light as the source of excitation and thus of
bulky optical equipment, and the real-time nature of DNA sensing. This provides an
opportunity to design and prototype a portable real-time DNA sensing microsystem for
point-of-care applications.
The polymerase chain reaction (PCR) is commonly utilized to increase the con-
centration of the target DNA in a sample, thus increasing the detection accuracy. As
described in Section 1.1.3, PCR is a procedure in biochemistry to amplify a single or a
few copies of a DNA sequence, generating up to millions of copies of the same DNA
sequence [62, 63]. This method relies on thermal cycling which consists of cycles of
repeated heating and cooling for DNA melting and enzymatic replication of the DNA.
This is generally costly and time consuming.
Signal transduction is typically performed by coupling the target sequence with a
reduction-oxidation (redox) or fluorescent label with a well-defined and easily detected
electrochemical or optical signature, respectively. This is known as label-based sensing
which requires sample labeling and the corresponding cost and time.
Label-free detection of DNA hybridization is also possible, by monitoring the elec-
trical signals, such as current, voltage, impedance, and conductance, at the sensor’s
solid-liquid interface. There are several label-free electrical nucleic acid detection plat-
forms, but most of these techniques rely on off-chip, expensive, and bulky instrumenta-
tion for signal readout and processing, characteristics that make the techniques unsuit-
able for many point-of-need and in-field applications.
CHAPTER 1. INTRODUCTION 19
Amperometric electrochemical DNA sensors [20, 21, 23–28, 99] have emerged as
a low-cost, high-throughput, and real-time alternative to conventional optical and pH-
based sensory methods. Electrochemical amperometric DNA analysis techniques have
the potential to provide real-time, label-free, PCR-free sensing in portable detection
platforms.
1.4.1 Three-electrode Sensing Configuration
Electrochemical sensing involves a chemical reaction that occurs on the surface of an
electrode in the presence of a potential difference. The chemical reaction results in a
charge transfer which results in a current or a voltage output. The current-mode output
has a wide dynamic range and is commonly used. Sensing methods in which a current
is measured are known as amperometric electrochemical sensing methods [66, 67].
A block diagram of a three-electrode electrochemical amperometric sensing sys-
tem is depicted in Fig. 1.8 [68]. It consists of a working electrode (WE), a reference
electrode (RE), a counter electrode (CE), a waveform generator and a current-to-digital
channel. In this configuration the waveform generator drives the reference electrode.
The working electrode is held at a known potential, VWE by the current-to-digital chan-
nel. In general the voltage difference between the working and reference electrode is set
to VREDOX . This is the voltage at which the chemical under test is reduced or oxidized.
The IWE current generated due to the voltage difference between the working and ref-
erence electrodes is recorded by the current-to-digital channel. If the voltage difference
is set to VREDOX , the resulted current is called IREDOX , which is the current generated
due to the reduction and oxidization of the chemical species.
In applications where the redox current level is high (greater than 1µA), there could
be a significant voltage drop between the working and reference electrode (due to the
finite impedance of the electrolyte solution) if the redox current is to flow from the
CHAPTER 1. INTRODUCTION 20
DIGITAL
OUTPUTCURRENT-TO
DIGITAL
CHANNEL
RE
RE CE
PROBE
DNA
TARGET
DNA
CE
WE
WAVEFORM
GENERATOR
REV
CEV
WEV
REDOXI
WEV
Figure 1.8: Conceptual view of an electrochemical amperometric sensing microsystem.
reference electrode to the working electrode. As a results of this voltage drop, the ac-
tual voltage in the vicinity of the working electrode will not be the same as the voltage
that was set by the reference electrode. This can significantly affect the redox poten-
tial required by the electrochemical experiment. To solve this issue a third (counter)
electrode is added to the electrochemical cell. In this configuration no current flows
from the reference electrode to the working electrode. The counter electrode provides
the current required to keep the voltage difference between the working and reference
electrodes exactly at the redox potential. In general the counter electrode is not required
for low-current applications (less than 1µA).
In amperometric electrochemical DNA sensing applications the working electrode
is coated with a single-stranded DNA (ssDNA) probe. Binding of the probe ssDNA
with the target ssDNA results in a variation of the working electrode surface properties
such as impedance or surface charge. The variation of the surface properties results in a
change in the recorded value and waveform features of the redox current, thus indicating
CHAPTER 1. INTRODUCTION 21
CWE
RS
RCT
CDB
+
+
-
-
-
-
-
+
+
+
WERE
RE WE
ELECTROLYTE
SOLUTION
DIFFUSION
LAYER
DOUBLE
LAYER
+
-PROBE
-TARGET
-
-
-
-
-
++
+ +
-REDUCTION
-OXIDATION+
+
-H O2
Figure 1.9: Charge profile of the electrode-electrolyte interface and its equivalent circuit model.
the thermodynamics and kinetics of chemical reactions at the sensory interface. In most
biochemical sensing applications the recorded redox current is in the range of pA to
nA [69–71].
Electrochemical amperometry provides a wide range of concentration measurement
because of its linear behavior. Cyclic voltammetry and impedance spectroscopy are
two common amperometric sensing methods. In general, the reference electrode poten-
tial, VRE , is set to a bidirectional ramp voltage for cyclic voltammetry or to a small-
amplitude sinusoid for impedance spectroscopy.
1.4.2 Electrode-Electrolyte Interface
When an electrode is in the presence of an electrolyte solution containing solvated and
unsolvated anions and cations and water molecules, a layer of charge forms at the
electrode-electrolyte interface in order to maintain the electrical neutrality, as shown
in Fig. 1.9. This layer can be electrically modeled by a set of resistors and capacitors.
A generic R-C biosensor impedance model is also depicted in Fig. 1.9. The counter
electrode is omitted for simplicity. In this model RS represents the electrolyte resistance
between the working and reference electrodes, CWE represents the diffusion layer ca-
CHAPTER 1. INTRODUCTION 22
pacitance, CDB models the interfacial double-layer capacitance at the WE-electrolyte
interface and RCT models the charge transfer resistance at the WE-electrolyte inter-
face [84]. The double layer consists of solvated positive ions physically separated from
the working electrode surface by a monolayer of water molecules and ions attached to
the working electrode surface. The diffusion layer is the layer beyond the double layer
where the capturing probe molecules are located. The ssDNA probes are negatively
charge, thus attracting the positively charged ions from the electrolyte solution.
1.4.3 Reduction-Oxidation Current
An example of the mechanism of generation of the redox current is also shown in
Fig. 1.9. In the presence of a potential difference between the working electrode and
the reference electrode the electroactive chemical species of the electrolyte solution
are either reduced (gain an electron) or oxidized (lose an electron). The amount of
charge transferred between the electrolyte solution and the working electrode depends
on the surface charge. The surface charge in term depends on the density of the target
molecules that are bonded to the probes. Measuring the current corresponding to the
charge transfer yields a measure of the concentration of the target molecules.
1.4.4 Labeled vs Label-free DNA Sensing
Amperometric electrochemical DNA sensors can be divided into two main categories:
label-based and label-free. In the label-based techniques an electroactive chemical label
is attached to the target DNA to provide a positive electrochemical signal. This signal
can also be derived from a redox indicator that associates with a ssDNA differently than
with a DNA duplex. On the other hand, a label-free DNA sensor does not require any
supporting redox species, indicator molecule or enzymes to provide the electrochemical
signal. This method relies on the oxidation of individual DNA bases or change in the
CHAPTER 1. INTRODUCTION 23
electrode-electrolyte interface properties such as capacitance and resistance changes
due to DNA hybridization.
1.4.5 Electrode Material
In general an electrochemical cell consists of a working electrode, a counter electrode
and a reference electrode.
A wide range of materials are used to fabricate working electrodes. These include
carbon, polymers and metals as shown in Fig. 1.10. The working electrode material is
chosen such that it provides a polarizable interface when it is in contact with the elec-
trolyte solution [26]. In this work gold (Au) has been chosen as the working electrode
material. Gold has the advantage of being electrochemically inactive and it does not
react with the electrolyte solution under normal operating conditions. Also, gold work-
ing electrodes can easily be fabricated on a CMOS chip, forming on-chip integrated
working electrodes. Another advantage of gold working electrodes is that it enables
straight-forward DNA attachment chemistry [26].
The reference electrode must meet the following criteria [78]:
• Must have a high exchange current density and be non-polarizable.
• The electrode potential must be reproducible.
• The electrode potential drift must be small during the life time of the electrode.
Depending on the application and solution chemistry, different reference electrodes
can be used. Due to its simple construction and the above-mentioned properties, silver
(Ag) coated by a layer of silver chloride (Ag/AgCl) is the most commonly used material
for a reference electrode. A conventional commercially available reference electrode
consists of a silver wire coated with silver chloride that is immersed in a Cl ion solution
CHAPTER 1. INTRODUCTION 24
such as KCl. The silver wire and the KCl solution are enclosed in a glass tube and
separated from the solution that is under test by an ion exchange membrane and a salt
bridge [79].
Another common type of the reference electrode is the saturated calomel electrode
(SCE), which consists of mercury covered by mercury chloride pates that are in contact
with saturated KCl solution and enclosed in a glass tubing. The advantage of this elec-
trode is its simple construction and stable potential. However, the toxicity of mercury
makes its use hazardous [80].
Another common reference electrode in electro-chemical sensing is the standard
hydrogen electrode. The electrode is constructed with a platinum surface that is covered
with black platinum. The black platinum helps with the reduction of protons and is a
catalyst for the electrode reaction. The platinum is submerged in an acidic solution.
The reference electrode cell is confined in a glass enclosure and is pressure- controlled
to keep the hydrogen pressure fixed and to maintain the electrode’s potential [81, 82].
Due to its non-reactive nature, platinum (Pt) is the most commonly used material
choice for counter electrode. Other nobel metals such as gold can be used as well [26].
In this work Au electrodes are used for the working and counter electrodes and off-
chip commercially available Ag/AgCl electrodes are used as the reference electrode.
In future work we are planning to implement the Ag/AgCl electrode on-chip using
the post-processing method described in [83] applied to the implemented on-chip Al
reference electrode bases.
1.5 Amperometric Electrochemical Sensing Methods
Different amperometric electrochemical sensing methods are distinguished by the wave-
form of the voltage difference between the working and reference electrode. The two
CHAPTER 1. INTRODUCTION 25
WE
(c)(b)
60µm
(a)
Figure 1.10: Working electrode material examples: (a) diamond [71], (b) polymer [20], and (c)flat gold [70].
OXIDATION
VOLTAGE
REDUCTION
VOLTAGE
VW
E-V
RE
REVERSE
SCAN
FORWARD
SCAN
TIME
Epc
Epa
VWE
I RE
DO
X
-VRE
OXIDATION
PEAK
REDUCTION
PEAK
BACKGROUNDTARGET CHEMICAL
REVERSE
SCAN
FORWARD
SCAN
Epa Epc
(a) (b)
Figure 1.11: Fast-scan cyclic voltammetry principle of operation (a) cyclic redox potential ap-plied between the working and reference electrode, and (b) cyclic voltammogram in the absenceand presence of the target chemical.
most commonly used electrochemical sensing methods are: cyclic voltammetry (CV)
and impedance spectroscopy (IS). Both have to be thermally regulated as the redox cur-
rent is highly dependent on the temperature [19]. In this work all these three methods
are implemented on-chip to perform thermally-regulated label-free DNA detection and
cross-validation.
CHAPTER 1. INTRODUCTION 26
1.5.1 Fast-Scan Cyclic Voltammetry
In the fast-scan cyclic voltammetry (CV) method, a cyclic ramp potential is intermit-
tently applied between the working and reference electrodes, as shown for one period
in Fig. 1.11(a). The time between scans, which can vary, determines the temporal reso-
lution of the technique. The halt time prevents successive scans from influencing each
other. The cyclic voltammogram shown in Fig. 1.11(b), represents the redox current
versus the applied redox potential and provides chemical information about the sub-
stance under measurement. For example, the location of the reduction and oxidation
peaks acts as a unique chemical identifier for different chemicals. The reduction and
oxidation peak heights change for different chemical concentrations. A background
current is generated due to the transient changes of the applied voltage (Fig. 1.11(b)).
This current occurs mainly because of the charging and discharging of the double layer
capacitance associated with the electrode-electrolyte interface. The background current
is proportional to the scan rate and also to the double layer capacitance.
1.5.2 Amperometric Impedance Spectroscopy
Impedance spectroscopy (IS) is a popular method of quantitative and qualitative moni-
toring of chemical reactions in many biosensors. A wide range of biosensors have been
developed which rely on impedance spectroscopy, including sensors for detection of
enzymes, antibodies and DNA [85, 86].
In the IS method, a small-amplitude perturbing sinusoidal voltage, as shown in
Fig. 1.12(a), is applied between working and reference electrode and the subsequent
current response is measured. The impedance is calculated as the ratio of the applied
voltage to the resulting current. A frequency response plot of a biosensor, as shown
in Fig. 1.12(b), is computed by applying a variable frequency excitation signal and
computing the biosensor impedance at each frequency point. The biosensor frequency
CHAPTER 1. INTRODUCTION 27
VWE-VRE
TIME FREQUENCY
CHANGE IN CHEMICAL
CONCENTRATION
I RE
DO
X
(a) (b)
Figure 1.12: Impedance spectroscopy principle of operation (a) potential difference betweenworking and reference electrode, and (b) change in redox current frequency response due to thechange in the chemical concentration.
response is directly related to the concentration of the electroactive substance.
1.5.3 Temperature Regulation for DNA Analysis
Temperature reliance is a key constraint in DNA sensing. DNA hybridization is highly
dependent on the temperature [4–6,19], and as a result, temperature plays an important
role in the sensitivity of DNA sensors. It is shown in [7] that the redox current is highly
dependent on the temperature. Generally the redox current drops by 10 percent, given
a 10 degree increase in the temperature above the ambient temperature. A temperature
regulation system for DNA sensing that has on chip-heating, temperature sensing and
regulation circuits, would be of great benefit for both the characterization and sensing
of DNA. Such a system would increase the accuracy of the measurements, improve the
speed and efficiency of the analysis, and also reduce reagent usage and improve the
sensitivity limit of the measurements.
CHAPTER 1. INTRODUCTION 28
1.5.4 Commercial Amperometric Instruments
Early Warning Inc. offers a beta-version of its amperometric biohazard water analyzer
that employs carbon nanotube DNA sensors to detect E.coli O157:H7, as shown in
Fig. 1.13(a). The analyzer weighs 200 kg and promises to detect up to 25 pathogens
in 10L of water within three hours, with the sensitivity of one bacterium cell per
100mL [38].
Palm Sens Inc offers Palmsens 3, Fig. 1.13(b), which is a portable wireless cur-
rent recording frontend (potentiostat) for electrochemical recording applications [39].
The unit consist of one current-to-digital recording channel and a wireless bluetooth
transmitter. The unit weighs 450g and achieves an 8 hour wireless battery powered
operation. It has an input dynamic range of 100pA to 10mA.
Metrohm Inc offers PSTAT mini [40], Fig. 1.13(c), which is a single channel portable
potentiostat weighting 0.985kg. The unit achieves an input dynamic range of 2nA to
200A and can be directly interfaced to DNA sensing microelectrodes.
The bulky and expensive commercially available instruments described in Sections
1.2.2, 1.2.4 and 1.5.4 generally attain very high sensitivity (0.1nM). Routine quantita-
tive pathogen detection with low-cost, fast (under one hour), highly specific (multiple
strands), and medium to high sensitivity (10µM down to 0.1nM) biosensors in an inte-
grated and portable lab-on-chip platform is a sought-after goal that is not yet commer-
cially available.
1.6 Integrated Circuits for Amperometric DNA Sensing
Several electrochemical DNA detection microsystems utilizing cyclic voltammetry (CV)
have been reported [20, 21, 26, 96, 97, 99]. The design in [99] is a 50-channel pro-
grammable electrochemical biosensor array implemented in a 0.13µm standard CMOS
CHAPTER 1. INTRODUCTION 29
(a) (b) (c)
Figure 1.13: Commercially available amperometric instruments: (a) Early Warning biohazardwater analyzer [38], (b) Palmsens 3 [39], and (c) PSTAT mini [40].
technology. The microsystem includes flat gold electrodes and analog recording chan-
nels, and utilizes impedance spectroscopy for DNA detection. The implementation
in [96] consists of one recording channel and 24×16 recording electrodes implemented
in a 0.5µm CMOS technology. The design, which consists of a three-electrode regula-
tion loop and an analog recording channel, utilizes CV for DNA detection and analysis.
A 128-channel DNA analysis microsystem implemented in a 0.5µm CMOS technology
is presented in [21]. The microsystem consists of on-chip gold electrodes, a three-
electrode regulation loop, and an in-pixel ADC. The design presented in [20] is imple-
mented in a 0.5µm CMOS technology and consists of 24 recording channels with an
in-channel ADC, 24×24 polymer-functionalized sensing electrodes, and a temperature
sensor. The microsystem utilizes CV for DNA detection and analysis. The design
in [26] presents the first fully-integrated CMOS DNA analysis microsystem, which
consists of 16 recording channels, a three-electrode regulation loop, a flat gold DNA
sensing microelectrode, and an in-channel ADC.
Several DNA analysis SoCs based on impedance spectroscopy have been reported
[84, 130, 131]. The design in [130] is a single-channel impedance extractor based on a
lock-in amplifier that extracts the sensor impedance from 1Hz to 10kHz. The implemen-
tation in [84] is a 100-channel impedance-to-digital converter based on a delta-sigma
modulator capable of extracting sensor impedance from 1mHz to 10kHz at the expense
CHAPTER 1. INTRODUCTION 30
of a long conversion cycle. A direct conversion receiver without an on-chip ADC [131]
extracts the electrode impedance from 10Hz to 50MHz at the cost of consuming 104mW
of power.
1.7 Main Specifications for Amperometric Electrochem-
ical DNA Sensing Integrated Circuits Design
1.7.1 Accuracy
Amperometric electrochemical DNA sensors use selective binding and interaction be-
tween probe molecules and target molecules to generate a detectable electrical signal
that correlates with the presence or absence of target molecules. The main components
of any electrochemical DNA sensing system are a molecular recognition layer, a sig-
nal generator, and a readout unit. To generate an electrical target-specific signal, first
the target molecule needs to come into proximity with the probe molecule. The target
molecule’s motion is typically dominated by diffusive spreading, which is a probabilis-
tic mass-transfer. In addition, the nature of the chemical bond forming the interaction
between the probe molecules and the analyte solution is also probabilistic, and the result
is more uncertainty (noise) in the biosensor [9–12, 23].
In addition to these processes, the interrogation signal generation circuits and the
readout circuits also add noise to the biosensor. On top of the noise generated by the
binding process and the electronic circuit, the binding of other species to the probe
molecules (non-specific binding) also adds noise to the system. Non-specific binding
is generally less frequent than specific binding if the concentration of both the specific
and non-specific analyte is in the same order of magnitude. If the concentration of non-
specific molecules becomes much larger than the concentration of the target molecules,
CHAPTER 1. INTRODUCTION 31
then the non-specific binding (noise) can overcome the detectable signal and thus in-
creases the minimum detectable level (MDL) [13]. Based on the biosensor noise, we
can define the MDL, the highest detection level (HDL), and the detection dynamic range
(DR).
In [23], a noise model was developed to capture the biosensor MDL, HDL, DR, and
SNR (signal-to-noise ratio). The model was applied to a simple biosensor that consisted
of a signal generator, an electrode-eletrolyte interface, and a generalized recording cir-
cuit. Fig. 1.14 plots the SNR of the generalized biosensor model vs. the analyte concen-
tration (DNA molecule concentration) in different background solution concentrations.
In an ideal system, increasing the concentration of analyte generates more noise, but, at
the same time, generates larger signals, a process that results in a linear increase of SNR
vs. analyte concentration. In a practical biosensor, this outcome is not possible, since
the probe saturation occurs, and, as a result, SNR drops as the analyte concentration
increases. As shown in Fig. 1.14, the HDL is limited by saturation and the finite value
of the probe density. The maximum SNR of the generalized biosensor model in [21]
is about 55dB, and the limiting factor is probe saturation and biochemical noise. The
overall dynamic range of the system for an SNR of 30dB varies from 84dB to 104dB,
depending on background solution concentration.
Based on Fig. 1.14 results, the conclusion may be drawn that the minimum SNR of
the interrogation signal generation and recording circuitry should be larger than 55dB,
which corresponds to approximately effective number of bits (ENOB) of 9-bits.
The model developed in [23] is only valid for flat gold electrode. The concentration
range that is presented in this plot is similar to the DNA concentration range that was
successfully detected by the CMOS DNA detection microsystem as presented in Section
3.8. The model developed in [23] is not valid for the nanostructured electrodes. Similar
model for the nanostructured electrodes have not been developed in the literature.
CHAPTER 1. INTRODUCTION 32
Figure 1.14: Biosensor SNR vs target analyte concentration [23].
1.7.2 Dynamic Range
In [26–28, 69, 70, 99], the output redox current level from a typical on-chip working
electrode (10µm to 100µm diameter) in electrochemical DNA-sensing applications is
shown to be in the range of low pA to mid nA. As a result, in this work, the current
range specification of the system is set to cover a range from one pA to 500 nA.
1.7.3 Frequency Range
In electrochemical sensing methods such as cyclic voltammetry and impedance spec-
troscopy, a high-frequency interrogation waveform generates a large background cur-
rent due to the charging and discharging of the double-layer electrode-electrolyte ca-
pacitor. The large background current puts stringent requirements on the accuracy and
dynamic range of the electrochemical recording front end. To relax the accuracy and
CHAPTER 1. INTRODUCTION 33
dynamic range requirements in general, the interrogation waveform frequency range of
electrochemical sensing systems is limited to a maximum 10 kHz [26].
1.7.4 Non-electrical Design Specification
The ultimate goal in this thesis is design and development of a low-cost semiconductor-
based technology for rapid and sensitive in-field DNA analysis of various DNA strands.
As a result, small form-factor, low-cost, high-throughput, as well as label-free and PCR-
free operation are the most important non-electrical design requirements of the DNA
sensing microsystem.
The size and the cost of the DNA sensing microsystem needs to be minimized to
enable widespread in-field utilization. This would require integration of both DNA
sensing electrodes and circuitry on the same chip. CMOS integrated circuits provide
versatile signal acquisition and processing functionalities at a low cost. The large scale
of integration allows for hundreds of sensors to be placed directly on a chip for simulta-
neous high-throughput sensing. On-chip microsensors are directly suitable for CMOS
surface functionalization yielding high sensitivity and selectivity thus eliminating the
need for the PCR amplification step which is both time consuming and expensive. Di-
rect deposition of sensors on a chip eliminates costly excessive wiring and minimizes
the interference noise. Label-free operation of the microsystem also eliminates the
costly and time-consuming labeling process further reducing the overall cost of the
DNA sensing microsystem.
CHAPTER 1. INTRODUCTION 34
WERE
ADC
AM
PE
RO
ME
TR
IC
RE
AD
OU
T
WE
CC
V
EXCITATION
WAVEFORM
IREDOX
TIA
CURRENT
ACQUISITION
CIRCUIT OUTV
OUTI
BIOSENSOR-
FUNCTIONALIZED
SURFACE
Figure 1.15: Conceptual view of current acquisition circuits for electrochemical amperometricsensory system.
1.8 Additional Considerations for DNA-sensing Amper-
ometric Integrated Circuits Design
1.8.1 Low-Level Current Acquisition for Amperometry
The use of a transimpedance amplifier (TIA) or a current-conveyer (CC) are the most
common approaches for redox current acquisition [26], as shown in Fig. 1.15.
In the TIA approach, the transimpedance amplifier sets a virtual potential at the
working electrode and at the same time generates an output voltage that is propor-
tional to the redox current. This can be implemented as a resistive feedback configu-
ration [89] or a switched-capacitor circuit [90]. The size of the resistor in the resistive
feedback configuration makes it impractical for massively-parallel array sensing imple-
mentations. Additionally in a resistive feedback configuration the thermal noise can be
injected back into the biosensor. The switched-capacitor TIA configuration is a com-
mon choice for arrayed implementations [99]. A circuit diagram of a conventional
transimpedance amplifier with a captive feedback is shown in Fig. 1.16.
A current conveyer (CC), as shown in Fig. 1.17, is another common compact circuit
for measuring the redox current [91]. The WE is held at a virtual potential. Instead
CHAPTER 1. INTRODUCTION 35
−
+
IIN
VWE
VOUTCWE
RS
RCT
RE
CDB
VRE
WE
Figure 1.16: Conventional transimpedance amplifier (TIA) interfaced to a biosensor model.
IIN
VWEIOUT
−
+
CWE
RS
RCT
WERECDB
VRE
Figure 1.17: Conventional current conveyer (CC) interfaced to a biosensor model.
of converting the redox current directly to voltage, it is conveyed from WE to a high-
impedance output node and can then be converted to a voltage. A number of current
conveyer designs for amperometric sensing applications have been reported [91–95]. In
general, these designs do not support bidirectional current recording or offer pseudo-
bidirectional current acquisition capability by means of adding a large bias current to
the input so the current becomes unidirectional [69].
1.8.2 Wireless Data Transmission of DNA Results
Wireless communications capabilities is necessary in applications such as at-home health
care, food safety and water quality monitoring where in-field DNA sensing and analysis
are required.
One such example is placement of the DNA analysis microsystem in a water treat-
ment plant to quickly and accurately monitors the water quality and adjust the treatment
methods to disinfect water for municipal, industrial and residential customers. For ex-
ample in Trojan UV 3000PTP waste water solution [31], shown in Fig. 1.18, the waste
They are 100.10nA and 17.8pA for input current of 100nA. Table III summarizes the
experimentally measured characteristics of the TIA.
2.2 Current Conveyer (CC)
Another common method to acquire a bidirectional current is to use a unidirectional
current conveyer and to add a DC offset current to its input as shown in Fig. 2.7 [7-
12]. This requires high resolution in the subsequent ADC and adds noise to the redox
current, given that the current mirror generating the DC currents directly contributes
thermal and flicker noise to the output noise of the current conveyer. Also, depending
on the WE impedance, a portion of the DC offset current, IERROR, can leak into the
electrochemical cell and disturb the charge balance on the WE-electrolyte interface.
CHAPTER 2. CURRENT ACQUISITION CIRCUITS FOR ELECTROCHEMICAL AMPEROMETRICBIOSENSORS 57
-4
-2
0
2
4
6
10p 100p 1n 100n 1µINPUT CURRENT (A)
10nRE
LA
TIV
E E
RR
OR
(%
)
(a)
-1µ -100n -10n -1n -100p -10pINPUT CURRENT (nA)
-2
0
2
4
6
-4RE
LA
TIV
E E
RR
OR
(%
)
(b)
Figure 2.5: Experimentally measured relative error of the output of the TIA for the input currentof (a) 10pA to 350nA, and (b) -350nA to -10pA. The results are measured from one typicalchannel on one chip.
2.2.1 Circuit Implementation
The top-level VLSI architecture of the presented bidirectional current conveyer is shown
in Fig. 2.8. It is comprised of a PMOS and an NMOS transistors Mn and Mp connected
in the feedback of the OTA. The negative feedback ensures a known potential at the
working electrode is set by the voltage at the non-inverting input of the OTA. It also
enables the current conveyer to source and sink input current without the need for a
DC offset current. The currents through Mn and Mp are mirrored to the output of the
current conveyer.
Internal OTA chopping has been utilized to reduce the effect of its flicker noise. The
current mirrors are implemented in a low-current regulated cascode topology to enable
accurate current copying down to the pA level. To facilitate a comparative analysis with
CHAPTER 2. CURRENT ACQUISITION CIRCUITS FOR ELECTROCHEMICAL AMPEROMETRICBIOSENSORS 58
CURRENT (pA)0 20 40 60 80 100 120 140
0
1
2
3
4
5
6MEAN= 80.89pA
SD(3σ) = 6.9pA
N= 16
NU
MB
ER
OF
OC
CU
RA
NC
ES
(a)
CURRENT (nA)100.00 100.05 100.10 100.15 100.20 100.25100.30012
3
45
67
MEAN= 100.10nA
SD(3σ) = 17.8pA
N= 16
NU
MB
ER
OF
OC
CU
RA
NC
ES
(b)
Figure 2.6: Experimentally measured TIA output current of 16 channels (from 16 chips, onechannel each) for the input current of (a) 100pA, and (b) 100nA.
the TIA presented in section IV, the output current is integrated on the load capacitor
CF of 10pF and sampled.
The schematic diagram of the current conveyer is shown in Fig. 2.9. The OTA is
implemented as a folded-cascode amplifier and the current mirrors are implemented as
a low-current regulated cascode current mirror comprised of the transistors M12 to M20
and M21 to M29 [11]. The regulated cascode current mirrors ensure high precision and
The current conveyer was fabricated in a 0.13µm CMOS process with a 1.2V supply
and occupies an area of 100µm×100µm.
The experimentally measured relative errors (absolute error divided by the magni-
tude of the exact value) of the digital output for the input current swept between ±10pA
and ±350nA are shown in Fig. 2.14. The relative error stays below 8 percent over the
whole operating range. The current conveyer achieves a dynamic range of 8.6pA to
350nA. The lower limit is defined by the ADC LSB and the higher limit is defined by
the input current that saturates the current conveyer. Fig. 2.15 shows the experimen-
tally recorded output current distribution for the input currents of 100pA and 100nA
measured from 16 channels on 16 chips (one channel per chip). The mean output cur-
rent and the corresponding standard deviation are 81.19pA and 20.31pA, respectively,
CHAPTER 2. CURRENT ACQUISITION CIRCUITS FOR ELECTROCHEMICAL AMPEROMETRICBIOSENSORS 67
CHOPPER ON
CHOPPER OFF
M 3,4
M 1,2
M 12
M 15
M 21
M 24
0
0.005
0.010
0.015
0.020
0.025
INT
EG
RA
TE
D I
NP
UT
NO
ISE
( p
A )
0.035
(a)
CHOPPER ON
CHOPPER OFF
M 3,4
M 1,2
M 21
M 12
0
0.005
0.010
0.015
0.020
0.025
0.030
INT
EG
RA
TE
D I
NP
UT
NO
ISE
( p
A )
(b)
Figure 2.13: Current conveyer noise summary: (a) flicker noise contributions, and (b) thermalnoise contributions.
for the input current of 100pA. They are 100.21nA and 29.0pA for the input current of
100nA. Table III summarizes the experimentally measured characteristics of the current
conveyer.
2.3 Comparative Analysis
The electrical characteristics of TIA and CC are compared first. When the chopper
is enabled, the TIA achieves and input-referred noise of 0.07pArms and CC achieves
an 0.13pArms. This is due to the fact that the TIA integrates noise over one sampling
CHAPTER 2. CURRENT ACQUISITION CIRCUITS FOR ELECTROCHEMICAL AMPEROMETRICBIOSENSORS 68
-8
-4
0
4
6
8
10p 100p 1n 100n 1µINPUT CURRENT (A)
10n
RE
LA
TIV
E E
RR
OR
(%
)
(a)
-1µ -100n -10n -1n -100p -10pINPUT CURRENT (nA)
0
2
4
6
8
−2RE
LA
TIV
E E
RR
OR
(%
)
(b)
Figure 2.14: Experimentally measured relative error of the output of the current conveyer forthe input current of (a) 10pA to 350nA, and (b) -350nA to -10pA. The results are measured fromone typical channel on one chip.
period.
The chip-to-chip output current variation of TIA, in Fig. 2.6, is lower compared
to that of the CC design shown in Fig. 2.12. The variation in CC is mainly due to
the mismatch in the output current mirrors. The maximum relative error of the output
over the operating current range is 5 percent for the TIA and 8 percent for the CC.
Mismatch in the regulated cascode current sources limits the linearity of the current
conveyer. The mismatch in the regulated cascode current mirrors is due to the mismatch
in the transistor pairs M12-M15 and M21-M24, as shown in Fig. 2.9. Dynamic element
matching (DEM) can be employed to reduce the effect of the mismatch in the current
mirrors [166].
To study the effect of charge injection into the working electrode, two sets of simu-
CHAPTER 2. CURRENT ACQUISITION CIRCUITS FOR ELECTROCHEMICAL AMPEROMETRICBIOSENSORS 69
CURRENT (pA)
NU
MB
ER
OF
OC
CU
RA
NC
ES
0 20 40 60 80 100 120 14001
2
3
4
5
6MEAN= 81.19pA
SD(3σ) = 20.31pA
N= 16
(a)
CURRENT (nA)100.15 100.2 100.25 100.3 100.35 100.401
234
5
67
MEAN= 100.21nA
SD(3σ) = 29pA
N=16
NU
MB
ER
OF
OC
CU
RA
NC
ES
100.5
(b)
Figure 2.15: Experimentally measured current conveyer output current of 16 channels (from 16chips, one channel each) for the input current of (a) 100pA and (b) 100nA.
lations were performed with the working electrode model in both Fig . 2.1 and Fig. 2.8
connected to the voltage VRE . The average current integration into the working elec-
trode, due to current integration and sampling, is calculated for the cases where the
sampling frequency is varied from 1kHz to 12kHz and the chopping frequency is set to
20kHz. As it can be seen from Figs. 2.16(a) and (b), the average current injected into
the working electrode at 6kHz for the case of the TIA is significantly higher compared
to the average current injected by the CC.
To compare the performance of the TIA and CC in electrochemical sensing appli-
cations, two sets of CV scans of a DNA reporter, potassium ferricyanide, have been
performed. Potassium ferricyanide K3[Fe(CN)6] is commonly used in electrochemi-
cal sensing systems as a redox reporter [87]. Cyclic voltammetry recordings of 2µM
potassium ferricyanide in a 1M potassium phosphate buffer (pH 7.3) have been carried
out. A 100mV/sec 0.7V peak-to-peak CV waveform with 50ms resting period was ap-
CHAPTER 2. CURRENT ACQUISITION CIRCUITS FOR ELECTROCHEMICAL AMPEROMETRICBIOSENSORS 70
0 2 4 6 8 10 12
INP
UT
CU
RR
EN
T (
fA)
130
180
230
280
330
SAMPLING FREQUENCY (kHz)
(a)
0 2 4 6 8 10 12
INP
UT
CU
RR
EN
T (
fA)
20
40
60
80
100
SAMPLING FREQUENCY (kHz)
(b)
Figure 2.16: Average current injected into the working electrode for (a) TIA and (b) CC.
−700 −600 −500 −400 −300 −200 −100 0−12
−10
−8
−6
−4
−2
0
2
4
6
8
V -V (mV)
CU
RR
EN
T (
nA
)
OXIDATION
VOLTAGE
RE WE
REDUCTION
VOLTAGE
TIA
CC
Figure 2.17: Cyclic voltammogram of 2µM potassium ferricyanide in 1M potassium phosphatebuffer solution experimentally recorded with the transimpedance amplifier (TIA) and the currentconveyer (CC) using a 50µm×50µm gold electrode.
CHAPTER 2. CURRENT ACQUISITION CIRCUITS FOR ELECTROCHEMICAL AMPEROMETRICBIOSENSORS 71
Table 2.3: Experimentally Measured Transimpedance Amplifier (TIA) and Current Conveyer(CC) Characteristics
TIA CCTechnology 0.13µm CMOS 0.13µm CMOSSupply Voltage 1.2V 1.2VArea 80µm×60µm 100µm×100µmSensitivity 135fA 8.6pAMax Relative Error 5.1 8.0(10pA to 350nA)Max Relative Error 4.8 7.8(10pA to 350nA)Input Referred Noise 0.07pA 0.13pA(0.01Hz to 1kHz)Charge Injection at 6kHz 221fA 53fA(Simulated)Power Consumption
plied between a 55µm×55µm on-chip gold working electrode and an off-chip Ag-AgCl
reference electrode. The resulting CV curves recorded by the chopper-stabilized TIA
and the chopper-stabilized CC are shown in Fig. 2.17. The CV curves for both TIA
and CC show two distinct peaks at the reduction and oxidation voltages of potassium
ferricyanide at -250mV and -450mV respectively. The measurements match as well.
Next the same set of CV recordings have been conducted using a 2µm×2µm on-chip
gold working electrode. The resulting CV curves recorded by the chopper-stabilized
TIA and the chopper-stabilized CC are shown in Fig. 2.18. The CV curve for the CC
shows two distinct peaks at the reduction and oxidation voltages of potassium ferri-
cyanide, as expected. The CV curve for the TIA shows no reduction or oxidation peaks.
In the TIA case the switching charge from switch S is injected into the working elec-
trode. This disturbs the charge balance at the electrode-electrolyte interface thus affect-
ing the electrochemical reaction required for reduction and oxidation of the potassium
ferricyanide. As expected, this effect is more pronounced for smaller electrode size,
given that the amount of the charged injected into the working electrode is comparable
to the double layer capacitance charge.
CHAPTER 2. CURRENT ACQUISITION CIRCUITS FOR ELECTROCHEMICAL AMPEROMETRICBIOSENSORS 72
−700 −600 −500 −400 −300 −200 −100 0
−0.6
−0.4
−0.2
0.0
0.4
0.6
0.8
1.0
V -V (mV)
CU
RR
EN
T (
nA
)
OXIDATION
VOLTAGE
RE WE
REDUCTION
VOLTAGE
TIA
CC
0.2
−0.8
Figure 2.18: Cyclic voltammogram of 2µM potassium ferricyanide in 1M potassium phosphatebuffer solution experimentally recorded with the transimpedance amplifier (TIA) and the currentconveyer (CC) using a 2µm×2µm gold electrode.
Both circuits inject small amount of charge generated by the chopper switches into
the biosensor. This injected noise is negligibly small compared to the switching noise
due to the feedback switch in the TIA and is due to the chopper switches mismatch.
2.4 Chapter Summary
Designs of two low-noise chopper-stabilized bidirectional current acquisition circuits
for electrochemical sensing applications have been presented. The first design has a
switched-capacitor transimpedance amplifier topology. The second one has a current
conveyer topology. Both designs are implemented in a 0.13µm CMOS technology.
Electrical and electrochemical performance of both design has been characterized. The
TIA and CC consume 3µW and 4µW from a 1.2V supply, respectively. It is shown
that the TIA marginally outperforms the CC for high-amplitude input currents. For
small input currents corresponding to low concentration of biochemicals the CC is the
preferred choice as it better isolates the working electrode from current injection.
73
Chapter 3
Cyclic Voltammetry and pH sensing
A fully integrated 54-channel wireless fast-scan cyclic voltammetry DNA analysis SoC
is presented. The microsystem includes 546 3D nanostructured and 54 2D gold DNA
sensing microelectrodes as well as 54 pH sensors. Each channel consists of a chopper-
stabilized current conveyer with dynamic element matching. It is utilized as the am-
perometric readout circuit with a linear resolution from 8.6pA to 350nA. The on-chip
programmable waveform generator provides a wide range of user-controlled rate and
amplitude parameters with a maximum scan range of 1.2V, and scan rate ranging be-
tween 0.1mV/sec to 300V/sec. A digital ultra-wideband transmitter based on a de-
lay line architecture provides wireless data communication with data rates of up to 50
Mb/sec while consuming 400µW. The 3mm×3mm prototype fabricated in a 0.13µm
standard CMOS technology has been validated in prostate cancer synthetic DNA detec-
tion with 10aM label-free PCR-free detection limit. Each channel occupies an area of
only 0.06mm2 and consumes 42µW of power from a 1.2V supply.
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 74
3.1 Introduction
We have reported in Nature Nanotechnology amperometric electrochemical sensors
fabricated on passive silicon, not on CMOS, that do not require cumbersome tagging of
DNA with chemical or optical labels [138]. These gold microelectrodes have fine-tuned
nanostructured patterns on their surface that yield an over 140dB input dynamic range
and 10aM detection limit sufficient for PCR-free DNA detection.
In this paper, we present a 0.13µm CMOS DNA analysis SoC with 600 such nanos-
tructured microelectrodes (NMA) grown directly on the die. This paper extends on
an earlier report of the principle and demonstration in [87], and offers a more detailed
analysis of the design and additional experimental results characterizing the circuit im-
plementation and the DNA detection performance. This SoC performs label-free PCR-
free DNA analysis using fast-scan cyclic voltammetry with a 10aM detection limit and
pH sensing for cancer detection. The microsystem consists of a fully programmable
arbitrary waveform generator with an on-chip memory and 54 chopper-stabilized cur-
rent recording channels. The chopper-stabilized current conveyer frontend, with an
input-referred noise of 0.13pArms over one kHz bandwidth, is utilized as the ampero-
metric readout circuit in each channel. The current conveyer achieves linear resolution
from 10pA to 400nA. A chopper-stabilized dual-slope ADC is utilized to digitize the
recorded current. The waveform generator provides stimulation waveforms with a max-
imum scan range of 1.1V and a scan rate ranging from 0.1mV/sec to 300V/sec. A fully
digital 10Mb/s ultra-wideband (UWB) transmitter performs wireless communication.
The rest of this paper is organized as follows. Section II provides background on
DNA detection principles. Section III describes the process of fabrication of the nanos-
tructured DNA sensing microelectrodes. Section IV presents the DNA analysis SoC
VLSI architecture. Section V details the circuit implementation of the VLSI archi-
tecture. Section VI demonstrates the electrical experimental results obtained from the
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 75
PR
OB
E D
NA
TAR
GE
T D
NA
K [Fe(CN) ] 6
K [Fe(CN) ] 6
e e
e
e e
e
e
e
e
e
e e
e
e
e
e
e
e
e
e
e
BA
RE
GO
LD
I RE
DO
X1
I RE
DO
X1
I RE
DO
X2
<
4 3
(a) (b)
Au WE
e e
Au WE
e
Au WE
(c)
VRE
I RE
DO
X1
-VWE
VRE
I RE
DO
X2
-VWE V
RE
I RE
DO
X3
-VWE
Ag/AgCl RE Ag/AgCl RE Ag/AgCl RE
I RE
DO
X2
I RE
DO
X3
<
REDUCTION OXIDATION
Figure 3.1: Label-free electrochemical DNA detection principle. (a) Bare electrode: maximumcharge transfer between working and reference electrode in the absence of negatively chargedprobe and target DNA; (b) non-complementary target DNA: reduction in the charge transfer ratedue to the presence of negatively charged probe DNA, and (c) complementary DNA: furtherreduction in the charge transfer rate due to the presence of negatively charged target and probeDNA.
0.13µm CMOS prototype. In Section VII, the results of on-chip electrochemical record-
ing of calibration chemicals are presented. In Section VIII, the results of on-chip CV
recording of a synthetic DNA marker in prostate cancer screening are presented.
3.2 DNA Detection Principle
The principle of the label-free DNA detection method based on potassium ferricyanide
reporter is shown in Fig. 3.1. Potassium ferricyanide K4[Fe(CN)6] is a negatively
charged redox complex with a well-defined electrochemical signature exhibiting oxi-
dation and reduction currents at VRE-VWE voltage of -450mV and -250mV, respec-
tively. Maximum electron transfer between the bare gold electrode and potassium fer-
ricyanide is achieved in the absence of both the DNA target and probe, as denoted by
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 76
IREDOX1, in Fig. 3.1(a). Electron transfer is decreased when a negatively charged self-
assembled monolayer of probe DNA (SSDNA) is deposited on the electrode, as shown
in Fig. 3.1(b). This corresponds to smaller redox current IREDOX2, which results in
relatively smaller reduction or oxidization peaks. Upon bonding of the probe DNA and
target DNA (if present) the resulting DSDNA is more negatively charged and causes
potassium ferricyanide to be repelled farther from the electrode surface reducing the
generated faradaic current, as shown in Fig. 3.1(c). The redox current IREDOX3 is sig-
nificantly smaller compared to the first two cases and lacks the reduction and oxidation
peaks. In other words, the presence of negatively charged DNA on the biosensor sur-
face translates to a decrease in the potassium ferricyanide oxidation/reduction current
creating a detectible signal change [140–142].
3.3 Integrated Sensors
3.3.1 DNA Sensing Microelectrodes
To improve the sensitivity and dynamic range of the DNA sensor, nanostructured micro-
electrodes (NMEs) [138] are grown on the CMOS aluminum working electrode base,
using a combination of electroless plating and electroplating techniques.
It is shown in [138] that nanostructuring the working electrode allows for fabri-
cation of DNA sensors on passive silicon that have a broad range of sensitivities and
dynamic ranges. Highly branched electrodes with fine nanostructuring are capable of
achieving a 10aM detection limit [138]. It is postulated that the DNA probes which are
functionalized on nanostructured electrodes are more accessible and, as a result, bond
much easier and faster with target molecules. Microelectrodes with different degrees
of nanostructuring result in different sensitivities and dynamic ranges. By placing an
array of different electrodes on the same CMOS chip the sensor system can achieve a
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 77
sensitivity of two to six orders of magnitude [138].
In this design DNA sensing working electrodes are created by forming 2µm×2µm
passivation openings on the top metal layer (aluminum) of the CMOS chip (as it is
commonly done for bond pads) as shown in Figs. 3.2(a) and (b). An electroless metal
plating technique is employed to sequentially deposit nickel (Ni), palladium (Pd) and
then flat gold (Au) base on the exposed Al surface to form an electrode foundation as
shown in Fig. 3.3(c). Next NMEs are grown electrostatically in a solution containing
69µL of gold solution (544385-10G Aldrich) diluted in 2.5mL of deionized (DI) water
2.5mL of 5µM HCl [138] as also shown in Fig. 3.3(c). The shape and the size (defin-
ing the sensitivity and dynamic range) of the NMEs depend on the potential difference
between the working electrode and the reference electrode and the duration of the elec-
troplating. Examples of NMEs grown on a CMOS chip for 60sec at 100mV, 0 and
-100mV voltage difference between an on-chip Au working electrode and an off-chip
(Ag/AgCl) reference electrode are shown in Figs. 3.3(d), (e) and (f), respectively. Two
examples of arrays of NMEs grown on a CMOS chip are also shown in Fig. 3.3(f),
middle and right. For comparison purposes, large flat (2D) working electrodes have
also been fabricated on-CMOS. These flat gold electrodes are fabricated using the same
electroless plating technique as that used for the NME foundation fabrication, as shown
in Fig. 3.2(a). For example, the SEM photographs of such a gold-plated 55µm×55µm
on-CMOS flat working electrode are shown in Figs. 3.2(b) and (c).
3.3.2 pH Sensors
The in-channel ion-sensitive-field-effect-transistor (ISFET) based pH sensor is imple-
mented by a floating gate PMOS with the size of 0.5µM×0.35µM. The poly-gate
of the PMOS is connected to the top metal layer to form a floating gate electrode,
and the CMOS passivation layer (Si3N4 and SiO2) is used as the pH-sensitive mem-
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 78
(a)
(b) (c)
(a)
(c)
FLAT Au WE
25µm 1µm
FLAT Au WE
1.0μm0.2μmPd
NiAl
Au
SiO2
Si N43
POLYAMIDE
Figure 3.2: (a) Passivation opening in standard CMOS and added metal layers of a flat (2D)microelectrode after electroless nickel-palladium-gold plating, (b) and (c) SEM photographs ofsuch 55µm×55µm working electrodes.
brane [101]. It is shown in [143] that the passivation layer (exposed section where
there is no polyamide) gives a linear pH response with a sensitivity of approximately
56mV/pH [101], depending on the stoichiometry of the passivation layer. The 54 pH
sensors are directly interfaced to the 54 current-recording channels. The source of the
PMOS is connected to the VDD (1.2V), and the drain is connected to the input of a
current conveyer. The pH sensor gate voltage is set by the on-chip reference electrode.
In this configuration, both the Vgs and the Vds of the pH sensor PMOS transistor are
fixed. Any change in the pH level effectively changes the PMOS threshold voltage.
This change results in a corresponding change in the drain current, which is digitized
by the recording channel.
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 79
5µm
5µm
30µm
VRE
-V = -100 mVWE
RE-V = 0 mV
WE
VRE
-V = -100 mVWE
555555µµµµµmmmmm3µm
6µm1µm
Al BASE
VRE
-V = 100 mVWE
VR
mV RE
-V
= 100 mV
WE
VRE
-V = 100 mVWE
(b)
V
(d)
(e)
(f)
VRE -V = -100 m
V
WE
VRE
-V = 0mVWE
5µm7µm
VRE
-V = 100 mVWE
GROWN @
GROWN @
GROWN @
1µm
Au BASE
PdNiAl
1.0μm0.2μmSiO2
Si N43
POLYAMIDE
~5.0μmAu WE
(c)
Al
SiO2
Si N43
POLYAMIDE
~2.0μm
(a)
VRE
-V = -100 mVWE
40µm
(d)
(e)
VRE
-V = 0 mVWE
5µm
RE-V = 0 mV
WEV
5µm
Figure 3.3: Nanostructured DNA sensing working electrodes (NMEs): (a) Cross-sectional viewof a 2µm×2µm passivation opening in standard CMOS, (b) SEM photograph of a 2µm×2µmworking electrode passivation opening over an aluminum base, (c) nanostructured 2µm×2µmworking electrode grown on the passivation opening over an aluminum base in standard CMOS,(c) (d) and (e) SEM photographs of nanostructured microelectrodes grown at different elec-trodeposition conditions on the passivation opening in (b).
3.4 VLSI Architecture
3.4.1 Top-Level VLSI Architecture
The top-level VLSI architecture of the wireless DNA analysis SoC is shown in Fig. 3.4.
The SoC consists of 54 current-to-digital recording channels. Each channel is multi-
plexed between a bank of DNA sensors and a pH sensor.
The sensors are interrogated by the on-chip arbitrary waveform generator that is
shared among all channels. The arbitrary waveform generator consists of a 8-bit R-2R
DAC, an 8-bit up-down counter and a 3-electrode-configuration RE voltage regulation
circuit [144]. The waveform generator provides stimulation waveforms with a maxi-
mum scan range of 1.1V and the scan rate ranging from 0.1mV/sec to 300V/sec. It
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 80
DNA
PH
×54
SRAM
WE
WE IREF
IIN
WA
VE
FO
RM
GE
NE
RA
TO
R
CE
RE
CLK 1.2 VIBIAS VBIAS
CCII
CLOCK
BIAS
GLOBAL
CLOCK
GLOBAL
BIAS
U
WB
TX
IREF
CHANNEL
DUAL-
SLOPE
ADC
TIMING
COEFFICIENTS
SRAM
WAVEFORM
GENERATOR SRAM
Figure 3.4: Wireless DNA analysis microsystem functional block diagram.
consumes 900µA from a 1.2V supply when driving a 5nF load at the maximum scan
rate of 300V/sec. This maximum rate is not required for the DNA sensing applica-
tion as the scan rate is limited to low 100s of mV/sec. Other amperometric biochemical
sensing applications (such as, for example, neurotransmitter sensing [70]) require much
higher scan rates of up to 300V/sec. The microsystem presented here is designed so that
it can also be used in applications other than DNA sensing. As a result, the waveform
generator is designed such that it meets requirements for a general purpose biochemical
sensing microsystem but with the power scaling with the frequency. The digital data
representing the stimulation waveform properties are stored in the on-chip waveform
generator SRAM (Fig. 3.4).
A current conveyer is placed at the frontend of each channel to acquire the result-
ing sensory current at a low impedance. A dual-slope ADC quantizes the input redox
current and outputs a corresponding digital word. The digital output of each channel is
serialized on the chip and is wirelessly transmitted at a data rate of up to 10Mbps, using
an all-digital ultra-wide band transmitter. To enable independent channel programma-
bility, each channel also includes a bias voltage generation circuit, a clock generation
circuit, and an in-channel SRAM for setting the channel dynamic range and sensitivity.
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 81
−
+
S1
S2
S2
VWE
CONTROL
LOGIC
IREF+
IREF-
CINT
VOUT2
(S1,S2)
−
+
VWE
CHOPPER
STABILIZATION
VLATCH
S2
S1
DIGITAL
OUTPUT
−
+
CURRENT CONVEYER DUAL-SLOPE ADC
WEs
VWE
Mn
Mp
Cc
DYNAMIC ELEMENT
MATCHING
−
+ IOUT
IIN
9-BIT
COUNTER
CHOPPER CLK
GENERATOR
−
+
Figure 3.5: Simplified top-level VLSI architecture of one chopper-stabilized integrated current-to-digital channel.
VWE
VbiasP
VbiasN
VcascP
cascNVIIN
IOUTCc
M1 M2
M3 M4
M5 M6
M7 M8
M9 M10M11
M12M13 M14 M15
M16M17
M18
M19 M20
M21 M22 M23 M24
M25
M26
M27
M28 M29
M
Mn
p
I-to-V
I-to-V
Figure 3.6: Detailed implementation of the current conveyer OTA with internal chopping anddynamically-matched low-current regulation.
3.4.2 Channel VLSI Architecture
The top-level VLSI architecture of one current-to-digital channel of the integrated elec-
trochemical sensory microsystem is shown in Fig. 3.5. Each channel consists of a
chopper-stabilized bidirectional current conveyer (Fig. 3.5, left) and a 9-bit dual-slope
ADC (Fig. 3.5, right).
The current conveyer buffers the input current and maintains the working electrode
at a fixed potential, VWE , as needed to induce a redox reaction. DNA analysis applica-
tions require both sourcing and sinking the redox current. A number of current conveyer
designs for electrochemical sensing applications have been reported [91–94]. In gen-
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 82
eral, existing designs do not support bidirectional current recording and suffer from the
amplifier flicker noise and the mismatch within current mirrors. Fig. 3.5(left) depicts a
low-noise and accurate current conveyer VLSI architecture that overcomes these lim-
itations. Internal OTA chopper stabilization is utilized to reduce the effect of flicker
noise. The current conveyer utilizes low-current regulated-cascode current mirrors to
record small (i.e., as small as 10pA) bidirectional currents. Dynamic element matching
is utilized to improve the accuracy by averaging the mismatch in the current mirrors.
The current conveyer is comprised of a PMOS and an NMOS transistors Mn and
Mp connected in the feedback of the chopper-stabilized OTA. The negative feedback
ensures a known potential, VWE , at the working electrode is set by the voltage at the
negative terminal of the OTA. It also enables the current conveyer to source and sink
input current without the need for a DC offset current [92], which can disturb the DNA
charge balance. The currents through Mn and Mp are mirrored by dynamically-matched
current mirrors to the output of the current conveyer and are added.
Based on previously published results of DNA hybridization experiments on NME
working electrodes [138], it is determined that the on-chip ADCs must be able to dig-
itize bidirectional current in the 10pA to 100nA range or greater, and to cover a fre-
quency range of 0.01Hz to at least 1kHz. The dual-slope ADC architecture is selected
for this purpose because its dynamic range, sampling frequency, and nominal resolution
suit these requirements and can all be easily adjusted.
The dual-slope ADC shown in Fig. 3.5(right) consists of an integrating on-chip
variable capacitor CINT (adjustable from 1pF to 10pF, all the measurements here are
done using a 2pF capacitor value), regulated-cascode current sources IREF+ and IREF−,
a four-stage track-and-latch comparator, a 9-bit digital counter and control logic. All
switches are implemented as low-leakage switches as shown in an inset in Fig. 3.5.
The reference current sources are implemented as regulated-cascode current mirrors
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 83
to ensure accurate current sourcing over the operating dynamic range. The IREF+ is
implemented with PMOS devices and the IREF− is implemented with NMOS devices.
This can result in some mismatch between the IREF+ and IREF−. The effect of the
mismatch between the positive and negative current source does affect the linearity of
the ADC. These effects are within the specification and are reflected in the measured
spectrum of the ADC output and its ENOB presented in Section VI. The first stage of
the comparator is chopper-stabilized to reduce the effect of its offset and low-frequency
noise.
The dual-slope ADC operates in two phases. In phase one, the integrating capac-
itor CINT is charged by the input current IIN for a predetermined period of time T1.
Next, during the second phase of the operation, the capacitor is discharged to zero
by a DC reference current IREF (IREF+ or IREF−). By counting the duration of the
second phase, the time T2, a digital representation of IIN can thus be obtained as -
sign(IREF )×(T2/T1)×|IREF |. In this design the value of the IREF is programmable
(using an off-chip variable resistor) between 100pA to 50nA.
The in-channel SRAM can also be used to adjust the duration of the charging and
discharging cycles of the dual-slope ADC for the purpose of channel gain calibration.
For example if IREF is higher in the first channel compared to the second channel, then
the duration of the charging time T2 can be reduced for the first channel to compensate
for larger IREF and thus generating the same output digital code for both channels for a
given input current. This effectively calibrates each channel independently and reduces
to reduce the effect of the mismatch in the current mirrors. The main source of mis-
match in the regulated cascode current mirrors is due to the mismatch in the transistor
pairs M12, M15 and M21, M24. To reduce the effect of the mismatch between these tran-
sistors, the DEM technique is applied by means of the chopper switches at the drains
of the current source transistors, so that the critical transistor pairs are dynamically
matched. In this method, the locations of the transistors M12, M15 and M21, M24 are
swapped periodically, at 500Hz, effectively averaging the current mirrors mismatch.
Ideally the error due to the mismatch in the current mirrors is reduced with a higher
DEM switching frequency which results in better averaging over one ADC conversion
cycle. Due to the non-ideality of the switches, an increase in the switching frequency
results in high-frequency switching noise and an increase in the charge injected into the
current path. This in turn causes an error at the output of the current conveyer. Based
on these considerations the 500Hz DEM frequency was chosen.
To achieve efficient flicker noise reduction, the chopper frequency needs to be higher
(at least twice) than the input signal maximum frequency (1kHz). The chopper clock
frequency was set to 10kHz to place the switching noise well outside the operating
frequency range. As a result the current conveyer bandwidth should be higher compared
to the case where no chopper stabilization is utilized so that the output settles in each
switching period. The current conveyer 3dB bandwidth is 35.7kHz.
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 86
V6
V5
V4
V3
V2
V1
OUTV
D
C
C
C
CBP
M
ON-OFF KEYING (OOK)
MANCHESTER MODULATION
D IN
OUTVINV
PV
NV
(a) (b)
Figure 3.8: (a) Ultra-wideband transmitter circuit schematic diagram, and (b) schematic of onecurrent-starved inverter.
V6
V5
V4
V3
V2
V1
OUTV
DM
Figure 3.9: Timing diagram of the ultra-wideband transmitter.
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 87
1Х6 BANK OF CROSS-VALIDATION CHANNELS
UWB TX
WAVEFORM
GENERATOR & SRAMCONTROL
LOGIC
9×6 ARRAY OF
CHANNELSRECE
WE
2X2μm2
55X55μm
ISFET
5X5μm2X2μm
2
2
2ONE CHANNEL
8Х8 ARRAY OF
2Х2μm WE2
ARRAY OF
WEs
Figure 3.10: Die micrograph of the 3mm×3mm 54-channel wireless DNA analysis SoC. TheSoC was fabricated in a 0.13µm standard CMOS technology.
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 88
3.5.2 Dual-slope ADC Comparator
The ADC comparator is implemented with three stages of pre-amplifiers, with a total
gain of 60dB and the last stage with a high-speed latch as shown in Fig. 3.7. The first
stage of the comparator is implemented as a cross-coupled diode-connected gain stage.
This topology provides a moderate gain and a high frequency bandwidth. Chopper-
stabilization suppresses the input offset and ensures 9-bit accuracy. The second and
third stages are identical to the first one but with no chopping. The high-speed latch is
implemented with an NMOS input pair gain stage and a NMOS-PMOS cross-coupled
load. This topology provides high accuracy, low offset and a high frequency bandwidth.
The comparator transistor sizes are listed in Table 4.1.
3.5.3 Ultra-wideband Transmitter
The circuit diagram of the all-digital pulsed UWB transmitter is shown in Fig. 3.8(a).
The input data are modulated using on-off keying (OOK) Manchester modulation.
UWB pulses are generated on the rising edge of the modulated data (DM ). A delay line
bank is employed together with a capacitively coupled output combiner [150] as shown
in Fig. 3.8(a). The modulated data are passed through a delay line, and a delayed ver-
sion of the data are passed through three pulse generators. The pulse generators shape
a first-order Gaussian pulse at the rising edge of the input data. The presented digital
UWB transmitter achieves both power efficiency and spectral compliance in a much
smaller chip area compared to earlier designs [151, 152].
As illustrated in Fig. 3.9, each pulse generator forms pulses that are delayed, and
have opposite signs. By capacitively combining the three paths, the opposite signs
are canceled, and the zero-DC double-differentiated Gaussian pulse propagates to the
single-ended antenna [151, 152]. The width of the output pulse depends on the delays
in the delay line. The delay cells in all the paths are implemented as current-starved
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 89
0 5 10 15 20 25 30 35 40 45 50−140
−120
−100
−80
−60
−40
−20
0
2nd HARMONIC
3rd HARMONIC
SFDR = 59 dB
NORMALIZED POWER (dB)
SNR = 57 dB
FREQUENCY (Hz)
I = 350nAin
ELECTROCHEMICAL RECORDING
CHANNEL OUTPUT SPECTRUM
Figure 3.11: Experimentally measured spectrum of the electrochemical recording channel out-put for a 15Hz sinusoidal full-scale (350nA) input.
inverters, shown in Fig. 3.8(b), to allow for tuning of the UWB pulse width.
3.6 Electrical Experimental Results
The fabricated prototype die micrograph is depicted in Fig. 3.10. The 54 channels are
arranged in a 9×6 array on a 3mm×3mm 0.13µm CMOS die. Two channel types with
two different WE aluminum base configurations are implemented. A set of 48 channels
of the first type scan 4 WEs each, in order to perform initial detection of DNA. They
have three different WE aluminum base sizes of 2µm×2µm (twice), 5µm×5µm and
55µm×55µm each as needed to cover a wide combined dynamic range. An additional
set of 6 channels of the second type (at the bottom of the array in Fig. 3.10) additionally
scan a sub-array of 8×8 2µm×2µm WEs each. These redundant-electrode sub-arrays
are utilized for DNA detection results cross-validation and for titer DNA concentration
measurements.
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 90
100
101
102
103
104
SNDR=48dB LSB=8.6pA
f = 23 kHzs
SNDR=53dBLSB=132pA
SNDR=59dB LSB=1.05nA
f = 2 kHzs
10p 100p 1n 10n 100n 1μ
DIG
ITA
L O
UT
PU
T C
OD
E
f = 11kHzs
INPUT CURRENT (A)
CHANNEL
TRANSFER
CHARACTERISTIC
Figure 3.12: Experimentally measured transfer characteristics of the current-to-digital channelfor three sampling frequencies.
Dynamic performance of the entire channel was measured by applying a 15Hz full-
scale (350nA) sinusoidal input current sampled at 23kHz. Fig. 3.11 shows the 65536-
point FFT of the measured ADC output. The strong second harmonic is due to the
single-ended nature of the architecture of the ADC. The resulting effective number of
bits (ENOB) is 9.1.
For static performance characterization the input DC current of one typical channel
was swept between 10pA and 350nA as shown in Fig. 3.12. The input dynamic range
is 93dB cumulatively for the three sampling frequency settings, or 48dB at one fixed
sampling frequency of 2kHz. The dynamic range for each setting is computed by taking
the ratio of the maximum signal that saturates the ADC to the LSB for a given sampling
frequency setting.
Two sets of ENOB measurements were conducted to study the effectiveness of the
in-channel gain calibration using the in-channel SRAM to adjust the ADC timing. In
the first measurement no calibration has been performed and the timing parameters
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 91
ENOB (BITS)
NU
MB
ER
OF
OC
CU
RA
NC
ES
8.2 8.4 8.6 8.8 9.0 9.2 9.4 9.60
2
4
6
8
10
12
MEAN= 9.01 BITS
SD(3σ) = 0.307 BITS
N= 32
(a)
ENOB (BITS)
NU
MB
ER
OF
OC
CU
RA
NC
ES
8.2 8.4 8.6 8.8 9.0 9.2 9.4 9.60
2
4
6
8
10
12
MEAN= 9.15 BITS
SD(3σ) = 0.252 BITS
N= 32
(b)
Figure 3.13: Experimentally measured output ENOB of 32 channels (from 16 chips, two chan-nels each) for a 15Hz 350nA sinusoidal input (a) without calibration (b) with in-channel cali-bration.
of all channels are set to a constant value (all the ADCs have the same charging and
discharging phases duration). Fig. 3.13(a) shows the experimentally recorded ENOB
for a 15Hz full-scale (350nA) sinusoidal input current from 32 channels on 16 chips
(two channels per chip), with the ADC clocked at 12MHz. The mean ENOB and the
corresponding standard deviation are 9.01 and 0.307 respectively. Next, the same set
of experiments were repeated with the calibrated channels, as described at the end of
Section IV.B. Fig. 3.13(b) shows the experimentally recorded ENOB for the same input
tone as the pervious case. The mean ENOB and the corresponding standard deviation
are 9.15 and 0.252 respectively. The calibration improves the ENOB standard deviation
by 17 percent.
Fig. 3.14 shows the ADC ENOB versus the frequency for a full-scale (350nA) si-
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 92
101
102
103
104
5
6
7
8
9
10
EN
OB
(B
ITS
)
FREQUENCY (Hz)
I = 350nAin
Figure 3.14: Experimentally measured ENOB vs. input frequency for the in-channel ADC.
nusoidal input current. The ADC maintains an ENOB of greater than 8.5 bits at up to
3.4kHz. The drop in the ENOB is due to the limited bandwidth of the frontend current
conveyer and high-frequency switching interference noise.
The experimentally measured relative errors of the digital output for the input cur-
rent swept between ±10pA and ±350nA are shown in Fig. 3.15. The relative error stays
below 6 percent over the whole operating range. This is an improvement of 33 percent
compared to the design without the DEM [155]. Fig. 3.15 illustrates an improvement
in the output relative error of approximately 25 percent due to the use of DEM in this
design as compared to a previously reported design without DEM [155]. The current
conveyer achieves a dynamic range of 8.6pA to 350nA or 93dB. The lower limit is de-
fined by the ADC LSB and the higher limit is defined by the input current that saturates
the current conveyer.
Fig. 3.16 shows the experimentally recorded output current distribution for the input
current of 100pA measured from 32 channels on 16 chips (two channels per chip) with-
out dynamic element matching [155] and with dynamic element matching implemented
in this design. The mean output current and the corresponding standard deviation with-
out dynamic element matching [155] are 81.26pA and 20.2pA, respectively. In this de-
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 93
−4
-2
0
2
4
6
10p 100p 1n 100n
INPUT CURRENT (A)
10n
RE
LA
TIV
E E
RR
OR
(%
)
(a)
-100n -10n -1n -100p -10p
INPUT CURRENT (nA)
−2
0
2
4
6
−4
RE
LA
TIV
E E
RR
OR
(%
)
(b)
Figure 3.15: Experimentally measured relative error of the output digital code of the currentconveyer connected with the dual-slope ADC for (a) 10pA to 350nA and (b) -350nA to -10pAinput current.
sign, with dynamic element matching added, they are 92.12pA and 9.2pA, respectively.
Adding DEM results in a 54 percent improvement in channel-to-channel accuracy.
As shown in Fig. 3.17 the same experiment is repeated with the input current level
of 100nA. The mean output current and the corresponding standard deviation without
dynamic element matching [155] are 100.26nA and 34pA, respectively. In this design,
with dynamic element matching added to the design, they are 100.18nA and 22pA,
respectively. Adding DEM results in a 35 percent improvement in channel-to-channel
accuracy.
The input Manchester-encoded data to the UWB transmitter and its measured output
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 94
CURRENT (pA)
NU
MB
ER
OF
OC
CU
RA
NC
ES
0 20 40 60 80 100 120 1400
2
4
6
8
10
12
MEAN= 81.26pA
SD(3σ) = 20.2pA
N= 32
(a)
CURRENT (pA)
NU
MB
ER
OF
OC
CU
RA
NC
ES
0 20 40 60 80 100 120 140
0
2
4
6
8
10
12
MEAN= 92.12pA
SD(3σ) = 9.2pA
N= 32
(b)
Figure 3.16: Experimentally measured output current of 32 channels (from 16 chips, two chan-nels each) for the input current of (a) 100pA without DEM [32] and (b) 100pA with DEM (thiswork).
UWB pulses are shown in Fig. 3.18. The UWB pulses are measured using custom-built
UWB antennas (5cm spacing between the transmitter and receiver) and an custom-built
receiver. A zoomed-in version one such the measured UWB pulse overlayed on a simu-
lated UWB pulse is shown in Fig. 3.19. As it can be seen the measured pulse resembles
the expected UWB pulse but includes minor ringing due to the package bondwire in-
ductance. The measured output power spectrum of the UWB transmitter is plotted in
Fig. 3.20. The power spectrum complies with the FCC-defined 0-1 GHz UWB spectrum
(mask) also shown. An example of the input data to the UWB transmitter Manchester-
encoded at the rate of 10Mb/s and the data received at the distance of 5cm using a
custom-built UWB receiver is shown in Fig. 3.21.
Table II provides a summary of experimentally measured characteristics of the
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 95
CURRENT (nA)100.10 100.15 100.20 100.25 100.30 100.35 100.400
1
2
3
4
5
6
7MEAN= 100.26nA
SD(3σ) = 34pA
N= 32
NU
MB
ER
OF
OC
CU
RA
NC
ES
(a)
CURRENT (nA)100.10 100.15 100.20 100.25 100.30 100.35 100.400
1
2
3
4
5
6
7
MEAN= 100.18nA
SD(3σ) = 22pA
N= 32
NU
MB
ER
OF
OC
CU
RA
NC
ES
(b)
Figure 3.17: Experimentally measured output current of 32 channels (from 16 chips, two chan-nels each) for input current of (a) 100nA without DEM [32] and (b) 100nA with DEM (thiswork).
integrated CMOS DNA analyzer SoC.
3.7 Experimental Electrochemical Results
To validate the performance of the channel in electrochemical sensing applications,
CV, first, scans of a DNA reporter potassium ferricyanide and a buffer solution were
performed. Potassium ferricyanide K4[Fe(CN)6] is commonly used in electrochemical
DNA detection systems as a redox reporter. Cyclic voltammetry recordings of 20µM
potassium ferricyanide solution and 1M potassium phosphate buffer (pH 7.3) have been
carried out. On-chip waveform generator was utilized to generate the CV excitation
waveform. A 500mV/sec 0.7V peak-to-peak ramp-up-ramp-down CV waveform with
Figure 3.18: Experimentally measured (a) Manchester-encoded input data to the UWB trans-mitter and (b) the output pulses.
a 50ms resting period was applied between a 55µm×55µm flat gold working electrode
in Fig. 3.2(b) and an off-chip Ag/AgCl reference electrode (Basi, RE-5B). The result-
ing CV curves recorded by the chopper-stabilized channel with DEM are shown in
Fig. 3.22. The phosphate buffer CV curve occurs mainly because of the charging and
discharging of the electrode-electrolyte double layer capacitance and thus has no peak.
In contrast, the potassium ferricyanide CV curve shows two distinct peaks at the re-
duction and oxidation voltages of potassium ferricyanide. Indeed, such flat electrodes,
typically produce such distinct redox peaks.
The recorded CV waveforms characteristics (redox peaks location and spacing) are
similar to those reported in the literature [141, 142]. A typical CV curve is shown in
Fig. 1.11(b). The separation between the two peak potentials, ∆Ep=Epc-Epa, can be
used determine the electrochemical reversibility for a redox couple. For a reversible
CV reaction one has [102]
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 97
80.42 80.43 80.44 80.45 80.46−0.12
−0. 1
−0.08
−0.06
−0.04
−0.02
0
0.02
0.04
0.06
0.08
TIME (μs)
AM
PL
ITU
DE
(V
)
SIMULATED
MEASURED
RINGING
Figure 3.19: Wirelessly measured UWB pulse at the distance of 5cm using a custom-built UWBreceiver.
109
−100
−90
−80
−70
−60
−50
−40
−30
108
FREQUENCY (Hz)
OU
TP
UT
PO
WE
R (
dB
m)
FCC MASK
Figure 3.20: Experimentally measured UWB transmitter output spectrum (direct output of thetransmitter driving a 50 ohm load). The output spectrum is compliant with the 0-1GHz FCCUWB band output power criteria
.
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 98
0 0.25 0.750.5 1
0
0.5
1
TIME (µS)
TRANSMITTED DATA @ 10Mb/s
MANCHESTER-ENCODED DATA
AM
PL
ITU
DE
(V
)
(a)
0 0.5 10.25 0.750
0.5
1
TIME (µS)
RECEIVED DATA (5cm)
AM
PL
ITU
DE
(V
)
(b)
Figure 3.21: (a) Manchester-encoded input data to the UWB transmitter and (b) the correspond-ing data received wirelessly at a 5cm distance.
−700 −600 −500 −400 −300 −200 −100 0−12
−10
−8
−6
−4
−2
0
2
4
6
8
V -V (mV)
CU
RR
EN
T (
nA
)
OXIDATION
VOLTAGE
RE WE
REDUCTION
VOLTAGE
FORWARD
SCAN
REVERSE
SCAN
1M POTASSIUM
PHOSPHATE BUFFER
20µM K [Fe(CN )]4 6
Figure 3.22: Experimentally recorded cyclic voltammograms of 1M potassium phosphatebuffer and 20µM potassium ferricyanide solution using the 55µm×55µm working electrodein Fig. 3.2(b)
.
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 99
−700 −600 −500 −400 −300 −200 −100 0−12
−10
−8
−6
−4
−2
0
2
4
6
8
V -V (mV)
CU
RR
EN
T (
nA
)
RE WE
20µM K [Fe(CN) ]
10µM K [Fe(CN) ]
30µM K [Fe(CN) ]
40µM K [Fe(CN) ] 4 6
4 6
4 6
4 6
OXIDATION
VOLTAGEREDUCTION
VOLTAGE
Figure 3.23: Experimentally recorded cyclic voltammograms of 10µM, 20µM, 30µM and40µM potassium ferricyanide solution using the 55µm×55µm working electrode in Fig. 3.2(b)
.
CU
RR
EN
T (
nA
)
10 20 30 40 600
2
4
6
8
K [Fe(CN) ] CONCENTRATION (µM)4 6
Figure 3.24: Calibration curve for the peak reduction current of potassium ferricyanide solutionfor the 55µm×55µm Au
Power Consumption (Channel)Current Conveyer 8µWComparator 19µWBiasing 4µWDigital 11µWTotal (channel) 42µW
CURRENT (nA)
NU
MB
ER
OF
OC
CU
RA
NC
ES
6.4 6.6 6.8 7.0 7.2 7.4 7.6 7.80
3
5
7
9
11
13
MEAN= 7.10nA
SD(3σ) =0.22nA
N= 48
Figure 3.25: Experimentally recorded peak reduction current of the 40µM potassium ferri-cyanide solution recorded using the 55µm×55µm working electrode shown in Fig. 3.2(b) by48-channel on the CMOS DNA analysis SoC.
∆Ep =0.058
n, (3.1)
where n is the number of electrons transferred between the redox complex. This value
is independent of the scan rate for fast electron transfer. Increasing values of ∆Ep as a
function of increasing scan rate indicates the presence of electrochemical irreversibil-
ity. In practice, the theoretical value of 58/n mV for ∆Ep is seldom observed. In all
experiments the potassium ferrocyanide solution was diluted in 1M potassium phos-
phate buffer. This combined with the slow electron transfer kinetics present in case of
our complex multi-material electrodes have caused the peak voltage difference to de-
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 101
∆ pH
CU
RR
EN
T (
nA
)
−2 −1.5 −1 −0.5 0 0.5 1 1.5 22
3
4
5
6
7
8
9
10
11
Figure 3.26: Experimentally measured on-chip pH sensor calibration curve relative to pH of 7.A total of 60 measurements from 3 chips, 20 measurements each, have been performed. Thecorresponding 3σ error bars are shown.
viate from the theoretical 58mV value [33, 34]. In all experiments, the first four CV
curves were discarded and the fifth curve was used as the recorded data. As a result,
the peak recorded redox current is consistent for different concentrations. Other record-
ings [33, 34] using a similar DNA detection method also achieve ∆Ep higher than the
theoretical value of 58mV.
Next, CV scans of a potassium ferricyanide solution with four different concen-
trations (10µM to 40µM) using a 55µm×55µm flat gold working electrode shown in
Fig. 3.2(b) have been performed to study the effect of a change in the DNA reporter
concentration on the recorded redox current. As shown in Fig. 3.23, the peak current
at the reduction and oxidation voltages of potassium ferricyanide increases with an in-
crease in its concentration. The corresponding calibration curve is shown in Fig. 3.24.
This curve demonstrates the linear relationship between the concentration of potassium
ferricyanide and the output redox current.
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 102
CV scans of a potassium ferricyanide solution at 40µM have been conducted on
all 48 channels with 55µm×55µm flat gold working electrodes shown in Fig. 3.2(b)
to study the effect of the channel-to-channel variation on the CV recording results. A
500mV/sec 0.7V peak-to-peak ramp-up-ramp-down CV waveform with a 50ms resting
period was used in this experiment. Fig. 3.25 shows the resulting peak reduction cur-
rents recorded by the 48 channels. The mean peak reduction current is 7.02nA, and the
three-sigma variation is 0.22nA.
To validate the performance of the pH sensors, the sensitivity of the ISFET is mea-
sured in response to change in the solution pH level. A preliminary analysis of the
ISFET characteristics indicated that the pH sensors have different threshold voltages,
due to the trapped charge on the floating gates of the ISFETs. The UV radiation and
bulk substrate biasing (for 8 hours) technique was used to remove the trapped charge
and thus remove the threshold voltage mismatch among the pH sensors. Before the
pH sensor sensitivity is measured, the sensor array must be etched for 10s in a 10%
buffered hydrofluoric acid solution. Measurements made without this step are gener-
ally very noisy and result in a low sensitivity. After the threshold voltage calibration,
the sensitivity of the pH sensor is measured in a 0.1M NaCl electrolyte by adding small
quantities of hydrochloric acid to change the solution pH from five to nine. Recording
the calibrated steps in the measured current leads to the finding that the array has a lin-
ear response of 1.8nA/pH. The corresponding calibration curve with error bars (from
three chips, 20 measurements each) is shown in Fig. 3.26.
3.8 Synthetic Prostate Cancer DNA Detection
The SoC has been validated in label-free amperometric detection of synthetic prostate
cancer DNA. The DNA sequences are synthesized by Integrated DNA technology [154].
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 103
−12
−10
−8
−6
−4
−2
0
2
4
6
8
BARE GOLD ELECTRODE
GOLD ELECTRODE WITH PROBE DNA
NONCOMPLEMENTARY TARGET DNA
COMPLEMENTARY TARGET DNA C
UR
RE
NT
(n
A)
DETECTION
NO DETECTION
NO HYBRIDIZATION
K [Fe(CN) ] 64
OXIDATION VOLTAGE
(SSDNA)
−700 −600 −500 −400 −300 −200 −100 0
V -V (mV)RE WE
(DSDNA)
10
12
Figure 3.27: Experimentally measured cyclic voltammetry results of 5µM prostate cancer syn-thetic DNA detection from the 55µm×55µm flat gold working electrode in Fig. 3.2(b)
.
The following synthetic DNA sequences have been used in the experiments: DNA
DNA (ATA AGG CTT CCT GCC GCG CT) and non-complementary DNA (TTT TTT
TTT TTT TTT TTT TT). All the DNA experiments were conducted at room tempera-
ture. In all the experiments a 500mV/sec 0.7V peak-to-peak ramp-up-ramp-down CV
waveform with a 50ms resting period was applied between the working electrode and a
commercially available off-chip Ag/AgCl reference electrode (Basi RE-5B) [153].
Fig. 3.27 shows cyclic voltammetry scans from an on-chip 55µm×55µm flat gold
electrode for the 5µM prostate cancer synthetic DNA cyclic voltammetry recording,
in a 40µM potassium ferricyanide solution. The CV scan rate and range were set
to 500mV/sec and 0.7V peak-to-peak, respectively, with a 40ms resting period. The
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 104
PROBE
DNA
NONCOMPLEMENTARY
TARGET DNA
COMPLEMENTARY
TARGET DNA
CU
RR
EN
T (
nA
)
−12
−10
−8
−6
−4
−2
0
AT V -V = -450mVRE WE
ABSTRACT TIME
BARE GOLD 55μm×55μm
FLAT ELECTRODE
DETECTIBLE SIGNAL
CHANGE
Figure 3.28: Experimentally measured 5µM prostate cancer synthetic DNA cyclic voltamme-try recording 3σ error bars from 3 chips 60 measurements each from 55µm×55µm flat goldworking electrodes in Fig. 3.2(b).
bare gold electrode CV scan demonstrates well-defined oxidation and reduction peaks,
whereas scans taken using 5µM single-stranded probe DNA attached to electrodes show
a reduction in the oxidation/reduction peaks. This is expected since thiolated DNA
probes create a negatively charged film on the electrode repelling the negatively charged
electrochemical reporter potassium ferricyanide as illustrated in Fig. 3.1. Further adding
a 5µM non-complementary DNA target does not change the CV signal oxidation peak
value significantly indicating that non-specific adsorption is negligible. On the other
hand, adding a 5µM complementary target single-stranded DNA onto the chip leads to
creation of double-stranded DNA on the biosensing electrode resulting in an additional
negative charge and elimination of potassium ferricyanide redox peaks. The corre-
sponding error bars (from 3 chips, 20 measurements each) are shown in Fig. 3.28. As it
can be seen, the detectible signal change in this case is 2.85nA.
The same set of experiments were repeated with the on-die nanostructured elec-
trodes to study their DNA detection capabilities. Fig. 3.29 shows the CV curves ob-
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 105
−700 −600 −500 −400 −300 −200 −100 0
V -V (mV)RE WE
−4
−3
−2
-1
0
1
2
3
4
5
6
CU
RR
EN
T (
nA
)
BARE GOLD ELECTRODE
NONCOMPLEMENTARY TARGET DNA
COMPLEMENTARY TARGET DNA
GOLD ELECTRODE WITH PROBE DNA (SSDNA)
(DSDNA)
DETECTION
NO DETECTION
NO
HYBRIDIZATION
7
Figure 3.29: Experimentally measured cyclic voltammetry results of 100aM prostate cancersynthetic DNA detection, from 2µm×2µm nanostructured working electrodes in Fig. 3.3(e)
.
BARE NANOSTRUCTURED
2μm×2μm MICROELECTRODE
PROBE
DNA
NONCOMPLEMENTARY
TARGET DNA
COMPLEMENTARY
TARGET DNA
CU
RR
EN
T (
nA
)
6
5
4
3
2
1
0
AT V -V = -200mVRE WE
ABSTRACT TIME
DETECTIBLE SIGNAL
CHANGE
Figure 3.30: Experimentally measured 100aM prostate cancer synthetic DNA cyclic voltam-metry recording 3σ error bars from 3 chips, 60 measurements each, from 2µm×2µm nanostruc-tured working electrodes in Fig. 3.3(e)
.
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 106
10−19
10−17
10−15
10−13
10−11
10−9
10−7
CONCENTRATION (M)
I -I
0
25
50
75
100
10−5
DS
DN
AS
SD
NA
SS
DN
AI
(%
)
NUMBER of DSDNA STRANDS
10 102
1040
106
108
1010
1012
1014
10aM
DETECTION
LIMIT
100fM
DETECTION
LIMIT
100nM
DETECTION
LIMITΔI
=
Figure 3.31: Experimentally measured microelectrode characteristics, detection limits and dy-namic ranges in prostate cancer synthetic DNA detection using the three electrodes types shownin Fig. 3.3(d) and (e) and Fig. 3.2(b). Error bars (3 sigma) are from 3 chips, 100 measurementseach.
tained for a nanostructured electrode grown at VRE-VWE = 0mV for 100aM prostate
cancer synthetic DNA concentration, in a 40µM potassium ferricyanide solution. As
expected, compared to the flat gold electrodes the nanostructured electrodes typically
do not exhibit the redox peaks [138]. As it can be seen from Fig. 3.29 the current level
in the presence of complementary target DNA (DSDNA) is smaller compared to the
case where only the probe DNA (SSDNA) is present. The corresponding error bars
(from 3 chips, 60 measurements each) are shown in Fig. 3.30. As it can be seen, the
detectible signal change in this case is 1.1nA.
DNA sensing experiments were conducted for the target DNA concentrations of
1aM to 10µM to study the detection limits of the on-die nanostructured electrodes and
the on-die flat gold electrode. The resulting characteristics, detection limits and dy-
namic ranges of the two nanostructured electrode types and the 55µm×55µm flat gold
electrode are given in Fig. 3.31. ∆I is computed as (IDSDNA-ISSDNA)/ISSDNA)×100,
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 107
where IDSDNA is the redox current after the hybridization and ISSDNA is the redox cur-
rent before the hybridization. All the current recordings for nanostructured electrodes
are taken at VRE-VWE of -200mV and at at VRE-VWE of -250mV for the flat gold
electrodes. The corresponding error bars (from 3 chips, 100 measurements each) are
also shown in Fig. 3.31.
The detection limit, defined as the lowest concentration for which the background-
subtracted signal is three times higher than the standard deviation at that concentra-
tion, for nanostructured electrodes grown at VRE-VWE = 0mV as shown in Fig. 3.31
is 10aM. The 10aM sensitivity achieved using the optimized on-CMOS nanostructured
electrode enables PCR-free detection for many applications. This limit corresponds to
the detection of fewer than 100 copies of the target sequence.
Another benefit of having several types of electrodes on the same chip is that dif-
ferent electrodes cover different concentration ranges. As it can be seen from Fig. 3.31,
the nanostructured electrodes grown at VRE-VWE = 0mV cover a dynamic range (de-
fined as the range at which the 3 sigma error bar of the given concentration is below 100
and above 0 on the y-axis) of 3aM to 100fM, the nanostructured electrodes grown at
VRE-VWE of 100mV cover a dynamic range of 100fM to 90pM, and the 55µm×55µm
flat gold electrodes cover a dynamic range of 1nM to 10µM. As a result, by fabricating
electrodes with different degrees of nanostructuring, we can significantly expand the
dynamic range of the CMOS DNA sensing microsystem (as wide as 140dB with these
types of nanostructured microelectrodes [138]).
Table 5.3 provides a comparative analysis of the presented design and existing
amperometric biochemical sensory microsystems. The design presented in this work
achieves the highest dynamic range and the lowest sensitivity in terms of ADC LSB.
We have shown successful detection of 20-base pair long synthetic prostate cancer DNA
from several types of on-chip Au electrodes. The 10aM detection limit is the lowest de-
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 108
tection limit reported in literature from an integrated circuit-based DNA sensor to date.
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 109
Tabl
e3.
3:C
ompa
rativ
eA
naly
sis
ofA
mpe
rom
etri
cSe
nsor
yM
icro
syst
ems
Syst
emIS
SCC
08JS
SC08
JSSC
09IS
SCC
10IS
SCC
10T
his
Wor
k[2
0][2
6][8
4][9
9][9
7]Te
chno
logy
(CM
OS)
0.18µ
m0.
25µ
m0.
5µm
0.35
µm
0.6µ
m0.
13µ
mPo
wer
25m
W16
0mW
0.6m
W84
.5m
WN
/A0.
35m
WSu
pply
Volta
ge5.
0V2.
5V3.
0V3.
3V3.
3V1.
2VC
hip
Are
a11
.2m
m2
15m
m2
2.25
mm
24m
m2
25.8
mm
29m
m2
Ele
ctro
deC
ount
576
5410
010
040
600
Cha
nnel
Sens
ing
Prot
ocol
CV
CV
ISIS
CA
CV
Cha
nnel
Cou
nt24
1610
010
040
54Ty
peof
Ele
ctro
des
2D2D
2D2D
2D2D
Flat
,3D
Poly
mer
Gol
dG
old
Gol
dpH
Nan
ostr
uctu
red
Gol
dPo
wer
N/A
10m
W6µ
W0.
84m
WN
/A42µ
WD
ynam
icR
ange
N/A
60dB
58dB
N/A
50dB
93dB
(3-m
ode)
Con
vers
ion
Rat
e10
Hz
10kH
z10
kHz
N/A
1Hz
10kH
zSe
nsiti
vity
97pA
240p
A10
kHz
330p
A25µ
V8.
6pA
EN
OB
11bi
ts9
bits
8bi
tsN
oA
DC
12bi
ts9.
1bi
tsW
avef
orm
Gen
erat
orN
oN
oY
esY
esN
oY
esTy
pe—
—Sq
uare
Wav
eSq
uare
Wav
e—
8-bi
tPro
gram
mab
leFr
eque
ncy
(Per
iod)
——
10kH
z50
MH
z—
10kH
zPo
wer
——
——
—1.
1mW
(5nF
Loa
d)Tr
ansm
itter
No
No
No
No
No
Yes
Prot
ocol
——
——
—0-
1,3-
10.6
GH
zU
WB
Dat
aR
ate
——
——
—10
Mbp
sPo
wer
——
——
—10
0µW
Sens
ors
Type
DN
AD
NA
Prot
ein
DN
AD
NA
DN
AO
n-di
eY
esY
esY
esY
esY
esY
esL
abel
-fre
eN
oY
esN
oN
oN
oY
esPC
R-f
ree
No
No
No
No
Yes
Yes
Bio
mol
ecul
eTy
pe30
Bas
e18
Bas
eB
ilaye
rLip
idB
ovin
eSe
rum
Sing
leN
ucle
otid
e20
Bas
ePa
irs
Pair
sM
embr
ane
Alb
umin
Poly
mor
phis
ms
Pair
sC
once
ntra
tion
10nM
-100
nM10
0nM
1µM
100m
MN
/A10
aM-1
0µM
CHAPTER 3. CYCLIC VOLTAMMETRY AND PH SENSING 110
3.9 Chapter Summary
A 54-channel 0.13µm CMOS fast-scan cyclic voltammetry DNA analysis SoC has been
presented. The microsystem includes 600 time-multiplexed DNA sensors and 54 pH
sensors. It also includes an arbitrary waveform generator, an on-chip memory, an in-
channel low-noise chopper-stabilized frontend current conveyer with dynamic element
matching, an in-channel dual-slope ADC and a fully digital ultra-wideband transmit-
ter. Chopper stabilization achieves input-referred noise of less than 0.13pA over the
operating bandwidth. Dynamic element matching improves current conveyer accuracy
by 54 percent at the 100pA input current level. The in-channel SRAM enables in-
channel calibration which results in a 17 percent improvement in channel-to-channel
ENOB variation. Each channel occupies an area of 0.06mm2 and consumes 42µW of
power from a 1.2V supply. The presented current-to-digital channel design achieves a
combined dynamic range of 93dB with the sensitivity of 8.6pA. Two types of nanostruc-
tured microelectrodes and one type of a flat gold electrode have been characterized in
on-CMOS DNA prostate cancer detection. The on-chip nanostructured microelectrodes
achieve label-free PCR-free detection limit of 10aM, which is the lowest reported on-
CMOS detection limit.
111
Chapter 4
Impedance Spectroscopy DNA
We present a 54-channel, mixed-signal CMOS DNA analyzer that utilizes frequency
response analysis (FRA) to extract the real and imaginary impedance components of
the biosensor. Two computationally intensive operations, the multiplication and in-
tegration required by the FRA algorithm, are performed by an in-channel dual-slope
multiplying ADC in the mixed-signal domain resulting in minimal area and power con-
sumption. Multiplication of the input current by a digital coefficient is implemented
by modulating the counter-controlled duration of the charging phase of the ADC. In-
tegration is implemented by accumulating output digital bits in the ADC counter over
multiple input samples. The 1.2mm×1.6mm prototype fabricated in a 0.13µm standard
CMOS technology has been validated in prostate cancer synthetic DNA detection. Each
channel occupies an area of only 0.06mm2 and consumes 42µW of power from a 1.2V
supply.
4.1 Introduction
This Chapter presents a scalable, multi-channel, compact and low-power CMOS impedance
spectroscopy DNA analyzer. Frequency response analysis (FRA) algorithm is utilized
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 112
to extract the real and imaginary components of the biosensor impedance [156]. The
proposed microsystem consists of a programable on-chip waveform generator and 16
impedance extraction channels. Each channel includes a current-mode input dual-slope
multiplying ADC. It efficiently performs multiplication and integration, two computa-
tionally intensive operations required to implement the FRA algorithm. Multiplication
of the input current by a digital coefficient is implemented by modulating the counter-
controlled duration of the charging phase of the ADC by that coefficient. Integration
is implemented by accumulating the output digital bits in the ADC counter. The dual-
slope multiplying ADC utilizes mostly the same circuits as a conventional dual-slope
ADC, and the multiplication and integration are achieved by modifying the ADC algo-
rithm. The rest of the chapter is organized as follows. Section II presents the principle
of DNA detection on a CMOS die. Section III presents the impedance spectroscopy
VLSI architecture. Section IV details the circuit implementation of the VLSI architec-
ture. Section V demonstrates the electrical experimental results obtained from a 0.13µm
CMOS prototype. In Section VI, results of on-chip impedance spectroscopy of DNA in
prostate cancer screening are presented.
4.2 DNA Detection Principle
The principle of the label-free DNA detection method is shown in Fig. 3.1. It employs
potassium ferrocyanide K4[Fe(CN6)] reporter. Potassium ferrocyanide is a negatively
charged redox complex with well-defined electrochemical signature exhibiting oxida-
tion and reduction currents at VWE-VRE of -450mV and -200mV respectively [140].
The maximum electron transfer between the electrode and potassium ferrocyanide is
achieved in the absence of DNA target and probe as illustrated in Fig. 3.1(a). The
oxidization current IOX drops when the Au electrode surface is hybridized with nega-
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 113
tively charged probe DNA, as illustrated in Fig. 3.1(b). When the complementary DNA
binds with the probe DNA, the surface negative charge further increases and the oxi-
dization current is further reduced, as shown in Fig. 3.1(c). This change in the current
is an indicator of the target DNA presence and concentration. In other words, the pres-
ence of negatively charged DNA on the biosensor surface is translated to a decrease
in the potassium ferrocyanide oxidation/reduction current creating a detectible signal
change [141, 142].
The DNA sensing electrodes are created by first forming passivation openings on
the top metal layer (aluminum) of the CMOS die similarly to how it is done for wire
bond pads. Electroless electroplating is then employed to deposit nickel (2µm), palla-
dium (0.2µm) and gold (0.1µm) on the exposed Al surface to form a bio-compatible
electrode surface. After the electrode fabrication, the die is wire-bonded and the bond-
ing wires are insulated with a biocompatible epoxy to enable on-chip electrochemical
experiments without damaging the bonding wires [157].
∫
∫
cos(ωt)
BIOSENSOR
sin(ωt)
FRA UNIT
REAL
IMAG
Asin(ωt+φ)
SIG
NA
L G
EN
ER
AT
OR
(Q
- O
SC
ILL
AT
OR
)
Figure 4.1: Block diagram of a frequency-response analyzer (FRA) system for biosensorimpedance spectroscopy.
4.3 Impedance Spectroscopy VLSI Architecture
A small-signal model of the electrode-electrolyte interface in an electrochemical cell is
shown in the center of Fig. 1.8. Fast fourier transform (FFT) and frequency-response an-
alyzer (FRA) are two methods widely used for characterizing the electrode impedance [156].
Figure 4.4: Principle of the sine wave generation.
In this work the FRA algorithm has been chosen to implement a sensory array
impedance spectroscopy microsystem. The two key components in this system are
the multiplier and the digital integrator. Both of these operations are implemented with
an in-channel multiplying dual-slope ADC that reuses the circuits of a conventional
dual-slope ADC [158].
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 116
0.00
0.25
0.50
0.75
1.00
1.25
|REA
L R
ELA
TIV
E ER
RO
R| (
%)
4−110
010
110
210
31010
4−110
010
110
210
31010
FREQUENCY (Hz)
0.00
0.25
0.50
0.75
1.00
1.25
|IMA
G R
ELA
TIV
E ER
RO
R| (
%)
FREQUENCY (Hz)(a)
(b)
NUMBER OF INTEGRATION CYCLES ( )4
100
101
102
103
105
10
NUMBER OF INTEGRATION CYCLES ( )4
100
101
102
103
105
10
N
N
Figure 4.5: Absolute value of the relative error of the biosensor R-C model impedance as afunction of frequency due to stepwise approximation of the interrogation signal for the : (a)real, and (b) imaginary components.
4.4 Circuit Implementation
4.4.1 Multi-channel System-Level Architecture
The functional block diagram of the impedance spectroscopy microsystem based on
the FRA algorithm is shown in Fig. 4.2. The microsystem is comprised of a pro-
gramable analog waveform generator, a programmable digital pattern generator and
an array of impedance extraction units. The waveform generator produces the interro-
gation waveform sin(ωt) and drives the reference electrode with it. The digital pattern
generator generates digital multiplication coefficients representing either sine wave or
cosine wave that are synchronized with the interrogation waveform. Each impedance
extraction unit consists of a dual-slope multiplying ADC (DS-MADC). The front-end
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 117
−
+
S1
S2
S2
IIN
WE
VWE
VWE
IREF+
IREF-
CFS3
VOUT
3
(S1 ,S3,S2 )
CLK
RESET
UP/
DOWN
OR
REAL
IMAG
D = sin(n) OR D = cos(n) (FROM PATTERN GENERATOR)
TIME A
VWES1
UP
/DO
WN
CO
UN
TE
R
IN/O
UT
LA
TC
H
READ_IN
READ_OUT
WRITE
6
CO
NT
RO
L
L
OG
IC
TIME B TIME C
−
+
−
+
−
+
OTA1
VOUT
S
S
UP/DOWN
READ_OUT
WRITE
READ_IN
1
2
3
DT1T
RDDT
2
RESET
A B C
S2OR
t
S
VLATCH
(a)
(b)
Figure 4.6: (a) Dual-slope multiplying ADC VLSI architecture, and (b) timing diagram of allrelevant signals.
current-to-voltage converter is an analog integrator that acquires an input current at
the low-impedance input node set to a controlled potential. The DS-MADC multiplies
the biosensor response IIN with a set of digital coefficients D representing sin(ωt) or
cos(ωt) that are synchronized with the analog sinusoidal interrogation voltage on the
reference electrode. Next the DS-MADC accumulates the results over one period of the
interrogation signal using a digital integrator (counter), thus extracting the real or the
imaginary components of the biosensor impedance.
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 118
4.4.2 Waveform Generator and Pattern Generator
The frequency response analysis implementation is the simplest when rectangular wave-
forms are used instead of sine wave for both the interrogation and the multiplication
signals. The problem is that severe systematic errors appear due to the higher order
harmonics existing in the rectangular waveforms. Stepwise approximation of the in-
terrogation waveform and the multiplication signals reduces the effect of higher or-
der harmonics and increases the measurement accuracy. It has been shown [130] that
representing both the interrogation signal and the multiplication signals by a coarsely
quantized approximation can significantly reduces the error due to the higher order har-
monics and can significantly reduce the measurement inaccuracy.
= TI IN
I REF
VOUT
TIME
PHASE I PHASE II
DT
2
T2
T1
1
1
DT
2DT
Figure 4.7: Timing diagram illustrating the ADC multiplication function.
In this work a programmable analog waveform generator is utilized to generate
the stepwise approximation of the interrogation signal. The block diagram of the pro-
grammable waveform generator is shown in Fig. 4.3. It is composed of an 8-bit R-2R
DAC, an on-chip SRAM1 and a bidirectional counter. The DAC coefficients are stored
in the on-chip SRAM1. The on-chip waveform generator generates both stepwise sinu-
soidal and cyclic voltammetry (CV) ramp waveforms. It provides a wide range of user-
controlled rate and amplitude parameters with a maximum CV scan range of 1.2V, and
scan rate ranging between 0.1mV/sec to 400V/sec and the sinusoidal frequency range
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 119
from 100mHz to 50kHz. An off-chip clock generator provides the variable frequency
clock signal enabling the waveform generator to generate the interrogation signal at dif-
ferent frequencies. The DAC occupies an area of 0.012mm2 and dissipates 1.1mW of
power from a 1.2V supply when driving a load of 5nF at 50kHz. At low frequencies the
interrogation sine wave is represented by 64 samples per period. The DAC coefficients
for the first 16 samples (D0 to D15) of the waveform generator output waveform and for
the corresponding digital sin/cos multiplication coefficients are stored in two on-chip
global SRAM banks (SRAM1 and SRAM2, respectively), as shown in Fig. 4.2. The
SRAMs occupy an area of 0.028mm2 and dissipate 0.9µW of power when clocked at
50kHz. By symmetry, 64 samples in one period are generated from the 16 samples
stored on-chip. As shown in Fig. 4.4, for a sine wave, the first 16 samples stored in the
SRAM1 generates the first quadrant of the sine wave. In this case the counter control-
ling the SRAM1 is counting up and S is set to 00, in Fig. 4.3. In the next quadrant the
counter controlling the SRAM1 counts down thus reversing the order of the samples
and generating the second quadrant of the sine wave. S is set to 00 in this case. The
third and forth quadrants are generated in the same manner but in this case the polarities
of the samples fed into the DAC are reversed (by setting S to 01). To generate a cosine,
the cycle starts from the second quadrant instate of the first one. If a CV waveform is
required the up/down counter is directly interfaced to the waveform generator DAC (by
setting S to 11).
As the interrogation frequency increases, the number of samples representing the
interrogation and multiplication signals decreases and at 10kHz both signals are repre-
sented by three samples. This greatly reduces the ADC speed requirement while the
error caused by the reduction in number of samples is kept low by averaging the results
over multiple cycles.
An ideal model of the impedance spectroscopy microsystem has been constructed
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 120
using verilog-A model components (ideal analog integrator and multiplying ADC) to
study the effect of step-wise approximation of the interrogation signal on the overall
system accuracy. First a simulation was performed where a sinusoidal voltage stimulus
(8-bit accurate generated by an ideal DAC) with the frequency swept from 0.1Hz to
10kHz was applied to the biosensor model shown in Fig. 4.2. The value of RS was set
to 1GΩ, CWE was set to 500pF, CDB was set to 300pF and RCT was set to 1MΩ. The
sensor response was recorded with the ideal analog integrator and multiplied, using an
ideal multiplying dual-slope ADC, by an 8-bit accurate digital multiplication coefficient
representing sine or cosine in the FRA algorithm in Fig. 4.1. Next, the same set of sim-
ulations were performed in which both digital multiplication coefficients and the analog
sinusoidal voltage interrogation signal are represented by 64 samples per period at low
frequencies. In this simulation as the interrogation frequency increases, the number of
samples representing the interrogation and multiplication signals decreases. Also, in
this simulation the results are averaged over multiple cycles (N) to reduce the errors
caused by reduction in the number of samples. The absolute value of the relative error
of the biosensor impedance computed using 8-bit accurate interrogation signal and 8-bit
accurate multiplication coefficients versus the simulated biosensor impedance obtained
from stepwise approximation of these signals is shown in Fig. 4.5. The absolute value
of the relative error stays below 0.9% and 1.0% for the real and imaginary component
of the biosensor respectively.
4.4.3 Dual-slope Multiplying ADC Channel
The VLSI architecture of one channel of the integrated spectrum analyzer is depicted in
Fig. 4.6(a). Each channel consists of an integrating amplifier with an on-chip 10pF ca-
pacitor CF , a high-speed latched comparator and digital blocks. The integrator switches
are implemented with low-leakage switches as shown.
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 121
The conventional dual-slope ADC operates in two phases as depicted in Fig. 4.7.
In phase I the integrating capacitor CF is charged for a predetermined period of time
T1. Next, during the second phase of the operation, the capacitor is discharged to zero
by a DC reference current. By counting the time T2, a digital representation of IIN
can thus be obtained as (T2/T1)×IREF . To implement multiplication of the input cur-
rent by a digital sin/cos coefficient (D1 to D15) as needed by the FRA algorithm, the
duration of phase I is scaled with a constant coefficient D<1 as shown in Fig. 4.7. In
this case by counting the time DT2, a digital representation of DIIN can be obtained as
D×(T2/T1)×IREF .
To extract the real and imaginary components of the biosensor impedance, the input
current IIN is multiplied by the reference sine or cosine digital coefficient denoted as D
(stored in SRAM2, in Fig. 4.2) and the results are integrated over one period by a 16-bit
counter as follows
Real =
∫ T
0
IIN × sin(ωt)dt =
N/2∑1
IIN × |sin[n]| −N∑
N/2+1
IIN × |sin[n]|(4.3)
Imag =
∫ T
0
IIN × cos(ωt)dt =
N/2∑1
IIN × |cos[n]| −N∑
N/2+1
IIN × |cos[n]|(4.4)
where N is the number of samples in one period and sin[n] and cos[n] are the digital
multiplication coefficients D that are stored in the SRAM2. The multiplying dual slope
ADC performs the multiplication required by equations (5.3) and (4.4), thus the need
for 16 digital multipliers is eliminated.
The timing diagram of the ADC for a typical conversion cycle is shown in Fig. 4.6(b).
First, the integrating counter is reset. At the same time, the sin/cos multiplication coeffi-
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 122
VIN+
VIN-
VBIASN
VCASCP
CASCNVV
OUT
M1 M2
M3 M4
M5 M6
M7 M8
M9 M10M11
chopM
Figure 4.8: Chopper-stabilized folded-cascode OTA1 in the analog integrator.
VIN+ IN-
V
VLATCH
V
OUT+V
OUT-
VLATCH VLATCH
VOUT- VOUT+
VIN+
VIN-
VBIAS
(a) (b)
M1
M2
M3
M4 M
5M
6 M7
M8 M9
M10
M11 M12
M13 M14M15 M16
Figure 4.9: Comparator circuit diagram. (a) One of the three identical gain stages, and (b) thehigh-speed latch.
cient, D, is loaded into the in-channel input latch (time A). Next, the in-channel counter
counts up from zero to time DT1 and the input current is integrated onto capacitor CF .
After time DT1, the voltage on the capacitor is held constant for a fixed time interval
TRD. During this time the content of the output latch (zero for the first conversion cycle)
is loaded into the counter (time B). During time DT2, depending on the comparator out-
put, the integrating capacitor is discharged using the appropriate current source, IREF+
or IREF−. During time DT2 the counter counts up or down depending on the sign of
the input current in phase I and the final value of the counter is written into the output
The electrode model shown in Fig. 1.9 was used first to emulate the biosensor. The
value of RS was set to 1GΩ, CWE was set to 500pF, CDB was set to 300pF and RCT
was set to 1MΩ. To verify the impedance extraction capability of the microsystem, a
sinusoidal voltage stimulus (generated by the on-chip DAC) with the frequency swept
from 0.1Hz to 10kHz was applied to the biosensor model. Figs. 4.19 (a) and (b)
demonstrate that the fabricated prototype tracks the theoretical model well over the
full range of frequencies. The absolute value of the relative error of the ideal biosen-
sor impedance verses the measured biosensor impedance as a function of frequency is
shown in Fig. 4.20. The absolute value of the relative error stays below 8.4% and 7.5%
for real and imaginary component of the biosensor respectively.
Table 4.4 provides a summary of experimentally measured characteristics of the
integrated impedance spectroscopy microsystem.
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 131
RE
AL
IMP
ED
AN
CE
(Ω
)
10 5
10 6
10 7
10 8
10 9
IDEAL
MEASUREDIM
AG
IMP
ED
AN
CE
(Ω
)
10 4
10 5
10 6
10 7
10 8
10 9
10 10
IDEAL
MEASURED
500pF1GΩ
1MΩ
WERE300pF
500pF1GΩ
1MΩ
WERE300pF
−110
010
110
210
3
FREQUENCY (Hz)
10 104
(a)
−110
010
110
210
3
FREQUENCY (Hz)
10 104
(b)
Figure 4.19: Off-chip biosensor model impedance as a function of frequency experimentallymeasured by the impedance spectroscopy microsystem: (a) real, and (b) imaginary components.
4.6 Analog vs Digital vs Mixed-Signal Multiplication
Multiplication of the input current by a digital coefficient is implemented by modulat-
ing the counter-controlled duration of the charging phase of the ADC, as described in
Section 4.4.3.
The number of extra logic gates required for implementing in-ADC multiplication
in this design is 23. They are not required to be clocked at the ADC clock speed. The
digital gates required for the multiplications make a state transition only twice during
one entire ADC conversion cycle. For example, if the ADC is generating 10ksps while
being clocked at 10MHz, then the digital logic will be clocked at 20kHz. As a result, the
extra logic cells do not have a significant contribution to the overall power consumption
of the recording channel.
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 132
0
2
4
6
8
10
|REA
L R
ELA
TIV
E ER
RO
R| (
%)
4−110
010
110
210
31010
4−110
010
110
210
31010
FREQUENCY (Hz)
0
2
4
6
8
10
|IMA
G R
ELA
TIV
E ER
RO
R| (
%)
FREQUENCY (Hz)(a)
(b)
Figure 4.20: Absolute value of the relative error of the off-chip biosensor model impedance asa function of frequency: (a) real, and (b) imaginary components.
Mixed-signal multiplication approaches presented here can be compared to the ana-
log and digital multiplication approaches in term of area, power and linearity.
Analog multiplication requires a frontend circuit that performs the multiplications
before the signal is passed to the ADC. This will result in added noise, linearity degra-
dation and area requirement which otherwise do not exist in the mixed-signal or digital
approach. Compared to the analog design shown in Fig. 1.16, our approach results in
32µm2 area saving (0.5 percent of total area) and 6.73µW power saving (15.2 percent
of total power) per recording channel. Also, in terms of non-linearity degradation, the
purely analog multiplication approach in Fig. 1.16 adds 4.2 percent linearity degrada-
tion to the channel. This would be in addition to the degradation of the ADC ENOB
(shown in Fig. 4.15) due to the reduction of the signal amplitude vs multiplication
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 133
−1
100
101
102
103
103
104
105
106
10 7
FREQUENCY (Hz)
IMA
G I
MP
ED
AN
CE
(Ω
)
10
103
104
105
106
107
104
−1
100
101
102
103
FREQUENCY (Hz)
10 104
RE
AL
IM
PE
DA
NC
E (
Ω)
(a)
(b)
1.0mM K [Fe(CN) ]3 6
0.1mM K [Fe(CN) ]3 6
1.0mM K [Fe(CN) ]3 6
0.1mM K [Fe(CN) ]3 6
Figure 4.21: Potassium ferricyanide solution impedance as a function of frequency experi-mentally measured by the impedance spectroscopy microsystem: (a) real, and (b) imaginarycomponents.
coefficients.
The main disadvantage of the digital multiplication vs mixed-signal multiplication
approach is the excessive area required by the digital multiplier. For example the digital
multiplier shown in Fig. 1.17 requires 436 logic cells. This results in an area of 0.015
mm2 (adding up the area required by the digital cells listed in Fig. 1.17 scaled to CMOS
0.13µm technology as listed in Table. 4.5 ). The total channel area in this design is 0.06
mm2, thus the area required by the digital multiplier (0.015 mm2) would correspond to
25 percent of the channel area. By comparison the digital overhead (23 gates with a
total area of 0.0009mm2) required by the ADC for multiplication in the mixed signal
approach corresponds to 1.5 percent of the total channel area.
The power consumption of the digital multiplier shown in Fig. 1.17 is 550pJ/MMPS.
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 134
Table 4.5: Serial Digital Multiplier Area Breakdown
Gate Type Number of Gates Area/Gate (µm2) Total Gate Area (µm2)AND12 64 1.74107E − 11 1.11429E − 09AOI21 37 3.48214E − 11 1.28839E − 09AOI22 16 5.22321E − 11 8.35714E − 10DFF 32 1.67143E − 10 5.34857E − 09
Simulations shows that the ADC logic required for digital multiplication consumes
40pJ/MMPS. The total power consumed by the fully digital multiplier or by the extra
logic required by the MADC is not a significant portion of total channel power con-
sumption. The main advantage of the digital multiplier vs the mixed-signal multiplier
is that the multiplication is done after the ADC and the multiplication does not degrade
the ADC ENOB. In our application the degradation of the ADC ENOB does not re-
sult in a significant error in the system level performance. As shown in Figs. 4.20, in
the impedance extraction experiment, the absolute value of the relative error due to the
ADC ENOB degradation stays below 8.4% and 7.5% for real and imaginary component
of the biosensor respectively.
A comparison of the mixed-signal vs analog vs digital multiplication is shown in
Table 4.6. The mixed-signal multiplication approach presented in this work results in
small area and power requirements compared to the fully analog and fully digital multi-
plication approach. The mixed signal approach does not result in linearity degradation
as in the case of the analog multiplication approach. The fully digital approach provides
better system accuracy at the cost of excessive area requirement.
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 135
Table 4.6: Comparison of Mixed-Signal vs Analog vs Digital Multiplication
Performance Metric Mixed-Signal Analog DigitalArea Small Small Large
Power Small High SmallLinearity Degradation No Yes NoENOB Degradation Yes Yes No
4.7 Electrochemical Experimental Results
A set of electrochemical experiments have been conducted to validate the function-
ality of the impedance spectroscopy microsystem. In these tests a 3-electrode setup
has been utilized with on-chip gold-plated 55µm×55µm working electrode, on-chip
gold plated counter electrode and an off-chip Ag-AgCl reference electrode. The refer-
ence and counter electrodes are driven by the on-chip three-electrode regulation loop as
shown in Fig. 4.11(a) and all the excitation waveforms are generated using the on-chip
waveform generator.
First, in order to validate the design, impedance spectroscopy recordings of 0.1mM
and 1mM potassium ferricyanide in 1µM potassium phosphate buffer (pH 7.3) were
carried out. A 9mV 0.1Hz to 10kHz sine wave was applied between the WEs and an
off-chip Ag-AgCl reference electrode. The real and imaginary impedance results ob-
tained from the two concentrations of the potassium ferricyanide solution are shown in
Figs. 4.21 (a) and (b). An increase in the concentration of the potassium ferricyanide
results in a decrease in value of RS and RCT and increase in value of CWE . The mea-
surements of potassium ferricyanide validate impedance spectroscopy microsystem in
DNA sensing applications.
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 136
Tabl
e4.
7:C
ompa
rativ
eA
naly
sis
ofR
epor
ted
Am
prom
etri
cSe
nsor
yM
icro
syst
em
Syst
emIS
SCC
08JS
SC08
JSSC
09IS
SCC
10IS
SCC
10T
BIO
CA
S07
Thi
sW
ork
[20]
[26]
[84]
[131
][9
7][6
9]Te
chno
logy
(CM
OS)
0.18
µm
0.25
µm
0.5µ
m0.
35µ
m0.
6µm
0.5µ
m0.
13µ
mPo
wer
25m
W16
0mW
0.6m
W84
.5m
W—
1.2m
W0.
35m
WSu
pply
Volta
ge5.
0V2.
5V3.
0V3.
3V3.
3V3.
0V1.
2VC
hip
Are
a11
.2m
m2
15m
m2
2.25
mm
24m
m2
25.8
mm
29m
m2
1.92
mm
2
Ele
ctro
deC
ount
576
4810
010
040
1664
Cha
nnel
Cha
nnel
Cou
nt24
1610
010
040
1616
Type
Of
2D2D
2D2D
—O
ff2D
Ele
ctro
des
Poly
mer
Gol
dG
old
Gol
d—
Chi
pG
old
Sens
ing
Prot
ocol
CV
CV
ISIS
CV
CA
ISPo
wer
—10
mW
6µW
0.84
mW
—4.
3µW
42µ
WD
ynam
icR
ange
—60
dB50
dB97
dB80
dB—
140d
B(3
-mod
e)C
onve
rsio
nR
ate
10H
z10
kHz
10kH
z—
1Hz
10H
z10
kHz
Sens
itivi
ty(L
SB)
97pA
240p
A—
330p
A25µ
V—
320f
AE
NO
B11
bits
9bi
ts8
bits
No
AD
C12
bits
8bi
ts9.
3bi
tsW
avef
orm
Gen
erat
or—
—Y
esY
es—
—Y
esTy
pe—
—Sq
uare
Wav
eSq
uare
Wav
e—
—8-
bitP
rogr
amm
able
Freq
uenc
y—
—10
KH
z50
MH
z—
—10
kHz
Pow
er—
——
——
—1.
1mW
(5nF
Loa
d)
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 137
The SoC was also extensively validated in DNA analysis for detection of prostate
cancer synthetic DNA. Fig. 4.22(a) shows real and imaginary impedance components
of an on-chip Au electrode hybridized with 5µM single-stranded prostate cancer syn-
thetic probe DNA. Adding a 5µM noncomplementary target DNA does not significantly
change the real and imaginary components of the electrode impedance indicating that
non-specific adsorption is negligible. Adding a 5µM complementary prostate cancer
synthetic target DNA leads to significant reduction of potassium ferrocyanide redox
current, resulting in increase in both real and imaginary components of the electrode
impedance. This is due to the additional negative surface charge resulting from forma-
tion of double-stranded DNA on the electrode surface. The 3-sigma error bars (from 3
chips) with the real and imaginary impedance detection noise margins of approximately
18.3KΩ and 20.9MΩ respectively are shown in Fig. 4.22(b).
Table 4.7 provides a comparative analysis of the presented design and existing
amperometric biochemical sensory microsystems. The design presented in this work
achieves the highest dynamic range and the lowest sensitivity in terms of LSB. Also,
the design presented in this Chapter requires less time to extract real and imaginary
components of the biosensor compared to the FRA implementations in which a square
waveform is utilized for multiplication and integration. The stepwise approximation of
the interrogation waveform and multiplication coefficients significantly reduces the sys-
tematic error caused by higher order harmonics present in the rectangular waveforms,
thus eliminating the need for excessive averaging that is time consuming at the lower
frequency range. This enables real-time monitoring and analysis of DNA hybridization.
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 138
104
105
106
107
108
109
1010
10−1
100
101
102
103
104
FREQUENCY (Hz)
RE
AL
IM
PE
DA
NC
E (
Ω)
PROBE
NONCOMPLEMENTARY TARGET COMPLEMENTARY TARGET
106
107
RE
AL
IM
PE
DA
NC
E (
Ω)
COMPLEMENTARY
TARGET DNA
NONCOMPLEMENTARY
TARGET DNA
PROBE
DNA
COMPLEMENTARY
TARGET DNA
NONCOMPLEMENTARY
TARGET DNA
PROBE
DNA
DETECTION
NOISE MARGIN
DETECTION
NOISE MARGIN
104
105
106
107
108
109
1010
107
108
IMA
G I
MP
ED
AN
CE
(Ω
)
AT 10 HzAT 10 Hz
(a)
(b)
IMA
G I
MP
ED
AN
CE
(Ω
)
Figure 4.22: (a) Impedance spectrum of 5µM prostate cancer synthetic DNA probe, comple-mentary, non-complementary targets and (b) the corresponding 3-sigma error bars (from 3 chips,20 measurements each).
4.8 Chapter Summary
A 54-channel, mixed-signal CMOS impedance spectroscopy DNA analyzer is pre-
sented. It consists of a programable waveform generator, on-chip memory and multiple
impedance extraction units. Multiplication and integration, two operations required
for frequency response analysis (FRA) algorithm, are efficiently performed by the in-
channel current-mode input dual-slope multiplying ADC with negligible resources over-
head. The ADC combines impedance extraction and analog-to-digital conversion into a
single conversion cycle. The impedance spectroscopy microsystem was implemented in
CHAPTER 4. IMPEDANCE SPECTROSCOPY DNA 139
a CMOS 0.13µm technology. Each channel occupies an area of 0.06mm2 and consumes
42µW of power from a 1.2V supply.
140
Chapter 5
Temperature Regulation
A 9×6-arrayed-cell mixed signal CMOS thermal controller for on-die DNA sensing is
presented. The SoC reuses the circuits for impedance spectroscopy and cyclic voltam-
metry to also perform temperature regulation. The on-chip, in-cell heating and temperature-
sensing elements are implemented in standard CMOS without any post-processing.
Using feedback proportional-integral-derivative (PID) control the temperature can be
regulated to within 0.5C of the desired value. The two computationally intensive oper-
ations, the multiplication and subtraction required by the PID algorithm, are performed
by the in-cell dual-slope multiplying ADC in the mixed-signal domain, resulting in
small area and low power consumption. Specifically, multiplication of the input current
by a digital coefficient is implemented by modulating the counter-controlled duration of
the charging phase of the ADC. Subtraction is implemented by the ADC counter. The
3mm×3mm prototype fabricated in a 0.13µm CMOS technology has been fully ex-
perimentally characterized. Each channel occupies an area of 0.06mm2 and consumes
42µW from a 1.2V supply.
CHAPTER 5. TEMPERATURE REGULATION 141
5.1 Introduction
In this Chapter we present a 0.13µm CMOS distributed temperature regulator with an
on-chip mixed-signal PID controller. The hybrid analog-digital nature of the PID con-
troller yields a compact implementation with a low power dissipation overhead. This
enables distributed localized temperature control in large-scale temperature-dependent
parallel sensing applications such as on-CMOS amperometric electrochemical DNA
analysis essays.
The presented temperature controller is a part of a highly-integrated multi-functional
SoC that performs amperometric electrochemical DNA analysis. The DNA analysis
SoC consists of 54 recoding channels and 600 gold microelectrodes. It performs label-
free DNA analysis and pH sensing for prostate cancer detection. We previously reported
on the DNA sensing functionality of this SoC [137]. Accurate DNA analysis requires
distributed temperature control. We reported a brief preliminary summary of how the
DNA sensing circuits and the temperature control circuits are integrated and shared on
the same die [87]. Here we present a detailed report on the VLSI architecture and circuit
implementation of the distributed temperature regulator that extends on [87] and offer
an in-depth treatment of the algorithm design and the circuit implementation, as well as
extensive experimental results illustrating temperature regulation characteristics.
The distributed temperature controller consists of a spatial array of 54 heating ele-
ments and 54 temperature-sensing elements, one such pair for each DNA sensing chan-
nel. The heaters and temperature sensors are interfaced to a distributed mixed-signal
PID controller that regulates the temperature of the chip. The PID controller re-uses
key circuits from the DNA-sensing channel, thus significantly saving silicon area. The
re-used DNA sensing circuits are reconfigured to perform both temperature sensing and
computation needed by the PID temperature control algorithm. This architecture allows
for independent thermal regulation of each channel. Channels can be further thermally
CHAPTER 5. TEMPERATURE REGULATION 142
isolated from each other using a CMOS post-processing technique as described in [170].
In that method a combination of thin-film metal deposition, photolithography and wet
etching is utilized to etch out the dielectric and the bulk silicon around the area of the
interest and thus thermally isolating it from the rest of the chip.
+u(k) = u(k-1) + c e(k)
+ c e(k-1) + c e(k-2)
DESIRED TEMP, t(k)
+
e(k)y(k)
u(k)(c ,c ,c )
0 1 2 PWM
ACTUATOR
PID CONTROLLER
0
1 2
RE
TEMPERATURE
SENSOR
RE
HEATER
TEMPERATURE CONTROLLER
Figure 5.1: Block diagram of PID implementation of the temperature regulation loop.
The rest of the Chapter is organized as follows. Section II provides background on
temperature regulation principles. Section III describes the temperature-sensing princi-
ple. Section IV presents the VLSI architecture of the temperature controller. Section V
details the circuit implementation of the VLSI architecture. Section VI demonstrates
the experimental results obtained from the 0.13µm CMOS prototype.
5.2 Thermal Control Principles
5.2.1 Temperature regulation
The block diagram of the temperature regulation loop is shown in Fig. 5.1. It consists
of a heater, a temperature sensor, a temperature controller, and an actuator as the feed-
back element. In this work, a proportional-integral-derivative (PID) algorithm has been
chosen for the temperature controller, since this class of algorithms is effective in the
regulation of thermal systems [162, 163]. As shown in Fig. 5.1, a discrete-time PID
controller calculates the error value, e(k), as the difference between a measured input
signal, y(k), and a desired input signal, t(k). The controller attempts to minimize the
CHAPTER 5. TEMPERATURE REGULATION 143
error by adjusting the input to the feedback element (actuator), u(k), using the process
control inputs (c0, c1, c2). The PID controller algorithm involves three separate con-
stant parameters, namely, the proportional, the integral, and the derivative components,
which are denoted as P, I, and D, respectively. The proportional component depends on
the present error value, the integral component depends on the past error value, and the
derivative component is a prediction of the future error value. The weighted sum of the
three components can be used to adjust the heat, and thus regulate the temperature of
the microsystem [164].
IREF
IN
VBE1
R1
ᾳ∆VBE
R2
DIG
ITA
L
TE
MP
ER
AT
UR
E
VBE1
VBE2
∆VBE
ᾳ∆VBE
R2
VBE1
R1
V
T T
I
T
I
VBE1
R1
ᾳ∆VBE
R2
(b) (c) (d)
(e)
CHANNEL
I
CTAT
PTAT
(a)
E
C
B
VBE2+
-E
C
B
VBE1
+
-IE1
IE2
IC1
IC2
Q 1
Q 2
IB1
IB2
∆VBE
+ -
Figure 5.2: (a) A pair of BJTs for generating ∆VBE and VBE , (b) Temperature dependenceof the base-emitter voltage, (c) generation of PTAT and CTAT current, (d) generation of tem-perature dependent current and (e) utilization of a current-to-digital channel for temperaturemeasurements.
In this work the temperature of the sensing site is measured using an on-chip temper-
ature sensor. The output of the temperature sensor, y(k), is subtracted from the desired
temperature, t(k), and the resulting error value, e(k), is fed to the PID controller. The
output of the PID controller, u(k), sets the duty cycle of a digital pulse that is gener-
ated by the on-chip digital PWM controlling the on-chip heater. The PID controller is
implemented as a recursive filter [136]. The continuous-time representation of the PID
CHAPTER 5. TEMPERATURE REGULATION 144
controller is given by
u(t) = Kpe(t) +Ki
∫ t
0
e(t)dt+Kdd
dte(t) (5.1)
where Kp is the proportional gain, Ki is the integral gain, Kd is the derivative gain, e(t)
is the error value, and u(t) is the actuation value. A discrete-time representation of (5.1)
in Z-domain is given by
D(z) =u(z)
e(z)= Kp +Ki
Ts
2
z + 1
z − 1+
Kd
Ts
z − 1
z(5.2)
where Ts is the sampling period. Equation (5.2) can be implemented numerically as
Power Consumption(Channel)Current Conveyer 8µWComparator 19µWBiasing 4µWDigital 11µWTotal (channel) 42µW
CHAPTER 5. TEMPERATURE REGULATION 162
NU
MB
ER
OF
OC
CU
RA
NC
ES
49.6 49.7 49.8 49.9 50.0 50.1 50.2 50.30
2
4
6
8
10
12
MEAN = 49.83 C
SD(3σ) = 0.20 C
N= 54
o
o
TEMPERATURE ( C)o
Figure 5.20: Experimentally measured temperature from 54 channels in one chip (in-air).
CHAPTER 5. TEMPERATURE REGULATION 163
Tabl
e5.
3:C
ompa
rativ
eA
naly
sis
ofR
epor
ted
The
rmal
lyR
egul
ated
Bio
sens
ory
Mic
rosy
stem
s
Syst
emIS
SCC
08B
IOC
AS
09T
BIO
CA
S07
ISSC
C10
ISSC
C09
Thi
sW
ork
[20]
[135
][1
32]
[25]
[97]
Tech
nolo
gy(C
MO
S)0.
18µ
m0.
5µm
0.5µ
m0.
13µ
m0.
6µm
0.13
µm
Pow
er25
mW
30m
W—
165m
W—
0.35
mW
Supp
lyVo
ltage
5.0V
5V5.
0V1.
2V3.
3V1.
2VC
hip
Are
a11
.2m
m2
3mm
26m
m2
7.5m
m2
25.8
mm
29m
m2
Ele
ctro
deC
ount
576
91
1640
600
Cha
nnel
Cha
nnel
Cou
nt24
91
1640
54Ty
peof
Ele
ctro
des
2D2D
2D2D
—3D
Gol
dPo
lym
erM
icro
hotp
late
Mic
roho
tpla
teM
agne
ticIS
FET
—D
ynam
icR
ange
—O
ff-C
hip
Off
-Chi
p55
dB80
dB12
8dB
(3-m
ode)
Con
vers
ion
Rat
e10
Hz
——
—1H
z10
kHz
Sens
itivi
ty97
pA—
—0.
3Hz
25µ
V8.
6pA
EN
OB
11bi
ts—
—D
irec
tCon
vers
ion
12bi
ts9.
1bi
tsSi
gnal
Gen
erat
orN
oN
oN
oY
esN
oY
esTy
pe—
——
Sine
Wav
e—
8-bi
tPro
gram
mab
leFr
eque
ncy
——
—1G
Hz
—10
kHz
Pow
er—
——
——
1.1m
W(5
nFL
oad)
Tran
smitt
erN
oN
oN
oN
oN
oY
esPr
otoc
ol—
——
——
0-1
GH
zU
WB
Dat
aR
ate
——
——
—10
Mbs
Pow
er—
——
——
100µ
WE
lect
roch
emic
alSe
nsin
gC
yclic
Volta
mm
etry
Yes
Yes
Yes
No
No
Yes
Impe
danc
eSp
ectr
osco
pyN
oN
oN
oY
esN
oY
espH
No
No
No
No
ISFE
TIS
FET
Tem
pera
ture
Sens
ing
Yes
Yes
Yes
Yes
Yes
Yes
Sens
orC
ount
19
11
4054
Con
trol
No
PID
PID
PID
PID
PID
—(A
nalo
g)(A
nalo
g)(A
nalo
g)(D
igita
l)(M
ixed
-Sig
nal)
—O
ff-C
hip
Off
-Chi
pO
ff-C
hip
On-
Chi
pO
n-C
hip
Acc
urac
y—
0.7o
C2
oC
1o
C0.
5o
C0.
5o
C
CHAPTER 5. TEMPERATURE REGULATION 164
An external heater was utilized to fix the chip temperature at 50 degrees Celsius and
the temperature at each recording channel was recorded to study the effect of channel-
to-channel mismatch. Experimentally measured temperature from 54 channels in one
chip is shown in Fig. 5.20. The mean digital output temperature and the corresponding
standard deviation are 49.83 and 0.20 degree Celsius, respectively.
An example of the temperature regulation cycle in liquid (5mL 1M potassium phos-
phate buffer solution), with steps at 35, 45, 55, and 65 degrees Celsius, is shown
in Fig. 5.21. In these experiments the PWM was clocked at 5kHz and the ADC was
clocked at 10MHz to provide 3 samples (as needed by the timing diagram shown
in Fig. 5.7) in each PID cycle. The solid line is the chip temperature regulated by the
on-chip temperature regulator and the dashed line is the desired temperature. It takes
roughly 10 seconds to achieve a 5 degree increase in the chip temperature. Measured
absolute value of the relative error of the PID regulation loop over the operating tem-
perature range is shown in Fig. 5.22. The absolute value of the error stays below 0.75
degree Celsius over the operating temperature range. In these experiments the CMOS
chip was thermally isolated using a silicone rubber thermal insulation pad.
0 10 20 30 9025
35
45
55
65
75
TIME (S)
TE
MP
ER
AT
UR
E (
C)
o
40 50 60 70 80
DESIRED
TEMPERATURE
REGULATED
TEMPERATURE
Figure 5.21: Temperature regulation cycle with steps at 35, 45, 55 and 65 degree Celsius (in-liquid).
Table 5.2 provides a summary of the experimentally measured characteristics of the
integrated impedance spectroscopy microsystem.
CHAPTER 5. TEMPERATURE REGULATION 165
20 30 40 50 60 70 800.10
0.30
0.60
0.90
TEMPERATURE ( C)o
90
0.15
0.45
0.75
|REL
ATI
VE
ERR
OR
| ( C
)
o
Figure 5.22: Measured absolute value of the relative error of PID regulation loop (in-air).
Table 5.3 provides a comparative analysis of the presented design and existing ther-
mally regulated biochemical sensory microsystems. The design presented in this work
utilizes a mixed signal temperature regulation loop that reuses components from the
DNA-sensing recording channel [137] to perform temperature regulation, thus saving
power and per-channel area as compared to the designs presented in Table 5.3.
5.6 Chapter Summary
A 54-channel, mixed-signal CMOS thermal-controlled, DNA-sensing SoC was devel-
oped. The DNA-sensing SoC shares circuitries to perform impedance spectroscopy,
cyclic voltammetry, and temperature regulation. The on-chip, in-channel heating and
temperature-sensing elements were implemented in standard CMOS without any post-
processing. Using feedback proportional-integral-derivative (PID) control, the temper-
ature can be regulated to within 0.5C of the desired point, a process that enables precise
control of on-chip DNA hybridization during characterization or sensing. Two compu-
tationally intensive operations, the multiplication and subtraction required by the PID
algorithm, are performed by an in-channel, dual-slope multiplying ADC in the mixed-
signal domain, and the result is minimal area and power consumption. The PID regu-
lation loop reuses circuitries from the DNA-sensing channel, and the result is minimal
CHAPTER 5. TEMPERATURE REGULATION 166
area. Each channel occupies an area of only 0.06mm2 and consumes 42µW of power
from a 1.2V supply.
167
Chapter 6
Conclusions and Future Work
6.1 Contributions and Related Publications
This thesis describes a wireless thermally regulated label-free DNA analysis SoC with
nanostructured on-die electrodes. The summary of all contributions by chapter is pre-
sented next.
• Chapter 2
Chapter 2 describes the design, analysis and experimental results of two low-noise
bidirectional current acquisition circuits for interfacing with electrochemical ampero-
metric biosensor arrays. The first design is a switched-capacitor transimpedance am-
plifier (TIA) and the second design is a current conveyer (CC) with regulated-cascode
current mirrors. Both designs were fabricated in 0.13µm CMOS technology. The elec-
trical and electrochemical recording properties of both circuits have been characterized.
It is shown that the current conveyer exhibits superior performance in low-concentration
electrochemical catalytic reporter sensing. This work was presented at the 2011 IEEE
International Symposium on Circuits and Systems [87]. The results from this chapter
are published in IEEE Transactions on Circuits and Systems I [88].
CHAPTER 6. CONCLUSIONS AND FUTURE WORK 168
The main contribution of this chapter is the comparison of the switched-capacitor
transimpedance amplifier to the current conveyer approach for use as a current sens-
ing frontend and providing experimental proofs that the current conveyer (CC) is more
suitable for DNA sensing applications. It is shown through on-chip electrochemical
recording that the CC results in lower charge injection into the working electrode. The
charge injection can disturb the charge balance at the electrode-electrolyte interface
thus affecting the electrochemical reaction. It is shown through extensive electrochem-
ical recordings that the CC is more suitable for DNA sensing from small (2µm×2µm)
working electrodes.
• Chapter 3
Chapter 3 describes the fully integrated 54-channel wireless label-free DNA anal-
ysis SoC configured for cyclic voltammetry and pH sensing. The SoC includes 600
nanostructured DNA sensors and 54 pH sensors, and reuses key circuits for cyclic
voltammetry and pH sensing. The 3mm×3mm prototype fabricated in a 0.13µm stan-
dard CMOS technology has been validated in prostate cancer DNA detection. Each
channel occupies an area of only 0.06mm2 and consumes 42µW of power from a 1.2V
supply. This work was presented in 2012 Symposia on VLSI Technology and Circuits.
The main contribution of this paper is the fabrication of the nanostructured micro-
electrodes (NME) for the first time on a CMOS die. Also, we have characterized the
NME electrodes for the first time for on-CMOS DNA sensing. The detection limit of
10aM is the lowest reported in literature by a CMOS DNA detection system. Another
contribution of this chapter is the design and experimental demonstration of an elec-
trochemical recording channel that achieves the highest sensitivity and dynamic range
compared to designs previously reported in literature and also integration of a fully
digital UWB transmitter. The frontend current conveyer employs chopper stabilization
and dynamic element matching to achieve input-referred noise of less 0.13pA over the
CHAPTER 6. CONCLUSIONS AND FUTURE WORK 169
operating bandwidth and to improve the current conveyer accuracy by 60 percent at the
100nA input current level. The in-channel SRAM allows for in-channel gain calibra-
tion (by adjusting the timing of the ADC) which results in a 17 percent improvement in
channel-to-channel ENOB variation. The UWB transmitter design is adopted from [77]
and the design presented here is optimized for our application by significantly simplify-
ing the previously reported implementations, thus achieving small area and low power
consumption.
• Chapter 4
Chapter 4 presents the fully integrated 54-channel wireless, label-free DNA analysis
SoC configured for impedance spectroscopy. The SoC utilizes frequency response anal-
ysis (FRA) to extract the real and imaginary impedance components of the biosensor.
Two computationally expensive operations, the multiplication and integration required
by the FRA algorithm, are performed by an in-channel dual-slope multiplying ADC
in the mixed-signal domain resulting in high accuracy, small area and low power con-
sumption. Multiplication of the input current by a digital coefficient is implemented
by modulating the counter-controlled duration of the charging phase of the ADC. In-
tegration is implemented by accumulating output digital bits in the ADC counter over
multiple input samples. The 1.2mm×1.6mm prototype fabricated in a 0.13µm standard
CMOS technology has been validated in prostate cancer DNA detection. Each channel
occupies an area of only 0.06mm2 and consumes 42µW of power from a 1.2V supply.
This work was presented in 2012 IEEE Biomedical Circuits and Systems Conference
(BioCAS) [158] and published in IEEE Transactions on Biomedical Circuits and Sys-
tems [144].
The main contribution of this chapter is the design and implementation of a low-
power mixed-signal VLSI approach to implement frequency response algorithm (FRA)
on-chip for DNA impedance extraction. Multiplication, which is the most area-consuming
CHAPTER 6. CONCLUSIONS AND FUTURE WORK 170
and power-intensive operation required by the FRA algorithm is implemented by modi-
fying the timing of the in-channel dual-slope ADC resulting in a low-power, small-area
implementation of the FRA algorithm.
• Chapter 5
Chapter 5 presents the fully integrated 54-channel wireless, label-free DNA anal-
ysis SoC configured for temperature regulation. The on-chip, in-channel heating and
temperature sensing elements were implemented in standard CMOS without any post-
processing using feedback proportional-integral-derivative (PID) control. The tempera-
ture can be regulated to within 0.5C of the desired point, which enables precise control
of on-chip DNA hybridization during characterization or sensing. Two computationally
expensive operations, the multiplication and subtraction required by the PID algorithm,
are performed by the in-channel, dual-slope multiplying ADC in the mixed-signal do-
main, similar to which is described in Chapter 4, and the result is minimal area and
power consumption. The 3mm×3mm prototype fabricated in a 0.13µm standard CMOS
technology has been validated in a portion of the PCR protocol. The PID regulation loop
reuses circuits from the DNA-sensing channel resulting in minimal area. Each channel
occupies an area of only 0.06mm2 and consumes 42µW from a 1.2V supply. This work
was presented at 2012 Symposia on VLSI Technology and Circuits [87].
The main contribution of this chapter is the efficient mixed-signal VLSI implemen-
tation of the PID algorithm for temperature regulation. The multiplication required by
the PID algorithm which is the most area-consuming and power-hungary operation is
efficiently implemented by the in-channel dual-slope ADC. This results is small area
and power consumption and enables a large channel count.
CHAPTER 6. CONCLUSIONS AND FUTURE WORK 171
6.2 Future Work
6.2.1 Power Harvesting and Wireless Communication Chip
An additional 6mm2 chip has been designed and fabricated in 0.13µm CMOS tech-
nology that consists of a wireless inductive power receiver, a command data receiver
and two high-speed transmitters. The chip was co-designed with Karim Abdelhalim, in
order to provide wireless power transfer and wireless data command transfer capabili-
ties to the DNA analysis SoC. The fabricated prototype die micrograph is depicted in
Fig. 6.1.
The power harvesting unit of the chip consists of a low-drop-out rectifier capable
of operating at up to 10MHz, two low-dropout regulators providing 2.5V and 1.2V
supply voltages with a maximum load current of 10mA and an on-chip programable
DAC generating up to 8 bias voltages. A 4MHz amplitude shift keying (ASK)/ on-off
keying (OOK) wireless data receiver was implemented to enable wireless transmission
of configuration and command data to the DNA analysis SoC. Finally, a 915MHz PLL-
based FSK transmitter and a UWB transmitter were also included in the CMOS chip.
This chip will be integrated with the presented SoC in the near future.
6.2.2 Improvements to the SoC
The presented SoC prototype contains a large number of input/output (I/O) and debug
pins and also requires some off-chip current and voltage bias sources. In the next gen-
eration of the SoC the bias voltages and current sources should be implemented on the
chip using on-chip band-gap voltage and current sources. Also, the I/O and debug pins
number needs to be reduced. One approach to reduce the debug pins number is to mux
all the analog test points into a single node and the same should be done for the digital
I/O pins resulting in only one or few analog and one digital I/O pins.
CHAPTER 6. CONCLUSIONS AND FUTURE WORK 172
915MHz OOK TX
3.1-10.6GHz
UWB TX
10MHz
COMMAND
RX
4MHz
POWER
RX
Figure 6.1: Die micrograph of the 2mm×3mm power harvesting and wireless communicationchip. The chip was fabricated in a 0.13µm standard CMOS technology.
Analog improvements to the SoC design include implementation of a differential
front end current acquisition circuitry to reduce the second harmonic interference and
common-mode noise. Another improvement is reduction of flicker noise in the front-
end current acquisition circuits.
Testing improvements include design of a better and more durable packaging for
the CMOS chips. Currently a fill-and-dam technique is used to protect the chip bond-
wires from the electrolyte solution under test. This technique is only protecting the
chip for a limited time and after roughly one hour of testing a chip may become non-
functional. There is a high chance that the pins are shorted due to leakage. Another
improvement to the testing is integration and miniaturization of the Ag/AgCl electrode
on the CMOS chip. Currently a somewhat bulky off-chip Ag/AgCl is used for all the
testing. A number of miniaturized Ag/AgCl electrode fabrication methods have been
reported in literature [78, 170]. These reports can be used as a starting point for full
integration of the reference electrode on the CMOS chip.
173
References
[1] B. Albert, A. Johonson, J. Lewis, M. Raff, K. Robert and P. Walter, Molecular Biology of the cell,
5th ed. New York, NY: Garland Sceince, 2008.
[2] M. Schena, Mircorarray Analyis. hoboken, NJ: Wiley, 2003.
[3] B. A. Flusberg, D. R. Webster, J. H Lee, K. J. Travers, E. C. Olivares, T. A. Clark, J. Korlach,
and S. W. Turner, “Direct Detection of DNA Methylation During Single-Molecule, Real-Time
Sequencing,” Nature Methods, vol. 7, pp. 461−465, 2010.
[4] G. Liu, Y. Wan, V. Gau, J. Zhang, L. Wang, S. Song and C. Fan, “An Enzyme-Based E-DNA Sensor
for Sequence-Specific Detection of Femtomolar DNA Targets,” Journal of American Chemical
Society, vol. 130, no. 21, pp. 6820−6825, 2008.
[5] B. Malorny, C. Bunge, R. Helmuth, “A Real-Time PCR for the Detection of Salmonella Enteritidis
in Poultry Meat and Consumption Eggs,” Journal of Microbiological Methods, vol. 70, no. 2,
pp. 245−251, 2007.
[6] A. Nouri, and C. F. Chyba, “DNA Synthesis Security,” Methods in Molecular Biology, Vol. 852,
Part. 4, pp. 285−296, 2012.
[7] N. A. Straus, T. I. Bonner, “Temperature dependence of RNA-DNA hybridization kinetics,”
Biochimica et Biophysica Acta (BBA) - Nucleic Acids and Protein Synthesis., Vol. 277, pp. 87−95,
1972.
[8] A. Hassibi, H. Vikalo and A. Hajimiri, “On Noise Processes and Limits of Performance in Biosen-
sors,” Journal of Applied Physics., 102, 2007.
[9] N. Levine, Quantum Chemistry, 5th ed. Prentice-Hall, New York, 1999 .
[10] N. G. Van Kampen, Stochastic Processes in Physics and Chemistry North-Holland, Amsterdam,
1981 .
REFERENCES 174
[11] J. I. Steinield, J. S. Fransisco, and W. L. Hase, Chemical Kinetics and Dynamics, 2nd ed. Prentice-
Hall, New York, 1998 .
[12] J. Hubble, “Monte Carlo simulation of biospecific interactions,” Biotechnology Letters, Volume
22, Issue 18, pp. 1483−1486, 2000.
[13] K. Linneti and M. Kondratovich, “Partly nonparametric approach for determining the limit of
detection,” Clin Chem. Volume 50, Issue 4, pp. 732−740, 2004.
[14] P. M. Levine and G. W. Roberts, “High-resolution flash time-to-digital conversion and calibration
for system-on-chip testing,” Embedded Microelectronic Systems: Status and Trends, B. M. Al-
Hashimi, Ed. London: IEE Press, 2006.
[15] S. Cagnin, M. Caraballo, C. Guiducci, P. Martini, M. Ross, M. S. Ana, D. Danley, T. West, and
G. Lanfranchi, “Overview of Electrochemical DNA Biosensors: New Approaches to Detect the
Expression of Life,” Sensors, vol. 9, no. 4, pp. 3122−3148, 2009.
[16] O. A. Loaiza, S. Campuzano, M. Pedrero, M. I. Pividori, P. Garci, and J. M. Pingarro, “Disposable
Magnetic DNA Sensors for the Determination at the Attomolar Level of a Specific Enterobacteri-
aceae Family Gene,” Anal. Chem., vol. 80, no. 21, pp. 8239−8245, 2008.
[17] C. Stagni, C. Guiducci, L. Benini, B. Ricc, S. Carrara, B. Samor, C. Paulus, M. Schienle, M.
Augustyniak, and R. Thewes, “CMOS DNA sensor array with integrated A/D conversion based on
label-free capacitance measurement, IEEE J. Solid-State Circuits, vol. 41, no. 12, pp. 29562964,
Dec. 2006.
[18] X. C. Zaho, L. Q. Huang, S. F. Li, “Microgravimetric DNA Sensor Based on Quartz Crystal
Microbalance: Comparison of Oligonucleotide Immobilization Methods and the Application in
Genetic Diagnosis,” Biosens Bioelectron, vol. 16, no.2, pp. 85−95, 2001.
[19] A. Sassolas, B. D. Leca-Bouvier, and L. J. Blum, “DNA Biosensors and Microarrays,” Chem. Rev.,
Vol. 108, pp. 109−139, 2008.
[20] F. Heer, M. Keller, G. Yu, J. Janata, M. Josowicz, and A. Hierlemann, “CMOS Electro-Chemical