Lab on a Chip - College of Engineering, Purdue University

Lab on a Chip
View Article Online
Published on 26 March 2014. Downloaded by Purdue University on 24/07/2014 18:31:38.
Cite this: Lab Chip, 2014, 14, 2469
View Journal | View Issue
Non-faradaic impedance characterization of an
evaporating droplet for microfluidic and
biosensing applications†
Piyush Dak,ab Aida Ebrahimiab and Muhammad A. Alam*ab
We have developed a general numerical/analytical theory of non-faradaic impedance of an evaporating
droplet, and validated the model by experiments involving droplets of various analyte concentrations
deposited on a surface defined by coplanar electrodes. The impedance of the droplet Z(n0,t,f) is analyzed
Received 13th February 2014,
Accepted 26th March 2014
DOI: 10.1039/c4lc00193a
as a function of the concentration (n0) of the ions in the solution, the measurement frequency (f ) and the
evaporation time (t). We illustrate the versatility of the model by determining the sensitivity enhancement
α(t) of the droplet-based impedimetric nano-biosensor under different regimes of operation. The model
should have broad applications in the characterization/optimization of droplet-based systems, especially
lab-on-chip components involving digital microfluidics.
Droplets occur in a broad range of natural and engineered
systems. In natural systems, for example, a drop of water on
a lotus leaf forms a spherical shape to minimize the surface
energy.1 When a drop of liquid with suspended particles
dries on a substrate, it leaves a ring-shaped stain on the
surface generally known as the “coffee-ring effect”.2–4 On the
other hand, in engineered systems, micro/nano-liter sized
droplets have been used in a broad range of applications
including drop-on-demand inkjet printing,5 molecular transport,6 single-cell analysis and sorting7 through microfluidic
channels, electrically-addressable biochemical reactions in
sub-nanoliter droplets,8 etc. Evaporating droplets have also
found a number of interesting applications. Jing et al. have
used tiny evaporating droplets to elongate and fix DNA molecules on derivatized surfaces;9 De Angelis et al. have reported
attomolar-detection of DNA concentration by concentrating a
few copies of DNA onto a localized SPR sensor by the evaporation of a droplet;10 and most recently Ebrahimi et al. have
reported a label-free on-chip non-faradaic impedance-based
detection of attomolar (aM) concentrations of DNA.11
The concentration of biomolecules was enhanced through
School of Electrical and Computer Engineering, Purdue University, West Lafayette,
IN 47907, USA. E-mail: [email protected]
Birck Nanotechnology Center, Purdue University, West Lafayette, IN 47907, USA
† Electronic supplementary information (ESI) available. The code for the
droplet impedance model can be downloaded from https://engineering.purdue.
edu/∼alamgrp/dak_files/ See DOI: 10.1039/c4lc00193a
This journal is © The Royal Society of Chemistry 2014
evaporation of the droplet and an enhanced signal was
obtained for even a few copies of DNA in micro-liter sized
Optical techniques such as high-speed imaging,12 confocal
microscopy13 and laser light scattering14,15 have been used to
characterize the geometry and composition of droplets. For
probing the dynamics of a droplet on a surface, an electrical
characterization technique such as impedance spectroscopy
can provide complementary information. In this regard, it is
desirable to have a theoretical model which can map system
parameters such as the droplet composition, shape and size
to an electrical signal (i.e. impedance) as the droplet evaporates. Faradaic impedimetric sensors16 have long been used
for highly selective detection of biomolecules. If the analyte
is known and only its concentration is desired, non-faradaic
impedance spectroscopy (NFIS) provides a simple nonintrusive way to obtain wealth of information regarding the
composition of the droplet and the kinetics of evaporation.
Important initial work on NFIS has already been reported.
For example, Sadeghi et al. performed on-chip impedancebased droplet characterization for a parallel plate electrode
system.17 For a broader range of applications, however, all
droplet models must be generalized to include the accumulation of ionic charges (double layer) near the electrode surface, the arbitrary geometry of electrodes, the time dynamics
and droplet shape dependence of impedance components,
including all the parasitic components.
In this paper, we formulate a comprehensive theory
for droplet impedance with a focus on nano-biosensing.9–11
We solve for the time dynamics of droplet evaporation and
relate the composition, size and shape of the droplet to the
Lab Chip, 2014, 14, 2469–2479 | 2469
View Article Online
Published on 26 March 2014. Downloaded by Purdue University on 24/07/2014 18:31:38.
time-varying impedance. We demonstrate that the approach
can be used to optimize the sensor design and to operate the
sensor in the optimal frequency range. Indeed, the model
is general and can be used in a broad range of microfluidic
The paper is arranged as follows. In section 1, we describe
the device structure and operation principle of the dropletbased sensor. In section 2.1, we describe the impedance/
admittance response of the system for a fixed droplet geometry.
In section 2.2, we describe the time dynamics of droplet evaporation and describe the geometry variation as a function of
time. In section 2.3, we provide the time dependence of
circuit components/impedance for the system. In section 3.1
and 3.2, we explain the sensitivity enhancement of the
droplet-based sensor in various operation regimes and
discuss the implications of parasitic impedance, respectively.
Finally, the model is validated with experiments on droplets
containing DNA molecules in section 3.3.
1. Device structure and principle
of operation
As a model system for the theoretical framework, we consider
an evaporating droplet containing chemical/biomolecules
resting on a substrate with co-planar electrodes, as shown in
Fig. 1(a) and (b). We assume that the surface is designed in
such a way that the droplet is pinned and maintains a constant contact line as it evaporates.11 The contact width (r)
Lab on a Chip
and the contact angle (θ) that the droplet makes with the
surface depends on the surface wettability and the droplet
volume. The electrical impedance of the droplet is measured
by applying a small ac signal (with a dc bias) between the
electrodes. The impedance of the droplet, Zdrop(n0, f,t),
depends on the time-dependent (t) shape of the droplet, the
initial concentration of ions (n0), and the characterization
frequency ( f ). As the droplet evaporates, Zdrop changes due to
two distinct but correlated effects: the increase in ionic concentration associated with a decrease in the droplet volume,
and the change of the droplet geometry due to evaporation.
The changes in Zdrop can be used as a characterization tool
for many droplet-based problems and applications discussed
earlier. For droplet-based nanobiosensors, the positive implications are obvious (see Fig. 1(b)): the shrinking droplet
brings the analyte biomolecules close to the sensor surface
faster than the diffusion limit.18 As a result, the concentration of the biomolecules increases inversely with the volume
of the droplet, and this increased concentration is reflected
in enhanced sensitivity,19 S(t), defined as change in conductance (ΔY(t)) with respect to a known reference solution
(DI water).
For simplicity, we assume that the droplet is self-aligned
to the coplanar electrodes, as shown in Fig. 1(b). The conclusions of the paper, however, are general and would apply to
any electrode geometry. The electrodes are multi-functional:
they define the superhydrophobic surface that pins the droplet and can also be used as an addressable heater. If the
Fig. 1 (a) Model system for numerical/analytical modeling; (b) evaporation dynamics of droplet: as the droplet evaporates, the contact angle (θ)
decreases while the contact line remains pinned. The concentration of the chemical/biomolecules ( ρ) increases as the volume (V) decreases with
time (t) with the number of chemical/biomolecules (N) remaining constant; (c) equivalent circuit representation of the system.
2470 | Lab Chip, 2014, 14, 2469–2479
This journal is © The Royal Society of Chemistry 2014
View Article Online
Published on 26 March 2014. Downloaded by Purdue University on 24/07/2014 18:31:38.
Lab on a Chip
electrodes are simultaneously used as a heater and a prober,
a complex interaction is likely. Therefore, for simplicity of
model development, we use the electrodes exclusively for
impedance measurement, and the heating effects are not
considered. The applied voltage is presumed small to suppress the faradaic current.20 However, if a higher applied
voltage is necessary, the electrodes may be coated with a thin
dielectric layer to block any charge transfer between the
electrodes and the solution (refer to ESI† Section 3 for
Finally, the substrate offers a parasitic path for signal to
travel between the electrodes (see Fig. 1(b)) and thereby
defines the upper limit of the frequency of operation. At
sufficiently high frequencies, the impedance of the overall
system, Znet, is dictated by the parasitic impedance, Zpar, and
becomes insensitive to the properties of the droplet itself.
Depending on the substrate (e.g. glass vs. silicon-on-insulator,
SOI), the parasitic impedance may change by orders of magnitude; therefore, the choice of the substrate is important in
defining the sensitivity of the sensor.
we call the geometry factor. This factor depends on the width
(W), separation (L) between the electrodes and droplet
contact angle θ and contact width r. Hz represents the length
of the cylindrical segment as shown in Fig. 1(a). The conductivity σ can be related to the ionic concentration (ni) by
σ = qni(μp + μn), where μp and μn are the ionic mobilities of
the positive and negative ions, respectively.
Similarly, the geometry capacitance, which depends on
the permittivity of the solution (ε) and the same geometry
factor, g, as in Rseries (refer to ESI† Section 1), can be written
as Cgeo = Hzε/g. Henceforth, unless explicitly specified, we
assume that the analyte concentration is small so that ε ≈
εfluid and is unaffected by the salt/analyte concentration.
Finally, Cdl originates from the adsorbed charge layer and
2. Numerical/compact modeling of
droplet impedance
2q 2 ni
 qV 
cosh  e  , where A = rHz
 2kT 
is the area of the electrode in contact with the solution, Ve is
the applied bias on the electrode, q is the electronic charge,
k is the Boltzman constant, and T is the temperature of the
solution. While the applicability of this analytical formula
is well established for bulk solutions,22 we show through
detailed numerical simulations (refer to ESI† Section 1(b))
that this can also be applied to micro-liter sized droplets. For
medium to low ionic concentrations (<100 mM), the Debye
length is much larger than the thickness of the Stern layer
(~0.4 nm)24 so that Cdiff ≪ CS and hence, Cdl ≈ Cdiff.
Once the droplet/electrode geometries are specified, the
fluid properties are given (e.g. εfluid), and the salt (n0)/analyte
concentrations ( ρ) are known, Zdrop is fully determined, and
can be plotted, among other variables, as a function of
frequency f.
The frequency response of an ideal system (with no parasitic losses) can be divided into three distinct regions (see
Fig. 2(a)), such that the impedance components Rseries, Cdl
and Cgeo are dominant in one of the three regions. For
f  f low 
, Cdl dominates the net impedance, for
Let us first consider the frequency dependence of impedance
of a droplet (see Fig. 1(a)) with constant contact angle θ
resting on a substrate with two planar electrodes. For an arbitrary electrode (faradaic/non-faradaic), the different components which can affect the impedance are shown in Fig. 1(c).
Here, Rct denotes the charge transfer resistance,20 Zw the
Warburg impedance,21 Cdl double layer capacitance, Rseries
denotes resistance of the solution and Cgeo the dielectric
(geometric) capacitance of the droplet. The net impedance of
the system is therefore given by
 1 
where Z dl   Rct  Z w  || 
 represents the double layer
 jCdl 
impedance and Zpar the parasitic impedance. For a nonfaradaic electrode, there is no charge transfer at the surface,
so that Rct → ∞ and hence the net impedance of the system
simplifies to
2   1 
Z net   Rseries 
 ||  Z par 
 || 
jCdl   jCgeo 
The rest of the paper will focus on this reduced ‘nonfaradaic’ model, with the understanding that it can be easily
generalized to include faradaic contributions as well.
Physically, Rseries originates from the finite conductivity of
the solution, σ, as Rseries = g/σHz where g = g(θ,r,W,L), which
This journal is © The Royal Society of Chemistry 2014
where CS is the Stern capacitance23 and Cdiff is the differential capacitance. For electrode separation/droplet dimensions
much larger than the Debye length (λd ~ 1 μm for pure
water), the diffuse layer capacitance can be described by the
analytical formula Cdiff  A
2.1. Frequency response of the droplet impedance
 1 
Z net   Rseries  2 Z dl  || 
|| Z ,
 jC   par 
geo 
 1
1 
diffuse layer charge22 and can be written as Cdl  
diff 
 S
f low  f  f high 
, Rseries is the dominant compo2RseriesCgeo
nent, and finally for f > fhigh, Cgeo dominates. The numerical/
analytical estimation of different circuit components and cutoff frequencies is described in ESI† Sections 1 and 2, respectively. For a conductivity-based sensor, we should be operating in either regime I or II, while detection can be performed
in regime III if the change in permittivity of the solution
upon the addition of biomolecules is considerable.
The admittance of the droplet (see Fig. 2(b)) is defined as
Ydrop = 1/Zdrop. We can define the limit of detection as the
Lab Chip, 2014, 14, 2469–2479 | 2471
View Article Online
Published on 26 March 2014. Downloaded by Purdue University on 24/07/2014 18:31:38.
Lab on a Chip
Fig. 2 (a) Impedance of the droplet as a function of frequency. Cdl dominates at f < flow, Rseries dominates for flow < f < fhigh and Cgeo dominates
the impedance at very high frequency (f > fhigh). A similar trend (b) is visible in the admittance vs. frequency response.
minimum measurable change in conductance ΔYdrop of the
droplet upon the introduction of salt/biomolecules.
In order to improve the limit of detection, several design
parameters can be considered, i.e. electrode separation (L),
electrode width (W), electrode length (Hz) in contact with
droplet. These factors have been considered by Hong et al.,
albeit for a bulk solution. The longer the electrode length and
the smaller the electrode spacing, the better the sensitivity.25
However, for ultra-low concentrations of biomolecules, the
diffusion of the ions limits the detection time. Therefore, in
order to improve the sensitivity and response time of the system, we need to explore droplet volume (V) (or contact angle
(θ)) as an additional design parameter. This can be achieved
by evaporation of the droplet, which is considered next.
2.2. Dynamics of droplet evaporation
A droplet forms as a result of the balance of surface tensions at
the triple contact line between air, liquid and the surrounding
medium. Equivalently, the shape of the droplet can be determined by energy minimization.26 Our earlier work11 showed
that a droplet placed on nanotextured-superhydrophobic
electrodes assumes a nearly ellipsoidal shape with pinned
contact lines at the edges of the droplet. Contact line pinning
of the droplet is critical for highly stable impedance characterization. A constant contact width of the evaporating droplet is also obtained using a chemically heterogeneous striped
In order to determine the time evolution of impedance of
such a pinned droplet, we must first determine the evolution
of droplet shape with time. Numerical calculations show and
high-speed images confirm that as the droplet evaporates, it
maintains the shape of an elongated ellipsoid, defined by a
constant contact width r and decreasing perpendicular (θ⊥)
and parallel (θ∥) contact angles.11 For analytical simplicity, we
approximate the elongated ellipsoid as a truncated cylinder
with contact width r and contact angle θ, while keeping all
other constraints (e.g. initial volume) unaltered, see Fig. 1(b).
Our model is directly applicable in scenarios where the
2472 | Lab Chip, 2014, 14, 2469–2479
elongation of the droplet parallel to the coplanar electrodes
is large as compared to that in a direction perpendicular to
the electrodes. However, the ‘cylindrical’ approximation is
not restrictive – the formulation is general and can be applied
to any system where the evolution of droplet shape (i.e. the
geometry factor, g(t) and droplet volume (V(t)) is known
through numerical simulation28 or high-speed imaging.12
Similar to Rowan et al.29 and Birdi et al.,30 we consider
droplet evaporation as a gas diffusion process and assume
that the rate of mass loss from the droplet is given by
m   J dS where J is the diffusion flux of liquid mole
cules away from the surface and the integral of the flux is
taken over the surface (Ω) of the droplet. The diffusion flux
can be written in terms of the concentration of liquid vapors
c(r,θ,z) as J   Dc where D is the diffusion coefficient of
liquid vapors in the ambient surroundings. Therefore, the
rate of mass loss would be m    DcdS . In order to evalu
ate this integral, we use the equivalence between the electric
potential (ψ) and vapor concentration (c), as discussed in
ESI† Section 4. For an electrical system, we can write the
charge as Q    dS  Ce  s     where Ce is the electrical
capacitance. Similarly, the diffusion flux of molecules can be
written as31 Φm = CD(cs − c∞) where cs is the saturation vapor
density of the liquid and c∞ is the vapor density of liquid far
away from the surface. CD is the diffusion equivalent capacitance of a truncated cylinder with finite length32 which is
given by,
CD 
2DH z  0.3069 0.2717  
1 
 
2  
H 
where   log  z  and Rs is the radius of curvature of the
 Rs 
droplet. Note that the diffusion equivalent capacitance of the
cylinder with finite length has been appropriately scaled for
the reduced surface area of the truncated cylinder. If we
This journal is © The Royal Society of Chemistry 2014
View Article Online
Lab on a Chip
volume evolution of the droplet can often be described by
a power-law33
V  t   V0 1  
 
Published on 26 March 2014. Downloaded by Purdue University on 24/07/2014 18:31:38.
where τ and n are empirical parameters defined by the shape
of the droplet and the mode of evaporation. For the experimental data obtained from Ebrahimi et al.,11 we find that the
parameter n = 3/2 and τ = 20 min.
2.3. Time evolution of impedance/conductance
The net impedance of the droplet is given by
Fig. 3 Evolution of droplet contact angle (θ) (right) and droplet
volume (V) (left) as a function of time. Symbols are the experimental
data obtained from Ebrahimi et al.11 The variation of droplet
volume as a function of time can analytically be approximated as
V  V0 1  
 
with n = 3/2 where V0 is the initial volume of
the droplet and τ the total evaporation time. Simulation parameter:
(cs − c∞)/cs = 0.88.
assume that the density of the liquid is constant as the droplet evaporates, the rate of mass loss can be expressed as,
m  
 P
where P is the density of the liquid, m is the mass of the
droplet, V is the volume of liquid for a given contact angle
and t is time.
 CD  cs  c 
D  cs  c   sin    0.3069 0.2717  1
1 
2 
Pr 2
1   cot  
  f  , r , H z 
D  cs  c 
captures the material parameters of the
droplet. This equation is numerically integrated to obtain θ(t)
and V(t). Fig. 3 shows the evolution of droplet contact angle
(θ) and volume (V) as a function of time (t). Simulation
parameters are listed in ESI† Tables 2 and 3. Interestingly,
despite the complexity of the equation, one finds that the
This journal is © The Royal Society of Chemistry 2014
Given the geometry dependence of the circuit components
and the time dependence of the geometry, we can determine
the time dependence of different circuit components as
1. Series resistance/conductance. Variation in series
resistance due to evaporation arises from two distinct effects.
First, the geometry factor g(r,θ) evolves with θ(t), so that
g(t) = g(θ(t)), see Fig. 3 and SFig. 1(b), ESI.† Second, the
concentration of ions in solution increases inversely with the
volume of the evaporating droplet, V(t). If the electrolyte is
fully ionized, we can assume that conductivity is directly
proportional to the ionic concentration. Therefore, the
conductivity σ(t) = σ0V0/V(t) increases as a function of time. At
any time, series resistance is given by,
Rseries 
For simplicity, we assume that the evaporation occurs at a
constant temperature so that cs is independent of time. Also,
the equation assumes that the evaporation from the front
and back surfaces of the cylinder are negligible, which is
justified as long as Hz ≫ r. Once we relate V and Rs to the
contact angle θ (see ESI† Table 4), the rate of change of
contact angle as a function of θ is given by,
where  
 
Z drop  , t    Rseries  t  
 || 
jCdl  t    jCgeo  t  
g t  V t 
g t 
g  t V  t 
 R0
g 0 V0
 t  H z
 0 H zV0
where R0 represents the resistance of the solution at time
t = 0 and g0 = g(t = 0). Here, V0 and σ0 are the initial volume
and conductivity of the droplet, respectively. Fig. 4(a) shows
the evolution of Rseries and Gseries  Rseries−1 as a function
of time.
2. Double layer capacitance. The increased concentration
of the evaporating droplet is also reflected in CDL, as follows:
since the concentration at any time t is given by ni(t) = n0V0/V(t),
the double layer capacitance would be,
CDL  t   A
2q 2 ni  t 
 qV 
cosh  e   CDL, 0
V t 
 2kT 
where CDL,0 is the double layer capacitance at t = 0. Fig. 4(b)
shows the evolution of the double layer capacitance as a function of time.
3. Geometry capacitance. The geometry (dielectric) capacitance
Cgeo  H z
g t 
is independent of the ion concentration
(except indirectly through the permittivity of the solution, ε),
Lab Chip, 2014, 14, 2469–2479 | 2473
View Article Online
Published on 26 March 2014. Downloaded by Purdue University on 24/07/2014 18:31:38.
Lab on a Chip
Fig. 4 (a) Time dependence of series resistance (left) and series conductance (right); (b) time evolution of double layer capacitance (left) and
geometry capacitance (right) for n0 = 10 μM and V0 = 3 μL.
but depends on the geometry of the droplet through g(t). The
variation of Cgeo as a function of time is shown in Fig. 4(b).
Our numerical simulations show that both g(θ) and θ(t) are
monotonically decreasing functions of θ (SFig. 1(b), ESI†) and
t (Fig. 3) respectively; therefore, g(t) increases monotonically
with time t. Therefore, the geometry capacitance decreases
with time, unlike Gseries and CDL.
To summarize, the impedance evolution is specified by
two parameters, g(r,θ(t)) and V(t)/V0. Once these two parameters are known either from experiments, or detailed numerical models such as surface evolver;28 or by approximate
analytical/numerical models discussed above, one can compute any electrical characteristics associated with evaporating
droplets. In the next section, we will illustrate the concept by
analyzing a droplet-based sensor.
3. Application of the model to a
droplet-based sensor
3.1 Frequency-dependent time response of biosensors
Our earlier work showed that a droplet-based sensor is more
sensitive compared to sensors based on bulk liquids.11 In
order to determine the relative improvement in sensitivity,
we define the sensitivity of the droplet-based sensor as the
normalized change in admittance of a droplet containing
analyte (Yρ) with respect to a reference solution i.e. DI water
(YDI). Therefore,
S t  
Y  t   YDI  t 
YDI  t  0 
  t  
Y  t  0   YDI  t  0 
YDI  t  0 
chemical/biomolecule (Cdl,ρ ) with respect to a reference
solution (DI water) (Cdl,DI). Using eqn (11) with Y(t) = jωCdl(t),
the sensitivity is given by
S t  ~
2474 | Lab Chip, 2014, 14, 2469–2479
Cdl,DI  t  0 
Cdl,DI  t  0 
The amplification in sensitivity relative to time t = 0 is
obtained by inserting eqn (10) in eqn (12) i.e.
 t  
V t 
t 2
1   
where we have used the empirical approximation of V(t)/V0
from eqn (7). Note that the amplification factor is independent of the contact angle of the droplet at any time. Fig. 5(a)
shows the sensitivity and amplification factor for very low frequency mode of operation of a sensor with initial ion concentration n0 = 10 μM.
b) Intermediate frequency operation. This regime of
operation occurs when flow(t) ≪ f ≪ fhigh(t), and therefore
Y(t) ~ Gseries(t). In this regime, the capacitive response of the
ions is no longer relevant and the in-phase response of the
ions with respect to the applied signal dictates the net
The sensitivity S(t) in this regime of operation can be
defined in terms of the conductance change upon addition of
chemical/biomolecule (Gρ) with reference to DI water (GDI), i.e.
S t  
a) Low frequency operation. In this range of frequency,
f ≪ flow(t) for all 0 < t < τ, the double layer capacitance is the
dominant component i.e. Y(t) ~ jωCdl(t). As the droplet
shrinks and the concentration increases, the reduction in the
double layer thickness is reflected in increasing Cdl. The
sensitivity of this mode of operation can be defined as the
change in the double layer capacitance upon addition of
Cdl,  t   Cdl,DI  t 
G  t   GDI  t 
GDI  t  0 
  t 
G0  t  0 
GDI  t  0 
where ΔG0 = Gρ(t = 0) − GDI(t = 0) and the amplification factor,
 t  
 0 ~
g t  V t  
1  
 
relates the sensitivity enhancement obtained as a function of
This journal is © The Royal Society of Chemistry 2014
View Article Online
Published on 26 March 2014. Downloaded by Purdue University on 24/07/2014 18:31:38.
Lab on a Chip
Fig. 5 Sensitivity as a function of evaporation time for (a) low frequency operation, (b) intermediate frequency operation, (c) high frequency
operation; (d) evolution of cut-off frequencies as a function of time for n0 = 10 μM and V0 = 3 μL.
time. Note that even though g(t) is monotonically increasing
as a function of time, the net amplification factor (α(t)) still
increases due to a considerable reduction in droplet volume
V(t). This equation suggests that a very high sensitivity can be
achieved if we operate the sensor in a frequency regime where
Gseries is dominant. Fig. 5(b) shows the sensitivity and amplification factor for intermediate frequency mode of operation of
a sensor with initial ion concentration n0 = 10 μM.
c) High frequency operation. This regime occurs when
f ≫ fhigh(t), so that Y(t) ~ jωCgeo(t). Again the sensitivity of the
system can be defined as
S t  
Cgeo,  t   Cgeo,DI  t 
Cgeo,DI  t  0 
  t 
Cgeo,  t  0 
Cgeo,DI  t  0 
where Cgeo,ρ and Cgeo,DI are respectively the geometry
capacitances for the droplet with chemical/biomolecules and
the reference solution (DI water). The amplification in
sensitivity relative to time t = 0, is given by
 t  
g t 
Since g(t) is a monotonically increasing function of time,
the sensitivity in this regime of operation decreases with time
i.e. α(t) ≤ 1. Fig. 5(c) shows the evolution of sensitivity (S(t))
and amplification factor (α(t)) as a function of time. It is
This journal is © The Royal Society of Chemistry 2014
assumed that the permittivity change of the solution upon
addition of chemical/biomolecules is 10%.
Furthermore, for a conductance-based sensor (with negligible change in solution permittivity), ΔCgeo,ρ = 0 and hence
this regime is unsuitable for biomolecule detection. However,
if one is interested in characterizing the time-dependent evolution of the geometry of the droplet (e.g. shape or volume),
this frequency regime is ideally suited, since the impedance
is independent of salt/analyte concentration and depends
exclusively on droplet geometry.
In general, as the droplet evaporates, the relative importance of a particular circuit component changes as well. This
is because the cut-off frequencies, f low  t  
and f high  t  
2Rseries  t  Cdl  t 
, themselves evolve with time;
2Rseries  t  Cgeo  t 
as the boundaries of the frequency band shift, the circuit
may become more resistive/capacitive at a given frequency of
operation. Fig. 5(d) shows the evolution of lower and upper
cutoff frequencies for a droplet containing salt solution with
initial concentration n0 = 10 μM. Given the time and frequency dependence as discussed in section 2.3, one can
determine the frequency of operation for which ΔY(t) is maximum for a given set of parameters, such as mobility of ions
(μ) and applied bias (Ve). For example, in the case that μ of
ions is large, it would be preferable to operate the sensor in a
Lab Chip, 2014, 14, 2469–2479 | 2475
View Article Online
Lab on a Chip
Published on 26 March 2014. Downloaded by Purdue University on 24/07/2014 18:31:38.
Fig. 6 Simulation of parasitic capacitance for two different substrates. Geometry used for the simulation for glass substrate (a) and SOI substrate
(d). Variation of parasitic capacitance as a function of (b), (e) electrode separation and (c), (f) electrode width for glass and SOI substrate respectively.
resistive regime for optimal sensitivity. For such an operation, a frequency choice, foptimal, such that
10 max  f low  t    f optimal 
min  f high  t  
would be appropriate, since this will ensure that the resistive
component at any time is at least 10 times larger (dominantly
resistive) than the capacitive component. When the applied
bias is large, so that double layer capacitance is significant,
a frequency of operation foptimal ≤ 1/10 min ( flow(t)) would
ascertain the operation in dominantly capacitive regime.
However, a very large applied bias may not be desirable
because it would yield unreasonably low frequencies for
capacitive operation and lead to faradaic currents20 that can
2476 | Lab Chip, 2014, 14, 2469–2479
contaminate the results of impedance spectroscopy. For a
more realistic case, when substrate parasitic capacitance must
be accounted for (discussed in the next section), the upper cutoff frequency is given by f high  t  
2Rseries  t   Cpar  Cgeo  t  
This implies that the upper cut-off frequency can be substantially lower if the parasitic capacitance (Cpar) is large. If
10 max ( flow) ≥ 1/10 min ( fhigh), it is impossible to operate
the sensor in a dominantly resistive regime and the parasitic
capacitance must be suppressed to increase fhigh.
3.2 Implications of parasitic impedance of the substrate, Zpar
So far, we have focused exclusively on Zdrop, assuming that
the parasitic capacitance/resistance of the substrate is negligible.
This journal is © The Royal Society of Chemistry 2014
View Article Online
Published on 26 March 2014. Downloaded by Purdue University on 24/07/2014 18:31:38.
Lab on a Chip
However, in real systems the parasitic capacitance can be a
major limitation to the sensitivity of the device and must be
accounted for. Parasitic capacitance dominates at intermediate to high frequencies and can limit the frequency range of
operation of the sensor. It can either be obtained from experiments with droplet-free measurements11 or through detailed
numerical simulation/analytical modeling. Here, we consider
numerical/analytical evaluation of parasitic capacitance for
two different substrates which are commonly used for
impedance-based sensors:
1. Glass as the sensor substrate. Due to its low dielectric
constant, glass is an ideal candidate for use as a substrate for
the sensor. The parasitic capacitance is estimated by
numerical simulation of the structure shown in Fig. 6(a)
using Sentaurus, an advanced multidimensional device
simulator.34 A bias Vdc is applied between the electrodes and
the Laplace equation (∇·(εglass∇ϕ) = 0) is solved to determine
the potential, ϕ and electric field, E inside the substrate.
Charge Qdc is estimated on the positive electrode by the
Gauss Law i.e. Qdc   DdS    glass Ed S . The capacitance of
the substrate is then given by Cpar = Qdc/Vdc. An analytical
expression for the capacitance of coplanar electrodes was
derived by Wei,35 i.e.
Cpar 
 glass H E
2 K  k  /K
1 k2
where K(k) is the complete elliptical integral of the first kind
with k = L/(L + W), εglass denotes the permittivity of the glass
substrate and HE is the electrode length.
Fig. 6(b) and (c) show the simulation (numerical/analytical)
of the parasitic capacitance for different electrode separations
(width = 900 μm) and for different electrode widths (separation = 20 μm). The capacitance depends weakly on the
electrode separation and width. Numerical simulation is in
good agreement with the analytical expression. The marginal
difference in simulation and analytical estimate comes from
the fact that Wei derived eqn (18) by neglecting the fringing
fields in the transformed coordinate system.
2. SOI as the sensor substrate. Silicon-on-insulator (SOI) is
a popular substrate in the semiconductor industry because it
minimizes leakage currents, radiation-induced photocurrents, latch-up effects, etc.36 in comparison to conventional
bulk substrates. However, the same leads to huge parasitic
losses for impedance sensors, as the electrodes can couple to
the doped silicon below the top oxide layer which leads to a
large parasitic capacitance. Fig. 6(e) and (f) show the numerical simulation results for parasitic capacitance of an SOI
substrate (Fig. 6(d)) for different electrode separations (with
width W = 700 μm) and different widths (with separation,
L = 20 μm) respectively. The parasitic capacitance of the SOI
structure (~0.1 nF) is nearly 3 orders of magnitude larger
than that of the glass substrate (~0.1 pF). Also, the parasitic
capacitance for the SOI substrate increases linearly with the
increase in width of the electrodes.
This journal is © The Royal Society of Chemistry 2014
A first order estimate of parasitic capacitance can be
obtained by assuming the top silicon layer to be metal, so
that net capacitance C 
WH E ox
 0.97 nF . However, since
the top silicon layer has finite conductivity, the actual capacitance is smaller than the estimate which is observed in the
simulation. Regardless, such a large parasitic coupling
f high ~
and confines the opti2Rseries  Cgeo  Cpar 
mum sensor operation close to the low/intermediate frequency regimes.11
If one must perform droplet characterization on an SOI
substrate at very high frequencies, a parallel plate detection
system as described in Sadeghi et al.17 may be used. This will
ensure that most of the electric field from the electrode is
confined within the droplet, resulting in lower sensitivity to
the substrate.
3.3 Experimental verification
In order to validate the model described in the numerical
section, both the time and frequency response of droplets
containing different DNA concentrations were analyzed. The
data was obtained from Ebrahimi et al.11 The frequency
response of the system at t = 2 min was calibrated with the
numerical model (see eqn (2)) to determine the ionic conductivity (σ) for different DNA concentrations (see Fig. 7(a–c)).37
Using this ionic conductivity (σ), the time response of the
system was determined using Z = Zdroplet(t)||Zpar (see eqn (8),
Fig. 7(d)). Zpar was obtained from the droplet free measurement on the substrate.11 The ionic conduction was assumed
to take place due to H+ and OH− ions as the experiments
were performed using DI water containing DNA molecules.
The DNA solution (purchased from Fermentas, Inc.) had
850 bp long synthetic molecules in 1× TAE buffer solution.
The DNA molecules were precipitated using an isopropanol
precipitation method and resuspended in nuclease-free
DI water. Additional experimental details are provided in
Ebrahimi et al.11
Despite the various simplifying assumptions made in
Section 1, the model (solid lines) predictions agree with the
experimental results (circles) remarkably well. Indeed, apart
from fitting the t = 2 min conductivity at various analyte concentrations, the model describes the time-evolution and
frequency dependence of the droplet impedance consistently
without any other fitting parameters. The key features of the
experiments are reproduced: First, the model correctly
estimates the frequency response of the system. At low
frequency, the impedance is dependent on the composition
of the droplet (DNA) (compare Fig. 7(a)–(c)). At high frequency, the impedance of the parasitic substrate dominates
and yields essentially same impedance for different DNA concentrations, making the high frequency regime unsuitable for
detection. Second, Fig. 7(d) shows that the time-evolution of
the impedance predicted by the theoretical model correctly
reproduces the features observed in the experiment. The
Lab Chip, 2014, 14, 2469–2479 | 2477
View Article Online
Published on 26 March 2014. Downloaded by Purdue University on 24/07/2014 18:31:38.
Lab on a Chip
Fig. 7 Impedance vs. frequency (calibration curves) at t = 2 min for different DNA concentrations (a) 330 fM, (b) 3.3 pM and (c) 33 pM.
(d) Impedance vs. time for different DNA concentrations: 330 fM (red), 3.3 pM (black) and 33 pM (blue). Lines and circles represent simulation and
experiment respectively. Experimental data was taken from Ebrahimi et al.11
impedance of the droplets with different DNA concentrations
converge at higher times, due to the decrease in droplet volume (ΔZ(t) ∝ V(t)η, where η ≈ 1/2 or 1 depending on whether
Cdl or Rseries is dominant (refer to eqn (8)–(10)). Due to large
parasitic capacitance, fhigh varies in the range of 350 Hz–960 Hz
from t = 0 to t = 18 min respectively. This limits the operation
of the device to the sub-kHz range for sensing operations
even at longer times. Additional results with phase plots
that validate the robustness and accuracy of the model are
presented in the ESI,† Section 5.
We have developed a comprehensive numerical and compact
modeling framework for the impedance of an evaporating
droplet. The model is simple, and yet the theoretical framework correctly predicts the complex, time-dependent electrical response of an evaporating droplet containing analyte
molecules. Indeed, once the geometry factor g(t) and the
volume evolution V(t) are determined, either through
experiments or through numerical/analytical modeling, the
response of the system is completely specified. As a result,
this physics-based model can be used to optimize a variety of
droplet-related systems (e.g. the operation of a droplet-based
sensor) once the system parameters, such as the mobility of
ions and applied bias, are specified. The model also
2478 | Lab Chip, 2014, 14, 2469–2479
highlights the critical importance of the substrate for highly
sensitive impedance-based chem-bio sensing. Specifically, for
example, the model suggests that, compared to typical SOI
substrates, the reduced parasitic impedance of a glass substrate would improve the overall sensitivity as well as provide
a broader bandwidth of operation. Furthermore, higher frequencies can be used to characterize the droplet shape and
size, since the impedance in that regime is independent of
the droplet composition. If one must use SOI substrates for
integration purposes, a comparable level of sensitivity is
obtained only if the operating frequency is reduced to an
extent that completely eliminates the effects of parasitic
impedance on the overall impedance of the system.
This work was supported by the National Science Foundation
through the NCN-NEEDS program (1227020-EEC) and the
National Institute of Health (R01-CA20003). The authors also
thank Prof. Rashid Bashir, Prof. Suresh Garimella and Prof.
Pradeep Nair for useful discussions.
1 A. Marmur, Langmuir, 2004, 20, 3517–3519.
2 R. D. Deegan, O. Bakajin, T. F. Dupont, G. Huber, S. R. Nagel
and T. A. Witten, Nature, 1997, 389, 827–829.
This journal is © The Royal Society of Chemistry 2014
View Article Online
Published on 26 March 2014. Downloaded by Purdue University on 24/07/2014 18:31:38.
Lab on a Chip
3 R. Deegan, Phys. Rev. E: Stat. Phys., Plasmas, Fluids, Relat.
Interdiscip. Top., 2000, 61, 475–485.
4 P. J. Yunker, T. Still, M. A. Lohr and A. G. Yodh, Nature,
2011, 476, 308–311.
5 A. U. Chen and O. A. Basaran, Phys. Fluids, 2002, 14, L1.
6 J. Berná, D. A. Leigh, M. Lubomska, S. M. Mendoza,
E. M. Pérez, P. Rudolf, G. Teobaldi and F. Zerbetto,
Nat. Mater., 2005, 4, 704–710.
7 L. Mazutis, J. Gilbert, W. L. Ung, D. A. Weitz, A. D. Griffiths
and J. A. Heyman, Nat. Protoc., 2013, 8, 870–891.
8 E. Salm, C. D. Guevara, P. Dak, B. R. Dorvel, B. Reddy,
M. A. Alam and R. Bashir, Proc. Natl. Acad. Sci. U. S. A.,
2013, 110, 3310–3315.
9 J. Jing, J. Reed, J. Huang, X. Hu, V. Clarke, J. Edington,
D. Housman, T. S. Anantharaman, E. J. Huff, B. Mishra,
B. Porter, A. Shenker, E. Wolfson, C. Hiort, R. Kantor,
C. Aston and D. C. Schwartz, Proc. Natl. Acad. Sci. U. S. A.,
1998, 95, 8046–8051.
10 F. De Angelis, F. Gentile, F. Mecarini, G. Das, M. Moretti,
P. Candeloro, M. L. Coluccio, G. Cojoc, A. Accardo,
C. Liberale, R. P. Zaccaria, G. Perozziello, L. Tirinato,
A. Toma, G. Cuda, R. Cingolani and E. Di Fabrizio,
Nat. Photonics, 2011, 5, 682–687.
11 A. Ebrahimi, P. Dak, E. Salm, S. Dash, S. V. Garimella,
R. Bashir and M. A. Alam, Lab Chip, 2013, 13, 4248–4256.
12 S. T. Thoroddsen, T. G. Etoh and K. Takehara, Annu. Rev.
Fluid Mech., 2008, 40, 257–285.
13 S. Ghosh, S. Chattoraj, T. Mondal and K. Bhattacharyya,
Langmuir, 2013, 29, 7975–7982.
14 G. Chen, M. Mohiuddin Mazumder, R. K. Chang,
J. Christian Swindal and W. P. Acker, Prog. Energy Combust.
Sci., 1996, 22, 163–188.
15 A. R. Glover, S. M. Skippon and R. D. Boyle, Appl. Opt., 1995,
34, 8409–8421.
16 J.-G. Guan, Y.-Q. Miao and Q.-J. Zhang, J. Biosci. Bioeng.,
2004, 97, 219–226.
This journal is © The Royal Society of Chemistry 2014
17 S. Sadeghi, H. Ding, G. J. Shah, S. Chen, P. Y. Keng,
C.-J. Kim and R. M. van Dam, Anal. Chem., 2012, 84,
18 P. R. Nair and M. A. Alam, Appl. Phys. Lett., 2006, 88, 233120.
19 P. R. Nair and M. A. Alam, Nano Lett., 2008, 8, 1281–1285.
20 A. J. Bard and L. R. Faulkner, Electrochemical Methods:
Fundamentals and Applications, 2001.
21 J. E. B. Randles, Discuss. Faraday Soc., 1947, 1, 11.
22 D. C. Grahame, Chem. Rev., 1947, 41, 441–501.
23 R. E. G. van Hal, J. C. T. Eijkel and P. Bergveld, Adv. Colloid
Interface Sci., 1996, 69, 31–62.
24 T. Hiemstra and W. H. Van Riemsdijk, Colloids Surf., 1991,
59, 7–25.
25 J. Hong, D. S. Yoon, S. K. Kim, T. S. Kim, S. Kim, E. Y. Pak
and K. No, Lab Chip, 2005, 5, 270–279.
26 Y. Chen, B. He, J. Lee and N. A. Patankar, J. Colloid Interface
Sci., 2005, 281, 458–464.
27 L. J. Lin, S. Y. Chuang, Y. C. Yang and Y. J. Chen, AIChE,
2013 Annu. Meet.
28 K. A. Brakke, Surface Evolver, version 2.70, Susquehanna
University, Selinsgrove, PA, 2013.
29 S. M. Rowan, M. I. Newton and G. McHale, J. Phys. Chem.,
1995, 99, 13268–13271.
30 K. S. Birdi, D. T. Vu and A. Winter, J. Phys. Chem., 1989, 93,
31 R. Picknett and R. Bexon, J. Colloid Interface Sci., 1977, 61,
32 J. D. Jackson, Am. J. Phys., 2000, 68, 789.
33 S. Dash and S. V. Garimella, Langmuir, 2013, 29,
34 Sentaurus, version H–2013.03, United States, 2013.
35 J. Wei, IEEE J. Quantum Electron., 1977, 13, 152–158.
36 G. K. Celler and S. Cristoloveanu, J. Appl. Phys., 2003,
93, 4955.
37 Y.-S. Liu, P. P. Banada, S. Bhattacharya, A. K. Bhunia and
R. Bashir, Appl. Phys. Lett., 2008, 92, 143902.
Lab Chip, 2014, 14, 2469–2479 | 2479