INAE Lett (2017) 2:55–63

DOI 10.1007/s41403-017-0022-z

ORIGINAL ARTICLE

COMSOL-Based Design and Validation of Dilution Algorithm

with Continuous-Flow Lab-on-Chip
Tapalina Banerjee1 • Sarmishtha Ghoshal2 • Bhargab B. Bhattacharya1

Received: 23 December 2016 / Accepted: 20 May 2017 / Published online: 6 June 2017
Ó Indian National Academy of Engineering 2017

Abstract Sample preparation is an essential component in Introduction

any biochemical analysis, which involves, among others,
the task of producing dilutions of raw sample, and fluid- Microfluidics is the study of micro-/nano-/pico-litre vol-
mixtures of different reagents with prescribed proportions. umes of fluids that are enabled to move through micro-
In this work, a dilution method using continuous-flow channels (for Continuous-Flow Microfluidics (CFM)) or
microfluidic (CFM) lab-on-chip (LoC) is proposed, and a travel on patterned-array of electrodes as droplets (for
universal fluidic network for its implementation is Digital Microfluidics (DMF)). Microfluidic technology has
designed. The mixing of fluids is effected by the convec- the potential of replacing conventional bench-top protocols
tive/diffusive transport through specially designed by combining the functionalities of biochemical analysis
Y-shaped serpentine microchannels. Our method does not with sensors on a single chip (Chakrabarty and Thewes
need any valve-control, and instead of emitting waste 2007; Ghoshal et al. 2010). One of the essential factors in
droplets at each mix-split step as practiced earlier in the biochemical analysis is fluid-sample pre-processing, which
case of digital microfluidics, we adjust fluid-injection involves preparation of dilution and/or mixture of several
velocities or the dimension of inlet tubes as needed. The fluids in a certain ratio. Several sample preparation tech-
target concentration factor of the fluid computed via niques based on the CFM technology appeared in the lit-
COMSOL simulation of the network matches favorably erature (O’Neill et al. 2006; Lee et al. 2009; Walker et al.
with the expected value. Furthermore, this investigation 2010; Friedrich et al. 2012; Weibull et al. 2013), and for
provides an effective procedure towards the design of CFM DMF technology (Thies et al. 2008; Roy et al. 2010, 2015;
biochips needed for automated sample preparation with Mitra et al. 2014).
any given input ratio. Although dilution of a raw fluid to any desired con-
centration can be prepared using a fixed DMF fluidic
Keywords Bit-scanning  Concentration factor  architecture (Thies et al. 2008; Roy et al. 2010), it cannot
Fluidic-network  Lab-on-chip  Microfluidics be so easily achieved by the CFM technology. The typical
fluidic networks used in CFM are hard-wired to produce a
pre-designated set of concentrations (Friedrich et al. 2012;
Weibull et al. 2013). Recently, a general 2D-array of flu-
& Tapalina Banerjee
[email protected]
idic network has been proposed (Wang et al. 2016) that can
be programmed for implementing any desired dilution
Sarmishtha Ghoshal
[email protected]
under valve-control. In other words, while DMF-based
biochips are easily programmable, CFM-based chips need
Bhargab B. Bhattacharya
[email protected]
either complex controllers or differently-designed fluidic
network for implementing each target dilution. In this
1
Indian Statistical Institute, Kolkata 700 108, India work, to overcome the above-mentioned shortcoming, a
2
Indian Institute of Engineering Science and Technology, method for diluting fluid sample using CFM LoC is pro-
Howrah 711 103, India posed, in tune with the bit-scanning (BS) or twoWayMix

123
56 INAE Lett (2017) 2:55–63

(TWM) algorithm (Thies et al. 2008), which has been Since, the dilution steps used in these algorithms, consist of
studied extensively in the context to DMF. We have only (1:1) mix-split operations, the final CF, thus achieved,
designed Y-shaped serpentine microchannels, where mix- contains a term in the denominator, which can be expressed
ing of fluids is performed by convective/diffusive transport as an integer-power of 2. Hence, for the sake of conve-
through the channels. No valve or metering technique is nience, each CF is rounded-off as an n-bit binary fraction
used for volume control. Also, instead of emitting waste number 2xn , where n is a positive integer greater than 0
droplets/units following each mix-split step as in the BS- denoting the chosen level of accuracy, and x is a positive
method, we inject fluid with velocities in geometric pro- integer such that 0 B x B 2n.
gression to the inlets of the proposed fluidic-network and/or
by adjusting the dimension of the cross-section of the inlet Prior Art
tube. We observe that the target concentration factor (CF)
obtained via COMSOL simulation matches with the Several earlier techniques for producing concentration
expected value within the allowable error range. Moreover, gradients/dilution of fluids with CFM biochips and the
this analysis leads to the design of a universal and pro- design of the underlying flow network were reported in the
grammable continuous-flow network, which can be used to literature (O’Neill et al. 2006; Lee et al. 2009; Walker et al.
implement any arbitrary dilution, as required, for auto- 2010; Friedrich et al. 2012; Weibull et al. 2013; Wang et al.
mated sample preparation. 2016). However, the design, integration, and monitoring of
continuous-flow systems pose a challenge as a number of
flow parameters, such as fluid resistance, pressure, velocity,
Automated Sample Preparation electric field strength may vary along the flow path and can
affect the flow. In addition to these parameters, various
Preliminaries physical characteristics of fluids and channel dimensions
govern the behavior and robustness of the entire system
One of the crucial components in any biochemical analysis (Nguyen et al. 2013). Also, the management of inter-
is sample preparation. In sample preparation, the amount of sample contamination in CFM chips is difficult. Mixing of
raw sample in a fluid-mixture is known as concentration fluids using microchannels with different geometries have
factor (CF). We assume that the raw sample has 100% been reported (Elabbasi et al. 2012; Das et al. 2014; Bruce,
concentration factor, and the buffer fluid has CF = 0. The http://courses.washington.edu/microflo/; Karthikeyan et al.
value of CF can be any real number within the closed 2014). Theoretical modeling and simulations studies on
interval [0, 1] (Thies et al. 2008; Roy et al. 2010). Thus, for fluid mixing also appeared in the literature (Dhondi and
the dilution problem, a target CF will lie between the open Pereira 2009; Karst et al. 2013; Wang et al. 2016).
interval (0, 1), whereas buffer (CF = 0) and raw sample
(CF = 1) are generally provided as input-fluids. In DMF, Bit-Scanning (BS) or twoWayMix (TWM)
the (1:1) mix-split model is very well known (Thies et al. Algorithm
2008; Roy et al. 2010). In this model, mixing of two unit
volume of droplets is followed by balanced splitting of the Thies et al. (2008) first proposed a mixing algorithm based
composite droplet into two equal unit-volume droplets. In on bit-scanning (BS)/Min-Mix method. Dilution is a special
the BS/TWM algorithm (Thies et al. 2008), the process of case of mixing where a sample is mixed with a buffer in a
diluting a raw sample can be envisaged using a directed certain ratio. As stated earlier, to implement dilution, the
acyclic graph, known as mix-split tree/graph. A traversal of given target CF is first approximated with a user’s defined
this mix-split tree/graph provides the desired sequence of n-bit binary fraction, where n is a positive integer that
(1:1) mix-split operations that are to be performed with a determines the maximum allowable error 2nþ1 1
in CF. Mix-
digital microfluidic biochip (DMFB) in order to achieve the ing is performed based on the (1:1) model, where one unit
target CF. The accuracy of the target CF is determined by of sample is mixed with one unit of buffer, and only one
the depth of the mixing graph, which is a positive integer n unit of the resulting mixture is used discarding the
specified by the user. In BS/TWM algorithm (Thies et al. remaining unit as waste. The algorithm twoWayMix (TWM)
2008), the error limit (a.k.a. accuracy level) in the target works on scanning the bits of the binary fraction from
1
CF is 2nþ1 which is obtained assuming that the number of right-to-left to decide whether a droplet (unit-volume) of
allowed (1:1) mix-split operations is at most n. Note that a sample (if the current bit = 1), or of buffer (if the current
sequence of ten mix-split steps provide a good approxi- bit = 0) is to be mixed with one unit of fluid that was
mation of a target CF with an error less than produced at the preceding step (see Fig. 1). Thus, except
1
211 ’ 0:000488281, which is very small (Roy et al. 2010). the last step where two units of fluid with target CF are

123
INAE Lett (2017) 2:55–63 57

Fig. 1 BS/TWM algorithm for fluid-dilution in DMF (Roy et al. 2010)

obtained, one fluid-unit is wasted at each mix-split step. algorithm used for DMF. We provide a solution in this
This algorithm requires the minimum number of (1:1) mix- paper by controlling the injection velocities of fluids and by
split steps needed to produce a target concentration with modifying the inlet dimensions. The proposed technique
accuracy level n, where reuse of waste-droplets is not mimics the underlying mix-split tree/graph that describes
performed. Figure 1 shows the mix-split sequencing steps the sequence of mix-split operations, but unlike the pre-
313 vious DMF-based methods, it does not need any ‘‘split
to generate the target concentration CF = ð1024 Þ10 =
(0.0100111001)2, where 0’s (1’s) appearing from right-to- step’’ or cause waste of fluids.
left govern the injection of buffer (sample) droplets in the
mixing graph. Note that all extra 0’s appearing on the right
end of the binary fraction are deleted. Also, corresponding Problem Description
to the least significant 1-bit, both buffer and sample dro-
plets are injected (see Fig. 1). In the proposed work, we Input A supply of sample fluid (100% concentration, i.e.,
present a method for diluting a fluid sample using CFM CF = 1), and buffer fluid (0% concentration, i.e., CF = 0),
LoC, in tune with the BS/TWM algorithm. In other words, and a positive integer n denoting the accuracy level.
given any target CF, we will inject sample or buffer Output A mix-sequence that will generate the target
1
through the inlets of the continuous-flow network follow- concentration CF = T not exceeding allowable error  2nþ1
ing the same sequence as mandated by the bits present in when implemented with a CFM biochip.
the binary fraction representing the CF.

Proposed Fluidic-Network
Motivation
We propose a continuous-flow fluidic network for imple-
In the field of CFM LoC, one of the most important issues menting the dilution algorithm, where each inlet will have
is to control the fluid-injection velocity. Also, for ease of a pre-determined cross-section and fluid injection velocity
implementation, we consider a scenario where no valve- regardless of the target CF. We consider two types of
control is required. In order to achieve a target-CF with geometric layout for the fluidic network, the dimensions of
CFM LoC, we need to calculate the volume of raw sample which are shown Fig. 2 (Model 1) and Fig. 3 (Model 2). In
and buffer fluids to be injected per unit time into the both models, we use a passive serpentine mixer (Lee et al.
mixing chamber. Our motivation is to design a universal 2011; Ober et al. 2015) with turning radius 6.5 lm as
fluidic architecture/model that can be used to produce an shown in Figs. 2 and 3 (Kane et al. 2008); additionally, in
arbitrary target concentration CF = T (0\T \1) with Model 2, we use special rectangular boxes to augment the
accuracy level n using CFM in tune with the BS/TWM inlets as shown in Fig. 3. We choose channel-dimensions

123
58 INAE Lett (2017) 2:55–63

Fig. 2 Geometric layout of the serpentine fluidic mixer for Model 1; width of all inlets = 25 lm

Fig. 3 Geometric layout of the serpentine fluidic mixer for Model 2

and fluid-injection velocities following certain rules (de- Reynolds number (r) (Howard 2007) is a dimension-
scribed later), and study the flow dynamics in 2D for dif- less parameter, used to characterize the nature of fluid
ferent input concentration profiles using COMSOL motion. It is the ratio of the inertial force to the viscous
Multiphysics Software (https://www.comsol.co.in/). force, i.e.,

123
INAE Lett (2017) 2:55–63 59

R ¼ Reynolds number Table 1 shows the required inlet boundary width of the
Inertial force rectangular area, applied velocities, and equivalent
¼ velocities appearing at the arms of Y-shaped serpentine
Viscous force ð1Þ
fluidic mixer microchannel. Since we have performed 2D
qu‘
¼ COMSOL simulation, the length of each inlet tube is
l
assumed to be unity, and the width is varied as shown in
where, q represents the density of fluid, l denotes fluid Table 1.
dynamic viscosity, ‘ stands for fluidic channel length, and u
is the speed of fluid-flow. A high value of r indicates
Simulation Framework
turbulent flow, whereas a low value indicates the presence
of laminar flow. In our proposed model, we choose these
We have studied the flow dynamics for different input
parameters (q; l; ‘ and u) in such a way that r is lower
concentration profiles using COMSOL Multiphysics Soft-
than 2000 so that laminar flow for miscible fluids is
ware. Figures 4 and 5 represent the flow dynamics for
guaranteed by the model (Howard 2007). 313
target concentration CF = ð1024 Þ10 = (0.0100111001)2 for
We deploy a cascade of Y-shaped junctions each of
which merges into a serpentine channel of rectangular n = 10, based on Model 1 and Model 2, respectively.
cross-section filled with inter-digitized orthogonal obsta- Initially, the channels are assumed to be filled with air. The
cles (Figs. 2, 3). The detailed dimensions of the channels fluids may have different parameters such as density, vis-
are shown in the figures. The inflow of sample/buffer is cosity, electrical and thermal conductivity, diffusion coef-
passed into the serpentine mixing chamber through dif- ficients, and surface tension. We have used single-phase
ferent microchannels via several inlets oriented with an laminar flow (SPF) to study the parameters of fluids such as
angle of 1208 to each other (Figs. 2, 3). The mixing of fluid velocity profile, fluid pressure profile, and transport of
fluids is performed by convective/diffusive transport diluted species (TDS) to study the concentration profile of
through the serpentine channels. Our method does not the fluids along the channels.
need any valve-control, and no waste management is SPF solves the Navier–Stokes equation (Howard 2007)
needed as in the BS/TWM algorithm (Thies et al. 2008). In that governs the motion of fluids, and is analogous to
our Model 1, we increase fluid-injection velocities in Newton’s second law of motion for fluids. In the case of an
geometric progression at each subsequent inlet so that the incompressible Newtonian fluid, this equation is:
 
volume of fluid inflow is doubled at each step. At the two ou
q þ u  ru ¼ rp þ r
inlets of the first stage, the velocity is set to 0.1 mm/s ot
each, and then it is multiplied by a factor of two at every  
T
2
following inlet.  l ru þ ðruÞ  lðr  uÞI þ F
3
The number of required Y-shaped serpentine channels is
ð2Þ
n and the total number of inlets is (n ? 1), where n is the
number of bits in the binary fraction used to approximate the where u is the fluid velocity, p is the fluid pressure, q is the
target CF. Hence, in Model 1, we need n fluid-injection fluid density, and l is the dynamic viscosity. The left-hand-
velocities in geometric progression (in the first stage, we use side term represents inertial forces; on the right-hand side,
two equal velocities). However, Model 2 allows us to reduce the term (-rp) represents pressure-force; viscous-force is



the number of required fluid-injection velocities by changing represented by r  l ru þ ðruÞT  23 lðr  uÞI , and
the dimensions of fluid-inlet channels. We keep the length of F denotes the external forces applied to the fluid. These
each inlet same, but double the width of the inlet channel equations are solved together with the continuity equation
when needed so as to supply the required fluid-volume per (Howard 2007):
unit time through an inlet-branch. We have used rectangular
boxes as extensions at the inlets (Fig. 3) to implement this. oq
þ r  ðquÞ ¼ 0 ð3Þ
Note that a larger value of n will necessitate the use of more ot
fluid-inlets and high injection velocities or wider input The two properties, namely ‘‘conservation of
channels. Unfortunately, in these cases, flow-control and momentum’’ and ‘‘conservation of mass’’ for incom-
homogeneous mixing may be harder to achieve in practice. pressible fluids are captured by the Navier-Stokes
Thus, a compromise is needed. In our experiments, we have equation (2) and the continuity equation (3), respec-
chosen n to lie between 5 and 10; such a choice provides tively. The relative concentration of the sample is rep-
sufficient accuracy in CF from users’ perspective as well as resented by Fick’s laws (Crank 1975). Using Fick’s
supports manageable flow dynamics. laws, TDS can provide the solution of the convection-

123
60 INAE Lett (2017) 2:55–63

Table 1 Generation of ten fluid-velocities using two injection-ve- rci = Ri, where Di is the diffusion coefficient of fluid
locities and five inlet-widths i in the solution, the conserved quantity ci is the con-
Required inlet Applied velocity at Equivalent velocity centration of the fluid i, and Ri is the reaction rate of the
boundary width inlet boundary in mm/s at inlet-arm in mm/s fluids. We have used ethanol (diffusivity 1.24 9 10-9
in lm m2/s) (Goranovic and Bruus 2003) and glycerol (diffu-
6.25 0.8 0.1 sivity 0.94 9 10-9 m2/s) as sample, and water as buffer.
12.5 0.8 0.2 The boundary conditions are specified on the surfaces
25 0.8 0.4 that define the computational domain. The typical boundary
50 0.8 0.8 conditions affect the values of velocity, pressure, and
100 0.8 1.6
concentration. Meshing is another important part of the
6.25 25.6 3.2
simulation. We have used the same meshing for both of our
fluidic-network models. We set the predefined element size
12.5 25.6 6.4
at extra-fine while calibrating fluid dynamics. Time-dis-
25 25.6 12.8
cretization is needed to simulate time-dependent flows.
50 25.6 25.6
Generally, smaller time steps are needed in order to achieve
100 25.6 51.2
more accurate results. A time range of 0–25 s is considered
that provides saturation for diffusive mixing in the
diffusion equation. In our model, there exists advection microchannels, and simulation is performed with a time-
(i.e., net transport of whole solution). Hence, we have step of 1 s. In Table 2, we describe the details of the com-
used TDS with transport equation: r(-Di  rci) ? u  putational platform used in our simulation.

Fig. 4 Flow dynamics (based

on Model 1) in a cascade of
Y-shaped serpentine channels
with velocities in geometric
progression for target
313
concentration CF = ð1024 Þ10 =
(0.0100111001)2 where n = 10

Fig. 5 Flow dynamics (based

on Model 2) in a cascade of
Y-shaped serpentine channels
with various cross-sections for
target concentration
313
CF = ð1024 Þ10 =
(0.0100111001)2 where n = 10

123
INAE Lett (2017) 2:55–63 61

Table 2 Computation platform

CPU Intel Core i7-3770 CPU @ 3.40 GHz, 4 cores
Operating system Windows 8
Software COMSOL Multiphysics 5.2 (https://www.comsol.co.in/)
Dimension of models 2D
Materials (incompressible Newtonian fluids) Water (buffer)
Ethanol (sample; diffusion coefficient of ethanol in water is 1.24 9 10-9 m2/s)
Glycerol (sample; diffusion coefficient of glycerol in water is 0.94 9 10-9 m2/s)
Used physics Laminar flow (SPF: Single Phase Flow)
Transport of diluted species (TDS)
Meshing Predefined element size at extra fine while calibrating fluid dynamics
Study Time dependent (Time range: 0– 25 s; time step: 1 s)

Results and Discussions Table 3 Simulation results

Target Allowable Simulated Within Test-case type
Our results show that all target CFs that we have studied CF (T ) error range result for the allowable
can be produced within the maximum allowable error e in (e) proposed error
1 models range?
CF, i.e., not exceeding 2nþ1 where n is the number of bits
used to represent the CF as a binary fraction CF = T is 33 0:5
 256 32:96 Yes Random
 0
 0
256 256
T  T ¼ jxx j
2n , which is observed not to have exceeded
313
1024
0:5
 1024 313:29
1024
Yes
83 0:5 82:98 Yes
the allowable error e ¼  2nþ11
(Table 3). Our goal is to 128  128 128
27 26:99
obtain T within the accuracy level e assuming laminar 32  0:5
32 32
Yes
132 0:5 131:81
flow. We have performed COMSOL simulation for gen- 256  256 256
Yes
erating different target CFs, such as 13%, 30%, 65%, and 61
64  0:5
64
60:99
64
Yes
84% for n = 8, 10, 7, and 5, respectively. In other words, 421 0:5
 512 421:25 Yes
33 313 83 512 512
these CF-values are represented as 256 , 1024 , 128 , and 27
32, 486 0:5 485:96 Yes Real-life
 512
respectively. Figure 6 shows that simulated values of CFs 512 512
358 0:5 358:01 Yes
do not exceed the allowable error bound, after saturation. 512  512 512

For these target-CFs, the steady state, i.e., saturation is 51 0:5

 512 50:55 Yes Bradford
512 512
reached after around 22 s. Thus, simulation results match 102 0:5 101:89 Yes
protein assay
512  512 512
quite favorably with the expected values. Table 3 shows 154 0:5 154:28
 512 Yes
the simulated results for some real-life test-cases such as 512 512
205 0:5 204:59 Yes
Bradford Protein Assay and Sucrose Gradient Analysis, 512  512 512
and for some arbitrary test cases. The correctness of the 256 0:5 255:78 Yes
512  512 512
procedure directly follows from that of the BS-algorithm. 51 0:5 50:55
512  512 512
Yes Sucrose
77 0:5 76:87 gradient
512  512 512
Yes
analysis
102 0:5 101:89 Yes
Conclusions and Future Work 512  512 512
128 0:5 127:87 Yes
512  512 512
We have suggested a general method that can be used to 154
512
0:5
 512 154:28
512
Yes
design a continuous-flow microfluidic network for 179 0:5
 512 179:44 Yes
512 512
preparing any target dilution of a fluid sample. We 205 0:5 204:59 Yes
512  512 512
determine the geometry of channels and the dimensions
of inlets and mixing channel for implementing the
diluter network. Note that the same network can be used different inlet dimensions that effectively produce ten
to perform dilution accurately, given any target con- different fluid-injection velocities at the inlets needed for
centration factor. Only the sequence of fluid injections achieving an accuracy level of (1/211) in CF (i.e., for
with the sample and buffer through the inlets is to be n = 10). Our technique is based on a variant of the BS/
changed depending on the target CF. In Model 2, we TWM algorithm (Thies et al. 2008), which is dominantly
have used only two fluid-injection velocities and five used for sample preparation in digital microfluidics. We

123
62 INAE Lett (2017) 2:55–63

Fig. 6 Variation of target CF

produced at the output with
time, and the final error with
respect to the expected CF

have reported COMSOL simulation of the mixing pro- Conference on CFD in Minerals and Process Industries CSIRO,
cess needed for preparing various dilutions of two dif- Melbourne, Australia, pp. 1–6
Diffusivity of Glycerol. http://www.hypertextbookshop.com/biofilm
ferent fluids and validated the correctness of our book/working_version/artifacts/tables/Module_004/Table4-1_
approach. The proposed technique can be conveniently DiffCoeffH2O.htm. Accessed 01 June 2016
used for the design and simulation of other sample- Elabbasi N, Liu XO, Brown S (2012) Modeling of laminar flow static
preparation tasks with CFM biochips. Implementation of mixers. COMSOL News
Friedrich D, Please CP, Melvin T (2012) Design of novel microfluidic
3D simulation of the proposed method using COMSOL concentration gradient generators suitable for linear and expo-
Multiphysics software will be taken up as future work. nential concentration ranges. Chem Eng J 193–194:296–303
Ghoshal S, Mitra D, Roy S, Dutta Majumder D (2010) Biosensors and
Acknowledgements One of the authors, T. Banerjee, is registered biochips for nanomedical applicationa: a review. Sens Trans-
for PhD degree at the Department of Computer Science and Engi- ducers 113(2):1–17
neering, University of Calcutta, and she would like to thank the Goranovic G, Bruus H (2003) Simulations in microfluidics. In:
Department of Science and Technology (DST), Government of India, Geschke O, Klank H, Telleman P (eds) Microsystem engineering
for supporting her research under the Indo-Taiwan Collaborative of lab-on-a-chip devices. Wiley-VCH Verlag GmbH & Co.
Research Project (2014–17). The work of B. B. Bhattacharya was KGaA, Weinheim, FRG. doi:10.1002/3527601651.ch5
supported, in part, with the grant provided by INAE Chair Howard AS (2007) Introduction to fluid dynamics for microfluidic
Professorship. flows. In: Lee H, Ham D, Westervelt RM (eds) CMOS
biotechnology. Springer, New York, pp 5–30
Kane SA, Hoffmann A, Baumga P, Seckler R, Reichardt G, Horsley AD,
Schuler B, Bakajin O (2008) Microfluidic mixers for the investigation
References of rapid protein folding kinetics using synchrotron radiation circular
dichroism spectroscopy. Anal Chem 80:9534–9541
Bruce AF (Professor Emeritus of Chemical Engineering, University Karst C, Storey B, Geddes JB (2013) Laminar flow of two miscible
of Washington). http://courses.washington.edu/microflo/. fluids in a simple network. Phy Fluids. doi:10.1063/1.4794726
Accessed 01 June 2016 Karthikeyan K, Sudharsan N, Sujatha L (2014) Design of high
Chakrabarty K, Thewes R (2007) Guest editors’ introduction: performance micromixer for lab-on-a-chip (LoC) applications.
biochips and integrated biosensor platforms. IEEE Design Test COMSOL Conference, Bangalore
24(1):8–9 Lee K, Kim C, Ahn B (2009) Generalized serial dilution module for
COMSOL (2016) Multiphysics Software. https://www.comsol.co.in/ monotonic arbitrary microfluidic gradient generation. Lab Chip
Crank J (1975) The mathematics of diffusion. Clarendon Press 9:709–717
Oxford, Ely House, London W.I. Lee CY, Chang CL, Wang YN, Fu LM (2011) Microfluidic mixing: a
Das SS, Patawari BK, Patowari PK, Halder S (2014) Computational review. Int J Mol Sci 12:3263–3287
analysis for mixixng of fluids flowing through micro-channels of Mitra D, Roy S, Bhattacharjee S, Chakrabarty K, Bhattacharya BB
different geometries. AIMTDR, pp. 236:1–236:6 (2014) On-chip sample preparation for multiple targets using
Dhondi S, Pereira GG (2009) Complex fluid mixing in microfluidic digital microfluidics. IEEE Trans Comput Aided Des Integr
devices: theory and simulations. Seventh International Circuits Syst 33(8):1134–1144

123
INAE Lett (2017) 2:55–63 63

Nguyen NT, Shaegh SA, Kashanineja N, Phan DT (2013) Design, application-specific digital microfluidic biochips. ACM Trans
fabrication and characterization of drug delivery systems based Design Autom Electron Syst 20(3):45:1–45:33
on lab-on-a-chip technology. Adv Drug Deliv Rev Thies W, Urbanski JP, Thorsen T, Amarasinghe SP (2008) Abstrac-
65:1403–1419 tion layers for scalable microfluidic biocomputing. Nat Comput
Ober TJ, Foresti D, Lewis JA (2015) Active mixing of complex fluids 7(2):255–275
at the microscale. PNAS 112(40):12293–12298 Walker GM, Monterio-Riviere N, Rouse J, Neill OT (2010) A linear
O’Neill AT, Monteiro-Riviere N, Walker GM (2006) A serial dilution dilution microfluidic device for cytotoxicity assays. Lab Chip
microfluidic device for cytotoxicity assays. In: International 7:709–717
conference of the IEEE engineering in medicine and biology Wang J, Brisk P, Grover H (2016) Random design of microfluidics.
society, New York, 2006, pp 2836–2839 Lab Chip 16(21):4212–4219
Roy S, Bhattacharya BB, Chakrabarty K (2010) Optimization of Weibull E, Matsui S, Sakai M (2013) Microfluidic device for
dilution and mixing of biochemical samples using digital generating a stepwise concentration gradient on a microwell
microfluidic biochips. IEEE Trans Comput Aided Des Integr slide for cell analysis. Biomicrofluidics 7(6):064115:1–064115:1
Circuits Syst 29(11):1696–1708
Roy S, Chakrabarty PP, Chakrabarty K, Bhattacharya BB (2015)
Layout-aware mixture preparation of biochemical fluids on

123