Cross-Contamination Avoidance for Droplet Routing in
Digital Microfluidic Biochips
∗
Yang Zhao and Krishnendu Chakrabarty
Department of Electrical and Computer Engineering
Duke University, Durham, NC 27708, USA
Abstract— Recent advances in droplet-based digital microfluidics
have enabled biochip devices for DNA sequencing, immunoas-
says, clinical chemistry, and protein crystallization. Since cross-
contamination between droplets of different biomolecules can lead
to erroneous outcomes for bioassays, the avoidance of cross-
contamination during droplet routing is a key design challenge
for biochips. We propose a droplet-routing method that avoids
cross-contamination in the optimization of droplet flow paths. The
proposed approach targets disjoint droplet routes and minimizes
the number of cells used for droplet routing. We also minimize the
number of wash operations that must be used between successive
routing steps that share unit cells in the microfluidic array. Two
real-life biochemical applications are used to evaluate the proposed
droplet-routing methods.
I. INTRODUCTION
Droplet-based digital microfluidics is an emerging technology
that provides fluid-handling capability on a chip. It has therefore
led to the automation of laboratory procedures in biochem-
istry [1]. By reducing the rate of sample and reagent consump-
tion, digital microfluidic biochips enable continuous sampling
and analysis for real-time biochemical analysis, with application
to clinical diagnostics, immunoassays, and DNA sequencing.
Discrete droplets of nanoliter volumes can be manipulated using
electrowetting in a “digital” manner under clock control on a
two-dimensional array of electrodes (“unit cells”). Synthesis and
droplet-routing methods have been developed recently for the
design of microfluidic biochips [3], [5], [9], [10], [12]–[16].
Many biomedical assays require the on-chip transportation
of biological substances that contain large molecules such as
proteins. However, proteins cannot be transported easily on
a microfluidic platform [1]. This is because proteins tend to
adsorb irreversibly to hydrophobic surfaces and contaminate
them. Silicone oil with its low surface tension and spreading
property has been advocated as a filler medium for protein assays
to prevent contamination [1]. However, it has also been reported
that the use of silicone oil alone is not sufficient for many types of
proteins [4]. A set of wash droplets is typically used in such cases
for surface cleaning between successive droplet transportation
steps, especially for unit cells that are shared by droplet routes.
A drawback of current automated droplet-routing methods is
that they are based on the unrestricted sharing of unit cells by
various droplet routes. Therefore, cross-contamination in such
cases is inevitable. Cross-contamination occurs when the residue
left behind by one droplet transfers to another droplet with
undesirable consequences, such as misleading assay outcome
∗
This work was supported in part by the National Science Foundation under
grant CCF-0541055.
(false position, incorrect diagnosis, etc.). As a result, the droplet-
routing problem for biochips must consider the avoidance of
cross-contamination during droplet transportation.
In this paper, we present a new droplet-routing method for
digital microfluidic biochips. A key goal here is to avoid overlap
between different droplet routes, thereby minimizing the likeli-
hood of cross-contamination. Our approach attempts to determine
disjoint droplet routes and minimize the number of cells used for
droplet routing, while satisfying both timing goals and fluidic
constraints. Droplet-routing time must be minimized to reduce
time-to-result for biochips, which is necessary to prevent sample
degradation and ensure real-time response. We also minimize the
number of wash operations inserted between successive routing
steps that share unit cells in the microfluidic array.
The remainder of the paper is organized as follows. Section
II provides an overview of the digital microfluidic platform
and related prior work on droplet routing. In Section III, we
formulate the problem of disjoint droplet routing. Based on this
formulation, we propose a new routing technique in Section IV. A
method to optimally insert washing steps between routing steps
is also presented. In Section V, we use two real-life bioassay
applications as case studies to evaluate the proposed method.
Finally, conclusions are drawn in Section VI.
II. B
ACKGROUND AND RELATED PRIOR WORK
A digital microfluidic biochip utilizes the phenomenon of elec-
trowetting to manipulate and move nanoliter droplets containing
biological samples on a two-dimensional electrode array [1]. A
unit cell in the array includes a pair of electrodes that acts as
two parallel plates. The bottom plate contains a patterned array of
individually controlled electrodes, and the top plate is coated with
a continuous ground electrode. A droplet rests on a hydrophobic
surface over an electrode. Droplets are moved by applying a
control voltage to a unit cell adjacent to the droplet and, at
the same time, deactivating the one just under the droplet. This
electronic method of wettability control creates interfacial tension
gradients that move the droplets to the charged electrode.
Fluid-handling operations such as droplet merging, splitting,
mixing, and dispensing can be executed in a similar manner. The
digital microfluidic platform offers the additional advantage of
flexibility, referred to as reconfigurability, since fluidic operations
can be performed anywhere on the array. Droplet routes and op-
eration scheduling result are programmed into a microcontroller
that drives electrodes in the array.
A number of techniques have been proposed in the literature
to solve the droplet-routing problem for digital microfluidics [5],
[9], [14]–[16]. These methods are based on concepts such as
path-planning for robots, prioritized A
∗
search, network flow,
integrated synthesis-and-routing, etc. However, the droplet routes
978-3-9810801-5-5/DATE09 © 2009 EDAA
Authorized licensed use limited to: DUKE UNIVERSITY. Downloaded on July 26, 2009 at 22:58 from IEEE Xplore. Restrictions apply.
may intersect or overlap with each other during different time
intervals. Therefore, cross-contamination between droplets may
happen at these sites, and this is a serious problem for all known
automated droplet-routing methods.
III. P
ROBLEM FORMULATION AND CONSTRAINTS
In this section, we describe the problem of droplet routing to
avoid cross-contamination.
A. Problem formulation
Given a schedule of bioassay operations (derived from
architectural-level synthesis [12]) and the locations of these
modules on the biochip floorplan (derived from module place-
ment [13]), routing determines the paths for droplet transportation
using the available cells in the microfluidic array. Droplets are
transported along these routes between modules, or between
modules and fluidic I/O ports (e.g., on-chip reservoirs and
sensors).
The fluidic ports on the boundary of microfluidic modules
are referred to as pins. The droplet routes between pins of
different modules or on-chip reservoirs are referred to as nets.
Therefore, a fluidic route on which a single droplet is transported
between two terminals (i.e., one source and one sink) can be
modeled as a 2-pin net. Two droplets from different terminals are
often transported to one common module (i.e., mixer or diluter)
for mixing or diluting. To allow droplet mixing simultaneously
during their transportation, we model such fluidic routes using
3-pin nets.
Cross-contamination is likely to occur when multiple droplet
routes intersect or overlap with each other. At the intersection
site of two droplet routes, a droplet that arrives at a later clock
cycle can be contaminated by the residue left behind by another
droplet that passed through at an earlier clock cycle. The more
cells that two droplet-routes share, the higher is the likelihood
of cross-contamination.
Our first goal is to avoid cross-contamination between different
droplet routes. The second goal is to minimize the time needed
for droplet routing. Therefore, we focus on disjoint droplet routes
and the minimization of the total path length over all routes,
where path length is measured by the number of cells in the
path from the source to the sink.
In a set of disjoint routes, a droplet route does not share
any cell in its path with each of the other droplet routes in
that set. Such routes eliminate the possibility of a droplet being
transported via a cell when another droplet has already passed
through it in an earlier time interval. Therefore, disjoint routes
avoid cross-contamination between different droplets. The mini-
mization of the length of the droplet routes leads to a reduction
in the droplet-transportation time. It also frees up more spare
cells for parallel fluidic operations and fault tolerance [14].
B. Fluidic and timing constraints
Although disjoint droplet routes avoid cross-contamination
between different droplets, some droplets may inadvertently
mix if they are routed too close simultaneously during their
transportation. Therefore, during droplet routing, a minimum
spacing between droplets must be maintained to avoid unintended
mixing, except for the case that droplet merging is desired (i.e.,
in 3-pin nets). A segregation region is added to wrap around
the functional region of microfluidic modules to avoid conflicts
between droplet routes and assay operations that are scheduled
at the same time. Moreover, fluidic constraint rules in [14] need
to be satisfied in order to avoid undesirable mixing.
Another constraint in droplet routing is given by an upper
limit on droplet transportation time. Since a droplet may be held
stationary in some clock cycles during its route, the delay for a
droplet route consists of the transportation time as well as the
idle time. The delay for each droplet route should not exceed
some maximum, e.g., 10% of a time-slot used in scheduling, in
order that the droplet-routing time can be ignored for scheduling
assay operations [12].
C. Problem decomposition
Since a digital microfluidic array can be reconfigured dy-
namically at run-time, a series of 2-D placement configurations
of modules in different time spans are obtained in the module
placement phase [13]. Therefore, the droplet routing is decom-
posed into a series of sub-problems. In each sub-problem, the
nets to be routed between the sources and the sinks are first
determined. The microfluidic modules that are active at the time
when droplets are transported are considered as obstacles.We
obtain a complete droplet-routing solution by solving these sub-
problems sequentially.
Cross-contamination can occur within one sub-problem, and
also across two sub-problems, e.g., a droplet route in the current
sub-problem may share the same cells with a droplet route in
the predecessor sub-problem. We next address the problem of
determining disjoint routes and minimizing the route lengths for
both cases.
IV. R
OUTING METHOD
In this section, we first present the basic concepts underlying
the proposed routing method.
A. Graph model and disjoint routing
The problem of finding feasible disjoint routes for 2-pin or
3-pin nets in the 2-D microfluidic array can be directly mapped
to the problem of finding disjoint paths (vertex-disjoint or edge-
disjoint) for pairs of vertices in a graph [6], [7]. We consider
a planar undirected graph G =(V,E), where V is the vertex
set and E is the edge set. Each vertex in the graph represents
an electrode in the 2-D microfluidic array, and there is an
edge between two vertices if their corresponding electrodes are
adjacent. The pins in a 2-pin net are represented by a pair
of vertices (t
i
,s
i
) in the graph. The pins in a 3-pin net are
represented by a set of three vertices (t
1,i
,t
2,i
,s
i
), where t
1,i
and t
2,i
represent the two source pins, and s
i
represents the
destination pin. A 3-pin net (t
1,i
,t
2,i
,s
i
) can be treated as two
2-pin nets (t
1,i
,t
2,i
) and (m
i
,s
i
), where m
i
is the mixing point
in the route of the 2-pin net (t
1,i
,t
2,i
) where the droplets from
the two source pins t
1,i
and t
2,i
mix together. The route for a
net is represented by a path consisting of a set of successive
edges in the graph, where the endpoints of a path represent the
corresponding pins. Each edge in the path denotes the fact that
the electrodes represented by the two endpoints of the edge are
adjacent in the droplet route.
Consider a set of disjoint routes where a droplet route does
not share any cell with each of the other droplet routes. The
Authorized licensed use limited to: DUKE UNIVERSITY. Downloaded on July 26, 2009 at 22:58 from IEEE Xplore. Restrictions apply.
corresponding paths in the graph are mutually vertex-disjoint
since a path does not share any vertex with each of the other
paths in the set. Similarly, in a set of disjoint routes where a
droplet route does not share any pair of adjacent cells with each
of the other droplet routes, their corresponding paths in the graph
are mutually edge-disjoint since a path does not share any edge
with each of the other paths in the set.
Since a 3-pin net can be treated as two 2-pin nets, henceforth
we only consider 2-pin nets for disjoint routing. In the 2-D
microfluidic array, given a set of n 2-pin nets with corresponding
pins (t
1
, s
1
), (t
2
, s
2
),...,(t
n
, s
n
), the problem of finding feasible
disjoint routes for these nets is equivalent to the problem of
finding mutually vertex-disjoint or edge-disjoint paths in G, such
that the endpoints of each path in G represent the corresponding
pins of each net. Given G =(V,E) and the vertex pairs (t
1
,
s
1
), (t
2
, s
2
), ..., (t
n
, s
n
), the problem of determining whether
mutually vertex-disjoint paths P
1
, P
2
, ..., P
n
exist such that
P
i
has endpoints t
i
and s
i
, is NP-complete [7]. Furthermore,
the problem of determining whether mutually edge-disjoint paths
exist is also NP-complete, even if the graph G is a 2-D mesh [6].
Therefore, we use a heuristic approach in this work. Furthermore,
it is often difficult to find vertex-disjoint paths in the underlying
graph model to solve the droplet-routing problem; such paths
might not exist for a given set of nets. It is therefore more
practical to relax the overlap restriction and search for edge-
disjoint routes in such cases. Edge-disjoint routes lead to a
reduction in the number of array sites that need to be washed.
B. Avoiding cross-contamination within one sub-problem
We next present a droplet-routing algorithm based on dis-
joint routes that minimizes route lengths and avoids cross-
contamination within a sub-problem. The input to the algorithm
is a list of nets to be routed in the sub-problem, and the output is
a set of vertex-disjoint (preferred) or edge-disjoint (as a design
compromise) routes with minimized lengths, subject to both
fluidic and timing constraints.
Within a sub-problem, the Lee algorithm, a popular technique
used in grid routing [8], can obtain a single droplet route for
each net. The Lee algorithm is guaranteed to find the shortest
path between two pins in a two-pin net. For 3-pin net, we use
the modified Lee algorithm from [14] to obtain a feasible route
connecting these 3 pins. Note that the interconnection obtained by
the modified Lee algorithm from [14] is not guaranteed to be of
minimum length. Nevertheless, it is a desirable route in practice,
allowing concurrent mixing during transportation. However, the
Lee algorithm does not avoid cross contamination.
The microfluidic modules that are active in this sub-problem
are considered as obstacles. The individual net in this sub-
problem is routed using the modified Lee algorithm sequentially.
After any net has been routed, the cells occupied by its path are
marked as obstacles for the unrouted nets. Therefore, the latter
route is disjoint with respect to all the previous routes. However,
the routing order in the sub-problem influences the routability of
all the nets. Even if each of the nets is individually routable, the
routing order may prevent successful completion of routing for
all the nets.
Therefore, we modify the net-routing ordering method pro-
posed in [8] to obtain an optimized order for the routing of n
nets in a sub-problem. We first define the bounding box of a
net. Assuming that two pins of a 2-pin net p are p
1
and p
2
with
coordinates (x
1
,y
1
) and (x
2
,y
2
) in a 2-D microfluidic array, the
coordinates of the four vertices of its corresponding bounding
box are (x
1
,y
1
), (x
1
,y
2
), (x
2
,y
1
) and (x
2
,y
2
). Next we define
pin(p) to be the number of pins (for other nets to be routed
in the current sub-problem) within the bounding box of net p.
Let Xrange(p)=|x
1
− x
2
| and Yrange(p)=|y
1
− y
2
|.The
bounding box for a 3-pin net is defined in a similar manner. The
priority number for a net p is given by the following equation,
where A is a user defined parameter. The larger the priority
number, the lower the priority of the corresponding net in the
net-routing order.
priority(p)=pin(p)+A · max{Xrange(p),Yrange(p)}. (1)
There are two stages in the net-routing process. In the first stage,
we apply the modified Lee algorithm to each net sequentially in
the net-routing order. The cells occupied by the routed paths
are marked as obstacles for the unrouted nets. Each route thus
obtained needs to pass the “timing delay constraint check”
(TDCC). Additionally, each route also need to go through the
“fluidic constraint rule check” (FCRC) with previously acquired
routes according to the net-routing order. If a rule violation is
found, we modify droplet motion using the modification rules
in [14] to override the violation. The shortest route with minimum
number of cells used is selected as the output for each net. If
all the shortest routes for some net do not satisfy the timing
constraint, or violate the fluidic constraint and fail the droplet-
motion modification, we put this net to a conflict list in order to
resolve it at the end. Note that all routes obtained in this stage
are not only edge-disjoint, but also vertex-disjoint (e.g., no two
droplet routes share the same cell).
In the second stage, for each net in the conflict list, we execute
the modified Lee algorithm to generate several shortest routes
without considering the previously routed paths in the first stage
as obstacles. From these shortest routes, we select a route that has
no common pair of adjacent cells with the previously obtained
routes as the output. Therefore, the resulting route may intersect,
but still be edge-disjoint, with the previously obtained routes.
As before, this route also needs to pass both TDCC and FCRC
constraints.
At this stage, we have obtained disjoint routes for all the
nets. If all the shortest routes for some net in this stage do not
pass either TDCC or FCRC constraint, placement refinement is
required to increase the corresponding routability.
C. Optimization of wash-droplet routing
Since cross-contamination can also happen across two sub-
problems, the droplet routes for the nets in the current sub-
problem should avoid sharing cells with the routes in the pre-
decessor sub-problem. Therefore, along with the active modules,
the routes for all nets in the predecessor sub-problem are also
treated as obstacles when routes in the current sub-problem are
generated. However, in order to bypass these additional obstacles,
some routes will become longer and may fail TDCC. Further-
more, a net that can be routed easily without these additional
obstacles may become unroutable. Therefore, after the droplet-
routing process in one sub-problem, a wash operation needs to
Authorized licensed use limited to: DUKE UNIVERSITY. Downloaded on July 26, 2009 at 22:58 from IEEE Xplore. Restrictions apply.
be introduced. In a wash operation, a wash droplet is routed
to traverse selected cells and remove residue from them. The
delay introduced by the wash operation is equivalent to the time
needed to route the wash droplet. For a sub-problem that includes
a follow-up wash operation, its droplet-routes will not be treated
as obstacles for the next sub-problem.
Suppose a bioassay can be decomposed into N droplet-routing
sub-problems. The parameter T
i
, 1 ≤ i ≤ N, is defined as the
maximum transportation time needed among the droplet routes in
sub-problem i if the wash operation is not performed after droplet
routing in sub-problem (i − 1). The parameter T
∗
i
, 1 ≤ i ≤ N,
is defined as the maximum transportation time needed in sub-
problem i if the wash operation is performed after the droplet
routing in sub-problem (i − 1).LetTw
i
, 1 ≤ i ≤ N,bethe
time needed for the wash operation after droplet routing in sub-
problem (i − 1). Therefore, (T
∗
i
+ Tw
i
) denotes the total droplet
transportation time that includes the wash operation after sub-
problem (i−1) and routing for sub-problem i. If a wash operation
is carried out between sub-problem (i − 1) and sub-problem i,
it adds to the routing time for sub-problem i, but it also frees
up more cells for routing and reduces the routing time in sub-
problem i. As a result, in most cases, T
∗
i
<T
i
. The decision
on whether to add a washing step between two sub-problems
therefore depends on the individual routing time for the sub-
problem and the nature of these routes.
We next describe an optimization model to determine when the
wash operation should be performed between successive routing
steps. Let x
i
be a binary variable defined as follows: x
i
=1if
the wash operation is performed between sub-problems (i − 1)
and i, otherwise x
i
=0. Our goal is to minimize the total time
needed for droplet routing for the sub-problems. The objective
function for the optimization problem can therefore be stated as
follows:
Minimize F = T
1
+
N
i=2
(x
i
(T
∗
i
+ Tw
i
)+(1− x
i
)T
i
).
We also need to incorporate the constraint that the number of
wash operations is less than the number of sub-problems. There-
fore, we now formulate the optimization problem as follows:
Minimize F, subject to:
N
i=2
x
i
≤ N −1. To solve this problem,
we simply note that to minimize F, x
i
needs to be set to 1 if
T
∗
i
+ Tw
i
<T
i
, otherwise x
i
=0.
V. E
XPERIMENTAL EVA L U AT I O N
In this section, we evaluate the proposed disjoint droplet-
routing methods for two real-life bioassays, namely multiplexed
in-vitro diagnostics on human physiological fluids, and DNA
sequencing.
A. Example 1: Multiplexed in-vitro diagnostics
The assay protocol is described as follows. Three types of
human physiological fluids, i.e., urine, serum and plasma, are
sampled and dispensed to the digital microfluidic biochip. Glu-
cose and lactate measurements are performed for each type of
physiological fluid.
The sequencing graph and the s chedule for the assay protocol
are presented in [14]. Note that one time-slot in the schedule is
set to 2 seconds. Module placement for different time-slots on a
S
3
S
1
S
2
R
1
R
2
Waste Reservoir
Time-slots 1 to 6
B
DI
1
DI
2
M
2
M
1
M
2
DI
3
Dt
2
Dt
1
M
3
M
4
Time-slots 6 to 9
Dt
2
Dt
1
S
M
5
M
6
S
Time-slots > 9
W
W
Fig. 1. Module placement for the multiplexed in-vitro diagnostics example.
M
1
R
1
Dt
1
DI
1
DI
2
DI
3
M
3
M
4
S
2
S
1
R
2
Waste Reservoir
(a)
(b)
(c)
S
3
B
(d)
D
1
D
2
D
3
L(Route 2)=35 > Td=20
Route 1
Route 3
Route 4
Route 5
Four Route
Intersections
L(Route 2)=18 < Td=20
Fig. 2. Disjoint routing and cross-contamination-oblivious routing for sub-
problem 3: (a) three 2-pin nets and two 3-pin nets; (b) Route 2 violates the timing-
delay constraint; (c) feasible routes for all the nets using the disjoint-routing
method; (d) routing results obtained using the cross-contamination-oblivious
routing method.
16×16 microfluidic array is shown in Fig. 1. M
1
-M
6
are mixers,
DI
1
-DI
3
are dilutors, S is a storage unit, Dt
1
and Dt
2
are two
detectors, S
1
-S
3
are reservoirs for three samples, R
1
and R
2
are
reservoirs for two reagents, B and W are reservoirs for buffer
and wash droplets, respectively.
The routing problem is decomposed into eleven sub-
problems [14]. We address these sub-problems serially by at-
tempting to determine the set of vertex-disjoint or edge-disjoint
droplet routes and minimize the number of cells used by these
routes, subject to both the timing and fluidic constraints. The
maximum delay constraint is set to 10% of one time slot, i.e.,
0.2 second. We assume a typical clock frequency of 100 Hz.
Therefore, the timing constraint Td is equal to 20 clock cycles.
Here we use sub-problem 3 to illustrate the proposed routing
method. As shown in Fig. 2(a), there are three 2-pin nets and
two 3-pin nets. Route 1 is defined as the path between Dt
1
and
M
1
(i.e., Net 1), Route 2 is defined as the path between DI
2
and
DI
3
(i.e., Net 2), Route 3 is defined as the path between DI
2
and the waste reservoir (i.e., Net 3). Route 4 is defined as the
Authorized licensed use limited to: DUKE UNIVERSITY. Downloaded on July 26, 2009 at 22:58 from IEEE Xplore. Restrictions apply.
path for 3-pin net whose pins are R
1
, DI
1
and M
4
(i.e., Net 4),
while Route 5 is defined as the path for 3-pin net whose pins are
R
2
, DI
1
and M
3
(i.e., Net 5). Module M
2
that is active during
this time interval is considered as an obstacle for routing.
First, we decide the order for routing these nets. The parameter
A in Equation (1) is set to 1. The priority number for Net 1 is
5, since the number of pins within its bounding box is 0 and the
maximum of its Xrange and Yrangeis 5. The priority number
for Net 4 is 13, since the number of pins within its bounding
box is 2 and the maximum of its Xrange and Yrange is 11.
The priority number for Net 2, Net 3 and Net 5 are 15, 3 and
15 respectively. Therefore, the net-routing order is Net 3, Net 1,
Net 4, Net 2 and Net 5.
While generating the route for Net 2, since the paths for
previous routed Net 3, Net 1 and Net 4 have been marked as
obstacles, the shortest paths between DI
2
and DI
3
(i.e., Route
2) violate the timing delay constraint, i.e., L(Route 2) = 35 (in
cells) >Td= 20 (clock cycles), as shown in Fig. 2(b). Thus,
we put Net 2 to the conflict list, then generate the route for
Net 5. After the route for Net 5 is generated, we generate the
route for Net 2 without considering previously generated paths
as obstacles. A route that has no common pair of adjacent cells
with the previously routed paths is selected. The obtained route
for Net 2 satisfies the timing constraint, i.e., L(Route 2) = 18
(in cells) <Td= 20 (clock cycles), as shown in Fig. 2(c).
The route for Net 2 induces two intersections with the routes
for Net 4 and Net 5 respectively. Therefore, the desirable edge-
disjoint droplet-route set with total 63 cells is finally obtained.
All routes satisfy both timing and fluidic constraints, as shown in
Fig. 2(c). Note that for droplets D
1
and D
3
in Route 4 and Route
5 respectively, fluidic constraint might be violated if they start
moving at the same time to their destinations. However, based on
the modification rules, we force D
1
to stay in the current location
until D
2
is transported to the mixing site and mixed with it, at
the same time we continue moving D
3
to its destination, thereby
overriding the constraint violation. Since the delay for Route 4
is determined by the transportation time for D
2
, no extra delay
is introduced during this process. The droplet-transportation time
for all the routes in this sub-problem is 19 clock cycles.
We compare the routing results obtained using the proposed
method for sub-problem 3 with a baseline method that utilizes the
modified Lee algorithm in [14] to solve a sub-problem without
considering cross-contamination between different routes. The
baseline method uses four route intersections, two more than that
obtained by the proposed method, as shown in Fig. 2(d). We refer
to an intersection of two routes as a cross-contamination site. The
number of cross-contamination sites can be used to evaluate the
likelihood of cross-contamination for a set of routes in a sub-
problem.
Consider an intersection of two droplet routes in a sub-
problem. When a droplet passes through the intersection, a wash
droplet is dispensed from the wash reservoir and transported via
this site to the waste reservoir. (We assume, without loss of
generality, that one wash droplet suffices; multiple wash droplets
can also be considered.) After that, the other droplet can be
transported via this cleaned site. Therefore, cross-contamination
between the above droplets is avoided. For the routing results in
Fig. 2(c) obtained using the proposed method, at first, droplets are
TABLE I
C
OMPARISON OF THE PROPOSED METHOD WITH THE BASELINE METHOD FOR
EXAMPLE
1(CONTAMINATION WITHIN SUBPROBLEM).
Sub- Result for Result for
problem proposed method baseline method
no. N
cs
N
cell
T
r
T
rw
N
cs
N
cell
T
r
T
rw
1 0461717 1 391755
2 1411247 1 411247
3 2631959 4 571897
4 06 5 5 06 5 5
5 0631717 3 551555
6 1571651 3 551652
7 0541818 1 511857
8 0241313 0 241313
9 0241414 0 241414
10 0141313 0 141313
11 0151414 0 151414
Total 4 407 158 268 13 381 155 422
Routes of sub-problem 7
as obstacles
(a) (b)
(c)
Fig. 3. Routes for sub-problem 7 with or without wash operation: (a) disjoint-
routing solution for sub-problem 7; (b) disjoint routes for sub-problem 8 when
no wash operation is performed before it; (c) disjoint routes for sub-problem 8
when a wash operation performed.
transported along Routes 1 and 3-5 at the same time. After that,
two wash droplets are dispensed from the wash reservoirs and
transported via the two cross-contamination sites to the waste
reservoir, respectively. After the wash operation, the droplet is
transported along Route 2 via the two sites that have already
been cleaned. The maximum droplet-transportation time with
washing steps is 59 clock cycles. For the routing results in
Fig. 2(d) obtained using the baseline method, the maximum
droplet-transportation time with washing steps (97 clock cycles)
is much higher than that for the proposed method. This is because
two wash operations must be inserted sequentially during droplet
routing to clean the residue on the four cross-contamination sites.
In Table I, for each sub-problem, we compare for the two
methods, the number of cross-contamination sites (N
cs
), the
number of cells used by the nets (N
cell
), and the maximum
droplet-transportation time for the nets with and without wash
steps in this sub-problem, i.e., T
rw
and T
r
. N
cs
=0denotes
the fact the routes obtained are vertex-disjoint. The results show
that the proposed method significantly reduces the overall routing
time, especially when wash droplets must be routed.
In Fig. 3, we illustrate the influence of cross-contamination
across two sub-problems on the droplet routing. Fig. 3(a) shows
the disjoint routing solution for sub-problem 7. If no wash
operation is performed between sub-problems 7 and 8, the
routes for all nets in sub-problem 7 are treated as obstacles
for sub-problem 8, as shown in Fig. 3(b). The disjoint routes
for sub-problem 8 are also shown in Fig. 3(b). The maximum
transportation time T
8
is 76 clock cycles. If a wash operation
Authorized licensed use limited to: DUKE UNIVERSITY. Downloaded on July 26, 2009 at 22:58 from IEEE Xplore. Restrictions apply.
