A Cost & Performance-Efficient Field-Programmable Pin-Constrained Digital Microfluidic Biochip
11 A Cost & Performance-EfficientField-Programmable Pin-ConstrainedDigital Microfluidic Biochip
Alireza Abdoli and Ali Jahanian
Abstract —Digital microfluidic biochips (DMFBs) constitutemodern generation of Lab-on-Chip (LoC) devices aimed at au-tomation, miniaturization and cost-affordability of biochemistryand laboratory procedures. Over the course of past few yearsthere have been various application-specific and general-purposeDMFBs aimed at reduced manufacturing costs; following thesame trend this study presents a general-purpose DMFB withhighly competitive characteristics compared with the state-of-the-art DMFBs. The proposed DMFB architecture provides lowerLayout/PCB fabrication costs thereby reducing the total manu-facturing costs. While more cost-affordable the proposed designis competitive with the state-of-the-art DMFB architectures.
Index Terms —Lab-on-Chip, Digital Microfluidic Biochip,Field-Programmable, Pin-constrained, Cost, Performance.
I. I
NTRODUCTION M ICROFLUIDIC biochips are modern revolutionary de-vices enabling a new paradigm in performing fluidicbio-chemistry and laboratory manipulations never existed be-fore; providing various ranges of applications among whichare DNA multiplexed Polymerase Chain Reaction (PCR) [1],in-vitro diagnostics [2] and protein crystallization [3] and DNAcomputing [6].The conventional bio-chemistry operations are mostly per-formed/controlled by human intervention. On the other handtraditional methods of accomplishing laboratory operationsrequire considerable amounts of experiment materials andreagents; which might be costly with experimental testing ofnew drugs. Yet, laboratory equipment consume large amountsof space thus a room might be dedicated to accommodationof the aforementioned equipment. Also, automated roboticlaboratory equipment cost much more than affordable to anend-user person.In order to address aforesaid issues the microfluidic biochipswere developed; these chips are mainly aimed at three crucialfactors of automation, cost-affordability and miniaturization.Microfluidic biochips are accompanied by a microcontrollerused for programming the chip to perform wide ranges oflaboratory procedures along with the bio-chemistry operationswithout any human interventions thus realizing automationfactor. These chips operate on the basis of manipulating
A. Abdoli is with the Department of Computer Science and Engineer-ing, Shahid Beheshti University, G.C., Tehran, 1983963113 Iran e-mail:[email protected]. Jahanian is with the Department of Computer Science and Engineering,Shahid Beheshti University, G.C., Tehran, 1983963113 Iran e-mail: [email protected] negligible amount of fluids thus providing significant cost-affordability in terms of experiment materials and reagentconsumption. Yet, these chips are manufactured at scalesresulting in much smaller area consumption as compared tothe conventional circuits [10] and laboratory equipment; thisimplies the miniaturization factor inherent in the microfluidicbiochips.A typical digital microfluidic biochip consists of two plates;the bottom plate is consisted of an array of equal-size elec-trodes while the top plate spans the bottom plate and acts asthe ground electrode. Figure 1 illustrates the top and cross-sectional view of a DMFB.The electrodes forming the array of electrodes are connectedto the pins of the microcontroller; the microcontroller can beprogrammed in order to turn on/off the electrodes such thatthe intended bio-assay is realized. Droplets are sandwiched inbetween the top and bottom plates; at the bottom plate there areelectrodes on which droplets would be actuated. Consideringthe structure of the bottom plate on top of electrodes there is adielectric layer; additionally a hydrophobic layer is placed ontop of the dielectric layer. The hydrophobic layer is used forfacilitating movement of droplets on the surface of the DMFB.In order to ease movement of droplets further the spacebetween the top and bottom plates is filled with some fillerfluid, typically silicon oil, which allows for better movementof droplets on the array of electrodes.This paper presents a low-cost yet general-purpose digitalmicrofluidic biochip; the proposed design is considerablysmaller than the previous state-of-art designs which in turnresults in significant improvements in terms of total numberof electrodes and control pins used for driving electrodes,along with cheaper fabrication costs; also the proposed im-provements in the hardware design results in shorter dropletrouting times thus improving the overall bio-assay completiontimes.The rest of the paper is organized as follows: section II isdevoted to review of underlying technologies and the designflow associated with digital microfluidic biochips. Section IIIprovides literature review of the previous works on DMFBdesigns. Section IV initially reviews the original general-purpose field-programmable DMFB design on the basis ofwhich the enhanced design proposed in this study is estab-lished. Section V is devoted to presentation of simulationresults in comparison with other previous notable designs.Eventually, section VI concludes the paper. a r X i v : . [ c s . ET ] A ug (a) (b) Fig. 1: (a) Cross-sectional and (b) top view of a closed DMFBII. DMFB TECHNOLOGY AND DESIGN FLOWThis section initially reviews the fundamental technologiesrelated with digital microfluidic biochips, then proceeding tothe DMFB design flow.
A. DMFB Technology Overview
Digital microfluidic biochips are based on the electro-wetting on dielectric (EWOD) phenomenon [4]; which is theelectromechanical actuation (wetting) of conductive fluids ona solid surface through electrical bias [5]. As a result ofwhich droplets are actuated by applying appropriate level ofvoltage to the desired electrode; thus creating an electrical fieldaffecting the droplet over the activated electrode. This happensbecause of the tension between the droplet and the electrode;Figure 2 depicts the electro-wetting on dielectric phenomenon.As depicted in Figure 2 in case no voltage is applied thedroplet remains in its normal form whereas in case appropriatelevel of voltage is applied the droplet is polarized; this phe-nomenon can be put to work for moving the droplets on thearray of electrodes. It must be noted that currently dropletscan be actuated to neighboring (top, bottom, left and right)electrodes; though droplets currently cannot be actuated todiagonally adjacent electrodes.A typical DMFB is capable of various microfluidic opera-tions including dispensing, transporting (movement), merging,mixing, and also detection/heating; Figure 3 depicts variousfundamental microfluidic operations.The most fundamental microfluidic operation is holding(storing) the droplets which is achieved by activating theelectrode beneath the droplet. On a large DMFB at any giventime there might be several droplets on the DMFB; whichmust be held steady for the duration of their presence on theDMFB.Second fundamental microfluidic operation is transporting(moving) the droplets on the array of electrodes. As statedearlier, droplets can be moved to adjacent (top, bottom, left orright) electrodes. This is achieved by deactivating the electrodebeneath the droplet and activating the desired neighboringelectrode. (a) (b)
Fig. 2: Electrowetting on Dielectric phenomenon Third fundamental microfluidic operation corresponds tomerging two droplets into a single larger droplet. Initiallythe two droplets are moved near each other; in this casethere is a distance of one electrode between the two droplets.Then the electrode between the two droplets is activated whileat the same time deactivating the two electrodes holdingdroplets. This causes both droplets to be moved towards thejust activated electrode; thus merging the two droplets into asingle larger droplet.Fourth fundamental microfluidic operation is splitting asingle droplet into two smaller, ideally equal sized, droplets.This is accomplished by activating neighboring electrodes(top/bottom or left/right) while at the same time deactivatingthe electrode beneath the droplet. This splits the droplet intotwo smaller, ideally equal sized, droplets.Fifth fundamental microfluidic operation isheating/detection/cooling which requires availability ofexternal equipment in the DMFB. Given the architecture ofthe DMFB external heaters/detectors/coolers are affixed ontop of designated electrodes during the manufacturing processso that enabling aforementioned capabilities.
B. DMFB Synthesis Flow
Performing any bioassay on the DMFB involves variousstages referred to as synthesis flow of the DMFB. Initially, theprocess is initialized by inputting the protocol of the bioassayand also architecture specifications of the DMFB.The protocol of a given bioassay is in the form of a directedacyclic graph (DAG). On the other hand, the architecturespecification of the DMFB incorporates information on di-mensions of the array of electrodes, locations of I/O reservoirson the periphery of the array of electrodes and also locationof any fixed modules (e.g. detectors/heaters/coolers). Figure 4illustrates several stages of the DMFB synthesis flow.
1) Scheduling:
Given the protocol of the bioassay and alsothe architecture specifications the synthesis flow is commencedby scheduling microfluidic operations within the bioassayprotocol. Scheduling is the first stage of the synthesis flowduring which every microfluidic operation is assigned withexact start and end times. The scheduling algorithm mustmake the best use of available resources to ensure microfluidicoperations are scheduled as efficient as possible thus producingthe shortest overall scheduling time.
2) Placement:
Following the scheduling stage every mi-crofluidic operation has exact start and end times. Next,the scheduled operations must be placed on the array ofelectrodes according to specific resource type required bythe operation. The placement algorithm must be performedsuch that all scheduled operations during any given time-stepare successfully placed on the array of electrodes. There aretwo main categories of placement algorithms; namely, freeplacement and fixed placement algorithms.In free placement the algorithm keeps account of availablespaces on the array of electrodes and attempts to find appro-priate area to place the operation on the array of electrodes.On the other hand, in case of fixed placement locations ofmodules on the array of electrodes are already specified; thus (a) Storage (b) Transportation (c) Merging (d) Splitting (e) Mixing (f) Heat/Detect
Fig. 3: Fundamental microfluidic operations (Please note that light gray electrodes are active whereas dark gray electrodes aredeactivated)there is no need to search for available areas. This reduces theplacement to a binding problem; in which operations are boundto first available module. The free placement is more time-consuming computationally; yet, free placement of modulesmight be performed such that there is no space for movementof droplets in between the placed modules thus causing dropletblockages and deadlocks.In case of fixed placement given the already fixed loca-tion of modules there are dedicated spaces for routing ofdroplets between the modules thus guaranteeing successfuldroplet routing and eliminating the possibility of blockagesand deadlocks associated with free placement algorithms.
3) Droplet Routing:
There are several scenarios involvedin the droplet routing stage; dispensing droplet into modules,transportation of droplet between the modules and outputtingof droplet from modules to output reservoirs. The dropletrouting algorithm must operate such that the shortest possibleroute is produced. Also, the task must be accomplished suchthat the droplet does not interfere with other droplets alreadypresent on board. Furthermore, droplets must be routed suchthat not any two droplets collide with each other while beingrouted to their respective destinations.
4) Pin-Mapping:
Early DMFBs applied direct-addressingscheme [13] in which every single electrode is dedicated withan independent control pin; this provides highest degree offlexibility and controllability, yet the scheme requires highestnumber of control pins. As an illustration a DMFB of size m × n requires m × n control pins which might be excessivelylarge in case of large DMFBs.A cheaper alternative to direct-addressing scheme would becross-referencing scheme [14]; in which every single columnand row is assigned with a controlling pin; thus drivingan array of size m × n requires m + n control pins. Yet,simultaneous activation of multiple rows and columns might cause unintended droplet movement.Figure 5 shows schematic of direct-addressing and pin-mapping schemes for 4 ×
5) Wire Routing:
The wire routing stage deals with thewirings from microcontroller pins to the array of electrodes.Given the type of pin-mapping the amount of wire routingwould vary. This directly affects the number of metal-layersused toward the wire routing stage; thus significantly affectingFig. 4: Synthesis flow of a typical digital microfluidic biochip (Please note that letter I denotes input while letter O denotesoutput reservoirs; also letter M represents mixing operations) (a) (b)
Fig. 5: (a) direct-addressing (b) pin-constrained pin-mappingschemesthe overall manufacturing cost of the DMFB.III. P
REVIOUS W ORKS
This section reviews major previous works addressing var-ious recent DMFB architectures and algorithms. Over thecourse of past few years there have been numerous studieson DMFBs.Xu et al. [13] proposed a broadcast-addressing DMFB; theirproposed design encompasses a multi-function pin-constrainedDMFB capable of executing a predefined set of bioassays. Luoet al. [18] proposed a pin-constrained pin-mapping schemeallowing concurrent movement of two droplets on the DMFB;their proposed design is targeted towards performing a prede-fined set of bioassays.Keszocze et al. [19] proposed a DMFB synthesis based ontheir general and exact routing methodology. Their proposeddesign results in significantly low number of control pins.While demanding small number of control pins their proposeddesign is solely applicable to small to mid-size bio-assays (e.g.PCR, In-vitro diagnostics); this is because the computationaltime for large complex bio-assays such a protein-split wouldbe prohibitively large.Wille et al. [20] proposed a one-pass synthesis scheme withmuch faster computation times compared with [19]; fastercomputation times allow for performing large bio-assays suchas protein-split not computationally feasible in [19]. Althoughmuch more efficient in terms of computation times the runtimeof the produced solutions is approximate and considerablyvariable among different runs.Abdoli et al. [8] [24] proposed a field-programmable pin-constrained design; their proposed design is considerablysmaller than previous pin-constrained designs also yieldingreduced overall number of electrodes and controlling pins.This paper enhances the DMFB design in [8]. Also, Abdoli etal. [9] proposed their architecture with a cellular structure pro-viding a regular expandable structure; their proposed structureis inspired by the widely popular FPGA devices.Grissom et al. [11] proposed a field-programmable pin-constrained design aimed at general-purpose bioassay execu-tion. Also, Grissom et al. [12] proposed their enhanced low-cost pin-constrained DMFB design so that requiring less num-ber of metal-layers towards wire-routing of the DMFB; thus,yielding significantly reduced overall manufacturing costs.IV. P IN -C ONSTRAINED S CHEME
The architecture layout and pin-mapping scheme have agreat impact on the overall manufacturing costs of the design. The field-programmable pin-constrained designs [8] [9] [11][12] employ a layout and scheme such that to provide general-purpose bioassay execution along with reduced pin-count sothat the overall manufacturing cost of the DMFB is signifi-cantly reduced. Given the considerable number of electrodesallocated to droplet routing paths then efficient addressing ofrouting paths is of significant importance in terms of pin-countand overall DMFB manufacturing costs.A typical general-purpose pin-constrained DMFB designconsists of mixing modules, storage/split/detection (SSD)modules and droplet routing paths towards movement ofdroplets; simulation pin-mapping scheme of the proposedDMFB design is depicted in Figure 6.Looking at Figure 6 there are 4 mixing modules and 8SSD modules; also there are two vertical routing paths at theleft and right of the design; used towards inputting/outputtingdroplets into/out of I/O reservoirs. Also, there are two hori-zontal routing paths at the top and bottom of the design; whichallow for inputting/outputting droplets into/out of mixing andSSD modules. Mixing modules are used for merging andmixing of droplets; while SSD modules are used for storing,splitting and detection of droplets.Description PinsRouting PathsLeft Vertical Routing Path 1 − − − − − − − − − A. Pin-Constrained Operation Execution
A typical general-purpose pin-constrained DMFB must becapable of performing various microfluidic operations in-volving droplet dispensing/outputting, merging, mixing andsplitting/heating/detection.
1) Droplet Dispensing/Outputting:
The typical bioassay ex-ecution on a DMFB initially involves droplets being dispensedfrom I/O reservoirs on the perimeter of the array of electrodes.Each I/O reservoir has an individual electrode leading a dropletto the edge of the array of electrodes [11]; since common toall DMFB designs these electrodes are omitted from DMFBdesigns.
2) Droplet Merging/Mixing:
Droplet merging operationwithin pin-constrained DMFB designs is performed insidemixing modules; during which initially the first droplet fluidis moved into an available mixing module and then moved tothe Hold pin of the module; then the second droplet is movedto the I/O electrode of the same module containing the firstdroplet while at the same time moving the first droplet offthe Hold electrode towards the I/O electrode. Moving the firstand second droplets to the I/O electrode causes merging of thetwo droplets. After successfully merging the two droplets theresulting droplet is moved back to the Hold electrode of themixing module; this completes the merging operation. Thenthe droplet is rotated around the module for a certain periodof time so that contents of the droplet are perfectly mixed;this accomplishes the mixing operation.
3) Storage/Splitting/Detection:
Sometimes it might happenthat droplets produced by some operations need to wait on-chip for certain number of time-steps before being usedtowards other microfluidic operations. SSD modules can beused for storing droplets; every SSD module is capable ofstoring one droplet at any time-step.Furthermore, SSD modules can be used for splitting adroplet into ideally two equal size droplets. This is achievedby moving a droplet onto the I/O electrode of an available SSDmodule; then the I/O electrode is deactivated while at the sametime activating the Hold electrode of the SSD module andalso the routing electrode leading to the SSD module. Thisideally causes the droplet to be split in half thus producingtwo smaller droplets. Then one of the droplets is stored inthe current SSD module while the second droplet is moved toanother available SSD module. Additionally, in case equippedwith external detector/heater/cooler the SSD module can beused for performing detection/heating/cooling operations.
B. Pin-Constrained Droplet Routing
Droplet routing is consisted of a set of tasks in order tomove droplets from input reservoirs to mixing/SSD modules,between mixing and SSD modules and from mixing/SSDmodules to output reservoirs. Given the considerable numberof electrodes devoted to routing paths using an individual pinper electrode requires large pin-count which in turn requiresadditional hardware for driving routing path electrodes. Inorder to reduce the pin-count associated with routing pathsthe 3-phase routing path is used [21]. The 3-phase routing,as the name implies, requires only 3 pins for addressing any routing paths of arbitrary length. It must be noted that the 3-phase pin-mapping can be used in case of intersecting routingpaths; however, this requires using different pin numbers foraddressing intersecting routing paths in order to avoid conflictsand unintended droplet movements. Looking at Figure 6 the 3-phase routing is used for pin-assignment of routing paths in thedesign. As stated earlier, different pins are used for differentrouting paths to avoid conflicts and unintended movement ofdroplets.V. THE PROPOSED DMFB DESIGN FLOWThis paper enhances the field-programmable pin-constrainedDMFB design originally proposed in [8]. The original designoffered numerous advantages compared with previous designsamong which are smaller array dimensions, lower number ofelectrodes and also reduced number of controlling pins. Lowerdimensions results in significantly lower droplet routing timessuch that the total bioassay execution times are considerablyreduced.
A. The Proposed DMFB Design
This section introduces several improvements comparedwith the original GFPC DMFB design [8] which makes theproposed design competitive to the state-of-art pin-constrainedDMFB designs.
1) Improved Pin-Mapping For Routing Paths:
The originalGFPC design used pin numbers 1-3 for vertical routing pathsand pin numbers 4-6 for horizontal routing paths; whileefficient in terms of pin-count this complicates the wire-routing of the original GFPC design which results in increasednumber of metal-layers used towards wire-routing of thedesign. Considering the significance of pin-mapping on thewire-routing stage of the DMFB synthesis flow the proposeddesign utilizes a revised pin-mapping scheme towards routingpaths; in which every routing path is assigned with 3 dedicatedpins. While increasing the pin-count this greatly simplifiesthe wire-routing of the proposed DMFB design which in turnresults in decreased number of metal-layers used towards wire-routing of the design. The revised pin-mapping of the proposeddesign for routing paths is illustrated in Figure 6.
2) Improved Pin-Mapping For Mixing Modules:
The origi-nal GFPC design [8] allocated a single shared pin to everyelectrode in the mixing modules. The scheme required 6pins to address the electrodes inside mixing modules; yet theproposed pin-assignment method for mixing modules reducesthe number of mixing pins from 6 to only 4 pins. Figure 7depicts the pin-mapping of mixing modules in [8] versus thepin-mapping of this study. As illustrated in Figure 7 (a) 6 pins(7 −
12) are used for mixing pins; whereas the enhanced pin-mapping proposed in this study requires only 4 pins (7 − (a) Original GFPC (b) Proposed Design Fig. 7: Pin-assignment of mixing modules in the originalGFPC design [31] compared with the proposed design(that is because as illustrated in Figure 7 (b) activating pins7 and 9 causes redundant electrode actuations). Thus powerconsumption of mixing modules in the proposed design isincreased by 25% compared with the original GFPC design[8].
3) Reduced Overall Manufacturing Costs:
Given the im-proved pin-assignment of the proposed design the wire-routingof the design is improved such that the total number of metal-layers required for wire-routing of the proposed design isreduced from 5 to 3 layers; while retaining the functionalitiesof the original GFPC design. This reduction in the total num-ber of metal-layers directly affects the overall manufacturingcost of the design. Figure 8 is devoted to the wire-routing ofthe original GFPC design [8] while Figure 9 illustrates theimproved 3 layers wire-routing of the proposed design in thisstudy.
4) Improved Fault-Tolerance:
Given the significant pin-count reduction provided with pin-constrained DMFB designsthe level of flexibility is negatively affected. The proposeddesign allocates different ranges of pin numbers for addressingrouting paths throughout the design. Though, allocating differ-ent pins for different routing paths increases the total pin-counthowever it significantly improves the flexibility of the design.Because of the availability of various different routing pathsa faulty electrode in a given routing path can be tolerated and (a) Overall (b) Layer 1(c) Layer 2 (d) Layer 3(e) Layer 4 (f) Layer 5
Fig. 8: Wire-routing of the original GFPC architecture [8] withorthogonal capacity of 2 (a) Overall (b) Layer 1(c) Layer 2 (d) Layer 3
Fig. 9: Wire-routing of the proposed designbypassed through other routing paths.Abdoli et al. [22][23] investigated fault-tolerance of theoriginal GFPC DMFB design. A major advantage of the designcompared with previous pin-constrained DMFB designs lieswith the availability of various routing paths; which allowsbypassing any faulty electrodes hindering the way of droplettowards the intended destination.
5) Improved Power Consumption in Routing Paths:
Theproposed design enjoys improved power consumption withindroplet routing paths. The original GFPC design [8] assignedtwo distinct 3-phase routing paths (i.e. total pin-count of 6) foraddressing all droplet routing paths throughout the design. Incase of the proposed design a distinct3-phase pin-mapping isdedicated to every individual routing path (i.e. total pin-countof 12 as illustrated in Figure 6); albeit the increased pin-count,assigning distinct 3-phase pin-mapping to individual routingpaths results in decreased power consumption within dropletrouting paths. The improved pin-mapping applied to dropletrouting paths of the proposed design results in at least 50%in terms of power consumption within droplet routing pathscompared with original GFPC [8] design.
6) No Need for Routing Buffer (RB) Module:
Prior general-purpose pin-constrained designs in [8] [9] [11] [12] allocateda routing buffer (RB) module for bypassing possible mutualdroplet routing deadlocks. Since designs in [11] [12] used asingle routing path for droplet routing it is necessary to allocatean RB module to resolve possible routing deadlocks. In case ofthe proposed DMFB design since there are two groups of topand bottom mixing/SSD modules connected by vertical dropletrouting paths at the left and right of the design there is no needto allocate separate resources in the form of routing buffermodule. This saves design space and pin-count; eliminatingthe RB module reduces the total pin-count by 2.
B. Synthesis Flow of the Proposed DMFB Design
This section briefly discusses various algorithms used to-wards synthesis flow of the proposed DMFB design.
1) Scheduling:
For the scheduling stage of the proposeddesign the list scheduling algorithm [2] a fast however greedyalgorithm is applied.
2) Placement:
Given the already fixed location of modulesthe placement stage of the design flow is reduced to a bindingproblem during which microfluidic operations are bound tomodules with already-fixed locations. The binding phase of theproposed design is performed using Grissom left-edge bind-ing algorithm for field-programmable pin-constrained DMFBdesigns [7].
3) Droplet Routing:
The droplet-routing stage of the pro-posed design works on the basis of the sequential dropletrouting algorithm originally proposed for the GFPC DMFBdesign [8]; the revised pin-mapping of the proposed designdiffers with the original GFPC design in that the routingpaths connecting the upper groups of mixing/SSD module tothe lower group is omitted from the proposed design. Omit-ting inter-module droplet routing paths significantly simplifieswire-routing stage of the design flow. Though, the minordifference the overall structure still remains the same thus thesequential routing algorithm proposed for the original GFPCdesign still applies the proposed DMFB design.
4) Pin-Mapping:
The pin-mapping of the proposed designis depicted in Figure 6 for 11 ×
11 architecture. The leftvertical routing column is allocated with pins 1-3. Next, theright vertical routing column is assigned with pins 7-9. In caseof horizontal routing columns the pins 10-12 are assigned (ascan be seen in Figure 6), the proposed design accommodatestwo horizontal routing columns; intended for providing accessto mixing/SSD modules located at the top and bottom tiers ofthe proposed design. Next, pin-mapping of mixing modulesis addressed; in which initially shared pins are assigned withpins 13-16. As discussed earlier pin-mapping of the proposeddesign solely requires 4 pins for addressing shared pins insidemixing modules; then every mixing module must be allocatedwith two independent pins for addressing I/O and Hold pins ofthe module. Looking at Figure 6 it can be seen that pins 17-20are used for addressing Hold pins of mixing modules whilepins 20-23 are allocated towards I/O pins of mixing modules.Given allocation of pins to mixing modules then the processcontinues with pin-mapping of SSD (Storage/Split/Detection)modules; which merely consists of two pins (I/O and Holdpins). As can be seen in Figure 6 there are 8 SSD moduleswith pins 25-32 assigned to I/O pins and pins 33-40 assignedtowards Hold pins.
5) Wire Routing:
Given the improved pin-mapping of theproposed design compared with the original GFPC design [8]the wire-routing stage of the design flow is simplified so thatthe total number of metal-layers is reduced to 3 layers. Thewire-routing stage of the proposed DMFB design utilizes thenegotiated-congestion wire-router algorithm [17].VI. HARDWARE COST ANALYSISThe overall manufacturing cost of a typical DMFB isaffected by various factors among which are: • Dimensions of array of electrodes • Number of metal-layers • Pin-countIn this section it is attempted to provide detailed costinformation on manufacturing costs of the proposed DMFB design while at the same time comparing with the state-of-art designs already available; this helps to show how theproposed DMFB design retains original capabilities while atthe same time remaining competitive with the state-of-artDMFB designs [8] [12].The original FPPC DMFB design [11] proposed a field-programmable pin-constrained DMFB design capable general-purpose bioassay execution. The Enhanced FPPC DMFB [12]proposed by Grissom et al., is a general-purpose DMFB designrequiring a one/two metal-layer(s) towards wire-routing; thenumber of metal-layers in [12] varies given the orthogonalcapacity parameter applied during the wire-routing stage. Incase of orthogonal capacity equal to 2 the wire-routing stageyields a two metal-layers wire-routing. Figure 10 illustratesthe two metal-layers wire-routing of their proposed DMFBdesign.Table I shows a summary on characteristics and wire-routingdetails of various general-purpose DMFB designs. The columnName denotes name of the DMFB design; while column WRAlg. refers to the wire-routing algorithm applied to the DMFBdesign. The column Array Dim., along with sub-columns Xand Y show array dimensions of the DMFB design. Columns
TABLE I: Illustrating number of wire-routing metal-layers forvarious orthogonal capacities
DMFB DetailsArrayDims.
DMFB design in this work and also higher pin-count comparedwith the present work the overall manufacturing cost ofthe proposed DMFB design would be competitive with theEnhanced FPPC DMFB design.To our best knowledge authors in [12] for the first time everattempted to provide detailed information on overall manufac-turing costs of DMFB designs. Considering the overall cost ofa typical DMFB design the first key element is the number ofmetal-layers used for wire-routing. The second and third keyelements are architecture dimensions and pin-count number,respectively. Larger architecture dimensions mean higher PCBcosts; on the other hand higher number of pin-count requiresmore equipment for driving those extra pins which again costsextra charges to the overall manufacturing costs of the DMFB.The authors in [12] used Advanced Circuits online instantquote feature [25] to obtain estimations on the cost of PCBaccording to dimensions of the architecture. Also, accordingto [12] it is assumed that DMFBs are driven by an Atmega1284 microcontroller with 32 general-purpose I/Os (GPIOs)[26]. In case a DMFB design requires more than 32 pins thenadditional circuitry is needed which accomplished by daisy-chaining arbitrary number of Fairchild 74VHC595MTC 8-bitshift registers [27]; quantities of 2500 of the aforementionedshift register can be purchase for $0.14 per unit from Mouser[28]. The following equation obtained from [12] shows thenumber of shift registers required for driving any DMFBdesigns.
Shif tRegs = (cid:24) numP ins − (cid:25) numPins > otherwise. (1)Also, according to [12] the overall manufacturing cost of atypical DMFB is obtained using the following equation Cost
W R = Cost
P CB + Cost SR (2)According to above equation wire-routing cost of a typicalDMFB is directly affected by the cost of PCB plus the costof additional circuitry in the form of shift registers used fordriving pins; additional circuitry is needed in case numberof pins is higher than the number to be accommodated bythe microcontroller so requiring shift registers. According to [25], using larger feature sizes tends to reduce the PCB cost.Equation 3 shows the elements effective on the PCB cost. Cost
P CB =( numLayers, width P CB , height
P CB , width
W T ) (3)Figure 11 shows the DMFB layout for PCB size estimation.As can be seen the array of electrodes is surrounded by a 0.5inch perimeter of empty space; also the PCB width is extendedto accommodate as many shift registers as necessary [12].Table II defines parameters used in Figure 11. As illustratedin the figure parameters width PCB and height
PCB denote widthand height of the PCB. Also, parameters width A and height A represent width and height of the array of electrodes. Theparameters width SR and height SR correspond to width andheight of the shift registers; additionally, parameter width SRS represents the spacing between shift registers.Equations 4 and 5 calculate the PCB dimensions; also, thePCB space devoted to accommodation of shift registers iscalculated through Equation 6. According to [12] shift registersare stacked vertically and in case necessary the width of PCBis increased to accommodate another column of shift registers. height
P CB = height Array + 1( inch ) (4) width P CB = width Array + width P CB − SR + 1( inch ) (5) (cid:24) Shif tRegs (cid:98) height
P CB / ( height SR + width SRS ) (cid:99) (cid:25) × ( width SR + width SRS ) (6)Table III indicates PCB price estimates for varying numberof layers and parameters. The figures have been obtained fromAdvanced Circuits online quote [25]; the online quote systemallows for specifying various parameters. For the sake of thisstudy solely parameters of width and height of PCB, trace/sizespace, via size and number of layers were specified; leavingother parameters at their default values.Looking at the figures provided in Table III PCB priceestimates are provided for three electrode sizes of 1, 2 andFig. 11: DMFB layout for PCB size estimation [12] TABLE II: PCB Parameters
Feature SymbolPCB Width width
PCB
PCB Height height
PCB
Array of Electrodes Width width
Array
Array of Electrodes Height height
Array
Shift Register Width width SR Shift Register Height height SR Shift Register Spacing Width width
SRS
ElectrodePitch Advanced CircuitMetrics 2 ” × O . C a p TABLE IV: Cost estimates for various architecture sizes ofEnhanced FPPC and the proposed Enhanced GFPC DMFBdesigns
DMFB Details Cost ($)DMFBName wire-routing of the DMFB.Interpreting results obtained in Table IV is seen that in caseof 4 module variation of enhanced FPPC and the proposeddesign the enhanced FPPC design provides lower overall PCBcosts given the lower layer-count; though, as the designsenlarge the proposed design would require smaller area andlower pin-count. Thus, in case the Enhanced FPPC design with8 mixing modules and the proposed design with 6 modules thehigher pin-count in the Enhanced FPPC design necessitates5 shift registers for driving the controlling pins not to beaccommodated by the microcontroller directly. Yet, in caseof the proposed design the lower pin-count leads to lowernumber of shift registers thus incurring lower costs in termsof shift registers. The lower cost of shift register in case of theproposed design yields an overall cost lower than the enhancedFPPC design.VII. PERFORMANCE SIMULATION RESULTSThe performance simulation results are conducted using theUCR Static Synthesis Simulator [7], an open-source DMFBframework, provided by the researchers at the University ofCalifornia Riverside.Table V details performance simulation evaluation of con-ducting various bio-assays on the DMFBs. The bio-assaysinvolve PCR (Polymerase Chain Reaction), In-vitro diagnos-tics (with variable number of samples and reagents), andProtein split; all of the bio-assays are available within theUCR SSS framework [7]. The column Name illustrates thename and type of the bio-assay. Column Scheduling showsthe scheduling time of the bio-assay; while column Routing isdevoted to the droplet routing times of the bio-assays. ColumnTotal is computed as the sum of the values of Schedulingand Routing columns which stands for the total time taken toperform the bio-assay.Looking at the figures provided in Table V it is seen thatthe scheduling times of the enhanced FPPC is superior in caseof In-vitro 2 to In-vitro 5 bioassays; thereby outperformingthe proposed design by 3 percent. Next, considering dropletrouting times it is noted that the proposed design performsremarkably lower droplet routing times compared with theenhanced FPPC design; this is due to architecture of theproposed design and also smaller size of the architecturecompared with the enhanced FPPC design. TABLE V: The performance simulation results (excludingwash droplets) of various bio-assays on the Enhanced FPPC(FP), the Proposed Design (PD)
Name Scheduling(s) Routing(s) Total(s)FP PD FP PD FP PDPCR 11 11 2.2 1.7 13.2 12.7In-vitro 1 14 14 3.4 1.9 17.4 15.9In-vitro 2 16 18 5.0 2.8 21.0 20.8In-vitro 3 16 18 7.6 4.5 23.6 22.5In-vitro 4 18 19 10.1 6.4 28.1 25.4In-vitro 5 21 25 13.2 9.1 34.2 34.1Protein split 1 52 52 2.7 1.6 54.7 53.6Protein split 2 62 62 8.4 7.4 70.4 69.4Protein split 3 83 83 18.2 16.0 101.2 99.0Average - 3% + 27% + 3%
TABLE VI: The performance simulation results (includingwash droplets) of various bio-assays on the Enhanced FPPC(FP), the Proposed Design (PD)
Name Scheduling(s) Routing(s) Total(s)FP PD FP PD FP PDPCR 11 11 2.2 1.7 13.2 12.7In-vitro 1 14 14 3.4 1.9 17.4 15.9In-vitro 2 16 18 5.0 2.8 21.0 20.8In-vitro 3 16 18 7.6 4.5 23.6 22.5In-vitro 4 18 19 10.1 6.4 28.1 25.4In-vitro 5 21 25 13.2 9.1 34.2 34.1Protein split 1 52 52 2.7 1.6 54.7 53.6Protein split 2 62 62 8.4 7.4 70.4 69.4Protein split 3 83 83 18.2 16.0 101.2 99.0Average - 3% + 27% + 3%
VIII. C
ONCLUSION