Digital Signal Processing for Molecular Communication via Chemical Reactions-based Microfluidic Circuits
11 Digital Signal Processing for MolecularCommunication via Chemical Reactions-basedMicrofluidic Circuits
Dadi Bi,
Student Member, IEEE,
Yansha Deng,
Member, IEEE,
Abstract —Chemical reactions-based microfluidic circuits areexpected to provide new opportunities to perform signal process-ing functions over molecular domain. To realize this vision, in thisarticle, we exploit and present the digital signal processing capa-bilities of chemical reactions-based microfluidic circuits. Aimingto facilitate microfluidic circuit design, we describe a microfluidiccircuit using a five-level architecture: 1) Molecular Propagation;2) Chemical Transformation; 3) Microfluidic Modules; 4) Mi-crofluidic Logic Gates; and 5) Microfluidic Circuits. We firstidentify the components at Levels 1 and 2, and present howtheir combinations can build the basic modules for Level 3. Wethen assemble basic modules to construct five types of logic gatefor Level 4, including AND, NAND, OR, NOR, and XOR gates,which show advantages of microfluidic circuits in reusability andmodularity. Last but not least, we discuss challenges and potentialsolutions for designing, building, and testing microfluidic circuitswith complex signal processing functions in Level 5 based on thedigital logic gates at Level 4.
I. I
NTRODUCTION
Molecular communication (MC) employs chemical signalsto exchange information and is considered as a promisingmethodology to facilitate a diversity of applications, rang-ing from personalized healthcare to manufacturing industry.Although the literature on the design of MC systems hasgrown considerably over the past few years, existing work hasbeen mostly theoretical in nature, and functioning MC testbedimplementation is few. In particular, [1]–[3] have developedfunctioning MC prototypes, where transmitted bit sequencewas modulated to concentrations of alcohol molecules [1],odor molecules [2], and protons [3]. It is noted that theirsignal processing functions on chemical concentration signalsare all achieved via external electronic devices, such as electricspray, odor emitter, and Arduino controlled LED. However,the utilization of electronic devices can hardly meet the bio-compatible, non-invasive, and size-miniaturized requirementsof biomedical-related applications.The above limitations of electronic devices inspire us to de-sign novel MC devices to perform signal processing functionsdirectly over chemical signals rather than electrical signals.In general, signal processing functions can be realized overmolecular domain in a twofold fashion, namely, 1) geneticcircuits [4] in engineered living cells, and 2) chemical circuits[5] based on “non-living” chemical reactions. Genetic circuitsengineer cell behaviors by embedding synthetic gene networks
D. Bi and Y. Deng are with the Department of Engineering,Kings College London, London WC2R 2LS, U.K. (e-mail: { dadi.bi, yan-sha.deng } @kcl.ac.uk). (Corresponding author: Yansha Deng). to produce desired responses. While genetic circuits offer bio-compatibility, non-invasiveness, and miniaturization, they stillcurrently face challenges, such as slow speed, unreliability,and nonscalability, which motivate us to use chemical circuitsas an alternative approach. Chemical circuits execute signalprocessing functions by means of chemical reaction networks(CRNs). A CRN is defined as a finite set of chemical reactionscomprising a finite number of species. These reactions occur ina well-stirred environment to realize a function or an algorithmvia mass action kinetics.The CRNs can integrate with microfluidic systems toconstruct chemical reactions-based microfluidic circuits. Amicrofluidic system processes or manipulates small amountof fluids using channels at dimensions of tens to hundredsof micrometers [6]. Therefore, chemical reactions-based mi-crofluidic circuits are not only endowed with advantages ofrapid analysis and low reagent costs due to size reduction, butcan also benefit from an additional space level of chemicalcontrol through applying and regulating chemical reactions indifferent regions of a microfluidic device.The concept of using chemical reactions-based microfluidiccircuits to execute signal processing functions was first intro-duced and investigated in [7], [8], where an MC transceiverwas designed to successfully realize binary concentration shiftkeying (CSK) modulation and demodulation functions. Know-ing the advantages of microfluidic circuits, we envision theutilization of microfluidic circuits to perform more complexsignal processing functions for MC.In nature, many cellular signaling processes can be inter-preted to be driven by digital signals with two discrete statesaccording to whether signals are stronger than thresholds. Atypical example is the bacterial coordination behavior afterthe detection of a sufficiently high autoinducer concentrationrepresenting bit-1 from a low autoinducer concentration repre-senting bit-0. In effect, digital signals are ideal for reliable statetransitions and signal integration, and are useful for decision-making circuits [9]. Furthermore, digital circuits can be easilyscalable and are popular in wireless signal processing. Mo-tivated by these facts, in this article, we exploit the abilityof chemical reactions-based microfluidic circuits to processdigital chemical signals. The contributions of this article aresummarized as follows: • To facilitate digital microfluidic circuit design, we pro-pose a five-level architecture to describe a microfluidiccircuit with a discussion of the components in each level. a r X i v : . [ c s . ET ] S e p Fig. 1. The illustration of levels of abstraction for an electronic processingsystem and a microfluidic processing system along with typical buildingcomponents at each level. • We present the microfluidic designs of AND, NAND, OR,NOR, and XOR logic gates, which are then validated bysimulation results. • Finally, we illustrate a roadmap for the developmentof microfluidic circuits with complex signal processingfunctions. Importantly, we also discuss the challengesduring circuit design and testing, and provide correspond-ing potential solutions.The remainder of this article is organized as follows. First,we present the five-level architecture of microfluidic circuitsand discuss the components at Levels 1–3 in Section II. Then,we detail the microfluidic designs of logic gates for Level 4in Section III. Finally, we identify challenges and potentialsolutions to build large-scale microfluidic circuits for Level 5in Section IV.II. M
ICROFLUIDIC C IRCUIT A RCHITECTURE AND M ODULE D ESIGN
In order to facilitate microfluidic circuit design, we abstracta microfluidic processing system into five levels as shown inFig. 1. The levels of the abstraction are as follows: • Level 1: Molecular Propagation – the movement ofchemical molecules in microfluidic channels. • Level 2: Chemical Transformation – the interactionbetween different species, i.e., the chemical reactions thatsupport various signal processing functions. • Level 3: Microfluidic Modules – the basic modulesperforming simple calculations. • Level 4: Microfluidic Logic Gates – the digital logicgates assembled from the microfluidic modules designedfor Level 3. Although the microfluidic modules at Level 3 process a continuous range of concentrations, an ap-propriate combination of them can lead to digital signaloperation. • Level 5: Microfluidic Circuits – the top level is mi-crofluidic circuit itself, which is built from the logicgates designed for Level 4 and can perform a specificsignal processing function, such as coding-decoding andmodulation-demodulation for MC.As shown in Fig. 1, we map our proposed levels of mi-crofluidic processing systems to those of electronic processingsystems [10]. The main differences lie in Levels 1 and 2.In electronic systems, Level 1 focuses on propagation ofelectrons. By contrast, Level 1 in microfluidic systems isbased on the movement of chemical particles. The Level 2of electronic systems is composed of transistors (e.g., diodeand triode), whereas that of microfluidic systems is based onchemical reactions for signal transformation. The Levels 3–5in these two systems have similar functions, but are realizeddifferently via either electronic components or microfluidiccomponents. In the following, we present the components atLevels 1 and 2, and then focus on how these componentssupport the construction of the microfluidic modules at Level3.
A. Level 1: Molecular Propagation
In Level 1, the movement of chemical molecules is boundedby channel geometry, and its dispersion is governed by diffu-sion and convection. A channel without reactions refers to aconvection-diffusion channel. According to the channel shape,we consider three types of geometry: Y junction, T junction,and straight convection-diffusion channel as shown in Fig. 2.The Y junction and T junction are both merging channelswith two inlets and one outlet, and they can facilitate themixing of different species injected through two inlets. Thestraight convection-diffusion channels only provide a pathwayfor chemical molecules.
B. Level 2: Chemical Transformation
In Level 2, we introduce chemical reactions into convection-diffusion channels. The channels with chemical reactions arenamed as convection-diffusion-reaction channels and are filledwith grey-gradient color as shown in Fig. 2. In this article, weconsider two forms of chemical reactions: I + J → K and I + Amp → I + O [11]. By introducing chemical reactions,microfluidic circuits are endowed with signal processing ca-pability. C. Level 3: Microfluidic Modules
By combining the components at Levels 1 and 2, we canconstruct the
Addition , Subtraction , and
Amplification modulesfor Level 3. As illustrated in Fig. 2, each module contains oneor two chemical reactions. In the following, we reveal thesignal transformation nature of the reactions at Level 2 anddiscuss the mechanism of each module.
I+J K I1+M NI2+M N N Addition
500 180 M I2I1 I+ThL W Subtraction ThLI I ThL I+Amp I+OAmp
Amplification
I O I+Amp I+O
Y Junction
T Junction
Straight Convection-Diffusion ChannelLevel 1MolecularPropagationLevel 2
Chemical
TransformationLevel 3
Microfluidic
Modules
Fig. 2. The components at Levels 1 and 2, and their construction to the addition, subtraction, and amplification modules for Level 3. The unit of the channellength is µ m.
1) Addition Module:
The addition module calculates thetotal concentration of two different molecular species and isachieved by converting them to the same molecular species.As shown in Fig. 2, it is composed of two Y junctions,two reaction channels, and convection-diffusion channels. Theinputs of an addition module are three chemical signals con-taining species I , M , and I , respectively. The output is thechemical signal with species N . In reaction channels, species I and I are transformed to species N via I M → N and I M → N . Due to the one-to-one stoichiometricrelationship between reactants and product, the generated con-centration of species N equals the consumed concentration ofspecies I /I . Moreover, the stoichiometric relationship alsoreveals that the amount of transformed species I /I dependson the concentration of species M . To ensure a completeconversion of species I /I to species N , the concentrationof species M needs to be greater than or at least equal to theconcentration of species I /I . After reactions, the species N generated in two reaction channels converge at convection-diffusion channels to generate the final output.
2) Subtraction Module:
The subtraction module calculatesthe concentration difference between two species and relies onthe depletion of one species by the other species. As shown inFig. 2, it is consisted of a T junction and a reaction channelwith species I and T hL as inputs and the remaining species I as output. In the reaction channel, input species I is consumedby species T hL via I + T hL → W , where species W represents a waste species whose concentration we do not keeptrack of. The module output, i.e., the remaining concentrationof species I , is determined by the concentration of species T hL . Under the condition that the concentration of species
T hL is greater than the concentration of species I , species I will be fully depleted. As a result, the concentration of species I will be set as zero due to the one-to-one stoichiometricrelationship between species I and species T hL .
3) Amplification Module:
The amplification module gen-erates a chemical signal whose width and amplitude aredetermined by two input signals. As shown in Fig. 2, it uses thesame geometry structure as the subtraction module but witha different reaction I + Amp → I + O . In the presence ofspecies I , species I acts as a catalyst to enable the conversion of the other input species Amp to output species O ; if species I is absent, species O will not be produced. In this way,species I determines the time period when output species O isgenerated, whereas reactant Amp influences the concentrationof species O . The higher the concentration of species Amp ,the higher the concentration of species O , which allows us toflexibly adjust the output concentration of species O .III. L EVEL
4: M
ICROFLUIDIC L OGIC G ATES
Flowing fluids in microfluidic channels allows for an easyserial processing operation, which endows microfluidic circuitswith the feature of integrating different functional modules tobuild the microfluidic logic gates at Level 4. In particular,we apply the microfluidic modules designed for Level 3to construct the AND, NAND, OR, NOR, and XOR gates.Throughout this paper, the HIGH state (bit-1) and the LOWstate (bit-0) are represented by non-zero concentration andzero concentration, respectively.
A. AND and NAND Gates
We first design the AND gate as shown in Fig. 3, which con-sists of the addition, subtraction, and amplification modules.The AND gate takes input signals I and I and producesa HIGH state for output species O only when both inputsare HIGH. The addition module first converts species I and I to an intermediate species N assisted by species M .According to the combination of input species I and I , thespecies N concentration C N at location x has a ladder-shapeddistribution (the purple line in the subtraction module in Fig.3) with three typical values: • C N = 0 when both input species are LOW, • C N = α when only one input species is HIGH (the reddot line in Fig. 3), • C N = α when both input species are HIGH (the reddash line in Fig. 3).To achieve an AND function, the concentration of species N in the amplification module is required to span over the timeperiod where both species I and I are HIGH. Therefore,the species N generated by the addition module flows into asubtraction module and undergoes a depletion by species T hL
Fig. 3. The chemical reactions-based microfluidic AND gate and OR gate. that is continuously supplied through the first T junction. Wehighlight that the concentration of species
T hL at location x , i.e., C T hL , must satisfy α < C T hL < α so that theremaining concentration of species N is larger than zero onlywhen both inputs are HIGH. Once the remaining species N arrives at the amplification module, reaction N + Amp → N + O is activated, inducing the conversion of species Amp to output species O . Thus, we complete AND logic operationin molecular domain. The AND gate can be converted to aNAND gate with the addition of a subtraction module. B. OR and NOR Gates
The OR gate can be designed using a similar geometrystructure as the AND gate, as shown in Fig. 3. Differentfrom the AND gate, an OR gate generates a HIGH state foroutput species O when one or both input species I and I are HIGH. The only difference in design parameters betweenAND gate and OR gate is the injected concentration of species T hL at the subtraction module. In theory, the concentration ofspecies N at location x should be zero when both species I and I are LOW. However, this value is likely to be slightlylarger than zero in practice. To mitigate this fluctuation, theoutput species N generated by the addition module is requiredto be further processed by a subtraction module in whichthe concentration of species T hL at location x (i.e., C T hL )should be larger than the fluctuation level and smaller than α (i.e., the concentration of species N at x when only oneinput is HIGH). When either one input is HIGH or both inputsare HIGH, the remaining concentrations of species N after N + T hL → W have two different values, and an amplificationmodule is used to ensure that these two values can lead to ageneration of the same amount of output species O . The ORgate can also be converted to a NOR gate with a cascade ofa subtraction module. C. XOR Gate
We design the XOR gate based on an AND gate, an ORgate, and a subtraction module. As shown in Fig. 4, inputsignals I and I first flow into the AND and OR gates that Fig. 4. The chemical reactions-based microfluidic XOR gate. operate parallelly to generate species O and O at location x , respectively. Then, the generated species O and O entera subtraction module to activate O O → W . In thisreaction, species O is completely depleted by species O so that the remaining concentration of specie O only showsa HIGH state when either input species I or I is HIGH,thus achieving the XOR operation. D. Microfluidic Logic Gate Design Validation
To examine the effectiveness of our designed logic gates,we simulate them in COMSOL. All the gates are constructedusing the channel lengths marked in Figs. 2 and 4 along withwidth µ m and depth µ m. As an exception, the width forY junction is µ m. We set the diffusion coefficient and meaninjection velocity for each species as − m /s and . cm/s,and the rate constant for each reaction as m /(mol · s).For all the gates, we set injected concentrations of species I , I , and M as C I ( t ) = 8[ u ( t − − u ( t − , C I ( t ) = 8[ u ( t − − u ( t − , and C M ( t ) = 8 u ( t ) with Heaviside step function u ( t ) . For other species, we set: C T hL ( t ) = 6 u ( t ) and C Amp ( t ) = 4 u ( t ) for AND gate; C T hL ( t ) = 2 u ( t ) and C Amp ( t ) = 4 u ( t ) for OR gate; C T hL ( t ) = 6 u ( t ) , C Amp ( t ) = 4 u ( t ) , C T hL ( t ) = 2 u ( t ) ,and C Amp ( t ) = 4 u ( t ) for XOR gate. The concentration unitis mol/m .The simulation results are plotted in Fig. 5. We observethat our designed microfluidic circuits show desired behavioras their corresponding electric circuits in terms of logicgate operation functionalities, which demonstrates that ourdesigned circuits are feasible and effective for digital signalprocessing in molecular domain.IV. C HALLENGES AND P OTENTIAL S OLUTIONS
The realization of signal processing functions via microflu-idic circuits shows significant advantages in reusability andmodularity. On the one hand, the reusability is reflected onthat the subtraction-amplification module and the AND-ORgate share the same microfluidic geometry structure but withdifferent reactions and species concentrations, as shown inFigs. 2 and 3. Hence, a microfluidic structure can performdifferent functions by using different design parameters, which
Fig. 5. The normalized concentrations of input species I and I and thecircuits’ outputs. reduces implementation cost and enables a separation of func-tion design from device manufacture. On the other hand, themodularity is embodied in the construction of different logicgates. As illustrated in Figs. 3 and 4, our designed microfluidiclogic gates can be constructed via the combinations of threemicrofluidic modules at Level 3, which is similar to Legosthat a construction of vehicles, buildings, or working robotsare built merely via interlocking plastic bricks.Due to reusability and modularity, more complex signalprocessing circuits at Level 5 are envisioned to be built throughcombinations of logic gates at Level 4. In order to clarify thesteps moving from Levels 1 to 5, we illustrate a roadmap inFig. 6, which includes microfluidic circuit design and testingstages. In the following, we highlight the main challenges ineach stage and identify corresponding potential solutions. A. Microfluidic Circuit Design
The complexity of a signal processing function largelydepends on the number of available gates; thus, the first stepin the design stage is to expand logic gate library. Second,logic synthesis should be performed to identify the circuitdiagram for a specified operation, which provides a basis forthe followed gate assignment to choose correct gates. Then,functional connecting gates should be theoretical analyzed andverified by simulation results.
1) Component Library Expansion:
Compared with thenumber of electronic logic gates in the literature, the number ofmicrofluidic logic gates is still limited. Thus, it is essential todesign more microfluidic logic gates, such as multiplexers anddecoders, and expand the component library to allow for morecomplex signal processing. As the microfluidic logic gates atLevel 4 are built from the microfluidic modules at Level 3,the library expansion includes designing or introducing morecomponents at lower levels. This includes but not limited tointroduce serpentine and herringbone-like geometry for Level1, design biological inspired chemical reactions for Level 2,and basic arithmetic operations (e.g., half adder) for Level 3.It is noted that our designed logic gates are all combinationalcircuits, where circuits’ outputs only depend on current inputsignals. However, the realization of many signal processing
SimulationValidationCircuit AnalysisLogic GateAssignmentLogicSythesisComponnetLibrary Expansion
Stage 1: Microfluidic Circuit Design I1 I2 0 1 Differential
Equation
Device
Fabrication
ReactionLocalization Reagents
Selection
ConcentrationMeasurement
Stage 2: Microfluidic Circuit Testing
COMSOL o C o C Multiphysics
HOHO NH xy y ’ Fig. 6. A roadmap for the development of microfluidic circuits with complexsignal processing functions. functions can be based on sequential logic circuits, wherecircuits’ outputs depend on both current and previous inputsignals. In microfluidic circuits, how to retain flowing chemicalinformation to achieve sequential circuits is a big challenge.
2) Logic Synthesis:
Logic synthesis produces a circuitdiagram with available logic gates to perform a specifiedoperation. A challenge here is how to determine the typesof gate for a desired operation. This procedure is usuallynot straightforward and explicit, but we can provide valuableinsights from some cases to facilitate logic synthesis. Forexample, to achieve the n th order CSK modulation, we cantranslate this operation into a control problem, which can beimplemented by a n : 2 n decoder so that one concentrationlevel is transmitted from n levels in terms of the combinationof n input bits [12]. In addition, it would be helpful to useBoolean algebra and Karnaugh maps to derive the simplestBoolean equations that reduce a circuit redundancy and leadto the simplest set of logic gates.
3) Logic Gate Assignment:
After determining a circuit dia-gram and the required gate types via logic synthesis, the nextstep is to select available gates from logic library to be usedin the circuit. We note that even for a single gate in a whole circuit, it is likely to have different microfluidic designs, whichintroduce a challenge to identify the optimal gate combinationfor the whole circuit. A straightforward method is to permuteall possible designs for each gate and identify the optimalcombination of the whole circuit. However, this approachbecomes intractable with the increase of library size and circuitscale. To address this challenge, attention needs to be givento design an assignment algorithm. Furthermore, quantifiedmetrics should be developed to analyze circuit complexity,such as the total size of a circuit, the total number of speciesand chemical reactions, and the speed to finish specified signalprocessing tasks.
4) Circuit Analysis:
An integral part of microfluidic circuitdesign is how to derive circuits’ outputs and theoreticallyanalyze circuit performance, such as the channel noise char-acterization and channel capacity calculation for a circuit withcommunication functionalities. By doing so, it would revealthe dependency of circuit performance on design parameters,and provide feedback for the circuit design.To derive circuits’ outputs, we can rely on the impulse response of a microfluidic channel so that output signals canbe written as the convolution of circuit inputs and a cascade ofthe impulse response of each channel. Unfortunately, there aremany challenges in merely deriving the impulse response ofan individual channel, and we highlight two main challengesas follows: • The turbulent flow: It has been revealed that flows inmicrofluidic channels are mostly laminar, and a 3D partialdifferential equation (PDE) can be approximated by a 1DPDE when laminar flows fall into the dispersion regime[8]. However, microfluidics may still exhibit turbulentfeature, which can be caused by unintended physicalbarriers owing to the imperfectness of device fabrication,or intended physical barriers (e.g., herringbone-like ge-ometry) designed to facilitate species mixing. It is notedthat turbulent flow is always 3D, and this characteristichinders the simplification of 3D PDEs. • The coupling of convection, diffusion, and reaction:Chemical reactions occur during molecular movementsuch that chemical reactions are fully coupled withconvection and diffusion process, leading to the non-linearity of the convection-diffusion-reaction equation. Todeal with this problem, we can rely on operator splittingmethod [12] that splits the original PDE into severaleasily solved subproblems, and perturbation method [13]that assumes the analytical solution in the form of aninfinite series of functions.
5) Simulation Validation:
Although theoretical analysis canpredict circuits’ outputs, it is necessary to validate well-designed circuits via appropriate simulation tools before pro-totyping. For microfluidic circuits, finite element simulationis the most suitable simulation approach. In finite elementsimulations, a microfluidic channel is partitioned into smallmeshes with geometrically simple shapes, and molecular con-centrations in each mesh are updated over a sequence of timesteps according to the corresponding PDEs. Not surprisingly,the mesh settings, such as the size, the density, and the numberof meshes, impose an impact on the time and the amount ofmemory required to compute a design, and finally influencethe accuracy of the solution. With an increase of circuit scale,the default meshing sequences provided by simulation toolsmay not guarantee an accurate solution within a short time,which encourages engineers to build a customized mesh thatis best-suited for their particular models. Therefore, how toquickly find a customized and an optimal mesh remains amajor challenge.
B. Microfluidic Circuit Testing
After verifying a well-defined microfluidic circuit, the nextstage is to build a circuit prototype, including circuit fabri-cation, chemical reaction localization, and reagents selection.Moreover, detection techniques should also be selected tovisualize the circuit’s outputs.
1) Device Fabrication:
Over the past two decades,a majority of microfluidic devices have been built inpoly(dimethylsiloxane) (PDMS) by soft lithography due tosome key properties, including biocompatibility, transparency, and water-impermeability. However, PDMS fabrication usuallyinvolves substantial human labor and layered molding, whichtends to hinder the dissemination of PDMS devices outsideof research labs and the production of complex 3D devices.Fortunately, the rapidly developed 3D-printing is a promis-ing technique for microfluidic fabrication with advantages ofautomation and assembly-free 3D fabrication. Moreover, thethermoplastic materials-based fabrication method is also analternative. Thermoplastic materials are not only more robustand easier to manufacture than traditional materials but canalso achieve a rapid and low cost fabrication.
2) Reaction Localization:
How to confine a chemical re-action within a region to achieve desired circuit behavioris a critical challenge. According to the
Arrhenius equationthat reveals a dependency of rate constant on temperature,temperature control provides an opportunity to address thischallenge. By means of cooling convection-diffusion channelswhile heating reaction channels, it would allow us to keepconvection-diffusion channels thermally isolated from reactionchannels, which ensures pre-mixed reactants do not react untilthey reach heated reaction regions [14].
3) Reagents Selection:
In the testing stage, a key stepis to choose appropriate reagents to map to the species inchemical reactions. The selection of reagents should considerthe following aspects: • The selected reagents and chemical reaction productsmust be non-toxic to human body and environment. • The interactions among selected reagents, reaction prod-ucts, and channel materials should be studied so as toprevent any side reactions. • The outputs of a logic gate are expected to be the inputsof the cascaded gate. By doing so, the interconnection oflogic gates can be automated. • The disposition or recycle of remaining solutions forfuture use should be taken into account.With the above requirements, the number of satisfied reagentsmay be limited, which imposes a restriction on circuit sizesand functions.
4) Concentration Measurement:
The outputs of microflu-idic circuits should be analyzed by suitable detection tech-niques, which presents another major challenge. There are twomain factors that affect the choice of concentration detectionmethod: the sensitivity to output species and the scalabilityto smaller dimensions. Among the common detection tech-niques, the optical-based technique is prone to satisfy therequirements of sensitivity and scalability. In particular, the“on-chip” detection mode, where some optical componentshave been fully integrated with or fabricated together withmicrofluidic devices, brings new opportunities for concentra-tion detection. This mode strongly ties optics and microfluidicstogether, which exhibits great advantages in high interactionefficiency compared with traditional “off-chip” optics config-uration where detection units are separated from microfluidicdevices [15]. V. C
ONCLUSION
The main hindrance for molecular communication (MC)applications stands in the lack of nano/micro-devices able to process chemical concentration signals in biochemicalenvironment. In this article, we proposed the vision of us-ing chemical reactions-based microfluidic circuits to performsignal processing functions. To manage the complexity ofmicrofluidic processing systems, we described a microfluidiccircuit from five levels of abstraction: 1) Molecular Propaga-tion; 2) Chemical Transformation; 3) Microfluidic Modules;4) Microfluidic Logic Gates; and 5) Microfluidic Circuits. Wefirst introduced channel geometry in Level 1 and chemicalreactions in Level 2. We then presented the designs of mi-crofluidic modules in Level 3 and microfluidic logic gates inLevel 4. Importantly, these designs demonstrated significantadvantages in reusability and modularity, and this motivatedus to design the microfluidic circuits in Level 5 with complexsignal processing functions based on the designs in Levels 1-4.We identified the challenges in designing and testing complexmicrofluidic circuits and proposed corresponding potentialsolutions. This article serves to inspire research to design,analyze, and test novel chemical reactions-based microfluidiccircuits that perform signal processing functions in moleculardomain, in order to support the advancement of MC-enabledapplications. R
EFERENCES[1] N. Farsad, W. Guo, and A. W. Eckford, “Tabletop molecular commu-nication: Text messages through chemical signals,”
PLoS One , vol. 8,no. 12, p. e82935, Dec. 2013.[2] S. Giannoukos, D. T. McGuiness, A. Marshall, J. Smith, and S. Tay-lor, “A chemical alphabet for macromolecular communications,”
Anal.Chem. , vol. 90, no. 12, pp. 7739–7746, May 2018.[3] L. Grebenstein, J. Kirchner, W. Wicke, A. Ahmadzadeh, V. Jamali,G. Fischer, R. Weigel, A. Burkovski, and R. Schober, “A MolecularCommunication Testbed Based on Proton Pumping Bacteria: Methodsand Data,”
IEEE Trans. Mol. Biol. Multi-Scale Commun. , vol. 5, no. 1,pp. 56–62, Oct. 2019.[4] R. Weiss, S. Basu, S. Hooshangi, A. Kalmbach, D. Karig, R. Mehreja,and I. Netravali, “Genetic Circuit Building Blocks for Cellular Compu-tation, Communications, and Signal Processing,”
Nat. Comput. , vol. 2,no. 1, pp. 47–84, Mar. 2003.[5] M. Cook, D. Soloveichik, E. Winfree, and J. Bruck, “Programmabilityof Chemical Reaction Networks,” in
Algorithmic bioprocesses . Berlin,Heidelberg: Springer, 2009, pp. 543–584.[6] H. Bruus,
Theoretical Microfluidics . London, U.K.: Oxford Univ. Press,2008.[7] Y. Deng, M. Pierobon, and A. Nallanathan, “A Microfluidic FeedForward Loop Pulse Generator for Molecular Communication,” in
Proc.IEEE GLOBECOM , Dec 2017, pp. 1–7.[8] D. Bi, Y. Deng, M. Pierobon, and A. Nallanathan, “Chemical Reactions-based Microfluidic Transmitter and Receiver Design for MolecularCommunication,”
IEEE Trans. Commun. , pp. 1–1, May 2020.[9] Y. Xiang, N. Dalchau, and B. Wang, “Scaling up genetic circuit designfor cellular computing: advances and prospects,”
Nat. Comput. , vol. 17,no. 4, pp. 833–853, Oct. 2018.[10] D. Harris and S. Harris,
Digital Design and Computer Architecture . CA,San Mateo: Morgan Kaufmann, 2010.[11] D. Scalise and R. Schulman, “Designing Modular Reaction-DiffusionPrograms for Complex Pattern Formation,”
Technology , vol. 2, no. 01,pp. 55–66, Mar. 2014.[12] D. Bi and Y. Deng, “Microfluidic QCSK Transmitter and ReceiverDesign for Molecular Communication,” arXiv:2005.01353 , May 2020.[13] H. Abin, A. Gohari, and M. Nasiri-Kenari, “An Analytical Modelfor Molecular Communication over a Non-linear Reaction-DiffusionMedium,” arXiv:2005.14296 , May 2020.[14] B. K. Yen, A. G¨unther, M. A. Schmidt, K. F. Jensen, and M. G.Bawendi, “A Microfabricated Gas–Liquid Segmented Flow Reactor forHigh-Temperature Synthesis: the Case of CdSe Quantum Dots,”
Angew.Chem. Int. Ed. , vol. 44, no. 34, pp. 5447–5451, Aug. 2005. [15] H. Yang and M. A. Gijs, “Micro-optics for Microfluidic AnalyticalApplications,”
Chem. Soc. Rev. , vol. 47, no. 4, pp. 1391–1458, Jan.2018.
Dadi Bi is currently pursuing the Ph.D. degree with the Department ofEngineering, King’s College London, U.K. His research interests includemolecular communication and microfluidics.