Graphene-based Nanoscale Molecular Communication Receiver: Fabrication and Microfluidic Analysis
Murat Kuscu, Hamideh Ramezani, Ergin Dinc, Shahab Akhavan, Ozgur B. Akan
11 Graphene-based Nanoscale MolecularCommunication Receiver: Fabrication andMicrofluidic Analysis
Murat Kuscu, Hamideh Ramezani, Ergin Dinc, Shahab Akhavan, Ozgur B. Akan
Abstract
Bio-inspired molecular communications (MC), where molecules are used to transfer information,is the most promising technique to realise the Internet of Nano Things (IoNT), thanks to its inherentbiocompatibility, energy-efficiency, and reliability in physiologically-relevant environments. Despite asubstantial body of theoretical work concerning MC, the lack of practical micro/nanoscale MC devicesand MC testbeds has led researchers to make overly simplifying assumptions about the implicationsof the channel conditions and the physical architectures of the practical transceivers in developingtheoretical models and devising communication methods for MC. On the other hand, MC imposesunique challenges resulting from the highly complex, nonlinear, time-varying channel properties thatcannot be always tackled by conventional information and communication tools and technologies (ICT).As a result, the reliability of the existing MC methods, which are mostly adopted from electromagneticcommunications and not validated with practical testbeds, is highly questionable. As the first step toremove this discrepancy, in this study, we report on the fabrication of a nanoscale MC receiver basedon graphene field-effect transistor biosensors. We perform its ICT characterisation in a custom-designedmicrofluidic MC system with the information encoded into the concentration of single-stranded DNA
Murat Kuscu and Hamideh Ramezani are with the Internet of Everything (IoE) Group and Cambridge Graphene Centre (CGC),Department of Engineering, University of Cambridge, Cambridge, CB3 0FA, UK (e-mail: { mk959, hr404 } @cam.ac.uk).Ergin Dinc and Ozgur B. Akan are with the Internet of Everything (IoE) Group, Department of Engineering, University ofCambridge, Cambridge, CB3 0FA, UK (e-mail: { ed502, oba21 } @cam.ac.uk).Shahab Akhavan is with the Cambridge Graphene Centre, (CGC) Department of Engineering, University of Cambridge,Cambridge, CB3 0FA, UK (e-mail: [email protected]).This work was supported by the ERC project MINERVA (ERC-2013-CoG July 7, 2020 DRAFT a r X i v : . [ c s . ET ] J u l molecules. This experimental platform is the first practical implementation of a micro/nanoscale MCsystem with nanoscale MC receivers, and can serve as a testbed for developing realistic MC methodsand IoNT applications. I. I
NTRODUCTION
Nanotechnology is enabling us to devise ever-smaller devices to interact with the universeat molecular resolution. Though small enough to penetrate into cells, these devices individuallyare of limited capability. To unleash the full potential of nanotechnology, communication amongnanomachines is a must. This would enable more complex applications, e.g., continuous healthmonitoring with intrabody nanosensor networks, theranostic applications with distributed sensorand actuator networks, and industrial nanosensor applications. Internet of Nano Things (IoNT),which defines these networks of nanomachines, such as nanobiosensors and engineered bacteria,integrated with the Internet infrastructure, has seen tremendous interest recently [1]–[3]. IoNTis set to transform the way we connect with and understand the world at the bottom. However,conventional electromagnetic (EM) communication techniques, proved impractical at nanoscaledue to the antenna size and power limitations, and large propagation losses. On the other hand,Nature itself already provides a robust way of nanocommunication, i.e., Molecular Commu-nications (MC), which is the common communication modality among living cells, rangingfrom our own neurons to bacteria [4]. Using the same language with living cells by encoding,transmitting and receiving information with molecules provides an energy efficient and reliablenanocommunication, even at harsh biological environments, where, we expect, the most impactfulmedical applications of IoNT would be implemented.Although there is tremendous interest in this field accompanied by a large body of theoreticalwork to develop models and devise communication methods for MC [5]–[11], researchers relyon overly simplifying assumptions about the physical constraints of the transceivers and thechannel conditions due to the lack of a testbed for MC at micro/nanoscale, where these modelsand methods can be validated. On the other hand, MC brings about unique challenges resultingfrom its highly complex, nonlinear, time-varying channel properties due to the discrete natureof information carriers (molecules), substantial channel memory and peculiarities of molecularinteractions at nanoscale, that cannot be always tackled by conventional ICT tools [4], [12], [13].This leaves a huge question mark over the reliability of the existing MC methods, which are
DRAFT July 7, 2020 mostly adopted from conventional EM communications and not validated with practical testbeds.Few studies in MC literature have focused on macroscale implementation of MC systemstaking into account the physical limitations of a receiver, although the utilized receivers aremade of off-the-shelf macroscale components. In [14], the isopropyl alcohol (IPA) is used asinformation carrier, and commercially available metal oxide semiconductor alcohol sensors areused as MC receiver. This study provides a testbed for MC with macroscale dimensions, whichis later utilised in [15] to estimate its combined channel and receiver model. This testbed isextended to a molecular multiple-input multiple-output (MIMO) system in [16] to improve theachievable data rate. In [17], the information is encoded in pH level of the transmitted fluid, anda pH probe sensor is used as the MC receiver. On the grounds that the use of acids and bases forinformation transmission can adversely affect the other processes in the application environment,such as in the human body, magnetic nanoparticles (MNs) are employed as information-carryingmolecules in microfluidic channels in [18]. In that study, a bulky susceptometer is used to detectthe concentration of MNs and decode the transmitted messages. In addition, the performance ofMN-based MC, where an external magnetic field is employed to attract the MNs to a passivereceiver, is analysed in [19]. However, the focus of these works is on macroscale MC usingoff-the-shelf sensors as receiver. Therefore, these studies do not contribute to the developmentof a design and optimisation framework for practical nanoscale MC receivers that can actuallybe integrated into micro/nanoscale devices.As the first step to overcome this challenge, in this work, we report on the first implementationof a nanoscale MC receiver based on graphene field-effect transistor-based DNA biosensors(graphene bioFETs), and its ICT performance tests in a custom-designed microfluidic MC system.The main objective of this work is to provide an experimental testbed at physically relevantdimensions for nanonetworks, which can be used to reveal and study the effects of intricatebiochemical and physical processes on the MC performance, and develop practical and realisticcommunication methods, including new MC detection techniques.Graphene, with its exceptional electrical, chemical and mechanical properties, such as highcarrier mobility at room temperature, one atomic layer thickness and two dimensional geometryexposing all its atoms to the sensing environment, provides very high sensitivity towards bio-chemical molecules especially in a bioFET configuration [20]–[22]. Owing to these properties,graphene has been extensively studied for selective sensing of a wide range of biomolecules
July 7, 2020 DRAFT ranging from carbohydrates [23] to proteins [24] and oligonucleotides [25], [26]. Meeting thefundamental requirements of an MC receiver, such as the capability of label-free and reversibledetection and high sensitivity [27], graphene bioFET stands as an ideal candidate for the imple-mentation of the MC receiver. Flexibility and nanoscale 2d geometry of graphene are particularlyfavourable for the integration of graphene-based MC receiver into functional nanoscale devices.Functionalisation of graphene with biomolecular probes can provide the selectivity againsttarget analytes, required for avoiding biochemical interference for MC applications in physio-logically relevant environments. In this work, graphene is functionalised with single-strandedDNA (ssDNA) probes (pDNAs) which undergo reversible hybridisation reaction with the com-plementary target DNAs (tDNAs). The reason for selecting DNA as the recognition element isthat DNA can be easily customized with different base sequences of different lengths, and can bedesigned to bind not only complementary DNAs but also peptides, proteins, carbohydrates andsmall molecules [28]. As a result, the implemented device can serve as a model system to provideinsight into a broad range of MC systems relying on detection of molecular messages throughaffinity-based ligand-receptor interactions [6], [12]. Moreover, integration of the fabricated MCreceiver into a pressure-regulated microfluidic testbed provides control over the fluid flow rate,and enables flexibility and practicality in testing different channel geometries, which can mimicthe most promising application environments of the MC inside human body, e.g., circulatorysystem [29], [30].In the remainder of this paper, we elaborate on the fabrication process of single layer graphene(SLG) bioFET-based MC receiver and its integration into a microfluidic testbed. The electricalcharacterisation of the device is performed at each step of functionalisation. Sensing responsecharacteristics are revealed to determine the affinity between the complementary tDNA-pDNApair. Selectivity of the device against complementary tDNAs is examined through real-timesensing response to non-complementary target DNAs (ntDNAs). Following the fabrication andsensitivity/selectivity analysis, we provide an MC detection performance analysis based on thetransmission of pseudo-random binary data encoded into the concentration of tDNAs. The time-varying response of the MC receiver is fitted by a previously developed microfluidic MC model.This analysis provides important insights particularly into the infamous ISI problem of MCresulting from the slow kinetics of ligand-receptor binding reactions. Similar to the existingapproaches in the MC literature [31], [32], a concentration difference-based detection method is
DRAFT July 7, 2020 utilised to overcome the ISI effects by obviating the need for channel state information (CSI).II. F
ABRICATION OF
MC R
ECEIVER
MC receiver is fabricated in three consecutive steps. First, a graphene field-effect transistor(GFET) with chemical vapour deposition (CVD)-grown SLG is fabricated on Si/SiO substratethrough optical lithography techniques. Then, a polydimethylsiloxane (PDMS)-based microfluidicchannel is produced to encapsulate the GFET for bio-functionalisation, and real-time microfluidicsensing and communication experiments. Finally, for selectivity of the MC receiver againstinformation-carrying target DNAs, bio-functionalisation of the GFET channels with probe DNAmolecules is performed inside the microfluidic channel connected to a pressure-regulated mi-crofluidic setup.MC receiver is fabricated with a CVD-grown SLG polycrystalline domain on an n-type Si/SiO substrate (525 µ m with 90 nm thermal oxide layer, obtained from Mi-Net Technology Ltd). TheCVD-grown SLG on Cu with PMMA coating (60 nm, 495K, A2) is obtained from GrapheneaInc.
1) Wet Transfer of CVD Graphene on Si/SiO Substrate:
The Cu layer of the PMMA/SLG/Custack should be removed before transfer onto Si/SiO substrate. The Cu side is partially coveredwith a graphitic film, which may result in poor Cu etching performance. This backside grapheneon Cu is etched away with O plasma at 3 W for 30 seconds in a low-power Reactive Ion Etcher(RIE) in NanoEtch (Moorfield Nanotechnology Ltd). After this step, Cu is etched by placingthe PMMA/SLG/Cu stack on the surface of a solution of ammonium persulphate (APS) (1.8 gof APS in 150 ml DI water (18.2 M Ω · cm)).The Si/SiO substrate is cleaned before the transfer by means of sonication for 10 minutesin acetone followed by immersion in isopropyl alcohol (IPA) for 5 minutes and drying withnitrogen (N ). Once the Cu is entirely dissolved, the floating PMMA/SLG stack is transferredonto the surface of DI water in a beaker by a glass slide to dilute the APS residuals, and then,the stack is fished onto the Si/SiO substrate. The resulting sample (PMMA/SLG/SiO /Si) isleft vertically to dry overnight, and then annealed over a hot plate at 150 ◦ C for 2 hours. Thesample is then transferred into a beaker with acetone for PMMA removal for 2 hours, and thenimmersed in IPA for 5 minutes and dried with N , leaving only the SLG film on the Si/SiO substrate (Fig. 1(a)). July 7, 2020 DRAFT
2) Patterning of SLG Channels:
MC receiver is designed to contain 7 GFETs having isolatedsource and drain contacts but being exposed to a common electrolyte gate. The individual SLGchannels are patterned via optical lithography with a laser writer according to the design shownin Fig. 17 in Appendix A. Prior to all laser writing processes in this work, the sample isspin-coated with a photoresist (AZ-5214E from Microchemicals GmbH) at 4000 rpm for 60seconds (Fig. 1(b)), and baked at 110 ◦ C for 50 seconds on a hot plate. The photoresist layer isexposed by direct laser writing (wavelength-405 nm 169 mJ/cm ) via laser writer (LW-405B+from Microtech Srl), and the pattern is successively developed in diluted developer solution (1:4,AZ-351B/DI Water) for 35-45 seconds followed by brief immersion of the sample in DI-waterfor 2 seconds and drying with N (Fig. 1(c)). The patterning of the individual SLG channelsis completed with RIE removing the undesired areas of SLG film, which are not covered withthe photoresist layer, via O plasma at 3 W for 60 seconds (Fig. 1(d)). Finally, to remove thephotoresist layer, the sample is successively immersed in acetone and IPA for 20 minutes and 5minutes, respectively, and dried with N , leaving the patterned SLG film on the substrate (Fig. (a) (d) (b) (c)(e) SiSiO SLGAZ-5214E
Fig. 1: Process flow for the patterning of SLG channels. (a) SLG transferred on Si/SiO substrate. (b) Coating of sample with photoresist layer. (c) SLG channel pattern defined by opticallithography. (d) SLG channel pattern after RIE etching. (e) Removal of residual photoresist layerfrom SLG surface, and the resulting SLG channel on the substrate. DRAFT July 7, 2020 (a)(c) (b)(d)
SiSiO SLG AZ-5214EAuCr
Fig. 2: Process flow for the deposition of contacts. (a) Coating of sample with photoresist layer.(b) Contact pattern defined by optical lithography. (c) Deposition of Cr and Au metal filmsthrough thermal evaporation. (d) Patterned contacts after lift-off process.1(e)).
3) Deposition of Contacts:
In the following step, metal contact areas (source and drain) aredefined on the sample through another optical lithography process (Figs. 2(a)-(b)). Depositions of5 nm Cr and 50 nm Au are performed successively over the sample covered with the patternedphotoresist layer by thermal evaporation at 10 − mbar (using MiniLab 060 from MoorfieldNanotechnology Ltd) (Fig. 2(c)). Once the metals are deposited uniformly, the sample is dippedin acetone for 2 hours for lift-off process, during which the metals over the photoresist layer areremoved leaving only the patterned contacts (Fig. 2(d)).
4) Deposition of Insulator:
Finally, the drain and source contact areas, which might beexposed to the electrolyte during microfluidic experiments, are insulated to prevent any parasiticcurrent between metal contacts through the electrolyte. For this, a thin layer of Al O (20nm) is uniformly deposited over the sample through atomic layer deposition (ALD) in TFS200(manufactured by Beneq) (Fig. 3(a)). This step is followed by another optical lithography process, July 7, 2020 DRAFT (a) (d) (b) (c)(e)
SiSiO SLGAZ-5214EAuCrAl O Fig. 3: Process flow for the deposition of insulator. (a) Uniform Al O film over the sampleafter ALD. (b) Coating of sample with photoresist layer. (c) Insulator pattern defined by opticallithography. (d) Exposed SLG channel and contacts after wet etching of Al O . (e) PatternedAl O film after removal of excess photoresist.which defines the windows over Al O to expose only the SLG channels to the electrolyte, andto expose the source and drain contact pads, which remain outside of the microfluidic channelfor electrical measurements (Figs. 3(b)-(c)). The exposed areas of Al O are then wet-etched inPhosphoric Acid (85% wt. in H O obtained from Sigma-Aldrich) at 60 ◦ C for 1 minute (Fig.3(d)). For removing any excess photoresist, the sample is dipped in acetone for 20 minutes andIPA for 5 minutes followed by drying with N , leaving the patterned Al O film on top (Fig.3(e)). The optical images of the fabricated GFET are shown in Fig. 4. A. Fabrication of Microfluidic Channels and Device Integration
Fabricated GFET is encapsulated with a PDMS microfluidic channel, as demonstrated in Fig.5. To this end, a 3d-printed mould is designed to define the geometry of the rectangular fluidicchannel within the PDMS layer (see Fig. 18(a) in Appendix A). The microfluidic channel hasa width of × µ m and a height of . × µ m. At one end, the channel bifurcates forconnection with the two channel inlets, which are designated for connection to the fluid reser-voirs containing the buffer and information-carrying tDNA solutions during the communication DRAFT July 7, 2020
200 μm
20 μm (a)(b) (c)
Fig. 4: (a) Optical micrograph of the fabricated GFET channels after Al O etching process(only six of the seven channels are visible). (b) A closer look into one of the GFET channels.(c) Overall view of the fabricated 7-channel GFET before the bonding of microfluidic PDMSlayer.experiments.PDMS prepolymer is prepared using a 10:1 mixture of PDMS base monomer (Sylgard 184Silicone Elastomer) and PDMS curing agent (obtained from Dow Corning Corporation). Airbubbles inside the PDMS are removed by degassing in a desiccator for 1 hour. The degassedmixture is poured onto the 3d-printed mould and left for curing overnight at room temperature.The cured PDMS is then carefully peeled off from its mould (see Fig. 18(b) in Appendix A).The inlet and outlet holes are punched through the PDMS layer for microfluidic connections bya biopsy punch (1.25 mm radius). A platinum (Pt) wire having a diameter of 0.5 mm acting asthe common solution gate is then mounted to the top of the PDMS channel right above the SLGchannels (Fig. 5, and Fig. 18(c) in Appendix A). The length of the Pt wire inside the channel July 7, 2020 DRAFT0 (a) (b)
Pt gateInlets OutletPDMSlayer Microfluidic channel Microfluidic channel Pt gateSLG
Fig. 5: (a) Microfluidic PDMS layer bonded to the GFET surface after the inlets and outlet aredefined, and the Pt gate electrode is placed on top. (b) Cross-sectional view of the MC receiverafter PDMS layer bonding.is set to 1 cm.In the next step, the patterned PDMS with the Pt solution gate is bonded to the surface of theMC receiver, ensuring that the graphene channels are well-aligned with the microfluidic channeland not placed under the PDMS walls. The most common method for bonding PDMS on SiO and glass substrates is based on the O plasma activation of the PDMS surface and the targetsubstrate. This requires the surfaces of both the PDMS and the target substrate to be smooth.In our case, however, the exposed graphene channels on the target SiO substrate prevents theapplication of the O plasma, as this would cause the removal of graphene channels throughplasma etching. Moreover, the plasma activation of only the PDMS surface is not sufficientbecause curing in the 3d-printed moulds made of Polylactic acid (PLA) results in PDMS layerswith a rough surface (see Fig. 18(c) in Appendix A) rendering O plasma activation ineffectivein bonding. Therefore, we apply an alternative method, which was first introduced in [33]for bonding porous membranes into PDMS devices. In this method, a thin layer of PDMSprepolymer, which is in liquid form, is coated on the bonding surface of the cured PDMS layer.Then, the PDMS is carefully placed on the sample, which is cleaned off any dust with N prior tobonding. After placement, it takes approximately 1 minute for the PDMS prepolymer to spreaduniformly and cover the entire area between the PDMS and substrate except for the empty area DRAFT July 7, 20201
Flow sensor 1Switch 1Reservoir 1Pressure tubes PTFEtubesReservoir 2 Flow sensor 2Switch 2
MC receiver
WastereservoirGateSource Drain
Source-MeasureUnitPressure Regulator
Gate Source DrainPtgate PDMSlayerInlet 1 Inlet 2Outlet (a)(b) (c)
Fig. 6: (a) Microfluidic measurement setup consisting of a 4-channel pressure regulator, ahigh-precision SMU, electrical probes and microfluidic accessories. (b) A closer look into thefabricated graphene-based MC receiver connected to the microfluidic setup. (c) Probe connectionsfor electrical tests of the device.
July 7, 2020 DRAFT2 defining the microfluidic channel. Once a uniform PDMS prepolymer layer is observed, thetemperature of the hot plate is increased to 150 ◦ C, and the prepolymer, serving as mortar, isquickly cured, resulting in a strong bonding. This method has consistently yielded leakage-freePDMS-substrate bonding during the fabrication process.After bonding process, the inlet and outlet tubes are placed on the predefined inlet/outlet holes,as shown in Fig. 6(b). Here, Teflon PTFE tubing (1/16” OD x 1/32” ID, obtained from DarwinMicrofluidics) is preferred because of its higher chemical stability compared to Tygon tubing,which reacts with DMF used in the functionalisation process. The placement of the inlet andoutlet tubes is followed by the application of PDMS prepolymer around the connection pointsof inlet, outlet and Pt gate over the cured PDMS layer for the complete sealing of the device.The dimensions of the microfluidic channel together with the fluid flow rate and fluid propertiesdetermine the Reynolds number, which is a dimensionless variable indicating the fluid flowregime in the channel [34]. Reynolds number is the ratio of the inertial forces to the viscousforces, and can be given by
Re = ρuD H µ , (1)where ρ is the fluid density, u is the linear flow velocity of the fluid, µ is the viscosity ofthe fluid, and D H is the hydraulic diameter, which can be obtained for rectangular channels asfollows D H = 4 A ch P . (2)Here A ch = w ch × h ch is the cross-sectional area of the channel, and P = 2( w ch + h ch ) is thecross-sectional channel perimeter. In the sensing and communication experiments of this work,water-based solutions are flowed at a constant volumetric flow rate u V = 80 µ l/min. The linearflow velocity can then be obtained as u = u V /A ch = 220 µ m/s. By using ρ ≈ kg · m − and µ ≈ . Pa · s for water, we can obtain the Reynolds number for the microfluidic MCsystem as Re = 0 . , indicating a strong laminar flow regime, where viscous forces overcomethe inertial forces resulting in non-crossing, parallel streamlines [34].
1) Microfluidic Setup:
For functionalisation and electrical characterisation in the next steps,the device was connected to a microfluidic test setup, as shown in Fig. 6. The setup consistsof a pressure regulator (OB1 MK3 - Microfluidic flow control system, obtained from Elveflow)
DRAFT July 7, 20203 with four pressure outlets, two of which are connected to fluid reservoirs through the pressureinlets. The fluid outlets of the fluid reservoirs are connected to the device through PTFE tubing.Throughout the bio-functionalisation, sensing and communication experiments, microfluidicflow sensors are partly utilised for feedback-controlled modulation of the inlet pressure, andmechanical flow switches are used in cases where immediate stop/start of the microfluidic flowis required.
B. Functionalisation of GFET
Due to its one atomic thickness and 2d structure, the electronic properties of the pristineSLG is highly sensitive to the biochemical environment in the vicinity of its surface. Therefore,it suffers from low-level selectivity. On the other hand, in order to suppress the interferencefrom other biochemical processes in physiologically relevant applications of the MC receiver,selectivity against information-carrying molecules is a must. Selectivity of the graphene can berealised through bio-functionalisation with recognition elements such as DNA and antibodies.As ssDNAs are preferred as target information-carrying molecules, i.e., tDNAs, in this work,the fabricated GFET is functionalised with probe DNAs (pDNAs), which are complementary totDNAs.
Bufferinlet Channel outletMicrofluidicPressure ControllerPressure ControlLines PBSbuffer tDNAsolution WasteGraphene bioFETMC receiver at the bottom surfacetDNAInlet
Fig. 7: Conceptual drawing of the microfluidic measurement setup with the practical implemen-tation shown in Fig. 6
July 7, 2020 DRAFT4
OON OO H N OON OO
OOHN OOHN PBASEpDNA tDNA (a) (b) (c) (d)
Fig. 8: (a) Molecular structure of PBASE, and conceptual drawing of probe DNA (pDNA) andcomplementary target DNA (tDNA). (b) Noncovalent binding of PBASE to graphene via π − π interaction. (c) Immobilisation of pDNA via conjugation reaction with the succinimide group ofPBASE. (d) pDNA-tDNA hybridisation.For increasing the strength of the probe DNA immobilisation, and reducing the effect ofnonspecific binding, the pristine SLG channels are first functionalised with 1-Pyrenebutyric acidN-hydroxysuccinimide ester (PBASE, obtained from Cambridge Bioscience Ltd), which has beenwidely utilised in the literature as linker molecules between graphene surface and DNA molecules[25], [26], [35], [36]. PBASE is an aromatic molecule having an aromatic pyrenyl group and anamine-reactive succinimide group (Fig. 8(a)). PBASE exhibits a strong affinity towards SLG asits aromatic pyrenyl group interacts with the basal plane of graphene through π - π interactionsresulting in a strong noncovalent binding (Fig. 8(b)). The noncovalent attachment of the PBASEdoes not alter the inherent electronic structure and physical properties of the graphene [37]. Forthe functionalisation of the SLG with PBASE molecules, 10 mM solution of PBASE in N,N-Dimethylformamide (DMF, anhydrous, 99.8%, obtained from Sigma-Aldrich) is prepared in aglass bottle, and sonicated for 30 seconds for mixing. The prepared PBASE/DMF solution isflowed through the microfluidic channel until the entire channel is filled with the solution. Thenthe flow is stopped, and the SLG channels are exposed to steady PBASE/DMF solution for 2hours. After functionalisation with PBASE, unbound PBASE molecules are removed from thechannel with pure DMF, followed by rinsing with phosphate buffered saline (PBS, pH 7.4). DRAFT July 7, 20205
The next step is the immobilisation of 18-mer 5’-amine-modified probe DNAs, which havethe base sequence H N-(CH ) -5’-AGG ACT TCA CCG TAT TGC-3’. The DNAs are customdesigned and obtained from Sigma Aldrich. 2 µ M of probe DNAs, prepared in PBS, is flowedthrough the microfluidic channel over the SLG channels. The device is left for immobilisationwith probe DNAs overnight at 4 ◦ C inside a wet chamber following the recipe given in [26]. Theamine group of the pDNA reacts with the succinimide group of PBASE through conjugationreaction (Fig. 8(c)). The excess pDNA is then removed from the channel with PBS rinsing. Notethat although the ssDNAs and the double-stranded DNAs (dsDNAs) shown in Figs. 8(c)-(d) aredepicted as vertically aligned over the linker molecules, the orientation of DNAs tethered tosurfaces through their single end can be influenced by the electrical potential of the surface,electrolyte flow conditions, the length of the DNAs, temperature, pH, ionic strength of theelectrolyte, and the existence of the linker molecules. It is shown through molecular dynamicssimulations that under zero potential of the surface and in the absence of the solution gatepotential and the linker molecules, the flexible nature of the ssDNAs results in tilted and near-parallel orientation on the surface [38]. On the other hand, dsDNAs attain more vertical alignmentunder zero potential due to their higher rigidity [38]. However, to the best of author’s knowledge,the effect of the solution gate potential and the linker molecules has not been studied in theliterature.After pDNA immobilisation, the passivation of the unbound PBASE molecules is necessaryto prevent nonspecific binding of target DNAs (tDNAs). This is performed by flowing 100 mMethanolamine (NH CH CH OH) solution prepared in DI water through the microfluidic channel.Ethanolamine reacts with amine-reactive succinimide group of unbound PBASE molecules. Withthe passivation of PBASE, the device becomes ready for sensing and communication experimentswith tDNAs. III. E
LECTRICAL C HARACTERISATION OF
MC R
ECEIVER
For the electrical characterisation of the fabricated devices, direct-current (DC) measurementsare taken using a high-precision source measure unit (SMU, Keysight B2902A), which is con-nected to the device electrodes via high-impedance passive probes, as shown in Fig. 6. Onthe other hand, the mobility of the GFET channels is measured before functionalisation in aback-gate configuration using EverBeing probe station. Based on the linear approximation of the
July 7, 2020 DRAFT6 transfer curve, the mobility is calculated as (240 . ± . cm /V · s. A. Transfer Characteristics
After each step of functionalisation, transfer characteristics of the devices are obtained witha constant drain-to-source bias V ds = 100 mV, and a solution gate potential V g varying between-0.2 V and 1.2 V. The sweep rate of V g is set to 140 mV/s. All data are obtained afterremoval of excessive functional molecules from the microfluidic channel and the SLG surfacesby rinsing with PBS. This ensures that no change occurs in transfer characteristics due to ongoingchemical reactions. The PBS (pH 7.4) is used as the electrolyte in all measurements of transfercharacteristics.The measurements are taken from four of the SLG channels in the MC receiver, and resultsare provided in Fig. 9. Hysteresis was negligible for all channels (see Fig. 15 in Appendix A),thus, only the forward sweep of V g is demonstrated. First measurement is taken with only thePBS electrolyte inside the microfluidic channel prior to the functionalisation process. The p-typebehaviour and ambipolar characteristics of the SLG-based devices are revealed with the chargeneutrality point (CNP), i.e., the gate voltage of the minimum conductance, observed at ∼ ∼ ± mV is observed, indicating n-type doping. The negative shift of the CNP after the PBASE functionalisation was previouslyattributed to the dominance of the n-type doping effect of DMF in competition with the p-type doping effect of the PBASE molecules over long incubation times [37], [39]. Note thatthe large standard deviation of the CNP shift is mostly resulting from the atypical behaviourobserved in the third GFET channel, the transfer characteristics of which are demonstrated inFig. 9(c). In this channel, only 67 mV-shift in CNP is observed with the functionalisation ofPBASE. The significantly smaller shift compared to other channels, which manifest consistenttransfer characteristics, can be indicative of the poor functionalisation of the PBASE linkers onthis particular GFET channel. The poor functionalisation can be due to the residual polymers orinsulator material on the graphene surface remaining from the fabrication process and preventingthe non-covalent attachment of the PBASE molecules.On the other hand, the immobilisation of pDNAs resulted in a positive shift of the CNP inconsistence with the previous literature [25]. DNA molecules are negatively charged at pH 7.4, DRAFT July 7, 20207 -0.2 0 0.2 0.4 0.6 0.8 1 1.2
Gate Voltage V g (V) D r a i n - S ou r c e C u rr en t I d s ( A ) PBSPBASEpDNAEthanolamine/0.01xPBS (a) -0.2 0 0.2 0.4 0.6 0.8 1 1.2
Gate Voltage V g (V) D r a i n - S ou r c e C u rr en t I d s ( A ) PBSPBASEpDNAEthanolamine/0.01xPBS (b) -0.2 0 0.2 0.4 0.6 0.8 1 1.2
Gate Voltage V g (V) D r a i n - S ou r c e C u rr en t I d s ( A ) PBSPBASEpDNAEthanolamine/0.01xPBS (c) -0.2 0 0.2 0.4 0.6 0.8 1 1.2
Gate Voltage V g (V) D r a i n - S ou r c e C u rr en t I d s ( A ) PBSPBASEpDNAEthanolamine/0.01xPBS (d)
Fig. 9: Transfer characteristics of four channels at different steps of functionalisation in termsof drain-source current I ds as a function of varying gate voltage V g with sweep rate 140 mV/s.In all measurements drain-source voltage is held constant at V ds = 0 . V.attracting hole carriers to the graphene surface, thus, contributing to the p-type doping [25]. Theshift of the CNP is observed as ± mV over the four channels. Again, the large standarddeviation can be attributed to the third channel given in Fig. 9(c), where the CNP shift is only11 mV. This is again indicative of the poor functionalisation of the PBASE molecules, whichin turn results in a very low concentration of immobilised pDNAs. Other channels, on the otherhand, show similar CNP shifts, indicating more consistent immobilisation of DNAs.The last measurements are taken after the passivation of excess PBASE linkers with the July 7, 2020 DRAFT8 ethanolamine, and the introduction of the 0.01xPBS to the microfluidic channel to be used forthe following sensing and communication experiments. 0.01xPBS is the 100-fold diluted versionof PBS with DI water (18.2 M Ω · cm). While the ethanolamine does not possess any charge, theobserved positive shift of the CNP is consistent with the previous literature reporting increasedp-type doping with decreasing ionic concentration of the buffer solution [40]. Also note thatin all of the measured GFET channels, a mobility reduction is observed upon passivation withethanolamine in 0.01xPBS. The reason for switching to the diluted version of PBS for sensingand communication experiments is to decrease the effect of the Debye screening for enhancingthe sensitivity of the device for the hybridisation events on the SLG surface. Note that in allmeasurements, leakage current I gs has been detected to be under 15 nA (see Fig. 16 in AppendixA), and therefore, its effect on the transfer curve characteristics is negligible.With the help of the transfer characteristics, we can deepen our analysis by determining thesurface density of the immobilised pDNAs. To this end, we need to first determine the electrolytegate capacitance, which can be approximated by the overall capacitance of three parallel platecapacitors connected in series: C G = (cid:18) C Gr + 1 C P t + 1 C Q (cid:19) − , (3)where C Gr is the EDLC between the graphene and the electrolyte, C P t is the EDLC between Ptgate electrode and electrolyte, and C Q is the quantum capacitance of graphene [25]. The EDLCof graphene to electrolyte can be calculated as C Gr = A Gr (cid:15) r (cid:15) /λ D , with A Gr = 40 µ m × µ m = 4 × µ m being the area of graphene surface exposed to electrolyte, (cid:15) is the vacuumpermittivity, and (cid:15) r is the relative permittivity of PBS electrolyte, which is only slightly lowerthan the one of water, thus, taken as (cid:15) r ≈ [25], [41]. Lastly, λ D is the Debye length which givesthe thickness of the EDLC, and it can be approximated in aqueous solutions as λ D ≈ . / √ ρ ion in nm, with ρ ion being the ionic density in M [42]. For 1xPBS buffer, the ionic density is ∼ λ D ≈ . nm. The resulting EDLC for graphene is C Gr ≈ . nF. The EDLCbetween the Pt electrode and the electrolyte can be obtained similarly as C P t = A P t (cid:15) r (cid:15) /λ D .However, since the surface area of the Pt wire inside the electrolyte ( A P t = l P t d P t π/ cm × . mm × π/ . × µ m , calculated as the area of an half sphere of 1cm-length and0.5mm-diameter) is significantly larger than the graphene surface area ( A gr = 4 × µ m ), C P t can be neglected. Lastly, the quantum capacitance of graphene per unit area has been reported
DRAFT July 7, 20209 as c q ≈ µ F · cm − [25], [43], which gives C Q = c q × A gr = 8 × − nF. The overall gatecapacitance given by (3) then becomes C G ≈ . × − nF.The effective electric charge of a single immobilised 18-mer pDNA screened by the EDL canbe written as q pDNA = 18 × q e × e − r/λ D , (4)where q e is the elementary charge, r is the effective length of the pDNA taken as the half ofits length, i.e., r = 18 (basepairs) × . (nm/basepair) × / . nm, by assuming a verticalorientation for the single-stranded pDNAs following the analysis in [25], where similar solutiongate potentials and the same type of linker molecules are used.Finally, the surface density of the immobilised pDNAs can be written as a function of theaverage shift in V CNP upon pDNA functionalisation: n pDNA = ∆ V CNP C G q pDNA A gr , (5)which gives n pDNA ≈ × µ m − . A similar surface density ( ∼ . × µ m − ) for pDNAwas previously reported in [25]. B. Sensing Response
For determining the sensing characteristics of the MC receiver, complementary 18-mer targetDNA (tDNA: 5’-GCA ATA CGG TGA AGT CCT-3’, obtained from Sigma Aldrich) is preparedin 0.01xPBS solution. tDNAs of varying concentrations (50 nM, 100 nM, . . . µ M) aresuccessively flowed through the microfluidic channel in the order of increasing concentration,and I ds is recorded in real time with V ds = 100 mV and V g = 0 V. During the experiment, thevolumetric flow rate is held constant at u V = 80 µ l/min. The results of the measurements areprovided in Fig. 10(a), where a decrease in the drain-source current is observed with increasingtDNA concentration, implying n-type doping effect in contrast to the p-type doping of pDNAs.The n-type doping effect upon target DNA hybridisation or probe DNA immobilisation waspreviously reported in [26], [35], [39], [44]–[46], where the effect is mainly attributed to thepartial interaction of the DNAs with the graphene surface through π − π stacking of the nucleobasearomatic rings resulting in direct electron transfer to graphene instead of electrostatic gating.The electron transfer from DNA upon immobilisation was also reported for CNT transistors July 7, 2020 DRAFT0 [47]. Moreover, it is known that DNA molecules immobilised on a surface with their singleend can be stretched in parallel to the surface under lateral flow [48], and the extent of thisconformational change can be increased by a positive surface potential attracting the negativelycharged DNA molecules to the surface [38], [46]. Note that in our case, the graphene surface iscontinuously exposed to a positive drain-source bias during the sensing measurements. Therefore,we speculate that the conformational change resulting from microfluidic flow, hybridisation, andthe positive surface charge of graphene brings the hybridised DNA molecules closer to thegraphene surface, causing their partial interaction and electron transfer. Also, compared to themeasurements taken with the single-stranded pDNAs under no-flow conditions, the ionic strengthof the electrolyte (0.01 × PBS), in which the sensing experiments are performed, is significantlylower (compared to 1 × PBS), such that the attractive electrostatic force caused by the positivesurface potential extends more into the electrolyte without being significantly screened [49].Given that the hybridised dsDNAs carry twice the amount of negative charge of the ssDNAs,it can be considered that the hybridised pDNA-tDNA pairs, in our case, are more stronglyattracted to the graphene surface compared to pDNAs [50]. The stronger electrostatic attractioncombined with the stretching effect of lateral microfluidic flow supports our argument. However,this requires further confirmation, potentially through molecular dynamics simulations of bothssDNAs and hybridised dsDNAs under similar conditions to understand the effect of lateral flow,surface potential, solution-gate potential, and the linker molecules.Each working concentration of tDNAs were propagated in the channel until I ds reaches aplateau. The value of I ds at these plateaus are used to construct the sensing response graph ofthe MC receiver, which is provided in Fig. 10(b). The response curve is fitted by the Langmuiradsorption isotherm, i.e., ∆ I ds / ∆ I ds,sat = 11 + K D /C tDNA , (6)where ∆ I ds,sat is the receiver response in saturation, which occurs when all the probe DNAsare hybridised. K D is the dissociation constant of pDNA-tDNA hybridisation, and C tDNA is theapplied concentration of tDNAs. The curve fitting gives the dissociation constant as K D = 730 nM for the DNA hybridisation on the fabricated MC receiver, and the receiver response atsaturation as ∆ I ds,sat = 1 . µ A. DRAFT July 7, 20201
Time (s) D r a i n - S ou r c e C u rr en t I d s ( A ) Baseline50 nM100 nM250 nM500 nM1 M2 M5 M10 M (a) tDNA concentration ( M) R e c e i v e r R e s pon s e | I d s | ( A ) Langmuir fit (b)
Fig. 10: (a) Real-time sensing response of the MC receiver in terms of drain-source current I ds with varying concentration of complementary target DNAs (tDNAs). (b) Equilibrium sensingresponse fitted by the Langmuir adsorption isotherm. Resulting dissociation constant for pDNA-tDNA hybridisation is K D = 730 nM. C. Specificity
The specificity of the MC receiver against the complementary tDNAs is evaluated by com-paring the receiver’s response to different ssDNAs, which are not complementary to the pDNAs.The ultimate specificity can be determined with the application of an ssDNA having only onesingle base-pair mismatch. For this, we use 18-mer ntDNA with the base sequence 5’-GCA ATACGG CGA AGT CCT-3’, which has the mismatch in its 10 th base pair, where T to C mutationoccurs. Another test is performed with the application of 18-mer ntDNA (5’-GCA CGT CGGCGT CGT CAT-3’), which has 7 base-pair mismatches. Complementary tDNA is also appliedfor comparison. All DNAs are dissolved in 0.01xPBS with 1 µ M working concentration. Themeasurement results before and after a moving mean filter of 21-second window length is appliedare provided in Fig. 11(a).The response curve of the MC receiver for different DNAs is fitted by the Langmuir model ofadsorption to determine the kinetic rates of the DNA hybridisation for the three DNA sequences.The solution of the Langmuir model gives the time-varying response of the MC receiver during
July 7, 2020 DRAFT2
Time (s) N o r m a li z ed D r a i n - S ou r c e C u rr en t I d s tDNAntDNA ntDNA (a) Time (s) N o r m a li z ed D r a i n - S ou r c e C u rr en t I d s tDNAntDNA ntDNA (b) Fig. 11: Specificity analysis of the MC receiver. (a) Real-time sensing response for complemen-tary tDNA, non-complementary ntDNA with single base-pair mismatch, and non-complementaryntDNA with 7 base-pair mismatches (see Table I). At t ≈ s, the DNA solutions arereplaced with 0.01xPBS solution to allow dissociation of the hybridised DNAs. (b) Real-timesensing response fitted by the Langmuir adsorption/desorption model, equations (7)-(8).TABLE I: Kinetic constants of DNA hybridisation measured by MC receiver DNA Sequence Binding Unbinding Dissociationrate k + rate k − constant K D (M − s − ) ( × − s − ) ( × − M)tDNA ntDNA ntDNA the association and dissociation phases of DNA hybridisation [25] as follows ∆ I ds ( t ) = ∆ I ds,eq (cid:0) − e − ( k + c in + k − ) t (cid:1) , for ≤ t ≤ t d , (7) ∆ I ds ( t ) = ∆ I ds ( t d ) e − k − t , for t > t d , (8) DRAFT July 7, 20203 where, c in is the input concentration, which is set to c in = 1 µ M for all DNAs. t d denotes thetime of dissociation, and ∆ I ds,eq is the asymptotic value of the sensing response, which occurswhen the hybridisation reaches equilibrium. The variables to be fitted are the binding rate k + ,unbinding rate k − , and ∆ I ds,eq . The fitted response is plotted in Fig. 11(b), and the resultingkinetic rates are provided in Table I, which shows that the binding rate of the target DNAssubstantially decreases with increasing number of base-pair mismatches.The nonlinear curve fitting gives the dissociation constant of the tDNA as K D ( tDNA ) = k − ( tDNA ) /k + ( tDNA ) ≈ nM, which is very close to the value obtained by the fitting of thesensor response, i.e., 730 nM. On the other hand, we obtain higher dissociation constants fornon-complementary DNAs, i.e., K D ( ntDNA ) = 3 . µ M and K D ( ntDNA ) = 26 . µ M,indicating the specificity of the MC receiver against the complementary tDNAs.IV. C
OMMUNICATION P ERFORMANCE
Time-varying communication experiments are performed in the microfluidic testbed to revealthe detection performance of the MC receiver. For these experiments, both inlets are utilisedas shown in Fig. 6. One of the inlets is connected to the reservoir containing the 0.01xPBSbuffer solution, and the other one is connected to the reservoir containing the tDNA solutionin 0.01xPBS buffer. Manually controlled mechanical switches are utilised as they proved moreeffective in stopping the fluid flow into the channel than the digital control due to the fact thatthe pressure controller can drift out of calibration as the experiments progress. Accordingly, forthe transmission of tDNAs, the buffer flow is rapidly stopped through the mechanical switchalong the buffer line, and the tDNA line is opened at the same time. When the transmissionends, flow in the tDNA line is stopped, and the buffer flow is simultaneously started again bymeans of mechanical switching.
A. Time-varying Response
To determine the time-varying response of the receiver, varying length pulses of µ M tDNAsare flowed through the microfluidic channel. The results of three independent measurementstaken from the same channel are provided in Fig. 12(a) and Fig. 12(b) for 30-second and 60-second pulses, respectively. As with the previous cases, a 21-second-length moving mean filteris applied for each measurement.
July 7, 2020 DRAFT4
The time-varying response of the MC receiver to finite-length concentration pulses can bedescribed by the analytical microfluidic MC model developed in [51]. This approximate modelgives the receiver response in terms of number of bound receptors, i.e., N R ( t ) = N R,eq (cid:32) − W (cid:2) α ∗ exp ( α ∗ − β ∗ ( t − t a )) (cid:3) α ∗ (cid:33) (cid:16) Θ [ t − t a ] − Θ [ t − t d − (cid:15) ] (cid:17) (9) − γ ∗ W (cid:34) − N R, γ ∗ exp (cid:18) − k + N R, − k ∗ T k − ( t − t d ) k + γ ∗ (cid:19) (cid:35) Θ [ t − t d − (cid:15) ] , where t a and t d denote the start times of association (i.e., pDNA-tDNA hybridisation in thiscase) and dissociation phases, respectively, N R ( t ) is the number of bound receptors at time t , N R,eq is the number of bound receptors at equilibrium, N R, = N R ( t d ) is the number of boundreceptors when the dissociation starts. Here W [ . ] denotes the principal branch of the Lambert W function, and Θ[ . ] denotes Heaviside step function with Θ[0] = 1 . The parameters α ∗ , β ∗ and γ ∗ are provided in [51, Eqs. (37-39)] as functions of N R,max , which is the upper limit ofthe receiver response in terms of bound number of receptors. In theory, N R,max can be taken asequal to the total number of receptors. Lastly, the transport parameter k ∗ T incorporating the effectof the microfluidic channel geometry and flow velocity on the molecular transport dynamics isalso provided in [51, Eq. (13)] .To use this model for fitting the electrical measurements obtained by the graphene-based MCreceiver, we need to transform the variables in (9) to obtain a function of ∆ I ds . The number ofbound receptors is proportional to the change in the drain-source current through the followingrelation: N R ( t ) = ∆ I ds ( t ) /Q tDNA , (10)with Q tDNA = g m q tDNA C G . Here, g m = ∂I ds /∂V g is the transconductance of the device, q tDNA isthe effective charge of a single tDNA molecule, and C G, . xPBS is the total gate capacitance in0.01xPBS buffer. Gate capacitance is obtained as C G, . xPBS = 6 . × − nF using (3) with thenew Debye length in diluted PBS electrolyte, i.e., λ D, . xPBS = 7 . nm. The effective chargeof a single tDNA is obtained similarly using (4) with λ D, . xPBS . The transconductance of thedevice g m = ∂I ds /∂V g before the communication experiments can be calculated by a linearapproximation of the V g − I ds curve around V g = 0 V bias. Averaging over the transfer curvesof the four graphene channels obtained at the last step of functionalisation in 0.01xPBS, shown
DRAFT July 7, 20205
Time (s) N o r m a li z ed D r a i n - S ou r c e C u rr en t I d s (a) Time (s) N o r m a li z ed D r a i n - S ou r c e C u rr en t I d s (b) Time (s) N o r m a li z ed D r a i n - S ou r c e C u rr en t I d s (c) Time (s) N o r m a li z ed D r a i n - S ou r c e C u rr en t I d s (d) Fig. 12: Normalised pulse response of the MC receiver in terms of drain-source current I ds with constant operating voltages set to V g = 0 V and V ds = 0 . V. (a, b) Three independentmeasurements taken from the same channel for 30-second-long and 60-second-long 1 µ M tDNApulses, respectively. (c, d) Pulse responses fitted by the normalised output of the transformedmodel given in (13).in Fig. 9, we obtain g m ≈ − . ± . µ A/V. Similarly, the following transformation is made: N R,eq = ∆ I ds,eq /Q tDNA . (11)Here, ∆ I ds,eq is the response of the receiver to 1 µ M tDNA at equilibrium, and obtained from
July 7, 2020 DRAFT6 the sensing response given in Fig. 10 as ∆ I ds,eq = − . µ A. Note that N R,max can also becalculated as a function of ∆ I ds,eq as follows N R,max = (( c avg + K D ) /c avg ) × N R,eq (12) = (( c avg + K D ) /c avg ) × ∆ I ds,eq /Q tDNA , with c avg being the average tDNA concentration passing over the receiver surface [51]. Thedissociation constant K D for pDNA-tDNA pair given in Table I. Accordingly, (9) can be rewrittenby substituting (10), (11), and (12) into (9) as follows ∆ I ds ( t ) =∆ I ds,eq (cid:32) − W (cid:2) α ∗ exp ( α ∗ − β ∗ ( t − t a )) (cid:3) α ∗ (cid:33) (cid:16) Θ [ t − t a ] − Θ [ t − t d − (cid:15) ] (cid:17) (13) − γ ∗ Q W (cid:34) − ∆ I ds ( t d ) γ ∗ Q exp (cid:18) − k + ∆ I ds ( t d ) /Q − k ∗ T k − ( t − t d ) k + γ ∗ (cid:19) (cid:35) Θ [ t − t d − (cid:15) ] , with the transformed parameters given as α ∗ = k + c avg ∆ I ds,eq /Qk − ∆ I ds,eq /Q + k ∗ T c avg , (14) β ∗ = k + c avg + k − k − ∆ I ds,eq /Qk ∗ T c avg , (15) γ ∗ = ( c avg + K D )∆ I ds,eq /Qc avg + k ∗ T k + . (16)The model developed in [51] includes many other input parameters concerning the microfluidicchannel geometry and molecular transport dynamics. The linear flow velocity is already obtainedas u = 220 µ m/s by transforming the volumetric flow rate u V = 80 µ l/min. Diffusion coefficientof tDNAs is taken as D = 100 µ m /s, which is in the range of previously reported values forssDNAs of similar lengths [52], [53].For comparison to the empirical MC signals, the model response is normalised by the meanempirical baseline current ( I ds,baseline = 31 . µ A, as plotted in Fig. 10(a)), i.e., ˆ I ds ( t ) = ( I ds,baseline + ∆ I ds ( t )) /I ds,baseline . (17)The normalised output of the transformed model for pulse lengths T p = 30 s and T p = 60 s isshown in Fig. 12(c) and Fig. 12(d), respectively, presented in comparison to the mean of threeindependent measurements taken for each pulse length. The error bars represent the standard DRAFT July 7, 20207
Time (s) N o r m a li z ed D r a i n - S ou r c e C u rr en t I d s Time (s) N o r m a li z ed D r a i n - S ou r c e C u rr en t I d s Time (s) N o r m a li z ed D r a i n - S ou r c e C u rr en t I d s (a)(b)(c) Fig. 13: Normalised receiver response for binary data transmission with fixed pulse length T p =30 s and varying bit intervals: (a) T s = 60 s, (b) T s = 120 s, (c) T s = 360 s. The grey linesdenote the received MC signals normalised by the baseline current, and the solid red lines aretheir low-pass filtered version by a moving average filter of 21-second length in MATLAB.Solid blue lines represent the normalised output of the model given in (17). Dashed orange linesindicate the time instants when bit-1 is transmitted, i.e., when the mechanical switch in the tDNAline is opened for 30 seconds. July 7, 2020 DRAFT8 deviation of the measurements. We can observe that the approximate model is highly accuratein calculating the propagation delay and I ds during the association phase. Although its accuracyis lower during the dissociation phase, it well captures the pulse amplitude and the pulse width,which are the two important parameters in evaluating the performance of a communicationsystem. The same model will be applied in the next section for binary data transmission. B. Data Transmission
To reveal the detection performance of the MC receiver, 20-bit-long pseudorandom binaryinformation encoded into the concentration of tDNAs is transmitted through the microfluidicchannel and detected by the fabricated MC receiver located at the bottom of the channel. Herebit-1 is represented by a 30-second pulse of µ M tDNA at the beginning of a bit interval, andbit-0 is represented by no pulse transmission during the entire bit interval, i.e., the buffer linestays connected into the microfluidic channel. The results are provided for bit intervals of 60-,120-, and 360-second length in Fig. 13. As expected, the slow rate of dissociation of the boundtDNAs from the immobilised pDNAs results in a significant amount of intersymbol interference(ISI), which can potentially complicate the decoding at the receiver. The ISI is more pronouncedfor shorter bit intervals, i.e., higher transmission rates, such that the baseline could not findenough time to recover. On the other hand, for 360-second-long bit interval (Fig. 13(c)), ISI issignificantly lower and I ds is able to return to the baseline.Another important observation is that the propagation delay t delay between the transmissiontime of the bit-1 and the receiver response is ∼
55 seconds in each test, and the delay remainsalmost constant during the entire data transmission period. This implies that the application ofa constant delay shift in the receiver response can be sufficient for the synchronisation of thereceiver with the transmitter during the decoding process.As the receiver response is away from saturation in the performed tests, we can assume thatthe response is linear and time-invariant (LTI). Based on this assumption, we can reconstructthe received signal by applying the superposition principle of LTI systems in the approximatemodel developed in [51]. Accordingly, using (13), the received signal can be written as R ( t ) = L (cid:88) i =1 s [ i ]∆ I ds ( t − t transmit − ( i − T b ) , (18) DRAFT July 7, 20209 where L = 20 is the number of transmitted bits, t transmit is the start time of transmission, T b isthe bit interval, and s [ i ] ∈ { , } is the i th transmitted bit. Similar to (17), the model responseis normalised by the mean empirical baseline current ( I ds,baseline = 31 . µ A), i.e., ˆ I ds ( t ) = ( I ds,baseline + R ( t )) /I ds,baseline . (19)The normalised model response is plotted in Fig. 13 over the empirical MC signal for varying bitintervals. We can see that the approximate model is very accurate in capturing the transmissiondelay and the overall trend of I ds , although it is not able to exactly reconstruct the original signal.The deviations can largely be attributed to the LTI approximation in writing (18), which neglectsthe effect of previously transmitted bits on the received signal corresponding to the current bit.Based on the observation of constant delay during data transmission, the delay-shifted versionof the bit intervals is indicated with dashed lines in Fig. 14 with the transmitted bits written insideeach bit interval. As the envisioned MC applications demand low-complexity communicationtechniques due to the resource and size limitations of the communicating nanodevices, constantthreshold detection has been favoured in the literature. In this scheme, the received signal issampled at a predefined sampling time instant, and its amplitude is compared to a thresholdvalue for deciding between bit-0 and bit-1. However, the high ISI of the MC channel, and theresulting drift of baseline, observed in these experiments, render the constant threshold detectionmethods ineffective, especially at high communication rates. Therefore, in line with some ofthe previous work in the MC literature concerning ligand-receptor binding systems, we utilise asimple difference-based detection method, which decodes the information based on the differencein I ds measurements taken at the start and end points of a bit interval. Recall that the receiver issynchronised with the transmitter through the application of a constant delay shift at the receiverside. The sampling time points are demonstrated in Fig. 14 with green dots. Accordingly, thedecoded bits can be written as follows ˆ s [ i ] = L (cid:88) i =1 s [ i ] (cid:2) r [ i + 1] − r [ i ] < (cid:3) , (20)where ˆ s [ i ] is the decoded bit, and r [ i ] = I ds ( t − t transmit − t delay − ( i − T s ) is the discrete timesample of I ds at decision points, and (cid:2) . (cid:3) is the indicator function, which outputs 1 if the insideexpression is true.Applying the difference-based detection method on the unfiltered MC signal yielded biterror rate (BER). We observed 1 bit error for 60 s bit interval, and 2 bit errors for 360 s July 7, 2020 DRAFT0
Time (s)
Time (s) N o r m a li z ed D r a i n - S ou r c e C u rr en t I d s S t a r t o f T r an s m i ss i on MC SignalFiltered MC Signal
Time (s) N o r m a li z ed D r a i n - S ou r c e C u rr en t I d s S t a r t o f T r an s m i ss i on MC SignalFiltered MC Signal
Time (s) N o r m a li z ed D r a i n - S ou r c e C u rr en t I d s S t a r t o f T r an s m i ss i on MC SignalFiltered MC Signal (a)(b)(c)
Fig. 14: Delay-shifted version of the receiver response for binary data transmission for varyingbit intervals: (a) T s = 60 s, (b) T s = 120 s, (c) T s = 360 s. Dashed orange line denotes thestart time of the data transmission, and grey dashed lines demarcate the individual bit intervals.Green dots on the received MC signal indicates the sampled current values for difference-baseddetection method with the decoding rule formulated in (20). Transmitted/decoded bits are notedabove each bit interval, with the red-coloured ones denoting the erroneously decoded bits. DRAFT July 7, 20201 -0.2 0 0.2 0.4 0.6 0.8 1 1.2
Gate Voltage V g (V) D r a i n - S ou r ce C u rr en t I d s ( A ) ForwardBackward (a) -0.2 0 0.2 0.4 0.6 0.8 1 1.2
Gate Voltage V g (V) D r a i n - S ou r ce C u rr en t I d s ( A ) ForwardBackward (b) -0.2 0 0.2 0.4 0.6 0.8 1 1.2
Gate Voltage V g (V) D r a i n - S ou r ce C u rr en t I d s ( A ) ForwardBackward (c) -0.2 0 0.2 0.4 0.6 0.8 1 1.2
Gate Voltage V g (V) D r a i n - S ou r ce C u rr en t I d s ( A ) ForwardBackward (d)
Fig. 15: Hysteresis analysis of the MC receiver: Drain-source current I ds with forward andbackward sweep of gate voltage V g at 140 mV/s sweep rate. (a) Before functionalisation. (b)After functionalisation with PBASE. (c) After immobilization of pDNA. (d) After passivationwith ethanolamine.bit intervals. The decoding of the unfiltered MC signal of 120 s bit interval yielded no error.The erroneous transmissions are indicated in Fig. 14. On the other hand, the difference-baseddetection applied on the low-pass filtered MC signal correctly decoded all the transmitted bits. July 7, 2020 DRAFT2 -0.2 0 0.2 0.4 0.6 0.8 1 1.2
Gate Voltage V g (V) -202468101214 Lea k age C u rr en t I g s ( n A ) Fig. 16: Leakage current analysis of the MC receiver: Gate-source current I gs with varying gatevoltage V g . V. C ONCLUSION
This proof-of-concept study reports the very first fabrication and characterisation of a nanoscalegraphene bioFET-based MC receiver. The ICT tests of the MC receiver is performed in a custom-designed micro/nanoscale MC system using a PDMS-based microfluidic channel as the control-lable propagation medium, and a pressure-regulated flow control system for transmitting binarydata encoded into the concentration of target DNA molecules. The time-varying hybridisationof the information-carrying target DNAs with the probe DNAs immobilised on the SLG surfaceof the MC receiver is selectively transduced into a change in drain-source current over the SLGchannel. In light of the experimental sensing results, this transduction mechanism is speculatedto be related to both electrostatic gating and direct electron transfer. The ICT performance ofthe MC receiver is evaluated by binary data transmissions at different bit intervals. The responseof the MC receiver is well-fitted with the previously developed approximate analytical modelof microfluidic MC. The slow hybridisation kinetics of DNA molecules is revealed to causesignificant ISI. A simple difference-based detection method is shown to overcome the ISI toa significant extent, and provide reliable detection performance. The fabricated graphene-based
DRAFT July 7, 20203 nanoscale MC receiver and the overall microfluidic MC system can be used as an experimentaltestbed for probing intricate dynamics of MC, and developing novel communication techniques,transceiver architectures, and applications for MC.A
PPENDIX AS UPPLEMENTARY I NFORMATION
This Appendix provides additional information the hysteresis and leakage current analysesof the MC receiver in Figs. 15-16. Moreover, computer-aided designs (CADs) for the opticallithography of the MC receiver and for the 3d printing of PDMS mould are provided in Figs.17-18.
July 7, 2020 DRAFT4 S L G l a y e r C o n t a c t l a y e r I n s u l a t o r l a y e r μ m μ m μ m μ m μ m μ m . mm F i g . : C o m pu t e r- a i d e dd e s i gn f o r op ti ca llit hog r a phyo f M C r ece i v e r . DRAFT July 7, 20205 mm mm mm . mm mm mm mm ( a )( b ) ( c ) F i g . : ( a ) C o m pu t e r- a i d e dd e s i gn f o r r i n ti ngo f P D M S m ou l d . ( b ) - p r i n t e d P D M S m ou l d . ( c ) P D M S m i c r o fl u i d i cc h a nn e l w it h t h e P t w i r e m oun t e don t op a s t h e g a t ee l ec t r od e . July 7, 2020 DRAFT6 R EFERENCES [1] I. F. Akyildiz and J. M. Jornet, “The internet of nano-things,”
IEEE Wireless Communications , vol. 17, no. 6, 2010.[2] I. Akyildiz, M. Pierobon, S. Balasubramaniam, and Y. Koucheryavy, “The Internet of Bio-Nano Things,”
IEEE Commu-nications Magazine , vol. 53, no. 3, pp. 32–40, 2015.[3] E. Dinc, M. Kuscu, B. A. Bilgin, and O. B. Akan, “Internet of everything: A unifying framework beyond internet ofthings,” in
Harnessing the Internet of Everything (IoE) for Accelerated Innovation Opportunities . IGI Global, 2019, pp.1–30.[4] O. B. Akan, H. Ramezani, T. Khan, N. A. Abbasi, and M. Kuscu, “Fundamentals of molecular information andcommunication science,”
Proceedings of the IEEE , vol. 105, no. 2, pp. 306–318, 2017.[5] M. Pierobon, I. F. Akyildiz et al. , “Diffusion-based noise analysis for molecular communication in nanonetworks,”
IEEETransactions on Signal Processing , vol. 59, no. 6, pp. 2532–2547, 2011.[6] M. Pierobon and I. F. Akyildiz, “Noise analysis in ligand-binding reception for molecular communication in nanonetworks,”
IEEE Transactions on Signal Processing , vol. 59, no. 9, pp. 4168–4182, 2011.[7] M. S. Kuran, H. B. Yilmaz, T. Tugcu, and I. F. Akyildiz, “Modulation techniques for communication via diffusion innanonetworks,” in
Communications (ICC), 2011 IEEE International Conference on . IEEE, 2011, pp. 1–5.[8] M. Kuscu and O. B. Akan, “Modeling and analysis of sinw fet-based molecular communication receiver,”
IEEE Transactionson Communications , vol. 64, no. 9, pp. 3708–3721, 2016.[9] B. A. Bilgin, E. Dinc, and O. B. Akan, “DNA-based molecular communications,”
IEEE Access , vol. 6, pp. 73 119–73 129,2018.[10] M. Kuscu and O. B. Akan, “Maximum likelihood detection with ligand receptors for diffusion-based molecularcommunications in Internet of Bio-Nano Things,”
IEEE Transactions on Nanobioscience , vol. 17, no. 1, pp. 44–54, 2018.[11] ——, “Channel sensing in molecular communications with single type of ligand receptors,”
IEEE Transactions onCommunications , vol. 67, no. 10, pp. 6868–6884, 2019.[12] M. Kuscu, E. Dinc, B. A. Bilgin, H. Ramezani, and O. B. Akan, “Transmitter and receiver architectures for molecularcommunications: A survey on physical design with modulation, coding, and detection techniques,”
Proceedings of theIEEE , vol. 107, no. 7, pp. 1302–1341, 2019.[13] V. Jamali, A. Ahmadzadeh, W. Wicke, A. Noel, and R. Schober, “Channel modeling for diffusive molecular communicationatutorial review,”
Proceedings of the IEEE , vol. 107, no. 7, pp. 1256–1301, 2019.[14] N. Farsad, W. Guo, and A. W. Eckford, “Tabletop molecular communication: Text messages through chemical signals,”
PloS One , vol. 8, no. 12, p. e82935, 2013.[15] N.-R. Kim, N. Farsad, C.-B. Chae, and A. W. Eckford, “A universal channel model for molecular communication systemswith metal-oxide detectors,” in
Communications (ICC), 2015 IEEE International Conference on . IEEE, 2015, pp. 1054–1059.[16] B.-H. Koo, C. Lee, H. B. Yilmaz, N. Farsad, A. Eckford, and C.-B. Chae, “Molecular mimo: From theory to prototype,”
IEEE Journal on Selected Areas in Communications , vol. 34, no. 3, pp. 600–614, 2016.[17] N. Farsad, D. Pan, and A. Goldsmith, “A novel experimental platform for in-vessel multi-chemical molecular communi-cations,” in
GLOBECOM 2017-2017 IEEE Global Communications Conference . IEEE, 2017, pp. 1–6.[18] H. Unterweger, J. Kirchner, W. Wicke, A. Ahmadzadeh, D. Ahmed, V. Jamali, C. Alexiou, G. Fischer, and R. Schober,“Experimental molecular communication testbed based on magnetic nanoparticles in duct flow,” in . IEEE, 2018, pp. 1–5.
DRAFT July 7, 20207 [19] W. Wicke, A. Ahmadzadeh, V. Jamali, H. Unterweger, C. Alexiou, and R. Schober, “Magnetic nanoparticle-based molecularcommunication in microfluidic environments,”
IEEE Transactions on Nanobioscience , vol. 18, no. 2, pp. 156–169, 2019.[20] A. C. Ferrari, F. Bonaccorso, V. Fal’Ko, K. S. Novoselov, S. Roche, P. Bøggild, S. Borini, F. H. Koppens, V. Palermo,N. Pugno et al. , “Science and technology roadmap for graphene, related two-dimensional crystals, and hybrid systems,”
Nanoscale , vol. 7, no. 11, pp. 4598–4810, 2015.[21] Y. Huang, X. Dong, Y. Shi, C. M. Li, L.-J. Li, and P. Chen, “Nanoelectronic biosensors based on CVD grown graphene,”
Nanoscale , vol. 2, no. 8, pp. 1485–1488, 2010.[22] B. Zhan, C. Li, J. Yang, G. Jenkins, W. Huang, and X. Dong, “Graphene field-effect transistor and its application forelectronic sensing,”
Small , vol. 10, no. 20, pp. 4042–4065, 2014.[23] Y. H. Kwak, D. S. Choi, Y. N. Kim, H. Kim, D. H. Yoon, S.-S. Ahn, J.-W. Yang, W. S. Yang, and S. Seo, “Flexibleglucose sensor using cvd-grown graphene-based field effect transistor,”
Biosensors and Bioelectronics , vol. 37, no. 1, pp.82–87, 2012.[24] Y. Ohno, K. Maehashi, Y. Yamashiro, and K. Matsumoto, “Electrolyte-gated graphene field-effect transistors for detectingph and protein adsorption,”
Nano Letters , vol. 9, no. 9, pp. 3318–3322, 2009.[25] S. Xu, J. Zhan, B. Man, S. Jiang, W. Yue, S. Gao, C. Guo, H. Liu, Z. Li, J. Wang et al. , “Real-time reliable determinationof binding kinetics of DNA hybridization using a multi-channel graphene biosensor,”
Nature Communications , vol. 8, p.14902, 2017.[26] R. Campos, J. Borme, J. R. Guerreiro, G. Machado Jr, M. F. Cerqueira, D. Y. Petrovykh, and P. Alpuim, “Attomolar label-free detection of dna hybridization with electrolyte-gated graphene field-effect transistors,”
ACS Sensors , vol. 4, no. 2, pp.286–293, 2019.[27] M. Kuscu and O. B. Akan, “On the physical design of molecular communication receiver based on nanoscale biosensors,”
IEEE Sensors Journal , vol. 16, no. 8, pp. 2228–2243, 2016.[28] N. S. Green and M. L. Norton, “Interactions of dna with graphene and sensing applications of graphene field-effecttransistor devices: A review,”
Analytica Chimica Acta , vol. 853, pp. 127–142, 2015.[29] D. Malak and O. B. Akan, “Molecular communication nanonetworks inside human body,”
Nano Communication Networks ,vol. 3, no. 1, pp. 19–35, 2012.[30] Y. Chahibi and I. F. Akyildiz, “Molecular communication noise and capacity analysis for particulate drug delivery systems,”
IEEE Transactions on Communications , vol. 62, no. 11, pp. 3891–3903, 2014.[31] Y. Deng, A. Noel, M. Elkashlan, A. Nallanathan, and K. C. Cheung, “Modeling and simulation of molecular communicationsystems with a reversible adsorption receiver,”
IEEE Transactions on Molecular, Biological and Multi-Scale Communica-tions , vol. 1, no. 4, pp. 347–362, 2015.[32] B. Li, M. Sun, S. Wang, W. Guo, and C. Zhao, “Low-complexity noncoherent signal detection for nanoscale molecularcommunications,”
IEEE Transactions on Nanobioscience , vol. 15, no. 1, pp. 3–10, 2015.[33] B.-h. Chueh, D. Huh, C. R. Kyrtsos, T. Houssin, N. Futai, and S. Takayama, “Leakage-free bonding of porous membranesinto layered microfluidic array systems,”
Analytical Chemistry , vol. 79, no. 9, pp. 3504–3508, 2007.[34] A. O. Bicen and I. F. Akyildiz, “System-theoretic analysis and least-squares design of microfluidic channels for flow-induced molecular communication,”
IEEE Transactions on Signal Processing , vol. 61, no. 20, pp. 5000–5013, 2013.[35] W. Yue, C. Tang, C. Wang, C. Bai, S. Liu, X. Xie, H. Hua, Z. Zhang, and D. Li, “An electricity-fluorescence double-checking biosensor based on graphene for detection of binding kinetics of dna hybridization,”
RSC Advances , vol. 7, no. 70,pp. 44 559–44 567, 2017.
July 7, 2020 DRAFT8 [36] M. T. Hwang, P. B. Landon, J. Lee, D. Choi, A. H. Mo, G. Glinsky, and R. Lal, “Highly specific snp detection using 2dgraphene electronics and dna strand displacement,”
Proceedings of the National Academy of Sciences , vol. 113, no. 26,pp. 7088–7093, 2016.[37] G. Wu, X. Tang, M. Meyyappan, and K. W. C. Lai, “Doping effects of surface functionalization on graphene with aromaticmolecule and organic solvents,”
Applied Surface Science , vol. 425, pp. 713–721, 2017.[38] M. Kabel´aˇc, O. Kroutil, M. Pˇredota, F. Lankaˇs, and M. ˇS´ıp, “Influence of a charged graphene surface on the orientation andconformation of covalently attached oligonucleotides: a molecular dynamics study,”
Physical Chemistry Chemical Physics ,vol. 14, no. 12, pp. 4217–4229, 2012.[39] S. Ghosh, N. I. Khan, J. G. Tsavalas, and E. Song, “selective detection of lysozyme biomarker utilizing large area chemicalvapor deposition-grown graphene-based field-effect transistor,”
Frontiers in Bioengineering and Biotechnology , vol. 6, p. 29,2018.[40] T.-Y. Chen, P. T. K. Loan, C.-L. Hsu, Y.-H. Lee, J. T.-W. Wang, K.-H. Wei, C.-T. Lin, and L.-J. Li, “Label-free detectionof dna hybridization using transistors based on cvd grown graphene,”
Biosensors and Bioelectronics , vol. 41, pp. 103–109,2013.[41] Y. Zheng, J. Nguyen, C. Wang, and Y. Sun, “Electrical measurement of red blood cell deformability on a microfluidicdevice,”
Lab on a Chip , vol. 13, no. 16, pp. 3275–3283, 2013.[42] B. R. Goldsmith, L. Locascio, Y. Gao, M. Lerner, A. Walker, J. Lerner, J. Kyaw, A. Shue, S. Afsahi, D. Pan et al. , “Digitalbiosensing by foundry-fabricated graphene sensors,”
Scientific Reports , vol. 9, no. 1, pp. 1–10, 2019.[43] I. Meric, M. Y. Han, A. F. Young, B. Ozyilmaz, P. Kim, and K. L. Shepard, “Current saturation in zero-bandgap, top-gatedgraphene field-effect transistors,”
Nature Nanotechnology , vol. 3, no. 11, p. 654, 2008.[44] A. K. Manoharan, S. Chinnathambi, R. Jayavel, and N. Hanagata, “Simplified detection of the hybridized DNA using agraphene field effect transistor,”
Science and Technology of Advanced Materials , vol. 18, no. 1, pp. 43–50, 2017.[45] X. Dong, Y. Shi, W. Huang, P. Chen, and L.-J. Li, “Electrical detection of dna hybridization with single-base specificityusing transistors based on cvd-grown graphene sheets,”
Advanced Materials , vol. 22, no. 14, pp. 1649–1653, 2010.[46] G. Xu, J. Abbott, L. Qin, K. Y. Yeung, Y. Song, H. Yoon, J. Kong, and D. Ham, “Electrophoretic and field-effect graphenefor all-electrical dna array technology,”
Nature Communications , vol. 5, p. 4866, 2014.[47] A. Star, E. Tu, J. Niemann, J.-C. P. Gabriel, C. S. Joiner, and C. Valcke, “Label-free detection of dna hybridization usingcarbon nanotube network field-effect transistors,”
Proceedings of the National Academy of Sciences , vol. 103, no. 4, pp.921–926, 2006.[48] J. van Mameren, E. J. Peterman, and G. J. Wuite, “See me, feel me: methods to concurrently visualize and manipulatesingle dna molecules and associated proteins,”
Nucleic Acids Research , vol. 36, no. 13, pp. 4381–4389, 2008.[49] W. Kaiser and U. Rant, “Conformations of end-tethered dna molecules on gold surfaces: influences of applied electricpotential, electrolyte screening, and temperature,”
Journal of the American Chemical Society , vol. 132, no. 23, pp. 7935–7945, 2010.[50] U. Rant, K. Arinaga, S. Fujita, N. Yokoyama, G. Abstreiter, and M. Tornow, “Dynamic electrical switching of dna layerson a metal surface,”
Nano Letters , vol. 4, no. 12, pp. 2441–2445, 2004.[51] M. Kuscu and O. B. Akan, “Modeling convection-diffusion-reaction systems for microfluidic molecular communicationswith surface-based receivers in Internet of Bio-Nano Things,”
PloS One , vol. 13, no. 2, p. e0192202, 2018.[52] E. Stellwagen and N. C. Stellwagen, “Determining the electrophoretic mobility and translational diffusion coefficients ofdna molecules in free solution,”
Electrophoresis , vol. 23, no. 16, pp. 2794–2803, 2002.
DRAFT July 7, 20209 [53] A. E. Nkodo, J. M. Garnier, B. Tinland, H. Ren, C. Desruisseaux, L. C. McCormick, G. Drouin, and G. W. Slater, “Diffusioncoefficient of dna molecules during free solution electrophoresis,”
Electrophoresis , vol. 22, no. 12, pp. 2424–2432, 2001., vol. 22, no. 12, pp. 2424–2432, 2001.