Applications of graphene to cell biology 2

3.1.2 Sample deviceI made a very simple and large device consisting of a ≈15 µm membrane of PDMS sandwiched between thick 1 mm-wide perpendicular lines of gold.First, bottom electrodes a

Chapter 3

Early sensor attempts

This chapter describes my attempts at developing a graphene-baseddevice that would allow measurements of forces exerted by cells on theirsubstrate As covered in Section 1.4, getting a better understanding ofthis process is useful for answering a number of biological processes, andexisting techniques have limitations

The basic concept behind of all these approaches is to build a parent, electronic device The transparency allows optical imaging of cellswith a microscope, while substrate deformation, and therefore forces ap-plied, are transceived into electrical signals Most of these ideas could befirst experimented with using gold electrodes, which are fairly opaque tolight [43], and moving to graphene electrodes later on when a proof ofconcept has been shown to be functional

trans-I describe 5 possible ways of converting forces into electrical signals:capacitive, piezoresistive, using graphene itself as a strain gauge, piezoelec-tric, and quartz crystal micro-balance principle (QCM)

All these approaches, except the one using graphene as a strain gauge,

Graphene Polymer film

Figure 3.1: Basic 2 layer grid design Cells and their medium are put ontop of the sensor The polymer film can be PDMS, PDMS withconductive particles, or PVDF, depending on the approach taken It can

be replaced by a Quartz crystal for the QCM approach

use 2 sets of parallel stripes of graphene, perpendicular to each other, arated with a thin film (see Figure 3.1) The thin film can be either adeformable dielectric, a piezoresistive/piezoelectric film, or a quartz crys-tal for the QCM

sep-These methods are mainly thought to measure forces perpendicular tothe surface (normal forces), but will probably also be sensitive to forces par-allel to the surface (Shear forces) A summary of the different approaches,with their advantages and disadvantages, is provided in Table 3.1

Some approaches were investigated in details, with some experimentsperformed, while others did not leave the design stage One of the mainchallenge is that electrical signals are very weak when we attempt to shrinkthe device to the sort of resolution desired for our device (≈ 1 µm resolu-

50

tion) I even considered looking at larger cells (Amoeba, see Appendix B),but the biological significance of studying such cells is less clear While ul-timately unsuccessful, these experiments led to the more interesting devicepresented in Chapter 4.

3.1 Capacitive

Figure 3.1 shows the basic capacitive sensing design: perpendicular lines

of electrodes creates a matrix of capacitors, that can be read individually.Each intersection of graphene stripes forms a parallel plate capacitor, whosecapacitance C is expressed by:

C = r0· A

where A is the plates area, d the distance between the plates, 0 is thedielectric constant (8.854 · 10−12 [Fm−1]), and r is the material-dependentrelative permittivity

When a pressure P (in pascals) is applied on the capacitor, the dielectric

is compressed, and the distance between the places is reduced by

∆d

P

E,

where E is the Young’s modulus of the dielectric

This deformation leads to a change in capacitance ∆C:

Using PDMS as dielectric (r ≈ 3 [24]), numerical computations forthe base capacitance lead to the results in Table 3.2 Even large area(100 × 100 µm), and small spacing (0.1 µm), produces small capacitances

(2.7 pF or 2.7 · 10−12 F), that may be problematic to measure, especially

in the presence of other parasitic capacitances Furthermore, standardmeasuring equipments usually have an input capacitance in the order of

25 pF [4], an order of magnitude larger than the capacitance expected to

3.1.2 Sample device

I made a very simple and large device consisting of a ≈15 µm membrane

of PDMS sandwiched between thick 1 mm-wide perpendicular lines of gold.First, bottom electrodes are sputtered on a glass slide using a stencilmask Then, a 15 µm membrane of 1:10 PDMS is spun on top of the glassslide: spinning at 2000 RPM for 5 minutes gives a repeatable thickness.Finally, top electrodes are sputtered, using a stencil mask again

This gives a theoretical base capacitance of around 1.7 pF for eachintersection I use 2 identical active area in differential mode: i.e theAD7745 chip measures the difference between the 2 capacitances, whichshould help to reduce electrical noise Finally, I use some sort of “table”,created using a 3D printer, to apply a somewhat controlled pressure on thedevice by adding metal rings around the center pillar (Figure 3.2)

Figure 3.2: 3D-printed “table” to apply a controlled pressure on thedevice: The total contact area is designed to be 4 mm2 Metal rings are

added around the central pillar to increase the weight

Linear fits

Figure 3.3: Capacitive readout in differential mode, adding weights on the

table (red), then removing them (blue)

Figure 3.3 shows results obtained with this sample device: I start withthe unloaded table, and slowly add weights, to increase the pressure (redline) Then, I unload the device by removing weights (blue line) The ex-pected approximately linear dependence between pressure and capacitancecan be observed However, the system is extremely sensitive to noise, andthe capacitance varies significantly: 2 measurements taken within a fewminutes interval give very different values, as seen between the loading andunloading curves

56

(a) Rest state.

(b) Shear stress parallel to the

the effective surface area of the capacitor changes

Measurement of shear forces One of the criticism of the capacitiveidea was that is it mostly sensitive to bulk forces, perpendicular to thesurface, while forces related to the cell motility, on the edges of cells, aremostly Shear forces, tangential to the substrate [55] This can be addressed

by using a design such as the one presented in Figure 3.4 [101], where theintersection between the top and bottom electrodes is changed due to Shearforces, as well as bulk forces

This approach, however, has the disadvantages of requiring precise

alignment of top and bottom electrodes of the device, which is technicallymore challenging.

Increase in sensitivity To make this approach more viable, the stiffness

of the membrane between the 2 sets of electrodes may be decreased bypatterning the PDMS between the electrodes (pyramids, pillars ), leading

to larger capacitance changes

Miniaturization Decreasing the sensor size to resolution required forour application (s = 1 or 2 µm in Table 3.2) leads to very small basecapacitance (< 100 aF), unless thickness can be reduced to a few hundrednanometers And even then, given our accuracy measurements around

60 aF, differences would be impossible to measure

This last issue is a blocker in our context, and is the main reason whythis approach was not pursued further

This approach uses a composite film of PDMS and conductive cles, to create a piezoresistive layer Piezoresistive materials change resis-tance when compressed, leading to a force-sensitive material Setting upelectrodes in a grid, as in Figure 3.1, would allow individual readout ofresistance

parti-One possible way is to add some conductive filler to PDMS, such aszinc [1], nickel [71], silver [85], multiwalled carbon nano-tubes (MWNT) [27],carbon black [79, 85], or even graphene flakes [65] (the latter approach usesepoxy instead of PDMS)

58

Figure 3.5: PDMS mixed with Carbon Black: the resulting composite

material is completely opaque

The particle size of such fillers need to be small, less than 10–100 nm,

as our objective is to obtain a sensing “pixel” size of 1 µm2: we need asignificant number of conductive fillers per sensing area Also, we need theresulting composite to be transparent to light

I tried using carbon black (Vulcan XC-72R, free sample provided byMaha Chemicals Singapore), whose particle size is specified to be around

40 to 100 nm [85] However, I experienced difficulties in preparing themixture at the required ratio (20 wt%): carbon black is very light, and alarge volume needs to be blended in the PDMS Furthermore, the resultingcomposite loses PDMS properties: it is very solid, therefore impossible tospin on a glass slide, and completely opaque

Further attempts with graphene flakes did not lead to better results

Even a low 3% wt/wt graphene flakes in PDMS concentration leads to avery solid non-conducting composite: one may need to increase the propor-tion of graphene to get a conductive material, but adding more graphenewill produce an even harder composite.

After these experiments, this approach appeared likely to be a end, unless a completely different kind of piezoresistive material is used,and was not pursued further

In collaboration with Henrik Andersen (Professor ¨Ozyilmaz group)

Another interesting option is to use intrinsic strain gauge properties ofgraphene: when subjected to strain, graphene changes resistance, up to a

10 times factor with 30% stretching [63]

A possible design is shown in Figure 3.6, the zig-zag patterns increasethe magnitude of the resistance changes when shear stress is applied in agiven direction [60] However this design has the disadvantage of requiringvias between the row and column electrodes, which presents significanttechnological challenges

This approach was pursed by collaborators Transferring graphene on

a soft PDMS substrate proved very challenging, with the resulting deviceoften breaking and rarely conducting electricity Finally, common transfertechniques involve oxygen plasma transfer of the PDMS surface, whichconsiderably hardens its surface I believe that this approach has potential,but graphene transfer technology needs to improve dramatically before thiscan be revisited

60

of charge produced (Q, in Coulombs) when a force F (in Newtons) is appliedperpendicularly to the surface [110]:

The grid design of Figure 3.1 is used again, and the electrodes on eachsides of the piezoelectric material also act as a parallel plate capacitor,with capacitance C as shown in Equation 3.1 The charges created by thepiezoelectric material are therefore charging the capacitor, which can then

be read out as a voltage:

d33· F

 r  0 ·A d

= d33· d

where P = F/A is the pressure applied

Similar relations can be established for Shear forces, using coefficients

d31 and d32

PVDF (Polyvinylidene Fluoride) is a potentially interesting polymerwith this piezoelectric property, and it is transparent, which is essentialfor my purpose The parameters of commercial films are given as follows(Piezotech [93]):

d33≈ 13 − 22 pC/N; (piezoelectric coefficient, normal forces)

d31,32 ≈ 6 − 10 pC/N; (piezoelectric coefficient, Shear forces)

Some experimental results are shown in Figure 3.9: When the device

is pressed, negative charges are created, leading to a voltage drop As theinput impedance is the circuit is finite, and due to some leakage in thepiezoelectric material, the capacitance slowly discharges [105], according

to an inverse exponential curve When the stress on the device is released,positive charges are released, leading to an inverted peak

Because of the discharge, piezoelectric materials are only good to detectchanges in the force applied It is however possible to integrate the signal(either using an analog circuit, or digitally) to obtain a static measurement

of the forces

The expected amplitude of the signal (1 mV for a kPa of pressure, seeEquation 3.4) should be possible to read using a charge amplifier, but I amfacing the following 2 issues

First, the capacitance of the active area is very small (see Table 3.3):even a square area of 100 µm sides only has a capacitance of 29 fF Thebest operational amplifiers have a stray input capacitance of at least 1 pF(1000 fF) [61] Therefore, any charge coming out of the sensor will first be

Top electrodeBottom electrode

Active area

Contacts

Figure 3.7: Schematics of sample shock sensor from Piezotech, 25 µmthick Chromium/Gold electrodes on both sides The overlap of the

electrodes (i.e., the active area), covers 9 mm2

Figure 3.8: Charge amplifier circuit, very commonly used to read out

strain information from piezoelectric sensor (e.g [108])

Figure 3.9: Sample output from the sample shock sensor, using a chargeamplifier (Figure 3.8: 100 MΩ resistor, 1 nF capacitor) connected to anArduino, with the data displayed using a simple Java application

64

“diluted” in a capacitor 35 times larger before being re-amplified.

Secondly, it is possible to amplify signals by large factors (e.g., EEGsignals are usually amplified 10000 times [91]), however, due to some con-siderations of the operational amplifiers (namely a small output offset), aswell as noise problems, one generally uses a high-pass filter to remove the

DC component of the signal This is not an option for me: the signal I need

to measure has a very low frequency content, as cells move very slowly, inthe order of a few tenth of a micron per minute (see Appendix A)

Because of these limitations, this approach does not seem so promising.However, Section 3.5 uses piezoelectric materials in a different fashion, byoscillating them at their resonant frequency

Finally, piezoelectric materials also show the reverse behavior: applying

an electric field creates a deformation of the material This could be used

to apply forces onto cells, and observe their behavior This is the startingidea of the work in Appendix C

Quartz crystals are commonly used as a frequency reference in electronicsystems A thin slice of quartz is cut at a precise angle (“AT-cut”), andelectrodes are deposited on both sides When subjected to an external elec-trical field, the piezoelectric quartz oscillates in a pure shear mode (i.e the

Figure 3.10: Frequency response (phase and amplitude) of a 4.433619 Mhzcrystal, as measured with a signal generator and lock-in amplifier.

electrodes only move parallely to the surface of the quartz crystal) [7] Thisoscillation has maximum amplitude at the resonant frequency:

crys-Quartz crystals are sensitive to external environment: they are affected

by temperature changes, and mechanical forces Circuit designers generallywant to avoid this, to get a stable frequency reference, which is why crystalsare protected in a sturdy metal case

In the context of quartz crystal micro-balances (QCM), these

pertur-66

bations are used to measure properties of samples [7] As indicated byits name, the main application is to weigh small masses, added onto thecrystal Furthermore, QCM can provides measurement of the sample vis-cosity [58].

When a sample, such as a thin film, is rigidly coupled with the QCM,the added film increases the effective thickness of the crystal, as the wholeQCM/sample system resonates This leads to a decrease in resonant fre-quency The frequency shift is governed by the Sauerbrey equation [58]:

This expression can be further simplified given that Zq = pρqGq =

For viscoelastic samples, the QCM response is more complex, as theassumption of a rigidly coupled sample does not hold anymore In thatcase, the deposited sample dampens the crystal oscillations, resulting in areduction in the Q factor (i.e the “sharpness” of the resonant frequencypeak) Measuring both the resonant frequency and the Q factor allows toderive some viscoelastic properties of the sample [58]

Figure 3.11: a) Original 4.433619 Mhz crystal in its casing b) After

opening, electrodes are connected using silver paste

A simple experiment was performed by opening a 4.433619 MHz crystal(Figure 3.11) One can easily measure a frequency shift when the quartz isloaded with various small objects: I was able to measure a 116 Hz variation

in the oscillating frequency under load, compared to a rest state (Table 3.4).Interpreting the relationship between object and frequency shift is non-trivial, as the object vary in geometry, and the assumptions of Equation 3.6(thin, rigidly-coupled film) do not hold

QCM have been used before to study cell biology ([35] is one example),but, to the best of my knowledge, is only used as a single-point analysis(cells are seeded on a large crystal, with a single set of electrodes) Beingable to build an array of QCM, and measuring both forces and viscositywould be a interesting new tool to study cell dynamics

This approach was not pursed further, as it deviated too much fromthe core idea behind this project, and I lacked appropriate measurementequipment at the time However, I believe it may have potential for furtherexploration

68

Unloaded Object 1 Object 2 Object 3 Object 4 Object 5

in each case, and no direct relationship between mass and frequency

response should be expected

3.6 Electrical interface

No matter which approach we choose, one of the challenges that wefaced is to design an interface between the device and measuring equip-ments The maximum area that can be written by e-beam lithography is

2 mm x 2 mm The common way to connect the device to the outside world

is to use wire-bonding, but this is known not to work on PDMS We alsowant a method that allows to connect many wires at a time, with minimaleffort

The solution we came up with is illustrated in Figure 3.12 First, the vice is patterned, gold electrodes are deposited using electron beam lithog-raphy Then, a 0.5 mm flexible ribbon is attached to the device, usingsilver paste or conductive epoxy to make sure contacts are mechanicallystable A custom PCB was built using Electrical Engineering facilities atNUS, that allows a 6-way ribbon to be connected on one side, and 6 BNCconnectors on the other side We can then connect BNC cables to a lock-inamplifier or any other measurement equipment

de-In the scope of this thesis, I never obtained a working multi-channeldevice, and this concept was only tested on dummy samples However, Ihope it can be put to use in the future

70

Chapter 4

Cell sensing device

In this chapter, I show that intrinsic graphene, without tion, can be used to detect cell metabolism in solution

functionaliza-In Section 4.1, I transfer a large piece of graphene onto a glass slip, and then connect 2 electrodes onto the graphene sheet Instead ofapplying the gate voltage from the back of the device, through a siliconoxide layer, as is commonly done [87], I use the fact that cell media is anelectrolyte to gate the device via a platinum electrode (see Section 1.1.2.2for details) This allows us to transfer graphene to transparent substrates(glass coverslips), making it possible to perform optical imaging while elec-trical readout is performed Also, the small voltages required for electrolytegating reduce the likelihood of perturbing cell behavior: Back-gating oftenrequires voltages in excess of 30 V, while this system never requires volt-ages larger than 1 V Finally, such low voltages can easily be applied andmeasured using cheap, off the shelf, microcontrollers and electronic parts

cover-I provide a complete analysis of the gate voltage-conductance curve toobserve that cell media (and in particular fetal bovine serum) perturbs

mostly the gate capacitance, while cells themselves cause a large shift ofthe Dirac peak towards more negative values, consistent with an addition

of positive charges on the surface of the graphene

Furthermore, I show that this behavior cannot be explained by a simplechange in pH, and that the device response is proportional to the number

of cells Furthermore, conditioning media by putting it in presence of cellsovernight, then removing this conditioned media and adding it to a deviceconfirms that I am detecting a cell by-product

Section 4.2 focuses on an improved device design, that is built to allowmeasurement of the number of cells in a dish I envision this as a new kind

of cell counter, complementing existing techniques such as hemocytometer,automated cell counters or impedance-based systems [11, 34, 59]

I acquire a 4 by 4 inches piece of monolayer graphene on copper phene Supermarket, Calverton, NY) Following the transfer method de-scribed in [97], one side of the copper foil is plasma etched to removegraphene (oxygen plasma, performed by Graphene Supermarket) I spincoat the graphene side of the copper foil with poly(methyl methacrylate)

(Gra-in anisole (PMMA, MicroChem 495K A4; 4000 RPM for 80 s), then bake

it for 2 minutes on a hotplate at 180◦C I then cut a small piece (10 by

5 mm), that I carefully place on the surface of a 0.7% ammonium persulfate(Sigma) solution in deionized (DI) water I leave the piece in this solutionovernight, in order to etch the copper layer I then scoop the graphene

74

supported by PMMA, using a glass coverslip, and keep it floating in DIwater for another night.

I then scoop the graphene piece using the destination substrate (a glasscoverslip), and leave it to dry on a hotplate at 70◦C for 30 minutes Ithen increase the temperature to 180◦C, for 30 minutes as well, to enhanceadhesion to the glass [106] Finally, I place the device in Acetone overnight,

to etch the PMMA layer, then rinse it with isopropanol, DI water, andblow-dry with nitrogen

Figure 4.1 shows the assembly process First, I fix 2 wires of electricalribbons (TE Connectivity FSN-23A-30) on each side of the graphene deviceusing conductive epoxy (ITW Chemtronics CW2400) The epoxy is cured

in an oven at 65◦C for 2 hours As a routine check, I verify the electricalresistance of the device using a multimeter: values between 5 and 10 kΩare typical

I then drill a 15 mm diameter hole in a 35 mm tissue culture dish(Dow Corning), and attach the graphene device to the bottom of the tissueculture dish using a silicon glue (WPI Kwik-Cast Sealant), making surethat the seal is waterproof, and that the conductive epoxy is not in directcontact with the liquid

Gating is performed with a platinum wire (Sigma 357367, diameter0.10 mm), dipping into the cell media It is attached to an electrical ribbon,and fixed to a 35 mm dish lid The lid is clamped with a metallic ring,weighing down the device to ensure minimal movement of the assembleddevice when setting it up in the microscope chamber A schematic of theassembled device is repeated in Figure 4.2

(a) Graphene device, transferred on a coverslip, electrodes attached with conductive

epoxy.

to the graphene device

(d) Gating wire A platinum wire is

attached to a 35mm dish lid A large

metal ring is used as a weight to keep it

the lid in place.

(e) Completed device, ready to be transferred to the microscope.

Figure 4.1: Device fabrication steps, after graphene transfer to a coverslip

76

+

+ -

-++

-++

platinum electrode dipping into buffer or cell media

4.1.1.1 Electrical read-out setup

I connect the 2 graphene electrodes and the gating wire to the surement setup, composed of an Arduino with few simple electronic parts

mea-on a breadboard (Figure 4.4) The resistance of the graphene device isconstantly measured, while the gate voltage is slowly swept back and forth

in an interval (usually, from −0.2 V to 0.8 V), allowing a good readout ofthe Dirac peak, and both sides of the curve At low gate voltages, verylittle current flows between the gate and the graphene (leakage current).However, as the voltage increases, the conductance of the electrolyte be-comes non-linear, and electrolysis may occur For this reason, I restrictthe voltage range to the minimum possible Furthermore, to reduce thecurrent flow as the gate voltage is modified (dynamic leakage), I sweep at

a low frequency of about 2 minutes per curve

Figure 4.3: Custom-built Processing software screenshot The softwareallows the operator to setup the experiment conditions, communicateswith the Arduino, setting up gate voltage, and reading back Analog toDigital Converter values Finally, it plots the information for immediateinspection, and saves the data to a text file for later analysis.

The Arduino is controlled from a PC using a custom-built Processingsoftware, shown in Figure 4.3: the virtual serial port interface, via USB, ofthe Arduino is used for communication between the PC and the Arduino.Gating is performed using a pulse-width modulated (PWM) output

of the Arduino, followed by an attenuating low-pass RC filter (see Gatecontrol part of Figure 4.4) The PWM quickly oscillates the output betweenminimum and maximum voltage (0 and 5 V on an Arduino), and allows

to configure the amount of time that the Arduino spends in the minimumand maximum state using a 8-bit register Filtering the AC component of

78

Plugging in numerical values (RG1 = RG2 = 10 kΩ, RGL = 30 kΩ and

CG = 10 µF), the gain is 0.4: since the maximum output voltage of theArduino is 5 V, this gives us a voltage range between 0 and 2 V The time

constant is 0.3 s, more than good enough to filter out PWM oscillation(≈ 1 kHz on the Arduino) Since the PWM resolution is 8-bit (256 values),

I can control the gate voltage value down to 8 mV precision

Furthermore, the gate voltage value is fed back to an analog to tal (ADC) pin of the Arduino, so that the actual gating voltage can bemeasured experimentally

digi-The graphene device resistance is measured using a simple voltage vider (see Resistance read-out part of Figure 4.4) The graphene sheet isconnected between 2 resistors of known values, one small resistor RL1 isgrounded (bias resistor), while RL2 is connected to the 5 V input voltage(VCC) The graphene resistance is obtained by a voltage divider equation:

di-RG = (RL1 + RL2) · VG

VCC − VG,

where VG is the measured voltage drop across the graphene sheet

The role of RL1 is to bias the graphene sheet to a slight positive voltage.This allows to gate the graphene to an effective negative voltage, despite thefact that the Arduino is unable to produce such negative voltage output.For example, assuming a voltage drop of 0.5 V on RL1, applying a gatevoltage of 0 V through the PWM circuit will lead to an effective gate voltage

of −0.5 V Since the range of the PWM circuit output is 0 V → 2 V, theeffective voltage range becomes −0.5 V → 1.5 V

I use the 1.1 V voltage reference of the Arduino for the ADC read-outs.With the 10-bits ADC, I get a maximum resolution of about 1 mV Foreach point, 160 ADC read-outs are performed The first half are ignored, tolet the system stabilize after a change in gate voltage The rest is summedtogether, which, assuming large enough white noise and stable signal [8],

80

Figure 4.5: Sample gate-voltage vs resistance curves for one experiment.

theoretically increases the resolution to about 16 bits (≈ 13 µV)

In terms of resistance, this means that, for a base graphene sheet tance of RG = 10 kΩ, and RL1 = 75 kΩ, RL2 = 390 kΩ, we have a voltagedrop of 105 mV, and a resistance resolution in the order of a few Ohm

resis-4.1.1.2 Data analysis

When constantly sweeping the gate voltage, I obtain a large set ofpoints, which is hard to interpret Figure 4.5 shows a sample of such data.Due to hysteresis [114], curves for increasing and decreasing gate voltagesare typically very different Therefore, as a first step, I split the dataset in

2 parts, according to sweep direction

Then, I analyze each gate voltage Vg vs resistance R curve separately

I fit 3 parameters for each sweep: Vmax, the gate voltage corresponding

to the maximum resistance (Dirac peak); µ0h+ /e − = Cg · µh+ /e −, Hole (h+)and electron (e−) mobility of the graphene times gate capacitance (seeSection 1.1.2 for details).

In this context, using electrolyte gating of graphene, I cannot separatethe contributions of gate capacitance and mobility Therefore, for the pur-pose of this discussion, I will call the slope µ0 “mobility”, and use “intrinsicmobility” when talking about the graphene material property µ

I first filter the data with a zero-lag moving average of length 5: i.e.,for each point, I average its value with the 2 values obtained on both sides(between -16 mV to 16 mV) This helps smoothing the curve to avoid spu-rious fitting I then invert the measured values to obtain the conductanceG(Vg) To obtain the Dirac peak position Vmax, I fit a cubic spline aroundthe conductance minimum, and extrapolate the peak position

For the mobility, I perform a linear least square fit of the conductanceslope for gate voltages smaller and larger than Vmax, around the region

of maximum absolute value of the slope, to obtain respectively hole andelectron mobilities Figure 4.6 shows a sample of such fitting

I also evaluate the gate leakage by comparing the measured gate voltage

as a function of the applied PWM value By comparing this value to abaseline experiment performed without any device connected to the setup,

I can evaluate the amount of leakage For successful experiments, theleakage is very low, at a maximum of a few µA

I repeat this fitting operation for each and every sweep, and therefore

am able to provide graphs showing variations of Dirac peak and mobilitiesover time After the analysis has run, I collate all plots together on onepage, as presented in Figure 4.7

82

and electron mobilities.

4.1.1.3 Experimental setup

For experiments with cells, I incubate the device with 20 µg/mL offibronectin (Sigma F1141) in phosphate buffered saline (PBS, Gibco), for 1hour at room temperature As graphene is hydrophobic [69], this makes itpossible for cells to adhere on the surface I then wash the device with PBS

3 times, and leave the device overnight in PBS before running experiments.The device could be reused across experiments, by washing with a 2%sodium dodecyl sulfate (SDS), followed by multiple washes in DI water,and finally PBS Care must be taken to not let the device dry up, assalt crystals easily break the graphene layer With careful handling, thedevice can be reused up to 6–7 times, and the integrity of the graphene is

Absmax leakage (uA)

Figure 4.7: Sample processed results for one experiment (same as

Figure 4.5) Columns are: Dirac peak position Vmax; hole mobility µ0h+;electron mobility µ0e− The first 2 rows correspond to sweeps with

increasing and decreasing mobilities Finally, leakage as a function of gatevoltage is also plotted, and maximum absolute leakage for a given sweep

84

evaluated based on its electrical characteristics A functional device shows

a resistance of less than 20 kΩ at the Dirac peak, while damaged deviceseither stop conducting, or show resistances in excess of 50 kΩ

For most experiments, I transfer DMEM media (Invitrogen) to the vice The media usually includes 10% fetal bovine serum (FBS, Invitro-gen) and 1% Antibiotic-Antimycotic (ABAM, Invitrogen), unless otherwisenoted I wait at least 1 hour, to make sure that the temperature is stable

de-I then start time-lapse optical imaging, using the phase contrast imagingmode of the microscope (Nikon Biostation) I then add a concentratedsolution of HeLa cells in DMEM media, freshly split with Trypsin/EDTA(0.25%, Invitrogen) and centrifuged at 100G for 2 minutes

For experiments requiring measurement of the cell surface area, I staincells for 45 minutes in 3 µM CMFDA (Invitrogen), before splitting andplating them Image acquisition is then done using a GFP filter, overmultiple locations Background is subtracted in post-processing, and thecovered area obtained by a simple threshold

Experiments on dead cells are performed by leaving the cells at roomtemperature overnight, then centrifuge and resuspend them as describedwith live cells

For experiments with conditioned media, I leave confluent cells night in a 35 mm dish, with 3 mL DMEM, in a incubator, and take outthe required amount of cell media

over-For pH calibrations experiments, I first perform the experiment in a

15 mL falcon tube and a pH meter (Sartorius PB-11), adding 0.1 M NaOH

or HCl to 5 mL of DMEM I then repeat the experiment on the device

In all cases, I transfer the device to the incubation chamber of the

microscope, which provides a controlled 37◦C temperature, 5% CO2, andhigh humidity vital to cells (Figure 4.8) Incidentally, this environmentcontrol also provides temperature and media pH stability required for theelectrical measurements

86

To confirm the quality of the graphene being used, I perform Ramanspectroscopy after transfer onto a glass coverslip This allows to probe notonly the quality of the graphene growth, but also possible damages duringthe transfer process.

Using a excitation wavelength of 532 nm, Figure 4.9 shows characteristic

G and 2D peaks at respectively 1580 cm−1 and 2680 cm−1 The high ratiobetween their amplitude (I2D/IG), as well as the sharpness of the 2D peak,confirms the presence of graphene [32] Figure 4.9a shows a I2D/IG ratio

of ≈ 2.7, indicative of single or 2-layer graphene [97] Another location onthe sample (Figure 4.9b) is probably composed of 2- or 3-layer graphene,

as the I2D/IG ratio is only ≈ 1.8 Such areas of multilayer graphene arecommon with CVD graphene [97]

Finally, the low magnitude of the D peak at 1340 cm−1, with a D peak

to G peak amplitude ratio smaller than 0.25 (average: 0.17), is typical forgood quality CVD graphene with a low number of defects [80]

88