Strain-Based Chemiresistive Polymer-Coated Graphene Vapor Sensors.

Suspended chemiresistive graphene sensors have been fabricated using well-established nanofabrication techniques to generate sensors that are highly sensitive to pyridine and with excellent discrimination between polar and nonpolar analytes. When coated with a polymer surface layer and suspended on 3-D patterned glass electrodes, a hybrid combination of polymer and graphene yields chemiresistive vapor sensors. Expansion and contraction of the polymer layer produces strain on the suspended graphene (Gr). Hence, when organic vapors permeate into the polymer layer, the high gauge factor of the graphene induces substantial electrical resistive changes as folds and creases are induced in the graphene. The hybrid suspended polymer/Gr sensor exhibits substantial responses to polar organic vapors, especially pyridine, while also exhibiting reversibility and increased discrimination between polar and nonpolar analytes compared to previous approaches. This sensor design also allows for potential tunability in the types of polymers used for the reactive surface layer, allowing for use in a variety of potential applications.

Introduction

Artificial olfactory electronic systems have attracted interest for air quality monitoring, medical diagnostics, explosives detection, and other applications.[1−3] Arrays of cross-reactive, chemically sensitive resistors provide an especially simple technological implementation of a vapor detection and classification device.[4] When exposed to volatile organic vapors, the analyte permeates and interacts with the sensing material, producing a change in the dc resistance of the film.[4,5] Different vapors can be classified and quantified using pattern recognition algorithms.[4−6] In medical applications, biomarkers for diseases found in exhaled human breath, such as increased levels of acetone, formaldehyde, and pyridine, are of particular interest for rapid, point-of-care diagnostics.[2,7] A variety of materials have been explored as chemiresistive vapor sensors, including intrinsically conducting or nonconducting polymers loaded with conducting material such as carbon black (CB), as well as capped metallic nanoparticles and other related systems.[4,6]

Two-dimensional nanomaterials such as graphene have shown potential for sensors as well as in various applications such as microelectronics and energy storage while also affording the opportunity to produce elastic and flexible architectures.[8−11] In a pristine monolayer, graphene is composed of sp2-hybridized carbon atoms that have unoccupied pz-orbitals oriented perpendicular to the basal plane. These pz-orbitals give rise to the impressive electronic sensitivity of graphene, which has led to the integration of graphene in pressure, piezoelectric, and chemical sensors. Hybrids of graphene with metals, polymers, and combinations of the two have also been widely explored.[12]

Suspension of graphene across a series of columns or trenches minimizes substrate effects on the resulting device and yields an approach to design sensitive graphene-based devices. Large chemiresistive responses have been observed upon exposure to polar analytes when monolayer graphene was used as a top contact on one-dimensional ZnO rods.[8] When suspended across a narrow trench, single-crystal graphene exfoliated from highly ordered pyrolytic graphite yielded devices with higher mobilities and sensitivities for dopamine than graphene monolayers on planar, flat substrates.[13] Suspension of the graphene monolayer typically produces effects on the nanoscale and requires time-intensive fabrication methods such as electron-beam lithography. Additionally, the resulting devices are delicate, and scale-up is challenging.

Strain responses have also been accessed by supporting or coating graphene with a polymer and then bending the subsequent sensor to induce strain in the graphene, detected as a change in resistivity. With a thick polymer overlayer, the graphene fully conforms to the polymer, so the graphene deforms as the shape of the polymer changes through mechanical or chemical deformation. Graphene pressure sensors are sensitive to small signals and are a well-developed area of study.[14] These strain-based graphene devices rely on changes in the resistance of graphene at the surface produced by compressive and tensile deformation of the lattice, which decreases and increases, respectively, electron localization. Devices are most commonly formed by encasing monolayer graphene in a stretchable polymer. This strain-based approach is feasible because the graphene lattice can undergo substantial deformation without breaking.[15] These polymer-coated monolayers are used less frequently as chemical sensors, in part because in these sensors, the graphene is typically integrated with a polymer that does not readily and reversibly respond to the presence of a vapor [i.e., polydimethylsiloxane (PDMS)].[16] However, these strain sensors typically have very high gauge factors, providing large measurable changes in resistance to small amounts of deformation, with some devices reported to have gauge factors of over 200.[17,18] Additionally, the presence of the polymer substantially improves the robustness of the suspended graphene due to the combined strength of the materials.

Vapor sensors involving monolayer graphene also often suffer from a lack of sensitivity because cleaned graphene is chemically inert and stiff and consequently offers limited tunability in its response and physical properties.[19] Typical transfer methods involve the use of a polymer overlayer, such as polymethylmethacrylate (PMMA) and PDMS, to support the graphene during etching of the supporting metal and transfer to the target substrate.[20] A number of polymer-free transfer methods have been developed,[21] but the polymer-supported transfer is widely used because it is relatively simple to implement, requires minimal skill during transfer, and can be scaled relatively easily on the lab scale to accommodate larger pieces of graphene. However, complete removal of the polymer overlayer is difficult after the polymer-supported transfer.[22] Extra cleaning steps are required to remove the polymer, including washes in a suitable organic solvent in which the supporting polymer is soluble. Frequently, a high-temperature anneal under a reducing atmosphere, such as forming gas, is also required. Typically, these methods do not fully remove the transfer polymer, leaving behind networks of hardened polymer on the surface of the graphene.[22,23]

When graphene that has been transferred using a polymer support is used in a sensor, multiple studies have demonstrated that the remaining polymeric residue gives rise to different responses than are observed from a pristine monolayer of graphene that does not have such a residue. For example, a pristine monolayer of graphene shows very low specificity towards most volatile organic compounds (VOCs).[19] An advantage of using graphene in a sensor derives from the ability to modify the graphene using physical or chemical methods, in conjunction with the inherent sensitivity to impurities and strain due to the underlying electronic structure of the graphene. Although chemical and physical modification of graphene through exfoliation from graphite, reaction with chemical precursors, or lithographic patterning can lead to increased sensitivity, these techniques rely on fundamentally changing the chemical structure of graphene through the attachment of new functional groups or introduction of defects to induce a response to a VOC of interest.[24]

In contrast, limitations associated with the polymer-supported transfer and any alteration of the monolayer of graphene can be minimized by retaining the polymer overlayer covering the graphene. The specificity that the polymer lends to the sensor and amplification of the sensor response by the strain dependence of the graphene can therefore both be exploited. Polymer composites have the advantage of better reproducibility and strength than either the polymer or graphene alone, offering remarkable transferable properties such as electrochemical reinforcement and stability.[9] These composites have been used in a variety of sensor device implementations but rely on integration with oxidized graphene flakes as opposed to pristine graphene.[9,25,26]

We describe herein a facile fabrication method that allows partially suspended monolayer graphene to be integrated with a vapor-responsive polymer, producing a sensor with high responsivity to pyridine as well as to a variety of other polar and nonpolar volatile organic analytes. A patterned sensor substrate was developed to support graphene as a partially suspended layer above the surface. By suspending the graphene, the material is allowed to expand and contract in response to the movement of the polymer overlayer. Additionally, the device configuration allows exploitation of the sensitivity of the physical and chemical properties of 2-D materials to the introduction of curvature, folds, and creases.[27] Numerous studies have reported the properties that result from folding and wrinkling of graphene, and the unique behavior due to the curvature of suspended graphene under mechanical stimulation.[8,17,28,29] The resulting chemiresistive sensors can consequently access both the sensitivity of the graphene monolayer and the specific response to organic vapors of different polymer overlayers.

Results and Discussion

Several sets of sensors were tested in this work, including 4% Poly(ethylene-co-vinyl acetate) (PEVA) mixed with CB (PEVA/CB) and 4% PEVA coated on graphene (PEVA/Gr) either on flat substrates (flat) or on substrates with columns (col) (see Methods for details on fabrication). The response from the sensors was quantified as the change in resistance (ΔRmax) with respect to the resistance of the baseline (Rb), where Rp is the peak resistance of the sensor

The sensitivity of the sensors (SR) was then quantified as the slope of the linear least-squares fit of ΔRmax/Rb (ΔR) versus P/Po. P̅ is the mean of the exposure partial pressures relative to the vapor pressure of the analyte (P/Po), is the mean of the ΔRmax/Rb values, p is the value of (P/Po) at the ith exposure, and ΔR is the ΔRmax/Rb value at the respective p value

The sensors developed in this work used a conductive monolayer of graphene coated with a layer of PEVA deposited onto a substrate patterned with columns to generate a signal in the presence of a VOC (Figure A). As the micrographs in Figures B,C show, the columns do not allow the graphene to fully adhere to the surface, leaving a gap between the bottom of the column and the monolayer of graphene. Less adherence is expected from 4% PEVA/Gr layers as this combined layer should be less flexible than graphene alone, increasing the gap between the columns and the 4% PEVA/Gr layer. Typical conductive polymer composite sensors contain CB as a conductive material. The percent composition of CB determines the baseline resistance and the optimal sensor response, so several control sensors were fabricated with CB and with monolayer graphene. These control sensors included a monolayer of graphene with no polymer on a flat substrate or columns, PEVA/CB transferred identically to graphene on a flat substrate or columns, and PEVA/CB sprayed onto a flat substrate (Figure ). Strikingly, the PEVA/CB composite sprayed onto the surface of a substrate showed a large negative response to ethanol and ethyl acetate, in accord with behavior that has been reported previously for such sensors,[30] whereas the PEVA/CB composites formed as uniform films showed a positive response to these analytes (Figure S2).[31] However, the PEVA/CB composites transferred identically to graphene in this work were composed of 4% PEVA as compared to 2% PEVA deposited by an airbrush because 4% PEVA was the lowest concentration that was sufficiently stable for the graphene transfer. Further optimization of the concentration of PEVA used to transfer the graphene used in these sensors was not pursued due to challenges with the transfer process.

Figure 1

Schematic of the sensor used in this work and scanning electron micrographs of the underlying monolayer of graphene on the sensor surface. (A) A monolayer of graphene is transferred to the sensor body by means of the supportive 4% PEVA overlayer. On exposure to a VOC, the 4% PEVA overlayer deforms the underlying monolayer of graphene, producing a detectable change in resistance. (B) Representative image of a single 150 nm tall column within the sensing region of a device. Some areas around the column are visible where the graphene had not completely adhered to the underlying surface. (C) Closer view of the same column. The monolayer of graphene is stretched out from the top of the column to the underlying surface.

Figure 2

SR values for the control sensors versus PEVA/Gr on columns [4% PEVA/Gr (col)] exposed to six different VOCs in a range of concentration (0.001 ≤ P/Po ≤ 0.005) at a flow rate of 3000 mL min–1 under N2 as the carrier gas. Substrates with columns were all 150 nm in height. Each value is the average of four vapor sensors per sensor type exposed to 4 different VOC concentrations, where the SR value was calculated using a linear least-squares fit to determine the slope of ΔRmax/Rb versus P/Po. For most of the analytes, the SR value was the highest for 4% PEVA/Gr (col). This geometry produced larger relative differential resistance changes than analogous PEVA/CB or bare graphene chemiresistive sensors under nominally similar test conditions.

PEVA/Gr (col) exhibited the largest sensitivity of all the sensors for most of the analytes, including pyridine (28–138 ppm), acetone (301–1506 ppm), ethyl acetate (124–618 ppm), and tetrahydrofuran (THF) (213–1066 ppm), except for toluene and ethanol. For both toluene and ethanol, PEVA/CB (col) showed the largest sensitivity of the sensors evaluated. Figure shows the linearity of responses of the sensor coated with Gr/PEVA to a single pulse of a range of concentrations of pyridine vapor (28–138 ppm). Upon exposure of the sensor to the analyte, the resistance steadily increased until the analyte was purged from the chamber, at which point the resistance decreased very slowly and flattened. The pulse peaks for the Gr/PEVA (col) were slower to respond and recover, whereas the other controls [bare Gr (col), 4% PEVA/CB (col), 4% PEVA/CB (flat)] exhibited much smaller but more rapid and reversible responses (Figure S3). The control sensors with 4% PEVA/CB on flat substrates or substrates with columns were more reversible than a bare monolayer of graphene on columns after the analyte exposure. This behavior suggests that the PEVA polymer overlayer with graphene not only makes the sensor less flexible and perhaps stiffer and slower in its recovery but also exhibits a higher resistance change when strained. In comparison, the remaining control sensors [bare graphene (flat), 4% PEVA/Gr (flat)] typically produced lower and noisier sensor responses, likely due to the strong adhesion of the graphene to the sensor substrate.

Figure 3

Typical sensor responses when exposed to 28, 55, 82, and 138 ppm of pyridine at a flow rate of 3000 mL min–1 under N2 as the carrier gas. The times at which the sensor was exposed to the analyte and purged with the carrier gas, respectively, are marked on the plot.

The column height was varied to ascertain the optimal response of the sensors, and changes to the column height were expected to change the degree to which the PEVA/Gr stack adhered to the underlying substrate. For most analytes, a 150 nm pillar height resulted in the highest sensitivity, except for pyridine and toluene, for which the sensitivity increased substantially as the column height increased, although the response of pyridine was an order of magnitude higher than that of toluene (Figure ). Pyridine and toluene are similar in chemical structure and size but differ in polarity, consistent with the observation that higher columns are required to obtain an optimal response from both of these VOCs. Aromatic compounds have noncovalent interactions available with graphene through π–π stacking, so the height dependence also suggests that these analytes benefit from larger areas of exposed graphene with which to interact.

Figure 4

(A) SR values for controls of column pillar height for suspended hybrid graphene sensors exposed to six different VOCs at 0.001 ≤ P/Po ≤0.005 at a flow rate of 3000 mL min–1 under N2 as the carrier gas. Pyridine characteristically exhibited an increase in response as column height increased. (B) An expanded ordinate of the responses except for pyridine. For the majority of VOCs, 150 nm column heights produced the largest response.

In addition to column height, the thickness of the polymer overlayer was varied to obtain the optimal response for the sensors to the VOCs evaluated in this work. Figure shows the responses for sensors with the polymer overlayer deposited at speeds between 1000 and 8000 rpm. 75 nm (6k rpm) thick films showed the largest sensitivity to most analytes except for toluene, with the responses decreasing substantially for sensors having thinner layers of PEVA. A PEVA layer of 320 nm in thickness exhibited comparable responses, but these layers were more difficult to fabricate uniformly due to the slower spin speed involved, leading to the wide range of responses reflected by the width of the shown error bars. To maintain ease of fabrication, high responsivity, and more flexibility in the polymer overlayer, the thinner 75 nm layer was selected for further investigation.

Figure 5

SR values of controls for polymer overlayer thickness exposed to 6 different VOCs at 0.001 ≤ P/Po ≤0.005 at a flow rate of 3000 mL min–1 under N2 as the carrier gas. The quoted spin speed in thousands (k) of rpm produced a polymer film of thickness: 1 k (320 nm), 2 k (220 nm), 3 k (160 nm), 4 k (130 nm), 5 k (80 nm), 6 k (∼75 nm), 7 k (∼73 nm), and 8 k (∼71 nm), respectively. A spin speed of 6 k (∼75 nm resulting film thickness) produced the best response for the majority of the exposures. SR values were larger for pyridine than for the other VOCs tested.

The sensors were similarly optimized for the number of columns on the substrate, as an increase in the number of columns was expected to increase the overall signal from a sensor (Figure ). The standard pattern had columns with a 3 μm diameter and a pitch of 7 μm. The pitch was then varied between 3 and 120 μm, with a constant thickness of polymer overlayer and size of the transferred Gr/PEVA sheet. As expected, the sensor response decreased as the number of columns decreased, likely due to fewer locations where the strain in the polymer overlayer could effectively translate to strain in the underlying graphene. Interestingly, the response of the sensors exhibited a plateau at ∼5 × 105 columns, indicating that further study of different patterns with reduced diameters and closer spacing could be valuable.

Figure 6

SR values of controls for columns/pitch of the substrate at various 6 different VOCs at 0.001 ≤ P/Po ≤0.005 at a flow rate of 3000 mL min–1 under N2 as the carrier gas. The number of columns correlated to the pitch as 4.3 million columns (3.5 μm), 2 mil (7.5 μm), 1 mil (15 μm), 500 k (30 μm), 250 k (60 μm), and 125 k (120 μm), respectively. SR values were largest at a 7 μm pitch for ethyl acetate, pyridine, and THF.

The reproducibility of the responses of the optimized sensor (PEVA/Gr on a substrate with 150 nm high columns with a 3 μm diameter and 7 μm pitch) was probed through repeated measurements of the response to the same concentration of the analyte. Figure A shows the response of an optimized sensor to repeated exposures of pyridine at 28 ppm. After repeated exposures, the response decreased over time to a plateau at 60% of the original signal (Figure B). However, after extended exposure to background N2(g), the sensor recovered the initially observed full response. The mean of the first 10 exposures prior to the extended rest was 0.99% with a 95% confidence interval (CI) of ±0.06%, whereas the mean of the first 10 exposures after an extended rest was 1.11% with a 95% CI of ±0.07%. One-way analysis of the variance yielded an F-value of 7.5417 and a p-value of 0.0133, indicating that the results before and after 100 consecutive exposures were statistically distinct. The higher mean after the first 100 exposures suggests that the performance of the sensor may improve over time.

Figure 7

(A) Optimized pyridine response at 28 ppm (nitrogen carrier gas at a flow rate of 3000 mL min–1) showing reproducibility in response to repeated exposure, with 200 s under N2 between exposures. (B) Comparison of initial 20 exposures (black) versus sensor response after being subjected to 100 pyridine exposures at 28 ppm (red) under N2 as the carrier gas at a flow rate of 3000 mL min–1 after being allowed to recover for 24 h under N2(g).

To better understand the ability of these sensors to distinguish between different analytes, the discrimination performance was analyzed using principal components analysis (PCA) (Figure ). The first, second, and third projections of the principal components (PCs) showed that the hybrid graphene polymer sensor array clearly separated polar from nonpolar vapors. Additional plots of this data illustrating the separation between VOCs can be found in the Supporting Information (Figure S8). Overlaps between data clusters were observed, especially for some groups of the polar aprotic vapors (group 1: THF, ethyl acetate, and acetone; group 2: isopropanol, ethanol), although these groups were mutually discriminated as were dimethylformamide (DMF) and pyridine. Moreover, pyridine generally exhibited the highest resistive response and produced a unique fingerprint relative to its aprotic polar counterparts. This behavior indicates that the nitrogen-based functional groups may have a specific effect on the PEVA/Gr sensors, likely due to interactions with the underlying graphene layer.

Figure 8

PCA of the response of an array of graphene polymer hybrid suspended on a patterned electrode. The responses showed a clear separation between polar and nonpolar analyte vapors. Although much less distinction was present between acetone, ethyl acetate, and THF, good separation was observed between aprotic and protic polar VOCs.

In comparison, previous PCA analysis of the response of an unmodified graphene sensor showed groupings between chemically similar compounds and separations between polar and nonpolar groups.[32] In this work, the PEVA/Gr sensors on substrates with columns showed equal, if not greater, separation between polar and nonpolar groups while also exhibiting unique fingerprints for compounds like pyridine, which was not evaluated on the unmodified graphene sensor. Pyridine is a known component of tobacco and cigarette smoke and can be used as a biomarker in disease detection.[33] The selective detection of pyridine suggests that this family of sensor designs could be used as part of a cross-reactive sensor array for environmental and occupational health sensing from human breath. Moreover, the optimized sensor in this work serves more generally as an example of a methodology to target a specific functionality or functional group with otherwise cross-reactive sensors.

The response of the optimized sensors described herein was a strong function of the number of columns (Figure ), with increases in the number of columns leading to a higher response until a plateau was reached at 5 × 105 columns. This plateau indicates that on this scale, the resistance change in the graphene is limited by the degree of strain that an analyte can impose on this configuration on the PEVA/Gr stack. The observed response also depended on the thickness of the polymer overlayer, scaling linearly with the square root of the film thickness (Figure ). The optimal thickness was ∼75 nm, whereas the optimal spacing was obtained using columns with a 3 μm diameter and 7 μm pitch. Although the signal degraded with time, the sensors exhibited highly reproducible responses, exhibiting complete recovery over repeated exposures (×100) with longer recovery periods. Further work on these sensors, such as changing the patterning on the underlying substrate or modifying the polymer overlayer could allow for further tuning of the performance and could potentially decrease the recovery time of the sensors.

Conclusions

Polymer-coated monolayer graphene can be integrated with a simple patterned electrode to produce a larger chemiresistive response to organic analyte vapors than a polymer-CB film or bare graphene. The use of hybrid materials can allow for programmed chemiresistive sensors with tunability in the various types of polymers that can be coupled with graphene. The response is controlled by the structure of the underlying substrate along with the thickness of the polymer overlayer. The sensor had a long recovery time in successive tests compared to the control sensors but recovered full functionality after extended exposure to a background gas. The sensitivity to pyridine increased as the column height increased, consistent with expectations for an enhanced ability of the sensor to expand and contract, resulting in large changes in the resistance. A hybrid graphene/polymer sensor array exhibited clear discrimination between polar, nonpolar, aprotic, and protic vapors with unique fingerprints for DMF and pyridine. Different polymer support layers on monolayer graphene within a larger array of cross-reactive sensors could allow for a similar sensor design to target other VOCs of interest.

Methods

Materials

CVD-grown monolayer graphene on Cu (Cu/Gr) was purchased from Advanced Chemical Supplier Materials (Medford, MA). Grains of graphene from this source were ∼50 μm in diameter, as reported by the manufacturer. PEVA (vinyl acetate, 18 wt %) was purchased from Sigma-Aldrich and used without further purification. Black Pearls 2000 CB was purchased from Cabot Corporation. All solvents were reagent grade from VWR and were used without further purification to generate the vapors tested herein.

Sensor Fabrication

The patterned sensor substrates were prepared in a Class 100 cleanroom. Glass slides were first cleaned with acetone and isopropanol before being baked at 170 °C to remove any residual solvent. Microposit S1813 photoresist (MicroChem) was spun onto the cleaned slide at 500 rpm for 30 s and then at 4000 rpm for another 60 s, followed by a 10 s exposure through the appropriate mask to a 425 nm lamp in a contact mask aligner (Suss MicroTech MA6/BA6). The pattern was developed in MF-319 developer (MicroChem) for 90 s. Columns of different heights were grown on the patterned slide by using an e-beam evaporator to deposit 50–300 nm of SiO2 (CHA Industries Mark 40). Lift-off was completed by sonicating the slides at 60 °C in Remover PG (MicroChem) for 45 min. Contacts were formed by sequentially evaporating 5 nm of Ti, followed by 45 nm of Au, onto masked glass slides. This process produced two metallic electrodes that were separated by a 0.3 cm gap.

Solutions of 4 wt % PEVA in toluene were sonicated for 2–4 h until the PEVA was well dispersed. To make the coated sensors, a strip of Cu covered by a monolayer of graphene (Cu/Gr) was coated with a supporting layer of PEVA at a specified rotation rate (1000–8000 rpm) for 60 s. The resulting stack (Cu/Gr/PEVA) was then cured for 1 min at 150 °C. Smaller pieces, ∼1 cm × 3 mm (active area ∼0.1–0.2 cm2), were cut and subsequently etched in an FeCl3 solution (Copper etch, Transene) until the Cu disappeared by visual inspection, generally requiring 1.5 h. This Cu-free piece (Gr/PEVA) was transferred for 1 h to a bath that contained ≥18.2 MΩ cm resistivity H2O before transfer to a second clean H2O bath, in which the sample was immersed for 12 h. After transfer to a final fresh H2O bath, the stack was pulled onto the appropriate sensor substrate and dried using a gentle stream of N2(g).

The sensors used as controls were fabricated using similar transfer techniques. Solutions of 4 wt % PEVA and 1 wt % CB were sonicated for 2–4 h until the CB was well-dispersed. The solution was then spun onto bare Cu and transferred as described above, or alternatively was applied to the sensors using an airbrush. Gr without a PEVA coating was transferred with a supporting layer of 495 K A4 polymethyl methacrylate (PMMA, MicroChem) spun at 3000 rpm for 60 s. After transfer, the PMMA was removed by soaking the sensor for 10 min in acetone.

Sensor Measurements

Sensors were tested using a custom setup that has been described previously.[4−6] Organic vapors were generated by sparging N2(g) at a flow rate of 3000 mL min–1 through 45 cm tall bubblers that had been filled with the appropriate solvents. The analyte concentration was controlled by adjusting the volumetric mixing ratio of the saturated analyte stream to the background N2(g) stream. The flow rates of the background and analyte gases, respectively, were regulated using mass flow controllers. Each run started with 700 s of background data collection. Each analyte exposure consisted of 200 s of a pure background gas, 80 s of the diluted analyte, and then 200 s of a background gas to purge the system, all at a flow rate of 3000 mL min–1. The sensors were loaded into a rectangular, 16-slot chamber that was connected by Teflon tubing to the gas delivery system. The resistance of each of the sensors in the array was measured by a Keysight Technologies 34970A data acquisition/switch unit with a Keysight 34903A 20 Channel Actuator. The measurement electronics were interfaced with a computer via a GPIB connection and were controlled with LabVIEW software.

The classification ability of the Gr/PEVA column sensors was visualized using PCA. PCA was performed on 16 individual hybrid graphene polymer arrays. For all volatile organic carbon (VOC) vapors, 20 test exposures were recorded at P/Po = 0.0050, where P is the partial pressure and Po is the vapor pressure of the analyte at room temperature.

Sensor Characterization

Profilometric data of the polymer overlayers were collected on a Bruker Dektak XT profilometer using a probe with a 2 μm tip radius. Atomic force microscopy (AFM) images of the sensors were collected using a Bruker Dimension Icon AFM. Raman spectra were collected with a Renishaw Raman microscope at a wavelength of 532 nm through an objective with a numerical aperture of 0.75. The laser power was ∼3 mW.

Signal Processing

All data processing was performed using custom routines in Origin and R. The relative differential resistance change, ΔRmax/Rb, was calculated from Rmax, the baseline-corrected response maximum upon exposure of the sensor to the test analyte and the baseline resistance under inert N2, Rb. A spline was fitted to the baseline data during the initial baseline preexposure period, and the values of ΔRmax/Rb were calculated by subtracting the values of the spline over the extrapolated exposure time from the observed resistance during the length of exposure. SR values were calculated using linear least-squares fitting of ΔRmax/Rb versus analyte concentration.

Sensor discriminant performance was visualized using PCA. The normalized data were mean-centered, and the diagonalized data set of the covariance matrix was transformed into sets of dimensions in terms of PCs. The largest amount of variance is captured in the first PC, while the second PC was orthogonal to the first PC and captured the second-most variance in the data. The normalized mean-centered data were projected onto the first and second PCs in accord with their respective coordinate vectors as observed through their corresponding eigenvalues and eigenvectors.