Tuning the Interfacial Chemistry of Nanoparticle Assemblies via Spin‐Coating: From Single Sensors to Monolithic Sensor Arrays

Sensor arrays based on gold nanoparticle (GNP) films are promising candidates for numerous applications, including medical diagnosis and health monitoring. Their economic fabrication, however, remains challenging. This study presents a facile route to GNP chemiresistors with tunable properties via layer‐by‐layer spin‐coating (LbL‐SC). Key steps involve the alternating deposition of dodecylamine‐stabilized GNPs and mixtures of monothiols (MTs) with 1,9‐nonanedithiol (9DT). The 9DT molecules serve to reinforce the growing film via GNP cross‐linking while the MT ligands are used to tune the interfacial chemistry of the GNP assembly. Hence, by employing differently functionalized MTs the sensors' chemical selectivity can easily be adjusted. Further, by varying the MT‐9DT ratio and adjusting the size of the MT ligands the sensitivity can be tuned along with the conductivity and optical properties of the films. In general, decreasing the 9DT fraction significantly enhances the sensitivity while the response isotherms change from Langmuir‐Henry to Henry type. Finally, the cross‐linked GNP films are robust enough to be patterned via photolithography. Hence, this study demonstrates the fabrication and application of monolithic sensor arrays. Different features of the responses to numerous analytes are used as input data for linear discriminant analyses (LDA), revealing that very similar analytes can be distinguished.


Introduction
Thin films of ligand-stabilized gold nanoparticles (GNPs) are well suited for applications as highly sensitive chemiresistors.Usually, these sensors respond with an increase in resistance when exposed to volatile organic compounds (VOCs). [1]This response is caused by sorption-induced swelling of the ligand matrix, leading to increased tunneling distances between neighboring GNP cores. [2]Besides their high sensitivity, these sensors benefit from very short response and recovery times, room temperature operation, low power consumption, and integrability.Another major advantage is their tunable chemical selectivity, which is achieved by the surface functionalization of the GNPs with a broad variety of commercially available ligands.It has been shown that arrays of selectivity-tuned GNP chemiresistors can enable the non-invasive diagnosis of various diseases via the detection of volatile compounds in breath and body fluids, [3,4] including the diagnosis of Alzheimer's and Parkinson's disease, [5] ovarian carcinoma, [6] lung cancer, [7][8][9][10] and COVID-19. [11]19] However, the efficient and eco-friendly fabrication of GNP chemiresistors with tunable chemical selectivity and their assembly as sensor arrays remains challenging.
Figure 1 shows the three major approaches that are currently used for the fabrication of GNP chemiresistors with tunable selectivity.][22][23] Hence, it is necessary to carefully select suitable solvents and to adjust the reaction conditions for each ligand individually.Further, the GNP batches are purified individually before depositing the GNPs on a suitable substrate.Hence, the fabrication of GNP chemiresistor arrays via Route A is laborious and requires extensive use of chemicals and solvents.
Route B provides a simplified approach. [16,24]Here, all chemiresistors are prepared from the same GNP batch.In order to tune the chemical selectivity, the GNP surface is modified by exchanging the initial ligands for differently functionalized ligands.To achieve highly efficient ligand exchange reactions, amines have been used as initial ligands that are easily exchanged for thiols, due to the formation of strong thiolate-gold bonds.Similar to Route A, however, it is necessary to adjust the conditions for the exchange reactions and to purify the GNP batches afterwards.Hence, Route B offers only few benefits compared to Route A.
5][26][27][28][29][30] Hence, the chemical selectivity of resulting chemiresistors is primarily controlled by the structure and chemical nature of the cross-linker molecules and added ligands.Since the cross-linked GNPs form fairly robust coatings, the films can easily be washed with arbitrary solvents.This robustness also enables the postpreparative adjustment of the selectivity via 2-phase linker-ligand exchange reactions. [31]n summary, while Routes A and B require laborious handling of colloidal solutions to provide GNPs with distinct ligand shells for chemiresistor fabrication, Route C can significantly simplify the process.Here, cross-linked GNP films are formed without prior surface modification.The obtained films can simply be washed with any solvent and re-functionalized, if desired.Furthermore, cross-linking improves the mechanical properties of the film and their durability.Additionally, cross-linking can reduce baseline drifts which have been attributed to electromigration of non-cross-linked GNPs. [32]On the other hand, crosslinked chemiresistors are usually less sensitive than their noncross-linked counterparts produced via Route A or B. The origin of this limitation has been attributed to the reduced ability of cross-linked GNP films to swell during analyte sorption. [1,33,34]owever, as shown in our present study, it is possible to solve this problem by reducing the degree of cross-linking to a level which still allows for the facile fabrication of chemiresistors via Route C but, at the same time, enables high sensitivities as reported previously for non-cross-linked GNP chemiresistors.
Currently, Routes A and B are prevalently used for the fabrication of GNP chemiresistors.Despite its potential advantages, Route C is being used less frequently.However, some variations of Route C have been reported.Zhong and coworkers reported the fabrication of GNP chemiresistors via a one-step exchangecross-linking-precipitation route. [7,17,25]In another study, Zamborini et al. fabricated chemiresistors using GNPs with mixed alkanethiol and -carboxylate alkanethiol ligand shells. [35]In a layer-by-layer self-assembly (LbL-SA) approach the GNPs were cross-linked by bridging the terminal carboxylate groups with Cu 2+ ions.In general, the LbL-SA process is based on dipping a surface-functionalized substrate alternately into solutions containing the GNPs and a cross-linking agent.Usually, the crosslinking agent has two or more functional groups with high affinity to the GNP surface.In order to avoid precipitation of the GNP solution by contamination with the cross-linking agent, it is necessary to wash the substrates extensively after each treatment with the cross-linking agent.[38][39][40][41] Following the LbL-SA approach of Bethell et al., [42] we fabricated chemiresistors by cross-linking amine-stabilized GNPs with various types of dithiols and dendrimers. [28,29,33,43,44][47] This method is compatible with industrial fabrication routines, including standard photolithographic patterning. [31,48]In another study, we demonstrated the fabrication of cross-linked and patterned GNP chemiresistors via layer-by-layer inkjet printing. [19]ere, we demonstrate the facile fabrication of highly sensitive GNP chemiresistors and monolithic chemiresistor arrays via Route C. First, we present a new protocol for the LbL-SC fabrication of GNP chemiresistors.Various ligand-linker mixtures are used to tune the interfacial chemistry and the level of crosslinking in resulting GNP films.Second, we show how the conductivity and the optical properties of the GNP films are controlled by the composition of used ligand-linker mixtures.Third, we demonstrate the application of the GNP films as chemiresistors.Our data reveal how their selectivity and sensitivity are controlled by the applied ligand-linker mixtures and how these properties correlate with respective electrical and optical properties.Further, it is shown how the shape of the response isotherms changes dramatically with variations of the films' interfacial composition and the level of cross-linking.Fourth, we present the fabrication of monolithic sensor array chips by combining the LbL-SC approach with conventional photolithography.Using resistive and kinetic features of the transient responses as input data for linear discriminant analyses (LDA), these sensor arrays can distinguish and classify very similar volatile organic compounds (VOCs), even when applying the analytes at very different concentrations.Hence, the methods and achievements presented in this study demonstrate a versatile and economic approach to the fabrication of GNP sensor arrays for various advanced sensing applications.

Fabrication of GNP Films and Individual Chemiresistors
Individual GNP chemiresistors were fabricated following our previously reported LbL-SC procedure, [47] see Figure 2. As detailed in the Experimental Section, mixtures of various monothiols (MTs) with the crosslinker 1,9-nonanedithiol (9DT) and dodecylamine-stabilized GNPs were alternatingly dropped onto a rotating glass substrate.The GNPs were synthesized using the method of Peng et al. [49] with minor adjustments and had a core diameter of ≈ 7.3 nm (± 5%).Details of the GNP characterization (UV/vis absorbance spectra, TEM images) are presented in the Supporting Information (Figure S1, Supporting Information).As reported previously, the initial amine ligands of the GNPs are efficiently exchanged by the stronger binding 9DT cross-linker and the MT ligands during the LbL-SC procedure. [50]9]25] Hence, to enable comparability with these earlier works, we deliberately used 9DT as the standard cross-linker.
For most MT-9DT mixtures, we observed that the LbL-SC fabrication of GNP films was feasible even after decreasing the 9DT molar fraction to only 5%.As reference experiments, we tested the LbL-SC film fabrication by applying pure alkanethiol solutions without the 9DT cross-linker.In this case, the LbL-SC film deposition failed.Hence, we infer that in the case of alkanethiol MTs, the stepwise LbL film growth requires a small amount of 9DT to enable covalent cross-linking of GNPs.In contrast, the LbL film fabrication was feasible when applying pure 3MPA or 8MOA solutions.In accordance with previous studies, [25] we attribute this finding to the formation of hydrogen bonds between the terminal carboxylic acid groups, which enable sufficient GNP cross-linking.However, in the case of the long 11MUA ligand, the LbL-SC film deposition required the use of a 2:1 molar ratio of the 11MUA-9DT mixture, revealing that for this ligand the formation of hydrogen bonds was insufficient to form a stable GNP network.Also, in the case of the long 14T alkanethiol, the LbL-SC film fabrication was feasible when applying the 2:1 molar ratio of the 14T-9DT mixture, but not with the higher MT molar fractions of 89 and 95%.Further, the LBL-SC fabrication of the 67%14T-9DT film was challenging.This material easily detached from the substrate and formed a rather inhomogeneous film.These findings suggest that the long alkyl chain of the 14T ligand sterically shields the much shorter 9DT cross-linker and, thereby, prevents sufficient GNP cross-linking.Hence, when mixing very long alkanethiols with 9DT, a high surface coverage of the GNPs with 9DT is required to enable the LbL-SC film fabrication.
In a first set of experiments, we prepared 32 GNP films (cf.Table 1) using 5-6 deposition cycles (cf. Figure 2, steps ii and iii).The films displayed a bluish to purple color, depending on the MT-9DT mixture used for film fabrication.Photographs, scanning electron microscopy (SEM) and atomic force microscopy (AFM) images of exemplary films are shown in Figure S2 (Supporting Information).The film thicknesses were determined by The glass substrate was first treated with the 9DT cross-linker or with an amino silane to provide an adhesion layer for GNP binding (step i).The GNP and MT-9DT solutions were dropped alternatingly onto the rotating substrate (steps ii and iii).After repeating steps ii and iii 4-5 times, the substrate was immersed in the MT-9DT solution overnight (step iv).The substrate was then washed with acetone (step v) and cut into quarters.Two electrodes were deposited onto the GNP film via PVD using a shadow mask (step vi).The electrode distance was ≈400 μm and the electrode width was ≈11 mm.b) Photograph of a chemiresistor and schematic showing the structure of the cross-linked GNP film.
AFM measurements and ranged from ≈34 to ≈45 nm.(see Table S1, Supporting Information).The fabrication of the chemiresistors was finished by depositing ≈100 nm thick gold electrodes onto the GNP films via physical vapor deposition (PVD), cf. Figure 2a,b.

Conductivity, UV-vis Absorbance of GNP Films, Grazing Incidence Small Angle X-Ray Scattering (GISAXS) Measurements
A widely used model for the conductivity  of GNP films considers thermally activated charge tunneling according to the following equation: [2,51]  =  0 e − e − E a RT (1)   Here,  0 is a preexponential factor,  the tunneling decay constant,  the edge-to-edge distance of neighboring GNPs, E a the activation energy, R the universal gas constant, and T is the absolute temperature.According to this model, the conductivity decreases exponentially with increasing interparticle distance.Further, it is well known that the spectral position of the localized surface plasmon resonance (LSPR) band of GNP films blueshifts with increasing interparticle distances, as described by the Maxwell-Garnett theory and confirmed in numer-ous experimental studies. [50,52]Hence, conductivity and UV-vis absorbance measurements are well suited to study how the interparticle distances of the GNP films are changed when varying the composition of the MT-9DT mixtures used for the LbL-SC process.
All GNP films prepared in this study showed linear currentvoltage (I-V) behavior within the voltage range of ±5 V. Figure 3 presents their room temperature conductivity as well as the spectral position of their LSPR band.In agreement with previous studies, [26,48,53] the conductivity of the 9DT cross-linked GNP film was ≈ 1 × 10 −2 S cm −1 .All GNP films prepared with the MT-9DT mixtures containing the short 6T and 8T ligands had conductivities that were comparable to that of the 9DT cross-linked film, suggesting that the interparticle distances along the percolating pathways of charge transport were similar.However, a slight blueshift of the LSPR band with increasing fraction of the MT ligands suggests that the average interparticle distances increased somewhat (see the UV-vis absorbance spectra shown in Figure S3, Supporting Information).Note, while the conductivity measurement probes the interparticle distances only along the percolating pathways of charge transport, the UV-vis data represent optical properties of the films sampled over the spot size of the spectrometer.
When using the 10T-9DT mixture with a 10T molar fraction of 89 and 95%, the LSPR blueshift increased (Figure S3, Supporting Information).Further, for the 10T molar fraction of 95%, a significant decrease in the film's conductivity was observed.These trends became even more obvious when the length of the added MT ligands (11T, 12T, 14T) exceeded the length of the 9DT crosslinker: When increasing the molar fraction of 11T from 67 to 89%, the conductivity decreased by nearly two orders of magnitude (from ≈10 −2 to ≈10 −4 S cm −1 ).At the same time, the LSPR band blueshifted from ≈615 to ≈590 nm (Figure S3, Supporting Information).Similarly, but even more pronounced, the conductivity of the films prepared with the 12T ligand progressively decreased by roughly three orders of magnitude (from ≈10 −2 to ≈10 −5 S cm −1 ) as the 12T molar fraction was increased from 67 to 95%.At the same time, the LSPR band shifted from ≈615 to 580 nm (Figure S3, Supporting Information).Finally, the conductivity of the film prepared with 14T-9DT mixture was approximately three orders of magnitude lower than that of the 9DT cross-linked film and the LSPR band was blue-shifted by ≈50 nm, even at the relatively low 14T molar fraction of 67%.(Figure S3, Supporting Information) Generally, the GNP films prepared with the mercaptocarboxylic acid-9DT mixtures showed similar trends.With increasing size of the MT ligands, the conductivity decreased while the LSPR band shifted to the blue (Figure S4, Supporting Information).Furthermore, the conductivities of the films prepared with the 6MHA-9DT and 8MOA-9DT mixtures decreased with increasing fraction of the MT ligand while the LSPR shifted to the blue (Figure S4, Supporting Information).In contrast, however, the conductivity of the films prepared with the short 3MPA ligands increased with increasing 3MPA fraction, while the LSPR band was redshifted compared to the 9DT crosslinked film (Figure S4, Supporting Information).The conductivities of the films prepared with the small 4MBA ligand followed a similar trend, while the shift of the LSPR band was not consistent (Figure S4, Supporting Information).We attribute this increase in conductivity to the increase in permittivity of the organic ligand matrix with increasing fraction of the polar 3MPA or 4MBA ligands.It is well known that an increase in permittivity decreases the activation energy E a for charge transport (cf.Equation 1). [2]In the case of the longer MTs (6MHA, 8MOA, 11MUA), our data suggest that this permittivity effect is outbalanced by the size of the ligands, resulting in increasing interparticle distances with increasing MT fraction.Hence, in this case, the conductivity declined with increasing MT fraction.
In summary, by adjusting the size and the type of MTs in combination with the 9DT cross-linker, the conductivity and the optical properties of resulting GNP films can be adjusted.The major trends displayed in Figure 3 suggest that in the case of alkanethiol-9DT mixtures, a sizeable increase in interparticle dis-tances is achieved when the length of the MT ligands (11T, 12T, 14T) exceeds the length of the 9DT cross-linker and when the MT ligand fraction is sufficiently high.In the case of the mercaptocarboxylic acid-9DT mixtures, an increase in interparticle distances is already achieved for MT ligands (6MHA, 8MOA) that are somewhat shorter or of similar length as the 9DT cross-linker.According to earlier works, the mercaptocarboxylic acid ligands cross-link the GNPs via hydrogen bonds. [25,54]Hence, the mercaptocarboxylic acid ligands of neighboring GNPs cannot interdigitate and, thus, demand more space than the interdigitating alkyl chains of the alkanethiol ligands.
In order to confirm the effect of the MT chain length and the MT-9DT mixing ratio on the interparticle distance variation, we conducted GISAXS measurements with selected samples.In qualitative agreement with our interpretation, these measurements revealed interparticle edge-to-edge distances of ≈1.2 nm, ≈0.9 nm, and ≈1.1 nm for the films prepared with the pure 9DT cross-linker, the 95% 6T-9DT and the 95% 3MPA-9DT mixtures, respectively.As expected, significantly longer edge-to-edge distances of ≈1.9 nm and ≈1.7 nm were determined for the films prepared with the 89% 12T-9DT and 95% 8MOA-9DT mixtures, respectively (Figure S5, Supporting Information).

Chemiresistor Screening Tests and Response Isotherms
According to Equation 1, the conductivity of GNP films depends exponentially on the interparticle distance  and the activation energy E a .Therefore, even minor perturbations of these parameters can be detected by measuring the relative change in resistance (ΔR∕R 0 ), where ΔR is the change in resistance and R 0 is the baseline resistance.Since analyte sorption causes swelling of the GNP film and, thus, an increase in the tunneling distance , the analyte is usually detected by measuring a resistance increase. [1,2]However, analyte sorption can also cause an increase in the permittivity of the GNPs' environment which leads to a decrease of the activation energy E a .If this effect overrides the effect of sorptioninduced swelling, the sensor response is observed as a decrease in resistance. [1,44]Generally, however, in three-dimensional GNP films, the effect of swelling predominates the permittivity effect and the overall sensor response is measured as an increase in resistance, i.e., a positive ΔR∕R 0 signal.
To screen the chemiresistive response characteristics, all 32 GNP films were exposed to vapors of toluene and 1-propanol.Toluene and 1-propanol represent nonpolar aprotic and polar protic VOCs, respectively.Because both solvents have a similar vapor pressure (≈27 and ≈37 mbar at room 25 °C, respectively), the differences of their partition coefficients for analyte sorption are predominantly controlled by their distinct chemical nature.Figure 4a shows the transient response amplitudes of the GNP films which were fabricated using the nonpolar n-alkanethiol-9DT mixtures.The amplitude was measured after exposing the films for 115 s to the analyte vapors at the concentration of 2000 ppm in nitrogen.As shown exemplarily in Figure 4b, the response transients had a nearly ideal rectangular shape with short response and recovery times.The transient responses of all films prepared with the n-alkanethiol-9DT mixtures are shown in Figure S6 (Supporting Information).
Compared to the film cross-linked with pure 9DT, the films prepared with the MTs 6T, 8T, and 10T showed similar response amplitudes.However, with increasing MT fraction a trend to slightly increasing response amplitudes was observed.As the length of the MT ligand exceeded the length of 9DT, i.e. for the films containing the MTs 11T, 12T, and 14T, this trend became much more obvious.Especially, the intensity of the responses to toluene increased significantly with increasing length and molar fraction of the MT ligands.We attribute this selective increase in sensitivity to toluene to, both, the nonpolar character of the longer MT's alkyl chains and a decreasing degree of covalent GNP cross-linking.Note, in comparison to the 89%12-9DT film, the optical appearance of the 95%12T-9DT film was somewhat inhomogeneous and we observed that this film occasionally detached from the substrate.Therefore, we attribute the somewhat lower sensitivity of this film to defects resulting from insufficient cross-linking.Qualitatively, however, the major trends in sensitivity enhancement to toluene correlate with the observed increase of the films' baseline resistance and the shift of the LSPR band discussed in the previous section (cf. Figure 3).Hence, the films' electrical and optical properties can enable the selection of chemiresistor coatings with high sensitivity.
It is well-known that the chemical selectivity of GNP-based chemiresistors can be adjusted by varying the polarity of the GNPs' ligand shell, or by using different cross-linkers for GNP film assembly. [1,6,19,26,28,55]][58][59] Here, we used carboxylic acid functionalized MTs and their mixtures with 9DT to systematically tune the selectivity of the GNP chemiresistors towards more polar analytes.Figure 5a shows the amplitudes of transient responses to toluene and 1-propanol vapors at the concentration of 2000 ppm in nitrogen.As seen by the exemplary response transients in Figure 5b, the GNP's surface modification with polar carboxylic acid groups slowed down the response and recovery kinetics.The response transients of all films prepared with the carboxylic acid MTs and MT-9T mixtures are presented in Figure S7 (Supporting Information).
The GNP films containing the MT-9DT mixtures with the short 3MPA ligand revealed decreasing response amplitudes to toluene and increasing response amplitudes to 1-propanol with increasing MT fraction.Interestingly, the GNP film fabricated with pure 3MPA revealed a significantly attenuated response to 1-propanol while a weak negative response was observed to toluene.As suggested by Snow et al., [58] we attribute the positive resistive response to 1-propanol to increased interparticle distances caused by the intercalation of analyte molecules into regions of polar carboxylic acid groups between neighboring GNPs.Toward toluene, the film behaves more like a rigid material.Hence, the negative resistive response is attributed to the increase in permittivity caused by the insertion of toluene molecules into free voids of the ligand matrix. [44]Similar trends were also observed for the GNP chemiresistors fabricated using the 6MHA-9DT and 4MBA-9DT mixtures (see Figure 5a).These data clearly show that the application of relatively short carboxylic acid functionalized MTs and their mixtures with 9DT enables the facile LBL-SC fabrication of chemiresistors with adjustable chemical selectivity.However, while the chemical selectivity could be tuned, the sensitivities to 1-propanol were comparable to that of the 9DT cross-linked GNP film.
In order to test if the sensitivity to 1-propanol can be enhanced by applying longer carboxylic acid functionalized MTs, we used 8MOA and 11MUA as well as their mixtures with 9DT for GNP film fabrication.As mentioned above, the LbL-SC deposition of films containing 11MUA was only possible when mixing this ligand with a relatively high amount of 9DT, i.e., when using the 67% 11MUA-9DT mixture.The response intensities of this GNP film were comparable to those of the 9DT cross-linked GNP film, but the relative response amplitudes to toluene and 1-propanol were reversed.However, with the 8MOA ligand the films could be assembled using different MT-9DT mixing ratios or even the pure 8MOA ligand.As shown in Figure 5a, a significant enhancement of the response to 1-propanol was observed for the highest molar fraction of 8MOA (95% 8MOA-9DT) and, especially, when using pure 8MOA for film fabrication.At the same time, the response amplitudes to toluene were attenuated at 8MOA fractions of 67% and 89%, but increased again with further increasing 8MOA fraction.We attribute this variation in sensitivity to the counteracting effects of increasing polarity and the decreasing degree of covalent cross-linking with in-creasing 8MOA fraction.The increasing polarity decreases partitioning of nonpolar toluene molecules.Hence, the sensitivity to toluene decreases at molar 8MOA fractions of 67% and 89%.However, as the degree of covalent cross-linking decreases, the ability of sorption-induced swelling increases.Therefore, the sensitivity to both toluene and 1-propanol increases when increasing the 8MOA fraction to 95% and 100%.Similar to the GNP films assembled with the alkanethiol-9DT mixtures, the sensitivity enhancement to 1-propanol with increasing 8MOA fraction correlates with the decrease of the films' baseline conductivity and the LSPR shift (cf. Figure 3).Note, to ensure a particularly robust comparison, the data presented in Figures 3, 4a, and 5a were acquired from the same film of each composition.To achieve this, the spin-coated substrates were diced to conduct the different measurements.Further, we studied the reproducibility of the chemiresistive responses.For this purpose, additional GNP films were prepared using selected MT-9DT mixtures.Figures S9 and  S10 (Supporting Information) present the response amplitudes of all GNP films studied.These data clearly confirm the major trends observed in Figures 4a and 5a.Furthermore, we investigated whether variations in the film thickness lead to changed sensor behavior.To this end, the 89%12T-9DT and 95%8MOA-9DT films were prepared by applying 2-6 LbL-SC cycles.Independent of the film thickness, the responses confirmed the major findings presented in Figures 4a and 5a (see Figure S11, Supporting Information).Additionally, the data suggest a somewhat decreasing sensitivity to 1-propanol with increasing film thickness.
Based on the above-discussed response characteristics, the GNP films prepared with pure 9DT, pure 8MOA, as well as the mixtures 89% 12T-9DT, 95% 3MPA-9DT, and 95% 8MOA-9DT, appeared to be promising candidates for the fabrication of chemiresistor arrays.Therefore, we studied their responses to toluene, 1-propanol, and water vapors over a broad range of concentrations (50 -10 000 ppm).Since the three analytes have similar vapor pressures, their sorption within the GNP films is mainly controlled by their chemical nature.Figure 6 presents the respective response isotherms.As reported previously, [18,60] the isotherms of the 9DT cross-linked GNP chemiresistors could be fitted using the Langmuir-Henry sorption model (see Figure 6a; for details see the Section S3, Supporting Information).Within the range of low concentrations (50 -500 ppm) the response amplitudes increased nearly linearly with increasing analyte concentrations (Figure 6f).Slope fits to these data revealed increasing sensitivities ranging from ≈ 1 × 10 −6 to ≈ 2 × 10 −5 ppm −1 , following the order water ≪ 1-propanol < toluene (Figure 7).This order is attributed to the hydrophobic nature of the 9DT's alkylene chain.Similar sensitivities have been reported previously for flexible 9DT cross-linked GNP chemiresistors fabricated on polymer substrates. [19,26]n striking contrast, the 89%12T-9DT chemiresistor showed significantly enhanced responses to toluene vapor and dramatically changed isotherms to the three analytes.As shown in Figure 6b the response isotherms were linear over a very broad range of analyte concentrations (50 -5000 ppm).Note, such linear response isotherms are ideal for sensors employed in sensor arrays because they simplify the analysis of signal patterns when the analytes are sampled at unknown concentrations.Only at concentrations above 5000 ppm, the isotherms for toluene and 1-propanol deviated somewhat from linearity with a slight positive curvature (upward bending).The slopes of the linear fits to the data presented in Figure 6g revealed qualitatively the same order of sensitivities (water ≪ 1-propanol ≪ toluene) as the 9DT cross-linked sensor (cf. Figure 7).However, the sensitivity to toluene was significantly enhanced and approached the range of 10 −4 ppm −1 .Previously, we observed such high sensitivity to nonpolar analytes, as well as response isotherms with a slight positive curvature, only for chemiresistors which were fabricated from non-cross-linked 12T-stabilized GNPs using a drop casting approach. [2]In that study, the positive curvature of the response isotherm was attributed to the exponential increase in resistance with increasing interparticle distances during analyte sorption.Accordingly, we attribute the pronounced sensitivity enhancement to toluene and the extended linear range of the response isotherms to the increased hydrophobic character of the GNP film (due to the high fraction of hydrophobic 12T ligands) as well as a low level of interparticle cross-linking due to the low fraction of the 9DT cross-linker and steric shielding of the GNPs by the longer 12T ligands.Note, compared to the 9DT cross-linked film, the enhanced sensitivity to toluene (and 1-propanol) becomes even more pronounced at higher vapor concentrations (>1000 ppm).Due to the saturation behavior of the 9DT cross-linked chemiresistor (cf. Figure 6a), its sensitivity (i.e. the slope of the response isotherms) decreases with increasing vapor concentrations, whereas the sensitivity of the film containing the 12T-9DT mixture remains constant up to ∼5000 ppm and increases slightly at even higher concentrations.
In order to test our hypothesis that the enhanced sensitivity of the 89%12T-9DT GNP film is due to a reduced degree of cross-linking enabling enhanced sorption-induced swelling, we performed comparative in situ GISAXS measurements.To this end, the GNP films prepared with pure 9DT and the 89% 12T-9DT mixture were subjected to GISAXS measurements before and during the exposure to saturated toluene vapor.As shown in Figure S8 (Supporting Information), a significant shift of the scattering curve to lower q-values upon exposure to toluene vapor was observed for the 89%12T-9DT film but not for the 9DT film.Hence, this finding proves that sorption-induced swelling of the 89%12T-9DT GNP film was much more pronounced.
Figure 6c shows the response isotherms of the GNP chemiresistor fabricated using the 95% 3MPA-9DT mixture.Similar as in Figure 7. Sensitivities of selected GNP films to toluene, 1-propanol, and water.The values were extracted as the slope of the response isotherms within the low concentration range (50 -500 ppm, cf. Figure 6).the case of the 9DT cross-linked reference sensor, the isotherms could be fitted using the Langmuir-Henry sorption model.This finding suggests that cross-linking by the added 9DT molecules and the formation of hydrogen bonds between the carboxylic acid groups of 3MPA ligands inhibited pronounced sorption-induced swelling.Hence, a similar saturation behavior was observed as in the case of the 9DT cross-linked reference film (cf. Figure 6a).Due to the presence of the polar carboxylic acid groups, however, the order of sensitivities to the three analytes was changed to toluene < water < 1-propanol.The sensitivities, which were extracted from slope fits to the data in the low concentration range (50 -500 ppm, cf. Figure 6h), are shown in Figure 7. Compared to the 9DT cross-linked film, the sensitivity to water was significantly enhanced whereas the sensitivity to toluene was significantly reduced.However, all sensitivities fall into the range of ≈ 1 × 10 −6 to ≈ 2 × 10 −5 ppm −1 and, hence, they are similar to those of previously reported cross-linked GNP chemiresistors. [19,26]igure 6d shows the response isotherm of the GNP film prepared with the 95% 8MOA-9DT mixture.In contrast to the 95%3MPA-9DT film (Figure 6c), the shape of these isotherms reveals no saturation behavior.Instead, the isotherms are almost linear up to vapor concentrations of 3000 ppm and a slight positive curvature is observed for higher analyte concentrations, similar as in the case of the 89%12T-9DT film, and as observed previously for non-cross-linked GNP chemiresistors. [2]This finding suggests that the degree of covalent cross-linking was lower than in the 95%3MPA-9DT film.We assume that the longer 8MOA ligand effectively shielded the GNPs and, hence, reduced the degree of covalent cross-linking by the 9DT cross-linker.However, the sensitivities extracted from slope functions fitted to the data in the low concentration range (50 -500 ppm, cf. Figure 6i) were comparable to those of the 95%3MPA-9DT film but the order changed to water < toluene < 1-propanol as shown in Figure 7.This change in selectivity is attributed to the longer hydrophobic alkylene chains of the 8MOA ligands.
As shown in Figure 6e, the shape of the response isotherms of the GNP film prepared with pure 8MOA ligands resembled the isotherms of the 95%8MOA-9DT film.The isotherms for toluene and water were nearly linear over the whole range of concentrations (50 -10 000 ppm).In contrast, the isotherm for 1-propanol was linear up to a concentration of ≈4000 ppm, above which a significant positive curvature was observed.Similar as for the 95%8MOA-9DT film, the sensitivities extracted from slope fits to the data in the low concentration range (50 -500 ppm, cf. Figure 6j) increased in the order water ≪ toluene < 1-propanol (cf. Figure 7).However, while the sensitivity to water was similar, the sensitivity to toluene and 1-propanol was significantly enhanced.We attribute this enhanced sensitivity to the fact that the GNPs were cross-linked only via hydrogen bonds and not covalently via 9DT, which was absent in the 100%8MOA film.
The results presented above clearly demonstrate that the LbL-SC method enables the facile fabrication of GNP chemiresistors with tunable response characteristics.Tuning the chemical selectivity of these sensors, their response kinetics, and significantly enhancing their sensitivity was achievable by selecting ligandlinker mixtures with properly adjusted mixing ratios.Furthermore, by reducing the fraction of the dithiol cross-linker, it is possible to not only enhance the sensitivity but to also tune the response isotherms to ideal linear characteristics over a broad range of analyte concentrations.As shown in the next section, combining the LbL-SC approach with conventional photolithographic methods enables the fabrication of monolithic sensor array chips.By choosing various adjustable features of the sensing elements, it is possible to discriminate and classify structurally very similar analytes.

Fabrication and Characterization of Monolithic Chemiresistor Arrays
For the fabrication of a monolithic chemiresistor array chip, we combined the LbL-SC method with a photolithographic patterning approach. [31]Essential steps of the process are illustrated in Figure 8a.A thermally oxidized silicon substrate with four pairs of interdigitated (ID) electrode structures was spin-coated with a thin layer of poly(methyl methacrylate) (PMMA).The areas of two ID electrode structures were selectively exposed to deep ultraviolet (DUV) radiation and the irradiated sections of the PMMA layer were removed by immersing the substrate in a mixture of 4-methylpentan-2-one (4M2P) and 2-propanol.The first GNP film was then deposited onto the patterned PMMA layer via LbL-SC.The GNP film on the PMMA-covered sections was removed by dissolving the PMMA layer with acetone.Afterwards, the whole substrate was again coated with a new PMMA layer and the photolithographic procedure was repeated, followed by the LbL-SC deposition of another GNP film onto the second pair of ID electrodes.By repeating the process three times, an array of four pairs of ID electrodes covered with different GNP films was obtained.For the LbL-SC fabrication of the four pairs of GNP chemiresistors, we used the pure 9DT crosslinker, the mixtures of 89% 12T-9DT and 95% 3MPA-9DT, and the pure 8MOA ligand.A photograph of a monolithic sensor array chip mounted on a printed circuit board (PCB) is shown in Figure 8b.
The two sensors of each sensor pair showed well reproducible responses to toluene, 1-propanol, and water (see Figure S12, Supporting Information).Furthermore, except for the 95%3MPA-9DT film, the sensors of the array showed similar selectivity patterns as the corresponding GNP films deposited onto individual substrates (cf. Figure 7).Again, the 89%12T-9DT and 100%8MOA films showed significantly higher sensitivities to toluene and 1-propanol than the 9DT and the 95%3MPA-9DT films, respectively.However, some deviations of the relative responses to the different analytes were noticeable (see the response transients presented in Figure S12, Supporting Information).We attribute these deviations to the different procedures used for the fabrication of sensors on individual substrates or of the sensor array on a single monolithic substrate.Since the latter required the repeated deposition and removal of the PMMA photoresist, repeated treatments with GNP solutions and ligandlinker mixtures, as well as treatments with different solvents, it is possible that this procedure altered the composition of the GNP films to some extent.Nevertheless, our findings clearly confirm that the selectivity and sensitivity of the array's sensors can be adjusted by applying mixtures of different ligand-linker compositions for their fabrication.
In order to test the capability of the sensor array to recognize and classify different analytes, the array chip was exposed to 17 volatile compounds with concentrations ranging from 50 to 1000 ppm. Figure 9a,b show the transient responses of the 100%8MOA-chemiresistor to ethyl acetate and n-heptane with a concentration of 500 ppm.As shown by these examples, the maximum response amplitudes as well as the response and recovery kinetics were different for different analytes.These kinetic features are well suited as input data for pattern analyses. [7,17,18,61,62]ere, the following transient features were extracted to perform a linear discriminant analysis (LDA) of the data set: i.The response amplitude (ΔR∕R 0 ) max measured at the end of analyte exposure (average value over the interval 110 -115 s).ii.The area A 20s below the transient curve limited to the first 20 s of analyte exposure.iii.The area A r underneath the recovery curve limited to the first ≈240 s of the recovery phase.For each analyte, the kinetic features A 20s and A r were first normalized by dividing them by the corresponding (ΔR∕R 0 ) max value to eliminate the variation of feature values due to different analyte concentrations.Instead of directly using the (ΔR∕R 0 ) max values as the maximum response feature, we took the corresponding square root values to avoid extremely high feature values for analytes with relatively low vapor pressure (i.e., high partition coefficients).Finally, for each analyte at each concentration, all three features were normalized by dividing each value by the corresponding highest feature value of the eight sensors.
Figure 9c shows the LDA plot for the 17 volatile compounds tested.Each circle corresponds to an exposure of the sensor array to the analyte indicated by the color code.Note, for each analyte different concentrations ranging from 50 -1000 ppm were applied, as indicated by different shades of color.The response transients, which were used to extract the features for LDA, are presented in Section S5 (Supporting Information).Although very different analyte concentrations were applied, the data points corresponding to different analytes group in separate clusters.Roughly, the clusters of polar analytes fall in the upper right area of the LDA plot, and can be assigned to water, polar protic, and polar aprotic solvents.Although the sensor array comprised only four different types of GNP films, the clusters of some very similar analytes, such as 1-butanol, 1-propanol, and 2-propanol, are well-separated.However, some overlapping clusters (e.g., 4M2P and 1-butanol, or 1-propanol and ethanol) are also observed.The clusters of the alkanes are clearly separated from the class of polar analytes and are found on the left-hand side of the LDA plot.Here, some similar analytes, (e.g., cycloctane and n-octane) are also well separated, whereas the clusters of other analytes are closely spaced or overlapping (e.g.cyclopentane and n-hexane, or methylcyclohexane and n-octane).Furthermore, the cluster of toluene is well separated from the classes of alkanes and polar analytes.
The performance of the LDA was evaluated by excluding analyte exposures with concentrations of 400 ppm from the training data set.By applying the linear transformation of the LDA model, the data sets of these concentrations were projected onto the LDA plot.These projections are represented by hollow circles in Figure 9c.It is clearly seen that the test data of the different analytes were correctly projected to their respective clusters.Further, the prediction method of the algorithm returned the correct assignment for all test data (colored crosses).
In another set of experiments, we expanded the diversity of sensors of the array.To this end, we prepared two monolithic sensor chips, each equipped with 4 different GNP films (Figure S30, Supporting Information).Four GNP films were prepared using the same MT-9DT mixtures as described above.The remaining four films were fabricated using different 95% MT-9DT mixtures.Based on their previously reported ability to tune the selectivity of chemical sensors, [31,[63][64][65] we selected the following ligands: 6-mercaptopyridine-3-carboxylic acid (6MNA), 4nitrothiophenol (4NTP), and 8-mercaptooctanol (8MOO).One sensor chip featured the 9DT, 95%6MNA-9DT, 95%8MOA-9DT, and 100%8MOA films.The other chip featured the 95%3MPA-9DT, 95%4NTP-9DT, 89%12T-9DT, and 95%8MOO-9DT films.The performance of the 8-sensor array was tested by exposing it to 19 different volatile compounds with concentrations ranging from 100 to 1000 ppm (100, 200, 300, 400, 500, 1000 ppm).The same resistive and kinetic features as described above were extracted from the response transients and used as input data for LDA.All response transients used for feature extraction are presented in Section S5 (Supporting Information).The LDA plot generated with all 19 analytes tested is presented in Figure 10.As expected, the clusters are even better resolved than in the case of the 4-sensor array.Structurally very similar analytes are wellseparated and only very few clusters show some overlap.Similar to the 4-sensor array, the polar analytes tend to cluster on the positive side of discriminant 1, whereas the nonpolar analytes tend to cluster on the negative side.The classification performance was tested by using the sensor responses to analyte concentrations of 600 ppm, which were excluded from the training data set.As indicated by hollow circles, the test data of all analytes were projected onto their respective cluster and assigned correctly by the prediction algorithm (colored crosses).Hence, these results clearly demonstrate that the GNP-based array chips developed in this study are well suited for the detection and classification of various volatile analytes.

Summary and Conclusions
We presented a facile process for the fabrication of chemiresistors with tunable selectivity and high sensitivity.The process is based on the LbL-SC deposition of cross-linked GNP films combined with in situ surface modification of the GNPs.The use of amine-stabilized GNPs enables rapid ligand exchange reactions with thiols during the LbL-SC film fabrication.Hence, when applying mixtures of different MT ligands (alkanethiols or mercaptocarboxlic acids) with 9DT, the MT ligands tune the chemical selectivity of resulting sensors.At the same time, the 9DT molecules cross-link the GNPs and enable the economic and well-controlled LbL-SC deposition of homogeneous GNP films.
Importantly, the size and chemical nature of the MT ligands as well as the MT-9DT mixing ratio have strong impact on the fundamental properties of the GNP films (optical properties, conductivity) as well as the selectivity and sensitivity of resulting chemiresistors.In general, with increasing length of the added MT ligands, the LSPR band blue-shifts and the conductivity of the films decreases significantly.For the films made with alkanethiol-9DT mixtures, these changes are accompanied by an increase in sensitivity to nonpolar analytes.These effects are especially pronounced when the length of the alkanethiols exceeds the length of the 9DT cross-linker and when the molar fraction of the 9DT cross-linker is reduced to 11% or 5%.As shown by GISAXS measurements, the decrease in conductivity and the blue-shift of the LSPR band are due to increasing interparticle distances.Furthermore, the increasing sensitivity to nonpolar analytes is due to a decreasing level of GNP cross-linking, enabling more pronounced film swelling during analyte sorption.In agreement with this interpretation, the response isotherms change from Langmuir-Henry to Henry type.In fact, sensors prepared with the longer alkanethiols and low 9DT fractions behave similar to previously studied chemiresistors based on noncrosslinked alkanethiol-stabilized GNP films. [2]However, the addition of small amounts of 9DT cross-linker is necessary to enable the facile LbL-SC fabrication of the films and subsequent fabrication of monolithic sensor arrays.
GNP films prepared with different mercaptocarboxylic acids-9DT mixtures show similar variations of conductivity and optical properties with increasing length of the MT ligand and decreasing 9DT fraction.However, in contrast to the films fabricated with alkanethiol-9DT mixtures, the LbL-SC film fabrication is feasible even without adding the 9DT cross-linker.In agreement with previous studies, [25,54] this finding was attributed to the formation of hydrogen bonds between the terminal carboxylic acid groups of neighboring GNPs.Furthermore, due to the high polarity of the carboxylic acid groups, the selectivity for polar protic analytes can be enhanced by simply increasing the MT fraction.Again, with decreasing 9DT fraction the shape of the response isotherms changes from Langmuir-Henry to Henry type.
Finally, by combining the LbL-SC method with sequential DUV-lithographic patterning of the GNP films, we demonstrated the fabrication of monolithic sensor arrays with 8 GNP chemiresistors.In a first example, the array consisted of 4 duplicates of sensors.Implementing our previous findings, the chemical selectivity of the sensing elements was adjusted by simply varying the composition of the MT-9DT mixtures used for the LbL-SC process.Different features of the transient responses to numerous analytes were used as input data for a LDA, revealing that even very similar analytes (e.g., 1-propanol, 2-propanol, 1butanol) could be distinguished and assigned correctly.Additionally, a sensor array consisting of 8 different sensing elements showed some further improvement of cluster separation in the LDA plot, which was generated using 19 different analyte vapors.Although the analyte concentrations used to produce the LDA training data varied between 100 and 1000 ppm, all analytes were assigned correctly using the LDA prediction algorithm.These findings clearly show, that the methods presented in this study are well suited for the facile fabrication of GNP chemiresistor arrays, which have shown great potential for numerous applications, such as medical diagnosis, personal health monitoring, food quality control, and hazardous materials detection.
Fabrication of GNP Films and Chemiresistors: Glass slides (2.2 × 2.2 × 0.02 cm, Carl Roth) were pre-cleaned in acetone for 15 minutes using an ultrasonic bath, and treated for 2 min in air-oxygen plasma (Harrick Plasma Cleaner PDC-002) directly before surface silanization and the deposition of GNP films.After the plasma process, the fabrication of GNP films prepared with 89% 12T-9DT, 95% 12T-9DT, 95% 8MOA-9DT mixtures, and 100% 8MOA required the surface functionalization of the substrates with amino groups to provide sufficient adhesion of the films.To this end, the glass substrates were immersed in a solution of 50 μL 3aminopropyldimethylethoxysilane solution in 5 mL toluene at 60 °C for 30 min. [33]Afterwards, the substrates were washed with toluene and acetone.9DT was dissolved in methanol to provide a stock solution with a concentration of 7.4 mM.The monothiols (MTs) were also dissolved in methanol to provide stock solutions with a concentration of 14.8 mM.The mixed ligand-linker solutions were prepared by mixing MT-and 9DTstock solutions at three volume ratios: 50/50, 80/20, and 90/10 (MT/9DT v/v), corresponding to molar fractions of MTs in resulting mixtures of 67%, 89%, and 95%, respectively.Note, mixtures with high molar MT fractions were chosen because we were especially interested to study GNP films with low levels of covalent GNP cross-linking.Using the GNP stock solutions and the MT-9DT mixtures, or alternatively, the 9DT stock solution (with the exception of 3MPA and 8MOA ligands, for which pure MT stock solutions were also used), the GNP films were produced by the LbL-SC protocol reported earlier. [47]The glass substrates were continuously rotated at 3000 rpm using a Karl Süss Labspin 6TT spin-coater.The substrates without silanization were treated two times with 100 μL of 9DT stock solution.Afterwards, 10 μL of the GNP colloid was deposited onto the substrate, followed by 2 × 10 μL of the MT-9DT solution.This sequential deposition of GNPs and MT-9DT solution represents one deposition cycle.GNP films on glass substrates were prepared using the GNP stock solutions GNP1 and GNP2.In total 5-6 deposition cycles were applied, and the as-prepared GNP films were immersed in their corresponding organic (pure 9DT or MT-9DT mixtures or pure mercaptocarboxylic acids) solutions overnight, resulting in film thicknesses of 34-45 nm, as determined by atomic force microscopy (AFM; DI Multimode, Nanoscope IV Controller; Table S1, Supporting Information).The absorbance of the films at the LSPR maximum was at least 0.8.UV-vis spectra of GNP films are presented in Figures S3 and S4 (Supporting Information).The GNPcoated glass substrates were cut into quarters.The fabrication of chemiresistor sensors was finished by depositing two ≈100 nm thick gold electrodes onto the GNP films via physical vapor deposition (Pfeiffer Classic 250).A gap of ≈0.4 mm between the two electrodes was kept using a cannula as shadow mask.To integrate multiple sensors into an array, a previously published lithographic patterning process was used. [31]xidized silicon wafers (725 μm thickness, 500 nm SiO 2 layer, Si-Mat-Silicon Materials) were used as substrate.The 4 pairs of interdigitated (ID) electrodes (5 pairs of fingers with 50 μm width, 50 μm spacing and 800 μm overlap) were lithographically fabricated on the substrate.The silicon substrate with the ID electrode structures was first spin-coated with a thin layer of PMMA by applying a solution of ≈33 mg mL -1 PMMA in chlorobenzene at 4000 rpm.Then, the substrate was soft-backed at 60 °C for 10 min.Afterwards, the PMMA layer was exposed for 140 min to deep ultraviolet (DUV) radiation through a shadow mask, and developed in 2-propanol/4M2P (3/1 v/v) solution for 70 s.The substrate was rinsed with 2-propanol yielding a PMMA mask with a pair of rectangular openings above the ID electrode structures.To strengthen the adhesion between the GNP films and the wafer, the opened rectangular sections of the substrate were functionalized with amino groups by subjecting the whole substrate to the silanization process, as described above for the glass substrates.The GNP films were then deposited onto the substrate with the PMMA mask using our LbL-SC protocol described above. [47]Afterwards, a lift-off step was performed by dissolving the PMMA mask in acetone.Hence, only rectangular sections of the GNP films remained on the substrate above the ID electrode structures.The whole lithographic process was then repeated three times to cover all eight ID electrode structures of the substrate pairwise with rectangular sections of GNP films with different MT-9DT compositions.The sensor array consisting of eight different GNP films was fabricated by splitting and combining two monolithic sensor chips with originally 4 pairs of different sensors.All monolithic sensor chips were fabricated using the GNPs of batch GNP3.
Characterization of GNP Films: The optical properties of the GNP films on glass substrates were analyzed by UV-vis absorption spectroscopy (Cary50 spectrophotometer).The nanostructure of GNP films was characterized by high-resolution scanning electron microscopy (SEM, Zeiss LEO Gemini 1550).For this purpose, the GNP films were deposited onto silicon wafers instead of glass slides, applying three to four LbL-SC deposition cycles.Charge transport properties of GNP films were characterized using a parameter analyzer (Agilent 4156C).The conductivity of the films was determined by taking into account the thickness of the films and the geometry of the electrode structures. [47]Current-voltage (I-V) curves were recorded in the range between -5 to +5 V. Further, to assess differences of the center-to-center nearest neighbor distance (NND) between GNPs in different samples and under different conditions, grazing-incidence smallangle X-ray scattering (GISAXS) measurements were conducted using our in-house SAXS setup with an Incoatec X-ray source I μ S and Quazar Montel optics.The X-ray wavelength was 0.154 nm and the focal spot size was 0.6 mm 2 .The samples were positioned at a distance of 1.2 m from a Rayonix SX165 CCD-detector.As the samples for SEM characterization, the samples for GISAXS measurement were prepared on silicon substrates.The volume scattering intensity profiles were extracted along radial cuts of the scattering patterns, focusing solely on the scattering resulting from the interaction between the incident beam and the particle layers.Other reflections generated by the specular scattering were not taken into account.Assigning the extracted volume scattering signal to a 3D face centered cubic (FCC) lattice model, the average NND of the GNPs was calculated using the software Scatter. [67,68]haracterization of GNP Chemiresistors and Sensor Arrays: Analyte vapors with varying concentrations (50-10 000 ppm) were generated using a commercial programmable gas calibration system (Kalibriersystem Modell CGM 2000, Umwelttechnik MCZ).All vapor-dosing measurements were programmed as a sequence of 2 min analyte exposure followed by 4 min purging with carrier gas.All experiments were carried out at room temperature.Nitrogen 5.0 was used as carrier gas.During the measurement, a constant flow (500 mL min -1 ) of analyte vapor or pure carrier gas was passed through the test-cell.The GNP chemiresistors, which were fabricated on glass substrates, were placed inside the test cell (aluminum, ≈20 mL) and connected to the parameter analyzer.The parameter analyzer supplied a constant bias of 5 V to the GNP films and the resistive responses were determined by measuring the change in current.The sensor signal was defined as the relative change in resistance ΔR∕R 0 , where ΔR is (R − R 0 ) and R 0 is the baseline resistance measured before starting the exposure to the analyte vapor.The maximum resistive response (ΔR∕R 0 ) max was measured after 115 s exposure to the analyte vapor.The transient responses of the chemiresistor arrays were measured using a different setup.The array chip was placed into a test-cell (aluminum, ≈10 mL), and each sensor was connected to a commercial surface mount device shunt resistor in series.The resistances of the shunt resistors was ≈1 to 10% of the baseline resistance of the corresponding sensor.A Keithley 2601A sourcemeter supplied a constant 5 V bias to all sensor/shunt pairs, and the voltage change at each shunt resistor was recorded using a custom-built multiplexer connected to a Keithley 2002 multimeter.The performance of the sensor arrays was studied by analyzing the resistive and kinetic features of the sensors' transient responses similar as described in our previous study. [18]The linear discriminant analysis (LDA) was performed using the LinearDiscriminantAnalysis algorithm provided by the sklearn (version 1.1.2)python (version 3.8.5)library.Details on the feature extraction are provided in the Results and Discussion Section (cf. Figure 9a,b).

Figure 1 .
Figure 1.Approaches to the deposition of GNP films for applications as resistive transducing elements in chemical sensors.Routes A and B require the solution-phase preparation of GNPs with different ligand shells to adjust the chemical selectivity of the sensor before depositing the particles onto a suitable substrate.In Route C, the surface modification of the GNPs and the film preparation are performed simultaneously by cross-linking the GNPs with different ligand-linker mixtures.

Figure 2 .
Figure 2. a) LbL-SC fabrication of GNP chemiresistors.The glass substrate was first treated with the 9DT cross-linker or with an amino silane to provide an adhesion layer for GNP binding (step i).The GNP and MT-9DT solutions were dropped alternatingly onto the rotating substrate (steps ii and iii).After repeating steps ii and iii 4-5 times, the substrate was immersed in the MT-9DT solution overnight (step iv).The substrate was then washed with acetone (step v) and cut into quarters.Two electrodes were deposited onto the GNP film via PVD using a shadow mask (step vi).The electrode distance was ≈400 μm and the electrode width was ≈11 mm.b) Photograph of a chemiresistor and schematic showing the structure of the cross-linked GNP film.

Figure 3 .
Figure 3. Room temperature conductivities (grey dots) and spectral position of the LSPR band (red dots) of GNP films assembled with different MT-9DT mixtures.

Figure 4 .
Figure 4. a) Resistive response amplitudes of GNP films prepared with pure 9DT and n-alkanethiol-9DT mixtures.Toluene vapor (black bars) and 1propanol vapor (green bars) were applied at the concentration 2000 ppm in nitrogen.b) Exemplary transient responses of the GNP films prepared with pure 9DT (dashed lines) and the 89% 12T-9DT (solid lines) mixture to toluene (left) and 1-propanol (right) vapor (2000 ppm in nitrogen).The films were exposed to the analyte vapors for 120 s.Start and end of the exposure are indicated as "on" and "off".

Figure 5 .
Figure 5. a) Resistive response amplitudes of GNP films prepared with pure 9DT and mercaptocarboxylic acid-9DT mixtures.Toluene vapor (black bars) and 1-propanol vapor (green bars) were applied at the concentration 2000 ppm in nitrogen.b) Exemplary response transients of the GNP films prepared with pure 9DT (dashed lines), pure 8MOA (pale lavender and pale green), 95% 8MOA-9DT (light gray and fluorescent green), and 95% 3MPA-9DT (gray and cobalt green) mixtures to toluene (left) and propanol (right) vapors (2000 ppm in nitrogen).The films were exposed to the analyte vapors for a duration of 120 s.Start and end of the exposure are indicated as "on" and "off".

Figure 6 .
Figure 6.a-e) Chemiresistive response isotherms of five selected GNP films.The films were exposed to toluene (black), 1-propanol (green), and water (blue) with concentrations ranging from 50 to 10 000 ppm.The data points represent the response amplitudes measured after 115 s exposure to the test vapor.Nitrogen was used as carrier gas.In figure parts a) and c) Langmuir-Henry isotherms (solid purple lines) were fitted to the data.In figure parts b), d), e) linear functions (solid orange lines indicate the fit range, dashed lines are extrapolated) were fitted to the data.f-j) Response isotherms at low vapor concentrations (50-500 ppm).Linear functions were fitted to the data (solid lines).

Figure 8 .
Figure 8. a) Lithographic process for the fabrication of GNP chemiresistor arrays: i) Silicon substrate with 4 pairs of interdigitated micro electrodes.The optical micrograph shows an interdigitated electrode structure (scale bar 200 μm).ii) A PMMA layer was deposited onto the substrate and one pair of electrodes was uncovered via DUV lithography.iii) A cross-linked GNP film was deposited onto the uncovered electrode structures via LbL-SC.On the PMMA-masked sections the GNP film was removed by dissolving the PMMA layer.Steps ii and iii were repeated using different ligand-linker compositions for GNP cross-linking.iv) The final sensor array comprised four pairs of chemiresistors with different chemical selectivity.The optical micrograph shows an example of an electrode structure covered with a cross-linked GNP film (scale bar: 400 μm).b) Photograph of the sensor array mounted on a printed circuit board (PCB).

Figure 9 .
Figure 9. Transient responses of the 100%8MOA chemiresistor to a) ethyl acetate and b) heptane at the vapor concentration 500 ppm.The three features used as input data for the linear discriminant analysis (LDA) are indicated.(ΔR∕R 0 ) max : response amplitude measured at the end of analyte exposure; A 20s : area below the transient curve limited to the first 20 s of analyte exposure; A r : area underneath the recovery curve limited to the first ≈240 s of the recovery phase.c) LDA of data sets acquired using a monolithic sensor array of four pairs of GNP chemiresistors: 9DT, 89%12T-9DT, 95%3MPA-9DT, and 100%8MOA.The concentration of each analyte ranged from 50 to 1000 ppm.Filled circles indicate data points used for computation of the model, empty circles indicate test exposures projected onto the discriminant coordinate system.The colored crosses indicate predictions of the test exposures.

Figure 10 .
Figure 10.Linear discriminant analysis (LDA) of a chemiresistor array comprising 8 different GNP films (see text).The concentrations of each analyte ranged from 100 to 1000 ppm.Filled circles indicate data points used for computation of the model, hollow circles indicate test exposures projected onto the discriminant coordinate system.The colored crosses indicate predictions of the test exposures.Using the LDA prediction function, all 19 analytes were assigned correctly.

Table 1 .
Cross-linker and monothiol (MT) ligands used in this study, molar fractions of MT ligands in MT-9DT mixtures used for sensor fabrication, and sensor notations.