Electronic noses and sensor arrays based on metal oxides semiconductors have recently been used in food quality analysis (Konduru et al., 2015a; Konduru et al., 2015b; Mei et al., 2015; Xu et al., 2017; Ghasemi-Varnamkhasti et al., 2018a, Ghasemi-Varnamkhasti et al., 2018b; Liu et al., 2018; Tohidi et al., 2018; Ghasemi-Varnamkhasti et al., 2019; Voss et al., 2019a; Voss et al., 2019b; Gamboa et al., 2019; Grassi et al., 2019; Hidayat et al., 2019; Jong et al., 2019; Liang et al., 2019; Zappa, 2019; Lee et al., 2020a; Mu et al., 2020; Rodriguez-Mendez and de Saja, 2020; Tang et al., 2020), medical diagnostics (Moon et al., 2018; Kononov et al., 2019; Liao et al., 2020; Tiele et al., 2020), environmental monitoring (Plawiak and Rzecki, 2014; Yi et al., 2014; Mondal et al., 2015; Akamatsu et al., 2017; Krivetskiy et al., 2017; Pan et al., 2017; Luo et al., 2018; Shi et al., 2018; Gancarz et al., 2019; Jaeschke et al., 2019; Lagod et al., 2019; Liu et al., 2019a; Palacín et al., 2019; Song et al., 2019; Kang et al., 2020; Khan et al., 2020; Prajapati et al., 2020; Rahman et al., 2020; Wu et al., 2020; Yang et al., 2020) and drug detection (Hernandez et al., 2020) applications, among others. While single metal oxide sensors are typically limited by poor selectivity, their incorporation into sensor arrays results in an improved ability to differentiate between single VOCs and VOC mixtures due to the unique response characteristics of different MOS sensing materials (Prajapati et al., 2020). Further selectivity optimization has been achieved through applying temperature modulation (Nakhleh et al., 2017; Prajapati et al., 2020) to the sensors, as MOS gas sensitivity is functional to temperature. Although some reports have discussed room temperature sensing mechanisms (Li et al., 2017; Abokifa et al., 2018; Li et al., 2019; Bindra and Hazra, 2019; Zhang et al., 2020a), MOS sensors typically require higher operating temperatures and thus the incorporation of a heater into the device architecture for operation. Additionally, MOS sensors tend to have nonlinear responses (Bochenkov and Sergeev, 2010), but highly linear results were obtained by applying dielectric excitation to an MOS sensor array recently (Potyrailo et al., 2020).

The sensing mechanism of MOS materials depends on temperature and the specific metal oxide material. Above 700°C, bulk defect effects dominate the conductivity response to changes in the partial pressure of oxygen, which achieves an equilibrium with the stoichiometric bulk material (Bochenkov and Sergeev, 2010). Surface conductance effects are most evident in the 400–600°C range and depend on the adsorption/desorption of monatomic oxygen (Bochenkov and Sergeev, 2010; Reghu et al., 2018). The active metal oxide material forms a sensing film which is designed to optimize mass transfer properties with both oxygen and analyte gases (Bochenkov and Sergeev, 2010), which has been accomplished by increasing the aspect ratio of these materials by depositing layers of hollow, porous, nanospherical active metal oxide material (Kanan et al., 2009b; Righettoni et al., 2015). MOS materials are polycrystalline in nature, with crystallographic faces functioning as the active sites for oxygen adsorption (Bochenkov and Sergeev, 2010) separated by grain boundaries (Korotcenkov, 2007; Reghu et al., 2018). Adsorbed oxygen attracts electrons from the bulk material, creating a space-charge layer which repels other electrons from the surface, thus forming an “electron depletion” layer. With fewer charge carriers (electrons) in the material, conductivity decreases and a potential barrier forms at the grain boundaries (Kanan et al., 2009b; Bochenkov and Sergeev, 2010). In the case of n-type metal oxides, exposure to reducing (electron-donating) gases and their subsequent adsorption to and reactions at the material surface results in more available charge carriers in the conduction band, resulting in reduction of the potential barrier at grain boundaries (Kanan et al., 2009b) and increased conductivity, with the inverse electrical effect observed with oxidizing gases. (Korotcenkov, 2007; Bochenkov and Sergeev, 2010; Righettoni et al., 2015; Reghu et al., 2018).

One strategy for tuning the gas sensing properties of nanoscale metal oxides is to selectively promote high-index crystallographic facets on the surface of the nanostructures which expose additional catalytically favorable active sites for oxygen adsorption and surface reactions with the target analytes, chiefly unsaturated metal ions with a large dangling bond density. The careful selection and control of synthesis conditions are important for promoting growth of the high-energy high index facets over low-energy, less-active low index facets. The interested reader is directed to a review which discusses synthesis strategies of and other information about high-index faceted metal oxides (Sun et al., 2019b) and a review which provides an overview of facet-engineered semiconducting metal oxides in gas sensors (Gao and Zhang, 2018). The engineering of surface facets and morphology for enhanced selectivity and sensitivity has been previously reported for WO₃ (Hu et al., 2020), TiO₂ (Liu et al., 2017; Wang et al., 2017), CuO (Hou et al., 2018) and SnO₂ (Liu et al., 2015a; Zhang et al., 2020c).

Beyond facet engineering, doping nanostructured metal oxides with homogenous, substitutional additives and heterogeneous nanostructures to modify the surface chemical reactivity and electrical properties to enhance sensitivity, selectivity and other gas sensing properties has been the subject of a significant body of research and publications. Only a small selection of works to demonstrate the previous progress or potential to use such techniques in electronic noses is discussed here; the reader is directed to the reviews by (Degler et al., 2019) and (Zhang et al., 2020b) for detailed reviews of homogeneous and heterogeneous metal oxide nanostructures for gas sensing, including the mechanisms by which dopants modify the chemical and electronic properties of metal oxides to improve sensitivity and selectivity.

Absent photoexcitation, noble and transition metal nanoparticle doping has been used as a successful strategy to improve the sensitivity and selectivity of metal oxides by creating Schottky barriers to increase electron-hole recombination time and modifying catalytic activity at the surface (Zhang et al., 2020b). Liu et al. demonstrated that decorating the surface of high-index {211}-faceted SnO₂ with Au nanoparticles improved the selectivity of the hybrid material toward ethanol over acetone, THF and methanol, and increased the sensitivity toward all analytes, albeit barely for methanol. (Liu et al., 2015a) The undecorated SnO₂ had similar responses toward acetone and ethanol, while Pd and Pt doping significantly reduced the sensitivity toward all analytes. However, despite the lowered sensitivity, Pt imparted somewhat greater selectivity toward methanol. In other works, a combination of doping Cr and/or Pt into a SnO₂ thin film with temperature modulation allowed for selectivity toward NO₂ and CO to be attained separately, (Kamble and Umarji, 2016) and greater selectivity toward triethylamine over several other VOCs was achieved by the addition of Pd nanoparticles (present as PdO) onto ZnO nanorods. (Meng et al., 2019) Tangirala et al. (Tangirala et al., 2017) investigated the effects of Cu-, Pt- and Pd doping in nanostructured SnO₂ powders, and determined that higher sensitivity toward CO was obtained using Cu nanoparticles as a dopant while also noting the impact that two different methods of introducing the dopants had on morphology of the SnO₂ structures and, consequently, the observed gas sensing performance. In summary, by imparting different catalytic activities and sensitivities toward different VOCs, metal nanoparticle doping is a potentially useful strategy to construct cross-sensitive and semi-selective metal oxide sensor arrays.

A virtual sensor array may also be constructed using a few, or even single, metal oxide sensing materials cycling through different operating temperatures. Metal oxides display an optimum operating temperature with respect to maximizing their response (or sensitivity) toward a specific analyte at a specific concentration due to the strong temperature dependence of and competition between oxygen adsorption, analyte adsorption and surface reaction kinetics (Ahlers et al., 2005). Thus, operating a metal oxide sensor at different temperatures is a viable strategy to distinguish between different analytes that may otherwise be difficult to distinguish at a single operating temperature, provided that the relationships between temperature and sensitivity for the target analytes and the metal oxide sensing material are sufficiently different. Several examples of such virtual sensor arrays and electronic noses using single or a few different metal oxide sensing materials with transient temperature cycling or variation in their operation have been previously reported (Martinelli et al., 2012; Amini et al., 2013; Fonollosa et al., 2013; Warmer et al., 2015; Hastir et al., 2016; Kneer et al., 2016; Krivetskiy et al., 2017; Durán et al., 2018; Krivetskiy et al., 2018; Tonezzer et al., 2018; Tonezzer et al., 2019a; Tonezzer et al., 2019b; Liu et al., 2019b).

Low-dimensionality carbon nanomaterials such as carbon nanotubes (CNTs) and graphene have demonstrated potential for applications in chemical sensor development, especially for artificial nose applications (Park et al., 2010; Wongchoosuk et al., 2010; Son et al., 2016; Orzechowska et al., 2019; Hayasaka et al., 2020). These carbon allotropes exhibit excellent carrier mobility and low thermal and electric noises owing to their bond structure, which is rich in sp² electrons. In addition, these carbon nanomaterials display high mechanical strength and thermal conductivity. Due to the conductive properties and optical transparency, these materials make great candidates for transparent devices for sensor applications (Yusof et al., 2019).

CNTs and graphene are composed of sp² bonded carbon atoms packed into honeycomb lattice structures (Varghese et al., 2015; Tran and Mulchandani, 2016; Shi et al., 2018); graphene has a planar (2-D) atomic monolayer structure, while CNTs assume a cylindrical pseudo 1-D structure resembling a seamlessly rolled sheet of graphene along the (m,n) lattice vector in the sheet (Aroutiounian, 2015; Liu et al., 2015c; Yusof et al., 2019). Divided into two categories based on the number of concentric atomic layers, single-walled carbon nanotubes (SWNTs) can exhibit either semiconducting or metallic electronic properties depending on chirality while and multi-walled carbon nanotubes (MWNTs) have metallic electronic properties (Gong et al., 2005; Tran and Mulchandani, 2016). Semiconducting SWNTs typically have small band gaps of 0.1–0.2 eV, which allow for room-temperature operation (Wongchoosuk et al., 2010; Aroutiounian, 2015; Zaporotskova et al., 2016). This suggests low-power requirement, which is an attractive performance feature for chemical sensors, especially in artificial nose applications that employ high-density sensor arrays. On the other hand, pristine graphene exhibit ambipolar electric field effect, which can operate in both n-type and p-type regimes (Novoselov, 2004; Geim and Novoselov, 2007; Castro Neto et al., 2009).

The chemical sensing mechanism of chemiresistive/FET-based SWNTs sensors can be categorized by intra-tube, inter-tube, and tube-electrode conduction pathways (Schroeder et al., 2019b). Intra-tube chemical sensing mechanisms are governed by modulations in charge carrier concentrations and mobility and which can occur through charge transfer, charge carrier trapping, charge scattering, and any of perturbations of ideal SWNTs structure by chemical and electrostatic interactions on the walls of SWNTs. Inter-tube conduction pathways in sensors based on a network of CNTs can also be modulated by small physical changes in tube-tube junction distance due to intercalation of analytes in the interstitial spaces, influencing the charge tunneling probability in the CNT network. Another sensing mechanism occurs through the modulation of the Schottky barrier at the tube-electrode junction (Heller et al., 2008; Salehi-Khojin et al., 2011; Lerner et al., 2012; Orzechowska et al., 2019; Schroeder et al., 2019b). Similar to SWNTs-based devices, graphene-based chemiresitive and FET-based sensors operate via charge transfer phenomena that occur between the sensing material and adsorbed gas analytes, where charge transfer direction and quantities depend on the electron-donating or electron-withdrawing nature of the analyte molecule (Yang et al., 2017; Donarelli and Ottaviano, 2018).

For chemiresistive and FET-based gas sensing applications with implications toward development of electronic nose system, semiconducting SWNTs and graphene have been extensively utilized as sensing materials, which was reviewed in various references (Dai et al., 2002; Zhang et al., 2008; Liu et al., 2012; Llobet, 2013; Varghese et al., 2015; Zaporotskova et al., 2016; Tang et al., 2017; Xu et al., 2018b; Orzechowska et al., 2019). Specifically for artificial nose applications, efforts to tune the selectivity and sensitivity of SWNT- and graphene-based chemiresistive and FET sensors take advantage of the rich library of covalent and non-covalent chemical modification strategies and the ease of integration or hybridization with a plethora of other chemicals and nanomaterials to impart additional receptor functionalities. Thus, the lack of selectivity of these graphitic nanomaterials have been overcome by introducing secondary sensing materials of various chemical composition and structures to the carbon structure to provide additional receptor functionality, ranging from macrocyclic molecules (e.g., porphyrins, phthalocyanines, pyrenes) (Su et al., 2020) DNA polymers, organic polymers), nanoparticles of various composition (e.g., metallic and semiconducting) (Mubeen et al., 2007; Mubeen et al., 2010; Mubeen et al., 2011), to larger peptide-based receptors (Kuang et al., 2010; Son et al., 2016).

One example of array-based artificial nose using carbon nanomaterials was demonstrated by Schroeder et al. using a 20-sensor array based on SWNTs noncovalently functionalized with 20 different “chemical selector” as sensing materials for identification of various food item categories (Schroeder et al., 2019a). These 20 chemical selectors, used to impart chemical selectivity to bare SWNTs, consisted of transition metal complexes, ionic liquids, porous polymers, cavitands molecules, and metalloporphyrins, which were selected based on their differential affinities to various classes of odorant molecules (Lin and Swager, 2018; Schroeder et al., 2019a). By combining machine learning approaches to guide the experimental design and characterization of the chemical sensor array, Schroeder et al. demonstrated the ability to differentiate between various samples of cheese, liquor, and oils by their odors (Schroeder et al., 2019a).

Due to their unique chemical structures, porphyrins have the ability to bind with different analytes through van der Waal forces, hydrogen bonding, electrostatic and coordination interactions with the central metal ion (Brunink et al., 1996). The interaction of porphyrins and their complexes with metals affects the delocalization of the electron charge in graphene and nanotubes, as well as the energy barrier and the size of energy gaps between valence and conductivity bands. In consequence, the conductivity in graphene can change significantly. Based on other experimental and theoretical studies, it was observed that the sensitivity of graphene-based sensors can be significantly improved by doping with Br, N, P, Ga, Cr, Mg, S, and Si (Orzechowska et al., 2019). By utilizing doping, new active sites are formed on the graphene’s surface, which have an ability to strongly adsorb gas molecules (Orzechowska et al., 2019). For example, doping graphene with Fe, N, and N and Si combined improves sensitivity toward H₂S, CO, and NO₂ (Zhang et al., 2014) while doping graphene with Mg and Cr results in an increase in sensitivity toward SO₂ (Shao et al., 2013; Liu et al., 2014; Orzechowska et al., 2019). The nitrogen atom is the active site of NO₂ adsorption in the N and Si doping, whereas doping graphene with Si significantly improves the sensitivity toward NO and NO₂. The sensor built of N-Si-G shows sensitivity toward 21 ppm NO₂ and the sensor response declines along with the gas concentration to ∼1 ppm (Niu et al., 2013). In addition, the introduction of defects in the graphene structure by doping with Br, S and N results in improved sensitivity toward formaldehyde (Zhou et al., 2014).

Another emerging strategy for achieving an artificial nose system is the integration of protein-based olfactory receptors (ORs) into the sensing materials which have been previously demonstrated by several groups to detect VOCs in gas phase and liquid phase down to the ppt level using CNT-based and graphene-based FET sensors (Goldsmith et al., 2011; Lee et al., 2012b; Son and Park, 2018). This biomimetic approach uses the ORs, which are protein-based receptors with multiple genotypic variants found in biological olfaction (over 350 different found in humans), to confer differential affinity of sensors to various gaseous odorants, depending on specific variants used (Lee et al., 2012b). The sensing mechanism using these OR-functionalized sensing materials was proposed to be attributed to the change in local ionization of specific amino acid residues in the OR upon binding to odorants, leading to local electrical modulation of the semiconducting sensing material (Lee et al., 2012b). This approach to integrate ORs onto transduction and sensing materials harnesses the natural selectivity of these polypeptide macromolecules to enable discrimination between odorants by electrical sensors and sensor arrays (Son and Park, 2018).

2D semiconducting transition metal dichalcogenide (TMDC) nanomaterials, which are binary compounds with the general formula MX₂ — M being a transition metal and X being a group VI element other than oxygen (i.e., S, Se, Te) — have been extensively engineered and evaluated for their gas sensing performance in FET and chemiresistive measurement configurations. Their use in multimodal sensor arrays and electronic noses is currently not as widespread compared to other materials, but they nonetheless present as promising candidates for the low to moderate temperature sensing of various analytes and incorporation into electronic noses. The few applications of TMDC-based chemiresistive sensor arrays reported in the literature include discriminating between atmospheric pollutants using Zn²⁺-doped MoS₂ nanosheets of variable composition (Shao et al., 2018), precursors to the explosive triacetone peroxide using variable-composition reduced graphene oxide/MoS₂ nanocomposites (Sun et al., 2019a) and oxidized/non-oxidized VOCs using organic ligand-functionalized and Au nanoparticle-doped MoS₂ (Kim et al., 2014; Cho et al., 2017). Kumar et al. (Kumar et al., 2020a) recently reviewed 2D nanostructured TMDCs as gas sensors on flexible substrates, with a focus on MoS₂, MoSe₂, WS₂, WSe₂, SnS₂ (which is technically not a TMDC but has similar properties and applications to gas sensing), TaS₂ and VS₂-based sensors. Many of the sensors they reviewed operated at ambient temperature and displayed higher selectivity toward NO₂, NH₃ or humidity. Additionally, 2D NbS₂ (Kim et al., 2019) and ReS₂ (Martín-García et al., 2019, 2) nanostructures have recently been demonstrated to act as sensitive room-temperature chemiresistors toward NO₂ and NH₃/VOCs, respectively. MoTe₂ chemiresistors and FETs have also been demonstrated to respond to NH₃ and NO₂ (Feng et al., 2017a; Shackery et al., 2018; Rani et al., 2019), with higher sensitivity and improved limits of detection under UV light excitation (Feng et al., 2017b; Wu et al., 2018b). These examples indicate that surface functionalization, doping and photoexcitation are among the tools that can be used to tune the selectivity of TMDC gas sensors and enable their incorporation into electronic nose sensor arrays.

TMDCs display a wide range of electrical properties with metallic, semiconducting and insulating behavior represented by different materials. Group VI TMDCs are narrow-bandgap semiconductors (approximately 1–2 eV optical bandgaps) whose electrical, optical and optoelectrical properties have enabled their use in electrical and optical sensors. The reader is referred to the review by Liu et al. (2015b) for a detailed review of the electronic structure and properties of molybdenum and tungsten 2D TMDCs, which are common TMDCs for gas sensing applications.

Top-down and bottom-up procedures have been employed to synthesize 2D TMDC nanomaterials for use in various gas sensors. Samadi et al. provide a detailed review of the various top-down and bottom-up synthesis methods used to make TMDC nanomaterials (Samadi et al., 2018). Of the various synthesis methods, liquid exfoliation is particularly suited for sensor array fabrication in an electronic nose. The result of liquid exfoliation is a suspension of nanosheets which is readily amenable to dropcasting or inkjet printing, both of which are applicable to the mass-fabrication of microelectronics. Typically, the suspension of nanosheets resulting from liquid exfoliation cannot be immediately inkjet printed, as its viscosity and surface tension must be modified with additives to make it printable with an inkjet printer. In a bid to simplify this process, Lee et al. recently demonstrated the inkjet printing of MoS₂, WS₂, MoSe₂ and WSe₂ nanosheets for optical sensor arrays using a simple aqueous exfoliation process with alkyl zwitterion additives; no organic solvents or other additives were required to adjust the fluid properties to make the suspensions inkjet-printable (Lee et al., 2020b). The interested reader is directed to two books about 2D TMDCs for more detailed information about their various properties and other applications (Kolobov and Tominaga, 2016; Arul and Nithya, 2020).

Similar to graphene, the gas sensing mechanism of TMDCs typically results from the reversible adsorption of an analyte followed by a charge-transfer process between the analyte and TMDC, which changes the charge carrier concentration in the TDMC, resulting in the conductivity of the material changing. Such a mechanism has been posited for the interaction between SnS₂ and NO₂ (Ou et al., 2015), WS₂ and NO₂ (Liu et al., 2020), MoSe₂ and NO₂ (Chen et al., 2019b), MoSe₂ and NH₃ (Late et al., 2014), and MoS₂ and various gases (Yue et al., 2013). This simple model of analyte adsorption and charge transfer is not applicable in all cases, however. Liu et al. demonstrated that an optimal concentration of absorbed oxygen mediates and enhances the response of WS₂ to H₂S likely via a surface redox reaction producing SO₂, similar to the mechanism of semiconducting metal oxide gas sensing, while the absorbed oxygen competes with the physisorption of NO₂ onto the surface (Figure 2). (Liu et al., 2020) Similarly, the competition between the physisorption/charge transfer and redox/direct electron transfer mechanisms of VOC gas sensing remains a matter of uncertainty with MoS₂ (Chen et al., 2019a). Thus, the sensing mechanism of TMDCs toward certain analytes in the presence of oxygen may shift to a redox/direct electron transfer-driven process compared to when it is absent and only physisorption/charge transfer can occur, a factor that should be carefully considered in the design and evaluation of TMDC-based gas sensors for a specific application. Physisorption dominates in an inert atmosphere, such as pure nitrogen, while the introduction of oxygen may enable a redox pathway for the oxidation of VOCs at the surface of and subsequent electron release into the TMDC material.

The performance and selectivity of 2D TMDC gas sensors may be improved through engineering the material surface and active sites of analyte interaction, adding trace dopants and photoexcitation. The first can be done by controlling the growth of the TMDC to expose a greater number of active sites on the surface. For example, Cho et al. increased the sensitivity of a MoS₂-based room temperature sensor toward ethanol and NO₂ by modifying the parameters of a CVD sulfurization process to induce different growth modes in the MoS₂ (depicted in Figure 3), leading to vertical growth along a higher-energy plane with a greater number of active defect sites on the surface compared to a smaller number of active defect sites present on the horizontal basal plane (Cho et al., 2015).

Differentiating between significantly different molecules such as NO₂ and ethanol with pristine MoS₂ nanosheets is clearly not difficult as seen in Figure 3, but differentiating between different VOCs, which is more applicable to some electronic nose applications, is more challenging with bare 2D TMDCs. One method of increasing the differentiability between VOCs using 2D TMDCs has been to add metal nanoparticles or organic ligands to the TMDCs. The functionalization of the MoS₂ surface with Au nanoparticles (Cho et al., 2017; Chen et al., 2019a) and mercaptoundecanoic acid ligands (Kim et al., 2014) has been demonstrated to modify the sensing response to different VOCs, with both types of functionalized MoS₂ nanosheets displaying higher increased selectivity and differentiability toward oxygen-containing VOCs vs. hydrocarbon VOCs—in one case even displaying a change in polarity of response when exposed to oxygen-containing VOCs (seen in Figures 4A,B)—when compared to bare MoS₂. The carboxylic acid group in the mercaptoundecanoic acid was suspected to interact with the negatively charged oxygen atoms in the polar oxygen-containing VOCs via hydrogen bonding, leading to a charge transfer effect into the MoS₂ nanosheets which affected their resistivity (Kim et al., 2014). On the other hand, different mechanisms for the sensitivity and selectivity enhancement toward oxygen-containing VOCs observed with the addition of Au nanoparticles were proposed: Au nanoparticles may act as a catalyst to increase the rate of or provide different pathways for the surface reactions with oxygen-containing VOCs (Chen et al., 2019a), while the p-type SiO₂/Si substrate’s interface with the MoS₂ nanosheets in combination with Au nanoparticles modified both the electronic structure and reactivity of the MoS₂ compared to the pristine MoS₂ (Cho et al., 2017). The experiments in (Cho et al., 2017) and (Kim et al., 2014) were carried out in nitrogen, while those in (Chen et al., 2019a) were carried out in air, leading to two different sets of sensing behaviors toward otherwise similar analytes; physisorption and dipole charge transfer alone are expected to be the dominant mechanisms in a nitrogen atmosphere, whereas adsorbed oxygen species potentially modify the mechanism in air by introducing a redox pathway. Within all three of these works, the Au nanoparticle and organic ligand-functionalization of the MoS₂ nanosheets enabled greater differentiation between oxygen-containing and hydrocarbon analytes. Similar developments have also been observed with WS₂ nanosheets, with Ag nanowires on the surface of WS₂ nanosheets improving their selectivity toward NO₂ over acetone and recovery performance by introducing NO₂-selective catalytic active sites (Ko et al., 2016) and Au-nanoparticle functionalized WS₂ nanosheets displaying higher selectivity toward CO over other VOCs due to the selectivity of active sites on the Au nanoparticles toward CO (Kim et al., 2020). This body of work shows that the functionalization of MoS₂ and WS₂ nanosheets with surface-exposed Au and Ag metal nanostructures and organic ligands is a viable strategy to differentiate between different VOC analytes and use in designing a cross-sensitive sensor array for an electronic nose. Of additional interest is the work of Tuci et al. (Tuci et al., 2018), which develops a method for the surface functionalization of MoS₂ with a diverse class of organic ligands; this approach may be useful for fabricating a multifunctional TMDC sensor array with mixed organic surface ligands that display different affinities toward different VOCs.

Another technique for engineering the gas sensing performance of 2D TMDCs in electronic nose applications is through doping with transition metal ions. Shao et al. developed a cross-sensitive MoS₂ nanosheet-based sensor array by varying the amount of Zn²⁺ incorporated into the nanosheets (Shao et al., 2018). A sensor array with MoS₂ sensing elements containing different compositions of Zn²⁺ (ranging from 1 to 15% atomic) and operating at room temperature could differentiate between ozone (O₃), N₂O, formaldehyde (CH₂O), ethanol and different mixtures thereof using a combination of kinetic and thermodynamic response features. Another application of transition metal ion doping included the substitutional doping of MoSe₂ nanosheets with a small amount of Nb²⁺; this slightly decreased the nanosheets’ sensitivity to NO₂ compared to the undoped MoSe₂, but significantly increased the long-term stability of the baseline and gas sensing response by slowing oxidative degradation (Choi et al., 2017). Panigrahi et al. have provided future direction for the development of higher-performance 2D MoSe₂ and MoTe₂ gas sensing materials with computational predictions that doping them with low concentrations of Ge and Sb, respectively, will significantly enhance the sensitivity of these materials toward NH₃, NO₂ and NO (Panigrahi et al., 2019).

Polymers are commonly used as insulators and are widely used in daily life due to their low-cost fabrication, optical and mechanical properties, thermal stability, and processability. The first polymeric sensors were obtained by dispersing conductive nanofillers into insulating polymers; however, the first conductive polymer was discovered by Green and Street in 1975 (Yi and Abidian, 2016). Conductive polymers are heterocyclic compounds with alternating single and double bonds along their backbone, creating conjugated π-electron systems (Rodriguez-Mendez and de Saja, 2020). Some commonly used conductive polymers include Polyaniline (PANI), Polypyrrole (PPy), Poly-3-methylthiophene (P3MT), Polythiophene (PT) (Wong et al., 2019).

These polymers can be arranged in various morphologies such as films (Joshi et al., 2011), thin films (Kumar et al., 2017), nanofibers (Yang and Li, 2010), nanowires (Zhang et al., 2009), nanowire arrays (Hassanzadeh et al., 2012), nanotubes (Yoon et al., 2006), nanoparticles (Lee et al., 2014), pellets (Chabukswar et al., 2001), nanoribbons (Lu et al., 2011), and other morphologies. The surface morphology of conductive polymers plays a significant role in the gas sensing performance, especially as nanostructures. Due to their larger specific surface area, 1D conductive polymers nanostructure are considered excellent material for gas sensing (Lu et al., 2011). It was observed that a thin layer of PPy sensor displayed a response of 1.38 in terms of the ratio of sensors resistance in air to its resistance in the presence of the target analyte to a concentration of 12,500 ppm of ammonia at room temperature (Chen et al., 2007), whereas a thin film of PPy displayed a response of 2.56 toward 10,000 ppm of ammonia at room temperature (Wong et al., 2019). Conductive Polymers are most commonly synthesized by chemical or electrochemical polymerizations of the corresponding monomers, but other methods can be used such as photochemical polymerization, metathesis, and plasma polymerization (Kus et al., 2018). The substrate and morphology have a significant influence on the sensitivity, reliability, accuracy, response time, susceptibility to interferences, and shelf life of the array. (Askim et al., 2013).

Conductive polymer-based sensors were used in the first commercial artificial nose system, and have demonstrated their effectiveness for vapor sensing (Yi and Abidian, 2016). Since then, polymers have been utilized in the development of a wide range of sensors due to their excellent transducing material and response to various chemical and physical stimuli (Guadarrama et al., 2000). The conjugated backbone consisting of alternating single and double bond structures displays electrical conductivity due to the sp² hybridized carbon atoms on the polymer backbone. The parallel p orbitals above and below the carbon atom form a π bond constructs a pathway for the charge carriers transported along the polymer chains (Yi and Abidian, 2016). However, organic conjugated polymers generally lack intrinsic charge carriers. Therefore, external charge carriers are introduced to the polymers via doping. Conductive polymers can either be partially oxidized by electron acceptors or partially reduced by electron donors (Dai, 2004). Based on band theory, it is known that insulators have a large band gap between the conduction band and valence band. When conducting polymers are doped, either the valance or conduction band is filled, or polarons are generated during the process.

The conductive properties of the polymers are attributed to the intra and inter chain transports (Hangarter et al., 2013). In p-type sensing material, oxygen molecules from the air are chemisorbed onto the surface, removing electrons from the conduction band. The adsorbed oxygen is then converted into double or single oxygen ions, leading to ionosorption on the surface. The removal of electrons leads to a decrease in electron density and an increase in hole concentrations, which leads to a diminution in the resistance. As a reducing gas such as ammonia reacts with the ionosorbed double oxygen ion species, electrons are absorbed by the conduction band of the p-type material. As a result, the hole concentration decreases, and the resistance increases. When an oxidizing gas such as NO₂ is introduced, electrons are depleted from the valence band, which leads to an increase in hole concentration and decrease in resistance (Arafat et al., 2012; Baharuddin et al., 2019). The sensing mechanism is the opposite for the n-type material. The sensing mechanism of carbon-polymer composites in artificial nose systems is observed in devices such as Cyrano Sciences’ Cyranose 320, which consists of 32 individual polymer sensors blended with carbon black composite (Dutta et al., 2002). The device is configured as an array of sensors, which swell when exposed to VOCs changing the conductivity of the carbon pathway. The swelling leads to a change in the resistance across the array, which is captured as a digital pattern that is a representation of the test smell. Each VOC regardless of its complexity has a distinct response signature and ‘smell print’, which is specific to the stimulus (Dutta et al., 2002).

Conductive polymers can be either n- or p-doped, and this doping provides charge carriers and modifies the band structure. Various counterions can be used as dopants to modify films to obtain various physicochemical properties. The dopants can also be deposited as thin films onto interdigitated electrodes using inkjet deposition, electrospinning, or electrodeposition to obtain films with various structures, hydrophobicity, thickness, and roughness (Rodriguez-Mendez and de Saja, 2020).

Due to their modifiable selectivity, short response time, ease of synthesis, mechanical properties, and capability to operate at room temperature, conductive polymers have been materials of interest for artificial nose application since the 1980s (Wilson and Baietto, 2009). Doped arrays of nano- and micro-structured conducting polymers doped with perchlorate (ClO⁻⁴), para-toluene sulfonate (pTS⁻), chloride (Cl⁻), Trichloroacetate (TCA⁻), dodecyl sulfonate (DS⁻), and dodecylbenzene sulfonate (DBS⁻) were capable of detecting acetone, methanol, ethanol, 1-propanol, 2-propanol, nitromethane, propylamine, pyridine, and gas mixtures of aliphatic alcohols (Alizadeh et al., 2016). Doping can also enhance the sensing mechanism of arrays of microchemiresistors. An electronic nose system was modified with bio-inspired nanofibrous artificial epithelium to produce a microchemiresistor covered with electrospun nano-fibrous structures that were prepared by blending doped poly emeraldine, with polyethylene oxide, polyvinylpyrrolidone, and polystyrene, which acted as the charge carriers for the conducting polymer (Alizadeh et al., 2016). Conductive polymer-based artificial nose systems have been utilized in the detection of bacterial wetwood detection in Fagus grandifolia and Prunus serotina Sapwood based on the detection of headspace volatile microbial and plant metabolites derived from sapwood. The sensor array of an Aromascan A32S conductive polymer along with principal component analysis and quality factor techniques could provide unique and identifiable aroma signature profiles for four healthy and wetwood-infected sapwood core types. The principal component analysis showed that not only were the healthy and wetwood-infected samples distinguishable, but there was a clear distinction between the profiles of the healthy American beech and black cherry sapwood cores (Wilson, 2014). In another example, conductive polymer-based E-nose systems were used for the discrimination of various olive oil samples based on their aromas. A system of eight different polymeric gas sensors were prepared by electrodeposition under varying conditions to produce and dope PPy, P3MT, and Polyacrylonitrile (PAN) thin films with different properties. The sensors were then repeatedly exposed to the headspace of olive oils and pattern recognition techniques were used to discriminate the signals. It was observed that each sensor had a unique response when exposed to the sample of olive oil. The response of P3MT sensors generated using different doping electrolytes to extra virgin olive oil is observed. This sheds light on the significant contribution of dopants to the sensitivity and selectivity of the generated sensors with respect to the different VOCs (Guadarrama et al., 2000). Furthermore, the response of the P3MT sensor array to various VOCs are observed. The sensor displays observable sensitivity to all the samples including the sample of flat olive oil. It is observed that various polymeric sensors prepared from different monomers and dopants displayed varying degrees of sensitivity when exposed to the headspace of an extra virgin olive oil. This is attributed to the different chemical natures of the monomers and the dopants used in the generation of the thin conductive polymer films.

Surface plasmon based gas sensor was first demonstrated in 1983 (Liedberg et al., 1983) and remarkable growth has since been made over the last few decades as an analytical route in not only the gas detection field but also biology, energy, environment, and catalyst related fields (Le Ru and Etchegoin, 2008). Surface plasmons are the collective charge oscillations of free electrons in the metal caused by the interaction with electromagnetic waves. With the excitation of the surface plasmon, light is either coupled into propagating or standing surface plasmon depending on its structure. With the 1D or 2D plasmonic structure, propagating surface plasmon (PSP) can be excited and its propagation constant of the surface plasmon is given by the following expression: β = k ε m ε d ε m + ε d (2) where k is the free space wave number, ε m   a n d   ε d are the dielectric constants of the metal and environment, respectively. As the refractive index ( ε d   o r   n d ) changes due to the gas molecules, the propagation constant changes, resulting in the change of the resonance condition between the surface plasmon and incident light. The change subsequently alters wavelength, frequency (phase), and/or polarization of the photon motion in the light in the output and allow for estimation of the refractive index of the adsorbed molecules on the metal. We refer reader to (Schasfoort, 2017) for more theoretical background of the surface plasmons.

For the excitation of the surface plasmon, its momentum must match that of the incident light. The momentum of the surface plasmon is larger than the incident light in air based on the dispersion relation. The increase of the momentum can be fulfilled by prism, grating, and waveguide couplings, which are the common configurations for the excitation of the surface plasmon (Figure 6). For the prism coupling, the light wave in a high refractive index prism with a certain angle can generate an evanescent wave. If the momentum of the evanescent wave matches that of the surface plasmons, a PSP is excited at the other interface between the metal film and environment with gas. As the result, a significant reduction of the reflected light occurs due to the excitation of the surface plasmon at the angle. Another type of SPR excitation method is grating coupling as shown in Figure 6B. Here, the diffraction by grating is employed to increase the momentum of the incident light. The excitation of the PSP by prism and grating couplings depends on the optical properties of the dielectric and any change of the refractive index by the adsorption of the analytes on the metal surface subsequently alters the resonance angle of surface plasmon. The configurations are highly sensitive, simple, and reproducible, and therefore the technologies have been intensively utilized in the past 3 decades as a common sensing platform (Homola and Piliarik, 2006; Brenet et al., 2018). However, identification of gaseous components from multiple orders is still challenging by monitoring RI change alone. Multiple sensing arrays with different chemical specificity must be utilized to discriminate groups of target species. Brenet et al. have demonstrated the detection of VOCs with 72 cross-reactive sensor microarrays (Figure 7). The microarrays were functionalized with 18 biomimetic peptides and organic molecules by thiol-gold chemistry. Each peptide and organic molecule with different physicochemical properties (e.g., hydrophobic, hydrophilic, charged, neutral, etc.) allowed for distinguishing between homologous VOCs with the difference of a single carbon atom (Brenet et al., 2018).

Recently, nanoplasmonic sensors have attracted attention in the field of the chemical sensors due to their high potential for multiplex sensing and miniaturization of the plasmonic sensors. The nanoplasmonic sensors employ tightly localized surface plasmon (LSP) in the vicinity of the engineered metallic nanoparticles or nanostructures instead of the PSP on the metallic film for the conventional surface plasmon techniques (Figure 6D). The nanoplasmonic sensors can reduce the sensor volume and cost by utilizing a simple colinear transmission or reflection illumination system instead of bulky prism-coupling system for PSP techniques (Haes and Duyne, 2004; Gao et al., 2014). Chemical sensing with LSP exploits the two unique effects of the LSP: 1) electric fields in the vicinity of the plasmonic nanoparticles or in the small gap generated by the nanostructure are drastically enhanced and 2) the field enhancement has the maximum at the resonance frequency which depends on the size, gap distance, and chemical composition of the nanoengineered materials. The LSP resonance frequency or the wavelength, which are related with the refractive index of the environment, is as follows: ω m a x = 2 π c λ = ω p 2 ε d + 1 (3) where ω p is the plasma frequency, c is the light velocity in air, and λ is the wavelength of the light. PSP based gas sensor is known to have sensitivity to the bulk refractive index four times higher than that of the LSP sensor. (Haes and Duyne, 2004) The higher sensitivity of the PSP results from the large surface area and the long decay length (∼200 nm). With respect to the same footprint, however, the LSP can offer higher sensitivity (Bingham et al., 2010). By harnessing the tunable resonance frequency and well-developed functionalization chemistry of the plasmonic nanomaterials, a single nanoparticle could possess a potential to serve as an independent sensor taking the advantage of its high sensitivity and the short decay length (30 nm) (McFarland and Van Duyne, 2003; Svedendahl et al., 2009). In addition, the extremely large surface area of nanomaterials substantially provides enough sites for adsorption or condensation of the vapor, leading to the surface refractive index change. (Cheng et al., 2007) There have been numerous proof of concept studies demonstrating the discrimination ability of the LSP among different gaseous molecules (Karakouz et al., 2008; Chen and Lu, 2010; Joy et al., 2012); however, Au is the only material potentially applicable for the artificial nose due to its high chemical stability and high Q factor of absorption frequency (Le Ru and Etchegoin, 2008). Thus, its tunable wavelength range of surface plasmon with high sensitivity is limited at red or near IR wavelength, where resonance peak absorption is broadening due to the inhomogeneous size distribution of the large particles. Although the difficulty in the synthesis of uniform NPs and fabrication of highly ordered arrays prevented the sensor from being tuned at the most desired wavelength to reduce the absorption and scattering bandwidth (sharpness) of LSP, the ongoing advances in optical lithography technology will solve the issues with the band broadening of LSP and eventually enhance the discrimination capability between a multitude of odorants.

Analysis based on the vibrational spectroscopy has many advantages in that it can provide the fingerprint of the molecules directly, and therefore labeling the sensing materials with a chemical indicator or dense sensing arrays to create unique patterns corresponding to the multitude of analytes is not necessary. Among the various vibrational spectroscopy, Raman spectroscopy is the most widely used technique in chemical analysis, and it utilizes visible light, which does not often face interference from the main components of ambient air (nitrogen, oxygen and water vapor). But the Raman scattering is an extremely weak phenomenon; hence, detection of the dilute analytes such as gaseous odorants in atmosphere is not an easy task. LPSR can enhance the weak Raman scattering by up to 11 orders of magnitude, and it allows for vibrational fingerprint of even a single molecule to be observed (Le Ru and Etchegoin, 2008). This technique is called surface enhanced Raman spectroscopy (SERS), which can occur near the surface of metallic nanostructures, especially within a plasmonic hotspot, which is a gap smaller than a couple of nm created by metallic nanostructures. Due to the technical difficulty in uniformly fabricating such a small gap distance, six to eight orders of magnitude enhancement is realistic values with the good SERS substrates (Le Ru and Etchegoin, 2008). Compared to fluorescence or QD, the narrow bandwidth of the Raman scattering enables the integration of multiple adsorbents with the SERS arrays and the distinct peaks at different wavenumber can offer multiplex molecular target detection. Due to the extremely low Raman cross-section of the gas, however, detection of the analyte freely moving around in air is still not quite feasible. For this reason, adsorption of the gaseous molecules on SERS substrates is prerequisite. The adsorption of the gaseous molecules has been enhanced by either introducing the materials with a high affinity to the analytes (Rae and Khan, 2010), prolonging the retention time of the analytes to be adsorbed using dendritic Ag nanocrystals (Zhang et al., 2017), or applying electric field to the Ag-coated SiO₂ beads to increase the adhesion of electrostatically charged molecules (Lin et al., 2013) with the gas sensitivity down to ppb level. Compared to other sensing techniques, the adsorption of the analytes without specificity is preferential for SERS-based sensor because the vibrational fingerprint of the molecules can be obtained directly. The high adsorption capacity for the enhanced SERS effects often causes the substrate contamination leading to the signal drift and loss of the function. Improvement of the reversibility is essential for continuous and effective operation over a long period of time.