Advancements and Strategies for Selectivity Enhancement in Chemiresistive Gas Sensors

Abstract

Chemiresistive gas sensors are extensively employed in environmental monitoring, disease diagnostics, and industrial safety due to their high sensitivity, low cost, and miniaturization. However, the high cross-sensitivity and poor selectivity of gas sensors limit their practical applications in complex environmental detection. In particular, the mechanisms underlying the selective response of certain chemiresistive materials to specific gases are not yet fully understood. In this review, we systematically discuss material design strategies and system integration techniques for enhancing the selectivity and sensitivity of gas sensors. The focus of material design primarily on the modification and optimization of advanced functional materials, including semiconductor metal oxides (SMOs), metallic/alloy systems, conjugated polymers (CPs), and two-dimensional nanomaterials. This study offers a comprehensive investigation into the underlying mechanisms for enhancing the gas sensing performance through oxygen vacancy modulation, single-atom catalysis, and heterojunction engineering. Furthermore, we explore the potential of emerging technologies, such as bionics and artificial intelligence, to synergistically integrate with functional sensitive materials, thereby achieving a significant enhancement in the selectivity of gas sensors. This review concludes by offering recommendations aimed at improving the selectivity of gas sensors, along with suggesting potential avenues for future research and development.

1. Introduction

Gas sensors are essential for various applications including environmental monitoring, industrial safety, disease diagnosis, smart agriculture and food quality assessment [1,2,3,4].

A variety of gas sensor technologies, each based on distinct operating principles, have been developed, including electrochemical, optical, catalytic combustion, and chemiresistive gas sensors [5]. Compared to these methods, chemiresistive sensors exhibit compelling advantages crucial for sensing applications. Electrochemical sensors offer good selectivity and sensitivity, but they often involve higher material costs and complex packaging [6,7]. Optical sensors provide high accuracy and stability but are typically bulky, expensive due to sophisticated light sources/detectors, and less amenable to miniaturization [8]. Catalytic combustion sensors are robust and simple but consume significant power, pose potential safety risks in explosive environments, and have limited scope for miniaturization [9]. In contrast, chemiresistive sensors stand out due to their significantly lower fabrication and operational costs, high sensitivity achievable through nanostructured materials, and inherent potential for miniaturization and integration into compact systems like portable devices and IoT nodes [10,11].

However, chemiresistive gas sensors often struggle to selectively detect and measure target gases in complex environments with multiple gases due to cross-sensitivity and poor selectivity. Therefore, the development of gas sensors with high sensitivity, excellent selectivity, and reliability is necessary for practical applications. The key to solving this problem lies in gaining an in-depth understanding of the strategies and mechanisms for enhancing gas sensing performance [12].

The working principle of chemiresistive gas sensors is based on the change in electrical conductivity of the sensitive material caused by redox reactions when exposed to the target gas environment. There are four main mechanisms based on different sensing material types (Figure 1). Among them, gas sensors based on semiconductor metal oxides (SMOs) are widely used due to their remarkable resistive modulation effect and well-established fabrication processes. To overcome the limitations of its intrinsic selectivity and relatively high operating temperature, researchers have developed a variety of refined modulation strategies, including noble metal nanoparticle catalysis [13], oxygen vacancy defect engineering [14] and heterojunction construction [15], aiming to optimize surface reaction activity and electron transport pathways [16]. Meanwhile, metal and alloy nanostructures, with their unique plasmonic effects and catalytic properties, demonstrate advantages in the detection of specific gases (such as H₂) [17,18]. Conjugated polymers (CPs), with their solution processability, potential for room-temperature operation, and ease of molecular structure modification to achieve “lock-and-key” type recognition, offer new insights for flexible and wearable sensors [19]. Finally, two-dimensional materials represented by transition metal dichalcogenides (TMDs), graphene, and their derivatives, with their large specific surface area, high carrier mobility, and tunable electronic band structures, have paved the way for the development of highly sensitive and low-power devices [20,21,22].

With continuous advancements in materials science, the convergence of data science and system integration technologies is driving the emergence of transformative innovations in the field of chemiresistive sensors [23]. In addition, the synergistic integration of bionic technology and machine learning techniques offers a robust solution to the persistent challenge of gas identification within complex background environments [24]. Furthermore, there is significant variation in the methods used to evaluate gas selectivity across studies, and a standardized approach for assessing gas selectivity is lacking. Existing indicators based on response ratios struggle to reflect interference effects in real-world complex environments such as humidity, temperature fluctuations, and mixed gases resulting in poor comparability between studies. Therefore, the establishment of a standardized testing system is needed [25].

While extensive reviews have been documented on chemiresistive gas sensors [26,27], there has been limited comprehensive discussion on selectivity—a critical parameter that directly dictates the accuracy and reliability of sensing data in real-world applications. This review systematically elaborates on innovative strategies to overcome the poor selectivity of chemiresistive sensors. We emphasize material-level modifications, including crystal structure engineering, defect regulation (e.g., oxygen vacancy creation, doping), surface functionalization (e.g., noble metal decoration), and heterostructure construction. In addition, we introduce integration systems such as sensor arrays, and multi-modal sensing platforms for the practical implications. It also explores the current status, potential, and challenges of emerging disruptive technologies.

2. Gas Sensing Mechanism of Chemiresistive Gas Sensors

The gas sensing mechanism of chemiresistive gas sensors lies in the changes in the electrical conductivity due to the interactions between sensitive materials and target gas species. This involves three primary principles of a sensor: the receptor function (interaction with gas), the transducer function (chemical to electrical signal conversion), and the utility factor (gas diffusion efficiency) [28].

(1)

Gas sensors based on semiconductor metal oxides

For typical semiconductor metal oxides, the sensing mechanism relies on oxygen species adsorption and redox reactions with the gas molecules. Initially, oxygen molecules adsorb onto the surface of semiconductor metal oxides (MOSs) and subsequently convert into various adsorbed oxygen species depending on the operating temperature. Below 150 °C, $O_{2\left(ads\right)}^{-}$ species exist as stable oxygen ions. Between 150 °C and 450 °C, $O_{\left(ads\right)}^{-}$ is stable. While above 450 °C, $O_{\left(ads\right)}^{2-}$ ions are the predominant species, as shown in the following Equations (1)–(4). This electron capture depletes near-surface carriers, increasing resistance. Reducing analyte gases react with these adsorbed oxygen species, releasing electrons and decreasing resistance.

$$O_{2\left(gas\right)} \to O_{2\left(ads\right)}$$

(1)

$$O_{2\left(gas\right)}+e^{-}\to O_{2\left(ads\right)}^{-} \left(T<150 °\mathrm{C}\right)$$

(2)

$$O_{2\left(ads\right)}^{-}+e^{-}\to 2O_{\left(ads\right)}^{-} \left(150 °\mathrm{C}<T<450 °\mathrm{C}\right)$$

(3)

$$O_{2\left(ads\right)}+4e^{-}\to 2O_{\left(ads\right)}^{2-} \left(450 °\mathrm{C}<T\right)$$

(4)

For n-type SMOs (e.g., ZnO, In₂O₃) [29,30], an electron depletion layer (EDL) forms due to surface oxygen adsorption and ionization, creating a “semiconductor core-high-resistance EDL shell” structure that requires overcoming a potential barrier for carrier transport. As shown in Figure 2a, when reducing gases such as CO contact the surface, they react with adsorbed oxygen, releasing electrons back into the EDL, which lowers its resistance and significantly decreases the sensor’s overall resistance [31]. For p-type semiconductor oxides (e.g., CuO and NiO) [32], surface oxygen adsorption captures electrons, leading to hole accumulation at the surface and forming a low-conductivity core with a hole accumulation layer (HAL) shell. This creates a dual-channel conduction system: a high-mobility HAL channel and a low-conductivity core channel. When reducing gases react, injected electrons recombine with HAL holes, reducing HAL conductivity and increasing the sensor’s overall resistance (Figure 2b) [31].

The gas sensitivity of SMOs is influenced by band structure, surface defects, grain size, chemical modifications, temperature, and humidity [33]. The semiconductor type (n-type or p-type) affects carrier transport direction, while micro/nano structures impact gas adsorption and reaction kinetics, affecting sensitivity, selectivity, and stability [26].

(2)

Gas sensors based on two-dimensional materials

The two-dimensional (2D) materials, such as graphene and TMDs, rely on direct charge transfer with gas molecules for sensing at room temperature [34,35,36], as displayed in Figure 3a. For n-type 2D semiconductors (e.g., MoS₂ and MoSe₂) [37], electron-accepting gases reduce electron concentration and increase resistance. For p-type 2D semiconductors (e.g., WSe₂ and black phosphorus), electron acceptor gases increase hole concentration and decrease resistance, while electron donor gases reduce hole concentration and increase resistance. Gas adsorption not only causes band bending (ΔE_g) and Fermi level (E_F) shifts through surface charge transfer but also alters the Schottky barrier height (V_built-in) at the material-metal electrode interface, impacting the transport efficiency of majority carriers and increasing resistance (Figure 3b) [38]. This dual mechanism, involving surface charge transfer and Schottky barrier modulation, underpins the high-sensitivity and reversible gas sensing capabilities of two-dimensional materials [39].

3. Selectivity Enhancement Strategies for Chemiresistive Gas Sensors

The performance of chemiresistive gas sensors is critically evaluated through several key parameters, including sensitivity, selectivity, long-term stability, response time, and the recovery characteristics signifying the device’s ability to return to its baseline resistance upon removal of the target gas [40]. The reversible response, underpinned by the efficient desorption of analyte molecules from the sensing material’s surface, is fundamental for the sensor’s practical utility and long-term reliability, enabling repeated and consistent measurements [41]. However, achieving high selectivity—the capability to distinguish a specific target gas accurately amidst complex mixtures containing potential interferents—remains one of the most persistent challenges for chemiresistive sensors. Overcoming this limitation necessitates deliberate modifications to the sensing materials or device architecture. Consequently, this chapter focuses on exploring various enhancement strategies specifically designed to augment the selectivity of chemiresistive gas sensors.

3.1. Semiconductor Metal Oxide-Based Gas Sensors

Gas sensors based on metal oxide semiconductor (MOS) primarily utilize n-type (e.g., ZnO, SnO₂) and p-type (e.g., NiO, CuO) as sensing materials, which detect gases through electrical resistance modulation upon surface adsorption [16]. There are various approaches to significantly improve the selectivity of gas sensors based on SMOs. We present a discussion in the following two aspects: (1) functionalization of SMOs with catalysts, and (2) defect introduction and phase control. Catalytic functionalization enhances reactivity by depositing noble metals that dissociate oxygen molecules into reactive intermediates, accelerating target-specific gas reactions [13]. Defect generation and phase control modify material structures to generate oxygen vacancies or lattice defects, augmenting charge carrier density and creating preferential adsorption sites [14]. These methods collectively optimize surface chemistry and electronic properties for selective gas identification.

3.1.1. Functionalization of Semiconductor Metal Oxides with Catalysts

Metal nanoparticles (NPs) demonstrate remarkable electronic properties and catalytic activity toward various gases [42,43,44]. The sensing performance of sensors is significantly enhanced by incorporating metal nanoparticles on the surface of SMOs. This is mainly attributed to the “chemical sensitization” achieved through the spillover effect of metal nanoparticles and the “electronic sensitization” resulting from the formation of metal–semiconductor Schottky contacts [45]. Particularly, noble metals such as platinum (Pt), palladium (Pd), gold (Au), and silver (Ag) are commonly used in catalysts to activate or dissociate gas molecules, accelerating their reactions with oxygen ions on the surface of SMOs [13]. Schottky barriers between these metals and SMOs can be adjusted by gas adsorption, thus affecting the resistance of the sensor.

In addition, metal catalysts enhance the selectivity of gas sensors by providing specific surface active sites that preferentially catalyze the adsorption and dissociation of target gas molecules while simultaneously suppressing the reactions of non-target gases through electronic or geometric effects. For example, Pd-specific dissociative adsorption of H₂ forming PdH_x intermediates for highly selective H₂ sensing [46], and gold (Au) is effective for oxidizing flammable gases detection [47]. Bae et al. [48] synthesized Au NPs-decorated ZnO composite by atomic layer deposition and thermal evaporation method for gas sensing (Figure 4a). By controlling the size and density of the Au NPs, the sensor based on Au-ZnO achieved a 4.99% response for 50 ppb methyl mercaptan, with a detection limit of 50 ppb, and high selectivity against H₂S. The improved sensing performance of the Au-ZnO sensor is attributed to the Au NPs catalyst, which enhances oxygen ionization, accelerates the oxidation of CH₃SH, and thus increases the conductivity of the sensor (Figure 4b). In addition, the Au-ZnO sensor exhibits high sensitivity even at 73% humidity, offering a new potential method for the early diagnosis of periodontal disease. Teng et al. [49] present a one-step electrospinning synthesis of mesoporous PdO-functionalized SnO₂ composite nanotubes (SPCTs) for highly selective NO₂ detection at room temperature. The optimized 3-SPCT sensor exhibits exceptional selectivity toward NO₂ over interfering gases (CO, H₂, CH₄, H₂S, NH₃), attributed to synergistic effects from PdO functionalization and unique nanostructure. The sensor based on SPCTs achieves a high response value of 23.3 toward 100 ppm NO₂ with a rapid response time (1.33 s) and ultra-low detection limit (10 ppb).

Apart from single metal catalysts, combining two metals as catalysts for gas sensors can not only produce “electronic effects” (where one metal alters the electronic state of the other) and “geometric effects” (the formation of new active sites at the interface), but also enable “series catalysis” (the sequential activation of reactants) [13,50]. This design improves selectivity for complex gases, reduces interference, and boosts catalytic stability through modulation of metal ratios or the formation of core–shell/alloy nanostructures. For example, Liu et al. [51] studied AuPd alloys deposited on SnO₂ hollow spheres for gas sensors. The AuPd-SnO₂ sensor enables ultra-selective dimethyl disulfide (DMDS) detection via sulfur spillover. Charge transfer from AuPd to SnO₂ strengthens DMDS adsorption (~2.28 eV), which is further confirmed by DFT. In situ DRIFT and XPS measurements reveal that DMDS can be decomposed to atomic S on AuPd and migrated to the surface of SnO₂, reacting with adsorbed oxygen to release electrons. This pathway grants 36.6 response to 10 ppm DMDS at 135 °C. Moreover, due to the weaker adsorption and distinct kinetic behavior of interfering gases (such as hydrogen sulfide and acetone), the sensor achieves highly selective detection of DMDS. Le et al. [52] developed PdAu alloy encapsulated within ZnO shells. The synergistic effects of the composite structure enable Au to enrich surface oxygen species, while Pd in the alloy resists oxidation, thereby preserving hydrogen adsorption sites. The PdAu-ZnO sensors yield a high response value of 80 toward 100 ppm H₂ at 300 °C and exceptional selectivity. Li et al. [53] functionalized PdPt bimetallic nanoparticles with a Pd-rich shell and Pt-rich core on SnO₂ nanosheets, achieving temperature-dependent dual selectivity: high response to CO at 100 °C and the ability to detect CH₄ at 320 °C. This enhanced selectivity is attributed to synergistic chemical sensitization, PdO-mediated electronic sensitization, and Schottky barrier modulation at the metal-semiconductor interface.

To further improve the utilization efficiency of metal catalysts, single-atom catalysts (SACs) have been proposed for gas sensors [54]. This sensing material design is composed of atomically isolated catalyst atoms stabilized on the surface of the support SMOs materials. SACs improve catalytic efficiency through chemical sensitization. A study demonstrated that using atomic layer deposition (ALD) to anchor platinum (Pt) atoms on a 9 nm tin dioxide (SnO₂) film forms a Pt-SnO₂ SAC system [55]. High-angle annular dark-field scanning transmission electron microscopy confirmed the uniform dispersion of Pt atoms, while X-ray photoelectron spectroscopy showed Pt in a metallic state (Pt 0) with strong bonding to SnO₂ (Figure 4c). The sensor based on Pt-SnO₂ exhibits exceptional sensing performance toward triethylamine (TEA) with a response value of 136.2 at 200 °C, 9 times higher than that of pure SnO₂ (8.76), as shown in Figure 4d. In addition, the sensor shows low detection limit of 7 ppb, and rapid response/recovery times of 3 and 6 s. As shown in Figure 4e, compared with pristine SnO₂ sensor, the Pt-SnO₂ sensor exhibited significantly enhanced selectivity toward TEA detection, demonstrating that the single-atom Pt catalysts stabilized on the surface of SnO₂ serve as favorable adsorption sites for TEA molecules over other interferent gas species.

Figure 4. (a) Schematic illustration of the surface design of ZnO nanofilms with the strategic incorporation of Au NPs. (b) Schematic representation of the distinct sensing mechanisms for ZnO based gas sensors and Au NPs-decorated ZnO-based gas sensors. Reproduced with permission from ref. [48]. Copyright 2022 Elsevier. (c) The XPS spectra of Pt 4f in Pt/SnO₂ thin films after annealing in Ar/H₂ at 500 °C, PtO/SnO₂ after annealing in air at 500 °C and as-deposited PtOx/SnO₂ without annealing. (d) Linear fitting responses of SnO₂ and Pt/SnO₂ thin films. (e) Selectivity coefficients of SnO₂ and Pt/SnO₂ sensors toward various 10 ppm gases relative to TEA. Reproduced with permission from ref. [55]. Copyright 2020 Royal Society of Chemistry.

Figure 4. (a) Schematic illustration of the surface design of ZnO nanofilms with the strategic incorporation of Au NPs. (b) Schematic representation of the distinct sensing mechanisms for ZnO based gas sensors and Au NPs-decorated ZnO-based gas sensors. Reproduced with permission from ref. [48]. Copyright 2022 Elsevier. (c) The XPS spectra of Pt 4f in Pt/SnO₂ thin films after annealing in Ar/H₂ at 500 °C, PtO/SnO₂ after annealing in air at 500 °C and as-deposited PtOx/SnO₂ without annealing. (d) Linear fitting responses of SnO₂ and Pt/SnO₂ thin films. (e) Selectivity coefficients of SnO₂ and Pt/SnO₂ sensors toward various 10 ppm gases relative to TEA. Reproduced with permission from ref. [55]. Copyright 2020 Royal Society of Chemistry.

3.1.2. Defects Generation and Phase Control

Experimental evidence proves that defect engineering in SMOs sensing materials plays an essential role in tuning the electronic structure and electrical conductivity of SMOs, as well as enhancing the number of active sites for chemical reactions [14]. These combined effects contribute to improved gas adsorption characteristics. In the field of defect engineering, methods such as ultraviolet irradiation [56,57], defect induction [58], plasma etching [59,60], and ion irradiation [61] have been demonstrated to introduce defects and reduce the coordination number of surface atoms. In particular, the introduction of controllable oxygen vacancies can narrow the band gap, increase the carrier concentration, and enhance the chemical adsorption of target gases [14]. The literature reports that the selectivity of ZnO-based gas sensors can be effectively enhanced via defect engineering. Li et al. [60] introduced oxygen vacancies into ultrathin ZnO films using Ar plasma (Figure 5a). The gas sensor exhibited a significantly enhanced selectivity toward triethylamine (TEA) (Figure 5b,c). This improvement is attributed to the relatively low C-N bond energy of TEA (307 kJ/mol), which makes it more likely to react at oxygen vacancy sites. Shin et al. [61] irradiated ZnO nanoparticles with Xe ions and achieved the optimization of oxygen vacancies at a dose of 1 × 10¹⁵ ions/cm². The sensor achieves a high response value of 88.5 toward 10 ppm NO₂ and excellent selectivity. This is mainly attributed to the high electron affinity (2.28 eV) and low O-NO bond energy (305 kJ/mol) of NO₂, which makes it more prone to adsorption on vacancy-rich surfaces. The above two methods both achieve highly selective responses to specific gases through defective engineering. Paolucci et al. [62] controlled oxidation of SnSe₂ yields amorphous 2D a-SnO₂ interfaces with exceptional stability. DFT reveals competitive dissociative chemisorption of H₂S and H₂O at identical sites, explaining humidity-induced H₂S signal reduction (RR = 2.4 → 1.9 at 1 ppm) and LOD increase (210 → 380 ppb), enabling selective H₂S detection amid humidity interference.

While defects provide active sites for target gas molecules-thereby improving sensitivity and response speed-they can also disrupt the material’s electronic structure, leading to increased noise, baseline drift, and reduced carrier mobility. For instance, ZnO nanorod arrays with moderate oxygen vacancies exhibit superior ethanol sensitivity due to tailored carrier behavior, whereas excessive vacancies induce unstable metal-semiconductor contacts and non-linear I-V characteristics, compromising accuracy [63]. Additionally, excessive defect generation may introduce charge carrier recombination centers, impairing carrier mobility and reducing conductivity. Thus, defect optimization—rather than maximization—is essential to harness beneficial sites while preserving the sensing matrix’s coherence and reliability.

Figure 5. (a) Ar plasma treatment of ZnO films. (b) Response recovery time of ZnO sensors to 10 ppm TEA at optimal operating temperature before and after Ar plasma treatment. (c) Responses of ZnO film sensors before and after 5 min Ar plasma treatment to 10 ppm gases. Reproduced with permission from ref. [60]. Copyright 2021 Elsevier. (d) XRD patterns of W₀ a, FW₁ b, FW₂ c, FW₃ d, FW₄ e and FW₅ f. (e) Dynamic curves of FW₃ sensor to acetone, ethanol, methanol, toluene, NH₃, CO, and NO. Reproduced with permission from ref. [64]. Copyright 2018 Elsevier.

Figure 5. (a) Ar plasma treatment of ZnO films. (b) Response recovery time of ZnO sensors to 10 ppm TEA at optimal operating temperature before and after Ar plasma treatment. (c) Responses of ZnO film sensors before and after 5 min Ar plasma treatment to 10 ppm gases. Reproduced with permission from ref. [60]. Copyright 2021 Elsevier. (d) XRD patterns of W₀ a, FW₁ b, FW₂ c, FW₃ d, FW₄ e and FW₅ f. (e) Dynamic curves of FW₃ sensor to acetone, ethanol, methanol, toluene, NH₃, CO, and NO. Reproduced with permission from ref. [64]. Copyright 2018 Elsevier.

Besides defect generation, phase control is another key strategy to realize high sensitivity and selectivity of gas sensors based on SMOs. Shen et al. [64] investigated Fe-C codoped ε-phase WO₃ for acetone sensing. The ε-phase of WO₃ is stabilized through doping with chromium (Cr) or silicon (Si). The XRD patterns in Figure 5d confirm the ε-phase of WO₃. The selectivity of the ε-phase WO₃ originates from its non-centrosymmetric crystal structure. This structure generates a spontaneous dipole moment, which attracts polar acetone molecules and promotes their oxidation. As shown in Figure 5e, the FW₃ sensor shows high sensitivity of 7.3 toward 0.9 ppm acetone as well as high selectivity.

3.2. Pure Metal and Alloy-Based Gas Sensors

The enhancement of selectivity in pure metal and alloy-based chemiresistive gas sensors primarily relies on the distinct electronic and catalytic properties of these materials upon interaction with target gases. Noble metal-based sensors, utilizing elements such as platinum and palladium, exhibit exceptional sensitivity and selectivity due to their high catalytic activity to dissociate gas molecules [65]. Meanwhile, metal alloy-based sensors benefit from synergistic effects and tailored electronic structures, enabling improved selectivity through composition optimization. This section is crucial as it highlights strategies for designing highly selective sensing interfaces without employing complex composite structures, thereby offering fundamental insights into material–gas interactions and guiding the development of efficient and stable gas sensors.

3.2.1. Noble Metal-Based Sensors

Noble metals such as platinum (Pt) [66] and palladium (Pd) [67] are sensitive to hydrogen and carbon monoxide due to their catalytic properties. However, the selectivity of pure metals is generally limited because their surface-active sites are relatively uniform and prone to cross-interference from various gases with similar chemical properties. Recently, ultra-fine metal nanowires (with diameters < 5 nm) have become a research hotspot due to their extremely high specific surface area and quantum confinement effects [68]. Ultra-fine platinum nanowires enhance gas adsorption kinetics by regulating surface states, emphasizing the importance of metal properties in achieving selectivity. Ding et al. [69] synthesized ligand-free ultra-fine Pt nanowires (~3 nm diameter, shown in Figure 6a) for H₂ sensing. The gas sensor based on platinum nanowires achieves high sensitivity with response of 261.4% toward 0.5% H₂, which increased to 768% with alkylamine modification. Their selectivity is due to enhanced surface scattering from Pt-H bonding, while thiol/amine modifications control gas diffusion and reduce humidity interference. Figure 6b illustrates the electrochemical cleaning setup, which removes impurities to achieve reversible response, solving a key issue for wet-synthesized sensors.

Research on Pd nanowires addresses the issue of oxygen interference in air. Koo et al. [70] introduced ZIF-8 metal–organic frameworks as a nanofiltration layer with 0.34 nm micropores for molecular sieving (Figure 6c). The nanofiltration layer allows H₂ with the kinetic diameter of 0.289 nm to pass while blocking O₂ with the kinetic diameter of 0.345 nm, achieving effective H₂ gas sieving, as shown in Figure 6d. This innovation improved sensor response and recovery speed by 20 times (7 s to respond, 10 s to recover at 1% H₂), despite a slight decrease from 5.9% to 3.5% in response, as shown in Figure 6e,f. The detection limit improved to 600 ppm, demonstrating the benefits of combining physical filtration and chemical sensing. Pure metal chemical resistance sensors enhance sensitivity and selectivity by controlling surface defects and phase interfaces, offering a reliable solution for hydrogen sensors in complex environments.

3.2.2. Metal Alloy-Based Sensors

Metal alloy-based sensors have been developed to overcome the limitations of pure metals and further improve the selectivity. Alloying multiple metals enables precise tuning of the electronic structure, surface energy, and catalytic properties, resulting in significant synergistic effects for gas sensors. Several studies reported that alloy/bimetallic structures (e.g., Pd-Mg [71,72], Pd-Y [73], Pd-Pt [74]) can optimize the sensitivity of chemiresistive gas sensors. By employing nanostructured designs, they address the issues of expansion, oxidation, and stability commonly found in traditional Pd-based sensors, thereby achieving highly sensitive and fast-response room-temperature hydrogen detection. For instance, Gautam et al. [71] used magnetron sputtering technology to fabricate Pd-Mg alloys and multilayer films structures (Figure 7a). The sensor based on Pd-Mg alloys achieves a rapid and reversible response in 3 s at room temperature (ΔR = 3.3% @ 2 bar H₂) (Figure 7b). This is because palladium (Pd) doping reduces the hydrogen binding energy in the magnesium lattice, thereby enhancing hydrogen desorption at the Pd/Mg interface (Figure 7c). Hassan et al. [72] investigated quantum-sized Pd/Mg bimetallic films (~6 nm) on a hydrophobic Al₂O₃ substrate for H₂ sensor. By exploiting nanoparticle-induced lattice expansion to modulate conductivity, this approach not only achieved ultrafast response and wide-range hydrogen detection (1–40,000 ppm), but also effectively suppressed humidity interference through the hydrophobic interface. As shown in Figure 7d,e, XPS and COMSOL simulations highlighted charge transfer at the Pd-Mg interface, supporting Gautam’s synergistic mechanism.

Furthermore, Wang et al. [73] proposed a Pd-Y (Pd_0.92Y_0.08) alloy for H₂ gas sensor, where Y doping expands the Pd lattice spacing, effectively preventing hydrogen-induced cracking while enhancing sensitivity (29.28% at 4% H₂). The ultra-thin structure significantly accelerates response time (83 s at 4% H₂) by shortening the hydrogen diffusion path. The nanocrystalline Pd (<10 nm) exhibits a gap effect that allows switch-type sensing. Moreover, Kim et al. [74] deposited a Pd@Pt layer on polyacrylonitrile nanofiber yarns, revealing that resistive switching behavior depends on hydrogen concentration: at high levels (>1%), electron scattering is dominant (ΔR/R₀ > 0), while at low levels (<0.5%), lattice expansion reduces the gap (ΔR/R₀ < 0). The core–shell structure also mitigates Pd volume expansion. These findings suggest that alloy-based chemiresistive sensors show great potential for enhanced selectivity and sensitivity.

3.3. Conjugated Polymer-Based Gas Sensors

The enhancement of selectivity in conjugated polymer-based gas sensors primarily revolves around strategic molecular engineering approaches, notably through doping and functionalization as well as composite structure design. Doping with oxidizing or reducing agents and introducing functional side groups can significantly modulate the electron density and binding affinity of polymer chains, thereby improving specificity toward target gases [75]. Meanwhile, compositing with inorganic nanoparticles or carbon nanomaterials further enhances selectivity via synergistic effects and interface-dominated sensing mechanisms. This section will focus on recent advances in these two strategic directions for achieving highly selective chemiresistive sensing using conjugated polymers.

3.3.1. Doping and Functionalization

Conjugated polymers (CPs) are ideal for gas sensing due to their customizable π-conjugated backbone, semiconductor properties, and structural flexibility. However, their practical applications are limited by the following factors: weak signals caused by inherently low electrical conductivity, a lack of selective recognition capability for different gases, and poor long-term stability. The strategies of doping and functionalization of conjugated polymers have been widely adopted to overcome the above issues [76,77]. Doping improves conductivity by adjusting the polymer’s energy levels and carrier concentration. For instance, Ngai et al. [78] developed a donor-acceptor polymer, PDEB, which transforms into PDNB at 200 °C, enhancing solvent resistance. PDNB’s narrow bandgap and high HOMO level allow effective p-type doping with HCl, resulting in a stable, conductive material (PDNB: HCl) with a conductivity of 0.24 S cm⁻¹. The sensor based on PDNB can identify 10 different types of volatile organic liquids at a low voltage of 1 V, with each liquid producing a unique current-time curve. Moreover, the sensor is reusable for hundreds of cycles. This study provides a high-performance material platform for reusable chemiresistive sensors.

To further address the issue of cross-sensitivity in complex environments, specific functional groups can be introduced into the polymer backbone or side chains. This enables the polymer to engage in specific interactions with gas molecules, thereby distinguishing the target gas in complex environments and enhancing the selectivity of gas sensors. Rath et al. [79] introduced diethylamine (DEA) into the CO₂-responsive copolymer P (D-co-M), where amine groups gradually protonate under CO₂, increasing ionic conductivity as pH drops (pH 4.4–7.3), as displayed in Figure 8a. The high pKa of DEA (10.64) enables the protonation of NH₃ (pKa = 9.25), significantly reducing ammonia interference and achieving CO₂/NH₃ discrimination (Figure 8b). The sensor performs wide-range CO₂ detection (10³–10⁶ ppm) in high-humidity environments (>80% RH) while substantially shortening the response time (Figure 8c). In another study, Rath et al. [80] demonstrated the effectiveness of functional group design for specific detection by developing a highly selective ammonia sensor using electrostatic interactions between polymers and NH₄⁺. Yang et al. [81] incorporated polar triethylene glycol (TEG) side chains into a heat-resistant matrix for a DPP-based organic mixed conductor-polyimide blend film for enhancing NO₂ selectivity and response. These studies demonstrate that systematic optimization of the conductivity, selectivity, and environmental tolerance of gas sensors can be achieved through backbone energy level engineering, side-chain polarity modification, and functional group regulation. This advancement promotes their application in complex scenarios such as industrial emissions and medical diagnostics.

3.3.2. Composite Structure Design

Composite structure design enhances performance by utilizing the synergistic effects of various materials through hybrid systems. This involves three main strategies: carbon-based hybridization, metal hybridization, and porous framework hybridization. For instance, combining carbon materials such as carbon nanotubes and graphene improves sensor sensitivity due to their high conductivity and abundant adsorption sites, which facilitate selective gas molecule adsorption and optimize charge transport. He et al. [81] creation of the PSATT-7NC hybrid material by combining amino-functionalized multi-walled carbon nanotubes (NH₂-MWCNTs) with conjugated microporous polymers (PSATT). This structure leverages the high electrical conductivity of NH₂-MWCNTs for electronic transport and enhances charge transfer through π-π interactions, achieving high response of 9766% toward 4 ppm NO₂, which is 2.5 times higher than that of pure PSATT and a low detection limit of 0.79 ppb at 100 °C, as well as excellent selectivity and stability.

In addition, incorporating noble metal nanoparticles or metal oxides such as gold, silver, platinum, and palladium into the polymer matrix enables highly selective and sensitive detection of specific reactive gases or biomolecules. Thangamani et al. [82] enhanced the polypyrrole/polyvinyl alcohol (PVA/WPPy) matrix with V₂O₅ nanorods to create flexible nanocomposites. These composites show strong ionic dipole interactions, enabling reversible LPG adsorption at room temperature. With 15 wt.% filler, the material achieves 1.16% response to 600 ppm LPG with a rapid response/recovery time of 10 s/8 s, and high selectivity over gases such as acetone and benzene.

The composite of conjugated polymers with metal–organic frameworks (MOFs) has been proven to improve the selectivity and sensitivity of sensors. These framework materials possess highly ordered pore structures, large surface areas, and abundant active sites, which modulate the resistance signals through interactions at the interface with the conjugated polymers. For example, Jo et al. [83] developed a composite structure using a mixed matrix membrane (MMM) for selective formaldehyde detection at room temperature. ZIF-7 nanoparticles (~139 nm) are dispersed in a PEBA polymer matrix to form an MMM coating on a TiO₂ thin film in Figure 9a. The hexagonal channels of ZIF-7 (~0.3 nm) selectivity block interfering with gases such as ethanol through molecular sieving effects, while allowing formaldehyde molecules to permeate. Combined with the photoactivation properties of TiO₂, the sensor enables specific detection of formaldehyde at concentrations as low as 25 ppb (Figure 9b). This design combines MOF/polymer membranes with photoactivated sensing, enhancing flexible device integration and addressing response intensity and mass transfer efficiency. Huang et al. [84] proposed a method for the conversion from insulation to conductive MOFs by creating 12 types of 2D c-MOFs with hierarchical nanostructures. As shown in Figure 9c, this work utilizes HKUST-1 and other 3D MbbOF as sacrificial templates, converting them into 2D conjugated MOFs with hollow nanostructures at room temperature. These materials exhibit both high crystallinity and large surface area, promoting exposure to the active site and gas diffusion, thereby making them ideal for the design of high-performance sensors.

3.4. Two-Dimensional Materials-Based Gas Sensors

This section delves into the enhanced selectivity of chemiresistive gas sensors utilizing two-dimensional materials, with a focus on transition metal dichalcogenides (TMDs) and graphene and its derivatives-based gas sensors. TMDs such as MoS₂ exhibit tunable electronic properties and layer-dependent sensitivity, which facilitate selective gas detection via metal doping, and heterojunction construction [85]. Graphene and its derivatives, including graphene oxide and reduced graphene oxide, are widely modified through chemical functionalization, elemental doping, and hybridization with other nanomaterials to promote specific gas adsorption and discrimination [86]. Key strategies including heteroatom doping, composite formation, and surface modification will be discussed to illustrate how material tailoring significantly boosts selective detection capabilities.

3.4.1. Transition Metal Dichalcogenides (TMDs)

Two-dimensional materials, with unique distinct physical and chemical properties, promise to construct high-performance chemiresistive sensors [87]. Transition metal dichalcogenides (TMDs, such as MoS₂, WS₂, and MoSe₂), with their tunable band gaps and abundant edge-active sites, are particularly suited for gas sensors. Studies have confirmed that incorporating noble metal nanoparticles (e.g., Pt [36], Ni [88], Co [89]) into TMDs can improve the sensitivity and stability of sensors at room temperature by increasing the specific surface area, providing active sites, catalyzing gas dissociation, and modulating the interfacial band structures. Bharathi team [88] conducted studies highlighting how different transition metal dopants affect MoS₂ nanostructures and enhance NO₂ sensing. They synthesized nickel (Ni)-doped MoS₂ nanosheets (with 3–7at.% Ni) via a hydrothermal method. This study revealed that Ni doping optimized material morphology, inducing the self-assembly of nanosheets into porous nanoflower structures and exposing more edge-active sites [88]. Among them, the sensor based on 7at.% Ni-MoS₂ showed a response of 45.25% to 200 ppm NO₂ and demonstrated long-term stability of 80.4% over 30 days. In addition, they further investigated the effect of cobalt (Co) doping and confirmed that 7 at.% cobalt doping in MoS₂ yields optimal NO₂ sensing performance due to maximized edge-active sites and vertically aligned nanosheet morphology, enhancing gas adsorption and charge transfer. Below 7 at.% (e.g., 3–5 at.%), insufficient defect sites limit response enhancement. The XPS measurements confirmed that 7 at.% cobalt doping in MoS₂ induces a Fermi level shift, which also enhances conductivity. However, excessive cobalt doping exceeding 7 at.% disrupt the MoS₂ lattice structure, compromising charge carrier mobility. Thus, an appropriate amount of 7 at.% cobalt doping in MoS₂ not only maintains the structural integrity of the gas-sensitive material but also maximizes the number of active sites, thereby significantly enhancing the gas-sensing performance [89].

In contrast, Tian et al. [90] developed a Pt NPs-modified MoS₂/polyaniline nanocomposite (Pt/MoS₂/PANI) for room-temperature NH₃ detection, as shown in Figure 10a. Figure 10b demonstrates the superior selectivity of PANI, MP-2, and Pt/MP-2 sensors toward NH₃ over other VOC gases (including TMA and TEA) at 50 ppm, while Figure 10c reveals through DFT calculations that despite TMA exhibiting higher adsorption energy, NH₃ facilitates the largest charge transfer (ΔQ) to both PANI and MoS₂ (002)/PANI models, explaining the experimental selectivity. Pt/MoS₂/PANI efficiently detects NH₃ through p-n heterojunctions, Schottky barriers, and Pt’s catalytic effect, with a fast response/recovery time of 15 s/103 s (Figure 10d). Li et al. [91] fabricated Pt-modified MoSe₂ for room-temperature NO₂ detection. This study combines experiments and theory to explain the gas-sensing mechanism of metal-modified TMDs, where target gas adsorption alters carrier concentration and mobility, enhancing sensor selectivity through metal oxide composite heterostructures. Niu et al. [92] developed a groundbreaking photovoltaic self-powered gas sensor achieving outstanding NO₂ detection at room temperature with a limit detection of 20 ppb and rapid response/recovery times of 23 s/178 s. As shown in Figure 10e–g, KPFM and Raman analysis showed that NO₂ reduces the short-circuit current by capturing MoS₂ conduction band electrons, causing Fermi level shifts. The vertically stacked MoS₂/GaSe heterojunction was fabricated with all-dry transfer method as illustrated in Figure 10h. This study introduces GaSe into heterojunction sensing for the first time, providing a new paradigm for ultra-low power ppb-level detection.

3.4.2. Graphene and Its Derivatives-Based Gas Sensors

Graphene, a single layer of carbon atoms in a two-dimensional honeycomb structure, is promising for sensing applications due to its large surface area and excellent conductivity [36,93,94]. However, graphene-based gas sensors respond to all gases and cannot achieve selective gas detection. Therefore, to enhance the sensing performance of graphene, its selectivity can be greatly improved through defect engineering, graphene oxide (GO) [66], reduced graphene oxide (rGO) [95], and functionalization can enhance selectivity. Eigler and Feicht et al. [96] discussed strategies for regulating defects in GO, suggesting that vacancy defects can serve as active sites for gas adsorption, while hole defects can regulate the molecular sieving effect. This research provides a theoretical basis for developing sensitive graphene-based gas sensors.

Furthermore, the practical application of graphene-based materials is often limited by issues such as poor interfacial contact and uneven thickness caused by insufficient precision in the assembly of nanostructures. Wang et al. [97] addressed this by using alternating current dielectrophoresis (DEP) with optimized parameters (500 kHz frequency, 10 V voltage, 30 s duration) for precise assembly of graphene oxide (GO) nanostructures. This method leverages the GO’s oxygen-containing groups for improved hydrogen gas selectivity and response. Additionally, atomic-level interface engineering can enhance the sensitivity of sensing materials. Yang et al. [98] developed a room-temperature DMMP sensor using polypyrrole-reduced graphene oxide (PPy-rGO) with isolated Cu-N₅ sites in Figure 11a. These sites enhance hydrogen bonding and use unsaturated Cu⁺ as Lewis acid sites to adsorb DMMP’s oxygen atoms (Figure 11b,c). Therefore, the sensor’s response increased by 4.5 times, the detection limit decreased by 2.5 times, and the response/recovery times were 49 s/106 s.

Additionally, rGO-encapsulated photosensitive heterojunctions can enhance sensor selectivity by promoting charge separation. The Shafiei team [99] developed a composite system of rGO-encapsulated Pd quantum dots on TiO₂ nanospheres (rGPT-NS)forming multiple heterojunctions that enhance hydrogen selectivity (Figure 11d,e). As quantified in Figure 11f, it exhibits 100% response to 50 ppm H₂ while exhibiting minimal cross-sensitivity to competing gases (<1% for 50 ppm CH₄/acetone), maintaining 99% ± 0.09 response under 65% humidity. Critically, Figure 11g demonstrates < 3% signal attenuation over 8 months toward 1000 ppm H₂, confirming long-term selectivity. The superior selectivity is attributed to the synergistic effects of Pd’s catalytic specificity for hydrogen dissociation, rGO’s selective gas adsorption properties, and visible-light-activated surface reactions that preferentially modulate hydrogen interactions over competing gases.

More importantly, the surface of graphene allows for the covalent grafting of MOFs. Through size matching, electrostatic interactions, or specific chemical reactions, it enables highly selective recognition of target molecules. Jayaramulu et al. [100] developed GA@UiO-66-NH₂ hybrids by covalently bonding graphene acid’s carboxyl groups with MOFs’ amino groups, ensuring stable chemical coupling. Graphene acts as both a charge transport network and enhances gas diffusion through its microporous-mesoporous structure. The sensor, leveraging amide bond and CO₂ interactions, achieves a 10% response at 200 °C with response time of 18 s. In situ Raman spectroscopy first confirmed the reversible chemical adsorption mechanism.

Therefore, the key to enhancing the selectivity of two-dimensional material-based chemiresistive sensors lies in the precise regulation of the intrinsic electronic structure and surface chemical properties of the materials, as well as in the design and optimization of specific interactions between the sensing interface and the analytes. This approach is fundamental for advancing next-generation gas and chemiresistive sensors with excellent selectivity.

As displayed in

Table 1. Recently reports of different modification methods on the gas-sensing performance of chemiresistive gas sensors.

Modification Method	Materials	Target Gas	Temperature (°C)	Interferents	Response/Conc (ppm)	Ref.
Noble Metal Loading	Pt/ZnO	H2S	310	VOCs and NH3	118 a/5.5	[101]
Noble Metal Loading	Pd/SnO2	Benzene	RT	benzene, m-xylene, toluene, acetone, acetaldehyde, isoprene, methanol, ethanol, CO, hdrogen, and ethylbenzene	2.1 a/1	[102]
Nanostructure Engineering	Au/WO3	Isoprene	275	not investigated	11.3 a/0.1	[103]
Nanostructure Engineering	Ag/SnO2	TEA	170	not investigated	1700 a/100	[104]
Bimetallic Core–Shell Loading	PdAu/W18O49	H2S	100	CH4	55.5 a/50	[105]
Bimetallic Noble Metal Loading	AgPd/In2O3	Toluene	180	Methanol, ammonia, acetone, ethanol and formaldehyde	15.9 a/1	[106]
Metal Oxide Loading	CuOx/Co3O4	formaldehyde	75	acetone, toluene, ethanol, NO, acetaldehyde, NH3, CO, CH4	5.2% c/1	[107]
Noble Metal Decoration	Pd hemitubes	H2	RT	NH3, CO, NO2, and H2S	2.1% e/1	[108]
Pd-Doping	WO3 nanomaterials	Acetone	175	Acetone, Ethanol, Ammonium hydroxide, Methyl cyanide, Dichloromethane, Trichloromethane, Ether, Acetic acid and Hexane	150% a/50	[109]
Photoelectronic Sensing	MoS2 with led light	NO2	RT	not investigated	12% e/200	[110]
Nanostructure Engineering	3D epitaxial graphene nanowalls	O2	RT	N2 and H2	77% d/500	[111]
Heterojunction Construction	5CeO2/Co3O4	CO	200	not investigated	184% b/50	[112]
Heterojunction Construction	Co3O4/ZnO	CO	210	H2, CH4 and NH3	38% c/10	[113]

^a: R_g/R_a. ^b: (R_g − R_a)/R_a. ^c: R_a/R_g × 100%. ^d: ∆I/I₀. ^e: (∆R/R₀) × 100%.

, we summarize the effects of different modification methods on the selectivity and gas-sensing performance of chemiresistive gas sensors as reported in the literature.

4. Novel Selective and System Integration Strategies

4.1. Bionics and Heterogeneous Structures

Among the strategies for enhancing the selectivity of chemiresistive gas sensors, inspired by the biological olfactory system, researchers have sought to improve the recognition of specific target molecules. This is achieved by mimicking its highly specific receptor arrays and signal processing mechanisms, leveraging energy band modulation between different materials, interfacial charge transfer, and synergistic sensitization effects. Chun et al. [114] proposed a chemiresistive gas sensor based on oxygen vacancy dynamics (Pt/TiO₂ nanorod/TiN structure), overcoming the limitations of traditional sensing mechanisms. This device uses redox reactions for quick response and recovery to reductive gases H₂ at room temperature and features an intrinsic memory effect, allowing gas pulse counting and risk grading without external storage. By alternately introducing reducing/oxidizing gases (H₂/NO), the device’s conductivity can be linearly modulated (linearity factor 1.02/1.31), mimicking synaptic enhancement/inhibition and achieving 92.76% accuracy in MNIST pattern recognition. This approach paves the way for bio-inspired olfactory systems that integrate sensing and computing.

On this basis, Wang et al. [115] further developed a single-chip bionic olfactory system (BOC) to overcome sensor diversity limitations through large-scale monolithic integration. This work employs vertically aligned PdO/SnO₂ nanotube arrays supported by porous alumina templates and combines atomic layer deposition (ALD) and mask-assisted sputtering techniques to construct a two-dimensional multi-metal oxide interface layer (ZnO/NiO/In₂O₃/WO₃) on the sensing layer. This enables a single chip to accommodate 10,000 independently addressable sensor arrays in Figure 12a. Through elemental gradient design (Figure 12b), the sensor array achieves spatial variation in material composition, mimicking the diversity of biological olfactory receptors. Using a convolutional neural network, the chip accurately classifies eight gases (99.04%), analyzes mixed gases (maximum 8.12% error), and distinguishes 24 complex odors (Figure 12c). It also integrates with a robot dog’s visual sensors for blind object recognition. These studies highlight the advantages of biomimetic strategies, advancing the development of chemiresistive sensors toward high discriminative and low-power integrated systems.

Additionally, creating material heterojunctions is an effective way to improve sensor selectivity for specific gases [116,117]. Yao et al. [118] developed a Ti₃C₂T_x/ZnO p-n heterojunction using a hydrothermal method to enhance room-temperature ammonia detection. By combining p-type MXene (Ti₃C₂T_x) with n-type ZnO at a 3:1 mass ratio, they achieved a sensor response of 196% for 50 ppm NH₃ at 28 °C, higher than that of Ti₃C₂T_x-based gas sensors (~2%) (Figure 12d). This is the first successful use of MXene/metal oxide heterojunctions for selective ammonia sensing, offering a new design approach for low-power sensing, as shown in Figure 12e,f.

Figure 12. (a) Structure of a monolithic BOC with the correlated circuit. (b) Spatial distribution of elements in the MCI layer visualized using the loci of ToF-SIMS depth profiles. (c) Confusion matrix of the actual class and predicted class when recognizing eight gases (A, acetone; C, carbon monoxide; E, ethanol; F, formaldehyde; N, nitrogen dioxide; T, toluene; H, hydrogen; I, isobutylene). Reproduced with permission from ref. [115]. Copyright 2024 Springer Nature. (d) Sensing mechanism of Ti₃C₂T_x/ZnO (3:1) sensor. (e) Comparison of Ti₃C₂T_x MXene, ZnO, and Ti₃C₂T_x/ZnO sensors’ response to ammonia, triethylamine, trimethylamine gases with the concentration of 50 ppm, and to ethanol, toluene, acetone, methanol, n-butanol, and isopropanol gases with the concentration of 500 ppm at a low operating temperature (28 °C). (f) the line chart of gas concentration to the response and the inset is an index-related curve of Ti₃C₂T_x/ZnO (3:1) about the correlation of gas concentration and response value. Reproduced with permission from ref. [118]. Copyright 2023 AIP Publishing.

Figure 12. (a) Structure of a monolithic BOC with the correlated circuit. (b) Spatial distribution of elements in the MCI layer visualized using the loci of ToF-SIMS depth profiles. (c) Confusion matrix of the actual class and predicted class when recognizing eight gases (A, acetone; C, carbon monoxide; E, ethanol; F, formaldehyde; N, nitrogen dioxide; T, toluene; H, hydrogen; I, isobutylene). Reproduced with permission from ref. [115]. Copyright 2024 Springer Nature. (d) Sensing mechanism of Ti₃C₂T_x/ZnO (3:1) sensor. (e) Comparison of Ti₃C₂T_x MXene, ZnO, and Ti₃C₂T_x/ZnO sensors’ response to ammonia, triethylamine, trimethylamine gases with the concentration of 50 ppm, and to ethanol, toluene, acetone, methanol, n-butanol, and isopropanol gases with the concentration of 500 ppm at a low operating temperature (28 °C). (f) the line chart of gas concentration to the response and the inset is an index-related curve of Ti₃C₂T_x/ZnO (3:1) about the correlation of gas concentration and response value. Reproduced with permission from ref. [118]. Copyright 2023 AIP Publishing.

4.2. Artificial Intelligence-Assisted Sensing

Artificial intelligence (AI) is revolutionizing sensor signal analysis and decision-making by using machine learning and deep learning to process sensor data and tackle cross-sensitivity. Techniques like linear discriminant analysis (LDA) [119], partial least squares regression (PLSR) [120], support vector machines (SVM) [121], principal component analysis (PCA) [122], parallel neural networks (PNN) [123], and convolutional neural networks (CNN) [124] are applied to chemiresistive sensor arrays for gas signal processing. This AI-sensor integration offers innovative approaches to enhance the selectivity of sensors. Chu et al. [125] developed a sensor array with four metal oxide semiconductor sensors and demonstrated that a GA-optimized BPNN algorithm significantly improved quantitative gas identification. Figure 13a,d show the normalized confusion matrices of BPNN and GA+BPNN, achieving 100% classification accuracy in the quantitative detection of CO and NO₂ mixed gases, but the mean squared error (MSE) of BPNN was 3.085 (Figure 13b,c). In contrast, the MSE of the GA+BPNN optimized by the genetic algorithm was reduced to 1.7207, as shown in Figure 13e,f. This optimization by the genetic algorithm enhanced the initial weights and bias parameters of the BPNN, improving detection accuracy. The study integrates multi-sensor data fusion with deep learning to create a framework for selective gas detection in complex environments.

The AI model effectively learns gas “fingerprints”, ensuring stable identification despite sensor drift or environmental changes. Pan et al. [126] developed the HATN-DA algorithm to tackle drift and batch variation in mixed gas detection. This model combines multi-head Transformers for temporal dependencies, channel attention for sensor optimization, and capsule networks to output gas types and concentrations. By introducing domain adversarial learning based on Wasserstein distance, it compensates for drift without target domain labels, achieving 98.79% transfer accuracy on public datasets and reducing ethanol/methane/ethylene prediction errors to 3.32%/3.25%/3.01%. This approach enhances system robustness, overcoming traditional gas sensor limitations and providing dynamic adaptability in complex environments.

4.3. Multimodal Integrated System

To address the limitations of single-sensor systems, multimodal integrated systems combining various sensor types and technologies have become essential for improving selectivity and reliability. Li et al. [127] combined data from an electronic nose, an electronic tongue, and a colorimeter with machine learning algorithms to predict the freshness of frozen mackerel. Using PAC, random forest regression, XGBoost, and artificial neural networks, the multi-sensor fusion significantly enhanced the accuracy of predicting lipid and protein oxidation, proving the effectiveness of the “data fusion + machine learning” approach. In addition, the system can significantly improve the selectivity and quantitative analysis capabilities of gas detection by integrating complementary sensing technologies. Yeon Sik Jung’s team developed a multimodal gas sensor using 3D cross-multifunctional nanostructures (3D-CMA) [128]. This sensor combines chemiresistive sensing and surface-enhanced Raman spectroscopy (SERS) to selectively and quantitatively detect gases with similar molecular structures in Figure 14a, such as nitrobenzene and toluene. The structure is constructed via solvent-assisted nano-transfer printing, forming quantum contacts at SnO₂ nanowire intersections to boost electrical sensitivity and using nano-gaps with Au NP modification as SERS hotspots (Figure 14b). This enables the sensor to provide both electrical signals and SERS fingerprint spectra for quantitative and qualitative analysis, with detection limits of <50 ppb and <5 ppm, respectively, as shown in Figure 14c,d. This design overcomes the bottlenecks of traditional sensors in selectivity, sensitivity, and mixed gas analysis, providing a new platform for environmental monitoring and biomedical diagnostics.

Furthermore, Burgués et al. [129] proposed the multimodal system for dynamic spatial monitoring by developing SNIFFDRONE, a drone-mounted chemical sensor system with 21 gas sensors and a 10 m sampling tube to minimize rotor wash interference. This system allows real-time odor concentration measurement at wastewater treatment plants. They implemented a dynamic calibration strategy, collecting transient sensor signals and odor samples according to EN13725 standards during drone hovering. Using PLSA, they created a model linking sensor responses to actual odor concentrations measured by dynamic olfactometry. This approach reduces prediction errors by 33%, increases the correlation coefficient with olfactometry to 0.86, and establishes 95% consistency limits at [0.25×, 3.91×]. The system was the first to spatially map odor concentrations in WWTPs, identifying major hotspots through 225 aerial survey points in a 200 × 100 m² area, demonstrating the practical value of multimodal sensing in complex environments. The multimodal integrated system evolved from enhancing classification and accuracy in the lab through multi-sensor fusion to achieving spatial monitoring outdoors with a drone-mounted setup. A dynamic calibration strategy has effectively addressed the mismatch between transient field signals and steady-state laboratory data.

5. Conclusions and Future Perspective

In this review, we focus on the strategies reported in the literature for enhancing the selectivity of various gas-sensitive chemiresistive materials, addressing a persistent challenge through material innovation and system engineering.

Significant advancements have been made in improving the selectivity of chemiresistive gas sensors through various material modification strategies. Key approaches include (i) nanostructuring and morphology control of semiconductor metal oxides (SMOs) to increase surface area and active sites; (ii) functionalization with noble metal catalysts (e.g., Pd, Pt) or dopants to enhance specific gas-catalyst interactions and promote catalytic reactions; (iii) defect engineering (e.g., oxygen vacancy manipulation) to tailor surface chemistry and charge transfer processes; (iv) construction of heterojunctions (e.g., p-n junctions, Schottky barriers) to modulate electron transport and energy barriers; and (v) development of hybrid materials such as phosphorene, conductive MOFs, COFs, and composites like rGO-Ru OEP, which offer tunable electronic properties and selective gas permeation.

Additionally, surface modification with organic ligands and the integration of molecular sieving or filter layers have shown promise in enhancing specificity. Despite the aforementioned enhancements, numerous challenges continue to exist. Many strategies suffer from poor reproducibility due to difficulties in controlling catalyst size/distribution, sensitivity to environmental conditions (e.g., humidity), and limited long-term stability. The selectivity mechanisms are often not fully understood at the atomic level, hindering the development of universal design principles. Future efforts must focus on combining advanced in situ characterization techniques (e.g., operando XRD, XPS) and computational modeling (DFT) to decouple the dynamics of gas adsorption, surface reactions, and charge transfer. This will help establish clearer structure–activity relationships. Moreover, for practical applications, a systems-level approach is essential. This includes integrating selective materials with sensor arrays, microfluidic pre-concentration or separation columns, and artificial intelligence (AI)/machine learning algorithms for pattern recognition to handle complex gas mixtures and mitigate cross-sensitivity. Addressing issues like power consumption, manufacturing scalability, and robustness in real-world conditions (e.g., variable humidity/temperature) is crucial for reliable deployment in areas such as medical diagnostics, environmental monitoring, and industrial safety. Ultimately, the goal is to develop adaptive, multifunctional sensing platforms that combine material innovation with intelligent systems for high-fidelity and reliable gas discrimination.

In future research, it is imperative to prioritize the development of unified evaluation frameworks that rigorously assess sensor performance under practical conditions, including encompassing complex gas mixtures and defined humidity ranges. Material innovation should focus on creating multifunctional systems that integrate self-healing biomimetic coatings with stimuli-responsive layers to mitigate drift and environmental interference. Advanced manufacturing techniques, such as spark ablation printing (VSP-P1), could facilitate the high-throughput fabrication of precisely engineered multi-component sensing films, thereby accelerating the discovery of optimal materials. At the system architecture level, incorporating AI-driven dynamic baseline correction within miniaturized “lab-on-a-chip” platforms which include micro-preconcentrators, chromatographic columns, and optimized sensor arrays-holds the potential to enhance specificity while reducing power consumption. Additionally, biomimetic designs that mimic the spatial patterning of olfactory receptors may further refine array topology for complex odor discrimination. In conclusion, the integration of these material, structural, and systemic strategies will facilitate advancements in noninvasive medical diagnostics such as breath-based disease detection and real-time environmental monitoring, bridging the critical selectivity-reliability gap for ubiquitous sensor deployment.