RBFNN-Based PPy/GaN sensor array with wide dynamic range and sub-ppb detection for accurate ammonia identification in human exhaled breath

Abstract

Detecting ammonia (NH₃) in human breath is vitally important for early warning and disease detection. Polymer composite nanomaterials, which serve as high-performance NH₃ gas sensors, have become a prominent research focus in this field. In this study, polypyrrole (PPy) and various gallium nitride (GaN) nanostructures were selected for synthesis. PPy/GaN gas sensors were successfully fabricated by employing the precisely controllable metal-organic chemical vapor deposition (MOCVD) process in conjunction with in situ oxidative polymerization methods. The gas sensing performance of the sensors was systematically analyzed. The PPy/GaN-1 sensor demonstrated an ultra-wide detection range for NH₃ (100 ppb–1000 ppm) at room temperature, along with exceptional moisture resistance and long-term stability. This can be attributed to the excellent uniformity of the film distribution, which enabled optimal synergy between GaN and PPy. Furthermore, an experiment detecting human exhaled breath was carried out using a sensor array to validate its high sensitivity. With the assistance of machine learning algorithms, high-precision prediction of gases within the low-concentration range was achieved (with an error of 1.17 ppm). Overall, this study offers valuable insights into the development of early warning systems for chronic kidney disease (CKD).

Introduction

Ammonia (NH₃), a colorless irritant gas¹, is widely utilized in fertilizers, refrigerants, and biofuels², yet its high toxicity demands precise monitoring³. The US Occupational Safety and Health Administration limits exposure to 25 ppm (8 h) and 35 ppm (15 min)⁴. Concentrations as low as 50 ppm cause respiratory, dermal, and ocular irritation^5,6. while 5000 ppm induces fatal laryngospasm and pulmonary edema^7,8. Furthermore, as an endogenous nitrogen metabolite, NH₃ serves as a biomarker for hepatorenal pathologies. Chronic kidney disease (CKD), which affects approximately 7–12% of the global population and ranks as the sixth-leading cause of mortality worldwide⁹, is associated with elevated exhaled NH₃ levels (>2 ppm). These levels exceed physiological concentrations (0.05–2 ppm) and demonstrate a strong correlation with blood urea nitrogen (BUN) (R² = 0.95)¹⁰. Given the established role of exhaled NH₃ as a biomarker for hepatorenal pathologies, the creation of ultra-responsive NH₃ sensors emerges as a crucial advancement in non-invasive health monitoring systems.

Numerous materials have been investigated as sensing components for NH₃ detectors to date, including single-walled carbon nanotubes (SWCNTs)¹¹, silicon nanowires (SiNWs)⁴, MoS₂¹², Ce₂Sn₂O₇¹³, and metal-oxide semiconductors such as SnO₂¹⁴, ZnO¹⁵, WO₃¹⁶, etc. Among them, metal oxides exhibit high sensitivity by detecting surface resistance changes induced by NH₃. However, their operational requirement of high temperatures (200–400 °C)¹⁷ limits applicability in human exhaled breath monitoring.

Advanced materials research increasingly focuses on intrinsically conductive polymers characterized by conjugated double-bond structures in recent years¹⁸. For instance, Duan et al. successfully prepared PANI/HNTs composite ammonium-sensitive materials by in situ polymerization and improved the NH₃ sensing performance of polyaniline-based sensors through morphological modification¹⁹. Luo et al. synthesized polyaniline/polyacrylonitrile composites by using a combined electrospinning, pre-oxidation and in situ self-assembly technique, which exhibited excellent NH₃ sensing performance within the range of 2–50 ppm at room temperature²⁰. Qin et al. prepared polydopamine-coated cobalt hydroxide oxide (CoOOH@PDA) composite nanomaterials and adopted a dual-channel three-signal design to solve the detection bottleneck of portable sensors in strong alkaline environments²¹. Among those conducting polymers, polypyrrole (PPy) is especially promising for commercial applications because of its good environmental stability, facile synthesis and higher conductivity than many other conducting polymers²². However, PPy’s cross-sensitivity to multiple gases limits its selectivity in sensing systems²³, prompting the development of room temperature PPy-based composites for enhanced NH₃ detection. Sun et al. deposited rGO-PPy hybrids on PET films to fabricate flexible NH₃ sensors, and the sensitivity to 10 ppm NH₃ was 2.5 times higher than that of pure PPy sensors²⁴. Zhou et al. demonstrated that the response of PPy/TiO₂ decorated SiOC honeycomb to 50 ppm NH₃ was 18.67%, the response and recovery times were 119 and 86 s respectively, and it had a detection range of 10-500 ppm²⁵. A. J. Heiner et al. reported PPy-SnO₂ nanocomposites exhibiting 55% response to 1 ppm NH₃ and 85.33% reproducibility over 50 days²⁶. However, to enable the practical application of sensors in the detection of human exhaled breath, further investigation into their sensitivity towards NH₃ at the ppb concentration level is essential.

Experimental studies have shown that the combination of controllable nanostructured materials and PPy can achieve an optimal balance between high selectivity and high sensitivity²⁷. Among them, Qin et al. successfully achieved uniform distribution of PPy nanoparticles on the entire surface of 3D-rGO nanowalls (pore walls) through in situ polymerization, thereby fabricating PPy/3D-RGO biomimetic nanocomposites with a three-dimensional macroporous architecture. The response of the PPy/3D-RGO sensor to 1–5 ppm NH₃ was approximately 4–5 times higher than that of pure PPy, which proved that the microstructure of the PPy/3D-rGO nanocomposite is extremely conducive to gas adsorption and rapid diffusion, thereby enhancing the gas sensing performance of the sensor²⁸. Sahebrao B. Pagar et al. synthesized ZTOAg@PPy nanocomposites with varying weight percentages. The 10% ZTO-Ag@PPy composite exhibited the highest selectivity toward H₂S gas, which can be attributed to the strong interaction between the nanoparticles and the polymer chains. This interaction facilitated the effective incorporation of ZTO-Ag nanoparticles into the macromolecular structure, thereby enhancing the gas sensitivity performance²⁹. Gallium nitride (GaN) nanomaterials represent advanced third-generation semiconductors, characterized by a wide bandgap (3.4 eV), high electron mobility, superior thermochemical stability, high-temperature tolerance, and radiation resistance. Notably, due to the presence of nitrogen vacancies (VN), unintentionally doped GaN is an N-type semiconductor, and their structure and composition can be precisely controlled by metal-organic chemical vapor deposition (MOCVD) methods. Recently, it has emerged as a promising gas-sensing material for trace detection of various types of gases by designing different structures and surface modification. For instance, Li et al. developed a H₂ sensor exhibiting high sensitivity and low power consumption by uniformly depositing Pd NPs onto the surface of GaN NWs using MOCVD combined with Pd NP sputtering. The sensor demonstrated a response of 15.2% upon exposure to 500 ppm H₂ at room temperature, with a detection limit (LOD) of 42.2 ppb. Furthermore, the sensor exhibits excellent repeatability, relatively high resistance to moisture, and long-term stability³⁰. However, the gas-sensitive performance of NH₃ sensors based on GaN nanostructures has rarely been reported, and the detection limit remains relatively high³¹.

In this study, PPy/GaN sensors were fabricated through heterojunction synergy, leveraging the tunable electronic properties of PPy and the exceptional physicochemical characteristics of GaN nanostructures, enabled by precisely controlled metal-organic chemical vapor deposition and in situ oxidative polymerization techniques. PPy films were uniformly grown on GaN nanostructures through in situ oxidative polymerization. GaN not only serves as the substrate for PPy but also facilitates the construction of heterostructures, leading to a significant enhancement in sensor sensitivity. The prepared PPy/GaN-1 sensor has excellent selectivity, moisture resistance, stability for NH₃ at room temperature, and an ultra-low detection limit of 100 ppb. GaN materials synthesized via precisely controlled MOCVD technology exhibit high stability, excellent reliability, and consistent batch-to-batch performance. The in situ oxidative polymerization method used for PPy synthesis offers advantages such as a simple fabrication process, mild reaction conditions, low manufacturing cost, and uniform film deposition. Based on these characteristics, the prepared sensors are expected to be suitable for large-scale production. In addition, the prepared sensor array was used to conduct exhaled gas detection experiments on simulated CKD patients. Meanwhile, combined with machine learning algorithms, high-precision prediction of gases within the low-concentration range has been achieved (with an error of 1.17 ppm).

Experimental section

Preparation of GaN nanostructures

GaN nanostructures were epitaxial grown on sapphire substrate by MOCVD. This method has high growth rate and good uniformity, which can accurately control the structure of the material and is suitable for industrial production.

In this work, distinct GaN nanostructures with varying thicknesses were controlleded by the growth time and the growth rate regulated through temperature, pressure and precursor flow rates. Firstly, the substrate was placed in the MOCVD reaction chamber and heated to 1100 °C under a flow of high-purity H₂ (99.999%) for 30 min to remove oxides and decompose organic residues. After cooling to 550 °C, the substrate was subjected to nitridation on its surface under an NH₃ (99.999999%) atmosphere, thereby achieving the desired pretreatment effect. Trimethylaluminum (TMAl, 410 sccm) was then introduced, and H₂ and NH₃ were simultaneously injected at 1100 °C under a pressure of 75 Torr for 3 min, resulting in the formation of the AlN buffer layer. Subsequently, trimethylgallium (TMGa, 80 sccm) was injected as a gallium source, and the flow rate of N₂, H₂ and NH₃ was adjusted to grow at 200 Torr pressure and 970 °C for 50 min to obtain GaN transition layer. Finally, silane (SiH₄, 2.57 sccm) was introduced as the n-type doping source, and triethylgallium (TEG, 666 sccm) was added to grow n-GaN active layer at 200 Torr pressure 900 °C for 135 s. The sample prepared by the above process was named GaN-1. By changing the above process parameters, the growth time of AlN buffer layer was reduced to 2 min while other conditions remained unchanged during the growth process. During the growth process of n-GaN active layer, the growth temperature was increased to 1010 °C and TMGa (80 sccm) was added, and the growth time was reduced to 95 s. The sample prepared by this process was named GaN-2. In the growth process of GaN-2 sample, keep the growth conditions of the AlN buffer layer and the GaN transition layer unchanged. Subsequently, TMGa (80 sccm) and TMAl (152 sccm) were simultaneously introduced and grown at a pressure of 75 Torr and 1010 °C for 95 seconds to obtain the AlGaN active layer. The sample prepared through this process was named GaN-3. The epitaxial growth structures of the three groups of samples are shown in Fig. 1.

Fig. 1. MOCVD epitaxial growth structure diagrams.

a GaN-1 b GaN-2 c GaN-3

Synthesis

GaN-1, GaN-2, and GaN-3 substrates were sectioned into 3 mm × 5 mm rectangular specimens and ultrasonically cleaned. The samples were subsequently immersed in a solution containing 0.02 mL pyrrole monomer and 20 mL HCl (0.2 mol/L). The mixture was immediately transferred to a 0–5 °C ice bath under continuous stirring for 30 min to ensure uniform monomer dispersion. Upon achieving homogeneous distribution, 20 mL ammonium persulfate solution (0.05 mol/L) was added dropwise, and the reaction proceeded under sustained ice-bath conditions for 4 hours. The samples were then retrieved from the reaction medium, rinsed with deionized water, and vacuum-dried at 50 °C. Finally, Ti (2 nm)/Au (50 nm) bilayer electrodes were deposited on both terminals of each sample using a 3 mm × 3 mm × 0.1 mm shadow mask via magnetron sputtering. The magnetron sputtering power was 35 W, and the cavity temperature was maintained between 30 and 40 °C. The resulting sensors were designated as PPy/GaN-1, PPy/GaN-2, and PPy/GaN-3.

Characterization techniques

Fabricated powder samples and composite films underwent comprehensive characterization using diverse analytical methods. The specific characterization equipment is shown as Supporting note 1.

The developed sensor devices were evaluated for gas detection performance using a CGS-MT intelligent gas analysis platform under controlled environmental conditions of 25 °C room temperature (RT) and 30% relative humidity (RH). During the testing process, the manufactured sensors were placed on the platform and connected to the test circuit through probes. Upon stabilization of the baseline electrical resistance (Ra), standard NH₃ gas was injected into a 5 L volume gas chamber using a syringe. The dynamic resistance during gas exposure was represented by Rg. The calculation method for the response was outlined as follows:

$$\mathrm{Response}\left(\%\right)=\left(\left|\mathrm{Rg}-\mathrm{Ra}\right|/\left(\mathrm{Ra}\right)\right)\times 100$$ 1

Response duration (T₁) quantifies the period for 90% resistance variation post gas adsorption, while recovery period (T₂) measures the equivalent restoration duration after desorption.

In the selective test, NH₃, CH₄ and CO₂ are dehumidifying standard gases purchased from Taineng Gas Co., Ltd. (Taiyuan, China). Trimethylamine (TMA), ethanol, acetone and n-butanol are injected into the gas chamber in liquid form through the evaporator (90 °C), and the evaporated gas generates gas of corresponding concentration in the sealed gas chamber. The calculation formula for the volume of solution required for a specific concentration of gas is as follows:

$$V_x=\left(V\times C\times M\right)/\left(22.4\times d\times p\right)\times {10}^{-9}$$ 2

Among them, $V_x$ represents the volume of the solution (mL), $V$ represents the volume of the sealed chamber (mL), $C$ represents the volume concentration of the gas (ppm), $M$ represents the molecular weight of the liquid (g), $d$ represents the density of the solution (g/cm^-3), and $p$ represents the volume fraction of the solution.

In the process of diluting gas concentration using the static gas distribution system, a stepwise dilution approach was employed. Taking the configuration of 200 ppb concentration NH₃ as an example, the experiment used NH₃ (20%) as the target gas and high-purity N₂ (99.999%) as the dilution gas. A 5 mL syringe was used to draw 0.5 mL of NH₃ (20%) and 4.5 mL of N₂. After thorough mixing of the gases within the syringe, approximately 4.5 mL of the mixture was gradually expelled, leaving 0.5 mL remaining. Subsequently, an additional 4.5 mL of N₂ was drawn into the syringe. Following another round of uniform mixing, the 0.5 ml final gas mixture was injected into a 5 L gas chamber to achieve a NH₃ concentration of 200 ppb.

Results and discussion

Characterization

Firstly, the morphologies and elemental compositions of PPy/GaN-1, PPy/GaN-2, PPy/ GaN-3 and three groups of GaN nanostructure samples were analyzed by SEM and EDS. The results are shown in Fig. 2. Figure 2a shows the morphology of GaN-1 nanostructure, from which GaN blocky-shaped granules with a diameter of about 500 nm can be seen uniformly growing on the surface. Figure 2b shows the SEM image of PPy/GaN-1 film obtained after composite at magnification, from which it can be seen that PPy nanoparticles are uniformly dispersed and attached to GaN surfaces, making good nanocontact with GaN. The distribution of Ga and C elements in Fig. 2c, d further indicates the successful construction of GaN and PPy heterojunction. The morphology of GaN-2 nanostructure is shown in Fig. 2e, different from the uniformly distributed particles on GaN-1, GaN-2 grows a layer of GaN columnar granules with a diameter of about 1 μm suspended on the upper surface of massive GaN. Figure 2f–h shows the morphology of PPy/GaN-2 film and the distribution of Ga and C elements on the surface of the film. It can be observed from Fig. 2f that PPy nanoparticles are polymerized in situ on the surface, however, some columnar GaN particles are not entirely covered by the PPy film. In conjunction with the elemental distribution depicted in Fig. 2g, h, it can be concluded that the uniformity of PPy nanoparticle distribution in PPy/GaN-2 is inferior to that in PPy/GaN-1. As shown in Fig. 2i, the GaN-3 nanostructure exhibits micro-hexagonal pits on its flat surface, with sizes uniformly ranging from 200 to 500 nm. The distribution density is lower compared to that of GaN-1 and GaN-2. The morphology and elemental distribution of the PPy/GaN-3 film are illustrated in Fig. 2j–l. As shown in Fig. 2j, the depth of the hexagonal pit nanostructures on the surface is reduced. This reduction is attributed to the uniform attachment of the PPy film onto the GaN surface, which partially fills the hexagonal pit nanostructures. Successful synthesis of the composite materials is collectively verified by the aforementioned analysis results.

Fig. 2. SEM and EDS images of sensor materials.

SEM images of a GaN-1 b PPy/GaN-1; elemental mapping of PPy/GaN-1: c C d Ga; SEM images of e GaN-2 f PPy/GaN-2; elemental mapping of PPy/GaN-2: g C h Ga; SEM images of i GaN-3 j PPy/GaN-3; elemental mapping of PPy/GaN-3: k C l Ga

The crystal qualities of three different composite films were characterized by XRD. As illustrated in Fig. 3, the (002) plane’s ω-rocking curves exhibit the full width at half maximum (FWHM) values of 2314.8 arcsec (GaN-1), 2426.4 arcsec (PPy/GaN-1), 979.2 arcsec (GaN-2), 1076.4 arcsec (PPy/GaN-2), 244.8 arcsec (GaN-3), and 324 arcsec (PPy/GaN-3). It can be seen that among the three groups of GaN nanostructure samples, GaN-1 exhibits the widest FWHM, which suggests that its internal grain size is the smallest. Additionally, after being combined with PPy, the FWHM values of PPy/GaN-1, PPy/GaN-2, and PPy/GaN-3 increased by 4.8%, 9.9%, and 32.4%, respectively. Notably, PPy/GaN-1 exhibits the smallest increase in FWHM, indicating that the introduction of PPy nanoparticles has the least impact on its dislocation density. In contrast, GaN-3 possesses a more planar structure, which facilitates higher PPy coverage and subsequently introduces greater dislocation density in PPy/GaN-3.

Fig. 3. XRD spectra of the (002) plane.

a GaN-1, b GaN-2, c GaN-3, d PPy/GaN-1, e PPy/GaN-2, and f PPy/GaN-3

To better investigate the interactions between GaN and PPy, the binding properties of the materials were analyzed by Raman spectroscopy. Figure 4a and Fig. S1 show the Raman spectra of three groups of GaN before and after compounding with PPy, as well as pure PPy. As shown in Fig. 4a, the absorption peak at 1589 cm^–1 is related to C = C stretching vibration, whereas the band at 1407 cm^–1 is ascribed to C-C stretching vibration³². The peak observed at 1241 cm^–1 is caused by C-H in plane bending vibration in PPy. The peak obtained near 1072 cm^–1 corresponds to the C-H in plane deformation. The small peaks near 933.47 cm^–1 are associated with the bipolaronic and quinoid polaronic structure²³. Combining Fig. 4a and Fig. S1, it is obvious that the Raman spectra of PPy/GaN is similar to those of PPy. The Raman scattering of PPy by the three groups of GaN nanostructured substrates is significantly enhanced, which proves the successful recombination of the three groups of GaN and PPy³³.

Fig. 4. Raman and XPS spectra.

a Raman spectra of PPy/GaN-1, PPy and GaN-1; b XPS survey spectrum of GaN-1 and PPy/GaN-1; c–f XPS spectra of Ga 3 d, N 1 s, O 1 s and C1s for GaN-1 and PPy/GaN-1

XPS characterization was employed to examine the elemental constituents and chemical environment of the composite films. Figure 4b shows the full XPS spectra of GaN-1 and PPy/GaN-1, confirming the presence of Ga, N, O, and C elements. The XPS spectra of GaN-2, PPy/GaN-2, GaN-3 and PPy/GaN-3 sensors are shown in Fig. S2 and Fig. S3. Figure 4c shows the Ga 3 d XPS spectra of GaN-1 and PPy/GaN-1, which can be deconvoluted into Ga-Ga, Ga-N, and Ga-O peaks. Compared to GaN-1, the three peaks in PPy/GaN-1 moved 0.84, 0.5, and 0.42 eV, respectively, reaching higher binding energies of 18.90, 19.78, and 20.64 eV. The shift in the sample peak position is attributed to charge transfer occurring at the hybrid interface between PPy and GaN, which induces alterations in the electron structure and electron density³⁴. Figure 4d shows the N 1 s XPS spectra of GaN-1 and PPy/GaN-1, where the N1s spectra of PPy/GaN-1 can be deconvolved into four characteristic spectral peaks attributed to Pyrrolic-N, N-Ga, and two Ga-LMM peaks, respectively. These four characteristic peaks are located at 398.18, 397.12, 395.48 and 392.57 eV respectively. Compared with GaN-1, the extra Pyrrolic-N characteristic peak clearly indicates that PPy molecules are successfully functioned into GaN. Combined with Fig. S2c and S3c, the area proportions of the Pyrrolic-N characteristic peaks in PPy/GaN-1, PPy/GaN-2, and PPy/GaN-3 are 16.7%, 14.4%, and 22.9%, respectively. Notably, the proportion of Pyrrolic-N characteristic peak in PPy/GaN-1 is moderate, providing an optimal balance between sufficient adsorption sites and avoiding excessive PPy coverage that could otherwise compromise conductivity. The O 1 s spectra of GaN-1 and PPy/GaN-1 in Fig. 4e can be deconvolved to ${\mathrm{O}}_2^{-}$, ${\mathrm{O}}^{2-}$, and ${\mathrm{OH}}^{-}$ peaks. The three peaks of PPy/GaN-1 are located at the binding energies of 531.00, 532.16 and 533.11 eV, respectively, and the area ratios of the three peaks are 50.8%, 29.5% and 19.7%, respectively. At the same time, the area ratios of the three peaks in GaN-1 are 15.1%, 54.5% and 31.4%. Evidently, the proportion of ${\mathrm{O}}_2^{-}$ (representing oxygen ions adsorbed on the material surface) in the composite has increased. Furthermore, the surface-adsorbed oxygen content of PPy/GaN-1 (50.8%) significantly exceeded that of PPy/GaN-2 (22%) and PPy/GaN-3 (21%). This disparity can be attributed to the larger specific surface area of GaN-1, which provides more adsorption sites for oxygen species. Consequently, the concentration of ${\mathrm{O}}_2^{-}$ in PPy/GaN-1 was substantially enhanced, demonstrating more favorable interactions and a stronger response to NH₃. In addition, the C1s spectra of PPy/GaN-1 and GaN-1 are shown in Fig. 4f, which can be convolved into C-C, C-N and C-O. The area ratio of C-N characteristic peak in PPy/GaN-1 is 0.34, which is significantly higher than 0.1 of C-N characteristic peak area in GaN-1, which also proves the successful combination of PPy and GaN.

Gas sensing characteristics

At 25 °C RH and 30% relative humidity, the gas-sensitive properties of PPy/GaN-1, PPy/GaN-2, PPy/GaN-3 and Pure PPy sensors were evaluated. Firstly, four groups of sensors were used for selective testing of seven typical gases at a concentration of 200 ppm, including NH₃, TMA(C₃H₉N), ethanol (C₂H₅OH), acetone (CH₃COCH₃), n-butanol (C₄H₁₀O), CH₄ and CO₂. The results are shown in Fig. 5a. These four sensors have the optimal response to NH₃ during the detection process, and the response values to interfering gases are all less than 10%. It can be concluded that PPy/GaN gas sensors not only exhibit high selectivity for NH₃, but also have significantly enhanced performance compared to pure PPy. This is attributed to the fact that PPy/GaN films provide abundant adsorption sites for NH₃. Then, the dynamic response curves of the three groups of sensors PPy/GaN-1, PPy/GaN-2 and PPy/GaN-3 under 200 ppm NH₃ environment were analyzed, and the results are shown in Fig. 5b. Among them, the response value of PPy/GaN-1 sensor to 200 ppm NH₃ is 59.7%, the response value of PPy/GaN-2 and PPy/GaN-3 are 44.5% and 24%, respectively. It is obvious that the PPy/GaN-1 sensor has a significant advantage over the other two sensors at 200 ppm NH₃ atmosphere.

Fig. 5. Performance comparison charts of three sensor groups.

a Selectivity of PPy/GaN-1, PPy/GaN-2, PPy/GaN-3 and Pure PPy sensors for different gases at concentrations of 200 ppm; b response and recovery curves of three sets of sensors to 200 ppm NH₃; c fitted response curves for PPy/GaN-1, PPy/GaN-2 and PPy/GaN-3 sensors; d–f Response curves of PPy/GaN-1, PPy/GaN-2 and PPy/GaN-3 sensors to different concentrations of NH₃ under RT; g-i PPy/GaN-1, PPy/GaN-2 and PPy/GaN-3 sensor resistance curve for different concentrations of NH₃ under RT

Subsequently, to further delineate and investigate the operational range, ultra-wide-range NH₃ tests were performed on the PPy/GaN-1, PPy/GaN-2, and PPy/GaN-3 sensors. The response values of three sensors to different NH₃ concentrations were fitted, and the results of the fitting equations and parameters are shown in Fig. 5c. It can be seen that the response values of the three sensors are consistent with the Langmuir isotherms for NH₃ molecules adsorbed on the PPy/GaN-1, PPy/GaN-2, and PPy/GaN-3 gas sensors. This behavior aligns with the fundamental gas adsorption principle, which states that gas adsorption on a sensor surface is constrained by the finite number of adsorption sites, leading to a saturation of adsorption capacity at higher gas concentrations. In the fitting equations, x is corresponding to the concentration of NH₃, and the correlation coefficients (R²) are 0.98 for PPy/GaN-1, 0.99 for PPy/GaN-2, and 0.98 for PPy/GaN-3, respectively. Figure 5d–f shows the response curves of PPy/GaN-1, PPy/GaN-2 and PPy/GaN-3 sensors at a high range of NH₃. It can be seen from Fig. 5d that when NH₃ concentration reaches 1000 ppm, the response value of PPy/GaN-1 sensor reaches 125.6%, and when the NH₃ concentration continues to drop to 100 ppb, the sensor can still make a relatively obvious response. This confirms that PPy/GaN-1 sensor can accurately and efficiently cover a wide operating range from 100 ppb to 1000 ppm. This wide range can be attributed to the synergistic interaction between GaN and PPy, which produces many heterojunctions as additional reaction sites. Figure 5e shows the response of the PPy/GaN-2 sensor at the NH₃ concentration of 1000 ppm-200 ppb, the sensor’s response value is 66.5% at 1000 ppm NH₃ concentration, which is obviously little improvement compared to 65.8% at 800 ppm concentration. This is due to the fact that as the NH₃ concentration increases, the active site reaches saturation and the availability decreases, making the sensor exhibit a threshold response or saturation effect. At the same time, the minimum detection limit of PPy/GaN-2 sensor for NH₃ is 200 ppb. According to Fig. 5f response curves of PPy/GaN-3 sensor to different concentrations of NH₃ under RT, the response value of PPy/GaN-3 sensor is lower than that of the other two sensors under each concentration gradient of NH₃, and the lower limit detection ability is the worst, the lower limit is 500 ppb, and the response value is 0.54%. Overall, due to the optimal distribution uniformity of the film, PPy/GaN-1 sensor demonstrates the optimal synergistic effects of GaN and PPy, achieving the lowest detection limit and a high sensitivity response to NH₃, thereby exhibiting superior performance compared to the other two groups of sensors, which is consistent with the analysis of the above characterization results.

Figure 5g–i shows the dynamic resistance curves of the three groups of sensors at RT and 30% RH to the corresponding concentration of NH₃. When in the air, the Ra values of the three groups of sensors were 2 × 10⁷, 7 × 10⁷, and 6.8 × 10⁴ Ω respectively. When the three sensors were exposed to an NH₃ reduction atmosphere, they exhibited P-type semiconductor behavior with increased resistance, which can be attributed to the successful in situ polymerization of PPy on GaN surfaces. Meanwhile, the responses of GaN-1, GaN-2, and GaN-3 to different concentrations of NH₃ at RT and 30% RH were tested. The response curves and dynamic resistance curves were shown in Fig. S4. The comparison between Fig. 5d–f and Fig. S4a–c shows that the gas sensitive performance of PPy/GaN-1, PPy/GaN-2 and PPy/GaN-3 sensors under each concentration of NH₃ has been greatly improved compared with that before the combination. By comparing Fig. 5g–i with Fig. S4d–f, it can be seen that pure GaN sensors have high initial resistance and resistance fluctuations at room temperature, and poor stability in the gas sensitive performance test, while after combining with PPy, the baseline resistance of the three groups of sensors is reduced by 2-3 orders of magnitude, and it is very stable in the gas sensitive performance test. This shows that after the introduction of PPy nanoparticles, the conductivity of the sensor is enhanced, and the stability is greatly improved.

Furthermore, in order to better demonstrate the stability, repeatability and reliability of PPy/GaN-1 sensor, the sensing performance of the sensor was systematically evaluated. Figure 6a and Fig. S5, respectively present the single dynamic response curve of the PPy/GaN-1 sensor to 200 ppm NH₃ and the response curves of the pure PPy sensor to NH₃ gas of different concentrations, it is evident that the sensor can rapidly generate the first distinguishable signal within 110 s and achieve desorption within 116 s. This demonstrates that NH₃ can readily enter and exit the structure during the adsorption and desorption processes. The baseline resistance of the sensor can essentially be restored to its initial value after a reaction cycle, thereby confirming its stability. Figure 6b shows three response cycles of the PPy/GaN-1 sensor to 200 ppm, 20 ppm, and 2 ppm NH₃ at room temperature, respectively. It can be seen that in each cycle, the sensor produces the same response pattern at the same concentration of NH₃, and the response deviation is minimal, verifying its repeatability. Considering the issue of uncontrollable temperature in practical working environments, the influence of operating temperature on the PPy/GaN-1 sensor was investigated. Figure 6c shows the effect of temperature near room temperature on both the resistance of the sensor and its response to 1 ppm NH₃. The test results indicate that as the operating temperature increases within the range of 20–40 °C, the resistance of the fabricated sensor exhibits a gradual upward trend. This behavior can be attributed to the temperature sensitivity of the carrier concentration. With increasing temperature, the Fermi level in GaN shifts towards the mid-bandgap, leading to an expansion of the heterojunction barrier, which consequently results in the observed increase in resistance. Since the adsorption/charge transfer kinetics typically follow Arrhenius behavior, the sensor’s sensitivity reaches a slight peak at 25 °C. However, as the temperature further increases, it adversely affects the π-π stacking structure of the conjugated polymer and enlarges the depletion layer width. This increases the hole transport barrier, making it difficult for electrons provided by NH₃ to effectively compensate for the hole loss, consequently leading to a decline in sensor sensitivity. Furthermore, although a decrease in sensitivity is observed, the maximum attenuation rate of the sensor’s sensitivity to low-concentration NH₃ within the near-room-temperature range is 9.3%, demonstrating that the sensor maintains acceptable stability under conditions close to room temperature. Figure 6d shows the change in the response value of the PPy/GaN-1 sensor at 20 ppm NH₃ observed every 10 days at RT and 30% RH, and the change in response to 20 ppm NH₃ over 90 days is less than 8.8%, indicating excellent long-term reliability of the sensor.

Fig. 6. The repeatability and stability of PPy/GaN-1 sensor.

a Response and recovery time of PPy/GaN-1; b repeatability of PPy/GaN-1 sensor for 200 ppm, 20 ppm, and 2 ppm NH₃; c the influence of operating temperature on PPy/GaN-1; d Stability curves of PPy/GaN-1 sensor to 20 ppm NH₃ for 90 days

Gas sensitive performance is affected by the actual working environment. Therefore, the influence of different humidity conditions on the sensing performance of PPy/GaN-1 sensor was studied. In the humidity performance test of the sensor, we accurately control the humidity inside the air chamber by combining a humidity generator with a humidity sensor. When the humidity levels in the gas chamber and the baseline resistance of the sensor reach a stable state, a specific concentration of NH₃ is introduced into the gas chamber to monitor the resulting change in the sensor’s resistance value. The Fig. 7a–d and Fig. S6 show the dynamic response and resistance curves of PPy/GaN-1 to different concentrations of NH₃ under various humidity conditions (20% RH, 40% RH, 60% RH, and 80% RH). Response characteristics and baseline resistance variations of PPy/GaN-1 under diverse RH conditions and NH₃ concentrations are illustrated in Fig. 7e, f. Combined with Fig. 7a–e, it can be concluded that both the highest response value and the lowest detection limit of the sensor to NH₃ are obtained at 30% RH. When the humidity increased from 20% RH to 30% RH, H₂O molecules preferentially occupied the defect sites on the surface of the sensing material via chemical adsorption¹³. At this stage, trace water molecules acted as electron donors (weak reducing agents) and engaged in electron transfer with the polaron/bipolaron structures of PPy. The conjugated chains of PPy possessed unoccupied orbitals capable of accepting electrons donated by water molecules, facilitating charge separation. This electron transfer process optimized the adsorption of NH₃: as NH₃ acted as an electron donor, its interaction with PPy was enhanced, allowing electrons to transfer more readily to the PPy backbone²⁵. As a result, the resistance of the sensing material increased rapidly in the presence of NH₃, thereby improving the sensor response³⁵. However, with a further increase in humidity, the formation of a physically adsorbed water multilayer film disrupts the conjugated structure of PPy and induces swelling of its film, thereby slowing the initial rate of resistance reduction (from 30% RH to 80% RH in Fig. 7f) Simultaneously, the adsorption of excessive water molecules on the surface of the sensing material shields the available NH₃ adsorption sites on the PPy/GaN surface, ultimately leading to a reduction in the sensor’s response capability³⁵. As can be seen from Fig. 7f, due to the continuous adsorption of water molecules, the baseline resistance of the PPy/GaN-1 sensor gradually decreases with the increase of humidity. This is due to the rapid migration of ions produced by water molecules through the PPy pore structure and PPy/GaN interface, significantly improving the ionic conductivity. Overall, the performance of the sensor suffers with increasing humidity, but even under high humidity conditions (80% RH), the NH₃ concentration of 1 ppm can still be accurately detected. The results above verify the reliability of sensing response under humid conditions, and prove that the sensor has great potential for practical application.

Fig. 7. The moisture resistance of PPy/GaN-1 sensor.

a–d Response curves of PPy/GaN-1 sensor to NH₃ concentration at 20%, 40%, 60% and 80% RH at room temperature; e response of PPy/GaN-1 sensor to different concentrations of NH₃ at different RH at room temperature; f influence of RH on the resistance of PPy/GaN-1 sensor at RT

Table 1 compares the performance of the proposed PPy/GaN-1 sensor with other recently reported NH₃ sensors. Material choices for sensors listed in Table 1 include MOF, Sulfide, Mxene, CNTs and rGO. As can be seen, the PPy/GaN-1 sensor exhibits a broad dynamic detection range for NH₃, with an ultra-low detection limit of 100 ppb and a suitable operating temperature. Compared with several recent PPy-based sensors, the PPy/GaN-1 sensor demonstrates a higher response and lower detection limit. Although it exhibits certain disadvantages in response relative to NH₃ gas sensors fabricated from other materials, its ultra-wide detection range (0.1–1000 ppm) still has certain advantages and can meet the detection requirements of trace levels of NH₃ (ppb concentration). This provides a valuable pathway for advancing high-performance NH₃ sensors through material optimization based on PPy/GaN composites.

Table 1.

Comparative analysis between the PPy/GaN-1 sensor and previously reported NH₃ sensors

Material	T (°C)	Response (%)	Concentration (ppm)	Detection range	Ref.
PPy/Zntpp	RT	25	200 ppm	20–1000 ppm	41
PPy/MoS2	RT	21.65	100 ppm	10–100 ppm	42
PPy urchins	RT	45.13	200 ppm	1–200 ppm	43
MXene/MoS2/PPy	RT	31	200 ppm	10–200 ppm	44
PPy-MWCNTs	RT	29.74	200 ppm	5–200 ppm	45
PPy/Ti3C2Tx	RT	31.9	100 ppm	5–100 ppm	46
GaN/rGO	RT	92	200 ppm	0.5–200 ppm	47
SnS2/Ti3C2Tx	RT	72	200 ppm	1–500 ppm	48
PEDOT:PSS/MXene	RT	62	200 ppm	0.6–1000 ppm	49
SnS2/graphene	RT	70	200 ppm	1–500 ppm	50
PPy/GaN-1	RT	59.7	200 ppm	0.1–1000 ppm	This paper

Gas sensing mechanism

Semiconductor-based gas detectors function through chemical-to-electrical signal conversion, where semiconductor elements gauge atmospheric pollutant levels by translating molecular interactions into resistance variations. This sensing mechanism fundamentally relies on charge transfer-triggered electrical conductivity adjustments³⁶. The NH₃ sensing mechanism of PPy is shown in Fig. 8a. The interaction mechanism of PPy with proton adsorbed ${\mathrm{H}}_{\mathrm{ad}}^{+}$ and NH₃ is presented by the whole equation as follows:

$$\left(\left(\mathrm{PPY}\right)·{\mathrm{H}}_{\mathrm{ad}}^{+}\right){\mathrm{Cl}}^{-}+{\mathrm{NH}}_3\to \left(\left(\mathrm{PPY}\right)·{\left({\mathrm{NH}}_4^{+}\right)}_{\mathrm{ad}}\right){\mathrm{Cl}}^{-}$$ 3

Fig. 8. Sensing mechanism diagrams.

a NH₃ sensing mechanism of PPy; b the energy band diagrams of PPy/GaN composite material in air and NH₃

The specific gas sensitivity mechanism is shown as in Supporting note 2³⁷.

The energy band diagrams of PPy/GaN composites in air and NH₃ are shown in Fig. 8b. The Fermi level (E_F) of N-type GaN is near the bottom of the conduction band (E_C), while that of P-type PPy is close to the top of the valence band (E_V)²⁷. At the interface, since the Fermi level of GaN is higher than that of PPy, electrons will flow from N-type GaN to P-type PPy, until the Fermi level is unified, forming a depletion layer^38,39. The resistance of pure PPy is directly determined by the hole concentration and mobility, and the holes in the composite need to overcome the PPy/GaN interface barrier to participate in conduction, so the initial resistance of PPy/GaN increases relative to the initial resistance of PPy. GaN exhibits high resistivity due to its wide bandgap. PPy, as a highly conductive polymer, forms a continuous conductive network in the composite material, significantly reducing the overall resistance and making it easier to detect NH₃.

In situ monitoring of sensors plays a critical role in elucidating the adsorption mechanisms of gases. To elucidate the sensing mechanisms and interactions between the reducing gas NH₃ and PPy/GaN composite materials, in situ XPS analyses were performed on PPy/GaN-1, PPy/GaN-2, and PPy/GaN-3 before and after 30-min exposure to a 500 ppm NH₃ atmosphere, aiming to investigate the role of adsorbed oxygen in the sensing mechanism. The O 1 s spectra of the PPy/GaN-1 sensor before and after NH₃ exposure are presented in Fig. 9. It can be observed that under NH₃ atmosphere, the concentration of adsorbed oxygen species (${\mathrm{O}}_2^{-}$) on the PPy/GaN-1 composite surface decreased significantly, from 50.8% in air to 21.6%. The O 1 s spectra of PPy/GaN-2 and PPy/GaN-3 after NH₃ exposure are displayed in Fig. S7. According to the data, the concentration of ${\mathrm{O}}_2^{-}$ on PPy/GaN-2 decreased from 22% in air to 12%, while for PPy/GaN-3, it decreased from 21% to 9%. The reduction in adsorbed oxygen content across all three sensor groups, to varying extents, further supports the involvement of adsorbed oxygen species in the NH₃ sensing mechanism.

Fig. 9. In-situ XPS analysis diagrams.

a The O 1S spectrum of the PPy/GaN-1 sensor before exposure to NH₃; b the O 1S spectrum of the PPy/GaN-1 sensor after exposure to NH₃

In PPy/GaN composite, the sensing mechanism for NH₃ can be attributed to electron transfer during the NH₃ adsorption and desorption processes, as well as the deprotonation of PPy. Under ambient air conditions, oxygen molecules are spontaneously adsorbed onto the surface of the PPy/GaN composite, capturing free electrons and forming adsorbed oxygen (${\mathrm{O}}_2^{-}$). Upon exposure to NH₃, two primary reactions occur. First, NH₃ molecules react with the ${\mathrm{O}}_2^{-}$ on the composite surface, leading to the release of a significant number of electrons. These electrons rapidly recombine with holes present in PPy, thereby increasing the width of the hole depletion layer at the PPy/GaN interface. Second, the reducing NH₃ gas interacts with PPy through a deprotonation process, during which electrons are transferred to PPy, resulting in the formation of ${\mathrm{NH}}_4^{+}$. These two processes occur concurrently¹³. As a result of electron-hole recombination, the overall hole concentration in the PPy/GaN composite decreases, leading to a substantial reduction in the sensor’s electrical conductivity. When the sensor is removed from the NH₃ environment, the composite gradually releases electrons, the hole carrier concentration is restored, the depletion layer width returns to its original state, and the sensor’s resistance recovers to its initial level. The reaction process of ${\mathrm{O}}_2^{-}$ and NH₃ can be described as follows:

$${\mathrm{O}}_{2\left(\mathrm{gas}\right)}\to {\mathrm{O}}_{2\left(\mathrm{ads}\right)}$$ 4

$${\mathrm{O}}_{2\left(\mathrm{a}\mathrm{d}\mathrm{s}\right)}+{\mathrm{e}}^{-}\to {\mathrm{O}}_2^{-}$$ 5

$$4{\mathrm{NH}}_{3\left(\mathrm{ads}\right)}+3{\mathrm{O}}_2^{-}\to 2{\mathrm{N}}_2+{6\mathrm{H}}_2\mathrm{O}+3{\mathrm{e}}^{-}$$ 6

In summary, the performance of the sensor toward NH₃ has been substantially enhanced through the construction of the PPy/GaN heterojunction system. Specifically, the direct coupling of holes in PPy with the electron-withdrawing property of NH₃ and the effect of adsorbed oxygen on the PPy/GaN surface significantly enhance the charge transfer efficiency during the NH₃ adsorption process and induce a significant change in resistance. Additionally, the PPy/GaN thin films offer abundant adsorption sites for NH₃, which accelerates the kinetics of NH₃ adsorption and desorption, thus optimizing the response and recovery times. Consequently, the sensor exhibits markedly improved sensitivity to NH₃.

Application

Exhaled breath analysis serves as an emerging diagnostic approach, evaluating human physiological health through monitoring changes in exhaled gas composition. Wenzheng Heng et al. demonstrated that in healthy individuals after high protein intake, NH₄⁺ in exhaled condensate was linearly correlated with serum urea concentration⁴⁰. This provides an idea for the exhaled breath detection of CKD patients simulated in this experiment.

In this experiment, a sensor array was constructed based on the different sensing characteristics of PPy/GaN-1, PPy/GaN-2 and PPy/GaN-3 sensors for NH₃. The composition of the detection system is as shown in Supporting note 3. The exhalation detection experiment process conducted in this study is illustrated in Fig. 10. The breath detection process of the simulated CDK patient is shown as Supporting note 4.

Fig. 10. Flowchart of the simulation experiment.

The experimental process diagram of exhaled breath detection for simulated CDK patients

In the exhalation detection experiment, the humidity of exhaled breath is monitored in real time using the humidity sensor integrated within the exhalation detection device. Humidity Sensor data reveals that when the exhaled breath collection duration is maintained at 30 s, the relative humidity remains stable within the range of 60% RH to 80% RH. Based on these findings, this study establishes a humidity correction model utilizing Particle Swarm Optimization combined with a Backpropagation neural network (PSO-BP) to enable real-time humidity compensation. The model aims to standardize the sensor response data by mapping them to the reference response values at 30% RH. The training dataset used by this model includes the response data of sensors to NH₃ collected within the humidity range of 30% RH to 80% RH.

Upon acquiring real-time resistance signals, the system first converts them into response values, which are then fed into the PSO-BP model alongside current humidity data. The model calculates standardized response values referenced at 30% RH, with outputs documented in Fig. 11a. This humidity compensation model employs a BP neural network featuring two hidden layers with 32 and 16 neurons respectively. To enhance convergence efficacy, the PSO is implemented for parameter tuning, configured with a population size of 40, learning factors c1 = 1.5 and c2 = 1.5, and a maximum inertia weight of 0.8. The mean squared error (MSE) progression during training is illustrated in Fig. 11b, while the final compensation outcomes are demonstrated in Fig. 11c, d. The R² exceeds 0.9 across both training and test sets, with prediction errors of response constrained within 5%, demonstrating the model’s capability to deliver precise, stable, and reliable humidity response compensation for sensors.

Fig. 11. The humidity compensation model and detection response data of the sensor array.

a Sensor humidity compensation model; b mean square error curve of PSO-BP neural network; c the model’s performance in the training set; d the model’s performance in the validation set; e expiratory data collection in the fasting state (without protein intake); f continuous expiratory monitoring data after protein intake; g data on the response of the sensing array to the detection of exhaled breath in different states of the target individual

The breath test results of the simulated CKD patients in the experiment are shown in Fig. 11e–g. Among them, Fig. 11e shows the response recovery curve of the sensor array of the tester (without protein intake) within 1 h in the fasting state. During this period, the 6 sets of data obtained by the sensor array have good repeatability and also verify that the array has a good anti-interference ability. Figure 11f illustrates the breath monitoring curve of the test subject over a 2.5-h period following the ingestion of a large quantity of protein within a short timeframe. Figure 11g presents the overall response result graph obtained from a series of breath test experiments. It can be inferred from the Figure that after protein consumption, the sensor array exhibited a pronounced signal enhancement. As time progressed, due to the test subject’s digestive and metabolic processes, the sensor array’s monitoring signal gradually diminished, demonstrating a tendency to revert to the fasting state. After the test subjects consumed high protein, the serum urea content in their bodies increased in a short period of time, resulting in an increase in the NH₃ content in exhaled breath. The experimental results indicate that the sensor array could accurately identify the trace changes in the test subjects’ exhaled breath and respond promptly, which also verified the high sensitivity of the sensor array.

Furthermore, in terms of gas concentration prediction, the prediction model of a single sensor has problems such as poor robustness and significant accumulation of errors. In this work, we have developed a concentration prediction model based on multi-sensor data fusion and an enhanced radial basis function neural network (RBFNN), using extensive response data from the three sensors exposed to varying NH₃ concentrations on a testing platform as training and test sets. The nodes and architecture of the model are shown in Fig. 12a. The specific algorithm is shown as in Supporting note 5.

Fig. 12. Model architecture and prediction results.

a The nodes and architecture diagrams of the model; b the prediction result graph of NH₃ concentration in the low concentration range

During the experimental verification stage, real monitoring data from the sensor array were utilized, and the mean absolute error (MAE), coefficient of determination (R²), root mean square error (RMSE), and mean relative error (MRE) were used as evaluation indicators to visually assess the performance of the model, as shown in Table 2. The results demonstrate that, on the test set, the model achieves a prediction accuracy with an error of 1.17 ppm. As illustrated in Fig. 12b, the distribution curve of actual concentration versus predicted concentration exhibits a high degree of fitting, visually highlighting the model’s capacity to track trends accurately. This experiment further proves that the sensor array prepared in this work, combined with machine learning, has great potential in the early warning and continuous monitoring of CKD patients.

Table 2.

Model performance parameters

Algorithm	MAE	RMSE	R2	ARE
RBFNN	1.17	1.59	0.9701	5.31%

Conclusion

In conclusion, in this paper, high-performance PPy/GaN gas sensors were successfully fabricated by using the precisely controllable MOCVD process and in situ oxidative polymerization. Notably, the PPy/GaN-1 sensor demonstrated exceptional gas-sensitive sensing performance for NH₃, characterized by an ultra-wide detection range (100 ppb to 1000 ppm), outstanding moisture resistance, and long-term stability. The gas sensitivity mechanism of the PPy/GaN composite materials was systematically analyzed. GaN nanostructure not only served as the substrate for PPy but also formed a heterostructure, significantly enhancing charge transfer efficiency and providing abundant NH₃ adsorption sites, thereby leading to a significant enhancement in sensor sensitivity. Furthermore, exhaled breath monitoring experiments were conducted on simulated CKD patients using a sensor array to validate its high sensitivity. By integrating machine learning algorithms, high-precision prediction of gases within the low-concentration range was achieved, with an error of 1.17 ppm. These experiments collectively confirm that the sensor array holds significant potential for early warning and continuous monitoring of CKD patients.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41378-025-01119-7.