Simultaneous Detection of CO and NO2 Gases using Interaction Analysis of SnS2 Sensor Array Response

Developing a multi-analyte gas sensing system that simultaneously detects trace levels of CO and NO2 at low temperatures is necessary for the Internet of Things (IoT) based air quality monitoring applications. Nevertheless, gas sensors operating at low temperatures are nonspecific and rarely detect target gases at lower ppb levels in the air. Herein, an array of two SnS2 sensors with different bias voltages has been developed and characterized upon exposure to individual and binary mixtures of CO and NO2 gases at different concentrations. The developed gas sensors array achieved the lower detection limit of 45 ppb for NO2 and 150 ppb for CO. Further, co-adsorption-induced interaction analysis was carried out to predict the target gas concentration in the binary mixture using the mixed gas response. The mean absolute percentage error of 7.86% is observed in predicting the target gas concentrations in the binary mixture, which indicates the high prediction accuracy of proposed method. As a minimal resource intensive approach, the proposed method can be used in air quality monitoring applications that require low-power and low-cost sensors.

Estimating the concentrations of pollutants such as NO 2 and CO is vital in air quality monitoring.2][3][4][5] In addition, they cause several environmental effects such as acid rain and ground-level ozone formation and even contribute to the formation of particulate matter.According to the US National Ambient Air Quality Standards (NAAQS), acceptable CO levels in the environment are 9 ppm on average for 8 h, whereas NO 2 levels are 100 ppb on average for 1 h. 6][9][10][11][12][13][14] However, unmodified metal oxide gas sensors require high operating temperatures and thus consume more power.On the other hand, carbon materials, metal sulfides and their composites with metal oxides require a low operating temperature to detect the trace levels of these toxic gases.Nevertheless, they suffer from cross-sensitivity to other gases.A few reports, such as low-frequency noise spectra analysis, 15 transient analysis, 16,17 temperature modulation, 18 adsorption-desorption noise analysis, 19 and machine learning with the steady-state response, [20][21][22][23][24][25][26][27] have tried to address the abovementioned issues.Though noise analysis is an excellent method to discriminate the gases, it is not efficient in determining their concentration.Besides, the noise measurement system is bulkier and too expensive to use in practical environmental monitoring applications.The transient analysis is applicable only for applications requiring step response analysis but cannot be used to monitor gases in the environment continuously.The temperature modulation method is power-hungry, whereas the adsorption-desorption noise method is ineffective in simultaneously detecting multiple gases.The machine learning-based approaches are computationally intensive and cannot be used where the power consumption should be too low.Therefore, developing a gas detection platform that simultaneously detects multiple gases in the air at lower concentrations with nonspecific gas sensors remains challenging.
To this end, a gas sensor array consisting of two-room temperature SnS 2 chemiresistive sensors is developed to detect CO and NO 2 gases at trace levels simultaneously.The sensors exhibited reversible, repeatable and stable responses upon exposure to single and binary analytes.Ideally, the response due to mixed gases should be the sum of the responses due to individual gases in their mixture.However, the above assumption is invalid in reality due to the mixed gases' co-adsorption-induced interaction effect. 28The deviation in the response due to co-adsorption can be ignored if the sensors' response to individual gases is a homogeneous function of the concentration of the gas.Nevertheless, the difference is notable if the response is inhomogeneous.In such cases, the deviation from the sum of the individual responses must be considered when analyzing the sensor's response to mixed gases.In this regard, the concentration of gases in the mixture is predicted using co-adsorption-induced interaction analysis.The mean absolute percentage (MAPE) error in predicting NO 2 and CO concentrations is close to 8%, acceptable in gas sensor applications.As a simple approach to predicting the concentration of gases in a mixed gaseous environment, it can be used to monitor the air quality in low-power IoT applications.

Experimental
The SnS 2 flowers with nano petals were synthesized using tin and sulfur precursors with a solvothermal method.An autoclave with a solution of 350 mg SnCl 4 •5H 2 O and 400 mg thioacetamide dissolved in ethanol (40 ml) was heated at 180 °C for 15 h.The resultant solution was washed and dried at 70 °C to get SnS 2 powder.The synthesized SnS 2 was characterized by FESEM (JEOL JSM 7800 F), TEM (JEOL 2100), powder XRD (Panalytical Xpert Pro, Cu Kα1 (λ = 1.5406Å)), Bruker Raman Spectroscopy (532 nm) to investigate its morphological and structural properties.
The sensors were made by drop-coating SnS 2 flowers solution in ethanol solvent on gold interdigitated electrode contacts on the glass substrate and drying the film at 60 °C.The gas sensing experiment was carried out using equipment containing the current measurement probes to get the device's resistance during the experiment.The synthetic air (79% N 2 and 21% O 2 ) was used as reference gas, whereas the experimental grade CO and NO 2 gases were utilized as target analytes.The required relative humidity (RH) is achieved by flowing a part of the synthetic air through a water bubbler.A LabView-based data acquisition software was employed to control the composition of target gas using Mass Flow Controllers (MFCs), read the data and control other manifold valves in the gas sensing experimental setup.It stores the data read from Keithly SMU, which contains the time stamp, applied voltage, and sensor current from which the device's resistance can be extracted.From the extracted z E-mail: sgsingh@ee.iith.ac.inECS Sensors Plus, 2023 2 045201 resistance values, the steady-state response corresponding to different compositions of the binary mixture was calculated for both the sensors.The total flow rate of the target gas and the time interval between the consecutive measurements are constant throughout the experiment.

Results and Discussion
Figure 1a shows the XRD pattern of the synthesized material.The diffraction peaks can be indexed to hexagonal structure SnS 2 (JCPDS data 23-0677). 29,30Figure 1b shows the Raman spectra of SnS 2 .The strong peak at 311 cm −1 represents A1g mode, and the broad peak between 500 cm −1 and 700 cm −1 can be attributed to the second-order effects.The downshift of the A1g peak of synthesized material from that of bulk materials (317 cm −1 ) can be attributed to nanosize effects.The FESEM (Figs. 2a and 2b) and TEM images (Figs.2c and 2d) show the morphology of the synthesized SnS 2 .The material has micro flowers with nano petals morphology.The high surface area to volume ratio of nano petals enhances the sensitivity of a chemiresistive gas sensor.The elemental analysis of SnS 2 using Energy Dispersive X-ray Spectroscopy (Figs.S1, S2 and Table S1) shows that the atomic ratio of Sn to S is close to 0.5.
A chemiresistive gas sensor array is constructed with SnS 2 sensors to detect CO and NO 2 concentrations in their mixed gaseous environment.Figure 3 depicts the grain boundary of SnS 2 with (a) O 2 under no bias, (b) O 2 with bias voltage and (c) CO and NO 2 under applied bias voltage.The E C and E V represent the bottom of the conduction band and the top of the valence band, respectively and E F denotes the Fermi level.The x d1 and x d2 are the widths of the depletion regions on two sides of the grain boundary (x = 0).Assuming the grains are of the same size, the widths of the depletion layer on two sides are equal (x d ) when there is no applied bias voltage in the oxygen environment (Fig. 3a).The grain boundary potential ( ) V B0 and the total depletion width (x d ) under no applied bias are given by Eqs. 1 and 2. 31 where N B is the charge density of grain boundary, q is the charge, N d is the donor density, and ε is the sensing film permittivity.Upon application of bias voltage (U), the potential barriers on two sides of the boundary vary as given by Eqs. 3 and 4 and N P is the quantity of potential barriers across the SnS 2 film (Fig. 3b).When U is high, the voltage-dependent potential barrier gives nonlinear current characteristics with applied bias voltage.Moreover, the depletion width increases on one side of the boundary and decreases on the other side of the grain when the bias voltage is applied; thus, the depletion widths on both sides are not equal.However, the total width on both sides is equal to agree with the electroneutrality condition despite bias voltage application (Eq.5).
In the presence of the target gases, the depletion layer width decreases due to the donation of electrons from the target gas to the sensing layer.Therefore, the total depletion width ( + ) varies in the presence of the target gas, and consequently, the current through the sensor changes.The current through the sensor under applied bias voltage can be expressed as Eq. 6. 31 where A is the area of the cross-section of the sensing film, μ is the charge carrier mobility, k B is the Boltzmann constant, and T is the temperature in Kelvin.As the potential barrier is a function of both the target gas and bias potential, the current carriers experience different potential barriers when bias potential varies (Fig. 3c).2][33][34][35][36] Thus, a sensor array with two SnS 2 gas sensors is formed to measure the target gas concentrations in their mixture.
To test the performance of the sensors in the array, they are exposed to various concentrations of CO and NO 2 individually at room temperature.The sensors showed a reasonable response to ppb levels of target gases (Fig. 4).The minimum levels of detection of CO and NO 2 at 5 V are 300 ppb and 150 ppb, respectively, whereas these levels are 150 ppb and 45 ppb, respectively, at 12 V bias ECS Sensors Plus, 2023 2 045201 voltage.The response for 45 ppb NO 2 is low but the signal to noise ratio is much greater than 3, which is the required condition for detection of the gas at a particular concentration.However, the sensors showed an abnormal response when exposed to NO 2 gas.Usually, the reducing gases such as CO donate the electrons, and oxidizing gases like NO 2 accept the electrons from the conduction band of n-type semiconducting materials such as SnS 2 .Therefore, the resistance decreases with exposure to reducing gases and increases with exposure to oxidizing gases.However, in this case, the sensor's resistance decreased with the exposure of both CO and NO 2 .It can be attributed to Fermi level shifting of the material and the orbital mixing. 37Nevertheless, it can be safely assumed that both NO 2 and CO donate the electrons at room temperature when SnS 2 is exposed to them, and consequently, the sensors' resistance  decreases.Further, to know the behavior of the sensors in the presence of more than one target gas, they were exposed to the binary mixtures of CO and NO 2 for multiple cycles, as shown in Fig. 5.The sensors exhibited a reversible and repeatable response.Moreover, the response is higher than that of individual target gases.However, the response may be more than the combined response or less based on the effect of adsorption of one gas on another gas, which depends on the composition or partial pressure of the target gases.
The sensor's response (δ ) R to mixed gas can be represented by the simple addition of the sensor's response to individual gases (Eqs.7 and 8) if the adsorption of one gas is independent of the    presence of other interfering gases in the environment.
Subscripts 5 V and 12 V represent bias voltages, NO 2 and CO represent the target gases, and S and C are sensitivity and concentration, respectively.However, the co-adsorption of interfering gases influences the adsorption of a target gas.Thus, an interaction term was introduced to address this issue, as shown in Eqs. 9 and 10.
As the sensors are reversible or point functions of concentration, the interaction coefficients can be written as the product of the sensitivities to the individual gases. 38,39Therefore, Eqs. 9 and 10 can be re-written as Eqs.11 and 12.
[ ] These equations can be re-arranged as Thus, the sensor's response to the mixed gases can be written as the product of its responses to individual gases.So, the steady-state response of sensors to CO and NO 2 was calibrated as in Fig. 6 and formulated as in 15 and 16 where Res 12v and Res 5v are mixed gas responses at 12 V and 5 V, respectively.The above two equations are plotted on XY space (CO concentration on X-axis and NO 2 concentration on Y-axis) to predict the individual gas concentrations present in the mixed gas, as shown in Fig. 7.The intersection coordinates on the XY plot represent the target gas composition.These concentrations for various mixed gas responses were measured 10 times for each combination, and the mean absolute percentage error from actual concentrations is calculated as represented in Table I.The error is slightly above 10% when one of the gases is low in concentration (45 ppb NO 2 and 1 ppm CO), but it is below 10% in all other cases.The overall mean absolute percentage error in gas concentration prediction when all the measurements are considered is 7.86%, which is practically acceptable.
Further, rapid tracking of gas concentrations in the air is necessary for continuous gas monitoring.The sensor should respond quickly when adsorption and desorption of gas molecules on semiconductor material takes place to fulfill the criterion mentioned above.So, the response-recovery times are measured from the transient response and plotted as shown in Fig. S3.It can be understood that the response and recovery of the sensors operating at 5 V and 12 V are less than 5 min, which is sufficient for practical use.Also, the response time is almost constant irrespective of the target gas concentration, which agrees with the literature. 40inally, the sensors' repeatability, stability, selectivity and the effect of sensing layer thickness were studied to know their possible utility in real-time gas sensing applications.The repeatability of the sensors was observed by exposing them to the same concentration of target gases for multiple cycles, as depicted in Fig. S4.The response is repeatable, with a minor change in response over multiple cycles.The sensor's response is measured once in 3 weeks (Fig. S5) to investigate its stability.The minor variation in the response with time indicates that the sensors are stable over time.The selectivity of the sensors to CO (3 ppm) and NO 2 (750 ppb) is investigated by exposing the sensors to other possible gases (CO 2 (500 ppm), H 2 S (500 ppb), NH 3 (500 ppb), CH 4 (100 ppm), SO 2 (500 ppb)) in the environment.The negligible sensors' response to other gases (Fig. S6) indicates the strong selectivity of the sensors to CO and NO 2 .The effect of humidity on the sensors' response is investigated by exposing the sensors to the target gases at different RH levels (Fig. S7).It is observed that there is no significant change in response levels (ratio of resistance before and after exposure of the gas) though the resistance varies with the RH levels.The thickness of the sensors used in the experiment is approximately 32 μm.The effect of sensing layer thickness on the sensor's response is shown in Fig. S8.It can be observed that the response is more or less the same when thickness varies from 19 μm to 62 μm, but it decreases slightly when the thickness is above 100 μm.The insignificant change in sensors' response with a change in thickness from 19 μm to 62 μm is attributed to the porous nature of dropcasted sensing layers.In the case of compact films, the gas cannot penetrate through the sensing film and interacts only with the surface of the sensing film.Thus, the response decreases significantly with the increase in the thickness of the sensing layer.However, in the case of porous films, the gas penetrates down to the substrate and interacts with individual grains. 41Therefore, the sensitivity is almost constant when the change in sensing layer thickness is not too high.ECS Sensors Plus, 2023 2 045201

Conclusions
An array of gas sensors comprising two SnS 2 sensors biased with different voltages exhibited notable response characteristics to the individual and binary mixture of NO 2 and CO gases.The sensors are non-specific to either NO 2 or CO in their unfunctionalized form.Besides, the sensors exhibited repeatability and stability at room temperature with a reasonable response and recovery times.Further, co-adsorption-desorption analysis was used to investigate the sensor's capability in predicting the composition of CO and NO 2 in their mixture.The mean absolute percentage error in predicting the composition of CO and NO 2 is less than 8%, which is acceptable in pollution monitoring applications.The acceptable predictability of gas concentrations indicates that the proposed approach has a promising future in air quality monitoring applications.

Figure 3 .
Figure 3. SnS 2 grain boundary in the presence of air with (a) no bias and (b) with bias; (c) upon exposure with the target gases (NO 2 and CO) under applied bias voltage.

Figure 4 .
Figure 4. Transient response of SnS 2 sensor with (a) NO 2 and (b) CO at 5 V and (c) NO 2 and (d) CO at 12 V.

Figure 5 .
Figure 5. Transient response of SnS 2 sensors with a mixed gas of NO 2 and CO at (a), (b) 5 V and (c), (d) 12 V.

Figure 6 .
Figure 6.SnS 2 sensors response as a function of gas concentration: (a) NO 2 and (b) CO at 5 V and (c) NO 2 and (d) CO at 12 V.

Figure 7 .
Figure 7. Graphical analysis of mixed gas responses of two sensors biased with 5 V and 12 V with (a) 0.3 ppm CO and 750 ppb NO 2 (b) 0.9 ppm CO and 450 ppb NO 2 (c) 0.9 ppm CO and 150 ppb NO 2 , and (d) 3 ppm CO and 450 ppb NO 2 .

Table I .
Comparison of the predicted composition of the gas mixture with the actual composition.