ppb level detection of NO2 using a WO3 thin film-based sensor: material optimization, device fabrication and packaging

In this study, we have investigated the thickness-dependent nitrogen dioxide (NO2) sensing characteristics of a reactive-ion magnetron sputtered tungsten trioxide (WO3) film, followed by morphological and electrical characterizations. Subsequently, the sensing material was integrated with an MEMS platform to develop a sensor chip to integrate with electronics for portable applications. Sputtered films are studied for their sensing performance under different operating conditions to discover the optimum thickness of the film for integrating it with a CMOS platform. The optimized film thickness of ∼85 nm shows the 16 ppb lower limit of detection and 39 ppb detection precision at the optimum 150 °C operating temperature. The film exhibits an extremely high sensor response [(Rg − Ra)/Ra × 100 = 26%] to a low (16 ppb) NO2 concentration, which is a comparatively high response reported to date among reactively sputtered films. Moreover, this optimum film has a longer recovery time than others. Thus, an intentional temperature overshoot is made part of the sensing protocol to desorb the NO2 species from the film surface, resulting in full recovery to the baseline without affecting the sensing material properties. Finally, the optimized film was successfully integrated on the sensor platform, which had a chip size of 1 mm2, with an inbuilt micro-heater. The minimum power consumption of the microheater is ∼6.6 mW (∼150 °C), which is practically acceptable. Later, the sensor device was packaged on a Kovar heater for the detailed electrical and sensing characterizations. This study suggests that optimization of the sensing material and optimum operating temperature help to develop a highly sensitive, selective, stable, and portable gas sensor for indoor or outdoor applications.


Introduction
Air pollution is one of the emerging problems in our surroundings. Thus, strict regulations on the emission of toxic gases require fast and highly selective gas sensors capable of detecting the sub-ppm level of gases. Carbon monoxide and nitrogen dioxide are the major pollutants, which play a major role in the formation of ozone and acid rain. Frequent exposure to NO 2 levels higher than 53 ppb may cause an increase in respiratory illness. 1 Therefore, inexpensive as well as simple fabrication procedures to develop sensors with high sensitivity, stability, and durability are in demand nowadays.
Thin lms are more suitable for resistive-based gas sensors due to their high surface-to-volume ratio as the gas reaction is a surface phenomenon. Moreover, if the lm morphology has a porous structure, gas molecules can easily react with the whole volume through the pores; this enhances the sensitivity. There are mainly two approaches for the improvement of the sensor sensitivity and selectivity. The rst is the optimization of the sensing material growth/deposition conditions. [2][3][4] The second is to quantify the operating conditions, such as operating temperature and bias voltage, of the sensor. 5,6 In this study, the NO 2 response is monitored by varying the thickness of a WO 3 lm with the impulse mode of temperature operation. Some reports have reported the effect of lm thickness on sensor response. [7][8][9][10][11][12][13][14] It can be understood that by controlling the microstructure shape and size of the WO 3 lm, the ppb level detection of NO 2 can be achieved. [15][16][17][18][19][20] In addition, not only an optimum sensing layer thickness helps to achieve a high response to test a gas but also the sensor operating conditions play an important role to dene the overall sensor performance. In the past decade, WO 3 nanostructures with large surface-tovolume ratios have been considered for gas sensing applications. Flower-like WO 3 nanosheets, synthesized by calcining an acid-treated hydrothermal precursor, showed minimum 2 ppb NO 2 level detection at a 90 C operating temperature. 21 Wojcik et al. 22 studied the NO 2 response of a drop cast-synthesized WO 3 material and showed minimum 10 ppb NO 2 detection at a 300 C operating temperature. Triple-shelled WO 3 spheres, prepared by ultrasonic spray pyrolysis, showed the minimum detection of 50 ppb NO 2 at 100 C, reported by Kim et al. 23 A fully gravure-printed WO 3 -PEDOT:PSS nanocomposite-based NO 2 sensor on a polyimide foil has been explored to detect minimum of 50 ppb NO 2 at room temperature, reported by Lin et al. 24 Recently, Zhang et al. 25 reported 10 ppb NO 2 detection at the 120 C operating temperature using Fe-doped WO 3 nanostructures synthesized by the hydrothermal method. Shen et al. 26 have concluded that a Au-doped hierarchical WO 3 microsphere nanostructure, prepared using the hydrothermal method, is capable of detecting a 1 ppm NO 2 concentration at a 50 C operating temperature. Although these nanostructures show a high response to NO 2 in the sub-ppm concentration range, they are prepared through chemical route processes such as hydrothermal, drop cast, spray pyrolysis, which are not CMOS compatible. Although many studies have been reported on the physical deposition of a WO 3 lm for NO 2 detection, [27][28][29][30][31] no study has been reported on the realization of a sensor product from the optimization of a sensing lm to the integration of the lm with a MEMS platform. Thus, in this study, the sensing lm of WO 3 is optimized by varying the lm thickness using a reactive-sputtering technique, followed by their sensing characterization to realize the best optimum lm for highly selective response towards NO 2 at the sub-ppb level. Later, using an MEMS platform with a low power integrated microheater, a large-scale production of a sensor chip, with a size of 1 mm 2 , is developed with integration of the optimized lm. Packaging of the sensor chip on the header using a wire bonding process is conducted for easy integration with the electronics for real-time monitoring of sub-ppm levels of NO 2 in air. The packaged sensor is highly sensitive and selective towards NO 2 as further investigated.
To fabricate the highly sensitive and selective NO 2 sensor device, the WO 3 lms of different thicknesses, deposited by an rfmagnetron-sputtering technique, were extensively investigated by sensing the characterizations at various operating temperatures ranging from 100 C to 300 C. Later, the optimized lm was integrated with a CMOS compatible sensor platform, which was integrated with a micro-heater for the on-chip operation of the sensor device. In brief, sensor fabrication is mostly carried out with the help of photolithography and sputtering, followed by dry etch processes in reactive ion etching (RIE) and a deep reactive ion etching (DRIE) tool. The packaging is conducted on a Kovar header, followed by wire bonding for the easy handling of the sensor device, a prototype NO 2 sensor.

Experimental
The conventional planar rf-magnetron sputtering system with a 3 00 target of tungsten in ambient oxygen is used to sputter WO 3 lms on top of the inter-digitated electrodes (IDEs). IDEs are patterned using photolithography, followed by Ti/Pt (10/80 nm) sputtering and a li-off process, as shown in Fig. S1. † The distance between the target and substrate is maintained at 8.5 cm. An Ar gas ow of 300 sccm was maintained in the chamber by a mass ow controller, and the deposition pressure was kept at $6.3 mTorr. Before deposition, the chamber was evacuated to a pressure of the order of 10 À6 Torr, and then, a presputtering process was conducted to clean the target surface. The lm thickness is controlled by adjusting the deposition time. The calculated average deposition rate of the WO 3 lm is $3.43 nm per minute, as shown in Fig. S2. † Film thickness was measured by a Dektak surface proler and cross-sectional scanning electron microscopy (SEM). Surface roughness and grain size were analysed by atomic force microscopy (AFM). Surface morphologies were characterized by eld emission scanning electron microscopy (FE-SEM). Finally, the as-deposited lms were subjected to NO 2 sensing characterization at different operating temperatures (100-300 C) and gas concentrations.

Results and discussion
3.1. Structural and morphological characterizations of the WO 3 lms X-ray photoelectron spectroscopy (XPS) is a widely used technique to investigate the chemical composition of thin lms. The obtained XPS data of the WO 3 lms is shown in Fig. S3. † The study concludes that the sputtered lms are pristine since there is no peak other than the characteristic peak for W and O. The doublet was observed at a binding energy of 33.9 eV and 37.0 eV corresponding to W 4f 7/2 and W 4f 5/2 , respectively, from the core-level spectra of W 4f , see Fig. S3. † This is in good agreement with other reported results. 32,33 Therefore, it is clear that the W oxidation state is +6, which conrms the WO 3 phase formation of the lms. In Fig. S3(b), † the peak of O 1s core level is found at 530.87 eV, which is quite close to the value reported in the literature. 34 Surface morphologies of the as-deposited lms of different thicknesses were studied using FE-SEM (Fig. 1). The topography of the lms shows that the lms have a porous structure with some black holes or zones on the surface. Films of lesser thickness have some minor cracks on the surface that provide direct conduits for gas molecules to ow inside the lm; this may inuence the sensor performance. 35 The WO 3 lm of thickness $85 nm has a smaller grain size and higher surface roughness, as conrmed by the AFM analysis of the lm grain size, as well as the surface roughness data, as shown in Fig. 2.

Electrical characterization of the WO 3 lms
The graph of the change in the electrical resistance of lms with temperature in the range of 25-400 C is shown in Fig. S4(a). † The resistance of the lms, except the 37.3 nm, 113 nm, and 154.9 nm lms, decreases rapidly with temperature up to 125 C and thereaer begins to fall slowly up to 225 C and aerward again decreases very slowly up to 400 C. The overall trend of the lm resistances indicates the semiconducting nature of the sputter-deposited WO 3 lms. In fact, two competing processes of thermal excitation of electrons and oxygen adsorption occur simultaneously. In the beginning, the decrease in the lm resistance with temperature is because of the thermal excitation of electrons that dominates over the oxygen adsorption process. The slow decrease of lm resistance in the temperature range from 125 C to 225 C is attributed to adsorption of atmospheric oxygen on the lm surface. Herein, oxygen adsorption is not more favourable for the WO 3 lm; thus, the resistance of lms decreases throughout the temperature range. A similar explanation has been reported by other authors. [36][37][38] The inverse absolute temperature of the electrical resistance of the lms is shown in Fig. S4(b). † Films exhibit two activation energies in different temperature ranges. The activation energy is calculated using the following relation: where DE is the activation energy, R o is a constant, k is the Boltzmann constant, and T is the absolute temperature. The activation energies thus obtained are listed in Table S1, † which indicates two energies levelsone deep and one shallow near the bottom of the conduction band in the band-gap.

Nitrogen dioxide (NO 2 ) sensing characteristics of the WO 3 lms
Room-temperature deposited lms were tested several times at each operating temperature to guarantee the reliability of the sensing data. The sensor response (S) of the lm is dened as     the ratio of change in lm resistance upon exposure to test gas to the lm resistance in air (at same operating temperatures) and is given by the equation where DR is change in the resistance of the sensing lm before and aer exposure to the test gas and R is the initial resistance of lm under an air atmosphere. To measure the sensing characteristics of thin lms, the sensing setup used is shown in Fig. S5. † The lm was mounted in a gas calibration chamber. The gas chamber had the ability to connect to the target gas cylinder along with the synthetic air (80% nitrogen and 20% oxygen) cylinder to set the appropriate concentration of the target gas using mass ow controllers (MFCs). The relative humidity was observed to be $45% inside the gas chamber during measurements. The resistance of the WO 3 lm was found to increase on exposure to NO 2 gas due to the oxidizing nature of the gas. The sensing measurements were conducted under dry gas conditions. To determine the optimum thickness of the WO 3 lm for the maximum response to NO 2 , the gas sensing characteristics of different lms towards 0.9 ppm NO 2 were measured at different operating temperatures ranging from 100 C to 300 C, as shown in Fig. 3. It is well known that a high response depends not only on the optimum lm thickness but also on the operating temperature. The present study concluded that the WO 3 lm thickness of $85 nm showed the highest response of $3102% to 0.9 ppm NO 2 concentration at 150 C, which was quite a low operating temperature. 13 To estimate the stability of the lm response towards NO 2 , the $85 nm thick lm is exposed multiple times to a 0.9 ppm NO 2 concentration to quantify the resistance change of the lm on each exposure. The lm exhibits degradation in response aer each exposure, as indicated by the dri observed in the baseline resistance of the lm. The lm is able to recover only $80% of the resistance, which is changed on exposure to NO 2 gas. This may be due to the accumulation of incompletely oxidized gas molecules on the lm surface. This results in an incomplete recovery of lm resistance upon switching to synthetic air (NO 2 exposure is off). To recover the sensor base line dri, the periodic shi to a higher temperature for a short duration is made the part of sensing protocol to desorb the gas molecules from the lm surface. 2,39 Thus, the temperature impulses of 50 C and 100 C of 50 seconds duration are implemented in between the sensing cycle, as shown in Fig. 4. From the initial two response cycles, it is clear that the recovery of lm resistance is poor on impulse of 50 C temperature, as can be seen from the obtained responses, as shown (red dots) in the inset of Fig. 4. This may be due to insufficient thermal energy for the gas molecules to desorb from the lm surface. However aer a temperature impulse of 100 C, the base  resistance is almost recovered, as shown (green dots) in the inset of Fig. 4. It can be concluded that the lm shows a higher dri in sensor response treated with the impulse of 50 C (red dots) as compared to the response dri in the case of a 100 C temperature impulse (green dots). This kind of temperature treatment for a short duration is really effective to obtain the reproducible sensor response. In conclusion, temperature pulse of 100 C is optimum to obtain the reproducible as well as the stable response towards NO 2 using the WO 3 thin lm.
To estimate the low order of detection (LOD), NO 2 gas concentration was tested from 16 ppb to 800 ppb at a 150 C operating temperature with an optimum impulse of 100 C temperature, as shown in Fig. 5. The lm was able to detect 16 ppb [(R g À R a )/R a Â 100 ¼ 26%] NO 2 concentration, which was comparatively low concentration than that reported in other studies. 40,41 Moreover, the WO 3 lm shows a linear response to different NO 2 concentration in the 16-800 ppb range and a detection resolution of 11 ppb for the optimum operating temperature (150 C) with help of impulsive mode of temperature. Theoretically estimated LOD is 1.6 ppb obtained from the linear t of response data of lm shown in inset of Fig. 5. The comparison of present study along with responses observed by other researchers using different nanostructures of WO 3 is shown in Table 1. [42][43][44][45][46][47][48][49][50][51] The present study on the WO 3 thin lm-based NO 2 sensor concludes that sub-ppb level NO 2 detection with high sensitivity and selectivity can be obtained by simple reactive-ion sputtered technique, a scalable process. The WO 3 lm selectivity towards NO 2 was tested in the presence of CO, CO 2 , SO 2 , and NH 3 gases at 150 C. The study clearly indicates the high selectivity of the lm towards NO 2 among other gas species, as represented in Fig. 6.
The present investigation of WO 3 thickness-dependent NO 2 characteristics suggest that a lm thickness of $85 nm is optimum to achieve a highly sensitive and selective NO 2 sensor, which signicantly shows the sub-ppb range detection with a quick response and recovery time. Furthermore, to realize the prototype NO 2 sensor, we fabricated a sensor device inbuilt onchip-integrated microheater to control the operating temperature of the WO 3 lm using MEMS surface micromachining processes, as explained hereinaer. The detailed optimization of the fabrication process of the sensor device is described elsewhere. 52

Fabrication of the NO 2 sensor device
A 4 00 wafer was cleaned prior to SiO 2 deposition rst by piranha solution and then dipped in hydrouoric acid followed by washing with DI water and drying in nitrogen. The front side SiO 2 (1 mm) is used to build the sensor device, and the back side SiO 2 (1 mm) is used as a mask for backside Si etching in deep reactive ion etching (DRIE). On top of SiO 2 , Ti/Pt (10/80 nm) is sputtered to a pattern microheater structure. Then, 200 nm PECVD SiO 2 is deposited on top of the microheater (200 mm Â 200 mm) to serve as an insulator between the micro-heater and sensing electrodes. Sensing electrodes of sputtered Ti/Pt (10/50 nm) are fabricated on top of the microheater. Later, as optimized, $85 nm thick WO 3 lm is deposited onto the sensing electrodes. Finally, bulk Si from the back side of the microheater is dry etched to form an air cavity to reduce the power consumption of the microheater. The deposition and lioff process of the WO 3 sensing material is conducted using sputtering and photolithography. The schematic of the fabricated sensor device is shown in Fig. 7(a and b). To estimate the WO 3 sensing layer thickness, a cross-sectional SEM image of the fabricated sensor chip is shown in Fig. 7(c), which indicates the sensing layer thickness of $88 nm on top of the sensor chip stacks. To check the response of the fabricated sensor device towards NO 2 gas, the measurement is obtained from the packaged sensor, as shown in Fig. 7(d and e).
The microheater characterization was conducted to calculate the heater power consumption to achieve different temperatures from the microheater, as shown in Fig. 8(a). The calculated temperature coefficient of resistance (TCR) is 1.35 Â 10 À3 C À1 . Initially, the gas sensitivity is measured under the xed (0.1 ppm) NO 2 concentration at different operating temperatures, ranging from 62 C to 228 C, to know the optimum operating temperature to achieve a high response, as shown in   8(b). The sensor device shows the high response of $74.6% at 157 C, which requires a power of $6.55 mW. To examine the repeatability of the fabricated sensor, the sensor is exposed multiple times to a xed NO 2 concentration (0.1 ppm), as shown in Fig. 9(a). The sensor shows an almost repeatable response, but the recovery of the sensor's base resistance is still an issue. Thus, to overcome this issue, the sensor was operated with an impulse mode of temperature by increasing the heater voltage for 20 seconds, which increased the operating temperature by $100 C. As a result, the sensor response is almost repeatable Fig. 9(b). Fabricated sensors are also exposed to different NO 2 gas concentrations from 0.1 ppm to 3 ppm under the same operating conditions. For a very low concentration range from 0.1 ppm to 0.5 ppm, the sensor shows a rapid change in response. However, for high concentrations, the increase in response is comparatively slow, as shown in Fig. 10. The rapid change at low concentrations may be because gas-molecules obtain enough thermal energy to react with the sensor surface; this leads to a fast reaction at the sensor sites. On the other hand, with an increase in gas concentration, the gas molecules may be covering the sensor surface very fast; this leads to a slow increase in response. Moreover, the fabricated NO 2 sensor shows a nearly linear response in the concentration range from 0.1 ppm to 0.5 ppm with a detection resolution of 100 ppb for the optimum operating conditions, as shown in the inset of Fig. 10. The results indicate that the sensor is capable of detecting a NO 2 gas concentration as low as 100 ppb. In fact, many models have been proposed to describe the sensitivity of the semiconducting metal oxide; thus, it can be represented empirically. 53 The lowest order of detection of the NO 2 sensor is 0.8 ppb, which is calculated by a linear t of the sensor response data in the concentration range from 0.1 ppm to 0.5 ppm, as shown in the inset of Fig. 10.
We have monitored the fabricated NO 2 sensor response characteristics to evaluate the sensor reproducibility and stability for a period of more than 6 months to estimate the sensor life. The as-fabricated sensor is found to be very stable during this period. Thus, we propose that the present sensor is a promising candidate for real-time monitoring of NO 2 gas in air.

Conclusions
In conclusion, lm surface morphology plays an important role in deciding the sensing characteristics of thin lm-based sensors. XPS analysis of sputter-deposited lms shows the desired chemical states. SEM images show that lms have a porous microstructure with small cracks, which helps to enhance the sensing reaction because of the deep interaction of gas molecules with the lm. The impulse mode of temperature is implemented successfully to produce a highly stable and reproducible sensor response. With these sensors, a detection limit of 16 ppb for NO 2 is achieved. This is the lowest detectable concentration with this pristine metal-oxide semiconductor to date. Sensors show high selectivity as well as sensitivity to NO 2 gas.  The NO 2 sensor device is fabricated successfully using an MEMS platform and tested under different operating conditions to evaluate the performance of the sensor. An impulse mode of temperature is found to be effective to recover the baseline dri in NO 2 sensor resistance. The choice of sensor elements on a single diaphragm exhibits fairly good crosssensitivity, long-term stability, as well as reproducibility towards NO 2 gas detection.

Conflicts of interest
There are no conicts to declare.