Gas sensors have been widely utilized in various applications, such as automobile, industrial, medical, agricultural, and environmental applications (Gas Sensor Applications, 2021) (SST Sensing Ltd., 2021). For instance, gas sensors could be installed in living rooms, bathrooms, kitchens, etc. for environmental monitoring to realize smart homes for improvement in the quality and safety of our daily lives. Gas sensors could also be employed to monitor the air quality of public and private domains. Thus, gas sensors are considered as a promising component for internet of things (IoT) applications and have the potential to be integrated onto smart phones and devices. Many gas detection technologies have been developed, such as electrochemical (Wan et al., 2018) (Paul et al., 2020), calorimetric (Bartlett and Guerin, 2003), optical-based (Wang et al., 2021) (Chang et al., 2018) (Jorge et al., 2004) (McDonagh et al., 2002), and acoustic-based (Arsat et al., 2009) devices. Among the aforementioned sensing mechanisms, optical detection technologies (including infrared and chemi-luminescence approaches) have the advantages of higher sensitivity and faster response. However, bulky size and relatively high cost are two concerns for optical sensing technologies.

The standard complementary metal-oxide semiconductor (CMOS) processes are mature fabrication technologies offered by many foundries. To date, the CMOS processes have been extended to implement micro-electro-mechanical-systems (MEMS) chips by adding several additional fabrication steps (post-CMOS processes, such as the wet and dry etching on thin films and the Si substrate) (Fedder, 2005) (Fang et al., 2021). In this regard, the monolithic integration of the integrated circuit (IC) and sensing components on a single chip can be achieved (Tzeng et al., 2016) (Lee et al., 2021a). This CMOS-MEMS technology exhibits various advantages, for instance, multiple metal and dielectric layers are available to offer the design flexibility for MEMS structures and electrical routing; and many sensing mechanisms can be realized by the inherent materials of the CMOS process such as the metal/dielectric films, and boron/phosphorus-doped poly-silicon. Various miniaturized environment sensors and sensing hubs have been demonstrated using the CMOS-MEMS approach (Lin et al., 2020), including the gas sensors in Chien et al. (2022). Thus, CMOS-MEMS is a promising technology to achieve the size and cost reduction requirements for gas sensors.

This study extends the concept in Lee et al. (2017); Lee et al. (2019) to adopt the CMOS-MEMS technology to design and implement optical gas sensors. Various key components such as photo-sensors, mechanical structures, and temperature sensors will be integrated on the sensing chip through CMOS-MEMS processes. Based on the TSMC (Taiwan Semiconductor Manufacturing Corp.) 0.35 µm 2P4M (two poly-Si layer and four metal layers) standard processes, the CMOS chip is designed and implemented. The in-house post-CMOS micromachining processes are also developed to realize the gas sensing chip. In addition, this study will also exploit the method of heterogeneous integration to bond an LED (light emitting diode) chip as the light emitting source on the CMOS-MEMS gas sensing chip, through which the size reduction of the optical gas sensor can be achieved. Sensing materials are prepared to realize the fluorescence quenching-based optical gas sensors. To further enhance the performance of the CMOS-MEMS chemi-luminescence gas sensor, this study also exploits the solution in Lee et al. (2021b) to encapsulate the CMOS-MEMS sensing chip by using light reflectors. Thus, the light emitted from the LED chip is reflected and redirected onto the CMOS-MEMS chip by an encapsulated optical reflector to increase the sensitivity and reduce the power consumption for the presented sensor.

The presented sensing chip is designed based on the design rules of standard CMOS processes offered by the foundry. The in-house post-CMOS micromachining and assembly processes are also employed to implement the presented gas sensor shown in Figure 1. As illustrated in Figure 1A, the presented gas sensor consists of the gas sensing chip attached to the PCB (printed circuit board), the blue LED bonded on the central stage of sensing chip, and the light reflector. This study presents and discusses the hemispherical shell and flat plate light reflector designs. The sensing chip consists of a cavity array filled with gas sensing films, the photo-sensors (photodiode element) embedded at the bottom of each cavity, and the RTD (resistance temperature detector). Furthermore, the trenches between the cavity and LED are designed to prevent the overflow of gas sensing materials during dispensing. The cross-section AA’ in Figure 1B indicates that the cavity array could be filled with different sensing films to detect multiple gases. As depicted by the cross-section BB’ in Figure 1C, the photo-sensor will generate the photo-current I ₀ when the LED light is incident on the reference cavity (cavity with no sensing film). An additional photo-current I _f will be introduced as the cavity is filled with the sensing material used in this study (sensing cavity). The photo-current I _f is contributed from the fluorescence of the sensing film. When the gas to be detected is absorbed by the sensing film, fluorescence quenching occurs, causing the photo-current I _f to drop to I _f ^’ . Therefore, the gas to be detected is determined by the photo-current I _f ^’ which can be measured by subtracting the sensing photo-current (I ₀ +I _f ^’ ) and the reference photo-current (I ₀ ).

This study exploits the principle of fluorescence quenching (Jorge et al., 2004) for gas sensing. In this approach, molecules of sensing material will transit from the ground state to the excited state when absorbing a given light. As molecules of sensing material transit back to the ground state, radiation (i.e., fluorescence) will be emitted. The fluorescence intensity will be decreased if the sensing material absorbed the gas to be detected. Thus, the gas concentration can be determined by the variation of fluorescence intensity. Accordingly, the gas to be detected acts as the quencher to inhibit the intensity of fluorescence and operates according to the Stern-Volmer relation (Lakowicz, 2006) (Chu and Lo, 2008) F 0 F = 1 + K [ Q ] (1) where F ₀ and F are the fluorescence intensities, respectively, for the absence and the presence of gas, (Q) is the gas (quencher) concentration, and K is the Stern-Volmer coefficient. Thus, as the fluorescence intensities are determined by the measured photo-current, the gas concentration can be extracted from Eq. 1. Note that photo-sensors in Figure 1 are realized using the Si substrate with doped N⁺/P⁺ inherent in the CMOS process. According to Olson et al. (2003), the upper cut-off wavelength λ _g of the photo-sensor can be expressed as λ g = h ⋅ c E g × 10 9 = 1.24 E g (2) where E _g is 1.12 eV in this study, h is the Planck’s constant, and c is the speed of light (2.998 × 10⁸ m/s). Thus, the upper cut-off wavelength λ _g is 1.11 µm. Moreover, sensing materials are fluorophore compounds which could emit fluorescence upon light excitation. In applications, O₂ and CO₂ detection is demonstrated by the presented gas sensors, and C₃₀H₂₄C_l2N₆Ru·6H₂O and C₁₆H₇Na₃O₁₀S₃ will be used as the sensing materials. Since the sensing materials have the light excitation wavelengths (λ _exc ) of 470 nm (for O₂) and 460 nm (for CO₂) (Jorge et al., 2004) (Nivens et al., 2002), the blue LED with a wavelength of 445–455 nm is selected. In addition, the fluorescence emitted from sensing materials has the wavelengths (λ _emi ) of 610 nm (for O₂) and 520 nm (for CO₂). These wavelengths fulfill the requirement of being smaller than λ _g .

The light source of the blue LED is mainly top-emitting (Chu et al., 2015) (Liou and Tsou, 2018). To enhance the sensing signal, an optical reflector (two different designs) is employed to cover the sensing chip through the packaging process. Thus, the light emitted from the top-side of the LED could be reflected back and redirected to the sensing chip and photo-sensors. This study employs the commercial software TracePro to predict the light path and intensity. To compare the light intensity and distribution of the designs with and without covers, as depicted in Figure 2, the same incident light emitted from the LED is established in simulation models. The optical reflector has a diameter of 20 mm (to enable manually assembly). Simulations in Figures 2A–C show the light intensity distribution on the sensing chip for these three designs. With the assistance of the optical reflector, the designs in Figures 2B,C have higher light intensity in the regions of photo-sensors. The results in Figure 2D further indicate that the maximum light intensities of these three designs within the photo sensing area are, respectively, 8.4 × 10⁷ lux (no light reflector), 1.07 × 10⁸ lux (hemispherical shell), and 1.82 × 10⁸ lux (flat plate). Thus, the proposed designs with the optical reflector could offer the advantage of improved responsivity.

Figure 3 shows the micromachining processes to implement the proposed gas sensor. Figure 3A depicts the chip fabricated using the standard 0.35 μm 2P4M CMOS processes by the TSMC. The stacking and patterning of multiple Al and SiO ₂ layers, and tungsten vias are observed. The sacrificial layers consisted of Al metal films; and tungsten vias were then removed by the metal wet etching solution (H₂SO₄ and H₂O₂) (Tsai et al., 2009). After that, the structures to act as the cavity array and the stage for LED mounting were realized, as illustrated in Figure 3B. The photo-sensors and RTDs are also exhibited in this chip. As depicted in Figure 3C, the dielectric film was removed by reactive ion etching (RIE) to expose bonding pads for electrical routing of photo-sensors and RTDs. The blue LED chip was then assembled on the stage, as indicated in Figure 3D. In this study, the LED was manually placed on the CMOS-MEMS chip and then bonded by glue. Note that the LED could be bonded on the CMOS-MEMS chip by using the mature pick-and-place technology provided by the commercial packaging houses. The wire bonding between the sensing chip and the PCB board was performed in the process of Figure 3E, and then the sensing chip was deposited with sensing material and baked at the 40°C for 12 h, as show in Figure 3F. Finally, the fabrication and bonding processes for two different optical deflector designs are illustrated in Figure 3G. As displayed in Figure 3Gi, the hemispherical shell optical reflector was bonded to the circular groove on the PCB to cover the fabricated sensing chip. The hemispherical shell plastic optical reflector was fabricated using the 3D printer and then evaporated with the metal film on its inner surface to enhance the light reflection. As shown in Figure 3Gii, the flat plate optical reflector was manually bonded to the PCB through a spacer to cover the fabricated sensing chip. The flat plate optical reflector uses the photolithography process to define the pattern of the reflector area then etched by deep RIE (Lee et al., 2021b).

Figure 4 shows micrographs of a typical fabricated sensing chip. The optical micrograph in Figure 4A displays the chip fabricated by the CMOS processes of TSMC. The scanning electron microscopy (SEM) micrograph in Figure 4B depicts the sensing chip after the post-CMOS metal wet etching. The inset zoom-in micrographs, respectively, show the cavity and the RTD routing at the bottom of the cavity. The SEM micrograph in Figure 4C exhibits the sensing chip with the blue LED bonded at the center stage. The micrograph in Figure 4D indicates the sensing chip with sensing materials dispensed in the cavity. The left micrograph in Figure 4E shows the hemispherical shell of different diameters prepared using the 3D printer, and the right micrograph in Figure 4E depicts the hemispherical shell coated with metal film for the light reflector. The flat plate reflector is shown in Figure 4F. Figure 5 shows a typical packaged sensing chip as the device under test (DUT). As displayed in the left micrograph of Figure 5A, alignment marks on the PCB were exploited to ensure the sensing chip was placed at the right location. The ventilation holes and circular groove on the PCB are, respectively, used to input the test gases and to mount the hemispherical shell optical reflector. The right micrograph in Figure 5A shows that both the sensing chip and the blue LED are electrically connected to the PCB by wire bonding. Micrographs in Figure 5B show the on/off states of the blue LED after the sensing chip is covered with the hemispherical shell optical reflector (20 mm in diameter). Micrographs in Figure 5C show the on/off states of the blue LED after the sensing chip is covered with the flat plate optical reflector (acrylic as the spacer). To show the operation of blue LED, parts of the optical reflectors were removed. In this study, the footprints of the sensing chip and the blue LED are, respectively, 2.48 × 2.68 mm² and 890 × 890 μm². The dimension of the circular groove in Figures 5A,B needs to match with the size of the hemispherical shell optical reflector.

Finally, the sensing materials were prepared using the standard sol-gel process. The precursor of sol-gel is tetraethyl orthosilicate (TEOS). As shown in Figure 6A, after dissolving TEOS in ethanol (solvent), ammonia was added to form the porous silica (Rao et al., 2005). Finally, the sensing materials were coated on the surface of the silica particles. The micrograph depicts the porous silica prepared by the sol-gel process. The diameter of these silica particles is near 100 nm. In order to obtain material sensitive to O₂, the silica was coated with C₃₀H₂₄C_l2N₆Ru·6H₂O so that λ _exc and λ _emi were, respectively, 470 and 610 nm (Jorge et al., 2004). Moreover, the silica coated with C₁₆H₇Na₃O₁₀S₃ was prepared for CO₂ detection, and λ _exc and λ _emi were, respectively, 460 and 520 nm (Nivens et al., 2002). Thermal analysis was performed to confirm that the sensing material was properly coated on the silica surface. In this test, the temperature range is 30°C–1,000°C and has a rising rate of 10°C/min. Measurements in Figures 6B,C and Figures 6D,E, respectively, show the thermogravimetric analysis (TGA) and the differential scanning calorimetry (DSC) analysis. According to TGA measurements in Figure 6B, the silica pyrolysis temperature changes from 581.4 °C to 584.3°C after the silica was coated with the O₂ sensing material. Moreover, TGA measurements in Figure 6C indicate the silica pyrolysis temperature changes from 581.4 °C to 553.6°C after the silica was coated with the CO₂ sensing material. The drift of silica pyrolysis temperature demonstrates that a force is generated between the sensing material and silica (Jaroniec et al., 1997) (Wu et al., 2020). Note that different sensing materials will generate intramolecular or intermolecular forces (Wu et al., 2020). For instance, the O₂ sensing material (ionic compound) will generate an ionic bond force with the silica surface. The CO₂ sensing material (molecular compound) will generate van der Waals forces and H-bonding with the silica surface. According to DSC measurements in Figure 6D, the energy peak of the silica drifts from 581.8 °C to 584.5°C after the silica was mixed with the O₂ sensing material. DSC measurements in Figure 6E depict the energy peak of silica drifts from 581.8 °C to 553.3°C after the silica was mixed with the CO₂ sensing material. These drifts are also caused by previously mentioned forces on the material (Jaroniec et al., 1997) (Wu et al., 2020). Since these two sensing materials have different bonding conditions with the silica surface, their energy peak drifts are also different. As a result, measurements show that the fluorescent materials have been coated onto the surface of silica effectively.

This study performed measurements to evaluate the sensing performances of the O₂ and CO₂ sensors with and without optical reflectors. Moreover, the temperature influence of the proposed gas sensor has also been characterized and investigated. Firstly, this study established the measurement setup in Figure 7A to characterize the RTD temperature sensing unit. The temperature controller under the fabricated sensing chip is used to specify a constant ambient temperature (at the range of 30°C–120°C) for testing. The source meter is used to measure the resistance of the RTD. Measurements in Figure 7B show the variation of resistance change with the ambient temperature. The results indicate that the sensitivity of the temperature is 0.07%/°C at the temperature range of 30°C–120°C. Secondly, the characteristics of the blue LED and photo-sensor used in this study are characterized. The spectrometer is employed to measure the wavelength of the LED. Typical measurement results in Figure 8A show that the central wavelength of LED is 452 nm. The overall emission wavelength covers the λ_exc (470/460 nm) of the sensing material. Thus, this commercially available blue LED fulfills the requirements in this study. Measurements in Figure 8B present the relationship between the reference photo-sensor (photo-sensor at the reference cavity) output current I ₀ and the LED driving current at room temperature for gas sensors with and without an optical reflector. The input current of the LED ranges from 10 to 250 mA. The results indicate that, for a given output current I ₀ from the reference photo-sensor (160 μA), the gas sensor without an optical reflector needs a higher input current (100 mA) on the LED, as compared with those for the gas sensor with optical reflectors (20 mA for the flat plate reflector and 70 mA for the hemispherical shell reflector). In other words, the power consumption of the presented sensors can be reduced by adding the light reflector.

The test setup in Figure 9 is established to characterize the performance of O₂/CO₂ sensors at different gas concentrations. The specimen is prepared and tested inside an opaque chamber with a standard O₂, CO₂, and N₂ gas analyzer. The mass flow controllers (MFCs) are employed to control the gas flow and further control the gas concentration. The LED is driven through a constant current source from a power supply. The source meters are employed to measure output currents from the sensing chip at different O₂/CO₂ gas concentrations. Measurements in Figure 10 show the variation of the output current with gas concentrations for sensors with and without optical reflectors. Note that, based on the results in Figure 8B, this study specified the same I ₀ for each gas sensor during tests. Thus, the input currents of the LED for gas sensors of three different designs are, respectively, 20 mA (with a flat plate reflector), 70 mA (with a hemispherical shell reflector), and 100 mA (without a reflector). In addition, tests were performed under the humidity condition of 60%–65% RH and room temperature. Measurements indicate that the increase of gas concentration will enhance the fluorescence quenching reaction, so as to decrease the fluorescence intensity. As depicted in Figure 10A, the sensitivities of O₂ gas concentration sensing (within the O₂ gas concentration range of 14%–50%) are 0.12 μA/% (O₂/N₂) and 0.24 μA/% (O₂/N₂), respectively, for sensors with hemispherical shell and flat plate optical reflectors. The sensitivity for the design without an optical reflector is only 0.023 μA/% (O₂/N₂). Moreover, as shown in Figure 10B, the sensitivities of CO₂ gas concentration sensing (within the CO₂ gas concentration range of 14%–50%) are 0.19 μA/% (CO₂/N₂) and 0.32 μA/% (CO₂/N₂), respectively, for sensors with the hemispherical shell and the flat plate optical reflectors. The sensitivity for the design without an optical reflector drops to 0.12 μA/% (CO₂/N₂). The results indicate, even when each sensor was normalized at the same I ₀ during tests (i.e., LED on the gas sensor without a reflector provides much higher light emission), the designs with optical reflectors still have much higher sensitivities. The redirect of the light from the LED by the reflector could be the reason for such sensitivity enhancements. As reported in Coppel et al. (2015), the incidence angle of the light source could change the intensity of the detected fluorescence. Measurements in Figure 11 exhibit the responses of the gas sensor at different ambient temperatures (−25°C–−60°C). As the ambient temperature increases, the response and sensitivity of the gas sensor decreases. For example, the sensitivities of the CO₂ sensor are 0.19 μA/% and 0.18 μA/%, respectively, at 25°C and 60°C. Thus, a monolithically integrated RTD temperature sensor can be exploited for temperature compensation. Measurements in Figure 12A indicate that O₂ gas was injected into the test chamber at t = 0 s (the background gas is N₂) to cause the fluorescence quenching of the O₂ sensor, and the induced current I_f ^’ of the O₂ sensing unit is dropped. However, the induced current of the CO₂ sensing unit remained the same. The results show that the presented CO₂ sensor has no crosstalk with the O₂ gas. Moreover, measurements in Figure 12B demonstrate that the presented O₂ sensor has no crosstalk with the CO₂ gas. In summary, comparisons of the presented CMOS-MEMS gas sensor with other existing optical gas sensing devices are summarized in Table 1.

In this work, micro fluorescence quenching gas sensors comprised of a CMOS-MEMS sensing chip, LED chip, optical reflector, and sensing materials are designed and implemented. By using the TSMC standard CMOS and the in-house post-CMOS processes, the mechanical structures, temperature sensor, and photo-sensors are fabricated and monolithically integrated on the CMOS-MEMS sensing chip. The LED chip is selected to provide the required wavelength for sensing. The optical reflector can reflect and redirect the light emitted from the blue LED to the sensing cavity, which can effectively reduce the power consumption and can also improve the sensitivity. The vertically integrated temperature sensor on the gas sensor enables on-site temperature monitoring for sensing signal compensation. The O₂ and CO₂ gas sensors are implemented and tested to demonstrate the feasibility of the presented designs. The sensing materials are, respectively, mixed with C₃₀H₂₄C_l2N₆Ru·6H₂O and C₁₆H₇Na₃O₁₀S₃ for O₂ and CO₂ detection. Moreover, two different light reflectors are, respectively, demonstrated using the micromachined Si flat plate and the 3D-printed polymer hemispherical shell. Note that the roughness of the reflector will cause the diffusion of the reflected light, and such an issue could be improved by using other processing methods such as injection molding. Measurements indicate that the designs with an optical reflector (hemispherical shell and flat plate) show an enhancement in sensitivity for sensor detection and also provide the benefit of lower power consumption for the LED. In addition, the flat plate reflector has a better performance (0.24 μA/% in O₂/N₂ and 0.32μA/% in CO₂/N₂) compared with the design with the hemispherical shell optical reflector (0.12 μA/% in O₂/N₂ and 0.19 μA/% in CO₂/N₂). In conclusion, the presented heterogeneous integration of CMOS-MEMS and LED chips could effectively reduce the size of optical gas sensors. The performance and power consumption of the optical gas sensor can be further improved by using the light reflector.