I. INTRODUCTION

CO₂ is one of the critical greenhouse gases. Concentrations of CO₂ in Earth’s atmosphere have risen rapidly from 315 ppm in 1958 to about 410 ppm today^[1]. It has important influence on the respiration of bio organisms and the photosynthesis of plants^[2]. Moreover, the excessively high indoor CO₂ content will induce inattention, headaches, nausea and other symptoms. Therefore, it is necessary to well monitor and control concentrations of CO₂ in rooms, and low-cost portable CO₂ sensors with high sensitivity and selectivity are desired^[3]. So far, many researches on various kinds of sensors and sensing materials have been conducted to detect CO₂. Nowadays, carbon-based materials, like CNT and graphene, have attracted many attentions because they can detect CO₂ even at room temperature^[4]. Most of reported sensors with carbon-based materials measure their resistance changes during target gas detection^[5-7]. However, they usually suffer significant drift in the sensing signals which greatly diminishes the sensing accuracy^[6,7]. Hence, further improvement is required to effectively reduce the drift and obtain reliable sensing results. In this paper, the sensing performance of an FET-type CO₂ gas sensor, which has a horizontal FG, is investigated at room temperature. The layer-by-layer PEI-coated SWNT (PEI-SWNT) sensing materials are formed by inkjet printing. I-V transfer characteristics and DC sensing properties are investigated. With PEI coating, I-V hysteresis of the sensor is significantly improved. Moreover, the drift of sensing signal is suppressed by applying a pulse measurement.

II. EXPERIMENTAL

The structure of the proposed FET-type CO₂ sensor is shown in Fig. 1. Fig. 1(a) and Fig. 1(b) show the microscopic images of sensors with printed SWNT network and PEI-SWNT network, respectively. Fig. 1(c) shows the microscopic image of the MOSFET platform. Fig. 1(d) and Fig. 1(e) show the 2D schematic cross-sectional views of sensors cut along line A-A’ in Fig. 1(c), which have random SWNT network and PEI-SWNT network, respectively. The interdigitated FG and CG are formed horizontally to increase the coupling ratio between them. In this paper, pMOSFET sensor platform is adopted. The length and width of the channel of the MOSFET are 2 μm and 2.4 μm, respectively. The detailed fabrication processes of the sensor platform can be found in our previous work^[8] and main fabrication steps are described as follows. First, the active region of the MOSFET is defined. A 550-nm-thick isolation oxide is grown by local oxidation of silicon (LOCOS) technique. After growing 10-nm-thick gate oxide, an n⁺-doped poly-silicon FG is formed and patterned. Next, the source and drain regions are created by ion implantation process. The SiO₂ (10 nm)/Si₃N₄ (20 nm)/SiO₂ (10 nm) (ONO) insulator layers are formed to cover the whole substrate for preventing any contaminations. After that, stacked metal layers of Ti (20 nm)/TiN (30 nm)/Au (30 nm) are deposited consecutively to serve as the electrodes. Finally, an SU-8 passivation layer is formed and patterned to only expose the interdigitated CG-FG area and the electrode contact pads.

Fig. 1. (a) and (b) show the microscopic images of the sensors with printed only SWNT network and PEI-SWNT network, respectively, (c) shows the microscopic top view of the MOSFET platform, (d) and (e) show the 2D schematic cross-sectional views cut along line A-A’ in (c). (d) is the sensor with only random SWNT network and (e) is the sensor with PEI-SWNT network.

The sensing material, PEI coated SWNT, is formed layer by layer using ink-jet printing process. Semiconducting SWNT ink (1mg 98%) is purchased from NanoIntrgris (Canada) and printed on the platforms without further processing. Then the device is annealed under vacuum at 250 ºC for 3 hours to evaporate the solvent. Since the ink having a viscosity of 1 to 10 cps is suitable for inkjet printing^[9], when producing PEI ink, a Wt. 50% branched PEI aqueous solution (number average molecular weight 1200) purchased from Sigma-Aldrich (USA) is diluted to 10^-3 mol/L with deionized water. As a result, the viscosity of the solution is reduced. After printing the PEI, the sensor is heated at 60 ºC for 2 hours and then cooled in air for 1 hour. The CO₂ sample gas is prepared by mixing 1% CO₂ diluted in N₂ with dry synthetic air (20 vol. % of O₂ and 80 vol. % of N₂). The sensing characteristics are tested by switching between the sample target gas and the reference gas. A humid gas sample is obtained by running the dry gas through a bubbler filled with deionized water. All electrical measurements are carried out with an Agilent B1500A at room temperature.

III. RESULTS AND DISCUSSION

Fig. 2 shows double-sweep DC I-V curves of the SWNT sensor and the PEI-SWNT sensor at room temperature. The SWNT sensor has larger I-V hysteresis than the PEI-SWNT sensor. According to literature^[10], as the SWNTs contact silicon dioxide, a great number of traps are generated at the interface and induce a significant I-V hysteresis due to charge trapping and detrapping. Whereas, by coating the SWNT with PEI, nanotubes are wrapped and dispersed by PEI^[11,12]. It reduces the contact of SWNT network with the oxide, resulting in less interface traps. Therefore, the I-V hysteresis is reduced.

Fig. 2. Double-sweep DC I-V curves of the SWNT sensor and the PEI-SWNT sensor at room temperature.

The FET sensor platform can be electrically programmed or erased by applying pulses to the CG because it has an FG that can store the charge. As a result, the threshold voltage of the sensor can be calibrated. Fig. 3(a) and Fig. 3(b) show the double-sweep pulsed I-V curves of the programmed SWNT sensor and PEI-SWNT sensor, respectively. The $V_{th} s$ of both sensors are shifted to positive direction after programming compared to the results in Fig. 2. The pulse pattern for measuring I-V is shown in the inset, where the pulse width and period are 50 μs and 0.1 s, respectively. Pulsed gate biasing limits the time for charge trapping and detrapping, resulting in the suppression of I-V hysteresis. In Fig. 3, the I-V hysteresis of the PEI-SWNT sensor is removed by pulse measurement but the I-V hysteresis of the SWNT sensor still exists. By coating SWNT network with PEI and applying the pulse measurement, the I-V hysteresis in the sensor is nearly eliminated.

Fig. 3. Double-sweep pulsed I-V curves of the SWNT sensor and the PEI-SWNT sensor at room temperature after programming.

Fig. 4 shows the dry CO₂ sensing properties of the PEI-SWNT sensor measured by applying DC bias at room temperature after one day (Fig. 4(a)) and after 2 weeks (Fig. 4(b)) after depositing PEI coated SWNTs on the sensor platform. In Fig. 4(a), the drain current decreases when the sensor is exposed to CO₂ target gas. During the sensing measurement, as CO₂ acts as a kind of reducing gas^[13,14], the adsorbed CO₂ molecules transfer electrons to SWNTs, reduce the work-function of the sensing material, and result in the decrease of the absolute drain current. However, after two weeks, the sensor has no response to dry CO₂ gases. This may be attributed to the effect of moisture on CO₂ detection. It seems that the water left in the sensing material is fully evaporated within two weeks and the response to CO₂ is degraded. Therefore, it is necessary to investigate the effect of moisture on CO₂ gas detection.

Fig. 4. Dry CO₂ sensing properties of the PEI-SWNT sensor measured by applying DC bias at room temperature after one day (a) and two weeks (b) after the sensing material is deposited.

Fig. 5 shows wet CO₂ sensing properties measured by applying a DC bias to the PEI-SWNT sensor with the sensing material formed two weeks before the measurement (the same device measured in Fig. 4). Before measuring, we humidify the sensing material by blowing humid air into the sensor for 3 hours. The relative humidity (RH) of the gas sample is set to be 30%. With the help of water vapor in the target gas and the humidification, the sensor responds to CO₂ again. The drain current decreases with the increase of CO₂ concentration. It indicates that the PEI-SWNT sensor requires moisture for the detection of CO₂ at room temperature. CO₂ gas and water vapor together react with the sensing material, change the threshold voltage of the sensor and decrease the drain current. Some research groups also reported the effect of moisture on the reliability of PEI-based sensing material. In literature^[15], PEI-PANI sensing material was investigated. As the dissolution and base-catalyzed hydration of CO₂ in the PEI layer require water, the response of the sensor to CO₂ is strongly dependent on the relative humidity^[15]. According to their results, at 10% RH, barely no response was observed at low concentrations of CO₂ ( < 2500 ppm), while at 35% RH, the response was dramatically increased (about 5% at 2500 ppm according to the figures). Note that, moisture places an important role for the detection of CO₂ not only when using PEI-based materials but also some other sensing materials, like metal oxides^[16,17], even though the sensing mechanisms are different. In the case of metal oxides, the water molecules will be pre-adsorbed on the surface of metal oxides forming the OH- group, so that CO₂ molecules can be absorbed with the help of the OH- group^[17].

Fig. 5. Wet CO₂ detection characteristics of the PEI-SWNT sensor measured by applying DC bias at room temperature. The relative humidity (RH) is 30%.

As shown in Fig. 4 and Fig. 5, there is drift in the drain current over time due to hysteresis. During the DC sensing measurement, traps in the sensing material and interface traps will be charged over time. It has a significant impact on the accuracy of CO₂ detection. Hence, a pulse scheme for transient measurement is introduced to obtain reliable measurement results without drift. The pulse scheme is shown in Fig. 6(a). An appropriate pre-bias ($V_{pb}$) and read bias ($V_{r}$) are applied sequentially to the CG. Note that when $V_{pb}$ is applied to the CG, $V_{DS}$ is set to 0 V simultaneously to remove unwanted power consumption. In read biasing period, both $Vr and $VDS are synchronized. During the read period, the drain current of the sensor will be read out and recorded as the sensing signal. Here, the width of $V_{pb}$ pulse ($t_{w}$) and $V_{r}$ pulse ($t_{r}$) are set to 1 s and 50 µs, respectively. Fig. 6(b) shows wet CO₂ (30% RH) sensing characteristics of the PEI-SWNT sensor obtained by applying the pulse scheme at room temperature. The drift is removed by the pulse measurement method. The response of the sensor to CO₂ gas is defined as

Fig. 6. (a) Voltage pulse scheme for measuring sensing characteristics, (b) Wet CO₂ detection characteristics of the PEI-SWNT sensor measured by applying the voltage pulse at room temperature.

where $I_{D_B}$ and $I_{D_G}$ represent the drain currents of the sensor exposed to the reference gas and the target gas mixture, respectively. We define response time ($t_{RES}$) as the rise time of |$I_{D}$| to 90% of its maximum value, and recovery time ($t_{REC}$) as the fall time to 10% of the difference between the maximum and reference currents. In Fig. 6, our sensor shows a response of 16%, a $t_{RES}$ of 81 s and a $t_{REC}$ of 202 s for 2000 ppm of CO₂ at room temperature. Table 1 lists the CO₂ sensing characteristics of some recently reported sensors based on various sensing materials^[17-20]. The information of the proposed sensor in this work is also included for comparison, which shows that the proposed sensor has a reasonable response and competitive response and recovery times.

IV. CONCLUSIONS

In this work, an FET-type CO₂ sensor working at room temperature has been investigated. It was shown that the PEI coating reduces the traps at the interface between SiO₂ and SWNT network resulting in less hysteresis in I-V characteristics. In the I-V curves of the PEI-SWNT sensor, the hysteresis is greatly reduced, but the finite hysteresis under DC bias conditions leads to drift of the drain current, degrading the sensing characteristics. In transient sensing characteristics, drift was removed by applying voltage pulses. It was confirmed that moisture plays an important role in detecting CO₂ gas when PEI-SWNT is used as a sensing material. The response, response time, and recovery time of our FET-type PEI-SWNT gas sensor are 16%, 81 s and 202 s, respectively, for 2000 ppm CO₂ at room temperature.