Humans are constantly developing technologies to pursue smart living. Currently, the Internet of Things (IoT) is among the most promising technologies for making our lives smarter. IoT refers to the networked interconnection of embedded devices that allows them to interact with each other, services, and people on a global scale [1]. Such level of connectivity can increase reliability, sustainability, and efficiency through the integration of every device for interaction via embedded systems to improve access to information [1,2]. Conceptual examples of IoT-based smart technologies are as follows. In houses, gas leakages generated in the kitchen are detected by smart gas sensors, and alerts are sent in the form of mobile messages via a wireless network. Simultaneously, self-inspection is performed to remove the leakage using a smart automation system based on IoT technology [3,4]. Depending on occupancy, the lights in a smart home are automatically turned on and off, and heating or air-conditioning systems operate to adjust the indoor temperature [5]. Moreover, through the monitoring of the home environment, power efficiency can be improved, and the well-being of residents can be enhanced [6]. In factories, various types of smart sensors, such as temperature, pressure, and vibration sensors, can contribute to automated process control to satisfy the quality and yield requirements of products [7,8]. Many concepts and research results on IoTbased smart technologies have been reported [3-11]. Furthermore, a considerable amount of cash flow is expected through the global market in the coming years for IoT [12]. For the development and realization of IoT technology, the development of sensor technology, which is one of the components of IoT system architecture and where the workflow of IoT systems begins, is critical [13]. These sensors detect external changes and transduce them into electrical signals. Subsequently, the electrical signals are conditioned by interface circuits, such as a noise filter and an amplifier, and digitally processed and transmitted to the data collection center or other devices through a wireless network [14]. Simultaneously, humans can monitor the transmitted data in real time using a user interface-based mobile application [14].
Sensor technology requires high sensitivity and high resolution. To improve sensitivity and resolution, extensive research on the morphological transformation or composition tuning of sensing materials has been conducted for a long time [15-17]. Recently, a novel electrical operation scheme was proposed to achieve high resolution without any transformation or tuning of the sensing materials which is complex and difficult to apply in device fabrication [18]. However, in the recent development of IoT sensor technology, reliability and high operating speed have gained increased importance over the sensing abilities described above [19]. Moreover, IoT sensors necessitate miniaturization and low power consumption for application to portable devices. CMOS technology can satisfy these requirements for IoT sensors [20]. In addition, CMOS technology is compatible with interface circuits and can reduce fabrication costs [20].
Humidity control technologies play an important role in human life. Accurate humidity monitoring and control is critical for maintaining a pleasant home environment. In factories and farms, the maintenance of a constant humidity level significantly affects the product yield. Accordingly, several studies on humidity sensors have been performed, but most of them have focused on resistive- [21,22] and capacitive-type [23,24] humidity sensors. Although resistive and capacitive sensors are inexpensive and easily fabricated, they have the disadvantages of a large size [21], output signal drift [22], and hysteresis [24]. Si field-effect transistor (FET)-type sensors are expected to be the most suitable candidates for overcoming these disadvantages owing to their scalability via the use of CMOS technology and compatibility with integrated circuits (ICs) and electrical control schemes which improve the sensing performance [18].
This study presents the sensing characteristics of a Si FET-type humidity sensor with a tungsten trioxide (WO3) sensing layer prepared using Si CMOS fabrication technology. The key process steps of the humidity sensor are explained, and the experimental setups for humidity generation and electrical measurements with pulse bias are described. The humidity-sensing characteristics are measured and analyzed based on the chemical reaction mechanism betweenWO3 and H2O. Subsequently, the chemicalto- electrical transduction that occurs in the FET part of the sensor during humidity sensing is explained comprehensively using schematic energy band diagrams. Finally, the transient drain current (ID) as a function of relative humidity (RH) is discussed.

 The humidity sensor is prepared using the conventional Si FET technology. An n-type Si (100) wafer is used as the substrate for the sensor because p-channel MOSFET is superior to n-channel MOSFET in terms of low-frequency noise [25]. Fig. 1(a) displays the top-view SEM image of the humidity sensor. The humidity sensor comprises two parts: sensing (top of Fig. 1(a)) and FET (bottom of Fig. 1(a)) parts. The sensing part comprises the controlgate (CG), the floating-gate (FG), and the WO3 film functioning as a humidity-sensing layer. The CG and FG are formed in an interdigitated structure and face each other with a gap of 500 nm. The $WO_3$ sensing layer partially covers them by filling the gap. The FET part comprises the FG, the source, the drain, and the Si active layer. Fig. 1(b) depicts a three-dimensional structure of the humidity sensor, viewed from above at an oblique angle. Figs. 1(c) and 1(d) present schematic cross-sectional views of the sensor along the FET channel width and length directions, respectively. The key fabrication process for the humidity sensor is described below. To define the channel region, a stack of pad oxide and silicon nitride is patterned on a bare Si wafer and a local field oxide is thermally grown. Subsequently, the pad oxide and silicon nitride are removed and a 10 nm thick gate oxide is grown back on the Si active. Thereafter, a 350 nm thick n+ poly-Si is deposited and patterned to form the FG. The source and drain (S/D) regions are defined by the self-aligned arsenic ion implantation. The wafer is covered with an oxide-nitride-oxide (O/N/O, 10 nm/20 nm/10 nm) stack to passivate the FET part of the sensor. The nitride layer of the O/N/O stack is known to prevent impurities from penetrating into the channel region [26]. The metal contact areas in the S/D regions are defined, followed by the formation of a Cr/Au metal stack for the source, drain, and CG electrodes. At the end, a 30 nm thick semiconducting $WO_3$ sensing layer is formed via radio frequency (RF) magnetron sputtering. The chamber pressure is 5 mTorr with Ar flow and the RF power is 230 W during sputtering.

  Fig. 2 shows a schematic diagram of the humiditysensing measurement system, which is composed of two parts: a humidity generator and a test chamber. Three dry $N_2$ gas lines controlled via a mass flow controller (MFC) are used. The first gas line is connected to the mixing container via a bubbler for $N_2$ gas to carry water vapor. The sidewall of the bubbler is wrapped with an isolated flow line from the heating circulator through which the heated water flows. This heated water is used to warm the water inside the bubbler to facilitate the generation of water vapor. The second gas line is directly connected to the mixing container, where $N_2$ gas from the second gas line is intermixed with that from the first gas line to adjust the RH. If the intermixed $N_2$ gas with a certain RH flows toward the test chamber for humidity sensing, it passes through a commercialized RH reader to recognize the RH before reaching the test chamber. The third gas line connected directly to the RH reader is used to recover the sensor.

1. Pulse Operation Schemes for Humidity Sensing

 The humidity-sensing properties of the sensor are demonstrated using pulse operation schemes. Figs. 3(a) and 3(b) show the pulse signal waveforms for operating the sensor to obtain the pulsed $I_D-V_{CG}$ (PIV) characteristics and the transient $I_D$ behaviors, respectively. The CG bias ($V_{CG}$) is swept ranging from 2 to -2 V for PIV measurement, while a fixed value of -0.2 V is applied to the CG read bias ($V_{rCG}$) for transient ID measurement. For both PIV and transient $I_D$ measurements, the base voltages ($V_{bases}$) of the CG and drain pulses are fixed at 0 V, and the drain read bias is -0.1 V. A single pulse signal is applied to the CG terminal with the high and low levels corresponding to $V_{CG} (or V_{rCG})$ and $V_{base}$, respectively. The durations of the high and low levels are defined as ton and toff, respectively, which are set to 30 μs and 100 ms in this study. The drain pulse signal is synchronized with the CG pulse signal, following the same timing scheme.
The pulse operation scheme in Fig. 3(b) has been previously proposed and its effect has been verified [27]. According to [27], through the application of a pulse operation scheme, the ID drift of a Si FET-type sensor caused by the DC bias is significantly reduced, which facilitates the obtaining of stable sensing characteristics.
Fig. 4 displays the transfer ($I_D-V_{CG}$) characteristics of the fabricated FET-type humidity sensor. The channel width and length of the FET are both 2 mm. The $I_D-V_{CG}$ curves are measured at an RH of 3.4% and room temperature using the DC and PIV methods. The two logarithmic $I_D-V_{CG}$ curves obtained by the DC and PIV methods coincide well with each other at |$I_D$| > 1 nA. This implies that the $t_{on}$ of 30 ms is appropriate to operate normally the FET of the humidity sensor.

2. Chemical Reaction and Chemical-to-electrical Transduction Mechanisms

 Fig. 5(a) shows the shift in the $I_D-V_{CG}$ curves of the sensor obtained using the PIV method with varying RH inside the test chamber. Each $I_D-V_{CG}$ curve is measured 300 s after injecting $N_2$ gas with a certain RH into the test chamber to saturate the ambient RH inside the test chamber. Fig. 5(b) illustrates the variation of threshold voltage ($V_{th}$) with respect to RH.$V_{th}$ is extracted from each $I_D-V_{CG}$ curve in Fig. 5(a) using the linear extrapolation method. The increase in RH from 3.4% to 80.3% leads to the change in $V_{th}$ (△$V_{th}$) by -134 mV.
Here, we explain the reason for the △$V_{th}$ being negative with the increase of RH. As shown in Fig. 6, adsorption of water molecules onto the surface of the $WO_3$ humiditysensing layer is divided into two steps. At a lower RH, water molecules dissociate into hydroxide (OH-) and hydrogen (H+) ions when exposed to the WO3 layer [28].

 The OH- ions interact with the tungsten cations on the $WO_3$ surface. Whereas, the H+ ions interact with the lattice oxygens or the oxygen ions existing on the surface of the WO3 layer, forming the hydroxyl groups that are chemically bonded to the tungsten cations [28]. Finally, both OH- and H+ ions are chemically adsorbed (chemisorbed) onto the surface of the WO3 layer in the form of a hydroxyl group (-OH). Note that the O-H chemical bond in this hydroxyl group is a polar covalent bond with the negatively polarized oxygen atom and the positively polarized hydrogen atom.
Meanwhile, we must recall a fact that has already been demonstrated experimentally in our previous study [18]. In [18], it was revealed that the sensor response changes depending on the pre-bias ($V_{pre}$) applied to the CG. When $V_{pre}$ is negative, the electrons of the ZnO sensing layer are accumulated near the interface between the ZnO layer and the O/N/O layer, and the chemical reaction between NO2 (target gas) and ZnO mainly occurs near the interface, which is relatively close to the FET channel. Therefore, the sensor response increases compared to that in the case of $V_{pre}$ = 0 V. However, when $V_{pre}$ is positive, the ZnO layer near the interface is depleted. Consequently, the chemical reaction mostly happens in the bulk region of the ZnO layer. In this case, the sensor response eventually decreases compared to that in the case of $V_{pre}$ = 0 V because the amount of chemical reaction at the interface between the ZnO layer and the O/N/O layer is reduced due to the depletion of the ZnO layer. In summary, the chemical reaction occurring in the sensing layer near the interface between the sensing layer and the O/N/O layer dominantly affects the FET of the sensor. Therefore, in this study, it is assumed that the chemical reaction between $H_2O$ and $WO_3$ occurs predominantly near the interface between the $WO_3$ layer and the O/N/O layer.
At the interface between the $WO_3$ layer and the O/N/O layer near the FG, the positively polarized hydrogen atoms of the hydroxyl groups generated by $H_2O$ chemisorption on the$WO_3$ layer induce a negative sheet charge at the interface of the FG in contact with the O/N/O layer. This negative sheet charge consists of the electrons, which are the majority carriers of the n-type poly-Si FG. Simultaneously, a positive sheet charge, which consists of the depletion charge of the n-type poly-Si FG, is induced at the interface of the FG in contact with the gate oxide. This positive sheet charge reduces the hole concentration of the ptype FET channel of the sensor. In summary, chemisorption of water molecules onto the surface of the $WO_3$ layer shifts Vth of the p-type FET sensor in the negative direction, causing |$I_D$| to decrease. At a higher RH, additional water molecules are physically adsorbed (physisorbed) onto the hydroxyl groups formed by chemisorption of water molecules [28]. The first physisorbed layer of water molecules are formed by the hydrogen bond between the hydrogen atoms of the hydroxyl groups already bonded to the $WO_3$ layer via chemisorption of water molecules at a lower RH, and the oxygen atoms of the additional water molecules [28]. From the second physisorbed layer, hydrogen bonds occur between the hydrogen atoms of the previously physisorbed water molecules and the oxygen atoms of added water molecules [28]. Similar to the chemisorbed layer, the molecular arrangement of these multiple physisorbed layers can be regarded as a dipole with the direction of the dipole moment being toward the FET channel of the sensor, thereby reducing the hole concentration of the channel. Therefore, $V_{th}$ and |$I_D$| of the sensor decrease as RH increases, regardless of whether water molecules are chemisorbed or physisorbed onto the surface of the $WO_3$ layer.

3. Change of Energy Band Structure during Humidity Sensing

 Next, we examine the change in energy band structure of the sensor when water molecules are adsorbed onto the WO3 sensing layer. Fig. 7(a) shows the schematic energy band diagram of the sensor under flat-band condition. Here, $Φ_{CG}, Φ_{WO3}, Φ_{FG}$, and $Φ_{sub}$ represent the work functions of CG,$WO_3$ sensing layer, FG, and Si substrate, respectively. $χ_{WO3}$, $χ_{FG}$, and $χ_{sub}$ stand for the electron affinities of $WO_3$ sensing layer, FG, and Si substrate, respectively. Because the Au of the CG occupies most of the direct contact area with the $WO_3$ layer, the $Φ_{CG}$ is regarded as the work function of Au (5.1 eV). The electron concentrations of the FG and the substrate are ~1x10$^21$ and ~1x10$^16$ cm$^{-3}$, respectively, thus the FFG and the Fsub are ~4.05 and ~4.26 eV, respectively. $WO_3$ is a wellknown n-type semiconductor [29] with an $χ_{WO3}$ value of ~3.3 eV [30]. The $χ_{FG}$ and the $χ_{sub}$ are both 4.05 eV. Note that the FG is in a fresh state, as shown in Fig. 7(a). This implies that electrons or holes are not stored in the FG. Fig. 7(b) shows the schematic energy band diagram under equilibrium condition where all the electrodes in the sensor are grounded (0 V). As the |$I_D$| at $V_{CG}$ = 0 V is ~0.2 μA in the $I_D-V_{CG}$ curve in Fig. 4, it can be inferred that the FET of the sensor is in weak inversion. In addition, considering the band structure shown in Fig. 7(a), it is evident that electrons are initially stored in the FG, as shown in Fig. 7(b). This phenomenon is similar to the programmed state of a flash memory device with a poly-Si floatinggate; we have confirmed the program and erase characteristics of the sensor in the previous study [18]. Fig. 7(c) illustrates the change of the energy band diagram before and after humidity sensing at $V_{CG} = V_{rCG}$ = -0.2 V. As described in Fig. 6, the chemical reaction between water molecules and the $WO_3$ sensing layer results in the formation of hydroxyl groups by chemisorption or dipoles by physisorption at the interface between the WO3 layer and the O/N/O layer. These hydroxyl groups and dipoles both exhibit the dipole moments in the same direction from the interface to the FET channel. This induces a negative sheet charge at one interface of the FG in contact with the O/N/O layer and a positive sheet charge at the opposite interface of the FG in contact with the gate oxide. The localization of these charges results in a change in the electric field, which consequently decreases |$I_D$| of the FET.

4. Sensor Response as a Function of Relative Humidity

 Fig. 8(a) shows the transient $I_D$ behaviors as a parameter of RH using the pulse measurement method presented in Fig. 3(b). Similar to the experimental method used when obtaining the results shown in Fig. 5, $N_2$ gas with a certain RH is injected into the test chamber for 300 s to saturate the humidity sensing before the pulse biases are applied to the sensor for 10 s to measure every transient $I_D$ curve. At this time, the $V_{rCG}$ and the $V_{rDS}$ are -0.2 V and -0.1 V, respectively. $I_D$ decreases with the increase in RH and there is no drift in any of the transient $I_D$ curves owing to the pulse measurement method. In the previous studies [31,32], drifts in output signals have been observed due to the application of DC bias to the sensors, which are undesirable because they degrade the accuracy of sensing and therefore induce errors in sensor operations. However, this study confirms that the pulse bias to the sensor can suppress $I_D$ drift. This method also offers the advantage of reduced power consumption during operation. Fig. 8(b) shows the sensor response (R) versus RH calculated from the $I_Ds$ in Fig. 8(a), where R is defined as the rate of change in $I_D$ divided by $I_D$ at an RH of 3.4%. The Rs are 22.3% and 45.5% at RHs of 54.9% and 80.3%, respectively. As RH increases from 3.4% to 54.9%, R increases and then appears to saturate. However, it linearly increases above an RH of 54.9%. It is speculated that a change in the adsorption process of water molecules on the WO3 layer from chemisorption to physisorption occurs at this inflection point (RH = ~54.9%), which needs to be proven through further research.
The dynamic response of the sensor is also demonstrated in Fig. 9 via the injection of $N_2$ gas with a certain RH and the reference $N_2$ gas with an RH of 3.4% alternately into the test chamber for 180 and 300 s, respectively. To measure the transient $I_D$ curve for the dynamic response, $N_2$ gases with RHs of 11.5%, 28.2%, 54.9%, and 68.5% are sequentially injected, and the reference $N_2$ gas is injected for sensor recovery. The $V_{rCG}$ and the $V_{rDS}$ are -0.2 V and -0.1 V, respectively. Similar to the results shown in Fig. 8(a), |$I_D$| decreases with the increase in RH of the injected $N_2$ gas. The response and recovery speeds of the sensor are also obtained by defining response time as a time duration for which |$I_D$| decreases by 90% of the change in $I_D$ during response, and recovery time as a time duration for which |$I_D$| increases by 90% of the change in $I_D$ during recovery. The response and recovery times of the sensor are 97 and 190 s, respectively, at an RH of 68.5%.

  In this study, a Si FET-type humidity sensor was fabricated using a $WO_3$ thin film as the sensing material. The sensor comprised a sensing part that functioned as a gate as well as an RH sensor and an FET part, wherein an electrical change was induced by humidity sensing. The humidity-sensing characteristics were obtained by measuring the $I_D-V_{CG}$ curves and transient $I_Ds$. In contrast to the general methods, stable performance was achieved without drift of $I_D$ by using the pulse measurement method. In addition, the signal transduction of the sensor generated by the reaction between the $WO_3$ sensing material and the water molecules was analyzed using energy band diagrams. The FET-type humidity sensor with Si CMOS technology and pulse-driving method proposed in this study is expected to be the most promising candidate for IoT-based smart humidity sensor.

  This work was supported by a New Faculty Research Grant of Pusan National University, 2024 and in part by NRF funded by the Korean government (RS-2024- 00405200).