Introduction

Recent years have witnessed a significant rise in demand for high-performance gas sensors. From conventional uses such as indoor/outdoor air quality monitoring, applications of gas sensors have expanded to include medical diagnosis and food quality control [1,2,3,4,5]. Among different types of gas sensors, metal oxide (MOX)-based gas sensors have merits in terms of the simple fabrication process, low cost, simplicity in use, and a large number of detectable gases [6, 7]. Despite the numerous advantages, MOX-based gas sensors still suffer from poor stability and reliability due to their susceptibility to temperature and humidity [8].

Accordingly, there are previous studies aimed at enhancing stability and long-term reliability. Most of the previous studies focused on improving the reliability of MOX sensing materials. Han et al. demonstrated that the optimization of the metal-to-oxygen ratio in the SnO2 could achieve good stability at a temperature above 350 °C [9]. Postica et al. reported that coating ZnO with an ultra-thin layer of SiO2 improves long-term stability and minimizes the humidity effects [10]. Besides these studies, various parameters are known to affect the stability and long-term reliability of MOX sensing materials. For example, small particles degrade long-term stability. This problem may be partly resolved by preparing the nanoparticles in a slurry form and then depositing them directly onto the microelectrodes by ink-jet printing and micro-dispensing. In addition, it is difficult to achieve thermal stability of the particle with a small size. This issue can be overcome to some degree by using sensors that are functional at low temperatures [11].

However, the reliability of the sensor is influenced not only by the sensing material (MOX), but also by the transducer of the sensor [12,13,14]. In particular, when configuring a sensitive amplifier circuit with n- and p-channel FET-type (nFET and pFET) gas sensors, the nFET-type gas sensor is exposed to iterative off-state stress (OSS) during the circuit operation [15,16,17]. When the input signal of the sensing amplifier circuit is low, the OSS is applied to the nFET (VG = GND, VD = VDD, VS = GND, VB = GND), as shown in Fig. 1a. It has been reported that the OSS degrades the gate oxide of the FET by generating interface and bulk traps [15]. When the 1/f noise is increased by the damaged gate oxide, the signal-to-noise ratio (SNR) and limit of detection (LOD) of the FET-type gas sensor are degraded because the FET transducer determines the low-frequency noise (LFN) characteristics [18,19,20]. Therefore, the impact of OSS on the sensing performances of the nFET-type gas sensor should be investigated.

Fig. 1
figure 1

a Schematic of sensitive amplifier composed of nFET- and pFET-type gas sensors. The nFET is exposed to the OSS when the input signal is ‘0’. b Schematic of the 3D bird’s eye view of the nFET-type gas sensor

Full size image

In addition, the method to improve the degradation induced by the OSS should be provided. Recently, self-curing methods that utilize the internal operation of the FETs, including PN-junction forward current and punch-through current, have been proposed to improve the reliability of the FETs [21,22,23]. Unlike global forming gas annealing, electrothermal annealing, also known as the self-curing method, can be effectively applied to a unit device without the need for bulky equipment. The self-curing methods make use of the Joule heat (JH) that is inherently generated during the operation of FETs. JH is used to cure the defects inside the gate oxide with the same principle as thermal annealing. Recent reports demonstrated the JH can be induced by PN-junction forward current, gate-induced drain current leakage, and punch-through current [21,22,23]. Self-curing methods have significant advantages over furnace annealing because they can repair the damage even after packaging without affecting the interconnection or device parameters of other devices. In this regard, the self-curing methods have been applied to various applications, including memory devices [24]. However, its application has not been demonstrated in sensor applications despite its advantage and potential.

Following the above discussion, in this work, we investigate the impact of OSS on the gas sensing, LFN, and SNR of the nFET-type gas sensor with horizontal floating gate (FG). The In2O3 is used as a sensing material, and NO2 is used as a target gas. It is revealed that the gate oxide is degraded by the band-to-band tunneling (BTBT) due to the increased vertical electric field at the gate-drain overlap. After the OSS is applied, the magnitude of the 1/f noise is significantly increased, degrading the SNR and LOD of the sensor. In order to cure the damaged gate oxide, the self-curing method based on PN-junction forward current is applied. The proposed self-curing method represents a major advancement in gas sensor research by presenting a novel method to improve the reliability of gas sensors.

Results and discussion

In a sensitive amplifier with nFET and pFET-type gas sensors, the nFET-type gas sensor is exposed to the OSS when the input signal is low, as shown in Fig. 1a. Figure 1b shows the schematic of the 3D bird’s eye view of the nFET-type gas sensor. The width/length (W/L) of the FET channel is 1 μm/0.6 μm. The detailed fabrication process of the sensor is explained in [13]. The simplified explanation of the sensor fabrication process is as follows: The devices are isolated using the local oxidation of silicon technique in an n-type Si wafer. A 10 nm-thick SiO2 is grown using dry oxidation as a gate oxide. A 350 nm-thick in-situ doped n+ polycrystalline Si is formed as FG. Source and drain are formed using self-align. The SiO2/Si3N4/SiO2 is successively formed as a passivation layer. Using a lift-off process, successive layers of Ti (30 nm), TiN (20 nm), and Al (100 nm) are deposited and patterned to form the electrodes for the drain, source, and body. These metal layers are also used as a control gate (CG). Finally, a 15-nm-thick n-type semiconducting In2O3 film is deposited using a sputtering method. Note that a detailed fabrication process of the sensor is explained in Ref. 13. In this work, the OSS is applied at 125 °C for the acceleration test. The VD is set to OSS stress bias (VD,OSS), and the other terminals are grounded for the OSS. The stress time (tOSS) and VD,OSS are changed to quantitatively investigate the impact of OSS.

Figure 2a shows the transfer characteristics (ID–VCG) of the nFET-type gas sensor as a parameter of tOSS. Note that the VD,OSS is set at 6.5 V. The threshold voltage (Vth) of the sensor is shifted to a negative direction with an increase in tOSS. As the electrical field is increased by the VD,OSS at the gate/drain overlap region, the valence band electrons tunnel to the conduction band. The strong electric field causes these electrons to gain energy, inducing impact ionization to generate electron/hole pairs. The energetic holes move toward the gate along the electric field and these holes are eventually injected into the gate oxide, causing a negative Vth shift of the sensor. Figure 2b shows the Vth shift (ΔVth) versus tOSS at VD,OSS of 5.0 V and 6.5 V measured at 125 °C (solid symbol) and 150 °C (open symbol). Figure 2c schematically illustrates the mechanism of OSS-induced gate oxide degradation. During the OSS, hot holes cause traps and defects inside the gate oxide. It has been reported that the Si–O bonds are broken by the hot holes during the OSS [15]. When the hot holes are injected into the gate oxide, Si–O and Si–H bonds can be damaged, leading to dangling bonds. As shown in Fig. 2b, the time exponent of the degradation at the early stress stage is high and then saturates with an increase in tOSS. Such a trend demonstrates that the defects generated by the OSS mostly originate from the broken Si–O bond.

Fig. 2
figure 2

a ID–VCG of the sensor as a parameter of tOSS measured at 125 °C (Solid symbols) and 150 °C (Open symbols). b ΔVth versus tOSS at VD,OSS of 5.0 V and 6.5 V. c Schematic illustration of the OSS-induced gate oxide degradation

Full size image

In sensor applications, the LFN determines the SNR and LOD of the sensor, which requires a systematic investigation [18]. The LFN characteristics of the FET-type gas sensors are determined by the FET transducer, whose properties are strongly affected by the gate oxide-channel interface quality.13 When the carriers (electron) in the FET channel moves from source to drain, the carriers are trapped/detrapped to/from defects inside the gate oxide (SiO2 in this work). A time constant of trapping/detrapping process generates the Lorentzian noise with a specific corner frequency, and the distribution of traps with different time constants produces a power spectral density (PSD) of 1/f noise. This can be modeled using the carrier number fluctuation (CNF) model [13].

The CNF model is expressed as [24]

$$\frac{{S_{{{\text{ID}}}} }}{{I_{D}^{2} }} = \left( {\frac{{g_{m} }}{{I_{D} }}} \right)^{2} S_{{{\text{Vfb}}}}$$
(1)

with

$$S_{{{\text{Vfb}}}} = \frac{{q^{2} kTN_{T} \lambda }}{{WLC_{{{\text{ox}}}}^{2} f}}$$
(2)

where gm is the transconductance, SVfb is the flat-band voltage noise spectral density, q is the electron charge, k is the Boltzmann constant, T is the temperature, NT is the volume trap density, \(\lambda\) is the tunneling attenuation coefficient in the gate oxide, Cox is the gate oxide capacitance, f is the frequency. The SVfb is proportional to the NT and inversely proportional to the dimension of the channel (WL). The reason why the SVfb is inversely proportional to WL is that the magnitude of noise is inversely proportional to the effective size of the noise source. The SVfb is reflected SID by the amplification of gm. In addition to CNF, the carrier mobility can be fluctuated by a trapped carrier in the gate oxide due to Coulomb interaction [25, 26]. This phenomenon is called correlated mobility fluctuation (CMF). The CNF model is modified by multiplying a CMF term \({(1\pm \alpha {\mu }_{\mathrm{eff}}{C}_{\mathrm{ox}}\frac{{I}_{D}}{{g}_{m}})}^{2}\) in Eq. (1). Note that α is the correlated mobility fluctuation parameter and μeff is the effective mobility.

Figure 3a shows the normalized drain current power spectral density (SID/ID2) of the before (pristine) and after the OSS. The 1/f noise is increased after the OSS. In order to find where such an increase stems from, the origin of 1/f noise is verified. As shown in Fig. 3b, the SID/ID2 and (gm/ID)2 show similar behavior with respect to ID, demonstrating that the 1/f noise stems from the CNF [27, 28]. Figure 3c shows the input gate-voltage noise power spectral density (SVG) versus ID/gm. An increase in SVG with ID/gm after the OSS is due to the CMF. Figure 3d shows trap density (NT) along the vertical depth (z) of the gate oxide extracted in the sensor before and after the OSS. It is clearly observed that the NT is increased after the OSS.

Fig. 3
figure 3

a SID/ID2 of the sensor before (pristine) and after the OSS. b SID/ID2 sampled at 1 Hz and constant \(\times\) (gm/ID)2 versus ID of the sensor before and after the OSS. c SVG1/2 versus ID/gm of the sensor before and after the OSS. d NT versus z of the sensor before and after the OSS

Full size image

Now we evaluate the SNR of the sensors. In most studies that evaluate the LOD of gas sensors, LODs are determined by simply extrapolating the sensitivity to the gas concentration at which the response becomes either one or zero (depending on the definition of response employed). However, the LoD of sensors is also determined by the noise of the sensing signal. Therefore, the noise of the sensor should also be considered in evaluating LOD; thus, the signal-to-noise ratio (SNR) rather than the response is a more valid metric. Note that the noise of gas sensors is determined by LFN due to a long time response (from a few seconds to hundreds of seconds).In this study, the SNR is defined as

$$\mathrm{SNR}=\frac{\Delta I}{\delta I}=\frac{{I}_{g}-{I}_{a}}{\sqrt{{\int }_{f1}^{f2}{S}_{I}\left(f\right)df}}=\frac{{I}_{g}-{I}_{a}}{\sqrt{BW}\times \sqrt{{S}_{I}(f=1\,\text{Hz})}}$$
(3)

where ΔI is signal generated from the gas reaction; δΙ is root-mean-square drain current noise amplitude; SI (f = 1 Hz) is PSD of the current noise at a frequency of 1 Hz; and BW is a bandwidth-related term that depends on the largest (f 2) and smallest (f 1) frequencies sampled; BW = ln(f2/f1). Note that \(\sqrt{BW}\) ranges from about 3.5 to 3.8 in typical gas sensing measurements. The LFN determines the δΙ as the magnitude of the 1/f noise is in several orders larger than that of thermal noise in the measuring BW range. Therefore, an evaluation of SNR has to be based on the accurate measurement of the PSD of 1/f noise.

The SNR per unit change of FG voltage (ΔVFG) due to the NO2 gas reaction of the FET-type gas sensor is gm/\(\sqrt{BW}\times \sqrt{{S}_{VG}}\) where SVG is gate-voltage noise power spectral density [18, 29]. The ΔVFG does not show a significant difference after the OSS because the ΔVFG is determined by the sensing material, not by the FET transducer (Fig. 4b). Figure 4c shows the SNR/ΔVFG versus ID of the sensor before and after the OSS. It is revealed that the increased 1/f noise due to the OSS significantly decreases the SNR. A larger decrease in SNR in the linear region is due to the increased impact of CMF after the OSS. Figure 4d shows the SNR of the sensors versus NO2 gas concentration in different operating regions.

Fig. 4
figure 4

a SNR/ΔVFG versus ID of the sensor before and after the OSS. b SNR versus NO2 concentration of the sensor before and after the OSS at different operating regions

Full size image

To improve the SNR degradation, the method to recover the damage originating from the OSS should be provided. Previous studies reported that the damage induced by the OSS can be cured by the thermal annealing [16]. However, the problem is that conventional thermal annealing using external equipment, including a furnace cannot be employed to repair OSS-caused damage because the heat treatment is difficult after being integrated on the system board. In contrast, the self-curing method employs Joule heat (JH) that is generated by the self-heating effect within the transistor itself [21,22,23,24]. Therefore, the self-curing method can be used selectively on the damaged device to recover the device even after packaging. In this study, the PN-junction forward current is used to generate JH in the transistor, thereby repairing the OSS-induced damage. When a large current is generated at the gate/drain overlap region by the PN-junction forward bias, the defects are cured by the JH from the forward junction current (IFWD). Figure 5a shows the IFWD versus VD. As a very large IFWD flows through the junction, the Joule heat (JH) is generated, which can be used to cure the damaged gate oxide. Figure 5b shows the ID–VCG of the OSS-damaged and self-cured sensor. The inset schematically illustrates the mechanism of self-curing. Figure 5c shows the SID/ID2 of the OSS-damaged and self-cured sensors. The 1/f noise is decreased by self-curing because the JH decreases the traps generated by the OSS. Figure 5d shows the SNR/ΔVFG of the OSS-damaged and self-cured sensors. It is also revealed that the impact of CMF is decreased by self-curing.

Fig. 5
figure 5

a IFWD versus VD of the sensor. b ID–VCG of the off-state stressed and self-cured sensors. c SID/ID2 of the off-state stressed and self-cured sensors. d SNR/ΔVFG versus ID of the off-state stressed and self-cured sensors

Full size image

Now we investigate the NO2 gas sensing properties of the sensors. When the sensor is exposed to NO2, the negative sheet chare (NO2−) is accumulated at the interface between the passivation layer and the sensing material. To compensate for the excessive NO2− charge, the electrons are depleted at the channel of the FET transducer. Accordingly, the ID of the sensor is decreased.

Figures 6a-1, a-2, and a-3 correspondingly show the response versus ID of the pristine, OSS-damaged, and self-cured sensors at different NO2 gas concentrations. The VDS is fixed at 0.1 V, and VCG is swept from 0.0 to 3.0 V. In this study, the response is defined as

Fig. 6
figure 6

Response versus ID of the (a-1) pristine, (a-2) OSS-damaged, and (a-3) self-cured sensors at various NO2 concentrations. b ΔVFG versus NO2 concentration of the pristine, OSS-damaged, and self-cured sensors. c Transient response behaviors of the self-cured exposed to 500 ppb of NO2 gas. The transient behaviors are measured at the subthreshold and linear regions

Full size image
$$\mathrm{Response }(\mathrm{\%})= \frac{\left|{I}_{\mathrm{D},\mathrm{air}}-{I}_{\mathrm{D},\mathrm{gas}}\right|}{{I}_{\mathrm{D},\mathrm{air}}} \times 100$$
(4)

where the ID,air and ID,gas are the ID of the sensors at reference gas (dry air) and NO2 gas, respectively. Because the \(\left|{I}_{\mathrm{D},\mathrm{air}}-{I}_{\mathrm{D},\mathrm{gas}}\right|\) is proportional to the gm \(\times\) ΔVFG [18], the response can be expressed as

$$\mathrm{Response }(\mathrm{\%})= \frac{{g}_{m}}{{I}_{\mathrm{D},\mathrm{air}}} \times \Delta {V}_{\mathrm{FG}}\times 100$$
(5)

where ΔVFG denotes the change in the FG voltage from the NO2 gas reaction. In all cases, the response exhibits its largest value at the subthreshold region owing to the largest transconductance efficiency (gm/ID). As shown in Fig. 6b, there is no change in ΔVFG after the OSS. The result demonstrates that the OSS does not affect the properties of sensing materials (In2O3). However, the decrease in response is observed in Fig. 6a-2 after the OSS. This originates from the degradation in transconductance efficiency (gm/ID) due to the damaged gate oxide, as demonstrated in Fig. 3. From these results, it is revealed that not only the SNR but also the response of the sensors is degraded by the OSS. Figure 6c shows the transient response behaviors of the self-cured sensor exposed to 500 ppb of NO2 gas. The transient behaviors are measured at the subthreshold and linear regions. IDs of 80 and 800 nA are used to operate the sensor in the linear and subthreshold regions.

Figure 7a shows the response of the sensor to different types of gases (NO2: 500 ppb, H2S: 10 ppm, C3H6O: 50 ppm, C3HOH: 50 ppm, SO2: 100 ppm, NH3: 125 ppm, CO: 1000 ppm, and CO2: 1000 ppm). The sensor shows the largest response to NO2 gas, exhibiting excellent selectivity. Figure 7b shows the long-term stability of the sensor. The response of the sensor is maintained for 100 days, demonstrating good long-term stability. Figure 7c shows the effects of humidity on the sensors. The humidifier produces humid air by passing dry air (3.4% RH) through it. The mass flow controller regulates the gas flow rates of the reference gas and the dry air going through the humidifier. The detailed measurement condition is explained in [30]. With an increase in relative humidity (R. H.) from 3.5 to 81.3%, the response decreases from 89.1 to 61%. It seems that the H2O molecules in humid air interfere with the interaction of gas molecules and In2O3 film.

Fig. 7
figure 7

a Response of the self-cured sensor to different types of gases (NO2: 500 ppb, H2S: 10 ppm, C3H6O: 50 ppm, C3HOH: 50 ppm, SO2: 100 ppm, NH3: 125 ppm, CO: 1000 ppm, and CO2: 1000 ppm). b Long-term stability of the self-cured sensor. c Response of the self-cured sensor versus R. H.

Full size image

Conclusions

In this study, we investigated the impact of OSS on the NO2 gas sensing performance of the horizontal FG FET-type gas sensors. The LFN of the sensor is determined by the 1/f noise of the FET transducer, and its magnitude is significantly increased by damaged gate oxide due to the hot holes. Accordingly, the SNR of the sensor is significantly degraded by the OSS. Such degradation is repaired by the self-curing method based on forward current at the body–drain PN junction. The proposed self-curing method can be successfully adopted to recover the damage caused by the OSS. The results of this study pave the way to improve the reliability of the FET-type gas sensor by proposing an efficient self-curing method.