Amorphous Oxide Thin Film Transistors with Nitrogen-Doped Hetero-Structure Channel Layers

Abstract

The nitrogen-doped amorphous oxide semiconductor (AOS) thinfilm transistors (TFTs) with double-stacked channel layers (DSCL) were prepared and characterized. The DSCL structure was composed of nitrogen-doped amorphous InGaZnO and InZnO films (a-IGZO:N/a-IZO:N or a-IZO:N/a-IGZO:N) and gave the corresponding TFT devices large field-effect mobility due to the presence of double conduction channels. The a-IZO:N/a-IGZO:N TFTs, in particular, showed even better electrical performance (µ_FE = 15.0 cm²·V⁻¹·s⁻¹, SS = 0.5 V/dec, V_TH = 1.5 V, I_ON/I_OFF = 1.1 × 10⁸) and stability (V_TH shift of 1.5, −0.5 and −2.5 V for positive bias-stress, negative bias-stress, and thermal stress tests, respectively) than the a-IGZO:N/a-IZO:N TFTs. Based on the X-ray photoemission spectroscopy measurements and energy band analysis, we assumed that the optimized interface trap states, the less ambient gas adsorption, and the better suppression of oxygen vacancies in the a-IZO:N/a-IGZO:N hetero-structures might explain the better behavior of the corresponding TFTs.

1. Introduction

Amorphous oxide semiconductors (AOS), including amorphous InGaZnO (a-IGZO), amorphous InZnO (a-IZO), amorphous InGaO (a-IGO), etc., have been widely investigated as the channel layers of thin-film transistors (TFTs) since 2004 [1]. In fact, AOS TFTs are one of the major candidates to replace silicon TFTs for driving active-matrix liquid crystal displays (AMLCDs) and active-matrix organic light-emitting diodes (AMOLEDs) because of their high field-effect mobility (µ_FE~10 cm²·V⁻¹·s⁻¹), good uniformity (substrate size could reach 2200 × 2500 mm or even larger), simple processing methods, etc. [2,3,4,5]. However, further improvements over the performance and stability of AOS TFTs are still needed for practical applications in flat panel displays (FPDs) and the other fields.

Arai et al. [6] used amorphous InSnZnO (a-ITZO) as the channel layers of TFTs and designed low RC-delay device structures to optimize the performance and reliability of pixel and driving circuits for AMOLEDs. Hsu et al. [7] improved the stability of a-IGZO TFTs by using the asymmetric device structure. Lin et al. [8] co-sputtered IGZO and Al to form stable IGZAO TFTs [8]. Mativenga et al. [9] prepared the rollable a-IGZO TFTs with good bias stability, good mechanical stability, and high operation frequency (~400 kHz) by using the solution-processed colorless-polyimide as the substrates [9].

Recently, nitrogen-doping (N-doping) was shown to effectively improve the electrical properties (e.g., sub-threshold swing (SS), bias-stress stability, etc.) of AOS TFTs (e.g., a-IGZO:N TFTs [10], a-IZO:N TFTs [11], etc.). The N-doping decreased the channel/dielectric interface trap density, as well as the density of deep states, and the concentration of oxygen vacancies (V_O) in the channel layers of a-IGZO TFTs [12]. Our previous work [13] indicated that N-doping improved the device performance and stability of a-IGZO TFTs by moderately adjusting the Vo concentration and largely suppressing the defects formation in the bulk channels, which was different from undoping and oxygen doping (O-doping). On the other side, N-doping unexpectedly degraded the mobility in a-IGZO:N [12], a-IZO:N [14], and ZnO:N [15], which might be attributed to the suppression of Vo, the main source of free electrons in AOS films [16].

Another effective improving method is to form double-stacked channel layers (DSCL) for AOS TFTs, which has been studied by our group [17,18], Chong et al. [19], and Jeon et al. [20]. Among these studies, the DSCL structure was shown to be valid at improving the electrical characteristics (e.g., µ_FE, bias-stress stability, etc.) of TFTs. These improvements might result from the hetero-structure channel layers in DSCL devices [21,22,23]. In our recent reports, we obtained the DSCL structure by combining the a-IZO film with a-IGZO:N films, which made the corresponding TFTs exhibit quite a large µ_FE (49.6 cm²·V⁻¹·s⁻¹) but too negative threshold voltage (V_TH ~−2.3 V) [24,25]. Therefore, further improvement is still necessary.

In this study, we used a-IZO:N rather than a-IZO to combine with a-IGZO:N and formed two novel DSCL structures (a-IGZO:N/a-IZO:N and a-IZO:N/a-IGZO:N), achieving a good electrical performance and good stability of the corresponding TFT devices.

2. Materials and Methods

The inverted staggered AOS TFTs were fabricated on the p++ silicon wafers with 100-nm-thick thermal oxide (SiO₂). The silicon wafers and thermal oxide films served as the gate electrodes and gate insulators, respectively. As shown in Figure 1, two types of N-doped DSCL TFTs were prepared for comparison, i.e., Samples I (a-IGZO:N/a-IZO:N) and II (a-IZO:N/a-IGZO:N).

Here, the channel layers and source/drain (S/D) electrodes were all deposited by a sputtering machine (SYSKEY Automatic Muti-targets Sputtering System, Hsinchu, Taiwan). The chamber was evacuated to base-pressure (lower than 3 × 10⁻⁶ Torr), and then the pressure was fixed at 3 × 10⁻³ Torr during film deposition. The a-IGZO:N films were deposited by RF magnetron sputtering at room temperature (RT) using an a-IGZO target (In₂O₃:Ga₂O₃:ZnO = 1:1:1 mol % ) with a plasma power of 60 W and an Ar flow rate of 10 sccm (sccm denotes cubic centimeter per minute at STP); the nitrogen gas (N₂) was introduced in situ into the chamber at a flow rate of 1.2 sccm. The deposition of a-IZO:N films was performed by RF magnetron sputtering at RT using an a-IZO target (In₂O₃:ZnO = 3:1 mol %) with a plasma power of 70 W. The N₂ was fed in situ into the chamber at a flow rate of 2.5 sccm; the flow rates of Ar and O₂ gas were fixed at 20 and 7 sccm, respectively. After depositing the channel layers, the 100-nm-thick indium-tin-oxide (ITO) layers were prepared as S/D electrodes by RF magnetron sputtering. During sputtering, the shadow masks were used to define the patterns of the channel layers and S/D electrodes. Finally, the samples were annealed at 573 K for 1 h in N₂ atmosphere.

In this study, the channel width (W) and length (L) of the tested TFTs were fixed as 1000 and 250 µm, respectively. The electrical measurements for all the devices were employed using an electrical analyzer (Keithley 4200, Keithley Instruments, Inc., Beaverton, OR, USA). AnX-ray photoemission spectroscopy (XPS, ESCLAB 250Xi, Thermal-fisher, Waltham, MA, USA) was used to analyze the O1s peaks for the a-IGZO:N and a-IZO:N thin films. An atomic force microscopy (AFM, Bruker ICON, Tucson, AZ, USA) was used to characterize the surface properties of a-IGZO:N and a-IZO:N thin films.

3. Results

Figure 2 shows the O1s XPS spectra of a-IGZO:N and a-IZO:N thin films. The same processing conditions for the film depositions were used as were adopted for the corresponding TFTs. The spectra were calibrated by tacking the C1s (~284.8 eV) reference. Figure 2a,b show the de-convolution of O1s XPS spectra for the a-IGZO:N and a-IZO:N thin films, respectively. Gaussian fitting was used to de-convolve the O1s peaks, which could be fitted by three components, centered approximately at 530 ± 0.1 eV (O₁ peak), 531 ± 0.3 eV (O₂ peak), and 532 ± 0.2 eV (O₃ peak). The O₁ peak is supposed to reflect the oxide lattices in the stoichiometric wurtzite structure without oxygen deficiency; the O₂ peak represents the oxide lattices in the oxygen-deficient region (V_O). The O₃ peak is usually considered to be related to the chemisorbed hydroxide (e.g., moisture etc.) [26,27,28].

To quantitatively study each component of the O1s peak, we defined the Area and %Area to denote the de-convolution area and the de-convolution area percentage of each sub-peak over the O1s peak as follows:

Area(O_total) = Area(O₁) + Area(O₂) + Area(O₃),

(1)

%Area(O₂/O_total) = Area(O₂)/Area(O_total) × 100%,

(2)

%Area(O₃/O_total) = Area(O₃)/Area(O_total) × 100%,

(3)

As shown in Figure 2c, the %Area (O₂/O_total) and %Area (O₃/O_total) of a-IGZO:N films were 13.7% and 14.3%, respectively. However, the %Area (O₂/O_total) and %Area (O₃/O_total) of a-IZO:N films were much larger (19.9% and 35.4%), implying that there were more V_O and trap states in the a-IZO:N films relative to those in the a-IGZO:N films. In addition, the a-IZO:N films tended to adsorb much more moisture gas at the film surface than the a-IGZO:N films.

Figure 3 shows the root mean square (RMS) roughness and resistivity of a-IGZO:N and a-IZO:N thin films. The roughness of a-IGZO:N film (0.28 nm) was a little larger than that of a-IZO:N film (0.21 nm). In addition, the a-IZO:N exhibited a much smaller resistivity (2.7 Ωcm) than the a-IGZO:N (58 Ωcm ). This might be because a-IZO:N contained more V_O (electrons) than a-IGZO:N. This agreed well with the XPS measurement data (as shown in Figure 2).

Figure 4 shows the transfer curves of the AOS TFTs with a-IGZO:N/a-IZO:N (Sample I) and a-IZO:N/a-IGZO:N (Sample II) channel layers, respectively. The corresponding performance parameters were extracted and listed in

Table 1. Extracted performance parameters of the N-doped DSCL TFTs.

Sample	µFE (cm2·V−1·s−1)	SS (V/dec)	VTH (V)	ION/IOFF
I	31.9	0.8	−5.0	1.2 × 108
II	15	0.5	1.5	1.1 × 108

. Here, the µ_FE was calculated from the maximum slope, that was extracted graphically from the square root of drain current, I_DS^1/2, versus gate voltage, V_GS, in the saturation region [29,30]. The V_TH was achieved from the gate voltage value where I_DS/(W/L) = 100 nA. The SS was obtained from the half value of the difference between the gate voltages corresponding to the drain current of 10⁻¹⁰ A and 10⁻⁸ A, respectively. The on-off current ratio (I_ON/I_OFF) was defined as the ratio of the maximum turn-on current to the minimum turn-off current.

Conventional AOS TFTs, i.e., the devices with single channel-layers (a-IGZO TFTs, a-IGZO:N TFTs, etc.), generally have µ_FE of less than 10 cm²·V⁻¹·s⁻¹ [10,12,31]. The N-doped DSCL TFTs (Samples I and II), however, exhibited apparent improvements in their field-effect mobilities, as shown in Table 1. If we compare these two samples, Sample II showed even better electrical performance on the whole than Sample I, although its field-effect mobility (15.0 cm²·V⁻¹·s⁻¹) was smaller than that of Sample I (31.9 cm²·V⁻¹·s⁻¹). To be specific, Sample II exhibited a more positive V_TH (1.5 V) and a smaller SS value (0.5 V/dec) than Sample I. As shown in Table 1, I_ON/I_OFF values of both the samples were larger than 10⁸.

Figure 5 shows the transfer curves evolution of the N-doped DSCL TFTs under positive bias stress (PBS) and negative bias stress (NBS) tests for 2500 s, respectively. Here, the gate electrodes were stressed at +30 V for PBS and −30 V for NBS for a period and then the transfer curves were instantly measured. As shown in Figure 5a,b, Sample I showed more stable PBS properties with a smaller V_TH shift of +0.5 V, whereas Sample II exhibited a little larger V_TH shift (+1.5 V) for the PBS duration. However, during NBS testing (Figure 5c,d), the V_TH shift of Sample I became much worse (−9.0 V) than that of Sample II (−0.5 V). In short, Sample I showed better PBS stability (V_TH shift of +0.5 V) and Sample II exhibited much more stableelectrical properties (V_TH shift of −0.5 V) for NBS tests.

Figure 6 shows the transfer curve evolution of the N-doped DSCL TFTs under thermal stress. From here, one might notice that these two samples exhibited markedly different negative V_TH shifts. As shown in Figure 6, the V_TH of Sample I shifted up to −11.5 V as the temperature increased from 298 to 398 K, indicating that Sample I had muchworsethermal stability. In comparison, Sample II was more stable under thermal stress, with a V_TH that only shifted −2.5 V.

Here we summarize the interesting results related to the N-doped DSCL TFTs. The µ_FE of AOS TFTs could be largely improved by using the DSCL composed of a high-defect-density channel layer and a low-defect-density channel layer to form the hetero-junction structure [19,21]. In this study, we deliberately designed and prepared the N-doped DSCL TFTs (Samples I and II) using the hetero-structure channel layers composed of a 15-nm-thick a-IZO:N layer and a 15-nm-thick a-IGZO:N layer, and finally obtained good electrical performance (e.g., the large µ_FE, the proper V_TH, the small SS, and the large I_ON/I_OFF), as shown in Table 1. On the other hand, although Sample I exhibited a larger field-effect mobility (µ_FE = 31.9 cm²·V⁻¹·s⁻¹), its V_TH was rather negative (−5.0 V), which was apparently not fit for driving FPDs. Sample IIshowed the optimum device performance (µ_FE = 15.0 cm²·V⁻¹·s⁻¹, V_TH = 1.5 V, SS = 0.5 V/dec) relative to Sample I. In addition, Sample I exhibited excellent PBS stability (V_THshift = +0.5 V), whereas its NBS (V_TH shift = −9.0 V) and thermal stability (V_TH shift = −11.5 V) were poor. Compared to Sample I, more stable properties were observed for Sample II during NBS (V_TH shift = −0.5 V) and thermal stress (V_TH shift = −2.5 V) tests. Therefore, Sample II, rather than Sample I, was preferred because of its better electrical performance and stability. However, what accounted for these results? The related physical mechanisms should be ascertained by analyzing the microstructures and energy bands of the channel layers for these TFT devices.

Figure 7 shows energy band diagrams of the two samples when V_GS>>V_TH. As mentioned in the literature [15], V_O was considered as the main source of free electrons in AOS films. According to the XPS result and the film resistivity (Figure 2 and Figure 3), the a-IZO:N film contained more V_O and exhibited a smaller resistivity with respect to the a-IGZO:N film, indicating that there could be even more electrons in the a-IZO:N film than those in the a-IGZO:N film. Due to the differences in carrier concentrations, the electrons injected from the a-IZO:N layer into the a-IGZO:N layer and formed the built-in electrical voltage (V_built-in) near the interfacial region between the a-IZO:N and a-IGZO:N films. The V_built-in might cause injection electrons to accumulate in the interface region between the a-IZO:N and a-IGZO:N layers. In addition, the polarity of V_built-in was opposite for Samples I and II, as shown in Figure 7.

The N-doped DSCL AOS TFTs (Samples I and II) exhibited large µ_FE, which resulted from the presence of double channels due to the hetero-structure channel layers in these samples. As shown in Figure 7a,b, some electrons might accumulate at the interfacial region between the a-IGZO:N layer and a-IZO:N layer due to the difference of the carrier concentration, as well as the formation of V_built-in. Apparently, these accumulated electrons could also participate in current conduction, forming a sub-channel at the a-IGZO:N/a-IZO:N or a-IZO:N/a-IGZO:N interfacial regions during the operation of the TFT devices. In other words, the current conduction operation of the DSCL AOS TFTs was dominated by not only the main-channel (the interface between channel layer and gate insulator), but also the sub-channel (the interface between a-IGZO:N and a-IZO:N films). The co-contribution of these two channels in current conduction led to the large µ_FE in the hetero-structure AOS TFTs.

In particular, among the DSCL devices, Sample I exhibited larger µ_FE (31.9 cm²·V⁻¹·s⁻¹) than Sample II (15.0 cm²·V⁻¹·s⁻¹), which could be explained by the hetero structures in their energy bands when V_GS>>V_TH. As shown in Figure 7a,b, moisture gas molecules could diffuse in and out of AOS thin films, donating extra electrons. This might largely influence the device performance (µ_FE, SS, V_TH, etc.) [32]. As mentioned before, from the XPS measurements, the a-IZO:N film adsorbed much more ambient moisture gas than the a-IGZO:N films. Thus, there could be many more extra electron donations caused by moisture gas adsorption in Sample I than those in Sample II. This implied that many more extra electrons accumulated at the sub-channel in Sample I, participating in the current conduction when V_GS>>V_TH, and thus leading to a larger µ_FE and a more negative V_TH in Sample I relative to those in Sample II. According to Reference [32], moisture adsorption could bring about SS deterioration in AOS TFTs. Hence, as shown in Table 1, Sample I exhibited a worse SS value (0.8 V/dec) than Sample II, possibly due to the more moisture adsorptions in its channel layer.

For the applications of TFTs in AMLCDs and AMOLEDs, it is important to keep V_TH of 0 V (or a little larger). In our previous study [25], we designed a-IZO/a-IGZO:N TFTs, whose µ_FE was quite large (49.6 cm²·V⁻¹·s⁻¹), but V_TH was too negative (−2.3 V). In this study, however, Sample II (a-IZO:N/a-IGZO:N TFT) had a positive V_TH of 1.5 V, which might result from the fact that N-doping decreased the trap states in a-IZO and at the front channel interface [12,13].

As shown in Figure 5a,b, the N-doped DSCL TFTs (Samples I and II) showed different PBS stability. Sample I had better PBS stability than Sample II. The stability of AOS TFTs could be influenced by the ambient O₂ gas during the PBS testing [10,25,32]. The reaction might be described as:

O_2(gas) + e⁻ → O_2(solid)⁻,

(4)

The better PBS stability of Sample I could be attributed to the more moisture gas adsorptions on the film surface, which weakened the O₂ adsorption to the channel.

NBS tests might cause a moisture reaction at the back channels of AOS TFTs, which could be described as [25,32,33]:

H₂O⁻ → H₂O + e⁻,

(5)

The XPS results indicated that the a-IZO:N films more easily adsorbed the ambient moisture gas than the a-IGZO:N films (Figure 2). Figure 5c,d indicate that Sample I showed very poor NBS stability (V_TH shift = −9 V). This was likely because the sub-channel of Sample I (a-IGZO:N/a-IZO:N interface) was greatly influenced by the ambient moisture gas during the NBS testing. In contrast, Sample II showed a much smaller V_TH shift (−0.5 V) under NBS tests. This might be due to the weak moisture adsorption effect of the a-IGZO:N layer, which was consistent with the XPS data (Figure 2).

Thermal stability data in Figure 6 indicated that Sample I was unstable (V_TH shift = −11.5 V) under thermal stressing; Sample II was much more thermally stable (V_TH shift = −2.5 V). We reported that larger concentration variation of V_O in AOS TFTs could lead to serious V_TH shift during the thermal testing [29]. In addition, the literature [16,34] indicated that N-doping suppressed V_O formation in the bulk channels for the thermal tests of a-IGZO:N TFTs. Hence, the better thermal stability of Sample II might be attributed to the a-IGZO:N films, which acted as a passivation layer and suppressed the V_O formation in the a-IZO:N layers.

4. Conclusions

We prepared and characterized the N-doped TFTs with the hetero-structure channel layers composed of 15-nm-thick a-IZO:N and 15-nm-thick a-IGZO:N. Two DSCL structures (a-IGZO:N/a-IZO:N and a-IZO:N/a-IGZO:N) were designed and studied to serve as the channel layers of AOS TFTs. Generally, the N-doped DSCL TFTs exhibited a large field-effect mobility (up to 31.9 cm²·V⁻¹·s⁻¹) due to the presence of double conduction channels during electrical operations. As compared with the a-IGZO:N/a-IZO:N TFT, the a-IZO:N/a-IGZO:N TFT showed better electrical performance (µ_FE = 15.0 cm²·V⁻¹·s⁻¹, SS = 0.5 V/dec, V_TH = 1.5 V, I_ON/I_OFF = 1.1 × 10⁸) and more stable properties (V_TH shift of +1.5, −0.5 and −2.5 V, for PBS, NBS, and thermal tests, respectively). These results were likely due to the optimized hetero structure, the less ambient moisture adsorption, and the better suppression effect of the V_O formation.