Effects of Deposition Oxygen Flow

Effects of the deposition oxygen flow had been studied previously [24-26] and known to greatly affect the electrical performance of TFTs. Some of those studies reported devices with transfer characteristics showing a hump phenomenon [47], but none of them explained the cause for the observation. In this work, effects of the oxygen flow are also explored. We deposited the channel layer with various oxygen flow rates which are 0, 1, 3 and 5 sccm, respectively. In addition to studying the basic electrical characteristics, we’ve also observed the hump phenomenon and attempted to understand its origin. Figs. 3.1 (a)-(d) show the variation of the transfer characteristics of a-IGZO devices with channel prepared with oxygen flow at 0 sccm, 1sccm, 3 sccm, and 5 sccm, respectively. As we can see from these figures, the variation of the transfer characteristics is the smallest at deposition O₂ flow rate of 1 sccm and getting worse with either increasing or decreasing the deposition oxygen flow.

In Fig. 3.2, the mean transfer curves of a-IGZO devices selected from Figs. 3.1 (a) ~ (d) with channels prepared with different O2 flow rate are shown and compared. The Vth (subthreshold swing, SS) are 2.65 (246), 2.42 (167.7), 2.83 (323.3) and 3.69 V (492.6 mV/dec) for devices with O2 flow of 0, 1, 3, 5 sccm, respectively. On/Off ratios are in the range from 10⁷ to 10⁹ and mobilities are 11.45, 12.13, 9.6 and 0.69 cm²/Vs for O₂ flow at 0, 1, 3, 5 sccm, respectively. The Vth shifts positively as O2 flow increases from 1 to 5 sccm or decreases from 1 to 0 sccm. The mobility (μ), SS and Ion/off ratio are shown as a function of O₂ in Figs.

3.3 (a)-(b). From the above results, we conclude that the condition with O2 flow of 1 sccm is the best in terms of the device performance and variation control.

16

Since the oxygen vacancies are found to act as the shallow donor states for AOS materials, the higher O₂ flow rate during deposition will compensate the oxygen vacancies, and accordingly reduce the free carriers and lower the conductivity of the channel layer [24-26]. This trend explains the increase with increasing O₂ flow rate shown above. This can also be proved by the voltage dependent mobility and series resistance (Rs) shown in Figs. 3.4 (a) and (b), respectively, which are extracted based on the method reported in [22-23]. It is seen that the mobility decreases while the Rs increases rapidly as we increase oxygen flow from 1 sccm to 5 sccm. This is due to the higher amount of free carriers in the channel layer when deposition oxygen flow is less. However, the trend is not obvious in the range of O₂ flow from 0 to 1 sccm. This might be related to the huge variation in device performance for the case with zero O₂ flow.

Figs. 3.5 (a)-(b) show the comparison of two transfer characteristics measured at (a) V_d=0.1, 1 V and (b) V_d=1, 4 V. An increase in V_d will boost the on current and the Vth shift is not significant when we increase the drain voltage. The tremendous increase in the off current at V_d of 4 V is due to the gate leakage current (analysis not shown). Fig 3.6 shows the transfer characteristics measured by positive and negative voltage sweeping and the Vth shift is not obviously, either. Also the measured C-V characteristics with multi-frequency for the four splits are shown in Figs. 3.7 (a)-(b) and 3.8 (a), (b), respectively. Fig. 3.9 shows the frequency independent Cg calculated from the procedure introduced in Chap. 2 [21-23]. Here we notice that in Fig. 3.2 the transfer curves show a hump in the subthreshold region when the oxygen flow is 3 sccm or 5 sccm. Here we use the modified conductance method (MCM) proposed by Bae et al.[21] mentioned in Sect. 2.2 to extract the DOS shown in Fig 3.10. We define an energy level as E₁ in the gap corresponding to the gate voltage at a specific I_d of 10^-11 A.

We find that the DOS of defects in the gap is generally much higher in the 3 and 5 sccm splits than the other two. This explains why the SS is worse for the two splits with high O₂

17

flow. Moreover, one or two peaks in the defect distribution are observed for these samples. As the surface potential is modulated by the gate bias so as to switch the device to the ON state, the induced electrons have to fill these states first. When the Fermi level moves through the region close to a DOS peak, the abundant defects will slow down the modulation of surface potential with increasing gate bias. This is believed to be the origin for the observed “hump”

in the subthreshold region of the two splits with high O₂ flow of 3 sccm and 5 sccm.

Finally, Figs. 3.11 (a)-(d) show the output characteristics of devices with O₂ flow of 0, 1, 3, and 5 sccm, respectively.

3.2 Effects of Active Layer Thickness

Effects of channel thickness have also been studied by many groups previously [27-31].

TFTs with different channel thickness were studied here, too. Figs. 3.12 (a)-(c) show the transfer characteristics of several a-IGZO devices with channel thickness of 10 nm, 15 nm, and 20 nm, respectively. As we can see from these figures, the variation of the transfer characteristics is comparable among the three splits, although it seems to be a little worse as the channel is thinner. This might be due to the influence of the back channel surface which is expected to be more pronounced as the channel is thinned down.

Typical transfer curves of different splits are shown in Figs. 3.13(a) and (b). The Vth (subthreshold swing, SS) are 5.12 (305), 4.59 (409), and 3.3 V (430 mV/dec) for channel thickness of 10, 15, and 20 nm respectively. On/Off ratios are in the range between 10⁷~10⁸ and mobilities are 8.12, 10.9, and 11.22 cm²/Vs for channel thickness at 10, 15, 20 nm, respectively. These parameters are shown as a function of the channel thickness in Figs. 3.14 (a) and (b).

As we can see from the I_d-V_g curves (Fig. 3.13 (a)), the differences of the on current among the samples are small, implying the surface conduction dominates. On the other hand,

18

the off current, on/off ratio, SS, mobility, and threshold voltage are all thickness dependent, as shown in Figs. 14 (a) and (b). Increasing the channel thickness would draw a higher off current and worse SS since it is harder to fully deplete the channel layer.

Fig 3.15 (a) shows the mobility with respect to gate voltage for the three splits [22-23].

In Figs. 3.14 (a) and 3.15 (a), we can find the mobility is higher for the device with a thicker channel. This can be explained by percolation conduction theory [12]. The random distribution of Ga³⁺ and Zn²⁺ will form potential barriers and the barriers are overcome when the carrier concentration is sufficiently large (~10¹⁹ cm^-3), thus the mobility increases accordingly. This also reflects on the series resistance (Rs) which was extracted by the method proposed in [22-23], as shown in Fig. 3.15 (b). The Rs starts to decrease at a higher V_g for the device with a thinner channel due to its lower mobility and reduced amount of carriers. Fig. 3.13 (b) also shows the comparison of transfer characteristics of the devices with different channel thickness at V_d=0.1 and 1 V. We can see that the on current is apparently improved without worsening the off current.

The measured C-V characteristics with multi-frequency are shown in Figs. 3.16 (a)-(c).

Fig. 3.17 shows the frequency-independent Cg [21-23] where the difference in the voltage as Cg begins to increase, reflecting the shift in Vth with different channel thickness. Finally, Figs. 3.18 (a)-(c) show the output characteristics of the devices with channel thickness of (a) 10 nm (b) 15 nm (c) 20 nm.

19

3.3 Effects of Surroundings Ambient

AOS materials are very sensitive to surrounding ambient. The devices we fabricated and characterized have no passivation layer in order to study the effect of the surrounding ambient.

The component of air includes oxygen and water vapor. Impacts of these species have been studied by Jin-Seong Park [25] and Kang et al. [26]. As we can see in Fig. 3.19, after exposure to humidity surroundings the Vth shifts negatively accompanied with a dramatic increase in the off-state leakage current by about 10³ ~10⁴ times [25]. Fig. 3.20 (a) shows the transfer characteristic of a device measured with different air partial pressure [26]. They found that the Vth shifts positively as the air partial pressure increases. A similar trend is also observed as the ambient is switched to pure oxygen, as shown in Fig. 3.20 (b) [26]. These observations can be explained by the figures shown in Figs. 3.21 (a) and (b). The adsorption of O₂ near the back surface, as shown in Fig. 3.21 (a) will tend to attract electrons in the channel as stated by the following equation:

O + e → O _{( )}， (Eq.3-1) and then form a depletion layer underneath the adsorption layer, causing the positive Vth shift.

In contrast to O₂, H₂O adsorption, as shown in Fig. 3.21 (b), will donate free electrons to the channel, causing a negative shift in Vth. Such a phenomenon is also called donor effect [25].

Figs. 3.22 and 3.23 show the transfer characteristics of the fabricated devices with various channel thickness and various O₂ flow during the channel deposition, respectively, measured right after the fabrication and three months later. We find that Vth of almost all splits shifts positively in the two figures as the measurements were done three months later.

According to the studies mentioned above [25-26], we conclude that the Vth shift is mainly due to the adsorption of the oxygen molecules from the surrounding atmosphere on the back surface of the devices.

20

In Fig. 3.22, we also notice that the Vth shift caused by adsorbing oxygen is related to the active layer thickness. That is, the thicker the active layer, the harder the oxygen molecules to affect the electrical properties. With the bottom-gate scheme of the fabricated devices, the oxygen molecules would draw more influence on the gated channel conduction as the channel is thinner. In Fig. 3.23, we find that irrespective of the deposition oxygen flow rate, the Vth of all splits shifts positively except for the one with O₂ flow of 5 sccm. As have been pointed out in previous section that the channel layer deposited at 5 sccm is deficient of oxygen vacancies and thus free electrons in the channel. In this case the adsorbed oxygen is harder to take away any of the electrons in the channel, hence the Vth shift isn’t apparent. In Fig. 3.23, the dramatic increase in the off current after 3 months is presumably due to the bad quality of gate dielectric.

3.4 Impacts of Different Gate Dielectric Thickness and Channel Materials

Reduction in gate dielectric thickness is a common method to improve the TFT performance including the SS and on current. This is demonstrated in Fig. 3.24 (a) which shows the transfer characteristics of a-IGZO TFTS with gate dielectric of 10 and 50 nm. The on current and SS are significantly improved with a thinner gate oxide due to the better controllability. This is further evidenced by the data shown in Fig. 3.24 (b) that the mobility increases with gate voltage more rapidly when the gate dielectric is thinner.

So far we know that the thicker channel layer provides more free carriers in the AOS materials as discussed in Sect. 3.2. Based on the presupposition and the fact that ITO is one of the AOS materials which had been commonly used in practical manufacturing of flat-panel display products, we compared the electrical performance of these two materials. Since ITO is more conductive than a-IGZO, we fabricated the ITO TFTs with an ultra-thin active layer of 10 nm and compare their performance with the IGZO TFTs. Fig. 3.25 shows the comparison

21

of the transfer curves of ITO and a-IGZO TFTs. The on current is larger for the ITO TFT but the SS and on/off ratio are better for the a-IGZO TFT. The larger on current displayed by the ITO device is due to the higher concentration of free carriers in the active layer even though the channel thickness is less than that of the a-IGZO film. By the same token, worse off current and SS are also resulted in the ITO one. On the other hand, the better on/off ratio and SS shown by the a-IGZO TFT are due to the better gate controllability.

3.5 Effects of Using Double Active Layer

In order to suppress the off current and improve the on current, double active layer structure was explored in this work (Fig. 2.2 (j)). As has been stated in last chapter that the key parameter investigated in the deposition of the double channel layers is the O₂ flow rate.

Fig 3.26 shows the transfer curves of devices with 1^st/2^nd layer of 1/3, 3/1, 1/5, and 5/1 sccm.

Note that the 1^st layer is in contact with the gate dielectric, and the thicknesses of both 1^st and 2^nd layers are 7.5 nm. Also shown in the figure is the control device with the channel prepared by O₂ flow rate of 1 sccm. The off currents are the same for all splits but, as compared with the control device, the on currents are apparently degraded for the devices with double active layer. This indicates that the on performance is majorly dependent on the layer with a higher O₂ flow. As mentioned in Section 3.2, the total amount of the free carrier concentration in the channel is proportional to the channel thickness. The above degraded results associated with the double-layer channel might be due to ultra-thin channel. If the first layer is thickened, expected benefits of the double-layer structure should appear.

22

3.6 Effects of Annealing

In order to resolve the hump phenomenon discussed in Sect. 3.1 and improve the electrical performance, we annealed the devices in the fabrication. Effects of the treatment on the electrical performance are addressed in the section. In Sect 3.3, we speculate that the surrounding oxygen adsorbed on the back surface of the device may cause the Vth shift, thus annealing done in vacuum is explored. The annealing temperature and time are 300 ^oC and 30 minutes, respectively, done on the devices after fabrication for 3 months.

Fig. 3.27 (a) shows the I_d-V_g curves of the devices with various O₂ flows measured at different stages. As indicated previously that after exposing to the air ambient for 3 months, Vth shifts positively apparently. But after the vacuum annealing, Vth shifts negatively. This trend agrees with our speculation that the adsorbed oxygen atoms on the back surface, which cause the positive Vth shift, tend to de-adsorb during the vacuum annealing. We also notice the Vth after annealing is even smaller than the fresh value, as shown in Fig. 3.27 (b). This can be attributed to the fact that some of the oxygen atoms that penetrate into the channel and form weak bonding wherein will be released back to the surrounding [49], causing more oxygen vacancies and thus more free carriers.

Devices with double active channel layers are also examined after vacuum annealing. As mentioned in Sect. 3.5, an obvious hump is observed in some of the as-fabricated samples with double active channel layers. Such a finding is, however, not found and reported in previous studies on devices with double active channel layers [64]. After checking the process sequence, we find that the devices characterized in those previous studies all received thermal annealing. To verify if the annealing is responsible for the disparity, we also performed annealing on some of the devices with double active layer. Here we only compare the devices with channel prepared with O₂ flow of 1, 1/3 and 1/5 sccm. Fig 3.28 (a) shows the transfer characteristics of the devices for positive and negative sweeping. It is seen that the hump is

23

indeed disappear after the annealing. Moreover, the hysteresis of those devices is not significant. Fig 3.28 (b) and (c) show the transfer characteristics of a-IGZO devices after vacuum annealing for single active layer and double active layer (1/3 and 1/5 sccm) at (b) V_d=0.1V and (c) V_d=4V. The Vth values of the devices become more positively since the thinner 1^st layer with less free carriers. The SS are slightly better for the devices using double active layer and the on current and on/off ratio look fine. For the results with V_d of 4V, the off current of the device with single active layer apparently increases while it remains low for the devices with double active layer. This is attributed to the use of the 2^nd layer which can suppress the back surface leakage current.

3.7 Effects of Bias Stress and Light Stress

Since a-IGZO TFTs are expected to be widely used as the switching transistors in AM-LCD and AMOLED, the devices are usually biased and exposed to light during normal operation. Any issues related to instability caused by bias stress and/or light stress must be crucially treated and resolved. These instability phenomena are typically associated with a shift in Vth and SS degradation. The former is majorly attributed to charge trapping in the gate dielectric and a change in the carrier concentration of the channel film, while the latter is largely caused by the generation of interface states. Light stress could generate electron-hole pairs in the channel and the generated carriers would participate in the process of defect creation inside the devices.

Effects of bias conditions and light exposure on IGZO TFTs have been explored by many groups [54-62]. A number of stress modes, including positive bias stress (PBS), negative bias stress (NBS), light stress (LS), positive bias light stress (PBLS), and negative bias light stress (NBLS), are categorized in the previous works. In this work we also focus on these categories. Figs. 29 (a)~(e) show the results of characterization performed on devices

24

deposited with O₂ flow of 1sccm and channel thickness of 10 nm. Fig. 3.29 (a) shows the effect of PBS. The gate voltage applied to the gate during the stressing is 10 V, while the source and drain are grounded. The Vth shifts positively with increasing stress time and basically no clear degradations on SS and mobility are observed. The rise in leakage current is also seen and has been identified to be due to the gate leakage current, indicating the poor quality of the gate dielectric. The positive shift in Vth is attributed to the electrons trapping in the gate dielectric and a schematic illustration describing the process is shown in Fig 3.30 [60].

Accumulated electrons under the high positive gate bias are trapped in the gate dielectric and cause the positive Vth shift. Fig 3.29 (b) shows the effect of NBS. As can be seen in the figure that there are no apparent Vth shift and SS variation under NBS. The gate voltage applied to the gate during the stressing is -10 V, while the source and drain are grounded. Lacking of holes in the n-type channel of a-IGZO material was considered as the reason for the small influence. However, Nathan et al. [58] found that the Vth shifted negatively under NBS. They attributed this to the de-population of donor-like traps. As the negative bias excludes the electrons away, the empty trap is with positive charge and causing the Vth shift negatively.

Fig 3.29 (c) shows the effect of LS. The Vth shift negatively as the light stress time increases.

Two possible mechanisms are proposed previously [54-62]. One is the light-induced electron-hole pairs increase the number of free electrons in the channel, thus Vth shifts negatively. Another is that a portion of the holes induced by light are trapped in the gate dielectric, also causing Vth shifts negatively. On the other hand, Chen et al. [59] claimed that the Vth shift is related to the surrounding ambient. They found the Vth shift due to LS is relatively small under oxygen ambient. As mentioned in Sect. 3.3, the adsorbed oxygen atoms will cause a positive Vth shift. However, under the LS, photon-generated e-h pairs will supply the holes to react with the O₂^- and desorb the oxygen molecules, causing the negative Vth

Two possible mechanisms are proposed previously [54-62]. One is the light-induced electron-hole pairs increase the number of free electrons in the channel, thus Vth shifts negatively. Another is that a portion of the holes induced by light are trapped in the gate dielectric, also causing Vth shifts negatively. On the other hand, Chen et al. [59] claimed that the Vth shift is related to the surrounding ambient. They found the Vth shift due to LS is relatively small under oxygen ambient. As mentioned in Sect. 3.3, the adsorbed oxygen atoms will cause a positive Vth shift. However, under the LS, photon-generated e-h pairs will supply the holes to react with the O₂^- and desorb the oxygen molecules, causing the negative Vth