3-3 Characteristics of MILC Poly-Si TFTs Fabricated with Type-B Process According to the discussion given in last section, we have learned two important facts:
1. Because the phosphorus is an effective gettering center of Ni, a high series S/D resistance is resulted if the MILC process and phosphorus dopant activation are done simultaneously [2-1].
2. If the MILC front propagates from a wider region to a narrower region, the MILC speed will be affected, while the film crystallinity also shows dependence on the channel width.
To address the issues encountered in process Type-A, we investigate the characteristics of MILC device fabricated with process Type-B. In this process, as introduced in Chapter 2, the MILC process was done on the blanket α-Si thin film.
Because the S/D implantation and post-activation process are performed after the MILC process while the surface Ni layer has been removed, the dopant activation
process will not be seriously affected. In the following sub-sections we investigate the characteristics of the poly-Si TFT fabricated by Type-B process and check if the results can resolve the issues associated with the Type-A splits.
3-3.1 Effects of channel length
The device structure is the same as that illustrated in Fig. 3-5. Figure 3.22 shows the transfer characteristics of MILC poly-Si TFTs (Type-B) with channel length ranging from 1μm to 10μm, and identical channel width of 20μm. The MILC process was performed at 550 oC for 24 hours. Figure 3.23 re-plots the data of Fig.
3.22 normalized to the corresponding channel length. Major performance data extracted from Fig 3.22 are shown in Table 3.3. From Fig. 3.23, we can see that the on-current decreases as the channel length L is shortened. Although the trend is similar to that observed in Fig. 3-7 for Type-A devices, however, the extent of degradation is dramatically reduced. This implies that Type-B can indeed help reduce the Rsd. To confirm this inference, we extract the Rsd and the results are shown in Fig.
3.24. As can be seen in the figure, a lower Rsd 1.13KΩ was measured. This value is ≒ close to the SPC samples with the same dopant activation process (Rsd 1.1~1.6KΩ≒ , see Fig. 3.25). Figure 3.26 shows normalized transfer characteristics of the SPC samples. In this figure, minor on-current degradation for shorter channel device is also found. By comparing Fig. 3.26 with Fig. 3.23, we can conclude that the on-current degradation for shorter channel devices fabricated by either MILC process Type-B or SPC process is due to the rather high Rsd of around 1KΩ. From Fig.3.22 and Table 3.3, we can see that there is no significant difference in SS and mobility for the devices with different channel length. This indicates that Type-B devices with channel length 1~10μm have similar channel film quality. According to the analysis done in Sec.3-1, for a MILC seeding window with W/L=5μm/20μm and proceeded the
process at 550℃ for 24hours, the MILC length can reach 30μm, which is much longer than the sum of the maxima channel length(10μm) and the MIC/MILC offset(5μm)(see Fig. 3-5). The results are in agreement with the electrical characteristics of Type-B devices.
3-3.2 Effects of channel width
The device structure and layout used to investigate the effect of channel width on the characteristics of Type-B devices are the same as that shown in Fig. 3-13.
Figure 3.27(a) shows transfer characteristics of MILC TFTs with different channel width. The normalized data are shown in Fig. 3.27(b). Table 3.4 summarizes the quantified data extracted from Fig. 3.23(a). From Table 3.4, we can see that the characteristics of the devices are almost the same except for the device with the narrowest width (e.g., 0.6μm), which exhibits the best performance among the four devices in terms of SS and on-current.
The trend is significantly different from the results shown in Fig. 3.14 and Table 3-2. One possible explanation for such improvement is the presence of the parasitic conduction channels appearing at the sidewalls of the channel, as shown in Fig. 3.28. However, when we re-calculate the mobility by replacing the channel width of 0.6 μm with 0.8 μm (assuming that all of the sidewalls of the active layer serve for the channel conduction), a result of about 69.88 cm2/V*s is obtained. The value is, however, still much larger than the other samples compared in the same plot. The disparity might be related to the corners of the active layer (see Fig. 3-28). A high field-strength is expected at the corner region owing to the larger curvature. As a consequence a higher amount of inversion electrons is induced there, leading to a higher effective mobility in the simplified calculation.
In order to identify this phenomenon more clearly, SPC devices are compared.
Fig. 3.29 shows the normalized transfer characteristic of SPC sample and Table. 3.5 lists the major performance data. From Fig. 3.29, there is no significant difference among the devices. And from Table. 3.5, major differences come from the mobility.
While the mobility increase is solely due to the increase of conduction width, since the mobility ratio coincides well with the width ratio when the sidewall conduction is taken into account. Obviously, unlike the MILC samples, the enhanced corner conduction is absent for the SPC samples. One possible explanation for this disparity is the high amount of trap states contained in the SPC channel, which may impede the modulation of surface potential.
3-3.3 Ammonia plasma treatment
Figure 3.30 shows the transfer characteristics of Type-B devices with and without NH3 plasma treatment, and Table. 3.6 summarizes the major performance data. In the figure we can see that the device characteristics are not improved after 1-hour treatment, and a positive shift in threshold voltage is observed. The cause for such shift is not clear at the moment, and might be related to the plasma radiation damage. After 2-hour plasma treatment, the leakage current is effectively lowered and the subthreshold swing is improved obviously. For the MILC type-A TFT, the MILC process was done after channel definition and the MILC front was confined in the channel region. More Ni contaminants are expected to be confined in the channel region and cause large leakage current. As a result, the main trap source is the metal contaminants which are not passivated by the plasma treatment. As a result the plasma treatment can not efficiently improve the leakage current. For the MILC type-B TFT, the MILC process was done on the plane α-Si layer. The MILC front is stemmed out from the seeding window and form an elliptic shape as shown in Fig. 3-2. Most of the Ni contaminants after the MILC process are expected to be located out of the channel
region and thus the concentration of the metal contaminants inside the channel is expected to be much lower than the type-A split. However, owing to the spreading out of the MILC front in the MILC type-B split, more α-Si residues between needle-like grains at the edge of the active region of the device is expected. After dopant activation at 600℃ for 24 hours, these α-Si residues become small SPC grains and contribute additional grain boundaries. As a result, the major type of defects in the MILC Type-B is that associated with the granular structure rather than the Ni contaminants. This explains why the SS is worse for the Type-B devices (Table xx).
This type of defects is also known to be more efficiently improved by the plasma treatment, explaining why the leakage is lower than the Type-A split.
3-4 Leakage Mechanisms
From the quantified data of Tables 3.1~ 3.4, we can easily find that Type-A devices exhibit much better swing and slightly better mobility. However, Type-A devices also show an anomalously high off-state leakage. To understand the possible conduction mechanisms for the off-state leakage, we measure and analyze the transfer characteristics of the devices at various temperatures.
The following equation describes the relationship between temperature, activation energy and drain current [3-10]:
exp( a )
off o
I I E
= −KT (3-2), where Ioff is the off-state drain current, I0 is the constant, K is the Boltzman constant, T is the absolute temperature, and Ea is the activation energy. Take the nature logarithm of both sides of Equation 3-2, it can be expressed as follow:
ln( off) ln( ) (0 Ea )
I I
= + −KT (3-3).
If we plot the Arrhenius plots of off-state drain current, the activation energy can be extracted from the slope.
Figure 3.31 shows and compares the activation energy of MILC Type-A, Type-B, and SPC samples. Figures 3.31(a) and (b) are the off-state activation energy of Type-A and Type-B devices, respectively, under forward and reverse modes of measurements at Vd = 5V. From the figures, we can see that Type-A has a high peak value of activation energy and the value is significantly affected by the gate bias.
Moreover, the difference between the forward and reverse modes is significant. For Type-B, the Ea in the off-state is in the range between 0.3 and 0.6 eV and shows much weaker dependence on the bias condition, while the difference between the two modes is small. As Vgd (=Vg-Vd) increases and a strong field is developed in the channel, the tunneling distance from the valence band to conduction band is reduced, and trap-assisted tunneling may dominate for the case with sufficiently high trap state density. This will lead to a lower Ea. In our samples, such situation is more significant in Type-A devices. This is attributed to the high amount of Ni residues left in the channel after the MILC treatment. In addition, owing to asymmetrical arrangement of the seeding window in the sample characterized in Fig.3-31(a), higher amount of Ni residues is expected in the channel region near the source under the forward mode (or near the drain under the reverse mode). As a consequence the reverse mode shows a higher rate in Ea lowering than the forward mode as the Vg becomes more negative, since the high-field region is located at or near the drain junction.
Such high-field dependence of Ea is relaxed in the case of Type-B devices. As has been discussed in previous sections, although Type-B devices exhibit worse crystallinity than Type-A counterparts, much reduced amount of Ni residues is also resulted due to the unconfined MILC direction. This is also evidenced from Fig.
3-31(b) in which the difference between the two measurement modes is small. Owing
to the Ni contamination issue, a high GIDL current is observed in Type-A devices, as shown in Fig. 3-32.
In Fig. 3-31(c), the Ea of SPC sample is also shown as a reference for comparison with MILC Type-A and Type-B. From this figure, we can see that the curve for SPC sample is closer to that of Type-B device especially in the lower field region, implying the two splits have similar leakage mechanisms. Since the SPC sample does not contain Ni species, the results support the aforementioned inference.
In the plot we can see that the Ea drops dramatically and begins to deviate from the result of SPC sample as Vg is smaller than -8V. This indicates that trace amount of Ni species remaining in the channel of Type-B devices may contribute to additional leakage as the field strength is sufficiently high.
3-5 Characteristics of HC-TFTs
For MILC TFTs fabricated by process Type-B, we also applied the HC-TFT structure developed by our group [1-7] to examine the device characteristics. The structure is introduced in Chapter 2. The test pattern is shown in Fig. 3.33, in which the MILC open window is designated as the source. Since the structure actually contains several transistors, for convenience, we denote the device with its S/D arrangement along the horizontal direction as the “main transistor”, and the three monitor transistors with their S/D arrangement along the vertical direction as the “DT’,
“MT”, and “ST” in Fig. 3.33. Major structural parameters of several splits of devices are listed in Table 3.7. Using the HC-TFT measurement, we can study the location-dependent MILC film properties.
3-5.1 HC-TFTs for Large Scale Devices (BE1_1)
At first, devices with larger dimensions are investigated. These devices are
labeled as BE1_1, and the detailed structural dimensions are given in Table. 3.7. The forward- and reverse-modes Id-Vg characteristics for the main transistor of a BE1_1 device are shown in Fig. 3.34. We can see that the characteristics of the device under forward mode almost coincide with those of reverse mode.
Fig. 3.35 compares the transfer characteristics of the main transistor and the three monitor transistors and the extracted performance data are shown in Table. 3.8.
The W/L ratio for the main transistor (about 0.47) and for terminal channels (about 0.42) are close so the transfer characteristics of different devices in the figure can be compared directly. From Fig. 3.35, we can see that the characteristics for the main transistor and the drain-side monitor transistor almost fall onto each other. Moreover, their characteristics are the worst in terms of the largest swing, the smallest on-current, and the highest threshold voltage, among the results shown in the figure. This implies the film quality in the main channel near the drain region is the worst. From the results presented in Sec. 3.1, the MILC length is estimated to be around 22 μm. Based on the structural parameters shown in Table. 3.7, the MILC front should stop in the channel region between the MT and the DT. This inference actually reflects on the device characteristics. As can be seen in Fig. 3.35 and the data listed in Table 3.8, the ST exhibits the best performance owing to its location being closest to the seeding window.
The off-state leakage characteristics are also investigated. Figure 3.36 shows the activation energy (Ea) of leakage current for the three monitor transistors measured at Vd=5V. In the figure, we can see that the decay rate of Ea is more significant for the ST and MT devices in the regime Vg < -10 V. From the analysis made in last section, we can consider this phenomenon an indication of higher Ni contamination. Such finding also supports the inference made in last paragraph that the MILC region actually does not cover the DT channel region. The analysis also
clearly demonstrates the effectiveness of the HC-TFT structure in resolving and identifying the film properties and the resultant electrical characteristics.
Figure 3.37 depicts the density of states (DOS) extracted from the three monitor transistors. The DOS value of the DT in the mid-gap range is larger than that of either ST or MT. This fact also agrees with the results discussed above.
Figures 3.38(a), (b), and (c) are the SEM pictures of ST, MT, and DT, respectively, taken after Secco etching [3-11]. We can see that the ST channel region is entirely MILC-dominated, but the grain direction is random. For the MT channel region, the grain size seems to be slightly larger than that in the ST channel region, and the direction of grains appears to be more uniform than ST. From the SEM image of the DT channel region, we can see the features of SPC grains [3-12]. This feature is believed to be the cause for the worse device performance of the DT.
3-5.2 HC-TFTs with Small Dimensions (BA1 and BB1 splits)
In this sub section, we present and discuss the characteristics of BA1 and BB1 HC- TFTs which have smaller dimensions than the BE1_1 split discussed above. First, let us concentrate on BA1 split, with its device dimensions detailed in Table. 3.7.
Figure 3.39 shows transfer characteristics of a BA1 device (main transistor) under forward and reverse modes of measurements, and Fig. 3.40 shows transfer characteristics of the three monitor transistors inside the BA1 structure. The major difference between the forward and reverse modes of measurements is the higher GIDL leakage current under the reverse mode in off-state regime. This phenomenon is similar to that shown in Fig. 3.16, owing to the identical cause that the junction closer to the seeding window contains higher Ni contamination. From Fig. 3.40, we can see that the on-state current of three transistors are almost the same, but we can easily tell that the DT has the largest subthreshold swing and leakage current among
the three monitor transistors. This fact suggests that, the ST channel has better film quality while the film quality decays when the channel location is farther away from the MILC seeding window. This is reasonable because, for the sub channels crystallized by MILC process later, they are affected by the SPC mechanism more seriously. And as mentioned in Sec. 3.1, Ni concentration at the MILC front is gradually reduced as the front is moving out from the seeding window. Moreover, as shown in Fig. 3.41, the boundary between the MILC and SPC looks pretty rough.
Some α-Si residues may be left behind at this boundary region. Such α-Si residues will become small SPC grains after the 600℃ (12 hours) dopant activation process and contribute to additional grain boundaries. Not only the on-characteristics but also the off-current will be affected if the device channel is built over this region. This is evidenced in Fig. 3-40. Figure 3.42 shows the activation energy (Ea) of drain leakage current for the three monitor transistors at Vd=5V. The difference observed in Fig.
3.36 shows up again in the figure.
Next, we discuss the BB1 devices. As shown in Table 3.7, this split has the same structural parameters as the BA1 split except with the three times larger MILC seeding window length and channel width for the main transistor (WG). Figure 3.43 shows the transfer characteristics of a BB1 device (main transistor) under forward and reverse modes of measurements, and Fig. 3.44 shows transfer characteristics of the three monitor transistors inside the BB1 structure. From the figures, we can see that the difference in the transfer curves between forward and reverse modes of the main transistor (Fig. 3.43) and among the monitor transistors (Fig. 3.44) is reduced as compared with the case of the BA1 device. In Sec.3.1, we have shown that the MILC length will be affected by the seeding window length. As a result, the MILC front is expected to pass the channels faster and farther in the BB1 device than in the BA1 device. In Fig. 3.4, for a 5μmX5μm MILC seeding window (similar to that in the
BA1 structure) annealed at 550℃, the MILC speed and length are clearly much smaller than those with a 5μmX15μm MILC seeding window (similar to those in the BB1 structure). The higher speed in the BB1 device suggests that the film properties inside the channel are much uniform as compared with that in the BA1 device. As a result, the three monitor transistors in the BB1 structure exhibit similar characteristics, although the DT still depicts a higher leakage current than the ST. Figure 3.45 shows Ea for the leakage current of the three monitor transistors. Similar trend as that shown in Fig. 3.42 is observed.
Chapter 4 Conclusions
4-1 Conclusions
In Section 3-1, the MILC length has been measured as a function of temperature, thin film thickness, and window size. These parameters are all important
In Section 3-1, the MILC length has been measured as a function of temperature, thin film thickness, and window size. These parameters are all important