3.1 Leakage Current and Mechanisms
The major off-state currents of fabricated NW-TFTs are related to several paths. Figure 3-1 illustrates two possible regions dominating the leakage conduction in our novel NW-TFTs, according to the previous study [28]. One is the drain/channel junction (path 1) and the other is the gate-to-drain overlap region (path 2).
The conduction mechanism in the drain/channel junction is via trap-assisted conduction, which is strongly dependent on the magnitude of the drain bias. Figures 3.2(a)-(c) illustrate three cases according to the strength of electric field at the drain/channel junction.
Under low electric field, electrons are thermally excited from the valence band into the midgap states. Then the trapped electrons are emitted to the conduction band (Fig. 3-2(a)). This is called “pure thermal emission” or “thermal generation”.
Under medium electric field, the drain bias pulls the energy band downward at drain side. Electrons via thermal excitation from the valence band to the trap states
can tunnel to the conduction band through the reduced barrier width (Fig. 3-2(b)).
This is called “thermionic field emission”.
Under high electric field, the energy band is pulled further downward at the drain side. The electrons can tunnel from the valence band to the conduction band with the aid of the trap states. This is called “field emission” or “tunneling”.
The conduction mechanism in the gate-to-drain overlap region is strongly dependent on the strength of local field which is determined by the gate and drain biases. This is known as the gate-induced drain leakage (GIDL). Figure 3-3 shows the mechanism in different local field strength.
When the voltage difference between the drain and gate (|VGD|) is high, the strong electric field would lead to trap-assisted tunneling (Fig. 3-3(b)) or band-to-band tunneling (Fig. 3-3(c)).
The major conduction path can be identified by investigating the dependence of device leakage current with the same channels on the gate-to-drain overlap area.
For the leakage mechanism through path 2, the dependence is linearly proportional to the gate-to-drain overlap area. For the leakage mechanism through path 1, the leakage current is independent of the gate-to-drain overlap area.
To identify the leakage path, the normalized leakage currents of the fabricated NW-TFTs measured at VG = -8V and VD = 3V are expressed as a function of “gate
width”, as shown in Fig. 3-4. The “gate width” refers to the planar width of the gate pattern and the main gate-to-drain overlap area is proportional to the “gate width”, as shown in Fig. 3-4(b). From Fig. 3-4, we can see that the off-state current is proportional to the gate-to-drain overlap area. This suggests that the major conduction of off-current is through the gate-to-drain overlap area (path 2).
3.2 Basic Transfer Characteristics
The operation principles of the suspended NW-TFTs are similar to the suspended gate (SG)-MOSFETs. In this work n+ poly-Si side-gate is used to modulate the channel potential. Thus the electrical force between the charges present at the gate and the suspended channels pulls the suspended channels toward the gate dielectric and turns on the device. Figs. 3-5 & 3-6 show the transfer characteristics of two suspended NW devices with different air-gap thickness (comparisons of the transistors’ major parameters are given in Table 3-1) but same nitride thickness of 19.8 nm. In Fig. 3-5, the device with 4.5 nm air gap thickness shows on/off current ratio of 4.64×105 and minimum S.S. of 185 mV/dec at the low-current subthreshold regime. In Fig. 3-6, the device with 25.5nm air gap thickness shows on/off current ratio of 4.7×105 on/off
ratio and minimum S.S. of 142 mV/dec. In the figures the extracted S.S. values are unusually small based on the previous experience of the related studies carried out in our group. To make this clear, Fig. 3-7 shows the transfer characteristics of a device without stripping the TEOS oxide (i.e., non-suspended channel). The device thus has gate dielectric consisting of 35nm oxide/11nm nitride. Although, due to the air gap, the nominal equivalent oxide thickness (EOT) of the device shown in Fig. 3-6 (110.4 nm) are much thicker than the device characterized in Fig. 3-7 (40.7 nm), the S.S. is not reduced accordingly. Such phenomenon is postulated to be related to the action of the suspended channels during operations.
However, the switching behaviors of the two devices shown in Figs. 3-5 and 3-6 are not as abrupt as that of the suspended gate (SG) MOSFET (i.e. S.S. almost 0) [18].
As can be seen in the figures, the fabricated devices show lower S.S. at the low subthreshold current regime. Figure 3-8 shows the S.S. as a function of ID for the devices characterized in Figs. 3-6 and 3-7. Throughout the ID range we can see the S.S.
of the suspended channel device is always lower than that of the device with oxide/nitride gate dielectric. For the suspended channel device, the lowest S.S. point (142 mV/dec) in Fig. 3-7 occurs when the suspended NW channels contact the gate nitride, and then the S.S. increases with ID when ID is below 1×10-10 A, similar to that reported on T-FET [29]. The increase in S.S. with ID implies the suspended NW
channels are further attracted toward the gate nitride and the portion of the NW channels in contact with the gate nitride increases gradually. One interesting point worth noting is that the S.S. seems to “oscillate” when ID is larger than 1×10-10 A for the suspended channel device, a phenomenon not clearly exhibited for the other device shown in Fig. 3-8. This is attributed to the oscillation of the NW channels due to the interaction of elastic force and electrostatic force as VG is sweeping, as shown in Fig. 3-9. This can be regarded as another indication of the action of the suspended channel.
Figs. 3-10 and 3-11 illustrate the transfer characteristics of NW devices with air gap of 4.5 nm and 22.5 nm, respectively. In the two figures the NW channels of the devices are formed with two different over-etch time. As mentioned in last chapter, the over-etch time during the NW-channel formation controls the dimension of the NW channel. Longer over-etch time implies smaller NW dimensions which may provide different elastic constant. Both Fig.3-10 and Fig. 3-11 show slightly lower pull-in voltage and lower S.S. for the devices with longer over-time (see Tables 3-2 and 3-3). Here the pull-in voltage is defined as the gate voltage with the smallest S.S.
in the measurements. Moreover, in Fig. 3-11 significant difference in on-current between the two devices is also observed: despite the slimmer NW channels, the one shows an order of magnitude higher in on-current (VG = 7V) than the other. This is
suspected to be caused by the fact that a larger portion of the slimmer NW channels is connected with the gate nitride as the gate voltage is high due to their higher flexibility, hence the drive current is larger. Such difference is not obvious in Fig. 3-10 owing to the much smaller air gap. The above discussion implies that the suspended NW channels are not entirely in contact with the gate nitride even when the device is turned on. The higher EOT in the channel regions near the drain side and source side may limit the on-current.
Figure 3-12 shows the transfer characteristics of the suspended-channel devices with different channel length. As can be seen in the figure, the pull-in voltage decreases as the channel length increases. This is owing to the presence of the air gap and therefore the large EOT (110.4 nm) which results in the severe short-channel effects.
3.3 Hysteresis Phenomenon
3.3.1 Characteristics of Hysteresis
In this section, we discuss the hysteresis phenomenon of the suspended-channel device under consecutive forward sweep and reverse sweep measurements. An
VG sweeping from -2 V to 7 V as those addressed in previous section. The reverse sweep measurement refers to the VG sweeping from 7 V to -2 V. It can be seen that the transfer characteristics of the reverse sweep do not coincide with the trace of forward sweep, resulting in the hysteresis phenomenon. In the example shown in Fig. 3-13, a Vth hysteresis window larger than 2 V is obtained.
Fig. 3-14 illustrates the transfer characteristics of a device with different sweeping range of VG. Again, each set of transfer characteristics consists of the forward and reverse sweeping curves. As can be seen in the figure, the three forward sweep curves with a narrower sweeping range are basically overlapping with the one with the widest sweeping range. For the reverse sweeping, the curves are separate and dependent on the largest VG of the sweeping range (i.e., the starting point). As can be seen in the figure, at a fixed VG the drain current is higher for the curve swept from a higher VG value, implying a larger portion of the NW channels contacting with the gate nitride. Starting from a higher VG implies a stronger electrostatic force exerted on the suspended NW channels due to a higher amount of charges induced in the NW channels. To release those charges (and thus the NW channels), a higher amount of shift in VG would be needed to release the suspended NW channels.
To test the reproducibility, Fig. 3-15 illustrates the characteristics of three consecutive operations performed on a device. Highly reproducible results are
demonstrated in the figure. It also shows that the suspended channel would not stick to the gate nitride in the reverse sweeping.
The characteristics of hysteresis window with different air gap thickness (4.5 nm and 22.5 nm) are compared in Fig. 3-16. The hysteresis window is defined by the VG
difference at Id = 1×10-9 A, and the extracted data are listed in Table 3-4. The device with 22.5 nm air gap thickness shows larger hysteresis window owing to the requirement for a larger VG to pull in the suspended NW channel. The initial current of the reverse sweep is reduced and not equal to the on-current of the forward sweep.
This is due to internal program setting that does not start the reverse sweep immediately after the forward sweep. This problem can be eliminated by setting a consecutive "dual sweep" option by which reverse sweep is executed immediately after forward sweep (as evidenced in Fig. 3-14).
Fig. 3-17 shows the transfer characteristics of devices with different channel length under consecutive forward and reverse sweeping measurements. The differences in transfer characteristics of forward sweeping among the devices are owing to Vth roll-off due the large EOT of gate dielectric. During the reverse sweeping, the long channel device (e.g., 1 μm) shows gradual transition in transfer
characteristics, while the two devices with shorter channel exhibit an abrupt pull-out behavior, as indicated by the arrows. This is presumably to be caused by a stronger
elastic (restorative) force in the shorter-channel devices, and the NW channels are pulled away from the gate nitride more easily as VG is swept to a sufficiently low value. Also can be seen in the figure is that the off-state current of reverse sweep is higher than that of forward sweep. A postulation about this observation is the flowing out of the remaining charges stored in the NW channels.
3.3.2 Hysteresis Mechanism
In the previous sections, the hysteresis characteristics of suspended-channel devices have been discussed. The origin of hysteresis in the poly-Si channel devices has been proposed recently [30]. The hysteresis phenomenon in poly-Si channel devices is owing to the electron trapping and de-trapping process in the traps of the poly-Si channel. However, the Vth difference between the forward and reverse sweeping is small (~ 0.4 V) [31] and cannot explain our observation in this study. As mentioned in the previous section, the fabricated devices show a much larger hysteresis window (at least 1.46 V). This is clearly related to the motion of the suspended channels during forward and reverse sweeping measurements. Based on the results presented in previous sections, the mechanism of the suspended NW
Figure 3-18 shows again the transfer curves illustrated in Fig. 3-13 to help illustrate the operation of the device described below. Figures 3-19(a)~(c) illustrate the pull-in mechanism of the suspended-channel device operated under forward sweeping. In the beginning, the device is in the off-state with a low VG (i.e. VG = 0V).
Hence the suspended NW channels and gate nitride are separated by the air gap (Fig.
3-19(a)). As the VG increases, the electrons are induced in the suspended NW channels and leads to the attractive electrostatic force between the suspended NW channels and gate. As such force is sufficiently large it could pull the suspended NW channels to contact the gate nitride. However, the contact with gate nitride occurs initially at the central region of the suspended NW channels (Fig. 3-19(b)). This leads to a sudden increase in drain current, as indicated by the point A in Fig. 3-18. As VG
increases further (region B in Fig. 3-18), the portion of NW channels contacted with the gate nitride increases gradually (Fig. 3-19(c)), further increasing the drain current.
In the beginning of the reverse sweep (region C in Fig. 3-17), the device is still in on-state and most of the NW channels remain in contact with the gate nitride. A lowering in VG tends to repel the charges stored in the channel and thus the attractive electrostatic force between the NW channels and the gate nitride is reduced. As VG is swept to a sufficiently low value (region D in Fig. 3-17), the device starts to turn off and release the electrons induced in the NW channels. The NW channels would then
be released from the gate nitride and returned to the suspended state. As the channel is shorter (see, for example, the devices with channel length of 0.4 and 0.7 micron meter in Fig. 3-17), such pull-out action may occur suddenly and lead to an abrupt decrease in drain current.
3.4 The Suspended NW Channel TFT with Larger Air Gap
From the results shown in the previous discussions, we can see that the thickness of the air gap plays an important role in the hysteresis window. In general, the transfer characteristic of the device with a thicker air gap shows a larger hysteresis window. In a pervious work [27] the results of simulation on analyzing the characteristics of SG-MOSETs also confirm this trend (Fig. 3-20). Here, we further discuss the characteristics of suspended NW channel device with a larger air gap. Figure 3-21(a) shows the transfer characteristics of a fabricated device with gate dielectric consisting of 11 nm gate nitride and 35 nm air gap. Owing to the large gap, the pull-in action of the suspended NW channel does not occur until VG = 13.6 V. Although the transfer characteristics show abrupt switching, the high VG requirement becomes a shortcoming. Another issue accompanied with the high VG is the high gate leakage
current (about 1×10-9 A), as shown in the figure. Figure 3-21(b) shows the results of the second measurement performed on the device. The overlap of the forward and reverse sweeping curves implies that the suspended NW channels are stuck to the gate nitride. This would be ascribed to the large surface adhesion force that the elastic force of the NW channels cannot overcome to return to the suspended state. The high VG during forward sweep leads to the existence of such high surface adhesion force, hence the channels cannot pull out.