1. Introduction
In recent decades, driven by the pursuit of enhanced performance in binary logic chips and increased information density, the MOSFET (metal-oxide-semiconductor field-effect transistor) dimension has been rapidly reduced. However, the modulation of carrier thermionic emission over the energy barrier controlled by the gate bias poses challenges for MOSFETs in surpassing the sub-threshold swing (SS) limitation of 60 mV/dec at room temperature. This limitation leads to a rapid increase in the leakage current and power density during the miniaturization process [
1,
2,
3,
4]. Consequently, this presents a significant obstacle to further advancement in information density and circuit performance. Although steep slope devices, such as negative-capacitance FETs (NCFETs) [
5,
6,
7,
8,
9] and tunneling FETs (TFETs) [
10,
11,
12,
13,
14], can mitigate this issue, the number of bits is inherently smaller than the number of gates in a binary Boolean logic system [
2].
Fundamentally, the most promising strategy for addressing the power challenge and significantly improving information density involves a shift from traditional binary logic to ternary logic [
15,
16,
17,
18,
19]. In binary systems, information is represented by {0, 1}, whereas in a ternary system, it is represented by {0, 1, 2} or {−1, 0, 1}. Consequently, in a ternary system, the number of ternary bits can exceed the number of gates, a crucial factor for increasing the information density. With the same number of gates, ternary logic can accommodate more information, leading to higher information density. Alternatively, to store equivalent information, ternary logic has the potential to reduce the number of required devices, pins, and connections. Ultimately, this transition can drastically decrease the overall system complexity to 63.1% [
1,
4,
20].
The advancement of the standard ternary inverter (STI) represents a crucial foundational element in the development of ternary logic systems. Its practical implementation holds immense significance and garners substantial attention within the field. Currently, two methods exist for realizing STIs. One approach involves the direct design of the STI using MOSFETs, initially preferred due to its compatibility with the CMOS platform. However, this method typically requires a larger number of MOSFETs for an STI, ranging from three to as many as five [
21,
22,
23]. Additionally, it necessitates multiple power supplies, including additional voltages such as –V
DD, 1/3V
DD, and 2/3V
DD [
22,
24], or passive elements like two additional resistors [
25]. Consequently, this not only results in a larger chip area and increased power consumption, but also elevates the complexity of system design. As a result, recent years have witnessed a decline in research focused on STIs based on this approach.
An alternative approach involves constructing the STI using ternary devices that exhibit an intermediate state between the on and off states. Unlike the method mentioned earlier, this approach avoids an increase in the required device number and design complexity. However, the intermediate state, often arising from negative differential resistance (NDR) [
26,
27] and negative differential transconductance (NDT) [
28,
29], necessitates the use of heterojunctions formed by relatively novel and immature materials. Examples of such materials include BP/MoS2 [
19,
30], BP/ReS2/HfS2 [
31], h-BN/WSe2/InSe [
32], and even certain organic materials like PTCDI-C8 [
33] and PTCDI-C13 [
34]. Unfortunately, this reliance on novel materials renders this approach incompatible with the CMOS platform and poses challenges for mass producibility.
Obviously, based on the MOSFETs, the ternary logic inverter can be compatible with the CMOS technology, but it requires a passive component or multi-valued power supply, which makes the circuit significantly complex. The other method, with the aid of the ternary device, requires only two devices for implementing one ternary inverter and does not need multi-valued power supply, but this method relies on the novel immature materials that are incompatible with CMOS technology.
To address the issues and implement a ternary inverter compatible with the CMOS platform, a novel device combining the tunneling and drift-diffusion mechanisms is proposed. The proposed TDFETs can present three states without involving any novel immature material. Based on the TDFETs, a ternary inverter is designed and investigated. Additionally, the influences of the supply voltage as well as the key parameters of the TDFETs are studied in detail.
2. Device Structure and Simulation Setup
Figure 1 shows the device structure of the proposed TDFET. The source consists of two regions with the same concentration, but different doping types. The gate is divided into two parts, one with work function WF
I and the other with work function WF
II. The corresponding channels are channel-I and channel-II. A pocket is inserted into channel-II, enabling the device to conduct a current through different mechanisms. For the convenience of subsequent analysis and description, the cut-lines that would be used later are also marked in
Figure 1. The cut lines AA
0 and BB
0 are 15 nm and 1 nm below the oxide/channel interface. The CC
0 is perpendicular to the channel direction and located at the midpoint of the pocket region. The device parameters adopted in the simulation are listed in
Table 1.
The proposed TDFET is studied using 2-D technology computer-aided technique (TCAD) tools(Sentaurus 2013). In order to account for the arbitrary tunneling barrier with a non-uniform electrical field, the BTBT model with a dynamic nonlocal path accounting for the arbitrary tunneling barrier with a non-uniform electrical field is adopted, and the nonlocal tunneling parameters [
35,
36] A
path and B
path are 4 × 10
14 cm
−3·s
−1 and 1.9 × 10
7 V·cm
−1, respectively. The Slotboom model is utilized to consider the influence of the high doping density on the band gap. Additionally, the doping-dependent mobility model, high-field velocity saturation model, and the Shockley–Read–Hall recombination models are also included.
Figure 2 presents the transfer characteristics of the n-type TDFET. It is evident that the characteristics of the TDFET differ significantly from those of the conventional TFETs and MOSFETs. A noticeable transition occurs at a turn voltage (V
turn) of about 0.6 V, resulting in three distinct states in the TDFET, namely, the off-state, the on-state, and an intermediate state between the on and off states. This tri-state behavior of the TDFET makes it highly suitable for the design of ternary logic.
This tri-state behavior is actually caused by two different conduction mechanisms, as exhibited in
Figure 3, showing the band diagrams along the cut-lines AA
0, BB
0, and CC
0. Owing to the relatively large WF
I, a high barrier of about 0.5 eV forms between the Source-N (SN) and channel-I regions at V
G = 0.3 V, as presented in
Figure 3a. This barrier blocks the electrons in the SN to the drain via the drift-diffusion mechanism. However, in this case, the conduction band (E
C) in the pocket region near the oxide overlaps with the valance band (E
V) further away, as depicted in
Figure 3b. This allows the electrons farther away from the oxide to tunnel to the region near the oxide along the cut-line CC
0, which can be verified in
Figure 4a by the high electron band-to-band tunneling rate (eBTBT) near the oxide and high hole band-to-band tunneling rate (hBTBT) a little further away from the oxide in the pocket region. The tunneling electrons near the oxide drift further right along the channel to the drain, while the holes further away from the oxide drift left to the Source-P (SP), as shown in
Figure 3c, forming the current path from the drain to the source along the dotted line with the arrow in the top figure of
Figure 4a.
With the increased V
G to 0.9 V, the E
C in the channel-I descends. The barrier between the SN and the channel-I regions decreases, and even to zero, as shown in
Figure 3a, allowing large amount of electrons in the SN region to be thermally injected into the channel-I and further drift to the drain. It should also be pointed out that the tunneling process in the pocket region still exists as shown in
Figure 3b. Therefore, in this case, the tunneling and the drift-diffusing mechanisms coexist in the device.
Figure 4b gives the distribution of the current density and tunneling rate at V
G = 0.9 V. Obviously, there are high eBTBT and hBTBT in the pocket region. But there are many more electrons thermally injected from the SN than from the tunneling process. Thus, the drift-diffusion current is much higher than the tunneling current, and the current density is mainly distributed along the drift-diffusion current path indicated by the thicker dotted line in the top figure of
Figure 4b.
Obviously, there are two conduction mechanisms in the TDFET. For VG < Vturn, the current is primarily dominated by the tunneling process in the pocket region and flows from the drain to the SP of the source. For VG > Vturn, the current is mainly governed by the drift-diffusion mechanism and flows from the drain to the SN of the source. As the drift-diffusion current is much greater than the tunneling current, a sudden increase in the current occurs near the Vturn and causes a noticeable transition on the transfer curve, forming the tri-state behavior of the proposed TDFET.
3. Ternary Inverter Based on the TDFET
To implement a ternary inverter, a p-type device with symmetrical characteristics to the n-type TDFET is designed. The structure of the p-type device is identical to that of the n-type device, except for the change in the doping type in all regions. The gate work functions WF
I and WF
II of the p-type device are adjusted as 4.47 and 5.64 eV, respectively. Additionally, the pocket concentration is set as 3.7 × 10
19 cm
−3. The other parameters remain the same as in
Table 1.
Figure 5a presents the obtained symmetrical n-type and p-type curves, which are desirable features for the inverter design.
Figure 5b presents the circuit diagram of the ternary inverter, showing a simple connection where the gates of the n-type and p-type devices are connected as the input (V
in), and the drain is connected as the output (V
out), with the source of the p-type device connected to V
DD (operating voltage of the inverter) and the source of the n-type device connected to GND, similarly to the binary inverter circuit.
Figure 6 shows the voltage transfer characteristic (VTC) of the designed ternary inverter based on the TDFETs. It can be seen that the inverter exhibits three distinct output states. The V
IL and V
IH are the maximum and minimum of the input voltage and can be considered as the logic L and H, respectively. The V
IML and V
IMH are the minimum and maximum of the input voltage and can be considered as the logic M. The V
IL, V
IH, V
IML, and V
IMH are all defined at the points where the slope equals −1. The input voltage ranges corresponding to the logic L, M, and H are calculated as R
L = V
IL, R
M = V
IMH − V
IML, and R
H = V
DD − V
IH, respectively.
The VTC of the ternary inverter with various V
DD from 0.6 V to 1.2 V is shown in
Figure 7a. Interestingly, with increased V
DD, the logic L and H show totally different changes compared to those of the intermediate logic M. The R
M gradually decreases, while the R
L and R
H gradually increase with V
DD. Even when V
DD = 1.2 V, the intermediate logics M totally disappears and the ternary inverter becomes a conventional binary inverter. The varied R
M and R
H with V
DD are extracted in
Figure 7b. Due to the symmetrical transfer characteristics of the p-type and n-type devices, the R
H ≈ R
L (as can also be seen in
Figure 7a), and therefore, R
L/V
DD is not present. We can see that as V
DD increases, the R
H/V
DD gradually increases, while the R
M/V
DD decreases. When V
DD = 1.45 V
turn = 0.87 V, R
H ≈ R
L = R
M = 0.27 V
DD, and the three levels are equiprobable, which is preferred for a ternary inverter.
To further study the effects of V
DD, the I
DS varied with the input voltage V
in is plotted in
Figure 8. Considering that in an inverter, the gate voltage of the n-type (V
Gn) and p-type TDFETs (V
Gp) are V
Gn = V
in and V
Gp = V
in − V
DD, respectively, all we need is to shift the transfer curve of the p-type TDFET by V
DD towards the positive direction to obtain the I
DS-V
in characteristics.
Figure 8a presents the I
DS-V
in curves at V
DD = 0.6 V. It can be seen that, as the V
in increases from 0.0 V to 0.2 V, the p-type device exhibits a larger current than the n-type device, and the resistance of the pull-up p-type device R
up is smaller than the pull-down n-type device R
down, which results in V
out > V
DD/2. Although the p-type current is larger than the n-type current, the maximum difference is less than an order of magnitude. This means the R
up is not small enough compared with the R
down. Therefore, the logic H cannot be built, as shown in
Figure 7a. The reason why the logic L disappears can be also explained in the similar way as the increase in Vin from 0.4 V to 0.6 V. For V
in increasing from 0.2 V to 0.4 V, both the devices operate in the transition region between the off state and the intermediate state, where the current and resistance of the devices are comparable. Thus, the output voltage is near 0.3 V (V
DD/2). This is why the inverter only presents the logic M at a low voltage of V
DD = 0.6 V
in (
Figure 7a).
To make sure the inverter can present three distinct logics, the p-type transfer curve need to be shifted further to the right. That is to say that a larger V
DD is required, as shown in
Figure 8b, in which V
DD = 0.9 V. It can be seen that when V
in is near 0V, the p-type TDFET is in the drift-diffusion region, while the n-type TDFET is in the off state. The R
up is much smaller than R
down, and the high logic, H, forms. As for 0.3 V < V
in < 0.6 V, both the n-type and p-type devices operate in the tunneling current region in which the R
up ≈ R
down and the V
out is near the V
DD/2. Thus, the intermediate logic, M, is obtained. When V
in closes to the V
DD, the p-type device enters the off-state while the n-type device enters the drift-diffusion region. Hence, the R
up is much larger than the R
down, and the V
out approaches 0 V, resulting in the low logic L. This explains why the inverter can present three distinct levels at a V
DD near 0.9 V.
However, as the V
DD continues to increase, the input voltage range R
M, over which both the n-type and p-type devices are in the tunneling current region, gradually decreases. The logic M on the VTC becomes more and more narrow. Until the V
DD = 1.2 V, as shown in
Figure 8c, the R
M decreases to 0 and the intermediate logic M completely disappears. While the V
in rises from 0.0 V to V
DD, there are only two distinct logics. Namely, when 0.0 V < V
in < V
DD/2, the p-type current is far larger than the n-type current and the R
up < R
down leading to logic H. When V
DD/2 < V
in < V
DD, the p-type current is far smaller than the n-type current and the R
up > R
down, resulting in logic L. Obviously, in this case, the ternary inverter becomes a conventional binary inverter.
Based on the analysis presented in
Figure 8, it is evident that V
DD significantly influences the performance of the inverter. If we aim to design a ternary inverter with R
L=R
M=R
H, we can deduce from
Figure 8b that V
turn = (2/3)V
DD, considering R
L + R
M + R
H = V
DD. In other words, the theoretical V
DD for an ideal equiprobable ternary inverter is V
DD = 1.5 V
turn. However, the practical scenario is that R
L + R
M + R
H < V
DD due to transitions between different logic states. This discrepancy is the reason why the equiprobable ratio R
H/V
DD = R
M/V
DD = 0.27 in
Figure 7b is smaller than the ideal value of 1/3. Additionally, it explains why the V
DD required for the equiprobable case is 0.87 V, slightly less than the theoretical value of 1.5 Vturn (0.9 V).
Obviously, we have the flexibility to adjust V
turn to achieve an equiprobable inverter for a specified V
DD of 0.9 V. Considering that the V
turn is the critical gate voltage over which the drift-diffusion mechanism starts to dominate the device current, we can modulate the WF
I to change the V
turn.
Figure 9a depicts the symmetrical transfer curves of the TDFETs with different WF
I. We can see that the absolute value of V
turn increases with the decreased WF
I for the p-type TDFET and the increased WF
I for the n-type device. As the absolute value of V
turn increases, the intermediate state of the transfer curve becomes flatter and more obvious. Thus, the input voltage range R
M over which both the devices are in the intermediate state increases and the logic M gradually widens, as presented in
Figure 9b, giving the variations in the VTC with V
turn at fixed V
DD = 0.9 V.
Figure 10a exhibits the varied R
H/V
DD and R
M/V
DD with V
turn, and we can see that, for V
DD = 0.9 V, the ternary inverter becomes equiprobable at V
turn = 0.62 V. In this case, R
H = R
M ≈ 0.27 V
DD. The corresponding VTC is plotted in
Figure 10b. R
M = R
H = R
L = 0.24 V indicates that the ternary inverter is equiprobable, and the subsequent simulations are based on this inverter.
In a TDFET, the tunneling process happens in the pocket region. Thus, the pocket doping density N
P and pocket length L
P may show important effects. The symmetrical transfer curves of the n-type and p-type TDFETs with different N
P are given in
Figure 11a. It can be observed that, as the N
P decreases, the tunneling-dominated current decreases, while the drift-diffusion-dominated current is hardly affected. This results in decreased V
turn and a weakened intermediate state between the on and off states, which further leads to gradually narrowing logic M on the VTC curve, as depicted in
Figure 11b, which shows the varied VTC curves with N
P. Until the N
P decreases to 8 × 10
18 cm
−3 for the n-type device and 5 × 10
18 cm
−3 for the p-type device, the tunneling current, as well as the tri-state behavior, completely disappear, and the characteristics of the TDFET become almost identical to those of MOSFETs. In this case, the logic M on the VTC curve also completely disappears, and the ternary inverter becomes a conventional binary inverter.
The influence of L
P is illustrated in
Figure 12. As L
P increases, the effective tunneling area and the tunneling current gradually increase. However, the drift-diffusion current, primarily influenced by the barrier between the SN and channel-I regions, exhibits almost no variation. Consequently, this results in a slight increase in V
turn, as depicted in
Figure 10a, and a subsequent slight widening of the logic M on the VTC, as shown in
Figure 10b.
Figure 13 illustrates the impact of WF
II. As the WF
II increases for the p-type device and decreases for the n-type device, the tunneling current undergoes a gradual increase, resulting in a flatter intermediate state on the transfer curve, depicted in
Figure 13a. Referencing the analysis in
Figure 8, it becomes apparent that the flatter intermediate state contributes to a wider and flatter logic M, as shown in
Figure 13b.
Figure 14a illustrates how WF
II impacts the transient characteristics of the inverter. With changes in WF
II, the tunneling current gradually decreases, leading to longer charging and discharging times for the capacitance. Consequently, the delay time for the inverter output to transition from level H to level M increases from 0.077 μs to 0.321 μs, causing a decrease in the inverter speed. Moreover, the variation in the intermediate state current affects the circuit power consumption.
Figure 14b depicts the current variation from the power supply to the ground during the inverter output switching between different levels. It is evident that when the inverter outputs levels H and L, the power supply current is relatively small, whereas for level M, the power supply current is larger. This is attributed to the fact that at levels H and L, one of the n-type and p-type devices conducts while the other is turned off, resulting in no direct current path from the power supply to the ground. However, at level M, both n-type and p-type devices are in a partially conducting state, leading to the inverter power consumption being the sum of the n-type and p-type power consumption. Furthermore, as nWF
II decreases and pWF
II increases, both the intermediate state current and the off-state current of the device increase, resulting in a gradual increase in the inverter current. In conclusion, WF
II significantly impacts the stability, speed, and power consumption of the inverter’s intermediate state. In practical design, inverters must balance stability, speed, and power consumption based on the circuit’s application scenarios.