1. Introduction
The current electronics era was made possible by digital systems that maximize their simplicity in converting, processing, and memorizing a wide range of data into only two available electrical signal levels (“0” and “1”). However, such a standard binary logic architecture seems to hardly meet the increasing requirements for larger information capacities, which laid out the foundation of multi-valued logic (MVL) systems with
n ≥ 3, where
n is the number of distinguishable data levels [
1,
2,
3]. To date, the realization of MVL has no fully established protocols; thus, various strategies are being pursued at the materials, device, and system levels [
4,
5,
6]. For instance, Kobashi et al., in 2018, proposed a ternary (
n = 3) inverter based on a lateral heterojunction of organic semiconductors [
7,
8]. In 2019, Yoo et al. proposed a full-swing ternary circuit using a negative transconductance heterojunction transistor with a modified structure [
9]. Such reports suggested that, up to
n = 3, a two-transistor topology can be effectively maintained (one unipolar and one anti-ambipolar). However, for a higher
n, previous research mostly adopted circuit-based implementations with the number of transistors roughly corresponding to the number of data levels [
10,
11], implying the huge technological interest in advanced designs that reduce the transistor count.
In this study, we demonstrate an aggressive projection of quaternary (n = 4) digital inverter (or QNOT gate) made of only two field-effect transistors (FETs), by carrying out physically based finite-element simulation. In addition to saving the number of transistors, our proposed logic gate circuit is built by vertical (or three dimensional, 3-D) stacking of its component transistors, also showing the high potential for area-efficient on-chip deployment. In short, this particular geometry combines the two important trending design concepts in current electronics research, namely MVL and 3-D integration, in enabling a greatly simplified QNOT gate with a more efficient use of raw materials, manufacturing resources, and physical support areas. Importantly, the systematical variation of simulation parameters provides an in-depth understanding of how such an operation is obtained and also offers practical engineering guidelines toward targeted performances, which are intended to motivate wide-ranging experimental follow-ups.
2. Methods
A finite-element drift-diffusion simulation package (Silvaco ATLAS) was used for this study. Organic semiconductors provide viable routes toward low-cost flexible electronics, by virtue of their secondary bonding nature. However, printed layers can have a feature size that is much larger than those patterned by conventional lithography. In this context, the vertical integration is of particular interest to printed organic electronics [
12,
13]. We, therefore, employed geometries and parameters relevant to organic materials and devices, yet some of the results will be extendable to other semiconductor platforms. The two-transistor, five-electrode circuit was defined according to the structure depicted in
Figure 1a. The single gate electrode embedded in the common dielectric serves as the input node, where the input voltage (
Vin) is applied. In the base structure, this input electrode is at the middle of the dielectric and is aligned to have no vertical overlap with the source and drain electrodes. The organic semiconductor thickness (
tosc) is 50 nm (base structure), and the dielectric thickness (
tdiel) is 100 nm. The channel length (
L) and width (
W) are 50 and 1000 µm, respectively. The dielectric constant of gate dielectric is 3.5. All electrodes are metallic with the work function of 4.5 eV. The p-type FET (PFET) and n-type FET (NFET) are made symmetric both structurally and energetically, with the carrier mobility of 1 cm
2 V
−1 s
−1, the bandgap of 2 eV, and the effective density of states of 10
20 cm
−3. The high-occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) energies of the respective semiconductors are systematically varied to generate different charge-injection barriers (
Eb) [
14]. The supply electrode is positioned on top of PFET, and the supply voltage (
VDD) is applied to this node, in reference to the ground on the NFET side. The two drain electrodes are declared as the single floating numerical node, where the output voltage (
Vout) is measured. The terminal characteristics (voltages and currents) along with the internal carrier distributions are self-consistently calculated for the whole circuit, by solving the coupled Poisson’s and drift-diffusion equations, under the Boltzmann carrier statistics and based on the Newton method.
3. Results and Discussion
The proposed layout in
Figure 1a is basically analyzed in our previous work on the vertical 3-D organic inverter, which focuses on the voltage-gain improvement by increasing the contact resistance of FETs [
13]. Here, we demonstrate for the first time that a substantial increase in
Eb at both FETs can dramatically transform the simple 2-level inverter into a fully functional QNOT gate, without any structural modification.
Figure 1b shows the simulation results of the voltage-transfer characteristics (VTCs) of the circuit in
Figure 1a, at
VDD = 5 V. The parametric variation of
Eb was introduced with no change in electrode work function, but by energetically moving the semiconductor HOMO and LUMO levels together keeping the same bandgap (
Figure 1b inset). The VTCs reveal that an initial departure from an ideal ohmic contact (
Eb = 0 eV) to a blocking one (
Eb = 0.4 eV) actually provides a sharper data transition from high to low
Vout [
13]. Surprisingly, it was found that an
Eb of 0.6 eV substantially changes the shape of VTC, where the four data levels (“0”, “1”, “2”, and “3”) are clearly identifiable. To gain a first insight into its mechanism, we additionally performed simulations of individual FETs. As shown in
Figure 1c, the normalized saturation-regime transfer curve of the PFET at a drain voltage (
VD) of −5 V shows a quadratic shape with regard to gate voltage (
VG) when
Eb = 0 eV, suggesting no contact resistance effects [
15,
16]. In contrast, increasing
Eb resulted in substantial deviation, exhibiting a quasi-linear behavior at
Eb = 0.4 eV. At the
Eb enabling the quaternary circuit, the drain current (
ID) showed an apparent kink (arrow in
Figure 1c), which is generally considered as the consequence of a substantial contact resistance [
17,
18,
19]. Albeit with different biasing, this abrupt slope change on the transfer curve in part rationalizes the QNOT behavior of
Figure 1b; this in fact translates into the existence of two distinct regimes with different resistance distribution ratios on a single transistor, during its on-off scan by
Vin.
As the second requirement for QNOT functionality (in addition to
Eb), the absence of geometrical overlap between the gate (input) and source/drain electrodes is addressed.
Figure 2a is the side-by-side comparison of the cases of no overlap (base structure) and the full overlap with an extended gate electrode. Clearly, there is no quaternary effect in the modified circuit with a full electrode overlap. To further our understanding of underlying mechanism, we extracted the charge-carrier distribution at each component device in a full simulated inverter (not from the individual transistor simulation such as that in
Figure 1c). Especially, the focus was placed on visualizing the internal distribution at two newly developed data states, one at
Vin = 2V (level “2”) and the other at
Vin = 3 V (level “1”) (in conjunction with
Figure 1b). Comparing
Figure 2b,c reveals one important difference, which is the apparent carrier pinch-off at the supply node (i.e., source electrode) indicated by the arrow in the no-overlap data of
Figure 2b. This actually reminds us of the special property of the source-gated transistor (SGTs), whose operation largely relies on this source pinch-off rather than the drain pinch-off prevailing in traditional FETs [
20]. While it was known that such a behavior is fundamentally associated with a high-barrier contact (which is the case here), our data additionally prove that the pinch-off is more accentuated when there is no sufficient gate-source overlap to attract the injected charge carriers toward the channel surface. More importantly, it justifies the presence and absence of the two intermediate voltage levels, since such strong pinching-off at the no-overlap circuit dictates the stable establishment of the constant resistance regime between the complete turning-on and off.
Figure 2d,e simply deliver the same message, based on the electron pinch-off at the no-overlap NFET (here the ground is the source) that forms the intermediate state between its full on-off switching.
Now, we return to the base structure (
Eb = 0.6 eV, no electrode overlap) and demonstrate the possibilities of VTC engineering. Firstly, the parameter
tosc was found to have a significant impact on the level formation, as illustrated in the simulation results of
Figure 3a upon changing
tosc from 50 to 200 nm. The four output data levels, “3”, “2”, “1”, and “0”, were extracted at
Vin = 0, 2, 3, and 5 V, respectively, from each VTC. As shown in
Figure 3b–e, the semiconductor thickness significantly influences the two intermediate data voltages, with a clearly monotonous trend, meaning that a desired value can eventually be obtained through fine tuning of the layer thickness. Meanwhile, little or no impact of
tosc was seen on the two extreme levels. This interesting phenomenon can be explained as follows. Since the QNOT behavior in our system originates from the
Eb and the contact resistance thereof [
13], an enlarged out-of-plane bulk film resistance at a thicker film can significantly modulate the VTC characteristics [
14]. More specifically, the FET contact resistance becomes larger with a higher
tosc in this staggered electrode architecture [
21,
22], affecting the establishment of the pinch-off region (see
Figure 2b,d), thus pushing the “2” and “1” levels toward
VDD/2 at the same time.
Secondly, the vertical positioning of the gate electrode was another efficient controlling factor. In fact, separately changing the PFET and NFET gate dielectric thicknesses is a traditional approach in a complementary metal-oxide-semiconductor (CMOS) binary inverter to relatively tune the strength of the transistors [
23]. In our simulation, the total thickness of the common dielectric (
tdiel) was fixed as 100 nm, yet the three in-dielectric positions of the gate electrode (center, high, and low) were compared as graphically illustrated in
Figure 4a. The high or low position was made by moving the electrode from the center position by 25-nm upwardly or downwardly, respectively.
Figure 4b shows that a systematic horizontal movement of the level “2” to “1” transition point was observed by changing the vertical gate position. This is understood as the outcome of the increase in the PFET strength (for the high position) or the NFET strength (for the low position) that redistributes the supplied
VDD at different ratios, thus modulating the mid-transition
Vin point. These results also re-confirm the contact-dominated QNOT mechanism, as evidenced by no practical change in the four data levels and the extreme transitions points, because these properties are mostly determined by the depleted source regions rather than the capacitively modulated channel regions [
19].
Regarding the quaternary inverting behavior in this study, we can make several additional notes. As the
Eb increases, the driving current (
IDD) systematically decreases because of contact resistance. This in turn provides an additional advantage of low power consumption (total power is
VDD multiplied by
IDD) in the QNOT regime (
Supplementary Figure S1). Another critical point is that a vertical coupling is essential to QNOT behavior. In our proposed structure, a strong electric field appears between the supply and ground nodes (see
Figure 1a), which helps establish an elongated accumulation channel at the source region even without a gate overlap, laying a foundation for the source pinch-off and associated quaternary operation. In stark contrast, comparative simulations showed that both the source and drain contact regions are fully depleted so that less currents flow in the horizontal inverter with no indication of pinch-off and QNOT functionality (
Supplementary Figure S2). Also note that a relatively large
tdiel is used for the relevance to organic electronics. Additional simulations revealed that the operation voltage can be scaled down even in this structure, yet the intermediate levels become indistinct at a
VDD below 3 V (
Supplementary Figure S3).