1 Introduction
Two-dimensional (2D) semiconductors-based devices are promising for electronic and optoelectronic fields, such as photodetectors [
1-
3], solar cells [
4-
6], memristive [
7-
9], and neuromorphic devices [
10-
13]. Moreover, the performances of 2D devices are greatly affected by dielectric engineering [
14,
15]. For Na
3Sb field-effect transistors (FETs), the HfO
2 dielectric is used to suppress the leakage current and decrease the off-state voltage [
16]. Robert
et al. [
17] exploited TiO
2 to reduce the impact of source-to-drain tunneling currents in black phosphorous metal-oxide-semiconductor FETs (MOSFETs), making subthreshold swing (SS) approach 60.2 mV/dec. Moreover, Choi
et al. [
18] designed hetero-gate-dielectric tunnel FETs to improve on-state current (
Ion) (~30%) and suppress ambipolar behavior. The hetero-gate-dielectric structure also enhances the
Ion of the gate-all-around tunnel FETs [
19]. Therefore, dielectric engineering can enhance the gate control ability to improve the
Ion and reduce leakage current in FETs.
Gallium selenide (GaSe) [
20,
21] is a stable
p-type semiconductor with exceptional properties such as high carrier mobility and tunable energy bandgap [
22]. Previously, GaSe showed field effect mobility of about 0.6 cm
2·V
−1·s
−1, with the good on/off current ratios in the range of 10
4‒10
5 [
23]. Furthermore, GaSe showed high 2D Young’s modulus of about 819 ± 127 N·m
−1 [
24]. Based on such excellent characteristics, GaSe is a good candidate for nanoelectronic devices. Recent advances in GaSe open up new opportunities towards novel nanoelectronics devices. Chen
et al. [
25] believed that the “Mexican hat” shape valence band is suitable for FETs. The GaSe/graphene heterostructure interface possesses a controllable Schottky barrier, which is desirable in electronic devices [
26]. In recent years, large area ultrathin layers of GaSe crystals were successfully synthesized on SiO
2/Si substrates by using micromechanical cleavage technique [
27,
28]. The GaSe stacking patterns result in different energy bandgap and effective mass [
29], which affects the transport performance. However, there is limited research on the role of the dielectric in GaSe FETs.
In this work, we choose GaSe as an example to illustrate the impact of dielectric engineering. We first study the relationship between the dielectric positions and transfer characteristics. In addition, we investigate the effect of four GaSe bilayer stacking patterns as channel on the transport properties in different dielectric environments. The outstanding results provide reasonable proposals for regulating 2D nanodevices by dielectric engineering.
2 Computational details
For 2D bilayer GaSe, the geometrical optimization and electronic properties are computed with the Vienna
ab initio simulation package (VASP) code [
30-
33]. Exchange correlation energies are described by the generalized gradient approximation (GGA) using the Perdew–Burke–Ernzerh (PBE) function [
34]. The force tolerance on each atom is less than 10
−2 eV/Å, and the converged energy is less than 10
−5 eV. A 500 eV plane wave cutoff energy and a 13 × 13 × 1 Monkhorst–Pack
k-point grid are chosen for geometrical optimization. The vacuum region of 20 Å is added to avoid interaction with the adjacent periodic structure.
The device performance is simulated in Atomistix Toolkit (ATK) [
35-
37] package. Ballistic transport properties are calculated by coupling the density-functional theory (DFT) and the nonequilibrium Green function (NEGF) methods. The exchange-correlation function uses the GGA in the form of PBE, and an additional Grimme DFT-D3 function [
38-
42] is used to correct the weak van der Waals interactions. The norm-conserving pseudopotential and the basis set are PseudoDojo and Medium, respectively. The electrostatic potential is treated by solving the Poisson equation self-consistently via a real space solver with the “multi-grid” type. We set the cutoff to 110 Hartree, the
k-point grid to 1 × 15 × 250 and the temperature to 300 K. Finally, the periodic, Neumann and Dirichlet boundary conditions are used along the transverse, perpendicular and transport direction of the GaSe plane [
43,
44]. The self-consistent calculation is performed until the energy and force converge to 10
−5 eV and 0.01 eV/Å, respectively. We use the Landauer−Büttiker formula [
45-
48] to calculate the drain current at a certain bias and gate voltage:
where T (E, Vds, Vg) is the transmission coefficient at the given Vds and Vg. The fS/D and μS/D are the Fermi‒Dirac distribution functions and the Fermi level of the source or drain electrode.
3 Results and discussion
3.1 Transmission direction and doping concentration test
Before investigating the effect of the dielectric engineering on the device performance, we first explore the role of concentration and transmission direction on device performance. Fig.1(a) presents the double gate (DG) device schematic of MOSFETs based on bilayer GaSe, in which gate dielectric architecture is broken into four portions to control the dielectric layer. The length and thickness of the dielectric are 2.5 and 0.5 nm, respectively. Also, the
p-doped bilayer GaSe is used as the source (S) and drain (D), the channel is an intrinsic bilayer GaSe and band structure is illustrated in Fig.1(b). It possesses an indirect gap of about 1.50 eV, as well as the conduction band minimum (CBM) and valence band maximum (VBM) are located at the Г point and between the Г and M points, respectively, which is consistent with the report [
49]. The valence band has a “Mexican hat” edge shape, which facilitates the generation of high density of states (DOS). The band structure of GaSe exhibits the typical “Mexican hat” shape at the top of the valence band, resulting in a high density of states (DOS) [
50,
51]. The “Mexican hat” shape of the valence band edge leads to sharp peaks in the density of states [
21]. In Fig.1(c), we show the transfer characteristic curves (
Ids‒
Vg) of DG MOSFETs at different doping concentrations along armchair (Arm) and zigzag (Zig) directions, where the gate insulators are SiO
2, bias voltage is 0.64 V and the gate length (
Lg) is 5 nm. The leakage current in the off-state region obviously decreases from 10
−2 to 10
−4 μA/μm with the decline of doping concentration. Especially, the
Ids‒
Vg of both directions are similar.
From the
Ids−
Vg, we can obtain the key figures of merit, shown in Fig.1(d)‒(f). The
Ion is obtained from the transfer characteristic curves at on-state gate voltage
Von =
Voff −
Vds, where the
Vds is the bias voltage. According to the high-performance (HP) standard of the International Technology Roadmap for Semiconductors (ITRS) 2028 Version [
52], the
Voff is defined by the off-state current (
Ioff) at 0.1 μA/μm. When the doping concentration changes from 1 × 10
20 to 3 × 10
20 cm
−3, the
Ion increases rapidly. However, when the concentration is greater than 3 × 10
20 cm
−3, the
Ion fluctuates around 2000 μA/μm. To describe the gate control ability of devices,
SS is defined as the required gate voltage to change the current by one decade and can be expressed by
SS = ∂
Vg/∂lg
Ids. With doping concentration increasing,
SS changes from 69 to 75 mV/dec and the gate control capability deteriorates. Moreover, intrinsic delay (
τ) and power dissipation product (PDP) are much lower than 0.423 ps and 0.24 fJ/μm, which are the HP standards of ITRS. Thus, in terms of two-direction comprehensive performance, we finally choose the armchair direction as the transport direction and 5 × 10
20 cm
−3 as the doping concentration of the source and the drain to further study the role of high-
k dielectric engineering.
3.2 The effect of high-k dielectric position
In order to assess the effect of high-k dielectric position on the device performance, we construct hetero-gate-dielectric structures in Fig.1(a), and can be divided into two structures. For the type-A structure in Fig. S1, different dielectric materials are used for the top and bottom insulators, respectively. For type-B structure in Fig. S1, transverse dielectric is divided equally into two parts, and dielectric material are not the same. In order to concisely state the characteristics of the two structures, the dielectric positions are coded in the order of TK1, K2BK3, K4, where Ki (i = 1, 2, 3, 4) represents the dielectric position. The high-k dielectrics of Al2O3, and HfO2 are chosen for the simulation, in addition, the air dielectric is considered for comparison. When the dielectric material is air, Al2O3, and HfO2, K(i) = 0, 1, and 2, respectively. The gate length is set to 5 nm, and the doping concentration is 5 × 1020 cm−3.
To compare the transport performance of devices, we plot the Ids−Vg of type-A structures in Fig.2(a). The device with higher dielectric constant has a lower leakage current and higher Ion. In addition, Fig.2(b) and (c) display the Ion and on/off-state current ratio (Ion/Ioff). The Ioff is defined by Voff, which is obtained by scanning from 0.1 to 0.6 V with a step size of 0.1 V. The Ion of the MOSFETs with the T11B22 and T22B11 structures reach 4580 and 4614 μA/μm, respectively, far exceeding the ITRS standard. However, the value is lower than HfO2 (5084 μA/μm) and higher than Al2O3 (3936 μA/μm) as the dielectric, which are improved comparing with the situation of SiO2 in Fig.1. The Ion/Ioff of these four devices is close to 108 together with Ion close to 103 μA/μm, and has similar change as Ion. The same transmission performance of T22B11 and T11B22 structures implies that device performance is independent of the position of the top and bottom dielectric. Hence the device can achieve excellent performance and reduce the difficulty of experimental production when employing high-k material at the bottom insulator.
Fig.2(d) presents the transfer characteristic curves of type-B structure. For the asymmetric structure, the device is able to obtain a lower leakage current when the insulator with a larger dielectric constant is used on the drain side. We extract the Ion and Ion/Ioff shown in Fig.2(e) and (f). The Ion of the device with T12B12 structure is approximately 5052 μA/μm, while that with T21B21 construction only reaches 4538 μA/μm. The Ion/Ioff of MOSFETs with T12B12 structure behaves excellently compared to that of T21B21 construction. Similar phenomena can also be found in T02B02 and T20B20 devices. This shows that the device has better performance when the high-k gate dielectric is located close to the drain side, and this architecture is more suitable for making high-performance electronic devices.
To clearly illustrate the mechanism between gate dielectric position and device performance, the projected local density of states (PLDOS) and spectral current are plotted in Fig.3(a, b) and (d, e). The effective barrier height (Фh) represents the energy barrier between the maximum value of the channel VBM and the source Fermi energy level. Additionally, spectral current consists of tunneling current (Itunnel) and thermionic current (Itherm), which is separated by the Фh. As can be seen from Fig.3(a), the Фh of the device with T22B22 structure is 0.19 eV in the off-state, and Itunnel is hardly generated. Fig.3(b) presents that the barrier heights of the Al2O3 (Фhs) and HfO2 (Фhd) part are 0.13 and 0.23 eV, respectively. In the on-state, the VBM shifts upward and enters the bias window for both structures, resulting in the disappearance of the barrier. Thus, the Itherm completely dominates the total spectral current, which is comparable for both structures. Different dielectrics are used on the left and right sides for the device with T12B12 constructions, the VBM moves up different distances in different regions. Therefore, the Ion of the MOSFETs with T22B22 and T12B12 structures are almost equal, indicating that high-k materials can be saved while maintaining performance.
We plot the valence band profiles of the device in Fig.3(c) and (f) to explain the better performance of the high-k materials close to the drain than the source. When the gate voltage is ‒0.2 V, both structures in Fig.3(c) have similar barrier heights, but the width of the T12B12 structure is significantly smaller, which makes the current larger. In Fig.3(f), the barrier of the MOSFETs with T02B02 structure disappears, but that with the T20B20 construction still exists. Therefore, when the high-k material is close to the drain side, the gate has stronger electrostatic control of the channel and better performance. Additionally, when the gate voltage is applied to 0.4 V (off-state), there is almost no difference in barrier height, but an obvious distinction in barrier width. In Fig.3(f), the device with T20B20 structure only has a barrier close to the source, while the MOSFETs with T02B02 construction has a barrier close to both the drain and source, thus it can better suppress carrier tunneling and achieve a lower drain current.
3.3 The high-k dielectric on devices with different stacking pattern channel
The dielectric location can have a large impact on the device performance, but the effect of the high-
k dielectric on devices with different stacking pattern channel needs further investigation. The T
22B
22 structure is chosen and SiO
2 insulator is used to compare with case of high-
k dielectric. Fig.4(a) displays the
Ids−
Vg of the MOSFETs based on the bilayer GaSe with AA, AB, AC, and AD stacking patterns. The stacking orientations and corresponding band structures are presented in Figs. S2 and S3. Table S1 lists the detailed structural parameters. The insets of Fig.4(a) show the enlarged curves for all stackings between
Vg = 0.2 V and
Vg = 0.3 V. When HfO
2 dielectric is used, the difference in the transport performance of the devices with four structures is more obvious, especially in the off-state region. When using SiO
2 as dielectric, the stacking orders have little effect on device performance, and the largest difference of the
Ion between devices with AA and AB stacking is only 349 μA/μm. Thus, the stacking patterns have a great impact on device performance when using HfO
2 as dielectric, the difference of the
Ion between MOSFETs with AA and AC stacking is the largest (1111 μA/μm), which is larger than the SiC [
53]. Furthermore, this result can be found from the
Voff of the devices with four stacking patterns, as shown in Fig.4(c). When SiO
2 is used, the
Voff fluctuates slightly around 0.49 V, while it fluctuates obviously around 0 V for HfO
2 dielectric. These results show that when high-
k dielectric is used to improve device performance, the choice of channel stacking pattern should be paid attention to.
To gain insight into the above observations, in Fig.4(b) and (d), the transmission spectra of devices with different insulators and four stacking patterns are depicted at the gate voltage of 0.2 and −0.2 V. Regardless of the gate voltage of 0.2 or −0.2 V, the difference between the VBM of MOSFETs with four stacking is smaller for SiO2 than that of HfO2 dielectric. And with the gate voltage changing from positive to negative, the difference between the VBM of devices with four stacking at high-k will become small. In addition, when Vg changes from 0.2 to −0.2 V, the VBM of high-k shifts about 0.3 eV to the right, while it changes about 0.2 eV for SiO2. It indicates that when high-k insulator is used, device has better gate control capability. When Vg = −0.2 V, the VBM of the device with AA stacking pattern and SiO2 dielectric appears to be farther away from the Fermi level. The Ion (1605 μA/μm) of MOSFETs with AB stacking is lower than the other three structures (1954, 1879, 1691 μA/μm correspond to AB, AC, AD stackings), while all the devices meet the HP standard of IRTS, but are below the case of high-k. Thus, it is necessary to choose the appropriate stacking order when high-k dielectric is used for MOSFETs to improve performance.
3.4 Scaling of the gate length
Device performance with AA and AB stacking patterns is chosen for different gate length to investigate the role of dielectrics in size scaling. Fig.5(a)−(c) and S4 display the transfer characteristic curves of the device. When the Lg is reduced from 6 to 4 nm, the minimum leakage currents of the devices with AA and AB stacking increase by approximately 3 orders of magnitude for the HfO2 dielectric, which is 2 orders of magnitude larger than the case for the SiO2 dielectric. Thus, the lower leakage current can be obtained in the off-state region by using high-k dielectric. Additionally, when Lg is reduced to 2 nm, the device performance degenerates drastically and is difficult to reach the off-state.
Fig.5(d) and (e) present the Ion and Voff in different dielectric environments. When Lg is 6 nm, the Ion of device with high-k insulator reaches 270% (260%) of that with SiO2 dielectric for AA (AB) stacking pattern, and the other Lg devices behave similar phenomenon. Moreover, Fig.5(d) also shows that whatever the size is, stacking pattern has more obvious effects on the Ion when high-k dielectric is used. At 4 nm, the Ion difference for the MOSFETs with the HfO2 dielectric is about 2509 μA/μm, while it is only 813 μA/μm for SiO2. A similar phenomenon also is found in off-state in Fig.5(e). At the same time, the switching characteristics of the device are also discussed. Fig.5(f) also shows the SS of devices with AA and AB stacking at different Lg. When HfO2 is used as an insulator and Lg is 4 nm, the SS is close to the limit of 60 mV/dec at room temperature, which shows better gate control ability compared to that of SiO2. When the Lg is less than 3 nm, the short channel effect is more severe and the SS increases sharply, approaching 114 mV/dec for AA stacking. The difference of SS and Ion between MOSFETs with AA and AB stackings for Lg = 6 nm (4 nm) is 1 (4) mV/dec and 550 (813) μA/μm, respectively. The results show that high-k dielectric is more favorable for size reduction and excellent performance.
In consideration of the HP standard of ITRS, the Lg of the device with AA stacking can be reduced to 2 nm and Ion is close to 978 μA/μm when high-k dielectric is used. While the Lg of the device with AB stacking is reduced to 3 nm, the Ion is up to 3604 μA/μm. Additionally, for AA or AB stacking devices with SiO2 dielectric, the Lg is only reduced to 4 nm considering the ITRS 2028 standard. The AA stacking is superior in the preparation of ultra-short channel MOSFETs in a high-k dielectric environment.
4 Conclusions
In summary, we have explored the theoretical quantum transport properties of bilayer GaSe MOSFETs, considering the impact of the gate dielectric location on device performance. The device behaves best performance when the high-k dielectric is placed close to the drain. This is mainly attributed to better control over the channel when the position of high-k material is near the drain. Moreover, stacking and dielectric environments have a great influence on device performance. The largest Ion difference is about 1111 μA/μm under high-k dielectric. The device holds the Ion about 978 μA/μm when Lg = 2 nm for AA stacking. This work indicates that the dielectric environment is closely related to device performance and provides an effective direction for dielectric engineering applications.