Characteristics of cylindrical surrounding-gate GaAsxSb1−x/InyGa1−yAs heterojunction tunneling field-effect transistors
Guan Yun-He, Li Zun-Chao†, , Luo Dong-Xu, Meng Qing-Zhi, Zhang Ye-Fei
Department of Microelectronics, Xi’an Jiaotong University, Xi’an 710049, China

 

† Corresponding author. E-mail: zcli@mail.xjtu.edu.cn

Project supported by the National Natural Science Foundation of China (Grant Nos. 61176038 and 61474093), the Science and Technology Planning Project of Guangdong Province, China (Grant No. 2015A010103002), and the Technology Development Program of Shaanxi Province, China (Grant No. 2016GY-075).

Abstract
Abstract

A III–V heterojunction tunneling field-effect transistor (TFET) can enhance the on-state current effectively, and GaAsx Sb1−x/InyGa1−yAs heterojunction exhibits better performance with the adjustable band alignment by modulating the alloy composition. In this paper, the performance of the cylindrical surrounding-gate GaAsx Sb1−x/InyGa1−yAs heterojunction TFET with gate–drain underlap is investigated by numerical simulation. We validate that reducing drain doping concentration and increasing gate–drain underlap could be effective ways to reduce the off-state current and subthreshold swing (SS), while increasing source doping concentration and adjusting the composition of GaAsx Sb1−x/InyGa1−yAs can improve the on-state current. In addition, the resonant TFET based on GaAsx Sb1−x/InyGa1−yAs is also studied, and the result shows that the minimum and average of SS reach 11 mV/decade and 20 mV/decade for five decades of drain current, respectively, and is much superior to the conventional TFET.

1. Introduction

With the continuous scaling down of the feature size of metal–oxide–semiconductor field-effect transistor (MOSFET), the switching speed, density and functionality of integrated circuits have been improved greatly. But further reducing the device dimensions will increase the leakage current, which results in increased power consumption. The higher power consumption under low supply voltage results from the not so steep subthreshold swing (SS) of MOSFET, with a 60 mV/decade limit at room temperature. In order to solve this problem, new devices such as tunneling FET (TFET)[14] and impact ionization FET[5] have been proposed. TFET benefits from the quantum mechanical band-to-band tunneling, which is different from the thermal injection of carriers in MOSFET, and can obtain SS beyond the 60 mV/decade limitation.[6,7]

The on-state current ION of TFET is determined by the tunneling probability, and is much lower than that of MOSFET. The Wentzel–Kramer–Brillouin approximation provides a way to calculate the tunneling probability TWKB:

where Eg is the bandgap, λ represents the tunneling length, m* is the effective carrier mass, q is electronic charge, h is the Dirac constant, and ΔΦ denotes the energetic difference between the valence band in the source and the conduction band in the channel in the n-type TFET.[6] Equation (1) implies some methods to boost ION. So various material TFETs with different structures have been studied, including carbon nanotube,[8] PNPN,[9] and heterojunction (Si/InAs, group III–V materials)[10,11] TFETs. Compared with the homojunction TFET, the heterojunction TFET shows superior performance.[12] GaAsxSb1−x/InyGa1−yAs heterojunction TFETs allows various band alignments through changing the alloy compositions of the source and channel material, which makes it more convenient to design and optimize TFET, such as increasing ION without deteriorating off-state current IOFF and SS. The ION can be improved even to the MOSFET level by reducing the effective tunneling barrier height (Ebeff),[13] so the TFET based on GaAsx Sb1−x/InyGa1−yAs heterojunction possesses great potential applications in the low-power and energy-efficient logical circuits. In order to alleviate short-channel effects, surrounding-gate (SG) structure which has the superior gate-controllability and higher scalability is widely adopted. Previous work focused on the planar GaAsx Sb1−x/InyGa1−yAs heterojunction TFETs.[13,14] Research on the performance of the SG mixed-As/Sb heterojunction TFET with cylindrical cross-section, which can eliminate the corner effect,[15] has not been found so far in the literature. Moreover, gate–drain underlap can be efficient to suppress the IOFF in TFET.

In this paper, we study the cylindrical surrounding-gate GaAsx Sb1−x/InyGa1−yAs heterojunction TFET with gate-drain underlap by numerical simulation. The effects of device parameters and resonant tunneling on the performance of the device are investigated.

2. Device structure

The designed structure is a lateral n-type cylindrical SG-TFET which is shown in Fig. 1. The P+ source region is GaAsxSb1−x with carbon-doped concentration NA, the channel is undoped InyGa1−yAs, the N+ drain region is InyGa1−yAs with silicon-doped concentration ND, and the dielectric material used as gate oxide layer is HfO2. R is the channel radius, Lg is gate length, and Lunderlap represents the underlap length between the gate and the drain. Unless otherwise specified, the channel length Lch is 20 nm, the oxide thickness tox is 2 nm, Lunderlap is equal to 10 nm, R is 7 nm, and x and y are equal to 0.35 and 0.7, respectively. The three-dimensional (3D) device simulator Sentaurus[16] is used for simulation, and the dynamic non-local band-to-band tunneling model is activated for current calculation. Besides, under the assumption of high doping concentration in the source and drain, the band-gap narrowing model is enabled and Shockley–Read–Hall recombination model is also included. In order to consider the effect of interface trap on band-to-band tunneling in TFET, the trap-assisted tunneling model is also activated.

Fig. 1. Schematic of SG-TFET with gate–drain underlap.
3. Results and discussion

Figure 2 shows the influence of source doping concentration on the characteristics of the heterojunction SG-TFETs. From Fig. 2, it can be observed that, by increasing NA, there is enhancement in ION which results from the fact that a higher NA will make the band steeper and enhance the electric field between the source and channel as shown in Fig. 3, thus tunneling probability increases. However, at VDS = 0.5 V and VGSVOFF = 0.5 V, where VOFF is the gate voltage corresponding to IOFF = 10pA/μm, ION increases more than 30 times when NA varies from 1019 cm−3 to 5 × 1019 cm−3, much larger than the two-fold enhancement when NA varies from 5 × 1019 cm−3 to 1020 cm−3. The smaller increase in ION in the latter case results from the higher degeneration of the source region at higher doping concentration, which makes the Fermi level EFS fall below the valence band maximum EV in the source region and electrons partially occupy the states in the valence band between EV and EFS, resulting in reduced paired-tunneling states and thus smaller increase in ION.[17] Figure 4 shows the dependences of transfer characteristics on ND. It can be seen that ND hardly influences the on-state performance, since ION is primarily controlled by the tunneling junction between the source and channel. Nevertheless, IOFF increases with increasing ND because of the ambipolar behavior.[18] From Figs. 2 and 4, in contrast to MOSFETs, the current of TFET does not change linearly with gate voltage on a semi-logarithmic scale, which results from the band-to-band tunneling mechanism leading to the complex dependences of the tunneling current on energy band profile and available quantum states. The dependence of SS on ND is plotted in Fig. 5. SS is lower than 60 mV/decade for a drain current range of more than 5 orders of magnitude, and achieves a minimum of 30 mV/decade when ND is equal to 1018 cm−3. It is evident that SS degrades as ND increases.

Fig. 2. Plots of drain current versus gate voltage for different values of carbon-doped concentration NA at VDS = 0.5 V and ND = 5 × 1018 cm−3.
Fig. 3. Plots of electric field versus distance along the channel for different values of carbon-doped concentration NA at VGSVOFF = 0.5 V, VDS = 0.5 V, and ND = 5 × 1018 cm−3.
Fig. 4. Variations of drain current with gate voltage for different values of silicon-doped concentration ND at VDS = 0.5 V and NA = 1020 cm−3.
Fig. 5. Plots of SS versus drain current for different values of silicon-doped concentration ND at VDS = 0.5 V and NA = 1020 cm−3.

The influences of the gate–drain underlap Lunderlap on the characteristics of SG-TFET are given in Fig. 6. It can be seen that ION is scarcely affected by Lunderlap, but IOFF decreases with increasing Lunderlap. When VDS = 0.5 V and VGS = − 0.3 V, IOFF decreases about 300 times as Lunderlap varies from 0 nm to 10 nm. In the TFET with gate–drain underlap, there is a high-resistance region in the non-gated channel region which makes the electrical field reduced at the channel–drain junction. As a result, the ambipolar behavior which makes the main contribution to IOFF is mitigated, thus IOFF is reduced. Because the high-resistance region is near the drain, the electric field in the source–channel tunneling junction is not influenced by Lunderlap. ION is controlled by the source–channel tunneling junction, so ION is insensitive to Lunderlap. Figure 7 gives the dependences of SS on Lunderlap. The advantage of the gate–drain underlap is more evident. When Lunderlap is 10 nm, SS is lower than the thermal limitation 60 mV/decade of MOSFET for more than 2-decade drain current. Besides, the gate–drain underlap also reduces the area of gate capacity, thus reducing the gate delay.[19]

Fig. 6. Curves of drain current versus gate voltage for different values of Lunderlap at VDS = 0.5 V, Lg = 10 nm, NA = 1020 cm−3, and ND = 5 × 1018 cm−3.
Fig. 7. SS versus drain current for different values of Lunderlap at VDS = 0.5 V, Lg = 10 nm, NA = 1020 cm−3, and ND = 5 × 1018 cm−3.

Figure 8 presents the transfer characteristics of SG-TFETs at different values of Lg, in which there is no gate–drain underlap. The plot indicates that increasing gate length results in the diminution of IOFF, but unnoticeable change of ION. When Lg is reduced from 15 nm to 10 nm, due to the direct tunneling from the source to the drain, IOFF increases more than 2 orders of magnitude at VDS = 0.5 V and VGS = − 0.3 V, which is much larger than the increment when Lg is changed from 20 nm to 15 nm. The weak dependence of ION on Lg is due to the fact that the current in TFET results from the quantum tunneling at the source–channel junction. Figure 9 shows the variations of SS with drain current at different gate lengths. SS decreases with increasing Lg and reaches a minimum value of 27.7 mV/decade when Lg is 20 nm. The current range for SS below 60 mV/decade almost reaches 6 decades.

Fig. 8. Variations of drain current with gate voltage for different values of gate length Lg at VDS = 0.5 V, NA = 1020 cm−3, and ND = 1018 cm−3.
Fig. 9. Variations of SS with drain current for different values of gate length Lg at VDS = 0.5 V, NA = 1020 cm−3, and ND = 1018 cm−3.

The band alignment can be designed through modulating the alloy compositions of GaAsx Sb1−x and InyGa1−yAs to easily obtain better device performances. Figures 10 and 11 depict the transfer and SS characteristics of SG-TFETs with different alloy compositions (x = 0.5, y = 0.53 and x = 0.4, y = 0.65), at which GaAsx Sb1−x and InyGa1−yAs are lattice-matched, and Ebeff values are 0.5 eV and 0.31 eV, respectively.[13,20] Here, ION is determined at VDS = 0.5 V and VGSVOFF (at IOFF = 1 nA/μm) = 0.5 V. ION increases from 33 μA/μm to 103 μA/μm, more than 3 times, when Ebeff varies from 0.5 eV to 0.31 eV as shown in Fig. 10. Reducing Ebeff is an effective way to enhance ION, because smaller Ebeff makes it much easier for electrons to tunnel in the n-type TFET. The simulation shown in Fig. 11 demonstrates that SS becomes better with the scaling down of Ebeff. We can obtain higher drain current and smaller SS at the same time due to the various band alignments in GaAsx Sb1−x/InyGa1−yAs system, so the mixed-As/Sb material system is attractive for low-power applications.

Fig. 10. Curves of drain current versus gate voltage for different material compositions (x = 0.5, y = 0.53, Ebeff = 0.5 eV and x = 0.4, y = 0.65, Ebeff = 0.31 eV) in (dashed) linear and (solid) log scales at VDS = 0.5 V, NA = 1020 cm−3, and ND = 1018 cm−3.

Next, the effect of the resonant tunneling is investigated. Compared with the conventional TFET, the resonant TFET (R-TFET) only exchanges the materials between the source region and channel (drain) region.[21] Here, the materials used in the source region and channel (drain) region of the R-TFET are In0.7Ga0.3As and GaAs0.35Sb0.65, respectively. Figure 12 shows the lateral band diagrams of the conventional TFET and the R-TFET. In the R-TFET, there is a narrow triangle quantum well between the source and channel due to the larger electron affinity of In0.7Ga0.3As (χ = 4.651 eV) than that of GaAs0.35Sb0.65 (χ = 4.07 eV),[22] which creates a double-barrier band structure[21] as shown in Fig. 12(b). According to quantum mechanism, there are discrete energy levels in a quantum well. The way of tunneling through the double-barrier via discrete energy levels is resonant tunneling.[23] The tunneling probability, as a function of tunneling carrier energy, is like Breit–Wigner-type distribution, and can achieve maximum when the energy of tunneling carriers is equal to a quantized energy level. In Fig. 13, the transfer characteristics of the conventional TFET and the R-TFET are plotted. It is obvious that the transfer curve of the R-TFET in the sub-threshold region is much steeper than that of the conventional TFET, which results from the fact that when the energy level of tunneling carriers in the R-TFET aligns with the quantized energy level, the tunneling probability and thus the drain current increases abruptly. The drain current of the R-TFET increases from 10−8 μA/μm to 10−2 μA/μm with a gate voltage increase of 110 mV. The minimum SS of the R-TFET is 11 mV/decade and SS is below 20 mV/decade over 2.5 decades of drain current as shown in Fig. 14. The average SS of the R-TFET calculated in a drain current range from 10−7 μA/μm to 10−2 μA/μm, is merely 20 mV/decade, 30 mV/decade lower than the conventional TFET. Because of its excellent sub-threshold performance, the R-TFET may have greater potential applications in very low-voltage and low-power though its on-state current is lower than that of the conventional TFET.

Fig. 11. Curves of SS versus drain current for different material compositions at VDS = 0.5 V, NA = 1020 cm−3, and ND = 1018 cm−3.
Fig. 12. Energy band diagrams of (a) the conventional TFET and (b) the R-TFET with Lg = 10 nm at VGS = 0 V.
Fig. 13. Transfer curves of the conventional TFET (C-TFET) and the R-TFET, where VDS = 0.5 V, R = 3 nm, NA = 1020 cm−3, and ND = 5 × 1018 cm−3.
Fig. 14. Curves of SS versus drain current of the C-TFET and R-TFET, where VDS = 0.5 V, R = 3 nm, NA = 1020 cm−3, and ND = 5 × 1018 cm−3.
4. Conclusions

The characteristics of cylindrical SG GaAsx Sb1−x/InyGa1−yAs heterojunction TFET with gate–drain underlap are investigated. The effects of doping concentration, gate–drain underlap, gate length, and the compositions of GaAsx Sb1−x and InyGa1−yAs on the performance are analyzed. It turns out that better SS can be obtained through lower drain doping concentration and longer gate–drain underlap, and the drain current can be enhanced by modulating the alloy composition or increasing source doping concentration. In addition, a novel TFET called R-TFET is studied based on GaAsx Sb1−x/InyGa1−yAs staggered-gap heterojunction, and the result shows that the sub-threshold performance is superior to that of the conventional TFET due to its resonant tunneling mechanism. The minimum SS of the R-TEFT is only 11 mV/decade, and its average SS is 20 mV/decade for 5 decades of drain current, 30 mV/decade lower than those of the conventional TFET.

Reference
1Li RLu Y QZhou G LLiu Q MChae S DVasen TWan S HZhang QFay PKosel TWistey MXing H LSeabaugh A 2012 IEEE Electron Dev. Lett 33 363
2Li Y CZhang H MZhang Y MHu H YWang BLou Y LZhou C Y 2013 Chin. Phys. 22 038501
3Liu YHe JChan MDu C XYe YZhao WWu WDeng W LWang W P 2014 Chin. Phys. 23 097102
4Choi W YLee W J 2010 IEEE Trans. Electron Dev 57 2317
5Björk M THayden OSchmid HRiel HRiess W 2007 Appl. Phys. Lett 90 142110
6Ionescu A MRiel H 2011 Nature 479 329
7Liu YWang H JYan JHan G Q 2013 Chin. Phys. Lett 30 088502
8Koswatta S OKoester S JHaensch W2009International Electron Devices MeetingDecember 7–9, 2009Baltimore, USA37.5.1
9Abdi D BKumar M J 2014 IEEE Electron Dev. Lett 35 1170
10Riel HMoselund K EBessire CBjörk M TSchenk AGhoneim HSchmid H2012International Electron Devices MeetingDecember 10–13, 2012San Francisco, USA16.6.1
11Dey A WBorg B MGanjipour BEk MDick K ALind EThelander CWernersson L 2013 IEEE Electron Dev. Lett 34 211
12Wang L QYu ETaur YAsbeck P 2010 IEEE Electron Dev. Lett 31 431
13Mohata D KBijesh RMujumdar SEaton CEngel-Herbert RMayer TNarayanan VFastenau J MLoubychev DLiu A KDatta S2011International Electron Devices MeetingDecember 5–7, 2011Washington DC, USA33.5.1
14Kazumi OMotohiko FShinjiro IYasuyuki M 2015 Jpn. J. Appl. Phys 54
15Najmzadeh MSallese J MBerthome MGrabinski WIonescu A M 2012 IEEE Trans. Electron Dev 59 3519
16Synopsys Sentaurus H-2013.03
17Wang L QAsbeck P2009IEEE Nanotechnology Materials and Devices ConferenceJune 2–5, 2009Traverse City, USA196
18Schmidt MMinamisawa R ARichter SSchäfer ABuca DHartmann J MZhao Q TMantl S 2012 Appl. Phys. Lett 101 123501
19Verhulst A SVandenberghe W GMaex KGroeseneken G 2007 Appl. Phys. Lett 91 053102
20Mohata DRajamohanan BMayer THudait MFastenau JLubyshev DLiu A W KDatta S 2012 IEEE Electron Dev. Lett 33 1568
21Avci U EYoung I A2013International Electron Devices MeetingDecember 9–11, 2013Washington DC, USA4.3.1
22http://www.ioffe.rssi.ru/SVA/NSM/Semicond
23Sze S MNg K K2007Physics of Semiconductor Devices3rd ednNew JerseyWiley454