Structure design criteria of dual-channel high mobility
electron transistors
Jia-Chuan Lin
a,*, Yu-Chieh Chen
b, Wei-Chih Tsai
c, Po-Yu Yang
daDepartment of Electronics Engineering, St. John’s University, Taiwan, ROC bInstitute of Automation and Mechatronics, St. John’s University, Taiwan, ROC c
Institute of Microelectronics, Department of Electrical Engineering, National Cheng Kung University, Taiwan, ROC
d
Department of Photonics & Display Institute, National Chiao Tung University, Taiwan, ROC Received 7 March 2006; received in revised form 17 October 2006; accepted 27 November 2006
Available online 16 January 2007
The review of this paper was arranged by Prof. Y. Arakawa
Abstract
The design criteria of dual-channel high electron mobility transistor (DHEMT) are proposed in this study. d-Doped In0.52Al0.48As/
In0.53Ga0.47As/InP material systems are concentrated in this article. The DHEMT structures are explored numerically and compared
with conventional single-channel high electron mobility transistor (SHEMT) structures. Some criteria of doping concentration and layer structure design are proposed. The simulation results reveal that DHEMT has a larger voltage swing, a lower gate leakage current, a better carrier confinement, a higher density of two-dimensional electron gas (2DEG) and an excellent transconductance than SHEMT. 2006 Elsevier Ltd. All rights reserved.
Keywords: Dual-channel; Single-channel; d-Doping; HEMT; 2DEG
1. Introduction
InP/InGaAs high electron mobility transistors (HEMTs) have superior electronic transport properties to GaAs/AlGaAs HEMTs due to the large Gamma–L band separation, low effective mass, higher low-field electron mobility, high electron saturation velocity, and higher sheet carrier densities in the InGaAs channel[1–3]. However, the high electron affinity in the InGaAs channel layer may induce impact ionization field under high electric field that would lead to a high leakage current, and hence, degrade output conductance, voltage gain, and on-state breakdown voltage considerably[4].
Dual-channel structure based on modulation-doped GaAs/AlGaAs material system was first proposed in 1984
[5]. By the well controlling on the two-dimensional electron gas (2DEG), a larger output current and higher transcon-ductance can be obtained in dual-channel high mobility electron transistors (DHEMTs) [6,7]. It attracted much attention for both analog and digital applications [1]. In this study, DHEMTs based on d-doped In0.52Al0.48As/
In0.53Ga0.47As/InP heterostructures are concentrated. It is
investigated numerically and compared with conventional single-channel high electron mobility transistor (SHEMT) structures. Some criteria of doping concentration and layer structure design are proposed. It reveals that the dual-channel structure could reduce leakage current remarkably and upgrade the device performance. The transconduc-tance may go up to 1836 ms/mm.
2. Structure design and simulation
Two-dimensional device simulator, MEDICI, is used to solve the Poisson’s equation and the electron/hole current
0038-1101/$ - see front matter 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.sse.2006.11.016
*
Corresponding author. Tel.: +886 02 28013131x6130; fax: +886 02 28013142.
E-mail address:jclin@mail.sju.edu.tw(J.-C. Lin).
continuity equations. Boltzmann transport theory is also applied. The carrier densities and electronic band struc-tures are calculated. The influences of d-doping concentra-tion and posiconcentra-tion, gate width, spacer thickness, etc. on the device performances are explored.
Fig. 1shows the schematic cross-section of (a) SHEMT and (b) DHEMT. InGaAs cap layer (250 A˚ ) doped with sil-icon at 3· 1017cm3to provide good source/drain Ohmic contacts are shown. The well designed pinch-off voltage can be obtained by the optimization of the thickness and doping concentration in the InAlAs barrier layer. In
Fig. 1a, the active layer is composed of an In0.52Al0.48As
undoped barrier layer (250 A˚ thick), the d-doping concen-trations are 1· 1018
cm3 for top d-doped layer, and 3· 1017cm3 for bottom d-doped layer., an undoped spacer layer (50 A˚ thick) for reducing Coulomb scattering [8], and an In0.53Ga0.47As undoped channel layer (50 A˚ thick).
In addition, an undoped InP buffer layer (1000 A˚ thick) is placed on the bottom layer. In order to keep the strong sen-sitivity of the gate control on the channel, the spacer must be as thin as possible. Furthermore, compared to
modula-tion-doped structures, d-doped structures have several merits, such as (1) high 2DEG concentration, (2) high elec-tron mobility, (3) large breakdown voltage, (4) more linear and higher transconductance, (5) high drain current capa-bility, and (5) low leakage current[9,4]. On the other hand,
Fig. 1b shows the layer structure with a duplicate of the active layer of which an InAlAs barrier layer (200 A˚ thick), an undoped spacer layer (50 A˚ thick) and a channel layer with thickness (50 A˚ ) are composed. Moreover, the perfor-mance of the devices depends strongly on the structures of the recess configuration. A narrow recess leads to low gate–drain breakdown voltage, whereas a wide recess introduces a current limiter, especially at the source. There-fore an asymmetric recess configuration is needed. A wider recess at the drain side is used to improve breakdown voltage and to reduce feedback effect. A narrow recess at the source side leads to a low parasitic source resistance
[10].
The electron concentration distribution, n, may be cal-culated by solving the Poisson’s and Schro¨dinger’s equa-tions self-consistently [11]. Poisson’s equation is shown in Eq.(1)where n and p are the electron and hole concentra-tions, respectively.
er2
w¼ qðp n þ Nþ
D NAÞ qS; ð1Þ
where Nþ
Dand NA are the ionized donor and acceptor
con-centrations, respectively, and qsis a surface charge density.
wis the electrostatic potential. For a HEMT structure, the threshold voltage VT can be derived from Poisson’s
equa-tion and is expressed as
VT ¼ Ub DEc q qNddd es dcþ dd 2 ; ð2Þ
where Ubis the Schottky barrier height of the gate metal on
the semiconductor; DEcis the conduction band
discontinu-ity between InGaP and InGaAs. Ndand ddare the doping
concentration and the thickness of the doped barrier layer, respectively. dc is the thickness of the undoped barrier
layer; and es is the permittivity of the Schottky layer[12].
Conduction band electron tunneling effects are included in our simulation. The net tunneling current across the bar-rier layer is calculated using the independent electron approximation[13] jDT¼4pqrDOSm1kBT h3 Z Eb 0 TCðEÞ ln eðEFn1Ec1EÞ=kBTþ 1 eðEFn3Ec1EÞ=kBTþ 1 dE; ð3Þ where the integral is over the vertical kinetic energy, E, of the incident electrons. EFn1, Ec1, and m1 are the electron
quasi-Fermi level, the conduction band edge, and the elec-tron effective tunneling mass, respectively, in the emitting layer. EFn3 and Ec3 are the corresponding electron
quasi-Fermi level and conduction band edge in collecting layer. The electron charge is given by q, h is Planck’s constant,
Fig. 1. Schematic cross-sections of the In0.52Al0.48As/In0.53Ga0.47As/InP
and kBT is the thermal energy. TC is the tunneling
coeffi-cient of an electron with energy E.
Tunneling through multiple barriers is evaluated numer-ically using the Airy Transmission Matrix Technique or AiryTMT[13,14]. Tunneling through the potential barrier is treated as a scattering problem. The tunneling coefficient is calculated as TCðEÞ ¼ m0 mNþ1 kNþ1 k0 jT j2 jIj2; ð4Þ
where m0and mN+1are the tunneling masses in the source
and destination regions, respectively, and k0and kN+1are
the wave vectors of the carrier in the source and destination region, respectively.
3. Simulation results and discussions
Fig. 2 shows the energy band diagrams and the corre-sponding electron concentration distributions of (a) SHEMT and (b) DHEMT in thermal equilibrium. A chan-nel layer that is near to the gate layer would keep a good sensitivity of gate control. On the other hand, a channel layer that is far from the gate layer would induce much
more electrons in the channel because of the downward conduction band edge of the channel layer to the Fermi level from the Schottky barrier. Then, the design of dual-channel structure could come to a compromise between them. It is also clearly showed inFig. 2that the electron concentration of both two channels in DHEMT is much more than SHEMT.Fig. 3shows the sheet carrier density (ns) with gate bias of top and bottom channels of DHEMT,
respectively. It shows that the top channel is more sensitive to the gate bias and exhibits a narrow range sheet carrier density distribution. In such a normally-on DHEMT, when a small negative bias is applied, the top channel turns off first. The bottom channel does not turn off until a large negative bias is applied. An excellent transconductance could be obtained when both of the two channels turn on. Fig. 4 shows the drain–source voltages (VDS) versus
drain–source current (IDS) of (a) SHEMT and (b) DHEMT
for various gate–source voltages (VGS). DHEMT has a
lar-ger drain saturation current than SHEMT at the same gate bias and keeps a good pinch-off and breakdown voltage performance. InFig. 5, VGSversus IDSand Gmof SHEMT
and DHEMT are shown. DHEMT has a higher drain cur-rent up to 1425 mA/mm and a wider DC operation voltage 2.7 V (1.4–1.3 V) than SHEMT. The Gm of DHEMT is
up to 1836 mS/mm and higher than SHEMT. The result reaches the same conclusion asFig. 2that a deeper conduc-tion band from Fermi level in DHEMT can lead to a higher electron concentration, a higher drain current and a higher Gm. The comparison of gate leakage current of SHEMT
and DHEMT are shown inFig. 6. The gate leakage current of DHEMT is much less than SHEMT, especially in large gate bias. To get a low leakage current in HEMT, some researchers take the InP channel instead of InGaAs chan-nel [15,16]. However, it would reduce the carrier density in the channel and hence the Gm. A dual-channel structure
would provide a good choice for it.Fig. 7shows the vari-ation of Gm under different top barrier thickness (T1) and
bottom barrier thickness (T2) in DHEMTs. The maximum
Gm can reach to 1836 mS/mm in most cases. However,
Fig. 2. Simulation conduction band and electron concentration distribu-tion diagram of (a) SHEMT and (b) DHEMT in thermal equilibrium.
Fig. 3. The sheet carrier densities under various gate biases in top channel and bottom channel of DHEMT.
when T1and T2both are 25 nm, the channel is difficult to
turn off. In addition, when T1is 25 nm and T2is 20 nm,
lar-ger voltage swing could be obtained. In addition, the threshold voltage is very sensitive to the sum of top and bottom barrier thicknesses (T1+ T2). When the total
bar-rier thickness decreases, the absolute value of a threshold voltage decreases.Fig. 8shows the variation of Gmunder
different doping concentrations cases. Best voltage opera-tion ranges of Gmcan be obtained. It shows that the doping
effect on the transconductance is dominant on the top d-doped layer. However, the threshold voltage and gate operation reign are effected by bottom d-doped layer dominantly.
The design criteria identifying the structure layer design for DHEMTs are expressed as follows:
1. The addition of the second channel in HEMT structure can improve the total channel carrier density, current driving capability and transconductance. Besides, it can keep a good pinch-off feature and reduce the leakage current.
2. Asymmetric dopant structure is suggested in DHEMT. Since the gate control is more sensitive on the top chan-nel, proper asymmetry between two dopant layers can obtain a higher gain than a symmetric one. A higher doping concentration on top d-doped layer is suggested.
Fig. 4. Simulation drain–source voltages (VDS) versus drain–source
current (IDS) of (a) SHEMT and (b) DHEMT.
Fig. 5. Gate–source voltages (VGS) versus drain–source current (IDS) and
transconductance (Gm) of SHEMT and DHEMT.
Fig. 6. The gate leakage current versus gate–source voltages (VGS) of
SHEMT and DHEMT.
Fig. 7. Gate–source voltages (VGS) versus transconductance (Gm) for
various top barrier thickness (T1) and bottom barrier thickness (T2) of
3. A higher doping concentration in d-doped layer can enhance the transconductance.
4. The increase of barrier thickness can decrease the leak-age current, and increase the turn-on voltleak-age.
5. Thick spacer layer can reduce Coulomb scattering but reduce the current density.
6. Increasing the total thickness of superlattice layers in DHEMT would enhance the transconductance. How-ever, the critical thickness for lattice matching should be concerned.
4. Conclusions
The design criteria of the DHEMTs are proposed in this paper. The InAlAs/InGaAs/InP DHEMTs are simulated and analyzed by MEDICI in which a self-consistently model based on the Poisson’s and Schro¨dinger’s equations are used. The results are also compared to that in SHEMT
structure. Also, the relationship between the transconduc-tance, the barrier thickness and the doping concentration are investigated. The results lead the conclusion that the doping concentration on the top d-doped layer affects the Gmdominantly. When the doping concentration increases,
the Gmincreases. However, the threshold voltage and gate
operation reign are effected by bottom d-doped layer dom-inantly. When the doping concentration decreases, the absolute value of a threshold voltage decreases and the gate operation region becomes narrow. The leakage currents can be drastically reduced in DHEMT structure because of the effective double block from electron tunneling by the design of dual-channel (dual barrier) structure. The simulation results show that DHEMT structures have higher 2DEG density, higher sheet carrier density, high voltage swing, good carrier confinement, lower gate leakage current, high maximum drain–source current (up to 1425 mA/mm), and high transconductance (up to 1836 mS/mm). It is good for the high power device application.
References
[1] Kasemsuwan V. Circuits and systems, 1998. IEEE APCCAS 1998. The 1998 IEEE Asia-Pacific conference, November 24–27 1998. p. 783.
[2] Gupta Ritesh, Aggarwal Sandeep Kumar, Gupta Mridula, Gupta RS. Solid State Electron 2005;49:167.
[3] Chen Yeong Jia, Hsu Wei Chou, Chen Yen Wei, Lin Yu Shyan, Hsu Rong Tay, Wu Yue Huei. Solid State Electron 2005;49:163. [4] Chen Yen Wei, Hsu Wei Chou, Hsu Rong Tay, Wu Yue Huei, Chen
Yeong Jia. Solid State Electron 2004;48:119.
[5] Chihiro Hamaguchi, Kazuo Miyatsuji, Hiroki Hihara. Jpn J Appl Phys Part II Lett 1984;23:132.
[6] Sheng NH, Lee CP, Chen RT, Miller D L, Lee SJ. IEEE Electr Device Lett 1985;6:307.
[7] Tomizawa M, Furuta T, Yokoyama K, Yoshii A. IEEE Electr Device 1989;36:2380.
[8] Frank Schwierz, Liou Juin J. Modern microwave transistors: theory, design, and performance. A John Wiley & Sons; 2002.
[9] Liu Wen Chan, Chang Wen Lung, Lour Wen Shiung, Pan His Jen, Wang Wei Chou, Chen Jing Yuh, et al. IEEE Electr Device Lett 1999;20:548.
[10] Geiger D, Dickmann J, Wo¨lk C, Kohn E. IEEE Electr Device Lett 1995;16:30.
[11] Thomas Kerkhoven, Raschke Michael W, Umberto Ravaioli. J Appl Phys 1993;74:1199.
[12] Li AZ, Chen YQ, Chen JX, Qi M, Liu XC, Chen J, et al. J Cryst Growth 2003;251:816.
[13] Taurus – MediciTMUser Guide Version V-2003.12, 2003: 2–121.
[14] Lui WW, Fukuma M. J Appl Phys 1986;60:1555.
[15] Ladner C, Berthelemot-Aupetit C, Decobert J, Harmand JC, Post G, Vigier P, et al. Indium phosphide and related materials, In: 1998 International Conference, 1998. p. 505.
[16] Maher H, Decobert J, Falcou A, Le Pallec M, Post G, Nissim YI, et al. IEEE Electr Device 1999;46:32.
Fig. 8. Gate–source voltages (VGS) versus transconductance (Gm) for