Diamond & Related Materials 80 (2017) 14–17

Contents lists available at ScienceDirect

Diamond & Related Materials

journal homepage: www.elsevier.com/locate/diamond

Prediction of highly scaled hydrogen-terminated diamond MISFET MARK

performance based on calibrated TCAD simulation
Hiu Yung Wong⁎, Nelson Braga, R.V. Mickevicius
Synopsys, Inc., United States

A R T I C L E I N F O A B S T R A C T

Keywords: TCAD simulation is calibrated to experimental IDVD, IDVG and fT data from an H-terminated diamond MISFET
Diamond from the literature. With one set of self-consistent parameters, experimental data are all well matched. In the
Hydrogen-terminated calibration process, TCAD is used to study the possible cause of current degradation in some of the devices
TCAD simulation fabricated in the same process and it is conjectured that the degradation of low field mobility of the two-
RF
dimensional hole gas (2DHG) under the gate is the main cause of current degradation. With the calibrated
Power device
parameters, a highly scaled MISFET is simulated and it is predicted that with the given fabrication process, one
can achieve fT > 100 GHz if the gate length can be scaled down to ~0.1 μm.

1. Introduction
Diamond is a promising material for power and radio frequency
(r.f.) applications due to its wide bandgap, high thermal conductivity,
and high electron mobility [1].
A promising and widely studied diamond device is the hydrogen
terminated Metal-Semiconductor Field Effect Transistor (H-MESFET),
in which hydrogen termination is used to induce a two-dimensional
hole gas (2DHG) as high as 1014 cm− 2, resulting in high drain current
[2–5]. To reduce the gate leakage and improve the stability of H-pas-
sivation, MISFETs with gate insulators are also widely studied [6–8]
and are expected to exhibit superior performance compared to MESFET
counterparts if a high quality dielectric can be found.
Currently, feasible fabrication steps for diamond devices are much
more limited compared to silicon devices. Moreover, the H-termination
process also requires special care in the fabrication flow, which further
increase the barrier to exploring device optimization through experi-
ments. For example, the lift-off process is generally used to define the
gate length (LG) and gate to source (LGS) and drain (LGD) distances
[6–8], which results in large LG, LGS and LGD. In addition to that, dia-
mond substrates are expensive. Therefore, TCAD simulations are very
useful to help optimize devices and guide future development [9].
In the literature, TCAD simulations have not yet been widely used in
the development of diamond devices because of some of the new unique
physics (e.g. hydrogen-termination with negative electron affinity) and
its novelty [5,10]. In [5], we proposed a model to explain the abnormal
Schottky gate behavior in H-terminated MESFET (i.e. without gate in-
sulator). In this paper, we calibrate TCAD simulations to experimental
data (ID-VG, ID-VD, fT) of a high performance H-diamond MISFET [6]
and then run further TCAD simulations to predict the performance of
highly scaled H-terminated MISFETs. The experimental data used is
based on [6].
2. Simulation setup
Hirama et al. have demonstrated record high drain current in a H-
diamond MISFET device using nitrogen-dioxide treatment and Al2O3
passivation with drain current reaching 1.3 A/mm [6]. This device is
dubbed as “device1” in the following discussion. However, a different
device with smaller drain current, dubbed as “device2”, was used for fT
measurement in the same paper. This might be due to the instability
and lack of reproducibility of the devices in this immature technology.
Moreover, the device dimensions are still large (LG is about 0.4 μm, LGS
and LGD are about 1 μm), probably due to fabrication limitations.
Therefore, TCAD simulations are calibrated to the available experi-
mental data and used to predict the performance of the best transistor
when it is highly scaled.
Fig. 1 shows the structure used in simulations. The structure is
constructed based on the process flow given in [6]. Since hydrogen
termination treatment in general will form an interfacial layer (IL), this
is included in the simulation. In the experiment, Au was used to form
drain and source contacts. Therefore, the workfunction of these con-
tacts in simulations is set to 5 eV and non-local barrier tunneling based
on WKB approximation is used for hole to tunnel between diamond
2DHG and contacts through the IL. The gate workfunction is set to
4.1 eV as Al is used. The 2DHG is induced by setting an interfacial fixed

⁎
Corresponding author.
E-mail address: hywong@synopsys.com (H.Y. Wong).

http://dx.doi.org/10.1016/j.diamond.2017.10.004
Received 22 August 2017; Received in revised form 15 September 2017; Accepted 5 October 2017
Available online 07 October 2017
0925-9635/ © 2017 Elsevier B.V. All rights reserved.
H.Y. Wong et al. Diamond & Related Materials 80 (2017) 14–17

Fig. 1. H-Diamond MISFET used in the simulations. The figure is not drawn to scale for
clarity. See Fig. 6 for figure drawn to scale.

charge at the IL/Diamond interface (Q1) as it was done in [11]. How-

Fig. 2. Experimental and simulation IDVG of the “device1” in [6], with VD stepping
ever, it is found that it is impossible to match the experimental through − 12 V, −10 V, − 6 V, − 3 V, −2 V and − 1 V. The experimental data was
threshold voltage (VTH) and Hall-effect measurement 2DHG density extracted from the experimental IDVD curves in [6].
(i.e. 2DHG outside the gate region) simultaneously if homogeneous
2DHG is induced within and outside the gate region at zero gate bias.
Therefore, additional charge (Q2) is added at the Al2O3/IL interface
(only under the metal gate as shown in Fig. 1) so that VTH and the hall
measurement of 2DHG can be matched simultaneously. Such setup can
be justified if we postulate that, during fabrication, the formation of Al-
gate on top of Al2O3 causes damage or structure modification to the
part under the gate, resulting in different effective fixed charges under
the gate and outside the gate (e.g. − 1.85 × 1013 cm− 2 and
− 4 × 1013 cm− 2 respectively, for “device1” in Table 2). Electrical
contact resistances (RC) are included as lumped elements in simulations
as the experimental S/D contact resistance is substantial.
TCAD Sentaurus is used in simulations [9]. Electron/hole con-
tinuity, Poisson, and thermodynamic (for self-heating) equations are
solved self-consistently with the tunneling equation. Fermi Dirac sta-
tistic is used and high field saturation of carrier mobility is turned on.
However, an effective constant low field mobility based on experiments
is used because the lack of understanding of scattering mechanisms in
such devices. An effective thermal resistance (RTH) is also attached to
the substrate to model the thermal resistance due to the substrate.
Fig. 3. Experimental and simulation IDVD of the “device1” in [6], with VG stepping from
− 5 V, 11 V at step of 2 V.
3. Calibration

Experimental IDVG curves, which are not given in [6], are extracted
Firstly, bulk diamond parameters were calibrated to the literature.
from IDVD curves. This step is essential because calibration is more ef-
This includes temperature dependent bangap [12], hole/electron ef-
fective if one calibrates the VTH using IDVG curves first. After IDVG ca-
fective density of states [13], hole/electron temperature dependent low
libration, the original IDVD curves are used to compare to simulation
field mobilities [14,15] and high field mobilities [14,16]. Table 1 shows
and shows good agreement (Fig. 3). Since the purpose is to calibrate the
some of the important calibrated parameters for bulk diamond.
device to predict both DC and r.f. performance for a highly scaled
Then, the IDVG (Fig. 2) and IDVD (Fig. 3) curves of the “device1” in
version of “device1”, we further calibrate model parameters using IDVG
[6] is fitted based on the bulk parameters and those given in [6].
and fT curves of “device2”.
Fig. 4 shows experimental and simulation of the IDVG curves of
Table 1
Extracted bulk diamond physical parameters from experiments in the literature. “device2” and Fig. 5 shows those of fT.
The same set of parameters are used in “device1” and “device2”
Parameters Values except for those necessary to account for the change/degradation in
“device2”. Parameters used are summarized in Table 2 and will be
Band gap at 300 K 5.46 eV
Hole density of states at 300 K 1.8 × 1019 cm− 3 discussed in Section 4.
Electron density of states at 300 K 6.41 × 1019 cm− 3
Hole low field mobility at 300 K 2483 cm2/Vs
Electron low field mobility at 300 K 2326 cm2/Vs
4. Discussion and prediction
Hole saturation velocity 〈100〉 1.1 × 107 cm/s
Beta 〈100〉 1.4 Figs. 2–5 show that TCAD can be calibrated to experimental DC and
Hole saturation velocity 〈110〉 1.1 × 107 cm/s r.f. data well. Based on Table 1 and Table 2, it is believed that the
Beta 〈110〉 1.1
physics is properly captured in the models and setup because most of
Electron saturation velocity 〈100〉 1.5 × 107 cm/s
Electron saturation velocity 〈111〉 1.6 × 107 cm/s the critical and ascertained experimental parameters are not changed or
need only slight adjustments in TCAD simulations.

15
H.Y. Wong et al. Diamond & Related Materials 80 (2017) 14–17

Fig. 6. H-Diamond MISFET temperature distribution at VD = − 12 V, VG = − 5 V which

gives ID = 1.3 A/mm.
Fig. 4. Experimental [6] and simulation of IDVG of “device2” for VD = − 10 V.

Fig. 5. Experimental [6] and simulation of fT as a function of gate voltage of “device2” in Fig. 7. Predicted peak drain current (VD = −12 V, VG = −5 V) and fT as a function of
ON-state. LG for LGS = LGD = 0.5 μm for “device1”.

Table 2
Experimental and simulation device parameter values used. N/A means not available or
not applicable. ⁎RSD is the resistance between the gate and source or drain (i.e. outside of
gate region) not including contact resistance (RC).

Device parameters Experimental value Simulation value

LG (μm) 0.4 0.4

LGS and LGD (μm) ~1 0.9
tox (nm) 17 17
tIL (nm) N/A 0.2
Al2O3 εr N/A 6.17
S/D RC (Ωmm) (per side) 2.2 – RSD⁎ 0.9
Hole Mobility cm2/Vs (device1, outside N/A 110
of gate)
Hole Mobility cm2/Vs (device2, outside N/A 110
of gate)
Hole Mobility cm2/Vs (device1, under 110 110
gate)
Hole Mobility cm2/Vs (device2, under N/A 29
gate)
Q1 (cm− 2) 2DHG = 4 × 1013 −4 × 1013
Q2 (cm− 2) (device1) N/A 2.15 × 1013 Fig. 8. Predicted peak drain current (VD = −12 V, VG = −5 V) and fT as a function of
Q2 (cm− 2) (device2) N/A 2.3 × 1013 LGS (= LGD) for “device1” with LG = 0.4 μm.
Rth (cm2 K/W) N/A 3 × 10− 5
Hole saturation velocity cm/s N/A 1.265 × 107
Beta N/A 0.84 Firstly, we discuss the difference between the extracted bulk para-
meters in Table 1 and device simulation parameters used in Table 2.
The largest difference is the low field hole mobility. The hole mobility
used in device simulation is much lower than that of bulk (110 cm2/Vs

16
H.Y. Wong et al.
vs. 2483 cm2/Vs). This is expected because in these devices, holes flow
at the surface where they are scattered severely due to defects at the IL/
Diamond interface. There can be also surface roughness scattering [17]
and remote Coulomb scattering [18]. However, these scattering me-
chanisms are not well-studied in Diamond MESFETs/MISFETs. For-
tunately, it appears from the matching that an effective low field mo-
bility is enough to fit experimental data well (110 cm2/Vs).
Then we discuss the difference between the device experimental
parameters and the parameters used in TCAD simulations in Table 2.
The largest difference and uncertainty is in the contact resistance used.
In [6], the contact resistance and source-to-gate and drain-to-gate re-
sistances (RSD in Table 2) are lumped together in the measurement and
the total value is 2.2 Ωmm on each of the source and drain sides. Since
it is not clear how much contribution is from source-to-gate and drain-
to-gate resistances, it is assumed that the mobility outside of the gate is
the same as that under the gate for “device1” (i.e.110 cm2/Vs). It is
then found that the contact resistance needed to fit the experiment is
0.9 Ωmm, which is smaller than the measured lumped value, and
doesn't contradict the measurement result.
As showed in Table 2, most parameters are the same for “device1”
and “device2” in the simulation and the experimental data of both
devices can be fitted well simultaneously. The major difference is the
low field mobility under the gate, that “device2” mobility is only about
26% of that of “device1”. From this, we conjecture that the degradation
of “device2” is mainly due to the degradation of the effective mobility.
However, the physical cause cannot be inferred.
With confidence in the TCAD simulation setup, we first examine the
self-heating in diamond MISFET. Fig. 6 shows the lattice temperature
distribution of “device1” at VD = − 12 V, VG = − 5 V which gives the
record high ID (1.3 A/mm). As expected, self-heating is not significant
in Diamond devices compared to other power devices such as those
made of GaN, SiC and Si. The peak temperature is only ~320 K.
However, self-heating helps fitting the experimental data (Figs. 2 and
3). It is found that the shape of IDVD family of curves cannot be fitted
well by tuning Beta alone without self-heating.
Then we use TCAD to study the change of peak ID at VD = −12 V,
VG = − 5 V and peak fT as a function of LG, LGD and LGS. Fig. 7 shows
the change of peak current and peak fT as a function of gate length with
fixed LGS = LGD = 0.5 μm and RC = 0 Ωmm (i.e. assume perfect con-
tact). It can be seen that it is possible to attain higher than 100 GHz of fT
when LG is short enough. It has been experimentally shown that fT starts
saturating to ~50 GHz when LG < 0.2 μm for a similar device but it is
suggested in the same paper that this is probably due to parasitic ca-
pacitance [19]. The simulation here shows that the intrinsic fT of highly
scaled device may be much higher than experimentally measured value.
It should also be noted that at LG = 0.1 μm, short channel effect be-
comes more significant and the VTH is about 2 V more positive than
long channel device, which also contributes to the large increase of
drain current and fT. Fig. 8 shows the change of peak current and peak
fT as a function of LGS (LGD) with LG = 0.4 μm and RC = 0 Ωmm. The
peak fT doesn't have significant increase even if LGS = LGD is reduced to
0.1 μm. It is also found that, compared to the experimental data from
“device2” (Fig. 5), fT of the better performing “device1” could be twice
as much as that of “device2” based on our simulations.
5. Conclusion
TCAD simulations have been calibrated to IDVG, IDVD and fT curves
17
Diamond & Related Materials 80 (2017) 14–17
of a high performance H-terminated diamond MISFET and a consistent
set of parameters is presented. It is found that self-heating is relatively
small compared to what is observed in devices made of other materials
due to diamond's high thermal conductivity. However, self-heating still
must be accounted for in the simulation to match experimental data
well. Based on TCAD simulations, it is conjectured that the degradation
of device is due to the degradation of 2DHG mobility. Using the cali-
brated data, TCAD simulations are used to predict the performance of
highly scaled devices and it is shown that scaling of LG is much more
important than scaling of LGD and LGS. An intrinsic fT as high as
100 GHz can be attained when LG ~0.1 μm.
References
[1] R.S. Sussmann, CVD Diamond for Electronic Devices and Sensors, 1st ed., Wiley,
London UK, 2009.
[2] Claudio Verona, Walter Ciccognani, Sergio Colangeli, Fabio Di Pietrantonio,
Ennio Giovine, Ernesto Limiti, Marco Marinelli, Gianluca Verona-Rinati, Gate-
source distance scaling effects in H-terminated diamond MESFETs, IEEE Trans.
Electron Devices 62 (4) (April 2015) 1150–1156.
[3] D. Kueck, A. Schmidt, A. Denisenko, E. Kohn, Analysis of passivated diamond
surface channel FET in dual-gate configuration - localizing the surface acceptor,
Diam. Relat. Mater. 19 (2010) 166–170.
[4] M. Kubovic, M. Kasu, Y. Yamauchi, K. Ueda, H. Kageshima, Structural and electrical
properties of H-terminated diamond field-effect transistor, Diam. Relat. Mater. 18
(5–8) (May–August 2009) 796–799.
[5] Hiu Yung Wong, Nelson Braga, R.V. Mickevicius, “A Physical Model of the
Abnormal Behavior of Hydrogen-Terminated Diamond MESFET,” To Appear in
Simulation of Semiconductor Processes and Devices (SISPAD), (2017).
[6] Kazuyuki Hirama, Hisashi Sato, Yuichi Harada, Hideki Yamamoto, Makoto Kasu,
Diamond field-effect transistors with 1.3 A/mm drain current density by Al2O3
passivation layer, Jpn. J. Appl. Phys. 51 (2012) 9R.
[7] K. Hirama, H. Sato, Y. Harada, H. Yamamoto, M. Kasu, Thermally stable operation
of H-terminated diamond FETs by NO2 adsorption and Al2O3 passivation, IEEE
Electron Device Lett. 33 (8) (Aug. 2012) 1111–1113.
[8] Zeyang Ren, Jinfeng Zhang, Jincheng Zhang, Chunfu Zhang, Shengrui Xu, Yao Li,
Yue Hao, Diamond field effect transistors with MoO3 gate dielectric, IEEE Electron
Device Lett. 38 (6) (June 2017) 786–789.
[9] Synopsys Inc., Mountain View, CA, USA, Sentaurus™ Device User Guide, (2017).
[10] Xi Zhou, Sacharia Albin, A new capacitance-voltage model for hydrogen-terminated
diamond Mosfet, Electrochemical Society Transactions 66 (2015) 287–296.
[11] Hiroshi Kawarada, Tetsuya Yamada, Dechen Xu, Hidetoshi Tsuboi,
Yuya Kitabayashi, Daisuke Matsumura, Masanobu Shibata, Takuya Kudo,
Masafumi Inaba, Atsushi Hiraiwa, Durability-enhanced two-dimensional hole gas of
C-H diamond surface for complementary power inverter applications, Sci Rep 7
(2017) 42368, http://dx.doi.org/10.1038/srep42368.
[12] C.D. Clark, P.J. Dean, P.V. Harris, Intrinsic edge absorption in diamond, Proc. Roy.
Soc. London A277 (1964) 312–329.
[13] C.J. Rauch, A.C. Stickland (Eds.), Proc. of Int. Conf. on the Physics of
Semiconductors, Exeter, UK, Institute of Physics and the Physical Society of London,
1962, pp. 276–280.
[14] F. Nava, C. Canali, C. Jacoboni, L. Reggiani, S.F. Kozlov, Electron effective masses
and lattice scattering in natural diamond, Solid State Commun. 33 (1980) 475–477.
[15] L. Reggiani, D. Waechter, S. Zukotynski, Hall-coefficient factor and inverse valence-
band parameters of holes in natural diamond, Phys. Rev. 6 (B28) (1983)
3550–3555.
[16] L. Reggiani, S. Bosi, C. Canali, F. Nava, S.F. Kozlov, Hole-drift velocity in natural
diamond, Phys. Rev. 6 (B23) (1981) 3050–3057.
[17] C. Lombardi, S. Manzini, A. Saporito, M. Vanzi, A physically based mobility model
for numerical simulation of nonplanar devices, IEEE Trans. Comput. Aided Des. 7
(11) (1988) 1164–1171.
[18] Hiroyoshi Tanimoto, Masaki Kondo, Toshiyuki Enda, Nobutoshi Aoki,
Ryosuke Lijimat, Takeshi Watanabe, Mariko Takayanagi, Hidemi Ishiuchi,
“Modeling of Electron Mobility Degradation for HfSiON MISFETs,” in International
Conference on Simulation of Semiconductor Processes and Devices (SISPAD),
Monterey, CA, USA, (September 2006).
[19] H. Kawarada, High-current metal oxide semiconductor field-effect transistors on H-
terminated diamond Surs and their high-frequency operation, Jpn. J. Appl. Phys. 51
(2012) 090111.