Microelectronics Reliability 110 (2020) 113698

Contents lists available at ScienceDirect

Microelectronics Reliability
journal homepage: www.elsevier.com/locate/microrel

Relative effectiveness of high-k passivation and gate-connected field plate T

techniques in enhancing GaN HEMT breakdown
⁎
Bhavana Prasannanjaneyulu , Shreepad Karmalkar
Department of Electrical Engineering, Indian Institute of Technology Madras, India.

A R T I C LE I N FO A B S T R A C T

Keywords: Prior work has shown that the OFF-state breakdown voltage, VBR, of AlGaN/GaN high electron mobility tran-
AlGaN/GaN HEMT sistors (HEMTs) can be raised by just extending the drain end of the gate as a field plate (FP) or a high-k
Breakdown voltage passivation over the AlGaN layer. However, the two techniques have not been compared, and the VBR for which
Buffer trap density their effectiveness was demonstrated was ≥120 V. High-k passivation avoids the lithography involved in rea-
Field plate
lizing a FP, and the VBR of practical devices can go down to a few 10's of volts. Hence the present work compares
High-k passivation
Impact ionization
the effectiveness of these techniques by TCAD simulations and establishes the following. At VBR of few 10's of
Volts, the high-k passivation is much less effective than the FP, e.g. for a device with VBR = 28 V, a 0.7 μm long
FP over 0.105 μm thick SiN doubles the VBR to 56 V, whereas even a 0.5 μm thick HfO2 passivation (εr = 20) has
almost no effect on VBR. However, such differences between the two techniques reduce progressively as VBR is
raised. The differences vanish for VBR ≥ 100 V, for which both techniques can improve the VBR by ≥ 2.5 times.
Although focused primarily on breakdown, our work gives quick calculations showing that the high-k passi-
vation has 5 – 10 % higher gate-drain capacitance than the gate-connected FP.

1. Introduction realization of a source connected FP requires an additional step of di-

electric deposition between the gate and the FP.
The GaN-based High Electron Mobility Transistors (HEMTs) are Another limitation of the prior work [2–4], [6–9] is that the VBR for
considered to be excellent candidates for radio-frequency (RF) power which the techniques' effectiveness was demonstrated was ≥120 V.
amplifier and high-speed power switching device applications due to However, at times, devices with high VBR under deep OFF-state have
better material properties such as high carrier concentration, high shown a low VBR for near-threshold OFF-state conditions, e.g. Ref. [11]
electron mobility in the channel and large critical electric field of GaN, reported devices with a threshold voltage of VT = −3.6 V, that showed
ECrit ≈ 3 MV/cm [1]. OFF-state breakdown voltage, VBR, of the device is a VBR ≥ 70 V for gate to source voltage, VGS = −7 V = VT − 3.4 V but
a key performance parameter from application point of view [1], e.g. low VBR = 28 V for VGS = −4.1 V = VT − 0.5 V (Ref. [12] has at-
the performance of the power switch is limited by VBR in the OFF-state tributed this VBR change to variation in surface and barrier traps due to
and on-resistance in the ON-state. It is well known that the VBR of GaN DC electric stress). It is necessary that a good device should have a high
HEMTs can be raised by the addition of a field plate (FP) near the drain VBR from near-threshold to deep OFF-state conditions swept during
edge of the gate [2–5]. This technique spreads the electric field over the turn-off, to reduce leakage and power loss. For this reason, the present
gate to drain region, thereby decreasing the peak electric field. Re- work compares the VBR enhancement techniques for a range of VBR that
cently, it has been reported that this field spreading and the resulting goes down to a few 10's of volts and for OFF-state near-threshold bias
VBR improvement can also be achieved just by depositing a high-k (high conditions. Prior work [2–9] has used a range of OFF-state VGS bias
dielectric constant, ɛr) passivation layer over the AlGaN layer [6–10]. conditions for studying VBR enhancement techniques without giving
One limitation of the prior work [2–10] is that it has discussed the any reason.
above two VBR enhancement techniques separately, i.e. it has not Our work is focused primarily on breakdown. However, we give
compared their effectiveness. The present work undertakes such a quick calculations of the parasitic gate to drain capacitance of the two
comparison since high-k passivation avoids the lithography step re- VBR enhancement techniques, since this capacitance degrades the fre-
quired to realize a FP. Further, a FP can be connected to either the gate quency response. It has been pointed out [8,9] that a FP raises the
[2–4] or the source [5]. We consider the gate connected FP since parasitic gate to drain capacitance, degrading the frequency response.

⁎
Corresponding author.
E-mail addresses: b.prasanna92@gmail.com (B. Prasannanjaneyulu), karmal@ee.iitm.ac.in (S. Karmalkar).

https://doi.org/10.1016/j.microrel.2020.113698
Received 16 February 2020; Received in revised form 8 April 2020; Accepted 13 May 2020
Available online 25 May 2020
0026-2714/ © 2020 Elsevier Ltd. All rights reserved.
B. Prasannanjaneyulu and S. Karmalkar Microelectronics Reliability 110 (2020) 113698

Fig. 2. (a) Calibration of near-threshold portion of ON-state ID-VGS simulation

with measured data at small VDS. (b) Calibration of the rapidly rising portion of
Fig. 1. Schematic of the device structure and biasing arrangement employed in
OFF-state ID-VDS simulations with measured data displaying 28 V breakdown
our study. The high-k passivated device has only the insulator over the AlGaN
due to SCLC. Also shown is the simulated ID-VDS curve displaying 100 V
layer and no FP; device width = 150 μm. The gate is placed centrally, resulting
breakdown due to avalanche; this is obtained by increasing the NAT of the ca-
in ungated lengths of 3.65 μm on either side of the gate.
librated device as described in the text.

However, no calculations of this parasitic capacitance have been re-

model, FLDMOB, and Selberherr's impact ionization model, SELB.
ported for high-k passivation. With the help of TCAD simulation of
Model parameters will be explained shortly. The VBR was defined as the
practical device structures biased in OFF-state near-threshold, we show
value of VDS @ ID = 0.15 mA = 1 mA/mm, where ID denotes the drain
the following: (i) At VBR of a few 10's of Volts, the high-k passivation is
current and VDS the drain to source voltage. The OFF-state drain to
much less effective than the FP. However, this difference reduces pro-
source current causing breakdown is sensitive to ns0, deep acceptor trap
gressively as VBR is raised, and vanishes for VBR ≥ 100 V. (ii) For a
density, NAT, and trap energy location, EAT. The ns0 depends on the
given VBR, the gate to drain capacitance of the high-k passivation is
polarization charges at the opposite faces of the AlGaN layer and the
higher than that of even the gate connected FP whose capacitance is
AlGaN surface charges. The separation between these charges, i.e. the
known to be higher than that of a source-connected FP [5]. These as-
barrier thickness, is several times smaller than the thickness of the gate
pects are discussed in Section 5, after descriptions of the device struc-
metal and the insulator. We have shown in Ref. [3] that, under such a
ture in Section 2, TCAD simulator calibration in Section 3, and the
condition, these charges can be coalesced into an effective fixed charge
qualitative physics underlying the two techniques in Section 4.
at the AlGaN/GaN interface equal to the measured ns0, to model the
VBR.
2. Device structure
We have not considered SRH statistics and energy levels of surface
states, since these are required for switching and transient character-
The device used to demonstrate our ideas have the structure shown
istics but not for DC breakdown simulations considered in our work.
in Fig. 1. Such a device with SiN (ɛr ~ 7.5) passivation thickness of
Two recent works [6,8] too have followed this approach. Further, we
t = 0.105 μm but without FP was fabricated at Solid State Physics
have not included any gate leakage models in the simulation, because at
Laboratory (SSPL), New Delhi. Epitaxial layers were grown on 6H-SiC
breakdown, i.e. at VDS @ ID = 1 mA/mm, which is the subject of the
using Metal Organic Chemical Vapor Deposition (MOCVD) system on 3"
present work, the measured gate current in our devices is very small
SiC substrate in the following sequence: AlN nucleation layer over SiC
compared to ID as shown in Fig. 3(a) of [11]. This implies that the
substrate, 2.5 μm thick unintentionally doped (UID) GaN layer, 1 nm
measured current flows from drain to source, or ID ≈ IS [11] in
AlN spacer layer, 24 nm thick Al0.23Ga0.77N barrier layer and 1 nm GaN
breakdown. This can be due to buffer leakage or impact ionization. In
cap layer. AlN is used as a spacer to suppress the alloy disorder scat-
latter case, the generated holes near the drain reach the source; though
tering and improve the 2-DEG mobility and its confinement. GaN cap
the gate is more negative than the source, the holes are stopped from
layer is used to protect the AlGaN surface from oxidation. Provides an
going to the gate by the potential barrier presented by AlGaN layer.
additional barrier at the schottky contact and reduces the leakage.
This has been observed in Ref. [1] as well.
MESA isolation regions of active layer were defined using photo-
lithography, dry etched with chlorine-ion bombardment and the photo-
resist layer was removed in oxygen plasma. Source/drain contacts were 3.1. Device with near-threshold VBR = 28 V without any enhancement
made of Ti/Al/Ni/Au. The metal layers were deposited by e-beam technique
evaporation, lithographed and annealed using Rapid Thermal Proces-
sing (RTP) to make them ohmic. A centrally placed Ni/Au Schottky type The device whose measured data is given in Fig. 2 is same as that
gate was deposited by e-beam evaporation and defined by lift-off. The studied in our prior work [11]. Hence, the models and parameters
SiN passivation layer was deposited by Plasma Enhanced Chemical employed to match the simulations to this data were decided following
Vapor Deposition (PECVD) technique. [11].
The 2-Dimensional Electron Gas (2-DEG) density, ns0, was measured We set ns0 = 9.67 × 1012 cm−2 obtained from Hall effect mea-
using Hall effect as 9.67 × 1012 cm−2. Fig. 2(a) shows the measured surements, the gate work function as 5.2 V, and the unintentional
ON-state ID-VGS data at small VDS around VT = −3.6 V, and Fig. 2(b) shallow donor density, ND, in the GaN buffer as 1 × 1015 cm−3. We
shows the measured OFF-state ID-VDS curve at VGS = −4.1 V, i.e. near- have assumed the same ND in the AlGaN layer as in the GaN buffer. We
threshold. also verified through TCAD simulation that results reported here are not
sensitive to ND upto ND = 1 × 1018 cm−3 in the AlGaN layer. Fig. 2
3. Simulator calibration shows that the VBR of the fabricated device (without FP) is soft and as
low as 28 V at VGS = −4.1 V. The analysis of [11] has established that
Our simulations were done using Silvaco ATLAS TCAD software such a VBR is the consequence of Space Charge Limited Current (SCLC)
[13]. Standard equations of current density, continuity and electro- from drain to source via a GaN buffer whose NAT is low and so the ND is
statics employed in device analysis, with field dependent mobility not compensated fully. Further, Ref [11] showed that device has

2
B. Prasannanjaneyulu and S. Karmalkar Microelectronics Reliability 110 (2020) 113698

NAT = 9.9 × 1015 cm−3 and located at 0.475 eV below the conduction
energy band, i.e. EC-EAT = 0.475 eV.
We used the FLDMOB mobility model since it captures the velocity
saturation effect at high fields encountered at large VDS in the OFF-
state. The low field electron mobility was set at μn = 700 cm2 V−1 s−1
by matching the simulated ON-state ID-VGS data at low VDS to the cor-
responding measured data around VGS = VT = −3.6 V, as shown in
Fig. 2(a). Matching the complete ON-state ID-VGS requires the use of VGS
dependent mobility as reported in Ref [14]. However, this is not ne-
cessary since our work concerns OFF-state behavior only. The OFF-state
ID-VDS curve at VGS = −4.1 V simulated using the above NAT and μn
matches the measured VBR as shown in Fig. 2(b). The mismatch be-
tween simulated and measured ID-VDS below breakdown is due to the
neglect of IG; measurements show that ID ≈ IG upto VDS = 15 V (see
Fig. 3(a) of [11]). The neglect of IG in simulation is justified since the
present work concerns breakdown, i.e. the ID-VDS segment near
ID = 1 mA/mm; in this segment ID ≫ IG as pointed out already, and so,
simulations and measurements match.

3.2. Devices with near-threshold VBR > 28 V without any enhancement

technique

To see the effectiveness of the FP and high-k techniques at higher

VBR using simulation, devices with VBR = 50, 75, 100 and 163 V with
SiN passivation but without FP were obtained by raising the NAT of the
above device (see Fig. 3(a) for values of NAT) to reduce the free carrier
density in the GaN buffer as proposed in [11]. This cuts down the SCLC
and thus raises the VBR. For high NAT, the VBR gets limited by impact
ionization induced avalanche rather that SCLC. Following [11], we
used the SELB impact ionization model [15] with parameters of AN
= AP = 2.4 × 108 cm−1, BN = BP = 1.9 × 107 V cm−1. Fig. 2(b)
shows that a simulated VBR = 100 V at VGS = −4.1 V is obtained for
NAT = 3.5 × 1016 cm−3. This is limited by avalanche, since simulation
without the impact ionization model yields a higher VBR = 166 V.
However, the VBR = 50, 75 V are due to SCLC since simulation without
impact ionization did not change the simulated VBR. Fig. 4. Simulated 2-D potential lines in a device section within ± 0.1 μm from
Fig. 3(a) shows that the VBR increases to 163 V at the AlGaN surface and 4 to 5.5 μm from the source. Bias conditions are:
NAT = 1 × 1017 cm−3. This can be explained with the help of Fig. 3(b), VGS = −4.1 V, VDS = 28 V (left column) and VGS = −4.1 V, VDS = 100 V
(right column). (a) 0.105 μm thick nitride passivation; (b) 0.5 μm thick high-k
which shows the distribution of the field directed from drain to source
passivation; (c) FP with l = 0.5 μm and t = 0.056, 0.12 μm nitride for
along 2-DEG. Here, for NAT ≤ 2.5 × 1016 cm−3, the peak field is seen
VDS = 28, 100 V.
to be less than the ECrit ≈ 3 MV/cm. For NAT > 2.5 × 1016 cm−3,
increased NAT reduces the electron concentration in the GaN buffer,
which in turn decreases the slope of the field distribution, i.e. spreads 4. Qualitative physics underlying VBR enhancement
the field, as per Gauss law. Hence, ECrit is reached at a higher voltage,
implying an increase in VBR. Note that, increase NAT beyond Refer to Fig. 4 which shows the potential lines in the VBR = 100 V
1 × 1017 cm−3 can cause reliability issues. device described above, at two different VDS = 28, 100 V which are VBR
values of the two devices considered, for high-k passivation and FP.
Focus on spread of the potential lines over the 2-DEG along the AlGaN/
GaN interface; more the spread, less the field. Since the
NAT = 3.5 × 1016 cm−3 is same in all the devices shown, the differ-
ences in field/potential spreading can be clearly attributed to variation
in VDS. High-k passivation thickness of t = 0.5 μm and the FP geometry
have been chosen to yield maximum reduction in the peak field. As
shown in [3], the t of the FP yielding the maximum reduction in peak
field increases with VDS; however a constant minimum t of high-k
passivation works for varying VDS. This is why in Fig. 4(c), t of the FP
for VDS = 100 V is higher than that for VDS = 28 V, while in Fig. 4(b), t
of the high-k passivation is same for both voltages. The potential and
field distributions of FP and high-k passivation methods have been
discussed individually in several prior works [3,7–9]. However, ours is
the first work which compares these two.
Fig. 3. (a) Simulated breakdown voltage at VGS = −4.1 V in a device without Fig. 4(b), (c) shows that at low VDS, high-k passivation cannot
field-plate or high-k passivation for increasing NAT in the GaN buffer layer. (b) spread the potential lines to the same extent as the FP; however, this
Distribution of the breakdown field directed from drain to source along the 2- difference reduces at high VDS. This behavior can be understood in
DEG for various NAT. Breakdown occurs at ID = 1 mA/mm. terms of the field lines which are perpendicular to the potential lines. In

3
B. Prasannanjaneyulu and S. Karmalkar Microelectronics Reliability 110 (2020) 113698

the case of the FP structure, the fringing field emanating from the gate
edge as well as the field plate helps spread potential lines crossing the 2-
DEG; moreover, the length and location of the field plate as well as the
εr of the insulator beneath can all be optimized to get the maximum
spread for a specific VDS, whether high or low. In contrast, in the high-k
passivation case, it is only the fringing field from the gate edge that
spreads the potential lines. At low VDS, this field is small and extends
little from the gate edge. Hence, the spread of the potential lines is less,
and the parameters t and εr of the passivation do not have much impact.
At high VDS however, the fringing field is large and extends farther.
Therefore, potential lines spread a lot, and parameters t and εr have a
strong impact on the fringing field and hence the spread. As mentioned
in prior works, e.g. [3,8] the spread of the potential lines saturates for
increase in l, t or εr beyond some critical values of these parameters.
The above discussion concludes the following. Both FP and high-k
passivation improve VBR by spreading the field. However, at low VBR,
high-k passivation cannot spread the field to the same extent as FP. This
difference reduces progressively as VBR is raised. Thus, it is the mag-
nitude and not the mechanism of VBR (SCLC or avalanche) that influ-
ences the effectiveness of the VBR enhancement techniques.
Consequences of the above physics on the effectiveness of high-k pas-
sivation and FP techniques are discussed below.

5. Breakdown voltage and capacitance simulations

Fig. 6. Simulated breakdown voltage data at VGS = −4.1 V and
Below, the VBR is taken as the value of VDS @ NAT = 3.5 × 1016 cm−3; (a) as a function of field plate length for different
ID = 0.15 mA = 1 mA/mm, as mentioned in Section 3. passivation layer thicknesses, and breakdown field distribution for 0.5 μm thick
SiN passivation, (b) as a function of passivation thickness for different dielectric
constants, and breakdown field distribution along 2-DEG for various HfO2
5.1. Device with near-threshold VBR = 28 V without any enhancing
passivation thicknesses.
technique

Fig. 5(a) shows that, for this device, the FP does not improve the VBR brought closer, for t = 0.105 μm, the VBR is seen to improve from 28 V
much for t = 0.15–0.2 μm, because the FP is not effective in spreading at l = 0 to 56 V at l = 0.7 μm. This is due to the increased spreading of
the field when it is far away from the 2-DEG. However, when the FP is the field with increase in l, when t = 0.105 μm. However, Fig. 5(b)
shows that even a 0.5 μm thick HfO2 passivation has almost no effect on
VBR because the field does not spread in spite of increase in t. This is
consistent with the conclusion of the previous section that mere pas-
sivation layer, even if high-k, cannot spread the field at low VDS and
improve a low VBR.

5.2. Devices with near-threshold VBR = 100, 163 V without any enhancing
technique

For the device with VBR = 100 V without FP, Fig. 6(a) gives the VBR
as a function of l for different t, and Fig. 6(b) the VBR as a function of t
for both SiN and HfO2 passivation. As explained above, VBR increases
with l, t and εr. Fig. 6(a) shows that a 0.7 μm long FP on 0.5 μm thick
SiN increases the VBR to 248 V. Fig. 6(b) shows that the same high VBR
can also be achieved simply using a 0.5 μm thick HfO2 passivation.
For the device with VBR = 163 V without FP, Fig. 7(a) shows that a
0.7 μm long FP on 0.5 μm thick SiN increases the VBR to 437 V. Further
increase in FP length does not increase VBR, since the two triangular
lobes of the field distribution (e.g. see Fig. 5(a)) no more overlap and so,
the area under this distribution which represents VBR, saturates as re-
ported in Ref [3]. Fig. 7(b) shows that the same high VBR can be also
achieved simply using a 0.5 μm thick HfO2 passivation layer. It is seen
that, the field spreads with increase in l of the FP or t of the high-k
passivation as explained in Section 4 and reaches the drain contact for
FP l ≥ 0.3 μm or HfO2 t ≥ 0.3 μm.

Fig. 5. Simulated breakdown voltage data at VGS = −4.1 V and 5.3. Comparison of FP and high-k passivation techniques
NAT = 9.9 × 1015 cm−3; (a) as a function of field plate length for different
passivation layer thicknesses, and breakdown field distribution for 0.105 μm Fig. 8 summarizes the simulated enhancement in VBR due to the two
thick SiN passivation, (b) as a function of passivation thickness for different techniques for different values of VBR without the technique obtained
dielectric constants, and breakdown field distribution along 2-DEG for various by varying the NAT. The results due to FP correspond to a FP geometry
HfO2 passivation thicknesses. which is optimized to yield maximum VBR enhancement. It is seen that,

4
B. Prasannanjaneyulu and S. Karmalkar Microelectronics Reliability 110 (2020) 113698

Table 1
Field plate and high-k passivation parameters yielding the same VBR. The 200 V
device has NAT = 3.5 × 1016 cm−3, while the 350 V device has NAT
= 1 × 1017 cm−3
VBR (V) High-k passivation Field plate

200 t = 0.32 μm, ɛr = 20 l = 0.53 μm, ɛr = 7.5, t = 0.3 μm

or
l = 0.45 μm, ɛr = 7.5, t = 0.5 μm
350 t = 0.28 μm, ɛr = 20 l = 0.36 μm, ɛr = 7.5, t = 0.3 μm
or
l = 0.25 μm, ɛr = 7.5, t = 0.5 μm

Above, t is passivation layer thickness, ɛr is relative permittivity of passivation

layer and l is field plate length.

VBR in the device with the FP but not so in the device with high-k
passivation. On the other hand for a device structure with VBR = 150 V
without any enhancement technique, the FP and high-k passivation can
spread the electric field equally (see Fig. 6). In this case, if LDG = 2 μm
by way of example, the same VBR may be obtained and the breakdown
field may touch the drain, in both techniques. Now, increasing LDG will
increase VBR in both cases, but to the same extent; the VBR will saturate
for some large LDG.

5.4. Capacitance simulations for devices with same VBR

Fig. 7. Simulated breakdown voltage data at VGS = −4.1 V and
NAT = 1 × 1017 cm−3; (a) as a function of field plate length for different
A proper assessment of the effectiveness of the FP and high-k pas-
passivation layer thicknesses, and breakdown field distribution for 0.5 μm thick
SiN passivation, (b) as a function of passivation thickness for different dielectric sivation requires a comparison of their parasitic gate to drain capaci-
constants, and breakdown field distribution along 2-DEG for various HfO2 tances, Cgd, which affect the device frequency response. This should be
passivation thicknesses. done for VBR ≥ 100 V in which range both techniques are equally ef-
fective in enhancing the VBR. From this point of view, we estimated the
parameters (t, ɛr) of the high-k passivation and (t, ɛr, l) of the FP re-
quired to get VBR = 200, 350 V. These are given in Table 1. In capa-
citance-voltage simulation, we set the small-signal frequency equal to
1 MHz. Fig. 9 gives the TCAD Cgd versus VDS simulations of devices with
these parameters at VGS = 0 (i.e. ON-state) and −4.1 V (i.e. OFF-state).
Similar range of capacitance values have been reported in the literature
[17]. It is seen that both ON-state and OFF-state capacitances are
5–10% lower for the devices with the FP, leading to conclude that FP is
more effective than high-k passivation.

6. Conclusion

We compared the high-k passivation and gate-connected field plate

Fig. 8. Simulated enhancement in VBR due to FP and High-k passivation tech- techniques of enhancing the OFF-state breakdown voltage of practical
niques. The x-axis is VBR without the technique. High-k passivation data cor- devices using TCAD simulations. We found that, independent of the
responds to HfO2 layer (εr = 20) of 0.5 μm thickness. FP data corresponds to a breakdown mechanism, at VBR of a few tens of Volts, the high-k pas-
FP of length l = 0.7 μm and an optimized SiN insulator thickness of t = 0.105,
sivation is much less effective than the FP, but this difference reduces
0.105, 0.25, 0.5 μm for VBR = 28, 50, 75, 100 V respectively. The value of NAT
progressively as VBR is raised and vanishes for VBR ≥ 100 V. Moreover,
for VBR on x-axis is given in Fig. 3(a).

difference in the effectiveness of the two techniques reduces progres-

sively as VBR is increased and vanishes for VBR > 100 V, where the two
techniques increase the VBR by > 2.5 times.
It is of interest to note the impact of varying the drain to gate
spacing LDG. Our prior work [16] has shown the following for a device
with a FP. If the field distribution along the 2-DEG extends up to the
drain electrode for some value of LGD, then VBR falls with reduction in
LGD below this value, rises for increase in LDG above this value, and
saturates beyond some LDG. This applies to high-k passivation as well,
since both FP and high-k passivation enhance the VBR by spreading the
electric field along the 2-DEG. Consider a device structure whose VBR
without any enhancement technique is as low as 30 V. Here, the FP can Fig. 9. Simulated bias dependence of the gate to drain capacitance in ON and
spread the field more than high-k passivation (see Fig. 5). If LDG = 1 μm OFF-states. Fig. 1 and Table 1 give the device parameters. (a) VBR = 200 V; (b)
by way of example, the breakdown field extends up to the drain with FP VBR = 350 V. Legend: high-k passivation (dashed line), field plate (dotted line).
but not with high-k passivation. Here, increasing LDG will increase the The dotted lines for the two sets of field plate parameters overlap.

5
B. Prasannanjaneyulu and S. Karmalkar Microelectronics Reliability 110 (2020) 113698

the high-k passivation has 5–10% higher gate to drain capacitance than [7] K. Nakamura, H. Hanawa, K. Horio, Analysis of breakdown voltages in AlGaN/
the gate-connected field plate, for a given VBR. GaNHEMTs with low-k/high-k double passivation layers, IEEE Trans. Device Mater.
Rel 19 (2) (Jun. 2019) 298–303, https://doi.org/10.1109/TDMR.2019.2903213.
[8] H. Hanawa, H. Onodera, A. Nakajima, K. Horio, Numerical analysis of breakdown
Declaration of competing interest voltage enhancement in AlGaN/GaN HEMTs with a high-k passivation layer, IEEE
Trans. Electron Devices 61 (3) (Mar. 2014) 769–774, https://doi.org/10.1109/TED.
2014.2298194.
The authors declare that they have no known competing financial [9] H. Hanawa, K. Horio, Increase in breakdown voltage of AlGaN/GaN HEMTs with a
interests or personal relationships that could have appeared to influ- high-k dielectric layer, Phys. Status Solidi A 211 (4) (Jan. 2014) 784–787.
[10] Q. Luo, Q. Yu, Electric field modulation by introducing a Hk dielectric film of tens of
ence the work reported in this paper. nanometers in AlGaN/GaN HEMT, Nanosci. Nanotechnol. Lett. 4 (9) (Sept. 2012)
936–939, https://doi.org/10.1166/nnl.2012.1406.
References [11] B. Prasannanjaneyulu, S. Bhattacharya, S. Karmalkar, Mechanism and enhancement
of the near-threshold low off-state breakdown voltage in gallium nitride high
electron mobility transistors, Jap. J. Appl. Phys. 58 (Apr. 2019), https://doi.org/10.
[1] M. Wang, K.J. Chen, Off-state breakdown characterization in AlGaN/GaN HEMT 7567/1347-4065/ab07a (pp. SCCD01-1-6).
using drain current injection technique, IEEE Trans. Electron Devices 57 (7) (July [12] B. Prasannanjaneyulu, S. Mishra, S. Karmalkar, Analysis of the significant rise in
2010) 1492–1496, https://doi.org/10.1109/TED.2010.2048960. breakdown voltage of GaN HEMTs from near-threshold to deep off-state gate bias
[2] E. Bahat-Treidel, O. Hilt, F. Brunner, V. Sidorov, J. Wurfl, G. Trankle, AlGaN/GaN/ conditions, IEEE Trans. Device Mater. Rel 19 (4) (Dec. 2019) 766–773, https://doi.
AlGaN DH-HEMTs breakdown voltage enhancement using multiple grating field org/10.1109/TDMR.2019.2950604.
plates (MGFPs), IEEE Trans. Electron Devices 57 (6) (Jun. 2010) 1208–1216, [13] ATLAS User’s Manual, Device Simulation Software, SILVACO Int, Santa Clara, CA,
https://doi.org/10.1109/TED.2010.2045705. 2015.
[3] S. Karmalkar, U.K. Mishra, Enhancement of breakdown voltage in AlGaN/GaN high [14] D. Pradeep, D.S. Rawal, S. Karmalkar, Comparison of two DC extraction methods
electron mobility transistors using a field plate, IEEE Trans. Electron Devices 48 (8) for mobility and parasitic resistances in a HEMT, IEEE Trans. Electron Dev 64 (4)
(Aug. 2001) 1515–1521, https://doi.org/10.1109/16.936500. (Apr. 2017) 1528–1534, https://doi.org/10.1109/TED.2017.2663764.
[4] H. Onodera, K. Horio, Analysis of buffer-impurity and field-plate effects on break- [15] S. Selberherr, Analysis and Simulation of Semiconductor Devices, Springer-Verlag,
down characteristics in small sized AlGaN/GaN high electron mobility transistors, Wien, New York, 1984.
Semicond. Sci. Technol. 27 (8) (Aug. 2012), https://doi.org/10.1088/0268-1242/ [16] S. Karmalkar, N. Soudabi, A closed - form model of the channel electric field under
27/8/08501 (pp. 085016-1-6). the field plate in a HEMT, IEEE Trans. Electron Devices 53 (10) (Oct. 2006)
[5] G. Xie, et al., Breakdown-voltage-enhancement technique for RF-based AlGaN/GaN 2430–2437, https://doi.org/10.1109/TED.2006.882273.
HEMTs with a source-connected air-bridge field plate, IEEE Electron Device Lett 33 [17] A. Zhang, L. Zhang, Z. Tang, X. Cheng, Y. Wang, K.J. Chen, M. Chan, Analytical
(5) (2012) 670–672 May. modeling of capacitances for GaN HEMTs, including parasitic components, IEEE
[6] T. Kabemura, S. Ueda, Y. Kawada, K. Horio, Enhancement of breakdown voltage in Trans. Electron Devices 61 (3) (Mar. 2014) 755–761, https://doi.org/10.1109/TED.
AlGaN/GaN HEMTs: field plate plus high-k passivation layer and high acceptor 2014.2298255.
density in buffer layer, IEEE Trans. Electron Dev. 65 (9) (Sep. 2018) 3848–3854,
https://doi.org/10.1109/TED.2018.2857774.