Sunteți pe pagina 1din 7

MR-12562; No of Pages 7

Microelectronics Reliability xxx (2017) xxx–xxx

Contents lists available at ScienceDirect

Microelectronics Reliability

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

TCAD simulation capabilities towards gate leakage current analysis of advanced AlGaN/
GaN HEMT devices
K. Mukherjee ⁎, F. Darracq, A. Curutchet, N. Malbert, N. Labat
IMS Laboratory, University of Bordeaux, CNRS UMR 5218, F-33405 Talence, France

a r t i c l e i n f o a b s t r a c t

Article history: 2D TCAD Sentaurus simulations based on Drift-Diffusion transport are performed to identify the modeling pa-
Received 24 May 2017 rameters that crucially affect the reliability characteristics of AlGaN/GaN HEMT devices, demonstrated by their
Received in revised form 5 July 2017 effects on the gate leakage characteristic. The behavioural nature and impact of each parameter on the leakage
Accepted 9 July 2017
performance is discussed. Schottky gate tunneling and trapping effects within the structure are two major reli-
Available online xxxx
ability issues that modulate the leakage characteristic. Hence, their contributions are precisely modeled. A simu-
Keywords:
lation methodology is presented to recognize the relative control of individual parameters on distinct regions of
GaN HEMT the leakage characteristic. This modeling approach is demonstrated for a GaN HEMT technology and can be fur-
TCAD ther applied to facilitate reliability comparisons across different device technologies. This validates TCAD simula-
Simulation tion to be an effective aiding tool in reviewing and interpreting GaN HEMT reliability performance and design
Gate current choices.
AlGaN/GaN;gate leakage © 2017 Elsevier Ltd. All rights reserved.
Reliability
Tunneling
Traps

1. Introduction behind choosing certain parameter values for modeling a particular de-
vice or technology are often unclear.
Technology Computer Aided Design (TCAD) simulation has emerged To address these issues, this work targets simulation of device pa-
as an advantageous tool to investigate various reliability concerns of rameters that majorly control the IG-VG characteristic. The process to
AlGaN/GaN HEMTs such as current collapse [1], gate lag and drain lag recognize the physical limits and establish the correct operational
[2]. However, simulating the unique III-N HEMT with its polarization ranges for these crucial parameters, and thus, to build a consistent
charges is a challenge and requires further work to isolate inherent cor- model for a GaN based HEMT is discussed. Two critical reliability issues
relations within factors controlling and modulating device behaviour. affecting leakage performance are carefully modeled: Gate tunneling
The gate leakage current describes the robustness of the pivotal described by the Non-Local tunneling model; trapping effects through
gate-drain region and is linked to GaN HEMT reliability. Reverse introduction of acceptor traps within the simulated structure. A meth-
gate leakage is often substantially higher than predicted by therm- odology is proposed to recognize the relative control of parameters on
ionic emission which deteriorates high voltage operation and stand- distinct regions of the IG-VG characteristic, aiding to identify primary pa-
by power dissipation [3,4]. Significant research has been performed rameters that are relevant to performance optimization across different
to interpret gate leakage mechanisms [5–10]. Many works have con- operational conditions or technologies.
firmed the contribution of tunneling through the Schottky barrier to
the increased leakage [6–8]. While Poole-Frenkel emission could as- 2. TCAD simulation approach
sist conduction at high temperatures [8,9], surface trap related hopping
conduction [10] and trap assisted tunneling mechanisms [6,9,10] have A physical AlGaN/GaN HEMT model is built to virtually represent in-
also been reported. The gate leakage current still requires improved ternal transistor operation using the Sentaurus tool from Synopsys [11].
modeling, probably because it is susceptible to various degradations The model is based on the Drift-Diffusion (D-D) transport model and
that could be undiscernible in other outputs. Also, the motivations Fermi statistics. A doping dependence model and a high velocity satura-
tion model, driven by a field computed as the gradient of the electron
quasi-Fermi level, are defined for electron mobility. The SRH model for
⁎ Corresponding author at: 351 Cours de la Libération, Laboratoire IMS, Bâtiment A31,
generation-recombination is activated. The source and the drain regions
Talence 33405, France. (see Fig. 1) are defined in Sentaurus as “modified ohmic” contacts [11].
E-mail address: kalparupa.mukherjee@ims-bordeaux.fr (K. Mukherjee). This definition while reproducing the electrical behaviour of an ohmic

http://dx.doi.org/10.1016/j.microrel.2017.07.049
0026-2714/© 2017 Elsevier Ltd. All rights reserved.

Please cite this article as: K. Mukherjee, et al., TCAD simulation capabilities towards gate leakage current analysis of advanced AlGaN/GaN HEMT
devices, Microelectronics Reliability (2017), http://dx.doi.org/10.1016/j.microrel.2017.07.049
2 K. Mukherjee et al. / Microelectronics Reliability xxx (2017) xxx–xxx

Fig. 1. Simulated 2D GaN HEMT structure details.

contact does not impose the charge neutrality condition at contact ver-
tices, thus helping to prevent inaccurate simulations around charged
Fig. 2. Forward IG-VG characteristics for varying RG.
depletion regions.
Defining reference values for major parameters is crucial in repro-
ducing a characteristic GaN HEMT. Each parameter is studied individu-
ally to identify its physical range through its impact on global IG-VG. To the IForw in the high current regime (VG N 1 V). However, as can be ex-
enable unambiguous inferences, parameter variations are performed pected, the activation of RG has no influence on reverse gate character-
under ideal or degradation-free initial conditions. The activated therm- istics (IRev). No reference RG is defined to better isolate the effects of
ionic emission (TE) model dominantly controls the gate current re- parameters described in the following sections.
sponse in this ideal model.

2.1. 2D TCAD structure details 2.3. Effect of surface donors

The simulated 2D TCAD structure in Fig. 1 is built to closely resemble The GaN HEMT is singular because of the piezoelectric and sponta-
the advanced GaN HEMT GH-25 process from United Monolithic Semi- neous polarization charges at device interfaces, which require detailed
conductors (UMS). Except the thin 2 nm GaN cap, which is highly n- modeling. The “Piezoelectric Polarization” model in Sentaurus [11] in-
doped, the rest of the structure has minimal doping. A non-uniform ternally calculates these interface charges, and is activated in our simu-
mesh [11] of nodes is created to discretize the physical properties and lations as a function of strain. Surface-Donors (S-D) at the top SiN/GaN
numerically simulate the HEMT behaviour. The mesh is markedly finer cap interface support and control the 2-dimensional electron gas
at gate edges, especially for the Gate-Drain, and all semiconductor inter- (2DEG) concentration formed in the quantum well at the AlGaN/GaN
faces, to ensure both robust and efficient simulation. hetero-interface. Hence, they are both inevitable and indispensable in
the GaN HEMT system. Both S-D density (NS-D in cm− 2) and energy
2.2. Simulation of gate Schottky contact (ES-D in eV) need careful optimization for a given structure to sustain
the complete activation of 2DEG, and thereby achieve maximal device
An essential step to develop a complete HEMT model is physical sim- performance. Works on simulation of GaN HEMTs [13,14] generally
ulation of the gate Schottky contact. The Schottky barrier ΦB at the sur- choose S-D definitions based on fits to experiments, or previous litera-
face is theoretically defined as the difference of the gate metal work ture. However, reasons for which specific NS-D or ES-D values might be
function and the electron affinity of the semiconductor. Standard physically more suitable than others remain abstruse. IG-VG characteris-
thermionic emission theory can however be applied to extract the effec- tics offer a coherent approach towards interpreting the physical origin
tive Schottky barrier height ΦB,eff directly from the device forward gate of changes due to variations in NS-D or ES-D.
characteristics IForw using:
 
ΦB;eff
I0 ¼ AA T2 exp ; ð1Þ
ηkB T

where I0 is the saturation current obtained as the zero-bias intercept


through extrapolation of the linear region of IForw. A is the contact
area, A* is the effective Richardson's constant, T is the absolute temper-
ature and η is the ideality factor.
It is a preferable simulation technique to choose Φms as a parameter
instead of directly specifying ΦB for Schottky contacts on semiconduc-
tors such that the ΦB,eff is calculated internally. An ΦB,eff ~ 0.9 eV has
been previously extracted from UMS GH-25 measurements in [12],
and also reported in other works [2,13–15]. For the mole fraction
range of x = 0.235–0.25, a χAlGaN of 3.425–3.45 eV and a χGaN =
3.9 eV is considered in Sentaurus material parameter files [11]. A gate
contact with Φms = 4.4 eV well approximates the measured IForw cur-
rent levels, resulting in an effective gate to channel barrier of ~ 1 eV.
Hence, it's chosen as the reference value.
A distributed gate resistance (RG) is the other major parameter de- Fig. 3. IRev characteristics for varying NS-D; inset: Trapped NS-D as a function of reverse VG
fining a Schottky gate. As observed in Fig. 2, an increase in RG reduces bias.

Please cite this article as: K. Mukherjee, et al., TCAD simulation capabilities towards gate leakage current analysis of advanced AlGaN/GaN HEMT
devices, Microelectronics Reliability (2017), http://dx.doi.org/10.1016/j.microrel.2017.07.049
K. Mukherjee et al. / Microelectronics Reliability xxx (2017) xxx–xxx 3

2.3.1. Surface donors: concentration (NS-D) The corresponding ESurf is stronger and creates depletion regions at
The primary factor governing NS-D validity at a given ES-D is the sur- the GD and GS edges (Fig. 4(c)), restricting electrons to only under
face Electric Field ESurf. IRev characteristics for varying NS-D at ES-D = the gate (Fig. 4(d)), and effectively suppressing surface leakage.
0.4 eV from MidBand (1.3 eV from CB) are presented in Fig. 3. Here, the IRev slope corresponds to electrons that steadily compensate
A rapid growth in IRev is observed for NS-D b 3 × 1013 cm− 2. These or de-trap the initial trapped NS-D due to the stronger ESurf (Fig. 4(c))
NS-D values also cause worsening of ID levels. Relative occupation of along LGS and LGD accompanying a higher │VG│ (Fig. 3 inset), until
the defined NS-D can be perceived from the inset of Fig. 3 through |VG | = Vth when NS-D reaches its minimum and the 2DEG is depleted.
the evolution of the trapped (occupied) NS-D with VG bias. For NS-D = A reference value of NS-D = 5 × 1013 cm−2 is thus chosen for further
2 × 1013 cm−2, the trapped NS-D remains fixed at the defined value simulations.
and shows no evolution with higher | VG |. This can be explained by
the evolution of absolute ESurf along the device length as described in 2.3.2. Surface donors: energy position (ES-D)
Fig. 4(a). An insufficient ESurf at the Gate-Source (GS) and Gate-Drain Energy positions above the Fermi level and closer to CB have a higher
(GD) edges allows electrons to leak and spread into the GS and GD pas- probability (≥50%) of being occupied by a hole. Hence, at a defined NS-D,
sivation regions (LGS, LGD) (Fig. 4(b)). This promotes surface leakage, S-D charges have a greater chance of being occupied for ES-D closer to
which further aggravated by higher |VG |, induces a higher IRev. Hence, the CB. B. Jogai in [16] indicated an ESD = 1.5 eV, pinning the surface
NS-D has a minimum threshold, in this case ≈3 × 1013 cm−2. Fermi level EF close to the mid-band. However, Gladysiewicz et al. [17]
An NS-D of 3 × 1013 cm− 2 was also chosen by Zhou et al. [13] for reported that the EF is pinned 0.55 eV below the CB for GaN/AlGaN/
modeling a GaN capped HEMT. For a defined NS-D ≥ 4 × 1013 cm−2, GaN structures, based on theoretical and experimental findings. Other
IRev growth is more gradual, with higher NS-D values slightly raising works [8,13,18] also support and validate an ES-D ≈ 0.5 eV in their stud-
IRev. They finally saturate beyond the threshold voltage Vth ~ − 4 V. ies. Fig. 5(a) and (b) displays the IRev and IForw characteristics for an ES-D
range of 0.6 eV–0.9 eV from the CB, compared to an ES-D = 1.3 eV.
An initial rise followed by a gentle saturation in both IRev and IForw
is observed at low | VG |, for ES-D values lower than a threshold ES-D ≈
0.9 eV. The effect gets stronger for ES-D closer to CB. For ES-D b 0.8 eV,
overall IRev is higher at | VG | N Vth. Fig. 6(a) and (b) presents the dis-
tribution of absolute values of ESurf and current density JSurf along the
device length, just under the GaN cap for VG = −1 V. For ES-D = 1.3 eV
from the CB, a high ESurf at the GS, GD edges and a low ESurf in LGS and
LGD produce the general IRev curve as previously discussed. In contrast,
for ES-D = 0.6 eV, the ESurf at the GS and GD edges are the lowest,
allowing electrons to flow into passivation regions. This, combined
with the higher ESurf along LGS and LGD, spikes JSurf and hence IRev. For
low | VG |, the S-D de-trapped by the incoming electrons can get re-
trapped owing to their proximity to CB, triggering an even higher IRev
in response.
Once |VG | is high enough, the ESurf at the gate edges blocks electrons
from leaking into the surrounding regions, so IRev softly saturates and
then merges into the general curve. For the reliability section, a refer-
ence ES-D = 1.3 eV has been chosen since it doesn't modify the leakage
characteristics considerably and hence helps to isolate other degrada-
tion effects from that of ES-D.

3. Reliability analysis

Gate leakage simulations are capable of clearly reflecting two major


degradation phenomena: the gate tunneling and various device trap-
ping effects. Hence, a deeper understanding of their mechanisms can
be achieved through precisely modeling their modulations of the IG-
VG characteristic.

3.1. Simulation of gate tunneling: effect on IG-VG

Tunneling current (ITun) is inherently dependent on the CB edge


profile along the entire tunneling path connecting any two points,
making tunneling a non-local process. Hence, the powerful “Non-
Local (N-L) Tunneling” [11] model in Sentaurus, which accurately
computes the complex band edge profile for each node from the
Poisson and D-D equations, is activated for modeling Schottky gate
tunneling. For this, a specialized N-L mesh is constructed which over-
laps over the original device mesh. Then, it connects all semiconductor
vertices (mesh nodes) within a chosen tunneling length (LTun) to the
gate electrode through tunneling. These connections are the focal points
of N-L lines which essentially comprise tunneling paths. A higher LTun
hence strengthens the tunneling in both X and Y directions. Choosing
Fig. 4. Abs(ESurf) and E-density for a X-Cut 1 nm under the GaN cap/AlGaN barrier interface an LTun for a technology depends on the Schottky robustness and any
for (a), (b) 2 × 1013 cm−2 and (c), (d) 5 × 1013 cm−2 (lines) respectively. corresponding surface degradations. In Fig. 7, a schematic band-

Please cite this article as: K. Mukherjee, et al., TCAD simulation capabilities towards gate leakage current analysis of advanced AlGaN/GaN HEMT
devices, Microelectronics Reliability (2017), http://dx.doi.org/10.1016/j.microrel.2017.07.049
4 K. Mukherjee et al. / Microelectronics Reliability xxx (2017) xxx–xxx

Fig. 5. (a) IRev and (b) IForw curves for varying ES-D.

diagram of the HEMT illustrates the N-L tunneling mechanism under additional tunneling area is too far from the gate to make a notable
varying LTun and bias. A higher LTun allows electrons to tunnel deeper impact.
into the AlGaN barrier while a higher | VG | shifts the CB higher which Fig. 8(b) demonstrates that a higher LTun increases IForw for lower
weakens the tunneling barrier and increases IRev. | VG | conditions. At higher | VG |, tunneling becomes less significant
The IRev characteristics for LTun ranging from 1 nm–10 nm are pre- [19], and even curves with high LTun start merging into the reference,
sented in Fig. 8(a). The reference curve (LTun = 0) is primarily governed corresponding to just thermionic emission.
by the activated TE model. For LTun = 1 nm or 2 nm, tunneling occurs
within the highly doped GaN cap, hence only a slight rise in IG is ob- 3.2. Simulation of device traps: effect on IG-VG
served compared to the reference. For LTun = 2.1 or 2.5 nm, the stronger
vertical electron E-field component causes a higher IRev. Vth is however Acceptor traps in various device locations majorly degrade the GaN
still preserved. Higher LTun curves however, begin displaying earlier sat- HEMT performance. Though different kind of traps impair the device re-
uration (lower Vth) due to faster S-D de-trapping while the stronger liability in distinct ways, their effect on output characteristics could
leakage flow now reaches deeper into the barrier layer and begins re- often be similar. Traps of different concentrations in different energy
ducing the GaN channel electron mobility. For LTun ≥ 5 nm, the tunneling and spatial locations could cause equivalent ID degradation or Vth shifts,
field is strong enough to raise IRev even for VG ≥ Vth. However, the rela- making it difficult to distinguish the presence of specific traps within the
tive IRev rise for higher LTun values reduces beyond LTun N 5 nm, since the device. IG-VG characteristics, however, are uniquely affected by each
trap type which aids in identifying individual trapping phenomena.
The presence of bulk traps in the AlGaN barrier [4,14,20,21] and in
the GaN buffer [1,2,13,14,20] has been substantiated by numerous
works. Research also corroborates the importance of traps close to the
AlGaN/GaN hetero-interface [4,21,22] and at the top SiN passivation/
cap surface [1,4,14]. Bulk trap concentrations considered usually range
from ≈1016 cm−3 to 1018 cm−3 while interface trap concentrations of
3 × 1012–9 × 1012 cm−2 have been studied in [14].
Traps at relevant structural locations have been simulated, with de-
tails as specified in Fig. 9. Bulk acceptor traps of level distribution in the
AlGaN barrier, the GaN buffer and the GaN channel, and interface traps

Fig. 6. (a) Abs (ESurf) and (b) Abs (JSurf) at VG = −1 V along the device length 1 nm below
the GaN cap/AlGaN barrier interface for varying ES-D. Fig. 7. Band-diagram illustrating N-L tunneling model.

Please cite this article as: K. Mukherjee, et al., TCAD simulation capabilities towards gate leakage current analysis of advanced AlGaN/GaN HEMT
devices, Microelectronics Reliability (2017), http://dx.doi.org/10.1016/j.microrel.2017.07.049
K. Mukherjee et al. / Microelectronics Reliability xxx (2017) xxx–xxx 5

Fig. 8. (a) IRev and (b) IForw for variation in LTun.


Fig. 10. (a) IRev and (b) IForw for variation in traps.

at the AlGaN/GaN hetero-interface and passivation/GaN cap interface


have been investigated. The chosen concentrations agree with literature probably since for low | VG | b Vth, the rise in IRev due to de-trapping of
while best reflecting the trapping influence on the leakage curve. For NS-D makes electron availability for trapping low. The impact on IForw
UMS devices, trap activation energies of 0.48 eV and 0.66 eV [23] have (Fig. 10(b)) is important only at high VG ≥ 1 V conditions. The defined
been identified by Drain Current Transient Spectroscopy (DCTS). Nu- barrier traps do not modify IRev characteristics but are found to signifi-
merous works have reported similar trap energy of 0.5 eV [1,2,4,13,21, cantly reduce the IForw slope for 0.8 V ≤ VG ≤ 1.7 V resulting in lower
22] or 0.6 eV [14,20] relative to the CB. Hence, all considered traps are IForw levels than the reference. Buffer and channel traps change the
located 0.5/0.6 eV from the CB. IRev slope near Vth, causing a slow IRev rise beyond Vth to finally reach
Each trap type has its own signature impact on the IG-VG charac- the reference current level at VG = − 6 V, however, their effects on
teristics when compared to the reference (no traps) as observed IForw are negligible. Traps at the AlGaN/GaN interface display a similar
from Fig. 10. Traps dominantly affect IRev (Fig. 10(a)) for | VG | ≥ Vth, slight rise in IRev slope after Vth, however they start deviating from the
reference earlier, causing a noticeable reduction in overall IRev current
levels and slope earlier for | VG | ≥ 2 V. The corresponding change in
IForw is minor in comparison, and is observable as a minor decrease
once VG crosses 1 V. Passivation interface traps increase IRev due to sur-
face leakage as previously discussed and decrease final IForw levels be-
yond VG ≥ 1 V. Donor traps defined in the same locations have
equivalent but contrary effects. While their activation supports conduc-
tion, the relative impact is dependent upon defined trap specifications
and the corresponding occupational dynamics.

4. Model optimization methodology

The above discussion recognizes the impact of critical simulation


parameters on the IG-VG characteristic. Based on careful assessment
of their limits and relative impact, a methodology is proposed to
achieve an optimal set of simulation parameters for a given technol-
ogy. Customizing the principal parameters, within their identified
thresholds, to govern the low and high current regimes of a particu-
Fig. 9. Band diagram illustrating simulated trap details. lar IG-VG curve can help extract a representative parameter set. IRev

Please cite this article as: K. Mukherjee, et al., TCAD simulation capabilities towards gate leakage current analysis of advanced AlGaN/GaN HEMT
devices, Microelectronics Reliability (2017), http://dx.doi.org/10.1016/j.microrel.2017.07.049
6 K. Mukherjee et al. / Microelectronics Reliability xxx (2017) xxx–xxx

and IForw characteristics of an UMS GH-25 technology are compared Table 1


to simulations in Fig. 11(a) and (b), respectively, which illustrate Relative control of parameters on IG-VG curve.

the relative control exercised by individual parameters over distinct Impact Φms LTun NS-D ES-D Traps RG
segments of the leakage characteristic. Region 1, IRev High Medium Low High Low Trivial
Table 1 summarizes the relative parameter effects on current (IG) Region 2, IRev Medium High Low Low High Trivial
levels in two different regions: Region 1 (IRev: VG = − 2 V to 0 V; Region 1, IForw High Low Trivial Medium Low Trivial
IForw: 0 V to 0.3 V) and Region 2 (IRev: VG = − 6 V to -3 V; IForw: 0.8 V Region2, IForw High Trivial Trivial Trivial Low Medium
to 1 V). The IForw characteristic is majorly controlled by TE [9] and
hence is less susceptible to modification by parameters other than
Φms. An ES-D closer to CB and a higher LTun slightly increases the IForw appreciable tunneling, but also donor-like traps. The simulated ID-VD
in Region 1, while RG can moderately reduce the level and slope of curves, however, were higher than measurements, thus reflecting the
IForw in Region 2. In contrast, IRev can be well controlled by the discussed impact of acceptor traps. Since the measured IRev exhibits a small
parameters. The initial value in Region 1 is defined by Φms and LTun. LTun, slope beyond Vth, buffer and AlGaN/GaN interface acceptor traps (as
however, can potentially alter the level and nature of IRev quite signifi- discussed for Fig. 10(a)) were considered. Also, since LTun is irrelevant
cantly in Region 2. While ES-D is relevant in obtaining a weak initial to ID-VD curves while traps significant control the ID current levels, the
slope in Region 1 (as for IForw), the effect of NS-D though small, extends ID-VD measurements helped in determining the relative impact of
over the entire VG range. Except in Region 1, traps can alter the IRev na- tunneling versus traps. Optimization of LTun and AlGaN/GaN interface
ture and final levels to a moderate extent. To demonstrate this method- trap parameters finally generated a dedicated parameter set for the
ology, the simulated IRev and IForw curves in Fig. 11(a) and (b) have been UMS technology, as specified in Table 2. Because of non-negligible dis-
optimized for the UMS GH-25 device. To verify that the chosen param- persion within GaN HEMT characteristics, the achieved parameter set,
eters do not cause deviation in device behaviour under active operation, though unique, is not exclusive. Further parameter tuning dedicated
corresponding ID-VD comparisons have been performed and are pre- to specific constraints for certain parameters could improve the tuning
sented in Fig. 12. of other parameters, and result in a slightly modified set.
Initially, Φms was chosen to approach IForw levels. The initial slope Final IG-VG simulations show good agreement excepting a minor
in IForw also suggested an ES-D value close to the CB and ESD = 0.53 eV mismatch in IForw for medium VG range. Using η = 1 and A* =
provides the best agreement, similar to findings in [17]. The simulated 23.2 A/cm 2 K 2 [24] in Eq. (1) for Fig. 11(b), a ΦB,eff of 0.88 eV and
IRev levels were however still significantly lower in comparison for 1.06 eV is extracted for the measured and simulated curves respectively.
both Region 1 and Region 2. A higher LTun cannot account for a pro- This slight disparity could be potentially improved by considering barri-
nounced increase in Region 1, thus indicating the presence of not just er height modulations [9,25] caused by transient trap effects, hopping
mechanisms or trap assisted tunneling. Understandably, ID-VD simula-
tions slightly deviate for high VG conditions, since self-heating induced
lowering of ID is not taken into account by the adopted Drift-Diffusion
model. However, the global accuracy is quite good, which validates
the chosen parameter set. Also, though hydrodynamic models allow
simulation of self-heating effects, they require sacrificing simulation
speed and simplicity. In contrast, since this methodology models IG-VG
behaviour at VDS = 0 V where self-heating effects are insignificant,
the D-D model is powerful enough to viably model reliability concerns
of GaN HEMTs.

5. Conclusion

Leakage current analysis presented in this work advances the discus-


sion on various degradation mechanisms plaguing the GaN HEMT. The
Drift–Diffusion model is found sufficiently capable of illustrating the be-
haviour and relative impact of all considered parameters including gate

Fig. 11. Model comparison of (a) IRev and (b) IForw with UMS GH-25 measurements. Fig. 12. Model comparison to ID-VD of GH-25 device.

Please cite this article as: K. Mukherjee, et al., TCAD simulation capabilities towards gate leakage current analysis of advanced AlGaN/GaN HEMT
devices, Microelectronics Reliability (2017), http://dx.doi.org/10.1016/j.microrel.2017.07.049
K. Mukherjee et al. / Microelectronics Reliability xxx (2017) xxx–xxx 7

Table 2
Simulated parameter values for UMS GH-25 technology.

SWF LTun NS-D ES-D Traps:donors ND, ED Traps:acceptors NA, EA

4.24 eV 4.5 nm 6 × 1013 cm−2 0.53 eV from CB AlGaN barrier: 5 × 1018 cm−3, 1.2 eV from CB AlGaN/GaN: 1012 cm−2, 0.1 eV from CB

tunneling and traps. Since the investigated IG-VG characteristic, in con- [9] S. Turukevere, et al., Gate leakage mechanisms in AlGaN/GaN and AlInN/GaN
HEMTs: comparison and modeling, IEEE Trans. Electron Devices 60 (10) (2013)
trast to ID-VD curves, is unique to each technology; precise physics- 3157–3165.
based optimization of the dominant parameters for each region of the [10] Arulkumaran, et al., Temperature dependence of gate-leakage current in AlGaN/GaN
IG-VG curve improves device simulation accuracy. Comparison of pa- high-electron-mobility-transistors, Appl. Phys. Lett. 82 (18) (2003) 3110–3112.
[11] “Sentaurus Device User Guide”, Synopsys, 2007.
rameter sets could provide insight into recognition of device vulnerabil- [12] Laurent Brunel, Contribution à l'assurance fiabilité de filières HEMTs à base de GaN
ities or design advantages relative to distinct reliability phenomena. sur substrat SiC : caractérisation électrique approfondie et modélisaton des effets
This can assist device engineers aiming to improve not only leakage parasites, 2015 (Doctoral Thesis).
[13] Zhou, et al., Transient simulation of AlGaN/GaN HEMT including trapping and ther-
but overall performance criteria of GaN HEMT technologies. mal effects, International Conference on Solid-State and Integrated Circuit Technol-
ogy 2014, pp. 1–3.
Acknowledgements [14] M. Faqir, et al., Investigation of high-electric-field degradation effects in AlGaN/GaN
HEMTs, IEEE Trans. Electron Devices 55 (7) (2008) 1592–1602.
[15] L. Efthymiou, et al., Modelling of an AlGaN/GaN Schottky diode and extraction of
The authors would like to thank UMS for their support by providing main parameters, International Semiconductor Conference 2015, pp. 211–214.
the GH-25 technology devices. [16] B. Jogai, et al., Influence of surface states on the two-dimensional electron gas in
AlGaN/GaN heterojunction field-effect transistors, J. Appl. Phys. 93 (3) (2003)
1631–1635.
References [17] M. Gladysiewicz, et al., The surface boundary conditions in GaN/AlGaN/GaN transis-
tor heterostructures, Appl. Phys. Lett. 98 (2011) 231902.
[1] M. Faqir, et al., Analysis of current collapse effect in AlGaN/GaN HEMT: experiments [18] A. Rizzi, et al., Surface and interface electronic properties of AlGaN(0001) epitaxial
and numerical simulations, Microelectron. Reliab. 50 (2010) 1520–1522. layers, Appl. Phys. A 87 (3) (2007) 505–509.
[2] C. Miccoli, et al., Trapping and thermal effects analysis for AlGaN/GaN HEMTs by [19] W. Mei, et al., Schottky forward current transport mechanisms in AlGaN/GaN
means of TCAD simulations, IEEE Electron Device Lett. 34 (9) (2008) 1121–1123. HEMTs over a wide temperature range, Chin. Phys. Soc. 23 (9) (2014) 097307.
[3] G. Meneghesso, et al., Reliability of GaN high-electron-mobility transistors: state of [20] D. Bisi, et al., Deep-level characterization in GaN HEMTs-part I: advantages and lim-
the art and perspectives, IEEE Trans. Device Mater. Reliab. 8 (2) (2008) 332–343. itations of drain current transient measurements, IEEE Trans. Electron Devices 60
[4] G. Meneghesso, et al., Reliability issues of gallium nitride high electron mobility (10) (2013) 3166–3175.
transistors, Int. J. Microw. Wirel. Technol. 2 (1) (2010) 39–50. [21] J. Yang, et al., Determination of trap energy levels in AlGaN/GaN HEMT, Device Re-
[5] Y. Yamaguchi, et al., Simulation study and reduction of reverse gate leakage current search Conference 2013, pp. 79–80.
for GaN HEMTs, IEEE Compound Semiconductor Integrated Circuit Symposium [22] Zhang, et al., Influence of the interface acceptor-like traps on the transient response
2012, pp. 1–4. of AlGaN/GaN HEMTs, IEEE Electron Device Lett. 34 (1) (2013) 45–47.
[6] R. Jos, Reverse Schottky gate current in AlGaN/GaN high-electron-mobility transis- [23] M. Rzin, et al., Investigation of gate and drain leakage currents of AlGaN/GaN HEMTs
tors, J. Appl. Phys. 112 (2012) 094508. at subthreshold regime for temperature range 300K–400K, European Microwave In-
[7] K. Hayashi, et al., Mechanism study of gate leakage current for AlGaN/GaN high elec- tegrated Circuits Conference 2013, pp. 236–239.
tron mobility transistor structure under high reverse bias by thin surface barrier [24] T. Sawada, et al., Electrical properties of metal/GaN and SiO2/GaN interface and ef-
model and technology computer aided design simulation, Jpn. J. Appl. Phys. 52 fects of thermal annealing, Appl. Surf. Sci. 159-160 (2000) 449–455.
(2013) 04CF12. [25] E. Arslan, et al., Tunneling current via dislocations in Schottky diodes on AlInN/AlN/
[8] Zhang, et al., Analysis of leakage current mechanisms in Schottky contacts to GaN GaN heterostructures, Semicond. Sci. Technol. 24 (7) (2009).
and Al0.25Ga0.75N∕GaN grown by molecular-beam epitaxy, J. Appl. Phys. 99
(2006) 023703.

Please cite this article as: K. Mukherjee, et al., TCAD simulation capabilities towards gate leakage current analysis of advanced AlGaN/GaN HEMT
devices, Microelectronics Reliability (2017), http://dx.doi.org/10.1016/j.microrel.2017.07.049