In order to analyze the trade-off between high-frequency performance and threshold voltage achieved by the gate recess technique [42], results from two-dimensional hydrodynamic simulations supported by experimental data [44] are presented [393].
AlGaN/GaN DHEMT and EHEMT structures with T-gates of 250 nm length
share the same layer specification and are processed on the same SiC
wafer. The devices consist of GaN buffer, 22 nm thick
AlGa
N (Al
GaN
N for the EHEMT
device) barrier layer, 3 nm thick GaN cap layer, and SiN passivation
(Fig. 5.46). The cap and part of the barrier layer under
the gate of the EHEMT are recessed by Cl
plasma etching. A
remaining AlGaN barrier thickness t
11 nm is
assumed. The Ohmic contacts are assumed to reach the 2DEG in the
channel.
The densities of the polarization charges at the channel/barrier
interface and at the barrier/cap interface are determined by calibration
against the experimental data to be 910
cm
and
10
cm
, respectively. Low Ohmic contact
resistances of 0.2
mm are considered
[44]. Self-heating effects are accounted with a substrate
thermal contact resistance of R
=5 K/W. This value lumps the
thermal resistance of the nucleation layer and the substrate.
Fig. 5.47 compares the simulated transfer characteristics to
experimental data. Both devices are simulated using the same set of
models and model parameters, including the interface charge
densities. A good agreement is obtained, both for the threshold
voltage and the transconductance
(Fig. 5.48).
The mismatch between the drain current at high gate voltages is due to
the high gate leakage current in the real device, for which the
simulation does not account. A possible explanation for the
underestimation of the peak
for the recessed device is an
overestimation of the sheet resistance under the gate. Simulated
output characteristics for both structures are compared to
measurements in Fig. 5.49 and
Fig. 5.50.
The RF simulations provide slightly higher
cut-off frequency
than the experiments for both structures
(Fig. 5.51). Note, that both the measured and simulation data
show an increase of
and
for the EHEMT structures.
Since the gate capacitance depends on the gate channel distance, several
simulations are performed with variable recess depths, which means a
variable barrier thickness t
under the gate. As expected a
shift in the threshold voltage is observed (Fig. 5.52)
and
increases with decreasing t
(Fig. 5.53) due to the lack of charge control for thicker
layers.
![]() |
![]() |
However, the simulated
characteristics do not show any
noticeable change (Fig. 5.54). Fig. 5.55
shows the gate source capacitance which increases with decreasing
t
. It compensates the increase in
, thereby
resulting in a constant
. Thus, the major reason for the rise of
and
of EHEMTs in comparison to DHEMTs
(Fig. 5.51) is the absence of barrier/cap negative interface
charges under the gate. The exact depth of the recess has less
influence on
and
, but has a significant impact on the
threshold voltage and the transconductance.
![]() |
![]() |