The investigations will be based on three manufactured SHPHEMTs
without a double recess, i. e. LDR = 0. The shape of
the Tgates of the devices are not know quantitatively, hence a simplified
cross section is used in the simulations as shown in Figure
6.6.
The epitaxial structure common to all devices from bottom to top consists
of a GaAs buffer on a S. I. GaAs substrate followed by a 12 nm In0.2Ga0.8As
channel layer, a 3 nm undoped Al0.23Ga0.77As spacer
layer, a 15 nm Al0.23Ga0.77As layer with an active
doping of about 3.5 1018cm-3 and a 7 nm undoped Al0.23Ga0.77As
Schottky barrier layer. The top layer is formed by a highly doped GaAs
cap to facilitate the formation of the source and drain ohmic contacts.
The different geometries of the three investigated SHPHEMTs A, B,
and C are given in Table
6.1.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
In Section 5.3.1 a qualitative discussion of the most important parameters and their impact on the transconductance was given. Based on these considerations only a small difference of gm max must be expected between devices A and B as they differ only in LG. Only for device C a slightly higher value for gm max is expected due to a significantly shorter LR which reduces the series resistance RS. The gate length variations are not likely to have a large impact on gm max but will have other consequences. The short LG of HEMTs A and C will lead to a small CGS and, thus, to a higher fT compared to HEMT B. However, further consequences of the parameter variations in Table6.1 cannot be easily estimated quantitatively. Such consequences are the increase of the output conductance g0 (and the decrease of the voltage gain gm/g0) that is expected with a decrease of LG or LR. A small LR will cause a small RS, but unfortunately, a large CGD undesirable for high fT and high power-gain cutoff frequency fmax. Numerical simulations are requested for the calculation of these effects.
All DC and RF measurements were performed on HEMTs with a gate width of 4 x 40 = 160 µm. For the simulation another fitting procedure had to be performed as the epitaxial structure is substantially different from the structure of HEMTref.
First, the simulation of HEMT A was fitted to the measurements as described in Section 5.3.2. A HD model was used only in the channel, whereas DD was applied on all other semiconductor layers. For the low field mobility µ in the InGaAs channel, the value obtained from Hall measurements of an equivalent layer structure was adopted.
The saturation velocity assumed for GaAs is unrealistically low. This
was deliberately chosen to compensate the overestimation of the buffer
current for the reason described in Section 5.2.
Other main fitting parameters are dGC, the concentration
of active dopant atoms and a constant interface charge density between
the passivation and the semiconductor [77].
The data given in Table
6.2 and Table
6.3 lead to the best simultaneous fit to the DC measurements
of threshold voltage VT, drain current ID
and transconductance gm and are well within their
respective ranges of uncertainty.
|
|
|
|
|
|
Interestingly enough, significantly better transport properties (vsat
and b) in the channel had to be used for the
simulation compared to HEMTref.
This assumption goes along with a very high activation (80 %) of dopands
assumed in the supply. As shown in the fitting procedure of Section
5.3.2 the uncertainty of the velocity in the channel is very high,
but the differences in the simulation are large enough to assume that the
use of a GaAs barrier under the channel improved the transport properties
compared to HEMTref.
Helmut Brech 1998-03-11