1
High Mobility Strained Germanium Quantum Well Field Effect Transistor
as the P-Channel Device Option for Low Power (Vcc = 0.5 V) III-V CMOS Architecture
R. Pillarisetty, B. Chu-Kung, S. Corcoran, G. Dewey, J. Kavalieros, H. Kennel, R. Kotlyar, V. Le,
D. Lionberger, M. Metz, N. Mukherjee, J. Nah, W. Rachmady, M. Radosavljevic, U. Shah, S. Taft, H. Then,
N. Zelick, and R Chau
Intel Corporation, Technology and Manufacturing Group, Hillsboro, OR 97124, USA
Abstract
In this article we demonstrate a Ge p-channel QWFET with
scaled TOXE = 14.5Å and mobility of 770 cm
2
/V*s at n
s
=5x10
12
cm
-2
(charge density in the state-of-the-art Si transistor channel at
Vcc = 0.5V). For thin TOXE < 40 Å, this represents the highest
hole mobility reported for any Ge device and is 4x higher than
state-of-the-art strained silicon. The QWFET architecture
achieves high mobility by incorporating biaxial strain and
eliminating dopant impurity scattering. The thin TOXE was
achieved using a Si cap and a low Dt transistor process, which
has a low oxide interface Dit. Parallel conduction in the SiGe
buffer was suppressed using a phosphorus junction layer,
allowing healthy subthreshold slope in Ge QWFET for the first
time. The Ge QWFET achieves an intrinsic Gmsat which is 2x
higher than the InSb p-channel QWFET. These results suggest
the Ge QWFET is a viable p-channel option for non-silicon
CMOS.
Introduction
Recently, III-V quantum well field effect transistor (QWFET)
research for future low power CMOS logic applications has
made significant progress [1,3]. While n-channel III-V studies
have shown significant drive current gains over state of the art
silicon at low Vcc [1], the corresponding p-channel transistor
with thin TOXE and high mobility (µ) has not yet been
demonstrated. In this study, we demonstrate a high mobility
strained germanium (Ge) p-channel QWFET suitable for low
power CMOS architecture with scaled TOXE = 14.5Å and hole
mobility = 770 cm
2
/V*s at n
s
=5x10
12
cm
-2
.
For TOXE < 40 Å,
this represents the highest hole mobility reported for any Ge
device and is 4x higher than state-of-the-art strained silicon.
These results suggest that the Ge QWFET is a viable p-channel
option for III-V CMOS realization.
Materials Growth and Device Fabrication
Figure 1 shows a schematic of a biaxially strained undoped Ge
QW structure. The boron modulation doping layer allows for
Hall measurement, but is optional for implanted S/D transistors.
The phosphorus doped layer is grown to suppress parallel
conduction in the SiGe buffers. A cross sectional TEM image of
a Ge QW grown by RTCVD on 300mm silicon is shown in Fig
2, highlighting both the 2-step SiGe buffer layers and biaxially
strained Ge QW layer bounded by Si
.3
Ge
.7
barriers. Although not
shown, we also grew relaxed Ge layers on this two layer buffer
to provide us with a Ge MOSFET reference structure. Figure 3
shows X-ray diffraction spectra of the symmetric (004) reflection
for both the Ge QW and relaxed Ge structures indicating 1.3%
biaxial strain in the Ge QW. The Hall mobility for RTCVD
grown Ge QW structures, plotted in Fig 4, matches MBE grown
Ge QW literature data [5-7] and shows gain over the InSb QW[3]
and strained Si [2]. Figure 5 shows a TEM of a fully processed
Ge QWFET utilizing shallow trench isolation (not shown),
HfO
2
/TiN high-k metal gate, self-aligned B-implanted S/D, W/Ti
contacts, a strained Ge QW channel, and a phosphorus isolation
layer. A TEM image of a Ge QWFET with an in-situ doped
Si
x
Ge
1-x
raised source/drain (RSD) is shown in Fig 6.
Silicon Cap and Gate Dielectric Interface
A thin Si cap layer is required to prevent carrier spill-out from
the Ge QW. This is demonstrated in Fig 7 where k*p-Poisson
simulations show that for a hole density (n
s
)
= 5x10
12
cm
-2
,
a
10Å Si cap layer confines carriers in the Ge QW, whereas
significant carrier spill-out occurs with a 100Å Si
.3
Ge
.7
barrier.
Figure 8 shows a TEM image of a high-k metal gate stack with a
thin silicon cap on a Ge QW. Part of the silicon cap is oxidized
due to thermal cycle (Dt) during the transistor fabrication
process. This is suggested by the EDS depth profile of the gate
stack, shown in Fig 9, indicating the presence of both Si and SiO
2
between the Ge and HfO
2
. CV data in Fig 10 indicates inversion
TOXE reduction with Si cap thickness scaling. Due to
asymmetry of the valence and conduction band offsets between
Si and Ge, the Si cap only contributes to C
inv
. Hence, the SiO
2
thickness (T
SiO2
) on the Si cap can be extracted from the
accumulation TOXE, and in this example is 6Å for all cases due
to constant thermal Dt. Since a body contact is needed to
measure the accumulation CV, this data was collected from the
Ge MOSFET reference device. The corresponding µ vs n
s
plotted
in Fig 11 shows that µ improves as Si cap thickness is reduced
due to reduction in carrier spill-out. However, µ is degraded
significantly without Si cap due to an increase in interface trap
density (Dit). Figure 12 shows that by lowering process Dt from
700°C to 635°C TOXE can scale to 14.5Å without loss of
mobility via T
SiO2
reduction on the Si cap.
Ge QWFET Device Analysis
The minimal CV frequency dispersion in Fig 13 indicates a good
quality interface for both the relaxed Ge MOSFET reference and
strained Ge QWFET with the same 14.5Å TOXE process. Figure
14 shows mobility vs n
s
for both devices. The experiments agree
with k*p simulations, which assume Dit and surface roughness
matched to state-of-the-art Si. This indicates a high quality oxide
interface on Ge. At n
s
= 5x10
12
cm
-2
, the QWFET exhibits 4x
mobility gain over state-of-the-art strained Si [2]. Furthermore, in
Fig 15 the Ge QWFET achieves the highest mobility (770
cm
2
/V*s) at the thinnest TOXE (14.5Å) compared to the best Ge
devices in literature [8-9]. Figs 16 and 17 plot the temperature
(T) dependence of the Ge QWFET mobility, which shows no
saturation of µ down to T=20K. This indicates minimal impact
from Coulomb scattering due to absence of doping in the QW
and low Dit.