Content uploaded by Fei Liu
Author content
All content in this area was uploaded by Fei Liu on Dec 24, 2014
Content may be subject to copyright.
Tunable normal incidence Ge quantum dot midinfrared detectors
Song Tong,a) Fei Liu, A. Khitun, and K. L. Wang
Device Research Laboratory, Department of Electrical Engineering, University of California
at Los Angeles, Los Angeles, California 90095-1594
J. L. Liu
Quantum Structures Laboratory, Department of Electrical Engineering, University of California
at Riverside, Riverside, California 92521
共Received 22 December 2003; accepted 13 April 2004兲
Midinfrared photodetectors in the 3–5
m region were demonstrated by using molecular beam
epitaxy grown self-assembled Ge quantum dots at normal incidence. The structure was a p-i-pwith
p-type doped Ge dots embedded in the intrinsic layer sandwiched in the two heavily p-doped
regions. The dark current density at 77 K is 6.4 mA/cm2at 1 V.The as-grown sample has a response
at normal incidence in the wavelength range of 2.2 to 3.2
m and peaked at 2.7
m. Thermal
annealing at 900°C for 10 min shifted the peak response to 3.6
m. Annealing effect was simulated
with the interdiffusion behavior of Ge and Si atoms to explain the shift of the response wavelength.
©2004 American Institute of Physics. 关DOI: 10.1063/1.1759081兴
I. INTRODUCTION
For midinfrared and far-infrared photodetection,
HgxCd1⫺xTe is currently the mainstream material due to its
suitable and tunable band gap, which makes it fit for these
wavelength ranges. However, this material is well known to
have many issues in controlling the growth, the processing,
and the fabrication of devices.1Thus quantum-well infrared
photodetector 共QWIP兲was proposed and investigated.2,3 But
QWIPs are a one-dimensional confinement system, and are
not sensitive to normal incidence light due to the selection
rule of intersubband transition. Thus optical coupling
through the use of diffraction gratings4or random scattering5
is necessary. In the recent years, quantum dots 共QDs兲such as
InGaAs6and Ge7dots have been successfully grown by self-
assembling processes. QDs are three-dimensional confine-
ment objects and thus can be used for the detection of normal
incidence illumination. Quantum dot infrared photodetectors
共QDIPs兲have attracted a great deal of interest.8–10 These
materials are also predicted to have many advantages such as
reduced phonon scattering, longer carrier lifetime, lower
dark current, increased optical absorption coefficient, and
higher detectivity. Ge quantum dots, comparing to their
III–V counterparts, are grown on Si substrates,6thus have
the potential to be monolithically integrated with the highly
advanced Si integrated circuits at low cost. In this paper, we
will present our photoresponse results on Ge quantum dots
and the controlling of the response wavelength.
II. EXPERIMENT
The sample used for experiment is very much like a
P-I-Pstructure, with p-type doped Ge dots in the intrinsic
Si region, sandwiched by two p⫹regions as contact layers.
Double-side-polished highly p-type doped Si 共100兲wafers
with a resistivity of 18–25 ⍀cm were used as substrates.
The substrates were cleaned by using a standard Shiraki
method followed by in situ thermal cleaning at 930°C for 15
min. The nominal growth rates are 1 and 0.2 Å/s for Si and
Ge, respectively. Boron doping was achieved by sublimation
of boron from a Knudsen cell. A 200 nm 5⫻1018 cm⫺3
p-doped Si buffer layer was first grown, followed by a 100
nm undoped separation layer with the substrate temperature
maintained at 600°C, then the temperature was reduced to
540°C during the following growth. Twenty layers of Ge
quantum dots were grown separated by 19 undoped Si layers
with each layer 20 nm. The nominal thickness of each Ge
layer was 1.4 nm and was grown with a boron doping level
of 1⫻1018 cm⫺3. Another 100 nm intrinsic and 200 nm 5
⫻1018 cm⫺3boron doped Si layers were grown on top. The
sample was then separated into small pieces for annealing.
Annealing was processed with rapid thermal annealing
共RTP兲in nitrogen at 900 °C for 10 min.
The as-grown and annealed samples were then processed
into mesa devices. Mesas were defined by dry etching with
CF4/O2 gases and a plasma power of 100 W. The wafers
were then treated with BOE:H2O2for 5 min to remove the
plasma damaged surface. Low-temperature LPCVD oxide
was then deposited for passivation. Contacts using Ti/Al
were finally formed by e-beam evaporation and lift-off tech-
nology. Ohmic contact was formed by 1 min annealing at
400 °C by RTP. The mesa size was 500⫻500
m2. The wafer
was diced and the chips were mounted on TO-5 packages for
further measurements. I–Vmeasurements were carried out
with HP4142 in dark. Photoresponse spectra were taken at
normal incident geometry.A glow bar operating at 70 W was
used as the light source. A 34 cm grating monochromator
was used to disperse the light. Photocurrent was measured
through a sampling resistor (R⫽1.5M⍀). A lock-in ampli-
fier was used to detect the voltage signal.
III. RESULTS AND DISCUSSIONS
A cross-sectional transmission electron microscopy
共TEM兲image of the dot region is shown in Fig. 1. The dot
a兲Author to whom correspondence should be addressed; electronic mail:
tong@ee.ucla.edu
JOURNAL OF APPLIED PHYSICS VOLUME 96, NUMBER 1 1 JULY 2004
7730021-8979/2004/96(1)/773/4/$22.00 © 2004 American Institute of Physics
Downloaded 24 Jun 2004 to 138.23.167.19. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
height ranges from 7 to 10 nm while the base ranges from 50
to 100 nm. One feature of the structure is that the dots are
vertically correlated and form stacks separated by Si separa-
tion layers. The vertical correlation shown is typical when
the separating Si layer is less than 50 nm. It was established
that the underlying dots provide maximum tensile strain for
the covering Si and these sites are favorable for forming
subsequent dots.11 As shown in the figure, the dot size in-
creases in upper layers, in both height and base dimensions.
The aspect ratio, height:base, is around 1:5–7. It can also be
seen that the dot density in the lower layers is large and
reduces in the upper layers. Some dot sites appearing in the
lower level disappear after one or several layers, as indicated
by the arrows in the image.
Figure 2 shows the dark current characteristics of the
as-grown QDIP device as a function of the bias voltage at 77
and 300 K. The dark current has an exponential increase with
increasing applied bias voltage at moderate biases. This
agrees with the theoretical analysis of Ryzhii et al.12 At 77
K, the forward current density at 1 V is 6.4 mA/cm2. The
total dark current consists of three parts, sequential resonant
tunneling, thermionic emission, and phonon assisted
tunneling.3Thermionic hole emission from QDs is the major
source for the dark current at these temperatures. The unsym-
metrical appearance of the I–Vcurves is due to the growth
induced unsymmetrical properties of the dot region, such as
the wetting layer and the dot shape. This is clearly shown in
Fig. 1. For the annealed sample, the dark current slightly
increases.
Figure 3 shows the normal incidence photoresponse
spectrum of the as-grown sample measured at 80 K. The
response ranges from 2.2 to 3.2
m and peaked at about 2.7
m. The full width at half maximum 共FWHM兲is 0.6
m,
which corresponds to 103 meV, resulting a relatively broad
absorption, a linewidth ⌬/of 22%. We ascribe this re-
sponse to the transition of holes in Ge dots from their bound
ground states to the continuum states on top of the barrier13
as illustrated in the inset of Fig. 3. There are many issues that
are responsible for the broadening of the spectrum, such as
the variation in dot size, Ge content, as well as local strain.
From the TEM image, we saw that the dot height and width
were different. Our PL results on samples grown at similar
conditions show a FWHM of 60 meV,14 indicating a salient
energy variation among the ground states for the dot en-
semble. This property is desirable for a wide range response
of the detector. On the other hand, the nonuniformity of the
dots results in degradation of the absorption intensity since
not all the dots participate in the absorption at a certain
wavelength.
To move the photoresponse to the more interesting 3–5
m range, we performed heat treatment to the sample. One
sample was annealed at 900°C for 10 min. The response
spectra at different temperatures from 80 to 120 K were
shown in Fig. 4. The response spectra range from 2.4 to 4.6
m and the peak response occurs at 3.6
m. This shows that
FIG. 1. TEM image of the as-grown sample. Arrows indicate where the dots
cease to continue in the upper layers.
FIG. 2. Current-voltage characteristic of as-grown sample measured at 77
and 300 K.
FIG. 3. Photoresponse of the as-grown sample at 80 K. Inset shows the
corresponding transition of the holes from the bound states to continuum
states.
774 J. Appl. Phys., Vol. 96, No. 1, 1 July 2004 Tong
et al.
Downloaded 24 Jun 2004 to 138.23.167.19. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
annealing may be an effective method to tune the spectral
range of quantum dot samples. The result also shows that the
response intensity decreases with increasing the temperature.
With increasing temperature, the bound holes can be excited
out of QD states through thermionic emission. This emission
reduces the number of bound carriers that can be photoex-
cited, thus reduces the photoresponsivity. Another reason
may come from the increasing phonon scattering, which can
reduce the excited carrier lifetime and reduce the optical
gain. The dips in the spectra are due to the interference effect
of the epilayer as well as the atmospheric absorption.
The shift of the response wavelength is the result of the
interdiffusion of Si and Ge during annealing. In order to
estimate the effect of heat treatment on Ge dot composition,
we carried out numerical calculations of Ge concentration
profile, taking into account Ge/Si interdiffusion processes.
The diffusion equation
c
t⫽Dⵜ2c共1兲
is solved by the finite differences method and forward time
centered step algorithm. In the equation, cis the concentra-
tion of Ge atoms, Dis the diffusion coefficient, and tis the
time. The algorithm is stable only if the so-called Courant-
Friedrichs-Lewy condition is met,
⭐h2
2 max
兵
Dx,Dy,Dz
其
,共2兲
where
and hare time and space steps, respectively. From
this condition, it follows that in the nanometer scale (h
⭓1 nm) and with the diffusion coefficient less than
10⫺15 cm2/s, the computational time step
must be less than
5s共
⭐5s兲, which is acceptable as typical annealing time is
about several minutes.
In Fig. 5, we show the numerical simulation results of
the Ge concentration profile for the SiGe dot embedded in a
100% silicon host matrix. At the initial time the dot is as-
sumed to have 75% Ge. Then, we simulated interdiffusion
process for the 10 min annealing. The in-plane diffusion co-
efficients Dxand Dyare taken to be 10⫺17 cm2/s(T
⬃1200 K) and the diffusion coefficient along the growth axis
zis taken to be ten times higher than the in-plane one (Dz
⫽10Dx), as a recent experimental investigation on Ge/Si
interdiffusion in GeSi dots has shown a significant diffusion
anisotropy.15 One can see a significant ‘‘broadening’’ of the
initial quantum dot region as a result of interdiffusion from
the initial steplike Ge distribution to a Gaussian-like shape.
After annealing, the average Ge content in the original dot
site decreases to 58% from the initial 75%. This can be con-
firmed by our former TEM results.16 Evidently, interdiffusion
substantially modifies the dot physical and electrical struc-
ture. In this simulation, we neglected the following effects.
First, we assumed that initially the dot has a uniform Ge
distribution, yet, in fact, the Ge content in the GeSi dots is
not uniform and usually the dots tend to have Ge-rich
cores.17 Second, we did not include the effects of strain and
Ge concentration on the diffusion coefficients, as there are no
sufficient experimental data on these parameters. Neverthe-
less, this simulation still provides useful information to pre-
dict the effect of thermal treatment on dot electronic and
optical properties.
As we have pointed out previously, the optical transi-
tions are from the bound heavy hole states to continuum
states. To evaluate this energy level, a three-dimensional va-
lence band finite deep potential box model with effective
mass approximation was used. We calculated the response
wavelength dependence on dot parameters, i.e., Ge content,
strain, QD height, and base width. Since quantum dots are
partially relaxed,18 a strain ratio (a
储
dot⫺aSi)/(aSiGe⫺aSi)is
used to describe the percentage of strain in QDs. First, the
effective masses of relaxed SiGe alloys were obtained by
using the Vegard’s law 1/mSiGe
*⫽x/mGe
*⫹(1⫺x)/mSi
*. Then,
the change of effective mass due to strain was obtained by
using a two band k*pmethod. The valence band offset of
Si1⫺xGexstrained layers on 共100兲Si substrate were taken
from the results of Rieger and Vogl.19 In Fig. 6 we show the
response wavelength dependence on Ge content and strain.
The dot height and base were taken to be 9 and 100 nm,
respectively. The four curves correspond to four values of
relative strain in the dot, 100%, 80%, 60%, and 40%. The
FIG. 4. Photoresponse of the 900 °C and 10 min annealed sample at differ-
ent temperatures.
FIG. 5. Simulation results of the Si and Ge interdiffusion at 900 °C for 10
min. Si0.25Ge0.75 were assumed for the dot before annealing. The dot was 9
nm in height and 100 nm in base.
775J. Appl. Phys., Vol. 96, No. 1, 1 July 2004 Tong
et al.
Downloaded 24 Jun 2004 to 138.23.167.19. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
calculation results show that lower Ge content in QD and
smaller QD strain ratio are necessary in order to achieve
longer wavelength response. At the growth temperature of
540°C, we can estimate the Ge content in the dot to be
75%.20,21 Considering this together with the response peak of
2.7
m, we can proceed to determine the relative strain to be
about 65%. This point is denoted by the diamond sign in the
plot. The point after annealing is also shown in the plot as a
square sign. We can see that the relative strain keeps about
the same value as before annealing. This means that the ab-
solute strain value decreased since the Ge content decreased.
The calculation also shows that the peak wavelength changes
about 17% as the dot height increases from 3 nm to 9 nm for
a constant Ge content and strain. The response wavelength
has a weak dependence on the base dimension since the base
dimension is much larger than the height.
As we have seen, one effect of annealing is the interdif-
fusion between Ge and Si, the Ge content in the quantum dot
decreases after annealing. The second effect is to make the
QDs become more relaxed. Both effects lead to a smaller
valance band offset and thus the response shifts to longer
wavelength. Another effect is that QD becomes larger and
this will give shorter wavelength response. However, our cal-
culations show that the variation of QD height at around 9
nm has a relatively weak effect on the response wavelength
for QD comparing to those due to Ge content and QD strain.
Besides the postannealing process that we investigated
here, by varying the Ge deposition thickness, the energy lev-
els can also be tuned. We observed a 28 meV peak energy
shift in photoluminescence spectra for samples grown with
1.2 and 1.5 nm Ge. Comparing to the 115 meV shift for the
rapid annealing, this method is relatively less effective.
The absorption coefficient of this kind of devices was
studied using identical structure grown on lowly doped Si
double side polished substrate with waveguide geometry.
The results show an absorption coefficient of 5800 cm⫺1.
This value is high comparing to the III–V quantum wells,
which is typically 400–1800 cm⫺1,3presumably due to the
high doping density in our structure. Research is going on to
investigate the absolute responsivity in our laboratory.
VI. CONCLUSION
Molecular beam epitaxy grown Ge dots photodetectors
were fabricated. The p-i-pstructure showed normal incident
photoresponse. As-grown samples had a response range from
2.2
mto3.2
m. Annealing was shown to be able to tune
the response to longer wavelength range, i.e., 2.5–4.8
m
for 900°C 10 min annealed samples. The normal incident
response from boron-doped Ge QDs was ascribed to transi-
tions from heavy hole ground states to continuum states in
the valence band. The dark current density at 1 V was 6.4
mA/cm2at 77 K, which could be optimized by tuning the
doping level and the spacing layer thickness in the active
region. The response wavelength calculations agree with the
experimental results. It was also pointed out that lower Ge
content and smaller dots are more favorable for the longer
wavelength range response. To achieve this, smaller nominal
Ge growth thickness and relatively higher growth tempera-
tures can be used. In addition, by using laser annealing, the
same tuning results may be obtained at selected areas, and
this can give a matrix of photodetectors with different re-
sponse peaks. Ge quantum dot arrays on Si substrate may
find potential application for midinfrared photodetectors.
ACKNOWLEDGMENTS
Financial support by MURI under the Centroid project
and phonon program is gratefully acknowledged.
1A. Sher, M. A. Berding, M. van Schilfgaarde, and C. An-Ban, Semicond.
Sci. Technol. 6,C59共1991兲.
2L. Esaki and H. Sakaki, IBM Tech. Discl. Bull. 20, 2456 共1977兲.
3B. F. Levine, J. Appl. Phys. 74,R1共1993兲.
4D. Heitmann and U. Mackens, Phys. Rev. B 33, 8269 共1986兲.
5E. Yablonovitch and G. D. Cody, IEEE Trans. Electron Devices ED-29,
300 共1982兲.
6D. Leonard, M. Krishnamurthy, C. M. Reaves, S. P. Denbaars, and P. M.
Petroff, Appl. Phys. Lett. 63, 3203 共1993兲.
7H. Sunamura, S. Fukatsu, N. Usami, and Y. Shiraki, J. Cryst. Growth 157,
265 共1995兲.
8V. Ryzhii, Semicond. Sci. Technol. 11, 759 共1996兲.
9P. Bhattacharya, A. D. Stiff-Roberts, S. Krishna, and S. W. Kennerly,
SPIE-Int. Soc. Opt. Eng. Proceedings of SPIE, International Society for
Optical Engineering, 46,100共2002兲.
10H. C. Liu, M. Gao, J. McCaffrey, Z. R. Wasilewski, and S. Fafard, Appl.
Phys. Lett. 78,79共2001兲.
11 O. G. Schmidt, O. Kienzle, Y. Hao, K. Eberl, and F. Ernst, Appl. Phys.
Lett. 74, 1272 共1999兲.
12V. Ryzhii, V. Pipa, I. Khmyrova, V. Mitin, and M. Willander, Jpn. J. Appl.
Phys., Part 2 39, L1283 共2000兲.
13B. F. Levine,A. Zussman, S. D. Gunapala, M. T. Asom, J. M. Kuo, and W.
S. Hobson, J. Appl. Phys. 72, 4429 共1992兲.
14S. Tong, J. L. Liu, J. Wan, and K. L. Wang, Appl. Phys. Lett. 80, 1189
共2002兲.
15J. Wan, Y. H. Luo, Z. M. Jiang, G. Jin, J. L. Liu, K. L. Wang, X. Z. Liao,
and J. Zou, J. Appl. Phys. 90, 4290 共2001兲.
16J. Wan, Y. H. Luo, Z. M. Jiang, G. Jin, J. L. Liu, K. L. Wang, X. Z. Liao,
and J. Zou, Appl. Phys. Lett. 79,1980共2001兲.
17J. Tersoff, Phys. Rev. Lett. 81, 3183 共1998兲.
18Z. M. Jiang, X. M. Jiang, W. R. Jiang, Q. J. Jia, W. L. Zheng, and D. C.
Qian, Appl. Phys. Lett. 76, 3397 共2000兲.
19M. M. Rieger and P. Vogl, Phys. Rev. B 48, 14276 共1993兲.
20J. L. Liu, J. Wan, Z. M. Jiang, A. Khitun, K. L. Wang, and D. P. Yu, J.
Appl. Phys. 92, 6804 共2002兲.
21G. Capellini, M. De Seta, and F. Evangelisti, Appl. Phys. Lett. 78,303
共2001兲.
FIG. 6. Calculated results on the peak response wavelength dependent on
the Ge content and Ge QDs strain ratio for QDs with the height of 9 nm and
the base of 100 nm. The diamond and square signs indicate the points for the
as-grown sample and the annealed sample, respectively.
776 J. Appl. Phys., Vol. 96, No. 1, 1 July 2004 Tong
et al.
Downloaded 24 Jun 2004 to 138.23.167.19. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp