Chapter 2 Patterned Si Substrate, Experimental Apparatus and Scanning Electron
2.5 SEM images of GaAs in trenches with variant widths
The SEM images of GaAs in trenches with variant widths are shown in figure 2.6
to 2.14, with the orientation the same as figure 2.1. Through the investigation of SEM
images, we observed the filling rate is near hundred percent. In addition, there are
rhombic/eye-shaped GaAs islands above the STIs, with lengths about 1µm and widths
about 300nm, as a result of relatively low growth temperature compared to a previous
work presented by A. Okamoto et al. [13]. Inside the trenches, there are [110] direction
breakings between the GaAs bars and separate the entire GaAs bar in trenches into
several-microns-length bars. The breaking facets are all Si (1-10) facets. The heights of
the bars are not the same according to the contrast difference of the images.
Table 2.1 Different trench widths
Figure 2.1 (a) SEM image of 55nm trench, the lightest parts are the STI SiO2. Between
the STI are the trenches. There are 16 trenches at the upper side. The lowers are 1
micron, 500nm, and two 250nm trenches. (b) The schematic graph of the cross section
of the trenches. The side walls are amorphous silica and the bottom of trench are silicon.
Si
STI GaAs STI
[1-10]
[110]
[001]
SiO
2(a)
(b)
Figure 2.2 (a) The source assembly of Ga and As port in VG-V80H GSMBE. (b)
The molecular beams were paralleled to the long side of the trenches.
Group V cracking cell
Ga SUMO cell RF plasma cell
(b)
(a)
Figure 2.3 (a) The schematic graph of electron distribution, where Xd is the
diffusion path and R the maximum range (b) The schematic graph of FEGSEM
with CL system, where C1 and C2 are the electromagnetic lenses.
C1
Field Emission Gun
C2
Sample Electron beam
Luminescence
Parabolic mirror
(b)
Si CCD detector
(a)
Figure 2.4 (a) Schematic graph of photon scattering (b) Schematic graph of the
Rayleigh scattering anti-Stokes scattering Stokes scattering E
2E
1E
0(a)
(b)
Figure 2.5 (a) VPEC GaAs room temperature CL with variant exposure time. We can
see the peak intensity is proportional the exposure time, which means the high energy
16kV and 18nA electron gun does not influence the luminescence mechanism of GaAs.
(b) The laser polarization of T64000 is X direction. The incident direction is set to be z
direction. (c) Different polarization of (001) VPEC GaAs Raman spectra.
(a) (b)
(c)
Figure 2.6 The SEM images of GaAs in 40nm Si trenches: (a-d) middle of the trench long side with different magnification (e-h) right of the trench of the trench long side
(a) (b)
(c) (d)
(e) (f)
(g) (h)
Figure 2.7 The SEM images of GaAs in 55nm Si trenches: (a-d) middle of the trench long side with different magnification (e-h) right of the trench of the trench long side
(a) (b)
(c) (d)
(e) (f)
(g) (h)
Figure 2.8 The SEM images of GaAs in 70nm-wide Si trenches: (a-c) middle of the
trench long side with different magnification (d-f) right of the trench of the trench long
side with different magnification
(a) (b)
(c) (d)
(e) (f)
Figure 2.9 The SEM images of GaAs in 80nm-wide Si trenches: (a-c) middle of the
trench long side with different magnification (d-f) right of the trench of the trench long
side with different magnification
(a) (b)
(c) (d)
(e) (f)
Figure 2.10 The SEM images of GaAs in 100nm-wide Si trenches: (a-c) middle of the
trench long side with different magnification (d-f) right of the trench of the trench long
side with different magnification (a)
(e)
(b)
(c) (d)
(f)
Figure 2.11 The SEM images of GaAs in 120nm-wide Si trenches: (a-c) middle of the
trench long side with different magnification (d-f) right of the trench of the trench long
side with different magnification
(a) (b)
(c) (d)
(e) (f)
Figure 2.12 The SEM images of GaAs in 140nm-wide Si trenches: (a-b) middle of the
trench long side with different magnification (d-f) right of the trench of the trench long
side with different magnification
(a) (b)
(c) (d)
(e)
Figure 2.13 The SEM images of GaAs in 160nm-wide Si trenches: (a-b) middle of the
trench long side with different magnification (c-e) right of the trench of the trench long
side with different magnification
(a) (b)
(c) (d)
(e)
Figure 2.14 The SEM images of GaAs in 180nm-wide Si trenches: (a-b) middle of the
trench long side with different magnification (c-d) right of the trench of the trench long
side with different magnification
(a) (b)
(c) (d)
Figure 2.15 The SEM images of GaAs in 200nm-wide Si trenches: (a-b) middle of the
trench long side with different magnification (c-d) right of the trench of the trench long
side with different magnification
(a) (b)
(c) (d)
Figure 2.16 The SEM images of GaAs in (a-b) 300nm-wide trenches: middle of the
trench long side with different magnification (c) 500nm-wide trenches (d) 1µm-wide
trenches (e) 1.25µm-wide trenches (f) 1.5µm-width trenches
(a) (b)
(c) (d)
(e) (f)
Chapter 3
Cathodoluminescence of GaAs in Si Nanotrenches
3.1 Excitation Volume of Electron
As shown in figure 2.3 (a), when the electrons incident into a solid, the electron
diffuse to a water-drop shape distribution. The max penetration distance according to K.
Kanaya et al. [14] found to agree well with experimental results in a wider range of
water-drop form. Finally, since GaAs is only about 300 nm thick on or in the planar Si
or patterned Si substrates, the effective diffusion diameter at the bottom of GaAs is
estimated to be 2.6 µm.
3.2 CL spectrum analysis of SiO
2as the shallow trench isolation
3.2.1 Typical CL spectrum of SiO2 as STI
Figure 3.1(a) shows typical CL spectra of STI measured by scan mode. We can see
that there three bands, which are located at 1.9eV (645 nm), 2.2eV (560 nm) and 2.7eV
(460 nm), respectively. It is worth to mention that the intensity of these three peaks is
proportional to exposure time, as shown in Fig.3.1 (a), indicating that the 18
keV-energetic electrons bombardment does not influence the luminescence mechanism.
Figure 3.1(b) and (c) show the CL spectra measured by spot mode at 11K from the
side and the bottom of the trench, respectively. We can see both of the spectra are
almost the same. Since the trench width is 1µm, implying the electron dispersion range
much lager than 1 µm, as we discussed in the end of section 3.1,
3.2.2 Luminescence theory of SiO2
The defects in glass, vitreous-SiO2 quartz and fused silica were characteristic by
electron spin resonator (ESR) or electron magnetic resonator (EPR), photoluminescence
(PL), optical absorption and SEM/CL technology. Scientists could recognize the energy
level of defects by the hyperfine structure with isotopes characterized by ESR method.
On the other hand, optical absorption, CL and PL still can do a lot favor for evaluating
the quality of various SiO2.
3.2.2.1 The red band (R-band) luminescence of 1.9 eV (645 nm)
The oxide hole center network defect structure of this band, widely observed on
silica optical fibers and gate oxides, was firstly deduced by M. Stapelbroek et al [16].
The origin of this luminescence is generally attributed to the electron-hole
recombination at the non-bridging oxide hole center (NBOHC). This intrinsic defect
structure in SiO2 network is represented by ≡Si-O．, where “ ≡ ” represents three oxygen
atoms connected to a silicon atoms and “．” represents a single electron. The structure
was characterized by the 29Si hyperfine structure using ESR method [17]. Due to the
variation of this red band, different precursors have been proposed, like energetic
electron induced defect [18], trapped Si3+ [19], or interstitial ozone [20]. We believe that
the 1.9eV band is attributed to NBOHC, which had been reported tremendously.
3.2.2.2 The 2.2 eV (560 nm) band
This band is not that common as 1.9eV peak and 2.7eV peak in a-SiO2. The
literatures provide some perspectives: e-beam induced defect luminescence [21], singlet
transition of the oxygen vacancy and the oxygen vacancy–interstitial pairs self trapped
exciton (VO;(O2)i) [22] and intensity increase with electron irradiation [21]. To
summarize, we believe that this band is originated with the oxygen vacancy and the
oxygen vacancy–interstitial pairs self trapped exciton (VO;(O2)i). We confirm the
postulation by the elimination of luminescence at low temperature, which represents
this kind of defect prone to break when the temperature decreases.
3.2.2.3. The 2.7eV (460nm) band
The luminescence mechanism of this band is mostly attributed to the so-called E’
-centers or related to trivalent pure Si [23]. The E’ center represents a Si attach to three
oxide atoms and a single electron represented by ≡Si． in the SiO2 composition network
[16]. Nevertheless, electron bombardment and temperature influence the luminescence
peak intensity [23]. A. V. Amossov et al. claim that increasing of temperature may
increase the luminescence peak intensity [20]. Another group mentioned that the peak
intensity increase when Si surplus [24]. A. N. Trukhin et al. point out this band is usual
for oxygen deficient center (ODC) luminescence in a-SiO2. In SiO2–Si, oxygen
deficiency is increased, so for that sample the mentioned luminescence is characteristic
of SiO(2-δ), where δ=2e-4 [25]. According to M. Watanabe, the recombination of oxide
vacancy and interstitial oxygen as (VO ;Oi) pair give rise to the luminescence [26].
Finally, A. Zatsepin et al. believe this luminescence is due to trivalent silicon, intensity
increasing when silicon is surplus and H+ implant will reduce the emission [27]. Finally,
we attributed this band to the E’ center.
3.2.2.4 The reduction of 2.2eV and 2.7eV bands at low temperature (11K)
After the temperature decrease, we have seen the reduction of 2.2eV and 2.7eV
band at low temperature, as shown in Fig.3.1 (c) and (d). Since the 2.2 band still exist,
2.7eV band almost vanished. At the same moment, the 1.9eV peak slightly red shift to
1.858eV. The peak form an asymmetry peak, the low energy side shows a
original SiO2 structure, then no luminescence would occur.
3.3 Growth Condition of GaAs on planar and patterned Si (001) wafer
Samples were grown by almost the same procedure: C2584 was grown at 580℃
with a growth rate of 0.33 µm/hr. After every 60 nm thick film was grown, sample was
rotated by 180° and raised to 650℃ for 10 minutes. During the annealing, the sample
was illuminated by hydrogen plasma from a plasma cell. The plasma power was 230W.
For convenience, the growth conditions of C2584, C2585, C2586, C2587 and C2588
are listed in table 3.1.
To discuss the surface configuration, figure 3.2 (c), (d) and (e) are the SEM top
view image of C2585, C2586, and C2588 on planar Si, respectively. We can see that
epi-layer of C2585 has much more “holes” on the surface, yet the GaAs grains are
continuum everywhere. The C2588 surface also has holes on it, but they are slimmer,
and we can see many facets, like wrinkled paper with slots. C2586 is mostly like film,
with triangle vines toward [110] direction.
3.4 RTCL spectra of GaAs on bulk/patterned Si
Luminescence phenomena in GaAs are caused by radiative recombination of holes
and electrons. The direct band gap nature of GaAs gives rise to a high efficiency
conduction band to valance band transition during the recombination. The typical RTCL
of GaAs which is fabricated by VPEC with molecular organic vapor phase epitaxial
(MOVPE) growth is shown in figure 3.2(a), with 1.42eV peak position and 37meV full
width half maximum (FWHM). Its high energy side decays slowly owing to the high
energy tail states, and low energy side decay more sharply, representing that the density
of state (DOS) of radiative impurities inner the energy band are relatively less.
3.4.1 The RTCL peaks discussion of GaAs grown on planar Si (001) samples
1. C2585
Figure 3.3 (a) is the RTCL spectrum of C2588 GaAs on planar Si. Due to the
zigzag of the data of the spectrum, we utilize Fourier low pass filter with frequency 500
and two peak Gaussian fitting, the original spectrum resolved two Gaussian-like peaks,
one peak at 1.440eV with 127meV FWHM and another peak at 1.285eV with 88 meV.
2. C2586
Figure 3.3 (b) is the RTCL spectrum of C2586 GaAs on planar Si. Using the same
method as C2585, the peak is at 1.438eV with 167meV FWHM.
3. C2588
Figure 3.3 (c) is the RTCL spectrum of C2588 GaAs on planar Si. Utilizing Fourier
low pass filter with frequency 150 and first order derivative, the peak of C2588 is 9meV
blue shift from the VPEC GaAs peak. We use Gaussian fitting and received one peak at
1.433eV with 107meV FWHM and another 1.287eV peak with 89meV FWHM. Note
that the deviation of the 1.43eV peak, which is not totally normal distribution yet has a
trend like heteroepitaxial GaAs. A 1.44eV band was reported at 4 K and was attributed
to deep acceptor level luminescence [28]. In our cases, the RT three times broadening
near 1.44eV band may be attributed to other mechanisms. The 1.28eV peak was
reported by M. F. Millea using electroluminescence (EL) at 80 K, recognizing as a
donor-acceptor pair model [29]. We also attributed this peak to the deep donors to the
deep acceptors inner band transition. With the SEM image observation, we consider that
the 1.429eV peak broaden is due to the structure of GaAs. Inside the excitation volume,
every grain is postulated to be compressed or tensile because of the slots on the Si
surface, then due to the Fermi level consistency, the band gap of every grain
overlapping mutually and many localized states giving rise to the finally broaden peak.
Finally, we consider that the Burstein-Moss shift cause the blue shift effect.
3.4.2 RTCL spectra of GaAs in patterned Si
The RTCL spectra of GaAs in different trench width are shown in figure 3.4-3.7.
The GaAs signals are not that clear when trench width is less than 100nm. In figure
3.4(a), (b), (c) and (d), the luminescence of GaAs in 50nm trench to GaAs in 70nm
trench have almost the same shape. Even though the GaAs-like bands are so weak, after
subtract the SiO2 RTCL signal, we can see a 1.4eV neighboring band, which is
larger than the SiO2 luminescence and the FWHM of the peak is much larger than the
one of intrinsic GaAs. The broaden phenomenon has also been seen on the C2588 bulk.
In figure 3.4 (d), 3.5 (b) and 3.6(c), we can see higher luminescence intensity of
spot mode rather than scan mode. Since the higher efficiency, we finally use spot mode
to measure the whole trench.
The spectrum of trench 11, figure 3.5 (a), with 80nm width, has a triangle form
GaAs peak, with intensity much lower than the 1.9eV SiO2 peak. After subtraction the
SiO2 signal, utilizing Fourier low pass filter with frequency 1000, forming a peak at
1.452eV with 292meV FWHM. From figure 3.5 (b), trench 12, with 90nm width, also
reveals a band belongs to GaAs, utilizing the same method as trench 11, forming a peak
at 1.4626eV with 224meV FWHM. After the trench is larger than 100nm, the GaAs
peak intensity starts to exceed the SiO2 luminescence intensity of 1.9eV peak, with
much wider FWHM and blue shifts than the heteroepitaxial one. Trench 13, with 100nm
trench width, minus oxide signals and using Gaussian fitting, we receive three bands:
1.4505eV with 137meV width, 1,2763eV with 72meV and 2.1645 with 148meV. Trench
17, with 180nm width, minus oxide signals and using Gaussian fitting, we receive two
bands: 1.4533eV with 136meV width and 1.2732 with 80meV. Trench 18, with 200nm
width, minus oxide signals and using Gaussian fitting, we receive two bands: 1.4464eV
with 136meV and 1.2742eV with 84meV. Trench 19, with 300nm width, minus oxide
signals and using Gaussian fitting, we receive two bands: 1.4516eV with 111meV and
1.2856eV with 72meV width. Trench 20, with 500nm width, is similar to the bulk GaAs
luminescence in spectrum line shape, performing the high energy tail and sharp low
energy side and still has the 1.27eV band, using Gaussian fitting, we receive two bands:
1.445eV with 103meV FWHM and 1.284eV with 83meV.
In the end of these RT measurements, we provide the figure 3.7 (c) to make sure
the reliability of these works. Meanwhile, figure 3.7 (d) shows the main two peaks of
trenches widths larger than 100 nm. We can see the positions of the two peaks of all the
trenches larger than 100 nm are near the same. This fact indicates that the luminescence
mechanism has nothing to do with the trench width, in other words, the strain induced
by side walls. Besides the strain model, high density of twinning and stacking faults
inducing inner band states may be the reason of the overwhelmingly broadening peak.
The blue shift may because of the cross-doping effect or the Burstein-Moss shift.
3.4.3 LTCL spectra discussion of C2588 bulk
At low temperature about 11K, we had done a series of CL experiments. Figure
3.2(b) reveals the 12K CL spectrum of VPEC GaAs, which shows the exciton bound to
neutral acceptor-like point defect at 1.510eV and exciton (doublet) bound to neutral
carbon at 1.512eV.
Surprisingly, the spot mode on the same spot produced different results of the
C2588 GaAs on planar Si, different from the STI RTCL. This phenomenon indicated
that the 16kV with 18nA energetic electron at 11K would break some bonding in the
grains and induced other bands of GaAs. The figure 3.8 (b) to (d) was set on the same
dot and (e), (f) set on another. Figure 3.8 (b) contains three parts of Gaussian fitting:
1.5084eV with 29meV width, 1.4734eV with 142meV width, 1.3001eV with 157meV
width, 2.707eV with 556meV width. Figure 3.8 (c) contains three parts of Gaussian
fitting: 1.4193eV with 150meV width, 1.2647eV with 111meV width, 1.5067eV with
21meV width. Figure 3.8 (d) contains four parts of Gaussian fitting: 1.5681eV with
70meV width, 1.4439eV with 102meV width, 1.3621eV with 47meV width, 1.2893eV
with 124meV. Figure 3.8 (e) contains two parts of Gaussian fitting: 1.4696eV with
106meV width, 1.3213eV with 162meV width. Figure 3.8 (f) contains four parts of
Gaussian fitting: 1.2609eV with 112meV width, 1.4348eV with 51meV width,
1.3977eV with 107meV width, 1.5398eV with 147meV. Figure 3.11 (a) shows the fitted
peaks and their FWHM.
The LTCL of the C2588 sample gives problems to discussion. First, the 1.508eV
band is attributed to bound exciton to neutral acceptor like point defect. Second, the
Gaussian fitted peaks separately distribute from 1.2 simplest two spectra both have near
1.45eV peak and a 1.32eV peak. The 1.45eV peak may be the donor to Si acceptor. The
adjacent 1.26eV and 1.32eV peaks may be the GaAs inner band defect level..
3.4.4 LTCL spectra discussion of patterned C2588
We can see from figure 3.9 (a) and (b) that the spectra only reveal the silicon
dioxide luminescence parts, since the trench widths are too small. From trench width
larger than 80nm, there bands gradually reveals around 1.4eV and 1.55eV while the
vitreous glass luminescence still plays a rule in the excitation volume.
Observing figure 3.9 (c), the 80nm-width trench luminescence additional
luminescent 1.35eV and 1.5eV peaks. By the trench width enlarging, the intensity of
these two peaks enlarging in the mean time. After the trench width larger than 160nm,
the Gaussian fitting results new peak near 1.25eV. These three or four peaks compared
to the VPEC GaAs 1.42eV are obscure. Figure 3.11 (b) shows the Gaussian fitted peaks
positions versus trench width. In our opinion, we attribute this luminescence to the
shape-induced band splitting.
Table 3.1 The growth condition of C2584, C2585, C2586, C2587 and C2588
Figure 3.1 (a) The CL spectra of SiO2 STI side wall at RT using scan mode with variant
exposure time: 3 seconds, 10 seconds and 30 seconds. The peak energy intensities of
spectra are proportional to exposure time. (b) A CL spectrum of SiO2 STI sidewall at
11K using spot mode (c) a CL spectrum of SiO2 as STI at 11K using spot mode, the
cursor set at the bottom of the trench (c)
(a) (b)
Figure 3.2 (a) RTCL of GaAs on GaAs substrate obtained by SEM/CL method with
exposure time to be 1 second (b) The 11K case (c) C2585 SEM top view of GaAs
grown on planar substrate (d) C2586 SEM top view of GaAs grown on planar substrate
(e) C2588 SEM top view of GaAs grown on planar substrate (a)
(c)
(b)
(e)
(d)
Figure 3.3 (a) RTCL of C2585 GaAs on planar Si with exposure time 10 seconds and 2
cycle (b) RTCL of C2586 GaAs on planar Si with exposure time 10 seconds and 2 cycle
(c) RTCL of C2588 GaAs on planar Si with exposure time 10 seconds and 2 cycle (c)
(a) (b)
(c)
(a) (b)
(d)
Figure 3.4 (a) 50 nm trench CL spectrum using scan mode at X20k (b) 55 nm
trench CL spectrum using scan mode at X20k (c) 60 nm trench CL spectrum
using scan mode at X20k (d) 70nm trench CL spectra, there are spot modes
results and scan modes results. The GaAs peak of scan mode at X40k slightly
red shifts and broadens compared to scan mode at X20k.
Figure 3.5 (a) 80 nm trench RTCL spectrum using spot mode (b) 90 nm trench RTCL
spectrum using spot mode (c) 100 nm trench RTCL spectrum using spot mode (d)
120nm trench RTCL spectrum using spot modes
(b)
(d) (a)
(c)
Figure 3.6 (a) 140nm trench RTCL peaks using variant measure modes and variant
exposure time (b) 160nm trench RTCL peaks using spot mode (c) 180nm trench RTCL
peaks using spot mode (d) 200nm trench RTCL peaks using spot mode
(a) (b)
(a)
(c) (d)
Figure 3.7 (a) 300nm trench RTCL peaks (b) 500nm trench RTCL peaks (c) no beam
emission spectrum (d) peaks positions by Gaussian fitting versus trench widths, after the
trench larger than 100nm, mainly 2 peaks: around 1.29eV and around 1.45eV are fitted.
(c)
(b) (a)
(d) (a)
(c)
Figure 3.8 (a) RTCL spectrum of C2588 GaAs on planar Si with Gaussian fitting, (b)-(d)
are LTCL spectra on the same spot, (e) (f) are on another spot; the green curves are
are LTCL spectra on the same spot, (e) (f) are on another spot; the green curves are