[ieee 2010 18th international conference on advanced thermal processing of semiconductors (rtp) -...
TRANSCRIPT
![Page 1: [IEEE 2010 18th International Conference on Advanced Thermal Processing of Semiconductors (RTP) - Gainesville, FL, USA (2010.09.28-2010.10.1)] 2010 18th International Conference on](https://reader030.vdocuments.us/reader030/viewer/2022020301/5750a8c11a28abcf0ccafc72/html5/thumbnails/1.jpg)
18th IEEE Conference on Advanced Thermal Processing of Semiconductors - RTP 2010
Location- and orientation-controlled, large single grain silicon induced by pulsed excimer laser crystallization
M.R. Tajari Mofrad, R. Ishihara, J. van der Cingel and C.I.M. Beenakker
Delft University of Technology, Faculty of Electrical Engineering, Mathematics and Computer Science (EEMCS)
Department of Microelectronics and Computer Engineering (ME&CE) Delft Institute of Microsystems and
Nanoelectronics (DIMES) Feldmannweg 17, POBox 5053, 2600 GB Delft, The Netherlands Tel: +31-(0)15-2787307 Fax: +31-(0)15-2787369
Abstract- We studied pulsed laser induced epitaxy of silicon using a seeding wafer to realize location
and orientation-control of silicon grain. Silicon grains as large as 4 I'm x 4 I'm with mostly the
preferred (100) orientation area were obtained on top of contact openings through Si02 to seeding
silicon (100) wafer. The orientation of the seed is inherited by a-Si during the solidification phase of
molten-Si. The maximum process temperature of this process is 545°C which is for LPCVD deposition
of a-Si. This layer is suitable for high mobility SOl CMOS devices which serve as building blocks for
monolithic 3D integration.
I. Introduction
Down-scaling of transistors increases the
complexity of the process and thus the
fabrication. Whether there will be a hard limit to
down-scaling, it still needs to be determined [1].
However, higher transistor density needs more
space reserved for routing purposes, which
decreases the advantage of the scaling.
Moreover, the interconnect delay increases in
such a way that it becomes the limiting factor for
circuit performance [2].
Three-dimensional (3D) integration is a solution
for these issues as it shortens the interconnect
length and lowers power consumption. It also
increases the functionality of a chip by enabling
integration of a sensor layer on top of it. It can
be considered low cost since the expenses
required for developing new equipment or new
deep sub-micron processes is not needed [3].
978-1-4244-8399-0/10/$26.00 © IEEE
Between several existing 3D IC technologies,
monolithic integration is promising to offer the
highest density and speed. However, obtaining
high quality crystalline silicon (c-Si) in low
temperatures, demanded by monolithic character
of this technology is usually a severe challenge.
In the past, we have reported location-controlled
large single grains (SG) silicon obtained by the
so-called J.l-Czochralski process [4]. By placing
the channel of the MOS devices inside such
large grain, we could obtain thin film transistors
with mobilities as high as 600 cm2Ns [5][6].
This principle is shown in Fig. 1. Here we see
two adjacent silicon grains with a channel of the
TFT positioned in a grain-boundary-less area.
But lacking a single seeding crystal orientation
caused the grain to grain crystal orientation
differences. The point for improvement lied in
the non-uniformity of the electrical
characteristics of devices, since mobility of a
MOS transistor is dependent on the
crystallographic orientation of the channel
![Page 2: [IEEE 2010 18th International Conference on Advanced Thermal Processing of Semiconductors (RTP) - Gainesville, FL, USA (2010.09.28-2010.10.1)] 2010 18th International Conference on](https://reader030.vdocuments.us/reader030/viewer/2022020301/5750a8c11a28abcf0ccafc72/html5/thumbnails/2.jpg)
material. Controlling this parameter can reduce
the variability of transistors in a 3D Ie.
Laser induced epitaxy is the perfect method to
achieve this goal. By opening a hole into a
seeding layer e.g. bulk c-Si the
created in this manner reduce the hole diameter
down to 300 nm. The cavities are filled with 250
om thick low-pressure chemical vapor
deposition (LPCVD) a-Si at 545 °C. The
schematics of the process is shown in Fig. 2.
Pulsed XeCI excimer laser (A,=308 nm) is used
D MOO transistor Channel to melt the silicon. The schematics of the laser
D D crystallized silicon
D seeding openning
Figure 1: Principle of high mobility TFTs built on SG
siliconamorphous silicon
(a-Si) contacting the seed layer can inherit the
crystallographic orientation of the seed while
solidifying from the molten state due to laser
irradiation. By using a pulsed laser system we
can minimize the thermal damage of this process
step to the underlying layers.
There has been some research done in the field
of laser induced epitaxy [7]. However, a
comprehensive study on the subject is still
missing. Here, we studied pulsed laser induced
epitaxy process of silicon using a seeding wafer
to realize the location- and orientation-control of
the silicon grain. 4 /lm x 4 /lm large grains are
obtained and characterized. The effect of laser
pulse duration, diameter of the openings to seed
layer and the type of the seed wafer have been
investigated to optimize the epitaxy process.
II. Experimental
We start with deposition of 700 nm thick Si02 by
plasma enhanced chemical vapor deposition
(PECVD) on top of a (lOO)-oriented wafer,
which is either bulk or SOL This is done at 350
°C by using tetra-ethy l-ortho-silicate (TEOS) as
precursor. We palOOttern holes with diameters
varying from 0.5 to 2 /lm by dry etching the
Si02. The diameter of the holes is reduced by
deposition of a 500 nm of Si02, followed by
mask-less anisotropic dry etching. The spacers
system is shown in Fig. 4. There are two laser
modules which can perform independently, but
also in combination with a certain delay set by
the pulse generator unit shown in Fig. 3. The
smoothed normalized envelope of several pulse
shapes which can be obtained by this system are
shown in Fig. 5. Each module has 25 ns wide
pulse. By increasing the delay between the two
laser units, we can obtain a pulse width with a
maximum envelope of 100 ns which is shown by
the green line in Fig. 5. Beside changing this
delay, we can also engage the pulse extender
unit which will extend the pulse width of each
unit up to 250 ns for each. This extends the
pulses up to 500 ns wide (FWHM) as shown by
the black dotted line. The substrate temperature
during the crystallization can be raised up to 450
°C. Figure 5 shows the smoothed normalized
envelope of the pulse intensities.
The thickness of the 500 nm underlying TEOS
layer is optimized for the minimum reflectance.
Thus variations due to exposure energy
fluctuations must be minimized. Figure 3 shows
the SEM pictures of the holes after etching. Left
has been etched by purely dry-etching, while the
right sample has been etched partly dry with a
wet landing step. The bright ring around the
opening is a measure for the slope of the oxide
sidewall, knowing that the diameter of the
patterned holes were initially equal after the
photo resist patterning. The angle of the side
wall varies from 65° to 84° with straight forward
geometric calculations. The focus of this paper
is the results of the seeding holes with 84 °
angle.
![Page 3: [IEEE 2010 18th International Conference on Advanced Thermal Processing of Semiconductors (RTP) - Gainesville, FL, USA (2010.09.28-2010.10.1)] 2010 18th International Conference on](https://reader030.vdocuments.us/reader030/viewer/2022020301/5750a8c11a28abcf0ccafc72/html5/thumbnails/3.jpg)
III. Simulation
We use two different simulation methods based
on finite element method (FEM) technique for
predicting the thermal profile and/or the tracking
the melt front. For fast predictions and
comparison, we use the enthalpy method
including the latent heat. For more precise
simulations we use the comprehensive phase
field approach which enable us to track the melt
and solidification front. For the enthalpy
method, thermal characteristics of each layer e.g.
heat capacity, thermal conductance and density
is used. Also laser related parameters
like reflection and absorption @308 nm is
considered. Phase-field approach uses all those
parameters accompanied by data about the
temperature dependent melt velocity and surface
tension. For more detail about this simulation
see [8].
(b)
Figure 2: Schematics of Pulsed laser induced epitaxy process with bulk wafer seeding (a), and SOl wafer seeding (b).
Figure 3: Two examples of dry-etching (left) and wetdry-etching combination and its effects on the oxide sidewall slope.
w. X-Y-Z1iIIIF f
Figure 4: Schematics ofXeCI pulsed excimer laser system
• 25 ns single pulse --·double pulse
double pulse with 2/8 pulse extender -double pulse with 4/8 pulse extender --·double pulse with full pulse extender
6 -7
x10
Figure 5: Measured pulse shapes with the excimer
laser system.
![Page 4: [IEEE 2010 18th International Conference on Advanced Thermal Processing of Semiconductors (RTP) - Gainesville, FL, USA (2010.09.28-2010.10.1)] 2010 18th International Conference on](https://reader030.vdocuments.us/reader030/viewer/2022020301/5750a8c11a28abcf0ccafc72/html5/thumbnails/4.jpg)
------ -- -- ---. 1600 c-SI meillemp_ -",'/-
;;< .. --, -"f 1400 /.,,, ....... : ... ..' � 1200 ........ ,,;:/.' � , ..... E 1000 --- 1 - - SUlk Wale' I 8OOL�_-=-=-- S=O=1 w=af =e ':J
0.6 0.8 I 12 Fluence (J/cm2( 1.4
� 1400 - - - -a:srmeTftomp�------ ----., �1300 e 8-�1200 11w"-______ -'
0.05 0.06 0.07 0.08 Fluence (J/cm2(
Figure 6: Maximum temperature as a function of laser energy density at a-Si/c-Si interface (left) and aSi surface and (right)_ This is done for the bulk wafer substrate with an elevated temperature of 450°C and a pulse width of 70 ns_
IV. Results and Discussion
With an absorption coefficient of le8 m-I, a-Si
absorbs the largest amount of laser light in its
first 10 nm and melts. This thin layer of silicon
then acts as the heat source for the underlying
layers. As the melting continues to sink deeper
in the a-Si layer, the c-Si seeding layer is
reached by the melt front. Since the melting
temperature of the c-Si is almost 200 °C higher
than a-Si, it is not always melted by just coming
in contact by the molten-Si front. In case the c
Si interface is not reached, the seed for
solidification will be fine grain polysilicon. Thus
the crystallized layer will have multiple
crystallographic orientations which means that
the epitaxial growth is not succeeded. We define
the onset of epitaxial growth to be a the laser
energy density by which the melting
temperature is obtained at the interface of
seeding c-Si and a-Si. This results in a
successful epitaxy process in which, the
crystallized a-Si layer will have the same
crystallographic orientation as the c-Si of the
seeding layer. In Fig. 6, simulated temperature
behavior at specific points in both SOl and bulk
wafer seeding structures have been plotted
against the increasing laser fluence. A substrate
temperature of 450 °C and a pulse width of 70
ns were considered in the phase-field simulation.
The schematic of each structure is shown in Fig.
2. Left figure shows the onset of epitaxial
growth with both SOl and bulk seed layer
against increasing laser energy density. The
structure with SOl wafer seeding reaches the
onset of epitaxial growth with approximately
100 [mJ/cm2] less energy density than that of
the bulk, which means that one needs less laser
energy to have an epitaxial growth for this
structure. The SOl wafer contains less silicon at
the seeding area. The high thermal conductivity
of Si and low thermal conductivity of Si02 cause
the SOl case to have a better heat confinement.
The right figure indicates the minimum laser
energy density needed to melt the a-Si on the
surface. It is rouphly the same for both structure
since it is mainly surface dependent.
A simple enthalpy simulation using two pulses
of 25 ns and 250 ns produces Fig. 7. Here the
temperature history is shown for several
coordinates in the structure. The X coordinate
stands for the lateral position in [m], from the
center of the epitaxy hole and y for the depth in
the structure. The schematic of each structure is
shown in Fig. 2. We observe that while the c
Si/a-Si interface reach the same temperature in
shortly after one another, the maximum surface
temperature is decreased by 500 °C using the
longer pulse. The ablation temperature of silicon
is assumed to be roughly around its boiling
temperature of 2600 K. Although this
temperature is not reached in this simulation, the
surface temperature of both structures can be
relatively compared. It was found that the
ablation threshold is increased due to more
uniform temperature profile in case of long pulse
causing more energy to be given to the structure,
resulting in a deeper melt. By using shorter
pulses, the heating is more abrupt, causing the
maximum surface temperature reach the ablation
temperature.
The experimental counter parts of these results
are summarized in Table I. The minimum
energy densities needed for epitaxial growth and
![Page 5: [IEEE 2010 18th International Conference on Advanced Thermal Processing of Semiconductors (RTP) - Gainesville, FL, USA (2010.09.28-2010.10.1)] 2010 18th International Conference on](https://reader030.vdocuments.us/reader030/viewer/2022020301/5750a8c11a28abcf0ccafc72/html5/thumbnails/5.jpg)
Temperature [K) 2000 �����������
1800
g1600
<II 3 1400
� �1200 E <II f--1000
800
-(4e-8,0) - (4e-8, -8e-8) -(4e-8, - 2 . S e - 7) -(4e-8, -7e-7)
6000 0.5 2 2 1 1.5 .5 3 3.5 4 4.5 5
Time x10-7
Temperature [K) 1 4 00 .-����������
1300
1200 Q �1100 :::> � 1000 <II c. E 900 <II f--
800
700 ILL-._-
6000 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time x10-7
Figure 7: Simulated temperature history, for two
different pulse widths of 25 ns (left) and 250 ns
(right), with the same energy density of 1.4 mJ/cm2.
Bulk seeding was considered in this calculation The
depicted coordinates are in [m]. X value stands for
the lateral position, zero begin the center of the
epitaxy hole, y for the depth in the structure
to reach ablation threshold are shown. Process
window is defined to be the energy density
difference between the onset of epitaxy and the
ablation threshold. Though ablation is not a very
abrupt happening at die level. By comparing the
onset of epitaxy between the short and long
pulses, it is obvious that the process window is
increased by using longer pulses. This is valid
for both SOl and bulk wafer seeding. This can
be justified by the more uniform temperature
profile in case of longer pulse, which is due to
the more uniform heat flow in to the system. We
observe that while using the shorter pulses, SOl
wafer still leads to successful epitaxy. The onset
of epitaxy is so close to ablation threshold of the
bulk wafer seeding structure, that no large area
epitaxy is observed. However, longer pulse
responds better with bulk wafer. The increase in
process window is larger in case of bulk wafer
seeding. Since heat sink of this structure is better
than the SOl wafer, the flux towards the heat
sink is also larger in magnitude. With other
words, once the surface is not ablated which
happens in case of short pulses, the surface
remains cooler and allows more heat to flow into
the system without reaching ablation
threshold. Silicon grains with 4 Ilm x 4 Ilm area
were obtained on top of the holes. Arrays of
Si02 openings of 84 0 sidewall with 4 Ilm pitch
were designed. Areas as large as 2.5 mm x 1.7
mm, which is the laser beam spot size, were
crystallized. By EBSD measurement, the
orientations of these grains have been
investigated. Preferred (l00) orientation have
been observed for samples of both bulk and SOl
seeding layers. In case of 250 ns pulse, energy
density for crystallization is less than 1500
mJ/cm2 for a SOl seeding wafer and 1600 mJ/cm
2 for
bulk wafer seeding, we do not melt the c-Si seed
layer. Thus the seed for nucleation consists of
several orientations.
Figure 8 shows the SEM image of an array of
grains crystallized by 1.4 mJ/cm2 laser energy
density. Since this energy is lower than onset of
epitaxial growth, the results is an unsuccessful
epitaxy. The red lines are drawn to emphasize
the contrasts that exist within each large grain,
which are indication for existence of several
sub-grains. Figure 9 shows a EBSD
measurement of 25 Ilm x 25 Ilm large array of
grains, similar to SEM picture of Fig. 8. It
indicates that many crystallographic orientations
exist in the surface of the crystallized layer. In
Fig. 10, the SEM image of an array of grains is
shown, which are crystallized by a 250 ns pulse
![Page 6: [IEEE 2010 18th International Conference on Advanced Thermal Processing of Semiconductors (RTP) - Gainesville, FL, USA (2010.09.28-2010.10.1)] 2010 18th International Conference on](https://reader030.vdocuments.us/reader030/viewer/2022020301/5750a8c11a28abcf0ccafc72/html5/thumbnails/6.jpg)
with an energy densitiy of 2 J/cm2• This energy is
higher than values mentioned as onset of epitaxy
in Table 1. One can observe that the epitaxy in
case of bulk wafer seeding appears less
defective. This could be due to the fact that
crystalline quality of bulk wafer is generally
higher than a SOl wafer. Figure 11 shows a
EBSD measurement of 25 /lm x 25 /lm large
grain array, similar to SEM picture of Fig. 10. It
indicates that the (l00) is the main
crystallographic orientation that exists in the
surface of the crystallized layer. In EBSD and
pole figure, the existence of four secondary sub
grains is visible. Figure 12 indicates the size and
position of these sub-grains. A TEM cross
sectional image of an epitaxy process with a
laser fluence of 2 J/cm2 is shown in Fig. 13. In
this figure, we can see a successful epitaxy
process, but also the formation of the sub-grains
due to twin formation originated at the oxide
sidewall.
Table 1: The onset of ablation and epitaxial growth shown for the shortest and longest pulse duration, for different substrate types. All energies have [mJ/cm2] unit. The stepping for the energy density increase of the laser has been 50 [mJ/cm2]. Substrate temperature was raised to 450 °e.
[0011 1 1 1
001 101
Figure 9: EBSD mapping (left) and pole figure (right) of an epitaxially grown sample with (100) bulk wafer seeding with a laser fluence of laser fluence of 1.4 J/cm2. No crystallographic orientation-control is achieved.
Figure 10: Successful epitaxial process for bulk (left) and SOl (right) wafer with (100) orientation by a laser fluence of 2 J/cm2 which appears to be sufficient.
,------,--------------,------------------, Pulse 25 [ns] 250 [ns]
width Ablation Epitaxy Ablation Epitaxy
Seed type [001] SOl 1500 1400 2000 1500 1 1 1 Bulk 1550 2200 1600
Figure 11: EBSD mapping (left) and pole figure ........... iooiiiioi----......
...J(right) of an epitaxially grown sample with ( 100) bulk wafer seeding. Laser fluence of 1600 mJ/cm2 is used. Figure 8: Unsuccessful epitaxial process for bulk
(left) and SOl (right) wafer with (100) orientation by a laser fluence of 1.4 J/cm2 which appears to be insufficient.
![Page 7: [IEEE 2010 18th International Conference on Advanced Thermal Processing of Semiconductors (RTP) - Gainesville, FL, USA (2010.09.28-2010.10.1)] 2010 18th International Conference on](https://reader030.vdocuments.us/reader030/viewer/2022020301/5750a8c11a28abcf0ccafc72/html5/thumbnails/7.jpg)
Figure 12: AFM (left) and surface TEM image (right) of epitaxially grown single cry stalls showing the existance of four sub-grains.
Figure 13: TEM a cross-sectional image of a epitaxy process with a laser fluence of 2 J/cm2• Formation of facets at oxide sidewalls is shown.
V. Conclusion
Orientation- and location-controlled grains as
large as 4 J.lm x 4 J.lm were obtained by the
pulsed excimer laser induced epitaxy process.
Longer pulse durations result in deeper melt
depths and higher surface ablation threshold.
Seeding from a SOl wafer seems to require less
laser energy density due to better heat
confinement of the structure, due to thinner
seeding silicon and the low thermal conductivity
of underlying buried oxide layer. However, by
increasing the pulse duration, bulk wafer has a
larger increase in process window and more
suitable to be used as seeding layer for epitaxy.
The quality of the crystallized layer appears to
be better in case of bulk wafer which can be
explained by the better quality silicon. The oxide
sidewall angle can be varied by changing the
etching type. However, the defect density
remained mostly equal for both sidewall angels.
This study shows successful location- and
orientation-controlled large 4 J.lm x 4 J.lm silicon
grains, with (100) orientation, which can be used
for transistor layer of the building blocks for
high quality 3D IC, applicable for e.g. SRAMs
and SoCs.
Acknowledgment
The authors would like to thank the ICP
members and staff of DIMES research center for
their support during the fabrication of these
devices and the Dutch Technology Foundation
STW for their financial support.
References
[I] International Technology Roadmap for Semiconductors 2009 edition, Process Integration, Devices, and Structures, pp. 5. [2] K. C. Saraswat and F. Mohammadi, "Effect of Interconnection Scaling on Time Delay of V LSI Circuits", IEEE Transaction Electron Devices, Vol. ED-29, pp. 645-650. ( 1982) [3] M. Koyanagi et aI., "Future System-on-Silicon LSI Chips"IEEE Micro, Vol. 18, NO. 4, pp.17-22. ( 1998) [4] P. C. van der Wilt, B. D. van Dijk, G. 1. Bertens, R. Ishihara, and C. I. M. Beenakker, Appl. Phys. Lett. 79, pp. 18 19. (2001) [5] R. Ishihara et aI., "Single-grain Si TFTs with ECRPECV D Gate Si02", IEEE Transactions on Electron Devices, Vol. 51, NO. 3, pp.500-502. (2004) [6] V. Rana et aI, "High Performance Single Grain Si TFTs inside a Location-Controlled Grain by J.lCzochraski Process with Capping Layer", Proceedings of International Electron Devices Meeting 2006, pp. 37-3. [7] Yong-Hoon Son, "Laser-induced Epitaxial Growth (LEG) Technology for High Density 3-D Stacked Memory with High Productivity", Proceedings of International Electron Devices Meeting 2007, pp. 5B-3. [8] M. R. Tajari Mofrad, A. La Magna, R. Ishihara, H. Ming, K. Beenakker, "A three-dimensional phasefield simulation of pulsed laser induced epitaxial growth of silicon", Journal of Optoelectronics and Advanced Materials, Vol. 12, NO. 3, pp.lOI-706. (2010)