adam microdiskfabcyclicrie after jsk comments final
Post on 13-Dec-2014
110 Views
Preview:
TRANSCRIPT
Cyclic deep reactive ion etching with mask replenishment
T.N. Adam, S. Kim, P. Lv, G. Xuan, S. K. Ray, R. T. Troeger, D. Prather, and J. Kolodzey
Department of Electrical and Computer Engineering University of Delaware
Newark, DE 19716, USA(T.N. Adam, presently at IBM)
ABSTRACT
A multi-step reactive ion etching (MS-RIE) process for silicon was developed
for the fabrication of deep anisotropic, closely packed structures with vertical sidewalls.
This process used repeated cycles of etching and the replenishment of masking layers,
similar to the Bosch process [1] that is employed in specialized etching tools. The process
described here, however, can be used on conventional RIE tools, and is based on the
isotropic deposition of an etch-inhibiting polymer to protect sidewalls, its anisotropic
removal from the bottom etch front, and a subsequent isotropic etch into deeper layers. A
conventional parallel plate etcher without fast gas management, cryogenic substrate
cooling, or inductively coupled plasma density enhancement, produced these steps. Each
process step was optimized for maximal etch rate, minimal mask erosion, deposition of the
thinnest polymer required to protect the sidewalls, and was tailored for use with 2 µm
thick photoresist as the initial mask layer. This cyclic RIE process was used to fabricate
photonic devices with high aspect ratios of etched depths over 100 µm and etch widths
near 1 µm
1
1. Introduction
For micro-electromechanical systems (MEMS) applications, the etching of
silicon is required to relatively large depths (tens of microns) with high aspect ratios
(depth of trench versus lateral opening). Conventional wet etchants, such as hydrofluoric
acid mixed with nitric acid [2], result in isotropic undercutting below the masking
material. Crystal-orientation dependent etchants, such as potassium hydroxide [3], sodium
hydroxide [4], ethylene diamine pyrochatechol [5], and tetramethyl-ammonium hydroxide
[6], produce sidewalls that are straight, but that are inclined along preferred crystal planes.
For most of these chemicals, the wet etching rate differs for the plane orientations,
resulting in shapes that are restricted to rectangular and pyramidal forms. The wet etching
rates of silicon also depend on doping concentration and conductivity type [7]. In contrast,
“dry” ion etching is somewhat independent of crystal orientation, produces nearly vertical
sidewalls, and only weakly depends on doping. Typical dry Reactive Ion Etching (RIE)
inside a vacuum chamber occurs due to surface erosion of the desired sample by reactive
species and/or ion bombardment. The most common and least expensive etching tool is
the Capacitively Coupled Reactive Ion Etcher (CC-RIE). Compared to other dry etching
systems, such as Inductively Coupled Plasma etchers (ICP) [8], or Magnetically Enhanced
RIE (ME-RIE) [9], the ion directionality and density in the CC-RIE plasma is relatively
low. Consequently, the etching rate of CC-RIE is comparatively low and the anisotropy is
poor at rf-powers and gas pressures that are similar to the other techniques. Highly
anisotropic shapes can be obtained through increased ion bombardment at lower pressures,
similar to that in Chemically Assisted Ion-Beam Etching (CAIBE) [10] or Reactive Ion
Beam Etching (RIBE) [11], but at the expense of lower etch rates and faster mask erosion
due to reduced mask selectivity. Unlike ICP or ME-RIE, CC-RIE does not allow for the
independent adjustment of plasma density and ion energy. With photoresist as a mask, it is
2
undesirable to use ion bombardment alone to achieve anisotropy because of the poor etch
rate selectivity between photoresist and silicon.
The process that is patented and licensed by Robert Bosch GmbH [12] relies
on the ion-bombardment-enhanced deposition of a polymer that protects the mask and the
sidewalls during the deep etching of the desired structure. This Bosch Process makes use
of certain gases and gas mixtures, such as CF3H, c-C4F8, or CF4 + H2, that can deposit
polymers anisotropically during etching, while the etch rate of silicon at the bottom etch-
front is greatly reduced. With increasing ion energy, the otherwise isotropically depositing
polymer is re-sputtered from all horizontal surfaces that are subject to ion flux, and it
tends to re-deposit on the sidewalls of emerging features, during a process sub cycle called
the deposition step. The ion energy during this deposition step is increased to enable the
so-called “forward scattering” of the sidewall polymer by ion bombardment, which is
employed to remove and re-distribute the accumulating polymer from the top of the
structures to along the sidewalls towards the etch front. The polymer on the sidewalls
prevents their undercutting during etching. This procedure creates a deposited profile for
homogeneous sidewall protection allowing higher aspect-ratio etching but requires the use
of robust “hard” masking materials such as oxides or metals. In the Bosch process, during
a subsequent low-energy, chemically driven process sub cycle called the etching step, the
deposited polymer acts as a micro-mask that temporarily prevents the sidewalls from
being etched. During the Bosch process, the deposition and etching steps are periodically
cycled until the required depth is reached. High etch rates are achieved using ICP systems
that are capable of switching gases and stabilizing pressures quickly (~1 second), and by
increasing the local plasma density using a remote inductive coil. The ion energy required
for polymer re-deposition is adjusted separately from the plasma density by changing the
power that is capacitively coupled to parallel plate electrodes. In contrast, conventional
CC-RIEs do not have this capability to independently adjust ion density and energy, and
consequently the Bosch process does not work on parallel-plate RIEs. Therefore it would
3
be useful to have an alternative cyclic etching technique that can be performed on
conventional CC-RIE systems, which lack inductively coupled plasma (ICP) excitation.
We describe here a novel anisotropic multi-step reactive ion etching (MS-RIE) technique
that combines the deposition of micro masking layers with their cyclical replenishing,
removal, and deep etching. This technique was developed to fabricate structures with deep
and smooth vertical walls and large vertical depth-to- horizontal opening aspect ratios for
MEMS and photonic applications.
2. Experimental Method
We have developed a novel cyclic process that can be performed on
conventional capacitively coupled RIEs, and that does not require an ICP reactor as does
the Bosch Process. The process described here was tailored for deep silicon structures
using photoresist masks, and was optimized for high etch rates and low mask erosion. It
has been tested to an etched depth of 140 m on a PlasmaTherm 790 RIE system. The
absence of inductive coils, separate energy and ion density adjustment, fast gas
management, and cooled sample stages makes this process, using conventional RIE
systems, an inexpensive alternative to ICP tools using the original Bosch process.
Compared to optimized Bosch recipes on fully equipped ICPs as reported by Aayón et al.
[13], however, the net etch rates of our cyclic process are much lower (1.25 µm /hr vs. 4
µm/min), and the etch depth non-uniformity across large samples (3 inch substrates) is
somewhat degenerated ( 60 % vs. 4 % for the optimized Bosch process). In our process,
the forward scattering of the sidewall polymer is intentionally avoided by using higher
pressures (several hundred millitorr) during the etching step that yield a greater etch
selectivity of silicon over photoresist by having lower energy ions, but at a lower net etch
rate. In the absence of polymer forward scattering, the polymer accumulates more heavily
at the top of mesas and produces a slightly tapered sidewall profile. We found that the
4
inclination of the sidewall depended on the final thickness of protective sidewall polymer
and the etch depth.
The reactive-ion etcher used for this process development was a parallel plate
PlasmaTherm 790 system [14], equipped with an 8-inch diameter water-cooled aluminum
substrate holder as shown in Fig. 1. The chamber was evacuated by a TMP-151C
turbomolecular pump (145 l/s) backed by a Leybold D25BCS mechanical pump (18.1
cfm), to pressures typically around 0.1 mTorr. Process gases were managed by a bank of
mass flow controllers and introduced through an 8-inch showerhead configuration nozzle
to ensure a homogeneous gas flow pattern. The processing gases were sulfur-hexafluoride
(SF6), tetrafluoro-methane (CF4), oxygen, hydrogen, argon, and helium. We chose an SF6
based etch chemistry due to its inherently high silicon etch rate compared to CF 4 etching.
Pressure stabilization and gas management were computer controlled using a feedback
controlled throttle valve. The RF-power (at 13.56 MHz) was capacitively coupled to the
aluminum substrate holder (bottom electrode). The grounded showerhead electrode was
located 7 cm above the powered substrate holder. Through quartz view ports, the plasma
glow and dark regions could be observed.
5
Figure 1: A schematic diagram showing the layout of the reactive-ion etcher
(PlasmaTherm 790 system). The RF-power (at 13.56 MHz) was capacitively coupled to
the aluminum substrate holder (bottom electrode). The grounded upper electrode was
perforated to allow gases to flow through in a showerhead configuration.
As shown schematically in Figure 2, the cyclic MS-RIE process consisted of
three main sub cycles or steps: (A) isotropic etching, (B) polymer deposition, and (C)
anisotropic polymer removal. To develop this process, first the isotropic etching (A) and
the polymer deposition (B) steps were optimized separately for maximum etching rate and
the selectivity ratio of mask/silicon etching. Subsequently, the anisotropic removal (C)
step was adjusted during cycling experiments in combination with the other two steps.
6
2.1 Isotropic Etch (STEP A)
The mask pattern of the microstructures to be fabricated was transferred onto
a Si substrate coated with 2µm thick positive photoresist (AZ5214E [15]), by contact-
exposure photolithography. Positive photoresists usually exhibit unwanted reflow
characteristics during the final hardbake, resulting in uneven thickness and sloped
sidewalls. Because straight resist sidewalls were crucial for the MS-RIE, the standard
photolithography process was modified by hard baking at 125ºC for 1 minute before
exposure and by using longer than normal developing times. These conditions produced
nearly straight sidewalls with negligible bowing. A fully processed
Figure 2: Left: Schematic depiction of the three steps in the MS-RIE cycle and their effect
on an etched mesa of silicon. (A): Isotropic etch at high pressures to produce depth with
a slight amount of lateral undercutting and ripple, which can be minimized until it is not
noticeable by reducing the duration of this step. (B): Isotropic polymer deposition to coat
all horizontal and vertical surfaces with protective polymer. (C): Anisotropic polymer
removal at low pressures optimized to remove polymer from horizontal surfaces only,
without attacking the original photoresist or the vertical sidewall protection. Right top:
7
Drawing of finished anisotropic etch showing the sidewall protecting polymer layer (not to
scale), and scalloped edges due to the slight undercutting of the isotropic etching step.
Right bottom: The scanning electron micrograph shows a 4.5µm deep edge after 10
cycles demonstrating scalloping (intentionally exaggerated in this sample for illustration).
The masking material was 0.2/5 kÅ of Ti/Au. [16]
8
Structure (microdisk) is displayed on the right side of Fig. 2, showing a scalloped vertical
sidewall that was intentionally exaggerated by using a longer than optimal isotropic etch
sub-cycle. Figure 2 shows a portion of a circular mesa structure before undercutting to
form the final pedestal base of the mesa. Compared to tetrafluoro-methane (CF4) as a
processing gas, the sulfur-hexafluoride (SF6) had much higher silicon etch rate. In Figure
3, the dependence of the silicon removal rate on RF-power and gas pressure in an SF6
plasma is displayed, together with the self-induced voltage on the powered electrode. The
lowest pressure for a stable plasma was 4 mTorr, at which high aspect ratios were
obtained by ion-assisted sputter etching.
At relatively low pressures (30 mTorr and below), the chemical driven etch
rate of silicon was suppressed relative to the anisotropic physical sputtering processes due
to the insufficient concentration of reactive etch molecules and the higher ion energy due
to the longer mean free path. The length of the Crooke's dark space in the plasma column,
and the magnitude of the self-induced voltage on the powered electrode were larger at low
pressures, but the etch selectivity of photoresist relative to silicon was extremely poor.
Under low-pressure, high-energy conditions, only hard masking materials, such as Ni, Cr,
Au, or Al withstood sufficient exposures so that deep etching (>10 µm) could be obtained,
even at moderate RF-powers.
In addition to the RF-power and gas pressure, the gas mixture also affected the
etch rate. Low pressure etching in pure SF6 resulted in rough surfaces, similar to that from
XeF2 etching [17]. Adding CF4 and/or H2 to SF6 decreased the etch rate and also
produced rough surfaces (rms surface roughness ≈ 500nm). Mixing SF6 with O2 reduced
the etch rate (see Figure 4), but produced smooth surfaces (rms surface roughness ≈
100nm). Unfortunately, the etch selectivity of photoresist over silicon decreased
drastically due to photoresist ashing by O2.
9
50 55 60 65 701.9
2.0
2.1
2.2
2.3
2.4
2.5
Etc
h R
ate
(k
Å/m
in)
4mT SF6
RF-Power (Watt)
220
240
260
280
300
Vo
lta
ge
(V
)
0 10 20 30 402
3
4
5
6
Etc
h R
ate
(k
Å/m
in)
Pressure (mTorr)
0
50
100
150
20050W, SF6
Vo
lta
ge
(V
)
(a)
(b)
50 55 60 65 701.9
2.0
2.1
2.2
2.3
2.4
2.5
Etc
h R
ate
(k
Å/m
in)
4mT SF6
RF-Power (Watt)
220
240
260
280
300
Vo
lta
ge
(V
)
0 10 20 30 402
3
4
5
6
Etc
h R
ate
(k
Å/m
in)
Pressure (mTorr)
0
50
100
150
20050W, SF6
Vo
lta
ge
(V
)
(a)
(b)
Figure 3: Data for the isotropic etching step. (a): The dependence of silicon etch rate and
self-induced electrode voltage on RF-power in an SF6 plasma showing near linearity. For
these measurements, the masking material was the photoresist NR5 [18], of thickness
2.2 µm. (b): The silicon etch rate and the self-induced electrode voltage showing the
effects of pressure in an SF6 plasma at constant power. The slight difference in etch rates
at 4 mTorr, between the top and bottom graphs, was most likely caused by employing
samples of different sizes for the two experiments (this macro-loading effect is described
in the text).
10
0 20 40 60 80 1000
2
4
6
8
10
Etc
h R
ate
(k
Å/m
in)
200mT and 50W
%He Flow in SF6 - He Mixture
0 10 20 30 40 50 60 700
1
2
3
4
5 90mT and 50W
Etc
h R
ate
(k
Å/m
in)
%O2 Flow in SF6-O2Mixture
(a)
(b)
0 20 40 60 80 1000
2
4
6
8
10
Etc
h R
ate
(k
Å/m
in)
200mT and 50W
%He Flow in SF6 - He Mixture
0 10 20 30 40 50 60 700
1
2
3
4
5 90mT and 50W
Etc
h R
ate
(k
Å/m
in)
%O2%O2 Flow in SF6-O2Mixture-O
2Mixture
(a)
(b)
Figure 4: Data for the isotopic etching (A) step. (a): Etch rate of silicon as a function of
SF6:He gas mixture, showing nonlinear dependence on He content. A maximum etch rate
was found at a gas composition of 1:2 SF6: He. (b): Silicon etch rate as a function of
oxygen percentage in SF6 plasma, showing the decrease in etch rate as the relative
amount of SF6 deceases. The process gas pressure and RF-power are given in the
insets.
11
The gases that increased the etch rates were argon and helium. While argon worsened the
surface roughness, helium produced relatively smooth surfaces (tens of nm scale).
Therefore for our isotropic etching step (a), we chose flow rates of 10 sccm (standard
cubic centimeter per minute) of SF6 and 20 sccm of He, resulting in a maximum etch rate,
as displayed in Figure 4a. To minimize the mask erosion in He:SF6 mixtures, the RF-
power was lowered to 50 W, where we observed the etch rate selectivity ratios of silicon
over photoresist to be as high as 30. Reducing the RF-power decreased the etch rate,
which could be compensated by increasing the gas pressure to 200 mTorr, at the expense
of etch uniformity across the wafer, as shown in the top graph of Figure 5. At pressures
near 200 mTorr, the self-induced electrode voltage was reduced to less than 20 V, and the
etch process was chemically driven rather than bombardment-enhanced. Consequently, the
silicon removal rate depended strongly on the area of the exposed silicon surface, as
shown in Figure 5b, where changes in etch rate by about a factor of five were observed
versus different amounts of exposed silicon. We addressed the dependence on exposed
area by adding “dummy” substrates of blank Si to maintain a nearly constant area during
processing.
As a technical comparison, ICP reactors typically perform the isotropic
etching step using pressures between 15 and 25 mT and RF-powers of 5 to 15 W that,
without further plasma density enhancement, would yield extremely low but very
homogeneous etch rates. In ICP reactors, the plasma density is usually enhanced by
microwave coil powers between 700 and 2000 W, yielding etch rates as high as 4µm /min
on 150 mm wafers. However, the heat generated within the sample during high-rate
etching has to be removed by a helium gas flow in a heat sink at the sample backside,
resulting in a more complex process than that described here. On the other hand, in our
cyclic RIE process, the etch rate is lower but does not cause significant sample heating.
12
Etc
h R
ate
(k
Å/m
in)
Etc
h R
ate
(k
Å/m
in)
(a)
(b)
0 10 20 30 400
24
68
1012
141618
Distance from Center (mm)
0 1000 2000 3000 40000
2
4
6
8
10
3inch wafer
1/4 3inch
Chips 200mT 50WSF6:He(1:2)
Exposed Area (mm2)
Etc
h R
ate
(k
Å/m
in)
Etc
h R
ate
(k
Å/m
in)
(a)
(b)
0 10 20 30 400
24
68
1012
141618
Distance from Center (mm)
0 1000 2000 3000 40000
2
4
6
8
10
3inch wafer
1/4 3inch
Chips 200mT 50WSF6:He(1:2)
Exposed Area (mm2)
Figure 5: Data for isotropic etching step (A). (a): Etch depth versus position on a 3-inch
wafer after 60 cycles, each consisting of: 360 sec (per step sub-cycle) of deposition (step
b) at 40 W in 200 mTorr of 25:15 sccm CF4:H2, followed by 54 sec of removal (step c) at
100 W in 5 mTorr of 16:8 sccm He:SF6, and 95 sec of isotropic etching (step a) at 50 W in
200 mTorr of 20:10 sccm He:SF6. (b): Etch rate versus exposed silicon area showing the
macro-loading effect: larger exposed silicon areas etched slower than centimeter sized
chips, due to the local depletion of reactants.
13
2.2 Isotropic Polymer Deposition (STEP B)
When tetrafluoro-methane (CF4) was used in our CC-RIE, much smaller
silicon etch rates were obtained as compared to SF6, even at high pressures and RF-
powers. When CF4 was mixed with hydrogen, however, polymer deposits were readily
visible on the chamber walls. With increased percentage of hydrogen flow, the silicon etch
rate decreased as displayed in Figure 6a. At relative H2 flows above 28%, silicon was no
longer etched, but a Teflon-like polymer was deposited isotropically on the sample. It was
realized that this process could be used to deposit and replenish the masking layers. For
instance, in one variant of the Bosch process, the polymer deposition step was performed
at ion energies sufficiently high to disrupt the deposition on horizontal surfaces and cause
selective deposition on vertical sidewalls, which in the two-step Bosch process was
followed immediately by the etching step. In our CC-RIE system, at a gas mixture of 25
sccm CF4 and 15 sccm H2 with pressures as low as 30 mTorr and powers as high as 500
W, which represents our upper power (and ion-energy) limit, the impinging reactive
species were not energetic enough to disrupt the polymer formation on horizontal surfaces.
As a consequence, when using CF4, even under strongly enhanced ion bombardment, the
polymer could not be selectively removed from the etch front while remaining on the
sidewalls for protection during the subsequent isotropic etch. It was therefore necessary to
insert a third step using SF6, (STEP C: polymer removal, described below) that
anisotropically removed the deposited polymer from the horizontal etch-front surfaces. As
shown in Fig. 6, a hydrogen flow of 38 % was chosen for our deposition step. An accurate
determination of the deposited thickness was difficult because the polymer was not etched
easily and therefore were not be cross-sectioned easily to display an accessible edge.
Subsequently, we found that the mask polymer could be removed by the
following techniques. To get rid of the polymer that originated from the CF4 gas based
chemistry during the deposition process of RIE, the samples were exposed to an oxygen
plasma at 200 mTorr for 30 min. and then cleaned with H2SO4/H2O2 (3:1, 120C, 10 min.)
14
or NH4OH/H2O2/H2O (1:1:5, 80C, 5 min.) chemicals. After this, the samples showed no
polymer left on the silicon surface.
0 10 20 30 40 50-200
-100
0
100
200
300
400
500
600 200mT 40W
Etc
h R
ate
(Å
/min
)
%H2Flow in CF4 -H2 Mixture
Resist
Silicon
Polymer
(a)
(b)
0 10 20 30 40 50-200
-100
0
100
200
300
400
500
600 200mT 40W
Etc
h R
ate
(Å
/min
)
%H2Flow in CF4 -H2 Mixture
Resist
Silicon
Polymer
0 10 20 30 40 50-200
-100
0
100
200
300
400
500
600 200mT 40W
Etc
h R
ate
(Å
/min
)
%H2Flow in CF4 -H2 Mixture
Resist
Silicon
Polymer
(a)
(b)
Figure 6 :(a): The net etch rate of silicon versus percentage of hydrogen in total flow of
CF4-H2 mixture. Negative etch rates indicate deposition. (b): Scanning electron
micrograph of etched edge that was cleaved. Scallops (intentionally exaggerated) and
deposited polymer from the multiple cycles are clearly visible on the vertical edge.
15
Using scanning electron microscopy (SEM), the deposited polymer thickness was
determined by measuring the thickness of cleaved sidewalls with polymer accumulated
after many cycles, as displayed in Fig. 6b. From its outgassing and reaction by-products,
the polymer itself could contribute contaminants to subsequent etches, especially during
the non-polymer based sub-cycle steps. For example, polymer particles from the removal
step were re-deposited onto the substrate to form micro-masks that, with subsequent
etching, left behind individual and clustered micro-pillars, commonly known as “grass” or
“black silicon”. This effect is shown in Figure 7 for a 20µm wide waveguide structure that
was etched in 100 cycles of the conditions designated as Process 2 of Table I, except with
a removal (step C) time that was reduced from 32 to 8 seconds in order to exaggerate the
amount of micro masking that was induced by particles. Other gases, such as C2F4 or
CH3F have been reported to exhibit similar polymer deposition properties and may
substitute for CF4.
2.3 Anisotropic Polymer Removal (STEP C)
During the anisotropic polymer removal step, the protective polymer mask
deposited in step B above was removed on all horizontal surfaces by a high ion-energy,
low-pressure plasma that was produced by lowering the process pressure and increasing
the RF-power. The RIE settings were 100 W power, 5 mTorr pressure, with flow rates of
8 sccm of SF6, and 16 sccm of He, and the self-induced electrode voltage was
approximately 370 V. The timing of this “smash” step duration was critical to minimize
both mask erosion and the amount of residual polymer on the bottom surface of the etch
front. Shorter durations resulted in less mask erosion but left behind a polymer residue,
which ultimately manifested as “grass” or “black silicon”, observable on a blank Si
scavenger wafer underlying and supporting the intended etch wafer. Longer than optimal
etch times prevented the formation of silicon grass but yielded considerable mask erosion.
16
A compromise was found by increasing the step duration until no black grass discoloration
of the scavenger wafer was visible after several etch cycles.
(a)
(b)
(a)
(b)
Figure 7: Scanning electron micrographs of multistep-RIE etched annular ring and linear
structure after several cycles, showing the effects of re-deposited polymer and its
incomplete removal. (a): A clean surface after 73 cycles with optimized polymer removal
times (Process 1 of Table I). Note the proximity of the annulus to the linear structure (b):
Rough surface after 100 cycles of Process 2 in Table I, with an intentionally reduced
polymer removal time of 10 seconds (instead of the optimal time of 32 seconds), and an
etch depth of approximately 8.5µm. Micro-masking as a result of incomplete polymer
removal produced a rough bottom surface with partial polymer coverage (grass), while
larger particles produced pillars having the height of the etch depth.
17
100µm
20µm
Table I: Parameters for the cyclic multistep RIE process that were optimized for minimum
required polymer sidewall protection, minimal bottom surface roughness, and the least
mask erosion. The Polymer Deposition, Anisotropic Polymer Removal, and Isotropic Etch
steps are designated (B), (C), and (A), respectively, as described in the text. The two
processes (1) and (2) are tailored for the use of 2 µm of photoresist as a mask.
18
Sub-
cycle
B C A
A step C(smash) etch time of 18 seconds resulted in grass-free and smooth etch fronts,
while maintaining a sufficient selectivity of silicon over photoresist (20:1) for a polymer
deposition step of 131 seconds (step B in Process 1, Table I). Under these conditions,
after the high energy smash step, the approximately 2.4 nm thickness of polymer
remaining on the sidewalls was able to withstand a high-pressure isotropic etch (step A) of
68-seconds duration that produced an etch depth of 100 nm per cycle. In comparison, a
210-sec and 275-nm deep isotropic etch required a 4-minute deposition step for sidewall
wall protection, during which approximately 4.4 nm polymer accumulated, and at least 32
seconds of high-energy etch were then needed to remove the protective polymer from the
bottom etch-front surfaces. These two cyclic process variations (1) and (2) are described in
Table I.
3. Results and Discussion
Before every cycling experiment, the RIE chamber was pre-cleaned in a 200
mTorr oxygen plasma at 500 W RF-power for at least 1 hr during which all surfaces
exposed to ion flux were cleaned of residual polymers. A three-inch, p-type silicon wafer
(5-10 Ohm-cm) was attached to the aluminum platen (powered bottom electrode) with
conductive carbon paint from the backside, and samples of different sizes to be etched
were placed immediately on top of it. The three process steps (A, B, and C) were cycled
with intermediate evacuation and stabilization steps until the target etching depth was
reached. After every process sub-cycle step, the residual reactants were removed from the
chamber by evacuating the reactor for 10 seconds using the turbo-pump without further
care of final pressure. At the beginning of each processing sub-cycle step, 5 to 15 seconds
were required to reach the desired gas flows and pressures, after which they were allowed
to equilibrate for an additional 3 seconds. During cycling experiments, it was found that
the system was not able to reliably ignite a plasma at low pressures below 5 mTorr.
19
Therefore, prior to etching at low pressure, a short ignition step consisting of 30 mTorr at
10 W was added along with a subsequent transient step during which the RF-power and
the pressure were ramped to the step C process values, for instance 100 W and 5 mTorr,
respectively. As a result, the plasma reliably ignited at 30 mTorr and remained active
during the subsequent steps.
It was found that the initial few cycles of the multi step RIE process produced
relatively higher etch rates and offered less sidewall protection than subsequent cycles.
Jagged features emerged near the top of our structures that were caused by the failure of
the protective polymer on the sidewall. Therefore, a preliminary procedure of four cycles
of polymer deposition (B) and smash removal (C) were processed without any isotropic
etching to establish thicker steady state polymer conditions for subsequent full-step cycles.
In addition, the first isotropic etching step (A) was performed using half the duration of the
remaining etching cycles, which produced a less rounded top edge.
After one hundred cycles of Process 1 in Table I (131 seconds deposition, 18
seconds removal, 68 secconds etching), the etch depth had progressed to 10 µm in the
center and 20 µm at the rim of a 3-inch diameter wafer. While the duration of the
isotropic etch (A) was fixed to achieve a certain ripple period (see Fig. 2), the duration of
the isotropic deposition (B) and anisotropic polymer removal (C) steps were fine-tuned to
realize sufficient sidewall protection, with smooth bottom etch fronts, and minimal mask
erosion. When the duration of deposition STEP B was too short, the temporary sidewall
protection failed during the isotropic etch step resulting in smooth, but undercut
structures. When the deposition duration was increased, jagged undercut occurred partially
protecting some parts and leaving gaps between polymer and sidewall. With further
increase in polymer deposition time, these gaps closed and fully protected the sidewall,
which then appeared to be microscopically smooth (≈ 100nm), as shown in Figure 8.
Increasing the deposition time beyond this point only resulted in a thicker sidewall
20
polymer layers and longer periods were required for anisotropic polymer removal. When
the duration of STEP C (polymer removal) was too long, the photoresist mask eroded
faster, while a downward slope of the horizontal etch front in the direction away from the
bottom of the vertical sidewall was formed due to area-dependent etching effects, as
shown in 8a.
21
Figure 8: Optical microscope digital images of cleaved edges taken at a magnification of
1250x showing cross sections after 60 cycles of MS-RIE with varying step times and
powers to demonstrate the process adjustments (see text). The mask was 2 µm of
photoresist. (a): Removal time (STEP C) was too long and a downward slope in the
horizontal direction away from the walls was visible at the bottom surface of the etch
front. (b): Removal time (STEP C) is slightly too short for complete polymer removal from
the etch front, and an upward slope is produced. While the process still works somewhat,
small amounts of deposit become visible far away from any structure walls. (c): The
polymer is not efficiently removed due to a shorter removal duration, and only trench
etching occurs near the sidewalls, but protective polymer remained on the etch-front
away from the sidewall.
22
(a)
(b)
Deposition: 1 minute 10 seconds 120W
Removal: 40 seconds 100W
Etch: 3 minutes 40 seconds 50W
Deposition: 5 minutes 30 seconds 40W
Removal: 52 seconds 100W
Etch: 1 minute and 15 seconds 50W
(c)
Deposition: 5 minutes 30 seconds 40W
Removal: 50 seconds 100W
Etch: 1 minute 15 seconds 50W
Down
Up
5 µm
If sufficient masking material were available, a long anisotropic polymer removal step (C)
would be preferred to ensure a complete removal of all polymer residue from the bottom
surface of the etch front. In contrast, when STEP C was too short, the polymer removal
was incomplete (see Figure 8b), with reduced etching of the front far from the sidewall. In
the case when STEP C was long enough to puncture the bottom polymer coverage but too
short for its complete removal, “grass” became readily visible after a few cycles. In
addition, an upward slope of the etch front in the direction away from all bottom corners
formed, and typically only a few µm around the structures became “trench-etched”, as
shown in Figure 8c. In extreme cases of short removal (C) times, the mask and silicon
substrate become enveloped in a thick layer of polymer after extensive cycling,
suppressing further etching. The duration of STEP C was carefully increased to a value
where no grass was formed while maintaining an etch selectivity of silicon over
photoresist of 20:1.
The optimized parameters for two different cycle etching processes with
different etch rates are given in Table I. With this cyclic multistep etching technique,
combined with selective wet etching, we have fabricated the free-standing silicon
microdisk shown in Figure. 9. Note the smoothness of the vertical walls along the
periphery of the disk with no scallops, demonstrating the utility of this technique. The
50µm diameter microdisk has periodically spaced elements around the periphery (Fig. 9).
4. Conclusion
We have described a novel multistep etching process, which operates on
conventional RIE tools. The dry etching was developed using a combination of anisotropic
and isotropic methods that were optimized to achieve very smooth and nearly straight
sidewalls. This technique was developed for use with conventional RIE systems and does
not require separate magnets for separate plasma confinement, nor specially licensed
commercial techniques. Etched depths over 100 µm were obtained with this technique.
23
(a)
(b)
(c)
(a)
(b)
(c)
Figure 9: Left: Process flow of structure fabrication. (a): After photolithography and MS-
RIE to a depth of 10 m. (b): 40 m porous silicon was electrochemically formed where
highly doped silicon is exposed to the electrolyte. (c): Porous silicon was selectively
oxidized by anodic current in oxidizing liquid and subsequently removed in an HF:C2H5OH
solution. Right: SEM photograph of finished microdisk (50 µm diameter) with periodic
perforations (5 µm wide) around circumference. [15]
ACKNOWLEDGEMENTS
We gratefully acknowledge the support from the Sarnoff Corporation [19]
for the CVD grown layers. This research was supported by DARPA-funded Air Force
Contract No. F19628-00-C-0005 under the Terahertz program, by Air Force Office of
Scientific Research Contract No. F49620-01-1-0042, and by the National Science
Foundation under Grant No. 9815775. We would like to express special thanks to S.
Saddow, J. Suehle, and N Sustersic for useful advice.
24
REFERENCES
[1] F. Laermer, A. Schilp, “Method for anisotropic plasma etching of substrates”, US.
Patent Number 5,498,312, 12 March, 1996
[2] S. K. Ghandi, “VLSI Fabrication Principles”, J. Wiley, New York, 1994
[3] Q. B. Vu, D. A. Stricker, P. M. Zavracky, “Surface Characteristics of (100) Silicon
Anisotropically Etched in Aqueous KOH“, Journal of the Electrochemical Society,
Volume 143, Issue 3, April 1996, pp. 1372
[4] J. G. Smits, “Methods for anisotropic etching of (100) silicon”, US. Patent Issue
5,441,600, 15 August, 1995
[5] A. K. Chu, K. M. Lee, I. J. Lan, “Silicon V grooves fabricated using Ta2O5 etch mask
prepared by room-temperature magnetron sputtering”, Journal of Vacuum Science
and Technology B, Volume 19, Issue 4, July 2001, pp. 1169
[6] M. Paranjape, A. Pandy, S. Brida, L. Landsberger, M. Kahrizi, M. Zen, “Dual-doped
TMAH silicon etchant for microelectromechanical structures and systems
applications”, Journal of Vacuum Science and Technology A, Volume 18, Issue 2,
March 2000, pp. 738
[7] I. L. Berry, A. L. Caviglia, “High resolution patterning of silicon be selective gallium
doping”, Journal of Vacuum Science and Technology B, Volume 1, Issue 4, October
1983, pp. 1059
[8] M. A. Blauw, T. Zijlstra, E. van der Drift, “Balancing the etching and passivation in
time-multiplexed deep dry etching of silicon”, JVSTB 19(6), November 2001, pp.
2930
[9] M. J. Buie, J. T. P. Pender, M. Dahimene, “Characterization of the etch rate non-
uniformity in a magnetically enhanced reactive ion etcher”, Journal of Vacuum
Science and Technology A, Volume 16, Issue 3, May 1998, pp. 1464
[10] M. W. Geis, G. A. Lincoln, N. Efremow, W. J. Piacentini, “A novel anisotropic dry
etching technique”, Journal of Vacuum Science and Technology, Volume 19, Issue 4,
November 1981, pp. 1390
25
[11] T. I. Cox, V. G. I. Deshmukh, “Use of optical emission spectroscopy to study
hexafluoroethane reactive ion beam etching of silicon in the presence of oxygen”,
Applied Physics Letters, Volume 47, Issue 4, 15 August, 1985, pp. 378
[12] Robert Bosch GmbH, US. Patent Numbers 5,498,312 and 5,501,893, and 4241045C1
(Germany), 1994
[13] A. A. Ayón, R. Braff, C. C. Lin, H. H. Sawin, M. A. Schmidt, “Characterization of a
Time Multiplexed Inductively Coupled Plasma Etcher”, Journal of the Electrochemical
Society, Volume 146, Issue 1, January, 1999, pp. 339
[14] Unaxis USA Inc., Division Semiconductors, 10050 16th Street North, St. Petersburg,
Florida 33716, USA
[15] AZ5214E photoresist vendor: Clariant Corporation, 70 Meister Avenue, Somerville,
New Jersey 08876, USA
[16] Adam, T.N., et al. The Design and Fabrication of Microdisk Resonators for
Terahertz Frequency Operation. in IEEE Lester Eastman Conference on High
Performance Devices. 2002. University of Delaware, Newark, DE 19716
[17] M. J. M. Vugts, M. F. A. Eurlings, L. J. F. Hermans, H. C. W. Beijerinck, “Si/XeF2
etching: Reaction layer dynamics and surface roughening”, Journal of Vacuum
Science and Technology A, Volume 14, Issue 5, September 1996, pp. 2780
[18] NR5 photoresist vendor: Futurrex Inc, 12 Cork Hill Road, Franklin, New Jersey
07416, USA
[19] Sarnoff Corporation, 201 Washington Road, Princeton, New Jersey 08543-5300,
USA
26
top related