264 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 11, NO. 3, JUNE 2002
Effect of Process Parameters on the Surface
Morphology and Mechanical Performance of Silicon
Structures After Deep Reactive Ion Etching (DRIE)
Kuo-Shen Chen, Arturo A. Ayón, Senior Member, IEEE, Xin Zhang, Member, IEEE, and S. Mark Spearing
Abstract—The ability to predict and control the influence of
process parameters during silicon etching is vital for the success
of most MEMS devices. In the case of deep reactive ion etching
(DRIE) of silicon substrates, experimental results indicate that
etch performance as well as surface morphology and post-etch
mechanical behavior have a strong dependence on processing
parameters. In order to understand the influence of these param-
eters, a set of experiments was designed and performed to fully
characterize the sensitivity of surface morphology and mechan-
ical behavior of silicon samples produced with different DRIE
operating conditions. The designed experiment involved a matrix
of 55 silicon wafers with radiused hub flexure (RHF) specimens
which were etched 10 min under varying DRIE processing condi-
tions. Data collected by interferometry, atomic force microscopy
(AFM), profilometry, and scanning electron microscopy (SEM),
was used to determine the response of etching performance to
operating conditions. The data collected for fracture strength was
analyzed and modeled by finite element computation. The data
was then fitted to response surfaces to model the dependence
of response variables on dry processing conditions. The results
showed that the achievable anisotropy, etching uniformity, fillet
radii, and surface roughness had a strong dependence on chamber
pressure, applied coil and electrode power, and reactant gases
flow rate. The observed post-etching mechanical behavior for
specimens with high surface roughness always indicated low
fracture strength. For specimens with better surface quality, there
was a wider distribution in sample strength. This suggests that
there are more controlling factors influencing the mechanical
behavior of specimens. Nevertheless, it showed that in order to
achieve high strength, fine surface quality is a necessary requisite.
The mapping of the dependence of response variables on dry
processing conditions produced by this systematic approach
provides additional insight into the plasma phenomena involved
and supplies a practical set of tools to locate and optimize robust
operating conditions. [684]
Index Terms—Deep reactive ion etching (DRIE), fracture
strength, MEMS, plasma etching, silicon, surface morphology.
Manuscript received April 23, 2001; revised September 26, 2001. This work
was supported by the U.S. Army Research Office (ARO) and DARPA under
Contract DAAH04-95-1-0093. This paper was presented in part at the MRS
Fall Meeting, Boston, MA, 1998. Subject Editor H. Fujita.
K.-S. Chen is with the Department of Mechanical Engineering,
National Cheng-Kung University, Tainan, Taiwan 70101 (e-mail:
kschen@mail.ncku.edu.tw).
A. A. Ayón is with Sony Semiconductor, San Antonio, TX 78245 USA.
X. Zhang was with the Microsystems Technology Laboratories, Department
of Electrical Engineering and Computer Science, Massachusetts Institute of
Technology (MIT), Cambridge, MA 02139 USA. She is now with the Depart-
ment of Manufacturing Engineering, Boston University, Brookline, MA 02446
USA.
S. M. Spearing is with the Department of Aeronautics and Astronautics,
Massachusetts Institute of Technology (MIT), Cambridge, MA 02139 USA.
Publisher Item Identifier S 1057-7157(02)04976-4.
I. INTRODUCTION
D
EEP reactive ion etching (DRIE) of silicon enables the
microfabrication of high-aspect ratio structures (HARS),
which, in turn, permit the fabrication of devices able to span
from 100 to 1000
m. HARS also allow the fabrication of struc-
tures that are compliant in the plane of the wafer but rigid in the
direction normal to its surface [1]. Furthermore, HARS in com-
bination with aligned silicon wafer bonding, make possible the
realization of novel and promising applications, such as Power
MEMS [2]. However, most applications place stringent require-
ment on HARS in terms of high silicon etching rate, good se-
lectivity to masking material, profile control and compatibility
with other processes [3].
STS licensed the DRIE technique patented by Robert Bosch
Gmbh [4]. It relies on the deposition of inhibiting films to
obtain anisotropic profiles. This approach utilizes an etching
cycle flowing only SF
[see Fig. 1, steps (ii) and (iv)] and then
switches to a sidewall passivating cycle using only C
F [see
Fig. 1, step (iii) ]. During the subsequent etching cycle, the pas-
sivating film is preferentially removed from the bottom of the
trenches due to ion bombardment, while preventing the etching
of the sidewalls. The alternating of etching and passivating
cycles forms scallops on the sidewalls of etched features. The
peak to valley height of those scallops, being a function of
operating conditions, can be controlled to some extent. Because
of the alternating between etching and passivating cycles, the
term time multiplexed deep etching (TMDE) describes more
closely this technique and it will be used in all subsequent
descriptions. The success of Bosch’s TMDE scheme relies on
the deposition of the inhibiting films to prevent the etching of
the sidewalls.
The Bosch approach uses the high etching rate of flu-
orine-rich plasmas to etch HARS. Some recently reported
applications are already exploiting this last alternative [1],
[5], [6]. Furthermore, by suppressing the time multiplexing,
the equipment can be run with continuous flows of SF
or
C
F . With SF it is possible to achieve isotropic profiles.
With C
F it is possible to deposit teflon-like films that have
been described elsewhere [7]. The large parameter space for a
DRIE etching tool has proven to be versatile enough to allow
prescribing the profile of etched features, uniformity across the
wafer, selectivity to masking material, silicon etching rate as
well as surface roughness.
As mentioned before, power MEMS is one important appli-
cation of HARS. The favorable scaling of the strength of brittle
1057-7157/02$17.00 © 2002 IEEE
Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on July 1, 2009 at 12:11 from IEEE Xplore. Restrictions apply.
CHEN et al.: EFFECT OF PROCESS PARAMETERS ON SILICON STRUCTURES AFTER DRIE 265
Fig. 1. TMDE scheme. (i) Patterned masking material on a silicon wafer.
(ii) Silicon etching cycle. (iii) Fluorocarbon film deposition. (iv) Silicon
etching cycle.
materials offers the potential to create MEMS capable of oper-
ating at high power densities. Such devices could be used for
electrical power generation, propulsion, flow control, pumping
or local cooling. In order to achieve useful power levels, these
machines require characteristic dimensions in the range from
0.1 to 10 mm. Examples under development at MIT include
micro gas turbine engines, micro motor compressor, micro
rocket engines [8], and micro solid-state hydraulic transducers
[9]. The success of these power MEMS relies on achieving high
mechanical strengths and the capability of controlling local
stress levels. At temperatures below about 800 K silicon is a
brittle, elastic material with an extremely high ideal strength
and low fracture toughness [10], [11]. The low toughness
implies a strong sensitivity of strength on processing conditions
and the presence of induced flaws or cracks. The performance
of a DRIE tool was investigated by Ayón et al. [12]. They found
that the etch rate and the anisotropy were strongly influenced
by the etching parameters. By varying these parameters, it is
possible to obtain a suitable recipe for a particular application.
In parallel, the surface morphology and fracture strength of
silicon after DRIE have been studied by Chen et al. [13]. They
investigated the strength of planar biaxial silicon specimens and
silicon specimens with fillets hereby referred to as RHFS [14].
The reference fracture strength of planar silicon specimens was
found to range from 1.2 to 4.6 GPa with a Weibull modulus
ranging from 2.7 to 12. This reported strength and distribution
agreed with some previous investigations at similar scale [15]
and the strength was slightly higher than most found in another
study at a smaller length scale and different surface treatment
[16]. However, the strength of RHFS was only around 1.5 GPa
(with a Weibull modulus of 9) due to surface roughening in
the fillet region. Chen et al. also found that by performing a
short secondary isotropic etch to remove the surface 2–3
m
on the silicon substrate, the fracture strength can be recovered
to around 60–80% of the test strength of biaxial specimens.
Ayón et al. [7], [12] have demonstrated the capability to con-
trol the etching performance and trench profile by adjusting
etching parameters. On the other hand, Chen et al. [13] have
demonstrated that the fracture strength of silicon after DRIE
could be acceptable for many applications. However, two issues
remained unsolved. First, the work performed by Chen et al.
focused on a particular set of operating conditions which was
typically used for micro turbine-engine development. The sen-
sitivity of fracture strength versus a wider range of process pa-
rameters was not investigated. Second, from the MEMS system
design point of view, an optimal process parameter set should
yield both satisfactory results in both microfabrication perfor-
mance and material strength. As a result, it was important to
conduct a systematic investigation to find the sensitivity of sur-
face morphology and mechanical strength on processing condi-
tions.
Experimental studies [7], [12] suggested that the achievable
fillet radii, surface quality, and etch rate are functions of the
etch conditions such as the flow rate of C
F and SF , elec-
trode power, chamber pressure, and etching cycle duration. In
order to optimize the fabrication process design, it is necessary
to build the corresponding database to address the sensitivities
of fabrication performance on etching parameters. To facilitate
the understanding of the observed characteristics, the develop-
ment of such a database via a systematic parametric study of
DRIE parameters is required.
This paper discusses the study of sensitivity of etching per-
formance on processing conditions. This parametric study uti-
lized 55 silicon wafers etched with different etching process
parameters. Each silicon wafer contained 25 RHF specimens
[14]. Etching rate, uniformity, anisotropy, profile control, re-
sulting fillet radii, surface roughness, and fracture strength from
these specimensweremeasured.Finally,data reduction wasper-
formed in order to establish the sensitivity of those variables on
processing parameters.
This paper is organized as follows: the experimental plan
is addressed in detail in Section II. The etching related
performance variables, including etching rate and surface
morphology, are discussed in Section III. Fracture strength
as well as other mechanical characteristics of silicon samples
after DRIE will be covered in Section IV. Section V discusses
some of the applications of this study. Finally, Section VI
summarizes the results and presents the final conclusions.
II. E
XPERIMENTAL APPROACH
The goal of this study was, therefore, to explore the sensi-
tivity of etching performance on processing parameters and
to provide guidance for locating optimal etching conditions
for MEMS fabrication utilizing DRIE. The performance
variables to be measured or tested included silicon etching
rate, anisotropy, uniformity, surface morphology and fracture
strength. In this study, the processing parameters were chosen
systematically and 50 sets of operating conditions were created.
These 50 recipes were then used to etch 55 wafers. The extra
five wafers were etched using operating conditions within the
matrix of 50 in order to evaluate the repeatability of the etches.
Each wafer contained 25 planar biaxial and 25 RHF specimens.
Since the etching rate of silicon in DRIE is around 1–3
m/min,
Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on July 1, 2009 at 12:11 from IEEE Xplore. Restrictions apply.
266 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 11, NO. 3, JUNE 2002
TABLE I
P
ROCESS CONTROLLED PARAMETERS FOR INVESTIGATION
the etching time for each wafer was fixed at 10 min and a
10–30
m each depth was expected. Although the etch depth
was much smaller than that for typical DRIE etched MEMS
structures, the results were useful for building the sensitivity
models to predict the behavior of silicon samples after DRIE.
This work was performed using a Surface Technology Sys-
tems Multiplex ICP (STS Multiplex ICP STS Made in Red-
wood, CA Serial # 041, Vacuum pumping Chamber is done by
Balzers TMH 1000C (Pfeiffer Vacuum Technology), Hudson,
NH) [17]. The equipment included two independent 13.56 MHz
RF power sources: one for the coil around the etching chamber
to create the plasma, and another connected to the wafer elec-
trode to control the RF bias potential of the wafer with respect
to the plasma. Backside helium pressurization was used to pro-
vide sufficient heat transfer between the wafer and the elec-
trode to maintain a constant wafer temperature. A set of eight
alumina fingers clamped the wafer to the electrode. The mea-
sured temperature on the surface of the wafer during processing
was 40
C. This was monitored using temperature sensitive dot
strips mounted on the surface of the wafer. At this temperature,
the etching rate of the photoresist was reduced and allowed its
use as a mask for etching silicon. It was possible to operate
with a predetermined common pressure for both etching and
passivating active cycles, or with a fixed angular position for
the throttle or automatic pressure control (APC) valve. In the
first case the position of the APC valve varied as the gas flow
changed during each cycle. In the latter case, the position of the
throttle valve was fixed and the pressure was determined by the
gas flow rate. The results presented in this paper used this latter
approach. Higher values of the APC valve position in degrees
corresponded to higher chamber pressures. However, the trip
pressure was fixed at 90 mT creating an upper chamber pressure
limit in these experiments. Another variable influenced by the
gas flow rate is the residence time,
, which is proportional to
, where is the chamber pressure, the chamber volume
and
the gas flow rate. This time is important in relation to the
removal rate of etching byproducts in the process chamber with
the corresponding effect on reactant concentration [18], as well
as in the replenishing of etching species. Thus, APC positioning
and gas flow rate influence the etching characteristics. The vari-
ables explored are listed in Table I.
The specimens were prepared in the following fashion:
400 -
m thick, 4-in single crystal silicon 100 wafers, n-type
with resistivity between 6 and 20
-cm, were coated with
(a)
(b)
Fig. 2. (a) The RHF specimen design and (b) final layout of the entire wafer.
photoresist, exposed to a mask that uncovered 18% of the
total wafer area, and developed. The photoresist removal rate
was determined by measuring the photoresist thickness before
and after the etch with an optical interferometer. Additionally
the depth achieved and the uniformity across the wafer were
measured using a DEKTAK profilometer and SEM inspection.
The actual thickness of each wafer was also measured by a
micrometer.
Once the samples were prepared, the designed set of experi-
ments adequate to fit a quadratic model was performed and an-
alyzed using the commercial software package ECHIP (a reg-
istered trademark). The measured data was fitted and the corre-
sponding response surfaces were generated [18]–[20]. The pre-
liminary validation of the model was determined by recording
the adjusted
of each variable. The adjusted is always 1
and it is a measure of the fit of the curve to the data. Higher
values indicate a better data fit. Additional validation criteria
included the comparison of predicted and measured values on
arbitrary operating conditions selected by the user and the com-
parison of predicted and measured values using processing con-
ditions suggested by the model. The large parameter space pre-
sented in this paper obviates the necessity for the utilization of
response surfaces as required tools to assess the combined effect
of modifying operating conditions. Without this aid, the final
Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on July 1, 2009 at 12:11 from IEEE Xplore. Restrictions apply.
CHEN et al.: EFFECT OF PROCESS PARAMETERS ON SILICON STRUCTURES AFTER DRIE 267
Fig. 3. Total depth achieved as a function of SF flow rate (sccm) and applied
coil power (W).
outcome after modifying operating conditions is difficult to pre-
dict.
For the mechanical strength determination, as shown in
Fig. 2(a), the RHF specimen design consists of square dies of
1cm
in size. An annular region with inner and outer radii
of 1 and 2.5 mm was etched. There were 25 RHF specimens
per wafer. Due to the stochastic nature of the fracture strength
of brittle materials, relatively large numbers of specimens are
required to obtain the strength in a statistically acceptable
manner. The schematic layout of the entire wafer is shown in
Fig. 2(b). 15 to 20 RHF specimens per wafer were mechanically
tested to determine the fracture strength. For the rest of the
specimens, two were used for measuring the etching rate, and
the remainder were used for SEM inspection and AFM mea-
surements. The fillet radii at the trench bottom, were defined
as the smallest radius of curvature at the trench fillet region.
This is an important parameter, since it defines the local stress
concentration, the values were estimated by fitting a circle to
SEM images. In some cases, the etch resulted in an elliptical,
not a circular shape. In these situations, the semi-minor axis
was treated as the fillet radius. The fracture load, combined
with measurements of wafer thickness, etched depth, and the
local fillet radius were used to construct finite element models
to convert the fracture loads to fracture strengths [13].
Surface roughness measurements were taken with a Digital
Instruments D3000 AFM using tapping mode. The scanned area
was a square of 10
m per side; multiple readings were taken
from each die in various locations between the hub and the outer
wall. Dies from each wafer were cleaved to provide a cross sec-
tion of the trench profile and these were examined using SEM,
all with the same magnification of 20 000
. Curves were fit to
determine the fillet radius at the etch base. The size of the scal-
lops on the vertical walls were also measured, and averagedover
20 to 30 scallops per die.
III. E
TCHING RATE AND SURFACE MORPHOLOGY
A. Silicon Etching Rate
Fig. 3 shows the etched depth achieved as a function of ap-
plied coil power and SF
flow rate for samples exposed 10 min-
utes to the glow discharge. All measurements graphed in Fig. 3
were taken for trenches of a nominal width of 64
m. This width
was chosen because this size is small enough to avoid loading
effects while it is sufficiently large so as to avoid problems oc-
curring in deep etching due to mass transportation of the etchant
and etching byproducts. These issues are discussed in more de-
tail elsewhere [12].
The adjusted
of the quadratic model for this response was
0.91 indicating that the fit was reasonably good [18]. The sil-
icon etching rate increases with coil power because the ion flux
density increases. Similarly, the etching rate increases with SF
flow rate because of increases in the concentration of etching
species
and because of a reduction of etching products
(SiF
) that redeposit [21]. Although not shown here, silicon
etching rate could also be increased by increasing the applied
electrode power during the etching cycle. This is due to the in-
crease in ion bombardment energy that results in higher etching
rates. As the flow rate increases, the etching rate increases with
both coil and electrode power [22].
The etching rate also increases with increases in the duration
of the etching cycle due to longer exposures to the glow dis-
charge. It also increases when the duration of the passivating
cycle decreases due to thinner fluorocarbon film depositions.
Chamber pressure also has a significant influence on etching
rate, which initially increases with pressure due to higher
con-
centrations [23], but as pressure increases even further, the ion
energy and/or ion flux is reduced and the etching rate drops.
B. Photoresist Etching Rate
Low photoresist removal rate is necessary for a robust oper-
ation. Fig. 4 shows the photoresist etching rate dependence on
applied electrode and coil power with TMDE suppressed. The
photoresist etching rate increases with applied electrode power
because of increases in ion bombardment energy, therefore, in-
creasing ion bombardment improves the etching anisotropy but
reduces the selectivity. During TMDE operation, this response
is also sensitive to pressure and the duration of the etching and
passivating cycles.
Although photoresist etching rate decreases as the pressure
increases, there are several other important implications associ-
ated with large settings of the APC, namely, the sputtering and
redeposition of masking material which promotes the formation
of micro-columns or “grass” [20], the damage of structures (see
Fig. 5) and excessive polymer deposition. These considerations
limit the range of useful settings to no more than 75
. The du-
ration of the active etching cycle determines the exposure time
of the masking material during the etch and therefore the longer
the cycle, the more the photoresist is etched. Similarly, during
Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on July 1, 2009 at 12:11 from IEEE Xplore. Restrictions apply.
268 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 11, NO. 3, JUNE 2002
Fig. 4. Photoresist removal (Å) as a function of electrode power (W) during
the etching cycle and coil power (W). The adjusted
R
for this variable is 0.88,
which indicates a good fit.
Fig. 5. SEM micrographs showing the deleterious effects observed with high
settings of the APC. Results obtained with the APC set at 80
.
the passivating cycle the thickness of the polymerization film
increases with time, thereby decreasing the photoresist removal
rate with increasing passivation cycle time.
C. Uniformity
The variation of etch rate uniformity with SF
flow rate and
pressure is shown in Fig. 6. Plotted values were obtained by
comparing the depth of trenches of nominal width of 64
m
in the middle of the wafer
, with trenches located
30 mm away
and expressed by the equation: uniformity
%.
This variable determines the extent of overetching required
to achieve a prescribed depth across the wafer. Although this
response is influenced by the temperature uniformity across the
Fig. 6. Variation of etching rate uniformity with SF flowrate(sccm)andAPC
valve position in degrees.
wafer, the local rates of plasma density loss and formation, the
exposed area (i.e., the loading effect), and the feature density,
etc., we can significantly simplify this picture by focusing on
plasma formation. Thus, the plasma density being higher at
points closer to the RF power coil or heating source, promotes
local increases in the etching rate. Therefore, for most operating
conditions the etching rate will be higher on the periphery of
the wafer compared to points closer to the center of the wafer.
Uniformity benefits from lower APC valve settings because
the diffusivity varies inversely with the pressure. Thus, as
the APC position is lowered, pressure drops, which results in
increased diffusivity and improved uniformity [24]. Therefore,
higher diffusivities reduce the plasma nonuniformity providing
a better distribution of the ion flux to the wafer. The diffusive
gas transport of the neutral reactants is also increased, reducing
concentration gradients in the gas phase and producing a more
uniform neutral flux on the surface. Similarly, uniformity
benefits from lower SF
flow rates because, for a fixed position
of the APC valve, pressure decreases when the flow rate
decreases.
D. Anisotropy and Profile Control
This response is of importance in every application, and the
ability to tailor the slope of trench walls is one of the more im-
portant characteristics of this deep silicon etching tool. It is fea-
sible toobtain anisotropic profiles [seeFig. 7(a)], positiveslopes
[see Fig. 7(b)] as well as reentrant profiles [see Fig. 7(c)] by
changing etching parameters.
Anisotropy has a strong dependence on several variables such
as coil and electrode power, the duration of the etching and pas-
sivating cycles, and chamber pressure. In general the combina-
tion of ion bombardment in conjunction with the formation and
Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on July 1, 2009 at 12:11 from IEEE Xplore. Restrictions apply.
