Upload
subrat-kumar-sahoo
View
213
Download
0
Embed Size (px)
Citation preview
7/26/2019 1.1621713
1/9
Modeling of fluorine-based high-density plasma etching of anisotropic silicon trenches
with oxygen sidewall passivation
M. A. Blauw, E. van der Drift, G. Marcos, andA. Rhallabi
Citation: Journal of Applied Physics 94, 6311 (2003); doi: 10.1063/1.1621713
View online: http://dx.doi.org/10.1063/1.1621713
View Table of Contents: http://scitation.aip.org/content/aip/journal/jap/94/10?ver=pdfcov
Published by theAIP Publishing
Articles you may be interested in
Atomic-scale cellular model and profile simulation of poly-Si gate etching in high-density chlorine-based plasmas:
Effects of passivation layer formation on evolution of feature profiles
J. Vac. Sci. Technol. B 26, 1425 (2008); 10.1116/1.2958240
Etching of polysilicon in inductively coupled Cl 2 and HBr discharges. III. Photoresist mask faceting, sidewall
deposition, and microtrenching
J. Vac. Sci. Technol. B 20, 1077 (2002); 10.1116/1.1481868
Level set approach to simulation of feature profile evolution in a high-density plasma-etching system
J. Vac. Sci. Technol. B 19, 701 (2001); 10.1116/1.1370174
Analytical modeling of silicon etch process in high density plasma
J. Vac. Sci. Technol. A 17, 2485 (1999); 10.1116/1.581986
Method for angular sputter yield extraction for high-density plasma chemical vapor deposition simulators
J. Vac. Sci. Technol. B 16, 1123 (1998); 10.1116/1.590020
se of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 14.139.196.7 On: Wed, 18
11:52:25
http://scitation.aip.org/search?value1=M.+A.+Blauw&option1=authorhttp://scitation.aip.org/search?value1=E.+van+der+Drift&option1=authorhttp://scitation.aip.org/search?value1=G.+Marcos&option1=authorhttp://scitation.aip.org/search?value1=A.+Rhallabi&option1=authorhttp://scitation.aip.org/content/aip/journal/jap?ver=pdfcovhttp://dx.doi.org/10.1063/1.1621713http://scitation.aip.org/content/aip/journal/jap/94/10?ver=pdfcovhttp://scitation.aip.org/content/aip?ver=pdfcovhttp://scitation.aip.org/content/avs/journal/jvstb/26/4/10.1116/1.2958240?ver=pdfcovhttp://scitation.aip.org/content/avs/journal/jvstb/26/4/10.1116/1.2958240?ver=pdfcovhttp://scitation.aip.org/content/avs/journal/jvstb/20/3/10.1116/1.1481868?ver=pdfcovhttp://scitation.aip.org/content/avs/journal/jvstb/20/3/10.1116/1.1481868?ver=pdfcovhttp://scitation.aip.org/content/avs/journal/jvstb/19/3/10.1116/1.1370174?ver=pdfcovhttp://scitation.aip.org/content/avs/journal/jvsta/17/5/10.1116/1.581986?ver=pdfcovhttp://scitation.aip.org/content/avs/journal/jvstb/16/3/10.1116/1.590020?ver=pdfcovhttp://scitation.aip.org/content/avs/journal/jvstb/16/3/10.1116/1.590020?ver=pdfcovhttp://scitation.aip.org/content/avs/journal/jvsta/17/5/10.1116/1.581986?ver=pdfcovhttp://scitation.aip.org/content/avs/journal/jvstb/19/3/10.1116/1.1370174?ver=pdfcovhttp://scitation.aip.org/content/avs/journal/jvstb/20/3/10.1116/1.1481868?ver=pdfcovhttp://scitation.aip.org/content/avs/journal/jvstb/20/3/10.1116/1.1481868?ver=pdfcovhttp://scitation.aip.org/content/avs/journal/jvstb/26/4/10.1116/1.2958240?ver=pdfcovhttp://scitation.aip.org/content/avs/journal/jvstb/26/4/10.1116/1.2958240?ver=pdfcovhttp://scitation.aip.org/content/aip?ver=pdfcovhttp://scitation.aip.org/content/aip/journal/jap/94/10?ver=pdfcovhttp://dx.doi.org/10.1063/1.1621713http://scitation.aip.org/content/aip/journal/jap?ver=pdfcovhttp://scitation.aip.org/search?value1=A.+Rhallabi&option1=authorhttp://scitation.aip.org/search?value1=G.+Marcos&option1=authorhttp://scitation.aip.org/search?value1=E.+van+der+Drift&option1=authorhttp://scitation.aip.org/search?value1=M.+A.+Blauw&option1=authorhttp://oasc12039.247realmedia.com/RealMedia/ads/click_lx.ads/www.aip.org/pt/adcenter/pdfcover_test/L-37/1329196761/x01/AIP-PT/JAP_ArticleDL_051816/AIP-APL_Photonics_Launch_1640x440_general_PDF_ad.jpg/434f71374e315a556e61414141774c75?xhttp://scitation.aip.org/content/aip/journal/jap?ver=pdfcov7/26/2019 1.1621713
2/9
Modeling of fluorine-based high-density plasma etching of anisotropicsilicon trenches with oxygen sidewall passivation
M. A. Blauwa) and E. van der DriftDelft Institute of Microelectronics and Submicron Technology, Delft University of Technology,Feldmannweg 17, P.O. Box 5053, 2600 GB Delft, The Netherlands
G. Marcos and A. RhallabiLaboratoire des Plasmas et des Couches Minces, Institut des Materiaux Jean Rouxel (IMN), UMR 6502,
CNRS-Universitede Nantes, 2 rue de la Houssiniere, BP 32 229, 44 322 Nantes Cedex 3, France
Received 14 April 2003; accepted 2 September 2003
The kinetics of high aspect ratio, anisotropic silicon etching in a SF6 O2 plasma is investigated with
a combination of Monte Carlo simulations and inductively coupled plasma etching experiments. The
spontaneous reaction of atomic fluorine is dominant at room temperature and Knudsen transport of
the radicals is the only limitation in narrow structures. At low temperaturestypically between 125
and95Coxygen passivation becomes effective and anisotropic profiles are obtained because the
oxygen passivation can only be removed by the directional ion bombardment. The input parameter
settings for the Monte Carlo model are based on measurements with plasma diagnostics.
Simulations show that anisotropy is controlled by the oxygen sidewall passivation which depends on
the oxygen flux, the oxygen adsorption coefficient, and the aspect ratio. The simulated trench
profiles and the aspect ratio dependent etch rate are consistent with the experimental results.
Experimentally the etch rate behavior can be tuned from aspect ratio dependent to aspect ratioindependent by decreasing the ion flux. This effect can be described well by the recently developed
chemically enhanced ion-neutral synergy model. It turns out that aspect ratio independent etching is
obtained if the downwards depletion of fluorine radicals due to Knudsen transport is compensated
by an increase of the available reaction sites. 2003 American Institute of Physics.
DOI: 10.1063/1.1621713
I. INTRODUCTION
Deep anisotropic plasma etching of silicon is an indis-
pensable tool for microfabrication. High etch rates 5
m min1 with high selectivity and good profile control are
achieved in fluorine-based high-density plasmas due to the
independent control of the radical and ion fluxes.1 5 This
way on-chip applications such as microelectromechanical
systems MEMS and passive radio frequency rf compo-
nents coupled to the electronic circuitry become realistic. In
particular, exploiting the third dimension with high aspect
ratio structures, i.e., structures with a high depth to width
ratio, yields highly sensitive inertial sensors based on capac-
ity measurement at low area consumption. However, the sup-
pression of the lateral etch is of utmost importance for de-
vices with high aspect ratio structures.
Oxygen radicals play a crucial role as etch inhibitor in
deep anisotropic silicon etching. The controlled cooling ofthe substrate down to 125 C assures that a solid oxide
layer is formed on the silicon surface. It prevents the reaction
of fluorine radicals with silicon and can only be removed by
the directional ion bombardment. As such, the addition of
oxygen is necessary to obtain anisotropic profiles, because
even at these very low temperatures atomic fluorine by itself
may result in isotropic etching in high-density plasmas.6 The
amount of oxygen has to be adjusted precisely in relation to
the local fluorine concentration to obtain perfectly vertical
sidewalls in high aspect ratio trenches. The role of the oxy-
gen radicals is hard to quantify, especially in deep structures
where surface analysis of the bottom and sidewalls is ex-
tremely difficult.The surface chemistry is a delicate interplay between
ions, oxygen radicals, and fluorine radicals. Many aspects of
the process have already been investigated in detail, such as
the chemical reaction of fluorine radicals with plain silicon
and the etching of silicon with fluorine radicals under Ar ion
bombardment.7,8 Surface analysis with x-ray photoelectron
spectroscopyXPS and Rutherford backscattering spectros-
copyRBS has resulted in a thorough knowledge of surface
processes, however, many questions still exist about the ex-
act interactions of all three species simultaneously.912
In this work Monte Carlo simulations including both the
fluorine and oxygen flux and inductively coupled plasma
ICP etching experiments are combined in order to clarifythe underlying reaction mechanisms. In this unique situation
the profile evolution of high aspect ratio trenches for
fluorine-based plasma etching with oxygen surface passiva-
tion is investigated. In previous work by Marcos et al.silicon
etching is the exclusive result of fluorine radicals and ions,
but the importance of oxygen radicals for lateral etch control
was also demonstrated.13 In Sec. II, the Monte Carlo simu-
lation model is described. Plasma diagnostics are used to
measure the density of fluorine radicals, oxygen radicals, and
ions for a quantitative approach of the simulations. WithaElectronic mail: [email protected]
JOURNAL OF APPLIED PHYSICS VOLUME 94, NUMBER 10 15 NOVEMBER 2003
63110021-8979/2003/94(10)/6311/8/$20.00 2003 American Institute of Physics
se of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 14.139.196.7 On: Wed, 18
11:52:25
7/26/2019 1.1621713
3/9
these data as input for the parameter settings deep aniso-
tropic trench etching is realistically simulated. In particular,
the role of the oxygen radical flux in the profile evolution is
investigated, because the oxygen radicals along with the ions
are mainly responsible for the passivation mechanism. In
Sec. III the experimental etch rate as a function of aspectratio is investigated. Aspect ratio dependent etching ARDE
is observed, which is primarily caused by the depletion of the
fluorine radicals in narrow structures. This is described by
the Knudsen transport model. Aspect ratio independent etch-
ingARIEis obtained by tuning the ion flux to an extremely
low value. To some extent the Knudsen transport model fails
to describe ARIE correctly. To improve this, the chemically
enhanced ion-neutral synergy model has been developed,
which includes the fluorine radical, oxygen radical, and ion
flux. The latter model explains both ARDE and ARIE well.
The results of the plasma diagnostics are also used for an
independent verification of the model parameters.
II. PROFILE EVOLUTION WITH A MONTE CARLOSIMULATION MODEL
A. Monte Carlo simulation model
An etching simulator based on the Monte Carlo method
is used. Details of the model have been described in a pre-
vious article13 and here it is only briefly summarized. A two-
dimensional structure is considered composed of a mask and
a silicon substrate as shown in Fig. 1. It is defined by a
cellular discretization, which describes both geometry and
surface composition. The size of each cell is 1 nm and ac-
cording to the silicon density this corresponds to fifty atoms
per cell. The etching process is treated by the introduction ofinteractions between the incident plasma species fluorine,
oxygen, and ionsand the surface compounds (SiOxFy cells.
The Monte Carlo approach allows the introduction of chemi-
cal and physical mechanisms, each with its probabilistic con-
siderations. Typical examples are neutral adsorption and de-
sorption, spontaneous chemical etching, isotropic and
specular reflection of ions, preferential sputtering, passiva-
tion layer formation, and redeposition of etched species. All
these mechanisms play a role in the profile evolution. 1416
In a SF6 O2 plasma the two main reactive neutrals are
fluorine and oxygen atoms. For these neutrals, isotropic
fluxes are considered above the mask. Their impact is calcu-
lated by virtue of their adsorption probability. In literature
there is much contradiction about the fluorine adsorption
probability which is between 0.001 68 and 0.1 at room
temperature.7,13 Because of this variance the fluorine adsorp-
tion probability was determined independently in this work.
The oxygen adsorption probability is close to unity. In the
model the adsorption process is automatically followed by a
chemisorption mechanism creating a series of SiOxFy cells
on the surface. Only if a silicon cell is completely saturatedwith fluorine, the SiF4 cell is immediately desorbed thus
modeling spontaneous chemical etching. In all other cases
some ion assistance is needed to remove the cell which is
described in the next paragraph.
To include ion-assisted effects, a model of the ion trans-
port through the plasma sheath, which is connected with the
surface module, calculates the angular and energetic distri-
bution function of the incident ions. When an ion strikes the
surface the preferential sputter yield is determined. The latter
depends on the ion energy, the angle of incidence, and the
chemical nature of the impacted surface site. The SiOxFysputter yield Yx ,y is calculated according to
Yx ,yx ,yAEIET 1
as a function of the ion energy EI . For the Si sputter yield
Y0,0 the results of SFx ion beam etching of silicon are used
where the threshold energy ETis 6 eV and the prefactor A is
0.15.17 For Si cells the factor 0,0 is equal to one. For other
SiOxFy cells the factor x ,y is modulated. The most extreme
values are the SiO2 factor of 0.4 and the SiF3 factor of 1.3.
Y0,0 is 0.79 for a typical ion energy of 60 eV. This is one
order of magnitude larger than for Si sputtering with Ar
ions. It incorporates the effect that the yield increases if fluo-
rine is involved in the sputtering process.
Redeposition of sputtered species is studied by consider-
ing their adsorption on the trench sidewalls. Local surface
displacement is thus modeled by the disappearance or reap-
pearance of a SiOxFy cell, when an etching or redeposition
process occurs, respectively.
B. Input parameters from experimental data
The equipment that has been used for the etching experi-
ments is an Alcatel MET ICP reactor and an Alcatel RCE
200 reactor.18 The latter is an electron cyclotron resonance
ECR system in which the ion flux can be tuned by placing
a quartz cylinder around the wafer, so that the direct path
between the ECR source region and the wafer is blocked.This effectively results in a downstream reactor geometry
with an extremely low ion-to-radical flux ratio at the wafer
level, because ion-electron recombination is much faster than
neutralneutral recombination. Ion densities are measured
with a Langmuir probe Scientific Systems Smart Probeand
radical densities are measured with actinometry optical
emission spectroscopy, ISA/Jobin Yvon-Sofie Digitwin
550.19,20 Radical fluxes are calculated assuming a plasma
temperature of 400 K.
The flux of ions, oxygen, and fluorine radicals have to be
known for a correct Monte Carlo simulation. The plasma
conditions for deep anisotropic trench etching in the ICP
FIG. 1. Two-dimensional cellular grid in the Monte Carlo simulation. The
structure is composed of a mask with an aperture and a silicon substrate.
6312 J. Appl. Phys., Vol. 94, No. 10, 15 November 2003 Blauw et al.
se of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 14.139.196.7 On: Wed, 18
11:52:25
7/26/2019 1.1621713
4/9
reactor were as follows: 300 sccm SF6 , 5.5sccm O2 , 3.4 Pa
pressure, 2500 W source power, 25 V bias voltage, and
120 C substrate temperature. A fluorine radical partial
pressure of 1.27 Pa and an oxygen radical partial pressure of
0.032 Pa were determined for a plasma nearly identical to
these plasma conditions. The fluorine radical flux Fis given
by
Fp F
2m FkT, 2
where pF is the fluorine radical partial pressure, m F the fluo-
rine atom mass, k the Boltzmann constant, and T the plasma
temperature. The fluorine radical flux to the substrate is thus
approximately a factor 40 higher than the oxygen radical
flux. In other words the oxygen percentage is 2.5%. In aparticular experiment the flow of oxygen is adjusted to ob-
tain perfect anisotropy. With a Langmuir probe, the ion cur-
rent density was measured. It was 0.39 mA cm2 on the sub-
strate position directly below the source and it was 0.07
mAcm2 on the substrate position 8.0 cm downstream. The
fluorine and oxygen radical flux were constant. This gives
fluorine radical-to-ion flux ratios of 1.48103 8.26103,
depending on the position.
The silicon etch rate was measured without the interfer-
ence of the ion flux at a substrate temperature of 25 C to
investigate the etching kinetics of fluorine radicals. The ver-
tical ion flux is excluded in a structure consisting of horizon-
tal lines in 1.5 m thick silicon on insulator SOI coveredby thermal silicon dioxide. The fluorine radicals reach the
silicon core through a window at one of the sides, and an
empty tube of silicon dioxide remains after etching. The av-
erage etch rate in these tubes has been plotted as a function
of the aspect ratio in Fig. 2 together with a view of the
structure after etching. A least squares fit of the data with the
Knudsen transport model described by Eq. 11 in Sec. III
gives a reaction probability of 0.47 together with an initial
etch rate of 10.6 min1. A fluorine radical partial pressure
of 2.5 Pa was measured with actinometry for the plasma
conditions used in this experiment. The etch rate RE is given
by
REFS
rF, 3
whereSis the reaction probability, rF the average number of
fluorine atoms per etched silicon atom, and the atomic
density of silicon. It is assumed that all silicon is removed in
the form of SiF4( rF4) , because in literature it has been
shown that SiF4 is the major reaction product below
300 C.21
However, above this temperature SiF2 is also animportant reaction product. In that case the etch rate in-
creases for a given fluorine flux and reaction probability,
because 2rF4. An initial etch rate of 10.2 m min1 can
be calculated from the independently measured fluorine radi-
cal partial pressure and the fitted reaction probability of 0.47
using Eqs.2 and 3 with rF4. This is in good agreement
with the fitted initial etch rate, which is derived for a sub-
strate temperature of 25 C.
C. Profile evolution results
For the profile simulations a radical-to-ion flux ratio of
0.950103 was chosen. It is of the same order of magnitude
as for the experiments and already high in comparison with
previous simulations.13 The pressure was 2.8 Pa. The adsorp-
tion probability of fluorine and oxygen was set to 0.5 and
0.7, respectively. The value for fluorine is based on the
Knudsen transport results of horizontal lines given in Sec.
II B. For oxygen it has been shown that the adsorption prob-
ability is near unity on bare silicon.22 For a covered silicon
surface the oxygen adsorption probability is reduced. Corre-
spondingly, an average value of 0.7 takes into account the
reduction of the oxygen adsorption probability in the simu-
lation model. Two sites per silicon atom are available for
oxygen adsorption.
The simulated trench profiles for an increasing oxygenpercentage of 0%, 5%, and 7.5% of the total neutral flux are
given in Figs. 3a, 3b, and 3c, respectively. The adsorp-
tion probability for sputtered species is initially set to 0.5.
The mask width and thickness are 0.50 and 0.25 m, respec-
tively. Intermediate profiles are shown for each 0.50 m etch
step. The depth of the profiles is 2 m except for the trench
in Fig. 3a, because the limits of the simulation grid are
reached in the lateral direction. The simulation shows that
oxygen is necessary for anisotropy. Without oxygen there is a
high lateral etch rate corresponding to purely isotropic
etching.13 The best anisotropy is obtained for an oxygen per-
centage of 5%, because most lateral etching is suppressed.
This percentage is rather high compared to the 2.5% in theexperiments. However, in the simulation the ion flux is
higher leading to more sidewall erosion, and moreover, in the
experiment molecular oxygen plays a role in the surface pas-
sivation. Spikes appear on the trench bottom for 7.5% oxy-
gen. This is similar to the formation of silicon grass due to
micromasking, which is experimentally observed as a result
of over-passivation. Reducing the adsorption probability for
sputtered species to 0.0 improves the trench profile, which is
shown in Fig. 3d for 5% oxygen.
With an adsorption probability for sputtered species of
0.0, the etching of a deep trench was simulated. The final
aspect ratio of the 0.5 m wide trench is 10. In Fig. 4 this
FIG. 2. Average etch rate as a function of the aspect ratio for horizontal
lines. The inset shows the hollow silicon dioxide structure after etching. The
continuous curve is a least squares fit of the Knudsen transport model with
a reaction probability of 0.47 for fluorine. The point at zero aspect ratio is
independently measured on a plain silicon sample.
6313J. Appl. Phys., Vol. 94, No. 10, 15 November 2003 Blauw et al.
se of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 14.139.196.7 On: Wed, 18
11:52:25
7/26/2019 1.1621713
5/9
trench is compared with deep anisotropic trenches etched in
the ICP reactor with the conditions given in Sec. II B. The
etch rate of the pattern consisting of 6.0 m lines on a 12.0
m pitch was measured by in situ laser interferometry. The
results are very similar regarding anisotropy and lateral etch-
ing emphasizing the accuracy of the Monte Carlo simulation.
The quantitative agreement of the simulated and experimen-
tal initial etch rate is very good with 4.62 and 5.20
m min1, respectively. Although, the lateral etch rate is
0.14m min1 for the simulation and 0.03 m min1 for the
experiment. In the simulation the trench bottom becomes
gradually faceted. A decreasing ion flux in the corners of a
trench due to the angular distribution of ions could be the
cause. Angle dependent sputter yield and crystal orientation
dependent etching are not included in the simulation. How-
ever, these effects may play a role in the plasma etching
experiments. The etch rate as a function of aspect ratio is
presented in Fig. 5 both for the Monte Carlo simulation and
the ICP etching experiment.
III. ASPECT RATIO DEPENDENT ETCHING
A. Chemically enhanced ion-neutral synergy model
The etch rate is evaluated with the chemically enhanced
ion-neutral synergy model. This model is obtained by con-
sidering the surface site balance for both fluorine and oxygen
radicals given by
dF
dt FF1OFFFFJF
4a
and
dO
dt OO1OFOJO , 4b
where is the surface site area density, F the surface cov-
erage of fluorine, O the surface coverage of oxygen, F the
fluorine flux, O the oxygen flux, J the ion flux, F the ad-
sorption probability for fluorine, O the adsorption probabil-
FIG. 3. Simulation results for a 0.5 m wide trench with different percent-
ages of oxygen. aWhen no oxygen is added the profile is strongly isotro-
pic; b the passivation is optimal for 5% oxygen and the profile is aniso-tropic; andcmicromasking appears for an increase to 7.5% oxygen; difthe adsorption probability of the sputtered species is reduced from 0.5 to 0.0
the sidewall taper changes from slightly positive to vertical for 5% oxygen.
FIG. 4. Comparison of deep trenches obtained with aICP etching and b
Monte Carlo simulation under comparable conditions. Both results show a
lateral etch in the top part directly below the mask, in the middle part a
straight anisotropic profile and an apex at the bottom of the trench.
FIG. 5. Aspect ratio dependent etching: a the etch rate is measured two
times byin situlaser interferometry during ICP etching; bthe etch rate for
the Monte Carlo simulation. The data points have been fitted with the Knud-
sen transport model continuous curves and the chemically enhanced ion-
neutral synergy model dotted curves.
6314 J. Appl. Phys., Vol. 94, No. 10, 15 November 2003 Blauw et al.
se of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 14.139.196.7 On: Wed, 18
11:52:25
7/26/2019 1.1621713
6/9
ity for oxygen, F the spontaneous reaction rate for fluorine,
Fthe ion-induced desorption coefficient for fluorine, and Othe ion-induced desorption coefficient for oxygen. For quasi-
steady state the derivatives are negligible and the set of linear
equations can be solved for F and O . The etch rate is
obtained by substituting these quantities in
RE1
FF
4
YFJFYOJO
, 5
where is the silicon atom density andYFand YOthe sputter
yield for silicon fluoride and silicon oxide, respectively.
When one silicon atom is sputtered, four surface sites are
cleared (F4YF and O4YO). Equation 5 is very
comparable to ion-induced etching with an etch inhibitor, but
an important difference is the incorporation of the spontane-
ous etch term.2325 The spontaneous reaction term is respon-
sible for chemical etching of silicon with volatile SiF 4 as the
reaction product. In high-density plasmas the ion flux is
much lower than the radical flux. If it is assumed that the
spontaneous reaction is much faster than the ion-induced de-
sorption of fluorine (
F
F
J) the etch rate can be writ-ten as
RE1
FF4OO
YO
O OJ
1 FFF
1 OJOO
. 6
In cryogenic ICP etching the fluorine flux is typically much
higher than the oxygen flux (FFOO). Actinometry
confirms this. It can also be assumed that the spontaneous
reaction is much faster than the supply of fluorine (FFF), because the reaction with silicon is found to be
proportional to the fluorine radical partial pressure.
26
Withthese assumptions Eq. 6 can be simplified to
RE1
FF4OO
OJ1
OJ
OO
. 7
Apart from the factor FF over OO the result is very
similar to the well-known ion-neutral synergy model. How-
ever, the ion-neutral synergy applies to the oxygen passiva-
tion mechanism only. The etch rate is strongly enhanced due
to the spontaneous chemical reaction of fluorine with silicon
and the subsequent desorption of SiF4
, which is expressed
by the factor FFoverOO . With this amplification the
extremely high etch yield in the order of 1000 observed in
cryogenic etching can be understood better.18 The etch rate
can also be expressed as
REFF
41O, 8a
where
O1
1OJ
OO
. 8b
So it can be seen that the etch rate is proportional to the
fluorine flux and the fraction of uncovered surface sites. The
fluorine surface coverage is negligible. When this model is
used for etch rate evaluation in deep trenches the decrease of
the neutral fluxes Fand Owith depth should be taken into
account. For this purpose F and Oare expressed as a con-
stant flux at zero aspect ratio times a factor as a function of
aspect ratio.27 These functions are defined by
ii0i
AR iF,O 9a
and
iAR
KAR
KARi1OKAR
i1O, 9b
where KAR is the Knudsen transport coefficient and i the
adsorption probability of species i. The parameters F0, O
J/O0 , F , and O are unknown. To reduce the number of
known parameters F is set to 0.5 and O is set to 0.7 as in
Sec. II. After substitution of Eq. 9 in Eq. 8 the etch rate
function is transformed to a function with the fit parameters
a and b given byREF
ARa1O, 10a
where
O1
1b
OAR
. 10b
The fit parameters are defined by
aFF
0
410c
and
bOJ
OO0
. 10d
One more step has to be taken before fitting can be per-
formed. The oxygen flux in Eq. 9 and the oxygen surface
coverage in Eq. 10 are interdependent. By substituting Eq.
9b in Eq. 10b a quadratic expression for O is obtained.
This gives a solution for O as a function ofKAR , b and O .
The solution is substituted in Eq. 10a. By varying a and b
in this final expression the least-squares fit is obtained.
B. Modeling of experimental and simulated results
Etch rate results obtained from etching experiments and
Monte Carlo simulations were fitted with both the Knudsen
transport model and the chemically enhanced ion-neutral
synergy model. The Knudsen transport model is given by
REFF
0
4
KAR
KARFKAR
F. 11
The main difference is that the Knudsen transport model
does not take into account the role of oxygen. The etch rate
is proportional to the fluorine flux and the adsorption prob-
ability of fluorine. For the Knudsen transport model the ad-
6315J. Appl. Phys., Vol. 94, No. 10, 15 November 2003 Blauw et al.
se of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 14.139.196.7 On: Wed, 18
11:52:25
7/26/2019 1.1621713
7/9
sorption probability is equal to the reaction probability. In
fitting the Knudsen transport model the fluorine flux at zero
aspect ratio F0 and the adsorption probability F are varied.
In the limit of zero oxygen surface coverage the chemically
enhanced ion-neutral synergy model is identical to the Knud-
sen transport model with an adsorption probability of 0.5.
Deep anisotropic trenches were made with ICP etching.
The etch rate was measured by in situ laser interferometry
during the ICP etching experiments. The results for thetrenches shown in Fig. 4aare plotted as a function of aspect
ratio in Fig. 5a. Corresponding results for the deep trench
Monte Carlo simulation are given in Figs. 4b and 5b.
For the ICP etching experiments the reaction probability
of fluorine at zero aspect ratio for the Knudsen transport
model and the chemically enhanced ion-neutral synergy
model is 0.26 and 0.32, respectively. Except for small differ-
ences the models fit equally well due to measurement inac-
curacies. For the simulation the reaction probability of fluo-
rine at zero aspect ratio is 0.41 and 0.44, respectively. There
are no visible differences between the two models. This is
due to the low oxygen surface coverage of 0.12 in the chemi-
cally enhanced ion-neutral synergy model. In this case theetch rate behavior is mainly governed by Knudsen transport
of fluorine.
With actinometry a fluorine radical partial pressure of
1.27 Pa was measured. An initial etch rate of 3.5 m min1
has been calculated with Eqs. 2 and 3 using the fitted
reaction probability of 0.32. This is lower than the experi-
mental value of 5.2 m min1. The discrepancy can be ex-
plained if it is assumed that the plasma temperature is equal
to the substrate temperature, because in that case the fluorine
flux and the calculated initial etch rate become a factor 1.6
higher. For the simulations the input is known, so that the
reaction probability can be calculated exactly from the etch
rate and the fluorine flux. In this case it is 0.333. This value
is lower than the fitted values of 0.41 and 0.44. Some effects
are not incorporated in the models. For example, sidewall
reactions that are responsible for the small lateral etch reduce
the fluorine flux. The etch rate decreases more quickly as a
function of aspect ratio, which effectively results in a higher
reaction probability in the fitted models.
Another remark concerns the fit parameter b. For the
simulation it is 7.3. The oxygen-to-ion flux ratio is 47.5. In
this case the ion-induced desorption coefficient of oxygen
must be 243 to obtain the right value ofb. However, in the
simulation this coefficient is only about 2.22 taking into ac-
count the sputter yield of the SiOxFy cells so a lot of oxygenmust be removed by other means. For example, the oxidized
cell can also be removed if the cell under it is etched away.
The chance of this is high because of the low oxygen surface
coverage and the high fluorine-to-oxygen flux ratio. The
boiling point of Si2OF6 is 23.3 C compared to 86.0 C
for SiF4 , so this oxyfluoride compound is slighty volatile
and can be removed without interference of ions during cryo-
genic etching.28
C. Aspect ratio independent etching
Plasma etching with a high and low ion flux was carried
out in the ECR reactor, i.e., without and with the quartz
cylinder, respectively. The plasma conditions were
40.0 sccm SF6 , 4.4sccm O2 , 0.32 Pa pressure, 400 W source
power, 13 V bias voltage and 100 C substrate tempera-
ture. The average etch rate for trenches in the range 0.4 10.0
m is given as a function of aspect ratio in Fig. 6. With a
high ion flux the etch rate is high and decreases relatively
fast with aspect ratio, whereas for a low ion flux the etch rate
is nearly constant.
The two data sets are fitted by the Knudsen transport
model using the least-squares method. For each data set aseparate reaction probability has been taken, but the fluorine
flux at zero aspect ratio is the same for both situations. The
best fit is obtained for a reaction probability of 0.42 and 0.17
and an initial etch rate of 1.92 and 0.78 m min1, respec-
tively. The result is plotted in Fig. 6a. The strong decrease
of the etch rate for the high ion flux is modeled well. How-
ever, the Knudsen model for the low ion flux still shows a
decrease for high aspect ratios.
The best fit of the chemically enhanced ion-neutral syn-
ergy model is plotted in Fig. 6b. The reaction probability of
fluorine at zero aspect ratio is 0.39 and 0.14 and the initial
etch rate is 1.79 and 0.62 m min1 for the high and the low
ion flux, respectively. In this case the aspect ratio indepen-dent data are modeled very well, because the decreasing
fluorine flux is compensated by a decreasing oxygen surface
coverage. In a deep trench the oxygen flux also decreases,
but the ion flux is constant, which leads to a larger fraction of
empty surface sites available for reaction with fluorine.
The fit results of the chemically enhanced ion-neutral
synergy model are tested with independent plasma diagnos-
tics. The fit parameterb is 3.55 and 0.39 for the high and the
low ion flux, respectively. The ion current densities measured
by the Langmuir probe are 0.083 and 0.005 mA cm2, re-
spectively. This makes clear that the difference found by the
model is largely in agreement with the change of the ion
FIG. 6. Aspect ratio dependent and independent etching by tuning the ion
flux. The results have been fitted by athe Knudsen transport model and bthe chemically enhanced ion-neutral synergy model.
6316 J. Appl. Phys., Vol. 94, No. 10, 15 November 2003 Blauw et al.
se of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 14.139.196.7 On: Wed, 18
11:52:25
7/26/2019 1.1621713
8/9
current density in the plasma see Eq.10d. A fluorine radi-
cal partial pressure of 0.31 Pa was determined with actinom-
etry. An initial etch rate of 1.05 and 0.38 m min1 has been
calculated with Eqs. 2 and 3 using this independent mea-
surement of the fluorine radical flux and the fitted reaction
probabilities for the high and low ion flux cases, respectively.
Compared to the fitted initial etch rate, it seems that the
calculated initial etch rate is systematically underestimated
with this method for the low temperature process. So someinaccuracy either in the actinometry or in the fitted adsorp-
tion probabilities must be supposed. The plasma temperature
decreases near the cooled substrate, so that the fluorine flux
increases see Eq. 2. If the plasma is locally in thermal
equilibrium the plasma temperature is 173 K 100 C in-
stead of the assumed 400 K. In that case the calculated initial
etch rate is 1.60 and 0.58 m min1, respectively, which
gives a much better agreement with the fitted results.
D. Further discussion
Only ARDE has been observed in ICP etching, but it
remains unclear why tuning of the plasma parameters does
not result in ARIE. In general, the fluorine radical flux deter-
mines the oxygen radical flux that is needed to prevent lat-
eral etching. Tuning the oxygen radical-to-ion flux ratio
causes the switching between ARDE and ARIE. However,
the ion flux and the ion energy are also factors that are im-
portant for profile control. Higher ion flux and ion energy
lead to a more negative taper due to sidewall erosion deeper
in the trench. So the ion flux and ion energy cannot be cho-
sen arbitrarily. Possibly, the oxygen radical-to-ion flux ratio
has to be lower in anisotropic ICP etching due to different
plasma conditions such as pressure and bias voltage. ARIE
has only been observed for the lowest bias voltages in ECRetching. The average angle of incidence is larger for lower
ion energy, so that erosion of the sidewall passivation is rela-
tively stronger. In that case the oxygen radical-to-ion flux
ratio has to be increased to prevent lateral etching, which
favors ARIE. In ICP etching it has not been possible to make
anisotropic profiles with lower bias voltages due to the form-
ing of surface roughness. Ion-neutral collisions in the plasma
sheath, which play a role in ECR etching, also increase the
average ion angle.18
Tuning of the etching process is also possible with the
substrate temperature. At a lower temperature species are
less volatile so a lower oxygen flux is sufficient for sidewall
passivation. However, surface species are also more difficultto remove by ion bombardment deeper in the trench and it is
hard to estimate whether the sidewall taper becomes more
positive or negative. In practice, a more negative taper is
observed for decreasing temperatures while keeping the oxy-
gen flux constant.29 This is explained by the depletion of
oxygen radicals, which is stronger than the depletion of fluo-
rine radicals, so that the sidewall passivation becomes
weaker in deep trenches. This is illustrated by the following
etching simulation with a radical-to-ion flux ratio of 0.1
103 and a fluorine-to-oxygen radical flux ratio of 9. The
fluorine adsorption probability was taken 0.1. The pressure
was 2.8 Pa and the bias voltage was 30 V. The mask aper-
ture was 0.50 m and the total etched depth was 10 m.
Table I gives the fraction of the incident ion, fluorine radical,
and oxygen radical fluxes that reach the bottom of a trench.
Values are given for etched depths between 0.5 and 2.5 m.
In the beginning of the process the passivation is important
with a fraction 0.31 of the oxygen flux impacting the trench
bottom. When the depth increases the plasma surface inter-
actions change. The oxygen radical-to-ion flux ratio becomes
lower. In the same time the depletion of fluorine radicals is
much less pronounced than the depletion of oxygen radicals.
The fluorine-to-oxygen radical flux ratio increases from 28 to
65. The passivation becomes thus weaker for larger depth. In
this simulation the surface is mostly covered by SiFx species
due to the low oxygen radical-to-ion flux ratio. In the begin-
ning this leads to a significant decrease of the simulated etch
rate due to Knudsen transport of fluorine radicals. In this
particular simulation preferential sputtering dominates for
depths larger than 2.5 m due to the very low fluorine
radical-to-ion flux ratio, which results in a smaller decrease
of the simulated etch rate.
IV. CONCLUSIONS
Monte Carlo simulations show that oxygen passivation
is necessary to obtain anisotropic profiles. Depletion of the
fluorine radical flux plays an important role in high aspect
ratio trenches both for the simulations and high-density
plasma etching experiments. The Knudsen transport model
describes the depletion of the fluorine radical flux in narrow
structures during etching. Simulated and experimental results
for comparable plasma conditions agree very well. Switching
from ARDE to ARIE is observed if the ion flux is decreased
in ECR etching. This behavior is explained by the chemically
enhanced ion-neutral synergy model, which takes into ac-count the depletion of both fluorine and oxygen radicals. For
a high ion flux the oxygen surface coverage is low and the
depletion of fluorine radicals dominates. On the other hand,
for a low ion flux the oxygen surface coverage is high, and
the depletion of fluorine radicals is compensated by a de-
crease of the oxygen surface coverage leading to ARIE.
ACKNOWLEDGMENTS
This work is supported by Technology Foundation STW,
the Applied Science Division of NWO Netherlands Organi-
zation for Scientific Research, under Project No. DEL 4577.
TABLE I. The fraction of ions I , fluorine radicals F , and oxygen radicals
O , that reach the bottom of a trench with a 0.50 m wide mask aperture
for depths between 0.5 and 2.5 m. The fluorine-to-oxygen radical flux ratio
increases with depth.
Depth
m I F O
0.5 0.48 0.97 0.31
1.0 0.35 0.62 0.15
1.5 0.42 0.60 0.162.0 0.24 0.31 0.06
2.5 0.28 0.29 0.04
6317J. Appl. Phys., Vol. 94, No. 10, 15 November 2003 Blauw et al.
se of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 14.139.196.7 On: Wed, 18
11:52:25
7/26/2019 1.1621713
9/9
1 K. Tsujimoto, S. Okudaira, and S. Tachi, Jpn. J. Appl. Phys., Part 1 30,
33191991.2 T. Pandhumsoporn, M. Feldbaum, P. Gadgil, M. Puech, and P. Maquin,
Proc. SPIE 2879, 941996.3 M. Francou, J. S. Danel, and L. Peccoud, Sens. Actuators A 4647, 171995.
4 T. Wells, M. M. El-Gomati, and J. Wood, J. Vac. Sci. Technol. B 15, 434
1997.5 S. Aachboun, P. Ranson, C. Hilbert, and M. Boufnichel, J. Vac. Sci. Tech-
nol. A 18, 18482000.6
M. Puech and P. Maquin, Appl. Surf. Sci. 100101, 579 1996.7 D. L. Flamm, V. M. Donnelly, and J. A. Mucha, J. Appl. Phys. 52, 36331981.
8 C. B. Mullins and J. W. Coburn, J. Appl. Phys. 76, 7562 1994.9 G. S. Oehrlein, J. Vac. Sci. Technol. A 11, 341993.
10 G. S. Oehrlein, S. W. Robey, and J. L. Lindstrom, Appl. Phys. Lett. 52,
11701988.11 G. S. Oehrlein, J. Appl. Phys. 59, 3053 1986.12 K. Takahashi, M. Hori, and T. Goto, J. Vac. Sci. Technol. A 14, 2004
1996.13 G. Marcos, A. Rhallabi, and P. Ranson, J. Vac. Sci. Technol. A 21, 1
2003.14 J. A. Levinson, E. S. G. Shaqfeh, M. Balooch, and A. V. Hamza, J. Vac.
Sci. Technol. A 15, 1902 1997.15 V. K. Singh, E. S. G. Shaqfeh, and J. P. McVittie, J. Vac. Sci. Technol. B
10, 10911992.
16 R. J. Hoekstra, M. J. Grapperhaus, and M. J. Kushner, J. Vac. Sci. Tech-
nol. A 15, 1913 1997.17 T. Chevolleau, P. Y. Tessier, C. Cardinaud, and G. Turban, J. Vac. Sci.
Technol. A 15, 2661 1997.18 M. A. Blauw, T. Zijlstra, R. A. Bakker, and E. van der Drift, J. Vac. Sci.
Technol. B 18, 3453 2000.19 V. M. Donnelly, D. L. Flamm, W. C. Dautremont-Smith, and D. J. Werder,
J. Appl. Phys. 55, 242 1984.20 J. P. Booth, O. Joubert, J. Pelletier, and N. Sadeghi, J. Appl. Phys. 69, 618
1991.21 H. F. Winters and J. W. Coburn, Surf. Sci. Rep. 14, 1611992.22 T. Engel, Surf. Sci. Rep. 18, 911993.23 A. D. Bailey III, M. C. M. van de Sanden, J. A. Gregus, and R. A.
Gottscho, J. Vac. Sci. Technol. B 13, 921995.24 R. A. Gottscho, C. W. Jurgensen, and D. J. Vitkavage, J. Vac. Sci. Technol.
B 10, 21331992.25 V. F. Lukichev, Microelectron. Eng. 4142, 423 1998.26 K. Ninomiya, K. Suzuki, S. Nishimatsu, and O. Okada, J. Appl. Phys. 58,
11771985.27 J. W. Coburn and H. F. Winters, Appl. Phys. Lett. 55, 2730 1989.28 R. C. Weast, D. R. Lide, M. J. Astle, and W. H. Beyer, Handbook of
Chemistry and Physics, 70th ed. CRC, Boca Raton, FL, 1989.29 S. Aachboun, P. Ranson, C. Hilbert, and M. Boufnichel, J. Vac. Sci. Tech-
nol. A 18, 1848 2000.
6318 J. Appl. Phys., Vol. 94, No. 10, 15 November 2003 Blauw et al.
se of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 14.139.196.7 On: Wed, 18