- Email: [email protected]
The surface chemistry of the thermal cracking of silane on silicon (111)
390
Surface
THE SURFACE CHEMISTRY OF THE THERMAL SILANE ON SILICON (111) *
Received
14 November
19X3; accepted
for publication
Science 145 (19X4) 390.-406 North-Holland. Amsterdam
CRACKING OF
17 May 1984
The kinetic of the thermal decomposition of silane gas during epitaxial growth on the (111) face of single crystal silicon were studied by modulated molecular beam mass apectrometry over the temperature range of 1000-1440 K and with beam intensities from 2~20’~ to 2 x 10lh molecules/cm*.s. An abrupt change in the apparent reaction probability was observed at - 1130 K. coinciding with a known surface structural transformation at about the same temperature. The molecular beam data for temperatures higher than 1130 K were used for construction of a reaction model. Additional experiments were conducted to fix certain featurea of the reaction mechanism. An attempt was made to measure the rate constant for desorption of unreacted silane molecules from the surface. The residence time of silane on the surface was below the sensitivity limit of the technique ( < 5 ~LS), which means that SiH, desorption is not rate-limiting. Applications of a simultaneous modulated beam of SiH, and a steady beam of SiD, did not product HD molcculca. thus ruling out hydrogen atom production in the surface reaction mechanism. According to the proposed model, silanc molecules which do not reflect from the surface undergo a branched reaction with two types of sites on the surface. producing two bound SiHz molecules in each chemisorption event. Subsequent surface decomposition of the adsorbed SiHz molecules occur with different pm-exponential factors, although with similar activation energies of about 17 kcal/mol for the two types of sites.
1. Introduction Epitaxial reaction surface atoms, second growth
growth
of silicon
via thermal
cracking
of silane
by the overall
SiH,(g) -+ Si + 2H,(g) can be divided into two subprocesses: (1) the decomposition of the silicon-bearing gas to produce adsorbed silicon and (2) the incorporation of silicon adatoms into an epitaxial layer. The aspect has received the most attention in the literature [l-3]; the process is driven by the supersaturation of the surface with silicon
* This work was supported by the Director, Office of Energy Research, Office of Basic Energy Sciences, materials Sciences Division of the US Department of Energy under Contract No. DE-AC03-76SFOO9X.
0039-6028/84/$03.00 Q Elsevier Science Publishers (North-Holland Physics Publishing Division)
B.V.
adatoms but does not depend upon their source (i.e. whether produced from a CVD process or by condensation of silicon vapor). The first subprocess contains
all the surface
the present
chemistry
of the overall process,
and is the subject
of
study.
Henderson
and Helm [4] investigated
conventional
kinetic
experiment
silane decomposition
at silane pressures
on silicon
in a
from 0.02 to 0.15 Torr and
at temperatures from 1070 to 1220 K. The change of pressure with time determined the rate of silane cracking. The silane decomposition rate and the film growth
rate depended
linearly
energy of 20 f 5 kcal/mol. cracking of SiH, to produce
on pressure
and exhibited
Their model of the chemistry gaseous H, and silicon adatoms.
an activation
involved simple The latter either
desorbed to the gas phase or joined the surface crystal structure. In a series of experiments, Joyce and coworkers used the molecular
beam
technique
spec-
trometry beam
to investigate
was used to follow
intensity
and a reaction
involved
an initial
adsorbed
surface H,
Farrow’s
reaction the H,
and substrate
kcal/mol
gaseous
silane
kinetics partial
on Si(ll1)
pressure
temperature.
species.
decomposition
An activation
Subsequent
step
surface
Mass
as a function energy
order near two were determined.
SiH,
[5-71.
which
of silane
of 18.7 f 2
Their reaction
produced
SiH,
model and
H
reactions
of these species
led to
technique
with molecular
beam
and silicon atoms. study
[8] employed
a gas-flow
sampling. The silane decomposition rate depended linearly and exhibited an activation energy of 17 f 2 kcal/mol.
on silane pressure The results were
interpreted with a mechanism similar to that proposed by Henderson and Helm [4]. Joyce et al. [5-71 described the advantages of applying the molecular beam technique to investigate this reaction. This method was chosen for the present study
with
the
measurement not
measured;
additional
feature
of the kinetics the primary
kinetics
and
mechanism
adatoms
from SiH,.
of
beam
of the surface goal
modulation
of the present
of the chemical
This was accomplished
to provide
steps. Substrate study
reactions
direct
growth rates were
was to elucidate
which
by monitoring
produced the H,
the
silicon
product
of
the decomposition. In the modulated
molecular
beam-mass
spectrometric
method,
phase-sensi-
tive detection electronics process the signals according to the shapes and magnitudes of the waveforms and produce the phase angles and amplitudes of the first Fourier components. The phase information between reactant incidence on and product emission
is related to the time lag from the surface and thus
contains valuable kinetic information. This feature constitutes the main advantage of the modulated beam technique over the conventional steady state kinetic methods. The relative amplitude of the product and reactant signals is a measure of the fraction of the incident molecules which react to form a particular product molecule. The apparent reaction probability (c) is the ratio
of the amplitudes of the first Fourier components of the product and scattered reactant signals. The phase lag (+) is the difference in the phase angles of the same Fourier components. These two quantities are determined experimentally as functions of the frequency of modulation (,f), the intensity of the reactant beam (I,), and the temperature of the solid (T). Theoretically, surface reaction models are analyz.ed to provide predictions of c and + as functions of the same three experimental variables. The parameters which characterize the elementary steps of a reaction model include the reactant gas sticking probability, reactive site densities on the surface, and the preexponential factors and activation energies of all temperature-dependent quantities such as rate constants and diffusion coefficients. Values of the model’s parameters are determined by fitting the theoretical calculation to the experimental results. It is customary to examine several theoretical models and choose the one which best fits the data and requires the least number of parameters. The experimental responses often serve as guides to model development. For example, the phase lag usually approaches zero as f- 0 and tends to a limiting value at high modulation frequencies. For simple product desorptioncontrolled kinetics, the limiting phase lag is 90°, while for a series process on the surface, the limit is 180”. If increasing the beam modulation frequency results in a decrease or a maximum in the phase lag, the surface reaction involves a branched process. For a reaction totally controlled by a diffusional step, the phase lag is 45 O, independent of all experimental parameters. These “signatures” of particular elementary steps in complex reaction sequences have been cataloged by a number of authors, along with dicussions of the advantages of the modulate beam method, its limitations, potential improvements and equipment alternatives (9-131. The technique has been applied in quite a few experimental studies of gas-solid reactions, some of which can be found in refs. [14-201. Residence time of silane molecules on the silicon surface has been a source of controversy among the kinetic studies described above. Joyce et al. [5-71 and Henderson and Helm [4] believe that cracking of silane occurs during collision with the crystal surface, whereas in Farrow’s model (81 cracking follows adsorption of silane on the surface. Whenever a silane molecule strikes the silicon surface, it either scatters back to the gas, cracks to release hydrogen, or absorbs with subsequent cracking or desorption. Although desorbed silane cannot be distinguished from scattered silane in a steady state experiment, these two classes of SiH, can be differentiated by virtue of modulation of the incident beam. The part of the silane flux which scatters from the surface does so instantaneously, or with no phase lag ascribable to residence on the surface. On the other hand, silane which physically adsorbs on the surface and then desorbs can be characterized by a mean surface lifetime r, which produces a phase lag 27rfT in this component of the silane flux leaving the silicon surface.
The amplitude
and phase lag of the silane signal observed
by the phase-sensi-
tive detector is the vector sum of the scattered and desorbed portions. Using a mixed beam of silane and an inert gas and measuring the phase difference between
the two as a function
analyzing
SiH,
Another previous
of modulation
adsorption-desorption
aspect
studies
of the mechanism
is the question
frequency
in the presence of SiH,
of whether
cracking hydrogen
provides
a means
of scattering. left unresolved atoms
by the
are produced
on
the surface during reaction. Were this to take place, part of the measured H, emission would arise from surface recombination of hydrogen adatoms. This form of H, can be distinguished on the surface directed
by an isotope
on the surface
comparable
intensity.
from H, mixing
simultaneously If hydrogen
produced
experiment.
directly
with a modulated
atoms
by silane cracking
A steady
are produced
beam beam
of SiD, of SiH,
on the surface
is of
then
there should be a production rate of HD comparable to that of Hz or D2 due to the recombination process. On the other hand, if the decomposition process involved simply splitting should be produced
SiH,
molecules
in the isotope
into Si and two H, molecules,
no HD
mixing experiment.
2. Experimental The source
modulated is quartz
diameter
molecular
beam
tube terminating
hole. The tube contains
apparatus
is shown
in fig. 1. The
in a 1 mm thick membrane semiconductor-grade
beam
with a 0.5 mm
silane gas at pressures
between 0.1 and 1 Torr. The gas flux from the source is collimated by a 0.8 mm diameter orifice into a thin beam which impinges on a solid target. The chamber containing the target is pumped to high vacuum (- 10e8 Torr), so impingement of background gases is. much smaller than the rate at which beam molecules strike the surface. The latter is equivalent to pressures up to 10P4 Torr. The pressure hole and hence The
relationship
of silane in the source tube controls
the intensity between
of the molecular the silane
beam
pressure
the flow rate from the which strikes
in the source
and
the target. the beam
intensity was determined in separate tests in which the target was removed to allow the molecular beam to enter a closed ion gauge tube at the rear of the chamber.
The beam intensity
the pressure
rise measured
at the position
of the target was calculated
by the ion gauge tube for various
from
gas pressures
in
the source tube. The range of SiH, beam intensities begins with a value large enough to distinguish the reaction product signal from noise. The maximum beam intensity is limited by the pumping power of the vacuum system and by corrosive attack by silane on sensitive components in the system. However, the order-of-magnitude accessible range of beam intensity is sufficient to reveal any nonlinearity in the reaction kinetic. Before formation into a molecular beam, the reactant gas effusing from the
source tube passes through the rotating blades of a toothed disk. The range of modulation frequencies is dictated by the mechanical stability of the chopper at low frequencies frequencies. when
and by the limitations
In addition,
the transit
target and detector
become
thereby
masking
characteristic
between
long compared
the measured
the small effects
modulation
frequency
response
times
was prepared
provide source
driver
and target
to the characteristic
due to surface
of the surface
power
a natural
phase lags are dominated
range (20-900
cracking. The target is a chip of a single-crystal material
of chopper
time effects
times of molecules
of the surface processes, accessible
transit
reactions.
at high
upper reaction
silicon produced
by the supplier by deposition
times
by the former, Fortunately,
Hz) was well-matched reactions
limit:
and between
responsible by Epitaxy
the to the
for silane Inc. This
of a 0.6 pm thick undoped
silicon expitaxial layer on a 375 pm thick single crystal silicon wafer with (111) faces. A triangular sample with 17 mm sides parallel to the (110) directions was broken from the wafer using a sharp object. Aside from a slight blow of
ION TITANIUM PUMP +SUBLIMATION / PUMP
Fig. 1. Apparatus
cross-section
(vertical).
nitrogen
gas to clean
condition. layer
facing
the incident
temperature surface. The
Emissivities
the crystal
and heated
by an optical
at the pyrometer
temperature
of the specimen
was used
on top of a tantalum
beam
was measured
maximum
heating
the surface,
The target was placed
(1443
by electron pyrometer
wavelength
K)
limit
Below
annealed
at temperatures
ambient vacuum sample appeared microscopy
pressures as bright
1400
and
target
from ref. [21].
for well-controlled
(principally 1600
K for
carbon),
it
an hour
at
of - lo-* Torr. Following this treatment, the as before and examination by scanning electron
revealed a featureless
which result from vaporization Similar
between
The
on the
- 1000 K, no reaction
was detectable. To verify that the sample was free of contamination was
bombardment.
sighted
were taken
is an upper
without risking melting.
in the as-received
ring with the epitaxial
tests were performed
surface.
The triangular
of a contaminated under deposition
pits or surface
surface
conditions,
[22,23]
lumps
were absent.
1400 K with a SiH,
beam of l-3 x 1015 molecules/cm2 +s, with the same results. The absence of surface morphology following these tests indicated that silicon atoms were added
or removed
samples
from the (111)
were therefore
judged
surface
by a layer-by-layer
to be sufficiently
clean
process.
for chemical
These kinetic
studies. Following the molecular beam experiments, the target surfaces were again examined by SEM; none showed any surface morphology, which is an indication of the adequacy of experimental conditions such as beam purity, background
gas pressure,
and contamination
of the surface
of impurities from the silicon or by surface diffusion supporting the target. In all cases, only two-dimensional, the silicon Near
crystal
the edges
impurities requirement evaporation
occurred
clips held
arising
from
of
featureless electron
of a monolayer.
the support surfaces spectroscopy Surfaces
in the central
the crystal,
is a much more stringent
ble with Auger percent
substrate where
portion
the surface
structure.
It should
following
high
test of surface which which
by bulk diffusion
from the structures featureless growth of of the specimen. was pitted be noted
temperature cleanliness
has a detection
appear
carbon-free
due to that
the
growth
or
than achievalimit
of a few
by AES
can
become pitted at high temperatures due to undetectable impurities. For the isotope mixing tests, provision was made for simultaneously injecting SiD, directly onto the target surface. This was accomplished by a copper doser tube located 1 cm away from the surface. The SiD, impingement rate on the target was calculated from the supply pressure and the flow resistance of the tubing. Chopping of the incident beam induces modulation at the same frequency in the product species desorbed from the target. Following recollimation, both the reflected reactant and desorbed products are detected by a quadrupole mass spectrometer located in the third chamber in fig. 1. The 1.6 mm diameter orifice separating the target chamber from the detection chamber insures that
the sampling spectrometer.
beam does not collide with the walls of the ionizer of the mass Neutrals and mass-analyzed ions travel along the mass filter but
at the exit the selected ions are deflected multiplier.
The neutral species continue
towards the input face of the electron undeviated
and are either deposited
on
the chamber walls or removed by the pumps. This separation significantly reduces the electronic noise which is produced when the neutral species is allowed to enter the electron multiplier input. The
output
of the electron
optical
switch
monitoring
multiplier
chopper
and
rotation
the reference are
fed
signal
from
to a lock-in
the
amplifier
augmented with a two-phase accessory. This combination Fourier-analyzes the signal from the electron multiplier and produces the phases and amplitudes of the various
modes
into which the complete
Only the fundamental The
output
determine
signals
SiH,.
of SiH,
molecules
produced the largest, tive
from
the mass
the relative proportions
reactant
This procedure
SiHi,
of
H,.
The
had
product
was not straightforward
by the 33 V electrons
but the presence
spectrometer
of the reaction
SiH: , SiHi , SiH’,
detection
signal waveform
is decomposed.
mode was used in this study.
of Hl
because
from SiH,
Of these, the SiHi
cracking
potential
of
to
fragmentation
used in the mass spectrometer
S’I+ and Hi.
ionization
to be analyzed
H2 and the scattered
complicates
H,
ionizer signal is quantita-
is higher
than
the
appearance potential of Hi from SiH,, so it is not possible to select an appropriately low ionizing electron energy to avoid cracking of SiH, without rendering
the H,
product
undetectable.
The signals at masses 28, 29, 30, 31 and 32 all exhibited the same phase lags and relative magnitudes at all temperatures, which indicates that they all arise from
the
same
Consequently,
parent
molecule
in
the only volatile product
the
ionizer
(namely,
of the reaction
scattered
is molecular
SiH,).
hydrogen.
This behavior is expected for the Si/H system. Contrary to the silicon-fluorine system, where SiF, is a stable species, SiH, has never identified in the gas phase of the silicon-hydrogen system. Consequently, we rule out the possibility of direct desorption Silicon Therefore
of SiH,
to the gas phase.
has three stable isotopes, “Si (92.2%) ‘“Si (4.7%) and ‘“Si (3.1%). the signal measured by tuning the mass spectrometer to mass 30
consists not only of 28SiHi ions, but as includes 29SiH+ and “Sir. However, knowing that the mass-30 signal represents 42% of the sum of the signals from all fragments of SiH, ionization, the former can be used as a direct measure of the silane content of the gas emanating from the target surface. The mass-2 but as noted above, only partly arising from signal consists solely of Hl, reaction product H,. Below about 1000 K the mass-2/mass-30 signal ratio was constant at a value of a few percent. At higher temperatures, this ratio increased significantly, indicating the onset of silane cracking on the silicon target. The mass-2 signal also exhibited a phase lag with respect to the mass-30 signal above 1000 K, but none at lower temperatures. The task of determining
the portions cracking
of the mass-2
of silane
signal (amplitude
by the surface
and phase)
which were due to
and in the mass spectrometer
was accom-
plished by a computer program which used the signals measured with the target at room temperature as well as at the high temperature of the experiment. This procedure
directly yielded the phase lag of the H, reaction
relative to the incident SiH, reactant. The final piece of data needed for determination probability
was the ratio of the ionization
former is known [24] and the latter was measured removing
the target and installing
of the apparent
cross sections
a nozzle directly
product reaction
of H, to SiH,.
in the apparatus
The
of fig. 1 by
into the mass spectrometer
ionizer. By comparing the mass spectrometer responses to known flow rates of N, and SiH, from this nozzle, the SiH, ionization cross section relative to that of N, was obtained. H, to SiH, probability
the ionization
This quantity
but not the phase lag deduced
Additional available
From this information
was determined. details
of the apparatus
affects
cross section ratio of
only the apparent
reaction
from the mass spectrometer
signals.
and data
interpretation
methods
are
in refs. [17] and [25].
3. Results The controllable variables in a modulated molecular beam experiment are the specimen temperature, modulation frequency and incident reactant beam intensity. An experimental campaign consists of fixing two of these variables and varying the third over its accessible range. For each experiment, the apparent reaction probability and phase lag of the reaction product H, relative to the SiH, reactant were determined. Some of the tests were repeated several times, which permitted
estimates
of the precision
to be made.
Based on these
replications, error bars have been assigned to the data points. The molecular beam data are shown in figs. 2-5 along with theoretical curves which will be discussed
in the next
section.
Two
frequency
scans
were performed,
1243 K (fig. 2) and the other at 1443 K (fig. 3). The behavior with respect
to modulation
frequency
at both temperatures
the existence of a branch process in the mechanism. Fig. 4 shows the lack of influence of the incident the reaction
product
that the elementary eliminates
vector.
one at
of the phase lag strongly
indicates
silane beam intensity
AT 1443 K and 20 Hz, this observation
steps of the model consist only of first order reactions,
coverage-dependence
of the sticking
probability
of SiH,
on
suggests and
on Si.
Gelain et al. [26] also reported a pressure-independent reaction probability for the decomposition of silane on silicon at 1183 K for silane pressures up to 1O-J Torr. The temperature dependence of the reaction product vector is shown in fig. 5. As can be seen from this plot, the apparent reaction probability is almost
I.0
I T = 1243 K I 49xl0~5olec”les/cm’-s
=
-
0.01
Theory
I
IO Modulol~o”
Fig. 2. Theoretical
20-
1
100
and experimental
Freqrency
IOOC?
(Hz\
values of c and $J versus modulation
frequency
at 1243 K.
frequency
at 1443 K.
T=l443K
I:
49x10’5
-
molec”les/cm2-s
Theory
IO-
w .= r” a
05-
:: e a ,!
0.2-
:: lz 5
a-
k z 4 0.05
-
0.02
-
O.OlI
I
II11J’
IO
Fig. 3. Theoretical
1000
100 Modulation
and experimental
Frequency
(Hz)
values of c and $J versus modulation
M. K. Fumuun~, D. R. Oiurtder / Thermul cracking
of s&me on Si(l1 I)
399
independent of temperature above about 1130 K and then decreases sharply at lower temperatures. An analogous change is also apparent in the phase lag. Madix and Schwarz [IS] reported similar abrupt changes in the reaction probability and phase lag at around 1050 K for the reaction of molecular chlorine with the (111) face of single crystal silicon. The chemical reactivity transition temperature of 1130 K in fig. 5 falls within the temperature region from 1130 to 1180 K where the (111) surface undergoes a transformation from an ordered 7 x 7 structure to a featureless 1 X 1 structure [27,28]. Beyond the surface transition temperature the apparent reaction probability is nearly constant and the phase lag is low. This behavior is characteristic of a reaction which is nearly supply-limited, by which is meant that all surface steps following initial sticking of the gaseous reactant are rapid and the emission of reaction product is directly dependent on the rate of reactant supply to the surface. In this limit, the apparent reaction probability is twice the sticking probability (because each interaction of an SiH, molecule with the surface produces two molecutes of H,) and, as seen from fig. 4, independent of
t’““‘40 os-
f = 20Hz T : 1443K -Theory
2 2
‘-
4---
0.2-
n ::
e
a
s
,_ ;
O.I-
B * E :: z
T 4
c
0.05-
0
1
0
d”
I 1
I
_I.
0.02-
o.ol
IO
10'5
IO'6 Beam Intensity
3xlo’6
(molecules/en?-s)
Fig. 4. Theoretical and experimental values of c and + versus beam intensity,
incident beam intensity. Fortunately, the surface reactions are not so fast that the phase lag vanishes. Fig. 5 shows that $I is between 10” and 20 o at 20 Hz and figs. 2 and 3 show that the phase lag rises to - 60” at high modulation frequencies. The constancy of the phase lag in fig. 4 is the true indicator of a mechanism with only linear steps. Therefore, kinetic information is still extractable from the phase lag data even though the high temperature reaction probability data serve only to give the total sticking probability of SiH, on Si(ll1). The attempt to characterize adsorptiondesorption of unreacted silane on the Si(ll1) surface was based on measuring the phase difference between the two species in an argon/silane mixed beam. Sufficiently long holdup of silane due to adsorption followed by slow desorption should cause the silane signal to Temperature
I = 4 9 x lOI f : 20
(Ki
molec”les/cm2.s
Hz
-Theory
I
4 70
8.0 Temperature
Fig. 5. Theoretical
and experimental
90
I
100”
(104/K)
values of c and + versus reciprocal
surface
temperature.
lag behind that of the reflected argon, Fig. 6 shows the results of such a test at two substrate temperatures. At low modulation frequencies, where most of the reaction data were taken, the relative phase lag is zero. At chopping frequencies greater
than 30 Hz, a consistent
lag continues
to decrease
approximately target
- 3’.
was near
negative tematic reactive
phase
until
Similar
ambient, lags
negative
phase lag is observed.
- 300 Hz, beyond behavior
and
occurs
when
with
silane
in fact,
are attributed
which it remains
to experimental
The phase constant
the temperature replaced
by neon.
uncertainties.
at
of the The
This
sys-
phase lag error in the mixed beam tests does not compromise the phase lags shown in figs. 2-5, which typical display error bars of
8- 10 O. The results of the mixed beam tests mean that the adsorption-desorption properties of silane on silicon were not observable. This finding has two interpretations. scatters
Either
or cracks)
SiH,
does not physically
or the physically
time (less than a few microseconds) frequencies
where transit
adsorbed
adsorb
on Si(lll)
(it either
silane has such a short residence
that it is not detectable
phase lags do not dominate
at the modulation
the entire process
(e.g.,
900 Hz). ‘The experiments utilizing simultaneous impingement of modulated and steady SiD, beams were conducted over a wide range of intensities
SiH, from
f<
each source and over the temperature modulated
HD detected
result it can be concluded
interval
in the hydrogen
1143 to 1443 K. In no case was
reaction
that atomic hydrogen
l
products.
From this negative
is not a surface intermediate
a. “,le OO*
00
in
A
0
‘0
-4 -
Oo
,O
G E 2
-64
Bv)_8_
-lO-
Corrected
I443 o
K
l
As-measured
I
1183 K A *
I
IllIll
I
I
I
100 Frequency, Fig.
6. Silane
experiments.
phase lag with respect Corrections
to argon
for mass-dependent
versus transit
I
III1
1000 Hz frequency effects
in mixed
are shown.
beam
high temperature
silane cracking, for if it were, it would have produced of the isotope mixing.
HD under the conditions
4. Discussion Construction
of a model of the surface
reaction
is based on the features
of
the experimental results discussed in the preceding section. Specifically, the mechanism appears to contain parallel pathways for production of the same product (Hz), all steps are linear, slow desorption of physically adsorbed silane is not important, cance.
Because
and hydrogen of the paucity
tures below the surface The following
atoms
are not surface
and poor quality
transition,
model conforms
species
of any signifi-
of the data taken at tempera-
the model applies only for T > 1130 K. to these restrictions:
Scattered
2 2 Si + 2 H,(g)
A silane surface
2 Si + 2 H,(g)
1 s,te 2 -+
2WH2
molecule
striking
the surface
is either
silicon atom to form two adsorbed
reflected
SiH,
can take place on either one of two types of surface Fig.
5 suggests
temperature, cracking
probabilities
into SiH,
different
are thermally
in atomic
takes place with different
structure
sites, denoted
activated. that
rate constants,
with a
The latter process as 1 and 2.
7, and n2 are independent
which implies that neither the surface densities
of SiH,
sufficiently SiH,
that the sticking
or combines
molecules.
Sites
subsequent denoted
of
of the sites nor the 1 and 2 must be decomposition
by k, and k,,
of
respec-
tively. This kinetic distinction might arise from differences in the geometrical arrangements of lattice atoms which are nearest neighbors of the SiH, molecule in the two types of sites. This could affect properties such as the Si-H vibrational frequency [29], which in turn could influence the SiH, decomposition
rate constant.
The
two types
of sites in the reaction
model
might
be
identified with kinks and ledges, respectively, on the unreconstructed (111) face of the silicon crystal during the two-dimensional layer-by-layer epitaxial growth process. Kinks and ledges have one of the attributes of the two reactive sites postulated in the model, namely, that of temperature-independent surface densities. Neglecting mass balance dn dt = 24,&g(t)
coverage-dependence equations -k,n,
of the sticking
for the model can be written dm x = 2nz4,g(t)
- k,m,
probabilities, as:
the surface
M. K. Furwam,
D. R. Olmder / Thermd trucking of dune
on Si(ll I)
403
where n and m are the surface number densities of the SiH, intermediates on sites 1 and 2, respectively, and Z, is the amplitude of the incident SiH, beam. The gating function of the modulated beam is represented by g(t), which is approximately a square wave of period l/f. The time dependence of n, m, and g are periodic, and Fourier expansion up to the first mode yields: n(t) m(t) g(t)
(24 (2b)
= n, + El ei2+, =
m.
+
Ci, ei2=“,
+(l + g,
=
ei2nf’),
where g, = 4/a for an incident square wave. The steady state densities no and are real whereas the quantities ii, and G,, representing the first Fourier components, are complex. The reaction product vector is defined as the ratio of the fundamental modes of the rate of production of H, to the rate of SiH, impingement: m,
c emi+ = (kin,
+ k2m,)/+g,Z0.
(3)
Eqs. (2) are substituted into eqs. (1) and the solutions for %, and Ti, are inserted into eq. (3). This procedure yields the following formulas for e and 9: E = 2(x*
+JJ’)~‘*,
+ = tan-‘(y/x),
(4)
where x=-+
771 1 +a:
7)2
1 +a:’
Y=
91%
1 + a,2
+
%a*
1 +a;’
(5)
and a, =: 2nf/k,,
a2 = 2rrf/k,.
(6)
In addition to the rate constants, the E and C#J formulas contain the modulation frequency. The beam intensity does not appear because the system of equations is linear. The temperature is implicit in the Arrhenius forms of the rate constants. Data fitting was performed by using a computer program which minimizes the sum of normalized squares of the errors in + and the logarithm of C. The fitting procedure resulted in the following set of parameters: TJi = 0.038,
72 = 0.068,
k, =: 5.1 x lo4 exp( - 16.3/RT),
s-l,
k, = 4.6 x lo6 exp( - 17.8/RT),
s-l,
where the activation energies are in kcal/mol and R is the gas constant. The theoretical curves for E and + based on these model parameter values are shown in figs. 2-5. Agreement with the data is satisfactory except for tempera-
tures below - 1200 K, which, however, is close to the temperature range of the surface structural transformation. Three other sets of reaction parameters were also found which fit the data in a visual sense nearly as well as the set given above. However, these alternate reaction parameters exhibited poorer least squares agreement with the data than the best-fit set. The rate constants for these sets of parameters are shown as lines A, B, and C in fig. 7 along with the rate constants of the best fit. Although the magnitudes of the rate constants in each group are within a factor four of each other, their preexponential factors and activation energies are dratically different. The insensitivity of the experimental reaction probabilities to temperature (fig. 5) leaves only the frequency responses of + at two temperatures (figs. 2 and 3) from which the temperature behavior can be inferred. This results in the substantial variances in preexponential factors and activation energies even though the absolute magnitudes of k, and k, are in close accord. Despite this shortcoming, the activation energies are in good agreement with those determined in earlier studies, which ranged from 17 to 20 kcal/mol [4-81. The total silane sticking probability of - 0.1 is also consistent with previous work
Temperature
7.0
Fig.
7. Rest-fit
and
8.0 Temperature runner-up
SiH,
(K)
9.0 (104/K)
decomposition
rate constants.
405
5. Conclusions Modulated results
molecular
interpreted
beam
experiments
have
in terms of a mechanistic
been
model
conducted
with the (111) face of single crystal silicon in the temperature 1443
K.
Isotope
mixing
and
producing
negative
results,
molecules
which do not reflect
silane
aided in model development.
with two types of sites on the surface. bound
SiH,
molecules
which
tests, a branched
Each chemisorption
decompose
into
although
In the model,
undergo
the
of silane
range of 1130 to
adsorption-desorption
from the surface
and
of the reaction
silane
reaction
event produces
two
with differen
rate
Si and H,
constants. Molecular beam sufficiently precise decrease
in reactivity
resulted about
data obtained at temperatures to warrant reaction modeling.
from
surface
below 1130 K were not However, the dramatic
compared
to the high
temperature
structural
transformation
which
reaction
is known
probably to occur
at
the same temperature.
Acknowledgements Both authors and
would like to thank Dr. B.A. Joyce
for his helpful
achieving Office
suggestions
contaminant-free
of Energy
Division
surfaces.
Research,
of the US
for his interest in this work
for overcoming
the persistent
This work was supported
Office of Basic Energy
Department
of Energy
Sciences,
under
problem
of
by the Director,
Materials
contract
No.
Sciences
DE-AC03-
76SF00098.
References [I]
H.C. Abbink,
[2] M. Shimbo,
R.M.
Broudy
J. Nishizawa
[3] E.I. Girargizov,
J. Crystal
[4] R.C. Henderson
and R.F.
(51 B.A. Joyce
and R.R.
[6] C.R. Booker
Farrow,
[9] R.H. Jones,
Growth
Bradley,
23 (1974)
301.
Phil. Mag. 15 (1967) Sot. 121 (1974)
310.
289. 1167.
899.
W.J. Siekhaus and J.A. Schwartz,
J. Vacuum
1429. [lo]
C.T. Foxon,
[ll]
J.A. Schwartz
[12]
R.L.
Palmer
M.R.
Foxon
and B.A. Joyce, Surface
J. CoIIoid Interface
and B.A. Joyce,
Surface
Surface
Sci. 44 (1974)
Sci. 46 (1974)
and J.N. Smith, Jr., Catalysis
1131 D.R. Olander, [14] CT.
Boudry
and R.J. Madix,
4673.
267.
194.
Phil. Mag. 14 (1966)
J. Electrochem.
Growth
Sci. 30 (1972)
Phil. Mag. 14 (1966)
Bradley,
D.R. Olander,
J. Appl. Phys. 39 (1968)
J. Crystal
52 (1981)
Helm, Surface
and B.A. Joyce,
[7] B.A. Joyce and R.R. [8] R.F.C.
and G.P. McCarthy,
and T. Terasaki,
317.
Rev. 12 (1975)
Sci. 58 (1977) Sci. 50 (1975)
169. 434.
279.
69.
Sci. Technol.
9 (1972)
[15]
S.L. Bernasek
[16]
R.J.
and G. A. Somorjai.
Madix.
in: Physical
J. Chem.
Chemistry
of Fast
Phys. 62 (1975) Reactions,
3149.
Vol. 2. Ed. D.O.
Hayward
(Plenum.
1975). [17] A.J.
Machiels
(181 R.J.
Madix
[19]
and D.R.
and J.A.
M. Balooch
and D.R.
[20] G.E.
Gdowski.
[21]
Allen,
F.G.
[22]
Private
R.C:Henderson,
[24]
L.J.
Kieffer,
Discharge M.K.
[26]
Lasers,
Marcus
JILA
J. Electrochem.
A. Cassuto
130 (1983)
Report
of Epitaxial
Cross Silicon
and W.D.
Robertson
Surface
Sci. 1 (1964)
G. Lucovsky,
J. Vacuum
1208.
Data
for Modeling
Gas
1973.
Deposition
by Thermal
Cracking
(1983).
and P. LeGoff,
Florio
Section
13, September
Abstract
of Bulletin
23.
Lander,
541.
J. Appl. Phys. 42 (1971)
Collision
Center
Chemistry
[27] J.V.
151.
Sci. 127 (1983)
1978. Polito,
of Electron
LBL-15448
Sot.
Surface
325.
264.
1510.
and W.J.
[28] J.J. [29]
Madix.
Information
The Surface
PhD Thesis,
C. Gelain,
Sci. 65 (1977)
Sci. 24 (1971)
from B.A. Joyce,
R.B.
A Compilation
Farnaam,
(1968)
Olander,
Fair and R.J.
communication
of Silane,
Surface
Surface
J. Appl. Phys. 28 (1957)
[23]
[25]
J.A.
Olander,
Schwartz,
Surface
Sci. 22 (1970)
125.
Sci. Technol.
16 (1979)
1225.
459.
of French
Ceramic
Society
80
Surface
THE SURFACE CHEMISTRY OF THE THERMAL SILANE ON SILICON (111) *
Received
14 November
19X3; accepted
for publication
Science 145 (19X4) 390.-406 North-Holland. Amsterdam
CRACKING OF
17 May 1984
The kinetic of the thermal decomposition of silane gas during epitaxial growth on the (111) face of single crystal silicon were studied by modulated molecular beam mass apectrometry over the temperature range of 1000-1440 K and with beam intensities from 2~20’~ to 2 x 10lh molecules/cm*.s. An abrupt change in the apparent reaction probability was observed at - 1130 K. coinciding with a known surface structural transformation at about the same temperature. The molecular beam data for temperatures higher than 1130 K were used for construction of a reaction model. Additional experiments were conducted to fix certain featurea of the reaction mechanism. An attempt was made to measure the rate constant for desorption of unreacted silane molecules from the surface. The residence time of silane on the surface was below the sensitivity limit of the technique ( < 5 ~LS), which means that SiH, desorption is not rate-limiting. Applications of a simultaneous modulated beam of SiH, and a steady beam of SiD, did not product HD molcculca. thus ruling out hydrogen atom production in the surface reaction mechanism. According to the proposed model, silanc molecules which do not reflect from the surface undergo a branched reaction with two types of sites on the surface. producing two bound SiHz molecules in each chemisorption event. Subsequent surface decomposition of the adsorbed SiHz molecules occur with different pm-exponential factors, although with similar activation energies of about 17 kcal/mol for the two types of sites.
1. Introduction Epitaxial reaction surface atoms, second growth
growth
of silicon
via thermal
cracking
of silane
by the overall
SiH,(g) -+ Si + 2H,(g) can be divided into two subprocesses: (1) the decomposition of the silicon-bearing gas to produce adsorbed silicon and (2) the incorporation of silicon adatoms into an epitaxial layer. The aspect has received the most attention in the literature [l-3]; the process is driven by the supersaturation of the surface with silicon
* This work was supported by the Director, Office of Energy Research, Office of Basic Energy Sciences, materials Sciences Division of the US Department of Energy under Contract No. DE-AC03-76SFOO9X.
0039-6028/84/$03.00 Q Elsevier Science Publishers (North-Holland Physics Publishing Division)
B.V.
adatoms but does not depend upon their source (i.e. whether produced from a CVD process or by condensation of silicon vapor). The first subprocess contains
all the surface
the present
chemistry
of the overall process,
and is the subject
of
study.
Henderson
and Helm [4] investigated
conventional
kinetic
experiment
silane decomposition
at silane pressures
on silicon
in a
from 0.02 to 0.15 Torr and
at temperatures from 1070 to 1220 K. The change of pressure with time determined the rate of silane cracking. The silane decomposition rate and the film growth
rate depended
linearly
energy of 20 f 5 kcal/mol. cracking of SiH, to produce
on pressure
and exhibited
Their model of the chemistry gaseous H, and silicon adatoms.
an activation
involved simple The latter either
desorbed to the gas phase or joined the surface crystal structure. In a series of experiments, Joyce and coworkers used the molecular
beam
technique
spec-
trometry beam
to investigate
was used to follow
intensity
and a reaction
involved
an initial
adsorbed
surface H,
Farrow’s
reaction the H,
and substrate
kcal/mol
gaseous
silane
kinetics partial
on Si(ll1)
pressure
temperature.
species.
decomposition
An activation
Subsequent
step
surface
Mass
as a function energy
order near two were determined.
SiH,
[5-71.
which
of silane
of 18.7 f 2
Their reaction
produced
SiH,
model and
H
reactions
of these species
led to
technique
with molecular
beam
and silicon atoms. study
[8] employed
a gas-flow
sampling. The silane decomposition rate depended linearly and exhibited an activation energy of 17 f 2 kcal/mol.
on silane pressure The results were
interpreted with a mechanism similar to that proposed by Henderson and Helm [4]. Joyce et al. [5-71 described the advantages of applying the molecular beam technique to investigate this reaction. This method was chosen for the present study
with
the
measurement not
measured;
additional
feature
of the kinetics the primary
kinetics
and
mechanism
adatoms
from SiH,.
of
beam
of the surface goal
modulation
of the present
of the chemical
This was accomplished
to provide
steps. Substrate study
reactions
direct
growth rates were
was to elucidate
which
by monitoring
produced the H,
the
silicon
product
of
the decomposition. In the modulated
molecular
beam-mass
spectrometric
method,
phase-sensi-
tive detection electronics process the signals according to the shapes and magnitudes of the waveforms and produce the phase angles and amplitudes of the first Fourier components. The phase information between reactant incidence on and product emission
is related to the time lag from the surface and thus
contains valuable kinetic information. This feature constitutes the main advantage of the modulated beam technique over the conventional steady state kinetic methods. The relative amplitude of the product and reactant signals is a measure of the fraction of the incident molecules which react to form a particular product molecule. The apparent reaction probability (c) is the ratio
of the amplitudes of the first Fourier components of the product and scattered reactant signals. The phase lag (+) is the difference in the phase angles of the same Fourier components. These two quantities are determined experimentally as functions of the frequency of modulation (,f), the intensity of the reactant beam (I,), and the temperature of the solid (T). Theoretically, surface reaction models are analyz.ed to provide predictions of c and + as functions of the same three experimental variables. The parameters which characterize the elementary steps of a reaction model include the reactant gas sticking probability, reactive site densities on the surface, and the preexponential factors and activation energies of all temperature-dependent quantities such as rate constants and diffusion coefficients. Values of the model’s parameters are determined by fitting the theoretical calculation to the experimental results. It is customary to examine several theoretical models and choose the one which best fits the data and requires the least number of parameters. The experimental responses often serve as guides to model development. For example, the phase lag usually approaches zero as f- 0 and tends to a limiting value at high modulation frequencies. For simple product desorptioncontrolled kinetics, the limiting phase lag is 90°, while for a series process on the surface, the limit is 180”. If increasing the beam modulation frequency results in a decrease or a maximum in the phase lag, the surface reaction involves a branched process. For a reaction totally controlled by a diffusional step, the phase lag is 45 O, independent of all experimental parameters. These “signatures” of particular elementary steps in complex reaction sequences have been cataloged by a number of authors, along with dicussions of the advantages of the modulate beam method, its limitations, potential improvements and equipment alternatives (9-131. The technique has been applied in quite a few experimental studies of gas-solid reactions, some of which can be found in refs. [14-201. Residence time of silane molecules on the silicon surface has been a source of controversy among the kinetic studies described above. Joyce et al. [5-71 and Henderson and Helm [4] believe that cracking of silane occurs during collision with the crystal surface, whereas in Farrow’s model (81 cracking follows adsorption of silane on the surface. Whenever a silane molecule strikes the silicon surface, it either scatters back to the gas, cracks to release hydrogen, or absorbs with subsequent cracking or desorption. Although desorbed silane cannot be distinguished from scattered silane in a steady state experiment, these two classes of SiH, can be differentiated by virtue of modulation of the incident beam. The part of the silane flux which scatters from the surface does so instantaneously, or with no phase lag ascribable to residence on the surface. On the other hand, silane which physically adsorbs on the surface and then desorbs can be characterized by a mean surface lifetime r, which produces a phase lag 27rfT in this component of the silane flux leaving the silicon surface.
The amplitude
and phase lag of the silane signal observed
by the phase-sensi-
tive detector is the vector sum of the scattered and desorbed portions. Using a mixed beam of silane and an inert gas and measuring the phase difference between
the two as a function
analyzing
SiH,
Another previous
of modulation
adsorption-desorption
aspect
studies
of the mechanism
is the question
frequency
in the presence of SiH,
of whether
cracking hydrogen
provides
a means
of scattering. left unresolved atoms
by the
are produced
on
the surface during reaction. Were this to take place, part of the measured H, emission would arise from surface recombination of hydrogen adatoms. This form of H, can be distinguished on the surface directed
by an isotope
on the surface
comparable
intensity.
from H, mixing
simultaneously If hydrogen
produced
experiment.
directly
with a modulated
atoms
by silane cracking
A steady
are produced
beam beam
of SiD, of SiH,
on the surface
is of
then
there should be a production rate of HD comparable to that of Hz or D2 due to the recombination process. On the other hand, if the decomposition process involved simply splitting should be produced
SiH,
molecules
in the isotope
into Si and two H, molecules,
no HD
mixing experiment.
2. Experimental The source
modulated is quartz
diameter
molecular
beam
tube terminating
hole. The tube contains
apparatus
is shown
in fig. 1. The
in a 1 mm thick membrane semiconductor-grade
beam
with a 0.5 mm
silane gas at pressures
between 0.1 and 1 Torr. The gas flux from the source is collimated by a 0.8 mm diameter orifice into a thin beam which impinges on a solid target. The chamber containing the target is pumped to high vacuum (- 10e8 Torr), so impingement of background gases is. much smaller than the rate at which beam molecules strike the surface. The latter is equivalent to pressures up to 10P4 Torr. The pressure hole and hence The
relationship
of silane in the source tube controls
the intensity between
of the molecular the silane
beam
pressure
the flow rate from the which strikes
in the source
and
the target. the beam
intensity was determined in separate tests in which the target was removed to allow the molecular beam to enter a closed ion gauge tube at the rear of the chamber.
The beam intensity
the pressure
rise measured
at the position
of the target was calculated
by the ion gauge tube for various
from
gas pressures
in
the source tube. The range of SiH, beam intensities begins with a value large enough to distinguish the reaction product signal from noise. The maximum beam intensity is limited by the pumping power of the vacuum system and by corrosive attack by silane on sensitive components in the system. However, the order-of-magnitude accessible range of beam intensity is sufficient to reveal any nonlinearity in the reaction kinetic. Before formation into a molecular beam, the reactant gas effusing from the
source tube passes through the rotating blades of a toothed disk. The range of modulation frequencies is dictated by the mechanical stability of the chopper at low frequencies frequencies. when
and by the limitations
In addition,
the transit
target and detector
become
thereby
masking
characteristic
between
long compared
the measured
the small effects
modulation
frequency
response
times
was prepared
provide source
driver
and target
to the characteristic
due to surface
of the surface
power
a natural
phase lags are dominated
range (20-900
cracking. The target is a chip of a single-crystal material
of chopper
time effects
times of molecules
of the surface processes, accessible
transit
reactions.
at high
upper reaction
silicon produced
by the supplier by deposition
times
by the former, Fortunately,
Hz) was well-matched reactions
limit:
and between
responsible by Epitaxy
the to the
for silane Inc. This
of a 0.6 pm thick undoped
silicon expitaxial layer on a 375 pm thick single crystal silicon wafer with (111) faces. A triangular sample with 17 mm sides parallel to the (110) directions was broken from the wafer using a sharp object. Aside from a slight blow of
ION TITANIUM PUMP +SUBLIMATION / PUMP
Fig. 1. Apparatus
cross-section
(vertical).
nitrogen
gas to clean
condition. layer
facing
the incident
temperature surface. The
Emissivities
the crystal
and heated
by an optical
at the pyrometer
temperature
of the specimen
was used
on top of a tantalum
beam
was measured
maximum
heating
the surface,
The target was placed
(1443
by electron pyrometer
wavelength
K)
limit
Below
annealed
at temperatures
ambient vacuum sample appeared microscopy
pressures as bright
1400
and
target
from ref. [21].
for well-controlled
(principally 1600
K for
carbon),
it
an hour
at
of - lo-* Torr. Following this treatment, the as before and examination by scanning electron
revealed a featureless
which result from vaporization Similar
between
The
on the
- 1000 K, no reaction
was detectable. To verify that the sample was free of contamination was
bombardment.
sighted
were taken
is an upper
without risking melting.
in the as-received
ring with the epitaxial
tests were performed
surface.
The triangular
of a contaminated under deposition
pits or surface
surface
conditions,
[22,23]
lumps
were absent.
1400 K with a SiH,
beam of l-3 x 1015 molecules/cm2 +s, with the same results. The absence of surface morphology following these tests indicated that silicon atoms were added
or removed
samples
from the (111)
were therefore
judged
surface
by a layer-by-layer
to be sufficiently
clean
process.
for chemical
These kinetic
studies. Following the molecular beam experiments, the target surfaces were again examined by SEM; none showed any surface morphology, which is an indication of the adequacy of experimental conditions such as beam purity, background
gas pressure,
and contamination
of the surface
of impurities from the silicon or by surface diffusion supporting the target. In all cases, only two-dimensional, the silicon Near
crystal
the edges
impurities requirement evaporation
occurred
clips held
arising
from
of
featureless electron
of a monolayer.
the support surfaces spectroscopy Surfaces
in the central
the crystal,
is a much more stringent
ble with Auger percent
substrate where
portion
the surface
structure.
It should
following
high
test of surface which which
by bulk diffusion
from the structures featureless growth of of the specimen. was pitted be noted
temperature cleanliness
has a detection
appear
carbon-free
due to that
the
growth
or
than achievalimit
of a few
by AES
can
become pitted at high temperatures due to undetectable impurities. For the isotope mixing tests, provision was made for simultaneously injecting SiD, directly onto the target surface. This was accomplished by a copper doser tube located 1 cm away from the surface. The SiD, impingement rate on the target was calculated from the supply pressure and the flow resistance of the tubing. Chopping of the incident beam induces modulation at the same frequency in the product species desorbed from the target. Following recollimation, both the reflected reactant and desorbed products are detected by a quadrupole mass spectrometer located in the third chamber in fig. 1. The 1.6 mm diameter orifice separating the target chamber from the detection chamber insures that
the sampling spectrometer.
beam does not collide with the walls of the ionizer of the mass Neutrals and mass-analyzed ions travel along the mass filter but
at the exit the selected ions are deflected multiplier.
The neutral species continue
towards the input face of the electron undeviated
and are either deposited
on
the chamber walls or removed by the pumps. This separation significantly reduces the electronic noise which is produced when the neutral species is allowed to enter the electron multiplier input. The
output
of the electron
optical
switch
monitoring
multiplier
chopper
and
rotation
the reference are
fed
signal
from
to a lock-in
the
amplifier
augmented with a two-phase accessory. This combination Fourier-analyzes the signal from the electron multiplier and produces the phases and amplitudes of the various
modes
into which the complete
Only the fundamental The
output
determine
signals
SiH,.
of SiH,
molecules
produced the largest, tive
from
the mass
the relative proportions
reactant
This procedure
SiHi,
of
H,.
The
had
product
was not straightforward
by the 33 V electrons
but the presence
spectrometer
of the reaction
SiH: , SiHi , SiH’,
detection
signal waveform
is decomposed.
mode was used in this study.
of Hl
because
from SiH,
Of these, the SiHi
cracking
potential
of
to
fragmentation
used in the mass spectrometer
S’I+ and Hi.
ionization
to be analyzed
H2 and the scattered
complicates
H,
ionizer signal is quantita-
is higher
than
the
appearance potential of Hi from SiH,, so it is not possible to select an appropriately low ionizing electron energy to avoid cracking of SiH, without rendering
the H,
product
undetectable.
The signals at masses 28, 29, 30, 31 and 32 all exhibited the same phase lags and relative magnitudes at all temperatures, which indicates that they all arise from
the
same
Consequently,
parent
molecule
in
the only volatile product
the
ionizer
(namely,
of the reaction
scattered
is molecular
SiH,).
hydrogen.
This behavior is expected for the Si/H system. Contrary to the silicon-fluorine system, where SiF, is a stable species, SiH, has never identified in the gas phase of the silicon-hydrogen system. Consequently, we rule out the possibility of direct desorption Silicon Therefore
of SiH,
to the gas phase.
has three stable isotopes, “Si (92.2%) ‘“Si (4.7%) and ‘“Si (3.1%). the signal measured by tuning the mass spectrometer to mass 30
consists not only of 28SiHi ions, but as includes 29SiH+ and “Sir. However, knowing that the mass-30 signal represents 42% of the sum of the signals from all fragments of SiH, ionization, the former can be used as a direct measure of the silane content of the gas emanating from the target surface. The mass-2 but as noted above, only partly arising from signal consists solely of Hl, reaction product H,. Below about 1000 K the mass-2/mass-30 signal ratio was constant at a value of a few percent. At higher temperatures, this ratio increased significantly, indicating the onset of silane cracking on the silicon target. The mass-2 signal also exhibited a phase lag with respect to the mass-30 signal above 1000 K, but none at lower temperatures. The task of determining
the portions cracking
of the mass-2
of silane
signal (amplitude
by the surface
and phase)
which were due to
and in the mass spectrometer
was accom-
plished by a computer program which used the signals measured with the target at room temperature as well as at the high temperature of the experiment. This procedure
directly yielded the phase lag of the H, reaction
relative to the incident SiH, reactant. The final piece of data needed for determination probability
was the ratio of the ionization
former is known [24] and the latter was measured removing
the target and installing
of the apparent
cross sections
a nozzle directly
product reaction
of H, to SiH,.
in the apparatus
The
of fig. 1 by
into the mass spectrometer
ionizer. By comparing the mass spectrometer responses to known flow rates of N, and SiH, from this nozzle, the SiH, ionization cross section relative to that of N, was obtained. H, to SiH, probability
the ionization
This quantity
but not the phase lag deduced
Additional available
From this information
was determined. details
of the apparatus
affects
cross section ratio of
only the apparent
reaction
from the mass spectrometer
signals.
and data
interpretation
methods
are
in refs. [17] and [25].
3. Results The controllable variables in a modulated molecular beam experiment are the specimen temperature, modulation frequency and incident reactant beam intensity. An experimental campaign consists of fixing two of these variables and varying the third over its accessible range. For each experiment, the apparent reaction probability and phase lag of the reaction product H, relative to the SiH, reactant were determined. Some of the tests were repeated several times, which permitted
estimates
of the precision
to be made.
Based on these
replications, error bars have been assigned to the data points. The molecular beam data are shown in figs. 2-5 along with theoretical curves which will be discussed
in the next
section.
Two
frequency
scans
were performed,
1243 K (fig. 2) and the other at 1443 K (fig. 3). The behavior with respect
to modulation
frequency
at both temperatures
the existence of a branch process in the mechanism. Fig. 4 shows the lack of influence of the incident the reaction
product
that the elementary eliminates
vector.
one at
of the phase lag strongly
indicates
silane beam intensity
AT 1443 K and 20 Hz, this observation
steps of the model consist only of first order reactions,
coverage-dependence
of the sticking
probability
of SiH,
on
suggests and
on Si.
Gelain et al. [26] also reported a pressure-independent reaction probability for the decomposition of silane on silicon at 1183 K for silane pressures up to 1O-J Torr. The temperature dependence of the reaction product vector is shown in fig. 5. As can be seen from this plot, the apparent reaction probability is almost
I.0
I T = 1243 K I 49xl0~5olec”les/cm’-s
=
-
0.01
Theory
I
IO Modulol~o”
Fig. 2. Theoretical
20-
1
100
and experimental
Freqrency
IOOC?
(Hz\
values of c and $J versus modulation
frequency
at 1243 K.
frequency
at 1443 K.
T=l443K
I:
49x10’5
-
molec”les/cm2-s
Theory
IO-
w .= r” a
05-
:: e a ,!
0.2-
:: lz 5
a-
k z 4 0.05
-
0.02
-
O.OlI
I
II11J’
IO
Fig. 3. Theoretical
1000
100 Modulation
and experimental
Frequency
(Hz)
values of c and $J versus modulation
M. K. Fumuun~, D. R. Oiurtder / Thermul cracking
of s&me on Si(l1 I)
399
independent of temperature above about 1130 K and then decreases sharply at lower temperatures. An analogous change is also apparent in the phase lag. Madix and Schwarz [IS] reported similar abrupt changes in the reaction probability and phase lag at around 1050 K for the reaction of molecular chlorine with the (111) face of single crystal silicon. The chemical reactivity transition temperature of 1130 K in fig. 5 falls within the temperature region from 1130 to 1180 K where the (111) surface undergoes a transformation from an ordered 7 x 7 structure to a featureless 1 X 1 structure [27,28]. Beyond the surface transition temperature the apparent reaction probability is nearly constant and the phase lag is low. This behavior is characteristic of a reaction which is nearly supply-limited, by which is meant that all surface steps following initial sticking of the gaseous reactant are rapid and the emission of reaction product is directly dependent on the rate of reactant supply to the surface. In this limit, the apparent reaction probability is twice the sticking probability (because each interaction of an SiH, molecule with the surface produces two molecutes of H,) and, as seen from fig. 4, independent of
t’““‘40 os-
f = 20Hz T : 1443K -Theory
2 2
‘-
4---
0.2-
n ::
e
a
s
,_ ;
O.I-
B * E :: z
T 4
c
0.05-
0
1
0
d”
I 1
I
_I.
0.02-
o.ol
IO
10'5
IO'6 Beam Intensity
3xlo’6
(molecules/en?-s)
Fig. 4. Theoretical and experimental values of c and + versus beam intensity,
incident beam intensity. Fortunately, the surface reactions are not so fast that the phase lag vanishes. Fig. 5 shows that $I is between 10” and 20 o at 20 Hz and figs. 2 and 3 show that the phase lag rises to - 60” at high modulation frequencies. The constancy of the phase lag in fig. 4 is the true indicator of a mechanism with only linear steps. Therefore, kinetic information is still extractable from the phase lag data even though the high temperature reaction probability data serve only to give the total sticking probability of SiH, on Si(ll1). The attempt to characterize adsorptiondesorption of unreacted silane on the Si(ll1) surface was based on measuring the phase difference between the two species in an argon/silane mixed beam. Sufficiently long holdup of silane due to adsorption followed by slow desorption should cause the silane signal to Temperature
I = 4 9 x lOI f : 20
(Ki
molec”les/cm2.s
Hz
-Theory
I
4 70
8.0 Temperature
Fig. 5. Theoretical
and experimental
90
I
100”
(104/K)
values of c and + versus reciprocal
surface
temperature.
lag behind that of the reflected argon, Fig. 6 shows the results of such a test at two substrate temperatures. At low modulation frequencies, where most of the reaction data were taken, the relative phase lag is zero. At chopping frequencies greater
than 30 Hz, a consistent
lag continues
to decrease
approximately target
- 3’.
was near
negative tematic reactive
phase
until
Similar
ambient, lags
negative
phase lag is observed.
- 300 Hz, beyond behavior
and
occurs
when
with
silane
in fact,
are attributed
which it remains
to experimental
The phase constant
the temperature replaced
by neon.
uncertainties.
at
of the The
This
sys-
phase lag error in the mixed beam tests does not compromise the phase lags shown in figs. 2-5, which typical display error bars of
8- 10 O. The results of the mixed beam tests mean that the adsorption-desorption properties of silane on silicon were not observable. This finding has two interpretations. scatters
Either
or cracks)
SiH,
does not physically
or the physically
time (less than a few microseconds) frequencies
where transit
adsorbed
adsorb
on Si(lll)
(it either
silane has such a short residence
that it is not detectable
phase lags do not dominate
at the modulation
the entire process
(e.g.,
900 Hz). ‘The experiments utilizing simultaneous impingement of modulated and steady SiD, beams were conducted over a wide range of intensities
SiH, from
f<
each source and over the temperature modulated
HD detected
result it can be concluded
interval
in the hydrogen
1143 to 1443 K. In no case was
reaction
that atomic hydrogen
l
products.
From this negative
is not a surface intermediate
a. “,le OO*
00
in
A
0
‘0
-4 -
Oo
,O
G E 2
-64
Bv)_8_
-lO-
Corrected
I443 o
K
l
As-measured
I
1183 K A *
I
IllIll
I
I
I
100 Frequency, Fig.
6. Silane
experiments.
phase lag with respect Corrections
to argon
for mass-dependent
versus transit
I
III1
1000 Hz frequency effects
in mixed
are shown.
beam
high temperature
silane cracking, for if it were, it would have produced of the isotope mixing.
HD under the conditions
4. Discussion Construction
of a model of the surface
reaction
is based on the features
of
the experimental results discussed in the preceding section. Specifically, the mechanism appears to contain parallel pathways for production of the same product (Hz), all steps are linear, slow desorption of physically adsorbed silane is not important, cance.
Because
and hydrogen of the paucity
tures below the surface The following
atoms
are not surface
and poor quality
transition,
model conforms
species
of any signifi-
of the data taken at tempera-
the model applies only for T > 1130 K. to these restrictions:
Scattered
2 2 Si + 2 H,(g)
A silane surface
2 Si + 2 H,(g)
1 s,te 2 -+
2WH2
molecule
striking
the surface
is either
silicon atom to form two adsorbed
reflected
SiH,
can take place on either one of two types of surface Fig.
5 suggests
temperature, cracking
probabilities
into SiH,
different
are thermally
in atomic
takes place with different
structure
sites, denoted
activated. that
rate constants,
with a
The latter process as 1 and 2.
7, and n2 are independent
which implies that neither the surface densities
of SiH,
sufficiently SiH,
that the sticking
or combines
molecules.
Sites
subsequent denoted
of
of the sites nor the 1 and 2 must be decomposition
by k, and k,,
of
respec-
tively. This kinetic distinction might arise from differences in the geometrical arrangements of lattice atoms which are nearest neighbors of the SiH, molecule in the two types of sites. This could affect properties such as the Si-H vibrational frequency [29], which in turn could influence the SiH, decomposition
rate constant.
The
two types
of sites in the reaction
model
might
be
identified with kinks and ledges, respectively, on the unreconstructed (111) face of the silicon crystal during the two-dimensional layer-by-layer epitaxial growth process. Kinks and ledges have one of the attributes of the two reactive sites postulated in the model, namely, that of temperature-independent surface densities. Neglecting mass balance dn dt = 24,&g(t)
coverage-dependence equations -k,n,
of the sticking
for the model can be written dm x = 2nz4,g(t)
- k,m,
probabilities, as:
the surface
M. K. Furwam,
D. R. Olmder / Thermd trucking of dune
on Si(ll I)
403
where n and m are the surface number densities of the SiH, intermediates on sites 1 and 2, respectively, and Z, is the amplitude of the incident SiH, beam. The gating function of the modulated beam is represented by g(t), which is approximately a square wave of period l/f. The time dependence of n, m, and g are periodic, and Fourier expansion up to the first mode yields: n(t) m(t) g(t)
(24 (2b)
= n, + El ei2+, =
m.
+
Ci, ei2=“,
+(l + g,
=
ei2nf’),
where g, = 4/a for an incident square wave. The steady state densities no and are real whereas the quantities ii, and G,, representing the first Fourier components, are complex. The reaction product vector is defined as the ratio of the fundamental modes of the rate of production of H, to the rate of SiH, impingement: m,
c emi+ = (kin,
+ k2m,)/+g,Z0.
(3)
Eqs. (2) are substituted into eqs. (1) and the solutions for %, and Ti, are inserted into eq. (3). This procedure yields the following formulas for e and 9: E = 2(x*
+JJ’)~‘*,
+ = tan-‘(y/x),
(4)
where x=-+
771 1 +a:
7)2
1 +a:’
Y=
91%
1 + a,2
+
%a*
1 +a;’
(5)
and a, =: 2nf/k,,
a2 = 2rrf/k,.
(6)
In addition to the rate constants, the E and C#J formulas contain the modulation frequency. The beam intensity does not appear because the system of equations is linear. The temperature is implicit in the Arrhenius forms of the rate constants. Data fitting was performed by using a computer program which minimizes the sum of normalized squares of the errors in + and the logarithm of C. The fitting procedure resulted in the following set of parameters: TJi = 0.038,
72 = 0.068,
k, =: 5.1 x lo4 exp( - 16.3/RT),
s-l,
k, = 4.6 x lo6 exp( - 17.8/RT),
s-l,
where the activation energies are in kcal/mol and R is the gas constant. The theoretical curves for E and + based on these model parameter values are shown in figs. 2-5. Agreement with the data is satisfactory except for tempera-
tures below - 1200 K, which, however, is close to the temperature range of the surface structural transformation. Three other sets of reaction parameters were also found which fit the data in a visual sense nearly as well as the set given above. However, these alternate reaction parameters exhibited poorer least squares agreement with the data than the best-fit set. The rate constants for these sets of parameters are shown as lines A, B, and C in fig. 7 along with the rate constants of the best fit. Although the magnitudes of the rate constants in each group are within a factor four of each other, their preexponential factors and activation energies are dratically different. The insensitivity of the experimental reaction probabilities to temperature (fig. 5) leaves only the frequency responses of + at two temperatures (figs. 2 and 3) from which the temperature behavior can be inferred. This results in the substantial variances in preexponential factors and activation energies even though the absolute magnitudes of k, and k, are in close accord. Despite this shortcoming, the activation energies are in good agreement with those determined in earlier studies, which ranged from 17 to 20 kcal/mol [4-81. The total silane sticking probability of - 0.1 is also consistent with previous work
Temperature
7.0
Fig.
7. Rest-fit
and
8.0 Temperature runner-up
SiH,
(K)
9.0 (104/K)
decomposition
rate constants.
405
5. Conclusions Modulated results
molecular
interpreted
beam
experiments
have
in terms of a mechanistic
been
model
conducted
with the (111) face of single crystal silicon in the temperature 1443
K.
Isotope
mixing
and
producing
negative
results,
molecules
which do not reflect
silane
aided in model development.
with two types of sites on the surface. bound
SiH,
molecules
which
tests, a branched
Each chemisorption
decompose
into
although
In the model,
undergo
the
of silane
range of 1130 to
adsorption-desorption
from the surface
and
of the reaction
silane
reaction
event produces
two
with differen
rate
Si and H,
constants. Molecular beam sufficiently precise decrease
in reactivity
resulted about
data obtained at temperatures to warrant reaction modeling.
from
surface
below 1130 K were not However, the dramatic
compared
to the high
temperature
structural
transformation
which
reaction
is known
probably to occur
at
the same temperature.
Acknowledgements Both authors and
would like to thank Dr. B.A. Joyce
for his helpful
achieving Office
suggestions
contaminant-free
of Energy
Division
surfaces.
Research,
of the US
for his interest in this work
for overcoming
the persistent
This work was supported
Office of Basic Energy
Department
of Energy
Sciences,
under
problem
of
by the Director,
Materials
contract
No.
Sciences
DE-AC03-
76SF00098.
References [I]
H.C. Abbink,
[2] M. Shimbo,
R.M.
Broudy
J. Nishizawa
[3] E.I. Girargizov,
J. Crystal
[4] R.C. Henderson
and R.F.
(51 B.A. Joyce
and R.R.
[6] C.R. Booker
Farrow,
[9] R.H. Jones,
Growth
Bradley,
23 (1974)
301.
Phil. Mag. 15 (1967) Sot. 121 (1974)
310.
289. 1167.
899.
W.J. Siekhaus and J.A. Schwartz,
J. Vacuum
1429. [lo]
C.T. Foxon,
[ll]
J.A. Schwartz
[12]
R.L.
Palmer
M.R.
Foxon
and B.A. Joyce, Surface
J. CoIIoid Interface
and B.A. Joyce,
Surface
Surface
Sci. 44 (1974)
Sci. 46 (1974)
and J.N. Smith, Jr., Catalysis
1131 D.R. Olander, [14] CT.
Boudry
and R.J. Madix,
4673.
267.
194.
Phil. Mag. 14 (1966)
J. Electrochem.
Growth
Sci. 30 (1972)
Phil. Mag. 14 (1966)
Bradley,
D.R. Olander,
J. Appl. Phys. 39 (1968)
J. Crystal
52 (1981)
Helm, Surface
and B.A. Joyce,
[7] B.A. Joyce and R.R. [8] R.F.C.
and G.P. McCarthy,
and T. Terasaki,
317.
Rev. 12 (1975)
Sci. 58 (1977) Sci. 50 (1975)
169. 434.
279.
69.
Sci. Technol.
9 (1972)
[15]
S.L. Bernasek
[16]
R.J.
and G. A. Somorjai.
Madix.
in: Physical
J. Chem.
Chemistry
of Fast
Phys. 62 (1975) Reactions,
3149.
Vol. 2. Ed. D.O.
Hayward
(Plenum.
1975). [17] A.J.
Machiels
(181 R.J.
Madix
[19]
and D.R.
and J.A.
M. Balooch
and D.R.
[20] G.E.
Gdowski.
[21]
Allen,
F.G.
[22]
Private
R.C:Henderson,
[24]
L.J.
Kieffer,
Discharge M.K.
[26]
Lasers,
Marcus
JILA
J. Electrochem.
A. Cassuto
130 (1983)
Report
of Epitaxial
Cross Silicon
and W.D.
Robertson
Surface
Sci. 1 (1964)
G. Lucovsky,
J. Vacuum
1208.
Data
for Modeling
Gas
1973.
Deposition
by Thermal
Cracking
(1983).
and P. LeGoff,
Florio
Section
13, September
Abstract
of Bulletin
23.
Lander,
541.
J. Appl. Phys. 42 (1971)
Collision
Center
Chemistry
[27] J.V.
151.
Sci. 127 (1983)
1978. Polito,
of Electron
LBL-15448
Sot.
Surface
325.
264.
1510.
and W.J.
[28] J.J. [29]
Madix.
Information
The Surface
PhD Thesis,
C. Gelain,
Sci. 65 (1977)
Sci. 24 (1971)
from B.A. Joyce,
R.B.
A Compilation
Farnaam,
(1968)
Olander,
Fair and R.J.
communication
of Silane,
Surface
Surface
J. Appl. Phys. 28 (1957)
[23]
[25]
J.A.
Olander,
Schwartz,
Surface
Sci. 22 (1970)
125.
Sci. Technol.
16 (1979)
1225.
459.
of French
Ceramic
Society
80