The reaction of fluorine atoms with silicon
`Daniel L. Flamm, Vincent M. Donnelly, and John A. Mucha
`Bell Laboratories, Murray Hill, New Jersey 07974
`
`(Received 19 August 1980; accepted for publication 26 January 1981)
`Fluorine atoms etch silicon with a rate, RF(Si\ = 2.91 ± 0.20X 1O- 12T I 12nFe - 0.108 eV/kT A/min,
`where n F (cm- 3 ) is the atom concentration. This etching is accompanied by a chemiluminescent
`continuum in the gas phase which exhibits the same activation energy. These phenomena are
`described by the kinetics: (1) F(II) + Sisurf-+SiF2(gl' (2) SiF21g) + Flg)-+SiF~g\' (3) SiF21g\ + F 21g)
`-SiF~g) + FigI' (4) SiF!lg)--+SiF3Ig) + hVcontinuum where formation ofSiFz is the rate-limiting step.
`A detailed model of silicon gasification is presented which accounts for the low atomic fluorine
`reaction probability (0.00168 at room temperature) and formation ofSiFz as a direct product.
`Previously reported etch rates ofSiOz by atomic fluorine are high by a constant factor. The etch
`rate ofSiOz is RFISiO,) = (6.14 ± 0.49) X 10- 13nr T 1/2e - O.I63/kT A/min and the ratio ofSi to Si02
`etching by F atoms is (4.74 ± 0.49)e - OOSS/kT.
`
`PACS numbers: 81.60. -
`
`j, 82.65.Nz
`
`I. INTRODUCTION
`
`A broad understanding of the rates and mechanisms by
`which free radicals react with various substrates is important
`for the development and selection of plasma-etching tech(cid:173)
`niques. Although fluorine atoms are the principal gaseous
`reactant in many common etching processes, relatively little
`information is available on the reaction of these radicals with
`common semiconductor substrates. Recently, several inves(cid:173)
`tigators I have shown that an apparent continuum centered
`at 632 nm accompanies the etching of silicon by fluorine.
`Donnelly and Flamm la studied the spectrum of the chemilu(cid:173)
`minescence and compared it with spectra from species in the
`discharge and afterglow regions of an SiF4 discharge; they
`ascribed the chemiluminescence to the reactions:
`
`where spectral, kinetic and thermodynamic considerations
`were consistent with the view that the emitting radical is
`SiF3• In other work the reaction probability for etching of
`silicon dioxide by F atoms and the etch ratio of Si to SiOz at
`room temperature was reported.2
`In the present investigation, the etching of silicon by F
`atoms and intensity of the concommitant luminescence were
`measured as a function of temperature (223-403K) and F(cid:173)
`atom concentration (nF = 1.6x 10 15
`- 7.7x 10 15 cm- 3
`).
`The etching ofSiOz at room temperature was also measured
`in order to determine simultaneously the ratio ofSi and Si02
`etch rates.
`
`II. EXPERIMENTAL
`
`Fig)}
`.
`.
`+ Slsurf-+S1surf - F chemisorbed ,
`F
`21g)
`. Fig) + Sisurf - F chemisorbed --+SiF 2(ads) ,
`SiF 2lads) --+SiF 2(g) ,
`F{g) + SiF2Ig)-+SiF!tg\,
`F 2111) + SiF2Ig)-+SiF~g) + F(II\'
`SiF~g)-+SiF 31g) + hVcon'inuum'
`
`(1 )
`
`(2)
`(3)
`
`(4)
`
`(5)
`
`(6)
`
`The discharge-flow apparatus used for the etching ofSi
`and Si02 by atomic fluorine is shown in Fig. 1. The discharge
`and flow arrangements have been described previously. 2a,la
`Single-crystal silicon samples (100) were bonded to the end of
`a 2. 54-cm-o.d. aluminum rod (6061-T3) with epoxy and posi(cid:173)
`tioned in-line with the wall of a 2.54-cm-i.d. aluminum reac(cid:173)
`tion cell.
`The uncoated aluminum reaction cell had five ports
`through which the inlet tube, outlet tube, and silicon sub-
`
`ALUMINA TUBE
`ALUMINUM TUBING
`
`SAPPHIRE WINDOW
`Sl SAMPLE
`- -,'1
`------ -
`~ I
`.--I----r'--""""""'I
`I
`
`(
`
`t F2FLOW
`
`FI G. I. Schematic of apparatus.
`
`HEATING TAPE OR COLO BATH
`
`3633
`
`J. Appl. Phys. 52(5), May 1981
`
`0021·8979/81/053633-07$01.10
`
`~ 1981 American Institute of Physics
`
`3633
`
`
`
`Page 1 of 7
`
`Samsung Exhibit 1007
`
`

`

`strate holder were sealed using Viton or Kalrez o-rings (Kal(cid:173)
`rez o-rings did not seal well at the lowest temperatures em(cid:173)
`ployed). The reaction ceIl was thermaIly insulated, while
`that part of the sample-holder rod (2.54-cm o.d. X 25.4-cm
`long) which extended out of the vacuum system was heated
`or cooled with electrical heating tape, a constant tempera(cid:173)
`ture bath, or a circulating refrigerant loop, depending on the
`desired temperature. A thermocouple was inserted into a
`thermowell just under the sample, and a second thermocou(cid:173)
`ple was attached to the surface of the substrate holder imme(cid:173)
`diately outside of the cell. The temperatures at these two
`stations were always in good agreement during the experi(cid:173)
`ments, indicating that temperature gradients along the sam(cid:173)
`ple holder near and within the cell were negligible. The sili(cid:173)
`con samples were 0.05-cm thick and the epoxy had a thermal
`conductivity of8 X 10-.1 W /sec cm. A heat-transfer calcula(cid:173)
`tion including conduction, convection, radiation, and the
`heat of reaction, revealed that the sample surface is essential(cid:173)
`ly isothermal in these experiments.
`Fluorine atoms were generated by dissociation of F2
`(Air Products Technical Grade) in a 14-MHz-rf discharge
`50-cm upstream of the reaction cell. Fluorine-atom concen(cid:173)
`trations were measured both upstream and downstream of
`the reaction ceIl by gas-phase titration with C12. 2n . .1 The
`average of these titrations ( < 10% difference for all condi(cid:173)
`tions) was used to interpret the data. Atom concentrations
`ranged from 1.0X 1015 atoms/cm' at 3.8-W discharge power
`to 5.1 X IOI.~ atoms/cm' at 78 W with constant pressure
`(0.40 Torr) and F2 flow (44 sccm), the conditions used in the
`etch-rate experiments. In emission experiments, the pressure
`was varied between 0.2 and 0.6 Torr (27-80 Pal and the F2
`feed rate was between 18 and 70 sccm.
`Chemiluminescence, originating in the gas phase above
`the Si (100) samples, was monitored through a l-in-diam
`sapphire window which faced the sample in a direction per(cid:173)
`pendicular to the flow-tube axis and the sample surface nor(cid:173)
`mal. A cooled photomultiplier tube (RCA C31 034) equipped
`with a 440-nm interference filter, 10 nm fullwidth at half
`maximum (FWHM) or Corning CS 2-61 long-pass red filter
`was used to monitor the emission.
`Silicon etch rates were measured using samples having
`10 OOO-A-thick, 0.2-cm-wide parallel bars of steam-grown
`thermal oxide that covered 50% of the sample surface area.
`These bars were prepared by etching through a
`photoresist mask with 8: 1 buffered HF, followed by acetone
`and methanol washes to remove the resist. In order to mini(cid:173)
`mize or eliminate the presence of native oxide during the
`silicon etch studies, the bonded samples were immersed in
`buffered HF (BHF) for 20 sec, washed with deionized water,
`and blown dry with N2 immediately before use. Samples
`were then sealed into the vacuum system and evacuated to
`several microns before starting the molecular fluorine flow.
`Several trials were also performed using samples that had
`been exposed to air for several weeks and not subsequently
`treated with HF before exposure to atomic fluorine (see be(cid:173)
`low). After etching, the samples were removed from the rod
`by immersion in boiling dimethylformamide. The oxide
`mask was then dissolved in HF, and the etch depth was mea(cid:173)
`sured using a Sloan Technology Model 90050 Dektak stylus
`
`thickness monitor.
`Room-temperature Si02 etch rates and Si:Si02 etch ra(cid:173)
`tios were determined using the same procedure, except that
`freshly patterned samples were always employed, and the
`photoresist-free pattern was not treated with BHF before
`use. Instead, long etch times were employed ( > 30 min) to
`insure that the effect of any native oxide was negligible. Ox(cid:173)
`ide thickness was measured before and after etching with a
`Nanospec AFT Model 174 microspectrophotometer; etch(cid:173)
`ing of the exposed silicon on these samples was measured as
`above.
`
`III. RESULTS AND DISCUSSION
`A. Silicon etch rates and chemiluminescence
`Silicon etch depths varied from 6000 to 200 000 A, de(cid:173)
`pending on temperature, F-atom concentration, and etch
`time. The depth was uniform from "stripe" to "stripe" and
`along each stripe on all of the samples analyzed. This pro(cid:173)
`vides experimental evidence that reactant depletion and con(cid:173)
`centration-boundary-Iayer effects (diffusion control) can be
`ignored. While the depth etched was proportional to time, in
`general the exposed silicon adjacent to the oxide barmasks
`was etched slightly more than exposed silicon in the center of
`the stripes. This difference was S 4000 A and did not vary
`with etch time. It appears that etching at the oxide-silicon
`boundary began somewhat earlier than etching at the center
`of the stripes. Perhaps a product of the BHF treatment tend(cid:173)
`ed to remain at this interface and minimized the initial re(cid:173)
`growth of surface oxide.
`Chemiluminescence was continuously monitored dur(cid:173)
`ing all of the runs on masked samples. In some of the early
`experiments, before special attention was given to surface
`cleaning, we noted a delay in chemiluminescense and etch(cid:173)
`ing. For instance, in one set of four runs (nF = 4.0X 1015
`cm - \ T = 296 K) the initial luminescence was relatively
`weak; but after a period of time there was a rapid and dra(cid:173)
`matic (10-30 fold) increase in chemiluminescent intensity.
`Three samples, which had been exposed to air for several
`weeks and were not treated with buffered HF before etching,
`showed a latency of ;::::: 8 min. When the etching time for
`these runs was plotted as a function of etch depth and ex(cid:173)
`trapolated to zero depth, a positive intercept was obtained
`(i.e., an induction period for etching). This induction period
`was in excellent agreement with the corresponding chemilu(cid:173)
`minescent latency. The fourth air-exposed substrate was
`dipped in buffered HF solution for 20 sec immediately before
`etching. The emission-versus-time trace was similar to the
`above runs, except that the latency was reduced from 8 to
`;:::::2 min. Thus, it may be concluded that the delay in lumi(cid:173)
`nescence was associated with the removal of a surface layer
`(native oxide or more likely some other contamination).
`In subsequent work, uncertainties in the onset of etch(cid:173)
`ing were overcome by using freshly prepared samples, al(cid:173)
`ways treating the samples with HF solution prior to etching
`(as described above), and etching for times much longer than
`the latencies noted in early experiments. Extrapolation of
`etch depth versus time to t = 0 (discharge turned on) showed
`little or no induction period in these runs. The latency in
`emission also was not present in later experiments; instead, a
`
`3634
`
`J. Appl. Phys., Vol. 52, No.5, May 1981
`
`Flamm, Donnelly, and Mucha
`
`3634
`
`
`
`Page 2 of 7
`
`

`

`very intense transient emission was observed at t = 0 which
`decayed in - 1 min.
`After the initial step change in luminescent intensity, a
`very gradual increase in luminosity (- 2%/min for the first
`20 min) was always observed. At room temperature and n p
`= 4 X 1015cm -3, fresh, unpatterned samples reached a con(cid:173)
`stant "saturation luminosity" after -40 min. As a sample is
`etched the surface roughness increases, and the slow in(cid:173)
`crease in chemiluminescence is probably associated with a
`change in surface texture.
`
`B. Emission and silicon etching versus temperature
`The saturated intensity (see Sec. III A) was measured as
`a function of temperature. Figure 2 shows a typical data set
`taken at constant pressure and mole fraction of F atoms
`(constant discharge power), which has been corrected for the
`effect of temperature on gas-phase F-atom density [nF(T)
`= n P(296) X (296/T)]. The intensity is well described by an
`Arrhenius expression:
`/ = /o(273/T)1/2e - E,IkT,
`
`(7)
`
`where the factor (273/T) I 12 corrects for the temperature de(cid:173)
`pendences of atom concentration and atom flux to the sur(cid:173)
`face; k is the Boltzman constant. The slope of these data
`corresponds to an activation energy Ei of 0.101 eV (2.33
`kcallmole). Table I presents Ei'S independently determined
`from regression of four different experimental conditions.
`The weighted average activation energy for these five experi(cid:173)
`ments is 0.116 ± 0.012 eV (2.68 ± 0.28 kcallmole). The
`maximum deviation of any one run from the mean is 15%,
`indicating little if any dependence on pressure, flow, atom
`concentration, or spectral region over a fairly wide param(cid:173)
`eter space.
`Figure 2 also shows the temperature dependence of etch
`rates similarly corrected for the effect of temperature on
`atom density. These etch rates are described by the regres(cid:173)
`sion equation
`) = 2.91 ± 0.20x 1O- 12nF T 1/2e - E"'hlkT.
`R 1Si
`Least-squares analysis yields an activation energy E etch
`= 0.108 ± 0.005 eV (2.49:0.12 kcallmole). Within experi(cid:173)
`mental error the etch rate and chemiluminescent intensity
`have the same activation energy.
`The activation energy derived from the etch-rate data is
`associated with the slowest step in the etching reaction. In
`emission experiments, the intensity at any given temperature
`is a measure of the yield of SiF 2 relative to all fluorosilicon
`desorption products. Regardless of which step is rate-limit-
`
`(8)
`
`en I--z
`
`~
`ai
`a::
`<{
`
`>-
`I--
`Vi z
`
`UJ
`I--
`~
`UJ
`~
`I--
`<{
`...J
`UJ a::
`
`10000
`
`5000
`
`c:
`E
`"-
`o<{
`
`UJ
`1000 I--
`<{
`a::
`:I:
`U
`I--
`500 w
`
`20
`
`2.5
`
`30
`
`3.5
`
`4.0
`
`45
`
`1000 (K -I)
`T
`
`FIG. 2. Silicon etch rate and chemiluminescence YS IOOO/Tfor
`n .. = 2.9X 10". Intensity data group corresponds to Run I, Table I.
`
`ing, the activation energies in the two sets of experiments can
`agree in only two circumstances: (1) the fraction ofSiF2 rela(cid:173)
`tive to all SiF x desorption products is temperature indepen(cid:173)
`dent, or (2) the chemiluminescence and etching are associat(cid:173)
`ed with the same rate-limiting process-implying that SiF2
`is the primary etch product. These two cases are treated in
`detail below (Sec. III.E.).
`The probability E F1Si ) of an impinging fluorine atom un(cid:173)
`dergoing reaction with silicon may be defined as
`
`EF(Si) = 4NaPsi R sJM Si U(nFv)],
`(9)
`where Na is Avogrado's number, PSi is the density of silicon,
`MSi is the atomic weight of silicon (28.09), nF is the gas-phase
`number density of atomic fluorine, v is the mean thermal
`velocity [equal to (8kT hrMp) I 12] and !nFv is the flux of
`atoms to the surface. In deriving Eq.(9), the factor of 4 in the
`numerator arises from the assumption that SiF2 is a minor
`desorption product and that the major reaction product
`formed on the surface from F atoms is SiF4 (EF1Si) would be
`half as large if SiF2 were the major product). This assump(cid:173)
`tion is dictated by the arguments presented in Sec. III E (but
`
`TABLE I. Chemiluminescence activation energies. Weighted mean: Ei = 0.116 ± 0.012 eV.
`
`Run E,
`(eV)
`
`Pressure
`(Torr)
`
`Flow
`(sccm)
`
`Power
`(W)
`
`Temperature
`Range(K)
`
`Filter
`Type
`
`No. of
`Points
`
`1
`2
`3
`4
`5
`
`0.102
`0.\20
`0.135
`0.123
`0.12\
`
`0.56
`0.58
`0.54
`0.54
`0.27
`
`70
`70
`18
`18
`18
`
`18
`40
`40
`40
`38
`
`221-373
`309-373
`243-393
`297-379
`233-373
`
`CS-2-6\
`CS-2-6\
`440nm
`440nm
`440nm
`
`72
`16
`40
`16
`62
`
`3635
`
`J. Appl. Phys., Vol. 52, No.5, May 1981
`
`Flamm, Donnelly, and Mucha
`
`3635
`
`Page 3 of 7
`
`

`

`is given by the regression equation
`see Ref. 16). €F(Si)
`€FISil = 0.1162 ± 0.0080e E"jkl,
`(10)
`where Eetch = 0.108 ± O.OOS and €F(Si) ranges from 0.00168
`at room temperature (23 C) to 0.0040 at 100 C.
`
`C. Silicon etching and emission versus F-atom
`concentration
`
`In the above discussion, it has been implicitly assumed
`that the etching of silicon is proportional to F-atom concen(cid:173)
`tration. Figure 3 does indeed show that the etch rate changes
`linearly with F-atom concentration as it is varied by chang(cid:173)
`ing the discharge power. Furthermore, the intercept is not
`significantly different from zero, in accord with the indepen(cid:173)
`dent experimental finding4 that the rate of silicon etching by
`molecular fluorine ( - 3 A/min at 300 K) is negligible com(cid:173)
`pared with present etch raJes (1()()()....4000 A/min).
`Unlike etching, the intensity of chemiluminescence
`does not increase linearly with F-atom concentration (Fig.
`3). Equations (2)-(S) suggest that the luminescent intensity
`should be given by5
`
`(11 )
`
`)/(k4nF
`)
`($4k4nF
`I
`+1
`- -+ 1 ,
`-=
`kSnF,
`nF
`$SkSnF2
`where the first term in the numerator arises from the reac(cid:173)
`tion ofSiF2 [nSiF, a: k2np (k4nF + kSnF,)] with atomic flu(cid:173)
`orine in Eq. (4), and the second term corresponds to reaction
`with molecular fluorine, Eq. (S). $4 and $s are the fractional
`yields of SiFT, relative to all products arising from the reac(cid:173)
`tion ofSiF2 with F and F 2, respectively. A plot of I /n F
`against n F/n 1", along with the least-squares fit to Eq. (11) are
`shown in Fig. 4; the data are in excellent agreement with this
`model. From the fit, k4/ks = 8.S, so that reaction (4) is from
`I.S-7.S times faster than reaction (S) over the range of our
`
`u:;
`I--
`Z
`::)
`cO
`ex::
`~
`>-
`I--
`U'i
`z
`W
`I--
`~
`W
`>
`ti
`.-J
`w
`ex::
`
`8
`
`6<=
`E
`"-
`oc:{
`~
`W
`I--
`4~
`X
`U
`I--
`w
`
`2
`
`~ ____ ~ ______ L -____ -L ______ ~ ____ ~ __ ~O
`
`o
`
`FIG. 3. Etch rate and chemiluminescence vs nF at room temperature
`(296 K).
`
`20r--------------------------------------,
`
`15
`
`'" !:::
`z
`::>
`)0-cr
`<f cr 10
`I--
`1i5
`cr
`<f
`
`"-c:
`
`"--
`
`5
`
`OL-----~~----~------~------~-----J
`0.0
`0.2
`0.4
`0.6
`0.8
`1.0
`
`FIG. 4. I/nrn r. vs n,/n F• at room temperature 1296 K). These intensity
`data are also shown in Fig. 3.
`
`data. The least-squares analysis also yields $4/$S::::: 100.
`The agreement is sensitive to the value of k4/ks, but does not
`depend strongly on $4/$S; hence it is concluded
`20 S $4/$5 S 120.
`At the present pressure, the reaction ofSiF2 with F
`could conceivably proceed by a termolecular mechanism,
`but a choice between bi- and termolecular models on the
`basis of the present data would be arbitrary. However, it is
`significant that the reaction ofSiF2 with both atomic and
`molecular fluorine must be included in either model to ex(cid:173)
`plain the relationship between luminescence and F-atom
`concentration. Our observations of a chemiluminescence
`during the etching of Si by molecular fluorine 4c provide fur(cid:173)
`ther evidence for reaction (S). Smolinsky and Flamm6 also
`presented indirect evidence for this process. Additional ex(cid:173)
`periments to confirm steps (4)-(6) are now in preparation.
`This will involve adding F and F z to a flow ofSiFz formed in
`the high-temperature reaction of SiF4 with solid Si.7
`
`D. Si02 etch rates and Si:Si02 etch ratio
`The etch rates ofSi02 and Si were simultaneously mea(cid:173)
`sured as a function of F-atom concentration at room tem(cid:173)
`perature. Both rates were proportional to nF with an inter(cid:173)
`cept at the origin (9S% confidence level). The silicon etch(cid:173)
`rate data from these experiments are included in Fig. 3.
`In the course of this work, we discovered a systematic
`inconsistency between our data and previously reported
`Si02 etch rates. 2. A number of significant improvements
`(such as the installation oflinear mass flowmeters and a pres(cid:173)
`sure controller) have been made in the present apparatus
`since that early study, and it was possible to reexamine the
`
`3636
`
`J. Appl. Phys., Vol. 52, No.5, May 1981
`
`Flamm, Donnelly, and Mucha
`
`3636
`
`Page 4 of 7
`
`

`

`previous raw etch-rate data and flow calibrations. A system(cid:173)
`atic error in Clz flow calibration was found to influence the
`previous results. When this is taken into account, the correct
`atom concentration applicable to the study of Flamm et al. 2a
`is found to be n F -9.3 X 10 15
`• It follows that the previously
`reported rates were high by a constant factor of -1.5.
`U sing corrected raw data from Ref. 2a together with the
`present room-temperature Si02 etch rates, the etch rate of
`Si02 by F atoms becomes
`R
`-614+049XlO- '3n T'/2e-o.'63IkTA/min (12)
`F(SiO,) - . _ .
`F
`,
`and the reaction probability for SiOz is
`0000
`.
`112
`-O.163/kT
`CF(SiO,) = 0.0 ±. ge
`,
`where it is assumed that the final reaction product formed on
`the surface of Si02 by F atoms is SiF4 ,
`Combining Eqs. (8) and (12), the ratio of the Si etch rate
`to that of SiOz is
`RF(Si/RF(SiO,) = 4.74 ± 0.49 eO.055IkT,
`which ranges from 41.0 at room temperature to 26.2 at
`100 C. The room-temperature etch ratio (42:1), which
`Flamm 2b calculated from preliminary Si etch data (based on
`the previous (high) atomic-fluorine calibrations) is in good
`agreement with the present result (41 ± 4).
`
`(13)
`
`(14)
`
`E. Reaction of a silicon surface with atomic fluorine:
`mechanism
`Recent work8 provides evidence that fluorine atoms
`form a stable chemisorbed layer on the surface of single crys(cid:173)
`tal silicon. Consider a < 1(0) silicon surface. Since Si atoms
`have two free bonding sites per atom, it can be anticipated
`
`that fluorine atoms will react to form a periodic array ofSiFz
`groups on the surface. Each of these groups is, in turn, bound
`to two Si atoms in the bulk crystal:
`
`F
`F
`\/
`Si
`surface
`---- --- --( --\----- -- --- -- --
`Si
`Si bulk
`\ /
`/
`\
`
`(IS)
`
`Electron spectroscopy for chemical analysis (ESCA)
`supports this detailed picture.H After silicon is etched by
`XeF2, fluorine remains bound to the surface with bonds that
`exhibit a chemical shift similar to that ofSiF2 molecules. A
`study of infrared emission from a silicon surface during
`9 The radi(cid:173)
`XeF2 etching also suggests a fluorinated surface. H
`•
`ation is polarized and wavelengths are consistent with
`known Si-F-bond absorptions.
`At steady state during etching, some incident F atoms
`may be expected to physisorb as a secondary layer on the
`SiF2-like surface. Since cFISil is small, it follows that most
`fluorine in this secondary layer simply desorbs. Our experi(cid:173)
`ments show that the Si etch rate is directly proportional to
`the impingement rate of F atoms from the gas phase (at least
`within the present range of n F :5 6 X lO '5cm- 3
`). This fact,
`together with the low value of cF(Si)' suggest that surface
`coverage (eF ) by the physisorbed fluorine layer is sparse and
`that recombination of physisorbed atoms is likely to be slow
`compared with desorption. The same conclusion can be
`reached by a statistical thermodynamic calculation of cover(cid:173)
`age lO giving eF :5 10- 4 for a Langmuir isotherm with param(cid:173)
`eters corresponding to physical adsorption ( :5 5 Kcal heat of
`adsorption and an oscillator frequency v_1O '2 sec-I).
`
`Chemisorbed SiF2 groups may be ultimately gasified by the reaction of impinging atoms with an Si-Si bond:
`
`{F} -
`
`F
`F
`\/
`---- --- --,--\--------------
`.urfec.
`Si
`Si bulk
`Si
`/ \ / \
`
`or by
`
`F
`F
`\/
`{Fl -
`Si
`surface
`---------1--\------- --- ----
`Si bulk
`Si
`/ \ / \
`followed by
`
`F
`F
`\ /
`{FI-
`F Si
`_____ -___ -----j __ 0 __ , \ ______ _
`Si
`Si
`/ \
`/ \
`
`F
`
`F
`
`\/
`Si
`°
`
`0
`
`F
`--------------j----'\---------------
`Si
`Si
`/ \
`/ \
`F
`F
`\/
`F Si
`--------------j-~--'\-------------
`Si
`Si
`/ \
`/ \
`
`F
`F
`\/
`Si . .
`
`F F
`-----------/----\---------------
`Si
`Si
`/ \
`/ \
`
`(16a)
`
`(16b)
`
`(17)
`
`3637
`
`J. Appl. Phys., Vol. 52, No.5, May 1981
`
`Flamm, Donnelly, and Mucha
`
`3637
`
`Page 5 of 7
`
`

`

`where IF 1 indicates an incident fluorine atom. Since the Si(cid:173)
`Si bond energy is - 54 Kcallmole and the Si-F bond energy
`is -140 Kcallmole, both mechanisms are sufficiently ex(cid:173)
`othermic to liberate SiF 2•
`The luminescence and etch rate have identical tempera(cid:173)
`ture dependences. As explained in Sec. III. B, this agreement
`suggests that either SiFz is the major desorption product, or
`that the ratio of desorbed SiF1 to the total desorbed product
`yield (SiF,.) is temperature independent. The latter circum(cid:173)
`stance imposes a very specific constraint: reaction (16a) must
`be replaced by a single branching reaction in which the pre(cid:173)
`cursors ofSiFx (x> 2) are also formed (e.g., reactions (16a)
`and (16b) are two channels of a single process); and the
`branching ratio for these channels must be independent of
`temperature. Alternative parallel or sequential mechanisms
`almost certainly imply that the proportionate yield ofSiFz1g,
`would be a function of temperature. For example, if the rate(cid:173)
`limiting step were reaction (16b) and fluorination of the sur(cid:173)
`face moiety proceeded in competition with reaction (17), the
`yield ofSiF Zig) should exhibit a temperature dependence, due
`to distinct activation energies for each reaction step. Like(cid:173)
`wise, processes in which SiF 2 is formed, physisorbed, and
`then subject to fluorination on the Si surface, should lead to
`an increase in SiF Z(gl with temperature, relative to more
`weakly bound SiF4 , As a result, emission and etch rate would
`each exhibit a different temperature dependence. This is con(cid:173)
`trary to observation.
`These requirements are so stringent that a priori we
`would tend to favor mechanisms in which SiF2 is the domi(cid:173)
`nant desorption product. However, Winters!! reported that
`SiF4 was almost quantitatively desorbed from silicon when
`etching with XeF1 , and Vasile!l recently observed SiF4 as a
`major desorption product while etching polycrystalline sili(cid:173)
`con with fluorine (F + F z) in high vacuum. Thus, we are
`forced to the conclusion that etching of single-crystal silicon
`generally proceeds by the single-reaction branching mecha(cid:173)
`nism [reaction (16)].
`While the branching ratio (x*) between the products
`SiFzlg) and SiF4 is unknown, we have estimated the absolute
`luminescent intensity, based on photomultiplier gain, filter
`transmission, and geometric factors. This corresponds to a
`"quantum yield", 1> *;:::: 10- 4 photons per gasified Si atom.
`To obtain x*, 1> * must be divided by the ratio ofluminescent
`to nonluminescent SiF3 found in reactions (4) and (5), and by
`the fluorescence quantum yield of SiF!. Since both of these
`factors are.;; 1, 1> * provides a lower limit for x*. According
`to this description, only 1 of the incident fluorine atoms that
`react with silicon are consumed in the rate-limiting step [re(cid:173)
`actions (16)]. If the measured activation energy (Eelch ) is as(cid:173)
`sociated with this reaction, a "steric" factor 5 can be defined
`for reaction with the bound moiety:
`
`( 18)
`
`and from Eq. (10),5 = 0.03. Correspondingly, -3% of the
`proportionate atom flux is effective in the rate-limiting step.
`The small reaction probability and low activation ener(cid:173)
`gy are consistent with the necessity for F atoms to reach an
`underlying Si-Si bond; this is characteristic of processes
`which require the coordinated motion of surface atoms and
`
`has previously been noted during the adsorption of CO on
`silicon 1.1 and chemisorption of oxygen on nickel. \4 It may be
`noted that the activation energy for this kind of mechanism
`is primarily associated with the substrate temperature rather
`than that of the gas. Unfortunately the present data provide
`no further insight into other significant details. For instance,
`reactions (16) may proceed preferentially at steps or disloca(cid:173)
`tions on the crystalline surface. Experimental observations
`of surface roughness or non uniformities could be explain(cid:173)
`able in terms of this type of attack.
`
`CONCLUSIONS
`
`Fluorine atoms etch silicon with a rate R 1'151)
`= 2.91 ± 0.20X 1O- 12T !!2 npe OI08lkTA/min in the ab(cid:173)
`sence of a plasma. This etching is accompanied by a chemilu(cid:173)
`minescent continuum centered at 632 nm. The intensity of
`the luminescence exhibits the same activation energy as the
`etch process, suggesting that the rate-limiting step for both
`processes is the formation of SiF 2 per reactions (16).
`The luminescence accompanying silicon etching by a
`gas consisting of F atoms and F 2 is closely described by the
`kinetics of reactions (l}-(6), and (11).
`A detailed mechanism has been proposed for the etch(cid:173)
`ing of silicon by fluorine atoms. In this model a layer of SiF 2
`groups are chemically bound to the silicon surface. The rate(cid:173)
`limiting reaction of impinging fluorine atoms with Si-SiF2
`bonds of the chemisorbed layer controls the gasification ofSi
`as SiF2 and higher fluorinated species.
`The previously reported etch rate, and reaction prob(cid:173)
`ability for the etching ofSi02 by F atoms are high by a con(cid:173)
`stant factor of - 1.5. Considering all results, the etch rate for
`the etching of Si02 is R FlSiO,) = 6.14 ± 0.49 X 10
`I:ln!
`T 1/2e
`0.10.1/1<"1 A/min. The ratio of Si to Si0 2 etching by F
`atoms is 4.74 ± 0.4geOO
`55/kr.
`The present rates and activation energy are consistent
`with those reported for in situ etching of Si and Si02 in F
`atoms containing plasmas at 0.3-0.5 Torr (CF4-02 , SiF4 -O},
`SiFh-02 , NFi 5
`; consequently the F atom solid reaction
`alone can generally account for these data. It appears that
`ion or electron bombardment does not play an essential role
`in the etching ofSi or Si02 by F-atom-containing plasmas in
`the few-tenths of a Torr pressure range, consistent with the
`isotropic nature of these etchants.
`
`ACKNOWLEDGMENTS
`
`We gratefully acknowledge helpful discussions with M.
`J. Cardillo and T. M. Duncan.
`
`I(a) V. M. Donnelly, D. L. Flamm,J. Appl. Phys.lO, 5273 (1980); (b)V. M.
`Donnelly, D. L. Flamm, "Optical Emission From Transient Species in
`Halocarbon and Fluorosilicon Plasmas," Extended Abstracts, 157th
`Meeting, Electrochemical Society (St. Louis, May, 1980), Vol. 80-1, p. 323;
`(c) V. M. Donnelly, D. L. Flamm, "Studies of Chemiluminescence Accom(cid:173)
`panying Silicon Etching by F-Atoms," Proceedings of the 88th National
`Meeting, Amer. Inst. Chern. Engrs. (Philadelphia, June, 1980); (d) C. I. M.
`Beenakker, J. H. J. van Dommelen, and J. Dieleman, "Origin of the LumI(cid:173)
`nescence Produced By the Reaction of Fluorine Atoms with Silicon,"
`
`3638
`
`J. Appl. Phys .. Vol. 52, No.5, May 1981
`
`Flamm, Donnelly, and Mucha
`
`3638
`
`Page 6 of 7
`
`

`

`Extended Abstracts, 157th Meeting, Electrochemical Society (St. Louis,
`May, 1980), Vol. 80-1, p. 330.
`'(a) D. L. Flamm, C. J. Mogab, and E. R. Sklaver, J. App\. Phys. 50, 624
`(1979); (b) D. L. Flam'll, Solid State Techno!. 22(9),109 (1979).
`Ie. J. Mogab, A. e. Adams, and D. L. Flamm J. Appl. Phys. 49,3796
`(1978).
`'(a) A. K. Kuriakose, J. L. Margrave, J. Phys. Chern. 68, 2671 (1964); (b) M.
`Chen, V. J. Minkiewicz, and K. Lee, J. Electrochem. Soc. 126, 1946 (1979);
`(c) 1. A. Mucha, V. M. Donnelly, and D. L. Flamm, unpublished results
`(1980).
`'I t should be noted that the quenching kinetics of excited SiF 3 are neglected
`in this simple expression. For the present range ofF-atom concentrations
`at constant pressure and temperature this appears to be a useful approxi(cid:173)
`mation, consistent with the scope of the experiment.
`6G. Smolinsky, D. L. Flamm, 1. Appl. Phys. 50, 4982 (1979).
`7p. L. Timms, R. A. Kent, T. C. Ehlert, 1. L. Margrave, 1. Am. Chern. Soc.
`87,2824 (1965).
`"T. J. Chuang, J. Appl. Phys. 51, 2614 (1980).
`"T. J. Chuang, Phys. Rev. Lett. 42,815 (1979).
`to A. Gelb, S. K. Kim, 1. Chern. Phys. 55,4935 (1971); T. L. Hill, Statistical
`
`Mechanics (McGraw-Hill, N.Y., 1956), Chap. 7.
`"1. W. Coburn, H. F. Winters, 1. Appl. Phys. 50, 3189(1979).
`I'M. J. Vasile, private communication, June (1980).
`IlH. F. Dylla, 1. G. King, M. S. Cardillo, Surface Sci. 74,141 (1978).
`14T. A. Delchar, F. e. Tompkins, Proc. Roy. Soc. A300, 141 (1967); F. S.
`Ham, ibid, 155 (1967).
`IS(a) E.P.G.T. van de Van, P. A. Zijlstra, "A Critical Comparison of
`SiF.IO, and CF.IO, As Plasma Etching Gases," "Extended Abstracts,
`157th Meeting, Electrochemical Society (St. Louis, May, 1980), Vol. 80-\,
`p. 253; (b) R. Horwath, e. B. Zarowin, and R. Rosenberg, "Characteriza(cid:173)
`tion of a High Pressure, Radial Flow, Plasma Etch Reactor for Silicon
`Etching in a CF4 Plasma," "Extended Abstracts, 157th Meeting, Electro(cid:173)
`chemical Society (St. Louis, May, 1980), Vol. 80-1, p. 294; (c) e. M. MeI(cid:173)
`liar-Smith and C. J. Mogab, in Thin Film Processes, edited by J. L. Vossen,
`W. Kern (Academic, New York, 1979).
`16More recent measurements by Vasile, using improved apparatus and tech(cid:173)
`niques, show a large fraction of lower fluorides (i.e., SiF ,) are desorbed.
`This is also consistent with our recent study of the SiF, and (F,F,) reac(cid:173)
`tion, but disagrees with the findings of