Remote plasma etching of silicon nitride and silicon dioxide
`using NF3/O2 gas mixtures
`B. E. E. Kastenmeier,a) P. J. Matsuo, and G. S. Oehrleinb)
`Department of Physics, The University at Albany, State University of New York, Albany, New York 12222
`J. G. Langan
`Air Products and Chemicals, Inc., Allentown, Pennsylvania 18195
`共Received 8 May 1997; accepted 2 January 1998兲
`The etching of silicon nitride (Si3N4) and silicon dioxide (SiO2) in the afterglow of NF3 and
`NF3 /O2 microwave discharges has been characterized. The etch rates of both materials increase
`approximately linearly with the flow of NF3 due to the increased availability of F atoms. The etch
`rate of Si3N4 is enhanced significantly upon O2 injection into the NF3 discharge for O2 /NF3 ratios
`of 0.3 and higher, whereas the SiO2 etch rate is less influenced for the same flow ratios. X-ray
`photoelectron spectroscopy of processed Si3N4 samples shows that the fluorine content of the
`reactive layer, which forms on the Si3N4 surface during etching, decreases with the flow of O2, and
`instead oxidation and nitrogen depletion of the surface occur. The oxidation of the reactive layer
`follows the same dependence on the flow of O2 as the etch rate. Argon actinometry and quadrupole
`mass spectrometry are used to identify reactive species in the etching of both materials. The atomic
`fluorine density decreases due to dilution as O2 is added to the discharge. The mass spectrometer did
`not detect NFx species (x⫽1 – 3) at any discharge parameter setting, which indicates the near
`complete dissociation of NF3. Nitric oxide 共NO兲 was detected by mass spectrometry, and the NO
`density shows the same dependence on O2 flow as the Si3N4 etch rate and the surface oxidation.
`Based on this observation, we propose that the etch rate enhancement for Si3N4 is due to the
`adsorption of the NO on the Si3N4 surface, followed by the formation of N2 with a N atom from the
`surface. The O atom can then attach to the same surface site, contributing to the oxidation. © 1998
`American Vacuum Society. 关S0734-2101共98兲00604-6兴
`
`I. INTRODUCTION
`
`The minimization of feature sizes forces the semiconduc-
`tor industry to constantly improve fabrication processes. For
`example, ion induced damage to oxide layers is not accept-
`able as the gate oxide thickness approaches 50 Å or less.
`Therefore, mask materials are increasingly stripped down-
`stream from a remote plasma source, avoiding the bombard-
`ment of the surface with energetic ions, which is typical for
`a direct plasma process. Also, reactors for chemical vapor
`deposition need to be cleaned periodically in order to ensure
`a constant high quality of the thin films deposited.1 Cur-
`rently,
`plasma
`enhanced
`chemical
`vapor
`deposition
`共PECVD兲 chambers are often cleaned in situ, which can re-
`sult in damage to chamber parts because of the presence of
`both fluorine and ion bombardment on electrodes. Low-
`pressure chemical vapor deposition 共LPCVD兲 tubes are
`cleaned using a wet chemistry, e.g., hydrofluoric acid for the
`cleaning of LPCVD Si and Si3N4 tubes.
`A procedure that minimizes tool downtime and chamber
`damage, avoids the disposal of wet chemicals, and poten-
`tially enables a higher level of cleanliness, is remote plasma
`cleaning. The reactive afterglow of etching gases can be used
`to strip deposited layers off reactor walls and to clean the
`
`a兲Electronic mail: bk7752@csc.albany.edu
`b兲Electronic mail: oehrlein@cnsibm.albany.edu
`J. Vac. Sci. Technol. A 16„4…, Jul/Aug 1998
`
`2047
`
`chamber. This method is applicable in CVD reactors for Si,
`SiO2, Si3N4, and tungsten compounds.1–3
`The etching characteristics of fluorocarbon gases like CF4
`and C2F6 have been widely studied. These gases are used for
`reactor cleaning, but since etching often occurs together with
`the formation of an undesired fluorocarbon polymer layer,
`they require the addition of O2. A clean alternative to those
`gases is nitrogen trifluoride, NF3, and mixtures of NF3 with
`O2. Discharges of NF3 are not polymerizing, and thus a good
`choice for cleaning applications. Nitrogen trifluoride is envi-
`ronmentally preferable to CF4 and C2F6 because it has a
`shorter atmospheric lifetime.4 Another advantage of NF3
`over fluorocarbon gases is that the dissociation of NF3 in a
`discharge can approach 100%, resulting in higher F atom
`concentrations and higher etch yields as compared to fluoro-
`carbon gases.
`In previous publications5–7 the etching of Si3N4 and SiO2
`in remote CF4 discharges with O2 and N2 additions has been
`examined. It was found that the etch rate of Si3N4 is strongly
`enhanced when both O2 and N2 are added to the CF4 dis-
`charge, but the SiO2 etch rate remains unchanged. A linear
`correlation between the Si3N4 etch rate and the density of
`NO was observed,5,7 and Blain et al.6 suggested three models
`for the chemical effect of the NO on the nitride surface, all
`incorporating enhanced removal of the nitrogen. Surface ef-
`fects of the NO molecule could also be observed for silicon
`etching.8 The thickness of the reactive layer that forms on the
`0734-2101/98/16„4…/2047/10/$15.00
`
`©1998 American Vacuum Society
`
`2047
`
`TSMC 1326
`
`

`

`2048
`
`Kastenmeier etal.: Remote plasma etching of Si3N4 and SiO2
`
`2048
`
`FIG. 1. Schematic of the chemical downstream etcher used in this investi-
`gation. The gases are fed into the sapphire applicator, where a microwave
`discharge is ignited. The species effluent from the plasma travel through
`tubing of variable length and lining material to the reactive chamber. The
`sample is placed on the center of an electrostatic chuck. A quadrupole mass
`spectrometer is mounted on the chamber on top of the sample, and mono-
`chromatic ellipsometry is used to determine etch rates.
`
`crystalline Si during etching is reduced when NO is present,
`leading to an enhanced etch rate.
`This article characterizes the etching of Si3N4 and SiO2 in
`the afterglow of NF3 /O2 microwave discharges. Etch rates
`are reported as a function of NF3 flow and gas composition,
`and the reported etch rates are explained by the generation
`rate of active species, determined by optical emission acti-
`nometry and mass spectrometry. Furthermore, the etching
`mechanism of Si3N4 in the presence of fluorine and NO in
`the gas phase is investigated in more detail by angular re-
`solved x-ray photoelectron spectroscopy 共XPS兲. Another
`article9 will characterize the etching of polycrystalline Si in
`NF3 /O2 mixtures.
`
`II. EXPERIMENT
`Figure 1 shows a schematic of the apparatus used for the
`experiments. Nitrogen trifluoride and mixtures of NF3 and
`O2 are excited using an Astex 2.45 GHz microwave applica-
`tor with a sapphire coupling tube. The pressure for all ex-
`periments was 1000 mTorr. The microwave power was var-
`ied from 600 to 1400 W, the flow of NF3 from 50 to 500
`sccm. All experiments involving O2 were conducted at a mi-
`crowave power level of 1400 W, with a constant flow of NF3
`of either 300 or 500 sccm. A fiberoptic cable for optical
`emission experiments of the discharge is mounted on the
`housing of the applicator. The spectrograph used in this in-
`vestigation is a 30 cm optical multichannel analyzer 共EG&G
`PAR Model 1470兲 which covers the spectrum between 250
`and 850 nm. The species produced in the plasma travel
`through a transport tube to the cylindrical reaction chamber.
`The length, geometry, and lining material of the transport
`tube can be varied. Samples of size 1 in.⫻1 in. are glued on
`a 5 in. carrier wafer, which is placed on an electrostatic
`chuck in the reaction chamber. The materials used for this
`investigation are LPCVD Si3N4 and thermally grown SiO2.
`The temperature of the sample is monitored with a fluoroptic
`probe which contacts the backside of the sample. It was kept
`constant at 10 °C for all experiments. A pressure of 5 Torr of
`helium was maintained between the surface of the electro-
`
`J. Vac. Sci. Technol. A, Vol. 16, No. 4, Jul/Aug 1998
`
`FIG. 2. The etch rate of Si3N4, SiO2, and polycrystalline silicon as a func-
`tion of the flow of NF3. The measurements were performed at a constant
`pressure of 1000 mTorr and with three different microwave power leads.
`The etch rate roughly increases linearly with the flow due to the increasing
`availability of reactive species.
`
`static chuck and the carrier wafer in order to obtain good
`heat conduction. Etch rates are measured in situ by mono-
`chromatic ellipsometry 共wavelength 632.8 nm兲. A quadru-
`pole mass spectrometer 共Leybold Inficon Transpector兲 is
`mounted on top of the reaction chamber such that the dis-
`tance from the orifice to the discharge is the same as that
`from the sample to the discharge. The ionization region of
`the mass spectrometer is in line of sight with the sampling
`orifice and the reaction chamber. The energy of the ionizing
`electrons is 35 eV. The pressure in the mass spectrometer
`during an experiment is around 1⫻10⫺6 Torr. The reaction
`chamber is connected to an ultrahigh vacuum 共UHV兲 wafer
`handling system which allows the samples to be moved to a
`multi-technique surface analysis chamber without exposure
`to air.
`
`III. RESULTS
`A. Etch rates
`The etch rates of Si3N4 and SiO2 were measured as a
`function of the flow of NF3 共see Fig. 2兲. The pressure was
`kept constant at 1000 mTorr, and the parameter for the
`curves in Fig. 2 is the microwave power. The flow range in
`which it was possible to obtain stable discharges depended
`on the power. At 600 W, for example, a stable discharge
`could be obtained only for 50 sccm of NF3, whereas at 1400
`W the flow of NF3 could be varied across the whole range
`permitted by the mass flow controller. The etch rates of all
`three materials increase linearly with the flow of NF3. Since
`all curves coincide, microwave power does not influence the
`etch rate significantly. The SiO2 etch rates, however, grow
`faster with the flow of NF3 than the corresponding Si3N4 etch
`rates, their slopes being greater by a factor of more than 2.
`
`

`

`2049
`
`Kastenmeier etal.: Remote plasma etching of Si3N4 and SiO2
`
`2049
`
`FIG. 4. The etch rate of silicon dioxide as a function of the flow ratio
`O2 /NF3.The etch rate decreases as oxygen is added to a flow of 300 sccm of
`NF3. If a higher NF3 flow is used 共500 sccm兲, the etch rate remains on a
`constant level up to a ratio of 0.4, and then increases slightly upon further
`increase of the flow ratio.
`
`power 共⬍10 W兲, the etch rates assumed regular values. Mass
`spectrometry measurements which were performed with an
`untuned discharge 共200 W reflected power兲, showed a higher
`NO signal
`than a tuned discharge and the presence of
`NFx (x⫽1,2) species in the reaction chamber. It is likely that
`the high NO density is responsible for the fast Si3N4 etch
`rate. The SiO2 etch rate is not influenced by the NO density.
`It is possible that NFx species enhance the etching of SiO2
`under these conditions.
`
`FIG. 3. The etch rate of silicon nitride vs the flow ratio of O2 and NF3. The
`effect of oxygen addition is most pronounced for a high flow of NF3 共500
`sccm兲.
`
`The etch rate of polycrystalline Si is proportional to the
`density of atomic F if no significant oxidation of the silicon
`surface occurs.8,10 Therefore, the F density can be calculated
`from the Si etch rate and published rate constants.11 The etch
`rate of polycrystalline silicon is shown on the bottom panel
`of Fig. 2. It is also proportional to the NF3 flow, and higher
`than the Si3N4 etch rate by a factor of 30. This comparison
`shows that F atoms, the primary etchants for Si and Si3N4,
`are available in abundance to sustain the etching of Si3N4,
`and that the etch rate of Si3N4 is not limited by the density of
`atomic F.
`Oxygen addition to a NF3 discharge strongly enhances the
`Si3N4 etch rates. Figure 3 shows the etch rates of Si3N4 as a
`function of the ratio O2 /NF3. Pressure and microwave power
`were kept constant at 1000 mTorr and 1400 W, respectively.
`The flow of NF3 was fixed at either 300 or 500 sccm. A
`small amount of oxygen increases the etch rate by a factor of
`2 for a NF3 flow of 500 sccm, and by a factor of 4.3 in the
`case of 300 sccm of NF3. As the flow of O2 is increased
`further 共up to O2 /NF3⫽0.3), the etch rates remain constant,
`and etch rates for 300 and 500 sccm of NF3 are identical. A
`significant difference is observed for high flows of oxygen
`(O2 /NF3⬎0.3). The etch rate for the low flow of NF3 re-
`mains on the same low level, whereas the etch rate for the
`high NF3 flow increases continuously until it saturates near
`O2 /NF3⫽1.
`Oxygen addition to a discharge of 300 or 500 sccm of
`NF3 does not affect the silicon dioxide etch rates as strongly
`as it does the silicon nitride etch rates. Figure 4 shows the
`etch rates of SiO2 for ratios O2 /NF3⫽0 to 1. In the case of
`500 sccm of NF3, the etch rate remains almost constant up to
`ratios O2 /NF3⫽0.5, and then increases slightly. If oxygen is
`added to a lower flow of NF3 共300 sccm兲, the etch rate of
`SiO2 actually decreases.
`For certain unstable plasma conditions, etch rates of Si3N4
`and SiO2 were found to be abnormally high. These condi-
`tions often occurred during the tuning of the discharge. After
`the discharge was tuned to a stable state with low reflected
`
`JVST A - Vacuum, Surfaces, and Films
`
`B. Optical emission and actinometry measurements
`We performed actinometry measurements to monitor the
`production of atomic fluorine and oxygen in the discharge as
`a function of the flow O2 in NF3. Actinometry with argon as
`a tracer gas has been widely used to determine the relative F
`atom density in CF4 /O2 and SF6 /O2 systems.12–15 The
`method has been validated by Donnelly et al.16 for the after-
`glow of CF4 /O2 and NF3 /Ar systems. The ground state den-
`sity of a species X in the discharge, nX , is proportional to the
`density of Ar and the ratio of the emission intensities17
`nX⬀nAr* IX /IAr ,
`共1兲
`where the Ar density is determined from the total gas density
`ntot and the gas flows
`QAr
`⫹QO2⫹QAr
`
`QNF3
`Typically, the intensity of the Ar 4s⬘关1/2兴0⫺4p⬘关1/2兴
`emission at 750.4 nm, whose upper level has an excitation
`energy of 13.48 eV, is set in relation with the F 3s2P
`⫺3p2P0 emission at 703.7 nm 共14.76 eV兲, or the O 3s3S0
`⫺3p3P triplet at 844.6 nm 共10.99 eV兲. In our experiments,
`however,
`the Ar 共750.4 nm兲 line overlapped with other
`peaks, and the numerical determination of the intensity is
`very likely to have a systematic error. In order to eliminate
`this error, we included the Ar 共763.5 nm, 4s关1 1/2兴0
`
`nAr⫽ntot⫽
`
`.
`
`共2兲
`
`

`

`2050
`
`Kastenmeier etal.: Remote plasma etching of Si3N4 and SiO2
`
`2050
`
`FIG. 5. The relative changes of the O atom concentration as determined by
`Ar actinometry. The flow of NF3 was kept constant at 300 and 500 sccm,
`respectively. As expected, the density of O atoms in the plasma region
`grows with the flow of oxygen. The production of atomic oxygen is higher
`by a factor of 1.9 for the low flow of NF3.
`
`⫺4p关1 1/2兴) and Ar 共811.5 nm, 4s关1 1/2兴0⫺4p关2 1/2兴)
`lines into our analysis. The energies of the upper levels of
`these emissions, 13.17 and 13.08 eV, are very close to that of
`the Ar 共750.4 nm兲 line. The determination of the emission
`intensities of these two emissions is straightforward, since no
`overlap with other emission lines occurs. In all our experi-
`ments, we find that the F/Ar 共763.5 nm兲 and F/Ar 共811.5 nm兲
`ratios have the exact same dependency on the O2 flow. We
`decided to use the Ar emission at 811.5 nm as actinometer
`for all graphs shown here. As discussed recently by Petrovic
`et al.18 and by Malyshev and Donnelly,19 cascading from
`metastable states into the 4p关2 1/2兴 level may contribute
`significantly to the Ar 811.5 nm emission, whereas the con-
`tribution to the Ar 750.5 nm emission is negligible. All
`changes in the plasma affecting the Ar metastable density
`will also affect the emission intensity of the Ar 811.5 nm
`line. This effect is not taken into account in our analysis.
`However, we estimate the error introduced by using the Ar
`811.5 nm line to be 15% or less. This estimate is based on a
`comparison of Ar 750.4 and 811.5 nm emission intensities
`from a CF4 /O2 /Ar microwave plasma ignited in the same
`applicator under similar conditions. The Ar emission lines
`from this gas mixture are free of overlap. The normalized Ar
`750.4 and 811.5 nm emission intensities, as a function of O2
`flow, vary by a maximum of 15% in the CF4 /O2 /Ar system.
`This is reflected by the error bars in Fig. 5. Furthermore, the
`Ar 811.5 nm emission deviates from the dilution curve no
`more than 10% and in a nonsystematic way, which supports
`the notion that the metastable contribution does not depend
`on the O2 flow.
`Walkup et al.20 have compared Ar actinometry of O in
`CF4 /O2 discharges with two photon laser induced fluores-
`cence measurements. They found that the O 3s3S0⫺3p3P
`triplet at 844.6 nm yields more reliable results for the ground
`state O atom density than the 3s5S0⫺3p5P triplet at 777
`⫹ molecules can
`nm, since dissociative recombination of O2
`significantly contribute to the population of the 3p5P level
`and the emission at 777 nm.
`In Fig. 5, the relative change of the O atom concentration
`
`J. Vac. Sci. Technol. A, Vol. 16, No. 4, Jul/Aug 1998
`
`FIG. 6. The relative changes of the F atom concentration as determined by
`Ar actinometry. The same experimental parameters were used as in Fig. 5.
`The dashed lines indicate the calculated density of F under the assumption
`that dilution is the only effect of O2 addition to NF3. At both flows of NF3,
`the production of F atoms is not enhanced significantly. The F concentration
`decreases due to dilution. This is in contrast to CF4 /O2 microwave plasmas,
`and also to NF3 /O2 low density plasmas.
`
`in the discharge region is shown as a function of the ratio
`O2 /NF3. As one expects, the density of O atoms increases
`with the flow of O2. However, in the case of 300 sccm of
`NF3, the O atom concentration grows faster than in the case
`of 500 sccm of NF3. The initial slopes of the curves differ by
`a factor of 1.9. At a fixed flow of 300 sccm of O2 共designated
`by crosses in Fig. 5兲, the density of ground state oxygen
`atoms is higher by a factor of 2.9 for the low flow of NF3.
`Figure 6 shows the behavior of the F atom concentration
`in the plasma as a function of oxygen addition. The dashed
`lines show the predicted F density under the assumption that
`dilution of gas phase species is the only effect on the pro-
`duction of F as O2 is added to the discharge. The concentra-
`tion of F decreases with the addition of O2. However, Fig. 6
`indicates that the decrease of the F density for the case of
`500 sccm of NF3 is less than predicted by the dilution effect
`by a margin significantly greater than the error. This indi-
`cates that the production rate of F atoms is slightly increased
`by the presence of oxygen in the discharge, but the total
`density decreases due to dilution.
`In order to gain information about the chemical effects of
`the O2 in the discharge, relative changes of the emission
`intensities from N2 and NF were determined. Oxygen atoms
`can be expected to quickly oxidize the lower fluorides of
`NF3,21,22 leading to a reduced density of NFx (x⫽1,2) in the
`plasma. The production of N2 is likely to decrease in favor of
`the generation of oxides of nitrogen.22 The b 1兺 ⫹⫺X 3兺 ⫺
`system of NF at 528.8 nm and the C 3兿 u⫺B 3兿 g system of
`N2 at 357.7 nm could be detected. The emission intensity of
`both N2 and NF in the plasma decreases more strongly than
`just due to the dilution effect 共see Fig. 7兲, and the emission
`from NF vanishes at O2 /NF3⫽2. This indicates the presence
`of chemical reactions of those species with O or O2.
`Optical emission from other species, e.g., F2, N, NO, or
`NO2, could not be identified in the present work. This is
`23
`consistent with the observation of other researchers for NF3
`
`

`

`2051
`
`Kastenmeier etal.: Remote plasma etching of Si3N4 and SiO2
`
`2051
`
`FIG. 7. The relative changes of the N2 and NF emission intensities from the
`NF3 /O2 discharge. The same experimental parameters were used as in Fig.
`5. Again, the dashed line indicates the dilution by O2. At both flows of NF3,
`the density of both species is reduced more than just by dilution alone.
`
`and NF3 /Ar16 discharges. Atomic nitrogen and F2 are sup-
`pressed by fast reactions with NFx species16
`N⫹NFx→NF⫹NFx⫺1 ,
`共3兲
`F2⫹NFx⫺1→NFx⫹F,
`共4兲
`therefore their emission intensity is below the detection limit
`of our spectrograph. So far, no explanation for the absence of
`emission from NO can be given.
`C. Mass spectrometry measurements
`We previously applied mass spectrometry to determine
`the relative changes in the concentration of reactive species
`in chemical dry etching.7 As in our previous work, we mea-
`sured the intensities of the species with and without a dis-
`charge ignited. The data is then plotted as the difference
`between the plasma-on and the plasma-off state, which rep-
`resents the production and dissociation of species in the
`plasma better than the approach where the plasma-on state
`only is measured.
`An analog spectrum of the afterglow of a microwave dis-
`charge in pure NF3 is shown in panel 共a兲 of Fig. 8, together
`with the spectrum obtained for no discharge ignited. This
`spectrum shows the NF3 peak at mass number 71 and the
`cracking products NF and NF2. Background signals of O2
`and N2 are also visible. These peaks disappear completely, as
`a discharge is ignited with 1400 W of microwave power.
`Instead, F, F2, and N2 are produced. SiF3 at 85 amu appears
`as a cracking product of SiF4, the product of etching reac-
`tions of quartz windows in the reactor. The difference spec-
`trum IPlasmaOn⫺IPlasmaOff is shown in panel 共b兲. The dissocia-
`tion of species, like NFx , is represented as a negative peak.
`Generation of species, like N2, F2, and the F radical leads to
`positive peaks in the difference spectrum. Panel 共c兲 of Fig. 8
`shows the difference spectrum obtained from a NF3/O2 mix-
`ture. Nitric oxide is produced in the discharge, and oxygen is
`visible as a negative peak.
`Figure 9 shows the normalized density of F2 and the in-
`
`JVST A - Vacuum, Surfaces, and Films
`
`FIG. 8. Typical mass spectra sampled from the downstream reactive cham-
`ber. For the top panel pure NF3 is used. The top panel shows analog spectra
`for no discharge and for the microwave discharge ignited, together with the
`difference spectrum. The NFx (x⫽1,2,3) peaks disappear completely as the
`discharge is ignited. Therefore they appear as negative peaks in the differ-
`ence spectrum. N2, F2, and F radicals are the main products of the NF3
`discharge. The bottom panel contains the difference spectrum of a NF3 /O2
`mixture showing NO production.
`
`tensity of the peak at 19 amu as a function of O2 addition to
`500 sccm of NF3. The density of F2 decreases with increas-
`ing flow of O2. The dashed line in Fig. 9 is the dilution
`curve. The peak at 19 amu is due to atomic fluorine and
`electron impact dissociation of F2 in the ionization region of
`the mass spectrometer. The data shown are not corrected for
`this effect, since an estimate for the F2 contribution was not
`available.
`It is known from previous work7 that the etch rate of
`Si3N4 is proportional to the density of NO in the reaction
`chamber. Figure 10 shows the normalized NO density down-
`
`FIG. 9. The intensity difference between the plasma-on and plasma-off val-
`ues, IPlasmaOn⫺IPlasmaOff , for the 19F and the 38F2 peaks. Both difference
`values decrease as the flow of O2 in NF3 is increased from a ratio O2 /NF3
`⫽0 to O2 /NF3⫽2.
`
`

`

`2052
`
`Kastenmeier etal.: Remote plasma etching of Si3N4 and SiO2
`
`2052
`
`FIG. 10. IPlasmaOn⫺IPlasmaOff for the 30NO peak at two different flows of NF3.
`Both curves have been normalized with their common maximum value,
`which occurred for the high NF3 flow. The two distinctively different Si3N4
`etch rate curves from Fig. 3 are well mirrored in the 30NO peak behavior.
`
`stream from the plasma for 300 and 500 sccm of NF3. The
`two curves show significantly different behavior, and both
`show a strong similarity to the Si3N4 etch rate curves for the
`same parameters from Fig. 3. In the case of 500 sccm of
`NF3, the NO density remains on a constant level for ratios
`O2 /NF3⬍0.5, and then increases strongly with the flow of
`O2. At 300 sccm of NF3, the NO density generally is signifi-
`cantly lower than at the high flow of NF3. It remains on a
`constant level up to O2 /NF3⫽0.6, and then grows steadily.
`The normalized change in the concentration of O2 is
`shown in Fig. 11. In previous work on the chemical down-
`stream etching with CF4 /O2 /N2 gas mixtures, a significant
`amount of oxygen atoms could be detected in the reaction
`chamber by the mass spectrometer. The common existence
`of O and NO leads to the appearance of the yellow-greenish
`air afterglow24–26 in the CF4 /O2 /N2 system. No atomic oxy-
`gen can be detected downstream from a NF3 /O2 discharge.
`The difference of the plasma-on and plasma-off intensities of
`the peak at 16 amu follows that of the parent molecule O2.
`
`FIG. 11. IPlasmaOn⫺IPlasmaOff for the 32O2 peak. O2 is destroyed in the dis-
`charge, thus the plasma-off intensity is higher than the plasma-on intensity,
`and the difference is in the negative. The 16O value follows the 32O2 value,
`indicating that the major contribution to the 16O intensity comes from O2
`split up in the ionization region of the mass spectrometer, not from O radi-
`cals produced in the discharge. The reaction chamber is free of atomic O.
`
`J. Vac. Sci. Technol. A, Vol. 16, No. 4, Jul/Aug 1998
`
`FIG. 12. The corrected peak intensity ratio of the F(1s) and the Si(2p)
`photoelectron emissions. The electron emission angle was 75° with respect
`to the surface normal. The fluorine content in the reactive layer initially
`decreases with the flow of O2, then remains on a constant level.
`
`Therefore, one is led to the conclusion that the atomic oxy-
`gen signal is exclusively due to O2 split up in the ionization
`chamber of the mass spectrometer, and that the afterglow of
`a NF3 /O2 discharge essentially contains no atomic oxygen.
`
`D. Surface analysis results
`X-ray photoelectron spectra were obtained from Si3N4
`and SiO2 samples immediately after processing. The electron
`emission angle for the element ratio measurements shown
`here was 75° with respect to the surface normal. The effec-
`tive electron escape depth under this angle, assuming an in-
`elastic mean free path of 20 Å, is only 5 Å. Thus, the XPS
`measurements are extremely surface sensitive. The areas of
`the different elemental peaks were corrected for different
`photoionization cross sections and the detector response.
`Figure 12 shows the fluorination of the Si3N4 surface after
`processing under conditions for which we have found the
`most pronounced effects of oxygen addition on the etch rate.
`These conditions are a flow of 500 sccm of NF3, a micro-
`wave power of 1400 W, and a chamber pressure of 1000
`mTorr. The fluorination of the surface decreases for O2 /NF3
`ratios up to 0.5, then remains on a constant level.
`The oxidation of the surface layer and the nitrogen con-
`tent are shown in Fig. 13 as the corrected ratio of
`O(1s)/Si(2p) and N(1s)/Si(2p) emission intensities. The
`surface oxidation initially is on a fairly constant level for
`O2 /NF3 ratios up to 0.5, then increases by a factor of 4 as
`O2 /NF3 approaches 1. The surface oxidation does not in-
`crease further, as the flow of O2 in NF3 is increased to a ratio
`of 2. The amount of surface oxidation is very similar to the
`Si3N4 etch rates and the NO concentration in the reaction
`chamber for the same parameters. The surface is depleted of
`N in the same way it is oxidized. A close correlation between
`surface oxidation and nitrogen depletion is shown by these
`data.
`The stoichiometry of the reactive layer can be deduced
`from Fig. 14. There, the corrected intensity ratios of emis-
`sions from O(1s) and F(1s) are shown. For all gas compo-
`sitions, F atoms are the dominant foreign species in the re-
`active layer. At low flows of O2, there are approximately ten
`
`

`

`2053
`
`Kastenmeier etal.: Remote plasma etching of Si3N4 and SiO2
`
`2053
`
`FIG. 15. The penetration depth of F and O atoms into Si3N4, and of F atoms
`into SiO2.
`
`FIG. 13. The corrected peak intensity ratios of the O(1s) and the N(1s) over
`the Si(2p) photoelectron emissions. The electron emission angle was 75°
`with respect to the surface normal. The surface is depleted of N atoms in the
`same way as it is oxidized. Both the depletion of N and the surface oxidation
`follow the behavior of the 30NO mass spectrometer signal for the same
`experimental parameters very well.
`
`times more F atoms in the surface than there are O atoms.
`However, as the flow of O2 is increased, the surface oxida-
`tion increases, and the corrected ratio of O and F emission
`intensities increases to 0.5.
`The thickness of the reactive layer which forms on the
`bulk silicon nitride during the etching has been determined
`by angular
`resolved XPS 共ARXPS兲. The conventional
`method27 to determine the overlayer thickness on a silicon
`substrate cannot be applied to reactive layers on Si3N4 and
`SiO2 films. The conventional method is based on the chemi-
`cal shift that Si(2p) core electrons suffer as Si–Si bonds are
`successively replaced with bonds with more electronegative
`atoms. The chemical shift of the Si(2p) emission from the
`overlayer on a crystalline or polycrystalline silicon substrate
`is high enough to be resolved by XPS from the unshifted
`emission from the bulk. However, for Si3N4 the binding en-
`ergy of a 2p electron from a bulk Si atom is 104 eV, and the
`binding energies of Si(2p) electrons from the reacted layer
`
`FIG. 14. The corrected peak intensity ratio of the O(1s) and the F(1s)
`photoelectron emissions. The electron emission angle was 75° with respect
`to the surface normal. The stoichiometry of the reactive layer surface can be
`deduced from this graph. Fluorine atoms are the dominant foreign species
`共ratios ⬍1兲. The oxidation increases strongly as the flow ratio of O2 in NF3
`is increased beyond 0.4. Thus, high etch rates in the case of Si3N4 etching
`can be achieved even with higher surface oxidation.
`
`JVST A - Vacuum, Surfaces, and Films
`
`are too close to this value to be resolved by the XPS. In the
`case of SiO2, the contribution to the Si(2p) emission from
`oxidized Si atoms from the bulk and from Si atoms in the
`reactive layer cannot be resolved, since each fluorine–
`oxygen substitution shifts the energy level of the Si(2p) core
`electrons by merely 0.2–0.3 eV.28
`In ARXPS, the emission angle of electrons with respect to
`the surface normal, ␪, is changed by rotating the sample. The
`depth of origin of the photoelectrons is thereby changed from
`the very surface at ␪⫽75° to the order of the inelastic mean
`free path at ␪⫽0°. The values for the inelastic mean free
`paths of electrons originating from F(1s) or O(1s) are ob-
`tained from Briggs and Seah.29
`The intensity as a function of the electron escape angle,
`I共␪兲, of the F(1s) and O(1s)emissions was measured at six
`angles from 75° to 0° with respect to the surface normal. The
`peak intensities were then corrected for instrumental effects
`as a function of the angle. Correction factors for the different
`angles were obtained from the emission intensities of a ho-
`mogeneous semi-infinite SiO2 film. The measured and cor-
`rected I共␪兲 was then least-square fitted with the output of a
`for different
`single-
`simulation, which calculated I共␪兲
`parametric model assumptions about the decay of I(x), the
`foreign species density in the reactive layer as a function of
`the depth x in the sample. The three models used for this
`investigation are: 共a兲 one-step, 共b兲 exponential, and 共c兲 linear
`decrease of the foreign species density with increasing d. For
`a given model assumption for I(x), I(␪)is given by
`1
`⬁
`
`cos ␪冕
`
`I共␪兲⫽
`
`I共x 兲e ⫺x/␭ cos ␪dx.
`
`共5兲
`
`0
`Models 共a兲 and 共b兲 yield results for the penetration depth
`of the foreign species with a deviation of 0.1 nm or less. The
`linear decay model gives results consistently higher by a
`factor of 2. We have therefore chosen to report the results
`obtained with the exponential decay model. In Fig. 15, the
`penetration depths of F and O atoms into Si3N4 are shown as
`a function of the ratio of O2 in NF3. Fluorine penetrates the
`silicon nitride about 0.4 nm deep, with only a 25% variation
`around this value as the gas composition is varied. Oxygen
`atoms penetrate the Si3N4 much less than F. The penetration
`
`

`

`2054
`
`Kastenmeier etal.: Remote plasma etching of Si3N4 and SiO2
`
`2054
`
`depth for O2 /NF3 ratios up to 1 is between 0.1 and 0.2 nm.
`For higher flows of O2, the penetration depth of O atoms
`increases substantially. We measure a penetration depth of
`0.7 nm at O2 /NF3⫽2. Also included in Fig. 15 is the pen-
`etration depth of F atoms into SiO2. Fluorine penetrates SiO2
`about 50% deeper than it does Si3N4.
`
`IV. DISCUSSION
`A. Gas phase effects of O2
`Nitrogen trifluoride is easily dissociated by electron im-
`pact. The threshold energy for the first step, dissociative
`electron attachment
`e ⫺⫹NF3→NF2⫹F⫺
`共6兲
`is