study of the sio -to-sin etch selectivity mechanism in ... · lam research corporation, fremont,...

Study of the SiO 2-to-Si 3N4 etch selectivity mechanism in inductivelycoupled fluorocarbon plasmas and a comparisonwith the SiO 2-to-Si mechanism

M. Schaepkens, T. E. F. M. Standaert, N. R. Rueger, P. G. M. Sebel,a)

and G. S. Oehrleinb)

Department of Physics, University at Albany, State University of New York, Albany, New York 12222

J. M. CookLam Research Corporation, Fremont, California 94538-6470

~Received 21 July 1998; accepted 2 October 1998!

The mechanisms underlying selective etching of a SiO2 layer over a Si or Si3N4 underlayer, aprocess of vital importance to modern integrated circuit fabrication technology, has been studied.Selective etching of SiO2-to-Si3N4 in various inductively coupled fluorocarbon plasmas (CHF3,C2F6/C3F6, and C3F6/H2) was performed, and the results compared to selective SiO2-to-Si etching.A fluorocarbon film is present on the surfaces of all investigated substrate materials during steadystate etching conditions. A general trend is that the substrate etch rate is inversely proportional to thethickness of this fluorocarbon film. Oxide substrates are covered with a thin fluorocarbon film~,1.5nm! during steady-state etching and at sufficiently high self-bias voltages, the oxide etch rates arefound to be roughly independent of the feedgas chemistry. The fluorocarbon film thicknesses onsilicon, on the other hand, are strongly dependent on the feedgas chemistry and range from;2 to;7 nm in the investigated process regime. The fluorocarbon film thickness on nitride is found to beintermediate between the oxide and silicon cases. The fluorocarbon film thicknesses on nitride rangefrom ;1 to ;4 nm and the etch rates appear to be dependent on the feedgas chemistry only forspecific conditions. The differences in etching behavior of SiO2, Si3N4, and Si are suggested to berelated to a substrate-specific ability to consume carbon during etching reactions. Carbonconsumption affects the balance between fluorocarbon deposition and fluorocarbon etching, whichcontrols the fluorocarbon steady-state thickness and ultimately the substrate etching. ©1999American Vacuum Society.@S0734-2101~99!03201-7#

sO-n

rgden

ehen

eicursalic

nt

ctstchand

fing--g.

ereedem-

of

s

ri-that

I. INTRODUCTION

Etching of via or contact holes into SiO2 to make electri-cal contact with an underlayer is an indispensable procesmodern integrated circuit fabrication technology. High Si2

etch rate and selectivity of SiO2-to-Si are important requirements for etch processes to be commercially viable in mafacturing. Etch processes employing fluorocarbon dischaare typically able to meet these demands, as first reporteHeinecke1 and an extensive number of subsequstudies.2–12

It is believed that the primary mechanism for highly slective SiO2-to-Si etching using fluorocarbon plasmas is tselective formation of a relatively thick passivating film othe Si surface during steady-state etching conditions. Thetch rate in that situation is limited by the arrival of atomfluorine that needs to diffuse through the film to the Si sface, where it chemically reacts.13–16 For the same procesconditions SiO2 surfaces stay clean of fluorocarbon materiand are etched directly through a mechanism of chemsputtering.6,17,18

The ability to achieve selective etching of SiO2 overSi3N4 is becoming an increasingly important requireme

a!On leave from Eindhoven University of Technology.b!Electronic mail: [email protected]

26 J. Vac. Sci. Technol. A 17 „1…, Jan/Feb 1999 0734-2101/99

in

u-esbyt

-

Si

-

,al

.

Silicon nitride is used as a passivating layer that protecircuits from mechanical and chemical attack, or as an estop layer, enabling the fabrication of certain damasceneself-aligned contact~SAC! structures. Selective SiO2-to-Si3N4 etching has been demonstrated in several systems.19–24

Correlations between the Si3N4 etch rate and the amount ofluorocarbon material present on the surface during etchsuggest a SiO2-to-Si3N4 selectivity mechanism that is analogous to the SiO2-to-Si etching mechanism. A detailed comparison between the two mechanisms, however, is lackin

This work summarizes results obtained in a study whSiO2, Si3N4 and Si were processed in an inductively couplplasma source fed with various fluorocarbon feedgas chistries (CHF3, C2F6/C3F6 and C3F6/H2). Etch rates of SiO2,Si3N4, and poly-Si samples and surface modificationscrystalline Si samples were measured usingin situ ellipsom-etry. The surface chemistry of processed SiO2 and Si3N4

samples was examined using postplasmain situ x-ray photo-electron spectroscopy~XPS!. The experimental results allowa direct comparison of SiO2, Si3N4, and Si etch mechanismand therefore a comparison of the SiO2-to-Si3N4 andSiO2-to-Si etch selectivity mechanisms. From this compason it can be understood why certain feedgas chemistriesgive SiO2-to-Si selectivity do not necessarily give SiO2-to-Si3N4 selectivity. However, it has been found that SiO2-to-

26/17 „1…/26/12/$15.00 ©1999 American Vacuum Society

esh

fiech

rr

is

ure

ucyis0–

Crinthoo

r-e

bl

ts.at

antredndtly.

mo-of

asestedwithy an

heh ans is

ofs is

lus-UHVceopy

to-ple

ingave

om-onro-ar-O

n-ition.

the-

s-ied

bon

dive

asthe

rc

27 Schaepkens et al. : Study of the SiO 2-to-Si 3N4 etch selectivity mechanism 27

Si3N4 selectivity in an etching process also providSiO2-to-Si selectivity. More importantly, a trend in the etcrate behavior of the different materials has been identiallowing a general description of fluorocarbon plasma eting to be formulated.

II. EXPERIMENTAL SETUP

The high-density plasma source used in this work isradio-frequency inductively coupled plasma~ICP! source ofplanar coil design. This plasma source has also been refeto in the literature as transformer coupled plasma~TCP!25

and radio frequency induction~RFI!26 source. A schematicoutline of the used ICP reactor is shown in Fig. 1. Itsimilar to the one described by Kelleret al.27

The apparatus consists of an ultrahigh vacuum~UHV!compatible processing chamber in which the plasma soand a wafer holding electrostatic chuck are located. The cter part of the ICP source is a planar, 160-mm-diam indtion coil that is separated from the process chamber b19.6-mm-thick, 230-mm-diam quartz window. The coilpowered through a matching network by a 13.56 MHz,2000 W power supply.

Wafers with a diameter of 125 mm can be placed onbipolar electrostatic chuck during processing. The chucklocated at a distance of 7 cm downstream from the Isource and allows the wafer to be rf biased and cooled duprocessing. A helium pressure of 5 Torr is applied tobackside of the wafer during the experiment to achieve gthermal conduction between the wafer and the chuck.28 Thechuck is cooled by circulation of a cooling liquid.

Samples placed at the center of a wafer can be monitoby an in situ He–Ne~632.8 nm! rotating compensator ellipsometer~RCE! in a polarizer-compensator-sample-analyz~PCSA! configuration. Plasma diagnostics like a retractaLangmuir probe and optical emission spectroscopy~OES!can be used for plasma gas-phase characterization. Withretractable Langmuir probe it is possible to make a scanthe ion current density over 70% of the wafer at a distance2 cm above the wafer surface.

FIG. 1. Schematic outline of the experimental setup of the used ICP sou

JVST A - Vacuum, Surfaces, and Films

d-

a

ed

cen--a

aisPg

ed

ed

re

theofof

A variable frequency rf power supply~500 kHz–40 MHz,0–300 W! is used to bias the wafer for etching experimenThe experiments reported in this work were all performed3.4 MHz. In the process regime investigated, no significinfluence of rf biasing on the ion current density measuwith the Langmuir probe was observed. The ion energy athe ion flux to the surfaces can thus be varied independen

The process chamber is pumped using a 450 l/s turbolecular pump backed by a roughing pumpstack, consistinga roots blower and a mechanical pump. The process gare admitted into the reactor through a gas inlet ring locajust under the quartz window. The pressure is measureda capacitance manometer. Pressure control is achieved bautomatic throttle valve in the pump line.

In order to remove deposited fluorocarbon films from twalls of the process chamber, the chamber is cleaned witO2 plasma after each experiment. The cleaning procesmonitored by taking real time OES data. The absenceoptical emission of fluorocarbon related gas phase speciea measure for the cleanliness of the chamber.

The ICP apparatus is connected to a wafer handling cter system. Processed samples were transported underconditions to a Vacuum Generators ESCA Mk II surfaanalysis chamber for x-ray photoelectron spectrosc~XPS! analysis using a polychromatic MgKa x-ray source~1253.6 eV!. Photoelectron spectra were obtained at phoemission angles of 90° and 15° with respect to the samsurface.

III. EXPERIMENTAL RESULTS

A. Fluorocarbon deposition and etch rates

In fluorocarbon plasma processing, deposition and etchoccur simultaneously. Deposition and etching processes hdifferent dependencies on the energy of the ions that bbard the surface on which they occur. By varying the ibombardment energy one can therefore switch from fluocarbon deposition to etching of earlier deposited fluorocbon material or other substrate materials, such as Si2 ,Si3N4, or Si.

At low self-bias voltages, i.e., low ion bombardment eergy, these processes produce net fluorocarbon deposThe deposition characteristics are strongly dependent onfeedgas chemistry. Figure 2~a! shows the fluorocarbon deposition rates obtained as a function of feedgas (CHF3, C2F6,C3F6, and C3F6/H2) in discharges at 6 mTorr operating presure, 1400 W inductive power, and no rf bias power applto the surface. The deposition rate in CHF3 and C2F6 is sig-nificantly lower than in C3F6 or C3F6/H2. The highly poly-merizing character of C3F6 or C3F6/H2 discharges may berelated to the size and chemical structure of the fluorocarparent molecules.2,24

Figure 2~b! shows the refractive index of the depositefluorocarbon films using the different gases. The refractindex of the fluorocarbon deposited using C2F6 and C3F6 arefound to be lower than if CHF3 or C3F6/H2 is used. Crystal-line Si samples on which 150 nm fluorocarbon material wdeposited were analyzed by XPS. It was found that

e.

igr-aaa

ltsoac

to

etndta

inae

olat

deboi

nfon

g

t 0der

p-si-

isis

po-is

ce,

00

ison-thethatthtothe

ar-

at

ern

uredwith

onondi-dedard-


fluorine-to-carbon ratio of the fluorocarbon material, see F2~c!, is inversely proportional to the refractive index. Futher, the fluorine-to-carbon ratio of the deposited fluorocbon material is lower if hydrogen is present in the feedgchemistry. This trend can be explained by the fluorine scenging effect of hydrogen in the gas phase,2 or more likely atthese operating pressures fluorine reduction as a resuhydrogen–fluorine recombination at the reactor walls. Alfluorine abstraction by hydrogen from a fluorocarbon surfcould explain the observations.29

The fluorocarbon films that deposit if no bias is appliedthe substrate, can be etched off the thin film substrateself-bias voltages above a certain threshold value. Therates of the fluorocarbon films deposited at unbiased cotions have been measured as a function of self-bias volby in situ ellipsometry in discharges of CHF3, C2F6, C3F6,and C3F6/H2 at 6 mTorr operating pressure and 1400 Wductive power. The fluorocarbon etch rates are plottedpositive values in Fig. 3. Also included in Fig. 3 are thfluorocarbon deposition rates~negative values! measured byin situ ellipsometry at self-bias voltages below the threshfor etching. It shows that the fluorocarbon deposition rdecreases as the self-bias voltage increases.

The fluorocarbon etch rates are found to be stronglypendent on the feedgas chemistry. First, the fluorocaretch rate is relatively low at conditions where hydrogenpresent in the feedgas mixture, e.g., compare CHF3 to C2F6

processing and C3F6 to C3F6/H2 processing. This observatiocan be attributed to the fact that fluorine is a precursoretching of fluorocarbon material.3 The presence of hydrogein the feedgas chemistry namely results in~a! a reduction offluorine in the plasma gas phase~observed when comparinoptical emission spectra from 40 sccm CF4 and 40 sccmCHF3 plasmas at identical conditions!, and ~b! a more fluo-

FIG. 2. ~a! Deposition rates,~b! refractive index, and~c! fluorine-to-carbonratio determined from XPS analysis of fluorocarbon material depositedmTorr operating pressure in discharges fed with 40 sccm of CHF3 , C2F6 ,C3F6 , or C3F6/H2 ~27%!. The fluorine-to-carbon ratios determined und15° electron escape angle are slightly higher than the ratio determined u90°. The values in~c! are averages of the two values.

J. Vac. Sci. Technol. A, Vol. 17, No. 1, Jan/Feb 1999

.

r-sv-

of,e

atchi-

ge

-s

de

-n

s

r

rine depleted fluorocarbon substrate~the fluorine-to-carbonratio of fluorocarbon is reduced, see Fig. 2!. A second obser-vation is that the higher the fluorocarbon deposition rate aW bias power, the lower the fluorocarbon etch rate unbiased conditions, e.g., compare C2F6 to C3F6 processing andCHF3 to C3F6/H2 processing. Net fluorocarbon etching aparently benefits from a reduction in fluorocarbon depotion.

The position of the threshold voltage for net etchingconsistent with the above observations. If hydrogenpresent in the feedgas chemistry or if the fluorocarbon desition rate at 0 W bias power is high, the threshold voltagerelatively high, and vice versa.

B. SiO2 , Si3N4 , and Si etch rates

At conditions where net fluorocarbon etching takes plaother substrate materials, such as SiO2, Si3N4, and Si, canalso be etched. Figure 4 shows the SiO2, Si3N4, and Si etchrates measured on blanket samples byin situ ellipsometry asa function of self-bias voltage in discharges of CHF3, C2F6,C3F6, and C3F6/H2 at 6 mTorr operating pressure and 14W inductive power.

At sufficiently high self-bias voltage, SiO2 etching of theblanket samples occurs at a relatively high rate, whichroughly independent of the feedgas chemistry. This is csistent with Langmuir probe measurements performed fordifferent discharges. The Langmuir probe data showedthe ion current density does not vary significantly wifeedgas chemistry. Since SiO2 etching has been suggestedoccur through a chemical sputtering mechanism in whichion flux is the limiting factor,18 the SiO2 etch rate is expectedto be relatively independent of the feedgas chemistry.~Note:this is the case if the average composition of the fluoroc

6

der

FIG. 3. Fluorocarbon etch rates as a function of self-bias voltage, measwith respect to ground, at 6 mTorr operating pressure in discharges fed40 sccm of CHF3 , C2F6 , C3F6 , or C3F6/H2 ~27%!. The fluorocarbon sub-strate was deposited at 0 W rf bias power at the same process conditiwhere the etch rates were determined. The plasma potential at these ctions typically varies between 20 and 30 V. This potential needs to be adto the self-bias voltage in order to estimate the actual average ion bombment energy.

em

arfTtcanib

isioen

boupon

gre

er

ti

tric

inrial

eei

The

ws

d-

ceshisies,. Atrate

k-

he

as

sion

ring


bon ion flux that bombards the SiO2 surface does not changsignificantly with the various conditions. If the average coposition of the ion flux changes, e.g., from CF2

1 to CF1, theaverage SiO2 sputter yield possibly changes as well.!

The blanket Si etch rates for the above conditionssignificantly lower than the SiO2 etch rates. The etch rates osilicon show a strong dependence on feedgas chemistry.dependence is similar to that observed for fluorocarbon eing, suggesting that the effects due to hydrogen additionfluorocarbon deposition rate ultimately are also responsfor the etch rate behavior of Si.

Etching of Si3N4 is intermediate between SiO2 and Si,both in etch rate and dependence on the feedgas chemFor chemistries that result in a low fluorocarbon depositrate, the Si3N4 etch rate is relatively independent of thfeedgas, similar to SiO2 etching. For chemistries resulting ia high fluorocarbon deposition rate, the Si3N4 etch rate de-pendence is similar to that observed for Si and fluorocaretching. In other words, hydrogen addition only helps to spress Si3N4 etching at conditions where the fluorocarbdeposition rate is sufficiently high.

C. Surface analysis: Steady-state fluorocarbon films

The surface chemistry of processed SiO2 and Si3N4

samples was investigated by XPS as a function feedchemistry in discharges at 6 and 20 mTorr operating psure. The feedgas chemistries used were CHF3, C2F6, C3F6,and mixtures of C2F6/C3F6 and C3F6/H2. At the same condi-tions the surface modifications of crystalline Si samples wmeasured byin situ ellipsometry.30

Figure 5 shows high-resolution C~1s! photoemissionspectra obtained from SiO2 and Si3N4 surfaces etched a2100 V self-bias voltage at 6 mTorr operating pressure

FIG. 4. Etch rates of SiO2 , Si3N4 , and Si substrates as a function of self-bivoltage at 6 mTorr operating pressure in discharges fed with 40 sccmCHF3 , C2F6 , C3F6 , or C3F6/H2 ~27%!.


-

e

heh-d

le

try.n

n-

ass-

e

n

discharges fed with CHF3, C2F6, C3F6, and C3F6 /H2. Thebinding energies were corrected for charging of the dielecsubstrate materials. The integrated intensity of the C~1s!spectrum, plotted in the upper left corner of each graphFig. 5, is a measure for the amount of fluorocarbon matepresent on the sample surface. It shows that the C~1s! inten-sity from processed SiO2 surfaces is typically lower than thintensities measured on Si3N4 samples processed under thsame process conditions, and compared to processed S3N4

samples relatively independent of the feedgas chemistry.intensity of the C~1s! spectrum from processed Si3N4 de-pends significantly on the feedgas chemistry. It clearly shothat the amount of fluorocarbon material on the Si3N4 surfaceshown in Fig. 5, is inversely proportional to the corresponing etch rate at2100 V self-bias from Fig. 4.

The C ~1s! intensity measured on the processed surfacan be expressed as a fluorocarbon film thickness. Tthickness can be calculated from photoemission intensitdue to the exponential decay of the XPS signal with depthmethod that uses angular resolved intensities from subselements is described by Ruegeret al.18 Standaertet al.16

describe a method of quantifying the fluorocarbon film thicness by comparing the C~1s! intensity from a processedsample to the C~1s! intensity of a semi-infinitely thick fluo-rocarbon film~e.g., a fluorocarbon film deposited at 0 W!. Acomparison of the two methods showed that at low C~1s!intensities, both methods give very similar values for tfluorocarbon thickness. At high C~1s! intensities, slightly

of

FIG. 5. C~1s! spectra obtained by XPS surface analysis under 90° emisangle on SiO2 and Si3N4 samples processed in CHF3 ~40 sccm!, C2F6 ~40sccm!, C3F6 ~40 sccm!, and C3F6/H2 ~20 sccm/15 sccm! discharges at 6mTorr and a self-bias voltage of2100 V. The C~1s! intensity is a measurefor the amount of fluorocarbon material present on the surfaces dusteady-state etching conditions.

ntwinouthtws

ti-

te

orboa

s,uori

teriaflue

ede

ceenick-s

r-the

uo--

ons

s

nsr-on

ngles

ge

asetateofate

byederess

tionmoutepely

en-

ac-ese

-to

ck-bonter-

be

onC

ofco

d-


higher values for the fluorocarbon film thickness are fouwith the latter method. The discrepancy between themethods results from the fact that the error in the determtion of the thickness in both methods increases as the amof fluorocarbon material on the surface increases. Inwork the average of the thicknesses determined by themethods is used to quantify the fluorocarbon film thickneIn Fig. 6 the etch rates of SiO2 and Si3N4 are plotted as afunction of the fluorocarbon film thickness for all invesgated chemistries (CHF3, C2F6/C3F6, and C3F6/H2) and pres-sures~6 and 20 mTorr! obtained at2100 V self-bias. Datafor silicon etched at the same conditions as SiO2 and Si3N4 isalso included. The film thicknesses on silicon were demined in real time byin situ He–Ne ellipsometry. In theconversion from ellipsometric anglesC andD to fluorocar-bon film thickness, it is assumed that the steady-state flucarbon film has the same refractive index as fluorocardeposited at the same process conditions without rf biasplied. This assumption is supported by XPS analysi16

which indicates that the composition of the steady-state flrocarbon films and passively deposited fluorocarbon matedoes not vary significantly.

Figure 6 clearly shows a general trend for the investigaprocess regime; i.e., the etch rate of a thin film matedecreases with increasing thickness of the steady-staterocarbon film on its surface. It can be seen that independof the process conditions the SiO2 surfaces stay relativelyclean of fluorocarbon material~fluorocarbon film thick-ness<1.5 nm!. The ion penetration depth in the investigatenergy range is around 1 nm.31 The SiO2 surfaces can thus betched directly by a mechanism of chemical sputtering,6,17,18

which explains the relatively high etch rate. The Si surfaare covered with a fluorocarbon film that is strongly depdent on the process conditions but typically relatively th~2–7 nm!. It is therefore unlikely that direct ion impact contributes to the Si etch rate. Instead, the etch rate is sugge

FIG. 6. Etch rates of SiO2 , Si3N4 , and Si samples plotted vs the thicknessthe fluorocarbon film present on the surface during steady-state etchingditions. The varying parameters are feedgas chemistry@CHF3 ~40 sccm!,C2F6 ~40 sccm!, C3F6 ~40 sccm!, and C3F6/H2 ~20 sccm/15 sccm!# andoperating pressure~6 and 20 mTorr!. The rf bias power level correspondeto a self-bias voltage of2100 V. It clearly shows that the thicker the fluorocarbon film, the lower the etch rate.


doa-nt

iso

s.

r-

o-np-

-al

dlo-nt

s-

ted

to be limited by the arrival of atomic fluorine at the Si suface. Since the fluorine needs to be transported throughrelatively thick fluorocarbon film by a diffusionmechanism;13–16 the Si etch rates are suppressed. The flrocarbon films on Si3N4 are of intermediate thickness compared to SiO2 and Si~1.25–3.75 nm!. For certain conditionsa thick fluorocarbon film is present and the Si3N4 etch rate issuppressed, similarly to the Si etch rate. For other conditithe fluorocarbon films are relatively thin, and the Si3N4 etchbehavior is similar to that of SiO2.

From Fig. 6 it is clear that SiO2-to-Si3N4 selectivity canbe achieved only if a thick enough fluorocarbon film formon the Si3N4 surface, while keeping the SiO2 surface clean.This is completely analogous to the SiO2-to-Si selectivitymechanism. However, Fig. 6 also shows that for conditiowhere Si is already covered with a relatively thick fluorocabon and the etch rate is suppressed, the fluorocarbon filmSi3N4 can be still thin enough for direct chemical sputterito occur. This explains why a process condition that enabSiO2-to-Si selectivity does not necessarily allow for SiO2-to-Si3N4 selectivity, while all process conditions enablinSiO2-to-Si3N4 etch selectivity, automatically also providSiO2-to-Si selectivity.

D. Surface analysis: Reaction layers

In the previous section it was suggested that in the cthat a substrate is covered with a relatively thick steady-sfluorocarbon film, the etch rate is limited by the arrivalatomic fluorine which needs to diffuse towards the substrsurface. This suggestion is based on the observationsOehrleinet al.1 who found that in the case of Si etching ththickness of a fluorinated silicon reaction layer located unthe steady-state fluorocarbon film decreases if the thicknof the fluorocarbon overlayer increases. This observarules in favor of the fluorocarbon film protecting the Si frofluorine attack, rather than that diffusion of etch productsthrough the fluorocarbon film would be the rate limiting st~which is also consistent with the etch rate being inversproportional to the fluorocarbon film thickness!.

Based on the observations in Fig. 6, which suggest a geral etch mechanism for SiO2, Si3N4, and Si in fluorocarbonplasmas, the presence of a fluorinated oxide or nitride retion layer is to be expected. To verify the presence of threaction layers, a detailed analysis of high resolution C~1s!,F ~1s!, Si ~2p!, and N~1s! or O~1s!, was performed.

The C ~1s! spectra from Fig. 5 can typically be deconvolved into four different Gaussian peaks correspondingdifferent chemical carbon bonds~C–CFn , C–F, C–F2, andC–F3), using a least-squares fitting routine and linear baground subtraction. In certain cases where the fluorocarfilms are relatively thin and the fluorocarbon–substrate inface is visible to XPS analysis, a C–C/C–Si peak needs toincluded for obtaining a proper fit of the C~1s! spectrum.The binding energies corresponding to the different carbbonds are given in Table I. From the deconvolving of the~1s! spectra fluorine-to-carbon ratios F/Cdec were calculatedas follows:

n-

h-

sit

onheby

i.e

feFog

lmh15ick

gleell,e

o-igi-.dedrbonto-lm

na-

lo-d toeso-

on

lesesian

r

erideFig.

onts,b-

.5–is-

er

e-

e

e

rom


F/Cdec5I C–F12•I C–F2

13•I C–F3

I C–C/C–Si/C–CFn1I C–F1I C–F2

1I C–F3

, ~1!

whereI bond the photoemission intensity~i.e., area under eacGaussian! originating from a specific bond. The fluorine-tocarbon ratio can also be calculated by dividing the intenof the F ~1s! spectrum~BE5687.1 eV!32 by the total C~1s!intensity, after correcting for differences in photoemissicrossections. In the rest of this work we will indicate tfluorine-to-carbon ratio calculated in the latter fashionF~1s!/C~1s!. Figure 7 shows the differenceDF/C between thefluorine-to-carbon ratios determined by the two methods,

DF/C5F~1s!/C~1s!2F/Cdec, ~2!

for both processed SiO2 and Si3N4 surfaces as a function ofluorocarbon film thickness. The XPS analysis was pformed at both 15° and 90° photoelectron take off angle.XPS spectra obtained under 15° photoelectron take off anwhich is a surface sensitive condition,DF/C is close to zero.At 90° photoelectron take off angleDF/C is also close tozero at conditions where a relatively thick fluorocarbon fiis present on the substrates. This data thus shows that t~1s! photoelectrons detected from all samples using thetake off angle as well as from samples with a relatively th

TABLE I. Listing of photoelectron binding energies.

Element BE Specification Referenc

C ~1s! 283.4 eV C–C/C–Si 16285.5 eV C–CFn288.0 eV C–F290.3 eV C–F2292.6 eV C–F3

F ~1s! 687.1 eV 32Si ~2p! 103.4 eV SiO2 32Si ~2p! 101.7 eV Si3N4 33O ~1s! 532.6 eV SiO2 33N ~1s! 397.5 eV Si3N4 33

FIG. 7. Difference between the fluorine to carbon ratios determined fF~1s!/C~1s! and C~1s! only, as a function of fluorocarbon film thickness.


y

.,

r-r

le,

e F°

fluorocarbon film on the surface using the 90° take off anprimarily originate from fluorine bonded to carbon in thfluorocarbon film. If the fluorocarbon film thickness is smahowever, the F~1s!/C~1s! ratio is found to be higher than thF/Cdec ratio, resulting in a value forDF/C that is larger thanzero. This means that a significant part of the F~1s! photo-electrons detected from samples with a relatively thin flurocarbon film on the surface at 90° take off angle must ornate from fluorine species that are not bonded to carbon

The above data suggest that the fluorine that is not bonto carbon is located underneath the steady-state fluorocafilm, since it can be detected only if the path of the phoelectrons from the underlayer through the fluorocarbon fiis significantly smaller than the electron escape depth~e.g.,90° escape angle and thin fluorocarbon overlayers!. Theelectron escape depth is typically in the order of a fewnometers.

The fluorine that is not bonded to carbon and that iscated underneath the fluorocarbon film can be expecteform fluorinated SiO2 and Si3N4 reaction layers. In the casof Si etching, the films could be observed in the high relution Si ~2p! XPS spectra. In the case of SiO2 and Si3N4

etching, it is possible to obtain information on the reactilayers from substrate elements, Si~2p! and N~1s! or O ~1s!.

Indeed, the Si~2p! spectra at 15° electron escape angobtained from processed Si3N4 samples were found to bantisymmetric and have been deconvolved into two Gauspeaks, as shown in Fig. 8~a!. The low binding energy peakcorresponds to Si~2p! in bulk Si3N4 @BE5101.7 eV; fullwidth at half maximum~FWHM!52.3 eV!.33 The high bind-ing energy peak is found to be shifted by 2.160.2 eV andhas a FWHM of 2.760.1 eV. The shift towards a highebinding energy with respect to bulk Si3N4 can be explainedby silicon bonding to electronegative fluorine. Similarly, thN ~1s! spectrum taken at glazing angles on processed nitsamples can also be deconvolved in two Gaussians, see8~b!. The low binding energy peak corresponds to N~1s! inbulk Si3N4 ~BE5397.5 eV; FWHM52.2 eV!.33 The highbinding energy peak is found to be shifted by 2.2760.35 eVand has a FWHM of 3.0060.25 eV, and can be attributed tsubstrate nitrogen bonded to more electronegative elemesuch as fluorine or carbon with fluorine neighbors. The oserved shift is consistent with this, since~iso!cyanides arereported to have binding energies in the range of 397400.2 eV.34 The above results are consistent with the extence of a fluorinated Si3N4 reaction layer. In the followingwe will associate the high binding energy peaks with eithreacted silicon or nitrogen.

The reacted intensity~Gaussian peak corresponding to raction with F! divided by the unreacted~Gaussian corre-sponding to bulk material! intensity is higher under grazingangles than under 90° electron take off angle in both Si~2p!and N ~1s! spectra. This indicates that the fluorinated Si3N4

layer is on top of the bulk Si3N4 owing to the fact that thesampling depth~measured perpendicular to the surface! at15° is a factor of sin~15°) smaller than at 90°. Both threacted and unreacted Si~2p! and N~1s! intensities decrease

e-rg

id

r

-ths

ha

at-

ion

o-.-ectaovi-of

usal

ithism

k-ma-

of

tion

there-heen-em-

o-itedennge of

of


with increasing fluorocarbon film thickness. This indicatthat both the reacted and unreacted Si3N4 are located underneath the fluorocarbon material. In other words, the fluonated Si3N4 reaction layer is, again analogous to Si etchinlocated at the interface between the Si3N4 sample and thefluorocarbon film.

The intensity ratioI reacted/I bulk of reacted to unreacted S~2p! or N ~1s! is related to the thickness of the fluorinateSi3N4 reaction layerdreacted:

14,35

dreacted5l lnS 11~ I reacted/I bulk!

K D , ~3!

wherel the photoelectron escape depth andK the ratio ofthe atomic density of reacted atoms in the reaction layethe atomic density of unreacted atoms in the bulk Si3N4.Since the intensity ratioI reactedI bulk decreases with fluorocarbon film thickness, see Fig. 9, it can be concluded thatthickness of the fluorinated Si3N4 reaction layer decreases athe fluorocarbon film thickness increases, similar to wOehrleinet al.14 observed for Si etching.

FIG. 8. ~a! Typical Si ~2p! and ~b! typical N ~1s! spectra obtained from aprocessed Si3N4 sample under 15° photoemission angles.


s

i-,

to

e

t

A similar detailed surface analysis was performed intempt to obtain information on fluorinated SiO2 reaction lay-ers. However, it was not possible in either the O~1s! or theSi ~2p! spectra to distinguish between intensities originatfrom a reaction layer or from the bulk SiO2. This is due tothe fact that the binding energies of silicon bonded to flurine and silicon in bulk SiO2 are too close to be resolvedFurther, it is rather unlikely that fluorine will bond to oxygen. Next to the data in Fig. 7, from which one can expthat part of the Si~2p! photoelectrons originate fromfluorinated SiO2 reaction layer in which fluorine is bonded tsubstrate silicon, we do not have additional experimental edence for the existence of such a layer. The formationa fluorinated SiO2 has however been observed by varioinvestigators using additional surface analytictechniques.32,36

The results presented in this section are consistent wthe results in Fig. 6, which suggest a general etch mechanfor SiO2, Si3N4, and Si in fluorocarbon plasmas: the thicness of the steady-state fluorocarbon film controls the fortion of a fluorinated reaction layer and thus the etch ratethe various substrates.

IV. DISCUSSION

The experimental results presented in the previous secsuggest that a general mechanism exists for SiO2, Si3N4,and Si etching in fluorocarbon plasmas. The formation ofetch rate controlling steady-state fluorocarbon film is thefore key in etch selectivity control. The mechanism for tformation of a steady-state fluorocarbon film and the depdence on substrate chemical composition and feedgas chistry conditions will be discussed below.

A. Mechanism of fluorocarbon film formation

Initially, after ignition of the plasma, fluorocarbon depsition takes place on the substrate material. The deposfluorocarbon film will be exposed to ion bombardment whan rf bias is applied. Ion bombardment will result in breakibonds in the fluorocarbon material and thus in the releas

FIG. 9. Ratio of reacted intensity to unreacted intensity as a functionfluorocarbon film thickness for both Si~2p! and N ~1s! signals under 15°photoemission angle.

thfreeu

ntet fo

oninitc

.e

heubprmo

aswcan

cheorb

rai

r-eiaue

hin.

stthp

eth

eensu

ate

forenhe-

tern

s

ub-tc.m-on

ted

toren-

te.nt-is-,dur-heing

ctthick

thineme-


fluorine, which is an etchant for both fluorocarbon andinvestigated substrate materials. It is also possible thatatomic fluorine radicals are produced at the surface duimpact dissociation of fluorocarbon ions. Additionally, thflux of gas phase species to the surface is a possible sofor fluorine radicals.

In addition to the production of the etchant fluorine, iobombardment provides the fluorocarbon/substrate syswith energy, which is the a second necessary componenetching of the investigated substrate materials as was shby Ding et al.37 The energy dissipated in the fluorocarbmaterial will be attenuated as the ions penetrate deeperthe fluorocarbon film. The energy dissipation combined wthe presence of atomic fluorine will result in chemical reations between atomic fluorine and the fluorocarbon film, ietching of the fluorocarbon material.

When the fluorocarbon film is thin enough, some of tion bombardment energy will also be dissipated in the sstrate material. This energy dissipation in the substratemotes chemical reactions of fluorine atoms and substrateterial, i.e., substrate etching. The total amountfluorocarbon etched per unit time will be reduced in this csince the available fluorine and energy has to be sharedthe substrate etching process. At the point that the fluorobon etch and deposition rates balance, steady-state cotions are reached and a steady-state fluorocarbon filmformed on the substrate. The substrate is then being etwith that part of the energy and fluorine that is not requirfor balancing of the fluorocarbon deposition. The detailshow a substrate is etched through a steady-state fluorocafilm are described by Standaertet al.16 for the case of Si.

In the above mechanism, the etch rate of the substmaterial will be reduced if the fluorocarbon deposition rateincreased, since more energy and etch precursors arequired for fluorocarbon removal. Similarly, if the fluorocabon material that is deposited is hard to remove, e.g., dua reduction in the fluorine content in the deposited materless energy and etch precursors will be available for net sstrate etching. This mechanism, therefore, successfullyplains the feedgas chemistry dependence of silicon etcand of Si3N4 etching at highly polymerizing conditionsEtching of SiO2 and Si3N4 at low polymerizing conditions,however, was found to be independent of feedgas chemiIn the following section, the substrate dependence ofetching is explained by the differences in carbon consumtion ability of SiO2, Si3N4, and Si substrates. The abovmechanism is extended such that it successfully explainssubstrate chemistry dependence.

B. Substrate dependence of fluorocarbon filmformation

In the above mechanism, fluorocarbon removal is pformed only by the plasma. In order to explain the depdence of the steady-state fluorocarbon film thickness onstrate type and the corresponding etch rates of SiO2, Si3N4,and Si, also reactions between the substrate materialfluorocarbon species need to be taken in account. The ra


eeeto

rce

morwn

toh-.,

-o-a-feithr-di-iseddfon

tesre-

tol,b-x-g

ry.e-

is

r--b-

ndof

fluorocarbon consumption by the substrate is differentSiO2, Si3N4, and Si, resulting in different balances betweetch and deposition for the various substrates. This is scmatically depicted in Fig. 10. The ability of SiO2 to reactwith carbon species in the fluorocarbon material is greathan for Si3N4 or Si.2–4,6 In other words, reactions betweefluorocarbon and SiO2 with typical volatile reaction productslike CO, CO2, or COF2, are more effective than reactionbetween fluorocarbon and Si3N4, resulting in the formationof volatile products such as CNF38,39 and FCN.40 The differ-ence in fluorocarbon consumption ability between SiO2 orSi3N4 is possibly due to differences in bond strengths, sstrate stoichiometry, volatility of the reaction products, eIn Si etching there is no volatile carbon containing copound at all that can be formed in reactions of fluorocarbspecies and the substrate.

The reactions between fluorocarbon and the SiO2 sub-strate results in consumption of the majority of the deposifluorocarbon material during etching reactions.18 Fluorocar-bon removal by the plasma is therefore not a limiting facfor net substrate etching. Provided that there is enoughergy for chemical sputtering, the SiO2 surfaces will stay rela-tively clean for any deposition rate or fluorocarbon etch raThe SiO2 etch rates will therefore be relatively independeof the feedgas chemistry~provided that the ion current density and the average ion flux composition in the various dcharges are comparable!. On the other hand in Si etchingwhere the substrate cannot assist in the carbon removaling etching, fluorocarbon can only be removed by tplasma. The fluorocarbon etch rate is therefore the limit

FIG. 10. Schematic view of fluxes incident on and outgoing from the~a! Si,~b! Si3N4 , and ~c! SiO2 substrates. In the case of Si, no volatile produbetween the substrate material and carbon exists, and a relativelysteady-state fluorocarbon film develops. In the case of SiO2 , most of thecarbon is consumed in reactions with substrate oxygen, and a relativelysteady-state fluorocarbon film forms. The Si3N4 substrate has a moderatability to react with carbon, and a steady-state fluorocarbon film of interdiate thickness results.

chndn

te

thb

-

ne

nur

7

,m

rbh

inrbinn

i

tenao

ha

aset behar-onee

ed

6 tondW.agedno

sethero-

eatk-thes inltsre-,ea-ar-hatieldss.ldsare

car-all

arly

co

00–g

-


factor, and the Si etch rate will follow the fluorocarbon etand deposition rate trends as a function of process cotions. Si3N4 is intermediate in its ability to consume carbospecies during etching. At low fluorocarbon deposition rathe nitrogen–carbon interaction for a Si3N4 substrate is suf-ficient to maintain a clean surface. The Si3N4 etch rate will inthis case be independent of the feedgas chemistry. Iffluorocarbon deposition is too high, fluorocarbon removalthe plasma becomes a limiting factor in the Si3N4 etchingprocess. The Si3N4 etch rate will then follow the fluorocarbon etch and deposition trends.

The suggested role of substrate oxygen and nitrogecarbon consumption is supported by the results from anperiment in which small amounts of O2 or N2 are added to aCHF3 ~40 sccm! discharge, see Fig. 11. The Si etch rates acorresponding fluorocarbon film thicknesses were measby in situ ellipsometry as a function of O2 and N2 flow at2100 V self-bias voltage, at 1400 W inductive power andmTorr operating pressure. First, Fig. 11~a! shows that whenO2 or N2 are added to the discharge, the etch rate of thesubstrate increases. Second, if O2 is added to the dischargethe etch rate of silicon increases more than if the saamount of N2 is added. Figure 11~b! shows again that theetch rate decreases with increasing steady-state fluorocafilm thickness. These results are consistent with the mecnism in which fluorocarbon deposition, fluorocarbon etchand substrate etching need to be balanced. The fluorocadeposition rate is found to be relatively constant for thevestigated conditions compared to the large changes iand fluorocarbon etch rates. The fluorocarbon etch ratecreases significantly both with O2 and N2 addition and theincrease is larger if a given amount of O2 rather than N2 isadded. The increase in fluorocarbon etch rate is consiswith the suggested role for substrate oxygen and nitrogeassist in the carbon consumption. The result that the increin fluorocarbon etch rate is larger for a certain amountadded O2 than N2, however, does not necessarily mean tatomic oxygen is more reactive than nitrogen, since N2 ismore difficult to dissociate than O2.41

FIG. 11. ~a! Silicon etch rates as a function of O2 and N2 flow added to a 40sccm CHF3 discharge at 7 mTorr at2100 V self-bias.~b! Silicon etch ratesas a function of corresponding fluorocarbon film thickness on the silisurface.


i-

s,

ey

inx-

ded

Si

e

ona-gon-Sin-

nttoseft

C. Further validation of the mechanism

The above suggested mechanism implies that in a cwhere no fluorocarbon deposition occurs, there should noa significant substrate chemistry dependence of the etch cacteristics. It is already observed that for low depositirates, Si3N4 and SiO2 etching are relatively close. In the casof no fluorocarbon deposition at all, Si should follow thSi3N4 in its behavior to etch similarly to oxide.

In order to validate this implication of the suggestmechanism, a set of Si etch experiments in C2F6 dischargeswas performed. The operating pressure was varied from20 mTorr, the total gas flow ranged from 10 to 40 sccm, athe inductive power level was adjusted from 600 to 1400The rf bias power was varied such that the self-bias voltwas kept at2100 V for all conditions. At high pressures anlow inductive power the plasma conditions were such thatnet fluorocarbon deposition occurs at 0 W rf bias power.

Figure 12~a! shows the measured Si etch rates for thisof data as a function of fluorocarbon film thickness. Tgeneral trend of decreasing etch rate with increasing fluocarbon film thickness is still observed, but there is a grdeal of scatter, especially for small fluorocarbon film thicness. However, the ion density varies significantly acrossinvestigated parameter space. To correct for the changeion density it is more useful to plot the experimental resuin terms of etch yield, i.e., number of substrate atomsmoved per incident ion. Figure 12~b! shows the etch yieldscalculated by using values for the ion current densities msured with a Langmuir probe, as a function of the fluorocbon film thickness. For a clean Si substrate it is found tthe etch yield is around 1.5 atoms per ion, and the etch ydecreases exponentially with fluorocarbon film thickneWhen converting the oxide etch rates in Fig. 6 to etch yiewe find values in the range of 0.67–0.86. These valuesclose to the values measured on Si covered with a fluorobon film that has a thickness in the range of 0.5–1 nm. Smdiscrepancies between Si and SiO2 etch yields of fluorocar-bon ions are to be expected. The results shown here cle

n

FIG. 12. ~a! Silicon etch rates measured in C2F6 plasma for a variety ofconditions~6–20 mTorr operating pressure, 10–40 sccm gas flow, and 61400 W inductive power! at 2100 V self-bias as a function correspondinsteady-state fluorocarbon film thickness.~b! Etch yields for the same conditions as a function of corresponding fluorocarbon film thickness.

etdyielmd

on

itletcre

nte

o

s

no

l-d

in

e..,

thin

ssbodwtinlm

ncetch

heg a

sed

er

ro-

ate

k-

nr

ndsthe

bon

-bon

t of

. Inallbe--

ed.the

lm

ribesheieldfilm


support the above suggested general fluorocarbonmechanism in which the key step is to control the steastate fluorocarbon film thickness. The fact that the etch yrather than the rate is controlled by the fluorocarbon fithickness indicates that each individual ion contributesrectly to the etch process, either by dissipating its energyby providing the system with fluorine~e.g., through ion im-pact dissociation or fluorine liberation in the fluorocarbfilm!.

D. Mathematical description

The essence of the above mechanism can be capturedsimple set of equations. It was suggested that the amounfluorocarbon material on a surface is a result from a compbalance between fluorocarbon deposition, fluorocarbon eing and substrate etching, which can aid in fluorocarbonmoval. This balance can be written as

d

dtd~ t !5DRCFx

2ERCFx2CRCFx

, ~4!

whered is the fluorocarbon thickness,DRCFxis the fluoro-

carbon deposition rate during etching,ERCFxis the total fluo-

rocarbon etch rate due to the plasma, andCRCFxis the fluo-

rocarbon consumption rate due to substrate etching.The fluorocarbon consumption rateCRCFx

will be propor-tional to the substrate etch rate and therefore be dependethe fluorocarbon film thickness on the surface. Fig. 6 showthat the substrate etch rate dependence decreases exptially with fluorocarbon film thickness, i.e.,

ERsub5ER0 exp~2d/l!, ~5!

whereER0 is a proportionality constant, which symbolizethe etch rate of a clean surface, andl is a typical attenuationlength for energy dissipation and fluorine concentration aestimated from Fig. 6 to be around 1 nm. The fluorocarbconsumption rate can thus be written as follows:

CRCFx5CR0 exp~2d/l!, ~6!

whereCR0 is a proportionality constant that in fact symboizes the rate at which fluorocarbon material is consumethe substrate keeps itself clean during etching.

Next to the carbon removal rate due to substrate etchthe total fluorocarbon etch rate due to the plasmaERCFx

inEq. ~4! can be assumed to have a thickness dependencthe etching of semi-infinitely thick fluorocarbon films, e.gfluorocarbon films deposited at 0 W rf bias power, all dissi-pated energy and atomic fluorine will be used to removefluorocarbon material. When the fluorocarbon film is thenough, some of the ion bombardment energy will be dipated in the substrate material instead of the fluorocarfilm. The etching of the fluorocarbon film will be reducesince the available fluorine and energy has to be sharedthe substrate etching process. Figure 6 showed thatamount of energy or fluorine used for substrate etching,creases exponentially with decreasing fluorocarbon fi


ch-

ld

i-or

n aofxh--

ond

nen-

dn

if

g,

In

e

i-n

ithhe-

thickness. If it is assumed that the exponential dependeholds for the fluorocarbon material, the fluorocarbon erate can be written as

ERCFx5ER`~12exp~2d/l!!, ~7!

whereER` is a proportionality constant that symbolizes ttotal fluorocarbon etch rate by the plasma when etchinsemi-infinitely thick fluorocarbon film. In order to findER`

for the different process conditions, it needs to be expresin measurable values. This is done by writing Eq.~4! foretching a semi-infinitely thick fluorocarbon film:

d

dtd5DRCFx

2ER`5ERnet, ~8!

whereERnet is the net change in fluorocarbon thickness ptime, which can be measured by ellipsometry.ERnet wasreferred to in the section experimental results as the fluocarbon etch rate.

When solving the fluorocarbon balance for steady-stsubstrate etching conditions:

DRCFx2ERCFx

2CRCFx50, ~9!

one finds an expression in which the fluorocarbon film thicness depends on the fluorocarbon deposition rateDRCFx

, thenet fluorocarbon etch rateERnet and the carbon consumptiorate of a clean surfaceCR0 . Substituting the expression fothe steady-state fluorocarbon film thickness in Eq.~5!, givesthe following expression for the substrate etch rate:

ERsub5ER0

11~~DRCFx2CR0 !/ERnet!

. ~10!

This expression captures the observed experimental trewell. It clearly shows that in the case of Si etching wheresubstrate cannot aid in fluorocarbon removal, i.e.,CR050,the substrate etch rate is a function of both the fluorocardepositionDRCFx

and net fluorocarbon etch rateERnet. Onthe other hand, in the case of SiO2 etching where the substrate consumes the majority of the deposited fluorocarmaterial during etching reactions, i.e.,CR0'DRCFx

, it isclear that the substrate etch rate is relatively independenthe fluorocarbon depositionDRCFx

and fluorocarbon etchrateERnet, and thus independent of the feedgas chemistrythe case of Si3N4 etching, the substrate can only consumedeposited fluorocarbon material if the deposition rate islow a certain level. In other words, for relatively low fluorocarbon deposition rateCR0'DRCFx

, until CR0 reaches itsmaximum and levels off.~Note: in the case of SiO2 there willalso be a maximum deposition rate that can be consumThis maximum deposition rate apparently lies outsideinvestigated process regime, although in the case of C3F6 orC3F6/H2 processing slight increases in the fluorocarbon fithickness on SiO2 surfaces were already observed.!

It needs to be noted that the above rate equation descthe situation for a constant ion flux to the surface. In tprevious section it was seen, however, that the etch yrather than the etch rate depends on the fluorocarbon

ohTfie

e

thatere

an

e

b

r-hldedingalin

/

hen ofors

r ato

on-

tion

by.

ishat

n-

or

o-

ilecer--n-

resd

etpe ing

eld


thickness. Equation~10!, however, can directly be used tcalculate the etch yields at a specific condition, provided tall rate parameters are converted into yield parameters.proportionality constantER0 that symbolizes the etch rate oa clean substrate must thus be replaced by the etch y~atoms removed per incident ion! of a clean substrateEY0 ,the fluorocarbon deposition rateDRCFx

is replaced by thesticking per incident ionS, the net fluorocarbon etch ratERnet changes into the net fluorocarbon etch yieldEYnet,and finally the fluorocarbon consumption rateCR0 is con-verted to the consumption yieldCY0 .

In order to quantitatively check the suggested modelfluorocarbon deposition rates and net fluorocarbon etch rwere determined for the same conditions as the Si etchperiments in Fig. 12. Figure 13 shows how the measuetch yields correlate to the calculated etch rates~Note: thefluorocarbon consumption yieldCY050 in these calcula-tions, since the substrate material is Si.! Although the modelqualitatively succeeds in predicting the Si etch yields, qutitative discrepancies exist. When a fit parameterf is intro-duced in the above model:

EYsub5EY0

11 f •~~S2CY0! /EYnet!, ~11!

quantitative agreement between the model and the expments is found for a value off59. This value could partly bedue to the fact that the actual fluorocarbon removal ratethe plasma is overestimated in Eqs.~7! and~8!. It is conceiv-able that theER` value is lower for steady-state fluorocabon films, that are deposited under conditions with higenergetic ion bombardment, than for fluorocarbon filmsposited under 0 W rf bias conditions. Formation of hard anhighly cross-linked films is possible under the actual etchconditions~Note: cross-linking does not necessarily chanthe average chemical composition observed by XPS ansis!. Further, the actual fluorocarbon deposition rate/stickper ion during substrate etching may be underestimatedtaking it to be equal to the 0 W rf bias deposition rate

FIG. 13. Comparison between predicted etch yields and calculatedyields. The solid points are calculated using the original model, the opoints correspond to the calibrated model.


athe

ld

eesx-d

-

ri-

y

y-

gey-gby

sticking. Enhanced implantation of fluorocarbon ions into tsubstrate during steady-state etching or additional creatiosurface sites for neutral fluorocarbon deposition precursmay enhance the total fluorocarbon deposition.

The introduction of fit parameterf appears to be useful tocalibrate the described model in its simplest form, i.e., foSi substrate. We applied the model in its calibrated formthe data from Fig. 6, and determined the fluorocarbon csumption yieldsCY0 for SiO2 and Si3N4 at the various con-ditions. This can be done by comparing the SiO2 and Si3N4

etch rates to the Si etch rates, while making the assumpthat the etch yield of a clean surfaceEY0 is similar for SiO2,Si3N4, and Si substrates. This assumption is supportedthe fact that the SiO2, Si3N4, and Si data in Fig. 6 line upFigure 14 shows the calculated consumption yieldCY0 as afunction of atoms deposited per incident ion, i.e., stickingS,for the various materials. The consumption yield for Siobviously equal to zero, since the model is calibrated in tfashion. The carbon consumption yield for SiO2 is found tobe close to the fluorocarbon sticking for all investigated coditions. The consumption yield for Si3N4 is only close to thefluorocarbon sticking if the fluorocarbon sticking is low. Fhigher sticking values, the consumption yield for Si3N4 lev-els off. The results in Fig. 14 nicely visualize that the flurocarbon consumption ability of SiO2 is such that it can keepits surface relatively clean of fluorocarbon material, whSi3N4 substrates can only consume fluorocarbon up to atain level, resulting in the formation of relatively thick fluorocarbon films if the carbon supply is too large to be cosumed.

V. CONCLUSIONS

Results of a study on selective SiO2-to-Si3N4 etching inICP discharges fed with different fluorocarbon gas mixtu~CHF3, C2F6/C3F6, and C3F6/H2) have been presented an

chnFIG. 14. Dependence of the carbon consumption yield for SiO2 , Si3N4 , andSi on fluorocarbon sticking calculated for the data from Fig. 6 by comparthe SiO2 and Si3N4 etch rates to Si. On the dotted line the consumption yiis equal to the sticking.

r-

ined

tendicorlmsu

oSim-

tomacuugera

ocotc

rkncrgto

ndoreroroA

an

urf.

ys.,

mp.

ci.

l, G.

l. B

M.l. A

C.

K.

ed-

ac.

L.s.

em.

pl.

N.


compared to selective SiO2-to-Si etching. A general etchmechanism for SiO2, Si3N4, and Si substrates has been fomulated.

XPS surface analysis of SiO2 and Si3N4 samples showsthat a fluorocarbon film is present during steady-state etchconditions. The thickness of the film is strongly correlatwith the etch rate of the substrate material.The thicker thefluorocarbon film the lower the etch rate of the substramaterial.At the interface between the fluorocarbon film athe substrate material, a reaction layer is present. The thness of this reaction layer decreases with increasing flucarbon film thickness, indicating that the fluorocarbon fiprotects the substrate from fluorine attack and thereforestrate etching.

The thickness of the fluorocarbon film is dependentboth the feedgas chemistry and the substrate material.2films are covered with a relatively thin fluorocarbon fil(,1.5 nm!, while Si is covered with a relatively thick fluorocarbon film~2–7 nm!. Si3N4 is intermediate between SiO2

and Si~about 1–4 nm!. The difference in film thickness onthe various substrate materials is suggested to be duedifference in the ability of a substrate material to consucarbon species from the fluorocarbon film in chemical retions during substrate etching. The dependence of the flrocarbon film thickness on the feedgas chemistry is sgested to be a result from a competition betwefluorocarbon deposition, fluorocarbon removal, and substetching.

A mathematical description that includes the fluorocarbdeposition rate, net fluorocarbon etch rate, and a carbonsumption factor allows to predict trends in the substrate eyields.

ACKNOWLEDGMENTS

The authors gratefully acknowledge support of this woby Lam Research Corporation, the New York State Scieand Technology Foundation, and the Department of Eneunder Contract No. DE-FG02-97ER54445. We would likeacknowledge R. J. Severens, M. C. M. v. d. Sanden, andC. Schram from Eindhoven University of Technology, aM. Chang and F. Wang from Advanced Micro Devices fhelpful discussions. M. F. Doemling, B. E. E. Kastenmeiand P. J. Matsuo are thanked for sharing the insightstained in their parallel studies performed in the Plasma Pcessing and Research Laboratories at the University atbany.

1R. A. H. Heinecke, Solid-State Electron.18, 1146~1975!.2J. W. Coburn and E. Kay, IBM J. Res. Dev.23, 33 ~1979!.3J. W. Coburn and H. F. Winters, J. Vac. Sci. Technol.16, 391 ~1979!.4J. W. Coburn, J. Appl. Phys.50, 5210~1979!.5L. M. Ephrath, J. Electrochem. Soc.126, 1419~1979!.6Y.-Y. Tu, T. J. Chuang, and H. F. Winters, Phys. Rev. B23, 823 ~1981!.7G. S. Oehrlein, G. J. Coyle, J. C. Tsang, R. M. Tromp, J. G. Clabes,


g

k-o-

b-

nO

ae-o--

nte

nn-h

ey

D.

,b--l-

d

Y. H. Lee, Mater. Res. Soc. Symp. Proc.68, 367 ~1986!.8G. S. Oehrlein and Y. H. Lee, J. Vac. Sci. Technol. A5, 1585~1987!.9S. Joyce, J. G. Langan, and J. I. Steinfeld, J. Chem. Phys.88, 2027~1988!.

10J. G. Langan, J. A. Shorter, X. Xin, S. A. Joyce, and J. I. Steinfeld, SSci. 207, 344 ~1989!.

11Ch. Cardinaud and G. Turban, Appl. Surf. Sci.45, 109 ~1990!.12T. Akimoto, S. Furuoya, K. Harasima, and E. Ikawa, Jpn. J. Appl. Ph

Part 133, 2151~1994!.13G. S. Oehrlein and H. L. Williams, J. Appl. Phys.62, 662 ~1987!.14G. S. Oehrlein, S. W. Robey, and M. A. Jaso, Mater. Res. Soc. Sy

Proc.98, 229 ~1987!.15G. E. Potter, G. H. Morrison, P. K. Charvat, and A. L. Ruoff, J. Vac. S

Technol. B10, 239 ~1992!.16T. E. F. M. Standaert, M. Schaepkens, N. R. Rueger, P. G. M. Sebe

S. Oehrlein, and J. M. Cook, J. Vac. Sci. Technol. A16, 239 ~1998!.17J. W. Butterbaugh, D. C. Gray, and H. H. Sawin, J. Vac. Sci. Techno

9, 1461~1991!.18N. R. Rueger, J. J. Beulens, M. Schaepkens, M. F. Doemling, J.

Mirza, T. E. F. M. Standaert, and G. S. Oehrlein, J. Vac. Sci. Techno15, 1881~1997!.

19A. M. Barklund and H.-O. Blom, J. Vac. Sci. Technol. A11, 1226~1993!.20J. Givens, S. Geissler, J. Lee, O. Cain, J. Marks, P. Keswick, and

Cunningham, J. Vac. Sci. Technol. B12, 427 ~1994!.21K. Ueno, T. Kikkawa, and K. Tokashiki, J. Vac. Sci. Technol. B13, 1447

~1995!.22H. Hayashi, K. Kurihara, and M. Sekine, Jpn. J. Appl. Phys., Part 135,

2488 ~1996!.23S. Fang, C. Chiang, D. Fraser, B. Lee, P. Keswick, M. Chang, and

Fung, J. Vac. Sci. Technol. A14, 1092~1996!.24Y. Zhang, G. S. Oehrlein, and F. H. Bell, J. Vac. Sci. Technol. A14, 2127

~1996!.25J. M. Cook, O. Turmel, T. Wicker, and J. Winniczek, Technical Proce

ings, Semicon Japan, SEMI, Chiba 1993~unpublished!, p. 414.26F. H. Bell, O. Joubert, G. S. Oehrlein, Y. Zhang, and D. Vender, J. V

Sci. Technol. A12, 3095~1994!.27J. H. Keller, J. C. Forster, and M. S. Barnes, J. Vac. Sci. Technol. A11,

2487 ~1993!.28J. A. Meyer, K. H. R. Kirmse, J.-S. Jenq, S. Y. Perez-Montero, H.

Maynard, A. E. Wendt, J. W. Taylor, and N. Hershkowitz, Appl. PhyLett. 64, 1926~1994!.

29D. C. Marra and E. S. Aydil, J. Vac. Sci. Technol. A15, 2508~1997!.30M. Haverlag, D. Vender, and G. S. Oehrlein, Appl. Phys. Lett.61, 2875

~1992!.31P. C. Zalm,Handbook of Ion Beam Processing Technology,edited by J.

J. Cuomo, S. M. Rossnagel, and H. R. Kaufman~Noyes, NJ, 1989!, p. 77;J. Kirschner and H. W. Etzkorn, Appl. Phys. A: Solids Surf.29, 133~1982!.

32S. W. Robey and G. S. Oehrlein, Surf. Sci.210, 429 ~1989!.33J. A. Taylor, G. M. Lancaster, A. Ignatiev, and J. W. Rabalais, J. Ch

Phys.68, 1776~1978!.34Handbook of X-ray Photoelectron Spectroscopy, edited by G. E. Muilen-

berg ~Perkin-Elmer, Eden Prairie, MN, 1979!.35G. S. Oehrlein, J. Vac. Sci. Technol. A11, 34 ~1993!.36N. Ikegami, N. Ozawa, Y. Miyakawa, and J. Kanamori, Jpn. J. Ap

Phys., Part 130, 1556~1991!; N. Ikegami, N. Ozawa, Y. Miyakawa, N.Hirashita, and J. Kanamori,31, 2020~1992!.

37J. Ding, J.-S. Jenq, G.-H. Kim, H. L. Maynard, J. S. Hamers,Herskowitz, and J. W. Taylor, J. Vac. Sci. Technol. A11, 1283~1993!.

38D. Field, D. F. Klemperer, and I. T. Wade, J. Vac. Sci. Technol. B6, 551~1988!.

39J. Dulak, B. J. Howard, and Ch. Steinbruechel, J. Vac. Sci. Technol. A9,775 ~1991!.

40K. Miyata, M. Hori, and T. Goto, J. Vac. Sci. Technol. A15, 568~1997!.41Y. Wang, R. J. Van Brunt, and J. K. Olthoff, J. Appl. Phys.83, 703

~1998!.

study of the sio -to-sin etch selectivity mechanism in ... · lam research corporation, fremont,...

Documents