Ion-radical synergy in HfO2 etching studied with a XeF2/Ar+
beam setup
Citation for published version (APA):
Gevers, P. M., Beijerinck, H. C. W., Sanden, van de, M. C. M., & Kessels, W. M. M. (2008). Ion-radical synergy in HfO2 etching studied with a XeF2/Ar+ beam setup. Journal of Applied Physics, 103(8), 083304-1/8. [083304]. https://doi.org/10.1063/1.2903058
DOI:
10.1063/1.2903058
Document status and date: Published: 01/01/2008 Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers) Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne
Take down policy
If you believe that this document breaches copyright please contact us at:
openaccess@tue.nl
providing details and we will investigate your claim.
Ion-radical synergy in HfO
2etching studied with a XeF
2/ Ar
+beam setup
P. M. Gevers, H. C. W. Beijerinck, M. C. M. van de Sanden, and W. M. M. Kesselsa兲Department of Applied Physics, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, Netherlands
共Received 29 November 2007; accepted 24 January 2008; published online 23 April 2008兲 To gain more insight into fundamental aspects of the etching behavior of Hf-based high-k materials in plasma etch reactors, HfO2films were etched in a multiple-beam setup consisting of a low energy Ar+ion beam and a XeF
2radical beam. The etch rate and etch products were monitored by real-time
ellipsometry and mass spectrometry, respectively. Although etching of HfO2in XeF2/Ar+chemistry
is mainly a physical effect, an unambiguous proof of the ion-radical synergistic effect for the etching of HfO2is presented. The etch yield for 400 eV Ar+ions at a substrate temperature of 300 ° C was
0.3 atoms/ion for Ar+sputtering and increased to 2 atoms/ion when XeF
2was also supplied. The
etch yield proved to follow the common square root of ion energy dependence both for pure sputtering and radical enhanced etching, with a threshold energy at room temperature of 69⫾17 eV for Ar+ions and 54⫾14 eV for Ar+ions with XeF
2. © 2008 American Institute of Physics.
关DOI:10.1063/1.2903058兴
I. INTRODUCTION
Due to aggressive scaling, current high-performance semiconductor devices based on silicon technology are close to the physical limits of the materials applied. Especially, the gate dielectrics of SiO2 or its derivatives in metal–oxide–
semiconductor field-effect transistors have reached a physi-cal thickness in which they suffer from quantum-mechaniphysi-cal tunneling of carriers. These gate dielectrics will lead to un-acceptably high leakage currents when scaling is continued. The solution to this problem, which has been recognized by industry for quite some time共e.g., illustrated by the Interna-tional Technology Roadmap for Semiconductors兲,1
is to in-crease the dielectric constant of the gate oxide, instead of decreasing the physical thickness. This can be done by uti-lizing other dielectrics, so called high-k materials. Hafnium-based dielectrics are among the leading candidates for the replacement of Si-based oxides because of the high-k values, thermal stability, and electrical reliability of the material with respect to device performance.2 From the processing per-spective, these dielectrics used in gate stacks, also in dy-namic random access memory capacitors, pose many chal-lenges for integrating them in the existing processes. For example, the source and drain regions diagonally below the gate oxide are sensitive to damage, therefore the etch recipes for the high-k material must prevent substantial modification and etching of the underlying layers. Selectivity is therefore one of the very important issues. From a more fundamental point of view, the question arises if we can learn more about the etching process by studying the etching of these materi-als. Likely, high-k material etching is different from SiO2 etching. For example, the boiling point of typical metal ha-lides is much higher than that of their silicon counterparts.3
Metal–oxide etching has been studied in plasma reactors in the past decade and has recently attracted even more
at-tention. A wide variety of chemistries have been tried for the etching of high-k metal oxides such as HfO2 and ZrO2.4–12
These studies have revealed that etching of these materials is dominated by momentum transfer from the ions to the sur-face atoms in most cases.7–9An etch selectivity toward Si of 10 and higher has been reported.12,13Furthermore, Sha et al.5 have reported that a functional transistor could be built with HfO2as a gate dielectric with the pattern transferring using a
BCl3plasma. Cl-based chemistries tend to be favored, due to
the higher volatility of the metal chlorides compared to metal fluorides. However, as long as it is not determined how gate stack formation and, in particular, metal–oxide etching will be implemented in the device production process in the long run, other chemistries are still of interest.14
In the early stages of semiconductor processing, beam etch studies proved to be extremely valuable in investigating and discovering fundamental aspects of etching of silicon-based materials.15They revealed details which were difficult to measure under real plasma conditions, owing to the com-plex nature of plasmas and the immense diversity of associ-ated surface reactions. The knowledge on these details is now utilized extensively in etch technology, e.g., the so-called ion-radical synergy effect16 is exploited in many etch processes. Besides etching studies in plasma reactors, beam studies have proven their merits for the plasma processing field. Given the importance of high-k materials in near-future applications and the historical success of beam setups, the next logical step is to study the etching properties of these materials in such systems under well-defined conditions.
In this paper we report on atomic layer deposited HfO2
etched in a beam setup with a XeF2/Ar+ chemistry. The
setup allows for control of the absolute value of the XeF2
flux, and Ar+ flux and energy, which are difficult to control
separately in a plasma reactor. The system serves as a model system for the etching of metal oxides, in particular for Hf-based dielectrics. The XeF2is used to mimic the F chemistry
like in previous beam studies16 but in general fundamental knowledge can be extended to other chemistries as well. The
a兲Author to whom correspondence should be addressed. Electronic mail:
w.m.m.kessels@tue.nl.
etching process is monitored by real-time ellipsometry and quadrupole mass spectrometry. Ion-radical synergy, ion en-ergy dependence of the etch yield, and etch products will be discussed.
II. EXPERIMENTAL DETAILS A. Multiple beam setup
Because the experimental beam setup has been described extensively in previous publications,17–19 only the most im-portant features will be described here. Briefly, the setup consists of a high vacuum chamber with a base pressure below 10−8 mbar. Samples of approximately 10⫻10 mm2
in size can be loaded in the chamber from a load lock which can store up to six samples. The temperature of the sample can be controlled between 20 and 600 ° C by resistive heat-ing. When loaded, the sample faces a well characterized ther-mal XeF2 neutral beam with an average flux⌽XeF
2 varying
from 3.5⫻1013to 1.4⫻1015 cm−2s−1. The absolute error in
these numbers is thought to be smaller than 10%, while the relative accuracy for different steady-state fluxes is much better, i.e., the error is below 2%. The XeF2beam has a 52°
angle of incidence with respect to the sample surface normal. In addition, Ar+ ions from a low energy共10−2000 eV兲 ion
beam 共Nonsequitur Technologies, customized version of model LEIG-2兲 impinge on the sample at 45° from the nor-mal. Typical average fluxes ⌽Ar+ are 1⫻1013 cm−2s−1 for
ion energies below 70 eV and 3⫻1014 cm−2s−1for 1000 eV.
The accuracy in flux is estimated to be better than 10%, while the energy spread is specified to be below 3 eV. Figure1gives a schematic representation of the beams with respect to the sample. A discussion on the beam profiles and overlay of the beam can be found in the Appendix.
B. HfO2sample
The experiments were performed on plasma-assisted atomic-layer-deposited共ALD兲 HfO2of 32.5 nm thick, which
includes an interfacial SiOx layer of 2⫾1 nm as deduced from capacitance–voltage measurements.20 The HfO2 was
grown at a substrate temperature of 290 ° C on 200 mm Czo-chralski 共100兲 p-type c-Si wafers with a resistivity of 10− 30 ⍀ cm. The Hf precursor was Hf关N共CH3兲共C2H5兲兴4
关tetrakis共ethylmethylamido兲hafnium 共TEMAH兲兴 and the
oxi-dation step in the ALD cycle was carried out with a remote O2plasma. Rutherford backscattering spectrometry showed a
nearly stoichiometric composition of 关O兴/关Hf兴=2.10⫾0.02 for the films. The annealed HfO2 layer had a dielectric
con-stant of k = 16.3 and a bulk density of= 9.68 g cm−3. C. Spectroscopic ellipsometry
The HfO2 was also characterized ex situ with a spectro-scopic ellipsometer 共J.A. Woollam, Inc., M2000D兲 in the range from 190 to 1000 nm to obtain the optical properties for the real-time single-wavelength ellipsometry measure-ments. The data were modeled with a three layer optical model using standard optical constants21 for the c-Si sub-strate, and the interfacial SiOx, and a Tauc–Lorentz model for the HfO2. The thickness of HfO2 could, to a large extent, be
increased by simultaneous decreasing the SiOxthickness and vice versa without reducing the quality of the fits. A thick-ness of 1.0 nm for the SiOx was assumed to extract optical constants of the HfO2, yielding a refractive index of
2.01⫾0.02 and an extinction coefficient equal to zero at 632.8 nm. Although the root-mean-square roughness mea-sured by atomic force microscopy 共AFM兲 was 1.3 nm, no surface roughness was assumed because it did not increase the quality of the fit. The Tauc band gap for HfO2 deduced
from the data is 5.8⫾0.1 eV.
D. Single-wavelength ellipsometry
During etching, the HfO2thickness was monitored by a single-wavelength ellipsometer 共SWE兲 using a He–Ne laser at 632.8 nm in a polarizer-compensator-sample-analyzer con-figuration with rotating compensator. SWE has the advantage over spectroscopic ellipsometry 共SE兲 in our setup that the beam size is considerably smaller, i.e., a 1/e2 beam size of
0.6 mm versus a collimated beam of⬃3 mm diameter. The angle of incidence onto the sample was around 74°, which is close to the Brewster angle of silicon. The compensator of the SWE rotated at 2/3 of the 50 Hz line frequency to reduce disturbances on the power line. The compensator was mounted on an 8-bit encoder. The periodic intensity signal was averaged over 50 revolutions, resulting in a time reso-lution of 1.5 s. The ellipsometry quantities ⌿ and ⌬ were calculated in real time from the second and fourth Fourier
FIG. 1.共Color online兲 共a兲 Experimental setup in horizontal cross section. The sample is rotated in place when loaded and faces the Ar+ion beam, XeF 2beam,
SWE, and QMS. The spherical coordinates共,兲 of the beams are depicted in 共b兲. Additional 共optical兲 ports, which are not used in the present study, are indicated by the dashed lines. The CDA, which is defined by the QMS, is also depicted.
components of the signal. The data were analyzed with the same model as mentioned earlier,22 assuming no change in optical properties of the HfO2. Again, surface roughness was
ignored because AFM measurements proved even a decrease of roughness after ion bombardment. To obtain the exact angle of incidence, which changed from sample to sample, data obtained prior to the experiment were fitted with the layer thickness and the angle as free parameters. Once the angle was determined, the only free fitting parameter during the etch experiment was the layer thickness of the HfO2. In
all experiments a reasonable fit was obtained, although they were better for the etching with XeF2than for the pure
sput-tering case. The accuracy of the thickness measurement is estimated to be around 5%.
E. Quadrupole mass spectrometry
During etching the etch products desorbing from the sample were monitored with a differentially pumped line-of-sight quadrupole mass spectrometer 共QMS兲 at normal inci-dence to the sample, see Fig.1. The etch products that desorb from the sample enter the ion filter of the QMS directly, i.e., in a straight line, mainly without wall collisions. This QMS is capable of monitoring mass-to-charge ratios m/z from 1 to 500 in steps of approximately 0.2. The electron energy in the ionizer was fixed at 70 eV. The time resolution for the real-time measurements ranged from 0.1 to 3 s depending on the number of different species measured and the desired signal-to-noise ratio with respect to the background signal for those species. The QMS is operated in pulse counting mode.
The QMS background signal for the complete mass range increased during etching of HfO2. For 1000 eV ions
with a flux of 3⫻1014 cm−2s−1 this effect was approxi-mately 1 kHz on the base background of around 300 Hz. The effect is less pronounced for lower ion energy and/or ion flux. After complete etching of the HfO2layer共i.e., the HfO2
is completely removed in the ion beam spot area兲, the back-ground signal decreased again to the same level as prior to bombardment. This remarkable behavior is not yet under-stood, but it is most likely an artifact from the system. A possible explanation is that deep ultraviolet photons created in the ion source are more effectively scattered into the QMS by HfO2 than by Si, due to a big difference in reflectance at those wavelengths. However, the detector of the QMS is not positioned in line-of-sight with the sample and the signal dependence on the ion source settings does not completely agree with this explanation. Although not further investi-gated, it should be noted that this background signal does not affect the conclusions presented in this paper.
In Sec. III, the mass spectrum is presented after subtrac-tion of both the background spectrum measured prior to the experiment and the aforementioned mass-independent in-crease in background signal. The same correction was ap-plied in the real time measurement in which the time-dependent presence of ions was taken into account. The real-time signals were normalized to their steady-state values.
F. X-ray photoelectron spectrometry
To investigate the composition of the as-deposited and etched HfO2 film, additional ex situ x-ray photoelectron
spectrometry共XPS兲 measurements were performed in a PHI Quantera scanning x-ray microprobe using monochromatic Al-K␣-radiation in high power mode. During the measure-ments the angle between the axis of the analyzer and the sample surface was 45°. The probing depth at this angle is approximately 6 nm. The Hf concentration was measured by examining the Hf 4f7/2peak at 16.9 eV, while the O 1s peak at 530.1 eV was used for oxygen. Because standard sensitiv-ity factors were used to convert peak areas to atomic concen-trations, the possible error in the concentration ratio is esti-mated to be less than 20%. However, when results are compared, the results are more accurate 共5%兲. The organic contamination on the samples were modeled according to Ref. 23.
III. EXPERIMENTAL RESULTS A. Ion-radical synergy
The ion-radical synergetic effect has been studied by ex-posing clean HfO2films on c-Si to XeF2and/or Ar+ ions at
two different temperatures: room temperature and 300 ° C. Figure2shows the HfO2 layer thickness during such an
ex-periment, as deduced from SWE. The experiment started at t = 0 min with a clean 32.5 nm thick HfO2 layer at 300 ° C,
which was exposed to a XeF2 flux of 7⫻1014 cm−2s−1. At t = t1⬅10 min, the sample was additionally exposed to
400 eV Ar+ at a flux of 0.7⫻1014 cm−2s−1, while still
ex-posed to the XeF2. Five minutes later, at t = t2⬅15 min, the
XeF2beam was switched off.
During the XeF2 exposure, the ellipsometry signal and
consequently the HfO2 thickness remained constant. In fact,
the signal did not change when the XeF2beam was turned on
FIG. 2. 共Color online兲 HfO2layer thickness during etching at a substrate
temperature of 300 ° C as deduced from real-time ellipsometry measure-ments. The HfO2 sample is subsequently exposed to a XeF2 beam only 共t⬍10 min兲, to a XeF2beam and Ar+beam共10⬍t⬍15 min兲, and to an Ar+
beam only 共t⬎15 min兲. The beam fluxes are 7⫻1014 cm−2s−1 for the
XeF2and 0.7⫻1014 cm−2s−1for the 400 eV Ar+ions. The solid line is a
curve fit to Eq.共1兲and the top frame shows the residue, i.e., the difference between the data and the curve fit.
共not shown兲. Upon exposure to Ar+, the layer thickness
ini-tially changed rapidly. After 15 s, a lower steady-state etch rate of 0.9 nm min−1set in. When XeF2was switched off at t2, the etch rate gradually decreased to a steady-state etch
rate of 0.17 nm min−1. The difference in the two steady-state etch rates proofs unambiguously that a synergy effect exists between XeF2and Ar+ions during etching of HfO2.
To investigate the dynamics of the etching process, when switching from one condition to another, the thickness is modeled with a time dependent expression for the etch rate ERSWE. This empirical ERSWEis the sum of an equilibrium
rate EReqSWE and an exponential decaying transient rate, as given by:
ERSWE共t兲 = EReq,xSWE+ ER0,xSWEexp
冉
−t − txx
冊
, 共1兲
where the subscript x denotes the respective time frame, ER0SWEis the magnitude of the decaying additional etch rate, tx= t1 and t2, respectively, and x is the time constant. The initial thickness at t1 is the average between t = 1 min and t = t1, while the initial thickness at t2 is taken from the fit to the previous time interval. In Fig.2, the resulting curve fit is shown, in which the difference between the data and the curve is also depicted. The parameters resulting from the fits are given in Table I. Equivalent experiments at room tem-perature showed similar behavior. A discussion of the trends
will follow after the presentation of the mass spectrometry data in the next subsection.
B. Reaction product analysis
In order to identify the reaction products, a mass spec-trum in the mass range of Hf and its fluorides was measured and is presented in Fig. 3. The spectrum is shown for a mass-to-charge ratio from m/z=168 to 252. The spectrum was measured while HfO2 was etched at room temperature
with 1000 eV Ar+at a flux of 3⫻1014 cm−2s−1and XeF 2at
a flux of 7⫻1014 cm−2s−1. This high ion energy was chosen
to obtain a signal-to-noise ratio better than 2 for this part of the spectrum. As discussed in Sec. II, the mass spectrum is corrected for the background spectrum. The spectrum shows peaks at m/z values corresponding to XeF2+, HfF+, HfF2+, and HfF3+, with a possible contribution of Hf+ inside the large XeF2+signal. The spectrum in Fig.3was not extended to the m/z values of HfF4+because HfF4+was not observed in any of our experiments.
Although no cracking patterns for dissociation by elec-tron impact of hafnium fluorides are available in the litera-ture and in some etch systems the etch products created dur-ing the collision cascade are not fully halogenated,15 we conjecture that the spectrum in Fig.3results from HfF4
mol-ecules. HfF4 is the only stable fluoride3 of Hf, so it can be
expected to be one of the major reaction products. As already mentioned, HfF4+ was not observed in any of our experi-ments, nor in plasma experiments as discussed for example by Takahashi et al.10The ions SiH4+and CF4+are known to be unstable.24,25 It is possible that HfF4+ is also unstable and therefore not present in the mass spectrum. The spectrum in Fig.3then represents the cracking pattern of HfF4. The rela-tive contributions of the cracking products are calculated by summation of the peak areas and given in TableIIfor future reference.
FIG. 3. Mass scan共m/z ratio from 168 to 252兲 recorded by the QMS during HfO2 etching by fluxes of 3⫻1014 cm−2s−1 1000 eV Ar+ ions and
7⫻1014 cm−2s−1 XeF
2. The spectrum has been corrected for the
back-ground signals. For comparison, the relative abundance of the species with respect to the isotopic composition is indicated in the top of the figure. The substrate was at room temperature.
TABLE I. The parameter values in Eq.共1兲for the fits of the etch rate to the data presented in Fig.1. Transient Equilibrium
Frame x Time共min兲 Species
ER0,xSWE
共nm/min兲 共min兲x
EReq,xSWE
共nm/min兲 0 t⬍10 XeF2 0共fixed兲 0共fixed兲 0.002⫾0.002
1 10⬍t⬍15 Ar+, XeF
2 6.0⫾0.2 0.05⫾0.02 0.900⫾0.004
2 t⬎15 Ar+ 0.72⫾0.01 0.7⫾0.2 0.17⫾0.02
TABLE II. The cracking pattern of HfF4as deduced from Fig.3. Due to the
background correction of the spectrum, the error in these numbers is rather large and estimated at 20% point.
Species Relative contribution共%兲
Hf+ 0
HfF+ 60
HfF2+ 40
HfF3+ 100
C. Real-time dynamics
Figure4共a兲presents the etch rate using Eq.共1兲with the parameters in Table Iand also as deduced from the data in Fig.2by numerical differentiation.26Together with the ellip-sometry results, the simultaneously measured QMS data for XeF+ 共m/z:151兲 and HfF
3
+ 共m/z:237兲 are depicted in Figs.
4共b兲 and4共c兲. The XeF+ signal was constant and relatively
high during the XeF2 exposure phase, which indicated that
no XeF2 reacted with the HfO2. The HfF3+ signal was at its background level during exposure to XeF2 only, as can be
expected when no spontaneous etching by XeF2took place. Once the ion beam was switched on at t1, the XeF+ signal
initially decreased to approximately 75% and reached a steady-state level of⬃82% of the value prior to Ar+impact.
The HfF3+ signal increased slowly after exposure to Ar+, reaching a steady state in 1 min. When the XeF2beam was
switched off at t2, the XeF+signal decreased within 1 min to
the background level, while the HfF3+signal decayed back to the background level in approximately 2 min.
At first glance, it appears that the initially high etch rate at t1 can be explained by a layer of adsorbed XeF2 on the surface prior to the Ar+ ion bombardment. Under the as-sumption that the additional etch rate, which accounts for the etching of 0.3 nm HfO2, leads to HfF4formation, this means
that there was a layer of XeF2 present with a surface
cover-age of XeF
2= 2⫻10
15 cm−2. However, two arguments are
not supporting this explanation. First, the real-time response of the HfF3+did not show a response that supports this view, i.e., HfF3+ shows a slow increase instead of an overshoot. Second, the initial high etch rate was also observed for Ar+
only experiments 共not shown兲, where no XeF2 was present. An alternative explanation could be that preferential sputter-ing of O allows for initially fast etchsputter-ing. Also ion bombard-ment might induce some minor optical effect not incorpo-rated in the ellipsometry model which means that the transient in the etch rate is an artifact of the measurement. Any combination of these effects could also be present. At this time it is not possible to distinguish these effects.
As no spontaneous etching took place during the XeF2
exposure before t1, the steady-state level of XeF+ acts as a
measure of the incoming XeF2. Thus, the change in the XeF+
signal after t1reflects the consumption of XeF2by the
etch-ing process. The flux of available F radicals for chemical reactions, calculated from the steady-state consumption of XeF2 between t1 and t2, was ⌽F= 2.5⫻1014 cm−2s−1. As
discussed in a previous section, the fluorine was likely used to create HfF4. With the bulk density of HfO2, the
consump-tion of XeF2 corresponds to an etch rate of
EReqQMS= 1.3 nm min−1by the creation of HfF4. This value is
on the order of magnitude as deduced from ellipsometry 共EReq
SWE= 0.9 nm min−1兲, although it is ⬃45% higher. A part
of this discrepancy can be explained by experimental error, however, the difference seems to be large.
A possible explanation for the discrepancy is the storage of F atoms on or in the sample, which is known to occur for Si during XeF2/Ar+ etching.15 The gradual decrease of the
etch rate and HfF3+signal after t2indicates that a buffer of F
is present in this case as well. A rough estimation of the depth at which the F is stored can be made by determining the decrease in layer thickness necessary to obtain the steady-state. The difference in HfO2 thickness between t = t2
and t = t2+ 32is⬃0.9 nm, where2is taken from TableIfor the appropriate time interval. The resulting thickness is ap-proximately the same as the 1.1 nm penetration depth of 400 eV Ar+ions in HfO
2共calculated withSRIM兲27and
there-fore seems credible when mixing is responsible for the pen-etration of F in the HfO2. Additionally, under the assumption
that for t⬎t2 the exponential part of Eq. 共1兲 is completely
due to creation of HfF4, the maximum amount of buried F can be estimated. The integration of the exponential part of the etch rate using ER0,2and2 from TableIyields a
maxi-mum amount of F atoms ofF= 3⫻1015 cm−2. This number
is about five times smaller than the difference needed to ex-plain the difference in etch rate as calculated from ellipsom-etry and XeF2consumption data.
FIG. 4. 共Color online兲 共a兲 The curve fit 共solid line兲 of the data in Fig.2to Eq.共1兲together with the real time etch rate deduced from the data in Fig.2
by numerical differentiation共Ref.26兲 共open symbols兲. The XeF+signal共b兲
and HfF3+signal共c兲, as obtained by mass spectrometry simultaneously with
the SWE data, are also given. Both QMS signals exhibited a disturbance at
t = 8 min, which might be related to a “particle burst” in the QMS. The
XeF+signal represents the XeF
2 flux reflecting from the sample and the
HfF3+signal is expected to originate from HfF
4as etch product. Other
Another hypothesis to explain the difference in etch rates is the consumption of F atoms to create products involving O atoms. Unfortunately, no direct proof is available in our case and oxyfluoride species are not stable at room temperature.28 However, Joyce et al.29 speculate on the presence of such species on a thermally grown SiO2surface exposed to XeF2.
Under the assumption that all O atoms are removed as O2F2,
the etch rate calculated by XeF2consumption goes down by
50% and the etch rates are the same within the experimental error. Further work is necessary to investigate this hypoth-esis.
Yet another possibility for the dissimilarity of the two etch rates is partial consumption of XeF2, i.e., release by
XeF2of one F atom to the substrate and consecutive
desorp-tion of XeF from the sample. Because XeF is weakly bound 共0.13 eV兲,30
partial consumption is unlikely to have a large influence on the overall consumption of XeF2. Moreover,
reasoning along the lines of cracking patterns, partial con-sumption will only lead to an overestimation of the etch rate when the contribution of XeF+to the XeF2mass spectrum is
more than twice the contribution of XeF+ to the XeF
spec-trum. Together, these arguments suggest that partial con-sumption is unlikely to cause the 45% difference in calcu-lated etch rate.
Note that if we assume the creation of lower hafnium fluorides during XeF2/Ar+etching, i.e., HfF
xwith x⬍4, the calculated etch rate EReqQMS would go up, yielding an even larger difference with EReqSWE. Therefore, the XeF2
consump-tion indirectly supports the assumpconsump-tion of creaconsump-tion of HfF4in
the XeF2/Ar+etching process.
D. XPS analysis
In order to investigate possible preferential sputtering, two samples were examined ex situ by XPS. One sample was as-deposited, the second sample was equal to the other sample, except that approximately 8 nm HfO2 was sputtered away by 1000 eV Ar+ ions, while the sample was at room
temperature. According toSRIM, 1000 eV ions have an aver-age penetration depth in HfO2 of 1.6⫾0.9 nm. The 关O兴/关Hf兴 ratio of the as-deposited sample was measured to be 2.1⫾0.4, which agreed with the aforementioned Ruther-ford backscattering spectroscopy results.20 The sputtered sample showed no sign of preferential sputtering, as it had the same 关O兴/关Hf兴 ratio of 2.1⫾0.4, while also no metallic Hf was observed. Both samples showed some contamination by hydrocarbons due to handling and exposure to the ambi-ent. The contamination on the sputtered sample was some-what lower than the as-deposited sample, i.e., 0.8 nm versus 1.2 nm equivalent thick organic layer.23Interesting to note is that small traces of Ar 共1.9⫾0.4 at. % with respect to the HfO2兲 were detected in the top layer of the sputtered HfO2.
Although this experiment does not confirm preferential sputtering, it does not rule out the possibility because the samples were in ambient air for some time before the XPS measurements were carried out. During this time a metallic top surface could have reoxidized. In XPS measurements done by Sha and co-workers,5metallic Hf was found in HfO2
samples sputtered by an Ar plasma.
E. Ion energy dependence
In general, the etch yield EY of sputtering and ion-assisted etching processes depend on the ion energy E. In the ion energy range relevant for dry etching the etch yield usu-ally follows the commonly found dependence:31
EY⬃
冑E −
冑E
th, 共2兲with Ethas the threshold ion energy for etching. The work of
several authors5,6,10suggests that etch yield of HfO2 in
sev-eral different chemistries is governed by the same equation. For measurement convenience the data in literature are pre-sented as the etch rate. However, when no precautions are taken the ion flux in a plasma setup can vary when the ion energy is changed. Furthermore, the ion energy in a conven-tional plasma reactor is not very well defined and often has a bimodal energy distribution due to the applied radio fre-quency voltage.32 These effects might explain why not all etch rate data4,5,8,9 agree well with Eq. 共2兲, while the etch yield may still follow this ion energy dependence. To inves-tigate the ion energy dependent behavior of HfO2etching in our setup several etch series were carried out.
Figure 5 presents the steady-state etch yields for two samples in which the ion energy was varied stepwise. One series was measured for pure Ar+ sputtering starting with
100 eV ions, increasing the ion energy up to 1000 eV. The other series was measured for radical-assisted ion etching, with the ratio between the XeF2and Ar+fluxes fixed at
ap-proximately 8. The latter series started with 1000 eV ions, with subsequent decrease of the ion energy down to 70 eV. The etch yield in Fig. 5 is defined as the total number of atoms共hafnium and oxygen兲 removed per incident ion. This etch yield was calculated from the ion flux ⌽ArSWE+ , the etch
rate EReqSWE as measured with ellipsometry, the bulk density of HfO2, and the atomic masses mHfand mOof Hf and O:
FIG. 5. 共Color online兲 HfO2etch yield vs the square root of the Ar+ion
energy E, both for Ar+sputtering and for XeF
2assisted ion etching. For the
XeF2 assisted ion etching the XeF2to Ar+flux ratio is kept constant at
approximately 8. The solid lines are curve fits to the data. The same data are plotted vs the ion energy in the inset. The substrate was at room temperature.
EY =EReq SWE ⌽ArSWE+ meff , 共3兲 with meff= mHf+ 2.1mO 3.1 , 共4兲
where meff denotes the effective mass per atom. The flux
⌽Ar+
SWE= 1.27⌽
Ar+ is the Ar+ flux in the area probed by the
ellipsometer as discussed in the Appendix. Equation 共4兲 is based on the关O兴/关Hf兴 ratio from the XPS analysis.
As the XeF2/Ar+ etch yield is always higher than the
Ar+ etch yield, the results in Fig. 5 show that the synergy
effect for the XeF2/Ar+chemistry is present for all relevant
ion energies. Both etch yield series can be reasonably well fitted with Eq.共2兲, as indicated in Fig.5. Interesting to note is the difference of the共fitted兲 threshold energy between both cases, i.e., 69⫾17 eV for pure sputtering versus 54⫾14 eV for the XeF2 assisted etching. Although the error in the
threshold ion energy is quite large, the difference can be significant. In that case, apparently, less energy is needed to form volatile species in the case of ion-radical synergy. For comparison,SRIMsimulations were performed for sputtering of HfO2 with Ar+ ions at several different energies. At the
high side of the ion energy range, i.e., at 1000 eV, the total calculated sputter yield was roughly four times higher than the measured etch yield. The threshold energy resulting from the simulations was 42⫾1 eV, which is lower than in the experiments. According toSRIM, the sputter yield for Hf
at-oms is approximately five times lower than the yield for O atoms, i.e., O atoms are preferentially sputtered. In theSRIM
calculation, however, the target composition is restored for each calculation, thus the total calculated sputter yield is overestimated. In order to see the maximum influence of this effect, the etch yield of pure Hf sputtering simulations is also calculated, resulting in a total etch yield that is approxi-mately a factor of 2 lower. As SRIM is currently not very accurate in predicting low ion-energy results,33 the overall agreement between simulations and measurements is consid-ered good.
IV. CONCLUDING REMARKS
In this paper a multiple beam setup with several diag-nostics was used to investigate the etching behavior of HfO2
in the XeF2/Ar+chemistry. Real-time ellipsometry was used
to measure the HfO2layer thickness which in turn was used
to extract the etch rate as a function of XeF2flux and Ar+ion
flux and energy. The results prove unambiguously that ion-radical synergy plays a significant role in the etching of HfO2
for all ion energies relevant in dry etching processes. Ion bombardment is needed to etch HfO2as XeF2alone does not
etch HfO2. When XeF2 is present during etching, a reaction
layer is created, which is around the same thickness as the penetration depth of the ions. Mass spectrometry data were used to investigate the etch products in the presence of XeF2
and it is discussed that the most abundant reaction product is likely HfF4.
In these experiments, the ion energy dependence of the etch yield was found to follow the square root of ion energy for both Ar+sputtering and for radical enhanced etching. The
threshold energy was determined to be lower for the XeF2/Ar+ combination than for sputtering alone, possibly
due to different reaction products that are released from the surface with different threshold energies.
Although this paper only reports on the etching of HfO2
in the XeF2/Ar+chemistry, the results are also relevant for
other high-k materials, both for other Hf-based dielectrics and other metal-oxides. Also with respect to the etching chemistry, which is probably going to be based on Cl rather than on F, the results presented in this paper can serve as a guideline for further research. Furthermore, it turned out that in beam studies the use of a layered system with high optical contrast facilitates the measurement of the etch yield.
ACKNOWLEDGMENTS
The authors would like to thank Oxford Instruments Plasma Technology for supplying the HfO2samples and M. J. F. van de Sande, A. B. M. Hüsken, H. M. M. de Jong, and R. F. Rumphorst for their skillful technical assistance. The research of W.K. was made possible by a fellowship from the Royal Netherlands Academy of Arts and Sciences共KNAW兲.
APPENDIX: BEAM PROFILES AND OVERLAY
In this work different beams with their respective beam profiles are used. In this Appendix the profiles and beam overlay procedures will be discussed.
The central detection area共CDA兲 in this setup is defined by the QMS. The etched surface is sampled by the QMS via a flow resistance and entrance aperture of the ionizer, which are 3 mm in diameter.17 The flat top profile that is thus probed has a 3 mm top diameter and 3.2 mm full width at half maximum diameter. In Fig. 6, the 共3 mm兲 CDA is de-picted on the exposed side of the sample.
The XeF2 beam consists of a temperature controlled
XeF2 container, a capillary used as a fixed flow resistance,
transport tubing, and a multichannel array. The procedure for the calculation of the flux is described extensively in a
pre-FIG. 6. 共Color online兲 The overlay of different beams on the sample with the central detection area defined by the QMS共solid line兲. The 1/e2
inten-sity contour line for the He–Ne laser共dashed line兲 and the XeF2flux profile
lines共dotted lines兲 are shown in 共a兲. The Ar+flux profile lines共dotted lines兲
are shown in共b兲. The contour lines for the XeF2and Ar+fluxes have 20%
intervals with respect to the maximum flux and the 50% contour line is tagged.
vious publication.17 For completeness the resulting contour lines are depicted in Fig. 6共a兲. It is noted that the average flux ⌽XeF2 on the CDA 共which is used as the flux in this
work兲 is 77% of the peak flux ⌽XeFMax2, while the lowest flux on
the CDA is ⌽XeF
2
Min = 0.47⌽ XeF2 Max
. The alignment of the XeF2
beam is optimized 共using micrometers attached to the source兲 by maximizing the XeF+ signal in the QMS while
exposing a Ni target.
The Ar+ beam has a Gaussian beam profile, which size
depends on the ion energy, the Ar gas pressure in the ion gun, and the focus voltage. All measurements are performed with a beam size at the center of the sample of = 1.1 mm. The ion flux distribution is depicted in Fig. 6共b兲. Because the beam angle is 45° with respect to the surface normal, the beam divergence is taken into account. From calibration measurements the divergence is estimated to be 0.04 rad maximum. The ion flux distribution is depicted in Fig.6共b兲. For the CDA this means that the average flux ⌽Ar+ on the
CDA is 71% of the peak flux ⌽ArMax+. The lowest Ar+ flux is
⌽ArMin+= 0.37⌽ArMax+. Alignment is facilitated by the fact that ion
bombardment of HfO2 samples produces a visible spot on
the sample. This spot is then overlapped with a light spot that marks the CDA. The light spot is produced by a light source in the QMS vacuum chamber.
The He–Ne laser has a specified 1/e2 beam waist of w0= 0.59 mm and a divergence of = 1.35 mrad. The
Ray-leigh range z0 associated with this beam waist is 1.7 m.
Be-cause the distance between the laser and the sample is only z = 0.43⫾0.03 m, the beam size w at the sample is still com-parable to the beam waist, which is given by
w共z兲 = w0
冑
1 +共z/z0兲2. 共A1兲The 1/e2 beam contour is depicted in Fig. 6共a兲. The
alignment of the ellipsometer is done on a HfO2 sample
which was etched previously with a tightly focused ion beam. The ion beam creates a small spot with a different optical response than the rest of the sample, providing high contrast for the alignment of the ellipsometer on the center of Ar+beam. Under normal operating conditions the area which is monitored by the ellipsometry receives⬃27% higher Ar+ flux and⬃15% higher XeF2 flux than specified by the
aver-age fluxes⌽Ar+and⌽XeF 2.
1International Technology Roadmap for Semiconductors, http://
www.itrs.net.
2G. D. Wilk, R. M. Wallace, and J. M. Anthony,J. Appl. Phys.89, 5243
共2001兲.
3CRC Handbook of Chemistry and Physics, 87th ed., edited by D. R. Lide
共CRC Press, Boca Raton, 2006兲.
4S. Norasetthekul, P. Y. Park, K. H. Baik, K. P. Lee, J. H. Shin, B. S. Jeong,
V. Shishodia, D. P. Norton, and S. T. Pearton,Appl. Surf. Sci.187, 75
共2002兲.
5L. Sha, R. Puthenkovilakam, Y. S. Lin, and J. P. Chang, J. Vac. Sci.
Technol. B21, 2420共2003兲.
6L. Sha and J. P. Chang,J. Vac. Sci. Technol. A22, 88共2004兲.
7T. Maeda, H. Ito, R. Mitsuhashi, A. Horiuchi, T. Kawahara, A. Muto, T.
Sasaki, K. Torii, and H. Kitajima, Jpn. J. Appl. Phys., Part 143, 1864
共2004兲.
8M. H. Shin, S. W. Na, N. E. Lee, T. K. Oh, J. Kim, T. Lee, and J. Ahn,Jpn.
J. Appl. Phys., Part 144, 5811共2005兲.
9K. Takahashi, K. Ono, and Y. Setsuhara,J. Vac. Sci. Technol. A23, 1691
共2005兲.
10K. Takahashi and K. Ono,J. Vac. Sci. Technol. A24, 437共2006兲. 11T. Kitagawa, K. Nakamura, K. Osari, K. Takahashi, K. Ono, M. Oosawa,
S. Hasaka, and M. Inoue,Jpn. J. Appl. Phys., Part 245, L297共2006兲.
12M. Hélot, T. Chevolleau, L. Vallier, O. Joubert, E. Blanquet, A. Pisch, P.
Mangiagalli, and T. Lill,J. Vac. Sci. Technol. A24, 30共2006兲.
13E. Sungauer, E. Pargon, X. Mellhaoui, R. Ramos, G. Cunge, L. Vallier,
and O. Joubert,J. Vac. Sci. Technol. B25, 1640共2007兲.
14J. H. Ko, D. Y. Kim, M. S. Park, N.-E. Lee, S. S. Lee, J. Ahn, and H. Mok,
J. Vac. Sci. Technol. A25, 990共2007兲.
15H. F. Winters and J. W. Coburn,Surf. Sci. Rep.14, 161共1992兲. 16J. W. Coburn and H. F. Winters,J. Appl. Phys.50, 3189共1979兲. 17M. J. M. Vugts, G. J. P. Joosten, A. Vanoosterum, H. A. J. Senhorst, and
H. C. W. Beijerinck,J. Vac. Sci. Technol. A12, 2999共1994兲.
18A. A. E. Stevens and H. C. W. Beijerinck,J. Vac. Sci. Technol. A23, 126
共2005兲.
19A. A. E. Stevens, W. M. M. Kessels, M. C. M. van de Sanden, and H. C.
W. Beijerinck,J. Vac. Sci. Technol. A24, 1933共2006兲.
20S. B. S. Heil, J. L. van Hemmen, C. J. Hodson, N. Singh, J. H. Klootwijk,
F. Roozeboom, M. C. M. van de Sanden, and W. M. M. Kessels,J. Vac. Sci. Technol. A25, 1357共2007兲.
21C. M. Herzinger, B. Johs, W. A. McGahan, J. A. Woollam, and W.
Paul-son,J. Appl. Phys.83, 3323共1998兲.
22Because the interfacial SiO
xlayer had very little contrast at this
wave-length towards the HfO2, the layer could in practice be ignored and still
yield the same etch rates.
23C. van der Marel, M. Yidirim, and H. R. Stapert,J. Vac. Sci. Technol. A
23, 1456共2005兲.
24H. F. Winters and M. Inokuti,Phys. Rev. A25, 1420共1982兲.
25J. Perrin, O. Leroy, and M. C. Bordage, Contrib. Plasma Phys. 36, 3
共1996兲.
26The procedure of the numerical differentiation in this case started by
grouping the data in sets of 16 data points and averaging each group共for both the thickness value and time兲. The final data were obtained by taking the difference between successive groups divided by the time difference.
27J. F. Ziegler, J. P. Biesack, and U. Littmark, The Stopping and Range of
Ions in Solids共Pergamon, New York, 1985兲.
28A. D. Kirshenbaum, A. V. Grosse, and J. G. Aston,J. Am. Chem. Soc.81,
6398共1959兲.
29S. Joyce, J. G. Langan, and J. I. Steinfeld,Surf. Sci.195, 270共1988兲. 30P. C. Tellinghuisen, J. Tellinghuisen, J. A. Coxon, J. E. Velazco, and D. W.
Setser,J. Chem. Phys.68, 5187共1978兲.
31C. Steinbrüchel,Appl. Phys. Lett.55, 1960共1989兲. 32C. Wild and P. Koidl,J. Appl. Phys.69, 2909共1991兲. 33K. Wittmaack,J. Appl. Phys.96, 2632共2004兲.
