High-density plasma patterning of low dielectric constant polymers: A comparison between polytetrafluoroethylene, parylene-N, and poly„arylene ether… T. E. F. M. Standaert,a) P. J. Matsuo, X. Li, and G. S. Oehrleinb) Department of Physics, State University of New York at Albany, Albany, New York 12222
T.-M. Lu and R. Gutmann Rensselaer Polytechnic Institute, Troy, New York 12180
C. T. Rosenmayer and J. W. Bartz W. L. Gore & Associates, Inc., Eau Claire, Wisconsin 54701
J. G. Langanc) and W. R. Entley Electronics Division, Air Products and Chemicals, Inc., Allentown, Pennsylvania 18195
共Received 24 November 1999; accepted 18 December 2000兲 The pattern transfer of SiO2 hard masks into polytetrafluoroethylene, parylene-N, and poly共arylene ether兲 共PAE-2兲 has been characterized in an inductively coupled plasma source. Selected results obtained with blanket parylene-AF4 films are included in this work. These dielectrics offer a relatively low dielectric constant 共k⬃2–3兲 and are candidate materials for use as intra- and interlayer dielectrics for the next generations of high-speed electronic devices. Successful patterning conditions were identified for Ar/O2 and N2 /O2 gas mixtures. It was found that the formation of straight sidewalls in Ar/O2 discharges relies on the redeposition of oxygen-deficient etch products on the feature sidewall. Furthermore, the etch rates of parylene-N, parylene-F, and PAE-2 for blanket and patterned films could be captured by a semiempirical surface coverage model, which balances the adsorption rate of oxygen and the ion-induced desorption rate of oxygenated etch products. © 2001 American Vacuum Society. 关DOI: 10.1116/1.1349201兴
I. INTRODUCTION Integration of low dielectric constant 共k兲 materials in upcoming generations of integrated circuits 共IC’s兲 becomes indispensable, because interconnect delays will limit the overall performance of sub-0.25 m devices based on conventional dielectrics and metals only. Low-k dielectrics also reduce cross-talk noise between adjacent lines. Even though many low-k materials have been introduced over the last years, only a very few are used in commercial devices. Successful integration puts high demands on thermal conductivity and stability, gap fill, low outgassing, production costs, and compatibility with processing steps like etching, stripping, cleaning, and chemical–mechanical polishing. Because of partial compatibility with conventional oxide technology, oxide-like materials were the first low-k dielectrics 共k⬃3–4兲 to be implemented in commercial devices. Some of the etching and cleaning issues for fluorinated oxide, hydrogen silsesquioxane 共HSQ兲, and methyl silsesquioxane 共MSQ兲 have been discussed in previous publications.1,2 Organic materials have a lower dielectric constant (k⬃2 – 3) and will a兲
Current address: IBM Microelectronics, 2070 Route 52, Mail Stop E40, Hopewell Junction, New York 12533; electronic mail:
[email protected] b兲 Author to whom correspondence should be addressed; current address: Department of Materials and Nuclear Engineering and Institute for Plasma Research, University of Maryland, College Park, Maryland 20742-2115; electronic mail:
[email protected] c兲 Current address: Air Products and Chemicals, Inc., Santa Clara, California 95054. 435
J. Vac. Sci. Technol. A 19„2…, MarÕApr 2001
enable the production of high-speed devices. The process flow of organic dielectrics, however, may differ significantly from the process flow developed for the integration of conventional metals and dielectrics. The challenge is thus to develop consistent process sequences for organic dielectrics that require minimum and low-cost modifications of the existing production lines. The aim of this article is to examine the patterning of organic thin films in oxygen-based high-density plasmas (O2 /Ar and O2 /N2 ). The use of oxygen in nonoptimized processes can result in serious erosion and bowing of feature sidewalls.3–8 However, the formation of straight sidewalls is feasible in two different process windows which will be discussed along with a deeper understanding of the etch mechanism for organic films. The studied organic dielectrics are polytetrafluoroethylene 共PTFE兲, parylene-N, and poly共arylene ether兲 共PAE-2兲. Additionally, a few results will be shown on parylene-AF4. The structure of each polymer is shown in Fig. 1. Parylene-N and parylene-AF4 are both vapor-deposited films with dielectric constants of 2.7 and 2.3, respectively. Even though both polymers can be successfully integrated, integration in commercial devices is unlikely due to relatively high production costs.9 PAE-2 and PTFE are deposited by spin-on and subsequent drying and curing techniques.10,11 PAE-2 (k⫽2.8) suffers from a relatively low glass transition temperature 共⬃270 °C兲. Recently, Schumacher introduced VELOX as a variant on PAE-2. The chemical structures of VELOX and PAE-2 are very similar, although the glass transition tem-
0734-2101Õ2001Õ19„2…Õ435Õ12Õ$18.00
©2001 American Vacuum Society
435
436
Standaert et al.: High-density plasma patterning
436
FIG. 2. Patterning scheme for organic dielectrics.
FIG. 1. Structure of parylene-N, parylene-AF4, PAE-2 and PTFE. The density and the dielectric constant of the films are indicated by and k, respectively.
perature of VELOX is much higher 共⬃400 °C兲. A comparison between PAE-2 and VELOX showed that both polymers are very similar in terms of etch rates and profile control.12 PTFE, trademarked as SPEEDFILM by W. L. Gore & Associates, has a dielectric constant of 2.0, which is extremely low for a nonporous material. II. EXPERIMENTAL SETUP AND PROCEDURES The transformer-coupled plasma 共TCP兲 reactor used for this study has been described in detail elsewhere.1,13 All experiments were carried out in discharges maintained at a source power of 600 W 共13.56 MHz兲. A constant gas flow of 40 sccm O2 /Ar/N2 mixtures was injected in the discharge region. The discharge is laterally confined by an alumina ring and vertically by the wafer and the quartz window. Due to the restriction in pumping speed a pressure differential builds up between the discharge and the area just outside the confinement. All experiments described here were conducted at a pressure of 4 mTorr, which was measured outside the J. Vac. Sci. Technol. A, Vol. 19, No. 2, MarÕApr 2001
confinement area. Even though unknown, it is expected that the pressure at the wafer surface runs higher than 4 mTorr. Samples with the polymer films were placed in the center of a 125 mm silicon wafer located 7 cm below the plasma generation region. The wafer was clamped on an electrostatic chuck and cooled at 10 °C. The ion energy at the wafer can be controlled independently of the plasma generation by rf biasing.13 The rf bias power on the wafer was varied between 0 and 100 W at a fixed frequency of 3.4 MHz. The self-bias voltage on the wafer was measured with a probe. A Balzers PPM422 plasma process monitor was installed next to the wafer surface and was utilized to obtain the composition of the ion flux. Furthermore, etch products of the blanket polymer films were monitored by spatially resolved mass spectrometry. The setup of this mass spectrometer is discussed by Li et al.14 A brief description of this spectrometer will be given in Sect. III B. Etch rates of blanket samples were obtained using in situ He–Ne ellipsometry. In the etching of these organic dielectrics only a marginal etch selectivity with respect to photoresist can be achieved. Hence, patterning requires an additional layer between the resist and the polymer. This layer is typically a spin-on glass or a nitride layer and is referred to as a hard mask. The patterning occurs in two steps and is schematically outlined in Fig. 2. First, the trilayer structure is processed in a CF4 discharge 共1000 W source power and 100 W bias power兲 where the resist pattern is transferred into the hard mask. The low polymerization rate of a CF4 discharge leaves a minimum amount of residual fluorocarbons on the wafer and reactor walls. The use of CF4 leads to high etch rates of the hard mask material. Dilution of CF4 in argon 共1:1 ratio兲 reduces the etch rate and allows for longer, more controllable, processing times on the order of 20–40 s. The pattern in the hard mask is subsequently transferred into the polymer using an oxygen-based chemistry. The resist erodes quickly, whereas the hard mask has a high etch resistance in the discharge. For the patterning of parylene-N, a 100 nm nitride mask was applied. PTFE and PAE-2 were both spin coated with a spin-on glass film of 100 and 200 nm, respectively. Cleaving patterned polymer films for cross-sectional secondary electron microscopy 共SEM兲 often requires special techniques such as focused ion beam 共FIB兲 to prevent tearing of the film. In this work, patterned samples were frozen by submerging them in liquid nitrogen and were then cleaved using a regular graphite tip.
437
Standaert et al.: High-density plasma patterning
FIG. 3. Characterization of the ion flux as a function of O2 -flow percentage: ion current density 共ICD兲 共a兲, and ion composition of the ion flux 共b兲. Source power, pressure, and total gas flow (Ar/O2 ) were fixed at 600 W, 4 mTorr, and 40 sccm, respectively.
III. RESULTS A. Gas-phase characterization
The ion flux to the wafer in high-density plasma systems is not affected by the rf bias power and a linear relationship between self-bias voltage and rf bias power is obtained. The ion current density can be calculated from the slope of this linear relationship.13,15 The ion current density in Ar/O2 plasmas is plotted as a function of the relative O2 flow in Fig. 3共a兲. The source power, pressure, and total gas flow of Ar/O2 mixtures were fixed at 600 W, 4 mTorr, and 40 sccm, respectively. The ion current density drops significantly as the O2 flow is increased. Ions were sampled by a Balzers PPM422 system. In the PPM system, ions pass a retarding electric field and subsequently a cylindrical mirror analyzer. The pass energy of the analyzer was fixed while the magnitude of the retarding electric field was varied. After the energy selection, ions were detected by a quadrupole mass spectrometer. Ion energy spectra for various O2 flows up to 25% showed an almost constant plasma potential of 20 V with respect to ground. Figure 3b shows the integrated mass spectrometer intensities ⫹ ⫹ ⫹ for O⫹ and O⫹ 2 divided by the total O , O2 , and Ar in⫹ ⫹ tensity. The contribution of O and O2 to the ion flux increases linearly with O2 flow. JVST A - Vacuum, Surfaces, and Films
437
Threshold-ionization mass spectrometry was applied to obtain the qualitative behavior of the radical densities in the Ar/O2 plasmas. Neutral species were sampled by the PPM422 system and subsequently bombarded by electrons with a controllable energy. For electron energies below the threshold for ionization from the ground state only excited species are ionized and detected in the mass spectrometer. It was found that both O* and O2 density are proportional to the O2 flow. This method, unfortunately, does not reveal absolute densities. In order to estimate the degree of dissociation other methods need to be applied. A rough estimate can be obtained by monitoring the pressure increase 共drop兲 when the plasma is turned on 共off兲 while the pumping speed is fixed. Using this method we noticed only a small pressure change 共⬍0.1 mTorr兲, suggesting a degree of dissociation below 3%. A different method compares the intensities of the parent ion ⫹ peaks (O⫹ 2 and Ar ) in the case of plasma on and plasma off using the line-of-sight mass spectrometer of the PPM422 system. This method leads to a similar result as was obtained by monitoring the pressure change. A similar degree of dissociation in inductively coupled Ar/O2 plasmas has also been measured by Wang, Van Brunt, and Olthoff16 using mass spectrometry. A low degree of dissociation in these discharges may be unexpected. Dissociation of the oxygen molecule in the ground state requires only 5 eV and modeling of inductively coupled Ar/O2 discharges suggests a very high degree of dissociation close to 100%. 17 Additionally, Fig. 3共b兲 shows that O⫹ can form up to 10% of the ion flux and suggests a reasonable degree of dissociation. Dissociative ionization of molecular oxygen requires almost 19 eV and is less likely to occur than nondissociative ionization of molecular oxygen which requires only 12 eV. Hence, O⫹ most likely originates from ionization of oxygen atoms. Finally, argon actinometry using the oxygen emission line at 777 nm showed an almost linear increase in the ground state oxygen density when the source power was increased from 600 to 1400 W. This measurement suggests that the degree of dissociation at 600 W is at least below 50%. Note that the contribution of dissociative excitation of oxygen can be neglected in inductively coupled discharges as it requires a high electron energy 关⬃15 eV 共Ref. 18兲兴 relative to direct excitation 关⬃11 eV 共Ref. 18兲兴.19 At this point it is not clear why the various estimates for the degree of dissociation are so different. One possibility may be that associative reactions of oxygen, including surface recombination, in inductively coupled discharges have been underestimated. Absolute density measurements need to be performed to measure the degree of dissociation directly. B. Etch rates and etch products of blanket films
The etch rates of parylene-N, parylene-AF4, PAE-2, and PTFE as a function of rf bias power are plotted in Fig. 4 for various Ar/O2 flow ratios. It is important to note that the PTFE etch rates are plotted on a different scale. All films show a relatively low etch rate when no O2 is added to the
438
Standaert et al.: High-density plasma patterning
438
FIG. 5. Setup of the linear motion mass spectrometer.
FIG. 4. Blanket etch rates of parylene-N 共Pa-N兲, parylene-AF4 共Pa-AF4兲, PAE-2, and PTFE as function of the self bias voltage. The etch rates were measured for various O2 percentages in Ar/O2 plasmas.
Ar discharge. In this case, the etch rate or, better, the physical sputter rate, may be enhanced by residual water and oxygen trapped in the chamber and by small amounts of oxygen released at the quartz-coupling window.20 The etch rates exhibit a maximum around 25–50 W corresponding to an average ion energy of approximately 15–25 eV. Above this energy, a high etch rate was observed which gradually slowed down to the constant values presented in Fig. 4. At these higher energies ions may induce cross linking of polymer chains which results in a more etch resistant overlayer. The presence of this overlayer can be seen by switching from pure Ar to a more reactive feed gas chemistry, for example, Ar/O2 . Initially, a slow etch rate is observed which gradually increases to the value that is measured without the preexposure to a pure Ar plasma. Unfortunately, the thickness of the overlayer cannot be determined by this experiment, since etching of the polymer may occur when this overlayer becomes thin enough. The addition of 6% O2 to the Ar discharge significantly enhances the polymer etch rates, which sublinearly increase as more O2 is admixed. The etch rate as function of the rf bias power exhibits also a sublinear relationship in the case of parylene-N, parylene-AF4, and PAE-2. In contrast, PTFE etches 2–3 times faster and is linearly dependent on the rf bias power. In order to detect etch products in Ar/O2 plasmas, the sampling system of the spatially resolved mass spectrometer was positioned 6 mm above the polymer sample 2.5 cm⫻2.5 cm in size, see Fig. 5. The mean-free path of neutral species J. Vac. Sci. Technol. A, Vol. 19, No. 2, MarÕApr 2001
at 4 mTorr is on the order of 1–2 cm and is larger than the distance between the orifice and sample. As a result, a significant portion of the species detected in the mass spectrometer were either reflected or created at the sample surface. The sheath thickness in high-density plasmas is on the order of 0.1–0.5 mm and is not affected by the position of the linear-motion mass spectrometer. The electron-impact ionizer of the mass spectrometer was set at 30 eV. Figure 6 shows the spectrometer output during the etching of a PAE-2 film. In the first step 共I兲 PAE-2 is etched in a pure Ar discharge at 0 W rf bias power. The initial high levels of CO can be attributed to oxygen and water that adsorbs at the reactor walls during loading of the sample. In the second step 共II兲 the bias power is increased up to 50 W, but no significant change in the mass spectrometer signals is observed. Finally, O2 is injected 共region III兲 which yields a significant increase in etch rate. CO, CO2 , and H2 O 共the latter ones are not shown in Fig. 6兲 were the major etch products detected by the linear-motion mass spectrometer. In the case of parylene-
FIG. 6. Ar, O2 , and CO detected by the linear motion mass spectrometer during the etch back of a PAE-2 film. 共I兲 1:0 Ar:O2 , 0 W bias; 共II兲 1:0 Ar:O2 , 50 W bias; 共III兲 3:1 Ar:O2 , 100 W bias; and 共IV兲 etching of SiO2 , 3:1 Ar:O2 , 100 W bias.
439
Standaert et al.: High-density plasma patterning
FIG. 7. O2 and C2 H2 detected by the linear motion mass spectrometer during the etch back of a PAE-2 film. 共I兲 1:0 Ar:O2 , 0 W bias; 共II兲 1:0 Ar:O2 , 50 W bias; 共III兲 3:1 Ar:O2 , 0 W bias; 共IV兲 3:1 Ar:O2 , 50 W bias; and 共V兲 etching of SiO2 , 3:1 Ar:O2 , 50 W bias.
AF4 and PTFE, fluorinated etch products, such as HF and COF2 , were also detected. At the end of region III the underlying SiO2 is reached. This is followed by a drop in CO intensity and an increase in O2 intensity in region IV, while the measured Ar signal remains at a constant level. Since a significant portion of the O2 detected in the mass spectrometer reflects from the sample surface, it can be concluded that oxygen is an important etchant. It cannot be concluded that molecular oxygen is an important etchant as the recombination rate of atomic oxygen may be high in the sampling tube of the spatially resolved mass spectrometer. Only if the degree of dissociation of the Ar/O2 discharge is low, then Fig. 6 shows that molecular oxygen is consumed at the polymer interface. This is consistent with other studies.21–25 The role of atomic oxygen could not be established with spatially resolved mass spectrometry due to the low detection levels and the possible recombination of atomic oxygen in the sampling tube of the mass spectrometer. Cook and Benson26 and Selwyn27 have established that atomic oxygen is an important etchant. Tepermeister and Sawin28 observed in a study on polyimide etching in Ar/O2 plasmas that the formation of acetylene (C2 H2 ) exhibits the strongest dependence on the ion energy. Hence, formation of acetylene could be an important reaction channel for the abstraction of aromatic groups from parylene-N, parylene-AF4, and PAE-2. To verify if the etch rate of these polymers is indeed governed by the formation of acetylene, a PAE-2 sample was etched back in four steps. The O2 and C2 H2 levels detected by the linear motion mass spectrometer are shown in Fig. 7. The electron-impact ionizer of the mass spectrometer was set at 16 eV. In the first step 共region I兲 PAE-2 is etched in a pure Ar discharge at 0 W rf bias power. In line with the observations made by Tepermeister and Sawin, the C2 H2 level increases significantly as the rf bias power is increased to 50 W in the second step 共region II兲 and is accompanied with an increase in etch rate, as was shown in Fig. 4. The formation of acetylene could JVST A - Vacuum, Surfaces, and Films
439
thus be an important reaction channel in a pure Ar discharge. In the third step 共region III兲 O2 is admixed with Ar 共3:1 Ar:O2 ) and the rf bias power is reduced down to 0 W and no significant amount of acetylene is detected. As the bias power is increased up to 50 W 共region IV兲 the C2 H2 increases, indicating that the C2 H2 formation is indeed strongly dependent on the ion energy. However, the C2 H2 level does not reflect the etch rates observed in Fig. 4. It is, therefore, unlikely that the formation of acetylene is a dominant reaction channel in Ar/O2 discharges. Instead, carbon and hydrogen are primarily abstracted from the polymer through the formation of CO, CO2 , and H2 O. The last region 共V兲 in Fig. 7 shows the acetylene and O2 level after the PAE-2 film is etched back to the underlying SiO2 film. The increase in O2 level corresponds to a higher reflection probability of 共atomic or molecular兲 oxygen from an oxide surface than from a reactive polymer surface. C. Surface analysis
The surface of PAE-2 films after a partial Ar/O2 etch was characterized by x-ray photoelectron spectroscopy 共XPS兲. The etching for these experiments was performed in a different TCP reactor, but the processing conditions were almost identical as for the experiments described above. Partially etched PAE-2 films were transferred under ultra-high vacuum conditions to a Vacuum Generators ESCA Mk II analysis chamber. Photoemission spectra at a grazing emission angle of 22° were acquired using a nonmonochromatized Mg K␣ source 共1253.6 eV兲. The resolution is in this case limited by the linewidth of the Mg K ␣ source and is approximately 1 eV. Figure 8 shows the C共1s兲 and the O共1s兲 photoemission spectra for PAE-2 films unprocessed and partially etched back at ⫹9, ⫺50, and ⫺100 V self-bias voltage and a 3:1 Ar/O2 flow ratio. The spectra are corrected for charging by positioning the C共1s兲 emission from the aromatic groups at 284.8 eV.29 It is interesting to note the shake up around 292 eV in the C共1s兲 spectrum of the unprocessed PAE-2 film. This shake up is typical for polymers with a high degree of aromaticity and arises from an inelastic loss process that excites the ground-state ring orbitals of the aromatic groups in the polymer.29 The shake-up feature is significantly reduced after processing in Ar/O2 and indicates that the polymer surface is severely modified by the plasma exposure. More-detailed information can be obtained from the O共1s兲 spectra. The photoemission for the unprocessed PAE-2 film is centered around 533.5 eV, corresponding to C–O–C bonds where the carbon atoms belong to an aromatic group.29 After the PAE-2 film is partially etched back at a floating potential of ⫹9 V 共with respect to ground兲, a significant emission at higher binding energies is observed. This can be attributed to the formation of a surface layer that is heavily oxidized and is rich in carbon-oxygen bonds where the carbon atom is bonded to two or three oxygen atoms. The peak positions for these bonds can typically be found in a range up to approximately 534 eV.29 In this case, however, an additional, unknown contribution at 535–536 eV is
440
Standaert et al.: High-density plasma patterning
440
FIG. 9. Two examples of parylene-N etched in an Ar:O2 discharge 共1:1兲 maintained at 4 mTorr, and 600 W source power. The rf bias power was 0 W for 共a兲 and 100 W for 共b兲. The plasma exposure was fixed at 200 s in both cases.
FIG. 8. Carbon 1s 共a兲 and oxygen 1s 共b兲 photoemission spectra for PAE-2 after processing in Ar/O2 as a function of self-bias voltage. The collection angle is 22° with respect to the sample surface.
clearly present. It could be speculated that it originates from hydroxyl groups which are bonded to the same carbon atom, i.e., C共OH兲2 and C共OH兲3 . As the bias power is increased, the O共1s兲 emission shifts towards lower binding energies. In addition, the integrated O共1s兲 intensity decreases and is accompanied by an increase in the integrated C共1s兲 intensity. The lower binding energies correspond to bonds such as C–OH 共⬃532.9 eV兲, C–O–C 共⬃532.6 eV兲 and C⫽O 共⬃532.3 eV兲, where the carbon atoms are aliphatic.29 Hence, it can be concluded that as the ion energy increases a less heavily oxidized surface is obtained where most carbon atoms are bonded to at most one oxygen atom. D. Undercutting and erosion of hard mask
It is well known in the literature that anisotropic etching of polymers in oxygen-based plasmas is hard to realize.3–8 An example is shown in Fig. 9共a兲, where parylene-N was patterned using an Ar/O2 discharge 共1:1兲 without applying a rf bias to the substrate. The hard mask is seriously undercut and sidewall bowing can be observed. It is interesting to note that parylene-N exhibits a high lateral erosion rate directly below the hard mask, even though this position is almost completely shadowed from ions and neutrals emerging from the gas phase. Hence, atomic and molecular oxygen reflecting from the feature bottom are likely to be involved in the lateral erosion process. Pons, Pelletier, and Joubert5 suggested that in the absence of ion impact the desorption of oxidized products is induced by the absorption of ultraviolet light. The sidewall erosion can be minimized by increasing the rf bias power as is demonstrated in Fig. 9共b兲. Since the J. Vac. Sci. Technol. A, Vol. 19, No. 2, MarÕApr 2001
plasma exposure time in Fig. 9共a兲 and 9共b兲 was fixed at 200 s, it can be concluded that the lateral erosion rate of parylene-N depends on the ion energy and is consistent with the work of Heidenreich et al.3 and Joubert et al.4 The spatially resolved mass spectrometry data in Fig. 7 show that less oxygen reflects from the feature bottom at higher ion energies. This will lower the flux of oxygen to the sidewalls and reduce the lateral erosion rate during the polymer etch. However, the polymer etch in Fig. 9共b兲 is completed in approximately 40 s. After this, the oxygen flux to the sidewall increases and there must be a second mechanism by which the sidewall is protected from the oxygen flux during the remaining 160 s overetch. We will argue in Sec. III E that in an Ar/O2 discharge the sidewall can be passivated by the redeposition of oxygen-deficient etch products. At higher ion energies it is more likely that partially oxidized and, hence, involatile products, desorb from the feature bottom and redeposit on the sidewall. In Sec. III E we will also show that the redeposited material has a relatively high etch resistance. Involatile etch products can, unfortunately, not be detected by spatially resolved mass spectrometry. Hence, the exact composition and flux of these etch products could not be established. Two comments on the effect of ion energy should be made here. First of all, the profile control is not always improved at higher ion energies. An example is shown in Fig. 10 where PTFE was patterned in an Ar/O2 discharge 共3:1兲 using two different rf bias powers, 0 and 50 W. The etching time was fixed at 200 s. In this case, both rf bias powers yield a similar amount of undercutting. A possible explanation may be found in terms of the difference in stoichiometry: CH for parylene-N and CF2 for PTFE. PTFE contains relatively less carbon and the etch products from the feature bottom may be more volatile than in the case of parylene-N. In Sec. III E it will be shown that the redeposition rate for PTFE can very high in oxygen-deficient discharges. It also should be noted that the ion energy cannot be increased too high. The sputter yield increases at higher ion energies and results in the tapering of the hard mask corners. This can result in the loss of the critical dimension, as is demonstrated in Fig. 10. Figure 11 shows the sputter yield of
441
Standaert et al.: High-density plasma patterning
441
FIG. 10. Two examples of PTFE etched in an Ar:O2 discharge 共3:1兲 maintained at 4 mTorr, and 600 W source power. The rf bias power was 0 W for 共a兲 and 50 W for 共b兲. The plasma exposure was fixed at 200 s in both cases.
blanket SiO2 films in Ar, N2 , O2 , and H2 discharges as a function of self-bias voltage. Due to its relatively high atomic mass, Ar exhibits the highest sputter rate. Hence, in the process design for the patterning of polymers, one should carefully choose the Ar flow and ion energy in order to prevent excessive tapering of the hard mask. For the patterning of PTFE, the erosion of the hard mask may be enhanced by the fluorine from the PTFE as long as the ion energy is high enough. E. Methods of sidewall passivation
Patterning of organic dielectrics in oxygen-based plasmas is favored for the relatively high etch rates and the good selectivity to hard mask and etch stop layers. Unfortunately, sidewall bowing and undercutting is often observed and
FIG. 11. Sputter yield of SiO2 as a function of self-bias voltage in Ar, N2 , O2 , and H2 plasmas maintained at a source power of 600 W, a gas flow of 40 sccm, and a pressure of 4 mTorr. JVST A - Vacuum, Surfaces, and Films
FIG. 12. Cross-sectional SEM photographs of PTFE patterned in a pure Ar discharge at 200 W for 80 s.
means of sidewall passivation are necessary for successful pattern transfer. Namatsu, Ozaki, and Hirata30 showed that adequate profile control can be obtained using hydrocarbon– oxygen mixtures as an etching gas. Joubert et al.4 proposed a different approach using SO2 /O2 mixtures. This section discusses two alternative mechanisms allowing for the formation of straight sidewalls. The first mechanism of sidewall passivation is observed for the patterning of PTFE. After the opening of the hard mask in an Ar/CF4 discharge, the pattern was transferred into the 500-nm-thick PTFE film at 200 W rf bias power in a pure Ar-discharge. Figure 12 shows that in this case straight sidewalls are obtained. The photoresist pattern eroded only partially and needs to be removed. The resist was stripped in a pure O2 plasma at a low source and bias power. Figure 13 shows that the resist is successfully removed but it also shows that veils are covering the sidewalls. A prolonged exposure to the O2 plasma reduces the thickness of these veils and eventually erodes the protruding parts completely. The protruding parts are first removed in the smaller features, after which the PTFE behind the remaining part of the veil starts to erode. Via cleaning in a high-density plasma source using O2 is, therefore, not successful for PTFE, since the veil removal is anisotropic and feature dependent. However, the fact that the veils have a relatively low erosion rate and the fact that the thickness of the veils increases as a function of
442
Standaert et al.: High-density plasma patterning
FIG. 13. Cross-sectional SEM of PTFE after patterning in a pure Ar discharge followed by an ashing step in a pure O2 discharge.
feature size, provide important information about the nature of the veils. It suggests that the veils are formed by redeposition of fluorine-deficient sputter products from the feature bottom. Assuming that the thickness is proportional to the volume etched in the feature and inversely proportional to the area of the sidewall, it can be argued that the thickness of veils scales with the feature width. This type of veil formation seems also to occur in Ar/O2 discharges. Figure 14 shows an unusual case for a partially etched contact hole in PAE-2 using an Ar/O2 discharge with 25% O2 addition. The veil can be seen near the corners of the hard mask, since it is partially eroded. As a result, the PAE-2 has been slightly undercut. We, therefore, suggest that also in Ar/O2 discharges redeposition of carbon-rich products occurs. However, the redeposition rate is too low to completely prevent sidewall bowing.
FIG. 14. Partially etched contact hole in PAE-2 where the veil is exposed at the corners of the hard mask. J. Vac. Sci. Technol. A, Vol. 19, No. 2, MarÕApr 2001
442
FIG. 15. Trench etched in PAE-2 using N2 /O2 discharge 共10:1兲 共40 sccm total gas flow, 4 mTorr, 600 W source power, and 50 W rf bias power兲.
Successful pattern transfer into PAE-2 is possible using N2 /O2 plasmas as is shown in Fig. 15. The N2 /O2 ratio was 10:1. In this case, no bowing is observed and straight sidewalls are obtained. The N2 /O2 flow ratio has also important implications for the bottom profile. Figure 16 shows trenches etched in N2 /O2 plasmas for various flow ratios while etching time and other parameters were fixed. A pure O2 discharge yields the highest etch rate, but results in a significant amount of sidewall bowing and undercutting. The PAE-2 etch rate decreases as the N2 /O2 flow ratio is increased. The bowing and undercutting disappears for high N2 /O2 flow ratios. It is interesting to note that microtrenching appears as soon as the sidewalls are in line of sight of the ions. Microtrenching has been explained by differential charging.31,32 This mechanism assumes that the areas near the mask material charge up negatively due to a relatively isotropic velocity distribution of the electrons, while the feature bottom is positively charged by ions which have a highly directional velocity distribution. Due to the charge buildup more ions are focused at the corners of the feature bottom. A second mechanism which can account for microtrenching is reflection of ions from slightly inclined sidewalls.33–35 Bogart et al.36 recently showed for Cl2 -based etching of silicon that differential charging is not the primary cause of microtrenching by demonstrating that the formation of microtrenching is similar for both conductive and oxide masks. A similar conclusion can be drawn here. Even though differential charging should occur for the trenches shown in Figs. 16共a兲 and 16共b兲, no microtrenching is observed. The microtrenching in Fig. 16共c兲 seems, therefore, more consistent with sidewall reflection than with differential charging. A study of the underlying mechanism of sidewall passivation in N2 /O2 discharges is currently underway and will be addressed in a future publication.
443
Standaert et al.: High-density plasma patterning
443
FIG. 17. Etched depth in trenches and via holes of various widths in PAE-2 as a function of time in an Ar/O2 共3:1兲 discharge 共40 sccm total gas flow, 4 mTorr, 600 W source power, and 50 W rf bias power兲. The solid lines show the model prediction for the smallest and largest features according to Eqs. 共5兲 and 共6兲.
behavior is consistent with neutral shadowing, the respective roles of ions and oxygen neutrals in the etching process have to be clarified first. IV. MECHANISM OF ETCHING IN ARÕO2 PLASMAS FIG. 16. Partially etched trenches in PAE-2 for various ratios of N2 /O2 : 共a兲 0:1, 共b兲 1:1, and 共c兲 10:1. 共40 sccm total gas flow, 4 mTorr, 600 W source power, 50 W rf bias power, and 150 s etching time兲.
F. Aspect-ratio-dependent etching
The results from spatially resolved mass spectrometry in Sec. III B clearly show that oxygen neutrals play an important role in the polymer etching process. Consequently, it can be anticipated that feature etching differs from blanket etching due to neutral shadowing. Figure 16共a兲 shows a partially etched trench in PAE-2 using a pure O2 discharge. In addition to mask undercutting and sidewall bowing, it can be seen that the feature bottom is nonuniformly etched. The curved bottom profile can be explained by the fact that the direct neutral flux from the gas phase is more shadowed in the corners than in the center of the feature bottom. The time dependence of the etched depth measured in the center of trenches and via holes in PAE-2 is shown in Fig. 17. The etching is performed in a 4 mTorr Ar/O2 discharge 共3:1 flow ratio兲 at 600 W source power and 50 W rf bias power. The resists pattern erodes completely in the first 50 s. As time progresses, the PAE-2 etch rate slows down more significantly in the narrower features and via holes. To verify if this JVST A - Vacuum, Surfaces, and Films
Oxygen-based etching of polymers in various types of reactors has been modeled by several researchers using Langmuir adsorption kinetics.22,37,38 Motivated by their work, we were able to successfully model the etch rate dependence on ion flux, oxygen flow and feature geometry in Ar/O2 plasmas. The organic dielectrics, especially PTFE, are known to have a high resistance to chemical attack.39,40 A chemical reaction between the polymer surface and oxygen occurs only at elevated temperatures or in conjunction with ion, electron, or photon bombardment. The etch rates presented in this article were obtained at a relatively low substrate temperature of 10 °C. It is assumed that the reaction between polymer and oxygen neutrals is predominantly initiated by ion impact and can be described by a surface coverage model. The model fails, therefore, to describe the etch rates of surfaces that are not in line of sight of the ion trajectories, for example, undercut sidewalls. The characterization of the gas phase in Secs. III A and III B is inconclusive about the absolute atomic oxygen density. In addition to atomic oxygen and ground-state molecular oxygen O2 ( 3 ⌺ ⫺ g ), the absolute density of excited O2 ( 1 ⌬ g ) may play an important role in the etching process. The low-lying state 1 ⌬ g is separated by only 1 eV from the ground state and is easily accessible through electron excitation. The transition from 1 ⌬ g to the ground state is strongly
Standaert et al.: High-density plasma patterning
444
444
forbidden. Hence, the internal energy of O2 ( 1 ⌬ g ) can be used during a reaction at the polymer surface. At this point, it is assumed that the flux of the various oxygen neutral species is linear in the O2 flow. The total oxygen flux will be denoted as 2⌫ O2 , where ⌫ O2 is the flux of molecular oxygen to the surface when no dissociation or excitation would take place in the gas phase. The energy from a single ion impact is dissipated on a time frame that is many orders of magnitude shorter than the characteristic time for the next impact of a gas particle.24 Hence, the interaction of ions and oxygen at the polymer surface has to occur in at least two steps. Gokan and Esho22 suggested that in the first step free-radical sites are created via hydrogen abstraction by energetic ions. In the second step oxygen reacts on the radical sites to form volatile species, such as CO. Vanderlinde and Ruoff37 and Joubert, Pelletier, and Arnal38 suggested a different reaction pathway where oxygen first adsorbs at the polymer surface. The reaction between adsorbed oxygen and the polymer surface is then initiated by the impact of energetic ions. The model presented here is based on this two-step reaction pathway, since it predicts that the surface oxidation is reduced as the ion energy is increased. The reaction pathway suggested by Gokan predicts the opposite and is in disagreement with the XPS results presented in Sec. III C. Other, more-complicated reaction pathways can be considered. However, based on the simple two-step reaction, where oxygen first adsorbs and then desorbs in an etch product upon ion impact, it is possible to explain the observed etch rate behavior of both blanket and patterned films. Given the oxygen flux 2⌫ O2 and total ion flux ⌫ i , the surface coverage of sites where oxygen is adsorbed can be calculated by balancing the adsorption and desorption rate of oxygen:
d ⫽2⌫ O2 S 共 1⫺ 兲 ⫺⌫ i Y i ⫽0, dt
共1兲
where is the highest attainable surface density of adsorbed oxygen. S is an effective sticking probability for all the various oxygen neutral species at unoccupied sites. Y i is the reaction probability between an adsorbed oxygen and the polymer during an ion impact. Since the energy of a single ion is dissipated by several surface atoms, reactions can be initiated at sites that are not directly at the center of impact. The area affected by ion impact and, consequently, Y i , are dependent on the ion energy. Here, it is assumed that Y i is a linear function of the ion energy E i : Y i⫽
Ei , E0
共2兲
where E 0 is a constant. The surface coverage is then given by
冉
⫽ 1⫹
⌫ iE i 2S⌫ O2 E 0
冊
⫺1
.
共3兲
Equation 共3兲 predicts that the surface coverage decreases as the ion energy is increased, which is consistent with Sec. J. Vac. Sci. Technol. A, Vol. 19, No. 2, MarÕApr 2001
III C. At this point, it is convenient to define the etch yield Y L as the amount of carbon atoms removed from the polymer surface per single ion impact event. The yield Y L according to Eq. 共1兲 is then given by
冉 冊 冉 冊
2S ⌫ O2 ␣ ⌫i , Y L⫽ E 0 ⌫ O2 1⫹2S Ei ⌫i
共4兲
where ␣ is a dimensionless constant accounting for the number of carbon atoms per site and for the number of oxygen atoms required to remove a single carbon atom. The composition of the ion flux has been neglected up to this point. However, it is very likely that the impact of oxygen ions (O⫹ and O⫹ 2 ) immediately results in the formation of volatile products and does not require the assistance of adsorbed oxygen neutrals. Baggerman and co-workers24 showed that the etching reactions occurring at the surface are exothermic. The reaction with oxygen ions should, therefore, already occur at relatively low ion energies. It also should exhibit a weak dependence on ion energy as the reaction is limited by the number of oxygen atoms that the ion carries. It is assumed that the rate constant for this reaction is independent of the ion energy for the experimental conditions used in this work. Including this reaction, the total etch yield Y T is now given by
冉 冊 冉 冊
1 ⌫ O⫹ 2 ⌫ O⫹2 Y T⫽ ⫹ ␣ ⌫i ␣ ⌫i
冉 冊 冉 冊
2S ⌫ O2 ␣ ⌫i ⫹ , E 0 ⌫ O2 1⫹2S Ei ⌫i
共5兲
where ⌫ O⫹ and ⌫ O⫹ are the fluxes of O⫹ and O⫹ 2 ions, 2 respectively. To verify this semiempirical model, the blanket etch rates in Fig. 4 have been reduced to yields in Fig. 18. The average ion energy was calculated from the self-bias voltage and the plasma potential. The total ion flux ⌫ i was calculated from the ion current density in Fig. 3共a兲 and is listed in Table I. The oxygen ion fluxes ⌫ O⫹ and ⌫ O⫹ are obtained from Fig. 2 3共b兲. The oxygen flux ⌫ O2 was calculated from the partial pressure assuming a gas temperature of 500 K and a wafer area pressure of 4 mTorr. Equation 共5兲 was then fitted to the data in Fig. 18 for each organic dielectric by varying the parameters ␣, E 0 , and S using a least-squares method. The best fits were obtained where the sticking coefficient was on the order of 1. This suspiciously high value may be explained in part by the semiempirical nature of the model. Additionally, the sticking coefficient may actually be lower since the wafer area pressure may run much higher than 4 mTorr as was mentioned in Sec. II. The fitted values for ␣ and E 0 vary around 0.6 and 70 eV, respectively. It was not possible to fit the yields for PTFE with Eq. 共5兲. In contrast to the other dielectrics, the yield for PTFE depends almost linearly on the ion energy, and an increase in the O2 -flow ratio from 6% to 25% has a relatively small
445
Standaert et al.: High-density plasma patterning
445
at the feature bottom due to shadowing. The direct flux of oxygen ⌫ Of at the center of the feature bottom for trenches 2 and via holes is given by41,42
⌫ Of ⫽ 2
FIG. 18. Etch yields of parylene-N 共Pa-N兲, parylene-AF4 共Pa-AF4兲, PAE-2, and PTFE as a function of the average ion energy for various O2 percentages in Ar/O2 plasmas. The solid lines are fits according to Eq. 共5兲, the dashed lines are just a guide for the eye.
impact on the etch yield. Clearly, PTFE is etched through a different mechanism which is more dependent on the ion energy. A possible explanation for the different etching behavior may be found in the relatively large amount of fluorine that is incorporated in the PTFE. Comparison between parylene-N and parylene-AF4 shows that the incorporation of small amounts of fluorine does not have a significant impact on the etch yield. Further validation of the suggested reaction kinetics is obtained by extending the blanket formalism to the feature etching in PAE-2, for which the experimental results were shown in Sec. III F, Fig. 17. Because of the relatively large feature sizes 共⬎1 m兲, it is assumed the ion flux at the center of the feature bottom is the same as for a blanket film and is not affected by charging or reflection from the sidewall. On the other hand, the neutral flux can be significantly reduced TABLE I. Ion and molecular oxygen fluxes in Ar/O2 plasmas maintained at 40 sccm total gas flow, 4 mTorr pressure, and 600 W source power ⌫ O⫹ O2 flow ⌫I ⌫ O⫹ ⌫ O2 2 共%兲 (1017 cm-2 s-1 ) (1017 cm-2 s-1 ) (1017 cm-2 s-1 ) (1017 cm-2 s-1 ) 0 6.25 12.5 25.0 a
1.28 1.11a 0.94 0.78
0.002 0.028a 0.053 0.079
Interpolated values.
JVST A - Vacuum, Surfaces, and Films
0.001 0.031a 0.061 0.114
0 0.69 1.38 2.76
冦
冉 冊 冉 冊
⌫ O2 1⫹
4L 2
⌫ O2 1⫹
⫺1/2
,
w2
4L 2 w2
trench, 共6兲
⫺1
,
hole,
where L and w are the feature depth and width, respectively, and ⌫ O2 is the oxygen flux on a blanket surface. The etch rate in features can now be calculated by replacing ⌫ O2 in Eq. 共5兲 with ⌫ Of . Subsequently, this depth-dependent etch 2 rate can be integrated over time to obtain the etched depth as function of time. The sacrificial resist layer on the oxide/ PAE-2 stack erodes within the first 50 s of the Ar/O2 etch. Since it is hard to account for this rapid erosion, the time integration of Eqs. 共5兲 and 共6兲 was started after 50 s using the measured depth as the boundary condition. The sticking probability S was set at 1. The parameters ␣ and E 0 were set at the values as obtained from the blanket fits in Fig. 18. The time integration was performed for the smallest and the largest features. The solid lines in Fig. 17 show that the model predictions are in good agreement with the observed aspectratio-dependent etching for PAE-2. It should be noted that for this integration no additional fitting parameter was introduced. Furthermore, this result is independent of the used value for the sticking coefficient. Namely, similar results were obtained by using other combinations of S, ␣, and E 0 that fit the data in Fig. 17 well. V. CONCLUSIONS The etching of PTFE, parylene-N, parylene-AF4, and PAE-2 films has been characterized in an inductively coupled plasma source employing Ar/O2 gas mixtures. It is possible to capture the observed etch rate dependence for parylene-N, parylene-AF4, and PAE-2 on O2 flow, ion flux, ion energy, and feature geometry by a simple surface coverage model. Sidewall bowing and mask undercutting in Ar/O2 discharges can be reduced or prevented by the redeposition of oxygen-deficient and carbon-rich etch products released at the feature bottom. Finally, it was demonstrated that successful pattern transfer is also possible in N2 /O2 discharges. The underlying mechanism for sidewall passivation in these discharges will be addressed in future work. ACKNOWLEDGMENTS This work was financially supported in part by the Semiconductor Research Corporation and Air Products and Chemicals, Inc. Air Products and Chemicals, Inc. is also acknowledged for the supply of PAE-2 wafers. The authors acknowledge Leybold Inficon and Balzers Instruments for supplying them with the mass spectrometers used in this work. M. A. Plano is acknowledged for providing the parylene-AF4 wafers. T. J. Dalton is thanked for helpful dis-
446
Standaert et al.: High-density plasma patterning
cussions. M. Schaepkens, E. A. Joseph, and K. H. J. M. Robben are thanked for their assistance with the data taking. 1
T. E. F. M. Standaert, P. J. Matsuo, S. D. Allen, G. S. Oehrlein, and T. J. Dalton, J. Vac. Sci. Technol. A 17, 741 共1999兲. 2 P. J. Matsuo, T. E. F. M. Standaert, S. D. Allen, G. S. Oehrlein, and T. J. Dalton, J. Vac. Sci. Technol. B 17, 1435 共1999兲. 3 J. E. Heidenreich, J. R. Paraszcak, M. Moisan, and G. Sauve, Microelectron. Eng. 5, 363 共1986兲. 4 O. Joubert, C. Martinet, J. Pelletier, M. Pons, J. M. Francou, J. P. Panabiere, A. Weill, S. Tedesco, F. Vinet, and T. Mourier, Proc. SPIE 1803, 130 共1992兲. 5 M. Pons, J. Pelletier, and O. Joubert, J. Appl. Phys. 75, 4709 共1994兲. 6 A. M. Goethals, F. van Roey, T. Sugihara, L. van den Hove, J. Vertommen, and W. Klippert, J. Vac. Sci. Technol. B 16, 3322 共1998兲. 7 T. E. F. M. Standaert, P. J. Matsuo, S. D. Allen, G. S. Oehrlein, T. J. Dalton, T.-M. Lu, and R. Gutmann, Mater. Res. Soc. Symp. Proc. 511, 265 共1998兲. 8 M. R. Baklanov, S. Vanhaelemeersch, H. Bender, and K. Maex, J. Vac. Sci. Technol. B 17, 372 共1999兲. 9 E. Korczynski, Solid State Technol. 42, 22 共1999兲. 10 W. F. Burgoyne, U. S. Patent Document No. 5,658,994 共1997兲. 11 C. T. Rosenmayer, U. S. Patent Document No. 5,889,104 共1999兲. 12 T. E. F. M. Standaert, M. Schaepkens, X. Li, P. J. Matsuo, and G. S. Oehrlein 共unpublished兲. 13 J. H. Keller, J. C. Forster, and M. S. Barnes, J. Vac. Sci. Technol. A 11, 2487 共1993兲. 14 X. Li, M. Schaepkens, G. S. Oehrlein, R. E. Ellefson, and L. C. Frees, J. Vac. Sci. Technol. 共submitted兲. 15 M. F. Doeming, N. R. Rueger, G. S. Oehrlein, and J. M. Cook, J. Vac. Sci. Technol. B 16, 1998 共1998兲. 16 Y. Wang, R. J. Van Brunt, and J. K. Olthoff, J. Appl. Phys. 83, 703 共1998兲. 17 C. Lee and M. A. Lieberman, J. Vac. Sci. Technol. A 13, 368 共1995兲. 18 R. E. Walkup, K. L. Saenger, and G. S. Selwyn, J. Chem. Phys. 84, 2668 共1986兲. 19 M. van Kampen, Internal Report No. VDF/NT 97-33, Eindhoven University of Technology 共1997兲. 20 M. Schaepkens, N. R. Rueger, J. J. Beulens, X. Li, T. E. F. M. Standaert,
J. Vac. Sci. Technol. A, Vol. 19, No. 2, MarÕApr 2001
446 P. J. Matsuo, and G. S. Oehrlein, J. Vac. Sci. Technol. A 17, 3272 共1999兲. G. Turban and M. Rapeaux, J. Electrochem. Soc. 130, 2231 共1983兲. 22 H. Gokan and S. Esho, J. Electrochem. Soc. 131, 1105 共1984兲. 23 CH. Steinbruchel, B. J. Curtis, H. W. Lehmann, and R. Widmer, IEEE Trans. Plasma Sci. PS-14, 137 共1986兲. 24 J. A. G. Baggerman, R. J. Visser, and E. J. H. Collart, J. Appl. Phys. 75, 758 共1994兲. 25 E. J. H. Collart, J. A. G. Baggerman, and R. J. Visser, J. Appl. Phys. 78, 47 共1995兲. 26 J. M. Cook and B. W. Benson, J. Electrochem. Soc. 130, 2459 共1983兲. 27 G. S. Selwyn, J. Appl. Phys. 60, 2771 共1986兲. 28 I. Tepermeister and H. Sawin, J. Vac. Sci. Technol. A 9, 790 共1991兲. 29 G. Beamson and D. Briggs, High Resolution XPS of Organic Polymers 共Wiley, New York, 1992兲, p. 234. 30 H. Namatsu, Y. Ozaki, and K. Hirata, J. Electrochem. Soc. 130, 523 共1983兲. 31 J. C. Arnold and H. H. Sawin, J. Appl. Phys. 70, 5314 共1991兲. 32 M. Schaepkens and G. S. Oehrlein, Appl. Phys. Lett. 72, 1293 共1998兲. 33 S. Van Nguyen, D. Dobuzinski, S. R. Stiffler, and G. Chrisman, J. Electrochem. Soc. 138, 1112 共1991兲. 34 T. J. Dalton, J. C. Arnold, H. H. Sawin, and D. Corliss, J. Electrochem. Soc. 140, 2395 共1993兲. 35 A. C. Westerheim, A. H. Labun, J. H. Dablash, J. C. Arnold, H. H. Sawin, and V. Yu-Wang, J. Vac. Sci. Technol. A 13, 853 共1995兲. 36 K. H. A. Bogart, F. P. Klemens, M. V. Malyshev, J. I. Colonell, V. M. Donnelly, J. T. C. Lee, and J. M. Lane, J. Vac. Sci. Technol. A 18, 197 共2000兲. 37 W. E. Vanderlinde and A. L. Ruoff, J. Vac. Sci. Technol. B 6, 1621 共1988兲. 38 O. Joubert, J. Pelletier, and Y. Arnal, J. Appl. Phys. 65, 5096 共1989兲. 39 T. A. Blanchet, in Handbook of Thermoplastics, edited by O. Olabisi 共Marcel Dekker, New York, 1997兲, p. 981. 40 P. K. Wu, G.-R. Yang, J. F. McDonald, and T.-M. Lu, J. Electron. Mater. 24, 53 共1995兲. 41 R. A. Gottscho, C. W. Jurgensen, and D. J. Vitkavage, J. Vac. Sci. Technol. B 10, 2133 共1992兲; note that this reference contains a typographical error in Eqs. 共7兲 and 共8兲 which is rectified in Ref. 42. 42 A. D. Bailey III, M. C. M. van de Sanden, J. A. Gregus, and R. A. Gottscho, J. Vac. Sci. Technol. B 13, 92 共1995兲. 21