CHARACTERIZATION OF COPPER ELECTROPLATING AND
ELECTROPOLISHING PROCESSES FOR SEMICONDUCTOR INTERCONNECT
METALLIZATION
by
JULIE MARIE MENDEZ
Submitted in partial fulfillment of the requirements
For the degree of Doctor of Philosophy
Dissertation Advisor: Dr. Uziel Landau
Department of Chemical Engineering
CASE WESTERN RESERVE UNIVERSITY
August, 2009 CASE WESTERN RESERVE UNIVERSITY
SCHOOL OF GRADUATE STUDIES
We hereby approve the thesis/dissertation of
______
candidate for the ______degree *.
(signed)______(chair of the committee)
______
______
______
______
______
(date) ______
*We also certify that written approval has been obtained for any proprietary material contained therein. TABLE OF CONTENTS
Page Number List of Tables 3 List of Figures 4 Acknowledgements 9 List of Symbols 10 Abstract 13
1. Introduction 15 1.1 Semiconductor Interconnect Metallization – Process Description 15 1.2 Mechanistic Aspects of Bottom-up Fill 20 1.3 Electropolishing 22 1.4 Topics Addressed in the Dissertation 24
2. Experimental Studies of Copper Electropolishing 26 2.1 Experimental Procedure 29 2.2 Polarization Studies 30 2.3 Current Steps 34 2.3.1 Current Stepped to a Level below Limiting Current 34 2.3.2 Current Stepped to the Limiting Current Plateau 36 2.3.3 Effect of Current Density 38 2.3.4 Effect of Rotation Speed 40 2.4 Highly Resistive Surface Film 42 2.5 Electrochemical Impedance Spectroscopy 44 2.6 Stability of the Film in Presence of Chloride 49 2.7 Two-Compartment Cell Experiments 51 2.8 Conclusions 52
3. A Mechanistic Model for Copper Electropolishing 53 3.1 Regime I – Buildup of Surface Copper Ion Concentration 53 3.2 Regime II – Controlling Transport through a Surface Layer 57 3.3 Model Verification 60 3.3.1 Time Delay Prior to the Onset of the Sharp Potential Increase 67 3.3.2 Effect of Water Concentration on the Limiting Current 67 3.4 Conclusions 72
4. Novel Polyether Suppressors Enabling Copper Metallization of High Aspect Ratio Interconnects 73 4.1 Experimental Procedure 75 4.2 Results and Discussion 76 4.2.1 Polarization Data 79 4.2.2 Modeled Via-fill Ratio 89 4.2.3 Interaction with the Anti-suppressor 96 4.3 Conclusions 98
1 5. Mechanistic Studies of Polyether Adsorption 99 5.1 Experimental Details 101 5.1.1 Attenuated Total Reflectance Fourier Transform Infrared Spectroscopy (ATR-FTIR) 101 5.1.2 Quartz Crystal Microbalance (QCM) 103 5.2 ATR-FTIR Studies of PEG Adsorption 106 5.3 Quartz Crystal Microbalance 118 5.4 Effect of Cu+ on Copper Deposition 125 5.5 Conclusions 127
6. Major Conclusions and Recommendations for Future Work 128 6.1 Major Conclusions 128 6.1.1 Electropolishing 128 6.1.2 Novel Polyethers Extending Gap-fill Capabilities 129 6.2 Recommendations for Further Studies 131
Bibliography 135
2 LIST OF TABLES
Page Number Table 2.1. Ohmic and polarization resistances measured by electrochemical 48 impedance spectroscopy.
Table 3.1. Copper electropolishing model parameters. Justification for the 61 estimated values is given in the text.
Table 4.1. Polyethers explored in the polarization studies. 77
Table 4.2. Kinetics parameters fitted to polarization data in Figure 4.1 and 90 Figure 4.2.
3 LIST OF FIGURES
Page Number Figure 1.1. Schematic detailing the dual Damascene process. Both (a) the 17 via and (b) the trench are etched into the insulator. (c) A diffusion barrier layer and copper seed layer are both deposited by physical vapor deposition. (d) Copper is electrodeposited to fill the via and trench. (e) The overburden copper is removed by chemical mechanical planarization (CMP).
Figure 1.2. Schematic of the transport and diffusion processes inside the 19 via. (a) PEG adsorbs quickly but is diffusion limited; therefore, it adsorbs primarily at the top of the via sidewalls. SPS diffuses quickly but adsorbs more slowly than PEG and adsorbs primarily at the via bottom. (b) As the fill progresses, the bottom surface contracts. The SPS becomes more concentrated at the bottom, bringing about rapid growth at the via bottom.
Figure 2.1. Schematic of concentration profiles in the two major proposed 28 mechanisms for copper electropolishing. In mechanism (a), an acceptor species (indicated as water) diffuses towards the anode, complexes there with the discharged cupric ions, and diffuses back (as a complex) toward the bulk solution. According to mechanism (b), the concentration of cupric ions increases at the anode until the solubility limit is reached, at which point, a solid film is formed on the anode. Cupric ions then diffuse towards the bulk across a mass transport boundary layer of thickness δ.
Figure 2.2. Polarization curves for electropolishing of a copper disk 32 electrode at various rotation speeds (A – 50 rpm, B – 100 rpm, C – 200 rpm, D – 400 rpm, E – 600 rpm) in 85 wt% phosphoric acid. The potential was scanned at 10 mV/s.
Figure 2.3. Limiting current density for phosphoric acid solutions of various 33 water concentrations at 800 rpm. The linear relationship between limiting current density and bulk water concentration has led numerous investigators to associate it with the transport of an acceptor species (water) toward the anode.
Figure 2.4. Potential response to a current pulse at 100 rpm (a) below the 35 limiting current (6.3 mA/cm2), and (b) at the limiting current (19.6 mA/cm2). The current is stepped up to the specified value at 100 s, held at that value for 400 s, and then stepped down to zero at 500 s. Note that the potential scales in (a) and (b) are quite different.
4
Figure 2.5. Potential transient responses to current steps from zero to three 39 values on the limiting current plateau. The current was stepped to (A) 18.8 mA/cm2; (B) 17.8 mA/cm2; (C) 17.0 mA/cm2 at 100 s. The disk was rotated at 100 rpm.
Figure 2.6. Potential responses to currents steps from zero to 19.6 mA/cm2 41 (at the limiting current) for various rotation speeds: A – 90 rpm, B – 100 rpm, C – 110 rpm. The time delay prior to the sharp potential increase shows a strong dependence on the rotation speed.
Figure 2.7. Nyquist plots at various applied potentials below the limiting 46 current at 400 rpm. The ohmic resistance remains approximately constant, but the polarization resistance decreases as the potential is increased.
Figure 2.8. Nyquist plots at an applied potential of 1.3 V vs. copper (at the 47 limiting current plateau) and various rotation speeds. These measurements indicate an ohmic resistance of approximately 2.8 Ω-cm2 and a polarization resistance between 1.6 and 3.4 Ω-cm2, which decreases with increasing rotation speed.
Figure 2.9. Potential response to current step of 14.2 mA/cm2 (near the 50 limiting current) in 85 wt% H3PO4 solution containing 100 ppm HCl. The potential oscillations suggest the formation and breakdown of a film.
Figure 3.1 Schematics representing (a) Regime I and (b) Regime II of the 54 proposed model. In Regime I, the concentration at the anode increases until Csat is reached. In Regime II, a flux imbalance leads to the buildup in thickness (x) of a surface film.
Figure 3.2. Comparison of measured and modeled overpotential response to 63 a current pulse in Regime I. The measured data are from Figure 2.4a, while the predicted response is based on the numerical solution of Eqs. [3.1] and [3.5]. The two curves are in reasonable agreement. The small deviation can probably be attributed to the spatial distribution of the copper ions.
Figure 3.3. Comparison of the model governing Regime II (Eq. [3.12]) to 65 the experimental potential response displayed in Figure 2.4b (region C-D). Note the highly expanded time scale.
Figure 3.4. Sensitivity analysis for parameters A and M (Eq. [3.12]). The 66 lines indicate values for these parameters such that the model correlates the data from Figure 2.4b (region C-D) within the indicated percentages.
5
Figure 3.5. The model (Eq. [3.22]) predicts a nearly linear relationship 71 between the limiting current density and the bulk water concentration for copper electrodissolution in phosphoric acid. Also indicated is a linear approximation for Eq. [3.22].
Figure 4.1. Polarization data for solutions containing 0.5 M CuSO4 (pH~2), 80 70 ppm Cl-, and 100 ppm of the specified polyether, all with molecular mass of approximately 1000 g/mol. The points for polyoxyethylene lauryl ether (diamonds) and polyoxyethylene cetyl ether (circles) fall nearly on top of one another.
Figure 4.2. Polarization data for solutions containing 0.5 M CuSO4 (pH~2), 82 70 ppm Cl-, and 100 ppm of the specified polyether with molecular mass of approximately 600 g/mol.
Figure 4.3. Polarization data for additional solutions containing 0.5 M 83 - CuSO4 (pH~2), 70 ppm Cl , and 100 ppm of the specified polyether with molecular mass of approximately 600 g/mol.
Figure 4.4. Polarization data for solutions containing 0.5 M CuSO4 (pH~2), 85 70 ppm Cl-, and 100 ppm of the specified polyether with molecular mass of approximately 300 g/mol.
Figure 4.5. Polarization data for solutions containing 0.5 M CuSO4 (pH~2), 86 70 ppm Cl-, and 100 ppm of the specified polyether with molecular mass in the range of approximately 2000 g/mol to 4000 g/mol.
Figure 4.6. Overpotentials at 5 mA/cm2 as a function of the number of ether 88 oxygen atoms in the polyether. The circles correspond to PEG, and the squares are all other polyethers studied (listed in Table 4.1). Although some trend is indicated between increased overpotential and the number of ethereal oxygen atoms, the data spread implies that other factors, including the chemical structure, are important.
Figure 4.7. Bottom-up fill ratio (iB/iSW) as a function of total current 93 simulated for a 300 mm wafer with 15% feature loading (pattern density) for solutions containing one of the following suppressors: PEG 1000, polyoxyethylene lauryl ether, or polyoxyethylated β-naphthol. Significant improvement (3 ~ 4x) is expected by replacing PEG with either of the above listed other polyethers.
6
Figure 4.8. SEM cross-sections of vias with aspect ratio close to 10 95 electroplated with copper after PVD seed deposition.48 Fill quality is compared for copper plating in the presence of two different suppressors: (a) PEG 1000, exhibiting center-line voids due to inferior bottom-up fill rate (Figure 4.7), and (b) polyoxyethylated (POE) β-naphthol, indicating void- free fill on account of its improved suppression and higher bottom-up fill rate.
Figure 4.9. Voltage transient response at 5 mA/cm2 to SPS injections into 97 polyether polarized deposition. The solution initially contained 0.5 M - CuSO4 (pH~2), 70 ppm Cl , and 100 ppm of the specified polyether. After steady-state was reached, 10 ppm SPS was injected into the solution. Polyoxyethylated β-naphthol has a similar voltage response to that of PEG.
Figure 5.1. Schematic of ATR-FTIR system (not to scale). The laser beam 102 is introduced from the back, through a thin Si wafer coated with Cu. Only a thin region (~1 μm) at the electrode/electrolyte interface is sampled.
Figure 5.2. Schematic of QCM cell (not to scale). The top electrode is 105 exposed to the solution, while the bottom electrode is in contact with air, serving as the reference.
Figure 5.3. Spectrum for solid PEG relative to ZnSe-air. The most 108 prominent peaks are the C-O-C stretch at 1109 cm-1 and the C-H stretch at 2885 cm-1. Since it is believed that the ethereal oxygen is important in the adsorption of polyethers, the C-O-C stretch peak is of interest.
Figure 5.4. Spectrum for solid cupric sulfate pentahydrate relative to ZnSe- 109 air. The most prominent peaks are at 670, 3630, and 3730 cm-1. There is also a peak at 1090 cm-1, which is near the C-O-C stretch peak of PEG at 1100 cm-1.
- Figure 5.5. Spectrum for 0.5 M CuSO4 and 70 ppm Cl on a Cu substrate, 110 relative to water. A prominent sulfate peak is evident at 1100 cm-1, near a peak of interest for PEG. The peaks at 3280 and 1640 cm-1 are due to error in referencing to the spectrum for water on Cu.
- Figure 5.6. FTIR spectra for solutions containing 0.5 M CuSO4, 70 ppm Cl , 112 and 100 ppm PEG 1000 on Cu (bottom) and on Si (top) substrates. The spectra indicate similar behavior on both Si and Cu. The spectra were shifted vertically for clarity.
7
Figure 5.7. Spectra for various solutions on Cu. The solutions contain the 114 specified components in the following concentrations: 0.5 M CuSO4, 70 - ppm Cl , 100 ppm PEG 1000, and 0.67 M Na2SO4. No peak is observed with a solution of 100 ppm PEG and 70 ppm Cl- in water.
Figure 5.8. FTIR spectrum for a solution containing 0.5 M CuCl2 and 100 115 ppm PEG 1000, relative to 0.5 M CuCl2. The C-O-C stretch peak of PEG at 1100 cm-1 can be detected in the absence of sulfate.
Figure 5.9. FTIR spectrum for a solution containing 0.5 M CuSO4, 70 ppm 117 - - Cl , and 100 ppm PEG 1000, relative to 0.5 M CuSO4 and 70 ppm Cl . Different solutions were introduced using a flow cell. A peak associated with the C-O-C stretch of PEG is evident at 1100 cm-1.
Figure 5.10. Frequency response upon addition of small volumes of 0.5 M 119 CuSO4 to the cell initially containing 0.5 M CuSO4 solution. The arrows indicate times when additional solution was introduced into the cell. The frequency signal stabilizes to nearly the original value, indicating that additional solution does not affect adsorption on the substrate.
Figure 5.11. Frequency response upon addition of Cl- and PEG to a 0.5 M 121 - CuSO4 solution. When 70 ppm Cl was added to the cell, a change in frequency difference of ~30 Hz, corresponding to ~130 ng/cm2, was observed. When 100 ppm PEG (molecular mass 1000 g/mol) was added to the Cl--containing solution, a similar change in frequency difference of ~30 Hz (~130 ng/cm2) was observed.
Figure 5.12. Frequency response upon addition of PEG and Cl- to a 0.5 M 123 CuSO4 solution. When 100 ppm PEG (molecular mass 1000 g/mol) was added, a small decrease in the frequency difference was observed. When 70 ppm Cl- was added to the PEG-containing solution, an increase in frequency difference of ~50 Hz (~220 ng/cm2) was observed.
Figure 5.13. Overpotential response to injection of 10 ppm Cu+ into 126 acidified (pH~2) 0.5 M CuSO4 solutions containing the indicated additives. The cathode was polarized at 10 mA/cm2. When Cu+ is added to a solution containing no additives, the overpotential decreases by ~30 mV. No change in overpotential is observed when Cu+ is added to solutions containing PEG or PEG and Cl-.
8 Acknowledgements
I would like to thank Professor Uziel Landau for his guidance in the completion of this dissertation. I would also like to thank the other members of my committee,
Professor Heidi Martin, Professor C. C. Liu, and Professor Frank Ernst for helpful comments and suggestions regarding this work.
Funding from the Intel Research Council supported this work. I would like to thank Rohan Akolkar at Intel for his collaboration in this work and for the opportunity to
spend a summer working at Intel. I also thank Tatyana Andryushchenko at Intel for
contributions to the electropolishing work.
I would like to thank Jim Adolf for helpful suggestions regarding the polyether
work and the students in the Landau Group for their helpful comments and support.
Professor Heidi Martin assisted with the FTIR measurements. Professor Kathleen Kash
provided assistance with electron beam deposition. Professor Jim Burgess and Professor
Robert Savinell provided use of QCM equipment. Craig Virnelson and Dan Shelberg
assisted with data acquisition for the QCM. Cliff Hayman and the students in the
Diamond Lab have helped with various things in the lab. Professor E. Gileadi and
Professor J. Newman are acknowledged for helpful discussions regarding
electropolishing.
I appreciate the support and encouragement from my parents, sister, and extended family. I am grateful to the friends I have made here in the last four years.
To J.D., my husband, thank you for your understanding and for always being there for me.
9 List of Symbols
A film resistance parameter (Eq. [3.11])
AB area at bottom of features
Asupp area at feature sidewalls and wafer top surface
B film resistance parameter (Eq. [3.11])
C Cu2+ concentration
2+ Cb bulk Cu concentration
2+ Ce Cu concentration near the electrode
CH2O concentration of water
CH3PO4 concentration of phosphoric acid
Cm constant dependant on properties of quartz crystal
Csat saturation concentration of copper ions
D Cu2+ diffusion coefficient f frequency measured at quartz crystal surface f0 resonant frequency of the fundamental mode of the crystal
F Faraday’s constant i current density iB current density at via bottom iL limiting current density i0 exchange current density i0,AS exchange current density in presence of anti-suppressor i0,S exchange current density in presence of suppressor iSW current density at via sidewalls
10 I current
Kd first dissociation constant of H3PO4
Ksp solubility product of copper salt
L via length
m mass change at quartz crystal surface
M film molecular mass
MH2O molecular mass of water
MH3PO4 molecular mass of phosphoric acid
n number of electrons transferred
nh number of the harmonic at which the crystal is being driven
N1 rate of generation of copper ions
N2 rate at which the film dissolves
r via radius
R ideal gas constant
t time
T temperature
V voltage drop across the film
x film thickness
yH2O weight fraction of water
yH3PO4 weight fraction of phosphoric acid
Greek
α transfer coefficient
11 αC,AS cathodic transfer coefficient in presence of anti-suppressor
αC,S cathodic transfer coefficient in presence of suppressor
δ boundary layer thickness
Δf change in frequency
Δm change in mass per area
η overpotential
μq shear modulus of quartz
ρ film density
ρq density of quartz
12
Characterization of Copper Electroplating and Electropolishing Processes for Semiconductor Interconnect Metallization
Abstract
by
JULIE MARIE MENDEZ
Research focusing on two topics central to metallization of semiconductor
interconnects by copper electroplating is reported. A systematic experimental study of
copper electropolishing in phosphoric acid, focusing on current stepping, yielded a model for the process. The model invokes two regimes. The first regime involves the build-up
of the copper ion concentration at the dissolving anode. The second regime is
characterized by the formation of an insoluble film on the anode when the concentration
of the copper phosphate species reaches the solubility limit. Subsequent transport of
dissolving copper ions through the film yields the observed high potential drop. This
model correlates well with experimental data and explains the various heretofore
unexplained phenomena observed in copper electropolishing.
The second focus of the research provides significant improvement to bottom-up
fill of copper features by identifying a new class of polyether additives (including
polyoxyethylated β-naphthol and polyoxyethylene lauryl ether) which provide stronger
inhibition than the currently used poly(ethylene glycol) (PEG). These novel polyether
inhibitors extend the present process capabilities to features smaller than the present limit
of about 60 nm. Attenuated total reflectance Fourier transform infrared spectroscopy
13 (ATR-FTIR) and quartz crystal microbalance (QCM) studies provide insight into the polyether adsorption mechanism.
14 CHAPTER 1
Introduction
1.1 Semiconductor Interconnect Metallization – Process Description
In an integrated circuit, metalized vias and trenches fabricated within the silicon
serve as the electrical connections between the semiconductor junctions and devices. As
the features become smaller, their resistance increases, leading to a larger RC time
constant. To maintain the desired fast processing speed, the resistance must be
decreased. This can be achieved by utilizing a metal with a lower resistivity. In the last
decade, the semiconductor industry has moved from aluminum interconnects
( = 2.65 μcm) to copper ( = 1.68 μcm), gaining more than a 40% advantage.1 In addition to the lower resistivity, copper has a higher resistance to electromigration as compared to aluminum and offers an opportunity for more efficient fabrication through the dual Damascene process, which enables the simultaneous metallization of the vias and the trenches, resulting in fewer process steps.2
The dual Damascene process steps are shown schematically in Figure 1.1. First a
via (Figure 1.1a) and then a trench (Figure 1.1b) are etched into the insulator, typically
SiO2. A barrier layer, typically consisting of TaN or TiN, which prevents the migration
of copper into the insulator, is deposited by physical vapor deposition (PVD). A copper
seed layer, providing conductivity, is then deposited on top of the barrier, also by PVD
(Figure 1.1c). The sub-micron features are then filled with copper by electrodeposition
(Figure 1.1d). In order to assure the fill of very wide features, typically capacitors and
metal pads, excess (‘overburden’) copper is electroplated. This excess copper is then
removed by chemical mechanical planarization (‘CMP’, which may be replaced in the
15 future by the electropolishing process studied herein) and followed by growth of a subsequent SiO2 layer (Figure 1.1e). The entire process is repeated for the fabrication of as many layers as required (typically 5-7 times for Pentium class processors).1
16
Figure 1.1. Schematic detailing the dual Damascene process. Both (a) the via and (b) the trench are etched into the insulator. (c) A diffusion barrier layer and copper seed layer are both deposited by physical vapor deposition. (d) Copper is electrodeposited to fill the via and trench. (e) The overburden copper is removed by chemical mechanical planarization (CMP).
17 The success of the electroplating process for copper metallization of semiconductor interconnects requires the vias to be filled without voids.2 This is achieved through “bottom-up” plating, controlled by an additive mixture consisting typically of ppm quantities of chloride ions, a polyether, [e.g., polyethylene glycol
(PEG)], and an organic sulfur compound, [e.g., bis(3-sulfopropyl) disulfide (SPS)].
Chloride ions (~70 ppm) are required for the polyether to effectively function as a suppressor in the via fill process. Often, a nitrogen compound is also present; however, this does not play a major role in gap-fill and will not be discussed in this work. The effectiveness of these additives in the via-fill process has been determined empirically.
While a detailed molecular level understanding of their roles in via-fill and their interactions with one another are lacking, the macroscopic mechanistic aspects have been determined.3
The functions of the polyether and the sulfur compound can be described by the following transport processes. The polyether, which suppresses copper deposition, is a large transport-limited molecule, and therefore it primarily adsorbs and inhibits deposition on the via rim, on the top portion of the sidewalls, and on the wafer top surface,3 as shown in Figure 1.2a. The faster diffusing sulfur compound, which is a deposition enhancer (‘anti-suppressor’), preferentially adsorbs on the via bottom and accelerates the plating at this location. As shown in Figure 1.2b, the decreasing surface area at the feature bottom due to deposition further concentrates the anti-suppressor, bringing about even higher bottom-up fill rates.4 Due to competitive adsorption, the anti- suppressor displaces the suppressor as time progresses, leading eventually to enhanced deposition at the wafer surface and the feature sidewalls.3
18
Figure 1.2. Schematic of the transport and diffusion processes inside the via. (a) PEG
adsorbs quickly but is diffusion limited; therefore, it adsorbs primarily at the top of the via sidewalls. SPS diffuses quickly but adsorbs more slowly than PEG and adsorbs
primarily at the via bottom. (b) As the fill progresses, the bottom surface contracts. The
SPS becomes more concentrated at the bottom, bringing about rapid growth at the via
bottom.
19 1.2 Mechanistic Aspects of Bottom-up Fill
Although the suppression effect of PEG is well characterized,5-23 the underlying
molecular mechanism is not. Two major classes of PEG suppression mechanisms are
reported in the literature. One mechanism proposes that chloride ions adsorbed on the
surface coordinate with cuprous ions and the oxygen atoms of the PEG, inhibiting the
reduction of Cu+ ions. Yokoi et al. suggested that Cu+ ions are bound to the PEG; when
chloride ions are present, they also bind to Cu+ ions, preventing them from being released
from PEG.5 This mechanism has served as a basis for further mechanistic studies by other investigators.12, 15, 16 Healy et al. suggest that PEG adsorbs in two forms: a copper chloride complex that forms near the open circuit potential and a neutral PEG layer that
adsorbs at more cathodic potentials.7 Long et al. characterized the transition from suppression to nonsuppression by a model for the suppressor complex that is dependent on the concentrations of Cu2+, Cl-, H+, and suppressor.19
A second mechanism found in the literature is that adsorbed PEG acts as a
physical diffusion barrier to cupric ions, slowing the rate of copper deposition.15 Quartz
crystal microbalance (QCM) data suggest that PEG adsorbs in a monolayer as spheres in the presence of Cl-.8 A related paper suggests a model by which PEG competes with
Cu2+ for surface sites when Cl- is also present.9 Doblhofer et al. suggested that an
inhibitive surface film is formed by PEG and Cl-, and neither Cu2+ nor Cu+ are required to
form this film.13 Jin et al. showed by AFM that PEG is adsorbed on copper surfaces in acid solutions in the absence of Cu2+ if in the presence of Cl-.14 Walker et al. reported
that an adsorbed PEG layer also forms on Ag and Au, and this layer does not block
20 electron transfer.17 Garrido and Pritzker reported that PEG can adsorb in the absence of
Cu2+ if Cl- is present.22
Other aspects of the role of PEG in via-fill have been reported. Dow et al. reported that the best filling performance occurs when PEG is used at molecular masses between 6000 and 8000 g/mol.18 Willey et al. studied time constants for PEG adsorption and desorption.20 Gallaway and West reported polarization studies using poly(propylene glycol) (PPG) and copolymers of PEG and PPG as alternative suppressors.23
The inhibition provided by the presently used PEG does not provide sufficient inhibition of the plating on the sidewalls to prevent their premature closure (before the bottom reaches the rim) in sub-60 nm vias, thus imposing a serious limitation on the current technology. In order to achieve void-free gap-fill, the average deposition rate on the feature bottom must exceed the rate at which the sidewalls merge. It has been shown that the via sidewalls have similar additives coverage to the via top (rim) and propagate at about the same rate, which is substantially lower than that of the via bottom.3
21 1.3 Electropolishing
As stated above, to ensure complete filling of interconnects of various sizes,
excess copper must be plated onto the surface. This overburden must be removed prior to
the next processing step. The currently utilized method for the removal of the copper overburden is CMP, whereby the copper is chemically oxidized and mechanically
abraded. A possible replacement or complement to CMP is electropolishing, the anodic
dissolution of copper under conditions which induce the generation of a smooth surface
texture. This process, which has long been the subject of scientific and technological
interest,24, 25 has recently been attracting attention for this application.26-28 Being a non- contact process, electropolishing is compatible with fragile low-k dielectric materials which may be damaged by abrasion. Electropolishing is typically attained by high-rate anodic dissolution of copper in concentrated (~85 wt%) phosphoric acid under limiting current conditions29 at moderate agitation. Despite its technological significance,
fundamental understanding of the mechanism governing the electropolishing process is
still lacking.
Two different mechanisms for the electropolishing process have been proposed.
The first implies a diffusion limited transport of an acceptor species to the dissolving
anode.27, 30 The acceptor species complexes with the copper ions at the anode, enabling their transport towards the bulk. Most investigators suggest that the acceptor species is water.31-33 According to a second proposed mechanism, upon dissolution, the copper ion
concentration increases at the anode until the solubility limit of a copper salt, e.g.,
2+ - 25 (Cu )x(H2PO4 )y, is reached, at which point, a film precipitates on the anode. In
addition to the uncertainty concerning the transport-limited species, debate persists over
22 the presence of a film on the anode surface. Some investigators claim that a film does not
exist,32 while others have reported evidence for the presence of a solid oxide film34 or a
resistive boundary layer.27 The nature of this resistive boundary layer has not been clearly specified.
23 1.4 Topics Addressed in the Dissertation
Several aspects regarding copper electropolishing and the void-free fill of interconnect vias remain undetermined and are specifically addressed in this dissertation:
(i). Requirement of more effective additives. PEG does not provide sufficient
inhibition of the plating on the sidewalls to prevent their premature closure in sub-
60 nm vias. This imposes a technological limit. More effective suppressors are
essential for the void-free fabrication of sub-60 nm vias.
(ii). Adsorption mechanism of PEG. A detailed understanding of the adsorption
mechanism of PEG and Cl-, their interactions, and how adsorption relates to
suppression is lacking. Additional mechanistic understanding would be beneficial
to the identification and development of more effective additives.
(iii). Role of phosphoric acid in electropolishing. Nearly all copper electropolishing
processes are operated in concentrated phosphoric acid. This implies that the
chemical nature of phosphoric acid is important, whereby electropolishing occurs
through a phosphate intermediate species or through a surface film. The exact
nature of the role of phosphoric acid has not been determined, and a film has not
been identified.
(iv). Characteristics of the electropolishing mass transport limit. The limiting current
plateau observed in the polarization curves indicates that a species is at its mass
transport limit. Electropolishing is carried out at conditions that correspond to
this plateau, suggesting that the transport-limited species is of importance to the
process. As described above, there is some debate over which species is
transport-limited: cupric ions or an acceptor species such as water.
24 The present dissertation addresses the above issues. The primary objectives for this work
are:
(i). To identify the mass-transport limited species and develop a mechanistic model
for copper electropolishing.
(ii). To identify and characterize suppressor molecules which would provide improved
suppression over the conventionally used PEG.
Chapter 2 of this dissertation presents experimental studies of the copper electropolishing process.
Chapter 3 presents a mechanistic model consistent with experimental observations detailed in Chapter 2. The model indicates a two-step process, whereby a highly resistive surface film is formed at the anode.
Chapter 4 describes the identification and characterization of numerous polyether molecules which have been studied as alternatives to PEG as suppressors in bottom-up fill of interconnect features.
Chapter 5 presents attenuated total reflectance Fourier transform infrared spectroscopy (ATR-FTIR) and quartz crystal microbalance (QCM) studies which provide insight into the suppression mechanism of PEG.
Chapter 6 provides the major conclusions of this work and recommendations for future work.
25 CHAPTER 2
Experimental Studies of Copper Electropolishing
The anodic dissolution of copper under conditions which induce the generation of
a smooth surface texture is referred to as electropolishing. This process, which has long
been the subject of scientific and technological interest,24, 25 has recently been attracting
attention as a possible replacement or complement to chemical mechanical polishing
(CMP) for the removal of excess electroplated copper in semiconductor interconnect
metallization.26-28 Being a non-contact process, electropolishing is compatible with
fragile low-k dielectric materials which may be damaged by abrasion. Electropolishing is
typically attained by high-rate anodic dissolution of copper in concentrated (~85 wt%)
phosphoric acid under limiting current conditions29 at moderate agitation. Despite its technological significance, fundamental understanding of the mechanism governing the electropolishing process is still lacking.
Two mechanisms proposed for the electropolishing process are depicted schematically in Figure 2.1. The first mechanism (Figure 2.1a) implies the diffusion- limited transport of an acceptor species to the anode.27, 30 The acceptor species
complexes with the copper ions at the anode, enabling their transport towards the bulk.
Most investigators suggest that the acceptor species is water.31-33 According to the
second proposed mechanism (Figure 2.1b), upon dissolution, the concentration of copper
ions increases at the anode until the solubility limit of a copper salt, e.g.,
2+ - 25 (Cu )x(H2PO4 )y, is reached, at which point a film precipitates on the anode.
26 The aim of this work is to shed additional light on the mechanistic aspects of copper electropolishing. The focus is the analysis of experiments involving the application of anodic current steps to a rotating copper disk electrode.
27
Figure 2.1. Schematic of concentration profiles in the two major proposed mechanisms for copper electropolishing. In mechanism (a), an acceptor species (indicated as water) diffuses towards the anode, complexes there with the discharged cupric ions, and diffuses back (as a complex) toward the bulk solution. According to mechanism (b), the concentration of cupric ions increases at the anode until the solubility limit is reached, at which point, a solid film is formed on the anode. Cupric ions then diffuse towards the bulk across a mass transport boundary layer of thickness δ.
28 2.1 Experimental Procedure
In order to quantify the transport rates, the working electrode consisted of a copper disk, 6 mm in diameter, rotated in phosphoric acid (Fisher, 85 wt% [14.7 mol/L], unless otherwise indicated) using a Pine Instruments rotator (Model A-SR2). The copper disk electrode was mechanically polished with silicon carbide paper (FEPA #1200 followed by #4000) prior to each experiment. The rotating disk electrode was polarized anodically using a Solartron 1280B potentiostat. The cathode was a copper disk (5.5 cm diameter) placed at the bottom of a 250 mL beaker. A copper electrode, placed far from the disk, in a region where the potential was essentially constant, served as the reference electrode. Potentials are reported versus Cu/Cu2+. No ohmic drop compensation was applied to the reported experimental data; however, the ohmic drop correction was introduced when comparing the measured overpotentials with those predicted by the model. The estimated error in the measurements (current density in polarization data and potential in current step data) is approximately 5% of the reported values.
29 2.2 Polarization Studies
The limiting current behavior of the copper electropolishing system has been
previously characterized.31, 32 In the current study, data were obtained which are in good agreement with that previously reported. Typical polarization curves, obtained by scanning the potential (at 10 mV/sec) at various rotation speeds, are shown in Figure 2.2.
As noted, initially, the current increases with the potential, forming a peak at about 0.3 V.
The magnitude of the peak decreases, and the potential at which it is observed increases with the rotation speed. The nature of this peak is not well characterized. It is uncertain whether the peak is caused by transport limitations32 or is due to the formation of a film.31
At potentials higher than about 0.5 V, the current forms a plateau, remaining
approximately constant over the range ~0.5 V to 1.6 V. Above 1.6 V, the current
increases again due to oxygen evolution. In this range, oxygen bubbles are clearly
observed.
The limiting current densities shown in Figure 2.2 are linearly dependent on the
square root of the rotation speed. This relationship, suggesting a mass transport limited
process, has been previously noted by other investigators.32 The limiting current density
is proportional to the bulk water concentration, as shown in Figure 2.3 and as reported by
Du and Suni.33 The data in Figures 2.2 and 2.3 indicate Levich behavior, described by
the following equation, indicating that the process is limited by mass transfer.35
211 362 inFDCL 0.62 b [2.1]
The current plateau characteristics have led numerous investigators to associate it with
the transport of an acceptor species toward the anode.32, 33 Other investigators31 link the plateau to the formation of a film on the anode. Since steady-state polarization data do
30 not provide sufficient insight into the process, the transient technique of current stepping was applied in order to probe the process.
31
Figure 2.2. Polarization curves for electropolishing of a copper disk electrode at various rotation speeds (A – 50 rpm, B – 100 rpm, C – 200 rpm, D – 400 rpm, E – 600 rpm) in
85 wt% phosphoric acid. The potential was scanned at 10 mV/s.
32
Figure 2.3. Limiting current density for phosphoric acid solutions of various water concentrations at 800 rpm. The linear relationship between limiting current density and bulk water concentration has led numerous investigators to associate it with the transport of an acceptor species (water) toward the anode.
33 2.3 Current Steps
A series of current steps were applied to a copper disk electrode rotating at
100 rpm, a typical rotation speed for operation, in concentrated (85%) phosphoric acid.
Initial tests involved stepping the current to a level below the limiting current. These tests were then compared with experiments in which the current was stepped to the limiting current, i.e., the condition under which electropolishing is commonly performed.
2.3.1 Current Stepped to a Level below Limiting Current
Figure 2.4a shows the potential response in a typical experiment in which the current was initially held at zero for 100 s and then stepped up to approximately 30% of the limiting current (6.3 mA/cm2). After 400 s, the current was stepped back down to
zero. The potential transient (Figure 2.4a) exhibits a response similar to that commonly
observed in electrochemical systems below the limiting current.36 Upon stepping up the
current, there is a sharp immediate increase in potential (~70 mV) attributed to ohmic
overpotential followed by a potential rise due to the activation overpotential. This is then
followed by a gradual potential increase characteristic of a mass transport transient. The
theoretical time scale for the mass transport transient, estimated by δ2/D (where δ is the boundary layer thickness and D is the diffusion coefficient), is about 50 s. This approximately corresponds to the time scale observed in Figure 2.4a over which most of the gradual potential change occurs. Following this time scale, the potential remains nearly constant.
34
Figure 2.4. Potential response to a current pulse at 100 rpm (a) below the limiting current (6.3 mA/cm2), and (b) at the limiting current (19.6 mA/cm2). The current is
stepped up to the specified value at 100 s, held at that value for 400 s, and then stepped
down to zero at 500 s. Note that the potential scales in (a) and (b) are quite different.
35 2.3.2 Current Stepped to the Limiting Current Plateau
A significantly different potential response was observed when the current was
stepped up to the limiting current plateau and held constant at that value. As a typical
example, data are presented for a current step similar to that shown earlier; however, the
current density was now stepped to the limiting current plateau (19.6 mA/cm2). The potential response shown in Figure 2.4b exhibits an initial sharp increase (point A to B), followed by a gradual increase of ~50 mV over a period of ~70 s (point B to C). At point
C, the potential very rapidly (in about 5 s) increases to a large value, approximately 1.6 V
(indicated by point D). When the current is stepped down to zero (point E), the corresponding potential decrease (E to F) is almost instantaneous (less than 1 s).
The potential response exhibited when the current is stepped to a value at the
limiting current plateau (Figure 2.4b) indicates a two-step process. The first step (A to C
in Figure 2.4b) consists of the initial sharp potential increase followed by a slow gradual
increase over ~70 s, similar to a common potential response to a current step. The second
step involves a large, rapid potential increase to approximately 1.6 V (points C to D).
This apparent two-step process is uncharacteristic of transport-limited behavior, and
cannot be explained as such. Stepping the current to values below the limiting current, as
shown in Figure 2.4a, did not produce this unique, two-step response. This suggests that
some additional process, leading to a significant potential increase, is initiated when the
current is held at the limiting current plateau.
It should be noted that the potential decreases almost instantaneously when the current is stepped down to zero (point E in Figure 2.4b). The rapid decrease of the potential can not be associated with the concentration relaxation within the mass transport
36 boundary layer, since the latter would be characterized by a much longer relaxation time- scale (t~50s), similar to that required for the buildup of the concentration profile. The very short time (~1 s) required for the potential to drop to nearly zero suggests the presence of a large surface resistance (offered possibly by a surface film) which disappears almost instantaneously upon switching off the current.
37 2.3.3 Effect of Current Density
Since the limiting current plateaus, as shown in Figure 2.2, exhibited a slight variation in the current densities along the plateau, current steps were performed to three different values on the limiting current plateau. The potential responses are shown in
Figure 2.5. All indicate a similar response; however, the time delay before the large
surge in potential is shorter when the current density is higher. This potential response is
similar to data reported earlier by Padhi et al.,27 who suggested that the large increase in
potential is caused by the formation of a stable resistive film.
38
Figure 2.5. Potential transient responses to current steps from zero to three values on the limiting current plateau. The current was stepped to (A) 18.8 mA/cm2; (B) 17.8 mA/cm2;
(C) 17.0 mA/cm2 at 100 s. The disk was rotated at 100 rpm.
39 2.3.4 Effect of Rotation Speed
The potential responses to current steps to the limiting current at different rotation speeds are shown in Figure 2.6. The rotation speed was varied in these experiments by a small amount (10% variation in rotation speed, corresponding to about 4% difference in the boundary layer thickness), yet a significant variation (~25%) is noted in the time delay before the large potential increase, with the lower rotation speed corresponding to the shorter time delay. This trend agrees with previously reported data.27
40
Figure 2.6. Potential responses to currents steps from zero to 19.6 mA/cm2 (at the
limiting current) for various rotation speeds: A – 90 rpm, B – 100 rpm, C – 110 rpm. The
time delay prior to the sharp potential increase shows a strong dependence on the rotation
speed.
41 2.4 Highly Resistive Surface Film
It is interesting to note that in all cases observed, the steep potential rise always
levels off at the same value (~1.6 V), corresponding to the potential at which oxygen
evolves in the copper electropolishing system, as shown in Figure 2.2. Performing
numerous galvanostatic experiments, it was observed that it was impossible to stabilize
the potential, which always drifted to ~1.6 V when the current was held at the limiting
current plateau. The resistance associated with the large potential drop (points E to F,
Figure 2.4b) when the current is stepped down to zero is quite high and can exceed
200 , depending on the magnitude of the current density. This is, by far, the dominant
resistance in the process; the ohmic resistance in the electrolyte and the activation
resistance are below 10 each. The ohmic resistance is estimated according to
Newman’s formula for resistance of a disk electrode,37 and the activation resistance is
estimated from the activation overpotential as determined by the slope of the polarization curve in the Tafel regime (below the limiting current). The overpotential for the oxygen
evolution is assumed to be negligible in the measured range. This conclusion is based on
the following observations. The 100 rpm curve in Figure 2.2 indicates that the current
begins to rise due to oxygen evolution at 1.61 V. In Figure 2.4b, the potential transient
stabilizes at 1.62 V. This difference is only 10 mV, while the entire potential rise in
Figure 2.4b is 1.4 V. Therefore, under the conditions of copper electropolishing, the
oxygen overpotential provides only negligible contribution to the total potential rise.
This high resistance suggests the presence of a highly resistive surface film. The
long induction time suggests that the creation of the film requires a concentration buildup
of a species at the anode. Under galvanostatic conditions, the film resistance is expected
42 to increase with time as the film grows until the potential drop across it reaches the oxygen evolution range. At that point, the process is no longer limited by the copper dissolution rate, and any additional potential increase will lead to enhanced oxygen evolution. When the current is stepped down to zero, the nearly instantaneous potential
decay indicates that the film resistance is of a nature different than that associated with
conventional mass transport. In the following chapter, this resistance is assumed to be
associated with ionic transport through a solid-like film, following a mechanism similar
to that described by the Mott-Cabrera model.38 In the latter, ionic transport resistance is known to respond much faster than in diffusion-related processes.
43 2.5 Electrochemical Impedance Spectroscopy
Electrochemical impedance spectroscopy measurements have been performed at
various rotation speeds (400-1800 rpm) by applying a 10 mV perturbation at frequencies between 1 and 20,000 Hz on potentials of various magnitudes, corresponding to current densities both below and at the limiting current plateau.
Figure 2.7 shows frequency impedance results at 400 rpm for various potentials below the limiting current plateau. The curves are nearly semicircles, indicating that the reaction is dominated by kinetics.39 The ohmic (solution) resistance remains
approximately constant at 2.8 Ω-cm2, but the polarization resistance varies with potential.
The polarization resistance decreases as the potential is increased, indicated faster reaction rate. The inductance loop observed in the low frequency limit, as well as similar inductance behavior for other dissolution systems, has not been well characterized.
These results below the limiting current plateau are in agreement with previously reported measurements.40
Frequency impedance results for an applied potential of 1.3 V (at the limiting
current plateau) at various rotation speeds (400-1800 rpm) are shown in Figure 2.8.
These measurements indicate an ohmic resistance of approximately 2.8 Ω-cm2 and a
polarization resistance between 1.6 and 3.4 Ω-cm2, indicating faster kinetics with
increased rotation speed. A mass transport resistance is denoted by lines at ~45° in the
low frequency limit, but this resistance is not purely diffusion based since the lines
deviate from 45°. These impedance measurements are in general agreement with
previously reported results.32 The very large film resistance (on the order of 70 Ω-cm2), measured in the current step experiments, is not seen in the frequency impedance
44 measurements. This is likely due to voltage perturbations (10 mV at 1-20,000 Hz) which
may disturb the formation of the film. As previously discussed, the latter process is
relatively long, of the order of 100 s.
A summary of the measured ohmic and polarization resistances is given in Table
2.1.
45
Figure 2.7. Nyquist plots at various applied potentials below the limiting current at
400 rpm. The ohmic resistance remains approximately constant, but the polarization resistance decreases as the potential is increased.
46
Figure 2.8. Nyquist plots at an applied potential of 1.3 V vs. copper (at the limiting current plateau) and various rotation speeds. These measurements indicate an ohmic resistance of approximately 2.8 Ω-cm2 and a polarization resistance between 1.6 and
3.4 Ω-cm2, which decreases with increasing rotation speed.
47 Table 2.1. Ohmic and polarization resistances measured by electrochemical impedance spectroscopy.
Potential Rotation speed Ohmic resistance Polarization resistance (V vs. copper) (rpm) (Ω-cm2) (Ω-cm2) 0.15 400 2.7 2.0 0.20 400 2.8 0.9 0.23 400 2.6 0.8 0.25 400 2.6 0.8 0.27 400 2.8 0.7 1.3 400 2.7 3.4 1.3 800 2.8 2.4 1.3 1200 2.8 2.0 1.3 1800 2.8 1.6
48 2.6 Stability of the Film in Presence of Chloride
Further indication of the presence of a surface film was provided by transient
experiments in the presence of trace amounts of chloride. In these experiments, the
current was stepped to the limiting current in an 85 wt% phosphoric acid solution
containing 100 ppm hydrochloric acid. The potential response is shown in Figure 2.9.
Initially, the potential response is similar to that observed in the absence of chloride
(Figure 2.4b); however, after about 230 s, rapid potential oscillations were noted for approximately 250 s, before stabilizing at ~1.6 V. This oscillatory behavior, which was not observed in the absence of chloride, can not be explained in terms of transport alone.
The trace concentration of Cl- should not appreciably affect the transport properties of the
system. The observed oscillations can be associated with the formation and breakdown
of a surface film in the presence of Cl-. Chloride ions have been implicated in the
breakdown of stable surface films in other systems, including those of passivating
oxides.41
49
Figure 2.9. Potential response to current step of 14.2 mA/cm2 (near the limiting current)
in 85 wt% H3PO4 solution containing 100 ppm HCl. The potential oscillations suggest
the formation and breakdown of a film.
50 2.7 Two-Compartment Cell Experiments
Additional experiments were performed in a two-compartment cell with glass
wool in the center section to separate the anode from the cathode. Upon application of a
constant current (at the limiting current plateau), the potential increases initially slowly
(~100 s) to ~0.3-0.4 V vs. copper, and then increases quickly (~10 s) to ~1.5 V vs.
copper. The potential remains stable at this value until the current is turned off; the
potential then rapidly decreases to nearly 0 V vs. copper. These observations correspond
to the results depicted in Figure 2.4b.
Additional observations have been made during electrodissolution in the
separated cell. When the current is held constant at a value initially on the limiting
current plateau and the solution near the anode is agitated, the potential decreases rapidly,
and then takes time to build up again to the previous constant value. This also occurs if
the anode is slightly moved in the solution. This behavior supports the hypothesis that a
film is formed at the anode under limiting current conditions. This film can be easily
disrupted by agitating the solution near the anode or moving the anode, causing shear
stress of the fluid. The film then again begins to rebuild over a longer time scale
(~100 s).
After several hours of dissolution, the potential did not decrease as rapidly when the current was suddenly turned off. The solution seemed more viscous close to the
anode and appeared darker in color near the bottom of the cell.
51 2.8 Conclusions
The copper electropolishing process was studied using the technique of transient current stepping. When the current was stepped to a value below the limiting current, the potential response exhibited expected behavior. Significantly different behavior was noted in the potential response when the current was stepped up to a value at the limiting current plateau. The potential response indicates a two-step process: an initial sharp potential increase followed by a slow gradual increase over ~70 s followed by a large, rapid potential increase (~1.5 V). This apparent two-step process can not be explained by a simple transport model. The steep potential increase suggests the presence of a highly resistive surface film.
52 CHAPTER 3
A Mechanistic Model for Copper Electropolishing
A mechanism, consistent with the experimental results reported in Chapter 2, is presented here. This mechanism considers the electropolishing process in terms of two distinct regimes. The first corresponds to the copper ion concentration buildup at the dissolving anode, and the second is associated with the thickness increase of a resistive film or layer on the anode surface.
3.1 Regime I – Buildup of Surface Copper Ion Concentration
When current is applied to a dissolving anode, copper ions are generated at the electrode/solution interface. Since the transport number of the copper ions in phosphoric acid is significantly less than 1, copper ions are generated at a rate faster than can be removed by migration, leading to the concentration buildup at the anode (schematic in
Figure 3.1a). The concentration continues to increase until the solubility limit of a copper
2+ - phosphate salt (e.g., (Cu )x(H2PO4 )y) is reached.
53 (b) x δ anode
i DCsat -C b nF bulk
film
Figure 3.1 Schematics representing (a) Regime I and (b) Regime II of the proposed
model. In Regime I, the concentration at the anode increases until Csat is reached. In
Regime II, a flux imbalance leads to the buildup in thickness (x) of a surface film.
54 The copper ion concentration profile near the electrode surface is governed in
Regime I by unsteady-state diffusion:
2CC D [3.1] x2 t
C is the copper ion concentration, and D is its diffusivity in phosphoric acid. The galvanostatic boundary conditions for Eq. [3.1] are
tCC 0, b [3.2]
x , CC b [3.3]
Ci x 0, [3.4] x nFD
The relationship between current density (i) and the total overpotential (η) at the anode can be described in terms of a modified Butler-Volmer equation.42
Ce nF 1 nF ii0 exp exp [3.5] CRTRTb
In Eq. [3.5], i0 is a concentration-dependent exchange current density, Ce and Cb are the copper ion concentrations at the anode and in the bulk, respectively, n is the number of electrons transferred in the reaction, F is Faraday’s constant, R is the universal gas constant, and T is the temperature. The first term on the right in Eq. [3.5] corresponds to the cathodic reaction, and the second term corresponds to the anodic reaction (thus a negative current density will represent a net anodic dissolution). The concentration ratio for the anodic reaction (second term on the right) involves the reduced species, i.e., metallic copper, and is set identically to unity.
The copper ion concentration profile as a function of time is obtained by solving
Eq. [3.1] subject to the boundary conditions given in Eqs. [3.2]-[3.4]. The relationship
55 between the current density and the overpotential given by Eq. [3.5] is used in conjunction with the boundary condition given in Eq. [3.4] to eliminate the current density. The finite difference solution provides the total overpotential as a function of time, which can be compared with the data in Figure 2.4a. Commercial software
(FEMLAB)43 has been used to solve Eq. [3.1] numerically.
56 3.2 Regime II – Controlling Transport through a Surface Layer
In Regime I, as described above, the copper ion concentration increases near the anode until the solubility limit is reached. At this point, a film begins to form on the anode surface. In Regime II, the growth of this film is determined by the imbalance between two copper fluxes (Figure 3.1b): (i) the dissolution of copper ions brought about by the current applied to the anode and (ii) the transport rate (by diffusion) of copper ions from the film into the bulk solution. When copper ions are produced by the externally applied current at a rate faster than can diffuse into the bulk, the thickness of the film and its resistance will increase with time. A quantitative analysis of the process follows.
Once the solubility limit of copper ions is reached, the copper ion concentration is given by:
Ksp Csat = 2 [3.6] - HPO24
In Eq. [3.6], Csat represents the copper ion concentration at saturation and Ksp is the solubility product. The exact nature of the precipitating copper salt has not been determined; however, its identification is not essential for the present analysis. A copper phosphate, e.g., Cu(H2PO4)2 (as represented in Eq. [3.6]), or a copper oxyphosphate are likely candidates.
The rate of generation of copper ions (N1) depends on the anodic current density according to:
i N [3.7] 1 nF
The rate at which the film dissolves is determined by the diffusion of copper ions towards the bulk:
57 DC C N sat b [3.8] 2
In Eq. [3.8] a linear (steady-state) concentration gradient across a boundary layer of thickness δ is assumed. The latter is determined by the electrolyte circulation (rotation speed of the disk electrode). Csat is the saturated copper ion concentration close to the anode surface as given by Eq. [3.6]. Since both the surface concentration and δ are assumed constant in Regime II, N2 will be constant. A material balance on the film, based on the difference between the (inflowing) copper dissolution flux, N1, and the (out- flowing) copper ion diffusion flux, N2, yields the growth rate of the film thickness, dx/dt:
dx i ()CC NN Dsat b [3.9] Mdt12 nF
In Eq. [3.9], ρ and M represent the density and the molecular mass of the surface film formed on the anode. Analytical solution of Eq. [3.9] indicates that the anode film thickness increases linearly with time:
Mi DC sat C b x t [3.10] nF
The ionic transport (of the dissolving copper ions) through the film is assumed to give rise to a Mott-Cabrera type resistance expressed as an exponential current-voltage relationship:38
V iA exp B [3.11] x
In Eq. [3.11], V/x is the electric field strength across the film, where V is the voltage drop across the film. A and B are constants related to the film properties. Their numerical values were assigned as discussed in the following section. Combining Eq. [3.10] and
[3.11] and solving for the potential provides:
58 Mii DC sat C b Vtln [3.12] BAnF
Under galvanostatic conditions, the model predicts that the potential increases linearly with time.
59 3.3 Model Verification
In this section, the model is quantitatively compared with the experimental current step measurements reported in Chapter 2, and qualitatively, with additional experimental observations. The model parameters are listed in Table 3.1. The values of most parameters are well known or have been independently determined. A few of the parameters, however, could not be evaluated, and their magnitudes were assigned. These relate primarily to the film properties: A, M, and ρ, as discussed below.
60
Table 3.1 Copper electropolishing model parameters. Justification for the estimated values is given in the text.
Parameter Name Value Units Source A film resistance parameter (Eq. 11) 4×10-3 A/cm2 fitted B film resistance parameter (Eq. 11) 8×10-6 cm/V Ref. 44 2+ -6 3 Cb bulk Cu concentration 10 mol/cm estimated -3 3 Csat saturation concentration of copper ions 1.2×10 mol/cm Ref. 30 D Cu2+ diffusion coefficient 10-7 cm2/s Ref. 31 F Faraday’s constant 96500 C/eq -4 2 i0 exchange current density 8×10 A/cm measured -6 3 Kd first dissociation constant of H3PO4 7.5×10 mol/cm Ref. 45 -8 3 9 Ksp solubility product of copper salt 10 mol /cm calculated from Csat M film molecular mass 250 g/mol estimated n number of electrons transferred 2 eq/mol R ideal gas constant 8.314 J/mol-K T temperature 298 K α transfer coefficient 0.6 - measured δ boundary layer thickness (100 rpm) 2.3×10-3 cm Ref. 35 ρ film density 2 g/cm3 estimated
61 Figure 3.2 compares the measured to the predicted overpotential response to a current pulse for Regime I. The measurements correspond to the experimental data depicted in Figure 2.4a. The model is based on the solution to Eqs. [3.1] and [3.5], as
-4 2 described above. The kinetic parameters (α = 0.6, i0 = 8×10 A/cm ) were independently obtained by fitting the Tafel equation to polarization data measured below the limiting
-3 3 current. Csat was taken as 1.2×10 mol/cm , based on data for copper dissolution in phosphoric acid.30 The only parameter that could not be independently determined for
-6 3 the Regime I model (Figure 3.2) was Cb. It was assigned a value of 10 mol/cm , since the bulk phosphoric acid is almost void of copper. The sensitivity of the model to changes in this parameter has been analyzed. Decreasing Cb by two orders of magnitude
(to 10-8 mol/cm3) has little effect on the shape of the curve and only shifts the curve by approximately 50 mV. Comparing (in Figure 3.2) the model to the experimental results indicates that the assumed value of 10-6 mol/cm3 is a good estimate. Both curves in
Figure 3.2 are in reasonably good agreement. The small deviation can probably be ascribed to the spatial distribution of the copper ions.
62
Figure 3.2. Comparison of measured and modeled overpotential response to a current pulse in Regime I. The measured data are from Figure 2.4a, while the predicted response is based on the numerical solution of Eqs. [3.1] and [3.5]. The two curves are in reasonable agreement. The small deviation can probably be attributed to the spatial distribution of the copper ions.
63 Figure 3.3 provides a comparison between the experimental data and the model for Regime II. The experimental data are those depicted in Figure 2.4b (section C-D), presented here, however, at a highly expanded time scale in order to display sufficient resolution. The model for Regime II corresponds to Eq. [3.12]. M, the molecular mass of the film, was assumed to be 250 g/mol, on the basis of the assumed composition of the film (copper phosphate). A value of 2 g/cm3 was assigned to ρ, based on the densities of similar salts, e.g., copper oxides and sulfates. The Mott-Cabrera parameter B (Eq. [3.11]) was taken as 8×10-6 cm/V, corresponding to an assumed 60 Å film thickness.44 The parameter A could not be independently measured and was treated as an adjustable parameter. Figure 3.3 indicates an experimentally observed and analytically predicted steep linear voltage ramp. It should be noted that while the magnitude of the modeled slope depends on the numerical value of the parameters, its linearity does not. Subject to the limitations discussed above, the model correlates very well the steep potential rise associated with stepping the current to the limiting current value.
As stated above, two adjustable parameters, A and M, are applied to the model.
The sensitivity of the model to these two parameters was considered by determining the percent variation in the slope of Eq. [3.12] for changes in both A and M. This analysis is summarized in Figure 3.4, indicating the allowable changes in A and M so that the model falls within 10% or 25% of the slope correlating the data. It is noted that variations of up to 25% in the value of M or up to 13% in the value of A from the selected values still correlate the data to within 25%.
64
Figure 3.3. Comparison of the model governing Regime II (Eq. [3.12]) to the experimental potential response displayed in Figure 2.4b (region C-D). Note the highly expanded time scale.
65
Figure 3.4. Sensitivity analysis for parameters A and M (Eq. [3.12]). The lines indicate values for these parameters such that the model correlates the data from Figure 2.4b
(region C-D) within the indicated percentages.
66 3.3.1 Time Delay Prior to the Onset of the Sharp Potential Increase
The model is consistent with and provides an explanation for the transient behavior of the potential responses depicted in Figures 2.5 and 2.6. Figure 2.5 indicates that the time delay prior to the sharp potential increase is shorter at higher current densities. At higher current densities, the rate of generation of copper ions is greater, while the rate of copper ions diffusing out of the boundary layer region remains the same
(since the rotation speed is constant). Hence, less time is required for the concentration buildup to the saturation level. Figure 2.6 displays a related phenomenon, whereby higher disk rotation speeds correspond to a longer time delay prior to the onset of the sharp potential increase. At higher rotation speeds, the boundary layer is thinner and hence the transport rate of copper ions into the bulk is greater. Consequently, since the copper ions influx is the same, more time is required for the concentration of copper ions to build up at the anode to the saturation level.
3.3.2 Effect of Water Concentration on the Limiting Current
A linear relationship between the limiting current and the concentration of water in the bulk has been reported previously.33 This and other observations have led several investigators to suggest that the electropolishing process is limited by the transport of water as an acceptor species.31-33 This linear relationship is, however, consistent with the proposed model. The model implies that the controlling transport is that of copper species, subject (per Eq. [3.8]) to the concentration driving force Csat-Cb, where Cb is a very small quantity that does not vary much. Furthermore, as shown below, the saturated copper salt concentration, Csat, is, to a very close approximation, almost linearly
67 dependent on the bulk water concentration. Accordingly, if Csat is substituted in Eq. [3.8] with its equivalent water concentration, it will appear as if the electropolishing process is linearly dependent on the water concentration, whereas in reality, the significant parameter is the saturation concentration of the copper salt. Details of the derivation are described here.
The dissociation of phosphoric acid can be represented by
HHPO+- 24 63 Kd 7.5 10 mol/cm [3.13] HPO34
45 where Kd is the first dissociation constant. Since the dissociation constants for the second and third protons are much smaller, these can be neglected. The saturation concentration of cupric ions is given by Eq. [3.6]. Ionic mass balance, accounting for electroneutrality, yields:
+- 2HHPO0Csat 24 [3.14]
+ 2- In deriving Eq. [3.14] it has been assumed that the concentrations of Cu , HPO4 , and
3- PO4 are negligible.
Combining Eqs. [3.6], [3.13], and [3.14] gives:
12 12 C K 20CKC sat sp [3.15] sat d H3PO4 KCsp sat
Eq. [3.15] relates the saturation concentration of copper ions to the bulk phosphoric acid concentration (both in mol/cm3).
Instead of relating Csat to the concentration of phosphoric acid as in Eq. [3.15], it is desired to relate Csat to the concentration of water in the bulk solution, CH2O. A
68 relationship between the concentration of H3PO4 and the concentration of water can be determined as follows.
The weight fraction of phosphoric acid, yH3PO4, can be described by Eq. [3.16], where MH3PO4 is the molecular mass of phosphoric acid, CH3PO4 is the concentration of phosphoric acid in mol/cm3, and ρ is the density of the solution.
M yC H3PO4 [3.16] H3PO4 H3PO4
An analogous equation can be written for the weight fraction of water, yH2O.
M yC H2O [3.17] H2O H2O
The weight fractions of the two components must add to unity:
M M CCH3PO4 H2O 1 [3.18] H3PO4 H2O
The density will vary for solutions of different compositions and is therefore a function of the phosphoric acid concentration. Density data46 as a function of phosphoric acid concentration for the range of interest (60-85 wt%) show a nearly linear increase in density as the phosphoric acid concentration increases. This density data can be approximated by a linear correlation. Substituting this density correlation and the values for the molecular masses into Eq. [3.18] and solving for CH3PO4 provides the following approximate correlation between the concentrations of phosphoric acid and water (in mol/cm3).
CCH3PO4 0.019 0.33 H2O [3.19]
This relationship is then substituted into Eq. [3.15] to derive Eq. [3.20], relating the water concentration to the saturation concentration of copper ions.
69 12 12 C K 2CK 0.019 0.33 C sat sp 0 [3.20] sat d H2O KCsp sat
While rigorously indicating a non-linear relationship, Eq. [3.20] can be closely approximated in the range of interest (phosphoric acid concentration exceeding 60 wt%) by a linear correlation between Csat and the water concentration:
43 CCsat3.0 10 H2O 1.3 10 [3.21]
Applying the linear approximation of Eq. [3.21] to substitute Csat in the limiting current expression results in the following equation.
43 nFD3.0 10 CH2O 1.3 10 C b i [3.22] L
The parameters D and δ in Eq. [3.22] are somewhat dependent on the water concentration. When this is accounted for, the linear relationship implied by Eq. [3.22] is no longer strictly valid. However, as noted in Figure 3.5, this variation is rather small and a linear approximation can still be applied as a close approximation.
Eq. [3.22] indicates a nearly linear transport dependence on the water concentration, although the mechanistic model identifies the saturated copper salt as the transport limited species. The substitution of Csat by the water concentration is not inherent to the mechanism; it is the consequence of the almost linear dependence of Csat on the water concentration in the range of interest and the specific stoichiometry.
70
Figure 3.5. The model (Eq. [3.22]) predicts a nearly linear relationship between the limiting current density and the bulk water concentration for copper electrodissolution in phosphoric acid. Also indicated is a linear approximation for Eq. [3.22].
71 3.4 Conclusions
Studies of copper electrodissolution using current step techniques indicate a potential response that cannot be explained in terms of a simple mass transport model. A two-step mechanism for the copper electropolishing process, which is consistent with the transient potential behavior, is presented. A flux imbalance between the dissolving copper ions and their transport out of the boundary layer region is shown to cause a buildup of copper ions near the anode surface. This concentration increases until the solubility limit of the copper salt is reached and a resistive film is formed. During the second stage, the resistive film increases in thickness due to the imbalance between the current-dependent dissolution and the removal of copper ions by transport. Under constant current, the film thickness increases linearly with time, giving rise to a corresponding increase in the potential. The proposed mechanism accounts for the potential response to current transients and is consistent with experimental observations, including those relating to the effect of the bulk water concentration.
The model described herein provides mechanistic insight into the electropolishing process on a flat electrode. With a more detailed understanding of the surface film properties, this mechanism could be extended to predict planarization on a patterned substrate. Some limitations of the electropolishing process remain. The planarization uniformity will vary over the wafer surface, which could lead to a loss of electrical contact if all of the copper is removed at some locations. The removal of islands of remaining copper formed by non-uniform planarization would still require a CMP step.
72 CHAPTER 4
Novel Polyether Suppressors Enabling Copper Metallization of High Aspect Ratio
Interconnects
In order to achieve void-free gap-fill of metalized copper features, the average rate of deposition on the feature bottom must exceed the rate at which the sidewalls merge, i.e.,
i L B [4.1] irSW iB and iSW represent the average current densities at the upward-propagating via bottom and at the merging sidewalls, respectively. L is the via length, and r is its radius. It has been shown that the via sidewalls have similar additives coverage to the via top (rim) and propagate at about the same rate, which is substantially lower than that of the via
3 bottom. As indicated below, PEG gives rise to a differential plating rate (iB/iSW) of about
15 (depending on the average current density, the geometry, and other process parameters), enabling, according to Eq. [4.1], the bottom-up fill of features with aspect ratio (L/2r) of up to about 7.5. Future generation semiconductor devices will incorporate smaller via radii (e.g., <30 nm for the 22-nm microprocessor technology node), making it increasingly challenging to meet the conditions indicated by Eq. [4.1]. An approach to overcome this barrier would be to identify stronger suppressors that provide greater inhibition of the deposition on the via sidewalls.
Not many guidelines are available for this endeavor. Compounds that contain polyether functionality are sought, since this group provides adsorption to the copper, yet is gradually displaced by the anti-suppressor (SPS) through competitive adsorption, an
73 essential requirement for the bottom-up fill.3 Furthermore, in order to exhibit the required diffusion limitation, a relatively large molecule is sought, with a weight of
~1000 g/mol, corresponding to a molecular radius in the 2-4 nm range, a size sufficiently small to enable penetration into the narrow features yet providing the necessary transport limitations.3 This relatively large molecular mass is typically indicative of a polymer.
Although the suppression effect of PEG is well characterized,3, 5-23, 47 the underlying molecular mechanism is not. One proposed mechanism is that the adsorbed PEG acts as a physical diffusion barrier to cupric ions, slowing the rate of copper deposition.15 A different mechanism proposes that chloride ions adsorbed on the surface coordinate with cuprous ions and the oxygen atoms of the PEG, inhibiting the reduction of Cu+ ions.12, 15, 16
The objective of this work is to identify more effective suppressors that can extend the gap-fill capability to narrower features.
74 4.1 Experimental Procedure
Polarization studies of copper deposition, comparing the inhibition provided by the tested compounds to that of PEG, offer the simplest and the most direct indication of the efficacy of the inhibitor. Galvanostatic experiments were carried out on a flat 6 mm diameter copper (OFHC) disk electrode rotating at 200 rpm, polarized using a Solartron
1280B potentiostat. The copper disk was mechanically polished with silicon carbide paper (FEPA #1200 followed by #4000) prior to each experiment. All polarization experiments were carried out in 100 mL solutions containing acidified 0.5 M CuSO4
(pH~2), 70 ppm Cl-, and 100 ppm of the specified suppressor. The anode was a 5.5 cm diameter copper disk placed at the bottom of a 250 mL beaker. An insulated copper wire with an exposed tip served as the reference electrode. Polarization curves were obtained by performing a series of galvanostatic experiments. A constant current was applied until a steady-state was reached (~1-2 min), and the potential was recorded. This procedure was repeated for numerous current densities below 10 mA/cm2, corresponding to the current density range expected within the via.
75 4.2 Results and Discussion
The various polyethers, listed in Table 4.1, were studied by comparing their polarization behavior to that of PEG. The rationale for narrowing the search to polyethers, as discussed above, has been the requirement for their gradual displacement by the anti-suppressor through competitive adsorption, a feature that polyethers exhibit.
Although the targeted molecular mass has been about 1000 g/mol, the range of studied molecular masses has been determined by the commercial availability of the compounds.
76 Table 4.1 Polyethers explored in the polarization studies.
Name Structure Molecular Overpotential Bottom- Mass (mV) at 5 up Fill (g/mol) mA/cm2 Ratio at 20 A TM H O H UCON O O 270 217 62 Lubricant 50-HB- n n H3C 55 (trademark of the Dow Chemical Company)
Poly(ethylene H O 300 86 2 O H glycol) n Poly(ethylene O 324 175 22 glycol) phenyl O H2C ether acrylate O n
CH3 Poly(ethylene CH3 404 170 21 glycol) mono[4- O OH H3C (1,1,3,3- CH3 CH3 n tetramethylbutyl) phenyl] ether
Poly(ethylene O O 410 145 11 glycol) dibenzoate O O n
Poly(ethylene O 526 168 24 glycol) diglycidyl O O n ether O
Tergitol NP-7 H C O 550 158 18 3 OH n
Poly(ethylene H O 600 151 21 O H glycol) n
Poly(ethylene H C O 600 171 24 3 OH glycol) n monolaurate O
Tergitol NP-9 H C O 617 175 28 3 OH n
Triton N-101 H3C 644 192 34 H3C OH H3C O n CH3 H3C
CH3 Poly(ethylene CH3 650 173 25 glycol) mono[4- O OH H3C (1,1,3,3- CH3 CH3 n tetramethylbutyl) phenyl] ether
77 O Polyoxyethylated OH 716 191 41 β-naphthol n
CH Poly(ethylene O 2 726 191 36 glycol) diacrylate O O n O
CH2 Poly(ethylene O 930 83 2 CH CH glycol) distearate 2 O 3 H C 3 16 O CH2 n 16 O
UCON Lubricant H O H 980 178 28 O O 75-H-450 m n H3C
Poly(ethylene H O 1000 158 16 O H glycol) n
O Polyoxyethylene H3C OH 1124 192 25 cetyl ether n
Polyoxyethylene O 1200 191 33 H3C OH lauryl ether n
CH3 Poly(ethylene CH3 1746 203 44 glycol) mono[4- O OH H3C (1,1,3,3- CH3 CH3 n tetramethylbutyl) phenyl] ether
Tergitol NP-40 H C O 1980 190 35 3 OH n
Poly(propylene CH3 3350 139 8 glycol) H OH O n
UCON Lubricant H O H 3930 236 66 O O 50-HB-5100 n n H3C
Poly(ethylene H O 4000 195 30 O H glycol) n
78 4.2.1 Polarization Data
Figure 4.1 shows polarization data for polyethers of molecular mass of about
1000 g/mol. For reference, also shown are the polarization data for a solution containing no suppressor. One polyether, poly(ethylene glycol) distearate, was identified as not being as effective a suppressor as PEG. This molecule has large alkyl chains on both sides of the oxyethylene chain. Three polyethers which show significantly improved suppression over PEG have been indentified: UCONTM Lubricant 75-H-450 (trademark of the Dow Chemical Company) which is a copolymer of oxyethylene and oxypropylene, polyoxyethylene lauryl ether, and polyoxyethylene cetyl ether, both of which have a large alkyl chain on one end of the oxyethylene segment.
79
- Figure 4.1. Polarization data for solutions containing 0.5 M CuSO4 (pH~2), 70 ppm Cl , and 100 ppm of the specified polyether, all with molecular mass of approximately
1000 g/mol. The data points for polyoxyethylene lauryl ether (diamonds) and polyoxyethylene cetyl ether (circles) fall nearly on top of one another.
80 Due to commercial availability, polyethers of molecular mass of about 600 g/mol were also studied, although the latter is somewhat lower than the target molecular mass of 1000 g/mol. Polarization data for solutions containing these polyethers are shown in
Figures 4.2 and 4.3. Several polyethers indicate improved suppression over PEG: polyethylene glycol diglycidyl ether, polyethylene glycol mono[4-(1,1,3,3-tetramethylbutyl)phenyl] ether, Tergitol NP-7, Tergitol NP-9,
Triton N-101, polyethylene glycol diacrylate, polyethylene glycol monolaurate, and polyoxyethylated β-naphthol. As shown in Table 4.1, these polyethers are structurally similar to PEG but have large groups on the end of the oxyethylene chain. These results suggest that the differences in chemical structure are likely to impact the suppression behavior.
81
- Figure 4.2. Polarization data for solutions containing 0.5 M CuSO4 (pH~2), 70 ppm Cl , and 100 ppm of the specified polyether with molecular mass of approximately 600 g/mol.
Several polyethers indicate improved suppression over PEG: polyethylene glycol diglycidyl ether, polyethylene glycol mono[4-(1,1,3,3-tetramethylbutyl)phenyl] ether,
Tergitol NP-7, and polyoxyethylated β-naphthol. These polyethers are structurally similar to PEG but have large groups on the end of the oxyethylene chain, suggesting that the differences in chemical structure are likely to impact the suppression behavior.
82
Figure 4.3. Polarization data for additional solutions containing 0.5 M CuSO4 (pH~2),
70 ppm Cl-, and 100 ppm of the specified polyether with molecular mass of approximately 600 g/mol. Several polyethers indicate improved suppression over PEG: polyethylene glycol monolaurate, Tergitol NP-9, Triton N-101, and polyethylene glycol diacrylate.
83 Polyethers of molecular mass of about 300 g/mol were also studied, although these are smaller molecules than the desired size. Polarization data for solutions containing these polyethers are shown in Figure 4.4. PEG 300 only exhibits slight polarization (~30 mV) compared to a solution containing no suppressor. In this molecular mass range, four polyethers have been identified as more effective suppressors than PEG 300. These polyethers are polyethylene glycol dibenzoate, polyethylene glycol mono[4-(1,1,3,3-tetramethylbutyl)phenyl] ether, polyethylene glycol phenyl ether acrylate, and UCON Lubricant 50-HB-55. As noted in Table 4.1, UCON Lubricant
50-HB-55 is a copolymer of oxyethylene and oxypropylene. Polyethylene glycol mono[4-(1,1,3,3-tetramethylbutyl)phenyl] ether is similar to PEG but with a large group on one end of the oxyethylene chain. Polyethylene glycol dibenzoate and polyethylene glycol phenyl ether acrylate have large groups on both ends of the oxyethylene chain. At this low molecular mass, these polyethers are more effective than PEG but not as effective as other polyethers of higher molecular mass.
One additional range of molecular masses was studied. Figure 4.5 shows polarization data for solutions containing polyethers of molecular mass of about
2000 g/mol to 4000 g/mol. Two polyethers exhibit nearly the same or slightly improved suppression compared to PEG 4000. These polyethers are Tergitol NP-40 and polyethylene glycol mono[4-(1,1,3,3-tetramethylbutyl)phenyl] ether, which both have large groups on one end of the oxyethylene chain. One polyether, polypropylene glycol, does not show an improvement over PEG of a similar molecular mass. However, a copolymer of oxyethylene and oxypropylene, UCON Lubricant 50-HB-5100, shows an improvement in suppression over PEG.
84
- Figure 4.4. Polarization data for solutions containing 0.5 M CuSO4 (pH~2), 70 ppm Cl , and 100 ppm of the specified polyether with molecular mass of approximately 300 g/mol.
85
- Figure 4.5. Polarization data for solutions containing 0.5 M CuSO4 (pH~2), 70 ppm Cl , and 100 ppm of the specified polyether with molecular mass in the range of approximately 2000 g/mol to 4000 g/mol.
86 The effects due to variations in molecular mass in the range explored (300 to
4000 g/mol) is not expected to be significantly large, as seen from the polarization data above. Figure 4.6 presents the extent of suppression as a function of the number of ether oxygen atoms in the molecule. Here, the suppression is quantified as the overpotential measured at 5 mA/cm2. The circles indicate PEG at various molecular masses, while the squares correspond to the other polyethers (Table 4.1). The left-most circle in Figure 4.6 corresponds to PEG 300, the next circle to the right is PEG 600, followed by PEG 1000, and the right-most circle is PEG 4000. As noted, the overpotential increases somewhat as the PEG molecular mass increases; however, it appears to reach an asymptote between molecular mass 1000 – 4000 g/mol.
Although some trend between increased suppression and the number of ethereal oxygen atoms is noted, the spread of the data indicates that no simple correlation exists.
This implies that not only the molecule size or the number of ethereal oxygen atoms is important, but also the chemical structure has a major influence on the suppression efficacy.
It appears that substituting the hydrogen atom on the end of PEG with a larger end group causes increased polarization and therefore is beneficial. One exception is poly(ethylene glycol) distearate. However, this compound has a double-bonded oxygen atom next to the ether chain, which may interfere due to its negative charge. Furthermore, a larger end group (e.g., as in polyoxyethylated β-naphthol and polyoxyethylene lauryl ether) appears to provide enhanced inhibition. However, when the end group is too large with respect to the polyether chain, the inhibition enhancement diminishes. This is observed in poly(ethylene glycol) distearate and poly(ethylene glycol) dibenzoate.
87
Figure 4.6. Overpotentials at 5 mA/cm2 as a function of the number of ether oxygen atoms in the polyether. The circles correspond to PEG, and the squares are all other polyethers studied (listed in Table 4.1). Although some trend is indicated between increased overpotential and the number of ethereal oxygen atoms, the data spread implies that other factors, including the chemical structure, are important.
88 4.2.2 Modeled Via-fill Ratio
To achieve bottom-up fill, the ratio iB/iSW must be large, exceeding L/r. The via- fill ratio can be approximated as the ratio of the current density at the via bottom (iB) to the current density at the via sidewalls (iSW). The current densities can be represented in terms of the Tafel approximation to the Butler-Volmer equation, using the kinetics parameters (i0, the exchange current density, and αC, the cathodic transfer coefficient) shown in Table 4.2, fitted from the polarization curves. The variation in the αC values indicates that the kinetics mechanism is somewhat different for the three compounds, i.e., the difference in the end groups affects the deposition mechanism.
The current density at the sidewalls is given by:
F CS, RT iieSW 0, S [4.2]
The current density at the via bottom is assumed to correspond to copper deposition kinetics in the presence of the anti-suppressor:3
F CAS, RT iieBAS 0, [4.3]
The subscripts ‘S’ and ‘AS’ in Eqs. [4.2] and [4.3] designate values for the suppressor and anti-suppressor, respectively.
Applying Eqs. [4.2] and [4.3] to Eq. [4.1]:
()F i i CAS,, CS L B 0,AS e RT [4.4] iiSW0, S r
89 Table 4.2. Kinetics parameters fitted to polarization data in Figure 4.1 and Figure 4.2.
2 Name i0 (A/cm ) αc Overpotential (mV) at 10 mA/cm2 Poly(ethylene glycol) 1000 1.9x10-4 0.55 180 Polyoxyethylene lauryl ether 1.8x10-4 0.46 224 Polyoxyethylated β-naphthol 5.5x10-5 0.60 224
90 Clearly, in order to achieve an effective bottom-up fill, a suppressor with low exchange current density, i0,S, and a small cathodic transfer coefficient, C,S, should be applied. Also, the overpotential, η, and the corresponding average current density should be as high as practically possible. iB and iSW can be related to the total current, I, through the approximate relationship:
I iAB B i SW A supp [4.5]
The area at the bottom of the features, which is covered primarily by the anti-suppressor, is AB. The total area that is covered predominately by suppressor, the feature sidewalls and the wafer top surface, is designated as Asupp. The pattern density on a given substrate is fixed and does not vary with deposition. The change in the via sidewalls due to deposition is assumed to be small with respect to the total suppressed area, Asupp. Thus, the areas AB and Asupp are also fixed. The changes in those areas during the fill process are not accounted; hence the approximation is strictly valid only for the initial stages of the bottom-up fill. When the fraction of features is small, the validity range of the approximation can be further extended. Substituting the current densities in Eq. [4.5] in terms of the Tafel approximations (Eqs. [4.2] and [4.3]):
FF CAS,, CS RT RT I ie0,AS Aie B 0, S A supp [4.6]
Selecting different overpotential values in the range of interest, the partial current densities iB and iSW can now be computed from Eqs. [4.2] and [4.3] and, applying
Eq. [4.6], also the total current, I. Figure 4.7 shows the expected via-fill ratio iB/iSW for
PEG 1000, polyoxyethylene lauryl ether, and polyoxyethylated β-naphthol as a function of the total current applied to a 300 mm wafer with 15% pattern density. This pattern density, averaged over the entire wafer, is typical of back-end copper metallization
91 wafers. The bottom-up fill ratios of various suppressors are compared at a given total current since commercial wafer metallization is done under galvanostatic conditions. The modeled via-fill ratios for polyoxyethylene lauryl ether and polyoxyethylated β-naphthol are both significantly larger than that for PEG 1000, with the via-fill ratio of polyoxyethylated β-naphthol being up to 4 times larger than that of PEG 1000. Although the polarization enhancement for these polyethers is only moderate (as noted in Table 4.1 and in Figures 4.1 through 4.5), the effect on the via-fill ratio is quite significant. For example, the smallest diameter of a 0.5 μm deep via that can be filled in the presence of
PEG is about 70 nm, even at relatively high current densities (iB/iSW ~ 15 from Figure
4.7). By comparison, polyoxyethylated β-naphthol is expected to provide bottom-up fill of 25 nm vias at the same total current (iB/iSW ~ 40), thus extending the bottom-up fill to the target 22 nm technology node (30 nm vias). The bottom-up fill ratio (iB/iSW) at a total applied current of 20 A was calculated for each suppressor; these values are given in the last column of Table 4.1.
92
Figure 4.7. Bottom-up fill ratio (iB/iSW) as a function of total current simulated for a
300 mm wafer with 15% feature loading (pattern density) for solutions containing one of the following suppressors: PEG 1000, polyoxyethylene lauryl ether, or polyoxyethylated
β-naphthol. Significant improvement (3 ~ 4x) is expected by replacing PEG with either of the above listed other polyethers.
93 The effectiveness of polyoxyethylated β-naphthol in the via fill was confirmed by the plating of vias with an aspect ratio (L/2r) of about 10. SEM cross-section images,48 shown in Figure 4.8, compare fill performance in the presence of two suppressors:
PEG 1000 and polyoxyethylated β-naphthol. The vias plated in the presence of PEG indicate center-line voids while the polyoxyethylated β-naphthol leads to void-free fill due to its improved suppression and higher bottom-up fill rate (Figure 4.7).
94 (a) PEG 1000 Bottom-up fill voids
(b) POE β-naphthol Void-free fill
Figure 4.8. SEM cross-sections of vias with aspect ratio close to 10 electroplated with copper after PVD seed deposition.48 Fill quality is compared for copper plating in the presence of two different suppressors: (a) PEG 1000, exhibiting center-line voids due to inferior bottom-up fill rate (Figure 4.7), and (b) polyoxyethylated (POE) β-naphthol, indicating void-free fill on account of its improved suppression and higher bottom-up fill rate.
95 4.2.3 Interaction with the Anti-suppressor
As stated above, the bottom-up fill process hinges on the displacement of the suppressor by the anti-suppressor over time due to competitive adsorption. For an alternative species to be considered as a replacement for PEG, it must undergo similar competitive displacement by SPS as the PEG does. To this end, the interactions of these polyethers with SPS have been investigated, using injection studies similar to those reported by Akolkar and Landau.3 A disk electrode is initially polarized at a current
2 - density of 5 mA/cm in 0.5 M CuSO4 (pH~2) with 70 ppm Cl and 100 ppm of a specified polyether. After steady-state has been reached, 2 mL of concentrated SPS solution was injected into the original solution such that the resulting solution contained
10 ppm SPS. The voltage response was recorded as a function of time. The data for
PEG 1000, polyoxyethylene lauryl ether, and polyoxyethylated β-naphthol are shown in
Figure 4.9. Polyoxyethylated β-naphthol is displaced by SPS in a similar time scale to that of PEG, while polyoxyethylene lauryl ether is displaced over a slightly longer time scale. These results indicate that the studied polyethers can be utilized as suppressors for gap-fill applications because they can be displaced by SPS in a similar manner to PEG.
96
Figure 4.9. Voltage transient response at 5 mA/cm2 to SPS injections into polyether
- polarized deposition. The solution initially contained 0.5 M CuSO4 (pH~2), 70 ppm Cl , and 100 ppm of the specified polyether. After steady-state was reached, 10 ppm SPS was injected into the solution. Polyoxyethylated β-naphthol has a similar voltage response to that of PEG.
97 4.3 Conclusions
Several polyether compounds, including polyoxyethylated β-naphthol, UCON
Lubricant 75-H-450, and polyoxyethylene lauryl ether, have been indentified, all exhibiting significantly improved suppression over PEG. Injection studies confirm that these molecules are displaced by SPS during gap fill similarly to PEG. The expected bottom-up fill ratio for these polyethers is about 4 times greater than that of PEG, making them attractive as replacements for PEG, thus extending the bottom-up fill process to features as narrow as 25 nm.
98 CHAPTER 5
Mechanistic Studies of Polyether Adsorption
Although the macroscopic mechanism of the role of additives in copper plating is well understood, allowing for the successful design of the bottom-up fill process, an understanding of the molecular mechanism by which these additives affect the copper plating process is lacking. As stated in Chapter 1, two mechanisms have been proposed to account for the PEG inhibiting action. The first assumes that the adsorbed PEG binds or interacts with the Cu+ ion, an intermediate in the copper plating process.5, 12, 15, 16 The second model assumes that adsorbed PEG only serves as a barrier to the depositing species, i.e., presenting a steric effect.15 It is known that PEG only slightly suppresses copper deposition in the absence of Cl-, while Cl- on its own is actually a slight deposition accelerator.3 By contrast, PEG and Cl- together provide significant suppression. It has not been determined how PEG and Cl- interact to provide this suppression to copper deposition. Greater insight into this mechanism may be beneficial to the understanding of why some polyether suppressors are more effective than PEG.
This knowledge could allow for the identification and possible synthesis of more effective additives. Two techniques used in the study of polyether adsorption mechanisms are discussed here: attenuated total reflectance Fourier transform infrared spectroscopy (ATR-FTIR) and quartz crystal microbalance (QCM).
FTIR spectroscopy has been previously applied to studies of the adsorption of molecules on copper. External reflection studies have been reported for the adsorption of thiourea49 and benzotriazole50 on copper electrodes. Internal reflection (or ATR)
99 techniques have been reported for the study of adsorption of organic compounds, including gum arabic,51 dextran,52 and pyridine,53 on copper.
QCM has been used by several researchers to monitor metal electrodeposition54-56 and the adsorption of molecules such as pyridine57 and PEG8 on metal surfaces.
In this chapter, the above techniques were applied to study the adsorption behavior of PEG.
100 5.1 Experimental Details
5.1.1 Attenuated Total Reflectance Fourier Transform Infrared Spectroscopy
(ATR-FTIR)
To quantify the adsorption of PEG and other polyethers identified in the previous chapter, ATR-FTIR (using a Bomem MB-157 spectrometer operated at 4 cm-1 resolution using a mercury cadmium telluride detector) was employed using a setup described previously.58 The advantage of this approach is that the beam is probing only a narrow region (~0.25 μm – 2 μm, depending on the wavelength) at the electrode/electrolyte interface, where the adsorption takes place, rather than probing the electrode from the solution side, sampling also a thick electrolyte region that contains significant amounts of water and dissolved species. As shown in Figure 5.1, the device consists of a ZnSe prism
(n = 2.89) against which a thin (50 μm) IR-transparent silicon wafer (n = 3.4) (Virginia
Semiconductor), coated with a very thin (~5 nm) copper layer (n = 2.43), is pressed. A laser beam is introduced through the ZnSe prism where it is internally reflected, creating an evanescent wave.59 This wave continues through the prism and Si from the substrate
(back) side and samples a region at the electrode/electrolyte interface, extending into the solution on the order of 1 μm. The copper layer, prepared by electron beam deposition, must be very thin in order to be sufficiently transparent to the laser beam. The thinness of the copper layer imposes limitations on the ability to conduct measurements under conditions where copper will dissolve or deposit. The 5 nm thick copper layer will completely dissolve or double its thickness due to deposition becoming non-transparent, in about 2 s even at low current density of 5 mA/cm2, while a typical scan takes ~5 s. A typical spectrum described here, obtained by collecting 100 scans, takes several minutes.
101
Figure 5.1. Schematic of ATR-FTIR system (not to scale). The laser beam is introduced from the back, through a thin Si wafer coated with Cu. Only a thin region (~1 μm) at the electrode/electrolyte interface is sampled.
102 5.1.2 Quartz Crystal Microbalance (QCM)
The technique is based on measuring the change in the vibration frequency of two metalized crystal surfaces, one of which is subject to the polyether adsorption. The change in frequency, Δf, is related to the change in mass per area, Δm, through the
Sauerbrey equation.60
2nf2 f h 0 m [5.1] 1 2 qq nh is the number of the harmonic at which the crystal is being driven, f0 is the resonant frequency of the fundamental mode of the crystal, ρq is the density of quartz, and μq is the shear modulus of quartz. An increase in mass at the mass-sensitive electrode will cause a decrease in the frequency of oscillation of this electrode. The solution density and viscosity and changes in hydrostatic pressure and temperature can also affect the frequency. In this work it is assumed that these effects on frequency are very small compared to the effect of mass change.
The crystals used for the QCM technique were 10 MHz quartz crystals, 14 mm in diameter, 0.166 mm thick, AT-cut, with 5 mm diameter deposited disk electrodes of
200 nm Au over 10 nm Ti (ELCHEMA). The Au disk is the mass-sensitive area of the crystal. A schematic of the cell is shown in Figure 5.2. The crystal is held in place between two o-rings, with one electrode in contact with solution and the other with air, serving as a (frequency) reference. A high-frequency oscillation is applied to the crystal, and the difference in frequency between the electrode in contact with the solution and the reference electrode is recorded. In the following experiments, the mass-sensitive electrode is at a lower frequency than the reference electrode. An increase in mass at the
103 mass-sensitive electrode results in a decrease in frequency at this electrode and therefore an increase in the frequency difference between the two electrodes. The cell and the electronics for signal generation and recording were custom built.
104
Figure 5.2. Schematic of QCM cell (not to scale). The top electrode is exposed to the solution, while the bottom electrode is in contact with air, serving as the reference.
105 5.2 ATR-FTIR Studies of PEG Adsorption
The spectrum for solid PEG, relative to ZnSe-air, is shown in Figure 5.3. This spectrum was used as a reference to identify the relevant peaks for PEG. Because this scan was measured for PEG directly on ZnSe, the magnitudes of the peaks in Figure 5.3 are expected to be larger than peaks for solution spectra on ZnSe-Si-Cu. The most prominent peaks are the C-O-C stretch at 1109 cm-1 and the C-H stretch at 2885 cm-1.61
Since it is believed that the ethereal oxygen is the active group in the adsorption of polyethers, the C-O-C stretch peak is of primary interest. The peak at 2361 cm-1, shown here and in subsequent spectra, is due to variations in atmospheric CO2 levels near the detector during and between experiments. 61
The spectrum for solid cupric sulfate pentahydrate on ZnSe is shown in Figure
5.4. The most prominent peaks are at 670, 3630, and 3730 cm-1. There is also a peak at
1090 cm-1, which is very close to the C-O-C stretch peak of PEG at around 1100 cm-1.
- Figure 5.5 depicts the spectrum for a solution containing 0.5 M CuSO4 and 70 ppm Cl on a Cu-coated Si wafer relative to water on Cu-coated Si. A prominent peak is evident at about 1100 cm-1, near the peak of interest for PEG. Due to their overlap, resolving those peaks is very difficult, particularly since the wafer/electrolyte set-up must be dismantled and reassembled when a new electrolyte is introduced, causing lack of reproducibility in the magnitude of the signal between experiments. The reproducibility of the technique was measured by performing a series of scans with a copper sulfate solution, dismantling and reassembling the wafer/electrolyte set-up each time. Referencing one scan to another scan in the series did not result in a flat line; peaks associated with water and sulfate were evident. This leads to some error when a spectrum is shown relative to a reference
106 spectrum (water in Figure 5.5). The peaks noted at 3280 and 1640 cm-1 in Figure 5.5 are associated with water61 and appear due to errors in referencing the spectrum of interest to the spectrum for water on Cu. These peaks are of no interest to measuring the PEG signal.
107
Figure 5.3. Spectrum for solid PEG relative to ZnSe-air. The most prominent peaks are the C-O-C stretch at 1109 cm-1 and the C-H stretch at 2885 cm-1. Since it is believed that the ethereal oxygen is important in the adsorption of polyethers, the C-O-C stretch peak is of interest.
108
Figure 5.4. Spectrum for solid cupric sulfate pentahydrate relative to ZnSe-air. The most prominent peaks are at 670, 3630, and 3730 cm-1. There is also a peak at 1090 cm-1, which is near the C-O-C stretch peak of PEG at 1100 cm-1.
109
- Figure 5.5. Spectrum for 0.5 M CuSO4 and 70 ppm Cl on a Cu substrate, relative to water. A prominent sulfate peak is evident at 1100 cm-1, near a peak of interest for PEG.
The peaks at 3280 and 1640 cm-1 are due to error in referencing to the spectrum for water on Cu.
110 - Figure 5.6 shows spectra for a solution containing 0.5 M CuSO4, 70 ppm Cl , and
100 ppm PEG 1000 on both Si and Cu substrates. The two spectra have been purposely shifted vertically in order to provide visual clarity. The spectra are relative to those of water on Si and Cu, respectively. The spectra on both substrates are nearly identical, suggesting similar absorption behavior on both Si and Cu. The strong peak at 1100 cm-1 may correspond to the C-O-C stretch of adsorbed PEG; however, as discussed earlier, a sulfate peak is also present in this region, and the effect of the sulfate in the bulk solution may overlap or mask the signal of the adsorbed PEG.
111
- Figure 5.6. FTIR spectra for solutions containing 0.5 M CuSO4, 70 ppm Cl , and
100 ppm PEG 1000 on Cu (bottom) and on Si (top) substrates. The spectra indicate similar behavior on both Si and Cu. The spectra were shifted vertically for clarity.
112 The effect of various ions on the adsorption of PEG on Cu has been measured and is shown in Figure 5.7. The four solutions contained some of the specified components
- as listed in the figure in the following concentrations: 0.5 M CuSO4, 70 ppm Cl ,
-1 100 ppm PEG 1000, and 0.67 M Na2SO4. The strong peak near 1100 cm is absent when the solution contains only PEG and Cl- in water. This peak is present in the absence of
- Cl when CuSO4 is present. This peak is also evident if CuSO4 is replaced by Na2SO4 in a concentration such that the ionic strength of the solution remains about the same
(0.67 M Na2SO4 in place of 0.5 M CuSO4). It is unclear whether this peak is indicative of PEG adsorption or if the signal of the adsorbed PEG is masked by the bulk sulfate peak. One solution to overcome this issue is to use a species such as CuCl2 in place of
CuSO4.
Figure 5.8 shows the adsorption of PEG on Cu in a solution containing 0.5 M
CuCl2 and 100 ppm PEG 1000. CuCl2 was used as an alternative copper salt to avoid the interference of the sulfate peak near 1100 cm-1. As shown in Figure 5.8, a peak is evident at 1100 cm-1. Since sulfate is not present in this solution, this peak is likely associated with the C-O-C stretch of PEG. This result indicates that it is possible to detect the adsorption of PEG in a solution that does not contain sulfate.
113
Figure 5.7. Spectra for various solutions on Cu. The solutions contain the specified
- components in the following concentrations: 0.5 M CuSO4, 70 ppm Cl , 100 ppm
PEG 1000, and 0.67 M Na2SO4. No peak is observed with a solution of 100 ppm PEG and 70 ppm Cl- in water.
114
Figure 5.8. FTIR spectrum for a solution containing 0.5 M CuCl2 and 100 ppm
-1 PEG 1000, relative to 0.5 M CuCl2. The C-O-C stretch peak of PEG at 1100 cm can be detected in the absence of sulfate.
115 One additional technique was used to separate the effect of the sulfate peak on the
C-O-C stretch peak of PEG. Using the previous set-up, the dismantling and reassembly of the cell introduced error when comparing a spectrum of interest to a reference spectrum. Because of this error, the C-O-C stretch peak of PEG could not be separated from the sulfate peak also present at about the same wavenumber. To overcome this, a flow cell was used to introduce new solution into the cell. Figure 5.9 shows the spectrum
- for PEG adsorption on Cu from a solution containing 0.5 M CuSO4, 70 ppm Cl , and
- 100 ppm PEG 1000, relative to 0.5 M CuSO4 and 70 ppm Cl . A peak associated with the
C-O-C stretch of PEG is evident at 1100 cm-1. Using this technique, it is possible to detect PEG adsorption from a copper sulfate solution.
116
- Figure 5.9. FTIR spectrum for a solution containing 0.5 M CuSO4, 70 ppm Cl , and
- 100 ppm PEG 1000, relative to 0.5 M CuSO4 and 70 ppm Cl . Different solutions were introduced using a flow cell. A peak associated with the C-O-C stretch of PEG is evident at 1100 cm-1.
117 5.3 Quartz Crystal Microbalance
A second technique used to study the adsorption behavior of PEG was QCM.
Two sets of experiments were done in order to shed light on the adsorption mechanism of
PEG and Cl-. These experiments were performed on gold substrates; future experiments on copper will extend the results to the Cu system; however, it is believed that the adsorption behavior of PEG on gold will be similar to that on copper. During these experiments the frequency difference between the substrate in contact with solution and the substrate in air was recorded. Prior to performing experiments with additives, the effect of adding solution to the cell was determined. After a stable frequency signal was obtained with 0.5 M CuSO4 in the cell, subsequent small volumes of CuSO4 were added.
As shown in Figure 5.10, after a sharp spike in the frequency difference due to introducing additional solution, the frequency reading stabilized (~2 min) to the original value. This suggests that just adding solution in small volumes has only a transient effect on the frequency difference, and any frequency shifts observed upon the introduction of additives are due to adsorption. The noise present in the data in the following figures is likely caused by the electrical connections in the circuit.
118
Figure 5.10. Frequency response upon addition of small volumes of 0.5 M CuSO4 to the cell initially containing 0.5 M CuSO4 solution. The arrows indicate times when additional solution was introduced into the cell. The frequency signal stabilizes to nearly the original value, indicating that additional solution does not affect adsorption on the substrate.
119 The first set of experiments involved filling the cell initially with 0.5 M CuSO4 and then adding first Cl- followed by PEG. Although repeating the experiment multiple times revealed some error in reproducibility and significant noise, a number of experiments indicated identical adsorption behavior. A typical result is shown in Figure
5.11. After a stable frequency reading was established with ~5 mL CuSO4 in the cell,
- 0.25 mL of concentrated Cl solution was added, such that the resulting 0.5 M CuSO4 solution contained 70 ppm Cl-. An increase in frequency difference of ~30 Hz, corresponding to ~130 ng/cm2, was observed. This is assumed to be consistent with the adsorption of a monolayer; a mass of ~100 ng/cm2 is expected for a surface fully covered with chloride atoms. After the frequency reading reached a stable value, 0.25 mL of concentrated PEG solution was added, such that the resulting solution also contained
100 ppm PEG (molecular mass 1000 g/mol). Upon addition of PEG, an additional increase in frequency difference of ~30 Hz (~130 ng/cm2) was observed. This change in mass corresponds approximately to the saturation surface coverage of PEG
(610-11 mol/cm2 for PEG molecular mass 4000 g/mol).3
120
- Figure 5.11. Frequency response upon addition of Cl and PEG to a 0.5 M CuSO4 solution. When 70 ppm Cl- was added to the cell, a change in frequency difference of
~30 Hz, corresponding to ~130 ng/cm2, was observed. When 100 ppm PEG (molecular mass 1000 g/mol) was added to the Cl--containing solution, a similar change in frequency difference of ~30 Hz (~130 ng/cm2) was observed.
121 A typical result for an experiment in which PEG was added before Cl- is shown in
Figure 5.12. After a stable frequency signal was established with ~5 mL 0.5 M CuSO4 in the cell, 0.25 mL of concentrated PEG solution was added, such that the resulting solution contained 0.5 M CuSO4 and 100 ppm PEG (molecular mass 1000 g/mol). A small decrease in frequency difference was observed. This indicates that PEG did not adsorb. The decrease in the frequency difference could be caused by changes in the solution properties near the substrate. After the frequency signal reached a stable value,
0.25 mL of concentrated Cl- solution was added, such that the resulting solution also contained 70 ppm Cl-. Upon addition of Cl-, an increase in frequency difference of
~50 Hz, corresponding to ~220 ng/cm2, was observed. This frequency shift is larger than
- the shift observed when Cl was introduced to CuSO4 solution without PEG (Figure
5.11), possibly indicating that both Cl- and PEG adsorb after Cl- is added.3
122
- Figure 5.12. Frequency response upon addition of PEG and Cl to a 0.5 M CuSO4 solution. When 100 ppm PEG (molecular mass 1000 g/mol) was added, a small decrease in the frequency difference was observed. When 70 ppm Cl- was added to the PEG- containing solution, an increase in frequency difference of ~50 Hz (~220 ng/cm2) was observed.
123 The above QCM experiments provide some mechanistic insight into the adsorption behavior of Cl- and PEG. As seen in Figure 5.11, Cl- adsorbs on the metal surface. PEG adsorbs in the presence of Cl-, and the magnitudes of the frequency shifts indicate that similar masses (~130 ng/cm2) of Cl- and PEG are adsorbed. Since the mass
- of a Cl atom is similar to the mass of one repeat unit of PEG (-CH2-CH2-O-), the results in Figure 5.11 suggest that approximately one Cl- atom adsorbs for each ether oxygen in the adsorbed PEG. While it is apparent that PEG adsorbs in the presence of Cl-, the results in Figure 5.12 suggest that very little PEG adsorbs in a CuSO4 solution without chloride. Once Cl- is added, it seems that both PEG and Cl- adsorb. These results suggest that Cl- helps bind PEG to the surface, allowing for the increase in polarization seen in electrochemical experiments.
124 5.4 Effect of Cu+ on Copper Deposition
One additional mechanistic aspect presented here is the effect of Cu+ on the rate of copper deposition. Injection experiments62 were performed in which 10 ppm Cu+ (as
CuO) was added to acidified (pH~2) 0.5 M CuSO4 solutions containing either no additives, 100 ppm PEG, or 100 ppm PEG and 70 ppm Cl-. The cathode was polarized at
10 mA/cm2, and the overpotential response, shown in Figure 5.13, was measured. When
Cu+ is injected into a solution containing no additives, the overpotential decreases by
~30 mV. No change in overpotential is observed when Cu+ is injected into solutions containing PEG or PEG and Cl-. These results suggest that the rate of copper deposition is increased by the addition of Cu+ to a solution containing no additives, but the presence of PEG (with or without Cl-) prevents this increased deposition rate. This suggests that the Cu+ ions may bind with the PEG molecules, which slow the rate of Cu+ transport to the surface.
125
Figure 5.13. Overpotential response to injection of 10 ppm Cu+ into acidified (pH~2)
0.5 M CuSO4 solutions containing the indicated additives. The cathode was polarized at
10 mA/cm2. When Cu+ is added to a solution containing no additives, the overpotential decreases by ~30 mV. No change in overpotential is observed when Cu+ is added to solutions containing PEG or PEG and Cl-.
126 5.5 Conclusions
Initial FTIR experiments were done to study the adsorption of PEG on Cu. The prominent peaks for PEG adsorption are expected at 1109 cm-1 (C-O-C stretch) and at
2885 cm-1 (C-H stretch). Because a sulfate peak also occurs near 1100 cm-1, the effect of sulfate may mask the peak associated with the C-O-C stretch of adsorbed PEG.
Additional experiments to achieve better resolution and greater reproducibility may resolve the complication of the sulfate and C-O-C peaks. Also, experiments without sulfate, using e.g., only chlorides, will resolve this issue.
- QCM results suggest that PEG does not adsorb in CuSO4 unless Cl is also
- present. Cl adsorbs on the electrode in CuSO4 and is required for the PEG molecules to bind to the surface. The magnitudes of the frequency shifts indicate that approximately one Cl- atom adsorbs for every ether oxygen atom of the adsorbed PEG.
127 CHAPTER 6
Major Conclusions and Recommendations for Future Work
6.1 Major Conclusions
Following is a summary of the major conclusions derived in this dissertation in regard to copper electropolishing and electroplating processes in applications relating to metallization of semiconductor interconnects.
6.1.1 Electropolishing
(i) Experimental results indicate the presence of a highly resistive surface film during
electropolishing. Different potential response is observed when the current is stepped
from zero to values below or at the limiting current. The potential response at the
limiting current indicates a two-step process: (i) an initial sharp potential increase
followed by a slow gradual increase over ~70 s followed by (ii) a large, almost
instantaneous potential increase (~1.5 V). This apparent two-step process can not be
explained by a simple transport model. The steep potential increase suggests the
presence of a highly resistive surface film.
(ii) A mechanistic model for electropolishing based on an insoluble film build-up on the
anode with cupric ions as the transport-limited species is proposed. A two-step
mechanism for the copper electropolishing process is presented. Cupric ions are
identified as the transport-limited species. During the first step of the mechanism, a
flux imbalance between the dissolving copper ions and their transport out of the
boundary layer region causes a buildup of copper ions at the anode surface. This
concentration increases until the solubility limit of the copper salt is reached and a
128 resistive film is formed. The film composition has not been identified but for analysis
is assumed to be a copper phosphate species. During the second step, the resistive
film increases in thickness due to the imbalance between the current-dependent
dissolution of the anode and the removal of copper ions by transport. Under constant
current, the film thickness increases linearly with time, giving rise to a corresponding
increase in the potential. The proposed mechanism accounts for the potential
response to current transients and is consistent with experimental observations,
including those relating to the effect of the bulk water concentration.
6.1.2 Novel Polyethers Extending Gap-fill Capabilities
(i) Novel polyether suppressors, which extend via-fill capability to 25 nm features, are
identified. Several polyether compounds, including polyoxyethylated β-naphthol,
UCON Lubricant 75-H-450, and polyoxyethylene lauryl ether, have been identified,
all exhibiting significantly improved suppression over PEG. Injection studies
confirm that these molecules are displaced by SPS during gap fill similarly to PEG,
an essential requirement for the bottom-up fill process. The expected bottom-up fill
ratio for these polyethers is about 4 times greater than that of PEG, making them
attractive as replacements for PEG, thus extending the bottom-up fill process to
features as narrow as 25 nm. SEM cross-section images confirm the advantage of
polyoxyethylated β-naphthol over PEG 1000 in the fill of features with aspect ratio of
10:1 (~50 nm width). The identification of these polyethers significantly extends via-
fill capability beyond the present limit of 60 nm using PEG. It becomes evident by
inspecting the different effective polyethers that the inhibition effects are not just
129 associated with steric effects, but relate, at least in part, to specific chemical
interactions with the specific chemical groups attached to the polyether backbone.
(ii) PEG adsorption studies: Attenuated total reflectance Fourier transform infrared
spectroscopy and quartz crystal microbalance provide insight into the adsorption
behavior of PEG. The results of preliminary ATR-FTIR experiments identify the
peaks associated with adsorbed PEG, although the measurements may be
compromised by the overlap from the sulfate species. QCM results indicate that PEG
- - does not adsorb on Au surfaces in CuSO4 unless Cl is also present. Cl readily
adsorbs on the electrode in CuSO4 and is required for PEG to adsorb. This suggests a
mechanism by which PEG is bound to the surface by interactions with Cl-. The
magnitudes of the frequency shifts indicate that approximately one Cl- atom adsorbs
for every ether oxygen atom of the adsorbed PEG.
130 6.2 Recommendations for Further Studies
The following recommendations are made for further study.
(i) The mechanistic model for electropolishing described herein applies to a flat
electrode. However, in order to extend this model to predict leveling or planarization
performance on a patterned substrate, two key aspects need further exploration: (a) A
more detailed understanding of the surface film properties, e.g., its physico-chemical
properties, thickness, and the charge transport mechanism through it, is required.
This requires the development of in-situ surface analysis and characterization
techniques, such as QCM, FTIR, atomic force microscopy (AFM) and ellipsometry.
(b) In the context of the surface film model for electropolishing described herein, the
wafer-scale planarization efficiency is expected to depend on the distribution of the
film thickness (and therefore the resistance offered by the film) over the wafer
surface. The film thickness distribution in turn depends on the interaction of the
wafer topography with the hydrodynamic conditions close to the wafer surface. To
understand these effects, and to develop a comprehensive wafer-scale planarization
model, advanced numerical modeling analysis will be required.
(ii) Although several polyethers are identified as more effective than PEG in the
suppression of copper deposition, the cause of their improved suppression behavior is
not well understood. No simple correlation exists between suppression and any easily
identifiable molecular parameter. Results suggest that the chemical structure of the
polyether is of importance to the suppression effect. Since the molecular mechanism
of PEG suppression is not well understood, it is difficult to develop an understanding
of the suppression effect of other polyethers. If the prevailing molecular mechanism
131 of PEG adsorption, whether by steric effects, complexation with Cu+, or a
combination of these, is determined, the mechanism can be extending to describe the
adsorption of other polyethers. Techniques which may provide insight into this
mechanism include FTIR and QCM.
(iii) Only preliminary ATR-FTIR studies of PEG adsorption have been completed. To
date, this technique has not been reported for the adsorption of PEG on copper.
Additional experiments are required to better understand the adsorption of PEG. The
current experimental set-up reported in this work, requires removal of the wafer in
each experiment and its re-assembly, resulting in different pressure between the wafer
and the prism, which in turn affects the signal, leading to a large signal to noise ratio.
One solution to this problem is to use a cell where the electrolyte can be pumped in
and out of the cell and the wafer does not need to be removed in between
experiments. This should eliminate the error associated with removing and replacing
the wafer in between each experiment. Additional experiments are required to
separate the effects of the sulfate peak and C-O-C peak of PEG. Minimizing the error
associated with reproducibility should help with this effort. An additional experiment
would be to collect the spectrum of 100 ppm PEG in a solution that does not contain
sulfate, such as 0.5 M NaCl or CuCl2, relative to 0.5 M NaCl or CuCl2 without PEG.
This eliminates the sulfate from solution and should provide sufficient ionic strength,
which may be necessary for PEG adsorption. Other possible experimental changes to
reduce error are to obtain scans at higher resolution (2 cm-1 or smaller) and to use a
liquid N2-cooled detector in place of the room temperature detector used in the
present work. Once results are obtained for the adsorption of PEG in the absence and
132 presence of Cl-, this technique can be extending to study the adsorption of the other
polyethers identified in Chapter 4.
(iv) QCM has proved to be a useful technique for studying the adsorption behavior of
PEG. Several additional experiments may provide insight into the polyether
adsorption mechanism.
(a) Effect of Cu2+ and Cu+ ions on the adsorption of PEG and Cl-. Experiments
similar to those reported in Chapter 5, using water in place of CuSO4 solution,
- 2+ could indicate whether Cl and/or PEG adsorbs in the absence of Cu . CuSO4
could then be added to the cell to determine if any changes in adsorption occur
due to the addition of Cu2+. If the addition of Cu2+ does indicate a change in
adsorption, similar experiments could then be performed using Cu+ in place of
Cu2+. If the adsorption behavior of PEG appears to be dependent on the presence
of Cu+, this would lend support to the proposed PEG/Cl-/Cu+ complex
mechanism.5, 12, 15, 16
(b) Adsorption behavior of other polyethers. Experiments similar to those reported in
Chapter 5, using polyethers identified in Chapter 4 in place of PEG, could provide
insight into the adsorption mechanism of these polyethers. If a larger adsorbed
weight is observed for one of these polyethers, e.g., polyoxyethylated β-naphthol,
in comparison to PEG, this could indicate that polyoxyethylated β-naphthol
adsorbs with a greater surface coverage than that of PEG or that it is bound more
closely to the surface. If no difference in adsorbed weight is observed, this could
indicate a more complex chemical interaction.
133 (v) The above techniques can also be applied to the study of SPS adsorption and its
displacement of PEG during via fill. One proposed QCM experiment is to initially
- fill the cell with 0.5 M CuSO4 and 70 ppm Cl . 100 ppm PEG can then be added to
the cell, with an expected increase of adsorbed weight. When 10 ppm SPS is added
to the PEG-containing solution, the frequency difference is expected to decrease as
SPS displaces PEG. The steady-state frequency reading can then be compared to an
- experiment where SPS is added to a CuSO4 and Cl solution without PEG. This
should provide insight into whether SPS displaces all of the PEG or if some of the
PEG remains on the surface.
134 Bibliography
1. D. C. Edelstein, G. A. Sai-Halasz and Y.-J. Mi, IBM J. of Res. and Dev., 39, 383
(1995).
2. P. C. Andricacos, C. Uzoh, J. O. Dukovic, J. Horkans and H. Deligianni, IBM J. Res.
Develop., 42, 567 (1998).
3. R. Akolkar and U. Landau, J. Electrochem. Soc., 151, C702 (2004).
4. D. Josell, B. Baker, C. Witt, D. Wheeler and T. P. Moffat, J. Electrochem. Soc., 149,
C637 (2002).
5. M. Yokoi, S. Konishi and T. Hayashi, Denki Kagaku, 52, 218 (1984).
6. J. D. Reid and A. P. David, Plat. Surf. Finish., 74, 66 (1987).
7. J. P. Healy and D. Pletcher, J. Electroanal. Chem., 338, 155 (1992).
8. J. J. Kelly and A. C. West, J. Electrochem. Soc., 145, 3472 (1998).
9. J. J. Kelly and A. C. West, J. Electrochem. Soc., 145, 3477 (1998).
10. K. R. Hebert, J. Electrochem. Soc., 148, C726 (2001).
11. V. D. Jovic and B. M. Jovic, J. Serb. Chem. Soc., 66, 935 (2001).
12. Z. V. Feng, X. Li and A. A. Gewirth, J. Phys. Chem. B, 107, 9415 (2003).
13. K. Doblhofer, S. Wasle, D. M. Soares and K. G. Weil, J. Electrochem. Soc., 150,
C657 (2003).
14. Y. Jin, K. Kondo, Y. Suzuki, T. Matsumoto and D. P. Barkey, Electrochem. Solid-
State Lett., 8, C6 (2005).
15. K. R. Hebert, J. Electrochem. Soc., 152, C283 (2005).
16. K. R. Hebert, S. Adhikari and J. E. Houser, J. Electrochem. Soc., 152, C324 (2005).
17. M. L. Walker, L. J. Richter and T. P. Moffat, J. Electrochem. Soc., 152, C403 (2005).
135 18. W. Dow, M. Yen, W. Lin and S. Ho, J. Electrochem. Soc., 152, C769 (2005).
19. J. G. Long, P. C. Searson and P. M. Vereecken, J. Electrochem. Soc., 153, C258
(2006).
20. M. J. Willey and A. C. West, J. Electrochem. Soc., 153, C728 (2006).
21. Y. Qin, X. Li, F. Xue, P. M. Vereecken, P. Andricacos, H. Deligianni, R. D. Braatz and R. C. Alkire, J. Electrochem. Soc., 155, D223 (2008).
22. M. E. H. Garrido and M. D. Pritzker, J. Electrochem. Soc., 155, D332 (2008).
23. J. W. Gallaway and A. C. West, J. Electrochem. Soc., 155, D632 (2008).
24. P. A. Jacquet, Trans. Electrochem. Soc., 69, 629 (1936).
25. W. C. Elmore, J. Appl. Phys., 10, 724 (1939).
26. R. Cantolini, A. Bernhardt and S. Mayer, J. Electrochem. Soc., 141, 2503 (1994).
27. D. Padhi, J. Yahalom, S. Gandikota and G. Dixit, J. Electrochem. Soc., 150, G10
(2003).
28. A. C. West, H. Deligianni and P. C. Andricacos, IBM J. Res. & Dev., 49, 37 (2005).
29. R. D. Grimm, A. C. West and D. Landolt, J. Electrochem. Soc., 139, 1622 (1992).
30. J. Edwards, J. Electrochem. Soc., 100, 189C (1953).
31. S. H. Glarum and J. H. Marshall, J. Electrochem. Soc., 132, 2872 (1985).
32. R. Vidal and A. C. West, J. Electrochem. Soc., 142, 2682 (1995).
33. B. Du and I. I. Suni, J. Electrochem. Soc., 151, C375 (2004).
34. T. P. Hoar and T. W. Farthing, Nature, 169, 324 (1952).
35. V.G. Levich, Physicochemical Hydrodynamics, Prentice-Hall, Englewood Cliffs, NJ
(1962).
136 36. A. J. Bard and L. R. Faulkner, Electrochemical Methods, 2nd ed., p. 306, John Wiley
& Sons, New York (2001).
37. J. Newman, J. Electrochem. Soc., 113, 501 (1966).
38. L. Young, Anodic Oxide Films, p. 19, Academic Press, London (1961).
39. A. J. Bard and L. R. Faulkner, Electrochemical Methods, 2nd ed., p. 386, John Wiley
& Sons, New York (2001).
40. S. H. Glarum and J. H. Marshall, J. Electrochem. Soc., 132, 2878 (1985).
41. L. Young, Anodic Oxide Films, p. 246, Academic Press, London (1961).
42. A. J. Bard and L. R. Faulkner, Electrochemical Methods, 2nd ed., p. 92, John Wiley
& Sons, New York (2001).
43. “FEMLAB”, multi-physics modeling package, COMSOL Ltd., Burlington, MA,
USA.
44. K. Kojima and C. W. Tobias, J. Electrochem. Soc., 120, 1202 (1973).
45. Handbook of Chemistry and Physics, 45th ed., R. C. Weast, Editor, Chemical Rubber
Co., Cleveland (1964).
46. J. H. Christensen and R. B. Reed, Ind. & Eng. Chem., 47, 1277 (1955).
47. D. Stoychev and C. Tsvetanov, J. Appl. Electrochem., 26, 741 (1996).
48. R. Akolkar, private communication (2009).
49. D. Papapanayiotou, R. N. Nuzzo and R. C. Alkire, J. Electrochem. Soc., 145, 3366
(1998).
50. M. E. Biggin and A. A. Gewirth, J. Electrochem. Soc., 148, C339 (2001).
51. T. Iwaoka, P. R. Griffiths, J. T. Kitasako and G. G. Geesey, Appl. Spectrosc., 40,
1062 (1986).
137 52. K. P. Ishida and P. R. Griffiths, J. Coll. Interfac. Sci., 213, 513 (1999).
53. H.-F. Wang, Y.-G. Yan, S.-J. Huo, W.-B. Cai, Q.-J. Xu and M. Osawa, Electrochim.
Acta, 52, 5950 (2007).
54. E. Gileadi and V. Tsionsky, J. Electrochem. Soc., 147, 567 (2000).
55. V. Tsionsky and E. Gileadi, Mater. Sci. Eng. A, 302, 120 (2001).
56. J. J. Kelly, K. M. A. Rahman, C. J. Durning and A. C. West, J. Electrochem. Soc.,
145, 492 (1998).
57. V. Tsionsky, L. Daikhin, G. Zilberman and E. Gileadi, Faraday Discuss., 107, 337
(1997).
58. H. B. Martin and P. W. Morrison, Electrochem. Solid-State Lett., 4, E17 (2001).
59. N. J. Harrick, Internal Reflection Spectroscopy, 3rd ed., p. 40, Harrick Scientific
Corporation, Ossining, New York (1987).
60. G. Sauerbrey, Z. Phys., 155, 206 (1959).
61. T. W. G. Solomons and C. B. Fryhle, Organic Chemistry, 7th ed., p. 79, John Wiley
& Sons, New York (2002).
62. J. Adolf, private communication (2009).
138